Datasets:

timaeus

/

pile-pubmed_central

like 0

Follow

Timaeus 9

The prevalence of childhood obesity is emerging as a major health problem^(^[@ref1]^)^. In several studies, a negative or U-shaped relationship between sleep duration and weight status has been observed in infants^(^[@ref2]^)^, children^(^[@ref3]^,^[@ref4]^)^, adolescents^(^[@ref4]^,^[@ref5]^)^ and adults^(^[@ref3]^,^[@ref6]^)^. Moreover, meta-analyses analysing data from studies in children have concluded that shorter sleep duration was associated with higher odds of being obese^(^[@ref3]^)^. This might have long-term implications^(^[@ref7]^)^ because tracking of overweight from childhood to adulthood occurs^(^[@ref8]^)^, thereby increasing the risk of developing certain non-communicable diseases, as overweight is associated with certain non-communicable diseases^(^[@ref9]^)^. The potential influence of shorter sleep duration on body weight balance is not fully understood. One possible underlying mechanisms is that shorter sleep duration makes an impact on the hormonal regulation of appetite. Indeed, lack of sleep has been reported to decrease leptin levels, increase ghrelin levels, alter glucose homeostasis, and activate the orexin system^(^[@ref10]^)^. Furthermore, shorter sleep duration might also promote overeating and weight gain by increasing the time available for eating, especially in the evening where sedentary activities, such as watching television, and snacking on highly palatable and energy-dense foods are common^(^[@ref11]^)^. When shortened sleep duration leads to sleepiness and/or fatigue during daytime this may limit the motivation for being physically active and promote sedentary behaviours instead. It has therefore also been speculated that there is an association between sleep duration, BMI and physical activity; however, results from studies concerning this have been contradictory^(^[@ref5]^,^[@ref12]^)^. Only a few studies have examined the association between sleep duration and dietary quality^(^[@ref13]^--^[@ref15]^)^. This issue is particularly relevant from a public health standpoint because increased energy intake appears to be the most plausible explanation as to why children with shorter sleep duration have a higher risk of becoming obese. The main aim of the present study was to examine the relationship between intake of different foods and nutrients, and sleep duration in Danish children. We hypothesised that short sleepers have higher energy intake and less healthy food intake compared with long sleepers. Materials and methods {#sec1} ===================== Sample {#sec1-1} ------ The present study was based on data from the Danish National Survey of Dietary Habits and Physical Activity (DANSDA) 2003--2008, which is a nationwide and representative cross-sectional survey. The data collection was evenly distributed during the period. The study population comprised a simple random sample of 4- to 14-year-old children retrieved from The Central Office of Civil Registration. In comparison with census data from Statistics Denmark, the distribution of sex and age of the participants could be characterised as representative and parental education as close to representative for the Danish population of children aged 4--14 years. Data from a total of 802 children were available and included in the present study. The study was conducted according to the guidelines laid down in the Declaration of Helsinki and was approved by the Danish Data Protection Agency. The Danish National Committee on Health Research Ethics decided that DANSDA did not require their approval. Anthropometric variables {#sec1-2} ------------------------ Information about height and weight was obtained through a personal face-to-face interview with one of the parents. The prevalence of overweight and obesity in the study population was defined according to international age- and sex-specific BMI cut-off values^(^[@ref16]^)^, corresponding to BMI values of 25 and 30 kg/m^2^, respectively, for adults aged ≥18 years. Sleep duration {#sec1-3} -------------- Sleep duration was reported in hours and minutes in a diary integrated in a food record. Every day, on the same 7 d as the food record was completed, a simple question about sleep duration during the last night and day was answered: 'what length of time did you sleep for during the last 24 h (please include daytime napping)?'. Children and their parents were instructed in person by trained interviewers on how to complete food and sleep recordings. The parents were responsible for completing the records and for deciding to what extent their children were capable of assisting. Thus, parents reported all or the majority of the dietary intake and sleep for the youngest children, but somewhat less for the older children. Mean sleep duration per d was calculated and used for subsequent analyses. Days where children were in bed with illness were omitted. For each child, a minimum of 4 d with measurements of diet and sleep duration was required for inclusion in the statistical analyses. Children with short, medium and long sleep duration were defined within each year of age according to the tertiles of sleep duration. Dietary intake {#sec1-4} -------------- Dietary intake was recorded every day for seven consecutive days in food records with pre-coded response categories, which included open-answer options. For food items not included in the pre-coded food record, the participants wrote the type of food and portion size eaten in open-answer categories. The amount of foods eaten was given in predefined household measures (cups, spoons, slices, etc.) or estimated from photographs of various portion sizes. Children also received a food record booklet to take to school or to other places outside of the home on the days of assessment. Details about the method and calculation of intake of food and nutrients are described elsewhere^(^[@ref17]^)^. In order to characterise the children\'s dietary intake, energy intake, macronutrients and major food groups as well as energy-dense foods were included in the analyses. Parental education {#sec1-5} ------------------ Parental educational level was recorded as the educational level of parent in the household with the highest education, and divided into four levels: (1) basic school (10 years or less of total education); (2) vocational education, upper secondary school (10--12 years); (3) short higher education (13--15 years) (primarily theoretical); and (4) long higher education (15+ years) (primarily theoretical). Statistical analysis {#sec1-6} -------------------- All data were analysed with SPSS (version 20; IBM SPSS Statistics, Inc.) and a significance level of *P* \< 0·05 was used (two-tailed tests). Normality was checked visually with histograms as well as with Kolgomorov--Smirnoff tests. As commonly found, the data for most foods and nutrients were not normally distributed (positively skewed). Logarithmic and square root transformations were attempted, but did not produce distributions with sufficient normality. For this reason and insufficient homogeneity of variance, non-parametric analyses were performed. BMI *Z*-scores were calculated. Spearman\'s ρ correlation coefficients were calculated to analyse associations between sleep duration and BMI and between sleep duration and age. The χ^2^ test was used to analyse differences between weight status, parental education and sleep duration and sex. The Kruskal--Wallis test was used to analyse differences in BMI between parental education. Multiple linear regression analyses were performed with Bonferroni adjustment for multiple tests separately for each dietary variable with sleep duration (h/d) as the dependent variable and with BMI (kg/m^2^), age (years), sex (male, female), energy intake (kJ/d) and parental education level (1, 2, 3, 4; see Parental education section above) as forced covariates in one model, and with age (years), sex (male, female), energy intake (kJ/d) and parental education level (1, 2, 3, 4) as forced covariates in another model. Normality of the residuals from the models was regarded as sufficient. In order to further elucidate the dietary intakes in relation to sleep duration, the Kruskal--Wallis test was applied to test for statistically significant differences in dietary intake between age-dependent tertiles of sleep duration. Results are provided as mean values and standard deviations. Results {#sec2} ======= Descriptive characteristics {#sec2-1} --------------------------- Characteristics of the study population are given in [Table 1](#tab01){ref-type="table"}, and are shown for the three age groups: 4--6 years, 7--10 years and 11--14 years. This is the standard age grouping of children in the National Survey of Dietary Habits and Physical Activity, since these age groups correspond to the Danish institutions for children: kindergarten (4--6 years), introductory school period (7--10 years) and intermediate school period (11--14 years). Within each age group, no differences between boys and girls were found regarding age (*P ≥* 0·29), height (*P ≥* 0·52), weight (*P ≥* 0·30), BMI (*P ≥* 0·17), parental education (*P ≥* 0·39) and sleep duration (*P ≥* 0·95). Therefore, further analyses were conducted with both sexes combined. Table 1.Characteristics of the study population(Mean values and standard deviations; number of subjects and percentages)Age group...4--6 years7--10 years11--14 yearsSubjects (*n*)212289301Sex (%)Male505445Female504655Height (cm)Mean118·6139·4161·4[sd]{.smallcaps}8·99·39·7Weight (kg)Mean22·133·050·5[sd]{.smallcaps}4·17·410·9BMI (kg/m^2^)Mean15·616·819·3[sd]{.smallcaps}1·92·63·1Weight status\*Overweight (%)Male7·512·919·4Female16·014·915·0Obesity (%)Male0·03·82·2Female3·84·54·2Sleep duration (h)Mean10·29·68·9[sd]{.smallcaps}0·60·50·7Parental education (%)†(1) Basic school9911(2) Vocational education414342(3) Short higher education81015(4) Long higher education423832[^1][^2] Prevalence of overweight and obesity {#sec2-2} ------------------------------------ Prevalence of overweight and obesity is given in [Table 1](#tab01){ref-type="table"}. No sex differences were observed with regard to weight status for the 7- to 10-year-old and 11- to 14-year-old children (*P* = 0·59 and *P* = 0·93), but there were more overweight and obese girls than boys in the 4- to 6-year-old children (*P* = 0·004). Sleep duration, age and BMI {#sec2-3} --------------------------- Age was negatively correlated to sleep duration (ρ = --0·68; *P* \< 0·001). An inverse relationship between sleep duration and BMI *Z*-scores (4--14 years: ρ = --0·41, *P* \< 0·001; 4--6 years: ρ = --0·19, *P =* 0·79; 7--10 years: ρ = --0·16, *P =* 0·008; 11--14 years: ρ = --0·14, *P =* 0·013) was also observed, but was not statistically significant for the youngest age group. Furthermore, length of parental education was negatively correlated with BMI of the child (*P* \< 0·05). However, there were no differences between short, medium and long sleepers regarding parental education (*P* = 0·44). Sleep duration and dietary intake {#sec2-4} --------------------------------- Multiple linear regression analyses were performed with Bonferroni adjustment for multiple tests separately for each dietary variable with sleep duration (h/d) as the dependent variable and with BMI (kg/m^2^), age (years), sex (male, female), energy intake (kJ/d) and parental education (1, 2, 3, 4) as forced covariates in model A, and with age (years), sex (male, female), energy intake (kJ/d) and parental education (1, 2, 3, 4) as forced covariates in model B. Sleep duration was not associated with energy intake (model A: b = --0·015, *P* = 0·20; model B: b = --0·015, *P* = 0·20), but there was a trend towards a positive association with intake of dietary fibre (model A: b = 0·006, *P* = 0·05; model B: b = 0·006, *P* = 0·05) and vegetables (model A: b = 0·011, *P* = 0·05; model B: b = 0·009, *P* = 0·05), and a negative association with intake of poultry (model A: b = --0·002, *P* = 0·02; model B: b = --0·002, *P* = 0·02) and a trend towards a negative association with liquid 'discretionary calories' (model A: b = --0·010, *P* = 0·05; model B: b = --0·011, *P* = 0·06), as shown in [Table 2](#tab02){ref-type="table"}. Excluding BMI as covariate did not change the results markedly. Table 2.Coefficients from multiple linear regression analyses performed separately for each dietary variable with sleep duration (h/d) as the dependent variable and with BMI (kg/m^2^), age (years), sex (male, female), energy intake (kJ/d) and parental education (1, 2, 3, 4)[\*](#tfn2_1){ref-type="table-fn"} as forced covariates in model A, and with age (years), sex (male, female), energy intake (kJ/d) and parental education (1, 2, 3, 4)[\*](#tfn2_1){ref-type="table-fn"} as forced covariates in model B (*n* 802)Model AModel Bb*P*†b*P*†Energy (MJ/d)−0·0150·20−0·0150·20Protein (E%)‡0·0060·53−0·0070·56Fat (E%)‡−0·0010·16−0·0080·99SFA (E%)‡−0·0020·82−0·0010·92MUFA (E%)‡−0·0090·47−0·0070·55PUFA (E%)‡−0·0020·90−0·0080·99Carbohydrate (E%)‡0·0020·630·0020·72Added sugars (E%)‡−0·0030·53−0·0030·55Dietary fibre (g per10 MJ)0·0060·050·0060·05High-fat milk and milk products (g per10 MJ)\<0·0010·47\<0·0010·40Low-fat milk and milk products (g per10 MJ)−0·0080·31−0·0080·27Rye bread (g per10 MJ)0·0010·140·0010·16Wheat bread (g per10 MJ)\<0·0010·99−0·0030·95Sugary breakfast cereals (g per10 MJ)§0·0010·980·0010·99Fruit (g per10 MJ)−0·0060·68−0·0070·65Vegetables (g per10 MJ)0·0110·050·0090·05Meat and meat products (g per10 MJ)0·0070·900·0040·94Poultry (g per10 MJ)−0·0020·02−0·0020·02Fish and seafood (g per10 MJ)0·0020·140·0020·13Fast food and light meals (g per10 MJ)∥0·0090·710·0100·71Snacks (g per10 MJ)¶−0·0010·49−0·0010·54Sweets and chocolate (g per10 MJ)0·0010·460·0010·47Solid 'discretionary calories' (g per 10 MJ)\*\*\<0·0010·62\<0·0010·66Liquid 'discretionary calories' (g per 10 MJ)††−0·0100·05−0·0110·06[^3][^4][^5][^6][^7][^8][^9][^10][^11] In Kruskal--Wallis analyses for differences between children in the lowest, medium and highest age-dependent tertiles of sleep duration, only intake of liquid 'discretionary calories' was significantly lower in long sleepers than in short and medium sleepers (*P =* 0·03), as seen in [Table 3](#tab03){ref-type="table"}. However, there was a tendency towards a higher intake of dietary fibre (*P* = 0·09) and lower intake of added sugars (*P* = 0·10) in long sleepers ([Table 4](#tab04){ref-type="table"}). Table 3.Intake of food groups in 4- to 14-year-old children with short (*n* 268), medium (*n* 267) and long (*n* 267) sleep duration(Mean values and standard deviations)Sleep duration...ShortMediumLongMean[sd]{.smallcaps}Mean[sd]{.smallcaps}Mean[sd]{.smallcaps}*P*\*High-fat milk and milk products (g/d)16622377251040·59Low-fat milk and milk products (g/d)1872562202471802170·25Rye bread (g/d)4936493453370·29Wheat bread (g/d)7242724371420·87Sugary breakfast cereals (g/d)†2·563·363·070·14Fruit (g/d)1811381771241781380·82Vegetables (g/d)1227512970131750·23Meat and meat products (g/d)9042923988410·34Poultry (g/d)1919181818200·42Fish and seafood (g/d)1215111412130·86Fast food and light meals (g/d)‡7875888380760·27Snacks (g/d)§8128157110·86Sweets and chocolate (g/d)2222211822220·97Solid 'discretionary energy' (g/d)∥9558905388530·32Liquid 'discretionary energy' (g/d)¶2442192422192142240·03[^12][^13][^14][^15][^16][^17] Table 4.Intake of nutrients in 4- to 14-year-old children with short (*n* 268), medium (*n* 267) and long (*n* 267) sleep duration(Mean values and standard deviations)Sleep duration...ShortMediumLongMean[sd]{.smallcaps}Mean[sd]{.smallcaps}Mean[sd]{.smallcaps}*P*\*Energy (MJ/d)8·32·38·32·08·22·20·69Protein (g/d)6920691868190·75Fat (g/d)7626742174250·84SFA (g/d)3211329·932120·83MUFA (g/d)269·2257·5258·90·66PUFA (g/d)113·8113·0113·50·92Carbohydrate (g/d)2667626467262720·68Added sugars (g/d)6333613159320·10Dietary fibre (g/d)152·7175·0186·10·09[^18] Discussion {#sec3} ========== This is the first publication with representative population data on sleep duration and its relationship with dietary intake in Danish children. We found a negative association between sleep duration and BMI in children aged 4--14 years, which is in accordance with several other studies^(^[@ref2]^--^[@ref4]^,^[@ref6]^)^. There was no significant association between sleep duration and energy intake, and children with short, medium and long sleep duration did not differ with regard to energy intake. This observation is in agreement with what has been found previously in adults^(^[@ref6]^)^, implying that a small chronic energy gap associated with short sleeping is difficult to capture with the dietary assessment methods that are normally used in epidemiological research. When assessing the relationship between sleep duration and dietary variables with multiple linear regression analyses with BMI, age, sex, energy intake and parental education as covariates, we only found tendencies that intakes of vegetables and dietary fibre were positively associated with sleep duration, and that intakes of poultry and liquid 'discretionary calories' were negatively associated with sleep duration. Furthermore, when assessing differences between children in the lowest, medium and highest age-dependent tertiles of sleep duration, only intake of liquid 'discretionary calories' was significantly lower in long sleepers than in short and medium sleepers. There was, though, a tendency towards a higher intake of dietary fibre and lower intake of added sugars in long sleepers. Mutual adjustment for vegetables when assessing the association between intake of dietary fibre and sleep duration and adjustment of dietary fibre when assessing the association between intake of vegetables and sleep duration weakened both associations markedly (data not shown), suggesting that vegetables and dietary fibre are interdependently associated with sleep duration. It might be speculated that discretionary sugar and/or caffeine from liquid 'discretionary calories' may act as a stimulant keeping children longer awake, thereby contributing to explaining the inverse association between liquid 'discretionary calories' and sleep duration. With increasing age, there is an increasing autonomy in the dietary intakes of children, and in the pubertal period, which is a period of high general autonomy, the propensity for buying energy-dense foods might be high. But in younger ages it can be speculated that less sleep and higher intake of energy-rich foods are reflecting an overall lifestyle and that healthy behaviours are less relevant in the family. Other factors could also be considered, such as family upbringing, personal values and attitudes towards healthy eating habits and sleep hygiene, late-night activities such as screen time, for example, television viewing and computer games, daytime napping, and physical activity. Furthermore, compensating for insufficient weekday sleep during the weekends may also in part ameliorate the risk of childhood overweight^(^[@ref18]^)^. In the present study, sleep duration was based on the average sleep duration during the week, and differences on weekdays compared with weekend days were not taken into account. The variability in sleep patterns during the week and the potential effect of 'catch-up' sleep need to be further studied. Only a few other studies have examined the relationship between sleep duration and dietary intakes^(^[@ref4]^,^[@ref13]^--^[@ref15]^,^[@ref19]^,^[@ref20]^)^. The results from these studies are concordant with those found in the present study, suggesting that inadequate sleep may partly be associated with less healthy food habits in children. In 10- to 11-year-old Finnish children, inadequate sleep was associated with a greater likelihood of consuming energy-rich foods and less likelihood of consuming nutrient-dense foods^(^[@ref15]^)^. Further, sleepiness during daytime was associated with daily snacking in Japanese school girls, and with skipping breakfast and evening snacks in boys^(^[@ref19]^)^. In another study on Taiwanese adolescents, adequate sleep was associated with adopting a healthy diet, including eating breakfast daily, eating three meals per d, choosing foods with little oil, and drinking at least 1·5 litres of water per d^(^[@ref20]^)^. In 14- to 18-year-old Americans, a positive association between daytime sleep, which may have reflected an increased need for nocturnal sleep, and greater food cravings was found^(^[@ref14]^)^. Finally, among Iranian school girls, those who were overweight had shorter sleeping times and consumed 'less nutritious food', such as candies, chocolates and potato chips, more often than normal-weight girls^(^[@ref13]^)^. Although based on differing populations with different age groups and dietary intake variables, the general findings from studies of sleep duration and diet indicate more unfavourable dietary patterns in short sleepers. However, in a study of German children and adolescents no association was observed between sleep duration and a nutrition quality score reflecting consumption of healthy and unhealthy foods^(^[@ref4]^)^. Since the hedonic value of food intake might play a crucial role in the association between short sleep duration and overweight, this is an important target of future research^(^[@ref11]^)^. The observed inverse association between BMI and parental education is consistent with the results from other studies in children and adolescents^(^[@ref21]^)^. However, the fact that parental education was not associated with sleep duration in the present study is inconsistent with the findings of others^(^[@ref22]^)^, and might be explained by the fact that only 10 % of the participating children had parents with educational level 1 (the lowest level, basic school). This might be insufficient to detect a difference. Although sex differences have been observed in some other studies examining the association between diet and sleep^(^[@ref4]^,^[@ref14]^,^[@ref15]^)^, there were no sex differences regarding sleep duration, and adjusting for sex in the multiple linear regression analyses did not change the results markedly. Among the limitations of the present study is its cross-sectional nature that cannot disentangle cause and effect. To confirm a causal relationship between sleeping habits and eating patterns, and to determine whether short sleep duration leads to greater dietary intake or whether greater intake leads to shorter sleep duration, prospective, longitudinal studies are needed. Moreover, BMI was calculated from parent-reported height and weight, and might therefore be slightly biased. Few studies have investigated this bias and findings have been inconsistent^(^[@ref23]^,^[@ref24]^)^. However, several studies have found relatively good relationships between measured and self-reported estimates of children\'s and adolescents\' height and weight^(^[@ref25]^,^[@ref26]^)^. Additionally, we have no data on pubertal status, which might also influence both BMI^(^[@ref16]^)^ and sleep duration^(^[@ref27]^)^ and maybe also eating habits as well. However, no association between sleep duration and pubertal status has been found in obese children^(^[@ref28]^)^. Possibly, under-reporting may have blurred the results particularly in the oldest age group, as it has been well documented that under-reporting increases with the age of children^(^[@ref29]^)^, and occurs especially with food items perceived as unhealthy^(^[@ref30]^)^. Finally, data were collected over a period of 5 years, in which some variables in the environment might have changed, such as more electronic devices that might take time from sleep. The strengths of the present study include its nationwide character, as it is based on a nationally representative population study. We also had access to very detailed dietary data that enabled us to analyse the association of sleep duration with specific dietary variables. An additional strength of the present study is that sleep duration was reported every day for seven consecutive days and reported as a mean of 7 d and not as a measure of usual sleep duration. However, sleep duration was self-reported and not measured, but self-reported sleep duration has been found to be positively correlated with polysomnographic measurements^(^[@ref31]^)^. There may be a potential overestimation of sleep duration in our data because parental reports of sleep duration may be determined by the time of going to bed and getting up rather than on actual time of sleeping. However, sleep diaries have been found to have better agreement with objective measurements than sleep questionnaires^(^[@ref32]^)^, and therefore the quality of the sleep data in the present study may be considered relatively high. In conclusion, there was a negative association between sleep duration and BMI in the present sample of 4- to 14-year-old Danish children. However, sleep duration was not associated with energy intake and the proposal that children with short sleep duration have less healthy eating habits than children with longer sleep duration was only weakly supported by the present findings. Although causality needs to be elucidated further, the present findings are contributory to gaining a better understanding of the link between sleep duration, dietary intake and body weight, which may be valuable for further research and health strategies related to the prevention of overweight and obesity. The authors are grateful to Tue Christensen, Karin Hess Ygil and Karsten Kørup (Division of Nutrition, National Food Institute, Technical University of Denmark) for developing the software system and databases used in the processing of the data and for contributing to data processing. The present study is part of the OPUS project (Optimal well-being, development and health for Danish children through a healthy New Nordic Diet) supported by a grant from the Nordea Foundation. The authors designed the analyses of the present study (C. H. and B. W. R.) and were responsible for the design and collection of data from The Danish National Survey of Dietary Habits and Physical Activity (M. V. G., A. B.-J. and J. M.). C. H. and B. R. were responsible for analyses of data and writing the manuscript. All authors participated in the discussion of the results and revision of the manuscript. The authors declare no conflict of interest. [^1]: \* Weight status according to international cut-off values^(^[@ref16]^)^. [^2]: † Parental educational level (*n* 731): (1) basic school (10 years or less of total education); (2) vocational education, upper secondary school (10--12 years); (3) short higher education (13--15 years) (primarily theoretical); (4) long higher education (15+ years) (primarily theoretical). [^3]: E%, energy percentage. [^4]: Parental educational level: (1) basic school (10 years or less of total education); (2) vocational education, upper secondary school (10--12 years); (3) short higher education (13--15 years) (primarily theoretical); (4) long higher education (15+ years) (primarily theoretical). [^5]: † With Bonferroni adjustment for multiple tests. [^6]: ‡ Macronutrient energy percentages are calculated including alcohol. [^7]: § Sugar content \>10 g per 100 g. [^8]: ║ Burger, toast, spring roll, hotdog, French hotdog, sandwich, filled croissant, pizza, falafel, humus, filled patty shell. [^9]: ¶ Chips, popcorn, groundnuts, pistachio nuts, almonds, cheese dippers, pretzels. [^10]: \*\* Sum of the variables snacks and sweets and chocolate as well as chips, confectionery, ice cream and desserts. [^11]: †† Sugar-sweetened beverages, cider, ice tea. [^12]: \* Kruskal--Wallis test for difference between age-dependent tertiles of sleep duration. [^13]: † Sugar content \>10 g per 100 g. [^14]: ‡ Burger, toast, spring roll, hotdog, French hotdog, sandwich, filled croissant, pizza, falafel, humus, filled patty shell. [^15]: § Chips, popcorn, groundnuts, pistachio nuts, almonds, cheese dippers, pretzels. [^16]: ║ Sum of the variables snacks and sweets and chocolate as well as chips, confectionery, ice cream and desserts. [^17]: ¶ Sugar-sweetened beverages, cider, ice tea. [^18]: \* Kruskal--Wallis test for difference between age-dependent tertiles of sleep duration.

1. Introduction =============== The terrestrial cyanobacterium *Nostoc commune* (*N. commune*) is a cosmopolitan species that is distributed from the tropics to the polar regions of the Earth \[[@B1-marinedrugs-11-03124]\]. *N. commune* adapts to terrestrial environmental conditions \[[@B1-marinedrugs-11-03124],[@B2-marinedrugs-11-03124]\]. In its natural habitats, it forms visually conspicuous colonies that are subjected to frequent cycles of desiccation and wetting. Desiccated colonies have no metabolic activity and retain the ability to grow for more than 100 years \[[@B3-marinedrugs-11-03124],[@B4-marinedrugs-11-03124]\]. Upon rehydration, *N. commune* cells rapidly recover respiration and photosynthesis \[[@B5-marinedrugs-11-03124],[@B6-marinedrugs-11-03124],[@B7-marinedrugs-11-03124],[@B8-marinedrugs-11-03124]\]. This phenomenon is termed anhydrobiosis \[[@B9-marinedrugs-11-03124],[@B10-marinedrugs-11-03124],[@B11-marinedrugs-11-03124],[@B12-marinedrugs-11-03124]\]. *N. commune* is considered to be a prokaryotic model anhydrobiote that retains oxygenic photosynthetic capabilities in vegetative cells and does not differentiate into akinetes (spores) \[[@B1-marinedrugs-11-03124],[@B12-marinedrugs-11-03124]\]. In addition to extreme desiccation tolerance, *N. commune* colonies are exposed to direct solar radiation and can tolerate UV radiation stress \[[@B13-marinedrugs-11-03124],[@B14-marinedrugs-11-03124]\]. The mechanisms behind this extreme environmental stress tolerance are thought to involve multiple processes, and the ability to produce a biochemically complex extracellular matrix appears to be a required factor \[[@B15-marinedrugs-11-03124]\]. The structural constituents of this matrix in *N. commune* include extracellular polysaccharides (EPS) \[[@B15-marinedrugs-11-03124]\], water stress protein (WspA) \[[@B16-marinedrugs-11-03124]\] and UV absorbing pigments \[[@B14-marinedrugs-11-03124]\]. EPS, which account for 80% of the dry weight of *N. commune* colonies \[[@B17-marinedrugs-11-03124]\], play a major role in mechanisms, including desiccation tolerance, that protect cells from various stresses in severe habitats \[[@B7-marinedrugs-11-03124],[@B15-marinedrugs-11-03124]\]. Mycosporine-like amino acids (MAAs) are water-soluble and absorb specific UV-B radiation in the range of 280--320 nm \[[@B18-marinedrugs-11-03124],[@B19-marinedrugs-11-03124],[@B20-marinedrugs-11-03124],[@B21-marinedrugs-11-03124],[@B22-marinedrugs-11-03124],[@B23-marinedrugs-11-03124],[@B24-marinedrugs-11-03124]\]. Structurally distinct MAAs have been observed in eukaryotic algae and cyanobacteria, including marine, freshwater and terrestrial species \[[@B18-marinedrugs-11-03124],[@B19-marinedrugs-11-03124],[@B20-marinedrugs-11-03124]\]. MAAs have an important role in the overall strategy to reduce the deleterious effects of desiccation and environmental UV radiation \[[@B25-marinedrugs-11-03124]\], especially for the adaptation to terrestrial environments exposed to higher levels of UV-radiation than aqueous environments \[[@B26-marinedrugs-11-03124]\]. With their photoprotective and antioxidative properties, MAAs are natural bioactive compounds attractive to cosmeceutical and pharmaceutical applications \[[@B19-marinedrugs-11-03124],[@B20-marinedrugs-11-03124]\]. Physiological responses to osmotic water stress induced by high salt concentrations are thought to overlap with responses to matric water stress in dry environments \[[@B27-marinedrugs-11-03124]\]. It has been suggested that MAAs may function as osmotic solutes because of the accumulation of MAAs in halophilic cyanobacteria \[[@B28-marinedrugs-11-03124]\] and overlap between anti-stress compounds produced by marine and terrestrial cyanobacteria can be expected. *N. commune* synthesizes UV-A/B absorbing compounds that are secreted to the extracellular matrix \[[@B14-marinedrugs-11-03124],[@B15-marinedrugs-11-03124],[@B16-marinedrugs-11-03124]\]. Recently, two novel glycosylated MAAs were found in naturally growing *N. commune* colonies \[[@B29-marinedrugs-11-03124]\]. An MAA with an absorption maximum at 335 nm and a molecular mass of 478 Da was identified as a pentose-bound porphyra-334 derivative. The other identified MAA had double absorption maxima at 312 and 340 nm and a molecular mass of 1050 Da. The 544-Da lipid-soluble pigment scytonemin absorbs UV-A radiation of 320--400 nm and occurs exclusively in cyanobacterial sheaths \[[@B13-marinedrugs-11-03124],[@B14-marinedrugs-11-03124]\]. These glycosylated MAAs \[[@B29-marinedrugs-11-03124]\] and scytonemin \[[@B30-marinedrugs-11-03124],[@B31-marinedrugs-11-03124]\] have radical scavenging activities *in vitro*. Additionally, a unique antioxidative compound, nostocionone, was reported in *N. commune* \[[@B31-marinedrugs-11-03124]\]. *N. commune* is considered to adapt to terrestrial environments with high levels of solar radiation, as it produces both glycosylated MAAs and scytonemin with antioxidative activities in its extracellular matrix \[[@B14-marinedrugs-11-03124],[@B15-marinedrugs-11-03124],[@B16-marinedrugs-11-03124],[@B29-marinedrugs-11-03124],[@B30-marinedrugs-11-03124],[@B32-marinedrugs-11-03124]\]. *N. commune* is known to be genetically diverse, and four major genotypes, which are hardly morphologically distinguishable and have genetic differences that are not great enough to be recognized as distinct species, have been reported in Japan. These genotypes have been observed within small areas, such as the Kakuma Campus of Kanazawa University \[[@B33-marinedrugs-11-03124]\]. During the investigation of MAAs in naturally growing colonies from different sampling locations, we found novel MAAs that were neither the 478-Da MAA nor the 1050-Da MAA. In this study, we purified and characterized the chemical structures and radical scavenging activities of these newly found glycosylated MAAs. These findings provide new insights into the diversity and chemotaxonomic features of MAAs, as well as the biological functions of MAAs in the adaptation of the cyanobacterium *N. commune* to terrestrial environments. 2. Results and Discussion ========================= 2.1. UV Absorption Spectra of Water Extracts and Genotypes of *N. commune* -------------------------------------------------------------------------- *N. commune* colonies were found for which the water extract showed a characteristic UV-absorbing spectrum with an absorption maximum at 325 nm. However, this spectrum was different from known *N. commune* spectra, and thus, the MAA profile and genotype of this particular *N. commune* sample were examined further. In a typical HPLC chromatogram of the water extract with an absorption maximum at 325 nm, three distinct MAAs were detected as major MAAs with different retention times ([Figure 1](#marinedrugs-11-03124-f001){ref-type="fig"}). Because the retention times and UV absorption maxima were different from those of other known MAAs from *N. commune*, these MAAs were purified and characterized as described below. According to an analysis of the 16S rRNA gene sequence, the *N. commune* colony that contained MAAs with absorption maxima at approximately 325 nm was identified as genotype D, as described by Arima *et al.* \[[@B33-marinedrugs-11-03124]\]. {#marinedrugs-11-03124-f001} 2.2. 508-Da MAA with an Absorption Maximum at 334 nm ---------------------------------------------------- A MAA with an absorption maximum at 334 nm ([Figure 1](#marinedrugs-11-03124-f001){ref-type="fig"}) was purified from the field-isolated natural *N. commune* colonies ([Table 1](#marinedrugs-11-03124-t001){ref-type="table"}, [Figure S1](#marinedrugs-11-03124-s001){ref-type="supplementary-material"}). The UV absorption spectrum of the purified MAA showed a single absorption peak at 334 nm ([Figure 2](#marinedrugs-11-03124-f002){ref-type="fig"}a). The absorption maximum shifted reversibly to 332 nm in a high acidic solution (pH \< 2) and to 334 nm in alkaline solutions. The molecular mass was 508 Da, according to MALDI-TOF MS ([Table 2](#marinedrugs-11-03124-t002){ref-type="table"}). The absorption coefficient of this MAA in water was 71.4 L·g^−1^·cm^−1^ at 334 nm and the calculated molar absorption coefficient at 334 nm was 3.63 × 10^4^ M^−1^·cm^−1^. Because no MAA with a molecular mass of 508 Da had been previously reported, the chemical structure of the 508 Da MAA was further characterized. marinedrugs-11-03124-t001_Table 1 ###### Purification of 508-Da, 450-Da and 612-Da mycosporine-like amino acids (MAAs) with absorption maxima at 334, 322 and 322 nm, respectively, from *N. commune* ^a^*.* Step Volume (mL) MAA concentration ^b^ (mg·L^−1^) MAA amount (mg) Yield (%) ---------------------- ------------- ---------------------------------- ----------------- ----------- ---- Water extract 640 19.4 12.4 100 70% Ethanol solution 1960 5.4 10.6 85 Vacuum concentration 3.5 2374 8.3 67 Reverse-phase HPLC 508-Da 0.5 3525 1.8 15 450-Da 0.5 1969 1.0 8 612-Da 0.5 2404 1.2 10 Gel filtration 508-Da 0.5 2474 1.2 10 450-Da 0.5 1782 0.9 7 612-Da 0.5 2226 1.1 9 ^a^ Dry colonies of *N. commune* (30 g) that contained the MAAs with absorption maxima at 334 and 322 nm were used as starting materials and the MAAs were purified as described in Experimental Section. HPLC chromatograms and absorption spectra of the purified MAAs are shown in [Figure S1 in Supporting Information](#marinedrugs-11-03124-s001){ref-type="supplementary-material"}; ^b^ MAA concentrations were determined with an absorption coefficient of 120 L·g^−1^·cm^−1^ \[[@B34-marinedrugs-11-03124]\]. {#marinedrugs-11-03124-f002} marinedrugs-11-03124-t002_Table 2 ###### Summary of MALDI-TOF MS analysis of the purified 508-Da MAA with absorption maximum at 334 nm. Mass of fragment Relative abundance Neutral loss Deleted fragment ---------------------- ------------------ -------------------- ------------------------------- ------------------ MS of purified MAA 509.12 100 465.13 81 44 CO~2~ 451.12 10 58 CO~2~ + CH~2~ 427.07 27 82 \- 425.04 58 84 \- 421.15 58 88 2CO~2~ 419.15 24 90 CO~2~ + 2CH~2~ + H~2~O 407.10 40 102 2CO~2~ + CH~2~ 403.13 37 106 2CO~2~ + H~2~O 387.11 57 122 \- 385.07 35 124 2CO~2~ + 2H~2~O MS^2^ of *m*/*z* 509 **509.12** 347.27 100 162 Hexose 303.24 6 206 Hexose + CO~2~ (or C~2~H~4~O) 279.20 7 230 \- MS^2^ of *m*/*z* 465 **465.13** 303.29 100 162 Hexose 285.27 3 180 Hexose + H~2~O 235.22 4 230 \- The IR spectrum of the 508-Da MAA ([Figure 3](#marinedrugs-11-03124-f003){ref-type="fig"}a) was compared to those of known MAAs, including porphyra-334 \[[@B35-marinedrugs-11-03124]\] and a glycosylated porphyra-334 with a molecular mass of 478 Da from *N. commune* \[[@B29-marinedrugs-11-03124]\] ([Table 3](#marinedrugs-11-03124-t003){ref-type="table"}). Similar IR absorption peaks were observed, which suggested that the 508-Da MAA had a similar structure to the known MAAs. The absorption peak at 1560 cm^−1^ in the 508-Da MAA indicated the presence of the conjugated imine as a characteristic structure of MAA chromophore, corresponding to those at 1540 cm^−1^ in porphyra-334 and at 1558 cm^−1^ in the 478-Da glycosylated porphyra-334 derivative. The characteristic absorption peak at 3334 cm^−1^ in the 508-Da MAA indicated the presence of several hydroxyl functional groups, and the absorption peaks in the range of 1200--1300 cm^−1^ suggested the presence of a sugar, which was consistent with the predicted structure as described below. MALDI-TOF MS/MS analysis was performed on the parent molecular ion fragment with *m*/*z* 509 and molecular ion fragment with *m*/*z* 465 ([Table 2](#marinedrugs-11-03124-t002){ref-type="table"}). A fragment with *m*/*z* 347, a molecular mass identical to that of porphyra-334 \[[@B36-marinedrugs-11-03124],[@B37-marinedrugs-11-03124]\], was detected in the second MS of the fragment with *m*/*z* 509. The neutral loss of 162 Da suggested the deletion of a hexose (C~6~H~10~O~5~) from the fragment with *m*/*z* 509. The fragment with *m*/*z* 303 suggested the deletion of CO~2~. The fragment with *m*/*z* 303 was also detected in the second MS of the fragment with *m*/*z* 465. The deletion of 162 Da from the fragment with *m*/*z* 465 to yield *m*/*z* 303 indicated the presence of a hexose in the MAA structure (Table 2). Additionally, in the MS analysis the fragments with *m*/*z* 465 and *m*/*z* 421 suggested the consecutive deletion of CO~2~ and the presence of two carboxyl groups in the MAA structure ([Table 2](#marinedrugs-11-03124-t002){ref-type="table"}). These fragmentation patterns were similar to those from the MS analyses of known MAAs \[[@B38-marinedrugs-11-03124]\]. These data suggest that the 508-Da MAA is a hexose-bound porphyra-334 derivative. {#marinedrugs-11-03124-f003} marinedrugs-11-03124-t003_Table 3 ###### Comparison of the IR spectra of the purified MAAs with that of porphyra-334. 508-Da MAA with A~max~ at 334 nm Porphyra-334 478-Da MAA with A~max~ at 335 nm ---------------------------------- -------------- ---------------------------------- 3400 3334 3300 2942, 2975 1718 1600 1606 1560 1540 1558 1379, 1395 1380 1382 1301, 1355 1310 1231, 1257 1273 1138 1132 1075 1080 1072 1030 1006 968 IR spectra were recorded with a Fourier transform infrared (FT-IR) spectrometer (Nicolet NEXUS 470 FT-IR) according to the KBr disk method. The wavenumbers of the absorption bands for the 508-Da MAA are compared to those for the known MAAs porphyra-334 \[[@B35-marinedrugs-11-03124]\] and 478-Da pentose-bound porphyra-334 \[[@B29-marinedrugs-11-03124]\]. The FT-IR spectrum of the 508-Da MAA is shown in [Figure 3](#marinedrugs-11-03124-f003){ref-type="fig"}a. This predicted structure was confirmed by NMR spectroscopic analysis. Both the known chemical shifts for porphyra-334 and hexose were observed in the 508-Da MAA ([Table 4](#marinedrugs-11-03124-t004){ref-type="table"}). Typical ^13^C-chemical shifts assignable to a cyclohexenimine chromophore (C1, 2, 3, 4, 5, 6) and amino acid substituents (C9, 10, 11, 12, 13, 14) of the 508-Da MAA were observed at the same regions in porphyra-334. The characteristic coupling pattern (AB quartet) of methylene protons at C4 and C6 was determined (*J* = 17.4 Hz), which is not shown in porphyra-334. It was hard to identify which carbon (C4 and C6) should be assigned to the glycine-substituted side and *vice versa* because chemical shifts at C9 and C11 protons were very close to recognize the correlation in heteronuclear multiple bond correlation (HMBC) spectrum. However, the ^13^C-chemical shifts in the 508-Da MAA were highly similar to those in the pentose-bound porphyra-334 derivative, which has been reported previously ([Table 4](#marinedrugs-11-03124-t004){ref-type="table"}). The correlations in COSY and HMBC spectra are summarized in [Table 5](#marinedrugs-11-03124-t005){ref-type="table"} and shown in [Figure 4](#marinedrugs-11-03124-f004){ref-type="fig"} as red arrows and blue dashed lines, respectively. Characteristic correlations assignable to the MAA backbone appeared and a probable HMBC correlation suggesting a putative hexose-binding site was observed (shown as red dashed arrow in [Figure 4](#marinedrugs-11-03124-f004){ref-type="fig"}). marinedrugs-11-03124-t004_Table 4 ###### Summary of the NMR analysis of the 508-Da MAA with an absorption maximum at 334 nm. C 508-Da MAA with A~max~ at 334 nm 478-Da pentose-bound porphyra-334 Porphyra-334 -------- ---------------------------------- ----------------------------------- -------------- --------------------- ------------------------------- -------------------- 1 ^a^ 161.4 \- 161.6 \- 161.6 \- 2 128.6 \- 128.4 \- 126.0 \- 3^a′^ 163.4 \- 163.2 \- 163.2 \- 4^b^ 36.1 2.82, 3.03 (ABq, 17.4) 36.4 3.00 32.5 2.75 5 73.3 \- 73.1 \- 71.3 \- 6^b′^ 36.6 2.80, 2.91 (ABq, 17.4) 35.9 2.83 33.0 2.77 7 76.1 3.47, 3.76 (d, 10.1, each) 77.7 3.90, 3.67 67.1 3.61 (s) 8 62.3 3.67 (s) 62.2 3.70 (s) 59.0 3.73 (s) 9 49.6 4.09 (d, 1.8) 49.5 4.06 (d, 2.4) 47.0 4.07 (s) 10^c^ 177.6 \- 177.7 \- 177.6 \- 11 67.5 4.10 (d, 4.6) 67.3 4.09 (d, 4.5) 64.0 4.12 (d, 5.0) 12^c′^ 178.2 \- 178.2 \- 178.0 \- 13 74.4 4.34 (dq, 4.6, 6.4) 71.0 4.31 (dq, 4.5, 6.5) 68.0 4.33 (m, 5.0, 6.4) 14 22.4 1.24 (d, 6.4) 22.3 1.26 (d, 6.5) 19.0 1.26 (d, 6.4) Methyl α- [d]{.smallcaps}-Man ^13^C ^1^H 1′ 101.8 4.96 (d, 3.2) 106.4 4.37 (d, 7.6) 102.2 4.66 2′ 71.4 3.88 73.5 3.60 (dd, 7.6, 9.6) 71.4 3.82 3′ 72.2 3.97 75.0 3.68 (dd, 9.6, 3.6) 72.1 3.65 4′ 70.9 3.82 71.1 3.95 (m) 68.3 3.53 5′ 72.3 3.80 69.1 3.92, 3.65 73.9 3.51 6′ 64.2 3.55, 3.72 62.5 3.79, 3.65 NMR spectra were recorded with a JEOL ECS400 spectrometer in D~2~O as a solvent. Chemical shifts (ppm) for the purified 508-Da MAA with A~max~ at 334 nm are compared with those for the known MAA 478-Da pentose-bound porphyra-334 \[[@B29-marinedrugs-11-03124]\], porphyra-334 \[[@B35-marinedrugs-11-03124]\] and methyl α-[d]{.smallcaps}-Man \[[@B39-marinedrugs-11-03124]\]. The coupling patterns and constants (Hz) are shown in parentheses. x, x′, Chemical shifts may be exchangeable for the 508-Da MAA. marinedrugs-11-03124-t005_Table 5 ###### Characteristic correlations in COSY and HMBC spectra assignable to the 508-Da MAA backbone. COSY HMBC ----- ---------- ---------------- H4 H4 C2, C3, C5, C6 H6 H6 C1, C2, C4, C5 H7 H7 C4, C5, C6 H8 C2 H9 C3, C10 H11 H13 C1, C12 H13 H11, H14 H14 H13 C11, C13 H1′ C7 ^a^ ^a^ Probable correlation. {#marinedrugs-11-03124-f004} The presence of a hexose moiety in the 508-Da MAA was confirmed by the ^13^C chemical shift in methyl α-[d]{.smallcaps}-mannose; chemical shift at the anomer carbon (C1′) was almost the same as that in methyl α-[d]{.smallcaps}-mannose, indicating that the substitution position is at the anomer position. Remaining peaks were assigned to one CH~2~ (C6′) and four CH carbon (C2′, 3′, 4′, 5′) from the result of Distortionless Enhancement by Polarization Transfer (DEPT) spectrum. The chemical shifts of these peaks were also consistent with those of methyl α-[d]{.smallcaps}-mannose. In the 508-Da MAA, the hexose binding position could be thought of as at C7 position because the CH~2~ protons at C7 position coupled to each other (*J* = 10.1 Hz). This coupling pattern was not observed in porphyra-334 but was observed in the pentose-bound 478-Da MAA ([Table 4](#marinedrugs-11-03124-t004){ref-type="table"}). In our previous report, the pentose-binding position at the C7 methylene was identified because of the apparent correlation between the C7 carbon and the anomer (C1′) proton in the HMBC spectrum \[[@B29-marinedrugs-11-03124]\]. In the 508-Da MAA, the correlation between C7 and H1′ was not observed apparently ([Table 5](#marinedrugs-11-03124-t005){ref-type="table"}, [Figure 4](#marinedrugs-11-03124-f004){ref-type="fig"} red dashed arrow), but geminal coupling at C7 protons suggesting magnetic anisotropy was observed. Thus the high rotation barrier around C5--C7 bond due to a large substituent can be predicted and this idea agrees with the presence of a monosaccharide at C7 position of the 508-Da MAA. The dihedral angle dependence of C5--C7 rotational energy was calculated by using Chem-3D software. The rotational barrier of the C7 substituted porphyra-334 was much greater than that of porphyra-334, which cannot be surmounted at room temperature (data not shown). This simulation can be supported by the previous report concerning neocedumoside derivatives who have a similar chemical structure around a hexose-binding site \[[@B40-marinedrugs-11-03124]\]. [Table 6](#marinedrugs-11-03124-t006){ref-type="table"} shows the chemical shifts of the hexose-binding sites of the 508-Da MAA, porphyra-334, neocedumoside and its hydrolysate. The geminal coupling of magnetically anisotropic methylene protons (*J* = 10.1 Hz) was detected due to the deshielding effect of the neighboring σ-electron. The equatorial proton was observed at lower magnetic field region by 0.43 ppm rather than the axial proton. The ^1^H-chemical shifts at C7 position in the 508-Da MAA showed two apparently separated signals by 0.29 ppm with geminal coupling between them. The hydrolysis of neocedumoside results in axial and equatorial peaks getting closer to their center of balance \[[@B40-marinedrugs-11-03124]\]. The center of balance of two separated signals in the 508-Da MAA was 3.61 ppm, which is identical to the corresponding chemical shift in porphyra-334. These data support the idea that magnetic anisotropy of methylene protons is caused by the substitution of hexose. In addition, the ^13^C chemical shift in neocedumoside is downfield shifted by 8.2 ppm from that of hydrolysate, suggesting a characteristic NMR feature on glycolysis. A similar shift (9.0 ppm) from porphyra-334 was also observed in the 508-Da MAA. These NMR analyses and molecular simulation support the substitution of hexose at C7 position. [Figure 4](#marinedrugs-11-03124-f004){ref-type="fig"} shows the predicted structure of the 508-Da MAA from *N. commune*. marinedrugs-11-03124-t006_Table 6 ###### Comparison of ^13^C and ^1^H chemical shifts at the hexose binding site of the 508-Da MAA, porphyra-334, neocedumoside and its hydrolysate. -------------------------------------------------------------------------------------- 508-Da MAA Porphyra-334 Neocedumoside Neocedumoside-hydrolysate ------- ----------------- -------------- ----------------- --------------------------- ^13^C 76.1 67.1 79.0 70.8 ^1^H 3.47 (d, 10.1)\ 3.61 (s) 3.40 (d, 10.1)\ 3.51 (d, 11.0)\ 3.76 (d, 10.1) 3.83 (d, 10.1) 3.54 (d, 11.0) -------------------------------------------------------------------------------------- Chemical shifts (ppm) for the purified 508-Da MAA are compared with those for the known MAA porphyra-334 \[[@B29-marinedrugs-11-03124]\], neocedumoside and its hydrolysate \[[@B40-marinedrugs-11-03124]\]. The coupling patterns and constants (Hz) are shown in parentheses. 2.3. 612-Da MAA with an Absorption Maximum at 322 nm ---------------------------------------------------- An MAA with an absorption maximum at 322 nm ([Figure 1](#marinedrugs-11-03124-f001){ref-type="fig"}) was purified from the same *N. commune* colonies ([Table 1](#marinedrugs-11-03124-t001){ref-type="table"}, [Figure S1](#marinedrugs-11-03124-s001){ref-type="supplementary-material"}). The UV absorption spectrum of the purified MAA showed a single absorption peak at 322 nm ([Figure 2](#marinedrugs-11-03124-f002){ref-type="fig"}b); similar to the 508-Da MAA, the absorption maximum reversibly shifted to 320 nm in a highly acidic solution. The molecular mass was 612 Da, according to MALDI-TOF MS ([Table 7](#marinedrugs-11-03124-t007){ref-type="table"}). The accurate molecular mass was determined by FAB MS to predict the elemental composition. A molecular ion fragment with *m*/*z* 613.2462 was detected and its predicted molecular formula was C~24~H~41~N~2~O~16~ within 1 ppm error. The absorption coefficient of this MAA in water was 46.07 L·g^−1^·cm^−1^ at 322 nm, and its calculated molar absorption coefficient at 322 nm was 2.82 × 10^4^ M^−1^·cm^−1^. Because no MAA with a molecular mass of 612 Da had been previously reported, the chemical structure of the 612 Da MAA was further characterized. marinedrugs-11-03124-t007_Table 7 ###### Summary of MALDI-TOF MS analysis of the purified 612-Da MAA with absorption maximum at 322 nm. Mass of fragment Relative abundance Neutral loss Deleted fragment ---------------------- ------------------ -------------------- ------------------------------------- ------------------ MS of purified MAA 613.17 18 569.19 95 44 CO~2~ 451.12 4 162 Hexose 407.13 17 206 CO~2~ + Hexose 389.12 100 224 CO~2~ + Hexose + H~2~O 349.09 21 264 Hexose + C~4~H~6~O~3~ 227.02 16 386 CO~2~ + 2Hexose + H~2~O MS^2^ of *m*/*z* 613 **613.17** 451.21 100 162 Hexose 407.21 1 206 Hexose + CO~2~ 389.23 8 224 Hexose + CO~2~ + H~2~O 317.20 6 296 \- 289.18 14 324 2Hexose 241.17 10 372 \- 185.14 6 428 \- 91.10 6 522 \- MS^2^ of *m*/*z* 569 **569.19** 407.37 100 162 Hexose 392.32 13 177 Hexose + CH~3~ 389.33 37 180 Hexose + H~2~O 375.31 10 194 Hexose + H~2~O + CH~2~ 245.25 26 324 2Hexose 227.24 16 342 2Hexose + H~2~O 209.22 21 360 2Hexose + 2H~2~O 191.21 8 378 2Hexose + 3H~2~O 177.19 18 392 2Hexose + 3H~2~O + CH~2~ 91.10 12 478 \- MS^2^ of *m*/*z* 451 **451.21** 407.00 2 44 CO~2~ 289.20 100 162 Hexose 245.17 6 206 Hexose + CO~2~ 227.19 2 224 Hexose + CO~2~ + H~2~O 170.13 1 281 Hexose + CO~2~ + H~2~O + C~3~H~5~O 151.14 1 300 Hexose + CO~2~ + 2H~2~O + C~3~H~6~O 91.10 2 360 \- MS^2^ of *m*/*z* 407 **407.13** 245.28 100 162 Hexose 230.24 3 177 Hexose + CH~3~ 227.25 15 180 Hexose + H~2~O 209.23 5 198 Hexose + 2H~2~O 199.21 4 208 Hexose + C~2~H~6~O 177.21 6 230 Hexose + 3H~2~O + CH~2~ 91.11 3 316 \- MS^2^ of *m*/*z* 389 **389.12** 227.29 100 162 Hexose 212.24 9 177 Hexose + CH~3~ 209.26 36 180 Hexose + H~2~O 191.24 54 198 Hexose + 2H~2~O 181.21 17 208 Hexose + C~2~H~6~O 170.19 8 219 Hexose + C~3~H~5~O 159.20 9 230 \- 91.11 12 298 \- MS^2^ of *m*/*z* 349 **349.09** 187.19 100 162 Hexose 169.18 8 180 Hexose + H~2~O 151.16 25 198 Hexose + 2H~2~O Similar IR absorption peaks were observed when the 612-Da MAA IR spectrum ([Figure 3](#marinedrugs-11-03124-f003){ref-type="fig"}b) was compared with those from the known MAA palythine triacetate \[[@B41-marinedrugs-11-03124]\] and the 1050-Da glycosylated MAA from *N. commune* \[[@B29-marinedrugs-11-03124]\] ([Table 8](#marinedrugs-11-03124-t008){ref-type="table"}), which suggested that the 612-Da MAA had a similar structure to the known MAAs. The characteristic absorption peak at 3370 cm^−1^, but not that at 3310 cm^−1^, in the 612-Da MAA suggested the presence of a large number of hydroxyl groups, and the absorption peaks in the range of 1200--1300 cm^−1^ suggested the presence of sugars, which was consistent with the presence of two hexose rings as described below. The absorption peak at 1542 cm^−1^ suggested the presence of the conjugated imine in the 612-Da MAA. MALDI-TOF MS/MS analysis was performed on the parent molecular ion fragment with *m*/*z* 613 and on the other main molecular ion fragments with *m*/*z* 569, 451, 407, 389 and 349 ([Table 7](#marinedrugs-11-03124-t007){ref-type="table"}). A plausible fragmentation pattern could be suggested with regard to the mass spectra ([Figure 5](#marinedrugs-11-03124-f005){ref-type="fig"}). The ion fragment with *m*/*z* 569 indicated the deletion of CO~2~ from the parent molecular ion fragment with *m*/*z* 613, suggesting the presence of a carboxyl group. Subsequent deletions of 162 and 18 Da from the fragment with *m*/*z* 569 to produce the fragments with *m*/*z* 407 and *m*/*z* 389 indicated the deletions of hexose (C~6~H~10~O~5~) and H~2~O, respectively. MS/MS analysis of the fragments with *m*/*z* 407 and 389 led to the fragment at *m*/*z* 227, which indicated the deletion of another hexose ([Table 7](#marinedrugs-11-03124-t007){ref-type="table"} and [Figure 5](#marinedrugs-11-03124-f005){ref-type="fig"}). marinedrugs-11-03124-t008_Table 8 ###### Comparison of IR spectra of the 612-Da MAA with known MAAs. 612-Da MAA with A~max~ at 322 nm Palythine triacetate 1050-Da MAA with A~max~ at 312 and 340 nm ---------------------------------- ---------------------- ------------------------------------------- 3370 3399 3310 2932 2929 1725, 1740 1660 1607 1590 1617 1542 1535--1555 1541 1388 1400 1346 1277, 1216 1275 1150 1067, 1036 1076, 1046 972 IR spectra were recorded with a Fourier transform infrared (FT-IR) spectrometer (Nicolet NEXUS 470 FT-IR) according to the KBr disk method. The wavenumbers of absorption bands for the 612-Da MAA are compared with those for the known MAA palythine triacetate \[[@B41-marinedrugs-11-03124]\] and the 1050-Da MAA \[[@B29-marinedrugs-11-03124]\]. The IR spectrum of the 612-Da MAA is shown in [Figure 3](#marinedrugs-11-03124-f003){ref-type="fig"}b. {#marinedrugs-11-03124-f005} The successive deletions of 162 Da from the parental molecular fragment with *m*/*z* 613 to produce *m*/*z* 451 and *m*/*z* 289 indicates the presence of two hexoses (C~6~H~10~O~5~) in the parent molecular ion with *m*/*z* 613; the resultant fragment with *m*/*z* 289 can be assumed to be a palythine-threonine (C~12~H~20~N~2~O~6~ + H^+^). Consistent with this assumption, the deletion of 44 Da from *m*/*z* 289 to produce *m*/*z* 245 indicated the deletion of CO~2~ and suggested the presence of a carboxyl group in the molecular fragment with *m*/*z* 289. The fragment with *m*/*z* 227 could lose two molecules of H~2~O and CH~3~ to produce the fragments with *m*/*z* 209, 191 and 177, respectively. The fragment with *m*/*z* 227 might also lose a part of the threonine chain (C~3~H~6~O) and H~2~O to produce the fragment with *m*/*z* 170 (or *m*/*z* 169) and 151 ([Table 7](#marinedrugs-11-03124-t007){ref-type="table"} and [Figure 5](#marinedrugs-11-03124-f005){ref-type="fig"}). In a MS/MS analysis of the molecular ion fragment with *m*/*z* 349, the neutral loss of 162 Da indicated the presence of hexose in the fragment with *m*/*z* 349, and the fragments with *m*/*z* 169 and *m*/*z* 151 suggested the subsequent deletions of two H~2~O molecules ([Table 7](#marinedrugs-11-03124-t007){ref-type="table"} and [Figure 5](#marinedrugs-11-03124-f005){ref-type="fig"}). These fragmentation patterns were similar to the results from MS analyses of the known MAAs \[[@B42-marinedrugs-11-03124],[@B43-marinedrugs-11-03124]\] and confirmed the proposed structure of the 612-Da MAA with the presence of two hexoses and a carboxyl group. ^1^H, ^13^C and 2D NMR experiments were performed in D~2~O. The ^13^C and ^1^H NMR spectra of the 612-Da MAA were compared to those of the known MAAs mycosporine-glycine and palythine-threonine sulfate as well as β-[d]{.smallcaps}-glucose and methyl β-[d]{.smallcaps}-glucose ([Table 9](#marinedrugs-11-03124-t009){ref-type="table"}). Signals with identical chemical shifts to 3-aminocyclohexenimine (C1, 2, 3, 4, 5, 6) and threonine (C9, 10, 11, 12) were observed in the 612-Da MAA, similar to those of palythine-threonine sulfate. In mycosporine-glycine the chemical shifts assignable to carbonyl and imine carbon are obviously distinguishable (159.7 ppm and 187.2 ppm in [Table 9](#marinedrugs-11-03124-t009){ref-type="table"}, \[[@B44-marinedrugs-11-03124]\]). However, the chemical shifts corresponding to imine carbon (C3) and amino alkene carbon (C1) in the 612-Da MAA overlapped each other because of the conjugation between them. In palythine-threonine sulfate these chemical shifts assigned to C1 and C3 carbons also overlapped each other ([Table 9](#marinedrugs-11-03124-t009){ref-type="table"}, \[[@B45-marinedrugs-11-03124]\]). Therefore, high-resolution ^13^C NMR measurement focusing on the X-range at around 160 ppm was performed to separate two distinct signals assignable to imine and amino alkene carbons at C1 and C3 positions in the 612-Da MAA (NMR data shown in [Supporting Information](#marinedrugs-11-03124-s001){ref-type="supplementary-material"}). These two separated signals demonstrated the presence of the 3-aminocyclohexenimine ring in the 612-Da MAA. In the 2D NMR spectra, characteristic correlations assignable to the MAA backbone were detected. The correlations in COSY and HMBC spectra are summarized in [Table 10](#marinedrugs-11-03124-t010){ref-type="table"} and shown in [Figure 6](#marinedrugs-11-03124-f006){ref-type="fig"} as red arrows and blue dashed lines. marinedrugs-11-03124-t009_Table 9 ###### Summary of the NMR analysis of the 612-Da MAA with an absorption maximum at 322 nm. C 612-Da MAA with A~max~ at 322 nm Palythine-threonine sulfate Mycosporine-glycine ------- ---------------------------------- -------------------------------- ------------------------------ ----------------------- ------- ------------------------ 1^a^ 163.6 \- 160.6 \- 159.7 \- 2 127.7 \- 125.0 \- 130.4 \- 3^a′^ 163.6 \- 160.6 \- 187.2 \- 4^b^ 37.0 2.85, 3.10 (ABq, 17.6) 34.0 2.81 (ABq, 17.0) 45.4 2.50, 2.73 (ABq, 17.0) 5 73.3 \- 69.8 \- 72.9 6^b′^ 38.9 2.76, 3.03 (ABq, 17.0) 36.0 2.77 (ABq, 17.0) 33.8 2.72, 2.83 (ABq, 17.0) 7 76.0 3.49 (d, 10.0), 3.78 (d, 10.0) 72.5 3.90 (s) 68.4 3.57 (s) 8 62.1 3.69 (s) 59.3 3.55 (s) 60.2 3.64 (s) 9 66.4 4.17 (d, 4.6) 64.5 3.98 (d, 4.5) 43.7 4.24 (s) 10 177.3 \- 174.9 \- 174.5 \- 11 75.7 4.34 (dq, 4.6, 6.4) 68.2 4.20 (dq, 4.5, 6.0) 12 18.9 1.28 (d, 6.4) 19.5 1.12 (d, 6.0) Methyl β-[d]{.smallcaps}-Glc β-[d]{.smallcaps}-Glc ^13^C ^1^H ^13^C ^1^H 1′ 107.4 5.09 (s) 104.0 4.27 96.8 4.64 1″ 101.7 5.00 (d, 3.7) NMR spectra were recorded with a JEOL ECS400 spectrometer in D~2~O as a solvent. Chemical shifts (ppm) for the purified 612-Da MAA with A~max~ at 322 nm are compared with those for the known MAA mycosporine-glycine \[[@B44-marinedrugs-11-03124]\], palythine-threonine sulfate \[[@B45-marinedrugs-11-03124]\] and methyl β-[d]{.smallcaps}-Glc and β-[d]{.smallcaps}-Glc \[[@B39-marinedrugs-11-03124]\]. The coupling patterns and constants (Hz) are shown in parentheses. x, x′, Chemical shifts may be exchangeable for palythine-threonine sulfate and the 612-Da MAA. marinedrugs-11-03124-t010_Table 10 ###### Characteristic correlations in COSY and HMBC spectra assignable to the 612-Da MAA backbone. COSY HMBC ----- --------- ---------------- H4 H4 C2, C3, C5, C6 H6 H6 C1, C2, C4, C5 H7 H7 C4, C5, C6 H8 C2 H9 H11 C10 H11 H9, H12 C9 H12 H11 C9, C11 H1′ C11 {#marinedrugs-11-03124-f006} Two separable signals that are characteristic for the anomer positions (C1′ and C1″) of distinguishable carbohydrates were observed ([Table 9](#marinedrugs-11-03124-t009){ref-type="table"}), demonstrating the presence of two carbohydrates. The downfield ^13^C chemical shifts of the carbohydrate anomer positions suggested hexose sugars bound to the palythine-threonine scaffold of the 612-Da MAA. As described in the section on the structure determination of the 508-Da MAA, the downfield ^13^C shifts of C7 and C11 were thought to be due to hexose binding via *O*-glycoside bonds in the 612-Da MAA. According to the HMBC spectrum, a correlation between the C11 carbon in the MAA backbone and the proton bound to the anomer carbon (C1′) of hexose was observed, suggesting the presence of hexose bound at the C11 position. The correlation between the C7 carbon and the anomer proton of the other hexose did not appear; however, geminal coupling at the C7 methylene protons due to the high rotational barrier was observed, suggesting the presence of a large functional group such as a hexose at the C7 position. The difference of ^13^C chemical shift at C7 position between the 612-Da MAA and palythine-threonine sulfate was only 3.5 ppm, which is smaller than that expected from the difference between the 508-Da MAA and porphyra-334. Because a sulfate group is a strong electron-withdrawing substituent, the ^1^H-chemical shift at the C7 position in palythine-threonine sulfate shifts downfield from that in palythine-threonine. These results are consistent with the MS analysis of the 612-Da MAA. After combining the data from the MS and NMR analysis, the predicted structure of the 612-Da MAA was generated and is shown in [Figure 6](#marinedrugs-11-03124-f006){ref-type="fig"}. 2.4. 450-Da MAA with an Absorption Maximum at 322 nm ---------------------------------------------------- Another MAA with an absorption maximum at 322 nm and a different retention time in the HPLC analysis elution profile ([Figure 1](#marinedrugs-11-03124-f001){ref-type="fig"}) was purified from the same *N. commune* colonies ([Table 1](#marinedrugs-11-03124-t001){ref-type="table"}, [Figure S1](#marinedrugs-11-03124-s001){ref-type="supplementary-material"}). The UV absorption spectrum of the purified MAA showed a single absorption peak at 322 nm ([Figure S1](#marinedrugs-11-03124-s001){ref-type="supplementary-material"}). A molecular mass of 450 Da was determined by MALDI-TOF MS and FAB MS. MALDI-TOF MS/MS analysis was performed on the parent molecular ion fragment with *m*/*z* 451 and the other main molecular ion fragments with *m*/*z* 407, 389 and 349. The fragmentation patterns were similar to those of the 612-Da MAA and suggested the presence of a hexose and the same palythine-threonine scaffold ([Table 11](#marinedrugs-11-03124-t011){ref-type="table"}). The ^1^H and ^13^C NMR analyses suggested that the structure consisted of a cyclohexenimine chromophore that was similar to that of the 612-Da MAA with one hexose sugar (data not shown). These data suggest that the 450-Da MAA is a hexose-bound palythine-threonine. Additionally, MALDI-TOF MS analysis suggested that the hexose was not bound to the threonine chain ([Table 11](#marinedrugs-11-03124-t011){ref-type="table"}). marinedrugs-11-03124-t011_Table 11 ###### Summary of MALDI-TOF MS analysis of the purified 450-Da MAA with absorption maximum at 322 nm. Mass of fragment Relative abundance Neutral loss Deleted fragment ---------------------- ------------------ -------------------- ----------------------------- ------------------ MS of purified MAA 451.22 82 407.23 100 44 CO~2~ 389.22 53 62 CO~2~ + H~2~O 349.18 29 102 C~4~H~6~O~3~ MS^2^ of *m*/*z* 451 **451.22** 289.21 100 162 Hexose 245.18 13 206 Hexose + CO~2~ 185.16 7 266 \- 91.09 8 360 \- MS^2^ of *m*/*z* 407 **407.23** 245.23 100 162 Hexose 230.20 7 177 Hexose + CH~3~ 227.21 8 180 Hexose + H~2~O 213.21 15 194 Hexose + H~2~O + CH~2~ 211.23 12 196 \- 209.18 7 198 Hexose + 2H~2~O 199.18 16 208 Hexose + C~2~H~6~O 197.21 16 210 \- 185.17 20 222 \- 177.17 11 230 Hexose + 3H~2~O + CH~2~ 169.17 5 238 Hexose + H~2~O + C~3~H~6~O 151.16 7 256 Hexose + 2H~2~O + C~3~H~6~O 91.09 8 316 \- MS^2^ of *m*/*z* 389 **389.22** 227.19 100 162 Hexose 209.18 16 180 Hexose + H~2~O 191.16 13 198 Hexose + 2H~2~O 181.16 14 208 Hexose + C~2~H~6~O MS^2^ of *m*/*z* 349 **349.09** 187.18 100 162 Hexose 169.16 8 180 Hexose + H~2~O 151.15 18 198 Hexose + 2H~2~O 2.5. Radical Scavenging Activity in Glycosylated MAAs ----------------------------------------------------- [Table 12](#marinedrugs-11-03124-t012){ref-type="table"} shows the radical scavenging activity found in the purified 508-Da and 612-Da MAAs from *N. commune*. Both the 508-Da MAA and the 612-Da MAA showed ABTS radical scavenging activity as determined by the decolorization of ABTS radicals. During the time course experiments, the decolorization of the ABTS radicals increased as the incubation time extended from 10 min to 2 h, which suggested that these MAAs were slow-acting radical scavengers. Trolox and ascorbic acid, which were used as standards, are known to be fast-acting scavengers that bring the decolorization reactions to completion within 10 min. Based on these results, the assay incubation time was fixed at 1 h as described by Matsui *et al.* \[[@B29-marinedrugs-11-03124]\]. When ESR was used to directly monitor the decrease in ABTS radical concentrations, the activity of the 612-Da MAA was comparable to that of Trolox and ascorbic acid, while the 508-Da MAA showed weak activity ([Table 12](#marinedrugs-11-03124-t012){ref-type="table"}). marinedrugs-11-03124-t012_Table 12 ###### Radical scavenging activity in MAAs ^a^. Assay Colorimetry ^b^ ESR ^c^ ---------------------------------- ----------------- --------- Ascorbic acid ^d^ 0.28 0.16 Trolox ^d^ 0.25 0.16 508-Da MAA with A~max~ at 334 nm 58 29 612-Da MAA with A~max~ at 322 nm 16 0.25 ^a^ Radical scavenging activity was measured with ABTS as the organic radical source; ^b^ Decolorization of ABTS was monitored with a spectrophotometer for 1 h; ^c^ ESR signals were monitored with a free radical monitor (JEOL JES-FR30EX); ^d^ Ascorbic acid and Trolox were used as standards; IC~50~ (50% inhibitory concentration) values are shown. The MAA-associated radical scavenging activity was examined in the *N. commune* water extract by HPLC fractionation ([Figure 7](#marinedrugs-11-03124-f007){ref-type="fig"}a). The fraction with MAA-associated radical scavenging activity accounted for approximately 45% of the total recovered activity ([Figure 7](#marinedrugs-11-03124-f007){ref-type="fig"}a). Interestingly, high radical scavenging activity, which was not MAA-associated, was detected and determined to account for approximately 40% of the total recovered activity ([Figure 7](#marinedrugs-11-03124-f007){ref-type="fig"}a). This result suggested that additional water-soluble antioxidants contributed highly to the total radical scavenging capacity of genotype D *N. commune*. This non-MAA radical scavenger will be characterized in future studies. {#marinedrugs-11-03124-f007} The recovered fractions with MAAs were further examined on an HPLC system with a gel filtration column ([Figure 7](#marinedrugs-11-03124-f007){ref-type="fig"}b). The 612-Da MAA accounted for approximately 60% of the total MAAs, and the radical scavenging activity eluted with the 612-Da MAA accounted for approximately 45% of the recovered activity ([Figure 7](#marinedrugs-11-03124-f007){ref-type="fig"}b). These results suggest that the 612-Da MAA is a major MAA in genotype D *N. commune* and is also a main component of the water-soluble radical scavengers, as it contributed approximately 20% of the total water-soluble radical scavenging activity in the water extract. 2.6. Specificity of the Genotypes and MAAs ------------------------------------------ *N. commune* is known to be genetically diverse, and four major genotypes of *N. commune* have been reported in Japan; however, the morphological features of the macroscopic colonies and microscopic trichomes are almost identical, and the genotypes are indistinguishable without a determination of molecular taxonomical markers \[[@B33-marinedrugs-11-03124]\]. During an investigation of MAA contents in field-isolated *N. commune* colonies, we observed different UV-absorption spectra in the water extracts from colonies at different sampling locations. We have reported two types of *N. commune*, those that specifically produce a 478-Da MAA with an absorption maximum at 335 nm or those that produce a 1050-Da MAA with double absorption maxima at 312 and 340 nm, although biological or physiological differences that would allow us to separate the different MAA producers could not be determined \[[@B29-marinedrugs-11-03124]\]. In addition to these two types of *N. commune* colonies, another type of *N. commune* colony, the water extract of which showed a characteristic UV-absorbing spectrum with an absorption maximum at 325 nm was found. According to the unique UV-absorbing spectrum and HPLC chromatogram of the water extract, we thought that this particular *N. commune* contained a structurally different MAA. Thus, in this study, the novel glycosylated MAAs were purified and characterized from organisms that did not produce either the 478-Da MAA or the 1050-Da MAA. Eight samples of *N. commune* from different sampling locations that had an absorption maximum at 325 nm in their water extracts were identified as genotype D, according to their 16S rRNA nucleotide sequences \[[@B46-marinedrugs-11-03124]\]; these data confirmed that genotype D is the 612-Da MAA producer*.* The glycosylated MAA patterns could be a feasible chemotaxonomic marker with which to characterize *N. commune*; it might be classified into three groups, namely, the 478-Da MAA producer, the 1050-Da MAA producer and the 612-Da MAA producer, as these groups are genetically different chemical races of *N. commune*. Further investigations into the genotypes and specificities of the types of the glycosylated MAAs are ongoing, and the results will be published in the future. 2.7. Glycosylated MAAs ---------------------- The chemical structures were determined for two glycosylated MAAs that were previously identified in *N. commune* \[[@B29-marinedrugs-11-03124]\]. The first was the 478-Da MAA with an absorption maximum at 335 nm that was identified as a pentose-bound porphyra-334 derivative. The second MAA, with a molecular mass of 1050 Da, had a characteristic UV absorption spectrum with an absorption peak at 312 nm, which was associated with a shoulder at 340 nm. The 1050-Da MAA is unique due to its high molecular mass and the presence of both 3-aminocyclohexen-1-one and 3-aminocyclohexenimine chromophores within a single molecule. These unique structural features with different chromophores are related to the expanded UV-absorbing window due to the double absorption maxima that cover both the UV-A and UV-B wavelengths. Interestingly, all of the MAAs characterized so far in *N. commune* have been glycosylated \[[@B29-marinedrugs-11-03124],[@B47-marinedrugs-11-03124]\]. This glycosylation is unique to the terrestrial cyanobacterium *N. commune*, although the function of this glycosylation has not yet been clarified. However, other glycosylated mycosporines, including mycosporine-glutaminol-glucoside and mycosporine-glutamicol-glucoside, have been reported in rock-inhabiting microcolonial fungi \[[@B48-marinedrugs-11-03124]\] and terrestrial cyanobacteria from rock surfaces \[[@B49-marinedrugs-11-03124]\]. The high concentrations of mycosporine-glutaminol-glucoside in desert rock-inhabiting fungi might be related to their survival potential in a terrestrial environment \[[@B50-marinedrugs-11-03124]\]. Hence, the glycosylated MAAs might be generally protective and allow adaptation to terrestrial environments in which the organisms are exposed to drastic changes in temperature and extreme desiccation, as well as direct solar radiation in the biologically harmful UV range. The details of these ecophysiological roles of the glycosylated MAAs remain to be demonstrated directly in future studies. 2.8. Porphyra-334 and Its Derivatives ------------------------------------- Porphyra-334, a UV-protective compound with absorption maximum at 334 nm and molecular mass of 346 Da, was first identified in the marine red alga *Porphyra tenera* \[[@B51-marinedrugs-11-03124]\] and was reported to be one of the most common MAAs in marine algal species \[[@B52-marinedrugs-11-03124]\]. To date, a limited number of studies have been published on the occurrence of porphyra-334 in cyanobacteria. Porphyra-334 has been reported in the marine cyanobacterial species of *Nodularia* \[[@B53-marinedrugs-11-03124]\], in the freshwater bloom-forming cyanobacterium *Microcystis aeruginosa* \[[@B54-marinedrugs-11-03124]\], in the aquatic cyanobacterium *Aphanizomenon flos-aquae* \[[@B35-marinedrugs-11-03124]\] and in the rice-field cyanobacterium *Anabaena doliolum* \[[@B55-marinedrugs-11-03124]\]. We have reported a 478-Da MAA pentose-bound porphyra-334 derivative in the terrestrial cyanobacterium *N. commune* \[[@B29-marinedrugs-11-03124]\], and in this study, we have identified a 508-Da MAA to be a hexose-bound porphyra-334 derivative. These studies have indicated the existence of glycosylated porphyra-334 in *N. commune*; however, the non-glycosylated form of porphyra-334 was not observed and the glycosylation of porphyra-334 in *N. commune* suggests a unique adaptation for terrestrial environments. The absorption maximum of the hexose-bound porphyra-334 derivative from *N. commune* reversibly shifted to 332 nm in a highly acidic solution, similar to porphyra-334 \[[@B56-marinedrugs-11-03124]\], which suggested that glycosylated porphyra-334 is a zwitterion in aqueous solution and thus is stable across a wide pH range. To confirm the stable conformation, a semi-empirical molecular simulation of the 508-Da MAA in neutral water at 25 °C was performed by molecular orbital package (MOPAC) 2011 with the PM6 Hamiltonian function ([Figure 8](#marinedrugs-11-03124-f008){ref-type="fig"}). A methyl group was substituted to mimic the hexose group in the 508-Da MAA and simplify the simulation (indicated by a white arrow in [Figure 8](#marinedrugs-11-03124-f008){ref-type="fig"}). The most stable form of the methyl-porphyra-334 derivative occupied a near-planar conformation, which allowed for a stable conformation with three hydrogen bonds in the molecule (depicted as green dashed bonds; heat of formation = −485.75 kcal/mol). In acidic conditions, the protonation on the carboxylate anion could cleave the one of the hydrogen bonds and thus would slightly prevent the resonance delocalization of π-electrons within the molecule. {#marinedrugs-11-03124-f008} The biosynthesis of porphyra-334 is thought to be genetically controlled, and porphyra-334 producers can be classified as a taxonomical group in cyanobacteria and also in marine algae. Biochemical and molecular biological studies of the biosynthesis of the porphyra-334 scaffold based 478-Da and 508-Da MAAs in *N. commune* will yield further understanding of the diversity and function of porphyra-334, which is produced and accumulated in taxonomically diverse marine, freshwater and terrestrial organisms \[[@B18-marinedrugs-11-03124],[@B19-marinedrugs-11-03124],[@B20-marinedrugs-11-03124]\]. 2.9. Palythine-Threonine and Its Drivatives ------------------------------------------- Palythine-threonine, an MAA with absorption maximum at 320 nm and molecular mass of 288 Da, was first identified in the corals *Pocillopora capitata*, *Stylophora pistillata* and *Pocillopora eydouxi* \[[@B42-marinedrugs-11-03124]\], however its sulfate ester had been reported in the reef-building coral *S. pistillata* \[[@B45-marinedrugs-11-03124]\]. To date, no studies have been published on the occurrence of palythine-threonine in cyanobacteria. In this study, we have identified the 612-Da MAA consisting of a cyclohexenimine chromophore conjugated with the substituent group of threonine ([Figure 6](#marinedrugs-11-03124-f006){ref-type="fig"}), and this scaffold was characterized as a palythine-threonine with a molecular mass of 288 Da. This is the first report of the unique glycosylated palythine-threonine, which was not listed for previously reported MAAs \[[@B18-marinedrugs-11-03124],[@B19-marinedrugs-11-03124],[@B22-marinedrugs-11-03124],[@B57-marinedrugs-11-03124],[@B58-marinedrugs-11-03124]\]. The absorption maximum of this molecule at 322 nm ([Figure 2](#marinedrugs-11-03124-f002){ref-type="fig"}b), the MS/MS results ([Table 7](#marinedrugs-11-03124-t007){ref-type="table"}) and the molecular mass of its scaffold were similar to those of the previously reported MAA palythine-threonine \[[@B42-marinedrugs-11-03124]\]. The identification of the glycosylated palythine-threonine from *N. commune* provides new insight into the molecular diversity of MAAs and the glycosylation of palythine-threonine in the terrestrial cyanobacterium suggests a unique adaptation for terrestrial environments that are harsh and drastically fluctuating in comparison to stable aquatic environments. However, the biosynthesis and glycosylation of palythine-threonine in *N. commune* remain to be elucidated in future studies. Biochemical and molecular biological studies of the biosynthesis of palythine-threonine in cyanobacteria will yield further understanding of this unique MAA originally identified in corals. 2.10. Antioxidative Role of MAAs -------------------------------- In terrestrial environments, *N. commune* colonies are subjected to desiccation and UV-irradiation. Protective compounds that are involved in reactive oxygen-scavenging mechanisms must have an important role in increasing cell tolerance to the oxidative stresses associated with desiccation and UV irradiation \[[@B12-marinedrugs-11-03124],[@B59-marinedrugs-11-03124]\]. MAAs have been suggested to be protective against UV-induced oxidative stress in algae \[[@B60-marinedrugs-11-03124]\] and in cyanobacteria \[[@B61-marinedrugs-11-03124]\]. In *N. commune*, the 478-Da and 1050-Da MAAs were reported to show radical scavenging activity *in vitro*, and the 1050-Da MAA, which has potent radical scavenging activity, contributes approximately 27% of the total water-soluble radical scavenging activity \[[@B29-marinedrugs-11-03124]\]. The glycosylated MAAs identified in this study also showed radical scavenging activities *in vitro* ([Table 12](#marinedrugs-11-03124-t012){ref-type="table"}). Although porphyra-334 was reported to have photoprotective but not antioxidant functions \[[@B62-marinedrugs-11-03124],[@B63-marinedrugs-11-03124],[@B64-marinedrugs-11-03124]\], both glycosylated derivatives of porphyra-334, the 508-Da MAA ([Table 12](#marinedrugs-11-03124-t012){ref-type="table"}) and the 478-Da MAA \[[@B29-marinedrugs-11-03124]\], had slow-acting radical scavenging activities *in vitro*. The glycosylation of porphyra-334 has been suggested to provide the antioxidant activities of these glycosylated porphyra-334 derivatives, although the molecular mechanism is unknown. The 612-Da MAA was the main water-soluble radical scavenger in the water extract, as it provided approximately 20% of the water-soluble radical scavenging activities in *N. commune*. The 612-Da and 1050-Da MAAs were the major MAAs in their respective colonies and demonstrated very similar water-soluble radical scavenging capacities \[[@B29-marinedrugs-11-03124]\]. The high *in vitro* radical scavenging activity of the 1050-Da MAA consisting of cyclohexenone chromophores is consistent with the antioxidative functions of the oxocarbonyl-type MAAs, such as mycosporine-glycine \[[@B64-marinedrugs-11-03124],[@B65-marinedrugs-11-03124],[@B66-marinedrugs-11-03124]\] and mycosporine-taurine \[[@B61-marinedrugs-11-03124]\]. Because the 612-Da MAA is not an oxocarbonyl-type MAA, the high radical scavenging activity may be due to its glycosylation, similar to the glycosylated porphyra-334 derivatives. These findings suggest the importance of these glycosylated MAAs to the oxidative stress response in *N. commune* and support the idea of multifunctional MAAs as UV-protectants and antioxidants. 2.11. Localization of Glycosylated MAAs in *N. commune* ------------------------------------------------------- The 36-kDa water stress protein (WspA) accounts for more than 70% of the extracellular matrix protein content of *N. commune* \[[@B16-marinedrugs-11-03124],[@B17-marinedrugs-11-03124],[@B67-marinedrugs-11-03124]\]. WspA binds to the scytonemin and MAA through non-covalent interactions and is assumed to be relevant to the structure and/or the function of the extracellular matrix \[[@B15-marinedrugs-11-03124],[@B16-marinedrugs-11-03124],[@B68-marinedrugs-11-03124]\]. UV irradiation stimulates the synthesis and secretion of WspA \[[@B16-marinedrugs-11-03124]\], as well as EPS production and scytonemin and MAA synthesis \[[@B69-marinedrugs-11-03124]\]. The xylose-containing MAA, which interacts with EPS via its sugar moieties, was reported to form multimeric complexes with WspA through strong ionic interactions in the absence of salt, which must be subject to the attenuation of UV-induced radiation damage during desiccation in *N. commune* \[[@B16-marinedrugs-11-03124],[@B68-marinedrugs-11-03124]\]. MAA glycosylation might be relevant to the architecture of the extracellular matrix in *N. commune*, but the details must be characterized in future studies. 2.12. Biosynthesis of Glycosylated MAAs in *N. commune* ------------------------------------------------------- More than 30 structurally distinct MAAs, including the mycosporine-derivatives, have been characterized to date \[[@B18-marinedrugs-11-03124],[@B19-marinedrugs-11-03124],[@B22-marinedrugs-11-03124],[@B57-marinedrugs-11-03124],[@B58-marinedrugs-11-03124]\]. These MAAs are formed from the precursor 4-deoxygadusol \[[@B70-marinedrugs-11-03124]\] and are conjugated with a nitrogen substituent (e.g., amino acids) \[[@B18-marinedrugs-11-03124],[@B19-marinedrugs-11-03124],[@B22-marinedrugs-11-03124],[@B71-marinedrugs-11-03124]\]. The gene product of Ava_3856 from *Anabaena variabilis* can convert 4-deoxygadusol and glycine into mycosporine-glycine \[[@B70-marinedrugs-11-03124]\], which is a metabolic precursor of the bi-substituted MAAs \[[@B70-marinedrugs-11-03124],[@B72-marinedrugs-11-03124]\]. Previous reports indicate that condensation of the mono-substituted mycosporine-glycine with an amino acid would be a common reaction in the generation of bi-substituted MAAs such as shinorine and porphyra-334 \[[@B70-marinedrugs-11-03124],[@B72-marinedrugs-11-03124]\]. Consistent with a previous report on the coral *P. capitata* which suggested that palythine-threonine is formed by decarboxylation of porphyra-334 followed by demethylation of mycosporine-methylamine-threonine \[[@B42-marinedrugs-11-03124]\], the simultaneous occurrence of palythine-threonine and porphyra-334 as scaffolds in *N. commune* suggests that porphyra-334 could be a metabolic precursor of palythine-threonine ([Figure 9](#marinedrugs-11-03124-f009){ref-type="fig"}). Supporting this idea, the glycosylated palythine-threonine accounted for 60% of the total MAAs and was the most abundant one in *N. commune*. According to the structures and fragmentation patterns of the glycosylated MAAs, a MAA biosynthetic pathway is predicted in *N. commune* ([Figure 9](#marinedrugs-11-03124-f009){ref-type="fig"}). In our predicted pathway, porphyra-334 could be generated via the addition of threonine to the core ring of mycosporine-glycine and the glycosylation of porphyra-334 could produce the 508-Da MAA. The 450-Da MAA could be formed via the elimination of a portion of glycine from C3 of the 508-Da MAA. The addition of another hexose to the 450-Da MAA could produce the 612-Da MAA ([Figure 9](#marinedrugs-11-03124-f009){ref-type="fig"}). This biosynthetic pathway for the production of the glycosylated palythine-threonine and porphyra-334 derivatives could occur particularly in genotype D of *N. commune*; however, the enzymes and genes involved in the biosynthesis of these glycosylated MAAs are unknown. Further molecular genetic studies are required to identify the enzymes and associated genes in the predicted biosynthetic pathway as well as the function of these glycosylated MAAs in adaptation to terrestrial environments. {#marinedrugs-11-03124-f009} 3. Experimental Section ======================= 3.1. Microorganisms ------------------- Colonies of field-grown *N. commune* were collected from the Kakuma Campus of Kanazawa University (N 36.544812, E 136.709635), Ishikawa, Japan. Wet colonies that were naturally swollen in response to rain were harvested, washed with tap water to remove soil, air-dried in the laboratory, and stored at room temperature until used. Water extracts from *N. commune* powders were examined spectrophotometrically to confirm the absorption maximum at 325 nm, and genotypes were characterized by the nucleotide sequence of the 16S rRNA gene as described previously \[[@B33-marinedrugs-11-03124]\]. 3.2. Purification of MAAs ------------------------- *N. commune* powder (30 g) that contained MAAs with absorption maxima at 334 or 322 nm was suspended in distilled water (2 L), and the MAAs were extracted by stirring at room temperature for 2 h. After centrifugation at 15,240× *g* for 20 min at 4 °C, the supernatant was vacuum-filtered with a Buchner sintered-glass filter funnel and then condensed to approximately 600 mL with a rotary evaporator under reduced pressure. Sufficient ethanol was added to the filtrate to yield a final concentration of 70% ethanol (v/v), and the mixture was kept at 4 °C for 1 h in the dark to precipitate the 70% ethanol-insoluble materials. After centrifugation at 15,240× *g* for 20 min at 4 °C, the supernatant was vacuum filtered with a Buchner sintered-glass filter funnel. The filtrate was evaporated and centrifuged at 21,500× *g* for 10 min at 4 °C. The supernatant was filtered through a 0.20-μm syringe filter (Minisart RC 15, Sartorius Stedim, Göettingen, Germany) and injected into an HPLC system with a Hitachi L-6200 pump that was equipped with a reverse phase column (IRICA C18, 20 × 250 mm). The mobile phase changed stepwise from distilled water during the first 40 min to 10% (v/v) methanol with 0.1% (v/v) acetic acid during the next 20 min to 100% methanol during the final 20 min. The flow rate was kept at 4 mL·min^−1^, and the A~330~ was monitored with a Hitachi L-4200 UV-VIS detector. The fractions with the MAAs were recovered separately, condensed with a lyophilizer and injected into an HPLC system equipped with a gel filtration column (TSKgel G2500PW, TOSOH, Tokyo, Japan). The mobile phase was water at a flow rate of 1 mL·min^−1^, and A~330~ was monitored with a Hitachi L-4200 UV-VIS detector. The MAA fractions were recovered, and the final MAA products were lyophilized. In each step, the total MAA concentration was estimated spectrophotometrically with an extinction coefficient of 120 L·g^−1^·cm^−1^ \[[@B34-marinedrugs-11-03124]\]. To determine the extinction coefficients of the purified MAAs with absorption maxima at 334 or 322 nm, diluted solutions were prepared in water, and the A~334~ or A~322~ was determined, respectively. The MAA dry weight in 1 mL of solution was measured after lyophilization. 3.3. MS Analysis ---------------- MALDI-TOF MS analysis was performed at the Division of Functional Genomics, Advanced Science Research Center, Kanazawa University, on a tandem mass spectrometer (4800 plus MALDI TOF/TOF™ Analyzer; Applied Biosystems, Foster City, CA, USA) with 2,5-dihydroxybenzoic acid (DHB) as a matrix. The secondary mass spectrum was recorded when applicable. FAB MS analysis to determine the accurate mass and predicted elemental composition was performed at the Research Institute for Instrumental Analysis in Kanazawa University on a mass spectrometer (JMS-SX102A, JEOL, Tokyo, Japan) with glycerol as a matrix. 3.4. Spectroscopic Methods -------------------------- UV-VIS spectra were recorded with a Hitachi U-2800 spectrophotometer. Fourier transformation infrared (FT-IR) spectra were recorded with a Nicolet NEXUS 470 FT-IR by the KBr disk method. NMR spectra in a D~2~O solvent were recorded with a JEOL ECS400 spectrometer at the Research Institute for Instrumental Analysis in Kanazawa University. 3-(Trimethylsilyl)-1-propanesulfonic acid-d6 sodium salt (TMP) was used as an internal NMR standard. In order to separate two overlapping signals assignable to imine carbons in the 612-Da MAA, high-resolution ^13^C NMR measurement was performed focusing on the X-range around 160 ppm and the NMR spectrum was recorded with a JEOL ECA-600 spectrometer. 3.5. Measurement of Trolox Equivalent Antioxidant Capacity (TEAC) ----------------------------------------------------------------- Radical scavenging activity was measured with 2,2′-azino-*bis*(3-ethylbenzothiazoline-6-sulfonic acid) (ABTS) as a substrate in a colorimetric assay \[[@B73-marinedrugs-11-03124]\]. Decolorization at A~734~ was monitored spectrophotometrically for 1 h. The electron spin resonance (ESR) signals of ABTS were recorded with a free radical monitor (JES-FR30EX, JEOL, Tokyo, Japan). Trolox (6-hydroxy-2,5,7,8-tetramethylchroman-2-carboxylic acid) and ascorbic acid were used as artificial and natural water-soluble antioxidant standards. 3.6. Chromatographic Separation and Detection of MAAs and Radical Scavenging Activities in *N. commun*e Water Extracts ---------------------------------------------------------------------------------------------------------------------- *N. commune* powder (1 g) was suspended in distilled water (100 mL) and extracted at room temperature by stirring for 4 h. After centrifugation at 21,500× *g* for 10 min at 4 °C, the supernatant was concentrated with a centrifugal concentrator (VC-360, TAITEC, Koshigaya, Japan) and filtered through a 0.20-μm syringe filter (Minisart RC 15, Sartorius Stedim). The concentrated water extract with 178 nmol TEAC was injected into an HPLC system with a Hitachi L-6200 pump and an L-4200 UV-VIS detector equipped with a reverse phase column (Wakosil 5C18, 4.6 mm × 250 mm; Wako, Osaka, Japan). The mobile phase changed stepwise from distilled water during the initial 14 min to 100% methanol during the next 36 min. The flow rate was kept at 0.5 mL·min^−1^, and a 1-mL fraction was collected every 2 min. The MAA-containing fraction with the highest radical scavenging activity was re-injected into another HPLC system with a Hitachi L-6200 pump and an L-4200 UV-VIS detector equipped with a TSKgel G2500PW column. The mobile phase was distilled water, and the flow rate was constant at 0.5 mL·min^−1^. The MAAs were detected at A~330~. Radical scavenging activity was measured by ABTS decolorization assay. 4. Conclusions ============== In this study, novel glycosylated MAAs with radical scavenging activities were identified in the terrestrial cyanobacterium *Nostoc commune*. The chemical structures of these glycosylated MAAs were different from those of previously reported MAAs \[[@B18-marinedrugs-11-03124],[@B19-marinedrugs-11-03124],[@B22-marinedrugs-11-03124],[@B29-marinedrugs-11-03124]\]. The major MAA, which had a molecular mass of 612 Da and an absorption maximum at 322 nm, was identified as a two-hexose-bound palythine-threonine derivative ([Figure 6](#marinedrugs-11-03124-f006){ref-type="fig"}). The second MAA, which had a molecular mass of 508 Da and an absorption maximum at 334 nm, was identified as a hexose-bound porphyra-334 derivative ([Figure 4](#marinedrugs-11-03124-f004){ref-type="fig"}). The MAA with a molecular mass of 450 Da and an absorption maximum at 322 nm was identified as a hexose-bound palythine-threonine derivative. The 450-Da and 612-Da MAAs were linked to one or two hexoses, and their scaffolds were related by a common palythine-threonine scaffold, suggesting that the 450-Da MAA is a likely intermediate from which the 612-Da MAA is produced via glycosylation ([Figure 9](#marinedrugs-11-03124-f009){ref-type="fig"}), although there are no data to rule out the possibility of a break-down product of the 612-Da MAA. *N. commune* excretes and accumulates MAAs in the extracellular matrix, and thus far, all characterized MAAs in *N. commune* have been glycosylated. Putatively, *N. commune* would be incapable of surviving terrestrial environments and sustaining viability in a long-term desiccated state without glycosylated MAAs. The glycosylation mechanism and physiological roles of these glycosylated MAAs with antioxidative activities in this unique anhydrobiotic organism, as well as their interactions with other components of the extracellular matrix architecture, remain to be elucidated in future studies. This work was supported by the Japan Society for the Promotion of Science (KAKENHI 22510227) and HABA Laboratories, Inc. We thank Kenji Takahashi for generous assistance with IR analysis and Chieri Tsuji for technical assistance. We are sincerely grateful to one of the anonymous reviewers for helpful comments concerning chemical structure analysis. The authors declare no conflict of interest. ###### Supporting Information (PDF, 2994 KB) ###### Click here for additional data file.

1. Background {#sec145611} ============= Dysglycemia is common in critically ill patients. Hyperglycemia is associated with adverse outcomes, including increased mortality, so insulin therapy and glucose control had been recommended to improve patient outcomes ([@A33849R1]-[@A33849R9]). However, intensive insulin therapy is associated with an increased risk of hypoglycemia, which is a possible predictor of morbidity and mortality in critical ill patients ([@A33849R10], [@A33849R11]) and a limiting factor for intensive insulin therapy. Over the past two decades, many studies have been conducted on intensive insulin therapy in critically ill patients, demonstrating controversial results. Spontaneous episodes of severe hypoglycemia are rare during the management of critical ill patients, and many factors contribute to its occurrence, such as underlying disease, malnutrition, infection, different glucose measurement methods, and chronic liver or kidney diseases ([@A33849R12]). Since the introduction of intensive insulin therapy strategies in intensive care units (ICUs), hypoglycemia has become a daily concern during the management of critically ill patients ([@A33849R13]). An absolute or relative insulin excess with inadequate energy intake together with limited exogenous glucose production and increased glucose utilization are the fundamental causes of hypoglycemia in ICUs. Several studies have shown that the number of hypoglycemic episodes may not be increased by intensive insulin therapy ([@A33849R14], [@A33849R15]). There is some evidence that hypoglycemic episodes are directly responsible for an increased mortality rate in critical ill patients ([@A33849R10], [@A33849R16]). However, another case control study in critically ill patients receiving insulin therapy showed that the occurrence of hypoglycemia was not associated with an increased risk of mortality ([@A33849R17], [@A33849R18]). As a result of several interventional clinical trials performed in critically ill patients, intensive monitoring and treatment of glucose levels in critically ill patients is emerging as a standard of care for these patients. So there are some concerns about which patients should be treated, the target level of hyperglycemia, and the incidence and risk factors for hypoglycemia, as hypoglycemia is a major limiting factor for the implementation of intensive insulin therapy and related mortality and was a reason for the early termination of the multicenter Glucontrol and VISEP trials ([@A33849R14]). The typical manifestations of hypoglycemia is not routinely seen in critical ill patients, mostly because of the masking of their clinical pictures and physiologically blunted response. Little is known about the pathophysiology and consequences of hypoglycemia in ICUs, in contrast with an extensive body of literature on the pathophysiology and consequences of hypoglycemia in diabetes mellitus. Therefore, recognizing the risk factors for hypoglycemia would help to identify patients who are at increased risk for hypoglycemia. 2. Objectives {#sec145612} ============= In this study, we aim to evaluate the incidence of hypoglycemia, its risk factors, and its relationship with mortality in critically ill patients. 3. Patients and Methods {#sec145617} ======================= 3.1. ICU Setting {#sec145613} ---------------- Five hundred critically ill patients from Feb 2011 to Sept 2013 were enrolled in this study. Two ICUs of Tabriz University of Medical Sciences (Shohada hospital and the ICU general of Imam Reza hospital) with mixed surgical and medical patients were included in this study. Cardiac surgeries were not performed at these two hospitals. All routine managements guided by protocols and a team of intensivists as directors of a multidisciplinary approach. The nurse patient ratio was 1:2 and full-time respiratory therapists accompanied the team. Inclusion criteria were all patients who admitted to theses ICUs. The study was approved by the ethics committee of Tabriz University of Medical Sciences. 3.2. Glucose Control Protocol {#sec145614} ----------------------------- A program of glycemic control with a target of 100 - 140 mg/dL was instituted. We used the threshold of 150 mg/dL for septic patients. This protocol was monitored by point of care devices for capillary blood measurement, and central laboratory values of venous blood were gathered to detect blood glucose at the specific time schedules and the accuracy of glucometers. Insulin therapy was performed by the frequent use of subcutaneous regular insulin as well as continuous intravenous regular insulin. We did not use any other type of insulin in our protocol. Blood glucose measurements were performed every hour, and if 4 consecutive measure were in the target range, the intervals were increased to 2 hours. If 3 consecutive measurements were in the target range, we performed measurements every 4 hours. If glucose was not in the target range, the measurement intervals were reduced to every hour. We detected hypoglycemia with blood sugar of less than 50 mg/dL, and with the detection of each hypoglycemia episode, blood glucose was measured every 30 minutes. All patients received energy from the enteral rout, except when they had contraindications, in which case we started parenteral nutrition. We calculated patients' daily caloric needs based on 25 kcal/kg. 3.3. Data Collection {#sec145615} -------------------- Patients' demographic characteristic were noted. Data collection consisted of acute physiologic and chronic health evaluation (APACHE) scores, previous history of diabetes, HbA1c, hypoglycemia episodes, diagnostic category (medical, surgical) sepsis, shock, liver or renal failure, previous history of renal replacement therapy, and drug history (beta blocker, pentamidine, aspirine, disopiramid, cotrimoxazol, coticsteroid, metfomin, glibenclamid). 3.4. Statistical Analysis {#sec145616} ------------------------- We used SPSS version 16 for statistical analysis. Data were presented as mean ± standard deviation. We used Student's t-test to compare two quantitative parameters. The chi-square test was used for the analysis of qualitative variables. To identify the predictors of hypoglycemia, we performed a stepwise logistic regression model for the mentioned variables. A p-value of less than 0.05 was considered significant. To assess the association between hypoglycemia and ICU mortality, we carried out a multivariate stepwise Cox proportional hazard regression model, adjusting for the abovementioned variables. Because the occurrence of hypoglycemia is also time-dependent, the time until the first occurrence of hypoglycemia was included. The results were expressed as adjusted hazard ratios (AHRs) and 95% CIs. Additionally, we carried out the same analyses stratified by selected variables. Continuous variables were categorized into two groups based on the median values. 4. Results {#sec145618} ========== A total of 500 critically ill patients who admitted to the ICUs of Shohada and Imam Reza hospitals (Tabriz University of Medical Sciences) from Feb 2011 to Sept 2013 were enrolled in this study. Patients' demographic characteristic are shown in [Table 1](#tbl37339){ref-type="table"}. As shown in [Table 1](#tbl37339){ref-type="table"}, their most common diagnoses were multiple traumas, brain tumors, and lower limb fractures. Almost half the patients received one of the following drugs: aspirin, corticosteroid, or beta blocker. All patients except 10 received enteral nutrition. Forty-six patients died during the study, of which 26 were men and 20 were female. Fifty patients experienced at least one episode of hypoglycemia, which was almost always on the third day. Of 15 expired patients who experienced one episode of hypoglycemia, the most common causes were multiple trauma and sepsis. The effects of age, sex, APACHE, sequential organ failure assessment (SOFA), admission blood sugar, blood urea nitrogen, acute kidney injury (AKI), and HbA1C on hypoglycemia occurrence were analyzed using the logistic regression method and after 8 times modeling SOFA, AKI, and hemoglobin A1c were recognized as effective (independent) variables on hypoglycemia. The analysis showed that increases in the SOFA number augmented the risk of hypoglycemia to 52% (P \< 0.001) ([Table 2](#tbl37340){ref-type="table"}). ###### Characteristics of Patients With and Without Hypoglycemia^[a](#fn40010){ref-type="table-fn"}^ Characteristic Hypoglycemia No Hypoglycemia P Value ----------------------------------------- --------------- ----------------- ---------- **No of patients** 50 (10) 450 (90) **Age** 72.14 ± 19.6 54.56 ± 17.8 \< 0.001 **APACHE** 28.50 ± 8.8 19.84 ± 6.9 \< 0.001 **SOFA** 13.72 ± 4.1 9.26 ± 3.5 \< 0.001 **Blood Sugar** 167.74 ± 38.5 131.27 ± 25,9 \< 0.001 **BUN** 30.94 ± 18.5 24.32 ± 15.2 \< 0.001 **Creatinine** 1.74 ± 0.8 1.91 ± 1.1 \< 0.001 **Hemoglobin** 10.9 ± 2.1 12.3 ± 3 \< 0.001 **Bilirubin** 0.69 ± 0.5 0.55 ± 0.3 0.110 **Diabetes Mellitus** 39 (78) 55 (12) \< 0.001 **Renal Failure** 2 (4) 2 (0.4) 0.007 **Liver Failure** 6 (12) 0 0.001 **HbA1C** 8.9 ± 4.5 5.8 ± 2.9 \< 0.001 **Acute Kidney Injury** 33 (66) 27 (6) \< 0.001 **Corticosteroid/Aspirin/Beta Blocker** 43 (86) 199 (44) \< 0.001 **Metformin/Glibenclamid** 40 (80) 52 (11) \< 0.001 **Mortality** 15 (30) 31 (7) \< 0.001 **Gender** 0.06 Male 24 276 Female 2 174 **Nutrition** \< 0.001 E 42 428 E+P 8 2 ^a^Values are expressed as No. (%) or mean ± SD. ###### Risk Factors Associated With Hypoglycemia Modeled Using Logistic Regression Analysis^[a](#fn40012){ref-type="table-fn"}^ B S.E. Wald df P Value Exp (B) 95.0% C.I. for EXP(B) -------------- ------- ------- -------- ---- --------- --------- ----------------------- ------- **SOFA** 2.20 0.06 12.497 1 0.000 1.52 0.988 1.053 **AKI** 2.9 0.05 10.887 1 0.000 10.3 3.162 33.63 **HbA1C** 1.21 0.019 8.70 1 0.031 3.2 1.93 5.88 **Constant** 3.264 0.746 12.866 1 0.00 0.002 NA NA Abbreviations: AKI, acute kidney injury; lower, lower bound for 95% C.I. for the OR; NA, not available; OR, odds ratio; upper, upper bound for 95% C.I. for the OR. ^a^The Hosmer and Lemeshow test showed an acceptable model fit (chi-square (4) = 6.52; P = 0.12). In addition, in patients with AKI, the risk of hypoglycemia is 10 times greater than in patients without AKI (RR: 10.3, CI: 3.16 - 33.6, P \< 0.001). HbA1c has a direct correlation with the occurrence of hypoglycemia, as with increasing HbA1c the risk of hypoglycemia is increased threefold ([Table 2](#tbl37340){ref-type="table"}). ICU admission blood sugar has a significant relationship with mortality (RR: 1.01, CI: 1.004 - 1.02, P = 0.006). Hypoglycemia increased the mortality rate 20%, but it was not significant (RR: 1.2, CI: 0.927 - 1.58, P = 0.221). HbA1c increased the mortality rate 20%, but it was not statistically significant (RR: 1.2, CI: 0.93 - 1.58, P = 0.161) ([Table 3](#tbl37341){ref-type="table"}). ###### The Effect Size of Hypoglycemia and Other Factors on Mortality Using Logistic Regression Analysis^[a](#fn40014){ref-type="table-fn"}^ B S.E. Wald df P Value Exp (B) 95.0% C.I. for EXP (B) ------------------ -------- ------- -------- ---- --------- --------- ------------------------ ------- **Age** 0.20 0.16 1.497 1 0.12 1.020 0.988 1.053 **Hypoglycemia** 1.9 0.45 1.887 1 0.22 1.2 0.927 1.58 **ICUad BS** 1.607 0.684 5.514 1 0.006 1.01 1.004 1.02 **HbA1C** 0.21 0.119 1.70 1 0.161 1.2 0.93 1.58 **APACHE** 0.129 0.080 2.553 1 0.11 1.137 0.971 1.332 **SOFA** 0.227 0.209 1.190 1 0.27 1.255 0.834 1.889 **Constant** -6.264 1.746 12.866 1 0.00 0.002 NA NA Abbreviations: ICUad BS, ICU admission blood sugar; lower, Lower bound for 95% C.I. for the OR; NA, not available; OR, odds ratio; upper, upper bound for 95% C.I. for the OR. ^a^The Hosmer and Lemeshow test showed an acceptable model fit (chi-square (7) = 5.22, P = 0.52). 5. Discussion {#sec145619} ============= Since the first leuven study intensive insulin therapy has led to improved inflammation and infection ([@A33849R19]) which is associated with improved patient survival ([@A33849R20]-[@A33849R24]). Although numerous studies have concluded that tight glycemic control can positively impact the clinical outcomes in ICU patients ([@A33849R25]), the apparent benefit of narrowly regulated tight glycemic control may come at the expense of an increased rate of hypoglycemia ([@A33849R26], [@A33849R27]). We determined that hypoglycemia is common in critically ill patients, and in this study hypoglycemic patients were significantly older and had higher HbA1c and APACHE scores, which was similar to the results of previous ([@A33849R28], [@A33849R29]) studies. Patients with type 1 diabetes and patients with longstanding type 2 diabetes may have an impaired counter-regulatory response. This may help to explain why patients who used insulin before ICU admittance were at a higher risk of developing hypoglycemia ([@A33849R30]). But in our study, logistic regression modeling after matching patients based on age, sex, and APACHE showed that only SOFA, AKI, and HbA1c are independent variables related to hypoglycemia. In Van den Bergh's study ([@A33849R19]), the most important risk factor for developing hypoglycemia was a discontinuation of nutrition or a reduction in glucose intake. ICU admission blood glucose, HbA1c, and hypoglycemia in our study increased the risk of death, but only ICU admission blood glucose is significantly related to increased mortality. As the lowering or discontinuation of nutrition without adjusting insulin therapy was associated with hypoglycemia, after any changes in the nutrition protocol, we decreased the blood sampling to 30 minutes to detect hypoglycemia more rapidly, which did not lead to higher mortality. The incidence of hypoglycemia was almost 10%, which differs from other studies ([@A33849R10], [@A33849R15], [@A33849R19]). The difference in hypoglycemia incidence might be related to the type of patients, the intensity of protocols, the sampling methods (arterial vs venous), and the measurement frequency. Our study results showed that there is no association between the first episode of hypoglycemia and mortality, but Egi et al. ([@A33849R31]) showed that an early onset of hypoglycemia following ICU admission is related to higher mortality levels. There are three explanations for the association between hypoglycemia and outcomes: first, the severity of hypoglycemia may be associated with the severity of the illness. Second, hypoglycemia may be a biomarker of imminent death. Third, hypoglycemia might have a deleterious biological effect on critically ill patients. This study showed that hypoglycemia did not have a significant effect on mortality, which was similar to the results of NICE SUGAR ([@A33849R32]) and Arabi ([@A33849R28]), but inconsistent with the results of Egi et al.'s study, which showed that the severity of hypoglycemia was significantly related to mortality ([@A33849R31]). Our study showed that there were no gender differences for hypoglycemia, which contradicts Merimee et al.'s finding of a lower counter-regulatory threshold in women compared to men ([@A33849R33]). Several studies have shown that severe hypoglycemia is independently associated with a higher risk of death with a greater duration of hospital stay ([@A33849R14], [@A33849R34], [@A33849R35]). These researchers have suggested that every hypoglycemic event may increase the mortality rate, which is in contrast with our results, possibly due to the delayed recognition and impaired counter-regulatory responses in critically ill patients, which leads to poor clinical outcomes. Our results suggest that the duration of hypoglycemia episodes may be short, largely due to intensive monitoring. A limitation of this study was that we did not examine the permanent neurologic dysfunction after hypoglycemia, secondary outcomes (i.e., renal replacement therapy, critical illness neuromuscular complications, nosocomial infections), or data on the blood glucose variability on outcomes. Moreover, we did not mention mechanical ventilation as a predisposing risk factor for hypoglycemia via a direct or indirect effect by sedation. Our results showed that the SOFA score, AKI, and HbA1c are the independent risk factors for the development of hypoglycemia and demonstrated that ICU admission blood glucose, HbA1c, and hypoglycemia increased the risk of death, but only ICU admission blood glucose is significantly related to increased mortality. Special thanks to the ICU staff of Shohada and Imam Reza hospitals, Sharokh Teshnehdel, Qorbanali Tarinezhad, and Mrs. Alikhani. **Authors' Contribution:**All authors have read and approved the manuscript. Ata Mahmoodpoor, Hadi Hamishehkar, Hassan Soleimanpour, Mohammadtagi Beigmohammadi, and Sarvin Sanaie performed the data collection, literature review, and drafting of the manuscript. Saeed Safari and Ahsan Rahimi undertook the major parts of the study design and performed the statistical analysis.

Introduction {#s1} ============ Coxsackievirus B3 (CVB3), a small RNA virus in the *picornaviridae* family, is an important human pathogen associated with various diseases, including myocarditis, aseptic meningitis, pancreatitis and possibly insulin-dependent diabetes. We and others have shown that CVB3 infection leads to activation of several intracellular signaling pathways [@pone.0002585-Opavsky1]--[@pone.0002585-Yuan1], and downregulation of host proteins likely through the ubiquitin/proteasome system (UPS) [@pone.0002585-Yuan1]--[@pone.0002585-Luo2]. It is well-established that the UPS is the major intracellular proteolytic system of all eukaryotic cells [@pone.0002585-Wolf1], [@pone.0002585-Schwartz1]. The ATP-dependent system begins with covalent attachment of ubiquitin to the ubiquitin-activating enzyme (E1). Then the ubiquitin is transferred to a ubiquitin-conjugating enzyme (E2). Finally, ubiquitin ligase (E3) transfers the ubiquitin to the substrate protein. After several cycles of ubiquitination, multiple ubiquitin molecules are attached to the substrate which is then quickly recognized and subsequently degraded by the 26S proteasome. Ubiquitin is recycled through the actions of deubiquitinating enzymes (DUBs) [@pone.0002585-Amerik1], [@pone.0002585-Nijman1]. There are at least two classes of deubiquitinating enzymes, the ubiquitin C-terminal hydrolases (UCHs) and ubiquitin-specific processing proteases family. In addition to the degradation of mutant, damaged and misfolded proteins, this system is responsible for the modulation of many regulatory proteins such as cyclins [@pone.0002585-Glotzer1], inhibitors of cyclin-dependent kinases (p21, p27) [@pone.0002585-Pagano1], tumor suppressors (p53) [@pone.0002585-Scheffner1], and inhibitor of NFκB (IκB) [@pone.0002585-Palombella1], which are essential for a variety of cellular functions, including cell-cycle regulation, apoptosis and host immune responses [@pone.0002585-RoosMattjus1]. Unlike polyubiquitination in the regulation of protein degradation, monoubiquitination of cellular proteins, such as histones, calmodulins, actin, proliferating cell nuclear antigen and receptor tyrosine kinases, plays more diversified roles involving in the regulation of chromatin remodeling, DNA repair, transcriptional regulation and endocytosis [@pone.0002585-Hicke1]. Since the first discovery that human papillomavirus protein E6 targets the cellular tumor suppressor protein p53 for the UPS-mediated degradation [@pone.0002585-Scheffner1], increasing studies, including those from our laboratory, have suggested that various viruses evolve different mechanisms to utilize or manipulate the host UPS for their own benefits [@pone.0002585-Luo2], [@pone.0002585-Bres1]--[@pone.0002585-Strack1]. We have previously shown that CVB3 infection results in downregulation of several host proteins [@pone.0002585-Yuan1], [@pone.0002585-Luo2], such as cell-cycle protein cyclin D1, tumor suppressor p53, and transcription activator β-catenin in infected HeLa cells. The downregulation of host proteins following CVB3 infection is most likely through the UPS. Specific inhibitors to 26S proteasome reverse the degradation of proteins in HeLa cells [@pone.0002585-Yuan1], [@pone.0002585-Luo2] and reduce CVB3 replication in murine cardiomyocytes [@pone.0002585-Luo3]. In this study, we investigated the possible underlying mechanisms by which the UPS regulates CVB3 replication. We demonstrated that protein ubiquitination was enhanced after coxsackievirus infection. We further showed that knockdown of ubiquitin expression by small-interfering RNA (siRNA) decreased CVB3 infection, likely through the downregulation of ubiquitination and subsequent alteration of protein function and/or degradation. In addition, we showed that inhibition of deubiquitinating enzyme increased the inhibitory effects of proteasome inhibitors on CVB3 replication. We also found that CVB3 RNA-dependent RNA polymerase 3D (3D^pol^) was modified by ubiquitination. Taken together, our study suggests an important role of ubiquitination in the regulation of coxsackieviral replication. Results {#s2} ======= Proteasome inhibition reduces CVB3 infection in HeLa cells {#s2a} ---------------------------------------------------------- To uncover the underlying mechanisms of the antiviral activities of proteasome inhibitors, we chose to use the well-characterized HeLa cells to further our study. We first examined the role of proteasome inhibition in CVB3 replication. As shown in [Fig. 1](#pone-0002585-g001){ref-type="fig"}, we found that proteasome inhibitor, MG132, significantly reduced CVB3 viral RNA synthesis ([Fig. 1A](#pone-0002585-g001){ref-type="fig"}). Both proteasome inhibitors used in the study, MG132 and lactacystin, decreased the synthesis of CVB3 capsid protein, VP1, in a dose-dependent manner ([Fig. 1B](#pone-0002585-g001){ref-type="fig"}). In addition, two inhibitors inhibited CVB3 viral titers by up to fifteen folds ([Fig. 1C](#pone-0002585-g001){ref-type="fig"}). Although MG132 and lactacystin significantly inhibited cellular 20S proteasome activities, we have previously demonstrated there was no apparent difference in proteasome activities between CVB3-infected and sham-infected HeLa cells [@pone.0002585-Luo2]. Together, these results suggest that efficient replication of CVB3 requires the intact UPS function rather than the core proteasome activity alone. {#pone-0002585-g001} We also performed cell viability assay and morphological examination to determine whether inhibiting viral replication by proteasome inhibitors is due to the toxicity. We found that there was no measurable cell death throughout the incubation period for all doses of proteasome inhibitors used in this study ([Fig. 1D](#pone-0002585-g001){ref-type="fig"}). On the contrary, virus-induced cell death was markedly inhibited after the treatment of proteasome inhibitors as a result of decreased viral replication ([Fig. 1D](#pone-0002585-g001){ref-type="fig"}). CVB3 infection promotes protein ubiquitination {#s2b} ---------------------------------------------- As alluded to earlier, two successive steps are involved in protein degradation: (1) covalent attachment of ubiquitins to the target protein substrate, and (2) degradation of the polyubiquitinated protein by the 26S proteasome with the release of ubiquitin for recycling. To dissect out the role of ubiquitination and degradation in CVB3 infection, we next decided to investigate the protein ubiquitination after CVB3 infection. As shown in [Fig. 2A](#pone-0002585-g002){ref-type="fig"}, protein ubiquitination was gradually increased along the time-course of CVB3 infection, which was accompanied by a decrease of free ubiquitin levels. Densitometric analysis further demonstrated that the increases in protein ubiquitination at 3 h, 5 h, and 7 h post-infection were statistically significant as compared to sham infection ([Fig. 2B](#pone-0002585-g002){ref-type="fig"}). We have previously demonstrated that 26S proteasome activities were unchanged during CVB3 infection [@pone.0002585-Luo2]. Thus, the finding of increased accumulation of ubiquitinated proteins is likely due to enhanced protein ubiquitination as opposed to reduced proteasome activity. Decreased levels of free ubiquitin could be a direct consequence of the increased protein ubiquitination. These results suggest that enhanced ubiquitin conjugation may be a prerequisite for efficient synthesis of CVB3 viral RNA and continuation of its lifecycle. {ref-type="sec"}". Protein levels of protein-ubiquitin conjugates (molecular weight starting from 82.2 kDa to approximately 230 kDa) and free ubiquitin were quantitated by densitometric analysis using NIH ImageJ program and normalized to the sham infection, which was arbitrarily set to a value of 1.0. Similar results were observed in two independent experiments. (B). Statistical analysis of protein-ubiquitin conjugates at 3 h, 5 h and 7 h after CVB3 infection. The data represent mean±SE of five different experiments. \* p\<0.05; ^&^ p\<0.01 as compared to protein expression in sham infection.](pone.0002585.g002){#pone-0002585-g002} Knockdown of ubiquitin by siRNA reduces CVB3 infection {#s2c} ------------------------------------------------------ In addition to blocking proteasome proteolytic activities, proteasome inhibitors are known to reduce free ubiquitin levels in treated cells [@pone.0002585-Mimnaugh1]. It has been suggested that proteasome inhibition negatively affects the budding of retroviruses through reducing free ubiquitin level and subsequently interfering with ubiquitination of viral Gag proteins [@pone.0002585-Patnaik1], [@pone.0002585-Schubert1], [@pone.0002585-Strack1]. Ubiquitin is generated in the cell by proteolysis of polyubiquitinated proteins or ubiquitin fused to carboxyl extension proteins (CEPs) [@pone.0002585-Jentsch1]. To investigate whether protein ubiquitination is beneficial to CVB3 replication in HeLa cells, we used the ubiquitin-specific siRNA to gene-silence the expression of human ubiquitin-CEP Uba80, which codes for ubiquitin fused to ribosomal protein S27a [@pone.0002585-Kirschner1]. As shown in [Fig. 3A](#pone-0002585-g003){ref-type="fig"}, both ubiquitin conjugates and free ubiquitin levels were markedly knocked down after the treatment of ubiquitin siRNA. We further showed that viral titers were significantly reduced in the ubiquitin siRNA-transfected cells as compared to scramble siRNA control ([Fig. 3B](#pone-0002585-g003){ref-type="fig"}), suggesting that protein ubiquitination is a critical process adopted by coxsackievirus for the successful completion of its lifecycle. {#pone-0002585-g003} DUB inhibition further enhances the inhibitory effects of proteasome inhibitor on CVB3 replication {#s2d} -------------------------------------------------------------------------------------------------- It has been demonstrated that protein ubiquitination can also be regulated by deubiquitinating enzymes that specifically cleave ubiquitin from ubiquitin-conjugated protein substrates [@pone.0002585-Amerik1], [@pone.0002585-Nijman1], [@pone.0002585-Wilkinson1]. To further explore the role of protein ubiquitination in viral replication, we examined the influence of DUB inhibition on viral protein expression. Two commercially available ubiquitin c-terminal hydrolase inhibitors, UCH L1 and UCH L3 inhibitors, were used for this study. As shown in [Fig. 4](#pone-0002585-g004){ref-type="fig"}, specific inhibition of UCH L1 or UCH L3 further reduced CVB3 protein expression and virus titers in proteasome inhibitor-treated cells, suggesting that these enzymes may be involved in the lifecycle of CVB3. Nevertheless, it was found that inhibition of the UCH L1 and L3 activities alone was not sufficient to block coxsackievirus replication since no significant changes in viral protein expression and CVB3 titers were observed in cells treated with two UCH inhibitors either separately or in combination ([Fig. 4](#pone-0002585-g004){ref-type="fig"}). As discussed earlier, stabilization of short-lived host proteins and prevention of protein ubiquitination by reducing recycled ubiquitin likely contribute to the inhibitory effect of proteasome inhibition on viral replication. Thus, it is speculated that DUB inhibition by UCHL1/L3 inhibitors alone, in the absence of apparent inhibition of protein degradation, is not sufficient enough to block viral replication. However, additional reduction of recycled free ubiquitin by DUB inhibition can further enhance the inhibitory effect of proteasome inhibitor. {#pone-0002585-g004} CVB3 RNA-dependent RNA polymerase 3D is ubiquitinated {#s2e} ----------------------------------------------------- Some virus RNA-dependent RNA polymerases including the sindbis virus and the turnip yellow mosaic virus RNA polymerases have been demonstrated to be phosphorylated and ubiquitinated [@pone.0002585-deGroot1]. Although the role of ubiquitination of these RNA polymerases in the regulation of virus replication remains to be determined, such observation raises the interesting possibility that the ubiquitin/proteasome system may regulate CVB3 replication through ubiquitinating viral polymerase 3D, which is essential for initiating viral RNA replication. To examine whether coxsackieviral proteins are subjected to ubiquitination during viral infection, we performed immunoprecipitation with anti-ubiquitin antibody, followed by immunoblots using antibodies against 3D^pol^ and viral capsid protein VP1, respectively. As shown in [Fig. 5](#pone-0002585-g005){ref-type="fig"}, immunoreactive bands of around 60 kDa were detected in CVB3-infected cells. Non-modified 3D^pol^ has a molecular weight of about 53 kDa, thus this observation suggests that 3D^pol^ likely undergoes post-translational modification by monoubiquitination. No protein ubiquitination was observed for VP1 (data not shown). Our results implicate that the ubiquitination process of CVB3 viral proteins might be required for successful replication of the virus. {#pone-0002585-g005} Effects of CVB3 infection on protein expression of several key enzymes involved in the process of ubiquitination and deubiquitination {#s2f} ------------------------------------------------------------------------------------------------------------------------------------- In trying to understand the mechanisms by which CVB3 manipulates the UPS, we examined the protein expression of several key enzymes involved in the process of protein ubiquitination and deubiquitination. We measured expression levels of ubiquitin-activating enzyme E1A/E1B, ubiquitin-conjugating enzyme Ubc H7, ubiquitin C-terminal hydrolase and two p53-related E3 ligases, human papillomavirus E6-associated protein and mouse double minute 2 homolog. However, no apparent changes were observed during the time-course of CVB3 infection (data not shown). These results indicate that the manipulation of the UPS by CVB3 is unlikely regulated by the above-examined ubiquitin-related key enzymes or molecules. Future studies will determine whether CVB3 infection targets on specific ubiquitin ligases or deubiquitinating enzymes. Discussion {#s3} ========== In the present study, we have provided further evidence that CVB3 manipulates the UPS for its infection. CVB3 infection results in increased protein polyubiquitination and a subsequent decrease in free ubiquitin levels. Knockdown of ubiquitin and ubiquitin-mediated protein modification and/or degradation by siRNA markedly reduces CVB3 replication in HeLa cells, further supporting the essential roles of the UPS in the replication of CVB3. It is increasingly apparent that viruses can evolve various strategies to utilize the host UPS for their own benefits. The UPS has been suggested to play a critical role in the different steps of viral lifecycle, including viral entry, viral replication, maturation, viral progeny release, and latent virus reactivation [@pone.0002585-Morita1]--[@pone.0002585-Furman1]. The mechanisms that the UPS regulates viral infection involve degrading intracellular proteins or excessive viral proteins that are against efficient viral replication and modulating viral protein function through ubiquitin-mediated modification or by directly encoding ubiquitin-related enzymes [@pone.0002585-Fan1]. The finding in this study that CVB3 infection stimulates protein ubiquitination without inhibition of the core 20S proteasome activity highlights the possibility that CVB3 manipulates the UPS to destabilize or modulate the host and viral proteins. Polyubiquitination and degradation of host antiviral proteins has been suggested to be a mechanism of HIV-1 replication [@pone.0002585-Yu1] . We have previously identified several proteins, such as cyclin D1, p53 and β-catenin, which are downregulated through the UPS after CVB3 infection [@pone.0002585-Yuan1], [@pone.0002585-Luo2]. Destabilization of these short-lived host proteins is likely required for CVB3 viral RNA and protein synthesis in its lifecycle. Moreover, it is speculated that nonstructural viral proteins of CVB3 could also be potential targets of the UPS for degradation. Previous studies on picornavirus have shown that several viral proteins, such as encephalomyocarditis virus (EMCV) 3C protease and hepatitis A virus (HAV) 3C protease, are ubiquitinated and present in low concentrations in infected cells [@pone.0002585-Losick1]--[@pone.0002585-Lawson2]. Several E3 ubiquitin ligases, such as human E3α ubiquitin ligase, have been shown to catalyze the ubiquitination of these viral proteins [@pone.0002585-Lawson1], [@pone.0002585-Lawson2]. Although the exact role of ubiquitination and subsequent degradation of nonstructural viral proteins of EMCV and HAV in infected cells remains elusive, such rapid turnover may be required for efficient viral RNA replication, viral protein synthesis and virus maturation. As alluded to earlier, DUBs are a large family of cysteine protease responsible for the removal of ubiquitin from substrate proteins [@pone.0002585-Love1]. It is estimated that the human genome encodes more than 100 DUBs. Although UCHL1 is identified as an important DUB, inhibition of UCHL1 alone has been shown to only partially block the activities of DUBs [@pone.0002585-Gong1]. Thus, the finding in this study that UCHL1/L3 inhibition is not as efficient in blocking viral replication as general inhibition of proteasome function or knockdown of ubiquitin is likely attributed to incomplete inhibition of DUBs by UCHL1/L3 inhibitors. In addition to protein degradation, ubiquitin-modification has been suggested to be involved in the regulation of protein function. It was reported that monoubiquitination of the Gag protein of retroviruses is required for virus budding [@pone.0002585-Patnaik1], [@pone.0002585-Schubert1], [@pone.0002585-Strack1]. Depletion of free ubiquitin by proteasome inhibitors prevents Gag ubiquitination, subsequently blocks virus progeny release/budding. In addition, ubiquitination of human immunodeficiency virus type 1 Tat protein and human T-cell leukemia virus type 1 Tax protein has been shown to modulate their transactivation activities [@pone.0002585-Bres1], [@pone.0002585-Peloponese1]. We speculate that monoubiquitination is also an important machinery for post-translational modification and activation of CVB3 viral proteins. In the current study, we have shown that CVB3 RNA-dependent RNA polymerase 3D is post-translationally modified by ubiquitination, suggesting a critical role of protein ubiquitination in the regulation of viral protein functions. Based on the results in the manuscript, in combination of our previous findings that CVB3 infection promotes host protein degradation, including cyclin D1, p53 and β-catenin, a model system on the role of the UPS in CVB3 replication is proposed in [Fig. 6](#pone-0002585-g006){ref-type="fig"}. Coxsackievirus infection facilitates host protein polyubiquitination, which subsequently increases intracellular protein degradation by the proteasome and/or viral protein modification, such as 3D^pol^, by monoubiquitination. Degradation of host antiviral proteins provides a favorable environment for virus to achieve successful replication. Knockdown of ubiquitin decreases host protein degradation and viral protein ubiquitination. Proteasome inhibition blocks host protein degradation and viral protein ubiquitination by reducing recycled ubiquitin. DUB inhibitors further decreases the viral replication when used together with proteasome inhibitors through the additional reduction of recycled free ubiquitin. {#pone-0002585-g006} In conclusion, we have demonstrated for the first time that CVB3 infection results in increased protein ubiquitination and consequent decreases in free ubiquitin levels. We further demonstrate that protein ubiquitination is required for the completion of viral lifecycle, likely through ubiquitin modification of viral polymerase. Materials and Methods {#s4} ===================== Cell culture, virus, and materials {#s4a} ---------------------------------- HeLa cells (American Type Culture Collection) were grown and maintained in complete medium \[Dulbecco\'s modified Eagle\'s media (DMEM) supplemented with 10% heat-inactivated newborn calf serum (NCS) (Invitrogen)\]. CVB3 (Kandolf strain) was propagated in HeLa cells and stored at −80°C. Virus titer was routinely determined by a plaque assay prior to infection as described below. The monoclonal anti-β-actin and anti-ubiquitin antibodies were purchased from Sigma-Aldrich. The monoclonal anti-VP1 antibody was obtained from DakoCytomation. The ubiquitin siRNA, scramble control siRNA, and horseradish peroxidase-conjugated secondary antibodies were obtained from Santa Cruz Biotechnology. The proteasome inhibitors, MG132 and lactacystin, the UCH L1 inhibitor (LDN-57444) and the UCH L3 inhibitor (4,5,6,7-Tetrachloroindan-1,3-dione), and the polyclonal anti-ubiquitin antibody were obtained from Calbiochem. The polyclonal anti-3D^pol^ antibody was a generous gift from Dr. Karin Klingel (University Hospital Tuebingen, Germany). Virus infection {#s4b} --------------- HeLa cells were grown in complete medium to 70--80% confluence, and then infected at a multiplicity of infection (MOI) of 10 with CVB3 or sham-infected with phosphate-buffered saline (PBS) for 1 h in serum-free DMEM. Cells were then washed with PBS and cultured in serum-free medium. For inhibition experiments, HeLa cells were infected with CVB3 for 1 h, washed with PBS, and then incubated with DMEM containing various concentrations of inhibitors. Immunoprecipitation and immunoblot analysis {#s4c} ------------------------------------------- Cell lysates were prepared using lysis buffer (50 mM pyrophosphate, 50 mM NaF, 50 mM NaCl, 5 mM EDTA, 5 mM EGTA, 100 µM Na~3~VO~4~, 10 mM HEPES (pH 7.4), 0.1% Triton X-100, and the protease inhibitor cocktail) as described previously [@pone.0002585-Luo1]. For immunoblot analysis, equal amounts of protein were subjected to sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and then transferred to nitrocellulose membranes (GE Healthcare). Membranes were blocked for 1 h with nonfat dry milk solution (5% in PBS) containing 0.1% Tween 20. Blots were then incubated for 1 h with the primary antibody followed by incubation for 1 h with the secondary antibody. Immunoreactive bands were visualized by enhanced chemiluminescence (GE Healthcare). When protein ubiquitination was examined, membrane was heat-activated by autoclaving at 121°C for 35 min prior to blocking with nonfat dry milk solution to enhance antigenic site recognition. For immunoprecipitation, cells were lysed using the above-described lysis buffer with freshly added 20 mM iodoacetamide. A total of 500 µg of cell lysates were incubated with a monoclonal anti-ubiquitin antibody (1∶100) at 4°C overnight, followed by 2 h incubation with protein G-agorose beads (Amersham). Immunocomplexes were washed five times with the lysis buffer containing 20 mM iodoacetamide, and then boiled for 5 min in the 2× non-reducing sample buffer which lacks both β-mercaptoethanol and DTT, but with addition of 20 mM iodoacetamide. After centrifugation, the precipitated proteins were separated by SDS-PAGE. Ubiquitin conjugates were analyzed by immunoblot using polyclonal anti-3D^pol^ antibody. Viral RNA *in situ* hybridization {#s4d} --------------------------------- HeLa cells were grown and maintained on two-chamber culture slides (Becton Dickinson Labware). Subconfluent cells were infected with either PBS or CVB3 (MOI = 10). Following 1 h of incubation at 37°C, cells were washed with PBS and replenished with complete medium in the absence and presence of MG132. HeLa cells were incubated for an additional 6 h. The culture slides were then washed gently with PBS, fixed with formalin buffer for 15 min, and then air-dried at room temperature. Culture slides were then subjected to *in situ* hybridization assays to detect the sense-strand of CVB3 genomic RNA as previously described [@pone.0002585-Luo3]. Plaque assay {#s4e} ------------ CVB3 titer in cell supernatant was determined on monolayers of HeLa cells by an agar overlay plaque assay in triplicate as described previously [@pone.0002585-Luo1]. Briefly, samples were serially diluted and overlaid on monolayer of HeLa cells. After 1 h incubation, medium was replaced with complete medium containing 0.75% agar. Cells were incubated for 72 h, then fixed with Carnoy\'s fixative (75% ethanol-25% acetic acid), and stained with 1% crystal violet. Plaques were counted and viral titer was calculated as plaque forming unit (PFU) per milliliter. Cell Viability Assay {#s4f} -------------------- MTS (3, 4-(5-dimethylthiazol-2-yl)-5-(3-carboxymethoxy phenyl)-2-(4-sulfophenyl)-2H-tetrazolium salt, Promega) assay was performed to determine cell viability as previously described [@pone.0002585-Yuan1]. Briefly, cells were incubated with MTS solution for 2 h prior to collection. Absorbance was measured at a wave length of 490 nm using an ELISA reader. Ubiquitin siRNA transfection {#s4g} ---------------------------- HeLa cells were grown to 50% confluency and then transiently transfected with ubiquitin-specific siRNA (200 nM) using oligofectamine according to the manufacturer\'s suggestion (Invitrogen). A scramble siRNA (200 nM) was used as a control. The silencing efficiency was detected by immunoblot analyses using the anti-ubiquitin antibody. After 24 h of transfection, cells were infected with CVB3 as indicated. Statistical analysis {#s4h} -------------------- Statistical analysis was performed using the paired *Student\'s t* test. A *p* value of less than or equal to 0.05 was considered statistically significant. **Competing Interests:**The authors have declared that no competing interests exist. **Funding:**This work was supported by grants from the Canadian Institutes of Health Research (CIHR) and the Heart and Stroke Foundation of Canada (HSFC) (to HL). XS is a recipient of the CIHR/HSFC IMPACT Post-Doctoral Fellowship, CIHR Michael Smith Fellowship and the HSFC Research Fellowship. GG is a recipient of a Doctoral Traineeship from the HSFC and the MSFHR. HL is a New Investigator of the CIHR/St. Paul\'s Hospital Foundation Award and a Scholar of the Michael Smith Foundation for Health Research (MSFHR). [^1]: Conceived and designed the experiments: XS HL. Performed the experiments: XS JZ GG JW YW. Analyzed the data: XS HL GG JW. Wrote the paper: XS HL.

Introduction {#s1} ============ Development of an invasive cancer is not only the result of genetic changes in tumor cells but also the result of the interplay between tumor and stromal cells [@pone.0052796-DeNardo1]. Tumors are infiltrated by a large number of immune cells that constitute the main cell population of tumor microenvironment, where they can account for up to 50% of the total tumor mass in invasive breast carcinomas. Historically, tumor-infiltrating leukocytes have been considered as an intrinsic defensive mechanism against developing tumors [@pone.0052796-Johnson1]--[@pone.0052796-Lin1]. However, increasing evidence indicates that leukocyte infiltration may favor tumor development by promoting angiogenesis, growth, and invasion [@pone.0052796-Coussens1]--[@pone.0052796-Daniel1]. This may be due to inflammatory cells that probably influence cancer promotion by secreting cytokines, growth factors, chemokines and proteases, which stimulate proliferation and invasiveness of cancer cells [@pone.0052796-Sica1]--[@pone.0052796-LeBitoux1]. Inflammatory cells have gained a renewed interest in breast cancer research due to our increased understanding of their role in tumor development, and also due to our increased ability to identify each cell type. Leukocyte infiltrate includes a variable representation of leukocytes, including macrophages, neutrophils, mast cells, and T and B-lymphocytes [@pone.0052796-Coussens1], [@pone.0052796-Lin2]. There are evidences indicating that different types of breast carcinomas may have different types of leukocyte infiltrate with distinct abilities to control tumor growth according to their tumor dissemination. Thus, whereas macrophages are known to have several pro-tumor functions and macrophage infiltration has also been associated with worse prognosis [@pone.0052796-Coussens1], [@pone.0052796-Bingle1]--[@pone.0052796-Lewis1], it has been reported that both T- and B-lymphocytes perform an important immunological response by inhibiting cancer development and progression [@pone.0052796-Alexe1]--[@pone.0052796-Mahmoud2]. Metastasis development is regulated not only by intrinsic genetic changes in malignant cells, but also by the tumor microenvironment. Matrix metalloproteases (MMPs) play an essential role in the degradation of the stromal connective tissue and basement membrane components, which are key elements in tumor invasion and metastasis. In fact, in the metastatic process across the axillary lymph node chain in breast cancer, MMP-1 expression by mononuclear inflammatory cells (MICs) from the sentinel lymph node (SLN) was significantly associated with metastatic spread to non-SLNs [@pone.0052796-Eiro1]. MMPs cleave proapoptotic factors and induce a more aggressive phenotype generating apoptotic resistant cells [@pone.0052796-Egeblad1], and also regulate cancer-related angiogenesis, both positively through their ability to mobilize or activate proangiogenic factors [@pone.0052796-StetlerStevenson1], or negatively through the generation of angiogenesis inhibitors, such as angiostatin and endostatin [@pone.0052796-Cornelius1]. The activity of MMPs is specifically inhibited by the so-called tissue inhibitors of metalloproteases (TIMPs). In previous reports we analyzed the expression of several MMPs and TIMPs (MMP-1, 2, 7, 9, 11, 13 and 14, and TIMP-1, 2 and 3), either at the invasive front or at the tumor center of breast carcinomas, in many of the women included in the present study [@pone.0052796-Vizoso1]--[@pone.0052796-DelCasar1]. Thus, we identified a phenotype of MICs characterized by the expression of specific MMPs and TIMPs (MMP-2, 9 11 and 14, and with TIMP-2) in the tumor center, associated with distant metastasis development [@pone.0052796-Vizoso1]--[@pone.0052796-Gonzalez1], suggesting that inflammatory cells at the invasive front can polarize their phenotype impacting on tumor progression [@pone.0052796-Gonzalez2]. These tumors also showed an up-regulation of inflammatory-related genes (IL-1, -5, -6 and -17, IFNβ and NFkB), which emphasize their importance in promoting disease metastasis and recurrence [@pone.0052796-Eiro2]. Considering that the invasive front is the area where some of the most important interactions between cancer cells and tumor supporting stroma take place [@pone.0052796-Giatromanolaki1], we investigate the relevance of the relative amount of macrophages (CD68), T-cells (CD3) and B-cells (CD20) in this tumor location from breast carcinomas. Also, we study their relationship with MMPs and TIMPs expression, either at the invasive front or at the tumor center. Thus, we found that a high CD68/(CD3+CD20) ratio (\>0.5) at the invasive front is associated with tumor aggressiveness and poor prognosis in patients. Materials and Methods {#s2} ===================== Ethics Statement {#s2a} ---------------- Women were treated according to the guidelines used in our Institution (Hospital de Jove). Written informed consent, approved by "Hospital de Jove Ethics and Investigation Committee", was obtained from all patients before the evaluation of tumor samples. The study adhered to National regulations and was approved by our Institution\'s Ethics and Investigation Committee. Patient selection, characteristics and tissue specimen handling {#s2b} --------------------------------------------------------------- This study comprises 102 women with a histological confirmed diagnosis of early invasive breast cancer and treated between 1990 and 2003. Many of these women have been included in previous studies of our group [@pone.0052796-Vizoso1]--[@pone.0052796-DelCasar1]. We selected women with the following inclusion criteria: invasive ductal carcinoma and a minimum of 5 years of follow-up for those women without tumor recurrence. The exclusion criteria were the following: metastatic disease at diagnosis, prior history of any kind of malignant tumor, bilateral breast cancer at diagnosis, have been treated with any type of neoadjuvant therapy, development of loco-regional recurrence during the follow-up period or development of a second primary cancer. From patients fulfilling these criteria, we randomly selected a sample size of 102 patients in accordance to 4 different groups stratified with regard to nodal status and to the development of metastatic disease, which were the key measure variables of the study. Thus, we included an important number of cases in both node-positive and node-negative patient subgroups in order to guarantee the statistical power of the survival analysis. Patient characteristics included in the two main groups, with or without distant metastases, are listed in [Table 1](#pone-0052796-t001){ref-type="table"}. Menopausal status was defined as "postmenopausal" if 1 year was elapsed since the last menstrual period. For reporting the Histological Grade we used the Nottingham combined histologic grade (Elston-Ellis modification of Scarff-Bloom-Richardson grading system) [@pone.0052796-Ellis1]. 10.1371/journal.pone.0052796.t001 ###### Basal characteristics of 102 patients with invasive ductal carcinoma of the breast. {#pone-0052796-t001-1} CHARACTERISTICS Without recurrence No. (%) With recurrence No. (%) ---------------------------------------- ---------------------------- ------------------------- **Total cases** 59 (100) 43 (100) **Menopausal status** Premenopausal 18 (30.5) 12 (27.9) Postmenopausal 41 (69.5) 31 (72.1) **Tumoral size** T1 31 (52.5) 19 (44.2) T2 28 (47.5) 24 (55.8) **Nodal status** N (−) 28 (47.5) 12 (27.9) N (+) 31 (52.5) 31 (72.1) **Histological grade** Well Dif. (I) 20 (33.9) 7 (16.3) Mod. Dif. (II) 31 (52.5) 16 (37.2) Poorly Dif. (III) 8 (13.6) 20 (46.5) **Nottingham prognostic index** \<3.4 25 (42.4) 8 (18.6) 3.4--5.4 25 (42.4) 22 (51.2) \>5.4 9 (15.3) 13 (30.2) **Estrogen Receptor** Negative 16 (27.1) 23 (53.5) Positive 31 (52.5) 18 (41.9) **Progesterone Receptor** Negative 20 (33.9) 27 (62.8) Positive 27 (45.8) 14 (32.6) **Adjuvant radiotherapy** No 44 (74.6) 21 (48.8) Yes 15 (25.4) 22 (51.2) **Adjuvant systemic therapy** Chemotherapy 18 (30.5) 18 (41.9) Tamoxifen 24 (40.7) 9 (20.9) Chemotherapy plus sequential Tamoxifen 10 (16.9) 7 (16.3) No treatment 7 (11.9) 9 (20.9) **HER2 Status** Negative 49 (83.1) 36 (83.7) Positive 8 (13.6) 7 (16.3) **Basal like phenotype** Non basal like 30 (50.8) 23 (53.5) Basal like 15 (25.4) 18 (41.9) The end-point of our study was distant metastatic relapse. The median follow-up period in patients without metastases was 85 months, and 52 months in patients with metastases. Tissue arrays and immunohistochemistry {#s2c} -------------------------------------- Breast carcinoma tissue samples were obtained at the time of surgery. Samples were removed from the tumors, avoiding grossly necrotic tissues, routinely fixed, paraffin-embedded and stored. Histopathological representative tumor areas of invasive front and tumor center were defined in hematoxylin and eosin-stained sections and marked on the slide. The invasive front was defined as the tumor advancing edge, which corresponds to a 2 mm margin surrounding the tumor and containing cancerous cells, and the tumor center was defined as the tumor area inside the invasive front. Tumor tissue microarray (TMA) blocks containing primary tumor samples were performed as described previously [@pone.0052796-Vizoso1]. We analyzed 2 cores of the invasive front and 2 cores of the tumor center in each case (double redundancy) as it has been demonstrated to correlate properly with conventional immunohistochemical staining methods [@pone.0052796-Vizoso1], [@pone.0052796-Gonzalez2]. Four composite high-density TMA blocks were performed, consecutively cut in 5 µm sections with a microtome (Leica Microsystems GmbH, Wetzlar, Germany) and transferred to adhesive-coated slides. One section from each TMA block was stained with hematoxylin and eosin, and these slides were then reviewed to confirm that the sample was representative of the invasive front and tumor center of the original tumor. Immunohistochemistry was performed using a TechMate TM50 autostainer (Dako, Glostrup, Denmark), where sections were incubated with the following antibodies (ready to use): CD3 (T-lymphocytes), CD20 (B-lymphocytes) and CD68 (macrophages) all purchased from Dako (Glostrup, Denmark). In previous reports from our group, we found a specific MICs phenotype characterized by high MMP-2, 9, 11, 14, and TIMP-2 expression, which correlated significantly with distant metastasis development [@pone.0052796-Vizoso1]--[@pone.0052796-DelCasar1]. Consequently, in the present study we performed a new staining set using antibodies against these specific proteins, in the tissue arrays from the invasive front and those from the tumor center. Antibodies for MMPs and TIMPs were purchased from Neomarker (Lab Vision Corporation, Fremont, CA, USA), and the dilution used was: 1/50 for MMP-2, -14 and TIMP-2; 1/100 for MMP-9; and 1/200 for MMP-11. To enhance antigen retrieval, tissue sections were treated in a PT-Link® (Dako) at 97°C for 20 min, in citrate buffer of pH 6.1 for MMP-14, in EDTA buffer of pH 9 for TIMP-2. Antibodies for MMP-2, -9 and -11 do not require antigen retrieval. The negative control was DakoCytomation mouse or rabbit serum diluted at the same concentration as the primary antibody. All the dilutions were made in Antibody Diluent, (Dako, Glostrup, Denmark) and incubated 30 min at room temperature. Endogenous peroxidase activity was blocked by incubating the slides in peroxidase-blocking solution (Dako) for 5 min. The EnVision Detection Kit (Dako) was used as the staining detection system. Sections were counterstained with hematoxylin, dehydrated with ethanol, and permanently coverslipped. Immunohistochemistry analysis {#s2d} ----------------------------- Five fields per core, corresponding to areas of higher immunostaining and without necrosis, were evaluated with a 400× power objective, counting CD3, CD20 and CD68 positive cells, in 1 mm^2^ final area, at the invasive front. If there was no tumor sample in a particular core, 10 fields were then evaluated in another one in order to obtain the same final area. We obtain a total score and this is the value of CD3, CD20 or CD68 for each tumor. For each MMP or TIMP antibody studied, we determined the immunoreactivity location, percentage of reactive area and intensity, at the invasive front and at the tumor center. An image analysis system composed of the Olympus BX51 microscope, digital camera system DP12 and soft analysis (analySIS®, Soft Imaging System, Münster, Germany) was used in the tumor sections (stained with antibodies and counterstained with hematoxylin), as described before [@pone.0052796-GonzalezReyes1]. To evaluate immunostaining intensity we used a numeric score ranging from 0 to 3, reflecting the intensity as follows: 0, no reactivity; 1, weak reactivity; 2, moderate reactivity; and 3, intense reactivity. Using an Excel spreadsheet, the mean score was obtained by multiplying the intensity score (I) by the percentage of reactivity area (PA) and the results were added together (total score: I×PA). This overall score was then averaged with the number of cores performed for each patient. If there was no tumor in a particular core, then no score was given. In addition, the mean score of two core biopsy samples was calculated for each tumor. This scoring evaluation was based on a global evaluation of staining areas corresponding to tumor cells as well as to stromal cells. Nevertheless, in the present work we also evaluated the immunohistochemical staining exclusively for mononuclear inflammatory cells (MICs). Statistical analysis {#s2e} -------------------- Differences in percentages were calculated with the chi-square test. Immunostaining score values for each protein were expressed as a median (range). Correlation between score values was calculated by using the Spearman correlation test. Comparison of immunostaining values between groups was made with the Mann-Whitney or Kruskall-Wallis tests. Statistical results were corrected applying Bonferroni\'s correction. For relapse-free survival analysis we used the Cox\'s univariate method. Cox\'s regression model was used to examine interactions between different prognostic factors in multivariate analysis. Only parameters that achieve statistical significance for distant relapse-free survival in the univariate analysis were included in the multivariate analysis. The PASW Statistics 18.0 software (SPSS Inc, Chicago, IL, USA) was used for all calculations. *p*\<0.05 was considered as significant. Results {#s3} ======= Immunostainings for CD3, CD20 and CD68 were performed in TMA blocks from invasive ductal carcinoma of the breast ([Figure 1](#pone-0052796-g001){ref-type="fig"}), showing a membranous staining for CD3 and CD20, whereas CD68 staining is found in the cytoplasm. Our results demonstrate a wide variability among tumors in the number of CD3^+^ T-cells (median: 214.00 (0--999), CD20^+^ B-cells (29.50 (0--1152) or CD68^+^ macrophages (141.00 (14--727), by 1 mm^2^ at the invasive front ([Figure 2](#pone-0052796-g002){ref-type="fig"}). We found direct correlations between the number of CD68^+^ macrophages and the number of CD3^+^ T-cells (r sub S = 0.57; p = 0.0001) or the number of CD20^+^ B-cells (r sub S = 0.51; p = 0.0001), and specially between the number of CD3^+^ T-cells and the number of CD20^+^ B-cells (r sub S = 0.71; p = 0.0001). {#pone-0052796-g001} {#pone-0052796-g002} We examined the possible relationship between the overall number of intratumoral MICs at the invasive front, or the relative ratio of these cells \[number of CD68^+^ macrophages/number of lymphocytes (number of CD3^+^ T-cells+number of CD20^+^ B-cells), further named as CD68/(CD3+CD20) ratio\], and the clinico-pathological characteristics of patients and tumors ([Table 2](#pone-0052796-t002){ref-type="table"}). Our results demonstrated a direct relationship between the number of CD3^+^ T-cells and premenopausal status (p = 0.009); whereas this same cell count was inversely associated with both ER^+^ and PgR^+^ status (p = 0.04 and p = 0.003, respectively). The number of CD20^+^ B-cells was directly associated with HER2^+^ status (p = 0.009). The number of CD68^+^ macrophages was inversely associated with PgR^+^ status and directly associated with HER2^+^ status (p = 0.027). However, our results showed no significant association between the CD68/(CD3+CD20) ratio and any clinico-pathological characteristics ([Table 2](#pone-0052796-t002){ref-type="table"}). 10.1371/journal.pone.0052796.t002 ###### Relationship between inflammatory cells count or ratio and clinico- pathological characteristics in 102 patients with invasive ductal carcinoma of the breast. {#pone-0052796-t002-2} CHARACTERISTICS No. CD3 CD20 CD68 CD68/(CD3+CD20) --------------------------------- ----- ---------------------- ---------------------- ---------------------- ----------------- **Total cases** 102 214 (0--999) 29 (0--1152) 141 (14--727) 0.5 (0--6.6) **Menopausal status** ***p*** ** = 0.009** Premenopausal 30 322 (9--999) 50 (0--1121) 158 (31--404) 0.3 (0.1--5.4) Postmenopausal 72 167 (0--987) 18 (0--1152) 128 (14--727) 0.5 (0--6.6) **Tumoral size** T1 50 207 (0--987) 22 (0--1152) 128 (15--727) 0.5 (0--5.4) T2 52 242 (12--999) 34 (0--1121) 154 (14--577) 0.6 (0.1--6.6) **Nodal status** N (−) 40 201 (9--987) 27 (0--1152) 136 (15--727) 0.5 (0.1--6.3) N (+) 62 250 (0--999) 32 (0--1121) 142 (14--577) 0.6 (0--6.6) **Histological grade** Well Dif. (I) 27 197 (9--987) 25 (0--1152) 140 (15--727) 0.5 (0.1--5.4) Mod. Dif. (II) 47 228 (12--999) 30 (0--1121) 142 (49--577) 0.6 (0.1--6.6) Poorly Dif. (III) 28 252 (0--542) 35 (0--156) 139 (14--416) 0.6 (0--5.4) **Nottingham prognostic index** \<3.4 33 172 (9--954) 7 (0--655) 122 (15--727) 0.5 (0.1--5.4) 3.4--5.4 47 267 (0--999) 41 (0--1152) 143 (21--577) 0.5 (0--6.6) \>5.4 22 250 (14--756) 40 (0--252) 170 (14--416) 0.5 (0.1--4.0) **Estrogen Receptor** ***p*** ** = 0.040** Negative 39 298 (0--987) 41 (0--1152) 181 (14--727) 0.6 (0--4) Positive 49 151 (9--895) 10 (0--1121) 122 (34--362) 0.6 (0.1--6.3) **Progesterone Receptor** ***p*** ** = 0.003** ***p*** ** = 0.002** Negative 47 267 (27--987) 40 (0--1152) 182 (14--727) 0.6 (0.1--6.1) Positive 41 144 (0--895) 10 (0--1121) 105 (35--314) 0.6 (0--6.3) **HER2 Status** ***p*** ** = 0.009** ***p*** ** = 0.027** Negative 85 209 (0--999) 16 (0--1152) 137 (14--727) 0.5 (0--6.6) Positive 15 359 (36--917) 101 (0--576) 186 (54--577) 0.6 (0.1--1.1) **Basal like phenotype** Non basal like 53 197 (9--895) 14 (0--1121) 137 (34--577) 0.7 (0.1--6.3) Basal like 33 251 (0--987) 40 (0--1152) 142 (14--727) 0.4 (0--4.0) Mann-Whithney or Kruskall-Wallis tests. We had previously identified a significant percentage of tumors with a MICs phenotype characterized by a molecular profile with specific MMPs and TIMPs increased expression, and associated with a high metastatic rate [@pone.0052796-Vizoso1]--[@pone.0052796-DelCasar1]. Thus, in the present work we determined the expression of these significant proteins (MMP-2, 9, 11, 14, and TIMP-2) in the tumor samples, and analyzed the possible relationship between the presence of different MICs phenotypes at the invasive front, and MMPs and TIMPs expressions by tumors both in the invasive front and in the tumor center. With regard to global expression (score values) of MMPs and TIMPs, our result showed a direct correlation between MMP-2 score values and CD3 (r = 0.21, p = 0.038), CD20 (r = 0.25, p = 0.011) or CD68 (r = 0.32, p = 0.001) counts at the invasive front; whereas MMP-9 score values correlated with CD68 count (r = 0.21, p = 0.041) in this same tumor location. On the other hand, TIMP-2 score values at the tumor center correlated inversely with CD3 (r = −0.23, p = 0.021) or with CD20 (r = −0.21, p = 0.036) count in the invasive front, but correlated directly with CD68/(CD3+CD20) ratio in this same tumor location (r = 0.24, p = 0.014). [Figure 3](#pone-0052796-g003){ref-type="fig"} shows examples of immunostaining for different MMPs and TIMPs, at tumor center and at the invasive front. We found several significant associations between the different MICs counts at the invasive front and the expression of MMPs and TIMPs by MICs from the invasive front or from the tumor center ([Table 3](#pone-0052796-t003){ref-type="table"}). Thus, high CD3, CD20 or CD68 counts were significantly associated with MMP-9 expression, at the invasive front; whereas high CD68 count was significantly associated with MMP-14 and TIMP2 in this same tumor location. Also, we found that high CD68 count and CD68/(CD3+CD20) ratio were associated with both MMP-11 and TIMP-2 expressions by MICs at the tumor center. In addition, it is interesting our finding indicating that if there is a high CD68/(CD3+CD20) ratio at the invasive front, most of MICs with a positive MMP-11 or TIMP-2 phenotype at the tumor center are macrophages ([Figure 3A and B](#pone-0052796-g003){ref-type="fig"}, respectively). In this figure, MMP-11 staining demonstrates that apart from tumor cells with large nucleus and an intense cytoplasmic staining, there are a small number of lymphocytes with rounded nucleus surrounded by a small positive cytoplasm, but the most abundant cells type in the tumor center are macrophages, which are the large, round cells that contain a central round nucleus and an abundant clear positive cytoplasm. {#pone-0052796-g003} 10.1371/journal.pone.0052796.t003 ###### Relationship between inflammatory cells count or ratio at invasive front and MMPs/TIMPs expression by mononuclear inflammatory cells at invasive front or tumor center. {#pone-0052796-t003-3} MICs at invasive front --------------------- ------------------------ ------------ ----------- ------------ ----------- ----- ----------- ------------ ----------- ----------- ------------ ----------- **CD3** 185 327 **0.006** 209 266.5 N.S 210 261 N.S 154 317.5 **0.001** (0--987) (76--999) (9--954) (0--999) (0--999) (12--542) (0--895) (27--999) **CD20** 15.5 85 **0.029** 25 38.5 N.S 23 55.5 N.S 8 74.5 **0.002** (0--1152) (0--576) (0--655) (0--1152) (0--1152) (0--302) (0--1121) (0--1152) **CD68** 130.5 184 **0.036** 128 166 N.S 132.5 186.5 **0.015** 118 178.5 **0.002** (14--727) (15--416) (21--727) (14--577) (15--727) (14--577) (21--416) (14--727) **CD68/(CD3+CD20)** 0.6 0.43 N.S 0.5 0.48 N.S 0.45 0.7 N.S 0.6 0.43 N.S (0--6.6) (0.1--0.9) (0.1--5.4) (0--6.6) (0--6.1) (0.2--6.6) (0--6.6) (0.1--6.1) MICs at tumor center --------------------- ---------------------- ------------ ----- ----------- ------------ ----------- ------------ ----------- ----- ----------- ------------ ----------- **CD3** 209 256 N.S 222.5 199 N.S 209 234 N.S 210 252.5 N.S. (0--999) (36--451) (0--999) (9--542) (12--999) (0--917) (0--999) (9--954) **CD20** 25 75 N.S 29.5 29 N.S 30 21 N.S 22.5 35.5 N.S. (0--1152) (0--211) (0--1152) (0--211) (0--1152) (0--576) (0--1152) (0--664) **CD68** 140 167 N.S 130.5 166.5 **0.041** 143 137 N.S 128 184.5 **0.025** (14--727) (62--577) (14--727) (40--577) (15.727) (14--577) (14--577) (40--727) **CD68/(CD3+CD20)** 0.5 0.6 N.S 0.44 0.9 **0.001** 0.43 0.6 N.S 0.4 0.7 **0.045** (0--6.6) (0.2--1.1) (0--6.61) (0.2--6.3) (0.1--6.6) (0--6.3) (0--5.37) (0.1--6.6) Mann-Whithney test. MICs: mononuclear inflammatory cells. Data are expressed as median (range). N.S: not significant. The possible influence of the number of the different inflammatory cell types on relapse-free survival was evaluated in all patients included in the present study. For this purpose, we took the corresponding median value of the total number of each cell type by 1 mm^2^ at the invasive front as cut-off point. Univariate analysis indicates that CD3, CD20, or CD68 count showed no significant associations with relapse-free survival ([Figure 4](#pone-0052796-g004){ref-type="fig"}). Nevertheless, our results showed that a high CD68/(CD3+CD20) ratio was significantly associated with a higher probability of shortened relapse-free survival (p = 0.002) ([Table 4](#pone-0052796-t004){ref-type="table"} and [Figure 4D](#pone-0052796-g004){ref-type="fig"}). Multivariate analysis according to Cox\'s model demonstrated that tumor stage (II: (relative risk (RR) (confidence interval (CI) = 1.8(0.7--4.5); III: 4.6(1.8--12.0); *p* = 0.003) and PgR status (positive: 0.4(0.2--0.8), *p* = 0.011) were significant and independently associated with distant relapse-free survival. Nevertheless, this same analysis also demonstrated that CD68/(CD3+CD20) ratio was significant and independently associated with distant relapse-free survival ([Table 4](#pone-0052796-t004){ref-type="table"}). {#pone-0052796-g004} 10.1371/journal.pone.0052796.t004 ###### Cox\'s univariate (HR) and multivariate (RR) analysis of the significant relationships between MMPs, TIMPs expression or CD68/(CD3+CD20) ratio at the tumor center or at the invasive front, and relapse-free survival. {#pone-0052796-t004-4} Tumor location Factor No. of patients Event frequency HR (95% CI) RR (95% CI) -------------------- ------------------------------ ----------------- ----------------- ----------------------------------------------------------- ----------------------------------------------------------- **TUMOR CENTER** **TIMP2** Score \< median vs. \>median 51/51 9/34 4.62 (2.21--9.65)[\*\*\*\*](#nt107){ref-type="table-fn"} 3.23 (1.51--6.92)[\*\*\*](#nt106){ref-type="table-fn"} MIC (−) vs. (+) 72/30 20/23 3.77 (2.06--6.89)[\*\*\*\*](#nt107){ref-type="table-fn"} 4.37 (2.31--8.25)[\*\*\*\*](#nt107){ref-type="table-fn"} **MMP11** MIC (−) vs. (+) 76/26 18/25 9.19 (4.73--17.85)[\*\*\*\*](#nt107){ref-type="table-fn"} 8.80 (4.40--17.61)[\*\*\*\*](#nt107){ref-type="table-fn"} **INVASIVE FRONT** **MMP9** Score \< median vs. \>median 50/49 16/25 2.03 (1.08--3.80)[\*](#nt104){ref-type="table-fn"} 2.22 (1.15--4.29)[\*](#nt104){ref-type="table-fn"} MMP14 MIC (−) vs. (+) 74/24 24/17 3.38 (1.81--6.31)[\*\*\*\*](#nt107){ref-type="table-fn"} 3.41 (1.75--6.63)[\*\*\*\*](#nt107){ref-type="table-fn"} **TIMP2** MIC (−) vs. (+) 49/50 15/26 1.89 (1.01--3.58)[\*](#nt104){ref-type="table-fn"} 2.51 (1.28--4.92)[\*\*](#nt105){ref-type="table-fn"} **CD68/(CD3+CD20) Ratio** 51/50 13/29 2.68 (1.39--5.17)[\*\*\*](#nt106){ref-type="table-fn"} 2.54 (1.23--5.24)[\*\*](#nt105){ref-type="table-fn"} Abbreviations: MIC: mononuclear inflammatory cells; HR: hazard ratio; RR: relative risk; CI: confidence interval. p\<0.05; p\<0.01; p\<0.005; p\<0.001. Discussion {#s4} ========== Inflammation is now considered a hallmark of cancer and can play a role in all aspects of tumor biology, including initiation, promotion, angiogenesis, and metastasis [@pone.0052796-Coussens1], [@pone.0052796-Gonzalez2], [@pone.0052796-Hanahan1]--[@pone.0052796-Grivennikov1]. It is known that the activation of oncogenes can trigger the production of inflammatory molecules and the recruitment of inflammatory cells. But the potential effects of the inflammatory cell infiltrate in breast cancer seem to be diverse and complex. Therefore, in this study we investigate the impact of different inflammatory cell types at the invasive front from breast carcinomas on distant metastasis development. We consider that this is of special interest because the invasive front is the area where some of the most important interactions between cancer cells and the tumor supporting stroma take place [@pone.0052796-Giatromanolaki1]. Our results showed a biological heterogeneity among breast tumors with regard to these cellular infiltrates at the invasive front. In addition, we found that a high CD68/(CD3+CD20) ratio at the invasive front is significant and independently associated with the occurrence of distant metastasis. There are data indicating that, depending on the cell type present and their functional profile, inflammatory cells can either suppress or promote tumor growth. We analyzed the expression profile of the individual inflammatory cell types, and our results are in accordance with other studies indicating that tumor-infiltrating lymphocytes correlate with hormone receptor-negative or HER2^+^ status, or with high grade/highly proliferative tumors, although we did not find correlation with favorable long-term prognosis [@pone.0052796-Alexe1]--[@pone.0052796-Mahmoud1]. In addition, it has been reported that activated B cells can mediate tumor regression by itself and confers host T-cell antitumor immunity. Likewise, it was suggested that effector B cells can serve as a useful adjunct in adoptive T-cell therapy [@pone.0052796-Li1]. Tumor-associated macrophages arise from circulating monocytes that migrate into tissues in response to chemical signals and differentiate into macrophages. In breast cancer, macrophages have been found to comprise up to 50% of the breast tumor mass [@pone.0052796-Kelly1]. Tumor-associated macrophages produce a variety of cytokines and chemokines, as well as growth factors for both epithelial and endothelial cells, which play a key role in tumor growth and metastasis [@pone.0052796-Coussens1], [@pone.0052796-Bingle1]--[@pone.0052796-Lewis1]. Our results are in accordance with previous studies reporting an association between macrophages density and PgR^−^ or HER-2^+^ status [@pone.0052796-Mahmoud3]. However, also in accordance with Mahmoud *et al.*, we found that overall macrophage numbers are not related to prognosis in breast cancer in a multivariate analysis [@pone.0052796-Mahmoud3]. This may be due the density of macrophages was correlated with higher tumor grade in the present study as well as in previous studies [@pone.0052796-Mahmoud3]--[@pone.0052796-Naukkarinen1]. Hence, multivariate analysis is thus essential when examining the relation between macrophage infiltration and survival. Nevertheless, this latter analysis led us to identify a high CD68/(CD3+CD20) ratio was a potent independent factor for predicting distant metastasis relapse-free survival in our patient population. Therefore, we describe here, for the first time, a study evaluating the relative amount of different MICs at the invasive front in breast carcinomas, using a new ratio that correlates with patient survival and could be useful in predicting patient outcome. We consider this is a relevant finding since the role of inflammatory cells in cancer seems to be complex, and this ratio can reflect a more objective result of the interactions between both anti-tumor and pro-tumor effects of the different inflammatory cells. The end point of the present study was the occurrence of distant metastasis, which is regulated not only by intrinsic genetic changes in malignant cells, but also by the microenvironment. MMPs play an essential role in tumor invasion and metastasis via degradation of the stromal connective tissue and basement membrane components, and are inhibited by TIMPs. In previous reports we identify a phenotype of MICs characterized by the expression of specific MMPs and TIMPs at the tumor center, and associated with distant metastasis development [@pone.0052796-Vizoso1]--[@pone.0052796-DelCasar1], which also showed an up-regulation of inflammatory-related genes [@pone.0052796-Eiro2]. According to this, in the present study we determined the expression of these significant proteins (MMP-2, 9, 11, 14, and TIMP-2) in those breast cancer samples and analyzed the possible relationship between the different inflammatory cells counts at the invasive front and the expression of MMPs and TIMPs, either at the invasive front or at the tumor center. Then, we found several associations between the inflammatory cell types and some of these factors. Nevertheless, the most relevant finding was the association between high CD68/(CD3+CD20) ratio and the expression of MMP-11 (stromalysin-3) or TIMP-2 by the MICs at the tumor center. This is a relevant finding considering that both MMP-11 and TIMP-2 are the two principal factors defining the pro-metastatic phenotype of MICs in our previous studies [@pone.0052796-Vizoso1]--[@pone.0052796-DelCasar1]. Therefore, these results may indicate that a high CD68/(CD3+CD20) ratio at the invasive front contributes to polarize macrophages to achieve a high metastatic phenotype at the tumor center. In addition, it is remarkable our finding indicating that if there is a high CD68/(CD3+CD20) ratio at the invasive front, most of MICs with a positive MMP-11 or TIMP-2 phenotype at the tumor center are macrophages. A limitation of the present study was the lack of a complete study of the count for the different MICs at the tumor center. It was due to the absence of enough tissue sample in many cases, because of their utilization in our previous expression studies on MMPs, TIMPs and other factors in breast carcinomas. Nevertheless, we observed that most of MICs in tumor center have macrophage-like morphology, indicating an important contribution of these stromal cells to tumor biology in this tumor location. In summary, our results contribute to characterize the inflammatory cell infiltrate in breast cancer, and their relationship with prognostic evaluation and MMPs/TIMPs expression. Further studies will be necessary to assess if this CD68/(CD3+CD20) ratio at the invasive front can contribute to identify patients with breast cancer candidates to different therapeutic strategies based on immuno-modulation. In fact, several strategies against tumor-associated macrophages have already been published [@pone.0052796-Griffiths1]--[@pone.0052796-Mukhtar1], and several reports indicate the effectiveness of activated B-cells in cellular immunotherapy of malignancies [@pone.0052796-Schultze1]--[@pone.0052796-DiLillo1]. Hence, to design breast tumor immunotherapy and vaccine strategies hereafter, it will be necessary to consider humoral immunity in addition to the cell mediated immunity, as a potential therapeutic tool. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: NE IP LOG ALM FJV. Performed the experiments: NE IP BFG SJ MLL JMDC LOG ALM FJV. Analyzed the data: NE IP BFG SJ MLL JMDC LOG ALM FJV. Contributed reagents/materials/analysis tools: NE IP BFG SJ MLL JMDC LOG ALM FJV. Wrote the paper: NE IP BFG SJ MLL JMDC LOG ALM FJV.

Data are available on the OSF: <https://osf.io/ptb92/?view_only=28a281db341f4ef8a5c8e2a84f3b5803>. Introduction {#sec001} ============ With several key scientific topics (e.g., climate change, safety of GMOs) \[[@pone.0207125.ref001]\] dividing the American public, and recent reports placing the United States at 27^th^ out of 64 countries in science performance \[[@pone.0207125.ref002]\], public support and trust in science are more important than ever. However, attitudes toward science and science literacy are usually viewed as an educational issue, a perspective that does not consider the role of religion as a potential detractor from the value of science. Although surveys show that the general public, and even scientists, may think that science and religion do not conflict \[[@pone.0207125.ref003], [@pone.0207125.ref004]\], some evidence to the contrary exists, both in research \[[@pone.0207125.ref005]\] and in reviews of the literature \[[@pone.0207125.ref006], [@pone.0207125.ref007]\]. Given the relative cultural importance of religion in the United States \[[@pone.0207125.ref008]\], and increasing skepticism towards science \[[@pone.0207125.ref007]\], this issue merits an empirical investigation. In the present studies, we used two large national datasets and two additional datasets collected by the authors to examine the relations between religiosity and both attitudes towards science and science literacy. The potential conflicts between science and religion can be viewed as limited or general. The limited version is that the conflict exists only for a few topics where science contradicts religious assertions, such as the origin of the Earth and the origin of humans \[[@pone.0207125.ref005],[@pone.0207125.ref009]\]. Additionally, some scientific research---such as stem cell research---may have moral and ethical implications to which religious people object \[[@pone.0207125.ref005], [@pone.0207125.ref010], [@pone.0207125.ref011]\]. Outside of these specific epistemological and moral contentions, according to the limited view, we would not expect religious teachings or believers to oppose science. However, a number of research findings suggest that the conflict between science and religion is more general, at least within the US. For example, greater religiosity is related to less favorable views toward scientific innovations and nanotechnology \[[@pone.0207125.ref011], [@pone.0207125.ref012]\], and religious people are less likely to choose careers in science \[[@pone.0207125.ref013], [@pone.0207125.ref014]\]. Survey data also shows that religious beliefs are negatively correlated with scientific consensus on a number of issues (e.g., vaccinations \[[@pone.0207125.ref015]\], climate change \[[@pone.0207125.ref016]\],) even when such issues do not directly challenge religious claims. While some researchers posit that objections towards nanotechnology and vaccinations may be driven by concerns about morality and scientists "playing God" \[[@pone.0207125.ref017], [@pone.0207125.ref018]\], the fact remains that these topics do not conflict with religious teachings per se---that is, we are not aware of religious texts which speak directly about nanotechnology, vaccinations, or climate change. Thus, religious people justify their opposition to some scientific concepts in terms of moral and religious beliefs in the same way that holders of a particular political ideology will oppose an idea in terms of economic or social justification. In contrast, Christian religious texts *do* discuss the creation of the world (versus evolution) and the sanctity of life (versus stem cell research). Viewed in this light, the conflict appears to address a more general epistemological dispute about whether science or religion is a better tool for understanding and explaining the world \[[@pone.0207125.ref019]\]. Thus, the general conflict hypothesis implies that religious people have more negative attitudes and possibly less trust towards science as a source of information. In the present studies, we tested whether this general conflict results in more negative attitudes towards science and in lower levels of scientific literacy. Three previous studies examined the relation between religion and scientific literacy, using data from the General Social Survey (GSS) in 2006 \[[@pone.0207125.ref005], [@pone.0207125.ref009]\] or in 2008 \[[@pone.0207125.ref010]\]. Two of the studies \[[@pone.0207125.ref005], [@pone.0207125.ref010]\] found no such relationship. The third study \[[@pone.0207125.ref009]\] found that religious fundamentalists and Catholics (but not mainline Protestants, Jews, and Muslims) possessed less science knowledge than those participants who stated they were not affiliated with any religious group. Some of the differences among these studies may be due to how science knowledge and religiosity were defined. For example, two of the studies \[[@pone.0207125.ref005], [@pone.0207125.ref009]\] included apparently "contested" items (questions about the Big Bang and continental drift) where a wrong answer may reflect a religious conviction rather than a lack of knowledge. The studies also differed in how religiosity was measured, such that the comparison between religious and non-religious people focused on religious affiliation and fundamentalism but did not involve the same groups. Religiosity is defined differently by many researchers. However, Allport's \[[@pone.0207125.ref020]\] construct of intrinsic religiosity, which refers to belief and identity, has been described as "...the *sine qua non* theory about what it means to be truly and appropriately religious" (\[[@pone.0207125.ref021]\], pg. 1). Indeed, widely used measures focus on religious beliefs, practice, and identity (e.g., \[[@pone.0207125.ref022], [@pone.0207125.ref023]\]), as those concepts are applicable to many different religious groups \[[@pone.0207125.ref023], [@pone.0207125.ref024]\]. Yet, the three studies just discussed \[[@pone.0207125.ref005], [@pone.0207125.ref009], [@pone.0207125.ref010]\] focused on specific combinations of religious affiliation and fundamentalism when investigating science attitudes and knowledge, which lends to a rather limited interpretation of the relation between religion and science. True, some religious groups do endorse teaching which conflict more directly with science (for example, literalist teachings about the creation of humans and the earth), but that is not the subject matter that we address. Instead, we aimed to investigate whether elements of religiosity which are common to most religions uniquely contribute to science rejection, not whether specific religious groups know more or less about science than others. Further, members of the same religious group can differ in the extent to which they endorse certain teachings, so a focus on affiliation ignores valuable individual differences. Finally, including affiliation and belief in an equation simultaneously creates problems of multicollinearity and makes interpretation of each variable difficult. For example, we assume that most people who identify themselves as Christian will score higher on a belief measure than people who identify themselves as atheists. For a minority, however, this difference will be reversed, raising the question of understanding where they are in terms of religiosity. In addition, any analysis of this minority will be hampered by low statistical power. We, therefore, proceed to examine only aspects of religiosity---religious beliefs and religious practice---which are common to most, if not all, religious groups \[[@pone.0207125.ref021]--[@pone.0207125.ref024]\]. Finally, whereas past research has found that the relation between religiosity and science depends on how the religiosity variable is measured or categorized \[[@pone.0207125.ref005], [@pone.0207125.ref009], [@pone.0207125.ref010]\], in the present research we demonstrate that this relation is robust and generalizable to broad, yet conceptually similar, measures of religiosity. Our present operationalization of religiosity also departs from past research due to differences between disciplines. Previous research on the relation between religiosity and science knowledge has so far come from sociologists; social psychologists have paid little attention to this area of research. Whereas social psychologists are interested in the individual differences in religious beliefs, sociologists may be more interested in how the religious traditions themselves influences attitudes towards science which results in more of a focus on religious affiliation. In our investigation, we sought to take a different approach simply to provide a different perspective on the relation between religion and science. It's also important to note that the studies reviewed above consider the relation between religion and science from a US perspective. While the above patterns may be stronger in the US---or may even only be true for the US---it is currently unknown whether these relations are true for other countries around the world. In evaluating the past and current research, these limitations should be kept in mind. The present studies {#sec002} ------------------- Taken as a whole, the relation between religiosity and science knowledge has not been empirically established; at the most, it has been examined in relation to specific religious affiliations and fundamentalism. We sought to address some limitations of the previous investigations. First, to increase statistical power, we used GSS data from all five separate surveys that included science knowledge questions (2006--2016). Second, in addition to the GSS \[[@pone.0207125.ref025]\] data (examined in Study 1), we also tested the relation between religiosity and science knowledge in another large, independently collected dataset (Study 2), the Longitudinal Study of American Youth (LSAY \[[@pone.0207125.ref026]\]) and two more original datasets (Study 3--4). Third, assuming that affiliation and fundamentalism are rather limited measures of religiosity, we used instead general religiosity scales that included a number of questions about both religious faith and practice. Thus, one important contribution of the present research is to operationalize religiosity differently from past research. Here we examined whether general measures of religiosity (focusing only on belief and practice) consistently predict negative science attitudes and lower levels of science literacy. Fourth, we examine the "limited" versus "general" conflict hypotheses by using two measures of science knowledge. First, in Studies 1 and 2, we present zero-order correlations between religiosity and a measure of science knowledge which includes questions about "contested" topics such as evolution---we expected religiosity to be negatively related to this measure. However, if the conflict between science and religion were more general and not limited to specific topics, we would expect the negative relation to persist after removing the contested items from the science knowledge scale. Therefore, following Johnson et al. \[[@pone.0207125.ref010]\], we used a science knowledge measure that excluded "contested" topics (e.g., evolution, the Big Bang, and continental drift), thus testing the general conflict hypothesis. Fifth, in Study 2, we examined whether the relation between participants' religiosity and attitudes towards science begins with the religiosity of their parents. A large research literature indicates that parents exert strong religious influence on their offspring's' beliefs and behavior in both childhood and adolescence \[[@pone.0207125.ref027]--[@pone.0207125.ref031]\] as well as in adulthood \[[@pone.0207125.ref032]--[@pone.0207125.ref034]\] (but see Hunsberger and Brown \[[@pone.0207125.ref035]\], who reported that parents' religious attendance and whether the children received religious education was related to their children's attendance when the children were in the high teens but not when they were in their early 30s). One problem with this evidence, however, is that in most cases it was based on children's reports of their own and of their parents' religiosity. Nevertheless, the findings suggest a possible path from parents' religiosity to children's religiosity. The present study examined whether this link exists, using measures of parents' religiosity that were obtained from the parents, and measures of children's religiosity and science knowledge that were obtained from the children, separately, over 20 years later. Finally, we also tested whether the predicted negative relation between religiosity and science knowledge is mediated by attitudes towards science. The rationale for such mediation is straightforward. To the extent that religious people view science as invalid, irrelevant, or morally suspect, they will be less interested in learning science, both formally and informally. Of course, this logic might also imply that another possible mediator is the amount of learning that a person has done. Education is linked to science knowledge because much of this knowledge is learned in schools. Religiosity is linked to education because it appears that those with lesser with lesser religious beliefs are more likely to seek out higher levels of education \[[@pone.0207125.ref036]\]. This logic suggests that the relation between science knowledge and religiosity will be mediated by education. On the other hand, religiosity may not necessarily translate into an overall anti-education bias; Catholic schools, for example, actually excel on the usual criterion of standardized tests \[[@pone.0207125.ref037]\]. Therefore, it is possible that the mediator is not general formal education, typically measured by the number of years spent on schooling, but the type of education one chooses to get. Indeed, we noted above that religious people are less likely to choose careers in science \[[@pone.0207125.ref013], [@pone.0207125.ref014]\]. Since the data available to us concerned *length* rather than *nature* of participants' education, we were not sure whether we would find support for mediation. We should note that our rationale implies a directional connection from higher religiosity to lesser science knowledge. We obviously could not obtain experimental evidence to test this prediction. However, whereas Study 1 is purely correlational, Study 2 is in part longitudinal, which is the preferable design to test directional hypotheses when experimental evidence cannot be collected. Study 1 {#sec003} ======= Method {#sec004} ------ ### Participants {#sec005} The National Opinion Research Center at the University of Chicago collects the GSS from a random sample of the English-speaking population in the US. We used the data from five different cohorts participating in the survey in 2006--2016. After deleting respondents with missing data on at least one variable, we examined responses of 4,627 participants (2,535 females). The participants had an average age of 47.78 (*SD* = 16.54) and an average education of 14.09 years (*SD* = 2.92). A majority of the participants (81%) were white, 10% were black, and 9% identified as other. A list of all the demographics that the analyses controlled for is presented in the Results section. ### Measuring religiosity {#sec006} We identified five items that measured religious beliefs and behaviors: confidence in the existence of God, how often respondent attends religious services, how often the respondent prays, how religious the respondent is, and strength of religious affiliation (questions about fundamentalism and the frequency of nonreligious activities at church were excluded). Beliefs about God were measured on a scale ranging from 1 (*don't believe*) to 6 (*know God exists without doubts*); frequency of attendance was indicated on a scale from 0 (*never*) to 8 (*more than once a week*); frequency of prayer was indicated on a scale from 1 (*never*) to 6 (*several times a day*). The remaining questions were answered on a scale from 1 (*not at all*) to 4 (*very much/very strong*). The five items were standardized and combined into a religiosity composite (α = .87). ### Measuring scientific knowledge {#sec007} We used the seven-item scientific knowledge measure that was previously used by Johnson et al., \[[@pone.0207125.ref010]\]. Participants received one point for each correct response and zero points for incorrect or "Don't know" responses. As participants answered question 7b (how long it takes for the earth to go around the sun) only if they correctly answered question 7a (whether the earth goes around the sun), they were given two points if they answered both questions correctly. As such, scores on the scale had a 0--8 range with higher scores indicating higher scientific knowledge. The questions are displayed in the upper half of [Table 1](#pone.0207125.t001){ref-type="table"}. However, we also report correlations for the total science knowledge measure, which includes 3 additional "contested" questions about evolution, the big bang, and continental drift. 10.1371/journal.pone.0207125.t001 ###### Science knowledge questions used in studies 1 and 2. Total correct responses were summed; 7b was displayed only if 7a was answered correctly. {#pone.0207125.t001g} -------------------------------------------------------------------------------------------------------------------- **Study 1: GSS** 1\. It is the father's gene that decides whether the baby is a boy or a girl.\ 2. Electrons are smaller than atoms.\ 3. The center of the earth is very hot.\ 4. Lasers work by focusing sound waves.\ 5. All radioactivity is man-made\ 6. Antibiotics kill viruses as well as bacteria\ 7a. Does the Earth go around the Sun, or does the Sun go around the Earth?\ 7b. How long does it take for the earth to go around the sun? **Study 2: LSAY** 1\. Antibiotics kill viruses as well as bacteria.\ 2. Nuclear plants destroy the ozone layer.\ 3. More than half of human genes are identical to those of mice.\ 4. Ordinary tomatoes, the ones we normally eat, do not have genes whereas genetically modified tomatoes do.\ 5. For the first time in recorded history, some species of plants and animals are dying out and becoming extinct.\ 6. Which travels faster: light or sound?\ 7a. Does the earth revolve around the sun, or does the sun revolve around the earth?\ 7b. How long does it take the earth to revolve around the sun? -------------------------------------------------------------------------------------------------------------------- ### Measuring science attitudes {#sec008} Except for the GSS administered in 2006, respondents in the remaining four cohorts answered questions assessing their interest in four relevant topics: medical discoveries, scientific discoveries, space exploration, and new technologies. For each topic, participants rated their interest on a scale ranging from 1 (*not at all interested*) to 3 (*very interested*). Participants also responded to four additional items regarding more general attitudes towards science on a scale from 1(*strongly disagree*) to 4(*strongly agree*): *Because of science and technology*, *there will be more opportunities for the next generation; science makes our way of life change too fast* (reverse coded); *even if it brings no immediate benefits*, *scientific research that advances the frontiers of knowledge is necessary and should be supported by the federal government;* and *would you say that*, *on balance*, *the benefits of scientific research have outweighed the harmful results*, *or have the harmful results of scientific research been greater than its benefits*? These eight ratings were standardized and averaged, yielding a science attitudes composite (α = .69) with higher scores representing more positive attitudes towards science. Results {#sec009} ------- ### Relation with science knowledge {#sec010} Inspection of zero-order correlations (see [S1 Table](#pone.0207125.s001){ref-type="supplementary-material"}) showed that religiosity was associated with lower levels of total (*r* = -.28, *p* \< .001) and non-contested (*r* = -.16, *p* \< .001) science knowledge and with more negative attitudes towards science (*r* = -.16, *p* \< .001). Because there are a number of variables that could account for this relationship, we repeated the analysis, controlling for relevant demographic variables. As our main interest was in science knowledge excluding contested items, we regressed the science knowledge scores of the non-contested items onto religiosity while controlling for age, education, sex (0 = male, 1 = female), household income, race (two dummy coded variables), father's education, mother's education, region of the country (south versus all others \[[@pone.0207125.ref007]\]), and cohort-years (four dummy coded variables). The model was significant and accounted for about 27% of variance, *R*^2^ = .27, *F* (15, 4611) = 113.15, *p* \< .001. Religiosity continued to negatively predict science knowledge while controlling for the demographic variables (see [Table 2](#pone.0207125.t002){ref-type="table"}), β = -.04, *F*(1,4611) = 10.25, *p* = .001. Thus, religiosity uniquely accounts for lower levels of non-contested science knowledge even after controlling for relevant demographic variables. 10.1371/journal.pone.0207125.t002 ###### Multiple regression coefficients and inferential statistics for all predictors and covariates in study 1. Standardized coefficients are displayed. {#pone.0207125.t002g} Predictor β *t* *p* *Partial r* ------------------------ ------ ------- --------- ------------- **Age** -.05 3.75 \< .001 -.06 **Education** .29 18.95 \< .001 .27 **Father's education** .10 5.57 \< .001 .08 **Mother's education** .06 3.27 \< .001 .05 **Sex** -.14 10.97 \< .001 -.16 **Income** .10 7.31 \< .001 .11 **2008** .02 1.53 .540 .02 **2010** .02 1.16 .582 .02 **2012** .01 .50 .449 .01 **2014** .01 .59 .737 .01 **2016** -.01 .59 .139 -.01 **White** .10 5.58 \< .001 .08 **Black** -.05 2.65 .004 -.04 **Region** -.04 3.36 .001 -.05 **Religiosity scale** -.04 3.19 .001 -.05 ### The role of spirituality {#sec011} Researchers often distinguish between religiosity and spirituality (e.g., \[[@pone.0207125.ref038]\]), and this distinction is relevant to the issue of science literacy. Citing unpublished research by A. Willard and A. Norenzayan, Norenzayan \[[@pone.0207125.ref039]\] noted that spirituality rejects religious dogma and taps into psychological experiences "...that are stripped away from the cultural baggage inherited from ...traditions" (p. 479). If spirituality is less tied to religious concepts (e.g., God, scriptures, creationism) that induce opposition to science, it may not share with religion its negative relation to science knowledge. Indeed, repeating the same multiple regression analysis as above, but including an item measuring spirituality in the religiosity composite (α = .87), yielded a somewhat weaker negative relation between science knowledge and religiosity, β = -.03 (vs. β = -.04), F(1, 4611) = 3.06, *p* = .080, partial *r* = -.03. ### Mediation analysis {#sec012} To examine whether the relation between religiosity and science knowledge was accounted for by negative attitudes towards science, we conducted a mediation analysis with 5,000 bootstrapped samples, again controlling for the same demographic variables ([Fig 1](#pone.0207125.g001){ref-type="fig"}). Because four of the questions regarding science attitudes were not asked during the 2006 GSS, we used only the 2008--2016 cohorts in this analysis (*N* = 3,479). As shown in [Fig 1](#pone.0207125.g001){ref-type="fig"}, the indirect effect from religiosity to science attitudes, and then to science knowledge was significant, β = -.02, 95% CI: -.0239, -.0117, accounting for 39% of the total relation between religiosity and science knowledge; including science attitudes in the model reduced the relationship between religiosity and science knowledge from β = -.05 to β = -.03, though it remained significant. Thus, controlling for relevant demographic variables, greater religiosity was associated with lower science knowledge, a relation that was partially accounted for by less positive attitudes towards science among religious people. {#pone.0207125.g001} It's also important to note that, while these models are consistent with our hypothesis that religiosity is linked to lower science knowledge because of less positive attitudes towards science, cross-sectional mediation models do not indicate causation. Thus, another interpretation of this model is simply that both religiosity and science attitudes are uniquely related to science knowledge. We also tested alternative models of these relations and describe these in the supplementary file (see Figure A in [S1 File](#pone.0207125.s003){ref-type="supplementary-material"}). ### The role of education {#sec013} In the above mediation analysis, education served as one of the covariates. When we tested education as a mediator alongside science attitudes, the indirect effect of education was also significant (β = -.01, 95% CI: -.0162, -.0018); the indirect effect through science attitudes was still significant (β = -.02, 95% CI: -.0250, -.0130), reducing the relation between religiosity and science knowledge from β = -.05, *p* \< .001 to β = -.03, *p* = .050. Discussion {#sec014} ---------- Results from Study 1 indicate that higher religiosity is associated with lower levels of science literacy even after excluding contested items (e.g. evolution). Further, religiosity is associated with more negative attitudes towards science---an effect that partially accounts for the relation between religion and science literacy. These results are in line with the general conflict hypothesis. Education accounted for a small portion of the relation between religiosity and science knowledge, suggesting that religiosity is associated with lower levels of science literacy resulting from less formal education. In Study 2, we further examined these relations using an independently collected dataset to replicate the effects observed here and to expand on the relation between religiosity, attitudes towards science, and science literacy. Study 2 {#sec015} ======= The goal of study 2 was twofold. First, we examined whether a general measure of religiosity was again related to science knowledge, as well as to lower trust in scientific sources of information, and whether these relations are mediated by science attitudes. Second, we examined whether the relation between a person's religiosity and science attitudes/knowledge (and presumably lower trust in scientific information) begins earlier, with the religiosity of that person's parents. Method {#sec016} ------ ### Participants {#sec017} The data were obtained from the Longitudinal Study of American Youth (LSAY \[[@pone.0207125.ref026]\]). This study collected two waves of data from over 3,000 American youth. The first wave, which began in 1987, collected data from two cohorts. Survey of Cohort 1 began with almost 3,000 10^th^ grade public school students and followed them for an initial period of seven years, ending about four years after high school graduation. Survey of Cohort 2 began with almost 3,000 public school students in 7^th^ grade and followed them for seven years, until about one year after high school graduation. Data collection for both cohorts ended in 1994. The second wave of data collection was carried out during 2007--2011 and included questions concerning science knowledge and attitudes towards science, as well as questions assessing religious beliefs and behaviors which were similar to the question used in the GSS. After deleting respondents with missing data on exogenous variables, we examined responses of the 2,839 participants (1,541 females) who responded to the relevant ballot of questions used in our analyses. In 2008, participants' ages ranged approximately from 33 to 37. Participants were 86% white, 7% African-American, and 7% Hispanic. In 2008, 3% had less than a high school education, 42% had achieved a high school education, 8% had earned a 2-year degree, 30% had earned a 4-year degree, and the remaining 17% had earned an advanced degree. A list of all the demographics controlled for in the analysis is presented in the Results section. ### Measuring scientific knowledge {#sec018} The 2008 survey included 15 science knowledge questions. Seven questions addressed the contested topics of evolution, the Big Bang, continental drift, and climate change. Accordingly, only responses to the remaining eight questions were examined in the main analysis (see bottom half of [Table 1](#pone.0207125.t001){ref-type="table"}). As in Study 1, participants received one point for each correct response and zero point for incorrect or "Don't know" responses; participants received an extra point for answering correctly question 7b (which was contingent on answering correctly question 7a). Overall scores had a 0--8 range with higher scores indicating higher science knowledge. As in Study 1, we also report the zero-order correlations between the total 15 item general science knowledge score in the supplementary materials (see [S2 Table](#pone.0207125.s002){ref-type="supplementary-material"}). ### Measuring religiosity {#sec019} Participants rated on 1(*strongly disagree*) to 4 (*strongly agree*) scales whether they believed that the bible is the literal word of God, and whether a personal God exists; they were also asked how often they attended religious services and activities on weekly basis. Responses to these three questions were standardized and combined into a religiosity composite (α = .69). ### Measuring science attitudes {#sec020} In the 2008 follow-up, participants completed the same measure of attitudes toward science that was administered in the GSS and used in Study 1. That is, participants rated their interest in science discoveries, science and technology, new medical discoveries, and space exploration on a scale from 1(*not at all interested*) to 3(*very interested*). In addition, participants also rated their attitudes toward science on eight more statements on a scale from 1(*strongly disagree*) to 4(*strongly agree*). Four of the items were identical to those used in Study 1: *Because of science and technology*, *there will be more opportunities for the next generation*, *science makes our way of life change too fast* (reverse coded), *even if it brings no immediate benefits*, *scientific research that advances the frontiers of knowledge is necessary and should be supported by the federal government*, *how would you assess the balance between the beneficial and harmful results of scientific research* (answered on a 5-point scale)? An additional four items were new, but similar to the previous four: *Science and technology are making our lives healthier and easier*, *most scientists want to work on things that make life better for the average person*, *would you say the world is better off or worse off because of science*?, *It is not important to know about science in everyday life* (reverse-coded). We standardized all 12 items and combined them into a composite score representing science attitudes (α = .78) with higher scores representing more positive attitudes towards science. This is the scale that was used in subsequent analyses. However, using only the same eight items as in Study 1 produced the same results, and we discuss this analysis in the section "Revisiting Studies 1 and 2," below. ### Parents' religiosity {#sec021} At various times, participants' parents (mothers and fathers) were asked four questions assessing their religious beliefs and the extent to which religion played a role in their children's upbringing during the years 1987--1994. The questions were (1) *how many times have you attended a religious service with your child during this last year*? Parents responded with a number; (2) *how often do you talk to your child about religious values and beliefs*? Parents responded on a scale from 1(*not at all*) to 3(*often*). For those in Cohort 1, these two questions were asked during the 11th and 12^th^ grade years. For those in Cohort 2, these questions were asked each year from 7^th^ to 11^th^ grade. Parents were also asked (3) if they *believe that a personal God exists* and (4) whether they *believe the bible is the literal word of God*. They responded to both questions on a scale from 1(*strongly disagree*) to 4 (*strongly agree*). For Cohort 1, these questions were asked during the 10^th^ and 11^th^ grades, and for Cohort 2, these questions were asked during the 7^th^ and 8^th^ grades. Scores for each of these questions were standardized and averaged across the years; the four averages were then standardized and combined into a composite measuring parents' religiosity (α = .70). ### Measuring trust in science {#sec022} During the follow-up surveys in 2009--2011, participants were asked about their trust in information from various sources. For example, participants were asked *If you wanted to get more information about the flu virus*, *how much would you trust information from each of the following sources*? The sources included news (e.g., *CNN*, *New York Times*, *NPR*), social sources (*YouTube*, *President Obama*), internet (e.g., *Wikipedia*), and scientific sources (e.g., *NASA*, *science museum*). The topic of the information changed each year and included the flu virus (2009), genetically modified organisms (GMO; 2010), the Hubble Space telescope (2011), nuclear power (2011), and climate change (2009, 2011). Participants responded on a scale ranging from 0 (*would not trust this source*) to 10 (*would definitely trust this source*). Excluding the questions about climate change, we identified 11 questions about scientific sources of information; those included information about the flu virus from a (1) science museum and (2) PBS Nova or Discovery; information about GMOs from (3) a science museum and (4) PBS Nova/Discovery; information about the Hubble from (5) Nasa, (6) a science museum, (7) PBS Nova/Discovery, and (8) an astronomy professor; information about nuclear power from (9) a science museum, (10) a science professor, and (11) PBS Nova/Discovery. Since the items were highly related, we standardized and combined them into a composite score representing trust in scientific sources of information (α = .90). Results {#sec023} ------- Inspection of zero-order correlations (see [S2 Table](#pone.0207125.s002){ref-type="supplementary-material"}) showed that religiosity was again related to more negative attitudes towards science (*r* = -.16, *p* \< .001) and lower levels of both total (*r* = -.26, *p* \< .001) and non-contested science knowledge (*r* = -.13, *p* \< .001). Using structural equation modeling, we created a model to examine the relation between parents' religiosity, participants' religiosity, attitudes towards science, trust in scientific sources of information, and science knowledge. We used parents' religiosity (as measured in 1987--1994) to predict participants' religiosity and attitudes towards science (as measured in 2008). We also used participants' religiosity and attitudes towards science to predict science knowledge scores and trust in scientific information. Thus, we expected to replicate the findings from Study 1, showing that the relationship between religiosity and science knowledge is mediated by attitudes towards science. Further, we examined the indirect (mediated) effects from both parents' religiosity and participants' religiosity to science knowledge and trust in scientific information. All analyses were conducted in MPlus v. 8 \[[@pone.0207125.ref040]\]. Missing data on endogenous variables were handled using full-information maximum likelihood. [Fig 2](#pone.0207125.g002){ref-type="fig"} shows the model and [Table 3](#pone.0207125.t003){ref-type="table"} presents the coefficients of all the indirect effects in the model. {#pone.0207125.g002} 10.1371/journal.pone.0207125.t003 ###### Coefficients for all possible indirect effects from SEM model in study 2. Standardized coefficients and 95% confidence intervals are displayed. {#pone.0207125.t003g} Path β *p* LLCI, ULCI -------------------------------------------------------------------------------------------- ------ --------- -------------- **Predicting Science Attitudes**     Parents' religiosity ➔ Religiosity (2008) ➔ Science attitudes -.06 \< .001 -.079, -.048 **Predicting Trust in Scientific Sources of Information**     Religiosity (2008) ➔ Science attitudes ➔ Trust in Science info. -.05 \< .001 -.067, -.040     Parents' religiosity ➔ Science attitudes ➔ Trust in Science Info. -.02 \< .001 -.035, -.012     Parents' religiosity ➔ Religiosity (2008) ➔ Trust in Science Info. -.04 \< .001 -.054, -.026     Parents' religiosity ➔ Religiosity (2008) ➔ Science attitudes ➔ Trust in Science Info. -.02 \< .001 -.061, -.027 **Predicting Science Knowledge Scores**     Religiosity (2008) ➔ Science Attitudes ➔ Science knowledge -.03 \< .001 -.038, -.020     Parents' religiosity ➔ Science attitudes ➔ Science Knowledge -.01 \< .001 -.019, -.006     Parents' religiosity ➔ Religiosity (2008) ➔ Science Knowledge -.03 \< .001 -.042, -.020     Parents' religiosity ➔ Religiosity (2008) ➔ Science attitudes ➔ Science Knowledge -.01 \< .001 -.047, -.022 We conducted the analysis using 5,000 bootstrapped samples (non-bias corrected). To control for demographic variables, we regressed each of our endogenous variables onto region (north vs. south), gender, race (two dummy codes for Hispanic and black), education (as of 2008), and parents' education levels; these effects are not depicted in the figure. Using standard fit criteria \[[@pone.0207125.ref041]\], the model fit the data well, χ^2^ (4) = 12.10, *p* = .017, RMSEA = .027, CFI = .996, SRMR = .007. Similar to Study 1, the model explained about 29% of variance in science knowledge (*R*^2^ = .29, *p* \< .001). At the request of a reviewer, a model controlling also for political orientation (see [S1 File](#pone.0207125.s003){ref-type="supplementary-material"}) and removing the literal belief items (see 'Revisiting Studies 1 and 2', below) was also tested; it showed highly similar relations to those reported here. As shown in [Fig 2](#pone.0207125.g002){ref-type="fig"} and [Table 3](#pone.0207125.t003){ref-type="table"}, this model provides a replication of the effects reported in Study 1. Participants' religiosity (in 2008) predicted lower science knowledge scores, β = -.09, *p* \< .001, and the indirect effect to science knowledge via science attitudes was significant (see [Table 3](#pone.0207125.t003){ref-type="table"}). Including science attitudes as a mediator accounted for about 54% of the total effect and reduced the relationship between religiosity and science knowledge from β = -.09 to β = -.06, though it remained significant, β = -.06, *p* \< .001. We also conducted additional analyses to directly replicate the mediation models using only the common items between the GSS and LSAY datasets. These analyses are reported in the section 'Revisiting Studies 1 and 2', below. Next, religiosity (in 2008) also predicted having less trust in scientific sources of information, β = -.11, *p* \< .001. Positive attitudes toward science predicted more trust in scientific sources of information, β = .29, *p* \< .001, and the indirect effect via science attitudes was significant (see [Table 3](#pone.0207125.t003){ref-type="table"}). Including science attitudes as a mediator accounted for about 54% of the total effect, reducing the relationship between religiosity and trust in scientific sources of information from β = -.11 to β = -.06, though it remained significant, β = -.06, *p* = .002. Parents' religiosity (as reported by the parents in 1987--1994) predicted participant's negative attitudes towards science (in 2008), β = -.08, *p* \< .001, and the indirect effect via the participant's religiosity (also in 2008) was significant. Including participant's religiosity as a mediator between parents' religiosity and science attitudes explained 75% of this effect, reducing the relationship to non-significance, β = -.02, *p* = .399. Finally, the indirect relation from parents' religiosity to science knowledge was significant through all three routes: via participants' religiosity (in 2008), via science attitudes (in 2008), and through participants' religiosity to science attitudes, and then to science knowledge. Similarly, the same three indirect paths from parent's religiosity to trust in scientific information were also significant (see [Table 3](#pone.0207125.t003){ref-type="table"}). ### The role of education {#sec024} In the analyses above, education served as one of the covariates. When we tested education as a mediator alongside science attitudes in a simple mediation model (including all covariates), the indirect effect of education was not significant (β = -.01, 95% CI: -.0243, .0039); the indirect effect through science attitudes remained significant (β = -.04, 95% CI: -.0509, -.0290), reducing the relation between religiosity and science knowledge from β = -.13, *p* \< .001, to β = -.09, *p* \< .001. Discussion {#sec025} ---------- These results both replicate and extend the effects found in Study 1. Specifically, religiosity was again associated with lower levels of both contested and non-contested science knowledge, with more negative attitudes towards science, and with less trust in scientific sources of information. These results suggest that religiosity is associated with a less positive view of science, which results in less science learning. These effects were obtained for non-contested science topics, lending additional support to the general conflict hypothesis. In addition to replicating the effects from Study 1, we also explored additional longitudinal relations between religiosity and science attitudes and knowledge. We found that negative attitudes towards science may begin early in life, with the religious beliefs of one's parents. That is, the data here suggests that growing up in a religious household may lead to less positive views of science, resulting in lower levels of science knowledge and less trust in scientific information. This sequence is consistent with a directional model in which religiosity leads to a more negative view of science and to less science knowledge. In the analyses that follow, we first examined a direct replication between the data from Studies 1 and 2, using only the common items between the studies. Studies 1 and 2 revisited {#sec026} ========================= The religiosity composites in studies 1 and 2 included two religiosity items that were common across the two data sets. In both surveys, participants were asked if they believed in a personal god, and how often they attend religious services. As these two items represent central themes in measures of religiosity, it was of interest to test the relation between a religiosity composite made of these two items and non-contested science knowledge. In addition to items regarding belief in God and attendance of religious services, Study 2 included a third item concerning participants' belief in the bible (i.e., *The bible is the actual word of God and is to be taken literally word for word*). As literal interpretation of the bible is a feature of fundamentalism \[[@pone.0207125.ref007], [@pone.0207125.ref042]\], it might be argued that the results are partly due to fundamentalism. Therefore, it is important to examine whether the data show a negative relation between religiosity and science knowledge, using a religiosity measure without the biblical literalism item. Both datasets included the same four-item measure of science interest, and almost the same four items measuring attitudes towards general science. We, therefore used these common items to construct the science attitudes composite. Finally, the science knowledge measure was identical across the two studies. For both the GSS and the LSAY data, we standardized and then combined responses to the questions about belief in God and attendance of religious services (αs = .64 and .46, respectively). We also standardized and then combined responses to the eight items that formed the science attitudes scale (for GSS: α = .69; for LSAY: α = .74). Using these measures of religiosity and science interest as well as our previous measure of science knowledge, we repeated the mediation analyses with both the GSS and the LSAY data, separately. We controlled for the same demographic variables as reported in the previous analyses. The results show significant negative relations between religiosity and science knowledge in both the GSS and the LSAY data (see [Fig 3](#pone.0207125.g003){ref-type="fig"}). The results also show significant indirect effects through science attitudes for both studies. For the GSS: β = -.02, 95% CI: -.0276, -.0141; for the LSAY: β = -.03, 95% CI: -.0448, -.0244; the indirect effects through science attitudes explain 40% and 31% of the total effects in the two studies, respectively. Controlling for the mediator, the direct relation between religiosity and science knowledge was reduced from β = -.05 to β = -.03 (Study 1) and from β = -.05 to β = -.02 (Study 2). As with Study 1, we also tested alternative models where science knowledge was treated as the mediator (see Figure A in [S1 File](#pone.0207125.s003){ref-type="supplementary-material"}.). {#pone.0207125.g003} Discussion {#sec027} ---------- These results replicate our previous findings for both Study 1 and Study 2 and demonstrate that the observed relations do not depend on fundamentalist or literalist forms of belief. Next, we turn to replication of these results in different samples and using another measure of religiosity, but one that also focuses on beliefs and practice. Study 3 {#sec028} ======= The purpose of Study 3 is to replicate the effects observed in Studies 1 and 2 using another independent dataset and another measure of religiosity; this measure is widely used \[[@pone.0207125.ref023]\] and does not include items about fundamentalist or literalist beliefs. Data for Study 3 were collected for a larger survey on general aspects of science rejection with a focus on beliefs about genetically modified foods. Other aspects of this survey will be reported elsewhere \[[@pone.0207125.ref043]\] but the relations described here are not described elsewhere. The surveys also included the relevant measures of science knowledge questions, religiosity, and demographic controls. Participants {#sec029} ------------ Participants (*n* = 747) were recruited simultaneously from the undergraduate subject pool at a northeastern university (*n =* 262) and from the Research Match online recruitment system (*n* = 485). There were 178 males and 532 females (4 identified as other). Participants ranged in age from 18--93 (*M* = 41.30, *SD* = 19.85) and were 76% white, 13% Asian, 3% Black, 3% Hispanic, and 4% mixed or other. Eight percent reported having a high school education, 32% some college, 5% a two-year degree, 19% a 4-year degree, 8% some post-graduate education, and 28% had an advanced degree. Materials and procedure {#sec030} ----------------------- Participants were recruited for a survey regarding beliefs about science where they filled out questionnaires regarding their knowledge and beliefs about genetically modified foods. All questionnaires were presented in randomized order, except that demographic questions were presented last. Among these questionnaires was the 18-item measure of science knowledge, composed of the 13 non-contested items, as well as the 5 contested items, common between Studies 1 and 2. The items were summed up to create a general science knowledge score ranging from 0 to 18, as well as a non-contested science knowledge score ranging from 0--13. Other measures are not relevant to the current study and are not discussed further. Demographics questions included a 6-item measure of religiosity (e.g., *I believe in God*; α = .96; \[[@pone.0207125.ref023]\]). Participants also answered questions about their socioeconomic standing during childhood and currently, their political orientation regarding social issues, economic issues, and voting history (α = .90), and their parents' education levels (α = .79). Throughout the survey, there were 6 attention check questions (e.g., *select 'somewhat agree' and continue on;* \[[@pone.0207125.ref044]\]); excluding those who missed at least one attention check question did not change the results. Both sets of results are reported below. Results and discussion {#sec031} ---------------------- Replicating the results from Studies 1 and 2, religiosity correlated negatively with general science knowledge (*r* = -.30, *p* \< .001) and with non-contested science knowledge (*r* = -.20, *p* \< .001). We then regressed the non-contested science knowledge onto religiosity while controlling for data source, age, education, gender (two dummy codes), parents' education, current and childhood SES, political orientation, and race (4 dummy codes). The model was significant and explained about 21% of variance in non-contested science knowledge, *R*^2^ = .21, *F* (14, 694) = 12.78, *p* \< .001. Religiosity continued to uniquely predict lower levels of non-contested science knowledge. Coefficients are displayed in [Table 4](#pone.0207125.t004){ref-type="table"}, below. An additional analysis with the same controls but excluding those who missed at least one attention check item (*n* = 32), also showed a significant relation between religiosity and non-contested science knowledge: β = -.13, *F* (1, 662) = 10.11, *p* = .002, *partial r =* -.12. These results replicate the negative relation between religiosity and science literacy, using a somewhat different measure of religiosity. 10.1371/journal.pone.0207125.t004 ###### Multiple regression coefficients indicating the negative relation between religiosity and science knowledge in study 3. Standardized coefficients are displayed. {#pone.0207125.t004g} Predictor β *t* *p* *partial r* ------------------------ ------ ------- --------- ------------- **Source** .11 1.60 .110 .06 **Age** -.06 -1.04 .300 -.04 **Education** .21 4.23 \< .001 .16 **Male** -.10 -0.55 .586 -.02 **Female** -.34 -1.87 .062 -.07 **Parents' education** .10 2.48 .013 .09 **Current SES** .08 2.08 .038 .08 **Childhood SES** .05 1.47 .142 .06 **White** .11 1.56 .120 .06 **Black** -.10 -2.36 .018 -.09 **Hispanic** -.08 -1.91 .057 -.07 **Asian** -.07 -1.02 .310 -.04 **Religiosity** -.19 -5.19 \< .001 -.19 ### The role of education {#sec032} In the analyses above, education served as one of the covariates. When we tested education as a mediator in a simple mediation model (including all covariates), the indirect effect of education was significant (β = -.01, 95% CI: .0034, .0339), reducing the relation between religiosity and science knowledge from β = -.15, *p* \< .001, to β = -.16, *p* \< .001. Study 4 {#sec033} ======= Data for Study 4 were collected as part of a larger study for scale development and is part of a currently ongoing research program \[[@pone.0207125.ref045]\]. This survey also included the relevant measures of religiosity and science knowledge, as well as the same measure of science attitudes used in Study 1. The present data have not been described or reported elsewhere. Participants {#sec034} ------------ Participants (*n* = 992) were recruited simultaneously from the undergraduate subject pool at a northeastern university (*n =* 375) and from the Research Match online recruitment system (*n* = 617). There were 231 males and 723 females (11 identified as other). Participants ranged in age from 18--86 (*M* = 38.78, *SD* = 19.21) and were 72% white, 13% Asian, 4% Black, 4% Hispanic, and 4% mixed or other. Ten percent reported having a high school education, 33% some college, 6% a two-year degree, 19% a 4-year degree, 6% some post-graduate education, and 25% had an advanced degree. Materials and procedures {#sec035} ------------------------ The survey collected information on general beliefs about science topics, such as science attitudes and beliefs about climate change, vaccinations, alternative medicine, etc. Again, only the items relevant to the current study are described here. Participants completed the same 18-item measure of science knowledge from Study 3; this was scored into an 18-item general science knowledge score and a 13-item non-contested science knowledge score. Participants also responded to a shortened (6-item) version of the science attitudes measure used in the previous studies (α = .66). The six items were: *Scientific research makes life change too fast*, *the benefits of scientific research outweigh any possible harms*, *the world is better because of science*, *science and technology make more opportunities for the next generation*, *scientists want to make life better*, and *it is not important to know about science in daily life*. Participants also responded to the same demographic controls as in Study 3: childhood and current SES, the three-item measure of political orientation (α = .89), parents' education levels (α = .80), and the 6-item religiosity scale used in Study 3 (α = .96; \[[@pone.0207125.ref023]\]). Finally, 11 attention check questions were included as in Study 3 \[[@pone.0207125.ref044]\], so those who missed at least one item will be excluded. Results with and without exclusion criteria were similar and both are reported below. Results {#sec036} ------- ### Main hypothesis {#sec037} Replicating the results from studies 1--3, religiosity negatively correlated with general (*r* = -.30, *p* = \< .001) and non-contested (*r* = -.18, *p* \< .001) science knowledge, and negatively with science attitudes (*r* = -.12, *p* \< .001). As before, we regressed the non-contested science knowledge score onto religiosity while controlling for data source, age, education, gender (two dummy codes), parents' education, current and childhood SES, political orientation, and race (4 dummy codes). The model was significant and explained about 19% of variance in non-contested science knowledge, *R*^2^ = .19, *F* (14, 943) = 15.68, *p* \< .001. Religiosity continued to uniquely predict lower levels of non-contested science knowledge. Coefficients are displayed in [Table 5](#pone.0207125.t005){ref-type="table"}, below. In an additional analysis with the same controls, excluding those who missed at least one attention check item (*n* = 32) also showed a significant negative relation between religiosity and non-contested science knowledge: β = -.13, *F* (1, 859) = 13.73, *p* \< .001, *partial r =* -.13. 10.1371/journal.pone.0207125.t005 ###### Multiple regression coefficients indicating the negative relation between religiosity and science knowledge in study 4. Standardized coefficients are displayed. {#pone.0207125.t005g} Predictor β *t* *p* *partial r* ------------------------ ------ ------- --------- ------------- **Source** -.05 -0.83 .409 -.03 **Age** -.09 -1.79 .074 -.06 **Education** .30 7.01 \< .001 .22 **Male** .02 0.13 .898 .00 **Female** -.18 -1.42 .157 -.05 **Parents' education** .10 2.76 .006 .09 **Current SES** .09 2.74 .006 .09 **Childhood SES** -.05 -1.41 .159 -.05 **White** .01 0.18 .859 .01 **Black** -.09 -2.16 .031 -.07 **Hispanic** -.07 -1.64 .102 -.05 **Asian** -.13 -2.27 .023 -.07 **Religiosity** -.16 -5.24 \< .001 -.17 ### Mediation analysis {#sec038} As in Studies 1 and 2, I conducted a mediation analyses using the religiosity scale as the independent variable, science attitudes as the mediator, and non-contested science knowledge as the dependent variables. We again controlled for data source, age, education, gender, race, parents' education, current and childhood SES. As shown in [Fig 4](#pone.0207125.g004){ref-type="fig"}, below, the indirect effect through science attitudes was significant, β **=** -.03, 95% CI: -.0440, -.0134. Controlling for science attitudes reduced the relation between religiosity and science knowledge from β **=** -.17 to β **=** -.14, and accounted for about 18% of the relation between religiosity and non-contested science knowledge. As with Studies 1 and 2, we also tested an alternative model where science knowledge was the mediator between science attitudes and science knowledge (see Figure A in [S1 File](#pone.0207125.s003){ref-type="supplementary-material"}). {#pone.0207125.g004} ### The role of education {#sec039} In the analyses above, education served as one of the covariates. When we tested education as a mediator alongside science attitudes in a simple mediation model (including all covariates), the indirect effect of education was not significant (β \< .01, 95% CI: -.0144, .0240); the indirect effect through science attitudes remained significant (β = -.02, 95% CI: -.0326, -.0017), reducing the relation between religiosity and science knowledge from β = -.16, *p* \< .001, to β = -.15, *p* \< .001. ### Discussion {#sec040} These results provide a direct replication of the negative relation between religiosity and science knowledge observed in Study 3, and a conceptual replication of the results observed in Studies 1--2. This relation was larger in Studies 3 and 4 than in studies 1 and 2, perhaps because of differences between the samples and or the religiosity measure that was used. Though the science attitudes scale was not available in Study 3, it was included in Study 4. Results suggest that a negative relation between religiosity and science attitudes partially accounts for the negative relation between religiosity and science knowledge; this indirect effect is similar in magnitude to the indirect effects observed in Studies 1 and 2. General discussion {#sec041} ================== The findings from these four studies show that religiosity is negatively related to science knowledge and is associated with more negative attitudes towards science. Importantly, these results were obtained while controlling for a large number of demographic variables, and after deleting contested portions of science knowledge. All four studies are correlational. However, the relation of parents' reports of their religiosity and the religious upbringing of their children with (some 20 years later) their children's attitudes toward science (Study 2) implies that religiosity may impact attitudes towards science, and thus science knowledge, later in life. We measured religiosity differently across the four studies. The religiosity measures in studies 1 and 2 were identical as were the measures used in studies 3 and 4. Previous researchers examined the relation between religiosity and science knowledge, measuring religiosity via specific combinations of religious affiliation \[[@pone.0207125.ref010]\] and fundamental/literalist interpretations \[[@pone.0207125.ref005], [@pone.0207125.ref009]\]. The present results demonstrate that the negative relation between religiosity is general and robust across different samples and broad religiosity measures that focus on belief in God and religious practices (attendance, prayer)---quintessential elements of religiosity \[[@pone.0207125.ref021]--[@pone.0207125.ref024]\]. Of course, the differences between the present study and past research may also be understood in light of differences between how sociologists and social psychologists approach research. Whereas sociologists may be more interested in the social structure of religious traditions or affiliations, we have approached this question with a focus on individual variation in the strength of religious belief and practice. These are two different ways of asking the same question which both shed light on the relation between religion and science. Echoing other researchers \[[@pone.0207125.ref006],[@pone.0207125.ref007]\], we hope for more research in psychology and in other disciplines on this issue, both in and outside of the US. Our findings indicate that one possible mediator of the negative relation between religiosity and science knowledge is negative attitudes towards science. Our interpretation of this relation is straightforward: to the extent that science is viewed as less interesting, useful, or valid, one would be less likely to learn about it. Longitudinal findings from Study 2 suggests these negative attitudes towards science may begin early in life when one is exposed to their parents' religious beliefs. Future research will be needed to examine exactly how religious upbringing may negatively impact science attitudes. However, it's important to note that cross-sectional mediation models do not indicate causation because we did not manipulate religiosity or science attitudes in these studies. In line with this consideration, we presented in the supplementary file formal tests of alternative models. Some resolution of this issue may be obtained only with longitudinal research. In each study, we also tested whether the relation between religiosity and science knowledge could be accounted for by formal education level. However, education level did not consistently account for additional variance above science attitudes. While some previous research suggests that religion may lead one to pursue less education \[[@pone.0207125.ref036]\], the present findings are not necessarily in conflict with such reports. In each of the four studies, religiosity was negatively correlated with education but the relation between education and science knowledge was weaker and inconsistent. This is likely because not all education focuses on science such that education in the humanities and arts, for example, may not contribute to science knowledge. However, much more research will be needed to disentangle the roles of education, science education, and influences of religion on interest in science, which likely would be related to whether one pursues science education. Another limitation to consider is that all of the studies described herein are based in the US. Future research will need to consider country-level and cultural differences as moderators of the relation between religiosity and science knowledge. Thus, the picture that emerges is that religiosity is associated with less interest in science and the believe that science is less important; such attitudes are related to somewhat lower levels of science literacy and less trust in scientific sources of information. Following Evans \[[@pone.0207125.ref005]\], this finding suggests that, as religion offers its own route to knowledge, people of faith might be less interested in what science has to offer. However, because the mediation was partial, it is clear there are other routes leading from religiosity to lower science knowledge. For example, one study \[[@pone.0207125.ref046]\] found that threatening stereotypes about Christians' low competence in science may lead them to underperform in science. Elsdon-Baker \[[@pone.0207125.ref047]\] suggested that the proliferation of such stereotypes may actually "create creationists" by reinforcing the belief that Christians don't belong in science, thus promoting identification with counter-scientific beliefs. Sherkat \[[@pone.0207125.ref042]\] suggested that tight religious communities might actually prevent access to science knowledge because they deem it untrustworthy. In view of the present findings, the challenge is how to increase science knowledge in the face of religious disinterest or actual opposition. Improving science knowledge and literacy will allow people to make more informed, evidence-based decisions about choices, beliefs, and activities, as well as about the products and services they use or avoid. According to the present result, the recent decline in religiosity \[[@pone.0207125.ref048]\] might help but, given that the majority of the world population still defines itself as religious, it behooves us to find a way for people to keep their religious beliefs and yet open their minds to science. Supporting information {#sec042} ====================== ###### Correlations between all variables in Study 1. \*\**p* \< .001, \**p* \< .05, ^†^*p* \< .10. (PDF) ###### Click here for additional data file. ###### Correlations between all variables in Study 2. \*\**p* \< .001, \**p* \< .05, ^†^*p* \< .10. (PDF) ###### Click here for additional data file. ###### Supplementary analyses document. (DOCX) ###### Click here for additional data file. [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

Trajectories of Health-Related Quality of Life and Perceived Social Support Among People Living With HIV Undergoing Antiretroviral Treatment: Does Gender Matter? {#sec1} ================================================================================================================================================================= Over the last two decades, cutting-edge progress has occurred in the treatment and prevention of HIV infection, which has not only significantly reduced mortality and morbidity among people living with HIV (PLWH; [@ref68]; [@ref21]) but has also been attributed to signaling the end of the HIV/AIDS epidemic ([@ref25]; [@ref16]). However, the enormous progress in bio-medical care for PLWH did not directly translate to the improvement of their psychological well-being. More specifically, both in the past (e.g., [@ref7]) and at present, PLWH systematically declare lower levels of well-being and higher psychological distress than the general population ([@ref53]), and especially worse health-related quality of life (HRQoL) with respect to other chronic diseases ([@ref62]). Although a large number of studies have been conducted on factors related to the HRQoL of these patients (e.g., [@ref13]; [@ref37]; [@ref39]; [@ref11]; [@ref48]; [@ref54]; [@ref70]), they failed not only to provide a consistent picture of the variables associated with the HRQoL of PLWH but also to produce a convincing answer to the aforementioned issue of low well-being among PLWH, especially at a time when their life expectancies are similar to the general population ([@ref26]). The only visible trend deals with the fact that clinical factors have been historically considered as the predominant predictor of the well-being of PLWH ([@ref50]), while there is now a growing recognition of psychosocial factors as the major determinant of their quality of life ([@ref20]; [@ref23]). Some authors underline that the plethora of inconclusive results in research on HRQoL among PLWH is derived from methodological shortcomings among most studies on this topic. Notably, the literature is dominated by cross-sectional frameworks using the variable-centered approach, which focuses only on the average values for an entire study sample and neglects the heterogeneity of well-being among PLWH and its relationship with sociodemographic, clinical, and psychological characteristics ([@ref60]; [@ref67]). Therefore, the present study aims to overcome these methodological drawbacks by applying a longitudinal design accompanied by a person-centered perspective that allows for the identification of subgroups with different levels of HRQoL and their changes during time. Specifically, we aim to enrich the literature by focusing on the relatively understudied issue of gender-based differences in HRQoL among PLWH. Surprisingly, although the global amount of HIV-infected women is similar to the number of HIV-infected men, both of which continue to grow worldwide ([@ref31]), the majority of studies on HRQoL in HIV infection were conducted on male populations only (e.g., [@ref43]; [@ref49]; [@ref69]; [@ref30]). Thus, research on HRQoL among HIV-infected women remains scarce (e.g., [@ref52]; [@ref33]). In addition, authors examining gender differences in HRQoL in this patient group consequently observed lower HRQoL among HIV-infected women than HIV-infected men (e.g., [@ref15]; [@ref56]; [@ref19]). There are several hypotheses on this consistent finding that have pointed to the limited access to antiretroviral treatment (ART) in some world regions ([@ref61]; [@ref3]), more intense HIV-related stigma, and an associated higher rate of mental disorders among HIV-infected women than HIV-infected men ([@ref14]; [@ref51]; [@ref32]). On the other hand, several studies have demonstrated that females living with HIV exhibit far greater adherence to treatment than their male counterparts and therefore report better HIV-related clinical outcomes ([@ref22]; [@ref57]; [@ref9]). In considering the aforementioned findings, it seems that no comprehensive explanation for the observed gender differences in HRQoL among PLWH exists to date. As such, the present study examines one variable that may provide important context on this topic -- social support ([@ref17]). HIV/AIDS is a chronic disease that promotes multidimensional psychological distress among PLWH, which is now primarily associated with persistent HIV-related stigma and social isolation ([@ref64]). Many studies have observed that social support -- particularly perceived social support (PSS) -- is one of the most important assets in coping with HIV infection and related distress (e.g., [@ref71]; [@ref34]; [@ref28]). More specifically, perceiving a high availability of support may enhance adjustment to HIV infection directly through improved adherence to treatment (e.g., [@ref2]; [@ref1]) and also indirectly through buffering the effect of HIV-related stigma on mental functioning and quality of life among these patients ([@ref4]; [@ref10]). Although the beneficial effects of perceived support on quality of life of PLWH are widely known, some authors have recently become increasingly skeptical regarding this unambiguously optimistic picture and have highlighted several methodological shortcomings of existing studies conducted using a cross-sectional framework, which is the most common ([@ref63]). Consequently, it is difficult to solve the *egg or chicken dilemma* in research on the link between perceived social support and HRQoL among PLWH. Notably, existing studies have been conducted in small samples (since the implementation of a repeated measurement design among PLWH may be challenging; [@ref12]) or using only a baseline assessment of PSS as a predictor for HRQoL changes ([@ref42]). Thus, the current literature has not examined possible heterogeneity of the dual trajectories of HRQoL and PSS. Furthermore, another topic that remains understudied in the literature deals with gender differences in both PSS and HRQoL among PLWH ([@ref35]). Current Study {#sec2} ------------- In considering the aforementioned research gaps, the aim of our study was three-fold. First, we aimed to examine whether heterogeneity of univariate change of HRQoL and PSS exists among PLWH, and if these trajectories are also gender-related, both with and without other possible sociodemographic and clinical covariates. Then, the probability of following a given pattern of the dual trajectories of HRQoL and PSS was explored. Specifically, we were interested in the co-occurrence of trajectories with the same direction of change, under the assumption of existing cross-sectional studies that a decrease or an increase in HRQoL corresponds to relevant changes in PSS. Finally, we examined whether any gender differences in joint probability for dual trajectories existed, supposing that combinations of changes may not be equally distributed due to gender-related patterns of social exchange as well as the social consequences of being diagnosed with HIV. Materials and Methods {#sec3} ===================== Participants and Procedure {#sec4} -------------------------- The participants were 252 persons with confirmed HIV-positive results undergoing antiretroviral therapy in an outpatient clinic. The majority of them were men, which is typical based on the gender-related prevalence rate of HIV infection in Europe and the United States ([@ref44]). Detailed characteristics of the sample are provided in [Table 1](#tab1){ref-type="table"}. ###### Sociodemographic and clinical variables in the studied sample (*N* = 252). Variable *N* (%) ------------------------------------------------------------- ------------------ Gender Male 208 (82.5) Female 44 (17.5) Age in years (*M* ± SD) 39.03 ± 10.40 Marital status Married 147 (58.3) Single 105 (41.7) Education Basic vocational 109 (43.3) Secondary and university degree 143 (56.7) HIV/AIDS status HIV+ only 215 (85.3) HIV/AIDS 37 (14.7) HIV infection duration in years (*M* ± SD) 7.23 ± 6.23 Antiretroviral treatment (ART) duration in years (*M* ± SD) 5.82 ± 5.25 CD4 count 575.48 ± 2248.89 *M, mean; SD, standard deviation*. The study design was longitudinal, with three measurements at 6-month intervals. After written informed consent was obtained from a participant, they filled in the self-descriptive questionnaire provided. For the next two measurements, they were approached during their control visit in the outpatient clinic after establishing the date *via* phone or email, based on their preference. All longitudinal data were collected by trained research assistants using a "paper-and-pencil" approach. Participation in the study was voluntary. The study was approved by the institutional ethics committee. Measures {#sec5} -------- *Health-related quality of life* was assessed using the WHO Quality of Life-BREF (WHOQOL-BREF), developed under a WHO initiative for cross-cultural assessment ([@ref75]). The tool consists of 26 items to measure four domains: physical health, psychological health, social relationships, and environment. Each item is rated on a five-point Likert scale (scores ranged from 1 to 5), and raw scores were used. Since correlations between domains in our study were stable and moderate (from 0.48 to 0.69), and followed the research indicating a possibility of assessing global HRQoL using this tool (Harsha et al., 2016), the overall indicator was obtained by summing and averaging all item scores. Higher values indicate a higher quality of life. The reliability, measured by the Cronbach's *α* coefficient, was 0.93, 0.92, and 0.92, from the first to third wave, respectively. *Perceived social support* was measured using the relevant subscale of eight items from the Berlin Social Support Scales developed by [@ref500]. The answers are provided on a Likert-type scale, from 1 (*not true at all*) to 4 (*entirely true*), then summed and averaged. The higher scores indicate higher PSS. The Cronbach's *α* coefficient was 0.93, 0.92, and 0.92, for first, second, and third measurement, respectively. Data Analysis {#sec6} ------------- We started the analysis with univariate latent class growth curve models to examine how HRQoL and PSS changed in our sample during the study period. From 1- to 5-class solutions were tested separately for HRQoL and PSS, and error variances and covariances were freely estimated across classes. The optimal solution was identified on the basis of several criteria widely identified in the literature ([@ref58]). Namely, we used Akaike as well as Bayesian information criterion (AIC and BIC, respectively), including the sample size-adjusted BIC (SABIC). The model with lower values was favored. Next, entropy as a measure of accuracy of classification was taken into consideration; in this case, the model with higher values was favored ([@ref18]). Finally, sample proportion per class was analyzed, since classes with very few individuals may be sample-specific and difficult to replicate. The practical rule is to favor a model with a fewer number of classes when at least one class has a frequency of less than 5% of the sample size ([@ref40]). Time was coded as 0 for the first measurement, 0.5 for the second, and 1 for the third ([@ref6]). Both linear and quadratic trends were explored, but we did not present them further since all quadratic terms were revealed as insignificant. To identify covariates of trajectories, the bias-adjusted three-step analysis ([@ref73]) was implemented in order (1) to separate modeling trajectories from their relationship with other variables and (2) to correct for probabilistic classification to classes. Namely, when univariate models were established (e.g., the number, shape, and membership of trajectories were fixed), we examined whether any differences existed in predicting this membership based on sociodemographic and clinical variables. We started with gender and then added other covariates to determine if they would modify the gender effect. In the next step, joint probabilities were computed since we used two sets of trajectories (one for HRQoL and the other for PSS). Specifically, we were interested in the probability of belonging to a given trajectory of HRQoL when simultaneously being a member of a given trajectory of PSS. Finally, by means of multinomial logistic regression, we assessed whether the probability of being a member of each combination of HRQoL and PSS trajectories was the same for women and men, both with and without additional covariates. All analyses were performed using IBM SPSS Statistics version 25 ([@ref41]) and LatentGOLD version 5.1 ([@ref73]). Results {#sec7} ======= Descriptive Statistics and Missing Values {#sec8} ----------------------------------------- [Table 2](#tab2){ref-type="table"} presents basic descriptive statistics for repeated measures of HRQoL and PSS. The dropout due to longitudinal design was 41% of the sample between the first and last measurements. Missing data analysis suggested that the pattern of missingness can be treated as random (Little's MCAR test: *χ*^2^ = 53.32, df = 54, *p* = 0.50); therefore, the option that included all available data was chosen, with missing values for indicators being handled by the maximum likelihood function ([@ref73]). Furthermore, regarding sociodemographic and clinical variables, there were also no significant differences between completers and non-completers. However, the result for gender was on the edge of significance (*χ*^2^ = 3.95, df = 1, *p* = 0.05), suggesting a tendency of higher dropout among women than men. ###### Descriptive statistics for health-related quality of life (HRQoL) and perceived social support (PSS). Variable Range Mean SD Skewness Kurtosis ---------- ------------ ------ ------ ---------- ---------- HRQoL Time 1 1.42--4.92 3.75 0.57 −0.82 0.99 Time 2 1.58--4.92 3.74 0.56 −0.59 0.61 Time 3 1.27--4.81 3.66 0.58 −0.65 1.47 PSS Time 1 1--4 2.31 0.68 −0.84 0.561 Time 2 1--3 2.26 0.71 −0.75 −0.37 Time 3 1--3 2.22 0.72 −0.75 −0.29 *SD, standard deviation; sample size for Time 1, Time 2, and Time 3 was 252, 201, 149, respectively*. Heterogeneity of Change: HRQoL and PSS Univariate Trajectories {#sec9} -------------------------------------------------------------- The model fit criteria indicate that model with three trajectories was the best fitted to the HRQoL data (see [Table 3](#tab3){ref-type="table"}). Specifically, although all the informative criteria indices scored lower with every added class, the drop in value became smaller. For models with more than three classes, the smallest class had only four members, which suggests the existence of outliners. Finally, entropy was relatively stable across all models, indicating that the perfect classification of participants was challenging. Average posterior probabilities were 0.79, 0.93, and 0.81 for classes 1, 2, and 3, respectively. This solution is plotted in [Figure 1](#fig1){ref-type="fig"}. ###### Summary of model selection indices of latent class growth curve analysis: Unconditional univariate models for health-related quality of life (HRQoL) and perceived social support (PSS). Model BIC AIC SABIC Number of parameters Entropy Smallest class --------- --------- --------- --------- ---------------------- --------- ---------------- ---- HRQoL 1-Class 3467.27 3456.69 3457.76 3 2-Class 3334.58 3309.88 3312.39 7 0.61 39.3 99 3-Class 3285.87 3247.04 3251.00 11 0.63 25.0 63 4-Class 3262.65 3209.71 3215.09 15 0.64 1.6 4 5-Class 3256.24 3189.18 3196.00 19 0.62 1.6 4 PSS 1-Class 4039.37 4028.78 4029.86 3 2-Class 3824.25 3799.55 3802.06 7 0.84 26.2 60 3-Class 3751.05 3712.22 3716.18 11 0.62 15.9 40 4-Class 3727.09 3674.15 3679.54 15 0.71 7.5 19 5-Class 3738.32 3671.26 3678.08 19 0.57 15.1 38 *BIC, Bayesian information criterion; AIC, Akaike's information criterion; SABIC, sample-size adjusted BIC*. {#fig1} The starting points differed significantly across trajectories (overall Wald statistics = 14101.81, *p* \< 0.001; for all pairwise comparisons *p* \< 0.001). The first class was the most numerous, containing 45.6% of the sample, for whom HRQoL significantly decreased during the study period (slope = −0.13, *z* = 2.55, *p* \< 0.02). The second class, with the lowest and most stable HRQoL (slope = −0.08, *z* = −0.86, ns), was represented by 29.4% of the sample. Finally, the third class included 25% of PLWH and exhibited the highest and increasing HRQoL (slope = 0.12, *z* = 1.98, *p* \< 0.05). For PSS, the 4-class solution was chosen as the best fitted to the data due to the lowest BIC and SABIC values, and highest accuracy of classification, while retaining a reasonable size for the smallest class. Moreover, this decision was supported by visibly worse performance on all these criteria by the 5-class model (see [Table 3](#tab3){ref-type="table"}). Average posterior probabilities ranged from 0.78 for class 2 to 0.89 for class 1. [Figure 2](#fig2){ref-type="fig"} presents the obtained trajectories. The majority of the sample (63.1%) was allocated to the first class and exhibited decreasing PSS trajectory (slope = −0.19, *z* = −2.21, *p* \< 0.05) with a middle starting point. The second class consisted of PLWH (14.3% of the sample) with a slightly higher starting point and increasing PSS (slope = 0.21, *z* = 2.36, *p* \< 0.05). The highest and stable trajectory was represented by 15.1% of the sample, identified as class 3. On the other side was the smallest class 4 (7.5% of the sample) with the lowest and stable PSS, albeit with a noticeable tendency to decrease (slope = −0.23, *z* = −1.08, ns). {#fig2} Gender as a Covariate of Univariate Trajectories {#sec10} ------------------------------------------------ When gender was added to the models as the only predictor of class membership, it was insignificant for both HRQoL (*W* = 4.71, ns) and PSS (*W* = 3.25, ns). When all the other covariates were included in the mode (i.e., age, education, relationship status, CD4 count, duration of being diagnosed with HIV infection, duration of antiretroviral therapy and being in the AIDS stage) for HRQoL, trajectories of gender remained insignificant. Significant effects were observed for age (*W* = 7.11, *p* \< 0.05) and education (*W* = 8.77, *p* \< 0.02). Specifically, PLWH in the decreasing HRQoL trajectory were older than the other two trajectories (41.2 vs. 36.8 and 37.7 years of age, respectively), and PLWH in the increasing HRQoL trajectory were better educated than those with the stable HRQoL trajectory. For PSS, both gender (*W* = 674.39, *p* \< 0.001) and relationship status (*W* = 9.02, *p* \< 0.05) were significant correlates of PSS trajectories. Thus, an interaction of these variables was examined, revealing a significant effect between classes 1 and 2 (*W* = 196.56, *p* \< 0.001). Compared to the decreasing trajectory, there was overrepresentation of single women in the increasing PSS trajectory. Dual Trajectories of HRQoL and PSS: Gender Differences {#sec11} ------------------------------------------------------ [Table 4](#tab4){ref-type="table"} illustrates joint probability while accounting for both sets of heterogonous trajectories (i.e., for HRQoL and PSS simultaneously). It is evident that no clear correspondence exists between the trajectories of HRQoL and PSS. Although the highest probability was noted for the dual decreasing trajectory (0.31), the relatively high probability also points to the simultaneous membership of stable HRQoL and decreasing PSS. Likewise, approximately 11% of PLWH followed the increasing HRQoL and decreasing PSS trajectories concurrently. However, no members in the lowest and stable PSS reported an increase in HRQoL. ###### Joint probability of membership for dual trajectories of health-related quality of life (HRQoL) and perceived social support (PSS). HRQoL PSS ------------ ---------- ---------- ---------- ---------- ---------- Decreasing 0.32 0.05 0.06 0.03 **0.46** Stable 0.20 0.03 0.01 0.05 **0.29** Increasing 0.11 0.06 0.08 0.00 **0.25** **Total** **0.63** **0.14** **0.15** **0.08** **1** Finally, considering joint probability, gender differences were examined with control for all the other sociodemographic and clinical variables. We focused on the dual trajectories combining the decreasing PSS and all HRQoL trajectories, since for all the other combinations the frequency was below 10% of the sample (see [Table 4](#tab4){ref-type="table"}). The results of multinomial logistic regression with dual decreasing trajectories as a reference category showed no significant effect of gender alone (*χ*^2^ = 3.88, df = 2, *p* = 0.14) nor of any of the remaining variables when added to the model (*χ*^2^ = 20.12, df = 16, *p* = 0.22). Discussion {#sec12} ========== The first aim of our study was partly achieved (i.e., we managed to observe the heterogeneity of change in HRQoL and PSS among PLWH), but no systematic gender differences were found with regard to these trajectories. Specifically, we identified three classes of HRQoL, and their members differed in terms of age and education only; therefore, clinical variables did not predict class membership. Thus, our study fits the current trend of HRQoL research, which points to the diminishing role of HIV-related clinical factors as determinants of quality of life for PLWH ([@ref23]). In a time of great progress in HIV/AIDS treatment and prevention HIV, infection has lost its fatal character and became a chronic and manageable health problem ([@ref25]). Therefore, it is unsurprising that socially valid resources such as education can be a significant predictor of HRQoL trajectories, which is in line with other studies (albeit cross-sectional and variable-centered; e.g., [@ref26]; [@ref59]). Moreover, we observed that older age was related to decreasing HRQoL trajectories, even if they started from a relatively high level. This corresponds to existing research on elderly PLWH who compared to younger PLWH faced more problems and impediments in their daily functioning due to higher HIV-related stigma ([@ref29]) as well as difficulties in distinguishing physical HIV symptoms from those associated with aging ([@ref36]; [@ref55]). However, one of the most important (yet null) results deals with the lack of gender differences in HRQoL change among our participants. Specifically, our finding may revise the long-lasting and relatively persistent trend in the literature, which points to lower HRQoL among female PLWH than male PLWH based on cross-sectional data only (e.g., [@ref15]; [@ref56]; [@ref680]; [@ref19]). Some authors observed that gender differences in HRQoL within this patient group may be apparent, i.e., they disappeared after careful adjustment of the results with regard to some clinical (e.g., longer illness duration; [@ref65]) or sociodemographic data (worse employment and education status; [@ref66]). In other words, lower quality of life among female PLWH does not necessarily reflect their more difficult or different adjustment to HIV/AIDS in comparison to male PLWH but may rather be a result of other factors that have not been carefully controlled for in other studies ([@ref8]; [@ref66]). However, the results for PSS do not support this explanation. Namely, the null effect for gender became significant only when other covariates were included in the model, resulting in gender and relationship status being identified as correlates of PSS trajectory membership. For a single woman, there was a higher probability of belonging to the increasing PSS trajectory than to the decreasing one, even if the starting points of both trajectories were only slightly different. Notably, this is inconsistent with results concerning female PLWH ([@ref47]), but it corresponds to research demonstrating that relationship status may have different consequences for men and women, with men benefiting more from marriage ([@ref580]). Among PLWH, men also benefit more from social support, while women are more likely to seek it ([@ref35]; [@ref4]). Thus, being a single woman is not necessarily a disadvantageous condition in this context. Such individuals may effectively receive support from other sources that do not require HIV disclosure, and they may also be less prone to abuse from an intimate partner ([@ref51]). However, we lack data on the relationship status of these women at the time of diagnosis; therefore, it cannot be excluded that they had been infected by a partner in a heterosexual relationship. The decision to be single could thus be a deliberate consequence of this mode of transmission -- the most frequent among women in Europe and the United States ([@ref24]; [@ref31]). Nonetheless, since no relevant published data exist with which to compare this gender-relationship status interaction, it could represent a sample-specific association. As such, this topic requires further research. The obtained combinations of dual trajectories added to a complexity of change in HRQoL and PSS among PLWH. Although the probability of being a member of dual decreasing trajectories was the highest, only 31% of the sample could be assigned to this group. Thus, some factors may indeed be responsible for the simultaneous change of both HRQoL and PSS; however, they do not respond in the same manner across the entire sample, since co-occurrence of the matching change direction in both variables was rather modest. This is especially pronounced within decreasing PSS, where we identified three combinations of dual trajectories: (1) congruent decreasing PSS and HRQoL, (2) decreasing PSS and increasing HRQoL, and (3) decreasing PSS with stable HRQoL. Therefore, a few non-exclusive explanations are possible for this mixture of HRQoL and PSS change. First, it is likely that an interrelation exists between PSS and HRQoL ([@ref12]) causing a downward spiral over time. In this case, some general factors may broadly affect the functioning of PLWH. The natural candidates are those related to sociodemographic resources and clinical characteristics; however, this group did not differ in this respect from the other combinations of trajectories. The primary explanation for this null effect could be that we did not assess the change of these characteristics, only baselines. Nevertheless, this result is congruent with most studies, which show its modest role in the functioning of PLWH after the inception of ART ([@ref70]). Second, for some PLWH, an increase in HRQoL may result in reduced perceived support since it is no longer needed to the same extent. Also, this shift in perception may additionally serve to conserve self-efficacy ([@ref74]), which is of particular importance for patients with chronic diseases that have a stigmatized social reception, thereby further improving quality of life ([@ref46]). Finally, in the third case (i.e., for PLWH with decreasing PSS and stable HRQoL), these two processes may have different temporal dynamics, as a change in PSS likely proceeded changes in HRQoL ([@ref42]). It would be rather exceptional to maintain such an incongruent dynamic-static status in light of findings that suggest cross-sectionally lower PSS is related to lower HRQoL among PLWH ([@ref27]; [@ref12]). Interestingly, being a member of each of these groups remained unrelated not only to gender but also to the other sociodemographic and clinical variables. There are several strengths of this study, including the longitudinal and person-centered approach with a relatively large clinical sample and three measurement points. However, certain limitations must be noted. First, this is a correlational study based on self-descriptive data; thus, no causal interpretation is allowed. Additionally, the separation of the univariate trajectory classes for HRQoL was only acceptable. Moreover, since the HRQoL measurement covered social domain, significant overlap with social support is likely to occur, which may lead to an overestimation of the relationship between these variables. Even if conceptually relevant, only weak and highly similar correlation has been noted across all the domains[^1^](#fn001){ref-type="fn"}. This indicates already well-recognized difference among perceived social support, satisfaction with social support, and their correlates and outcomes ([@ref72]). Next, although the sample reflects the gender-related prevalence of HIV infection in Europe and United States, the study could be underpowered to detect gender differences. Furthermore, since there was a tendency for higher dropout among women, a recruitment bias cannot be excluded, though it was not observed in a careful examination of the general pattern of missingness. Finally, it must be underlined that the findings are restricted only to PLWH who are formally diagnosed and under medical treatment, which is a typical characteristic for most studies with clinical samples. Despite these limitations, our study adds to the HIV/AIDS literature by investigating the heterogeneity of change in HRQoL and PSS with a special emphasis on gender differences. The present study demonstrates the complexity of dual changes and identifies groups of PLWH with mismatched HRQoL and PSS trajectories. The limited role of baseline sociodemographic and clinical characteristics of PLWH in predicting these changes was also highlighted, particularly a lack of significant differences between men and women. In fact, the findings may suggest that gender is no longer a crucial factor beyond their HRQoL and PSS change if after being diagnosed access to treatment is equal. Data Availability {#sec13} ================= The datasets for this study will not be made publicly available because although anonymized, it concerns sensitive issues (being infected with HIV). The informed consent did not include the consent to the publication of the data. Ethics Statement {#sec14} ================ All procedures performed in this study were in accordance with the ethical standards of the Research Ethics Committee of the University of Economics and Human Sciences, Warsaw, Poland, and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Informed consent was obtained from all individual participants included in the study. The protocol was approved by the Research Ethics Committee of the University of Economics and Human Sciences, Warsaw, Poland. Author Contributions {#sec15} ==================== EG and MR conceived the study, designed the study, supervised the data collection and database organization, conducted the interpretation of the data, drafted, and revised the manuscript. EG conducted the statistical analysis. EG and MR approved the submitted version of the manuscript. Conflict of Interest Statement {#sec16} ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** This study was funded by the National Science Center, Poland, research project no. 2016/23/D/HS6/02943. Open access of this article was financed by the Ministry of Science and Higher Education in Poland under the 2019--2022 program "Regional Initiative of Excellence", project number 012/RID/2018/19. ^1^We would like to thank the anonymous reviewer for this comment. The time-averaged correlations with PSS for each domain of HRQoL were as follows: 0.25 for somatic domain, 0.40 for psychological domain, 0.40 for social domain, and 0.43 for environmental domain. [^1]: Edited by: Sabrina Cipolletta, University of Padova, Italy [^2]: Reviewed by: Serena Giunta, University of Palermo, Italy; Rytis Pakrosnis, Vytautas Magnus University, Lithuania [^3]: This article was submitted to Health Psychology, a section of the journal Frontiers in Psychology

Introduction ============ Teachers in the United States face new levels of accountability due to persistent student achievement and opportunity gaps ([@B2]; [@B3]; [@B33]; [@B35]; [@B34]). Consequently, teacher preparation programs face mounting pressure to prepare effective teachers, particularly for diverse learners ([@B9]). An emerging body of research indicates that rigorous teacher evaluation increases teacher effectiveness and student achievement ([@B31]; [@B27]). Increased scrutiny of teacher evaluation has promoted an emphasis on the design of reliable and valid observation-based evaluation models that delineate the competencies of an effective teacher ([@B7]). The purpose of this study was to assess the reliability and validity of a newly developed observation-based measure of K-12 teaching proficiency called the Framework for Equitable and Effective Teaching (FEET). Since the FEET is observation-based, understanding its susceptibility to rater bias is critical. This study was designed to assess rater bias along with FEET item and scale function to find whether the new measure shows promise for general use in evaluation of apprentice teacher competencies. To support more general use, the measure should be easy to use and easy to score -- thus while a many facet Rasch model was used to assess measure psychometric quality in this study, the ideal result would show little rater bias and allow item scores to be simply added if the measure is easy to use as a summative evaluation of apprentice teacher competency. Theoretical Development of the FEET ----------------------------------- Efforts to define equitable and effective teaching have long permeated teacher education reform efforts ([@B6]; [@B5]; [@B4]). The FEET evaluation model emerges from positivist and humanist approaches to defining equitable and effective teaching. ### A Positivist Approach to Defining Effective Teaching A positivist approach promulgates a view of teaching based on the development of concrete, observable criteria that result in the enactment of measureable behaviors, or competencies, of effective teaching ([@B15]). This approach is influenced by behavioral theories of teacher learning developed by John Watson in the early 1900s ([@B21]). Current research in teacher evaluation indicates a trend toward behavioral approaches to defining and measuring effective teaching ([@B15]; [@B16]). A number of contemporary models attempt to provide behavior, or performance-based, definitions of effective teaching, including: the InTASC Model Core Teaching Standards and Learning Progressions ([@B14]), the [@B25], the Marzano Evaluation Model ([@B19]), and the Danielson Framework ([@B32]). The FEET is based on a positivist approach to defining effective teaching in the sense that competencies and indicators are defined, and a rating scale allows for quantitative measurement of proficiency. However, positivist approaches are insufficient. These lack "attention to specific local contexts, human complexity, emotion, and agency" ([@B30], pp. 214--215), indicating a need for a humanizing approach to teaching. ### A Humanist Approach to Defining Equitable Teaching [@B20] define equitable teaching as "teaching strategies and classroom environments that help students from diverse racial, ethnic, and cultural groups attain the knowledge, skills, and attitudes needed to function effectively within, and help create and perpetuate, a just, humane, and democratic society" (p. 152). Equitable teachers grasp the importance of providing diverse learners with access to values, beliefs, and ways of knowing needed to function in the dominant culture. The FEET incorporates the following: (a) integrate skills for college and career readiness; (b) set high academic expectations; (c) communicate a belief in students' capacity to achieve at high levels; (d) develop students' academic language; (e) facilitate the acquisition of content knowledge and skills through discovery, application, and higher-order thinking skills; (f) design units and lessons based on state and national content standards; and (g) implement a classroom management system that facilitates learning. Diverse learners also need to maintain and develop their cultural resources ([@B28]). The FEET model is infused with culture and prepares teacher candidates to: (a) build relationships with students and parents; (b) engage with communities; (c) incorporate multiple learning styles; (d) engage students in collaborative learning; (e) use instructional strategies to support English language learners and special needs students, (f) incorporate multicultural materials and resources; (g) develop relevant lessons that reflects the cultures of students, counteract stereotypes, and incorporate the histories and contributions of diverse populations; (h) connect content to students' background experiences, prior knowledge, skills, and/or interests; and (i) and incorporate students' native language into instruction. Instrument Development Procedure -------------------------------- Positivist and humanist theory guided the development of the FEET evaluation model. The FEET includes four dimensions of effective and equitable teaching, with rubrics using a four-level rating scale created with detailed performance indicators. The FEET is used to evaluate pre-service, or apprentice teachers, however, it is applicable to practicing teachers as well. University faculty used the FEET to evaluate pre-service teaching proficiency through classroom observations. The framework includes four teaching dimensions: Engage, Plan, Teach, and Lead. The Plan domain is not part of the observation. Each domain comprises multiple competencies. Raters assign a numerical score for each competency based on the behavior indicators in the rubric. [Table 1](#T1){ref-type="table"} shows an excerpt of FEET competency 3.1 and its associated rubric. Each competency has a separate rubric with performance indicators. The FEET development process is described below. ###### FEET rubric excerpt. Competency Unsatisfactory Indicators (1) Developing Indicators (2) Proficient Indicators (3) Advanced Indicators (4) ----------------------------- ----------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------- 3.1 Set context for lesson. • Delivers lesson without posting, previewing, or reviewing content and language objectives (CLOs). • Posts content objective only, and/or does not share objective with students during the lesson. • Posts, previews, and reviews clear, rigorous, measureable content and language objectives (CLOs). • Engages students in previewing and reviewing standards and content and language objectives (CLOs). • Begins lesson without providing a rationale for lesson. • Shares rationale for lesson that is focused on content knowledge and skills rather than big ideas relevant to students' lives. • Provides rationale that connects content to students' background experiences, prior content knowledge, skills, and/or interests. • Facilitates student development of the rationale for lesson related to big ideas and essential questions. • Lesson is disconnected from real-world application, focusing on rote skills. • Focuses lesson on content that is missing connections to real-world application, including college and career readiness. • Promotes real-world application that facilitates college and career readiness. • Engages students in making real-world connections to the content through their own lenses, and emphasizes college and career readiness. • Lacks clarity when communicating performance expectations. • Communicates performance expectations orally, although expectations are not clearly defined and/or explained in student-friendly language. • Clearly defines performance expectations orally and in writing using student-friendly language. • Clearly defines performance expectations and encourages students to provide input into performance expectations. The first phase of research was completed from 2007 to 2010 through a three-year exploratory qualitative research project. The purpose of the research was to define performance expectations for equitable and effective teaching through the design of a framework for teaching. Frameworks for teaching are commonly used observation-based evaluation models that define, assess, and develop effective teaching ([@B8]; [@B26]). The research question posed in this phase was: What are the dimensions, competencies, and indicators of equitable and effective teaching? This phase included the following procedures: (1) identify performance-based expectations for apprentice teachers; (2) determine the structure and organization of the framework; (3) develop rubrics of performance; and (4) design standardized field-based observation instruments. First, the researchers identified performance-based expectations for equitable and effective apprentice teachers. The researchers began by analyzing available standards, models, and readiness requirements for apprentice teachers entering the field. The researchers conducted purposeful selection and analysis of public documents related to models, instruments, and research on effective teaching. The data sources included: the Interstate New Teacher Assessment and Support Consortium (InTASC) Model Core Teaching Standards, the National Board for Professional Teaching Standards; two nationally recognized frameworks for teaching (i.e., Danielson Framework and Teach for American Leadership Framework); and 165 peer-reviewed journal articles related to effective and equitable teaching. The articles were selected based on targeted key words in abstracts related to teaching, including: effective, quality, culturally responsive, equitable, multicultural, linguistically responsive, and humanizing. A significant proportion of the articles, 70%, highlighted pedagogical practices that promote the academic achievement of diverse learners by building on their sociocultural resources. The researchers then analyzed and coded the data through a macro-level deductive content analysis to identify general themes of effective teaching. Subsequently, the researchers used the software, ATLAS.ti, to conduct micro-level inductive content analysis and develop open, axial, and selective coding schemes used to generate themes and sub-themes of equitable and effective teaching ([@B1]). The emerging data transformation resulted in codes by tallying the number of times concepts occurred in the textual data. This approach revealed key themes and subthemes of effective teaching that recurred across the data sources. The researchers determined how the emerging themes and subthemes would be represented as domains, competencies, and indicators based on degree of specificity. The researchers then conducted an extensive review of the dimensions, competencies, and indicators for alignment, coherence, clarity, appropriate sequence, and practical usage. Next, the researchers compared the data with literature on humanist approaches to defining effective teaching in order to strengthen the focus on equity. Last, the researchers enlisted three faculty members and 10 mentor teachers to establish the content validity of the dimensions, competencies, and indicators. This process helped to establish the FEET's relevance, representativeness, and accuracy. Second, once the performance-based expectations were defined, the researchers analyzed the structures of two national frameworks for teaching, the Danielson Framework (2007) and the Teach for America Teaching as Leadership Framework (2015). The frameworks were compared to the emerging FEET dimensions, competencies, and indicators in order to identify strengths and rectify gaps in the FEET, and provide a template for the structure and organization of the Framework. The FEET is structured in a way that moves from the simple themes related to equitable and effective teaching (e.g., dimensions), to more detailed descriptions of performances (e.g., competencies), and evidence of behaviors indicating the performances are evident (e.g., indicators). Third, once the researchers identified performance-based expectations for apprentice teachers, and determined the structure and organization of the framework, the next step was the development of rubrics of performance. Numerical rating scales are often used to quantify observations resulting in greater accuracy and objectivity of observational reports ([@B22]). The rubrics provide exemplars of performance at four levels of proficiency. Last, after developing the rubrics, the researchers developed an observation instrument to facilitate the practical implementation of the FEET, and to allow for summative and formative assessments of apprentice teachers. Raters use the rubrics to provide a quantitative assessment of apprentice teacher performance. The observation instrument is intended to be utilized by experts, or supervisors, in the field. These supervisors have the experience and understanding of the content skills and knowledge to judge an apprentice teacher's mastery level. They are raters or judges and they play a central role in rater-mediated assessments. But, raters can contribute undesired variance in ratings ([@B12]). If the variability contributed by raters is substantial it manifests in various forms of rater errors and is referred to as construct irrelevant variance ([@B10]). Although these rater errors are irrelevant to the construct, they affect ratees' performance scores. Raters can vary in terms of the severity/leniency in their ratings, consistency in ratings, and can display biases on items, subjects, or rating categories ([@B12]). These different sources of variability can be collectively addressed as rater error or rater effects. The purpose of this study was to assess the psychometric quality of the FEET. The FEET was completed by the apprentice teachers' supervisors twice per quarter during their first year of coursework and student teaching. Supervisors' ratings were analyzed in this multi-faceted study. The intent of this work was to provide direction for student supervisor training and for item revision of the instrument. Variability in (1) rater judgments, (2) item difficulty, (3) time of assessment and (4) apprentice's proficiency levels were evaluated. The research questions that directed this study were: 1. Do the items vary sufficiently in difficulty? 2. Do supervisors differ in the severity or leniency with which they rate teacher apprentice performance in teaching? 3. Do supervisors exhibit bias when using the items in the instrument? 4. Is progression over time seen with use of the measure? 5. Does the instrument provide evidence of convergent validity? Materials and Methods {#s1} ===================== Participants: Apprentices and Supervisors ----------------------------------------- ### Participants The participants in this research project included eight TEP supervisors and 59 teacher apprentices. Of the eight supervisors, or raters, four were appointed faculty members, and four were adjunct faculty. The supervisors' areas of teaching and research expertise included: urban education, cultural and linguistic diversity, bilingual education, teacher evaluation and coaching, aesthetics, and teacher renewal. Seven of the eight supervisors were White and one was Latina. Seven of the supervisors were female. The supervisors all had 3--5 years of experience evaluating teacher candidates. They had a combined expertise of 77 years teaching in K-12 schools. Two of the supervisors held doctoral degrees in education, three were doctoral candidates in education, and three held master's degrees in education. Of the 59 teacher apprentices, 23% of the students self-identified as students of color. Male participation was 38%. Instrument ---------- The FEET, described in detail above, comprises 11 items and all the items were measured using a 4-point rating scale with categories 1 = unsatisfactory, 2 = developing, 3 = proficient, and 4 = advanced. See [Supplementary Appendix A](#SM1){ref-type="supplementary-material"} for the FEET measure. The FEET administration data summarized herein are from the 2015--2016 academic year. Two validation measures were administered at the end of the program using one statewide measure and a local measure created specifically for the teacher preparation program. The instruments were the Core Competencies of Novice Teachers Survey ([@B29]) and the Teacher Education Program Satisfaction Survey (2015). Core Competency Survey (CCS) ---------------------------- The *Core Competency Survey* (CCS) ([@B29]) was administered to teacher program graduates as a self-report of teaching competencies. The instrument contains 46 statements related to eight core competencies related to effective teaching: (1) demonstrating mastery of and pedagogical expertise in content taught; (2) managing the classroom environment to facilitate learning for students; (3) developing a safe, respectful environment for a diverse population of students; (4) planning and providing effective instruction; (5) designing and adapting assessments, curriculum and instruction; (6) engaging students in higher order thinking and expectations; (7) supporting academic language development and English language acquisition; and (8) reflection and professional growth. The response scale asks participants to report how well prepared they are by their teacher education program on a 1--4 response scale with 1 = not well prepared and 4 = very well prepared. Exploratory factor analysis of the development sample yielded a dominant first factor with Cronbach's alpha for the total score exceeding 0.85 ([@B29]). The total score was used in this study. Teacher Education Preparation Satisfaction Survey (TPS) ------------------------------------------------------- Additionally, the *TEP Satisfaction Survey* (TPS) was specifically oriented to the University of Denver's Teacher Education Program. The TEP Satisfaction Survey (2015) assesses self-reported proficiency based on coursework. The 46 items relevant to candidate performance are comprised of a 22-item subscale asking for self-reported competence related to fieldwork, a 22-item subscale asking for competence related to coursework, a single item global self-rating of overall teaching competence, and a single-item related to how the candidate thought the field supervisor would rate him/her. Two adjunct faculty members and one appointed faculty member were selected to conduct an expert review of the survey content in order to establish the content validity of the survey. Each of the reviewers was a current TEP supervisor and was familiar with all aspects of TEP and the FEET. Expert reviewers assessed each survey item for relevance, difficulty, and clarity. The expert review indicated survey items were generally low in difficulty, high in relevance, and high in clarity. Reliabilities for the two multi-item subscales were 0.95 and 0.96. Procedure --------- Supervisors were assigned to conduct two observations of non-supervisees per quarter, for a total of six observations throughout the 3-quarter instructional sequence. This was in addition to observations of pre-assigned supervisees per quarter, for a total of six additional observations. In total, each supervisor completed 12 observations of teacher candidates for the 2015--2016 academic year. More than one supervisor rated each candidate in a rating scheme designed to ensure connectivity or linkage -- that is, all supervisors overlapped in rating apprentices so the data were connected. Linkage is needed between all elements of all facets so that all parameters can be estimated without indeterminacy within one frame of reference. In the rating scheme used, connectivity was adequate and there were no disconnected subsets. Observations were scheduled jointly by the supervisor and teacher candidate and occurred throughout the academic quarter at the teacher candidate's school site. Observations occurred in K-12 classroom settings and took on average of 45--60 min. The researchers designed and implemented standardized protocols and training for supervisors in order to minimize rater bias. This included the following: (a) design training protocols, including procedures, training manual, and benchmark and rangefinder videos used for scoring practice; (b) delineate protocols for candidate observations; (c) establish scoring parameters or guidelines; and (d) develop a standardized training for supervisors. The supervisors participated in the first FEET evaluation training in September of 2015. The researchers correlated candidate position on the FEET with scores from the two validation measures in order to establish convergent validity estimates. Validation measures were administered via an on-line survey given in spring 2016. Analysis -------- The Many-Facet Rasch Model (MFRM) was used to model the variability and bias in ratings. The variance in ratings that can be attributed to raters and other sources can be compared. The MFRM is an extension of the Rasch model which simultaneously calibrates all rating facets in a single common scale that can be used to estimate a person's score. The MRFM not only allows monitoring the effects of differing severity of raters but offers adjustments for systematic rater error in the ratee final score ([@B10]). The MFRM can also help to determine whether raters are exhibiting effects besides severity ([@B23], [@B24]). The bias analysis in MFRM uncovers the interaction between the rater and other facets in the rating schema. While the MFRM allows apprentice measures to be corrected for facets such as rater bias, measurement occasion, and item difficulty, the focus of the present study was to assess the impact of those facets, in particular rater bias, rather than to generate corrected scores for apprentices. That is, apprentice scores corrected for other facets in the design were not used to inform apprentice grades in this study. An MFRM was used to evaluate the apprentice teacher performance over 1 year of coursework using the FEET. In this study, a four facet Rasch model was used. The facets were: (1) apprentice, (2) item, (3) supervisor, and (4) evaluation occasion (time). The probability of an apprentice (*n*) with competence (*B*) obtaining a rating of *x* (*x* = 1, 2, 3, 4) on item *D* from supervisor C with item category difficulty *F* at time *T* (*t* = 1, 2,...,6) is expressed as the following: Log ( P n i j k l / P n i j ( k − 1 ) l ) = B n − D i − C j − F k − T l ( 1 ) The performance on each facet on the rating responses was evaluated. Empirical indices were examined for each facet to ensure that the facets were performing as intended. The assessment of individual facets helps to provide direct facet-related feedback for improvement. In total, there were eight raters assessing 59 apprentices on 11 items over 6 occasions. The FACETS (version 3.71.2, [@B18]) software was used to analyse the four-facet model. Chi-square tests, fit indices, separation ratio, and reliability of separation indicators were used to determine the performance of each individual facet. All of the statistical indicators were examined for each facet. These statistical indicators were used as evidence to draw conclusions about the quality and deficiencies of the instrument. The chi-square tests for facets, facet measures (logit), facet separation ratio, and the facet reliability indicators were examined to understand dispersion and fit of elements in a facet (e.g., raters, items). The fixed chi-square statistic (fixed effects) provides information about the heterogeneity/difference between the elements of a facet. A statistically significant chi-square result rejects the null hypothesis that elements are at the same position. The random chi-square statistic provides information about whether facet elements can be regarded as a random sample from a normal distribution. If non-significant, elements can be regarded as coming from a normal distribution sample. Separation gives the spread of the facet measures relative to the precision of those measures. Elements are similar to each other in terms of their position if this value is closer to zero. This index helps to determine if the differences are larger than random measurement error. A higher separation ratio (*G~j~*) shows greater spread of the measures. Unlike internal consistency reliability, higher "reliability" represents greater variability among the raters/supervisors. Reliability here is the variance of the rater severity measure over the measurement error. Greater variance of the rater severity indicates the presence of variability among the raters. For the remaining three facets, reliabilities is interpreted similarly to interpretation of Cronbach's alpha. Researchers generated a variable map that presents the position of all the facets in a single layout, also known as a Wright map. This is used to represent the calibration (position) of each facet in the analysis; thus, the researcher is able to make visual comparisons within and between various facets, and gain an overall understanding of the measure. The first column represents the range of the measure in logits. The facets were set to be negatively oriented except for the candidate facet. Therefore, supervisor, time, and items with negative measure means that the supervisors are lenient, candidates are rated lower, and items administered are easier. The positive measures identify supervisors who are more severe raters, candidates with higher ratings, and more difficult items. The second column corresponds to supervisor severity or leniency exercised when rating the apprentice. The third and fourth columns present the identification number assigned to the candidate and the distribution of the candidate related to teaching skill proficiency. The fifth column displays candidate proficiency across time of evaluations. The sixth column indicates the item difficulty. The variable map presents a visual representation of the individual facets and the associations between the facets. Results ======= The variance explained by the Rasch measure, an indicator of dimensionality of the item set, was 41.75%, suggesting a unidimensional construct underlying the 11 FEET items. The response scale was used appropriately, with no inversions in Rasch-Andrich thresholds or observed average, though response option 1 (unsatisfactory) was used only 1% of the time. [Table 2](#T2){ref-type="table"} lists the scale function indices. ###### Scale use indices. Rasch-Andrich Average --- ------ ---- --------------- --------- 1 30 1 -- --1.91 2 787 28 --4.57 --0.28 3 1762 62 --0.03 1.76 4 251 9 4.60 3.47 Item Facet ---------- The chi-square test statistic for items, *X*^2^(10) = 731.6, *p* \< 0.001, indicated significant differences in the item difficulties The fit statistics identified all except one of the 11 item's mean square values as fitting within a range of 0.5 to 1.5 and so were *productive of measurement* ([@B17]). Item 6 (rigorous academic talk) evidenced some misfit and would be a candidate for revision or added supervisor training. An item separation ratio of 6.45 shows the variability between the administered items. The logit measure of item difficulty ranged from a low of --1.95 (easy item) to a high of 1.06 (difficult item). Item reliability of separation (0.98) supports existence of variability in level of difficulty among the items. FEET shows the ability to identify and distinguish different levels of proficiency. Although FEET items ranged along the proficiency continuum they were generally clustered at mid-range. Thus, the items are not spread out along the entire difficulty continuum, with a range of approximately 3 logits. In general, the items need to be reviewed again to ensure that the different levels of teaching skills proficiencies are well-represented. [Table 3](#T3){ref-type="table"} provides the sample RMSE, separation, reliability, and results of the fixed and random chi-square tests for this and the other facets. [Table 4](#T4){ref-type="table"} presents item difficulty measures, standard error of the measures, and infit and outfit mean squares. [Figure 1](#F1){ref-type="fig"} is the Wright map and presents positions of all facet elements. ###### RMSE, separation, reliability of separation, and fixed and random chi-square tests by facet. Item Supervisor Time Apprentice --------------------------- -------------- -------------- -------------- -------------- RMSE 0.16 0.12 0.10 0.34 Separation 6.45 5.87 10.54 3.91 Reliability of Separation 0.98 0.97 0.99 0.94 Fixed Chi-square *p* \< 0.001 *p* \< 0.001 *p* \< 0.001 *p* \< 0.001 Random Chi-square *p* = 0.36 *p* = 0.34 *p* = 0.29 *p* = 0.57 {#F1} ###### Item difficulty measure, standard error, and fit indices. Infit Outfit -------------------------------------- ------- ------ ------- -------- 11-Demonstrate growth -1.95 0.12 0.72 0.70 10-Meet professional standards -1.64 0.12 0.75 0.72 1-Develop respectful relationships -1.04 0.12 1.03 1.01 3-Actively engage students -0.30 0.12 1.13 1.17 7-Make content and language 0.34 0.23 0.60 0.57 comprehensible 5- Facilitate rigorous learning 0.36 0.15 1.28 1.30 6-Rigorous academic talk 0.55 0.23 1.62 1.66 4-Set context for lesson 0.59 0.12 1.09 1.11 2-Equitable classroom management 0.97 0.12 1.06 1.07 8-Use formal and informal assessment 1.05 0.12 0.97 0.98 9-Differentiate instruction 1.06 0.22 0.88 0.88 SE = Standard error . Supervisor Facet ---------------- The fit statistics identified all eight supervisors' mean square values as fitting within the range of 0.5 to 1.5. The fixed chi-square, *X*^2^(7) = 270.3, *p* \< 0.001, showed that the supervisors' severity ratings were significantly different. The Rasch-kappa ([@B11]) was nearly zero (κ = 0.01). Findings from the chi-square test indicate that the supervisors did not have the same level of severity/leniency in evaluating the apprentices. The supervisor's reliability of separation (0.97) supports the presence of distinctive levels of severity/leniency among the sample of supervisors. The logit measure of supervisor severity ranged from a low of --1.57 (lenient supervisor) to a high of 0.80 (severe supervisor) (see [Table 5](#T5){ref-type="table"}). But, a closer evaluation of the levels of severity/leniency showed rater's logit position as not far from each other except for Rater 3. Since the raters showed significant differences in their logit position, the difference in the levels of severity/leniency is considered a call for further rater training in this context, especially for Rater 3. ###### Summary of supervisor measure and fit statistics. Measure SE Infit Mean Outfit Mean --------- --------- ------ ------------ ------------- Rater 3 -1.57 0.13 1.07 1.09 Rater 2 -0.50 0.11 0.92 0.92 Rater 5 -0.23 0.11 1.17 1.19 Rater 4 0.16 0.13 0.92 0.91 Rater 7 0.23 0.11 0.81 0.79 Rater 8 0.55 0.16 0.96 0.97 Rater 6 0.56 0.10 1.12 1.13 Rater 1 0.80 0.10 0.95 0.94 SE = Standard error . Time Facet ---------- The fit statistics identified all time ratings as fitting, or perhaps fitting too well for the fall observations (with fit indices \< 0.50). The fixed chi-square, *X*^2^(7) = 689.5, *p* \< 0.001, showed that the ratings were significantly different over time. The logit measure by time ranged from a low of --1.35 (post-test in spring) to a high of 1.17 (post-test in fall) (see [Table 6](#T6){ref-type="table"}). The difference in performance scores, which generally increased from the first observation in fall quarter to the last observation in spring quarter, supported the notion that (1) apprentice performance improved over the course of the year, (2) apprentices learned what observers were looking for in their performance, or (3) observers expected better performance with time and were more familiar with both the FEET tool and the apprentices. ###### Summary of time measure and fit statistics. Measure SE Infit Mean Outfit Mean ----------------- --------- ------ ------------ ------------- Pretest-Fall 1.57 0.10 0.83 0.83 Posttest-Fall 1.05 0.12 0.83 0.82 Pretest-Winter 0.27 0.10 1.05 1.06 Posttest-Winter -0.46 0.09 1.09 1.08 Pretest-Spring -1.08 0.10 1.05 1.06 Posttest-Spring -1.35 0.09 1.06 1.06 SE = Standard error . Apprentice Facet ---------------- The fit statistics identified all except one apprentice as fitting within the range of mean square fit from 0.5 to 1.5. The fixed chi-square, *X*^2^(7) = 903.1, *p* \< 0.001, showed that the performance ratings differed across apprentices. The reliability of separation (0.94) supports the presence of distinct levels of performance among the sample of apprentices. The logit measure of apprentice performance ranged from a low of --2.77 (least proficient) to a high of 2.94 (most proficient). This difference in apprentice logit positions reflects raters' ability to use the items to distinguish among apprentices' teaching skills proficiency. Rater by Item Interaction ------------------------- The objective of the bias-interaction analysis was to determine if some supervisors had specific biases for some of the items. A statistically non-significant chi-square, *X*^2^(88) = 108.6, *p* \> 0.05 indicates that raters did not differ significantly overall in using the items. The item ratings were generally invariant across the raters though there were some significant bias-interactions that explained a total of 3.93% of residual variance. The finding helps to support the quality of the items and ratings in the instrument. Validity Assessment ------------------- The Pearson correlations between scores on the FEET measure, CCS, and the TPS were calculated and are presented in [Table 7](#T7){ref-type="table"}. Visual inspection of scatterplots showed no evidence of non-linearity. The highest correlation (*r* = 0.43, *p* \< 0.01) was found between the FEET score and how students perceived they would be rated by their field supervisor. The only other statistically significant correlation was between self-reported global performance and FEET score (*r* = 0.37, *p* \< 0.01). How well-trained students perceived themselves to be as reported on the CCS was not related significantly to observer ratings of performance. These results suggest self-perceptions of teaching proficiency are statistically significantly but not strongly related to supervisor perceptions of teaching proficiency. ###### Correlations between FEET and TPS and CC scales. Instrument used for convergent validity FEET ------------------------------------------ ---------- **Teacher Performance Survey subscales** TPS Field 0.25 TPS Course 0.21 TPS Global 0.37\*\* TPS Field Supervisor Rating 0.43\*\* **Core Competency Survey** CCS total score 0.19 ∗∗ p \< 0.01 . In summary and in response to the research questions, items varied in difficulty though the construct coverage could be improved, supervisors varied significantly in rating severity though little overall bias (supervisor--item interaction) was shown, higher ratings were shown over time as apprentices progressed through the program, and some evidence of convergent validity was found though FEET ratings were clearly not strongly related to apprentice self-reports of competencies. Discussion ========== The results from item fit indices, severity/leniency of the raters, and the interaction between the raters and the items were used to assess the quality of the FEET instrument. In terms of the items, the 11 items covered a 3-logit range. One misfitting item was detected in the analysis. Bias analysis indicates that the items were generally invariant across the raters with approximately 4% of the variance explained by bias (rater/item interaction) terms. The findings from item, rater, and the interaction between the rater and items analyses showed support for the FEET as yielding reliable ratings. The objective of this research project was to investigate the psychometric properties of the FEET (e.g., scale use, fit, consistency, convergent validity); and identify implications for revising the FEET evaluation model and its effectiveness to train supervisors to evaluate apprentice teacher competencies. Overall, the supervisor, apprentice, time, and item facet analysis indicate that the FEET has adequate measurement quality, with apprentices progressing over time. Ratings of apprentices improved by nearly three logits -- a substantial change -- over the course of the program, in a coherent progression, suggesting competencies were gained over the course of the year and, more importantly for the purpose of this manuscript, were reflected by the measure. The supervisors showed a good understanding and use of the FEET evaluation instrument. There was no randomness in the way the supervisors assigned the ratings. The supervisors also showed evidence of distinguishing the apprentices' abilities and rating them at different performance levels. While the supervisor ratings were fitting, they also had significant differences in the severity of the apprentice ratings. The variability in supervisor ratings indicates a need for improved supervisor training; this may include the use of range finder videos for practice scoring, a review of scoring rubrics, and frame of reference training to an agreement criterion. If the FEET is implemented broadly and scores are used for summative purposes, it is critical to train raters to a criterion *or* to continue use of an MFRM analysis so that rater bias can be controlled in obtaining apprentice final scores. If rating severity or leniency is a trait for a particular rater, it may be difficult to remediate. The apprentices' ratings were similar to the ratings expected from the model. This suggests minimal error from the apprentice facet to the measurement model. This indicates that the majority of apprentices demonstrated proficiency in the development of teaching skills, as rated by the FEET. Moreover, their teaching proficiency increased over time. Two students were overfitting, indicating that supervisors may be overestimating or underestimating the skills of some students. Moreover, the separation reliability of items was adequate although there were few items with intermediate levels of difficulty. It is suggested that the items or scale response options potentially be revised to obtain a more diverse spread of item difficulty levels. Last, while some evidence of convergence with external measures (CCS, TPS) was found, it was clear that self-perceptions were not strongly related to supervisor perceptions of teaching proficiency. This result is similar to those found in other content areas where there is potential for the influence of perception and also clarity in construct definition that differs from self-report to observation (e.g., [@B13]). The results of the study indicate that supervisors showed adequate reliability and the FEET demonstrated adequate measurement quality, thereby indicating the success of the FEET evaluation model in assessing apprentice teacher proficiency. The results also point to specific areas of improvement for the supervisor training and FEET evaluation model, including: (a) improve supervisor training through the review of FEET rubrics and the use of a range finder video to decrease the variability of ratings among supervisors; (b) provide individual training to the most severe and most lenient supervisors; (c) examine the FEET item difficulty progression and potentially revise one item; (e) analyze the data on overfitting students to see if there are patterns or contextual factors that may have impacted apprentice ratings. At this point in its development, results suggest that the FEET may be useful for formative evaluation but for summative purposes, apprentice scores would need to be adjusted, particularly for rater severity through use of a MFRM. Future research initiatives include a second MFRM study, replication with a larger number of supervisors, and the completion of a project in which the researchers estimate the predictive validity of the FEET evaluation measure by comparing pre-service teacher summative evaluation ratings to their in-service teacher effectiveness ratings and student outcomes. This research and future research are important because the FEET can be used to prepare apprentice teachers or develop practicing teachers. Initial research on the FEET demonstrates that this model shows support for reliability and validity in the preparation and development of effective and equitable teachers. Ethics Statement ================ This study was carried out in accordance with the recommendations of the Office of Research Integrity and Education, Human Research Protection Program. The protocol was approved by the Institutional Review Board, University of Denver. Written, informed consent was obtained from all observers and apprentice teachers who served as participants in the study. Author Contributions ==================== PG consulted on data collection, conducted analyses, and wrote portions of the manuscript. MdCS obtained funding to support the project, conducted the research to develop the measure, and wrote portions of the manuscript. KG supervised the data collection, consulted on analyses, and wrote portions of the manuscript. JL participated in measure development research and wrote portions of the manuscript. Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** This project was supported by the University of Denver's PROF Grant. Supplementary Material ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fpsyg.2019.01363/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: Jason C. Immekus, University of Louisville, United States [^2]: Reviewed by: Craig Velozo, Medical University of South Carolina, United States; Jodi Nickel, Mount Royal University, Canada [^3]: This article was submitted to Educational Psychology, a section of the journal Frontiers in Psychology

1. Introduction {#sec1-molecules-23-01561} =============== *Zizania latifolia* is the sole member of the tribe Oryzeae Dum., subfamily Oryzoideae Care, family Gramineae. It is an important gramineous crop, genetically related to the *Oryza* species \[[@B1-molecules-23-01561]\]. Four *Zizania* species are found worldwide: *Zizania aquatica*, *Zizania palustris*, and *Zizania texana* in North America (U.S.A. and Canada), and *Z. latifolia* in East Asia (China, Japan, and Korea) \[[@B2-molecules-23-01561],[@B3-molecules-23-01561]\]. *Z. aquatica* and *Z. palustris* are annual species, whereas *Z. texana* and *Z. latifolia* are perennial species \[[@B3-molecules-23-01561],[@B4-molecules-23-01561]\]. The genetic diversity of the North American *Zizania* species is higher than that of *Z. latifolia* \[[@B5-molecules-23-01561]\]. The reasons for the low genetic diversity of the latter include restricted gene flow among populations, self-fertilization as a mode of reproduction, and genetic drift within populations \[[@B6-molecules-23-01561]\]. Rivers, mountains, and habitat fragmentation influence the landscape-scale genetic structure of *Z. latifolia* \[[@B7-molecules-23-01561]\]. Molecular evidence indicates that *Zizania* species likely originated in North America, thereafter spreading to East Asia and other parts of the world via the Bering Strait \[[@B8-molecules-23-01561]\]. In contrast to many review articles on the nutritional constituents and health benefits of wild rice (*Zizania* spp.) \[[@B3-molecules-23-01561],[@B4-molecules-23-01561]\], few review articles on the morphological characteristics, nutrients, and bioactive compounds of *Z. latifolia*, and health benefits of its seeds have been published, despite the fact that *Z. latifolia* is valued as a potential source of desirable phytochemicals and nutrients. Therefore, the main objective of the present review was to summarize the current state of knowledge regarding the morphological characteristics of the *Z. latifolia*, the history, present situation, nutrients, bioactive compounds, bioactivities, and health benefits of *Z. latifolia* seeds, and the nutrients and bioactive compounds from the swollen culm and aerial parts of *Z. latifolia*, and to explore the functional food and nutraceutical potential of this plant. 2. Methods Used in Searching Research Paper {#sec2-molecules-23-01561} =========================================== Our review paper mainly focuses on the morphological characteristics, nutrients, and bioactive compounds of *Z. latifolia*, and health benefits of its seeds. First, we used "*Zizania latifolia*" as the key word to search in the Web of Science, and obtain the related English research papers. Next, we used "*Zizania latifolia*" as the key word to search in the China National Knowledge Infrastructure, and obtained related Chinese research papers. Finally, because of the scarcity of research papers on the bioactive compounds of the seeds of *Z. latifolia*, we have supplemented research papers on the bioactive compounds and health benefits of the seeds of *Z. palustris* and *Z. aquatic*. 3. Morphological Characteristics of the *Z. latifolia* {#sec3-molecules-23-01561} ====================================================== As a perennial herbaceous plant, *Z. latifolia* grows to 1.5--3.0 m ([Figure 1](#molecules-23-01561-f001){ref-type="fig"}A) and its height increases with water depth \[[@B9-molecules-23-01561]\]. It exhibits high water tolerance and normally grows in clusters on wet marshlands, often with emergent plants such as *Phragmites australis* and *Typha angustifolia* \[[@B10-molecules-23-01561],[@B11-molecules-23-01561],[@B12-molecules-23-01561]\]. *Z. latifolia* produces underground as well as surface stems (rhizomes) that form multiple tillers and enable vegetative propagation ([Figure 1](#molecules-23-01561-f001){ref-type="fig"}B). Infection of *Z. latifolia* by the endophytic smut fungus *Ustilago esculenta* leads to loss of flowers and seeds, with the stem swelling at the base and forming an edible swollen culm ([Figure 1](#molecules-23-01561-f001){ref-type="fig"}C). The swollen culm is the second most-cultivated aquatic vegetable in China, known as *Jiaobai* \[[@B13-molecules-23-01561],[@B14-molecules-23-01561],[@B15-molecules-23-01561],[@B16-molecules-23-01561],[@B17-molecules-23-01561]\]. Asexual propagation via infected rhizomes is the only means of *Jiaobai* production \[[@B18-molecules-23-01561]\]. Long-term artificial selection maximizes the desirable generation of *Jiaobai* and modulates the evolution of the fungus from a plant pathogen to an entophyte \[[@B19-molecules-23-01561]\]. Similar to other endophytic fungi, *U. esculenta* infection does not cause leaf chlorosis or death \[[@B14-molecules-23-01561],[@B15-molecules-23-01561],[@B16-molecules-23-01561]\]. During the period of culm gall formation, *U. esculenta* secretes plant hormones, such as auxin and cytokinin, to stimulate gall formation \[[@B20-molecules-23-01561],[@B21-molecules-23-01561]\]. Environmental factors such as temperature, sunlight, and fertilizer use affect the swollen culm yield \[[@B16-molecules-23-01561],[@B22-molecules-23-01561]\]. *Z. latifolia* not infected by *U. esculenta* forms flowers and seeds under suitable conditions, of which the latter is collected as food \[[@B23-molecules-23-01561]\]. The inflorescence is a 30--50-cm-long panicle that branches out multiple times either upwards or sideways. The plant is monoecious, i.e., female and male flowers are present on the same branch, which are observed among the branches in the middle portion of the inflorescence ([Figure 1](#molecules-23-01561-f001){ref-type="fig"}D). The flowering times vary because of their arrangement on the branch, with the female (located above the male) flowering before the male. The seeds of *Z. latifolia* are sparsely arranged on the ear and mature at different times, after which they fall off easily \[[@B23-molecules-23-01561]\]. They can be deshelled either manually or mechanically to obtain the caryopsis, a cylindrical grain ca. 1.0--1.5 cm long and 1.0--2.0 mm wide, with tapered ends and a light brown color ([Figure 1](#molecules-23-01561-f001){ref-type="fig"}E). Compared to the control (cultivation at 20-cm water depth), cultivation at 50-cm and 100-cm water depths increases the number of inflorescences on *Z. latifolia* \[[@B9-molecules-23-01561]\], which increases the yield of the seeds of *Z. latifolia*. 4. History and Present Situation of the *Z. latifolia* Seeds (Chinese Wild Rice) {#sec4-molecules-23-01561} ================================================================================ The *Z. latifolia* seeds have been consumed as a cereal in China for \>3000 years, and are one of the earliest cultivated crops. The earliest record of the species comes from the annals of the Zhou Dynasty. '*Zhou Li*', a classical book of ancient China, lists Chinese wild rice as one of the 'six grains' (the other five being rice, broomcorn millet, panicled millet, wheat, and beans) offered to the emperor for his enjoyment \[[@B24-molecules-23-01561]\]. However, the habitat of *Z. latifolia* has changed drastically with the rapid increase of the Southern Chinese population and the drying of lakes for rice cultivation, both of which have reduced the area occupied by this species. It has gradually been outnumbered by regular rice and is consumed less frequently, mostly for relieving hunger. Currently, it is rarely cultivated as a crop and mainly survives untended in the wild. It should be noted that *Z. latifolia* is prevalent throughout China, except in Xinjiang and Tibet, and grows in lakes, ponds, rivers, marshes, and the wetlands \[[@B6-molecules-23-01561],[@B7-molecules-23-01561]\]. It is particularly abundant in the middle and lower Yangtze River basin and Huai River basin \[[@B5-molecules-23-01561],[@B25-molecules-23-01561],[@B26-molecules-23-01561]\]. A *Z. latifolia* survey of 10 lakes of the Eastern China plain over 5 consecutive years revealed growth over a 488-km^2^ area, with the generation of ca. 1.8 × 10^7^ kg seeds and 5.5 × 10^8^ kg aerial parts \[[@B26-molecules-23-01561]\]. Between 1981 and 2009, the area covered by *Z. latifolia* in the Wuchang Lake increased linearly, with 53% of the lake surface occupied by 2009; this overgrowth was mainly associated with changes in the water level \[[@B27-molecules-23-01561]\]. 5. Nutrients and Bioactive Compounds of Chinese Wild Rice {#sec5-molecules-23-01561} ========================================================= 5.1. Nutrients of Chinese Wild Rice {#sec5dot1-molecules-23-01561} ----------------------------------- The *Z. latifolia* seed (Chinese wild rice) is a type of whole grain with high nutritional value; it contains proteins, minerals, and vitamins, as well as bioactive compounds \[[@B3-molecules-23-01561],[@B4-molecules-23-01561],[@B28-molecules-23-01561]\]. According to laboratory and population-based studies, Chinese wild rice is a safe and unique tasting food, the consumption of which shows a protective effect \[[@B24-molecules-23-01561]\]. The nutrients of Chinese wild rice are shown in [Table 1](#molecules-23-01561-t001){ref-type="table"}. The vitamin (vitamin B1, vitamin B2, and vitamin E) content in Chinese wild rice from five lakes of Jiangsu Province is appreciably higher than in polished *Indica* rice from Jiangsu Province \[[@B29-molecules-23-01561]\]. In addition, it contains higher amounts of essential amino acids than white rice, barley, and maize \[[@B30-molecules-23-01561],[@B31-molecules-23-01561]\]. The amino acid score of Chinese wild rice is 66.6, which is considerably higher than that of polished *Indica* rice (51.1) \[[@B28-molecules-23-01561]\]. The first and second limiting amino acids of Chinese wild rice are originally considered to be threonine and lysine, respectively \[[@B31-molecules-23-01561]\]. However, a subsequent study by Jiang et al. \[[@B28-molecules-23-01561]\] shows that lysine and isoleucine are the first and second limiting amino acids of Chinese wild rice, respectively. The protein efficiency ratio of Chinese wild rice is 2.75, which is higher than that of enriched white flour (0.6), rice (2.18), and soybean (2.32) \[[@B29-molecules-23-01561]\]. Therefore, proteins derived from Chinese wild rice are considered to be high quality. The lipid content of North American wild rice ranges from 0.7 to 1.1%, compared to 2.7% in regular brown rice \[[@B35-molecules-23-01561]\]. The main lipids in North American wild rice are linoleic (35--37%), linolenic (20--31%), palmitic (14.1--18.4%), and oleic (12.8--16.2%) acids \[[@B35-molecules-23-01561]\]. The amount of moisture, protein, fat, ash, and dietary fiber in Chinese wild rice is similar to that in North American wild rice; however, iron, vitamin B1, and vitamin E content of Chinese wild rice is higher, whereas methionine, zinc, and vitamin B2 content is lower \[[@B30-molecules-23-01561],[@B31-molecules-23-01561]\]. The amount of vitamin B1 in Chinese wild rice is 0.52--0.63 mg/100 g, whereas that in North American wild rice is 0.36--0.50 mg/100 g; the total vitamin E content of North American wild rice is reportedly 0.2 mg/100 g lipids, whereas it is 0.48 mg/100 g lipids in Chinese wild rice \[[@B31-molecules-23-01561]\]. Although the total carbohydrate content of Chinese wild rice is lower than that in polished *Indica* rice, the levels of other components and elements are higher, including protein (2.16-fold), fat (1.54-fold), total dietary fiber (7.62-fold), and total mineral levels (1.91-fold), as well as calcium (3.32-fold), chromium (4.00-fold), copper (1.83-fold), iron (2.39-fold), magnesium (3.14-fold), manganese (1.31-fold), phosphorus (2.93-fold), potassium (2.83-fold), sodium (1.57-fold), and zinc (1.39-fold) \[[@B28-molecules-23-01561]\]. Therefore, Chinese wild rice is a rich source of minerals ([Table 1](#molecules-23-01561-t001){ref-type="table"}). 5.2. Bioactive Compounds in Chinese Wild Rice {#sec5dot2-molecules-23-01561} --------------------------------------------- In addition to nutrients, Chinese wild rice also contains bioactive compounds ([Table 1](#molecules-23-01561-t001){ref-type="table"}). The flavonoid, saponin, and phytosterol content in Chinese wild rice are 52.25-, 12.12-, and 4.29-times higher, respectively, than in white rice \[[@B28-molecules-23-01561]\]. Importantly, anthocyanins and chlorophyll, which were not detected in white rice, are present in high quantities in Chinese wild rice. Although the bioactive compounds of Chinese wild rice have not been extensively studied, relatively more data are available regarding the flavonoid, phenolic acid, cell wall hydroxycinnamate, γ-oryzanol, and phytosterol of North American wild rice \[[@B35-molecules-23-01561],[@B36-molecules-23-01561],[@B37-molecules-23-01561],[@B38-molecules-23-01561],[@B39-molecules-23-01561]\]. Among these, the total phenolic content of North American wild rice varies from 2472 to 4072 mg of ferulic acid equivalents (FAE)/kg, which is higher than that in white rice (279 mg of FAE/kg) \[[@B38-molecules-23-01561]\]. Flavonoid glycosides (diglucosyl apigenin, glucosyl-arabinosyl apigenin, and diarabinosyl apigenin) and flavan-3-ols (catechin, epicatechin, and oligomeric procyanidin) are the main antioxidants identified in North American wild rice \[[@B38-molecules-23-01561]\]. Ferulic acid (FA) is the most abundant phenolic acid (up to 355 mg/kg), followed by sinapic acid, in North American wild rice; the other monomeric phenolic acid compounds present in North American wild rice are *p*-coumaric acid, vanillic acid, syringic acid, *p*-hydroxybenzoic acid, *p*-OH-benzaldehyde, and vanillin \[[@B39-molecules-23-01561]\]. Phenolic acid dehydrodimers are cell wall-bound, and are only detectable in the methanol-insoluble fractions; they are represented by diferulic and disinapic acids \[[@B39-molecules-23-01561]\]. FA is the most abundant phenolic acid (394.2 mg/kg) among the insoluble dietary fibers in North American wild rice, although significant amounts of sinapic acid (51.8 mg/kg) and *p*-coumaric acid (14.2 mg/kg) were also detected \[[@B37-molecules-23-01561]\]. Five feruloylated oligosaccharides were isolated and identified as arabinoxylan ferulate fragments, and the new feruloylated tetrasaccharide {\[5-*O*-(*trans*-feruloyl)\]\[*O*-*β*-[d]{.smallcaps}-Xyl*p*-(1→2)\]-*O*-*α*-[l]{.smallcaps}-Ara*f*-(1 → 3)}-*O*-*β*-[d]{.smallcaps}-Xyl*p*-(1 → 4)-[d]{.smallcaps}-Xyl*p* was isolated from North American wild rice \[[@B37-molecules-23-01561]\]. North American wild rice contains more γ-oryzanol, a natural mixture of FA esters of triterpene alcohols and sterols, and an important bioactive component in rice bran oil, than regular brown rice (1352 vs. 688 mg/kg); cycloartenol ferulate is the most abundant sterol ferulate in North American wild rice, whereas 24-methylenecycloartenol ferulate is the most abundant in regular brown rice \[[@B36-molecules-23-01561]\]. There are more saturated ferulates in regular brown rice compared to North American wild rice; four additional γ-oryzanol compounds are identified as caffeates and cinnamates of cycloartenol and campesterol in North American wild rice \[[@B36-molecules-23-01561]\]. To our knowledge, North American wild rice is a source of bioactive phytosterols; the main sterols of North American wild rice found in an unsaponified fraction were: campesterol (14--52%), *β*-sitosterol (19--33%), Δ^5^-avenasterol (5--12%), and cycloartenol (5--12%) \[[@B35-molecules-23-01561]\]. Moreover, the level of omega-3 fatty acids in North American wild rice is up to 18 times higher than in brown rice \[[@B35-molecules-23-01561]\]. Anthocyanins are thought to be the major functional components in pigmented rice \[[@B40-molecules-23-01561]\], which may be the major functional components in North American wild rice \[[@B38-molecules-23-01561]\]. The bioactivities and health benefits of wild rice may be attributed to the synergistic effects of its functional components. Notably, the inhibitory effects of dietary polyphenols against α-glucosidases, α-amylases, and aldose reductases have attracted interest among researchers \[[@B41-molecules-23-01561],[@B42-molecules-23-01561],[@B43-molecules-23-01561]\]. 6. Bioactivity and Health Benefits of Chinese Wild Rice {#sec6-molecules-23-01561} ======================================================= 6.1. Antioxidant Activity {#sec6dot1-molecules-23-01561} ------------------------- Whole grains are rich in bioactive substances, such as antioxidants, which protect the body against oxidative stress \[[@B44-molecules-23-01561]\]. As a whole-grain food, Chinese wild rice contains high levels of antioxidants \[[@B28-molecules-23-01561]\], which can effectively remove free radicals from the body, and enhance the function of antioxidant systems. When incorporated into different meat products, wild rice retards lipid oxidation, with phytic acid being one of the potent antioxidants of wild rice \[[@B45-molecules-23-01561]\]. The antioxidant activity of raw *Z. palustris* seeds is 30 times higher than that of control white rice, and the total phenolic content is highly correlated with the total antioxidant activity of *Z. palustris* seeds \[[@B38-molecules-23-01561]\]. The diphenylpicrylhydrazyl (DPPH) free radical-scavenging activities of the soluble and insoluble phenolic acids suggest that the antioxidant activity of *Z. palustris* seeds is partially associated with its phenolic acid profile \[[@B39-molecules-23-01561]\]. The main contributors to the antioxidant activity of free flavonoids in *Z. aquatica* seeds are epigallocatechin, epicatechin, and rutin, and the main contributors to the antioxidant activity of free phenolic acids in *Z. aquatica* seeds are ferulic, vanillic, ellagic, sinapic, and syringic acids \[[@B46-molecules-23-01561]\]. Chinese wild rice also contains the trace elements selenium and copper, as well as vitamin E, which contribute to its enzymatic and non-enzymatic antioxidant activities \[[@B29-molecules-23-01561]\]. Consumption of Chinese wild rice prevents the accumulation of oxidative stress in rats fed a high saturated fat and cholesterol diet; this is because of increased antioxidant and superoxide dismutase activities, and reduced malondialdehyde concentration in the serum and liver \[[@B47-molecules-23-01561],[@B48-molecules-23-01561]\] ([Table 2](#molecules-23-01561-t002){ref-type="table"}). Owing to its superior antioxidant capacity, Chinese wild rice can inhibit the deleterious effects of lipid peroxidation to ameliorate chronic metabolic diseases. 6.2. Alleviation of Insulin Resistance and Lipotoxicity {#sec6dot2-molecules-23-01561} ------------------------------------------------------- Chinese wild rice is abundant in dietary fiber, resistant starch, and phytosterols, as well as polyphenols, such as flavonoids, saponins, and anthocyanins \[[@B28-molecules-23-01561]\] ([Table 1](#molecules-23-01561-t001){ref-type="table"}). Replacing white rice and processed wheat starch with Chinese wild rice as the main source of dietary carbohydrate in rats fed high saturated fat and cholesterol diet ameliorated abnormal glucose metabolism by suppressing diet-induced insulin resistance \[[@B49-molecules-23-01561],[@B51-molecules-23-01561]\]. Animal studies revealed that Chinese wild rice prevented increase in body weight and fat accumulation in insulin-resistant mice, and lowered the serum levels of glucose, insulin, and free fatty acids \[[@B49-molecules-23-01561]\]. Chinese wild rice inhibits the expression of protein tyrosine phosphatase 1B in the liver to promote the phosphorylation of insulin receptor substrate-2, thereby enhancing insulin signal transduction and reducing insulin resistance ([Table 2](#molecules-23-01561-t002){ref-type="table"}). Dietary carbohydrate replaced with Chinese wild rice reduced triglyceride and free fatty acid levels in the liver, leading to increase in serum adiponectin levels, reduction in serum lipocalin-2 and visfatin concentrations, up-regulation of the expressions of adiponectin receptor 2, peroxisome proliferator-activated receptor alpha, and gamma, and down-regulation of leptin and lipocalin-2 expression \[[@B51-molecules-23-01561]\] ([Table 2](#molecules-23-01561-t002){ref-type="table"}). Because of its low glycaemic index, Chinese wild rice alleviated insulin resistance in rats induced by high saturated fat and cholesterol diet when it was used to replace 50% of the enriched rice and flour content in the diet \[[@B50-molecules-23-01561]\]. The total dietary fiber content of Chinese wild rice is 7.62 times that of polished *Indica* rice; its resistant starch content (11.73 g/100 g) is significantly higher than that of rice flour (7.71 g/100 g) and wheat flour (7.79 g/100 g) \[[@B28-molecules-23-01561]\]. The ratios of omega-6 to omega-3 fatty acids in *Z. palustris* seeds range from 1.1 to 1.8, whereas those in regular brown rice range between 20.2 and 22.4 \[[@B35-molecules-23-01561]\]. Reports indicate the beneficial effects of the lower ratios of omega-6 to omega-3 fatty acids in North American wild rice on human blood lipid level. In rats fed high saturated fat and cholesterol diet, Chinese wild rice effectively reduced fat deposition, and lowered serum total cholesterol, triglyceride, tumor necrosis factor (TNF)-α, C-reactive protein, and free fatty acid levels; in contrast, it increased high-density lipoprotein cholesterol levels by enhancing lipid metabolism. Lipotoxicity is a metabolic syndrome that results from the accumulation of lipid intermediates in non-adipose tissue, leading to cellular dysfunction and death, and is involved in heart failure, obesity, and diabetes \[[@B56-molecules-23-01561]\]. Notably, Chinese wild rice reduced hepatic lipotoxicity, lowered blood lipid levels in rats with dysregulated lipid metabolism, and reduced low-level inflammation caused by hyperlipidaemia \[[@B52-molecules-23-01561],[@B53-molecules-23-01561]\]. It also prevented the up-regulation of sterol-regulatory element binding protein-1c, fatty acid synthase, and acetyl-CoA carboxylase, and down-regulated lipoprotein lipase and adipose triglyceride lipase induced by high saturated fat and cholesterol diet \[[@B54-molecules-23-01561]\] ([Table 2](#molecules-23-01561-t002){ref-type="table"}). Therefore, Chinese wild rice belongs to foods of low glycaemic index, which can alleviate insulin resistance and lipotoxicity. 6.3. Cardiovascular Disease Prevention {#sec6dot3-molecules-23-01561} -------------------------------------- Soluble dietary fiber can reduce blood glucose levels and bind to bile acid for excretion in feces, thereby lowering cholesterol levels and reducing the risk of heart disease, atherosclerosis, and other conditions associated with high cholesterol levels \[[@B52-molecules-23-01561]\]. Owing to its high resistant starch content, Chinese wild rice lowers cholesterol and blood lipid levels, which is associated with a reduced incidence of cardiovascular diseases, including coronary heart disease and atherosclerosis \[[@B57-molecules-23-01561]\]. Consumption of *Z. palustris* seeds appreciably reduced the size and severity of atherosclerotic lesions in the aortic roots of male and female low-density lipoprotein receptor-knockout (LDLr-KO) mice, by 71% and 61%, respectively, compared to the gender-matched controls \[[@B58-molecules-23-01561]\]. The phytosterol content of Chinese wild rice (71.28 mg/100 g) was 4.29 times higher than that of polished *Indica* rice \[[@B28-molecules-23-01561]\]. Consumption of 60% (*w*/*w*) *Z. palustris* seeds in combination with 2% (*w*/*w*) phytosterols appreciably reduced the size and severity of atherosclerotic lesions in the aortic roots of LDLr-KO mice compared to the control group \[[@B59-molecules-23-01561]\]. This effect was associated with an appreciable reduction of total plasma, low-density lipoprotein, and very low-density lipoprotein cholesterol levels, as well as increase in fecal cholesterol excretion. Chinese wild rice lowered blood cholesterol concentration in rats with high lipid levels, and inhibited the formation of atherosclerotic plaques and the occurrence of fatty liver, while simultaneously increasing antioxidant activity in serum and suppressing inflammation associated with atherosclerosis \[[@B55-molecules-23-01561]\] ([Table 2](#molecules-23-01561-t002){ref-type="table"}). When compared to white rice, *Z. palustris* seeds inhibit monocyte adhesion to the aorta, atherosclerosis, and the concentration of inflammatory and fibrinolytic regulators in the cardiovascular tissue of LDLr-KO mice \[[@B60-molecules-23-01561]\]. The anti-atherosclerotic effect of wild rice in LDLr-KO mice may be associated with its inhibition of monocyte adhesion and inflammatory modulators \[[@B60-molecules-23-01561]\]. 7. Nutrients and Bioactive Compounds of the Swollen Culm of *Z. latifolia* {#sec7-molecules-23-01561} ========================================================================== The swollen culm, induced by the smut fungus *U. esculenta*, is consumed as an important aquatic vegetable in China, Japan, and other Asian countries \[[@B13-molecules-23-01561],[@B14-molecules-23-01561],[@B15-molecules-23-01561],[@B16-molecules-23-01561],[@B61-molecules-23-01561],[@B62-molecules-23-01561]\]. Nutrients in the swollen culm of *Z. latifolia* are shown in [Table 1](#molecules-23-01561-t001){ref-type="table"}. Qian et al. \[[@B32-molecules-23-01561]\] determined the nutrient content of the swollen culm of *Z. latifolia* to be as follows: moisture, 92.001 g/100 g fresh weight (FW); protein, 2.018 g/100 g dry weight (DW); fat, 2.258 g/100 g FW; starch, 1.429 g/100 g DW; total soluble sugars, 35.719 g/100 g DW; reducing sugars, 30.854 g/100 g DW; ascorbic acid, 0.60 g/100 g DW; polyphenols, 1.050 g/100 g FW; ash, 0.531 g/100 g DW; total dietary fiber, 4.220 g/100 g FW; soluble dietary fiber, 0.700 g/100 g FW; insoluble dietary fiber, 3.510 g/100 g FW; lignin, 0.567 g/100 g FW. Another study reported the following nutrient composition of the swollen culm of *Z. latifolia*: carbohydrate, 13.82%; protein, 8.13%; fat, 1.0%; lignin, 7.0%; sodium, 0.02%; iron, 0.85 ppm; magnesium, 3.34 ppm; copper, 0.12 ppm; zinc, 4.71 ppm; and arsenic, 0.09 ppm \[[@B63-molecules-23-01561]\]. These reported differences between nutrients in the swollen culm of *Z. latifolia* may be associated with crop variety, growth environment, cultivation techniques, and harvest time. Various compounds (**1**--**10**, [Figure 2](#molecules-23-01561-f002){ref-type="fig"}) have been isolated from the swollen culm of *Z. latifolia* \[[@B64-molecules-23-01561],[@B65-molecules-23-01561],[@B66-molecules-23-01561]\], including osteoclast-inhibiting compounds **1** (white crystal) and **2** (colorless amorphous crystal). The combination of **1** (25 μg/mL, 53 μM) and **2** (25 μg/mL, 58 μM) reduced the respective number of TRAP-(+) multinucleated cells to 49% and 19% without cytotoxicity \[[@B65-molecules-23-01561]\]. Makomotine A (**3**, white amorphous powder), makomotine B (**4**, yellow amorphous powder), makomotine C (**5**, white amorphous powder), makomotine D (**6**, colorless oil), and two other compounds (**7**, **8**) were also isolated from the swollen culm; 25 μg/mL of **3** inhibited osteoclast formation without cytotoxic effects in vitro \[[@B64-molecules-23-01561]\]. The [l]{.smallcaps}-glucoside isomer (**10**) of makomotindoline (**9**, colourless amorphous solid) reduced the number of rat glioma cells after 24 h when applied at a concentration of 10 μM (92.4 ± 9.1%) and 100 μM (83.9 ± 8.0%) \[[@B66-molecules-23-01561]\]. Water- and alkali-extractable polysaccharides (ZLPs-W and ZLPs-A, respectively) were recently extracted from the swollen culms of *Z. latifolia* \[[@B67-molecules-23-01561]\]. Both compounds exhibit radical-scavenging activity; the half-maximal effective concentrations of the DPPH, superoxide radical-, and hydroxyl radical-scavenging activities of ZLPs-A were 1.87, 1.13, and 0.38 mg/mL, respectively, whereas the corresponding values for ZLPs-W were 2.95, 3.99, and 0.5 mg/mL, respectively. ZLPs-W show no cytotoxicity, and their immunomodulatory activity is higher than that of ZLPs-A, as indicated by higher phagocyte stimulation and nitric oxide (NO) production in RAW 264.7 macrophages \[[@B67-molecules-23-01561]\]. ZLPs-W were further separated into three purified polysaccharides (ZLPs-W1, ZLPs-W2, and ZLPs-W3), which effectively enhanced the proliferation, phagocytosis, and NO production of RAW 264.7 macrophages, suggesting that they possess a potent immunostimulatory activity and might be developed as immunomodulators in drugs or foods \[[@B68-molecules-23-01561]\]. 8. Nutrients and Bioactive Compounds of the Aerial Parts of *Z. latifolia* {#sec8-molecules-23-01561} ========================================================================== Although the aerial part of *Z. latifolia* is not eaten directly as food, several studies have reported the main nutrients contained therein. Qian et al. \[[@B32-molecules-23-01561]\] reported the following nutrients in the leaf sheath: moisture, 870.00 g/kg FW; protein, 11.71 g/kg DW; fat, 18.10 g/kg FW; starch, 25.86 g/kg DW; total soluble sugars, 220.53 g/kg DW; reducing sugars, 182.73 g/kg DW; ascorbic acid, 0.01 g/kg DW; polyphenols, 6.30 g/kg FW; ash, 10.10 g/kg DW; total dietary fiber, 106.20 g/kg FW; soluble dietary fiber, 28.90 g/kg FW; insoluble dietary fiber, 77.40 g/kg FW; lignin, 8.19 g/kg FW. Another study reported the biomass chemical composition of the aerial parts of *Z. latifolia* to be as follows: cellulose, 28.5%; hemicellulose, 12.9%; lignin, 31.0%; ash, 3.0%; extractives, 18.9% \[[@B2-molecules-23-01561]\]. Considering their high protein, cellulose, hemicellulose, and lignin content, the aerial parts of *Z. latifolia* could be used as animal feed, culture material for edible mushrooms, feed stock for bioethanol production, and activated carbon \[[@B2-molecules-23-01561],[@B32-molecules-23-01561],[@B69-molecules-23-01561]\]. The aerial parts of *Z. latifolia* have been shown to exert anti-fatigue, anti-inflammatory, and anti-allergic effects \[[@B70-molecules-23-01561],[@B71-molecules-23-01561],[@B72-molecules-23-01561],[@B73-molecules-23-01561],[@B74-molecules-23-01561]\]. A methanol extract of the aerial parts of *Z. latifolia* inhibits compound 48/80-induced degranulation, antigen-induced β-hexosaminidase release, and PMA plus A23187-induced TNF-α production in RBL-2H3 mast cells \[[@B71-molecules-23-01561]\]. In contrast, the chloroform fraction of the plant tissue suppresses allergic inflammatory response by blocking the release of β-hexosaminidase and TNF-α from RBL-2H3 cells stimulated with dinitrophenyl and bovine serum albumin, and inhibiting the expression of cyclooxygenase-2 and activation of the mitogen-activated protein kinase \[[@B74-molecules-23-01561]\]. Tricin (**11**, pale yellow powder), salcolin A (**12**, yellow amorphous powder), salcolin B (**13**, yellow amorphous powder), salcolin C (**14**, yellow amorphous powder), and salcolin D (**15**, yellow amorphous powder) were recently isolated from the aerial parts of *Z. latifolia* ([Figure 3](#molecules-23-01561-f003){ref-type="fig"}). Compounds **12**--**15** are tricin derivatives with more potent anti-inflammatory and anti-allergic activities than tricin (**11**) \[[@B72-molecules-23-01561]\]. In particular, compound **15** strongly inhibits lipopolysaccharide-induced NO production in RAW 264.7 cells, as well as β-hexosaminidase release in IgE-sensitized RBL-2H3 cells. Other compounds isolated from the aerial parts of *Z. latifolia* include tricin-7-*O*-β-[d]{.smallcaps}-glucopyranose (**16**, pale yellow powder), tricin-4′-*O*-(threo-β-guaiacylglyceryl) ether 7-*O*-β-[d]{.smallcaps}-glucopyranose (**17**, yellow amorphous powder), tricin-4′-*O*-(erythro-β-guaiacylglyceryl) ether 7-*O*-β-[d]{.smallcaps}-glucopyranose (**18**, yellow amorphous powder), tricin-4′-*O*-(threo-β-guaiacylglyceryl) ether 7″-*O*-β-[d]{.smallcaps}-glucopyranose (**19**, yellow amorphous powder), and tricin-4′-*O*-(erythro-β-guaiacylglyceryl) ether 7″-*O*-β-[d]{.smallcaps}-glucopyranose (**20**, yellow amorphous powder) \[[@B73-molecules-23-01561]\] ([Figure 3](#molecules-23-01561-f003){ref-type="fig"}). To develop and use these bioactive compounds, they must be isolated from the aerial parts of *Z. latifolia*. 9. Conclusions {#sec9-molecules-23-01561} ============== Chinese wild rice contains proteins, minerals, vitamins, and bioactive substances. It exhibits various bioactivities, including antioxidant activity. A number of health benefits are associated with its consumption, e.g., alleviation of insulin resistance and lipotoxicity, and protection against cardiovascular disease. Therefore, Chinese wild rice may be used to prevent and treat metabolic disease, such as obesity, diabetes, and cardiovascular diseases. However, no human dietary intervention studies have been performed, although studies where human subjects will be administered realistic doses of Chinese wild rice as a part of their daily diet are required. Although considerable progress has been made toward understanding the bioactivities and health benefits of Chinese wild rice, relatively fewer studies focus on the bioactivities and structure-activity relationship of the isolated compounds. Consequently, future studies will focus on the isolation of compounds with high bioactivities and stable chemical structures, and elucidate their structure-activity relationship. Compounds from the swollen culm of *Z. latifolia* have been shown to suppress osteoclast formation and inhibit the growth of glioma cells. Their role as potential antioxidants and immunomodulators for use in drugs or food products have also been investigated. Furthermore, compounds from the aerial parts of *Z. latifolia* can be used to treat fatigue, inflammation, and allergic reaction. **Sample Availability:** Samples of the compounds are not available from the authors. N.Y. and Z.Z. initiated the writing of this review and designed the structure of this review, interpreted results, and drafted the manuscript. Y.D., X.L., C.C., J.S., H.Z., and Y.L. compiled information and contributed to the revision of the manuscript. This work was supported by the Science Foundation for Young Scholars of Tobacco Research Institute of Chinese Academy of Agricultural Sciences (2018A01), the Fundamental Research Funds for Central Non-Profit Scientific Institution (1610232018003), and the Agricultural Science and Technology Innovation Program (ASTIP-TRIC05). The authors declare no conflicts of interest. {#molecules-23-01561-f001} {#molecules-23-01561-f002} {#molecules-23-01561-f003} molecules-23-01561-t001_Table 1 ###### Nutrients and bioactive compounds of the seed and swollen culm of *Z. latifolia*. Unit per 100 g Seed \[[@B28-molecules-23-01561],[@B31-molecules-23-01561]\] Swollen Culm \[[@B32-molecules-23-01561],[@B33-molecules-23-01561],[@B34-molecules-23-01561]\] ------------------------- -------------------------------------------------------------- ------------------------------------------------------------------------------------------------ Moisture (g) 9.51 ± 0.13 92.001 ± 0.90 Protein (g) 13.30 ± 1.38 2.818 ± 0.019 Fat (g) 1.08 ± 0.12 2.258 ± 0.071 Total carbohydrates (g) 73.18 ± 1.68 2.43 ± 0.19 Total dietary fibre (g) 7.24 ± 1.29 4.22 ± 0.006 Total minerals (g) 1.30 ± 0.04 0.531 ± 0.001 Vitamin B1 (mg) 0.59 ± 0.04 0.02 Vitamin B2 (mg) 0.11 ± 0.03 -- Vitamin E (mg) 0.29 ± 0.11 0.99 Alanine (g) 0.69 ± 0.07 0.06 ± 0.008 Arginine (g) 1.13 ± 0.13 0.04 ± 0.005 Aspartic acid (g) 1.19 ± 0.15 0.15 ± 0.028 Cysteine (g) 0.37 ± 0.02 -- Glutamic acid (g) 2.40 ± 0.24 0.11 ± 0.010 Glycine (g) 0.59 ± 0.06 0.04 ± 0.005 Histidine (g) 0.38 ± 0.03 0.05 ± 0.005 Isoleucine (g) 0.50 ± 0.05 0.04 ± 0.003 Leucine (g) 0.95 ± 0.10 0.07 ± 0.007 Lysine (g) 0.66 ± 0.07 0.07 ± 0.007 Methionine (g) 0.28 ± 0.01 0.01 ± 0.000 Phenylalanine (g) 0.65 ± 0.08 0.04 ± 0.005 Proline (g) 0.37 ± 0.05 0.04 ± 0.005 Serine (g) 0.66 ± 0.07 0.06 ± 0.007 Threonine (g) 0.44 ± 0.04 0.05 ± 0.007 Tryptophan (g) 0.21 ± 0.03 -- Tyrosine (g) 0.44 ± 0.05 0.03 ± 0.003 Valine (g) 0.70 ± 0.08 0.05 ± 0.005 Calcium (mg) 23.74 ± 0.47 4.00 Chromium (mg) 0.12 ± 0.03 -- Copper (mg) 0.22 ± 0.13 -- Iron (mg) 2.80 ± 0.27 0.40 Magnesium (mg) 114.74 ± 7.21 8.00 Manganese (mg) 1.34 ± 0.11 0.49 Phosphorus (mg) 291.20 ± 37.64 36.00 Potassium (mg) 218.47 ± 11.06 209.00 Sodium (mg) 4.48 ± 0.87 5.80 Zinc (mg) 2.40 ± 0.20 0.33 Flavonoids (mg) 352.00 ± 3.12 383.70 Saponins (mg) 354.11 ± 22.70 -- Anthocyanins (mg) 258.00 ± 17.31 -- Chlorophyll (mg) 108.40 ± 2.41 -- Phytosterols (mg) 71.28 ± 8.12 -- molecules-23-01561-t002_Table 2 ###### Bioactivity and health benefits of *Z. latifolia* seeds (Chinese wild rice). ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Bioactivity or Health Benefit Putative Functional Compounds Potential Mechanism Reference ---------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------- Antioxidant activity Vitamins, minerals, phytosterols, phenolic acid, phytic acid, flavonoids, saponins, anthocyanins, and chlorophyll Suppressing oxidative stress by increasing antioxidant capacity and superoxide dismutase activity, and reducing malondialdehyde concentration \[[@B47-molecules-23-01561],[@B48-molecules-23-01561]\] Alleviation of insulin resistance and lipotoxicity Dietary fibre, resistant starch, vitamins, minerals, polyunsaturated fatty acids, phytosterols, phenolic acids, flavonoids, saponins, anthocyanins, and chlorophyll \(1\) Improving glucose metabolism and insulin sensitivity by inhibiting the expression of protein tyrosine phosphatase 1B and enhancing the expression of insulin receptor substrate-2 in the liver\ \[[@B49-molecules-23-01561],[@B50-molecules-23-01561],[@B51-molecules-23-01561],[@B52-molecules-23-01561],[@B53-molecules-23-01561],[@B54-molecules-23-01561]\] (2) Decreasing triglyceride and free fatty acid levels in a liver homogenate; increasing adiponectin, and decreasing lipocalin-2 and visfatin concentrations in the serum\ (3) Increasing the expression of adiponectin receptor 2, and peroxisome proliferator-activated receptors alpha and gamma; inhibiting the expression of leptin and lipocalin-2\ (4) Improving lipid metabolism and liver lipotoxicity, and suppressing low-grade inflammation induced by a high saturated fat and cholesterol diet in hyperlipidaemic rats\ (5) Suppressing increase in lipid droplet accumulation, and free fatty acid and leptin concentrations; reducing lipoprotein lipase and adipose triglyceride lipase levels\ (6) Inhibiting expression of sterol-regulatory element binding protein-1c, and fatty acid synthase and acetyl-CoA carboxylase induced by high saturated fat and cholesterol diet Cardiovascular disease prevention Dietary fibre, resistant starch, vitamins, phytosterols, phenolic acids, flavonoids, saponins, anthocyanins, and chlorophyll \(1\) Reducing blood lipid levels in rats fed high-fat diet, and inhibiting the formation of atherosclerotic plaques and the occurrence of fatty liver\ \[[@B55-molecules-23-01561]\] (2) Enhancing antioxidative capacity and preventing atherosclerosis ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Introduction ============ Despite the undoubted benefits that endocrine therapies have brought for breast cancer patients in terms of increased survival, *de novo*and acquired resistance to such treatments presents a major clinical problem; not all patients with oestrogen-receptor (ER) positive disease benefit and a significant number of initially-responsive patients ultimately relapse on such treatments \[[@B1]\]. The selective ER modulator tamoxifen has been the mainstay of therapy for almost two decades, and much has been learned about acquired resistance to this anti-oestrogen. To date, mechanistic studies have revealed important roles for growth factor signalling pathways such as those regulated by the epidermal growth factor receptor (EGFR) and human epidermal growth factor receptor (HER) 2, as contributors to endocrine resistance \[[@B2]\]. Significantly, in addition to antagonising oestrogen (E2)-regulated gene expression, tamoxifen can promote the re-expression of E2-repressed genes and, importantly, regulate the expression of a unique subset of E2-independent genes \[[@B3]\]. The consequences of such events are only now becoming clear, with recent data suggesting that the ability of selective ER modulators, such as tamoxifen, and the steroidal anti-oestrogen, fulvestrant, to induce expression of signal transduction genes normally repressed by oestrogen/ER signalling may play an important role in the ability of breast cancer cells to evade their growth inhibitory effects \[[@B4],[@B5]\]. Moreover, such treatments may modulate the expression of genes associated with an adverse cell behaviour; for example, in ER-positive breast cancer cells, tamoxifen has been reported to increase expression of 14-3-3, a marker of poor prognosis in breast cancer patients \[[@B6]\]. In addition to their genomic effects, selective ER modulators may also exert non-genomic effects on target cells; for example, tamoxifen has been demonstrated to induce activation of mitogen-activated protein kinase (MAPK) \[[@B7]\], focal adhesion kinase (FAK) \[[@B8]\] and Src \[[@B8],[@B9]\], signalling elements frequently linked to tumour migration and invasion \[[@B10],[@B11]\]. Interestingly, Src kinase is also implicated in limiting the response of tamoxifen, where it stimulates the weak AF-1 function of the tamoxifen-ER complex through its tyrosine kinase activity \[[@B12]\]. Furthermore, in 3Y1 rat fibroblasts, which overexpress Src kinase, tamoxifen cooperates with Src to cause cellular transformation through induction of DNA synthesis and anchorage-independent cell proliferation \[[@B13]\]. E-cadherin is an intercellular adhesion protein important for maintenance of cell-cell adhesion and tissue integrity \[[@B14]\] and much evidence links alterations in its expression with the advent of invasive growth in epithelial tumours \[[@B15]\]. The functional disruption of E-cadherin using monoclonal antibodies can promote Src-dependent cellular invasion \[[@B16]\] and conversely, Src activation has been demonstrated to be essential to the progression of early diffuse gastric tumours, where it is associated with a loss of E-cadherin and the development of local invasion \[[@B17]\]. Significantly, in breast cancer cells, the presence of a functional ER has been shown to be necessary for expression of E-cadherin \[[@B18]\], thus suggesting that its expression and subsequent cell-cell adhesion, may be modulated by anti-oestrogens and may have a bearing on the invasive growth of tumours. In the current report, we propose a further mechanism that may provide a route for progression of breast cancer in the face of ER inhibition with anti-oestrogens. We have studied the effects of tamoxifen on MCF-7 cells deficient in E-cadherin-mediated homophilic interactions and demonstrate that in a cellular context of reduced intercellular adhesion, tamoxifen promotes an invasive cell phenotype through a process involving the activation of Src kinase. Although also noted with the steroidal anti-oestrogen fulvestrant, this effect of invasion in tamoxifen treated cells was not seen under conditions of oestrogen withdrawal. This evidence suggests that, *in vivo*, a loss of E-cadherin may predict a poor patient outcome on tamoxifen therapy, and may account for some of the additional benefits seen with aromatase inhibitors. Materials and methods ===================== Cell culture ------------ Tamoxifen-responsive, wild-type MCF-7 cells (wtMCF-7) were routinely cultured in RPMI medium (Invitrogen, Paisley, UK) supplemented with 5% foetal calf serum (FCS), antibiotics (10 IU/ml penicillin and 10 μg/ml streptomycin), 2.5 μg/ml fungizone, 200 mM glutamine and incubated at 37°C in a 5% carbon dioxide atmosphere. For experimental analysis, the medium was changed to experimental medium, which is oestrogen depleted, (phenol-red-free RPMI containing 5% charcoal-stripped, steroid-depleted FCS, glutamine and antibiotics as above) for 24 hours before undertaking the relevant assays as described below. These experimental conditions were maintained for oestrogen withdrawal studies, although for anti-oestrogen and hormone treatments the culture medium was supplemented with 10^-7^M 4-hydroxy-tamoxifen (\'tam\'), 10^-9^M oestrogen (\'E2\') or 10^-7^M fulvestrant (\'Faslodex\', AstraZeneca, UK). All tissue culture media and constituents were obtained from Life Technologies Europe Ltd (Paisley, UK) and tissue culture plasticware was obtained from Nunc (Rosklide, Denmark). Antibodies and reagents ----------------------- The antibodies used were as follows: anti-phospho Src kinase (Y418) from Cell Signalling Technologies (New England Biolabs, Herts, UK); pan-Src antibody from Biosource (Invitrogen, Paisley, UK); glyceraldehyde 3-phosphate dehydrogenase (GAPDH) antibody from ABCAM (Cambridge UK); an anti-E-cadherin antibody with neutralising activity (HECD-1) was purchased from R & D Systems Ltd. (Oxford, UK). For immunofluorescence microscopy, the anti-E-cadherin antibody, SHE-78, (Invitrogen, UK) was used. siRNA-mediated suppression of E-cadherin in MCF-7 cells ------------------------------------------------------- E-cadherin protein expression was suppressed by siRNA transfection of MCF-7 cells as follows: SMARTpool siRNA against human E-cadherin gene (CDH-1) was obtained from Dharmacon Ltd (Perbio Science UK Ltd., Northumberland, UK) and used according to the manufacturer\'s instructions. Briefly, MCF-7 cells were seeded into six-well plates at 5 × 10^5^cells/well in antibiotic-free medium with or without anti-hormone where appropriate. After 24 hours culture, the medium was replaced with fresh, antibiotic-free medium or medium containing transfection lipid, 100 nM non-targeting siRNA control or 100 nM SMARTpool siRNA specific for CDH1 (E-cadherin). Cells were assayed for E-cadherin protein expression after 24, 48 and 72 hours post-transfection by Western blotting to confirm protein knockdown. For invasion assays and Western blotting analysis, cells were treated with CDH1 siRNA for 72 hours before performing the experiments in the presence or absence of the agents as detailed. Basement membrane invasion assay -------------------------------- Cell invasion was determined using invasion chambers possessing 8 μm porous membranes (BD Biosciences, Oxford, UK) coated with Matrigel (0.4 μg/ml). Cells (treated as above) were seeded into the chambers (10^5^cells/well) with or without anti-hormone and 600 μl of medium was added to the outside of the well. Inserts were cultured at 37°C in a tissue culture incubator for 48 hours, after which the non-invasive cells and Matrigel were removed from the inside of the wells with a cotton swab. After fixing the invaded cells with 3.7% formaldehyde, the porous membranes were removed form the invasion chamber using a scalpel blade and mounted onto a glass microscope slide using Vectashield (Molecular Probes, Eugene, OR, US) containing the nuclear stain 4\',6-diamidino-2-phenylindole. Cell invasion was quantified by viewing five separate fields per membrane at a magnification of × 20 and counting the number of cells in each field. Data were then plotted at mean cells per field ± SD for a minimum of three independent experiments, each performed in triplicate. Cell lysis and Western blotting ------------------------------- After cell cultures were treated as described above, the cells were washed twice with ice-cold PBS and lysed in lysis buffer (50 mM Tris, pH 7.5, 5 mM ethylene glycol tetraacetic acid, 150 mM sodium chloride and 1% Triton X100) containing protease inhibitors (2 mM sodium orthovanadate, 20 mM sodium fluoride, 1 mM phenyl-methylsulfonyl fluoride, 20 μM phenylarsine, 10 mM sodium molybdate, 10 μg/ml leupeptin and 8 μg/ml aprotinin). The lysates were placed on ice for 20 minutes with intermittent mixing and clarified by centrifugation (10 minutes, 13,000 rpm, 4°C). The concentration of solubilised proteins was then determined using the DC protein assay kit (BioRad, Hemel Hempstead, UK). From these lysates, 50 μg of total protein was separated by SDS-PAGE using 8% gels and transferred to nitrocellulose membranes by electroblotting. Membranes were subsequently blocked with 5% (w/v) milk protein in Tris-buffered saline containing 0.05% Tween-20. Blots were then incubated with primary antibodies as indicated, washed in Tris-buffered saline containing 0.05% Tween-20 and incubated with horseradish peroxidase-conjugated secondary antibodies. An enhanced chemiluminescence system (\'West Dura\' reagent, Pierce and Warriner Ltd, Chester, UK) was used for subsequent detection of bound antibodies and the blots exposed to X-ray film (Kodak, UK). Blots shown are representative of a minimum of three separate experiments. Each blot was scanned using a densitometer in order to obtain data for statistical analysis. Immunofluorescent staining -------------------------- Cells were cultured on eight-well chamber slides and allowed to reach log phase growth before being treated with CDH1 siRNA ± tamoxifen as above for a further period of 72 hours. Cells were then fixed with 3.7% formaldehyde, permeablised using 0.05% TritonX-100 and blocked with 10% normal goat serum for 30 minutes. Staining for E-cadherin was carried out using 1 μg/ml SHE78 antibody for 60 minutes followed by anti-mouse immunoglobulin (Ig) G:fluorescein isothiocyanate conjugate (Molecular Probes, Ugene, Oregon, USA) at 1:2000 for 30 minutes. Co-staining for actin was performed by incubating the cells with phalloidin (Invitrogen, Carlsbad, California, USA), in PBS containing 1% BSA. Cells were then washed and mounted onto microscope slides using a hard-set mounting medium (Vectashield, Molecular Probes, Invitrogen, Carlsbad, California, USA). Cells were viewed at a magnification of × 63 with an oil-immersion objective and representative cells photographed. Immunocytochemical staining --------------------------- Log-phase cells, grown on glass cover slips in experimental medium, were left untreated (control) or treated with tamoxifen (10^-7^M) or oestrogen (10^-9^M) for four days. Cells were fixed in 2% paraformaldehyde/vanadate for 20 minutes before being washed three times for five minutes with PBS. After blocking with PBS-Tween (0.02% v/v), cells were incubated in primary antibody overnight followed by a peroxidise-labelled secondary antibody (DAKO enVision system (DAKO UK Ltd, Ely, Cambridge, UK) for one hour. After washing, bound antibodies were detected using DAB substrate with counterstaining using 0.02% methyl green solution. Stained cells were rinsed, air dried and mounted onto glass slides using a xylene soluble mounting medium. Cell growth assay ----------------- Cells were seeded into 24-well plates before treating with CDH1 siRNA in the presence or absence of tamoxifen and a range of inhibitors, as described, for 72 hours. After this time, wells were gently washed and cells fixed with formaldehyde (3.7% in PBS). The numbers of cells in each well were then determined using a coulter counter. Data was obtained as cell counts (mean of three separate wells) for each treatment, with experiments performed in duplicate. Statistical analysis -------------------- Statistical significance was determined with paired student\'s t-test of the data using MiniTab14. Significance was observed at p ≤ 0.05. Results ======= Tamoxifen, but not E2-withdrawal, promotes invasion of MCF-7 cells in the absence of intercellular contacts ----------------------------------------------------------------------------------------------------------- The capacity of ER-positive, MCF-7 breast cancer cells to invade through Matrigel in response to tamoxifen or oestrogen withdrawal (-E2) was determined and compared with E2 alone. MCF-7 cells are poorly invasive *in vitro*\[[@B19]\] and their invasive capacity was not significantly affected by -E2, while tamoxifen alone had only a minor effect. Interestingly, a non-significant trend emerged, with tamoxifen treatment resulting in increased cell invasion compared with -E2, which in turn caused more invasion than in cells exposed to E2 (Figure [1a](#F1){ref-type="fig"}). {#F1} Whereas neutralising E-cadherin function using the HECD-1 antibody promoted a modest increase in cell invasion, invasion in response to combined treatment (tamoxifen + the antibody) was significantly higher (figure [1b](#F1){ref-type="fig"}). The use of this antibody in combination with either E2 or under E2-withdrawal conditions (\'-E2\'), however, did not promote a increase in invasion (Figure [1b](#F1){ref-type="fig"}). These changes were not due to changes in cell proliferation, because cell proliferation rates were unchanged over the short timescale of the experiment (data not shown). To further confirm that tamoxifen could induce cellular invasion in the absence of intercellular adhesion, we transfected MCF-7 cells with siRNA for the CDH1 (E-cadherin) gene. Cells treated in this way displayed a significant loss of E-cadherin protein, an effect maintained over a period of 72 hours (Figure [2a, b](#F2){ref-type="fig"}) in contrast to cells treated with either the siRNA delivery lipid or a non-targeting siRNA (\'L\' and \'NT\' respectively). Interestingly, these data appeared to show that tamoxifen increased E-cadherin expression compared with -E2. However, a caveat to this was that these samples were run on separate gels. Thus, to confirm the effects of endocrine manipulation on E-cadherin expression, immunoblots were performed using samples treated with -E2, +E2 or tamoxifen. These data showed that none of the treatments tested had a significant effect on E-cadherin expression (Figure [2c](#F2){ref-type="fig"}). {#F2} siRNA effects were further confirmed with immunofluorescence analysis of E-cadherin expression. Whereas strong staining for E-cadherin could be detected in untreated MCF7 cells (Figure [2d](#F2){ref-type="fig"}, left two panels), very little or no E-cadherin was observed following 48 hours siRNA treatment (right panels). By co-staining for actin (bottom row), changes in cellular morphology were observed following siRNA-mediated CDH1 suppression. Interestingly, in the samples which were treated with siRNA and tamoxifen combined, the numbers of cells with a more spherical appearance appeared to increase, an event previously linked to enhanced cellular migration \[[@B20]\] (Figure [2d](#F2){ref-type="fig"}). As was the case with the antibody, siRNA-mediated suppression of E-cadherin in the absence of oestrogen resulted in an increase in the numbers of cells invading through the Matrigel basement membrane, although the extent of this was variable (Figure [3a](#F3){ref-type="fig"}) and overall not significant. However, when tamoxifen was included in this system, the numbers of invading cells was considerably higher (Figure [2a](#F2){ref-type="fig"}). No effects on invasion were observed following treatment of the cells with either transfection lipid or the non-targeting siRNA, or the addition of E2 (data not shown). Whilst cell growth assays confirmed that these cells were responsive to E2 and antihormones irrespective of E-cadherin expression (Figure [3b](#F3){ref-type="fig"}), the increases in invasion following antihormone treatment were again not attributable to changes in cellular proliferation as there were no significant changes in cell growth as a consequence of any treatment over the short experimental time period (6 days) (Figure [3c](#F3){ref-type="fig"}). To determine whether these effects were non-specific and independent of the ER, ER-negative MDA-MB-231 cells were treated similarly. However, no increase in invasion of these cells was seen (Figure [3d](#F3){ref-type="fig"}). {#F3} Tamoxifen-induced invasion in E-cadherin deficient MCF7 cells involves increases in Src kinase activity ------------------------------------------------------------------------------------------------------- Tamoxifen can activate Src kinase \[[@B8]\] known to promote the invasive phenotype of tamoxifen-resistant MCF-7 cells \[[@B21]\]. Thus, in an attempt to elucidate the mechanism by which tamoxifen may promote invasive behaviour in the absence of intercellular adhesion, we investigated the expression and activity of Src kinase. Src phosphorylation at Y418 was increased in all cells treated with tamoxifen compared with -E2, and was significantly higher in cells treated with the combination of CDH1 siRNA and tamoxifen (Figure [4a, b](#F4){ref-type="fig"}). {#F4} To determine the functional relevance of elevated Src activity, invasion assays were performed following siRNA and tamoxifen treatment and in the presence of the Src kinase inhibitor, SU6656. Treatment of cells with SU6656 (2.5 μM) inhibited Src kinase activity in all treatment samples while having no effect on total Src levels (Figure [5a](#F5){ref-type="fig"}). The invasive capacity of these cells was then determined in the presence or absence of SU6656, alongside inhibitors of the EGFR (1 μM gefitinib) and HER2 (100 nM traztuzumab), both previously reported to be upregulated following prolonged tamoxifen treatment in MCF-7 cells \[[@B5],[@B22],[@B23]\]. In these experiments, only inhibition of Src kinase activity was able to significantly suppress the invasion of tamoxifen-treated, E-cadherin-deficient MCF-7 cells (Figure [5b](#F5){ref-type="fig"}). This was not due to inhibition of cellular growth, as none of the inhibitors tested significantly altered cellular growth over a period of 72 hours, the time period over which the invasion assays were carried out (Figure [5c](#F5){ref-type="fig"}). {#F5} The steroidal anti-oestrogen fulvestrant also enhances breast cancer cell invasion in the absence of E-cadherin expression -------------------------------------------------------------------------------------------------------------------------- To determine whether the invasion-promoting effects of tamoxifen could also be achieved with a different class of anti-oestrogen, experiments were repeated using the steroidal agent fulvestrant instead of tamoxifen. Our data revealed that fulvestrant is also able to induce cellular invasion in the absence of E-cadherin (Figure [6a](#F6){ref-type="fig"}), whereas oestrogen withdrawal did not. Moreover, in the absence of E-cadherin, only fulvestrant treatment was accompanied by an apparent increase in Src kinase activity (Figure [6b](#F6){ref-type="fig"}). As was the case for tamoxifen, no gain in invasion was seen after fulvestrant treatment of the ER-negative, E-cadherin-negative MDA-MB-231 cells (Figure [6d](#F6){ref-type="fig"}). {#F6} Discussion ========== Adjuvant endocrine therapy has been a major contributor to the decline in breast cancer mortality in the Western world, with the effectiveness of tamoxifen in women with ER-positive cancers clearly demonstrated in many trials \[[@B24]\]. More recently, the dominance of tamoxifen has been challenged by the advent of the aromatase inhibitors, anastrazole, letrozole and exemestane. A number of large randomised trials (e.g. BIG98, MA17 and ATAC \[[@B25]-[@B27]\]) have examined the efficacy of aromatase inhibitors in the adjuvant therapy of ER-positive postmenopausal patients, assessing upfront therapy versus tamoxifen, sequential therapy and extended therapy beyond five years. The encouraging results of these studies have led to aromatase inhibitors becoming a standard component of care for these patients. There is also continuing interest in steroidal anti-oestrogens, such as fulvestrant in breast cancer, currently being explored through trials at various stages of the clinical management of the breast cancer patient. However, despite these advances, the optimal endocrine therapy for an individual patient remains uncertain. In addition, a significant number of patients will still experience a recurrence during endocrine therapy and ultimately die of their disease. It is becoming apparent that administering endocrine agents such as tamoxifen has effects far beyond their initially described mechanism of action and thus attempting to understand the failure of such therapy is a complex problem. Emerging evidence suggests that tamoxifen can modulate cellular processes linked to migratory and invasive responses *in vitro*. For example, tamoxifen can induce FAK-mediated cytoskeletal remodelling \[[@B9]\] and the expression of matrix metalloproteinases \[[@B28]\] by non-genomic mechanisms. Additionally, since the majority of changes in gene expression after oestrogen stimulation are largely thought to be of a repressive nature \[[@B3]\], anti-hormones themselves may promote the re-expression of such genes. Although many of these oestrogen-repressed genes may be growth inhibitory, and thereby contribute to tamoxifen-induced suppression of cell growth, it is increasingly recognised that anti-hormones also induce many genes with an ontology that is not easily reconciled with growth inhibition. Indeed, expression of such genes may contribute towards limiting maximal anti-tumour activity of these agents in ER-positive breast cancer cells \[[@B29],[@B30]\] as is suggested by the ability of tamoxifen and further anti-oestrogens to promote early induction of EGFR and HER2 \[[@B5]\], genes linked to tamoxifen resistance and an adverse cell phenotype. Furthermore, in the light of reports demonstrating that oestrogens confer a protective effect on invasiveness and motility \[[@B31]\] one might expect anti-hormones to reverse this process. However, despite data from ourselves and others demonstrating a significant, tamoxifen-dependent induction of pro-invasive genes in anti-hormone responsive cells, this does not subsequently translate out into a significant increase in cellular invasive capacity *in vitro*, although a previous study by Mathew and colleagues \[[@B32]\] demonstrated increased healing of MCF-7 cell wounds *in vitro*in response to tamoxifen. Intriguingly, however, our data here indicate the full impact of anti-oestrogen-induced genes may only be manifested under conditions of poor cell-cell contact. Thus, siRNA-induced depletion of E-cadherin-mediated intercellular adhesion, while promoting a modest increase in cellular invasion, greatly enhances the ability of tamoxifen to induce invasive behaviour in MCF-7 cells. Clearly, while this anti-oestrogen confers only small increases in invasiveness under conditions of good cell-cell contact, this may become substantial where cell-cell contact is compromised. These observations are interesting clinically, given that studies have shown that up to 40% of non-lobular breast cancers show reduced or absent E-cadherin expression \[[@B33]\], associated with a poorer prognosis \[[@B34],[@B35]\]. To date, however, it is unknown as to whether E-cadherin status correlates with survival on, or response to, tamoxifen treatment. Our data suggests that underlying these phenomena is an increase in the activity of Src kinase. Although it has been previously shown that inhibition of E-cadherin function can lead to elevated Src activity \[[@B16]\], our study indicated only a small, non-significant increase in Src phosphorylation following knockdown of the E-cadherin gene. These apparent discrepancies may arise from the fact that in our study, Src activity was monitored after long-term (72 hours) knockdown of E-cadherin expression, compared with short-term (30 minutes), antibody-mediated E-cadherin disruption in the previous case. The effect of tamoxifen on Src activity in this system is interesting in light of the role of Src in cellular invasion. Elevated Src activity is reported in a variety of solid tumours, including breast cancer, and its expression has been shown to increase with disease progression, suggesting an important role in invasion and metastasis \[[@B36]\]. *In vitro*studies have also shown that elevated Src kinase promotes an aggressive and invasive phenotype in tamoxifen-resistant cells \[[@B10]\] and has been linked to the induction of epithelial-to-mesenchymal transition \[[@B37],[@B38]\]. The relevance of Src activity to tamoxifen-induced cellular invasion in our model is further demonstrated by the fact that inhibition of Src phosphorylation significantly reduced tamoxifen-induced invasion in the absence of E-cadherin. The observation that Src inhibition did not completely reverse tamoxifen-induced invasion suggests there are other important mechanisms involved. Interestingly, the adverse inductive effects do not appear to be specific to the selective ER modulator tamoxifen, because our studies reported here showed that similar invasive responses, and a corresponding increase in Src activity, could also be observed after treatment with fulvestrant (a steroidal anti-oestrogen). Critically, we observed that such events were absent under conditions of oestrogen withdrawal and were not seen in an ER-negative, E-cadherin negative cell line (MDA-MB-231). In total, these data indicate the induction of aggressive cellular behaviour in this model system is a unique consequence of anti-oestrogen occupancy of the oestrogen receptor. The mechanism by which anti-oestrogens, such as tamoxifen, promote increased Src activity in the absence of E-cadherin is currently unclear. Tamoxifen, itself, can activate signalling pathways via growth factor receptors \[[@B39]\]. Conversely, E-cadherin expression suppresses the activity of receptor tyrosine kinases upstream of Src, such as the EGFR \[[@B40]\]. It is evidenced by the positive regulation of E-cadherin in *in vivo*systems such as the Apc+ mouse facilitating negative regulation of EGFR and associated signalling molecules including Src \[[@B41]\]. Moreover, mice bearing germline mutations in Apc, display impaired E-cadherin-mediated adhesion together with augmented EGFR and Src activity \[[@B42]\]. Thus, the loss of E-cadherin may allow the subsequent activation of growth factor receptors by tamoxifen acting to augment Src kinase activity and leading to an enhanced invasive phenotype. To this end we have investigated the expression and activity of EGFR family members in MCF-7 cells under conditions of E-cadherin deficiency and tamoxifen but have not observed any alteration in expression or activity (A. Borley, unpublished data). However, this does not rule out the role of other growth factor receptors in this process and current studies are underway to address this issue and identify potential candidates. Our data show that tamoxifen and fulvestrant significantly induce invasion in MCF-7 cells where the E-cadherin intercellular-mediated intercellular adhesion is disrupted, and that activation of Src kinase plays an important role in this process. The clinical implications of these findings are that patients with breast cancers exhibiting aberrant expression in adherens junction components may have a poorer long-term outcome on therapy with a selective ER modulator and that such observations may ultimately extend to steroidal anti-oestrogens. Such changes in expression might be inherent in a subpopulation of breast cancer patients, or induced by long-term endocrine treatment, where prolonged exposure to tamoxifen can promote growth factor pathway activation and the resultant modulation of E-cadherin/catenin function \[[@B43]\]. In such cases, adjuvant tamoxifen therapy may induce expression of pro-invasive genes and promote an adverse cell phenotype given the absence of cell-cell adhesion. We suggest this may lead to an increased risk of aggressive behaviour of recurrences on therapy and hence impact substantially on prognosis. Significantly, our recent studies in an additional ER-positive, E-cadherin-positive breast cancer cell model (T47D cells) also demonstrate that similar pro-invasive effects can be elicited by tamoxifen following suppression of E-cadherin; importantly, this is also accompanied by an increase in Src activity \[see Additional data file [1](#S1){ref-type="supplementary-material"}\] thus suggesting that these events are not specific to MCF-7 cells, but may represent a generic effect in such breast cancer subtypes. Importantly, although it might be tempting to speculate that suppression of E-cadherin in MCF-7 cells may represent a model of lobular carcinoma, a subset of breast cancers characterised by low/absent E-cadherin expression \[[@B44]\], this is unlikely because lobular carcinomas are likely to have a multitude of additional changes in gene expression and signalling pathway activity separate from that of a loss in E-cadherin expression. Indeed, a recent report by Rakha and colleagues \[[@B45]\] shows that about 20% of invasive ductal carcinomas have no E-cadherin but are still classed as ductal rather than lobular. Moreover, a number of studies have shown that lobular cancer is a distinct entity of breast cancer that differs from ductal not only in histological and clinical features but also in global transcription programmes \[[@B46]\] and genomic profiles \[[@B47]\]. Lobular carcinomas are generally considered to have more favourable pathological characteristics than ductal carcinomas, including being more likely to express hormone receptors. Although in the early years after diagnosis the prognosis for patients with lobular cancers who receive endocrine therapy appears better than for those with ductal cancers, the overall outcome for lobular cancers may be worse as these show a persistent early relapse rate \[[@B45]\]. Indeed, a recent paper by Pestalozzi and colleagues \[[@B48]\] reports that lobular cancers, despite fairing slightly better than invasive ductal cancers initially, are significantly associated with a reduction in disease-free survival when compared with invasive ductal carcinomas. Such data might suggest that anti-oestrogens may have a beneficial effect on proliferative responses initially but pro-invasive genes expressed over time may promote an invasive phenotype and thus a poor prognosis in cancers lacking E-cadherin. However, no data is currently available concerning the anti-oestrogen-induced expression of pro-invasive genes in lobular cancers. The future availability of *in vitro*cell models of lobular cancer will allow the further examination of anti-oestrogen effects in cells representative of this breast cancer subtype. Whether tamoxifen-induced cell invasion is a direct result of E-cadherin manipulation or arises from changes within other junction proteins (eg, β-catenin) whose function can be altered after E-cadherin deregulation, or as a result of a loss of cell-cell adhesion *per se*, is unknown at present. We are currently examining a series of clinical breast cancers from the Adjuvant Breast Cancer trial \[[@B49]\] to investigate if E-cadherin (and/or its associated adherens-junction proteins) is a marker of earlier relapse and poorer prognosis on tamoxifen therapy. We suggest further similar studies, correlating E-cadherin expression with outcome, be carried out within the large adjuvant aromatase inhibitor trials. Conclusion ========== Taken together, our pre-clinical data generates the hypothesis that in patients whose primary breast cancers show reduced or aberrant E-cadherin expression, tamoxifen therapy may promote the development of an adverse cell phenotype that may have an impact on disease relapse, its invasive behaviour and hence patient survival. Although our observations may only apply to a relatively small subset of patients, they may account for some of the superiority seen with aromatase inhibitors in the large adjuvant studies. If these findings are borne out, E-cadherin expression could be used a biomarker to help guide the choice of adjuvant endocrine therapy. Abbreviations ============= BSA: bovine serum albumin; ER: oestrogen receptor; EGFR: epidermal growth factor receptor; FAK: focal adhesion kinase; FCS: fetal calf serum; HER2: human epidermal growth factor receptor 2; MAPK: mitogen activate protein kinase; PBS: phosphate buffered saline; siRNA: small interfering RNA. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= SH made the original observations, conceived the study, performed the initial experiments and drafted the manuscript. AB participated in study design, carried out the experimental work, analysed the data and contributed to the manuscript. CS and VS assisted with Western blotting and functional assays. JG, PBL and RIN contributed to study design, interpretation of the results and to the writing of the paper. All authors read and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1 Figure that demonstrates the ability of tamoxifen to promote the invasion, and an increase in Src activation, of an additional ER+ cell line, T47D, following E-cadherin knockdown. T47D cells were treated with non-targeting siRNA control (NT control) or siRNA for E-cadherin (CDH1) in the absence of oestrogen (-E2) or presence of tamoxifen (tam) as described for MCF-7 cells. **(a)**Cell invasion was assessed using Matrigel invasion assays while changes in E-cadherin expression and Src activity were determined by **(b)**Western blotting **(c)**with subsequent densitometry. Tamoxifen promoted significant cell invasion in the absence of E-cadherin expression which was accompanied by an increase in Src kinase activity (Src phosphorylated at Y418). ###### Click here for file Acknowledgements ================ The authors would like to thank Lynne Farrow for statistical analysis, Carol Dutkowski for assistance with tissue culture and Rebecca Goode for manuscript co-ordination. This research was funded by In The Pink and the Tenovus charity.

Introduction {#H1-1-ZOI190072} ============ Drug overdoses were the leading cause of accidental death in the United States in 2016.^[@zoi190072r1]^ The opioid overdose epidemic has been attributed, in part, to a rapid increase in opioid analgesic prescribing in the period between 1990 and 2010.^[@zoi190072r2]^ After peaking in 2012, US opioid analgesic prescribing rates have decreased slightly,^[@zoi190072r3]^ although the amount prescribed in 2015 was still approximately 3 times as high as in 1999.^[@zoi190072r4]^ Transmucosal immediate-release fentanyl (TIRF) products are potent, rapid-acting opioid analgesics only approved for use for breakthrough pain in patients with cancer who are already receiving and who are known to be tolerant to around-the-clock opioid analgesics. Reports have, however, suggested there is substantial off-label prescribing of TIRF products.^[@zoi190072r5],[@zoi190072r6]^ To address their growing use and potential harms, the US Food and Drug Administration (FDA) has required TIRF manufacturers to institute programs to restrict prescribing and promote appropriate opioid prescribing and safe use of opioid analgesics, including Risk Evaluation and Mitigation Strategies (REMS). A REMS is a drug safety program that the FDA can require for certain medications with serious safety concerns to help ensure the benefits of the medication outweigh its risks. In December 2011, the FDA approved the TIRF-REMS Access Program, mandating prescribers, distributors, pharmacies, and patients to enroll in the program before they can prescribe, distribute, dispense, or use TIRF products.^[@zoi190072r7]^ The TIRF-REMS Access Program was fully implemented in March 2012 and is a shared system REMS that includes all TIRF products. Under the TIRF-REMS, prescribers and pharmacists are required to complete an educational program, patients must be enrolled and are required to sign an agreement attesting to their understanding of the risks and safe use of TIRFs, and distributors are required to agree to ship TIRF products only to enrolled pharmacies. Education for prescribers and pharmacists focuses on key safety information for minimizing the drug's risks and on the safe and appropriate use of these medications. The association of TIRF-REMS with TIRF prescribing has, to our knowledge, not been previously described. We sought to evaluate the association of the TIRF-REMS Access Program with TIRF prescribing using interrupted time series analyses. We used 2010 through 2014 Medicare Part D claims to evaluate TIRF prescribing overall, among patients without cancer, and among patients without known opioid tolerance. We examined whether implementation of the TIRF-REMS was associated with changes in prescribing. Methods {#H1-2-ZOI190072} ======= Data Sources {#H2-1-ZOI190072} ------------ The Yale University institutional review board exempted this study from review. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology ([STROBE](http://www.equator-network.org/reporting-guidelines/strobe/)) reporting guideline. The data were obtained in April 2017 and analyzed from August 2017 through July 2018. We assessed TIRF prescribing using prescription claims data submitted to Medicare Part D for 2010 through 2014. Medicare Part D provided drug coverage for 29.5 million beneficiaries in 2010, including adults aged 65 years or older, as well as disabled adults and those with end-stage renal disease younger than 65 years. Part D grew to cover 40 million beneficiaries in 2014, which represented 70% of eligible Medicare beneficiaries.^[@zoi190072r8]^ We linked the pharmacy claims to beneficiary-level administrative claims data over the study period using beneficiary identifiers. Of note, Part D claims represent dispensed prescriptions; a clinician's act of prescribing does not by itself generate a claim. Opioid and TIRF Prescriptions {#H2-2-ZOI190072} ----------------------------- Using a 100% Part D file, we identified all prescription claims for opioids (eTable 1 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}) using the opioid class identifier of the Medi-Span Master Drug Database classification system, which is used to categorize drug claims (eAppendix 1 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}). We then identified claims for TIRF products (eTable 2 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}) using the drug component (fentanyl) and route (oral, intranasal, sublingual, or buccal). We excluded prescriptions missing the route and prescriptions for injectable medications, as detailed in eAppendix 2 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}. Outcomes {#H2-3-ZOI190072} -------- Our main outcome measures were TIRF prescription fills per 100 000 Medicare Part D beneficiaries, overall and stratified by cancer status, drug brand, and age group; the percentage of TIRF prescriptions for patients without cancer, overall and by brand; and the percentage of TIRF prescriptions for patients without known opioid tolerance, overall and by brand. As a secondary outcome, we characterized the number of unique prescribers before and after program implementation. To determine TIRF prescribing per 100 000 Medicare Part D beneficiaries, we calculated aggregate monthly opioid and TIRF prescribing rates per 100 000 Part D beneficiaries, accounting for the monthly variation in covered beneficiaries using Part D monthly enrollment data.^[@zoi190072r9]^ We excluded prescriptions for sublingual fentanyl tablets marketed as Abstral, fentanyl nasal spray marketed as Lazanda, and fentanyl buccal film marketed as Onsolis (0.96%, 0.49%, and 0.24% of all TIRF prescriptions, respectively), as these drugs had separate REMS implemented prior to the classwide TIRF-REMS implementation. We identified unique prescribers using national provider identifiers or US Drug Enforcement Agency registration numbers included in the prescription data. Of note, the only brands for which there were prescriptions during each month of the study period to allow for brand-level analysis were Actiq and Fentora. To calculate the percentage of TIRF prescriptions for patients without cancer, we first needed to differentiate patients with and without cancer. We categorized prescriptions filled by beneficiaries who did not have a cancer diagnosis associated with any visit during the calendar year the prescription was filled as TIRF prescriptions for patients without cancer. In contrast, prescriptions filled by beneficiaries with a cancer diagnosis associated with any visit during the same calendar year were categorized as prescriptions for patients with cancer. We also performed sensitivity analysis for this primary outcome measure using a broader cancer definition, defined as beneficiaries who had a cancer diagnosis associated with any visit during any year of the study period. To determine cancer diagnoses, we used 4 hierarchical condition category indicators^[@zoi190072r10]^ that encompass all cancer diagnoses (indicators 7-10). These hierarchical condition category indicators are determined annually for all Medicare beneficiaries using inpatient and outpatient claims and are part of a model used by CMS to risk adjust Medicare beneficiaries and determine payments for Medicare Advantage insurers. For each month, we calculated the percentage of TIRF prescriptions for patients without cancer, overall and by brand. We again excluded prescriptions for Abstral, Lazanda, and Onsolis, and brand-level analyses were only conducted for Actiq and Fentora. To calculate the percentage of TIRF prescriptions for patients without known opioid tolerance, we first needed to define opioid tolerance. To do this, we used the definition of opioid tolerance in the approved TIRF-REMS education program,^[@zoi190072r11]^ which defines it as those patients receiving opioids averaging 60 morphine milligram equivalents (MME) per day for at least 7 days, or who have been receiving 30 mg of oxycodone or 8 mg of hydromorphone daily for at least 7 days. We calculated the total MME for each prescription using Centers for Disease Control and Prevention published conversion factors (eTable 3 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}).^[@zoi190072r12]^ We then calculated the average daily prescribed MME for each patient using all opioid prescriptions filled in the 7-, 14-, 30-, 60-, and 90-day periods prior to the patient's initial TIRF prescription, including prescriptions for Abstral, Lazanda, and Onsolis. Patients whose daily prescribed average was less than 60 MME in all of the lookback periods and whose average daily dose of oxycodone or hydromorphone was less than 30 mg and 8 mg, respectively, were considered patients without known opioid tolerance. We then calculated the monthly percentage of initial TIRF prescriptions to patients without known opioid tolerance, overall and by brand. Of note, for these analyses, the only brand for which there were prescriptions during each month of the study period to allow for brand-level analysis was Fentora. Statistical Analysis {#H2-4-ZOI190072} -------------------- After descriptive analysis, we performed interrupted time series analyses using segmented ordinary least squares regression^[@zoi190072r13]^ with robust Newey-West errors to account for autocorrelation and heteroskedasticity.^[@zoi190072r14],[@zoi190072r15]^ Interrupted time series analysis is best used to analyze data collected over time and at regular intervals.^[@zoi190072r16]^ This method of analysis calculates independent tests of the level (the intercept) and trend (slope) before and after an interruption, or intervention, and then evaluates for differences between the 2 levels and slopes. This allows the preintervention level and trend to serve as a predictor of the counterfactual, ie, what would have been without the intervention.^[@zoi190072r17]^ Additional details on the models used are included in eAppendix 3 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}. We assessed whether REMS implementation in March 2012 was associated with changes in either the level or trend for the 3 main outcome measures. Changes in level would be expected around intervention implementation, while trend tracks outcomes in the preintervention and postintervention periods. We specified a level change model,^[@zoi190072r18]^ expecting the strict implementation and mandatory requirements and controls of the TIRF-REMS program to have an immediate impact on TIRF prescribing overall, as well as on prescribing to patients without cancer and to patients without known opioid tolerance. Analyses for the main outcome measures were performed using single-group time series analysis. Single-group time series models assume that the preintervention trend captures all unmeasured time-varying confounders.^[@zoi190072r14],[@zoi190072r17]^ To account for potential unmeasured confounders that could have affected all opioid prescribing, such as a growing national awareness of opioid harms, we performed sensitivity analyses using all-opioid prescribing as a control. For these analyses we used prescriptions for any product containing opioids during the study period, and also repeated the analysis excluding prescriptions for buprenorphine-containing products, as an increase in buprenorphine prescribing for opioid use disorder treatment could mask a decrease in overall opioid prescribing, as detailed in eAppendix 4 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}. We used multiple-group time series models to look for differences between all-opioid prescribing and TIRF prescribing. We also used 2-group models to assess for differences in outcome measures between brands, using generic prescriptions as a control. All models were adjusted to account for autocorrelation, seasonal variation, and the number of days in the month, as detailed in eAppendix 5 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}. Statistical significance was considered as 2-sided *P* \< .05. We report adjusted relative percentage changes in the text, with adjusted absolute level and slope changes in the [Table](#zoi190072t1){ref-type="table"}, as detailed in eAppendix 6 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}. Analysis was performed using Stata statistical software version 14.1 (StataCorp LLC^[@zoi190072r14]^). ###### Interrupted Time Series Regression Analysis for the Primary Outcomes[^a^](#zoi190072t1n1){ref-type="table-fn"} Outcome Baseline Trend (95% CI) Postintervention (95% CI) --------------------------------------------------------------------------------- ------------------------- --------------------------- ------------------------ Per 100 000 Part D beneficiaries[^b^](#zoi190072t1n2){ref-type="table-fn"} All opioid prescriptions 11.2 (−8.3 to 30.8) 39.5 (−299.2 to 378.3) −64.9 (−92.8 to −37.0) TIRF prescriptions −0.06 (−0.09 to −0.03) −1.25 (−1.70 to −0.80) 0.10 (0.06 to 0.13) \% Of TIRF prescriptions for patients[^c^](#zoi190072t1n3){ref-type="table-fn"} Without cancer −0.22 (−0.34 to −0.10) −0.21 (−3.79 to 3.36) 0.12 (−0.03 to 0.28) Without known opioid tolerance 0.16 (−0.11 to 0.43) −6.43 (−10.00 to −1.96) −0.46 (−0.74 to −0.17) Abbreviation: TIRF, transmucosal immediate-release fentanyl. Primary outcomes include monthly prescribing for all opioids and transmucosal immediate-release fentanyl drugs, the percentage of TIRF prescriptions for beneficiaries without cancer, and the percentage of TIRF prescriptions for patients without known opioid tolerance from 2010 to 2014. The table presents absolute values, with relative percentage changes in the text. The table presents absolute percentage changes, with relative percentage changes in the text. Results are adjusted for seasonal trends and autocorrelation. Results {#H1-3-ZOI190072} ======= Overall Rates of TIRF Use {#H2-5-ZOI190072} ------------------------- During calendar years 2010 through 2014, there were 99 601 TIRF prescriptions filled, written by 8619 clinicians to 10 472 patients. For context, over the same period there were 372 023 319 opioid prescriptions filled, written by 2 001 523 clinicians to 27 409 105 patients. Prescriptions for TIRF products represented 0.03% of all opioid prescriptions. The mean (SD) age of patients who filled TIRF prescriptions was 56 (13) years. More than three-quarters (79%) of TIRF prescriptions were for patients younger than 65 years. The TIRF products, including brand names, prescribed during the study period are listed in eTable 3 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"} and prescribing trends by brand are shown in [Figure 1](#zoi190072f1){ref-type="fig"}. {#zoi190072f1} In the period prior to TIRF-REMS program implementation, TIRF prescribing decreased by 1.0% per month (95% CI, −1.5% to −0.6%; *P* \< .001) ([Table](#zoi190072t1){ref-type="table"}, [Figure 2](#zoi190072f2){ref-type="fig"}A; eFigure 1A in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}; for clarity, figures in the article show unadjusted models, whereas adjusted models are shown in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}). The TIRF-REMS implementation was associated with a 26.7% level decline in TIRF prescribing (95% CI, −33.3% to −19.4%; *P* \< .001) but was followed by a change in the trend from 1.0% monthly decreases in prescribing to 2.0% monthly increases (95% CI, 1.3%-2.7%; *P* \< .001). This trend change was driven by prescribing increases that began approximately a year after TIRF-REMS implementation. {ref-type="supplementary-material"}.](jamanetwopen-2-e191340-g002){#zoi190072f2} In contrast to TIRFs, the rate of all-opioid prescribing did not vary significantly prior to the TIRF-REMS program implementation (slope = 0.06%, 95% CI, −0.04% to 0.20%; *P* = .24) ([Figure 2](#zoi190072f2){ref-type="fig"}B; eFigure 1B in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}). In addition, there was no significant level change in all-opioid prescribing associated with TIRF-REMS implementation (level change, 0.26%; 95% CI, −1.7% to 2.2%; *P* = .79), with 0.36% monthly declines during the postimplementation period (95% CI, −0.52% to −0.21%; *P* \< .001). A sensitivity analysis using a 2-group interrupted time series model with all-opioid prescribing as a control showed similar results (eFigure 2 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}); repeat analyses that excluded buprenorphine from the all-opioid control group showed similar results. Supplementary analyses showed no significant difference in prescribing patterns by age group associated with TIRF-REMS implementation (eFigure 3 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}). Similarly, there were no significant brand-related prescribing differences associated with TIRF-REMS implementation (eFigure 4 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}). However, post hoc, supplementary analyses that exclude Subsys prescriptions are shown in eFigure 5 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}. Notably, we observed a discordance between prescribing levels and number of prescribers. There was an initial decrease in TIRF prescribing accompanied by a similarly steep decrease in the number of unique TIRF prescribers. Subsequently we observed a return to baseline prescribing levels, despite a sustained decrease in the number of prescribers (eFigure 6 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}). TIRF Prescribing to Patients With and Without Cancer {#H2-6-ZOI190072} ---------------------------------------------------- Most of the patients who filled TIRF prescriptions (67%) did not have a diagnosis of cancer associated with their claims during the calendar year of the filled prescription, and most prescriptions (72%) were for patients without cancer. There was a 0.29% monthly decline in the percentage of TIRF prescriptions for patients without cancer during the pre--TIRF-REMS period (95% CI, −0.45% to −0.14%; *P* = .001) ([Figure 3](#zoi190072f3){ref-type="fig"}). There were no significant changes associated with REMS implementation in the level (−0.47%; 95% CI, −5.36% to 4.69%; *P* = .85) or trend (0.16%; 95% CI, −0.06% to 0.37%; *P* = .15) of the percentage of TIRF prescriptions for patients without cancer. Estimated declines in the level of TIRF prescribing after TIRF-REMS implementation were similar for patients without cancer (−27.0%; 95% CI, −36.1% to −16.6; *P* \< .001) to those with cancer diagnoses during the calendar year of the prescription (−26.9%; 95% CI, −34.0% to −19.0%; *P* \< .001) (eFigure 7 and eFigure 8 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}). {ref-type="supplementary-material"}.](jamanetwopen-2-e191340-g003){#zoi190072f3} Sensitivity analyses using a broader cancer definition (a claim with a cancer diagnosis during any year of the study) found a similar pre-REMS trend but also found REMS implementation was associated with a 7.2% level decrease in the percentage of TIRF prescriptions for patients without cancer (95% CI, −13.5% to −0.48%; *P* = .04) and with a change in the prescribing trend, from monthly decreases of 0.29% (95% CI, −0.49% to −0.10%; *P* = .003) to monthly 0.63% increases (95% CI, 0.28%-0.98%; *P* = .001) (eFigure 9 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}). As with the findings for the overall TIRF prescribing rate, in this sensitivity analysis the increase in prescribing began approximately a year after TIRF-REMS implementation. Sensitivity analysis comparing brand to generic prescriptions found no consistent differences in prescriptions for patients without cancer for Actiq prescriptions compared with generic, while Fentora had an associated 8.3% level decline in filled prescriptions (95% CI, −14.8% to −1.27%, *P* = .02) (eFigure 10 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}). The monthly percentage of TIRF prescriptions for patients without cancer by brand are shown in eFigure 11 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}. Prescribing to Patients Without Known Opioid Tolerance {#H2-7-ZOI190072} ------------------------------------------------------ Prior to TIRF-REMS implementation, a mean of 30% of initial TIRF prescriptions were for patients without known opioid tolerance as defined by the REMS guidelines, which was nonsignificantly increasing monthly (0.60%, 95% CI, −0.46% to 0.17%; *P* = .26) ([Figure 4](#zoi190072f4){ref-type="fig"}; eFigure 12 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}). Implementation of TIRF-REMS was associated with a 22.5% relative level decline in the percentage of TIRF prescriptions filled by patients without known opioid tolerance (95% CI, −36.1% to −5.95%; *P* = .01) followed by 1.98% monthly decreases (95% CI, −3.19% to −0.80%; *P* = .001). The decline in the percentage of prescriptions filled by patients without known opioid tolerance persisted even when the TIRF-REMS--associated change in the rate of prescribing had diminished. Trends of prescribing to patients without known opioid tolerance using various lookback periods are shown in eFigure 13 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}. Sensitivity analyses found no brand-specific differences associated with TIRF-REMS implementation for Fentora, the only brand with enough prescriptions to analyze this outcome (eFigure 14 in the [Supplement](#note-ZOI190072-1-s){ref-type="supplementary-material"}). {ref-type="supplementary-material"}.](jamanetwopen-2-e191340-g004){#zoi190072f4} Discussion {#H1-4-ZOI190072} ========== In this cohort study using interrupted time series analysis of TIRF prescribing for Medicare Part D beneficiaries between 2010 and 2014, we found that implementation of the TIRF-REMS Access Program, a mandatory restrictive drug distribution program, was associated with a temporary decline in the rate of overall TIRF prescribing, with a possible temporary decline in the percentage of TIRF prescribed to patients without cancer, as well as a lasting decrease in the percentage of TIRF prescribed to patients without known opioid tolerance. To our knowledge, this is the first study examining TIRF prescribing following TIRF-REMS implementation. A closer review of the post-REMS implementation data reveals 2 distinct periods: the 12-month period following implementation, during which prescribing remained depressed relative to pre-REMS, followed by a period of prescribing increases that returned TIRF prescribing to near pre-REMS levels beginning mid-2013. One explanation for the observed attenuation of the decrease in prescribing associated with the program could be that it may have taken a year for patients, prescribers, and pharmacists to familiarize themselves with REMS requirements, and to register and complete the educational program. However, the subsequent return to baseline prescribing levels occurred despite a persistent decrease in the number of prescribers. This points to other possible contributing factors to the subsequent return of increased TIRF prescribing. Notably, the increase in prescribing coincided with a marked increase in filled prescriptions for Subsys, a fentanyl sublingual spray that became the most commonly prescribed TIRF during this period. There have been reports that the Subsys manufacturer used aggressive,^[@zoi190072r5],[@zoi190072r19]^ possibly illegal,^[@zoi190072r20],[@zoi190072r21]^ strategies to increase drug sales and made efforts to guide prescribers and patients through the REMS program requirements.^[@zoi190072r6]^ These efforts may have attenuated the prescribing decrease associated with TIRF-REMS implementation. The overall decline in TIRF prescribing following TIRF-REMS implementation was associated with either similar or close to similar declines in prescription claims for both patients with and without cancer, depending on the cancer definition used. This highlights a potential unintended consequence of the TIRF-REMS. Two additional study findings are notable. First, TIRF prescriptions represented only 0.03% of all opioid prescriptions, highlighting the relatively narrow scope of the TIRF-REMS program. Second, 79% of TIRF prescriptions were written for patients younger than 65 years, who typically receive Medicare benefits owing to disability or end-stage renal disease. These findings are consistent with previous studies that have shown a high rate of opioid prescribing in patients with end stage renal disease^[@zoi190072r22]^ and Medicare disability beneficiaries^[@zoi190072r23]^ and point to a population that may benefit from focused interventions. Limitations {#H2-8-ZOI190072} ----------- Our study has important limitations. First, it is possible that unmeasured policy and/or clinical practice changes, such as the increasing use of prescription drug monitoring programs, or rescheduling of narcotic drug classes, impacted TIRF and other opioid prescribing. While time series methodology, along with the all-opioid control group we used, help mitigate external confounders, they do not eliminate them. Similarly, demographic shifts among Part D enrollees could have affected prescribing. While we did not have monthly demographic data for all Part D enrollees to include as covariates in the models, annual data show no notable year-to-year shifts in sex, race, age, or dual-eligible status of Part D enrollees.^[@zoi190072r24]^ Second, we only had access to prescription data for Medicare beneficiaries who participate in Part D. Medicare Part D beneficiaries receive approximately 25% of all prescriptions written in the United States,^[@zoi190072r25]^ but demographic differences may result in dissimilarities from national prescription rates. Furthermore, given the relatively small number of TIRF prescriptions for Part D beneficiaries, many of the TIRF brands did not have stable monthly prescription numbers to enable brand-level analysis for all outcomes measured. For those that did have enough prescriptions to enable analysis, we may not have had the statistical power to find significant brand-specific differences from generic. Third, our data only include prescriptions paid for by Medicare and do not include prescriptions paid for out of pocket, although the relatively high cost of TIRF drugs makes non--insurance coverage use less likely. Fourth, formulary changes could have affected prescribing. However, midyear, restrictive formulary changes are generally prohibited, so they are unlikely to have caused the immediate level changes noted after TIRF-REMS implementation, which occurred in March, although they may have affected the pretrend and posttrend changes. Fifth, the results from the 2-group time series analyses that compared all opioids to TIRFs may be biased because there were statistically significant differences in the preintervention level as well as trend.^[@zoi190072r14]^ Still, even as an imperfect control, we believe all-opioid prescribing helps control for unmeasured confounders. Conclusions {#H1-5-ZOI190072} =========== We found that a mandatory educational and restrictive distribution program for TIRF products, the TIRF-REMS Access Program, was associated with a temporary decrease in overall TIRF prescribing, possibly with a temporary decrease in off-label TIRF prescribing, and with a lasting decline in the percentage of TIRF products prescribed to patients without known opioid tolerance. ###### **eAppendix 1.** Source and Processing of Prescribing Data **eAppendix 2.** Exclusions **eAppendix 3.** Interrupted Time Series Models **eAppendix 4.** Sensitivity Analysis Excluding Buprenorphine **eAppendix 5.** Model Adjustment **eAppendix 6.** Interpretation of Coefficients **eTable 1.** Medicare Part D Opioid Prescriptions, 2010-2014 **eTable 2.** Part D Transmucosal Immediate-Release Fentanyl Prescriptions by Brand, 2010-2014 **eTable 3.** Opioid Morphine Equivalent Conversion Factors Used for Study Morphine Milligram Equivalent (MME) Calculations **eFigure 1.** Adjusted Interrupted Time Series Models for TIRF Prescribing (Panel A) and All-Opioid Prescribing (Panel B), 2010-2014 **eFigure 2.** Adjusted Two-Group Interrupted Time Series Model of All-Opioid Prescriptions and Transmucosal Immediate-Release Fentanyl (TIRF) Prescriptions per 100,000 Part D Participants **eFigure 3.** Adjusted 2-Group Interrupted Time Series Model of Monthly Transmucosal Immediate-Release Fentanyl Prescribing, by Age Group **eFigure 4.** Adjusted, Interrupted Time Series Models of Transmucosal Immediate-Release Fentanyl Prescribing, 2010-2014, by Brand **eFigure 5.** Post-hoc Analysis of the Three Primary Outcomes With and Without Subsys-Brand Transmucosal Immediate-Release Fentanyl (TIRF) Prescriptions **eFigure 6.** Monthly Transmucosal Immediate-Release Fentanyl Prescriptions and Prescribers per 100,000 Part D Beneficiaries, 2010-2014 **eFigure 7.** Adjusted Time Series Models of Monthly Transmucosal Immediate-Release Fentanyl (TIRF) Prescribing to Patients Without Cancer (Panel A) and With Cancer (Panel B), 2010-2014 **eFigure 8.** Two-Group Adjusted Interrupted Time Series Model of Monthly Transmucosal Immediate-Release Fentanyl Scripts per 100,000 Part D Beneficiaries by Cancer Status **eFigure 9.** Adjusted Interrupted Time Series Models of the Monthly Percentage of Transmucosal Immediate-Release Fentanyl Prescriptions Filled by Patients Without Cancer **eFigure 10.** Adjusted Interrupted Time Series Models of Transmucosal Immediate-Release Fentanyl Prescribing to Patients Without Cancer, 2010-2014, by Brand **eFigure 11.** Monthly Percentage of Transmucosal Immediate-Release Fentanyl (TIRF) Prescriptions Filled by Patients Without Cancer Diagnosis, by TIRF Brand **eFigure 12.** Adjusted Interrupted Time Series Model of Monthly Percentage of Transmucosal Immediate-Release Fentanyl Prescriptions to Patients Without Known Opioid Tolerance, 2010-2014 **eFigure 13.** Percent of Transmucosal Immediate-Release Fentanyl Prescriptions to Patients Without Known Opioid Tolerance Using Various Lookback Periods, 2010-2014 **eFigure 14.** Adjusted Interrupted Time Series Model of the Monthly Percentage of Generic vs Fentora-Branded Transmucosal Immediate-Release Fentanyl Prescriptions to Patients Without Opioid Tolerance, 2010-2014 **eReferences** ###### Click here for additional data file.

Introduction {#Sec1} ============ Oocyte maturity is an essential prerequisite for its successful fertilization and the viability of the resulting embryo. In clinical practice, it is customary to assume that each oocyte displaying a polar body (PB) is a mature metaphase II (MII) oocyte. However, live imaging of microtubule dynamics during human oocyte maturation revealed that PB becomes visible a couple of hours before the bipolar MII spindle is assembled \[[@CR1]\]. Due to this asynchrony, MII-arrested eggs and telophase I oocytes cannot be distinguished by their appearance under a conventional microscope. If the sperm is injected in an unphysiological time, the developmental potential of late-maturing oocytes will be diminished \[[@CR2]--[@CR5]\]. Avoiding untimely intracytoplasmic sperm injection (ICSI) is particularly important in the population of poor/slow responders with a low number of eggs available for fertilization. Only PB-displaying oocytes are subjected to ICSI while germinal vesicle and metaphase I (MI) oocytes are typically discarded. However, immature oocytes, which spontaneously extruded PB in vitro, have been successfully used in fertility treatment \[[@CR6]--[@CR10]\]. Cumulative evidence suggests that the developmental potential of late-maturing oocytes could be better exploited if the timing of ICSI is adjusted to their developmental stage \[[@CR2]--[@CR5], [@CR11]--[@CR14]\]. The advent of computer-assisted polarized light microscopy (PLM) has enabled real-time assessment of egg maturity prior to ICSI. Detectable birefringence is generated by the interaction of a polarized light with the highly ordered microtubular mass building up the bipolar spindle. Due to its non-invasiveness, this technique could be used in clinical settings to monitor the dynamics of division apparatus in a living state \[[@CR3], [@CR15], [@CR16]\]. Numerous studies have explored the possibility of predicting oocyte developmental ability by the evaluation of presence \[[@CR11], [@CR17]--[@CR20]\], morphology \[[@CR21]--[@CR23]\], birefringence intensity \[[@CR24]--[@CR26]\], and positioning of the MII spindle with respect to the PB \[[@CR17], [@CR27]--[@CR29]\]. The majority of these studies reported that the absence of the spindle compromises the ability of the oocyte to be fertilized, undergo preimplantation development, and give rise to full-term pregnancy \[[@CR11], [@CR17]--[@CR19], [@CR21]--[@CR24]\]. Importantly, an experiment performed by Montag and colleagues demonstrated that the absence of the spindle birefringence might be only temporary, corresponding to the physiological transition from MI to MII stage \[[@CR2]\]. In light of these findings, it is assumed that at least some late-maturing oocytes would benefit from the postponement of ICSI instead of being subjected to immediate sperm injection. The primary objective of this study was to determine if PLM could be used as a tool to optimize the timing of ICSI with respect to the maturational stage of the oocyte. We sought to test the hypothesis that postponing ICSI of late-maturing oocytes might be beneficial because it provides more time for the emergence of the MII spindle, which is known to be associated with better clinical results. Materials and methods {#Sec2} ===================== Study population and ethical approval {#Sec3} ------------------------------------- This prospective case series study was carried out in a single private IVF center, Reprofit International (Brno, The Czech Republic). It was comprised of a total of 916 oocytes from 229 IVF patients (aged 25--48 years, average 36.96 ± 4.28 years) undergoing 234 ICSI cycles between May 2016 and May 2018. Study participants were included only if the number of PB-displaying oocytes retrieved in stimulated cycles was lower than 6 and if there was a large proportion of immature MI oocytes which extruded PB in vitro within 3--4 h after retrieval (MI/MII oocytes). Severe male infertility factor was present in 14.96% of analyzed cases. In 15.38% of cycles, only immature oocytes, not MII oocytes, were collected. All included cycles were assigned to ICSI and extended embryo culture. Apart from the study population, 404 MII oocytes from 50 egg donors (aged 20--32 years, average 26.42 ± 2.99 years, no previous history of infertility) were used as a positive control of the spindle imaging procedure. Written informed consent was obtained from all participants. The study design was approved by the institutional Ethics Committee. Ovarian stimulation and oocyte retrieval {#Sec4} ---------------------------------------- Ovarian stimulation was induced with either recombinant follicle-stimulating hormone (Gonal-f, Merck Serono, Switzerland; Puregon, MSD, USA) or highly purified human menopausal gonadotropin (Menopur, Ferring, Switzerland). Pituitary suppression was achieved by administration of the gonadotropin-releasing hormone antagonist (Cetrodite, Merck Serono, Switzerland). Ovulation was triggered with human chorionic gonadotropin (hCG) (Ovitrelle, Merck, Switzerland). Oocyte pick-up (OPU) was scheduled 35--36 h after hCG injection. Cumulus-oocyte complexes (COCs) were collected in a MOPS/HEPES-buffered medium (MHM, Irvine Scientific, USA). Oocyte denudation was undertaken immediately after retrieval. COCs were briefly exposed to hyaluronidase (\#90101, Irvine Scientific, USA) and cumulus-corona cells were mechanically removed by gentle pipetting. Developmental status of the oocyte was assessed according to the presence or absence of the first PB. MI and MII oocytes were incubated separately in Continuous Single Culture (CSC) medium (\#90164, Irvine Scientific, USA) for an additional 3--4 h at 37 °C in a humidified atmosphere of 5% O~2~ and 6% CO~2~. Oocytes, which extruded PB in vitro during the pre-incubation period, are hereby referred as "MI/MII oocytes" while term "MII oocytes" is restricted to the oocytes displaying PB at the time of denudation/retrieval. Polarized light microscopy {#Sec5} -------------------------- All PB-displaying oocytes (MII and MI/MII) were subjected to maturity assessment prior to ICSI that was scheduled 39--40 h after hCG trigger (3--4 h after denudation). The oocytes were placed individually into numbered 5-μl droplets of a prewarmed HEPES/MOPS-buffered medium covered with equilibrated mineral oil (\#9305, Irvine Scientific, USA) on glass-bottomed dishes (WPI Fluorodish FD 5040 or WillCo GWST-5040). PLM examination was performed on a Nikon Eclipse TE 2000-U microscope (Tokyo, Japan) equipped with OCTAX polarAIDE™ (MTG, Germany). Imaging software OCTAX Eyeware ™ (MTG, Germany) combined bright field (green) and birefringence (red) visions of individual oocytes while they were gently rotated around each axis to achieve spindle alignment with the path of polarized light. All micromanipulation procedures were carried out in a temperature-controlled environment maintaining 37 ± 0.5 °C in the culture droplet. PLM-examined oocytes were categorized based on the pattern of detected birefringence (Fig. [1](#Fig1){ref-type="fig"}a). Grade A oocytes featured a bipolar barrel-shaped spindle with clearly delineated boundaries and even distribution of birefringence, while grade B oocytes displayed dysmorphic, apolar, and translucent spindles with irregular boundaries and uneven distribution of the signal. Oocytes with no visible spindle birefringence were classified as a grade C. Anaphase I/telophase I oocytes, showing a microtubular bridge (a connective strand between first PB and the oocyte) instead of MII spindle, were marked as grade D oocytes. If the majority of oocytes in the treatment cycle were found to lack the MII spindle (grades C and D), sperm injection was postponed and performed 1--5 h (on average 2 h) later after additional PLM examination. The spindle imaging procedure was video recorded and the grading of each oocyte was performed by two independent evaluators.Fig. 1Categories and incidence of oocyte grades. (a) Representative examples of different oocyte grades based on PLM-detected pattern. Scale bar; 20 μm. (b) The incidence of oocyte grades (A--D) in the control group of oocytes (egg donors) and study group (poor/slow responders). Distribution of oocyte grades in subpopulations of MII and MI/MII oocytes is shown along with overall distribution. The total number of oocytes in each group is shown in brackets ICSI, embryo culture, and embryo transfer {#Sec6} ----------------------------------------- All eggs underwent ICSI immediately after PLM (re-)assessment. They were placed into 5-μl droplets of a prewarmed HEPES/MOPS-buffered medium, covered by mineral oil on plastic microinjection dishes (Nunc™ IVF Petri Dishes, \#150270, Thermo Fisher Scientific, Waltham, USA). ICSI was routinely performed according to the standard protocol using ICSI/holding micropipettes (\#002-5-30/\#001-120-30, Microtech IVF, The Czech Republic), polyvinylpyrrolidone (\#90121, Irvine Scientific, USA), and Eppendorf (Hamburg, Germany) micromanipulation system equipped with thermoplate (TokaiHit, Japan). After ICSI, injected oocytes were placed into 30-μl droplets of CO~2~-dependent CSC medium covered with mineral oil in micro-droplet culture dishes (\#16003, Vitrolife, Sweden) and individually cultured at 37 °C in a humidified atmosphere of 5% O~2~ and 6% CO~2~. Embryo development was followed until day 5 or day 6 when good-looking blastocysts were either transferred in a fresh cycle or cryopreserved (Rapid-i, \#10119, Vitrolife, Sweden). Blastocysts were chosen for transfer based on morphological criteria by clinical embryologists who were blinded to the PLM assessment results. Study endpoints and data analysis {#Sec7} --------------------------------- Primary outcomes were spindle visualization during PLM (re-)examination, fertilization, embryo blastulation, and utilization. ICSI outcome was analyzed with respect to the oocyte grades (A--D). Besides, developmental potential of oocytes with (grade A or B) and without (grades C and D) the MII spindle was compared. Fertilization was defined as the presence of two pronuclei 16--20 h post-ICSI. Blastulation (BR) and utilization rates (UR) were calculated as the total number of blastocysts/utilized embryos (i.e., transferred in the fresh cycle or cryopreserved) divided by the number of injected oocytes. Secondary outcome measurements included biochemical pregnancy rate (PR---number of positive hCG tests on day 10 post embryo transfer per number of embryo transfers) and clinical pregnancy rate (CPR---number of pregnancies with detected fetal heartbeat after 12 weeks of gestation per number of embryo transfers). Information about treatment outcomes, including abortions (loss of clinical pregnancies) and live births, was extracted from the electronic registration system. Statistical methodology {#Sec8} ----------------------- Standard measures of summary statistics were used to describe the data: relative and absolute frequencies for categorical variables and arithmetic means with standard deviations for continuous variables. The chi-squared distribution test was used to compare fertilization, blastulation, and utilization (yes/no) of oocytes with/without detectable spindles. Evaluation of the association between the presence of the spindle and the formation of blastocyst was the main endpoint of the statistical analysis. For this purpose, generalized estimating equations (GEE) logistic regression with repeated measurements was used. According to the quasi-likelihood under independence model criterion (QIC), an independent correlation structure was chosen. The dependent variable was the formation of the blastocyst (yes/no); independent variables were the presence of a spindle (yes/no), the presence of severe male infertility (yes/no), and the age of female patients. Results with a *p* value \< 0.05 were considered statistically significant. Statistical analysis was performed using SAS software version 9.4. Results {#Sec9} ======= PLM assessment prior to scheduled ICSI {#Sec10} -------------------------------------- A total of 916 PB-displaying oocytes were PLM-examined and classified as A, B, C, or D based on the observed birefringence pattern (Fig. [1](#Fig1){ref-type="fig"}a, criteria described in the "[Materials and Methods](#Sec2){ref-type="sec"}" section). Only 512/916 (55.90%) of analyzed oocytes exhibited PB already at the time of retrieval (MII oocytes), while 404 oocytes (44.10%) extruded PB in vitro shortly after retrieval (MI/MII). The MII spindle (oocyte grade A or B) was detected in 89.84% MII oocytes and 38.86% MI/MII oocytes (Fig. [1](#Fig1){ref-type="fig"}b). Together, nearly a third (299/916, 32.64%) of examined oocytes from poor/slow-responding patients showed no MII spindle (grade C or D) at 3--4 h after OPU. This result was in sharp contrast with the data from the maturity assessment of the control group of donor eggs. Oocytes from egg donors typically displayed PB at the time of retrieval and the MII spindle was absent in only 3.22% (13/404) of collected oocytes (Fig. [1](#Fig1){ref-type="fig"}b). If the majority of oocytes within the treatment cycle exhibited an MII spindle, fertilization took place immediately after PLM examination. A total of 682 oocytes (432 MII and 259 MI/MII oocytes) were injected in standard ICSI time---39--40 h after hCG (Supplementary Table [1](#MOESM2){ref-type="media"}). Delayed ICSI and PLM re-examination {#Sec11} ----------------------------------- PLM examination revealed that the absence of a spindle signal was abundant in MI/MII oocytes (Fig. [1](#Fig1){ref-type="fig"}b). Thus, we decided to test whether the absence of the MII spindle might be only temporary. We assumed that the spindle signal would develop as late-maturing oocytes progressed in maturation. If the majority of oocytes within the treatment cycle was found to lack MII spindle signal, ICSI was deliberately rescheduled to a later time (Supplementary Table [1](#MOESM2){ref-type="media"}). A total of 234 oocytes (89 MII oocytes and 145 MI/MII oocytes) from 86 cycles were not injected immediately after the first PLM assessment (time 1). Instead, the group of oocytes was kept in culture and re-inspected for presence/absence of the MII spindle prior to delayed ICSI \~ 2 h later (time 2). The PLM pattern has changed in 70.51% of analyzed oocytes and the presence of the spindle dramatically increased from 29.49% at time 1 to 71.37% at time 2 (Fig. [2](#Fig2){ref-type="fig"}a). The fact that 98/165 (59.39%) of initially spindle-negative oocytes displayed a spindle signal at a later time demonstrated that birefringence pattern is not a fixed quality of the given oocyte but may change over time (Fig. [2](#Fig2){ref-type="fig"}a, b). Interestingly, 86.73% of eggs that developed a spindle signal during extended incubation period were MI/MII oocytes (Fig. [2](#Fig2){ref-type="fig"}b). Provided with the extra incubation time, C-graded oocytes were more likely to assemble a detectable MII spindle than D-graded oocytes which represents an earlier stage of oocyte development (Fig. [2](#Fig2){ref-type="fig"}b). Together, this data indicates that late-maturing oocytes, which failed to complete development in vivo, require extra time to achieve full maturity in vitro.Fig. 2Oocyte grades in cycles with postponed ICSI. (a) The percentage of oocyte grades and (b) number of MII and MI/MII oocytes within oocyte grade categories detected during the initial (time 1) and later (time 2) PLM examination Fertilization, blastulation, and utilization {#Sec12} -------------------------------------------- Postponing the ICSI in indicated cycles, we increased the overall proportion of oocytes with an MII spindle from 67.36 to 77.95% (Fig. [1](#Fig1){ref-type="fig"}b, Supplementary Table [2](#MOESM3){ref-type="media"}). As expected, oocytes in which an MII spindle signal was detected before ICSI showed significantly higher developmental competence than oocytes without a spindle (Fig. [3](#Fig3){ref-type="fig"}a). A-graded oocytes had higher fertilization, blastulation, and utilization rates than grade B oocytes. Spindle-negative oocytes (grades C and D) were less successful in all analyzed parameters (Supplementary Table [2](#MOESM3){ref-type="media"}). The same trend was apparent in subpopulations of MII and MI/MII oocytes (Fig. [3](#Fig3){ref-type="fig"}b, c; Supplementary Table [2](#MOESM3){ref-type="media"}). The MI/MII oocytes showed generally lower developmental potential than the MII oocytes. However, the rate of fertilization and embryonic development was acceptable if they managed to assemble a spindle prior to ICSI (Fig. [3](#Fig3){ref-type="fig"}c, Supplementary Table [2](#MOESM3){ref-type="media"}).Fig. 3Impact of oocyte grade on fertilization and embryo development. (a) Comparison of fertilization (FR), blastulation (BR), and utilization (UR) rates of oocytes with (+) or without (−) spindle in the total of analyzed oocytes, (b) only MII oocytes, and (c) only MI/MII oocytes. \**p* ˂ 0.001 Importantly, the results of logistic regression proved that there is an association between the presence of the spindle and the formation of the blastocyst. According to the odds ratio, oocytes with an MII spindle signal have 3.4 times higher probability of formation of blastocyst than oocytes without a detectable MII spindle. The presence of the spindle is also the only statistically significant variable (*p* \< 0.001) in multivariate analysis. Confounding factors, namely severe male infertility and female age, turned out not to be statistically significant (Table [1](#Tab1){ref-type="table"}). Logistic regression using different oocyte grades (A, B, C, or D) as a potential predictor of formation of blastocyst was performed as well, choosing type A as a reference type. Odds ratios show that blastocysts have a significantly (*p* \< 0.001) lower probability of formation if they originate from oocytes graded B, C, and D (odds ratios 0.492, 0.216, and 0.113, respectively) instead of A-graded oocytes (Table [1](#Tab1){ref-type="table"}). For other variables, results are very similar to previous regression, and male infertility factor and female age remain statistically insignificant.Table 1GEE logistic regression evaluation. Evaluation of the association between blastocyst formation and presence/absence of the spindle or blastocyst formation and oocyte grades (A-D)ParameterOR (95% CI)*p* valuePresence of the spindle3.439 (2.301, 5.140)\< 0.001Male infertility factor0.767 (0.469, 1.253)0.2890Female age0.974 (0.937, 1.011)0.1652Grade A1 (reference)Grade B0.492 (0.358, 0.676)\< 0.001Grade C0.216 (0.135, 0.345)\< 0.001Grade D0.113 (0.042, 0.301)\< 0.001Male infertility factor0.743 (0.449, 1.230)0.2483Female age0.974 (0.939, 1.011)0.1626*OR* odds ratio, *CI* confidence interval Clinical outcome {#Sec13} ---------------- A total of 197 transfers of 216 embryos were carried out. A vast majority (90.74%) of the transferred embryos were derived from oocytes which developed a detectable spindle (grade A or B) prior to ICSI (Supplementary Table [3](#MOESM4){ref-type="media"}). Implantation of A-graded oocytes was slightly higher than B-graded oocytes (PR 48.15% vs. 42.68%) but their chance of producing clinical pregnancies was comparable (CPR 35.80% vs. 36.59, respectively). Embryos derived from grade B oocytes showed a higher risk of abortion at later stages of pregnancy (Supplementary Table [3](#MOESM4){ref-type="media"}). Only 3/15 (20%) of individually transferred embryos derived from oocytes without a spindle produced clinical pregnancies. Only two embryos derived from D-graded oocytes were transferred along with their sibling embryos from spindled oocytes. Interestingly, the combination of embryos derived from spindle-positive and spindle-negative oocyte always resulted in singleton pregnancies. Twin pregnancies were recognized only in two patients who had two embryos derived from A-graded oocytes transferred. In summary, 47/49 live births originate from transfers involving at least one embryo derived from the spindle-positive oocyte (Supplementary Table [3](#MOESM4){ref-type="media"}). Further 8 ongoing clinical pregnancies are still being followed, 7 of them are produced by spindled oocytes. Notably, 28.24% of transferred embryos were derived from MI/MII oocytes and 11/51 healthy children reported so far by study participants were born from these late-maturing oocytes extruding PB in vitro (Supplementary Table [1](#MOESM2){ref-type="media"}). Discussion {#Sec14} ========== In this study, we explored the clinical value of spindle imaging in a group of patients with unexpectedly suboptimal response to ovarian stimulation. We used PLM to inspect whether oocyte completed nuclear maturation prior to ICSI. Non-invasive monitoring of microtubular dynamics during MI to MII transition assisted us to time ICSI according to the developmental stage of the late-maturing oocytes. Combining "rescue in vitro maturation" of MI oocytes with PLM-navigated optimization of timing of ICSI, we were able to effectively utilize oocytes extruding PB in vitro. Immature oocytes retrieved in stimulation cycles are generally regarded as of poor quality and rejected for fertility treatment because of the risk of abnormal fertilization \[[@CR30], [@CR31]\]. Nevertheless, published case reports indicate that immature oocytes can acquire developmental competence during overnight culture \[[@CR6]--[@CR10]\]. Here, we focused on clinical utilization of developmentally delayed oocytes that extruded PB in vitro shortly after retrieval. Our data demonstrated that these MI/MII oocytes can produce viable embryos and full-term pregnancies if the time of ICSI is adjusted to their maturational stage. Hence, poor developmental outcomes of late-maturing oocytes reported in previous studies might be, at least partially, attributed to untimely sperm injection \[[@CR2]--[@CR5]\]. Although being of paramount clinical importance, the timing of fertilization remains a subject of controversy. Clinical data support the notion that a pre-incubation interval between oocyte retrieval and ICSI improves the outcome of IVF treatment \[[@CR12], [@CR14], [@CR32], [@CR33]\]. However, other studies recommended limiting the cultivation period to prevent post-ovulatory aging of the oocyte \[[@CR34], [@CR35]\]. In this study, ICSI was performed on the day of retrieval and total pre-incubation time never exceeded 9 h, the period associated with a decline of post-fertilization outcome \[[@CR5], [@CR32]\]. Oocyte denudation was carried out immediately after retrieval, so we could discriminate between (1) in vivo matured MII oocytes and (2) MI/MII oocytes which extruded PB in vitro. Our results confirmed previous findings that these two subpopulations differ in incidence of the MII spindle and their developmental potential \[[@CR28]\]. If denudation was performed just before ICSI, MII and MI/MII oocytes would not be distinguishable based on morphology. Confocal microscopy clearly demonstrates that the oocyte appearance in transmitted light can be misleading. Some seemingly mature, PB-displaying oocytes might be engaged in chromosome segregation or early phases of spindle reconstitution (interkinesis), and are thus not ready for fertilization. Polarized light microscopy can serve as a tool to discriminate oocytes arrested in an MII stage from those undergoing the maturational transition from the MI to MII stage ([Supplementary Figure](#Sec16){ref-type="sec"}). It has been previously suggested that non-invasive spindle imaging might be used for egg quality control, and to optimize the timing of ICSI in clinical practice \[[@CR4], [@CR11], [@CR17]--[@CR19]\]. Strikingly, reported incidence of spindle-negative oocytes varied from 12 to 56% \[[@CR11], [@CR17]--[@CR28]\]. One of the reasons for this inconsistency could be that the authors worked with different sources of oocytes. In this study, nearly one-third of oocytes from poor/slow responders lacked a detectable MII spindle. This was in contrast to control oocytes from egg donors which typically showed a spindle signal. Also, the representation of late-maturing MI/MII oocytes in an analyzed sample has not been taken into account in most studies. Our data show that not only a maternal age \[[@CR18], [@CR26], [@CR36]\] and infertility factor \[[@CR37]\], but also the suboptimal response to the ovarian stimulation and the timing of observation affect the incidence of spindle-positive oocytes in an analyzed population. In addition, procedural diversities and different laboratory conditions could have influenced PLM examination results. In this study, a great deal of effort was dedicated to optimizing the spindle imaging procedure and minimizing factors known to adversely affect the stability of delicate division machinery, namely pH and temperature fluctuation \[[@CR3]\]. Published studies also differ in the classification of the spindle pattern, study endpoints, and interpretation of results. Some studies only compared oocyte as with and without a spindle \[[@CR11], [@CR17]--[@CR20]\], and others recognized dysmorphic and translucent spindle morphology \[[@CR21]--[@CR23]\]. Here, we classified oocytes into four categories (A--D) based on the observed birefringence pattern. We distinguished between bipolar (grade A) and dysmorphic/translucent spindles (grade B), as well as the absence of spindle birefringence (grade C) and presence of microtubule bridge (grade D). When interpreting a detected spindle pattern, we took advantage of the knowledge of spindle dynamics during oocyte maturation \[[@CR1], [@CR2]\]. Our presumption was that, at least in some oocytes observed shortly after PB extrusion, the absence of spindle birefringence might be only temporary. Indeed, more than half of spindle-negative oocytes managed to form a detectable MII spindle when ICSI was postponed to a later time. The delay in spindle signal emergence is explained by the fact that, in the absence of centrosome, microtubule nucleation is slow and it may take a couple of hours for a bipolar MII spindle to be reformed after PB extrusion \[[@CR1], [@CR2]\]. Capability to assemble the MII spindle is likely to reflect overall oocyte fitness because spindle-positive oocytes have been shown to contain more mtDNA copies and ATP content than oocytes without a spindle \[[@CR38]\]. Multifactorial constitution of egg quality makes it difficult to evaluate in clinical practice. Nevertheless, this study confirmed that the presence of the spindle is significantly associated with the capacity of the oocyte to form a blastocyst. Late-maturing MI/MII oocytes have generally lower developmental competence than in vivo matured MII oocytes. However, the rate of embryo development was acceptable if they managed to assemble a detectable spindle prior to ICSI. Importantly, 10 out of 42 children born in this study were conceived by individually timed fertilization of late-maturing MI/MII oocytes which are typically discarded. We have shown that the "rescue in vitro maturation" of immature oocytes can be used as a salvaging method and viable option to the cancelation of the IVF cycle if there are no MII oocytes available for ICSI. Conclusions {#Sec15} =========== This study demonstrates the benefits of spindle imaging in IVF cycles with a suboptimal response to conventional stimulation. In particular, we used PLM to determine the maturation status of late-maturing oocytes and, when required, postponed ICSI so that the oocytes could complete maturation in vitro. The timing of sperm injection needs to be individually fine-tuned especially for MI oocytes which extrude PB in vitro, and are thus at risk of premature fertilization. Therefore, we suggest the employment of egg maturity assessment in indicated patients with a low number of MII oocytes available for ICSI. Based on our experience, a subtle change in laboratory procedures, namely individual adjustment of the timing of ICSI, can make a major difference for poor prognosis patients. Electronic supplementary material ================================= {#Sec16} Supplementary FigureCorrelation of chromosome-microtubule organization with birefringence pattern in oocytes representing stages of MI to MII transition during oocyte maturation. The appearance of oocytes in transmitted light (top row), combined with a fluorescent signal for chromosomes (cyan) and microtubules (magenta) (middle row), and in polarized light (bottom row) are shown. Each oocyte was first PLM-examined and immediately fixed. Fixed oocytes were (immuno) labeled with Hoechst (DNA) and anti-α-tubulin antibody (microtubules). Fluorescent and transmitted light images were acqured by Zeiss LSM 800 confocal laser scanning microscope. Scale bar, 20 μm. The yellow arrow indicates the presence of PB and the white arrow highlights the position of birefringent microtubules. In vitro matured, supernumerary immature oocytes donated for research were used for this experiment. (PNG 2828 kb) High resolution image (TIF 32126 kb) Supplementary Table 1Overview of standard/delayed ICSI outcomes in MII, MI/MII, and the total of oocytes. (XLSX 11 kb) Supplementary Table 2Impact of oocyte grade on fertilization, embryo development, and clinical outcome. Overview of the total of analyzed oocytes and subpopulations of MII and MII oocytes. (XLSX 12 kb) Supplementary Table 3Oocyte grades and clinical outcome. Overview of clinical outcomes of single/double embryo transfers with respect to oocyte grade. (XLSX 10 kb) **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The authors wish to thank the embryology laboratory and staff of Reprofit International for being instrumental in all procedures and documentation related to this research project. We acknowledge the core facility CELLIM of CEITEC supported by MEYS CR (LM2015062 Czech-Bioimaging) for their support with obtaining supplementary imaging data presented in this paper. Author's roles {#FPar1} ============== Z.H.: conception and design, implementation and supervision, data analysis and interpretation, manuscript writing. D. K, M.M., D.P., and T.K.: collection and/or assembly of data. P.O.: recruitment of eligible patients, informed consent collection. R.S.: statistical analysis, S.K., A.H.: data analysis and interpretation, manuscript proofreading. All authors have read and approved the final manuscript. This research was co-funded by internal R&D funds of the clinic, and a research grant from the Faculty of Medicine, Masaryk University, to the junior researcher, Zuzana Holubcová. No external sponsorship was used. Written informed consent was obtained from all participants. The study design was approved by the institutional Ethics Committee. Conflict of interest {#FPar2} ==================== The authors declare that they have no conflict of interest.

{#sp1 .451} {#sp2 .452} {#sp3 .453}

Introduction {#s1} ============ Sustainable agriculture is of utmost importance to feed the ever-increasing world population. Over the past few decades, higher agricultural yields have been achieved through an increase in, among others, fertilizer application, which increased 7.3-fold from 1960 to 2002 ([@CIT0012]). However, agriculture is currently facing the challenge of reducing nitrogen (N) input due to legislative and public pressure in developed countries. Nitrogen fertilization causes pollution of ground, surface and coastal waters via NO~3~ ^-^ leaching ([@CIT0015], [@CIT0016]; [@CIT0010]) and increases N~2~O emission, which contributes to global warming ([@CIT0024]; [@CIT0046]). In this context, it is a crucial goal to decrease such N losses without reducing crop yield. A potentially interesting method to increase the fraction of N that is taken up by a crop is growing it under row fertilized conditions, which is the patch-wise application of fertilizer in the sowing row. An effective placement of roots would be needed to fully exploit the N patch supplied in this agrotechnique. However, studying root placement in the field in detail is difficult, due to neighbouring plants and the need for an invasive excavation of the root system. In this study, we therefore quantified the response of maize roots to a high N patch in a rhizoslide system ([@CIT0027]) that allowed high resolution measurements of root elongation and architecture over time. Plants generally respond to a local high nutrient patch by increasing lateral root length specifically within the patch, which is referred to as selective root placement ([@CIT0014]; [@CIT0019]; [@CIT0047]). Selective root placement is known to occur in a broad range of plant species ([@CIT0039]; [@CIT0003]), including various cereals, in which increased lateral root elongation and lateral root density was observed in response to N patches ([@CIT0018]; [@CIT0007]; [@CIT0006]; [@CIT0017]; [@CIT0048]). Such an increase in root length and root density is important for nutrient uptake. Yet, the dynamics at which the roots respond to spatially variable nutrient supply is poorly understood. Hence, more detailed measurements are needed to determine which roots respond in which time frame to the locally increased nutrient concentrations. [@CIT0050] showed for *Arabidopsis* seedlings that lateral root elongation rate within the N patch increased linearly for the first 5 d to 3-fold of the original rate, while at the same time the elongation of the laterals outside the patch was only about half this rate and levelled off. This suggests a rapid re-allocation of growth towards the part of the root system directly in contact with the local high nutrient concentration. However, it remains doubtful if responses of seedlings can be easily translated to more complex stages of root architecture. This is particularly true for maize where the post-embryonic root system (crown roots) develops \~14 d after germination and plays a major role in nutrient and water uptake ([@CIT0030]). A promising number of collocations were found between quantitative trait loci (QTLs) for the number of embryonic roots and traits related to grain yield ([@CIT0021]). However, only a few quantitative genetic studies focus on the characteristics of the crown root system ([@CIT0021]). The limited research conducted on the post-embryonic root system leaves a big gap in our understanding of the plasticity and functioning of this most prominent root type of maize. The embryonic root system is typically still strongly influenced by seed properties ([@CIT0011]; [@CIT0023]; [@CIT0027]), and growth and responses may not be representative for roots that develop later. Phenotyping systems, which allow the screening of a large number of plants at later stages of development, such as GROWSCREEN-Rhizo ([@CIT0034]) and rhizoslides ([@CIT0027]), tackle this problem. There are different suggestions how the root system may optimize resource allocation to optimize the uptake of N and water. [@CIT0028] proposed an ideotype of maize to optimize water and N acquisition from deep soil layers. For this ideotype, the different root types specialize in different functions: embryonic roots are foraging for shallow soil resources, while crown roots expand rapidly deep into the soil, thereby being unresponsive to N availability. An alternative concept would be a strongly interactive framework of roots, of which the allocation towards different functions depends on the temporal and spatial availability of soil resources and the demand of the shoot. This would require nutrient sensing and coordinated growth regulation of the various components of the root system, as suggested by [@CIT0026]. Indeed, in *Arabidopsis* low availability of N in parts of the root system is first signalled to the shoot which in turn triggers the expression of NO~3~ ^-^ transporter genes within root regions with high NO~3~ ^-^ content ([@CIT0045]). Most likely, there will be an optimum, balancing between foraging of locally available N and the need to acquire a deep root system for sufficient water availability during grain filling. To evaluate the responsiveness of root growth on spatially varying N supply, methods are needed that enable measurement of the different root type elongation over time. Rhizoslides enable the evaluation of crown root development non-destructively and in detail by combining non-destructive regular imaging with subsequent image analysis ([@CIT0027]). Here, we present a further development of the rhizoslide system, enabling split-nutrient application. We aimed to test this system for its suitability to (i) observe crown root development in response to local N application in high temporal resolution, (ii) model and parameterize this response, and (iii) use the model parameters to evaluate the independence of growth of individual roots, growing on the same plant, when a couple of these roots were subject to an N patch. Moreover, this method allowed us to determine (iv) development of which root type was tightly linked to shoot growth and (v) the importance of the primary root on shoot and root development. Materials and methods {#s2} ===================== Rhizoslide construction {#s3} ----------------------- Experiments were carried out in so called split-nutrient rhizoslides that were developed by [@CIT0027] and facilitate the two-dimensional growth of root systems over time alongside germination paper surfaces. The rhizoslides consisted of Plexiglas plates (length 65cm; width 57cm; thickness 4mm) with blue germination paper on each side (Anchor steel blue seed germination blotter; Anchor Paper Co., Saint Paul, Minnesota, USA; 61×48.4cm) covered with oxygen permeable transparent, polypropylene foil with micro holes of 70 µm (Maag GmbH, Iserlohn, Germany). The split-nutrient system was constructed by creating a water impermeable barrier from top to bottom in the germination paper by ironing into it a narrow band of wax (Wax pen, Knorr Prandell GmbH, Lichtenfels, Germany) ([Supplementary Fig. S1A](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). After this, the germination paper was sterilized by heating in an oven at 80°C for 2h for three consecutive days (tyndallisation). During the time spans between the heating periods, the germination paper was kept in the oven at 30°C with a petri dish filled with deionized water to keep air humidity high (\~50%). The germination paper was then impregnated with a 2.5g l^-1^ Captan solution (Malvin WG, Syngenta Agro AG, Dielsdorf, Switzerland) to suppress the development of fungi. Two PVC plates (60×10×1cm) were attached to the short side of the rhizoslides using flash washers as spacers between the Plexiglas plate and PVC plates. The PVC plates enabled the vertical placement of the rhizoslides in a rack protected against light. Two plants were grown in one rhizoslide, on each side of the Plexiglas plate one ([Supplementary Fig. S1](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). Seedlings were placed in oval cylinders of black foil, filled with water-retaining material (Seramis clay granules, Seramis GmbH, Mogendorf, Germany) between the transparent foil and germination paper on top of the wax barrier ([Supplementary Fig. S1B, C](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). To provide the plant shoots with sufficient space, the wax barrier was not placed exactly in the middle of the germination paper, but slightly (2.2cm) shifted to one side leaving 4.4cm room between the shoot of the plant growing on the front of the rhizoslide and the shoot of the plant growing on the back of the rhizoslide. This enabled the placement of seedlings on top of the wax barrier and at the same time next to each other. To prevent roots from being squashed, rubber flash washers were placed between the PVC plates and transparent foil on each side of the black foil cylinder ([Supplementary Fig. S1B](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). The rhizoslides were each covered from the top with a flat PVC plate with two holes for the plants to grow through, to prevent incidence of light from above ([Supplementary Fig. S1A](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). The four compartments (two on each side of the Plexiglas plate) that were created by the wax barrier in the germination paper were watered separately. For this, four tubes were fixed at the top side of each compartment between the germination paper and the transparent foil. The tubes were attached to drippers (Gardena pressure compensating row drippers supplemented with pumps and transformers from the Gardena Micro-drip Terrace irrigation system with additional timers; Gardena GmbH, Ulm, Germany) that were set to drip 35ml of nutrient solution twice a day. To prevent PO~4~ ^3-^ from precipitating, PO~4~ ^3-^ was kept together with the micronutrients and Fe-EDTA solution separate from the other nutrients and both solutions were mixed 30min before flush through (see 'Plant cultivation' for solution composition). To prevent roots suffering from hypoxia, the mixed nutrient solutions were bubbled with air prior to supplying them to the plants. To prevent fungal infections during plant growth, 0.066g l^-1^ of the fungicide Captan was added to the unmixed nutrient solutions (Malvin WG, Syngenta Agro AG, Dielsdorf, Switzerland). The pH of the unmixed nutrient solutions was adjusted to 6.0 with NaOH (final concentration Na^+^ \~20 µM). Plant cultivation {#s4} ----------------- Maize seeds (*Zea mays* L., genotype B73×UH007, supplied by Delley Seeds and Plants Ltd, Delley, Switzerland) were sterilized in bleach (1.5% in water; v:v) for 15min and incubated in Petri dishes lined with filter paper soaked with a soil bacteria mixture (0.0001% RhizoPlus 42, Andermatt Biocontrol AG, Grossdietwil, Switzerland and 0.01% FZB24, Bayer AG CropScience, Zollikofen, Switzerland) to promote the development of a healthy rhizosphere and protect against seed-borne infections. During germination, seeds were kept in the dark at 26°C. After 70h, seedlings were transferred into the rhizoslides and placed in a climate chamber at 26°C during the 14h light and 20°C during the 8h dark period. Air humidity was 60% and light intensity 720−760 µmol m^-2^ s^-1^ photosynthetic photon flux density (PPFD) at plant canopy level supplied by 2/3 cool daylight tubes (Philips Master TL 5 HO 54W/840; Philips, Amsterdam, The Netherlands) supplemented with 1/3 Grolux tubes (Sylvania Grolux FHO 36W/T5/GRO; Havells India Ltd, Noida, India). Plants received nutrient solution without N containing 5mM KCl, 5mM CaCl~2~, 2mM MgSO~4~, 0.5mM KH~2~PO~4~, 0.04mM FeEDTA and micronutrients (9 µM MnCl~2~, 0.2 µM CuSO~4~, 46 µM H~3~BO~3~, 0.58 µM Na~2~MoO~4~ and 0.77 µM ZnSO~4~) in both compartments for 12 d after transplantation. The first 7 d after transplantation, plants also received nutrient solution without N twice a day at the shoot (40ml) to prevent the young roots from drying. The 12 d period without N was meant to minimize the influence of seed reserves and to create N deficient repression of selective root placement due to a high N content of the shoot. Twelve days after transplantation (set as time point 0; [Fig. 1A](#F1){ref-type="fig"}), the nutrient solution in one compartment was changed to high N containing 5mM KNO~3~, 5mM Ca(NO~3~)~2~, 2mM MgSO~4~, 1mM NH~4~NO~3~, 0.5mM KH~2~PO~4~, 0.04mM FeEDTA and micronutrients (same as for the nutrient solution without N). Ion strength and free ion activity for both the solution with and without N were estimated using the chemical speciation programme GEOCHEM-EZ ([@CIT0040]), and the solutions were adjusted accordingly to keep them as similar as possible with exception of the N concentration. The high N solution was used as a basis, and the nutrient concentrations in the high N solution were adjusted for which KNO~3~ and Ca(NO~3~)~2~ were replaced by KCl and CaCl~2~ (final concentration of Cl^-^ 15mM, which is considered non-toxic for maize; [@CIT0001]) and NH~4~NO~3~ completely removed. {#F1} Daily image capture and analysis {#s5} -------------------------------- In each run, four plants were grown in the split-nutrient rhizoslide setup. From two of these, the primary root was cut off 2 d after transplantation into the rhizoslides. This was replicated five times over time resulting in 20 plants; 10 with and 10 without primary root. From 4 d after transplantation until the end of the growing period, daily images were taken of the whole root system with a 21 mega pixel full-frame digital single-lens reflex camera (EOS 5D Mark II, Canon, Tokyo, Japan) equipped with a 50mm lens (compact macro 50mm f/2.5, Canon, Tokyo, Japan) and circular polarizer (Hama, Augsburg, Germany). For this, plants were placed in a frame with the camera fixed at 125cm distance from the root system. Imaging and image pre-processing was carried out as described by [@CIT0027]. As soon as a root reached the side or bottom of the germination paper, it was not traced any further. Tracing was stopped to avoid modelling of effects that were due to the size of the rhizoslide rather than the effect of the nutrient solution or root age. Crown roots reaching the bottom appeared not to influence elongation rates of associated laterals or other parts of the root system (data not shown). Next, root elongation of the crown roots over time was calculated using R ([@CIT0037]). Crown roots that were not present at one day after solution change (DASC) were discarded, as well as roots that had less than four measurement points, and roots showing zigzag patterns or negative root elongation rates. In a few cases, roots were growing on the wax layer. In these cases, root elongation values were removed during these periods. Crown lateral roots were traced with SmartRoot after solution change (time point 0). For this a 5cm zone on the crown root axis was chosen, the lateral initiation zone, beneath the last formed laterals and taking a 5cm buffer zone from the PVC plate into account ([Fig. 2](#F2){ref-type="fig"}). This lateral initiation zone could be determined at 1−2 DASC, and from then on, the number of laterals was determined each day. Additionally, the five laterals that initiated first in this zone, were traced with SmartRoot until the end of the experimental period. Root elongation was calculated for these laterals using R. {#F2} CCD image capture and analysis {#s6} ------------------------------ From -1 to 1 DASC, a crown root tip in the left compartment and a crown root tip in the right compartment of the rhizoslide of half the number of plants (one with and one without the primary root) was monitored with monochrome Charge Coupled Device (CCD) cameras (Scorpion SCOR-20SO, Point Grey Research Inc, Richmond, British Columbia, Canada) equipped with a 25mm lens (Cosmicar/Pentax, The Imaging Source, Bremen, Germany) and supplemented with a 940nm infrared filter (Edmund Optics, Karlsruhe, Germany). Plants were placed in a black box in which the roots to be monitored were illuminated with near-infrared LED lights (880/940nm) to avoid interference with root growth ([Fig. 1B](#F1){ref-type="fig"}). Images were taken every 90sec at a resolution of 720×1280 pixels. After 22h (time point 0) the low N solution in one compartment was changed to high N solution. For this, the desired nutrient solution in both compartments was flushed through for 3min. Images were analysed with a slightly adapted version of the Martrack Leaf software ([@CIT0031]) that traced the root tip throughout the image sequences and a custom routine in Matlab calculated root elongation rates over time. Final harvest {#s7} ------------- After 30 d (15 DASC), plants were harvested ([Fig. 1A](#F1){ref-type="fig"}). The shoot was cut off at the base and fresh weight was determined. After this, the shoot was bagged and dried in an oven at 60°C for at least 48h until weight stabilised, after which dry weight was determined. Roots were removed from the blue germination paper. The primary root, seminal roots from each compartment, crown roots from each compartment and roots growing within or on the other side of the germination paper were bagged and dried separately at 60°C for at least 48h, after which dry weight was determined. Statistical analyses {#s8} -------------------- The development of elongation rates of the different root traits was plotted using R and appropriate models were chosen according to the shape of the developmental curves ([Supplementary Figs S2−S4](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). For crown root elongation before solution change, a simple linear regression was fitted ([Fig. 3](#F3){ref-type="fig"}, Slope 1). For crown roots growing in the compartment without N after solution change, a regression model with segmented relationships and one breakpoint was fitted using the function *segmented*() of the R-package segmented ([@CIT0032], [@CIT0033]). The estimated parameters were the slope after solution change ([Fig. 3](#F3){ref-type="fig"}, Slope 2), the time at which the breakpoint occurred and the new slope after this breakpoint ([Fig. 3](#F3){ref-type="fig"}, Slope 3). For the high N roots after solution change, a simple logistic model of the form {#F3} F ( x ) = β 1 1 \+ e − ( ( β 2 − x ) / β 3 ) was fitted, were *ß* ~*1*~ is the asymptote ([Fig. 3](#F3){ref-type="fig"}), *ß* ~*2*~ are the days after solution change at which half of the asymptotic value was reached, and *ß* ~*3*~ is a scaling factor. Model fitting was done using the function *nls*() in combination with the self-starting function *SSlogis*() of the R-package stats ([@CIT0037]). Spearman's rank correlations were performed to determine the relationship among model parameter estimates and traits measured at final harvest. Significance of the differences per measured time point was tested with paired *t*-tests after Bonferroni correction between the no N and high N roots. Between plants, unpaired *t*-tests were performed to test for significant differences. Results {#s9} ======= Selective root placement {#s10} ------------------------ To determine whether local application of N resulted in selective root placement, N was applied to only one half of the root system of 15-day-old maize seedlings previously grown without N. This resulted in a root biomass that was 3.6 times higher in the compartment that received N than in the compartment without N ([Fig. 4A](#F4){ref-type="fig"}, [Table 1](#T1){ref-type="table"}). At the time of solution change, the crown root system had already started to develop while the seminal roots were 2 weeks old. Accordingly, the N placement had a strong effect on the developing crown roots which made up 85% of the selectively placed roots. The other 15% were represented by the seminal roots. The importance of the crown root system was also suggested by a positive correlation between dry weight of the part of the root system in the high N compartment and the dry weight of the shoot, whereas no correlation was found between biomass of the seminal root system in the same compartment and dry weight of the shoot ([Fig. 4B](#F4){ref-type="fig"}). Based on these data, further analyses were done solely on the crown root system. {#F4} ###### Effects of local nitrogen application (NN, no nitrogen; HN, high nitrogen) on dry weight (DW) of the shoot and root of *Zea mays* plants DW Shoot (g) DW Seminal (g) DW Crown (g) -------- -------------- ---------------- ------------------ 0.730±0.06 **NN** 0.032±0.01\* 0.100±0.01\*\*\* **HN** 0.072±0.01 0.399±0.04 Plants were grown without N for 15 d after which half of the root system received 17mM nitrogen. Values are means ±SE (*n*=18). Asterisks indicate significance between LN and HN (paired *t*-test: \*\*\*, P\<0.001; \*\*, P\<0.01; \*, P\<0.05; (\*), P\<0.1; NS, P\<1.0). Crown root development {#s11} ---------------------- Individual crown root tip elongation was monitored in high time-resolution from one day before until one day after the onset of local high N supply (i.e. solution change), to investigate if selective root placement was an immediate response to increased N. We only analysed those six plants for which at least one root on the high N side and one root on the side without N could be tracked. Crown axile root elongation rates before solution change were on average \~3cm d^-1^ ([Fig. 5A](#F5){ref-type="fig"}), with some diurnal variation (lowest rates in the afternoon, at 2.0cm d^-1^, and fastest rates in the late morning, at 4.0cm d^-1^). Crown axile roots that received high N showed a relative decrease of elongation rates in the first half hour after solution change, but recovered thereafter. Overall, crown axile roots that received high N did not show a clear change in elongation rate within the first day compared to roots of the same plant that did not receive N ([Fig. 5A](#F5){ref-type="fig"}). Apparently, if crown axile roots respond to an increase in local N, this needs a longer initiation period than 1 d. Therefore, root elongation was monitored at longer time intervals during the following 15 d. In contrast to the first day, crown axile root elongation rates over the remainder of the experimental period showed a striking response to locally applied N ([Fig. 5B](#F5){ref-type="fig"}, [Table 2](#T2){ref-type="table"}). The slightly slowing elongation rates just before solution change continued for roots kept without N until the breakpoint at \~6.4 DASC, after which root elongation rates more or less stabilized at a rate of 1.2cm d^-1^. Crown axile root elongation rates of roots supplied with high N, however, showed a dramatic increase in growth rates, levelling off to a maximum asymptote of on average 5.3cm d^-1^. To estimate the rapidness of this response, we calculated the time at which the roots reached 95% of their asymptotic value. This 95% of the maximum elongation rate was reached at approximately 4 DASC and for the majority of roots stayed at this maximum until the end of the experiment at 15 DASC ([Fig. 5B](#F5){ref-type="fig"}, [Table 2](#T2){ref-type="table"}, [Supplementary Fig. S2](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). {#F5} ###### Parameters describing crown axile and crown lateral root elongation over time of split-root no N (NN) and high N (HN) roots Crown Lateral -------- ------------ --------- ------ ---------- ---------------- ---------- ---------------- ------------- 1 2 3 x (DASC) y (cm day^-1^) x (DASC) y (cm day^-1^) **NN** -0.26 ^NS^ -0.41 0.02 6.4 1.2 0.00 \*\*\* **HN** -0.10 4.0 5.3 0.14 DASC, days after solution change. Crown axile and lateral root parameters were estimated from plotted models for [Fig. 5B](#F5){ref-type="fig"}, with exception of the x-value of the asymptote, which was estimated directly from [Fig. 5B](#F5){ref-type="fig"}, and crown lateral root initiation parameters, which were estimated from [Fig. 5D](#F5){ref-type="fig"}. Values are means (*n*=6−13); asterisks indicate a significant difference between NN and HN roots (paired *t*-test: \*\*\*, P\<0.001; \*\*, P\<0.01; \*, P\<0.05; (\*), P\<0.1; NS, P\<1.0). See [Fig. 3](#F3){ref-type="fig"} for an explanation of the parameters. Lateral root development {#s12} ------------------------ A typical response of cereals to locally applied N is the development of more lateral roots within the N rich patch ([@CIT0018]; [@CIT0007]; [@CIT0006]). To quantify this response over time, crown lateral root initiation was monitored after solution change, in both the part of the root system with high N and the part without N. Lateral root density on the crown root axis reached a maximum that was two to three times higher for the high N roots compared to the roots without N (1.0 laterals cm^-1^ without N, and 2.6 laterals cm^-1^ with N, respectively; [Fig. 5C](#F5){ref-type="fig"}). Moreover, the time to reach this maximum differed considerably (4 d without N, and 2 d with N, respectively; [Fig. 5C](#F5){ref-type="fig"}, [Supplementary Fig. S3](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). Elongation rates of the initiated laterals were also influenced by the presence of N: without N, laterals initiated, but did not elongate beyond 1−5mm. Laterals in the compartment with N had already a higher elongation rate at 1 DASC, and showed every day a further increase in elongation rate of, on average, 0.14cm d^-1^ ([Fig. 5D](#F5){ref-type="fig"}). This resulted in lengths of up to 18cm at 12 DASC, whereas the laterals without N only reached lengths of up to 1cm ([Supplementary Fig. S5](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). The maximum asymptote for lateral root elongation for most laterals was reached at approximately 8 DASC, at on average 1.5−2cm d^-1^. A few roots, however, did not show a clear maximum asymptote within the experimental period ([Supplementary Fig. S4](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). Correlation of root growth parameters with plant performance {#s13} ------------------------------------------------------------ We tested the correlation of the root dynamics during the different phases with final measurements at harvest. The stronger the elongation rate of crown axile roots declined before solution change (slope 1, [Fig. 3A](#F3){ref-type="fig"}), the lower was the final root and shoot dry weight. Per gram of increase in dry weight of the total plant, slope 1 decreased with 0.20cm d^-1^ ([Table 3](#T3){ref-type="table"}). This indicates that the strong differences in dry weight between plants were established during the first 2 weeks of the experiment. After solution change, no significant correlations occurred anymore between biomass parameters and elongation rates of crown axile roots (data not shown) supporting the hypothesis that variation between plants was established early. After solution change, the time point at which 50% of all crown laterals in the high N compartment were formed, was positively correlated with dry weight of the shoot. A 0.1g increase in dry weight of the shoot was associated with an almost half day earlier formation of 50% of all crown laterals in the high N compartment ([Table 3](#T3){ref-type="table"}). Thereafter, the acceleration of elongation of these laterals in the high N compartment was negatively correlated with dry weight of the roots in the same compartment (seminal plus crown root system; [Table 3](#T3){ref-type="table"}). With each gram of roots that was more present in the high N compartment, the slope of lateral root elongation decreased with 0.6cm d^-1^. ###### Correlations between parameters describing root development in a split-root system with half the root system supplied with no N (NN) and the other half with high N (HN) Parameter 1 Parameter 2 r~s~ Significance ---------------- ---------------- ------- -------------- Slope 1 DW total plant -0.76 \* Slope 1 DW roots -0.67 \* Slope 1 DW shoot -0.60 (\*) Slope 1 FW shoot -0.71 \* Reached 50% HN DW shoot 0.89 \* Slope lat HN DW roots HN -0.89 \* DW, dry weight; FW, fresh weight; 'Reached 50% HN', time point at which 50% of all laterals were formed in the high N compartment; 'Slope lat HN', acceleration of elongation of laterals in the high N compartment. If not indicated otherwise, no N and high N root parameters were taken together. Spearman's rank correlation coefficient (r~s~) indicates statistical dependence (*n*=6−9; (paired *t*-test: \*\*\*, P\<0.001; \*\*, P\<0.01; \*, P\<0.05; (\*), P\<0.1; NS, P\<1.0). See [Fig. 3](#F3){ref-type="fig"} for model estimated parameters. Effect of primary root removal {#s14} ------------------------------ In split-root experiments, the primary root may cause a bias, since it can only be placed in one of the compartments and thus affect root biomass in that compartment. In this study, the primary root of half of the plants was removed 2 d after transplantation to determine the influence on selective root placement. Significant effects occurred neither on the biomass of the shoot and total root system nor on the dry weight of the crown root system ([Supplementary Fig. S6](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). A significant 2-fold increase in dry weight of the total seminal root system was found when the primary root was removed ([Supplementary Fig. S6](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). However, when seminal root biomass in the separate compartments without and with high N was compared between plants with and without primary root, this difference disappeared (data not shown). The model-derived parameters, as well as crown root emergence, were not significantly affected by primary root removal. A trend indicated, however, a 2-fold higher elongation rate from 1.0 to 2.0cm d^-1^ at the breakpoint for crown roots grown without N, when the primary root was removed (data not shown). Despite the comparably high number of replications, insufficient data were available to proof the significance of the primary root placement in either the low or high N compartment for the growth dynamics of the remaining root system. Discussion {#s15} ========== N patches cause faster elongation of crown roots {#s16} ------------------------------------------------ The elongation of crown roots increased asymptotically upon exposure to high N to an average of 5.3cm d^-1^ at \~4 DASC. It is well established, that a heterogeneous distribution of N in the soil can lead to selective root placement resulting in a higher root length in the N patch than in the surrounding soil (e.g. [@CIT0018]; [@CIT0006]; [@CIT0014]; [@CIT0049]; [@CIT0038]). This response could, however, only be observed on the slightly long term, i.e. from daily images, and not on the short term, i.e. on the high-time-resolution CCD measurements. The CCD measurements did not reveal a short-term response by changes in root growth, whereas root growth started to increase 24−48h after solution change. This timing is in line with observations of [@CIT0045] where NO~3~ ^-^ transporters were up-regulated \~24h after exposure of Arabidopsis roots to N-rich medium. Similar dramatic increases in elongation rates were not observed in seminal roots in experiments by [@CIT0007] on barley. These roots grew from a low NO~3~ ^-^ compartment into a high NO~3~ ^-^ compartment, and showed only a slight increase 17.5 d after entering the high NO~3~ ^-^ compartment. This difference in response might be related to the internal N status of the shoot. Our maize plants were likely to be N starved from growing without N for 12 d, while the roots of the barley plants in the experiments of [@CIT0007], with exception of one seminal root, were growing on high NO~3~ ^-^, causing a higher internal N status, and therefore possibly a slower and less prominent response. This hypothesis is quite likely, as N starvation on one side of the root system, leading to an up-regulation of NO~3~ ^-^ uptake on the other side of the root system, is mediated by the shoot ([@CIT0045]). The reason for this indirect pathway may be a check for the N status of the shoot. We found that root elongation on the side without N declined until a 'change point' of 1.2cm d^-1^ \~6.3 d after solution change. This decrease is consistent with the decrease in elongation rates of the seminal root axes of barley growing from a low NO~3~ ^-^ compartment into another low NO~3~ ^-^ compartment ([@CIT0007]). Also here, a breakpoint was suggested for the elongation rate of these roots. Possibly, allocation of resources from the shoot to these roots in N-deficient conditions was gradually declining. N patches strongly stimulate lateral root production with most dramatic effects on lateral root elongation {#s17} ---------------------------------------------------------------------------------------------------------- The total lateral root length increased within the high N patch, which resulted from both longer lateral roots and an increase in lateral root density on the crown root axis. Selective root placement has most often been reported as an increase in lateral root production (e.g. [@CIT0018]; [@CIT0007]; [@CIT0014]; [@CIT0049]; [@CIT0038]). The increase of both length and densities of lateral roots is consistent with findings for barley ([@CIT0007]; [@CIT0006]), wheat ([@CIT0018]) and maize ([@CIT0017]), but is in contrast with findings for *Arabidopsis* where only lateral root length increased and not lateral root density ([@CIT0049]; [@CIT0051]). However, *Arabidopsis*, in contrast to the cereals barley, wheat and maize, only has laterals developing from the primary root, and therefore an entirely different type of root architecture. For maize it has, however, also been suggested that a local high NO~3~ ^-^ concentration does not affect lateral root density on the primary root axis ([@CIT0048]). This divergent observation might be caused by a difference in developmental stage, since very young seedlings were used in the latter study (7 d after germination), with a strong influence of seed reserves on root morphology as demonstrated by [@CIT0011]. The lateral root density in the 5cm observation zone increased to a maximum within 3−4 d. The final density under high N was with 2.3 roots cm^-1^ about two to three times higher compared to the density at the side without N. These values are still lower than those found in solution and sand culture for maize ([@CIT0048]). One explanation could be the delayed formation of new lateral roots. To verify this hypothesis, we counted lateral roots in the second half of the rhizoslide and found lateral root densities of 4.5 laterals per cm on the high N side and 1.1 laterals per cm on the side without N. However, the branching density observed by [@CIT0048] was still higher. This could probably be a result of the use of a different genotype, and cultivation systems with more contact between roots and substrate. In rhizoslides, only about half of the root is in contact with the paper substrate while the other half is exposed to the foil cover, which may limit development of laterals. [@CIT0023] however used a similar system and found \~7 lateral roots per cm primary axile root for most extreme genotypes. Since primordia numbers were not determined in the present study, it is unclear whether this was resulting from a higher percentage of primordia emergences or from the presence of more primordia. However, [@CIT0025] suggested that maize typically lacks dormant root primordia, favouring the second hypothesis. In the compartment without N, it took twice as long to reach the maximum lateral root density of the first order laterals on the crown axis compared to in the high N compartment. After reaching this maximum, no additional laterals were formed on that particular part of the crown root axis in both the compartments with and without N. This suggests that once first order laterals were formed, no additional laterals could be produced in response to the prevailing conditions, as also found for wheat ([@CIT0018]). Carbohydrate allocation may play an important role in the increase in lateral root densities on the root axis, as suggested by the increase in lateral root density in both *Arabidopsis* and wheat in response to glucose or sucrose feeding ([@CIT0002]; [@CIT0005]). Tobacco roots, however, did not show a similar response ([@CIT0044]). To date it is unknown if these effects were results of previous carbon limitation or if the sugars acted as signal molecules ([@CIT0013]). The most remarkable change in the whole root system was the prolonged elongation of the lateral roots exposed to high N, while those without N stayed short. The elongation rates for most laterals under high N in our experiment increased linearly to values as high as 2.0cm d^-1^. Some growth rates even proceeded to increase linearly until the end of the experimental period (15 DASC; [Supplementary Fig. S4](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1)). This is clearly not in line with previous work, where first order lateral roots of crown and seminal roots of maize were short, reaching only \~3cm in length ([@CIT0030]; [@CIT0022], [@CIT0021]) and only \~10% of the laterals reach a length greater than 10cm ([@CIT0036]). Usually an exponential decrease in elongation rates is assumed for first order laterals of maize ([@CIT0035]). Prolonged elongation rates are usually only known for the primary lateral root of some genotypes of maize ([@CIT0020], [@CIT0022]). It seems that the response to a spatially varying N concentration allows the root system to respond with prolonged and fast elongation of crown lateral roots. Compared to a study on barley ([@CIT0007]), we observed an elongation rate that was more than three times greater. An important factor in this may have been, again, the stronger N deficiency of the maize plants in our study. Removal of the primary root affects growth of the other root types {#s18} ------------------------------------------------------------------ The position of the singular primary root in a split-root setup, being either in the compartment with or without N, could potentially affect the response of other parts of the root system. This is why the primary root is often partially or completely removed at the start of the experiment (e.g. [@CIT0017]; [@CIT0041]; [@CIT0029]; [@CIT0042]; [@CIT0038]). Such removal of the primary root in our experiments had a significant positive effect on the size of the seminal root system, suggesting an early competition between primary and seminal roots for carbon. However, the final total size of the crown root system was not significantly influenced. Still, a 2-fold higher elongation rate of the crown roots without N at the breakpoint of the segmented regression was apparent for plants without their primary root. Removal of the primary root thus influenced the seminal and, to a lesser extent, the crown root system. A thorough consideration of how to cope with primary root position in split-root setups is therefore important, especially when studying the embryonic root system. No direct dependence of the dynamic of roots exposed to differential N supply {#s19} ----------------------------------------------------------------------------- The two to three times greater lateral root density on the crown root axis, the striking increase in lateral root elongation, and the fast increase in crown axile root elongation clearly indicate that the plants prioritized N uptake from the N patch. On the other hand, only few laterals were formed in the compartment without N, which ceased to elongate within the first day after formation. We correlated the estimated model parameters of the different root-response models to evaluate their dependency. There was no correlation between these model parameters determined under no N conditions and those determined under high N. For example, the time until 95% of the asymptotic value was reached for crown root elongation on the high N side was not correlated with the time until the break point was reached on the no N side. This indicates a large degree of independency. In the case of a direct dependency within the network of roots, we would have expected a correlation among different parameters. The lack of direct dependencies may be plausible when considering the shoot as an organ integrating signals from roots without N to roots with high N ([@CIT0045]). We measured shoot traits only at the end of the experiment. Nevertheless, the correlation between the early decline in root elongation rates and plant weight indicates that differences among plants established before solution change. After solution change, later larger plants seemed to be able to form lateral roots in response to N more rapidly. Clearly, a similar dynamic for the shoot would benefit the interpretation of these root-shoot-root interactions. For this reason, we added shoot imaging to the next version of the rhizoslide system. Perspectives for application {#s20} ---------------------------- The strong selective root placement demonstrates the exceptional plasticity of the maize crown roots, and the strong ability of the B73×UH007 hybrid to respond to local N patches. The mother inbred line B73 is one of the most successful inbred lines, being incorporated in many breeding programmes throughout the world. It may be speculated that the strong responsiveness of crown roots to N is part of its success, which would somehow contradict the philosophy to only select for unresponsive root systems, as proposed by [@CIT0028]. However, we think that the extreme responses observed here were likely caused by severe N-starvation in the first 2 weeks of growth. In a follow-up study, with 1.7mM N during the first 2 weeks, the response was more moderate and the length of the lateral roots was \~3cm as observed usually (data not shown). Most plants will seek for an optimization of N uptake to maintain organ growth depending on their nutrient status. Such foraging behaviour would support the concept that also a generally high responsiveness to locally varying nutrient supply could benefit the plant if coping with heterogeneous nutrient conditions. A modelling study by [@CIT0009] showed that a root system capable of strong selective root placement to N displayed a \>2-fold higher N uptake efficiency throughout the whole growing season, compared to a root system that was only poorly capable of selective root placement to N. This suggests that particularly when N fertilizer is applied side-banded during sowing, varieties with the ability of a strong selective root placement may benefit most and, subsequently, decrease NO~3~ ^-^ leaching ([@CIT0008]). Maize seems to be a suitable species to start such experimental systems, since it has been shown that it is capable of finding and colonizing a fertilized furrow in the field ([@CIT0043]; [@CIT0004]). Great potential thus seems to exist for improved N management through patch-wise N fertilization in such a cropping system. Only a few quantitative genetic studies focus on the characteristics of the crown root system ([@CIT0021]) and the limited research conducted on the post-embryonic root system leaves a big gap in our understanding of the plasticity and functioning of this most prominent root type of maize. The rhizoslide setup combined with the root growth modelling approach presented could be a first step in the direction of high-throughput screening allowing detection of QTLs responsible for favourable root traits under limited N; whereas the most time-intensive step is still the image analysis. As the lateral root tracing is time-intense, we would suggest counting branching densities and determination of lateral root length only at critical time points. For example, this study demonstrated that the last time point would be representative of branching density. To screen crown roots of a population of 1000 individual plants in the manner proposed here, the image analysis would entail around one month's work, a feasible option in our opinion. Supplementary data {#s21} ================== Supplementary data is available at *JXB* online. [Supplementary Figure S1](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1). Rhizoslide construction. [Supplementary Figure S2](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1). Root elongation rates of individual maize crown roots grown in a split-root rhizoslide system. [Supplementary Figure S3](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1). Lateral root density on the crown root axis of maize for roots grown in a split-root rhizoslide system. [Supplementary Figure S4](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1). Lateral root elongation rates of individual roots of maize of crown roots grown in a split-root rhizoslide system. [Supplementary Figure S5](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1). Crown lateral root length of maize plants over time grown in a split-root rhizoslide system with half the root system in no N and the other half in high N. [Supplementary Figure S6](http://jxb.oxfordjournals.org/lookup/suppl/doi:10.1093/jxb/erv307/-/DC1). Dry weight (DW) of the seminal and crown root system of maize plants grown in rhizoslides. ###### Supplementary Data We would like to thank Achim Walter for valuable advice on diurnal root growth measurements and Bertrand Muller and Philippe Nacry for their help and discussion concerning the method. Thanks to Delley Semences et Plantes SA for hybrid production and to Albrecht Melchinger, University of Hohenheim for supplying the UH007 inbred line. This research received funding from the European Community Seventh Framework Programme FP7-KBBE-2011--5 under grant agreement no. 289300 and the Erasmus Mobility Grant from the Lifelong Learning Programme of the European Union. [^1]: Editor: Roland Pieruschka

Introduction ============ *Helicobacter pylori*is a microaerophilic gram-negative organism involved in many digestive system diseases, such as peptic ulcer, gastritis, or mucosa-associated lymphoid tissue (MALT) lymphoma, or acting as a risk factor in the development of gastric cancer.^[@B1]^ The prevalence of *H. pylori* infection varies greatly among different countries, as in many developing countries it is over 70%, while in most industrialized nations it is 20% to 50%.^[@B2]^ Eradication of *H. pylori*is an important component of treatments for peptic ulcer disease and other gastrointestinal disorders.^[@B3]^Triple or quadruple therapy regimen containing a proton-pump inhibitor (PPI) and antibiotics, mainly clarithromycin and metronidazole, are currently in use.^[@B4]^ The inhibition of protein synthesis is the functional mechanisms of the macrolides, causing the separation of peptidyl-tRNA from the ribosome during the elongation reaction.^[@B5]^ One of the most common components of the *H. pylori* infections therapy regimens is clarithromycin. The resistance to macrolides such as clarithromycin in *H. pylori*has been demonstrated to occur at different rates (1 to 10%) in different countries, and is an important cause of *H. pylori* therapeutics regimens failure. Furthermore, macrolide-resistant *H. pylori*mutants are simply obtained by in vitro selection.^[@B5]^ Macrolide resistance is caused by several mechanisms such as the lack of macrolide binding to the ribosomal target, inactivation of the macrolides by enzymes, reduced or lack of bacterial membrane permeability, and macrolides active efflux.^[@B5]^ The widespread use of clarithromycin for the treatment of *H. pylori*infection has resulted in the development of resistance.^[@B6]^ Clarithromycin resistance (Cla^R^) of *H. pylori*is mainly caused by point mutations of the genomic 23s *rRNA*, the main component of the 50S subunit, mostly at position 2142/43 (A2142 to G/C/T; A2143 to G/C) in the peptidyl-transferase region of the V domain, thereby preventing drug binding. Cla^R^ is increasing due to widespread use of macrolides for other diseases in the western world.^[@B7]^ There are some methods to detect the point mutations in genes such as sequencing, and amplification and restriction fragment length polymorphism (RFLP). In this study we used the RFLP method to detect the point mutations in 23s *rRNA* gene in our local *H. pylori* isolates.^[@B8]^ Clarithromycin is recognized as the key antibiotic for the treatment of *H. pylori* infections, as has a powerful bactericidal effect in vitro compared with the other available macrolides.^[@B9]^ Therefore, the present study aimed at evaluating the Cla^R^ rates in local *H. pylori* isolates and the probable molecular mechanisms of such a resistance. Specifically, the study aimed at determining the most important point mutations in 23s *rRNA* gene that are closely related to clarithromycin resistance among *H. pylori*isolates in Kerman, Iran. Materials and Methods ===================== *Bacteria* Sixty three *H. pylori* isolates were obtained from 191 patients\' biopsy samples referred to the Endoscopy Division Unit of Afzalipour Hospital in Kerman, Iran. The biopsy samples were cultivated in Brucella Agar medium (Merck, Germany) supplemented with 10% defibrinated sheep blood (Darvash, Iran) and three antibiotics including Vancomycin (10 mg/l), Amphotricin B (10 mg/l) and Trimetoprim (5 mg/l) (Sigma, USA). The inoculated plates were incubated at 37°C under microaerophilic atmosphere provided by anerocult C (Merck, Germany) for 3-5 days. The isolates were recognized as *H. pylori* by urease, catalase, oxidase positive and gram negative staining tests.^[@B10]^ *Antibiotic Susceptibility Tests* The susceptibility of the isolates to clarithromycin was evaluated by disc diffusion method. There is no an standard method to evaluate the susceptibility of *H. pylori* to antibiotics. We used the clinical and laboratory standards institute (CLSI) -recommended method called Modified Disc Diffusion method. In this method a microbial suspension with turbidity equals to four McFarland (12 x10^8^CFU/ml) and cultivated in Muller-Hinton agar (Merck, Germany) supplemented with 10% defibrinated sheep blood (Darvash, Iran). The 2 μg clarithromycin disc (Mast, England) were placed in the plates and incubated in 37°C under microaerophilic atmosphere for three days. Any inhibition zone was considered susceptible.^[@B10],[@B11]^ *DNA Extraction* DNA was extracted from all 63 *H. pylori* isolates using Bioneer genomics kit for DNA extraction (Bioneer, South Korea) according to the manufacturer's instruction. *Amplification and Restriction Fragment Length Polymorphism (RFLP)* Two sets of primers were used in this study ([table 1](#T1){ref-type="table"}). ###### Primers used for amplifications. Primers CLA 18 and CLA 21 were used in polymerase chain reaction-amplification and restriction fragment length polymorphism (PCR-RFLP) to obtain a 1.4 kbp amplified fragment. Primers CLA 18 and CLA 3 were used in 3′-mismatched PCR to obtain a 700 bp amplified fragments. **Set 1** ----------- ----------------------- --------- ---- Cla18 AGTCGGGACCTAAGGCGAG 1400 bp 7 Cla21 TTCCCGCTTAGATGCTTTCAG (set2) Cla18 AGTCGGGACCTAAGGCGAG 700 bp 11 Cla3 AGGTCCACCACGGGGTCTTG The first set (cla18, cla21) was used to amplify a 1400 bp fragment from an internal region of 23s *rRNA* gene followed by digestion with *BsaI* & *MboII* (Fermentas, Lithuania). The 1400 bp fragment normally has one restriction site for *BsaI*enzyme. If the gene is wild type, the enzyme produces a 1000 bp and a 400 bp fragments. If the A2143G point mutation occurs in 1400 bp fragment, the enzyme find two restriction sites and produces three fragments: a 700 bp, a 400 bp, and a 300 bp one. The 1400 bp fragment normally has no restriction site for MboII enzyme, therefore, if the gene is wild type, the 1400 bp remains undigested. But, if the A2142G point mutation exist, the enzyme find one restriction site in the 1400 bp fragment and digest it to two 700 bp fragments that look as one overlapping band in electrophorsis gel.^[@B7]^ The second set (cla18, cla3) was used to do 3^\'^-mismatch PCR to detect A2142C point mutation. In this case, if the gene was of wild type there was no fragment, and if the A2142C point mutation took place, a 700 bp fragment was produced.^[@B12]^ Polymerase chain reaction condition was as follows for the amplification of the 1400 bp fragment: reactions were carried out in Primus thermo cycler (MWG-Biotech, Germany) in 50 μl mixtures containing 25 μl PCR master mix (CinnaGen Inc, Iran), 19 μl sterile deionized water, two μl template DNA and two μl of each oligonucleotide primer (4 μl totally). Initial denaturation at 94°C for five min followed by 30 cycles of denaturation at 94°C for one min, annealing for one min at 58°C, extension at 72°C for one min. The final extension step was extended to five min at 72°C. The RFLP protocol was as follows: 10 μl of the 1400 bp fragment was added to two PCR microtubes, and five units of each enzyme was added to the micritubes and incubated at 37°C for 16 hours. 3\'-mismatch PCR condition was as follow: reactions were carried out in Primus thermo cycler (MWG-Biotech, Germany) in 25 μl mixtures containing 12 μl PCR master mix (CinnaGen Inc, Iran), 10 μl sterile deionized water, one μl template DNA and one μl of each oligonucleotide primer. Initial denaturation at 94°C for five min followed by 30 cycles of denaturation at 94°C for one min, annealing for one min at 55°C, extension at 72°C for one min. The final extension step was extended to five min at 72°C. *Electrophorsis* The PCR products were separated on 1.5% and the PCR-RFLP products were separated on 2% agarose gels (Cinna gen, Iran) after being stained with ethidium bromide (Merck, Germany) in TBE 1X (Tris/borate/EDTA) buffer under 100 volts electricity flow. Bands were visualized under UV gel documentation and photographed. Results ======= Twenty out of 63 (31.7%) of the *H. pylori* isolates were resistant to clarithromycin. There was no significant relation between gender, age or the history of antibiotic consumption by the patients and resistance to calrithromycin. All of the 20 Cla^R^ isolates had at least one of the three common point mutation in 23s *rRNA* gene, while none of the Cla^S^ isolates had such a point mutation ([table 2](#T2){ref-type="table"}). ###### The frequency and (rate) of clarithromycin susceptibility test for *H. pylori* isolates in both resistant and sensitive isolates in Kerman, Iran. ------------------------------------------------------------------------------------------- **CLA susceptibility** **Number (%)** **23 s** ***rRNA*** **point**\ **Number (%)** **mutation** ------------------------ ---------------- -------------------------------- ---------------- R 20 (31.7%) \+ 20 (100%) \- 0 (0%) S 43 (69.3%) \+ 0 (0%) \- 43 (100%) ------------------------------------------------------------------------------------------- CLA: Clarithromycin, R: Resistant, S: sensitive All of the 63 *H. pylori* isolates were positive for the 1400 bp fragment ([figure1](#F1){ref-type="fig"}). Fifteen percent of the Cla^R^ isolates (three out of 20 isolates) had the A2143G point mutation ([figure 2](#F2){ref-type="fig"}). There was a significant relation between the gender of the patients and the A2143G point mutation. Three out of 38 (7.9%) of the strains isolated from the female population had this point mutation, whereas no such a mutation was found in the strains isolated from the male population. There was no significant relation between age or the history of antibiotics consumption and the A2143G point mutation. {#F1} {#F2} Fifty five percent of the Cla^R^ isolates (11 out of 20 isolates) had the A2142G point mutation ([figure 3](#F3){ref-type="fig"}). There was no significant relation between gender, age or the history of antibiotics consumption of the patients and this mutation. {#F3} ###### Results obtained with the PCR-RFLP and the 3′-mismatched PCR methods for the clinical isolates tested according to the clarithromycin resistance. **mutation** **Number (%)** **Digestion with: BsaΙ/MboΙΙ** **3\'mismatch PCR** **Cla susceptibility** -------------- ---------------- -------------------------------- --------------------- ------------------------ A2143G 3 (15%) +/- \- Cla^R^ A2142G 11 (55%) -/+ \- Cla^R^ A2142C 6 (30%) -/- \+ Cla^R^ CLa: Clarithromycin, R: Resistant Thirty percent of the Cla^R^ isolates (six out of 20 isolates) were positive for the A2142C point mutation ([figure 4](#F4){ref-type="fig"}). There was no significant relation between age, gender or the history of antibiotics consumption of the patients and this mutation. {#F4} The A2142C point mutations occurred only in Cla^R^ isolates without A2142G or A2143G ([table 3](#T3){ref-type="table"}). Discussions =========== Resistance of *H. pylori* to antibiotics has been increasing in most parts of the world including Iran.^[@B11],[@B13]-[@B15]^ Clarithromycin resistances is a serious concern for doctors who are using the drug as one of the most important therapeutic components for *H. pylori*-induced gastric ulcer. There are ever-increasing requests from physicians for a reliable standard antimicrobial susceptibility test for *H. pylori* against clarithromycin, but that would be hard to do because of its fastidious properties and its time-consuming culture. Furthermore, success in *H. pylori* culture is dependent on the microbiology laboratory technicians\' skills.^[@B16]^ Clarithromycin resistance rates are varied across the world. For example Elviss et al in London reported 11% resistance to clarithromycin,^[@B17]^ or Bagalan et al announced 27.6% resistance.^[@B18]^ Also, the rate of clarithromycin resistance varies in different cities in Iran. For example Kohanteb et al reported 9.4% resistance in Shiraz (2007),^[@B19]^ while Mohammadi et al showed 20% resistance to clarithromycin in Tehran (2005), and in more recent studies Siyavoshi et al (2010) in Tehran reported 7.3% resistance.^[@B11],[@B15]^ Clarithromycin is a macrolide, that due to its high prices, was not used commonly in Iran in the past years. However, after its production in the country in recent years, it has been used routinely in the treatment of *H. pylori* infections. So, the emergence of Cla^R^ isolates is inevitable. It is also has been shown that countries with a high consumption of other macrolides have a higher rates of clarithromycin resistance.^[@B20]^ Macrolides such as erythromycin and clarithromycin inhibit nascent peptide chain elongation by interacting with the 50S ribosomal subunit and stimulating the release of peptidyl-tRNA from the A site.^[@B21]^ Biochemical studies have demonstrated a direct interaction of clarithromycin and its chief metabolite, 14-hydroxyclarithromycin, with 50S ribosomal subunits isolated from *H. pylori*.^[@B22],[@B23]^ The antibacterial activity of clarithromycin is better than that of erythromycin. One reason for such a difference is the synergistic phenomenon between clarithromycin and one its metabolites 14-hydroxyclarithromycin, which leads to a considerable post antibiotic effect. The second reason is higher hydrophobicity of clarithromycin, which leads to a better penetration through the cell membranes than that of erythromycin. The third reason is clarithromycin activity, which is less influenced by acidity than that of erythromycin.^[@B23]^ Versalovic and colleagues were the first to announce that the clarithromycin resistance of *H. pylori*was associated with a point mutation in the V domain of 23S *rRNA*. They discovered A to G point mutations at positions identical to *E.coli*23S *rRNA* positions 2058 and 2059, and then called these positions 2143 and 2144 according to the entire *H. pylori*23S *rRNA* sequence.^[@B24]^ The present study focused on the three common point mutations, namely A2143G, A2142G and A2142C, which according to a sizable number of previous reports are the most common mutations associated with clarithromycin resistance. All of 20 Cla^R^ isolates had at least one of these three mutations. Therefore, there was an absolute association between these three point mutations in 23s *rRNA* gene and Clarithromycin resistance in the isolates. In agreement with the findings by Alarcon et al,^[@B16]^ the present study showed that the A2142C point mutation in 23s *rRNA* existed only on Cla^R^ isolates without A2142G or A2143G point mutations in 23s *rRNA* ([table 3](#T3){ref-type="table"}). A number of other investigator reported other point mutations in 23s *rRNA* gene that were associated with clarithromycin resistance as well. For example Hao et al. in China reported three novel point mutations including C2245T, G2244A and T2289C that were associated with clarithromycin resistance in their local isolates.^[@B25]^ Also, Khan et al. showed that T2182C point mutation in 23s *rRNA* was associated with clarithromycin resistance in Bangladesh.^[@B26]^ Therefore, it is important to realize that the three common point mutations that the present study focused on are not the only reason of clarithromycin resistance, and there could be some other point mutations in 23s *rRNA* gene associated with such a resistance. Some other mechanisms have been suggested for clarithromycin resistance, of which one is efflux pumps. Hirata et al. suggested a contribution of efflux pumps to the clarithromycin resistance in Japan.^[@B27]^ Since there was no significant relation between gender, age, or the history of antibiotics consumption of the patients and resistance to clarithromycin, it seems that spontaneous mutations are responsible for such a resistance among the microbial population. The importance of such a resistance was revealed when a number of studies reported that resistance to clarithromycin was equal to the whole therapeutic regime failure.^[@B28]^ Conclusion ========== The high rate of clarithromycin resistance in the isolates in the present study is a serious alarm, and in agreement with clinical colleagues\' views that many of their patients do not respond to clarithromycin anymore. Point mutations in 23s *rRNA* are closely related to such a resistance. With daily increase in the use of clarithromycin in therapeutic regime for *H. pylori* in Iran, the rate of *H. pylori* resistance rate to the drug is increasing. Therefore, it seems necessary to do antibiotic susceptibility tests for *H. pylori* before therapy begins. The authors would like to thank the Research Council of Kerman University of Medical Sciences for their financial supports. **Conflict of Interest:** None declared

1. Introduction {#sec1} =============== Simultaneous multisite recordings using Microelectrode Arrays (MEAs) coupled to cultured neuronal networks are a widely applied approach in the field of*in vitro* electrophysiology \[[@B1]--[@B3]\]. This technology overcomes the drawbacks of single-cell recording techniques, by sampling the electrical activity of a neuronal culture from multiple sites in a noninvasive way. This allows long-term studies of extracellular neuronal action potentials (i.e., spikes) and high frequency sequences of spikes (i.e., bursts) recorded at each electrode. Moreover, neuronal dynamics at the network level, such as array-wide bursts barrages (i.e., network bursts), can be observed thanks to the multiple sites of recording \[[@B1]\]. Given that typical electrode diameters are comparable to cells size or bigger (i.e., 20--30 *μ*m), each electrode is able to sense the extracellular activity of multiple cells simultaneously. As a consequence, specific signal processing methods (i.e., spike sorting algorithms) are needed to identify the network activity at a single-cell level, before proceeding with the analysis of spike trains (e.g., burst and network burst detection). Spike sorting algorithms are designed to address this task, based on the assumption that the coupling between an individual cell and its respective electrode creates a unique spike shape \[[@B4], [@B5]\]. Applications where the focus is on the spike timing analysis \[[@B6]\] or that require discrimination between the activities of neurons of different origin \[[@B7]--[@B9]\] are examples of studies that particularly benefit from spike sorting. Many algorithms with different levels of complexity and automaticity have been proposed to sort neuronal spikes \[[@B5]\]. Most of the methods start with a feature extraction step, where prominent spike waveform features are computed. After an optional dimensionality reduction, a clustering step is applied on the extracted spike features to identify groups of spikes belonging to the same cell \[[@B5], [@B10]\]. Other methods rely on the temporal matching or on the correlation of spike waveforms with spike templates, without the feature extraction phase \[[@B11], [@B12]\]. Despite many efforts to tackle the spike sorting problem, it is still difficult to identify the best algorithm with large generality and also to define which spike sorter is the most appropriate under specific circumstances \[[@B13]\]. As reported earlier, the experimental protocol, the acquisition setup, the culture type, and the unique coupling of cells and electrodes within the same culture shape the electrode data in a peculiar way, making it inevitable to use data-centric methods for the classification of neuronal spikes \[[@B10]\]. [Table 1](#tab1){ref-type="table"} reports the heterogeneity of some of the spike sorting algorithms found in the literature and in custom or commercial software \[[@B14]--[@B34]\]. Except for a few instances \[[@B31]\], these tools do not incorporate alternative methods for spike sorting process (i.e., feature extraction and clustering) but usually apply one method indiscriminately to all the electrodes \[[@B20], [@B33]\]. This prevents an immediate comparison of different algorithms on the same data. Neuronal spikes recorded with MEAs can be sorted with two different approaches: (i) off-line, which means that spikes are sorted after the acquisition and storage of raw voltage traces, or (ii) on-line, which means that spikes are sorted during data acquisition. In the first case, the information about the spikes collected throughout the recording is available to the algorithm; in the other case, only information available up to the current point in time can be exploited to sort a spike. Although in off-line modality spike waveforms are classified with a better accuracy \[[@B4]\], an extensive amount of time, together with a massive data transmission rate and storage space, is often required for this task \[[@B11]\]. An immediate data analysis is crucial not only in closed-loop situations (on-line real-time processing) \[[@B35]\] but also for open-loop long-lasting and/or high-throughput experiments, taking into account that standard MEAs can generate tens of gigabytes of raw data per day \[[@B24], [@B36]\]. Moreover, this approach facilitates for experimental scientists having a glance at the results of their ongoing experiment. On-line spike classification methods have been incorporated into few commercial (e.g., McRack software by Multi Channel Systems GmbH, Spike2 by Cambridge Electronic Design Ltd.) or custom acquisition software \[[@B30], [@B32]\], as shown in [Table 1](#tab1){ref-type="table"}. However, as stated for the vast majority of the off-line toolboxes, the available on-line implementations do not allow the user to select the appropriate algorithms to apply from a pool of methods. This aspect may be detrimental for sorting accuracy, due to the poor generality of spike sorting methods. To our knowledge it is difficult to find a spike sorting framework that incorporates alternative methods for all the spike sorting processing steps and that allows on-line analysis with any selected method. To this aim, in this work different spike sorting methods proposed in the literature and suitable for on-line analysis have been selected and integrated within a software environment familiar to the MEA users, that is, Matlab. The tool is to provide the users with the possibility to select a method according to the data set at hand, optionally different for each electrode of the same MEA. Here, the working principle of each algorithm is described together with metrics and the evaluation flow used to assess the performance of each method. To validate the tool, accuracy performances of the implemented methods on neuronal signals (both simulated and real recordings) are reported and discussed. Finally, the work reports a comparison of runtimes of the implemented spike sorting algorithms during MEA data acquisition with a test-bed setup, showing that on-line operations are feasible. 2. Materials and Methods {#sec2} ======================== 2.1. Algorithms {#sec2.1} --------------- An overview of the processing blocks and implemented methods is presented in [Figure 1](#fig1){ref-type="fig"}. ### 2.1.1. Spike Detection and Alignment {#sec2.1.1} To attain accurate spike sorting, spike waveforms have to be properly detected and aligned. The most common methods to detect spikes apply a threshold to the voltage of the input signal, computed as a multiple of the standard deviation of the signal over a predefined window. In this work, spike waveforms data provided as inputs to spike classifiers were obtained by means of an adaptive threshold-based algorithm, that is, "AdaBandFlt," fully described in \[[@B24]\]. This method updates the current noise level esteem every 1 second and detects spikes by comparison with both a positive and negative threshold (set as ±4 times the noise level), with appropriate checks to avoid detecting twice a single event \[[@B24]\]. When a sample crosses the threshold, a 3 ms window (1 ms before the threshold crossing and 2 ms after it) is applied to the signal and the result is saved as a spike. Spikes are then aligned to the point of the maximum amplitude \[[@B5], [@B20], [@B37]\]. ### 2.1.2. Algorithms for Building Blocks of Spike Sorting Process {#sec2.1.2} Four feature extraction algorithms, coupled to suitable dimensionality reduction methods and to three feature clustering methods, and one waveform clustering method (i.e., without a feature extraction phase) have been selected for implementation ([Figure 1](#fig1){ref-type="fig"}). *(1) Feature Extraction (FE)*. The aim of FE step is to extract representative features from a set of events to judge the differences between spike waveforms. The following clustering is facilitated if spikes are projected into compact and distant groups in the feature space \[[@B5]\]. [Table 2](#tab2){ref-type="table"} reports details about the four FE methods included in this work, pertaining their analysis domain, their occurrence in spike sorting publications, and the need of an off-line training phase before the on-line application of the methods. *(a) Principal Component Analysis (PCA)*. Principal component analysis is the benchmark feature extraction method in spike sorting of neural signals \[[@B38]\]. To obtain spike features, eigenvalue decomposition of the covariance matrix of the data (i.e., spike waveforms) is computed. Then, eigenvectors associated with the directions onto which projected data display the largest variance are used as the "principal components" (PCs). Each spike is represented by a series of PC coefficients *c* ~*i*~:$$\begin{matrix} {c_{i} = {\sum\limits_{n = 1}^{N}{{PC}_{i}\left( { n} \right)\ast s\left( { n} \right)}},} \\ \end{matrix}$$where *s*(*n*) is the waveform of the spike, PC~*i*~ is the *i*th PC, and *N* is the number of samples contained in a spike waveform. The dimensionality is usually reduced by only keeping the first few PCs, thus decreasing the computation time of PC coefficients calculation \[[@B5]\]. The computation of PCs matrix is usually performed using the entire set of acquired spikes. However, the method can be applied on-line by computing the PCs on a first set of acquired spikes (training set) and then performing a dot product between the PCs matrix and each new incoming waveform \[[@B5], [@B30]\]. *(b) First and Second Derivative Extrema (FSDE)*. A feature extraction method based on the first and second derivative calculation of the incoming spike profile was recently proposed as an efficient and low computational complexity solution \[[@B33]\]. The derivatives are computed as the difference between successive sample points, according to ([2](#EEq2){ref-type="disp-formula"}) for the first derivative (FD) and ([3](#EEq3){ref-type="disp-formula"}) for the second derivative (SD). Consider $$\begin{matrix} {FD\left( { n} \right) = s\left( { n} \right) - s\left( { n - 1} \right),} \\ \\ \end{matrix}$$ $$\begin{matrix} {SD\left( { n} \right) = FD\left( { n} \right) - FD\left( { n - 1} \right),} \\ \\ \end{matrix}$$where *s*(*n*) is the waveform of the spike. The positive peak of the first derivative (FD~max~) and the negative and the positive peaks of the second derivative (SD~min~ and SD~max~) are then extracted as spike features. This method does not require a training phase or a dimensionality reduction step, which makes it very suitable for on-line operation. *(c) Geometric Features (GEO)*. Geometric features relate to the shapes of the spike waveforms and usually require low complexity operations \[[@B4]\]. In this work, seven geometric features have been considered, including (i) the positive amplitude, (ii) the negative amplitude, (iii) the peak-to-peak amplitude, (iv) the positive area, (v) the negative area, (vi) the ratio between the positive and negative area, and (vii) the maximum slope. After feature computation, a dimensionality reduction step determines which subset of features best represents the differences among waveforms. As for PCA, on-line GEO feature extraction can be computed after a training step on a first set of spikes. *(d) Discrete Wavelet Transform (DWT)*. The DWT is a multiresolution technique which provides good time resolution at high frequencies and good frequency resolution at low frequencies. It has been shown to have the potential of outperforming PCA when spikes differ mainly in small details which are not captured by the few first components \[[@B39]\]. As in PCA, performing the DWT on a spike waveform provides a set of coefficients which can then be reduced and clustered to achieve spike classification. The coefficients obtained are the result of subsequent high- and low-pass filtering of the spike waveform at *j* different scales (i.e., filter bandwidths). The spike waveforms can then be described with a vector *v*:$$\begin{matrix} {v = \left\{ { a_{- j},d_{- j},d_{- {\{{j - 1}\}}},\ldots,d_{- 1}} \right\},} \\ \end{matrix}$$where *a* ~−*j*~ is the output of the low-pass filter (approximation coefficients) at the *j*th scale and *d* ~−*j*~, *d* ~−(*j*−1)~,..., *d* ~−1~ are the outputs of the high-pass filter at each scale (detail coefficients) \[[@B40]\]. In this work, Haar wavelet has been chosen as the default wavelet option, thanks to its computational efficiency and similarity to biphasic spike shapes \[[@B5], [@B20]\]. Sincethe DWT yields the same number of coefficients as samples in the original spike, a dimensionality reduction step is needed. As for PCA and GEO, the DWT-based FE can be applied on-line: the best subset of coefficients is chosen on a training data set and then these coefficients are retrieved from the DWT of incoming new spikes. *(2) Dimensionality Reduction*. Dimensionality reduction is a critical step in spike sorting. Indeed, adding dimensions improves the accuracy of the subsequent clustering only up to a certain point, after which adding more dimensions can cause the performance of the clusterer to degrade \[[@B5]\]. One reason for this may be that dimensions along which the data points are not well separated introduce noise or confusion into the clusterer. In this work, after PCA, the first three PCs are retained to reduce dimensionality, which were known to capture most of the spike waveform variance \[[@B13]\]. To select the most suitable combination of both GEO and DWT features we implemented the "maximum difference" method proposed by Gibson and colleagues \[[@B13]\]. This method provided good accuracy and low complexity, thus reducing the time needed for data reduction. Briefly, given a vector containing features from an incoming spike, this algorithm computes the element by element difference between this vector and the vector related to the previous spike and memorizes the indices corresponding to the three largest difference values. The coefficients which presented the highest variability over a training set of spikes are identified as the coefficients that will be given as input to the clustering phase. *(3) Spike Clustering*. An ideal clustering algorithm for on-line analysis should be (i) automatic (i.e., no need to set the number of clusters*a priori*), (ii) nonparametric (i.e., the user does not have to set arbitrary threshold values for algorithm parameters), and (iii) able to classify the incoming spikes without*a priori* knowledge. In this work, we selected for implementation two feature clustering methods (fuzzy-*C*-means and density-based clustering) and one template matching method (O-sort) fulfilling most of these requirements and the *K*-means, a benchmark off-line feature clustering method. [Table 3](#tab3){ref-type="table"} summarizes the features of the selected methods. The following section details the implementation of each method in the proposed framework. *(a) K-Means*. *K*-means clustering assigns a data point to a cluster according to the minimum Euclidean distance between the point and the centroid of the cluster. The main benefit of using this method in spike sorting is that it is a very simple and fast algorithm \[[@B5]\], yet it is supervised, since it requires inputting the number of clusters, usually unknown in advance. Different approaches have been proposed to automatically determine the number of clusters \[[@B41], [@B42]\]. Here we chose to maximize a structural index, that is, the "PBM index":$$\begin{matrix} {PBM\left( { k} \right) = \left( {\frac{1}{K}\ast\frac{E_{1}}{E_{k}}\ast S_{k}} \right)^{2},} \\ \end{matrix}$$where *K* is the number of clusters; *S* ~*k*~ is the maximum separation (i.e., distance) between cluster centers; *E* ~1~ is the sum of distances between all the points and the cluster center when *k* = 1; *E* ~*k*~ is the sum of distances between the feature points and the *k*th cluster center \[[@B42]\]. In principle, it would be possible to adapt the algorithm for on-line operation, including a training period which defines the cluster centroids on a set of spikes and determines the number of clusters, followed by an on-line classification, which computes the distance of each new spike from the centroids. However, this would only be appropriate for stationary data because of its "hard" clustering \[[@B5]\]. *(b) Fuzzy-C-Means (FCM)*. In fuzzy clustering, every spike belongs to all possible classes, with different degrees of membership. The higher the membership of a spike to a given class is, the more likely the spike belongs to that class. Thus, the classification is performed according to the membership values of a spike to each cluster, which depends on the Euclidean distance from the cluster centroid and on the degree of fuzziness \[[@B43]\]. In the defuzzification phase, a spike is labeled according to the maximum membership value, provided that it overcomes a minimum threshold, otherwise the spike is not classified \[[@B16]\]. Unlike the *K*-means clustering, the "fuzziness" of the classification makes the method more suitable for on-line clustering, because it accounts for the varying nature of the data and does not classify outliers (noise) thanks to defuzzification. After a training period in which the centers of the cluster are identified, it is possible to perform the classification of the incoming spike by defuzzification \[[@B30]\]. To automatically define the number of clusters (*C*) during the training, the classification can be performed with different values of *C* and then, for each cluster configuration, the fuzzy clustering validity (*S*) is computed and minimized with respect to *C* \[[@B16]\]. This index is expressed as the ratio between the compactness and the separation of the clusters and can be explicitly written as in$$\begin{matrix} {S = \frac{\sum_{i = 1}^{c}{\sum_{j = 1}^{N}{\mu_{ij}^{m}\left\| {c_{i} - s_{j}} \right\|^{2}}}}{N\ast{\min_{ik}\left\| {c_{i} - c_{j}} \right\|^{2}}},} \\ \end{matrix}$$where *μ* ~*ij*~ ^*m*^ is the membership value of spike *j* belonging to cluster *i*, *m* is the degree of fuzziness (*m* \> 1), *N* is the total number of spikes, and *c* ~*i*~ and *s* ~*j*~ are the center of cluster *i* and the position of spike *j*, respectively. In this work, the membership threshold for defuzzification was empirically set to 1/*C* + 0.1 (e.g., if *C* = 2 and the maximum membership value of a spike is lower than 0.6, the spike is not classified). *(c) Density-Based Clustering (DBC)*. This feature clustering method was proposed by Cheng and colleagues \[[@B26], [@B27]\] as a classification method suitable for on-line operation and not requiring the setting of any threshold value (i.e., nonparametric). Moreover, in contrast to FCM and *K*-means algorithms, which provide round shaped clusters due to the Euclidean metric, no assumption on the shape of the cluster is made by DBC. The algorithm starts with a training phase in which a density distribution is built from the feature space. To this aim, each dimension of the feature space is firstly quantized into *N* different levels; then the feature space is divided into *N* ^*d*^ cells (*d* = dimensionality of the features). When a spike is detected, the density value of the cell containing the projected spike and of its surrounding cells increases according to a discrete spatial kernel. After computing the density distribution of the training spikes, a label is given to the cells corresponding to a local maximum of the density and to their surrounding cells. As the result, the entire density map is divided in an unsupervised way into several clusters corresponding to the local peaks of the density. Then, a further step can be introduced to merge the initial clusters \[[@B26], [@B27]\]. In the proposed implementation, two clusters are merged if the distance between their peaks and the peak density values are below a threshold. During on-line classification, an incoming spike is classified immediately by the look-up-table, provided that it falls within the mapped volume of the feature space (otherwise it is not classified). *(d) O-Sort*. O-sort is a template matching algorithm presented by Rutishauser and colleagues \[[@B11]\]. The method is unsupervised, automatic, and designed to run in on-line mode, without the need of a training phase. The distance between an incoming spike and the already stored mean waveforms is computed. Then, the minimum of this distance indicates the class of the spike, provided that this is lower than a value (*T* ~*M*~), otherwise a new cluster is introduced. If the distance between any cluster pair is lower than a value (*T* ~*S*~), the clusters are merged into a single one. The authors use *T* ~*M*~ = *T* ~*S*~ = *N*〈*σ*〉^2^ to compute the thresholds, where *N* is the number of samples for each spike and 〈*σ*〉^2^ is the variance of the data computed over a sliding window (\~1 min). Moreover, they suggest to multiply *T* by a correction factor, *c*, to take into account systematic changes of spike shapes \[[@B11]\]. Different from FCM and DBC, which may not include a new spike in any class, O-sort assigns a class to every input spike. To evaluate the method we discarded clusters with a number of spikes lower than a threshold (i.e., 5 spikes in 1-minute long signal), in order to account for the presence of small clusters of noise. 2.2. Test Data {#sec2.2} -------------- To validate the implemented code, the methods have been applied to simulated signals and real MEA recordings. *(1) Simulated Data*. Simulated raw data traces were generated by using average spike waveforms obtained from spontaneous activity recorded at 25 kHz from hippocampal and cortical dissociated cultures in our laboratory (see \[[@B44]\] for a reference about culturing and recording protocol), as done in \[[@B13], [@B20], [@B31]\]. In order to have a "ground truth" against which algorithm outcomes were compared, each waveform had been previously labeled according to the classification performed by experienced researchers. To simulate background noise, MEA signals without spikes were randomly selected from a set of 42 electrodes raw data. Noise was band-pass-filtered (2nd order Butterworth filter) between 200 Hz and 3000 Hz, and then it was normalized and overlapped to the simulated waveforms. Each signal contains waveforms belonging to either 2 or 3 different groups and a number of spikes between \~600 and 1000 over 60 seconds (i.e., average firing rate of single units between 3 Hz and 8 Hz, to mimic real recordings \[[@B44]\]). The occurrence of spikes were determined randomly from a uniform distribution, setting a refractory period of 2 ms. Three different levels of signal-to-noise ratio (SNR) were simulated (i.e., 2, 3, and 4 \[[@B20], [@B24]\]) rescaling the noise on the simulated signal to mimic the realistic levels of*in vitro* MEA recordings. SNR was defined as the ratio between the mean amplitude of the spike waveforms in the signal and the mean of the peak-to-peak amplitude of noise calculated over a 1-second window. A total of 36 data sets were thus obtained, whose features are reported in [Figure 2](#fig2){ref-type="fig"}, together with the Bray-Curtis similarity index, a measure of the similarity among the waveforms in each data set (see ([7](#EEq7){ref-type="disp-formula"})). Consider $$\begin{matrix} {BCS\left( { x,y} \right) = 1 - \frac{\sum_{i}^{}\left| {x\left| i \right| - y\left| i \right|} \right|}{\sum_{i}^{}{\left| {x\left| i \right|} \right| - \left| {y\left| i \right|} \right|}},} \\ \end{matrix}$$where *x* and *y* are the two spike waveforms being compared and *N* is the number of sample points. BCS is computed between all the possible neuron pairs and then averaged. BCS lies in the range (0-1), with 1 corresponding to identical signals \[[@B33]\]. *(2) Real Data*. To test the algorithm performances in a realistic scenario, where the "ground truth" is unavailable, data sets of real MEA recordings were used. Extracellular recordings from hippocampal and cortical cultures grown on 60-channel MEAs (Ti200/30iR, Multi Channel Systems GmbH) had been previously carried out with a 60-channel MEA acquisition system (MEA1060 and USB-ME64, Multi Channel Systems GmbH) \[[@B44]\]. Recordings were sampled at 25 kHz and filtered (200 Hz--3 kHz, 2nd order Butterworth) before spike detection. The data set used in this work is composed of waveforms from 10 signals 120 seconds long. 2.3. Implementation of the Framework {#sec2.3} ------------------------------------ The algorithms were implemented and evaluated in Matlab (version R2008b, The Mathworks). Scripts are in Matlab native language apart from part codes written in C language and running in Matlab as MEX-files. Source code for MEX-files was written using Microsoft® Visual C++ 2008 Express Edition. Graphical user interfaces (GUI) were designed using the graphical user interface development environment (GUIDE) of Matlab. To convert the file format generated by the acquisition software of our commercial acquisition platform (*∗*.mcd) to Matlab format (*∗*.mat), the "Neuroshare" API library of functions has been employed (freely downloadable at <http://neuroshare.sourceforge.net/>), as done in other MEA analysis frameworks \[[@B45]\]. 2.4. Performance Evaluation {#sec2.4} --------------------------- The evaluation scheme adopted to assess the performance of the algorithms on simulated and real signals is depicted in [Figure 3](#fig3){ref-type="fig"}. For methods requiring a training phase, the training was performed on the first detected spikes (\~1/3 of the total number of spikes) for each signal, and then spikes were classified on the fly. ### 2.4.1. Indexes of Performance {#sec2.4.1} To test the effectiveness of spike sorting methods on simulated data sets, two indexes were employed: \(1\) Cluster validity (CV), which is the ratio of the between-cluster to the within-cluster distance. This index was calculated as in ([8](#EEq8){ref-type="disp-formula"}) using the "ground truth" labels of the data points:$$\begin{matrix} {CV = \frac{\min_{i = 1\cdots K,\,\, j = i + 1\cdots K}\left\| {{CC}_{i} - {CC}_{j}} \right\|^{2}}{\left( {1/N} \right){\sum_{i = 1}^{K}{\sum_{A \in N_{i}}\left\| {A - {CC}_{i}} \right\|}}^{2}},} \\ \end{matrix}$$where CC~*i*~ is the center of cluster of spikes produced by neuron *N* ~*i*~, *N* is the total number of spikes, *K* is the number of neurons simulated in the recording, and *A* is the feature vector. The higher the cluster validity, the better the classes separation, which eases the task of the following clustering step \[[@B42]\]. \(2\) The rate of classification accuracy (CA) is defined as the percentage of the number of correctly classified spikes over the total number of input spikes \[[@B29]\], as in$$\begin{matrix} {CA\% = \frac{\text{number  of  correctly  classifies  events}}{\text{total  number  of  spikes}}\ast 100} \\ \end{matrix}$$The accuracy of the methods was evaluated without discarding the false positives of the detection (total number of spikes = true positives + false positives). In this way, we tried to mimic the real condition, where unavoidably a small percentage of the detected spikes represents noise. Thus, CA could reach 100% only if all true spikes are correctly clustered and if no false positives exist or all the false positives are left unclassified by the sorter. To compare algorithm performances on real signals, two measures were used: \(1\) Intracluster variance (ICV), given by the following \[[@B12]\]:$$\begin{matrix} {ICV = \frac{1}{N_{i}}{\sum\limits_{j = 1}^{N_{i}}\left( { v_{j} - \mu_{i}} \right)^{2}},} \\ \end{matrix}$$where *v* ~*j*~ is the *j*th spike in the *i*th cluster, *μ* ~*i*~ is its mean template, and *N* ~*i*~ is the number of spike in the cluster *i*. Thus, the lower the value of ICV (meaning that the cluster is compact), the lower the probability of misclassification in each cluster. \(2\) Comparison with expert\'s visual inspection of the results \[[@B46]\], for which two parameters were defined. An expert was requested to judge the number of true spike clusters in each raw data set, before classification, and the number of identified clusters containing evidently similar and spike-like waveforms. The ratio of the latter over the first number (between 0 and 1) was used as an indication of the ability to group together true similar spikes, leaving false spikes into isolated clusters or outlier points (which could be excluded in a later off-line analysis). A second parameter was the number of unlabeled data points (i.e., spikes not assigned to any group). ### 2.4.2. Statistical Analysis {#sec2.4.2} A statistical analysis has been carried out to highlight relevant differences in the performance of the different methods on the data sets by means of Statistica (StatSoft Inc.). Each group presented as input to the statistical analysis consisted in the values of a performance index ([Section 2.4.1](#sec2.4.1){ref-type="sec"}) obtained after applying a certain method to all signals. Having assessed nonnormality of distributions (the Lilliefors test), a nonparametric multiple dependent comparison of the different groups has been performed (Friedman\'s test with Wilcoxon\'s matched pair test as*post hoc*) \[[@B42]\]. The significance level was established at *p* \< 0.05 for Friedman\'s test and *p* \< 0.01 for Wilcoxon\'s test. Data are given as median and its variation is stated as differences between 75th and 25th percentile (i.e., interquartile range, IQR). 2.5. Runtime Evaluation {#sec2.5} ----------------------- Runtimes of the different spike sorting algorithms were compared in Matlab on the same dedicated desktop computer (quad-core 3.3 GHz CPUs with 4 GB RAM running Windows 7 64-bit). The algorithms were launched from a custom script including code for the real-time communication with a MEA A/D device (USBME-64, Multi Channel Systems GmbH), through a proprietary dynamic-link library distributed by Multi Channel Systems ("McsUsbNet.dll"). Thus, it was possible to evaluate the effective feasibility of an on-line implementation, taking into account the time required for raw data reading, filtering, spike detection, spike sorting, and storage of results. Runtimes were computed in a worst-case scenario, simulated by the occurrence of a high frequency spiking signals simultaneously in all the 60 channels (i.e., 250 Hz \[[@B44]\]). The evaluation was performed with a sampling frequency of 25 kHz and with varying the length of serially transmitted data blocks from the A/D device (i.e., 100 ms--3 s). Simulated signals were repeatedly loaded in the code at each new iteration. Time needed to perform the training for spike sorting (i.e., in every method apart from FSDE and O-sort) was not considered in this evaluation, supposing performance of the training off-line on a first data stream. Runtimes were computed with the Matlab functions "tic" and "toc.". Reported data are averaged over 100 repetitions and related to one Matlab process running on one CPU (i.e., no parallel Matlab processes) \[[@B47]\]. 3. Results {#sec3} ========== 3.1. Graphical User Interface {#sec3.1} ----------------------------- The algorithms described in [Section 2.1](#sec2.1){ref-type="sec"} were incorporated into a GUI to speed up and ease the application and the evaluation of the methods by nonexpert users. [Figure 4](#fig4){ref-type="fig"} sums up the main functionalities of the GUI. The GUI is fully described in the guide provided as Supplementary Material available online at <http://dx.doi.org/10.1155/2016/8416237>. From a main menu (Figure 1 of Supplementary Material), two different modalities can be selected: (i) test data and (ii) real data. The "test data" interface is dedicated to the application and evaluation of spike sorting algorithms on simulated signals (Figure 2 of Supplementary Material). Specifically, the user can build a simulated signal giving as inputs the number of sources in the signals, the neurons average firing rates, and the SNR. The "real data" modality is dedicated to the analysis of real MEA data off-line (Figures 5, 6, and 7 of Supplementary Material) or on-line (Figure 10 of Supplementary Material). Both the test data and the real data "off-line" GUI allow selecting which algorithms to apply and setting all the necessary parameters for algorithm functioning (default values are the ones used in this work and reported in [Section 2](#sec2){ref-type="sec"}). After loading a signal, it is possible to perform a training on an arbitrary chosen number of spikes or to load the results of a previously performed clustering and use them for classification. Besides manual selections of the methods, the GUI embeds an automatic routine which runs all the possible combinations of spike sorting blocks on a selected signal and displays the performance indexes (i.e., CV and CA for simulated data, ICV and CV for real data) (Figure 4 of Supplementary Material). For a selected method, the GUI shows the performance indexes, the spikes projected and clustered in the feature space, the aligned spike waveforms, color-coded according to the clustering results, and the raster plots of each identified unit, as shown in [Figure 5(a)](#fig5){ref-type="fig"}. A *∗*.txt file can be generated as a report of the analysis, together with *∗*.JPEG figures of spike sorting results (e.g., raster plots of one or multiple electrodes, grouped spike waveforms superimposed to the MEA layout, and interspike-interval histograms for each source) as in Figures [5(b)](#fig5){ref-type="fig"} and [5(c)](#fig5){ref-type="fig"} (see also Figures 3, 8, and 9 of Supplementary Material). The real data "on-line" GUI (Figure 10 of Supplementary Material) allows the user to select the length of data flow blocks and the sampling frequency, establishes an on-line communication with the A/D device, and performs on-line spike detection, sorting, and data storage. The GUI and the scripts were written with Matlab R2008b, but they are compatible with all the following releases up to version R2014a. The framework is freely available upon request to <alessandra.pedrocchi@polimi.it>. 3.2. Spike Sorting Accuracy {#sec3.2} --------------------------- ### 3.2.1. Simulated Data Sets {#sec3.2.1} *(1) Feature Extraction*. [Figure 6(a)](#fig6){ref-type="fig"} shows an example of the projection of simulated neuronal spikes on different feature spaces (i.e., PCA, DWT, GEO, and FSDE). Concerning the DWT-based feature extraction method, DWT with 3, 4, or 5 decomposition levels was performed and the feature extraction performance in terms of cluster validity (CV) was evaluated projecting 3, 4, or 5 coefficients in the feature space. Since three and four decomposition levels provided statistically highest CV values, the three-level DWT algorithm has been selected to keep a lower complexity. The same consideration led us to the selection of a 3D feature space for the GEO method. Thus, results reported here about DWT and GEO method were obtained with the above-mentioned settings. The FE effectiveness assessed on each signal with the different feature extraction methods is shown in [Figure 6(b)](#fig6){ref-type="fig"}. Overall, the DWT method yielded the best performance on 12 signals (including 5 signals with SNR = 4, 3 signals with SNR = 3, and 4 signals with SNR = 2). Also the GEO method provided good results at different noise levels, being the best method on 14 signals including 4 with SNR = 4, 6 with SNR = 3, and 4 with SNR = 2. The PCA yielded the best separability for 9 signals (2 with SNR = 4, 3 with SNR = 3, and 4 with SNR = 2). Finally, the FSDE method yielded the poorest performance, being susceptible to the noise corrupting the extracted waveforms. But it was the best solution for 1 high SNR signal (i.e., signal \#10). All the methods showed a similar trend of reduced performance when the noise level was increased ([Figure 6(c)](#fig6){ref-type="fig"}). In particular, CV values yielded by PCA decreased by 30% when the SNR was lowered from 4 to 2, while CV values yielded by the other FE methods decreased by 40%. For PCA, DWT, and FSDE FE methods, the waveforms similarity index (see [Section 2.2](#sec2.2){ref-type="sec"}) was not a distinguishing factor to obtain a good separability. This parameter yielded a visible impact on the GEO method, whose CV values dropped in cases of high waveforms similarity (e.g., signals \#7-8-9 and \#25-26-27). Overall, the DWT and the GEO feature selection yielded a comparable CV (*p* \> 0.01), higher than the temporal PCA feature selection and the FSDE method (*p* \< 0.01), as shown in [Figure 6(d)](#fig6){ref-type="fig"}. The application of the most effective feature extraction method for each signal resulted in the statistically highest CV as compared to the application of the same method to all the signals ([Figure 6(d)](#fig6){ref-type="fig"}, black box-plot). *(2) Clustering*. Besides the specific clustering algorithm, the accuracy of a clustering method depends on the separability of features and their distribution/shape in the feature space. To first focus on the effect of the previous FE step, all the considered FE algorithms were coupled to the benchmark *K*-means clustering and then compared. As an example of this, [Figure 7(a)](#fig7){ref-type="fig"} illustrates the classification performed by *K*-means on the same data set of [Figure 6(a)](#fig6){ref-type="fig"} in each of the feature spaces. [Figure 7(b)](#fig7){ref-type="fig"} shows the trend of classification accuracy (CA) yielded by *K*-means coupled to the 4 FE methods across all the simulated data sets. *K*-means coupled to both DWT and PCA achieved statistically the highest and comparable classification performances (*p* \> 0.01). The high accuracy of DWT+*K*-means combination is ascribable directly to the high separability provided by DWT ([Figure 6(d)](#fig6){ref-type="fig"}). Even if PCA did not provide the best CV values ([Figure 6(d)](#fig6){ref-type="fig"}), the high accuracy of the PCA+*K*-means combination can be explained by the almost spherical shapes of clusters in the feature space (e.g., [Figure 7(a)](#fig7){ref-type="fig"}), which best matches the *K*-means classification. On the contrary, *K*-means clustering applied on GEO features yielded poor accuracy, most likely because the clouds of spikes created in the GEO feature spaces, even if compact ([Figure 6](#fig6){ref-type="fig"}), have no spherical shapes (e.g., in [Figure 7(a)](#fig7){ref-type="fig"}). Clustering of FSDE features provided the worst results when compared to any other method, due to the poor separability already shown in [Figure 6](#fig6){ref-type="fig"}. As mentioned earlier, part of the results shown in [Figure 7](#fig7){ref-type="fig"} depends on the specific properties of the *K*-means clustering. We have compared the classification accuracy of all the tested FE methods combined with different clustering methods, and obtained the performances shown in [Table 4](#tab4){ref-type="table"}. The performances of FCM clustering are shown with respect to two different degrees of fuzziness (i.e., *m* = 1.1 and *m* = 3). Overall, FCM with *m* = 3 provided statistically higher or comparable accuracy compared to FCM with *m* = 1 (i.e., a value which makes FCM similar to the *K*-means method \[[@B43]\]). The performances of the DBC method are reported in the case of two different resolution levels of the 3D-LUT (i.e., *N* = 16 and *N* = 32) and of resolution *N* = 32 with an additional automatic cluster merging step. The highest accuracy levels were obtained in the third configuration. The lower resolution of the density map led also to acceptable clustering, but very close clusters could not be distinguished. The worst performance was obtained building the LUT with *N* = 32 without the cluster merging step. In this case the outcome is affected by overclustering, since a lot of little clusters were identified. Results of the statistical comparison of clustering methods applied after a given FE method are presented in [Table 4](#tab4){ref-type="table"}. After PCA, the application of DBC or FCM provided an accuracy comparable to the benchmark *K*-means clustering (median CA \> 97%). Concerning DWT, the combination with FCM yielded accuracy values comparable to *K*-means (median CA \> 95%). GEO features achieved the best results (median CA \> 88%, statistically higher than CA achieved with *K*-means) if coupled to DBC, which is suitable to identify the clusters even if they have no spherical shapes. Only the FSDE method could not reach satisfactory accuracy with any of the clustering methods (median CA \< 70%) on our data. The statistical analysis applied to all the possible combinations of methods confirmed the absence of a unique method outperforming the other when applied indiscriminately to all the signals. Indeed, PCA+*K*-means, PCA+FCM, DWT+*K*-means, DWT+FCM, PCA+DBC, and O-sort yielded a comparable accuracy ([Figure 8](#fig8){ref-type="fig"}). However, these performances were statistically lower compared to the utilization of the best clusterer for each signal ([Figure 8(a)](#fig8){ref-type="fig"}), as illustrated by box-plots in [Figure 8(b)](#fig8){ref-type="fig"} (box-plot on the right). ### 3.2.2. Real Data Sets {#sec3.2.2} Spike sorting performances measured on real data by visual inspection and by quantitative assessment (i.e., intracluster variance, ICV) were proven to be in good agreement with results on simulated signals. For each combination of FE and clustering algorithms and O-sort, [Figure 9(a)](#fig9){ref-type="fig"} shows the scatter plot of the percentage of unclassified spikes and the ratio between the number of identified clusters containing evidently similar and spike-like waveforms and the true number of clusters (mean values across *N* = 10 signals). DWT and PCA features, combined with both FCM and DBC, provided the best performances, since they lie in the upper left area of the graph. GEO features of real spikes combined to either DBC or FCM performed worse than PCA and DWT. Finally, cluster methods combined with FSDE features could not correctly identify almost any cluster, mostly because real spikes and noise were merged inside poorly compact clusters. The figure also shows that DBC left a higher number of unclassified spikes than FCM. Indeed, if the shape of the spikes changes after the definition of the LUT, due to bursting events or an increase of the noise \[[@B48]\], the incoming spikes could be not labeled at all. Also O-sort provided a high percentage of unclassified spikes, due to the tendency to overclustering (i.e., creating many small clusters). As shown in [Figure 9(b)](#fig9){ref-type="fig"}, *K*-means and DBC clusterers coupled to PCA and DWT yielded a statistically lower ICV than FCM, but higher than O-sort, which created the most compact clusters, comparable to the application of the best method for each signal. FSDE and GEO coupled to any clusterers resulted in the worse (i.e., highest) ICV. 3.3. Spike Sorting Runtimes {#sec3.3} --------------------------- An evaluation aimed at comparing Matlab execution time relative to the feature extraction and clustering steps was performed. Parameter values set for this evaluation were the ones which allowed the best performance for each method (see [Section 3.2](#sec3.2){ref-type="sec"}). [Table 5](#tab5){ref-type="table"} reports the number of operations (i.e., additions, multiplications, and if-operations) required to process on-line a single spike in the implemented Matlab code and the resulting runtimes in Matlab. Extraction of features from each input spike takes on average a comparable time for PCA, FSDE, and GEO methods (i.e., 5-6 *μ*s). In spite of the higher number of additions and multiplication compared to the other methods, PCA is time effective thanks to the absence of comparison operations. The DWT-based method requires a time one order of magnitude higher (i.e., 64 *μ*s) because of the significantly highest number of operations. FCM classification of one spike is fourfold slower than DBC classification (i.e., 27 *μ*s versus 7 *μ*s). This difference is ascribable to the fact that DBC requires only the consultation of a look-up-table. O-sort algorithm takes a higher time (on average 210 *μ*s per spike), since its implementation involves not only arithmetic operations but also large amounts of memory and control logic. 3.4. On-Line Feasibility Evaluation {#sec3.4} ----------------------------------- The experimental test performed with the setup described in [Section 2.5](#sec2.5){ref-type="sec"} showed that an on-line processing of MEA signals is possible using the implemented framework. To allow that, the Matlab execution time spent to process a data block containing 64-channel raw data should be lower than the length of serially transmitted data blocks from the A/D device. The most time effective coding turned out to be a sequential processing over the channels. Given as input data buffer, each channel is filtered before being scanned for spikes; then each detected spike is given as input to the feature extractor and classifier. [Figure 10](#fig10){ref-type="fig"} reports the runtimes for the processing of 64 channels in a worst-case input scenario (i.e., 1 spike every 4 ms at every channel). The runtime is related to input data block length (i.e., the time available for any processing before the buffer update) and is expressed as its percentage (e.g., a runtime percentage equal to 60% for a 1-second block means that there is a margin of 400 ms for further operations). Runtimes include the time for raw data reading, filtering, and saving (gray dashed line), plus spike detection (black dashed line), plus feature extraction and clustering (lines colored according to the combinations of the algorithms). In the example, reported spike detection runtime relates to a negative threshold-based spike detection method. Generally, "AdaBandFlt" takes more time (i.e., up to 4-fold in the case of maximum data block length), due to its higher complexity, suggesting that further optimization is needed. In the simulated scenario of high spiking frequency simultaneously in all the channels, all the feature extraction methods (apart from DWT), coupled to both clustering methods, can process data before buffer overwriting, if the data block length is higher than \~300 ms. When using shorter data blocks, the operation of raw data reading from the acquisition device (black dashed line in [Figure 10](#fig10){ref-type="fig"}) takes the highest percentage of the available time, so that the only feasible processing operation is spike detection (gray dashed line). When using data blocks as long as 1 second, the runtime of all the methods (except DWT) is lower than 600 ms (i.e., 60% of data block length), indicating that further operations are feasible. As expected from evaluations on single-spike runtimes, the DWT-based feature extraction, although highly accurate in many cases, is hard to be performed on-line in the actual implementation. The tests showed that, only in the case of data block length between \~700 and \~1500 ms, a classification with DWT coupled to DBC was possible in the worst-case scenario but leaving a poor margin (lower than 10%) for additional operations. Finally, the O-sort method could not run on-line in the proposed implementation, as expected from results on single-spike runtimes reported in [Table 5](#tab5){ref-type="table"}. 4. Discussion and Conclusions {#sec4} ============================= 4.1. Usefulness of the Present Work {#sec4.1} ----------------------------------- The present work addressed two important issues in the field of spike sorting of neuronal signals collected by means of MEAs, which are the very limited availability of data-centric and on-line spike sorting tools. Our aim was to provide a framework for an easy comparison of different spike sorting algorithms on the same data which would be suitable for off-line and on-line data analysis. Rather than proposing a new sophisticated algorithm, we exploited the modularity of existing spike sorting processes, that is, the presence of several steps and different techniques that can be mixed and matched to adjust the process to the data set. Therefore, the implemented toolbox integrates different spike sorting blocks (i.e., four feature extraction methods, three feature clustering methods, and one template matching clustering), which have been selected from the literature. Thus, it provides the possibility to choose the algorithm that optimally performs on a specific channel data most suitably, in contrast to commonly used tools which apply one predefined method to all electrodes. The pool of algorithms integrated in the framework presents features of automaticity and simplicity that are important requirements in spike sorting and facilitate on-line implementations \[[@B31]\]. Besides the modular software tool, the work has presented an extensive evaluation of the different combinations of feature extraction and clustering methods integrated in the framework. In order to help the users in the selection of their best methods for data processing and to guide the evaluation of multiple algorithms on other data sets, the algorithms were tested both on simulated data sets (as most commonly done to assess the performance of spike sorting methods \[[@B13], [@B20]\]) and on real multiunit recordings from neuronal dissociated cultures, that is, in more realistic conditions \[[@B24], [@B46]\]. As expected, the selection of channel-centric methods (i.e., the best method for each channel) provided the statistically highest accuracies level. These results support the usefulness of different spike sorting options in the same framework. Furthermore, the realized framework lends itself to be exploited also to explore possible associations between signal features (e.g., waveforms similarity and SNR value) and the accuracy of spike sorting methods. 4.2. Advantages of the Matlab Framework for Spike Sorting {#sec4.2} --------------------------------------------------------- Matlab has been primary chosen because the MEA users often share algorithms written in Matlab language, which can facilitate the utilization and extension of the framework. Despite existing open-source alternatives (e.g., Python), Matlab is still a very common framework for neurophysiologists and research institutions working with MEAs \[[@B49], [@B50]\]. The spike sorting codes implemented in this work could thus be easily tested and incorporated into existing Matlab-based frameworks offered by the community and dedicated to different kinds of analysis of MEA data (e.g., "Spycode" \[[@B45]\], "Find" \[[@B51]\], "Wave_clus" \[[@B20]\], "NeuroQuest" \[[@B31]\], "DrCell" \[[@B34]\], "SigMate" \[[@B50]\], "QSpike tools" \[[@B52]\], and "Manta" \[[@B32]\]). For running in the off-line mode, the GUI integrates the Neuroshare API library, which is a community-supported vendor-neutral library. Therefore, it would be possible to import into Matlab neural data files acquired by different platforms from the one we used (Multi Channel Systems GmbH) with minor modifications. For running in the on-line mode, actually two types of acquisition boards from Multi Channel Systems were tested (i.e., USB-ME64/128 and MEA2100). However, the framework is expected to be compatible with other acquisition boards provided that an*ad hoc* communication interface is built in Matlab. This would be easily performed exploiting the Data Acquisition toolbox of Matlab, which allows acquiring data from a variety of DAQ hardware \[[@B32]\]. 4.3. Discussion of On-Line Spike Sorting Operations {#sec4.3} --------------------------------------------------- Taking into consideration that Matlab is slower in performing some operations compared to low level languages \[[@B52]\] and thus it is not the best choice for real-time analysis (e.g., closed-loop MEA experiments), we explored its potentiality to run MEA data classification (including also the preceding filtering and spike detection) during on-line open-loop recordings. We demonstrated that the realized implementation (60 channels) was able to respect the time constraints running most of the evaluated spike sorting algorithms in a worst-case realistic scenario, where simultaneously a high firing activity is collected by all the channels. The runtime evaluation took into account also times for accessing data sent by the A/D device that we use in our lab and showed that a good temporal margin is achievable with data blocks length longer than 500 ms. However, considerations about the length of data blocks reported in [Section 3](#sec3){ref-type="sec"} can be generalized to any A/D board which sends dataflow to the acquisition computer by means of data blocks. Clearly, a much lower data block length would reduce the delay between acquisition and processing, but this issue is not relevant when the aim is to speed up analysis of open-loop recordings. Moreover, a too short data buffer would increase the probability of splitting spike waveforms or burst events into two different time epochs. 4.4. Future Work {#sec4.4} ---------------- Further code optimization activity will be performed to reduce the time for spike sorting blocks currently requiring too long computational runtime. To this aim, a more careful optimization of codes, an implementation of MEX-files of all the spike processing steps, and the resorting to the Parallel Computing toolbox \[[@B49]\] will reveal the easiest options to provide improvements over the numbers reported here. If this is not suitable, we plan to exploit the Parallel Computing toolbox to access the graphical processing unit (GPU) of the computer \[[@B34], [@B53]\]. Moreover, an issue of the current implementation is that on-line parallel data visualization is not possible, due to speed limits of Matlab graphics \[[@B32]\]. Possible solutions could be to test a range of efficient coding strategies to speed up data plotting \[[@B32]\] or to integrate a plotting plugin written in a faster GUI programming language such as C\# \[[@B31], [@B35]\]. Besides improvements centered on shortening algorithm runtimes, a future activity will be focused on the automatization of the training phase. To this aim, an evaluation of the time needed to train the data (e.g., acquire a data fragment until a minimum number of spikes have been detected) and the most reasonable frequency of the training during acquisition will be performed. A repetition of the training would be preferred especially for long-lasting experiments since (i) only neurons which fire during the learning phase can be classified and (ii) the physical relations between neurons and electrodes may change due to cell growth (nonstationarity). A possible solution could be to perform a training step whenever a metric referring to the goodness of clustering detects the fact that data have changed their features to a considerable extent, as suggested by a recent work \[[@B30]\]. Furthermore, a possible enrichment of the tool would be to provide the possibility to combine heterogeneous features extracted from the spikes (e.g., DWT, PCA, and GEO) allowing taking advantage of the strengths of each feature extraction method to achieve better performances \[[@B4]\]. This implementation would likely require a longer time for the initial training phase and, during on-line mode, the computation of several features within the time constraints. Thus, it will be incorporated after the afore-mentioned optimization steps. Finally, the integration of methods to resolve the issue of spike overlapping \[[@B19], [@B54]\] may be considered. Supplementary Material {#supplementary-material-sec} ====================== ###### The supplementary material includes the description of the Matlab toolbox for neuronal spike sorting presented and validated in the work (see Results 1.1). In particular, guidelines on how to use the 'test data\' GUI as well as the 'real data\' off-line and on-line GUIs are here provided together with graphical examples. The authors thank Alembic facility and Dr. Andrea Menegon for providing them with lab facilities. They are grateful for Dr. Ludovico Minati for his help in setting up the real-time communication between Matlab and the MEA acquisition device. Disclosure ========== Emilia Biffi is currently working at the Bioengineering Laboratory, Scientific Institute IRCCS Eugenio Medea, Bosisio Parini, Italy. Competing Interests =================== The authors declare that they have no competing interests. ![Scheme of the spike sorting processing algorithms incorporated in this work. For each electrode the raw signal is preprocessed before the subsequent spike detection by a threshold-based algorithm (i.e., AdaBandFlt \[[@B24]\]). The feature extraction can be performed with four different methods (i.e., principal component analysis (PCA), Discrete Wavelet Transform (DWT), geometric features (GEO), and First and Second Derivative Extrema (FSDE)) followed by a dimensionality reduction step that retains the relevant features. Three clustering algorithms are implemented to automatically cluster spike features (i.e., *K*-means, fuzzy-*C*-means (FCM), and density-based clustering (DBC)). As an alternative, a template matching algorithm (O-sort) groups the spikes as soon as they are detected.](CIN2016-8416237.001){#fig1} ![Features of the simulated data set. Spike waveforms were selected from a database of averaged spike waveforms obtained from spontaneous activity recorded in hippocampal and cortical*in vitro* neuronal networks by MEA. For each group of waveforms depicted in the figure, signals with three different SNR (4, 3, and 2) were simulated, obtaining a total of 36 signals. Each set of waveforms is associated with an ID name (e.g., S2(A)), the ordinal position in the data set (e.g., \#1-2-3), and the mean Bray-Curtis similarity (BCS) between the waveforms (e.g., \[0.52\]).](CIN2016-8416237.002){#fig2} {#fig3} {#fig4} {#fig5} {ref-type="fig"}) in each feature space (PCA: Principal Components Analysis, DWT: Discrete Wavelet Transform, GEO: geometric features, and FSDE: First and Second Derivative Extrema), colored according to the real labels. (b) Cluster validity (CV) values obtained after the application of the 4 feature extraction methods to the 36 simulated extracellular signals. (c) Cluster validity dependence on different noise levels (median of CV values for each SNR group). (d) Box-plots (median and IQR with whiskers delimited by the maximum and minimum nonoutliers values) of CV values on all the simulated signals (*N* = 36). The asterisks above each method indicate statistically highest CV values of the current method compared to the method(s) coded by the asterisks\' color (Wilcoxon\'s matched pair test with *p* \< 0.01).](CIN2016-8416237.006){#fig6} {ref-type="fig"}) in each FE space and results of *K*-means clustering of the features. Spikes are colored according to the real label. Superimposed circles are the clusters found by *K*-means. (b) Classification accuracy values for the 36 simulated signals.](CIN2016-8416237.007){#fig7} {#fig8} {#fig9} {#fig10} ###### Overview of spike sorting algorithms. ------------------------------------------------------------------------------------------------------------------------------------------------------------------ Reference Feature extraction Clustering ---------------------------------------------- ------------------------------------ ------------------------------------------------------------------------------ Letelier and Weber \[[@B14]\] Wavelet Fuzzy-*C*-means Harris et al. \[[@B15]\]^a^ PCA^c^ Expectation maximization Zouridakis and Tam \[[@B16]\] Waveforms Fuzzy-*C*-means Hulata et al. \[[@B17]\] Wavelet *K*-means Egert et al. \[[@B18]\]^a^ PCA Manual cluster cutting Shoham et al. \[[@B19]\] PCA Expectation maximization Quiroga et al. \[[@B20]\]^a^ Wavelet packet coefficients Superparamagnetic clustering Rutishauser et al. \[[@B11]\]^a^ --- Template matching Cho et al. \[[@B21]\] LDA^d^ Fuzzy-*C*-means Adamos et al. \[[@B22]\] PCA Expectation maximization Awais and Andrew \[[@B23]\] Zero crossing *K*-means Biffi et al. \[[@B24]\] PCA Hierarchical clustering Takekawa et al. \[[@B25]\] Wavelet Bayes Gibson et al. \[[@B13]\] Discrete derivative Fuzzy-*C*-means Cheng et al. \[[@B26], [@B27]\] PCA Density-based clustering Liu et al. \[[@B28]\] PCA Valley-seeking Lai et al. \[[@B29]\] Wavelet Gray relation analysis Bestel et al. \[[@B4]\] PCA, wavelet, geometrical features Expectation maximization Yuan et al. \[[@B12]\] Wavelet *K*-means, template matching Oliynyk et al. \[[@B30]\]^a^ PCA Fuzzy-*C*-means^f^ Kwon et al. \[[@B31]\]^a^ DWT^e^, PCA, peak-to-peak Expectation maximization, *K*-means, fuzzy-*C*-means, manual cluster cutting Englitz et al. \[[@B32]\]^a^ Geometrical features 1D clustering^f^ Paraskevopoulou et al. \[[@B33]\] FSDE *K*-means Nick et al. \[[@B34]\]^a^ PCA, DWT, geometrical features Expectation maximization MCRack (Multi Channel Systems GmbH)^b^ --- Manual amplitude window^f^ Spike2 (Cambridge Electronic Design Ltd.)^b^ PCA Template matching^f^ \ Manual cluster-cutting, *K*-means, Gaussian mixture models Off-Line Sorter (Plexon Inc.)^b^ PCA Expectation maximization, *K*-means, valley-seeking ------------------------------------------------------------------------------------------------------------------------------------------------------------------ Overview of the literature about spike sorting algorithms, including published papers about methods, custom toolboxes, and commercial software. ^a^Custom toolbox available to the community. ^b^Software coupled to a commercial acquisition platform. ^c^Principal component analysis. ^d^Linear Discriminant Analysis. ^e^Discrete Wavelet Transform. ^f^On-line mode. Keywords used for literature search were "spike sorting", "spike detection", "spike classification", "*in vitro*", "real time/on-line", "Microelectrode arrays", and "toolbox" (PubMed and Google Scholar). ###### Properties of the selected feature extraction methods. -----------------------------------------------------------------------------------------------------------------------------------------------------   Domain Percentage of publications\ Need of training before on-line FE (with respect to [Table 1](#tab1){ref-type="table"}) -------------------------------------------- ------------ ------------------------------------------------------ ------------------------------------ Principal Components Analysis (PCA) Time 36% Yes First and Second Derivative Extrema (FSDE) Time 3% No Geometric features (GEO) Time 13% Yes Discrete Wavelet Transform (DWT) Time/scale 26% Yes Other methods --- 22% --- ----------------------------------------------------------------------------------------------------------------------------------------------------- The "domain" column refers to the analysis domain in which each method works, for example, time domain or time/scale domain. The "percentage of publications (with respect to [Table 1](#tab1){ref-type="table"})" is the ratio between the number of publications dealing with a certain FE method and the total number of analyzed publications (reported in [Table 1](#tab1){ref-type="table"}). To esteem this parameter, works dealing with *N* methods were counted as *N* different works in the denominator. Different works of the same authors using the same FE method were counted as 1. The "need of training before on-line FE" column indicates whether a preliminary training phase on a first set of acquired spikes is needed in order to run the method in on-line mode. ###### Properties of the implemented and evaluated clustering algorithms. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------   Input Percentage of publications\ Automaticity Parametric Need of training before on-line clustering (with respect to [Table 1](#tab1){ref-type="table"}) ----------------------- ---------------- ------------------------------------------------------ -------------- ------------ -------------------------------------------- *K*-means Spike features 30% Yes^a^ Yes --- Fuzzy-*C*-means (FCM) Spike features 25% Yes^a^ Yes Yes Density-based (DBC) Spike features 4% Yes No Yes O-sort Spikes 11% Yes Yes No Other methods^b^ --- 30% --- --- --- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ The "percentage of publications (with respect to [Table 1](#tab1){ref-type="table"})" is the ratio between the number of publications dealing with a certain clustering method and the total number of analyzed publications (reported in [Table 1](#tab1){ref-type="table"}). "Automaticity" refers to the possibility not to define a number of clusters a priori. "Parametric" refers to the need to set one or more threshold values for parameters involved in the algorithm. The "need of training before on-line clustering" column refers to the necessity of an off-line phase to train the algorithms on data before an on-line classification can be performed. ^a^The optimization of a proper index during training phase is needed. ^b^Including nonautomatic methods (e.g., manual methods) and methods not suitable for on-line mode. ###### Classification accuracy of all the tested methods. *K*-means FCM DBC O-sort ------ ------------------- ------------------- ------------------- --------------- --------------- -------------------- -------------- PCA 97.16^*∗*^ (4.09) 98.46^*∗*^ (9.64) 97.11^*∗*^ (6.45) 94.76 (12.15) 91.04 (27.45) 97.66^*∗*^ (5.92) --- DWT 96.42^*∗*^ (4.78) 77.18 (31.08) 95.29^*∗*^ (8.81) 93.60 (12.68) 76.13 (24.47) 93.90 (13.39) --- FSDE 68.42 (29.64) 69.83 (14.07) 63.66 (33.26) 67.01 (25.22) 36.97 (50.62) 56.71 (21.24) --- GEO 79.57 (17.82) 72.25 (28.26) 77.43 (28.63) 82.35 (19.77) 84.99 (18.29) 88.10^*∗*^ (20.04) ---   --- --- --- --- --- --- 94.37 (4.75) Spike sorting classification accuracy, CA (%), on the simulated data sets for all the possible combinations of FE (rows) and clustering algorithms (columns) and for O-sort. CA is presented as median and (IQR) over the different signals (*N* = 36). For each FE method (i.e., first 4 rows of the table) the asterisk points out the clustering method with the statistically highest CA performance (*p* \< 0.01, the Friedman + Wilcoxon test). Accuracies lower than 70% (e.g., FSDE method) were not considered acceptable. ###### Computational effort and runtime to process a single spike. Method Number of additions Number of multiplications Number of if-operations Time (*μ*s) ----------- --------------------- --------------------------- ------------------------- ------------- PCA^*∗*a^ 3*n* ^e^ 3*n* 0 5.8 ± 3.3 DWT^*∗*b^ 4*n* 6*n* 0 64 ± 33 FSDE^*∗*^ 2*n* − 3 0 2(*n* − 1) 5.2 ± 4.3 GEO^*∗*c^ 2*n* 1 6*n* − 2 6.5 ± 3.9 FCM 5*C* ^f^ 6*C* *C* 27 ± 0.6 DBC 0 0 6*N* ^g^ 7.3 ± 0.2 O-sort^d^ 6*Cn* + 2*n* 6*Cn* + 2*n* + 1 8 210 ± 200 Computational requirements to classify a single spike (3 ms waveform sampled at 25 kHz). Columns from 2 to 4 indicate the number of operations for each spike included into the implemented Matlab code, showing their dependence on algorithms parameters. The last column reports the resulting execution times (averaged over 100 repetitions and reported as mean ± standard deviation). Times are for Matlab running on a quad-core 3.3 GHz CPUs desktop computer with 4 GB RAM and Windows 7 64-bit. Asterisks in the first column indicate methods for which an implementation in C language with MEX-files was performed. ^a^Projection onto 3 principal components. ^b^3-level wavelet decomposition. ^c^Extraction of 7 geometric features from the spike waveform. ^d^Worse case in which all the cycles/iterations involved in O-sort method are entered. ^e^ *n*: number of samples per spike (*n* = 75 in the numerical time example). ^f^ *C*: number of clusters or templates (*C* = 4 in the numerical time example). ^g^ *N*: resolution of the look-up-table for density-based clustering method (*N* = 32 in the numerical time example). [^1]: Academic Editor: Gaetano D. Gargiulo

Introduction ============ Disasters are of significant encounters for human societies worldwide and can have devastating social, medical, and public health consequences [@ref1]. Iran is highly vulnerable to different types of natural disasters such as earthquake, flood, and droughts. On average, 2000 to 3000 people are killed annually due to such incidences [@ref2]. The Rudbar-Manjil earthquake (1990), Bam earthquake (2003), Golestan flash floods (2000-2005), Lorestan earthquake (2006), Azarbaijan earthquake (2012), and Bushehr earthquake (2013) were of the most destructive disasters during last decades. People's life after natural events and disasters has not been studied sufficiently [@ref3] ^,^ [@ref4]. Most of the studies in this field have only focused on short periods of disasters and long term processes were completely missed out. However, getting back to normal life after a disaster can be as significant as life in pre or during disaster stages. Although many experimental and theoretical works have been conducted on consequences of disasters, an in-depth understanding of rehabilitation and what exactly forms this process is still evolving [@ref5]. It is also suggested that the concept of rehabilitation cannot be decontextualized and investigated out of complicated economic, political, and social systems in which it happens[@ref6] ^,^ [@ref7]. Hence, to evaluate the challenges of rehabilitation process, it is necessary to have a multi-dimensional look rather than a one-dimensional perspective. Nevertheless, many of the studies conducted in this field have not taken a comprehensive look at this process, and have usually focused only on one aspect such as psychological interventions after disasters, stress disorders after disasters [@ref8] ^-^ [@ref11], physical injuries [@ref12] ^,^ [@ref13], the role of social capital in reconstruction [@ref14], and the role of community participation in physical reconstruction [@ref15]. Consequently, the majority of rehabilitation plans are not comprehensive and there are limitations in adopted social approaches. They usually do not appropriately consider the variable range of long-term physical and social needs of societies after disaster occurrence [@ref16]. Furthermore, research indicates that some people experience a disaster posttraumatic growth in terms of greater appreciation of life, deeper meaningfulness, closer relationships, and deeper religious faith; however, these studies do not describe the process of moving from disaster period to posttraumatic growth [@ref17]. On the other hand, there is currently no comprehensive study on the concerns of life after disasters in Iran. Previous works have mainly focused on post-disaster mental health interventions [@ref18], disaster relief problems and management inconsistencies [@ref19] ^-^ [@ref22], and healthcare challenges in disasters [@ref23] ^-^ [@ref25]. However, most of these studies have not comprehensively evaluated the challenges of getting back to normal life after disasters, and more importantly, they have not used a qualitative approach to better understand the experiences and perceptions of disaster survivors. Thus, there is still a significant gap in this field. Clearly, this stage is a complicated, systematic, and interactive process which needs a more complete understanding of the long-term recovery experiences of people involved. The current study, therefore, investigated the challenges of returning to the normal life after earthquake in rural areas of Iran. Materials and Methods ===================== **Study design:**A qualitative approach using content analysis (Graneheim approach) was considered appropriate for our study. In this method, required data was gathered directly from participants, without any pervious hypothesis. Produced knowledge was based on unique viewpoints of participants. Codes and categories were derived through an inductive process and then conceptually ordered properties and dimensions were developed [@ref26]. **Setting and participants**: The study was conducted in rural areas of Azarbaijan where there were two, somehow, massive quakes in 2012. According to their firsthand experience or expertise in earthquake and also their willingness to participate in the study and to gather in-depth and rich experiences, the participants were chosen using a purposive sampling method with maximum diversity. Sampling continued until data saturation was achieved. Of participants, 20 people experienced the earthquake firsthand and 7 people, including 3 social workers, 2 psychologists, and 2 local health workers, had scientific expertise in disaster recovery. Also, 2 focused group discussions were held to complete the data collection and understand the meaning of primary results. 11 local survivors attended the first group discussion session, 10 disaster recovery experts participated in the second session. **Data Collection**: Semi-structured interviews, focused group discussions and field notes were used to collect data. Data collection was done by the main researcher who lived in earthquake-stricken area for about 18 months (August 2013 -- January 2014). Before the interviews, by introducing himself and expressing the aim of the study, the researcher obtained the informed written and oral consent of the participants. The interviews lasted for 30 to 60 minutes and were tape recorded and transcribed verbatim; each interview began with a broad question about participants' experiences of events they had observed. Probing was conducted according to reflection of each participant on life after disaster, such as perception of reconstruction process, services and their needs, facilitators and barriers of providing services, role of social networks, and organization of disaster recovery. The focus groups were run by the researcher and local assistants and attempts were made to facilitate discussions and consider all relevant sides of the issue. **Data Analysis**: Qualitative content analysis was used to analyze the data. Systematic stages were followed and simultaneous analysis was undertaken: first, recorded interviews were transcribed verbatim. Then, before coding, the transcribed text was read several times for familiarization. Codes and categories were extracted by an inductive process via open coding (through line by line reading of the text and devoting relevant codes to it). Then, categories emerged by constant comparison. Peer check and constant comparison were used to reach a consensus in coding. In fact, data analysis was performed simultaneously and continually with the data collection. After completion of coding and assuring accuracy of coding, concepts were identified. **Trustworthiness:** we used the strategies recommended by Lincoln and Guba for trustworthiness. According to this recommendation, four criteria of creditability, dependency, conformability, and transferability are necessary for trustworthiness. To increase data creditability, the researcher engaged with data and the environment for 18 months while constantly making observations and compiling field notes. Dependency of data was assessed by peer check strategies. Peer check was performed on a monthly basis so that the research team had a thorough discussion about emerged data. The background and personal interest of researcher on the subject and maintaining documents of study were used for conformability of data. The context of the interviews, codes, and the extracted categories were reviewed by the research team and other professional colleagues in the field of qualitative research. Using sampling with a maximum variation, the researchers were able to collect a wide variety of different comments, observations, and interpretations. **Ethical Considerations**: Informed consent was obtained through explaining the aim and process of the study orally and in written. The study was approved by the ethical committee of University of Social Welfare and Rehabilitation Science (USWR) in Tehran, Iran. Information was kept confidential and Participants had the right to withdraw at each stage of the study. . Results ======= Of 27 participants of the study, including survivors of the earthquake-stricken areas and the qualified people with relevant scientific expertise, the mean age was 41 years. 21 of participants were men, three were single, and the literacy level ranged from illiterate to postgraduate qualifications. Quotations mentioned in this article are all translated from Persian into English by the researcher (FA). Accuracy of translations was independently checked by another author (PSF). Findings indicated that "social uncertainty and confusion" was the main concern of people after disaster, a matter which was directly and indirectly hidden in most of study codes. People reported that they lived with significant uncertainty about the future and that they experienced a significant decline in perceived ability to recover and in motivation. One of participants explained the situation as follows: "...*before the earthquake we were all busy with our everyday life. We knew what to do, where to go, what we were looking for ...but after this long period from the earthquake, we just spend days and nights and think about all of the our troubles. We feel that we do not know what we are doing, where we are going, and finally what will happen to us. We are tied up to these troubles and unfinished jobs. Our Houses are ruined and agriculture and husbandry are destroyed. Finally, we will be forced to leave this place (Participant No. 8; resident, man, 4o years old)."* This study also identified six other main concepts related to participants' main concerns: social vulnerability, lack of comprehensive rehabilitation plan, ignorance of local social capital, waste of assets, and psychological problems. These main categories and their subcategories are explained below: **1. Social Vulnerability** Findings showed that disasters are associated with everyday life of people. The severity of damages and their consequences depend on various factors such as socioeconomic vulnerability of people. Experiences and perceptions of the participants indicated that natural events are not the only obstacles impeding their return to normal life and different direct and indirect factors affect the speed and quality of this process. Three emerged subcategories about social vulnerabilities were as follows: 1) inefficiency of local non-governmentalsocial institutions, 2) lack of public awareness, and 3) social problems before the earthquake. **1.1 Inefficiency of local non-governmental social institutions** According to the results, inefficiency of non-governmental social institutions had left the government alone with the responsibility of restoration, and caused various problems in returning to normal life. Inefficiency of institutions such as councils and associations caused challenges for people in providing assistance and communicating with governmental institutions. From the viewpoint of participants, intermediary institutions could facilitate the rescue works, make them more goal-oriented, and prevent from lots of financial waste. It is noteworthy that competency of intermediary institutions is also an important factor. If people do not trust them or if they are incompetent, then their effectiveness is lost. One of the participants explained the situation as: "...in some areas we don't have any intermediary institutions to play an active role in post disaster recovery and in other areas there are problems in relation between people and these institutions. From my viewpoint, since the rescue work should be mainly done by these people and their systems, their competency is important. Their incompetency was one of the problems in some areas, which made rescue work and goods distribution more difficult ..." (Participant No. 18; resident, woman, 49 years old). **1.2 Lack of public awareness** Lack of public awareness was another subcategory which caused problems for people at disaster time and even during the process of getting back to normal life. Most of the participants talked about lack of in advance knowledge and skills to respond to and recover from a disaster; they also believed that the current public education was not effective enough and made them unable to play any active role. A participant noted the issue as: "...it is noteworthy that shortly before the earthquake, they held earthquake training here and we participated. Of course, it was a fake training and was not held seriously. Most of the people even were not aware of it and those who participated were only observers. They burned a car's tire and then extinguished it while laughing and joking ... and called it the readiness training and exercise for natural disasters"(Participant No. 23; community health worker, man, 4o years old). **1.3 Social problems before the earthquake** Pre-earthquake social problems significantly increase vulnerability and hinder people's return to normal life in a timely fashion. Most of participants believed that factors such as poverty, inequality, and unemployment were factors which increased their social vulnerability and created problems in returning back to normal life in an efficient manner: "...the problems and challenges that we had before the earthquake are still with us, making our lives more difficult. The condition for people who did not have a good financial situation before the earthquake is worse than others and they still couldn't get back to their normal life."(Participant No. 4; resident, man, 65 years old). Participants believed that people who had physical, social and, personal vulnerabilities suffered the most losses: "...people who did not have a good financial situation, or had shouldered more losses, or have lost one of their close relatives are currently having more difficulties and should be protected..." (Participant No. 21; rescuer, man, 38 years old). **2. Lack of comprehensive rehabilitation plan** Findings showed that there were times of progress and yet times of frustration in the process of helping people to get back to normal life. In some areas, the management was very effective, yet in others it was very poor having no leadership or even being totally ignored. Participants believed that it is necessary to determine responsibility of the teams and consider appropriate measures before, during, and after disasters. Regarding this issue, findings revealed four subcategories of: 1) lack of comprehensive data and information, 2) negligence of vulnerable groups, 3) Concentration on reconstruction and overlooking of rehabilitation, and 4) improper distribution of resources. **2.1 lack of comprehensive information** One of the obstacles impeding the process of rehabilitation was the lack of comprehensive information about the demographics of the people and residents of each district. This lack of this information was significantly noted in views of a rescuer: "...unfortunately, since the responsible organizations lacked a detailed statistical system the rescue problems were intensified, and unfortunately some of the people who are residing here in the shelters are not the permanent residents of the village and only used to live here seasonally; we do not have any detailed statistical data about them. For example, a village had 80 households and we only had 45 tents, which caused a chaos among the people\..." (Participant No. 27; rescuer, man, 42 years old). **2.2 Negligence of vulnerable groups** Neglecting vulnerable groups was another subcategory extracted in the study. Findings showed that certain groups had not been considered in the distribution of goods, nor had their specific requirements been considered. Women, children, the elderly, and people with disabilities were among the groups that, according to the participants, were greatly neglected. This significantly led to intensification of social confusion and delayed rehabilitation. Rescuers had firsthand views of negligence of these subgroups. One rescuer explained the situation as: "...we were in a village to distribute women basic hygiene items. However, women felt embarrassed to receive them from us because we were men. Finally, at our return, many of them thanked us and stated that these items were more important for them than rice, food and etc..." (Participant No. 24; recovery service provider, man, 26 years). One of the female participants said: "...most of the times, the relief goods were distributed in a way that we couldn't get them. Because men were stronger and could get more goods, but women, the elderly and the others who did not have such power couldn't get them. Even sometimes we were embarrassed..." (Participant No. 5; resident, woman, 42 years). **2.3 Improper distributions of resources** Another extracted subcategory was the improper distribution of resources. In many cases, because of improper management much of resources were wasted thereby leading to more difficulties among the survivors: "Sometimes they provide us some food, but there is no regulation in distribution. Our relatives also bring us some food from the city. Sometimes benefactors also provide us something. But the most outstanding defect is lack of an organized program. Packs of water or canned fishes are stacked in one place and wasted, while in some other places people are in a great need of these items..."(Participant No. 24; recovery service provider, man, 26 years). **2.4 Concentration on reconstruction and overlooking of rehabilitation** Another subcategory extracted was mere concentration on reconstruction and overlooking of rehabilitation. People, authorities and recovery services focused only on reconstruction, with little attention to rehabilitation needs of the people. This led to intensification of social uncertainty and confusion in the earthquake-stricken areas, as explained by one of the participants: "...the government has done its best to reconstruct the houses and this has not been completed yet and most of the houses are incomplete, but unfortunately no attention has been primarily paid to agriculture, husbandry, employment, and mental, and social issues \... "(Participant No. 19; resident, man, 66 years). The people of earthquake-stricken areas also concentrated on reconstruction and did not attend to other aspects of recovery. In this regard one of the participants said: "...but here people are concentrated on the houses. If the houses are completed, people can get back to their normal life. We are only busy with reconstructing our homes and working. Therefore, we cannot do anything else. We urge the government only to take the reconstruction seriously. If they complete our houses, we will be relaxed..." **3. Incomplete reconstruction** Participants believed that getting back to normal life depends on the completion of reconstruction and therefore they were worried about the delay of this process. Most of the participants emphasized the importance and priority of physical reconstruction and believed that their return to normal life significantly depended on completion of physical reconstruction. Our analysis revealed two subcategories regarding incomplete reconstruction: 1) failure of timely reconstruction, and 2) Construction without considering the culture **3. 1 Failure of timely reconstruction** Failure of timely reconstruction due to different reasons was one of the main factors contributing to confusion in returning to normal life. Based on participants' viewpoint, one of the main reasons for this problem was the failure to control and supervise reconstruction and also break of contractors who had left the projects incomplete during the cold season so that people could not get back to their normal life: "...we are currently involved in reconstruction of our houses and there is nothing else and we cannot have proper food. The only thing I have to think about now is finishing my house to stay safe and sound against cold. If I get stuck with construction and don't finish it soon, I will not be able to start working on agriculture and husbandry next year" (Participant No. 15; resident, man, 50 years old). From the people\`s point of view, completion of reconstruction would be a necessary relief from their current situation and they believe that it would be a way to gradually forget the bitter memories of earthquake. However, a mass of rubble on the road side of many villages was a reminded of earthquake which left people with feelings of depression: "...every time I pass this road I see the rubble on the road side and the bitter memories of the quake come to my mind again..." (Participant No. 6; resident, man, 18 years). **3. 2 Reconstruction without considering the culture** The second subcategory, the inability to consider local cultural needs, led to problems in recovery. Negligence of special requirements in rural houses was one of the main aspects of reconstruction that led to dissatisfaction and uncertainty in people's lives. In other words, architecture and interior design of the reconstructed houses did not meet the needs of people in earthquake-stricken areas. They were just shelters and in many cases had no bedrooms, workshop, or barn. The reconstruction itself was very satisfying for people, but lack of attention to cultural aspect of reconstruction was one of the main concerns. A participant argued that they have especial problems with houses: "We need large rural houses. We usually have many guests and these sizes are insufficient. They have built us some civic houses. These houses might be suitable for people who live in cities, but are not enough for us. The houses have not enough bedrooms and no place for carpet weaving. Husbandry and agriculture have been also neglected. As you enter the village these houses look beautiful, but if you put yourself in my place, you will see these houses are like a cage\..." (Participant No. 14; resident, man, 34 years old). ****4.**** **Ignorance of local social capital** An important aspect in disaster recovery and return to normalcy is that survivors need to be active participants in the process. Self-efficacy is an important concept in psychological health, but when people are not included in their recovery; their sense of self-efficacy can be substantially undermined. Regarding the issue, following subcategories were extracted: 1) a top-down paternalistic approach, 2) undermining of trust, and 3) undermining of social networks and generation of self-centering. **4.1 A top-down paternalistic approach** Findings support the idea that even regarding the minor issues in recovery, people were not asked for their ideas and input, nor were they asked to participate in reconstruction and other aspects of rebuilding. Most of the plans have been performed following a top-down paternalistic approach. It caused people to feel dissatisfied and not to have the sense of ownership and belonging regardless of several governmental efforts that were made for them. In this regard, one of the participants said: "...only during the first days after the earthquake people sometimes helped each other, and then we just received relief goods and people played no role. We were all waiting for the Red Crescent and other institutions to come and help us..." (Participant No. 9; resident, man, 60 years old). Another participant explained the situation as bellow: "...the situation is not like before at all. They only built us some 60 m^2^ houses and now they want us to go and live in them. We have no choice but to tolerate this situation, but have no sense of ownership to these houses. These houses cost us a lot and we were forced to sell everything (to buy the houses)..." (Participant No. 11; resident, woman, 58 years old). **4.2 undermining of trust** Another extracted subcategory was undermining of trust. Before and after the disaster, a kind of distrust was built up between the earthquake-stricken people and the government, and also between benefactors and the government for various reasons. According the participants, these conditions paved the way for confusion and social uncertainty: "...if the government did nothing and just gave us the benefactors' reliefs, we had no problem and everything was good. The government seized all the money and relief goods of benefactors' and instead gave us loans. Instead of receiving an aid from the government, I am in debt of a 25 million Tomans loan to the government" (Participant No. 15; resident, man, 50 years old). In addition, there was another kind of distrust between benefactors and the government that led to special problems. In fact, the donations were distributed by benefactors and these distributions were not targeted and organized. A participant described the situation as: "...We must accept the fact that many people do not trust government agencies. Why not to trust is another matter. In fact, the government agencies should figure out why people intend to directly aid earthquake-stricken areas. It is necessary to study why they do not deliver their aids to the governmental agencies..." (Participant No. 21; psychologist, man, 38 years old). One of the community disaster recovery staff described the situation as: *"* *Unfortunately, benefactors* *treat quite emotionally so that they directly deliver their aids to the villages or those who pretend to be needy villagers. In Iran, in the early stage of disaster recovery management, aids of donors and responsible organizations are like a flood, they suddenly come and go. If it is not effectively managed, it can lead to waste of aids"(Participant No. 20; resident, man, 27 years old).* **4.3 undermining social networks and generation of self-centering** The third subcategory of social capital neglect was the undermining of social networks and creation of self-centering. According to some participants, this was a significant obstacle in their ability to gain social effectiveness and a successful return to normal life: "...during the first days after the earthquake people helped each other, especially they rescued many victims by helping each other. Unfortunately, as time passed, especially during the reconstruction of houses and receiving of relief goods, people just thought about themselves and everyone tried to get more goods..."(Participant No. 7; resident, man, 41 years old). **5. Waste of Assets** Waste of people\`s critical assets was another factor that impeded recovery. Some people were forced to sell their cattle at low prices after the earthquake because they did not have adequate barns and shelters necessary to take care of their livestock. These assets later were very important for getting back to normal life. Yet, most people were not able to make the loss up after some months because they could not afford the higher prices for cattle: One of the participants explained the situation as: "...since all infrastructures were destroyed and there were no facilities, people were forced to sell their cattle. May people came here and bought our assets. We waited a month for a barn and when they didn't build it, we were forced to sell our cattle. Dealers took advantage of our situation..." (Participant No. 2; resident, man, 28 years old). Loss of properties and assets caused people to think the earthquake has seized everything they had, so they lose their hope to get back to the normal life. Shortly after earthquake, conditions had changed in a way that people thought they cannot revive their lost assets. For example one participant said: *"...of course we want to get back to our home, but how? We do not have anything. I had 10 cows and was forced to sell them in those days, and now having those cattle is like a dream, which cannot be achieved. These small houses and high prices do not let us buy even two cows ..." (Participant No. 14; resident, man, 34 years old)* *.* **6. Psychological Problems** Most of participants were challenging with significant psychological problems as one of the consequences of earthquake. Even after several months, different groups of people suffered from various mental problems such as depression, anxiety, and fear. In many cases, fear and anxiety led to sleep disorder among children. One of the participants described it as: "The earthquake mentally affected us, especially the children. Now children are afraid of every little sound. They are very sensitive. We cannot have a good sleep, every moment we are afraid of another earthquake. There were numerous aftershocks; there was one some nights ago ..." (Participant No. 18; resident, woman, 49 years). 18 about months after the earthquake, many participants talked of lack of desire to get back to the normal life, feeling sad and ...: "...we are not like before. We have lost our motivation and most of the people are not happy and are mostly sad. If you look at our situation from distance, you cannot recognize the sheer reality of our lives. It takes a life time to get back to previous point" (Participant No. 11; resident, woman, 58 years old). Discussion ========== The current study, which was the first of its kind using a qualitative and comprehensive approach, evaluated the challenges of people in return to a normal life after a significant earthquake in rural areas of Iran. By analyzing viewpoints of earthquake victims and disaster experts dealing with rehabilitation interventions, the study provided a much needed perspective of the most important problems of post-disaster recovery process. The key element explored in this study was "social uncertainty and confusion". Furthermore, the findings identified six inhibiting factors or barriers to successful recovery, namely, lack of a comprehensive rehabilitation plan, social vulnerability, ignorance of local social capital, incomplete reconstruction, waste of assets, and psychological problems. These issues greatly prolonged the rehabilitation process thereby causing people to suffer from social uncertainty and confusion. One of the main obstacles on the way back to normal life after earthquake was the lack of a comprehensive rehabilitation plan. Participants indicated that the recovery process started with an emotionally-laden management approach with a lot of energy, but gradually the momentum declined until it reached a level of no further activity thus leaving people bewildered, confused, and forgotten. Another important issue was the presence of unresolved social problems before the earthquake which led to intensification of social uncertainty and confusion in the aftermath of the disaster. This finding was consistent with those of other similar studies [@ref27] ^-^ [@ref30]. Social capital can be conceptualized as social connections between people occupying homophiles networks, across heterogeneous networks and organizations, and with those of higher status and power [@ref31]. Accordingly, our findings also indicated that failure to fully engage community in recovery strategies and efforts was another reasons for social uncertainty and confusion. As relief agencies did not encourage participation, people felt a reduced sense of belonging, greater dissatisfaction, and more dependency on governmental assistance. Importance of public participation in post-disaster rehabilitation has been also emphasized in other studies [@ref14] ^,^ [@ref15]. Lack of occupational infrastructures was another issue in return to normal life. Absence of a systematic program in this regard and lack of income led to financial distress among people. Financial distress was also occurred as a result of forced sale of assets, properties, and livestock at low prices immediately after the earthquake and the subsequent failure to recover them due to high pricing in the recovery period. Pyles, (2007) have similarly emphasized issues surrounding restoration of jobs and businesses after disasters [@ref32]. Findings indicated that rebuilding of homes and infrastructures were of main preconditions of reconstruction in other spheres of rehabilitation. Delays in process of rebuilding led to delays in other social, psychological, and economic aspects. Speed and quality of reconstruction were specifically identified as the core problems. These findings were similar to other disaster related studies [@ref33] ^-^ [@ref35]. This was very interesting that all governmental and lay people focused only on reconstruction and did not care about rehabilitation. Gender is an important variable among social factors; disaster research has indicated that women generally are more vulnerable to disasters [@ref36]. The study also found problems and limitations in providing services, specifically for women and other vulnerable groups. Due to lack of awareness or insensitivity to needs of these groups, they were greatly underserved which further compounded their social uncertainty and confusion. Zahran and colleagues have concluded that domestic violence would increase after disasters [@ref37]. Such an issue was not mentioned by the present study\`s participants. Authors think that such issues need to be studied in future and with relevant methodologies. Finally, as typically seen in all disasters, people experienced various kinds of psychological and emotional disturbances. Depression, anxiety, and fear of future earthquakes were the most common problems indicated [@ref17]; however, in some groups, especially children, sleep disturbance was a significant symptom. Conclusion ========== According to the results of the current study, an effective rehabilitation plan needs a comprehensive management system that first of all should be approved and understood by people, intermediary and governmental institutions, and other beneficiaries. In addition, an effective rehabilitation plan also needs a common understanding, considering the complicated and multi-dimensional, dynamic and long term nature of getting back to the normal life with maximum rate of people's participation. Rehabilitation is prior to reconstruction, but still most of the people and disaster management affiliated organizations in Iran believe that the rehabilitation is a set of governmental interventions aiming merely at reconstruction. According to the results of the current study, this approach has to be reformed and considered in the further policies of rehabilitation as a social and developmental process. Rehabilitation should be considered as a comprehensive process to support affected communities with their maximum representation and aiming people to achieve the highest degree of independency and sufficiency. Hence, policy-makers are recommended to change their viewpoints about rehabilitation from a linear and outcome-oriented approach to a continuous, prolonged, and comprehensive process. It is also necessary for policy makers to consider psychosocial interventions issues raised in the current study as a main part of disaster recovery plans, and especially they must change their lens and see psychosocial interventions as a main part of the disaster recovery programs not as a luxury and unnecessary intervention. This study is one of the few studies on recovery and rehabilitation process that has employed the social approach. The biggest strength of the present study was that the main researcher, though facing many challenges, lived in the affected area and observed closely the process of community recovery for almost 18 months. However, data were collected from a limited sample of individuals using purposeful sampling, as dictated by the employed qualitative method. Therefore, the findings cannot be generalized to other locations. The present study marks a starting point for clarifying and describing the process of resuming normal life after earthquake. Further study on this process is needed. Studies including the perspectives of more affected people, policy-makers, and local authorities could yield increased understanding of this process. There is a need for further investigation of the process of life after disaster using Grounded Theory approach to develop strategies for improving disaster recovery systems. The authors wish to thank all participants for their support and involvement in this study, including affected people, local health workers, and all study participants. PhD in Social Work Ph.D. in Counseling Psychology from Texas A&M University, 1992. Currently Associate Professor of Psychology & Counseling at the University of Central Arkansas and training director of the Mental Health Counseling masters program.

1. Introduction {#sec1} =============== Chronic recurrent multifocal osteomyelitis (CRMO) is a rare nonpyogenic inflammatory bone condition with unclear aetiology. It was first described by Giedion et al. in 1972 as a subacute chronic symmetrical osteomyelitis \[[@B1]\]. CRMO mainly affects the metaphyses of long bones, along with the pelvis, shoulder girdle, and less commonly the spine \[[@B2]\]. CRMO represents 2%--5% of all osteomyelitis cases, mainly affecting young girls with female-to-male ratio of 5 : 1. A 5-year follow-up study revealed the mean age of onset to be 10, with a range of 4--14 years \[[@B3]\]. The natural history of the disease is unpredictable with varying severities and time courses, oscillating between acute exacerbations and spontaneous remission \[[@B1], [@B4]\]. Other organs are often involved, namely the skin and gastrointestinal tract. Cutaneous manifestations include synovitis, acne, pustulosis, hyperkeratosis, and osteitis (SAPHO) syndrome \[[@B5], [@B6]\]. In a letter in 1990, Kahn first reported the association of CRMO with inflammatory bowel disease (IBD) reporting that, of the 30 CRMO cases, 4 cases also had Crohn\'s disease and 1 case had ulcerative colitis \[[@B7]\]. CRMO is also associated with other rare autoimmune disease processes. Acute febrile neutrophilic dermatosis (Sweet\'s syndrome) is characterised by painful erythematous plaques over the face, trunk, and limbs and has been linked with CRMO. Majeed\'s syndrome is a very rare genetically linked disease, which presents with congenital dyserythropoietic anaemia, Sweet\'s syndrome, and CRMO seen in a small case series involving a Middle Eastern family \[[@B8]\]. We present a 15-year-old girl with an acute scoliosis secondary to CRMO. There appear to be only 2 previously documented similar cases. Moreover, this case has additional features of pustular psoriasis and anaemia, which have required further specialist management. The case is illustrated with a coronal MRI slice, which demonstrates the foci of inflammation in the spine, with a very pronounced thoracic scoliosis ([Figure 1](#fig1){ref-type="fig"}). 2. Case Presentation {#sec2} ==================== TM is a 15-year-old girl who presented with a 6-month history of steadily deteriorating intermittent back pain mainly affecting her cervicothoracic spine. Her parents had also noticed an asymmetry in her shoulders with abnormal posture. TM was born at term and developed within normal limits. Menarche was at the age 13. She was fully vaccinated and had no previous childhood illnesses. There was no history of trauma. TM also complained of a recurrent itchy rash affecting her palms and soles over the past 2 years. TM has a positive family history of psoriasis. On examination, TM stood tall for her age at 167.5 cm. Her right scapula was slightly raised. There was a 10-degree midthoracic rib hump. Upper and lower limb neurology was entirely normal. Respiratory, cardiovascular, and abdominal examinations were normal. Standing spinal X-rays showed a 36-degree T5--10 right atypical thoracic scoliosis with a 23-degree T10-L3 and T1--T4 compensatory curve. There were no obvious bony deformities. At this stage, it was planned to follow up TM after a spine MRI and to refer to physios for truncal support exercises. Blood tests showed a mild normocytic anaemia, and HLA B27 was negative. Spine MRI showed abnormal signal on the STIR sequences in C7, T5, and T8 vertebrae and the right sacral ala. At this point, differential diagnoses were Langerhan\'s cells histiocytosis, insufficiency fractures, chronic multifocal osteomyelitis, and a leukaemic bone malignancy. The case was referred for a paediatric oncology opinion and to the Bone Tumour Unit in Birmingham for discussion at the MDT. Both recommended a CT-guided bone biopsy. TM was admitted for biopsy of her right greater trochanter and skin lesions. The bone biopsy showed evidence of fibrosis and chronic inflammatory changes with no specific diagnosis. There was no evidence of Langerhan\'s cell histiocytosis or suppurative infection. A punch biopsy of the skin lesions showed pustular psoriasis. All blood cultures were negative. Under the guidance of the Birmingham Spinal Centre, TM\'s symptoms were managed conservatively with simple analgesia and truncal physiotherapy. Her symptoms appeared to be improving, and she was clinically examined in the outpatient department at 3 monthly intervals. Whole spine radiographs were taken at 6 monthly intervals to monitor severity of scoliosis, and the MRI was repeated annually to monitor the disease process. Follow-up whole spine MRIs at 6 months showed some new oedematous lesions around L2 spinous process and sternum. There was also a marked increase in oedema in the lesions previously seen around the right sacroiliac (SI) joint, C7, T5, T7, and T8. These radiological findings correlated with an acute deterioration whilst the patient was on holiday with no clear precipitant. The patient required a wheelchair at one stage due to the pain in her legs. The interval MRIs at 18 months showed similar signal intensity and size for many of the lesions previously noted in C7, T1, T2, T4-5, and T7-8. There was some improvement in the L2 lesion. New lesions were identified in the left clavicle and multiple ribs bilaterally. Fortunately, symptoms had improved considerably, now tolerating moderate exercise, and the limp had entirely resolved. TM described some mild pain in the lower back and sternum after long periods of sitting. Her psoriasis improved with the local PUVA treatment managed by a dermatologist. There appeared to be a strong correlation between the improvement in the musculoskeletal and dermatological symptoms. 3. Discussion {#sec3} ============= It is important to consider CRMO as a differential diagnosis when presenting with an acute atypical scoliosis. MRI typically shows multifocal increases in signal on the STIR sequences. High signals can present concurrently or sequentially; therefore, interval scanning is essential to monitor progression of the disease \[[@B9]\]. Malignancy, pyogenic infections, and atypical presentations of juvenile arthritis need to be ruled out, correlating the history, clinical findings and biopsy results. The use of allied professionals in specialist centres is key in formulating this diagnosis. CRMO\'s association with other immune-modulated diseases such as inflammatory bowel disease and psoriasis indicates a likely autoimmune process \[[@B10]\]. Unlike bacterial osteomyelitis, CRMO is not treated with antimicrobials, and anti-inflammatories are shown to improve symptoms. Azithromycin has been used for its anti-inflammatory properties. Indomethacin is an inhibitor of ossification, and a study has shown it to be effective in the treatment of CRMO, with clinical and radiological improvement \[[@B11]\]. The use of systemic steroids during flares and low-dose courses to prevent relapse has been shown to be effective \[[@B8]\]. The natural history of CRMO is poorly understood and may have multiple organ features that require specialist management. This paper fortifies the small amount of the currently available literature on a difficult multifaceted disease to manage. This 24-month clinical and radiological followup of scoliosis caused by CRMO appears to be the longest in the literature. The case highlights the relapsing and remitting nature of CRMO with new lesions developing and older lesions burning out. Clinically, our patient is improving with less pain, increased exercise tolerance, and no systemic upset. Radiologically, the scoliosis appears to be stable. We advise close radiological surveillance and symptomatic control as the mainstay of management. The authors declare that there is no conflict of interests. Consent has been obtained from the patient. {#fig1} [^1]: Academic Editors: C. Aldana-Valenzuela, A. W. Kamps, and M. Moschovi

Introduction {#sec1} ============ Superhydrophobicity, as inspired by the "Lotus effect" in nature, features a water contact angle (WCA) above 150°. The water repellency of a surface is governed by a combination of chemical composition and geometrical surface structure. Superhydrophobic thin films have attracted great attention in a wide range of applications including antifouling paints,^[@ref1]^ waterproof clothes,^[@ref2]^ corrosion inhibition,^[@ref3]^ water and oil separation,^[@ref4]^ and so forth. Recently, superhydrophobic thin films incorporating wide band gap semiconductor oxide nanoparticles such as TiO~2~ (band gap: 3.2 eV), ZnO (3.4 eV), and CeO~2~ (3.1 eV) have attracted increased interest because of their ultraviolet (UV) absorption and self-cleaning properties.^[@ref5]−[@ref12]^ These superhydrophobic thin films were converted to hydrophilic films because of the photocatalytic properties of these nanoparticles. In other words, these thin films changed their wetting properties from the Cassie state^[@ref13]^ to the Wenzel state^[@ref14]^ via the exposure to UV radiation.^[@ref5]−[@ref12]^ Nishimoto et al.^[@ref8]^ reported the hydrophilic transformation of a superhydrophobic surface based on TiO~2~ porous layers passivated with a self-assembled monolayer of octadodecylphosphonic acid because of UV radiation in just 30 min. Similarly, polydimethylsiloxane (PDMS)-coated TiO~2~ nanoparticles turned hydrophilic in 6 h because of the UV exposure as reported by Zhang et al.^[@ref15]^ In the recent years, several works have been performed on UV-protective superhydrophobic thin films on substrates such as metals, cottons, and fabrics.^[@ref6],[@ref16]−[@ref24]^ Most works used TiO~2~ nanoparticles as a UV absorber and organic compounds to lower the surface energy to fabricate UV-protective superhydrophobic surfaces.^[@ref6],[@ref16]−[@ref21]^ Recently, Gao et al. used fluoroalkylsilane (FAS)-modified TiO~2~ nanoparticles to fabricate superhydrophobic thin films that converted to hydrophilic \[contact angle (CA) = 2°\] following an exposure to UV radiation for 24 h.^[@ref16]^ This conversion from superhydrophobicity to hydrophilicity was an expected phenomenon as the fabricated thin film contained only a monolayer of FAS on TiO~2~. However, contradictory reports over FAS-coated superhydrophobic surfaces state that it is possible to maintain the superhydrophobic properties upon prolonged exposure to UV radiation.^[@ref24]^ Wang et al. have demonstrated stable superhydrophobic properties on steel surfaces even after 50 h of exposure to UV radiation.^[@ref24]^ The UV durability was ascribed to the long chain of FAS-17 on the surface, which could provide a large number of C--F bonds (i.e., −(CF~2~)~7~--CF~3~ chain). This C--F bond, with a bond energy of 485 kJ/mol, cannot be broken by the UV light (314--419 kJ/mol).^[@ref24]^ In another work by Gao et al., epoxy resin-covered TiO~2~ nanoparticles were used to fabricate superhydrophobic filter papers. These authors showed that the WCA on the surface of the filter paper slightly decreased to 151° ± 0.5° from 153° ± 1° after more than 300 min of UV exposure.^[@ref17]^ However, there was no information on the stability of the superhydrophobic properties after a long-term exposure as presented in our work. In another work by Wang et al., the superhydrophobic surfaces prepared by TiO~2~/polybenzoxazine nanoassemblies became hydrophilic upon exposure to UV radiation of only 300 s.^[@ref19]^ Qing et al. used TiO~2~ nanoparticles covered by PDMS layers to make superhydrophobic thin films which demonstrated corrosion resistance. These films, however, failed to maintain their superhydrophobic properties after only 20 min of exposure to UV radiation, and their CA reduced drastically from 162.3° to 75.6° converting to hydrophilic surfaces. Surprisingly, a postheat treatment of these surfaces at 120 °C for 30 min restored the superhydrophobicity.^[@ref18]^ In another work, the reversibly switchable wettability of the FAS--TiO~2~/poly(vinylidene difluoride) composite surface was investigated through the serial alternating of UV exposure and heat treatment.^[@ref6]^ They observed that the CA decreased sharply as the UV irradiation time increased, converting a superhydrophobic surface with a CA of 160° to superhydrophilic with a CA of 0° just after 60 min. Interestingly, it was observed that it took only 40 min to revert to the original CA when these surfaces were heated at a temperature of 180 °C.^[@ref6]^ Several smart ideas have been utilized to obtain UV-durable superhydrophobic surfaces.^[@ref5],[@ref7],[@ref10],[@ref25]−[@ref27]^ Ding et al.^[@ref26]^ reported a UV-durable superhydrophobic fluorinated polysiloxane/TiO~2~ nanocomposite coating because of the fact that Si--O and C--F bonds in the fluorinated polymer matrix were stable against photocatalytic reactions of TiO~2~ nanoparticles. It is noteworthy to mention that fluorinated silanes are quite expensive, environmentally unfriendly, and not appropriate for practical applications. Similarly, in a study of Gao et al.,^[@ref5]^ a transparent and UV-durable superhydrophobic surface was fabricated by passivating the arrays of the SiO~2~-coated ZnO nanorods with perfluorodecyltriethoxysilane. The physical barrier of the SiO~2~ layers was responsible for the UV resistance property. However, both of their fabrication processes are complex, and a further modification of the hydrophilic SiO~2~ shell with low-surface-energy materials is always inevitable. In the present study, a facile, low-cost, and effective sol--gel spin-coating method has been developed to fabricate UV-durable superhydrophobic thin films. With this method, initially, TiO~2~ nanoparticles are coated with cobalt stearate (CoSA) and this core--shell (TiO~2~\@CoSA) material is further incorporated in the ethanolic polymethylhydrosiloxane (PMHS) solution. The sol--gel spin-coated thin films prepared with this mixture are found to be superhydrophobic with high UV durability properties. A model has been proposed to explain the UV durability mechanism of this material. Results and Discussion {#sec2} ====================== [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}a-I,II shows the X-ray diffraction (XRD) patterns of pure stearic acid (SA) powder and CoSA films, respectively, on aluminum substrates in the 2θ scan range of 3°--30°. [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}a-I shows the characteristic peaks at the 2θ values of 6.76°, 20.35°, 21.61°, and 24.29°, assigned to the SA (\#JCPDS stearic acid: 00-009-0618). The XRD pattern of CoSA ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}a-II) shows a series of equidistant diffraction peaks situated at 3.61°, 5.42°, 7.22°, 9.02°, 10.81°, 12.62°, 14.44°, and 16.24°. The origin of these equidistant peaks in the diffraction pattern is attributable to the layered structure of CoSA. The average distance between the two peaks has been calculated to be 1.80° ± 0.01°. Hence, an initial peak would be expected to be situated at 2θ equal to 1.81°. Luo et al.^[@ref28]^ have reported the appearance of a peak in a position similar to that in the Langmuir--Blodgett thin film of CoSA. We have not observed this first peak in our study as we have performed the experiment at the initial angle of 3°. Lowering the diffraction angle may cause permanent damage to the detector because of the possibility of direct bombardment of the high intensity of X-ray to the detector. It is noteworthy to mention that we have observed 12 distinct characteristic diffraction peaks of CoSA as compared to only the 4 peaks reported by Luo et al.^[@ref28]^ {#fig1} The inset of [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}a shows the ultraviolet--visible (UV--vis) absorption spectrum of the drop-coated CoSA films on quartz substrates in the light wavelength range of 200--400 nm. A strong absorption peak is observed at around 210 nm. This result is comparable to the UV--vis spectrum of copper stearate.^[@ref29]^ It is noteworthy to mention that the UV--vis spectrum of CoSA is not available in the literature. [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}b-I,II reveals the attenuated total reflectance Fourier transform infrared (ATR--FTIR) spectra of pure SA powder and the drop-coated CoSA film, respectively. In the high-frequency region of the two spectra, the adsorption peaks at 2847 and 2914 cm^--1^ are attributed to the asymmetric and symmetric stretching vibrations of −CH~2~ groups, respectively. One tiny peak assigned to the asymmetric stretching of −CH~3~ is observed at 2939 cm^--1^.^[@ref3],[@ref30]^ In the mid-frequency region, a distinct sharp peak of the carboxyl (−COOH) group of SA is observed at 1701 cm^--1^, as shown in [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}b-I. It is noteworthy to mention that this peak, associated with the carboxyl group, disappears in the CoSA thin films ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}b-II). Interestingly, and on the contrary, [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}b-II shows two distinct peaks observed at 1410 and 1550 cm^--1^ corresponding to the carboxylate (−COOCo) symmetric and asymmetric stretching vibrations, respectively.^[@ref31]^ Another peak in between these two peaks appearing at 1446 cm^--1^ is ascribed to −CH~2~ bending.^[@ref31],[@ref32]^ In the low-frequency range, a sharp absorption peak at 720 cm^--1^ is attributed to the in-plane rocking vibrations of −(CH~2~)~*n*~ long carbon chains in the CoSA molecules.^[@ref31]^ In brief, XRD, ATR--FTIR, and UV--vis results confirm that CoSA has been successfully synthesized. Anatase phase of the TiO~2~ nanoparticle is well-known for the use of photodegradation^[@ref11],[@ref33],[@ref34]^ of organic compounds because of the wide photonic band gap of 3.2 eV. Therefore, we have used anatase TiO~2~ nanoparticles of size 100 nm to study photodegradation as well as to counter the photodegradation using our synthesized CoSA in the thin films. It is extremely significant to develop UV-durable superhydrophobic thin films especially for painting the exterior surfaces such as building walls which are normally exposed to the UV light directly from the sun.^[@ref35]^ A comparative UV degradation study has been performed on two superhydrophobic thin films prepared by a sol--gel process using (i) TiO~2~ nanoparticles and PMHS (coded as PMHS/TiO~2~) and (ii) CoSA-modified TiO~2~ nanoparticles and PMHS (coded as PMHS/TiO~2~\@CoSA) in our UV chamber developed in-house. The red plot of [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}a shows the variation of the WCA on the PMHS/TiO~2~ thin films with respect to the UV irradiation time. The initial superhydrophobic PMHS/TiO~2~ thin films exhibit a WCA of ∼152° (shown in the inset) and a contact angle hysteresis (CAH) of ∼6°. After 1 h of exposure to UV radiation, the WCA of this film decreased to ∼119°, losing the rolling-off properties. With the UV irradiation time increasing to 2 h, the WCA of the PMHS/TiO~2~ thin films was found to be only ∼44°. It is important to note that the static and dynamic CAs may be influenced by the size, shape, and movement of the water drop.^[@ref36],[@ref37]^ {#fig2} The superhydrophobic PMHS/TiO~2~ thin films were completely converted to superhydrophilic films after UV irradiation of 4 h because of the photocatalytic effect of anatase TiO~2~ nanoparticles.^[@ref33]^ The low-surface-energy PMHS was decomposed by highly reactive superoxide and hydroxyl radicals generated by TiO~2~ nanoparticles under UV irradiation.^[@ref33],[@ref38]^ Similar photodegradation and polymerization processes induced by ZnO nanoparticles have been studied recently by Schmitt.^[@ref39],[@ref40]^ He further reported the formation of silanes and siloxanes by Raman spectroscopy.^[@ref41]^ A model has been presented in [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}b to explain the degradation mechanism. By contrast, [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}a (black square) shows that the WCA remains nearly constant after more than 300 h of UV irradiation for the PMHS/TiO~2~\@CoSA superhydrophobic thin films. Although the data have been presented for 300 h, we have observed that these films retain their superhydrophobic properties even after more than 1000 h of UV irradiation. The initial WCA of 160° (CAH = 2°) reduced only slightly to 156° (CAH = 5°) after 300 h of UV irradiation. The intensity of the UV light in our in-house UV chamber is much stronger than that in the direct sunlight, indicating that the PMHS/TiO~2~\@CoSA superhydrophobic thin films have great potential in outdoor applications. This highly UV-durable superhydrophobic thin film with water roll-off properties can be regarded as a potential candidate for creating an excellent self-cleaning material as well. It is to be noted that the photo-Kolbe reaction leads to radicals which might result in a protective layer around the embedded TiO~2~ particles similar to that reported by Schmitt on ZnO nanoparticles.^[@ref39]^ {#fig3} [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}b shows the ATR--FTIR spectra of the chemical groups of the pure PMHS organosilane liquid ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}b-I) and the superhydrophobic PMHS/TiO~2~\@CoSA thin films ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}b-II). In the high-frequency region, the spectrum of PMHS displays a single peak at 2969 cm^--1^, which is assigned to the asymmetric stretching mode of −CH~3~ groups in the PMHS molecule. The spectrum of the superhydrophobic PMHS/TiO~2~\@CoSA thin film (the inset top graph of [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}b) shows two peaks at 2914 and 2853 cm^--1^ attributed to the symmetric and asymmetric stretching modes of −CH~2~ groups, respectively, arising from CoSA, beside the absorption peak of −CH~3~. In the mid-frequency region, a tiny peak is observed at 1450 cm^--1^ because of the −COOCo bonds. It can be seen that the intensity of this peak is very low because the number of −COOCo bonds in the composite films is calculated to be hardly 2% of that of Si--O--Si or Si--CH~3~ bonds. It is a known fact that the signals of FTIR peaks are proportional to the number of corresponding chemical bonds. In the mid-frequency region, the peak at 2162 cm^--1^ arising from the Si--H groups of PMHS shown in [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}b-I is no longer observed in the spectrum of the superhydrophobic PMHS/TiO~2~\@CoSA surfaces in [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}b-II. This indicates that the cross-linked polymer matrix has been formed by PMHS organosilanes through condensation reactions of Si--H groups during the sol--gel process.^[@ref42]^ The two peaks at 1270 and 764 cm^--1^ correspond to Si--CH~3~ groups. In addition, the double peaks present at 1112 and 1026 cm^--1^ are ascribed to the Si--O--Si groups of PMHS organosilanes after the formation of oligomers.^[@ref42]^ Furthermore, the absorption peak of Ti--O groups is displayed at quite a low frequency, in the region near 500 cm^--1^, indicating the existence of TiO~2~ nanoparticles in our UV-durable superhydrophobic thin films. [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}c depicts the XRD spectra of the as-received aluminum substrates ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}c-I) and PMHS/TiO~2~ ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}c-II) and PMHS/TiO~2~\@CoSA superhydrophobic thin films ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}c-III) in the scan range of 22°--58°. The two distinct peaks observed at 38.4° and 44.7° on the XRD pattern of the as-received aluminum substrates are attributed to the characteristic peaks of Al(111) and Al(200). In addition, the XRD patterns of PMHS/TiO~2~ and PMHS/TiO~2~\@CoSA superhydrophobic thin films coated with aluminum substrates show the characteristic peaks of aluminum as well as the characteristic peaks of TiO~2~(101), TiO~2~(200), TiO~2~(105), and TiO~2~(211) at 25.3°, 48.1°, 53.9°, and 55.1°, respectively, confirming the incorporation of TiO~2~ nanoparticles in the thin films. [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}d shows the energy-dispersive spectrometry (EDS) spectra of different elements in the superhydrophobic thin films PMHS/TiO~2~\@CoSA ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}d-I) and PMHS/TiO~2~ ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}d-II). The spectra show the common elements in each of the two thin films, which include C, Ti, O, and Si. The element Co, in particular, can only be detected in the EDS spectra of PMHS/TiO~2~\@CoSA, confirming the existence of CoSA in this superhydrophobic nanocomposite film. The atomic composition of the coating is found to be C/O/Si/Ti/Co = 25.7:54.1:7.9:12.1:0.1, as revealed from the EDS data. [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}a shows the morphology of the superhydrophobic PMHS/TiO~2~\@CoSA thin film on the aluminum substrate, as revealed by the scanning electron microscopy (SEM) measurements. PMHS molecules, CoSA sheets (as found from XRD), and TiO~2~ nanoparticles (100 nm) have agglomerated together producing microsized random particles during the sol--gel and spin-coating processes. The roughness of the thin film is found to be ∼3.3 μm with a WCA as high as 160°. The hydrophilic TiO~2~ nanoparticles covered with a low-surface-energy CoSA sheet as well as PMHS molecules form binary structures, providing superhydrophobic as well as UV-protective thin films. [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}b shows a schematic model of the thin film composed of a core--shell structure of TiO~2~\@CoSA covered with a PMHS network (PMHS/TiO~2~\@CoSA). The CoSA shell on the TiO~2~ nanoparticles stops the photocatalytic reaction of TiO~2~ with PMHS by suppressing the movements of the electron--hole pairs generated from TiO~2~ because of UV absorption. In the literature, SiO~2~ and Al~2~O~3~ shells have been used around TiO~2~ to stop such reactions to fabricate UV-protective superhydrophobic thin films.^[@ref7],[@ref27],[@ref43]^ However, it should be pointed out that the adoption of the hydrophilic SiO~2~ shell around the hydrophilic TiO~2~ nanoparticles would increase the risk of producing hydrophilic surfaces rather than the superhydrophobic ones. In our case, the naturally hydrophobic CoSA shell covering the cores of the TiO~2~ nanoparticles not only improved the hydrophobic properties of the nanocomposite surfaces but also inhibited the photoinduced degradation by TiO~2~ nanoparticles, providing a dual protection against UV degradation as well as water adhesion. As a result, the hydrophobic PMHS matrix will not be decomposed by the photoinduced degradation reactions of TiO~2~. Therefore, the superhydrophobic nanocomposite thin films fabricated with a TiO~2~\@CoSA core--shell structure embedded in the hydrophobic PMHS matrix sustain UV irradiation while exhibiting and maintaining their superhydrophobic properties. In the literature, few works have been reported on the simultaneous UV degradation study of the nanoparticles dispersed in the liquid of methylene blue (MB) or rhodamine B (RhB) as well as UV degradation in the thin films.^[@ref20],[@ref21]^ For instance, in a work by Xu et al.,^[@ref20]^ an organically modified silica (ormosil) aerogel with a high surface area and high porosity was loaded with TiO~2~ nanocrystals to synthesize TiO~2~--SiO~2~ nanocomposites. Both ormosil and TiO~2~--SiO~2~ nanocomposites show superhydrophobic properties when coated on fabric surfaces. The authors further studied the photocatalytic performance of the TiO~2~--SiO~2~ composite particles and observed decolorization of RhB diluted in water.^[@ref20]^ It is interesting to note that these catalyst nanoparticles are more UV active when they are dispersed in the MB or RhB solutions than that in the form of a solid in the thin films. Although the decolorization of RhB was due to photocatalytic reactions of TiO~2~--SiO~2~ nanocomposites under UV radiation, the coatings on fabrics remained superhydrophobic after UV radiation.^[@ref20]^ Similarly, PDMS-coated SiO~2~ and N-TiO~2~ were used by Lee et al.^[@ref21]^ to prepare UV-resistive superhydrophobic coatings. They reported that nearly 50:50 mixture of PDMS-coated SiO~2~ and N-TiO~2~ provided a film with stable superhydrophobicity; however, it suffered the maximum photocatalytic activity when dispersed in the MB solutions, degrading upon exposure to UV. In the present case, we have shown that our coatings retain their superhydrophobic properties even after more than 300 h of UV exposure, which is much longer than those of other works reported previously.^[@ref6],[@ref16],[@ref18]−[@ref21],[@ref24]^ Further work is in progress to understand the UV degradation mechanism of MB or RhB by our newly developed core--shell CoSA\@TiO~2~ (TiO~2~ in the core) nanoparticles in the liquid as found in the literature.^[@ref20],[@ref21]^ Conclusions {#sec3} =========== CoSA thin films have been prepared and characterized by XRD, ATR--FTIR, and UV--vis spectroscopy. The synthesized CoSA has been utilized to prepare a UV-durable superhydrophobic nanocomposite thin film by embedding TiO~2~ nanoparticles in a hydrophobic PMHS matrix on AA6061 aluminum substrates. This PMHS/TiO~2~\@CoSA superhydrophobic thin film exhibited a nearly constant WCA of 160° under continuous UV irradiation for 1 month, whereas the superhydrophobic PMHS/TiO~2~ thin films, without CoSA shells on TiO~2~ nanoparticles, were drastically converted to the superhydrophilic state after UV irradiation of only 4 h. The synthesized CoSA not only increased the hydrophobicity in the surface of TiO~2~ nanoparticles but also confined the photocatalytic efficiency of TiO~2~. The UV durability mechanism of the superhydrophobic TiO~2~ thin films has been explained using a reasonable schematic model. These UV-durable superhydrophobic thin films formed by incorporating TiO~2~\@CoSA nanoparticles have potential to be utilized in a wide range of practical applications. Experimental Section {#sec4} ==================== Materials and Reagents {#sec4-1} ---------------------- An AA6061-T6 aluminum alloy with a chemical composition of Al 97.9 wt %, Mg 1.08 wt %, Si 0.63 wt %, Mn 0.52 wt %, Cu 0.32 wt %, Fe 0.17 wt %, Ti 0.02 wt %, and V 0.01 wt % has been used as substrates for coating. The aluminum (AA6061 alloy) substrates of dimension 1″ × 1″ were ultrasonically degreased in a soap solution and cleaned in ethanol and deionized water for 30 min each. The ultrasonically cleaned aluminum samples were dried for 24 h at 70 °C before further coating processes. The reagents including PMHS, cobalt nitrate, SA, and ammonium hydroxide were purchased form Sigma-Aldrich. TiO~2~ nanoparticles (average particle size: 100 nm) were purchased from MKnano. All of the reagents were of analytical grade. Deionized water with a resistivity of 18.2 MΩ·cm was used in all of the experiments. Preparation of PMHS/TiO~2~ Sol {#sec4-2} ------------------------------ In the first beaker, a mixture was prepared using 2 mL of PMHS and 40 mL of ethanol. Similarly, in the second beaker separately, 3 mL of ammonium hydroxide (28 wt %) was diluted in 10 mL of ethanol. This diluted ammonium hydroxide solution was added drop by drop into the first beaker to make the sol of PMHS. The prepared PMHS sol was aged for 48 h at room temperature while stirring continuously using a magnetic Teflon stirrer. In the third beaker, 1.5 g of anatase TiO~2~ nanoparticle was dispersed in 50 mL of ethanol, by ultrasonication for an hour. In the same beaker, 25 mL of PMHS aged sol was added drop by drop while constantly stirring to prepare the sol of TiO~2~/PMHS. This mixed solution was further aged for 5 days at room temperature while being constantly stirred prior to spin-coating on clean aluminum surfaces. Preparation of CoSA {#sec4-3} ------------------- CoSA was prepared using 17 mL of cobalt nitrate (0.01 M) and 34 mL of SA (0.01 M) ethanolic solution as well as 150 μL of ammonia (28 wt %) catalytic solution; the mixture was stirred overnight. Preparation of PMHS/TiO~2~\@CoSA Sol--Gel {#sec4-4} ----------------------------------------- Anatase TiO~2~ nanoparticles (1.5 g) were dispersed in 51 mL of ethanolic CoSA solution prepared as above. Further, 25 mL of PMHS aged sol as prepared above was added drop by drop while constantly stirring to prepare the sol of TiO~2~\@CoSA--PMHS. This mixture was also aged further similar to the PMHS/TiO~2~ sol for 5 days at room temperature prior to spin-coating on clean aluminum surfaces. Spin-Coating Process {#sec4-5} -------------------- The spin-coating process was carried out using a Single Wafer Spin Processor (Laurell Technologies WS-650SZ-6NPP/LITE) to prepare thin films from PMHS/TiO~2~ and PMHS/TiO~2~\@CoSA sol. UV Degradation Process {#sec4-6} ---------------------- A UV degradation chamber developed in-house, composed of two UV lights with wavelengths of 302 and 365 nm, was used to study the UV degradation process of the coated thin films. Characterization {#sec4-7} ---------------- The morphological and elemental analyses of the thin films prepared by the spin-coating process were performed using a scanning electron microscope (JEOL JSM-6480LV) equipped with an energy-dispersive X-ray spectroscopy instrument. The chemical compositional and structural analyses were carried out using XRD (D8 discover with a Cu Kα wavelength of 0.154 nm) as well as by ATR--FTIR (Agilent Technologies Cary 630 FTIR). UV--visible absorption properties were characterized using a UV--vis spectrophotometer (Agilent 8453 UV--visible Spectroscopy System). The surface wettability investigations of the coated samples were conducted by measuring static and dynamic CAs of a 10 μL water drop using a First Ten Angstrom CA goniometer. The dynamic WCA was measured by holding the water drop with a stationary needle in contact with the sample surface and moving the goniometer stage in one direction. The CAH, which is the difference between the advanced and receding CAs, was measured as described in previous reports.^[@ref3],[@ref36],[@ref44],[@ref45]^ The roughness measurements were carried out using the MicroXAM-100-HR 3D surface profilometer. The Supporting Information is available free of charge on the [ACS Publications website](http://pubs.acs.org) at DOI: [10.1021/acsomega.7b01579](http://pubs.acs.org/doi/abs/10.1021/acsomega.7b01579).ATR--FTIR spectra of the PMHS/TiO~2~\@CoSA film of as-received and after one-month UV irradiation; a schematic chemical model of CoSA; comparison of morphological features of the PMHS/TiO~2~\@CoSA and PMHS/TiO~2~ films revealed by SEM images; ATR--FTIR spectra of PMHS liquid, PMHS/TiO~2~, and PMHS/TiO~2~\@CoSA superhydrophobic thin films; and studies of UV degradation of the PMHS/TiO~2~ thin film by ATR--FTIR spectra and WCA ([PDF](http://pubs.acs.org/doi/suppl/10.1021/acsomega.7b01579/suppl_file/ao7b01579_si_001.pdf)) Supplementary Material ====================== ###### ao7b01579_si_001.pdf ^§^ Body & Exterior Dept., Pan Asia Technical Automotive Center Co., Ltd., 3999, Longdong Avenue, Shanghai, China 201201 (J.X.). The authors declare no competing financial interest. The authors are very grateful for the financial support provided by the Natural Science and Engineering Research Council of Canada (NSERC) and Aluminum Research Center (REGAL). The authors thank Saleema N., NRC-Saguenay, Canada for the critical reading of the manuscript.

{#fig1} {#fig2} {#fig3} Introduction ============ The promoter of the mouse mammary tumor virus (MMTV) provirus is a well-characterized example of transcriptional control by steroid hormones in which the chromatin organization plays an important role \[[@r43]\]. The provirus integrated in the host cell chromatin is virtually silent in the absence of hormones, but responds with rapid transcriptional activation to the addition of either glucocorticoids or progestins. The receptors for these hormones bind to a cluster of HREs in the MMTV promoter and facilitate the interaction of ubiquitous transcription factors including Nuclear Factor 1 (NF1) \[[@r21]\] and the octamer transcription factor, Oct1/OTF1 \[[@r11]\] with their target sites located between the HREs and the TATA box. This results in a synergistic activation of transcription by the hormone receptors and NF1 (for a review see \[[@r5]\]). How synergism between PR and NF1 occurs is a question that has attracted considerable attention, but the mechanism is not simply cooperative DNA binding of the various proteins to the MMTV promoter DNA \[[@r12]\]. Chromatin organization and factor binding ========================================= The LTR region of MMTV is organized into positioned nucleosomes \[[@r43]\] and hormone induction leads to the appearance of a DNase I-hypersensitive region over the promoter chromatin \[[@r63]\], suggesting an influence of hormone induction on the chromatin organization of the promoter ([Figure 1](#fig1){ref-type="fig"}). A role for nucleosome phasing in MMTV regulation has been postulated based on studies with breast cancer cell lines carrying a single copy of MMTV reporter stably integrated and on *in vitro* nucleosome assembly studies \[[@r51]\]. Though the exact positioning of nucleosome over the MMTV promoter has been debated \[[@r25]\], a dominant nucleosome phase in breast cancer cells precludes binding of NF1, but permits steroid hormone receptors (SHRs) to recognize one properly oriented HRE within the HRE cluster \[[@r51]\] ([Figure 1](#fig1){ref-type="fig"}). The different affinities of SHRs and NF1 for nucleosomally-organized target sites can be reproduced *in vitro* \[[@r22]; [@r40]\] and reflect the different ways in which the two proteins recognize their cognate DNA sequences \[[@r4]\]. SHRs only contact a narrow region of the HRE DNA double helix and can therefore bind if this section is exposed, while NF1 embraces the complete circumference of the helix and thus cannot interact with target sites within nucleosomes. When both SHRs and NF1 are added simultaneously to isolated MMTV mononucleosomes, the receptors bind to the accessible HREs, but NF1 is unable to recognize its target sites ([Figure 1](#fig1){ref-type="fig"}) \[[@r40]\], suggesting that additional components are required for simultaneous factor binding as detected in intact cells by genomic footprinting analysis following hormone treatment \[[@r51]\]. When introduced in *Saccharomyces cerevisiae* engineered to express GR or PR, the MMTV promoter is organized into positioned nucleosomes, is silent in the absence of hormone, and responds poorly to expression of NFI or to a NFI-VP16 fusion, but can be induced by hormone treatment \[[@r17]\]. Deletion of the HREs disrupts nucleosome positioning and makes the promoter responsive to NFI-VP16, suggesting that the HRE region would repress access to the NFI site by positioning a nucleosome \[[@r14]\]. The nucleosomal organization of the MMTV promoter *in vitro* is similar in yeast and animal cells, implying that the DNA sequence contains conformational or topological information which determines nucleosome positioning \[[@r41]\] and thus modulates the accessibility of transcription factors to *cis*-acting elements. Hormone induction was believed to result in a displacement of the nucleosome over the promoter, thus allowing free access of NF1 to its binding site and transcriptional activation \[[@r43]\]. However, genomic footprinting of the chromosomal MMTV promoter shows that hormone induction does not lead to displacement, but rather to a rearrangement of nucleosomes, which is necessary for simultaneous binding of receptors, NFI and OTF1 \[[@r51]\]. Thus, contrary to the situation in the test tube, the presence of a positioned nucleosome *in vivo* is not an obstacle, but rather a prerequisite for simultaneous factor binding and optimal induction of the MMTV promoter ([Figure 1](#fig1){ref-type="fig"}) \[[@r51]\]. This is strongly supported by the observation that nucleosome depletion in yeast, while enhancing the accessibility of the MMTV promoter for hormone receptors and NF1, eliminates the functional synergism between these two factors \[[@r16]\]. Moreover, in a cell-free system, synergistic binding of receptors and NF1 to the MMTV promoters depends on its previous assembly into minichromosomes with positioned nucleosomes and on preincubation in the presence of ATP \[[@r21]\]. FRET experiments and footprints =============================== Fluorescence photobleaching experiments in living cells containing a cluster of 200 MMTV promoters in tandem have provided direct evidence that the hormone-occupied GR undergoes rapid exchange between chromatin and the nucleoplasmic compartment with a half-life of less than a minute \[[@r33]\]. This rapid exchange has also been reported for the ERα \[[@r47]; [@r48]\]. In agreement with the difficulties in visualizing a footprint of GR on the MMTV promoter *in vivo*, this has led to the formulation of a "hit-and-run" model of receptor action \[[@r44]\]. However, following dexamethasone treatment of hepatic cells, such footprints have been reported on the GREs of the tyrosine aminotransferase enhancer \[[@r6]; [@r24]\]. Similarly, after progestin addition to breast cancer cells, DNase I and DMS footprints are observed on a single copy chromosomal MMTV promoter \[[@r51]\]. Moreover, following incubation of MMTV minichromosomes with purified PR in *Drosophila* embryo extracts, clear DNase I footprints are detected over the HREs \[[@r21]\]. These results demonstrate that a significant fraction of target promoters must be occupied by the receptors during the time of treatment with the footprinting agents, DMS or DNase I. How these findings can be reconciled with the very rapid exchange of GR-GFP observed *in vivo* remains to be elucidated. One possibility is that the majority of the GR binding events observed in the MMTV cluster *in vivo* are non-productive and do not lead to stable complexes containing receptors and NF1. Non-productive does not imply irrelevant. Possibly, many of these weak interactions are needed to form a stable complex of PR and NF1. Indeed, in the absence of NF1, it is exceedingly difficult to generate a receptor footprint on MMTV chromatin *in vitro*. This can only be achieved after addition of very high concentrations of receptor that push the equilibrium to the DNA bound state \[[@r21]\]. Nevertheless, even in the absence of NF1, receptor concentrations which do not generate a footprint can activate chromatin transcription to some extent, indicating that an interaction is taking place. That this short-lived interaction could correspond to the rapid exchange observed *in vivo* is suggested by the observation that after hormone induction only a small proportion of the MMTV clusters exhibit the phenomenon of chromatin decondensation, indicative of active transcription \[[@r39]\]. Hormone-dependent chromatin remodeling ====================================== Role of histone H1 ------------------ The nature of the hormone-induced nucleosomal change that permits simultaneous binding of transcription factors to the MMTV promoter remains obscure. Depletion of histone H1 from the MMTV promoter could be a possibility \[[@r7]\] and a role for histone H1 phosphorylation in modulating MMTV activation has been postulated \[[@r31]\]. However, the MMTV promoter is regulated in budding yeast, which lacks linker histones \[[@r17]\], and also in minichromosomes assembled in the absence of histone H1 \[[@r21]\]. Incorporation of H1 into MMTV minichromosomes increases nucleosome spacing and reduces access of general transcription factors to the promoter, thus inhibiting basal transcription \[[@r29]\]. However, transcription activation by PR and NF1 is enhanced in H1 containing minichromosomes \[[@r29]\]. This unexpected effect is due to a more homogeneous positioning of nucleosomes in the presence of H1 \[[@r57]\] and, as a consequence, a better binding of PR \[[@r29]; [@r57]\] and a higher proportion of promoters participating in transcription \[[@r29]\]. Moreover, in the presence of bound PR in an extract from *Drosophila* embryos, H1 is phosphorylated and subsequently removed from the promoter upon transcription initiation \[[@r29]\]. Histone acetylation ------------------- Additional possibilities for chromatin remodeling are recruitment by the receptors of chromatin remodeling activities, either ATP-dependent, such as the SWI/SNF complex \[[@r20]\], or ATP-independent, such as histone acetyltransferases (HATs) \[[@r30]\]. This latter possibility has received considerable attention following the discovery that several steroid receptor coactivators exhibit HAT activity, including members of the SRC-1 family, P/CAF and p300/CBP \[[@r10]\]. The role of histone acetylation in hormone induction of the MMTV promoter in not clear. High doses of histone deacetylase inhibitors, butyrate or TSA, lead to intense hyperacetylation of core histones and inhibit hormone induction \[[@r3]; [@r8]\] without altering nucleosome positioning \[[@r9]\]. However, low doses of the inhibitors activate the MMTV promoter in the absence of hormone and generate a DNAse I-hypersensitive site similar to that observed following hormone treatment \[[@r3]\]. Derepressing concentrations of the inhibitors of histone deacetylases enhance hormone-activated transcription, and thus the extent of hormonal induction is not affected \[[@r3]\]. Similar results are obtained with these inhibitors in the cell-free transcription assay with MMTV minichromosomes, suggesting that inhibitor-sensitive acetylation is not involved in mediating the synergism between receptors and NF1. More recently, a role for histone deacetylase 1 (HDAC1) has been proposed for MMTV activation \[[@r42]\]. HDAC1, generally considered as a corepressor, acts as a coactivator of the glucocorticoid receptor (GR) and its histone deacetylase activity is regulated by a p300-dependent acetylation \[[@r42]\]. The mechanism of this effect remains unclear. ATP-dependent remodeling ------------------------ The SWI/SNF complex, or other ATP-dependent chromatin remodeling factors, could destabilize the interaction between the DNA double helix and the histone octamer and thus facilitate access to otherwise masked *cis* elements. Hormone-induced chromatin remodeling might also facilitate the assembly of the transcription initiation complex (TIC) on the TATA box, and/or contribute to efficient transcription elongation through downstream nucleosomes. There is evidence for a requirement of the SWI/SNF complex in glucocorticoid gene regulation in yeast \[[@r62]\] and in animal cells \[[@r38]\], and hSWI/SNF seems to be required for chromatin remodeling initiated by glucocorticoids \[[@r26]\]. The situation appears to be different for progestins, which act through the same HREs of the MMTV promoter \[[@r26]\]. We have reproduced the synergism between SHR and NF1 in a cell-free system driven by MMTV minichromosomes assembled in extracts from *Drosophila* embryos \[[@r21]\]. These minichromosomes exhibit nucleosomes positioned in the same way as in cultured cells and the NF1 site in the promoter is not accessible \[[@r53]\]. Addition of hormone-activated GR or PR leads to ATP-dependent remodeling of these minichromosomes, as demonstrated by topoisomer analysis \[[@r21]\], and to simultaneous binding of receptors and NF1. This activity can be partly reproduced with purified MMTV minichromosomes with recombinant dISWI, but not with purified dCHRAC or ySWI/SNF complex \[[@r21]\]. Moreover, incubation of the minichromosomes in the *Drosophila* extract in the presence of receptors and NF1 leads to recruitment of dISWI, and dNURF301, two subunits of the NURF complex \[[@r27]\]. Thus, in the *Drosophila* embryo extract, PR recruits NURF to the MMTV promoter chromatin, and NURF remodels the minichromosomes in an ATP-dependent fashion. In mammalian cells with a single MMTV promoter integrated in chromatin we have recently found the hormone-induced recruitment of both Brg1- and ISWI-containing complexes to the promoter \[[@r56]; [@r58]\]. Moreover, in experiments with MMTV mononucleosomes assembled *in vitro*, PR is able to recruit purified yeast Swi/Snf to the promoter, where it catalyzes the ATP-dependent displacement of H2A/H2B dimers \[[@r58]\]. Though we cannot formally exclude the possibility of a replacement of the H2A/H2B dimers by dimers containing histone variants, our results strongly suggest that the activated MMTV nucleosome with bound SHRs and NF1 is a H3/H4 tetramer particle. This assumption is further supported by our previous findings showing that MMTV promoter sequences assembled *in vitro* on H3/H4 tetramers, adopt similar positions as on a histone octamer, but in contrast with sequences assembled on the octamer, are able to bind SHRs and NF1 \[[@r46]\]. There are two human SWI/SNF-like complexes and both contain ATPase subunits similar to yeast Swi2/Snf2, hBRM (human Brahma) or BRG1 (Brahma-Related Gene 1), as well as a series of other subunits, some of which differ in various cell types \[[@r60]\]. The hSWI/SNF-α complex, also called BAF (BRG1/hBRM-associated factor), contains either hBRM or BRG1 as the central ATPase subunit and is an orthologue of the yeast SWI/SNF. The hSWI/SNF-β complex, also called PBAF (Polybromo-associated BAF), contains only hBRM and is an orthologue to the yeast RSC complex \[[@r37]\]. The BAF and PBAF complexes share many subunits, but also contain subtype-specific subunits: BAF250 and hBRM are only found in BAF, whereas BAF180 and BAF 200 are only found in PBAF \[[@r37]; [@r61]\]. In breast cancer cells that contain BAF and PBAF, we have recently found that BAF is essential during the activation of MMTV and other endogenous progesterone-induced genes. BAF is recruited via an interaction between PR and the BAF57 subunit of the complex and is anchored by H3K14ac, generated by PCAF, suggesting a mechanism for the cooperation between two types of chromatin remodeling activities. The recruited BAF catalyzes the ATP-dependent displacement of histones H2A/H2B required for NF1 to gain access to the promoter site. Thus, synergism between the two transcription factors PR and NF1 is mediated by the cooperation between two chromatin-remodeling machines, BAF and PCAF \[[@r59]\]. Link between protein kinases and transcriptional control -------------------------------------------------------- In addition to their transcriptional effect, steroid hormones also crosstalk to kinase signaling pathways, and this crosstalk is essential for the proliferative response of breast cancer cells to estrogens \[[@r35]\] and progesterone \[[@r36]\]. What could be the connection between the rapid hormonal activation of cytoplasmic signaling cascades and the events taking place at the MMTV promoter? We have investigated the function of Erk activation by progesterone, which is mediated via an interaction between PR and ERα ([Figure 2](#fig2){ref-type="fig"} and \[[@r2]\]). We observed that progestin activation of the MMTV promoter can be blocked by inhibiting Erk activation either with antiestrogens, with small molecular inhibitors (PD 98059), with dominant negative kinase mutant, or with RNAi against Erk \[[@r55]\]. As Erk phosphorylated PR at ser294 in response to progesterone, it was possible that the transcriptional inhibition of MMTV induction was due to a lack of PR phosphorylation. However, a similar inhibition of MMTV induction was observed when the activation of Msk1 was prevented by small molecular inhibitors (H89), a dominant negative mutant of Msk, or down-regulation of Msk1 with an specific RNAi \[[@r55]\]. Unexpectedly, after 5 minutes of hormone treatment, we detected a ternary complex of phosphorylated PR and activated Erk and Msk1 ([Figure 2](#fig2){ref-type="fig"}), which is selectively recruited to the MMTV promoter nucleosome containing the HREs ([Figure 3](#fig3){ref-type="fig"}) \[[@r55]\]. Concomitant with the recruitment of the ternary complex, histone H3 becomes phosphorylated at serine 10 and acetylated at lysine 14, only on the nucleosome containing the HREs and not on adjacent nucleosomes ([Figure 3](#fig3){ref-type="fig"}) \[[@r55]\]. Blocking H3 phosphoacetylation precludes displacement of a repressive complex containing HP1γ and recruitment of BAF, preventing H2A/H2B displacement and promoter activation. Microarray analysis of the progesterone response in breast cancer cells (Ballare et al., manuscript in preparation), showed that about 25% of the hormonally-regulated genes are sensitive to the inhibition of Erk activation. This shows that the behavior of the MMTV promoter does not represent a rare exception, but also suggests that other signaling pathways are involved in mediating progesterone action. We know that the PI3K/Akt pathway is involved in the regulation of stromal cell proliferation in the endometrium \[[@r52]\], and also in the proliferative response of breast cancer cells to estrogens \[[@r15]\]. There is also evidence that the cAMP/PKA pathway is implicated in estrogen action (\[[@r1]; [@r18]\]), and that the JAK/STAT5 plays an important role in glucocorticoid and progesterone effects in mammary gland (\[[@r13]; [@r49]; [@r50]\]). It remains to be established whether the crosstalk of the SHR with these different pathways acts merely by regulating the phosphorylation of key transcription factors and coregulators, or whether, as in the case of the Erk pathway, modifications of structural components of the chromatin also play an important role. Conclusions =========== The results summarized in this review contribute to a better understanding of the molecular mechanisms of promoter activation by progesterone. During activation of the MMTV promoter, the chromatin constitutes a key component essential for regulation of transcription by allowing synergism between the nuclear receptors and NF-1, causing full activation of the promoter. The linker histone H1, known as a repressive component of chromatin needed for condensation, increases MMTV transcription by improving the positioning of nucleosomes. PR interacts with an exposed HRE on the surface of a nucleosome and initiates chromatin remodeling as a prerequisite for the subsequent steps of gene activation. The nucleosomal DNA sequence, as part of chromatin, contains the information required to direct the outcome of the subsequent remodeling event. The SWI/SNF complex, depending on the DNA sequence, can displace histones H2A and H2B or slide the nucleosome toward one end of the DNA fragment. Thus, the "final outcome" in each case is completely different, depending on structural features of the underlying DNA, which confer the differential response to SWI/SNF action and should be further developed. The results with the MMTV promoter underline the significance of incorporating a precise knowledge of the structure of native DNA sequences in chromatin. Evidence for the positive contributions of chromatin structure and dynamics to the process of gene regulation continues to accumulate at an increasing pace. A classical example is a situation in which a positioned nucleosome brings in close proximity *cis*-regulatory elements located some 200 bp apart and in this way facilitates the interaction between DNA-bound factors. There are several examples of this type of regulation (for a recent report see \[[@r64]\]), and one case involves an interaction of the estrogen receptor with the transcription factor NF1 \[[@r45]\]. A particularly illustrative example of the significance of chromatin is the regulation of the Human Immunodeficiency Virus type 1 (HIV-1), the etiologic agent of AIDS. Recent advances have indicated that HIV-1 encoded proteins interact with chromatin-remodeling complexes and histone-modifying enzymes, implying that chromatin remodeling plays an important role in the HIV-1 life cycle \[[@r28]; [@r32]\]. Nucleosomes are positioned on the HIV-1 LTR and act as barriers to transcription. Following cellular activation, these nucleosomes are modified and repositioned, allowing for activation of viral gene expression \[[@r23]; [@r54]\]. Another example where chromatin has been shown to be essential is the gene regulation of the rat albumin gene enhancer. In its active state, in liver nuclei, the enhancer is organized into a nuclease-accessible array of three positioned nucleosome-like particles N1, N2, and N3, with bound transcription factors \[[@r34]\]. The N1 nucleosome-like particle at the albumin enhancer is positioned over the FoxA binding sites and is DNase I-hypersensitive in hepatic cells. In contrast, in non-liver tissues, where the FoxA sites are not occupied, nucleosome-like particles are not precisely positioned over the enhancer, and the enhancer is not hypersensitive \[[@r34]\]. It has been observed that FoxA and GATA-4 bound their sites in compacted chromatin and opened the local nucleosomal domain in the absence of ATP-dependent enzymes \[[@r19]\]. These are just a few examples of the intricacies that the chromatin organization of promoters and regulatory regions impose on gene regulation. This complexity is partly due to the fact that one of the key events during cell differentiation in complex organisms is the progressive chromatin-mediated silencing of genes that are not required in a particular cell lineage and the proper chromatin organization of those genes that will be needed at some later stage of differentiation or in response to external signals. Understanding this differential chromatin organization of particular genome regions is the main challenge for epigenetic research in the coming years. G.P.V. was a recipient of a fellowship of the Ramón y Cajal Programme. The experimental work was supported by grants from the European Union (HEROIC integrated project), the Departament d´Innovacio Universitat I Empresas (DIUE), Ministerio de Educación y Ciencia (MEC) BMC 2003-02902 and Fondo de Investigación Sanitaria (FIS) PI0411605 and CP04/00087. BAF : BRG1/hBRM-associated factor BRG1 : Brahma-related gene 1 CBP : CREB binding protein CHRAC : chromatin accessibility complex Erk : extracellular signal-regulated kinase ERα : estrogen receptor α GR : glucocorticoid receptor HATs : histone acetyltransferases HDAC1 : histone deacetylase 1 HIV-1 : human immunodeficiency virus type 1 HP1γ : heterochromatin protein 1γ HREs : hormone responsive elements ISWI : imitation SWI remodeling factor MMTV : mouse mammary tumor virus Msk1 : mitogen and stress response kinase-1 NF1 : nuclear factor 1 NURF : nucleosome remodeling factor PBAF : polybromo-associated BAF PCAF : P300/CBP-associated factor PR : progesterone receptor SHRs : steroid hormone receptors SRC-1 : steroid receptor coactivator-1 SWI/SNF : Switch/Sucrose nonfermentable nucleosome remodeling complex TIC : transcription initiation complex

1. Introduction {#sec1-marinedrugs-18-00224} =============== An increase in viral infections among all human infectious diseases and a lack of effective antivirals is one of the most significant problems of healthcare worldwide. Quite a few of currently used antivirals have a relatively low antiviral activity with a narrow spectrum of action (\<one virus/ one medicine\>), and viruses frequently become resistant to such antivirals. Therefore, the search for antiviral compounds that can effectively block the reproduction of a broad spectrum of human viruses and the development of new drugs based on such antivirals is quite essential. Global pharmaceutical experience shows that marine hydrobionts have significant potential as raw materials for the development of drug products. In recent years, the sulfated polysaccharides from various marine algae, including fucoidans from brown algae, have attracted researchers' attention. The main components of the fucoidans molecule are the sulfated L-fucose residues. In addition to fucose residues, fucoidans often contain other monosaccharides such as galactose, uronic acids, xylose, mannose and others \[[@B1-marinedrugs-18-00224]\]. These polysaccharides have various biological activities---anticoagulant, anti-inflammatory, antitumor, antioxidant, adjuvant, and antiviral \[[@B2-marinedrugs-18-00224],[@B3-marinedrugs-18-00224],[@B4-marinedrugs-18-00224],[@B5-marinedrugs-18-00224]\]. Their biological properties are determined by the structure of the main chain, molecular weight, content, and location of sulfate and acetate groups \[[@B6-marinedrugs-18-00224],[@B7-marinedrugs-18-00224],[@B8-marinedrugs-18-00224],[@B9-marinedrugs-18-00224]\]. A study of the pharmacokinetics and tissue distribution of fucoidans isolated from different types of brown algae showed that high-molecular-weight fucoidans are determined for a long time in the blood, mainly accumulating in the spleen, kidneys and liver, and excreted in the urine \[[@B10-marinedrugs-18-00224],[@B11-marinedrugs-18-00224]\]. Fucoidans are the basis of biologically active compounds \[[@B12-marinedrugs-18-00224]\] used as drug carriers \[[@B13-marinedrugs-18-00224],[@B14-marinedrugs-18-00224]\]. Therapeutic characteristics of fucoidans have been assessed in clinical trials \[[@B15-marinedrugs-18-00224],[@B16-marinedrugs-18-00224]\]. However, despite the apparent progress in studying the properties of fucoidans, there are no registered fucoidans-based drug products, in part due to difficulties in obtaining standard polysaccharide samples. One of the strategies for the development of simpler biologically active fucoidans with a standardized and reproducible structure is the enzymatic transformation of fucoidans \[[@B17-marinedrugs-18-00224],[@B18-marinedrugs-18-00224]\]. Previously, the enzymatic depolymerization of the native fucoidan from brown alga *Fucus evanescens* was carried out using recombinant fucoidanase FFA1 from marine bacteria *Formosa algae* to obtain a standardized fragment of fucoidan molecule with the regular structure \[[@B19-marinedrugs-18-00224]\]. In the present study, we used both fucoidans:---native fucoidan isolated from *F. evanescens* --- FeF (*F. evanescens* Fucoidan) and its derivative obtained after enzymatic treatment --- FeHMP (*F. evanescens* High-Molecular-Weight Product). The FeF (molecular weight 160 kDa), as previously shown, built up from randomly alternating α-(1→3)- and α-(1→4)-linked residues of sulfated fucose residues \[[@B20-marinedrugs-18-00224],[@B21-marinedrugs-18-00224],[@B22-marinedrugs-18-00224]\]. The FeHMP (molecular weight---50.8 kDa) has a regular structure and built from repeating fragments of (\[→3)-α-L-Fucp(2,4OSO~3~−)-(1→4)-α-L-Fucp(2OSO~3~−)-(1→\]~n~) \[[@B19-marinedrugs-18-00224],[@B20-marinedrugs-18-00224],[@B21-marinedrugs-18-00224],[@B22-marinedrugs-18-00224]\]. The aim of this study was the comparative analysis of the antiviral activity of the native fucoidan from *F. evanescens* (FeF) and its derivative (FeHMP) with the regular structure against several DNA- and RNA-containing viruses. Investigation of the antiviral activity of native fucoidan (FeF) with nonregular structure and standardized fucoidan (FeHMP) can give us the answer of which structural fragment of these unique polysaccharides is important against viruses. 2. Results {#sec2-marinedrugs-18-00224} ========== 2.1. The Cytotoxicity and In Vitro Antiviral Activity of Fucoidans against DNA and RNA Viruses {#sec2dot1-marinedrugs-18-00224} ---------------------------------------------------------------------------------------------- The data for cytotoxicity determined by MTT (methylthiazolyltetrazolium bromide) assay showed low toxicity of the native (FeF) and modified (FeHMP) fucoidans and Acyclovir^®^ against Vero cells---their 50% cytotoxic concentrations (CC~50~) were above 2000 μg/mL, while CC~50~ of Ribavirin^®^ was 750 μg/mL. Antiviral activity assay was performed at the concentrations below 500 µg/mL. In the case of MT-4 cells, the CC~50~ of fucoidans were 200 μg/mL, and CC~50~ of the Retrovir^®^ and Epivir^®^ were 50 μg/mL. Thus, their anti-HIV-1 activity was assessed at the concentrations below their CC~50~. The antiviral effect of fucoidans against herpes simplex viruses type 1 and 2 (HSV-1 and HSV-2), enterovirus (ECHO-1) and human immunodeficiency virus (HIV-1) was assessed using cytopathic effect (CPE) inhibition assay. To study the inhibitory effect of tested compounds on the stage of virus infection, the fucoidans were added before virus infection (pretreatment of cells), directly to the virus suspension (pretreatment of virus), concurrently with the initiation of virus infection (simultaneous treatment), and after penetration of the virus into host cells (treatment of infected cells), respectively. The results of the virus-inhibitory activity of tested compounds were used for calculations of the 50% inhibitory concentration (IC~50~) and the selectivity index (SI) for each of the compounds ([Table 1](#marinedrugs-18-00224-t001){ref-type="table"}). The pretreatment of DNA-containing viruses (HSV-1 and HSV-2) with fucoidans (direct virucidal action) showed a moderate antiviral activity of FeF and FeHMP (their mean values of SI were \~ 19). In the case of RNA-containing viruses, the virucidal action of fucoidans against HIV-1 was modest (SI = 8), and minor against nonenveloped RNA-containing ECHO-1 virus (SI\~3). This method of application of fucoidans showed no significant difference between FeF and FeHMP (*p* ˃ 0.05) ([Table 1](#marinedrugs-18-00224-t001){ref-type="table"}). The treatment of cells with fucoidans before infection (preventive effect) and just after the virus inoculation (0 h) (simultaneous treatment) revealed the highest antiviral activity of tested compounds ([Table 1](#marinedrugs-18-00224-t001){ref-type="table"}). The native fucoidan effectively inhibited the replication of both HSV types when compared with modified fucoidan---SI of FeF was 1.5--1.9 times higher than with FeHMP. In the case of ECHO-1, the antiviral activity of fucoidans was moderate (SI\~22), and the difference between the antiviral effect of FeF and FeHMP was not significant (*p* ˃ 0.05). The anti-HIV-1 activity of fucoidans in this application mode was modest (SI\~8). The application of sulfated polysaccharides after virus adsorption and penetration to cells (at 1 h postinfection) (treatment of infected cells) showed moderate replication inhibition against HSV-1, HSV-2, and ECHO-1 (the average SI was 23) and modest virus-inhibitory activity of fucoidans against HIV-1 (SI\~8). The difference between FeF and FeHMP was not significant (*p* ˃ 0.05) ([Table 1](#marinedrugs-18-00224-t001){ref-type="table"}). Our results demonstrated the ability of native and modified fucoidans to inhibit the replication of HSV-1, HSV-2, ECHO-1, and HIV-1; herpes simplex viruses, especially HSV-2, were the most sensitive to tested sulfated polysaccharides. 2.2. In Vivo Efficacy of Fucoidans against HSV-2 Infection in a Mouse Vaginitis Model {#sec2dot2-marinedrugs-18-00224} ------------------------------------------------------------------------------------- Having identified the antiviral potency of native and modified fucoidans in vitro, we evaluated their protective efficacy against the intravaginal HSV-2 challenge in mice. The clinical symptoms of infection were observed to start on the 5th day postinfection and included loss of body weight, vaginal swelling, hyperemia, and discharge. Also, the decreased motor activity, food, and water intake were observed, followed with hind limb paralysis on the 7th day postinfection. The average survival time of infected animals in the virus group was 9.7 ± 2.6 days ([Table 2](#marinedrugs-18-00224-t002){ref-type="table"}). The treatment of HSV-2-infected mice with fucoidans from *F. evanescens* (FeF and FeHMP) resulted in a dose-dependent antiviral effect, leading to reduced clinical symptoms and mortality ([Table 2](#marinedrugs-18-00224-t002){ref-type="table"}, [Figure 1](#marinedrugs-18-00224-f001){ref-type="fig"}A). Fucoidans (FeF and FeHMP) given intraperitoneally at 10 mg/kg/d provided a survival rate of 44--56% in infected mice. The average survival time of fucoidans-treated mice increased by 4--6 days compared to the virus group (*p* ≤ 0.05). Furthermore, the administration of these doses of fucoidans to HSV-2-infected animals effectively prevented the loss of their body weight ([Figure 1](#marinedrugs-18-00224-f001){ref-type="fig"}B). The treatment of infected mice with Acyclovir^®^ (50 mg/kg/d) protected all animals from the lethal outcome, so the protection index was 100%. On the 7th day postinfection, the replication of HSV-2 in the vaginal epithelium of mice in the virus group reached 3.63 ± 0.18 lg TCID~50~/mL ([Figure 1](#marinedrugs-18-00224-f001){ref-type="fig"}C). The intraperitoneal administration of fucoidans at 10 mg/kg/d significantly reduced virus replication in the infected vagina compared to the virus group (*p* ≤ 0.05) ([Figure 1](#marinedrugs-18-00224-f001){ref-type="fig"}C). By the seventh day of the treatment with 10 mg/kg/d of native fucoidan (FeF), a reduction of HSV-2 titers in the vaginal lavages by 2 lg TCID~50~/mL was observed. Similarly, the modified fucoidan (FeHMP, 10 mg/kg/d) and Acyclovir^®^ (50 mg/kg/d) significantly reduced HSV-2 titers by 3 lg TCID~50~/mL. Based on these data, it can be concluded that fucoidans FeF and FeHMP effectively protect mice from intravaginal HSV-2 challenge. 3. Discussion {#sec3-marinedrugs-18-00224} ============= It is well-known that fucoidans from different brown algae have antiviral activity against DNA-containing \[[@B24-marinedrugs-18-00224],[@B25-marinedrugs-18-00224],[@B26-marinedrugs-18-00224]\] and RNA viruses \[[@B27-marinedrugs-18-00224],[@B28-marinedrugs-18-00224],[@B29-marinedrugs-18-00224],[@B30-marinedrugs-18-00224]\], and are promising candidates for the development of fucoidan-based drug products. In the meantime, the standardization of fucoidan extraction is a complicated process \[[@B31-marinedrugs-18-00224],[@B32-marinedrugs-18-00224]\], mainly due to the structural diversity of fucoidans \[[@B33-marinedrugs-18-00224]\] and sometimes because of biotechnological problems \[[@B34-marinedrugs-18-00224]\]. Therefore, the critical task is the identification of the structural fucoidans fragments responsible for their biological activity as well as the development of standardized fucoidan-based drug products with a pronounced antiviral effect. Our study aimed to investigate the effect of the native fucoidan and its regular derivative at the different stages of the virus life cycle, including significant human pathogens HSV-1, HSV-2, HIV-1, and ECHO-1. The results showed the ability of both fucoidans to increase the resistance to virus infection (preventive effect), directly affect virus particles (virucidal effect), and inhibit the early stage of virus replication (virus-inhibiting effect). The multifaceted mechanisms of action of tested fucoidans are similar to the mechanisms of action of other sulfated polysaccharides from different algae \[[@B35-marinedrugs-18-00224],[@B36-marinedrugs-18-00224]\]. Nevertheless, we showed some specific features of the antiviral activity of FeF and FeHMP against the aforementioned viruses. For example, tested fucoidans more effectively inhibited the replication of the DNA-containing HSV-1 and HSV-2, compared with a moderate antiviral effect against RNA viruses ECHO-1 and HIV-1. It was found that fucoidans affected different stages of HSV-1 and HSV-2 replication more effectively at the stage of virus adsorption and penetration to the host cells, as demonstrated by the treatment of cells with fucoidans before virus infection and simultaneous treatment of cells with the mixture of fucoidan and virus. We hypothesize that one of the possible mechanisms of high anti-HSV activity of studied fucoidans can be related to the sulfated polysaccharides' ability to interact with cells to competitively inhibit binding sites usually used by herpesviruses for cell entry \[[@B35-marinedrugs-18-00224],[@B37-marinedrugs-18-00224],[@B38-marinedrugs-18-00224]\]. Comparative analysis of antiviral activities of native (FeF) and modified (FeHMP) fucoidans showed that FeF more effectively inhibits the replication of both types of HSV than FeHMP. On the other hand, both tested fucoidans had similar antiviral activity against ECHO-1 and HIV-1. It was previously demonstrated \[[@B19-marinedrugs-18-00224]\] that FeHMP had lower molecular weight and was significantly more sulfated than FeF. Some authors reported that the antiviral activity of sulfated polysaccharides increased with molecular weight and sulfate content \[[@B37-marinedrugs-18-00224],[@B38-marinedrugs-18-00224],[@B39-marinedrugs-18-00224]\]. However, we believe that the revealed differences in the anti-HSV activity of FeF and FeHMP can be associated with the fine structure of both polysaccharides rather than with a decrease in the molecular weight or increase in the number of sulfate groups in FeHMP fucoidan. It is possible that enzymatic hydrolysis brings about the decrease of structural fragments' diversity in the fucoidan molecule, leading to the reduction in the number of potential fucoidan targets on the cell surface. Presumably, it leads to the lower anti-HSV activity of the modified fucoidan FeHMP compared with the native fucoidan FeF. In our opinion, the antiviral activity of tested fucoidans from *F. evanescens* (FeF and FeHMP) against nonenveloped RNA viruses (for example, ECHO-1 enterovirus), is very important. Currently, there are only a few reports concerning the antienteroviral activity of sulfated polysaccharides from marine algae \[[@B40-marinedrugs-18-00224],[@B41-marinedrugs-18-00224],[@B42-marinedrugs-18-00224]\]. Our results suggest that mechanisms of anti-ECHO-1 activity of fucoidans can be related to the inhibition of virus adsorption and early stages of viral infection. It should be noted that in the case of enterovirus, the fucoidans antiviral effect was greater than that of Ribavirin^®^ ([Table 1](#marinedrugs-18-00224-t001){ref-type="table"}). The last ten years were characterized by the appearance of numerous reports about the anti-HIV activity of sulfated polysaccharides isolated from different algae \[[@B43-marinedrugs-18-00224],[@B44-marinedrugs-18-00224],[@B45-marinedrugs-18-00224]\]. It is believed that the primary mechanism of anti-HIV action is the ability of these compounds to block the virus penetration to the sensitive host cells via binding to positively charged amino acids of virus envelope glycoprotein gp120, which mediates viral attachment \[[@B4-marinedrugs-18-00224],[@B38-marinedrugs-18-00224],[@B46-marinedrugs-18-00224]\]. Several authors reported inhibition of the HIV-1 life cycle both before and after virus penetration to cells with some sulfated polysaccharides \[[@B47-marinedrugs-18-00224],[@B48-marinedrugs-18-00224]\]. Our study results also showed that native and modified fucoidans inhibited different stages of HIV-1 replication in human T cells (MT-4 cell line) when applied at the same inhibitory concentration (IC~50~ = 25 μg/mL) ([Table 1](#marinedrugs-18-00224-t001){ref-type="table"}). The high worldwide prevalence of HSV-2 infections, the severity of complications and their close linkage with cervical cancer and HIV-1 infection allow attributing the diseases caused by HSV-2 to global healthcare and social issues \[[@B49-marinedrugs-18-00224],[@B50-marinedrugs-18-00224],[@B51-marinedrugs-18-00224]\]. Therefore, the search for anti-HSV-2 drugs with different mechanisms of action is still very important. Previously, the anti-HSV-2 activity of sulfated polysaccharides from seaweed and mushroom was shown in the mouse model of genital herpesvirus infection \[[@B52-marinedrugs-18-00224],[@B53-marinedrugs-18-00224]\]. In our study, the protective efficacy of tested fucoidans against intravaginal HSV-2 challenges in mice was demonstrated ([Table 2](#marinedrugs-18-00224-t002){ref-type="table"}, [Figure 1](#marinedrugs-18-00224-f001){ref-type="fig"}). Intraperitoneal administration of native fucoidan FeF and its derivative FeHMP significantly improved the survival rate, alleviated symptoms of the disease, prevented the weight loss, and reduced vaginal virus load induced by HSV-2 infection compared to the virus group. We assume that the protective antiviral effect of fucoidans is associated not only with selective influence on a different stage of viral infection but also with antioxidant, anti-inflammatory, and immunomodulatory properties of these compounds. The results of our study showed that standardized fucoidan with a regular structure as well as native fucoidan have comparable potency against a range of DNA- and RNA-containing viruses associated with severe human pathology. We suggest that the antiviral properties of these unique polysaccharides may be due to structural features, and in particular, with their highly sulfated fragment (\[→3)-α-L-Fucp(2,4OSO~3~−)-(1→4)-α-L-Fucp(2OSO~3~−)-(1→\])~n~. Obtaining standardized fucoidan preparations can be a basis of a successful strategy for the development of promising broad-spectrum antivirals. 4. Materials and Methods {#sec4-marinedrugs-18-00224} ======================== 4.1. Viruses and Cell Cultures {#sec4dot1-marinedrugs-18-00224} ------------------------------ The following viruses were used for the study---herpes simplex virus type 1 (HSV-1) strain VR3 was obtained from the National Collection of US Viruses (Rockville, MD, USA). The herpes simplex virus type 2 (HSV-2) strain G ATCC VR-734 was obtained from the Smorodintsev Research Institute of Influenza (Sankt-Petersburg, Russia). The strain IP91 of ECHO-1 enterovirus was obtained from the Chumakov Federal Scientific Center for Research and Development of Immune and Biological Products (Moscow, Russia). The human immunodeficiency virus (HIV-1) strain M070, subtype A6, was from the collection of HIV strains of the Mechnikov Research Institute of Vaccines and Sera (Moscow, Russia). HSV-1, HSV-2, and ECHO-1 were grown in African green monkey kidney (Vero) cells using Dulbecco's Modified Eagle's Medium (DMEM) supplemented with 10% Fetal Bovine Sera (FBS) and 100 U/mL of gentamycin at 37 °C in a CO~2~ incubator; in the maintenance medium, the FBS concentration was decreased to 1%. HIV-1 was propagated in the human T-cell line MT-4 using RPMI-1640 medium supplemented with 10% FBS, 0.06% L-Glutamine, and 50 U/mL gentamycin. 4.2. Animals {#sec4dot2-marinedrugs-18-00224} ------------ Female outbred mice (16--20 g) were obtained from the Scientific Center for Biomedical Technology of the Federal Medical and Biological Agency (Andreevka, Moscow Region, Russia) and used for in vivo experiments. All procedures were performed strictly following the (European Convention for the Protection of Vertebrate Animals Used for Experimental and other Scientific Purposes) of 18 March 1986. All animal experiments comply with bioethical standards; the study protocol was approved by the local institutional bioethics committee of the Mechnikov Institute (protocol N 4, of 20.08.2019). 4.3. Studied Compounds {#sec4dot3-marinedrugs-18-00224} ---------------------- Native fucoidan (FeF)---sulfated polysaccharide from brown algae *Fucus evanescens*. Fucoidan from *F. evanescens* (FeF) was purified as described early \[[@B54-marinedrugs-18-00224]\] and characterized \[[@B20-marinedrugs-18-00224],[@B21-marinedrugs-18-00224],[@B22-marinedrugs-18-00224]\]. According to NMR spectroscopy data, native fucoidan FeF was built up from alternating α-(1→3)- and α-(1→4)-linked residues of sulfated fucose residues: (→3)-α-L-Fucp(2,4OSO~3~−)-(1→4)-α-L-Fucp(2OSO~3~−)-(1→) and (→3)-α-L-Fucp(2OSO~3~−)-(1→4)-α-L-Fucp(2OSO~3~−)-(1→). Modified fucoidan (FeHMP)---high molecular weight product (HMP) of enzyme hydrolysis of native fucoidan. To obtain a fucoidan derivative with a regular structure, we used the method described in \[[@B19-marinedrugs-18-00224]\]. In brief, fucoidan from *F. evanescens* (FeF, 0.5 g) was dissolved in 49 mL of the buffer (0.04 M Tris-HCl, pH 7.0 with 5 mM CaCl~2~) and 1 ml of enzyme (fucoidanase FFA1, 0.1 mg/mL) was added. The reaction mixture was incubated at 34 °C for 72 h, and then heated at 80 °C for 10 min, and the precipitate was removed by centrifugation. The high-molecular-weight reaction products (FeHMP) (in supernatant) were precipitated with ethanol at the ratio of 1:3 (v/v), and precipitate was separated by centrifugation at 10,000× *g* for 40 min. ^1^H NMR spectra of both native and modified fucoidans are presented in [supplementary data (Figure S1)](#app1-marinedrugs-18-00224){ref-type="app"}. FeHMP has a regular structure and consists of a repeating fragment: (\[→3)-α-L-Fucp(2,4OSO~3~−)-(1→4)-α-L-Fucp(2,4OSO~3~−)-(1→\])~n~ \[[@B19-marinedrugs-18-00224]\]. Summarized information with structural characteristics of fucoidans is presented in [Table 3](#marinedrugs-18-00224-t003){ref-type="table"}. Fucoidans (FeF and FeHMP) and Acyclovir^®^ for experiments in vitro and *in vivo* were diluted in DMEM. Stock solution (10 mg/mL) of Ribavirin^®^ for in vitro experiments was dissolved in dimethylsulfoxide (DMSO, Sigma, Saint-Louis, MO, USA), stored at −20 °C and was diluted with a suitable cell culture medium to a final concentration of 0.5% DMSO. Antiretrovirals were diluted with RPMI-1640 (Pan-Eco, Moscow, Russia). 4.4. Cytotoxicity Assay of the Fucoidans {#sec4dot4-marinedrugs-18-00224} ---------------------------------------- The cytotoxicity of fucoidans was estimated by MTT assay in Vero cells \[[@B55-marinedrugs-18-00224]\]. Cell monolayers (1 × 10^4^ cells/well) in the 96-well plates were treated with different concentrations of compounds (from 0.2 to 2000 μg/mL) and incubated at 37 °C in a CO~2~-incubator for three days; untreated cells were used as controls. MTT solution (methylthiazolyltetrazolium bromide, Sigma, Saint-Louis, MO, USA) was added to cells in the concentration 5 mg/mL, following incubation for 2 h at 37 °C. Then, the MTT solution was removed, and isopropanol was added to dissolve the insoluble formazan crystals. The optical density was measured at 540 nm using an ELISA microplate reader (Labsystems Multiskan RC, Vantaa, Finland). Cytotoxicity was expressed as 50% cytotoxic concentration (CC~50~) of each studied compound that reduced the viability of treated cells by 50% compared with untreated cells and was calculated using regression analysis of dose-response curve \[[@B56-marinedrugs-18-00224]\]. The cytotoxicity of fucoidans for lymphoblast cell line MT-4 was also assessed by an MTT assay. The cells were incubated in the 96-well plates, treated with different concentrations of fucoidans (0.25--250 μg/mL), and cultured at 37 °C in a CO~2~ incubator for five days; the untreated MT-4 cells were used as controls. 4.5. Antiviral Activity Assay of Fucoidans In Vitro {#sec4dot5-marinedrugs-18-00224} --------------------------------------------------- The antiviral activity of fucoidans against HSV-1, HSV-2, ECHO-1, and HIV-1 was evaluated using the cytopathic effect (CPE) inhibition assay in the Vero and MT-4 cells, respectively. The monolayers of Vero cells grown on 96-well plates (1 × 10^4^ cells/well) were infected with 100 TCID~50~/mL (the 50% tissue culture infectious dose) of the corresponding virus (HSV-1, HSV-2, ECHO-1). Suspension of MT-4 cells (3 × 10^5^cells/mL) was cultured in 96-well flat-bottomed plates followed by infection of 100 TCID~50~/mL of HIV-1. Some schemes of fucoidans and appropriate pharmaceutical application were investigated; each of them was carried out in three independent replicates in triplicate, with different concentrations of compounds (0.25--250 μg/mL). The plates were incubated at 37 °C in a CO~2~ incubator for 72 h for HSV-1, HSV-2, and ECHO-1 and 120 h for HIV-1 until 80--90% CPE was observed in virus control compared with cell control. The pretreatment of cells with fucoidans. The monolayer of cells was pretreated with different concentrations of studied compounds for 2 h at 37 °C. After washing, the cells were infected with 100 TCID~50~/mL of the virus at 37 °C for one hour. Then, unabsorbed virus was removed by washing with phosphate-buffered saline (PBS), and cells were incubated in the maintenance medium until CPE appeared. The pretreatment of the virus with fucoidans. An infectious dose of virus (100 TCID~50~/mL) was mixed with different concentrations of studied compounds at a ratio 1:1 (v/v), incubated for one hour at 37 °C; then, the mixture was applied to the monolayer of cells. After 1 h adsorption at 37 °C, the cells were washed with PBS, overlaid by the maintenance medium, and followed by incubation until CPE was observed. The simultaneous treatment of the cells with fucoidans and viruses. The monolayer of cells was infected with the virus (100 TCID~50~/mL) and simultaneously treated by different concentrations of the studied compounds (virus: compound, 1:1 v/v) for one hour at 37 °C. After virus adsorption, the mixture was removed; the cells were washed with PBS and incubated in maintenance medium until CPE appeared. The treatment of virus-infected cells with fucoidans. The monolayer of cells was infected with the virus (100 TCID~50~/mL) at 37 °C for 1 h, and then the cells were washed with PBS and treated with different concentrations of the studied compounds and incubated before apparent CPE. After the incubation, the cell supernatants were collected, virus titers were calculated using the Reed-Muench method \[[@B57-marinedrugs-18-00224]\] and expressed as TCID~50~/mL. The antiviral effect of fucoidans was determined by the difference of viral titers between the treated infected cells and untreated infected cells and expressed as virus inhibition rate (IR, %). IR was calculated according to the following formula \[[@B58-marinedrugs-18-00224]\]: IR = (1 − *T*/*C*) × 100%, where *T* is the antilog of the compound-treated viral titers, and C is the antilog of the control (without compound) viral titers. The fifty per cent inhibitory concentration (IC~50~) of each compound was determined as the compound concentration that inhibited virus-mediated CPE by 50% and was calculated using a regression analysis of the dose-response curve \[[@B56-marinedrugs-18-00224]\]. Selectivity index (SI) was calculated as the ratio of CC~50~ to IC~50~ for each compound. 4.6. Antiviral Activity Assay of Fucoidans against HSV-2 In Vivo {#sec4dot6-marinedrugs-18-00224} ---------------------------------------------------------------- To evaluate the anti-HSV-2 activity of fucoidans *in vivo*, we used a mouse vaginitis model by vaginal infection with the HSV-2. Female outbred mice (16--20 g) were randomly divided into 7 groups with 10 mice in each group as follows---normal group (control uninfected mice), virus group (infected and untreated mice), and infected and treated groups, including FeF-treated (5 mg/kg/d and 10 mg/kg/d), FeHMP-treated (5 mg/kg/d and 10 mg/kg/d), and Acyclovir-treated (50 mg/kg/d). The mice in the normal and virus groups were administered saline. Genital herpes virus infection of mice was established by vaginal inoculation of 30 μL HSV-2 (10^5^ TCID~50~/mL). At 1 h postinoculation and subsequently once a day for five consecutive days, all tested compounds were administered intraperitoneally, and the animals were observed daily for 21 days to calculate the survival rate, death protection, average survival time, and body weight change. Also, the change of virus titer was analyzed---vaginal lavages were collected from the animals in each group on the 5^th^ and 7^th^ days postinfection. Briefly, 30 μL precooled DMEM was introduced to the vagina of infected mice and pipetted, collected vaginal lavages were adjusted to 150 μL by DMEM and centrifuged (500 g, 10 min). Following this, 10-fold dilutions of supernatants were placed on Vero cells monolayer in 96-well plates, incubated for three days at 37 °C in a CO~2~-incubator, and virus titer (lg TCID~50~/mL) was determined by the Reed-Muench method \[[@B57-marinedrugs-18-00224]\]. 4.7. Statistical Analysis {#sec4dot7-marinedrugs-18-00224} ------------------------- Statistical analysis was done with Statistica 10.0 software (StatSoft Inc, Tulsa, OK, USA). The results are given as mean ± standard deviation (SD). The differences between parameters of control and experimental groups were estimated using the Wilcoxon test. Differences were considered significant at *p* ≤ 0.05. ###### Click here for additional data file. The following are available online at <https://www.mdpi.com/1660-3397/18/4/224/s1>, Figure S1: ^1^H NMR spectra of native fucoidan FeF from F. evanescence (A) and its derivative FeHMP obtained after treatment with fucoidanase FFA1 from marine bacterium *Formosa algae* (B). N.V.K., S.P.E., V.F.L., M.N.N. and T.S.Z. planned and summarized the work and manuscript preparation; N.V.K. wrote the manuscript; A.S.S. prepared fucoidans; G.G.K., O.V.I. and L.K.E. investigated the HSV antiviral activity; M.N.N. examined the HIV antiviral activity; I.A.L., N.V.K. and I.N.F. examined in vivo the HSV-2 antiviral activity. All authors have read and agreed to the published version of the manuscript. This work was supported by the Ministry of Science and Education of the Russian Federation (2019/№ 0545-2019-0006). The authors declare no conflict of interest. {#marinedrugs-18-00224-f001} marinedrugs-18-00224-t001_Table 1 ###### Spectrum of antiviral activity of fucoidans. Viruses Compounds Pretreatment of Virus Pretreatment of Cells Simultaneous Treatment Treatment of Infected Cells ------------ ----------- ----------------------- ----------------------- ------------------------ ----------------------------- ----------- ------------ ---------- -------- **HSV-1** FeF 106 ± 13 19 ± 2 53 ± 7 38 ± 4 59 ± 8 34 ± 4 80 ± 9 25 ± 3 FeHMP 127 ± 15 16 ± 2 100 ± 15 \* 20 ± 2 \* 95 ± 12 \* 21 ± 2 \* 100 ± 13 20 ± 2 Acyclovir NA NA 2.1 ± 0.3 ˃950 0.1 ± 0.01 ˃20.000 **HSV-2** FeF 95 ± 10 21 ± 2 45 ± 6 44 ± 5 50 ± 7 40 ± 5 65 ± 8 31 ± 4 FeHMP 110 ± 13 18 ± 2 77 ± 8 \* 26 ± 3 \* 80 ± 10 \* 25 ± 3 \* 85 ± 11 24 ± 3 Acyclovir NA NA 1.6 ± 0.2 ˃1200 0.1 ± 0.01 ˃20.000 **ECHO-1** FeF 710 ± 80 2.8 ± 0.2 105 ± 12 19 ± 2 90 ± 12 22 ± 2 110 ± 13 18 ± 1 FeHMP 580 ± 65 3.4 ± 0.2 83 ± 10 24 ± 2 85 ± 11 24 ± 2 93 ± 11 21 ± 2 Ribavirin NA NA ˃500 ≤1.5 ˃500 ≤1.5 **HIV-1** FeF 25 ± 3 8 ± 1 25 ± 3 8 ± 1 25 ± 3 8 ± 1 25 ± 3 8 ± 1 FeHMP 25 ± 3 8 ± 1 25 ± 3 8 ± 1 50 ± 6 4 ± 0.5 25 ± 3 8 ± 1 Retrovir 5 ± 0.6 10 ± 1 NA NA 1.3 ± 0.2 38 ± 4 2.5 ± 0.3 20 ± 3 Epivir 1.5 ± 0.2 33 ± 4 NA NA 1.5 ± 0.2 33 ± 4 1.5 ± 0.2 33 ± 4 Note: Values represent the means ± standard deviations of three or more independent experiments; FeF, native fucoidan from brown alga *F. evanescens*; FeHMP, modified fucoidan fragment; Acyclovir, Ribavirin, Retrovir and Epivir were used as positive controls; IC~50~, concentration that inhibited 50% of virus's replication; SI, selectivity index (CC~50~/IC~50~); NA, no activity; \* Statistically significant differences between tested compounds (FeF and FeHMP) (*p* ≤ 0.05). marinedrugs-18-00224-t002_Table 2 ###### Effect of fucoidans treatment on the protection of mice from intravaginal herpes simplex virus 2 (HSV-2) infection. Groups Survivors/Total Protection Index (%) Average Survival Time (d) ------------------------ ----------------- ---------------------- --------------------------- FeF (5 mg/kg/d) 2/10 11.1 11.0 ± 2.1 FeF (10 mg/kg/d) 5/10 \* 44.4 14.1 ± 1.4 \* FeHMP (5 mg/kg/d) 3/10 22.2 13.4 ± 1.8 FeHMP (10 mg/kg/d) 6/10 \* 55.6 16.0 ± 1.3 \* Acyclovir (50 mg/kg/d) 10/10 \* 100 21.0 ± 0.5 \* Virus group 1/10 \- 9.7 ± 2.6 Normal group 10/10 100 ˃21 Note: Protection index: (Mv − Mc) / Mv ×100%, where Mv and Mc are mortality (%) in the virus and compounds-treated groups, respectively. The tested drugs were administered intraperitoneally for 5 days postinfection. \* Doses of tested fucoidans (FeF and FeHMP) were comparable with doses of fucoidans from other brown algae used in experimental viral infections \[[@B23-marinedrugs-18-00224],[@B24-marinedrugs-18-00224]\]. Statistically significant differences between values in the compounds-treated group and virus group (*p* ≤ 0.05). marinedrugs-18-00224-t003_Table 3 ###### The structure of native (FeF) and modified (FeHMP) fucoidans from *Fucus evanescens* \[[@B19-marinedrugs-18-00224]\]. Polysaccharide Molecular Weight, kDa OSO~3~^−^ % Monosaccharide Composition ---------------- ----------------------- ------------- ---------------------------- ----- --- FeF 160 28 0.9 0.1 0 FeHMP 50.8 40 1.0 0 0 Antivirals used as positive controls: - Acyclovir^®^, freeze-dried powder for injections (GlaxoSmithKline Manufacturing, Italy) used for herpes virus infections; - Ribavirin^®^ (Sigma-Aldrich, USA) used for enterovirus infections; and - Retrovir^®^ (ViiV Healthcare UK Limited, UK), Epivir^®^ (ViiV Healthcare, UK) used for HIV infection.

Slow pace of digital health governance in Europe {#Sec1} ================================================ In October of 2018, the Department of Health and Social Care in the United Kingdom published a policy paper on the future of digital data and technology, which highlighted the widening gap between the implementation of digital healthcare and the original aspirations for the service. It clearly stated, "the gap between where we are and where we want to be is only getting bigger" \[[@CR1]\]. In 2019, the Topol Review (prepared by the NHS) reached similar conclusions and highlighted the divisive nature of Artificial Intelligence (AI) within healthcare \[[@CR2]\]. The 2019 EU guidelines on the ethics of AI raised concerns about the regulation of digital governance among other trepidations \[[@CR3]\]. It is widely accepted that digital healthcare and AI will forever change the ways in which healthcare is delivered to patients, as well as the relationship between doctors and patients. There are multiple objectives of digital health services: achieve cost efficiency gains, improve the effectiveness of medical treatments, boost the early diagnosis of illnesses, enhance surgeries with robotic systems, increase positive healthcare outcomes, and help good governance of the public sector by enhancing transparency and accountability \[[@CR4]\]. This article aims to take stock of the opportunities offered by digital technologies within the healthcare sector against the backdrop of structural bureaucratic rigidities and organizational resistance to embracing innovation. Some have argued that AI has triggered a process of "innovative disruption", because it will radically transform the healthcare model from a business-to-business model to a business-to-consumer model \[[@CR5]\]. In these new systems, doctors will no longer be the bosses. In a recent article, Richard Saltman discussed the impact of the digital revolution (referring mainly to computer technology) on the tax-funded healthcare systems of Europe. His primary focus was on the future challenges and organizational constraints that will create obstacles to the successful implementation of digital health governance \[[@CR6]\]. He particularly emphasized the factors that determine the slow progress of organizational innovation when implementing digital technologies and new integrated delivery arrangements within European tax-funded healthcare systems. He criticized the organizational stasis of public sector organizations, which he labelled as a structural blockage, citing resistance from public hospitals to organizational changes, internal reforms, and innovation. In regards to AI, the fear is that robotic systems will replace doctors when it comes to performing healthcare diagnoses. For instance, retinal scans and targeted radiotherapy use AI systems to carry out diagnoses with increasing confidence. Doctors will need to be trained in data science and upgrade their technology skills to be able to "control" these systems. However, the fear of job losses is exaggerated. In fact, AI is likely to create new jobs, for instance, in the field of machine learning and data science analysis. The article by Richard Saltman pointed to the need for more dynamic organizational change within public sector hospitals. Saltman also offered an insightful discussion of some of the best practices of service delivery in Europe and how they succeeded in overcoming the tendencies towards inertia and ultimately embraced the digital transformation of service delivery. We need to take a step back and further elaborate on what "bureaucratic inertia" means in the context of European healthcare services. The issue of change in a bureaucratic system is often framed in dysfunctional terms (see, for instance, the term "disruption"), emphasising the traits of public administration that lead inevitably to a pathological state of institutional inefficiency and malaise, a so-called "vicious cycle" in Crozier's terms \[[@CR7]\]. Saltman's article poses the important question of distinguishing between "dynamic" and "stable" healthcare systems, attributing to the former the capacity to innovate and to the latter a sort of pathological immobilisme. European tax funded systems belong to the latter category. In many European healthcare systems, administrative and managerial change does not happen in big and radical steps, and is only feasible through small, incremental reforms that allow for institutional learning and effective implementation \[[@CR8]\]. Incrementalism is not very different from inertial change, which refers to a uniform rate of motion done in a predictable series of increments \[[@CR9]\]. This definition of inertia emphasises the cumulative effect of incremental changes. Inertia is conceptualised as a "force in motion", which Richard Rose describes as a "juggernaut". Despite the legacy of past decisions, incremental changes made by governments can cumulatively produce big changes in the end \[[@CR10]\]. Digital healthcare is likely to be adopted in incremental steps within tax-funded European systems. On one hand, technology keeps an eye on patients and is often in the background of medical data and patient experiences. On the other hand, AI is technically less developed and less autonomous from human intelligence than we assume. Further technological advances are needed to develop "strong" systems of AI \[[@CR11]\]. Furthermore, digital healthcare is increasingly subject to regulatory governance by national and international bodies, while ethical concerns are, for good or for worse, significantly slowing down the pace of AI and other technologies. The path to digital innovation: from information technology (IT) to patients' empowerment through artificial intelligence (AI) {#Sec2} ============================================================================================================================== Technology, particularly digital technology, is increasingly playing a key role in leveraging future health policies. In the last decade, national governments in Europe have designed and implemented policies aimed at restructuring the delivery of healthcare services by using new digital health infrastructures and services \[[@CR12]\]. Electronic health records and patient data have been at the core of this digitization process. Artificial Intelligence (AI) requires medical health data, which is the most valuable asset for AI producers and companies. Healthcare data is not merely a preference for marketing purposes, but refers to symptoms, treatments, and patient experiences. In fact, this term refers to highly sensitive patient data protected by national legal systems and regulations. In the earlier stages on the path towards improved digital healthcare, governments used ICT (Information and Communication Technologies) to manage internal administrative processes. The main purpose of digitization was cost efficiency and improved effectiveness of procedures. The administration of health units has been using ICT for reimbursement procedures and e-procurement, for instance. However, digitization in the early days did not entail the participation of users and patients in the process of service delivery. More recently, e-health and e-government policies have substantially redesigned the role of communication with patients and their relatives \[[@CR13]\]. These policies have allowed for the development of more user-friendly, personalized services that rely on digital technologies, such as the use of smartphone applications to monitor patients' health conditions. On the path towards digital governance, the use of digital technologies is now moving beyond the back office of administrative processes and health data collection, and being consistently applied to direct service delivery and direct citizen engagement. For instance, NHS Highland recently launched a new programme that uses virtual reality to help patients prepare for upcoming MRI scans in Scotland. This virtual device simulates the MRI experience for patients, thus helping them be more comfortable during the procedure and making the MRI more bearable. This is just one example of how AI improves the experience of patients. Our future challenge is not merely to introduce ICT to public organizations, but also to reframe the traditional relationship between professionals and patients through the use of data sharing and digital technologies. Thus, the changes affecting healthcare systems will be transformative and not just marginal. The doctor-patient relationship will change, which has given rise to legitimate concerns. The empowerment of patients through digital platforms has been adopted in order to promote self-management of diseases and improve patients' engagement in the decision making process \[[@CR14]\]. Digital platforms allow patients and their families to have access to a wealth of information about their conditions and to connect more easily with professionals. The use of digital platforms provides the tools for data-based personalization of medicine and significantly increases the focus on the customer and their specific needs \[[@CR15]\]. However, as AI takes over low-level work, such as monitoring heart rate or sleeping patterns, there are concerns about the loss of human touch, which is so integral to the patient-doctor relationship. It is possible that AI will serve as a gatekeeper and patients will interact with a machine for low-level and routine practices. Despite the possible costs of losing some human touch for these services, AI will free up more time for doctors to treat patients who need specialized care. For instance, Ichilov Hospital in Tel Aviv, one of the largest hospitals in Israel, launched a medical trial in September 2019 to remotely monitor patients' conditions using sensors and cameras. The trial used an early detection instrument for patients' deteriorating conditions, which reduced the nurses' workload significantly. Ichilov Hospital collaborated with two Israeli AI companies, which specialize in monitoring systems (Innovision and BioBeat). In 2019, the Israeli government launched a massive reform plan to unleash AI technologies in healthcare organizations, making use of the country's advances in cyber security technology. In the field of oncology, digital healthcare has increased the confidence of early stage diagnosis. For instance, it provides quicker, better, and more cost-efficient image analysis of prostate biopsies. Precision medicine and new technologies promise to support doctors in matching the immunoprofiles of patients using targeted drugs for individual patients \[[@CR16]\]. Robotic systems are used effectively and widely for complex surgeries. Another example of a widely used digital technology is IBM's Watson, an expert system developed by the technological giant IBM. Watson is used as an instrument for doctors to identify individualized, evidence-based treatments. It collects a huge amount of medical data and presents doctors with treatment options, then recommends personalized drugs. The fees range from \$200 to \$1000 USD per patient. Hybrids and accountability {#Sec3} ========================== Saltman's article does not discuss the benefits of digital healthcare in terms of precision and personalized medicine. Instead, the piece is more concerned with the organizational impact of information technologies on public sector providers and public hospitals. In his article, Saltman advocates greater structural diversity within European NHS systems and better links with intermediate non-public providers and non-governmental actors. This is a highly contested proposal for tax-funded Beveridge healthcare systems, for which public hospitals are central elements of accountability for public health and wellbeing \[[@CR17]\]. Their institutional objectives depend on equity and universal access-oriented efforts. The majority of hospitals are still publicly owned and managed, despite a wide range of managerial reforms and leadership changes over the last few decades \[[@CR18]\]. Major organizational changes began in the late 1980s in Sweden and England, due to growing political pressure on public authorities to restructure the hierarchal governance models of healthcare providers \[[@CR19]\]. According to Richards and Smith, new forms of hospital governance pushed the state to focus on directing and regulating rather than managing hospitals \[[@CR20]\]. This meant, for instance, the establishment of new public-private partnerships between different stakeholders. In these arrangements, the private sector is involved in delivering public goods and services and serves as a vehicle for coordinating with non-governmental actors in order to undertake integrated efforts to meet healthcare needs \[[@CR21]\]. The post-2008 economic crisis in Europe reinforced the shifting role of the state from rowing to steering \[[@CR22]\]. As a result, we now have a diverse provider environment within the European NHS systems, as well as hybrid forms of accountability \[[@CR23]\] and performance measurement systems \[[@CR24]\]. AI producers and developers are for-profit businesses in many countries, which provide products and services to public sector organizations. Traditional mechanisms of accountability need to adjust to the new AI environment, which is characterized by multiple stakeholders sharing medical data, research, and systems. The new digital environment for healthcare services requires collaborative infrastructures and partnerships across different stakeholders in the public, private, and not-for-profit sectors. Hybrid forms of management have emerged and will only continue to grow \[[@CR25]\]. For instance, data sharing agreements between different providers are central to AI systems. Hybrid accountability describes the integration of accountability arrangements between and across public, market, and social accountability regimes. Hybrid forms of accountability layer and mix actors, logics, norms, and mechanisms from public, market, and social accountability regimes by concurrently applying two or more of these differing regimes to the same situation \[[@CR26]\]. Ethical concerns and regulations {#Sec4} ================================ We need clearer guidance about the accountability of AI, in order to foster an effective regulatory environment that does not stifle innovation and govern hybrid forms of arrangements without eschewing human agency and oversight. This is probably the most critical concern associated with digital healthcare in the future. Regulation is indeed necessary to protect patients' safety, privacy, and fundament rights and freedoms. As with the pharmaceutical industry, AI developers should be regulated by licensing and critical appraisal to remove unsafe systems and procedures. A delicate balance needs to be found between fostering innovation and regulating the environment to implement AI in healthcare \[[@CR27]\]. On the one hand, digital healthcare will disrupt the existing delivery of healthcare services, enhancing a redesign of internal organizational systems and procedures for public accountability and transparency. For instance, medical data storage associated with machine learning is subject to new legal frameworks and risk assessment obligations, such as the data protection impact assessment \[[@CR28]\]. Hospitals and health organizations must revise internal procedures and design new employee training programs to comply with European data privacy regulations. Machine learning is particularly opaque, thus it is highly difficult to ensure transparency about its use and deployment. The EU ethical guidelines recommend that AI systems should be identifiable and humans need to be aware of the content of their interactions with machines. Furthermore, patients need to be able to understand how AI systems work and how these technologies can be applied to their conditions. Transparency and accountability requirements are changing the governance systems of health organizations. NHS healthcare systems in Europe have adopted and deployed artificial intelligence (AI), machine learning, big data analytics, wearables, and other new technologies to improve clinical accuracy, patient safety, precision medicine, and personalized medicine. Yet, government regulations may risk slowing down rapid developments over concerns for accountability and the protection of fundamental human rights. The EU is a frontrunner in this policy area, as well as the ethical guidelines regarding AI. The European Parliament has expressed many concerns about the use of patients' data and the security of data generated through AI. In 2017, they initiated actions to legislate and regulate the new AI environment. Digital healthcare certainly poses new challenges to accountability arrangements that had previously been based on a (human) principal and (human) agents. The old arrangements of accountability will certainly be challenged by AI, given the ethical and philosophical implications associated with clinical choices and the ranking of options, not made by humans. The EU guidelines recommend that clinical decisions rest in the hands of doctors and that machines should be subject to human oversight. Some scholars argue that a new concept of information accountability is necessary to manage these processes \[[@CR29]\]. For instance, what happens when IBM's Watson for Oncology (or other expert systems described earlier in this article) is simply wrong? Legal frameworks associated with medical malpractice apply to people (the medical professional or organization), but not machines. Existing legal frameworks in many countries are not prepared for sanctioning errors by machines that make mistakes whilst establishing a treatment ranking or identifying drug prescriptions for individual patients. Personhood is the legal basis for liability concerning malpractice in both civil and tort law. What happens when a machine makes an error that has detrimental consequences for individual patients? Clearly a machine cannot be in full control of itself. Thus, it is essential that regulation provides for human oversight of these technologies and that doctors remain in charge of clinical decision-making. The EU has promoted a "human centric-approach" to AI that centrally focuses on the respect of fundamental rights, including those set out in the Fundamental Rights of the European Union. In April of 2019, the EU Commission drew up non-binding guidelines for AI design, development, deployment, and implementation. The thrust of this ethical approach is to protect human dignity and agency throughout the use of these new technological systems by ensuring human control when interacting with AI. Other ethical principles set out in the EU Commissions' guidelines are privacy and data protection to ensure that citizens are always in full control of their own data. Thus, AI systems in medical care should be designed with techniques, such as data encryption and anonymisation to protect patient data. Although the EU is a global frontrunner in setting principles for the AI community (including for giants such a Google, IBM, and Microsoft), the EU ethics guidelines thus far are not legally binding and the regulatory oversights that support their implementation are weak. A 2019 study by the European Parliament recommends the creation of a regulatory body for algorithmic decision-making, which will handle specific liability and certification regimes \[[@CR30]\]. To conclude, tax-funded European healthcare systems operate in a provider environment that is much more organizationally diverse than it was 20 years ago. Partnership arrangements with non-governmental actors and civil society are today an established governance model in many countries. Hybrid organizational forms that mix private and public delivery systems are the result of waves of internal organizational restructuring \[[@CR31]\]. AI will consolidate the process of hybridization in healthcare governance. National governments in Europe and the European Commission have embraced digital healthcare with enthusiasm, followed by legislative impetus and regulatory frameworks that improve patients' dignity, transparency of governance processes, and enhance patient engagement. The pace of change may still appear slow to observers in other parts of the world, but one should not be misguided by this slow rate of progress. Major change is in the works, even though it is advancing slowly. The pace of change is a result of the orientation of NHS healthcare systems towards values of equity, universal access, and organizational values that are fundamentally different from those of privatized medicine. Changes are incremental because of the desire to align the population's health needs with equity, efficiency, and safety. Digital governance is widely accepted as the way to achieve these objectives. The pace of change will remain uncertain because new hybrid accountability regimes will need to be designed and AI will need to be regulated at the national and European level. The outcome will be a transformative change that will integrate digital technologies into public sector modernization. AI has great potential to empower patients, while improving their health, personalized experiences, and individual wellbeing. **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The author read and approved the final manuscript. The author declares that she has no competing interests.

原发免疫性血小板减少症（ITP）是最为常见的良性出血性疾病。ITP患者血小板数量减少与出血严重程度并不完全相关。虽然血小板计数低于20×10^9^/L时患者更可能出现严重出血，但有些血小板计数高于20×10^9^/L的患者仍有出血表现，而有些血小板计数极低的患者却无出血表现。因此，ITP患者出血程度不能单独用血小板计数解释。我们前期研究结果显示血小板聚集功能与ITP患者出血程度相关[@b1]，我们推测血小板活化状态与ITP患者的出血程度相关。本研究中，我们采用流式细胞术检测ITP患者血小板膜糖蛋白（GP）Ⅰb、GPⅡb/Ⅲa和P选择素表达，探讨血小板活化和出血程度之间的关系。 病例与方法 {#s1} ========== 1．研究对象：2014年12月至2015年3月于中国医学科学院血液病医院门诊或住院治疗的ITP患者43例纳入研究。男21例，女22例；中位年龄44（18\~68）岁，中位PLT 15（2\~99）×10^9^/L。参照文献[@b2]标准进行诊断。以20名健康志愿者作为正常对照组，男8名、女12名，中位年龄41（22\~65）岁，中位PLT 183（118\~238）×10^9^/L。本研究经医院伦理委员会批准，患者均知情同意。 2．出血评分方法：运用2013年ITP国际工作组（IWG）制定的ITP特异性出血评价工具（ITP-BAT）1.0版[@b3]对患者进行出血评分，出血症状包括皮肤（S）、黏膜（M）、器官（O）三项指标，得出SMO评分值，用SMO评分值界定不同程度的临床出血情况，严重出血（影响生命体征）：O≥3；大量出血：M≥3；中度出血：S=3、M=2、O=2；轻度出血：S\<3、M=1、O=1；无症状：S=0、M=0、O=0。由2名血液科医师完成所有ITP患者的出血评分。 3．标本采集：抽取ITP患者肘静脉血2\~4 ml，取2 ml全血加入EDTA抗凝管，用于未成熟血小板比例（IPF）测定；取全血1 ml加入枸橼酸钠抗凝管，用于流式细胞术测定GPⅠb、GPⅡb/Ⅲa、P选择素。正常对照组同法采集外周静脉血标本。 4．流式细胞术检测血小板GPⅠb、GPⅡb/Ⅲa、P选择素表达水平：取1 ml枸橼酸钠抗凝血，轻轻混匀后静置，离心10 min（150×*g*）取富含血小板血浆，Falcon管内血小板计数用PBS调整至1×10^7^/ml，加入20.0 µmol/L二磷酸腺苷（ADP），立即旋涡1 s混匀，室温24 °C，激活时间5 min[@b4]。CD61-PerCP（美国Biolegend公司产品）和侧向散射角双参数确定血小板群，实验管加入CD42b-APC（美国Biolegend公司产品）、CD62p-PE（美国Biolegend公司产品）、PAC-1-FITC（美国BD Biosciences公司产品），对照管加入同型对照IgG1抗体（美国Biolegend公司产品）和GPⅡb/Ⅲa受体拮抗剂（RGDS）（美国BD Biosciences公司产品），4 °C避光孵育45 min，PBS洗涤2次，最后加300 µl PBS重悬，上流式细胞仪检测。GPⅠb、GPⅡb/Ⅲa、P选择素的表达分别以CD42b、PAC-1、CD62p的平均荧光强度表示。 5．IPF测定：收集EDTA抗凝静脉血2 ml，在6 h内采用Sysmex XE-5000血液分析仪检测IPF（自动进行模式）。网织血小板中的RNA在网织红细胞通道内被核酸荧光染料染色，再用XE-pro IPF master软件根据荧光强度和前向角散射光强度在网织红细胞散点图上较高荧光强度的区域圈出网织血小板，通过分析得到IPF。 6．统计学处理：采用SPSS19.0进行数据分析。符合正态分布计量资料的组间比较采用*t*检验及配对*t*检验。各指标与出血分级之间的关系采用多元有序回归分析。采用二元回归法分析各指标与中度以上出血风险的相关性，计量资料间相关性采用Pearson线性相关分析。*P*\<0.05为差异有统计学意义。 结果 {#s2} ==== 1．ITP患者血小板活化指标：ITP患者激活前血小板GPⅠb表达低于正常对照组（*t*=2.417，*P*=0.020），GPⅡb/Ⅲa、P选择素表达高于正常对照组（*t*=4.130，*P*\<0.001；*t*=2.070，*P*=0.043）；ITP患者和正常对照组ADP激活后血小板P选择素、GPⅡb/Ⅲa表达均高于激活前（*P*\<0.05），GPⅠb表达均低于激活前（*P*\<0.05）；ADP激活后ITP患者血小板GPⅡb/Ⅲa表达高于正常对照组（*t*=5.256，*P*\<0.001）,而P选择素和GPⅠb表达低于正常对照组（*t*=3.901，*P*\<0.001；*t*=2.070，*P*=0.043）。ADP激活前后血小板GPⅡb/Ⅲa和P选择素表达比值在ITP患者和正常对照组间差异有统计学意义\[（2.58±1.21）%对（7.75±3.57）%，*t*=7.297，*P*\<0.001；（2.97±2.19）%对（10.12±9.37）%，*t*=4.084，*P*\<0.001\]；两组间GPⅠb表达比值差异无统计学意义\[（0.79±0.59）%对（0.78±0.39）%，*t*=0.005，*P*=0.995\]。ITP患者ADP激活前后血小板GPⅡb/Ⅲa、P选择素、GPⅠb表达比值和激活前表达呈负相关（*P*=0.001，*r*=−0.540；*P*=0.002，*r*=−0.524；*P*=0.010，*r*=−0.434）。详见[表1](#t01){ref-type="table"}。 ###### ITP患者ADP激活前后血小板活化指标GPⅠb、GPⅡb/Ⅲa和P选择素表达（平均荧光强度，*x*±*s*） 组别 例数 GPⅠb GPⅡb/Ⅲa P选择素 ---------- ------ ----------------- ----------------- --------- -------------- --------------- --------- --------------- ----------------- --------- ITP 43 240.11±24.93 149.06±19.14 0.022 65.69±10.73 133.96±12.17 \<0.001 15.43±1.41 37.09±3.94 \<0.001 正常对照 20 295.11±22.15^a^ 205.73±21.00^b^ 0.044 7.16±0.99^c^ 39.67±4.99^d^ \<0.001 12.55±1.03^e^ 109.77±23.66^f^ \<0.001 注：ITP：原发免疫性血小板减少症；ADP：二磷酸腺苷；GP：血小板膜糖蛋白。与ITP组比较，^a^*t*= 2.417，^a^*P*=0.020，^b^*t*= 2.070，^b^*P*=0.043，^c^*t*= 4.130，^c^*P*\<0.001，^d^*t*= 5.256，^d^*P*\<0.001，^e^*t*= 2.070，^e^*P*=0.043，^f^*t*= 3.901，^f^*P*\<0.001 2．ITP患者血小板活化指标与血小板计数、IPF的相关性：ITP患者中位IPF为21.1%（4.4%\~44.7%），中位PLT为15（2\~99）×10^9^/L，血小板计数与IPF呈负相关（*r*=−0.475，*P*=0.011）。ADP激活后GPⅠb表达、ADP激活前P选择素表达、ADP激活后P选择素表达、ADP激活前后GPⅠb表达比值、P选择素表达比值与血小板计数显著相关（*P*\<0.05），ADP激活后P选择素表达、ADP激活前后P选择素表达比值与IPF显著相关（*P*\<0.05）。详见[表2](#t02){ref-type="table"}。 ###### ITP患者血小板GPⅠb、GPⅡb/Ⅲa、P选择素与血小板计数、未成熟血小板比例的相关性分析 血小板活化指标 血小板计数 未成熟血小板比例 ---------------------------- ------------ ------------------ ------- -------- ADP激活前GPⅠb表达 0.761 0.057 0.354 −0.182 ADP激活后GPⅠb表达 0.024 −0.404 0.468 0.143 ADP激活前后GPⅠb表达比值 0.017 −0.424 0.065 0.354 ADP激活前GPⅡb/Ⅲa表达 0.174 −0.255 0.207 0.151 ADP激活后GPⅡb/Ⅲa表达 0.863 −0.032 0.547 0.119 ADP激活前后GPⅡb/Ⅲa表达比值 0.324 0.018 0.172 −0.165 ADP激活前P选择素表达 0.045 −0.378 0.846 0.038 ADP激活后P选择素表达 0.002 0.536 0.011 −0.472 ADP激活前后P选择素表达比值 \<0.001 0.639 0.037 −0.396 注：ITP：原发免疫性血小板减少症；ADP：二磷酸腺苷；GP：血小板膜糖蛋白 3．ITP患者血小板活化指标与出血程度的相关性：43例患者根据ITP-BAT评分结果进行出血分级：无症状12例，轻度出血16例，中度出血9例，大量及严重出血6例。血小板计数、IPF在不同出血程度之间差异均有统计学意义（*P*=0.001、*P*=0.020）。 ADP激活前血小板P选择素表达在不同出血程度患者之间差异有统计学意义（*P*=0.040）（[表3](#t03){ref-type="table"}）。中度以上出血患者血小板P选择素表达高于轻度出血和无症状患者（20.20±8.54对13.16±6.57，*t*=2.526，*P*=0.017）。ADP激活后血小板P选择素、GPⅠb表达在不同出血程度患者之间差异有统计学意义（*P*=0.006、*P*=0.001）（[表3](#t03){ref-type="table"}）。中度以上出血患者ADP激活后血小板P选择素表达低于轻度出血和无症状患者（25.6±9.27对42.56±24.27，*t*=2.802，*P*=0.009），而中度以上出血患者GPⅠb表达高于轻度出血和无症状患者（121.73±61.25对46.19±23.42，*t*=3.771，*P*=0.003）。 ###### ADP激活前后ITP患者血小板GPⅠb、GPⅡb/Ⅲa、P选择素表达与出血程度的相关性（平均荧光强度，*x*±*s*） 组别 例数 GPⅠb GPⅡb/Ⅲa P选择素 ---------------- ------ -------------- -------------- -------------- -------------- ------------ ------------ 无症状 12 129.36±24.94 41.97±6.83 46.00±7.08 129.36±24.94 12.94±2.54 54.52±9.27 轻度出血 16 116.39±20.46 49.35±7.45 45.83±5.72 116.39±20.46 13.33±1.73 33.59±5.02 中度出血 9 126.38±28.26 127.10±18.94 121.55±39.43 126.38±28.26 19.71±3.85 25.43±3.54 大量及严重出血 6 92.03±35.04 120.86±46.00 109.20±54.80 90.03±35.04 21.33±2.18 26.66±6.36 *P*值 0.611 0.001 0.082 0.109 0.040 0.006 注：ITP：原发免疫性血小板减少症；ADP：二磷酸腺苷；GP：血小板膜糖蛋白 4．ITP患者ADP激活前后血小板活化指标比值与出血程度的相关性：ADP激活前后ITP患者血小板P选择素、GPⅡb/Ⅲa、GPⅠb表达比值在不同出血程度患者之间差异有统计学意义（*P*=0.003、*P*=0.001、*P*=0.003）（[表4](#t04){ref-type="table"}）。中度以上出血患者ADP激活前后P选择素、GPⅡb/Ⅲa表达比值低于轻度出血和无症状患者\[（1.51±1.01）%对（3.66±2.27）%，*t*=3.634，*P*=0.003；（1.36±0.62）%对（3.16±0.97）%，*t*=5.332，*P*\<0.001）\]。中度以上出血患者ADP激活前后GPⅠb表达比值高于轻度出血和无症状患者\[（1.34±0.68）%对（0.53±0.29）%，*t*=3.568，*P*=0.006\]。 ###### ADP激活前后ITP患者血小板GPⅠb、Ⅱb/Ⅲa及P选择素表达（平均荧光强度）比值与出血程度的相关性（%，*x*±*s*） 组别 例数 GPⅠb GPⅡb/Ⅲa P选择素 ---------------- ------ ----------- ----------- ----------- 无症状 12 0.44±0.24 3.40±1.04 4.74±2.43 轻度出血 16 0.59±0.32 2.98±0.92 2.84±1.85 中度出血 9 1.37±0.73 1.27±0.60 1.62±1.19 大量及严重出血 6 1.25±0.70 1.56±0.72 1.22±0.37 *P*值 0.003 0.001 0.003 注：ITP：原发免疫性血小板减少症；ADP：二磷酸腺苷；GP：血小板膜糖蛋白 5．ITP患者血小板活化指标与中度以上出血风险的相关性：与出血程度相关的因素包括：ADP激活后GPⅠb表达，ADP激活前P选择素表达，ADP激活后P选择素表达，ADP激活前后GPⅠb、P选择素、GPⅡb/Ⅲa表达比值。评估这些指标与中度以上出血风险的相关性，结果显示：ADP激活前后GPⅠb表达比值是ITP患者中度以上出血的正性影响因素\[*OR*=3.05（95% *CI* 1.29\~7.24），*P*=0.011\]；ADP激活前后P选择素、GPⅡb/Ⅲa表达比值是ITP患者中度以上出血的负性影响因素\[*OR*=0.32（95% *CI* 0.12\~0.85），*P*=0.023；*OR*=0.04（95% *CI* 0\~0.39）, *P*=0.006\]。 讨论 {#s3} ==== ITP患者出血的严重程度与血小板计数减少并非呈正相关，有的患者血小板计数尽管很低但不出血或出血并不严重。我们最近的临床研究发现ITP患者外周血IPF和血小板聚集功能与患者出血程度相关，提示血小板功能在ITP患者出血过程中发挥一定的作用[@b1]。本研究进一步证实了血小板活化状态与ITP患者出血情况存在相关性。 血小板的黏附、聚集和活化功能在血小板止血过程中起重要作用，其功能发挥均依赖其表面膜糖蛋白功能而实现。血小板表面GPⅠb-Ⅸ、GPⅡb/Ⅲa、P选择素均参与了血小板的聚集、活化和黏附功能，反映了体内血小板的功能及活化状态。PAC-1（血小板糖蛋白Ⅱ b/Ⅲa受体）是血小板活化时GPⅡb/Ⅲa暴露的纤维蛋白原结合位点。P选择素在血小板活化时随着颗粒内容物的释放而与质膜融合，能持久地存在于活化血小板表面。GPⅠb在血小板活化后由膜表面进入开放管道系统或裂解，使其表达水平降低[@b5]。本研究结果显示ITP患者外周血中血小板膜表面P选择素和GPⅡb/Ⅲa表达增高、GPⅠb表达减低，提示ITP患者血小板内在活性增高，其促凝功能增强。Olsson等[@b6]研究显示，ITP患者自身抗体可以通过Fc依赖的形式与血小板上FcgRⅡ受体结合而诱导血小板活化信号或通过Fc非依赖形式促进血小板活化。另外，ITP出血患者血管内皮损伤后胶原暴露也可导致血小板活化[@b7]。我们进一步研究发现ITP患者和正常对照ADP激活后血小板P选择素和GPⅡb/Ⅲa表达增高、GPⅠb表达减低，但是ADP激活前后ITP患者血小板GPⅡb/Ⅲa和P选择素表达比值低于正常对照组；ITP患者ADP激活前后P选择素、GPⅡb/Ⅲa、GPⅠb表达比值均与激活前表达水平呈负相关，以上结果提示ITP患者血小板经体外ADP诱导后的活化反应明显减弱，激活前血小板P选择素、GPⅡb/Ⅲa表达越高或GPⅠb表达越低，经ADP诱导后的活化反应越弱，推测原因：①ITP患者血小板表面P选择素、GPⅡb/Ⅲa、GPⅠb表达异常，说明ITP患者体内血小板已经处于活化状态，且接近活化界值，对于体外ADP刺激活化的反应减弱[@b4]。②P选择素、GPⅡb/Ⅲa、GPⅠb表达异常的血小板其表面ADP受体和ADP下游凝血酶信号途径抑制，导致ADP激活反应减弱[@b8]。 本研究结果显示ITP患者激活前血小板P选择素表达与血小板计数呈负相关，与Panzer等[@b9]研究结果一致，提示血小板胞质内的颗粒膜糖蛋白P选择素暴露在膜表面作为一种代偿机制，以减少因血小板数目少而发生出血的风险。Psaila等[@b10]研究认为血小板对ADP的反应不受血小板计数的影响，但我们研究发现ADP激活后GPⅠb、P选择素表达与血小板计数相关，具体原因有待进一步研究。 ITP患者IPF增高，反映了ITP患者血小板生成能力代偿性增加，未成熟血小板在外周血中发挥止血作用[@b1]。血小板完全活化后会发生构象变化，导致膜表面出现GPⅠb-Ⅸ、GPⅡb/Ⅲa变化，我们研究发现IPF与GPⅡb/Ⅲa、GPⅠb表达无相关性，而与P选择素表达相关，说明成熟血小板较未成熟血小板活化后更容易发生构象改变，可能因为未成熟血小板代偿性通过增加膜表面P选择素来增强血小板功能，导致构象改变能力减弱[@b8]。 本研究中ITP患者激活前血小板P选择素表达和出血程度相关，而GPⅡb/Ⅲa、GPⅠb表达与出血程度无相关，与van Bladel等[@b11]研究结果一致。分析原因：①严重出血患者通过增加血小板聚集、活化和黏附功能发挥止血作用。完全活化的血小板膜表面表达P选择素和GPⅡb/Ⅲa，通常分布于血管壁周围以保持血管完整性，外周血循环较少。②血小板完全活化后发生的构象变化不是持续的，很快就恢复静息状态，而血小板可以继续吸附血浆中脱落的P选择素[@b8]。中度以上出血患者与轻度出血和无症状患者比较，ADP激活后血小板表面GPⅠb、P选择素表达及ADP激活前后P选择素、GPⅡb/Ⅲa、GPⅠb表达比值的差异均有统计学意义，提示出血较重患者体外ADP诱导的血小板活化反应减弱，可能的原因：①出血较重患者因出血量大和血小板严重减少导致体内血小板代偿性的过度活化，体外对激活剂反应减弱。②出血较重患者体内高滴度抗体和炎症因子干扰了血小板的功能[@b12]。我们通过回归分析发现ADP激活前后P选择素、GPⅠb、GPⅡb/Ⅲa表达比值与中度以上出血风险相关，P选择素、GPⅡb/Ⅲa表达比值越大严重出血风险减少，GPⅠb表达比值越大严重出血风险增加，进一步证实ITP患者血小板体外活化功能缺陷、血小板黏附和聚集代偿性增强能力的减弱与出血程度相关。 本研究结果初步表明，血小板活化指标能比较准确地反映ITP患者出血程度，可以作为干预和疗效观察指标。本研究病例数较少，血小板活化指标对于ITP患者出血风险的预测价值尚需进一步临床研究加以验证。 [^1]: 吕明恩（现在南京医科大学附属苏州市立医院血液科，215002）

###### **Appendix:** Supplementary tables A-D and flow chart (fig A) ###### Click here for additional data file. Introduction ============ Quality measures are metrics used to assess the quality of healthcare. The definitions, or specifications, for quality measures are commonly derived directly from clinical practice guidelines. One example is the Healthcare Effectiveness Data and Information Set (HEDIS) measure, which seeks to decrease underuse of screening for colorectal cancer in the United States. The current HEDIS measure is based on the screening recommendations of the US Preventive Services Task Force (USPSTF) guideline, which suggests routine screening in patients at average risk starting at the age of 50 and continuing through to the age of 75.[@ref1] The United Kingdom National Health Service (NHS) has proposed a similar age based measure for the bowel cancer screening program in the UK.[@ref2] Large integrated healthcare organizations in the US, such as the Veterans Affairs (VA) Health Care System, have also introduced age based quality measures as part of their accountability and improvement programs; indeed, VA has an active measurement and feedback system in place to encourage screening in age eligible individuals.[@ref3] Despite the use of these well established age based quality measures, experts recommend that the decision to screen be informed by estimated life expectancy rather than age alone.[@ref4] [@ref5] Age is a reliable proxy for life expectancy in young patients (long life expectancy) and very elderly patients (short life expectancy). But in most elderly patients eligible for screening, life expectancy varies considerably according not only to age but also health status. For instance, a 74 year old man who is in excellent health has a life expectancy of almost 15 years.[@ref4] Evidence suggests that screening for colorectal cancer is likely to be of benefit in such a patient and should therefore be encouraged.[@ref6] On the other hand, a 74 year old man who is in poor health has a life expectancy of less than five years.[@ref4] Evidence suggests that screening by any method is unlikely to be of benefit and might even be harmful in such a patient and should therefore be discouraged.[@ref6] [@ref7] In fact, the American College of Physicians (ACP) guideline for screening for colorectal cancer specifically recommends that screening should not be performed in individuals with a life expectancy of less than 10 years.[@ref5] Yet, as operationalized, the quality measure for screening for colorectal cancer encourages screening equally for both healthy and frail patients aged 74, potentially leading to overuse. Conversely, the specification of a maximum age beyond which performance of screening is not "counted" could promote underuse in healthy patients who are older than 75 but are still likely to benefit because of good life expectancy.[@ref8] [@ref9] As highlighted by Walter and colleagues in 2004, such an approach fails to sufficiently consider "who received the test, why it was performed, or whether the patient wanted it."[@ref10] We examined whether the upper age cut off of the quality measure for colorectal cancer screening is associated with overuse of screening among patients aged 70-75 in poor health and underuse in those aged over 75 in good health. We used data from the VA Health Care System, the largest integrated healthcare system in the US. Specifically, we compared screening between two groups of patients: older individuals who were within the target age range in the quality measure (aged 70-75) and older individuals who were beyond the age where screening "counts" toward the quality measure (age \>75). We hypothesized that patients who were within the target age range but were in poor health (low likelihood of benefit from screening and potential for net harm) would be more likely to undergo screening than patients who were beyond the target age range but were in good health (higher likelihood of benefit from screening). We sought to examine how quality measures that utilize "one size fits all" targets without specific regard to individual risk and benefit influence both appropriate and inappropriate use of care. Methods ======= VA Health Care System --------------------- We performed a retrospective cohort study using electronic data from the VA Health Care System. This system serves nearly nine million veterans (that is, people who previously served in the US military) across over 150 medical centers and nearly 1400 community based outpatient clinics. Though VA is under a single centralized leadership, its organizational structure allows for substantial regional and local autonomy in terms of budgetary decisions and approaches to delivery of care. The healthcare system has a robust centrally controlled quality measurement program, with a wide array of quality measures across various healthcare conditions.[@ref3] In parallel, medical centers have adopted computerized reminders that prompt providers to fulfill quality measures for services such as screening for colorectal cancer. Audits of performance are also reported to providers at many medical centers, further encouraging screening. Partly as a result of such approaches, VA is considered a leader in the delivery of high quality care in the US.[@ref11] VA physicians are salaried, but a small percentage of their full compensation comes in the form of "performance pay," which can reward physicians for productivity, performance on quality metrics such as screening for colorectal cancer, or other locally determined factors. Many older veterans are eligible for Medicare insurance, which covers non-VA outpatient care. Thus, a considerable proportion of veterans receive care through both VA and private physicians.[@ref12] Although guidelines suggest that screening is not universally required in older patients, neither VA, Medicare, nor other private insurers currently limit coverage of screening for colorectal cancer in older adults in the US. Study design and dataset ------------------------ VA electronic data are available from fiscal year 2000 to the present time, including data on demographics, utilization of primary care and screening for colorectal cancer, and comorbidity status. We identified veterans who were regular users of VA primary care and preventive care and were due for repeat screening in the fiscal year 2010 (allowing for sufficient time to determine whether screening was subsequently utilized). Patients were included if they met the following criteria: primary care visit in 2010 when the patient was due for repeat colorectal cancer screening (age ≥50, no fecal occult blood test (FOBT) in the 12 months before the 2010 visit, no sigmoidoscopy in the five years before the 2010 visit, and no colonoscopy in the 10 years before the 2010 visit); and at least one additional primary care visit in the 12-24 months before the 2010 visit (to focus on patients who regularly use VA primary care). A similar approach has been used in other studies to identify regular users of VA primary care.[@ref13] To identify individuals who underwent screening for colorectal cancer through VA, we chose to focus on individuals who had undergone a previous fecal occult blood test with negative results in the 12-24 months before the 2010 visit. We made this decision on the basis of pilot work at a single VA medical center where we had access to more detailed clinical data, which showed that many veterans who seemed to be due for screening based on VA electronic data had in fact undergone previous colonoscopy screening outside VA (that is, VA electronic data are insensitive for detecting historical exposure to colonoscopy). Inclusion of these patients who were not due for screening in our cohort would lead to substantial underestimation of screening utilization. We carried out a sensitivity analysis to examine the impact of this inclusion criterion related to fecal occult blood testing on our results. Fecal occult blood tests were identified by using Logical Observation Identifiers Names and Codes (LOINC), and sigmoidoscopies and colonoscopies were identified by using ICD-9 (international classification of disease, ninth revision), Current Procedural Terminology (CPT), and Level II Healthcare Common Procedure Coding System (HCPCS) codes (appendix table A). Patients were excluded if CPT or ICD-9 codes showed any of the following diagnoses between 2000 and the qualifying 2010 visit (appendix table B): previous colectomy; history of colon polyps; history of colorectal cancer; history of inflammatory bowel disease; or family history of colorectal cancer. These exclusion criteria were selected to ensure that the cohort comprised individuals who were at average (rather than increased) risk for colorectal cancer. Outcome and predictor variables ------------------------------- We determined whether any screening test (fecal occult blood test, sigmoidoscopy, or colonoscopy) was completed within 24 months after the qualifying 2010 visit (primary outcome variable). A 24 month follow-up window has been used in previous work on this topic as it allows patients ample time to schedule and complete a screening test.[@ref14] Secondary outcome variables included fecal occult blood test, sigmoidoscopy, or colonoscopy performed within 24 months specifically for screening (that is, tests that were potentially performed for diagnostic purposes were excluded); fecal occult blood test within 24 months; sigmoidoscopy within 24 months; and colonoscopy within 24 months. To identify screening sigmoidoscopies and colonoscopies, we utilized an algorithm previously developed by Fisher and colleagues, using VA administrative data.[@ref15] To identify screening fecal occult blood tests, we excluded patients in whom ICD-9 codes showed a diagnosis of anemia or bleeding in the three months before the fecal occult blood test (indicating that it might have been performed for diagnostic purposes) (appendix table B). Data were also extracted on several predictor variables, including demographic characteristics (age and sex), number of primary care visits in 2010, and health status. Race was not utilized as a covariate as it has been shown to have limited reliability in VA data.[@ref16] Health status was measured with the Deyo adaptation of the Charlson comorbidity index,[@ref17] based on inpatient and outpatient codes from the 24 months before and including the qualifying visit in 2010. The Charlson comorbidity index is a weighted score of 17 common comorbid conditions. Originally developed as a predictor of one year mortality after hospital discharge,[@ref18] subsequent work has shown that it can also be used as a predictor of long term mortality.[@ref19] In our study, the index was categorized according to the approach used by Walter and colleagues, in studies of screening in veterans (Charlson index 0, 1-3, or ≥4).[@ref7] [@ref14] In a 75 year old man, a Charlson index of 0 indicates good health, with an estimated life expectancy over 10 years.[@ref7] [@ref19] An individual with a Charlson index of 0 might have mild comorbid conditions, such as uncomplicated hypertension or osteoarthritis, which do not substantially affect life expectancy. Such patients are likely to benefit from screening for colorectal cancer.[@ref6] [@ref20]. On the other extreme, a 75 year old man with a Charlson index ≥4 has poor health, with an estimated life expectancy of less than five years.[@ref7] [@ref19] A patient with a Charlson index ≥4 is likely to have multiple comorbid conditions that impact life expectancy, such as diabetes with end organ damage, congestive heart failure, and emphysema. Such patients are unlikely to benefit from (and might even be harmed by) screening.[@ref6] [@ref7] [@ref20] Statistical analysis -------------------- Screening completion rates were stratified by age at the time of the qualifying 2010 visit (50-69, 70-75, or \>75), sex, Charlson comorbidity index, and number of primary care visits in 2010. We grouped 70-75 year olds together because previous research suggests that for individuals in this age range with a high burden of comorbidity life expectancy is similar and is less than the 10 years needed for benefit from screening.[@ref4] [@ref7] [@ref20] [@ref21] We carried out multivariable log binomial regression to identify independent predictors of screening utilization within 24 months of the qualifying 2010 visit and to calculate adjusted screening rates. Robust standard errors were used to adjust for clustering within facility. We also performed sensitivity analyses using our secondary outcome variables and broadening our inclusion criteria to include patients both with and without previous fecal occult blood testing. Analyses were performed with the Stata 12 statistical package (StataCorp, College Station, TX). The proportion of patients with missing data was negligible (date of birth was missing for 400 out of 4 756 477 patients). These 400 patients were excluded (appendix fig A). Results ======= We identified 399 067 veterans who met our inclusion and exclusion criteria (table 1[](#tbl1){ref-type="table"}). The mean age was 67.2 (SD 9.7) years, and most were men. The median Charlson comorbidity index was 1 (interquartile range 1-3). Participants made a median of three visits to primary care clinics in fiscal year 2010. ###### Baseline characteristics in study of role of quality measurement in inappropriate use of screening for colorectal cancer. Figures are numbers (percentage) of patients (n=399 067) Characteristic Participants Age 50-69 Age 70-75 Age \>75 ------------------------------------- -------------- -------------- ------------- ------------- Age (years): 50-69 247 900 (62) 247 900 --- --- 70-75 53 351 (13) --- 53 351 --- \>75 97 816 (25) --- --- 97 816 Men 385 781 (97) 236 832 (96) 52 499 (98) 96 450 (99) Women 13 286 (3) 11 068 (4) 852 (2) 1366 (1) Charlson comorbidity index: 0 127 968 (32) 89 259 (36) 14 621 (27) 24 088 (25) 1-3 213 814 (54) 129 210 (52) 29 916 (56) 54 688 (56) ≥4 57 285 (14) 29 431 (12) 8814 (17) 19 040 (19) Primary care visits in 2010: 1 49 191 (12) 28 086 (11) 6968 (13) 14 137 (14) 2 88 099 (22) 51 686 (21) 12 561 (24) 23 852 (24) 3 75 115 (19) 46 694 (19) 10 351 (19) 18 070 (18) ≥4 186 662 (47) 121 434 (49) 23 471 (44) 41 757 (43) Screening utilization\*: Colonoscopy, sigmoidoscopy, or FOBT 151 850 (38) 113 835 (46) 22 757 (43) 15 258 (16) Colonoscopy 37 237 (9) 31 430 (13) 3549 (7) 2258 (2) Sigmoidoscopy 2008 (1) 1614 (1) 201 (0.5) 193 (0.2) FOBT 129 103 (32) 94 725 (38) 20 806 (39) 13 572 (14) FOBT=fecal occult blood test. \*Categories are not mutually exclusive. About 38% (151 850/399 067) of identified veterans (age ≥50 with no upper age cut off) underwent a screening test for colorectal cancer (fecal occult blood test, sigmoidoscopy, or colonoscopy) in the 24 months after their qualifying 2010 primary care clinic visit (table 1[](#tbl1){ref-type="table"}). Screening was more common among younger and healthier veterans (and among those with a greater number of visits to a primary care clinic) (table 2[](#tbl2){ref-type="table"}). Screening was relatively stable from age 50 to age 75, but utilization declined abruptly after 75, matching the age cut off promoted by the existing VA quality measure (table 2 and figs 1-2[](#fig1){ref-type="fig"} [](#fig2){ref-type="fig"}). This finding was independent of comorbidity status, number of primary care visits, and demographic characteristics (fig 3[](#fig3){ref-type="fig"}). Of 53 346 veterans aged 70-75, 16.5% (8814/53 346) had a Charlson comorbidity index ≥4 (indicating poor health and shortened life expectancy). Though these veterans were unlikely to benefit from screening, 40% (3516/8814) underwent a screening test within 24 months. On the other hand, of 97 786 veterans aged \>75, 24.6% (24 088/97 786) had a Charlson comorbidity of 0 (indicating excellent health and good life expectancy). Though many of these veterans would likely benefit from screening, only 16.5% (3964/24 088) underwent a screening test within 24 months. In fact, a veteran who was aged 75 and unhealthy (Charlson index ≥4) was significantly more likely to undergo screening than a veteran who was aged 76 and healthy (Charlson index 0) (unadjusted relative risk 1.64, 95% confidence interval 1.36 to 1.97; fig 2[](#fig2){ref-type="fig"}). {#fig1} {#fig2} {#fig3} ###### Unadjusted and adjusted\* relative risks† and rates of screening for colorectal cancer by age and health status (n=399 067) Predictor variable Unadjusted Adjusted ------------------------------ --------------------- --------------------- ---------- --------------------- --------------------- Age (years): 50-69 1 (reference) 45.9 (45.7 to 46.1) 1 (reference) 45.6 (43.7 to 47.4) 70-75 0.93 (0.91 to 0.95) 42.7 (42.2 to 43.1) 0.94 (0.92 to 0.96) 42.9 (40.9 to 44.9) \>75 0.34 (0.30 to 0.39) 15.6 (15.4 to 15.8) 0.35 (0.30 to 0.40) 15.8 (13.6 to 18.1) Charlson comorbidity index‡: 0 1 (reference) 41.0 (40.7 to 41.3) 1 (reference) 38.7 (36.8 to 40.5) 1-3 0.92 (0.90 to 0.93) 37.5 (37.3 to 37.7) 0.94 (0.93 to 0.95) 36.0 (34.2 to 37.7) ≥4 0.82 (0.80 to 0.83) 33.4 (33.0 to 33.8) 0.87 (0.85 to 0.89) 33.1 (31.6 to 34.6) \*Adjusted for sex and number of primary care visits in 2010; unadjusted results reflect simple proportions within each age or comorbidity group. †P\<0.001 for all comparisons. ‡In a 75 year old man, Charlson index of 0 indicates life expectancy \>10 years, 1-3 indicates life expectancy 5-10 years, and ≥4 indicates life expectancy \<5 years. We performed several additional analyses to determine whether our inclusion and exclusion criteria (which were designed to identify regular users of VA primary care who seemed to be receiving colorectal cancer screening in VA) and primary outcome variable (any fecal occult blood test, sigmoidoscopy, or colonoscopy within 24 months of the qualifying 2010 visit) might have resulted in spurious associations between characteristics of the quality measure and screening utilization. First, we repeated our analysis for all veterans with regular use of VA primary care irrespective of prior fecal occult blood testing. Second, we repeated our analysis using secondary outcome variables, including any fecal occult blood testing within 24 months of the 2010 visit; any sigmoidoscopy within 24 months; any colonoscopy within 24 months; and fecal occult blood testing, sigmoidoscopy, or colonoscopy within 24 months for screening purposes only (that is, diagnostic fecal occult blood tests, sigmoidoscopies, and colonoscopies were excluded). These sensitivity analyses showed similar results to our primary analysis, with an abrupt decrease in screening in veterans after age 75 independent of Charlson comorbidity index and number of primary care visits (appendix tables C and D). Discussion ========== Principal findings and comparison with other studies ---------------------------------------------------- Quality improvement efforts that incorporate quality measures have been shown to be effective in promoting utilization of appropriate care.[@ref22] But as our work and that of others has shown, quality measures also have the potential to encourage utilization of inappropriate care.[@ref10] [@ref13] [@ref23] [@ref24] This is particularly true of measures that use a "one size fits all" or "treat to target" approach that is designed to encourage adherence to clinical practice guidelines (like the measure for screening for colorectal cancer).[@ref10] As our study suggests, such an approach can unintentionally discourage thoughtful clinical decision making. Fisher and colleagues found similar results in a 2005 study of veterans, reporting that quality improvement efforts to increase screening for colorectal cancer were accompanied by high rates of inappropriate use of screening fecal occult blood tests.[@ref25] Work in other clinical domains also supports our results. For example, Kerr and colleagues recently reported that high facility level performance on "treat to target" quality measures for veterans with diabetes was associated with overtreatment of both hypertension and hyperlipidemia.[@ref13] [@ref23] Efforts such as the *Too Much Medicine* and *Choosing Wisely*campaigns have sought to limit overuse by encouraging patients and providers to discuss the appropriateness of care when the value of a service might be low.[@ref26] But a patient and provider cannot have a meaningful discussion about whether to utilize a clinical service if they do not fully understand the value of care for that individual patient, nor if the provider is incentivized to order the service regardless of a patient's characteristics or preferences. Thus, refining quality measures to be more individualized and sensitive to preferences is critical if we want to promote care that is both more appropriate and patient centered.[@ref27] A patient centered approach to quality measurement, one that incorporates health status and individual preferences, is particularly needed in older patients in whom age alone is an inadequate proxy for life expectancy (such as those aged 70-80). This need will only grow as the population ages (yielding increasing numbers of healthy and unhealthy older patients) and as quality improvement initiatives that incorporate quality measures are more widely adopted. Recent work by Cho and colleagues has shown that use of administrative data to estimate life expectancy can be further tailored not only by health status but also by race and sex.[@ref21] Such an approach could be utilized by healthcare systems to better target screening to those individuals who are more likely to benefit. Strengths and limitations ------------------------- Our study has several strengths and limitations. First and foremost, we utilized data from the largest integrated healthcare system in the US and analyzed data on nearly 400 000 patients. Furthermore, the use of VA data allowed us to examine the impact of a longstanding, comprehensive, and highly effective implementation of quality measures on both underuse and overuse. Because performance of screening for colorectal cancer among patients aged 50-75 is now a standard measure in many healthcare systems, it is likely that similar patterns of underuse and overuse can be found in other high performing healthcare systems.[@ref28] The use of VA data also presents some limitations. Specifically, because veterans might undergo procedures outside VA, our data probably underestimate true screening rates. Indeed, previous work that utilized abstraction of medical records to assess receipt of screening for colorectal cancer has shown that about 80% of VA patients aged 50-75 are up to date for screening.[@ref29] We sought to resolve this limitation in our primary analysis by including only veterans who had undergone previous fecal occult blood test screening in VA, but even these veterans might undergo subsequent screening outside VA. Nonetheless, it is unlikely that veterans aged 75 differ substantially from those who are 76 in terms of use of screening outside VA (fig 2[](#fig2){ref-type="fig"}). It is also important to note that our study is retrospective in design, and the observed results could, at least partly, be caused by unmeasured confounding. For instance, we could not adjust for some variables that are known to impact screening utilization, such as socioeconomic status, race, and education, though these factors are also not likely to vary substantially among veterans aged 75 versus 76. Additionally, electronic data are subject to imprecision in coding, which could result in both overestimation or underestimation of screening utilization. Conclusions and policy implications ----------------------------------- Our study suggests that the way quality measures are specified can have a profound effect on screening utilization. In the case of colorectal cancer screening, the translation of the US Preventive Services Task Force guideline into a quality measure has resulted in high adherence to guideline recommendations. But screening does not seem to be utilized according to expected benefit across the population. Specifically, screening is likely being overused in sick patients aged 75 and younger and underused in healthy patients aged over 75. Of note, both the task force guideline and the VA Clinical Preventive Services Guidance Statement encourage an individualized approach to screening for colorectal cancer in patients aged 76-85, but this nuance had not explicitly been incorporated into the quality measure. In the past year, however, VA has modified the quality measure specifications to permit exceptions for veterans age ≤75 with shortened life expectancies or serious comorbid medical conditions. Though this is an important first step towards a more individualized approach, future patient centered quality measures will need to incorporate more explicit measures of clinical benefit and ultimately help guide clinical decision making. In this way, we can ensure that patients who are likely to benefit from a service receive it (regardless of age) and those who might incur more harm than benefit are spared unnecessary and costly tests and treatments. ### What is already known on this topic 1. Screening for colorectal cancer is an evidence based widely recommended preventive service that has traditionally been underused 2. Efforts to increase screening through quality measurement have been successful, and such efforts are now being implemented widely in the US and the UK 3. The implementation of "one size fits all" age based measures might have unintended effects on screening ### What this study adds 1. In the largest integrated healthcare system in the US, age based measures for screening for colorectal cancer (which encourage screening in patients aged 50-75) promoted overuse of screening in unhealthy older patients and underuse in healthy older patients 2. Future quality measures should focus on clinical benefit rather than simply chronological age to ensure that patients who are likely to benefit from a service receive it (regardless of age), and that those who are likely to incur harm are spared unnecessary and costly care We thank the following individuals for their valuable input on earlier drafts of this manuscript: Joseph Francis (VA Clinical Analytics and Reporting); Linda Kinsinger and Kathleen Pittman (VA National Center for Health Promotion and Disease Prevention); Jason A Dominitz, (VA Puget Sound Health Care System and the University of Washington School of Medicine); and Deborah Fisher (Durham VA Center for Health Services Research in Primary Care and Duke University Medical Center). Contributors: SDS was involved in all aspects of this study, including study conception, study design, data analysis, and manuscript writing and revisions. EAK and SV contributed to study design and manuscript writing and revisions. SM contributed to study design, data analysis, and manuscript writing and revisions. AAP and PS contributed to manuscript revisions. SDS is guarantor. Funding: This study was funded by the Veterans Health Administration. The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or final approval of the manuscript. The opinions expressed in this paper are of the authors and do not necessarily reflect those of the Department of Veterans Affairs. Competing interests: All authors have completed the ICMJE uniform disclosure form at [www.icmje.org/coi_disclosure.pdf](http://www.icmje.org/coi_disclosure.pdf) and declare: SDS had financial support from the Department of Veterans Affairs for the submitted work; no financial relationships with any organizations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work. Ethical approval: The study was approved by the Institutional Review Board of the VA Ann Arbor Healthcare System (RO: 2011-100615) on 28 March 2012. Transparency: SDS affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained. All authors had full access to study data and take responsibility for the integrity of the data and the accuracy of the data analysis. Data sharing: Statistical code is available on request from the corresponding author. Cite this as: *BMJ* 2014;348:g1247

INTRODUCTION ============ Polyglandular autoimmune syndrome (PAS) is defined as the functional disorder of two abnormal endocrine glands. In 1926, Schmidt reported a case in which nontuberculosis hypoadrenalism with autoimmune thyroiditis, and Neufeld et al. \[[@B1]\] divided the PAS into type I, II, and III according to the endocrinopathy occurrence age, heredity method, relationship with human leukocyte antigen (HLA), and the characteristics of accompanying disease. In 1980 \[[@B1]\], PAS type I corresponds to cases with at least two of the following: autoimmune Addison\'s disease, chronic mucocutaneous candidiasis, and idiopathic hypoparathyroidism, while PAS type II was defined as a disease group including Addison\'s disease, autoimmune thyroid disease, and type 1 diabetes \[[@B2]\]. PAS type III is not accompanied by Addison disease but includes other autoimmune diseases, such as type 1 diabetes together with autoimmune thyroid disease. In addition, Neufeld et al. \[[@B1]\] subdivided cases with both type 1 diabetes and autoimmune thyroiditis into PAS type IIIa. However, type I PAS is so rare that only 100 cases have been reported worldwide \[[@B3]\]. In Korea, some PAS cases with various clinical features including PAS III have been reported \[[@B4]-[@B8]\]. Albright\'s hereditary osteodystrophy (AHO) is a growth impairment-related congenital disease, accompanied by several pathognomonic features \[[@B9],[@B10]\]. These somatic features are associated with resistance to parathyroid hormone (PTH) including pseudohypoparathyroidism (PHP) I or pseudopseudohypoparathyroidism (PPHP) in an autosomal dominant inheritance \[[@B11]\]. Primary hypoparathyroidism is characterized as a deficiency or secretory disorder of PTH without underlying disorders including any prior neck surgery, chronic renal failure, alcoholism, gastrointestinal absorption disorders, or pathognomonic musculoskeletal abnormality such as AHO \[[@B12]\]. In the present case, the 32-year-old female with type I diabetes had autoimmune thyroiditis, and therefore we could diagnosed her illness as having PAS type IIIa. At the same time, the present case was accompanied by primary hypoparathyroidism with clinical features similar to AHO and thyroid papillary cancer, and this is a rare case. Therefore, we report this valuable experience. CASE REPORT =========== Patient ------- 32-year-old female. Chief complaint --------------- General weakness. Present illness --------------- She was receiving insulin injections for type 1 diabetes, but her symptoms had not been improved at all, which prompted her decision to come to our hospital. Past history ------------ She had been using 32 U insulin every day for type 1 diabetes for 11 years before her visit to our hospital. Family history -------------- All family members, including both parents and all siblings, had diabetes. Physical examination -------------------- When she appeared, her blood pressure was 130/80 mm Hg, pulse rate was 74 beats per minute, and body temperature was 36.6℃. She had no other symptoms, except for general weakness and central obesity (body mass index 23.4 kg/m^2^). No tumor-like mass was checked at her head and neck, nothing was discovered in chest auscultation. When we checked her abdomen, neither any pain nor hepatosplenomegaly was observed. No skin or oral mucosa pigmentation was found. There was no edema in lower leg. The neurological examination results were also normal. She weighed 51.5 kg and was 148.2 cm tall, and she was somewhat short woman with a short neck and round face, similar to a clinical features of AHO, but there was no pathognomonic brachydactyly. Laboratory results ------------------ The result of complete blood count was the hemoglobin 9.3 g/dL, the erythrocyte average volume 96 fL, leukocyte 5,720/mm^3^, and platelets 310,000/mm. The results of anemia work up were the total iron binding capacity 447.5 µg/dL, serum ferritin 35.11 ng/mL, reticulocyte count 2.89%, average erythrocyte hemoglobin density 32.6 g/dL, and vitamin B12 839.65 pg/mL. Her blood glucose level was 212 mg/dL, the level of hemoglobin A1c was 7.3%, and the simultaneous measurement of serum level of C-peptide showed decreased level of 0.13 ng/mL. Serum concentration of total calcium were 5.5 mg/dL (range, 8.7 to 1.4); ionized Ca 1.56 mg/dL (range, 2.3 to 2.58); phosphorus 6.7 mg/dL (range, 2.5 to 4.5); total protein 6.7 g/dL; albumin 3.3 g/dL; alkaline phosphatase 329 IU/L; aspartate aminotransferase 119 IU; alanine aminotransferase 65 IU/L. Intact PTH level was 13.32 pg/mL (range, 13 to 54). Serum adrenocorticotropic hormone (ACTH) and cortisol levels were 11.92 pg/mL (range, 10 to 60) and 0.94 (range, 9.4 to 26.1). The results of rapid ACTH stimulation test for adrenal function were normal. The results of thyroid function test were T3 99.01 ng/mL, free T4 0.85 µg/dL, and thyroid stimulating hormone 7.99 µIU/mL. The results for antimicrosomal and antithyroglobulin antibody were positive (52.91 IU/mL, 132.57 IU/mL), and, both islet cell antibody and antiglutamic acid decarboxylase antibody tested negative. In the Ellsworth-Howard examination, urinary phosphoric acid was slightly increased (12.3 mg/dL) and the increase of cyclic adenosine monophosphate (cAMP) was sufficient (4,400 µmol/mL). There was no GNAS mutation. Serum levels of sodium, potassium, blood urea nitrogen, and creatinine were within normal limits at 144 mmol/L, 5.3 mmol/L, 8.2 mg/dL, and 1.0 mg/dL. Radiological results -------------------- The chest X-ray showed no abnormal lesion. An brain computed tomography scan showed prominent calcifications in both basal ganglion and hypothalamus ([Fig. 1](#F1){ref-type="fig"}). The thyroid sonogram discovered several nodules, sized 4.5×3.7 mm and 3.9×2.9 mm on each side that were suspected to be thyroid papillary cancer. Treatment and clinical course ----------------------------- The patient came to the hospital experiencing general weakness, and she was diagnosed with hypocalcaemia on laboratory examination. At that time, both the levels of PTH and vitamin D were within normal range, but she had previously experienced spasm symptoms and the levels of previous intact PTH were low (4.38 to 10.66 pg/mL) 7 years ago. She has taken oral calcium medicine for treatment of idiopathic hypoparathyroidism because she had no underlying secondary disorders for hypocalcemia and low PTH level. Although she has no pathognomonic brachydactyly, her appearance was similar to AHO in clinical features including short stature, intelligence decline, round face, calcification on basal ganglion and hypothalamus, or obesity. Furthermore, she had a lower normal PTH level and hypocalemia on admission, and we decided to investigate the PTH resistance to identify PHP Ia or PPHP with AHO. In the Ellsworth-Howard examination for investigation of the PTH resistance, it was confirmed that there was no PTH resistance at the kidney and there was no sign of GNAS mutation. We diagnosed this case as hypocalcemia due to idiopathic hypoparathyroidism for lower serum calcium and traced intact PTH levels and began calcium and calcitriol medicine to treat the hypocalcaemia. Subsequently, her symptoms were improved with normalization of ionized Ca at 2.32 mg/dL (range, 2.3 to 2.53). The results of fine needle aspiration on two nodules of thyroid were suggestive malignancy and she underwent a total thyroidectomy. The biopsy results were papillary microcarcinoma with lymphocytic thyroiditis for two suspicious nodules (sized 1.5×1.0 cm on the left and 2.0×1.0 cm on the right). Among the four parathyroid glands, the left inferior parathyroid gland was not discovered, and the remaining three parathyroid glands were remained during surgery. We used levothyroxine 0.05 mg medication for Hashimoto thyroiditis and insulin determir 12 U and aspart 16 U for glucose control. DISCUSSION ========== PAS was reported by Schmidt in 1922 when he conducted an autopsy of a patient who had died of adrenal insufficieny and discovered that the lymphocyte had infiltrated into the thyroid and adrenal cortex, and thereafter, many cases have been reported in which both autoimmune endocrinopathy and nonendocrinopathy have been found in the same patient. In 1980, Neufeld et al. \[[@B1]\] classified the PAS into type I, II, and III according to age of endocrinopathy occurrence, heredity method, relationship with HLA, and accompanying disease, and this contributed greatly to screening and early diagnosis of PAS. PAS type I is also called autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy, and it has been the topic of recent genetic research. This revealed that PAS type I is known to be associated with the abnormality of a single gene such as autoimmune regulator gene (AIRE) through autosomal recessive heredity \[[@B13],[@B14]\]. In addition, the immune dysfunction, polyendocrineopathy, enteropathy, X-Linked (IPEX) syndrome is casued by serious immune damage with forkhead box P3 gene mutation at the regulatory T cell. PAS may be classified into three types including type I related with AIRE, type II related with HLA, and type III realted with IPEX. Particularly, PAS type I is so rare that only 100 cases have been reported worldwide \[[@B3]\]. In Korea, some cases of PAS with various clinical features have been reported \[[@B4]-[@B8]\], but the type III case is not common. Seven cases of PAS type III have been reported; three cases of PAS with autoimmune thyroid disease and systemic lupus erythematosis, two cases with autoimmune thyroid disease and rheumatoid arthritis, one case with pernicious anemia and insulin-dependent diabetes, and one case with autoimmune thyroiditis and leukopathy. The patient in the present case was diagnosed with type I diabetes (C-peptide 0.13) accompanied by insulin deficiency 11 years ago and thyroiditis was already under development according to thyorid ultrasonogram examination that was taken 7 years ago. After then, antoimmune thyroiditis was confirmed with the positive results of the thyroid autoantibodies and biopsy. Therefore the patient was diagnosed with PAS III with autoimmune thyroid disease and type 1 diabetes. Primary hypoparathyroidism is characterized as a deficiency or secretory disorder of PTH without underlying disorders, including prior neck surgery, chronic renal failure, alcoholism, gastrointestinal absorption disorders, and pathognomonic musculoskeletal abnormality as stated by Albright \[[@B12]\]. AHO is known to be relatively rare in primary hypoparathyroidism, but one case with AHO and primary hypoparathyroidism has been reported in Korea \[[@B15]\]. AHO is a congenital disease associated with short stature, obesity, rounded face, subcutaneous ossification, and characteristic shortening and widening of long bones in the hands and feet, and less frequently, mental retardation \[[@B9],[@B10]\]. Usually, AHO is associated with PHP I or PPHP with GNAS mutation \[[@B11]\]. Six cases of PPHP or PHP with AHO and two cases of PPHP or PHP without AHO have been reported in Korean \[[@B16],[@B17]\]. The PPHP is considered when a patient has definite AHO evidences, a family history of PPHP, normal results on the Ellsworth-Howard examination, reductions in erythrocyte membrane Gs activity, GNAS mutation or a patient\'s father has AHO evidences \[[@B18],[@B19]\]. The AHO patients with paternal inheritance may be diagnosed with PPHP without hormone resistance of PTH, but the patients with maternal inheritance may have PHP by hormone resistance \[[@B11]\]. The patient in the present case was confirmed with hypocalcemia, low PTH, and a previous history of spasms that might be asoociated with hypocalcemia. Basen on laboratory results, we could consider this case as primary hypoparathyroidism. Interistingly, she had several clinical features similar to AHO including short stature, intelligence decline, round face, calcification on basal ganglion and hypothalamus, and obesity except pathognomonic brachydactyly, which might be considered as PHP or PPHP. Therefore, the Ellsworth-Howard examination was performed to investigate the resistance to PTH actions on target organs \[[@B20]\] but we diagnosed the patient with idiopathic hypoparathyroidism due to the increase of urinary phosphoric acid, sufficient increase of cAMP by 20 times (4,400 µmol/mL), and no GNAS mutation. The subnormal response of urinary phosphoric acid might be due to variations of the basal level or diurnal execretion. In conclusion, it is thought that this is a case of PAS type IIIa with autoimmune thyroiditis and type 1 diabetes required insulin treatment. At the same time, this case was accompanied by thyroid papillary cancer and primary hypoparathyroidism with clinical features that could be mistaken for AHO. This is a rare case and instructive experience for us and we report the case with the review of reference literature. In the present case, the patient was young female who was diagnosed with type I diabetes as an absolute deficiency of insulin and required insulin supply for glucose control. She had also autoimmune thyroid disease, thyroid papillary cancer and primary hypoparathyroidism with clinical features resembled AHO. Here, we report a rare case of PAS type IIIa with autoimmune thyroiditis and type 1 diabetes, accompanied by thyroid papillary cancer and primary hypoparathyroidism with clinical features that could be mistaken for AHO. No potential conflict of interest relevant to this article was reported. {#F1}

Introduction {#Sec1} ============ The efficacy of 5-fluorouracil (5FU)-based adjuvant chemotherapy and the associated survival benefit have been firmly established for patients with stage III colorectal cancer (CRC) \[[@CR1]\]. However, the use of adjuvant therapy for stage II colon cancer patients remains controversial \[[@CR2]\]. Typically, only high-risk stage II CRCs are treated. Hence, there is a need for predictive factors to support treatment decisions in this group of patients, the majority of whom will be cured by surgery alone \[[@CR3]\]. This is especially important because a wider implementation of prophylactic programs in recent years has resulted in an increased number of patients with stage B2 tumors. 5-Fluorouracil has been used to treat colorectal cancer patients for a long time \[[@CR4]\]. The target enzyme for 5FU is thymidylate synthase (TS). This enzyme catalyzes the conversion of deoxyuridine-5′monophosphate to deoxythymidine-5′-monophosphate and is therefore essential for DNA synthesis and repair. Resistance to 5FU-based treatment might depend both on the level of expression of TS and other enzymes involved in the metabolism of 5FU and on mechanisms involved in cell growth or apoptosis. p21^WAF1^ is a multifunctional cell cycle-related protein that inhibits cyclin-dependent kinases (CDKs), which results in cell cycle arrest in the G1 phase \[[@CR5]\]. p21^WAF1^ expression is known to be inversely related to cell proliferation and directly related to terminal differentiation \[[@CR6]\]. Recently, it has been shown that TS expression was upregulated in p21^WAF1^-/- human colorectal cancer HCT116 cells, and TS promoter activity was downregulated by ectopic p21 expression \[[@CR7]\]. Furthermore, a CDK inhibitor reduced expression of TS in a dose-dependent manner, and the reduction resulted in enhancement of sensitivity to 5FU in cultured human colon cancer cells \[[@CR8]\]. Taken together with other reports \[[@CR9]\], this suggests a great importance of p21^WAF1^ for in vitro response to chemotherapeutic agents and that the CDK inhibitor p21^WAF1^ regulates thymineless stress-induced DNA damage. Hence, p21^WAF1^ expression might have a predictive significance in 5FU-treated CRC patients. In several \[[@CR10]--[@CR19]\] but not all \[[@CR20]--[@CR24]\] previous studies, positive p21^WAF1^ expression has been suggested as an indicator of good prognosis in patients with stage III/IV CRC. However, it is not known whether the same is true for stage B2 patients. Therefore, our purpose was to assess the influence of p21^WAF1^ expression on disease-free survival (DFS) and overall survival (OS) of Astler--Coller stage B2 patients with CRC who either were or were not treated with 5FU-based adjuvant chemotherapy, and of stage C patients treated with adjuvant chemotherapy. Materials and methods {#Sec2} ===================== **Patients** This retrospective study was based on tumor tissue from 275 unselected patients who underwent potentially curative colorectal resection for sporadic CRC (defined as an absence of relevant family history at the time of admission to the hospital). Distant metastases at the time of operation were excluded by preoperative liver ultrasonography, chest x-ray, and intraoperative exploration. Table [1](#Tab1){ref-type="table"} lists the clinico-pathological details of the 275 tumors and patients. Astler--Coller staging of CRC based on the extent of local invasion and the status of lymph nodes was used \[[@CR25]\]. In this system, stage B2 denotes the tumor penetrating through the muscularis propria and uninvolved nodes, stage C---tumors extending into the muscularis propria or penetrating through it with lymph node metastases. The mean age of the patients was 61.1 ± 10.8 years (range, 33--83; median, 62). The splenic flexure was regarded as the border between the proximal and distal colon. The Research Ethics Committee of Pomeranian Medical University approved this study. Table 1Relations between p21^WAF1^ expression and clinico-pathological parametersParameterp21^WAF1^ Expression*nn* (%)*p*Age (years)\<6213887 (63)0.39\>6213779 (58)Sex Females12075 (63)0.54 Males15591 (59)GradeG1 + G215691 (58)0.46G3^a^11975 (63)Astler--CollerB213372 (54)*0.049*C14294 (66)SiteProximal5337 (70)0.16Distal222129 (58)SiteRectum13881 (59)0.62Colon13785 (62)Radiotherapy (rectal tumors)Preoperative4326 (61)0.85Postoperative RT and no RT^b^9353 (57)^a^Including mucinous carcinoma^b^Radiotherapy status unknown in two casesTumors were resected between July 1997 and April 2004 in the Departments of Surgery of the teaching hospitals of Pomeranian Medical University in Szczecin and the Regional Oncological Center in Szczecin, Poland. The operations consisted of either a resection with lymphadenectomy or a total mesorectal excision for rectal carcinomas. A total of 203 (74%) patients (83 high-risk stage B2 and 120 stage C) were treated with adjuvant chemotherapy (six 5-day courses of bolus infusion of 5FU (425 mg/m^2^) every 4 weeks combined with leucovorin, 20 mg/m^2^). A total of 72 patients (22 stage C patients) did not receive adjuvant chemotherapy due to internist contraindications.A total of 41 patients with rectal cancer received postoperative radiotherapy (50.4 Gy), and 43 patients, preoperative radiotherapy (5 × 5 Gy). Of the 133 Astler--Coller stage B2 tumors, there were 65 rectal tumors, and of these, 24 (18%) received preoperative and 11 (8.3%) received postoperative radiotherapy. Radiotherapy did not influence DFS (*p* = 0.61) nor OS (*p* = 0.83) of patients with rectal tumors. Since we did not find a statistically significant difference in p21^WAF1^ expression between the group of patients subjected to preoperative radiotherapy and the remaining 93 rectal tumors not treated with preoperative radiotherapy (Table [1](#Tab1){ref-type="table"}), the former was included in the study.Time from the surgery until the time of death due to cancer or to last known follow-up was regarded as OS, and the time until the first appearance of metastasis or local recurrence was regarded as DFS. The median follow-up was 54 months (mean, 57.3 ± 30.5 months; range, 5--143 months). During the follow-up, 66 of the 275 (24%) patients died of their disease, and 165 (60%) were alive without symptoms from the disease. Recurrences were found in 106 patients. Four patients died of non-cancer-related causes, and they were treated as censored observations. **Tissue microarray (TMA) construction** Tumor areas with the highest mitotic activity at the outer invasive zone of the cancer were chosen for tissue microarrays which were constructed as previously described \[[@CR26]\]. **Immunohistochemistry** Slides with tissue microarrays were deparaffinized, rehydrated, and had the endogenous peroxidase activity blocked. Slides were immersed in pH 9.0 buffer, and heat-induced antigen retrieval was performed in a pressure cooker (Pascal, Dako, Glostrup, Denmark). Monoclonal p21^WAF1^ antibody (Dako) was used (dilution, 1:25; incubation time, 30 min), and the slides were immunostained using a Dako EnVision+ kit according to the manufacturer's instructions. We used the sensitive EnVision+ visualization system because the detection system used is regarded as a critically important variable in immunohistochemical analysis, and detection methods using signal amplification with HRP-labeled polymers have been shown to be more sensitive than methods without such a layer of amplification \[[@CR27]\]. The reaction was developed with diaminobenzidine substrate--chromogen solution, and the slides were counterstained with hematoxylin. Positive controls included colorectal adenocarcinoma previously shown to have a high level of p21^WAF1^ expression. Negative controls omitted the primary antibody. The immunohistochemical procedure for all tissue microarrays from 275 tumors was performed at the same time in identical conditions because, instead of 275 histological slides of whole tissue sections, only four slides containing tissue cores from all 275 tumors were processed. **Scoring** Immunohistochemical staining for each tumor core was independently assessed by two observers (PD and WD) who were blinded to the clinical and pathological data. In cases of disagreement, the result was reached by consensus. The percentage of tumor cell nuclei with unequivocal staining was recorded for each core. P21^WAF1^ expression in tumors was variable, and so, tumors were classified as negative (\<1% of positive tumor cells) or positive (if ≥1% of tumor cells showed nuclear immunoreactivity). **Statistical analysis** Associations between the presence of p21^WAF1^ expression in tumors and other categorical variables were analyzed with the Fisher exact test. The Kaplan--Meier method was used for the univariate survival analysis, and the differences between compared groups were assessed by the log-rank test. A Cox proportional hazards model was used for univariate and multivariate analyses of factors associated with OS and DFS. The independent variables included in the model were: age, gender, tumor site, Astler--Coller stage, histological grade, 5FU-based adjuvant chemotherapy, and presence of p21^WAF1^ expression. A *p* \< 0.05 was considered statistically significant. STATISTICA version 9.1 (StatSoft Inc., Tulsa, USA) was used for the statistical analysis. Results {#Sec3} ======= Expression of p21^WAF1^ (Fig. [1a--d](#Fig1){ref-type="fig"}) was found in the nuclei of 60.4% (166/275) of all cases, in 54.1% (72/133) of stage B2 cancers, and in 66.2% (94/142) of stage C tumors. Of all the parameters examined, only Astler--Coller stage was associated with p21^WAF1^ expression (Table [1](#Tab1){ref-type="table"}). Fig. 1Expression of p21^WAF1^in representative cores of CRC. **a**, **b** High expression of p21^WAF1^ in the nuclei of tumor cells. **c**, **d** Low expression of p21^WAF1^ in the nuclei of cancer cells. **b**, **d** Fragments of the cores from a and c, respectively, at high magnification **Group B2** No association was found between p21^WAF1^ expression and DFS or OS in the whole B2 group, neither in univariate or multivariate analysis (data not shown). However, 83 patients in this group were treated with 5FU-based adjuvant chemotherapy, and 50 patients were not. Further analysis of the association of p21^WAF1^ expression with DFS and OS in these subgroups showed that expression of p21^WAF1^ was associated with DFS (*p* = 0.025) and OS (*p* = 0.0079) only in the subgroup of stage B2 patients that were treated with adjuvant chemotherapy (Fig. [2a](#Fig2){ref-type="fig"}). Moreover, p21^WAF1^ expression remained in this subgroup as the only independent prognostic parameter in the multivariate analysis in relation to DFS and OS (*p* = 0.035 and *p* = 0.02, respectively; Table [2](#Tab2){ref-type="table"}). In the stage B2 subgroup of patients not treated with chemotherapy, none of the parameters studied were associated with DFS or OS (data not shown). Fig. 2DFS (*black curves*) and OS (*red curves*) of stage B2 patients treated with adjuvant chemotherapy categorized according to the p21 expression (*n* = 83) (**a**) and of stage B2 patients with p21-positive tumors categorized according to adjuvant chemotherapy (*n* = 72) (**b**). DFS (*black curves*) and OS (*red curves*) of stage C patients (*n* = 120) (**c**) and of stage B2 + C patients (*n* = 203) (**d**) treated with adjuvant chemotherapy categorized according to p21 expressionTable 2Multivariate analysis of disease-free and overall survival of stage B2 and C patients treated with adjuvant chemotherapyCategoryB2 (*n* = 83)C (*n* = 120)Disease-free survivalOverall survivalDisease-free survivalOverall survivalHazard ratio (95% Cl)*p*Hazard ratio (95% Cl)*p*Hazard ratio (95% Cl)*p*Hazard ratio (95% Cl)*p*Sex0.97 (0.37--2.53)0.950.99 (0.24--4.04)0.991.21 (0.71--2.08)0.491.50 (0.73--3.05)0.27Age0.99 (0.95--1.03)0.631.00 (0.94--1.07)0.991.02 (0.99--1.05)0.121.02 (0.99--1.05)0.29Grade0.56 (0.20--1.57)0.270.39 (0.08--1.86)0.241.93 (1.12--3.33)*0.018*2.87 (1.37--6.01)*0.005*Tumor site0.99 (0.29--3.47)0.102.55 (0.64--10.15)0.190.57 (0.27--1.17)0.130.35 (0.10--1.15)0.08P21^WAF1^0.36 (0.14--0.93)*0.035*0.16 (0.03--0.76)*0.02*0.60 (0.35--1.04)0.070.48 (0.24--0.93)*0.03*In the subgroup of 72 stage B2 patients with p21^WAF1^ expression, adjuvant chemotherapy was associated with better DFS (85% 5-year survival versus 65% without chemotherapy, *p* = 0.03; Fig. [2b](#Fig2){ref-type="fig"}) and OS (96% 5-year survival versus 82% without chemotherapy, *p* = 0.014; Fig. [2b](#Fig2){ref-type="fig"}). In the subgroup of 61 stage B2 patients with negative p21^WAF1^ expression, a trend for the association of chemotherapy with worse survival (both DFS and OS) was observed, but the differences were not statistically significant (data not shown). **Group C** The expression of p21^WAF1^ in tumors of patients treated with adjuvant chemotherapy was associated with DFS in a univariate analysis (HR = 0.58, 95% CI = 0.34--0.99, *p* = 0.047) and was on the verge of statistical significance in the multivariate analysis (*p* = 0.07) (Table [2](#Tab2){ref-type="table"}). In terms of OS, there was a statistically significant association with p21^WAF1^ expression in both the univariate (HR = 0.41, 95% CI = 0.21--0.80, *p* = 0.008) and multivariate (*p* = 0.03) (Table [2](#Tab2){ref-type="table"}) analyses. Kaplan--Meier survival curves of patients with p21^WAF1^ expression in the nuclei of cancer cells indicate better DFS (*p* = 0.055) and significantly better OS (72% versus 45% 5-year survival of patients with negative p21^WAF1^ expression, *p* = 0.0079; Fig. [2c](#Fig2){ref-type="fig"}). **Stage B2 + C patients treated with adjuvant chemotherapy** Expression of p21^WAF1^ was associated with DFS and OS (*p* = 0.03 and *p* = 0.002, respectively) in a univariate analysis. The Astler--Coller stage and p21^WAF1^ expression were found to be independent prognostic factors for DFS and OS (Table [3](#Tab3){ref-type="table"}). In multivariate analysis, there was no statistically significant association between site of CRC (neither rectum versus colon nor proximal versus distal as defined by the splenic flexure) and DFS (*p* = 0.77, *p* = 0.28, respectively) or OS (*p* = 0.77, *p* = 0.28, respectively). Kaplan--Meier survival curves showed significantly better DFS and OS of patients with tumors showing p21^WAF1^ expression as compared to those negative for p21^WAF1^ (*p* = 0.03 and *p* = 0.0018, respectively; Fig. [2d](#Fig2){ref-type="fig"}). Table 3Multivariate analysis of disease-free and overall survival of patients treated with adjuvant chemotherapy (group B2 + C, *n* = 203)CategoryDisease-free survivalOverall survivalHazard ratio (95% Cl)*p*Hazard ratio (95% Cl)*p*Sex1.20 (0.75--1.90)0.441.42 (0.77--2.65)0.27Age1.01 (0.99--1.03)0.321.01 (0.98--1.04)0.58Astler--Coller3.16 (1.86--5.36)*\<0.0001*3.60 (1.77--7.33)*0.0004*Grade1.48 (0.94--2.32)0.091.95 (1.07--3.56)*0.03*Tumor site0.64 (0.34--1.20)0.170.62 (0.26--1.49)0.28P21^WAF1^0.52 (0.33--0.83)*0.0062*0.36 (0.20--0.66)*0.0009* Discussion {#Sec4} ========== We found p21^WAF1^ nuclear expression in 60.4% CRCs which is comparable to the literature data (16--87%) \[[@CR11], [@CR13]--[@CR15], [@CR28], [@CR29]\]. The majority of reports (but not all) concerning the prognostic role of p21^WAF1^ expression in CRCs indicate better OS and/or DFS for patients with tumors showing p21^WAF1^ expression (Table [4](#Tab4){ref-type="table"}). In some of these reports, p21^WAF1^ expression was an independent prognostic factor for OS and/or DFS (Table [4](#Tab4){ref-type="table"}). However, in those reports, patients with various stages of the disease who were subjected to different treatment protocols (surgery alone, surgery and adjuvant chemotherapy, radiochemotherapy) were often grouped together for the analysis, or no information on the mode of therapy is given. Thus, it is difficult to infer whether stage II patients with p21^WAF1^-positive tumors may receive any benefit from adjuvant chemotherapy. Table 4Major reports on p21^WAF1^ expression and survival of patients with CRCsSourceStudy groupStage of CRC and type of treatmentResultsCheng J.D., et al. \[[@CR10]\]*n* = 39Metastatic CRC 5FU CHTH5FU responders had greater p21 expressionRopponen K.M, et al. \[[@CR11]\]*n* = 1620--D incl. 62--B majority: surgery only; 22--CHTHBetter OS and RFS for patients with p21+ tumors; p21: independent prognostic parameter for OS and RFS (MA)Viale G., et al. \[[@CR12]\]*n* = 191I--IV CHTH?↓p21 → poor OS and DFS (UA)Bukholm I.K., et al. \[[@CR13]\]*n* = 61B--D CHTH?Low p21 → increased risk of metastases and deathZirbes T.K., et al. \[[@CR14]\]*n* = 294I--IV incl. 90--II. surgery CHTH?Better OS for patients with p21+ tumors; p21 independent prognostic parameter (MA)Holland T.A., et al. \[[@CR15]\]*n* = 126A--D RCHTH?Better OS for patients with high p21 expressionPasz-Walczak G., et al. \[[@CR16]\]*n* = 122I--IV CHTH?Better OS for patients with p21+ tumors (UA)Watanabe T., et al. \[[@CR20]\]*n* = 460II (B)--105 III (C)--355, 5FU CHTHNo association of p21 expression with survival in all and in stage B tumorsHoss A., et al. \[[@CR21]\]*n* = 100T2--3, N0 rectum, surgery onlyNo association of p21 expression with survivalSchwandner O., et al. \[[@CR17]\]*n* = 160I--III rectum, surgery, *n* = 69 surgery + CHTH + RT, *n* = 91Better RFS but not OS for patients with p21+ tumors; p21 independent prognostic parameter for RFS but not OS (MA)Rau B., et al. \[[@CR22]\]*n* = 66T3--4, N0--2, M0--1 rectum, RCHTHNo prognostic significance of pretreatment p21 expression. Post-treatment increase of p21: shorter DFSPrall F., et al. \[[@CR18]\]*n* = 184I--IV incl. 55-III, III: *n* = 32 -5FU, *n* = 23-5FU + RTBetter OS for patients with p21 + tumors (MA)Mitomi H., et al. \[[@CR19]\]*n* = 211B--D incl. 83-B CHTH for C--DBetter OS for patients with high p21 expression; p21 independent prognostic parameter (MA)Ioachim E., et al. \[[@CR23]\]*n* = 97Dukes B, C, surgery, CHTH?No association of p21 expression with OS and DFSNoske A., et al. \[[@CR24]\]*n* = 116II--III RCHTHBetter RFS and OS for patients with p21- tumors↓ down regulation, *UA* univariate analysis, *MA* multivariate analysis, *OS* overall survival, *DFS* disease-free survival, *RFS* relapse-free survival, *RT* radiotherapy, *CHTH* chemotherapy, *RCHTH* radiochemotherapy, *?* not known The benefit of adjuvant 5FU-based chemotherapy has been firmly established for patients with stage III CRC. However, in the stage B2 group, it is not known whether the survival benefit from chemotherapy is sufficient to outweigh the toxicity and cost of the treatment \[[@CR2]\]. Only one study \[[@CR20]\] has addressed the issue of the influence of 5FU-based adjuvant chemotherapy on the survival of patients with stage B2 CRC in relation to p21^WAF1^ expression. In that study, pretreatment levels of p21^WAF1^ were not related to survival of patients with stage II (and III) CRC treated with adjuvant chemotherapy. However, increased levels of p21^WAF1^ were associated with the sensitivity of metastatic CRC to 5FU-based chemotherapy \[[@CR10]\]. Our results indicate for the first time that p21^WAF1^ expression in CRC tumor cells is an independent factor that is associated with favorable DFS and OS in patients with stage B2 tumors treated with 5FU-based adjuvant chemotherapy. Striking survival benefits were seen for stage B2 patients who received adjuvant chemotherapy compared with those who did not. Conversely, chemotherapy did not significantly influence DFS or OS of stage B2 patients with p21^WAF1^-negative tumors. Rather, a statistically non-significant trend towards worse survival was observed for stage B2 patients with p21^WAF1^-negative tumors treated with chemotherapy. The differences between our results and those of Watanabe et al. \[[@CR20]\] may be attributed to the different scoring systems and different cutoff points used for the interpretation of immunohistochemical staining, as well as to differences in immunohistochemical methods. The major advantage of tissue microarrays is that tens of cases can be processed in identical laboratory conditions which greatly improves the reproducibility of the immunohistochemical method. Prall et al. \[[@CR18]\] used tissue microarrays (*n* = 184) and found better OS for combined group of stage I--IV patients with p21^WAF1^-positive tumors. Our results (from the whole group) are well in accord with this report and give further support for the association of p21^WAF1^ expression and longer survival of patients with stage C/III CRCs treated with adjuvant 5FU-based chemotherapy as has been reported previously (Table [4](#Tab4){ref-type="table"}). One limitation of TMA technology is that "punched" cores from donor tissues may not always be representative of the entire tumor. In this report, we applied one core from carefully identified, histologically relatively homogenous area with the highest mitotic activity in the outer invasive zone of each CRC. Using this approach, we found 60.4% of p21^WAF1^-positive CRCs which is within the range reported in the literature. Hoos et al. \[[@CR30]\] reported that correlations between phenotypes and clinical outcome were not significantly different between full sections and triplicate 0.6-mm core tissue microarrays. However, on the other hand, they were not significantly different when only one 0.6-mm core tissue microarray was used \[[@CR31]\]. The authors of the latter report conclude that tissue microarray "with a single core per specimen ensures full biological representativeness to identify the associations between biomarkers and clinico-pathological parameters, with no significant associated sampling bias." So, careful sampling of the representative region of the tumor is regarded as the key step in the construction of tissue microarrays. TS, the target enzyme for 5FU, is essential for DNA synthesis \[[@CR32]\], and it may function as an oncogene \[[@CR33]\]. Inhibition of TS induces apoptosis and cytotoxicity in human colorectal cancer cells \[[@CR7]\]. The importance of p21^WAF1^ in the response of CRC to chemotherapeutic agents is supported by in vitro studies \[[@CR7]--[@CR9]\]. It has been reported that p21^WAF1^ (a CDK inhibitor) regulates thymineless stress-induced cytotoxicity of human colon carcinoma cell lines \[[@CR7]\]. Also, TS expression is mediated through the inhibition of CDK: TS expression was upregulated by the knockout of the p21^WAF1^ gene in a CRC cell line \[[@CR8]\]. In addition, reduction of TS expression results in enhancement of the sensitivity to 5FU in human CRC cell lines \[[@CR8]\]. Our results are in line with other reports that show that p21^WAF1^ is a critical mediator of the cytotoxic action of TS inhibitors in cultured human colorectal cancer cells \[[@CR7]\] and that CDK inhibitor enhances the sensitivity to 5FU in colorectal cancer cell lines \[[@CR8]\]. In fact, p21^WAF1^ is required for maximal cytotoxicity induced by thymineless stress in colorectal cancer cells in culture \[[@CR7]\]. Poor survival of patients with p21^WAF1^-negative tumors treated with adjuvant 5FU-based chemotherapy may perhaps be attributed to inherent resistance to 5FU. It has been shown that development of resistance to 5FU by colon cancer cell lines is associated with downregulation of the *CDKN1A* gene, along with other genes engaged in DNA damage response/repair pathway \[[@CR34]\]. In summary, we found that p21^WAF1^ expression in CRC tumor cells identifies a subgroup of Astler--Coller stage B patients who would benefit significantly from 5FU-based chemotherapy and may therefore allow for better selection of patients for adjuvant chemotherapy. We believe that, in order to maximize the benefit of 5FU-based adjuvant therapy and to spare patients from unnecessary toxicity, stage B2 patients should be stratified according to p21^WAF1^ status. However, because this is a retrospective study, our results should be confirmed by further prospective randomized investigations. This work was supported by the grant KBN 2P05B 174 28 from the Committee for Scientific Research. **Conflict of interest** We declare that we have no conflict of interest. **Open Access** This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.

Introduction ============ Doxorubicin (DOX) is an effective chemotherapeutic agent that is widely used to treat numerous types of neoplasms. However, serious cardiotoxic side effects, which may cause arrhythmia and heart failure, limit its clinical application ([@b1-ol-0-0-8020]--[@b4-ol-0-0-8020]). Multiple mechanisms are involved in DOX cardiotoxicity, including lipid peroxidation and decreased glutathione levels ([@b5-ol-0-0-8020]), calcium overloading and mitochondrial dysfunction via increases in mitochondrial calcium and the generation of reactive oxygen species (ROS) ([@b6-ol-0-0-8020],[@b7-ol-0-0-8020]). Increased oxidative stress and an antioxidant deficit serve key roles in DOX-induced cardiotoxicity. Previous data suggests that the dysregulation of calcium handling and mitochondrial function contribute to DOX-induced cardiotoxicity ([@b8-ol-0-0-8020]). Multiple cardioprotective drugs may be combined to eliminate its cardiotoxicity or reduce doses to an acceptable level are expected to increase its efficacy ([@b8-ol-0-0-8020]). The majority of these attempts produced beneficial effects, but the search for more effective strategies against DOX-induced complications achieved little success ([@b6-ol-0-0-8020]). Therefore, additional adjuvant drugs are co-administered with DOX to patients with neoplastic diseases to reduce DOX-induced cardiotoxicity or enhance its therapeutic effects ([@b8-ol-0-0-8020]). Berberine (Ber) is an isoquinoline alkaloid that was originally extracted from the traditional Chinese plant *Coptis chinensis* (Huang Lian), and is an established treatment for diarrhoea in traditional Chinese medicine ([@b9-ol-0-0-8020]). Ber exhibits a wide range of pharmacological activities, including: Antioxidant properties to attenuate reactive oxygen species (ROS) formation in various tissues; anti-diabetic; anti-hyperlipidaemic; anti-inflammatory; anti-tumour; and cardio-protective effects ([@b9-ol-0-0-8020],[@b10-ol-0-0-8020]). We hypothesized that the antioxidant and cardio-protective effects of Ber may exhibit protective effects against DOX-induced cardiomyopathy. The present experimental study investigated the possible protective effects of Ber against acute DOX-induced cardiotoxicity induced in a rat model. The effect of Ber on changes in known indicators of cardiotoxicity and oxidative stress including serum creatine kinase (CK), creatine kinase isoenzyme (CK-MB) activities and serum and myocardial superoxide dismutase (SOD), malondialdehyde (MDA) and catalase (CAT) contents was investigated. The present study may support the utility of Ber as a safe, clinically-approved drug in the treatment of cancer. Materials and methods ===================== ### Drugs and chemicals DOX was provided by Lingnan Pharmaceutical, Ltd. (Guangzhou, China). Ber was provided by Acros Organics (Geel, Belgium). Fluo3-AM was purchased from Sigma-Aldrich; Merck KGaA (Darmstadt, Germany). MDA, CAT and SOD assay kits were purchased from Nanjing Jiancheng Bioengineering Institute (Nanjing, China). CK and CK-MB assay kits were purchased from Sysmex Corporation (Kobe, Japan). Rhodamine (Rh-123) and Rhod-2-acetoxymethyl (rhod-2-AM) were purchased from Molecular Probes; Thermo Fisher Scientific, Inc., (Waltham, MA, USA). The working solutions of Harris Haematoxylin and eosin were purchased from Baso Diagnostics Inc. Zhuhai (New Taipei City, Taiwan). ### Animals and treatments All experiments were performed in compliance with the Guide for the Care and Use of Laboratory Animals ([@b11-ol-0-0-8020]) and were reviewed and approved by the Ethics Committee for the Use of Experimental Animals at Hebei Medical University (Shijiazhuang, China). Sprague-Dawley (SD) rats weighing 200--250 g were obtained from the medical laboratory of Hebei Medical University. The animals were acclimated to the laboratory environment for 1 week in standard experimental conditions (12 h light:12 h dark schedule) and allowed access to food and water *ad libitum*. Animal experiments were performed in accordance with the National Institutes of Health guidelines for the experimental use of animals ([@b11-ol-0-0-8020]). Rats were randomly assigned to the following five groups of 10 animals each: Water-treated (control) group, DOX-treated group and DOX plus Ber treatment at doses of 5, 10 and 20 mg/kg. The control and DOX groups received distilled water orally for 10 consecutive days. Ber was administered at the aforementioned doses orally once daily for 10 consecutive days. The dosing volume was 1 ml/100 g body weight. All rats were intraperitoneally injected with a single dose of DOX 20 mg/kg on day 8, with the exception of the control group. The selected dose was based on our previous study ([@b12-ol-0-0-8020]). ### Sample collection and biochemical assays All rats were anaesthetized 48 h after DOX injection, and blood samples were collected prior to sacrifice. Serum was separated for CK and CK-MB assays. The heart was quickly isolated, blotted dry on filter paper, and weighed. The heart was cut and prepared for histopathological examination, and one part was prepared as 10% homogenates in ice-cold saline for the determination of MDA, CAT activity and SOD contents. The total protein content was detected with a bicinchoninic acid protein Assay kit (Pierce, Thermo Fisher Scientific Inc.), and the activities of MDA, CAT and SOD in the cardiac tissue were expressed as units/mg protein. ### Histopathological examinations To analyse the histopathologic changes of the cardiac tissue, one part of the heart was fixed in 10% buffered formalin, embedded in paraffin and dehydrated in an ascending series of ethanol (70, 80, 96, and 100%). Tissue samples were embedded in paraffin and cut into 5-µm thick slices. The sections were stained at room temperature with haematoxylin working solution and 1% eosin (H&E) for 2 min respectively for histological analysis under a light microscope (magnification, ×200, Olympus BX-50; Olympus Corporation, Tokyo, Japan). ### General toxicity observation A total of 10 animals were used to determine mortality in each group. The humane endpoints were a weight loss above 15% of initial weight or animal in a state of prostration. General conditions, mortality and body weight of the animals were observed daily until the end of the experiment. Fluid accumulation in the abdominal cavity was assessed at the end of the experiment subsequent to abdominal opening and scored on a graded scale of 0 to 3^+^, where: 0, none; 1^+^, mild; 2^+^, moderate; and 3^+^, severe ([@b13-ol-0-0-8020]). ### Acute isolation of cardiac myocytes Single cardiac myocytes were enzymatically isolated from adult male rat hearts as described previously ([@b14-ol-0-0-8020]). Briefly, SD rats (200--250 g) were anaesthetized using pentobarbital sodium (50 mg/kg, intraperitoneally). The hearts were rapidly excised, mounted through the aorta, and perfused on a modified Langendorff apparatus with a calcium-free Tyrode solution at 37°C, followed by 0.6% collagenase II in a Ca-free Tyrode solution to digest the heart. Isolated ventricular myocytes were maintained in Krebs solution. All experiments were performed within 6 h of ventricular myocyte isolation. ### Measurement of cytosolic calcium concentration \[Ca^2+^\]~i~ Isolated myocytes were prepared as aforementioned and loaded with 20 µmol/l membrane-permeable Fluo3-AM working solution containing 0.03% pluronic F-127 at 37°C for 60 min. The cells were superfused with fresh Tyrode solution three times at 25°C to allow de-esterification of Fluo3-AM. Cells were mounted in a chamber on the stage of an inverted microscope. Only cells with a rod shape and visible striations were used. Fluo3-AM fluorescence in cells was excited at 488 nm, and fluorescence emission was recorded at 530 nm using a photomultiplier. The fluorescence signal was detected using a confocal laser scanning system (Leica TCS-SP II; Leica Microsystems GmbH, Germany). Calcium measurement is represented as relative fluorescence intensity ((FI-FI~0~)/FI~0~, %; FI~0~: Control; FI: Administration of drugs). Cells were perfused with normal Tyrode solution for 2 min and perfused with Tyrode solution containing Ber for 15 min. A total of 50--100 images were scanned from each cell, and the data were analysed by a confocal laser microscopic system (Leica TCS SP2, Mannheim, Germany). ### Mitochondrial Ca^2+^ concentration and membrane potential measurements The mitochondrial membrane potential (ΔΨ~m~) was visualized in cardiomyocytes stained with the fluorescent probe Rh-123. Myocytes were incubated with 0.5 µM Rh-123 for 10 min at 37°C. Rh-123 fluorescence was detected at the emission wavelength of 490 nm, and the green emission fluorescence was excited at 530 nm to analyse ΔΨ~m~. The mitochondrial Ca^2+^ concentration \[(Ca^2+^)~m~\] was measured using the Ca^2+^-sensing dye rhod-2-AM in loaded cells. Myocytes were exposed to 10 µM rhod-2 acetoxymethyl ester for 120 min at 4°C, and transferred to 37°C for an additional 30 min in the Tyrode solution ([@b14-ol-0-0-8020]). Rhod-2 fluorescence was excited via excitation at 540 nm, and a 570 nm bandpass barrier filter was used to analyse \[Ca^2+^\]~m~. The fluorescence signal was detected using a confocal laser scanning system mentioned before. Regions of rod-shaped cardiomyocytes were outlined/highlighted to represent changes in fluorescence intensity over time, and the intensity of the selected region of the image was analysed using a confocal laser microscopic system. The fluorescence intensity was normalized to the initial value (F~0~) recorded in normal Tyrode solution in each experiment, and the density ratio represented the relative fluorescence: (FI-FI~0~)/FI~0~ ×100, where FI~0~ represents the control and FI represents the drug-treated groups. Rat cardiomyocytes were treated with 0.1 or 1 µM Ber for 20 min and exposed to 1 µM DOX for 15 min, which is the minimum time required for effective changes in \[Ca^2+^\]~m~ and ΔΨ~m~ ([@b15-ol-0-0-8020]). ### Statistical analysis All data are presented as the mean ± standard deviation. Statistical tests were conducted using Microsoft Excel 2010 (version number, 14.0.4760.1000; Microsoft Corporation, Redmond, WA, USA) and SPSS for windows (version 11.0; SPSS Inc., Chicago, IL, USA). Following a one-way analysis of variance, quantitative data were analysed with post hoc contrasts by Fisher\'s least significant difference test. A chi-squared test was used to analyse the difference of the injury severity scores of the heart among the various treatment groups. P\<0.05 was considered to indicate a statistically significant difference. Results ======= ### Effects of Ber on animal body weight and survival in DOX-induced cardiotoxicity DOX treatment for two days reduced rat body weight compared with the control group (235±35 vs. 266±15 g; P\<0.01). Treatment with 5, 10 and 20 mg/kg Ber exerted no marked increment effects on body weight (231±36, 226±16 and 253±54 g, respectively; P\>0.05 vs. DOX alone). There was no significant difference in the heart weight/body weight ratio between the five groups (3.0±0.4 mg/g in the control group vs. 3.1±0.3 mg/g in DOX treatment and 2.8±0.5, 3.3±0.3 and 2.9±0.3 mg/g in the Ber 5, 10, and 20 mg/kg groups, respectively). ### Effects of Ber on general toxicity in DOX-induced cardiotoxicity A total of 2 rats in the DOX-only treated group succumbed 2 days following DOX injection. No mortality was observed in the Ber+DOX-treated groups. All surviving rats in the DOX-only treated group appeared weak, with scruffy fur and developed a light-yellow tinge. These animals exhibited a significant decrease in body weight compared to the Ber+DOX-treated groups. Notably, these animals developed congestive heart failure, which manifested as marked ascites during necropsy. The hallmark gross pathological changes in DOX-only treated rats were excessive amounts of pericardial, pleural and peritoneal fluids. The maximum ascites volume exhibited by a single animal in DOX-only treated group was 11.5 ml. Compared with DOX alone (2.9±1.8%), the percentage of mean maximum ascites volume to animal body weight were decreased obviously in 5, 10 and 20 mg/kg Ber+DOX-treated groups (1.8±0.9, 1.0±0.1 and 1.4±1.1%: P\<0.05, respectively). The effusion intensity score was severe in 100% of the DOX-only treated animals compared with 20--30% severe scores in the Ber+DOX groups ([Table I](#tI-ol-0-0-8020){ref-type="table"}). ### Ber alleviates DOX-induced oxidative damage Serum CK-MB and CK levels are widely used as clinical markers for oxidative stress, which indicate myocardial injury ([Table II](#tII-ol-0-0-8020){ref-type="table"}). Oxidative stress was evident in the cardiac myocytes of the DOX groups, with significant increases in serum CK and CK-MB levels compared with the control (P\<0.01), and Ber treatment attenuated these increases in a dose-dependent manner (P\<0.05). CK and CK-MB values in the Ber 10 and 20 mg/kg groups were almost equivalent to control levels. Serum and heart tissues of DOX-only treated rats revealed a significant increase in MDA and decreased CAT and SOD contents compared to the control group. MDA, CAT and SOD concentrations recovered markedly in the Ber+DOX groups compared to the DOX-only group. However, 5 mg/kg Ber did not prevent the DOX-induced increase in cardiac enzymes ([Tables II](#tII-ol-0-0-8020){ref-type="table"} and [III](#tIII-ol-0-0-8020){ref-type="table"}). ### Histopathological examination DOX-induced cardiotoxicity was additionally assessed using H&E-stained sections. Heart tissue from the control groups exhibited regular cell distribution and normal myocardium architecture. Histological examination of cardiac sections using H&E stain revealed that DOX induced cardiomyocyte cytoplasmic vacuolization, interstitial oedema and inflammatory cell infiltration. Myocardial lesions were attenuated in the Ber+DOX groups compared to the DOX-alone group ([Figs. 1](#f1-ol-0-0-8020){ref-type="fig"} and [2](#f2-ol-0-0-8020){ref-type="fig"}). ### Effect of Ber on the DOX-induced increase in \[Ca^2+^\]~i~ \[Ca^2+^\]~i~ is critically important in the contraction and relaxation of cardiomyocytes ([@b14-ol-0-0-8020]). DOX-induced alterations in Ca^2+^ homeostasis is one possible mechanism of cardiotoxicity ([@b15-ol-0-0-8020]). Therefore, the effect of DOX on \[Ca^2+^\]~i~ in isolated cardiomyocytes was examined using changes in the fluorescence intensity of cardiomyocytes loaded with Fluo-3 AM. [Fig. 3A](#f3-ol-0-0-8020){ref-type="fig"} demonstrates the effects of 1 µM DOX on the increase in basal Fluo-3 fluorescence in the presence or absence of 1 µM Ber in rat cardiomyocytes. Treatment of cardiomyocytes with 1 µM DOX markedly increased \[Ca^2+^\]~i~ levels, and the cardiomyocytes gradually became 'rounded' into a contracture state in 15 min ([Fig. 3A-a](#f3-ol-0-0-8020){ref-type="fig"}). However, the cardiomyocytes treated with Ber did not become rounded ([Fig. 3A-b](#f3-ol-0-0-8020){ref-type="fig"}). [Fig. 3B](#f3-ol-0-0-8020){ref-type="fig"} summarizes the quantitative data. DOX (1 µM) increased \[Ca^2+^\]~i~ levels to 186.6±18.3% at 5 min and 252.1±11.1% at 15 min DOX, and the extent of the increase in \[Ca^2+^\]~i~ at 15 min was significantly \>5 min (P\<0.05). Pre-treatment with 1 µM Ber significantly attenuated the 1 µM DOX-induced elevation of \[Ca^2+^\]~i~ to 150.6±4.7% at 5 min and 164.2±10.8% at 15 min. These results demonstrated that DOX increased \[Ca^2+^\]~i~ and that Ber attenuated this increase. ### Effects of Ber on DOX-induced ΔΨ~m~ Mitochondria are involved in the maintenance of Ca^2+^ homeostasis primarily due to their capacity to buffer cytosolic Ca^2+^ ([@b14-ol-0-0-8020],[@b15-ol-0-0-8020]). ΔΨ~m~ is the central factor that controls the accumulation of Ca^2+^ within the mitochondrial matrix, cell respiration and ATP synthesis ([@b14-ol-0-0-8020]). The measurement of ΔΨ~m~ is a powerful tool for evaluating mitochondrial damage using the fluorescent probe Rhodamine 123 (Rh-123). Mitochondrial fluorescence intensity correlates quantitatively with changes in ΔΨ~m~ ([@b14-ol-0-0-8020]). Therefore, the effects of DOX on ΔΨ~m~ in rat ventricular myocytes were examined by evaluating ΔΨ~m~ in cardiomyocytes using the Rh-123 staining assay. The corresponding changes in Rh-123 fluorescence intensity in cardiomyocytes were measured 5 and 15 min following perfusion of 1 µM DOX. [Fig. 4A-a](#f4-ol-0-0-8020){ref-type="fig"} indicates the confocal images of ΔΨ~m~ following cell treatment for 15 min with 1 µM DOX. The results revealed that DOX induced a marked decrease in mitochondrial membrane potential. However, the pre-treatment of myocytes with 1 µM Ber significantly restored the DOX-induced reduction in ΔΨ~m~ (P\<0.05, [Fig. 4B](#f4-ol-0-0-8020){ref-type="fig"}) and attenuated the DOX-induced decrease in Rh-123 fluorescence intensity ([Fig. 4A-b](#f4-ol-0-0-8020){ref-type="fig"}). These results demonstrated that Ber prevented the DOX-induced loss of the mitochondrial membrane potential. ### Effects of Ber on DOX-induced \[Ca^2+^\]~m~ overload Ca^2+^ overload in mitochondria may lead to mitochondrial dysfunction and be an important determinant of myocyte toxicity ([@b14-ol-0-0-8020]). Whether Ber attenuated the rise in DOX-induced \[Ca^2+^\]~m~ overload was investigated. [Fig. 5A](#f5-ol-0-0-8020){ref-type="fig"} demonstrates confocal images of rat ventricular myocytes loaded with Rhod-2 in the presence or absence of 1 µM DOX or 1 µM DOX+1 µM Ber. The resting \[Ca^2+^\]~m~ in intact cells was low, as revealed by the dim baseline signal of Rhod-2 fluorescence. Treatment with 1 µM DOX significantly elevated \[Ca^2+^\]~m~ ([Fig. 5A-a](#f5-ol-0-0-8020){ref-type="fig"}). Cells exhibited a high intensity of Rhod-2 fluorescence in the presence of 1 µM DOX, and this fluorescence gradually increased. However, pre-incubation with 1 µM Ber markedly decreased the intensity of the DOX-induced rise in Rhod-2 fluorescence ([Fig. 5A-b](#f5-ol-0-0-8020){ref-type="fig"}). [Fig. 5B](#f5-ol-0-0-8020){ref-type="fig"} suggests that Rhod-2 fluorescence following 15 min of exposure to 1 µM DOX was 303.2±65.1% of the baseline (P\<0.001). Pre-incubation of 1 µM Ber apparently lowered the 1 µM DOX-induced rise in Rhod-2 fluorescence, which was reduced to 119.3±20.2% of the baseline (P\<0.05). These data indicate that the application of Ber prior to DOX treatment prevented the elevation in mitochondrial Ca^2+^ overload induced by DOX alone, and enabled normal mitochondrial function. Discussion ========== DOX is one of the most effective anti-tumour antibiotics, and novel methods to reduce or prevent the cardiotoxic side effects of DOX are expected. DOX-induced cardiotoxicity manifests as acute effects (atrial and ventricular dysrhythmias) that vary from transient electrocardiographic abnormalities to dose-dependent cardiomyopathy and congestive heart failure ([@b15-ol-0-0-8020],[@b16-ol-0-0-8020]). A previous study of DOX cardiotoxicity demonstrated that the single high-dose model is widely used since it manifests certain characteristics of DOX-induced cardiotoxicity, which is equivalent to a single high-dose injection in patients with cancer ([@b17-ol-0-0-8020]). Ber hydrochloride is an effective antioxidant and free radical scavenger that prevents ROS formation and exerts protective effects on cardiac, hepatic and renal functions ([@b18-ol-0-0-8020],[@b19-ol-0-0-8020]). Our previous study also demonstrated that Ber exhibited a protective effect on DOX-induced acute hepatorenal toxicity in rats ([@b12-ol-0-0-8020]). Whether Ber exerts a protective effect against DOX-induced cardiac injury is unknown. A number of cardioprotective agents, including dexrazoxane, amifostine and probucol, are of limited value in counteracting DOX cardiotoxicity and improving its clinical utility ([@b20-ol-0-0-8020],[@b21-ol-0-0-8020]). However, these scavengers exhibit clinical disadvantages, including a diminished anti-tumour effect and failure to protect the heart against DOX-induced injury in clinical settings. Tong *et al* ([@b22-ol-0-0-8020]) revealed that Ber treatment potentiated the sensitivity of cancer cells to DOX. Patil *et al* ([@b23-ol-0-0-8020]) also demonstrated that Ber suppressed tumour growth via the induction of apoptosis and cell cycle arrest in cancer cells. Taken together, based on these data, we hypothesized that combining DOX with Ber as a novel strategy for tumour therapy would not only increase the effect of DOX, but also prevent the cardiotoxicity induced by DOX. The present study investigated the mechanism of the protective effects of Ber against DOX-induced cardiotoxicity in rats. A rat model of DOX-induced cardiotoxicity was developed *in vivo* and *in vitro* to determine the potential protective action of Ber against the cardiotoxic effects of DOX. The doses of Ber used *in vivo* were those that has exhibited the maximum cardio-protective effect in preliminary optimisation experiments (data not shown). Also, this result is consistent with a previous study that indicated that Ber demonstrated significant protective effects on heart tissue at 10 mg/kg in rats ([@b24-ol-0-0-8020]). The data from the present study demonstrated that DOX administration was accompanied by a high mortality compared with the control group. Surviving animals suffered from an excessive degree of pericardial, pleural and peritoneal effusion, which indicated severe cardiac injury. The ability of Ber to protect against DOX-induced high mortality and effusion intensity score was considered an early sign of cardio-protection ([@b25-ol-0-0-8020]). Ber also attenuated DOX-induced histopathological and ultrastructural deteriorations. These data indicate that Ber is a potential protective agent against DOX injury. Oxidative damage and antioxidant deficiencies in cardiomyocytes are widely implicated as a primary cause for DOX-induced cardiac toxicity ([@b1-ol-0-0-8020],[@b15-ol-0-0-8020]), and these factors were utilized in the present study to examine the oxidant/antioxidant status of the rats. We used the single high-dose model and investigated whether Ber protected the heart from acute DOX toxicity. The current data demonstrated that cardiac and serum levels of MDA, CK and CK-MB were significantly elevated, and that the activities of the cardiac antioxidant enzymes CAT and SOD were significantly reduced, following DOX administration compared to the control group. These data clearly indicate a status of overt oxidative stress. Ber significantly decreased the DOX-associated elevation of serum MDA, CK and CK-MB activities, which are classical biomarkers of cardiotoxicity ([@b4-ol-0-0-8020]). It is well-known that the expression levels of B-type natriuretic peptide (BNP) and troponins are specific and sensitive indicators for cardiac damage ([@b20-ol-0-0-8020]). However, in the present study, severe cardiac damage induced by doxorubicin was observed, manifested by marked ascites and notable symptom of congestive heart failure. Since it was sufficient to analyse the cardiotoxicity by detecting CK and CK-MB levels in serum, it was not necessary to measure the expression levels of BNP and troponins. Nevertheless, means of measuring the expressions of BDP and troponins will be included in future studies, for full clarification of the protective mechanism of Ber on DOX-induced cardiotoxicity. The activities of CAT and SOD were significantly elevated in BER+DOX-treated groups compared with DOX-only administration in rats. These data are consistent with the current understanding of elevated cardiac lipid peroxidation accompanied by deteriorating antioxidant status was evident in the DOX-only treated rats. Ber pre-treatment significantly alleviated the oxidative stress in the DOX group, and Ber therapy completely prevented the biochemical and histopathological deteriorations caused by DOX These results suggest that Ber therapy during DOX treatment for cancer significantly protected the heart against DOX-induced injury. These data are in agreement with a study by Chen *et al* ([@b12-ol-0-0-8020]) and Hao *et al* ([@b26-ol-0-0-8020]). The initial damage of DOX-induced cardiotoxicity is likely oxidative in nature as it undergoes a one-electron reduction, resulting in the corresponding semiquinone, to form free radicals and superoxide radicals ([@b27-ol-0-0-8020]). Oxidative damage to the cardiac mitochondria and cardiomyocyte is a cornerstone of DOX-induced cardiotoxicity, but a previous study suggested calcium overload as an additional important mechanism of DOX-induced cardiotoxicity ([@b28-ol-0-0-8020]). Previous studies have demonstrated that DOX opened sarcoplasmic reticulum calcium release channels and increased the maximal amount of calcium release ([@b29-ol-0-0-8020],[@b30-ol-0-0-8020]), inhibited Na^+^-Ca^2+^ exchange ([@b31-ol-0-0-8020]) or activated the L-type cardiac calcium channel ([@b32-ol-0-0-8020]). DOX may lead to calcium overload in cardiac cells, causing inadequate contraction and impairing mitochondrial calcium homeostasis, which alters energy metabolism and the generation of ROS ([@b33-ol-0-0-8020]) and may trigger the calcium-dependent mitochondrial permeability transition via opening of the permeability transition pores ([@b14-ol-0-0-8020],[@b15-ol-0-0-8020]). The opening of these pores induces the release of cytochrome c, which is a critical step for apoptosis ([@b34-ol-0-0-8020],[@b35-ol-0-0-8020]). Mitochondria are key targets for anthracycline cardiotoxicity ([@b36-ol-0-0-8020],[@b37-ol-0-0-8020]), and may also be crucial for effective cardio-protection. The decrease in ΔΨ~m~ reduced the capacity for Ca^2+^ influx into mitochondria, as ΔΨ~m~ primarily drives mitochondrial calcium uptake ([@b38-ol-0-0-8020]). Therefore, the present study investigated intracellular calcium and mitochondrial calcium in cardiac myocytes following DOX treatment, and examined whether the protective effect of Ber on DOX-induced Ca^2+^ overload was associated with ΔΨ~m~ depolarization. The results demonstrated that the high local \[Ca^2+^\]~i~ level produced by DOX caused mitochondrial Ca^2+^ overload and a decrease in ΔΨ~m~, which may alter cardiac metabolism and lead to cardiomyocyte death. Incubation with Ber for 1 h significantly reduced mitochondrial Ca^2+^ overload and significantly elevated cardiac energy metabolism. Ber treatment significantly suppressed the DOX-induced increase in \[Ca^2+^\]~i~. These results demonstrate that Ber inhibited the DOX-induced acute modifications in calcium homeostasis in cardiomyocytes, and suggest a protective role of Ber against DOX-induced cardiotoxicity. A previous study demonstrated that Ber may also attenuate DOX-induced cardiomyocyte apoptosis by inhibiting caspase-3 activation, adenosine 5′-monophosphate activated protein kinase α and tumour protein 53 phosphorylation ([@b39-ol-0-0-8020]). Ber is a potential candidate agent for co-administration with DOX to ameliorate its cardiotoxicity. Additional studies are required to thoroughly evaluate the potential protective effect of Ber in DOX-induced cardiotoxicity. The authors would like to thank Professor Lian-Shan Zhang (Department of Pathology, Hebei Medical University, Shijiazhuang, China) for her help in histopathological analysis of the cardiac tissues. Funding ======= The present study was supported by the Natural Science Foundation of China (grant nos. 81273600, 81302773 and 81773828) and the Natural Science Foundation of Hebei Province (grant nos. C2011206145, H2014206319 and H2018206297). Availability of data and materials ================================== The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Authors\' contributions ======================= SWS and CX conceived and designed the study. YZ, CX, ZXW, YZW, XYC, HCG, KRX, KXW, and PJ performed the experiments. CX, YZ, YZW, and SWS wrote the paper. SWS and CX reviewed and edited the manuscript. All authors read and approved the manuscript. Ethics approval and consent to participate ========================================== All experiments were performed in compliance with the Guide for the Care and Use of Laboratory Animals and were reviewed and approved by the Ethics Committee for the Use of Experimental Animals at Hebei Medical University (Shijiazhuang, China). Consent for publication ======================= Not applicable. Competing interests =================== The authors declare that they have no competing interests. DOX : doxorubicin Ber : berberine ROS : reactive oxygen radical species ΔΨ~m~ : mitochondrial membrane potential CK : creatine kinase CK-MB : creatine kinase isoenzyme MDA : malondialdehyde SOD : superoxide dismutase CAT : catalase {#f1-ol-0-0-8020} {#f2-ol-0-0-8020} ![Effect of Ber on DOX-induced \[Ca^2+^\]~i~ elevation in isolated rat ventricular myocytes. (A) (a) Representative image of changes in fluorescence images of \[Ca^2+^\]~i~ in single isolated rat ventricular myocytes following Dox exposure, recorded at 15 min using laser confocal microscopy. (b) Effects of Ber incubation on DOX-induced \[Ca^2+^\]~i~ levels (magnification, ×40). (B) Effects of Ber incubation on DOX-induced \[Ca^2+^\]~i~ elevation. Changes in \[Ca^2+^\]~i~ are represented as the ratio of FI-FI~0~/FI~0~. (n=7 in each group) (^∆∆^P\<0.01 vs. control, ^\#\#^P\<0.01 and ^\#^P\<0.05). DOX, doxorubicin; Ber, berberine; con, control group; FI~0~, control; FI, the drug-treated groups; \[Ca^2+^\]~i~, cytosolic calcium concentration.](ol-15-04-5721-g02){#f3-ol-0-0-8020} {#f4-ol-0-0-8020} ![Effect of Ber on mitochondrial Ca^2+^ (\[Ca^2+^\]~m~) overload in Rhod 2-AM-loaded rat ventricular myocytes. (A) Confocal images of Rhod 2-AM fluorescence intensity from a single myocyte (magnification, ×40). Cells were treated with DOX for 15 min as follows: (a) DOX-induced \[Ca^2+^\]~m~ elevation and (b) Effects of Ber (1 µM) on \[Ca^2+^\]~m~ caused by DOX. (B) The effect of Ber on DOX-induced \[Ca^2+^\]~m~ overload at 5 min and 15 min on changes in relative Rhod 2-AM fluorescence intensity. (n=6 in each group) (^∆∆^P\<0.01 vs. control, ^\#^P\<0.05). Con, control; DOX, doxorubicin; Ber, berberine; Rhod-2-AM, Rhod-2-acetoxymethyl; \[Ca^2+^\]~i~, cytosolic calcium concentration.](ol-15-04-5721-g04){#f5-ol-0-0-8020} ###### Effect of Ber on DOX-induced abdominal, pleural and pericardial effusion intensity scores in surviving rats. Effusion intensity score ---------------- ---- -------------------------- ----- --- ---- --- ------ --- --------------------------------------------------------------------------------------------- Con 10 10 100 0 0 0 0 0 0 DOX 8 0 0 0 0 0 0 8 100^[a](#tfn2-ol-0-0-8020){ref-type="table-fn"}^ Ber (5 mg/kg) 10 0 0 4 40 3 30 3 30^[a](#tfn2-ol-0-0-8020){ref-type="table-fn"},[b](#tfn3-ol-0-0-8020){ref-type="table-fn"}^ Ber (10 mg/kg) 10 1 10 4 40 2 20 3 30^[a](#tfn2-ol-0-0-8020){ref-type="table-fn"},[b](#tfn3-ol-0-0-8020){ref-type="table-fn"}^ Ber (20 mg/kg) 10 1 10 5 50 2 33.3 2 20^[a](#tfn2-ol-0-0-8020){ref-type="table-fn"},[b](#tfn3-ol-0-0-8020){ref-type="table-fn"}^ Con, control; DOX, doxorubicin; Ber, berberine. 0, none; +, mild; ++, moderate; +++, severe. Statistical evaluation was performed using the χ^2^-test. P\<0.01 vs. Con. P\<0.01 vs. DOX. ###### Serum levels of CAT, SOD, MDA, CK, and CK-MB activities following acute DOX intoxication and protective activity by Ber. Group n CAT (U/ml) SOD (U/ml) MDA (mmol/ml) CK (U/l) CK-MB (U/l) ---------------- ----- ---------------------------------------------------------- ----------------------------------------------------------------------------------------------------- ---------------------------------------------------------- -------------------------------------------------------- -------------------------------------------------------- Con 10 29.72±11.68 41.35±3.60 40.42±6.73 727±261 1,843±659 DOX   8 19.79±4.53^[a](#tfn5-ol-0-0-8020){ref-type="table-fn"}^ 23.41±12.10^[b](#tfn6-ol-0-0-8020){ref-type="table-fn"}^ 55.37±6.20^[b](#tfn6-ol-0-0-8020){ref-type="table-fn"}^ 1,426±540^[b](#tfn6-ol-0-0-8020){ref-type="table-fn"}^ 3,404±939^[b](#tfn6-ol-0-0-8020){ref-type="table-fn"}^ Ber (5 mg/kg) 10 28.53±7.72^[d](#tfn8-ol-0-0-8020){ref-type="table-fn"}^ 36.44±9.65^[c](#tfn7-ol-0-0-8020){ref-type="table-fn"}^ 46.93±7.90^[c](#tfn7-ol-0-0-8020){ref-type="table-fn"}^ 1,069±366^[a](#tfn5-ol-0-0-8020){ref-type="table-fn"}^ 2,694±571^[a](#tfn5-ol-0-0-8020){ref-type="table-fn"}^ Ber (10 mg/kg) 10 25.70±5.16^[c](#tfn7-ol-0-0-8020){ref-type="table-fn"}^ 44.76±3.02^[a](#tfn5-ol-0-0-8020){ref-type="table-fn"},[d](#tfn8-ol-0-0-8020){ref-type="table-fn"}^ 47.98±8.81^[c](#tfn7-ol-0-0-8020){ref-type="table-fn"}^ 991±271^[c](#tfn7-ol-0-0-8020){ref-type="table-fn"}^ 2,395±618^[c](#tfn7-ol-0-0-8020){ref-type="table-fn"}^ Ber (20 mg/kg) 10 27.51±10.20^[c](#tfn7-ol-0-0-8020){ref-type="table-fn"}^ 41.51±3.77^[d](#tfn8-ol-0-0-8020){ref-type="table-fn"}^ 40.47±18.21^[c](#tfn7-ol-0-0-8020){ref-type="table-fn"}^ 792±378^[c](#tfn7-ol-0-0-8020){ref-type="table-fn"}^ 2,123±865^[c](#tfn7-ol-0-0-8020){ref-type="table-fn"}^ Data are presented as mean ± standard deviation. Con, control; DOX, doxorubicin; Ber, berberine; CAT, catalase; MDA, malondialdehyde; SOD, superoxide dismutase; CK, creatine kinase; CK-MB, creatine kinase isoenzyme. P\<0.05 P\<0.01 vs. Con P\<0.05 P\<0.01 vs. DOX. ###### Effect of Ber on peroxidative alterations induced by acute DOX intoxication in rat cardiac tissue. Group n CAT (U/mg prot) SOD (U/mg prot) MDA (nmol/mg prot) ---------------- ----- --------------------------------------------------------- ---------------------------------------------------------- ------------------------------------------------------------------------------------------------------- Con 10 52.0±4.5 54.3±8.5 28.73±3.94 DOX   8 37.9±13.4^[b](#tfn11-ol-0-0-8020){ref-type="table-fn"}^ 42.1±6.2^[b](#tfn11-ol-0-0-8020){ref-type="table-fn"}^ 42.59±4.93^[b](#tfn11-ol-0-0-8020){ref-type="table-fn"}^ Ber (5 mg/kg) 10 48.3±9.9 46.15±7.25^[a](#tfn10-ol-0-0-8020){ref-type="table-fn"}^ 42.50±7.22^[b](#tfn11-ol-0-0-8020){ref-type="table-fn"}^ Ber (10 mg/kg) 10 52.6±1.7^[d](#tfn13-ol-0-0-8020){ref-type="table-fn"}^ 49.32±7.12^[c](#tfn12-ol-0-0-8020){ref-type="table-fn"}^ 35.41±5.38^[a](#tfn10-ol-0-0-8020){ref-type="table-fn"},[d](#tfn13-ol-0-0-8020){ref-type="table-fn"}^ Ber (20 mg/kg) 10 52.9±2.6^[c](#tfn12-ol-0-0-8020){ref-type="table-fn"}^ 50.97±6.77^[d](#tfn13-ol-0-0-8020){ref-type="table-fn"}^ 33.00±1.15^[b](#tfn11-ol-0-0-8020){ref-type="table-fn"},[d](#tfn13-ol-0-0-8020){ref-type="table-fn"}^ Con, control; DOX, doxorubicin; Ber, berberine; CAT, catalase; MDA, malondialdehyde; SOD, superoxide dismutase; mg prot, mg protein. P\<0.05 P\<0.01 vs. Con P\<0.05 P\<0.01 vs. DOX. Data are presented as mean ± standard deviation.

Dear Sir, Dengue is a viral infection caused by an arbovirus of the Flaviviridae family. Epidemics frequently occur in tropical and subtropical countries, and it is a grave public health problem. It may present in many forms. The most common classical form of dengue causes fever, myalgias, headache, nausea, vomiting, diarrhea, arthralgia, and rash lasting for a week. Dengue hemorrhagic fever is a severe form of the disease that can progress to severe hemorrhagic manifestations, hypovolemic shock, and death. Various neurologic complications of dengue viral infection have been reported, including central and peripheral nervous system involvement. Encephalopathy, encephalitis, seizures, mononeuropathy, polyneuropathy, and Guillain--Barré or Miller--Fisher syndromes have been seen. However, spinal cord involvement associated with dengue viral infection has been rarely mentioned.\[[@ref1][@ref2]\] An 8-month-old male child presented with fever for 3 days. Initial complete blood counts showed severe anemia and thrombocytopenia, and the blood sample was positive for IgM antibody of dengue. Five days later, the child developed areflexia along with urinary retention and bowel incontinence and severe respiratory distress. On examination, the pupils were sluggishly reactive, and there was no response to pain. There were generalized petechiae in both lower limbs. Other systems were completely normal. Other laboratory parameters showed mildly raised liver enzymes and normal blood sugar, electrolytes, and kidney function tests. Serologic testing for malaria, typhoid, HIV, and syphilis were negative. Due to the acute presentation of the bowel and bladder dysfunction, magnetic resonance imaging of the spinal cord was done, which showed diffuse hyperintensity on T2-Weighted and short tau inversion recovery images extending from C4 level up to the termination of spinal cord \[[Figure 1a](#F1){ref-type="fig"}\]. There was also formation of syrinx above the C4 level \[[Figure 1b](#F1){ref-type="fig"}\]. On postcontrast images, there was mild patchy enhancement of the spinal cord at C4 and C5 levels \[[Figure 1c](#F1){ref-type="fig"}\]. {#F1} On the basis of imaging findings, a diagnosis of diffuse acute transverse myelitis with syringohydromyelia in association with dengue viral infection was made. Due to severe respiratory distress, the child was given ventilatory support, and intravenous (IV) methylprednisolone was started. After 6 days of ventilatory support, the child improved. Hence, gradually, he was extubated and after that he recovered well. In the past few years, numerous neurological complications related to dengue fever have been reported, and these can lead to a significant morbidity and mortality. Neurological complications occur in 0.5--6% of the cases with dengue fever.\[[@ref3]\] Dengue encephalopathy could be caused by cerebral edema, cerebral hemorrhage, hyponatremia, hypoxia, renal and hepatic insult. Acute myelitis as a consequence of dengue fever is a rare phenomenon and has been published through anecdotal case reports. Transverse myelitis is a clinical syndrome, which can be associated with a number of different conditions. It is, therefore, necessary to identify a direct infection, a systemic disease, or an autoimmune (postinfectious or postvaccinal) process.\[[@ref4]\] The clinical picture includes partial or complete paraplegia or quadriplegia, decrease or loss of deep reflexes, sensory impairment, and varying degrees of bladder and bowel disturbance. Usually, the full-blown disease is reached within 4 weeks after onset, but in most cases, the peak occurs in the 1^st^ week with the level of involvement set at the onset.\[[@ref5]\] In a few patients, however, the disease has an ascending course with the risk of asphyxia when the upper cervical segments (C3--C5) are involved. In the present case, the onset was sudden and the upper cervical spinal cord was immediately involved. Early diagnosis of the condition and early initiation of corticosteroid treatment are very important as it results in rapid recovery.\[[@ref4]\] Timely diagnosis and treatment of the condition with IV corticosteroids decreases mortality and morbidity and favors good outcome as seen in our case. Financial support and sponsorship {#sec2-1} ================================= Nil. Conflicts of interest {#sec2-2} ===================== There are no conflicts of interest.

1. Introduction {#sec1-foods-09-00784} =============== Whey is the main byproduct obtained from the manufacture of cheese, yogurt, and milk protein concentrates \[[@B1-foods-09-00784]\]. It is the yellowish liquid separated from curds, and it is mainly made of water (\~94%), lactose (\~5%), proteins (\~1%), minerals (\~1%), and milk fat (\~0.5%) \[[@B2-foods-09-00784]\]. As a byproduct, whey is concentrated and further fractionated to produce a wide array of products and ingredients with food and pharmaceutical applications. Examples of ingredients derived from whey are milk serum protein concentrate, whey powder, whey protein concentrate, whey protein hydrolysate, and whey protein isolate. Manufacturing details on the production of milk protein ingredients can be found elsewhere \[[@B3-foods-09-00784],[@B4-foods-09-00784]\]. Over the last few years, the number of food formulations containing milk proteins has considerably increased due to the functional and nutritional properties of milk proteins \[[@B5-foods-09-00784]\]. Nowadays, products such as sport drinks, desserts, beverages, snacks, dietary, and weight management supplements are formulated with milk proteins. Overall, the manufacture of concentrates of milk proteins involves numerous unit operations, such as thermal treatment, selective fractionation, concentration, and drying. Selective fractionation by means of membrane technology is the key processing step in which the protein fraction is sequentially separated from the whey stream. The separation is performed by multiple filtration stages, where the proteins are concentrated in the retentate, while lactose and minerals are concentrated in permeate stream. The byproduct derived from the production of milk proteins is known as whey permeate, and it contains a considerable amount of lactose (75--80% on dry basis). Lactose (LA) presents technological challenges, such as poor solubility and low sweetness index, as well as malabsorption by a certain population, which has limited its further utilization \[[@B6-foods-09-00784]\]. Consequently, there is a considerable industrial interest to further utilize LA as a feedstock for the production of lactose-based ingredients \[[@B7-foods-09-00784],[@B8-foods-09-00784]\]. As a feedstock, lactose has the potential to undergo different reactions leading to the formation of value-added compounds. [Figure 1](#foods-09-00784-f001){ref-type="fig"} illustrates the simultaneous formation of lactulose (LAU) and lactobionic acid (LBA), two of the most versatile lactose-derived ingredients, with applications in food, dairy, and pharmaceutical formulations. Lactulose (4-*O*-β-[d]{.smallcaps}-galactopyranosyl-[d]{.smallcaps}-fructofuranose), an isomer of LA, is a disaccharide made of galactose and fructose. It is used in the treatment of hepatic encephalopathy and chronic constipation \[[@B9-foods-09-00784]\]. Industrially, LAU is synthesized by either homogenous catalysis under alkaline conditions, or by enzymatic synthesis using glycosyltransferases and glycosidases \[[@B10-foods-09-00784]\]. Reaction schemes and conditions for LAU synthesis can be found elsewhere \[[@B11-foods-09-00784]\]. On the other hand, LBA (4-*O*-β-galactopyranosyl-[d]{.smallcaps}-gluconic acid) is an aldonic acid or sugar-acid made up of galactose linked to a gluconic acid \[[@B12-foods-09-00784]\]. The LBA molecule has various functionalities, such as antioxidant, chelating, and humectant, arising from the high number of hydroxyl groups. The partial oxidation of LA results in the formation of LBA, where the aldehyde group is converted into its corresponding carbonyl group via selective oxidation \[[@B13-foods-09-00784]\]. Most of the LBA commercialized in the world is produced via 1) biocatalytic conversion using dehydrogenase or oxidoreductase, or 2) heterogeneous catalysis using Pd, Pt, or Ru as insoluble catalysts. The manufacturing methods of LBA can be found elsewhere \[[@B12-foods-09-00784]\]. One common strategy employed in the industrial production of lactose-derived ingredients is the use of purified lactose (monohydrate or pharmaceutical grade) as a feedstock, which minimizes the formation of secondary products. Fundamentally, such a strategy aims to produce a single compound via multi-step flow, involving the formation, separation, and isolation of secondary and primary products. Alternatively, the conversion of lactose through one-pot synthesis represents an innovative utilization of lactose. One-pot conversion involves the synthesis of a mixture of compounds that can be used with minimal separation in the final formulation. This has been exemplified by our group, where a sweetening syrup from aqueous lactose was produced via one-pot synthesis (enzymatic hydrolysis followed by catalytic isomerization) \[[@B14-foods-09-00784],[@B15-foods-09-00784]\]. Similarly, Zaccheria \[[@B16-foods-09-00784]\] produced a mixture of sorbitol and dulcitol from lactose via one-pot catalytic conversion. Gallezot \[[@B17-foods-09-00784]\] conceptualized and exemplified the production of a pool of molecules from biomass via one-pot catalytic synthesis. Under the one-pot approach, the high concentration of lactose present in whey permeate can be used as an inexpensive feedstock for a variety of chemical modifications, such as hydrogenation, isomerization, and oxidation. During the catalytic oxidation of lactose, the first products formed are LBA and LAU, followed by 2-keto-lactobionic acid as a secondary product \[[@B18-foods-09-00784]\]. Preliminary results showed that increasing the oxygen pressure improved the yield of LAU during the oxidation of lactose over Ru/C \[[@B19-foods-09-00784]\]. It should be highlighted that the production of LAU and LBA was evaluated using lactose monohydrate as a feedstock. Nevertheless, these results present an opportunity for the simultaneous production of LAU and LBA directly from whey permeate. This work aimed at producing LAU and LBA via a one-pot catalytic oxidation from sweet and acid whey permeate. 2. Materials and Methods {#sec2-foods-09-00784} ======================== 2.1. Materials {#sec2dot1-foods-09-00784} -------------- Firstly, α-[d]{.smallcaps}-Lactose monohydrate (98%, Acros Organics, Fair Lawn, NJ, USA), lactulose (99%, Alfa Aesar, Haverhill, MA, USA), lactobionic acid (97%, Sigma-Aldrich, St. Louis, MO, USA), D-(+)-glucose (99%, Sigma-Aldrich), D-(+)-galactose (99%, Sigma-Aldrich), D-gluconic acid sodium salt (99%, Sigma-Aldrich), pyruvic acid (98%, Sigma-Aldrich), L-(+)-lactic acid (85%, Sigma-Aldrich), formic acid (99%, Fisher Scientific, Waltham, MA, USA), citric acid (99%, Sigma-Aldrich), and orotic acid (98%, Sigma-Aldrich) were purchased from commercial suppliers. Reduced ruthenium supported on activated carbon (5% Ru/C, Alfa Aesar) was purchased from commercial suppliers, and it was used without further preparation. Sweet whey permeate (SWP) was obtained from a regional cheese factory (Valley Queen Co., Milbank, SD, USA), while the acid whey permeate (AWP) was obtained from a Greek yogurt plant (Chobani Co., Twin Falls, ID, USA). 2.2. Preparation of Samples {#sec2dot2-foods-09-00784} --------------------------- A 10% wt./wt. solution of lactose (LAS) was prepared by dissolving D-lactose monohydrate in distilled water. Samples of SWP and AWP were analyzed for pH, total protein, fat content, total solids, and total non-volatile solids. The pH was measured in 10 mL of the sample using a triode epoxy electrode (Orion Versa Star Pro, Thermo Fisher Scientific). The protein content was determined by the Kjeldahl method, while the fat content was measured using the Mojonnier fat extraction, following the guidelines of the Association of Official Analytical Chemists (AOAC) methods 989.05 and 991.20, respectively \[[@B20-foods-09-00784]\]. Total solids (TS) and total non-volatile solids (TNVS) were determined using the AOAC method 990.20 \[[@B20-foods-09-00784]\]. [Table 1](#foods-09-00784-t001){ref-type="table"} shows the compositional characteristics of SWP and AWP. 2.3. Experimental Treatments {#sec2dot3-foods-09-00784} ---------------------------- A set of experiments was first conducted to evaluate the influence of pressure (15, 40, 60, and 80 bar) and temperature (50, 60, 70, and 80 °C) on the conversion rate of lactose in a 10% wt./wt. solution of lactose. Initial rates (r~o~) of lactose conversion were calculated using Equation (1):$$\left. {r_{o} = \frac{d\left\lbrack {La} \right\rbrack}{dt}} \right|_{t = 0}$$ At selected conditions, the second set of experiments was conducted to evaluate the formation of LAU and LBA directly from sweet whey permeate (SWP), and acid whey permeate (AWP). 2.4. One-Pot Synthesis {#sec2dot4-foods-09-00784} ---------------------- The one-pot conversion of LA was carried out in a continuous stirred-tank reactor (BR-300, Berghof Products & Instruments, Berghof, Germany). [Figure 2](#foods-09-00784-f002){ref-type="fig"} schematically represents the one-pot conversion of lactose directly from whey permeate. Features and characteristics of the reactor, as well as the working conditions, have been explained thoroughly in previous work \[[@B21-foods-09-00784]\]. Briefly, the reactor vessel was loaded with 400 mL of either LAS, SWP, or AWP, containing 0.5 g/L of Ru/C catalyst. Then, the vessel was lidded, clamped, and heated up to the target temperature. During heating, the vessel was pressurized with compressed air as an oxidizing agent (99.99% purity, Praxair, Sioux Falls, SD, USA). The initial time of the catalytic reaction was considered when the solution reached the target temperature and pressure. Once the working conditions were obtained, the reaction was monitored over 210 min, withdrawing samples (\~15 mL) every 30 min and stored at −20 °C until further analysis. Samples were withdrawn from the sampling port ((6) in [Figure 2](#foods-09-00784-f002){ref-type="fig"}), consisting of a set of needle valves that allowed the subtraction of samples without losing pressure inside the reactor. 2.5. Quantification of Reaction Products {#sec2dot5-foods-09-00784} ---------------------------------------- The reaction products derived were quantified by liquid chromatography-mass spectrophotometry (LC-MS). Firstly, samples withdrawn from the reactor were filtered through a 20-25 μm filter (NO 541 Hardened Ashless, and 110 mm diameter, WhatmanTM Co., Marlborough, MA, USA) to remove the spent catalyst. Then, the filtered solution was diluted 10-fold with HPLC grade water, followed by centrifugation using ultracentrifugation filters (Amicon^®^ filters, Merc Millipore, Billerica, MA, USA) at 9000 rpm for 10 min (accuSpin Micro 17R, Fisher Scientific, Waltham, MA, USA). After centrifugation, the precipitate was discarded, while the supernatant was used for analysis. An aliquot of 10 µL from the supernatant was injected into a Shimadzu LC system (LC-20AD, Shimadzu Corp, Kyoto, Japan), combined with a Qtrap 5500 triple quadrupole mass spectrometer (AB Sciex, Foster City, CA, USA). Analytes were eluded by means of an HPX-87C column (250x4.0 mm, Bio-Rad Aminex, Hercules, CA, USA) operated at 80 °C with a mobile phase consisted of acetonitrile: water (20:80 v/v), at a flow rate of 0.2 mL/min. Mass spectrophotometer analysis was operated in negative mode at a temperature of 500 °C, a curtain gas of 30 psi, an ion source gas of 15 psi for nebulizer (GS1), and heater (GS2). Samples were quantified according to HPLC-grade analytical standards. 2.6. Quantification of Organic Acids {#sec2dot6-foods-09-00784} ------------------------------------ The organic acids formed during the catalytic oxidation of LA were analyzed by HPLC, according to the methodology reported by Zeppa et al. \[[@B22-foods-09-00784]\], with some modifications. An HPLC instrument (Beckman Coulter, Inc., Fullerton, CA, USA) equipped with a solvent delivery module (System Gold^®^ 125), a multichannel wavelength scanning detector (190--600 nm, System Gold 168 detector), and a refractive index detector (RI-2031, Jasco Corporation, Hachioji, Japan) were used for the analysis. The separation of organic acids was performed using an ion exclusion column (ROA-Organic Acid H+ 8%, Phenomenex, Torrance, CA, USA) heated at 60 °C. The mobile phase was a sulfuric acid solution (0.013 N), at a flow rate of 0.5 mL min^−1^. 2.7. Data Analysis {#sec2dot7-foods-09-00784} ------------------ Lactose conversion was expressed as a percentage of converted LA into its derivatives, according to Equation (2). $$C_{LA} = \frac{\left\lbrack {LA_{o}} \right\rbrack - \left\lbrack {LA_{t}} \right\rbrack}{\left\lbrack {LA_{o}} \right\rbrack} \times 100$$ where $C_{LA}$---conversion of lactose (%), $\left\lbrack {LA_{o}} \right\rbrack$---initial concentration of lactose (mM), and $\left\lbrack {LA_{t}} \right\rbrack$---concentration of lactose at a given reaction time. The formation of LAU and LBA was determined as the product yield ($Y_{i}$) according to Equation (3). $$Y_{i} = \frac{Concentration~of~target~product}{Initial~concentration~of~lactose} \times 100$$ Experimental runs were conducted in duplicate, and all the figures were made with SigmaPlot software V11 for Windows (Systat Software, Inc., Chicago, IL, USA). 3. Results and Discussion {#sec3-foods-09-00784} ========================= 3.1. Reaction Conditions {#sec3dot1-foods-09-00784} ------------------------ [Figure 3](#foods-09-00784-f003){ref-type="fig"} shows a graphical representation of the temperature-pressure history during the conversion of LA. Each experimental treatment started with the heading time, which includes loading, pressurization (arrow (1)), and heating of the vessel. Start of the reaction time (arrow (2)) was considered when both the target temperature and pressure were achieved. Afterward, samples (\~15 mL) were withdrawn for analysis at interval times of 30, 90, 150, and 210 min. The end of the reaction time (arrow (3)) marked the beginning of the cooling and depressurization (arrow (4)). Similar characterization of the temperature-pressure history was reported elsewhere \[[@B21-foods-09-00784]\]. A detailed description of the main variables involved during the one-pot conversion allows one to meaningfully interpret the formation of reaction products, and facilitates comparison with the literature. 3.2. Initial Rates {#sec3dot2-foods-09-00784} ------------------ The initial rate of LA conversion is shown in [Figure 4](#foods-09-00784-f004){ref-type="fig"}. Pressure influence on the initial rates was evaluated at a constant temperature (60 °C, [Figure 4](#foods-09-00784-f004){ref-type="fig"}a), where the initial rates increased with the pressure until it reached a maximum value of 1.95 ± 0.09 mmol/min at 60 bar, while a further increment of pressure resulted in lower values of initial rates (0.70 ± 0.06 mmol/min at 80 bar). This finding is important from an economic and operational perspective, since the application of high pressure will increase the operational costs and be difficult to operate. Influence of temperature on the initial rates of lactose conversion was evaluated at a constant pressure (60 bar, [Figure 4](#foods-09-00784-f004){ref-type="fig"}b). The highest values of initial rates were obtained at 60 °C, and further elevation of the temperature did not significantly increase the initial rates. Overall, the magnitude of the initial rates was higher by increasing the temperature than by the increment of the pressure. This observation is in agreement with the Arrhenius and Eyring theory \[[@B23-foods-09-00784]\]. 3.3. Conversion and Yield {#sec3dot3-foods-09-00784} ------------------------- [Figure 5](#foods-09-00784-f005){ref-type="fig"} shows the kinetic curves of lactose conversion for SWP and AWP. During the first 30 min of reaction, about 30 and 24% of lactose was converted for SWP and AWP, respectively. After 210 min, the conversion of lactose reached a maximum value of about 37 and 34% for SWP and AWP, respectively. Mäki-Arvela et al. \[[@B24-foods-09-00784]\] reported conversion values of about 90% during the oxidation of LA (0.86 M) under alkaline conditions (pH = 8), using Pd/C as a catalyst. The difference in the conversion values can be explained by the deactivation of the catalyst and the pH of the solution. Indeed, Mäki-Arvela et al. \[[@B25-foods-09-00784]\] reported that Ru/C rapidly underwent deactivation during the oxidation of lactose monohydrate solution (0.86 M), achieving conversion values of 28--32%. In SWP and AWP, the presence of minerals and residual proteins seems to impact the activity of the catalyst negatively. However, the deactivation of Ru/C due to minerals and residual protein needs further experimental evidence. [Figure 5](#foods-09-00784-f005){ref-type="fig"}a,b showed the formation curves for LAU and LBA corresponding to SWP and AWP, respectively. In untreated samples of SWP and AWP, the concentration of LAU and LBA was below the detection limit. An increment in the yield values of LAU was observed within the first 30 min of reaction, 22.99 ± 0.81 and 17.29 ± 0.96% for SWP and AWP, respectively. The relatively high yield values for SWP is not surprising, since the pH of the SWP (6.23 ± 0.01) favored the formation of LAU. Seo et al. \[[@B26-foods-09-00784]\] reported yield values of LAU of 29% (90 °C and 20 min of reaction) during the isomerization of cheese whey, using Na~2~CO~3~ (0.5%) as a catalyst. As the reaction proceeded, the yield values decreased to 14.13 ± 0.13 and 10.93 ± 0.07% for SWP and AWP, respectively. The observed reduction in the yield of LAU by increasing the reaction time is due to the hydrolysis of LAU and subsequent formation of organic acids. The mechanism of lactose isomerization consisted of epimerization and aldose--ketose interconversion. Such a network of reactions is known as Lobry de Bruyn-van Ekenstein transformation. Hajek et al. \[[@B27-foods-09-00784]\] have studied the isomerization of lactose in an alkaline environment. In the case of LBA, the yield for the SWP ([Figure 5](#foods-09-00784-f005){ref-type="fig"}a) increased asymptotically with the reaction time, reaching a value of 5.23 ± 0.02%. Similar behavior, but less pronounced, was observed in the yield values of LBA for AWP (2.15 ± 0.15%). The relatively high yield of LBA for SWP, nearly 2-fold than AWP, is not surprising, since the pH of SWP (6.23 ± 0.01, [Table 1](#foods-09-00784-t001){ref-type="table"}) favored LBA formation. It is thought that acidic conditions hinder the interaction between molecular oxygen and LA moieties, affecting LBA formation. The role of pH during the catalytic oxidation of LA has been discussed elsewhere \[[@B28-foods-09-00784],[@B29-foods-09-00784]\]. Mechanistically, the oxidation of aldehydes is consensually thought to occur in two steps, hydration of the aldehyde group and subsequent dehydrogenation to produce their corresponding acid \[[@B30-foods-09-00784]\]. 3.4. Formation of Glucose, Galactose, and Organic Acids {#sec3dot4-foods-09-00784} ------------------------------------------------------- The conversion values of lactose, as well as the yield values of LAU and LBA ([Figure 5](#foods-09-00784-f005){ref-type="fig"}), suggest the formation of other products derived from hydrolysis, oxidation, and degradation. These products were divided into three groups---monosaccharides, sugar acids, and organic acids ([Figure 6](#foods-09-00784-f006){ref-type="fig"}). The first group of compounds is monosaccharides (glucose and galactose) that are formed due to the hydrolysis of lactose, breakage of the glycosidic linkage. The initial concentration of glucose and galactose was higher in AWP (about 21 and 78 mmol/L, respectively) than that of SWP (about 4.61 and 2.21 mmol/L, respectively). Acid whey permeate is the byproduct of yogurt manufacture \[[@B31-foods-09-00784]\], and it is characterized by a relatively high concentration of organic acids, [Table 1](#foods-09-00784-t001){ref-type="table"}. As the reaction proceeded, the glucose in SWP ([Figure 6](#foods-09-00784-f006){ref-type="fig"}a) gradually increased with time, reaching a maximum of about 13 mmol/L after 90 min. Subsequent progression of the reaction resulted in a lower concentration of glucose (about 9 mmol/L after 210 min). A similar trend was observed for galactose in SWP ([Figure 6](#foods-09-00784-f006){ref-type="fig"}a), where a maximum concentration (24 mmol/L) was obtained after 90 min, followed by a reduction in the concentration with the progression of the reaction. On the other hand, the concentration of glucose and galactose in AWP ([Figure 6](#foods-09-00784-f006){ref-type="fig"}b) gradually decreased with time from about 21 and 78 mmol/L at the beginning of the reaction to 16 and 42 mmol/L after 210 min, for glucose and galactose, respectively. It appears that lactose underwent hydrolysis to produce glucose and galactose that further oxidized to form their respective sugar acids, gluconic and galactonic acid. The sugar acids represent the second group of components derived from the catalytic conversion of lactose. In SWP, the concentration of gluconic acid increased with time until it reached a plateau at 150 min (60 mmol/L). Afterward, the concentration of gluconic acid marginally increased up to 66 mmol/L after 210 min. In the case of AWP, the concentration of gluconic acid reached a maximum value of 65 mmol/L only after 30 min, followed by a reduction to about 30 mmol/L. Chemical process for the production of gluconic acid consisted of catalytic oxidation of a concentrated glucose solution under alkaline conditions \[[@B32-foods-09-00784]\]. The third group of components under consideration corresponds to organic acids. Overall, the concentration of organic acids in AWP was higher than that of SWP. Formic and lactic were the most predominant organic acids (299 and 270 mmol/L, respectively) in AWP. This observation is not surprising, since acid whey is derived from the formation of lactic acid bacteria, whose main product is lactic acid. A general trend was observed in SWP, where the concentration of organic acids increased slightly with the reaction time ([Figure 6](#foods-09-00784-f006){ref-type="fig"}a). Citric, acetic, and uric acids were found at relatively low concentrations. These findings are in agreement with the earlier studies on the formation of isosaccharinic acids and organic acids \[[@B26-foods-09-00784]\]. Organic acids are formed from the breakdown of glucose and galactose. More specifically, lactic acid is formed as the primary byproduct of the catalytic oxidation of glucose \[[@B33-foods-09-00784]\]. The catalytic oxidation of glucose over bimetallic catalysts has been recognized as an efficient alternative for the production of gluconic acid. The oxidation of glucose occurs via a carbonyl conjugated radical mechanism due to the formation of H~2~O~2~, which in turns is formed by the presence of oxygen in the aqueous media \[[@B34-foods-09-00784]\]. However, the mechanisms of glucose oxidation strongly depend on the pH of the medium and the type of catalysts \[[@B33-foods-09-00784]\]. On the other hand, the concentration of organic acids in AWP showed an increase with time for citric and pyruvic acid, while the concentration of formic and lactic acid decreased with time ([Figure 6](#foods-09-00784-f006){ref-type="fig"}b). Citric acid and pyruvic acid are formed due to glucose breakdown, which subsequently formed oxalic acid and others. 3.5. Product Distribution {#sec3dot5-foods-09-00784} ------------------------- [Figure 7](#foods-09-00784-f007){ref-type="fig"} exemplifies the one-pot synthesis of lactose derivatives, where a pool of molecules synthesized from SWP ([Figure 7](#foods-09-00784-f007){ref-type="fig"}a) and AWP ([Figure 7](#foods-09-00784-f007){ref-type="fig"}b) was plotted at different reaction times. In SWP, the more predominant compounds were lactose (43--35%), gluconic acid (6--18%), LAU (14--9%), formic acid (8--13%), and lactic acid (8--7%), whose concentration varied with the reaction time. For AWP, lactose (27--23%), lactic acid (21--24%), formic acid (19--20%), gluconic acid (11--7%), and LAU (5--3%) were the most predominant compounds. It is worth mentioning that these results should be considered cautiously because the final concentration is influenced by a number of factors, including pH, temperature, the composition of the stream, the type of catalyst, and catalyst load. Thus, the concentration of a given group of components may be optimized in order to maximize their production according to the study of reaction kinetics. The concept of one-pot was first introduced by Kolb et al. \[[@B35-foods-09-00784]\], who described a click-chemistry in which a set of chemical transformations yielded higher efficiency, fast rates, and simple product isolation. Soon after, the concept of click chemistry quickly found widespread application in various research areas, including organic chemistry, polymer synthesis, petroleum, and biorefinery. Currently, there are several terminologies to describe multi-step reactions that take place in one pot, such as domino reaction, cascade reaction and tandem reaction. One-pot synthesis is effective because several chemical transformation steps can be carried out in a single pot, while circumventing several purification procedures at the same time. Thus, a one-pot procedure can minimize chemical waste, save time, and simplify practical aspects. 4. Conclusions {#sec4-foods-09-00784} ============== The one-pot conversion of lactose streams resulted in the formation of four main group of components---rare carbohydrate (lactulose), monosaccharides (glucose and galactose), sugar acids (gluconic acid), and organic acids. The final distribution and their concentration differ from SWP and AWP, where lactulose was favored in SWP, and organic acids were favored in AWP. The synthesis of a pool of molecules through a one-pot approach represents an alternative approach for the utilization of streams of lactose. Upon further separation, organic acids can be used as building blocks of numerous applications in the manufacture of herbicides, bioplastics, and biofertilizers. Conceptualization, S.I.M.-M.; methodology, M.E.; analysis, M.E.; writing---original draft preparation, M.E.; writing---review and editing, S.I.M.-M.; funding acquisition, S.I.M.-M. All authors have read and agreed to the published version of the manuscript. This work has been made possible through the financial support of the Midwest Dairy Foods Research Center (St. Paul, MN) and partial support from the USDA National Institute for Food and Agriculture (HATCH project SD00H607-16). The authors declare no conflict of interest. {#foods-09-00784-f001} {#foods-09-00784-f002} {#foods-09-00784-f003} {#foods-09-00784-f004} {#foods-09-00784-f005} {#foods-09-00784-f006} {#foods-09-00784-f007} foods-09-00784-t001_Table 1 ###### Physicochemical characteristics of sweet whey permeate and acid whey permeate. Parameters Sweet Whey Permeate Acid Whey Permeate --------------------------------- --------------------- -------------------- pH 6.23 ± 0.01 4.38 ± 0.02 Total solids (g/L) 87.38 ± 0.25 85.35 ± 0.06 Total non-volatile solids (g/L) 8.23 ± 0.17 9.22 ± 0.16 Fat (g/L) 0.85 ± 0.23 1.81 ± 0.10 Total protein (g/L) 3.02 ± 0.01 3.95 ± 0.08 Lactose (g/L) 76.42 ± 4.3 61.73 ± 6.29 Organic acids (g/L) 7.11 ± 0.35 17.91 ± 0.89 Mean ± standard deviation (*n* = 3).

Introduction {#section1-2050313X17726911} ============ Aneurysm is defined as a focal dilatation of an artery to at least 1.5 times its normal diameter. True aneurysms involve all three layers of the arterial wall.^[@bibr1-2050313X17726911]^ True femoral artery aneurysms are uncommon and are most often seen in male patients more than 70 years old. They are often associated with additional peripheral aneurysms.^[@bibr2-2050313X17726911]^ We describe what is, to our knowledge, the first reported case of true femoral artery aneurysm concomitant with bilateral iliac aneurysms that was treated with a staged surgical and endovascular approach. Case report {#section2-2050313X17726911} =========== A 67-year-old man who was previously healthy was admitted because of a mass in his right groin. He was asymptomatic. Physical examination revealed a palpable, pulsating mass in the right groin. There was no history of distal embolization or trauma to the groin, and distal pulses were present in the right leg. Ultrasonographic examination of the belly and groin showed an aneurysm of the right common femoral artery and bilateral aneurysms of the common iliac arteries. Blood parameters and biochemical and coagulation profiles were all within their normal ranges. Contrast-enhanced computed tomography confirmed multiple dilatations of both common iliac arteries and the right femoral artery. In addition, the external iliac arteries were highly tortuous and small in caliber ([Figure 1](#fig1-2050313X17726911){ref-type="fig"}). The thoracoabdominal aorta and popliteal arteries were of normal diameter, but the diameter of the left common femoral artery measured 2 cm. {#fig1-2050313X17726911} Informed consent was obtained from the patient. The patient was treated with a staged procedure. In the first stage, the right femoral aneurysm was repaired in open surgical fashion. Under intratracheal general anesthesia, a vertical incision was performed in the right groin to expose the aneurysm. Operative findings confirmed that the aneurysm extended from the common femoral artery to the proximal parts of the deep and superficial femoral arteries ([Figure 2](#fig2-2050313X17726911){ref-type="fig"}). Then, the patient underwent aneurysmectomy, placement of a knitted Dacron Y graft, and reconstruction of the superficial and deep femoral arteries ([Figure 3](#fig3-2050313X17726911){ref-type="fig"}). The proximal parts of the deep and superficial femoral arteries were resected and excluded, and all nondiseased distal parts were anastomosed to the arms of the Dacron graft in an end-to-end configuration. Another, 10-mm Dacron graft was anastomosed to the body of the Y graft in an end-to-side configuration to create an access point for planned endovascular repair of the iliac aneurysms ([Figure 4](#fig4-2050313X17726911){ref-type="fig"}). {#fig2-2050313X17726911} {#fig3-2050313X17726911} {#fig4-2050313X17726911} In the second-stage procedure, the bilateral iliac aneurysms were repaired with stent grafts. A stiff guide wire was used because of the tortuosity of the external arteries. Stented grafts were passed through the side branch of the Y graft. The internal iliac arteries were embolized with coils. After the endovascular repair, the side branch of the Y graft was ligated and resected. There were no intraoperative or postoperative complications. Microbiological cultures of the aneurysm wall were negative for bacterial infection. Discussion {#section3-2050313X17726911} ========== True femoral artery aneurysms are often associated with other peripheral aneurysms. They co-occur with multiple aneurysms in more than 50% of cases and are bilateral in 36%--72% of cases. At least a quarter to a half of all patients with femoral artery aneurysms have an associated abdominal aortic or iliac aneurysm.^[@bibr3-2050313X17726911],[@bibr4-2050313X17726911]^ After a peripheral artery aneurysm is diagnosed, all aortoiliac, femoral, and popliteal arteries should be evaluated by imaging studies to rule out other aneurysms before treatment is planned. Most femoral artery aneurysms are seen in association with arteriosclerosis, although they have also been reported in patients with Behcet disease, Marfan syndrome, and acromegaly.^[@bibr5-2050313X17726911][@bibr6-2050313X17726911]--[@bibr7-2050313X17726911]^ Other related risk factors are smoking, hypertension, and trauma. At the time of initial presentation, 30%--40% of patients are asymptomatic, whereas 10%--65% have complications. Symptoms may be local, resulting from pressure on surrounding structures (e.g. leg swelling, pain, and tenderness); in addition, limb ischemia may occur as a result of distal embolization.^[@bibr3-2050313X17726911]^ Rupture is uncommon, occurring in 10%--14% of cases.^[@bibr3-2050313X17726911],[@bibr4-2050313X17726911]^ Cutler and Darling^[@bibr4-2050313X17726911]^ classified true femoral artery aneurysms according to their relationship to the common femoral bifurcation. Type I aneurysms involve only the common femoral artery and terminate proximal to the femoral bifurcation, whereas Type II aneurysms involve the deep femoral artery. All symptomatic femoral aneurysms should be repaired. Available data show that most of these aneurysms remain silent even after the onset of life-threatening complications such as limb ischemia or rupture, which elevates the mortality and morbidity rates associated with such aneurysms. Therefore, although there are no reports of a strong correlation between aneurysm size and complications, repair has been recommended for all common femoral artery aneurysms with a diameter greater than 2.5 cm.^[@bibr8-2050313X17726911]^ In patients with multiple aneurysms, any symptomatic aneurysms should be treated first; in asymptomatic patients, the aneurysms can be repaired either simultaneously in a single operation or serially in a staged procedure with a combination of open surgical and endovascular techniques. If the aneurysm involves the deep femoral artery and the patient may have distal emboli, every effort should be made to maintain deep femoral artery patency through bypass grafting or reimplantation. In this case, we focused first on the femoral aneurysm in order to prevent embolic events during endovascular repair of the iliac aneurysms. We also attempted to maintain deep femoral artery patency by reimplantation. The literature includes only a few reported cases of the treatment of femoral artery aneurysm associated with other peripheral arterial aneurysms.^[@bibr4-2050313X17726911],[@bibr9-2050313X17726911]^ As far as we can ascertain, ours is the first reported case of a true femoral artery aneurysm combined with bilateral iliac artery aneurysms, which were repaired with a staged surgical and endovascular approach. The authors thank Stephen N. Palmer, PhD, ELS, and Sally P. Timko, BA, for editorial support. **Declaration of conflicting interests:** The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. **Ethical approval:** Our institution does not require ethical approval for reporting individual cases or case series. **Funding:** The author(s) received no financial support for the research, authorship, and/or publication of this article. **Informed consent:** Written informed consent was obtained from the patient(s) for their anonymized information to be published in this article.

1. Introduction {#sec1-ijms-20-05342} =============== In the past few decades, the significant diagnostic and therapeutic progress made in the field of oncology has improved the survival rates of children and young adults. However, it is now clearly established that these excellent results are achieved through treatments that have potentially deleterious impacts on reproductive function. In order to protect gametes from the gonadotoxic effects of chemotherapy and/or radiotherapy, several fertility preservation (FP) techniques, such as oocyte or embryo cryopreservation with or without ovarian stimulation or cryopreservation of the ovarian cortex, should be proposed \[[@B1-ijms-20-05342]\]. However, the application of these methods may be limited by age, pubertal status, disease, and emergency. In addition, these procedures may be difficult to perform, and the future use of cryopreserved germ cells remains uncertain. Therefore, improving the FP methods currently available and developing new FP strategies represent major challenges in oncofertility. Chemotherapy exerts toxicity on the ovaries directly. It is important to distinguish between the short- and long-term effects of drugs on the ovaries. Soon after the beginning of treatment, chemotherapy induces apoptosis of growing follicles, leading to temporary amenorrhea. The impact of drugs on fertility after healing concerns the effects of chemotherapy on the primordial follicular reserve as these treatments may lead to a premature loss and, at worst, primary ovarian insufficiency (POI). POI is a well-known long-term side effect of cancer chemotherapy treatment. The extent of ovarian damage depends upon several factors, of which the most important are the type of drug, its dosage, and the protocol \[[@B2-ijms-20-05342]\]. Chemotherapeutic agents can be divided into five categories: alkylating agents, antitumor antibiotics, platinum-based drugs, antimetabolites, and taxanes. The mechanisms implicated in the gonadotoxicity of these molecules have been explored in various experimental models, such as analysis of histological female ovary sections after chemotherapy, animal models treated with injections, xenograft models, or cell cultures in the presence of active metabolites of chemical agents, and are not yet fully understood \[[@B3-ijms-20-05342]\]. Several hypotheses have been proposed and could coexist. On the one hand, chemotherapeutic agents could exert direct toxicity on primordial follicles, inducing DNA damage and subsequent apoptosis. On the other hand, it has been suggested that these drugs could trigger an indirect depletion of primordial follicles by over-recruitment. Increasing knowledge of the possible mechanisms implicated in chemotherapy-induced ovarian damage will facilitate the development of new therapies, called fertoprotective agents \[[@B4-ijms-20-05342]\], aimed at protecting the follicular reserve \[[@B5-ijms-20-05342]\]. 2. Follicular Ovarian Reserve and Its Regulation {#sec2-ijms-20-05342} ================================================ In mammals, the follicular ovarian reserve, constituted by primordial follicles, is established early on in life then keeps declining regularly throughout the reproductive period. The pool of primordial follicles serves as a source of growing follicles and fertilizable eggs for the entire female reproductive life. Actually, each primordial follicle can remain quiescent for years; be activated and enter the growing process; or undergo atresia directly from the dormant stage \[[@B6-ijms-20-05342]\]. To produce mature oocytes, activated primordial follicles develop through primary and secondary stages before acquiring an antral cavity. At the antral stage, most follicles undergo apoptotic degeneration and only a few of them grow further to reach the preovulatory stage under the cyclic gonadotropin stimulation that occurs after puberty \[[@B7-ijms-20-05342]\]. The maintenance of female reproductive function implies the presence of a vast majority of quiescent primordial follicles and continuous repression of primordial follicle activation into early growing follicles. This activation, starting during fetal life, is finely controlled though maintaining a balance between inhibitory and stimulatory factors. In vitro experiments and genetically modified mouse or sheep models have enabled the decoding of the molecular mechanisms that control follicular activation. Numerous factors, such as growth factors, hormones, transcription factors, or cytokines, produced by oocytes and/or granulosa cells, can act in an autocrine, paracrine, or endocrine manner \[[@B8-ijms-20-05342]\]. The quiescence of primordial follicles is maintained by several molecules including phosphatase and tensin homolog deleted on chromosome 10 (Pten), tuberous sclerosis complexes 1--2 (Tsc1--Tsc2) complex, Forkhead box protein O3A (Foxo3A), p27, anti-Müllerian hormone (AMH), and Forkhead box L2 (FoxL2) \[[@B6-ijms-20-05342]\]. Many studies have highlighted the crucial roles of the phosphatidyl-inositol-3-kinase (PI3K) signaling pathway in oocytes in controlling follicular activation \[[@B9-ijms-20-05342]\]. Indeed, in genetically modified mouse models, it was observed that the PI3K--Akt--mammalian target of rapamycin (mTOR) signaling pathway is crucial for the control of survival and activation of primordial follicles \[[@B10-ijms-20-05342]\]. For example, Pten and Tsc1--2 are negative regulators of this signaling pathway and, in mice, the deletion of these genes from oocytes leads to primordial follicle activation and subsequent early follicular depletion. In these models, mTOR activity is accelerated within the oocyte, highlighting the critical role of this serine/threonine kinase in primordial follicle activation. The transcription factor FoxO3A, mainly expressed in the oocytes of resting follicles, acts downstream of the PI3K signaling pathway and appears to be the main actor involved in follicular activation \[[@B8-ijms-20-05342]\]. At the same time, the survival of primordial follicles is maintained by other mechanisms involving PDK1 signaling, rpS6. Several studies have suggested the involvement of the autophagy process in the regulation of the ovarian reserve of primordial follicles. For example, autophagy is implicated in maintaining the primordial oocyte pool in murine newborns \[[@B11-ijms-20-05342]\], and the induction of autophagy at birth seems to be a crucial step to preserve the stock of primordial follicles \[[@B12-ijms-20-05342],[@B13-ijms-20-05342]\]. In the same manner, primordial follicle survival or apoptosis results from a balance between the expression of survival (antiapoptotic) and proapoptotic factors. Among these factors, the proteins B-cell lymphoma 2 (Bcl-2) and Bcl-2-associated X protein (BAX) likely play a critical role. Thus, a synergistic and coordinated suppression of follicular activation, provided by multiple inhibitory and activator molecules, is necessary to preserve the primordial follicular stockpile in association with the process maintaining dormancy. Any disorder in these complex mechanisms can lead to a premature loss of the follicular reserve \[[@B14-ijms-20-05342]\]. 3. Follicular Atresia and Apoptosis {#sec3-ijms-20-05342} =================================== 3.1. Pathophysiology: DNA Alteration, Follicular Atresia, and Apoptosis {#sec3dot1-ijms-20-05342} ----------------------------------------------------------------------- The molecules used in chemotherapy induce alterations in the DNA. Double-stranded breaks (DSB) are one of the main DNA lesions caused by these cytotoxic agents and the most severe. DSB can, in turn, lead to either DNA repair pathways allowing cell survival or cell death by apoptosis \[[@B15-ijms-20-05342]\]. DNA repair pathways differ according to the type of chemotherapeutic agent and may involve, for example, pATM, RAD51, or PARP1 proteins \[[@B15-ijms-20-05342]\]. When the repair pathways are not sufficiently activated, DNA damage induces cellular apoptosis. This mechanism is mainly mediated by p63 protein (and, more specifically, the TAp63 isoform), which activates Bcl2-associated X (BAX) protein and the Bcl-2 antagonist killer (BAK) protein. BAX/BAK activation can be transmitted by TAp73 or secondary to p53 up-modulator of apoptosis) and phorbol-12-myristate-13-acetate--induced protein 1 activation \[[@B16-ijms-20-05342]\]. These mechanisms are particularly complex within the ovaries and differ according to the type of chemotherapy molecule. A recent and extensive review discusses the induction and repair of DNA damage in the ovaries \[[@B15-ijms-20-05342]\]. The impact of chemotherapeutic agents on growing follicles is well known, and apoptotic pathways have been well documented \[[@B5-ijms-20-05342],[@B17-ijms-20-05342]\]. Almost all classes of drugs induce DNA alteration of granulosa cells and/or oocytes, leading to either apoptosis of growing follicles or the survival of mutagenic oocytes. This phenomenon often induces temporary amenorrhea \[[@B18-ijms-20-05342]\]. More rarely, if fertilization occurs during drugs' exposure, it can lead to spontaneous abortion or congenital abnormalities in the offspring \[[@B19-ijms-20-05342]\]. These complications are closely related to the timing of oocyte exposure to cytotoxic treatment. Fertilization months or years after the end of protocol seems to be safe for offspring as these pregnancies are achieved from oocytes exposed in a dormant state, which remained genetically undamaged \[[@B20-ijms-20-05342]\]. While apoptosis and atresia in growing follicles in response to chemotherapeutic agents have been well investigated, the nature of these mechanisms in dormant follicles is still under debate \[[@B15-ijms-20-05342]\]. According to several studies, the main chemotherapeutic agents induce follicular depletion by directly affecting the primordial follicles entering massively into atresia \[[@B15-ijms-20-05342]\]. Overall, rodent models as well as models of human ovarian xenograft or in vitro ovary cultures were used to investigate the impact of chemotherapy on primordial follicles. Cyclophosphamide is a widely used alkylating agent and is recognized as one of the most gonadotoxic drugs. It has been shown to induce DNA double-stranded breaks and activate the DNA damage response in a human ovarian xenograft model \[[@B21-ijms-20-05342]\]. These results were confirmed in in vitro ovarian cultures with cyclophosphamide active metabolite \[[@B22-ijms-20-05342],[@B23-ijms-20-05342],[@B24-ijms-20-05342]\] or after in vivo cyclophosphamide injection \[[@B25-ijms-20-05342]\]. In the same manner, in vitro analysis of newborn mouse ovaries revealed DNA damage and apoptosis of primordial follicles after cisplatin treatment \[[@B26-ijms-20-05342],[@B27-ijms-20-05342]\], which were further confirmed after in vivo injection in newborn or adult mice \[[@B16-ijms-20-05342],[@B25-ijms-20-05342]\]. Recently, a model of a xenograft of human cortex ovaries in nude mice revealed the same results \[[@B28-ijms-20-05342]\]. Furthermore, similar effects with similar models were also found following doxorubicin exposure \[[@B29-ijms-20-05342],[@B30-ijms-20-05342]\]. The improvement of knowledge of the specific apoptotic and DNA repair pathways involved in the ovarian damage induced by chemotherapy will reveal targets for protective agents to reduce or prevent ovarian damage ([Figure 1](#ijms-20-05342-f001){ref-type="fig"}) \[[@B3-ijms-20-05342]\]. 3.2. Fertoprotective Agents {#sec3dot2-ijms-20-05342} --------------------------- ### 3.2.1. Sphingosine 1 Phosphate and Ceramide 1 Phosphate {#sec3dot2dot1-ijms-20-05342} Sphingosine-1-phoshate (S1P) is a membrane sphingolipid involved in several physiological processes, including apoptosis of ovarian follicles. Indeed, it was shown that the sphingomyelin pathway regulates the developmental death of oocytes and S1P protects the ovarian reserve from radiation injuries \[[@B31-ijms-20-05342]\]. Later, in an animal model, S1P injection directly into the ovaries was shown to decrease the apoptosis of primordial follicles induced by chemotherapy and thus protect fertility \[[@B32-ijms-20-05342],[@B33-ijms-20-05342]\]. In a human ovarian xenograft model, S1P can block the human apoptotic follicle death induced by cyclophosphamide and doxorubicin and preserve the primordial follicle stockpile \[[@B34-ijms-20-05342],[@B35-ijms-20-05342]\]. Moreover, S1P seems to reduce the atresia of primordial follicles that occurs during the slow freezing and thawing of human ovarian cortical strips, confirming its protective role \[[@B36-ijms-20-05342]\]. Recently, ceramide 1 phosphate (C1P), another sphingolipid, was also found to be a potential ovarian protective agent as ovarian administration of this drug reduces the ovarian damage induced by cyclophosphamide and protects the ovarian reserve via the inhibition of apoptosis and improvement of stromal vasculature \[[@B37-ijms-20-05342]\]. However, one study had conflicting results, demonstrating that S1P was not effective against apoptosis in rats after intraperitoneal cyclophosphamide treatment \[[@B38-ijms-20-05342]\]. One of the major limitations of these treatments is that S1P and C1P must be administered by continuous administration or injection directly into the ovaries. Nevertheless, recently, a long-acting oral form of an S1P analog has been developed and its impact on the ovaries was evaluated in a rat model \[[@B39-ijms-20-05342]\]. It was suggested that this treatment might decrease spontaneous follicular apoptosis, making these molecules potentially appropriate for human use. However, even if S1P might be a promising fertility preservation strategy in the future, additional studies have to be conducted to confirm the protective role of this molecule, to evaluate possible interference with chemotherapy, and to evaluate the impact of this strategy on offspring. ### 3.2.2. Imatinib {#sec3dot2dot2-ijms-20-05342} Imatinib is a competitive tyrosine kinase inhibitor and, more specifically, a c-Abl kinase inhibitor. This protein is implicated in the apoptotic pathway induced by DNA damage in activating TAP63 transcriptional activity. Clinically, it is used for the treatment of hemopathies or other cancers. Based on its role as a c-Abl kinase inhibitor, imatinib was evaluated as a molecule to prevent the primordial follicle loss caused by cisplatin as this drug was shown to induce DNA damage and subsequent apoptosis in primordial follicles via TAP63 activation. It was hypothesized that imatinib could prevent the TAP63 accumulation and activation induced by cisplatin and thus impede follicle apoptosis. This molecule was first evaluated in 2009 by Gonfloni et al. in a mouse model \[[@B26-ijms-20-05342]\]. In this study, the authors observed the occurrence of massive primordial and primary follicle depletion in cisplatin-treated mice, whereas they noted a significant rescue of these follicles in the ovaries of mice simultaneously treated with cisplatin and imatinib. Furthermore, they showed that this treatment had a long-term impact on fertility and reproductive outcomes. Similar results were found by the same team in 2012 \[[@B40-ijms-20-05342]\], while others have confirmed these results using in vitro newborn ovary cultures \[[@B41-ijms-20-05342]\] and in vitro culture and subrenal grafting of mouse ovaries \[[@B42-ijms-20-05342]\]. Nevertheless, two studies have also contested these results, finding that imatinib did not protect primordial follicles from cisplatin-induced apoptosis and did not prevent impaired fertility \[[@B28-ijms-20-05342],[@B43-ijms-20-05342]\]. Thus, due to the existence of conflicting results, additional studies are needed to evaluate whether imatinib could be a new treatment to limit cisplatin gonadotoxicity. Moreover, as imatinib interferes with the apoptotic pathway, it will be crucial to show that imatinib does not interfere with the antitumor activity of cisplatin. ### 3.2.3. Molecules Interfering with the DNA Repair Pathway {#sec3dot2dot3-ijms-20-05342} Following spontaneously occurring, or chemotherapy-induced, DNA damage, the efficiency of the DNA repair pathway is a critical determinant of a cell's survival. Thus, several studies have tried to develop molecules aiming to induce DNA repair instead of the apoptosis pathway to encourage follicle survival and limit follicular depletion. For example, Rad 51 is a protein implicated in DNA repair after double-strand breaks. It was shown that, in an in vitro oocyte culture model, DNA damage in oocytes can be induced by doxorubicin and that oocytes possess the machinery and capability for repairing such DNA damage through Rad51 activation \[[@B44-ijms-20-05342]\]. So, strategies manipulating Rad51 could be potential candidates to limit follicle depletion due to chemotherapy. Recently, Rossi et al. reported the protective effect of luteinizing hormone (LH) on the primordial follicle pool of prepubertal mouse ovaries against cisplatin-induced follicular depletion \[[@B27-ijms-20-05342]\]. First, these authors conducted an in vitro analysis and showed that LH treatment of prepubertal ovarian fragments generated antiapoptotic signals, reducing the oocyte level of proapoptotic TAp63 protein and favoring the DNA repair pathway in the oocytes. Thereafter, they showed that the administration of a single dose of LH to prepubertal female mice, concomitantly with cisplatin injection, inhibited the depletion of the primordial follicle reserve caused by the drug. If this protective role of recombinant LH is confirmed, it could be a very interesting candidate as this molecule is already available to women. Thus, clinical studies could be conducted relatively quickly. 4. Follicular Activation {#sec4-ijms-20-05342} ======================== 4.1. Physiopathology of Ovarian Reserve Depletion Due to Follicular Activation {#sec4dot1-ijms-20-05342} ------------------------------------------------------------------------------ A more recent theory suggests that chemotherapy, such as cyclophosphamide or cisplatin, induces follicular depletion through the massive growth of resting follicles, occurring simultaneously with the apoptosis of growing follicles \[[@B45-ijms-20-05342]\]. Recruitment of primordial follicles would be secondary to the activation of the PI3K signaling pathway, whose role in follicle quiescence has been well-established by many knockout mouse models as well as in vitro studies on human ovarian cortex fragments \[[@B8-ijms-20-05342],[@B46-ijms-20-05342]\]. In addition, cytotoxic agents destroy growing follicles, resulting in a reduction in AMH secretion. As this hormone is supposed to inhibit primordial follicles' recruitment, its decrease amplifies follicular activation and the subsequent depletion of the follicular reserve. In the first study revealing this hypothesis, no primordial follicles showed signs of apoptosis \[[@B45-ijms-20-05342]\]. Other studies, using the same mouse model, confirmed this hypothesis, called the "burnout effect" \[[@B47-ijms-20-05342],[@B48-ijms-20-05342]\]. Thus, the burnout effect consists of the trigger of recruitment of dormant follicle growth, mediated by an upregulation in the PI3K/PTEN/Akt pathway, occurring simultaneously with large follicle apoptosis and resulting in a reduction of AMH secretion. The route by which chemotherapy induces the activation of this signaling pathway remains unclear. It may be via the direct influence on the oocytes and pregranulosa cells of primordial follicles \[[@B49-ijms-20-05342]\]. This theory was also supported when using other cytotoxic agents such as cisplatin \[[@B50-ijms-20-05342],[@B51-ijms-20-05342]\]. In consideration of this theory, Lande et al. showed that, in vitro, phosphoramide mustard, a cyclophosphamide metabolite, enhances human primordial follicle activation in developing follicles \[[@B52-ijms-20-05342]\]. This theory could also explain the alteration of the follicular reserve induced by the presence of an ovarian endometrioma \[[@B53-ijms-20-05342]\] or massive follicular loss secondary to ovarian cortex transplantation \[[@B54-ijms-20-05342],[@B55-ijms-20-05342]\]. Nevertheless, the molecular mechanism by which chemotherapy activates the PI3K pathway within primordial follicles is not known. In accordance with this hypothesis, in the last few years, several investigations have been carried out to develop new molecules that would preserve the ovarian reserve by inhibiting the PI3K pathway and follicular activation ([Figure 2](#ijms-20-05342-f002){ref-type="fig"}) \[[@B3-ijms-20-05342],[@B56-ijms-20-05342]\]. 4.2. Fertoprotective Agents {#sec4dot2-ijms-20-05342} --------------------------- ### 4.2.1. AS101 {#sec4dot2dot1-ijms-20-05342} AS101 \[ammonium trichloro(dioxoethylene-o,o′)tellurate\] is a nontoxic immunomodulatory compound that modulates the PI3K--Pten--Akt pathway \[[@B57-ijms-20-05342]\]. It has been shown to reduce the negative hematologic and dermatologic side effects of chemotherapy \[[@B58-ijms-20-05342]\]. As cyclophosphamide was found to activate the PI3K pathway, inducing primordial follicle recruitment and subsequent follicular depletion of ovarian reserve, AS101 was investigated as a treatment to prevent cyclophosphamide-induced follicle loss in a mouse model \[[@B45-ijms-20-05342]\]. Ultimately, in vivo treatment of mice with AS101 was found to reduce the cyclophosphamide-induced depletion of primordial follicles. No increase in fetal malformation was observed in mice previously treated with AS101, indicating the safety of this treatment for offspring. This treatment was the first one tested and yielded encouraging results; nevertheless, to date, no other studies using this molecule to prevent cyclophosphamide ovarian damage have been performed. ### 4.2.2. Anti-Müllerian Hormone {#sec4dot2dot2-ijms-20-05342} AMH is a glycoprotein hormone expressed by the granulosa cells surrounding the oocytes. It is produced by follicles from the primary stage of development until selection for dominance, and plays a key role during folliculogenesis. As it has been shown to limit the activation of primordial follicles in in vivo or in vitro mouse models \[[@B59-ijms-20-05342],[@B60-ijms-20-05342],[@B61-ijms-20-05342]\], it was suggested in three recent studies that this hormone could be an effective treatment option to limit chemotherapy-induced gonadotoxicity \[[@B48-ijms-20-05342],[@B62-ijms-20-05342],[@B63-ijms-20-05342]\]. Kano et al. reported that, in mice, superphysiologic doses of AMH delivered either by a recombinant protein via osmotic pumps or gene therapy could limit the primordial follicle loss induced by cyclophosphamide, doxorubicin, or cisplatin \[[@B62-ijms-20-05342]\]. The protective effects of AMH vary between drugs, suggesting that different mechanisms for ovarian damage are induced by different chemotherapeutic agents. Recently, we assessed the protective effect of AMH in pubertal mice treated with cyclophosphamide \[[@B48-ijms-20-05342]\]. In this model, we showed that the ovaries of cyclophosphamide-treated mice were depleted of primordial follicles, whereas the number of primordial and early-growing follicles was similar to that in controls among the ovaries of mice treated with concomitant injections of cyclophosphamide and AMH. Then, we showed that 15 weeks after the end of the treatment, the number of ovulated eggs after ovarian stimulation was significantly reduced in cyclophosphamide-treated mice and rescued by AMH co-administration. The molecular mechanisms underlying these effects were explored. Interestingly, an investigation of the PI3K signaling pathway showed that the phosphorylation of FoxO3A was significantly lower in mouse ovaries treated with AMH. This transcription factor, expressed in the nucleus of primordial follicles, plays an essential role in the maintenance of primordial follicles in a quiescent state \[[@B6-ijms-20-05342]\]. The phosphorylation of FoxO3A induces the protein nuclear export, leading to the activation of primordial follicles \[[@B8-ijms-20-05342]\]. Our results suggested that AMH might inhibit primordial follicle recruitment by preventing cytoplasmic shuttling of FoxO3A induced by cyclophosphamide. Moreover, in this study, we also provide evidence of a possible role of autophagy in the preservation of the follicular pool reserve. Indeed, we showed that AMH administration was able to induce autophagy in ovaries and a possible mechanism to explain the modulation of AMH-induced autophagy might implicate FOXO3A as this factor was shown to be related to autophagy activation \[[@B64-ijms-20-05342]\]. These results are in accordance with those of other studies suggesting the involvement of autophagy in the regulation of follicular ovarian reserve \[[@B13-ijms-20-05342],[@B65-ijms-20-05342]\]. Further, Roness et al. confirmed the fertoprotective role of recombinant AMH in the same mouse model as pharmacological administration of AMH during chemotherapy treatment reduced follicle activation and primordial follicle loss and significantly improved reproductive outcomes \[[@B63-ijms-20-05342]\]. Interestingly, they also showed that AMH does not interfere with the therapeutic actions of chemotherapy. These data indicate that AMH represents a potential novel treatment for limiting the primordial follicle depletion induced by chemotherapy. Nevertheless, these promising results need to be confirmed further. As AMH is produced only by the ovaries and acts through a specific receptor mainly expressed by the ovaries, this hormone might be of particular interest since it could act as a targeted therapy without interfering with physiological mechanisms or the efficacy of chemotherapy. ### 4.2.3. Melatonin {#sec4dot2dot3-ijms-20-05342} Primarily revealed as a secretory product of the pineal gland, melatonin (N-acetyl-5-methoxytryptamine) is commonly used in various biological processes such as treating insomnia. Moreover, melatonin can be used as a potential therapeutic adjuvant during chemotherapy as it has been shown to reduce some adverse effects of drugs \[[@B66-ijms-20-05342]\]. Interestingly, melatonin is also produced in various tissues including reproductive tissues such as the ovaries and the placenta \[[@B67-ijms-20-05342]\], and melatonin receptors are present in the oocytes and granulosa cells of various species, including humans \[[@B68-ijms-20-05342],[@B69-ijms-20-05342]\]. Some studies have revealed that melatonin treatment could limit the depletion of germ cells in the gonads during chemotherapy. In rats, melatonin administration prevents cisplatin-induced testicular toxicity and reduces sperm motility \[[@B70-ijms-20-05342]\]. Recently, it was suggested as a new fertoprotective agent option against the ovarian damage induced by chemotherapy \[[@B51-ijms-20-05342],[@B56-ijms-20-05342],[@B71-ijms-20-05342]\]. Jang et al. evaluated the protective effect of melatonin on cisplatin-treated ovaries in a mouse model \[[@B71-ijms-20-05342]\]. They demonstrated that combined treatment with melatonin and cisplatin significantly prevented primordial follicle loss in cisplatin-treated ovaries. The molecular mechanisms implicated were also analyzed. In accordance with the burnout theory, these authors showed that the protection effect of melatonin was mediated by suppressing follicular recruitment through activation of the PI3K--Akt--FoxO3a signaling pathway. Nevertheless, this protective effect was partial. In addition, the same authors recently confirmed these results and revealed that ghrelin enhances the protective effect of melatonin against cisplatin-induced ovarian failure in a mouse model \[[@B51-ijms-20-05342]\]. The molecular mechanisms implicated were evaluated and revealed that the coadministration of ghrelin and melatonin inhibited cisplatin-induced phosphorylation of Pten and FoxO3A. As FoxO3A phosphorylation induces its cytoplasmic translocation and subsequent follicular activation, the inhibition of this process maintains the primordial follicles in a dormant state. This treatment seems promising, but knowledge of the details of the molecular mechanism of melatonin's protective response against chemotherapy-induced ovarian damage and the need for evaluation of this impact on ovaries requires further studies. ### 4.2.4. mTOR Inhibitors {#sec4dot2dot4-ijms-20-05342} mTOR is a serine/threonine kinase implicated in several crucial processes such as cell growth, proliferation, autophagy, and survival \[[@B72-ijms-20-05342]\]. In mice, accelerated mTOR activity in the oocyte activates the primordial follicles, resulting in POI \[[@B73-ijms-20-05342]\]. mTOR stimulators increase the activation of primordial follicles in animal models and mTOR inhibitors block the primordial-to-primary follicle transition \[[@B74-ijms-20-05342]\]. According to these data, and after having confirmed the burnout theory, recent studies used, in a mouse model, mTOR inhibitors to preserve the ovarian reserve from cyclophosphamide-induced follicular depletion \[[@B75-ijms-20-05342],[@B76-ijms-20-05342],[@B77-ijms-20-05342]\]. Goldman et al. explored the use of the clinically approved drug everolimus (RAD001) or the inhibition of mTORC1/2 with the experimental drug INK128, showing that mTOR inhibition preserves the ovarian reserve, as measured through primordial follicle counts and serum AMH levels \[[@B75-ijms-20-05342]\]. Moreover, cyclophosphamide-treated mice had significantly fewer offspring, whereas cotreatment with mTOR inhibitors preserved normal fertility. The protective effect of everolimus was also demonstrated against cisplatin-induced gonadotoxicity in an in vivo mouse model \[[@B77-ijms-20-05342]\]. As everolimus can be used in the treatment of some breast cancers, this approach represents a very interesting option for fertility preservation during conventional chemotherapy. On the other hand, Zhou et al. observed that cotreatment of chemotherapy with rapamycin, another mTOR inhibitor, prevented the follicle growth wave caused by cyclophosphamide treatment and significantly reduced primordial follicle loss \[[@B76-ijms-20-05342]\]. Rapamycin is an inhibitor of the mTOR pathway, shown previously to inhibit the accelerated activation of primordial follicles of *Pten*−/− rat ovaries \[[@B78-ijms-20-05342]\]. 5. Vascular Damage {#sec5-ijms-20-05342} ================== 5.1. Physiopathology {#sec5dot1-ijms-20-05342} -------------------- Alterations in the ovarian stroma and vascularization are another mechanism potentially implicated in chemotherapy-induced follicle loss \[[@B3-ijms-20-05342],[@B18-ijms-20-05342]\]. Indeed, vascular damage, revealed by decreased ovarian blood flow and reduction in ovarian size, has been demonstrated in women \[[@B79-ijms-20-05342]\] and in mice following doxorubicin administration \[[@B80-ijms-20-05342]\]. In addition, the histological analysis of human ovaries previously exposed to chemotherapy revealed that a thickening and hyalinization of cortical stromal blood vessels had occurred in association with the disorganization of blood vessels in the ovarian cortex and cortical fibrosis \[[@B81-ijms-20-05342]\]. 5.2. Fertoprotective Agents {#sec5dot2-ijms-20-05342} --------------------------- ### G-CSF In light of the vascular damage induced by chemotherapy, granulocyte colony-stimulating factor (G-CSF) was tested as a fertoprotective agent. Subsequently, it was determined that treatment with G-CSF decreased chemotherapy-induced ovarian follicle loss and extended the time to premature ovarian insufficiency in female mice treated with cyclophosphamide and busulfan \[[@B82-ijms-20-05342]\]. Later, similar protective effects were found, as follicle counts and serum AMH levels were significantly increased in mice treated with cisplatin and G-CSF as compared with mice treated with cisplatin alone \[[@B83-ijms-20-05342]\]. 6. Other Molecules as Candidate Fertoprotective Agents {#sec6-ijms-20-05342} ====================================================== 6.1. GnRH Analogs {#sec6dot1-ijms-20-05342} ----------------- Tested in 1995 in rhesus monkeys, gonadotropin-releasing hormone (GnRH) analogs were the first agents considered as possible chemoprotective molecules against cyclophosphamide ovarian damage \[[@B84-ijms-20-05342]\]. Subsequent studies evaluated the possible protective effects in a rodent model with contradictory results \[[@B85-ijms-20-05342],[@B86-ijms-20-05342],[@B87-ijms-20-05342],[@B88-ijms-20-05342],[@B89-ijms-20-05342],[@B90-ijms-20-05342]\]. In a more recent study, it was proven that ovarian damage occurred even in the absence of FSH, suggesting that the inhibition of the pituitary--gonadal axis is not involved in ovarian protection during GnRH agonist treatment \[[@B90-ijms-20-05342]\]. Nevertheless, other mechanisms potentially implicated in this protective influence were suggested to be vascular effects or the upregulation of antiapoptotic molecules \[[@B91-ijms-20-05342],[@B92-ijms-20-05342],[@B93-ijms-20-05342]\]. Several clinical studies have been performed to assess, in women, the ability of GnRH analogs to protect ovaries from chemotherapy ovarian damage. POI incidence, chemotherapy-induced amenorrhea, menses recovery, or pregnancy rates were evaluated in cancer patients who received GnRH analogs or not at the time of chemotherapy treatment. Conflicting results were reported \[[@B92-ijms-20-05342],[@B94-ijms-20-05342]\]. Elsewhere, a meta-analysis of randomized clinical trials revealed diverse conclusions about the ability of GnRH analogs to preserve fertility \[[@B95-ijms-20-05342],[@B96-ijms-20-05342],[@B97-ijms-20-05342],[@B98-ijms-20-05342]\]. Nevertheless, while clinical evidence for the efficacy of this treatment is still being debated, the safety of this strategy has already been clearly demonstrated. Thus, this treatment could be proposed for all young women requiring chemotherapy, although gamete cryopreservation should be performed, if possible, for women who want to preserve their fertility. 6.2. Tamoxifen {#sec6dot2-ijms-20-05342} -------------- Tamoxifen is an estrogen receptor antagonist and is currently used as an adjuvant therapy for hormone-sensitive breast cancer. In a rodent model, the administration of tamoxifen significantly decreased doxorubicin- or cyclophosphamide-induced follicle loss \[[@B99-ijms-20-05342]\]. Similar results were obtained in cultured rat ovaries \[[@B100-ijms-20-05342]\]. Nevertheless, the molecular mechanisms of this protective effect during chemotherapy have not been discovered yet. 6.3. Other Molecules {#sec6dot3-ijms-20-05342} -------------------- In the past few years, several other molecules have been observed to decrease chemotherapy-induced ovarian damage in the drive to preserve fertility, including Chinese herbal medicine \[[@B101-ijms-20-05342]\], fennel \[[@B102-ijms-20-05342]\], sildenafil citrate \[[@B103-ijms-20-05342]\], tocotrienol \[[@B104-ijms-20-05342]\], genistein \[[@B105-ijms-20-05342]\], and erythropoietin \[[@B106-ijms-20-05342]\]. 7. Conclusions {#sec7-ijms-20-05342} ============== Improving the knowledge of the molecular mechanisms involved in chemotherapy-induced ovarian damage can lead to the development of treatments to limit follicular depletion in vivo \[[@B3-ijms-20-05342],[@B18-ijms-20-05342],[@B56-ijms-20-05342]\]. The molecular mechanisms implicated in the protective role of these different agents are more or less clear. [Table 1](#ijms-20-05342-t001){ref-type="table"} summarizes the main fertoprotective agents that have been evaluated in a mouse model, their mechanism(s) of action, and the proposed mechanism(s) to explain ovarian protection. Although histological studies of human ovaries were carried out years ago to assess the impact of the disease and treatments on the gonads \[[@B107-ijms-20-05342],[@B108-ijms-20-05342]\], this type of research is more difficult to perform today due to ethical concerns. Nowadays, the assessment of the gonadotoxicity of chemotherapy is often based on organotypic or cell culture models in vitro. In vivo studies in rodents as well as models of human ovarian xenograft are also commonly used to investigate the impact of chemotherapy on primordial follicles and the potential protective role of fertoprotective agents. Moreover, the main chemotherapies used in these fundamental studies were cisplatin, cyclophosphamide, or doxorubicin. A recent review has been published that critically discusses the damaging effects of the most common chemotherapeutic compounds (cyclophosphamide, cisplatin, and doxorubicin) on the ovaries \[[@B109-ijms-20-05342]\]. In clinical practice, the protocols applied incorporate a combination of several drugs. Therefore, the results obtained should be extrapolated to women, but caution should be used when interpreting the clinical relevance of such findings. Indeed, it is difficult to mimic the doses and protocols used, and ovarian physiology and responses to treatments can differ. For these molecules to be used in clinical practice and studied in women, it is essential that they do not interfere with the therapeutic action of chemotherapy or important physiological processes. However, as apoptosis represents the main mechanism of anticancer action, apoptosis inhibitors could reduce the anticancer effect of chemotherapy. In addition, by blocking the death of oocytes with DNA alterations, some molecules could facilitate the survival of damaged germ cells and thus promote infertility, an increased risk of spontaneous miscarriages, or fetal malformations. Finally, the PI3K pathway is a ubiquitous pathway and the molecules modulating this pathway could interfere with various physiological processes. New therapies aimed at limiting follicular loss and protecting the ovaries would be of great interest. They could be used in combination with the currently available fertility preservation techniques and administered regardless of age, pathology, or proposed treatment. Moreover, they would also prevent against hormonal deficiencies and their consequences (e.g., pubertal delay, osteoporosis). Finally, these treatments may be of particular interest to women with altered ovarian reserve parameters, in whom no fertility preservation method can be proposed. C.S. write the article; I.B., N.B. and M.G. carefully revised the manuscript. This review received no external funding. The authors declare no conflict of interest. {#ijms-20-05342-f001} {#ijms-20-05342-f002} ijms-20-05342-t001_Table 1 ###### Main molecules evaluated in an in vivo rodent model to limit chemotherapy-induced follicular depletion. --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Fertoprotective Mechanism Fertoprotective Agent References ---------------------------------------------------------------------------------------- ---------------------------------------------------------------- ----------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------- Inhibition of primordial follicular apoptosis Sphingosine 1 phosphate\ Membrane sphingolipid \[[@B32-ijms-20-05342],[@B33-ijms-20-05342],[@B34-ijms-20-05342],[@B35-ijms-20-05342],[@B37-ijms-20-05342],[@B38-ijms-20-05342]\] ceramide 1 phosphate Imatinib Competitive tyrosine-kinase inhibitor (c-Abl kinase inhibitor) \[[@B16-ijms-20-05342],[@B26-ijms-20-05342],[@B40-ijms-20-05342],[@B42-ijms-20-05342]\] GNF2 c-Abl kinase inhibitor \[[@B40-ijms-20-05342]\] LH Gonadotrophine \[[@B27-ijms-20-05342]\] Inhibition of primordial follicle recruitment AS101 PI3K modulator \[[@B45-ijms-20-05342]\] Melatonin Pineal hormone \[[@B51-ijms-20-05342],[@B71-ijms-20-05342]\] Rapamycin mTOR inhibitor \[[@B76-ijms-20-05342]\] Everolimus (and INK128) mTORC1/mTORC2 inhibitor \[[@B75-ijms-20-05342],[@B77-ijms-20-05342]\] AMH Ovarian hormone \[[@B48-ijms-20-05342],[@B62-ijms-20-05342],[@B63-ijms-20-05342]\] Several mechanisms proposed (e.g., vascular effect, follicular recruitment inhibition) GnRH analogs Inhibition of the pituitary-gonadal axis \[[@B85-ijms-20-05342],[@B86-ijms-20-05342],[@B88-ijms-20-05342],[@B89-ijms-20-05342],[@B90-ijms-20-05342]\] Vascular effect G-CSF Granulocyte colony-stimulating factor \[[@B82-ijms-20-05342],[@B83-ijms-20-05342]\] Prevention of chemotherapy nuclear activation Bortezomib Proteasome inhibitor \[[@B30-ijms-20-05342]\] ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

The online version of this article contains supplemental material. C.R. Hauck\'s present address is Zentrum für Infektionsforschung, Universität Würzburg, Röntgenring 11, D-97070 Würzburg, Germany. J. Leng\'s present address is Chemicon International Inc., Temecula, CA 92590. Introduction ============ Focal adhesion kinase (FAK)[\*](#fn1){ref-type="fn"} is a protein tyrosine kinase involved in the regulation of cell cycle progression, cell survival, and cell migration (for review see [@bib48]). Through FAK COOH-terminal domain interactions with integrin-associated proteins, FAK is activated upon cell binding to extracellular matrix proteins and forms a transient signaling complex with Src family protein tyrosine kinases (for review see [@bib51]). FAK-null (FAK^−/−^) fibroblasts form an abundance of actin stress fibers and focal contact sites in cell culture that result in refractory motility responses ([@bib24]). FAK^−/−^ reconstitution studies have shown that cellular Src (c-Src) recruitment to FAK is an initial event promoting focal contact turnover and enhanced cell motility ([@bib40]; [@bib54]). FAK also associates with activated growth factor receptors through its NH~2~-terminal FERM domain and the FAK--Src complex is important in the regulation of growth factor--stimulated cell migration ([@bib55]). Whereas activation of the FAK--Src complex facilitates the association with and/or phosphorylation of multiple signaling proteins (for review see [@bib51]), both FAK ([@bib44]) and Src act to transiently inhibit p21 Rho--GTPase activity in part through targets such as p190RhoGAP ([@bib3]). Notably, inhibition of Rho signaling can partially reverse the morphological and motility defects of FAK^−/−^ cells ([@bib9]). Many malignant human tumor samples exhibit increased FAK expression and tyrosine phosphorylation ([@bib41]). These changes are correlated with the acquisition of an invasive cell phenotype and increased metastasis ([@bib33]; [@bib7]). Since specific signaling pathways associated with FAK and cell invasion have not been elucidated, it has been hypothesized that FAK-mediated cell invasion may represent FAK effects on tumor cell motility ([@bib18]; [@bib56]). Notably, the viral Src (v-Src) oncogene can directly transform a variety of cell types and can promote an invasive cell phenotype in vitro and experimental metastases in vivo ([@bib16]; [@bib1]). FAK is a v-Src substrate, and the formation of a v-Src--FAK signaling complex is associated with increased cell invasion and the generation of invadopodia ([@bib19]). Stable expression of a dominant-negative (DN) inhibitor of FAK termed FAK-related nonkinase (FRNK) in v-Src--transformed NIH3T3 fibroblasts resulted in the inhibition of Matrigel cell invasion in vitro and experimental metastasis in vivo without effects on cell motility or v-Src--enhanced cell growth ([@bib20]). These studies with v-Src and FRNK suggest that the role of FAK in promoting cell motility and/or invasion might be distinguishable events. Cell invasion is a complex process that can be initiated by alterations in integrin surface expression ([@bib21]), by the release or activation of proteases that degrade the extracellular matrix ([@bib37]), and by changes in gene expression during cell transformation ([@bib42]). Although FAK may not be a critical determinant regulating v-Src--altered cell morphology or cell growth ([@bib46]), we show that v-Src transformation reverses the integrin-stimulated motility defects of FAK^−/−^ fibroblasts as does FAK reexpression. However, FAK^−/−^ v-Src cells fail to exhibit an invasive phenotype as do v-Src--transformed FAK-reconstituted cells. For the first time, we show that serum stimulation promotes FAK accumulation at lamellipodia and that this is associated with increased Rac and c-Jun NH~2~-terminal kinase (JNK) activation. Our studies support the novel hypothesis that FAK recruitment to lamellipodia/invadopodia promotes cell invasion and FAK localization to focal contacts promotes integrin-stimulated cell motility, each in part through the activation of distinct signaling pathways. Results ======= v-Src transformation promotes fibronectin-stimulated FAK**^−^** ^/**−**^ cell motility -------------------------------------------------------------------------------------- To establish a model system where the role of FAK in cell transformation could be defined, primary FAK^−/−^, FAK-reconstituted (DA2), or FAK^+/+^ fibroblasts were transformed by stable v-Src expression ([Fig. 1](#fig1){ref-type="fig"}) . Importantly, v-Src protein levels ([Fig. 1](#fig1){ref-type="fig"} A) and in vitro kinase (IVK) activity (unpublished data) were equal in pooled populations of FAK^−/−^ and DA2 cells. FAK^−/−^ cells possess a rounded cell shape when plated onto fibronectin (FN), and FAK reexpression rescues this defect by promoting DA2 cell spreading ([Fig. 1](#fig1){ref-type="fig"} B). v-Src promoted a transformed fusiform morphology with reduced levels of actin stress fibers in both FAK^−/−^ and DA2 cells ([Fig. 1](#fig1){ref-type="fig"} B). Pooled populations of FAK^+/+^ v-Src cells exhibited an intermediate morphological conversion ([Fig. 1](#fig1){ref-type="fig"} B) that was associated with lower levels of v-Src expression ([Fig. 1](#fig1){ref-type="fig"} A). These results support the conclusion that FAK is not a critical determinant of v-Src--stimulated morphological changes ([@bib46]). {#fig1} Refractory FAK^−/−^ cell motility is due in part to the enhanced stability of focal contact structures ([@bib24]) and to defects in Rho regulation ([@bib44]; [@bib9]). Surprisingly, FAK^−/−^ v-Src cells exhibited an approximately sixfold increase in FN-stimulated cell migration compared with FAK^−/−^ cells. This level of FAK^−/−^ v-Src cell motility was equivalent to DA2, DA2 v-Src, and FAK^+/+^ v-Src cell migration ([Fig. 1](#fig1){ref-type="fig"} C). As integrins suppress Rho activity through Src-dependent p190RhoGAP tyrosine phosphorylation ([@bib3]), p190RhoGAP tyrosine phosphorylation was analyzed after FN stimulation of cells ([Fig. 1](#fig1){ref-type="fig"} D). Elevated p190RhoGAP tyrosine phosphorylation was observed in lysates from FN-stimulated FAK^−/−^ v-Src, DA2, and DA2 v-Src compared with FAK^−/−^ cells. Our results show that both v-Src and FAK can promote cell motility, enhanced p190RhoGAP tyrosine phosphorylation, and morphological alterations of FAK^−/−^ cells. v-Src localizes to lamellipodia in FAK-containing cells ------------------------------------------------------- High levels of v-Src expression result in the turnover of peripheral focal contacts and the formation of ventral adhesion structures termed podosomes ([@bib58]). In the absence of serum, the fusiform-shaped FAK^−/−^ v-Src cells ([Fig. 1](#fig1){ref-type="fig"} B) make podosomes (unpublished data). In the presence of serum, FAK^−/−^ v-Src cells assume a more spread morphology and make peripheral focal contacts as visualized by integrin-linked kinase (ILK) immunofluorescence (IF) staining ([Fig. 2](#fig2){ref-type="fig"} A). Interestingly, v-Src was not strongly localized to peripheral focal contact sites in FAK^−/−^ cells but instead exhibited a punctate perinuclear distribution ([Fig. 2](#fig2){ref-type="fig"} A). Nevertheless, v-Src expression increased FAK^−/−^ cell motility in scratch assays ([Fig. 2](#fig2){ref-type="fig"} B). However, FAK^−/−^ v-Src cell repopulation of the wounded area was less than DA2 cells ([Fig. 2](#fig2){ref-type="fig"} B). {#fig2} In contrast to FAK^−/−^ v-Src cells, DA2 v-Src cells plated on FN in the presence of serum exhibited extensive spreading and lamellipodia formation ([Fig. 2](#fig2){ref-type="fig"} A). Both v-Src and ILK were similarly distributed within lamellipodial regions of DA2 v-Src cells ([Fig. 2](#fig2){ref-type="fig"} A). However, DA2 v-Src cells also exhibited lower levels of wound closure activity compared with DA2 cells (unpublished data). As analyzed by Boyden Chamber chemotaxis assays, DA2 v-Src cell motility was ∼2.5-fold less than DA2 cells ([Fig. 2](#fig2){ref-type="fig"} C). Since cell motility responses are dependent on a balance of focal contact formation and turnover ([@bib34]), the lower levels of DA2 v-Src cell motility are consistent with the combined actions of FAK and v-Src in promoting focal contact turnover and reduced cell adhesion ([@bib13]; [@bib10]). Interestingly, although FAK^−/−^ v-Src and DA2 v-Src cells exhibited distinctly different cell morphologies in the presence of serum ([Fig. 2](#fig2){ref-type="fig"} A), they exhibited equivalent levels of serum-stimulated motility ([Fig. 2](#fig2){ref-type="fig"} C). FAK expression promotes Matrigel invasion ----------------------------------------- In addition to altered growth and motility responses, v-Src can promote an invasive phenotype ([@bib16]). This can be evaluated in three-dimensional (3-D) invasion assays wherein the cells degrade and move through a reconstituted (Matrigel) extracellular matrix barrier. Surprisingly, FAK^−/−^ v-Src cells did not significantly invade through Matrigel ([Fig. 3, A and B](#fig3){ref-type="fig"}) . Normal DA2 cells exhibited greater cell invasion than FAK^−/−^ v-Src cells. Both DA2 v-Src and FAK^+/+^ v-Src cells exhibited two- to threefold increased invasion compared with DA2 cells and eight to tenfold increased invasion compared with FAK^−/−^ v-Src cells ([Fig. 3, A and B](#fig3){ref-type="fig"}). The differences in DA2 v-Src compared with FAK^−/−^ v-Src cell invasion were not due to altered motility responses, since these cells exhibited equal wound closure activity on Matrigel-coated glass slides in the presence of serum ([Fig. 3](#fig3){ref-type="fig"} C). These results show that v-Src promotes an invasive phenotype in FAK-reconstituted cells but not in FAK-deficient cells. {ref-type="fig"}.](200212114f3){#fig3} v-Src--FAK signaling enhances basal and serum-stimulated JNK activation ----------------------------------------------------------------------- v-Src transformation results in the activation of multiple downstream targets such as the Ras--extracellular signal--regulated kinase (ERK)2/MAP kinase cascade, the Rac-JNK/SAP kinase cascade, phosphatidylinositol (PI) 3-kinase--Akt, and the phosphorylation of STATs (for review see [@bib27]). To determine the contribution of FAK- to v-Src--stimulated signaling, whole cell lysates from serum-starved cells were analyzed with a series of phospho-specific antibodies ([Fig. 4](#fig4){ref-type="fig"} A). Anti-phosphotyrosine (pTyr) blotting and phospho-specific antibodies to STAT3 (pSTAT3) revealed that v-Src promoted similar increases in tyrosine phosphorylation in both FAK^−/−^ and DA2 cells ([Fig. 4](#fig4){ref-type="fig"} A, lanes 2 and 4). Phospho-specific antibodies directed to Akt (pAkt) and to ERK (pERK) showed that these targets were equally activated in v-Src--transformed FAK^−/−^ and DA2 cells ([Fig. 4](#fig4){ref-type="fig"} A, lanes 6 and 8). However, blotting with phospho-specific antibodies to JNK (pJNK) revealed that this pathway was constitutively activated only in DA2 v-Src cells ([Fig. 4](#fig4){ref-type="fig"} A, lane 8). ###### **FAK expression promotes elevated v-Src and serum-stimulated JNK activity.** (A) Whole cell lysates from the indicated serum-starved cells were blotted with phospho-specific antibodies to tyrosine (pTyr), Tyr-705 of STAT3 (pSTAT3), Ser-473 of Akt (pAkt), Thr-202 and Tyr-204 of ERK2 (pERK), and Thr-183 and Tyr-185 of JNK (pJNK). Equal protein expression was verified by reprobing blots with antibodies to STAT3, Akt, ERK2, or JNK1. Numbers represent *M* ~r~ in kD. (B) ERK2 IP/IVK assays were performed with lysates from the indicated starved or 20% FBS-stimulated cells. Values are means ± SD from two independent experiments that were normalized for protein levels in IPs and represent fold differences in ERK2 activity from starved FAK^−/−^ cells. (Inset) Representative image of MBP phosphorylation. (C) JNK IP/IVK assays were performed as described for ERK2 above. (Inset) Representative image of GST-ATF2 phosphorylation.   Since phospho-specific blotting provides only a qualitative measure of target protein activation, IVK assays were performed with lysates from starved or serum-stimulated cells ([Fig. 4, B and C](#fig4){ref-type="fig"}). Although no significant differences in basal ERK2 activity were found between v-Src--transformed FAK^−/−^ and DA2 cells ([Fig. 4](#fig4){ref-type="fig"} B, lanes 2 and 4), normal and v-Src--transformed DA2 cells exhibited slightly increased serum-stimulated ERK2 IVK activity compared with FAK^−/−^ and FAK^−/−^ v-Src cells, respectively ([Fig. 4](#fig4){ref-type="fig"} B, lanes 5--8). This result is consistent with previous findings that FAK enhances the integrin signaling requirement for serum-stimulated ERK2 activation ([@bib45]). Analyses of basal and serum-stimulated Akt IVK activity yielded similar results as shown for ERK2 (unpublished data). However, JNK IVK activity was elevated four- to fivefold in DA2 v-Src lysates compared with FAK^−/−^ v-Src cells under both starved and serum-stimulated conditions ([Fig. 4](#fig4){ref-type="fig"} C). Additionally, the serum-stimulated JNK IVK activity in DA2 lysates was approximately twofold higher than FAK^−/−^ v-Src cells ([Fig. 4](#fig4){ref-type="fig"} C, lanes 6 and 7). Since the level of serum-stimulated JNK activity correlated with increased DA2 and DA2 v-Src cell invasion ([Fig. 3](#fig3){ref-type="fig"} A and [Fig. 4](#fig4){ref-type="fig"} C), our results support the hypothesis that serum-stimulated and FAK-enhanced signaling to JNK may promote cell invasion. Adenoviral-mediated FAK expression rescues FAK**^−^** ^/**−**^ v-Src invasion defects ------------------------------------------------------------------------------------- To establish the importance of FAK in promoting cell invasion, a transient adenovirus (Ad)-mediated expression strategy was employed to express various hemagglutinin (HA)-tagged FAK constructs in FAK^−/−^ v-Src cells ([Fig. 5](#fig5){ref-type="fig"}) . HA-FAK expression is controlled by the coinfection of cells with a recombinant Ad construct producing the tetracycline transactivator (TA) protein, and this serves as a control for the effects of Ad uptake into cells. Previous studies showed that FAK^−/−^ fibroblasts can be readily infected with Ad constructs at low multiplicity of infection (MOI) ([@bib35]). Ad infection of FAK^−/−^ v-Src cells yielded equal expression of HA-tagged FAK, Tyr-397--mutated FAK (Y397F), kinase-inactive FAK (K454R), or FAK mutated within both proline-rich SH3-binding sites in the COOH-terminal domain (Pro-null) ([Fig. 5](#fig5){ref-type="fig"} A). Surprisingly, all FAK constructs were tyrosine phosphorylated in FAK^−/−^ v-Src cells, and phospho-specific antibody blotting revealed that wild-type (WT), K454R, and Pro-null FAK were phosphorylated at Tyr-397 ([Fig. 5](#fig5){ref-type="fig"} A). However, only WT FAK expression promoted an approximately fourfold increase in FAK^−/−^ v-Src Matrigel cell invasion compared with Ad-TA--infected cells ([Fig. 5](#fig5){ref-type="fig"} B). This corresponded to an ∼80% rescue of FAK^−/−^ v-Src cell invasion compared with DA2 v-Src cells. {#fig5} Since the v-Src--FAK signaling complex is localized within invadopodia that emerge from the Matrigel barrier ([@bib19]), IF staining with HA tag or FAK phospho-specific antibodies was used to analyze the activation state of Ad-FAK within FAK^−/−^ v-Src cells during Matrigel invasion ([Fig. 5](#fig5){ref-type="fig"} C). WT FAK expression was localized to the tips of invadopodia and was highly phosphorylated at Tyr-397 ([Fig. 5](#fig5){ref-type="fig"} C). However, the FAK mutants that did not function to promote cell invasion also did not show invadopodia emergence through the Matrigel-filled 8-μm pores. Together, the IF results support the importance of FAK Tyr-397 phosphorylation during the processes of cell invasion. To determine whether FAK-enhanced cell invasion was associated with increased signaling to JNK, endogenous JNK IVK activity was measured after Ad-FAK infection of FAK^−/−^ v-Src cells ([Fig. 5](#fig5){ref-type="fig"} D). WT FAK expression promoted a greater than fourfold increase in JNK activity compared with Ad-TA--infected FAK^−/−^ v-Src cells, whereas the FAK mutants did not activate JNK. The approximately fourfold increase in FAK-stimulated JNK activity ([Fig. 5](#fig5){ref-type="fig"} D) was correlated with similar changes in FAK^−/−^ v-Src cell invasion ([Fig. 5](#fig5){ref-type="fig"} B) and was consistent with the differences in JNK activity between FAK^−/−^ v-Src and DA2 v-Src cells ([Fig. 4](#fig4){ref-type="fig"} C). The importance of JNK activation promoting cell invasion of FAK-expressing cells was confirmed by the dose-dependent inhibition of DA2 v-Src Matrigel invasion by SP600125, a pharmacological inhibitor of JNK kinase activity ([Fig. 5](#fig5){ref-type="fig"} E). FAK localizes to lamellipodia and promotes Rac activation --------------------------------------------------------- In many cells, JNK activation is regulated in part through small GTPases such as Rac1 ([@bib11]). To determine whether there were differences in Rac activity between FAK^−/−^ and DA2 cells, the Rac-binding region (residues 67--150) of p65PAK kinase (GST-PAK) was used as an affinity probe for activated Rac in pull-down assays ([Fig. 6](#fig6){ref-type="fig"} A). Under serum-starved conditions, activated Rac was only detected in lysates from DA2 v-Src cells, whereas total Rac expression levels was equal in all cells. After serum stimulation, Rac activation was greater in DA2 and DA2 v-Src cells compared with FAK^−/−^ and FAK^−/−^ v-Src cells, respectively ([Fig. 6](#fig6){ref-type="fig"} A). To evaluate the contribution of FAK expression in promoting Rac activation, FAK^−/−^ v-Src cells were infected with Ad-FAK constructs. WT FAK promoted endogenous Rac activation in FAK^−/−^ v-Src cells as determined by GST-PAK binding ([Fig. 6](#fig6){ref-type="fig"} B). No Rac activation was detected in cells infected with Y397F, K454R, or Pro-null FAK. These results support the conclusion that FAK-mediated Rac activation is associated with increased cell invasion. ###### **FAK activates Rac and serum stimulation promotes GFP-FAK localization to lamellipodia.** (A) Rac activation was visualized in the indicated cells by pull-down assays using GST-PAK (67--150) followed by Rac blotting. Total Rac expression is shown in whole cell lysates. (B) FAK^−/−^ v-Src cells were infected with Ad-TA or the indicated Ad-FAK constructs, and Rac activation was evaluated by GST-PAK binding. (C) FAK^−/−^ v-Src cells were transfected with GFP-FAK, starved, and then stimulated with 10% FBS. Time-lapse confocal video-microscopy at 1-min intervals was obtained over 3 h (Video 1 available at <http://www.jcb.org/cgi/content/full/jcb.200212114/DC1>). Panels show a 9-min time series where GFP-FAK is stably localized to focal contacts and transiently recruited to lamellipodia extensions (arrows).   Since Rac activation occurs in membrane-associated regions in close proximity to lamellipodia, whereas integrin-stimulated FAK signaling is initiated from focal contact sites, FAK^−/−^ v-Src cells were transfected with GFP-FAK to visualize the real-time localization of FAK ([Fig. 6](#fig6){ref-type="fig"} C). Not surprisingly, GFP-FAK was localized to ventral and perimeter focal contact sites in FAK^−/−^ v-Src cells. However, upon serum stimulation a transient accumulation of GFP-FAK was also visualized at distinct plasma membrane regions that were associated with the initiation of lamellipodial projections ([Fig. 6](#fig6){ref-type="fig"} C and see Video 1 available at <http://www.jcb.org/cgi/content/full/jcb.200212114/DC1>). This GFP-FAK lamellipodial association lasted from 1 to 3 min, and it was not detectably accompanied by the loss of GFP-FAK from focal contact sites. Instead, the lamellipodial-associated FAK became nucleated at numerous point contact sites that formed behind lamellipodial projections. The GFP-FAK--containing point contacts were transient structures with a small percentage becoming mature focal contact sites that moved toward the center of the cell ([Fig. 6](#fig6){ref-type="fig"} C). These analyses show that serum stimulation promotes the transient localization of FAK to lamellipodia. This places FAK at an appropriate intracellular site to facilitate signaling leading to Rac and JNK activation. p130Cas as a mediator of FAK-initiated signals to JNK ----------------------------------------------------- Several different cellular proteins associate with FAK through SH3-mediated binding interactions at two proline-rich sites within the FAK COOH-terminal domain ([@bib48]). Since a trimeric complex between Src, FAK, and p130Cas is stabilized by Src SH2 binding to FAK Tyr-397, Src SH3 binding to p130Cas, and p130Cas SH3 binding to FAK ([@bib50]), recombinant Ad constructs were used to overexpress WT p130Cas or SH3 domain-mutated p130Cas (Cas ΔSH3) in FAK^−/−^ v-Src cells ([Fig. 7](#fig7){ref-type="fig"} A). Although Ad-FAK expression increased endogenous p130Cas pTyr levels, both WT Cas and Cas ΔSH3 were highly tyrosine phosphorylated in FAK^−/−^ v-Src cells ([Fig. 7](#fig7){ref-type="fig"} A). However, only WT Cas functioned to rescue FAK^−/−^ v-Src cell invasion defects approximately equal to Ad-FAK expression ([Fig. 7](#fig7){ref-type="fig"} B). Importantly, Cas-stimulated cell invasion was associated with elevated JNK IVK activity in FAK^−/−^ v-Src cells ([Fig. 7](#fig7){ref-type="fig"} C). Cas ΔSH3 was highly tyrosine phosphorylated but did not stimulate JNK activity ([Fig. 7](#fig7){ref-type="fig"} C). {#fig7} Our results show that Cas SH3 domain function is needed for v-Src--stimulated cell invasion. Although Crk SH2 binding to Cas within the substrate domain can promote signaling to JNK, overexpression of a Cas substrate domain-deleted mutant blocked FN- but not v-Src--stimulated JNK activation ([@bib12]). Therefore, to elucidate a Cas SH3 domain--mediated and potential Crk-independent linkage to Rac and JNK, GST pull-down assays were performed with the Cas SH3 domain and compared with the binding activities of SH3 domains from other proteins ([Fig. 7](#fig7){ref-type="fig"} D). Whereas the Cas SH3 domain can bind to targets such as guanine nucleotide exchange factor C3G in addition to FAK ([@bib28]), we identified the guanine nucleotide exchange factor Dock180 as a target of both the Crk and Cas SH3 domains within lysates of FAK^−/−^ v-Src cells ([Fig. 7](#fig7){ref-type="fig"} D). FAK expression facilitates the formation of a Dock180 signaling complex ----------------------------------------------------------------------- To determine the potential participation of Dock180 in FAK-mediated signaling, serum-stimulated FAK^−/−^ v-Src and DA2 v-Src cells were analyzed by co-immunoprecipitation (IP) analyses with antibodies to Crk, p130Cas, Dock180, or FAK ([Fig. 7](#fig7){ref-type="fig"} E). Strikingly, highly tyrosine phosphorylated multiprotein complexes containing Crk, v-Src, p130Cas, Dock180, and HA-tagged FAK were isolated from DA2 v-Src but not FAK^−/−^ v-Src lysates ([Fig. 7](#fig7){ref-type="fig"} E). Importantly, equal amounts of Crk, p130Cas, or Dock180 were immunoprecipitated using the same primary antibodies, whereas multiprotein complexes were isolated only from DA2 v-Src lysates. Additionally, antibodies to FAK did not co-IP protein complexes in lysates from FAK^−/−^ v-Src cells. For Crk IPs, there was a strong association with v-Src and a weak association with highly tyrosine-phosphorylated p130Cas in FAK^−/−^ v-Src lysates. In DA2 v-Src lysates, there was an increased association of FAK, Cas, and Dock180 with Crk without a change in either Crk tyrosine phosphorylation or v-Src association. For p130Cas, the Cas immunoreactive band was super shifted in DA2 v-Src lysates, and this was correlated with increased Cas tyrosine phosphorylation and the strong association with HA-FAK and v-Src ([Fig. 7](#fig7){ref-type="fig"} E). Both Dock180 and Crk were associated with Cas only in DA2 v-Src lysates. For Dock180 IPs, strong associations with Cas and Crk and weaker associations with v-Src and HA-FAK were detected only in DA2 v-Src cell lysates. For FAK IPs, there was a strong association with v-Src and weaker associations with Cas and Crk. No association of Dock180 was visualized in FAK IPs, indicating that FAK expression in DA2 v-Src cells facilitates the formation of a Dock180-containing signaling complex likely through direct and indirect multiprotein interactions. FAK and Rac regulate cell invasion and matrix metalloproteinase-9 secretion --------------------------------------------------------------------------- Since the FAK^−/−^ v-Src cells exhibit equivalent motility responses as DA2 v-Src cells, experiments were performed to determine whether enhanced matrix metalloproteinase (MMP) secretion and/or activity were associated with FAK-mediated cell invasion ([Fig. 8](#fig8){ref-type="fig"}) . Cells were analyzed by Matrigel invasion assays ([Fig. 8](#fig8){ref-type="fig"} B) and by gelatin zymography of cell conditioned media ([Fig. 8](#fig8){ref-type="fig"} C). FAK^−/−^ and FAK^−/−^ v-Src--conditioned media contained secreted but not activated 72 kD MMP-2 ([Fig. 8](#fig8){ref-type="fig"} C, lanes 1 and 2). In zymography assays, pro--MMP-2 exhibits high levels of apparent activity due to the dissociation of TIMP inhibitor--MMP complexes during gel electrophoresis. Activation of pro--MMP-2 occurs in part through proteolytic cleavage ([@bib38]). DA2-conditioned media exhibited both pro- and activated MMP-2 ([Fig. 8](#fig8){ref-type="fig"} C, lane 3). {#fig8} Additional bands of gelatinase activity at ∼92 kD were also weakly visualized in DA2 cell conditioned media, prominently visualized in DA2 v-Src cell media, but not detected in FAK^−/−^ v-Src cell media ([Fig. 8](#fig8){ref-type="fig"} C, lanes 2--4). This activity corresponds to the predicted size of MMP-9. By protein blotting of conditioned media samples, MMP-9 was secreted only from DA2 v-Src cells, whereas equal levels of MMP-2 were secreted by DA2 v-Src and FAK^−/−^ v-Src cells ([Fig. 8](#fig8){ref-type="fig"} D). RT-PCR analyses revealed that MMP-9 mRNA levels were 6.5-fold higher in DA2 v-Src compared with FAK^−/−^ v-Src cells ([Fig. 8](#fig8){ref-type="fig"} E). The importance of increased MMP-9 secretion as a contributing factor promoting DA2 v-Src invasion was verified by the dose-dependent inhibition of DA2 v-Src cell invasion by recombinant TIMP-1 addition ([Fig. 8](#fig8){ref-type="fig"} F). Since only low levels of Rac activation were detected in lysates of FAK^−/−^ v-Src cells ([Fig. 6](#fig6){ref-type="fig"} A), Ad infection was used to deliver activated (Gln-61 mutated to Leu, Q61L) Flag-tagged Rac to cells to test whether activated Rac could promote cell invasion ([Fig. 8](#fig8){ref-type="fig"} A). However, Rac Q61L expression in FAK^−/−^ cells did not promote Matrigel invasion or detectable changes in MMP secretion ([Fig. 8, B and C](#fig8){ref-type="fig"}, lane 5). Additionally, Q61L Rac expression led to only modest increases in JNK activity within FAK^−/−^ cells ([Fig. 8](#fig8){ref-type="fig"} G). Since Rho activity is elevated in FAK^−/−^ cells ([@bib44]), it is possible that Rac-mediated signaling may be inhibited through the constitutive activation of Rho targets such as p160Rock ([@bib9]; [@bib60]). In contrast to FAK^−/−^ cells, equivalent Rac Q61L expression in DA2 cells increased cell invasion approximately threefold compared with mock-infected DA2 cells ([Fig. 8](#fig8){ref-type="fig"} B, lanes 3 and 7). The level of Rac-induced DA2 cell invasion was equal to DA2 v-Src cells and was associated with elevated MMP-9 secretion as detected by zymography ([Fig. 8](#fig8){ref-type="fig"} C, lane 7). Rac Q61L also functioned to promote high levels of JNK activity in DA2 compared with FAK^−/−^ cells ([Fig. 8](#fig8){ref-type="fig"} G). Since the JNK SP600125 inhibitor blocked DA2 v-Src cell invasion ([Fig. 5](#fig5){ref-type="fig"} E) and reduced MMP-9 secretion from DA2 v-Src cells (unpublished data), the Rac Q61L results with DA2 cells support the hypothesis that FAK expression coupled with Rac activation can promote an invasive cell phenotype in part by enhancing MMP-9 expression and secretion. Rac Q61L expression in FAK^−/−^ v-Src cells also resulted in the rescue of FAK^−/−^ v-Src invasion defects ([Fig. 8](#fig8){ref-type="fig"} B, lane 6). High levels of JNK IVK activity were stimulated by Rac Q61L expression in FAK^−/−^ v-Src cells ([Fig. 8](#fig8){ref-type="fig"} G) compared with equivalent low levels of JNK activity in lysates of FAK^−/−^ and FAK^−/−^ v-Src cells ([Fig. 4](#fig4){ref-type="fig"} C and unpublished data). These results are consistent with the hypothesis that either v-Src or FAK expression within FAK^−/−^ cells creates a permissive environment for maximal Rac activation. To this end, zymography analyses showed that Rac Q61L expression promoted both MMP-2 activation and increased MMP-9 secretion in FAK^−/−^ v-Src cells compared with mock-infected cells ([Fig. 8](#fig8){ref-type="fig"} C, lanes 2 and 6). Since zymography evaluates only relative MMP activity in the absence of associated TIMP inhibitors, solution gelatinase activity measurements were also performed and showed that total gelatinase activity in conditioned media samples paralleled the level of cell invasion (unpublished data). Although Rac Q61L expression slightly increased DA2 v-Src cell invasion further, the number of invasive cells enumerated on the lower cell membrane ([Fig. 3](#fig3){ref-type="fig"} B) was close to saturation ([Fig. 8](#fig8){ref-type="fig"} B). In contrast, Ad-mediated expression of DN Rac (Thr-17 mutated to Asn) in DA2 v-Src cells blocked cell invasion and reduced both MMP-2 and MMP-9 expression (unpublished data). These results show that FAK and Rac signaling can promote changes in MMP-2 activity or MMP-9 expression associated with cell invasion. The importance of JNK activity in enhancing cell invasion --------------------------------------------------------- Since DA2 v-Src cell invasion ([Fig. 5](#fig5){ref-type="fig"} E) and MMP-9 gene expression are sensitive to SP600125 JNK inhibition ([@bib53]), Ad-mediated expression of WT or DN JNK1 (Thr-180 and Tyr-182 mutated to Ala and Phe) was used to test the role of JNK activation as a critical Rac target promoting cell invasion ([Fig. 9](#fig9){ref-type="fig"}) . WT JNK overexpression in FAK^−/−^ v-Src cells led to its activation as evaluated by phospho-JNK blotting ([Fig. 9](#fig9){ref-type="fig"} A). Serum stimulation in combination with WT JNK overexpression was sufficient to promote increased signaling to downstream targets such as the c-Jun transcription factor as measured by phospho-specific blotting ([Fig. 9](#fig9){ref-type="fig"} B). These signaling changes were sufficient to promote a four- to fivefold increase in FAK^−/−^ v-Src cell Matrigel invasion equivalent to DA2 v-Src cell invasion ([Fig. 9](#fig9){ref-type="fig"} C). {#fig9} In contrast, DN JNK overexpression resulted in the inhibition of JNK in DA2 v-Src cells as detected by phospho-JNK blotting ([Fig. 9](#fig9){ref-type="fig"} A) and the 2.5-fold inhibition of DA2 v-Src cell invasion ([Fig. 9](#fig9){ref-type="fig"} C). By RT-PCR analyses, WT JNK overexpression promoted a twofold increase in MMP-9 mRNA levels in FAK^−/−^ v-Src cells ([Fig. 9](#fig9){ref-type="fig"} D). These JNK-induced changes were dependent on a serum stimulus as only minor changes in MMP-9 mRNA levels were detected in unstimulated FAK^−/−^ v-Src cells overexpressing JNK (unpublished data). Notably, gelatin zymography analyses of Ad-JNK--infected FAK^−/−^ v-Src cells showed elevated secretion of MMP-9 and activation of MMP-2 compared with mock-infected FAK^−/−^ v-Src cells ([Fig. 9](#fig9){ref-type="fig"} E). These results show the importance of JNK activity in promoting increased cell invasion. Discussion ========== FAK plays important roles in promoting growth factor-- and integrin-stimulated cell motility in both normal and transformed cells. FAK expression and tyrosine phosphorylation are elevated as a function of human tumor cell malignancy, and these changes are associated with increased tumor metastasis ([@bib7]). Whereas FAK-increased cell motility is a fundamental component of cell invasion, the significance of our findings is that defects in FAK^−/−^ cell motility are distinct from the role of FAK in promoting cell invasion. This connection was determined by the fact that both FAK and v-Src can rescue FAK^−/−^ motility defects, yet FAK expression was required for the generation of an invasive cell phenotype. Localization influences signaling connections: FAK at focal contacts -------------------------------------------------------------------- FAK associates with a variety of cell surface receptors, cytoskeletal-associated proteins, and SH2/SH3-containing adaptor proteins. Multiple connections link FAK to the activation of ERK2 kinase, PI 3-kinase, or JNK kinase signaling cascades ([@bib51]; [@bib48]). However, it is not known whether FAK activation leads to either generalized or specific downstream signaling events. Our results support the hypothesis that FAK localization to focal contacts promotes integrin-stimulated cell motility and that FAK accumulation at lamellipodia/invadopodia promotes cell invasion, each in part through the activation of distinct signaling pathways. FAK^−/−^ fibroblasts form an abundance of actin stress fibers and focal contacts in cell culture, and these abnormalities limit cell motility responses ([@bib24]). As summarized in [Fig. 10](#fig10){ref-type="fig"} A, FAK reconstitution reverses FAK^−/−^ cell defects, FAK activity is dependent on its localization to focal contacts, and FAK--Src signaling enhances the extent and duration of FN-stimulated ERK2 kinase activation ([@bib54]; [@bib32]). v-Src transformation of FAK^−/−^ cells resulted in the constitutive activation of ERK2 kinase activity, and FAK^−/−^ v-Src cells exhibited equivalent haptotaxis motility as FAK-reconstituted cells. MEK-ERK2 signaling is required for haptotaxis motility ([@bib32]), and ERK2 can promote cell contractility through increased myosin light chain phosphorylation ([@bib30]). However, FAK^−/−^ cell motility defects are associated with excessive contractility and are partially reversed through the inhibition of myosin light chain kinase or Rho kinase signaling ([@bib9]). ![**Models of differential FAK signaling connections at focal contacts or lamellipodia/invadopodia.** (A) Integrin-associated FAK and Src activation promotes increased p190RhoGAP tyrosine phosphorylation or ERK2/MAP kinase activation. FAK--Src modulation of p21 Rho activity in FAK^−/−^ cells is associated with increased cell motility, focal contact turnover, and actin cytoskeletal rearrangements. (B) In either EGF-stimulated human adenocarcinoma cells ([@bib18]) or serum-stimulated v-Src--transformed fibroblasts, FAK functions to coordinate a Src-p130Cas signaling complex localized to lamellipodia in two-dimensional cell culture or to invadopodia in 3-D invasion assays. Through direct Cas SH3-mediated and indirect Crk-mediated interactions with the Dock180 guanine-nucleotide exchange factor, the FAK--Src signaling complex promotes Rac and JNK activation with effects upon gene expression. Although PI 3-kinase is a direct target of both v-Src and Rac ([@bib36]; [@bib5]), FAK^−/−^ v-Src cells exhibit defects in JNK but not Akt activation. Overexpression of JNK in FAK^−/−^ v-Src cells coupled with serum stimulation promoted c-Jun Ser-63 phosphorylation, MMP-9 expression, MMP-2 activation, and rescued the FAK^−/−^ v-Src cell invasion defects.](200212114f10){#fig10} v-Src--mediated ERK2 activation also has been associated with the disruption of actin stress fibers in part through the down-regulation of Rho kinase expression ([@bib43]). However, transfection of FAK^−/−^ cells with activated MEK1 did not rescue FAK^−/−^ haptotaxis defects ([@bib32]). Since FAK reexpression in FAK^−/−^ cells transiently inhibits Rho activity upon FN stimulation cells ([@bib44]), p190RhoGAP was investigated as an upstream regulator of Rho. Both v-Src and FAK expression enhanced FN-stimulated p190RhoGAP tyrosine phosphorylation compared with FAK^−/−^ cells, and v-Src--mediated p190RhoGAP phosphorylation is associated with actin and focal contact reorganization ([@bib14]). Although integrins can suppress Rho activity through c-Src--dependent p190RhoGAP tyrosine phosphorylation ([@bib3]), it is possible that other FAK-associated proteins, such as Graf, may additionally contribute to Rho regulation ([@bib59]). The Graf SH3 domain binds to FAK COOH-terminal proline-rich sites that are required for full FAK function in promoting FAK^−/−^ haptotaxis motility ([@bib54]). Our results showing that v-Src does not strongly localize to focal contacts in FAK^−/−^ cells yet rescues haptotaxis motility is also consistent with recent studies, showing that Src catalytic but not scaffolding function is needed for integrin-regulated cell migration ([@bib8]). FAK signaling at lamellipodia or invadopodia -------------------------------------------- Although transient suppression of Rho activity may alleviate contractile forces that would otherwise impede protrusion at the leading edge of migrating cells ([@bib3]), FAK^−/−^ v-Src cells did not exhibit serum-stimulated membrane ruffling activity or possess an invasive phenotype. Notably, both Rac and JNK activation were attenuated in cells lacking FAK. In contrast, transient or stable FAK expression in FAK^−/−^ v-Src cells promoted membrane ruffling, and serum stimulation resulted in the localization of both FAK and v-Src either to lamellipodia or invadopodia cell projections. Importantly, the accumulation of FAK at lamellipodia was not accompanied by the loss of FAK at focal contacts, indicating that distinct intracellular pools of FAK may differentially respond to particular extracellular signals. Although the signals and mechanism(s) promoting FAK recruitment to lamellipodia are not known, the localization of FAK to membrane ruffles places FAK at an appropriate intracellular site to facilitate Rac activation. As summarized in [Fig. 10](#fig10){ref-type="fig"} B, we showed that FAK expression within FAK^−/−^ cells facilitated the formation of a multiprotein signaling complex comprised of v-Src, p130Cas, Crk, and Dock180. The role of FAK in linking serum stimulation to Rac and JNK activation is not limited to reconstituted FAK^−/−^ cells, since FAK antisense treatment of human A549 carcinoma cells disrupted the EGF-stimulated formation of a c-Src--p130Cas signaling complex ([@bib18]). Loss of FAK expression in A549 cells also inhibited EGF-stimulated JNK activation and invasive cell motility. Our Ad-mediated rescue assays performed with FAK^−/−^ v-Src cells showed that FAK phosphorylation at Tyr-397, FAK kinase activity, and the SH3-binding sites in the FAK COOH-terminal domain were individually required to facilitate Rac and JNK activation and cell invasion. Similar results linking FAK to JNK with respect to cell cycle and cell survival signaling have been obtained through the expression of DN FAK-related constructs ([@bib39]; [@bib2]). However, our results with Ad-p130Cas expression in FAK^−/−^ v-Src cells revealed the unexpected importance of p130Cas SH3 domain function in cell invasion that is distinct from the role of p130Cas in promoting Src-stimulated cell growth ([@bib22]). We found that SH3 domain--mutated p130Cas was highly tyrosine phosphorylated in FAK^−/−^ v-Src cells but that it did not function to promote either JNK activation or cell invasion as did WT p130Cas. Although p130Cas--Crk interactions form a scaffolding complex promoting Rac-induced JNK activation ([@bib15]) and have been proposed as a molecular "switch" promoting cell motility ([@bib31]), overexpression of a Crk-binding site mutant of p130Cas does not block v-Src--induced JNK activation ([@bib12]). We found that FAK expression facilitated the formation of a v-Src--p130Cas--Dock180 signaling complex and that the Cas SH3 domain bound to Dock180 in pull-down assays. As a consensus (PXXPXR), Cas SH3-binding site ([@bib51]) lies within the Dock180 COOH-terminal domain and Crk SH3 binding to Dock180 facilitates both Rac and JNK activation ([@bib29]), we speculate that Dock180 may function as both a Crk-dependent and -independent guanine nucleotide exchange factor promoting v-Src--stimulated Rac activation. Notably, FAK expression was required for maximal Rac activation in both reconstituted normal and v-Src--transformed cells. Moreover, we found that activated Rac expression in normal FAK-reconstituted cells yielded a highly invasive phenotype equivalent to v-Src transformation. The findings suggest that FAK, v-Src, and Rac may coordinately regulate common targets involved in promoting cell invasion. Interestingly, activated Rac expression in normal FAK-reconstituted cells yielded a similar profile of MMP-2 activation and increased MMP-9 secretion as did v-Src transformation. The importance of these MMPs was supported by our findings that exogenous TIMP-1 addition blocked Matrigel cell invasion in a dose-dependent manner. Although previous studies have shown that both FAK and Rac can modulate both MMP activity and expression ([@bib52]; [@bib18], [@bib20]; [@bib61]), gene array analyses reveal that these targets are part of a larger set of FAK-regulated and metastases-associated gene products (unpublished data). FAK to JNK signaling promoting an invasive cell phenotype --------------------------------------------------------- Our results using the SP600125 JNK kinase inhibitor and DN JNK expression support the importance of JNK activation downstream of Rac as being necessary for cell invasion. This is consistent with studies using Rac effector domain mutants where cell invasion was disrupted by the Rac Y40H mutation that uncouples Rac-mediated JNK activation ([@bib4]). Although FAK has been shown to recruit JNK to focal contacts ([@bib2]), we hypothesize that FAK--Src signaling through JNK also acts to alter the transcriptional regulation of cell invasion-associated gene targets such as MMP-9 ([@bib52]; [@bib42]; [@bib53]). Since Src activation contributes to the metastatic spread of carcinoma cells ([@bib6]) and FAK is a key regulator of cytotrophoblast invasion of the uterus during placental formation ([@bib26]), we propose that FAK--Src signaling may synchronize MMP-mediated extracellular matrix proteolysis and cell motility, thereby facilitating an invasive phenotype in both normal developmental and neoplastic cell settings. Materials and methods ===================== Cells and constructs -------------------- FAK^−/−^p53^−/−^ and FAK^+/+^p53^−/−^ fibroblasts were generated as described ([@bib24]). DA2 cells express HA-tagged FAK and were generated from FAK^−/−^ cells as described ([@bib54]). Prague C v-Src retrovirus was used as described ([@bib19]). Cells were infected for 24 h in medium containing 5 μg/ml polybrene, selected for growth in 3 μg/ml puromycin, and drug-resistant colonies were pooled and expanded. Pro-null FAK ([@bib17]), GFP-FAK ([@bib25]), and GST-SH3 domains for c-Src, Grb2, and PLCγ ([@bib49]) were generated as described. Antibodies and reagents ----------------------- Anti-pTyr (4G10) mAb, avian-specific mAb to v-Src (EC10), and Rac mAb (23A8) were from UBI. HA epitope tag mAb (16B12) was from Covance Research. p130Cas, p190RhoGAP, and Crk mAbs were from BD Biosciences. Flag tag mAb (M2) and β-actin mAb (AC-74) were from Sigma-Aldrich. Polyclonal antibodies to Dock180 (C19), p130Cas (C20), MMP-2 (C19), MMP-9 (C20), JNK1 (C17), ERK2 (C14), and Akt1 (C20) were from Santa Cruz Biotechnology, Inc. Phospho-specific mAbs to activated ERK2 (E10 to pT202/pY204), activated JNK1 (G9 to pT183/pY185), activated Akt (pS473), phosphorylated STAT-3 (pY705), t-phosphorylated c-Jun (pS63), and polyclonal antibodies to STAT-3 and c-Jun were from Cell Signaling Technology. Phospho-specific antibodies to FAK pTyr-397 motif were from Biosource International. Polyclonal antibodies to FAK were used as described ([@bib54]). Recombinant human TIMP-1 was from Chemicon. The pharmacological inhibitor to JNK (SP600125) was from Calbiochem. Blotting and pull-down assays ----------------------------- Cells were solubilized in RIPA lysis buffer containing 1% Triton X-100, 1% sodium deoxycholate, and 0.1% SDS. Antibodies (2.5 μg) were incubated with lysates for 3 h at 4°C and collected by binding to protein G plus (Oncogene Research Products) or protein A (Repligen) agarose beads. Blotting and sequential membrane reprobing was performed as described ([@bib20]). GST-PAK (residues 67--150) or GST-SH3 domains were purified from bacterial lysates by glutathione-agarose affinity binding, and 15 μg of agarose bead-associated GST fusion protein was incubated with cell lysates for 3 h at 4°C. The beads were washed three times in lysis buffer containing 1% Triton X-100, bound proteins were resolved by SDS-PAGE, and target protein association was determined by blotting. IVK assays ---------- ERK2 IVK activity was measured using myelin basic protein, JNK IVK activity was measured using GST-ATF2, and Akt IVK activity was measured using GST--GSK-3 as described ([@bib20]). After SDS-PAGE and electrophoretic transfer to PVDF membranes (Millipore), labeled proteins were quantified by a PhosphorImager (Molecular Dynamics), and the equal recovery of the immuno-isolated kinase was verified by blotting. Migration and invasion assays ----------------------------- Millicell (12 mm diameter with 8-μm pores; Millipore) haptotaxis assays with 10 μg/ml FN (Sigma-Aldrich) were performed as described ([@bib54]). Chemotaxis assays with using 10% FBS added to the lower chamber were performed as described ([@bib55]). Wound-healing scratch assays and growth factor--reduced Matrigel (BD Biosciences) invasion assays were performed as described ([@bib20]). One-way analysis of variance (ANOVA) was used to determine significance. Gelatinase activity ------------------- Conditioned media samples from 5 × 10^6^ cells were separated in nonreducing gels containing 0.1% (wt/vol) gelatin and processed for zones of clearing activity by zymography as described ([@bib20]). For blotting analyses, proteins in the conditioned media were precipitated by cold acetone and separated by SDS-PAGE. Analysis of soluble MMP-2 + MMP-9 activity was performed using the Chemicon Gelatinase Activity kit (ECM700) as per the manufacturer\'s instructions. Ad production and infection --------------------------- Murine HA-tagged FAK cDNA constructs were subcloned into pADtet7 containing Tet-responsive enhancer sequences within a minimal cytomegalovirus (CMV) promoter. pADtet7 also contains SV40 late poly(A) cassette, Ad E1A, and a single loxP site. Recombinant Ads were produced by pADtet7-FAK cotransfection of 293-Cre cells with Ad DNA (Ad5-ψ5) that contains an E1A/E3-deleted Ad genome. Recombinant Ads were expanded on 293-Cre cells, and stocks were titered by limiting dilution. FAK protein expression was driven by coinfection with Ad-TA expressing the Tet transactivator as described ([@bib57]). Cells were infected at an MOI at 10 or 30 plaque forming units (pfu)/cell for Ad-TA or Ad-TA with Ad-FAK, respectively. Cells were analyzed 48 h postinfection. The CMV promoter and bovine growth hormone poly(A) cassette from pCI-Neo (Promega) were subcloned into pE1sp1B (Microbix) to generate pAd/CMV. Flag-tagged Rac1 mutants (Q61L and T17N) were cloned into pAd/CMV, and recombinant Ads were generated by cotransfection of 293 cells with pJM17 (Microbix) containing the E1A-deleted Ad genome. Recombinant Ads were isolated by plaque formation, and Rac-producing virus were identified by flag tag blotting. MOI at 50 pfu/cell was used. A mutation was introduced into the ATG start site of the rat cDNA encoding the alternate spliced short form of p130Cas ([@bib47]). Translation of this construct begins at Met-30 within the p130Cas SH3 domain (CasΔSH3). Both WT and CasΔSH3 were cloned into pShuttle-CMV, and recombinant Ads were produced using the Ad-Easy System (Stratagene) as described ([@bib19]). Cells were infected at an MOI of 50 pfu/cell. Ads for WT JNK or DN JNK (APF) were produced as described ([@bib23]). An MOI of 80 pfu/cell was used. RT-PCR ------ Total RNA was isolated using Trizol (Invitrogen). Reverse transcription was performed using 2 μg RNA and the SuperScript first strand cDNA synthesis kit (GIBCO-BRL). 1.5 μl of the RT product was used for coamplification of MMP-9 with β-actin primer pairs (R&D Systems) as an internal standard. Semiquantitative MMP-9 PCR was performed for 37 and 39 cycles, and β-actin primers were added after PCR cycle 11. Values were quantified using ImageQuant and normalized to β-actin (Molecular Dynamics). IF staining ----------- 2 × 10^4^ cells were plated on FN-coated (5 μg/ml) glass coverslips for 6 h. Cells were fixed with 4% PFA in PBS for 15 min, permeabilized for 10 min with 0.2% saponin in PBS containing normal goat serum (blocking solution), and incubated for 1 h in blocking solution containing TRITC-phalloidin (Molecular Probes). Anti--v-Src mAb or ILK rabbit polyclonal antibody (Zymed Laboratories) costaining was performed using biotinylated donkey anti--mouse (Jackson ImmunoResearch Laboratories) followed by streptavidin-FITC (Vector) and TRITC-conjugated donkey anti--rabbit as described ([@bib19]). Staining of invadopodia emerging through Matrigel was performed at hour 16 of 24 during the invasion assay as described ([@bib19]). Live cell microscopy -------------------- FAK^−/−^ v-Src cells were transfected with GFP-FAK using Lipofectamine Plus (GIBCO-BRL), FACS^®^ sorted for GFP expression after 24 h, and plated onto FN-coated (10 μg/ml) 25-mm coverslips in DME with 0.5% BSA. After 18 h, 20 mM Hepes, pH 7.4, and 10% FBS were added, cells were placed into an Attofluor chamber (Molecular Probes), covered with light mineral oil (Sigma-Aldrich) to prevent evaporation, and placed into a 20/20 Technologies Bionomic microscope stage chamber maintained at 37°C. GFP-FAK was monitored using a 40× Plan-Apochromat objective on a ZEISS Axiovert 100TV microscope, coupled to a Bio-Rad Laboratories 1024 laser-scanning confocal focused at the ventral surface of cells. Images were taken at 1-min intervals, adjusted using Adobe Photoshop^®^, and exported to QuickTime^®^. Online supplemental material ---------------------------- Time-lapse images are included as an online video available at <http://www.jcb.org/cgi/content/full/jcb.200212114/DC1>. Video 1 shows the distribution of GFP-FAK transiently transfected into FAK^−/−^ v-Src cells. The first image was taken ∼15 min after serum stimulation. The time interval for Video 1 is 15 s at 12 frames/s. Supplemental Material ===================== ###### \[Supplemental Material Index\] We thank Gary Bokoch (The Scripps Research Institute) for GST-PAK (67-150), Jean-Francois Cote (Burnham Institute, La Jolla, CA) for GST-CAS SH3 and GST-Dock180 SH3 constructs, Kristiina Vuori for helpful discussions, Stephan Feller (Oxford University, Oxford, UK) for the GST-Crk SH3 construct, and Amanda Moore for administrative assistance. C. Hauck was supported in part by a fellowship from the Deutsche Forschungsgemeinschaft (HA-2856/1-1). This work was supported by grants from the National Institutes of Health to D. Cheresh (CA50286, CA45726, and CA78045), J. Nelson (HL65754), G. Nemerow (HL54352), and D. Schlaepfer (CA75240 and CA87038). This work was initiated with support from the American Cancer Society (RPG-98-109-TBE) to D. Schlaepfer. This is manuscript number 15085-IMM from The Scripps Research Institute. Abbreviations used in this paper: Ad, adenovirus; CMV, cytomegalovirus; c-Src, cellular Src; 3-D, three dimensional; DN, dominant-negative; ERK; extracellular signal--regulated kinase; FAK, focal adhesion kinase; FRNK, FAK-related nonkinase; FN, fibronectin; HA, hemagglutinin; ILK, integrin-linked kinase; IF, immunofluorescence; IP, immunoprecipitation; IVK, in vitro kinase; JNK, c-Jun NH~2~-terminal kinase, MMP, matrix metalloproteinase; MOI, multiplicity of infection; pfu, plaque-forming units; PI, phosphatidylinositol; pTyr, phosphotyrosine; TA, transactivator; v-Src, viral Src; WT, wild-type. [^1]: Address correspondence to David D. Schlaepfer, The Scripps Research Institute, Dept. of Immunology, IMM26, 10550 N. Torrey Pines Rd., La Jolla, CA 92037. Tel.: (858) 784-8207. Fax: (858) 784-8227. E-mail: <dschlaep@scripps.edu>

1. Introduction {#s0005} =============== Germline pathogenic variants in a variety of genes are associated with an increased lifetime risk of breast and gynecologic cancers. Patients with a pathogenic variant in *BRCA1*, for example, have a lifetime cumulative risk of 72% and 44% for developing breast and ovarian cancer, respectively; for *BRCA*2 carriers, those risks are 69% and 17% ([@b0030]). Although pathogenic variants in *BRCA1* and *BRCA2* account for the majority of hereditary breast and gynecologic cancers, pathogenic variants in a number of other high and moderate penetrance genes, including DNA mismatch repair genes, *TP53*, *PALB2*, *ATM*, *CHEK2*, *BARD1, BRIP1, CDH1, NBN, NF1, PTEN, RAD51C, RAD51D,* and *STK11* have also been implicated. Recent improvements in knowledge and accessibility of genetic testing has enhanced the detection of hereditary breast and ovarian cancer variants, leading to more widespread use of high-risk screening tools and risk-reducing surgeries. Breast magnetic resonance imaging (MRI) has become standard of care in breast cancer screening for high risk women (those with a deleterious mutation, prior therapeutic chest radiation, or 20--25% or greater lifetime risk of breast cancer, per the American Cancer Society), due to the increased sensitivity of identifying early breast cancers compared to mammogram, albeit at the expense of an increased false positive rate, with positive predictive values ranging from 24 to 71% ([@b0070], [@b0035]). The addition of breast MRI screening has been validated in high risk populations ([@b0025], [@b0040]). The American Cancer Society specifically recommends that patients with germline pathogenic variants in *BRCA* 1 or 2 begin breast cancer screening with MRIs at 25 years old and that they add surveillance mammography at 30 years old ([@b0065]). However, mammography, like MRI, is not without potential harms of both increased radiation exposure as well as the benefit-harm trade-off of overdiagnosis versus mortality reduction. Recent data from the radiology literature suggest that *BRCA1* and *BRCA2* carriers under 40 years old may not benefit from mammography in addition to MRI screening ([@b0080], [@b0085]. There are a number of studies that address breast cancer screening specifically in patents that carry a pathogenic variant in BRCA, but few assess broader high-risk populations. The objectives of the present study were to evaluate screening modalities utilized in a High Risk Breast and Ovarian Cancer (HBOC) clinic, to assess abnormalities found on mammogram and breast MRI screening, and to specifically compare *BRCA* and non-*BRCA* carriers with respect to callbacks, biopsies, and cancer diagnoses. A secondary objective was to analyze screen detected cancers in *BRCA* patients younger than 40 years old to determine outcomes of mammography use in this population. 2. Materials and methods {#s0010} ======================== Following Institutional Review Board review at the University of Virginia, all patients who were seen in the HBOC clinic at the University of Virginia from January 1, 2007 through March 1, 2017 were identified using an institutional Clinical Data Repository (CDR). Patients followed in this clinic were deemed to be high risk if they carried a known genetic mutation, met clinical criteria for a potential hereditary cancer syndrome, had a first or second degree relative with ovarian cancer, or met high risk breast criteria (over 20--25% lifetime risk of breast cancer). The CDR contains patient demographics and known genetic mutations. Through electronic medical record (EMR) review, patients who were only seen once in clinic for consultation and were deemed to not be truly high risk for breast cancer were excluded. Those who elected not to pursue their high risk screening at the University of Virginia were also excluded. Patients were considered to be high risk if they were known mutation carriers *or* had over a 20% lifetime risk on Tyrer-Cuzick (T-C) model *or* if they were deemed high enough risk to have a screening breast MRI recommendation as part of their follow-up in the high risk clinic. Details on frequency and results (e.g. callbacks, biopsies, cancer diagnoses) of breast cancer screening were abstracted by EMR review. Patients with a personal history of breast cancer were not included in this screening population. Characteristics of *BRCA* gene mutation carriers with screening-detected cancers were then examined granularly. Univariate analyses were used to compare baseline patient characteristics and breast cancer screening outcomes by *BRCA* mutation carrier status. Data were compared using Chi-square tests for categorical variables and appropriate parametric and non-parametric tests for continuous variables. A p-value \< 0.05 was used for statistical significance. IBM SPSS Statistics (Version 24) was used for all statistical analyses. 3. Results {#s0015} ========== The HBOC clinic saw 1348 patients over the ten-year study period. Six hundred thirty-one patients (46.8%) were deemed to be at high-risk for breast cancer; of the high-risk patients, 496 patients had no known pathogenic variant (79%), 128 (20%) had a pathogenic variant in *BRCA*1 or *BRCA*2, and 7 patients had other pathogenic variants in known breast cancer genes (1 *TP53*, 1 *PALB2*, 3 *ATM*, 2 *CHEK2*). The differences in patient characteristics of *BRCA* carriers (N = 128) versus high-risk non-*BRCA* carriers (N = 503) are listed in [Table 1](#t0005){ref-type="table"}. Those with a known *BRCA* variants were more likely to have had genetic testing at our institution versus those who are not known to have a *BRCA* pathogenic variant (non-*BRCA* carriers) (30% vs 45%; p = 0.002). Compared to non-*BRCA* patients, *BRCA* patients were on average younger (44.0 vs 46.0; p = 0.024), had higher rates of oral contraceptive use (88% vs 80%; p = 0.037), had more total screening mammograms (6.0 vs 4.0; p \< 0.001), and were younger at first screening MRI (44.0 vs 47.0; p = 0.03). Additionally, *BRCA* mutation carriers were more likely to undergo risk-reducing mastectomy (45% vs 14%; p \< 0.001) and risk-reducing bilateral salpingo-oophorectomy (56% vs 8.9%; p \< 0.001) compared to non-*BRCA* mutation carriers. Of note, there were no statistically significant differences in body mass index, Ashkenazi Jewish ancestry, race, insurance status, level of education, known breast cancer risk factors (e.g. age at menarche, age at menopause, parity, etc.), utilization of genetic counseling services, and age at first screening mammogram between *BRCA* and non-*BRCA* high-risk patients (all p \> 0.05).Table 1Patient characteristics.No Known Variant n = 503 (%)*BRCA1* and *BRCA2* n = 128 (%)p-valuePatient Age (years)[\*](#tblfn1){ref-type="table-fn"}46.0 (37.0--55.0)44.0 (33.0--53.0)0.024BMI (kg/m2)[\*](#tblfn1){ref-type="table-fn"}25.9 (22.7--31.9)26.5 (22.8--32.5)0.94Ashkenazi Jewish22 (4.4)10 (7.8)0.11Race0.10White432 (86)111 (87)Black48 (9.5)9 (7.0)Hispanic10 (2.0)5 (3.9)Asian3 (0.6)1 (0.8)Other10 (2.0)2 (1.6)Education Level0.26\<8th grade6 (1.2)3 (2.3)Some high school15 (3.0)2 (1.6)High school diploma46 (9.1)9 (7.0)Some college73 (15)19 (15)College degree134 (27)39 (30)Graduate school/degree183 (36)31 (24)Unknown46 (9.1)25 (20)Insurance Status0.55Medicare43 (8.5)9 (7.0)Medicaid18 (3.6)7 (5.5)Private407 (81)96 (75)Self-pay14 (2.8)5 (3.9)Unknown21 (4.2)11 (8.6)*Reproductive Factors*Age at menarche (years)12.0 (12.0--13.0)12.0 (12.0--14.0)0.45Age at menopause (years)49.0 (45.0--51.0)48.0 (43.5--53.0)0.97Parity2.0 (2.0--3.0)2.0 (2.0--3.0)0.073Age at first birth (years)25.0 (21.0--29.0)23.0 (20.5--29.0)0.86Breastfeeding duration (mo.)4.0 (0.0--14.0)1.0 (0.0--13.0)0.11SERM (preventive) use48 (9.5)9 (7.0)0.38Oral contraceptive use404 (80)113 (88)0.037HRT (past or present)81 (16)16 (13)0.31GeneticsGenetic counseling[\*\*](#tblfn2){ref-type="table-fn"}235 (47)62 (48)0.73Genetic testing[\*\*](#tblfn2){ref-type="table-fn"}152 (30)57 (45)0.002*Breast Cancer Screening/Prevention*Screening mammogram (number)6.0 (3.0--9.0)4.0 (2.0--7.0)\<0.001Age first screening mammo (yr.)42.0 (36.0--50.0)41.0 (32.0--51.0)0.23Screening breast MRI (number)2.5 (2.0--5.0)3.0 (1.0--4.0)0.64Age first screening MRI (yr.)47.0 (40.0--54.0)44.0 (33.0--54.0)0.030Risk reducing mastectomy[\*\*](#tblfn2){ref-type="table-fn"}14 (2.8)45 (35)\<0.001Risk reducing BSO[\*\*](#tblfn2){ref-type="table-fn"}45 (8.9)72 (56)\<0.001[^1][^2][^3][^4] Results of screening mammograms and MRIs, including callback, biopsy, and cancer rates, are shown in [Table 2](#t0010){ref-type="table"}, [Table 3](#t0015){ref-type="table"}, respectively. A total of 531 patients completed at least one screening mammogram and 399 patients completed at least one screening breast MRI at our institution. In sum, there were 3258 screening mammograms and 1218 screening breast MRIs performed over the study period. In the entire cohort over the ten-year course, 3.8% of non-*BRCA* patients were ultimately diagnosed with breast cancer compared to 10.2% of *BRCA* carriers (p = 0.0061). Of the high-risk patients who received mammograms at UVA, 91 *BRCA* and 440 non*-BRCA* carriers, 45% and 54%, respectively, received callbacks (p = 0.13). The rate of biopsies following callbacks after screening mammograms was not significantly different between *BRCA* and non-*BRCA* patients (29% vs 16%; p = 0.30). However, patients with a pathogenic variant in *BRCA* were more likely to have cancers diagnosed after mammogram callbacks (15% vs 3.8%; p = 0.0046) and biopsies (50% vs 13%; p = 0.0026) compared to non-*BRCA* mutation carriers. *BRCA* mutation carriers were also more likely to have cancers diagnosed after biopsies following screening MRI (27% vs 11%; p = 0.045), but rates of MRI callbacks, biopsies, and cancers diagnosed after callbacks was not statistically significantly different between the two groups (all p \> 0.05).Table 2Results of screening mammograms.No Known Variant\ n = 440 (%)*BRCA1* or *2*\ n = 91 (%)p-valueMammogram callbacks237 (54)41 (45)0.13Biopsy after mammogram68 (16)12 (29)0.30Cancer diagnosed after mammogram9 (1.8)6 (6.6)0.017Cancer diagnosed after callback9/237 (3.8)6/41 (15)0.0046Cancer diagnosed after biopsy9/68 (13)6/12 (50)0.0026Table 3Results of screening MRIs.No Known Variant n = 305 (%)*BRCA1* or *2*\ n = 94 (%)p-valueMRI callbacks117 (38)35 (37)0.84Biopsy after MRI90 (30)26 (28)0.73Cancer diagnosed after MRI10 (3.3)7 (7.4)0.080Cancer diagnosed after callback10/117 (8.5)7/35 (20)0.059Cancer diagnosed after biopsy10/90 (11)7/26 (27)0.045 *BRCA* patients were diagnosed with cancer (n = 13) at an average age of 51 (range 29--70). Characteristics of screening-detected cancers in *BRCA1* and *BRCA2* carriers are shown in [Table 4](#t0020){ref-type="table"}. Of the cancers diagnosed after abnormal MRI, three were ductal carcinoma in situ (DCIS); in all three cases, there had been a normal mammogram within the 4--6 months prior to the MRI that found the cancer. In those found after abnormal mammogram (n = 6), follow up MRI was performed in four cases; all demonstrated the lesion identified on mammogram. Only one of these cases had a preceding MRI and it was normal one year prior to the abnormal mammogram. Three patients were diagnosed younger than 40, one on mammogram and two on MRI. The patient diagnosed on mammogram had no prior MRI and the lesion was seen on immediate follow-up diagnostic MRI.Table 4Characteristics of screening-detected cancers in *BRCA1* and *BRCA2* gene mutation carriers.*BRCA*AgeImagingPathologyGradeSize (cm)ER/PR/HER2pTN[\*](#tblfn3){ref-type="table-fn"}*1*38MammoIDC, DCISIII7.5-/-/+T3N2a*1*58MammoIDCIII0.5-/-/-T1aN0*1*51MammoIDC, DCISIII1.6+/-/-T1cN1*2*50MammoIDC, DCISII0.9+/+/-T1bN0*2*46MammoDCISIINA+/NA/-TisN0*2*56MammoIDC, DCISII1.3+/+/-T1cN0*1*48MRIIDCIII1.5+/-/-T1cNx*1*59MRIDCISI0.2+/NA/NATisN0*1*34MRIIDCIINA-/-/-NA*1*58MRIDCISII/IIINA+/NA/NATisN0*1*66MRIIDC, DCISIII1.4-/-/-T1cN0*2*29MRIDCISIII0.8+/NA/NATisN0*2*70MRIIDC, DCISII0.4+/+/-T1aN0[^5][^6] 4. Discussion {#s0020} ============= Women with *BRCA* mutations in our patient population were more likely to have breast cancers diagnosed after both MRI and mammogram compared to patients with family history alone. Despite the differences in rates of cancer diagnoses between *BRCA* and non-*BRCA* mutation groups, the present study did not find a difference in callback or biopsy rates following both MRI and mammography. Therefore, there were more false positive recalls for those without a *BRCA* mutation, limiting the positive predictive value of MRI in this cohort. This has been reported scantly in the literature but certainly may hold clinical relevance, as false positives add to healthcare costs by necessitating further workup and may cause emotional harm by generating breast cancer anxiety in the patient ([@b0050]). A recent study found the false positive recall rates following mammogram or MRI to be 22.2% and 26.3% in *BRCA* mutation carriers and others at increased risk without a mutation, respectively ([@b0085]). A similar trend is observed in the current study. This study aimed to look specifically at cancers diagnosed in *BRCA* mutation carriers under the age of 40 in light of recent literature that calls into question the added utility of screening mammography in this population ([@b0080], [@b0085], [@b0020], [@b0010]). Of the breast cancers diagnosed in *BRCA* patients (n = 13) in our study population, three patients were younger than 40 years old, two *BRCA1* carriers and one *BRCA2* carrier. Of these patients, two of the cancers were diagnosed on MRI and one on mammogram (the oldest patient of the three), and the patient whose cancer was initially seen on mammography had no prior MRI and the lesion was seen on follow-up MRI. The utility of different imaging modalities in this younger age group of *BRCA* patients has been examined by a research group in the Netherlands over the past few years ([@b0080], [@b0085]). This group found, in a population of *BRCA* mutation carriers, that 3 of 61 cancers were detected only on mammogram (with none in those younger than 40) and that the addition of mammogram to MRI resulted mostly in the detection of a small number of DCIS cases that were occult on MRI. A primary argument for utilizing mammography (in addition to MRI) in breast cancer screening for *BRCA* mutation carriers is that it is better than MRI for identifying DCIS ([@b0075], [@b0005]). However, in the current study, the three *BRCA* patients with DCIS all had normal mammograms 4--6 months prior to DCIS being detected on MRI. This is supported by a prospective study that found that 48% of high-grade DCIS cases were missed on mammography but diagnosed by MRI alone; conversely, only two cases were missed by MRI and detected on mammography ([@b0045]). In women under 40 years old, the number of screening mammograms needed to detect an MRI occult cancer was 1829. These results are also supported by a meta-analysis of four breast cancer screening trials of high risk women that found only one invasive cancer detected by mammography alone in *BRCA1* mutation carriers ([@b0015]). Besides potentially not adding much screening benefit in this cohort, mammography has a number of risks that could be reduced by delaying when this imaging modality is started in high-risk patients. Potential harms include unnecessary costs, callback procedures, and the risk of radiation-induced breast cancer ([@b0010], [@b0055]). *BRCA* mutation carriers may be particularly susceptible to the cumulative effect of yearly mammograms, as they have impaired repair of the double-strand DNA breaks that are caused by low-dose X-rays ([@b0060]). Therefore, the potential benefit of discovering an occasional MRI occult cancer in this younger age group must be balanced with the potential harms of repeated mammography. This study has several limitations. It is a single-institution, retrospective study. This did allow for more thorough chart review and consistency, but it also resulted in a relatively small study size of patients, especially those who were diagnosed with cancer, which limits the generalizability of the results. Finally, a proportion of patients did not follow the recommended breast cancer screening schedule and a small number were lost of follow up, both of which may have affected the data. In conclusion, patients with a pathogenic variant in *BRCA 1* or *2* were more likely to be diagnosed with breast cancer following all screening modalities compared to high-risk non-*BRCA* carriers. In addition, MRI was able to effectively identify DCIS in the *BRCA* population. In *BRCA* mutation carriers younger than 40 years old, there were no MRI occult cancers found. These findings begin to address the question of whether MRI alone is a reasonable breast cancer screening strategy for *BRCA* mutation carriers under 40 years old. Larger studies are warranted to further investigate this question. CRediT authorship contribution statement {#s0025} ======================================== **Anne T. Knisely:** Conceptualization, Data curation, Formal analysis, Investigation, Writing - original draft. **Martha E. Stewart:** Data curation, Methodology, Writing - review & editing. **Christine Garcia:** Conceptualization, Writing - review & editing. **Martha H. Thomas:** Data curation, Methodology, Writing - review & editing. **Susan C. Modesitt:** Conceptualization, Writing - review & editing. **Kari L. Ring:** Conceptualization, Formal analysis, Investigation, Project administration, Supervision, Writing - review & editing. Declaration of Competing Interest ================================= The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. [^1]: Categorical variables are reported as N (%) and continuous variables as median (IQR). [^2]: SERM = selective estrogen receptor modulator; HRT = hormone replacement therapy; MRI = magnetic resonance imaging; BSO = bilateral salpingo-oophorectomy. [^3]: As reported at initial High-Risk Clinic visit. [^4]: At the University of Virginia. [^5]: Mammo = mammogram; MRI = magnetic resonance imaging; ER = estrogen receptor; PR = progesterone receptor; HER2 = human epidermal growth factor receptor 2; IDC = invasive ductal carcinoma; DCIS = ductal carcinoma in situ. [^6]: pTN = pathologic staging; M0 for all cases.

Dry eye is a common condition with symptoms that can considerably impact patients\' quality of life.^[@R1],[@R2]^ It is perceived to be as distressing as chest pain^[@R3],[@R4]^ and imposes considerable health care (U.S. \$0.27 million to U.S. \$1.1 million per 1,000 persons annually)^[@R5]^ and productivity costs.^[@R2],[@R6]^ These costs are especially pronounced in Asia, where there is a high prevalence and severity of dry eye.^[@R7],[@R8]^ Meibomian gland dysfunction (MGD) is a major cause of dry eye that affects 46.2% to 69.3% of Asians and 3.5% to 19.9% of whites.^[@R9],[@R10]^ A chronic abnormality of the meibomian glands in the eyelids, MGD, affects tear film stability^[@R11]^ and can cause eye discomfort.^[@R12]^ As a result, it is believed to lead to faster tear evaporation and poorer visual function, exposing the cornea and conjunctiva to more damage.^[@R13]^ Recent opinion, however, suggests that its role may be more accurately described as assisting in tear film spreading^[@R14]^ rather than retarding evaporation. The cornerstone therapy for MGD is the use of warm compresses.^[@R15]^ Various forms of eyelid-warming therapy have been shown to improve patients\' symptoms^[@R16]--[@R20]^ and tear film stability,^[@R16]--[@R21]^ slow down tear evaporation,^[@R20],[@R21]^ and reduce ocular surface damage.^[@R18],[@R22]^ For more severe MGD, other treatments, such as antibiotics and anti-inflammatory cyclosporin A, are prescribed in addition to warm compresses.^[@R15]^ The recommended regimen for warm compresses is daily treatment.^[@R15]^ However, clinical experience suggests that many patients cannot sustain routine warm compresses for an extended duration of time. Poor patient compliance is further complicated by difficulty in delivering therapeutic range of temperatures to the meibomian glands. To improve patient compliance, more convenient treatment modalities are necessary. Several eyelid-warming devices have been developed recently with features that improve convenience and deliver heat at safe, calibrated temperatures.^[@R17],[@R19],[@R23]^ Yet, little evidence exists on the efficacy of these devices, especially in their long-term use. Most studies^[@R18],[@R19],[@R21]--[@R23]^ lacked control groups for comparison, were short term (up to 3 weeks), or had just one application of lid warming. Among them, only two studies^[@R22],[@R23]^ reported effects in healthy participants. Currently, there are no data about the benefit of lid warming in patients with a baseline of significant meibomian gland loss. In clinical practice, evaluation of the pretreatment meibographic status may be needed to exclude patients with significant glandular loss from lid warming because they may be unresponsive to treatment, as it is believed that loss is a result of gland atrophy and these glands do not regenerate. As a practical consideration, outcomes of heat treatment for patients with different baseline meibographic status need to be compared. LipiFlow (TearScience Inc., Morrisville, NC) thermal pulsation, which provides heat and mechanical stimulation to inner eyelids, has been used in clinical settings to treat the blocked meibomian glands. This is a new promising treatment that uses heat to unblock glandular plugging by increasing the temperature of meibomian glands in patients with MGD from the inner surface of the eyelid. A single LipiFlow treatment has been demonstrated to be efficacious in patients with MGD.^[@R19],[@R24]--[@R26]^ In randomized controlled trials (RCTs) conducted in the United States,^[@R19]^ Germany,^[@R24]^ and France,^[@R25]^ patients with MGD who received a 12-min treatment showed significantly improved tear film breakup time and dry eye symptoms. In the U.S. RCT,^[@R19]^ these effects were sustained at even 12 months after treatment in a subgroup of white patients.^[@R27]^ As existing trials have only been conducted for Western populations, there is no published report on the efficacy, safety, and convenience of LipiFlow treatment in an Asian population. As the prevalence of MGD is high in Asia (46.2%--69.3% in Asian studies) compared with the West (3.5%--19.9% in western studies),^[@R9],[@R10]^ the severity of the disease may also be greater in the former ethnic group; thus, it cannot be assumed that equivalent efficacies are achieved in both populations. To address this issue, we conducted an interventional study of thermal pulsation compared to conventional warm compresses in patients with MGD in Singapore. METHODS {#s1} ======= Overview of Study Design {#s1-1} ------------------------ This is a 3-month single-institution trial comparing two eyelid warming methods in patients with MGD. The study has been approved by the SingHealth Centralised Institutional Review Board and adhered to the Tenets of the Declaration of Helsinki. The study was registered at the [clinicaltrials.gov](http://clinicaltrials.gov) database (NCT01683318). Allocation to treatment groups was not randomized (Fig. [1](#F1){ref-type="fig"}). Recruitments for the thermal pulsation (intervention) group and warm compresses (control) group were commenced at different times, but the evaluation periods greatly overlapped, and the participants were recruited from the same clinic by the same study team under similar recruitment criteria. Nonrandomization was due to a delivery delay of the LipiFlow thermal pulsation machine, together with the LipiView ocular surface interferometer and standardized force meibomian gland evaluator (Details in Supplemental Digital Content 1, <http://links.lww.com/ICL/A32>). {#F1} Participants {#s1-2} ------------ From February 2012 to March 2013, all patients with dry eye and MGD at the Singapore National Eye Centre dry eye clinic who met the eligibility criteria were briefed about this study and invited for screening. Meibomian gland dysfunction is diagnosed as the presence of dry eye symptoms with visible meibomian gland plugging on examination (exact recruitment criteria shown below). Eligible participants were then enrolled with written informed consent. Inclusion criteria were the following:At least one of eight dry eye symptoms (Supplemental Digital Content 2, <http://links.lww.com/ICL/A33>---questionnaire modified after Schein et al.^[@R28],[@R29]^) was experienced "often" or "all the time"At least one meibomian gland opening with a visible plugging over the eyelid margin^[@R30]^No ocular pathology requiring treatment other than eye lubricant and conventional eyelid hygiene within the last month and during the studyBaseline irritation symptom of more than 0 millimeter as reported on the modified Symptom Assessment in Dry Eye (SANDE) visual analog scale (Supplemental Digital Content 3, <http://links.lww.com/ICL/A34>---in addition to irritation symptom, the modified version also assessed blurring of vision and light sensitivity^[@R31],[@R32]^). The exclusion criteria were the following:Known diagnosis of thyroid dysfunction and rheumatoid arthritis and ocular surgery within the previous 6 monthsLaser-assisted in situ keratomileusis within the previous yearCentral nervous system and hormonal drugs required within the last month and during the studyActive ocular infection or presence of pterygiumAny need to wear contact lens during the studyAny need to use antibiotics, steroid, and anti-inflammatory drugs such as RestasisHaving another household member who also participated in this study. Baseline Data {#s1-3} ------------- For the purpose of characterizing the study population and evaluating treatment efficacy, participants were evaluated for their baseline characteristics before intervention. Dry eye symptoms, namely, irritation, visual blurring, and photophobia, were each assessed for their frequency and severity using two visual analog scales as previously described in a modified SANDE questionnaire.^[@R29]^ A global score was calculated for each symptom by taking the square root of the product of the two scales. For outcome evaluation, only the irritation global score (the only attribute assessed by the original SANDE questionnaire^[@R31]^) was used. The modified questionnaire of Schein et al.^[@R28],[@R29]^ described under the inclusion criteria was only used for screening of eligibility, not for the assessment of study endpoints. Objective clinical signs, namely, tear breakup time (TBUT), fluorescein corneal staining, and Schirmer I test, were also documented at baseline in both control and thermal pulsation groups. Tear breakup time was performed as in previous studies.^[@R29],[@R33]^ A wetted fluorescein (1 mg) strip (Fluoret Fluorescein Sodium Ophthalmic Strip; Laboratoire Chauvin, Aubenas, France) was introduced to the inferior palpebral conjunctiva to stain the ocular surface.^[@R33]^ Three readings were taken for each eye, and the average TBUT was calculated. Corneal fluorescein staining was graded using the previously published Cornea and Contact Lens Research Unit (CCLRU) scheme.^[@R34]^ Each of the five corneal zones was scored between 0 (no staining/scarring) to 4 (severe staining). The components of corneal staining, namely, type, depth, and extent, were assessed individually, and the grade of staining for each eye was determined using the component with the most severe grade. The presence of clinically relevant staining in each corneal zone was taken as a CCLRU staining grade of 1.0 or greater. The Schirmer I test without anesthesia was performed with 5-mm wide Schirmer Tear Test strips (4701001; Clement Clarke International, Harlow, Essex, United Kingdom) as described earlier.^[@R35]^ The length of wetting after 5 min was measured from the notch of the strip.^[@R30]^ The meibomian gland morphology was also evaluated using noncontact infrared meibography equipment modified from Topcon slit lamp SL-D7 (Topcon Corporation, Tokyo, Japan)^[@R32]^ and was graded as healthy, intermediate, or unhealthy based on the amount of glandular loss and glandular characteristics (Supplemental Digital Content 4, <http://links.lww.com/ICL/A35>---grading of meibomian gland).^[@R32]^ Two trained optometrists graded each image independently, and in the case of a disagreement, a third investigator was involved in determining the grade. As part of the exploratory outcomes in this study, the baseline tear lipid layer thickness (LLT) and the number of glands expressing liquid secretion were measured with the LipiView ocular surface interferometer (TearScience Inc., Morrisville, NC)^[@R36]^ and the standardized force meibomian gland evaluator (TearScience Inc.). Using a standardized force meibomian gland evaluator (TearScience Inc.), the number of glands yielding liquid secretion was counted in the lower lid under slit-lamp microscopy. The number of plugged glands was also counted in both upper and lower lids under slit-lamp microscopy. These meibomian gland secretion variables were only assessed in the thermal pulsation group. Interventions {#s1-4} ------------- In the thermal pulsation group, on the first visit, participants underwent a single 12-min treatment session of LipiFlow thermal pulsation after instillation of topical local anesthetic as instructed by the manufacturer. In the control group, participants received a towel on the first visit and were instructed to use it for twice daily warm compresses lasting 10 min per session. They were instructed to warm the towel in warm water before placing it over closed eyes and to rewarm it once it cools. Patients in both intervention groups were asked to do lid hygiene, which consisted of daily manual lid massage and lid cleaning with Blephagel (Spectrum Théa Pharmaceuticals Limited, Fernbank House, Cheshire, United Kingdom). Warm compresses (in the warm compresses group) and lid hygiene (in both groups) were performed for the entire study duration (12 weeks). To monitor treatment compliance, participants from both groups were given a diary to record details of eye lubricant use, lid hygiene with Blephagel, and warm compresses (the latter two for control group only). Outcomes {#s1-5} -------- Participants were assessed before treatment (baseline) and also 4 weeks and 12 weeks after the start of treatment. At week 4, SANDE symptom score and TBUT were evaluated, whereas at week 12, SANDE symptom score, TBUT, and Schirmer I test were evaluated. The primary outcome measure was the SANDE global score for irritation symptom one month after treatment. As a secondary outcome, irritation score changes after treatment over the period of 12 weeks were evaluated. The other secondary outcome measures were (1) the change in the TBUT and (2) Schirmer I test results after treatment over the period of 12 weeks. The exploratory outcomes, namely, (1) LLT, (2) number of meibomian glands yielding liquid secretion, and (3) number of plugged meibomian glands, were evaluated at week 4 and week 12 in the thermal pulsation group only. Safety was assessed through a prespecified safety outcome measure (best-corrected spectacle visual acuity \[VA\] changes after 3 months of treatment). The best-corrected spectacle VA was tested at 4 meters using a distance VA number test (cat. no. C 102; Lighthouse International, New York, NY). The test tool was presented in Snellen format and then converted to the corresponding logMAR equivalent for statistical analysis. A smaller logMAR value represents better vision. At all visits, other safety outcomes were documented if present, including signs of ocular and periocular irritations or other complaints. Sample Size Calculation {#s1-6} ----------------------- We endeavored to detect a 20% difference in the primary outcome of improved global symptom score between thermal pulsation and control groups. Based on an online calculator, 21 participants in each study arm were required for 80% power and a two-sided significance level of 5%. We aimed to recruit 25 participants per arm to allow for 4 cases of lost to follow-up or withdrawals per arm. Statistical Analyses {#s1-7} -------------------- Data were checked for normality with the skewness and kurtosis test to determine the appropriate parametric or nonparametric tests. To test for differences among groups for baseline characteristics and the various outcomes, we used the relevant chi-square test, independent *t* test, two-sample *t* test, and Wilcoxon rank-sum test. Linear mixed model analyses were conducted to analyze the repeated measures taken at baseline, 4 weeks, and 12 weeks after treatment. A fixed effect model was used to compare the thermal pulsation and control groups using compound, autoregressive (AR(1)), and unstructured variance--covariance structures to select the best model fit using information criterion (based on the smallest Akaike information criterion \[AIC\] and Bayesian information criterion \[BIC\] values). The AIC and BIC are measures of the relative quality of a model compared to other models for a particular data set, and the AIC/BIC values are a trade-off between the goodness-of-fit and complexity (number of terms) of the model.^[@R37]^ Interaction terms, along with age and sex adjustments, were also modeled. Bonferroni correction was applied for multiple comparisons. In subgroup analysis focusing on the thermal pulsation group, we evaluated the TBUT, Schirmer test results, LLT, and number of glands with liquid secretion at the two time points (weeks 4 and 12) stratified by baseline meibographic status. Statistically significant difference was defined as an alpha of 0.05. All analyses were performed with SPSS, version 21 (SPSS Inc., Chicago, IL). RESULTS {#s2} ======= Baseline Characteristics of Participants {#s2-1} ---------------------------------------- Twenty-five patients were recruited into the thermal pulsation group, and 25 patients were recruited into the warm compresses group. Four participants were lost to follow-up as shown in Figure [1](#F1){ref-type="fig"}. The study population was 56.2 (mean) ± 11.7 (SD) years, 24.5% male (12 participants), and 89.1% Chinese (41 participants) (Table [1](#T1){ref-type="table"}). This was similar to the previously reported profile of patients from the dry eye clinic at the Singapore National Eye Center.^[@R35]^ We present only the data for the right eye (baseline TBUT, Schirmer I test, LLT, and number of glands with liquid secretion were not significantly different from the left and right eyes \[*P*\>0.05\]). ###### Baseline Characteristics of Clinical Trial  As both groups were recruited at different time points, the baseline characteristics of their participants were checked for any significant differences (Table [1](#T1){ref-type="table"}). In the superior and inferior corneal zones, the presence of fluorescein staining was significantly more common in the control group than in the thermal pulsation group. All other baseline characteristics were not significantly different between both groups. Symptom Assessment {#s2-2} ------------------ The SANDE irritation global score was significantly reduced at 4 weeks from baseline for both groups (Fig. [2](#F2){ref-type="fig"}A). However, there was no significant difference between both groups at 4 weeks (*P*=0.30) (Table [2](#T2){ref-type="table"}). {#F2} ###### Relative Change in Symptom Score, TBUT, and Schirmer  Based on a linear mixed model with an AR(1) structure (AIC: 1137; BIC: 1142) adjusted for age and sex, irritation score significantly improved over the 12 weeks (*P*\<0.01), whereas the difference in score improvement was not significantly different between treatment groups (*P*=0.22). There was no significant interaction between treatment group and time (*P*=0.79). Assessment of Clinical Signs of Dry Eye {#s2-3} --------------------------------------- At 4 weeks, TBUT significantly improved from baseline (*P*=0.015) on the whole. When the treatment groups were examined (Fig. [2](#F2){ref-type="fig"}B---showing absolute values of TBUT), TBUT significantly increased from baseline in subjects who received thermal pulsation (*P*=0.048), but it was not significantly changed in the warm compresses group (*P*=0.14). However, the difference between both groups at 4 weeks was not statistically significant (*P*=0.66) (Table [2](#T2){ref-type="table"}---showing relative change in TBUT over time). At week 12, TBUT changes in both groups were not statistically significant from baseline. For the change in TBUT, a mixed model with an unstructured structure (AIC: 627; BIC: 645) was used and adjusted for age and sex. There was a significant trend of TBUT improvement over the 12 weeks (*P*=0.035) with no significant difference between treatment groups (*P*=0.88). At week 4, TBUT modestly improved by 1.10 sec (95% CI: 0.11--2.10), whereas at week 12, the improvement in TBUT was not significantly different from baseline (mean difference: 1.03; 95% CI: −0.29 to 2.36). Age was also found to be significantly associated with TBUT (0.028 sec decrease in TBUT for every year older, *P*=0.043) in the mixed model. There was no significant interaction between treatment group and time or between treatment group and age (*P*\>0.05). For the Schirmer test (Fig. [2](#F2){ref-type="fig"}C), the control group had a significant change of −3.95 mm (95% CI: −7.16 to −0.75; *P*=0.018), whereas the thermal pulsation group had a nonsignificant improvement of 1.00 mm (mean difference \[mm\]: 1.00; 95% CI: −3.03 to 5.03; *P*=0.612) at week 12. Meibomian Gland Secretion Variables (Exploratory Outcomes) {#s2-4} ---------------------------------------------------------- The meibomian gland secretion variables were only assessed in the thermal pulsation group. For the change in the number of plugged meibomian glands (Fig. [3](#F3){ref-type="fig"}A), a mixed model with a compound structure (AIC: 302; BIC: 307) was used and adjusted for age and sex. Over the 12 weeks, there was a significant and decreasing trend in the number of plugged meibomian glands (*P*\<0.01) in the thermal pulsation group. {#F3} For the number of glands with liquid secretion in the patients treated with thermal pulsation (Fig. [3](#F3){ref-type="fig"}B), a mixed model with an unstructured structure (AIC: 341; BIC: 354) showed that there was a significant increase over the 12 weeks (*P*\<0.01) in the number of glands with liquid secretions, and age was also significantly associated with a smaller number of glands (*P*=0.005), but there was no significant interaction between age and time (*P*\>0.05). Lipid layer thickness (Fig. [3](#F3){ref-type="fig"}C) did not significantly change over time in the thermal pulsation group over the 12 weeks (*P*=0.088) in a mixed model with a compound structure (AIC: 589; BIC: 593). Outcomes Stratified by Baseline Meibography {#s2-5} ------------------------------------------- The irritation improvement was not different between the baseline meibography classifications (*P*=0.98) at 4 weeks and 12 weeks (*P*=0.98). At 4 weeks, TBUT (*P*=0.60), number of glands yielding liquid secretion (*P*=0.53), and LLT (*P*=0.90) outcomes were not different between patients with varying meibographic severities. At 12 weeks, TBUT (*P*=0.78) and Schirmer test (*P*=0.90), and number of glands yielding liquid secretion (*P*=0.30) and LLT (*P*=0.44) were not influenced. None of the patients who underwent thermal pulsation had total loss of meibomian glands before treatment. In the thermal pulsation group, participants were stratified based on their baseline meibomian gland function (defined as number of meibomian glands with liquid secretion ≥5 or \<5) and analyzed for differences in the relative improvements in symptom score, TBUT, and Schirmer I score at 4 weeks. These treatment outcomes were not significantly different between both groups (*P*=0.60, 0.19, and 0.30, respectively, for between-group comparisons for symptom score, TBUT, and Schirmer I score). Safety Outcomes {#s2-6} --------------- Participants who received LipiFlow thermal pulsation all had transient redness in the eyes, and also mild puffiness of the eyelids, for a few minutes after treatment. Over the 12 weeks, there was no significant change in the best-corrected VA in either group (*P*\>0.05). There was also no report of unexpected adverse event either related or unrelated to the study treatment. DISCUSSION {#s3} ========== Our study found that a single session of LipiFlow thermal pulsation improved dry eye outcomes (irritation symptom and modest TBUT improvements) and was similar in efficacy as twice daily warm compresses over the 12 weeks. The improvements in symptoms and TBUT were not affected by the baseline meibographic status or meibomian gland function (assessed by the number of meibomian glands with liquid secretion). Both forms of treatments were found to be safe and well tolerated. LipiFlow thermal pulsation is a relatively novel treatment option for patients with MGD. As such, there are limited data on its efficacy at the moment. When our study was conducted, there were only one 3-month noncontrolled trial^[@R26]^ and 1 multicenter controlled trial with crossover at 1 month^[@R19]^ conducted in the United States. In the latter trial, after crossing over, the participants were reassessed at 3 months^[@R19]^ and recruited for two longer duration follow-up studies.^[@R27],[@R38]^ During the course of our study, two 3-month RCTs, conducted by Finis et al. (2014)^[@R24]^ in Germany and Baumann et al. (2014)^[@R25]^ in France, respectively, have also been published. The study by Finis et al. compared the one-month and 3-month outcome of LipiFlow against conventional lid warming and massage, whereas the study by Baumann et al. compared the 3-month outcome of LipiFlow against a commercial lid warming eye mask (MeiboPatch). These studies found that participants of both groups experienced symptomatic improvement at 1 month, and the effect was sustained up to a period of 3 months.^[@R24],[@R25]^ In our study, we found a similar trend of symptomatic improvement at 1 month with sustained effect at 3 months in both groups. In addition, the study by Finis et al. showed that symptomatic improvement, measured with the Ocular Surface Disease Index, was significantly greater in the LipiFlow group than the conventional lid warming group at one and 3 months. In our study, symptomatic improvement tended to be greater in the thermal pulsation group, although this difference was not statistically significant. In terms of the thermal pulsation group alone, all aforementioned trials and this study showed a significant posttreatment improvement in symptoms at 1 month, which was sustained at 3 months.^[@R19],[@R24],[@R25],[@R38]^ In terms of objective clinical signs, after thermal pulsation treatment, although TBUT was significantly improved, similar to the findings of most previous trials,^[@R19],[@R25],[@R38]^ it should be noted that the increase in TBUT in our study was modest and may not be clinically significant. The U.S. multicenter RCT also demonstrated an increase in TBUT of 1.9 sec at 1 month.^[@R19]^ In the study by Finis et al.,^[@R24]^ the improvement in TBUT at 1 month was significant but not sustained to three months. With respect to our exploratory outcomes, the number of meibomian glands yielding liquid secretion improved at 1 month in our study and in all previous studies.^[@R19],[@R24],[@R25],[@R38]^ However, the median number of meibomian glands yielding liquid secretion only increased by 2 to 3 across the follow-up period and may not be clinically significant. We did not assess this in the warm compresses group. The magnitude of increase was much higher in the U.S. noncontrolled trial (by a mean of 5.4 at 1 month).^[@R26]^ The improvement in symptoms and meibomian gland function may be attributed to the novel design of the LipiFlow device. The advantages of thermal pulsation may be its use of the right temperature and site of heating, as heat is applied from both inner and outer surfaces of the eyelids, and the mechanical massage of glands helps relieve obstruction. Further mechanistic findings after lid warming/thermal pulsation, that is, tear evaporation rates and lipid changes related to this trial, will be published separately. The main strength of our study was the high follow-up rates on reassessment of participants. Furthermore, we included parameters such as the number of plugged glands in the upper and lower eyelids, which have not been reported previously. The use of a recording diary also reinforced adherence to intervention and aided monitoring of treatment compliance. Standardized force expression of meibomian gland is not routinely assessed in most ophthalmology clinics; thus, symptom assessment, which is more commonly used by eye practitioners, was used as the main outcome of this study. The limitations of this study were nonrandomization of interventions, nonblinding of assessors and participants, and lack of meibomian gland secretion evaluation in the control group (Supplemental Digital Content 1, <http://links.lww.com/ICL/A32>---describing late delivery of LipiFlow and evaluation equipment). Given the lack of randomization, there is a potential bias in the patient selection in this study. As participants could not be masked to the type of treatment, it is possible that their perception of symptom severity or frequency was affected by this knowledge. Nevertheless, these issues are of greater concern had the study results show a significant difference between groups, which is not the case. A clinical implication of this study is that evaluation of MGD should consider the severity of inflammation and fibrosis, and not just glandular obstruction. Appropriate treatment for inflammation and fibrosis should be provided in combination with lid warming.^[@R39],[@R40]^ Another implication is that regardless of the amount of meibomian glands (provided that there is no total loss), treatment with lid warming will still be effective. In MGD treatment, sustained long-term improvement in symptoms and tear function is desired. It was suggested that a retreatment strategy with thermal pulsation, scheduled at 9 to 12 months after the first session, might sustain both symptomatic and objective tear function improvements.^[@R27],[@R38]^ Studies using more than 1 thermal pulsation treatment may therefore be useful in the future. Equipment and consumables for LipiFlow are overwhelmingly more expensive than lid warming. Nevertheless, as the effects of a single treatment may be sustained for months, thermal pulsation is more convenient to patients and may encourage better compliance as compared with daily lid warming. Therefore, future studies on long-term efficacy of LipiFlow and cost effectiveness of thermal pulsation treatment are necessary. CONCLUSIONS {#s4} =========== This trial demonstrated that in Asian patients, LipiFlow thermal pulsation had clinical efficacy in improving patient symptoms and produced modest increase in TBUT and meibomian gland function. In general, one session of thermal pulsation was comparable to three months of twice daily lid warming. In addition, the benefit from both forms of therapy may extend to cases with glandular loss, and if thermal pulsation is performed, there may be improvement in glandular function. However, affordability of thermal pulsation treatment will influence its degree of adoption in routine clinical practice. Supplementary Material ====================== ###### SUPPLEMENTARY MATERIAL The authors thank Htoon Hla Myint for his help in statistical analyses. In addition to the authors, CORIM also consists of Markus R Wenk and Manfred Rada from National University of Singapore; Lee Hwee Kuan from Singapore Bioinformatics Institute (Agency of Science and Technology Singapore); Guanghou Shui and Sin Man Lam from Institute of Genetics and Developmental Biology, Chinese Academy of Sciences, China; Sim Hui Shan from Duke-NUS Medical School, Singapore; and Cynthia SK Boo, Samantha SY Lee, and Cheah Loon Too from Singapore Eye Research Institute. The authors have no conflicts of interest to disclose. Supported by National Medical Research Council, Singapore (NMRC/CSA/045/2012), Biomedical Research Council, Singapore (BMRC(TCRP)10/1/35/19/670 R828), and MOE Innovation Fund TIF 11105. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal\'s Web site ([www.eyeandcontactlensjournal.com](http://www.eyeandcontactlensjournal.com)).

Introduction ============ In the last decade, carbon nanotubes--nanofibers (CNT--CNF) have aroused much interest in the bioengineering community because of their unique mechanical, electrical, and surface properties. For example, CNF have exceptional mechanical properties, including a high strength to weight ratio and a high Young's Modulus ([@b12]). This, coupled with tailorable surface energy and biologically inspired nanoscale dimensions, makes CNF intriguing materials for various implant applications ([@b12]; [@b8]; [@b5]). Indeed, research groups have shown that CNF outperform traditional implant materials for bone and neural applications ([@b12]; [@b8]; [@b5]). In particular, adhesion and deposition of calcium-containing mineral by bone cells were significantly greater when cultured on CNF compared with the currently used orthopedic implant materials of commercially pure Ti, Ti6Al4V, and CoCrMo. Directed axonal extension from neurons along CNF has also been observed ([@b9]). Moreover, previous work has shown that aligned patterns of CNF in polymers facilitate the adhesion of osteoblasts (bone-forming cells) on the CNF region ([@b5]). In fact, when cultured for up to 21 days, osteoblasts produced linear patterns of calcium-containing mineral directly on top of linear patterns of CNF placed on polymers ([@b5]). Since the CNF regions are the electrically active regions on such substrates, applied voltages have also been used to further the functions of these cells ([@b7]; [@b13]). For example, osteoblast deposition of calcium-containing mineral was up to 3 times greater when simultaneously cultured on CNT-based materials and subjected to an alternating current of 10 μA at a frequency of 10 Hz with a duty cycle of 50% for 6 hours daily up to 21 days (Supronowicz et al 2000). However, to design the next generation of implants, selective and spatial cell adhesion must be achieved ([@b15]). Specifically, for orthopedic applications, while the functions (such as adhesion, proliferation, and deposition of calcium-containing mineral) of osteoblasts are beneficial, the functions of other cell types (like fibroblasts which contribute to fibrous soft tissue formation) are not ([@b15]). In addition, for vascular applications, while the adhesion of vascular smooth muscle cells and endothelial cells to an implant surface are important, their spatial location must be different since vascular smooth muscle cells and endothelial cells are present in the inner and outer regions of vascular tissue, respectively ([@b3]; [@b2]; [@b1]; [@b6]). It was for these reasons that the objective of the present in vitro study was to examine the adhesion of a wide range of cells to linear CNF patterns formulated on one particular polymer, polycarbonate urethane (PCU), in order to ascertain their ability to serve as tissue engineering constructs. Polycarbonate urethane has been widely used in numerous tissue engineering applications as it is FDA approved for implantation ([@b14]; [@b10]). Materials and methods ===================== Substrates ---------- CNFs were produced by catalytic and chemical vapor deposition techniques and were obtained from Applied Sciences, Inc. (Cedarville, OH, USA). In this work, high surface energy, small diameter (60 nm) CNF were used. PCU was used as a model polymer in this study and was obtained from Thermedics, Inc. (Cleveland, OH, USA). PCU was chosen since it is biocompatible, has a good oxidative stability, and is FDA approved ([@b11]). Copper grids with linear grooves of 30μm were obtained from Electron Microscopy Sciences (Hatfield, PA, USA). To obtain the alignment of CNF on PCU, first PCU was melted by sonicating it in pellet form with chloroform for about 60 minutes. Then, PCU was applied as the base layer on a coverslip (Fisher Scientific, Pittsburgh, PA) and allowed to dry for 2 hours in a vacuum oven at 60°C. Copper grids were then placed on top of the dried PCU by using the principle of surface tension between PCU and CNF. A suspension of CNF was then made by sonicating CNF with 70% ethanol for about 60 minutes; 25 μL of this suspension was placed in the copper grids. The entire sample was allowed to dry again for 1 hour. After the prescribed timed period, the copper grids were removed and a highly aligned pattern of CNF was expected on the PCU layer ([@b5]). The substrates were then treated with 100% ethanol for 1 day under UV light for sterility purposes. Substrate characterization -------------------------- Scanning electron microscopy (SEM, JEOL JSM-840) was used to characterize the alignment of CNF on PCU according to standard techniques ([@b5]). Cell adhesion assays -------------------- To ascertain how a wide range of cells would adhere on the substrates of interest, fibroblasts, vascular smooth muscle cells, and endothelial cells were used. All of these cells would be found at a typical implant insertion site, specifically for the case of orthopedic and vascular implants. Fibroblasts (embryo 3T3; ATCC CRL 1658) and rat aortic smooth muscle cells (RASMC; Vec Technologies, Rensselaer, NY, USA) were cultured in Dulbecco's Modified Eagle Medium (Gibco) supplemented with 10% fetal bovine serum (Hyclone) and 1% penicillin--streptomycin (Hyclone) under standard cell culture conditions (a humidified, 95/5% air/CO~2~, 37°C environment). Rat aortic endothelial cells (RAEC; Vec Technologies) were cultured in MCDB media (Vec Technologies). 2500 cells/cm^2^ were separately seeded on the CNF--PCU samples placed in a 6-well plate and were cultured for 4 hours. At that time, the respective media was aspirated from the wells and the substrates were rinsed with phosphate buffered saline twice to remove non-adherent cells. Adherent cells were then fixed with formaldehyde (Fisher Scientific) for 10 minutes. The cell nuclei were stained with Hoechst stain (Sigma) and visualized under a fluorescence microscope. Substrates were also visualized under light microscopy to determine PCU compared with CNF regions. Fluorescent and light images (Leica) of each substrate were then merged using Image Merger (BMT Micro, Inc., Wilmington NC, USA) software. Digital images (up to 400 μm^2^ field of view) were taken using Image Pro Software. Experiments were repeated in triplicate, three times each. To ascertain the total number of cells that adhered onto the substrates, cells were also counted on these images and reported as cells/cm^2^. Results and discussion ====================== Substrate characterization -------------------------- As expected, results of this study confirmed the alignment of CNF on PCU through the techniques described here ([Figure 1](#fig1){ref-type="fig"}). {#fig1} Fibroblast adhesion ------------------- Results of fibroblast adhesion revealed that about 85% selectively adhered to the PCU region and about 15% selectively adhered to the CNF region ([Figures 2](#fig2){ref-type="fig"} and [3](#fig3){ref-type="fig"}). This is in contrast to previous studies which found close to 90% of osteoblasts selectively adhering to the CNF (not PCU) regions ([Figure 4](#fig4){ref-type="fig"}) ([@b5]). Such results highlight the ability of cells to recognize and adhere differently to PCU compared with CNF regions. Practically, such results may also give researchers the ability to spatially control osteoblast and fibroblast functions to maximize calcium-containing mineral and collagen extracellular matrix formation. This is because long bones of the body naturally possess a unique anisotropic blend of calcium-containing mineral on linear patterns of collagen ([@b4]). Future research will be needed to determine the CNF compared with PCU width spacing that can maximize bone production for such orthopedic applications. {#fig2} {#fig3} {#fig4} Smooth muscle cell and endothelial cell adhesion ------------------------------------------------ For vascular smooth muscle cells, about 80% selectively adhered to the PCU region and about 20% selectively adhered to the CNF region ([Figures 2](#fig2){ref-type="fig"} and [5](#fig5){ref-type="fig"}). In contrast to fibroblast and vascular smooth muscle cell adhesion, no statistical preference was observed for endothelial cell adhesion ([Figures 2](#fig2){ref-type="fig"} and [6](#fig6){ref-type="fig"}). Specifically, about 45% of the endothelial cells adhered to the PCU regions while about 55% adhered to the CNF regions. These findings are of interest to both orthopedic and vascular tissue engineering applications. For orthopedic applications, angiogenesis (mediated by endothelial cells) is needed for maintaining bone health ([@b4]). Thus, these materials could provide for new bone growth on CNF regions and angiogenesis on PCU regions (since CNF regions were almost exclusively favored by osteoblasts, leaving some favorability for PCU regions by endothelial cells). In addition, for vascular applications, these patterned materials may provide for select regions favorable for vascular smooth muscle adhesion--growth (almost exclusively on PCU regions) with different regions favorable for endothelial cell adhesion--growth (CNF regions favored somewhat by endothelial cells but not at all by smooth muscle cells). Future studies are needed to determine competitive cell adhesion on the CNF--PCU substrates to test the hypotheses stated above. {#fig5} {#fig6} Total cell numbers ------------------ When considering the total number of cells that adhered to the patterned surfaces, a significant portion of the original cell seeding density attached ([Figure 7](#fig7){ref-type="fig"}). Specifically, of the original cell seeding density (2500 cells/cm^2^), approximately 60%, 84%, and 90% of the fibroblasts, vascular smooth muscle cells, and endothelial cells adhered to the entire patterned surfaces, respectively. Previous studies ([@b5]) demonstrated that 85% of the original seeding density of osteoblasts adhered on these same substrates. Importantly, when considering total cell numbers, vascular cells and osteoblasts ([@b5]) adhered in significantly greater numbers than fibroblasts ([Figure 7](#fig7){ref-type="fig"}). Although the total number of adherent cells is critical in tissue engineering applications, it is important to note that a small seeding density was utilized in this study for experimental purposes. In addition, it is the selective attachment of these cells to different regions of the PCU--CNF patterned substrates that may be of unique advantage in tissue engineering applications. ![Significant cell adhesion on PCU--CNF patterned materials. Cell numbers represent the total on PCU--CNF patterned substrates. Of the original cell seeding density, approximately 60%, 84%, and 90% of the fibroblasts, RASMC (rat aortic smooth muscle cells), and RAEC (rat aortic endothelial cells) adhered, respectively. Previous studies ([@b5]) demonstrated that 85% of the original seeding density of osteoblasts adhered on these same patterned substrates. Data = mean +/− SEM; n = 3; \* p \<0.01 (compared with fibroblasts).](nano0103-361-07){#fig7} Possible reasons for selective cell adhesion -------------------------------------------- Based on the above data, it is intriguing to ponder why these cells adhered differently to CNF versus PCU regions. Previous studies have provided some insights into this as they have highlighted altered initial protein interactions from serum on PCU and CNF regions ([@b5]). Specifically, previous studies have shown greater vitronectin and fibronectin adsorption to CNF compared with PCU regions ([@b5]). On some surfaces, greater vitronectin and fibronectin adsorption has been correlated with increased osteoblast but decreased fibroblast adhesion ([@b16]); the same events of altered initial protein adsorption from serum that controls subsequent cell adhesion may be happening here. It is believed that proteins are interacting differently with CNF compared with PCU regions since these materials possess altered surface energetics (with the particular CNF used in this study having higher surface area and roughness compared with PCU). Conclusions =========== Collectively, these studies provide for patterned substrates that are not chemically functionalized (with entities like amino acids or other bioactive groups that may become compromised once implanted ([@b7])) to spatially direct cell adhesion. Specifically, osteoblast adhesion was directed to CNF patterns on PCU while fibroblast and vascular smooth muscle adhesion was directed towards PCU regions. Endothelial cell adhesion showed no preference. Coupled with the novel attributes of high mechanical strength and conductivity, CNF patterns on PCU may be novel substrates for numerous tissue engineering applications. The authors would like to thank the SURF (Summer Undergraduate Research Fellowship) program by Purdue University for financial support and Ms Deborah Sherman for help with scanning electron microscopy.

Introduction {#Sec1} ============ How are our decisions affected in situations where we experience acute pain? The majority (70%) of Europeans report having one or more physical pain experiences per month (Vowles et al. [@CR38]) and almost 20% report having chronic pain (Breivik et al. [@CR10]). Over the past few years, the pain reliever Acetaminophen has remained the most or second most sold medication in the United States (Aitken et al. [@CR1]). These data suggest that pain is a highly prevalent condition that potentially influences many everyday behaviors and decisions. In the present research, we investigated the effect of pain on risky and intertemporal choices involving money. Our findings have implications for understanding decision making in real life environments, where decisions are frequently made under non-optimal conditions. Dual-process theories hold that decision making is based on an interaction between intuition ("system 1") and reflection ("system 2"; Evans [@CR17]; Evans and Stanovich [@CR18]; Kahneman and Frederick [@CR23]). Intuitive processes are typically characterized as fast, automatic, effortless, and emotional. Reflective processes are characterized as slow, controlled, effortful, and deliberative. In this study, we used pain as a means to temporarily inhibit the reflective system 2, thus making decisions more intuitive and system 1 based. Pain is a suitable manipulation of system 1/system 2 processing, because painful stimuli are highly salient and attention-demanding (Eccleston and Crombez [@CR13]; Legrain et al. [@CR26]). When in pain, reducing or eliminating the pain becomes first priority and other tasks receive less attention. Pain could thus be viewed as a form of cognitive load. Loewenstein ([@CR28]) refers to pain as a "hot" feeling state that causes people to "take extreme actions" (p. 429). The more intense the state, the greater the gap between what one feels compelled to do (system 1) and what one should do based on the consequences of one's actions (system 2). The perception of pain also activates the insula, a brain region that is known to respond to monetary ("painful") losses (Kuhnen and Knutson [@CR25]; Paulus et al. [@CR30]; Samanez-Larkin et al. [@CR34]; Wu et al. [@CR41]). Research on chronic pain and decision making has mostly used the Iowa Gambling Task (Bechara et al. [@CR3]), an emotional learning task in which players pick cards from decks of varying gain/loss ratios. Advantageous decks contain cards with small gains and small losses. Disadvantageous decks contain larger gains but also larger losses. Healthy subjects normally begin by sampling cards from each deck and then learn to stick to the advantageous, low-risk decks and to avoid the disadvantageous, high-risk ones. Patients with chronic pain keep choosing cards from both kinds of decks and therefore end up with less money than controls (Apkarian et al. [@CR2]; Biagianti et al. [@CR7]; Tamburin et al. [@CR35]; Verdejo-García et al. [@CR37]; Walteros et al. [@CR40]). This learning impairment seems to be due to a lack of somatic markers, because chronic pain patients' physiological arousal does not increase when they pick cards from disadvantageous decks, whereas it does in healthy controls (Elvemo et al. [@CR15]). Patients with chronic pain are also more risk taking on tasks that do not involve a learning component, especially when high potential gains (as opposed to losses) are at stake (Berger et al. [@CR5]). The prediction that follows from research on chronic pain is that acute pain, like chronic pain, will increase risk taking overall. This prediction is somewhat in contrast with dual-process theories, which predict that inhibiting system 2 increases risk taking in the loss domain only, whereas it reduces risk taking in the gain domain. In a study by Porcelli and Delgado ([@CR32]), participants immersed their dominant hand in ice-cold water for 2 min, a procedure known as the cold-pressor task, before completing a decision-making task in which they chose between either two potential losses or two potential gains. The cold-pressor task is known to induce stress, which should lead to more system 1 processing. Indeed, participants who underwent the cold-pressor task were less risk taking in the gain domain but (marginally) more risk taking for equivalent gambles in the loss domain. Similar results have also been observed with time pressure (Kirchler et al. in press). An important distinction between the present study and the one by Porcelli and Delgado ([@CR32]) is that their participants experienced the painful stimulus before, rather than during, the decision phase. Thus, the authors explain their results in terms of stress rather than ongoing pain. In the present study, we investigated decisions made during the experience of a painful stimulus. On the one hand, we believe that our manipulation inhibits the reflective system 2 more effectively than common manipulations such as cognitive load, time pressure, and stress. On the other hand, pain has a unique neural signature (Wager et al. [@CR39]), implying that its effect on behavior and decision making might differ from the effect of other types of hot system 1 states. Another prediction from dual-process theories is that inhibition of system 2 leads to a greater preference for immediate over future rewards, a phenomenon known as temporal discounting. This prediction is supported by research showing that the evolutionarily old limbic and paralimbic systems are activated when participants choose immediate rewards, whereas evolutionarily newer prefrontal areas are activated when they choose delayed rewards (McClure et al. [@CR29]). Moreover, cognitive load has been shown to lead to greater discounting of delayed monetary rewards (Hinson et al. [@CR22]), and heroin addicts in the "hot" state of drug craving not only show a greater preference for immediate over delayed delivery of heroin but also for immediate over delayed delivery of monetary rewards (Giordano et al. [@CR20]). Given that pain could be viewed as a hot feeling state that reduces self-control (Loewenstein [@CR27], [@CR28]), participants in pain should show greater preference for immediate over future rewards, even if those rewards are not directly related to the pain. This prediction is in line with research showing that participants who report greater anticipation of pain prior to painful experiences (which is linked to pain sensitivity; Brañas-Garza et al. [@CR8]) are more impatient for monetary rewards (Brañas-Garza et al. [@CR9]). In the present study, pain was delivered at participants' subjective pain threshold while they made dichotomous decisions with real monetary consequences. We explored the effect of pain on risky choice (both gains and losses) and intertemporal choice. Participants completed the decision-making tasks both with and without pain (in counterbalanced order), which allowed us to explore the effect of pain both between and within subjects. Method {#Sec2} ====== Participants and materials {#Sec3} -------------------------- 109 participants (35% female; *M* age = 23.4 years, *SD* = 3.5) were recruited using the Online Recruitment System for Economic Experiments (ORSEE; Greiner [@CR21]) at Linköping University, Sweden. A sample-size calculation based on means and standard deviations from a previous study from our lab (Kirchler et al. in press) and with 70% power showed that 50 participants were needed in each condition. Data collection continued until all scheduled experimental sessions for the week of the 100th participant had been completed. All participants gave their informed consent prior to participation. Individuals were not allowed to participate if they were taking anxiolytic, antidepressive, or pain relieving medication. Participants were paid 100 SEK (approx. 12 USD) as a show-up fee plus or minus the amount from one randomly selected decision. Delayed payoffs from the intertemporal choice task were paid using Swish, a free, popular smartphone app that facilitates immediate money transfers between bank accounts. Participants who had not already installed Swish on their phone were required to do so before participating in the study. Two male participants were excluded due to technical problems during the experiment, leaving 107 participants in the final sample. Painful heat stimulation was delivered to the distal part of participants' left dorsal forearm using a 3 × 3 cm Thermal Stimulator Probe (Q-sense, Medoc). Prior to the experiment, participants' pain threshold for heat was determined by a procedure following Perini et al. ([@CR31]), in which the thermode had a baseline temperature of 32 °C and increased at a speed of 1 °C/s. Participants' task was to press a mouse button positioned in their right hand when the stimulation reached the border between painful and too painful. For safety reasons, the temperature never exceeded 50 °C. After they had pressed, the temperature returned to baseline. The procedure was repeated four times. The highest achieved temperature was selected as the pain threshold (max. 49 °C). Participants then completed a 1 min trial block in which the temperature varied between their pain threshold and 2 °C below the threshold, just as it would during the experimental trials (see below). Participants were told that the stimulation was meant to be painful but endurable and were allowed adjust to a lower or higher temperature if they perceived the stimulation as too painful or not painful enough, respectively. Pain thresholds after adjustments ranged between 40 °C and 49 °C (*M* = 48.07, *SD* = 1.38). Fifty-four participants had the maximum pain threshold of 49 °C. Pain thresholds did not differ as a function of the order in which participants were in the pain and control conditions, *t*(105) = .80, *p* = .425. A manipulation check was conducted on the last 56 participants, who indicated how painful they perceived the stimulation on a scale from 1 (*not at all painful*) to 10 (*extremely painful*). This confirmed that the stimulation was perceived as very painful (*M* = 8.43, *SD* = 1.36) and that the subjective experience of the stimulation did not differ as a function of the order of conditions, *t*(54) = .56, *p* = .578. Experimental design {#Sec4} ------------------- We used a crossover design in which participants performed three decision-making tasks twice: once with pain and once without pain. Participants thus served as their own controls. The order of the tasks was the same for all participants, but the order of the pain and control conditions was randomized between participants so that about half the participants (n = 57, 32% female) were in the pain condition first and the other half (n = 50, 40% female) were in the control condition first. In the pain condition, painful heat stimulation as described above was delivered to participants' left forearm continuously for 60 s while they completed each task. In the control condition, the thermode was placed on participants' forearm but the temperature remained at baseline (32 °C). Everything else was identical between conditions. Each task lasted for 60 s and was followed by a break of at least 30 s during which the thermode was removed from participants' arm. Thus, pain was only delivered during the decision phase. General instructions were given before the experiment started and task-specific instructions were given before each task. Participants were informed that one of their decisions would be randomly selected for actual payment. The tasks were presented on a computer screen and were programmed in Qualtrics. A translation of the complete instructions for the experiment is provided in the Supplemental Materials. ### Risky gains task {#Sec5} On each of five trials, participants chose between a safe option (receiving a sum of money with certainty) and a risky option (receiving a sum of money with 50% probability). Participants had 12 s to respond on each trial and could not proceed to the next trial until the given time had elapsed. This was done to minimize potential confounding effects of time pressure and response times. ### Risky losses task {#Sec6} On each of five trials, participants chose between a safe option (losing a sum of money with certainty) and a risky option (losing a sum of money with 50% probability). Participants again had 12 s to respond on each trial. ### Intertemporal choice task {#Sec7} On each of four[1](#Fn1){ref-type="fn"} trials, participants chose between receiving an immediate (smaller) monetary reward and a delayed (larger) monetary reward. The immediate reward was delivered either on the same day or the day after the experiment and the delayed reward was delivered one, two, five, or six days following the experiment. Participants had 10 s to respond on each trial. Data analysis {#Sec8} ------------- We first performed paired samples t-tests to investigate whether the proportion of risky and impatient choices, compared to the total number of trials in each task, was greater in the pain condition than in the control condition. We then performed regression analyses in order to confirm the results from the t-tests while also controlling for age and gender. Our regression model for the risky choice tasks was specified as follows:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_{ik} = \beta_{0} + \beta_{1} Pain + \beta_{2} Round + \beta_{3} Ratio + \beta_{4} X_{i} + \in_{ik}$$\end{document}$$where the dependent variable $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_{ik}$$\end{document}$ is a dummy variable indicating whether participant *i* chose the risky option in trial *k*. *Pain* is a dummy for the pain condition and *Round* is a dummy for the *second* round of the tasks, i.e. the second time participants performed the tasks. An alternative model also included the interaction term *Pain* × *Round*, which allows the effect of pain to differ across the two task rounds. *Ratio* is the ratio between the expected value of the risky option and the expected value of the safe option on each trial (standardized). **X** ~*i*~ is the control variables age and gender. The model was estimated using OLS and standard errors were corrected for clustering on the individual level. Our regression model for the intertemporal choice task was identical to the model above except the dependent variable $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_{ik}$$\end{document}$ was a dummy variable indicating whether participant *i* chose the immediate option in trial *k* and *Ratio* was replaced with *Delay*, which denotes the difference in days (one or five) between the immediate and the delayed reward. Thus, the regression model for the intertemporal choice task was specified as follows:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_{ik} = \beta_{0} + \beta_{1} Pain + \beta_{2} Round + \beta_{3} Delay + \beta_{4} X_{i} + \in_{ik}$$\end{document}$$For all choice tasks, we also investigated whether the effect of pain differed between male and female participants. To do this, we performed regression analyses using the models specified above where we included the interaction term *Pain* × *Female*, which allows the effect of pain to differ across genders. We also estimated risk aversion and discounting parameters for the pain and control conditions. For the estimation of risk aversion, we assumed constant relative risk aversion (CRRA) and the following utility function:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ u\left( x \right) = \left\{ {\begin{array}{*{20}c} {\frac{{x^{1 - r} }}{1 - r}\, \, {\hbox{if}}\,\, x \ge \,0} \\ { - \frac{{\left( { - x} \right)^{1 - r} }}{1 - r}\,\, {\hbox{if}}\,\, x \,< \,0} \\ \end{array} } \right. $$\end{document}$$where *r* is the coefficient of relative risk aversion and *x* is the monetary outcome. With this parameterization, *r* = 0 denotes risk-neutral behavior, *r* \> 0 denotes risk aversion (risk-seeking behavior), and *r* \< 0 denotes risk-seeking behavior (risk aversion) for gains (losses). Using this utility function, we denoted the expected utility of each alternative (A) as:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$EU\left( A \right) = \mathop \sum \limits_{x \in A} p\left( x \right)u\left( x \right)$$\end{document}$$We calculated the difference in expected utility between the safe option (S) and the risky option (R) using:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta EU = EU\left( S \right) - EU\left( R \right)$$\end{document}$$The likelihood function is:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ L = \left\{ {\begin{array}{ll} {\varPhi \left( {\Delta EU} \right) }&\quad {\hbox{if}}\, \hbox{Safe} \\ { 1 - \varPhi \left( {\Delta EU} \right)} &\quad {\hbox{if}}\, \hbox{Risky} \\ \end{array} } \right. $$\end{document}$$where Φ is the cumulative distribution function of the standard normal distribution. The likelihood function was estimated using maximum likelihood with the Broyden-Fletcher-Goldfarb-Shanno algorithm. We estimated the average risk factor for the whole sample in the pain and control conditions in the following way (allowing for heterogeneity in risk preferences):$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat{r} = r_{0} + \alpha_{0} Pain + \alpha_{1} Round$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r_{0}$$\end{document}$ represents the estimate of the constant, *Pain* is a dummy for the pain condition, and *Round* is a dummy for the second round of the task. An alternative model also included the interaction term *Pain* × *Round*, which allows the effect of pain to differ across the two task rounds. Standard errors were clustered at the individual level. Estimates were provided separately for gains and losses in order to allow for different curvature of the utility function for positive and negative outcomes. For the discount factor, we assume exponential discounting and the following function for discounted utility:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u\left( {\text{x}} \right) =\updelta^{t} u_{t} \left( {\text{x}} \right)$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document}$ denotes the discount factor and *t* denotes the number of days until the outcome will be realized. In function ([8](#Equ8){ref-type=""}) we assume that the utility function at time t, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u_{t} \left( x \right)$$\end{document}$, takes form as in ([3](#Equ3){ref-type=""}) with estimated average risk factor as described above. Using ([8](#Equ8){ref-type=""}) we calculated discounted utility separately for the immediate and the delayed reward. We then calculated the difference in discounted utility between the delayed (D) and the early (E) options using:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta U = U\left( D \right) - U\left( E \right)$$\end{document}$$We can then describe the likelihood function as:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ L = \left\{ {\begin{array}{ll} \varPhi \left( {\Delta U} \right) &\quad \hbox{if}\quad \hbox{Delayed} \\ 1 - \varPhi \left( {\Delta U} \right) & \quad \hbox{if}\quad \hbox{Early} \\ \end{array} } \right. $$\end{document}$$where Φ is the cumulative distribution function of the standard normal distribution. We used the same estimation method as for risk preferences. Similarly, we estimated the average discount factor for the whole sample in the pain and control conditions, allowing for heterogeneity in time preferences:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat{\delta } = \delta_{0} + \alpha_{0} Pain + \alpha_{1} Round$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta_{0}$$\end{document}$ represents the estimate of the constant, *Pain* is a dummy for the pain condition, and *Round* is a dummy for the second round of the task. An alternative model also included the interaction term *Pain* × *Round*, which allows the effect of pain to differ across the two task rounds. Standard errors were clustered at the individual level. We finally considered data at the individual level by investigating whether pain had the same effect on all individuals in each task and whether individuals who changed their behavior in response to pain on one task also changed their behavior in a similar fashion on the other tasks. For the latter, we calculated the difference in the proportion of risky and impatient choices in the pain condition compared to the control condition, for each individual and task. We then performed Pearson correlations of the difference scores to explore the relationship between choices in the three tasks. Results {#Sec9} ======= The effect of pain on risky choice {#Sec10} ---------------------------------- Figure [1](#Fig1){ref-type="fig"}a displays the proportion of participants' risky choices compared to the total number of trials in each condition (pain vs. control) and domain (gain vs. loss; for trial-by-trial results, see Table S1 in the Supplemental Material). A paired samples *t* test showed that the proportion of risky choices in the gain domain was greater in the pain condition than in the control condition, *M* ~*pain*~ = .68 (95% CI \[.63, .74\]), *M* ~*control*~ = .63 (95% CI \[.58, .69\]), *t*(106) = 2.06, *p* = .042, *d* = .12. The regression analyses confirm this finding (see Table [1](#Tab1){ref-type="table"}). That is, participants were more likely to choose the risky option in the pain condition than in the control condition, *β* = .049, *p* = .035. Age, gender, and *Round* (i.e., whether it was the first or second time participants performed the task) had no significant effect on decisions. However, there was a significant interaction between *Pain* and *Round*, indicating that participants who were in the pain condition first were on average more risk-seeking than participants who were in the control condition first. We return to this interaction later. There was no evidence that the effect of pain differed between the genders (see Table S2 in the Supplemental Material). The estimated parameter of risk aversion in the gain domain, when controlling for *Round*, was .133 in the pain condition and .146 in the control condition. This difference was not statistically significant, *p* = .348 (see Table S3 in the Supplemental Material). A paired samples *t* test found no difference in response times between the pain and control conditions, *M* ~*pain*~ = 4.23 s (95% CI \[3.93, 4.53\]), *M* ~*control*~ = 4.32 s (95% CI \[4.05, 4.58\]), *t*(106) = --.53, *p* = .599, *d* = --.05.Fig. 1Proportion of participants' **a** risky and **b** impatient choices as a function of condition (pain vs. control), with *error bars* showing 95% confidence intervals. \**p* \< .10, \*\**p* \< .05, \*\*\**p* \< .01 Table 1Regression analyses of risky and intertemporal choicesRisky gainsRisky lossesIntertemporal choice(1)(2)(1)(2)(1)(2)Pain.049\*\*.184\*\*\*.021.072.071\*\*\*.084(.023)(.055)(.025)(.056)(.018)(.058)Round.011.146\*\*\*.028.079.034\*.047(.023)(.054)(.025)(.052)(.018)(.058)Pain × round--.270\*\*\*--.101--.027(.099)(.099)(.116)Female--.078--.064.116\*\*.121\*\*--.033--.032(.049)(.051)(.049)(.049)(.056)(.057)Age.003.002.003.002--.006--.006(.008)(.008)(.010)(.009)(.007)(.007)Ratio.202\*\*\*.202\*\*\*--.194\*\*\*--.194\*\*\*(.014)(.014)(.016)(.016)Delay.076\*\*\*.076\*\*\*(.009)(.009)Constant.577\*\*\*.536\*\*\*.207.192.122.118 (.193)(.193)(.227)(.226)(.167)(.170)This table reports OLS coefficient estimates (robust standard errors corrected for clustering on the individual level in parentheses). For the risky choice tasks, the dependent variable is a dummy variable indicating whether participants chose the risky option. For the intertemporal choice task, the dependent variable is a dummy variable indicating whether they chose the immediate reward. "Pain" is a dummy for the pain condition. "Round" is a dummy for the *second* round of the tasks, i.e. the second time the participants performed the tasks. "Pain × Round" is the interaction between the pain condition and the task round, allowing the effect of pain to differ across the two task rounds. "Female" is a gender dummy. "Age" is the participant's age in years. "Ratio" is the ratio between the expected value of the risky option and the expected value of the safe option on each trial (standardized). "Delay" is the difference in days (1 or 5) between the immediate and the delayed reward in the intertemporal choice task\* *p* \< .10, \*\* *p* \< .05, \*\*\* *p* \< .01 In the loss domain, there was no statistically significant difference in the proportion of risky choices between the two conditions, *M* ~*pain*~ = .36 (95% CI \[.30, .42\]), *M* ~*control*~ = .33 (95% CI \[.28, 39\]), *t*(106) = 1.00, *p* = .322, *d* = .07. The regression analyses did not find a significant effect either (see Table [1](#Tab1){ref-type="table"}). That is, participants were not more likely to choose the risky option in the pain condition than in the control condition, *β* = .021, *p* = .400. There was, however, a significant effect of gender, such that women were more likely than men to choose the risky option, *β* = .116, *p* = .020. Age and *Round* had no significant effect on decisions and there was no significant interaction between pain and *Round* or between pain and gender (see Table S3 in the Supplemental Material). The estimated parameter of risk aversion in the loss domain, after controlling for *Round*, was .136 in the pain condition and .137 in the control condition. This difference was not statistically significant, *p* = .976 (see Table S3 in the Supplemental Material). There was no difference in response times between the pain and control conditions, *M* ~*pain*~ = 4.48 s (95% CI \[4.19, 4.76\]), *M* ~*control*~ = 4.58 s (95% CI \[4.27, 4.89\]), *t*(106) = --.57, *p* = .569, *d* = --.06. In short, the results from the risky choice tasks suggest that acute pain increases risk seeking in the gain domain, but not in the loss domain. However, there was also a significant interaction between pain and the order in which participants were in the pain and control conditions---that is, in the gain domain, participants who were in the pain condition first were always more risk seeking than participants who were in the control condition first. An additional analysis examined only the first round of each task (i.e., the first time participants completed each task), where choices were not influenced by choices from previous rounds. Results are presented separately for each trial in Fig. [2](#Fig2){ref-type="fig"} (see also Table S4 in the Supplemental Material). An independent samples t-test showed that the overall proportion of risky choices in the gain domain was significantly greater in the pain condition than in the control condition, *M* ~*pain*~ = .74 (95% CI \[.67, .81\]), *M* ~*control*~ = .55 (95% CI \[.47, .64\]), *t*(105) = 3.47, *p* \< .001, *d* = .69. The difference in the proportion of risky choices in the loss domain was not statistically significant, *M* ~*pain*~ = .37 (95% CI \[.28, .45\]), *M* ~*control*~ = .30 (95% CI \[.22, .38\]), *t*(105) = 1.09, *p* = .280, *d* = .23. The regression analyses corroborate these findings (see Table [1](#Tab1){ref-type="table"}). That is, pain increased the likelihood of choosing the risky option in the gain domain, *β* = .184, *p* = .001, but not in the loss domain, *β* = .072, *p* = .160. In the gain domain, the estimated parameter of risk aversion was .088 in the pain condition and .190 in the control condition. This difference was statistically significant, *p* \< .001. In the loss domain, the estimated parameter of risk aversion was .146 in the pain condition and .124 in the control condition. This difference was not statistically significant, *p* = .452 (see Table S3 in the Supplemental Material). Thus, the between-subjects results confirm the finding that pain increases risk seeking for gains but not for losses.Fig. 2Percent frequency of **a** risky choices in the gain domain, **b** risky choices in the loss domain, and **c** impatient choices in the intertemporal choice task, presented per trial as a function of condition (pain vs. control) for the first round of each task, with error bars showing 95% confidence intervals The effect of pain on intertemporal choice {#Sec11} ------------------------------------------ Figure [1](#Fig1){ref-type="fig"}b shows the proportion of participants' impatient choices in the pain vs. control condition compared to the total number of trials in the intertemporal choice task (for trial-by-trial results, see Table S1 in the Supplemental Material). A paired samples t-test showed that the proportion of impatient choices was greater in the pain condition than in the control condition, *M* ~*pain*~ = .28 (95% CI \[.22, .35\]), *M* ~*control*~ = .22 (95% CI \[.22, .27\]), *t*(106) = 3.84, *p* = .0002, *d* = .16. The results from the regression analyses corroborate this finding (see Table [1](#Tab1){ref-type="table"}). That is, participants were more likely to choose the impatient option in the pain condition than in the control condition, *β* = .071, *p* = \< .001. Age, gender, and *Round* had no significant effect on the proportion of impatient choices and there was no interaction between pain and *Round* or between pain and gender (see Table S2 in the Supplemental Material). The estimated discount factor, after controlling for *Round*, was .969 in the pain condition and .971 in the control condition. This difference was statistically significant, *p* \< .001. There was no difference in response times between the pain and control conditions, *M* ~*pain*~ = 3.62 s (95% CI \[3.38, 3.48\]), *M* ~*control*~ = 3.71 s (95% CI \[3.48, 3.93\]), *t*(106) = --.58, *p* = .566, *d* = --.06. Results from the first round of the intertemporal choice tasks are presented in Fig. [2](#Fig2){ref-type="fig"}c (see also Table S4 in Supplemental Material). The difference in the proportion of impatient choices was in the same direction as in the above analyses but was not statistically significant, *M* ~*pain*~ = .27 (95% CI \[.19, .35\]), *M* ~*control*~ = .19 (95% CI \[.11, .27\]), *t*(105) = 1.46, *p* = .147, *d* = .27. The regression analyses did not yield a significant effect either, *β* = .084, *p* = .144, (see Table [1](#Tab1){ref-type="table"}). The estimated discount factor was .969 in the pain condition and .971 in the control condition. This difference was not statistically significant, *p* = .119 (see Table S3 in the Supplemental Material). However, the non-significance might just reflect the decrease in statistical power when investigating between-subjects differences only. In contrast to the risky choice tasks, the order of conditions did not significantly influence the effect of pain on decision making in the intertemporal choice task. Thus, we conclude that pain increases preferences for immediate over future rewards. Individual differences in the effect of pain across tasks {#Sec12} --------------------------------------------------------- We found no evidence that the same individuals were driving the effect across all tasks, because the pain vs. control difference scores for the three tasks were only weakly correlated at best (gain and loss, *r* = −.181, *p* = .063; gain and intertemporal choice, *r* = −.071, *p* = .467; loss and intertemporal choice, *r* = .175, *p* = .074). Thus, we conclude that there was not just one subgroup of participants whose decisions were influenced by the painful stimulation. See Figure S1 in the Supplemental Material for a graphical depiction of the effect of pain on each task at the individual level. Discussion {#Sec13} ========== Pain is a highly aversive and attention-demanding experience (Eccleston and Crombez [@CR13]; Legrain et al. [@CR26]). Although responses to pain have evolutionarily adaptive value, performance on cognitively demanding tasks such as decision making may be compromised. Here we showed that acute physical pain influences risky and intertemporal choices involving money. Participants experiencing a painful stimulus were more risk seeking for gains but not for losses and showed greater preferences for immediate (smaller) over future (larger) monetary rewards than participants experiencing a non-painful stimulus. We interpret these results as a motivation to compensate for the negative state induced by pain. The results indicate that pain increases risk seeking for monetary gains but not for monetary losses. Although in line with previous research on chronic pain patients (Apkarian et al. [@CR2]; Berger et al. [@CR5]; Biagianti et al. [@CR7]; Elvemo et al. [@CR15]; Tamburin et al. [@CR35]; Verdejo-García et al. [@CR37]; Walteros et al. [@CR40]), these findings partly go against dual-process theories, which predict that inhibition of system 2 leads to greater reliance on automatic biases such as the reflection effect of prospect theory (Evans [@CR17]; Evans and Stanovich [@CR18]; Kahneman and Frederick [@CR23]; Kirchler et al. in press; Porcelli and Delgado [@CR32]). Thus, our findings suggest that pain does more than inhibit the reflective system 2---it produces risk seeking behavior for potential rewards. This suggestion is supported by previous research showing that pain increases the motivation to obtain a monetary reward, as indicated by faster response times in the monetary incentive delay (MID) task during pain than during a control condition (Gandhi et al. [@CR19]). Seeking risky rewards may even be an adaptive response to pain, because monetary gains can reduce the subjective intensity and unpleasantness of a painful experience (Becker et al. [@CR4]). Thus, a risky choice in the present study could be interpreted as an attempt to relieve pain. Pain also increased preferences for immediate (smaller) over future (larger) rewards. These results are in line both with dual-process approaches to decision making and with previous evidence that monetary rewards can offset the pain-induced negative state (Becker et al. [@CR4]). If monetary rewards act as pain relievers, and pain is experienced temporarily at the moment the decision is made, then it makes sense to choose an immediate over a delayed reward. After all, a pain reliever is most useful in a moment of pain. These findings have implications for understanding how "hot" feeling states influence decision making. Feeling states such as pain have previously been predicted to only influence decisions that are directly related to those states (Loewenstein [@CR27]). For example, drug addicts temporally discount drugs to a greater extent than other kinds of rewards (Giordano et al. [@CR20]). However, the current study shows that the hot state of pain influences temporal discounting of money, a reward that is not directly related to pain. Our general findings that pain influences risk seeking and impatience can be linked to a literature on the role of incidental emotions in decision making. Raghunathan and Pham ([@CR33]) noted that sad participants were more risk seeking whereas anxious participants were less risk seeking than control participants and suggested that emotions may have different effects on judgments and decisions depending on their informational content and implicitly activated goals. Feelings of sadness can be interpreted as a lack of something rewarding, which activates the goal of reward-seeking. Feelings of fear or anxiety involve uncertainty and lack of control, which activate the goal of uncertainty reduction. In line with this "mood repair" account, anticipation of painful, as opposed to non-painful, electric shocks has been found to reduce risk taking in an investment task (Cohn et al. [@CR12]). Furthermore, activity in the ventromedial prefrontal cortex and ventral striatum predict choices in a risk task in the absence of threat, whereas activity in the insula predicts choices in the presence of threat (Engelmann et al. [@CR16]). It should be noted that these studies involve anticipation of painful shocks, which are short-lived and induce a state of anxiety and uncertainty. In contrast, the present study involved continuous delivery of thermal stimulation, which induces an ongoing state of pain but which lacks the uncertainty component. This difference between ongoing, actual pain and anticipation of painful shock might explain why we observed an effect of experienced pain on choices but Cohn et al. ([@CR12]) did not. Furthermore, a speculative interpretation of our findings that is in line with the mood repair literature is that pain induces a negative emotional state and that participants attempt to repair their mood by opting for risker but higher reward options or rewards that are delivered closer in time. A limitation of the present study is that the choice tasks were presented in the same order for all participants (i.e., risky gains, risky losses, intertemporal choice). We therefore cannot rule out the possibility of order effects. For instance, the reason we did not find a significant effect of pain in the loss domain could be that participants habituated to the pain and were more familiar with the type of choice task. However, we did observe an effect in the intertemporal choice task even though it was placed last in the experiment. Moreover, Berger et al. ([@CR5]) found that patients with chronic pain were more risk seeking than control participants, but only for gains and not for losses, which is in line with our findings. A second limitation is the possibility that only individuals with a high tolerance for pain signed up to participate. It is unclear how such self-selection bias may have influenced our results. If we were to speculate, it seems most likely that the effect of pain is stronger among individuals who did not sign up, because anticipation of pain has been shown to be positively correlated with both sensitivity to pain (Brañas-Garza et al. [@CR8]) and impatience for monetary rewards (Brañas-Garza et al. [@CR9]). Future research needs to identify the exact processes behind the effects observed in the present study. For instance, does pain influence the value function or the probability weighting function of prospect theory? Furthermore, in the present study, we compared the influence of "hot" (painful) and "cold" (non-painful) states on decision-making while keeping the decisions constant. However, pain can also result from the decision itself, as when an individual experiences a monetary loss (Kuhnen and Knutson [@CR25]; Paulus et al. [@CR30]; Samanez-Larkin et al. [@CR34]; Västfjäll et al. [@CR36]; Wu et al. [@CR41]). Probability weighting for painful electric shocks has been found to be similar to probability weighting for monetary losses (Berns et al. [@CR6]), Future research should investigate the relative contributions of incidental pain, which is unrelated to the decision at hand, and integral pain, which results from the decision itself. Additionally, experiences such as social rejection (Eisenberger et al. [@CR14]) and economic insecurity (Chou et al. [@CR11]) can feel physically painful. The question remains whether the effect of pain on risky and intertemporal choice generalizes to pain from non-physical sources. Finally, it would be good if future replication studies were conducted, especially for risky choices given that our results were less conclusive for this domain when modeled within subject. In conclusion, the present study demonstrates a behavioral effect of acute, physical pain on economic decision making. Participants were more risk seeking (for gains, but not for losses) and more impatient when experiencing a painful, compared to a non-painful, stimulus. It is possible that the effect is even greater in real-life settings outside the lab, where individuals have little to no control over the pain they experience and where they may be less aware that the pain influences their behavior and decisions. Thus, our findings contribute to the understanding of decision making in everyday life, which is filled with painful experiences. Electronic supplementary material ================================= {#Sec14} Below is the link to the electronic supplementary material. Supplementary material 1 (DOCX 217 kb) We originally included a fifth question (20 s) in which participants indicated how much a sum of money (1000 SEK in the first round and 100 SEK in the second) would be worth if given to them a month from now. However, because 27% of participants failed to respond on time in the first round of the tasks, we have omitted that question from our data analysis. **Electronic supplementary material** The online version of this article (doi:10.1007/s10683-017-9515-6) contains supplementary material, which is available to authorized users. This research was funded by the Ragnar Söderberg Foundation and Marianne and Marcus Wallenberg Foundation.

Introduction {#sec1} ============ The worldwide incidence of geriatric hip fracture is increasing. This phenomenon can be explained largely through the extreme increase of incidence of hip fracture in most Asian countries in the last decades. It is expected that by 2050 more than 50% of all osteoporotic fractures will occur in Asia.[@bib1], [@bib2] Ageing is one of the major contributors to this phenomenon, as well as the incidence of osteoporosis and the process of urbanization.[@bib3] On the contrary in most developed parts of the world the incidence of hip fracture has decreased or developed a plateau in the age-adjusted rates in the last decades.[@bib3], [@bib4], [@bib5] Hip fracture is one of the most common orthopedic causes leading to hospital admission in the geriatric population and is associated with a high morbidity and mortality rate.[@bib6], [@bib7], [@bib8], [@bib9] Surgical treatment options depend greatly on fracture type which can be divided into intracapsular fractures (femoral neck) and extracapsular fractures (trochanteric fractures). Extracapsular fractures can be stabilized by extramedullary and intramedullary implants. Intramedullary nailing devices allow load sharing with immediate mobilization and show less intraoperative blood loss, shorter operation time and lower postoperative complication rates. In unstable fracture patterns, intramedullary devices appear to have a biomechanical advantage over extramedullary devices, lowering the force imposed on the implant due to the shorter lever arm of the fixation.[@bib10], [@bib11], [@bib12] The most common implant-related complication is lag screw migration with an incidence of 2%--13%.[@bib13] The dual lag screw intramedullary nail systems[@bib14], [@bib15] were designed to improve the rotational stability and bony purchase within the femoral head, thus resisting cut-out and subsequent fixation failure.[@bib16] The dual lag screw design provides equivalent rigidity and stability compared to an intramedullary nail with a single lag screw, and has a significantly higher failure strength.[@bib16] This implant design also brings a new type of failure: the Z-effect phenomenon, which describes the appearance of a lateral migration of the inferior lag screw and medial migration of the superior lag screw during the weight bearing rehabilitation period.[@bib10], [@bib17], [@bib18], [@bib19], [@bib20] The reversed Z-effect describes the opposite effect: migration of the inferior lag screw medially and the superior lag screw laterally. Most studies about intramedullary nailing report the incidence of (reversed) Z-effect but very few studies have studied this phenomenon specifically. The cause of the Z-effect is thought to be biomechanical, possibly due to a misbalance in head and neck compressive strength leading to varus collapse.[@bib21] The precise etiology of the Z-effect requires further clarification.[@bib20] Suggested explanations for the development of the Z-effect were medial migration due to lateral buttress deficiency, unstable medial cortex[@bib21] and constant friction within the femoral head and axial loading in varus.[@bib22] Another study[@bib19] found that a cervicodiaphyseal angle of \<125° of the postoperative X-rays was associated with the development of a Z-effect. An inadequate fracture reduction or entry point and osteoporotic bone might also be contributing factors, but convincing evidence lacks. We hypothesize that long nails provide more neck strength due to a longer lever than short nails and are therefore less likely to develop a Z-effect. Materials and methods {#sec2} ===================== This retrospective study was conducted in the department of Trauma Surgery in the Maxima Medical Center in the Netherlands, a large rural teaching hospital. From the 1st January 2006 till 31st December 2007 all patients treated with a dual lag screw intramedullary nailing for trochanteric hip fracture were analysed. In this period two types of nails were used: the Recon nail (Stryker, USA) and the Trigen nail (Smiths & Nephews, UK). A follow-up period of 2 years was chosen for implant-related failures. We selected all patients who received surgery for hip fracture and excluded the polytrauma patients and other fixation types. Patients with a malignancy in their history or pathological fractures were not excluded. A protocolised treatment algorithm regarding hip fractures was used. Standard work-up after admission to the emergency department consisted of a detailed history, a complete physical examination, an electrocardiography and standard biochemical and hematologic tests. All medical records were evaluated for patient and operation characteristics such as sex, age, American Society of Anesthesiologists (ASA) score, fracture type by the AO-classification system (Arbeitsgemeinschaft für Osteosynthesefragen, Switzerland), anesthesia type, delay to surgery, operation time and days of hospitalization. Furthermore we investigated postoperative complications and re-interventions during the 2-year follow-up. Radiological analyses were performed to measure the position of the superior and inferior lag screw, the cervicodiaphyseal angle and the Tip Apex Distance (TAD). The TAD is a clinically useful way to describe the position of the lag screws by the sum of the distance from the tip of the lag screw to the apex of the femoral head on the anteroposterior radiograph and the same distance on the axial radiograph with correction for magnification.[@bib23] The amount of radiographic magnification was determined by dividing the diameter of the projected shaft of the lag screw as seen on the radiograph by its known diameter. By measuring the TAD over time we studied the incidence of (reversed) Z-effect. The maximum TAD (TAD^max^) was measured during the follow-up period or before re-operation. To study migration of lag screws over time we calculated the TAD difference (TAD^diff^), which describes the maximum TAD minus the immediate postoperative TAD. The quality of the fracture reduction was based on the fracture alignment and the fracture displacement. This was judged by the cervicodiaphyseal angle, the degree of angulation and the displacement between fracture fragments and shortening. The fracture reduction status was subsequently scored as good (scored as: 2), acceptable (scored as: 1) and poor (scored as: 0).[@bib24], [@bib25] The position of the inferior lag screw in the femoral head was determined and judged as correct when placed as inferior or centrally on the anteroposterior radiograph and central placement on the axial radiographs.[@bib26] All radiological measurements are displayed in [Fig. 1](#fig1){ref-type="fig"}.Fig. 1Detailed radiological measurements. 1. Tip Apex Distance (TAD) of the superior screw, 2. Tip Apex Distance (TAD) of the inferior screw, 3. Collodiaphyseal angle (Anterior posterior view), 4. Angulation of the inferior lag screw (\<20% on the Axial view), 5. Fracture dislocation (\>80% overlapping in both planes AP and Ax; \<5 mm of shortening), a--f represent the zones in which the inferior lag screw can be placed in the femoral head.Fig. 1 Statistical analysis {#sec2.1} -------------------- All analyses were performed with SPSS 16.0 statistical software (SPSS Inc., Chicago, Illinois, USA). *p* \< 0.05 was considered to be statistically significant. Data were presented as mean for normally distributed or median for non-normally distributed variables. Percentages were used when appropriate. One-way ANOVA was used to compare normally distributed variables and the Mann Whitney *U* test for non-normally distributed continuous variables with Bonferroni correction for multiple testing. A Pearson\'s chi-square (χ^*2*^) test was used to investigate whether distributions of categorical variables differ from one another. To investigate potential predictors for re-intervention we performed a multivariate logistic regression analysis. All theoretically important variables were entered in the model. Re-intervention was chosen as outcome of the model so that other implant-related complications like cut-out, cut-through and non-union could be included in the analysis. Results {#sec3} ======= During the study period of 2006--2007, 224 patients were operatively treated for trochanteric hip fracture. Of these patients 103 received an intramedullary nail system with dual lag screws. In 23/103 (22%) patients there was only one X-ray control postoperatively. These patients were regarded as loss to follow-up. In these 23 patients there were no signs of a Z-effect during hospitalization. The remaining 80 patients enrolled in this study had data of 2-year follow-up. [Table 1](#tbl1){ref-type="table"} shows the baseline data and operation characteristics. There were significantly more unstable fracture types (A2.2, A2.3 A3.3) and A3 fracture types treated with a long nail (*p* \< 0.001). There were no other differences in baseline or operation characteristics between patients who received a short nail or long nail. The long nail group patients had a significant longer hospital stay (13 vs 21 days, *p* \< 0.05).Table 1Baseline data and operation characteristics.Table 1VariablesShort nail (*n* = 58)Long nail (*n* = 22)*p*Age (years)76 (SD 14)78 (SD 18)0.8Male sex \[*n* (%)\]14 (24)8 (36)0.3ASA score ASA (average)2.2 (SD 0.7)2.4 (SD 0.9)0.2 ASA 1 \[*n* (%)\]7 (12)2 (9)0.7 ASA 2 \[*n* (%)\]31 (53)10 (46)0.5 ASA 3 \[*n* (%)\]19 (33)8 (36)0.8 ASA 4 \[*n* (%)\]1 (2)2 (9)0.1AO classification \[*n* (%)\] A1.1 -- A1.314 (24)2 (9)0.1 A2.1 -- A2.333 (57)3 (14)0.001 A3.1 -- A3.311 (19)17 (77)\<0.001Spinal anesthesia \[*n* (%)\]43/51 (84)14/19 (74)0.3General anesthesia \[*n* (%)\]8/51 (16)5/19 (26)0.3Delay to surgery (h)10 (IQR 13)6 (IQR 13)0.6Operation time (min)71 (SD 31)103 (SD 26)0.6Number of patient days13 (IQR 11)21 (IQR 29)\<0.05[^1][^2] [Table 2](#tbl2){ref-type="table"} lists the radiological measurements displayed and classified by nail length. The TAD, TAD^max^, TAD^diff^ were not significantly different between the short nail group and the long nail group. The quality of the fracture reduction, expressed by a score (0--2), was significantly worse in the long nail group (*p* \< 0.05). Furthermore, the inferior lag screw was inserted at a larger angle in the long nail group in comparison with the short nail group (*p* \< 0.05). The correct position of the inferior lag screw in the femoral head was not different between both groups. Furthermore there were no differences in the collodiaphyseal angle or average screw migration between both groups.Table 2Radiological results classified by nail length.Table 2VariablesShort nail (*n* = 58)Long nail (*n* = 22)*p*Superior screw TAD (mm)26 (IQR 16)25 (IQR 16)0.9Superior screw TAD^max^ (mm)28 (IQR 17)27 (IQR 16)0.7Superior screw TAD^diff^ (mm)4 (IQR 5)2 (IQR 8)0.3Inferior screw TAD (mm)24 (IQR 16)26 (IQR 13)0.5Inferior screw TAD^max^ (mm)27 (IQR 21)25 (IQR 14)0.9Inferior screw TAD^diff^ (mm)3 (IQR 7)4 (IQR 6)0.8Quality fracture reduction (score 0--2)2 (IQR 1)1.5 (IQR 2)\<0.05Angle inferior lag screw (Ax, °)4 (IQR 9)9 (IQR 12)\<0.05Collodiaphyseal angle (AP, °)134 (IQR 12)133 (IQR 8)0.7Correct placement inferior lag screw in femoral head[a](#tbl2fna){ref-type="table-fn"} \[*n* (%)\]37 (64)12 (55)0.4[^3][^4][^5] Seven patients developed a (reversed) Z-effect, leading to an incidence of 9%. Six of them were treated with a short nail, which did not reach significance in comparison with the long nail group. Radiological measurements displayed and classified by the presence of (reversed) Z-effect are summarized in [Table 3](#tbl3){ref-type="table"}. The TAD^diff^ of the inferior lag screw was significantly different between the Z-effect group vs no Z-effect group (*p* = 0.001). [Table 4](#tbl4){ref-type="table"} shows the radiological measurements classified by implant failure (cut-out, cut-through and Z-effect). Significant displacement occurred of the superior and inferior lag screw over time in the implant failure group vs no implant failure (*p* \< 0.001 and *p* = 0.003 respectively). Four patients needed a re-operation because of the presence of a (reversed) Z-effect. A reversed Z-effect occurred in 2 patients (1 in short nail group, 1 in long nail group) and was always a reason for re-operation. Patients who were treated with a long nail had a significant larger number of complications in comparison with the short nail group (median 2 vs 0.5, *p* = 0.001). The long nail group received more often erythrocytes blood transfusions (82% vs 31%, *p* \< 0.001) ([Table 5](#tbl5){ref-type="table"}).Table 3Radiological results classified by the presence of a (reversed) Z-effect.Table 3VariablesZ-effect[\*](#tbl3fn1){ref-type="table-fn"} (*n* = 7)No Z-effect (*n* = 73)*p*AO classification \[*n* (%)\] A1.1 -- A1.32 (29)14 (19)0.5 A2.1 -- A2.34 (57)32 (44)0.5 A3.1 -- A3.31 (14)27 (37)0.2Superior screw TAD (mm)21 (IQR 17)26 (IQR 17)0.5Superior screw TAD^max^ (mm)26 (IQR 41)28 (IQR 15)0.9Superior screw TAD^diff^ (mm)6 (IQR 7)3 (IQR 5)0.2Inferior screw TAD (mm)24 (IQR 14)25 (IQR 15)0.6Inferior screw TAD^max^ (mm)26 (SD 41)28 (SD 15)0.4Inferior screw TAD^diff^ (mm)13 (SD 66)3 (SD 5)0.001Quality fracture reduction (score 0--2)2.0 (IQR 1.0)2.0 (IQR 1.0)0.8Angle inferior lag screw (Ax, °)9 (SD 8)7 (SD 9)0.7Collodiaphyseal angle (AP, °)131 (SD 8)133 (SD 8)0.6Correct placement inferior lag screw in femoral head[a](#tbl3fna){ref-type="table-fn"} \[*n* (%)\]4 (57)46 (63)0.3[^6][^7][^8][^9]Table 4Radiological results classified by the presence of implant failure (cut-out, cut-through or the presence of a Z-effect or reverse Z-effect).Table 4VariablesImplant complication (*n* = 17)No implant complication (*n* = 63)*p*AO classification \[*n* (%)\] A1.1 -- A1.32 (12)14 (22)0.4 A2.1 -- A2.311 (65)25 (40)0.1 A3.1 -- A3.34 (24)24 (38)0.3Superior screw TAD (mm)21 (IQR 25)26 (IQR 15)0.8Superior screw TAD^max^ (mm)26 (IQR 43)28 (IQR 14)0.7Superior screw TAD^diff^ (mm)10 (IQR 43)3 (IQR 4)\<0.001Inferior screw TAD (mm)22 (IQR 12)25 (IQR 16)0.3Inferior screw TAD^max^ (mm)23 (IQR 51)25 (IQR 18)0.2Inferior screw TAD^diff^ (mm)13 (SD 51)3 (IQR 5)0.003Quality fracture reduction (score 0--2)2.0 (IQR 1.0)2.0 IQR 1.00.5Angle inferior lag screw (Ax, °)7 (SD 6)8 (SD 8)0.7Collodiaphyseal angle (AP, °)131 (SD 8)134 (SD 8)0.3Correct placement inferior lag screw in femoral head[a](#tbl4fna){ref-type="table-fn"} \[*n* (%)\]11 (65)38 (60)0.7[^10][^11][^12]Table 5Postoperative complications.Table 5ComplicationsShort nail (*n* = 58)Long nail (*n* = 22)*p*Number of complications\*0.5 (IQR 1)2.0 (IQR 2)0.001Cardiac complications3 (5)1 (5)0.9Pulmonary complications3 (5)2 (9)0.5Blood transfusion18 (31)18 (82)\<0.001Urinary tract infections13 (22)12 (55)0.006Z-effect5 (9)0 (0)0.2Reversed Z-effect1 (2)1 (5)0.5Re-operation11 (19)6 (27)0.4Cut-out1 (2)1 (5)0.5Cut-through8 (14)3 (14)1.0Non-union1 (2)[a](#tbl5fna){ref-type="table-fn"}2 (9)[b](#tbl5fnb){ref-type="table-fn"}0.1[^13][^14][^15][^16] There were 17 re-operations performed within two years after primary surgery: 4/17 (23%) of cases due to a (reversed) Z-effect and 13/17 (76%) of cases due to a screw-related implant failure. All reasons for re-operation are mentioned in [Table 6](#tbl6){ref-type="table"}. There were significantly more deep infections in the long nail group (*p* \< 0.05).Table 6Reasons for re-operation classified by nail length \[n (%)\].Table 6Risk factors for re-operationShort nail (*n* = 58)Long nail (*n* = 22)*p*Z-effect2 (3)0 (0)0.4Reversed Z-effect1 (2)1 (5)0.5Cut-out1 (2)0 (0)0.5Cut-through5 (9)1 (5)0.5Deep infection0 (0)2 (9)\<0.05Avascular necrosis of the femoral head (AVN)1 (2)00.5Pain from screws/breakage of screws0 (0)2 (9)0.1Periprosthetic fracture or break-out nail1 (2)0 (0)0.5 All possible predictors for re-operation were entered in a logistic regression analysis ([Table 7](#tbl7){ref-type="table"}). Migration of the superior/inferior lag screw over time and unstable fracture type were the predicting variables for re-operation. All other radiological variables used in this study were not associated with re-operation.Table 7Regression analyses for re-operation.Table 7Variables*p* valueBStd. errorSex0.4−0.0020.003Age0.60.090.1ASA0.70.030.07Nail length0.60.050.1Unstable fracture type[a](#tbl7fna){ref-type="table-fn"}\<0.050.20.1Superior screw TAD^diff^0.0090.0090.003Inferior screw TAD^diff^0.0070.0070.003Quality fracture reduction (score 0--2)0.90.0130.07Angle inferior lag screw (Ax)0.7−0.0030.006Collodiaphyseal angle (AP)0.5−0.0040.005[^17][^18][^19] Discussion {#sec4} ========== The incidence of Z-effect in this study was 9%. The length of the nail was not associated with the development of a Z-effect. Migration of lag screws after intramedullary nailing is common, and a risk factor for re-operation. The re-operation rate within two years after surgery was 21%, of which one fourth was due to a Z-effect. Unstable fracture type was a predictor for re-operation. We hypothesized that long nails provides more neck strength due to a longer lever and are therefore less likely to develop a misbalance of a higher head compressive strength than neck compressive strength, contributing to the Z-effect. In this study we did not find any convincing evidence for this hypothesis. There was no association between nail length and the development of a Z-effect, although 6 out of 7 Z-effects occurred in the short nail group. These findings suggest that other variables are responsible for the occurrence of a Z-effect. More unstable fracture types and more high-energy trauma in the long nail group could explain the difference in quality of the osteosynthesis. This might also be the reason for the higher incidence of blood transfusion in the long nail group. The operation time was also longer in the long nail group but did not reach significance. The overall incidence of Z-effect in this study was 9%, which is in line with other studies that report an incidence of 7.1%--13% after the proximal femoral nail fixation with antirotation screw (PFNa, dual screw device).[@bib18], [@bib19], [@bib27] The migration of lag screws in the Z-effect group ([Table 3](#tbl3){ref-type="table"}) is probably underestimated because Z-effect (*n* = 5) and reversed Z-effect (*n* = 2) occur in the opposite direction. Therefor all patients with implant failure (cut-out, cut-through and Z-effect) were grouped to study migration of lag screws more closely ([Table 4](#tbl4){ref-type="table"}). Patients with implant failure did not have worse baseline characteristics of the primary placed dual lag screw system regarding fracture type, fracture reduction, collodiaphyseal angle or position of the lag screws in the femoral head. Significant migration of superior and inferior lag screws occurred over time in the case of implant failure (*p* \< 0.001 and *p* = 0.003 respectively); the maximum migration for the superior lag screw was 82 mm, and for the inferior lag screw was 97 mm. In 5 cases screw migration occurred with perforation of the acetabulum. Among them, 4 were due to cut-through and one developed a reversed Z-effect. The cause of this phenomenon is multifactorial. Osteoporotic bone and unstable fracture types have been previously mentioned to be risk factors.[@bib28], [@bib29], [@bib30] The principles behind the second or antirotation screw are clear, but strong clinical evidence for advantages lacks. In a randomized controlled trial the Dynamic Hip Screw (DHS) was compared to the PFNa, but no statistically significant differences were found regarding implant failure.[@bib31] Another study compared PFNa with conventional gamma nails and concluded no difference in clinical outcome. However, the PFNa demonstrated better results biomechanically in terms of less sliding of lag screw, less change of neck shaft angle, and less complications for the treatment of reverse obliquity intertrochanteric fractures.[@bib32] A previous study reported that the PFNa did not improve the position of the implant in the femoral head compared with the PFN without anti-rotation screw.[@bib33] This suggests that due to the use of a dual screw system, there might be an increased risk for suboptimal placement. If the inferior lag screw is placed centrally, it may result in an unavoidable cranially placed superior lag screw.[@bib34] The key to prevent Z-effects might be a optimal position of the lag screws in the head by inserting the inferior lag screw over the calcar of the femoral neck in order to achieve better bony anchoring, thus preventing screw cut-out. Strauss et al[@bib21] suggested in their biomechanical study that in cases of intertrochanteric hip fractures with significant medial cortical comminution, surgeons may wish to avoid the use of a dual lag screw intramedullary nail. We acknowledge a few limitations for interpreting the results of our study. Of 23 patients there was a limited follow-up available; therefore an underestimation of the results could be made. The loss to follow-up could be explained by early death or when no further radiologic follow-up was performed. Most patients loss to follow-up were suffering from dementia and discharged to a nursing home. Furthermore selection bias cannot be excluded because of the retrospective study design. Although this is one of very few studies that specifically report about Z-effect after intramedullary nailing, the number of patients is relatively small, making the risk of type II error higher. The measured TAD depends on the angle the X-rays are shot, which makes it difficult to compare accurately, but we did use all available X-rays in the follow-up and corrected for radiographic magnification. There exists an overall difficulty of comparing long and short nails with another as well as A1 fractures with A2 or A3 fractures. We corrected our analysis for the presence of unstable fracture types. In our study we did not only report implant-related complications but also looked at predictors for re-operation. Unstable fracture type was a predictor for re-operation as well as migration of the superior or inferior lag screw over time. The contributing effect of osteoporosis to implant failure was not included in our analysis. To prevent re-operations after intramedullary nailing, a stabile implant resistant to migration of lag screws is needed, even more in unstable comminuted trochanteric fractures. Dual lag screw systems have improved over the years. Henschel et al[@bib35] found that the stresses were distributed more equally between the two cephalic screws in the bone and the implant if a longer inferior screw was used. The Targon nail^®^ (B. Braun, Aesculap) interlocking nail system was developed on the same biomechanical principle. In a randomized trial the Targon nail seems to be superior to the sliding hip screw (SHS) in the case of unstable A3 trochanteric fractures.[@bib36] Fractures treated with SHS were more likely to show \>50% medicalization compared with the Targon nail in A3 fractures (38.1% vs 3.8%, *p* = 0.006). Medialization \>50% was a predictor for revision surgery. Another dual lag screw nail was reintroduced (Trigen Intertan, Smith-Nephew) and designed with two integrated lag screws to overcome Z-effect complications, and provides immediate intraoperative linear compression and rotational stability. Several studies have shown reliable outcome without the presence of Z-effect.[@bib34], [@bib37], [@bib38] A prospective randomized comparative study of the Endovis^®^ Cephalomedullary dual lag screw nail vs DHS showed comparable outcome, with implant failure rates of 9% and 11% respectively.[@bib39] Lag screw migrations only occurred in patients with unstable trochanteric fractures with comminution of the medial cortex. Finally, a recent prospective, randomized, multicenter clinical trial compared the Trigen Intertan nail (Smith-Nephew) with DHS. This study showed that most patients with intertrochanteric femur fractures could expect similar functional results whether treated with an intramedullary or extramedullary device. Sub-group analysis showed that active and functional patients with unstable trochanteric fractures have better outcome regarding less shortening, better Functional Independence Measure (FIM) and Timed Up and Go test (TUG) after one year follow-up when the Trigen Intertan nail is used.[@bib40] More prospective randomized comparative studies of recent dual lag screw systems with other fixation types are warranted to conclude what osteosynthesis is preferable in unstable trochanteric fractures. Peer review under responsibility of Daping Hospital and the Research Institute of Surgery of the Third Military Medical University. [^1]: Note: Data are presented as mean when normally distributed and as median when non-normally distributed. [^2]: SD = Standard deviation; IQR = Interquartile range; ASA score = American Society of Anesthesiologists score. [^3]: Note: Data are presented as mean when normally distributed and as median when non-normally distributed. [^4]: TAD^diff^ = Tip Apex Distance difference of TAD^max^ (maximal measured TAD distance over time) minus the TAD (first postoperative measured TAD). AP = anteroposterior view, Ax = axial view. [^5]: Correct placement established as inferior or centrally placement of the lag screw on the anteroposterior radiograph and central placement on the axial radiographs. [^6]: Note: Data are presented as mean when normally distributed and as median when non-normally distributed. [^7]: SD = Standard deviation; IQR = Interquartile range; ASA score = American Society of Anesthesiologists score. TAD^diff^ = Tip Apex Distance difference of TAD^max^ (maximal measured TAD distance over time) minus the TAD (first post-operative measured TAD). AP = anteroposterior view, Ax = axial view. [^8]: Correct placement established as inferior or centrally placement of the lag screw on the anteroposterior radiograph and central placement on the axial radiographs. [^9]: Z-effect or reversed Z-effect. [^10]: Note: Data are presented as mean when normally distributed and as median when non-normally distributed. [^11]: SD = Standard deviation; IQR = Interquartile range; ASA score = American Society of Anesthesiologists score. TAD^diff^ = Tip Apex Distance difference of TAD^max^ (maximal measured TAD distance over time) minus the TAD (first post-operative measured TAD). AP = anteroposterior view, Ax = axial view. [^12]: Correct placement established as inferior or centrally placement of the lag screw on the anteroposterior radiograph and central placement on the axial radiographs. [^13]: Note: Data are presented as mean when normally distributed and as median when non-normally distributed. [^14]: Data are listed as n (%) except\*. [^15]: Nonunion due to avascular necrosis of the femoral head. [^16]: nonunion due to deep infection. [^17]: Variables entered in Logistic regression analysis. *p* \< 0.05 means significant difference. [^18]: ASA score = American Society of Anesthesiologists score, TAD^diff^ = Tip Apex Distance difference of TAD^max^ (maximal measured TAD distance over time) minus the TAD (first post-operative measured TAD). [^19]: Unstable fracture types: A2.2, A2.3 A3.3 (AO-classification). AP = anteroposterior view, Ax = axial view.

1. Introduction {#sec1-genes-10-01021} =============== Consumption of caffeine, an adenosine receptor antagonist, is the most widely consumed stimulant in the world \[[@B1-genes-10-01021],[@B2-genes-10-01021]\]. It is found in a variety of foods and beverages, such as coffee, tea, chocolate, and energy drinks; coffee being the primary dietary caffeine source in Western Europe and the United States. It is used to enhance performance by athletes and as a countermeasure against fatigue in shift workers, airline pilots, and truck drivers \[[@B3-genes-10-01021]\]. Caffeine attenuates waking and sleep electroencephalographic markers of sleep homeostasis in humans \[[@B4-genes-10-01021]\]. It is generally accepted that caffeine promotes wakefulness by unselectively antagonizing adenosine receptors in the brain \[[@B2-genes-10-01021],[@B5-genes-10-01021]\]. Adenosine is a metabolic intermediate of the energy-rich molecule adenosine-tri-phosphate (ATP), acting as an endogenous homeostatic sleep factor, with increased levels in the basal forebrain during prolonged wakefulness and resolves during sleep \[[@B6-genes-10-01021]\]. The sleep-inducing effects of extracellular adenosine are mediated mainly through the inhibitory G protein-coupled adenosine A1 receptor and the excitatory G protein-coupled adenosine A2A receptor (A2AR) that are ubiquitously distributed throughout the brain \[[@B5-genes-10-01021]\]. Accumulating evidence suggests that among these receptors responsible for sleep induction, the role of A2A is predominant in sleep regulation, whereas A1 contributes to sleep induction in a region-dependent manner but may not be absolutely necessary for sleep homeostasis \[[@B7-genes-10-01021]\]. In our previous study, we described an upregulation of A2AR expression in human leukocytes from healthy young subjects after 24 h of sleep deprivation while A1R was not changed \[[@B8-genes-10-01021]\]. The metabolism, clearance, and pharmacokinetics of caffeine is affected by many factors (age, smoking, diet, medications, sex and hormones, liver disease, obesity) and also by genetic variability affecting caffeine levels and effects at the pharmacokinetic and pharmacodynamic levels that both critically drive the level of caffeine consumption \[[@B9-genes-10-01021]\]. Both at pharmacodynamic and consumption levels, and those of tolerance to sleep deprivation and side effects, several polymorphisms of the ADORA2A gene, the major target of caffeine action in the brain, have been shown to be implicated \[[@B10-genes-10-01021],[@B11-genes-10-01021],[@B12-genes-10-01021],[@B13-genes-10-01021],[@B14-genes-10-01021]\]. The ADORA2A locus is located on human chromosome 22 and contains a set of SNPs (single nucleotide polymorphism), the most studied being rs5751876 (formally designated as 1976C/T or 1083C/T). Other polymorphisms such as rs2298383 and rs3761422 are in high linkage disequilibrium with the rs5751876 \[[@B12-genes-10-01021]\]. The 1976T\>C polymorphism of ADORA2A modulates the EEG activity of the wake--sleep continuum \[[@B15-genes-10-01021]\], contributes to individual sensitivity to caffeine effects on sleep \[[@B11-genes-10-01021]\] and greater sensitivity to caffeine-induced anxiety \[[@B12-genes-10-01021]\]. Recently, an association was found between caffeine consumption and objective sleep variables in the T allele genotype carriers and not in the CC genotype of the ADORA2A rs5751876 single nucleotide polymorphism (SNP) \[[@B16-genes-10-01021]\]. Furthermore, among several ADORA2A polymorphisms, rs5751876 and several other variants in high LD (linkage disequilibrium) (rs5751862, rs2298383, and rs3761422), as well as the corresponding haplotypes, were found associated with anxiety, and of these variants, rs2298383 was shown to possess functional potential and might, therefore, represent the true underlying causal variant \[[@B17-genes-10-01021]\]. In a second study, these authors showed that the rs5751876 SNP is highly correlated with in vivo adenosine A1 receptor binding in the human brain \[[@B18-genes-10-01021]\]. Our aim is to assess the influence of six ADORA2A SNPs on self-reported sleep characteristics and caffeine consumption in a French active worker population (*n* = 1023 participants). The six ADORA2A SNPs, selected because of their involvement in caffeine consumption, sensitivity to caffeine effects on sleep, sleep disorders and anxiety in literature, were rs5751876, rs2298383, rs3761422, rs5751862, rs2236624, and rs4822492. The principal conclusions are that: (1) in low caffeine consumers (less than 300 mg per day) a combination of ADORA2A polymorphisms influences TST (total sleep time) and the risk of sleep complaints and insomnia, and (2) at caffeine daily consumption higher than 300 mg/day, total sleep time (TST) decreases and prevalence of insomnia and sleep complaints increases, whatever the ADORA2A polymorphism. This opens perspectives on the diagnosis and pharmacology of sleep complaints and caffeine chronic consumption. 2. Results {#sec2-genes-10-01021} ========== 2.1. Subjects {#sec2dot1-genes-10-01021} ------------- The questionnaire was completed by 1083 participants of European ancestry. We excluded 60 participants, of which 22 provided a saliva sample that was not usable, one did not sign the informed consent, 34 had at least one missing response on the questionnaire, and 3 presented an exclusion criteria. Finally, a total of 1023 questionnaires (618 men and 405 women) were analyzed. 2.2. Sociodemographic Data and Lifestyle Habits {#sec2dot2-genes-10-01021} ----------------------------------------------- The participants were aged between 18 and 60 years (32.5 ± 9.6) and were divided into 60.4% (*N* = 618) male and 39.6% (*N* = 405) female ([Table 1](#genes-10-01021-t001){ref-type="table"}). Most of the participants (43.3%) were single, and 56.6% were married or living with a partner. About 37.8% of them had children. The mean BMI was 23.6 ± 3.5 kg/m^2^; 56 (5.5%) were obese and 225 (22.0%) were overweight. About 192 (18.8%) of the participants are current smokers and 81.0% have never been smokers. About 76 (7.4%) of the participants consumed alcohol and 671 (65.6%) exercised more than 2 h per week. 129 (12.6%) regularly took pharmaceuticals treatments. The most frequent medications were contraceptive (26.5%), levothyrox (7.5%), blood pressure (7.5%), allergy (7.4%), asthma (7.5%) and proton pump inhibitor (PPI, 6.5%) treatments. The use of sedatives concerned only 9 participants (i.e., 0.9%). The mean daily caffeine consumption was 243 ± 208 (SD) mg/d. Two hundred and two (19.7%) pertain to the low (0 to 50 mg/day) caffeine consumer group, 478 (46.7%) to the moderate (51 to 300 mg/day), and 343 (33.5%) to the high (\>300 mg/day) caffeine consumers. Age varied among the groups, with younger subjects in low compared to high caffeine consumers (28.7 ± 8.7 years vs. 35.9 ± 9.1 years). Smokers were overrepresented (*p* \< 0.01) in the high (30.8%) and moderate (12.1%) caffeine consumption groups compared to the low caffeine group (7%). In the moderate and the high caffeine consumer groups, 5.5% and 19.0% of smokers consumed more than 5 cigarettes per day, respectively. Of the 1023 participants, 46.8% reported sleep complaints and 10.7% insomnia. 2.3. Sleep Duration, Sleep Complaints and Insomnia According to Caffeine Consumption {#sec2dot3-genes-10-01021} ------------------------------------------------------------------------------------ The self-reported nocturnal total sleep time (TST) significantly decreased with the increase of caffeine consumption, whatever the genotype. The ANOVA analysis showed that TST was significantly lower in high consumers compared with low and moderate consumers (*F* = 13.85, *p* \< 0.01; [Table 2](#genes-10-01021-t002){ref-type="table"}). Sleep complaints were significantly higher (ORa = 1.9 \[1.1--3.3\]) in high caffeine consumers compared to low and moderate consumers, and insomnia was higher in high compared to low consumers (ORa = 1.5 \[1.1--1.9\]). The percentage of participants reporting caffeine side effects significantly increased with the daily caffeine consumption (*p* \< 0.01), with 50% reporting at least one side effect when caffeine consumption is higher than 300 mg. The diuretic effect of caffeine is prevalent when caffeine consumption is higher than 300 mg. 2.4. Genotype Prevalence of the Six ADORA2A Polymorphisms {#sec2dot4-genes-10-01021} --------------------------------------------------------- The genotype prevalence of the six ADORA2A SNPs in our study is similar to the 1000 Genomes Project data on the GRCh38 reference assembly ([Table 3](#genes-10-01021-t003){ref-type="table"}). Genotype frequencies of all SNPs conformed to the Hardy--Weinberg equilibrium (*p* \> 0.16 for all). The pairwise linkage disequilibrium analysis showed moderate to strong LD between ADORA2A variants ([Figure 1](#genes-10-01021-f001){ref-type="fig"}B). We observed a *R*-square value \>0.8 between rs2298383, rs3761422, rs5751876, and rs4822492. 2.5. Impact of the Six ADORA2A SNPs on Total Sleep Time (TST) According to Caffeine Consumption {#sec2dot5-genes-10-01021} ----------------------------------------------------------------------------------------------- The two-way ANOVA analysis showed no significant effect of genotype but a significant effect of caffeine consumption on TST for the six ADORA2A SNPs, with significant interaction between genotype and caffeine consumption for 4 SNPs ([Table 4](#genes-10-01021-t004){ref-type="table"} and [Table 5](#genes-10-01021-t005){ref-type="table"}). Post hoc analysis showed that genotype is associated with a significant difference in TST only for low caffeine consumers (\<50 mg per day, i.e., less than one expresso coffee per day). In low caffeine consumers, TST level was significantly lower in the T/C and T/T carriers compared to C/C (ancestral) carriers of two ADORA2A SNPs (rs5751876 and rs3761422) while TST level was higher in the T/T (vs. C/C) carriers of the rs2298383 and the G (G/C and G/G) compared to C/C carriers of the rs4822492 ([Table 5](#genes-10-01021-t005){ref-type="table"}). 2.6. Impact of ADORA2A Haplotypes on Total Sleep Time (TST) According to Caffeine Consumption {#sec2dot6-genes-10-01021} --------------------------------------------------------------------------------------------- Six (1 to 6) different haplotypes (ATCCCG, GCTTTC, GCTCTC, GTCCCG, GCCCTC, GCCCCC) and one Rare were found, the higher frequency being 45.3% for haplotype 1 ([Table 6](#genes-10-01021-t006){ref-type="table"}). The haplotypes 2 (21.2%) and 3 (15.7%) in our study respectively corresponded to the haplotypes HT1 and HT4 in the Bodenmann et al. (2012) study \[[@B14-genes-10-01021]\]. In the Bodenmann study, HT4 haplotype carriers consistently performed at a higher level on a PVT task during sleep deprivation and presented less rebound of SWA (slow-wave activity) in recovery sleep than non-HT4 carriers. The two-way (caffeine and haplotype) ANOVA analysis showed no significant effect of haplotype (*p* = 0.10; *F* = 1.42) but a significant effect of caffeine consumption (*p* = 0.02; *F* = 4.51) on daily TST, with a significant interaction (*p* = 0.03; *F* = 2.42). All haplotypes are associated with a significant effect of caffeine consumption, with the lower TST level in high consumers in comparison with low consumers ([Table 7](#genes-10-01021-t007){ref-type="table"}). In addition, the TST was significantly lower in low caffeine consumers in the haplotypes 2 and 3 (GCTTTC and GCTCTC) compared with the most frequent haplotype 1 (ATCCCG) ([Table 7](#genes-10-01021-t007){ref-type="table"}, [Figure 2](#genes-10-01021-f002){ref-type="fig"}). In moderate compared with low caffeine consumers, TST was not significantly different in the haplotypes 2 and 3 (GCTTTC and GCTCTC), whereas it was significantly lower for four other haplotypes (1 and 4 to 6). 2.7. Impact of the Six ADORA2A SNPs on Sleep Complaints and Insomnia According to Caffeine Consumption {#sec2dot7-genes-10-01021} ------------------------------------------------------------------------------------------------------ The risk of sleep complaints was significantly lower in the C/T and T/T compared to C/C (ancestral) genotype carriers for rs5751876 in moderate caffeine consumers and all subjects. However, in moderate and in all subjects, the risk of sleep complaints was higher in T/T compared to C/C for rs2298383 and in G/G compared to C/C (ancestral) genotype carriers for rs4822492 in moderate and in all subjects ([Table 4](#genes-10-01021-t004){ref-type="table"} and [Table 8](#genes-10-01021-t008){ref-type="table"}). The results on the impact of six ADORA2A SNPs on reported insomnia, according to the three groups of caffeine consumers, are not shown because less than 10% of subjects were concerned and the numbers of subjects carrying each of the genotype SNPs in the different caffeine groups were low. However, if we consider all subjects, the risk of insomnia is significantly higher for the T/T and G/G compared to C/C ancestral genotype carriers of rs2298383 and rs4822492, respectively ([Table 4](#genes-10-01021-t004){ref-type="table"}). 3. Discussion {#sec3-genes-10-01021} ============= In the present study, we investigate the effect of six genetic polymorphisms of ADORA2A on sleep (duration and disorders) and the association with caffeine consumption groups (low, moderate, and high consumers) in 1023 French active workers aged between 18 and 60 years. We selected polymorphisms that were associated previously with the interindividual variability in brain electrical activity during sleep and wakefulness \[[@B14-genes-10-01021],[@B15-genes-10-01021]\], with the individual sensitivity to subjective and objective caffeine effects on sleep \[[@B11-genes-10-01021],[@B19-genes-10-01021]\], anxiogenic response to caffeine \[[@B12-genes-10-01021],[@B20-genes-10-01021]\], with caffeine consumption \[[@B10-genes-10-01021]\], with individual differences in anxiety-related personality and panic disorder \[[@B17-genes-10-01021]\], and sleep disturbances \[[@B19-genes-10-01021],[@B21-genes-10-01021]\]. We delineate three groups of caffeine consumers according to several criteria. There is a consensus that the daily ingestion of 300--400 mg caffeine (around 4--5 cups of expresso coffee) does not raise any health concern, and that high doses of caffeine (400--800 mg in one sitting) may have negative effects (anxiety, nervousness, insomnia, tachycardia, and trembling) \[[@B22-genes-10-01021]\]. Moreover, 300 mg is also the dose of caffeine contained in the pharmacological form of the slow-release caffeine that was shown to be an advisable alternative to other psychostimulants for long work schedules, and night shifts \[[@B23-genes-10-01021]\]. The low caffeine consumers (0--50 mg/day) were in accordance with Rogers et al. (2010) \[[@B12-genes-10-01021]\]. Our study focuses on healthy and active workers of European ancestry, excluding retired and medical workers, who are neither overweight nor obese according to WHO criteria, who practice physical activity for the majority, consume little alcohol, with very few smokers. As the genotype frequency analysis shows, our population is very similar to the 1000 Genomes project population \[[@B24-genes-10-01021]\]. The significant impact of genetic variants on TST is only shown in low caffeine consumers (0--50 mg, corresponding to less than one expresso coffee per day). Several genetic mutations are associated with lower or higher TST. Individuals carrying T allele (homozygous or heterozygous) of rs5751876 and rs3761422 exhibit a significant lower TST whereas T allele of rs2298383 is associated with higher TST when compared to homozygous C ancestral allele carriers. In addition, TST is higher for the G allele (homozygous or heterozygous) carriers of rs4822492 compared to the ancestral homozygote C/C. The haplotype analysis revealed a significant lower TST for two haplotypes (2 and 3) combining T allele carriers of rs5751876 and rs3761422, and C allele carriers of rs2298383 and rs4822492, in comparison to the more frequent haplotype 1 (45.3%). Furthermore, TST was significantly lower in high caffeine consumers (\>300 mg per day) compared to low and moderate, regardless of the selected ADORA2A polymorphism. Secondly, our results showed that the risk of sleep complaints and insomnia is higher in participants carrying mutations on rs2298383 and rs4822492 (after adjustment for age, tobacco, and sex) in moderate and all caffeine consumers, but not in high consumers. There were significantly more subjects with sleep complaints in the high caffeine group (52.5%) than in the two other groups, and more subjects with insomnia comparatively with the low caffeine group (12.5 % vs. 7.4%). Therefore, biological follow-up of genetic variants of the G-protein coupled A~2A~ receptor can provide additional insights into molecular processes underlying sleep duration and disorders, valuable for enhancing scientific knowledge of sleep and for informing new therapeutic approaches. Subsequent pharmacological studies from several laboratories have demonstrated that adenosine and its receptor agonists promote, but antagonists such as caffeine inhibit, both NREM and REM sleep stages \[[@B25-genes-10-01021]\]. There are four G protein-coupled adenosine receptor subtypes, A~1~, A~2A~, A~2B~, and A~3~ \[[@B26-genes-10-01021]\], and accumulated findings indicate a predominant role of A2AR in sleep regulation and caffeine-induced wakefulness \[[@B7-genes-10-01021],[@B27-genes-10-01021]\]. We previously showed that one night of sleep deprivation induced an up-regulation of the A2AR gene expression in leukocytes from healthy subjects \[[@B28-genes-10-01021]\]. The genetic variation of ADORA2A was found to be an important determinant of wake promotion by caffeine, particularly the rs5751876 polymorphism \[[@B11-genes-10-01021],[@B29-genes-10-01021]\]. Using polysomnography EEG in healthy males in the recovery night after sleep deprivation, Retey et al. \[[@B11-genes-10-01021]\] observed that caffeine increases spectral power in the beta band (16.625--20.125 Hz) in C/C compared to T/T genotype carriers of the ADORA2A c.1083T/C genotype (also designated 1976C/T or rs5751876 now). As in our study, the C/C genotype for rs5751876 seems associated with an increased likelihood of being sensitive to caffeine and an increased likelihood of insomnia when exposed to caffeine. Moreover, caffeine was shown to attenuate the rebound of SWA (slow-wave activity) occurring in the recovery night following sleep deprivation in subjects with non-HT4 haplotype ADORA2A only (this does not occur in HT4) \[[@B14-genes-10-01021]\]. Our results, in particular, the haplotype analysis, confirm that caffeine sensitivity is associated with several SNP mutations. We showed that the haplotypes carriers which are insensitive to caffeine effects on sleep were T allele carriers for rs5751876 and rs3761422, while being C carriers for rs4822492 and rs2298383. This profile is similar to the HT4 profile described by Bodenmann et al. \[[@B14-genes-10-01021]\] and confirms that T allele carriers of rs5751876 seem insensitive to caffeine effects on sleep duration \[[@B11-genes-10-01021],[@B14-genes-10-01021]\]. However, Nunes et al. 2017, with a polysomnographic study, failed to demonstrate an association between ADORA2A genotypes and sleep variables, as well as the lack of stronger correlations between caffeine load and other sleep variables \[[@B16-genes-10-01021]\]. In our study, we can suggest that once the C/C genotype carriers of rs5751876 and rs3761422 have taken caffeine, their sleep time will match that of T/T and C/T. Also, when G allele carriers for rs4822492 have taken caffeine, their TST will match the one of C/C, which pertains to the HT4 haplotype of Bodenmann et al. (2012) \[[@B14-genes-10-01021]\]. The rs5751876 SNP is a synonymous variant (it does not cause an amino acid change in the encoded protein) located on the exon 4 position, while rs2298383 and rs3761422 are located in the 5'UTR, and rs4822492 in 3'UTR, respectively. We evidenced that both rs2298383, rs3761422, and rs4822492 are in strong linkage disequilibrium with the most studied rs5751876 SNP. Thus, these SNPs may represent potential functional variants \[[@B30-genes-10-01021],[@B31-genes-10-01021]\] implicated in the relationship between sleep duration and caffeine consumption. Alternative splicing of the 5'UTR could play a crucial role in the post-transcriptional regulation of G protein-coupled receptors, including the modulation of translational efficiency, message stability, and subcellular localization, and differential expression of 5'UTR splice variants of the adenosine A~2A~ receptor gene was shown in vitro in human stimulated granulocytes in sepsis patients compared to healthy volunteers \[[@B32-genes-10-01021]\]. The 3'UTR region contains binding sites for regulatory proteins and miRNA which can down-regulate post-transcriptional genes expression by inhibiting protein translation or by destabilizing target transcripts. A second study of Kreth et al. \[[@B33-genes-10-01021]\] team has described the regulation of A2AR mRNA translation in human polymorphonuclear leukocytes by three miRNA. Barett et al. (2012) \[[@B31-genes-10-01021]\] reviewed the different mechanism by which non-coding regions, including the 5' and 3' UTR, introns, and intergenic regions, are vital for the precise regulation of gene expression. In our study, TST was significantly lower in participants with the haplotypes 2 and 3 compared with the more frequent protective haplotype 1 (ATCCCG, 45.3% of participants). Haplotype 3, which corresponds to the HT4 haplotype of Bodenmann et al. (2012) \[[@B14-genes-10-01021]\], confirmed the SNPs analysis for rs5751876 and rs3761422 (T variant sleeping less) and for rs4822492 and rs2298383 (C variant sleeping less). The alleles for rs5751876, rs3761422, rs2298383, and rs4822492 in haplotype 1 are in higher frequency compared to the ones in haplotype 2 and 3. In this study, we evidenced significant differences between total sleep time (TST) among variants of four SNPs of ADORA2A and two haplotypes but in low caffeine consumers only, and three of them have been previously demonstrated as risk variants for anxious personality (i.e., rs5751876, rs2298383, and rs3761422, in high LD) \[[@B17-genes-10-01021]\]. In this study, after Bonferroni adjustment for age, sex, and caffeine consumption, the risk haplotype for the anxiety-related personality score (HA1 subdimension, anticipatory worry, and pessimism vs. uninhibited optimism) included alleles rs5751876-T, rs3761422-T, and rs2298383-C. We found two similar haplotypes (2 and 3) associated with lower TST in comparison with the more frequent protective haplotype 1 which suggested that lower TST in low caffeine consumers may be related to anxiety-related personalities. Moreover, in our study, the risk of sleep complaints was significantly higher for participants carrying the T/T and G/G compared with C/C ancestral genotype for rs2298383 and rs4822492, respectively, and this was observed when considering all subjects and also in moderate caffeine consumers. A recent cross-sectional population-based study on the relationship between depression and symptoms and the rs2298383 SNP for ADORA2A (1253 participants aged 18--35 years, including 228 in current episode of depression) demonstrated that the TT genotype of rs2298383 (compared to CC plus CT) is associated with reduced risk depression and reduced sleep disturbances and less difficulty in concentrating after adjustment for confounding variables such as smoking, gender, socioeconomic class, and ethnicity \[[@B21-genes-10-01021]\]. This result differs from ours likely because we estimated sleep complaints risk by comparison with the rs2298383 ancestral C/C genotype, as previously described by Hohoff et al. (2010) \[[@B17-genes-10-01021]\]. Furthermore, our results are consistent regarding genotype frequencies for the six SNPs with those of the 1000 Genomes Project \[[@B24-genes-10-01021]\] and are consistent with the results on the risk of insomnia. Indeed, the risk of insomnia is also significantly higher for the T/T compared to the ancestral C/C genotype for rs2298383 and also for G/G carriers of rs4822492, when considering all subjects. In our study, it was difficult to analyze the impact of ADORA2A SNPs on insomnia on the three groups of caffeine consumers because only 10.7% of subjects were identified with insomnia. To our knowledge, the rs4822498 ADORA2A has been identified only once in a genome-wide study on the association between caffeine-induced sleep disturbance and several genetic variants, and the result was inconclusive \[[@B19-genes-10-01021]\]. With respect to sleep complaints, our results showed no significant difference in sleep complaints prevalence between genotypes in the six ADORA2A SNPs among high caffeine consumers, who are more than half of those affected (52.5%). This result suggests that other genetic mutations may be implicated in the high caffeine dose effect on sleep complaints \[[@B34-genes-10-01021]\], such as SNPs for ADORA1A. However, multiple variants of the ADORA1A were shown not to be implicated in caffeine-related sleep disturbances \[[@B35-genes-10-01021]\] nor caffeine-related anxiety (Rogers et al., 2010), and not associated with the in vivo variation of A1AR availability in the human brain compared while ADORA2A rs5751876 was \[[@B18-genes-10-01021]\]. For us, genetic variants for the cytochrome P450 1A~2~ (CYP1A2), the main system responsible for caffeine metabolism, and variants for pro-inflammatory cytokines should be investigated. Indeed, mutations on the CYP1A2 gene were recently shown to influence abstract reasoning in a large population (1374 participants) under free caffeine intake in everyday life \[[@B36-genes-10-01021]\], and genetic variations of inflammatory cytokines IL-6 and TNF-α have been associated in several studies with sleep disturbances \[[@B37-genes-10-01021]\] and susceptibility and severity of obstructive sleep apnea \[[@B38-genes-10-01021],[@B39-genes-10-01021]\]. Moreover, several studies have also shown that T/T genotype carriers for rs5751876 of ADORA2A are associated with higher anxiogenic responses to caffeine \[[@B12-genes-10-01021],[@B20-genes-10-01021]\]. The differential response of T/T vs. C/C and C/T genotypes was apparent at 100 mg dose of caffeine, and similar results were found for the nearly completely linked SNP rs3761422 and rs2298383 (but not significant after adjustment for multiple testing for the latter) \[[@B12-genes-10-01021]\]. For Childs et al. (2008) \[[@B20-genes-10-01021]\], individuals with the C/C genotype for rs2298383 and rs4822492 reported significantly higher anxiety than those with the T/T genotype and G/G, respectively. We can suggest that as soon as caffeine is consumed, anxiety may be implicated in the relationship with sleep duration, particularly in participants carrying the T allele (homozygote and heterozygote) for rs5751876, rs3761422, and rs2298383, and being in the lower frequency C/C genotype for rs4822492. However, our study also has several limitations that need to be addressed in future studies. We have relied on self-report data for all of the variables analyzed, which can be influenced by perception bias \[[@B40-genes-10-01021]\]. To this end, it will be important to investigate the impact of ADORA2A SNPs on sleep duration using polysomnography or actigraphy to get an objective measure of sleep. These methods provide greater accuracy of measurement, but they are much more time-consuming and less conducive to our large sample sizes study. However, we confirmed the association between sleep and rs5751876 that was initially discovered in a sample that used polysomnography \[[@B15-genes-10-01021]\], which suggests that self-report data, while imperfect, is a somewhat reliable measure of the true phenotype and may be of particular use in large-scale genetic studies. 4. Materials and Methods {#sec4-genes-10-01021} ======================== 4.1. Study Design and Participants {#sec4dot1-genes-10-01021} ---------------------------------- This is a cross-sectional population-based study of active French workers (men and women) aged 18 to 60. The sample consisted of 1023 participants recruited from Hotel Dieu sleep center (APHP), the medical department of EY (First Tower, La defense) and the armed forces biomedical research institute (IRBA). The study received the agreement of the Cochin -- CPP Ile de France 1 (Paris) Ethics Committee and was approved by the Agence Nationale pour la Sécurité du Médicament (ANSM) (2017-A00234_49). It was conducted according to the principles expressed in the Declaration of Helsinki of 1975, as revised in 2001 after obtaining written informed consent for all the participants. The study has been registered on the [ClinicalTrials.gov](ClinicalTrials.gov) public database (NCT number: NCT03855774). Exclusion criteria were unemployed, retired, medical leave, and pregnant women volunteers, lack of social insurance, and under-18 years old subjects. 4.2. Questionnaire {#sec4dot2-genes-10-01021} ------------------ The participants answered a questionnaire for sociodemographic characteristics, assessment of sleep complaints and disorders, and caffeine intake. All data were obtained by means of a paper questionnaire and done at the same time as the saliva sampling. The questionnaire was anonymous, and participants have to refer to the last month to evaluate sleep and wake disorders. The questionnaires had different types of questioning: open-ended, closed-ended, and multiple-choice questions. ### 4.2.1. Sleep Habits and Complaints {#sec4dot2dot1-genes-10-01021} For sleep evaluation, the questionnaire was constructed based on the Sleep Complaints Questionnaire--French version (SDQFV) \[[@B41-genes-10-01021]\], and the Epworth Sleepiness Scale (ESS) \[[@B42-genes-10-01021]\]. In addition, sleeping pills, stimulated substances, and other drug consumption were informed in the questionnaire. The SDQFV is a 42-item questionnaire based on the Stanford Sleep Questionnaire and Evaluation of Wakefulness (SQAW). The French version has been validated in several epidemiological studies \[[@B41-genes-10-01021]\]. It covers sleep habits and complaints. We focused on four specific sleep complaints which are commonly associated with insomnia: (1) difficulties in falling asleep, (2) frequent nocturnal awakenings, (3) early awakenings, and (4) nonrestorative sleep. The ESS, a self-administered questionnaire, is a subjective tool to assess sleepiness \[[@B42-genes-10-01021]\]. To assess sleepiness, we also introduced several commonly associated items: (1) napping habits; (2) taking more than five cups of coffee every day; (3) snoring loudly; (4) feeling sleep deprived. ### 4.2.2. Sleep Complaints {#sec4dot2dot2-genes-10-01021} Their definition was based on the International Classification of Sleep complaints (ICSD, 3rd edition) and the Diagnostic and Statistical Manual of Mental Disorders, 4th revision (DSM-IV) \[[@B43-genes-10-01021]\]. Insomnia was defined by the presence of at least one of the four sleep complaints over at least one month and with daytime consequences. Hypersomnia was measured on a daily basis according to whether patients "fall asleep during the day, during work, while listening to the radio or music, while traveling, in front of the TV" and scored \>10 in the ESS. Severe insomnia, regular use of sedatives, and sleep apnea were also considered. ### 4.2.3. Caffeine Consumption {#sec4dot2dot3-genes-10-01021} The questionnaire included the following beverages and caffeine-containing foods: coffee with caffeine, tea, cola, and other carbonated beverages with caffeine, and chocolate. For each item, participants were asked how often, on average, they had consumed a specified amount of each beverage or food over the past year. The participants could choose from nine frequency categories (never, 1--3 per month, 1 per week, 2--4 per week, 5--6 per week, 1 per day, 2--3 per day, 4--5 per day, and 6 or more per day). Typical milligram doses (Mayo Clinic--<http://www.mayoclinic.com/health/caffeine/AN01211>) were assigned to each and an approximate daily intake was obtained \[[@B44-genes-10-01021]\]. Based on previous criteria \[[@B12-genes-10-01021],[@B22-genes-10-01021]\], participants were then characterized as having low (0--50 mg·day^−1^), moderate (51--300 mg·day^−1^), and high (\>300 mg·day^−1^) caffeine intake. 4.3. Saliva DNA Extract and Genotyping {#sec4dot3-genes-10-01021} -------------------------------------- Genotypes were determined by investigators who were blind for trait anxiety, sleep and waking EEG, subjective sleepiness, and behavioral results. Oragene DNA kits OG-500 (DNAgenotek, Ottawa, Canada) were used to collect whole saliva samples from healthy adult volunteers (*n* = 1069) after rinsing the mouth with water and at least 30 min after eating or drinking. DNA from saliva collected in Oragene containers are stable for at least five years at ambient temperature. After manual cell lysate preparation, the genomic DNA purification was performed by the Autopure LS instrument (Qiagen, Hilden, Germany). DNA quantity, integrity, and ability to PCR were evaluated by quality controls. One thousand and twenty-three DNA extracts were transferred on 384-well plates. SNP selection was based on previous studies. Participants were genotyped by predesigned or customized probes; TaqMan SNP genotyping assays were provided by Thermo Fisher Scientific (Whaltham, USA). PCR was performed on GeneAmp PCR System 9700, and a 7900HT system with SDS software version 2.4 (Applied Biosystems, Foster City, USA) was used for fluorescence detection and allelic discrimination. A check by Sanger sequencing of PCR products was carried out for rs5751876 because the ADORA2A gene includes many SNPs in this region. The genomic organization of the ADORA2A gene with SNP positions investigated in the 1023 participants is illustrated in [Figure 1](#genes-10-01021-f001){ref-type="fig"}A. 4.4. Statistical Analysis {#sec4dot4-genes-10-01021} ------------------------- Statistical tests were performed using R studio (R software (version 3.6.0, 04.24.2019) ([www.r-project.org](www.r-project.org)), and significance (α risk) was fixed at *p* \< 0.05. Qualitative variables are presented as occurrence and percentage \[*n* (%)\] and quantitative variables as mean ± 95% CI (confidence interval). The genotype distributions in both alleles were examined using the Hardy--Weinberg equilibrium. The difference in TST was calculated using a 2-way analysis of variance (SNP alleles and caffeine consumption). If a significant interaction was observed, a post-hoc Bonferroni test was used to compare the allele (heterozygote and homozygote mutation) versus the ancestral allele and caffeine consumption (moderate and high versus low) for the 6 SNPs. Qualitative variables (sleep complaints) association with SNP alleles were made using a X^2^ test. Odds ratio adjusted for age, sex, and tobacco use and the 95% confidence intervals (95% CI) were calculated. The SNPStats program \[[@B45-genes-10-01021]\] was adopted for linkage disequilibrium (D, D, and R square) haplotype analysis. All haplotypes identified were computed, but in order to minimize loss of power, the six haplotypes with a frequency higher than 1% were taken into account for statistical analysis. The most common haplotype was selected as the reference for posthoc analysis. 5. Conclusions {#sec5-genes-10-01021} ============== In conclusion, our results showed ADORA2A genetic variations related to the duration of nocturnal sleep in a large population of European ancestry aged 18 to 60 years but only in low caffeine consumers (below 50 mg per day, i.e., less than one expresso). As soon as participants became moderate caffeine consumers (51--300 mg per day), the nocturnal sleep decreases whatever the SNPs ADORA2A variants. At caffeine consumption above 300 mg per day, the six genetic variants of ADORA2A studied here do not influence the total sleep time which remains significantly lower compared to low caffeine consumption. These results suggested that caffeine consumption above 300 mg per day may have completely blocked A2A receptors. The herein presented results confirmed previous results on the synonymous rs5751876 variant, and added information on several potential functional variants such as rs3761422, rs2298383 and rs4822492. On the other hand, ADORA2A genetic variations influenced the risk of sleep complaints and insomnia in all caffeine consumers, notably the rs2298383, rs4822492 and rs5751876. Indeed, the risk of sleep complaints is higher in rs2298383 TT and rs4822492 GG compared to the ancestral CC genotypes, and lower in rs5751876 CT and TT ([Table 4](#genes-10-01021-t004){ref-type="table"}), and this remained significant in the moderate caffeine consumers. We planned in our future studies to analyze A2A receptors density, function or gene expression in human blood in relation with the three genotypes of rs2298383, rs4822492 and rs5751876 ADORA2A variants. It will be interesting in future studies to investigate in a large population of European ancestry sleep duration and risk of sleep complaints and disorders concomitantly with multiple candidate SNP including dopamine- and cytokines-related genes \[[@B39-genes-10-01021],[@B46-genes-10-01021]\], and genes involved in the pharmacokinetic and pharmacodynamic of caffeine effects \[[@B35-genes-10-01021],[@B36-genes-10-01021]\]. Also, analysis of changes in A2A receptors density, function or gene expression in human granulocytes under low and high doses of caffeine consumption will aid in understanding the relationship between genetic mutations and sleep complaints and disorders, and will have implications for the treatment of sleep complaints \[[@B47-genes-10-01021]\]. We thank Robert Olaso, Céline Derbois and Bertrand Fin from the CNRGH (Centre National de Recherche en Génomique Humaine, CEA, Institut François Jacob, Evry, France) for their technical support and expertise. We thank Stéphanie Bérot from the Medical service of EY (Tour First, La Défense, 92037 Paris - France). We thank Anne Sophie Pellen and Catherine Verret from sponsor clinical research unit for their methodological and legacy considerations of the protocol. We thank all participants of our study, and Laurent Rocco, André Pisani, Michel Génot, Marc Pernot of the CNSD (Centre National des Sports de la Défense). All authors contributed significantly to this work: Conceptualization, M.C., F.S., D.G.-M., C.D., M.E., A.M., A.B. and J.F.D.; Methodology, D.G.-M., F.S.; Software, F.S.; Validation, C.D., D.G.-M. and M.C.; Formal Analysis, M.E., C.D. and A.B.; Investigation, M.E., C.D., A.M., F.S.; Resources, F.S. and M.C.; Data Curation, F.S. and M.E.; Writing---Original Draft Preparation, D.G.-M., C.D. and M.E.; Writing---Review & Editing, D.G.-M., F.S. and M.C.; Visualization, D.G.-M., M.C., A.B., J.F.D.; Supervision, M.C., C.T. and J.F.D.; Project Administration, F.S.; Funding Acquisition, F.S. This research has been funded in operating cost by the DGA (General Directorate for Armament, Ministry of the Armies) (contract PDH-1-SMO-2-509). The authors declare no conflict of interest {#genes-10-01021-f001} {#genes-10-01021-f002} genes-10-01021-t001_Table 1 ###### The sociodemographic data, lifestyle habits, sleep duration, and sleep disorders of subjects. Women or Men 405 (39.6%) or 618 (60.4%) ---------------------------------------- ---------------------------- Age, years 32.5 ± 9.6 BMI, kg/m^2^ 23.6 ± 3.5 Alcool consumption (min 1 glass/day) 76 (7.4%) Current smoker (\>1 cigarette per day) 192 (18.8%) Physical activity (\>2 h/week) 671 (65.6%) Shift work (3 × 8 h) 19 (1.9%) Medication (\>5 years) 129 (12.6%) Caffeine consumption, mg/day 243 ± 208 Epworth sleepiness scale, score 7.7 ± 3.9 Total sleep time, h 7.0 ± 1.0 Sleep complaints (\>3 months) 479 (46.8%) Sleep disorders Insomnia 109 (10.7%) Severe insomnia 4 (0.4%) Sleep apnea 115 (11.2%) Use of sedative 9 (0.9%) Values are mean ± SD (quantitative variable) or occurrence (%) (qualitative variable), *N* = 1023. genes-10-01021-t002_Table 2 ###### Total sleep time (TST), sleep complaints, and insomnia, according to caffeine consumption. -------------------------------------------------------------------------------------------------------------------------------------------------- Low Caffeine Consumers\ Moderate Caffeine Consumers\ High Caffeine Consumers\ Statistical Analysis (0--50 mg/day) (51--300 mg/day) (\>300 mg/day) ----------------------------------- ------------------------- ------------------------------ -------------------------- -------------------------- *N* 202 (19.7%) 478 (46.7%) 343 (33.5%) TST(total sleep time), hours ± SD 7.15 ± 1.16 7.04 ± 0.89 6.75± 0.93 \*\# *F* = 13.9, *p* = 0.01 Sleep complaints\ 93 (46.0%) 206 (43.1%) 180 (52.5%) \*\# *X*^2^ = 7.1, *p* = 0.03 \>3 months, *n* (%) Insomnia, *n* (%) 15 (7.4%) 51 (10.7%) 43 (12.5%) \* *X*^2^ = 3.5, *p* = 0.05 -------------------------------------------------------------------------------------------------------------------------------------------------- *N*: number of subjects in the three groups with caffeine consumption. SD: standard deviation; \* *p* \< 0.01, vs. low caffeine consumers (0--50 mg/day); ^\#^ *p* \< 0.01 vs. moderate caffeine consumers (51--300 mg/day). genes-10-01021-t003_Table 3 ###### The six ADORA2A SNPs characteristics and genotypes frequencies. -------------------------------------------------------------------------- Genotype Frequency ----------- ---------------------- ------------ ---------- ------- ------- rs5751862 5' flank, regulation 22/\ G/G (a)\ 28.8\ 25\ 24,406,596 G/A\ 48.2\ 50.5\ A/A 21.9 24 rs2298383 5'UTR 22/\ C/C (a)\ 20.6\ 17.1\ 24,429,543 C/T\ 45.3\ 48.5\ T/T 32.4 34.4 rs3761422 5'UTR 22/\ C/C (a)\ 38.9\ 38.4\ 24,430,704 C/T\ 44.8\ 46.3\ T/T 14.6 15.3 rs2236624 Intron 3--4 22/\ C/C (a)\ 61.1\ 54.7\ 24,440,056 C/T\ 32.8\ 36.8\ T/T 5.1 8.5 rs5751876 Exon 4 22/\ C/C (a)\ 35\ 37.4\ 24,441,333 C/T\ 44.5\ 47.1\ T/T 18.1 15.5 rs4822492 3' UTR\ 22/\ C/C (a)\ 20.5\ 17.1\ (antisense) 24,447,626 C/G\ 45.1\ 48.5\ G/G 33 34.4 -------------------------------------------------------------------------- \(a\) ancestral genotype; *N* = 1023 in this study. genes-10-01021-t004_Table 4 ###### Impact of the six ADORA2A SNPs on total sleep time (TST), sleep complaints, and insomnia. SNP Genotype *N* Total Sleep Time (TST) Sleep Complaints Insomnia --------------- ------------- ----- ------------------------ ------------------ ------------------- ------------ ------------------- Mean ± 95% CI *n* (%) ORa (95% CI) *n* (%) ORa (95% CI) **rs5751862** **G/G (a)** 293 7.02 ± 0.12 124 (42.3%) 1 32 (10.9%) 1 G/A 493 6.98 ± 0.14 226 (46.0%) 1.2 (0.9--1.5) 46 (9.3%) 1.2 (0.73--2.13) A/A 224 6.89 ± 0.16 123 (54.7%) 1.6 (0.9--2.3) 30 (13.4%) 0.8 (0.52--1.34) **rs2298383** **C/C (a)** 210 6.99 ± 0.14 88 (41.9%) 1 23 (10.9%) 1 C/T 463 6.94 ± 0.12 207 (44.7%) 1.2 (0.6--1.4) 40 (8.6%) 0.8 (0.4--1.3) T/T 332 7.01 ± 0.14 174 (52.4%) 1.9 (1.2--3.1) \* 44 (13.3%) 1.5 (1.3--2.5) \* **rs3761422** **C/C (a)** 398 6.99 ± 0.10 200 (49.1%) 1 49 (12.0%) 1 C/T 458 6.91 ± 0.10 204 (44.5%) 0.7 (0.6--1.0) 38 (8.3%) 0.6 (0.4--0.9) \* T/T 149 7.08 ± 0.16 64 (43.0%) 0.8 (0.5--1.1) 19 (12.8%) 0.6 (0.6--1.7) **rs2236624** **C/C (a)** 625 6.97 ± 0.04 299 (47.8%) 1 67 (10.7%) 1 C/T 337 6.94 ± 0.05 153 (45.4%) 0.9 (0.7--1.2) 35 (10.4%) 0.9 (0.6--1.6) T/T 51 7.03 ± 0.14 21 (41.2%) 0.8 (0.4--1.4) 7 (13.7%) 1.4 (0.6--3.2) **rs5751876** **C/C (a)** 359 6.98 ± 0.05 185 (51.5%) 1 46 (12.8%) 1 C/T 454 6.93 ± 0.04 198 (43.6%) 0.7 (0.5--0.9) \* 36 (7.9%) 0.5 (0.3--0.9) \* T/T 182 6.99 ± 0.08 80 (43.2%) 0.7 (0.5--0.9) \* 21 (11.5%) 0.9 (0.5--1.5) **rs4822492** **C/C (a)** 210 6.98 ± 0.05 89 (42.4%) 1 23 (10.9%) 1 C/G 461 6.93 ± 0.04 207 (44.9%) 1.1 (0.5--1.2) 39 (8.4%) 1.3 (0.7--2.2) G/G 339 7.01 ± 0.07 178 (52.5%) 1.5 (1.1--2.1) \* 46 (13.6%) 1.9 (1.3--3.2) \* A = ancestral. \* *p* \< 0.05 significant difference vs. the ancestral genotype (C/C for rs5751876, rs2298383, rs3761422, and rs4822492). *N* is the total number of subjects with the alleles, *n* is the number of subjects with sleep complaints and insomnia for each genotype. ORa: adjusted odds ratio (age, tobacco, sex). genes-10-01021-t005_Table 5 ###### Impact of the six ADORA2A SNPs on total sleep time (TST) according to caffeine consumption. ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- SNP Genotype Low Caffeine\ Moderate Caffeine Consumers (51--300 mg/day) High Caffeine\ *ANOVA (F (p))* Consumers (0--50 mg/day) Consumers (\>300 mg/day) ----------- ---------- -------------------------- ---------------------------------------------- -------------------------- ----------------- ------------- ---------------- --------------------- ------------------ ------------------ rs5751862 G/G (a) 52 (17.8%) 7.23 ± 0.44 148 (50.5%) 7.09 ± 0.22 93 (31.7%) 6.95 ± 0.14 7.4 (*p* =\< 0.001) 0.2 (*p* = 0.78) 1.2 (*p* = 0.26) G/A 106 (21.5%) 7.28 ± 0.26 223 (45.1%) 7.10 ± 0.14 \# 164 (33.1%) 6.81 ± 0.12 \# A/A 39 (17.3%) 7.46 ± 0.44 102 (45.3%) 6.94 ± 0.18 83 (36.9%) 6.77 ± 0.18 \# rs2298383 C/C (a) 62 (18.7%) 6.93 ± 0.44 155 (46.8%) 7.10 ± 0.14 114 (34.4%) 6.96 ± 0.18 8.2 (*p* \< 0.001) 0.2 (*p* = 0.81) 1.9 (*p* = 0.03) C/T 92 (19.8%) 7.13 ± 0.44 198 (42.7%) 7.09 ± 0.14 172 (37.1%) 6.79 ± 0.12 \# T/T 44 (20.9 %) 7.52 ± 0.32 \* 113 (53.8%) 6.99 ± 0.16 \# 53 (25%) 6.88 ± 0.14 \# rs3761422 C/C (a) 79 (19.3%) 7.56 ± 0.28 187 (45.8%) 7.00 ± 0.16 \# 131 (32.1%) 6.84 ± 0.14 \# 7.1 (*p* =\< 0.001) 0.4 (*p* = 0.66) 1.2 (*p* = 0.04) C/T 89 (19.4%) 7.00 ± 0.32 \* 204 (44.5%) 7.06 ± 0.14 164 (35.8%) 6.79 ± 0.12 \# T/T 28 (18.8%) 6.85 ± 0.45 \* 76 (51.0%) 7.27 ± 0.32 45 (30.2%) 6.99 ± 0.10 \# rs2236624 C/C (a) 130 (20.8%) 7.30 ± 0.26 284 (45.4%) 7.00 ± 0.12 \# 210 (33.6%) 6.86 ± 0.10 \# 3.5 (*p* = 0.002) 0.3 (*p* = 0.73) 1.4 (*p* = 0.18) C/T 60 (19.6%) 7.17 ± 0.36 159 (47.2%) 7.21± 0.20 117 (34.7%) 6.74 ± 0.14 \# T/T 11(21.6%) 6.62 ± 0.80 27 (52.9%) 7.12 ± 0.32 13 (25.4%) 7.05 ± 0.28 \# rs5751876 C/C (a) 68 (18.9%) 7.53 ± 0.30 165 (46.0%) 6.99 ± 0.14 \# 125 (34.8%) 6.85 ± 0.14 \# 7.5 (*p* \< 0.001) 0.1 (*p* = 0.91) 1.8 (*p* = 0.04) C/T 91 (20.0%) 7.05 ± 0.32 \* 199 (36.5%) 7.09 ± 0.20 163 (35.9%) 6.80 ± 0.12 \# T/T 38 (20.9%) 6.92 ± 0.48 \* 98 (53.8%) 7.01 ± 0.28 49 (26.9%) 6.97 ± 0.20 rs4822492 C/C (a) 44 (21.0%) 6.93 ± 0.44 113 (53.8%) 6.99 ± 0.14 53 (25.2%) 6.86 ± 0.14 8.2 (*p* \< 0.001) 0.1 (*p* = 0.89) 1.8 (*p* = 0.04) C/G 91 (19.7%) 7.26 ± 0.32 \* 200 (43.3%) 7.09 ± 0.14 169 (36.7%) 6.79 ± 0.12 \# G/G 64 (18.9%) 7.52 ± 0.30\* 159 (46.9%) 7.10 ± 0.24 \# 115 (33.9%) 6.96 ± 0.18 \# ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- A = ancestral. \* *p* \< 0.05 significant difference vs. the ancestral genotype (C/C for rs5751876, rs2298383, rs3761422, and rs4822492). \# *p* \< 0.05 versus low caffeine consumers (0--50 mg/day). *N* is the total number of subjects with the alleles, *n* is the number of subjects with sleep complaints and insomnia for each genotype (% are calculated according to the number of subjects in each caffeine group; see [Table 5](#genes-10-01021-t005){ref-type="table"}). F~2,1021~ and F~2,1019~ are results of the 2-way ANOVA (caffeine consumption groups X SNP alleles) analysis. genes-10-01021-t006_Table 6 ###### The ADORA2A haplotypes frequencies (*n* = 1023 participants). rs5751862 rs2298383 rs3761422 rs2236624 rs5751876 rs4822492 \% ------- ----------- ----------- ----------- ----------- ----------- ----------- ------- 1 A T C C C G 45.3% 2 G C T T T C 21.2% 3 ^a^ G C T C T C 15.7% 4 G T C C C G 11.1% 5 G C C C T C 2.8% 6 G C C C C C 2.6% Rare 1.30% Rare are haplotypes \<1%; ^a^ corresponding to HT4 of Bodenmann et al. (2012) \[[@B14-genes-10-01021]\]. genes-10-01021-t007_Table 7 ###### Impact of ADORA2A haplotypes on total sleep time (TST) according to caffeine consumption. TST (h) --------- ------------- --------------- --------------- ---------------- -------------- ATCCCG 6.81 ± 0.08 7.54 ± 0.1 6.99 ± 0.14\# 6.60 ± 0.18 \# *p* \< 0.001 GCTTTC 6.63 + 0.19 6.92 ± 0.25\* 6.88 ± 0.17 6.46 ± 0.28\# *p* \< 0.01 GCTCTC 6.70 + 0.17 7.04 ± 0.26\* 6.89 ± 0.34 6.55 ± 0.23\# *p* \< 0.05 GTCCCG 6.87 + 0.18 7.38 ± 0.32 7.05 ± 0.38\# 6.69 ± 0.3\# *p* \< 0.001 GCCCTC 6.83 + 0.4 7.51 ± 0.27 7.05 ± 0.28\# 6.60 ± 0.52\# *p* \< 0.001 GCCCCC 6.76 + 0.23 7.30 ± 0.27 7.03 ± 0.42\# 6.54 ± 0.42\# *p* \< 0.001 Rare 6.79 + 0.46 7.31 ± 0.57 7.01 ± 0.62 6.60 ± 0.62\# *p* \< 0.001 Mean ± 95% CI; \* versus ATCCCG (most frequent haplotype 1); \# versus low caffeine consumers (0--50 mg/day). genes-10-01021-t008_Table 8 ###### Impact of the six ADORA2A SNPs on sleep complaints according to caffeine consumption. ------------------------------------------------------------------------------------------------------------------------------------------------------------------- SNP Genotype Low Caffeine Consumers\ Moderate Caffeine Consumers\ High Caffeine Consumers\ (0--50 mg/day) (51--300 mg/day) (\>300 mg/day) --------------- ---------- ------------------------- ------------------------------ -------------------------- ------------------- -------------- ----------------- *n* (%) ORa (95% CI) *n* (%) ORa (95% CI) *n* (%) ORa (95% CI) **rs5751862** G/G (a) 21 (40.4%) 1 57 (37.3%) 1 56 (52.3%) 1 G/A 52 (49.1%) 1.5 (0.6--3.9) 96 (42.1%) 1.3 (0.9--2.6) 79 (49.4.5%) 0.85 (0.5--1.2) A/A 18 (46.2%) 1.3 (0.9--8.7) 59 (55.7%) 1.4 (0.8--2.1) 46 (57.5%) 1.27 (0.8--2.0) **rs2298383** C/C (a) 19 (43.2%) 1 44 (38.9%) 1 24 (45.3%) 1 C/T 46 (50.6%) 1.3 (0.6--2.7) 71 (35.9%) 1.1 (0.7--1.8) 87 (50.6%) 1.1 (0.7--1.8) T/T 26 (41.9%) 0.9 (0.6--1.6) 85 (54.8%) 1.5 (1.1--2.8) \* 60 (52.6%) 1.3 (0.9--2.9) **rs3761422** C/C (a) 35 (44.3%) 1 100 (51.8%) 1 65 (51.6%) 1 C/T 44 (49.4%) 1.3 (0.3--1.6) 74 (35.4%) 0.7 (0.4--1.1) 86 (53.8.%) 1.1 (0.6--1.6) T/T 11 (39.3%) 0.3 (0.1--1.1) 33 (41.8%) 0.7 (0.4--1.4) 20 (47.6%) 0.9 (0.5--1.5) **rs2236624** C/C (a) 59 (45.4%) 1 132 (45.5%) 1 108 (52.7%) 1 C/T 31 (51.7%) 1.1 (0.5--2.6) 63 (38.4%) 0.9 (0.6--1.4) 59 (52.2%) 0.9 (0.6--1.2) T/T 2 (18.2%) 0 14 (48.3%) 0.9 (0.3--2.2) 5 (45.5%) 1.1 (0.5--2.5) **rs5751876** C/C (a) 31 (45.6%) 1 90 (52.6%) 1 64 (56.3%) 1 C/T 43 (47.3%) 1.2 (0.6--2.6) 72 (35.3%) 0.6 (0.4--0.9) \* 83 (52.2%) 0.9 (0.6--1.2) T/T 17 (44.7%) 0.8 (0.3--1.6) 41 (49.6%) 0.8 (0.3--1.2) 22 (47.8%) 0.8 (0.5--1.3) **rs4822492** C/C (a) 19 (43.2%) 1 45 (39.8%) 1 24 (45.3.%) 1 C/G 46 (50.5%) 0.7 (0.3--1.6) 72 (36.0%) 0.8 (0.5--1.6) 86 (50.9%) 1.1 (0.6--2.3) G/G 28 (43.8%) 1.0 (0.4--2.1) 87 (54.7%) 1.8 (1.1--2.9) \* 61 (53.0%) 1.3 (0.6--2.5) ------------------------------------------------------------------------------------------------------------------------------------------------------------------- \(a\) = ancestral. \* *p* \< 0.05 significant difference vs. the ancestral genotype; CI; \* versus ATCCCG (most frequent haplotype 1).

Background {#Sec1} ========== Rationale {#Sec2} --------- Atherosclerotic vascular disease is a chronic disease that causes blockages of arteries throughout the body. Conditions resulting from atherosclerotic disease are the most common cause of death in North America. The location of the vascular blockage dictates the disease; blockages in coronary arteries may lead to heart attacks, blockages in cerebral arteries may lead to strokes, and blockages in peripheral arteries may lead to gangrene. The prevalence of atherosclerotic peripheral vascular disease in the North American general population over 50 years of age is estimated at 17.4% \[[@CR1]\] and is rising in association with the increasing prevalence of diabetes. While bypass surgery is reserved for select patients with severe forms of atherosclerotic disease, the minimally invasive options of angioplasty are the initial treatment of choice for most patients. Angioplasty is the foundational treatment of endovascular therapy, which may be augmented by treatments such as stenting or atherectomy. Unfortunately, the 2-year patency of plain balloon angioplasty for peripheral vascular disease has been poor, reported between 50 and 80%, depending on lesion location and characteristics \[[@CR2]\]. Various techniques have been proposed to minimize restenosis following angioplasty; however, many have not been thoroughly evaluated in randomized trials. One such technique is prolonged angioplasty balloon inflation time, which is theorized to reduce post-interventional dissection and induce smooth muscle dysfunction, thereby reducing vasospasm and resulting stenosis. While animal studies have not revealed short-term mechanical advantage for prolonged balloon inflation \[[@CR3]\], smooth muscle dysfunction has been observed \[[@CR4]\]. Both of these studies compared brief balloon inflation times less than 60 s with inflation times greater than 60 s. Alternatively, prolonged balloon inflation may incur risk to the patient by simply mask flow-limiting dissections, which could have been identified if only transient balloon inflation was used. Identification of these dissections is relevant as they are at high risk of causing target lesion occlusion and necessitate stent placement. Conversely, in addition to being a therapeutic treatment for atherosclerotic disease, balloon angioplasty is known to cause vascular damage to previously healthy arteries in animal models after an inflation time of 60 s \[[@CR5]--[@CR7]\]. These articles present several proposed mechanisms for arterial injury during angioplasty; however, they all ultimately refer to intimal hyperplasia which occurs within hours to days after the initial angioplasty, due to inflammatory response following endothelial injury caused by the balloon inflation. This contradiction raises concern that although angioplasty may result in esthetic immediate results, there is a fundamental limitation to preserving long-term results after angioplasty. In particular, prolonged angioplasty balloon inflation may result in superior immediate results, but may also cause more profound vascular injury. The potential risks to patients of long-term vascular injury due to prolonged angioplasty is currently unknown. When angioplasty was first introduced in the last twentieth century, early randomized trials revealed immediate benefit of prolonged balloon inflation, although the sustainability of the patency benefit is conflicting \[[@CR8]--[@CR10]\]. However, the immediate benefit of prolonged angioplasty has not been consistently reported in randomized trials \[[@CR11]\]. Initial studies describing the methods of arterial angioplasty in peripheral arteries use 1 min duration inflations without justification for the inflation duration \[[@CR12]\]. Despite the ubiquitous use of angioplasty in atherosclerotic disease and ongoing primary investigation, there are no systematic reviews evaluating prolonged angioplasty balloon inflation. Currently synthesized evidence is insufficient to confidently direct clinical decision-making, and the current variation in operator preference of balloon angioplasty duration suggests ongoing clinical equipoise. Objectives {#Sec3} ---------- The objective of this systematic review is to evaluate the risks and benefits of prolonged angioplasty balloon inflation duration in the diverse patients with atherosclerotic disease. Specifically, our primary objective will be to determine if in patients receiving elective arterial angioplasty for stenotic or occlusive atherosclerotic disease, does prolonged balloon inflation of greater than 1 min duration, when compared with brief balloon inflation 1 min or less, improve residual stenosis in the immediate post-procedure angiogram. Secondary questions that will be addressed by this review will include:Is the effect of prolonged balloon inflation modulated by the location of atherosclerotic disease: coronary, cerebrovascular, and peripheral?Does prolonged balloon inflation, when compared with brief balloon inflation, affect immediate adverse radiographic events such as major vascular dissection?Does prolonged balloon inflation, when compared with brief balloon inflation, affect immediate clinically apparent adverse events secondary to vessel territory ischemia, such as heart attack or stroke?Does prolonged balloon inflation, when compared with brief balloon inflation, affect long term radiographic vessel patency?Does prolonged balloon inflation, when compared with brief balloon inflation, affect long term resolution of clinical ischemic symptoms? Methods {#Sec4} ======= Eligibility criteria {#Sec5} -------------------- ### Study designs {#Sec6} We will include randomized controlled trial (RCT), including cluster RCT, non-randomized controlled trials, cluster trials, interrupted time series studies, controlled before-after studies (CBA), and prospective or retrospective cohort studies. Case series, nested case-control, cross-sectional studies, and case reports will be excluded. ### Participants {#Sec7} We will include studies examining human adults (age 18 or older) who received angioplasty for atherosclerotic stenotic or obstructive vascular disease. Subgroup analysis examining clinical symptoms will exclude all trials that do not reference ischemia-based symptoms and only report radiographic results. Of note, when multiple arterial lesions are intervened upon within the same patient, each lesion will be counted as a separate "participant." ### Intervention and comparators {#Sec8} We will examine studies investigating elective arterial angioplasty. The angioplasty must be the primary purpose of the intervention and not be performed concurrently with a hybrid open vascular procedure on an in-line flow artery. The intervention may be performed on any arterial structure, including coronary, extracranial, intracerebral, or extremity arteries. The balloons may be drug-coated or lined with cutting ribs. However, we will exclude studies which employed adjunctive endovascular procedures prior to measurement of the outcome of interest, including but not limited to stent placement, orbital atherectomy, laser atherectomy, rotational atherectomy, or directional atherectomy. We will exclude venous angioplasty, arteriovenous fistula angioplasty, and studies examining emergency settings. We will only include studies examining de novo stenosis and therefore exclude studies examining angioplasty of restenosis, as the histopathology and outcomes related to restenosis are vastly different than de novo atherosclerosis \[[@CR13]\]. The intervention of interest is at least one single prolonged angioplasty balloon inflation greater than 1 min duration, and the control will be brief balloon inflation, with no single balloon inflation duration greater than 1 min. ### Outcomes {#Sec9} The primary outcome of interest is residual stenosis immediately following angioplasty, as determined by post-inflation angiogram. The definition threshold for residual stenosis must be defined and is generally between 20 and 50% stenosis. Secondary endpoints are categorized into immediate and long-term outcomes. Immediately, we will collect reported information on radiographic adverse events; this excludes residual stenosis and will be defined as a vascular anatomic abnormality that did not exist prior to balloon inflation. This includes arterial dissection, vessel perforation, and acute occlusion where there was a degree of patency prior to intervention. In addition to radiographic adverse events, we will collect newly developed clinically significant adverse events noted during the day of intervention, which broadly encompasses ischemic symptoms secondary to the territory of vascular intervention, for instance, a myocardial infarction following coronary artery angioplasty, stroke following cerebral vessel angioplasty, or acute limb ischemia following peripheral arterial angioplasty. Long-term secondary endpoints will also be differentiated into radiographic or clinical outcomes. We will collect data from any study reporting radiographic vessel patency by any validated modality, when performed on the day following the initial intervention: interventional angiogram, CT-angiogram, MRI-angiogram, and duplex ultrasound will be considered. Clinically, symptoms indicating long term resolution of ischemia will include angina pectoralis, cerebral hypoperfusion syndromes, claudication, or critical limb ischemia. ### Timing {#Sec10} Studies will be selected for inclusion if they report immediate post-angioplasty angiographic results, performed on the same day as the procedure. Long-term endpoint data will also be collected, as described above, if available. ### Setting {#Sec11} There are no restrictions regarding the setting of the study. ### Language {#Sec12} We will include English and French studies, with a list of potentially relevant translated titles in other languages included in an Additional file [1](#MOESM1){ref-type="media"}. Information sources {#Sec13} ------------------- A literature search strategy using medical subject headings and text words has been developed. We will search MEDLINE (OVID interface), EMBASE (OVID interface), and the Cochrane Central Register of Controlled Trials (Wiley interface). To ensure capture of all relevant trials, all selected studies will also undergo ancestry search, in addition to citation search using SCOPUS. OpenGrey will be interrogated for unpublished relevant literature. Search strategy {#Sec14} --------------- Both qualitative and quantitative studies will be collected. All searches will be limited by date of publication (January 1977--February 2018). The initial year of 1977 has been chosen as the first in-human use of angioplasty was performed that year. No language limit will be placed on the search; however, only English or French studies will be included in the analysis, with titles of potentially relevant studies in alternate languages included in the Additional file [1](#MOESM1){ref-type="media"}. The search strategy and syntax will be guided by a Health Sciences librarian with systematic review experience. After the MEDLINE search syntax has been finalized, it will be adapted to accommodate the remaining database searches. Please see Additional file [2](#MOESM2){ref-type="media"} for a complete search syntax used for the MEDLINE search. Of note, the PROSPERO database has been searched, and no ongoing or recently completed systematic review on this topic has been performed. Study records {#Sec15} ------------- ### Data management {#Sec16} Literature search results will be aggregated in EndNote, including where duplicate articles will be removed. The results will then be uploaded to the Distiller SR software, which will facilitate collaboration among all reviewers. The two screening authors will independently screen titles and abstracts resulting from the combined search of all selected databases. The full text of an article will be obtained for any articles that appear to meet eligibility criteria, at which point the full text will be screened and confirmation of article inclusion will be made. Any reasons for exclusion following full text screening will be explicitly documented and listed in an Additional file [1](#MOESM1){ref-type="media"}. Once both reviewers have created a complete list of eligible articles, the lists will be compared. Discrepancies in article selection will be addressed with discussion with a third-party author experienced in systematic review conduct. No authors will be blinded to journal titles, study authors, or study location of origin. ### Data collection process {#Sec17} A standardized form created in Distiller SR will be used as the data collection method. Both reviewers will have a separate form for each article, which will be compared for consistency after data collection has completed. Any discrepancy will be addressed with discussion with a third-party author experienced in systematic review conduct. Study authors will not be contacted to resolve unclear or inadequate reporting of data. Data items {#Sec18} ---------- Generic article data collected will include year of publication, trial design methodology, trial size (both in number of patients and number of lesions), duration of follow-up, financial support sources and involvement, and publication status. Patient-specific data will include average age, gender, symptom status, and the anatomic location of arterial lesions (classified as coronary, cerebrovascular, peripheral, and other). Intervention-specific data will include the trial protocol for angioplasty balloon inflation duration, the total number of repeated inflations, target balloon inflation pressure, use of adjunctive endovascular therapy following angioplasty, and the total number of lesions intervened upon in each intervention arm. We will also document the type of angioplasty device used; specifically, we will record any use of drug-eluting balloons, cutting rib balloons, and perfusion catheters. Outcome-specific data will include the blinding status of outcome adjudication, the definition of restenosis, definition of acute or chronic radiologic or symptomatic adverse events, and any long-term vessel patency or symptom status outcomes. Outcomes and prioritization {#Sec19} --------------------------- ### Primary outcome {#Sec20} The primary outcome will be the number of lesions who do not suffer from residual stenosis immediately following angioplasty balloon inflation, with a threshold as defined specifically by the individual trials but not exceeding the limits of 20--50% residual stenosis. We will record whether the treating physician or a blinded adjudicator determined the end-point. Secondary outcomes {#Sec21} ------------------ Immediate outcomesRadiographici.Defined as a vascular anatomic abnormality that did not exist prior to balloon inflation. We will record whether the treating physician or a blinded adjudicator determined the end-point. These will be recorded as dichotomous outcomes.Vessel dissectionVessel perforationDistal embolizationComposite endpoint, which may include dissection, perforation, or severe residual stenosisClinicali.Defined as newly developed ischemic symptoms, temporally related to intervention, and anatomically related to ischemia of the end-organ perfused by the artery receiving angioplasty. These will be recorded as dichotomous outcomes.Myocardial infarction or angina pectoralisClinical or biochemical diagnostic confirmationStroke or transient ischemic attackClinical diagnosis, with or without radiographic confirmationAcute limb ischemiaClinical diagnosis, with or without radiographic confirmationLong-term outcomesRadiographici.Defined as any record of radiographic vessel patency by any validated modality, when performed on the day following the initial intervention. The length of follow-up will be noted, and multiple records kept if follow-up images were obtained at multiple time point following the intervention. We will record whether the treating physician or a blinded adjudicator determined the end-point. These will be recorded as dichotomous outcomes.Interventional angiogramCT angiogramMR angiogramDuplex ultrasoundClinicali.Defined as the recurrence or development of chronic ischemic symptoms related to the end-organ perfused by the vessel intervened upon by angioplasty. These will be recorded as dichotomous outcomes.Angina pectoralisCerebral hypoperfusion syndromeClaudicationCritical limb ischemia. Risk of bias of individual studies {#Sec22} ---------------------------------- To assess individual studies for potential risk of bias, we will collect information guided by the Cochrane Collaboration tool for assessing risk of bias. In summary, this includes sequence generation, allocation concealment, blinding, incomplete outcome data, and selective outcome reporting. For each category, each study will be determined to be at either low or high risk. Alternately, if the report includes insufficient information to determine the level of risk, the category will be labeled as unclear. Determination of the level of bias will be made by the two reviewers independently and compared following complete assessment of all studies. Any discrepancy will be addressed with discussion with a third-party author experienced in systematic review conduct. The resulting risk of bias for each study, in each category, will be graphically represented by RevMan software. Data synthesis {#Sec23} -------------- ### Quantitative synthesis {#Sec24} Following data collection, if the studies are homogenous in terms of design, subjects, interventions, and outcomes, we will conduct a meta-analysis with random effects model. All outcomes that may undergo meta-analysis will be dichotomous in nature. Measurement effects will be determined by using risk ratio with 95% confidence intervals. As mentioned, the proposed unit of analysis will be arterial lesions; one subject may lend multiple units of analyses if they have multiple lesions undergoing angioplasty. Clustering of data for purposes of meta-analyses will not be considered due to variable reporting patterns, resulting in statistical limitations. However, for cluster randomized trials, we will account for the interclass correlation coefficient to appropriately modify results, as guided by the Cochrane Handbook for Systematic Reviews of Interventions. Due to the anticipated heterogeneous nature of clinical symptomatic status presentation and reporting, in particular given the broad anatomic inclusion of this study, clinical symptom outcomes will not be assessed using quantitative meta-analysis. ### Issues relating to data quality {#Sec25} In cases of unclear or inadequate data reporting, the authors will not be contacted for further data, and the data will be absent from meta-analysis. For trials that did not report on an intention-to-treat basis, or otherwise are at unique risk for missing or high risk of bias, sensitivity analysis will be used to assess the effect of inclusion of these trials. Trials will not be excluded due to the number of participants, however, will be assessed with sensitivity analysis if the size of trials raises heterogeneity concerns. To assess for study heterogeneity, we will first subjectively assess variability in patient and study baseline factors, intervention type, and outcome assessments. Statistical heterogeneity will be tested using the *I*^2^ statistic, with a threshold of 40--60% possibly representing heterogeneity and greater than 60% likely representing heterogeneity. In cases of likely heterogeneity, we will attempt to identify the source of heterogeneity through the use of pre-defined subgroup or post hoc sensitivity analysis. ### Quantitative data synthesis {#Sec26} All quantitative data syntheses will be performed on RevMan software as guided by the Cochrane Handbook for Systematic Reviews of Interventions. If acceptable levels of homogeneity are observed, cumulative effect estimates will be calculated with the Mantel-Haenszel method using fixed effects model. Alternately, a qualitative synthesis will be performed. ### A priori subgroup analyses {#Sec27} Planned subgroup analyses include:Anatomic location of arterial intervention (coronary, cerebrovascular, peripheral, and other)Number of balloon inflations (single inflation versus repeated inflations)Era of study (prior to year 2000 and following year 2000)Angioplasty device (plain balloon, drug-coated balloon, cutting balloon, perfusion catheter) ### Qualitative synthesis {#Sec28} All reported outcomes will be synthesized and reported in a qualitative manner. Furthermore, clinical outcomes will only be synthesized in a qualitative manner, as the expected heterogeneity in clinical situations and reporting will preclude quantitative analysis. ### Meta-bias {#Sec29} In addition to individual study assessment of risk of bias, all studies will be evaluated for indications of meta-bias. We will search for preceding published or registered protocols prior to study publication and evaluate for selective outcome reporting. Potential for reporting bias will be assessed by a funnel plot. Confidence in cumulative estimate {#Sec30} --------------------------------- The quality of all outcomes will be judged subjectively as a consensus among study authors, using the standardized Grading of Recommendations Assessment, Development and Evaluation methodology. Discussion {#Sec31} ========== This study intends to focus on a procedural technique for a systemic disease and encompasses a broad scope of specialties and anatomic regions. This breadth is intentional and intended to account for the relatively low anticipated body of evidence on this topic. However, the breadth of the study also introduces limitations during the conduct and interpretation of this review. While a meta-regression to determine the relative impact of angioplasty inflation duration on restenosis could theoretically demonstrate an ideal inflation duration, this will not be feasible due to the anticipated breadth of disease and reporting variability. Therefore, the primary conclusion of this study will determine whether balloon inflation of greater than 60 s is favorable; the specific ideal duration will not be determined in the context of these limitations. Furthermore, case-specific characteristics that are known to affect angioplasty success, such as other systemic diseases and lesion characteristics, will not be feasible to introduce into the analysis. Therefore, this study relies on the integrity of study allocation procedures to ensure the treatment arms are comparable. Despite these limitations, the potential results of this study remain insightful and necessary in light of the ubiquity of angioplasty and variations in practice. Adherence to the PRISMA-P guidelines (see Additional file [2](#MOESM2){ref-type="media"}) will assist in performing a transparent and thorough investigation. Additional files ================ {#Sec33} Additional file 1:Proposed search syntax for MEDLINE, using OVID interface. (LOG 100 bytes). (DOCX 24 kb) Additional file 2:PRISMA-P 2015 Checklist. (DOCX 33 kb) CBA : Controlled before-after trial RCT : Randomized control trial Funding {#FPar1} ======= There are no funding contributions to declare for this study. Availability of data and materials {#FPar2} ================================== The datasets generated and analyzed during the current study will be available from the corresponding author on reasonable request. MR conceived of the project and developed the first draft of the protocol. KR was a major contributor in writing the manuscript and will act as the second reviewer. PJ was a major contributor of project concept and subgroup analysis overview. GW guided conceptual and statistical aspects of the trial. DF oversaw all stages of the manuscript. All authors read and approved the final manuscript. Ethics approval and consent to participate {#FPar3} ========================================== Ethics is waived for a systematic review of published papers. Consent for publication {#FPar4} ======================= Not applicable Competing interests {#FPar5} =================== The authors declare that they have no competing interests. Publisher's Note {#FPar6} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1. Introduction {#sec1-ijms-18-02492} =============== Metabolic syndrome is an emerging epidemic worldwide that consists of an association of multiple cardiovascular risk factors \[[@B1-ijms-18-02492]\]. These factors, including hypertension, insulin resistance and obesity, directly contribute to a higher incidence of cardiovascular disease and type 2 diabetes. A rise in oxidative stress is one of the main suggested hypotheses to explain the increased risks associated with metabolic syndrome \[[@B1-ijms-18-02492],[@B2-ijms-18-02492],[@B3-ijms-18-02492]\]. Studies have suggested that oxidative stress notably enhanced superoxide anion (O~2~^●−^) production and decreased antioxidant reserve may be implicated in the pathogenesis and complications of hypertension, diabetes and obesity \[[@B2-ijms-18-02492],[@B3-ijms-18-02492],[@B4-ijms-18-02492],[@B5-ijms-18-02492],[@B6-ijms-18-02492]\]. Indeed, studies undertaken in animals have shown that vascular superoxide anion production was elevated in spontaneously hypertensive rats (SHR) \[[@B7-ijms-18-02492],[@B8-ijms-18-02492]\], insulin resistant rats \[[@B2-ijms-18-02492]\] and obese rats \[[@B4-ijms-18-02492]\]. In addition, our previous investigations have shown that vascular NADPH oxidase activity was increased concomitantly with enhanced O~2~^●−^ production in metabolic syndrome rat model induced by 10% [d]{.smallcaps}-glucose feeding in the beverage \[[@B9-ijms-18-02492],[@B10-ijms-18-02492]\] and in obese Zucker diabetic fatty rats \[[@B11-ijms-18-02492]\]. In addition, studies have suggested that oxidative stress is implicated in the development of neuropathic pain in hypertensive insulin-resistant rats induced by high glucose feeding \[[@B12-ijms-18-02492]\]. Kinin B~1~ receptor, a biomarker of inflammation, which is virtually absent in healthy tissues, is induced by the cytokine pathway and oxidative stress via the transcriptional nuclear factor κ B (NF-κB) \[[@B13-ijms-18-02492]\]. We have shown that oxidative stress is implicated in the induction and upregulation of the kinin B~1~ receptor in insulin-resistant and hypertensive glucose-fed rats \[[@B12-ijms-18-02492]\] and in obese Zucker diabetic fatty rats \[[@B11-ijms-18-02492]\]. Studies have demonstrated that α-tocopherol (vitamin E) supplementation decreases oxidative stress, lipid peroxidation and the elevated blood pressure in SHR \[[@B14-ijms-18-02492],[@B15-ijms-18-02492]\] and improves insulin action in obese Zucker rats \[[@B16-ijms-18-02492]\] and in type 2 diabetic patients \[[@B14-ijms-18-02492],[@B17-ijms-18-02492]\]. Additionally, investigations have shown that melatonin, a potent antioxidant, exerts beneficial effects on systolic blood pressure, glucose levels, lipid profile and body weight in the fructose-induced metabolic syndrome rat model \[[@B18-ijms-18-02492],[@B19-ijms-18-02492]\]. Melatonin was also found to improve oxidative stress, glucose homeostasis, lipid profile and to reduce body weight gain in young Zucker diabetic fatty rats \[[@B20-ijms-18-02492],[@B21-ijms-18-02492],[@B22-ijms-18-02492]\]. Another potent antioxidant, α-lipoic acid, prevented neuropathic pain and B~1~ receptor upregulation in insulin-resistant hypertensive rats \[[@B12-ijms-18-02492]\] and counteracted the increase in body weight gain by preventing the rise in adipose tissue NADPH oxidase activity and the hepatic upregulation of the kinin B~1~ receptor in obese Zucker diabetic fatty rats \[[@B11-ijms-18-02492]\]. Virgin argan oil is harvested from the fruit of the argan tree (*Argania spinosa* L. Skeels, 1911), which naturally grows in Morocco. Virgin argan oil is obtained by a cold-pressing technique \[[@B23-ijms-18-02492]\] and, consequently, is not altered during the extraction step \[[@B24-ijms-18-02492]\]. Virgin argan oil is characterized by high levels of linoleic and oleic acids, tocopherols \[[@B25-ijms-18-02492],[@B26-ijms-18-02492]\] and melatonin \[[@B27-ijms-18-02492],[@B28-ijms-18-02492]\], which exhibit significant antioxidant activities. It is noteworthy that argan oil is essentially rich in tocopherols particularly in its γ-isoform \[[@B25-ijms-18-02492]\], which is converted to α-isoform in the animal organism \[[@B29-ijms-18-02492]\]. Studies have reported that argan oil-supplemented healthy subjects exhibited lower plasma triglycerides, LDL-cholesterol and total cholesterol levels in association with lower lipid peroxidation and higher α-tocopherol concentrations \[[@B30-ijms-18-02492],[@B31-ijms-18-02492],[@B32-ijms-18-02492]\]. Moreover, investigations have shown that phenolic-extract from argan oil inhibits LDL-cholesterol oxidation in humans. More recently, we have shown that argan oil reduced systolic blood pressure, hyperglycemia and insulin resistance through its antioxidant properties in five-week glucose-fed rats \[[@B33-ijms-18-02492]\]. Argan oil also reduced the increase in plasma triglycerides, total cholesterol and LDL-cholesterol levels, the rise in plasma markers of lipid peroxidation and improved antioxidant status in high fat diet-induced obese rats \[[@B34-ijms-18-02492]\]. To the best of our knowledge, no study has investigated simultaneously the effects of argan oil on the three main cardiovascular risk factors associated with metabolic syndrome (hypertension, insulin resistance and obesity) and on one of its main complications, neuropathic pain. Thus, the present study was designed to investigate whether a chronic treatment of twelve weeks with argan oil can prevent or ameliorate arterial hypertension, insulin resistance, obesity, allodynia, alterations in plasma levels of triglycerides, free fatty acids, leptin and adiponectin in twelve-week glucose-fed rats. The impact of argan oil treatment was also determined on adiposity (number and size of epididymal adipocyte cells), oxidative stress (increase in basal superoxide anion production and NADPH oxidase activity) and on a key marker of inflammation (kinin B~1~ protein expression) in thoracic aorta, gastrocnemius muscle and epididymal fat. 2. Results {#sec2-ijms-18-02492} ========== 2.1. Blood Pressure {#sec2dot1-ijms-18-02492} ------------------- As shown in [Figure 1](#ijms-18-02492-f001){ref-type="fig"}, twelve weeks of treatment with glucose combined or not with corn oil resulted in a significant increase in systolic blood pressure in comparison to control rats. The synchronous feeding with argan oil reduced the rise in systolic blood pressure in glucose-treated rats, so that the systolic blood pressure in argan oil glucose-fed rats was significantly decreased in comparison to glucose or glucose combined with corn oil-fed rats. Chronic glucose feeding combined or not with corn oil had no significant effect on final body weight ([Figure 1](#ijms-18-02492-f001){ref-type="fig"}B) and body weight gain ([Figure 1](#ijms-18-02492-f001){ref-type="fig"}C). However, the chronic treatment with argan oil reduced significantly the final body weight and body weight gain in glucose-fed rats in comparison to control or glucose-fed rats. 2.2. Plasma Glucose and Insulin Levels, Insulin Resistance Index and Adiponectinemia {#sec2dot2-ijms-18-02492} ------------------------------------------------------------------------------------ As shown in [Figure 2](#ijms-18-02492-f002){ref-type="fig"}A, the twelve weeks of glucose feeding combined or not with corn oil significantly increased the blood glucose levels in comparison to control rats. The chronic treatment with argan oil significantly reduced the rise in blood glucose levels in glucose-fed rats. As shown in [Figure 2](#ijms-18-02492-f002){ref-type="fig"}B, the plasma insulin levels were increased significantly in glucose-fed rats in comparison to control animals. The simultaneous oral feeding with argan oil or corn oil prevented (*p* \< 0.05) this increase so that the plasma insulin values of argan and corn oil glucose-fed rats did not statistically differ from that in control rats ([Figure 2](#ijms-18-02492-f002){ref-type="fig"}B). The chronic glucose feeding significantly increased the insulin resistance index, as estimated by HOMA ([Figure 2](#ijms-18-02492-f002){ref-type="fig"}C). The chronic treatment with argan oil or corn oil prevented (*p* \< 0.05) this rise so that the insulin resistance index values of argan and corn oil glucose-fed animals did not statistically differ from that in control rats ([Figure 2](#ijms-18-02492-f002){ref-type="fig"}C). As shown in [Figure 2](#ijms-18-02492-f002){ref-type="fig"}D, the twelve weeks of treatment with glucose increased the plasma adiponectin levels by 287% (*p* \< 0.05) in comparison to control rats. The synchronous treatment with argan oil or corn oil further increased this parameter by 866% and 915% (*p* \< 0.05), respectively, in comparison to control animals ([Figure 2](#ijms-18-02492-f002){ref-type="fig"}D). The plasma adiponectin levels of argan and corn oil glucose-fed rats were significantly higher in comparison to those of rats fed with glucose ([Figure 2](#ijms-18-02492-f002){ref-type="fig"}D). 2.3. Epididymal Fat Weight per Body Weight, Lipid Profile and Leptinemia {#sec2dot3-ijms-18-02492} ------------------------------------------------------------------------ Twelve weeks of treatment with glucose combined or not with corn oil caused a significant increase in epididymal fat weight per body weight in comparison to control rats ([Figure 3](#ijms-18-02492-f003){ref-type="fig"}A). The oral treatment with argan oil prevented the increase in epididymal fat weight per body weight in glucose-fed rats, so that this value did not statistically differ from that in control rats ([Figure 3](#ijms-18-02492-f003){ref-type="fig"}A). As shown in [Figure 3](#ijms-18-02492-f003){ref-type="fig"}B, the plasma triglyceride levels were significantly increased in glucose-fed rats. The simultaneous oral feeding with argan oil or corn oil prevented this increase so that the plasma triglyceride values of argan and corn oil glucose-fed rats did not statistically differ from that in control rats ([Figure 3](#ijms-18-02492-f003){ref-type="fig"}B). As shown in [Figure 3](#ijms-18-02492-f003){ref-type="fig"}C, the plasma free fatty acids' levels were not significantly modified either by glucose feeding, or by corn oil, or by argan oil. Chronic glucose feeding resulted in a significant increase in plasma leptin levels ([Figure 3](#ijms-18-02492-f003){ref-type="fig"}D). This latter effect was completely prevented by the treatment with argan oil or corn oil. The plasma leptin values of argan and corn oil glucose-fed rats did not statistically differ from that in control rats ([Figure 3](#ijms-18-02492-f003){ref-type="fig"}D). 2.4. Number and Size of Epididymal Adipocyte Cells {#sec2dot4-ijms-18-02492} -------------------------------------------------- As shown in [Figure 4](#ijms-18-02492-f004){ref-type="fig"}A, the epididymal adipocyte cells number was not significantly affected either by glucose feeding, or by corn oil, or by argan oil. Moreover, chronic glucose feeding combined or not with corn oil had no effects on epididymal adipocyte cells size ([Figure 4](#ijms-18-02492-f004){ref-type="fig"}B). The twelve weeks of treatment with argan oil, however, significantly decreased the epididymal adipocyte cells size in glucose-fed rats in comparison to control rats ([Figure 4](#ijms-18-02492-f004){ref-type="fig"}B). 2.5. Tactile and Cold Allodynia {#sec2dot5-ijms-18-02492} ------------------------------- As shown in [Figure 5](#ijms-18-02492-f005){ref-type="fig"}A, the twelve weeks of glucose feeding resulted in significant tactile allodynia in comparison to control rats. Corn oil treatment exacerbated this neuropathic pain syndrome in glucose-fed rats. In contrast, the simultaneous treatment with argan oil prevented (*p* \< 0.05) the occurrence of tactile allodynia in glucose-fed rats ([Figure 5](#ijms-18-02492-f005){ref-type="fig"}A). Glucose-fed rats displayed significant cold allodynia when compared to control rats ([Figure 5](#ijms-18-02492-f005){ref-type="fig"}B). Corn oil treatment reduced by 31% this syndrome, although not significantly in glucose-fed rats, while argan oil treatment significantly decreased cold allodynia in glucose-fed rats ([Figure 5](#ijms-18-02492-f005){ref-type="fig"}B). Values of cold allodynia in corn and argan oil glucose-treated rats were not significantly different from control values. 2.6. Oxidative Stress Parameters {#sec2dot6-ijms-18-02492} -------------------------------- The basal superoxide anion production was significantly increased in the thoracic aorta ([Figure 6](#ijms-18-02492-f006){ref-type="fig"}A), gastrocnemius muscle ([Figure 6](#ijms-18-02492-f006){ref-type="fig"}C) and epididymal fat ([Figure 6](#ijms-18-02492-f006){ref-type="fig"}E) of glucose-fed rats and glucose-fed rats co-treated with corn oil. Administration of argan oil prevented this increase in the three tissues of glucose-fed rats, so that the basal superoxide anion production did not significantly differ from that in control rats. Likewise, NADPH oxidase activity was significantly increased in thoracic aorta ([Figure 6](#ijms-18-02492-f006){ref-type="fig"}B), gastrocnemius muscle ([Figure 6](#ijms-18-02492-f006){ref-type="fig"}D) and epididymal fat ([Figure 6](#ijms-18-02492-f006){ref-type="fig"}F) of glucose-fed rats treated or not with corn oil. The simultaneous treatment with argan oil blocked (*p* \< 0.05) the rise in NADPH oxidase activity to control values in the three tissues of glucose-fed rats. 2.7. Kinin B~1~ Receptor Protein Expression {#sec2dot7-ijms-18-02492} ------------------------------------------- As shown in [Figure 7](#ijms-18-02492-f007){ref-type="fig"}A, kinin B~1~ receptor protein expression was significantly increased in thoracic aorta of glucose-fed rats, and this increase was prevented (*p* \< 0.05) by corn oil or reduced by argan oil to values not significantly different from the control. Chronic glucose feeding also resulted in a significant increase in B~1~ receptor protein expression in the gastrocnemius muscle ([Figure 7](#ijms-18-02492-f007){ref-type="fig"}B). Treatment with argan oil prevented (*p* \< 0.05) the rise in B~1~ receptor protein expression in glucose-fed rats to values that did not significantly differ from control ([Figure 7](#ijms-18-02492-f007){ref-type="fig"}B). In contrast, corn oil failed to affect this rise in the gastrocnemius muscle in glucose-fed rats significantly ([Figure 7](#ijms-18-02492-f007){ref-type="fig"}B). B~1~ receptor protein expression was not detectable in the epididymal fat of the four groups of rats. 3. Discussion {#sec3-ijms-18-02492} ============= This study showed enhanced basal superoxide anion production and NADPH oxidase activity in epididymal fat, gastrocnemius muscle and thoracic aorta in association with an upregulation of kinin B~1~ receptor protein expression in gastrocnemius muscle in rats treated for 12 weeks with glucose combined or not with corn oil. This was accompanied by an increase in epididymal fat weight per body weight, plasma levels of triglycerides and leptin together with the development of hyperglycemia, hyperinsulinemia, insulin resistance, hypertension and allodynia, a feature of neuropathic pain. Importantly, all these metabolic syndrome features were prevented or reduced by argan oil feeding. Data on blood pressure are consistent with previous studies in which we have shown that oral glucose feeding for three, four or five weeks caused an increase of systolic blood pressure \[[@B33-ijms-18-02492],[@B35-ijms-18-02492],[@B36-ijms-18-02492]\]. Since high blood pressure and enhanced basal superoxide anion production and NADPH oxidase activity from aortic tissue in glucose-fed rats were not affected by corn oil, this may suggest an implication of the vascular oxidative stress in the sustained elevation of systolic arterial pressure induced by chronic glucose feeding. This is in accordance with previous studies, which have shown an antihypertensive effect of argan oil in obese insulin-resistant rats \[[@B37-ijms-18-02492]\], SHR \[[@B38-ijms-18-02492]\] and five-week glucose-fed rats \[[@B33-ijms-18-02492]\]. This is in keeping with the beneficial effect of argan oil supplementation on plasma lipid profile and oxidant-antioxidant status in the rat model of high-fat diet-induced obesity \[[@B34-ijms-18-02492]\]. Argan oil was found to be rich in compounds with significant antioxidant activities particularly tocopherols \[[@B25-ijms-18-02492],[@B26-ijms-18-02492]\] and melatonin \[[@B27-ijms-18-02492],[@B28-ijms-18-02492]\]. Interestingly, antioxidant treatment with α-tocopherol was found to reduce oxidative stress and the increase in blood pressure in SHR \[[@B15-ijms-18-02492]\]. Moreover, investigations have reported that phenolic extract from argan oil inhibits human low-density lipoprotein (LDL) oxidation \[[@B39-ijms-18-02492]\]. Additionally, melatonin counteracted the increase in systolic blood pressure in the fructose-induced metabolic syndrome rat model \[[@B19-ijms-18-02492]\]. Therefore, the present study suggests that argan oil, which is rich in antioxidant compounds notably tocopherols, melatonin and sterols, reduced blood pressure through the prevention of oxidative stress that may also involve kinin B~1~ receptor upregulation at the vascular level in chronically-glucose-fed rats. To support this hypothesis, kinin B~1~ receptor antagonism and the use of other antioxidant therapies (α-lipoic acid, *N*-Acetyl-[l]{.smallcaps}-cysteine) reversed arterial hypertension together with the oxidative stress and B~1~ receptor upregulation in this model of metabolic syndrome \[[@B12-ijms-18-02492],[@B40-ijms-18-02492],[@B41-ijms-18-02492]\]. In the present study, the preventive effect of argan oil on hyperinsulinemia and insulin resistance might be associated with the blockade of the kinin B~1~ receptor protein overexpression in the gastrocnemius muscle and its composition in fatty acids notably in linoleic and oleic acids \[[@B42-ijms-18-02492]\] in glucose-fed rats. This is in accordance with a previous study \[[@B34-ijms-18-02492]\], which has reported that four weeks of treatment with argan oil blunted hyperglycemia and hyperinsulinemia in obese insulin-resistant rats. Argan oil normalized blood glucose levels and improved insulin sensitivity in fat and liver tissues of obese insulin-resistant rats \[[@B43-ijms-18-02492]\]. α-tocopherol, among the main compounds found in argan oil, improved insulin action and reduced plasma lipid peroxidation in obese insulin-resistant Zucker rats \[[@B16-ijms-18-02492]\]. Additionally, melatonin, another compound of argan oil, was found to improve oxidative stress and glucose homeostasis in young Zucker diabetic fatty rats \[[@B21-ijms-18-02492],[@B22-ijms-18-02492]\], although it remains uncertain whether the low quantity of melatonin in argan oil (60.5 ng·kg oil^−1^) contributes to the beneficial effects on oxidative stress and glucose metabolism in glucose-fed rats. The beneficial effect of corn oil on hyperinsulinemia and insulin resistance in glucose-fed rats is in agreement with our recent study in five-week glucose-fed rats \[[@B33-ijms-18-02492]\] and with previous studies that have reported that oleic and linoleic acids, the principal compounds of corn oil \[[@B44-ijms-18-02492]\], improved insulin sensitivity in skeletal muscle cells \[[@B45-ijms-18-02492]\]. Argan oil treatment resulted in a high increase in plasma adiponectin levels in chronically-glucose-fed rats. Adiponectin, a protein secreted by adipocytes, increases the sensitivity of insulin by activating muscle glucose uptake via promoting Glut 4 translocation and inhibiting hepatic gluconeogenesis \[[@B46-ijms-18-02492]\]. Adiponectin was also found to act as an agonist of peroxisome proliferator activated receptor γ (PPARγ), leading to additional plasmatic glucose uptake \[[@B46-ijms-18-02492]\]. PPARγ exerts an inhibitory effect on NF-κB, a nuclear factor involved in transcription of many genes encoding inflammatory proteins \[[@B47-ijms-18-02492]\] such as the kinin B~1~ receptor \[[@B13-ijms-18-02492]\]. Thus, the inhibition of NF-κB by the marked increase of adiponectin levels in rats fed with argan oil may account for the suppression of kinin B~1~ receptor expression in glucose-fed rats. Therefore, the present study suggests that argan oil prevents the development of hyperinsulinemia and insulin resistance by counteracting the kinin B~1~ receptor upregulation in skeletal muscle and by increasing plasma adiponectin levels independently of an action on oxidative stress in chronically-glucose-fed rats. Furukawa et al. \[[@B48-ijms-18-02492]\] have suggested that increased oxidative stress in accumulated fat is an important pathogenic mechanism of obesity-induced metabolic syndrome. Argan oil prevented the increase in epididymal fat weight and plasma triglyceride levels and the rise in superoxide production and NADPH oxidase activity in the epididymal fat of glucose-fed rats. This is in agreement with a previous study \[[@B34-ijms-18-02492]\], which has reported that a diet supplemented with argan oil reduced the increase in adipose tissue weight, body weight and serum triglyceride concentrations in high fat diet-fed rats. Treatment with NADPH oxidase inhibitor reduced ROS production in adipose tissue, improved hypertriglyceridemia, hyperglycemia, hyperinsulinemia and adiponectin levels in obese mice \[[@B48-ijms-18-02492]\]. Studies have reported that adiponectin inhibits the synthesis of fatty acids and stimulates their oxidation \[[@B46-ijms-18-02492]\]. These findings are in accordance with our results showing a reducing effect of argan oil on adipocyte size, fat pads and body weight in chronically-glucose-fed rats. Hence, the lower body weight could be responsible for the improvement of the metabolic parameter, notably plasma triglyceride levels in argan oil glucose-fed rats. Moreover, the fact that argan oil and NADPH oxidase inhibitor exert comparable beneficial effects on body weight, plasma levels of triglycerides, glucose, insulin and adiponectin suggest that adipose tissue oxidative stress is implicated in the visceral obesity and insulin resistance observed in the chronically-glucose-fed rat model. Treatment with glucose caused a significant increase in plasma leptin levels. These results are in accordance with previous studies showing elevated serum leptin levels in this model of glucose-fed rats \[[@B49-ijms-18-02492]\] and in obese rats and humans \[[@B34-ijms-18-02492],[@B50-ijms-18-02492]\]. Studies have suggested that the obese state is characterized by leptin resistance \[[@B50-ijms-18-02492]\]. Interestingly, argan oil prevented the increase in plasma leptin levels in chronically-glucose-fed rats. This may suggest that argan oil increases the sensitivity to leptin in the hypothalamus. Sour et al. \[[@B34-ijms-18-02492]\] did not observe any effect of supplemented diet with argan oil on the rise in plasma leptin levels in high fat-treated rats. This discrepancy may be explained by the difference in the animal model of metabolic syndrome, the dose and the duration of argan oil treatment used between the two studies. Melatonin was shown to decrease plasma leptin levels and body weight in obese rats \[[@B18-ijms-18-02492],[@B51-ijms-18-02492]\], and α-tocopherol decreased plasma triglyceride levels in obese high fat-treated animals \[[@B52-ijms-18-02492],[@B53-ijms-18-02492]\]. Nevertheless, the fact that corn oil prevented the rise in plasma leptin levels with no change in superoxide anion production and NADPH oxidase activity in adipose tissue suggests that argan oil exerts its beneficial effect on leptin levels through a mechanism unrelated to oxidative stress. The present study also showed that argan oil prevented tactile and cold allodynia in chronically-glucose-fed rats. This is in agreement with previous studies \[[@B12-ijms-18-02492],[@B40-ijms-18-02492]\], which have shown that α-lipoic acid and *N*-acetyl-[l]{.smallcaps}-cysteine, two potent antioxidants, alleviated tactile and cold allodynia in glucose-fed rats. Therefore, one may suggest that argan oil exerts its beneficial effects on sensory abnormalities associated with metabolic syndrome through its blockade of oxidative stress. Indeed, corn oil, which was devoid of antioxidant properties, exacerbated tactile allodynia and was less effective than argan oil at reducing cold allodynia. 4. Materials and Methods {#sec4-ijms-18-02492} ======================== 4.1. Animals and Protocols {#sec4dot1-ijms-18-02492} -------------------------- All research methods and animal care procedures were approved by the Animal Care Committee of the Université de Montréal (Protocol 15-084; Approval date: 3/09/ 2015) in compliance with the guiding principles as enunciated by the Canadian Council on Animal Care and the ARRIVE guidelines \[[@B54-ijms-18-02492],[@B55-ijms-18-02492]\]. Male Sprague-Dawley (SD) rats weighing 70--80 g (Charles River Laboratories, St-Constant, QC, Canada) were housed two per cage, under controlled conditions of temperature (22 °C) and humidity (43%), on a 12-h light-dark cycle. One week after their arrival, rats were randomly divided into four groups of 10 rats and treated for 12 weeks as follows: Group 1 had free access to a drinking solution of 10% [d]{.smallcaps}-glucose (Sigma-Aldrich, Oakville, ON, Canada) and to a standard chow diet (2018 Teklad Global 18% Protein Rodent Diet); Group 2 had free access to 10% [d]{.smallcaps}-glucose and was fed daily by gavage with argan oil (5 mL kg^−1^); Group 3 had free access to 10% [d]{.smallcaps}-glucose and was fed daily by gavage with corn oil (5 mL kg^−1^); Group 4, represents the control group and had free access to tap water only and to the same standard chow diet. It is noteworthy that animals were handled every day, one week before the beginning of gavage and through the period of treatment. Moreover, our laboratory staff is well trained with the gavage procedure that we used in previous articles for other treatments \[[@B33-ijms-18-02492],[@B41-ijms-18-02492]\]. Therefore, we did not feel at any time that animals were stressed by the gavage procedure. Argan oil was obtained from Argan3 Inc. (Montreal, QC, Canada) and was 100% pure with the following composition: palmitic acid 14%, stearic acid 5%, oleic acid 43.5%, linoleic acid 37%, linolenic acid 0.6%, sterols, schottenol 45%, spinasterol 35% and tocopherols 1034 mg·kg^−1^ with no additives or ingredients. A recent study has shown that virgin argan oil also contains melatonin (60.5 ng kg oil^−1^), which is a potent antioxidant \[[@B27-ijms-18-02492]\]. The duration of treatment with argan oil was based on our previous investigations showing that fourteen weeks of treatment with the antioxidant α-lipoic acid prevented the development of arterial hypertension, insulin resistance and reduced the plasma lipid levels in glucose-fed rats \[[@B9-ijms-18-02492]\]. Corn oil was purchased from ACH Food Companies Inc. (Oakville, ON, Canada) and has the following composition: palmitic acid 13%, stearic acid 3%, oleic acid 31%, linoleic acid 52% and linolenic acid 1%. Corn oil was found to be safe with no clinical signs or toxicity at 5 mL kg^−1^ day^−1^ for 12 weeks in treated SD rats \[[@B56-ijms-18-02492]\]. We compared corn oil to argan oil because they contain approximately the same composition of fatty acids with the exception that argan oil also contains tocopherols, sterols and, to lesser extent, melatonin \[[@B42-ijms-18-02492]\]. The dose of argan and corn oil was selected on the basis of our previous study \[[@B33-ijms-18-02492]\], which failed to show any negative effects when administered for a period of 5 weeks in this model and on the basis of other studies using 5 mL kg^−1^ day^−1^ chronically administered by oral gavage in other animal models \[[@B37-ijms-18-02492]\]. If one takes into consideration differences in the body surface area between rat and human, the dose of 5 mL kg^−1^ day^−1^ has to be divided by 6.2 to achieve the equivalent dose in human \[[@B57-ijms-18-02492]\]. This dose is comparable with doses used in human studies \[[@B58-ijms-18-02492]\]. After 12 weeks of treatment, systolic blood pressure was measured by tail-cuff plethysmography based on an average of 5 readings per animal (ADI Instruments Inc., Colorado, CO, USA) and registered with the ADI Instruments Program (Lab Chart Pro7.Ink). The rats were sacrificed by decapitation after light anesthesia with isoflurane. The blood was collected in a vacutainer tube early in the morning after fasting overnight (16 h) for the subsequent measurements of plasma insulin, triglycerides, free fatty acids, leptin and adiponectin. Three representative tissues involved in hypertension, insulin sensitivity and visceral obesity (thoracic aorta, gastrocnemius muscle and epididymal fat) were removed and kept frozen at −20 °C until the subsequent measurement of oxidative stress and kinin B~1~ receptor expression. 4.2. Measurement of Metabolic Parameters {#sec4dot2-ijms-18-02492} ---------------------------------------- Blood glucose concentrations in overnight-fasted rats were measured with a glucometer (Accu-Chek Aviva, Roche Diagnostics, Laval, QC, Canada). Plasma insulin, adiponectin and leptin were measured by Rat/Mouse ELISA kits from Millipore Canada Limited (Etobiocoke, ON, Canada). Plasma triglycerides and free fatty acids were measured by enzymatic kits from Cayman Chemical Company (Ann Arbor, MI, USA). To evaluate the degree of insulin resistance, the Homeostasis Model Assessment (HOMA) was used as an index of insulin resistance and calculated by the following formula: insulin (µU mL^−1^) × glucose (mM) ÷ 22.5 \[[@B59-ijms-18-02492]\]. 4.3. Measurement of Allodynia {#sec4dot3-ijms-18-02492} ----------------------------- Tactile and cold allodynia were assessed with the rats placed on a wire mesh floor beneath an inverted plastic cage. The rats were allowed to adapt for about 15 min or until explorative behaviour ceased. Tactile allodynia was assessed by measuring the hind paw withdrawal threshold to the application of a calibrated series of six von Frey filaments (bending forces of 2, 4, 6, 8, 10 and 15 g) applied perpendicularly to the mid-plantar surface as described previously \[[@B12-ijms-18-02492],[@B40-ijms-18-02492],[@B60-ijms-18-02492]\]. Cold allodynia was assessed using the acetone drop method applied to the plantar surface of the hind paws as previously described \[[@B12-ijms-18-02492],[@B40-ijms-18-02492],[@B60-ijms-18-02492]\]. The frequency of paw withdrawal was expressed as a percentage (the number of paw withdrawals ÷ number of trials × 100). 4.4. Measurement of Superoxide Anion and NADPH Oxidase Activity {#sec4dot4-ijms-18-02492} --------------------------------------------------------------- Superoxide anion (O~2~^●−^) production was measured from frozen thoracic aorta, gastrocnemius muscle and epididymal fat using the lucigenin-enhanced chemiluminescence method as described previously \[[@B11-ijms-18-02492],[@B61-ijms-18-02492],[@B62-ijms-18-02492]\]. Briefly, small slices of tissues were preincubated in Krebs-HEPES buffer (saturated with 95% O~2~ and 5% CO~2~, at room temperature) for 30 min and then transferred to a glass scintillation vial containing 5 µM of lucigenin for the determination of basal O~2~^●−^ levels. The chemiluminescence was recorded every minute for 10 min at room temperature in a liquid scintillation counter (Wallac 1409, Turku, Finland). Lucigenin counts were expressed as cpm mg^−1^ of dry weight tissue. Moreover, NADPH oxidase activity in the samples was assessed by adding 0.1 mM NADPH to the vials before counting \[[@B63-ijms-18-02492]\]. The background was counted using a vial containing the solution with no tissue. The basal superoxide production was evaluated by subtracting the background count from the luminescence value induced by the addition of tissue slices while the NADPH oxidase activity was measured by subtracting the background count from the luminescence value induced by the addition of tissue slices plus NADPH. 4.5. Western Blot Analysis {#sec4dot5-ijms-18-02492} -------------------------- Western blot was performed as described previously \[[@B49-ijms-18-02492]\]. Total proteins were loaded (20--30 μg) in each well of 10% SDS-PAGE. Dynein was used as the standard protein. Detection of kinin B~1~R was made with a specific polyclonal rabbit antiserum (1:1000) \[[@B64-ijms-18-02492],[@B65-ijms-18-02492]\]. Detection of dynein was made with anti-dynein antibody (1:4000 mouse, SC-13524; Santa Cruz Biotechnology, Santa Cruz, CA, USA). Secondary antibodies were horseradish peroxidase (HRP)-linked goat anti-rabbit SC-2004 and HRP-linked goat anti-mouse SC-2005 (Santa Cruz) used at a dilution of 1:25,000 (for B~1~R) and 1:5000 (for dynein). A quantitative analysis of proteins was provided by scanning densitometry using the MCID-M1 system (Imaging Research, St. Catharines, ON, Canada). 4.6. Adipocyte Morphometry {#sec4dot6-ijms-18-02492} -------------------------- After overnight fixation in 4% paraformaldehyde, epididymal fat wedges were embedded in paraffin and cut into 20-μm sections. Slides were stained with methylene blue after deparaffinisation. Sections snapshots, taken using a DAGE-MTI CCD72 digital camera, were analyzed with MCID-M1 software (Imaging Research). Adipocyte cells size, expressed in µm^2^, represents the average of 20 cells per section in 5 sections per animal for 5--6 rats per group. Adipocyte cells number, expressed as the number per mm^2^, represents the average of 20--30 mm^2^ per section in 5 sections per animal for 5--6 rats per group. Color photomicrographs were taken with a camera (QImaging Retiga-2000R, Burnaby, BC, Canada). 4.7. Statistical Analysis of Data {#sec4dot7-ijms-18-02492} --------------------------------- Data are expressed as the mean ± SEM of values obtained from *n* rats in each group. Statistical analysis was performed using Prism^TM^ Version 5.0 (GraphPad Software Inc., La Jolla, CA, USA); data and statistical analysis comply with the recommendations on experimental design and analysis in pharmacology \[[@B66-ijms-18-02492]\]. Data were subjected to one-way ANOVA, followed by the Bonferroni/Dunn multiple comparison test when F achieved *p* \< 0.05, and there was no significant variance in homogeneity. Significance was considered when *p* \< 0.05. 5. Conclusions {#sec5-ijms-18-02492} ============== Data show that argan oil prevented the increase in visceral obesity, plasma triglyceride levels, hyperinsulinemia, insulin resistance, neuropathic pain and reduced hyperglycemia and high blood pressure in a model of metabolic syndrome induced by high glucose feeding. These beneficial effects of argan oil are probably explained, at least in part, by (1) improvement of the metabolic status due to the loss of body weight, (2) its antioxidant properties, (3) the suppression of kinin B~1~ receptor upregulation and (4) the consequent inhibition of hyperleptinemia and the increase of plasma adiponectin levels. Thus, the present study supports the use of argan oil as a potential nutri-therapeutic agent in the prevention of hypertension, insulin resistance and obesity. This work was supported by a grant from the Canadian Institutes of Health Research (CIHR, MOP-119329) to Réjean Couture. The funder CIHR had no role in the design, analysis or writing of this article. The authors acknowledge the technical assistance of Jacques Sénécal. Argan oil was offered graciously by Argan3 Company. Adil El Midaoui and Réjean Couture conceived of and designed the experiments. Adil El Midaoui and Youssef Haddad performed the experiments, analysed the data and made the figures. Younes Filali-Zegzouti contributed to the discussion of the manuscript. Adil El Midaoui supervised the study and drafted the paper. Réjean Couture edited the final version of the paper. All authors approved the final manuscript. The authors declare no conflicts of interest. BK Bradykinin GLUT 4 Glucose transporter 4 HOMA Homeostasis model assessment NF-κB Transcriptional nuclear factor κ B NADPH Nicotinamide adenine dinucleotide phosphate hydrogen O 2 ●− Superoxide anion PPARγ Peroxisome proliferator activated receptor γ SHR Spontaneously hypertensive rat {#ijms-18-02492-f001} {#ijms-18-02492-f002} {#ijms-18-02492-f003} {#ijms-18-02492-f004} {#ijms-18-02492-f005} ###### Effects of 12 weeks of glucose feeding combined or not with orally and daily administered (5 mL kg^−1^) corn oil or argan oil on (**A**) superoxide anion production expressed in cpm mg^−1^ of thoracic aorta, (**B**) NADPH oxidase activity expressed in cpm mg^−1^ of thoracic aorta, (**C**) superoxide anion production expressed in cpm mg^−1^ of gastrocnemius muscle, (**D**) NADPH oxidase activity expressed in cpm mg^−1^ of gastrocnemius muscle, (**E**) superoxide anion production expressed in cpm mg^−1^ of epididymal fat and (**F**) NADPH oxidase activity expressed in cpm mg^−1^ of epididymal fat. Values are the mean ± SEM of 4--6 rats per group. \* *p* \< 0.05 vs. control, ^\#^ *p* \< 0.05 vs. glucose, ^+^ *p* \< 0.05 vs. glucose + corn oil.   {#ijms-18-02492-f007} [^1]: These authors contributed equally to this work.

Introduction {#s1} ============ It has been challenging to investigate microglia during inflammatory contexts primarily because distinguishing them from infiltrating monocytes and macrophages has been unreliable due to similar expression of cell-surface markers ([@B1]--[@B6]). Microglia contribute to an immunosuppressive environment, are derived from the yolk-sac, reside within two distinct niches in the retina, and show differential regulation because of inherent microglial heterogeneity ([@B7], [@B8]). Nonetheless conflicting data remains on how microglia regulate or promote inflammation depending on insult ([@B9], [@B10]). During inflammation microglia alter their homeostatic state ([@B8], [@B11]--[@B13]). In acute inflammation, markers including *C5ar1* show promise in delineating sub-populations of microglia mounting an LPS response ([@B13]). We wished to determine if the microglial transcriptome resets after an acute and resolving insult, or if homeostatic thresholds have been reset or altered permanently. Recent advancements in transgenic mouse lines, but also in identification of markers that are "microglial-specific," for example *Cd44, P2ry12, Siglech*, and *Tmem119*, have contributed in various degrees toward microglial identification ([@B12], [@B14]--[@B20]). However, loss of these markers at the transcript level are observed when microglia are perturbed and some of these markers are lost at the protein level, highlighting the need for careful validation of these markers as microglial-specific given their expression is dependent upon the disease context ([@B8], [@B11]--[@B13]). The *Cx3cr1*^*CreER*^*:R26-tdTomato* mouse strain permits binary discrimination of the microglia from other immune cells. The model utilises the high expression of *Cx3cr1* in microglia and the longevity (as low level self-replication) of microglia in comparison to other immune cells ([@B14]). In this study, we validate the *Cx3cr1*^*CreER*^*:R26-tdTomato* strain as sensitive and specific for tagging retinal microglia and perform mRNA-Sequencing on microglia obtained from individual retina during and after resolution of acute inflammation induced by intravitreal injection of LPS \[the endotoxin-induced uveitis (EIU) model\]. We show that the retinal microglia undergo acute transcriptional changes which resolve to their original homeostatic state by 2 weeks and support microglial heterogeneity in response to inflammatory signals. Materials and Methods {#s2} ===================== Mice ---- *Cx3cr1*^*CreER*^*:R26-tdTomato* mice on a C57BL/6J background were provided by Clemens Lange (University of Freiburg, Germany). Breeding colonies of homozygotes were established, and offspring crossed with C57BL/6J mice to generate heterozygotes for experiments. Genotyping (via PCR) of breeding pairs was performed. Mice were confirmed as negative for the *Rd8* mutation ([@B21]). All mice were housed at the University of Bristol Animal Services Unit under specific pathogen free conditions with food and water *ad libitum*. All procedures were conducted in concordance with the United Kingdom Home Office licence (PPL 30/3281) and were approved by the University of Bristol Ethical Review Group. The study also complied with the Association for Research in Vision and Ophthalmology (ARVO) Statement for the Use of Animals in Ophthalmic and Visual Research. ### Tamoxifen Preparation and Administration Tamoxifen (T5648; Sigma-Aldrich, Poole, UK) was dissolved in corn oil (C8267; Sigma-Aldrich) to a concentration of 20 or 5 mg/mL, for subcutaneous injection and topical administration, respectively. The solutions were freshly prepared by overnight incubation in an orbital shaker at 42°C and 300 rpm. Mice were injected with 200 μL subcutaneously \[100 μL into both the lower (inguinal) left and right quadrants using a 25G needle\] on days 1 and 3; alternatively, mice were administered 10 μL topically to the eye three times daily (minimum gap of 2 h between dosing) for up to 4 days. ### Induction of EIU Prior to anesthesia, pupils were dilated using topical tropicamide 1% w/v and phenylephrine 2.5% w/v (Minims; Chauvin Pharmaceuticals, Romford, UK). Mice were anaesthetised by intraperitoneal injection of 90 μL/10 g body weight of a solution containing 6 mg/mL ketamine (Ketavet; Zoetis Ireland Ltd., Dublin, Ireland) and 2 mg/mL Xylazine (Rompun; Bayer plc, Newbury, UK) mixed with sterile water. Mice were selected for injection (or to be used as a control) in a constrained randomised order within blocks using Excel 2016 (Microsoft, Redmond, WA); blocks were dependent on cage allocations, which itself was dependent on the litter they were derived from. The allocations ensured that all experiments had littermate controls. Intravitreal injections were performed as previously described ([@B22]). In brief, 2 μL volume of PBS containing 10 ng LPS from *E. coli* 055:B5 (Sigma-Aldrich) was delivered into the intravitreal space via the pars plana, using an operating microscope and a 33-gauge needle on a microsyringe (Hamilton Company, Reno, NV) under direct visualisation. Immediately following injection, 1% w/w chloramphenicol ointment (Martindale Pharma, Romford, UK) was applied topically, with the animals monitored and kept on a heat-pad during recovery. ### EIU Clinical Assessment At selected time-points (4 h, 18 h, and 2 weeks) post-injection, pupils were dilated and mice anaesthetised for clinical assessment. The Micron IV retinal imaging microscope (Phoenix Research Laboratories, Pleasanton, CA) was used to capture optical coherence tomography (OCT) scans, and brightfield and fluorescence fundal images. Prior to imaging, the Micron IV CCD and OCT were calibrated in accordance with the manufacturer\'s protocol. The gain was set to +3 dB and the FPS to 15, or +12 dB and 2 for brightfield and tdTomato fluorescence imaging, respectively. For tdTomato imaging, a 550/25 nm bandpass excitation and 590 nm longpass emission filter were used (Edmund Optics, Barrington, NJ). For OCT, the parameters were defined according to the manufacturers protocol, and scans were taken 30 times in rapid succession and averaged. Full-length B-scans were taken horizontally and vertically with the optic disc centered. Images were stored in the TIFF file format. Isolation and Flow Cytometric Phenotyping of Retinal Immune Cells ----------------------------------------------------------------- Eyes were dissected in 100 μL ice-cold PBS with aqueous, vitreous, and retina extracted by a limbal incision, lens removal and transfer into a 1.5 mL microcentrifuge tube. The tissue was mechanically disrupted by rapping the tube along an Eppendorf rack 12 times before transfer into a 96-well 60 μm nylon mesh filter plate (Merck Millipore, UK). The plate was centrifuged at 400 *xg* for 5 min and the supernatant aspirated. The remaining cell pellet was resuspended in FACS buffer (PBS supplemented with 3% v/v Foetal calf serum and 0.9 mg/mL sodium azide) and transferred to a 96-well V-bottom plate for FACS staining. Single cell suspensions from spleen and brain were generated by mechanical dissociation through a 70 μm cell strainer with a syringe plunger. Cells were centrifuged at 400 *xg* for 3 min, erythrocytes lysed using ammonium-chloride-potassium (ACK) buffer, and cells resuspended in FACS buffer. Cells were incubated with purified rat anti-mouse CD16/32 Fc block \[Clone 2.4G2; Becton Dickson (BD) Biosciences, Oxford, UK\] and 7-aminoactinomycin D (7AAD; Thermo Fisher Scientific, Waltham, MA) diluted in FACS buffer for 20 min at 4°C. Cells were then incubated with an antibody cocktail containing fluorochrome-conjugated monoclonal antibodies (see [Supplementary Table 1](#SM1){ref-type="supplementary-material"}) against various mouse cell-surface markers for 20 min at 4°C. All mAbs were titrated and tested using control tissues. Cell suspensions were acquired using a 4-laser Fortessa X-20 flow cytometer (BD Cytometry Systems, Oxford, UK). Compensation was performed using OneComp eBeads (Thermo Fisher Scientific), Anti-Rat Ig, κ/Negative Control Compensation Particles Set (BD Biosciences), or AbC total antibody compensation bead kit (Thermo Fisher Scientific). To compensate tdTomato, cell suspensions prepared from *Cx3cr1*^*CreER*^*R26-tdTomato* homozygote brains were used. Fluorescence-minus-one (FMO) controls were used to assist in gating of the markers selected for validation. Following acquisition, analysis was performed using FlowJo software (Treestar, San Carlos, CA). ### Fluorescence-Activated Cell Sorting Retinal single cell suspensions were re-suspended in 200 μL FACS buffer containing DRAQ7 (DR77524; Biostatus, Shepshed, UK). Live microglial cells (tdTomato^hi^ DRAQ7^−^) were immediately sorted using a BD Influx Cell Sorter. Retinal samples were prepared in small batches in order to maintain cell viability and ensure high quality RNA. Cells were sorted into 0.2 mL endonuclease-free tubes containing 0.05 μL RNase inhibitor, 0.95 μL lysis buffer, and a variable amount of nuclease-free water depending on the number of cells collected: $9.5 - {({\frac{x}{850}*3})}$ μL, where *x* is the number of cells isolated, using components of the SMART-Seq v4 Ultra Low Input RNA Kit for Sequencing (Takara Bio USA, Inc., Mountain View, CA). Samples were sorted in a constrained randomised order in blocks; blocks were made as small as possible and consisted of a balance of every experimental group to ensure that time- or order-dependent (batch) effects of sorting were mitigated. mRNA-Sequencing --------------- ### Sample and Library Preparation Samples were prepared using the SMART-Seq v4 Ultra Low Input RNA Kit for Sequencing, according to the user manual, to convert the mRNA into cDNA and amplify by Long Distance (LD) PCR (16 cycles for 600 cells). cDNA was isolated using the Agencourt AMPure XP Kit (Beckman Coulter, Brea, CA) and quantified using the Agilent High Sensitivity DNA Kit on an Agilent 2100 Bioanalyser (Agilent Technologies, Santa Clara, CA). The library preparation was performed using the Nextera XT DNA Library Preparation Kit (Illumina Inc., San Diego, CA). ### Sequencing and Analysis Samples were sequenced to depths of up to 16.7 million single-end 75 nt length reads per sample using the Illumina NextSeq 500/550 High Output v2 kit (75 cycles) on an Illumina NextSeq 500 Sequencing System. Image analysis, base calling, and generation of sequence reads were produced using the NextSeq Control Software v2.0 (NCS) and Real-Time Analysis Software v2 (RTA). Data was converted to FASTQ files using the bcl2fastq2 v2.20 software (Illumina Inc.). Sequencing data was initially quality-checked using FastQC[^1^](#fn0001){ref-type="fn"}, before alignment and initial analysis. The data was processed through an analysis pipeline using the Partek Flow (Build version 6.0.17.0614; Partek Inc., St. Louis, MO) software with the following task nodes (non-default parameters are specified in brackets): Trim adapters (inputting Nextera XT Index Kit v2 adapter sequences[^2^](#fn0002){ref-type="fn"}, Trim bases (From 3′ end, 1 base), Trim bases \[from 3′ end with minimum quality score (Phred) of 30\], Align reads using STAR (2.5.3a using mm10 as the reference index), Quantify to transcriptome (Partek E/M using mm10 -- Ensembl Transcripts release 89 as the reference index) ([Supplementary Figures 1, 2](#SM1){ref-type="supplementary-material"}). The Partek Flow data output was further analysed using Partek Genomics Suite (PGS) (Version 6.6, Build 6.16.0812). PGS normalises data using the reads per kilobase million (RPKM) approach and performs differential gene expression analysis using an ANOVA model; a gene is considered differentially-expressed (DEG) if it has an FDR step-up *p* ≤ 0.05 and a fold-change ≥±2. The data was subsequently analysed for enrichment of GO terms and the KEGG pathways; a pathway is considered significantly-enriched if the enrichment score is ≥3 (equivalent to a *p* ≤ 0.05). The fold-change and *p*-values of all genes were then imported into Ingenuity Pathway Analysis (IPA) version 01-13 (Qiagen Bioinformatics, Aarhus, Denmark) and analysed according to the manual. Pathways were considered significantly altered if they had a *p* ≤ 0.05 and a z-score (directionality score) ≥±2. PGS and IPA were both used to generate figures. Quantitative PCR ---------------- The remaining cDNA generated from the sorted cells was used for transcript-level validation. qPCR was performed using the TaqMan Universal Master Mix II, with UNG (4440038) and TaqMan gene expression probes (4331182) on a Quantstudio 3 Real-Time PCR system (A28137; all products from Thermo Fisher Scientific). Samples were run in technical duplicate, using 1 ng as the input amount, and analysed using the equation: 2^Cq(mean\ (control))−Cq(sample)^. The probes used were: *Bst2* (mm1609165_g1), *C5ar1* (mm00500292_s1), *Cd44* (mm01277161_m1), *Fas* (mm01204974_m1), *Lair1* (mm00618113_m1), *Mertk* (mm00434920_m1), *Milr1* (mm01242703_m1), *P2ry12* (mm01950543_s1), *Siglech* (mm00618627_m1), *Slamf1* (mm00443317_m1). Immunohistochemistry -------------------- Eyes from euthanised mice were enucleated and fixed in 4% v/v PFA for 1 h before dissection. The anterior portion of the eye (cornea, iris, ciliary body, and lens) was carefully removed and an eyecup prepared. The eyecup tissue was blocked in 5% v/v normal goat serum (Vector Laboratories, CA, USA), 1% v/v BSA, and 3% v/v Triton x-100 (both Sigma Aldrich) in PBS for 4 h at room temperature with gentle shaking. Eyecups were then incubated at 4°C overnight with a rabbit anti-mouse anti-RFP mAb (600-401-379; Rockland Immunochemicals Inc., Limerick, PA) and for target validation experiments a Super Bright 600-conjugated anti-mouse CD44 mAb ([Supplementary Table 1](#SM1){ref-type="supplementary-material"}) was used in combination. After thorough washing with PBS, samples were incubated overnight with the secondary antibody goat anti-rabbit Alexa-633 (A21070; Thermo Fisher Scientific). The eyecups were washed again, and the retina carefully removed and flatmounted in Vectashield hard-set antifade mounting media (H-1400; Vector Laboratories Ltd., Peterborough, UK) and imaged on a Leica SP5-AOBS confocal laser scanning microscope (Leica Microsystems Ltd., Wetzlar, Germany). Images were acquired with an xy pixel size ≤ 200 nm, and a z-step size of ≤ 400 nm. Statistical Analysis and Image Processing ----------------------------------------- Data were analysed using GraphPad Prism 7 software (GraphPad Software Inc., San Diego, CA). The One-way ANOVA with Tukey\'s multiple comparisons test was used to compare multiple groups of data to a control group. A *p* ≤ 0.05 was considered significant. Huygens professional software (Scientific Volume Imaging B.V., Hilversum, The Netherlands) was used to deconvolve the Micron IV fluorescent images and fluorescence microscopy ([Supplementary Figure 3](#SM1){ref-type="supplementary-material"}). For the Micron images, the following parameters were used: lens immersion = 1.343 \[refractive index of the 0.2% w/w carbomer eye gel ([@B23])\], embedding = 1.377 ([@B24]), peak emission = 581 nm, numerical aperture = 1.25, and xy pixel size of 130 nm; the background was estimated at 2 and a signal-to-noise ratio of 15 was used. Hot pixel correction (with a sensitivity of 4) was used prior to deconvolution. For fluorescence microscopy, the parameters were imported from the microscope and the default settings were used. Microscopy images were processed using the Leica LAS X software (Leica Microsystems Ltd.) and FIJI \[a distribution of ImageJ ([@B25])\]. Other images, and figures, were processed using Photoshop (Adobe Inc., San Jose, CA). Results {#s3} ======= Subcutaneous and Topical Tamoxifen Administration Regimes Tag Retinal Microglia in the *Cx3cr1^*CreER*^:R26-tdTomato* Mouse Line -------------------------------------------------------------------------------------------------------------------------------- Activation of the CreER system requires administration of tamoxifen or 4-hydroxytamoxifen, and various methods of administration are frequently employed such as subcutaneous injection and oral gavage ([@B14], [@B26]). Moreover, an efficacious 4-day topical regime for activating CreER systems within the eye was recently described ([@B27]). Therefore, we initially tested which tamoxifen administration regime tagged retinal microglia efficiently while minimising systemic immune cell labelling to preclude analysing cells subsequently recruited during inflammation. Five regimes were tested, spanning from 1- to 4-day topical and subcutaneous administration. A no-tamoxifen control was included. At 4 weeks post-tamoxifen treatment, flow cytometric analysis of the retinas quantified the proportion of CD45^int^ CD11b^+^ microglia (in a naïve retina) that were tdTomato^hi^ as corroborated with fluorescent fundal imaging ([Figures 1A,B](#F1){ref-type="fig"}). Microglia displayed a "resting" ramified morphology ([Figure 1C](#F1){ref-type="fig"}). The 3- and 4-day topical, and the subcutaneous regime tagged ≥95% of microglia whereas the no-tamoxifen controls had 45% tagged constitutively ([Figure 1D](#F1){ref-type="fig"}). All tdTomato^hi^ cells were microglia, indicating specificity in the normal adult mouse retina. {#F1} *Cx3cr1^*CreER*^:R26-tdTomato* Mice Exhibit Typical Kinetics of the EIU Model ----------------------------------------------------------------------------- The stability of microglial tagging during inflammation is pivotal to the isolation of pure populations of microglia for downstream transcriptomic assessment before, during, and after inflammation. To determine this, we utilised endotoxin-induced uveitis (EIU), a self-resolving model of acute Toll-like receptor 4 (TLR4)-mediated ocular inflammation, that following a single inflammatory insult generates acute immune cell tissue infiltration ([@B28], [@B29]). We confirmed disease kinetics in the *Cx3cr1*^*CreER*^*:R26-tdTomato* heterozygotes replicated our previous data in C57BL/6J mice ([@B22]). A time-course using OCT, deconvolved fluorescent fundal imaging, and confocal microscopy confirmed tdTomato expression throughout the expected time course of disease ([Figure 2](#F2){ref-type="fig"}). Intravitreal administration of LPS results in peak cellular infiltrate at 18 h, which then resolves by 2 weeks ([Figure 2A](#F2){ref-type="fig"}). Fluorescent fundal imaging demonstrates that the uniform distribution of microglia is altered in response to LPS, with defined areas of tdTomato^+^ cell accumulation observed at peak disease ([Figure 2B](#F2){ref-type="fig"}). Confocal microscopy confirmed tdTomato^+^ cells in the naïve retina with a ramified morphology that became amoeboid at peak and recovered to a ramified appearance by 2 weeks post-EIU induction ([Figure 2C](#F2){ref-type="fig"}). {#F2} Three-Day Topical Tamoxifen Induction Ensures Specificity of tdTomato Labelling in Microglia -------------------------------------------------------------------------------------------- With tdTomato expression confirmed as stable in the microglial population, we sought to determine whether this was exclusive to microglia in the context of immune cell infiltration, validating accurate isolation of the microglial transcriptome. To confirm the specificity for retinal microglial tagging during inflammation, we examined tamoxifen regimes which efficiently tagged microglia (3- and 4-day topical, and subcutaneous) in addition to a no-tamoxifen control. Flow cytometry was performed on peripheral tissues and a small portion of myeloid cells were observed as tdTomato^hi^, primarily when tamoxifen was administered via the subcutaneous route ([Supplementary Figure 4](#SM1){ref-type="supplementary-material"}). However, to test the specificity within the retina, inflammation was required. Retinas were retrieved at the peak cellular infiltrate stage of EIU (18 h) and flow cytometry was performed ([Supplementary Figure 5](#SM1){ref-type="supplementary-material"}). Single, live cells identified as tdTomato^hi^ were assessed by conventional microglia markers of CD45^int^ and CD11b^+^ ([Figures 3A,B](#F3){ref-type="fig"}). A small proportion (5%) of all single live retinal cells express very low levels of tdTomato ([Figure 3A](#F3){ref-type="fig"}, boxed), as described in the tdTomato reporter mice (in the absence of Cre/Cre^ER^)[^3^](#fn0003){ref-type="fn"} due to a failure of a small proportion of ribosomes to terminate translation when reaching the stop codon that precedes *tdTomato* ([@B30]). Using this gating strategy, analysis from the 4-day topical and subcutaneous regimes demonstrates the presence of an additional population of non-microglial (CD45^hi^ CD11b^−^) cells that were identified in the tdTomato^hi^ subset ([Figure 3B](#F3){ref-type="fig"}). Gating with CD45^int^ and CD11b^+^ but not tdTomato leads to the inclusion of non-microglial cells ([Figure 3C](#F3){ref-type="fig"}) and greatly compromises specificity. The 3-day topical and no-tamoxifen control retained the specificity of tdTomato for microglia ([Figures 3B,D](#F3){ref-type="fig"}), but as shown previously a no-tamoxifen control does not fully label the microglial population ([Figure 1](#F1){ref-type="fig"}). As our approach for distinguishing microglia (CD45^int^ CD11b^+^ tdTomato^hi^) from the tdTomato^hi^ group could include non-microglia that possess a similar transcriptional profile to retinal microglia, the total counts of the CD45^int^ CD11b^+^ tdTomato^hi^ group, from naïve and peak EIU retinas (3-day topical tamoxifen), were compared and no significant difference was observed confirming a pure microglial population ([Figure 3E](#F3){ref-type="fig"}). All subsequent experiments performed used the 3-day topical tamoxifen regime. {#F3} Microglia Undergo Acute Transcriptional Changes During EIU That Fully Normalise to a Baseline State by 2 Weeks -------------------------------------------------------------------------------------------------------------- The 3-day tamoxifen regime in the *Cx3cr1*^*CreER*^*:R26-tdTomato* strain provides the sensitivity and specificity required for reliable identification and isolation of distinct retinal microglia populations from other populations of infiltrating immune cells. To characterise changes in the microglial transcriptome during and after resolution of EIU, FACS (based on tdTomato) of 600 live microglia was performed from individual retinas collected at 4 h, 18 h, and 2 weeks post-injection in addition to naïve mice. The majority of these samples (28/30) yielded high-quality cDNA for sequencing using a validated pipeline ([Supplementary Figure 1](#SM1){ref-type="supplementary-material"}). Our approach identified 1,069 unique differentially expressed genes (DEGs; 613 at 4 h, 537 at 18 h, and 0 at 2 weeks) visualised by hierarchical clustering, revealing a highly plastic transcriptome with most up-regulated genes being mutually exclusive at different timepoints. Boxes highlight clusters of genes that were normal at 4 h but up-regulated at 18 h (yellow), those which were up-regulated at 4 h but not 18 h (brown), those which were up-regulated at both time-points (green), those which were down-regulated at 4 h but recovered to pre-EIU levels by 18 h (light blue), and those which were down-regulated at both 4- and 18 h (black). Restoration back to a homeostatic signal was observed by 2 weeks because unsupervised clustering failed to distinguish naïve and 2-week post-injection samples ([Figure 4A](#F4){ref-type="fig"}). A multitude of expected gene changes based on published literature were observed, including downregulation of microglial homeostatic genes (*Gpr34, Mafb*), expression of microglial activation and LPS-response genes (*Cxcl10, Map3k8*), in addition to no change in the "primed microglia" gene *Axl* ([@B11], [@B12], [@B17], [@B31]) ([Figure 4B](#F4){ref-type="fig"}). A heatmap demonstrates the change in z-score over time of canonical pathways identified as significantly different ([Figure 4C](#F4){ref-type="fig"}). The original canonical pathways (including bars showing exact *p*-values) are presented in [Supplementary Figure 6](#SM1){ref-type="supplementary-material"}. {ref-type="supplementary-material"}. EIU, endotoxin-induced uveitis. ^\*^*p* ≤ 0.05, ^\*\*^*p* ≤ 1\*10^−3^, ^\*\*\*^*p* ≤ 1\*10^−6^, ^\*\*\*\*^*p* ≤ 1\*10^−9^.](fimmu-10-03033-g0004){#F4} Enriched GO terms (enrichment score) when naïve and 4 h EIU groups were compared included: immune system process (60.9), regulation of cytokine production (56.0), and response to stress (52.7). Similarly, the enriched KEGG pathways included: NF-Kappa B signaling pathway (34.1), toll-like receptor signaling pathway (31.7), and TNF signaling pathway (25.3). For the naïve and 18 h EIU comparison GO terms included: cytosolic part (89.7), extracellular organelle (70.67), translation (65.8), and immune system process (29.8); enriched KEGG pathways included: ribosome (58.8) and proteasome (42.9). LPS was found to be a "master regulator" by IPA, indicating agreement of our data with the curated lists and pathways. Flow Cytometry and Fluorescence Microscopy Validate Key Transcriptional Alterations ----------------------------------------------------------------------------------- Our next aim was for orthogonal validation of key and novel transcriptional changes, both at the RNA and protein level. Markers for validation were selected by systematic assessment based on magnitude of the relative change in expression, novelty, lack of prior validation at the protein level, whether they were a previously suggested microglial marker, were in contrast to or appeared crucial in light of other reports, and lastly the availability of testing reagents ([Supplementary Figure 7](#SM1){ref-type="supplementary-material"}). Ultimately, the final set of 10 markers selected represent a variety of expression patterns and kinetics to match the plastic landscape identified by hierarchical clustering. In line with published reports and an activated state, pro-inflammatory markers (*Slamf1, C5ar1, Fas*, and *Cd44*) were all upregulated at 4 h following LPS challenge. In addition, a novel microglial associated transcript, *Milr1* (a negative regulator of mast cell activation) and *Bst2* (a previously validated marker of late activation) were elevated by 18 h. In contrast, constitutively expressed microglial genes, including homeostatic genes (e.g., *P2ry12, Siglech, Mertk*, and *Lair1*) were down-regulated at the early time-point. In general, qPCR analysis validated the transcript-level changes observed at each time-point, confirming resolution and return to baseline levels by 2 weeks ([Figure 5](#F5){ref-type="fig"}). {#F5} Flow cytometric analysis demonstrates increased expression in SLAMF1, MILR1, C5AR1, CD44, BST2, and LAIR1 at 18 h post-injection ([Figures 6A--D](#F6){ref-type="fig"}). Furthermore, differences in the proportion of marker-positive cells were evident, with C5AR1, CD44, and BST2 upregulated in the majority of microglia (\>50%), in contrast to the other makers which were elevated in a smaller fraction (\<20%) of cells. The upregulation of CD44 was also confirmed in retinal flatmounts at 18 h using fluorescence microscopy ([Figure 6E](#F6){ref-type="fig"}). Whilst P2RY12 was highly expressed in naïve microglia (\>80%), no change in expression in response to LPS were observed. Similarly, low level SIGLECH, MERTK, and FAS expression in naïve populations remained unchanged and restricted to a small percentage of the microglia (\<10%). We also compared expression of P2RY12 and SIGLECH on CD45^+^ cells, as both are previously suggested markers that differentiate microglial populations from other immune cells ([@B17], [@B20]). Flow cytometric analysis clearly demonstrates that both markers are equally expressed on CD45^+^ infiltrating cells, indicating these markers exhibit poor specificity for retinal microglia during the acute response ([Figure 6F](#F6){ref-type="fig"}). {#F6} Stratifying Microglia Using C5AR1 Identifies Both Generalised and Restricted Microglial Responses ------------------------------------------------------------------------------------------------- Recent reports show that *C5ar1* was one of several markers that was enriched in a subset of brain microglia (identified by sc-mRNA-Seq data) responding to systemic LPS challenge *in vivo* ([@B13]). Furthermore, mounting evidence from numerous reports identify heterogeneity in the microglial response during other pathological states ([@B8], [@B32]). We therefore examined whether stratifying microglia based on C5AR1 expression would delineate differences in the markers selected for validation, highlighting specificity to this subset of C5AR1-expressing cells, or generalised expression across the whole population of microglia. Microglia were stratified into three main groups: C5AR1^neg^, C5AR1^lo^, and C5AR1^hi^. The C5AR1-expressing microglia were sub-stratified based on whether the C5AR1 expression was equivalent to microglia observed within a naïve mouse (C5AR1^lo^) or whether expression was elevated (C5AR1^hi^; [Figure 7A](#F7){ref-type="fig"}). C5AR1^hi^ expression correlated to the extent of immune cell (CD45^+^ TdTomato^−^) infiltrate within the retinas and represents a potential microglial marker for disease scoring ([Figure 7B](#F7){ref-type="fig"}). Flow cytometric analysis compared expression of the other surface markers in C5AR1^neg^ and C5AR1^hi^ subsets, as two distinctive populations, in naïve and peak EIU retinas. Delineating the two populations on this basis demonstrates elevation of several markers (SLAMF1, FAS, MILR1, and LAIR1) which are restricted to the C5AR1-expressing microglia ([Figure 7C](#F7){ref-type="fig"}). In contrast CD44 and BST2 were enriched within the C5AR1-expressing population but also expressed by a large proportion of the C5AR1^neg^ microglia, thus representing more generalised markers of microglial perturbation ([Figure 7D](#F7){ref-type="fig"}). {#F7} Discussion {#s4} ========== Single eye mRNA-Seq revealed inflammation-responsive transcriptional changes in retinal microglia following LPS stimulation which resolve within 2 weeks, confirming the potential for these cells to reset their homeostatic state. In line with the literature, our data confirms recognised patterns of altered gene expression in homeostatic and activation pathways, an absence of changes in "primed microglia" genes, and identifies enriched pathways relating to immune function. Hierarchical clustering of the 1,069 DEGs identified reveals a high level of plasticity in the microglial transcriptome, with the majority of up-regulated genes at early (4 h) and peak (18 h) being mutually exclusive across the time-points. At the transcript level at least, this highlights a consensus group of microglial LPS-response genes including (*Slamf1, C5ar1, Fas, Cd44, Milr1*, and *Bst2*). In contrast, we highlight a common cluster of down-regulated genes, including homeostatic regulators (*P2ry12, Siglech, Mertk, Lair1*). Orthogonal validation by qPCR of the 10 markers selected representing key and novel transcripts, confirmed the RNA-Seq findings. However, altered expression of some of these membrane-associated makers did not translate to changes at the protein level, as determined by flow cytometry. In part, this may reflect the presence of intracellular protein that our flow cytometric approach did not detect, or represents genuine discrepancies between the transcriptome and proteome, as these are not always in direct proportion ([@B33]--[@B35]). Our results emphasise caution in reading out RNA-Seq alone as a true representation of the cell\'s activity and highlight the need for orthogonal validation. Nonetheless, the single eye mRNA-Seq approach identified key and novel transcriptional changes which informed subsequent testing on a smaller number of markers via low-throughput approaches. We confirmed and validated some of the transcriptional changes by flow cytometry but found that the differences were enhanced, and an additional marker (FAS) was identified as significantly different, when microglia were first stratified based on their C5AR1 expression as suggested via a transcripts-led report ([@B13]). Furthermore, we identified markers which were exclusive to the C5AR1-expressing microglia (SLAMF1, FAS, MILR1, and LAIR1), but also generalised markers which were not (CD44 and BST2). This agrees with other reports which suggest that C5AR1 is needed for microglial polarisation to pro-inflammatory states, and that its knock-out improved outcomes in an Alzheimer\'s model ([@B36]). Furthermore, microglial heterogeneity has already been reported in Alzheimer\'s disease, light-damage models, and in response to LPS stimulation *in vivo* ([@B8], [@B13], [@B32]). Understanding microglial heterogeneity, and identifying changes which are exclusive to subpopulations, is critical for developing targeted therapies. Conflicting with previous reports investigating LPS-responses and experimental autoimmune encephalomyelitis (EAE), we did not find significant down-regulation of P2RY12 upon microglial activation ([@B11], [@B13]). In EAE, the microglia exhibit a chronic inflammatory state different to the acute LPS response, whilst the report investigating the LPS response used a systemic dose 400 times greater than our own local dose and had an endpoint 24 h post-injection which could explain the discrepancy. We suggest microglial loss of P2RY12 as context-dependent, for example when subject to a significant immune stimulus or persistent inflammation. For this investigation, a clear marker to distinguish long-lived, yolk-sac derived microglia from infiltrating myeloid cells was critical, particularly during peak ocular inflammation at 18 h post-LPS injection. The *Cx3cr1*^*CreER*^*:R26-tdTomato* line has been validated for this approach previously ([@B14], [@B37]), but as we demonstrate the systemic tamoxifen administration also labels peripheral leukocyte populations which are recruited to the eye during inflammation. Modifying a recently published protocol ([@B27]) demonstrates that the 3-day topical tamoxifen administration regime is robust in labelling retinal microglia in both a sensitive ([Figure 1](#F1){ref-type="fig"}) and specific ([Figure 3](#F3){ref-type="fig"}) manner, superior to the established subcutaneous route. Shortening of the originally-published topical protocol from 4 days to three is likely possible due to the very high expression of *Cx3cr1* in microglia and reduces confounders of animal handling and stress to the mice as a refinement[^4^](#fn0004){ref-type="fn"} The high expression of *Cx3cr1* by microglia also explains why mice underwent tamoxifen-independent (or constitutive) recombination in 45% of the microglia as tamoxifen-independent recombination (constitutive receptor activation of the CreER) is a well-characterised phenomenon ([@B38]--[@B40]). With other administration routes, we found that the specificity for microglia was reduced during ingress of immune cells with inflammation. This supports recent studies of microglia that highlight the need to confirm the specificity for microglia in their disease model and employ techniques with single-cell resolution to resolve the non-microglial cell populations ([@B8], [@B41]). Taking all our data together, we show that three previously suggested markers (P2RY12, CD44, and SIGLECH) exhibit poor specificity for microglia. However, with the *Cx3cr1*^*CreER*^*:R26-tdTomato* line it remains possible to validate potential markers, assuming the line retains specificity for microglia across disease contexts. Microglial subtypes can exhibit both differential and generalised responses to LPS. In summary, we demonstrate that the homeostatic threshold of retinal microglia is reset following an acute inflammatory insult and identify potential markers for delineating the heterogeneity of microglia that may be used depending on context of retinal perturbation. Data Availability Statement {#s5} =========================== The datasets generated for this study can be found in the Gene Expression Omnibus (GEO) repository under the accession number [GSE138247](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE138247). Spreadsheets of gene expression values, in addition to the lists of DEGs (with *p*-values and fold-changes) are included as part of this upload. Ethics Statement {#s6} ================ All procedures were conducted in concordance with the United Kingdom Home Office licence (PPL 30/3281) and were approved by the University of Bristol Ethical Review Group. The study also complied with the Association for Research in Vision and Ophthalmology (ARVO) Statement for the Use of Animals in Ophthalmic and Visual Research. Author Contributions {#s7} ==================== OB, DC, AW, and CC performed the experiments. CL provided materials. OB, DC, CC, and AD planned and analysed the experiments. OB, DC, CC, LN, CL, and AD wrote the manuscript and planned the research project. Conflict of Interest -------------------- The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The authors wish to thank Andrew Herman and Lorena Ballesteros (Flow Cytometry Facility, University of Bristol), Christy Waterfall and Jane Coghill (Genomics Facility, University of Bristol), Tom Batstone and Alex Paterson (Bioinformatics arm of the Genomics Facility), and Dominic Alibhai and Stephen Cross (Wolfson Bioimaging Facility, University of Bristol) for the technical assistance. ^1^Simon Andrews, Brian Howard, Phil Ewels. FastQC: a quality control tool for high throughput sequence data 2015. Available online at: <http://www.bioinformatics.babraham.ac.uk/projects/fastqc/>. ^2^Illumina. Illumina Adapter Sequences 2017. Available online at: <https://support.illumina.com/content/dam/illumina-support/documents/documentation/chemistry_documentation/experiment-design/illumina-adapter-sequences-1000000002694-11.pdf>). ^3^The Jackson Laboratory. 007914 - B6.Cg-Gt(ROSA)26Sor 2019. Available online at: <https://www.jax.org/strain/007914>. ^4^National Centre for the Replacement R, Reduction of Animals in R. The 3Rs \| NC3Rs 2018. Available online at: <https://www.nc3rs.org.uk/the-3rs>. **Funding.** OB was funded as a National Eye Research Centre (NERC) Ph.D. studentship \[R103433-101\]. CL was supported by the SFB/TRR167. This work was supported by additional funding from the NERC and the Underwood Trust. Supplementary Material {#s8} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fimmu.2019.03033/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: Andrew W. Taylor, Boston University School of Medicine, United States [^2]: Reviewed by: Gianluca Matteoli, KU Leuven, Belgium; Thomas Langmann, University of Cologne, Germany [^3]: This article was submitted to Immunological Tolerance and Regulation, a section of the journal Frontiers in Immunology

1. Introduction {#s0005} =============== The word propolis has been derived from Greek in which the morpheme 'Pro' means "in front of" or "at the entrance to") and the morpheme 'polis' means 'community' or 'city' ([@b0075], [@b0170], [@b0670]) which means hive defensive substance ([@b0055], [@b0075], [@b0105], [@b0150], [@b0160], [@b0260], [@b0430], [@b0455], [@b0490], [@b0505], [@b0555], [@b0605], [@b0645]) Propolis is collected by worker-bees, from numerous plant resinous secretions ([@b0130], [@b0180]) such as mucilage, gums, resins ([@b0110]) and lattices and also from leaf buds ([@b0295], [@b0390], [@b0435], [@b0650]) of different plant species like palm, pine, alder, poplar, beech, conifer and birch ([@b0680]) and then mixed with salivary and enzymatic secretions ([@b0545]). It is also called "Bee-glue" which is a natural resin (wax-like) substance ([@b0010], [@b0615]) present in bee hives and used by Honey bees as a cementing material to close open spaces and cracks occurring in their hives ([@b0035], [@b0115], [@b0105], [@b0160], [@b0235], [@b0285], [@b0305], [@b0340], [@b0365], [@b0390], [@b0450], [@b0565], [@b0600], [@b0605], [@b0640], [@b0645]). Honey bees use propolis not only to protect their hives, by blocking the cracks, sealing the spaces and smoothening the internal walls of their hives but they also use it as an antiseptic ([@b0050], [@b0055], [@b0065], [@b0250], [@b0290], [@b0320], [@b0485], [@b0510], [@b0540], [@b0570], [@b0625]) to protect the bees larvae, honey stores and comb from microbial infections ([@b0020], [@b0220], [@b0425], [@b0495], [@b0555], [@b0585]). It is also applied to those areas of the hive where the comb is attached to an object so as to create germ-free, smooth surface ([@b0545], [@b0590]). The reason is that honey bees live together, and therefore infection occurring in one bee can spread quickly in the whole colony ([@b0075], [@b0365], [@b0535], [@b0585], [@b0635]). In addition, it is also used by honey bees to prevent decomposition of the carcass of the intruder in the hives ([@b0360], [@b0565]) and to maintain internal hive temperature ([@b0410], [@b0665]) around 35 °C ([@b0240], [@b0460]). Similarly, propolis prevents entry of water to the hive which maintains constant humidity and also serves as a control over the air flow towards the hive ([@b0025], [@b0365]). Propolis was identified and used by man since long ([@b0365], [@b0375]). Currently propolis is used as an anti-bacterial, anti-inflammatory, anti-viral ([@b0560]), anti-oxidant, anti-protozoal, anesthetic, anti-tumoural, anti-cancer, anti-fungal ([@b0485], [@b0555]), anti-septic, anti-mutagenic, anti-hepatotoxic in addition to being used for cytotoxic activity, etc. ([@b0605]). 2. History of propolis {#s0010} ====================== Propolis was used by man as a traditional medicine since 300 BC ([@b0280], [@b0540], [@b0600], [@b0620], [@b0625], [@b0630]). Researchers stated that the healing activities of propolis were identified by Roman ([@b0380]) and Greek doctors ([@b0245], [@b0255]) as well as other scientists, such as Dioscorides, Galen, Aristoteles and Pliny ([@b0160], [@b0185], [@b0215], [@b0605]). Similarly, physicians used propolis efficiently for the treatment of injuries during Anglo-Boer battle ([@b0205], [@b0545]) as well as in World War II ([@b0375], [@b0490]). Early Egyptians used propolis to preserve their corpses from decomposition ([@b0400], [@b0530], [@b0675]) and to heal wounds ([@b0240], [@b0455], [@b0545]), because they were aware of the putrefactive properties of propolis ([@b0045], [@b0160], [@b0375], [@b0555], [@b0645]). Furthermore, propolis was recognized as an anti-bacterial agent during 17th and 20th century in Europe ([@b0605], [@b0645]). In England, Propolis was recognized as a better medicine for the treatment of wounds during 17th century ([@b0365], [@b0405], [@b0415]). Similarly in China, propolis was recognized as an anti-cancer and anti-infection medicine ([@b0165]). The first scientific report about propolis, its composition and chemical actions was announced to the public in 1908 ([@b0260], [@b0310]). In 17th century used as for varnishing ([@b0260], [@b0400]). 3. Physical properties of propolis {#s0015} ================================== The color of propolis varies according to area and the plant source ([@b0035], [@b0205], [@b0375], [@b0435], [@b0530]). It melts on 60 °C to 70 °C while some of its kinds melt on 100 °C ([@b0400], [@b0490], [@b0645]). Hard at low while soft at high temperature ([@b0020], [@b0150], [@b0260], [@b0300], [@b0365], [@b0390], [@b0400], [@b0455], [@b0645]). It is extracted commercially with suitable solvents i.e. ethanol, methanol, chloroform, ether and acetone but Ethanol is the best ([@b0400], [@b0490], [@b0620]). The biological activity of propolis sample varies, due to its different geographical origin ([@b0645]). It is found commercially in the form of dentifrices, lozenges, mouth rinses, creams, gels, cough syrups, wine, cake, powder, soap, chewing gums and tablets ([@b0170], [@b0455], [@b0485], [@b0645]) as well as candies, shampoos, chocolate bars, skin lotions, toothpastes ([@b0255], [@b0670]), antiseptic mixtures ([@b0490]) and is also used for the preservation of flesh ([@b0165]). The estimated colony collection per year is 150--200 g ([@b0345], [@b0365], [@b0400]). 4. Composition of propolis {#s0020} ========================== Chemically propolis is composed of more than 180 different types of chemicals ([@b0365]) and season to season. Propolis is collected both in temperate zone and tropical zones and slightly different ([@b0475], [@b0590]). As a result more than 300 different components have been identified in propolis ([@b0045], [@b0055], [@b0130], [@b0165], [@b0200], [@b0205], [@b0220], [@b0240], [@b0260], [@b0365], [@b0485], [@b0490], [@b0545], [@b0555], [@b0605], [@b0640]). In general, propolis contains polyphenol (flavonoids, phenolic acids and esters) ([@b0305], [@b0350], [@b0360]), phenolic aldehydes and ketones, etc. ([@b0175], [@b0190], [@b0195], [@b0245], [@b0420]). The percentage of these substances is as follows: resins and vegetable balsam 50%, Bee wax 30%, pollen 5%, essential and aromatic oils 10%, and some other substances which include organic compounds as well ([@b0005], [@b0055], [@b0135], [@b0140], [@b0150], [@b0170], [@b0240], [@b0260], [@b0275], [@b0360], [@b0365], [@b0375], [@b0400], [@b0440], [@b0445], [@b0455], [@b0470], [@b0485], [@b0490], [@b0495], [@b0510], [@b0545], [@b0555], [@b0605], [@b0625], [@b0640], [@b0645]). The percentage of diverse material present in propolis depends upon the time of its collection and also on the geographical origin ([@b0025], [@b0145], [@b0190], [@b0370], [@b0580], [@b0645]). 4.1. Gross composition of propolis {#s0025} ---------------------------------- CompoundsReferences***Flavonoids, flavanones, flavones & flavonols:***\ Islapinin, Ermanin, Pectolinarigenin, Sakuranetin, Isosakuranetin, Quercetin- 3,3′-dimethyl ether, 3-acetyl pinobanksin, Betuletol, Isorhamnetin, Kaempferide, Rhamnazin, Rhamnetin, Alnusin, Alpinetin, Alnusitol, Pinostrobin, Pinocembrin, Chrysin, Tectochrysin, Acacetin, Rhamnocitrin, Quercetin, Galangin, Apigenin, Pinobanksin, Kaempferol,[@b0650]Rutin, Catechin, Luteolin, Naringenin[@b0375]***Benzoic acid and derivatives:***\ Benzoic acid, Salicylic acid, Gentisic acid, Gallic acid, Phenylmethyl ester of benzoic acid, Phenylmethyl ester of salicylic acid, Trans-coniferyl benzoate, Trans-p-coumaryl benzoate, Protocatechuic acid[@b0650]***Benzaldehyde derivatives:***\ Vanillin, Caproic aldehydes, Isovanillin p-hydroxybenzaldehyde, Protocatechualdehyde[@b0005], [@b0050], [@b0390], [@b0650]***Cinnamyl alcohol, cinnamic acid & its derivatives:***\ Cinnamyl alcohol, Hydrocaeffic acid, Isoferulic acid, Cinnamic acid methyl ester, Cinnamic acid ethyl ester, Cinnamylidene acetic acid, Cinnamic acid, Caffeic acid, Ferulic acid[@b0650]***Aliphatic hydrocorbons:***\ Eicosine, 1-octadecene, Tricosane, Pentacosane, Eicosane, Heneicosane***Sugar:***\ d-ribofuranose, d-fructose, d-glucitol, d-gulose, Talose, Sucrose, d-glucose***Vitamins:***\ B1, B2(complex), B6, C, E[@b0365]***Nicotinic acid, Pantothenic acid**Chalcones & dihydrochalcones***Alpinetin chalcone, Naringinen chalcone, Pinobanksin chalcones, Pinobanksin-3-acetate chalcone, Pinostrobin chalcone, Pinocembrin chalcones, Sakuranetin chalcone, 2′,6′,a-trihydroxy-4′-methoxy chalcone, 2′,6,dihydroxy-4′-methoxydihydro chalcone, 2′,4′,6-trihydroxydihydro chalcone[@b0390]***Amino acids:***\ Alanine, β-alanine, α-amino butyric acid, δ-amino butyric acid, Arginine, Asparagine, Aspartic acid, Cystine, Cystein, Glutamic acid, Glycine, Histidine, Hydroxyproline, Isoleucine, Leucine, Lysine, Methionine, Ornithine, Phenylalanine, Proline, Pyroglutamic acid, Sarcosine, Serine, Threonine, Tryptophane, Tyrosine, Valine***Esters:***\ Methyl palmitate, Cinnamyl-*trans*-4- coumarate, Ethyl palmitate, Stearic acid methyl ester, Phthalate ester, Benzyl benzoate, Benzyl-*trans*-4- coumarate, 3-Methyl-3-butenyl isoferulate, 3-Methyl-2-butenyl isoferulate, 3-Methyl-3-butenyl caffeate, 2-Methyl-2-butenyl caffeate, 3-Methyl-2-butenyl caffeate, Benzyl caffeate, Phenylethyl caffeate, Cinnamyl caffeate, Tetradecyl caffeate, Tetradecenyl caffeate, Tetradecenyl caffeate (isomer)^b^, Tetradecanyl caffeate^b^, Hexadecyl caffeate[@b0235]***Other acids and derivatives***Phenylmethyl ester of 14- methylpentadecanoic acid, Ethyl ester of palmitic acid, Myristic acid, Sorbic acid, Butyl-2-methylpropyle ester of Phthalic acid, Stearic acid, Methyl ester of alnustic acid[@b0650]***Alcohol, ketones, phenols and heteroaromatic compounds:***\ Benzyl alcohol, Hexadecanol acetate, Coumarine, Pterostilbene, Xanthorrhoeol, Scopoletol***Terpene, Sesquiterpene, alcohol & derivatives:***\ Geraniol, Neroledol, β-bisabolol, Guaiol, Farnisol, Dihydroeudesmol, α-acetoxybetulenol***Sesquiterpene & Triterpene hydrocorbons:***\ β-patchoulene, β-bisabolene, Squalene, β-bourbonene, Copaene, Calarene, Calamenene, Caryophyllene, Patchoulane, Selenene, Aromadendrene***Sterols & steroid hydrocarbons:***\ Cholestrilene, Cholinasterol, Stigmasterol, β-dihydrofucosterol, Lanosterol, Cholesterol***Minerals:***\ Sr, Ba, Cd, Sn, Pb, Ti, Ag, Co, Mo, Al, Si, V, Ni, Mn, CrNa, Mg, Cu, Ca, Zn, Fe, K[@b0375], [@b0460], [@b0650]***Enzymes:***\ Glucose-6-phosphatase, Acid phosphatase, Adenosine triphosphatase, Succinic dehydrogenase***Ketones:***\ Acetophenone, p-acetophenolacetophenone, Dihydroxy-acetope9i9inone, Methylacetophenone, Hept-5-en-2-one, 6-methylketone[@b0390]***Waxy acids:***\ Archid acid, Behenic acid, Cerotic acid, Lauric acid, Linoleic acid, Lignoceric acid, Montanic acid***Aliphatic acids & aliphatic esters:***\ Acetic acid, Angelic acid, Butyric acid, Crotonic acid, Fumaric acid, Isobutyric acid, Methylbutyric acid, Isobutyl acetate, Isopentyl acetate, Isopentinyl acetate***Alcohol:***\ Benzene methanol, Cinnamyl alcohol, Glycerol, α-glycerophosphate, Phenethyl alcohol, Isobutenol, Hydroquinone, Prenyl alcohol***Aliphatic acids:***\ Lactic acid, Hydroxyacetic acid, Malic acid, 5-Hydroxy-n-valeric acid, 2,3-Dihydroxypropanoic acid, Pentonic acid- 2-deoxy-3,5-dihydroxy-γ-lactone^b^, Pentonic acid- 2-deoxy-3,5- dihydroxy-γ-lactone (isomer)^b^, Succinic acid, 2,3,4,5-Tetrahydroxypentanoic acid- 1,4-lactone, 2,3,4,5-Tetrahydroxypentanoic acid- 1,4-lactone(isomer)^b^, Nonanoic acid, Palmitic acid, Oleic acid, Decanoic acid, Dodecanoic acid, Tetradecanoic acid, Heptadecanoic acid, Octadecenoic acid, Tetracosanoic acid, Eicosanoic acid, Hexacosanoic acid, 2- Hydroxy hexacosanoic acid^b^[@b0235]***Fatty acids (C7-C18 acids)**Other compounds:***\ Phosphoric acid, 1,4-Dihydroxy benzene, 4 -Hydroxy-benzaldehyde, 4-Hydrory acetophenone,\ 1,2,4-trihydroxy butane, 1,2,3-trihydroxy butanal, 1,2,3-trihydroxy butanal (isomer)^b^, Myristicin, 2,4-bis(dimethyl benzyl)-6-t- butyl phenol, 1,8-dihydroxy-3-methyl anthraquinone, Myristicin (isomer) 4.2. Geographical origin of propolis and composition {#s0030} ---------------------------------------------------- Poplar tree (*Populus nigra*), is found in North America, Europe, non-tropical regions of Asia as well as New Zealand. Propolis collected from Egypt was known to possess constituent of poplar tree as well as esters of caffeic acid and long-chain fatty alcohols including tetradecanol, hexadecanol and dodecanol ([@b0105]). Birch propolis collected from Russia in hold flavonols and Flavones (distinct from poplar propolis) of *Betula verrucosa* (plant source). Similarly, the main source of Brazilian propolis is *Baccharis dracunculifolia* leaf resin, including constituents, such as diterpenes, lignans, prenylated derivatives of *p*-coumaric acid as well as of acetophenone and flavonoids (distinct from poplar type). Brazilian propolis contains artepillin C in large amount as compared to CAPE (caffeic acid phenethyl ester) ([@b0165], [@b0255]). There are some compounds, found only in tropical zones, e.g. sesquiterpenoids including germacren d, ledol and spatulenol ([@b0105]). The main source of Cuban propolis is *Clusia rosea,* and contains polyisoprenylated benzophenone which is distinct from both European and Brazilian propolis ([@b0095], [@b0100], [@b0105]). 5. Biomedical application of propolis {#s0035} ===================================== Propolis use has great effect on human health and is used for various purposes. Nowadays, it is used as an antibacterial, antifungal, anti-inflammatory, antiviral, anesthetic, antioxidant ([@b0140], [@b0425]), antitumoural, antiprotozoal, anticancer ([@b0010], [@b0105], [@b0110], [@b0260], [@b0340], [@b0365], [@b0475], [@b0485], [@b0490], [@b0555]) antihypertensive, anticarcinogenic and anti-hepatotoxic in addition to possessing cytotoxic activity, etc. ([@b0605]). 5.1. Anti-bacterial activity of propolis {#s0040} ---------------------------------------- Propolis has a significant effect against bacteria such as *Enterococcus* spp., *Escherichia coli* and *Staphylococcus aureus* ([@b0060], [@b0285], [@b0320], [@b0355], [@b0365], [@b0395], [@b0550], [@b0575]). It also surfaced that, ethanolic extracts of propolis were more effective against gram-positive bacteria and showed limited effect against gram-negative bacteria ([@b0045], [@b0205], [@b0260], [@b0285], [@b0295], [@b0375], [@b0400], [@b0625], [@b0645]) but only stop growth by gram-negative bacteria due to high concentration of propolis ([@b0550]). The mode of action of propolis is due to the interaction between phenolic with other compounds such as pinocembrin, galangin, and pinobanksin ([@b0160], [@b0615], [@b0645]). Similarly, the antibacterial activity takes place due to its active compounds such as, aromatic compounds (caffeic acid) and flavonoids ([@b0455])*.* Moreover, propolis acts as a bactericidal agent, to stop division of bacterial cell, destroy cell wall, bacterial cytoplasm ([@b0455]) and stop protein synthesis ([@b0375], [@b0380]). The component of propolis such as Pinocembrin shows antibacterial activity towards *Streptococcus* spp. artepillin C, p-Coumaric acid and 3-phenyl-4-dihydrocinnamylocinnamic acid towards *Helicobacter pylori* and Apigenin strongly restrain bacterial glycosyltransferase ([@b0400]). Brazilian propolis was more effective against Gram-positive bacteria as compared to Gram-negative ones ([@b0520]). Propolis show antibacterial activity against some aerobic bacteria such as, *Bacillus cereus, B. subtilis, Enterococcus faecalis, Micrococcus luteus, Nocardia asteroids, Rhodococcus equi, Staphylococcus auricularis, S. epidermidis, S. capitis, S. haemolyticus, S. warnerii, S. mutans, S. hominis, Streptococcus cricetus*, *St. faecalis, St. pyogenes*, *St. pneumioniae*, *St. sobrinus* and *St. viridians* ([@b0260]). 5.2. Anti-fungal activity of propolis {#s0045} ------------------------------------- Propolis showed activity against different fungi ([@b0015], [@b0030], [@b0060], [@b0070], [@b0265], [@b0315], [@b0390]). It was investigated that, propolis inhibit the aflatoxigenic fungi, and also decreases conidial growth in *Aspergillus flavus*. Propolis from different areas show activity against *Candida guilliermondii, C. guilliermondii, C. krusei, C. albicans*. In another investigation, a French propolis was effectively used against human fungal pathogen *C. albicans, C. glabrata, Aspergillus fumigates*. A constituent of propolis called pinocembrin shows activity against *Penicillium italicum*, which stops mycelial growth and acting on the pathogen respiration and energy homeostasis leading to the rupturing of cell membrane and metabolism disorder ([@b0555]). Propolis also showed fungicidal effect against yeast ([@b0550]). The presence of flavonoids in propolis shows fungicidal activity against *C. pelliculosa, C. parapsilosis,* and *Pichia ohmeri, C. famata, C. glabrata* ([@b0645]). Australian propolis showed antifungal activity towards *C. albicans*, due the greater amount of pinocembrin ([@b0130]). It was reported that, constituents of propolis such as, 3-acetylpinobanksin, pinobanksin-3-acetate, pinocembrin, p-coumaric acid and caffeic acid out of 26 or more constituents show anti-fungal activity. Furthermore, caffeic acid showed antimycotic activity towards *Helminthosponum carbon* ([@b0435]). Propolis showed good result against *Mycobacteria*, *Candida, Trichophyton*, *Fusarium* as well as other skin infecting fungi ([@b0260]). 5.3. Anti-tumoural activity of propolis {#s0050} --------------------------------------- The components of propolis possess anti-tumoural property ([@b0125], [@b0155], [@b0330], [@b0335], [@b0595], [@b0630]) and components such as caffeic acid phenethyl ester (CAPE) ([@b0160]) and artepillin C were investigated, and found to possess anti-tumoural effects ([@b0165]). These compounds of propolis are involved in the cell-cycle arrest, inhibition of matrix metalloproteinases, anti-angiogenesis effect and also inhibit disease transferring from one body part to another ([@b0555]). Propolis has the ability to stop DNA synthesis in tumor cells, has the property to cause aging of tumor cells (Apoptosis) and possesses the capability to put into action the white blood cells for generating those agents which are able to regulate the function of B, T and natural killer cells respectively ([@b0515], [@b0645]). Other compounds such as galangin, cardanol, nemorosone and chrysin are involved to prevent the rapid division of tumor cells ([@b0555]). The cytotoxic activity of natural killer cell (NK) against murine lymphoma increased with the use of propolis for 3 days ([@b0260], [@b0545]). The presence of tumor suppresser proteins in Caffeic acid phenethyl ester causes the C6 glioma cells apoptosis ([@b0205], [@b0555], [@b0655]). Caffeic acid and esters as well as diterpenoids and phenolic compounds had the destructive capability against tumor cells. The anti-tumor effect of propolis is due to its polyphenolic constituents combined function ([@b0545]). Decrease in the production of glutathione in tumor cell due to radiations, is consequently fulfilled by propolis, as the synthesis of glutathione in haematopoietic tissues ([@b0170]). Turkish propolis acts as an anti-tumor by enhancing program cell death it also showed retardation of leucine, thymidine and uridine from becoming cancer causing cells, by restraining the synthesis of DNA ([@b0655]). 5.4. Anti-protozoal activity of propolis {#s0055} ---------------------------------------- Antiprotozoal activity of propolis against many protozoan that cause diseases in human and other animals such as, giardiasis ([@b0270]), chagas disease, leishmaniasis ([@b0225]), trichomoniasis, toxoplasmosis ([@b0205], [@b0515], [@b0610]) has been reported ([@b0030]). However propolis showed antiprotozoal activity against *Leishmania donovani, Trypanosoma cruzi, Giardia lamblia, Trichomonas vaginalis*, *Toxoplasma gondii* and *G. duodenalis* ([@b0075], [@b0260], [@b0645]). Some components of propolis showed antimicrobic activity such as, caffeic acid, chrysin, moronic acid, protocatechuic acid, p-coumaric acid, apigenin and other constituents as terpenoids, esters and phenols ([@b0555]). Propolis show better activity (coccidostatic) toward *Chilomonas paramecium* ([@b0390]). Dimethyl-sulphoxide extract and ethanolic extracts of propolis were more effective toward *Trypanosoma cruzi* while showed killing effect against *Trichomonas vaginalis* ([@b0375]). [@b0385] separated 4 phenolic compounds from Brazilian propolis such as, 2,2-dimethyl-6-carboxyethenyl-2H-1-benzopyran, 3,5-diprenyl-4-hydroxycinnamic acid, 3-(2,2-dimethyl-8-prenylbenzopyran-6-yl) propenoic acid and 3,5-diprenyl-4-hydroxycinnamic acid, showing anti-microbial activity toward *Trypanosoma cruzi* ([@b0130], [@b0385])*.* 5.5. Anti-inflammatory activity of propolis {#s0060} ------------------------------------------- Propolis possess the anti-inflammatory capability due to the presence of flavonoids ([@b0210], [@b0215], [@b0230], [@b0265], [@b0325], [@b0380], [@b0160], [@b0400], [@b0455], [@b0490], [@b0640]). It controls NADPH-oxidase ornithine decarboxylase, myeloperoxidase activity, hyaluronidase from guinea pig mast cells and tirosine-protein kinase ([@b0375]). The mode of action of these compounds is that, to restrain leukotrienes and prostaglandins production by white blood cells ([@b0380]) and retarding myeloperoxidase activity, ornithine decarboxylase, tyrosine-protein-kinase and NADPH-oxidase ([@b0490]). CAPE and galangin, both are components of poplar propolis demonstrated the anti-inflammatory property and inhibit carrageenan pleurisy, carrageenan oedema and helpful arthritis inflammations in rats ([@b0645]). The effect of Brazilian propolis and Chinese on the pathogenesis of collagen-induced arthritis in mice was also reported ([@b0555]). Propolis regulate those inflammatory substances which are produced in the cell, as a result of pressures, poisonous material or pathogenicity ([@b0350]). 5.6. Hepatoprotective activity of propolis {#s0065} ------------------------------------------ Propolis acts as a hepatoprotective agent ([@b0090], [@b0120], [@b0325], [@b0465], [@b0480], [@b0525], [@b0660]). It increases the glutathione level while stop lipid peroxidation and oxidized glutathione level. Consequently, propolis increase antioxidant activity against mercury induced-toxicity and acts as a hepatoprotective agent. Studies also showed that, extract of propolis possess protective role against CCL4-entice hepatorenal oxidative stress and resultant injury ([@b0645]). Propolis showed hepatoprotective effect toward liver damage in rats caused due to allyl alcohol CCL4 and paracetamol ([@b0160], [@b0615], [@b0645]). [@b0120] isolated phenolic components as well as diterpenic acids from Brazilian propolis, showing hepatoprotective property ([@b0100], [@b0120]). 5.7. Dental action of propolis {#s0070} ------------------------------ Ethanolic extract of propolis collected from four different regions of Brazil and Turkey, were used against different anaerobic bacterial strains, such *as Actinomyces naeslundii, Porphyromonas gingivalis, Fusobacterium nucleatum, Veillonella parvula, Lactobacillus acidophilus, Peptostreptococcus anaerobius, Peptostreptococcus micros, Prevotella oralis* and *Prevotella melaninogenica* gave result of low inhibitive and also low bacteriocidal activity through agar dilution method. Due to the presence of flavonoids, such as galangine, chrysin, pinobanksin, quercetin, naringenin, galangine and aromatic acids present in propolis are more effective against mouth abnormalities (especially dental diseases) ([@b0205], [@b0555]). It had significant role in dental pulp repairing ([@b0170], [@b0365]). The combination of ethanolic extract of propolis with mouthwash and toothpastes enhanced the prevention of microbial infection as well as to treat inflammation of gums ([@b0205], [@b0460]). 5.8. Anti-oxidant activity of propolis {#s0075} -------------------------------------- It was noted that, propolis had an antioxidant property due to its components galangin and pinocembrin ([@b0040], [@b0055], [@b0230], [@b0380], [@b0400], [@b0500], [@b0640]). Due to higher polyphenols contents, the aqueous extract of propolis was more effective than ethanolic extracts. Galangin showed more activity in both extracts as compared to pinocembrin, due to structural difference in both ([@b0645]). The antioxidants had the ability to refuse free radical also prevent vitamin C, lipids and other compounds from destruction or oxidization. Because free radical and other factors are the primary cause of aging of cells and deterioration in such states as parkinson's disease, alzheimer's disease, arthritis, cancer, diabetes, cardiovascular diseases ([@b0365]) and also insufficient liver function. Components of propolis such as vanillin and phenolic acids had the capability to enter into the epidermis as well as dermis and protect them from free radicals, produced due to radiations or before maturation of dermal cells aging ([@b0350]). The mechanism of anti-oxidant property of propolis is due to phenolic compounds which donate hydrogen ions to free radicals to protect cell from oxidation reactions and also food storage from oxidation and poisoning. Propolis had the capability to remove free radicals, which are the primary cause of lipids, nucleic acids and proteins oxidation ([@b0170]). Portuguese Propolis had the antioxidant property and prevent lipid peroxidation in red blood cells of human ([@b0675]). 5.9. Anti-viral activity of propolis {#s0080} ------------------------------------ It is known that, propolis showed antiviral activity ([@b0080], [@b0085]), by inhibiting the virus entry into the cells, create disturbance in viral replication which cause destruction of RNA before or after its (RNA) release in the cells ([@b0555]). Among other factors, propolis showed antiviral capability against genital herpes infection (HSV-2) ([@b0365]). Flavonoids which include kaempferol, acacetin, quercetin, galangin and chrysine were reported as a cytotoxic ([@b0390]). Some other researcher reported a compound, separated from poplar propolis called 3-methyl-but-2-enyl caffeate inhibit titration and DNA synthesis of herpes simplex virus (type1) ex vivo. Another compound called isopentyl ferulated showed activity against influenza virus A1 Honey Kong (H3N2) ex vivo ([@b0375]). Propolis showed antiviral activity against avian influenza virus, rift valley fever virus, newcastle disease virus, herpes bursal disease virus and influenza virus ([@b0235]). 5.10. Wound healing activity of propolis {#s0085} ---------------------------------------- Propolis components also has therapeutic capability of propolis on tissue repairing and regeneration of injury ([@b0365]). These are due to its immunomodulatory, anti-inflammatory and antimicrobial features ([@b0400], [@b0555]). It was also noted that, propolis lower the quantity of free radicals in inflammatory injury and to increase collagen and its constituents development ([@b0350], [@b0400]). It accelerate different enzymatic reactions, metabolism of cells, blood circulation and also formation of collagen fibers, due to the presence of bioflavonoids, arginine, vitamin C, provitamin A, B complex as well as some minerals ([@b0455]). 5.11. Anti-cancer activity of propolis {#s0090} -------------------------------------- Due to its antioxidant activity, propolis (ethanolic extract) of Indian stingless bees had capability of anticancer against four various cancer cell lines at various concentrations; which gave result of apoptosis and cytotoxic of these cancer cells ([@b0350]). Flavonoids in propolis stop breast cancers, lung cancer, oral cancer as well as esophagus, stomach, colorectal, prostate, and skin cancer ([@b0400]). Brazilian propolis had the property of angiogenesis as well as stops the increase number of human umbilical vein endothelial cells ([@b0675]). Ethanolic extract of Brazilian propolis had the anti-cancer capability, examined upon 1,2 dimethylhydrazine which causing colon carcinogenesis in mice ([@b0655]). The water extract of propolis from Thailand was examined by researchers, showed greater anti-cancer activity against colon carcinoma cell line SW620 as compare to methanolic extract of propolis ([@b0655]). Ethanolic extract of propolis had the cytotoxicity against HT-29 colon adenocarcinoma cells as well as toward HT-1080 human fibrosarcoma, however demonstrate no cytotoxicity toward typical human skin fibroblasts ([@b0655]). 6. Conclusion {#s0095} ============= Propolis; a honey-bee hive product, possesses a wide range of pharmacological potentials including anti-bacterial, anti-fungal, anti-protozoal, hepatoprotective, anti-oxidant, anti-inflammatory, anti-viral, anti-cancer and anti-tumor properties. Besides, the addition of ethanolic extract of propolis in the composition of mouthwashes and toothpastes enhances the prevention of microbial infection and is effective in the treatment of gums inflammation. Moreover, the presence of bioflavonoids, arginine, vitamin C, provitamin A, B complex along with some minerals possesses wound healing property and therefore enhances injury cure. Instead of individual component, there may be combined action, which leads propolis to have diverse biological performance. Finally, the development of new propolis compounds from propolis coming from diverse geographical origins is vital in controlling various pathogenic diseases. The current literature review suggests that propolis may be explored further for its potential properties against human pathogen. Peer review under responsibility of King Saud University.

Introduction ============ Kefir is an acidic-alcoholic fermented milk product with little acidic taste and creamy consistency that was originated in the Balkans, in Eastern Europe, and in the Caucasus ([@B23]; [@B72]). Kefir can be produced by fermenting milk with commercial freeze-dried kefir starter cultures, traditional kefir grains, and the product that remains after the removal of kefir grains ([@B5]). Kefir grains are a kind of yogurt starter, which are white to yellow -- white, gelatinous, and variable in size (varying from 0.3--3.5 cm in diameter) and are composed by a microbial symbiotic mixture of lactic acid bacteria (10^8^ CFU/g), yeast (10^6^--10^7^ CFU/g), and acetic acid bacteria (10^5^CFU/g) that stick to a polysaccharide matrix ([@B29]; [@B14]). After successive fermentations, kefir grains can break up to new generation grains, which have the same characteristics as the old ones ([@B24]). Commercial kefir is produced by two methods: The "Russian method" and the pure cultures. In the "Russian method" kefir is produced on a larger scale, using a series fermentation process, beginning with the fermentation of the grains and using the percolate. The other method employs pure cultures isolated from kefir grains or commercial cultures ([@B48]). Also, the industrial or commercial process uses direct-to-vat inoculation (DVI) or direct-to-vat set (DVS) kefir starter cultures. In addition, *Bifidobacterium* sp., *Lactobacillus* sp. and probiotic yeast (*Saccharomyces boulardii*) may be used as adjunct cultures when blended with kefir grains or kefir DVI cultures ([@B87]). On the other hand, whey may be a practical base for kefir culture production, and fermented whey has shown to be a suitable cryoprotective medium during freeze-drying. The freeze-dried culture retains a high survival rate and shows good metabolic activity and fermentation efficiency, indicating a good potential for its use as a value-added starter culture in dairy technology. All of these studies have shown promising perspectives for the application of kefir grains in whey valorization strategies ([@B5]; [@B10]). Traditionally, kefir is manufactured using cow, ewe, goat, or buffalo milk. However, in some countries, animal milk is scarce, expensive, or minimally consumed due to dietary constraints, preferences, or religious customs. Therefore, there have been many attempts to produce kefir from a variety of food sources such as soy milk ([@B7]). Historically, kefir has been linked with health, for example, in Soviet countries, kefir has been recommended for consumption by healthy people to restrain the risk of some diseases ([@B71]; [@B75]; [@B22]). The consumption of this fermented milk has been related to a variety of health benefits ([@B79]; [@B54]; [@B67]) not only linked to its microflora, but also due to the presence of some metabolic products as organic acids ([@B28]; [@B37]). In addition, kefir cultures have the ability to assimilate cholesterol in milk ([@B90]). On the other hand, there is a growing commercial interest in using kefir as a suitable food matrix for supplementation with health-promoting bacteria. Kefir may not only be a natural probiotic beverage, but also acts as an effective matrix for the delivery of probiotic microorganisms ([@B78]; [@B55]; [@B59]). In kefir grains the main polysaccharide is kefiran, which is a heteropolysaccharide composed by equal proportions of glucose and galactose and is mainly produced by *Lactobacillus kefiranofaciens* ([@B95]). It has been demonstrated that kefiran improves the viscosity and viscoelastic properties of acid milk gels ([@B65]), and is able to form gels that have interesting viscoelastic properties at low temperatures, because of that, kefiran can also be used as an additive in fermented products. Besides, kefiran can enhance the rheological properties of chemically acidified skim milk gels increasing their apparent viscosity ([@B94]). Compared with other polysaccharides, kefiran has outstanding advantages such as antitumor, antifungal, antibacterial properties ([@B9]; [@B81]) immunomodulation or epithelium protection ([@B72]), anti-inflammatory ([@B68]), healing ([@B67]), and antioxidant activity ([@B14]). This review presents the most recent advances about kefir and kefiran, their production and microbial cultures involved, biological activities and potential applications in health and food industries. Microbial Composition of Kefir Grains and Kefir =============================================== Kefir grains have a complex composition of microbial species such as the predominance of lactic acid bacteria, acetic bacteria, yeasts, and fungi ([@B38]; [@B61]). This microbial species are classified into four groups: homofermentative and heterofermentative lactic acid bacteria and lactose and non-lactose assimilating yeast ([@B10]). In that way, *Lactobacillus paracasei* ssp. *paracasei*, *Lactobacillus acidophilus, Lactobacillus delbrueckii* ssp. *bulgaricus*, *Lactobacillus plantarum*, and *L. kefiranofaciens* are predominant species. However, these species represent only 20% of the *Lactobacillus* in the final fermented beverage, with the remainder consisting of *Lactobacillus kefiri* (80%; [@B93]; [@B96]). *Acetobacter aceti* and *A. rasens* have also been isolated, such as the fungus *Geotrichum candidum*. More than 23 different yeast species have been isolated from kefir grains and from fermented beverages of different origins. However, the predominant species are *Saccharomyces cerevisiae, S. unisporus, Candida kefyr*, and *Kluyveromyces marxianus* ssp. *marxianus* ([@B85]; [@B18]; [@B96]; **Table [1](#T1){ref-type="table"}**). ###### Microbial compositions found in kefir and kefir grains of different origins. Microorganism Source -- Country Reference ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------- --------------------------------- *Lactobacillus kefir, Lactobacillus kefiranofaciens, Lactobacillus paracasei, Lactobacillus plantarum, Lactococcus lactis* ssp. *lactis, Kluyveromyces marxianus, Lactobacillus parakefir, Saccharomyces cerevisiae, Saccharomyces unisporus, Leuconostoc mesenteroides, Acetobacter* sp., *Saccharomyces* sp., *Lactococcus lactis* ssp. *lactis* biovar *diacetylactis*, *Lactococcus lactis, Lactobacillus kefiri, Lactobacillus parakefiri* Kefir grains and beverage -- Argentina [@B28]; [@B50]; [@B34]; [@B18]. *Lactobacillus kefiri, Lactobacillus kefiranofaciens, Leuconostoc mesenteroides, Lactococcus lactis, Lactococcus lactis* ssp. *cremoris, Gluconobacter frateurii, Acetobacter orientalis, Acetobacter lovaniensis, Kluyveromyces marxianus, Naumovozyma* sp., *Kazachastania khefir* Kefir grains and beverage -- Belgium [@B45] *Lactobacillus kefiri, Lactobacillus kefiranofaciens, Leuconostoc mesenteroides, Lactococcus lactis, Lactobacillus paracasei, Lactobacillus helveticus, Gluconobacter japonicus, Lactobacillus uvarum, Acetobacter syzygii, Lactobacillus satsumensis, Saccharomyces cerevisiae., Leuconostoc* sp., *Streptococcus* sp., *Acetobacter* sp., *Bifidobacterium* sp., *Halococcus* sp., *Lactobacillus amylovorus, Lactobacillus buchneri, Lactobacillus crispatus, Lactobacillus kefiranofaciens* ssp. *kefiranofaciens, Lactobacillus kefiranofaciens* ssp. *kefirgranum, Lactobacillus parakefiri* Kefir grains -- Brazil [@B56]; [@B47]; [@B96] *Lactobacillus brevis, Lactobacillus delbrueckii* ssp. *bulgaricus, Lactobacillus helveticus, Streptococcus thermophilus, Lactobacillus casei* ssp. *pseudoplantarum, Kluyveromyces marxianus* var. *lactis, Saccharomyces cerevisiae, Candida inconspicua, Candida maris, Lactobacillus lactis* ssp. *lactis* Kefir grains and beverage -- Bulgaria [@B73] *Lactobacillus paracasei, Lactobacillus parabuchneri, Lactobacillus casei, Lactobacillus kefiri, Lactococcus lactis, Acetobacter lovaniensis, Kluyveromyces lactis, Kazachstania aerobia, Saccharomyces cerevisiae, Lachancea meyersii* Kefir beverage -- Brazil [@B53] *Lactobacillus kefiranofaciens, Leuconostoc mesenteroides, Lactococcus lactis, Lactobacillus helveticus, Kluyveromyces marxianus, Saccharomyces cerevisiae, Pseudomonas* sp.*, Kazachstania unispora, Kazachstania exigua, Lactobacillus kefiri, Lactobacillus casei, Bacillus subtilis, Pichia kudriavzevii, Leuconostoc lactis, Lactobacillus plantarum, Acetobacter fabarum, Pichia guilliermondii, Lactococcus* sp., *Lactobacillus* sp., *Acetobacter* sp., *Shewanella* sp., *Leuconostoc* sp., *Streptococcus* sp, *Acinetobacter* sp., *Pelomonas* sp., *Dysgonomonas* sp., *Weissella* sp., *Shewanella* sp. Kefir grains (Tibet)-- China [@B38]; [@B24], [@B25] *Acetobacter acetic, Enterococcus faecalis, Enterococcus durans, Lactococcus lactis* ssp. *cremoris, Leuconostoc pseudomesenteroides, Leuconostoc paramesenteroides, Lactobacillus brevis, Lactobacillus acidophilus, Saccharomyces* sp., *Brettanomyces* sp., *Candida* sp., *Saccharomycodes* sp., *Acetobacter rancens* Kefir beverage -- China [@B89] *Lactobacillaceae* and *Streptococcaceae* Kefir grains and beverage -- Ireland [@B19] *Lactobacillus kefiranofaciens, Dekkera anomala, Streptococcus thermophilus, Lactococcus lactis, Acetobacter* sp.*, Lactobacillus lactis, Enterococcus* sp., *Bacillus* sp., *Acetobacter fabarum, Acetobacter lovaniensis, Acetobacter orientalis* Kefir grains -- Italy [@B27] *Leuconostoc* sp.*, Lactococcus* sp.*, Lactobacillus* sp., *Lactobacillus plantarum, Zygosaccharomyces* sp.*, Candida* sp.*, Candida lambica, Candida krusei, Saccharomyces* sp., *Cryptococcus* sp. Kefir grains and beverage -- South Africa [@B86] *Lactobacillus* sp., *Leuconostoc* sp., *Lactococcus* sp., *Zygosaccharomyces* sp., *Candida* sp., *Saccharomyces* sp. Kefir grains -- South Africa [@B85] *Lactobacillus kefiri, Lactobacillus kefiranofaciens, Leuconostoc mesenteroides, Lactococcus lactis, Escherichia coli, Pseudomonas* sp., *Saccharomyces turicensis*, Kefir grains -- Taiwan [@B88]; [@B12]; [@B80]; *Lactobacillus kefiri, Leuconostoc mesenteroides, Lactococcus lactis, Streptococcus thermophilus, Lactobacillus kefiranofaciens, Lactobacillus acidophilus* Kefir grains and beverage -- Turkey [@B33]; [@B40] *Lactobacillus helveticus, Lactobacillus buchneri, Lactobacillus kefiranofaciens, Lactobacillus acidophilus, Lactobacillus helveticus, Streptococcus thermophilus, Bifidobacterium bifidum, Kluyveromyces marxianus* Kefir grains -- Turkey [@B43]; [@B58] *Lactococcus cremoris, Lactococcus lactis, Streptococcus thermophilus, Streptococcus durans* Kefir beverage -- Turkey [@B93] The microbial composition may vary according to kefir origin, the substrate used in the fermentation process and the culture maintenance methods. Tibetan kefir, which is used in China, is composed of *Lactobacillus, Lactococcus*, and yeast. Additionally, acetic acid bacteria have been identified in Tibetan kefir, depending on the region in China from where it was obtained ([@B24]), additionally, Tibetan kefir composition differs from that of Russian kefir, Irish kefir, Taiwan kefir, Turkey fermented beverage with kefir; however, it is known that this microbial diversity is responsible for the physicochemical features and biological activities of each kefir ([@B38]; [@B39]; [@B24]; [@B2]). [@B80] examined a section of a whole kefir grain and found in the outer layer of the grain, lactococci, and yeasts, and, in the inner layer of the grain, the quantity of lactobacilli were much higher and more yeasts cells were found. There are little information about the mechanism of grain formation, so the same authors, proposed a hypothesis to explain that. "Initially, *Lactobacillus kefiranofaciens* and *Saccharomyces turicensis* start to auto-aggregate and co-aggregated to small granules." The aggregation is enhanced when the pH drops. The biofilm producers, *Lactobacillus kefiri*, *Kluyveromyces marxianus* HY1, and *Pichia fermentans* HY3 then adhere to the surface of these small granules due to their cell surface properties and their strong aggregation ability, which gives rise to thin biofilms. After biofilm formation, the kefir yeasts and *Lactobacillus* continue to co-aggregated with the granule strains and associate with the granule biofilm to become a three dimensional microcolony. As the cell density due to the growth of kefir yeasts and *Lactobacillus* increases, cells and milk components that are present in the liquid phase accumulate on the granule surface and the kefir grains are formed. There is a symbiotic relation between the microorganisms present in kefir grains, wherein the bacteria and yeast survive and share their bioproducts as power sources and microbial growth factors. This microorganism association is responsible for lactic and alcoholic fermentation ([@B86]; [@B80]; [@B34]). After receiving its actual/present denomination, some of the microorganisms isolated and identified in kefir cultures were classified using the product name, as in *Lactobacillus kefiri, L. kefiranofaciens, L. kefirgranum, Lactobacillus parakefir*, and *Candida kefyr* ([@B88]; [@B46]; [@B89]; [@B43]). **Table [1](#T1){ref-type="table"}** demonstrates the microbial composition, which has been isolated from kefir and kefir grains of different origins. Biological Activity of Kefir ============================ Due to its composition, kefir is mainly considered a probiotic resource ([@B58]). "Probiotics are microbial cell preparations or components of microbial cells with a beneficial effect on the health of the host" ([@B51]). Some studies suggest that probiotic bacteria in kefir consumers' gut are abundant and are correlated with health improvement ([@B1]; [@B98]); in that way, it had been demonstrated that the cell-free fraction of kefir enhances the ability to digest lactose relieving symptoms ([@B21]; [@B66]). Another reason for the increased interest in probiotic strains from kefir is its capacity to lower cholesterol levels. There are different ways in which bacteria can alter serum cholesterol: (i) through the binding to and absorption into the cell before it can be absorbed into the body; (ii) producing free and deconjugating bile acids; (iii) inhibiting the enzyme HMG-CoA reductase ([@B90]). The microorganisms in the kefir grains produce lactic acid, antibiotics and bactericides, which inhibit the development of degrading and pathogenic microorganisms in kefir milk ([@B49]). Kefir acts against the pathogenic bacteria *Salmonella, Helicobacter, Shigella, Staphylococcus, Escherichia coli, Enterobacter aerogenes, Proteus vulgaris, Bacillus subtilis, Micrococcus luteus, Listeria monocytogenes, Streptococcus pyrogenes*, ([@B51]), *Streptococcus faecalis* KR6, *Fusarium graminearum* CZ1 ([@B37]), and the fungus *Candida albicans*. On the other hand, it has been demonstrated that a mixture of kefir isolated bacteria and yeast is able to prevent diarrhea and enterocolitis triggered by *Clostridium difficile* ([@B6]). Besides, kefir showed good efficacy in inhibiting spore formation and aflatoxin B1 produced by the fungus *Aspergillus flavus*, which is a toxic compound formed either in the field or during food storage. Therefore, kefir appears as a promising safe alternative natural food preservative offering protection against intoxication with aflatoxin B1 ([@B37]). It had been proved that many species of lactobacilli present in kefir have S-layer proteins. Surface layers (S-layers) can be aligned in unit cells on the outermost surface of many prokaryotic microorganisms ([@B57]). It has been demonstrated that these S-layer proteins can apply a protective action inhibiting the grown of *Salmonella enterica* serovar *Enteritidis* in Caco-2 cells, and also have the ability to antagonize the effects of toxins from *Clostridium difficile* on eukaryotic/eukaryotic cells *in vitro* ([@B8]). However, there are other important bioactivities that have been tested with kefir grains, the cell-free fraction of kefir or acid lactic bacteria isolated from kefir, such as antitumoral ([@B26]), anti-inflammatory ([@B17]), antimicrobial ([@B3]) immunoregulatory ([@B35]), antiallergenic ([@B83]), wound healing ([@B36]), antidiabetic ([@B92]) antimutagenic ([@B32]), and antigenotoxic ([@B31]). In that way, it had been demonstrated that kefir cell-free fraction has antiproliferative effects on human gastric cancer SGC7901 cells ([@B26]), colon adenocarcinoma cells ([@B41]), HuT--102 malignant T lymphocytes, sarcoma 180 in mice, Lewis lung carcinoma and human mammary cancer ([@B66]), and reduce oxidative stress ([@B63]). Another study has shown that suspensions after 24 h fermentation and mechanically disintegrated kefir grains cause a significant inhibition of granuloma tissue formation and a 43% inhibition of the inflammatory process ([@B17]). Nevertheless, there are other important studies performed with some microorganisms isolated from different types of kefir. Some microorganisms with their biological activities and origin are shown in **Table [2](#T2){ref-type="table"}**. ###### Kefir microorganisms and their biological activities. Organism of interest Origin Biological activity Reference ---------------------------------------- -------------------------------- ---------------------------------------------------------------------------------------------------------------------------------- ---------------- *Lactobacillus plantarum* MA2 Tibetan kefir Hypocholesterolemic effect [@B90] *Lactobacillus plantarum* Lp27 Tibetan kefir Inhibited cholesterol absorption [@B91] *Lactobacillus plantarum* CIDCA 83114 Kefir grains -- Argentina Inhibit the growth of *Shigella sonnei in vitro* and also the cytotoxicity of *C. difficile* toxins on eukaryotic cells [@B6] *Lactobacillus kefir* CIDCA 8348 Kefir grains -- Argentina Inhibit the growth of *Shigella sonnei in vitro* and also the cytotoxicity of *C. difficile* toxins on eukaryotic cells [@B6] *Lactobacillus plantarum* ST8KF Kefir grains -- South Africa Bactericida effect against: *Lactobacillus casei*, *Lactobacillus salivarius*, *Lactobacillus curvatus*, *Listeria innocua* [@B62] *Lactobacillus kefiranofaciens* K1 Kefir grains -- Taiwanese milk Antiallergenic effect [@B12]; [@B83] *Lactobacillus kefiranofaciens* M1 Kefir grains -- Taiwanese milk Immunoregulatory effects -- anticolitis effect [@B35]; [@B13] *Lactobacillus lactis* CIDCA 8221 Kefir grains -- Argentina Inhibit the growth of *Shigella sonnei in vitro* and also the cytotoxicity of *Clostridium difficile* toxins on eukaryotic cells [@B6] *Kluyveromyces marxianus* CIDCA 8154 Kefir grains -- Argentina Inhibit the growth of *Shigella sonnei in vitro* and also the cytotoxicity of *Clostridium difficile* toxins on eukaryotic cells [@B6] *Saccharomyces cerevisiae* CIDCA 8112 Kefir grains -- Argentina Inhibit the growth of *Shigella sonnei in vitro* and also the cytotoxicity of *Clostridium difficile* toxins on eukaryotic cells [@B6] *Lactobacillus lactis* ssp. *cremoris* Kefir grains -- India Activity against food spoilage bacteria [@B64] Source: Soccol et al., 2014 . Kefiran, A Potential Exopolysaccharide ====================================== The increased search for natural polysaccharides has been very significant due to their use in the food, pharmaceutical, and cosmetic industries as additives, bio-absorbents, metal removal agents, bioflocculants, and medicine delivery agents, among other functions ([@B16]; [@B84]; [@B4]). Many microorganisms, such as bacteria, fungi, and weeds, have the capacity/ability to synthesize and excrete extracellular polysaccharides, and these polysaccharides can be either soluble or insoluble ([@B82]; [@B4]). The polysaccharides that are commonly used as food additives are xanthan, dextran, gellan, and alginates, while the exopolysaccharides (EPSs) produced by lactic acid bacteria show good physicochemical characteristics for their use as food additives. In addition to these characteristics, EPSs are obtained from microorganisms classified as GRAS (generally recognized as safe), such as lactic acid bacteria ([@B81]; [@B70]; [@B4]). Many reports have demonstrated that the quantity and properties of EPSs depend on the microorganisms used in the fermentation process and on the fermentation conditions and the composition of the culture media ([@B42]). EPSs have physicochemical and rheological properties that make them suitable as additives, which can be used as stabilizers, emulsifiers, gelling agents, and viscosity improvers. Additionally, EPSs possess biological properties suggesting their use as antioxidants, antitumor agents, antimicrobial agents, and immunomodulators, among other roles ([@B76]; [@B5]; [@B60]). The EPS kefiran is produced by *Lactobacillus kefiranofaciens* ([@B44]; [@B82]) from kefir grains, which are composed of proteins, polysaccharides, and a complex symbiotic microbial mixture ([@B86]; [@B38]). These microorganisms grow in kefiran, which is a polysaccharide matrix consisting of glucose and galactose. Despite good kefiran production by *L. kefiranofaciens* alone, it has been observed that the addition of *Saccharomyces* sp. to the culture improves the net quantity of kefiran, illustrating the importance of the symbiosis between the bacteria and yeast that are present in kefir ([@B11]). Lactic acid bacteria can synthesize homopolysaccharides or heteropolysaccharides. The synthesized homopolysaccharides are glucans or fructans, which are composed of only one type of monosaccharide (glucose or fructose, respectively; [@B77]; [@B4]), whereas the heteropolysaccharides contain different types of monosaccharides in different proportions (mainly glucose, galactose, and rhamnose), ([@B15]; [@B69]). Similarly to lactic acid bacteria, *Lactobacillus* sp. also produces glucan and fructan. The homopolysaccharides show a much higher performance compared with heteropolysaccharide production ([@B84]; [@B4]). The heteropolysaccharides excreted by *Lactobacillus delbrueckii*, *Lactobacillus bulgaricus*, *Lactobacillus rhamnosus*, and *Lactobacillus helveticus* contain galactose, glucose, and rhamnose as the main monosaccharides, with other monosaccharides being present in smaller concentrations. They are also highly branched with different types of linkages, and their denominations are complex and generally dependent on the main monosaccharide ([@B15]; [@B4]). *Lactobacillus plantarum* isolated from Tibetan kefir excretes EPS classified as heteropolysaccharides composed of galactose, glucose, and mannose. This EPS has the capacity/ability to reduce blood cholesterol and form a biofilm shape ([@B97]; [@B82]). Kefiran is an EPS classified as a heteropolysaccharide comprising glucose and galactose in high concentrations, and it is classified as a water-soluble glucogalactan, which makes it suitable to be used as an additive ([@B81], [@B82]). Kefiran has excellent rheological properties and can significantly improve the viscosity of lacteous products by favoring and maintaining gel properties and avoiding the loss of water during storage ([@B65]). With respect to the biological activity of kefiran, several studies have demonstrated that this EPS can be used as a nutraceutical, as described in **Table [3](#T3){ref-type="table"}**. ###### Biological activity of kefiran. Exopolysaccharide Biological activity Reference ------------------- ----------------------------------------------------------------------------------------------------------------------------- ---------------- Kefiran Reduction of blood pressure induced by hypertension [@B52] Favors the activity of peritoneal macrophages Increase in peritoneal IgA [@B20] Antitumoral activity [@B49] Antimicrobial activity [@B67] Modulation of the intestinal immune system and protection of epithelial cells against *Bacillus cereus* exocellular factors [@B55]; [@B60] The first study about kefiran structure was published by [@B44], who proposed a structure composed of two units: kefiran (polysaccharide) and kefirose (pentasaccharide). Then, some authors analyzed the polysaccharide structure with current techniques such chromatography and infrared spectroscopy ([@B81]; [@B14]) and nuclear magnetic resonance (NMR; [@B30]). The kefiran structure, according to them, is shown in **Figure [1](#F1){ref-type="fig"}**. {#F1} Kefir-Based Products ==================== Nowadays, the interest in developing functional foods is increasing because people want to improve their health and prevent diseases. Keeping in mind that kefir is a beverage with high probiotic activity, among other bioactivities, new companies are emerging around the world. One of the biggest kefir companies known is Lifeway, which started in 1986; their products can be obtained in the United States, Canada, and Great Britain, all of them based in kefir beverages, frozen, and cheese. Other companies are Evolve Kefir with its principal product, a smoothie; Wallaby Yogurt Company with Low Fat Kefir; and CocoKefir LLC, which provides drinks/beverages based mainly on coconut water cultured with a comprehensive blend of probiotics. **Table [4](#T4){ref-type="table"}** summarizes the products provided these companies with some general information about each one. ###### Marketed kefir-based products and their information. ---------------------------------------------------------------------------------------------------------------------------- Companies Product General information ------------------ ----------------------------- --------------------------------------------------------------------------- Lifeway\ Low Fat Kefir\ All-natural\ • United States\ \ 99% lactose-free\ • Canada\ Non-Fat Kefir\ Gluten-free\ • Great Britain \ 12 probiotic cultures\ Veggie Kefir High in protein and calcium Kefir Oats All-natural\ 99% lactose-free\ Gluten-free\ 12 probiotic cultures\ Oat fiber enriched\ High in protein and calcium Perfect 12 Kefir\ All-natural\ \ 99% lactose-free\ Traditional Kefir\ Gluten-free\ \ 12 probiotic cultures\ Greek Style Kefir\ No added sugar\ High in protein and calcium Low Fat Kefir (Organic)\ USDA Certified Organic\ Oregon Tilth Certified Organic\ 99% lactose-free\ Gluten-free\ 12 probiotic cultures\ High in protein and calcium Whole Milk Kefir (Organic) USDA Certified Organic\ Oregon Tilth Certified Organic\ 99% lactose-free\ Gluten-free\ 12 probiotic cultures\ No added sugar Helios Kefir (Organic) USDA Certified Organic\ Oregon Tilth Certified Organic\ 99% lactose-free\ Gluten-free\ Seven probiotic cultures\ Contains Inulin Green Kefir (Organic) USDA Certified Organic\ Oregon Tilth Certified Organic\ 99% lactose-free\ Gluten-free\ 12 probiotic cultures\ Phytoboost = 1 serving of vegetables ProBugs (organic) USDA Certified Organic\ Oregon Tilth Certified Organic\ 99% lactose-free\ Gluten-free\ 12 probiotic cultures\ No-spill pouch ProBugs Blast (Organic) USDA Certified Organic\ Oregon Tilth Certified Organic\ 99% lactose-free\ Gluten-free\ 12 probiotic cultures\ High in protein and calcium Frozen ProBugs (Organic) All-natural\ 99% lactose-free\ Gluten-free\ 10 probiotic cultures\ High in protein and calcium Frozen Kefir All-natural\ 99% lactose-free\ 10 probiotic cultures\ 90 calories per serving\ 1 g of fat Frozen Kefir Bars All-natural\ 99% lactose-free\ Gluten-free\ 10 probiotic cultures\ 60 calories per serving\ 0.5 g of fat BioKefir All-natural\ 20 Billion units of probiotics\ 12 probiotic cultures\ 99% lactose-free\ Gluten-free\ High in protein and calcium Farmer Cheese 99% lactose-free\ Gluten-free\ High in protein and calcium Evolve Kefir\ Evolve Kefir 11 probiotic cultures.\ • United States Natural fruit flavors.\ Fiber.\ Protein and calcium Wallaby Organic\ Lowfat Kefir 12 different strains of Live and Active Kefir cultures. • Australia CocoKefir\ CocoKefir\ Dairy, gluten, soy, and fat free\ • United States App\ Low calorie\ le Cinnamon CocoKefir\ Contains valuable nutrients such as potassium, manganese, and magnesium.\ Citrus CocoKefir\ Beneficial probiotic strains CocoYo\ Body Ecology Coconut Kefir\ ---------------------------------------------------------------------------------------------------------------------------- Conclusion ========== Kefir, the traditional beverage, is now recognized as a potential source of probiotics and molecules with highly interesting healthy properties. The careful and detailed characterization of kefir composition has helped the scientific community to find new possibilities for its application. Kefiran, the EPS of kefir, has very important physicochemical and rheological properties. Besides, its biological properties suggest its use as antioxidant, antitumor agent, antimicrobial agent, and immunomodulator, among other roles. Research is constantly being conducted to consolidate kefir and kefiran properties for the development of new important products to preserve consumer's health. Acknowledgment ============== Authors want to thank CNPq and CAPES for the financial support. Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. [^1]: Edited by: *Maria De Angelis, University of Bari Aldo Moro, Italy* [^2]: Reviewed by: *Theo Varzakas, Technological Educational Institute of Peloponnese, Greece; Cristiana Garofalo, Università Politecnica delle Marche, Italy* [^3]: This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

{.61} {.62} {.63} {.64}

Background ========== Thrombopoietin (Thpo) and its receptor (Mpl), which primarily regulate megakaryopoiesis, are also expressed in the central nervous system (CNS). Although a 39 amino-acid fragment of the N-terminal domain of Thpo shares significant homology with erythropoietin (Epo), which acts as neuroprotective substance \[[@B1]\], and with various neurotrophins \[[@B2]\], current data show pro-apoptotic effects of recombinant Thpo on newly generated neuronal cells in models of brain hypoxia/ischemia \[[@B3]\]. Expression of Thpo transcripts and/or protein has been described in the developing and adult brain, including the cerebral cortex, hippocampus, brain stem, and cerebellum \[[@B3]-[@B7]\]. Thpo mRNA and protein have been detected in neurons, astrocytes and microglia derived from the cerebral cortex and the hippocampi, as well as in Purkinje cells and in granule cells of the cerebellum \[[@B3],[@B6]\]. Thpo seems to be involved in the inflammatory response of the brain, because some patients with meningitis demonstrated elevated Thpo levels in the cerebrospinal fluid \[[@B8]-[@B10]\]. Thpo is most likely produced locally, since its high molecular weight (about 70-80 kDa) makes its crossing over the intact blood-brain barrier rather unlikely \[[@B11]\]. Under experimental conditions of hypoxia, *Thpo*mRNA expression is down-regulated \[[@B3],[@B6],[@B8]\]. Thpo is thought to mediate its effects by binding to its receptor Mpl. However, data on the expression and regulation of Mpl in the CNS are very limited. So far, expression of *Mpl*mRNA has been described in the CNS of mice, rats and humans \[[@B3],[@B4]\]. Region-specific analysis revealed *Mpl*transcripts in the adult rat cortex and hippocampus. In primary rat neurons and astrocytes, Mpl expression has been confirmed by RT-PCR and immunostaining. Under hypoxia *Mpl*mRNA levels increased in astrocytes, but decreased in hippocampal neurons, suggesting a differential regulation under certain conditions \[[@B3]\]. Additionally, we recently showed that Mpl is expressed on transcript and protein levels in cultured brain-derived endothelial cells of adult mice \[[@B12]\]. Neither malformation nor functional deficits of the CNS have yet been reported in transgenic mice with homozygous deletion of the *Thpo*(*Thpo*^-/-^) or *Mpl*(*Mpl*^-/-^) gene \[[@B13]-[@B15]\]. However, cerebral and cerebellar hypoplasia have been described in some patients with congenital amegakaryocytic thrombocytopenia (CAMT) \[[@B16],[@B17]\], which is caused by mutations in the ligand-binding domain of the MPL. In the thrombocytopenia and absent radii (TAR) syndrome, for which a defect in Mpl signalling has been shown \[[@B18]\], cerebellar dysgenesis either with/without agenesis of the corpus callosum and delayed myelinisation have been reported \[[@B19]-[@B21]\]. Herein, we analysed the Mpl expression pattern in the developing and adult murine CNS by immunohistochemistry in order to highlight cells sensitive to Thpo and to identify the time frame for the highest activity of the Thpo/Mpl system. Results ======= Specificity of the Mpl antibody ------------------------------- To test the different anti-Mpl antibodies for specificity in the brain, we compared the recognition pattern between wild-type and *Mpl*^-/-^mice. A monoclonal antibody directed against the murine Mpl (clone AMM2; Kirin Brewery) and a commercially available antibody directed against human Mpl (anti-human TpoR, clone 167639, R&D Systems) both displayed specific labelling of brain parenchymal cells of neonatal wild-type mice, but not in *Mpl*^-/-^pups (Figure [1A](#F1){ref-type="fig"}). Labelling of the choroid plexus may be unspecific (Figure [1A](#F1){ref-type="fig"}) as often the case in this structure. Of note, consecutive sections were also stained without using the primary antibody. Additionally, we confirmed the specificity of the anti-human Mpl antibody by comparing *in situ*hybridization and immunohistochemistry in the same section of the neonatal wild-type or *Mpl*^-/-^brain (Figure [1B](#F1){ref-type="fig"}). Of note, various clones of this anti-human antibody differed in the quality of staining, and the best results were obtained using the 167639 clone. Because of the commercial availability of this antibody, we used it for the rest of the analyses. The combined testing has been quoted as a golden standard for validation of antibodies to be used in immunohistochemistry \[[@B22]\]. {#F1} Expression pattern of Mpl in the developing brain ------------------------------------------------- Our analysis revealed no specific Mpl labelling in any resident cells of the CNS at developmental stages E12 and E15 (Figure [2A-C](#F2){ref-type="fig"}). Signals detected in the dorsal medulla at E12 may be an artefact (Figure [2A](#F2){ref-type="fig"}), since these structures were larger in size than normal neuroblasts at E12 and had ragged shape; they were also detected on consecutive sections incubated with secondary antibody alone. {#F2} From E18 onwards Mpl protein expression was detectable in the prosencephalon, midbrain, and rhombencephalon as well as in the spinal cord (Table [1](#T1){ref-type="table"}). In the telencephalon, Mpl-positive cells were detected in the cerebral cortex, in the subventricular zone of the IV^th^ventricle and in the inner layers of developing olfactory bulb, but not in the hippocampus and basal ganglia. Widespread Mpl expression was found in the diencephalon (thalamus and hypothalamus) and mesencephalon (superior and inferior colliculus); other areas included the pons (metencephalon), medulla (myelencephalon), and the grey matter of the cervical spinal cord (Figure [2](#F2){ref-type="fig"} D-G, Table [1](#T1){ref-type="table"}). There was no labelling within the lateral recesses of the secondary rhombic lip that generates the external granule layer of the cerebellum or granular cells of the vestibulo-cochlear anlage (*data not shown*). ###### Expression pattern of Mpl protein in the developing and adult mouse brain. ------------------------------------------------------------------------------------------ Subdivisions E18.5 P0 P4 P7 Adult --------------------- -------------------------------------- ----- ---- ----- ------- ---- **Spinal cord** Cervical Grey matter ++ ++ ++ +/- ++ White matter \- \- \- \- \- **Rhombencephalon** Myelencephalon Roof plate, cochlear complex \+ ++ \+ \- ++ Medulla, r2-r8 \+ ++ \+ +/- ++ Metencephalon Pons, r1 \+ ++ \+ \+ ++ Cerebellar white matter \- ++ \+ \+ \- Cerebellar cortex +\ \- \- \- \- \- external granule layer Purkinje cells of cerebellum \- \- \- \- \+ **Midbrain** Mesencephalon Prepontine hindbrain \- \+ \+ \+ \+ Tegmentum \+ ++ \- \+ ++ Inferior colliculus \+ ++ \+ +/- ++ Superior colliculus ++ ++ \+ +/- ++ **Prosencephalon** Diencephalon Prethalamus ++ ++ \+ +/- \+ Thalamus ++ ++ ++ \+ ++ Pretectum \+ ++ \+ \+ \+ Preoptic area ++ ++ \+ \- ++ Hypothalamus \+ ++ \+ \+ \+ Telencephalon Basal ganglia \- \+ \+ +/- \+ Hippocampal formation \- \- +/- ++ \+ Cerebral cortex, subventricular zone -/+ ++ \- \- \- Cerebral cortex \+ \+ \+ \+ \+ Olfactory bulb \+ \+ \- \- \+ ------------------------------------------------------------------------------------------ The rough estimates of the staining intensity are presented in different brain regions between E18.5, one day before birth, early postnatal stages as well as in the adult brain. Even considering that the staining intensity does not necessarily represent the strongest protein expression, it appeared that Mpl expression peaked in the early neonatal period. At birth the number of Mpl-positive cells increased as observed through many sections of different stages in parallel. In the telencephalon, the newly developing cortical subventricular zone contained numerous Mpl-positive cells with tangentially oriented processes, but the developing white matter was devoid of any labelling (Figure [2H](#F2){ref-type="fig"}). In addition, the caudal cortex contained a densely packed positive cell cluster (Figure [2I](#F2){ref-type="fig"}). Other clusters of Mpl-positive cells were located in the diencephalon (caudal thalamus; Figure [2J](#F2){ref-type="fig"}) and in the mes- and metencephalon (pons and medulla; *data not shown*). The cerebellar white matter became Mpl-positive, but the lateral recessus of the IV^th^ventricle and granule cell streams remained Mpl-negative (*data not shown*). Furthermore, deep regions of the vestibulo-cochlear nuclei also contained Mpl expressing cells (Figure [2K](#F2){ref-type="fig"}). At P4, the abundant labelling of the cortical ventricular zone disappeared, leaving behind a few Mpl-positive cells in the interface between the developing white matter and the inner cortical plate (Figure [3A](#F3){ref-type="fig"}). The subventricular zone surrounding the IV^th^ventricle had also become Mpl-negative. The expression of all other areas remained almost identical compared to P0 (Figures [3B-D](#F3){ref-type="fig"}, Table [1](#T1){ref-type="table"}) {#F3} Three days later, at P7, the number of the Mpl-positive cells further diminished. Interestingly, we observed diffuse labelling in the alveus and in some cells within the hippocampal formation scattered underneath pyramidal cells of CA1 (Figure [3E](#F3){ref-type="fig"}). Other positive areas included the ventral medulla, pons, cerebellar white matter and various regions of the midbrain. Similarly, Mpl-expression is maintained in most areas of the diencephalon, including thalamus and hypothalamus (Figures [3F-H](#F3){ref-type="fig"}, Table [1](#T1){ref-type="table"}). Finally, immunohistochemical analysis revealed Mpl-expressing cells also in the adult CNS. Mpl expressing regions in the telencephalon included the cerebral cortex, the hippocampus and the mantle zone of the olfactory bulb (*data not shown*). Large areas in diencephalon (preoptic area; Figure [3I](#F3){ref-type="fig"}), midbrain and medulla oblongata also contained Mpl positive cells at different density (Table [1](#T1){ref-type="table"}). In contrast to the earlier stages, Mpl was detectable in Purkinje cells of the cerebellum, but not longer in the white matter (Figure [3J](#F3){ref-type="fig"}). Mpl-expressing cells in the brainstem were found in dorsal cochlear nucleus (Figure [3K](#F3){ref-type="fig"}) and in the grey matter of the spinal cord (Figure [3L](#F3){ref-type="fig"}). Characterization of Mpl expressing cells ---------------------------------------- To characterize the cells, which express Mpl in the brain, we performed double immunohistochemistry with antibodies directed against Mpl as well as with established markers for neurons and astrocytes. All brain areas on parasagittal sections through the middle of the brain were examined for a possible overlap. The majority of Mpl-positive cells in the developing brain co-expressed the pan-neuronal marker NeuN with one exception: Mpl-expressing cells in the cerebellar white matter did not co-express NeuN (Figure [4A](#F4){ref-type="fig"}), indicating their possible glial origin. Since the Mpl-positive cells appeared in the brain during the generative period of small granule cells, we also performed double labelling of Mpl with the early granule cell marker NeuroD, but found no overlap in any of the examined regions (see Figure [4B](#F4){ref-type="fig"} for representative examples). {#F4} Moreover, we performed double labelling with the Mpl antibody and the astrocyte marker GFAP to investigate Mpl expression in astrocytes under physiological conditions. However, none of the Mpl-positive cells co-expressed GFAP. Figure [4C](#F4){ref-type="fig"} shows three representative examples at the developmental stages P0 and P7, as well as in adult mice. Thus, the Mpl-positive cell population in the perinatal and early postnatal murine CNS is mostly neuronal, but definitely heterogenic. For example, Mpl-positive cells in the developing cerebellar white matter had elongated cell bodies and projections parallel to the cerebellar cortex. These cells were negative for both NeuN and GFAP, ruling out the populations of outside-projecting deep cerebellar neurons and astrocytes (Figure [4A](#F4){ref-type="fig"}). Almost all cells of the developing cerebellar cortex, including granule neurons and small inhibitory neurons were Mpl-negative during perinatal and early postnatal development (Figure [3C](#F3){ref-type="fig"} and [3G](#F3){ref-type="fig"}; Figure [4A](#F4){ref-type="fig"} and [4B](#F4){ref-type="fig"}). However, in the adult cerebellum, Mpl expression was located in Purkinje cells (Figure [3J](#F3){ref-type="fig"}). Developmental changes in *Mpl*and *Thpo*mRNA expression throughout development ------------------------------------------------------------------------------ To further clarify the data on the developmental pattern of Thpo/Mpl expression, we analysed expression of both genes in the brain of wild-type mice (Additional file [1](#S1){ref-type="supplementary-material"}). RT-PCR analysis indicated a developmental up-regulation of *Thpo*mRNA. *Mpl*transcript levels appeared to be higher at very early stages (E9.5 to E13.5) compared to the perinatal period and adulthood. However, it needs to be considered that the tissue specimens of early embryonic mice (prior to e15.5) could not be efficiently perfused through the heart for technical reasons. Thus, *Mpl*mRNA levels might be false high, if mRNA from circulating hematopoietic *Mpl*expressing progenitors has been amplified in PCR analysis. This has been experimentally proven by comparing *Mpl*mRNA levels in perfused *vs*. non-perfused specimens of the adult mouse brain (Additional File [1](#S1){ref-type="supplementary-material"}). Beyond E 13.5, however, *Mpl*mRNA levels remained stable, which is concordant with the data obtained by immunohistochemistry. Mpl and Thpo mRNA expression in primary neurons, astrocytes, microglia, and neuronal stem cells ----------------------------------------------------------------------------------------------- Furthermore, we examined the expression of *Mpl*mRNA in cultured primary brain cells, including neurons, astrocytes, and microglia prepared from the developing rat brain, and in neuronal stem cells prepared from adult mouse brain. Amplification of *Mpl*transcripts showed the presence of *Mpl*expression not only in neurons, but also in cultured astrocytes (Figure [5A](#F5){ref-type="fig"}). A faint band for *Mpl*transcripts has been found in microglia, most likely resulting from contamination of this culture with astrocytes, which is a normal issue in preparing these cultures. In murine neuronal stem cells, *Thpo*mRNA was highly expressed, while these cells did not express detectable amounts of *Mpl*mRNA (Figure [5B](#F5){ref-type="fig"}). {#F5} Gross morphology of the Mpl^-/-^brain ------------------------------------- Investigating the intracranial morphology of *Mpl*^-/-^mice *in vivo*in neonatal (P7) as well as in adult stages, we did not detect any structural abnormalities on the macroscopic level in comparison to wild-type littermates. All animals presented with normal development and distribution of the white and the grey matter, inconspicuous gyrification, proper width of the inner and outer cerebral fluid interspaces, and normal anatomical morphology of the cerebrum as well as the cerebellum, as presented exemplarily in Figure [6](#F6){ref-type="fig"}. {#F6} Discussion ========== To further elucidate the role of the Thpo/Mpl system in the brain, we analysed herein the expression pattern of the Mpl protein in the murine brain by immunohistochemistry. It becomes evident that the Mpl is developmentally regulated, since Mpl is not detectable on protein levels at the very early stages (E12 and E15) of brain development (Figure [2A-C](#F2){ref-type="fig"}). Mpl protein expression seems to peak around birth (Figure [2D-K](#F2){ref-type="fig"}) , but is maintained in the adult murine brain (Figure [3](#F3){ref-type="fig"}; Table [1](#T1){ref-type="table"}). Previous data obtained by conventional RT-PCR showed slightly higher *Mpl*mRNA expression in the fetal rat brain compared to adult rat hippocampus and cortex \[[@B3]\]. Considerably high Mpl transcript levels may result in part from circulating hematopoietic cells as shown in our analysis of non-perfused brain tissue specimens (Additional File [1](#S1){ref-type="supplementary-material"}). However, our data also indicate a spatial and temporal expression pattern of Mpl within various areas of the developing and adult brain (Table [1](#T1){ref-type="table"}). This may be important for future dissection of the regulation and function of the Thpo/Mpl system in the brain. During late gestation, Mpl-positive cells are located in the inner layer of the cortex, in the subventricular zone of the IV^th^ventricle (Figure [2E](#F2){ref-type="fig"}) and in the olfactory bulb, but not in the hippocampal formation of the telencephalon. Furthermore, Mpl-positive cells are located in various areas of the diencephalon (including thalamus and hypothalamus), in the inferior and superior colliculus of the mesencephalon, in the pons and medulla, and in the grey, but not in the white matter of the spinal cord (Figures [2F-G](#F2){ref-type="fig"}; Table [1](#T1){ref-type="table"}). The lack of Mpl-positive cells in the secondary rhombic lip, which generates the external granule layer of the cerebellum, and in granule cells of the vestibulo-cochlear anlage may indicate that the Thpo/Mpl system is active in cells generated from the residential ventricular zone of the IV^th^ventricle rather than from the secondary subventricular zone. This hypothesis is supported by the observation that the lateral recesses of the IV^th^ventricle and granule cell streams, invading into the external cerebellar granule layer and between the cochlear nuclei, remain Mpl-negative at later stages. During the neonatal period and in adulthood of mice, Mpl expression remains robust in the diencephalon, mesencephalon, myelencephalon and the grey matter of the spinal cord (Table [1](#T1){ref-type="table"}). Notably, the strongest variations of Mpl expression occur in the telencephalon and in the metencephalon: In the telencephalon, Mpl-positive cells are initially located in the cortical subventricular zone and in the caudal cortex (Figures [2H, I](#F2){ref-type="fig"}), but not in the developing white matter. Later, the abundant labelling of cells in the cortical subventricular zone disappears. Only a few Mpl-positive cells can be detected in the interface between the white matter and inner cortical plate at P4 (Figure [3A](#F3){ref-type="fig"}). However, from P4 onwards, Mpl is expressed in the hippocampus. Mpl receptor is not expressed throughout the hippocampus, but in some multipolar cells in the stratum lacunosum/moleculare. This may be interesting, since the Thpo/Mpl system plays a role in selecting neurons by neurotrophins during ongoing neurogenesis \[[@B3]\]. The second major developmental change in Mpl expression affects the cerebellum. Here, we observe Mpl expression in the cerebellar white matter only during the perinatal period, but no longer in the adult (Figure [3C, G, J](#F3){ref-type="fig"}; Table [1](#T1){ref-type="table"}). Furthermore, Mpl expression in Purkinje cells is obviously silenced during development, but active in the adult cerebellum (Figure [3J](#F3){ref-type="fig"}). Both observations are of particular interest, since some patients with CAMT or TAR-syndrome, both resulting in impaired Mpl function, exhibit structural and functional abnormalities of the brain, affecting particularly the cerebellum \[[@B16],[@B19]-[@B21]\]. Double-labelling with various cell lineage markers suggest that most Mpl expressing cells are neurons, confirmed by staining with the pan-neuronal marker NeuN (Figure [4A](#F4){ref-type="fig"}). This is consistent with previous data indicating Mpl expression in primary rat hippocampal neurons as well as human SH-SY5Y and rat PC12 cells, which both exhibit a neuronal phenotype \[[@B3],[@B8],[@B23]\]. We could not detect Mpl protein in astrocytes by GFAP, although Mpl has been previously identified in isolated cortical astrocytes by RT-PCR and immunohistochemistry \[[@B3]\]. Thus, we re-evaluated *Mpl*expression in isolated rat astrocytes and could indeed confirm expression of *Mpl*transcripts (Figure [5](#F5){ref-type="fig"}). Either the sensitivity of immunohistochemistry is too low to detect Mpl expression in brain tissue specimens, or *in vitro*conditions cause up-regulation of *Mpl*mRNA expression. The later hypothesis appears to be more likely, since *Mpl*mRNA expression in astrocytes is known to be induced by hypoxia \[[@B3]\]. The hypoxic up-regulation of Mpl expression in astrocytes contrasts the down-regulation of the Mpl in neuronal cells under the same conditions \[[@B3]\], raising the question whether the Thpo/Mpl system exhibits a different role in astrocytes *vs*. neurons. Ehrenreich *et al*. identified a proapoptotic role of Thpo that seems to be restricted predominately to immature neuronal cells, suggesting a role for Thpo in the selection of differentiated neurons \[[@B3]\]. In line with this report, we detected *Mpl*mRNA expression in primary cortical neurons, but not in neuronal stem cells (Figure [5](#F5){ref-type="fig"}). Thus, the combined data support the current model that Thpo exhibits its proapoptotic role in maturating neurons, whereas neuronal stem or progenitor cells are largely unaffected. Thereby, Thpo may indeed help to select neurons that acquire target-derived neurotrophic support \[[@B3]\]. Notably, Mpl up-regulation coincides with a period of very active neurogenesis and establishment of connectivity. Mpl expression peaks in some crucial brain areas at a period, when normal neurogenesis prunes. This raises the question whether it may render neonates more susceptible than adults to the pro-apopotic effect of Thpo. Such a condition may be of particular significance in the case of intracranial or intraventricular haemorrhage, since it leads to a massive release of circulating Thpo into the brain. Experiments in a mouse model of uni-lateral stroke showed that recombinant Thpo exhibited significantly higher damage scores and extent of apoptosis \[[@B3]\]. To the best of our knowledge, further data on the function of Thpo in any model of ischemic or traumatic brain injury are not available yet. The biological relevance of the Thpo/Mpl system still needs to be further proven. Due to the developmental changes in Mpl expression in the SVZ of the telencephalon and in the cerebellum (downregulation in the white matter; upregulation in Purkinje cells) as well as structural and functional abnormalities in the brain of some patients with impaired Mpl function (CAMT, TAR syndrome) affecting particularly the cerebellum \[[@B16],[@B19]-[@B21]\], we expected structural abnormalities in the brain of mutant mice with homozygous *Mpl*deficiency. Surprisingly, MRI scans of *Mpl*^-/-^mice did not show gross abnormalities in areas which predominately express the Mpl during brain development or in adulthood. However, ultrasound and/or MRI examinations and assessment of neurodevelopment (for example Bayley Scales of Infant Development) should be undertaken in patients with CAMT or TAR syndrome to get further insights into the role of Thpo and Mpl in the brain. We did not observe a reduced life span of *Mpl*^-/-^mice. Preliminary data on H&E sections do not indicate cytomorphological changes in the brain of the *Mpl*^-/-^mice (not shown). Conclusions =========== In summary, our immunohistochemical analysis shows specific expression pattern of Mpl in various regions of the brain. Most strikingly, Mpl expression showed two major developmental changes in the telencephalon (affecting the subventricular zone and the hippocampus) and the cerebellum (affecting the white matter and Purkinje cells). Furthermore, we observed that Mpl protein expression in the brain is largely restricted to neuronal cells and in the adult brain to Purkinje cells of the cerebellum. We could not detect Mpl expression in resting astrocytes, pointing either to expression levels below the detection limit of immunohistochemistry or to an expression limited to certain *in vitro*conditions. The lack of *Mpl*expression in neuronal stem cells and the pattern of Mpl expression in the brain parenchyma support the concept that Thpo is critical for the selection of neurons to undergo apoptosis during a crucial period of neurogenesis. One should be aware of these findings prior to clinical trials on the use of the second generation of thrombopoiesis-stimulating agents in neonates, since these thrombopoietin mimetic peptides or non-peptide molecules are very small in size and molecular weight \[[@B24]\]. Thus, these promising drugs need to be tested whether they cross the blood-brain-barrier in neonates. Methods ======= Animal experiments ------------------ Wild-type CD1 mice (Charles River Laboratories, Sulzfeld) were mated, and noon on the day of discovery of the vaginal plug was designated embryonic day 0 (E0). Mpl expression was analysed on embryonic day 12 (E12), E15, E18, at birth (P0), on postnatal day P1, P4 and P7, as well as in adult mice. Embryos E12-15 were fixed in ice-cold 4% PFA overnight. For better penetration of the fixative, more mature embryos and pups were additionally perfused through the heart under lethal intraperitoneal thiopental anaesthesia. At least 3 animals have been used for each time point. For double-labelling experiments we have examined at least 5 animals per one pair of markers. Transgenic mice with homozygous deletion of the *Mpl*gene (kindly provided by Warren S. Alexander, Walter & Eliza Hall Institute of medical Research, Melbourne, Australia) were used to validate immunohistochemistry and *in situ*hybridization \[[@B13]\]. The tissue was treated identical to that of wild-type animals. Two animals of each genotype have been examined at three time points. Homozygous *Mpl*deletion was confirmed by PCR analysis of genomic DNA. For PCR analysis, we applied the following primers directed against DNA fragments of the transgenic and the wild-type alleles (GenBank Accession No. [NM_001122949](NM_001122949)): a) 2l Oligo lvg17 5\'-TCCAAGGTAAA GCACTGAAGTCCA-3\', b) 2l Oligo lvg15 5\'-GTCTCCATGGAGGCTTAGGTGGGA-3\', and c) 1l Oligo lvg19 5\'-GAAGAGCTTGGCGGCGAATGGGCT-3\' \[[@B13]\]. All animal procedures fully complied with institutional and state guidelines and were approved by the Institutional Review Board. Immunohistochemistry -------------------- After overnight cryoprotection of the brain tissue in 20% ice-cold sucrose, the specimens were frozen in OCT (Cryo-M-Bed, Bright Instrument Co. Ltd., Huntingdon, Cambridgeshire, UK) and sectioned at 14 μm. Whole embryos or brains of older animals were cut in parasagittal direction, making twelve series of sections. Two of them containing all sections between the midline and lateral edge of the large hemispheres per animal have been subjected to immunohistochemistry with and without primary antibody and examined under the fluorescent microscope. The sections were washed in PBS containing 0.1% (v/v) Triton X-100 (Sigma), then blocked in PBS containing 0.1% bovine serum albumin Fraction V (Roth, Karlsruhe, Germany), 10% FBS, 0.1% Triton X-100 and 0.05% sodium azide (Sigma). Sections from adult brain have been additionally postfixed with ice-cold 4% PFA/PBS for 10 min and washed three times with PBS containing Triton prior to the permeabilization/blocking step. For immunohistochemical analysis the sections were incubated at 4°C with the following primary antibodies diluted in blocking solution: mouse anti-human MPL (clone 167639; IgG~2a~; R&D Systems Minneapolis, MN; final concentration 2.5 μg/ml), rat anti-mouse Mpl (clone AMM2, 1:200; Kirin Brewery, Tokyo, Japan; final concentration 5 μg/ml), mouse anti-mouse NeuN (A60, IgG~1~; Millipore, Billerica, MA; final concentration 2 μg/ml), goat anti-mouse NeuroD (N-19; Santa-Cruz Biotechnology, Santa Cruz, CA; final concentration 0.4 μg/ml) and rabbit anti-cow GFAP (6F2; DAKO, Glostrup, Denmark; final concentration 5.8 μg/ml). After 3 washes, tissue specimens were incubated with the appropriate secondary antibody for 1 hour at room temperature and counterstained with DAPI (2 μg/ml; Roche, Basel, Switzerland). Cy-2 anti-mouse IgG, Cy-3 anti-goat or anti-rabbit IgG (Jackson ImmunoResearch, West Growe, PA) served as secondary antibodies. The two subclasses of IgGs were differentiated by either Alexa 594 anti-mouse IgG~2a~or Alexa 488 anti-mouse IgG~1~(Invitrogen,) as secondary antibodies. After detection of the Mpl, sections were thoroughly washed 3 times, fixed with 4% PFA for 5 min to crosslink the signal and processed for the second antibody the same way, including appropriate blocking. Microscopic analysis was performed using an epifluorescence microscope (AxioPlan 2 Imaging System; Carl Zeiss, Jena, Germany). Photographs were taken with a digital camera (AxioCAM MRc) using AxioVision 4.2 software (Carl Zeiss). While photographing the same region of the knockout animal and its wild-type littermate the same exposure and post processing has been used to allow faithful comparison. One may note that this resulted in a slightly different look of pictures due to deviations in DAPI intensity, but we refrained from adjusting it in Figure [1A](#F1){ref-type="fig"}. *In situ*hybridization ---------------------- To validate the Mpl antibody, freshly-cut 14 μm cryosections from wild-type and *Mpl*^-/-^mouse brain were first subjected to *in situ*hybridization with an antisense riboprobe against mouse *Mpl*followed by immunostaining with the anti-human Mpl antibody. *In situ*hybridization was performed as described \[[@B25]\]. The detailed protocol is available at <http://www.ucl.ac.uk/~ucbzwdr/richardson.htm>. The digoxigenin-labeled riboprobe was transcribed from a \~2kb cDNA encoding the full-length murine *Mpl*cDNA (GenBank Accession No. [NM_001122949](NM_001122949)) cloned into pSPT18 vector (kindly provided by Michelle Souyri, Hôpital Paul Brousse Villejuif, France) \[[@B26]\]. Preparation of primary cells from the brain ------------------------------------------- Primary neuronal cultures of cerebral cortex from Wistar rat embryos (E17) were prepared as previously described \[[@B27]\]. Astroglial and microglial cell cultures were prepared from newborn rats as described elsewhere \[[@B28]\]. Microglial cells were seeded at a density of 100,000 cells/cm^2^and harvested after 24 h. Astrocytes were seeded at a density of 120,000 cells/cm^2^and harvested after 48 h. Adult neural stem cells (NSC) were prepared from C57/Bl6 mice as described \[[@B29]\]. NSC were cultivated in NBMA (Invitrogen) containing 2% B27 without retinoic acid (Invitrogen), 1% L-glutamine, 10 ng/ml bFGF, and 20 ng/ml EGF (Biochrom, Berlin, Germany). RT-PCR analysis of Mpl expression --------------------------------- *Mpl*mRNA expression was analysed in primary brain-derived cells by conventional RT-PCR analysis. Total RNA was prepared using TRIzol reagent^®^(Invitrogen), and first strand cDNA synthesis was performed with 2 μg of total RNA using oligo(dt) primers (Promega,) and M-MLV reverse transcriptase (Invitrogen). One-twentieth of reaction product was used for PCR amplification in a thermal cycler, which was carried out under the following conditions: DNA denaturation at 94°C for 60 seconds, primer annealing at 60°C (or 55°C rat Mpl) for 60 seconds, extension of double-stranded DNA at 72°C for 90 seconds. The following primers were used: mouse or rat *Mpl*5\'-CTAGCTCCCAAGGCTTCTTC-3\' (forward primer), 5\'-GGCTCCAGCACCTTCCAGTCC -3\' (reverse primer), product size 392 bp (GenBank Accession No. [NM_001122949](NM_001122949)); mouse *Thpo*5\'-CTCTGTCCAGCCCCGTAGC-3\' (forward primer), 5\'-CCCCAA GAGGAGGCGAAC-3\' (reverse primer), product size 314 bp (GenBank Accession No. [NM_009379](NM_009379)); mouse *Actb*5\'-ACTGCTCTGGCTCCTAGCAC-3\' (forward primer), 5\'-ACATCTGCTGGA AGGTGGAC-3\' (reverse primer), product size 115 bp (GenBank Accession No. [NM_007393](NM_007393)); rat *Actb*5\'-ATCGTGGGCCGCCTAGCACC-3\' (forward primer), 5\'-CTCTTTAATGTCACGCACGATTTC-3\' (reverse primer), product size 542 bp (GenBank Accession No. [NM_031144](NM_031144)). Amplified PCR fragments were visualized on 1.5% agarose gel stained with ethidium bromide. MRI analysis ------------ Cerebral MRI was performed on a 7 Tesla rodent MRI scanner (Pharmascan 70/16AS, Bruker BioSpin, Ettlingen, Germany), applying a 20 mm RF-Quadrature-Volume head coil. Animals received anesthesia via facemask induced with 3% and maintained with 1.5 - 2.0% isoflurane (Forene, Abbot, Wiesbaden, Germany) delivered in 100% O~2~under constant ventilation control (Bio Trig System, Bruker BioSpin, Ettlingen, Germany). Mice were placed on a heated circulating water blanket to keep up body temperature at 37°C. IRB protocol G0240/06, Landesamt für Gesundheit und Soziales, Berlin, Germany. Triplanar fat-suppressed turbo spin echo T2-weighted (RARE; TE1 14.5 ms, TE2 65.5 ms, TR 4500 ms, 0.5 mm slice thickness, Matrix 256 × 256, FOV 2.8 cm, eight averages, 40 coronal slices, scan time 28 minutes, and 20 axial slices, scan time 28 minutes) and T2\*-weighted (GEFI; TE 5.6 ms, TR 1200 ms, flip angle 35°, 0.5 mm slice thickness, Matrix 256 × 256, FOV 2.8 cm, four averages, 40 coronal slices, scan time 20 minutes, and 20 axial slices, scan time 13 minutes) images were acquired. Identical slice positions were used for all sequences applied: coronal slices were aligned to the olfactory bulb/frontal lobe fissure and covered the entire brain up to the cervical spinal cord. Axial slices were positioned parallel to a plane through the most frontal tip of the olfactory bulb and the most rostral cerebellar part. MRI data were analysed using the MEDx3.4.3 software package (Medical Numerics, Virginia, USA) on a LINUX workstation. Authors\' contributions ======================= AI established, performed and analyzed *in situ*hybridizations as well as immunohistochemistry; AI also drafted and revised the manuscript. JW performed MRI and edited the manuscript. JZ performed Mpl expression analysis by PCR technique. OH isolated neuronal stem cells and revised the manuscript. MB provided the *Mpl*^-/-^mice, various experimental tools and revised the manuscript. CD designed the project, supervised the experimental work and wrote the manuscript. All authors read and approved the final version of the manuscript. Supplementary Material ====================== ###### Additional file 1 **Analysis of expression of *Thpo-mRNA*and *Mpl-mRNA*during the development of mouse brain**. Conventional RT-PCR analysis of *Mpl*and Thpo mRNA expression in the developing and adult brain. Concerning relatively high *Mpl*transcript levels during early embryonic development, one might take into account that the tissues overlying the developing central nervous system contain circulatory hematopoietic cells that carry Mpl. Transcardiac perfusion, often used to minimize that, cannot be efficiently used before E15 for technical reasons. This is supported by higher *Mpl*transcript levels in non-perfused *vs*. perfused brain specimens. ###### Click here for file Acknowledgements ================ This work was supported by a Lydia-Rabinowitsch Fellowship from the Charité - Universitätsmedizin Berlin to AI and by a grant from the Wilhelm-Sander-Stiftung (Az 2007.111.1) to CD and OH. We appreciate the gift of the murine Mpl antibody from Kirin Brewery to MB and of the murine Mpl cDNA from Michelle Souyri, Hospital Paul Brousse, Villejuif, France. The help from Dorette Freyer (Department of Experimental Neurology) in preparing primary brain derived cells is appreciated. We also thank Victoria Hoene, Malte Cremer, and Hannes Sallmon (Department of Neonatology) for critical discussion of data.

###### Significance Statement Psychostimulant addiction (e.g., cocaine and methamphetamine) remains a prevalent problem in the United States. Prior studies suggest that the serotonin~1B~ receptor (5-HT~1B~R) modulates cocaine reinforcement and incentive motivation dependent on whether there is a period of abstinence from cocaine in male rats. This study demonstrates that measures of methamphetamine drug taking in rats are attenuated by pretreatment with 5-HT~1B~R agonists, including the FDA-approved agonist zolmitriptan, regardless of abstinence history. Our findings suggest that 5-HT~1B~R agonists may be useful for treating psychostimulant use disorders. Introduction {#s1} ============ Psychostimulant addiction remains a prevalent problem worldwide ([@CIT0028]; [@CIT0029]), and yet there are still no FDA-approved, effective pharmacological treatments for psychostimulant use disorders. We and others have suggested that the serotonin~1B~ receptor (5-HT~1B~R) may be a useful target for medication development for these disorders ([@CIT0006a]; [@CIT0044]; [@CIT0024]; [@CIT0030]). Advances in medicinal chemistry have discovered drugs with high selectivity for 5-HT~1B~Rs ([@CIT0016]; [@CIT0048]; [@CIT0026]; [@CIT0045]). The agonist 5-propoxy-3-(1,2,3,6-tetrahydro-4-pyridinyl)-1*H*-pyrrolo\[3,2-*b*\] pyridine (CP 94,253) and the antagonist 1'-methyl-5-\[\[2'-methyl-4'-(5-methyl-1,2,4-oxadiazol-3-yl)biphenyl-4-yl\]carbonyl\]-2,3,6,7-tetrahydrospiro\[furo\[2,3-*f*\]indole-3,4'-piperidine (SB 224,289) have high affinities (K~i~ = 2 and 8.2 nM, respectively) for 5-HT~1B~Rs ([@CIT0016]; [@CIT0048]). The selectivity profiles of CP 94,253 and SB 224,289 have established these drugs as useful tools for studying the role of 5-HT~1B~Rs in psychostimulant addiction. Another group of 5-HT~1B~R agonists are tryptamine based, including zolmitriptan (Zomig), which is an FDA-approved drug used to treat migraine headaches. Although zolmitriptan is not as selective for 5-HT~1B~Rs as CP 94,253, it has a high affinity (K~i~ = 5.01 nM) for 5-HT~1B~Rs. Initial experiments examining the effects of 5-HT~1B~R agonists found that these drugs facilitated cocaine self-administration ([@CIT0034]). However, we found that the effects of 5-HT~1B~R agonists vary depending on whether or not animals have undergone abstinence. Specifically, CP 94,253 shifts the cocaine self-administration dose-response curve leftward when given as a pretreatment prior to a daily self-administration session (preabstinence) but produces a downward shift when given as a pretreatment prior to resumption of self-administration after prolonged (i.e., 21 days) abstinence ([@CIT0036]; [@CIT0030]). In addition, CP 94,253 pretreatment increases breakpoints and cocaine intake on a progressive ratio (PR) schedule compared with vehicle pretreatment when tested during daily maintenance sessions. In contrast, following a 21-day period of forced abstinence (postabstinence), CP 94,253 decreases cocaine intake and response rates on the PR schedule. Furthermore, CP 94,253, attenuates cocaine-seeking behavior in tests of both cue-induced and cocaine-primed reinstatement following a few weeks of extinction training during which the rats were abstinent ([@CIT0036]). Cocaine-seeking behavior under these conditions reflects incentive motivational effects produced by the cues and cocaine priming injections ([@CIT0021]). Therefore, these results suggest that preabstinence administration of 5-HT~1B~R agonists facilitates the reinforcing and motivational properties of cocaine while postabstinence 5-HT~1B~R agonists attenuate these effects. This study examined if CP 94,253 produces a similar abstinence-dependent decrease in methamphetamine intake. First, we examined CP 94,253 pretreatment effects on the methamphetamine self-administration dose-response function using low ratio schedules of reinforcement (i.e., fixed and variable ratio 5; FR5 and VR5). Second, we examined CP 94,253 pretreatment effects on a PR schedule of methamphetamine reinforcement as this more demanding schedule is particularly sensitive to changes in motivation for a drug. Third, we examined if the effects of CP 94,253 pretreatment on methamphetamine intake were 5-HT~1B~R mediated by administering the 5-HT~1B~R antagonist SB 224,289 to reverse the agonist effects. Fourth, we examined if CP 94,253 and SB 224,289 affected locomotor activity. Finally, we examined if acute and intermittent repeated treament with zolmitriptan affected methamphetamine intake. Since both methamphetamine and cocaine enhance monoaminergic neurotransmission by an action at monoamine transporters, we hypothesized that CP 94,253 would increase methamphetamine intake preabstinence and decrease methamphetamine intake postabstinence similar to that observed with cocaine intake. Methods {#s2} ======= Animals {#s3} ------- Male Sprague Dawley rats (Charles River) weighing 225 to 250 g were single-housed in a climate-controlled environment on a 14:10 reverse light/dark cycle (lights off at 6:00 [am]{.smallcaps}). Rats had ad libitum access to food except for initial self-administration training when they were food restricted to 90% of their ad libitum weights. The experiments proceeded in accordance with a protocol approved by the Arizona State University Institutional Animal Care and Use Committee. Drugs {#s4} ----- Methamphetamine hydrochloride (Sigma-Aldrich) was dissolved in bacteriostatic saline (Hospira Inc.) and filtered with 0.2-µm membrane Acrodisc syringe filters (PALL Corporation). CP 94,253 hydrochloride (Tocris Bioscience) was dissolved in saline, and SB 224,289 hydrochloride (Tocris Biosciences) was dissolved in 10% (2-hydroxypropyl)-β-cyclodextrin (Sigma-Aldrich) in saline and sonicated for 2 minutes. Zolmitriptan (Sigma-Aldrich) was dissolved in 10% dimethyl sulfoxide in saline and sonicated for 5 minutes. CP 94,253, SB 224,289, and zolmitriptan were prepared fresh daily. Vehicle refers to the respective solvent. All drug injections, with the exception of self-administered methamphetamine, were injected at a volume of 1 mL/kg body weight. Surgery {#s5} ------- Rats underwent surgery for implantation of chronic indwelling catheters into the jugular vein as detailed previously ([@CIT0041]). Rats had 6 to 7 days of recovery before commencing self-administration training. Catheters were flushed daily with 0.1 mL of either timentin (experiment 1; 66.67 mg/mL; GlaxoSmithKline) or cefazolin (experiment 2, 3, 4, and 5; 10 mg/mL; WG Critical Care, LLC) mixed with heparin/saline (70 U/mL; APP Pharmaceuticals). In addition, catheter patency was tested periodically by administering 0.05 mL of methohexital sodium (16.7 mg/mL; Jones Pharma Inc.), a dose that produces brief loss of muscle tone when administered i.v. Apparatus {#s6} --------- The operant conditioning chambers (Med Associates) contained an active and inactive lever, a cue light, and a tone generator as previously described ([@CIT0037]). All chambers had infusion pumps (Med Associates) that connected to liquid swivels (Instech) fastened to an outlet polyethylene tubing sheltered within a metal leash (PlasticsOne) that attached to the rat's catheter. All operant conditioning chambers were housed within sound attenuating boxes that contained a ventilation fan. General Procedures {#s7} ------------------ Experimental sessions commenced at approximately the same time of day, 6 days/week during the rats' dark cycle. Rats were first trained to self-administer 0.1 mg/kg (i.v.) methamphetamine on an FR1 schedule of reinforcement, and they later progressed to either a FR5, VR5, or PR schedule based on individual performance. In all experiments, schedule advancement during training occurred once rats received at least 10 reinforcers/session for 2 consecutive sessions, and testing commenced once rats reached a stability criterion of \<15% variability in the number of infusions obtained across 3 consecutive sessions. The training dose was choosen as it has been shown to be effective in producing methamphetamine acquisition ([@CIT0015]; [@CIT0007]; [@CIT0017]). The VR5 schedule was chosen for its tendency to sustain consistent steady response rates ([@CIT0009]). Session length was 2 hours, except for training and testing on the PR schedule for which sessions were 4 hours long to capture break points in most rats. In the PR schedule, the number of active lever responses required to obtain each subsequent infusion increased exponentially (e.g., 1, 2, 4, 6, 9, 12, 15, etc.), identical to the exponential equation from [@CIT0042]. The PR schedule was chosen because it is progressively more effortful to obtain reinforcement, reflecting in part how *motivated* a rat is to work for methamphetamine ([@CIT0021]). Therefore, the pattern of changes across various schedules of reinforcement can help inform how treatments affect behavioral processes. Completion of any of the operant schedules activated the tone and light cue followed 1 second later by an infusion of 0.1 mL methamphetamine delivered across 6 seconds (dose is given in the specific experiments section). The tone, light, and pump were then turned off, and simultaneously a house-light was activated for 20 seconds to signal a timeout period during which additional lever responses were recorded but had no programmed consequences. After the timeout period, the house-light turned off and methamphetamine was available again. Abstinence from methamphetamine occurred for 21 or more consecutive days, during which the rats were maintained in the home cage, handled, and weighed daily for i.v. administration of timentin or cefazolin to maintain catheter patency. Experiment 1: Effects of CP 94,253 on the Methamphetamine Self-Administration Dose-Response Function Pre- and Postabstinence {#s8} ---------------------------------------------------------------------------------------------------------------------------- The timeline for this experiment is outlined in [Figure 1](#F1){ref-type="fig"}. Beginning 2 days prior to the start of training and continuing throughout the experiment, rats were food restricted to 90% of their initial free-feeding body weight (291 ± 4.6 g). During training, rats progressed from a FR1 to FR5 schedule (\~17--20 sessions) after reaching a criterion of ≥10 infusions for 2 consecutive sessions. Once reinforcement rates met the stability criterion on the FR5, the rats commenced training on a within-session dose-response procedure. For these sessions, each methamphetamine dose (0.003, 0.01, 0.03, 0.1, and 0.30 mg/kg, i.v.) was available in ascending order for 30 minutes with a 5-minute time-out period between doses. After again meeting the stability criterion for reinforcement rates across the within-session dose-response training days (20--28 sessions), rats were randomly assigned to receive an injection of either CP 94,253 (5.6 mg/kg, s.c.) or vehicle 15 minutes prior to their daily dose-response session. The dose of CP 94,253 was selected based on our previous research demonstrating that it selectively reduces cocaine intake without affecting sucrose intake ([@CIT0036]). After this first test session, additional training sessions were given until rats met the stability criterion. Then rats were tested again, receiving the treatment opposite from their first treatment (i.e., animals that received vehicle first, now received CP 94,253 and vice versa). After this test session, rats underwent abstinence for 21 days, during which they remained in their home cages but continued to receive daily i.v. administration of heparin/antibiotic to maintain catheter patency. {#F1} For the postabstinence tests, rats received CP 94,253 or vehicle, and 15 minutes later, they were given access to methamphetamine using the same within session dose-response procedure as used during preabstinence tests. Rats then remained abstinent in the home cage for 3 days to allow time for CP 94,253 to be eliminated and to reinstate an abstinence period prior to the second test. On the second test day, the rats received the treatment opposite from that given prior to the first postabstinence test. A total of 10 rats ran through the entire experiment and their average final weight was 468 ± 4.94 g. Experiment 2: Effects of CP 94,253 on Methamphetamine Self-Administration on a Progressive Ratio Schedule Pre- and Postabstinence {#s9} --------------------------------------------------------------------------------------------------------------------------------- The timeline for this experiment is outlined in [Figure 2](#F2){ref-type="fig"}. A new cohort of experimentally naïve rats was food restricted and trained to self-administer methamphetamine, progressing from a FR1 to a VR5 schedule during 2-hour sessions (11--13 sessions) using the same procedure as the previous experiment, with the exception that food restriction was gradually discontinued once rats were on the VR5 schedule. After meeting the stability criterion on the VR5 schedule, rats were trained on the PR schedule of methamphetamine reinforcement during 4-hour sessions until again meeting the stability criterion (10--18 sessions). We capped session length to 4 hours to ensure that CP 94,253 would remain effective throughout the test ([@CIT0034]). We also reduced the methamphetamine dose to one-half the training dose (0.05 mg/kg, i.v.) with the intention that more rats would reach break point during the 4-hour session. Breakpoint was defined as the last schedule of reinforcement completed prior to a 1-hour period during which the next required ratio failed to be completed, or 4 hours had elapsed, whichever came first. After rats met the stability criterion on the PR schedule, they were randomly assigned to either a CP 94,253 (10 mg/kg, s.c.; n = 8) or a vehicle group (n = 7), counterbalanced for similar number of total drug infusions during training. The dose of CP 94,253 was higher than the previous experiment to sustain CP 94,253 levels throughout testing, and based on previous research this dose selectively decreases cocaine intake while not affecting sucrose intake ([@CIT0036]). These groups received their respective treatments 15 minutes before testing on the PR schedule. After testing, both groups of rats were placed into abstinence for 21 days as described in experiment 1. Postabstinence, both groups received their injections, which were identical to those given preabstinence, 15 minutes prior to a test for resumption of methamphetamine self-administration on the PR schedule. {#F2} Experiment 3: Reversing the Attenuating Effects of CP 94,253 on Methamphetamine Self-Administration with SB 224,289 {#s10} ------------------------------------------------------------------------------------------------------------------- The timeline for this experiment is outlined in [Figure 3](#F3){ref-type="fig"}. A new cohort of experimentally naïve rats was trained to self-administer methamphetamine, progressing from a FR1 to a VR5 schedule (\~12--17 sessions) of methamphetamine (0.1 mg/kg, i.v.) reinforcement. Rats were food restricted only during acquisition of self-administration and all sessions lasted for 2 hours. Once reinforcement rates stabilized under free feeding conditions (\~4--12 sessions), rats were randomly assigned to 1 of 2 groups (n = 14 and 17, respectively); Group 1 received an i.p. injection of either vehicle or SB 224,289 (10 mg/kg, i.p.) 30 minutes prior to the first test and the opposite treatment injection 30 minutes prior to the second test (i.e., rats that received vehicle on test 1 received SB 224,289 on test 2 and vice versa). Group 1 also received a vehicle injection 15 minutes prior to both tests. For Group 2, identical procedures were followed except that rats received vehicle followed by CP 94,253 (5.6 mg/kg, s.c.) on one test day and SB 224,289 (10 mg/kg, i.p.) followed by CP 94,253 (5.6 mg/kg, s.c.) on the other test day. {#F3} In addition to testing whether the antagonist would reverse the effects of the agonist in this experiment, we also verified that these rats showed a CP 94,253-induced decrease in methamphetamine intake postabstinence (not included on timeline). A subset of rats, randomly selected from both Group 1 and Group 2, underwent 21 days of abstinence as described above. They were then assigned to 2 groups, counterbalanced for the number of infusions obtained during training. One group (n=11) received an injection of CP 94,253 (5.6 mg/kg, s.c.) while the other group (n=11) received an injection of vehicle 15 minutes prior to a test for resumption of methamphetamine self-administration (0.1 mg/kg, i.v.). Experiment 4: Effects of 5-HT~1B~R Drugs on Spontaneous Locomotion {#s11} ------------------------------------------------------------------ The timeline for this experiment is outlined in [Figure 4](#F4){ref-type="fig"}. Rats from experiment 2 were used and they had a history of methamphetamine self-administration (38 sessions) and had undergone an abstinence period (23 days). On abstinence day 24, rats were placed into locomotor test chambers for a 60-minute habituation period. The test chambers (45.72 x 25.4 x 20.32 cm) were similar to the home cages and had a camera mounted above to record horizontal movement with Topscan software (Clever Systems). The rats were then tested twice for locomotor activity with 3 rest days intervening the 2 test days. They remained in their home cages during rest days. Thirty minutes prior to the first test, rats were pretreated with either vehicle or SB 224,289 (10 mg/kg, i.p.) and 30 minutes prior to the second test the rats received the opposite treatment as that given on the first test (i.e., rats that received vehicle on test 1 received SB 224,289 on test 2 and vice versa). They were also randomly assigned to 1 of 2 groups. Group 1 (n=7) received a vehicle injection 15 minutes prior to both tests, and Group 2 (n=7) received CP 94,253 (10 mg/kg, s.c.) 15 minutes prior to both tests. The tests began by placing the rat into the test chamber, and distance traveled was measured for 2 hours. {#F4} Experiment 5: Effects of Zolmitriptan on Methamphetamine Self-Administration Pre- and Postabstinence {#s12} ---------------------------------------------------------------------------------------------------- The timeline for this experiment is outlined in [Figure 5](#F5){ref-type="fig"}. Rats that were used in experiment 3 were tested for the effects of zolmitriptan on methamphetamine after achieving stable SA rates on a VR5 schedule (\~3--11 sessions) of methamphetamine reinforcement (0.1 mg/kg, i.v.) across 2-hour training sessions. Rats were then randomly assigned to either a zolmitriptan (10 mg/kg, s.c.; n=9) or a vehicle treatment group (n=6), counterbalanced for similar number of total drug infusions. Rats received their assigned treatment 15 minutes prior to the start of a self-administration session. Then rats underwent a period of abstinence for 29 to 36 days followed by a test phase. During the test phase, the rats again received their assigned treatment of either zolmitriptan or vehicle 15 minutes prior to each of 3 self-administration sessions with 2 to 3 self-administration sessions without treatment between each of the treatment test days. {#F5} Data Analysis {#s13} ------------- Statistical analyses were conducted with IBM SPSS Statistics v. 23. Descriptive statistics are reported as the mean ± SEM. Self-administration data, including active and inactive lever responses, and infusions obtained were analyzed by either repeated-measures or a mixed-design ANOVA with drug pretreatment(s) and dose of methamphetamine as between-subject or within-subject factors depending on the experimental design. In addition, breakpoint and total distance travelled were analyzed as described above for experiments 2 and 4, respectively. All sources of significant effects were further analyzed by Tukey's posthoc tests. There was some attrition in each experiment due to catheter failure or failure to acquire SA. Results {#s14} ======= Experiment 1 {#s15} ------------ Methamphetamine produced an inverted U-shaped dose-effect function, and CP 94,253 decreased methamphetamine infusions and active lever responses both pre- and postabstinence ([Figure 1](#F1){ref-type="fig"}). For the preabstinence tests, there were main effects of methamphetamine dose for both infusions \[*F* (4, 36) = 17.67, *P* \< .05\] and active lever responses \[*F* (4, 36) = 8.40, *P* \< .05\]. Posthoc tests indicated that the 0.01- and 0.03-mg/kg doses produced higher values than the lowest dose (0.003 mg/kg) and the highest dose (0.30 mg/kg) produced lower values than all doses except 0.10 mg/kg (Tukey's comparisons, *P \<* .05) ([Figure 1A](#F1){ref-type="fig"},[B](#F1){ref-type="fig"}). There were also main effects of treatment, which indicated that averaged across methamphetamine dose, rats exhibited lower infusion and active lever response rates when pretreated with CP 94,253 than when pretreated with vehicle ([Figure 1A](#F1){ref-type="fig"},[B](#F1){ref-type="fig"},insets). There were no significant effects for inactive lever responses during preabstinence tests ([Figure 1C](#F1){ref-type="fig"}). Analysis of postabstinence infusions showed main effects of treatment \[*F* (1, 9) = 30.74, *P* \< .05\] and methamphetamine dose \[*F* (4, 36) = 35.27, *P* \< .05\], as well as a treatment by methamphetamine dose interaction \[*F* (4, 36) = 2.87, *P* \< .05\]. CP 94,253 pretreatment decreased infusions at the 3 lowest doses of methamphetamine compared with vehicle pretreatment (Tukey's comparisons, *P* \< .05) ([Figure 1D](#F1){ref-type="fig"}). The analysis of postabstinence active lever responses also revealed main effects for treatment \[*F* (1, 9) = 10.38, *P* \< .05\] and methamphetamine dose \[*F* (4, 36) = 22.20, *P* \< .05\], but no treatment by methamphetamine dose interaction. The main effect of methamphetamine dose was due to the inverted U-shaped dose-response function similar to preabstinence. The main effect of treatment shows that averaged across methamphetamine doses, rats exhibited lower infusions and active lever response rates when pretreated with CP 94,253 than when pretreated with vehicle ([Figure 1D](#F1){ref-type="fig"},[E](#F1){ref-type="fig"}, insets). There were no significant effects for inactive lever responses during the postabstinence tests ([Figure 1F](#F1){ref-type="fig"}). Experiment 2 {#s16} ------------ CP 94,253 decreased methamphetamine infusions, breakpoint, and active lever responses ([Figure 2](#F2){ref-type="fig"}, [A](#F2){ref-type="fig"}-[C](#F2){ref-type="fig"}). Most rats reached breakpoint within the 4-hour test during both the preabstinence (80%) and postabstinence (73%) tests. The analyses showed main effects of pretreatment across pre- and postabstinence tests for infusions \[*F* (1, 13) = 9.86, *P* \< .05\], breakpoints \[*F* (1, 13) = 6.46, *P* \< .05\], and active lever responses \[*F* (1, 12) = 11.09, *P* \< .05\]. In each case, CP 94,253 pretreatment decreased these measures compared with the vehicle pretreatment (Tukey's comparisons, *P* \< .05). There were no main effects for inactive lever responses ([Figure 2D](#F2){ref-type="fig"}) and no pretreatment by time interaction effects for any of the measures. Experiment 3 {#s17} ------------ CP 94,253 decreased methamphetamine infusions, and this effect was blocked by SB 224,289 ([Figure 3](#F3){ref-type="fig"}). Data analysis showed a main effect of the first pretreatment (i.e., vehicle vs SB 224,289) \[*F* (1, 29) = 6.86, *P* \< .05\] and an interaction between the first and second (vehicle vs CP 94,253) pretreatments \[*F* (1, 29) = 12.52, *P* \< .05\]. Posthoc tests showed that pretreatment with vehicle + CP 94,253 decreased methamphetamine infusions compared with all other pretreatment conditions, including SB 224,289 + CP 94,253 (Tukey's comparisons, *P* \< .05). This finding indicates that administration of the antagonist SB 224,289 blocked the attenuating effect of CP 94,253 on methamphetamine intake, suggesting this effect was mediated by 5-HT~1B~Rs. There were no effects for active or inactive lever responses (data not shown). The rats in this experiment underwent 21 days of abstinence and were then tested for the effects of CP 94,253 (5.6 mg/kg, s.c.) on resumption of methamphetamine self-administration (0.1 mg/kg, i.v.). Similar to the results from experiment 1, pretreatment with a 5.6-mg/kg dose of CP 94,253 decreased the number of infusions and active lever responses compared with pretreatment with vehicle during the postabstinence tests (data not shown). The mean number of methamphetamine infusions obtained in the vehicle vs CP 94,253 pretreatment groups was 12.40 ± 0.88 and 8.90 ± 0.71, respectively \[*t* (9) = 7.72, *P* \< .05\]. The mean number of active lever responses in the vehicle vs CP 94,253 pretreatment groups was 70.30 ± 20.81 and 49.20 ± 17.89, respectively \[*t* (9) = 2.62, *P* \< .05\]. There was no group difference for inactive lever responses. Experiment 4 {#s18} ------------ Neither CP 94,253 or SB 224,289 pretreatment altered locomotor activity ([Figure 4](#F4){ref-type="fig"}). There were no main or interaction effects between treatment and the total distance traveled by rats. Experiment 5 {#s19} ------------ Acute zolmitriptan treatment decreased methamphetamine infusions and active lever responses during preabstinence tests ([Figure 5A](#F5){ref-type="fig"},[B](#F5){ref-type="fig"}). Comparisons between vehicle and zolmitriptan showed a difference in infusions \[*t* (10) = 3.50, *P* \< .05\] and active lever responses \[*t* (10) = 2.90, *P* \< .05\]. There were no effects on inactive lever responses after vehicle or zolmitriptan treatment with means of 39.18 ± 16.37 and 16.55 ± 6.41, respectively. Zolmitriptan pretreatment given intermittently across 3 postabstinence tests consistently decreased infusions and active lever responses ([Figure 5C](#F5){ref-type="fig"},[D](#F5){ref-type="fig"}). The ANOVA showed a main effect of treatment group for infusions \[*F* (1, 11) = 21.92, *P* \< .05\] but no effect of treatment day or treatment group by treatment day interaction. For active lever responses, there were significant main effects of treatment group \[*F* (1, 10) = 21.17, *P* \< .05\] and treatment day \[*F* (2, 20) = 6.08, *P* \< .05\], but no treatment group by treatment day interaction. Posthoc tests for treatment day showed that zolmitriptan treatment produced lower active lever response rates on day 2 compared with treatment days 1 and 3. There were no effects on inactive lever responses after vehicle or zolmitriptan treatment with means of 7.89 ± 3.68 and 3.56 ± 0.68, respectively. Discussion {#s20} ========== Unlike the abstinence-dependent modulatory role of 5-HT~1B~R agonists on cocaine intake that we observed previously ([@CIT0038], [@CIT0030]), this study found that 5-HT~1B~R agonists attenuated methamphetamine intake when given either pre- or postabstinence. Specifically, a moderate dose of CP 94,253 (5.6 mg/kg, s.c.) decreased methamphetamine intake and active lever response averaged across methamphetamine dose available both when administered during maintenance of daily self-administration sessions and following a period of abstinence (main effect of pretreatment, [Figure 1](#F1){ref-type="fig"}, insets). After abstinence, the effect of CP 94,253 was more pronounced at lower doses of methamphetamine, primarily because intake appeared higher under the vehicle pretreatment condition postabstinence compared with preabstinence. This enhancement of cocaine intake postabstinence is consistent with sensitized cocaine self-administration reported previously ([@CIT0050]). Thus, the findings suggest that CP 94,253 attenuates expression of the abstinence-induced, enhanced sensitivity to methamphetamine observed with vehicle pretreatment. CP 94,253 (10 mg/kg, s.c.) also decreased methamphetamine intake (0.05 mg/kg, i.v.) under a PR schedule of reinforcement both pre- and postabstinence ([Figure 2](#F2){ref-type="fig"}), further suggesting attenuation of methamphetamine reinforcing and/or motivational effects. Importantly, administration of the 5-HT~1B~R antagonist, SB 224,289 (10 mg/kg, i.p.), blocked the attenuating effects of CP 94,253 (5.6 mg/kg, s.c.) on methamphetamine intake in rats tested during maintenance of self-administration on a VR5 schedule ([Figure 3](#F3){ref-type="fig"}), suggesting that the effects of the agonist are mediated by 5-HT~1B~Rs. Finally, we report that zolmitriptan (10 mg/kg, s.c.) also attenuated methamphetamine intake on a VR5 schedule. Zolmitriptan treatment given acutely during maintenance, as well as given intermittently following abstinence, decreased methamphetamine intake and active lever responses ([Figure 5](#F5){ref-type="fig"}). There are a number of possible reasons for the 5-HT~1B~R agonist-induced decreases in methamphetamine self-administration, including an effect on methamphetamine reinforcement value and/or incentive motivation, an effect on anxiety, or an effect on motor capability. The decrease in methamphetamine intake is unlikely due to impairments in motor capability, as treatment with the 5-HT~1B~R ligands did not alter spontaneous locomotion ([Figure 4](#F4){ref-type="fig"}) or inactive lever responses at the doses used in the present study. Furthermore, previous research from our laboratory has shown that CP 94,253 (0.3--10.0 mg/kg, s.c.) has no effect on sucrose reinforcement ([@CIT0036]). Although the antagonist SB 224,289 decreases cocaine-induced locomotion in drug naïve rats ([@CIT0014]), it has no effect on locomotion in rats with a history of cocaine self-administration ([@CIT0036], [@CIT0030]). Thus, it seems unlikely that CP 94,253 or SB 224,289 altered methamphetamine intake by impairing motor capability. We cannot rule out the possibility that CP 94,253 may have influenced methamphetamine intake nonspecifically by increasing anxiety-like behaviors rather than attenuating reinforcement per se. Indeed, previous studies have found that either 5-HT~1B~R agonists or antagonists can increase baseline and cocaine-induced anxiety-like behaviors ([@CIT0019]; [@CIT0014]; [@CIT0036]). It is important to note that the rats in these previous studies were naïve to the experimental procedures used to assess anxiety-like behaviors, which would likely maximize any potential drug effect. In contrast, rats tested for CP 94,253 effects on methamphetamine self-administration in the present study were well habituated to the testing environment, which would likely minimize potential anxiogenic effects. Furthermore, animals with increased exposure to stress, such as foot-shock, often increase rather than decrease drug intake ([@CIT0012]; [@CIT0001]; [@CIT0040]; [@CIT0020]), mitigating the idea that CP 94,253 may have induced anxiety-like states that interfered with responding. Finally, it seems that potential anxiogenic effects of CP 94,253 on reinforcement would manifest as a decrease in both sucrose and psychostimulant intake, yet CP 94,253 has been shown to selectively decrease cocaine intake ([@CIT0036]). We suggest that the most likely explanation for the agonist-induced decreases in methamphetamine intake in the present study is that 5-HT~1B~Rs modulate psychostimulant reinforcement and/or incentive motivation ([@CIT0038], [@CIT0030]). The decrease of methamphetamine intake in response to 5-HT~1B~R stimulation following abstinence may result from an attenuation of the expression of enhanced sensitivity to the reinforcing effects of methamphetamine. This explanation is consistent with our previous findings that 5-HT~1B~R agonists or 5-HT~1B~R overexpression by viral gene transfer attenuate cocaine self-administration following a period of abstinence, as well as reduce cocaine-seeking behavior ([@CIT0036], [@CIT0038], [@CIT0030]). Furthermore, these agonist effects are reversed by 5-HT~1B~R antagonists, including SB 224,289, supporting a 5-HT~1B~R mechanism. We were surprised that CP 94,253 attenuated methamphetamine intake prior to any abstinence given that this same treatment enhances cocaine intake prior to abstinence ([@CIT0036], [@CIT0030]). However, others have shown that 5-HT~1B~R agonists attenuate d-amphetamine intake without prolonged abstinence, as well as attenuate d-amphetamine-induced responding for conditioned reward ([@CIT0011]; [@CIT0010]; [@CIT0023]; [@CIT0025]). Therefore, the 5-HT~1B~R agonist enhancement of cocaine intake prior to abstinence may be due to differences in the pharmacological actions of cocaine vs amphetamines. While both amphetamines and cocaine inhibit and downregulate 5-HT, dopamine, and norepinephrine transporters ([@CIT0002]; [@CIT0043]), they interact differently with the transporters. Amphetamines, including methamphetamine, not only inhibit monoamine transport, but also redistribute intracellular monoamines by acting at the vesicular monoamine transporter causing release of monoamines into the cytosol while at the same time reversing monoamine transport across the plasma membrane, resulting in monoamine release ([@CIT0049]; [@CIT0046]; [@CIT0033]). Cocaine and amphetamines also interact at different sites on the dopamine transporter and produce differential effects on the releasable vesicular pool and on regulation of vesicular monoamine transporter ([@CIT0051]). The latter effects may result in differences in the releasable pool of dopamine after cocaine vs methamphetamine following acute or subchronic administration ([@CIT0004]). Although the specific mechanisms responsible for the attenuating effect of 5-HT~1B~R agonists on cocaine and methamphetamine self-administration are unclear, we hypothesize that such mechanisms may involve a dysregulation of 5-HT~1B~R functions ([@CIT0030]). 5-HT~1B~Rs are widely distributed in the brain ([@CIT0005]; [@CIT0018]; [@CIT0052]; [@CIT0006]) and are expressed as either 5-HT terminal autoreceptors or heteroreceptors on terminals of non-5-HTergic cells. In both cases, these receptors negatively couple to adenylyl cyclase activity via G-proteins and function to inhibit neurotransmitter release ([@CIT0013]; [@CIT0047]; [@CIT0022]; [@CIT0003]). Several manipulations in the mesolimbic system have provided evidence for a modulatory role of 5-HT~1B~Rs in psychostimulant addiction; specifically, overexpression of 5-HT~1B~Rs in the nucleus accumbens of rats facilitates the rewarding and reinforcing effects of cocaine ([@CIT0031]; [@CIT0038]). Furthermore, local activation of 5-HT~1B~Rs in the ventral tegmental area alters cocaine-induced increases in dopamine levels in the nucleus accumbens and cocaine-induced decreases in gamma-aminobutyric acid levels ([@CIT0035]; [@CIT0032]). Similarily, activation of 5-HT~1B~Rs in the nucleus accumbens dose-dependently decreases the rewarding and reinforcing effects of amphetamine ([@CIT0011]; [@CIT0010]). The present findings suggest that 5-HT~1B~Rs are potential targets for developing pharmacotherapies for psychostimulant addiction. Here we show that the clinically available anti-migraine medication zolmitriptan, which is a 5-HT~1B/1D~R agonist, attenuated methamphetamine intake. The attenuation of methamphetamine intake was likely mediated, at least in part, by stimulation of 5-HT~1B~Rs, although it is possible that 5-HT~1D~Rs may have also contributed to the attenuation effect. Zolmitriptan, unlike CP 94,253, has a higher affinity for 5-HT~1D~Rs (Ki = 0.63 nM) than for 5-HT~1B~Rs (Ki = 5.01 nM; [@CIT0027]). CP 94,253 also has affinity for 5-HT~1D~Rs (Ki = 49 nM; [@CIT0016]), and therefore 5-HT~1D~Rs may also contribute to its effects on methamphetamine self-administration. The effects of zolmitriptan on methamphetamine self-administration were not likely due to a decrease in general activity, as we did not observe any differences in inactive lever responses in our treatment groups. Furthermore, previous research found that zolmitriptan (1--30 mg/kg, i.p.) attenuates alcohol-induced aggression in mice but has no effect on locomotion ([@CIT0008]). In conclusion, this study provides evidence that the selective 5-HT~1B~R agonist, CP 94,253, attenuates methamphetamine self-administration pre- and postabstinence under several schedules of reinforcement and in an antagonist-reversible manner. These findings build upon previous research demonstrating similar effects of 5-HT~1B~R agonists on d-amphetamine self-administration ([@CIT0011]; [@CIT0010]; [@CIT0023]; [@CIT0025]), and together the effect of the agonists on self-administration of amphetamines contrasts with the enhanced cocaine self-administration that has been observed prior to any abstinence ([@CIT0036], [@CIT0030]). These results suggest that 5-HT~1B~R agonists may differentially modulate cocaine and methamphetamine self-administration initially, but that after a period of abstinence, the agonists inhibit the reinforcing effects of both psychostimulants. The latter findings suggest that 5-HT~1B~R agonists may have potential therapeutic effects for psychostimulant abuse. In addition, we have provided evidence that the FDA-approved 5-HT~1D/1B~R agonist, zolmitriptan, also attenuates methamphetamine self-administration both pre- and postabstinence. Our findings suggest that 5-HT~1B~R agonists warrant further investigation as putative treatments of psychostimulant use disorders. Important future directions include determining the neural circuitry involved in the agonist effects, whether the effects are also observed in female rats, and whether the effects are observed in rats given more extensive access to the stimulants and more extensive abstinence. Statement of Interest ===================== None. We thank Nathan Pentkowski for his input on this study and Juliette Venault, Allegra Campagna, Katelin Ennis, Thomas Benson, Jennifer Taylor, and JP Bonadonna for their technical assistance. This work was supported by the National Institute on Drug Abuse (DA011064 to J.L.N. and DA025606 to M.F.O.), and the National Institutes of General Medical Sciences for ASU Post-baccalaureate Research Education Program for Biomedical Research and Initiative to Maximize Student Development (GM071798 and GM099650 to R.G.). [^1]: Correspondence: Janet L. Neisewander, PhD, School of Life Sciences, ISTB1, Rm 429, 427 E Tyler Mall, Arizona State University, Tempe, AZ 85287-4501 (<janet.neisewander@asu.edu>).

Dear Colleagues, We have appreciated your insightful comments concerning our work,^[@CR1]^ in particular your observation that there may be a distal radio-ulnar joint dislocation. Your comments prompted us to review all cases,^[@CR1]^ and we thank you for that. Dislocation of the distal radio-ulnar joint is described as when the position of the ulna is volar or dorsal to the radius.^[@CR2]^ After reviewing all cases, we found some patients with chronic Monteggia fracture treated by ulnar osteotomy and lengthening by external fixator that seemed to have dislocation of the distal radio-ulnar joint, on radiographs, at last follow-up visit.^[@CR3]^ However, it is very difficult to make the diagnosis of distal radio-ulnar joint dislocation from radiographs alone as the position of the forearm under radiographic examination may also contribute to this appearance. We performed posteroanterior and lateral radiographs of the forearm (elbow and wrist included) in patients with chronic Monteggia fracture during follow-up.^[@CR4]^ In order to maintain the forearm in such a position, the wrist may require additional pronation or supination, which will potentially contribute to the radiographic appearance of distal radio-ulnar joint dislocation. In addition, and most importantly, clinical evaluation did not reveal any wrist pain nor limitation of movement in any of the patients. Moreover, following your remarks, we also did a comparison between radiographs of normal children and age-matched patients with operated chronic Monteggia fracture, and we could not detect any significant difference between the two sets of radiographs ([Fig. 1](#F1){ref-type="fig"}). {#F1} In particular, we paid special attention to the case labelled as 'Figure 5' in the original article.^[@CR1]^ We examined the patient again, and clinical evaluation did not reveal any wrist pain or limitation of movement. We would like to thank our colleagues again for their very interesting remarks, and we confirm the absence of distal radio-ulnar joint dislocation. We are currently following this cohort of patients for the long term, and we will publish our results, with particular attention to both the function and the radiographic appearance of the wrist. Yours sincerely The Authors This article is distributed under the terms of the Creative Commons Attribution-Non Commercial 4.0 International (CC BY-NC 4.0) licence (<https://creativecommons.org/licenses/by-nc/4.0/>) which permits non-commercial use, reproduction and distribution of the work without further permission provided the original work is attributed. Compliance with ethical standards {#Sec1} ================================= ICMJE Conflict of interest statement {#Sec1.1} ------------------------------------ The authors declare no conflict of interest.

Introduction ============ Low back pain is a significant health problem with approximately one-half of adults reporting back pain during a given year \[[@ref1]-[@ref3]\]. Low back pain that persists for longer than 3 months is considered chronic \[[@ref4]\], and the longer the pain persists the greater the risk for long-term disability \[[@ref5]\]. Chronic back pain is associated with functional limitations, social isolation, unemployment, and lost productivity \[[@ref5]-[@ref7]\], making it one of the most costly health conditions in the United States \[[@ref8]-[@ref11]\]. Exercise therapy has proven benefits for managing chronic back pain \[[@ref12]-[@ref14]\]. Specifically, exercise can prevent recurrence, reduce pain, improve function, and decrease disability for patients with chronic back pain \[[@ref12],[@ref13],[@ref15]-[@ref19]\]. It is also generally recognized that, to be effective, patients have to be willing and able to perform the recommended exercise and for continuing benefits remain adherent to the exercise program \[[@ref18],[@ref20],[@ref21]\]. However, there are few efficient and effective strategies to help patients engage in exercise therapy for managing their chronic low back pain. Internet-based programs are an increasingly popular option for promoting healthy behaviors, such as those related to diet and exercise, and for delivering behavior change interventions \[[@ref22]-[@ref24]\]. Studies have shown that the Internet can be used to successfully promote weight loss \[[@ref25]\], increase physical activity \[[@ref26]\], and improve patient self-activation \[[@ref27]\] or self-management behaviors \[[@ref22]\]. Studies of Internet-based interventions for pain, while somewhat limited, show a generally positive effect on pain levels and, to some extent, activity \[[@ref27]-[@ref30]\]. No studies, however, have focused primarily on exercise to reduce pain-related disability and improve patient function. We conducted a randomized trial to investigate whether a pedometer-based, Internet-mediated intervention designed to assist patients with initiating and maintaining a regular walking program would reduce pain-related disability and functional interference among patients with chronic back pain at 6 months and over a 12-month timeframe. Methods ======= Design Overview --------------- We conducted a parallel group randomized controlled trial with participants allocated in a 1:1 ratio to the intervention or enhanced usual care (NCT00694018). This research was approved by the Department of Veterans Affairs (VA), Ann Arbor Healthcare System institutional review board. The study protocol, including conceptual framework, is described in detail elsewhere \[[@ref31]\], with key elements summarized below. There were no significant changes in methods following study initiation. Setting and Participants ------------------------ Participants were recruited from one VA Healthcare System between May 2009 and March 2011. Eligible participants were aged 18 years or older and identified through provider referrals to back class and use of the VA electronic medical record system. Specifically, we identified patients who had 2 or more outpatient encounters in the previous 12 months with a diagnosis of back pain with no neurologic findings (ICD-9-CM codes 724.2, 724.5, 846.0-846.9). Study staff used a standardized protocol to screen potential participants by phone or, for a minority of patients who could not be reached by phone, in person when they arrived for back class. Eligibility criteria included: (1) persistent back pain \>3 months, (2) self-reported sedentary lifestyle (defined as \<150 minutes of physical activity per week in accordance with the US Department of Health and Human Services 2008 Physical Activity Guidelines for Americans \[[@ref32]\]), (3) weekly access to a computer with a USB port and Internet access, (4) ability to provide written informed consent and communicate in English, (5) community residence, (6) ability to walk at least one block, and (7) report they are not pregnant. Prior to participation, all eligible patients had to attend back class and obtain medical clearance. Back class, led by a physical therapist, provided general education about managing back pain. Participants also performed back-specific strengthening and stretching exercises under the supervision of a physical therapist. Eligible participants then attended a study enrollment session at which time they provided written informed consent and were told they were helping test an Internet-based program and would be assigned to one of two groups: (1) an enhanced care group that would upload pedometer data weekly and have access to a study website and computer discussion group (Internet support group), or (2) a usual care group that would upload pedometer data monthly (monthly upload group). All participants received an uploading pedometer (the Omron HJ-720ITC, which stores 42 days of step-count data and has an embedded USB port \[[@ref33]\]), along with general guidance on using the pedometer and instructions for logging onto and uploading data to the study website. To establish a baseline step count that was not influenced by use of the pedometer information, participants were instructed to wear their pedometer for 7 days with the display covered before completing their first upload. Randomization ------------- After completing the baseline survey, uploading 7 days of useable pedometer data, and receiving medical clearance, each participant was randomly allocated in a 1:1 ratio to the intervention or usual care group by a computer program (using a random number generator). The program also generated an email message to inform participants about their group assignment (Internet support or monthly upload) and instructions to remove the sticker covering the pedometer display. Intervention ------------ The study intervention, based on the Stepping Up to Health program \[[@ref31],[@ref34]\], consisted of three primary components: (1) the uploading pedometer, (2) a website that provided automated goal setting and feedback, targeted messages, and educational materials, and (3) an e-community \[[@ref31]\] (see [Multimedia Appendix 1](#app1){ref-type="app"}). The conceptual framework and more detailed description of the intervention components are published elsewhere \[[@ref31]\]. Briefly, participants were instructed to wear their pedometer from the time they got up in the morning until they went to bed. Intervention participants then received weekly email reminders to upload their pedometer data, which was used to establish weekly individualized walking goals. Each participant's goal was based on their average total step count in the prior week with a fixed number of steps (800) added to promote a gradual increase in walking for the following week. The step count goal was emailed to the participant each week and posted on the study website. The study website, which was fully accessible to intervention participants, also included graphical and written feedback about their progress toward their walking goals and contained pain- or activity-related motivational and informational messages. These messages included quick tips, which changed every other day, and weekly updates about topics in the news. Back class materials, which included handouts about topics such as body mechanics, use of cold packs, lumbar rolls, and good posture, as well as a video demonstrating specific strengthening and stretching exercises were also available on the website. Finally, the website based e-community or forum allowed participants to post suggestions, ask questions, and share stories. Topics discussed included mental health concerns, such as depression, strategies for walking such as walking the dog or interesting hiking trails, walking during hot weather and cold weather, and use of alternative pain management strategies such as massage. Research staff participated in and monitored the forum posts as well as used the forum as a venue to generate competitions to encourage meeting walking goals. Enhanced Usual Care ------------------- Usual care participants also received the uploading pedometer and monthly email reminders to upload their pedometer data. However, they did not receive any goals or feedback and their access to the study website was limited to completing surveys and reporting adverse events only. Monitoring of Adverse Events ---------------------------- Both groups were encouraged to report any health problems via the website, email, or phone. Four weeks after randomization and every 8 weeks thereafter, participants were prompted to complete a survey that asked about specific adverse events (eg, heart attack) and symptoms such as shortness of breath. This information was closely monitored and participants with potentially serious health-related problems were contacted for further assessment and follow-up. Outcomes and Follow-Up ---------------------- Outcomes were measured at baseline, 6 months, and 12 months using a survey administered through the study website, or by a mailed questionnaire if the participant could not complete the computerized instrument. The prespecified primary outcome was pain-related disability at 12 months, as measured using the back pain-specific Roland Morris Disability Questionnaire (RDQ) \[[@ref35]\], and a generic pain-related function measure from the Medical Outcomes Study (MOS) \[[@ref36]\]. The RDQ, a 24-item scale with higher scores indicating greater disability, has been widely used in back pain studies as a measure of self-perceived disability \[[@ref35],[@ref37]-[@ref39]\]. The MOS measure assesses the effect of pain on mood and behaviors as well as pain severity, with higher scores also indicating greater functional interference \[[@ref36]\]. Pain intensity, a secondary outcome, was evaluated using a numeric rating scale with standard anchors (0="no pain" and 10="worst pain imaginable") \[[@ref40]\]. Walking, also a secondary outcome, was measured as the average number of steps per day over the past 7 days using step-count data collected through the pedometer uploads. Other secondary outcomes included pain-related fear-avoidance, measured using the Fear-Avoidance Beliefs Questionnaire physical activity subscale (higher scores reflect higher levels of fear-avoidance) \[[@ref41]\], and self-efficacy for exercise, measured using the Exercise Regularly Scale, with higher scores indicating higher levels of self-efficacy \[[@ref42]\]. Additional data collected at baseline included age, gender, race, employment status, education level, relationship status, average household income, body mass index, and use of narcotic medications for pain management. An administrative interface to the website provided data on the number of pedometer uploads and website log-ins. Sample Size ----------- Sample size was based on the RDQ score as the primary outcome with a minimally detectable and clinically meaningful effect size determined as a difference of 0.4 standard deviation (SD) in change scores or a 2-point difference, based on published data \[[@ref38],[@ref43],[@ref44]\]. To detect a difference of 0.4 SD with 80% power using a two-sided 0.05 level 2 group *t* test, we sought to enroll 130 subjects in each group, to allow for an attrition rate of 25% at 1 year. Statistical Analysis -------------------- The analyst assessing final trial outcomes was blinded to study assignment. All analyses were conducted using an intent-to-treat approach with participants analyzed according to original group assignment. We conducted both complete and all case analyses to assess differences between groups in change in RDQ at 6 and 12 months. The complete case analysis was conducted using multiple linear regression models with adjustment for baseline values of the RDQ. The all case analysis was conducted using linear mixed-effects models, allowing us to use data from all participants and provide an unbiased estimate of the outcome, assuming data are missing at random \[[@ref45]\]. For example, for our 12-month analysis, RDQ scores at baseline and 12 months were used as dependent variables, with the primary independent variables consisting of an indicator for the intervention group and an interaction term of time by intervention group. Each participant's data was modeled using a random intercept to allow within-patient correlation of the repeated measures. Adjustment for covariates was only planned if an imbalance was found between groups at baseline. We also conducted a post hoc subgroup analysis of participants with baseline RDQ scores of ≥4. As a pragmatic trial we did not screen based on RDQ scores, and some participants had baseline scores that were very low or even 0. Thus, to assess the effect of the intervention on participants reporting at least modest levels of back pain-related disability at baseline, we conducted a subgroup analysis of those with baseline RDQ scores of ≥4 using the same methods previously described. Analyses were conducted using Stata 11.2 and all reported *P* values are from adjusted analyses. Results ======= Summary ------- Over 1400 potential participants ([Figure 1](#figure1){ref-type="fig"}) were assessed for eligibility. Primary reasons for ineligibility were lack of regular access to a computer or the Internet (n=310) and being too physically active (n=159). Of those determined to be eligible, 229 completed all of the steps in the enrollment process, with 111 randomly allocated to the Internet-mediated intervention and 118 to enhanced usual care. Primary outcome data were provided by 91% of intervention and 90% of usual care participants at 6 months, and by 92% of those in the intervention group and 89% receiving usual care at 12 months. Baseline Characteristics ------------------------ Participants were predominantly male and white, with an average age of 51 years ([Table 1](#table1){ref-type="table"}). The majority had completed some college, were either married or living with someone as a couple, and the mean body mass index was over 30. At baseline, less than 40% of participants reported being employed full- or part-time and over 40% reported taking narcotic medications for their back pain. None of the observed differences in baseline characteristics were statistically significant. Primary Outcomes ---------------- At baseline, mean RDQ scores were greater than 9 in both groups ([Table 1](#table1){ref-type="table"}), indicating moderately severe back pain-related disability. The mean RDQ score at 6 months was 7.2 for intervention participants compared to 9.2 for those in usual care ([Figure 2](#figure2){ref-type="fig"}), an adjusted difference of 1.6 (95% CI 0.3-2.8, *P*=.02) for the complete case analysis and 1.2 (95% CI -0.09 to 2.5, *P*=.07) for the all case analysis ([Table 2](#table2){ref-type="table"}). When restricted to the subgroup with at least moderate back pain at baseline (RDQ score ≥4) ([Figure 2](#figure2){ref-type="fig"}, [Table 2](#table2){ref-type="table"}), patients in the intervention had a significant improvement in back pain-related disability compared to the control group, an adjusted difference of approximately 2 in both the complete (1.9, 95% CI 0.5-3.3, *P*=.01) and all case (1.7, 95% CI 0.3-3.0, *P*=.02) analyses. RDQ scores continued to decline between 6 and 12 months in both groups and, while scores for the intervention group remained lower than for usual care, at 12 months these differences were no longer statistically significant. The MOS function measure also suggested greater improvements in function for intervention compared to usual care participants at 6 months ([Figure 2](#figure2){ref-type="fig"}), but none of the adjusted differences were statistically significantly different. Secondary Outcomes ------------------ At baseline, pain severity was rated at approximately 6 on a 0-10 scale by both intervention and usual care participants ([Table 1](#table1){ref-type="table"}). Reported pain levels decreased in both groups at 6 months and remained lower than baseline at 12 months. The greatest change occurred between baseline and 6 months among those in the intervention group (6.0-4.7 vs 6.1-5.2 in the control group), although the adjusted difference between arms of 0.5 was not significant ([Table 3](#table3){ref-type="table"}). Average step counts of slightly more than 4000 steps per day at baseline in each group increased at 6 months for intervention patients, with an adjusted difference between groups of more than 700 steps. By 12 months, however, the adjusted difference between groups was only 100-200 steps. Exercise self-efficacy scores appeared to be the same or lower (worse) for both groups at 6 months, although the decrease was significantly less for those in the intervention compared to the control group, an adjusted difference of 0.8 (95% CI 0.24-1.4, *P*=.01) in the complete case analysis and 0.7 (95% CI 0.12-1.2, *P*=.02) for the all case analysis ([Table 3](#table3){ref-type="table"}). This difference did not persist at 12 months. There was no difference between groups in the physical activity fear-avoidance scale at any time point. Intervention Engagement ----------------------- Intervention participants uploaded pedometer data at least once per week for a median of 32 weeks (62% of the recommended time), although more than 25% of participants uploaded data for at least 42 weeks (80% compliance). However, intervention participants logged into the website at least once per week for a median of only 20 weeks (38% of the recommended time), with approximately 20% logging in for at least 42 weeks. Adverse Events -------------- During the study, approximately 600 adverse events were reported by participants (250 by those in usual care and nearly 350 by those in the intervention). These events ranged from calluses to chest pain. Worsening back pain, the most frequently reported event, accounted for 29% of events reported by the usual care group and 25% of those reported by the intervention group. Overall, more musculoskeletal events (n=112) were reported than cardiovascular events (n=85), and musculoskeletal injuries were more likely to be reported by participants in the intervention group compared to those in usual care. However, no major study-related adverse events (eg, heart attack) were identified for either group. ###### Participant baseline characteristics. --------------------------------------------------------------------------------------------------------------------------------------------------------------- Characteristic \ Internet-mediated intervention\ Enhanced usual care\ (n=111) (n=118) ---------------------------------------------------------------------- ------------------------------- --------------------------------- ---------------------- Age (y), mean (SD) \ 51.2 (12.5) 51.9 (12.8) Male (%) \ 89 86 **Race (%)** \ \ \ \ White 74 86 \ Black 13 9 \ Other or prefer not to answer 14 5 **Education level (%)** \ \ \ \ High school or less 29 25 \ Some college 56 59 \ 4 years of college or more 16 16 Married or living with a partner (%) \ 59 68 Employed full-time or part-time (%) \ 39 31 **Annual household income (%)** \ \ \ \ \<US \$10,000 18 13 \ US \$10,000-\$39,999 61 54 \ ≥ US \$40,000 21 33 Take narcotic medications for back pain (%) 41 49 General health status, fair or poor (%) 41 43 Body mass index, mean (SD) \ 30.6 (5.7) 31.6 (5.5) RDQ score (0-24)^a^, mean (SD) 9.1 (6.0) 9.8 (5.7) MOS pain-related functional interference score (0-100)^a^, mean (SD) 48.5 (18.6) 51.8 (16.3) Level of pain severity, 0-10 scale^a^, mean (SD) 6.0 (1.9) 6.1 (1.6) Daily step counts, mean (SD) 4492.9 (2749.9) 4321.9 (2285.4) Exercise self-efficacy score, 1-10^b^, mean (SD) 6.8 (2.1) 6.5 (2.3) Physical activity fear-avoidance behavior scale, 0-28^a^, mean (SD) 13.9 (5.9) 15.1 (6.0) --------------------------------------------------------------------------------------------------------------------------------------------------------------- ^a^lower scores are better ^b^higher scores are better {#figure1} ###### Primary outcomes, back pain-specific and general pain-related function. ------------------------------------------------------------------------------------------------------------------------------------------------------- Primary outcome \ Adjusted between-group difference^a^\ (95% CI) ------------------------------------------------------------------ ----------- --------------------------------------- ----- -------------------- ----- **RDQ score (0-24)** \ \ \ \ \ \ 6 months 1.6 (0.3 to 2.8) .02 1.2 (-0.09 to 2.5) .07 \ 12 months 1.2 (-0.3 to 2.7) .11 0.7 (-0.8 to 2.2) .38 **MOS pain-related functional interference score (0-100)** \ 6 months 3.6 (-0.51 to 7.7) .09 2.5 (-1.5 to 6.5) .23 \ 12 months 0.1 (-4.0 to 4.2) .97 -1.4 (-5.4 to 2.5) .48 **Subgroup with RDQ scores** ≥**4 at baseline RDQ score (0-24)** \ 6 months 1.9 (0.5 to 3.3) .01 1.7 (0.3 to 3.0) .02 \ 12 months 1.1 (-0.6 to 2.7) .20 0.8 (-0.8 to 2.4) .34 **MOS pain-related functional interference score (0-100)** \ 6 months 4.6 (-0.1 to 9.3) .05 3.8 (-0.7 to 8.3) .10 \ 12 months -0.5 (-5.0 to 4.0) .83 -1.5 (-5.8 to 2.8) .49 ------------------------------------------------------------------------------------------------------------------------------------------------------- ^a^Adjusted for baseline values and calculated as pain or function in enhanced usual care group minus Internet-mediated intervention group so that positive scores reflect greater improvement in the intervention group. ###### Secondary outcomes. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Secondary outcome \ Mean (SD) Adjusted between-group difference^a^\ (95% CI) --------------------------------------------------------- ----------- ----------------- --------------------------------------- -------------------- ----- ------------------- ----- **Level of pain severity, 0-10 scale** ^b^ \ 6 months 4.7 (2.1) 5.2 (2.1) 0.5\ .06 0.5\ .07 (-0.01 to 0.98) (-0.03 to 0.9) \ 12 months 5.4 (2.2) 5.6 (2.0) 0.1\ .81 0.04\ .86 (-0.4 to 0.5) (-0.4 to 0.5) **Daily step counts** ^c,d^ \ 6 months 5370.0 (3180.8) 4682.5 (2925.0) 725.5\ .12 724.0\ .08 (-193.6 to 1644.7) (-75.2 to 1523.2) \ 12 months 4681.8 (3000.6) 4758.1 (2991.1) 122.4\ .75 143.4\ .64 (-623.9 to 868.6) (-460.2 to 747.1) **Exercise self-efficacy score** ^c^ \ 6 months 6.7 (2.4) 5.7 (2.5) 0.8\ .01 0.7\ .02 (0.24 to 1.4) (0.12 to 1.2) \ 12 months 6.4 (2.6) 5.9 (2.3) 0.3\ .32 0.2\ .55 (-0.3 to 0.9) (-0.4 to 0.74) **Physical activity fear-avoidance behavior scale** ^b^ \ 6 months 13.2 (6.0) 14.0 (5.9) 0.6\ .42 -0.1\ .94 (-0.88 to 2.1) (-1.6 to 1.5) \ 12 months 13.3 (6.7) 15.1 (6.1) 1.1\ .18 0.6\ .50 (-0.5 to 2.7) (-1.1 to 2.2) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ^a^Adjusted for baseline values and calculated as pain or function in enhanced usual care group minus Internet-mediated intervention group so that positive scores reflect greater improvement in the intervention group. ^b^lower scores are better ^c^higher scores are better ^d^Pedometer data: intervention (n=84 at 6 months, n=78 at 12 months), usual care (n=70 at 6 months, n=68 at 12 months). {#figure2} Discussion ========== Principal Findings ------------------ Improving management of chronic pain is a significant public health challenge and moral imperative according to a recent Institute of Medicine report \[[@ref8]\]. More than 1 million adults in the United States have chronic pain, with low back pain being the most frequently reported condition \[[@ref8]\]. Our findings show that an automated, Internet-mediated walking intervention may help to reduce back pain-related disability among patients with chronic back pain, although the benefits did not persist for the entire 12-month study period. Improvement was greatest for those individuals reporting moderate to severe levels of pain-related disability at baseline. The functional results observed are generally similar to those found in other recent studies of non-invasive interventions, such as yoga and massage \[[@ref46],[@ref47]\]. These studies also tend to show more rapid improvements for those receiving the intervention but with gradual improvements over time for those in usual care. Moreover, if we employ the criteria proposed by Jordan and colleagues \[[@ref48]\] to classify patients as clinically improved or at least possibly improved (compared with not improved), as defined by a reduction in the RDQ score of at least 30% at 6 months, we find that 46% of those in the intervention versus 27% in the control group would meet this definition. Although we did not have a global health question and so are unable to isolate what proportion would qualify as definitely improved, this classification generally corresponds with other measures that suggest clinical improvement, such as return to work, less pain, improved function, and fewer physician visits \[[@ref48]\]. Thus, we believe that our findings suggest that automated, remotely delivered interventions can be effectively used to promote a more rapid reduction in back pain-related disability and supplement care for patients with chronic low back pain. Further investigation is needed, however, to understand the characteristics of patients who had an early or enduring response to the intervention so that we may better target patients most likely to benefit and broaden the response. Given the proven benefits of exercise for managing low back pain \[[@ref19]\], a key component of the intervention focused on increasing daily step counts (ie, walking). During the first 6 months of the study, we saw an increase of nearly 700 more steps or one-third of a mile per day among intervention compared to usual care participants. Although not a statistically significant difference, we believe that even modest increases in activity can be beneficial. As one intervention participant noted: "I didn't know what the walking could do for me. But\...it seemed to alleviate my back pain\...the true test came when I had to go off the program because of my illness and the back pain returned. In fact, just up until recently when I had resumed walking." On the other hand, step-count improvements were not sustained for the entire 12 months and poor adherence or declining engagement, as demonstrated by the percentage of patients who uploaded or logged into the website, could in part account for the lack of sustained benefit or added improvement over time. Although we do not know specific reasons for this lack of participation, these data suggest that additional strategies to keep people active and engaged may be needed. This could include, for example, an online coaching component, which has been shown to improve adherence to other types of behavioral changes \[[@ref47]-[@ref49]\]. Our monitoring of adverse events showed a higher number of reported events by intervention participants. This information was, however, collected solely through self-report and we expect that some of the difference in the overall number of events reported between groups could be due to our more frequent contact with intervention participants via email and through the website. In addition, despite the higher level of musculoskeletal events reported by intervention participants, we found no evidence that the intervention led to excessive harms. Thus, even though more work to understand the circumstances for those reporting musculoskeletal problems or worsening back pain may be required, these findings add to the evidence base to support walking as a generally safe and potentially effective intervention for some patients with chronic low back pain \[[@ref49]-[@ref52]\]. Other potential mechanisms of action are less clear. Despite a marginally greater decrease in pain levels among intervention participants at 6 months, this effect did not persist at 12 months. In addition, while there was a significant difference between groups in self-efficacy for exercise at 6 months, rather than the hypothesized improvement for those in the intervention, both groups reported lower levels of self-efficacy. However, the decline was smaller for those receiving the intervention. The reason for the decrease is not entirely clear but may be largely due to an unrealistic assessment of self-efficacy at baseline \[[@ref53]\]. Limitations ----------- Among the strengths of our study are the high rate of participant follow-up and our collection of detailed adverse event information. This study also has several limitations. First, patients were recruited from only 1 medical center and the sample was predominantly male. Although more than 10% of participants were female, which is relatively high for studies using a general VA patient population, the number is not sufficient for a formal subgroup analysis. However, based on trials of similar types of interventions, we expect this approach could be even more effective among women \[[@ref54]\]. Second, we are not able to directly compare our results to other types of back pain interventions (eg, yoga), although as previously noted the general trajectory of our primary outcome (RDQ score) appears consistent with recent trials in this area. Third, although a consistent data collection format is generally recommended \[[@ref55]\], we used both Internet-based and paper surveys. However, prior research has demonstrated similar psychometric properties between Internet and paper-and-pencil questionnaires \[[@ref55]\] and specifically equivalence for our primary outcome \[[@ref56]\]. We also believe that using both modes helped to ensure a high follow-up rate. Finally, as a multifaceted intervention, we are not able to determine which elements were most effective and can only draw conclusions about the program as a whole. Nonetheless, our results highlight the importance of providing active support (eg, goal setting and feedback) to encourage walking as compared with simply giving someone a pedometer to track step counts. Conclusions ----------- In sum, our findings indicate that a facilitated walking intervention that uses an uploading pedometer and the Internet may help to reduce back pain-related disability among patients with chronic back pain, at least in the short term. Additional support, however, is likely needed to ensure continuing improvements long term. Nevertheless, this type of primarily automated intervention can be used to deliver care with broad reach and could be an efficient way of delivering or supplementing care provided through traditional facility-based programs. This project was funded through a grant from the Department of Veterans Affairs, Health Services Research and Development Service (IIR 07-177). The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the US government. The study sponsors had no role in the design or conduct of the study; the collection, management, analysis, and interpretation of the data; or the preparation, review, or approval of the manuscript. Dr Piette is a VA Senior Research Career Scientist. Dr Richardson was supported in part by a career development award from the National Heart, Lung, and Blood Institute (K23HL075098) and a Physician Faculty Scholars Program award from the Robert Wood Johnson Foundation (57408). Development of the Stepping Up to Health intervention platform was supported by pilot grant funding from the following University of Michigan centers: Michigan Diabetes Research and Training Center (P60 DK020572), the Center for Health Communications Research (P50 CA101451), and the Michigan Institute for Clinical and Health Research (NIH \#UL1RR024986). Technical development of the Stepping Up to Health website could not have been accomplished without the expertise of our Web developers, Michael Hess, with assistance from Elizabeth Wilson and Adrienne Janney. Thanks also to Jill Bowdler for assistance with manuscript preparation. Finally, the authors would like to thank Robert Kerns, Jack Rosenberg, and Robert Werner for their assistance during the design of this study, as well as the Physical Therapy Service at the VA Ann Arbor Medical Center for their help during the conduct of the project. Authors\' Contributions: SLK, JDP, EAK, HMK, and CRR were involved in the conception and design of the study. RK and MH were involved in study data collection. SLK, RH, and HMK conducted study analyses. All authors were involved in drafting the manuscript and revising it for critically important content. All authors have read and approved the final manuscript. Conflicts of Interest: None declared. Intervention components. CONSORT-EHEALTH checklist V1.6.2 \[[@ref57]\]. MOS : Medical Outcomes Study RDQ : Roland Morris Disability Questionnaire SD : standard deviation VA : Department of Veterans Affairs

Introduction {#sec1-1} ============ Cutaneous botryomycosis is a rare and chronic suppurative bacterial infection with commonest causative organism being *S. aureus*.\[[@ref1][@ref2]\] Most of published literature on botryomycosis are case reports. It has been commonly reported in immunocompromised adults and usually occurs over trauma prone areas following local injury or at post-operative sites.\[[@ref3]\] Skin lesions usually present as papules, nodules, fistulas, abscesses, and ulcers with seropurulent discharge.\[[@ref3]\] However, characteristic morphological features giving clue to the diagnosis of cutaneous botryomycosis are unknown. It is usually treated with specific antibiotic based on culture sensitivity either single or combination of multiple drugs for prolonged duration. We present three immunocompetent patients of cutaneous botryomycosis with lesions present on various parts of body with different distribution patterns and evolution of lesions; and were successfully treated with sensitive antibiotic monotherapy. Case Report {#sec1-2} =========== Three immunocompetent patients (one child and two adults) presented with subcutaneous swellings with seropurulent discharge at various sites of body. Case 1 and 2 had localized lesions whereas case 3 had lesions in a sporotrichoid distribution. Case 1, 15-year old female child came with multiple erythematous mildly painful, subcutaneous swellings over left side of forehead and upper eyelid with ulceration and seropurulent discharge since 4 years \[[Figure 1a](#F1){ref-type="fig"}\]. There was preceding history of trauma over forehead 6 years back. The lesions started as subcutaneous swellings at one end of the scar and underwent ulceration on surface of swelling with few lesions developing sinuses giving seropurulent discharge after around 4 to 6 weeks. Surgical drainage was done on two occasions within one year without any significant improvement. Based on the history and clinical examination, differential diagnoses of botryomycosis, actinomycetoma or eumycetoma, scrofuloderma were considered. Routine haematological parameters were within normal limits. HIV 1and 2 were negative. X-ray of skull didn't reveal any bony abnormalities. Mantoux test was negative. Culture of discharge and tissue showed growth of coagulase negative staphylococcus sensitive to doxycycline. Histopathology from the nodule showed bacterial colonies in dermis with an eosinophilic rim at the periphery suggestive of Splendore--Hoeppli phenomenon \[Figure [1b](#F1){ref-type="fig"} and [c](#F1){ref-type="fig"}\]. A diagnosis of botryomycosis was made and oral doxycycline 100 mg twice daily was started. The lesions resolved completely after 4 weeks \[[Figure 1d](#F1){ref-type="fig"}\]. We continued the drug for 6 more weeks after the clinical cure. The patient didn't develop any side effects to doxycycline. Currently the patient is under follow up without any recurrence in last 5 months. Demographic details, and morphology and evolution of lesions in all cases have been described in [Table 1](#T1){ref-type="table"}. {#F1} {#F2} {#F3} {#F4} {#F5} ###### Demographic details, history, evolution, morphology, treatment and outcome of patients Characteristics Case 1 Case 2 Case 3 ----------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Age/Sex 15 years, Female 38 years, Female 28 years, Female Duration 4 years 5 years 5 years Sites Forehead and left upper eye lid Upper back Left knee and inguinal area Prior history of trauma Yes No Yes Evolution Subcutaneous swellings followed by ulceration, sinus on surface, seropurulent discharge after around 4 to 6 week Subcutaneous swellings, with subsequent appearance of multiple papules and nodules on the surface of the swellings at one to two weeks interval, Surface ulceration of individual papules and nodules with seropurulent discharge at 7 to 10 days interval Subcutaneous swelling with subsequent appearance of multiple papules and nodules on the surface of large swelling after two months .Surface ulceration of individual papules and nodules, sinus formation with seropurulent discharge. Morphology Forehead and left upper eye lid-Multiple swellings with ulceration and crusting Diffuse swelling on upper back and nape of the neck, Multiple papules, nodules, nodules with surface ulceration, sinuses, seropurulent discharge, Nodules with pedunculated granulation tissue (pyogenic granuloma like morphology), hypopigmented and depressed scars on the surface of large swelling \[Figure [2a](#F2){ref-type="fig"} and [b](#F2){ref-type="fig"}\]. Subcutaneous lobulated swellings over left knee Studded with multiple papules, nodules, nodules with ulceration, ulcers with granulation tissue (pyogenic granuloma like) and multiple sinuses \[Figure [3a](#F3){ref-type="fig"} and [b](#F3){ref-type="fig"}\]. Proximal thigh- a plate like mass studded with papules, nodules, scars and sinuses \[[Figure 3c](#F3){ref-type="fig"}\]. Plate like subcutaneous mass on the nape of neck \[[Figure 2b](#F2){ref-type="fig"}\]. Normal skin in-between knee and proximal thigh \[[Figure 3d](#F3){ref-type="fig"}\] Investigations Pus and tissue culture- coagulase negative staphylococcus sensitive to doxycycline Pus and tissue culture-coagulase negative staphylococci sensitive to ciprofloxacin. Pus and tissue culture-methicillin sensitive*staphylococcus* *aureus* sensitive to doxycycline HIV 1 and 2-negative Mantoux-negative KOH and fungal culture of pus-negative HIV 1 and 2-negative Mantoux-negative KOH and fungal culture of pus-negative HIV 1 and 2-negative Mantoux-negative KOH and fungal culture of pus-negative Histopathology- bacterial colonies in dermis with an eosinophilic rim at the periphery suggestive of Splendore-Hoeppli phenomenon Histopathology-clustered basophilic granules surrounded by prominent eosinophilic club of Splendore-Hoeppli material deep in dermis \[[Figure 2c](#F2){ref-type="fig"}\]. Histopathology- basophilic bacterial granules in dermis with eosinophilic rim suggestive of SHP \[[Figure 4](#F4){ref-type="fig"}\]. Treatment Oral doxycycline 100 mg twice daily Oral Ciprofloxacin 500 mg twice daily Oral doxycycline 100mg twice daily Outcome Complete resolution following 4 weeks of doxycycline Continued doxycycline for 6 more weeks after the clinical cure. No side effects to doxycycline Complete resolution following 8 weeks of ciprofloxacin \[[Figure 2d](#F2){ref-type="fig"}\]. Continued ciprofloxacin for 6 weeks more after clinical cure No side effects to ciprofloxacin. 60% and 80% improvement of knee and thigh lesions respectively following 20 days of doxycycline \[[Figure 5](#F5){ref-type="fig"}\]. Discussion {#sec1-3} ========== Botryomycosis can involve skin and viscera. Around 200 cases have been reported from all over the world.\[[@ref3]\] The present report is series of three cases of cutaneous botryomycosis. Majority of reported cases of cutaneous botryomycosis are immunocompromised adults and has been reported rarely in children.\[[@ref4]\] All our cases were immunocompetent with one child and two adults. Common organisms causing botryomycosis include *Staphylococcus aureus, Pseudomonas aeruginosa*, coagulase-negative staphylococci, *Streptococcus spp, Escherichia coli* and *Proteus spp*.\[[@ref3]\] In the present series, the organism isolated were coagulase negative staphylococcus in child and adult female and MSSA (methicillin sensitive *staphylococcus aureus*) in adult male. Various predisposing factors and co-morbid conditions associated with botryomycosis include altered immune function, diabetes mellitus, liver disease, alcoholism, lupus, cystic fibrosis, asthma, malnutrition, immunoglobulin deficiency, hyperglobulinemia E (Job syndrome), glomerulonephritis, long term treatment with corticosteroids and human immunodeficiency virus/acquired immunodeficiency syndrome.\[[@ref5][@ref6]\] Factors like accident and post-operative skin trauma help in entry of the organism and thereby leading to localized infections.\[[@ref7]\] Except prior history of local trauma in case 1 and 3, there was no other associated co-morbid condition in our cases. Described morphological presentation of cutaneous botryomycosis include nodules, sinus, abscesses, and ulcers giving a sero-purulent exudate with occasional granules of bacteria.\[[@ref4][@ref8]\] In the present series, besides the appearance of nodule, subcutaneous swelling, sinus and ulcer, we observed few nodules with ulceration and granulation tissue giving morphology of pyogenic granuloma in case 2 and 3. Secondly, the third case was interesting since the lesions were in a linear sporotrichoid pattern which has not been described previously in literature. In botryomycosis, there occurs formation of a rim of eosinophilic matrix composed of antigen-antibody complex, tissue debris and fibrin around the bacterial granules and colonies called Splendore--Hoeppli phenomenon (SHP), which prevents phagocytosis and intracellular destruction of the bacteria leading to chronic infection.\[[@ref7][@ref9][@ref10]\] The grains in botryomycosis are large and lobulated which are cluster of bacteria that does not have a filamentous structure (like grains of mycetoma and actinomycosis). All our cases had bacterial granules which demonstrated SHP in histopathology. Botryomycosis should be clinically differentiated from other conditions like mycetoma, actinomycosis, actinomycetoma, nocardiosis and tuberculosis that presents as swellings with sinus and discharge. However, there has been no clear cut morphological differentiation of above entities. In the present case series, we observed evolution of lesions of botryomycosis which could help in reaching at clinical diagnosis of botryomycosis. All cases developed subcutaneous swelling, followed by papule and nodule over the swelling which either developed, erosion, ulceration or sinus giving seropurulent discharge. In case 2 and 3 few sinuses and nodules developed pedunculated granulation tissue giving an appearance of pyogenic granuloma. Such an evolution noted in our cases could be possibly due to localized inflammation occurring around the organisms giving rise to subcutaneous swelling, which later on extruded out by formation of nodules, papule, and sinus on the surface of subcutaneous swelling. These lesions underwent secondary changes like erosion, ulceration and sinus on the papules and nodules. The sporotrichoid distribution in case 3 could be due the lymphatic spread. The diagnosis of botryomycosis is usually confirmed by demonstration of specific bacteria in discharge, tissue culture and characteristic histopathology findings. The causative organisms detected in tissue and pus culture were coagulase negative staphylococci in case 1 and 2 and MSSA in case 3 which ruled out other differentials like mycetoma, nocardiosis and tuberculosis. The treatment with prolonged course of antibiotics depending upon pus culture and sensitivity is recommended for botryomycosis. However, appropriate antibiotics in combination with surgical excision is the most effective therapy for botryomycosis.\[[@ref3]\] All our cases were treated with prolonged treatment with single sensitive antibiotic. Two cases were treated with doxycycline monotherapy whereas the adult female was treated with ciprofloxacin similar to the report by Pradhan *et al.*\[[@ref4]\] None of the cases needed any surgical intervention. Conclusion {#sec1-4} ========== From the present series of three cases, we suggest the evolution of lesions along with demonstration of bacteria in the discharge and tissue culture and characteristic histopathology findings can help to diagnose botryomycosis. Secondly, we also highlight new morphological findings like pyogenic granuloma like lesions and sporotrichoid distribution that can occur in the botryomycosis. {#sec2-1} ### Declaration of patient consent {#sec3-1} The authors certify that they have obtained all appropriate patient consent forms. In the form the patient(s) has/have given his/her/their consent for his/her/their images and other clinical information to be reported in the journal. The patients understand that their names and initials will not be published and due efforts will be made to conceal their identity, but anonymity cannot be guaranteed. ### Financial support and sponsorship {#sec3-2} Nil. ### Conflicts of interest {#sec3-3} There are no conflicts of interest.

Introduction ============ Glaucoma is a multifactorial optic neuropathy characterized by apoptotic cell death of the retinal ganglion cells (RGCs) in the optic disc or retinal nerve fiber \[[@r1]\]. This irreversible retinal deterioration results in progressive visual field loss along with decreased contrast and color sensitivity \[[@r2]\]. Quigly et al. \[[@r3]\] have estimated that by the year 2010, approximately 60.5 million individuals will be affected by this disease, which causes bilateral blindness, and this number may rise to 79.6 million by the year 2020. Glaucoma is classified as a silent disease because patients usually do not have any signs and symptoms until the end stage, when considerable damage has been done to the eye \[[@r4], [@r5]\]. Increased intraocular pressure (IOP) is an established risk factor of the disease, along with old age, race and refractive error \[[@r6]\]. In addition to these factors, vascular \[[@r7]\], immunological \[[@r8]\] and neurotoxic \[[@r9]\] factors are also believed to cause glaucoma. Wiggs et al. \[[@r10]\] point to genetics as an additional risk factor for the disease. Different strategies used to understand the genetic risk factors have helped to define the molecular events responsible for some Mendelian forms of the disease. In addition, some of the chromosomal locations of the genes that are likely to be involved in common forms of glaucoma have also been identified. Galassi et al. \[[@r11]\] have shown that as a result of a variety of physiologic stresses, a highly conserved mechanism of gene regulation is activated. Stress or hypoxia in the tissues stimulates increased synthesis of Nitric oxide (NO), which results in an increased blood flow in the vessels of the tissue that helps to overcome the stress. Although several isoforms of nitric oxide synthase (NOS) have been reported in abundance in almost all layers of the retina, the circulating NO is synthesized solely in the vascular endothelium through the action of the endothelial nitric oxide synthase (eNOS) on the substrate L-arginine. Studies to identify the *eNOS* promoter element have revealed that many cis- and trans-acting factors together regulate the expression of the *eNOS* gene, and the level of *eNOS* expression, in turn, corresponds directly to the amount of NO in the blood \[[@r12]\]. Experiments have revealed that polymorphisms in the non-coding regions of *eNOS* may alter *eNOS* expression and thus cause a decrease in NO synthesis \[[@r13]\], which may predispose patients to hypertension, vasospasm and atherosclerosis. In the vascular endothelium, an increase in the synthesis of NOS has been shown to be neuroprotective through the promotion of vasodilation and blood flow in the ocular tissue \[[@r11]\]. In contrast to previous models, which associated neuroprotection with higher eNOS activity, it has been observed that higher NO levels that correspond to an excess *eNOS* expression have pathological relevance and could be generally toxic. This pathogenicity model states that the excess NO freely diffuses to adjacent neurons and combines with O~2~**^-^** to form peroxynitrite anions (ONOO**^-^**), an extremely potent toxin which can set cell-death programs into motion, such as neuronal apoptosis \[[@r14]\]. An abundance of NOS has been found in the optic nerve head vessels of primary glaucoma patients, supporting the idea that the optic nerve damage in glaucoma can be related to *eNOS* overexpression \[[@r15]\]. In determining the risk factors associated with increased *eNOS* expression, it has been recognized that beside the many environmental and other physiologic factors, there is also a significant genetic contribution. One of the genetic risk factors that has been identified is the 27-bp variable number of tandem repeat (VNTR) polymorphism present in intron 4 of *eNOS*, which is thought to alter NO production and contribute to vascular deregulation \[[@r16]\]. In addition, it has been shown that vascular deregulation and abnormal IOP in POAG might together be responsible for causing optic nerve damage, and both of these factors cause a reduction in ocular perfusion \[[@r11]\]. No study has been conducted to define the relationship of the vascular deregulation and PCAG. The mechanical stress resulting from elevated IOP, ischemia, excessive excitatory amino acids and excessive TNFα release increases stress in the optic nerve head \[[@r17]\]. Another mechanism elicited in response to stress is the recruitment of specific cellular proteins called "heat shock proteins" (HSPs), which have been shown to have a protective function \[[@r18]\], as the synthesis of the HSPs is upregulated in response to many forms of metabolic stress. These proteins are molecular chaperones that function in the proper refolding of the dysfunctional proteins and prevent protein aggregation, a mechanism that could be critical to the survival of RGCs; thus, a low expression of HSPs will abolish the defense mechanism against heat shock and stress in general \[[@r18]\]. Several investigators have shown that anti-HSP antibodies, through the induction of the apoptotic mechanism or the HSPs themselves via the stimulation of an immune response, may have pathogenic significance for patients with glaucomatous optic neuropathy \[[@r19]\]. The present study was conducted to determine whether *HSP70* and *eNOS* polymorphisms were associated with POAG and PCAG in the Pakistani population. We hope that our results will contribute toward a better understanding of the disease's causative agents. Methods ======= This was a multicenter case-control study in which the patients and controls were randomly selected from different hospitals in the area. The participants included 166 unaffected controls, 159 POAG patients and 111 PCAG patients. The patients were selected on the basis of their clinical history, cup--to-disc ratio (CDR) evaluation, visual field evaluation and elevated IOP, and categorized into POAG and PCAG groups based on gonioscopic findings. In addition, to rule out any ocular anomaly, the controls also underwent applanation tonometery, slit lamp examination, CDR measurement and visual field assessment. Blood samples from all of the individuals were collected by venipuncture after informed written consent was obtained, and genomic DNA was extracted from these blood samples through a conventional phenol chloroform method \[[@r20]\]. This study has been approved by the Departmental Ethics Committee and conforms to all of the norms of the Helsinki Declaration. The 27-bp insertional VNTR in intron 4 of *eNOS* was typed with the use of PCR amplification. The VNTR region was amplified using the Gene amp^®^ PCR system 2700 (ABI, Foster City, CA), with the help of the forward primer 5′-AGG CCC TAT GGT AGT GCC TTT-3′ and the reverse primer 5′TCT CTT AGT GCT GTG CTC AC-3′. Briefly, the protocol consisted of 35 cycles of DNA denaturation at 95°C for 1 min, primer annealing at 60°C for 45 s and chain extension at 72°C for 1 min, followed by a final extension cycle at 72°C for 5 min. The constituents of the reaction consisted of: 1.2 µM of each primer, 10 mM of dNTPs, 2 mM of MgCl~2,~ 1 U of Taq and 1× PCR buffer, along with 40--50 ng of DNA. The PCR products were electrophoresed on 4% agarose gels for 1 h and the bands were visualized under ultraviolet light ([Figure 1](#f1){ref-type="fig"}). Two alleles were obtained when this region was amplified: "eNOS4a," which was 393 bp long and consisted of four 27-bp repeating units, and "eNOS4b," which was 420 bp long and consisted of five 27-bp repeating units. {#f1} A 488-bp fragment encompassing the *HSP70* polymorphism ([rs1043618](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=1043618)) was PCR-amplified by using the forward and reverse primers, 5′-CGC CAT GGA GAC CAA CAC CC-3′and 5′-GCG GTT CCC TGC TCT CTG TC-3′, respectively. Each of the 35 cycles consisted of a DNA denaturation step at 95°C for 1 min followed by primer annealing at 60°C for 45 s and chain extension at 72°C for 1 min. The concentration of both primers was 0.04 µM, and the reaction mixture also contained 0.2 mM dNTPs, 1.5 mM MgCl~2~, 1 U Taq, 1× PCR buffer, and approx 90--100 ng of DNA. The PCR-amplified products (17 µl) were then digested overnight with BrsB1, as recommended by the manufacturer (Fermentas, Burlington, Ontario). After digestion, the restriction enzyme was inactivated at 65°C for 20 min, and the digestion products were electrophoretically separated on 4% agarose gel for 2 h and visualized under UV transillumination ([Figure 2](#f2){ref-type="fig"}). The wild-type G allele was recognized based on the presence of a single BrsB1 restriction enzyme recognition site that, after digestion, yielded a 461-bp and a 27-bp fragment, while the C allele that lacks the restriction site remained uncut. ). The digested products were separated by electrophoresis on 4% agarose gel. Lane L, 100-bp DNA ladder; lanes 1, 2 and 4, homozygous wild-type G allele (461 bp and 27 bp \[not shown\]); and lane 3 and 5, homozygous C risk allele (488 bp).](mv-v16-18-f2){#f2} Genotype and allele frequencies in the unaffected controls and the patients (POAG and PCAG) were compared with the Pearson χ^2^ test. All of the data were analyzed by using SPSS (ver. 16; SPSS, Chicago, Il). The statistically significant differences were further analyzed using the statistical software Minitab (ver.15; Minitab, Chicago, Il). A p value of less than 0.05 was considered statistically significant. Results ======= Patients with POAG and PCAG and the controls were genotyped for the *eNOS4*a/b and *HSP70* G+190C polymorphisms. In the controls, the frequencies of the *eNOS4*b/b, *eNOS4*b/a and *eNOS4*a/a were 83.7%, 12.7%, and 3.6%, respectively, whereas in the POAG group, these values were 70%, 24% and 6%, respectively, showing a significant association of the disease with the polymorphism (χ^2^=8.89, p*\<*0.05). In the PCAG group, 71%, 22% and 7%, respectively, were also found to be significantly associated with the disease phenotype (χ^2^=6.33, p*\<*0.05; [Table 1](#t1){ref-type="table"}). The *eNOS4*b/a genotype was found to be significantly associated with POAG (χ^2^=6.92 \[p*\<*0.01\]), as well as with PCAG (χ^2^=3.93 \[p*\<*0.05\]). As opposed to this finding, the *eNOS4*b/b genotype was found to be at a significantly higher frequency in the control samples when compared with the POAG group (χ^2^=8.87, p*\<*0.01) or the PCAG group (χ^2^=6.26, p=0.01; [Table 1](#t1){ref-type="table"}). In addition to the genotype associations, the *eNOS4*a allele frequency was found to be significantly associated with both POAG (\[χ^2^=9.29, p*\<*0.01\], OR=2.02 \[95%CI=1.25--3.28, p=0.001\]) and PCAG (\[χ^2^=7.59, p\<0.01\], OR=1.99 \[95%CI=1.18--3.37, p\<0.01\]; [Table 1](#t1){ref-type="table"}). ###### Genotype and allele frequencies of intron 4 polymorphism of *eNOS* in controls and patients. **Genotype** **Controls (n=166)** **POAG (n=159)** **p (χ^2^)** **p (χ^2^)** **Odds ratio (95% CI)** **PCAG (n=111)** **p (χ^2^)** **p (χ^2^)** **Odds ratio (95% CI)** -------------- ---------------------- ------------------ --------------- ------------------------------ ------------------------------ ------------------ ------------------------------ -------------- ------------------------------ b/b 139 (83.7%) 111 (70%) \<0.05 (8.89) \<0.01 (8.87) DM=2.23 (1.26--3.39) p\<0.01 79 (71%) \<0.05 (6.33) 0.01 (6.26) DM=2.09 (1.12--3.89) p\<0.01 b/a 21 (12.7%) 38 (24%) \<0.01 (6.92)   24 (22%) \<0.05 (3.93)   a/a 6 (3.6%) 10 (6%) \>0.05 (1.24) RM=1.79 (0.58--5.69) p\>0.05 8 (7%) \>0.05 (1.79) RM=1.16 (0.40--3.31) p\>0.05 **Allele** **Controls** **POAG** **p (χ^2^)**   **Odds ratio (95% CI)** **PCAG** **p (χ^2^)**   **Odds ratio (95% CI)** b 299 (90%) 260 (82%) \<0.01 (9.29)   2.02 (1.25--3.28) p=0.001 182 (82%) \<0.01 (7.59)   1.99 (1.18--3.37)
p\<0.01 a 33 (10%) 58 (18%)     40 (18%)   DM=dominant model, RM=recessive model The univariate logistic regression analysis revealed that the *eNOS4*a/b polymorphism was significantly associated with glaucoma (both POAG and PCAG), based on the dominant model of inheritance (for POAG, OR=2.23 \[95% CI=1.26--3.39, p\<0.01\] and for PCAG, OR=2.09 \[95% CI=1.12--3.89, p\<0.01\]), but not according to the recessive model (for POAG, OR=1.79 \[95% CI=0.58--5.69, p\>0.05\] and for PCAG, OR=1.16 \[95%CI=0.40--3.31, p\>0.05\]; [Table 1](#t1){ref-type="table"}). To study the association of *HSP70* polymorphism G+190C with glaucoma, the POAG and PCAG patients, as well as the control samples, were genotyped. In the control group, the frequency of GG, GC and CC genotypes was 58%, 27% and 15%, respectively. For POAG patients, this distribution was 48%, 32% and 20%, respectively, with no significant association found for the genotype distribution in the patients, as compared to the controls (χ^2^=2.96, p\>0.05). In PCAG patients, the frequency of the GG, GC and CC genotypes were 34%, 39% and 27%, respectively, and the genotype distribution was found to be significantly associated with the disease (χ^2^=15.29, p*\<*0.001). The genotype CC was found to be significantly associated with PCAG (χ^2^=5.26, p=0.01), but not with POAG (χ^2^=1.12, p*\>*0.05). In addition to the genotype distribution, the frequency of the G allele was found to be at a significantly lower frequency in both patient groups compared to in the control group, whereas the frequency of the C allele was high in the POAG (χ^2^=3.57, p=0.05, OR=1.38 \[95% CI=0.97--1.94, p\<0.05\]) and PCAG groups (χ^2^=18.32, p*\<*0.001, OR=2.16 \[95% CI=1.49--3.13, p*\<*0.001\]; [Table 2](#t2){ref-type="table"}). ###### Genotype and allele frequencies of G+190C polymorphism of *HSP70* in controls and patients. **Genotype** **Controls (n=166)** **POAG (n=159)** **p (χ^2^)** **p (χ^2^)** **Odds ratio (95% CI)** **PCAG (n=111)** **p (χ^2^)** **p (χ^2^)** **Odds ratio (95% CI)** -------------- ---------------------- ------------------ --------------- ------------------------------ ------------------------------ ------------------ ------------------------------ ----------------- ------------------------------- G/G 96 (58%) 77 (48%) \>0.05 (2.96) \>0.05 (2.88) DM=1.44 (0.91--2.29) p\>0.05 38 (34%) \<0.001 (15.29) \<0.001 (14.38) DM=2.63 (1.55--4.48) p\<0.001 G/C 45 (27%) 51 (32%) \>0.05 (0.96)   43 (39%) \<0.05 (4.15)   C/C 25 (15%) 31 (20%) \>0.05 (1.12) RM=1.37 (0.74--2.54) p\>0.05 30 (27%) 0.01 (5.26) RM=2.09 (1.10--3.96) p\<0.01 **Allele** **Controls** **POAG** **p (χ^2^)**   **Odds ratio (95% CI)** **PCAG** **p (χ^2^)**   **Odds ratio (95% CI)** G 237 (71%) 205 (64%) 0.05 (3.57)   1.38 (0.97--1.94) p\<0.05 119 (54%) \<0.001 (18.32)   2.16 (1.49--3.13) p\<0.001 C 95 (29%) 113 (36%)   103 (46%)   DM=dominant model, RM=recessive model Univariate analysis of the *HSP70* G+190C polymorphism revealed that the polymorphism was not associated with POAG when the data was analyzed according to a dominant model (OR=1.44 \[95% CI=0.91--2.29, p\>0.05\]) or a recessive model (OR=1.37 \[95% CI=0.74--2.54, p*\>*0.05\]). However, an analysis of PCAG revealed the data to be significantly associated based on both the dominant (OR=2.63 \[95% CI=1.55--4.48, p*\<*0.001\]) and recessive models (OR=2.09 \[95% CI=1.10--3.96, p\<0.01\]). Discussion ========== Glaucoma is a disease leading to optic neuropathy and the second leading cause of blindness in the world. The slow, progressive loss of the retinal nerve ganglion cells in glaucoma results in peripheral vision loss. Glaucoma is a multifactorial disease and its exact etiology is still unknown. It has been reported that both genetic and environmental factors are significant in the progression of the disease. A multitude of recent genetic linkage studies have led to the identification of genetic loci, while genome wide search techniques have been used to identify single nucleotide polymorphisms, which have been observed to be associated with the disease. In the present study, we provide data on two different polymorphisms of two genes, the 27-bp repeat *eNOS* VNTR polymorphism and the *HSP70* G+190C polymorphism. The *eNOS* and *HSP70* genes have been shown to elicit enhanced transcription levels of the respective mRNA when the body is under stress \[[@r18],[@r21]\]. eNOS is involved in different processes, like neurotransmission, the regulation of vascular tone, vasodilatation and apoptosis. In addition, it also regulates blood flow to the ocular tissues and has been implicated in the pathogenesis of cardiovascular diseases and different neurodegenerative disorders, like diabetic retinopathy \[[@r22]\], glaucoma \[[@r23]\] and migraines \[[@r24]\]. In focusing on glaucoma, Liu and Neufeld \[[@r21]\] demonstrated that RGC degeneration in the glaucomatous optic nerve head of POAG patients clearly corresponds to excess plasma NO-mediated neurotoxicity. However, the authors did not conduct any genetic studies to find the susceptible loci. Plasma NO levels are regulated by eNOS, thus any environmental or genetic risk factor that enhances *eNOS* expression would contribute to NO mediated toxicity. It has been established that the VNTR in the intron 4 of *eNOS* significantly influences the plasma NO levels. Functional studies have shown that the variant a allele (with four 27-bp repeats) in homozygous form is strongly correlated with increased plasma NO levels, which were twice as high as the level found in the b/b genotype (with five 27-bp repeats), indicating the *eNOS*' underlying potential for excess NO-related pathogenicity \[[@r25]\]. It was hypothesized that the molecular mechanism of *eNOS* expression was regulated at the transcriptional levels by microRNAs, which can be derived from the intron during pre-mRNA splicing \[[@r26]\]. Based on this hypothesis, Zhang et al. \[[@r27]\] conducted experiments and established the role of intron 4 VNTR repeat with *eNOS* expression by identifying a 27-nucleotide microRNA sequence derived from the 27-nucleotide repeat within intron 4 of *eNOS*. This microRNA was shown to represses *eNOS* expression by interfering with the gene transcription efficiency. In addition, they demonstrated that this microRNA acted as an endogenous sequence-specific feedback regulatory molecule that reversibly inhibited this process and thus enabled the rapid turnover of eNOS molecules to minimize the damage associated with stress and hypoxia. The presence of the endogenous expression of the microRNA from intronic sequences was further demonstrated by the unexpected finding that the 5′ and 3′ ends of the 27-nucleotide repeat unit in intron 4 of *eNOS* had splice donor and acceptor sites (AG and GA) similar to the 5′ and 3′ bases at the exonic ends in the *eNOS* genomic sequence. This further supported the hypothesis that the intact microRNAs are produced when pre-mRNA splicing of *eNOS* occurs and that the amount of microRNA is determined by the number of repeats in intron 4 \[[@r27]\]. Zang et al. \[[@r27]\] also postulated that the amount of microRNA would vary from individual to individual, depending on the number of repeats at this position, which would account for the variations observed between the individuals and their expression of *eNOS* and its association with cardiovascular risk. Consequently, compared to the b allele with five 27-bp repeats, variant a allele with 4 repeats produced fewer microRNAs and hence, reduced transcriptional repression would occur, correlating with the increased plasma NO. Our current results are in line with the above results, as we found a significant association of the *eNOS4* a/b genotype with both forms of glaucoma in the Pakistani groups, which is probably a result of high plasma NO levels that are toxic to the optic nerve and ultimately cause its degeneration, a classic hallmark of glaucoma. Earlier studies conducted to establish a possible association between glaucoma and altered NOS expression focused mainly on POAG patients \[[@r11],[@r12]\] and were able to show a positive correlation between this polymorphism and the disease, matching our results. In the current work, we also studied PCAG as an impairment in the ocular blood flow that might lead to closure of the angle, which can be partially associated with altered *eNOS* expression and the resultant NO-mediated neurotoxicity. Our results indicate that there is indeed a correlation between PCAG and the polymorphism in the *eNOS* gene, as we found a significantly higher frequency of the variant *eNOS4*a allele in PCAG patients compared to the controls. Thus, in the Pakistani PCAG and POAG groups, the disease could be associated with increased NOS activity. Our findings are consistent with the results of Sakai et al. \[[@r28]\], who found the T(−786)C polymorphism of *eNOS* to be a risk factor in patients developing non-arteritic anterior ischemic optic neuropathy (NAION) disease \[[@r28]\]. However, while studying glaucoma, Logan et al. \[[@r13]\] were unable to show a significant association between *eNOS* polymorphism T786C and VNTR repeat polymorphism and the disease. Similarly, Sena et al. \[[@r29]\] and Lin et al. \[[@r30]\] did not find any association between POAG and *eNOS* intron 4 VNTR. As opposed to this finding, however, Colombo et al. \[[@r31]\] have shown an association between Glu289Asp and T788C polymorphism of *eNOS* and cardiac myopathies, wherein patients who carried both polymorphisms had a higher risk of developing the disease. Polak et al. \[[@r32]\] studied the role of the NO system in glaucoma patients by giving POAG patients and normal healthy controls NG-monomethyl-L-arginine, which is an NOS inhibitor. They observed that glaucoma patients did not respond to the drug due to abnormal NOS activity, whereas healthy controls showed reduced choroidal and optic nerve head blood flow \[[@r32]\]. NO is responsible for maintaining arteries' vasodilation, which keeps the ocular blood flow constant. When the endothelial function is deregulated, the blood supply to the tissue is altered and impaired blood flow damages the optic nerve and leads to the development of glaucomatous changes in the optic nerve, which then results in an increase in the CDR. In our study group, an increased CDR, as compared to the controls (C/D=0.29±0.11), was observed in the POAG group (C/D=0.75±0.24, p\<0.001), as well as in the PCAG group (C/D=0.67±0.28, p\<0.001). This increased CDR in both patient groups is evidence of excessive optic nerve damage, possibly as a result of NO activity. In addition to the differences in the CDR between the controls and the two patient groups, the POAG patients had a significantly higher CDR compared to the PCAG patients (p\<0.001). It has been suggested that the HSPs are involved in the defense mechanism of stress-mediated neuron injury, caused by an increased IOP in glaucoma patients. In addition, these proteins recruit antigen presenting cells, thus they also act as immune stimulatory molecules. Therefore, in addition to their neuroprotective function, when they are overexpressed in the cell, they potentially contribute to disease progression by activating the autostimulatory response that leads to optic nerve neuropathy \[[@r17],[@r18]\]. This is supported by the observation that in glaucoma patients, increased titers of anti-HSP antibodies are present, which facilitate the disease's progression by diminishing the protective abilities of the native HSPs \[[@r33]\]. This tight regulation of the HSPs' expression pattern determines the fate of the cell, where underexpression leads to an inefficient stress response and overexpression drives the immunodestruction, with the outcome in both cases being destruction ------ driven by the oxidative free radicals in one case and by cytokines in the other, respectively. This is supported by a study conducted by Tezel et al. \[[@r33]\], who have previously shown HSPs to be significant enhancers of both the cytoprotective and neurodegenerative functions of the immune system in RGCs and glial cells, and clearly demonstrated that these proteins were critical in facilitating the glaucomatous changes caused by optic nerve damage and neurodegeneration \[[@r33]\]. Because optic nerve neuropathy in glaucoma is the major cause of blindness in Pakistan, we were therefore interested in studying the possible association between the *HSP70* polymorphism and POAG and PCAG in the Pakistani population. Our specific interest was to study the stress-mediated responses and their association with glaucoma. We thus chose the G+190C polymorphism for genotypic studies, as this polymorphism has previously been shown to be associated with reduced *HSP70* expression, which could represent an inability to cope with cellular stress and perform cytoprotective functions. Functional analyses of the G+190C polymorphism, which maps to the 5′ UTR region of *HSP70*, have revealed that compared with the G allele, the variant C allele causes reduced promoter activity and lower HSP70 protein levels \[[@r34]\]. Tosaka et al. \[[@r19]\] had previously reported a significant association between the A-110C polymorphism of *HSP70--1* and POAG (p=0.026) in the Japanese population; this polymorphism is known to increase *HSP70* expression, thus supporting the hypothesis that neuropathy is associated with autoimmunity. However, Tosaka et al. \[[@r19]\] did not find any association between glaucoma and the G+190C polymorphism, which supports the alternate model of pathogenicity related to HSP70 cytoprotective functions. As in the Japanese population, we also did not observe any association between the G+190C polymorphism and POAG, but we did find a highly significant association between the polymorphism and PCAG in the Pakistani study group. One possible manner in which *eNOS* and HSPs might contribute to PCAG is by affecting the expression of matrix metalloproteinases (MMPs). MMPs have been found to be involved in the regulation of IOP and might play a significant role in the normal and pathogenic functions of the eye \[[@r35]\]. The malfunction of the MMP9 protein has previously been shown to be associated with PCAG \[[@r36]\]. The *NOS* gene plays an important role in controlling the activity of MMPs in flow-induced remodeling \[[@r37]\] and hence, the polymorphism we studied could potentially be associated with the overactivation of MMPs, which could then possibly result in PCAG. Other lines of evidence have also suggested a role for *HSP70* in the expression of the *MMP9* gene \[[@r38]\]. It must be pointed out that to fully understand the role of HSPs in the regulation of MMPs, it is important to conduct further expression and activity assays. In conclusion, glaucoma is a heterogeneous disorder with different environmental and genetic factors driving the etiology of the disease. Various polymorphisms have been reported to be associated with this disease in different populations. Our study represents a step forward in terms of clarifying the role of the stress-related cellular components eNOS and HSP70 and their role in causing glaucoma, particularly in regard to Pakistani patients, and in terms of contributing to the current body of knowledge. There is a further need to conduct studies in different populations across the world to fully unravel the genetic basis of glaucoma, which will ultimately lead to a better understanding of the disease's underlying mechanisms. Once the disease's mechanisms are completely understood, better therapeutic interventions can be devised. The authors wish to thank all the study subjects for donating their blood samples. The authors would like to acknowledge Mr. Joseph Lal and his staff at the Taxila Eye Hospital for their help in the study.

Correction to: *Nature Communications*; 10.1038/s41467-017-00867-z; published online: 13 Oct 2017. The original version of this Article contained an error in Figure 3, where panel d was inadvertently replaced with a duplicate of panel c during typesetting. Also, the legend of Figure 5f incorrectly read '310 AD patients (blue dots, *r* = --0.4) and 157 non-demented individuals (green dots, *r* = --0.1)', and should have read '310 AD patients (blue dots, *r* = --0.1) and 157 non-demented individuals (green dots, *r* = --0.4)'. Both of these errors have now been corrected in both the PDF and HTML versions of the Article.

Introduction {#Sec1} ============ To respond with anxiety to a threatening situation or stimuli is normal and healthy; however, when such a reaction becomes excessive and impairs daily functioning, one is classified as having an anxiety disorder (APA [@CR1]). Anxiety disorders are among the most common disorders in childhood, with prevalence rates between 8.3 and 20.9 % (Costello et al. [@CR11]). In addition, comorbidity among anxiety disorders as well as with other DSM-IV disorders is substantial (e.g., Kendall et al. [@CR24]). Symptoms of anxiety disorders may include preoccupations and repetitive behaviors (e.g., obsessions and compulsions), avoidance of (social) situations, and speech problems (e.g., dysfluency), which are also commonly seen in children with autism spectrum disorders (ASD) (Hartley and Sikora [@CR17]; Wood and Gadow [@CR40]). ASD is characterized by impairments in three domains; (a) the social domain (e.g., impairments in the use of nonverbal behaviors, lack of sharing, lack of social or emotional reciprocity), (b) the communication domain (e.g., impairments in the ability to initiate or sustain a conversation, stereotyped and repetitive language), and (c) the domain of repetitive, stereotyped interests and behaviors (e.g., preoccupations, nonfunctional routines or rituals) (APA [@CR1]). In addition, prevalence rates of anxiety disorders in youth with ASD are much higher (nearly 40 % estimated by a recent meta-analysis; van Steensel, Bögels and Perrin 2011) compared to those found in typically developing children, and the differentiation between ASD and anxiety disorder symptoms, particular in the case of obsessive compulsive disorder and social anxiety disorder, can be difficult. A study of Hartley and Sikora ([@CR17]) explored which criteria effectively discriminate ASD from anxiety. They administered a semi-structured interview of DSM-IV-TR criteria for autistic disorder to the parents of children with ASD, children with anxiety disorders, and children with ADHD. In this study it was found that the domain of restricted/repetitive/stereotyped patterns could not discriminate between children with ASD and children with anxiety disorders, while the best differentiation between children with anxiety and children with ASD could be made through the communication domain. Within the social relatedness domain it was found that the children with ASD had higher endorsements of impaired non-verbal behavior and lack of seeking to share compared to the children with anxiety disorders, however, social/emotional reciprocity was a poor indicator for ASD (Hartley and Sikora [@CR17]). In addition, recent studies have found elevated scores of ASD in children with anxiety and/or mood disorders (Towbin et al. [@CR38]; Pine et al. [@CR29]). The study of Towbin et al. ([@CR38]) examined ASD traits in a sample of children with mood and anxiety disorders with three ASD measures, namely the Social Communication Questionnaire (SCQ; Berument et al. [@CR3]), the Social Responsiveness Scale (SRS; Constantino et al. [@CR10]), and the Children's Communication Checklist (CCC-2; Bishop [@CR4]). In this study, 48.0 % of the sample scored in the ASD range on at least one measure. A subsequent study of Pine et al. ([@CR29]) examined ASD scores in youths with anxiety and/or mood disorders. The same three instruments were used to assess ASD; however, in this study a comparison group of healthy participants was added. Results revealed that 4.0--24.6 % of the clinically anxious children scored above ASD thresholds depending on which measurement was used (note that of the children with mood disorders these percentages ranged between 7.4 and 75.0 %). In addition, it was found that children with anxiety and/or mood disorders had significantly higher ASD scores compared to the children in the healthy sample (Pine et al. [@CR29]). In sum, research so far has found (a) high prevalence rates of anxiety disorders in children with ASD (e.g., van Steensel et al. [@CR35]), (b) several similarities (and differences) between children with ASD and children with anxiety disorders with respect to the DSM-IV-TR-criteria of autistic disorder (Hartley and Sikora [@CR17]), and (c) elevated scores on ASD measures in children with anxiety and/or mood disorders (Towbin et al. [@CR38]; Pine et al. [@CR29]). What is not yet known is whether these ASD-like behaviors in children with anxiety disorders (and mood disorders) are precursors of the anxiety disorders or not. Do current anxiety disorders manifest, in part, as ASD-like behavior (but possibly representing methodological artifact or phenocopy rather than true ASD), or, alternatively (or additionally) do ASD symptoms in some children with anxiety disorders begin early in life, consistent with the clinical course of true ASD, and perhaps cause anxiety and mood disorders later in childhood via stress generation (see Wood and Gadow [@CR40]). Of note, the study of Hallett et al. ([@CR16]) examined the relationship between autistic-like and internalizing traits in children from the normal population using a longitudinal design. The authors found evidence for an asymmetric bi-directional relation between the two; autistic-like traits measured at age 7 contributed to internalizing traits measured at age 12 and vice versa (although the latter relation---early internalizing traits contributing to latter autistic-traits---was found to be somewhat smaller compared to the first; Hallett et al. [@CR16]). The aim of this study was to compare children with anxiety disorders to typically developing children with respect to current ASD-like behaviors as well as ASD symptoms in early development (rated retrospectively). We studied a group of children with mixed anxiety disorders. Although a 'pure' group of, for example, children with social anxiety disorders only may also be of interest, comorbidity among anxiety disorders is high and therefore 'pure' cases hardly exist in the real world of children seeking treatment for anxiety (e.g., Kendall et al. [@CR24] reported that almost 80 % of the anxiety disordered youth with a principal diagnosis of generalized anxiety disorder, separation anxiety disorder or social anxiety disorder, was also diagnosed with at least one of these three disorders). Method {#Sec2} ====== Participants {#Sec3} ------------ Forty-two children were referred to several mental health care centers in the Netherlands for treatment of clinical anxiety disorders. They participated in a study that compares treatment effectiveness for anxiety disorders in children with and without ASD. None of the children with anxiety disorders in the study had received an ASD diagnosis or were suspected of having ASD at intake. All of the clinically referred children were diagnosed with a DSM-IV anxiety disorder by the multi-disciplinary team of the mental health care centers. In addition, anxiety disorders were assessed with the Anxiety Disorder Interview Schedule---Child/Parent version (ADIS-C/P; Silverman and Albano [@CR32]), which possesses good psychometric qualities (e.g., Silverman et al. [@CR33]; Wood et al. [@CR41]). The mean number of anxiety disorders in the clinical group was 3.52 (*SD* = 2.61). For an overview of the primary anxiety disorders and comorbid diagnoses of the clinically referred children, see Table [1](#Tab1){ref-type="table"} (a combined diagnosis of the ADIS-C/P was used).Table 1Primary anxiety disorder and comorbid disorders for the clinically anxious childrenPrimary disorderComorbid disorders (*n*)*n%*SADSOCSPHGADOCDPANAGORPTSDMOODSAD716.7--13210121SOC1228.62--7601212SPH921.422--200002GAD614.3024--00022OCD49.50133--0002PAN37.112101--200AGOR12.4011100--10*SAD* separation anxiety disorder, *SOC* social anxiety disorder, *SPH* specific phobia, *GAD* generalized anxiety disorder, *OCD* obsessive compulsive disorder, *PAN* panic disorder, *AGOR* agoraphobia, *PTSD* post traumatic stress disorder, *MOOD* mood disorder (dysthymic disorder/depressive disorder) Typically developing children (*n* = 42) were recruited by graduate students via schools, daycare facilities, and convenience sampling. Children were excluded if they had a classified DSM-IV diagnosis, or if they were currently referred to mental health care centers for anxiety or other behavioral-emotional problems. Anxiety disorders (as measured with the ADIS-C/P), however, were present in five children (11.9 %): two had an ADIS-C/P diagnosis for generalized anxiety disorder, one for social anxiety disorder, one for separation anxiety disorder and one had an ADIS-C/P diagnosis for specific phobia. It was decided not to exclude those cases because: (a) anxiety disorders are very common in children and the rate found in this study is in accordance with the prevalence rates of anxiety disorders found in other studies (e.g., Costello et al. [@CR11]), which strengthens our believe that the sample is representative for the population of typically developing children, and (b) the children were not clinically referred for their anxiety problems, nor did they seek any treatment for their anxiety problems following disclosure of the research diagnoses. The group of children with anxiety disorders consisted of 21 boys and 21 girls with a mean age of 12.50 (*SD* = 2.90). The group of typically developing children consisted of 22 boys and 20 girls with a mean age of 11.38 (*SD* = 3.34). Groups did not differ with respect to gender, *X*^*2*^\[1\] = 0.05; *p* = .827, or age, *F* \[1, 82\] = 2.71; *p* = .104. Instruments {#Sec4} ----------- ### Scared-71 {#Sec5} The parent version of the Screen for Child Anxiety Related Emotional Disorders (SCARED-71; Bodden, Bögels and Muris [@CR5]) was used in the current study to measure anxiety symptoms. The SCARED-71 is a questionnaire that consists of 71 items that have to be rated on a 3-point scale (almost never---sometimes---often). Besides calculating a total anxiety score, the 71 items can be grouped in various anxiety subscales measuring symptoms of panic disorder (13 items), social anxiety disorder (9 items), separation anxiety disorder (12 items), specific phobia (15 items), generalized anxiety disorder (9 items), obsessive compulsive disorder (9 items) and post traumatic stress disorder (4 items). Good reliability and validity is reported by Bodden and colleagues (2009). ### ADI-R {#Sec6} The Autism Diagnostic Interview-Revised (ADI-R; Lord, Rutter, and Le Couteur [@CR26]) is a semi-structured interview developed as a tool to diagnose ASD. The interview is conducted with the caregiver(s) who is asked about their child's (a) current behavior and (b) behavior in the past (usually concerning the age period of 4--5 years). The behavioral items are grouped in three domains following the DSM-IV criteria for autistic disorder (APA [@CR1]); reciprocal social interaction, communication, and restricted and repetitive behaviors. A diagnostic algorithm is applicable for the items that focus on behaviors in the past (i.e., all scores have to exceed certain thresholds) to establish a diagnosis of ASD. Psychometric properties of the ADI-R were investigated in the study of Lord et al. ([@CR26]) that found high intra-class correlations for the three domains, fair to good internal consistencies and adequate reliability over time. Further, specificity and sensitivity of the interview was found adequate (Lord et al. [@CR25]). In the current study the items of the diagnostic algorithm were used (further referred to as 'early ASD symptoms'). For several cases, the ADI-R was administered to the parents after their child received treatment for anxiety problems. Of note, no significant differences with respect to the scores for early ASD symptoms were found between those with an ADI-R report administered before or after treatment. ### CSBQ {#Sec7} The Children's Social Behavioral Questionnaire (CSBQ; Luteijn et al. [@CR27]) was used in the present study to measure current ASD-like behaviors. The 49 items of the CSBQ were rated by the parents, who are asked to what extent each description applies to their child, on a three-point scale (does not apply---sometimes or somewhat applies---clearly or often applies). The items can be summed into a total score and into six subscales, namely (a) behaviors not tuned to situation, (b) withdrawal, (c) orientation problems, (d) difficulties understanding social information, (e) stereotyped behaviors, and (f) fear of and resistance to change. Psychometric properties (validity, internal consistency, inter-rater reliability and test--retest reliability) were studied in a large Dutch sample study, and the CSBQ was found a valid and reliable instrument. Furthermore, the CBSQ was found to discriminate groups of children with ASD (high-functioning autism, PDD-NOS) from clinical and non-clinical groups (Hartman, Luteijn, Serra and Minderaa [@CR18]). Procedure {#Sec8} --------- The study in which the clinically anxious children participated was approved by a medical ethics board. The comparison group of typically developing children was recruited in a later stage of the research, for which approval was given by an ethics committee of the University of Amsterdam. Written consent was obtained and assessments were carried out by the first author, several diagnosticians/psychologists and graduate students (under the supervision of the first author). Assessments took place at the families' homes or at mental health care centers. All interview administers were extensively trained in the ADI-R by the first author, who is certified for administering the interview. The training consisted of role-plays, watching and coding several videotapes (including interviews with parents of who's children were not diagnosed or suspected of ASD), and round-table discussions about coding and administration. All coders were trained to achieve at least 80 % agreement with the first author. Inter-rater reliability was assessed and was found fair to excellent (inter-correlation coefficients between the first author and the different coders ranged between .73 and .94). Data-Analysis {#Sec9} ------------- Because the vast majority of the ADI-R was administered to both parents (i.e., both mother and father were present during the interview), parental report about their child's anxiety (SCARED-71) and ASD-like behaviors (CSBQ) were averaged (of note, reports between mothers and fathers correlated significantly for each subscale; *r* = .38--.88, all *p*'s \< .05 for the different SCARED-71 subscales, and *r* = .49--.82, all *p*'s \< .05 for the different CSBQ subscales). The SCARED-71 and CSBQ scores were aggregated in part because ADI-R interviews were administered to both parents in most families, leading to an integration of mother's and father's ratings of early ASD symptoms. Also, in general, aggregated scores have been found to increase the reliability of assessments (Bögels and van Melick [@CR8]). If one parent report was missing, the other was used. In total two reports of the SCARED-71 (2.4 %) and five reports of the CBSQ were missing (5.9 %). Missing values were estimated with 2-way imputation (based on group mean and post-treatment assessment) and analyses were run twice (with and without missing values), however, results were similar. Variables were transformed into Z-scores to explore possible outliers. One case was identified as having an outlier for the CSBQ total score, one case for the ADI-R total score and the social domain, and one case was identified as having an outlier for the ADI-R repetitive domain (Transformed Z's \> 3.29; Tabachnick and Fidell [@CR36]). To minimize the possible confounding effects of these outliers, we replaced the outliers by the sample mean with the addition of three standard deviations (Field [@CR12]). In addition, we ran analyses with and without the transformed outliers, however, results were similar. Some (sub-)scales were not normally distributed. Therefore, we also applied non-parametric tests and analyses yielded similar results. Results {#Sec10} ======= Early ASD Symptoms (ADI-R) {#Sec11} -------------------------- Compared to typically developing children, children with anxiety disorders had a higher mean total score of the ADI-R as well as for all ADI-R domains (see Table [2](#Tab2){ref-type="table"}). Of the clinically anxious children, 15 children (35.7 %) met at least one out of three ADI-R thresholds: three (7.1 %) met the threshold for the social domain, seven (16.7 %) for the communication domain, and nine (21.4 %) for the repetitive domain. One child (2.4 %) met thresholds for all three domains. Of the typically developing children, none met ADI-R thresholds for any domain. Exploratory post hoc analyses using Chi-squares revealed that children with anxiety disorders differed from typical developing children with respect to their scores for various items of the ADI-R (a p value of .01 was considered significant for these analyses in an effort to control for multiple comparisons; however, the exact p value for each item is given in Table [3](#Tab3){ref-type="table"}). The most commonly endorsed ADI-R items (by 25 % or more) of the children with anxiety disorders were (in sequential order): spontaneous imitation, imaginative play with peers, imaginative play, compulsions/rituals, offering to share, social verbalization/chat, offering comfort, group play, conventional gestures, circumscribed interests, appropriateness of responses, interests in children, and response to approaches.Table 2ASD symptoms in children with anxiety disorders, and control childrenAnxietyControl*MSDMSDESFpEarly ASD symptoms (ADI-R)*Total (range 0--74)10.267.782.123.311.3638.91\<.001\*\*Social (range 0--32)4.484.461.432.610.8314.64\<.001\*\*Communication (range 0--26)4.313.610.601.191.3840.06\<.001\*\*Repetitive behavior (range 0--16)1.481.660.100.301.1628.29\<.001\*\**Current ASD-like behaviors (CSBQ)*Total score (range 0--98)19.0814.916.235.561.1423.28\<.001\*\*(1) Behaviors not tuned to situation (range 0--22)5.764.672.462.500.8813.17\<.001\*\*(2) Withdrawal (range 0--24)4.624.890.901.121.0520.97\<.001\*\*(3) Orientation problems (range 0--16)1.862.600.751.080.564.38.013\*(4) Difficulty understanding social information (range 0--14)2.953.131.601.730.545.08.016\*(5) Stereotyped behaviors (range 0--16)1.672.340.290.640.8112.31\<.001\*\*(6) Fear of and resistance to changes (range 0--6)2.231.980.250.581.3639.80\<.001\*\*\* *p* \< .05; \*\*^ ^*p* \< .01*ES* Effect size (Cohen's d)Table 3Exploratory post hoc analyses of the number of children with a 0, 1, or 2 score on ADI-R items representing early ASD symptoms: children with anxiety disorders compared to controlsAnxietyControls*X*^2^*p*^a^012012*Social domain*Use of other's body357041104.79.057Imaginative play with peers24612402016.80\<.001\*\*Direct gaze337239303.50.061Social smiling346236510.34.558Showing/directing attention400242002.05.494Offering to share296735342.36.124Seeking to share enjoyment372341102.87.202Offering comfort3010238404.94.026Quality of social overtures392142003.11.241Range of facial expressions3282420011.35.001\*Inappropriate facial expressions366041103.90.109Appropriateness of responses317440117.37.007\*Interest in children319234530.61.434Response to approaches319238313.98.046Group play3011135611.70.192Friendships381340110.72.676*Communication domain*Stereotyped utterances348042008.84.005\*Social verbalization/chat295840209.82.002\*Reciprocal conversation3273420011.35.001\*Inappropriate questions390338400.161.00Pronominal reversal366042006.46.026Neologisms347142008.84.005\*Pointing to express interest344438131.56.212Nodding346241106.10.029Head shaking346242008.84.005\*Conventional gestures307539126.57.010Spontaneous imitation22911411022.92\<.001\*\*Imaginative play25107384010.73.001\*Imaginative social play334541107.27.007\**Repetitive domain*Verbal rituals402042002.05.494Unusual preoccupations411042001.011.00Circumscribed interests30111411011.01.001\*Repetitive use of objects401142002.05.494Compulsions/rituals2688411016.59\<.001\*\*Unusual sensory interests384042004.20.116Hand/finger mannerisms329140206.22.013Other complex mannerisms384042004.20.116\* *p* \< .01; \*\* *p* \< .001^a^For analyses, ADI-R item scores of 1 and 2 were collapsed and the corresponding *p* value of Fisher's Exact Test was reported if the expected cell count of at least one cell was less than 5 A cluster analysis was run on the scores of the three ADI-R domains (social, communication and repetitive behavior) using the sample of children with anxiety disorders only (*n* = 42) to examine how many children would display an ASD-like profile. Ward's hierarchical cluster analysis using squared Euclidean distances as the distance measure was used. To establish the number of clusters, the percentage change in agglomeration coefficients and inspection of the dendrogram was used. Using this approach, three clusters were identified; all clusters differed significantly with respect to the mean ADI-R social and communication scores, and cluster one and cluster two differed with respect to the mean ADI-R repetitive scores (see Fig. [1](#Fig1){ref-type="fig"}). The first cluster (*n* = 13) is characterized by low scores in every ADI-R domain. The second cluster (*n* = 25) is characterized by medium scores in the social and communicative domain and a somewhat higher score in the repetitive domain. Finally, the third cluster (*n* = 4) is characterized by a PDD-NOS-like profile: high scores on both the social and communication domain, and medium scores for the repetitive domain.Fig. 1Profiles of the three clusters of children with anxiety disorders for their mean ADI-R algorithm scores in the early childhood social, communication, and repetitive domains None of the children in the first cluster met any ADI-R threshold. Of the second cluster, none of the children met the ADI-R threshold for the social domain, however three children (12.0 %) met the threshold for the communication domain and eight children (32.0 %) met the threshold for the repetitive domain. Most commonly endorsed problems (ADI-R items) for the children in this cluster were: spontaneous imitation (score 0 = 40 %; score 1 = 28 %; score 2 = 32 %), imaginative play with peers (score 0 = 52 %; score 1 = 16 %; score 2 = 32 %), compulsions/rituals (score 0 = 52 %; score 1 = 24 %; score 2 = 24 %), offering to share (score 0 = 56 %; score 1 = 16 %; score 2 = 28 %), and imaginative play (score 0 = 60 %; score 1 = 20 %; score 2 = 20 %). With respect to the children in the third cluster it was found that three children (75.0 %) met the threshold for the social domain, all four children (100.0 %) met the cutoff for the communication domain and one child (25.0 %) met the cutoff for the repetitive domain. Most commonly endorsed ADI-R items for this cluster concerned problems with: imaginative play with peers (score 1 = 25 %; score 2 = 75 %), social verbalization/chat (score 1 = 25 %; score 2 = 75 %), conventional gestures (score 1 = 25 %; score 2 = 75 %), reciprocal conversation (score 1 = 50 %; score 2 = 50 %), and group play (score 1 = 75 %; score 2 = 25 %). Current ASD-Like Behaviors (CSBQ) {#Sec12} --------------------------------- Compared to typically developing children, those with anxiety disorders were found to have significantly higher mean scores for ASD-like behaviors in general (CSBQ total score) and for all subscale scores (see Table [2](#Tab2){ref-type="table"}). Thirteen children with anxiety disorders (31.0 %) had scores that fell in the ASD range. None (0.0 %) of the typical developing children had scores falling in the ASD range. Correlations {#Sec13} ------------ For the clinically referred children only (*n* = 42), correlations were calculated between early ASD symptoms (ADI-R), current ASD-like behaviors (CSBQ), and anxiety symptoms (SCARED-71). The correlation between early ASD symptoms and current ASD-like behaviors was .36 (*p* = .020), indicating that higher scores of early ASD symptoms were related to higher scores of current ASD-like behaviors. For current ASD-like behavior (CSBQ), correlations with the following SCARED-71 subscales were significant (in sequential order from lowest to highest); specific phobia, separation anxiety disorder, generalized anxiety disorder, panic disorder, and social anxiety disorder (varying from .38 to .59; see Table [4](#Tab4){ref-type="table"}). All correlations were positive, indicating that higher current ASD-like behaviors were related to higher anxiety symptoms. For early ASD symptoms only one correlation, with symptoms of social anxiety disorders, was found significant (Table [4](#Tab4){ref-type="table"}). Although this correlation did not statistically differ from the second highest correlation, the magnitude of this correlation was medium (*r* = .38), while the other correlations were found to be small (*r* ≤ .24).Table 4Correlations between current anxiety symptoms (SCARED-71) and early ASD symptoms (ADI-R), and current ASD-like behavior (CSBQ) in clinically referred children with anxiety disorders (n = 42)SCARED-71PANGADSOCSADOCDPTSDSPHEarly ASD symptoms (ADI-R).06.02.38\*.24.14.10.24Current ASD-like behavior (CSBQ).58\*\*.50\*\*.59\*\*.49\*\*.29.21.38\**PAN* panic disorder, *GAD* generalized anxiety disorder, *SOC* social anxiety disorder, *SAD* separation anxiety disorder, *OCD* obsessive compulsive disorder, *PTSD* post traumatic stress disorder, *SPH* specific phobia\* *p* \< .05; \*\* *p* \< .01 Discussion {#Sec14} ========== This study explored autistic symptoms and ASD-like behaviors in children with anxiety disorders. The main results were: (a) Parents reported that their school-aged children with anxiety disorders had significantly more ASD symptoms in early childhood than typically developing children; (b) over one-third of children with an anxiety disorder, but no known history of ASD, exceeded at least one of the three ADI-R thresholds for early childhood clinically significant ASD symptoms; and (c) early ASD symptoms were found to be related to current ASD-like behaviors as well as current symptoms of social anxiety disorder. It was found that ASD symptoms seem to be present early in life (based on retrospective reports of parents) in some children who have developed clinical anxiety but who are not recognized by professionals as having current ASD. Cluster analysis revealed that a small number of children (*n* = 4) showed a classic, if moderate, ASD-like profile (high scores on both the social and communication domains, as well as moderate repetitive behaviors), while the majority of the children with anxiety disorders (*n* = 25) were characterized by a milder ASD-like early childhood phenotype. Commonly endorsed items for the latter group included items such as early childhood impairments in spontaneous imitation, imaginative play (with peers), and offering to share, but at a low-severity level. Interestingly, the most commonly endorsed symptoms, and significant differences between clinically anxious and controls, were found for items in the communication domain. Similarly, the study of Hallett et al. ([@CR16]) found that early communication difficulties in children from the general population contributed more strongly to internalizing traits in later life than social difficulties and repetitive behaviors. With the exception of symptoms of obsessive compulsive disorder and post traumatic stress disorder, all other anxiety symptoms were found to be associated with current ASD-like behaviors. In contrast, only one (symptoms of social anxiety disorder) was found to be associated with early ASD symptoms (as measured with the ADI-R). It should be noted that the instruments to assess anxiety symptoms (SCARED-71) and current ASD-like behaviors (CSBQ) are more similar (likely resulting in higher correlations between these two instruments) compared to the instrument used to assess early ASD symptoms (ADI-R). However, it may also be that children with a moderate or high degree of ASD symptoms early in life, may be more prone to develop (symptoms of) some (social anxiety) rather than other anxiety disorders. In addition, although different instruments were used to assess early ASD symptoms and current ASD-like behaviors, evidence for moderate ASD-symptom stability was found in the current study (i.e., the correlation between early ASD symptoms and current ASD-like behaviors was found to be medium and significant, *r* = .36), however, it was not as large as in other studies that used the same instrument to asses ASD traits over time (e.g., the correlation between ASD traits over a 5-year period of time was found to be .59 for boys and .55 for girls in the study of Hallett et al. [@CR16]). With respect to current ASD-like behaviors, this study found almost one-third (31.0 %) of the clinically anxious sample falling in the ASD range on the current ASD-like behavior instrument (CSBQ). This finding is in accordance with previous studies (Towbin et al. [@CR38]; Pine et al. [@CR29]), although Pine and colleagues reported a somewhat lower percentage for the children with anxiety disorders (i.e., 4.0--24.6 %). The relatively high percentage may partly be explained by symptom overlap in the instrument used to assess current ASD-like behaviors (e.g., items of the CSBQ such as 'makes little eye-contact' or 'does not initiate to play with others' may very well be present in socially anxious children, while those behaviors may not necessarily be attributable to ASD). Nevertheless, it may be important in clinical practice to explore current ASD-like behaviors in children with anxiety disorders as the study of Puleo and Kendall ([@CR30]) found that the presence of ASD traits in anxious children not identified to have ASD may have consequences for treatment choice. That is, children with moderate ASD traits were found to profit more from family oriented cognitive behavioral therapy (CBT) compared to individual CBT (Puleo and Kendall [@CR30]). Note also that Sofronoff et al. ([@CR34]) found that for the treatment of anxiety in children with Asperger, active parental involvement enhanced the effects of this intervention. Contradictory, in typically developing children with anxiety disorders an additional effect of a family component is not always found (e.g., In-Albon and Schneider [@CR21]; Bodden et al. [@CR6]); however, those studies did not explore the possible role of ASD traits. Finally, limitations of the study have to be addressed. The first limitation is that ASD symptoms in early development were assessed retrospectively, and although the ADI-R has good psychometric properties (e.g., Lord et al. [@CR26]), recall bias cannot be excluded. In addition, it cannot be ruled out that recall-bias is different in the two groups. That is, it might be that parents of a child with an anxiety disorder report more negative aspects of their child's development (e.g., due to the stress associated with having a child with an anxiety disorder) compared to parents of control children. However, it was also found that the early ASD scores from parents who were interviewed after their child received treatment (perhaps being more in a positive state of mind because of successful treatment outcome) did not differ from the scores of parents interviewed before their child received treatment. Furthermore, the distinct clusters of early ASD symptoms revealed in the cluster analysis suggest that only a minority of children were reported to have a pattern of significant ASD symptomatology in early childhood, suggesting at the least that such a recall bias could not have affected the parents of children with anxiety disorders uniformly. Nonetheless, as recall bias cannot be ruled out when using a retrospective measure, a potential supplemental research strategy for future studies would be to ask the parents to provide home videotapes about their child's early development, and then rate the ASD-related behaviors (note that ASD-research already made use of such an approach; e.g., Osterling et al. [@CR28]; Werner and Dawson [@CR39]). A second limitation of the study is that we could not establish temporal precedence for we did not assess anxiety early in life. It may well be that anxiety disorders were already present early in life and may have not been noticed or treated until later in life. For example, social anxiety disorder may have its onset early in development and behavioral inhibition is thought to be a precursor (e.g., Bögels et al. [@CR7]; Rapee and Spence [@CR31]). Behavioral inhibition is viewed as a temperament style that consists of a pattern of behaviors such as avoidance, withdrawal, shyness, and reticence that manifest in response to novelty or unfamiliarity (Hirshfeld-Becker et al. [@CR19]). Although there may be some overlap between items of the ADI-R and behavioral inhibition under select novel conditions (e.g., items of the ADI-R asking about 'seeking to share one's enjoyment with others' or 'initiating social talk'), the ADI-R primarily includes items that share (very) little overlap with behavioral inhibition (e.g., 'imaginative play' or 'facial expressions'). In addition, post hoc analyses revealed significant differences between the clinically anxious children and the typically developing children for such non-overlapping items, suggesting that the early life ASD symptoms found in clinically anxious children are not solely attributed to the diagnostic overlap with other constructs like behavioral inhibition. However, considering the cross-sectional nature of this study, results about the relation between elevated ASD symptoms in early development and anxiety disorders later in life should be interpreted with caution and viewed with the perspective that these are preliminary findings that should be further explored. Third, relatively few children (*n* \< 10) were diagnosed with obsessive compulsive disorder, panic disorder, agoraphobia and post traumatic stress disorder. It is possible that this confounded the results. For example, certain autistic-like behaviors are common in children with obsessive compulsive disorder (e.g., Ivarsson and Melin [@CR22]), and autistic-like behaviors in adults were found to be more common in those with obsessive compulsive disorder than in those with social anxiety disorder (Bejerot and Mörtberg [@CR2]). It could be that differences between the clinical group and controls would have been even larger if more cases of obsessive compulsive disorder were included. Larger sample sizes of 'pure' cases (e.g., children with just social anxiety disorder or obsessive compulsive disorder) would in some ways have been preferable to explore the specific associations between anxiety disorders and ASD symptoms early in life. However, as noted in the introduction, clinically anxious children are often diagnosed with multiple anxiety disorders (Kendall et al. [@CR24]), making such comparisons complicated and unrepresentative of treatment-seeking children with clinical anxiety. A final limitation is that we did not have IQ data available for the two groups and that the clinically anxious children consisted of a very heterogeneous sample. That is, our sample of children with anxiety disorders consisted of a variety of anxiety disorders and most of the children had multiple anxiety disorders. On the other hand, all were highly verbal and able to respond appropriately to our anxiety disorders diagnostic interview, denoting roughly age-appropriate levels of verbal communication. Despite the addressed limitations, the relationship between anxiety and ASD is interesting and warrants further investigation. Not only are anxiety disorders highly prevalent and perhaps somewhat (phenomenologically) endemic in individuals with ASD, but also---as found in this and other studies---early ASD symptoms and current ASD-like behaviors are significantly more prominent in children with anxiety disorders. Interesting, the two disorders may share some deficits in the same brain region. For example, for both children with ASD as well as clinically anxious children, anxiety symptoms were found to be associated with abnormalities in the functioning of the amygdala (Thomas et al. [@CR37]; Juranek et al. [@CR23]). In addition, similar genetic markers are found for anxiety in children with autism and typically developing children (e.g., Gadow et al. [@CR13]). However, results concerning the amygdala are inconclusive, other brain areas (as well as other biomarkers) are found to be abnormal in ASD (e.g., Brambilla et al. [@CR9]; Hughes [@CR20]), and apparent similarity may disguise underlying differences. Gregory and Eley ([@CR14]) concluded that environmental factors are at least equally important as genetic factors for the differentiation of individual anxiety levels in children. Furthermore, Hallett et al. ([@CR15]) found little evidence for genetic influences with respect to the phenotypic correlation between autistic traits and internalizing traits in the general population, and propose that anxiety may be a response to autistic-like difficulties. Likely, the stress experienced by many children with moderate to high ASD symptoms promotes anxiety and mood disorders (Wood and Gadow [@CR40]), however, more empirical research is needed. Although part of the results (found here and elsewhere) may be conflicted by diagnostic overlap between ASD and anxiety, the present findings do warrant further investigation of the relationship between the two disorders, and more specifically of the role of ASD symptoms in the development, maintenance, and treatment of anxiety disorders. Conflict of interest {#d29e2595} ==================== The authors declare that there is no conflict of interest. Open Access {#d29e2600} =========== This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.

Introduction {#Sec1} ============ The technology of optical manipulation and trapping of microscopic particles^[@CR1]^ has been developed since the beginning of 1970s and has resulted in designing of numerous types of optical traps^[@CR2]^. Most of them operate within a setup of an inverted microscope, where the optical trap is created by an objective of a large numerical aperture (NA), which strongly focuses the light beam. Additionally, setups of this kind often utilize spatial light modulators^[@CR3]^, or deflected mirrors^[@CR4]^ in order to manipulate the trap or to create a more traps. In the recent years, a strong trend could be observed for miniaturization of optical traps, for example in order to use them in lab-on-chip setups^[@CR5]^. Miniaturized optical traps do not feature large microscope objectives, but optical fibers. Apart from miniaturization, this trend results in a simplification of the optical setup, avoiding beam propagation in free space and avoiding the use of immersion oil. Further, it allows for manipulating particles anywhere in the solution. Optical fibers are also biocompatible, mechanically resistant and cheap. Several solutions are used in order to achieve a focused or a suitably formed beam, which involve modifying fiber ends: spherical tapering^[@CR6]^ or sharp tapering of the fiber end^[@CR7]^, inserting a lens or a sphere to the fiber end^[@CR8]^, inserting an axicon lens^[@CR9]^, inserting a spiral phase plate^[@CR10]^, adding a Fresnel zone or phase plates^[@CR11]^, annular exposure of the end^[@CR12]^, and using a fiber with two cores^[@CR13]^. The main drawback of such setups is either a very short manipulation distance, typically allowing to trap only the particles directly at the fiber tip, or obtaining a weakly focused two-dimensional optical trap due to the reduced numerical aperture (NA). To solve these problems graded index (GRIN) lenses can be used^[@CR14],[@CR15]^. In this paper, we present the modified stack-and-draw method that allows to fabricate a nanostructured GRIN lens with record large refractive index gradients. We present both a theoretical and an experimental study of a novel compact lensed fiber system utilizing the fabricated nanostructured GRIN lens. As a proof-of-concept of the new GRIN lensed fiber system, we demonstrate an experiment of an optical trapping of micrometer-sized glass beads. We also show that the proposed method for fabricating GRIN elements is compatible with optical fiber technology. Moreover, it allows for fabricating elements with any shape of the refractive index distribution in 2D. Thanks to that a new functionality could be achieved by replacing the GRIN lens with an axicon lens, vortex type elements, micro-lenses arrays, diffraction elements and others. Principles of the GRIN lens {#Sec2} --------------------------- The standard GRIN lens has a parabolic profile of the refractive index in a cross-section. The radial dependence of the refractive index is given by the equation:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n={n}_{g}(1-\frac{g\,{r}^{2}}{2})$$\end{document}$$where *g* is a gradient constant (units: mm^−2^), *r* is a distance from the optical axis (units: mm), and *n*~*g*~ is the index of refraction in the center of the element^[@CR16]^. In this type of material, the focusing takes place in the bulk of the element and not on its surface. Because of this property the pitch remains constant, regardless of the environment of the element. The pitch Λ of a GRIN lens is the length of the distance between the consecutive focal planes, described by the following equation:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{\Lambda }}=\frac{2\pi }{\sqrt{g}}$$\end{document}$$ Depending on the length of the GRIN lens, we can design any imaging system which can be realized using the same index profile. The half-pitch element works as a lens, while the quarter-pitch element acts as the collimator for a point source located on the front surface of the lens. GRIN lenses are currently produced from various types of materials: axial GRIN, radial GRIN and spherical GRIN materials^[@CR17],[@CR18]^. The GRIN lens can also be fabricated with the use of different methods. The most common method for the fabrication of GRIN glass is ion exchange^[@CR19]^, where it is possible to obtain *Δn* = 8 × 10^−2^, and ion stuffing^[@CR20]^ where *Δn* = −6 × 10^−2^ at the radial distance of 250 µm. These methods have some limitations, with the most important disadvantages being very small contrasts of refractive index, only monotonic refractive index distribution^[@CR18]^, and the fact that these methods often require the use of toxic ingredients. Also, the neutron irradiation technique can be used, where it is possible to obtain *Δn* = 2.28 × 10^−3^. However, it is limited due to the use of a neutron source and allows for fabricating structures with a monotonic change of refraction index^[@CR21]^. The most significant change in the refraction index is achieved by modified chemical vapor deposition (MCVD), where typically it is possible to obtain *Δn* = 0.1 at the radial distance of 250 µm^[@CR22]^. It is also possible to fabricate 1D multilayer GRIN with a refractive index gradient of *Δn* = 0.25 per 6.5 µm^[@CR23]^. The disadvantage of this method is that it is time-consuming due to the multi-step process, but it allows to obtain any non-monotonic distribution of the refraction index. Presently, various solutions, i.e. taper structures^[@CR24]^, forming a focusing lens directly on the tip of an optical fiber^[@CR25]^, or lens axicon^[@CR26]^, are being tested in order to reduce the size of the optical elements so that they can be more easily integrated with other components, such as fibers. Nano-structuration and miniaturization of GRIN elements allows to integrate them with optical fibers and telecommunication systems. At the same time, it makes integration more difficult, because the size of such a lens is in the order of tens of micrometers^[@CR27]^. Usually, the method is to glue or splice a glass element called spacer to the fiber, together with a piece of a gradient fiber, which plays the role of a lens^[@CR28]^. This fiber system is usually called a "fiber probe". The spacer separates the fiber and the lens and is used to ensure that the spot size of propagating light is the same as the diameter of the lens. Mao *et al*.^[@CR29]^ proposed using the fiber system for biomedical imaging. They performed experiments for two lengths of the fiber spacer and different types of GRIN fiber lenses of different core sizes and outer diameters. They measured the working distance (WD) between 80 ÷ 630 µm and the spot size, i.e. the beam diameter (BD) 13 µm for the GRIN lens of 90 µm in length, and 34 µm for the GRIN lens of 420 µm in length. Recently, Bi *et al*.^[@CR28]^ examined theoretically and experimentally the GRIN fiber lenses of the length of 90 ÷ 120 µm. The measured working distances varied in the range of 500 ÷ 630 µm, and the spot diameters changed in the range of 23 ÷ 34 µm. In both those reports, a relatively large BS is notable, ranging from several to a dozen of micrometers. This is due to the technological constraints associated with the fabrication of GRIN components and, consequently, the relatively small value of the constant gradient *g* in Eq. [1](#Equ1){ref-type=""}. Therefore, we propose a method which allows to significantly increase the constant g, from 14.14 mm^−2^ or 30.25 mm^−2^ (as in the abovementioned publications) to more than 80 mm^−2^. The modified stack-and-draw method potentially allows to fabricate the GRIN lens with constant gradient *g* over 17500 mm^−2^ for a 3 µm diameter GRIN element with the maximal refraction index in the center equal 1.5581, and the minimal refraction index equal 1.5273^[@CR27],[@CR30]--[@CR32]^. This would reduce the size of the BS to less than 1 µm. What is more, the proposed method is compatible with optical fiber technology and permits to fabricate a compact lensed fiber system utilizing the nanostructured GRIN lens. As a proof-of-concept of the technological novelty we demonstrate a fiber-nanostructured lens system used for two-dimensional optical trapping of dielectric elements. Results and Discussion {#Sec3} ====================== Numerical analyzes {#Sec4} ------------------ The theoretical part of the study consisted in running several numerical analyses. The main objective of the simulations was to compare the properties of (1) the nGRIN lens with the ideal GRIN lens, and (2) the lensed fiber element containing the nGRIN with a lensed fiber containing an ideal GRIN lens. In Fig. [1](#Fig1){ref-type="fig"}, we present simulations of light propagation through lenses of infinite length, for both the ideal GRIN lens and nGRIN lens. For the wavelength λ = 976 nm (used in all simulations and experiments), the results of those simulations show the half-pitch (Λ/2) length of the lens to be 351 ± 1 µm for the ideal lens, and 355 ± 1 µm for the nGRIN lens. This indicates a good correlation with the theoretical predictions, but for the nGRIN lens the half-pitch is longer, which suggests that the discretization of the gradient resulted in a lens with a slightly different gradient constant, or a slightly different refractive index distribution. This topic is discussed in more detail in the section on the nGRIN lens design.Figure 1Normalized intensity of the beam propagation along the optical axis of an infinite GRIN lens for wavelength λ = 976 nm: (**a**) Ideal GRIN lens, (**b**) nGRIN lens. Planes A and B correspond to propagation distances in the lens, where the beam reaches the minimum and maximum width, respectively. Λ/2 denotes half-pitch length. In the consecutive simulations, we investigated the quality of the beam propagating inside the lens. The two-dimensional light intensity distributions in planes A, B marked on Fig. [1](#Fig1){ref-type="fig"} are shown in Fig. [2](#Fig2){ref-type="fig"}. These results reveal a small discrepancy between the full width at half maximum (FWHM) values for the ideal lens and for the nGRIN lens. There are also differences in the distribution of light intensity, but their nature and size confirm that the nGRIN element can work as a lens.Figure 2Comparison of light intensity distribution inside ideal GRIN and nGRIN lenses illuminated with a beam of FWHM equal 9.2 μm (Planes A and B denoted on Fig. [1](#Fig1){ref-type="fig"}). The next set of numerical analyses was performed for a lensed fiber element schematically shown in Fig. [3](#Fig3){ref-type="fig"}. The aim was to determine the optimum thickness of the spacer and the nGRIN lens for its use in an optical trapping setup, as discussed in the following section.Figure 3The scheme of the single mode nGRIN fiber lens system. In the simulations, we assumed that the FWHM of the output beam from the single mode fiber equals 5.5 μm and the spacer was made of glass with the same refractive index as the glass in the center of the nGRIN lens (*n*~0~ = *n*~*g*~ = 1.57675 for λ = 976 nm). It was also assumed that a 976 nm wavelength would be used for optical trapping in water. The length of the spacer was chosen such as to ensure that all the nGRIN lenses with a diameter of 22.8 μm (defined as the diameter of the circle in which the nanostructure can be inserted) are used to transform the light coming out of the single mode fiber (SMF). The optimum spacer thickness was set at *L*~*s*~ = 105 μm. For shorter spacer lengths, the beam is not widened sufficiently, whereby only the central part of the lens is lighting. However, due to the limitations of technology, in its central part, the refractive index distribution is characterized by a constant value. For a longer spacer length, the incident beam will be too wide and the whole beam will not be focused by the lens. The results achieved (Fig. [4](#Fig4){ref-type="fig"}) show that the minimal beam spot (BS) diameter is equal 5.8 μm and is obtained with 175 μm length of the nGRIN lens (the same BS diameter can also be obtained for the nGRIN lens of 175 + Λ/2 = 526 μm in length). Unfortunately, in this case the focus is on the end facet of the nGRIN lens, which means that the working distance (WD) equals zero. From the geometrical analysis of the optical trapping system based on the inverse microscope (described in the section: Optical trapping with the use of the nGRIN lens), and assuming the lensed fiber system is inclined at an angle of 45°, the resulting WD must be greater than 62.5 μm. Therefore, the nGRIN lens thickness was arbitrarily set to *L*~*G*~ = 102 μm. In this case BS is 7.3 μm and WD~water~ is 73.3 μm. Figure [4](#Fig4){ref-type="fig"} also shows the results for an nGRIN lensed fiber system operating in the air. The results obtained show that the BS is the same for both systems working air and water. However, the working distance (WD) changes and WD~air~ is 55.0 μm. Magnification of our nGRIN lensed fiber system, defined as the ratio of the image size to the object size, is equal M = 1.33 for both environment: water and air.Figure 4Dependence of the: (**a**) working distance (WD), and (**b**) beam spot diameter (BS) on thickness of the nGRIN lens for wavelength λ = 976 nm and immersed lensed fiber system in water and air. The red points denotes nGRIN lenses with a length of *L*~*G*~ = 102 μm. Experimental verification of the nGRIN lensed fiber system {#Sec5} ---------------------------------------------------------- In the experimental part, we first fabricated the nGRIN lens with the use of the stack-and-draw method and then built the compact nGRIN lensed fiber system (see Materials and Methods). Next, the system was characterized to verify its working distance. Measurements were made in the air, and a 976 nm laser coupled into our lensed fiber system was used as the light source. We measured the beam size as a function of the distance from the nGRIN facet and estimated the working distance of the system, as 54 ± 2 µm (Fig. [5b](#Fig5){ref-type="fig"}), which is concurrent with the simulation results (Figs [4](#Fig4){ref-type="fig"}, [5a](#Fig5){ref-type="fig"}). At that distance, the FWHM of the focal spot is equal to 8.15 µm (Fig. [5c](#Fig5){ref-type="fig"}). This is slightly larger than the minimal spot size of 7.3 µm for a beam focused by such a lens, as calculated and presented in Fig. [4b](#Fig4){ref-type="fig"}.Figure 5Beam propagation in air after the end plane of the *L*~*G*~ = 102 µm long nGRIN lens, with spacer *L*~*s*~ = 105 µm long. Normalized light intensity profile for: (**a**) simulation, and (**b**) experimental setup, (**c**) FWHM of the beam. Proof of concept: Optical trapping with the use of the nGRIN lensed fiber system {#Sec6} -------------------------------------------------------------------------------- In order to verify the applicability of the fabricated nGRIN lensed fiber system, we used it for 2D trapping, i.e. for pushing and pulling a single dielectric bead on the glass surface (Fig. [6](#Fig6){ref-type="fig"}). In the experiment, we utilized silica beads of 2 µm diameter immersed in water. Because the spot size was larger than 7 µm, which is larger than the diameter of a single silica bead being trapped, the proposed nGRIN lensed fiber system also trapped clusters of nearby beads. Figure [7](#Fig7){ref-type="fig"} shows the examples of trapping of two and three glass beads simultaneously.Figure 6Demonstration of the 2D optical trapping of a single glass bead (results for 120 mW output light power): (**a**) Scheme of the trapping setup, (**b**) observed area (a sequence of movement of the trapped beads is presented in the lower left corner, letters indicate sequence of figures) (**c**--**g**), the trapped elements are surrounded by a yellow dashed line circle, two reference beads are surrounded by a white dashed line rectangle, and the yellow frame marks an enlarged field of observation), (**c**--**g**) subsequent positions of the silica beads (the arrows in the lower left-hand corner indicate the direction of movement of the trapped beads towards position indicated at the next figure). (Supplementary materials: [Movie 1.avi](#MOESM2){ref-type="media"}).Figure 7Simultaneous trapping (results for 120 mW output light power) of: (**a**,**b**) two glass beads, and (**c**,**d**) three beads (directions of movement of the trapped beads is indicated by an arrow at the lower left corner, the trapped elements are indicated with a yellow dashed line circle, two reference beads are indicated with a white dashed line rectangle). (Supplementary materials: [Movie 2.avi](#MOESM3){ref-type="media"}). Discussion {#Sec7} ========== This paper presented the theoretical and experimental verification of a novel compact nGRIN lensed fiber system based on an nGRIN lens fabricated by the modified stack-and-draw technique. In the theoretical part, we showed how computational simulations supported the design of the nGRIN lens and the production process of the lensed fiber system. A high compatibility of the numerical analysis with the consecutive experimental results proves that it is possible to produce an nGRIN element according to the design and successfully use it in a fiber system. We produced an nGRIN lens with numerical aperture NA = 0.16 and a parabolic distribution of the refraction index from two thermally matched glasses with gradient constant *g* greater than 80 mm^−2^ (22.8 μm nGRIN in diameter, *n*~*g*~ = 1.56895, *n* = 1.56078). This allowed for focusing light in the air at wavelength λ = 976 nm to a spot diameter of 8.15 μm, at a distance of 54 μm from the end facet of the nGRIN lens. In water, for the same wavelength, the beam was focused to a spot diameter of approximately 7.5 μm at a distance of 73 μm. The beam spot diameter obtained is sufficient to perform the trapping in 2D, which has been confirmed experimentally. The lensed fiber system shows two main advantages of the nGRIN lens over conventional GRIN lenses. These are the small working distance and small beam spot diameter values, both stemming from the high value of gradient constant *g* of the nGRIN lens. Moreover, the characteristic of the nGRIN lens, which cannot be achieved with any other conventional method, was obtained in a relatively simple manner thanks to using the stack-and-draw technology. This indicates that lenses fabricated in such technology can potentially allow for optical trapping in 3D in a system with single GRIN lens. The stack-and-draw method allows to fabricate elements with any shape of the refractive index distribution in the direction perpendicular to the length of the fiber, and in particular, diffraction elements and GRIN-type elements such as: elliptical lenses, axicons, micro-lenses arrays and optical vortex. Additionally, the elements obtained with this method are compatible with fibers optics, i.e. their size is comparable to standard fibers, so they can be combined into more complex optical systems such as the nGRIN lensed fiber system presented in this paper. The stack-and-draw method also allows for to combine several functionalities within a single fiber (e.g. a micro-lens, a drug delivery channel, electrodes), which can be useful in optofluidic applications. Moreover, by combining materials with appropriately selected dispersion properties the method allows to create nGRIN lenses whose performance is nearly independent on the wavelength. The stack-and-draw method also allows for a very high repeatability of the fabricated elements. First, a microstructured fiber is obtained characterized by identical parameters along its length, which is similar to the fabrication of standard optical fibers. Then, thousands of identical elements are obtained from one long fiber through cutting, which is cost-effective. Future research {#Sec8} --------------- As shown above, the beam spot equal to 7.3 µm enables optical trapping in 2D. The presented nGRIN lens, however, is insufficient to obtain the BS of less than 0.7 µm, which would be necessary for 3D stable traps^[@CR33]^. However, since the nGRIN fabrication method presented in this work potentially allows to fabricate such a lens, thus, we assume that 3D optical trapping can be obtained in a system with a single GRIN lens. For this purpose, a lens with large gradient constant *g* is required. The way to do this is to increase the difference between the refractive index of the glass in the center and outside of the GRIN lens, and reduce the diameter of the lens. Buczyński *et al*.^[@CR32]^ have shown that borosilicate soft glasses with the refraction index equal 1.5581 and 1.5273 can be used for this purpose. This work also shows the possibility of reducing the diameter of the nGRIN element to 3 μm. This, potentially, allows to fabricate GRIN lens with constant gradient *g* over 17500 mm^−2^ and to reduce the size of the BS to less than 0.7 µm. The analysis of how the working distance (WD) and the beam spot (BS) diameter depends on the diameter and length of the nGRIN lens is shown in Fig. [8](#Fig8){ref-type="fig"}.Figure 8Relationship of the: (**a**) working distance (WD), and b) beam spot diameter (BS) on diameter and thickness of the nGRIN lens for wavelength λ = 976 nm and immersed lensed fiber system in water. For the red areas, BS is less than 0.7 μm. Since the stack-and-draw method used for the fabrication of the nGRIN lens opens up the possibility of producing elements with any refractive index distribution, in further work, we plan to extend the scope of the use of fiber optic systems with new functionalities by replacing the nGRIN lens by: diffraction elements, axicons, micro-lenses arrays and optical vortex. Materials and Methods {#Sec9} ===================== Design of the nGRIN lens {#Sec10} ------------------------ A nanostructured gradient index component consists of two different types of glasses which are assembled^[@CR34]^ in a certain pattern, which results in the continuous refractive index distribution. The calculations for the distribution of the two glass types are based on the Effective Medium Theory (EMT)^[@CR35]^, where two properties such as the conductivity δ and dielectric constant ε of the medium are taken under consideration. In this case, we need to describe the permittivity of the medium or the refractive index of the mixed materials. For this purpose, the Maxwell-Garnett (M-G) mixing formula for homogenized effective medium^[@CR28]^ can be used. It defines the effective permittivity of the medium by averaging over a neighborhood around a certain point *r*, according to the formula^[@CR27],[@CR36]^:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varepsilon }_{eff}=\langle \varepsilon \rangle -f({U}_{r})\langle {\varepsilon }_{1}-{\varepsilon }_{2}\rangle \frac{{\varepsilon }_{1}-\langle \varepsilon \rangle }{3\langle \varepsilon \rangle }$$\end{document}$$where *ε*~1~ and *ε*~2~ are the permittivity of the two glasses, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \varepsilon \rangle ={\varepsilon }_{1}-{\varepsilon }_{2}$$\end{document}$ and *f*(*U*~*r*~) is the fill factor of *ε*~1~ in a certain neighborhood *U*~*r*~ around point *r*. Technically, the nanostructured optical component consists of subwavelength sized glass rods what enables the heterogeneous materials (two different types of rods) to be considered as the homogeneous one^[@CR35]^. Therefore the M-G mixing formula is accurate. The calculated pattern of the two glass rods is presented in Fig. [9b](#Fig9){ref-type="fig"}. The simulated annealing^[@CR37]^ optimization process was used to change the continuous refractive index distribution to the discrete pattern of rods, corresponding to the GRIN lens with a parabolic distribution of the refractive index (Fig. [9a,d](#Fig9){ref-type="fig"}). Due to the finite number of rods, the average refractive index distribution did not perfectly reflect the distribution as in a lens with a continuous change in refractive index. Deviations from the ideal refractive index distribution are shown in Fig. [10](#Fig10){ref-type="fig"}. However, as shown in the simulation and experimental results, this did not significantly affect the performance of both GRIN lens and nGRIN lensed fiber system.Figure 9nGRIN lens: (**a**,**d**) ideal GRIN lens; (**b**,**e**) design of a preform composed of 9919 rods made from two glasses with refraction index *n*~*g*~ and *n*; (**c**) (detailed distribution of rods is shown in Supplementary materials), (**f**) SEM image of the final nGRIN structure.Figure 10Refractive index distribution: (**a**) in the ideal lens, and (**b**) in the microstructural lens, (**c**) a comparison of the refractive index profiles for both lenses. Fabrication of the nGRIN lens {#Sec11} ----------------------------- The stack-and-draw process was used to develop the nGRIN microlens. The technique consists of several steps shown in Fig. [11](#Fig11){ref-type="fig"}.Figure 11Scheme of the modified stack-and-draw technique: (**a**) glass rods preparation, (**b**) hexagonal preform stacked, (**c**) drawing hexagonal sub-preform, (**d**) surrounded preform by the glass rods, (**e**) drawing the final structure, (**f**) cut and polished final nGRIN elements. First, we prepared round rods, approximately 0.5 mm in diameter, made from two types of borosilicate soft glasses (Fig. [11a](#Fig11){ref-type="fig"}) named NC42 (ng = 1.56895 for the sodium D line) and NC34 (n = 1.56078 for the sodium D line). The refraction index characteristics for glasses are shown in Fig. [12](#Fig12){ref-type="fig"}. Both glasses are thermally matched and have been successfully used previously^[@CR38],[@CR39]^.Figure 12The refractive index as a function of the wavelength for the NC34 and NC42 borosilicate soft glasses. Next, the rods from both glasses were stacked to a hexagonal preform (Fig. [11b](#Fig11){ref-type="fig"}) according to the desired pattern (Fig. [9a](#Fig9){ref-type="fig"}) (detailed distribution of rods is shown in Supplementary materials). In this case, the preform consisted of 9919 rods (115 rods ordered on a diagonal, 5797 and 4122 rods with glass with higher and lower refraction index, respectively) and their pattern was chosen such that the average refractive index corresponded to the parabolic refractive index distribution in GRIN lens (Eq. [1](#Equ1){ref-type=""}). Subsequently, the preform was first drawn to a hexagonal structure of 3 mm in diameter (Fig. [11c](#Fig11){ref-type="fig"}). Next, the hexagonal sub-preform was surrounded by the glass rods of low index of refraction (Fig. [11d](#Fig11){ref-type="fig"}), to assure cylindrical shape of cladding and to facilitate the process of final element diameter scaling during the drawing process (Fig. [11e](#Fig11){ref-type="fig"}). In the last step, the fiber structures were cut into slices, grounded and polished to the required thickness (Fig. [11f](#Fig11){ref-type="fig"}). In our final structures, the diameter of the individual rods (\~190 nm) was equal or smaller than 1/5 of the propagating light wavelength (Fig. [9b](#Fig9){ref-type="fig"}). The final nGRIN lens used for the lensed fiber system had the diameter of 22.8/125 µm, where the first value corresponds to the graded area diameter and the second value is the total fiber diameter. The gradient constant *g* is equal 80.14 mm^−2^ (Eq. [1](#Equ1){ref-type=""}) and the pitch Λ (Eq. [2](#Equ2){ref-type=""}) is equal 702 μm. The measured value of the numeric aperture NA = 0.16. The stack-and-draw method allows to fabricate elements with unique properties. First of all, it allows to freely shape the distribution of the refractive index in the direction perpendicular to the length of the fiber. In particular, we have shown that we are able to fabricate with this technology diffraction elements (DOE)^[@CR38]^ and GRIN-type elements such as: elliptical lenses^[@CR30]^, lenses with an extended focal length -- axicon^[@CR40]^, micro-lenses array^[@CR41]^ and optical vortex^[@CR42]^, where the refractive index changes not along the fiber radius but perpendicularly. Moreover, the fiber with optical vortex structure cannot be fabricated with the use of any other methods available. Additionally, the elements obtained in this way are compatible with fibers optics, i.e. they have comparable size to standard fibers and, as shown in the article, can be combined into more complex optical systems such as the nGRIN lensed fiber system. The presented system allows for optical trapping in a single trap, but adding a DOE element or using an array of microlenses may allow to trap more elements. Similarly, a new functionality could be achieved by replacing the nGRIN lens with an extended focal length lens (axicon) or an element that allows the creation of an optical vortex. Because the method is compatible with optical fiber fabrication, it is possible to combine several functionalities within a single fiber (micro-lens, drug delivery channel, electrodes), which can be useful in optofluidic applications. Moreover, stacking optical elements from two glasses and drawing them, allows to obtain elements with a very high gradient of the refractive coefficient, which, as described in the introduction, is not possible with other methods. The method described here is characterized by very high repeatability of fabricated elements. Similarly to the fabrication of optical fibers where a fiber with identical parameters along its length is obtained (with a good control of the drawing parameters), here the fiber structure with constant parameters is obtained. Then, from one long fiber thousands of identical elements are obtained through cutting. The proposed method allows to create nGRIN lenses with performance nearly independent of the wavelength. These properties depend, to a large extent, on the chromatic properties of the glasses from which the GRIN element is fabricated. As opposed to the ion exchange method^[@CR19]^, which does not grant full control over the chromatic properties of the fabricated elements due to the limited set of ions that can be exchanged, the method proposed here, allows to select glasses for their dispersive properties. In our case, the difference in the refractive index of NC34 and NC42 (Fig. [12](#Fig12){ref-type="fig"}), in the range from 0.5 to 2 μm, is at level of 10^−4^ RIU, which results in a very low dependence of the lens focal length on the wavelength. Fabrication of the nGRIN lensed fiber system {#Sec12} -------------------------------------------- Usually, the splicing method is used to connect elements in fiber optics. However, in our case gluing was used due to the thermal mismatch between materials from which the fiber, spacer and nGRIN lens are made. The SM980 (Thorlabs) fiber was made of pure silica for which the glass-transition temperature is equal 1200 °C. The spacer and the lens were made of borosilicate soft glasses NC34 and N42, with a glass-transition temperature of 530 °C. Splicing of components made of materials with such a large thermal mismatch is challenging, however it can be done using a specialist splicing machine^[@CR43]^. The nGRIN lensed fiber system consisted of a standard single mode fiber SM980 (Thorlabs) with diameter of 125 µm and a flat ending, and a bulk spacer of the same diameter attached to the fiber facet using optical glue whose refractive index was about 1.56. The nGRIN lens (125 µm in diameter) was also attached to the end facet of the spacer with glue (Fig. [13](#Fig13){ref-type="fig"}). The final nGRIN lensed fiber system allowed for focusing light in the air at wavelength λ = 976 nm to a spot diameter of 8.15 μm, at a distance of 54 μm from the end facet of the nGRIN lens. In water, for the same wavelength, the beam was focused to a spot diameter of approximately 7.5 μm at a distance of 73 μm.Figure 13Microscopic image of 102 µm nGRIN lens, assembled with 105 µm spacer and SM980. The system was assembled in the following way. We began to attach the spacer and the nGRIN lens to the fiber by dipping the end of the fiber in the optical glue. The plate with the spacer was placed on the microscope table. Then, the fiber and the spacer were centered in one axis and placed adjacent to each other. The setup was illuminated with the UV light for several minutes to harden the glue. After the illumination, the same steps were executed with the nGRIN lens. After the second illumination, the probe was ready for measurements. Numerical simulations and calculations {#Sec13} -------------------------------------- Numerical simulations of the beam propagation both for the ideal and the nGRIN lenses were based on the fast Fourier transform beam propagation method (FFT BPM)^[@CR44]^. The analysis was carried out for the Gaussian beam propagating at a distance of 400 µm in a conventional SM optical fiber with the core diameter of 4.6 µm. Then, the light passed through the bulk glass spacer of a fixed length (105 µm) and refractive index *n*~*S*~ = 1.56895, and then through the nGRIN lens. The two glasses used in the nGRIN structure of the lens had refractive indices of *n*~*g*~ = 1.56895 and *n*~0~ = 1.56078. In order to ensure the convergence of the algorithm in the case of an ideal GRIN lens simulation, it is sufficient to sample every 0.25 μm in the plane perpendicular to the beam propagation and every 1 μm in the direction of the propagation. However, in the case of the nGRIN lens, it was necessary to sample with a resolution of 0.057 μm in the perpendicular plane and 0.2 μm in the direction of the beam propagation. Calculations for the working distance (WD) and the beam spot diameter (BS) were based on the matrix formulation of Gaussian optics^[@CR29]^ according to equations^[@CR45]^:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$WD=\frac{{n}_{w}[(1+{(\frac{{a}_{0}{L}_{s}}{{n}_{s}})}^{2}-{(\frac{{a}_{0}}{{n}_{g}g})}^{2})\sin (2g{L}_{g})-2\frac{{a}_{0}^{2}{L}_{s}}{{n}_{s}n{}_{g}g}\,\cos (2g{L}_{g})]}{2{n}_{g}g[{\sin }^{2}(g{L}_{g})+{(\frac{{a}_{0}}{{n}_{s}{n}_{g}g})}^{2}{({n}_{s}\cos (g{L}_{g})-{n}_{g}g{L}_{0}\sin (g{L}_{g}))}^{2}]}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$BS=\frac{{a}_{0}{w}_{0}}{{n}_{g}g[{\sin }^{2}(g{L}_{g})+{(\frac{{a}_{0}}{{n}_{s}{n}_{g}g})}^{2}{({n}_{s}\cos (g{L}_{g})-{n}_{g}g{L}_{0}\sin (g{L}_{g}))}^{2}]}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${a}_{0}=\frac{\lambda }{\pi {n}_{f}{w}_{0}^{2}}$$\end{document}$$where: *n*~*s*~ is the refractive index of the spacer between the optical fiber and the GRIN lens, *L*~*s*~ is the length of the spacer, *n*~*g*~ is the refractive index of the GRIN lens at the center, *L*~*g*~ is the length of the GRIN lens, *g* is the gradient constant of the GRIN lens, *n*~*w*~ is the refractive index of an imaged specimen, *w*~o~ is the initial beam radius from the optical fiber and, *n*~*f*~ is the refractive index of the optical fiber. Characterization of the nGRIN lensed fiber system {#Sec14} ------------------------------------------------- The nGRIN lensed fibre system was characterized to verify its working distance (WD) and beam spot diameter (BS). All measurements were made in the air. We used 976 nm laser (BL976-SAG300. 300 mW) coupled into our lensed fiber system as the light source. At the output, the beam was magnified with an ×20 microscope objective (NA = 0.35) and then projected onto a CCD camera that works with a fixed gain in the linear regime (Fig. [14](#Fig14){ref-type="fig"}). The focal plane of the lens was determined by scanning the beam along the optical axis with the imaging system, with the translation resolution of ±50 nm. In contrast, the measurement of the absolute distance WD was made with an accuracy of ±2 μm.Figure 14Schematic setup for measuring the working distance (WD) and the beam spot (BS). Optical trapping {#Sec15} ---------------- To test the nGRIN lensed fiber system for optical trapping, the setup depicted in Fig. [15a](#Fig15){ref-type="fig"} was used. The setup featured a part dedicated to imaging and image acquisition (tinted blue in Fig. [15a](#Fig15){ref-type="fig"}), a part devoted to building the optical trap (tinted green in Fig. [15a](#Fig15){ref-type="fig"}), and a part allowing for manipulating the sample (tinted red in Fig. [15a](#Fig15){ref-type="fig"}). The image acquisition system was based on the inverted microscope with an ×40, NA = 0.65 objective, and a CMOS camera (SC50), before which a dichroic filter was located. A precise manipulation of the sample was possible thanks to a 3-axis piezo micromanipulator with 20 μm range of motion. The optical fiber with a glued nGRIN lens was placed in a metal holder and fixed on a computer-steered rotating table which allowed for precisely setting the tilt angle of the fiber (Fig. [15b](#Fig15){ref-type="fig"}). A 976 nm pigtailed laser diode was used for the trapping (BL976-SAG300. 300 mW) wherein the measured light power at the nGRIN lensed fiber output was 120 mW.Figure 15Optical trapping setup with nGRIN lensed fiber system: (**a**) the whole setup, and (**b**) enlarge trapping head. A modular structure of the setup allows for its simple integration with the existing inverted microscope systems. Moreover, joining the sample shifting module with the optical trapping system makes the setup independent from the microscopic system. The setup does not use any microscope optical path and can be added to the majority of microscopes. Electronic supplementary material ================================= {#Sec16} Supplementary materials Movie1 Movie 2 **Electronic supplementary material** **Supplementary information** accompanies this paper at 10.1038/s41598-018-23464-6. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This work was supported by the project TEAM TECH /2016-1/1 operated within the Foundation for Polish Science Team Programme co-financed by the European Union under the European Regional Development Fund and EU-H2020-ICT-2014 grant MIREGAS: Programmable multi-wavelength Mid-IR source for gas sensing, Grant No. 644192. Concept of nGRIN lensed fiber and its application is created by R.B. and R.K., A.F., P.S. and R.B. designed the nGRIN element and fabricated nGRIN lensed fiber system. D.P., R.S., K.K. and A.F. fabricated the nGRIN elements. A.F., P.S., A.A., G.S., M.K. and R.K. measured the properties of the nGRIN lensed fiber system. R.K. and A.F. did the computational simulations and numerical analysis. R.K. conducted the experiments with 2D optical trapping. R.B. and R.K. supervised the work. The manuscript was written by R.K., A.F., P.S., R.B. and M.K. Competing Interests {#FPar1} =================== The authors declare no competing interests.

1. Introduction {#sec1-molecules-23-00708} =============== In humans, *N*-acetyl-[d]{.smallcaps}-hexosaminidases (HexA, HexB, and HexS) play essential roles in the lysosomal processing of degradation-bound glycolipids as well as glycans \[[@B1-molecules-23-00708]\]. *O*-GlcNAcase \[[@B2-molecules-23-00708]\] removes O-linked GlcNAc residues from serine or threonine in nucleocytoplasmic proteins. Another *N*-acetylhexosaminidase, HexD, was recently reported and has been mentioned in context with rheumatoid arthritis \[[@B3-molecules-23-00708]\]. Whereas the retaining lysosomal *N*-acetyl-α-[d]{.smallcaps}-hexosaminidases of GH89 follow the standard double-displacement mechanism, *N*-acetyl-β-[d]{.smallcaps}-hexosaminidases of GH20, as well as the GH84 *O*-GlcNAcase, GH18 chitinases, and GH85 *endo*-*N*-acetyl-β-[d]{.smallcaps}-glucosaminidases exploit anchimeric assistance by the sugar's *N*-acetyl group, which upon intramolecular attack of the intermediary oxocarbenium ion forms an α-configured oxazoline intermediate. This, in turn, is attacked from the β-face by an activated water molecule \[[@B4-molecules-23-00708]\]. Interesting reviews on hexosaminidases are available \[[@B5-molecules-23-00708],[@B6-molecules-23-00708]\]. Potent and highly selective inhibitors are required for studying these enzymes and their physiological significance. In particular, potential therapeutic applications in context with lysosomal disorders \[[@B7-molecules-23-00708]\], cancer \[[@B8-molecules-23-00708]\], and Alzheimer's disease \[[@B9-molecules-23-00708]\] require high degrees of selectivity for any of the enzymes mentioned above over the other respective *N*-acetylhexosaminidases. Amongst *N*-acetylhexosaminidase inhibitors ([Figure 1](#molecules-23-00708-f001){ref-type="fig"}), 2-acetamido-1,2-dideoxynojirimycin, **1** \[[@B10-molecules-23-00708]\] and diastereomers such as the [d]{.smallcaps}-*galacto* (**2**) \[[@B11-molecules-23-00708]\] and the [d]{.smallcaps}-*allo* (**3**) \[[@B12-molecules-23-00708]\] analogs, PUGNAc (**4**) \[[@B13-molecules-23-00708]\], and NAG-thiazoline (NGT, **5**) \[[@B14-molecules-23-00708]\] have attracted considerable attention. Furthermore, Thiamet G (**6**) \[[@B15-molecules-23-00708]\], nagstatin (**7**) \[[@B16-molecules-23-00708]\], and 6-acetamido-6-deoxycastanospermine (**8**) \[[@B17-molecules-23-00708]\], various pyrrolidine derivatives (for example, compound **9** \[[@B18-molecules-23-00708]\]), as well as 2-*N*-acetyl glycals including **10** \[[@B19-molecules-23-00708]\] have been reported. Amongst carbacyclic hexosaminidase inhibitors, pyranoid carbasugar acetamidodeoxy-β-valienamine (**11**) has to be mentioned \[[@B20-molecules-23-00708]\]. These inhibitors are either substrate/product analogs or, in the case of bicyclic systems such as NGT, are chemically stable structural analogues of the above-mentioned intermediate generated by anchimeric assistance of the *N*-acetyl group at C-1 at the first transition state of enzymatic *N*-acetylhexosaminide hydrolysis. Jäger and co-workers \[[@B21-molecules-23-00708]\] have directed our attention toward cyclopentanoid basic sugar analogs as potentially useful inhibitors of lysosomal glycosidases. Relying on the pioneering synthetic work by Vasella and co-workers \[[@B22-molecules-23-00708]\] based on guiding contributions by Padwa \[[@B23-molecules-23-00708]\] as well as Oppolzer \[[@B24-molecules-23-00708]\], we had thus investigated cyclopentane-based β-galactosidase inhibitors \[[@B25-molecules-23-00708]\] and have recently extended our range of new compounds by addition of *N*-acetyl-β-[d]{.smallcaps}-galactosaminide and -β-[d]{.smallcaps}-glucosaminide analogs. 2. Results and Discussion {#sec2-molecules-23-00708} ========================= 2.1. Synthesis {#sec2dot1-molecules-23-00708} -------------- Starting from known \[[@B26-molecules-23-00708]\] *N*-benzylisoxazolidine **12** ([Scheme 1](#molecules-23-00708-sch001){ref-type="scheme"}), by a simple oxidation/reduction sequence, ketone **13** provided epimer **14** in high yield. Its structural identity can unambiguously be verified by X-ray structure determination ([Figure 2](#molecules-23-00708-f002){ref-type="fig"}). The corresponding triflate **15** provided azidodeoxy derivative **16** by clean inversion of configuration. Intermediate **16**, upon reduction with Zn under slightly acidic conditions and subsequent conventional *N*-acetylation, highly selectively furnished, via free amine **17**, desired acetamido compound **18**, which yielded crystals of sufficient quality for XRD ([Figure 2](#molecules-23-00708-f002){ref-type="fig"}). From tricycle **18**, by hydrogenolysis over Perlman's catalyst, cyclopentane **19** featuring a free amine was obtained in good yield. Subsequent acidic deprotection yielded free aminotriol **20** ([Scheme 2](#molecules-23-00708-sch002){ref-type="scheme"}). Chemoselective *N*-alkylation of intermediate **19** provided the corresponding *N*-hexyl (**21**), *N*-methoxycarbonylpentyl (**22**), as well as *N*-cyanopentyl (**23**) derivatives. By reduction of the nitrile function in **23**, primary amine **24** became available, which was converted into fluorescent dansylaminohexyl derivative **25** with the aid of dansyl chloride. 2.2. Biological Evaluation {#sec2dot2-molecules-23-00708} -------------------------- New compounds turned out particularly potent inhibitors of *Streptomyces plicatus N*-acetyl-β-hexosaminidase (SpHex) with *K*~i~-values in the sub-nanomolar range ([Table 1](#molecules-23-00708-t001){ref-type="table"}). By introduction of a dansylamido moiety into the alkyl chain, this activity was further improved as shown by the 60 pM value determined for inhibitor **25** when compared to analogue **21**. With the same enzyme, compound **1** exhibited *K*~i~ = 80 μM \[[@B26-molecules-23-00708]\]. The latter two compounds were also screened with Tay-Sachs disease-related human lysosomal *N*-acetyl-β-hexosaminidase A. For comparison, pyrimethamine \[[@B27-molecules-23-00708]\], which under the same screening conditions exhibited IC~50~ = 62 μM, as well as 2-acetamido-1,2-dideoxynojirimycin,**1** (IC~50~ 31 μM) \[[@B28-molecules-23-00708]\] were included in these studies. Inhibitors **21** and **25** exhibited excellent properties with this vital human enzyme, considerably exceeding the activities of the reference compounds probed in this study. In conclusion, compounds of this new aminocyclopentane-derived family of *N*-acetylgalactosaminide mimetics represent a potentially interesting class of *N*-acetyl-[d]{.smallcaps}-hexosaminidase inhibitors and pharmacological chaperones for treatment of Tay Sachs disease, in particular, when also considering their simplicity of synthesis. 3. Materials and Methods {#sec3-molecules-23-00708} ======================== 3.1. General Methods {#sec3dot1-molecules-23-00708} -------------------- Optical rotations were measured at 20 °C on a Perkin Elmer (Waltham, MA, USA) 341 polarimeter at a wave length of 589 nm and a path length of 10 cm. NMR spectra were recorded on a Varian (Palo Alto, CA, USA) INOVA 500 operating at 499.82 MHz (^1^H), and at 125.894 MHz (^13^C), or on a Bruker (Billerica, MA, USA) Ultrashield spectrometer at 300.36 and 75.53 MHz, respectively. CDCl~3~ was employed for protected compounds and CD~3~OD as well as D~2~O for unprotected inhibitors. Chemical shifts are listed in delta employing residual, non-deuterated solvent as the internal standard. Signals were assigned unambiguously by COSY and HSQC analysis. The signals of the *N*-dansyl group are located in the expected regions and are not listed explicitly. For easier comparison with other *N*-acetylgalactosaminide analogues, interpretation of NMR-spectra was performed according to the carbohydrate-related numbering system depicted in [Scheme 2](#molecules-23-00708-sch002){ref-type="scheme"}. MALDI-TOF and EI-TOF mass spectrometry were performed on a Micromass (Waters Corporation, Milford, MA, USA) TofSpec 2E Time-of-Flight mass spectrometer. Analytical TLC was performed on pre-coated aluminum plates silica gel 60 F254 (E. Merck, Darmstadt, Germany 5554) and detected with UV light (254 nm). For staining, a solution of vanillin (9 g) in a mixture of H~2~O (950 mL)/EtOH (750 mL)/H~2~SO~4~ (120 mL) or ceric ammonium molybdate (100 g ammonium molybdate/8 g ceric sulfate in 1 l 10% H~2~SO~4~) were employed, followed by heating on a hotplate. For column chromatography, silica gel 60 (230--400 mesh, E. Merck 9385) or silica gel 60 (Acros Organics (Thermo Fisher Scientific Inc., Waltham, MA, USA), AC 24036) were used. CCDC contains the [supplementary crystallographic data](#app1-molecules-23-00708){ref-type="app"} for this paper. These data can be obtained free of charge via <http://www.ccdc.cam.ac.uk/conts/retrieving.html> (or from the CCDC, 12 Union Road, Cambridge CB2 1EZ, UK; Fax: +44 1223 336033; E-mail: <deposit@ccdc.cam.ac.uk>). 3.2. Biochemical Methods {#sec3dot2-molecules-23-00708} ------------------------ *Streptomyces plicatus N*-acetyl-β-hexosaminidase, SpHex, was expressed and purified in *E. coli* as described previously \[[@B27-molecules-23-00708]\]. Kinetic studies were performed at 25 °C in the assay buffer (sodium phosphate (50 mM), sodium citrate (50 mM), NaCl (100 mM), BSA (2 mg/mL), pH = 6.0). The enzyme was incubated with different concentrations of the inhibitors for 2--5 min prior to the start of the reaction by addition of the substrate (4-nitrophenyl *N*-acetyl-β-[d]{.smallcaps}-glucosaminide) and the initial rates were measured by monitoring the increase in absorbance at 405 nm for three to five minutes using a microplate reader (Synergy H1, BioTek, VT, USA). *K*~i~ determinations were performed using two or three different substrate concentrations. For each one of these substrate concentrations a range of five to eight different inhibitor concentrations bracketing the ultimately determined *K*~i~ value were used. Dixon plots (1/rate vs \[I\]) were constructed to validate the use of a competitive inhibition model. The data were then fit to a competitive inhibition model using non-linear regression analysis with GraFit 7.0 (Erithacus Software, UK). Assays were done twice using enzyme concentrations of 0.3 nM and 0.03 nM respectively in order to check compliance with the assumptions of Michaelis Menten kinetics (\[E\] \<\< \[I\]). In addition, for compound **25**, the assays were done a third time with the concentration of enzyme lowered to 0.003 nM. In all cases, the *K*~i~ values were in good agreement. Human skin fibroblasts (wild type) were grown in minimal essential medium (MEM) with Earle's Salts (Sigma Aldrich, St. Louis, MO, USA) containing 10% fetal bovine serum, 400 µM [l]{.smallcaps}-glutamine, and 50 µg/mL gentamycin at 37 °C and 5% CO~2~. All cells used in this study were between the third and nineteenth passages. All inhibitors were dissolved in DMSO in a concentration of 10 mM and diluted in 10 mM phosphate buffer (pH 7.0) containing 100 mM NaCl, 0.01% NaN~3~, and 0.01% Triton for the IC~50~-measurements. Human *N*-acetyl-β-hexosaminidase A activity measurements were performed in duplicate assays unless otherwise stated. Fibroblast cells of three single 80 cm^2^ flasks were harvested by trypsinization in 500 µL (each of them) 0.9% NaCl containing 0.01% Triton, homogenized by sonication (4 times 15 s, Bandelin (Berlin, Germany) Sonopuls ultrasound homogenator mini 20) and centrifuged at 13.000 rpm for 1 min in a table top centrifuge (Biofuge Pico, Heraeus, Hanau, Germany). Protein amounts were determined according to the method of Lowry. For assessment of *N*-acetyl-β-hexosaminidase A activity, 10 µL (diluted 1:10) of cell homogenate were mixed with 90 µL 0.9% NaCl and 200 µL of 1 mM (4-methyl)umbelliferyl *N*-acetyl-β-glucosaminide-6-sulfate. Na (Glycosynth, Warrington, UK) in McIlvains phosphate/citrate puffer, pH 4,4. After incubation at 37 °C for 60 min, the reaction was stopped by adding 2.5 mL 400 mM glycine/NaOH (pH 10.4). The amount of hydrolyzed 4-methylumbelliferone was determined with a fluorescence spectrometer (F7000 Hitachi, Chiyoda, Japan). Modified β-hexosaminidase A assays were used to estimate the half maximal inhibitory concentration (IC~50~) of the particular inhibitor. For IC~50~ determination, 0.001 to 100 µM of inhibitor was added to the assay mixture. Activity was measured in normal fibroblasts. Data analysis was performed with Microcal™ Origin^®^ v6.0 (Origin Lab, Northampton, MA, USA) using a non-linear curve fitting module based on sigmoid curve fitting. 3.3. (3aR,3bS,6aR,7R,7aR)-Hexahydro-5,5-dimethyl-1-phenyl-1H-\[1,3\]dioxolo\[3,4\]cyclopent\[1,2-c\]isoxazol-7-ol or 1-[l]{.smallcaps}-(1,2,3,4,5)-1^1^,2^1^-Anhydro-1-hydroxymethyl-2-(N-hydroxy)benzylamino-4,5-O-isopropylidene-3,4,5-cyclopentanetriol **14** {#sec3dot3-molecules-23-00708} ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- \(a\) *Via Swern oxidation*: To a solution of oxalyl chloride (1.12 mL, 13.0 mmol) in CH~2~Cl~2~ (20 mL), DMSO (1.11 mL, 15.7 mmol) was added dropwise at −60 °C. After 15 min, a 50% solution (*w*/*v*) of alcohol **12** (1.52 g, 5.22 mmol) in CH~2~Cl~2~ was added and the reaction was stirred for 15 min when Et~3~N (2.89 mL, 20.9 mmol) was added. The reaction mixture was allowed to reach ambient temperature and methanol (50 mL) and NaBH~4~ (0.39 g, 10.4 mmol) were added. When completed conversion of cyclopentanone **13** was observed (tlc, 30 min), solvents were removed under reduced pressure and crude alcohol **14** was dissolved in CH~2~Cl~2~. The organic layer was extracted with saturated aqueous NaHCO~3~, dried (Na~2~SO~4~), filtered and concentrated under reduced pressure. Purification on silica gel (cyclohexane/ethyl acetate 3:1 *v*/*v*) provided compound **14** as a pale yellow syrup (848 mg, 2.91 mmol, 55.8% from epimer **12**). \(b\) *Via Dess-Martin oxidation*: A 10% solution (*w*/*v*) of alcohol **12** (1.05 g, 3.60 mmol) in CH~2~Cl~2~ was stirred with Dess-Martin periodinane (1.68 g, 3.96 mmol) at ambient temperature for 10 min. After completed conversion, the reaction mixture was washed with saturated aqueous NaHCO~3~, dried (Na~2~SO~4~), and filtered. Removal of solvents under reduced pressure gave the crude product **13**. To a solution of crude ketone **13** in methanol (20 mL), NaBH~4~ (0.273 g, 7.21 mmol) was added. When completed conversion was detected (tlc, 30 min), solvents were removed under reduced pressure, and the crude product was dissolved in CH~2~Cl~2~. The organic layer was extracted with saturated aqueous NaHCO~3~, dried (Na~2~SO~4~), filtered, and concentrated under reduced pressure. Purification on silica gel (cyclohexane/ethyl acetate 3:1 *v*/*v*) provided compound **14** as a pale yellow syrup (0.769 g, 2.64 mmol, 73.2% over two steps). After extended storage, a compound sample provided a few minute crystals, one of which could be exploited for X-ray structure determination (CCDC 1826202). ${\lbrack\mathsf{\alpha}\rbrack}_{D}^{20}$: +44.6 (*c* = 0.86, CHCl~3~); ^1^H-NMR (300 MHz, CDCl~3~) δ = 7.42--7.23 (m, 5H, aromatic NBn), 4.58 (dd, 1H, *J*~2,3~ = *J*~3,4~ = 5.5 Hz, H-3), 4.50 (dd, 1H, *J*~4,5~ = 7.5 Hz, H-4), 4.33 (d, 1H, *J*~5,6a~ \< 1 Hz, *J*~6a,6b~ = 8.7 Hz, H-6a) 4.12 (d, 1H, *J* = 13.3 Hz, N-[CH~2~]{.ul}-Ph), 4.04 (m, 1H, H-2), 3.97 (dd, 1H, *J*~5,6b~ = 6.4 Hz, H-6b), 3.91 (dd, 1H, N-[CH~2~]{.ul}-Ph), 3.68 (dd, 1H, *J*~1,2~ = 7.6 Hz, H-1), 3.48 (bs, 1H, 6-OH), 3.07 (ddd, 1H, H-5), 1.57, 1.25 (2s, 3H each, C([CH~3~]{.ul})~2~). ^13^C-NMR (75.5 MHz, CDCl~3~): δ = 137.2 (ipso NBn), 129.2, 128.6, 127.6 (aromatic NBn), 112.8 ([C]{.ul}(CH~3~)~2~), 81.7 (C-3), 78.3 (C-4), 73.2 (C-1), 71.2 (C-2), 65.6 (C-6), 62.1 (N-[CH~2~]{.ul}-Ph), 49.6 (C-5), 25.4, 25.0 (C([CH~3~]{.ul})~2~). MS (MALDI): Calcd for \[C~16~H~21~NO~4~Na\]: *m/z* 314.1368 \[M + Na\]^+^; Found \[M + Na\]^+^ 314.1368. 3.4. (3aR,3bS,6aR,7S,7aR)-Hexahydro-7-azido-5,5-dimethyl-1-phenyl-1H-\[1,3\]dioxolo\[3,4\]cyclopent\[1,2-c\]isoxazol or 1-[l]{.smallcaps}-(1,2,4,5/3)-1^1^,2^1^-Anhydro-3-azido-1-hydroxymethyl-2-(N-hydroxy)benzylamino-4,5-O-isopropylidene-4,5-cyclopentanediol **16** {#sec3dot4-molecules-23-00708} ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- A solution of alcohol **14** (848 mg, 2.91 mmol) in CH~2~Cl~2~ (20 mL) was cooled to 0 °C. Pyridine (0.940 mL, 11.6 mmol) and trifluoromethanesulfonyl anhydride (0.637 mL, 3.78 mmol) were added. When completed conversion of the starting material was observed (10 min), the reaction mixture was washed consecutively with HCl (6%) and saturated aqueous NaHCO~3~. After drying with Na~2~SO~4~, the suspension was filtered, and the solvent was removed at room temperature under reduced pressure. Resulting crude triflate **15** was dissolved in DMF (20 mL), NaN~3~ (1.14 g, 17.5 mmol) was added and the mixture was stirred at ambient temperature for 60 min. The reaction mixture was then concentrated under reduced pressure, the residue was dissolved with CH~2~Cl~2~, and the solution was washed with brine. The organic layer was dried (Na~2~SO~4~), filtered, and concentrated under reduced pressure. Purification of the remaining residue on silica gel (cyclohexane/ethyl acetate 10:1 *v*/*v*) provided azidodeoxy compound **16** (568 mg, 1.80 mmol, 61.7% from **14**). ${\lbrack\mathsf{\alpha}\rbrack}_{D}^{20}$: +74.8 (*c* = 1.09, CHCl~3~); ^1^H-NMR (300 MHz, CDCl~3~) δ = 7.44--7.23 (m, 5H, aromatic NBn), 4.59 (dd, 1H, *J*~3,4~ = *J*~4,5~ = 6.9 Hz, H-4), 4.31 (dd, 1H, *J*~5,6a~ = 3.5 Hz, *J*~6a,6b~ = 8.9 Hz, H-6a), 4.28 (m, 1H, H-3), 4.07 (d, 1H, H-6b), 4.01 (d, 1H, *J* = 12.6 Hz, N-[CH~2~]{.ul}-Ph), 3.78 (dd, 1H, *J*~1,2~ = *J*~2,3~ = 7.5 Hz, H-2), 3.70 (d, 1H, N-[CH~2~]{.ul}-Ph), 3.61 (dd, 1H, *J*~1,5~ = 8.2 Hz, H-1), 3.29 (dddd, 1H, H-5), 1.54, 1.29 (2s, 3H each, C([CH~3~]{.ul})~2~. ^13^C-NMR (75.5 MHz, CDCl~3~): δ = 136.5 (ipso NBn), 129.1, 128.6, 127.8 (aromatic NBn), 113.4 ([C]{.ul}(CH~3~)~2~), 84.5 (C-3), 77.4 (C-4), 75.6 (C-1), 69.9 (C-2), 64.9 (C-6), 59.5 (N-[CH~2~]{.ul}~-~Ph), 45.6 (C-5), 27.7, 25.7 (C([CH~3~]{.ul})~2~). MS (EI): Calcd for \[C~16~H~20~N~4~O~3~\]: *m*/*z* 316.1535 \[M\]^+^; Found \[M\]^+^ 316.1532. 3.5. (3aR,3bS,6aR,7S,7aR)-Hexahydro-7-acetamido-5,5-dimethyl-1-phenyl-1H-\[1,3\]dioxolo\[3,4\]cyclopent\[1,2-c\]isoxazol or 1-[l]{.smallcaps}-(1,2,4,5/3)-1^1^,2^1^-Anhydro-3-acetamido-1-hydroxymethyl-2-(N-hydroxy)benzylamino-4,5-O-isopropylidene-4,5-cyclopentanediol **18** {#sec3dot5-molecules-23-00708} --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- To a stirred suspension of zinc (1.17 g, 18.0 mmol) and NH~4~Cl (0.961 g, 18.0 mmol) in methanol (20 mL) a 50% solution (*w*/*v*) of azidocyclopentane **16** (848 mg, 2.91 mmol) in methanol was added. After completed conversion of the starting material (2 h), the mixture was filtered and concentrated under reduced pressure. The resulting crude amine **17** was dissolved in pyridine (20 mmol) and treated with acetic anhydride (0,255 mL, 2.69 mmol) and 4-DMAP (5 mg) at 0 °C. After completed consumption of amine **17**, the reaction was quenched with methanol, and the solvents were removed under reduced pressure. The residue was dissolved in CH~2~Cl~2~ and consecutively washed with HCl (6%) and saturated aqueous NaHCO~3~, dried (Na~2~SO~4~) and filtered. Purification over silica gel chromatography (cyclohexane/ethyl acetate 2:1 *v*/*v*) provided acetamide **18** (422 mg, 1.27 mmol, 70.7% from **16**) as a pale yellow syrup. ${\lbrack\mathsf{\alpha}\rbrack}_{D}^{20}$: +11.1 (*c* = 0.82, CHCl~3~); ^1^H-NMR (300 MHz, CDCl~3~) δ = 7.38--7.23 (m, 5H, aromatic NBn), 6.11 (d, 1H, N[H]{.ul}COCH~3~), 4.95 (dd, 1H, *J*~2,3~ = *J*~3,4~ = 6.3 Hz, H-3), 4.75 (dd, 1H, *J*~4,5~ = 7.1 Hz, H-4), 4.29 (dd, 1H, *J*~5,6a~ = 4.1 Hz, *J*~6a,6b~ = 8.7 Hz, H-6a), 4.22 (dd, 1H, *J*~1,2~ = *J*~1,5~ = 8.0 Hz, H-1), 4.06 (dd, 1H, *J*~5,6b~ = 8.9 Hz, H-6b), 3.98 (d, 1H, *J* = 12.9 Hz, N-[CH~2~]{.ul}-Ph), 3.67 (d, 1H, N-[CH~2~]{.ul}-Ph), 3.42 (m, 1H, H-5), 3.34 (dd, 1H, H-2), 1.83 (s, 3H, NHCO[CH~3~]{.ul}), 1.51, 1.29 (2s, 3H each, C([CH~3~]{.ul})~2~). ^13^C-NMR (75.5 MHz, CDCl~3~): δ = 170.7 (NH[CO]{.ul}CH~3~), 137.0 (ipso NBn), 129.2, 128.5, 127.6 (aromatic NBn), 112.6 ([C]{.ul}(CH~3~)~2~), 83.2 (C-3), 78.4 (C-4), 74.3 (C-1), 65.4 (C-6), 63.9 (C-2), 59.9 (N-[CH~2~]{.ul}-Ph), 47.1 (C-5), 27.3, 25.4 (C([CH~3~]{.ul})~2~), 23.6 (NHCO[CH~3~]{.ul}). After extended storage, a compound sample provided small crystals which could be employed for X-ray structure determination (CCDC 1826203). MS (EI): Calc for \[C~18~H~24~N~2~O~4~\]: *m/z* 332.1736 \[M\]^+^; Found \[M\]^+^ 332.1737. 3.6. (3aS,4R,5R,6S,6aR)-5-Amino-tetrahydro-6-acetamido-2,2-dimethyl-4H-cyclopenta-1,3-dioxole-4-methanol or 1-[l]{.smallcaps}-(1,2,4,5/3)-3-Acetamido-2-amino-1-hydroxymethyl-4,5-O-isopropylidene-4,5-cyclopentanediol **19** {#sec3dot6-molecules-23-00708} ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ A 5% solution of acetamide **18** (422 mg, 1.27 mmol) in methanol was stirred with Pearlman's catalyst (Pd(OH)~2~/C, 20%) under an atmosphere of H~2~ at ambient pressure. After completed conversion (1 hour), the catalyst was filtered off, the filtrate was concentrated under reduced pressure, and the residue was chromatographically purified (chloroform/methanol/NH~4~OH (25%) 14:1:0.01 *v*/*v*/*v*) to obtain intermediate **19** as a pale yellow syrup (253 mg, 1.04 mmol, 81.6%). ${\lbrack\mathsf{\alpha}\rbrack}_{D}^{20}:$ +7.5 (*c* = 0.85, CHCl~3~); ^1^H-NMR (300 MHz, CDCl~3~) δ = 7.29 (d, 1H, N[H]{.ul}COCH~3~), 4.68 (dd, 1H, *J*~3,4~ = 6.0 Hz, *J*~4,5~ = 5.7 Hz, H-4), 4.47 (d, 1H, H-3), 4.05 (d, 1H, *J*~1,2~ = 6.5 Hz, *J*~4,5~ \< 1 Hz, H-2), 3.90 (m, 2H, H-6a, H-6b), 3.25 (d, 1H, *J*~1,5~ \< 1 Hz, H-1), 3.00 (bs, 3H, 6-OH, 1-NH~2~), 2.41 (m, 1H, H-5), 1.95 (s, 3H, NHCO[CH~3~]{.ul}), 1.45, 1.25 (2s, 3H each, C(CH~3~)~2~). ^13^C-NMR (75.5 MHz, CDCl~3~): δ = 170.7 (NH[CO]{.ul}CH~3~), 111.0 ([C]{.ul}(CH~3~)~2~), 85.4 (C-3), 80.8 (C-4), 63.2 (C-2), 59.3 (C-1), 58.1 (C-6), 47.2 (C-5), 26.5, 23.2 (C([CH~3~]{.ul})~2~), 23.1 (NHCO[CH~3~]{.ul}). MS (MALDI): Calcd for \[C~11~H~20~N~2~O~4~H\]: *m/z* 245.1501 \[M + H\]^+^; Found \[M + H\]^+^ 245.1506. 3.7. (1S,2R,3S,4R,5R)-3-Acetamido-4-amino-5-hydroxymethylcyclopentanetriol or "1-amino-2-acetamido-2-deoxy-β-[d]{.smallcaps}-galacto-cyclopentane" **20** {#sec3dot7-molecules-23-00708} --------------------------------------------------------------------------------------------------------------------------------------------------------- A solution of compound **19** (34.8 mg, 0.142 mmol) in methanol (1 mL) was treated with HCl (12 M 100µL). After completed deprotection, the solvent was removed under reduced pressure, and the remaining residue was purified by silica gel chromatography (chloroform/methanol/NH~4~OH (25%) 8:4:1 *v*/*v*/*v*) to furnish aminopolyol **20** as the free base (22.4 mg, 0.110 mmol, 77.0%). Treatment with HCl~g~ in methanol in the presence of small amounts of ethyl acetate as co-solvent afforded the corresponding hydrochloride (**20**^.^HCl) as a white solid. ${\lbrack\mathsf{\alpha}\rbrack}_{D}^{20}:$ +57.6 (*c* = 0.90, H~2~O) (hydrochloride); ^1^H-NMR (500 MHz, D~2~O) (free base): δ = 4.21 (dd, 1H, *J*~3,4~ = *J*~4,5~ = 3.7 Hz, H-4), 4.12 (dd, 1H, *J*~1,2~ = 5.7 Hz, *J*~2,3~ = 9.1 Hz, H-2), 4.03 (dd, 1H, *J*~3,4~ = 3.8 Hz, H-3), 3.96 (dd, 1H, *J~5~*~,6~ = 7.4 Hz, *J*~6a,6b~ = 11.2 Hz, H-6a), 3.90 (dd, 1H, *J*~5,6b~ = 7.7 Hz, H-6b), 3.44 (dd, 1H, *J*~1,5~ = 8.7 Hz, H-1), 2.48 (dddd, 1H, *J*~4,5~ = 3.9 Hz, H-5), 2.10 (s, 3H, NHCO[CH~3~]{.ul}). ^13^C-NMR (125.9 MHz, D~2~O) (free base): δ = 174.9 (NH[CO]{.ul}CH~3~), 75.7 (C-3), 72.2 (C-4), 62.0 (C-2), 57.0 (C-6), 55.6 (C-1), 42.9 (C-5), 21.9 (NHCO[CH~3~]{.ul}). MS (MALDI): Calcd for \[C~8~H~16~N~2~O~4~H\]: *m/z* 205.1188 \[M + H\]^+^; Found \[M + H\]^+^ 2051184. 3.8. (1S,2R,3S,4R,5R)-N-(1-Hexyl)-3-acetamido-4-amino-5-hydroxymethylcyclopentanetriol or "2-Acetamido-2-deoxy-1-(hexyl)amino-β-[d]{.smallcaps}-galacto-cyclopentane" **21** {#sec3dot8-molecules-23-00708} ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Amine **19** (32.2 mg, 0.132 mmol) was dissolved in DMF (1 mL) and treated with 1-bromohexane (22.1 µL, 0.158 mmol) in the presence of NaHCO~3~ (53.2 mg, 0.633 mmol) at 60 °C. After completed consumption of the starting material, the mixture was concentrated under reduced pressure. The residue was diluted with methanol and treated with HCl (100 µL, 12 M) and stirred for one hour. After evaporation of the solvents, the remaining precipiate was purified by chromatography on silica gel (chloroform/methanol/NH~4~OH (25%) 8:1:0.01 *v*/*v*/*v*) to yield *N*-hexyl carbacycle **21** (25.4 mg, 881 µmol, 66.8% over two steps). ${\lbrack\mathsf{\alpha}\rbrack}_{D}^{20}$: +49.8 (*c* = 0.97, MeOH); ^1^H-NMR (500 MHz, CD~3~OD): δ = 4.16 (dd, 1H, *J*~1,2~ = 5.0 Hz, *J*~2,3~ = 7.7 Hz, H-2), 4.08 (dd, 1H, *J*~3,4~ = *J*~4,5~ = 4.1 Hz, H-4), 3.96 (dd, 1H, *J*~5,6b~ = 7.4 Hz, *J*~6a,6b~ = 11.2 Hz, H-6a), 3.91 (m, 2H, H-3, H-6b), 3.21 (dd, 1H, *J*~1,5~ = 8.2 Hz, H-1), 2.94 (m, 1H, H-1′a), 2.73 (m, 1H, H-1′b), 2.44 (dddd, 1H, H-5), 2.04 (s, 3H, NHCO[CH~3~]{.ul}), 1.58 (m, 2H, H-2′), 1.47--1.33 (m, 6H, H-3′, H-4′, H-5′), 0.97 (t, 3H, H-6′). ^13^C-NMR (125.9 MHz, CD~3~OD): δ = 173.3 (NH[CO]{.ul}CH~3~), 78.9 (C-3), 74.0 (C-4), 64.6 (C-1), 63.0 (C-2), 59.1 (C-6), 47.9 (C-1′), 45.6 (C-5), 32.7, 29.6, 27.8, 23.6 (C-2′, C-3′, C-4′, C-5′) 22.8 (NHCO[CH~3~]{.ul}), 14.3 (C-6′). MS (MALDI): Calcd for \[C~14~H~28~N~2~O~4~H\]: *m/z* 289.2127 \[M + H\]^+^; Found \[M + H\]^+^ 289.2126. 3.9. (1S,2R,3S,4R,5R)-N-(Methoxycarbonyl)pentyl-3-acetamido-4-amino-5-hydroxymethyl-cyclopentanetriol or "2-Acetamido-2-deoxy-1-(methoxycarbonylhexyl)amino-β-[d]{.smallcaps}-galacto-cyclopentane" **22** {#sec3dot9-molecules-23-00708} ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Amine **19** (25.7 mg, 0.105 mmol) was dissolved in DMF (1 mL) and NaHCO~3~ (42.4 mg, 0.505 mmol) followed by methyl 6-iodohexanoate (20.8 mg, 0.505 mmol) were added. The reaction mixture was heated to 60 °C until completed consumption of the starting material was observed (tlc). The mixture was then concentrated under reduced pressure, and MeOH was added to obtain a ca. 50% solution. This was added to a mixture of methanol (5 mL) and acetic chloride (100 µL) at 0 °C, and the mixture was stirred for 15 min. Removal of solvent *in vacuo* and followed by chromatography of the residue (chloroform/methanol/NH~4~OH (25%) 8:1:0.01 *v*/*v*/*v*) gave free methyl ester **22** (23.8 mg, 71.6 µmol, 68.1% over two steps) as a colourless syrup. ${\lbrack\mathsf{\alpha}\rbrack}_{D}^{20}$: +41.0 (*c* = 0.99, MeOH); ^1^H-NMR (300 MHz, CD~3~OD): δ = 4.07 (dd, 1H, *J*~1,2~ = 5.0 Hz, *J*~2,3~ = 7.4 Hz, H-2), 4.02 (dd, 1H, *J*~3,4~ = *J*~4,5~ = 4.1 Hz, H-4), 3.97--3.80 (m, 3H, H-3, H-6), 3.66 (s, 3H, H-1″), 3.06 (dd, 1H, *J*~1,5~ = 8.1 Hz, H-1), 2.83 (m, 1H, H-1′a), 2.56 (m, 1H, H-1′b), 2.42 -- 2.28 (m, 3H, H-5, H-5′), 1.98 (s, 3H, NHCO[CH~3~]{.ul}), 1.72--1.28 (m, 6H, H-2′, H-3′, H-4′). ^13^C-NMR (75.5 MHz, CD~3~OD): δ = 175.8 ([CO]{.ul}OMe), 173.1 (NH[CO]{.ul}CH~3~), 79.1 (C-3), 74.0 (C-4), 64.5 (C-1), 63.6 (C-2), 59.1 (C-6), 52.0 (C-1″), 47.9 (C-1′), 46.0 (C-5), 34.7, 29.9, 27.7, 25.8 (C-2′, C-3′, C-4′, C-5′) 22.8 (NHCO[CH~3~]{.ul}). MS (MALDI): Calcd for \[C~15~H~28~N~2~O~6~H\]: *m/z* 333.2026 \[M + H\]^+^; Found \[M + H\]^+^ 333.2024. 3.10. (1S,2R,3S,4R,5R)-N-(6-Amino)hexyl-3-acetamido-4-amino-5-hydroxymethyl-cyclopentanetriol or "2-Acetamido-2-deoxy-1-(6-aminohexyl)amino-β-[d]{.smallcaps}-galacto-cyclopentane" **24** {#sec3dot10-molecules-23-00708} ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Amine **19** (68.2 mg, 0.279 mmol) was dissolved in DMF (3 mL) and treated with 6-bromohexanoic nitrile (44.4 µL, 0.335 mmol) in the presence of NaHCO~3~ (93.8 mg, 1.34 mmol). The reaction mixture was heated to 60 °C until completed consumption of the starting material was observed. The solvents were removed under reduced pressure. The residue was diluted with methanol, 2 M HCl (100 µL) was added, and the mixture was stirred for one hour. After evaporation of the solvents, the residue was purified by column chromatography (chloroform/methanol/NH~4~OH (25%) 8:1:0.01 *v*/*v*/*v*) to yield nitrile **23** (30.9 mg, 0.102 mmol, 37.0% from compound **19**), which was directly used in the next step. ${\lbrack\mathsf{\alpha}\rbrack}_{D}^{20}$: +24.3 (*c* = 0.975, MeOH); ^1^H-NMR (300 MHz, CD~3~OD): δ = 4.32 (dd, 1H, *J*~1,2~ = 5.2 Hz, *J*~2,3~ = 7.2 Hz, H-2), 4.10--3.94 (m, 4H, H-3, H-4, H-6a, H-6b), 3.64 (dd, 1H, *J*~1,5~ = 7.8 Hz, H-1), 3.23 (m, 2H, H-1′), 2.61 (m, 1H, H-5), 2.51 (t, 2H, H-5′), 2.05 (s, 3H, NHCO[CH~3~]{.ul}), 1.84--1.45 (m, 6H, H-2′, H-3′, H-4′). ^13^C-NMR (75.5 MHz, CDCl~3~): δ = 174.1 (NH[CO]{.ul}CH~3~), 121.0 (CN), 77.8 (C-3), 73.8 (C-4), 64.5 (C-1), 60.3 (C-2), 58.5 (C-6), 48.0 (C-1′), 44.5 (C-5), 26.6, 26.0, 25.9, (C-2′, C-3′, C-4′) 22.8 (NHCO[CH~3~]{.ul}), 17.2 (C-5′). A 10% solution of nitrile **23** (30.9 mg, 0.102 mmol) in methanol was stirred with small amounts of Raney-Ni under an atmosphere of H~2~ at ambient temperature. After full conversion of the starting material (20 min), the catalyst was filtered off, and the filtrate was concentrated under reduced pressure. Chromatographic purification (chloroform/methanol/NH~4~OH (25%) 8:4:1 *v*/*v*/*v*) afforded amine **24** as pale yellow syrup (30.9 mg, 0.102 mmol, 77.4%). Treatment with HCl~g~ provided the corresponding dihydrochloride **24·**HCl as a white solid. ${\lbrack\mathsf{\alpha}\rbrack}_{D}^{20}$: +38.5 (*c* = 1.115, H~2~O) (hydrochloride); ^1^H-NMR (300 MHz, CD~3~OD) (free base): δ = 4.14--4.03 (m, 2H, H-2, H-4), 3.86 (dd, 1H, *J*~2,3~ = 8.5 Hz, *J*~3,4~ = 4.2 Hz, H-3), 3.81 (d, 2H, H-6), 3.08 (dd, 1H, *J*~1,2~ = 5.6 Hz, *J*~1,5~ = 8.9 Hz, H-1), 2.76--2.56 (m, 3H, H-6′a, H-6′b, H-1′a), 2.47--2.34 (m, 2H, H-5, H-1′b), 2.00 (s, 3H, NHCO[CH~3~]{.ul}), 1.57--1.21 (m, 8H, H-2′, H-3′, H-4′, H-5′). ^13^C-NMR (75.5 MHz, D~2~O) (free base): δ = 173.2 (NH[CO]{.ul}CH~3~), 77.1 (C-3), 72.0 (C-4), 61.5 (C-2), 60.8 (C-1), 57.4 (C-6), 46.5 (C-1′), 43.8 (C-5), 40.2, 30.3, 28.0, 26.2, 25.7 (C-2′, C-3′, C-4′, C-5′, C-6′) 22.2 (NHCO[CH~3~]{.ul}). MS (MALDI): Calcd for \[C~14~H~29~N~3~O~4~H\]: *m/z* 304.2236 \[M + H\]^+^; Found \[M + H\]^+^ 304.2234. 3.11. (1S,2R,3S,4R,5R)-N-(6-Dansylamino)hexyl-3-acetamido-4-amino-5-hydroxymethyl-cyclopentanetriol or "2-Acetamido-2-deoxy-1-(6-dansylaminohexyl)amino-β-[d]{.smallcaps}-galacto-cyclopentane" **25** {#sec3dot11-molecules-23-00708} ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ A solution of amine **24** (23.2 mg, 61.6 µmol) in methanol (1mL) was treated with Et~3~N (38.5 µL, 277 mmol) and dansyl chloride (18.3 mg, 67.8 µmol). After completed conversion of the starting material (30 min), the solvent was removed under reduced pressure. Purification on silica gel (chloroform/methanol/NH~4~OH (25%) 8:1:0.01 *v*/*v*/*v*) provided compound **25** (16.1 mg, 48.6 µmol, 78.9%) as light yellow, fluorescent syrup. ${\lbrack\mathsf{\alpha}\rbrack}_{D}^{20}$: +23.7 (*c* = 0.960, MeOH); ^1^H-NMR (300 MHz, CD~3~OD): δ = 4.07 (dd, 1H, *J*~1,2~ = 4.9 Hz, *J*~2,3~ = 7.6 Hz, H-2), 4.01 (dd, 1H, *J*~3,4~ = *J*~4,5~ = 4.0 Hz, H-4), 3.93--3.79 (m, 3H, H-3, H-6a, H-6b), 3.09 (dd, 1H, *J*~1,5~ = 8.1 Hz, H-1), 2.90--2.71 (m, 3H, H-1'a, H-6'a, H-6'b), 2.53 (m, 1H, H-1'b), 2.35 (ddd, 1H, *J*~5,6a~ = *J*~5,6b~ = 7.2 Hz), 1.97 (s, 3H, NHCO[CH~3~]{.ul}), 1.40--1.05 (m, 8H, H-2′, H-3′, H-4′, H-5′). ^13^C-NMR (75.5 MHz, CD~3~OD): δ = 173.2 (NH[CO]{.ul}CH~3~), 78.9 (C-3), 74.0 (C-4), 64.6 (C-1), 63.1 (C-2), 59.0 (C-6), 47.9 (C-1′), 45.6 (C-5), 43.7, 30.4, 29.5, 27.5, 27.2 (C-2′, C-3′, C-4′, C-5′, C-6′) 22.8 (NHCO[CH~3~]{.ul}). MS (MALDI): Calcd for \[C~26~H~40~N~4~O~6~SH\]: *m/z* 537.2747 \[M+H\]^+^; Found \[M+H\]^+^ 537.2750. P.W. is the recipient of a DOC stipend (2017--2020) by the Austrian Academy of Sciences, Vienna. S.G.W. thanks GlycoNet, the Canadian Network of Centres of Excellence in Glycoscience, for financial support. **Sample Availability:** Samples of compounds are available from the authors. NMR spectra of compounds are available online. ###### Click here for additional data file. P.W. performed the syntheses; A.T. and P.M. provided crystal structures; S.A.N., B.M.P., and M.T. performed biochemical and medicinal experiments; A.E.S. and T.M.W. conceived and designed the synthetic experiments; E.P., W.W., and S.G.W. supervised and evaluated biochemical and medicinal studies. The authors declare no conflict of interest. Figures, Schemes and Table ========================== ![Examples of established as well as recently reported *N*-acetyl-[d]{.smallcaps}-hexosaminidase inhibitors.](molecules-23-00708-g001){#molecules-23-00708-f001} {#molecules-23-00708-sch001} {#molecules-23-00708-f002} ![Conversion of crucial intermediate **19** into free inhibitor **20** and *N*-alkyl derivatives thereof. f: (1) resp. halogenoalkane, NaHCO~3~, DMF, (2) HCl/MeOH; g: Raney-Ni, H~2~, MeOH; h: dansyl chloride, NaHCO~3~, MeOH; i: HCl/MeOH. (Structures show the numbering system applied for NMR-analysis for easier comparison with other [d]{.smallcaps}-galactosaminide related compounds).](molecules-23-00708-sch002){#molecules-23-00708-sch002} molecules-23-00708-t001_Table 1 ###### Inhibitory activities of new compounds with *N*-acetylhexosaminidases. --------------- ---------------------------------- ---------------------------------- ---------------------------------- ---------------------------------- ---------------------------------- **Enzyme**      SpHex 0.0007 0.0006 0.0007 0.0007 0.00006 HexA (h.lys.) n.d. 0.88 n.d. n.d. 0.30 --------------- ---------------------------------- ---------------------------------- ---------------------------------- ---------------------------------- ---------------------------------- ^1^*K*~i~-values \[μM\] of compounds with: SpHex = *Streptomyces plicatus N*-acetyl-β-[d]{.smallcaps}-hexosaminidase; HexA h. lys. = IC~50~ \[μM\] with human lysosomal *N*-acetyl-β-[d]{.smallcaps}-hexosaminidase A.; n.d., not determined.

Approach and avoidance behavior constitutes a core component of animal decision-making, where, in mammals, emotionality shapes the animals' behavioral tendencies to approach or avoid by modulating appetitive-aversive motivation[@b1][@b2][@b3][@b4]. Fear or anxiety, for example, elicits a set of defensive behavioral responses toward an imminent or anticipatory aversive stimulus, respectively[@b5]. Recent studies utilizing optogenetic tools have identified various important brain regions and circuitries underlying fear/anxiety responses[@b5]. The amygdala is thought to be central to the regulation of fear/anxiety, and activation of basolateral amygdala (BLA) projections to the centromedial amygdala (CeA) elicits anxiolytic responses, whereas somatic activation of BLA neurons is anxiogenic[@b6]. Nevertheless, how the regulatory machineries converge to effectuate emotional modulation of behavior remains largely unknown. Molecular mechanisms underlying anxiety regulation are much less well understood, and currently available anxiolytic medications produce their effects through pharmacological mechanisms of action yet to be fully understood[@b7]. In order to better understand mammalian anxiety, addressing its regulation from an alternative perspective is of importance. In light of this, recent findings on the functional relationship between the circadian clock and emotionality-related behaviors could provide an important clue. The circadian clock is an organism's internal pacemaker system with an intrinsic period of *circa* 24 hours, where the "master" clock in the hypothalamic suprachiasmatic nucleus (SCN) receives input from retinal photoreceptors and accordingly synchronizes peripheral clocks distributed throughout the body, driving diverse physiological phenomena. Dysfunctions of the circadian clock such as those arising from shift work or jet lag have been linked to a variety of mood disorders[@b8]. Conversely, abnormalities in the circadian rhythmicity of various physiological measures have been observed in patients diagnosed with major mood/anxiety disorders[@b9][@b10]. In rodents, perturbations of the circadian clock by means of surgical, genetic, pharmacological, light-induced, or behavioral manipulations lead to a spectrum of abnormalities in emotionality-related behaviors, including elevated or attenuated anxiety-like behaviors[@b11]. Recent evidence points to a mechanism by which dysfunctions in the circadian clockwork lead to abnormal emotionality through aberrant dopaminergic activity in the ventral tegmental area (VTA), a major dopaminergic nucleus[@b12][@b13]. Despite the established roles of dopamine and other monoamine systems in anxiety regulation, their causality in mood/anxiety disorders and sufficiency in the regulation of emotionality *per se* have been questioned[@b14][@b15][@b16][@b17]. Furthermore, while these studies provide important insights into affective abnormalities arising from clock dysfunction, much remains unknown as to how the circadian clock maintains emotionality-related behaviors at physiological levels. In humans, both positive and negative affect are reported to display diurnal variation[@b18], whereas excessive diurnal variations in mood states are a hallmark of major depressive and bipolar disorders[@b19], implicating the physiological importance of precise time-of-day-dependent regulation of emotionality. In the present study, we sought to unravel the mechanisms governing mammalian anxiety regulation and characterized temporal regulation of mouse anxiety-like behaviors by the circadian clock. We examined the involvement of SCOP (SCN circadian oscillatory protein), a signaling molecule originally identified as a gene product whose expression oscillates in a circadian manner in the rat SCN[@b20]. SCOP is a 183-kDa protein comprising pleckstrin homology (PH), leucine-rich repeat, protein-phosphatase 2C-like, glutamine-rich, and PDZ-binding domains, and SCOP has been shown to regulate a range of intracellular signaling pathways[@b21][@b22][@b23]. In the mouse hippocampus, SCOP plays an essential role in the consolidation of long-term object recognition memory[@b24]. Here, we describe SCOP-mediated time-of-day-dependent regulation of anxiety-like behaviors. Results ======= Anxiety-like behaviors in wild-type mice are under circadian regulation ----------------------------------------------------------------------- In order to examine the temporal regulation of anxiety by the circadian clock, we profiled time-of-day-dependent variations in anxiety-like behaviors of wild-type (WT) mice. To evaluate mouse anxiety-like behaviors, we utilized the elevated plus maze (EPM) and open field (OF) tests. These paradigms are based on rodents' intrinsic conflict between the drive to explore novel environments and the tendency to avoid open space; thus, increased time spent in the open arms of the EPM or in the center area of the OF is thought to represent reduced anxiety[@b25][@b26]. One group of mice ("LD" group, *n* = 24) were tested under a light-dark (LD) cycle at one of four or eight zeitgeber times (ZTs; lights on at ZT0, off at ZT12). The other group ("dLL" group, *n* = 76) were housed under a constant dim light condition (dLL) for \>24 hrs and tested on the second day under dLL at one of four or eight projected circadian times (CTs; subjective day starts at CT0 and ends at CT12) ([Fig. 1a](#f1){ref-type="fig"}). This enables us to eliminate the effects of cyclic light conditions. Testing CTs were projected using a circadian period (*τ*) of 25.0 hrs, as our mouse strains consistently exhibited free-running activity rhythms with a *τ* of 25.05 ± 0.07 hr SEM (*n* = 6). All mice were handled daily for \>1 week prior to behavioral assays for acclimation. Housing under dLL had no observable effects on sleep/wake cycles or general activity rhythms. In both the EPM and OF tests, anxiety-like behaviors exhibited diurnal variation and circadian rhythmicity under LD and dLL, respectively ([Fig. 1b--f](#f1){ref-type="fig"}). Anxiety-like behaviors in the EPM and OF tests (hereafter referred to as "EPM-anxiety-like" and "OF-anxiety-like" behaviors, respectively) showed rhythms almost anti-phasic to each other: EPM-anxiety-like behavior was high at early subjective night (active phase) and low at early subjective day (resting phase), while OF-anxiety-like behavior was high at early subjective day and low at early subjective night ([Fig. 1b,c,e](#f1){ref-type="fig"}). These results support the notion that EPM- and OF-anxiety-like behaviors reflect distinct physiological phenomena[@b25][@b26][@b27] (See Discussion). No significant variation in the numbers of total arm entries or in general locomotor activity levels was found across the day except for OF total distance under LD ([Fig. 1d,f](#f1){ref-type="fig"}), rendering it unlikely that circadian changes in general activity levels play a major role in the circadian expression of anxiety-like behaviors. The diurnal profiles were similar between the LD and dLL groups, suggesting that mouse anxiety-like behaviors are dynamically regulated by the intrinsic circadian clock rather than by external light conditions. The circadian clock in the dorsal telencephalon drives rhythmic anxiety-like behaviors -------------------------------------------------------------------------------------- To examine the regulation of anxiety-like behaviors by the circadian clock, we used *Bmal1*^*fl/fl*^ *Emx1*^*Cre/*+^ conditional knockout (cKO) mice. *Emx1*^*Cre*^ expression is restricted to glutamatergic neurons and astrocytes in the dorsal telencephalon (dTel), which includes the neocortex, hippocampus, and BLA[@b28]. *Bmal1*^*fl/fl*^ *Emx1*^*Cre/*+^ mice lack BMAL1 protein in the dTel[@b29]. BMAL1 and CLOCK, both bHLH transcription factors, heterodimerize to activate E-box-mediated transcription and hence constitute the core transcriptional-translational feedback loop of the molecular circadian clock. Cells lacking BMAL1 lose circadian rhythmicity[@b30]. This cKO enables us to eliminate the effects from clock dysfunctions in the SCN, which leads to a systemic loss of circadian rhythmicity, and in the monoamine-producing nuclei in the midbrain. We first examined the effects of *Bmal1* cKO on the circadian expression of clock genes in amygdala subnuclei: BLA, a dTel subnucleus involved in the regulation of anxiety-like behaviors[@b5][@b6], and CeA, a ventral telencephalic subnucleus. Both in the BLA and CeA of littermate WT mice, mRNA levels of *Bmal1*, *Dbp*, and *Rev-erb*α exhibited circadian variations ([Fig. 2a--c](#f2){ref-type="fig"}). In the BLA of *Bmal1* cKO mice, *Bmal1* mRNA levels were downregulated by \>3 fold, and *Dbp* and *Rev-erb*α expression was constantly low across the day ([Fig. 2a--c](#f2){ref-type="fig"}, blue). Clock gene expression in the CeA was unaffected by *Bmal1* cKO ([Fig. 2a--c](#f2){ref-type="fig"}, red), consistent with the lack of *Emx1* expression in the CeA[@b28]. *Bmal1* cKO mice exhibited normal activity rhythms ([Supplementary Fig. 1](#S1){ref-type="supplementary-material"}) and appeared physically normal. Whereas littermate WT mice (*Bmal1*^*fl/fl*^ *Emx1*^+*/*+^) reproduced the circadian variations in anxiety-like behaviors of WT mice ([Fig. 1](#f1){ref-type="fig"}) at CT2 (early subjective day) and CT14 (early subjective night), anxiety-like behaviors remained high in *Bmal1* cKO mice both during the day and night at levels comparable to the peak levels in littermate WT mice ([Fig. 2d--h](#f2){ref-type="fig"}). Circadian variations in anxiety-like behaviors between CT2 and CT14 were not observed in *Bmal1* cKO mice ([Fig. 2d,e,g](#f2){ref-type="fig"}). *Bmal1* cKO had no significant effect on general locomotor activities ([Fig. 2f,h](#f2){ref-type="fig"}). SCOP is expressed in a circadian manner in the amygdala ------------------------------------------------------- Given the involvement of the circadian clock in the temporal regulation of anxiety-like behaviors, we examined the roles of SCOP, a clock-controlled signaling protein. We first examined the expression of SCOP protein in dTel subregions. The prefrontal cortex, hippocampus, and BLA appeared to express SCOP in a circadian manner with higher expression at night ([Supplementary Fig. 2](#S1){ref-type="supplementary-material"}). We then narrowed our focus on the BLA, a region known to play major roles in anxiety regulation[@b5][@b6], and more fully profiled SCOP expression. Both mRNA and protein levels of SCOP exhibited circadian variations in the BLA, with mRNA peaking at CT8 and protein peaking at CT14 ([Fig. 3a--c](#f3){ref-type="fig"}, blue). SCOP was also rhythmically expressed in the CeA, exhibiting a profile anti-phasic to that in the BLA ([Fig. 3a--c](#f3){ref-type="fig"}, red), consistent with an earlier study reporting anti-phasic diurnal expression of the clock component PERIOD2 between the BLA and CeA[@b31]. In the BLA of *Bmal1* cKO mice, *Scop* mRNA levels were constantly high across the day with no significant circadian rhythmicity ([Fig. 3d](#f3){ref-type="fig"}). Combined with the constantly low expression of E-box-regulated clock genes *Dbp* and *Rev-erb*α across the day in the BLA of *Bmal1* cKO mice ([Fig. 2g,h](#f2){ref-type="fig"}), these results suggest indirect *Bmal1-*mediated transcriptional repression of *Scop* presumably through REV-ERB-mediated repression at the REV-ERB binding sites found in the intron 1 of *Scop* gene[@b32], a genomic region conserved among placental mammals. SCOP in the dorsal telencephalon is required for circadian expression of anxiety-like behaviors ----------------------------------------------------------------------------------------------- To examine the involvement of SCOP in anxiety regulation, we used dTel-specific *Scop* cKO mice (*Scop*^*fl/fl*^ *Emx1*^*Cre/*+^)[@b29]. In the BLA, *Scop* mRNA levels were undetectable both at CT8 and CT20 in *Scop* cKO mice ([Fig. 4a](#f4){ref-type="fig"}, blue), whereas *Scop* expression in the CeA was not significantly affected ([Fig. 4a](#f4){ref-type="fig"}, red). *Scop* cKO mice exhibited no observable defects in circadian activity rhythms or sleep/wake cycles[@b29] or in general locomotor activity ([Fig. 4d,f](#f4){ref-type="fig"}). Circadian expression of *Bmal1* was intact in the BLA and CeA of *Scop* cKO mice ([Supplementary Fig. 3](#S1){ref-type="supplementary-material"}). In both the EPM and OF tests, *Scop* cKO mice failed to express circadian changes in anxiety-like behaviors: While littermate WT mice (*Scop*^*fl/fl*^ *Emx1*^+*/*+^) reproduced the circadian variations in anxiety-like behaviors of WT mice ([Fig. 1](#f1){ref-type="fig"}) between CT2 (day) and CT14 (night), anxiety-like behaviors remained constantly low at both CTs in *Scop* cKO mice ([Fig. 4b--f](#f4){ref-type="fig"}), a phenotype distinct from that of *Bmal1* cKO mice ([Fig. 2](#f2){ref-type="fig"}). This is in line with the elevated *Scop* expression observed in the BLA of *Bmal1* cKO mice ([Fig. 3d](#f3){ref-type="fig"}), leading us to hypothesize that changes in SCOP levels in the BLA have direct effects on anxiety-like behaviors. SCOP in the BLA has an anxiogenic function and is essential for circadian expression of anxiety-like behavior in the EPM test ----------------------------------------------------------------------------------------------------------------------------- To investigate the function of SCOP in the BLA, we examined the effects of *Scop* knockout (KO) in the BLA. To this end, we constructed adeno-associated virus (AAV) expressing Cre recombinase fused to EGFP ("AAV-Cre") or EGFP alone ("AAV-GFP") driven under human synapsin (hSyn) promoter ([Fig. 5a](#f5){ref-type="fig"}). *In vitro* analyses using cultured cells confirmed the recombinase activity of the fusion protein, Cre expression in neuronal cells, and the infectious ability of the viral constructs in neurons ([Supplementary Fig. 4](#S1){ref-type="supplementary-material"}). We then injected AAV-Cre or AAV-GFP bilaterally into the BLA of *Scop*^*fl/fl*^ mice (*Scop* BLA KO) and examined the effects of BLA-specific *Scop* KO on anxiety-like behaviors ([Fig. 5b](#f5){ref-type="fig"}). Transduction with AAV-Cre resulted in a \>3 fold reduction in *Scop* mRNA levels in the BLA of *Scop*^*fl/fl*^ mice compared to the BLA of *Scop*^*fl/fl*^ mice transduced with AAV-GFP ([Fig. 5c--e](#f5){ref-type="fig"}). Behaviorally, -BLA-specific *Scop* KO mice phenocopied *Scop* cKO mice in the EPM test, exhibiting lower anxiety-like behavior both at CT2 and CT14 ([Fig. 5f--h](#f5){ref-type="fig"}). BLA-specific *Scop* KO abrogated the circadian variation in open arm entries in the EPM test between the CTs ([Fig. 5f,g](#f5){ref-type="fig"}). The effects of BLA-specific *Scop* KO on open arm entries at CT14 were significant compared to mice bilaterally transduced with AAV-GFP. The total numbers of entries were unaffected ([Fig. 5h](#f5){ref-type="fig"}). AAV-GFP-transduced mice exhibited circadian changes in open arm entries at levels comparable to that in wild-type ([Fig. 1b](#f1){ref-type="fig"}) or *Scop*^*fl/fl*^ mice ([Fig. 4b](#f4){ref-type="fig"}). BLA-specific *Scop* KO had no statistically significant effect on anxiety-like behavior in the OF test ([Fig. 5i,j](#f5){ref-type="fig"}), suggesting region-specific roles of SCOP in regulating anxiety-like behaviors in the EPM and OF tests (See Discussion). These results demonstrate that SCOP in the BLA functions to elevate anxiety-like behavior in the EPM test at early subjective night (CT14), when SCOP level in the BLA peaks. Discussion ========== Our results demonstrate that the anxiogenic function of SCOP in the BLA drives circadian rhythms in EPM-anxiety-like behavior in mice. We show that the clock machinery and SCOP in the dTel are required for the circadian expression of anxiety-like behaviors in both the EPM and OF tests ([Figs 2](#f2){ref-type="fig"} and [4](#f4){ref-type="fig"}). *Bmal1* cKO leads to upregulation of SCOP expression in the BLA ([Fig. 3d](#f3){ref-type="fig"}) and has anxiogenic effects, whereas *Scop* cKO has anxiolytic effects. Deletion of SCOP in the BLA abolishes the diurnal elevation of EPM-anxiety-like behavior at early subjective night (CT14) ([Fig. 5f,g](#f5){ref-type="fig"}), signifying the role of SCOP in the BLA in driving the anxiety-like behavior in the EPM test. Collectively, we conclude that SCOP is a regulator of mouse anxiety-like behaviors functioning in anxiogenesis, driving circadian changes in these behaviors, and that the rhythmic expression of SCOP in the BLA is required for the circadian expression of anxiety-like behavior in the EPM test. BLA-specific deletion of SCOP resulted in a loss of the circadian variation in anxiety-like behaviors in the EPM test ([Fig. 5f,g](#f5){ref-type="fig"}) but not in the OF test ([Fig. 5i,j](#f5){ref-type="fig"}), while dTel-specific deletion of SCOP abolished the rhythms in both behaviors ([Fig. 4b--f](#f4){ref-type="fig"}). These data suggest that while SCOP in the dTel is required for the circadian variation in anxiety-like behaviors in the OF test, SCOP in the BLA, which is part of the dTel, is not. It is therefore likely that a dTel subregion(s) other than the BLA is responsible for the circadian regulation of anxiety-like behaviors in the OF test. Our results also demonstrate an anti-phasic relationship in the circadian expression of SCOP in the BLA and CeA ([Fig. 3](#f3){ref-type="fig"}). Antiphasic expression of the clock gene *Period2* in these amygdala nuclei have previously been reported[@b31], and our data demonstrate that these nuclei also exhibit distinct phase distributions in the circadian expression of clock genes *Bmal1*, *Rev-erbα*, and *Dbp* ([Fig. 2a--c](#f2){ref-type="fig"}). Previous studies have reported roles of tissue-specific transcriptional co-regulators, such as CBP/p300, in regulating tissue- and cell type-specific circadian transcription and phase distributions[@b33][@b34]. We thus speculate that BLA- and CeA-specific co-activators and/or co-repressors may regulate distinct rhythms in mRNA and protein levels of SCOP and clock genes, although we have been unable to directly examine protein levels of core clock genes in the amygdala subregions potentially due to their low abundance. SCOP is a clock-controlled multi-domain signaling molecule regulating various intracellular signaling pathways[@b20][@b21][@b22][@b23][@b24][@b35]. Notably, SCOP and its shorter variant PHLPP1α have been shown to dephosphorylate AKT and thereby suppress PI3K/AKT/GSK3β signaling, where signaling cascades downstream of dopamine D~2~, 5-HT~1A~ and 5-HT~2~ receptors converge[@b36]. In humans and mice, AKT1, AKT2, and GSK3β have been associated with abnormal anxiety[@b37][@b38][@b39]. These data point to a model in which SCOP regulates anxiety by modulating AKT/GSK3β signaling. In addition, SCOP has been shown to regulate K-Ras/ERK signaling in the hippocampus[@b24][@b40]. MAPK/ERK signaling in the amygdala has been repeatedly shown to contribute to anxiety regulation in rodents. For example, exposure to novel environments or repeated intraperitoneal injection of corticosterone, which enhances anxiety-like behavior in the EPM test, increases ERK activation in the rat amygdala[@b41][@b42], whereas ERK inhibition in the mouse BLA abolishes restraint stress-induced increase in anxiety-like behavior in the EPM test[@b43]. Given the crosstalk reported for MAPK/ERK and AKT/GSK3β signaling pathways, it is therefore of interest to examine how SCOP regulates these anxiety-related signaling pathways in the BLA *in vivo*. Importantly, the anxiogenic and anxiolytic effects of *Bmal1* and *Scop* cKO, respectively, did not exceed the range of the circadian variations of anxiety-like behaviors expressed by wild-type mice ([Figs 2](#f2){ref-type="fig"} and [4](#f4){ref-type="fig"}). Put another way, the loss of circadian rhythmicity in the dTel appears to "halt" the circadian expression of anxiety-like behaviors at a specific time of day. We therefore speculate that the circadian clock in the dTel confers circadian rhythmicity on mouse anxiety-like behaviors. This, together with our finding that the anxiety-like behaviors in the EPM and OF tests peak at distinct times of day ([Fig. 1b,c,e](#f1){ref-type="fig"}), suggests that the rhythmic expression of anxiety-like behaviors may play an important survival function. In this regard, examining whether and how this circadian expression is conserved in other species, especially in diurnal animals, will be of particular interest. It has to be noted that the distinct temporal profiles in the expression of anxiety-like behaviors in the EPM and OF tests are not an unexpected finding, as these behaviors have been frequently suggested to reflect distinct physiological phenomena. For example, factor analysis studies using rodents revealed that anxiety-related behaviors evaluated in the EPM and OF tests do not load on a common factor[@b44][@b45]. Consistently, multiple pharmacological manipulations have been shown to produce test-selective effects on these anxiety-like behaviors[@b46]. Hence, a current model considers the anxiety-like behaviors evaluated in the EPM and OF tests, as well as in other anxiety-related tests, as representing partially overlapping but distinct aspects of the multidimensional "anxiety", likely underlain by partially overlapping mechanisms[@b26]. Not only are our results ([Fig. 1b,c,e](#f1){ref-type="fig"}) compatible with such a model, our finding that clock dysfunction in the dTel abolishes the distinct phase relationship between the EPM- and OF-anxiety-like behaviors ([Fig. 2](#f2){ref-type="fig"}) suggests that the circadian clock may be actively involved in the differentiation of these anxiety-like behaviors. Although further study is necessary to elucidate the underlying mechanisms, these results, combined with the selective effects of *Scop* BLA KO on the EPM but not the OF test ([Fig. 5f--j](#f5){ref-type="fig"}), could provide a foundation for unraveling the complexity of anxiety-like behaviors. In summary, we have described the circadian expression of mouse anxiety-like behaviors, which requires clock machineries and SCOP function in the dTel, and demonstrated that SCOP in the BLA is expressed in a circadian manner and exerts anxiogenic effects on the EPM. While the molecular mechanisms remain to be established, we conclude that circadian regulation of SCOP levels in the dTel plays an essential role in generating the circadian rhythmicity in anxiety-like behaviors, which could represent an important function in animal survival. By characterizing the involvement of a temporal axis in the regulation of mammalian anxiety-like behavior, as well as by identifying key molecular and neuroanatomical players therein, our present study provides a new foundation for future research on animal emotionality. Further understanding of the intrinsic plasticity in anxiety regulation described here could potentially enable the development of new classes of treatment for anxiety/mood disorders. Methods ======= Animals and housing ------------------- All experiments were conducted in accordance with guidelines set by The University of Tokyo and approved by the Committee on Animal Care and Use of the Graduate School of Science at The University of Tokyo. Wild-type male C57BL/6J mice were purchased from Tokyo Laboratory Animals Science Co., Ltd (Tokyo, Japan). *Emx1*^*Cre/*+* *^[@b28] mice were kindly provided by Dr. Atsu Aiba. All animals were initially housed under a 12 hr light/12 hr dark cycle (lights on at 8 am) in temperature- and humidity-controlled compartments with food and water available *ad libitum*. More information is available in [Supplementary Information](#S1){ref-type="supplementary-material"}. Surgery ------- Male *Scop*^*fl/fl*^ mice aged 8--10 weeks were deeply anesthetized with a mixture of ketamine (140 mg/kg) and xylazine (8.8 mg/kg) in bacteriostatic saline given intraperitoneally (20 ml/kg) and placed on a stereotactic apparatus (Narishige, Tokyo, Japan). The skull was exposed, and holes were drilled bilaterally above the basolateral amygdala. The coordinates relative to bregma were: anteroposterior, −1.65 mm; lateral, +3.30 mm; dorsoventral, −4.45 mm. Mice were bilaterally injected with 0.5 μl of either AAV-Cre or AAV-GFP over 5 min, and the needles were kept in place for an additional 5 min to ensure infusion. Behavioral assays ----------------- Male mice aged 12--16 weeks (wild-type and AAV-injected *Scop*^*fl/fl*^) or 20--25 weeks (*Bmal1* and *Scop* cKO) were subjected to behavioral tests. Littermate *Bmal1*^*fl/fl*^ *Emx1*^+*/*+^[@b46] or *Scop*^*fl/fl*^ *Emx1*^+*/*+^ mice were used for control. Prior to testing, all mice were singly housed, entrained to the LD cycle for \>2 weeks, and handled daily for acclimation at random times of day for \>1 week. All behavioral assays were conducted under dim light at 4.0 ± 0.1 lux at the center of the apparatuses. Mice were picked up on the operator's palms and released into the apparatuses such that the mice voluntarily walk into the maze or open field from the palms. For BLA-specific KO experiments, behavioral data from mice with clear bilateral GFP signal in the basolateral amygdalar complex were adopted as AAV-Cre or AAV-GFP data. For fluorescent microscopy, AAV-injected mice were sacrificed by rapid cervical dislocation, and the brain was removed, mounted on a brain matrix (ASI Instruments), and sliced into 1-mm thick sections. The sections were analyzed for GFP signal under a fluorescent stereoscopic microscope (Leica). Detailed procedures are described in [Supplementary Information](#S1){ref-type="supplementary-material"}. Elevated Plus Maze ------------------ Mice were placed in the center of an elevated plus maze (O'Hara & Co., Ltd., Tokyo, Japan) facing one of the closed arms. The maze has four arms (5 × 25 cm), the opposing two of which are protected with clear walls (16 cm high), and is elevated 50 cm from the ground. Mice were allowed to freely explore the maze for 5 min; their behavior was monitored with an automated video tracking system, and the time spent on open arms and entries into open/closed arms were determined using TIME_EP software (O'Hara & Co., Ltd.) or manually by an analyst blinded to time-of-day information (see Data analysis). Open Field ---------- Mice were placed into one of the corners of a 45 × 45 cm open field (O'Hara & Co., Ltd.), and their behavior was monitored for 5 min with an automated video tracking system. The time spent in the center of the field (30% area, circular) and the distance traveled were determined using ImageJ software (NIH, MD) with OpenField plug-in (O'Hara & Co., Ltd.). Data analysis ------------- Behavioral analyses were automated except for the elevated plus maze (EPM) test for wild-type C57BL/6J mice, for which all recorded behavioral data were shuffled into random orders, and entries into open and closed arms and the time spent on each arm were recorded by an operator blinded to time-of-day information for each subject. One-way ANOVA tests were used to examine the statistical significance of data consisting of three or more groups. Unpaired two-tail Student's *t*-tests were used for analysis on data with two groups. All data are presented as means with SEM. *P* values are presented as "*P*" for ANOVA tests and "*p*" for *t*-tests. Additional Information ====================== **How to cite this article**: Nakano, J. J. *et al*. SCOP/PHLPP1β in the basolateral amygdala regulates circadian expression of mouse anxiety-like behavior. *Sci. Rep.* **6**, 33500; doi: 10.1038/srep33500 (2016). Supplementary Material {#S1} ====================== ###### Supplementary Information This work was supported by KAKENHI Grants-in-Aid for Scientific Research (Grant numbers 24227001 to Y.F. and 15K12767 to K.S.) from Ministry of Education, Culture, Sports, Science and Technology, Japan. We thank Drs Atsu Aiba, Takahiko Matsuda, and Andras Nagy for the generous gifts of *Emx1*^*Cre/*+^ mice, pCAG-iCre plasmid, and pCCALL2 plasmid, respectively. **Author Contributions** J.J.N., K.S. and Y.F. designed the research, analyzed data, and wrote the main manuscript. J.J.N. performed the research. S.S. provided *Bmal1*^*fl/fl*^ mouse. All authors reviewed the manuscript. {#f1} {#f2} {#f3} {#f4} {ref-type="supplementary-material"}). hSyn: human synapsin promoter; WPRE: woodchuck hepatitis virus posttranscriptional regulatory element; pA: poly-A; ITR: inverted terminal repeat. (**b**) Timeline for the viral injection, behavioral tests (EPM and OF) and histological inspection. (**c**) *Scop* and *EGFP* mRNA levels in the BLA of *Scop*^*fl/fl*^ mice injected with AAV-GFP (open bars) or AAV-Cre (filled bars). (**d**) Histological reconstruction of the injection sites in coronal sections of the mouse brain. Values are distances in millimeters from Bregma (minus indicates a position posterior to Bregma) according to the mouse stereotaxic atlas by Paxinos and Watson[@b47]. AAV-Cre (magenta) or AAV-GFP (blue) was injected bilaterally to the BLA (shown in cyan) for each hemisphere of an individual mouse. Each colored circle represents the site of the strongest GFP signal. (**e**) A representative image of the infection are in the BLA. (**f--j**) Anxiety-like behavior in BLA-specific *Scop* KO mice. EPM % Open entries (**f**), time spent on open arms (**g**), total numbers of entries into closed and open arms (**h**), OF center time (**i**), and OF Total distance (**j**) at CT2 (shaded bars) and CT14 (filled bars) for mice bilaterally infected with AAV-Cre ("Cre") and mice bilaterally infected with AAV-GFP ("GFP"). \**p* \< 0.05 by unpaired Student's *t*-test. *n* for behavioral assays: Cre, 6 (CT2) and 8 (CT14) for EPM, 8 (CT2) and 6 (CT14) for OF; GFP, 6 (CT2) and 6 (CT14). Data are means with SEM. Scale bars, 500 μm.](srep33500-f5){#f5}

Introduction ============ Gastric cancer (GC) is the fourth most common cancer and the second leading cause of cancer-related death in the world [@B1]. Although GC mostly occurs in the 50-70 age groups and the overall incidence of GC has declined in last decade, the incidence rate in younger patients has increased [@B2], [@B3]. Compared to the older age group, the clinicopathological features and prognosis of young GC patients remain controversial. Some studies have revealed that the prognosis of GC is worse in younger patients due to delayed diagnosis or more advanced stage [@B4]-[@B6]. However, other studies state that the prognosis in the young patients is comparable to the older patients [@B7]-[@B9], or even better[@B10], [@B11]. The controversy findings may result from many reasons, including sample size, ethnicity, and different definitions of young age group. Most literatures defined young age as below 40 or 45 years[@B5]-[@B7], [@B9], [@B12]-[@B19], while others used 30, 50, 60 or even 70 [@B4], [@B8], [@B10], [@B11], [@B20]-[@B23]. The difference of cut-off value for ages may lead to different outcomes of analysis. Besides, in clinical practice, we found that the proportion of extremely young patients (20-30 years old) was increasing, and prognosis of this age group was poor. To investigate the association between age and prognosis, we reviewed the epidemiologic and clinicopathological characteristics of GC patients based on Surveillance, Epidemiology, and End Results (SEER) population-based data. Methods ======= Database -------- The SEER database is the largest publicly available cancer dataset. It is a population-based cancer registry across several disparate geographic regions. The SEER research data include cancer incidence and prevalence as well as demographic information tabulated by age, sex, race / ethnicity, year of diagnosis and Tumor-Node-Metastasis (TNM) stage. The exact dataset we used for this analysis was SEER Program ([www.seer.cancer.gov](http://www.seer.cancer.gov)) Research Data (1973-2014), National Cancer Institute, DCCPS, Surveillance Research Program, Surveillance Systems Branch, based on the November 2016 submission. Another cohort from the Sun Yat-sen University Cancer Center (SYSUCC) was used as external validation data. It included all the gastric adenocarcinoma cancer patients who received therapy and had full record of follow-up in SYSUCC during 2000 and 2013 (Supplementary table [1](#SM0){ref-type="supplementary-material"}). The study protocol for the Chinese cohort was approved by the independent Ethics Committees at SYSUCC. Outcome variables ----------------- Variable definitions information on age at diagnosis, sex, year of diagnosis, race/ethnicity, primary site, tumor grade and differentiation, histology, lymph node involvement, AJCC 7^th^ TNM stage, insurance status and overall survival were coded and available in SEER database. The primary site was defined by the following International Classification of Diseases for Oncology (ICD-O-2) codes: C16.0-C16.9. Cardia, (C16.0), fundus (C16.1), body (C16.2), antrum (C16.3), pylorus (C16.4), lesser curvature (C16.5), greater curvature (C16.6), overlapping lesion (C16.7) and stomach, NOS (C16.9). Grade and differentiated was defined by the following ICD-O-2 codes; well differentiated (Code 1), moderate differentiated (Code 2), poorly differentiated (Code 3) and undifferentiated (Code 4). Histological types were defined by the following ICD-O-3 codes: 8140 to 8147, 8210 to 8211, 8220 to 8221, and 8260 to 8263 for adenocarcinoma, 8480 and 8481 for mucinous adenocarcinoma, and 8490 for Signet ring cell carcinoma. For the Race/Ethnicity, we reclassified the patients into 5 groups: "Caucasian", "African American", "Asian", "Others" and "Unknown". Patient Population ------------------ The study population was based on the SEER cancer registry. We restricted eligibility to patients with gastric adenocarcinoma (including mucinous adenocarcinoma and signet ring cell carcinoma) from 1973 to 2014 with an age≥18. We excluded cases without follow-up records (survival time code of 0 months). Statistical Methods ------------------- The patients\' demographic and tumor characteristics were summarized with descriptive statistics. Comparisons of categorical variables were performed using the Chi square test, and continuous variables were compared using Student\'s t test. The primary endpoint of this study was 5-year cause specific survival (CSS), which was calculated from the date of diagnosis to the date of cancer specific death. Deaths attributed to gastric cancer were treated as events and deaths from other causes were treated as censored observations. Survival function estimation and comparison among different variables were performed using Kaplan-Meier estimates and the log-rank test. The independence of the prognostic factors was adjusted for other known factors including age at diagnosis and tumor stage. All of statistical analyses were performed using the Intercooled Stata 13.0 (Stata Corporation, College Station, TX). Statistical significance was set at two-sided P \< 0.05. Informed consent and Protection of Human Subjects ------------------------------------------------- This study with SEER data was deemed exempt from institutional review board approval by Sun Yat-sen University Cancer Center and informed consent was waived. The study protocol for the Chinese cohort was approved by the independent Ethics Committees at SYSUCC. Results ======= Cut-off points of age groups ---------------------------- We identified 79,505 GC patients diagnosed between 1973 and 2014 with an age≥18. The median age of the whole population was 68 years old (mean±SD: 66.43±13.74). We divided the patients into 11 age groups (every 5 years of age). Since the number of patients younger than 30 was small, we combined these patients into one group. If the median overall survival between two adjacent groups were significantly different, the youngest age of the latter group was set as the cut-off value (Table [1](#T1){ref-type="table"}). Based on the cut-off points, all the patients were divided into 4 groups as follows: young age patients: 18-34 years; middle-age patients: 35-64 years; elderly patients: 65-74 years; extremely elderly patients: \>74 years. Young age patients accounted for 1.72% (1369/79505) of all GC patients. Epidemiologic and clinicopathological characteristics ----------------------------------------------------- General epidemiologic and clinicopathological characteristics among different age groups were summarized in Table [2](#T2){ref-type="table"}. The proportion of young age patients had increased faster than the other 3 groups during the last decade. The proportion of female was significantly higher in young age group, compared to the middle-age and the elderly (46.8%, 31.6% and 32.4%, respectively. P\<0.001). The ethnic composition and location of lesions were different between young age group and the other 3 groups (All P\<0.001). The diffuse type GC was much more common in the young age group (47.6%) than in the other 3 age groups (26.8%, 17.6%, 12.8%, respectively, all P \<0.001). Besides, poorly differentiated tumors were remarkably more common in the young age group (66.9%), and signet-ring cell carcinoma was more frequently seen in the young age patients (40.8%). Moreover, less patients in the young age group received surgery (45.4%) than the middle-age and the elderly groups (P\<0.001), which was similar with the extremely elderly group. More young patients were diagnosed with distant metastasis (28.1%) than the other 3 groups. Survival analysis ----------------- The median CSS of the young age group was 12 months, which was significantly lower than the middle-age and the elderly groups (15 and 16 months, respectively, both P\<0.001). The median overall survival between young age group and extremely elderly group was comparable (12 months vs. 13 months, P=0.153) (Fig [1](#F1){ref-type="fig"}). To explore the prognostic factor for young age GC patients, we carried out the univariate and multivariate analysis in young GC group (Table [3](#T3){ref-type="table"}). There were 952 deaths (69.54%) in this group with a 5-year CSS of 14.61% (95%CI: 12.80%-16.54%). Variables showing a trend for association with survival (P \< 0.05) were selected in the cox proportional hazards model. Tumor location, ethnicity, tumor size (≥5cm), surgery and TNM stage were identified as independent prognostic factors in young GC patients by both univariate and multivariate analysis. Validation using GC patients from SYSUCC ---------------------------------------- We found that patients younger than 35 years old still had the worst prognosis among the four age groups (Figure [2](#F2){ref-type="fig"}). The median survival for these 4 groups patients were as follow: 25 months for patients younger than 35 years old, 57 months for patients who were 35-64 years, 43 months for patients who were 65-74 years and 38 months for patients \>74 years old, P=0.0003. Discussion ========== The prognosis of GC in different age groups is still controversial. This may result from different definitions of young age. Most literatures defined young age as below 40 or 45 years[@B5]-[@B7], [@B9], [@B12]-[@B19], while others used 30, 50, 60 or even 70 [@B4], [@B8], [@B10], [@B11], [@B20]-[@B23]. Some studies simply used median age as the cutoff value to analyze the prognosis of age in GC patients [@B24]-[@B27]. Most of the studies come to the conclusion that young patients had a better prognosis than elderly patients with the possible explanation that young patients had better performance status and better tolerance to therapy, while elderly patients might have more complication and could not stand effective therapy [@B10]. In our clinical practice, we realized that young patients should be treated as two distinct types, pretty young patients who actually had worse prognosis than older patients and middle-age patients who had better prognosis. How can we figure out the young patients who have poor prognosis? In our study, we compared the survival of different groups GC patients according to their ages with an interval of 5 years. By combining patients with similar prognosis, we figured out 3 cutoff values, 35 years old, 65 years old and 75 years old. By this, we divided patients into 4 groups: young age patients: 18-34 years; middle-age patients: 35-64 years; elderly patients: 65-74 years; extremely elderly patients: \>74 years. Young patients had the worst prognosis, while the elderly group had the best prognosis. The median survivals for these two groups were 12months and 16months, respectively, P\<0.001. The poor prognosis for young patients remained valid in the SYSUCC database. During the last few decades, the incidence of GC has declined in most countries. However, there is a rising trend in the incidence of GC in young age patients [@B2]. Our study also showed that the proportion of young GC patients had increased in the last 4 decades. It has been suggested that GC in young patients has different clinical features and tumor behavior from older gastric cancer. Most common reported features by literatures included female dominance, location in upper third area, undifferentiated and diffuse histologic types, advanced stage at diagnosis [@B5]-[@B7], [@B9], [@B10], [@B15], [@B23], [@B28], [@B29]. In our study, the percentage of female patients in the young patient subgroup was higher than that in other age groups. The male-to-female ratio in the young age group was 1.14, in contrast to 2.16 in middle age group and 2.08 in elderly group. The reason for higher proportion of female in the young age group remains controversial. Sex hormone especially estrogen has been suggested to play a role in pathogenesis of GC [@B14], [@B30]-[@B33]. Some studies reported the protective effect of estrogen in GC [@B30], [@B32], [@B33], while others claimed the opposite [@B14], [@B31]. Chung et.al suggested that the excessive exposure of estrogen without the counter effect of progesterone may increase the risk of GC [@B14]. Further studies are needed to clarify the underlying mechanism of female dominance in young age GC patients. Regarding histologic classifications, previous studies have showed poor-differentiated or undifferentiated and diffuse-type gastric cancer predominated in young age patients [@B7], [@B9], [@B14], [@B15], [@B23], [@B34]. Our study also showed a higher proportion of diffuse type and poor differentiated gastric cancer in the young age group. Previous studies indicated that intestinal-type and well differentiated GC originates from intestinal metaplasia or atrophic gastritis, which is associated with infection by *Helicobacter pylori*. This progression from infection of *H.pylori* to pre-malignant stage and finally malignant tumor may take many years [@B35]-[@B37]. However, the diffuse-type adenocarcinoma or poor-differentiated gastric cancer may develop without passing all these stages. Young age patients have fewer years to pass through these stages, which may partially explain why diffuse-type and poor-differentiated is more common in the young age group. Moreover, it has been reported that diffuse type GC is associated with E-cadherin deregulation upon genetic alterations [@B38], and E-cadherin gene (CDH1) mutations are more common in young age GC [@B39]. According to the definition of hereditary diffuse gastric cancer, diffuse gastric cancer patients who are younger than 40 years old have met the criteria [@B40]. Hence the genetic factors may play an important role in the development of GC in young age patients. To explore the factors that contribute to the worse survival in pretty young GC patients, we carried out the univariate and multivariate analysis. We found that location of tumor, ethnicity, tumor size, surgery and TNM stage were independent risk factors in young age patients. Previous studies have suggested various clinicopathological factors, including tumor location, tumor size, depth of invasion, lymphovascular invasion, peritoneal metastasis, distant metastasis, curative resection, CEA and CA125 level, and TNM stage [@B5]-[@B7], [@B10], [@B14], [@B17]-[@B19], [@B23], [@B34], [@B41]. Curative resection was the most common risk factor seen in these studies. Several studies found that the prognosis of young age patients after curative surgery was comparable or better than the old age patients [@B5], [@B23], [@B41], [@B42], and prognosis of advanced stages was worst in young age patients than the old age patients [@B29], [@B43]. However, few young age patients present with alarm symptoms and they do not receive regular screen by gastroscopy. More patients in young age group were diagnosed at stage IV and lost the chance for curative resection. Considering genetic factors may play a role in development of early gastric cancer, it is reasonable to offer earlier screening to young people with family history. Further studies are warranted to figure out the high risk factor for gastric cancer and based on these to design the screen strategy in high risk young people. It is also important to find out more effective screen methods, such as liquid biopsy or superfine gastroscopy. There were some limitations to this study. Firstly, it was a retrospective study. Secondly, as this study overlapped long time frame and AJCC staging system for gastric cancer changed, the current study couldn\'t reflect the stages and clinical practice of gastric cancer in previous years. Thirdly, the SEER database does not include data on signs and symptoms, family history, level of tumor markers, chemotherapy, and other factors that might affect prognosis. Further studies are needed to elucidate the mechanism for poor prognosis of young age patients. Conclusion ========== This study demonstrated that gastric cancer patients \<35 years old had a poor prognosis. Location of tumor, ethnicity, tumor size, surgery and TNM stage were independent risk factors on prognosis for young patients. Early detection of gastric cancer is very important in all patients but in young patients it is of paramount importance. Supplementary Material {#SM0} ====================== ###### Supplementary table. ###### Click here for additional data file. We would like to thank the staff members of the National Cancer Institute and their colleagues across the United States and at Information Management Services, Inc., who have been involved with the Surveillance, Epidemiology and End Results (SEER) Program. Funding ======= This work was supported by National Natural Science Foundation of China (Grant number. 81602066 and 81772587); the Fundamental Research Funds for the Central Universities (Grant number. 16ykpy25); the third outstanding young talents training plan and Medical Scientist program of Sun Yat-sen University cancer center; Science and Technology Planning Project of Guangzhou, China (Grant number. 201510161726583); National Science and Technology Major Project, China (Grant number. 2016ZX09101004). {#F1} {#F2} ###### Survival analysis among different age groups Age group Age Number (%) Median CSS (Months) 5-year CSS rate (%) 95% CI (%) P value ----------- --------------- --------------- --------------------- --------------------- ------------- --------- Group 1 \<30 499 (0.63) 14 21.53 17.55-25.79 0.0872 30-34 870 (1.09) 15 25.46 22.31-28.71 0.0166 Group 2 35-39 1539 (1.94) 17 30.56 28.09-33.06 0.3868 40-44 2623 (3.30) 17 28.68 26.80-30.58 0.3833 45-49 4097 (5.15) 18 29.87 28.37-31.39 0.6340 50-54 6065 (7.63) 18 28.75 27.51-30.01 0.1731 55-59 7821 (9.84) 19 29.49 28.39-30.61 0.2634 60-64 9470 (11.91) 19 29.27 28.26-30.29 0.0086 Group 3 65-69 10887 (13.70) 19 31.75 30.80-32.72 0.2732 70-74 11212 (14.10) 20 31.37 30.43-32.32 0.0071 Group 4 75+ 24422 (30.72) 16 27.34 26.70-27.98 CSS: Cause specific survival, CI: Confidential Interval. ###### Comparison of clinic-pathologic features among different age groups ------------------------------------------------------------------------------------------------- \<35 (%)\ 35-64 (%)\ 65-74 (%)\ \>74 (%)\ P value (N=1369) (N=31615) (N=22099) (N=24422) ------------------------- ------------- --------------- --------------- --------------- --------- Sex Male 728 (53.18) 21618 (68.38) 14935 (67.58) 13169 (53.92) Female 641 (46.82) 9997 (31.62) 7164 (32.42) 11253 (46.08) \<0.001 Year 1973-1983 115 (8.40) 4295 (13.59) 3292 (14.90) 3328 (13.63) 1984-1993 175 (12.78) 4305 (13.62) 3826 (13.71) 3752 (15.36) 1994-2003 399 (29.15) 8145 (25.76) 6118 (27.68) 7208 (29.51) 2004-2014 680 (49.67) 14870 (47.03) 8863 (40.11) 10134 (41.50) \<0.001 Lauren Diffuse 652 (47.63) 8473 (26.80) 3883 (17.57) 3138 (12.85) Intestinal 668 (48.79) 21927 (69.36) 17330 (78.42) 20469 (83.81) Unknown 49 (3.58) 1215 (3.84) 886 (4.01) 815 (3.34) \<0.001 Grade Well 31 (2.26) 1162 (3.68) 1117 (5.05) 1376 (5.63) Moderate 123 (8.98) 6233 (19.72) 5363 (24.27) 6724 (27.53) Poorly 916 (66.91) 17728 (56.07) 11414 (51.65) 11202 (45.87) Undifferentiated 44 (3.21) 779 (2.46) 493 (2.23) 426 (1.74) Unknown 255 (18.63) 5713 (18.07) 3712 (16.80) 4694 (19.22) \<0.001 Ethnicity Caucasian 888 (64.86) 22027 (69.67) 15654 (70.84) 17795 (72.86) African-American 180 (13.15) 4306 (13.62) 2592 (11.73) 2404 (9.84) Asian 225 (16.44) 4399 (13.91) 3353 (15.17) 3849 (15.76) Others 76 (5.55) 883 (2.79) 500 (2.26) 374 (1.53) \<0.001 Primary site Cardia 295 (21.55) 10469 (33.11) 6611 (29.92) 5228 (21.41) Fundus 43 (3.14) 1065 (3.37) 795 (3.60) 949 (3.89) Body 133 (9.72) 2165 (6.85) 1598 (7.23) 2019 (8.27) Antrum 229 (16.73) 5556 (17.57) 4471 (20.23) 6026 (24.67) Pylorus 46 (3.36) 987 (3.12) 764 (3.46) 982 (4.02) Lesser curvature 121 (8.84) 2894 (9.15) 2225 (10.07) 2537 (10.39) Greater curvature 61 (4.46) 1234 (3.90) 906 (4.10) 1014 (4.15) Overlapping 153 (11.18) 2629 (8.32) 1728 (7.82) 1923 (7.87) NOS 288 (21.04) 4616 (14.60) 3001 (13.58) 3744 (15.33) \<0.001 Surgery Yes 720 (52.60) 19799 (62.63) 14516 (65.69) 13149 (53.84) No 621 (45.36) 11245 (35.57) 7112 (32.18) 10642 (43.58) Unknown 28 (2.05) 571 (1.81) 471 (2.13) 631 (2.58) \<0.001 Histology Adenocarcinoma 786 (57.41) 23721 (75.03) 18372 (83.13) 21403 (87.64) Signet Ring cell 558 (40.76) 7169 (22.68) 3208 (14.52) 2548 (10.43) Mucinous adenocarcinoma 25 (1.83) 725 (2.29) 519 (2.35) 471 (1.93) \<0.001 Tumor size (mm) Median 50 45 40 40 Mean±SD 158.0±300.3 91.6±197.1 71.9±156.7 65.2±138.8 \<0.001 AJCC 7^th^ T stage T1 119 (8.69) 3011 (9.52) 2194 (9.93) 2951 (12.08) T2 42 (3.07) 1225 (3.87) 863 (3.91) 967 (3.96) T3 129 (9.42) 3545 (11.21) 2160 (9.77) 2020 (8.27) T4 222 (16.22) 4404 (13.93) 2233 (10.10) 1987 (8.14) Tx 857 (62.60) 19430 (61.46) 14649 (66.29) 16497 (67.55) \<0.001 AJCC 7^th^ N stage N0 216 (15.78) 5273 (16.68) 3629 (16.42) 4851 (19.86) N1 243 (17.75) 5511 (17.43) 3114 (14.09) 2773 (11.35) N2 64 (4.67) 1659 (5.25) 845 (3.82) 708 (2.90) N3 30 (2.19) 666 (2.11) 363 (1.64) 254 (1.04) Nx 816 (59.61) 18506 (58.54) 14148 (64.02) 15836 (64.84) \<0.001 AJCC 7^th^ M stage M0 276 (20.16) 9010 (28.50) 6184 (27.98) 7512 (30.76) M1 385 (28.12) 5452 (17.24) 2403 (10.87) 1955 (8.01) MX 708 (51.72) 17153 (54.26) 13512 (61.14) 14955 (61.24) \<0.001 AJCC 7^th^ TNM stage IA 36 (2.63) 1290 (4.08) 1132 (5.12) 1364 (5.59) IB 15 (1.10) 755 (2.39) 612 (2.77) 730 (2.99) IIA 21 (1.53) 844 (2.67) 660 (2.99) 761 (3.12) IIB 38 (2.78) 1531 (4.84) 999 (4.52) 987 (4.04) IIIA 51 (3.73) 1490 (4.71) 934 (4.23) 744 (3.05) IIIB 40 (2.92) 1207 (3.82) 629 (2.85) 543 (2.22) IIIC 28 (2.05) 710 (2.25) 434 (1.96) 320 (1.31) IV 385 (28.12) 5452 (17.24) 2403 (10.87) 1955 (8.01) Unknown 755 (55.15) 18336 (58.00) 14296 (64.69) 17018 (69.69) \<0.001 ------------------------------------------------------------------------------------------------- NOS: Not otherwise specified; SD: Standard Deviation; AJCC: The American Joint Committee on Cancer; TNM: Tumor Node Metastasis ###### Survival analysis for young patients with gastric cancer Univariate analysis Multivariate analysis --------------------------- --------------------- ----------------------- --------- ----------- -------------- --------- Sex \-- Male 14 22.09% \-- Female 16 26.45% 0.0667 \-- Location Cardia 13 16.26% Reference Fundus 10 14.64% 0.989 0.887-1.103 0.845 Body 16 28.66% 0.986 0.912-1.067 0.735 Antrum 22 35.72% 0.974 0.918-1.032 0.373 Pylorus 22 32.83% 1.042 0.930-1.167 0.478 Lesser curvature 26 33.57% 0.870 0.806-0.939 \<0.001 Greater curvature 17 31.61% 1.060 0.958-1.173 0.261 Overlapping 10 10.38% 1.094 1.013-1.180 0.021 NOS 12 21.68% \<0.001 1.056 0.977-1.142 0.171 Race/Ethnicity Caucasian 14 22.57% Reference African-American 11 20.29% 0.987 0.928-1.050 0.684 Asian 22 30.15% 0.805 0.760-0.853 \<0.001 Others 13 32.33% 0.0005 0.931 0.807-1.074 0.327 Tumor size \<5cm 40 38.47% Reference \>=5cm 15 18.27% \<0.001 1.124 1.078-1.172 \<0.001 Lauren \-- Diffuse 15 23.94% \-- Intestinal 14 23.99% 0.5443 \-- Histology subgroup \-- Adenocarcinoma 15 24.21% \-- Signet Ring cell 15 23.48% \-- Mucinous adenocarcinoma 13 34.89% 0.9155 \-- Grade \-- Well differentiated 21 31.52% \-- Moderately differentiated 17 26.35% \-- Poorly differentiated 15 23.88% \-- Undifferentiated 11 16.35% 0.0951 \-- Surgery Yes 25 34.50% Reference No 9 11.51% \<0.001 2.082 1.970-2.200 \<0.001 AJCC 7^th^ TNM stage IA NR 93.15% Reference IB NR 60% 2.007 1.744-2.311 \<0.001 IIA 80 54.62% 2.733 2.402-3.109 \<0.001 IIB 60 50.51% 4.325 3.845-4.864 \<0.001 IIIA 27 21.50 5.837 5.191-6.564 \<0.001 IIIB 27 40.19% 7.311 6.481-8.247 \<0.001 IIIC 19 31.28% 10.190 8.971-11.574 \<0.001 IV 9 4.97% 10.640 9.473-11.951 \<0.001 NOS: Not otherwise specified; CSS: Cause Specific Survival; AJCC: The American Joint Committee on Cancer; TNM: Tumor Node Metastasis [^1]: \*Wen-Long Guan, Lu-Ping Yuan and Xiang-Lei Yan contributed equally to the manuscript. [^2]: Competing Interests: The authors have declared that no competing interest exists.

Introduction {#s1} ============ Tropical flowers are produced worldwide because of their ornamental values. Postharvest handling of tropical flowers (*e.g.* Anthurium) is usually difficult due to their sensitivity to low temperatures. Anthurium (*Anthurium andraeanum*) is a tropical plant used in ornamental industry for its colorful spathes and green leaves. It is produced in wide ranges of climates; in locations far away from their original habitats in greenhouses. Although cut Anthurium has a long vase life when compared to the vase life of many other cut flowers ([@B33]), postharvest exposure to low temperatures can reduce its vase life. Temperatures below 12°C induce symptoms of chilling injury such as brown spots on the Anthurium spathe ([@B41]). In the winter time of temperate and cold climates, to prevent negative effects of cold temperatures on Anthurium cut flowers, the growers transfer the flowers to higher temperatures immediately after harvest. The importance of chilling stress for Anthurium flowers can be highlighted during their transport, storage, and distribution in the winter time. Anthurium flowers may be exposed to low temperatures as they are often part of mixed transport or storage with other flowers that need lower temperatures. This may cause a decrease in the quality of the spathe in customer locations. Browning and blueing of spathe and wilting of spadix are observed in Anthurium cut flowers when they face temperatures lower than 12°C ([@B41]; [@B48]). Much effort has been made during the last decade to decrease the susceptibility of Anthurium cut flowers to low temperatures ([@B41]; [@B48]; [@B50]). These efforts were mainly focused on the application of chemical solutions to the stems of cut flowers. For instance, salicylic acid ([@B41]) and *γ*-aminobutyric acid (GABA) ([@B48]; [@B50]; [@B51]) have successfully decreased the negative effects of low temperatures on cut flower quality and vase life. Application of these chemicals caused approximately 10% reduction in occurrence of browning of Anthurium spathes. In the other cases no effect of chemical solutions was observed. For instance, using calcium in holding solution did not result in vase life improvement in cold-stored Anthurium cut flowers ([@B19]). There is still a need for preventing the problems related to chilling injury beyond the use of chemicals. In some cut flowers with long vase life such as Anthurium, long distance transport to customers is common. During the postharvest period the cut flowers are usually held in darkness and may be exposed to cold (\<12°C). A positive role for the presence of light during postharvest storage has been proposed for keeping quality of flowers with photosynthetic leaves such as rose, chrysanthemum, and protea ([@B42]; [@B43]). To the best of our knowledge, no study has been done on the effect of light (spectra) on postharvest quality of cut flowers such as Anthurium. Nowadays, light emitting diodes (LEDs) lighting technology is widely used in horticulture ([@B23]). The LEDs have long lifetimes, small mass, low heat production, high radiant efficiency, physical robustness, and narrow spectrum ([@B55]; [@B20]; [@B5]; [@B32]). By using LEDs it is possible to apply specific wavelengths to plant material to study particular plant responses. Red (R) and blue (B) wavelengths are the main light spectra influencing water relations and gas exchange features of the plants ([@B30]; [@B14]; [@B36]; [@B37]). Research on the effect of spectral wavelengths on plant responses is still ongoing. After harvest, cut flowers still respond to environmental cues such as light spectra. Many plant processes are influenced by light spectrum in the range between 380 and 750 nm. For instance, B light induces water influx into the guard cells and thus, stomatal opening through activation of plasma membrane H^+^-ATPases ([@B21]). Red light increases phosphorylation levels of the H^+^-ATPase in response to the B light and therefore has inductive effects on B light-dependent stomatal opening and water loss by the leaf ([@B35]; [@B52]). The importance of spathe water loss during postharvest storage under different temperatures has been previously reported ([@B45]). It was shown that at temperatures above 13°C when transpiration is more than the water uptake; the product would not be marketable anymore. At 8°C both water uptake and transpiration will be hampered and chilling symptoms (*e.g.* blueing and browning of the spathes) will appear ([@B45]). Spectrum is one of the light characteristics that considerably influence plant responses. Light quality not only influences photosynthesis but also cellular integrity, water relations, pigmentation, carbohydrate, and antioxidant status of the plants ([@B3]; [@B15]; [@B16]). We hypothesized that different light spectra through their effects on cellular metabolism can influence the cold tolerance and quality of the spathe and hence the vase life of the Anthurium flower. The main aim of this study was to investigate the effects of different light spectra on postharvest performance of Anthurium cut flowers at a chilling temperature. Materials and Methods {#s2} ===================== Flowers and Lighting Treatments {#s2_1} ------------------------------- Cut flowers of Anthurium (*Anthurium andraeanum*) cultivars with red ('Calore') and white ('Angel') spathes were obtained from a commercial Anthurium greenhouse in the morning. Both cultivars were said to have long vase life in temperatures over 13°C, but 'Angel' was considered more sensitive to cold than 'Calore' (personal communication with Anthurium growers). Anthurium cut flowers were harvested at the commercial harvest stage when 40--50% of the spadix true flowers were fully opened ([@B41]). In the greenhouse, flowers with no deformities, no bruises, and straight peduncles were cut and placed into 50 ml tubes containing water. They were immediately transported (within less than 1 h after harvest) at 21°C to the laboratory where the flower stems were recut to 30 cm length. Each flower was placed in flasks containing 500 ml tap water. Sixty flasks with cut flowers (30 cut flowers from each cultivar) were placed into the light treatment compartments (l × w × h = 0.8 m × 0.5m × 0.5m). Compartments were positioned in a climate room with relative humidity of 80--81% and fixed temperature of 4°C provided by constant ventilation inside the cold rooms. Twelve flasks (six per each cultivar) were placed in each of the aforementioned compartments in the dark (D) and under continuous light intensity of 125 ± 5 μmol m^−2^ s^−1^ photosynthetic photon flux density (PPFD) but with different light spectra including white (W), blue (B), red (R), and 70% R + 30% B (RB) provided by LED production modules (24W, Iran Grow Light Co, Iran). Light intensity and spectra were measured using a light meter (Sekonic C-7000, Japan) ([**Figure 1**](#f1){ref-type="fig"}). {#f1} Visual Quality and Vase Life {#s2_2} ---------------------------- Each flower was inspected daily, and its vase life was estimated based on the visual symptoms such as loss of spathe glossiness, desiccation of spadix, blackening and wilting of spathe ([@B39]) as well as browning and desiccation of spadix. Visual symptoms on the spathes were recorded using a scale from 1 to 5; 1 = no chilling injury; 2 = mild injury (1--20% of spathe affected); 3 = moderate injury (21--50% of spathe affected); 4 = severe injury (51--80% of spathe affected); 5 = very severe injury (81--100% of spathe affected). Relative Water Loss Percentage {#s2_3} ------------------------------ Every day during the experiment, fresh weight (FW) of individual Anthurium cut flower was measured. FW differences between two subsequent days were calculated (ΔFW). Accumulative water loss of the Anthurium cut flowers was expressed as percentage water loss relative to the initial weight. Electrolyte Leakage {#s2_4} ------------------- For measuring EL, 20 discs (8 mm) per spathe (three spathes) were floated in 10 ml deionized water in closed vials and incubated at 25°C on a rotary shaker for 24 h. Thereafter, electrical conductivity of the solution (C~1~) was determined using a conductivity meter (Metrohm, Switzerland). Subsequently, samples were autoclaved at 120°C for 20 min, and the electrical conductivities of the solution (C~2~) were again recorded after equilibration at 25°C. EL was calculated based on the following equation. EL = C 1 C 2 × 100 Osmotic Potential {#s2_5} ----------------- Osmotic potential of the spathe was determined using the method described by [@B29]. The measurement was performed on three flowers per treatment. Spathes were cut into small segments; tissue pieces were placed in Eppendorf tubes perforated with four small holes and immediately frozen in liquid nitrogen. After being individually encased in a second intact Eppendorf tube, the samples were allowed to thaw for 30 min and centrifuged at 15,000 g for 15 min at 4°C. The extracted sap was collected and used for *ψ*s determination. Osmolarity (c) was measured with a vapor pressure osmometer (Osmomat 030-gonatec) and converted from mOsmole kg^−1^ to MPa according to the Van't Hoff equation ([@B29]). Proline Concentration {#s2_6} --------------------- Free proline content was measured based on the method described by [@B2]. This measurement was performed on three spathes per treatment. Grounded spathe samples (0.5 g) were homogenized in 3% (w/v) sulphosalicylic acid and then filtered through filter papers. Following addition of acid-ninhydrin and glacial acetic acid, the mixture was heated at 100°C for 1 h in a water bath. Reaction was stopped using an ice bath. The mixture was extracted with toluene, and the absorbance of the fraction with toluene aspired from the liquid phase was read at 520 nm (Perkin Elmer Lambda 25 UV-VIS Spectrometer). Proline concentration was determined using a calibration curve and expressed as μmol proline g^−1^ FW ([@B2]). Hydrogen Peroxide {#s2_7} ----------------- Hydrogen peroxide (H~2~O~2~) content was spectrophotometrically (Perkin Elmer Lambda 25 UV-VIS Spectrometer) measured after reaction with potassium iodide (KI). Ground spathe samples (0.25 g) were homogenized in 0.1% trichloroacetic acid (TCA) and centrifuged at 5,000 g for 10 min. The reaction mixture contained 0.5 ml of 0.1% trichloroacetic acid (TCA), spathe extract supernatant, 0.5 ml of 100 mm K-phosphate buffer and 2 ml reagent (1 m KI w/v in fresh double-distilled H~2~O). The blank contained 0.1% TCA in the absence of leaf extract. The reaction was developed for 1 h in darkness, and absorbance was measured at 390 nm. The amount of H~2~O~2~ was calculated using a standard curve prepared with known concentrations of H~2~O~2~ according to the method described by [@B38]. This measurement was performed on three spathes per treatment. Carbohydrate Contents {#s2_8} --------------------- Ground spathe samples (300 mg FW) were mixed with 7 ml of 70% ethanol (w/v) for 5 min on ice and centrifuged at 6,700 *g* for 10 min at 4°C. After adding 200 ml of the supernatant to 1 ml of an anthrone solution (0.5 g anthrone, 250 ml 95% H~2~SO~4~, and 12.5 ml distilled water), the absorbance was spectrophotometrically (Perkin Elmer Lambda 25 UV-VIS Spectrometer) recorded at 625 nm ([@B59]). This measurement was performed on three spathes per treatment. Pigments {#s2_9} -------- The Chl and carotenoid contents of the spathes were measured according to the method described by [@B1]. For measuring the anthocyanin content of the spathes, 1 g of ground spathe tissue was homogenized in 10 ml methanol, and the extract was incubated at 4°C in the dark overnight. The slurry was centrifuged (SIGMA-3K30) at 4,000 *g* for 10 min. The anthocyanin in the supernatant was spectrophotometrically (Perkin Elmer Lambda 25 UV-VIS Spectrometer) determined at 520 nm according to the method described by [@B61]. This measurement was performed on three spathes per treatment. Statistical Analysis {#s2_10} -------------------- Nondestructive measurements (water loss and vase life) were done on six spathes as six replicates. Destructive measurements were done on three spathes as three biological replicates. The experiment was repeated twice in precisely controlled cold room with fixed temperature of 4°C. Since the results of both experiments for the vase life were identical between the two experiments, part of the destructive measurements was performed in flowers from the first experiment and part of the measurements in flowers from the repeated experiment. In both experiments visual judgement of spathe quality and daily water loss were measured. Apart from these measurements, in the first experiment, EL, chlorophyll and carotenoids content, photosynthetic activity, and spathe characteristics were determined; in the second experiment osmotic potential, carbohydrates, anthocyanins, proline and hydrogen peroxide contents were determined after 14 days of treatment. Osmotic potential and EL were determined in fresh spathe samples; the other analyses were done in spathe samples ground in liquid nitrogen and kept in −80°C. In all the cases measurements were performed in three or more biological replications. The data were subjected to two-way analysis of variance (ANOVA), and Tukey was used as a means separation test, and P \> 0.01 was considered not significant. The percentage of spathe water loss data was fitted with linear regression, and F test was used for comparing the slopes of the curves. GraphPad Prism 7.01 for Windows (GraphPad software, Inc. San Diego, CA) was used for statistical analysis and comparisons among treatments. Results {#s3} ======= Vase Life of Anthurium Depends on Postharvest Light Spectra and Cultivar {#s3_1} ------------------------------------------------------------------------ Vase life of Anthurium cut flowers at 4°C was significantly (P ≤ 0.01, [**Table 1**](#T1){ref-type="table"}) influenced by the interaction between light spectra and cultivar ([**Figures 2**](#f2){ref-type="fig"} and [**3**](#f3){ref-type="fig"}). Among the light spectra, the longest vase life was observed in spathes exposed to R in both cultivars ([**Figure 3**](#f3){ref-type="fig"}). In 'Angel', the shortest vase life was observed under W, B, and RB spectra ([**Figures 2**](#f2){ref-type="fig"} and [**3**](#f3){ref-type="fig"}). After 14 days exposure to B spectrum, the spathe and spadix of 'Angel' had become dark brown ([**Figure 2**](#f2){ref-type="fig"}). In 'Calore', exposure to D, B, and RB spectra resulted in shorter vase life when compared to the vase life of flowers exposed to R and W spectra. These results indicate that R spectrum is able to prolong the vase life of Anthurium cut flowers in cold environments and that vase life under light was roughly dependent on the percentage of B wavelengths in the overall spectrum: the more B, the shorter the vase life. ###### Analysis of variance (*F* values) for assessed parameters for Anthurium cut flowers exposed to different light spectra under cold storage condition (4°C). Independent variables ----------------------- ----------------------- --------------- ----------------- Vase life 13.8^\*\*^ 18.2^\*\*^ 6.8^\*\*^ EL 7.01^\*\*\*^ 57.6^\*\*^ 13.1^\*\*^ Osmotic potential 4.5^\*\*\*^ 13.1^\*\*\*^ 1.4^ns^ Proline 3.8^\*^ 71.1^\*\*^ 6.7^\*\*^ Anthocyanin 17^\*\*\*\*^ 45^\*\*\*\*^ 4.7^\*\*^ Carotenoids 0.55^ns^ 0.29^ns^ 0.54^ns^ Soluble carbohydrates 2.9^\*^ 0.35^ns^ 13.8^\*\*^ Hydrogen peroxide 16.01^\*\*\*\*^ 296^\*\*\*\*^ 17.39^\*\*\*\*^ ns, Nonsignificance. \*Significance at 0.05 probability level. ^\*\*^Significance at 0.01 probability level. ^\*\*\*^Significance at 0.001 probability level. ^\*\*\*\*^Significance at 0.0001 probability level. ![Chilling injury symptoms in Anthurium cut flowers ('Angel' and 'Calore') held in the dark (D) or continuously exposed to 125 ± 5 μmol m^−2^ s^−1^ of different light spectra \[white (W), red (R), red and blue (RB), blue (B)\] after 14 days vase life at 4°C.](fpls-11-00846-g002){#f2} ![Vase life of Anthurium cut flowers ('Angel' and 'Calore') held in the dark (D) or continuously exposed to 125 ± 5 μmol m^−2^ s^−1^ of different light spectra \[white (W), red (R), red and blue (RB), blue (B)\] at 4°C. End of vase life was determined by visual symptoms such as loss of spathe glossiness, desiccation of spadix, blackening and wilting of spathe as well as browning and desiccation of spadix. Vase life was terminated when flowers rated 3 for spathe browning or 4 for spadix necrosis. Values are the means of six biological replicates, and bars indicate means ± SEM. Bars with different letters are significantly different (ANOVA, P \< 0.01).](fpls-11-00846-g003){#f3} Spathe Water Loss Is Controlled by Postharvest Light Spectra {#s3_2} ------------------------------------------------------------ Relative water loss percentage during 14 days of experiment was lowest under D and R spectrum in both cultivars ([**Figure 4**](#f4){ref-type="fig"}). In 'Calore' ([**Figure 4A**](#f4){ref-type="fig"}), the highest relative water loss percentage was observed under W; in 'Angel' under B spectrum ([**Figure 4B**](#f4){ref-type="fig"}). The slope of the water loss curve under W light in 'Calore' and under B lights in 'Angel' was approximately doubled in comparison with the slope of the curves in D ([**Table 2**](#T2){ref-type="table"}). In both cultivars, no statistical differences for the slope of water loss were observed between D and R spectrum. ![Relative water loss percentage in Anthurium cut flowers \['Calore' **(A)** and 'Angel' **(B)**\] held in the dark (D) or continuously exposed to 125 ± 5 μmol m^−2^ s^−1^ of different light spectra \[white (W), red (R), red and blue (RB), blue (B)\] during 14 days storage at 4^°^C. Values are the means of six biological replicates, and bars indicate means ± SEM. Data were fitted with linear regression, and the lines indicate the fitted line. Color of each line corresponds to the light spectrum of each treatment, gray stand for W spectrum.](fpls-11-00846-g004){#f4} ###### Slope of the curve for accumulative percentage of water loss over time and spathe osmotic potential after 14 days of treatment in two Anthurium cut flowers ('Angel' and 'Calore') exposed to different light spectra \[white (W), red (R), red and blue (RB), blue (B) and dark (D)\] at 4°C. -------------------------------------------------------------------- Cultivar Light spectrum Slope Osmotic potential\ (Mpa) ---------- ---------------- ------------------- -------------------- **W** 0.34 ± 0.007 ^a^ −1.41 ± 0.32^A,b^ **D** 0.19 ± 0.005 ^e^ −1.71 ± 0.27^A,ab^ Calore **R** 0.21 ± 0.006 ^de^ −1.13 ± 0.33^A,b^ **B** 0.26 ± 0.007 ^bc^ −2.02 ± 0.12^A,a^ **RB** 0.28 ± 0.012 ^b^ −1.03 ± 0.25^A,b^ **W** 0.19 ± 0.006 ^e^ −1.34 ± 0.02^B,a^ **D** 0.12 ± 0.002 ^g^ −0.91 ± 0.05^B,b^ Angel **R** 0.13 ± 0.005 ^g^ −0.82 ± 0.13^B,b^ **B** 0.24 ± 0.005 ^c^ −1.21 ± 0.08^B,a^ **RB** 0.15 ± 0.007 ^f^ −0.71 ± 0.12^B,b^ -------------------------------------------------------------------- Values are the means of six biological replicates for the Slope and three biological replicates for the Osmotic potential, and bars indicate means ± SEM. In the case of osmotic potential, different capital letters indicate the significant differences between cultivars and different small letters indicate significant difference among light spectra. Cellular Integrity and Osmoregulation Is Influenced by Light Spectra and Cultivar {#s3_3} --------------------------------------------------------------------------------- Electrolyte leakage (EL) after 14 days of treatment was considerably influenced by the interaction between the light spectra and cultivar (P ≤ 0.01, [**Table 1**](#T1){ref-type="table"}). EL was generally higher in 'Angel' than in the 'Calore'. In 'Calore', EL was significantly lower under all light treatments compared to D. The lowest EL in 'Calore' was detected in spathes exposed to the R spectrum. In 'Angel' EL was highest in spathes exposed to W and B spectra and was lowest in spathes exposed to R spectrum ([**Figure 5**](#f5){ref-type="fig"}). ![Electrolyte leakage from two cultivars of Anthurium cut flowers ('Calore' and 'Angel') held in the dark (D) or continuously exposed to 125 ± 5 μmol m^−2^ s^−1^ of different light spectra \[white (W), red (R), red and blue (RB), blue (B)\] following 14 days storage at 4°C. Values are the means of three biological replicates, and bars indicate means ± SEM. Bars with different letters are significantly different (ANOVA, P \< 0.01).](fpls-11-00846-g005){#f5} Osmotic potential was influenced by the single effects of the light spectra (P ≤ 0.001, [**Table 1**](#T1){ref-type="table"}) and cultivar (P ≤ 0.001, [**Table 1**](#T1){ref-type="table"}). Osmotic potential of the spathe in 'Calore' after 14 days of treatment was 32% higher (more negative) than osmotic potential of spathe in 'Angel' ([**Table 2**](#T2){ref-type="table"}). Among the light spectra, osmotic potential of 'Calore' spathes under the B spectrum was the highest. Under the B spectrum, osmotic potential was 40% higher than the osmotic potential under the R spectrum. In 'Angel' osmotic potential was highest under the W and B spectra ([**Table 2**](#T2){ref-type="table"}). Under all light spectra, proline concentration in the spathes of 'Calore' was higher than in 'Angel' ([**Figure 6**](#f6){ref-type="fig"}). In the spathe of 'Calore' the highest proline concentration was detected under the R spectrum; in 'Angel' the lowest proline concentration was measured under R light. There was no clear trend in proline concentration with the proportion of B spectrum. ![Proline contents in the spathes of two cultivars of Anthurium cut flowers ('Calore' and 'Angel') held in the dark (D) or continuously exposed to 125 ± 5 μmol m^−2^ s^−1^ of different light spectra \[white (W), red (R), red and blue (RB), blue (B)\] following 14 days storage at 4^°^C. Values are the means of three biological replicates, and bars indicate means ± SEM. Bars with different letters are significantly different (ANOVA, P \< 0.01).](fpls-11-00846-g006){#f6} Concentration of soluble carbohydrates in the spathe was affected by the interaction between light spectra and cultivar (P ≤ 0.01, [**Table 1**](#T1){ref-type="table"}). In 'Calore', the lowest concentration of soluble carbohydrates was detected under the R, B, and RB spectra; the highest concentration of carbohydrates was detected under the W spectrum and D. In 'Angel', the concentration of soluble carbohydrates in W was lower than in the other treatments ([**Figure 7**](#f7){ref-type="fig"}). There was no clear trend in carbohydrate concentration with the proportion of B spectrum. ![Soluble carbohydrate concentrations in the spathes of two cultivars of Anthurium cut flowers ('Calore' and 'Angel') held in the dark (D) or continuously exposed to 125 ± 5 μmol m^−2^ s^−1^ of different light spectra \[white (W), red (R), red and blue (RB), blue (B)\] following 14 days vase life at 4°C. Values are the means of three biological replicates, and bars indicate means ± SEM. Bars with different letters are significantly different (ANOVA, P \< 0.01).](fpls-11-00846-g007){#f7} Hydrogen peroxide content of the spathe was influenced by the interaction between light spectra and cultivar (P ≤ 0.00001, [**Table 1**](#T1){ref-type="table"}). Under all light spectra, higher hydrogen peroxide content was observed in 'Calore' in comparison with its content in 'Angel'. In 'Calore', the highest hydrogen peroxide content was detected under the W and RB spectra. The hydrogen peroxide content of 'Calore' spathes was respectively six and three times higher than its content in 'Angel' spathes under the W and RB spectra ([**Figure 8**](#f8){ref-type="fig"}). ![Hydrogen peroxide content in the spathes of two cultivars of Anthurium cut flowers ('Calore' and 'Angel') held in the dark (D) or continuously exposed to 125 ± 5 μmol m^−2^ s^v1^ of different light spectra \[white (W), red (R), red and blue (RB), blue (B)\] following 14 days vase life at 4°C. Values are the means of three biological replicates, and bars indicate means ± SEM. Bars with different letters are significantly different (ANOVA, P \< 0.01).](fpls-11-00846-g008){#f8} Spathe Anthocyanin Content Depends on Cultivar and Light Spectra {#s3_4} ---------------------------------------------------------------- Chlorophyll was not detected in the two studied Anthurium cultivars. No significant difference in carotenoid content was detected between the two cultivars or among the light spectra and their interactions ([**Table 1**](#T1){ref-type="table"}). Anthocyanin content of the spathe was affected by the interaction between light spectra and cultivar (P ≤ 0.01, [**Table 1**](#T1){ref-type="table"}). Under all light spectra, except B, higher anthocyanin concentration was detected in 'Calore' in comparison with 'Angel'. In 'Calore', the highest anthocyanin concentration was detected under R spectrum; the lowest concentration was found under W and B spectra. Similarly, in 'Angel', the lowest anthocyanin concentration was detected under W and B spectra ([**Figure 9**](#f9){ref-type="fig"}). ![Anthocyanin concentrations in the spathes of two cultivars of Anthurium cut flowers ('Calore' and 'Angel') held in the dark (D) or continuously exposed to 125 ± 5 μmol m^−2^ s^−1^ of different light spectra \[white (W), red (R), red and blue (RB), blue (B)\] following 14 days vase life at 4°C. Values are the means of three biological replicates, and bars indicate means ± SEM. Bars with different letters are significantly different (ANOVA, P \< 0.01).](fpls-11-00846-g009){#f9} Vase Life of Anthurium Depended on Cellular Integrity and Anthocyanin Content {#s3_5} ----------------------------------------------------------------------------- In 'Angel', after 14 days of light treatment, a positive relationship (R^2^ = 0.82) was found between the light effects on the osmotic potential and the EL of the spathe ([**Figure 10A**](#f10){ref-type="fig"}). The EL of the spathe tissue was positively correlated with the slope of the water loss curve (R^2^ = 0.87) ([**Figure 10B**](#f10){ref-type="fig"}). Both EL (R^2^ = 0.66) and the slope of the water loss curve (R^2^ = 0.68) were negatively correlated with the vase life ([**Figures 10C, D**](#f10){ref-type="fig"}). {#f10} In contrast to 'Angel', in 'Calore', there were no clear relations between the above discussed parameters. Only a significant negative correlation was observed between the vase life and the EL (R^2^ = 0.89) ([**Figure 10C**](#f10){ref-type="fig"}). In both cultivars, anthocyanin content of the spathe at day 14 was negatively correlated with EL (R^2^ = 0.55 for the combined data); anthocyanin was positively correlated with the vase life (R^2^ = 0.65 for the combined data) ([**Figure 11**](#f11){ref-type="fig"}). {#f11} Blue Light Is a Determinate Factor for the Vase Life of Anthurium at Chilling Temperature {#s3_6} ----------------------------------------------------------------------------------------- The relation between vase life and EL depended on the percentage of B (400--500 nm) in the light spectrum ([**Figure12**](#f12){ref-type="fig"}). A negative relationship was observed between the percentage of B in the light spectrum and the vase life in both Anthurium cultivars \[R^2^ = 0.75 for 'Angel' ([**Figure 12A**](#f12){ref-type="fig"}) and R^2^ = 0.89 for 'Calore' ([**Figure 12B**](#f12){ref-type="fig"})\], and a positive relationship was found between the percentage of B in the light spectrum and EL \[R^2^ = 0.69 for 'Angel' ([**Figure 12A**](#f12){ref-type="fig"}) and R^2^ = 0.72 for 'Calore' ([**Figure 12B**](#f12){ref-type="fig"})\]. ![Relationships between percentage of blue light (400--500 nm) in overall spectral light composition \[B spectrum (%)\] and electrolyte leakage (EL) at day 14 and vase life for spathes of Anthurium cut flowers ('Angel' \[panel **(A)**\] and 'Calore' \[panel **(B)**\]) at 4°C. Gray symbols stand for electrolyte leakage (EL), and black symbols stand for vase life. Values are the means of six biological replicates for the vase life and three biological replicates for EL ± SEM.](fpls-11-00846-g012){#f12} Discussion {#s4} ========== Production of Anthurium cut flowers in cold seasons is challenged by the low temperatures during postharvest ([@B17]). Cold storage is usually used to regulate supply of cut flowers ([@B9]). However, exposure to cold negatively affects quality of Anthurium ([@B40]). Results obtained from current experiment showed that postharvest lighting is an important aspect determining the quality of Anthurium cut flowers during exposure to cold. In most of the studies on postharvest quality, the different light treatments were applied before harvest ([@B11]; [@B44]; [@B7]). As far as we are aware there is no reported study on the effects of postharvest light spectra on quality of cut flowers especially under cold storage. In the present study we have used two cultivars differing in spathe color; 'Calore' with red and 'Angel' with white spathe. Overall, the response to the postharvest light treatments was comparable in the two cultivars. More B in the spectrum correlated with a shorter vase life and higher EL ([**Figure 12**](#f12){ref-type="fig"}). Lower anthocyanin content was correlated with higher EL and shorter vase life ([**Figure 11**](#f11){ref-type="fig"}), and more water loss was correlated with the higher EL and shorter vase life ([**Figure 10**](#f10){ref-type="fig"}). The latter aspect, however, was evident in 'Angel' but not in 'Calore'. In the current study, postharvest exposure to B spectrum resulted in higher water loss from spathes of Anthurium especially when compared with R spectrum and D environment ([**Figure 4**](#f4){ref-type="fig"}). Spathes of Anthurium contain stomata ([@B8]), and it has been shown that spathe water content depends on its stomatal conductance ([@B10]). B spectrum can affect different processes in plants including stomatal opening and photosynthesis ([@B56]; [@B27]; [@B63]), and a wide range of publications supported the role of the B spectrum in stimulating stomatal opening ([@B22]; [@B54]; [@B12]). Although promotion of stomatal opening has been reported by both B and R spectra, the effect of R spectrum on stomatal opening is indirect. B spectrum can directly induce ion influx into the guard cells and as a result promotes stomatal opening ([@B21]), whereas the promoting role of R spectrum on stomatal opening is *via* its effects on mesophyll photosynthesis and guard cell chloroplasts ([@B35]; [@B52]). Although many studies reported the importance of light intensity during postharvest storage of horticultural crops, there are not many reports regarding the effects of light spectrum on quality of horticultural products ([@B18]; [@B34]; [@B28]; [@B65]; [@B60]). Many of these studies related the effects of light on postharvest quality to its effects on photosynthesis process ([@B57]; [@B34]; [@B28]). In the current study we did not detect any photosynthetic activity in the spathe of Anthurium (data not shown), which is probably due to lack of chlorophyll pigments in the spathe of these two Anthurium cultivars. The observed effects of light therefore cannot be attributed to photosynthetic processes. As a consequence, the R spectrum cannot promote stomatal opening, which led to lower water loss from Anthurium spathes in comparison with B-containing spectral compositions (B, W, and RB) ([**Figures 5**](#f5){ref-type="fig"} and [**6**](#f6){ref-type="fig"}). In the present study, measuring stomatal conductance was not attainable due to low temperature. However, water loss is often closely related to stomatal conductance. At first glance, it may be speculated that B spectrum triggers stomata opening and thus promotes water loss and EL, which is not favorable for the vase life. Although this idea holds true for 'Angel', it did not apply to 'Calore' where water loss was not correlated with the EL and vase life ([**Figure 10**](#f10){ref-type="fig"}). [@B66] studied the relationships between stomatal functionality and vase life of two rose cultivars during low temperature storage. They found that depending on the cultivar, stomatal functionality (closing response following dehydration) is associated with water stress and vase life of the rose cultivars ([@B66]). Similar results were obtained for 'Angel'; however, no distinct correlation was found between water loss and vase life of 'Calore' ([**Figure 10D**](#f10){ref-type="fig"}) indicating that other mechanisms are involved. EL was correlated with osmotic potential ([**Figure 10A**](#f10){ref-type="fig"}) and percentage of water loss ([**Figure 10B**](#f10){ref-type="fig"}) in the spathe of Anthurium (especially for 'Angel'). Water loss induces alterations in cellular metabolism ([@B62]), resulting in accumulation of soluble carbohydrate and proline ([@B6]). In our experiment, no correlation was found between water loss percentage and proline or soluble carbohydrate concentrations among different light spectra. This partial contradiction in our findings compared to the previous reports can be due to the impact of low temperature on induction of osmotic solutes ([@B26]). Accumulation of different osmotic substances results in a decline in osmotic potential ([@B6]) and as a result membrane deterioration ([@B46]). Loss of membrane integrity increases EL from the tissues. The relationship between the rate of water loss and changes in EL indicates that water loss in Anthurium spathes is associated with membrane injury and a decline in vase life ([**Figure 10C**](#f10){ref-type="fig"}). However, the relationship between the rate of water loss with EL in 'Calore' was not the same as in 'Angel', and it seems the vase life is controlled by other mechanisms than those regulated with water loss. [@B10] showed that different mechanisms are involved in determining vase life of different cultivars of Anthurium cut flowers, and they concluded that compared to Anthurium cultivars with long vase life, cultivars with a short vase life have higher stomatal conductance and lower spathe RWC. Although, 'Angel' is considered as a long vase life Anthurium cultivar (more than 40 days), its vase life is dramatically decreased by exposure to low temperature during storage (3--20 days) ([@B41]; [@B19]). Therefore, it can be considered as a short vase life cultivar under cold storage conditions. Our results in the case of 'Angel' are in accordance with [@B10], who showed that spathes water holding capacity affects EL and as a result vase life of Anthurium spathes. In the current study, negative relationships (R^2^ = 0.66 for 'Angel' and R^2^ = 0.89 for 'Calore') were detected between EL and vase life for both Anthurium cultivars ([**Figure 10C**](#f10){ref-type="fig"}). Similarly, increase in EL due to low temperature exposure has been reported in spathes of different Anthurium cultivars ([@B10]; [@B48]; [@B49]). ROS over-accumulation is one of the main reasons compromising cellular integrity ([@B31]). However, no correlation was found between hydrogen peroxide levels and the vase life in Anthurium cultivars. 'Calore' with red spathe color had higher hydrogen peroxide content than 'Angel' under all studied light spectra ([**Figure 8**](#f8){ref-type="fig"}), which further disapproves the involvement of hydrogen peroxide in determination of Anthurium vase life in cold storage. Anthocyanins are plant pigments that usually accumulate in plant tissues in response to a wide range of stressors ([@B25]). These plant pigments can help plants to cope with light stress in two ways: by serving as a sun screen and through scavenging of free radicals ([@B25]). Apart from their sunscreen properties that restrict penetration of light into the plant tissues ([@B58]), anthocyanins through oxyradical scavenging activities can protect membranes and limit cell disruption ([@B53]; [@B69]; [@B58]; [@B3]). The ameliorative role of anthocyanins in different plant parts (*e.g.* leaf and fruit skin) in response to cold and light stresses has been reported ([@B47]; [@B4]). In Begonia leaves, elevation in transcript levels of the anthocyanin biosynthesis genes and as a consequence accumulation of anthocyanin was detected following six days of co-exposure to low temperature and light (300 μmol m^−2^ s^−1^), while no increase in transcript levels of the anthocyanin biosynthesis genes and anthocyanin level was detected in plants exposed to the same condition and treated with 3-(3,4-dichlorophenyl)-1,1-dimethylurea (inhibitor of light mediated anthocyanin accumulation) or in plants exposed to 25/15°C (day/night) and 300 μmol m^−2^ s^−1^ light intensity ([@B4]). These findings are indicative of the production of anthocyanins possibly in response to light under low temperatures as protectants against excessive ROS production and subsequent chilling injury. In agreement with these findings, a negative relationship between anthocyanin concentration and EL of the spathe and a positive relationship between anthocyanin concentration and vase life were observed in the both studied Anthurium cultivars ([**Figure 11**](#f11){ref-type="fig"}). The protective role of anthocyanins on cellular integrity has been previously reported in different plant species following exposure to different abiotic stresses ([@B53]; [@B58]; [@B3]). Furthermore, the protective role of anthocyanins on membrane integrity during co-exposure of plant leaves to light and low temperature stresses has been reported ([@B24]; [@B13]). However, in our research no significant relationship between anthocyanin concentration and H~2~O~2~ in the spathe was observed in both studied cultivars. It has been shown that anthocyanins mainly accumulate in the vacuole and not in the chloroplasts where ROS accumulation occurs ([@B64]). This suggests that antioxidant activity is not the main function of anthocyanins during cold--light stress ([@B13]). In our study, the anthocyanins were mainly detected in the vacuole of the epidermal cells ([**Supplemental Figure 1**](#SM1){ref-type="supplementary-material"}). Confirming this result we did not detect any chlorophyll and as a result no photosystem II activity in both Anthurium cultivars. Therefore, due to the absence of chloroplast, we can conclude that anthocyanin mainly accumulated in the vacuole. It has been reported that anthocyanin accumulation is a response to high ROS level in the leaves ([@B67]). High original H~2~O~2~ content ([**Supplemental Figure 2**](#SM1){ref-type="supplementary-material"}) in the spathe of 'Calore' can be a reason for high anthocyanin content for this cultivar ([**Supplemental Figure 3**](#SM1){ref-type="supplementary-material"}). Our result showed that a higher percentage of B in the spectral light composition during postharvest phase resulted in higher EL and as a result shorter vase life in Anthurium cut flowers under cold storage conditions ([**Figure 4**](#f4){ref-type="fig"}). In accordance with our results, [@B68] reported that in *Camptotheca acuminata* B spectrum can cause damage to the membrane and an increase in EL, while R spectrum can prevent ROS accumulation in the cells which prevents electrolyte leakage from the cells. The different processes that are associated with the senescence in the two different cultivars in response to the percentage of B in the spectrum may be explained by assuming a dual role of B spectrum in cold stored Anthurium flowers ([**Figure 13**](#f13){ref-type="fig"}). Original high ROS content causes accumulation of anthocyanins, as it is occurred for the Calore cultivar. Anthocyanin may be degraded by the B light but at the same time be used to neutralize the oxidative stress. B spectrum may induce oxidative stress, indirectly through its negative effects on anthocyanin content or directly leading to loss of membrane integrity, electrolyte leakage, and senescence. In both cultivars there were significant correlations between the vase life and ion leakage and anthocyanin. This indicates that this route is important in explaining the effect of B light. The magnitude of the effect of B light was less in 'Calore' than in 'Angel'; this may be related to the higher amount of anthocyanin naturally present in the spathe of 'Calore' ([**Supplemental Figure 3**](#SM1){ref-type="supplementary-material"}). On the other hand, the B spectrum may affect stomatal conductance leading to increased water loss, changes in osmotic potential and as a result loss of membrane integrity and senescence. Both routes may interact with each other. {#f13} In 'Angel', in addition to the oxidative stress route, there were also good correlations between the vase life and water loss, which was much less in the case of 'Calore'. This is caused by the relation between the percentage of the B light spectrum and the water loss. In 'Angel', water loss shows a linear relation with the percentage of the B light spectrum, indicating that stomatal conductance increases with increasing B spectrum; in 'Calore', however, maximum water loss occurs in 40% of B in the spectrum ([**Figure 4**](#f4){ref-type="fig"}). The observations that in both cultivars there are good correlations between the vase life and the factors of the oxidative stress route indicate that the oxidative stress route is more important than the water loss route for the explanation of the effect of B light under these specific conditions (lighting in the cold condition). This study is the first report showing the importance of light quality on determination of vase life of Anthurium under cold temperatures. However, further investigations regarding the effects of different temperatures, different intensities of light spectra and a wide range of Anthurium cultivars are needed to reach a practical solution for low temperature storage of cut Anthurium. Conclusion {#s5} ========== Anthurium is a tropical cold sensitive plant, which needs to be stored at 12.5--20°C. In the winter time and in cold environments, chilling injuries to Anthurium spathes decrease its marketability. Here, we showed that the spectral light composition during storage of Anthurium cut flowers is an important environmental factor determining postharvest flower performance. A high percentage of B in the light spectrum was associated with increased EL of the spathe and a shorter vase life. Therefore, postharvest handling of Anthurium cut flowers should preferably be performed in an environment with limited B spectrum when exposures to low temperatures are inevitable. Data Availability Statement {#s6} =========================== All datasets generated for this study are included in the article/[**Supplementary Material**](#SM1){ref-type="supplementary-material"}. Author Contributions {#s7} ==================== SA made substantial contributions to conception and design, also performed statistical analysis, drafted the manuscript, and critically revised the final version. ZF carried out the experiments, collected and critically analyzed the scientific literature, and helped in the writing of the manuscript. SD took part in designing and planning the experiments, preparing of material for the experiment. TL helped in the preparation of material for the experiment and took part in designing and planning of the experiments and critical revision of the final manuscript. EW contributed to conception and design of the final manuscript, scientific discussion, and critical revision of the final manuscript. Conflict of Interest {#s8} ==================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. We thank Iran Grow light company for providing LEDs and projectors. We would also like to thank The National Natural Science Foundation of China (No.31872955), Iran National Science Foundation (INSF) (grant number 96006991), and University of Tehran for their support. Supplementary Material {#s9} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fpls.2020.00846/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: Margherita Irene Beruto, Istituto Regionale per la Floricoltura (IRF), Italy [^2]: Reviewed by: Marcello Militello, Istituto Regionale per la Floricoltura (IRF), Italy; John Dole, North Carolina State University, United States [^3]: This article was submitted to Crop and Product Physiology, a section of the journal Frontiers in Plant Science [^4]: †These authors have contributed equally to this work

Background ========== Malaria remains one of the most important causes of morbidity and mortality in endemic areas, primarily affecting children under five years of age \[[@B1]\]. The highest death burden occurs in young children who have not yet developed protective immune mechanisms against the parasite. A minority of children appear to have a natural biological advantage thought to partially impede parasite growth \[[@B2]\]. Malaria is a complex disease that depends on many host genetic factors \[[@B3]\]. Indeed, resistance to *Plasmodium falciparum* is an important adaptive trait of human populations living in endemic areas \[[@B4]\]. Haemoglobin S (HbS) has become a stable polymorphism within malaria-endemic regions, associated with a limited life expectancy among homozygous individuals who suffer from sickle cell disease, and an extended life expectancy of heterozygous individuals who are more likely to evade malaria \[[@B5]-[@B7]\]. HbAS is widely known to confer significant protection from severe and uncomplicated malaria \[[@B6]-[@B12]\] although underlying mechanisms not precisely defined. Similar protection afforded by haemoglobin C (HbC) was more recently demonstrated although findings are less conclusive. Clinical studies performed in Nigeria and Mali has found no protection \[[@B13]-[@B15]\], while other Malian study and Burkina study indicated an association between HbAC and clinical malaria \[[@B16],[@B17]\]. Several innate or immune mechanisms have been hypothesized to explain malaria-protective effects of HbS or HbC \[[@B2],[@B18]-[@B20]\] Erythrocytes containing HbS or HbC may impede parasite growth and replication relative to normal red cells when subject to low oxygen tensions \[[@B18]\]. Protein targets of specific antibodies may be more rapidly exposed in HbS-containing red blood cells \[[@B21]\] resulting in an enhanced immune response to infection \[[@B18],[@B22]\]. It is also possible that unknown innate protective processes may up-regulate the malaria-specific immune response \[[@B23]\] or enhance nonspecific immunity to malaria \[[@B24]\]. The comparison of malaria indicators among populations with different genetic backgrounds that are uniformly exposed to the same parasite strains is one approach to the study of human heterogeneities in response to the infection \[[@B9],[@B17]\]. To characterize malaria risk in the population residing in the malaria vaccine trial site in Saponé, Burkina Faso, a haemoglobin genotyping study was conducted in children under five years of age living in this malaria endemic region. The study aimed to describe the relationship between abnormal haemoglobin genotypes and malaria in children under five years in a site being characterized for future malaria vaccine trials. Age-specific patterns of association were hypothesized to reflect the development of acquired immunity throughout early childhood. Methods ======= Study area and period --------------------- The study was conducted in four villages (Dawelgue, Tanghin, Kounda, and Watenga) in the Saponé Health District, located 50 km south-west of Ouagadougou, the capital city of Burkina Faso, in the Bazega province (Figure [1](#F1){ref-type="fig"}). The region is populated almost exclusively by the Mossi ethnic group, and farming is the main subsistence activity. The area has a rainy season lasting from June to October, corresponding to the high malaria transmission season, and a long dry season from November to May. The main malaria vectors are *Anopheles gambiae, Anopheles arabiensis*, and *Anopheles funestus*. The entomological inoculation rate (EIR) in 2001 was estimated at 0.3 and 44.4 infective/bites/person/month during the dry and rainy seasons, respectively in study area \[[@B25]\]. A demographic surveillance system (DSS) for monitoring vital events has been operating in the villages since 2002. {#F1} Enrolment was initiated on 23 January, 2007 and the final study visit for the last subject was completed on 29 February, 2008. Study participants ------------------ The study population consisted of children under five years of age identified from the above four villages whose parents were also resident in the study area. The sample was identified from a census list generated from the DSS census list of all children under five years living in the study region. Children were recruited approximately equally among age groups as follows: nought to 11 months, 12--23 months, 24--35 months, 35--47 months and 48--49 months. Study design and sample collection ---------------------------------- Exclusion criteria from the cohort included major congenital defects, any chronic disease, or anaemia defined as measured haemoglobin \<6 g/dl. Inclusion criteria were as follows: i) written/thumb-printed informed consent obtained from the parent or legal guardian of the child, ii) permanent resident in the study area and expected to remain so for at least one year following enrolment, iii) age between nought and five years. Prevalence of infection and malariometric indices were evaluated during bi-annual cross-sectional surveys within the low and high transmission seasons. The incidence of clinical malaria was documented through active and passive surveillance from the first survey through one year follow-up. Cross-sectional surveys ----------------------- Each cross-sectional survey consisted of a brief clinical examination and collection of blood by finger prick. The sampled blood was used for thick and thin blood film preparation for malaria microscopic examination. Haemoglobin measurements were done using a HemoCue machine. A venous blood sample of 5 ml was additionally collected at the first survey for haemoglobin genotyping. Active case detection --------------------- During the 12-month period of longitudinal surveillance, each child was visited twice per week by nurses living in the villages. The visit consisted of a brief history and clinical examination. Among children with an axillary temperature ≥ 37.5°C or reported history of fever in the past 24 hours, finger-prick blood samples were collected for haemoglobin measurement and malaria microscopic examination. Subjects with fever and positive rapid diagnostic test (RDT, OptiMAL) were recommended to visit the medical centre for treatment with artemether-lumefantrine. Passive case detection ---------------------- Parents of the children were encouraged to report to the nearest community clinic or hospital any time their child showed a sign of sickness in between home visits. Each child in attendance at either health facility received a brief clinical examination; any child with an axillary temperature ≥ 37.5°C or reported history of fever in the past 24 hours was tested for malaria and treated accordingly. Parasitological diagnosis ------------------------- Blood slides from either cross-sectional or longitudinal surveys were stained with Giemsa for microscopic identification of the *Plasmodium* species and determination of parasite density. Thick and thin blood films were air-dried; thin blood films were fixed with methanol, and both were stained with 3% Giemsa. One hundred high power fields were examined and the number of malaria parasites of each species and stage were recorded. The number of parasites per microlitre of blood was calculated by assuming that there were 20 white blood cells per high-power field and a fixed cell count of 8,000 per μl. Each film was read twice by two experienced technicians and a third reading was undertaken if discrepancies exceeding 30% occurred between the two readers. Haemoglobin typing ------------------ DNA was extracted using a QIAGEN kit (QIAamp DNA blood mini kit) and the haemoglobin was typed by polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP). Briefly, DNA samples were amplified using a 5'-AGG AGC AGG GAG GGC AGG A-3' forward primer and a 5'-TCC AAG GGT AGA CCA CCA GC-3' reverse primer. The 358 base pair bp fragment obtained was digested with MnlI for discrimination between HbAA (173 pb, 109 pb, and 60pb), HbSS/HbCC andHbSC (173 pb, 109 pb, and 76 pb), HbAS/HbAC (173 pb, 109 pb, 76 pb and 60). A second digestion with DdeI allowed for further discrimination for ambiguous results between HbSS (331 pb), HbCC (201 pb and 130 pb), HbSC (130 pb, 201 pb and 331 pb), HbAS (130 pb, 201 pb and 331 pb) and HbAC (201 pb and 130 pb). All digestion was carried out for three hours at 37°C and the products were run on a 3% agarose gel \[[@B17]\]. Malariometric indices --------------------- Malariometric indices were evaluated during the low and high transmission seasons using measures obtained from the biannual surveys. A clinical *P. falciparum* malaria episode was defined as an axillary temperature of ≥37.5°C or history of fever in the past 24 hours and *P. falciparum* trophozoite count \>2,500 parasites/μl. Splenomegaly of any size was identified by Hackett's classification \[[@B26]\]. Haemoglobin was measured using a HemoCue machine and reported as g/dl. Child age was defined as age (years) at enrolment. Statistical analysis -------------------- Demographic characteristics (age, sex and health district) and season-specific malariometric indices were compared among normal (AA) and abnormal Hb genotypes using Kruskal-Wallis tests for continuous and Chi-square or Fischer's exact tests for categorical measures. Clinical malaria incidence was calculated within strata defined by commonly occurring (AA, AC, and AS) genotype and season of follow-up as the total number of clinical malaria episodes observed by active and passive surveillance divided by the total number of person-years at risk. Children were excluded from incidence analysis for 28 days after recording an episode, to ensure that the infection causing the episode was only recorded once. The study period ended 365 days after the first visit or on the date of death/emigration from the study locale (N = 2). Age-specific risks associated with Hb genotype were evaluated using a Poisson regression model of haemoglobin genotype as a predictor of rate of malaria episodes observed over one year follow-up. The model was used to estimate rate ratios associated with AC, AS or either genotype relative to AA within age strata after adjustment for health district. Analyses were repeated among the full cohort to formally test for interactions of AC and AS genotype with age group. Cumulative incidence of clinical malaria within genotype strata was estimated using the Kaplan-Meier method. Malaria-free survival curves were compared among genotypes using the log-rank test. All data were double entered in Epi Info Version 6.04 (CDC, Atlanta, USA) and statistical analyses were performed in SAS Version 9.2. Ethical processes ----------------- The study protocol was explained to the local population prior to study initiation. Written informed consent was obtained from the parents or guardians of all participating children before enrolment. The protocol of this study was approved by the Burkina Faso Health Research Ethics Review Committee. Results ======= Baseline characteristics of the study cohort -------------------------------------------- The study cohort was divided into five predefined age groups: nought to one year (*n* = 75), one to two years (*n = 93*), two to three years (*n = 97*), three to four years (*n = 102*), and four to five years (*n = 85)*. Mean age of the cohort was 32 months; 47.8% were female (Table [1](#T1){ref-type="table"}). Frequencies of each haemoglobin genotype were 73.2% AA; 15.0% AC; 8.2% AS; 2.2% CC; 1.1% CS and 0.2% SS. Age and sex distributions were similar among genotype groups. ###### Demographic characteristics by Age, sex and haemoglobin genotype **Haemoglobin type**   **Gender** **All** ---------------------- -------------------- ------------ --------- ------ AA N 175 156 331   Age (months), mean 33.2 32.0 32.6   Age (months), SD 17.1 15.9 16.5 AC N 33 35 68   Age (months), mean 29.0 29.5 29.3   Age (months), SD 18.0 17.2 17.5 AS N 18 19 37   Age (months), mean 31.3 32.7 32.0   Age (months), SD 11.7 16.5 14.2 CC N 6 4 10   Age (months), mean 19.1 17.1 18.3   Age (months), SD 17.9 17.2 16.6 SC N 4 1 5   Age (months), mean 41.7 1.7 33.7   Age (months), SD 15.4 0 22.3 SS N 1 0 1   Age (months), mean 43.6 0 43.6   Age (months), SD 0 0 0 Non-AA N 62 59 121   Age (months), mean 29.8 29.2 29.5   Age (months), SD 16.4 17.3 16.8 Note: Age and sex distribution were similar among genotype. Malariometric indices --------------------- Prevalence of *P. falciparum* infection decreased from 59.7% to 50.4% from the low to high season; geometric mean parasite density increased from 1,579 (1,315--1,896) to 2,748 (2,104--3,589) trophozoites/μl. Prevalence of clinical malaria varied minimally (11.3% to 12.9%) (Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). Clinical malaria prevalence and mean haemoglobin did not vary by Hb genotype within either season; however *P. falciparum* infection was less prevalent among children with CC relative to AA genotype at each survey (p \< 0.05 for each). ###### Malariometric indices by haemoglobin genotype **Infection characteristic** **AA (N = 331)** **AC (N = 68)** **AS (N = 37)** **CC (N = 10)** **SC (N = 5)** **SS (N = 1)** **All (N = 452)** --------------------------------------------------------- ------------------ ------------------- ------------------ ------------------- ------------------ ---------------- ------------------- Symptomatic malaria (fever and \>0 parasites/μl), N (%) 39 (11.8%) 7 (10.3%) 4 (10.8%) 1 (10.0%) 0 (0.0%) 0 (0.0%) 51 (11.3%) *P. falciparum* infection (\>0 parasites/μl), N (%) 205 (61.9%) 39 (57.4%) 20 (54.1%) **2 (20.0%)\*** 4 (80.0%) 0 (0.0%) 270 (59.7%) Symptomatic malaria (fever and \>2500 parasites/μl) 22 (6.6%) 3 (4.4%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 25 (5.5%) *P. falciparum* infection (\>2500 parasites/μl), N (%) 82 (24.8%) 15 (22.1%) 6 (16.2%) 1 (10.0%) 1 (20.0%) 0 (0.0%) 105 (23.2%) Gametocyte carriage, N (%) 101 (30.5%) 21 (30.9%) 10 (27.0%) 3 (30.0%) 3 (60.0%) 0 (0.0%) 138 (30.5%) Splenomegaly, N (%) 67 (20.2%) 9 (13.2%) 4 (11.1%) 1 (10.0%) 0 (0.0%) 0 (0.0%) 81 (18.0%) Haemoglobin (g/dl), mean (SD) 9.7 (1.5) 9.5 (1.6) 9.5 (1.5) 9.4 (1.5) 11.1 (0.8) 10.1 9.6 (1.5) *P. falciparum* density (per/μl), GM (95% CI) 1546 (1259,1898) 1904 (1159, 3128) 1305 (527, 3231) 3088 (51, 185428) 1418 (104,19382) \-\-- 1579 (1315, 1896) Gametocyte density (per/μl), GM (95% CI) 51 (41, 64) 48 (30, 75) 95 (54, 166) 81 (2, 3110) 22 (0, 62183) \-\-- 52 (43, 64) Low transmission season. \*P for difference vs AA \< 0.05. \*\*P for difference vs AA \< 0.01. ###### Malariometric indices by haemoglobin genotype **Infection characteristic** **AA (N = 305)** **AC (N = 64)** **AS (N = 35)** **CC (N = 9)** **SC (N = 5)** **SS (N = 1)** **All (N = 419)** ---------------------------------------------------------- ------------------- -------------------------- ----------------------- ----------------- ------------------ ---------------- ------------------- Symptomatic malaria (fever and \> 0 parasites/μl), N (%) 37 (12.1%) 13 (20.3%) 2 (5.7%) 0 (0.0%) 2 (40.0%) 0 (0.0%) 54 (12.9%) *P. falciparum* infection (\> 0 parasites/μl), N (%) 152 (49.8%) 34 (53.1%) 19 (54.3%) **1 (11.1%)\*** 4 (80.0%) 1 (100.0%) 211 (50.4%) Symptomatic malaria (fever and \>2500 parasites/μl) 27 (8.9%) 9 (14.1%) 1 (20.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 37 (8.8%) *P. falciparum* infection (\> 2500 parasites/μl), N (%) 84 (27.5%) 22 (34.4%) **4 (11.4%)\*** 0 (0.0%) 2 (40.0%) 1 (100.0%) 113 (27.0%) Gametocyte carriage, N (%) 60 (19.7%) 11 (17.2%) 3 (8.6%) 2 (22.2%) 0 (0.0%) 0 (0.0%) 76 (18.1%) Splenomegaly, N (%) 26 (8.5%) 7 (10.9%) 2 (5.7%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 35 (8.4%) Haemoglobin (g/dl), mean (SD) 10.2 (1.5) 9.9 (1.4) 9.9 (1.5) 10.0 (1.5) 11.1 (1.1) 10.5 10.1 (1.5) *P. falciparum* density (per/μl), GM (95% CI) 2690 (1978, 3659) **5823 (3117, 10878)\*** **986 (383, 2541)\*** 364 1632 (7, 375567) 9740 2748 (2104, 3589) Gametocyte density (per/μl), GM (95% CI) 41 (31, 53) **20 (11, 37)\*** 23 (3, 173) 34 (4, 263) (NA) (NA) 36 (28, 45) High transmission season. \*P for difference vs AA \< 0.05. \*\*P for difference vs AA \< 0.01. Incidence of clinical malaria ----------------------------- Incidence of clinical malaria (95% CI) among the full cohort was 1.8 (1.7-1.9) episodes per person-year after one year follow-up. The season-specific incidence rate was 3.1 times higher in the high relative to the low transmission season (Table [4](#T4){ref-type="table"}). Within each season incidence rates were higher among children with AA genotype relative to abnormal genotypes, although differences were not statistically significant. ###### Incidence of malaria episodes (fever and parasitemia \>2500/μl) by haemoglobin genotype and season **Season** **Haemoglobin genotype** **Number of children** **Number of episodes** **Mean number of episodes per child** **Total person- years** **Incidence rate (95% CI) (per person-year)** -------------- -------------------------- ------------------------ ------------------------ --------------------------------------- ------------------------- ----------------------------------------------- Low AA 331 184 0.6 177.0 1.0 (0.9, 1.2)   AC 68 31 0.5 37.1 0.8 (0.6, 1.2)   AS 37 15 0.4 20.4 0.7 (0.4, 1.2)   CC 10 1 0.1 5.8 0.2 (0.0, 1.0)   SC 5 1 0.2 2.8 0.4 (0.0, 2.0)   SS 1 0 0.0 0.6 0.0 (0.0, 6.4)   Total 452 232 0.5 243.6 1.0 (0.8, 1.1) High AA 305 363 1.1 112.4 3.2 (2.9, 3.6)   AC 64 74 1.1 23.0 3.2 (2.5, 4.0)   AS 35 35 0.9 13.1 2.7 (1.9, 3.7)   CC 9 8 0.8 3.7 2.2 (0.9, 4.3)   SC 5 5 1.0 1.8 2.9 (0.9, 6.7)   SS 1 0 0.0 0.4 0.0 (0.0, 8.8)   Total 419 485 1.1 154.4 3.1 (2.9, 3.4) Low and High AA 331 542 1.6 288.0 1.9 (1.7, 2.0)   AC 68 104 1.5 59.8 1.7 (1.4, 2.1)   AS 37 50 1.4 33.1 1.5 (1.1, 2.0)   CC 10 9 0.9 9.3 1.0 (0.4, 1.8)   SC 5 6 1.2 4.5 1.3 (0.5, 2.9)   SS 1 0 0.0 1.0 0.0 (0.0, 3.7)   Total 452 711 1.6 395.7 1.8 (1.7, 1.9) Note: Incidence of malaria episodes was calculated as the number of malaria cases. divided by total person-years (p-yrs) at risk). In age-stratified analyses adjusted for health district, AC genotype was associated with lower incidence of clinical malaria relative to AA among children aged 1--2 years \[rate ratio (95% CI) = 0.66 (0.42, 1.04); p = 0.07\] and 2--3 years \[rate ratio (95% CI) = 0.37 (0.18, 0.75); p = 0.01\] (Table [5](#T5){ref-type="table"}). An association of opposite direction was however apparent among children aged 3--4 years: rate ratio (95% CI) = 1.61 (1.08, 2.41); p = 0.02. An association of AS genotype with lower incidence of clinical malaria relative to AA approached significance among children aged 2--3 years: rate ratio (95% CI) = 0.63 (0.40, 1.101); p = 0.06. ###### Incidence of malaria episodes (fever and parasitemia \>2500 parasites/μl) by haemoglobin genotype and age at enrollment **Age** **Haemoglobin genotype** **Number of children** **Number of episodes** **Incidence rate (95% CI) (per person-year)** **District-adjusted rate ratio (95% CI)** **P value** --------- -------------------------- ------------------------ ------------------------ ----------------------------------------------- ------------------------------------------- ------------- 0-1 yrs AA 52 115 2.7 (2.2, 3.2) \-\-- \-\--   AC 15 31 2.5 (1.7, 3.5) 0.93 (0.62, 1.38) 0.72   AS 2 2 1.1 (0.1, 3.9) 0.41 (0.10, 1.65) 0.21 1-2 yrs AA 66 131 2.3 (2.0, 2.8) \-\-- \-\--   AC 16 22 1.5 (1.0, 2.3) 0.66 (0.42, 1.04) 0.07   AS 9 19 2.5 (1.5, 3.9) 0.93 (0.56, 1.54) 0.77 2-3 yrs AA 70 145 2.5 (2.1, 2.9) \-\-- \-\--   AC 10 8 0.9 (0.4, 1.7) 0.37 (0.18, 0.75) **0.01**   AS 14 20 1.6 (1.0, 2.5) 0.63 (0.40, 1.01) 0.06 3-4 yrs AA 77 95 1.4 (1.1, 1.7) \-\-- \-\--   AC 17 32 2.2 (1.5, 3.1) 1.61 (1.08, 2.41) **0.02**   AS 6 9 1.7 (0.8, 3.2) 1.05 (0.53, 2.09) 0.89 4-5 yrs AA 66 56 0.9 (0.7, 1.2) \-\-- \-\--   AC 10 11 1.2 (0.6, 2.2) 1.27 (0.66, 2.44) 0.48   AS 6 0 0.0 (0.0, 0.6) 0.00 1.00 In analyses inclusive of all children aged 1--3 years and adjusted for health district, presence of either AS or AC genotype was associated with reduced risk of similar magnitude: rate ratio (95% CI) = 0.63 (0.49, 0.83); p = \<0.001. Among the full cohort, risk associated with AC genotype varied significantly among children younger versus older than three years after adjustment for health district (p for interaction \< 0.001). Time to first malaria episode did not significantly differ by AA, AC or AS genotype among the full cohort (p = 0.21) (Figure [2](#F2){ref-type="fig"}). Among children aged 1--3 years, however, delayed first malaria was apparent among children with AC (241 days) or AS (201 days) genotype relative to AA (198 days) (p = 0.04) (Figure [3](#F3){ref-type="fig"}). Associations of genotype with time to first malaria were not apparent among other age groups. {#F2} {#F3} Discussion ========== In this cohort of children under five years of age living in a high malaria-transmission region of Burkina Faso, AC or AS Hb genotype was associated with lower risk of clinical malaria relative to the AA genotype among children aged one to three years. This association was attenuated though approached significance among the full cohort. These data suggest that Hb genotype should be considered a potentially important confounder, particularly among younger children, in evaluations of clinical malaria risk in endemic regions. The haemoglobin S and C genes occurred in the cohort with similar frequencies to those previously published \[[@B17],[@B27],[@B28]\]; 26% were carriers of either HbS or HbC. The mechanism by which HbS protects against malaria has been the subject of speculation for more than 50 years. While protection may largely be conferred by the physical characteristics of HbS erythrocytes, a number of studies suggest that HbAS may also enhance the acquisition of naturally acquired immunity \[[@B19],[@B29]\]. Associations between the HbC trait and malaria risk, however, have not been uniformly established \[[@B30]\]. The finding that parasitaemia prevalence varied only minimally among AA, AC and AS genotypes in the full cohort is consistent with previous studies conducted in various parts of Africa \[[@B9],[@B31]-[@B33]\]. Malaria premunition has also been found comparable among haemoglobin groups among patients greater than six years of age \[[@B30]\]. Age-specific associations of malariometric indices with Hb gentoypes may be hypothesized to reflect the development of anti-malarial immunity \[[@B30]\] as the protective action of haemoglobin may be masked by maternally transmitted antibodies among children less than six months of age \[[@B34]\]. While age-specific patterns of association were not apparent in the current cross-sectional measures, further inferences are limited due to the small number of children within age and genotype strata. Mean parasite density was markedly lower in children AS relative to AA genotype normal haemoglobin, consistent with previous reports \[[@B17],[@B27]\]. Indeed, abnormal haemoglobin may not allow for optimal development of *Plasmodium* in deep organs where oxygen pressure is reduced. Parasite density was however higher among AC relative to AA genotype, suggesting potential mechanistic variation among protection afforded by abnormal genotypes in early childhood. Malaria-risk reduction associated with HbAS genotype has been reported in Mali\[[@B27],[@B35]\] Burkina \[[@B17]\], Ghana \[[@B36]\] and Kenya\[[@B11]\]; a similar protective advantage of HbAC has been less consistently supported. Abnormal Hb genotype in this cohort was further associated with lower incidence of clinical malaria episodes and delayed first episode among children aged one to three years. These effects were more subtle in infants and older children. Several studies have identified a protective effect of abnormal Hb against clinical malaria \[[@B17],[@B35]\]. Although HbAS was found unassociated with time to first malaria episode in Gabon \[[@B37]\], HbAS was associated with reduced time to all-cause mortality in Kenya, an effect attributed to reduction in malaria-specific outcomes including severe malarial anaemia and high-density parasitaemia\[[@B38]\]. Proposed mechanisms of protection \[[@B39]\] include decreased red blood cell invasion or poor growth under low-oxygen tension\[[@B40]\]; and accelerated acquisition of antibodies specific for *P. falciparum* erythrocyte membrane protein-1 (PfEMP-1) and other variant surface antigens\[[@B41]\]. Alternative findings among cohorts may in part reflect variation in age distribution and associated development of acquired immunity, malaria endemicity, and administration of anti-malarial drugs. Estimates of malaria burden in the current study are likely underestimates, due to prompt treatment of malaria episodes identified by active surveillance. This study is further limited by the small number of children with abnormal genotypes and restriction of follow-up to one year, preventing stronger inferences related to changing risk associated with alternative genotypes throughout early childhood. Conclusions =========== In this cohort of children under five years of age, AC or AS Hb genotypes was associated with lower risk of clinical malaria relative to normal genotype among children aged one to three years. This age-specific association may suggest influences of HbC and HbS genotypes in the development of naturally acquired immunity in early childhood. Evaluations of anti-malarial interventions in endemic regions should consider Hb genotype as a potentially important confounder, particularly among young children. Abbreviations ============= AA: Wild type; SS: Sickle cell; AS: Sickle trait; AC: Heterozygous for haemoglobin C; CC: Homozygote for haemoglobin C; SC: Haemoglobins S and C; Non-AA: Abnormal haemoglobin (AS, AS, SC, SS and CC); PCR: Polymerase chain reaction. Competing interest ================== The authors declare that they have no competing interests. Authors' contributions ====================== ECB designed the study, collected data and coordinated the study, performed statistical analysis and wrote the first draft of the manuscript. SBS, AT, IN, AO and ATK coordinated and participated in the design of the study, participated in the statistical analysis and procedures and the drafting of the manuscript. EBC, IS, SS and AD participated in the laboratory work and data interpretation. JBY, OE and ECB carried out the study. NW, MS and TJTD contributed to analysis and interpretation and to contribute to writing the paper. All the authors read and approved the final version. Acknowledgements ================ We thank the population of the study villages and local authorities of the health district of Sapone for their cooperation, and the Ministry of Health, Burkina Faso. We are grateful to the staff of the Centre National de Recherche et de Formation sur le Paludisme (CNRFP) whose participation has made this study possible and particularly to SERME Samuel for skilful technical assistance. We are also grateful to Mr. Walter Jones and Dr. Steven Rosenthal from NIAID/DMID for their helpful support. This study was supported by the regular budget of the National Institutes of Allergy and Infectious Diseases, National Institutes of Health, Department of Infectious Disease, Bethesda, MD (NIH, NIAID,DMID (DMID Contract HHSN266200400016C).

Basu J, Malhotra A, Papadakis M. Exercise and hypertrophic cardiomyopathy: Two incompatible entities? Clin Cardiol. 2020;43:43:889--896. 10.1002/clc.23343 1. INTRODUCTION {#clc23343-sec-0001} =============== The benefits of exercise on cardiovascular health and all‐cause mortality are well established.[1](#clc23343-bib-0001){ref-type="ref"}, [2](#clc23343-bib-0002){ref-type="ref"}, [3](#clc23343-bib-0003){ref-type="ref"}, [4](#clc23343-bib-0004){ref-type="ref"}, [5](#clc23343-bib-0005){ref-type="ref"}, [6](#clc23343-bib-0006){ref-type="ref"}, [7](#clc23343-bib-0007){ref-type="ref"}, [8](#clc23343-bib-0008){ref-type="ref"}, [9](#clc23343-bib-0009){ref-type="ref"} However, studies have demonstrated a nearly threefold increased risk of sudden cardiac death (SCD) in young athletes with underlying cardiovascular conditions when compared to sedentary individuals.[10](#clc23343-bib-0010){ref-type="ref"} Hypertrophic cardiomyopathy (HCM), first described by Teare[11](#clc23343-bib-0011){ref-type="ref"} is a genetic disorder, characterized by the presence of increased left ventricular wall thickness (≥15 mm) that is not solely explained by abnormal loading conditions.[12](#clc23343-bib-0012){ref-type="ref"} Historically, HCM was considered to be the leading cause of SCD in athletes.[13](#clc23343-bib-0013){ref-type="ref"}, [14](#clc23343-bib-0014){ref-type="ref"}, [15](#clc23343-bib-0015){ref-type="ref"} International recommendations therefore stood firm on the exclusion of athletes with HCM from most competitive sports.[16](#clc23343-bib-0016){ref-type="ref"}, [17](#clc23343-bib-0017){ref-type="ref"} Although scientific bodies approach the issue of participation of nonathletic patients with HCM in recreational sport in a more liberal fashion, guidance regarding safe levels of exercise remains vague.[18](#clc23343-bib-0018){ref-type="ref"} Consequently, patients adopt a sedentary lifestyle, with significant impact on their physical and psychological well‐being.[19](#clc23343-bib-0019){ref-type="ref"}, [20](#clc23343-bib-0020){ref-type="ref"}, [21](#clc23343-bib-0021){ref-type="ref"}, [22](#clc23343-bib-0022){ref-type="ref"}, [23](#clc23343-bib-0023){ref-type="ref"}, [24](#clc23343-bib-0024){ref-type="ref"}, [25](#clc23343-bib-0025){ref-type="ref"}, [26](#clc23343-bib-0026){ref-type="ref"} HCM is the most common ICC encountered in clinical practice and is thought to affect around 1 in 500 adults.[27](#clc23343-bib-0027){ref-type="ref"}, [28](#clc23343-bib-0028){ref-type="ref"}, [29](#clc23343-bib-0029){ref-type="ref"}, [30](#clc23343-bib-0030){ref-type="ref"} Recent data suggest that the prevalence may even be as high as 1 in 200.[31](#clc23343-bib-0031){ref-type="ref"} Extrapolating such figures indicate that approximately 120 000 people in the United Kingdom and 20 million people worldwide are affected,[32](#clc23343-bib-0032){ref-type="ref"} representing a significant global burden.[33](#clc23343-bib-0033){ref-type="ref"} The implementation of primary and secondary preventative strategies; preparticipation screening,[34](#clc23343-bib-0034){ref-type="ref"}, [35](#clc23343-bib-0035){ref-type="ref"}, [36](#clc23343-bib-0036){ref-type="ref"}, [37](#clc23343-bib-0037){ref-type="ref"} mandated familial evaluation following SCD or diagnosis of ICC in a family,[38](#clc23343-bib-0038){ref-type="ref"} education around cardiopulmonary resuscitation and automated external defibrillators[39](#clc23343-bib-0039){ref-type="ref"}, [40](#clc23343-bib-0040){ref-type="ref"} have resulted in an expansion of the HCM population. Furthermore, advances in management such as implantable cardioverter defibrillator (ICD) implantation have led to a reduction in mortality.[41](#clc23343-bib-0041){ref-type="ref"} As a result, HCM mortality rates are now comparable to the general population.[42](#clc23343-bib-0042){ref-type="ref"} Consequently, the need to reduce cardiovascular risk over a normal lifespan becomes a priority in these individuals. The risk of SCD during exercise in individuals with HCM, may not be as high as initially perceived. Recent postmortem studies have demonstrated that HCM accounts for a much smaller proportion of SCD in athletic individuals.[43](#clc23343-bib-0043){ref-type="ref"}, [44](#clc23343-bib-0044){ref-type="ref"} Moreover, recent studies of cardiac rehabilitation programs in older patients with HCM in their sixth decade of life, suggest that exercise in moderation may be safe.[45](#clc23343-bib-0045){ref-type="ref"}, [46](#clc23343-bib-0046){ref-type="ref"} Murine models and clinical studies in athletes with HCM suggest that exercise may even lead to favorable cardiac remodeling.[47](#clc23343-bib-0047){ref-type="ref"}, [48](#clc23343-bib-0048){ref-type="ref"} In a small study of lifelong competitive athletes with HCM there was no difference in terms of outcomes in athletes who adopted a more sedentary lifestyle compared to those who continued competitive sport.[49](#clc23343-bib-0049){ref-type="ref"} The shifting opinion regarding safety of exercise in HCM is reflected in the most recent European Association of Preventative Cardiology (EAPC) guidelines which support a more liberal approach towards sports participation in low risk individuals with HCM.[50](#clc23343-bib-0050){ref-type="ref"} Figure [1](#clc23343-fig-0001){ref-type="fig"} demonstrates the evolution in exercise recommendations from the European Society of Cardiology (ESC) and American Heart Association/American College of Cardiology (AHA/ACC) alongside key contributory studies of exercise in HCM. There still, however, remains intense debate among specialists as to whether the pendulum has now swung too far in the other direction. {#clc23343-fig-0001} 1.1. SCD and HCM {#clc23343-sec-0002} ---------------- Although the benefits of exercise are compelling, SCD is the leading medical cause of death in athletes.[10](#clc23343-bib-0010){ref-type="ref"}, [13](#clc23343-bib-0013){ref-type="ref"}, [14](#clc23343-bib-0014){ref-type="ref"}, [51](#clc23343-bib-0051){ref-type="ref"} The incidence is between 1 and 2 per 100 000 person years[10](#clc23343-bib-0010){ref-type="ref"}, [13](#clc23343-bib-0013){ref-type="ref"}, [14](#clc23343-bib-0014){ref-type="ref"}, [51](#clc23343-bib-0051){ref-type="ref"}, [52](#clc23343-bib-0052){ref-type="ref"}, [53](#clc23343-bib-0053){ref-type="ref"} and may be as high as 6.8 per 100 000 person years among adolescent athletes.[54](#clc23343-bib-0054){ref-type="ref"} Maron et al demonstrated that HCM was the leading cause of SCD in US athletes accounting for up to 36% of deaths.[13](#clc23343-bib-0013){ref-type="ref"}, [14](#clc23343-bib-0014){ref-type="ref"}, [15](#clc23343-bib-0015){ref-type="ref"} In a prospective cohort study of SCD in young individuals in the Veneto region of Italy the incidence of SCD was 2.3 per 100 000 per year in athletes compared to 0.9 per 100 000 per year in nonathletes. The higher risk of sudden death in athletes was attributable to underlying cardiovascular conditions, therefore suggesting a 2.8‐fold increased risk in athletes with a predisposing cardiac pathology.[10](#clc23343-bib-0010){ref-type="ref"} These studies have largely formed the basis for conservative exercise recommendations in individuals harboring serious cardiac conditions such as HCM. The 2005 ESC guidelines[16](#clc23343-bib-0016){ref-type="ref"} and the 2015 AHA guidelines[17](#clc23343-bib-0017){ref-type="ref"} restrict those individuals with a definite diagnosis of HCM from competitive sport with the exception of low intensity sports (bowling, billiards, curling, riflery, and golf). Although the same scientific bodies approach the issue of participation of HCM patients in recreational sport in a more moderate way, the guidance regarding the intensity of exercise is vague.[18](#clc23343-bib-0018){ref-type="ref"} 1.2. The impact of conservatism {#clc23343-sec-0003} ------------------------------- Results from the Italian data[55](#clc23343-bib-0055){ref-type="ref"} have been presumptively extrapolated to all ICC and all levels of exercise. As a result, physicians often encourage patients to adopt a sedentary lifestyle or provide vague and impractical advice with regards to exercise.[21](#clc23343-bib-0021){ref-type="ref"}, [23](#clc23343-bib-0023){ref-type="ref"} This practice is compounded by the paucity of data relating to the beneficial effects in this subset of patients with ICC and most importantly, the safety of a structured exercise program. Presently, 55% of patients fail to meet the minimum physical activity recommendations.[19](#clc23343-bib-0019){ref-type="ref"}, [20](#clc23343-bib-0020){ref-type="ref"} A sedentary lifestyle cultivates obesity and increases cardiovascular risk. In large cohort studies, up to 70% of individuals with HCM were found to be preobese or obese.[26](#clc23343-bib-0026){ref-type="ref"} Obesity in HCM leads to a significant increase in left ventricular mass, obstructive physiology, worsening heart failure symptoms as well as an increased likelihood of atrial fibrillation when compared to nonobese individuals.[25](#clc23343-bib-0025){ref-type="ref"}, [26](#clc23343-bib-0026){ref-type="ref"} Moreover, a sedentary lifestyle promotes the development of additional risk factors for atherosclerosis and coronary artery disease (CAD), both associated with excess mortality. In a cohort of 425 patients with HCM, hypertension was present in 58% of black patients and 32% of white patients, and was an independent risk factor (HR 2.02, 95% CI 1.05 to 3.88, *P* = .036) for a composite outcome of cardiovascular death, cardiac arrest or appropriate ICD therapy.[56](#clc23343-bib-0056){ref-type="ref"} Sorajja et al demonstrated that individuals with HCM and severe CAD were at greater risk of death, compared to those with mild to moderate or no CAD, and their risk far exceeded historical death rates of CAD patients with normal left ventricular function.[24](#clc23343-bib-0024){ref-type="ref"} Additionally, the psychological impact of a diagnosis of HCM should not be underestimated. In both competitive, as well as recreational sportsmen and women, sport restrictions can lead to anxiety and significantly worsen social functioning and stress management.[19](#clc23343-bib-0019){ref-type="ref"}, [21](#clc23343-bib-0021){ref-type="ref"}, [22](#clc23343-bib-0022){ref-type="ref"}, [23](#clc23343-bib-0023){ref-type="ref"} 1.3. Are we overestimating the risk? {#clc23343-sec-0004} ------------------------------------ The risk of SCD during exercise in HCM may not be as high as initially thought. Although, Maron et al cited HCM as the most common cause of cardiovascular death in athletes,[13](#clc23343-bib-0013){ref-type="ref"}, [14](#clc23343-bib-0014){ref-type="ref"}, [15](#clc23343-bib-0015){ref-type="ref"} other studies have shown contrary evidence. Contemporaneous reports from Italy demonstrated that coronary artery anomalies, arrhythmogenic right ventricular cardiomyopathy (ARVC), and premature cardiovascular disease were the most common causes of SCD.[10](#clc23343-bib-0010){ref-type="ref"} In a study of consecutive cases of athletes referred to a national center for cardiovascular pathology[43](#clc23343-bib-0043){ref-type="ref"} sudden arrhythmic death syndrome was the most common cause of death, in 42% of cases, while HCM contributed to only 6% of the deaths. Similar findings have been reported in unselected cohorts of U.S. collegiate athletes and young sudden deaths in Australia.[44](#clc23343-bib-0044){ref-type="ref"}, [53](#clc23343-bib-0053){ref-type="ref"} In addition, the association between HCM and SCD during exertion is based predominantly on circumstantial evidence and contrasts to the mounting evidence of adverse outcomes in individuals with ARVC/carriers who participate in regular moderate or intense exercise.[57](#clc23343-bib-0057){ref-type="ref"}, [58](#clc23343-bib-0058){ref-type="ref"} One, however, should exercise caution as we lack risk stratification protocols for HCM in athletes and extrapolating data from predominantly sedentary cohorts may underestimate the risk. In a large study of adolescent football players who underwent cardiac screening, three of the eight deaths were due to HCM, two of whom had been identified during screening and continued to exercise.[54](#clc23343-bib-0054){ref-type="ref"} 1.4. Potential beneficial effects of exercise in hypertrophy cardiomyopathy {#clc23343-sec-0005} --------------------------------------------------------------------------- Physical inactivity may exacerbate pathological processes already limiting exercise capacity. Exercise intolerance in individuals with HCM is mediated by a number of factors. These include a predisposition to arrhythmias including atrial fibrillation (AF)[59](#clc23343-bib-0059){ref-type="ref"}, [60](#clc23343-bib-0060){ref-type="ref"} and exercise‐induced arrhythmias.[61](#clc23343-bib-0061){ref-type="ref"} Pathological structural and vascular changes in the heart may give rise to myocardial ischaemia,[62](#clc23343-bib-0062){ref-type="ref"} diastolic dysfunction[63](#clc23343-bib-0063){ref-type="ref"}, [64](#clc23343-bib-0064){ref-type="ref"} and left ventricular outflow tract obstruction.[65](#clc23343-bib-0065){ref-type="ref"} Individuals may demonstrate an inability to respond to increases in activity due to an abnormal blood pressure response[66](#clc23343-bib-0066){ref-type="ref"} and/or chronotropic incompetence.[67](#clc23343-bib-0067){ref-type="ref"} Peripheral deconditioning is also a major contributor to reduced exercise tolerance. Exercise may help offset some of these limiting factors. In patients with HF, improvements in exercise capacity have been attributed to peripheral adaptation, including reduction in endothelial dysfunction in the skeletal muscle vasculature as well as increased oxidative capacity.[68](#clc23343-bib-0068){ref-type="ref"}, [69](#clc23343-bib-0069){ref-type="ref"} Exercise may also increase stroke volume and favorably remodel cardiac dimensions.[70](#clc23343-bib-0070){ref-type="ref"} Exercise interventions have demonstrated beneficial outcomes in patients with HF[71](#clc23343-bib-0071){ref-type="ref"}, [72](#clc23343-bib-0072){ref-type="ref"} reducing mortality and hospital admissions. Exercise may also exert a beneficial effect on diastolic filling and improve exercise capacity in patients with HF with preserved ejection fraction.[72](#clc23343-bib-0072){ref-type="ref"} The effects on diastology suggest that exercise may benefit individuals with established HCM in whom diastolic dysfunction is a significant contributor to exercise limitation. This notion is supported by a study in 106 athletes with HCM which showed that athletes exhibited normal or supranormal indices of diastolic function when compared to sedentary patients.[47](#clc23343-bib-0047){ref-type="ref"} Moreover, experimental studies in murine models of HCM demonstrate that exercise may halt or even reverse the cardiomyopathic process.[48](#clc23343-bib-0048){ref-type="ref"} Mice with HCM initiated exercise, in a voluntary cage wheel, at either 2 or 6 months. Pathological evaluation at 8 months of those who began exercising prior to disease expression, demonstrated inhibition of myocardial fibrosis and disarray, whereas older mice showed only reversal of disarray. Exercise also prevented the induction of hypertrophic markers and favorably affected apoptotic pathways in both groups. Notably, there was no difference in mortality between mice with HCM and non‐transgenic mice. 1.5. Cardiac rehabilitation in the general HCM population {#clc23343-sec-0006} --------------------------------------------------------- Two studies in HCM patients suggest that exercise may improve fitness without a concomitant increase in arrhythmic burden. In a prospective nonrandomized study of 20 symptomatic patients (mean age 62 years), exercise intensity was increased from 50% to 85% of an individual\'s heart rate reserve (HRR). Functional capacity, assessed by a graded exercise test, improved from 4.7 to 7.2 METS (*P* = .01). NYHA functional class also improved from baseline by ≥1 grade in 10 patients (50%).[45](#clc23343-bib-0045){ref-type="ref"} The most recent randomized controlled trial, RESET‐HCM,[46](#clc23343-bib-0046){ref-type="ref"} assigned patients (mean age 50 years) to 16 weeks of an unsupervised home‐based moderate‐intensity exercise training program (*n* = 67) or usual activity (*n* = 69). Exercise intensity in the intervention group increased from 60% to 70% of the HRR. Mean peak oxygen consumption improved significantly (*P* = .02) in the exercise training group. There were no adverse events in either study. These studies demonstrate that exercise benefits the cardiovascular fitness of HCM patients and may be safer than initially anticipated. It is important to note that these studies were not powered for safety. In the RESET‐HCM trial, although there was no death, aborted SCD, ICD shocks, or sustained ventricular tachycardia (VT), the authors reported that several participants experienced symptomatic nonsustained VT (NSVT). One patient who had previously had NSVT, experienced a further 30 s before starting the exercise program, while another patient (who had an ICD) demonstrated an increase in burden of symptomatic NSVT. A third patient experienced NSVT de novo 1 hour after exercising. All three individuals were withdrawn from the study. Although occurrence of nonfatal arrhythmias was not significantly different between groups, these episodes of symptomatic arrhythmia warranted ICD implantation, and raised concerns of further adverse events with continuation of exercise. With this in mind, careful consideration should be given to an individual\'s SCD risk score prior to provision of exercise recommendations. This also highlights the importance of careful monitoring of these individuals. Further work is needed in larger cohorts and younger age groups with the adoption of more vigorous exercise regimes. 1.6. Athletes with HCM {#clc23343-sec-0007} ---------------------- The recent position statement from the Sport Cardiology Section of EAPC recommends that participation in intense exercise or competitive sport may be considered after comprehensive clinical evaluation in asymptomatic adults, with low ESC‐SCD risk score and a mild HCM phenotype. This excludes those sports where syncope may cause harm or death.[50](#clc23343-bib-0050){ref-type="ref"} These changing opinions are particularly pertinent to those individuals who continue to compete or exercise above recommendations provided by their physicians.[73](#clc23343-bib-0073){ref-type="ref"} Data supporting the change in recommendations stem from the clinical profiling of athletes with HCM who are capable of competing at an elite level. Sheikh et al compared 106 elite athletes with HCM to 101 sedentary individuals with HCM. Athletes demonstrated a milder phenotype with regard to LVH, and a third demonstrated the apical variant of HCM, indicating a lower risk of SCD.[47](#clc23343-bib-0047){ref-type="ref"} Retrospective data from competitive athletes with HCM have also shown that higher levels of exercise may not significantly increase the risk of adverse outcomes. Results from a multinational, prospective ICD registry of 440 athletes with cardiovascular disease (17% with HCM), participating in high risk and organized sports, demonstrated no deaths related to arrhythmia, no aborted SCD or injury following syncope or shock during sport, during 44 months follow‐up period.[74](#clc23343-bib-0074){ref-type="ref"} In the original registry athletes with HCM demonstrated the lowest number of VT/VF shocks during competition/practice *n* = 1 (2%), in clear contrast to ARVC which was associated with increased risk of ICD shock during competition or practice.[75](#clc23343-bib-0075){ref-type="ref"} Pelliccia et al[76](#clc23343-bib-0076){ref-type="ref"} compared 15 athletes with HCM who continued to engage in regular exercise or competitive sport to athletes who had suspended exercise. Athletes had been engaged in sporting activities for a mean of 15 ± 8 years at a level equivalent to regional or above. Only two of these individuals were deemed to be high risk according to the ESC SCD risk calculator, two were intermediate risk and the remainder were low risk. None of these individuals had an ICD. Over the course of 9 ± 6 years of follow‐up only an amateur tennis player suffered a cardiac arrest while walking. In terms of symptoms seven individuals experienced these, the most significant being syncope (*n* = 3). The event and symptom rate were not significantly different between the two groups. Although Pelliccia\'s group[76](#clc23343-bib-0076){ref-type="ref"} present optimistic findings there are several issues to be mindful of. We know that the risk of SCD is influenced by an athlete\'s demographics and sporting discipline. Higher risk individuals include males, adolescents, black athletes, and those competing in high intensity start‐stop sports such as football and basketball.[13](#clc23343-bib-0013){ref-type="ref"}, [14](#clc23343-bib-0014){ref-type="ref"}, [15](#clc23343-bib-0015){ref-type="ref"}, [54](#clc23343-bib-0054){ref-type="ref"} In this study, the majority of these athletes were male and Caucasian, therefore the results do not account for ethnic variation in SCD. The sample size was small, with relatively short follow‐up and therefore not powered to detect a significant difference in outcomes, particularly given the low‐event rate of SCD. Additionally, the population may not reflect the consideration that should be given to risk according to age. This was highlighted in a recent study of outcomes in competitive adolescent football players who underwent cardiac screening.[54](#clc23343-bib-0054){ref-type="ref"} Of the eight SCDs recorded, three were due to HCM, of which two had been identified and were advised not to continue playing competitively. These findings highlight that estimation of the risk of an event is not an easy task. Although the great majority of individuals with HCM are expected to have a normal lifespan[41](#clc23343-bib-0041){ref-type="ref"}, [42](#clc23343-bib-0042){ref-type="ref"} the risk of SCD appears to vary with age. The highest risk is conferred to young individuals,[77](#clc23343-bib-0077){ref-type="ref"}, [78](#clc23343-bib-0078){ref-type="ref"}, [79](#clc23343-bib-0079){ref-type="ref"} with a notable decrease in risk above the age of 60 years.[80](#clc23343-bib-0080){ref-type="ref"} In the absence of previous hemodynamically compromising VT or ventricular fibrillation, an individual\'s risk may be calculated using the ESC‐SCD risk score calculator.[12](#clc23343-bib-0012){ref-type="ref"} However, the timing of adverse events is not predictable. The heterogeneity of outcomes is highlighted in studies of ICD therapies. Maron et al[81](#clc23343-bib-0081){ref-type="ref"} demonstrated that the interval between implantation and discharge varied widely, even by up to a decade.[82](#clc23343-bib-0082){ref-type="ref"} In secondary prevention cases, individuals may not experience further events for up to 30 years.[83](#clc23343-bib-0083){ref-type="ref"} Longer term follow‐up is required and most importantly considerations such as age, gender, ethnicity, sporting discipline, and ESC‐SCD risk score should be factored into advice regarding an individual\'s exercise recommendations. 1.7. Current practice and future perspectives {#clc23343-sec-0008} --------------------------------------------- Based on the best available evidence, it seems reasonable to offer all patients with HCM a comprehensive evaluation (Figure [2](#clc23343-fig-0002){ref-type="fig"}). This will allow for a more individualized approach guided by the patient\'s risk, symptoms, and baseline fitness. The exercise prescription should be specific and abide by the "FITT" principle (frequency, intensity, time \[duration\], type of exercise).[84](#clc23343-bib-0084){ref-type="ref"} Unless significant concerns exist, the exercise prescription should satisfy current WHO recommendations for physical activity.[85](#clc23343-bib-0085){ref-type="ref"} Maximal intensity should not exceed 70% of HRR, which approximately equates to 80% of maximal predicted HR or 14 to 16 of the Borg scale. Individuals at high risk of SCD, limited by symptoms or with low baseline fitness should be optimized on medical therapy and commenced at a lower exercise level which could then be gradually increased based on their response. It seems reasonable that for individuals who would qualify for competitive sport, based on the Sport Cardiology Section of EAPC recommendations,[42](#clc23343-bib-0042){ref-type="ref"} not to restrict the intensity level of recreational exercise. For competitive athletes, recommendations should be guided by the Sport Cardiology Section of EAPC recommendations.[42](#clc23343-bib-0042){ref-type="ref"} Ultimately, decisions relating to exercise prescription should take into consideration the wishes of a well‐informed patient. {#clc23343-fig-0002} Future research relating to competitive athletes should focus on large, multicenter registries as the option of randomized studies does not seem pragmatic. Large, randomized, multicenter studies, however, are possible in the general HCM population and should explore benefits and safety of different exercise regimes including higher intensity programs and high intensity interval training, which are likely to be more appealing to the expanding population of younger patients. 2. CONCLUSION {#clc23343-sec-0009} ============= It is clear that we are in the early stages of gathering data on safety of exercise in HCM, but the advances in our understanding of the condition and its relation to exercise as a trigger of fatal arrhythmias should not be ignored. Previous consensus guidelines regarding exercise prescription were based on historical data and were overly restrictive. The new guidelines propose a less restrictive approach, under specific circumstances and after detailed discussion with the athlete and other stakeholders. Although this is a welcomed perspective, one must remain mindful that adequately powered, large cohort studies with long‐term follow‐up are currently lacking. Therefore, at present, an individualized rather than a one‐size fits all approach is preferable. CONFLICT OF INTEREST {#clc23343-sec-0010} ==================== The authors declare no potential conflict of interests.

{#sp1 .573} {#sp2 .574} {#sp3 .575} {#sp4 .576}

{.257} {.258}

Introduction ============ The 2011 IEEE International Workshop on Genomic Signal Processing and Statistics (GENSIPS 2011) was organized in San Antonio, Texas from December 2-4th. GENSIPS\'11 provided a forum for researchers in the signal processing community, other related computational experts, and biomedical scientists to exchange ideas and discuss signal processing challenges due to the high modality of disparate high-throughput data, high variability of the data acquisition, high dimensionality of data, and high complexity of genomics and proteomics systems. The theme of GENSIPS\'11 was cancer and computational biology and GENSIPS featured prominent plenary speakers including Dr. John N. Weinstein from UT MD Anderson Cancer Center, Dr. Stephen Wong from Methodist Hospital at Cornell University, Dr. David Nelson from Baylor College of Medicine and Dr. Chung-I Wu from the Beijing Institute of Genomics. Articles ======== This supplement contains extended versions of selected articles from GENSIPS conference proceedings. Each submitted article to the conference was reviewed by a minimum of two reviewers and the top twenty favorably reviewed papers from sixty selected papers (33%) were invited to submit the extended versions for this supplement. The extended journal versions were further reviewed according to rigorous peer-review criteria. The accepted articles can be broadly categorized into four groups: Gene network inference and cancer therapy design ------------------------------------------------ Models of genetic regulatory networks can belong to the class of deterministic or stochastic, discrete or continuous, fine-scale or coarse-scale quantitative models. The availability of experimental data, prior biological knowledge and purpose of modeling frequently guide the selection of the quantitative model to represent the biological network. Sridharan *et al.*\[[@B1]\] provides an approach for Boolean modeling and analysis of the oxidative stress response pathway based on prior biological knowledge. Oxidative stress has been implicated in a variety of diseases including, but not limited to aging and age-related diseases such as cancer, cardiovascular disease, chronic inflammation, and neurodegenerative disorders. Lin and Khatri \[[@B2]\] considers the aberrant behavior of signaling pathways during cancer as faults in an electronic circuit and presents a Max-SAT based automatic test pattern generation (ATPG) algorithm for cancer therapy. In Haider and Pal \[[@B3]\], a Boolean Network inference algorithm from limited time series data utilizing prior biological knowledge on connectivity is presented. The algorithm is validated on synthetic data and experimental transcriptomic measurements from Human Mammary Epithelial Cell line. As compared to two existing Boolean network inference approaches, the proposed algorithm performs better in terms of robustness and estimation of state transitions. Wu *et al.*\[[@B4]\] provides a linear regression approach to model transcription regulation following estrogen stimulation in breast cancer cells. The proposed technique is validated using gene expression and ER*α*Chip-seq data from the MCF-7 cell line. Fine-scale modeling of genetic regulatory networks using stochastic master equation models entails a huge computational complexity and Karim *et al.*\[[@B5]\] presents a computationally inexpensive way to generate the steady state distributions for stochastic master equation models. In Wang *et al.*\[[@B6]\], a quantitative mathematical model to predict macrophage activation patterns following myocardial infarction is reported. The model was validated on experimental data from adult C57 mice. A single type of genomic data is typically not suitable to understand regulation of cell behavior. Vicente *et al.*\[[@B7]\] presents an approach to assess the gain in predictive performance by integrating various types of biological information in network inference. Genomic data analysis --------------------- For analysis of DNA methylation profiles, Zhang *et al.*\[[@B8]\] presents a non-parametric infinite beta mixture model to cluster DNA methylation expression profiles produced by Illumina Infinium Beadchip. For genome-wide association studies (GWAS), Jia and Zhao \[[@B9]\] applies a dense module search algorithm for locating network modules that are jointly associated with a disease. A restricted search approach is applied for reducing the computational complexity and the strategy is demonstrated on CATIE GWAS dataset for schizophrenia. Taslim *et al.*\[[@B10]\] proposes a quantitative approach using mixture models to characterize patterns of promoter regions and predict novel and alternative promoters. Jahid and Ruan \[[@B11]\] proposes a method for biomarker discovery by combining microarray gene expression profiles and protein-protein interaction networks. Prediction of drug effectiveness -------------------------------- Targeted cancer therapy is considered to be a cornerstone for personalized medicine. Research problems in this area include understanding and modeling the mechanisms of action of molecularly targeted drugs and design of combination drug therapies. Li *et al.*\[[@B12]\] presents an integrated experimental and theoretical approach to investigate the mechanism of action and identify pharmacodynamic characteristics of targeted agents based on cell-line platforms. Specifically, tumor cell response is analyzed via the use of fluorescent reporters; dynamics of drug efficacy for different dosages are studied using dynamic modeling; and time-varying parameters are estimated using system identification techniques. Lin *et al.*\[[@B2]\] proposes to predict the ffectiveness of targeted drugs and guide drug selection by modeling the cancer pathway as a Boolean circuit. Kim and Yoon \[[@B13]\] presents an adaptive reference update (ARU) algorithm to search for the optimal drug combination by comparing the response of the current combination against that of the reference combination and beneficially updating the drug concentrations. Application to real and synthetic examples shows that the ARU algorithm outperforms existing algorithms in terms of effectiveness, efficiency, and robustness. Analysis workflows ------------------ Advancement in measurement technologies provides vast quantities of experimental data on various components of the regulome. This requires new approaches for the systematic generation and analysis of the experimental data. For large-scale protein profiling, Sun *et al.*\[[@B14]\] provides an integrative model for Liquid Chromatography coupled Mass Spectrometry, and apply it to the systematic analysis of key factors that impact the number of identified peptides and quantified proteins, protein quantification error, differential expression results, and classification performance. The presence of extraneous variables caused by the sample variability can significantly affect the statistical analysis of high-throughput genetic data. Hsu *et al.*\[[@B15]\] proposes a novel approach using batch effect correction, a sample selection process, and a semi-supervised clustering method for reducing the confounding and suppression effects induced by the extraneous variables. Rodriguez *et al.*\[[@B16]\] presents an analysis workflow that permits researchers to compare DNA methylation profiles across multiple biological or patient groups. Comparisons can be made at particular regions of a chromosome or across the genome as a whole. This workflow and its suite of features can assist biomedical scientists in conducting methylation profiling projects and facilitate meaningful biological interpretation of their data. Characterizing copy-number variation is a basic method to profile normal and diseased tissue samples, but challenges remain in accurately interpreting the data from a single genome and comparative measurements from groups of sequenced genomes. Janevski *et al.*\[[@B17]\] studies several variations of copy-number analysis approaches to assess the significance and impact of each methodology choice. Huang *et al.*\[[@B18]\] proposes C2Maps platform as an online bioinformatics resource providing biologists with directional relationships between drugs and genes/proteins in specific disease contexts based on network mining, literature mining, and drug effect annotations. Doderer *et al.*\[[@B19]\] analyzes pathway consolidation approaches and provides a user-friendly web-accessible tool that can enable users to extract functional relations of genes across multiple pathway databases. A case study of performing personal genome analysis on a cloud computing environment is presented in Evani *et al.*\[[@B20]\]. The approach can assist researchers in applying existing cloud computing technologies to analyze enormous amount of data generated by next-generation sequencing technologies. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= All authors served as editors for the supplement, with RP serving as the Lead Editor. All authors helped write this editorial. Acknowledgements ================ This supplement will not have been possible without the support of the reviewers and program committee members of IEEE Gensips\'2011. This article has been published as part of *BMC Genomics*Volume 13 Supplement 6, 2012: Selected articles from the IEEE International Workshop on Genomic Signal Processing and Statistics (GENSIPS) 2011. The full contents of the supplement are available online at <http://www.biomedcentral.com/bmcgenomics/supplements/13/S6>.

Introduction ============ Rice stripe disease is the most devastating viral disease of rice in China, Japan, and Korea (Wei et al., [@B39]). The causal agent, *Rice stripe virus* (RSV), is the type member of the *Tenuivirus* genus and the viral genome consists of four single-stranded RNA segments (RNAs 1, 2, 3, and 4; Hibino, [@B14]). RNA 1 is negative-sense and encodes a putative RNA-dependent RNA polymerase. RNAs 2, 3, and 4 are ambisense, and each of which encodes two open reading frames (ORFs) with one on viral RNA (vRNA) and another on viral complementary RNA (vcRNA). RSV vRNA 2 encodes a membrane-associated protein that reportedly is an RNA silencing suppressor and interacts with SGS3 (Du et al., [@B7]). The vcRNA 2 encodes a glycoprotein with unidentified functions (Zhao et al., [@B47]). The vRNA 3 and vcRNA 3 encode a gene silencing suppressor and a nucleocapsid (NC) protein, respectively (Hibino, [@B14]; Xiong et al., [@B42]). RSV vRNA 4 encodes a disease-specific protein that accumulates in both infected plant and insect cells (Toriyama, [@B33]). The protein encoded by vcRNA 4 was identified as the RSV movement protein (MP; Xiong et al., [@B41]). RSV is transovarially transmitted by small brown planthopper (SBPH), *Laodelphax striatellus*, in a circulative-propagative manner (Falk and Tsai, [@B8]; Li et al., [@B19]). After RSV infection, rice plants often show chlorotic stripes in the newly expanded leaves, and the stripes progress into pale streaks in infected plant leaves. Because of global environment changes and the extensive increases in distribution of the transmission vector (*L. striatellus*) in the south and southeastern parts of China, RSV has caused significant losses in rice production in the past decade. To infect a host plant successfully, viruses must overcome two obstacles; they must be capable of replicating in host cells and moving between cells and then be able to move systemically throughout the plant via the vasculature. To carry out these functions, viruses encode MPs that often interact with viral genomic (g) RNAs to form ribonucleoprotein complexes that mediate intra- and inter-cellular movement. At the plasmodesmata (PD), the MPs modify PD size exclusion limits to enable transit of the ribonucleoprotein complexes to adjacent cells. In some examples, the viral MPs form tubules that penetrate through the PD and serve as conduits for whole virus cell-to-cell transport. It has been shown that viruses often co-opt plant cellular processes to carry out specific functions required for infection (Scholthof, [@B27]; Shen et al., [@B28]). In addition to host factors that interact directly with viral MPs (Paape et al., [@B22]; Shimizu et al., [@B29]), host cytoskeleton, and endoplasmic reticulum (ER) networks also play critical roles in virus movement in hosts (Ashby et al., [@B2]; Harries et al., [@B12]). For example, both microtubules and microfilaments have been implicated in supporting cell-to-cell movement of *Tobacco mosaic virus* (TMV) in *Nicotiana benthamiana* (Brandner et al., [@B5]; Harries et al., [@B11], [@B12]). The MP of *Abutilon mosaic virus* (AbMV) is known to have an anchor domain that allows the MP to localize to the ER (Aberl et al., [@B1]). Association of viral MPs with the secretory pathway was also reported for viruses whose MPs form tubules. For example in cells infected with *Cowpea mosaic virus* or *Cauliflower mosaic virus*, tubule formation was independent of microtubules or microfilaments, but tubule formation required a functional secretory pathway (Huang et al., [@B16]; Pouwels et al., [@B23]). Trafficking of P3N-PIPO and CI of *Turnip mosaic virus* (TuMV) to PD has also been shown to be dependent on the host secretory pathway (Wei et al., [@B38]). Interestingly, in *Grapevine fanleaf virus* infected cells both the secretory pathway and the cytoskeleton networks were reported to be involved in tubule formation and in intra-cellular targeting of virions (Laporte et al., [@B18]). Thus, plant virus may utilize the host cytoskeleton, the ER network, or both for PD targeting. Genomes of plant viruses are small and each virus encodes only a few proteins. Consequently, virus-encoded proteins are often multi--functional proteins. For example, the coat protein of *Turnip crinkle virus* (TCV) not only functions in movement between cells and in virion assembly, but also functions as a suppressor of gene silencing (Qu et al., [@B25]; Cao et al., [@B6]). Viral MPs also have varied functions: BC1 of AbMV accumulates preferentially at the cell periphery or around the nucleus in plant cells, and hence may participate in distinct functions (Zhang et al., [@B46], [@B45]). The *Barley stripe mosaic virus*-encoded triple-gene block (TGB) 1 protein has similar localization patterns, and TGB2 can localize to both ER membranes and chloroplasts, indicating it also has distinct functions (Torrance et al., [@B34]; Lim et al., [@B20]). Several other viral MPs have been reported to accumulate in chloroplasts and are considered to have important roles in virus replication, viral transport, or symptom development. For example, mutation of the chloroplast-targeting signal in the *Alternanthera mosaic virus* (AltMV) TGB3 impaired the virus cell-to-cell movement and eliminated the long distance movement of the virus (Lim et al., [@B21]). A number of biochemical and subcellular localization activities are associated with the TGB proteins of other flexiviruses, including intra-cellular targeting, gene silencing activities, and host membrane remodeling (Verchot-Lubicz et al., [@B36]; Tilsner et al., [@B32]). The 66 K protein of Turnip yellow mosaic virus (TYMV) was reported to localize to virus-induced chloroplastic membrane vesicles, which are thought to function as TYMV RNA replication factories (Prod'homme et al., [@B24]). The TuMV 6 K also has been shown to target chloroplasts to result in aggregation and elicitation of membrane invaginations (Wei et al., [@B37]). Former work demonstrated that NSvc4 rely on the early secretory pathway and actin-myosin VIII motility system for plasmodesmatal localization and could induce foliar necrosis from a TMV-NSvc4 hybrid vector (Yuan et al., [@B43]; Zhang et al., [@B44]). Here we present new evidence indicating that NSvc4 exerts its movement functions by trafficking on actin filaments and ER networks to reach the PD and shown that the N-terminal 125 amino acids (AAs) determine the PD localization. We also demonstrate that the NSvc4 protein targets chloroplasts in infected cells and is a symptom determinant in plant. Materials and Methods ===================== Plasmids constructions ---------------------- The full length ORFs of NSvc4 protein and the N- and C-terminal deletion mutants were amplified from pBin438-NSvc4 (Xiong et al., [@B41]) using the Phusion High-Fidelity DNA polymerase (New England Biolabs, Ipswich, USA). The NSvc4 deletion mutant (lacking AAs 106--125) was first amplified via an overlap PCR method with the primers MP-Fol and MP-Rol (See Table [A1](#TA1){ref-type="table"} in Appendix for all the primers used in this study). The resulting PCR fragments were ligated individually into the pCHF3-eGFP plasmid and used for agroinfiltration into *N. benthamiana* (Xiong et al., [@B41]). To construct Potato virus X (PVX) NSvc4 expression vectors, full length and deletion mutants of NSvc4 were PCR amplified with primers containing *Cla*I and *Sal*I restriction sites. The PCR fragments were cloned individually into the pGEM-Teasy vector. After digestion using the *Cla*I and *Sal*I enzymes, the resulting fragments were ligated individually to the PVX pgR107 vector (provided by Dr D. C. Baulcombe, Sainsbury Laboratory, John Innes Centre, Norwich Research Park, Norwich, UK). All the plasmids were verified by DNA sequencing before further use. Plant inoculation and confocal microscopy ----------------------------------------- *N. benthamiana* plants were grown in a growth chamber set at 25 ± 1°C and 16 h light and 8 h dark conditions. RSV infectivity trials were carried out by rub-inoculating leaves with crude extracts from RSV-infected *O. sativa* leaves ground in phosphate buffer (0.2 M). After a 12-h incubation in the dark, the plant were transferred to a culture room set at 25 ± 1°C, 80% relative humidity, and 16 h light and 8 h dark cycle. Local and systemic leaf infections were evaluated at 3, 7, and 10 days post inoculation by RT-RCR (data not shown). Leaves of 4-week-old plants were infiltrated with *Agrobacterium tumefaciens* (strain GV3101) harboring either the full length NSvc4 sequence or one of the mutant NSvc4 plasmids using needleless syringes as described previously (Batoko et al., [@B3]). Leaf tissue was harvested at 48 h post agro-infiltration and examined for GFP fluorescence under a Leica TCS SP5 confocal microscope equipped with a 20× objective lens. Conditions set to excite GFP and monitor the emission were as described by Brandizzi et al. ([@B4]). Chloroplast autofluorescence was detected using a 670-nm emission filter according to the manufacturer's instructions. Confocal images were processed using the LCS Lite Leica software. Inhibitor treatments -------------------- Latrunculin B (LatB), oryzalin, and brefeldin A (BFA) were purchased from Sigma-Aldrich (St. Louis, USA) and dissolved in dimethyl sulfoxide (DMSO) to make stock solutions at 10 mM, 2 mM, and 200 μg/ml, respectively. Immediately prior to use, the stocks were diluted to 5 μM LatB, 50 μM oryzalin, and 50 μg/ml BFA using double-distilled water (ddH~2~O). Three hours before agroinfiltration, diluted LatB, oryzalin, or BFA solutions were infiltrated into *N. benthamiana* leaves using needleless syringes as described (Harries et al., [@B10]). Diluted DMSO (1:1000 in ddH~2~O) was infiltrated into *N. benthamiana* leaves and used as a control. The MAN1-RFP (from soybean, which is known to localize to cis-Golgi) was used to monitoring BFA in function in our system (data not shown). For virus inoculation assays, leaves of six-to-eight leaf stage *N. benthamiana* were rub-inoculated with 5 μM LatB, 50 μM oryzalin, or diluted DMSO (1:1000 in ddH~2~O). One day after the chemical treatments, the leaves were rub-inoculated as described previously with crude extracts prepared from RSV-infected *O. sativa* leaves (Xiong et al., [@B41]). After 12 h incubation in the dark, the plants were transferred to a culture room set at 25 ± 1°C, 80% relative humidity, and 16 h light (5000 lux) and 8 h dark. Immunocytochemistry and electron microscopy ------------------------------------------- Small tissues (approximately 1 mm wide and 3 mm long) were excised from *N. benthamiana* leaves agroinfiltrated with the bacteria harboring the pgR107 or pgR107-NSvc4 vectors. Harvested tissues were fixed with 50 mM phosphate-buffered saline (PBS), pH 6.8, containing 1% glutaraldehyde and 2% formaldehyde for 3 h at 4°C. After dehydration in a graded series of ethanol (30, 50, 70, 90, and 100%), the fixed samples were embedded in Lowicryl K4M resin as described previously (Xiong et al., [@B41]). Construction of baculovirus plasmids and transfection of Sf-9 cells ------------------------------------------------------------------- The full length *NSvc4* sequence was PCR amplified from the pgR107-NSvc4 using the primers MP-(BamH1)-F and MP-(Sal1)-R. The amplified fragments were digested with the *Bam*HI and *Sal*I restriction enzymes, and then inserted between the *Bam*HI and *Sal*I sites within the pFastBacHTGFPT transfection vector (kindly supplied by Chuanxi Zhang, Zhejiang University, Zhejiang, China) under the control of the polh promoter. The recombinant plasmid pFastBacHTGFPT-NSvc4 was transformed into *E. coli* *DH10Bac* as instructed (Invitrogen, Carlsbad, USA). After transformation, the gene cassette from the recombinant plasmid was transferred to the bacmid genome by site-specific transposition and the recombinant bacmid DNA was then isolated following the manufacturer's instructions. Recombinant bacmid DNA was transfected into 1.0 × 10^6^ cells *Spodoptera frugiperda* 9 (Sf-9) cell using Cellfectin Reagent (Invitrogen, Carlsbad, USA), and transfected cells were incubated at 27°C for 72 h. Supernatant of culture medium TNM-FH (Sigma-Aldrich, USA) was collected from the transfected Sf-9 cell cultures and transferred to fresh Sf-9 cells followed by an additional 48--72 h incubation period before confocal microscopy observations. Computational analyses of the NSvc4 protein ------------------------------------------- The transmembrane helices of NSVc4 were predicted using the Membrane Protein Explorer (MPEx, version 3.2[^1^](#fn1){ref-type="fn"}) and the ΔG prediction server[^2^](#fn2){ref-type="fn"} (Hessa et al., [@B13]; Snider et al., [@B30]). The ΔG server provides predictions of the corresponding apparent free energy differences. In principle, a negative ΔG value by the Sec translocon predicts that a protein sequence has a TM helix and that proteins can integrate into membranes. Results ======= Actin filaments and Golgi apparatus disruption abolish PD localization of NSvc4 and delay RSV systemic infections in *N. benthamiana* ------------------------------------------------------------------------------------------------------------------------------------- To investigate roles of the cytoskeleton and secretory membranes in NSvc4 intra-cellular transport and PD targeting, three pharmacological inhibitors were applied to *N. benthamiana* leaves before agroinfiltration-mediated transient expression of NSvc4-eGFP. LatB and oryzalin treatments were used to disrupt actin filaments and microtubules, as previously reported (Harries et al., [@B10]; Yuan et al., [@B43]). Expression of NSvc4-eGFP under control of the PVX pgR107-NSvc4 vector in the DMSO (control) treated *N. benthamiana* leaves yielded punctate spots at the PD (Figure [1](#F1){ref-type="fig"}A). When NSvc4-eGFP was expressed in LatB treated *N. benthamiana* leaves, the number of punctate spots at the PD were clearly reduced, and fluorescence was more generally dispersed throughout the cell (Figure [1](#F1){ref-type="fig"}B), indicating that a functional actin cytoskeleton is important for targeting NSvc4-eGFP to punctate foci at the PDs. When the NSvc4-eGFP was expressed in the oryzalin treated *N. benthamiana* leaves, abundant punctate spots similar to those noted in the DMSO controls were evident at the PD (Figure [1](#F1){ref-type="fig"}C). These experiments indicate that depolymerizing microtubules does not have obvious interference on formation of punctate spot at the PD. BFA is known to interfere with the ER/Golgi secretory pathway by inhibiting COPI vesicle production (Tse et al., [@B35]). We therefore treated *N. benthamiana* leaves with BFA and noted the NSvc4-eGFP fluorescence was more generally distributed in the cytosol than in the DMSO treated controls and that the localization at the PD and the peripheral membranes was greated reduced (Figures [2](#F2){ref-type="fig"}A,B). These results thus suggest that an intact Golgi secretory system has a substantial positive effect on PD targeting of NSv4-eGFP. {#F1} {#F2} In order to determine whether the pharmacological affects were correlated with RSV infection, we inoculated *N. benthamiana* leaves with extracts from RSV-infected rice. We had noted earlier (Xiong et al., [@B41]) that RSV results in systemic infections in *N. benthamiana* after mechanical inoculation. Therefore to test the effects of the DMSO, LatB, and oryzalin treatments on RSV infection, we inoculated RSV from infected rice extracted to *N. benthamiana* leaves 1 day after application of the drug treatments. The results show that disruption of actin filaments using LatB strongly inhibits systemic infection of RSV, whereas oryzalin treatments were similar to those of DMSO on RSV systemic infection in *N. benthamiana* (Table [1](#T1){ref-type="table"}). ###### **Effect of different inhibitors on RSV infection in *N. benthamiana***. Treatment 3 dpi^a^ inoculation leaf 3 dpi systemic leaf 7 dpi systemic leaf 10 dpi systemic leaf ----------- --------------------------- --------------------- --------------------- ---------------------- LatB 15/15^b^ 0/15 3/15 3/15 Oryzalin 15/15 0/15 11/15 12/15 DMSO 15/15 0/15 13/15 15/15 *^a^Days post inoculation of RSV in *N. benthamiana*. ^b^ The denominator shows the number of *N. benthamiana* plants used in these treatment; the numerator represents the number of *N. benthamiana* plants with symptom*. Computational analysis of NSvc4 and domains responsible for NSvc4 PD localization --------------------------------------------------------------------------------- Using the Membrane Protein Explorer program, AAs spanning positions 106--125 of the NSvc4 protein have properties of a transmembrane domain (Figure [3](#F3){ref-type="fig"}). To confirm this prediction, we deleted AAs 106--125 from the NSvc4 to create NSvc4~Δ106--125~-eGFP, and expressed the mutant transiently from the PVX vector in the epidermal cells of *N. benthamiana* leaves via agro-infiltration. Confocal microscopy observations revealed that PD localization by the deletion mutant was substantially reduced compared to the DMSO controls. Only a few apparently intact foci were evident and most of the fluorescence was diffuse and appeared to be associated with the cytosol (Figure [2](#F2){ref-type="fig"}C). These observations provide evidence suggesting that the predicted transmembrane region (AAs 106--125) in the NSvc4 protein provides an important anchor domain that is required for NSvc4 trafficking on the endomembrane network. {#F3} To determine the domain responsible for NSvc4 PD localization, a series of NSvc4 deletion mutants were constructed, inserted into the pCHF3 vector and expressed transiently by agroinfiltration into *N. benthamiana* leaf cells. The fluorescence patterns in cells at 48 h after infiltration revealed that the NSvc4~1--54~-eGFP, NSVc4~1--73~-eGFP, NSvc4~1--106~-eGFP, and NSvc4~Δ125--286~-eGFP mutant derivatives each elicited GFP expression patterns similar to those produced by pCHF3-eGFP, the GFP control vector (Figure [4](#F4){ref-type="fig"}). However, fluorescence from the NSvc4~1--125~-eGFP deletion mutant protein accumulated in punctate foci at the PD that appeared to be similar to the fluorescence elicited in cells expressing NSvc4-eGFP (Compare Figures [4](#F4){ref-type="fig"}B,G). In marked contrast, the NSvc4~Δ\ 106--125~-eGFP proteins accumulated as small punctate bodies in the cytoplasm and were not observed at the PD (Figure [4](#F4){ref-type="fig"}E). These observations indicate that the N-terminal 1--125 AA fragment contains a PD localization signal that is sufficient for PD targeting. {#F4} Localization of NSvc4 in sphere-like compartments and chloroplast ----------------------------------------------------------------- Image analysis indicated that as well as localizing at the PD, the NSvc4 protein also accumulated in discrete, sphere-like compartments of approximately 4 mm in diameter in cells (Figure [5](#F5){ref-type="fig"}). To determine the subcellular localization of these spheres, epidermal cells expressing NSvc4-eGFP were analyzed by confocal microscope. A lambda scan set at 5 nm intervals between 595 and 755 nm for analysis of the sphere-like compartments had emission peaks at 500--530 nm and at 650--700 nm (Figure [5](#F5){ref-type="fig"}B). The spectral characteristics of the 650--700 nm emission peak were similar to the chlorophyll spectrum (maximum at 680 nm). So, the fluorescence spectra were collected simultaneously, with one photon multiplier-tube bandwidth set at 500--530 nm and a second one at 660--700 nm, and the NSvc4-eGFP and chlorophyll autofluorescence was merged to generate yellow fluorescent of the overlapping foci (Figure [5](#F5){ref-type="fig"}). {#F5} To confirm the presence of NSvc4 in the chloroplasts, NSvc4-eGFP was expressed in *N. benthamiana* leaf cells using the PVX vector (pgR107). By 7--8 days post agroinfiltration (dpi), leaves with systemic symptoms were sampled and examined by confocal microscopy. In cells expressing the NSvc4-eGFP, the GFP signal co-localized with the chlorophyll autofluorescence, suggesting that a fraction of the expressed NSvc4-eGFP protein accumulates in the chloroplasts (Figure [6](#F6){ref-type="fig"}). {#F6} To determine which region of NSvc4 is required for chloroplast-targeting, we agroinfiltrated plasmids harboring the wild type or mutant NSvc4-eGFP fusions (Figure [7](#F7){ref-type="fig"}). The results showed that NSvc4~1--73~-eGFP accumulated in the sphere-like compartments and in the chloroplasts of the epidermal leaf cells (Figure [7](#F7){ref-type="fig"}B). However, NSvc4~1--54~-eGFP and NSvc4~54--73~-eGFP localized around the nuclei and in the cytoplasm, but were not obvious in the chloroplasts (Figures [7](#F7){ref-type="fig"}A,E). The remaining NSvc4 mutants localized exclusively in the cytoplasm. These observations suggest that the N-terminal 73 AAs contain a chloroplast-targeting signal. {#F7} The NSvc4 PVX vector induces more severe symptoms in *N. benthamiana* than PVX ------------------------------------------------------------------------------ By 7 days after agroinfiltration of *N. benthamiana* plants for expression of the wtPVX vector (pgR107) or the NSvc4 (pgR-NSvc4), all plants developed systemic symptoms in the upper emerging leaves. Symptoms in plants infiltrated with pgR-NSvc4 were more severe than in plants infiltrated with the wtPVX vector. By 20 dpi, virus symptoms in plants infected with wtPVX vector disappeared, whereas symptoms in the PVX-NSvc4 infected plants remained intense and developed foliar necrosis (Figure [8](#F8){ref-type="fig"}). Reverse transcription PCR result showed that NSvc4 was accumulated in leaves of the PVX-NSvc4 infected plants in both the early and the late infection stages (data not shown). Examination of thin sections prepared from the PVX or PVX-NSvc4 infected *N. benthamiana* leaf tissues by electron microscopy revealed major malformations of chloroplast grana and electron lucent bodies beneath the membranes of PVX-NSvc4 infected cells, but similar malformations were not evident in wtPVX infected cells. In addition, proliferations radiating from the chloroplasts into the cytoplasm were observed in the PVX-NSvc4 and RSV rub-inoculating infected *N. benthamiana* leaves, but not in the cells infected with wtPVX (Figure [9](#F9){ref-type="fig"} and Figure [A1](#FA1){ref-type="fig"} in Appendix). {#F8} {#F9} Symptom development is independent of NSvc4 chloroplast localization -------------------------------------------------------------------- To determine the correlation between NSvc4 chloroplast localization and symptom development, PVX vectors expressing various mutants of NSvc4 were agroinfiltrated individually into *N. benthamiana* leaves. The results demonstrate that NSvc4~1--73~ and NSvc4~1--106~ are capable of targeting chloroplasts (Figure [10](#F10){ref-type="fig"}; Table [2](#T2){ref-type="table"}). Interestingly, plants infected with two mutant viruses developed phenotypes similar to those caused by the wild wtPVX at 7dpi, and the disease phenotype also recovered by 20 dpi. Interestingly, NSvc4~106--286~ was predicted not to localize to chloroplasts, but, the mutant still elicited a severe symptom phenotype in infiltrated *N. benthamiana* plants that was maintained for up to 20 dpi. These results indicate that NSvc4 chloroplast localization is dispensable for the exacerbated symptoms. Hence, it is possible that the NSvc4 transmembrane domain has a role in chloroplast malformations, membrane proliferations from the chloroplasts and symptom development. {#F10} ###### **Targeting of NSvc4 and its mutants to plasmodesmata and chloroplast and their roles in symptom development**. Localization/symptom Plasmodesmata Chloroplast Symptom ---------------------- --------------- ------------- --------- NSvc4 \+ \+ \+ NSvc4~1--54~ -- -- -- NSvc4~1--73~ -- \+ -- NSvc4~1--106~ -- \+ n NSvc4~1--125~ \+ \+ \+ NSvc4~125--286~ -- -- \+ NSvc4~106--286~ -- -- \+ *n, The symptom data was not record*. NSvc4 protein did not mediate tubule formation in Sf-9 cells ------------------------------------------------------------ Our earlier research has shown that NSvc4 accumulated at PD in the walls of RSV-infected cells (Xiong et al., [@B41]). Because the NSm MP of *Tomato spotted wilt virus* (TSWV) formed tubule-like structures in insect cells (Storms et al., [@B31]), we decided to investigate the possibility of tubule formation by RSV NSvc4. In these experiments, the TSWV NSm (AcNPV/NSm-GFP) protein elicited numerous tubule-like extensions on Sf-9 cell surface by 36--48 h post transfection. However, Sf-9 cells transfected with the RSV NSvc4 protein (AcNPV/NSvc4-GFP), failed to develop similar tubules by 48 hpi. In contrast to the free GFP protein (AcNPV/GFP), which was distributed uniformly in the nuclei and in the cytoplasm, the NSvc4-GFP protein (AcNPV/NSvc4-GFP) accumulated as globular structures at the cell periphery and, in this regard, was similar to the localization patterns of NSvc4 in plant cells (Figure [11](#F11){ref-type="fig"}). However, it has been reported in an abstract that the NS2 protein encoded by RSV vRNA 2 can induce tubule-like structures in insect cells (ITMGCM, [@B17]), but this report has not been verified in a peer reviewed paper. Nevertheless, it is possible that NS2 may interact with NSvc4 to facilitate RSV movement, so in future experiments, we plan to investigate possible roles of NS2 and NSvc4 protein interactions to determine whether they may act together to facilitate RSV cell-to-cell transport. {#F11} Discussion ========== Our previous studies of RSV NSvc4 have indicated that NSvc4 belongs to the 30 K MP superfamily, and have shown experimentally that the protein interacts with single-stranded RNA *in vitro*, traffics to the PD of dicot cells and can move to adjacent cells after bombardment (Xiong et al., [@B41]). Within the 30 K superfamily, the TMV MP has been studied most intensively and is known to target PD via trafficking on cortical ER and actin cytoskeleton (Wright et al., [@B40]; Hofmann et al., [@B15]). A recently report indicates that targeting of NSvc4 to PDs utilizes the actin microfilament pathway and the myosin VIII rather than myosin XI motility system (Yuan et al., [@B43]). Our results complement and verify this study by demonstrating that actin microfilament dissociation by LatB and Golgi disruption by BFA interfere with PD targeting of the NSvc4-eGFP fusion protein, whereas disruption of microtubules by oryzalin has little effect on PD targeting. Moreover, microfilament disruption but not microtubule disruption inhibits infection of RSV in *N. benthamiana*, Thus, our combined results clearly suggest that targeting of NSvc4 to PDs depends on a functional ER and actin network. Many viral MPs within the 30 K superfamily have a hydrophilic region at their C-termini. Deletion or alanine-scanning mutations within the C-termini of several viral MPs have demonstrated that this region is dispensable for cell-to-cell movement (Schoelz et al., [@B26]). We have now extended previous RSV studies through transient expression of wtNSvc4 and NSvc4 mutants in cells, and have determined that NSvc4 differs from the general MP rule because the N-terminal 125 AAs are sufficient to target the truncated fragment to the PD. Furthermore, we have shown that AAs 106--125 contain a predicted transmembrane domain and that deletion of this domain abolishes the PD targeting ability of NSvc4. These results strongly suggest that the 20 deleted residues serve as an integral membrane signal that facilitates insertion into the ER. Several previous reports have shown that some viral MPs accumulated in chloroplast and thus might have an important role in virus replication, movement, and/or symptom development (Prod'homme et al., [@B24]; Torrance et al., [@B34]). For example AltMV TGB3 was shown to be responsible for AltMV movement between cells and contained a novel signal which was required for chloroplast membrane localization. Here we provide definitive evidence that RSV NSvc4 has a chloroplast-targeting signal within its N-terminal 73 residues, and that this signal targets the NSvc4-GFP chloroplast in both the agro-mediated and PVX-based expression systems. We anticipate further studies to elucidate the potential involvement of chloroplast-targeting in the RSV life cycle. Expression of the NSvc4 through PVX-based vector exacerbated disease symptoms in *N. benthamiana* than the symptoms elicited by PVX alone. Electron microscope observations suggested that disease symptoms correlated with chloroplast malformations and cytoplasmic membrane proliferations in cells. However, expression of mutants of NSvc4 indicated no direct connection between chloroplast localization of NSvc4 and symptom development. We propose that the chloroplast-targeting phenomenon may be involved in RSV replication or other unidentified activities. Considering the chloroplast malformations and membrane proliferations in the PVX-NSvc4 infected *N. benthamiana* cells, it is reasonable to propose that the transmembrane activity of the NSvc4 may play a pivotal role in development of disease symptoms. Because viral MPs modify PD structures and increase PD size exclusion limits, transgenic plants expressing viral MPs often show alterations in plant development. Plant developmental anomalies have also been demonstrated through infection of *N. benthamiana* plants using TMV-based vector expression NSvc4 (Zhang et al., [@B44]), so the phenomena we have observed are not virus specific. The authors also found that region D17--K33 was recognized as a crucial domain for leaf necrosis response using TMV-based vector expression NSvc4 (Zhang et al., [@B44]). In our experiment, we also observed foliar necrosis expressed of PVX-NSvc4. More detailed work is needed to determine the regions responsible for formation of foliar necrosis expressed from PVX vector. From these accumulated data, we conclude that RSV NSvc4 is a symptom determinant that affects the host phenotype, but the mechanisms whereby the protein functions in symptom development remain to be elucidated. One of the major questions unique to RSV movement is the form in which infectious entities might move from initial infection foci to adjacent cells. Because RSV is a negative strand "ambisense" virus, it is obvious that the NC must be involved in intra-cellular transit in order to facilitate nascent transcription and replication in newly invaded plant cells. Similar events also must function during RSV infections of planthopper vectors. Therefore, we carried out preliminary experiments to determine the location of NSvc4 and the TSWV NSm MP in insect cells. Our results show that in contrast to NSm, NSvc4 failed to produce tubule-like structures after plasmid transfection into Sf-9 cells, but instead formed large foci at the surface of the cells. We previously were unable to detect NSvc4 binding to the RSV NC protein, but have shown that NSvc4 exhibits non-specific RNA binding in gel shift assays. These results suggest that NSvc4 may be able to access RNA encapsidated in the NC, and such a mechanism is compatible with recent experimental data for Vesicular stomatitis virus (VSV), the most intensively studied negative strand virus (Green et al., [@B9]). Interestingly, the VSV NC (N) protein is thought to undergo conformational changes to permit access by the polymerase protein during transcription and replication. Moreover, the matrix protein of negative strand viruses has mechanisms to discriminate genomic NCs from antigenomic NCs during morphogenesis, and these likely are RNA sequence specific. Therefore, we posit that NSvc4 specifically recognizes RNA in RSV gNCs and ferries these complexes to the cell wall and then enlarges the PD complexes sufficiently to facilitate NC transit to adjacent cells. Although, different mechanisms, possibly cell fusion, may be involved in systemic spread in infected planthoppers, we envision that NSvc4:NC associations with NCs likely function during insect infection processes. Therefore, we are planning further investigations to elucidate the complicated mechanisms whereby RSV moves between plant and insect cells. Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. This research was supported by the Hi-tech Research and Development Program of China (2012AA101505) and the National Natural Science Foundation of China (Grant No. 30870110). We thank Prof. Chuanxi Zhang in Zhejiang University for supply of pFastBacHT-eGFPT and *E. coli* *DH10Bac*, Qianli An for supplying Man1-RFP plasmids. We apologize to colleagues whose articles were not cited because of space limitations. TSWV NSm was kindly provided by Prof. Jian Hong in Zhejiang University. ^1^<http://blanco.biomol.uci.edu/mpex/> ^2^<http://dgpred.cbr.su.se/index.php?p=home> ###### **Primers used in our experiments**. ---------------- ---------------------------------------- MP(Kpn1)-F 5′-ggggtaccATGGCTTTGTCTCGACTTTTG-3′ MP(BamH1)-R 5′-CGGGATCCcatgatgacagaaacttcag-3′ MP(BamH1)-R54 5′-GGATCCtgtggcagcttggtcaatc-3′ MP(BamH1)-R73 5′-GGATCCATCATACTTGTTCACCTTGACAT-3′ MP(BamH1)-R106 5′-CGGGATCCatgggtgagaggttgatg-3′ MP(kpn1)-F125 5′-gggtaccATGagtggaataactaccctcc-3′ MP(Kpn1)-F106 5′-gggtaccATGtatccattctttagagtggc-3′ MP(BamH1)-R125 5′-CGGGATCCagctctacccttgattcct-3′ MP-Fol 5′-cctctcacccatagtggaataactaccctc-3′ MP-Rol 5′-agttattccactatgggtgagaggttgatg-3′ eGFP(Sal1)-R 5′-GTCGACTTACTTGTACAGCTCGTCCAT-3′ MP(sal1)-R54 5′-GTCGACCTAtgtggcagcttggtcaatc-3′ MP(sal1)-R73 5′-GTCGACCTAATCATACTTGTTCACCTTGACAT-3′ MP(sal1)-R125 5′-GTCGACCTAagctctacccttgattcct-3′ MP(Cla1)-F125 5′-CCATCGATATGagtggaataactaccctcc-3′ MP(Cla1)-F 5′-ATCGATATGGCTTTGTCTCGACTTTT-3′ MP(sal1)-R 5′-GTCGACCTACATGATGACAGAAACTTC-3′ MP-(BamH1)-F 5′-GGATCCATGGCTTTGTCTCGACTTTT-3′ MP-(Sal1)-R 5′-GTCGACCTACATGATGACAGAAACTTC-3′ ---------------- ---------------------------------------- {#FA1} [^1]: Edited by: Aiming Wang, Agriculture and Agri-Food Canada, Canada [^2]: Reviewed by: Taiyun Wei, Fujian Agriculture and Forestry University, China; Weimin Li, Chinese Academy of Agricultural Sciences, China; Andrew O. Jackson, University of California, Berkeley, USA [^3]: This article was submitted to Frontiers in Plant-Microbe Interaction, a specialty of Frontiers in Plant Science.

1. Introduction {#sec0005} =============== T2DM is one of the most common non-communicable diseases in the world, and its levels are progressively increasing, particularly in developing countries [@bib0005]. It imposes a heavy burden on individuals and health care systems. The disease is associated with severe complications (e.g. blindness, lower limb amputations, and chronic renal failure) which affect health and productivity [@bib0010]. In 2011, diabetes caused around 4.6 million deaths globally in the 20--79 age group, and at least US\$ 465 billion in healthcare expenditures, which was equivalent to 11% of total healthcare expenditures in adults [@bib0015]. Therefore, it is important for countries to have credible data on the trends in T2DM prevalence and its likely future projections. These data are required for proper health policy planning and resource allocation for the prevention and control of T2DM. The Kingdom of Saudi Arabia (KSA) is one of the largest and wealthiest countries in the region of Middle East and North Africa (MENA). It is a leading oil-producing country, and has witnessed massive socioeconomic developments in the past five decades with rapid urbanisation and changes in the population lifestyles. KSA is now classified by the IDF to be among the top 10 countries globally with the highest projected prevalence of diabetes in 2011 (16.2%) and 2030 (20.8%) [@bib0005]. Furthermore, the prevalence of some risk factors for T2DM in KSA (e.g. obesity) has also been estimated to be among the highest in the world [@bib0020]. Epidemiological modelling is a valuable tool for estimation and future prediction of T2DM prevalence. The IDF [@bib0005] and other international modelling studies [@bib0025; @bib0030; @bib0035; @bib0040; @bib0045] have generated estimates of the prevalence of diabetes in different countries of the world, including KSA, at different time points. However, for KSA, most of such estimates appear to noticeably underestimate the true situation, as they have been well surpassed by the local 'observed' data. We developed the 'Saudi IMPACT Diabetes Forecast Model' to estimate the trends and likely future projections of T2DM in KSA, based primarily on the trends in the prevalence of obesity and smoking as risk factors. Obesity has been well recognised as one of the most important risk factors for T2DM [@bib0050]. In addition, smoking is not only a reflection of unhealthy lifestyle (with low physical activity and unhealthy diet), but it has also been recognised as an independent risk factor for T2DM in several large prospective studies that have adjusted their results for many potential covariates (e.g. obesity, physical activity, age, etc.) [@bib0055]. We validated our model against local observed data from the STEPS (STEPwise approach to non-communicable diseases Surveillance) survey [@bib0060] in 2005. This study aims to provide a detailed comparison of the estimates of T2DM prevalence in KSA by the Saudi IMPACT Diabetes Forecast Model against the estimates of the IDF Diabetes Atlas [@bib0005] and the GBD study [@bib0025]. 2. Methods {#sec0010} ========== 2.1. The model {#sec0015} -------------- The Saudi IMPACT Diabetes Forecast Model is a discrete-state Markov model, implemented in Microsoft Excel spreadsheets. It was originally designed for the MedCHAMPS project [@bib0065; @bib0070] for use in middle and low income countries, where extensive data on T2DM risk factors, prevalence, and complications may not be available. It integrates information on the trends in the adult Saudi population structure by age group and sex (obtained from the Central Department of Statistics and Information (CDSI) [@bib0075] and the United Nations (UN) population estimates [@bib0080]), and the trends in the prevalence of two risk factors for T2DM: obesity (BMI ≥ 30 kg/m^2^) and current active smoking (obtained from local population-based surveys). The model estimates the trends in prevalence of T2DM in the Saudi adults aged ≥25 years during the 30-year period of 1992--2022. The age- and sex-specific prevalence of T2DM in the starting year of modelling (1992) was obtained from a nationwide population-based study, which used the WHO 1985 diagnostic criteria and oral glucose tolerance test (OGTT) for diagnosis of T2DM [@bib0085]. [Fig. 1](#fig0005){ref-type="fig"} shows a simple illustration of the model structure. The model assumes that the population is divided into three discrete pools (health states): those who are obese, those who are smokers, and those who are 'healthy' (i.e. non-obese, non-smokers, and not having T2DM). Individuals can make transitions from the three main health states or remain in the same state during each modelling cycle (one year). Individuals can make transitions to the (Diabetes) state (i.e. they develop T2DM), or die due to other causes. Individuals in the (Diabetes) state can die as a result of T2DM or due to other causes. Individuals with T2DM cannot make transition back to the three main states, assuming a zero remission rate. The size of the (Obese) and (Smokers) states were determined by the age- and sex-specific prevalence of obesity and active smoking, as obtained from nationally-representative local surveys. Three surveys were used to obtain the prevalence of obesity in 1992 [@bib0085], 1997 [@bib0090], and 2005 [@bib0060]. In addition, two surveys were used to derive the prevalence of active smoking in 1992 [@bib0095] and 2005 [@bib0060]. Data for missing years were estimated through linear interpolation, while data for future trends were estimated through linear extrapolation, assuming similar rates of increase as that observed from surveys. [Fig. 2](#fig0010){ref-type="fig"} demonstrates the trends in prevalence of obesity and smoking in the adult Saudi men and women over the modelling period (1992--2022), assuming a linear increasing trends. The potential overlaps between the model health states were handled in three different ways. First, smoking prevalence was multiplied by obesity prevalence in order to estimate the proportion of population who were both obese and smokers. Then, such a proportion was subtracted from the 'original' smoking prevalence, to leave in the (Smokers) state only those individuals who were smokers but not obese. Second, we estimated the number of individuals with T2DM (in the Diabetes state) in whom the disease was assumed to be 'caused' by obesity as an exposure, through multiplying the '*population attributable risk*' [@bib0100] by the size of (Diabetes) state. Then, the number of such individuals was subtracted from the total obese individuals in population, to leave in the (Obese) state only those obese individuals who do not have T2DM. Finally, we applied the same previous approach of the *population attributable risk*, to leave in the (Smokers) states only those people who are smokers, but not having T2DM. The transition from (Healthy) to (Diabetes) states is informed by the *incidence of T2DM* (incidence in 'healthy' people only), while the transition from (Obese) to (Diabetes) states is informed by the *incidence of T2DM* × *relative risk (RR) of diabetes in obese individuals*, and the transition from (Smokers) to (Diabetes) states is informed by the *incidence of T2DM* × *RR of diabetes in smokers*. Moreover, the transitions from any of the three main states (Healthy, Obese, and Smokers) to the state of (Deaths due to other causes) are informed by the *total mortality rate*. On the other hand, transition from the (Diabetes) state to the state of (Diabetes-related deaths) is informed by the *case fatality rate*. The model does not show explicit transitions from the (Healthy) state to (Obese) or (Smokers), because of lack of data needed to inform such transitions. In addition, as a simplifying assumption, the prevalence of overweight (BMI 25--29.9 kg/m^2^) is not explicitly modelled, and we only considered the trends in obesity prevalence. The age- and sex-specific incidence and case fatality rate of T2DM, in addition to the total mortality rate of population were estimated using DISMOD 2 [@bib0105], which is a validated generic disease model designed to supplement data on a disease epidemiology by exploiting the causal relations between the various available parameters. DISMOD 2 provides 'internally-consistent' estimates of diabetes incidence, case fatality rate, and total mortality rate, based on a set of differential equations that describe the disease process [@bib0105]. DISMOD 2 has been widely used by the WHO\'s Global Burden of Disease (GBD) Study and other modelling studies. DISMOD 2 provides an estimated incidence rate for the overall population (i.e. for obese, smokers, non-obese and non-smokers). However, assuming that the overall incidence of diabetes is a weighted sum of the incidence among exposed (Obese and Smokers) and non-exposed (Healthy) [@bib0110], we calculated the incidence rate of diabetes among (Healthy) people only to inform the transition between the (Healthy) state and (Diabetes) state. We then adjusted this incidence rate by applying the RRs of diabetes in obese and in smokers in order to inform the transitions between the (Obese) and (Smokers) states to (Diabetes) state. The RR of diabetes in obese and in smoker individuals were obtained from two recent systematic reviews and meta-analyses [@bib0050; @bib0055]. [Table 1](#tbl0005){ref-type="table"} shows the age- and sex-specific parameters used to set the transition probabilities in the model. We conducted two main methods of sensitivity analyses in order to test the potential uncertainties around the model parameters. First, we used the 'analysis of extremes' method [@bib0115; @bib0120], where all model parameters (except population structure) were conservatively set at 20% higher and 20% lower values than the base-case scenario. The model was run with this distribution of extreme values and the uncertainty intervals (UIs) were estimated accordingly. Second, we used the 'scenario analysis' method, where two reasonable scenarios were assumed for the projected obesity prevalence. For *scenario 1*, we assumed that the obesity trends would continue to increase at the same annual rate as that observed from the national surveys. In *scenario 2*, we assumed that the projected obesity trends would be 'capped' at the highest 'observed' value in any age group for each sex separately. The highest observed value for obesity prevalence in men was 34.5% in those aged 35--44 years, while in women was 58.8% in the age group 45--54 years. Therefore, the capping point was assumed to be 35% in men and 60% in women. 2.2. Comparison with other models {#sec0020} --------------------------------- We compared our model estimates with those recently published by two independent modelling exercises, which generated estimates and projections of T2DM prevalence in a large number of countries around the world, including KSA. ### 2.2.1. The IDF Diabetes Atlas, fifth edition [@bib0005; @bib0125] {#sec0025} The IDF used logistic regression to model diabetes prevalence rates in KSA, which were derived from five national population-based surveys. The smoothed age- and sex-specific prevalences were then applied to the national population distribution for the years 2011 and 2030 (using the UN population estimates and the world population distribution) to estimate national prevalence of diabetes. The IDF methodology used changes in age, sex and urbanisation as covariates for estimating diabetes prevalence. ### 2.2.2. The GBD model [@bib0025] {#sec0030} The GBD study (Global Burden of Metabolic Risk Factors of Chronic Diseases Collaborating Group (Blood Glucose)) used three national population-based surveys of T2DM prevalence in KSA [@bib0060; @bib0085; @bib0130] and used a multi-level statistical approach (Bayesian hierarchical modelling) to model the trends in fasting plasma glucose (FPG) and T2DM prevalence in adults aged ≥25 years. The estimates were informed by several country-level covariates. These covariates were national income (natural logarithm of the per-head gross domestic product), urbanisation (proportion of population that lived in urban areas), age-standardised mean BMI (from a previous GBD systematic analysis of country data [@bib0020]), and national availability of multiple food types for human consumption (from the food balance sheets of the Food and Agriculture Organisation (FAO) of the UN). The presented estimates were age standardised to the WHO reference population. 3. Results {#sec0035} ========== The prevalence of T2DM among the Saudi population aged ≥25 years is estimated to rise substantially from 8.5% (UI: 6.8--10.2%) in 1992 to 31.4% (UI: 25.5--37.0%) in 2013 and 44.1% (UI: 35.4--52.5%) by 2022, assuming that the observed prevalence rates of both smoking and obesity will continue to increase. In comparison, the prevalence of T2DM is estimated to increase to 30.8% (UI: 25.2--36.2%) in 2013 and 39.5% (UI: 32.5--45.9%) by 2022, assuming capped obesity trends at 35% in men and 60% in women. The estimated number of people with T2DM in KSA will increase substantially from around 555,000 in 1992 to approximately 7.4 million by 2022, assuming a continuing increase of obesity levels, and nearly 6.6 million, assuming capped obesity levels. We compared the estimates of T2DM prevalence in KSA from the Saudi IMPACT Diabetes Forecast Model against the estimates of the IDF Diabetes Atlas (fifth edition) and the GBD Study ([Table 2](#tbl0010){ref-type="table"}). 3.1. Comparison with the IDF Diabetes Atlas, fifth edition [@bib0005; @bib0125] {#sec0040} ------------------------------------------------------------------------------- The Saudi IMPACT Diabetes Forecast Model estimated the total T2DM prevalence in the Kingdom (in adults aged ≥25 years) at 29.2% in 2011 and 44.1% in 2022 (assuming a continuing linear increase in obesity trends "scenario 1"), or at 28.9% in 2011 and 39.5% (assuming capping of future obesity trends "scenario 2"). In comparison, the IDF has estimated the prevalence of diabetes in KSA in adults aged 20--79 years at 16.2% in 2011 and 20.8% in 2030. 3.2. Comparison with the GBD estimates [@bib0025] {#sec0045} ------------------------------------------------- The Saudi IMPACT Diabetes Forecast Model estimated the prevalence of diabetes in KSA among adult men and women respectively as follows: 17.7% and 16.4% in 2000, and 26.7% and 24.7% in 2008. The model results for 2000 and 2008 were the same with assuming a continuing linear increase (scenario 1), or capping (scenario 2) of obesity trends. On the other hand, the estimates of the GBD model for men and women respectively were 17.5% and 17.7% in 2000, and 22.0% and 21.7% in 2008. 4. Discussion {#sec0050} ============= We developed and validate the Saudi IMPACT Diabetes Forecast Model, which is a Markov model with relatively few data input requirements. The model might be most suitable for use in less developed settings with limited data on T2DM. In addition to the demographic trends, the model also used the trends in two risk factors for T2DM in KSA, as well as literature-derived transition hazards. This study presents a comparison between the results from this model against that of the IDF Diabetes Atlas (fifth edition) [@bib0005] and the GBD study [@bib0025]. In general, reliability and accuracy of T2DM prevalence estimates are highly dependent on the data sources used in modelling process and the model structure and methodology. Thus, comparing the results of different T2DM prevalence estimates may be difficult, as different models often utilise different data sources, apply different methodologies for estimation and projections, and use different assumptions. Furthermore, Danaei et al. [@bib0025] reviewed the available global diabetes estimates and reported some other potential reasons for variations in their results. For instance, the definition of diabetes varied in different studies, as diagnostic criteria are repeatedly changeable over time. In addition, studies were different in their populations of interest, and some of them used data sources with subnational samples, regarding them as equally representative of national populations. This could lead to biased results, as those specific subnational groups might differ from the general populations in many aspects such as prevalence of some risk factors for diabetes. This study revealed that the Saudi IMPACT Diabetes Forecast Model resulted in significantly higher estimates and projections of T2DM prevalence in KSA than that estimated by the IDF for 2011 and 2030 [@bib0005]. These considerable differences between the two studies could be mainly attributed to the different methods of modelling of T2DM prevalence and the covariates used for that purpose. According to the IDF, the estimation approach was deliberately kept simple and conservative. The IDF model used logistic regression method and based its predictions for 2030 on predicted demographic changes (urbanisation and ageing). Moreover, the IDF model did not attempt to directly account for the effects of changes in T2DM risk factors (e.g. obesity). The IDF has acknowledged this as a limitation which was likely to result in underestimation of T2DM prevalence if the levels of obesity and other risk factors continue to rise [@bib0005]. Also, several previous modelling studies [@bib0030; @bib0035; @bib0040; @bib0045] resulted in lower estimates and projections of diabetes prevalence in KSA at different time points than the Saudi IMPACT Diabetes Forecast Model ([Table 3](#tbl0015){ref-type="table"}). Shaw et al. [@bib0030] and Wild et al. [@bib0035] used local population-based surveys (also used by the IDF and GBD models) as data sources of diabetes prevalence in KSA. On the other hand, the oldest two studies (King et al. [@bib0040] and Amos et al. [@bib0045]) used old data from Oman [@bib0135] (a neighbouring country to KSA with similar socioeconomic characteristics) and extrapolated such data to KSA. In general, all these four modelling studies used only demographic changes and level of urbanisation as informants of diabetes prevalence estimates and projections. On the contrary, the Saudi IMPACT Diabetes Forecast Model utilised a different estimation approach (Markov modelling). It used only the prevalence of T2DM for the starting year (1992) along with the demographic trends of the Saudi population (1992--2022) and the trends in prevalence of two risk factors (obesity and smoking) over the same 30-year-period. In addition, the model used a number of transition parameters, such as the estimated incidence of T2DM, case fatality rate, general mortality and evidence-based RRs. The Saudi IMPACT Diabetes Forecast Model used published nationally representative, population-based studies to obtain data on the prevalence of obesity and smoking in KSA, considering two reasonable scenarios of the future obesity trends. The results of the GBD model [@bib0025] and the Saudi IMPACT Diabetes Forecast Model are very comparable, in spite of differences in the general methods of estimation. This similarity in results of the two models could be attributed primarily to the several covariates used in both models to estimate T2DM prevalence. In contrast to the other models which used only ageing and level of urbanisation as covariates, the GBD model incorporated more covariates to inform its estimates. However, among all these covariates, mean BMI is likely to have the most important contribution to the higher estimates of the GBD model than that of the IDF model; because mean BMI could serve as a 'direct' informant of the trends in obesity levels. Another GBD modelling study for trends in the global BMI [@bib0020] showed that the estimated mean BMI in Saudi Arabia followed a linear increase between 1980 and 2008. The estimates of mean BMI (kg/m^2^) in men and women in Saudi Arabia were respectively as follows: 25.0 (uncertainty interval (UI): 23.8--26.3) and 26.3 (UI: 24.8--27.8) in 1980, 25.9 (UI: 25.6--26.2) and 27.3 (UI: 26.9--27.8) in 1990, 27.0 (UI: 26.6--27.4) and 28.5 (UI: 28.0--29.0) in 2000, and 27.9 (UI: 27.2--28.6) and 29.6 (UI: 28.7--30.5) in 2008. Furthermore, the GBD estimates showed that the region of 'North Africa and Middle East' has witnessed the largest increase in mean BMI in men and women between 1980 and 2008 after the region of 'Oceania'. Saudi Arabia was among the top countries with the highest increase in mean BMI within its region [@bib0020]. Country-specific estimates of the prevalence of obesity (BMI ≥ 30 kg/m^2^) were not reported. However, the region of 'North Africa and Middle East' had the seventh (among the 21 GBD regions of the world) highest prevalence of obesity in men, and the second highest in women between 1980 and 2008. In men, the estimated prevalence of obesity in that region increased substantially from \<10% in 1980 to 20--30% in 2008. On the other hand, obesity prevalence in women increased from 10--20% in 1980 to 30--40% in 2008 [@bib0020]. Taking that into account, the trends in obesity prevalence seems then to be a better option when modelling the future burden of T2DM. The Association of Public Health Observatories (APHO) Diabetes Prevalence Model [@bib0140] is a recent consistent example that reported similar variations in the estimated prevalence of diabetes when compared to other models. The APHO Diabetes Prevalence Model used the trends in overweight and obesity in England to estimate diabetes prevalence, which was approximately one third higher than that estimated by the IDF for 2010. The Saudi IMPACT Diabetes Forecast Model offers to the health policy planners in KSA robust and validated estimates and projections of the burden of T2DM, and shows that the currently high levels of the disease are expected to continue increasing during the next decade, even assuming some degree of levelling off of obesity trends. Therefore, there is an urgent need for applying effective, aggressive and multisectoral prevention measures to promote healthy diet, physical activity, and smoking cessation. It is recommended to improve public awareness to increase consumption of fruit and vegetables, restrict consumption of fat and sugar, practice a regular physical activity, and quit tobacco smoking [@bib0145; @bib0150]. It is also important to establish (or maintain) relevant legislation to enforce food industry to make clear and informative food labelling, to incorporate physical education into curriculums at girls schools, to strictly prohibit smoking at public and work places, and to increase tobacco prices and ban its selling to children and adolescents [@bib0145; @bib0150]. Beside the primary prevention measures which aim to reduce the prevalence of T2DM, it is also recommended to implement mass screening programmes to diagnose those people with 'occult' disease for early treatment and prevention of disease progression [@bib0145; @bib0150]. 5. Conclusion {#sec0055} ============= The Saudi IMPACT Diabetes Forecast Model and the GBD model produced comparable estimates of T2DM prevalence in Saudi Arabia, as both models directly incorporated the trends in obesity prevalence and/or BMI to inform the estimates. In contrast, the IDF and other modelling studies relied solely on demographic trends and urbanisation as covariates, and produced lower estimates of T2DM prevalence in Saudi Arabia. Hence, it appears that the direct incorporation of obesity trends (as a strong predictor of T2DM) in modelling studies results in higher estimates of the future prevalence of T2DM, particularly in countries and regions with high levels of and increasing trends in obesity prevalence. Conflict of interest statement {#sec0060} ============================== None. This work is part of a PhD at Newcastle University, United Kingdom, sponsored by the Saudi Ministry of Health and the Saudi Cultural Bureau in London. M O'Flaherty was supported by the UK Medical Research Council (Grant G 0900847), and by the European Union MedCHAMPS project funded by EC FP7 grant no. 223705. We would like to thank Dr Ziad Ahmed Memish, the Assistant Deputy Minister for Public Health and Dr Mohammad Abubakr Alhamid, the Director of Diabetes Control Programme in Saudi Arabia for their kind cooperation and facilitating the field work. {#fig0005} {#fig0010} ###### Transition hazards used in the Saudi IMPACT Diabetes Forecast Model. Transition parameter Data source Men -- age group (years) Women -- age group (years) ------------------------------------------------------ ------------------------- -------------------------- ---------------------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- Estimated incidence rate of diabetes/1000 population DISMOD 2 12.90 17.70 18.90 20.70 22.40 26.70 12.90 15.00 15.90 16.70 19.70 30.70 Estimated case fatality rate (%) DISMOD 2 0.15 0.39 0.67 1.20 1.35 2.10 0.16 0.43 0.62 0.96 1.90 4.62 Estimated total mortality rate/1000 population DISMOD 2 0.10 0.50 1.10 2.50 3.30 6.10 0.10 0.50 1.00 1.80 4.10 11.60 

 RR of diabetes if obese Guh et al. [@bib0050] 6.74 12.41 RR of diabetes if a smoker Willi et al. [@bib0055] 1.44 1.44 ###### Comparison of the Saudi IMPACT T2DM Forecast Model against the IDF (2011) model and the GBD (2011) model. ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- IDF (2011) [@bib0005; @bib0125] GBD (2011) [@bib0025] Saudi IMPACT Diabetes Forecast Model ----------------------------------------------------- --------------------------------- ----------------------------------------------------- ---------------------------------------------------------------- --------------- ---------------- ---------------- Estimated DM prevalence in Saudi Arabia (%) 2011 Total: 16.2 2000 Males: 17.5\ 2000 Males: 17.7\ Females: 17.7 Females: 16.4\ Total: 17.2 2030 Total: 20.8 2008 Males: 22.0\ 2008 Males: 26.7\ Females: 21.7 Females: 24.7\ Total: 25.9 2011 Males: 29.8\ Females: 28.1\ Total: 29.2 2022 Males: 41.3\ Females: 47.7\ Total: 44.1 

 Age of study population (years) 20--79 25+ 25+ 

 Main data sources for DM prevalence in Saudi Arabia Al-Nuaim et al. [@bib0155]\ Warsy and El-Hazmi [@bib0085]\ Warsy and El-Hazmi [@bib0085] (for starting year prevalence)\ El-Hazmi et al. [@bib0160]\ Al-Nozha et al. [@bib0130]\ WHO STEPS [@bib0060] (for validation) Warsy and El-Hazmi [@bib0085]\ WHO STEPS [@bib0060] Al-Nozha et al. [@bib0130]\ WHO STEPS [@bib0060] 

 Estimation methodology Logistic regression modelling Complex multi-level Bayesian hierarchical modelling Markov modelling 

 Covariates used for estimating DM prevalence • Urbanisation\ • National income\ • Trends in population structure\ • Ageing • Urbanisation\ • Trends in obesity prevalence\ • National availability of multiple food types\ • Trends in smoking prevalence\ • Age-standardised mean BMI • Estimated incidence of T2DM\ • Estimated case-fatality rate\ • Evidence-based estimates of RRs for transition probabilities ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ###### Comparison of the Saudi IMPACT T2DM Forecast Model against other modelling studies. --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Shaw et al. [@bib0030] Wild et al. [@bib0035] King et al. [@bib0040] Amos et al. [@bib0045] Saudi IMPACT Diabetes Forecast Model ----------------------------------------------------- ------------------------------- ---------------------------- ---------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------- ------------ ------------- ------------- ------------- ------------- Estimated DM prevalence in Saudi Arabia (%) 2010 Total: 13.6 2000 Total: 6.2 1995 Total: 8.7 1995 Total: 10.0 1995 Total: 11.1 2030 Total: 17.0 2030 Total: 8.1 2000 Total: 9.1 2000 Total: 12.0 2000 Total: 17.2 2025 Total: 10.1 2010 Total: 13.8 2010 Total: 28.1 2022 Total: 44.1 

 Age of study population (years) 20--79 20+ 20+ 20+ 25+ 

 Main data sources for DM prevalence in Saudi Arabia Al-Nuaim et al. [@bib0155]\ El-Hazmi et al. [@bib0160] Asfour et al. [@bib0135]*(Study from Oman)* El-Hazmi et al. [@bib0165]\ Warsy and El-Hazmi [@bib0085] (for starting year prevalence)\ El-Hazmi et al. [@bib0160]\ Asfour et al. \[27\]*(study from Oman)* WHO STEPS [@bib0060] (for validation) Al-Nozha et al. [@bib0130] 

 Estimation methodology Logistic regression modelling DISMOD 2 Age-specific diabetes prevalence estimates were applied to UN population estimates and projections Country-specific diabetes prevalence data were applied to the corresponding national age distribution Markov modelling 

 Covariates used for estimating DM prevalence • Demographic changes\ • Demographic changes\ • Trends in population size and age structure\ • Level of economic development (GNP per capita)\ • Trends in population structure\ • Urbanisation • Urbanisation • Urbanisation • Urbanisation • Trends in obesity prevalence\ • Trends in smoking prevalence\ • Estimated incidence of T2DM\ • Estimated case-fatality rate\ • Evidence-based estimates of RRs for transition probabilities ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

{#sp1 .198} {#sp2 .199}

Influenza lethality is due to three main causes: First, direct viral origin, leading to viral pneumonia of high severity, with an acute respiratory distress syndrom (ARDS), associated with 30 to 50% lethality in intensive Care Unit (ICU) [@ref-2593790109]; Second, bacterial surinfection, due to pneumococal, staphylococal, streptococal, or meningococal pneumonia, usely curable with appropriate antibiotics, provided they are administered early enough; and Third, decompensation of severe underlying conditions, often in elderly people or in vulnerable chronic patients. The last one is the predominant cause of death attributable to seasonal influenza in developed countries, and has reached for years now, despite of the use of mass immunization in elderly, a level around 1 death per 1,000 infections. This cause of lethality is rarely reported as attributed to influenza in death certificates, and therefore is not easily assessable during outbreaks, and is usely measured as excess mortality when time series of mortality become available, sometimes in real time, most often several months after the season [@ref-271392059]. The second cause, i.e. bacterial surinfections, is much less seen in developed countries due to antibiotics availability. However, in the developing world, it may still be a cause of concern, since bacterial pneumonia may rapidely become lethal if treatment is not given appropriately in time. The direct lethality due to viral pneumonia gives probably the best estimate of influenza strains\' virulence, since it may vary from strain to strain, and is not due to the level of health development of a country. Furthermore, ARDS always leads to ICU, and is easy enough to diagnose and to report. Death from ARDS due to seasonal influenza There are not many assessments available with regards with the incidence of death from ARDS due to seasonal influenza in the litterature. Empirical evidence suggests in France that less than 5 to 10 of such cases are identified each year, when an average of 6 million seasonal infections are estimated from the French Sentinelles system [@ref-1898629806]. We may therefore assume, waiting for better estimates in the future, that deaths from ARDS due to seasonal influenza is an exceptional event, which occurs once every million infected patients. Death from ARDS due to H1N1pdm It is too early to provide precise figures of deaths of ARDS due to the new strain of pandemic influenza. In the USA there are a too large uncertainty reported on the denominator (i.e. number of influenza infections), when in the UK or Argentina too. In smaller areas, and in particular in islands such as French New Caledonia or Mauritius, we may have first rough estimates which may be more relevant and easier to catch. French Government estimated last week that 20,000 cases of influenza had occured in New Caledonia in the emerging outbreak, leading to a rough estimate of 30,000 infections and by August 21st, 3 deaths of confirmed cases were reported, in patients aged 8 yr, 27 yr, 58 yr [@ref-1030737724]. Mauritius Government estimated last week that 15,000 cases of influenza had occured in the last two weeks in  Mauritius, however some reports from the media and recent personal communication from local doctors lead me to estimate that about 50,000 cases actually occured, and therefore about 70,000 infections, taking acount mild illnesses. By August 19th, 7 deaths had occured, 5 of them were already virologically confirmed in patients aged 4, 6, 28, 46, 53 [@ref-1240742045]. New Caledonia and Mauritius figures are consistant to a rough estimate of about 1 death from ARDS due to H1N1pdm per 10,000 infections, i.e. a virulence of an order of magnitude of 100 times that observed for seasonal strains. We acknowledge that we are producing very rough estimates of direct lethality associated to H1N1pdm. However, it may be useful to deliver such estimates as early as possible in the course of this pandemic, with the purpose of helping health authorities to check availability of ICUs and artificial ventilation devices in their countries, in case of a wave of similar virulence this fall in the Northern Hemisphere. When assuming that half of ARDS are of lethal outcome, we may assume that H1N1pdm may generate 1 case of ARDS every 5,000 infections. ARDS is always treated in ICUs, and three weeks of treatment is probably an average period to be taken into account. In case of attack rates of 30% to 50% this fall and winter in Northern Hemisphere [@ref-48887779], these figures may help refining preparedness. This work is funded by EHESP School of Public Health, Rennes, FranceThe author has declared that no competing interests exist

Background ========== Malaria in pregnancy is a major public health problem in tropical and subtropical regions of the world. In Africa, millions of women living in malaria-endemic areas become pregnant each year \[[@B1],[@B2]\]. Malaria in pregnancy contributes to significant maternal and perinatal morbidity and mortality. Each year, more than 500,000 women die during pregnancy or childbirth \[[@B1]\]. Severe malaria is a medical emergency associated with high mortality, especially in cases with multiple organ dysfunction \[[@B3]\]. Cerebral malaria and severe malarial anaemia are two major syndromes causing malaria-related mortality \[[@B4]\]. Children and pregnant women are the most vulnerable groups to the severe form of *P. falciparum*malaria \[[@B3]\]. While much literature and many publications are available on severe malaria in children, few published data exist on severe malaria during pregnancy \[[@B5]\]. Pregnant women are more attractive to the main malaria vector and the disease, including its severe form, than their non-pregnant counterparts \[[@B6]-[@B8]\]. Understanding the interactions that underlie the disease and its control should be helpful to investigate the epidemiology of severe malaria. Thus, such study is vital and may be of great interest for providing health planners and caregivers with fundamental guidelines for the implementation of preventive measures. In Sudan, the largest country in Africa, high maternal and perinatal mortality have been observed in different regions, and both malaria and anaemia were the major causes of these high levels of mortality \[[@B9]-[@B11]\]. Thus, the present study was conducted at the Kassala and Medani maternity hospitals in Sudan, which are located in areas characterized by unstable malaria transmission \[[@B12]\], and where malaria is a substantial burden affecting pregnant women irrespective of their age or parity \[[@B13]\]. Methods ======= This study was conducted at the Kassala and Medani (Figure [1](#F1){ref-type="fig"}) maternity hospitals in Sudan during the period from July to November 2010 for investigating the epidemiology of severe *P. falciparum*malaria among pregnant women. Pregnant women with symptoms of *P. falciparum*malaria were included in this study after informed consent was obtained from the patient or guardian. Those women with one or more of the manifestations of severe *P. falciparum*malaria according to the World Health Organization (WHO) criteria, which include cerebral malaria (unarousable coma), convulsion (more than two per 24 hours), hypotension (systolic blood pressure \< 90 mmHg with cold extremities), severe anaemia (haemoglobin \< 7 gm/dl), jaundice (detected clinically or bilirubin \> 1 mg/dl), hypoglycaemia (blood glucose \< 40 mg/dl) and hyperparasitaemia (parasite count \> 100,000 ring forms/μl), were managed according to the WHO guidelines, and the rest were considered as uncomplicated cases \[[@B3]\]. Questionnaires were used to gather socio-demographic, medical and obstetrical data. {#F1} Blood films were prepared and stained with Giemsa, and 100 oil immersion fields were examined. The parasite density was evaluated by counting the number of asexual *P. falciparum*parasites for every 200 leukocytes, assuming a leukuocyte count of 8000 leukocytes/μl. All slides were double-checked in a blinded manner and only considered negative if no parasites were detected in 100 oil immersion fields. If gametocytes were seen, then the count was extended to 500 oil immersion fields. Haemoglobin concentrations were estimated using a HemoCue haemoglobinometer (HemoCue AB, Angelhom, Sweden). Blood glucose was measured at baseline before quinine infusion, two hours after quinine infusion and if there was clinical suspicion of hypoglycaemia using the bedside device Accu-Chek™ Multiclix (Roche diagnostics, Mannheim Germany). The Accu-Chek™ machine was calibrated weekly and every time a new box of test strips was opened. Resuscitation and supportive management were given according to the WHO guidelines \[[@B3]\]; i.e. quinine infusion at 10 mg/kg three times a day over 2-3 hours changed to oral quinine tablet when the patient could tolerate them, correction of hypoglycaemia with 10% glucose, termination of convulsions with intravenous diazepam if they were persisting for more than three minutes. Paracetamol was given every 6 hours until defervescence. Those with severe anaemia (haemoglobin \< 7 g/dl) and respiratory distress were transfused with blood screened for hepatitis and HIV. Vital signs were measured every 15 minutes for the first hour, then every 2 hours until 24 hours, and thereafter every 6 hours until the discharge from the hospital. Baseline investigations were performed for every patient on admission and repeated when clinically indicated. These included levels of haemoglobin, serum urea, serum creatinine, and serum bilirubin as well as the white blood cell count. Statistics ========== Data were entered into a computer database and SPSS software (SPSS Inc., Chicago, IL, USA) and double checked before analysis. Means (SD) and proportions for the socio-demographic and biochemical variables were calculated. ANOVA was used to compare the means (SD) between the different groups of severe malaria. Correlations between the different continuous clinical and biochemical variables were calculated. *P*\< 0.05 was considered significant. Ethics ====== The study received ethical clearance from the Research Board at the Faculty of Medicine, University of Khartoum. Results ======= Among 222 pregnant women diagnosed with malaria at the two hospitals, 40 (18.0%) fulfilled one or more of the WHO criteria for severe *P. falciparum*malaria. These were hypoglycaemia (14; 35.5%), severe anaemia (12; 30%), hypotension (10; 25%), jaundice (9; 22.5%), cerebral malaria (6; 15%), repeated convulsions (4; 10%), hyperparasitaemia (4; 10.0%) and more than one manifestation (9; 22.5%), Table [1](#T1){ref-type="table"}. Ten patients with severe anaemia received blood transfusion. Out the 12 patients with severe anaemia; three patients (25%) had jaundice, one patient had hypotension and hyperparasitaemia (parasite count was 133333 rings/μ) and one patient had hypogylcaemia. The mean (SD) of the age, gravidity and gestational age of these 40 women were 28.4 (6.1) years, 3.5 (2.3) and 29.3 (6.7) weeks, respectively, Table [2](#T2){ref-type="table"}. The parity ranged from 1 to 9, (median = 2) and 10 (25%) of these women were primigravidae. None of the patients had used antimalarial chemoprophylaxis. There were no maternal deaths. All women were febrile. Different symptoms such as sweating, aches, vomiting and diarrhoea were observed among these women, Figure [2](#F2){ref-type="fig"}. None of the patients developed hypoglycaemia during quinine treatment. All patients started oral quinine tablet within two days. ###### presentations of severe *P. falciparum*malaria at Kassala and Medani Maternity Hospitals, Sudan Criteria of severe *P. falciparum*malaria Number of women Percentage ------------------------------------------- ----------------- ------------ Hypoglycaemia 14 35.5 Severe anaemia 12 30.0 Hypotension 10 25.0 Jaundice 9 22.5 Cerebral malaria 6 15.0 Convulsions 4 10.0 Hyperparasitaemia 4 10.0 More than one complication 9 22.5 ###### presenting clinical and biochemical data of the pregnant women with severe *P. falciparum*malaria at Kassala and Medani Maternity Hospitals, Sudan Variables range, mean ± SD ------------------------------- -------------------------------- Age, years 28.4 ± 6.1 \[18\--40\] Gravidity 3.5 ± 2.3 \[1\--9\] Gestational age, weeks 29.3 ± 6.7 \[14\--38\] Duration of illness, days 2.5 ± 1.1 \[1\--6\] Weight, kg 59.3 ± 7.1 \[43\--75\] Temperature, °c 38.5 ± 0.7 \[37.8\--40.0\] Haemoglobin, gm/dl 8.3 ± 1.7 \[5\--11.0\] White blood cells, cell/mm^3^ 6300 ± 2927 \[2700\--14000\] Parasite count, rings/μl 30717 ± 5270 \[2080\--335242\] Blood glucose, mg/dl 71.4 ± 3.0 \[29\--120\] serum bilirubin, mg/dl 2.0 ± 0.7 \[1\--3\] Serum creatinine, mg/dl 1.0 ± 0.2 \[0.7\--2.0\] {#F2} While the mean (SD) presenting temperature was significantly lower in women presenting with hypoglycaemia \[38.2(0.6) versus 38.8(0.7) °C, *P*= 0.04\], other clinical and biochemical characteristics were not significantly different among women with different manifestations of severe *P. falciparum*malaria, Table [3](#T3){ref-type="table"}. The haemoglobin level was significantly lower in the anaemic group, Table [3](#T3){ref-type="table"}. There was no significant correlation between blood glucose levels and temperature, haemoglobin levels and parasite counts, Table [4](#T4){ref-type="table"}. ###### comparison of clinical and biochemical data in the subgroups of the pregnant women with severe *P. falciparum*malaria at Kassala and Medani Maternity Hospitals, Sudan -------------------------------------------------------------------------------------------------------- Variable Hypoglycemic group (*N*= 14) Anaemic group (*N*= 12) Other group\ *P* (*N*= 14) ------------------------ ------------------------------ ------------------------- -------------- ------- Age, years 27.1(8.1) 30.1(5.9) 28.8(4.4) 0.1 Gravidity 3.8(2.9) 3.6(2.7) 3.5(1.9) 0.3 Gestational age, weeks 26.9(6.4) 29.2(8.4) 32.0(4.5) 0.6 Weight, kg 59.0(6.6) 61.3(7.7) 58.3(7.6) 0.2 Temperature, °C 38.2(0.6) 38.5(0.5) 38.8(0.7) 0.04 Haemoglobin, gm/dl 8.4(1.4) 6.2(0.7) 9.2(1.1) 0.001 parasite count, ring/μ 32057(62169.1) 27191(37961) 29244(3633) 0.9 -------------------------------------------------------------------------------------------------------- ###### Correlation between various clinical and biochemical measures including haemoglobin Variable Temperature Blood glucose Parasite count --------------- ------------- --------------- ---------------- ------ ------- ----- Haemoglobin 0.242 0.1 0.146 0.3 0.129 0.4 Temperature 0.290 0.07 0.013 0.9 Blood glucose 0.079 0.6 Discussion ========== The main findings of the present study were that hypoglycaemia and severe anaemia were the predominant presenting manifestations of severe *P. falciparum*malaria observed during pregnancy in this epidemiological setting. With the exception of lower temperatures, other clinical and biochemical criteria were not different among women with different criteria of severe *P. falciparum*malaria. Previously, severe anaemia and jaundice had been observed as the presenting manifestations of severe *P. falciparum*malaria in pregnant women in central and eastern Sudan \[[@B5],[@B14]\]. In neighbouring Ethiopia, cerebral malaria, convulsions, altered mental state and prostration were the common manifestations of severe malaria observed in pregnant women \[[@B15]\]. Generally, pregnant women are more susceptible to severe malaria and hypoglycaemia than their non-pregnant peers \[[@B3],[@B7]\]. The glucose metabolism during malaria infection is affected by several factors, including drug treatment, fever, parasite metabolism, hormonal changes, cytokines, fasting and gastrointestinal disturbances \[[@B16],[@B17]\]. It has been reported that patients with severe malaria-induced hypoglycaemia have higher mortality rates \[[@B18]\]. Thus, the recognition of patients with falciparum malaria and hypoglycaemia by blood glucose estimation at the time of admission could significantly affect the ultimate outcome. Interestingly, some comatose patients regained consciousness with intravenous fluid infusion of 25% dextrose only without receiving any specific antimalarial treatment \[[@B18]\]. None of these patients developed hypoglycaemia after quinine treatment. In this study blood glucose was investigated at base line, two hours following quinine and if hypoglycaemia was clinically suspected. Ideally, blood glucose should be investigated every four hours if possible especially in comatose patients according to the WHO guidelines \[[@B19]\]. Therefore, this is one of the limitations of this study where quinine -induced hypoglycaemia was not investigated as should be. Previously, only one out of 33 pregnant Sudanese women developed hypoglycaemia following quinine treatment for severe *P. falciparum*malaria \[[@B5]\]. Hyperinsulinaemic hypoglycaemia is the most important adverse effect in the quinine treatment of severe malaria which is particularly common in pregnancy (50% of quinine-treated women with severe malaria in late pregnancy) \[[@B20],[@B21]\]. Intravenous artesunate is superior to quinine in the treatment of severe malaria \[[@B22]\]. Compared to intravenous quinine, intravenous artesunate has been shown to have; a lower risk of hypoglycaemia, significantly reduce the risk of death from severe malaria, and it is not requiring rate controlled infusion or cardiac monitoring \[[@B19]\]. Patients in this series were in their second and third trimester of pregnancy; therefore intravenous artesunate would have been given to these women instead of quinine. However, intravenous artesunate is not yet registered and available in Sudan. In the present study, 12 (30%) and 6 (15%) patients presented with severe anaemia and cerebral malaria, respectively. Cerebral malaria and severe malarial anaemia are two major syndromes causing malaria-related mortality \[[@B4]\]. The pattern of these two severe forms varies depending on the intensity of transmission; cerebral malaria is more common in older children in areas with lower intensity of transmission, whereas severe malarial anaemia is often seen in children below two years of age in areas with intense transmission \[[@B4]\]. Maternal anaemia and malaria have been reported in areas of unstable malaria transmission in Thailand and in Ethiopia, as well as in areas of stable malaria transmission \[[@B23],[@B24]\]. Regardless of the transmission level and the level of pre-pregnancy immunity against malaria, maternal anaemia remains the most frequent consequence of malaria during pregnancy \[[@B25]\]. Interestingly, we have recently observed a high prevalence of anaemia in pregnant women in these two hospitals and anaemic women were at a higher risk of stillbirth and low birth weight deliveries \[[@B26]-[@B29]\]. Interestingly, seven out 12 patients in the current study had severe anaemia without evidence of multiorgan dysfunction or other manifestations of severe of malaria. Although, these patients fulfilled the WHO criteria for severe malaria \[[@B3]\], perhaps some of these women had severe anaemia and concurrent uncomplicated malaria rather than severe *P. falciparum*malaria. Thus, in such situation these patients would have received blood transfusion and artemisinins combination therapy rather than quinine treatment. There were no maternal deaths in this study, early diagnosis; prompt effective treatment could explain this observation. These women would appear to represent a milder spectrum of disease e.g. severe anaemia and hypotension. It have been shown that, within the broad definition of severe *P. falciparum*malaria there are syndromes associated with mortality rates that are lower (e.g. severe anaemia) and higher (cerebral malaria and metabolic acidosis) \[[@B19]\]. We previously observed that malaria was one of the main causes of high maternal mortality in these two hospitals \[[@B9],[@B30]\]. Maternal mortality is approximately 50% in pregnant women with severe *P. falciparum*malaria, which is higher than in non-pregnant adults \[[@B19]\]. The other limitation of this work is that we could not follow up these women and investigate/report the maternal and perinatal outcomes, and compare them to women with uncomplicated *P. falciparum*malaria and healthy controls. Conclusion ========== Preventive measures for pregnant women such as insecticide-treated bednets and chemoprophylaxis may be beneficial in areas of unstable malaria transmission. Early detection and prompt treatment of severe malaria, especially in pregnant women with hypogylcaemia, are needed. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= AAA and EME carried out the study and participated in the statistical analysis and procedures. MMM carried out the biochemical tests. IA and MIE coordinated and participated in the design of the study, statistical analysis and the drafting of the manuscript. All the authors read and approved the final version. Acknowledgements ================ Authors are very grateful to the women who participated in the study and to all the staff of Medani and Kassala maternity hospitals. This work was funded by The National Fund for promotion of Medical Service, Khartoum, Sudan.

**Specifications Table**TableSubject area*Biochemistry*More specific subject area*Membrane protein, oocyte electrophysiology*Type of data*Graph and figure*How data was acquired*Two electrode voltage clamp and FRAP Assay using a Zeiss LSM 510 confocal microscope*Data format*Filtered and analyzed*Experimental factors*Application of lipid analogue detergent*Experimental featuresThe stability and functionality of solubilized nAChR was examined by fluorescence recovery after photobleaching and two electrode voltage clamp techniquesData source locationN/AData accessibility*Data is supplied in this article* **Value of the data**•The unique approach used to assess functional activity of an ion channel-detergent complex provides a practical and rapid method for screening activity of other membrane protein detergent complex prior to crystallization trials.•The result provided here may forewarn some researchers who are using traditional detergent for the solubilization of membrane protein about the possible effects of detergent structure on channel functionality.•The data can be useful for other researchers investigating the effects of different detergent head groups on the stability of solubilized membrane proteins. 1. Data {#s0005} ======= We provide additional data about the stability and functionality of nAChR solubilized from *Torpedo californica* with the lipid analog detergent, 1-hexadecanoyl-*sn*-glycero-3-phospho-(1′-*rac*-glycerol) (LFG-16). The stability of the affinity purified nAChR-LFG-16 detergent complex was determined after it incorporation into lipid cubic phase (LCP) of 1-(cis-9-Octadecenoyl)-rac-glycero for a period of 30 days using Fluorescence Recovery after Photobleaching (FRAP)([Fig. 1](#f0005){ref-type="fig"}). The functionality of the purified nAChR-LFG-16 detergent complex was studied after reconstitution into *Xenopus* oocyte by mean of two electrode voltage clamp ([Fig. 2](#f0010){ref-type="fig"}). 2. Experimental design, materials and methods {#s0010} ============================================= 2.1. Crude membrane protocol {#s0015} ---------------------------- nAChR extraction was performed homogenizing 60 g of *Torpedo californica* tissue for 4 min in cold room with 120 ml of buffer A (100 mM NaCl, 10 mM Sodium Phosphate, 5 mM EDTA, 5 mM EGTA, 5 mM DTPA, 0.02% Sodium Azide, pH 7.4) mixed with 120 μl of phenyl methane sulfonyl fluoride (PMSF) and 0.112 g of Iodoacetamide. The mixture of buffer and tissue was homogenized using a blender on high-liquefy for 4 min. The homogenate was transferred to centrifuge tubes and centrifuged for 30 min at 6500 rpm at 4 °C. The supernatant was filtered through gauze and centrifuged for 30 min at 40,000 rpmat 4 °C. Consequently, the pellet from this spin was resuspended in 100 mL Buffer B (10 mM Sodium Phosphate, 5 mM EDTA, 5 mM EGTA, 5 mM DTPA, 0.02% Sodium Azide, pH 7.8) mixed with 100 μL PMSF. This mixture was once again spun for 30 min at 6000 rpm at 4 °C. The supernatant from this spin was centrifuged for 30 min at 40,000 rpm at 4 °C. The pellet was resuspended in 25 mL of 40% sucrose storage solution and these crude membranes were properly labeled and stored at −80 °C until ready to use [(Fig. 3)](#f0015){ref-type="fig"}. 2.2. Affinity column purification of solubilized nAChR {#s0020} ------------------------------------------------------ All steps were carried out in the cold room or on ice. In order to solubilize the crude membranes, these were thawed and mixed with a 10% (w/v) detergent solution and DB-1X Buffer (100 mM NaCl, 10 mM MOPS, 0.1 mM EDTA, 0.02% NaN~3~) for a final concentration of detergent 1--4%. The DB-1X buffer was added first, followed by the detergent and finally the crude membranes, which were added drop by drop. This solution was shaken slowly for 1 h and then centrifuged at for 1 h at 40krpm and 4 °C. The supernatant was extracted and used immediately for the affinity-column purification. Approximately 12 mL of previously prepared bromoacetylcholine affinity resin (Bio-Rad Laboratories, Hercules, CA) in a 1.5×15 cm Econocolumn (Bio-Rad Laboratories, Hercules, CA) was drained of storage buffer (40% Sucrose, 2 mM PMSF) was conditioned with 50 mL of ddH~2~O and 50 mL of 1.5 CMC detergent buffer before the supernatant prepared previously was added to the column. The column was washed with 50 mL of 1.5 CMC detergent buffer ([Fig. 1](#f0005){ref-type="fig"}) before the nAChR was eluted with 50 mL of elution buffer. The sample was then concentrated using centrifuge filter with a 100 K cutoff (Amicon Ultra Centrifugal Filters Ultracel 100 K, Millipore Co., Billerica, MA)) and run through a P-10 desalting column (GE Healthcare, Uppsala, Sweden) to remove the carbamylcholine ligand. Our sample was eluted with 5 mL of 1.5 CMC detergent buffer and finally concentrated to 250 μL. Protein concentration was determined using a BCA Protein Concentration Assay (Pierce biotechnology, Rockford, IL) and a sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) was run to verify receptor purity. 2.3. FRAP assays {#s0025} ---------------- FRAP experiments were performed according to the conditions and protocols described by Cherezov et al. (2008) [@bib7], with the following modifications [@bib2], [@bib3]: 50 μl of a solution containing 2.0 mg/ml of ligand-affinity purified nAChR was incubated with (α-BTX) conjugated with Alexa-488 (Invitrogen, Carlsbad, CA) in a 1:2.5 ratio for 1.5--2 h in the dark at 4 °C. The nAChR-detergent-α-BTX complex was mixed with molten monoolein in a 2:3 volume ratio, using a syringe lipid mixer, and mixed until it was completely clear. The nAChR-detergent-α-BTX complex in LCP was placed on a 75×25 mm slides and washed with 1.5 ml of 1.5 CMC detergent buffer solution three times before recovering the LCP-nAChR-detergent-α-BTX complex with a syringe. The LCP- nAChR-detergent-α-BTX was transferred into an automatic sampler, and approximately 0.2 μl of LCP- nAChR-detergent-α-BTX was dispensed into 7 mm diameter wells formed by punching holes into 50 lm thick transfer tape (9482 PC; 3 M, Minneapolis, MN) and pressing onto a glass slide. The LCP-FRAP wells were covered immediately by pressing a coverslip against the slide and flattening with a rubber roll [@bib6]. This procedure was performed quickly to form a tight seal; otherwise, the LCP could dry out and compromise matrix integrity. The entire experimental procedure was performed in an environment with a relative humidity range of 60--80%. 2.4. FRAP Instrument setup and data collection {#s0030} ---------------------------------------------- All FRAP data was collected 24 h after plates were assembled. Data collection for FRAP assays was performed at room temperature using a Zeiss (Thornwood, NY) LSM 510 confocal microscope with an objective of ×40. Five pre-bleach images were used to establish baseline fluorescence, and the laser was triggered to bleach at 75% power, immediately followed by a sequence of 500 images scanning at 2.6% power with a 0.6-s laser scanning delay. All images were obtained and processed using the Zeiss ZEN software. For data analysis each sample was integrated within a 14.0-μm-diameter circular region of interest (ROI~1~). Averaged integrated intensity of another 14.0-μm circular region of interest (ROI~2~), positioned near the bleached ROI~1~, was used to correct for photobleaching from irradiation during the image-acquisition sequence. Fluorescence intensity was corrected by dividing the value of the integrated intensity ROI~1~ in the bleached spot by the average integrated intensity of the ROI~2~. As described by Cherezov et al. [@bib7]. The fractional fluorescence recovery curves, *F*(*t*), were calculated according [@bib1]. 2.5. Injection of oocytes with crude or nAChR detergent complex and two electrode voltage clamp assays {#s0035} ------------------------------------------------------------------------------------------------------ We used a modified version of the protocol originally used by the Miledi and Morales research group [@bib4], [@bib5]. *The Xenopus laevis* oocyte were obtained by surgical extraction, defolliculated and the selected oocyte were microinjected with with 50 nL of 6 mg/mL of crude membrane or 3 mg/mL of 1.5 fold critical micellar concentration nAChR detergent complex, affinity purified from *Tc*. In Padilla et al. 2015 we provide a complete description of both protocols [@bib1] ([Fig. 3](#f0015){ref-type="fig"}). Appendix A. Supplementary material {#s0045} ================================== Supplementary material This research was supported by the National Institutes of Health NIGMS Grants 1R01GM098343, the RISE Program (2R25GM061151), MARC Program (5T34GM007821) and Neuroimaging and Electrophysiology Facility Grant NIH P20 GM 103642. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Supplementary data associated with this article can be found in the online version at [doi:10.1016/j.dib.2015.12.010](http://dx.doi.org/10.1016/j.dib.2015.12.010){#ir0005}. {#f0005} {#f0010} {#f0015}

Introduction {#S1} ============ Age-associated impairment of spatial learning and memory (AISLM) is a common phenomenon in humans and presents a substantial challenge to our health and society ([@B17]). However, the factors and neurobiological mechanisms underlying AISLM remain poorly understood. In humans, viral or bacterial infections during pregnancy can delay the normal development of the central nervous system in the offspring ([@B53]; [@B13]). Chronic inflammation due to maternal infection during pregnancy contributes to the activation of astroglia and microglia and subsequently an increase in the production of proinflammatory cytokines in the brains of offspring. Studies using animal models have shown that inhibition of astrocytic activation through administration of minocycline, an anti-inflammatory, contributes to relieving postoperative cognitive impairment in aged mice ([@B25]). Lipopolysaccharide (LPS), the main toxic components of the cell wall of Gram-negative bacteria and other microorganisms such as *Chlamydia*, *Rickettsia*, and the spirochetes, can induce neuroinflammatory responses, including activation and proliferation of microglia and excessive production of proinflammatory cytokines such as interleukin-6, interferon-gamma, and tumor necrosis factor ([@B11]). These cytokines can affect the normal function of the brain and accelerate brain aging through specific signaling pathways ([@B43]; [@B1]; [@B37]). These proinflammatory factors can also enter the fetal blood circulation and brain through the placental barrier and the blood--brain barrier, leading to increased levels of inflammatory cytokines in the fetal brain. Increased inflammatory cytokine levels can stimulate glial cells, thereby causing lasting adverse effects on fetal growth, fetal development, and postnatal neurobehaviors, or lead to accelerated aging of progeny and AISLM ([@B19]; [@B2]; [@B7]). This may align with an age-related decline in synaptic function, such as damaged synaptic plasticity and neurotransmission, in some brain regions ([@B34]; [@B36]). These changes can impair the synaptic connections between axonal buttons and dendritic spines by affecting the levels of synaptic proteins or the size and number of synapses or dendritic spines in neurons ([@B59]). We have previously shown that exposure to LPS during pregnancy can accelerate AISLM in middle-aged CD-1 mice ([@B28]) and that maternal inflammation due to LPS administration during pregnancy can affect AISLM in their offspring from midlife to senectitude ([@B27]). Moreover, the levels of some synaptic proteins in the hippocampus also change with age, such as an increase in the levels of synaptotagmin-1 (Syt1) and a decrease in those of syntaxin-1 ([@B27]). Together, these observations indicate that early life exposure to inflammation has a long-term effect on AISLM, especially after midlife. Stress is a commonly encountered psychosocial factor. Maternal stress during pregnancy can affect fetal brain development and exert profound neurobiological effects on postnatal motor and affective development ([@B50]; [@B21]). Studies have shown that dams suffering stress during pregnancy exhibit hippocampal neuronal loss and reduced neurogenesis ([@B60]; [@B31]), while prenatal stress in the rat causes long-term spatial memory deficits and hippocampal abnormalities as detected by magnetic resonance imaging ([@B30]). These studies on animals demonstrate the importance of exploring whether stress in adolescence accelerates the AISLM induced by prenatal exposure to inflammation. Synaptic function is extensively impaired with aging, especially in the hippocampus, and may contribute to abnormal neuronal activity and ultimately to cognitive impairment ([@B49]; [@B5]). This synaptic dysfunction may be associated, at least partially, with altered synaptic plasticity resulting from changed hippocampal levels of synaptic proteins such as Syt1, Munc18-1, and SNAP-25, which are associated with AISLM ([@B6]; [@B12]; [@B58]). Although numerous synaptic proteins have been identified, which of these may be involved in AISLM remains poorly understood. Synaptotagmin-1 is abundantly localized at presynaptic vesicles in the brain and acts as the major calcium sensor for mediating fast and synchronous neurotransmitter release ([@B23]; [@B7]). This indicates that Syt1 is important for cognitive function, and further suggests that it may also play a role in AISLM. Additionally, Syt1 expression in the brain can be adversely affected by factors such as aging, inflammation, and stress. Current evidence suggests that dorsal hippocampal levels of Syt1 increase with age, as demonstrated in SAMP8 and CD-1 mice, and this increase is correlated with AISLM ([@B8]; [@B27]). Moreover, maternal exposure to LPS during pregnancy can exacerbate age-related upregulation of hippocampal levels of Syt1 in both the mother and offspring ([@B27]; [@B28]). However, it is not known whether prenatal exposure to inflammation affects *Syt1* gene transcription, or how stress during adolescence affects the normal or accelerated age-related increase in Syt1 protein expression. Activity-regulated cytoskeleton-associated protein (Arc, also known as Arg3.1) is a postsynaptic protein that shuttles between dendrites and nuclear compartments, and is essential for synaptic plasticity and synapse elimination ([@B3]; [@B14]; [@B35]). Arc controls the transport of AMPA receptors (AMPARs), thereby regulating synaptic plasticity, and is necessary for spatial memory consolidation ([@B54]; [@B24]). However, the relationship between the changes of Arc and cognition is difficult to draw. For instance, increased Arc levels may interfere with learning by altering the shape of spinous processes in transgenic mice ([@B26]). However, reducing Arc levels using genetics or antisense oligonucleotides also leads to impaired long-term memory formation ([@B38]; [@B39]). In addition, a few studies have indicated that Arc expression in the brain can be affected by factors such as early life experience, age, and cranial irradiation. In Wistar rats, hippocampal Arc expression declines with age, and stress due to separation from the mother in the neonatal period exacerbates this age-related reduction ([@B46]). To date, it is not known whether prenatal exposure to inflammation changes Arc expression in the brain at different ages, and how stress in adolescence affects normal age-related Arc expression. Accordingly, in this study, we explored whether prenatal exposure to inflammation either with or without stress during adolescence affected spatial learning and memory in young (3 months old) or aged (15 months old) CD-1 mice. We also evaluated whether the protein and mRNA levels of Syt1 and Arc were altered in different hippocampal subregions of mice of different ages and treatments. Finally, we determined the correlations between spatial learning and memory and the measured neurobiological indicators in the groups of different ages and treatments. Materials and Methods {#S2} ===================== Animals and Drugs {#S2.SS1} ----------------- All CD-1 mice (6 weeks old, 20 males and 40 females) were purchased from the Medical Experimental Animal Center of Anhui Province, China. Adaptive feeding was provided for 2 weeks before the experiment. The mice were maintained under a 12-h light/dark schedule (lights on at 07:00) and a temperature of 24 ± 1°C with 55 ± 5% humidity. Males and females were mated at a 1:2 ratio. The next day, the presence of a vaginal plug was designated as gestational day (GD) 0. During GDs 15--17, the mice received a daily intraperitoneal injection of LPS (50 μg/kg) or the same volume of normal saline. After normal childbirth and breastfeeding, offspring were separated from their mothers on postnatal day 21. The offspring of mothers that had received LPS were randomly assigned into two groups (LPS groups), one of which was additionally exposed to stress for 28 days from 2 months of age (LPS + S group). The offspring of mothers receiving normal saline were randomly assigned into two control (CON) groups, one of which was additionally exposed to stress in adolescence (CON + S) for 28 days from 2 months of age. When they reached 3 and 15 months, six males from each group were used for experiments. The schematic representation of the experimental timeline was shown in [Figure 1](#F1){ref-type="fig"}. All animal procedures were performed in compliance with the guidelines published in the National Institutes of Health (NIH) Guide for the Care and Use of Laboratory Animals. The protocol was approved by the Experimental Animal Ethics Committee of Anhui Medical University (No. LLSC20160338). {#F1} Stress Induction {#S2.SS2} ---------------- The stress scheme was performed according to our previous study ([@B29]). The LPS + S and CON + S groups were randomly and daily exposed to one of the following stress modes. They comprised of binding, suspension, lighting, or fasting at night. Four days constituted one cycle, and it lasted for seven cycles. For binding, the mice were fixed in a mesh made of soft wire (35 × 40 cm) to restrict their movement, resulting in psychological anxiety that did not affect their breathing. Binding lasted 30 min on the first day, and was increased by 10 min each time in the next time of binding. In the suspension test, the tail of the mouse was fixed on a crossbar (1.2 m in height) for 30 min on the first day and then increased by 10 min each time in the next time of suspension. For lighting at night, the light was switched on for 30 min at 30-min intervals from 19:00 to 07:00 the next morning. During fasting at night, the feed was removed from 19:00 to 07:00 the next morning, while drinking water was still provided. Morris Water Maze {#S2.SS3} ----------------- The maze consisted of a circular tank (150 cm in diameter, 30 cm in height) placed on a steel frame. The tank was filled with water (21--22°C, depth of 25 cm). A cylindrical platform (10 cm in diameter and 24 cm in height) was fixed in the center of a quadrant in the pool to allow the mouse to escape. The platform was 1 cm below the surface of the water. The periphery of the tank was surrounded by a white curtain, forming a cylindrical shape with a diameter of 3.5 m. Three black conspicuous markers (circles, squares, and triangles) were suspended equidistantly inside the curtain, 150 cm above the ground. A camera system was installed directly above the tank to record the performance of the mouse in the experimental task. In the positioning navigation (learning) phase, the mice were placed on the platform for 30 s on the first day. Then, and in subsequent days, the mice were randomly placed into the water from different quadrants (except for the quadrant of the platform) facing the pool wall. The mice were allowed to swim for 60 s to find the escape platform, and were allowed to rest on the platform for 30 s if they could not find the platform within the 60 s. They were then put back in their cages. The test was performed four times daily with 15-min intervals for 7 days. During the probe trial (to test the memory, on the last day of learning phase), the platform was withdrawn 2 h after the positioning navigation experiment, and the mouse was placed into the water from the opposite quadrant of the target quadrant and allowed to probe the pool for 60 s. Because the distance swam can better reflect the learning ability of aged and elderly mice in the learning phase ([@B51]), the average swimming distance was used as the learning ability, and the percent distance swam in the target quadrant was used as the memory performance. The distance swam was recorded using ANY-maze software (Stoelting, United States). Tissue Preparation {#S2.SS4} ------------------ Approximately 15 days after the behavioral test, the mice were sacrificed by cervical dislocation, decapitated, and their brains quickly removed from the skull. The brains were bisected in the midsagittal plane on dry ice. The left hippocampus was separated from the left hemisphere and stored at −80°C for western blotting. The right hemisphere was fixed in 4% paraformaldehyde at 4°C for 3 days and paraffin-embedded into blocks for hippocampal immunohistochemistry and RNAscope. Immunohistochemistry {#S2.SS5} -------------------- Paraffin-embedded sections of fixed brains (3 μm thick, in the coronal plane) were prepared on a Leica Microtome (Leica RM 2135, Germany). The hippocampal sections were deparaffinized and rehydrated through a series of xylene and ethanol washes. Antigen retrieval was performed by heating the samples in sodium citrate buffer (0.01 mol/L, pH 6.0) for 20 min in a microwave. Sections were treated with a 0.4% Triton X-100, H~2~O~2~, and 5% bovine serum albumin solution to minimize non-specific binding, and then incubated with anti-Syt1 (1:800; S2177, Sigma, United States) and anti-Arc (1:100; Proteintech, United States) primary antibodies overnight at 4°C. The next day, the sections were rewarmed for 45 min at 37°C, washed three times with PBS (Zsbio, ZLI-9062), incubated with a secondary antibody (biotin-labeled goat anti-rabbit IgG) for 20 min at 37°C, and then treated with a streptavidin-biotin-peroxidase complex (Zhongshan Golden Bridge Biotechnology, Beijing, China) at 37°C for 30 min. Sections that were not incubated with the primary antibody served as negative controls. Finally, the sections were washed three times with PBS, stained with diaminobenzidine (Zsbio, ZLI-9018) at room temperature until coloration became visible, and then quickly washed with tap water to stop color development. Two sections from each animal were stained for each protein. All slices were mounted with neutral gum. Images of the whole hippocampus (4 × 10) and subfields (20 × 10) were acquired using a digital scanner (Panoramic MIDI). Images of three hippocampal subregions (cornu ammonis \[CA\]1, CA3, and dentate gyrus \[DG\]) were obtained. Western Blotting {#S2.SS6} ---------------- For protein isolation, the left hippocampus was added to a lysis buffer (RIPA buffer, Haiji Biotechnology Co., Ltd., China), disrupted by ultrasonication, and centrifuged at 4°C for 15 min (13,000 rpm). The supernatant was taken as the extracted protein. Protein concentration was measured using a bicinchoninic acid assay kit (Pierce Biotechnology, United States). The protein was solubilized into an equal concentration and added (4:1) to 5× protein electrophoresis loading buffer, mixed, and boiled (100°C, 10 min). The samples were then applied to a 10% SDS-polyacrylamide gel. Beta-actin (TA-09; Zhongshan Golden Bridge Bio-technology) was used as an internal standard. After cooling to room temperature, the protein samples were applied to polyacrylamide gels. Each gel contained hippocampal protein samples and a pre-stained molecular weight marker (Thermo Fisher, United States). The voltage used for the concentrated gel was 80 V for 30 min, and 120 V for 1 h for separating gels. After electrophoresis, the proteins were transferred to polyvinylidene fluoride membranes (Millipore, United States). The membranes were first blocked with 5% skimmed milk in TBS for 2 h at room temperature, and then incubated with rabbit anti-Arc polyclonal (16290-1-AP; Proteintech) and rabbit anti-Syt1 (S2177; Sigma) antibodies overnight at 4°C. After washing with TBS containing 0.1% Tween 20 (TBS-T; Solarbio, T8220; 3 × 10 min), the membranes were incubated with horse radish peroxidase (HRP)-conjugated anti-rabbit IgG (Zhongshan-Golden Bridge Bio-technology; 1/10000) for 2 h at room temperature. Then, the membranes were rinsed with PBS-T (3 × 10 min) and the immunoreactive protein bands were visualized using an enhanced chemiluminescent ECL reagent (Thermo Fisher, United States). Each antibody revealed a single immunoreactive band corresponding to Arc (45 kDa), beta-actin (43 kDa), or Syt1 (47 kDa). Densitometric quantification of band intensities was performed within the range of linear exposure of the film using Image-Pro Plus 6.0 software (Media Cybernetics, United States). Duplicate samples were averaged for each subject. To control for equal loading, ratios of the optical density for the antibody of interest to the optical density of the antibody directed against beta-actin was calculated for each sample. RNAscope Assay for mRNA Detection {#S2.SS7} --------------------------------- ### Experimental Procedure {#S2.SS7.SSS1} For formalin-fixed, paraffin-embedded (FFPE) tissue, 5-μm thick sections were deparaffinized in xylene and then dehydrated using an ethanol series. A HybEZ hybridization oven was heated to 40°C and a humidity control tray containing distilled water was placed in the oven. Fresh 1× RNAscope Target Retrieval Reagent (ACD, 322000) was prepared and heated to boiling for use. A slide was placed on the slide holder and 5--8 drops of RNAscope H~2~O~2~ were added to the slide, followed by incubation for 10 min at room temperature. Then, the H~2~O~2~ was removed and the slide washed three times with distilled water. The slide was then immersed in 1× RNAscope Target Retrieval Reagent for 15--30 min, washed three times with distilled water, transferred to 100% ethanol for 3 min, and dried at room temperature. Subsequently, a hydrophobic circle was drawn three times around each slice with an Immedge pen and left to dry for 1 min at room temperature. The slide was placed on the holder and five drops of RNAscope protease Plus reagent (ACD, 322381) was added to completely cover the slide. The slide was placed in the HybEZ hybridization oven and incubated at 40°C for 30 min. Then, excess liquid was removed and the slide washed five times with distilled water. Excess liquid was removed and five drops of either the *Arc* (ACD, 316911) or *Syt1* (ACD, 491831) mRNA probe mixture was added to the slide, which was placed in the HybEZ hybridization oven, followed by incubation for 2 h at 40°C. The slide was then washed with 1× wash buffer (2 × 1 min) at room temperature. The liquid on the slide was removed and 4--6 drops of RNAscope multichannel fluorescent second-generation AMP1 (contained in the Multiplex Fluorescent Reagent Kit) were added dropwise onto the slide, which was then incubated at 40°C for 30 min, followed by washing with 1× wash buffer for 2 min at room temperature. AMP2 was added dropwise using the same procedure, followed by incubation at 40°C for 30 min. AMP3 was added dropwise, followed by incubation at 40°C for 15 min. The slide was washed with 1× wash buffer for 2 min, excessive liquid was removed, and 4--6 drops of HRP-C1 were added to completely cover the entire section, following which the slide was placed in the HybEZ hybridization oven and incubated for 15 min at 40°C. Then, the slide was rinsed with fresh 1× wash buffer for 2 min at room temperature. After removing excess liquid, 150--200 μL of TSA plus fluorescent dye was added dropwise to the slide, which was then placed in the HybEZ hybridization oven, incubated at 40°C for 30 min, and rinsed with 1× wash buffer for 2 min. After removing the excess liquid, 4--6 drops of RNAscope multichannel fluorescent second-generation HRP blocker was added to the slide, which was again placed in the HybEZ hybridization oven, incubated for 15 min at 40°C, and rinsed with 1× wash buffer for 2 min. After removing excess liquid, four drops of DAPI were added to the slide which was then incubated for 30 s at room temperature. After removing the DAPI, 1--2 drops of Prolong Gold antiquenching seal was added, and the slide was dried for 30 min in the dark overnight. Slides were stored in the dark at 4°C. Brain sections were observed under a fluorescence microscope. ### Morphometric Analysis and Quantification {#S2.SS7.SSS2} Images were acquired using an Olympus IX71 fluorescent microscope (Olympus, Tokyo, Japan) equipped with a PXL37 CCD camera (Photometrics, Tucson, AZ, United States). For multiplex RNAscope staining and for studies with FFPE tissue specimens, images were acquired using a Zeiss Axioplan M1 microscope (Carl Zeiss Micro Imaging, Göttingen, Germany) equipped with a CRi Nuance multispectral imaging system (Caliper Life Sciences, Cambridge, MA, United States). Overlapping signals from different fluorophores were separated by comparing composite signals against a reference spectral library generated with samples stained with a single color. RNAscope hybridization fluorescence was imaged using a ×40 objective on a laser-scanning confocal microscope (Zeiss LSM700 or LSM780). During each imaging session, both *Arc* and *Syt1* slices that were processed simultaneously in the same hybridization experiment were imaged. The microscope settings were fixed in each imaging session. Images were obtained for six brains per group (at least 6 sections/mouse) and Image J software was used to calculate the mean fluorescence intensity in the three subregions of the hippocampus, i.e., the CA1, CA3, and DG. Statistical Analysis {#S2.SS8} -------------------- All results were expressed as means ± standard error of the mean (SEM) for the parametric data or as 50th (25th/75th) quartiles for non-parametric data. For the performances in the learning phase of the Morris water maze (MWM) task, the data were analyzed using repeated measures analysis of variance (rm-ANOVAs) with day, age, or treatment as independent variables. For comparison of the results among the different groups, *post hoc* analysis was performed using Fisher's least-significant difference test. The parametric data were analyzed using two-way ANOVA with age or treatment as independent variables. For non-normally distributed data, the Kruskal--Wallis H test was used, followed by an extended *t*-test for pair-wise analysis. Pearson's correlation test was used to analyze the correlations between the relative Arc and Syt1 protein/mRNA levels in hippocampal subregions and the performance in all trials of the learning or memory phase in the MWM. Significance was assumed at *P* \< 0.05. All the analyses were conducted using SPSS 21.0 for Windows. Results {#S3} ======= Performances in the MWM {#S3.SS1} ----------------------- ### Learning Phase {#S3.SS1.SSS1} #### Age effects {#S3.SS1.SSS1.Px1} The distance swam declined progressively for all the control mice (*F*~\[6,60\]~ = 173.18, *P* \< 0.001), indicating that these mice were able to learn the task. There was a significant effect of age and interaction of ages × days on distance swam (*F*~\[1,10\]~ = 24.67, *P* = 0.001; *F*~\[6,60\]~ = 2.89, *P* = 0.015). *Post hoc* analysis indicated that 15-month-old mice from the CON group swam significantly longer distances than the 3-month-old mice from the same group (*P* \< 0.01; [Figure 2A](#F2){ref-type="fig"}). Similarly, 15-month-old mice swam significantly longer distances than 3-month-old mice in all the treatment groups (CON + S, LPS, and LPS + S) (*Ps* \< 0.01, *n* = 6 per group; [Supplementary Figures S1A--C](#SM1){ref-type="supplementary-material"}). {#F2} #### Treatment effects {#S3.SS1.SSS1.Px2} For the 3-month-old mice, there were significant differences in swimming distances among the different treatment groups (*F*~\[3,23\]~ = 26.40, *P* \< 0.001). The *post hoc* analysis showed that mice from the LPS and LPS + S groups swam significantly longer distances than the mice from the CON group (*Ps* \< 0.01); however, the differences were only marginal compared with the CON + S group (*P* = 0.075). Meanwhile, there were significant differences between the LPS and CON + S groups (*P* = 0.001) and between the LPS and LPS + S groups (*P* = 0.025; [Figure 2B](#F2){ref-type="fig"}). For the 15-month-old mice, significant differences in swimming distances were found among the different treatment groups (*F*~\[3,23\]~ = 15.08, *P* \< 0.001). Furthermore, the learning swimming distances in the LPS (*P* = 0.001) and LPS + S (*P* \< 0.001) groups were significantly longer than those in the CON and CON + S groups; the difference between the CON and CON + S groups were not significant (*P* = 0.374). Moreover, mice from the LPS + S group exhibited significantly longer swimming distances than those from the LPS group (*P* \< 0.05; [Figure 2C](#F2){ref-type="fig"}). ### Memory Phase {#S3.SS1.SSS2} #### Age effects {#S3.SS1.SSS2.Px1} The 15-month-old mice in the CON group had a significantly lower percentage of distance swam in the target quadrant than the 3-month-old mice from the same group (*t* = 5.64; *P* \< 0.01; [Figure 2D](#F2){ref-type="fig"}), as did those in the CON + S, LPS, and LPS + S groups (*Ps* \< 0.05, *n* = 6 per group; [Supplementary Figures S1D--F](#SM1){ref-type="supplementary-material"}). #### Treatment effects {#S3.SS1.SSS2.Px2} There were significant differences in the percent distance swam among the four groups at both 3 months (*F*~\[3,23\]~ = 55.09, *P* \< 0.001) and 15 months of age (*F*~\[3,23\]~ = 45.32, *P* \< 0.001). The *post hoc* analyses showed that the distance percentage was significantly smaller in the CON + S, LPS, and LPS + S groups than in the CON group (*n* = 6 per group, *Ps* \< 0.01), irrespective of age. Specifically, the distance percentage in the LPS group was smaller than in the CON + S group (*Ps* \< 0.01), but larger than in the LPS + S group (*Ps* \< 0.01; see [Figures 2E,F](#F2){ref-type="fig"}). Levels of Arc and Syt1 in the Hippocampus {#S3.SS2} ----------------------------------------- ### Immunohistochemical Analysis {#S3.SS2.SSS1} Representative photomicrographs of immunolabeled Arc and Syt1 proteins in hippocampal subfields (CA1, CA3, and DG) of 15-month-old mice from the LPS + S group are shown in [Figure 3A](#F3){ref-type="fig"}. Punctate staining is distributed throughout every layer of the CA1, CA3, and DG subregions. Moreover, the distribution of immunoreactivity for both proteins from the different treatment groups differed between older (15 months old) and younger (3 months old) mice, but was more evident in 15-month-old mice (*n* = 6 per group; [Supplementary Figure S2](#SM1){ref-type="supplementary-material"}). The immunoreactivity in both 3- and 15-month-old mice was greater with than without embryonic exposure to inflammation. Meanwhile, stress exposure in adolescence increased the immunoreactivity in the mice also exposed to embryonic inflammation ([Supplementary Figure S2](#SM1){ref-type="supplementary-material"}). {#F3} ### Western Blotting Analysis {#S3.SS2.SSS2} The protein levels of Arc and Syt1 were significantly higher in 15-month-old mice from the CON group than in 3-month-old mice from the same group ([Figures 3B--D](#F3){ref-type="fig"}). Interestingly, the different treatments significantly affected the hippocampal levels of Arc and Syt1, both in the young (*n* = 6 per group, *F*~\[3,23\]~ = 626.78, 371.98; *Ps* \< 0.001) and older mice (*n* = 6 per group, *F*~\[3,23\]~ = 1460.14, 1357.49; *Ps* \< 0.001). Among the 3-month-old mice, those in the CON + S, LPS, and LPS + S treatment groups exhibited significantly higher Arc and Syt1 protein levels than the 3-month-old mice from the CON group (*Ps* \< 0.001). The Arc and Syt1 protein levels in the LPS treatment group were significantly higher than those in the CON + S group (*n* = 6 per group, *Ps* \< 0.001), but lower than those in the LPS + S group (*Ps* \< 0.001). Similarly, among the 15-month-old mice, the Arc and Syt1 protein levels were significantly higher in the three treatment groups than in the CON group (*Ps* \< 0.001). However, the levels of both proteins in the LPS group were significantly higher than those in the CON + S group (*Ps* \< 0.01), but significantly lower than those in the LPS + S group (*Ps* \< 0.01; [Figures 3E,F](#F3){ref-type="fig"}). *Arc and Syt1* mRNA Levels in Different Hippocampal Subregions {#S3.SS3} -------------------------------------------------------------- *In situ* hybridization fluorescent immunostaining showed that *Arc* and *Syt1* transcripts were primarily localized in the pyramidal cell layer ([Figure 4](#F4){ref-type="fig"}). In 3-month-old mice, a significant intergroup difference was found for *Arc* transcripts in the CA1 (*F*~\[3,23\]~ = 38.71, *P* \< 0.001) and CA3 (*F*~\[3,23\]~ = 5.81, *P* \< 0.01) subregions, and for *Syt1* mRNA in the CA3 (*F*~\[3,23\]~ = 124.261, *P* \< 0.001) subregion. Similarly, in 15-month-old mice, significant intergroup differences in *Arc* and *Syt1* mRNA levels were also found in the CA1 (*F*~\[3,23\]~ = 13.98, 20.59; *Ps* \< 0.001) and CA3 (*F*~\[3,23\]~ = 21.42, 142.11; *Ps* \< 0.001) subregions. In addition, for both mRNAs, significant differences were found in the corresponding hippocampal subregions for each treatment group in the 15-month-old mice compared with the 3-month-old mice (*n* = 6 per group; [Supplementary Figures S3](#SM1){ref-type="supplementary-material"}, [S4](#SM1){ref-type="supplementary-material"}). {#F4} ### *Post hoc* Analysis {#S3.SS3.SSS1} In the 3-month-old mice, *Arc* and *Syt1* transcript levels were both significantly increased in all the treatment groups (*n* = 6 per group, *Ps* \< 0.05) compared with those in the CON group. Mice in the LPS + S group presented significantly higher *Arc* mRNA levels than the LPS, CON + S, and CON groups in the CA1 (*Ps* \< 0.001) and CA3 (*Ps* \< 0.05) subregions, as did *Syt1* mRNA levels in the CA3 (*Ps* \< 0.001). In the CA1 and DG, significant differences in *Syt1* mRNA levels were found only between the LPS + S and CON groups (*P* = 0.039, *P* = 0.021). The 15-month-old mice from the LPS and LPS + S groups showed significantly increased *Arc* and *Syt1* mRNA levels in the CA1 and CA3 subregions (*n* = 6 per group, *Ps* \< 0.01) compared with the CON group, with similar changes observed in the LPS + S group relative to the LPS and CON + S groups (*Ps* \< 0.05). The *Arc* and *Syt1* mRNA levels in the CA3 subregion were higher in the LPS group than in the CON + S group (*Ps* \< 0.01). However, in the DG, significant differences in *Syt1* mRNA levels were found only between the LPS + S and CON groups (*P* = 0.024; [Figure 5](#F5){ref-type="fig"}). {#F5} Correlations Between Performance in the MWM Test and Arc and Syt1 Expression Levels {#S3.SS4} ----------------------------------------------------------------------------------- ### Correlations Between Performance and Protein Levels {#S3.SS4.SSS1} In the 3-month-old mice, hippocampal Arc and Syt1 levels showed a significant positive correlation with the learning swimming distance (*r* = 0.908, 0.925; *Ps* \< 0.01) and a negative correlation with the percent of distance swam in the target quadrant (*Ps* \< 0.01) for the four groups combined. For each group, the learning swimming distance was positively correlated with Arc levels in the LPS + S group (*P* = 0.027) and with Syt1 levels in the LPS and LPS + S groups (*Ps* \< 0.05), and the memory distance percentage was negatively correlated with Arc (*r* = −0.940, 0.890; *P* = 0.005, 0.017) and Syt1 levels (*r* = −0.976, −0.914; *P* = 0.001, 0.011) in the LPS and LPS + S groups. In the 15-month-old mice, the levels of Arc and Syt1 were also positively correlated with the learning swimming distance (*r* = 0.855, 0.837; *Ps* \< 0.001), and negatively correlated with the percent distance swam in the target quadrant (*Ps* \< 0.05) for all the groups combined. For individual treatments, there were negative correlations between performance and protein levels in almost four groups (*Ps* \< 0.05; [Table 1](#T1){ref-type="table"}). ###### The correlations between the performance in the MWM test and hippocampal synaptic protein levels. Cognitive phase Age Group Synaptic protein ----------------- ----------- --------- ------------------- ------------------- Swimming 3 months CON 0.784 (0.065) 0.631 (0.179) distance CON + S 0.609 (0.200) 0.802 (0.055) LPS 0.805 (0.053) 0.838(0.037)\* LPS + S 0.862(0.027)\* 0.872(0.024)\* 15 months CON 0.876(0.022)\* 0.974(0.001)\*\* CON + S 0.842(0.036)\* 0.874(0.023)\* LPS 0.900(0.015)\* 0.897(0.015)\* LPS + S 0.904(0.013)\* 0.981(0.001)\*\* Distance 3 months CON −0.013(0.98) −0.127(0.810) percentage CON + S −0.635(0.175) −0.608(0.200) in target LPS −0.940(0.005)\*\* −0.976(0.001)\*\* quadrant LPS + S −0.890(0.017)\* −0.914(0.011)\* 15 months CON −0.929(0.007)\*\* −0.954(0.003)\*\* CON + S −0.928(0.008)\*\* −0.946(0.004)\*\* LPS −0.990(0.000)\*\* −0.917(0.010)\* LPS + S −0.934(0.006)\*\* −0.908(0.012)\* \* P \< 0.05; \*\* P \< 0.01. CON, untreated control group; LPS, lipopolysaccharide treatment group; S, group of mice exposed to stress. ### Correlations Between Performance and mRNA Levels {#S3.SS4.SSS2} As shown in [Table 2](#T2){ref-type="table"}, in the 3-month-old mice, the swimming distances in the learning phase showed significant positive correlations with the levels of *Arc* mRNA in the CA1 of mice from the LPS and LPS + S groups (*Ps* \< 0.05), and in the CA3 of mice from the CON + S group (*r* = 0.871; *P* = 0.024). There were significant positive correlations for *Syt1* mRNA levels in the CA3 of mice from the LPS (*r* = 0.884, *P* = 0.019) and LPS + S groups (*r* = 0.882, *P* = 0.02). The memory distance percentage was negatively correlated with the CA1 levels of *Arc* mRNA in the LPS + S and LPS groups (*Ps* \< 0.05), CA1 levels of *Syt1* mRNA in the LPS + S, LPS, and CON + S groups (*Ps* \< 0.05), and CA3 levels of *Syt1* mRNA in the LPS + S group (*P* \< 0.01). ###### The correlations between the performance in the MWM test and hippocampal mRNA levels. Cognitive phase Age Group *Arc* mRNA *Syt1* mRNA ----------------- ----------- --------- ------------------- ------------------- --------------- ------------------- ------------------- --------------- Learning 3 months CON 0.630 (0.18) 0.401 (0.431) −0.018(0.973) 0.605 (0.203) 0.547 (0.261) −0.338(0.513) swimming CON + S 0.514 (0.297) 0.871(0.024)\* −0.186(0.724) 0.418 (0.410) 0.410 (0.420) −0.232(0.658) distance LPS 0.922(0.009)\*\* 0.357 (0.487) −0.616(0.193) 0.884(0.019)\* 0.763 (0.077) −0.096(0.857) LPS + S 0.901(0.014)\* 0.367 (0.474) −0.017(0.975) 0.789 (0.062) 0.882(0.020)\* −0.398(0.435) 15 months CON −0.176(0.738) −0.630(0.180) −0.567(0.241) 0.416 (0.412) 0.601 (0.198) 0.778 (0.069) CON + S 0.953(0.003)\*\* 0.981(0.001)\*\* −0.049(0.926) 0.982(0.000)\*\* 0.756 (0.082) 0.321 (0.535) LPS 0.878(0.022)\* 0.800 (0.056) −0.450(0.370) 0.879(0.021)\* 0.938(0.006)\*\* −0.322(0.534) LPS + S 0.875(0.022)\* 0.825(0.043)\* 0.637 (0.174) 0.925(0.008)\*\* 0.897(0.015)\* 0.496 (0.317) Distance 3 months CON −0.030(0.955) −0.543(0.266) −0.545(0.264) 0.438 (0.385) 0.480 (0.335) 0.579 (0.228) percentage CON + S −0.633(0.178) −0.102(0.847) 0.329 (0.524) −0.813(0.049)\* 0.673 (0.143) −0.243(0.643) in target LPS −0.867(0.025)\* −0.776(0.07) 0.309 (0.551) −0.851(0.032)\* −0.336(0.515) 0.473 (0.343) quadrant LPS + S −0.975(0.001)\*\* −0.434(0.39) −0.258(0.621) −0.903(0.014)\* −0.973(0.001)\*\* −0.124(0.816) 15 months CON 0.163 (0.758) 0.583 (0.224) 0.349 (0.498) −0.533(0.288) −0.790(0.062) −0.707(0.117) CON + S −0.661(0.153) −0.817(0.047)\* −0.324(0.531) −0.783(0.066) −0.992(0.000)\*\* −0.181(0.731) LPS −0.988(0.000)\*\* −0.933(0.007)\*\* 0.096 (0.856) −0.967(0.002)\*\* −0.816(0.048)\* −0.164(0.756) LPS + S −0.918(0.01)\*\* −0.849(0.032)\* −0.315(0.544) −0.966(0.002)\*\* −0.952(0.003)\*\* 0.045 (0.932) \* P \< 0.05; \*\* P \< 0.01. CA, cornu ammonis; DG, dentate gyrus; CON, untreated control group; LPS, lipopolysaccharide treatment group; S, group of mice exposed to stress. In the 15-month-old mice, a significant positive correlation was found between *Arc* mRNA levels and the learning swimming distance in the CA1 of the LPS + S, LPS, and CON + S groups (*Ps* \< 0.05) and CA3 of the LPS + S and CON + S groups (*Ps* \< 0.05). Additionally, a significant negative correlation was found between *Arc* mRNA levels and the memory distance percentage in the CA1 of the LPS + S and LPS groups (*Ps* \< 0.05) and CA3 of the LPS + S, LPS, and CON + S groups (*Ps* \< 0.05). A significant positive correlation was found between *Syt1* mRNA levels and the learning swimming distance in the CA1 of the LPS + S, LPS, and CON + S groups (*Ps* \< 0.05) and the CA3 of the LPS + S and LPS groups (*Ps* \< 0.05). A significant negative correlation was also found between *Syt1* mRNA levels and the memory distance percentage in the CA1 of the LPS + S and LPS groups (*Ps* \< 0.05) and CA3 of the LPS + S, LPS, and CON + S groups (*Ps* \< 0.05; [Table 2](#T2){ref-type="table"}). Discussion {#S4} ========== Embryonic exposure to inflammation resulting from infections in pregnant mothers may be an important mechanism underlying age-related behavioral impairment, especially AISLM, in both humans and rodents. Several studies have demonstrated that there is a gender-dependent effect on memory in normal aging as well as on prenatal exposure to different insults; however, the results have not been consistent. For instance, we have previously shown that AISLM exists in older female, but not male, Kunming mice ([@B9]). [@B20] demonstrated that there are no significant differences between the sexes in age-related cognitive impairment (including spatial and working memory) in rats. [@B52] showed that prenatal exposure to inflammation leads to increased anxiety-related behavior in adult male offspring when compared with adult female offspring. [@B15] also reported a male-specific effect for intrauterine LPS exposure on synaptic pruning in the hippocampus of adult offspring, while the observed LPS-induced anxiety-related behavior deficits did not differ between the sexes. Our recent study showed no difference between female and male offspring in the effect of maternal prenatal LPS exposure on AILSM in CD-1 mice ([@B27]). In the present study, we only selected male offspring, and the results indicated that stress in male offspring during adolescence will accelerate the AISLM caused by exposure to inflammation during the embryonic stage. Moreover, we also demonstrated that there was an age-related increase in the expression of Arc and Syt1 at both the protein and mRNA levels in the dorsal hippocampus, which could be further increased following embryonic exposure to an inflammatory environment, with or without exposure to stress during adolescence. Interestingly, these altered expression levels of synaptic proteins might be linked to impaired spatial learning and memory caused by different treatments in embryos and/or adolescents. The Effects of Prenatal Exposure to Inflammation Coupled With Stress Exposure in Adolescence on AISLM in Midlife {#S4.SS1} ---------------------------------------------------------------------------------------------------------------- Several studies have confirmed that exposure to adverse environments in early life may lead to permanent changes in key organs or tissues during the developmental "window" period; in turn, these changes may trigger preset procedural and structural changes in relevant brain regions related to processes such as memory formation and stress response ([@B4]; [@B55]). Our results indicated that learning and memory ability decreases gradually with age. Moreover, mice that were exposed to inflammatory environments as embryos showed a significantly worse performance in the MWM test than unexposed mice, regardless of age (3 or 15 months). Studies have shown that stressors presented in the late prenatal or early postnatal periods, the periods when the formation of brain circuits associated with early development occurs, have long-term effects on the behavior of offspring ([@B57]). In this study, all the mice from the LPS, LPS + S, and CON + S treatment groups exhibited worse memory performance than the CON group. Furthermore, mice from the LPS and LPS + S treatment groups had worse memory performance than those in the CON + S group at both ages. These results suggested that exposure to an inflammatory environment during the embryonic period is an important accelerator of AISLM, and exposure to stress during adolescence may further aggravate this effect. The Age-Related Changes in Hippocampal Arc and Syt1 Expression {#S4.SS2} -------------------------------------------------------------- Arc is a postsynaptic protein critical for memory consolidation, especially that of spatial memory ([@B38]; [@B41]). However, relatively few studies have reported on age-related changes in *Arc* gene expression in the brain, and the reported results have been inconsistent. The hippocampus from aged (24--27 months) male Fischer-344 rats has a similar number of *Arc* mRNA-expressing pyramidal cells in the CA1 and CA3 regions compared with that in the adult (10--12 months) hippocampus, as labeled by fluorescence *in situ* hybridization ([@B32]). However, the hippocampus of aged mice (24 months old) has reduced levels of *Arc* mRNA, as detected with RT-RNA and *in situ* hybridization, in C57BL/6J mice relative to that in young (6 months old) mice ([@B40]). In contrast, aged (24 months of age) Long-Evans male rats have increased basal Arc protein levels in the CA1 field of the hippocampus when compared with young rats ([@B16]). Our results suggested that in the hippocampi of aged CD-1 mice, Arc transcription and translation are both upregulated, in agreement with previous results ([@B16]; [@B33]). Syt1 plays a modulatory role in endocytosis, and has recently been implicated in the calcium-sensitive trafficking of postsynaptic AMPARs to facilitate long-term potentiation ([@B22]; [@B56]). Therefore, Syt1 and Arc can synergistically regulate the number of AMPARs on the postsynaptic membrane to regulate synaptic plasticity and, consequently, learning and memory. In the current study, the hippocampal Syt1 and *Syt1* mRNA levels were significantly higher in 15-month-old mice than in 3-month-old mice from the CON group. These findings about Syt1 protein levels were consistent with the results from CD-1 and SAMP8 mice ([@B48]; [@B27]; [@B28]), but *Syt1* mRNA levels were inconsistent with the result from SAMP8 mice ([@B8]). This discrepancy may be due to different strains of rodents and experimental manipulation. The Effect of Prenatal Inflammatory Insult Coupled With Stress Exposure in Adolescence on Hippocampal *Arc and Syt1* Gene Expression {#S4.SS3} ------------------------------------------------------------------------------------------------------------------------------------ To date, no study has investigated the effect of maternal inflammatory insult during gestation on the transcript levels of synaptic protein-related genes in the offspring, or the effects of prenatal exposure to inflammation coupled with stress exposure in adolescence on the expressions of synaptic protein-related genes in offspring of different ages. In the current study, aged mice had higher levels of Arc and Syt1 at both the protein (detected by western blotting) and mRNA (detected by RNAscope) levels in the whole hippocampus than young mice. Interestingly, different treatments significantly affected hippocampal Arc and Syt1 levels, regardless of age. These results suggested that prenatal exposure to inflammation can enhance hippocampal expression of the *Arc* and *Syt1* genes in the offspring, irrespective of age, and additional exposure to stress in adolescence can further increase these levels. Notably, although exposure to stress (without LPS treatment) during adolescence could also significantly increase the expression levels of the *Arc* and *Syt1* genes in the hippocampus from youth to midlife, the extent of this effect was significantly less than the effect of prenatal exposure to inflammation. The fact that the aged mice had higher hippocampal expression levels of the *Arc* and *Syt1* genes than the young mice, and that the above-mentioned treatments elicited different effects, indicated that a prenatal inflammatory insult can aggravate the age-related increase in the hippocampal levels of Arc and Syt1, and exposure to stress during adolescence can further exacerbate this change in the levels of synaptic proteins. The Correlation Between Cognitive Performance and *Arc* and *Syt1* Gene Expression {#S4.SS4} ---------------------------------------------------------------------------------- Memory loss due to aging is accompanied by a downregulation of AMPARs that mediate fast excitatory synaptic transmission. The dynamic regulation of AMPARs is crucial for supporting basal synaptic transmission and plasticity. Arc-mediated regulation of the endocytic pathway modulates the basal levels of AMPARs and increased Arc expression leads to reduced AMPA transmission through GluR2/3 removal. To date, studies investigating the correlation between Arc expression and memory have shown inconsistent results. For example, Arc protein expression is decreased in the hippocampus during memory loss in C57/BL6 mice ([@B18]), while a decline in *Arc* gene expression might be associated with age-dependent memory decline in male Swiss albino mice ([@B45]). Aged Long-Evans male rats with memory impairment have increased basal Arc protein levels in the CA1 field of the hippocampus ([@B16]). In addition, cells with increased Arc levels exhibit impaired long-term depression, while chronic Arc overexpression is linked to abnormal spine structure ([@B26]). Syt1 may be a key regulator of AMPAR insertion in postsynaptic spine membranes ([@B22]). Increased Syt1 levels can change synaptic transmission and also have a marked effect on the morphology of neurons ([@B23]). In rats, stress can increase hippocampal Syt1 expression, which may be partially responsible for changes in neuronal morphology, biochemistry, and behavior, including learning and memory deficits ([@B47]). Therefore, we hypothesized that Arc and Syt1 overexpression are likely not to be beneficial for learning and memory. Behavioral testing can induce stress-related changes in plasticity markers as well as in gene expression ([@B10]). To mitigate this potential bias, we allowed 2 weeks of "recovery" from behavioral testing before performing biochemical analysis. In this study, the spatial learning and memory ability of 15-month-old mice decreased significantly and the protein and mRNA levels of Arc and Syt1 increased in parallel when compared with their younger (3 months old) siblings. This suggests that increased hippocampal expression of the *Arc* and *Syt1* genes may be involved in the impaired spatial learning ability and memory resulting from conditions such as aging, stress, and prenatal exposure to inflammation. Indeed, the correlation analysis indicated that performance in the MWM test was significantly correlated with changed Arc and Syt1 protein and mRNA levels for all the mice. Specifically, in the normally aging mice, the learning swimming distance positively, and memory distance percentage negatively, correlated with the hippocampal levels of Arc and Syt1 in 15-month-old mice, but not in 3-month-old mice ([Table 1](#T1){ref-type="table"}). Interestingly, because prenatal exposure to inflammation or exposure to stress in adolescence affected the cognitive behaviors and hippocampal levels of Arc and Syt1 in the 3- and 15-month-old mice, the hippocampal protein levels of Arc and Syt1 in the treatment groups were positively correlated with the learning swimming distance and negatively correlated with the percent distance swam ([Table 1](#T1){ref-type="table"}). Overall, the pattern of correlation between MWM performance and mRNA or protein levels was similar. However, in the untreated controls, no significant correlation was recorded between the MWM performance and *Arc* and *Syt1* mRNA levels, irrespective of age. For the groups undergoing different treatments, learning swimming distance was positively, and memory distance percentage negatively, correlated with the levels of *Arc* and *Syt1* mRNA in the CA1 and CA3, but not DG, hippocampal subregions of the 3-month-old mice ([Table 2](#T2){ref-type="table"}). Similar correlations were observed in the CA1 and/or CA3 subregions at 15 months of age in all the treatment groups (LPS, LPS + S, and CON + S; [Table 2](#T2){ref-type="table"}). The above results indicated that only the increased hippocampal expression of the Arc and Syt1 proteins was associated with AISLM. Moreover, increased hippocampal expression of Arc and Syt1 at both the protein and mRNA levels was associated with a decline in spatial learning and memory during "pathological" aging due to prenatal inflammatory insult, whether or not this course was exacerbated by exposure to stress in adolescence. In conclusion, LPS administration to CD-1 mice during pregnancy can lead to long-lasting enhanced hippocampal expression of the *Arc* and *Syt1* genes in offspring from adolescence onward, which can affect the behavior of the mice. In particular, increased expression of these genes can lead to impaired spatial learning and memory as well as the normal aging process. Moreover, exposure to stress during adolescence may further accelerate the AISLM and enhance the hippocampal expression of the *Arc* and *Syt1* genes. Notably, functional differences exist between the left and right hippocampi of rodents, depending on the demand for short-term or long-term memory ([@B44]; [@B42]). Due to experimental limitations, we did not consider the left--right anatomical and functional differences of the rodent hippocampus. However, all the animal experiments were performed using the same standards. Although this study was limited by the specimen number and gender, it did reveal that adverse events during pregnancy might contribute to accelerated aging and AISLM in male offspring. Further research is needed to explore whether similar effects occur in females. Data Availability Statement {#S5} =========================== All datasets generated for this study are included in the article/[Supplementary Material](#SM1){ref-type="supplementary-material"}. Ethics Statement {#S6} ================ The animal study was reviewed and approved by the Association of Laboratory Animal Sciences and the Center for Laboratory Animal Sciences at Anhui Medical University. Author Contributions {#S7} ==================== Z-ZZ and Z-QZ conceived and designed the study and performed the western blotting and RNAscope tests. Z-ZZ drafted the manuscript. S-YS, H-HG, and Y-FW performed immunohistochemistry and the behavioral test. LC, LX, and Q-GY participated in the study design and statistical analysis. G-HC and FW revised the manuscript. All authors read and approved the final manuscript. Conflict of Interest {#conf1} ==================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** This work was financially supported by the National Natural Science Foundation of China (81370444 and 81671316), Natural Science Foundation for the Youth of China (81301094), College Natural Science Foundation of Anhui Province (KJ2016A3), Natural Science Foundation for the Youth of Anhui Province (1708085QH182), and Research Project of Anhui Medical University (2018xkj029). This funding played an important part in the design of the study and collection, analysis, and interpretation of the data and in the writing of the manuscript. Supplementary Material {#S9} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fnagi.2020.00157/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: Yu-Min Kuo, National Cheng Kung University, Taiwan [^2]: Reviewed by: Ana María Genaro, Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET), Argentina; Deep R. Sharma, Albert Einstein College of Medicine, United States; Akash Gautam, University of Hyderabad, India; Meghraj Singh Baghel, Banaras Hindu University Varanasi, India, in collaboration with reviewer AG [^3]: ^†^These authors have contributed equally to this work

1. Introduction {#sec1} =============== Neuroinflammation plays a key role in the regulation of aging, Alzheimer\'s disease (AD), Parkinson\'s disease (PD), Huntington\'s disease (HD), multiple sclerosis (MS), stroke, depression, dementia, and metabolic disorders such as hypertension and diabetes \[[@B1]\]. Neuroinflammation is also the pathogenic hallmark of aging-related neurodegenerative conditions \[[@B1], [@B2]\]. Therefore, anti-inflammatory strategies could be efficient prophylactic and therapeutic management strategies for a number of central nervous system (CNS) disorders \[[@B3], [@B4]\]. CNS disorders may develop owing to chronic microglial activation. Glial cells, particularly microglia, are immune cells in the CNS that are responsible for the maintenance of normal homeostasis as well as repair after injury in the brain \[[@B5]\]. Although activated microglia are required for host defense and debris clearance in the brain, chronic microglial activation is toxic to the CNS \[[@B6]\]. Conversion of normal microglia to the toxic microglial phenotype (known as the M1 phenotype) is responsible for the initiation of inflammatory cascades in the CNS, particularly the production of reactive oxygen species (ROS), nitric oxide (NO), proteases, arachidonic acids, excitatory amino acids, and cytokines \[[@B7]\]. These neurotoxic substances trigger the oxidative stress and are responsible for the disruption of the architecture and functions of neurons, consequently leading to synaptic degeneration and neurodegeneration \[[@B8]\]. These neuroinflammatory and neurotoxic cascades are responsible for the hippocampal neuronal damage leading to cognitive dysfunction \[[@B9]\]. Inhibition of the neuroinflammation or activation of the endogenous antioxidant system might be a better alternative for the repairment of this damage. Particularly, activation of the Nrf2 and its mediated antioxidant enzyme, HO-1, can not only inhibit the inflammatory cascades but also increase the neuronal survival and hippocampal neurogenesis \[[@B10]\]. Extensive neuropharmacology research has succeeded in finding novel drug candidates for the treatment of CNS disorders. However, they have failed to prove their efficacy in human biological systems during clinical trials. Potential reasons for these failures were suggested to be differences in tissue physiology of the CNS, unstable pharmacokinetics, or difficulty in crossing the blood-brain barrier (BBB) \[[@B11], [@B12]\]. Given the lack of proper allopathic medications to treat neuroinflammatory disorders, there is a growing interest in complementary and alternative medication, including nutraceuticals. Consumption of dietary nutraceuticals with neuroprotection could prevent CNS diseases, overcoming the limitations of allopathic drug delivery to the CNS. Previous reports highlight the beneficial effects of broccoli and its active compound, sulforaphane, in neurodegenerative disorders. Although broccoli has a mild effect against age-related neuroinflammation, it does not show a significant effect against lipopolysaccharide- (LPS-) induced inflammatory conditions \[[@B13]\]. Another independent study has revealed that sulforaphane can reverse the hyperammonemia-induced glial activation, neuroinflammation, and disturbances in neurotransmitter receptors in the hippocampus that impaired spatial learning \[[@B14]\]. The anti-inflammatory effect of sulforaphane has also been previously reported in rat primary microglia \[[@B15]\]. These previous studies suggest that bioconversion of glucoraphanin in the broccoli may increase the yield of sulforaphane. It could lead to better health benefits for patients with neuroinflammatory disorders. Either the induction of myrosin production or the increase in myrosinase activity can increase the bioconversion of glucoraphanin into sulforaphane \[[@B16]\]. The presence of some other proteins, such as epithiospecifier protein (ESP), can also convert glucoraphanin to sulforaphane nitrile, without affecting sulforaphane\'s anti-inflammatory effects \[[@B17]\]. An increase in myrosinase activity with reduced ESP activity can selectively enhance sulforaphane yield in the broccoli sprout, making it a better candidate for the treatment of several pathological conditions, including cancer, inflammation, neurodegeneration, and aging \[[@B18]\]. Pharmacokinetic studies have revealed that, in mice, sulforaphane has good absorption and distribution patterns, in various tissues of the body, including the brain, lungs, heart, liver, kidneys, and muscles, after oral administration \[[@B19]\]. Previous reports suggest that sulforaphane and its metabolites do not readily cross the BBB \[[@B20]\]. However, the disruption of the BBB during neuroinflammation and neurodegenerative conditions allows sulforaphane to permeate the BBB and enhances its anti-inflammatory effects along the brain axis \[[@B20]\]. Former reports have revealed that activation of myrosinase in steamed broccoli sprouts can enhance the yield of sulforaphane \[[@B20]\]. Similarly, pulsed electric field (PEF) pretreatment enhances the production of glucosinolate in the broccoli flower and stalk. It also increased anthocyanin\'s production in red cabbage \[[@B21], [@B22]\]. Thus, we postulated that PEF might also be effective in enhancing the production of sulforaphane in broccoli sprouts. In the current study, we are to reveal whether PEF treatment can increase the sulforaphane yield in broccoli sprouts or not. We also try to figure out the difference in the biological activity of broccoli sprout before and after PEF treatment. To determine the molecular mechanism underlying the sulforaphane-enriched broccoli sprout-mediated neuroprotective effect, we assessed the effect of sulforaphane-enriched broccoli sprouts on lipopolysaccharide- (LPS-) induced proinflammatory responses in both acute and chronic microglial activation *in vitro*. *In vivo* study will be done whether sulforaphane-enriched broccoli sprouts improve the scopolamine-induced memory impairment in mice or not. 2. Materials and Methods {#sec2} ======================== 2.1. Reagents {#sec2.1} ------------- The pulsed electric field (PEF) was purchased from HVP 5 (Elea, DIL, Quakenbrueck, Germany). Lipopolysaccharide (LPS) and sulforaphane were purchased from Sigma Chemical (St. Louis, MO). Dulbecco\'s modified Eagle medium (DMEM), Fetal bovine serum (FBS), and penicillin-streptomycin were purchased from Invitrogen (Carlsbad, CA, USA). Antibodies for cyclooxygenase 2 (COX-2), *β*-actin, GAPDH, and histone-3 were obtained from Santa Cruz Biotechnology (Dallas, Texas, United States). The antibody of inducible nitric oxide synthase (iNOS) was purchased from Abcam (Cambridge, UK). The primary antibody for *α*-tubulin, C-Jun N-terminal kinase (JNK), p38, extracellular signal-regulated kinases (ERK), NF-*κ*B, I-*κ*B, pI-*κ*B, C-Jun, pC-Jun, C-Fos, and pC-Fos were purchased from Cell Signaling (Beverly, MA, USA). Various ELISA kits like interleukin-6 (IL-6), prostaglandin E2 (PGE2), tumor necrosis factor alpha (TNF-*α*), and interleukin-*β* (IL-1*β*) were acquired from R&D Systems (Minneapolis, MN, USA). 2.2. Animals {#sec2.2} ------------ Male ICR mice (6 weeks; 25-30 g) were purchased from Orient Bio, Seoul, Korea. Four or five mice were placed in each cage and acclimatized for one week in laboratory conditions with food and water ad libitum. They were maintained under temperature (23 ± 1°C) and relative humidity (60 ± 10%) conditions and in a 12/12 h light and dark cycle. Animal handling and experimental procedures were conducted in accordance with the Principles of Laboratory Animal Care (GIACUC-R2017016) and the Animal Care and Use Guidelines of Gachon University, Korea. The experimental protocol is shown in [Scheme 1](#sch1){ref-type="fig"}. 2.3. PEF-Broccoli Preparation and Extraction {#sec2.3} -------------------------------------------- Broccoli sprout extraction was performed according to a previous study with slight modifications \[[@B23]\]. Each whole broccoli sprout (SB) as well as the hypocotyl (H), cotyledon (C), and radicle (R) were treated with PEF, at 0--7 kJ, for 3 s each, followed by freeze-drying and grinding to make fine powder ready for myrosinase activity induction. The powdered broccoli or its parts were incubated with 10x volume of distilled water, and the suspension was kept at 37°C for a total of 2 h for the induction of myrosinase activity, yielding a greater amount of glucoraphanin to sulforaphane as a final product. During this conversion, sulforaphane nitrile could also be formed with the help of epithiospecifier protein (ESP). This conversion was prevented by placing/heating the mixture at 80°C for 10 min. The mixture was cooled down by keeping it in the ice. As a final step, the suspension was mixed with methanol making the solvent as 70% methanol in total and the extraction was performed. Extraction was performed with continuous stirring and sonication for about 3 h in total. We used each sample as the following; broccoli sprout cotyledons (C) or sprout cotyledons with both PEF and myrosinase activity (C-P) and hypocotyls from broccoli sprouts (H) or hypocotyls from broccoli sprouts treated with both PEF and myrosinase activity (H-P). The supernatant was filtered (HYUNDAI Micro No. 20 filter paper, Korea) and evaporated using a rotary evaporator to remove the methanol. The extract was free-dried, and the lyophilized powder was used for subsequent experiments. Preparation of PEF-broccoli samples is shown in [Scheme 2](#sch2){ref-type="fig"}. 2.4. Broccoli Extract HPLC Standardization {#sec2.4} ------------------------------------------ High-performance liquid chromatography (HPLC) analysis was performed to measure the sulforaphane content in the broccoli sprout extract. The HPLC analysis was performed using a Waters system (Waters Corp., Milford, MA), consisting of a separation module (e2695) with an integrated column heater, autosampler, and photodiode array detector (2998). The UV absorbance was monitored at 200--400 nm. Quantification was conducted by integrating the peak areas at 235 nm. The injection volume was 10 *μ*L. A column (250 × 4.6 mm; particle size, 5 *μ*m; Phenomenex, USA) was installed in a column oven and maintained at 25°C. The mobile phase was composed of water containing 1% acetic acid (solvent A) and acetonitrile (solvent B). The flow rate was 0.5 mL/min. The gradient was 0.0 min, 10% B; 5 min, 20% B; 15.0 min, 40% B; 25.0 min, 60% B; 35.0 min, 90% B; and 40.0 min, 10% B. The re-equilibration time between runs was 20 min. 2.5. Cell Culture {#sec2.5} ----------------- The murine microglia cell (BV2) was used as a representative cell of brain microglia. The neuroblastoma cell (N2a) was used as a representative cell line of the neuron \[[@B24]\]. BV2 cells were kindly obtained by Dr. E. Choi at Korea University (Seoul, Korea), while N2a cells were obtained from the Korean Cell Line Bank (Seoul, Korea). BV2 & N2a cells were maintained in high-glucose DMEM supplemented with 10% heat-inactivated FBS and 1% penicillin (1 × 10^5^ U/L) and streptomycin (100 mg/L), in a humidified incubator with 5% CO~2~ at 37°C. 2.6. Cell Treatment and Cytotoxicity Assay {#sec2.6} ------------------------------------------ Cytotoxicity of the samples was evaluated through 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay also called the cell viability assay \[[@B25]\]. BV2 cells were seeded in 96-well plates overnight, and they were treated (pre/post) with different concentrations of broccoli extract with or without LPS. LPS (100 ng/mL) was added 30 min after sample treatment in case of pretreatment of prophylactic treatment condition. LPS activation in the seeded cells occurred 30 min before the sample treatment in case of posttreatment/therapeutic treatment strategy. Cells were incubated for 24 h after LPS activation, for the nitrite and cell viability assays. In case of neuronal cell viability, conditioned media from 24 h treated BV2 cells were transferred to the seeded N2a cells and incubated for another 24 h as described previously \[[@B26]\]. Treated cells were incubated for an hour with the MTT solution of 0.5 mg/mL concentration that will stain the viable cells into a blue color. The MTT solution was suctioned out, and 200 *μ*L of dimethyl sulfoxide was added to each well that converts the blue-stained cells to a purple-colored solution. The optical density (OD) was measured using a plate reader at 570 nm. The results were expressed as a percentage of the LPS-treated cells (LPS-treated group). 2.7. Nitric Oxide (NO) Measurement {#sec2.7} ---------------------------------- Nitric oxide production inhibition by the treatment of broccoli extract and sulforaphane on LPS-stimulated BV2 cells was performed through Griess assay as described previously with slight modification \[[@B27]\]. BV2 cells were seeded in a 96-well plate (4 × 10^4^ cells/well) and activated with 100 ng/mL LPS, in the presence or absence of different concentrations of broccoli extract or sulforaphane, for 24 h. The nitrite level in the culture media was measured using Griess reagent (1% sulfanilamide and 0.1% N-1-napthylethylenediamine dihydrochloride in 5% phosphoric acid). A total of 50 *μ*L of supernatant was mixed with an equal volume of Griess reagent, and OD was measured at 570 nm. NG-Mono-methyl-L-arginine (L-NMMA), a well-known nitric oxide synthase (NOS) inhibitor, was used as a positive control. The pretreatment and posttreatment conditions were performed to evaluate its effect for prophylactic and therapeutic purposes. 2.8. Measurement of TNF-*α*, IL-1*β*, IL-6, and PGE2 Production {#sec2.8} --------------------------------------------------------------- Cells were plated in a 6-well plate at a density of 1.5 × 10^6^ cells/well in DMEM and incubated for 24 h, to measure TNF-*α*, IL-1*β*, PGE2, and IL-6 production. The cultures were prepared and stimulated with LPS in the presence or absence of sample. After a 24 h incubation, the supernatant from the culture medium was harvested. The levels of PGE2, TNF-*α*, IL-1*β*, and IL-6 were measured. PGE2 was measured using a competitive enzyme immunoassay kit (Cayman Chemical, Ann Arbor, MI, USA), and TNF-*α*, IL-1*β*, and IL-6 were measured using ELISA development kits (R&D Systems, Minneapolis, MN, USA). The % of CV was set below 10 for every ELISA assay. 2.9. NF-*κ*B Assay {#sec2.9} ------------------ Nuclear and cytosolic extracts from treated microglial cells were prepared using a Nuclear/Cytosolic Extraction Kit (Active Motif, Carlsbad, CA) according to the manufacturer\'s protocol. Protein levels of NF-*κ*B, histone-3, I-*κ*B, and pI-*κ*B were determined using western blot analysis. Expression of nucleolic and cytosolic NF-*κ*B was measured using histone-3 and *β*-actin as loading controls, respectively. The expressions of I-*κ*B and pI-*κ*B in the cytosolic fraction were observed. The absence of *β*-actin expression in the neucleolic fraction suggested the clear separation of the neucleolic and cytosolic fractions during fractionation, without any contamination. Densitometry analysis of the bands was performed using ImageMaster™ 2D Elite software (version 3.1, Amersham Pharmacia Biotech, Buckinghamshire, UK). 2.10. Western Blot Analysis {#sec2.10} --------------------------- Western blot analysis was conducted as previously described \[[@B26]\], with slight modification. Proteins obtained from BV2 cells (6 × 10^5^ cells/well), which were seeded in a 6-well plate, were used for western blot analysis. Total proteins (30 *μ*g) from each group were separated by 10% SDS-PAGE gel electrophoresis, transferred to nitrocellulose membranes, and incubated with primary antibodies against tubulin, iNOS, COX-2, ERK, pERK, JNK, pJNK, p38, pp38, NF-*κ*B, histone-3, *β*-actin, I-*κ*B, pI-*κ*B, C-Fos, pC-Fos, C-Jun, pC-Jun, and *α*-tubulin. Membranes were incubated with horseradish peroxidase-conjugated secondary antibodies, and protein bands were visualized using ECL Western Blotting Detection Reagents (Amersham Pharmacia Biotech). Densitometry analysis of the bands was performed using ImageMaster™ 2D Elite software (version 3.1, Amersham Pharmacia Biotech). 2.11. Y-Maze Test {#sec2.11} ----------------- The Y-maze test was performed to evaluate the spatial memory or perception of the mice after scopolamine-induced neuronal injury. The Y-shaped maze having 5 : 20 : 10 of width : length : height was prepared, and the three arms of Y shape were allocated as A, B, and C. Mice were trained in the Y-maze before the start of the experiment. Mice were randomly divided into groups of saline, donepezil (2 mg/kg, p.o.), and different extracts of the broccoli sprout (200 mg/kg, p.o.). Mice for the donepezil and extract-administered groups were challenged with scopolamine (1.2 mg/kg, i.p.) 30 min after drug administration. Donepezil is a well-known acetylcholine esterase inhibitor and is widely used as a positive control for the scopolamine-induced memory impairment model \[[@B28], [@B29]\]. Scopolamine-administered and sample treated mice were kept in the center of the maze, and the mice were allowed to enter into the arms of the maze. Mice were let to habituate for 2 min, and their entry in the arms for 8 min was evaluated. Mouse entry is set when the mouse body (from nose to tail) was fully entered into the arms. Only the entry of the mouse in all the 3 different arms consecutively was assigned with point 1 for each arm entered. Alternation behavior was defined as 3 consecutive entries into 3 different arms of the maze. Spatial perception ability was calculated according to the formula below \[[@B30]\]. $$\begin{matrix} {\text{Voluntary alternation behavior rate }\left( \% \right) = \frac{N_{\text{alteration}}}{N_{\text{entries}} - 2} \times 100,} \\ \end{matrix}$$where *N*~alterations~ is the number of times alternation behavior was observed (scored by points), and *N*~entries~ is the total number of arm entries. 2.12. Novel-Object Recognition Test (NORT) {#sec2.12} ------------------------------------------ In order to determine the role of the broccoli sprout extract enriched with sulforaphane in memory boost, a novel object recognition test (NORT) was performed. The mouse was placed into a 45 cm × 45 cm × 50 cm box containing the novel object for about 5 minutes to habituate in the testing environment. Following the habituation, response of the mouse (time) for object recognition was recorded for 3 minutes. On the third day of testing, 1 of the objects was replaced with a new object, and the response time for the new object recognition was recorded. The concentration of mouse to recognize the new object was evaluated by the recognition index (%) as follows: $$\begin{matrix} {\text{Recognition index}\left( \% \right) = \left\lbrack \left. {\text{time novel}/\text{time novel} + \text{time familiar}} \right) \right\rbrack \times 100.} \\ \end{matrix}$$ Here, time novel is the time spent exploring the novel object, and time familiar is the time spent exploring the familiar object. 2.13. Passive Avoidance Task {#sec2.13} ---------------------------- A passive avoidance test was performed using identical boxes which are illuminated or nonilluminated of the size (20 × 20 × 20 cm), separated by a guillotine door (5 × 5 cm) as described elsewhere \[[@B31]\]. The illuminated compartment contained a 50 W bulb, and the floor of the nonilluminated compartment (20 × 20 × 20 cm) was composed of 2 mm stainless steel rods spaced 1 cm apart. For an acquisition trial, mice were placed in the illuminated compartment and, after 10 s, the door between the two compartments was opened. When mice entered the dark compartment, the door automatically closed and an electric foot shock (0.25 mA) was delivered for 3 s through the stainless steel rods. Twenty-four hours after the acquisition trial, the retrieval trial was conducted by placing the mice in the illuminated compartment. The latency to enter the dark compartment after opening the door was recorded. Cut-off latency was set at 600 s to avoid ceiling effects. 2.14. Acetylcholinesterase Activity Assay {#sec2.14} ----------------------------------------- Colorimetric assay was performed to determine the acetylcholinesterase activity using acetylthiocholine enzyme and acetylthiocholine iodide substrate as described previously \[[@B32]\], with slight modifications. The mouse brains were quickly harvested after CO~2~ euthanasia and well homogenized in a homogenization buffer (12.5 mM sodium phosphate buffer, pH 7.0, 400 mM NaCl) using a glass Teflon homogenizer (EYELA, Japan), and the supernatant was used for the acetylthiocholine activity assay as described previously \[[@B32]\]. In brief, 0.02% Tanshinone congeners were prepared in dimethyl sulfoxide and mixed with acetylthiocholine iodide solution (75 mM), buffered Ellman\'s reagent (10 mM), 5,5′-dithio-bis(2-nitrobenzoic acid), and 15 mM sodium bicarbonate. The total mixture was incubated for 30 min for reaction at room temperature. Absorbance was measured at 410 nm immediately after adding the enzyme source to the reaction mixtures, and readings were taken at 30 s intervals for 5 min. Donepezil was used as a positive control. 2.15. Data and Statistical Analysis {#sec2.15} ----------------------------------- All results are expressed as mean ± standard error of the mean (SEM). Statistical significance between experimental groups was determined by using one-way analysis of variance (ANOVA) followed by the Tukey post hoc test using GraphPad Prism 5 (GraphPad Software Inc., La Jolla, CA, USA). Statistical significance was set at *P* \< 0.05. Each experiment was performed in triplicate. 3. Results {#sec3} ========== 3.1. PEF Exposure Enriched Sulforaphane Content in Broccoli Sprouts {#sec3.1} ------------------------------------------------------------------- We confirmed the content of sulforaphane as a surrogate marker, using HPLC, for the quality control of broccoli sprouts. [Figure 1](#fig1){ref-type="fig"} shows the chromatogram of broccoli sprouts compared to that of sulforaphane. The linearity of the compound was calculated using five concentrations. The sulforaphane content of adult broccoli is found to be almost 4.5-fold lesser than that of the broccoli sprouts. Enzymatic activation of broccoli sprouts increased the sulforaphane content almost 1.2-fold more than that of the normal broccoli sprout. Sulforaphane content was increased by 2.5-fold in a broccoli sample treated with PEF only. When the enzymatically activated broccoli sprout was treated with PEF, sulforaphane yield was 4.2-fold, in particular in the cotyledon, in comparison to the broccoli sprouts. Our experiment revealed that the sulforaphane content of broccoli sprouts, particularly in the hypocotyls and cotyledons, was increased to the highest extent with the combined treatment of PEF and enhanced enzymatic activity. The radicle of the broccoli had the lowest amount of sulforaphane, which was not altered, even with exposure to PEF and enzymatic activity. Compared to broccoli sprouts, adult broccoli contains less amount of sulforaphane, which was increased with activation of enzymatic activity, PEF exposure, and the combination of both conditions as shown in [Figure 1](#fig1){ref-type="fig"}. Overall, increased enzymatic activity and PEF exposure showed a synergistic effect in upregulating the amount of sulforaphane in broccoli sprouts. 3.2. Broccoli Sprouts Exposed to PEF Inhibited Nitrite Production in LPS-Activated Microglia {#sec3.2} -------------------------------------------------------------------------------------------- The ability of young and adult broccoli to inhibit NO was more pronounced when the broccoli sprouts were exposed to PEF and enzyme activity. Broccoli plants exposed to this combination were approximately 2-fold more effective at inhibiting nitrite production, compared to those exposed to either PEF or enzymatic activity only. Compared to the hypocotyls, broccoli plant cotyledons exposed to enzyme and PEF showed the highest potency. After confirming the effect of PEF and enzyme activity on broccoli, we evaluated differences in its activity, during pre-treatment and post-treatment, on LPS-activated microglial cells. The pattern of activity, as well as the extent of potency, was similar to its effects on nitrite production during pre- and post-treatment. Broccoli sprout cotyledons (C) exposed with PEF (C-P) showed better potency in comparison to normal hypocotyls (H) or PEF-exposed hypocotyls (H-P) as shown in [Figure 2](#fig2){ref-type="fig"}. From this screening, we select 100 *μ*g/mL of the concentration of C, C-P, H, and H-P for further experiment and mechanism study. 3.3. Broccoli Sprouts Exposed to PEF Inhibited the Expression of iNOS and COX-2, during 6 and 24 h LPS Activation {#sec3.3} ----------------------------------------------------------------------------------------------------------------- LPS-mediated inflammation is mainly characterized by increased nitrite production and the expression of iNOS and COX-2 \[[@B27]\]. The expression of iNOS and COX-2 significantly increased in BV2 cells treated with LPS \[[@B26]\]. The PEF-enzyme activity increased the inhibitory effect of the cotyledons and hypocotyls of broccoli sprouts on the expression of these inflammatory proteins. Further, the expression of these proteins in the BV2 cells returned to almost normal levels with cotyledon and hypocotyl treatment, even in the presence of LPS in microglial cells. Sprout broccoli cotyledons (C) and cotyledon-P (C-P) were less effective at inhibiting the expression of COX-2 than that of iNOS. However, C-P and H-P significantly inhibited COX-2 expression after LPS activation as shown in [Figure 3](#fig3){ref-type="fig"}. Interestingly, normal broccoli H showed a higher potency to inhibit COX-2 expression 24 h following microglial activation. Taken together, untreated broccoli plants as well as those exposed to PEF effectively inhibited LPS-induced iNOS and COX-2 expression. 3.4. Broccoli Sprout Cotyledons/Hypocotyls Exposed to PEF Modulated Effector Signaling, Specifically ERK and p38 Phosphorylation {#sec3.4} -------------------------------------------------------------------------------------------------------------------------------- MAPK effector signaling pathways control the production of inflammatory mediators and proinflammatory cytokines. The activation (phosphorylation) of MAPK proteins (i.e., p38, JNK, and ERK) is responsible for controlling transcription factors and ultimately the production of inflammatory mediators \[[@B33]\]. LPS significantly increased JNK, ERK, and p38, and this effect was inhibited by C-P. C-P inhibited the phosphorylation of ERK and p38 to an unexpectedly high extent as shown in [Figure 4](#fig4){ref-type="fig"}. 3.5. Broccoli Sprout Cotyledons/Hypocotyls Exposed to PEF Modulated Effector Signaling Short-Term and Chronic LPS Activation in BV2 Microglial Cells {#sec3.5} ---------------------------------------------------------------------------------------------------------------------------------------------------- Both 6 h and 24 h LPS activation sufficiently activated MAPK phosphorylation, which was slightly or significantly altered by C, C-P, H, and H-P treatment. Although both C-P and H-P extensively inhibited ERK phosphorylation during the 6 h (short-term) and 24 h (chronic) LPS activation, the inhibitory effect of cotyledons was weak. H samples, however, increased ERK phosphorylation during chronic activation. The inhibitory effect of C and C-P treatment on p38 was very promising during the 6 h short-term activation and 24 h chronic activation as shown in [Figure 5](#fig5){ref-type="fig"}. Tubulin was used as a loading control for these experiments. 3.6. Broccoli Sprout Cotyledons Exposed to PEF Inhibited NF-*κ*B- and AP-1-Mediated Transcription of Inflammatory Proteins {#sec3.6} -------------------------------------------------------------------------------------------------------------------------- NF-*κ*B and AP-1 are the major transcription factors that are responsible for altering the production of inflammatory proteins and proinflammatory cytokines in LPS-activated BV2 cells. Both the cotyledons and hypocotyls of broccoli sprouts can inhibit NF-*κ*B activity; however, this effect was more prominent following exposure to PEF. LPS treatment significantly upregulated nuclear NF-*κ*B and decreased cytosolic NF-*κ*B. This effective translocation was necessary for further transcription. However, this cascade was reversed by PEF-treated broccoli sprout treatment. In addition to inhibiting NF-*κ*B translocation, broccoli C-P and H-P also inhibited the phosphorylation of I-*κ*B and increased its inactive form. Histone-3 was used as a loading control for nuclear proteins, while B-actin was used for cytosolic proteins as shown in [Figure 6](#fig6){ref-type="fig"}. No significant changes were measured in AP-1 signaling after broccoli treatment. 3.7. Broccoli Sprout Cotyledons Exposed to PEF Inhibited Production of Proinflammatory Cytokines and Inhibited Neuronal Death Induced by Activated Microglia {#sec3.7} ------------------------------------------------------------------------------------------------------------------------------------------------------------ Next, we evaluated the effects of broccoli cotyledons and hypocotyls on the production of inflammatory cytokines. The cotyledons and hypocotyls of normal broccoli sprouts, as well as those exposed to PEF and enzyme, significantly inhibited the production of proinflammatory cytokines, such as TNF-*α*, IL-6, IL-1*β*, and PGE2. We demonstrated that broccoli plants showed the highest effect inhibiting IL-6 and IL-1*β* production and the lowest efficacy in inhibiting TNF-*α* and PGE2 production. The low potency by that in inhibiting COX-2 production is due to weak potency in inhibiting PGE2 production as shown in [Figure 7](#fig7){ref-type="fig"}. After confirming the anti-inflammatory potential of C and C-P, we checked its role in the neuronal survival against activated microglia-induced toxicity to neurons. LPS-induced activation of the microglia resulted in the production of various inflammatory mediators which are lethal to neuronal cells, but treatment of C and C-P to the activated microglia lowered the inflammatory cascades and hence made it possible to inhibit neuronal death by lowering the Bax/Bcl2 ration and expression of cleaved caspase-3. In both of the cases, C-P showed a slightly higher potential to that of C alone. This result provides a strong cue that C and C-P might prevent neuronal death, and hence, they can prevent neurodegeneration induced either by neuroinflammation/activated microglia or by other toxicities. 3.8. Broccoli Samples with or without PEF Increased the Nuclear Factor Erythroid 2--Related Factor 2 (Nrf2) and Heme Oxygenase- (HO-) 1 Expression in the Normal BV2 Cells and against Scopolamine-Induced Amnesia in Mouse Brain Tissue Samples {#sec3.8} ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Sulforaphane is known as an Nrf2 activator; in our study, we also observed that treatment of C and C-P significantly increased the expression of Nrf2 and antioxidant protein HO-1 in normal as well as LPS-activated microglia as shown in [Figure 8](#fig8){ref-type="fig"}. The increased HO-1 and Nrf2 are higher in the case of the PEF-exposed sample, and it must be because of the increased sulforaphane content in it. This suggests that the better potency of C-P to induce Nrf2/HO-1 activation in direct as well as LPS-activated microglia not only can show its antioxidant effect directly in the brain cells but also can lower the activated microglia-induced inflammatory cascades. In addition to this, SB-PEF or SB-Enz-PEF has the highest amount of sulforaphane, which is responsible for the inhibition of the neuroinflammation and neurodegeneration against scopolamine toxicity. In the animal brain, we clearly noticed that SB-Enz-PEF showed a sharp increase in Nrf2 expression ([Figure 8](#fig8){ref-type="fig"}) suggesting that increased Nrf2 was responsible for the improved memory, cognition, and increased latency time in the passive avoidance test. 3.9. Sprout Broccoli Samples Repair the Scopolamine-Induced Memory and Cognitive Impairment In Vivo {#sec3.9} --------------------------------------------------------------------------------------------------- Oral administration of the sprout broccoli sample 30 min before scopolamine administration significantly inhibited the spontaneous alteration almost 3-fold in comparison to the untreated control group ([Figure 9](#fig9){ref-type="fig"}). The sprout broccoli treated with PEF and sprout broccoli with enzyme activation and PEF treatment showed the highest and significant ability to repair the memory impairment. SB-PEF and SB-Enz-PEF bring the value of spontaneous alteration almost similar to that of untreated control mice. The role of SB-PEF and SB-Enz-PEF on memory impairment was reaffirmed through the NORT assay. The broccoli samples showed a significantly improved ability to recognize the novel object compared to that of scopolamine-administered mice. Interestingly, the effects of SB-Enz and SB-Enz-PEF for the novel object recognition were higher than that of the well-known positive control, donepezil. Additionally, acetylcholine esterase activity in the brain samples revealed that only the sprout broccoli with enhanced enzyme activity and SB-Enz-PEF significantly inhibited the level of AchE in the brain which might be responsible for the memory improvement against scopolamine toxicity. Beside these changes, normal/PEF broccoli further improved the cognitive ability of mice against scopolamine-induced memory impairment as evidenced by the passive avoidance test in scopolamine-induced acute and chronic models of memory impairment. Scopolamine dramatically reduced the latency time while the sprout broccoli with PEF and SB-Enz-PEF treatment remarkably recovered the latency time in the acute model. In case of chronic impairment, the broccoli samples increased the latency time suggesting their capacity to inhibit scopolamine toxicity. In both cases, SB-PEF and SB-Enz-PEF showed better potency than did the positive control donepezil ([Figure 9](#fig9){ref-type="fig"}). Donepezil and all the broccoli samples lowered the Bax/Bcl2 ration while only the broccoli samples lowered the expression of cleaved caspase-3 suggesting that the broccoli sample, especially SB-Enz-PEF, possesses higher potency to increase cell survival and decrease neuronal death in the dementia model like scopolamine treatment. Inhibition of the apoptosis-related proteins in SB-ENZ, SB-PEF, and especially SB-Enz-PEF further clarified that inhibition of neuronal death induced survival which could also take part in the higher potency to improve the cognitive function against scopolamine-induced amentia mice. 4. Discussion {#sec4} ============= Sulforaphane is an active compound of broccoli, *Brassica oleracea* var. italic, which is produced after conversion of glucoraphanin in the presence of the myrosinase enzyme \[[@B34]\]. Increased amounts of glucoraphanin or myrosinase activity can enhance sulforaphane production in broccoli or broccoli sprouts. We selected broccoli sprout for the current study since it contains more glucoraphanin than does the adult broccoli \[[@B16]\]. Previous studies suggested that steaming broccoli sprouts increases the enzymatic conversion of glucoraphanin to sulforaphane by lowering sulforaphane nitrile \[[@B35]\]. PEF pretreatment promotes the production of glucosinolate, including glucoraphanin, in broccoli flowers and stalk \[[@B21]\]. Thus, we exposed broccoli sprouts to PEF and myrosinase and evaluated its effect on sulforaphane production. The amount of sulforaphane was increased after enzymatic activity, which was further elevated following PEF exposure. In this study, the elevated sulforaphane content in the broccoli sprout showed better anti-inflammatory activities than did the untreated control, emphasizing the importance of enzymatic activation and PEF treatment. Accumulating evidences regarding the anti-inflammatory potential of natural products and their isolated compounds focus on the scavenging of oxidative stress and downregulating the inflammatory cascades \[[@B36]\]. One of the key molecules participating in such inflammatory disease conditions, especially in neuroinflammatory disorders, is NO \[[@B37]\]. NO in the inflammatory condition is produced by iNOS, which mainly exaggerates oxidative stress in the neurons leading to their degeneration \[[@B38]\]. Inhibition of iNOS or NO production, therefore, could mitigate oxidative stress and further inflammatory cascades in neurodegenerative diseases. Previous studies reported that broccoli sprout inhibited inflammation in endothelial cells \[[@B39]\] and in LPS-treated mice \[[@B13]\]. Additionally, sulforaphane inhibited iNOS-mediated NO production in activated microglia \[[@B40]\]. Along with iNOS-mediated NO production, COX-2-mediated prostaglandin release is also the key pathogenic event in various inflammatory conditions \[[@B41]\]. Elevated expression of iNOS and COX-2 can occur with toxin exposure and in neuroinflammatory diseases. The chronic overexpression of iNOS and COX-2 is typical in AD, PD, neuropathic pain, etc., in which overexpression of these proteins mainly occurs in microglia \[[@B42], [@B43]\]. Therefore, minimizing the activation of COX in neuroinflammatory conditions is a desirable approach to achieving the neuroprotective effects. In the current study, enhanced expression of sulforaphane in broccoli sprout dramatically downregulated the COX-2-mediated PGE2 production in activated microglia. Although C-P did not significantly alter the expression of COX-2 in long-term microglial activation, its potency in inhibiting iNOS was much better in long-term LPS exposure than that in short-term microglial activation. Previously, sulforaphane is reported to downregulate COX-2 expression \[[@B44]\]. Our study not only confirms these existing evidences but also provides noble insights exploring that PEF treatment further increased the NO inhibiting potential of broccoli sprout through elevated sulforaphane content. MAPK effector signaling is responsible for altering the expression and secretion of inflammatory mediators as well as proinflammatory cytokines \[[@B45]\]. MAPK proteins such as ERK, JNK, and p38 play essential roles in neuroinflammation \[[@B46]\]. P38, in particular, is responsible for the induction of apoptosis, differentiation, and regulation of inflammatory responses. LPS activates p38 downstream from TLR4 activation, promoting proinflammatory cytokines, such as TNF-*α*, IFN-*γ*, IL-1*β*, IL-12, IL-6, and IL-23, which are responsible for neuroinflammation \[[@B47], [@B48]\]. Similarly, JNK pathways can further phosphorylate c-Jun and respond to cytokines, like TNF-*α* and IL-1*β*, and growth factors \[[@B49]\]. The activation of JNK pathways plays a significant role in Tau pathology; therefore, inhibition of JNK/NF-*κ*B pathways is beneficial for AD and other neuroinflammatory conditions \[[@B50]\]. A growing body of evidences also suggests that MEK/ERK pathways are responsible for the elevated level of TNF-*α*, IL-1*β*, IL-6, and iNOS following ischemia, which further highlights their neuroinflammatory role \[[@B50]\]. Thus, modulation of p38, JNK, and ERK could be the potential therapeutic strategy in various inflammatory and neurodegenerative conditions in the brain. In the present study, the broccoli sprouts exposed to PEF enzyme significantly inhibited the phosphorylation of ERK, JNK, and p38, during the 30 min and 6 h LPS activation; p38 activation alone was maintained for 24 h of LPS activation. The sulforaphane-enriched broccoli did not change in the activation (phosphorylation) of c-Jun. However, H-P showed the significant inhibition of C-Fos in LPS-activated microglia. This result suggests that PEF-exposed broccoli hypocotyls might have a role in inhibiting pC-Fos-mediated signaling for inflammatory cascades. MAPK signaling controls NF-*κ*B-mediated transcription of inflammatory mediators such as cytokines and chemokines. NF-*κ*B/AP-1 and MAPK pathways play a key role in the production of cytokines such as TNF-*α*, IL-6, and IL-8 in BV2 microglial cells \[[@B51], [@B52]\]. In our study, C-P significantly inhibited the phosphorylation of ERK, JNK, and P38 nonspecifically, during the 30 min, 6 h, and 24 h LPS-induced microglial activation. This alteration might induce the significant inhibition of nuclear NF-*κ*B while increasing cytosolic NF-*κ*B. The inhibition of NF-*κ*B-mediated transcription of inflammatory proteins was further confirmed by the increased and decreased expression of I-*κ*B and pI-*κ*B, respectively. The inhibition of MAPK-NF-*κ*B was further characterized by the significant inhibition of TNF-*α* and IL-6 production in LPS-activated BV2 microglial cells. C-P showed the highest potency for all these events, and this might be due to the higher amount of sulforaphane. Previously, broccoli sprout extract and sulforaphane are reported as activators of well-known antioxidant molecules Nrf2/HO-1 \[[@B53]\]. Treatment of C and C-P activated the expression of Nrf2 and HO-1 as did by the sulforaphane treatment in normal and LPS-activated microglial cells. This effect of broccoli extract and sulforaphane can further inhibit the inflammatory cascades either through inhibiting TLR4-mediated inflammatory cascades or by their antioxidant potentials. Therefore, it is thought to be a potential mechanism that indicates the antineuroinflammatory effect of sulforaphane. Nrf2 inhibits the neuroinflammation as well as neurodegeneration against various models of brain disorders in both *in vitro* and *in vivo* experimental models \[[@B54], [@B55]\]. Previous reports suggested that sulforaphane protected neurons against rotenone-induced toxicity in an *in vivo* model which is mediated through the activation of the Nrf2 pathway \[[@B56]\]. In this study, SB-Enz-PEF significantly improved the scopolamine-induced spontaneous alteration as determined by the Y-maze test, novel object recognition test, and AchE activity inhibition test indicating that sulforaphane could improve memory impairment in different neurological disorders. SB-Enz-PEF also increased latency time in PAT assay. Moreover, through *in vitro* analysis, we found that the SB-Enz-PEF-treated conditioned medium from LPS-stimulated BV2 cells not only increased the neuronal cell survival but also attenuated the apoptotic proteins in neurons. Our data demonstrated that the protective effects of SB-ENZ-PEF are mediated through the Nrf2 signaling pathway; however, the overall neuroprotection of broccoli and sulforaphane might be mediated through the antiapoptotic effects, in particular downregulating caspase-3 activation. Sulforaphane-enriched broccoli extract, therefore, could mediate the neuroprotection in various neurodegenerative models, and the plausible pathway for the neuroprotection could be due to the combination of Nrf2 activation and its antiapoptotic effects. Taken together, in this study, we provided the new insight of increased-sulforaphane-mediated protective effects in neuroinflammatory models. Being multifunctional, identification of a particular therapeutic target for phytochemicals/nutraceuticals to show desired biological activity is a key issue for the new drug development \[[@B57], [@B58]\]. However, the technique that increased the particular bioactive compounds, as PEF-induced sulforaphane in this study, could enhance the therapeutic benefit of phytochemicals. Additionally, a time point study of iNOS/COX-2 and MAPK signaling in this study could be the magnificent cues for the further *in vivo* experiment including both acute (such as ischemia, traumatic brain injury) and chronic (such as AD, PD) inflammatory conditions. Our findings demonstrated that sulforaphane-enriched broccoli sprout showed antineuroinflammatory and neuroprotective effects *in vitro* and showed the protective effects in mice against scopolamine-induced amnesia *in vivo* through Nrf2 activation. Finally, in most studies, the biological activity of sulforaphane, specifically of (−)L-isothiocyanato-4R-(methyl-sulfinyl)-butane, is demonstrated using the racemic mixture, despite the fact that humans are exposed only to the R-enantiomer through their diet. Therefore, it would be tempting to determine the role of R- and S-sulforaphane as future independent studies. The overall findings of our current study are summarized in [Figure 10](#fig10){ref-type="fig"}. This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (NRF-2014M3A9B6069338) and also by a grant from BK bio (R0005491), Republic of Korea. Data Availability ================= All the data have been included in the manuscript and it can also be provided by the corresponding author on request. Conflicts of Interest ===================== No competing financial interests exist. Authors\' Contributions ======================= Hyuk Joon Choi, Sun Yeou Kim, and Lalita Subedi hypothesized and designed the experiment; Lalita Subedi, KyoHee Cho, and Young Un Park performed the experiment; and Lalita Subedi wrote the manuscript. Hyuk Joon Choi and Sun Yeou Kim revised and finalized the manuscript. {#sch1} {#sch2} {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} {#fig7} {#fig8} {#fig9} {#fig10} [^1]: Guest Editor: Francisco J.B. Mendonça Júnior

Background {#Sec1} ========== Currently, two out of three Australians will develop some form of skin cancer before the age of 70 \[[@CR1]\]. In 2010 there were over 778,000 for non-melanoma skin cancer and 11,545 new cases of melanoma diagnosed. More than 2000 people die from skin cancer every year \[[@CR2]\] and 99% of non-melanoma skin cancers and 95% of melanomas are caused by overexposure to solar ultraviolet (UV) radiation \[[@CR3]\]. Protecting skin from exposure to UV radiation is the simplest and most effective way to prevent skin cancer \[[@CR3], [@CR4]\]. Research suggests that reducing children's exposure to UV radiation, particularly in the first 15 years of life, significantly reduces their risk of developing skin cancer later in life \[[@CR4]\] The World Health Organization's (WHO) *Sun Protection and Schools* report \[[@CR5]\] outlines that schools need to implement both structural and organisational strategies to ensure sustainable sun protection for students and staff. In New South Wales (NSW) approximately 650,000 children aged between five and 12 years spend up to 7 h per day, up to 40 weeks per year attending primary school \[[@CR6]\]. As students participate in both organised and recreational outdoor activities during peak UV times, the NSW Department of Education provide schools with sun protection guidelines \[[@CR7]\] and recommend they join Cancer Council NSW (CCNSW) SunSmart Program \[[@CR8]\]. The SunSmart Program, launched in NSW in 2008, is based on the Health Promoting Schools principles \[[@CR9]\] and supports childcare services and schools to develop and implement a comprehensive sun protection policy \[[@CR8]\]. To be eligible for SunSmart membership, schools must submit a sun protection policy addressing 10 recommendations within three key areas: influencing the environment, modifying behaviour and integrating education \[[@CR8]\]. To maintain SunSmart membership, schools must review and resubmit their sun protection policy to CCNSW every 3 years \[[@CR8]\]. The SunSmart Program has been evaluated in three phases over the last 10 years via national survey of Australian primary schools' sun protection policy and practices that monitors trends in primary schools' sun protection policies and practices, and provides comparisons between SunSmart and non-SunSmart schools across Australia \[[@CR10], [@CR11]\]. Results demonstrated improved sun protection practices being adopted in SunSmart primary schools compared to non-SunSmart schools. This included them having more comprehensive sun protection policies and practices, sport lessons being held earlier in the day to avoid peak UV times and parents being provided with sun protection information \[[@CR11]\]. As of April 2015, 78% (*n* = 1983) NSW primary schools were members of the SunSmart Program. The national survey identified opportunities for strengthening sun-safe behaviours of primary school children attending SunSmart schools. This included increasing the number of students wearing sun-safe hats, the availability and promotion of sunscreen and an increase in staff role-modelling sun-safe behaviours themselves \[[@CR5]\]. These three behaviours were identified for inquiry in this study as evidence has shown the wearing of sun-safe hats, clothing and sunscreen in children has the potential to reduce the risk of skin cancer \[[@CR4]\]. There is some research to suggest that adult role modelling may also influence a child's sun protection behaviours \[[@CR12], [@CR13]\]. In any case, an effective intervention to modify these behaviours would need to translate to school health policy that enables practical application, with little cost to the school. Past evaluations of primary school sun protection policies have found that there is a lack of data supporting the link between sun protection policies and observations of sun-safe behaviour \[[@CR14]\]. A recent observational evaluation in primary schools has been conducted in the state of Queensland to address this \[[@CR13]\]. With approximately half of all students and adults observed wearing hats, it was identified that there is room for improvement in hat wearing and role modelling behaviour of students and adults. The SunSmart Evaluation and Policy Intervention Study aims to collect and measure objective data around student use of sun-safe hats, student sunscreen consumption, and adult role modelling of sun-safe behaviours in primary schools. This study will build on the above findings through observing students and teachers located in primary schools in the state of NSW, and identifying opportunities to develop and implement a policy-based intervention. This study reports on objective baseline data collected as part of a randomised cluster controlled intervention to assess students' use of sun-safe hats and sunscreen, and teacher role modelling of sun-safe behaviours in NSW primary schools. Methods {#Sec2} ======= Study design {#Sec3} ------------ The SunSmart Evaluation and Policy Intervention Study is an 18-month primary school-based intervention and is being evaluated using a cluster randomised controlled trial. The rationale and study protocol have been published previously \[[@CR15]\]. This paper reports on the baseline data of that study and their analyses. Ethics approval was obtained from an Australian University Human Ethics Committee (HREC: 2014/062) and the New South Wales Department of Education (SERAP: 2014148). The SunSmart Evaluation and Policy Intervention Study is registered with the Australian and New Zealand Clinical Trials Registry (ACTRN12614000926639). Following the initial recruitment processes, researchers conducted baseline assessments at participating schools. Principals provided written informed consent in order for their schools to participate in the study and for the results to be published in peer-reviewed journals. Recruitment and study participants {#Sec4} ---------------------------------- Schools recruited to the study were government primary schools and a current member of the SunSmart Program in the Greater Western Sydney Region, NSW, Australia (approx. 33.75 deg S,150.70 deg E). All eligible schools (*n* = 167) were sent an initial email with an invitation to participate in the study. CCNSW and the researchers identified a short-list of schools that may be receptive to participating in the study based on their response to the recruitment email (*n* = 40). Schools that were short-listed schools were pooled and received a follow up call from the project researchers to ascertain whether they would like to participate in the study. A power calculation was conducted to determine the sample size and number of observations required, which resulted in the first 20 schools that demonstrated interest being recruited into the study. Trained research assistants (RAs) conducted all assessments. All RAs completed training sessions prior to assessment to maintain consistency and the same RAs were used during the data collection phase. Outcomes and instruments {#Sec5} ------------------------ The primary outcome variable of this study was the wearing of sun-safe hats during break periods of the school day (i.e. recess and lunch) by students. Whilst sun-safe behaviours have been observed in a beach setting \[[@CR16], [@CR17]\] these were deemed inappropriate for school settings due to the variables being observed (i.e. type of swimwear). There have been a few studies to date that used a mix of surveys and on-site evaluations of SunSmart programs in Hawaii (Glanz et al, \[[@CR18]\]) and observations of public recreation venues (Dobbinson et al, \[[@CR19]\], Nikles & Harrision, \[[@CR20]\]). Very few however record sun-safe behaviours of children in school settings.. In this study, we adapted the System for Observing Play and Recreation in Communities (SOPARC) \[[@CR21]\] to capture sun-safe behaviour data in children. The SOPARC \[[@CR21]\] is based on momentary time sampling techniques in which systematic and periodic scans of individuals and contextual factors within predetermined target areas are made. Computer tablets (Apple Inc, USA) installed with the iSOPARC Application Version 1.75 (CIAFEL, Portugal: <https://ciafel.fade.up.pt/isoparc/>) were used to provide an objective measure of hat wearing by children during recess and lunch. In the traditional application of the iSOPARC tool, a scan of each subject is electronically coded and identified by: sex (male or female), intensity of activity (Sedentary, Walking, or Very Active), and whether they are a Child, Teen, Adult or Senior. For this study, given all the subjects were children, the third battery of coding (Child, Teen, Adult or Senior) was changed to detect whether the student was Unprotected (i.e. wore no hat), Partially-Protected (i.e. was wearing a baseball cap), Fully-Protected (i.e. was wearing a sun-safe hat: broad-brimmed, bucket hat or legionnaire). Separate scans were made for females and males, and entries are also made for time of day, temperature, UV radiation level, area accessibility, area usability and presence of supervision. Each observation was conducted twice during the recess and lunch breaks for both females and males. A single scan involved a researcher observing each student, of one sex, within a pre-defined Target Area sequentially from left to right without pausing. Direct observations were made in the designated Target Areas that had been identified by school principals (or their proxy) as areas that were likely to provide opportunities for students to have sun exposure during their recess and lunch periods (i.e one shaded and one unshaded play area). It is important to note here that shaded areas had to be man-made structures and were all metal structures with solid roofing. No areas were covered with 'shade cloth' or other permeable materials. Additional data recorded prior to the direct observation scans included; temperature and UV level at the start and end of the observation period; whether the observation was made at recess or lunch; start and finish times of recess and lunch; and whether the area was shaded or not. Researchers also recorded whether the teacher supervising the children's behaviour in the Target Area was role-modelling sun-safe behaviour. An observation note was added to the final iSOPARC data on whether the teacher was wearing a) a sun-safe hat; b) sunglasses or other appropriate eye protection (i.e. transition lensed optical glasses; and c) a sleeved shirt and collar. As there was no more than 4 teachers present during any observation period, research assistants were able to record all of the teacher's sun protective behaviours. Student clothing was not recorded as State requirements are that all students wear a school uniform that requires a collared shirt that covers the shoulders and upper arm as a minimum. Thirty two field-based inter-rater reliability checks were conducted during the 10-week observation period. During reliability checks, two observers independently coded the same students in the same lunch or recess period. A high degree of reliability was found between measurements. The average measure Intraclass Corelation Coefficient (ICC) was .912 with a 95% confidence interval from .885 to .932 (*F* = 11.324, *p* \< .001). In an effort to minimise bias, the inter-rater reliability checks on 4% of the iSOPARC observations were randomised in order to prevent possible collusion. Recess and lunch break observations were randomly selected and observers and schools were given limited notice of when a reliability check was going to occur (usually less than 24 h). Sunscreen consumption was also recorded during the observation period. All Stage 3 (i.e. Grades 5--6 or children aged 10--12 years) classes within the 20 schools were given accurately weighed and filled one-litre sunscreen containers (1 l = 1.07kg) at the start of the period. These containers were placed near the door of the students' classroom. The containers were removed at the end of the observation period and weighed again. This consumed weight of sunscreen was divided by the number of students within the class to produce a baseline consumption rate per student. Demographic data including socioeconomic status based on postcode of the school (Socio-Economic Indexes for Areas---SEIFA) \[[@CR22]\] and national school enrolment data collected by the federal government (Index of Community Socio Educational Advantage---ICSEA) \[[@CR23]\] was recorded for each school. The quantitative variables reported in this paper include: student sex, student hat wearing behaviour, teacher sun safety behaviour, environmental factors, school SEIFA and ICSEA status, and student activity levels. Student sex was reported as male or female. Student hat wearing behaviour was reported as a) no hat; b) baseball cap; c) legionnaire hat; d) 360° broad brimmed or bucket hat; or e) sun-safe hat which was the sum of (c) and (d). Teacher sun safety behaviours were reported as a) hat wearing; b) sunglass wearing; c) sun protective clothing (i.e. covered shoulders and collar); and d) shade seeking. The remaining quantitative variables were reported as correlation alphas to student wearing, or not wearing, a sun-safe hat during recess or lunch break. Environmental factors were shaded area (dichotomous), UV levels and temperature (continuous). The school factor was the school ICSEA value (continuous) and student enrolment numbers. Student activity levels were the proportion of time spent in each of the activity levels (Sedentary, Walking, or Vigorous). Percentages were for the entire sample and then stratified by sex. Pearson and bi-serial correlations were then calculated for the continuous and dichotomous variables accordingly. This was done to ascertain the relationship between the primary outcome variables and the covariates at the smallest discernible level. All data were analysed using Statistical Package for Social Science (SPSS) version 22. Results {#Sec6} ======= The first round of data collection occurred as planned in school Term 4, 2014 (September-December). The collection involved teams of two RAs randomly visiting the 20 consenting schools three times. The RAs each observed a separate predetermined covered and sun exposed playing area during recess and again during lunch time at each school. This resulted in 240 potentially observable sessions, of which 238 were completed. One session (i.e. two observations) were cancelled due to wet weather on the last observation day of the term. During each observable session, data was collected twice for male students in the target areas and twice for female students. It should be noted that when there were no males or no females in the playing area, observations were not recorded for that sex. This resulted in 839 individual observations being recorded out of a possible 960 (87%). Demographic Data {#Sec7} ---------------- An overview of the demographic characteristics of the schools involved in the study can be seen in Table [1](#Tab1){ref-type="table"}.Table 1Demographic characteristics of the 20 schools involved in the studyBaselineSchools stratified by SEIFA Index (% of schools) based on postcode of school10 (0%)21 (5%)33 (15%)45 (25%)511 (55%)Distribution of school students stratified by the ICSEABottom Quarter30.6%Middle Two Quarters49.8%Top Quarter19.6%Total student enrolment in schools*n* = 7971By SexBoys*n* = 4082Girls*n* = 3886 Table [1](#Tab1){ref-type="table"} provides an overview of the 20 schools that consented to being in the study. The tables shows that 55% of the schools participating in the study (*n* = 11) are located in the 5th decile (highest) of the Socio-Economic Indexes for Areas (SEIFA), with the remaining schools being in the lower deciles. This trend is highlighted when the students in the schools are stratified by the Index of Community Socio-Educational Advantage (ICSEA). This analysis reveals that 30.6% of the students enrolled in the participating schools are in the bottom quarter of the Australian population, with only 19.6% being in the top quarter. Main Results {#Sec8} ------------ Table [2](#Tab2){ref-type="table"} reports the unadjusted means and standard deviations of the students and teachers observed during the 240 observable sessions occurring during recess and lunch breaks in Term 4, 2014 (September---December).Table 2Unadjusted means and standard deviations for student and teacher factors of sun safe behaviourCategoryRecess & lunch behaviour (%) (*n* = 839)Recess & lunch behaviour (%) Males Obs. (*n* = 420)Recess & lunch behaviour (%) Females Obs. (*n* = 419)M (n)SDM (n)SDM (n)SDStudent Hat Wearing Behaviour No Hat19.3 (162)26.015.0 (63)21.223.8 (99)29.5 Baseball Cap (No side brim or back flap)20.5 (172)33.724.5 (103)35.816.5 (69)31.1 Legionnaire hat10.6 (89)26.212.1 (51)27.29.1 (38)25.0 360° Brimmed hat (Bucket or Broad)49.6 (416)41.848.3 (203)42.550.9 (213)41.1 Sun-safe hat\^60.1 (504)40.160.2 (253)41.259.9 (251)39.1Teacher sun-safe behaviour (%)M (n)Number of observations (*n* = 514) with: Hat wearing behaviour - Sun-safe hat^a^35.2 (181) - Baseball cap11.1 (57) - No hat53.7 (276) Sunglass wearing69.9 (359) Sun protective clothing50.8 (261) Shade seeking49.0 (252)^a^Sun-safe hat is the sum of the peak cap with back flap and 360° brimmed hat The table shows that 839 observations were conducted during these recess and lunch breaks with 60.1% (*n* = 504) of the students seen wearing a sun-safe hat (i.e. either a legionnaire hat or a 360° broad-brimmed or bucket hat). The table also shows that 19.2% (*n* = 161) of the students were wearing no hat at all during these times. Table [2](#Tab2){ref-type="table"} also highlights that there is a substantial difference between student hat wearing behaviours of male and female students, whilst 19% students did not wear any form of hat during recess and lunch, 24% (*n* = 99) of female students would not wear a hat, compared with 15% (*n* = 62) of male students. However, the differences of wearing of sun-safe hats between the sexes during recess and lunch was consistent at 60% across both groups. Further analysis of these results not revealed in this table but in the data showed that in 55% (*n* = 11) of the schools more than 80% of the students were wearing a sun safe hat, however the results also showed that in 20% (*n* = 4) of the schools less than 7% of the observed students were wearing a sun safe hat. This goes some way to our understanding of the amount a variance being reported in this table. The lower section of Table [2](#Tab2){ref-type="table"} displays the sun-safe behaviours of the teachers on playground duty during the observational periods. Only teacher behaviour was observed because in NSW, primary school students are required to wear a school uniform. As such, very little, if any, variability in their clothing is evident. Teacher on the other hand are able to exercise choice in this outcomes with 35% of the teachers observed wearing a sun-safe hat (but 54% wearing no hat at all) and 70% of the teachers were observed wearing sunglasses. Sun protective clothing was defined if the teacher wore clothing with a collar and covered their shoulders and upper arm to the mid-bicep. Only 51% of teachers wore items of clothing that met this description when observed during recess and lunch playground duties. In terms of sunscreen consumption, of the 141 one-litre containers of sunscreen issued to schools for use during this study, only 57 had their tamper seals broken. Of the 57 opened containers, only 5 containers exceeded more than 20% on the contents being consumed. The mean consumption per opened container per school ranged between 0 and 100g for 18 of the 20 schools. One school consumed an average of 130g per opened container (*n* = 2) and another school consumed 280g per opened container (*n* = 1). A series of Pearson product-moment correlation coefficients were computed to assess the relationship between the wearing of sun safe hats and a range environmental, school, teacher and student behaviours. We have described the strength of the correlation using the guide that Evans \[[@CR24]\] suggests for the absolute value of *r* being: .00-.19 "very weak"; .20-.39 "weak"; .40-.59 "moderate"; .60-.79 "strong" and; .80-1.0 "very strong". A summary of these correlations can be seen in Table [3](#Tab3){ref-type="table"} where they are also displayed by sex.Table 3Correlation of varying recess and lunch break contexts with the wearing of sun safe hats by sexMales and Females Observations (*n* = 839)Male Observations (*n* = 420)Female Observations (*n* = 419)r*p* valuer*p* valuer*p* valueEnvironmental factors Temperature0.000.910.020.710.030.58 UV Index0.040.200.090.080.000.99School factors School ICSEA0.26\<0.01^\*\*^0.28\<0.01^\*\*^0.23\<0.01^\*\*^Teacher on duty factor Teacher wearing hat0.15\<0.01^\*\*^0.14\<0.01^\*\*^0.16\<0.01^\*\*^Student activity levels Sedentary0.030.420.000.990.060.23 Walking0.12\<0.01^\*\*^0.14\<0.01^\*\*^0.100.04^\*^ Vigorous0.09\<0.01^\*\*^0.13\<0.01^\*\*^0.040.42*r* Pearson correlation, *ICSEA* Index of Community Socio-Educational Advantage^\*^ Statistically significant at \<0.05^\*\*^ Statistically significant at \<0.01 There were only "very weak" and no statistically significant correlations found between the wearing of sun-safe hats and daily temperature or UV Index. However, there was a "weak" positive correlation (*r* = 0.26) between a school's ICSEA and students wearing of sun-safe hats. Whilst this was statistically significant at *p* \< 0.01, the two variables only had a shared variance of 6.8%. The results did not significantly vary when analysed further by the sex of the students observed when ICSEA was the independent variable. The analysis found only "very weak", though statistically significant correlations, between the teacher wearing of a sun-safe hat and the students wearing of sun-safe hats. The analysis also revealed that in only 10% (*n* = 2) of the schools, teachers modelled hat-wearing behaviour more that 65% of the time. Again, only "very weak" though statistical significant correlations were observed between sun-safe hat wearing by students and when they engaged in vigorous or walking levels of physical activity. Consumption of sunscreen by students in Stage 3 was negligible. Cancer Council Australia recommends children apply 25ml per limb, on the head and neck, and on the torso (reference), which is equates of 25g of sunscreen. An analysis of students' consumption of sunscreen during the 9-week observation period found that each student applied an average of 9.9 ml (SD = 18.61) of sunscreen. No significant or meaningful correlation was found to exist between a school's ICSEA, daily temperature and UV Index or student activity levels with their students' consumption of sunscreen. Discussion {#Sec9} ========== The purpose of this study was to assess students' use of sun-safe hats and sunscreen, and teacher role modelling of sun-safe behaviours in NSW primary schools. The main findings of the study were that 60% of students wear sun-safe hats (i.e. a legionnaire hat, a broad brimmed or a bucket hat) during recess and lunch periods, while 20% wear a baseball cap and 19% of students do not wear any type of hat during these times. Current national survey data on self-reported sun-safe hat wearing by primary school children as reported by school Principals states that 80% of students wear sun-safe hats during recess and lunch periods \[[@CR11]\]. This 20% discrepancy appears indicative of the contrasts between self-report and objective guideline adherence to public health recommendations for school-aged children \[[@CR25]\]. We recognise that self-reported instruments still remain a practicality in large-scale studies, however, we hope this study provides some further evidence for the need to validate self-report data from large-scale studies with at least some additional objective measures to understand the degree of bias that is possibly being reported in such studies. When examining the data further, a weak positive relationship was found between a school's ICSEA and the sun-safe hat wearing behaviours of its students. No relationship was found between the teachers and students sun-safe behaviours. This is of interest considering it has been established that children's behaviours are influenced by what they observe and learn \[[@CR26]\]. CCNSW's SunSmart policy \[[@CR27]\] requires that schools ask their staff to role model good sun-safety behaviours, including the wearing of sun-safe hats when they are outside. Similar findings were reported in the observational research conducted by Turner and colleagues \[[@CR14]\] which noted schools may assume teachers are acting as role models without being asked, resulting in schools potentially placing less importance on this component. To date, CCNSW has no data pertaining to the amount of sunscreen students consumed during school hours. Consumption of sunscreen by Stage 3 students in this study was negligible suggesting that these recommendations were not met. External incentives have been recommended as a motivator to improve compliance with SunSmart policy recommendations and sun-safe practices in schools \[[@CR13]\]. As this study has identified opportunities to improve sun-safe us, the use of incentives will be considered moving forward for the intervention phase of this trial. Strengths and limitations {#Sec10} ------------------------- Strengths of this study are the high rate of observations recorded (87%). This was partially attributed to favourable weather conditions that meant only two observations were missed due to inclement weather. This appears to be one, of only a few objective studies of sun-safe behaviours and even fewer of those conducted within school premises and during the school day. We imagine that this should allow for a clearer understanding of routine sun protection practices that may be occurring. Earlier observation studies conducted by Turner and colleagues \[[@CR13], [@CR14]\] occurred from outside the school premises. The main limitation of this study was that it reports findings from a number of schools within a relatively small geographic area. Future studies need to recruit more schools and use diverse geographical settings that are more indicative of Australian population demographics. Other limitations are that it reports on cross-sectional data. Further longitudinal research is needed to ascertain whether trends in these reported behaviours are consistent. Another limitation is that data were only collected over a 3-month period of the academic year. As such, we were unable to determine with these data if annual sessional issues, such as weather, influence these findings to a greater extent. Furthermore, children were not directly observed using the sunscreen dispensers. So who, when or how the sunscreen dispensers were utilised remains a possible limitation. Conclusions {#Sec11} =========== Although it appears that a large percentage of NSW primary school students wear sun-safe hats during recess and lunch breaks, it is apparent that this varies considerably across schools and it is well-below the percentages being reported in national surveys as a whole \[[@CR10], [@CR11]\]. It is also apparent that sunscreen use among certain students may be well below current health recommendations \[[@CR27]\]. This study identifies specific areas within school sun-protection practices that could benefit from an intervention that promoted both the wearing of sun-safe hats and sunscreen. Given the finite resources of schools and the correlation, though small, with ICSEA for these behaviours, it behoves researchers to investigate low-cost solutions to these problems. Further qualitative data will also be needed to inform the enablers and barriers for sun-safe behaviour interventions to be adopted in NSW primary schools. ACTRN : Australian and New Zealand Clinical Trials Registry Number CCNSW : Cancer Council NSW CONSORT : Consolidation Standards of Reporting Trials HPS : Health Promoting Schools NSW : New South Wales NSWDEC : New South Wales Department of Education and Communities SCT : Social Cognitive Theory SERAP : State Education Research Approvals Process SOPARC : System for Observing Play and Recreation in Communities SPSS : Statistical Package for the Social Sciences UV : Ultraviolet WHO : World Health Organization The authors are grateful for the support and cooperation of the NSW Department of Education and the participating schools, students, and teachers. Funding {#FPar1} ======= The funds for this project were made available through the NSW Cancer Institute's Evidence to Practice Grant and CCNSW (MQU IRIS No: 9201401102). Availability of data and materials {#FPar2} ================================== As per ethics committee instructions, all data and materials for this study must remain in the secure possession of the lead researcher for a period of 7 years. For this reason, the data cannot be shared. Authors' contributions {#FPar3} ====================== DD, MW, WC and BW obtained funding for the research. All authors contributed to reviewing, editing and approving the final version of the paper. DD is the guarantor and accepts full responsibility for the conduct of the study. All authors have read and approved the final manuscript. Competing interest {#FPar4} ================== KJ, AB and VR are employees of CCNSW. The remaining authors have no other conflicts of interest to declare. Consent for publication {#FPar5} ======================= Not applicable. Ethics approval and consent to participate {#FPar6} ========================================== Ethics approval was obtained from the Charles Sturt University Human Ethics Committee (HREC 2014/062) and the New South Wales Department of Education and Communities (SERAP: 2014148). Written consent was also obtained from all school principals for the observations to occur in their respective schools and parents/guardians of students at schools were informed of the school's consent and the nature of the study. The SunSmart Evaluation and Policy Intervention study is registered with the Australian and New Zealand Clinical Trials Registry (ACTRN12614000926639). The study protocol was also reviewed internally by the research committees of the funding agencies.

Background ========== Hypertension is a highly prevalent risk factor for coronary heart disease (CHD), stroke, heart failure (HF), renal disease, and recurrent cardiovascular events. It has been shown to reduce the number of life-years lost and the number of years lived with disability by 64.3 million globally.[@b1-vhrm-5-1043] The European Society of Hypertension (ESH) classifies optimal blood pressure at \<120 mmHg systolic blood pressure (SBP), and \<80 mmHg diastolic blood pressure (DBP).[@b2-vhrm-5-1043] A therapeutic reduction of elevated blood pressure (BP) levels has been shown to decrease cardiovascular morbidity and mortality.[@b3-vhrm-5-1043] For the pharmacological management of hypertension, lowering BP below 140/90 mmHg in all patients is requested. Specific patients, those with diabetes mellitus or chronic kidney disease, need further BP reduction (\<130/80 mmHg or \<125/75 mmHg, respectively). However these BP targets are difficult to meet despite the availability of a number of effective antihypertensive drugs. Consequently hypertension control is largely ineffective as the achievement of guideline defined treatment targets with about 20% of patients in Europe and up to 50% of patients in the US being finally controlled when treated.[@b4-vhrm-5-1043],[@b5-vhrm-5-1043] BP target achievement is even worse in patients with comorbid disease like diabetes mellitus. Patient perspective ------------------- Hypertension is a rather unspectacular disease with unspecific symptoms that are, from a patient perspective, in many cases not perceived to occur in relation to hypertension. Patients therefore are reluctant to accept physicians' recommendations to adopt life-style changes (weight reduction, reduction of sodium intake, and increased physical activity), which are the recommended first steps in the treatment of hypertension. It becomes even more difficult to convince patients for the need of action when antihypertensive pharmacotherapy is introduced. A number of patient-related factors leading to nonadherence like frequent dosing,[@b6-vhrm-5-1043],[@b7-vhrm-5-1043] drug-related adverse events (AEs),[@b8-vhrm-5-1043] health beliefs,[@b9-vhrm-5-1043] drug--drug interactions and associated medical conditions interfere with the patients' willingness to take drugs as prescribed. For an overview of terms used to describe adherence see [Figure 1](#f1-vhrm-5-1043){ref-type="fig"}. Actually approximately half of the patients on antihypertensive drug therapy discontinue therapy by the end of the first year. Hence, from a patient perspective, there is a need for effective, highly tolerable and convenient medication that does not interfere with daily life while controlling hypertension-associated risk. Physician perspective --------------------- Recent data suggest that also physicians' attitudes and treatment strategies hamper the effectiveness of current therapy.[@b10-vhrm-5-1043]--[@b12-vhrm-5-1043] In a recent global survey in a random sample of primary care physicians, 41% of physicians aimed to reduce BP to "acceptable levels" only, although generally agreeing with guideline recommended treatment goals. Physicians further believed that 62% of their patients had their BP controlled. However, in fact only 6% of patients with hypertension in the UK had their BP lowered to the recommended levels.[@b13-vhrm-5-1043] In France, Germany, Italy, and Spain only 13% of hypertensive patients have their BP controlled.[@b14-vhrm-5-1043] The physicians' needs in the treatment of hypertension are partially overlapping with patients' needs. However, in the face of low patient compliance, the chronic nature of the disease, and increasing budget constraints, a possible solution seems to be difficult to determine. Treatment patterns ------------------ Globally about one-third of patients receive monotherapy, one-third dual combination therapy, and one-third 3 or more antihypertensive drugs. The NICE for example calculated that 36% of patients in the UK receive monotherapy, 38% dual combination therapy, and 26% 3 or more drug--drug combinations.[@b15-vhrm-5-1043] In a recent drug utilization analysis in primary care in Germany, 29.2% of treated patients received one, 43.7% received 2, and 27.2% received 3 or more antihypertensive drugs.[@b16-vhrm-5-1043] According to the aforementioned study[@b16-vhrm-5-1043] 40.8% of patients in Germany received ACE inhibitors (ACEi), 36.1% beta blockers, 31.7% diuretics, 22.3% calcium channel blockers (CCBs) and 14.1% angiotensin receptor blockers (ARBs).[@b16-vhrm-5-1043] Frequent drug--drug combinations were ACEi or ARBs in combination with diuretics, CCBs, and beta blockers which is mostly in accordance with recent guidelines in which 4 out of 6 recommended combinations are ACEi/ARB based.[@b17-vhrm-5-1043] Guidelines ---------- The 2007 ESH/ESC guideline recognize 5 major antihypertensive drug classes -- thiazide diuretics, CCBs, ACEi, ARBs, and betablockers -- to be suitable for the initiation and maintenance of antihypertensive treatment.[@b17-vhrm-5-1043] The JNC VII guidelines[@b18-vhrm-5-1043] on the other hand recommend using thiazide diuretics first. Both guidelines agree however in recommending the use of particular drug classes based on the presence of compelling indications. ACEi and ARBs are recommended for the largest variety of compelling indications (for a detailed overview see [Table 1](#t1-vhrm-5-1043){ref-type="table"}), with only minor compelling contraindications in patients with pregnancy, hyperkalemia, bilateral renal artery stenosis, and angioneurotic edema (ACEi only). Therefore both drug classes are used, as illustrated by data from different drug utilization studies,[@b16-vhrm-5-1043],[@b19-vhrm-5-1043] in 50% to 60% of patients. Drug--drug combination therapy ------------------------------ According to the ESH/ESC guidelines,[@b17-vhrm-5-1043] the following drug--drug combinations have been found to be effective and well tolerated in randomized efficacy trials: Thiazide diuretic plus ACEi, thiazide diuretic and ARB, CCB and ACEi, CCB and ARB, CCB and thiazide diuretic, and beta blocker and dihydropiridine CCBs. Thus, 4 out of 6 recommended dual antihypertensive combinations are ACEi/ARB based ([Figure 2](#f2-vhrm-5-1043){ref-type="fig"}). The ACCOMPLISH study triggered a lively discussion about the relative importance of drug--drug combinations.[@b20-vhrm-5-1043],[@b21-vhrm-5-1043] It was designed to test whether benazepril 40 mg combined with amlodipine 10 mg would result in stronger cardiovascular event reduction than benazepril 40 mg/HCTZ 25 mg. Inclusion and exclusion criteria favored the selection of patients with compelling indications for the use of CCBs. The composite primary endpoint of cardiovascular morbidity and mortality was reduced by 19.6% in patients receiving benazepril/amlodipine versus benazepril/HCTZ (9.6 versus 11.8%; hazard ratio \[HR\] 0.80, 95% CI 0.72--0.90). The secondary endpoint of death from cardiovascular causes, non-fatal myocardial infarction, and non-fatal stroke was reduced by 21% with a HR of 0.79 (95% CI 0.67--0.92). Side effects were generally however more frequent with amlodipine than with the thiazide diuretic. Unfortunately no similar comparative study of the ACCOMPLISH type exists for patients with compelling indications for thiazide diuretic use.[@b22-vhrm-5-1043] At present the evidence base is weak for deciding which patient would benefit the most from either combination. Law and Wald have suggested combining ACEi/ARBs with a low dose of any of the other drug classes to maximize BP lowering efficacy while maintaining a placebo-like tolerability.[@b23-vhrm-5-1043] This recommendation was based on the metaanalytic observation that both ACEi and ARBs maintain a particular low AE profile in doses up to 4 times standard dose. On the contrary all other drug classes (CCBs, betablockers, thiazides) showed a steep incline of side effects at higher doses, while the tolerability was good at half-standard or even at standard dose. The common approach to control BP in hypertensive patients is to titrate monotherapy to full dose and to add another agent if BP is still high ([Figure 3](#f3-vhrm-5-1043){ref-type="fig"}, green). Drugs might be exchanged if there is indication of non-response to a particular agent. More recently the ESH/ESC guidelines[@b17-vhrm-5-1043] introduced the concept of first-line combination therapy at low dose in patients with marked BP elevation, low target BPs, and high or very high cardiovascular risk ([Figure 3](#f3-vhrm-5-1043){ref-type="fig"}, red). This has been shown to be effective and safe and tolerability of first-line combination therapy is excellent.[@b24-vhrm-5-1043],[@b25-vhrm-5-1043] Requirements for antihypertensive drugs --------------------------------------- In summary, from a patient, physician, and societal perspective there is a clear need for drug--drug combinations which provide effective BP lowering, and display a low side effect profile and a high adherence of both physicians and patients with treatment. This would enable BP control to be increased considerably and would in turn not only save on hypertension-related morbidity (stroke, ischemic heart disease) and mortality but also on costs. It has been calculated for the UK that achieving a systolic BP of 140 mmHg on a large scale would decrease stroke incidence by up to 44% depending on age group considered and ischemic heart disease incidence up to 35%.[@b15-vhrm-5-1043] Assuming that a reduction of stroke incidence of 9% and ischemic heart disease of 4% across age groups might actually be achievable, the annual saving to the NHS would be £255 million for stroke and £25 million for ischemic heart disease while investing £58 million into drugs (net benefit to the NHS £222 million) ([Figure 4](#f4-vhrm-5-1043){ref-type="fig"}). Candesartan/HCTZ ================ The fixed drug--drug combination of candesartan cilexetil in a dose up to 32 mg and HCTZ in a dose up to 25 mg fulfils most of the above-mentioned requirements for improving BP control and related morbidity on a larger scale. Candesartan ----------- Candesartan is an ARB that is administered orally as candesartan cilexetil; it is rapidly and completely converted to candesartan, the active compound, during absorption from the upper gastrointestinal tract.[@b26-vhrm-5-1043] It is characterized by a strong binding affinity to the angiotensin II type 1 receptor and its slow dissociation. Its binding to the AT~1~ receptor is insurmountable, meaning that it cannot be overcome by high concentrations of angiotensin II and, under physiological conditions, may even not dissociate until the AT~1~ receptor is recycled.[@b26-vhrm-5-1043] Twenty-four hours after administration to healthy volunteers, the angiotensin II inhibiting activity per milligram of candesartan was stronger than that shown by other ARBs.[@b27-vhrm-5-1043] The trough-to-peak ratio is almost 90% (mean of all doses available). After a missed dose of candesartan, losartan, or placebo, 48-hour post-dose significant reductions of BP have been observed with candesartan 16 mg daily but not with losartan or placebo.[@b28-vhrm-5-1043] Candesartan, applied as oral monotherapy, results in a strong dose-dependent reduction of both SBP and DBP between 4 and 16 mg, levelling off at 32 mg, and reaching its maximum at 8 weeks after treatment initiation.[@b29-vhrm-5-1043],[@b30-vhrm-5-1043] In a direct comparison of 16 mg candesartan and 20 mg enalapril candesartan was significantly more effective in reducing SBP and DBP (−13.5/−8.7 versus −9.9/−5.8 mmHg; *P* = 0.008).[@b31-vhrm-5-1043] It was also shown that BP returned to baseline after a missed dose of enalapril (−7.2/−4.5 mmHg) earlier than after a missed dose of candesartan (−11.4/−8.0 mmHg; *P* = 0.0002). Candesartan (up to 16 mg) and losartan (up to 50 mg) were compared in an 8-week study.[@b32-vhrm-5-1043] Candesartan reduced diastolic BP by 8.9 and 10.3 mmHg with the 8 and 16 mg doses, respectively, while the BP reduction with losartan 50 mg was 3.7 mmHg, the latter comparison reaching statistical significance (*P =* 0.013). Twenty-four hours after the ingestion of candesartan 100% of the peak SBP/DBP lowering effect was preserved (trough/peak ratio about 100% both systolic and diastolic) while only 70% of the losartan effect was preserved (trough/peak ratio 70% both systolic and diastolic).[@b32-vhrm-5-1043] These data were essentially confirmed by Lacourciere et al which also demonstrated, that after a missed dose of 16 mg candesartan, the effect was well preserved after 36 hours, while the effect of 100 mg losartan was significantly reduced.[@b28-vhrm-5-1043] It is tempting to speculate that differences in effectiveness of these ARBs may reflect pharmacologic and pharmacokinetic differences. The elimination half-life of candesartan is longer than that of losartan and its active metabolite. Candesartan cilexetil produces clear dose-dependent antihypertensive effects, whereas it has been difficult to demonstrate this property for losartan.[@b26-vhrm-5-1043],[@b32-vhrm-5-1043] Hydrochlorothiazide ------------------- HCTZ mainly acts within the lumen of the distal nephron, blocking the luminal transmembrane-coupled sodium chloride transport system. The mechanism by which thiazide diuretics reduce BP is however not completely understood. It has been proposed that during long-term therapy, thiazides act by reducing total peripheral resistance probably through a direct vascular effect.[@b33-vhrm-5-1043] It is important to note however that in vivo vasodilation was achieved at higher doses than those reached during long-term oral treatment.[@b34-vhrm-5-1043] HCTZ treatment in patients with hypertension induced changes in plasma volume, cardiac output, mean arterial pressure, stroke volume, heart rate, and total peripheral resistance.[@b33-vhrm-5-1043],[@b35-vhrm-5-1043] HCTZ has a half-life of 8 to 15 hours on chronic use and a duration of action that is slightly longer.[@b36-vhrm-5-1043] HCTZ 50 mg twice daily for 12 or 36 weeks, after a 4-week placebo run-in period, lowered mean arterial pressure in 13 patients with untreated essential hypertension and DBP \> 100 mmHg.[@b35-vhrm-5-1043] Compared with the mean baseline value (177.2 mmHg), these reductions were significant throughout the study duration. BP reduction with candesartan/HCTZ versus placebo ------------------------------------------------- Fixed dose combinations of candesartan and HCTZ are available in various doses. Candesartan 32 mg once daily has to be regarded as a high dose (4 times standard dose).[@b23-vhrm-5-1043] HCTZ 12.5 and 25 mg have to be regarded as a half-standard and standard dose, respectively. Therefore the available combinations fulfil the requirements suggested by Law and Wald[@b23-vhrm-5-1043] for maximizing efficacy while maintaining a high tolerability. The extent of BP reduction with candesartan/HCTZ depends on baseline BP and the dose used. A variety of combinations including 2, 4, 6, 8, 16, or 32 mg candesartan and 6.25, 12.5, or 25 mg HCTZ, respectively, has been tested in clinical trials versus respective monotherapies or placebo.[@b25-vhrm-5-1043],[@b37-vhrm-5-1043]--[@b41-vhrm-5-1043] Uen et al for example demonstrated that replacing previously ineffective antihypertensive drugs by candesartan/HCTZ in patients with uncontrolled arterial hypertension significantly reduced both BP and ST-segment depression during daily life.[@b41-vhrm-5-1043] Taken together these studies have consistently shown that combinations of candesartan with HCTZ, administered orally once a day for 4 to 52 weeks, induced significant reductions in SBP and DBP from baseline in patients with mild, moderate, or severe hypertension. In a recent study by Edes et al (baseline DBP 90--114 mmHg), mean reductions in SBP and DBP were significantly greater with candesartan 32/HCTZ 25 mg (21/14 mmHg) than with candesartan 32 mg alone (13/9 mmHg), HCTZ 25 mg alone (12/8 mmHg), or placebo (4/3 mmHg) (*P* \< 0.001 for all comparisons).[@b39-vhrm-5-1043] The proportion of patients with controlled BP (SBP \< 140 mmHg and DBP \< 90 mmHg) at the end of this study was also significantly greater in the candesartan 32/HCTZ 25 mg group (63%) than in the other treatment groups (*P* \< 0.001 for all comparisons). Bönner investigated the efficacy of candesartan 32 mg in combination with HCTZ 12.5 or 25 mg in patients not optimally controlled using candesartan monotherapy.[@b42-vhrm-5-1043] A total of 3521 patients with treated or untreated hypertension and sitting DBP of 90 to 114 mmHg were included. After a single blind run-in phase (2 weeks candesartan 16 mg followed by a 6-week treatment with candesartan 32 mg) 1975 patients who still had DBP readings of 90 to 114 mmHg were randomized to an 8-week double-blind treatment with either candesartan 32 mg alone or in combination with HCTZ 12.5 mg or 25 mg respectively. Mean BP (153/97 mmHg at baseline) was further reduced during the double-blind treatment phase by 6.1/5.6 mmHg in the candesartan monotherapy group, by 13.0/8.8 mmHg in the fixed combination with HCTZ 12.5 mg group, and by 15.5/10.0 mmHg in the fixed combination with HCTZ 25 mg group (*P* \< 0.01 for all between treatment comparisons) ([Figure 5](#f5-vhrm-5-1043){ref-type="fig"}). Bönner et al tested the first-line use of candesartan 16 mg/HCTZ 12.5 mg in 166 patients with no prior pharmacotherapy for a treatment duration of 6 weeks.[@b25-vhrm-5-1043] Blood pressure was reduced by 38.1/29.4 mmHg with 40% of patients achieving a normalization of BP. Tolerability was good showing that first line combination therapy is feasible and safe. BP reduction with candesartan/HCTZ versus losartan/HCTZ ------------------------------------------------------- Ohma et al compared fixed dose combinations of candesartan 16/HCTZ 12.5 mg and losartan 50/HCTZ 12.5 mg in patients insufficiently controlled on previous monotherapy.[@b43-vhrm-5-1043] BP at randomization was 159.5/98.4 mmHg and 160.5/98.5 mmHg, respectively. After 12 weeks there was a greater reduction of BP with candesartan/HCTZ (−19.4/−10.4) than with losartan/HCTZ (−13.7/−7.8 mmHg), the differences being statistically significant. Twelve patients withdrew in the candesartan/HCTZ group (8 due to AEs), and 17 in the losartan/HCTZ group (12 AEs). König compared candesartan/HCTZ and losartan/HCTZ in a 6-week study.[@b44-vhrm-5-1043] Twenty-four-hour postdose mean seated BP was reduced by 32.2/21.1 mmHg (systolic/diastolic) in the candesartan/HCTZ group and 23.8/14.9 mmHg in the losartan/HCTZ group (*P* \< 0.001). Blood pressure reductions 48 hours postdose were 25.6/16.4 mmHg for candesartan/HCTZ and 9.2/4.2 mmHg for losartan/HCTZ, with differences between treatments being highly significant in favor of candesartan/HCTZ (16.5/12.2 mmHg; *P* \< 0.001). Both treatments were well tolerated. Tolerability/compliance ----------------------- ARBs are generally regarded to be a drug class with high compliance/persistence.[@b45-vhrm-5-1043] Persistence with antihypertensive medication (including candesartan) was compared between different drug classes and between substances within one drug class in an Australian analysis covering the years 2004 to 2006.[@b46-vhrm-5-1043] The database yielded information relating to 48,690 patients prescribed antihypertensive medication. The median persistence time was 20 months, which was also the median persistence with ARBs or ACEi. The median persistence with CCBs was considerably lower (median persistence time 7 months; −57%, *P*\< 0.001). There were further differences in persistence between individual drugs in the respective classes, the best outcomes being with candesartan and telmisartan (10%--20% better than the other ARBs considered), perindopril (ACEi; 25% better other ACEi) and lercanidipine (CCB; 25% better than other CCBs). This high persistence was reflected in the recent DIRECT trial in that about 80% of patients were compliant with 32 mg candesartan even when being nominally normotensive.[@b47-vhrm-5-1043],[@b48-vhrm-5-1043] Candesartan/HCTZ is generally well tolerated in patients with mild to moderate hypertension. Combined data from 5 randomised, double-blind, placebo-controlled clinical trials indicated that AEs during candesartan/HCTZ therapy (up to 16 mg/25 mg once daily) are uncommon and only few were serious.[@b49-vhrm-5-1043] Among patients receiving candesartan/HCTZ or placebo the incidence of serious AEs was 1.6 and 2.1%, respectively, while 3.3 and 2.7% of patients discontinued treatment because of AEs. The most common AEs were headache, back pain, dizziness, and respiratory infections. Recent trials indicated that the AE profile of candesartan 32 mg in combination with 12.5 or 25 mg HCTZ is comparable to the aforementioned observations.[@b39-vhrm-5-1043],[@b42-vhrm-5-1043] Bönner et al reported about 1% serious AEs that were independent of whether monotherapy with candesartan or combination therapy including HCTZ was considered. For metabolic parameters, a slight increase of serum ureate and serum creatinine was observed with the fixed combinations while other parameters were essentially unchanged ([Table 2](#t2-vhrm-5-1043){ref-type="table"}).[@b42-vhrm-5-1043] Edes et al reported a rate of serious AE for the fixed dose combination that was even lower compared to placebo (0.2% versus 3.1%), with overall AE rate ranging between 23% and 25% for placebo, HCTZ, candesartan, and their combination.[@b39-vhrm-5-1043] Mengden et al compared drug regimen compliance (DRC) with antihypertensive combination therapy in patients whose BP was controlled versus uncontrolled after 4 weeks of self-monitored BP measurement.[@b50-vhrm-5-1043] Whether switching one drug of the combination therapy to candesartan/HCTZ (16 mg/12.5 mg) in uncontrolled patients with and without compliance intervention program would improve BP normalization was also evaluated. It was found that normalization of BP was associated with superior drug regimen compliance in previously uncontrolled patients treated with a combination drug regimen. Switching still-uncontrolled patients to candesartan/HCTZ significantly improved BP control and stabilized a declining DRC. Patient types ============= Patients with heart failure --------------------------- Heart failure is a frequent comorbidity in patients with hypertension. It is characterized by a decline in systolic or diastolic function, the latter being a typical complication of long-term uncontrolled hypertension. The fixed dose combination of candesartan/HCTZ has never been formally tested in this patient population, but the benefits of blocking the RAS and enhancing diuresis are basic concepts in the treatment of HF.[@b51-vhrm-5-1043] HCTZ is recommended for the treatment of patients with HF in doses of 25 mg to initiate treatment and maintenance doses of between 2.5 and 100 mg daily.[@b51-vhrm-5-1043] Candesartan is recommended to be started at a dose of 4 or 8 mg daily and uptitrated to a target dose of 32 mg. The basis for the recommendation of candesartan was the results of the CHARM trial program.[@b52-vhrm-5-1043] In CHARM candesartan was tested in patients with systolic HF given either on top of an ACEi[@b53-vhrm-5-1043] or in cases of ACEi intolerance.[@b54-vhrm-5-1043] A third trial investigated the effect of candesartan in diastolic HF (HF with preserved systolic function).[@b55-vhrm-5-1043] Overall, 7601 patients were randomly assigned candesartan (titrated to 32 mg once daily) or matching placebo, and followed up for at least 2 years. In the overall CHARM trial program 82.8% of patients in the candesartan arm and 82.6% of patients in the placebo arm received diuretics and a further 16.9 and 16.6% respectively spironolactone. In CHARM Alternative (candesartan given instead of an ACEi) treatment resulted in a relative risk reduction (RRR) of death from cardiovascular cause or hospital admission for worsening HF of 23% (ARR 7%, NNT 14, over 34 months of follow-up, adjusted *P* \< 0.0001 ([Figure 6](#f6-vhrm-5-1043){ref-type="fig"}).[@b54-vhrm-5-1043] In the CHARM Added trial (candesartan on top of existing ACEi therapy) candesartan cotreatment resulted in a 15% RRR of cardiovascular death or hospital admission for CHF (ARR 4%, NNT 25, over 41 months of follow-up, adjusted *P* = 0.010).[@b54-vhrm-5-1043] The CHARM Preserved trial (candesartan in patients with HF but preserved systolic function) candesartan did not show a significant reduction in the risk of the primary composite endpoint (adjudicated death from cardiovascular causes or admission with HF) but did show a significant reduction in the number of patients admitted to hospital with CHF (ARR 3.3%, NNT 30, over 37 months follow up, *P* = 0.017).[@b55-vhrm-5-1043] In summary it appears possible to initiate drug treatment of HF with both combination agents at low dose, uptitrate candesartan and HCTZ as warranted and, should the doses fit, switch to a fixed dose combination of candesartan 32/HCTZ 25 mg for maintenance treatment. Patients with stroke -------------------- Stroke is a frequent, serious, and finally costly complication of hypertension. Candesartan was tested in 2 trials with respect to this indication, one testing the capability of preventing stroke or related disabilities (SCOPE,[@b56-vhrm-5-1043] see section Elderly patients), the other testing cerebro- and cardiovascular endpoints in patients with a history of stroke (ACCESS).[@b57-vhrm-5-1043] ACCESS was designed to evaluate the safety of early antihypertensive treatment in patients with acute cerebral ischemia).[@b57-vhrm-5-1043] Patients with motor paresis and initial SBP \> 200 mmHg and/or DBP \> 110 mmHg or mean BP of 2 measurements \> 180 mmHg and/or 105 mmHg, respectively, were included. The trial was stopped prematurely after the recruitment of 349 patients due to an imbalance in endpoints. Cumulative 12-month mortality and the number of vascular events differed significantly in favor of the candesartan group (OR 0.475; 95% CI 0.252--0.895). There were no cardiovascular or cerebrovascular events as a result of hypotension. Treatment was started with 4 mg candesartan daily on day 1. On day 2, dosage was increased to 8 or 16 mg candesartan if BP exceeded 160 mmHg systolic or 100 mmHg diastolic. In patients in the candesartan group who were still hypertensive on day 7 (mean daytime BP \> 135/85 mmHg), candesartan was increased or an additional antihypertensive drug (HCTZ, felodipine, metoprolol) was added. The control group received placebo for the first 7 days and 8 to 16 mg candesartan throughout the rest of the study. In the ongoing SCAST trial (NCT00120003) candesartan is tested in patients with stroke (ischemic or hemorrhagic) and SBP ≥ 140 mmHg. Patients receive 4 mg candesartan on day 1; 8 mg on day 2; 16 mg on days 3 to 7. Dose adjustment in cases of SBP \< 120 mmHg, or symptomatic fall in BP are mandated. From day 8 therapies can be supplemented with any antihypertensive agent including diuretics. Taken together there is good evidence that early candesartan treatment after acute stroke might be able to prevent vascular events and mortality. The fixed dose combination might be useful after several days of candesartan mono-therapy uptitration after which HCTZ in low dose is added to maintain or achieve BP control. Patients with diabetes mellitus ------------------------------- Antihypertensive treatment in diabetic patients is complicated by the fact that baseline BP readings are usually high, while having to meet lower BP goals (\<130/80 mmHg in most cases).[@b17-vhrm-5-1043] ARBs are beneficial within the context of diabetes because they have been shown to delay the development of diabetes more than any other drug class ([Figure 7](#f7-vhrm-5-1043){ref-type="fig"})[@b58-vhrm-5-1043],[@b59-vhrm-5-1043] and to be at least neutral or even beneficial with respect to metabolic parameters. As has been shown in a number of clinical trials, the reduction of cardiovascular morbidity following antihypertensive treatment is usually, but not always, pronounced in patients with diabetes.[@b60-vhrm-5-1043],[@b61-vhrm-5-1043] Candesartan reduced the number of patients developing diabetes in the CHARM,[@b62-vhrm-5-1043] SCOPE,[@b63-vhrm-5-1043] and ALPINE trials.[@b64-vhrm-5-1043] When administered to a group of hypertensive subjects it reduced C-reactive protein and increased adiponectin and markers of insulin sensitivity, as measured by QUICKI (Quantitative Insulin-Sensitivity Check Index).[@b65-vhrm-5-1043] It also reduced BP effectively in diabetic patients.[@b66-vhrm-5-1043] Bramlage et al demonstrated in an observational study in primary care that candesartan 16/HCTZ 12.5 mg lowered BP effectively in patients with and without diabetes.[@b45-vhrm-5-1043] The absolute amount of BP lowering (−27.2/−1--3.4 mmHg) appeared to be dependent on baseline BP but did not differ among patient types (diabetes, metabolic syndrome, or neither condition). Microalbuminuria in diabetes is strongly predictive of nephropathy, end-stage renal disease, and premature cardiovascular morbidity and mortality. Effective preventive therapies are therefore a clinical priority. The effect of candesartan in the prevention of microalbuminuria was tested in a pooled analysis of the DIRECT trial program in which normotensive patients with type 1 (n = 3326) and 2 (n = 1905) diabetes were included.[@b67-vhrm-5-1043] Due to the study design the incidence of microalbuminuria was low in this analysis and no differences in the risk for albuminuria were noted (HR 0.95; 95% CI 0.78--1.16). Pooled results showed that the annual rate of change in albuminuria was 5.53% lower (CI, 0.73%--10.14%; *P* = 0.024) with candesartan than with placebo. Studies conducted by Trenkwalder[@b68-vhrm-5-1043] and Mogensen[@b69-vhrm-5-1043] have however shown that candesartan is effective in lowering the level of albumin excretion in patients with hypertension, diabetes, and already existing microalbuminuria. Taking a much higher than recommended dose of the hypertension drug candesartan was shown to effectively lower the amount of protein excreted in the urine of patients with kidney disease in a study by Burgess et al.[@b70-vhrm-5-1043] 269 patients with persistent proteinuria despite treatment with 16 mg candesartan were randomized to receive 16, 64, or 128 mg daily of candesartan for 30 weeks. It was found that patients taking 128 mg of candesartan experienced a 33% reduction in proteinuria compared with those receiving 16 mg candesartan by the end of the study. There is however a missing link between microalbuminuria reduction and morbidity and mortality endpoints which have so far been reported only from post-hoc analyses. A respective study is however already underway to provide this link.[@b71-vhrm-5-1043] Important in this respect are the results of the GUARD study that combined an ACEi with either amlodipine or HCTZ and demonstrated that with HCTZ the nephroprotective effect of benazepril was preserved while it was reduced when amlodipine was chosen as the combination partner.[@b72-vhrm-5-1043] The DIRECT trial program was a series of clinical trials investigating the effect of candesartan on the development and progression diabetic retinopathy who were either normotensive or had treated hypertension.[@b47-vhrm-5-1043],[@b73-vhrm-5-1043] DIRECT consisted of three randomized, double-blind, placebocontrolled multicenter studies designed to investigate the potential for candesartan in halting the progression of, and possibly prevent, diabetic retinopathy. Results showed that candesartan was beneficial for patients with type 2 diabetes who had established mild to moderate retinopathy, because candesartan had an additional, BP-independent effect on improvement of retinopathy ([Figure 8](#f8-vhrm-5-1043){ref-type="fig"}). Candesartan was also shown to be indicated for patients with type 1 diabetes without retinopathy, in order to reduce their risk of developing retinopathy. In summary there is an abundance of evidence for the use of candesartan in patients with diabetes or a high risk for developing such. Data on the use of a fixed dose combination of candesartan/HCTZ are scarce, leaving it unproven that a combination treatment is likewise beneficial. Data addressing the dysmetabolic potential have however shown that the metabolic profile of HCTZ is neutralized when adding candesartan. Because of the need for multiple drug--drug combinations there is a clear need for combination therapy including diuretics, which are favored in common co-morbidities of diabetes, eg, CHF or diabetic nephropathy. Elderly patients ---------------- The pharmacokinetics of candesartan were investigated after single and repeated once-daily doses in a trial by Hübner et al[@b74-vhrm-5-1043] of candesartan in the dose range 2 to 16 mg in both younger (19--40 years) and elderly (65--78 years) healthy volunteers. The area under the curve (AUC) and maximal concentration (C~max~) of candesartan showed dose-proportional increases in the dose range of 2 to 16 mg candesartan after both single and repeated once-daily tablet intake, indicating linear pharmacokinetics in both younger and elderly healthy subjects. The time to peak candesartan concentrations after tablet intake was consistently approximately 4 hour at all dose levels. Only mild AEs were recorded, with 'headache' the most commonly reported event, and no increase in the number of reported AEs was observed with higher doses of candesartan cilexetil.[@b74-vhrm-5-1043] Results of the SCOPE study implied that candesartan treatment reduces cardiovascular morbidity and mortality in old and very old patients with mild to moderate hypertension.[@b75-vhrm-5-1043],[@b76-vhrm-5-1043] Candesartan-based antihypertensive treatment may also have positive effects on cognitive function and quality of life. SCOPE was a multi-center, prospective, randomized, double-blind, parallel-group study. The primary objective was to assess the effect of candesartan 8--16 mg once daily, on major cardiovascular events in elderly patients (70--89 years of age) with mild hypertension (DBP 90--99 and/or SBP 160--179 mmHg). The main analysis showed that non-fatal stroke was reduced by 28% (*P* = 0.04) in the candesartan group compared with the control group, and there was a non-significant 11% reduction in the primary endpoint, major cardiovascular events (*P* = 0.19). Significant risk reductions with candesartan in major cardiovascular events (32%, *P* = 0.013), cardiovascular mortality (29%, *P* = 0.049) and total mortality (27%, *P* = 0.018) were observed in patients who did not receive add-on therapy after randomization, and in whom the difference in BP was 4.7/2.6 mmHg. Other analyzes suggest positive effects of candesartan-based treatment on cognitive function, quality of life and new-onset diabetes. Results of SCOPE strongly suggested that candesartan treatment reduces cardiovascular morbidity and mortality in old and very old patients with mild to moderate hypertension. Candesartan-based antihypertensive treatment may also have positive effects on cognitive function and quality of life.[@b76-vhrm-5-1043] Subgroup analyses from the CHARM study in patients with HF showed that older patients were at a greater absolute risk of adverse CV mortality and morbidity outcomes, but derived a similar RRR and, therefore, a greater absolute benefit from treatment with candesartan, despite receiving a somewhat lower mean daily dose of candesartan.[@b77-vhrm-5-1043] Adverse effects were more common with candesartan than with placebo, although the relative risk of adverse effects was similar across age groups. The benefit to risk ratio for candesartan was thus favorable across all age groups. In summary, given that diuretics are frequently indicated in elderly patients there appears to be a role for fixed dose combinations of candesartan/HCTZ. But again evidence has been acquired with free combinations of candesartan with other antihypertensive drugs including thiazide diuretics. Economic evaluation =================== The addition of candesartan to standard therapy for CHF provided important clinical benefits at little or no additional cost in France, Germany, and the UK, according to a detailed economic analysis focusing on major cardiovascular events and prospectively collected resource-use data from the CHARMAdded and CHARM-Alternative trials in patients with CHF and left ventricular (LV) systolic dysfunction.[@b78-vhrm-5-1043] Results of a corresponding cost-effectiveness analysis showed that candesartan was either dominant over placebo or was associated with small incremental costs per life-year gained, depending on the country and whether individual trial or pooled data were used. Preliminary data from a US cost-effectiveness analysis based on CHARM data also showed favorable results for candesartan cilexetil. Two cost-effectiveness analyses of candesartan cilexetil in hypertension have been published, both conducted in Sweden. Data from the SCOPE trial in elderly patients with hypertension, which showed a significant reduction in non-fatal stroke with candesartan-based therapy versus non-candesartan based treatment, were incorporated into a Markov model and an incremental cost-effectiveness ratio of 12,824 per QALY gained was calculated (2001 value). In conclusion, despite some inherent limitations, economic analyses incorporating CHARM data and conducted primarily in Europe have shown that candesartan cilexetil appears to be cost-effective when added to standard CHF treatment in patients with CHF and compromised LV systolic function. The use of candesartan cilexetil as part of antihypertensive therapy in elderly patients with elevated BP was also deemed to be cost effective in a Swedish analysis, primarily resulting from a reduced risk of non-fatal stroke (as shown in the SCOPE study); however, the generalizability of results to other contexts has not been established. Cost-effectiveness analyses comparing candesartan cilexetil with ACE inhibitors or other angiotensin receptor blockers in CHF or hypertension are lacking, and results reported for candesartan cilexetil in a Swedish economic analysis of ALPINE data focusing on outcomes for diabetes require confirmation and extension. Conclusions =========== The fixed dose combination of candesartan and HCTZ is a valuable addition to the armamentarium of drugs in the treatment of hypertension, because of its high efficacy in reducing BP, its tolerability, and the high compliance of patients with treatment. Comparative studies with losartan/HCTZ have consistently shown a higher clinical efficacy with the candesartan/HCTZ combination. Candesartan/HCTZ therefore assists both physicians and patients in achieving long-term treatment goals. Data on the reduction of cardiovascular endpoints with fixed dose combinations of antihypertensive drugs are scarce, as are the data for candesartan/HCTZ. However many trials have tested candesartan versus a non-RAS blocking comparator based on a standard therapy including thiazide diuretics. The indications tested were HF and stroke, and particular emphasis was put on elderly patients or those with diabetes. In patients with HF, for example, the fixed dose combination might be applied in patients in whom individual titration resulted in a dose of 32 mg candesartan and 25 mg HCTZ which can then be combined into one tablet to increase compliance with treatment. Also in patients with stroke the fixed dose combination might be used in patients in whom maintenance therapy with both components is considered. The preparation of this manuscript has been financially supported by Takeda Pharma, Aachen, Germany. Takeda had no role in outlining, editing or determining the final content of this review. This was left at the discretion of the authors who take full responsibility for its content. **Disclosures** TM, SU and PB disclose receipt of lecture honoraries and research support from Takeda Pharma, Aachen, Germany. ACCESS : The Acute Candesartan Cilexetil Therapy in Stroke Survivors; ACCOMPLISH : Avoiding Cardiovascular Events in Combination Therapy in Patients Living with Systolic Hypertension; ACEi : ACE inhibitor; AE : adverse event; ALPINE : Antihypertensive treatment and Lipid Profile In a North of Sweden Efficacy Evaluation; ARB : angiotensin receptor blocker; ARR : absolute risk reduction; AT : angiotensin; AUC : area under the curve; BP : blood pressure; CANDIA : CANdesartan and DIuretic vs Amlodipine in hypertensive patients; CCB : calcium channel blocker; CHARM : Candesartan in Heart Failure-Assessment of Reduction in mortality and Morbidity; CHD : coronary heart disease; CHF : chronic heart failure; CI : confidence interval; CV : cardiovascular; DBP : diastolic blood pressure; DDD : defined daily doses; DIRECT : The DIabetic REtinopathy Candesartan Trials; DRC : drug regimen compliance; ESC : European Society of Cardiology; ESH : European Society of Hypertension; HCTZ : hydrochlorothiazide; HDL : high-density lipoprotein; HF : heart failure; HR : hazard ratio; JNC : Joint National Committee; LIFE : Losartan Intervention For Endpoint reduction in hypertension; LV : left ventricular; MOSES : Morbidity and Mortality after Stroke -- Eprosartan vs Nitrendipine for Secondary Prevention; NHS : National Health Service (UK); NICE : National Institutes of Clinical Excellence; NNT : number needed to treat; OR : odds ratio; QUICKI : Quantitative Insulin-Sensitivity Check Index; PRA : plasma renin activity; QALY : quality adjusted life year; RAS : renin--angiotensin system; RRR : relative risk reduction; SBP : diastolic blood pressure; SCAST : Scandinavian Candesartan Acute Stroke Trial; SCOPE : Study on COgnition and Prognosis in the Elderly; UK : United Kingdom; US : United States. {#f1-vhrm-5-1043} ![Four out of 6 recommended dual antihypertensive combination therapies include blockers of the renin--angiotensin system.\ **Notes:** Red, recommendation including an angiotensin receptor blocker; green, recommendations including an ACE inhibitor.\ Reproduced with permission from Mancia G, De Backer G, Dominiczak A, et al. Guidelines for the Management of Arterial Hypertension: The Task Force for the Management of Arterial Hypertension of the European Society of Hypertension (ESH) and of the European Society of Cardiology (ESC). *J Hypertens*. 2007; 25(6):1105--1187.[@b17-vhrm-5-1043] Copyright © Lippincott Williams & Wilkins.](vhrm-5-1043f2){#f2-vhrm-5-1043} ![Combination therapy as an escalation option and as first-line therapy.\ **Notes:** Green, 2-drug combination therapy as an option for treatment escalation; red, 2-drug combination as a first line option in patients with marked elevation of blood pressure and high cardiovascular risk.\ Reproduced with permission from Mancia G, De Backer G, Dominiczak A, et al. Guidelines for the Management of Arterial Hypertension: The Task Force for the Management of Arterial Hypertension of the European Society of Hypertension (ESH) and of the European Society of Cardiology (ESC). *J Hypertens*. 2007; 25(6):1105--1187.[@b17-vhrm-5-1043] Copyright © Lippincott Williams & Wilkins.](vhrm-5-1043f3){#f3-vhrm-5-1043} ![Cost savings in the UK by intensifying antihypertensive drug treatment.[@b15-vhrm-5-1043]\ **Notes:** Optimizing antihypertensive therapy in the UK will cost *£*58.4 million and will prevent costs due to prevented stroke/ischemic heart disease (IHD) of *£*280.4 million resulting in a net benefit of *£*221.9 million.](vhrm-5-1043f4){#f4-vhrm-5-1043} ![Blood pressure reduction with 32 mg candesartan alone or in combination with 12.5 or 25 mg HCTZ in patients not sufficiently controlled on monotherapy.[@b42-vhrm-5-1043]](vhrm-5-1043f5){#f5-vhrm-5-1043} ![Results of the CHARM trial program.[@b52-vhrm-5-1043]--[@b55-vhrm-5-1043]\ **Notes:** \*p for heterogeneity 0.37; ^†^p for heterogeneity 0.33.\ Reprinted from Pfeffer MA, Swedberg K, Granger CB, et al. Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme. *The Lancet*. 362:759--766.[@b52-vhrm-5-1043] Copyright © 2003, with permission from Elsevier.](vhrm-5-1043f6){#f6-vhrm-5-1043} ![Development of diabetes -- results of a meta-analysis.\ Reprinted from Lam SK, Owen A. Incident diabetes in clinical trials of antihypertensive drugs. *The Lancet.* 369:1513--1514.[@b59-vhrm-5-1043] Copyright © 2007, with permission from Elsevier.](vhrm-5-1043f7){#f7-vhrm-5-1043} ![Results of the DIRECT trial program.[@b47-vhrm-5-1043],[@b73-vhrm-5-1043]](vhrm-5-1043f8){#f8-vhrm-5-1043} ###### Compelling indications and contraindications in the use of antihypertensive drug classes[@b17-vhrm-5-1043] **Conditions favoring use** **Compelling or *possible* contraindications** ------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Thiazide diuretics Isolated systolic hypertension (elderly)Heart failureHypertension in blacks GoutMetabolic syndromeGlucose intolerancePregnancy Beta blockers Angina pectorisPost-myocardial infarctionHeart failureTachyarrhythmiasGlaucomaPregnancy AsthmaA-V block (grade 2 or 3)Peripheral artery diseaseMetabolic syndromeGlucose intoleranceAthletes and physically active patientsChronic obstructive pulmonary disease Calcium antagonists (dihydropyridines) Isolated systolic hypertension (elderly)Angina pectorisLV hypertrophyCarotid/coronary atherosclerosisPregnancyHypertension in blacks TachyarrhythmiasHeart failure Calcium antagonists (verapamil/diltiazem) Angina pectorisCarotid atherosclerosisSupraventricular tachycardia A-V block (grade 2 or 3)Heart failure ACE inhibitors Heart failureLV dysfunctionPost-myocardial infarctionDiabetic nephropathyNon-diabetic nephropathyLV hypertrophyCarotid atherosclerosisProteinuria/microalbuminuriaAtrial fibrillationMetabolic syndrome PregnancyAngioneurotic edemaHyperkalemiaBilateral renal artery stenosis Angiotensin receptor antagonists Heart failurePost-myocardial infarctionDiabetic nephropathyProteinuria/microalbuminuriaLV hypertrophyAtrial fibrillationMetabolic syndromeACEi-induced cough PregnancyHyperkalemiaBilateral renal artery stenosis Diuretics (antialdosterone) Heart failurePost-myocardial infarction Renal failureHyperkalemia Loop diuretics End stage renal diseaseHeart failure ###### Laboratory values at baseline, and mean change (±SD) from baseline^[a](#tfn1-vhrm-5-1043){ref-type="table-fn"}^ after 8 weeks of treatment[@b42-vhrm-5-1043] **Candesartan 32 mg n = 653** **Candesartan 32 mg 12.5 mg HCTZ n = 654** **Candesartan 32 mg 25 mg HCTZ n = 664** -------------------------- ------------------------------- -------------------------------------------- ------------------------------------------ S-urate (μmol/L)   Baseline 334.5 (84.6) 335.2 (82.6) 339.1 (80.6)   Change 5.1 (52.0) 27.4 (52.3) 43.9 (57.8) S-potassium (mmol/L)   Baseline 4.5 (0.4) 4.5 (0.4) 4.5 (0.4)   Change 0.0 (0.4) 0.1 (0.4) −0.1 (0.4) S-glucose (mmol/L)   Baseline 6.0 (1.4) 6.2 (1.9) 5.9 (1.5)   Change 0.0 (1.0) 0.4 (1.0) 0.0 (0.9) S-triglycerides (mmol/l)   Baseline 2.0 (1.4) 1.9 (1.8) 2.0 (1.5)   Change 0.0 (1.2) 0.0 (1.6) 0.2 (2.2) S-cholesterol (mmol/L)   Baseline 5.8 (1.1) 5.8 (1.1) 5.9 (1.0)   Change 0.1 (0.7) 0.0 (0.8) 0.1 (0.8) S-creatinine (μmol/L)   Baseline 79.8 (15.8) 79.9 (14.8) 80.3 (15.4)   Change 0.0 (10.5) 1.8 (9.8) 4.5 (13.1) Baseline: after a run-in phase with 16 mg candesartan for 2 weeks and 32 mg candesartan for 6 weeks.

This talk is inspired by the story of Matthew about whom you will hear in a few moments. This ceremony acknowledges a transition in your medical education: A rite of passage from the study of basic medical science in classrooms to new encounters with patients at bedsides and in clinics. On choosing and being chosen ============================ You are an exceptional group of women and men. In the style of praise we have come to know from Lake Woebegone one could say that all the men here are strong, all the women are beautiful, and being the children of your parents, you are, of course, all well above average. You chose to study medicine I assume it was a matter of free will that led you to this choice, but what was your motivation to make it? Was it to serve your patients, or was it for the prestige of being a doctor, or to be blunt, was it for the money and the lifestyle? Did you have an epiphany of vocational passion and compassion, as described perhaps in your application essay to wherever it was sent, and will that essay make interesting and honest reading 25 years hence? A better question: Who will you serve all your days as a physician and in so doing, will you be true to yourself? While you made a choice and applied accordingly, you were also chosen. Many were called, but fewer were chosen to become members of the Class of 2008. Some of you may wish to question and evaluate the criteria by which the selection process operates, but having been chosen you are privileged; and as a result, you have acquired responsibilities: to yourself; to your family and friends who helped you to succeed; and also to your mentors -who have expectations of you. At this stage of your life, you have chosen to acquire a body of knowledge about the illnesses and diseases from which your fellow human beings can suffer; in due course, you too will suffer from one or other of those illnesses and diseases. Therefore, it is fair to say that while they are conditions that undermine our wellbeing, the diseases also reveal the importance of our health. Accordingly, the good physician has two roles: one to diagnose, to explain and to treat disease; the other to listen, to heal the person with the illness, and to bring that person back to a state of health. Life and now ============ "Life can only be understood backward but it must be lived forward." The philosopher Kierkegaard, whose aphorism this is, was echoing an enduring theme about life and its trajectory, equally apparent in a little folk rhyme: Yesterday is history, Tomorrow is a mystery; Today is a gift, and we call it - the Present. Whichever version you prefer, it refers to the gift of life in past, present and future terms. You have chosen to serve the present lives of your patients. Your encounters with those lives may primarily involve events around procreation, birth and childhood; or they may be encounters contained within the odyssey of an individual's lifespan; or they may be encounters mainly with the ending of a life. Ever since humankind began to describe and record its view of life, for example in the Sumerian epic Gilgamesh, or in Homer's The Odyssey, we have been told that our mortality (call it death) is what makes life itself so interesting and important. In our profession of medicine, there is a counterpart to that awareness: it is disease and illness that make health so important. But life with disease is suffering, and if, as physicians, we do not address the suffering, we will have ignored the place of healing in our profession. Our medical expertise, no matter how extraordinary it may be, has assisted in only part of the journey back to health. Physicianship is a term that embraces both the fixing of the disease and the healing of the person. We use knowledge about the illness to understand its effect on health; we use compassion and empathy for the process of healing; and when we are complete physicians, we are also aware of the role that culture plays in our views of health and disease. Poets have ways of saying things that get to the core of a theme. Here is one example - from TS Eliot ([@b1-mjm0902p084]): Where is the Life we have lost in living? Where is the Wisdom we have lost in knowledge? Where is the Knowledge we have lost in information? It has taken me a fair portion of my lifespan to appreciate the density of the messages in those subtle questions. When I graduated with my MDCM in 1955, I was full of information; I was very proud of it and I was probably arrogant. I know now that I had too little knowledge at the time and very little wisdom. I would have been better prepared as a physician if I had been more humble; the poet again had something to say about that ([@b2-mjm0902p084]). The only wisdom we can hope to acquire is the wisdom of humility - humility is endless. Mystery, illness and individuality ================================== Life and its emergent properties, for example the assembly of the fetus in the womb, remain imperfectly known; there is mystery here. It would be well for physicians to recognize that there is mystery because we can then begin to appreciate the hold that disease and illness have on the emotions of our patients; and, if we are honest, on ourselves. With our medical expertise we will diagnose and treat disease. If we do only that, we may forget that there is a patient who has the disease, and in the forgetting we will dehumanize our practice of medicine, and at the same time, erode the dignity of our patient and of ourself. The disease will become the object of our interest and the patient will only be an appendage. We can do better than that and physicianship will help us to do it better by seeing things differently, for example: i. It is a person, with an illness, who has come to you for help. ii. The person with the illness, and the disease in the person, are not equivalent. iii. Every person (every individual human being) has her or his particular form of any nominal disease or illness. iv. Good doctors know that each patient is a different person, that medicine is a science of the individual, and that to treat only the disease is to treat the patient as an object. A famous medical anecdote illustrates these ideas: Coleridge, the poet, has already written his great poems such as Kubla Khan and The Ancient Mariner. He is famous, he is in his mid-20s and he is profoundly addicted to opium. Kubla Khan was composed in an opium dream. His friends notice deterioration in Coleridge's health, they fear for him, and they arrange for him to be seen by Caleb Parry, the great physician in the nearby city of Bath. Coleridge goes, Parry receives, Parry cures Coleridge's opium addiction within the year. When asked how he did it, Parry replies: I did not treat his opium addiction, I treated Mr. Coleridge. If I were to name the chief concerns I have for you, citizens of the Class of 2008, among them, I would name depersonalizing those you have come to serve. You will be drowning in a sea of information. You will be swimming hard to keep up with evolving facts and technologies, and you will be struggling hard to keep up with practice norms and recurrent accreditation requirements. As a result, you may have too little time to listen to your patients, to let them tell you about their own mystery and worries. Somehow, your education, your personal view of life, your profession, and the society that sets the standards for your professional practices, must allow you the opportunity to practice physicianship. Yet, it is possible to keep body and soul in focus even under the most technocratic conditions. Let me illustrate with the story of Matthew. Matthew's story =============== The story is told using terms you will learn in the clinical curriculum that lies ahead. The story is simplified for clarity. Matthew was born on Feb. 24th, 2005, the second child of older parents. His life in utero was monitored by the full array of modern technologies. Nothing adverse was observed except for mild polyhydramnios. Amniotic membranes ruptured at 36 weeks and premature delivery occurred by cesarean section. Matthew weighed 3.26 kg, a normal weight for the fetal age. Extrauterine life did not begin well; severe respiratory distress required immediate intubation, and congenital anomalies were recognized. An operation on the first day of life identified a tracheo-oesophageal fistula, segmental tracheomalacia, and atresia of the lower oesophagus. The fistula was repaired and the stomach joined to the existing oesophagus. Matthew survived. Post-operative monitoring confirmed suspicion of a cardiac problem and on the third day of life a second trans-thoracic operation was performed to repair his total anomalous pulmonary venous return and to close a large ventricular septal defect. Again -Matthew survived. Friends of the family were correct in applauding both the effectiveness of the anesthesia that made surgery possible, and the awesome surgical and medical expertise that rescued Matthew; in their lay language - it was "a miracle". Matthew required assisted airway ventilation for various reasons, and because of his oesophegeal problem, it was not possible to feed him on his mother's breast milk by the normal enteral route. He was fed intravenously by total parenteral nutrition (TPN) -another neonatal technology. He gained weight and showed normal cognitive parameters. It was even possible for the parents to cuddle Matthew occasionally, despite all the tubes, monitors and attached equipment. By the third week of life, it was impossible to ignore another problem. Matthew had chylothorax, probably due to anomalous flow of lymph, explained by extending the field of congenital anomalies to include the thoracic duct. A drain to the exterior was installed to prevent lung compression; and because of the drain, there was a portal for infection; serious infection occurred. Antibiotics were effective and again Matthew survived. Some of his caregivers began to call him "little cat with all those lives"; others referred to a series of "rebirths". Meanwhile, a large community of friends and family praying for Matthew recognized the Spirit at work. Matthew's set of congenital anomalies might be explained by a single post-zygotic somatic mutation. There was no evidence for an inherited germ-line mutation or syndrome in this scenario. The parents were counselled accordingly. The story continues. Matthew had received only TPN for his nutrition. TPN can be a poisoned chalice and some infants harboring unknown susceptibility alleles will develop a particular form of lethal liver damage. Matthew developed hepatic cirrhosis and TPN feedings had to be terminated. In the 5th month of life, a jejunostomy tube was inserted and enteral feeding with his mother's milk initiated by this artificial route, because it was still not safe to feed by mouth. All was well, until the bowel perforated, perhaps because its structural integrity had been compromised early by the failure in utero to ingest amniotic fluid and thus condition the bowel. A reaction to the perforation set in and Matthew developed an Abdominal Compartment Syndrome where pressure compromises abdominal organ function and will cause death. The pressure was relieved by opening Matthew's abdomen and covering the gap with artificial material. Once again, Matthew survived. He was 5 months old. How to feed Matthew without unhappy consequences became the ultimate awesome challenge. The difficulty was never overcome, his muscles to support respiration did not strengthen and a life connected forever to tubes, ventilator and catheters lay ahead. The family and caregivers chose to end his suffering and Matthew was taken home where for the last day of his life he was free of the technologies. He was held in the arms of his family where he died peacefully. And who is Matthew? Matthew was our grandson who lived and died in Adelaide, Australia. My wife and I have contemplated again and again the little life of Matthew, all 7 months of it, minus one day, spent in an intensive care unit. Matthew became a hub for physicianship and intensive care. Matthew became the maker of links between the members of his extended family who came from different places on Planet Earth to be with him in Adelaide. Matthew was the sum of us, of the parents who made him, of the grandparents who made his parents, and so on back through the generations; Matthew in life and death is the sum of us all. His role on Earth, has been to make that sum much more than a bit of biological arithmetic and in so doing to bring out the best in expertise and compassion in all those who cared for him. And what is the role for grandparents who fly half way around the world to be with grandson and family -day after uncertain day. My wife and I discovered the wisdom in Milton's sonnet on his own illness ([@b3-mjm0902p084]); we became aware that: They also serve who only stand and wait. Just being there, we discovered, was - to serve. Matthew's story has multiple references to being reborn, an unmedical term but perhaps allowable when the person is an infant recently born and rescued back to life. Here, the term reborn implies there was a death or near death, as there would have been had there not been remarkable medical and surgical expertise and intervention. To return then to the theme of an odyssey in life's journey; where there is a birth, a life, and then a death. You will have noticed that a death gave you the cadaver from which you learned some anatomy. Indeed that person gave you a gift of his or her body at the beginning of your education in medicine. From another viewpoint, we have each been born and we have survived that perilous journey that began in the union of two haploid genomes. With the awareness of our own odyssey, of the journey, and of the end that comes to all of us, we are privileged to know about the gift of Life. And as I said earlier, with privilege comes responsibility. Each one of us will meet those responsibilities, and experience an odyssey, in our own individual way. Nonetheless, there are but two ways to approach the end, as was the case with Matthew; in one, we move on into eternal, ethereal life after this one ends; in the other, we return to the void from which we came. In each case, we did experience Life - a privilege never to be known by the inanimate objects of the universe from which we were made. Whichever view I may hold about my own odyssey, I can still enrich my life and that of my patients, and of all other persons known to me, by making my life and their lives as meaningful as possible. In its simplest form, it can be done by being kind to others, to all living creatures and to life on Earth; and by opposing bad things; and in the physician's case, by healing the suffering associated with illness and disease. This essay has the title: Who do I serve? Several types of service exist in this room at this moment: Service in medicine: to patients, families and communities. Service in education; to you and to the lifelong students that we are. Service in research: to the knowledge of tomorrow's medicine. Service to the unknowable: to the mystery of Life and Self. Envoie ====== I mentioned my lack of humility when I graduated with my own medical degree. When I joined the faculty of McGill University some years later, I became aware that my life had to be lived forward and I became more humble because of the responsibilities I now had to family, myself, and my new appointment. I was also aware of global contexts which for you today are particular, and which for me then included the Cold War, the nemesis of nuclear war, the social crimes of segregation and poverty, the emerging war in Viet Nam and the assassination of a President, among other things. I needed a moral guide and I found him in the person and writings of Thomas Merton, a cloistered Trappist monk. He reached across denominational boundaries and cultures and touched people like me. Merton kept a diary and in it he describes an epiphany he experienced en route to an appointment with - his physician. He wrote ([@b4-mjm0902p084]): "At the corner of Fourth and Walnut, in the center of the shopping district, I was suddenly overwhelmed by the realization that I loved all those people, that they were mine, and I theirs, that we could not be alien to one another even though we were total strangers. It was like waking from a dream of separateness ...". Merton goes on to describe "those people" in all their diversity and individuality, and ends with this extraordinary insight: "And if only everybody could realize this! But it cannot be explained. There is no way of telling people that they are all walking around shining like the sun. " As I stand here, and gaze at you, I see extraordinary men and women. What is there to say to you, as you go forth to meet your patients, except this: Remember your humanity, listen to your patients, explain their mysteries to them, be kind, love your fellow beings, be informed, be humble and wise - and shine like the sun. The White Coat Ceremony of McGill Faculty of Medicine is entitled "Donning the Healer's Habit" and pays homage to the late Dr. Joseph Wener, a cardiologist, internal medicine specialist, and a popular teacher. Faculty, students and their loved ones are gathered in a ceremony which focuses on the white coats, a symbol of compassion and patient care rather than power. The Ceremony takes place during the second year of the medical students' education at McGill Faculty of Medicine and marks their transition from being in the classroom to being in regular contact with patients. Dr. Charles Scriver was invited to give the keynote address at this Ceremony. Dr. Rita Charon, Keynote speaker at the White Coat Ceremony to the class of 2007, spoke about choosing and being chosen to study medicine; her address was titled "Levitation of Care". Her thoughts find echoes here. [^1]: **Dr. Scriver** is currently Professor Emeritus of Pediatrics, Biochemistry, Biology and Human Genetics in the Faculties of Medicine and Science at McGill University and Alva Professor Emeritus of Human Genetics. Dr. Scriver received his M.D.C.M. degree from McGill University in 1955. He completed residencies in both medicine and pediatrics at the McGill University teaching hospitals and the Harvard University Children's Medical Centre, respectively. He first became interested in human genetics as a McLaughlin Travelling Fellow at the University College Hospital Medical School's Human Metabolism Unit. Throughout his career, Dr. Scriver has been President of many Societies devoted to scientific research in pediatrics and human genetics, including the American Pediatric Society (1995) and the American Society of Human Genetics (1987). He has also been actively involved in many other organizations such as the National Academy of Sciences, the World Health Organization, and the American Association for the Advancement of Science. Dr. Scriver was Director of the Medical Research Council Group in Genetics (until 1994) and co-Director of the Canadian Genetic Diseases Network (until 1997). His current research focuses on, among others, the human genetic and phenomic variations at the PAH (PKU) locus and new ways to treat genetic diseases. Dr. Scriver has over 600 publications.

Background ========== One of the principal ways in which we make sense of the world is to group things and events into classes that share common characteristics. Human beings learn to do this intuitively in early childhood, and quickly develop an understanding of classes for commonly encountered objects and concepts which is shared by all those around them. However, for more specialised areas, the description of things or events in terms of classes tends to be a far less intuitive process that demands a carefully thought-out, explicitly articulated framework. Such classification frameworks make it much easier to compare information about entities and concepts, and to discern their similarities and differences. Within the domain of public health, little conceptual work has been done to develop shared definitions, terminologies or classifications. As a result, we have limited ability to compare public health activity across jurisdictions and countries, or even to ascertain whether we share common notions of what constitutes \'public health\'. This in turn hinders our ability to collect comparable, time-series data on expenditure, workforce, or performance, and to set and monitor benchmarks for these. In Australia, the governments of the six states and two territories are the major providers of public health services, while the responsibility for funding these services is shared between the Australian (national) Government and state and territory governments \[[@B1]\]. Local government (municipal and shire councils) also plays a role in delivering public health services, particularly in the areas of environmental health, urban planning, food safety and immunisation; this role varies among the states and territories. The National Public Health Partnership, a body set up in 1996 to strengthen collaboration between the Australian Government and state and territory governments, adopted the following definition for public health: *the organised response by society to protect and promote health, and to prevent illness, injury and disability. The starting point for identifying public health issues, problems and priorities, and for designing and implementing interventions, is the population as a whole, or population sub-groups*\[[@B2]\]. However, in Australia -- as in other countries \[[@B3]\] -- the term \'public health\' is a source of confusion, because it is also often used to refer to health services provided by the state or otherwise paid for by taxpayers out of \"the public purse\", as opposed to services provided by the private sector, or paid for by individuals or nongovernmental health insurance or health maintenance funds. Some jurisdictions use the alternative term \'population health\' to refer to the same domain, but this term is also poorly understood. The lack of widespread understanding about what constitutes public health hampers efforts to advocate for more resources for the sector. If we as public health practitioners cannot clearly describe the activities of our sector, the resources expended by it, and its outcomes, it will remain difficult to convince the public, politicians and other decision-makers that greater investment is needed \[[@B4]\]. The lack of basic conceptual development within public health has been recognised internationally. An expert panel convened by the United States (US) Centres for Disease Control in 1999 identified the use of common definitions and comparable data sources as being among the most important issues for achieving the goal of quality improvement in public health through performance measurement \[[@B5]\]. Five years later, lack of terminological and conceptual consensus was cited as obstructing even basic work in the area of public health finance in that country \[[@B6]\]. In Australia, a 2002 project that set out to develop a key set of performance indicators for public health practice recommended, as a priority, the development of a common classification system that could be used for measuring expenditure as well as for organising performance measurement activities \[[@B7]\]. A 2003 review of the financing of population health (defined as a subset of public health with a whole-of-population focus) in eight Organisation for Economic Co-operation and Development (OECD) countries, noted that comparability was hampered by differing definitions and categorisations of activity, lack of reliable data, and lack of uniformity in methods for extracting information \[[@B8]\]. Several conceptual models describing \'core\' or \'essential\' functions of public health exist, including the framework described in the Institute of Medicine\'s 1988 report on the status of public health in the US \[[@B9]\], the \'10 essential public health services\' proposed by an expert panel convened by the US Department of Health and Human Services \[[@B10]\], the Australian National Public Health Partnership\'s \'core functions for public health practice\' \[[@B2]\], a set of core functions promulgated by the Chief Medical Officer in the UK \[[@B11]\], and another developed by the World Health Organization (WHO) in 1996 \[[@B12]\] as well as \'essential public health functions\' developed from a three-country study in WHO\'s Western Pacific Region in 2003 \[[@B13]\]. The Pan American Health Organization developed \'essential public health functions\' and public health \'roles\' in a conceptual renewal of public health in 2002 \[[@B14]\], and revised these in 2007 \[[@B15]\]. A list of \'essential functions\' was recommended by the Canadian National Advisory Committee on Population Health in 2003 \[[@B16]\] (See Additional file [1](#S1){ref-type="supplementary-material"}). These conceptual models, in particular the \'10 essential public health services\', have proved valuable for deriving performance indicators, standards and associated measurement instruments \[[@B17]-[@B23]\]. However, all are essentially \'flat\' lists, or at best hierarchical taxonomies, which conflate discrete dimensions such as the purpose of public health activities, the health issues and problems addressed and the settings in which services are delivered, into single \'functions\'. None presents a well-defined theoretical framework for multiple aspects of effective public health practice \[[@B24]\]. Some of the many standard classifications that are already in use in health fields address aspects of public health. For example, WHO\'s international classifications of diseases \[[@B25]\], functioning and disability \[[@B26]\], and external causes of injury \[[@B27]\] can be used to classify morbidity and mortality data in terms of diseases, disability and injury of interest to public health. The recently created OECD system of health accounts \[[@B28],[@B29]\] classifies health care in three dimensions for the purposes of international comparisons of health care spending: sources of funding; service providers; and functions of care (the goals or purposes of health care; e.g. disease prevention, health promotion). While these dimensions are clearly separate, the functional activity category of \'Prevention and public health services\' \[[@B29]\] consists of a list of only six, non-exclusive, items (see Additional file [1](#S1){ref-type="supplementary-material"}). A multidimensional approach was adopted by the Eastern Region Public Health Observatory in the United Kingdom (UK), for the construction of their Public Health Information Tagging Standard (PHITS). PHITS was developed to categorise and provide structure to information provided on websites, and to improve the efficiency of the retrieval of web-based public health resources. PHITS has seven dimensions: \'Person\'; \'Time\'; \'Place\'; \'Determinants\'; \'Morbidity and Mortality\'; \'Services\'; and \'Policy\' \[[@B30]\]. It has now been integrated into a UK National Public Health Language thesaurus \[[@B31]\]. PHITS was designed primarily to categorise web-based information resources, rather than as a multi-purpose classification for public health. Like other existing classifications, it does not capture all the important dimensions of public health, which include its functions, the methods and interventions used to achieve these, the health issues and determinants of health which public health activities address, the resources and infrastructure they use, and the settings in which the activities occur. A multidimensional classification of public health that describes all these dimensions and their relationships, and adopts elements from existing classifications where appropriate, will serve multiple purposes. It will have utility for standardising the collection of information about public health programs, expenditure, workforce and performance. It will facilitate aggregate reporting and analysis of this information in ways that suit particular perspectives; for example, according to the health problem addressed, or the setting where public health activity occurred. This paper presents an initial version of such a multidimensional classification, and describes the process that we used to develop it \[[@B32]\]. Methods ======= We used an ontology-building process to develop the public health classification. The term \"ontology\" is used in several ways, but here we use it in the computer science or knowledge engineering sense of an explicit formal specification of the concepts in a domain (in this case, public health), their attributes and the relationships among them, which allows people (and computers) to share a common understanding of the structure of information. We chose this process, rather than a more traditional method for developing a classification, in recognition that a flexible, multidimensional classification structure would be needed in order to suit a variety of uses and user groups, and to exploit the near future capabilities of the Semantic Web. The Semantic Web is an initiative which aims to give meaning (semantics), in a manner understandable by machines, to the content of documents on the World Wide Web (and elsewhere) \[[@B33]\]. We used published methods for frame-based ontology development \[[@B34]\], and Protégé open-source, ontology-building software from Stanford University \[[@B35]\] as a development tool. After scanning available methods and software, we chose Protégé because it is freely available to everyone and does not require a commercial license, supports current Semantic Web standards such as the Resource Description Framework (RDF) \[[@B36]\] and Web Ontology Language (OWL) \[[@B37]\], and has an active community of interest, with strong representation from researchers in the health, biomedical and related sectors. Although Protégé provides support for the three types of OWL, we chose to develop our ontology in simpler CLIPS format in the first instance, as this is slightly easier to use, with a view to transforming it to OWL format at a later date. The four steps in the ontology-building process were as follows: 1\. Determine the domain and scope of the ontology. 2\. Consider reusing existing ontologies. 3\. Enumerate important terms in the ontology. 4\. Define the classes and class hierarchy \[[@B34]\]. A reference group of public health experts (see Acknowledgements) drafted initial responses to steps 1 to 4. Definitions of public health and of core and essential public health functions, including those used by the WHO \[[@B13]\], the OECD system of health accounts \[[@B29]\], the Australian National Public Health Partnership \[[@B2]\], the US Department of Health and Human Services \[[@B10]\], the US Association of Schools of Public Health \[[@B38]\] and the Pan American Health Organisation \[[@B14]\], the UK Department of Health \[[@B11]\] and the Canadian National Advisory Committee on SARS and Public Health \[[@B16]\] (see Additional file [1](#S1){ref-type="supplementary-material"}) were reviewed for step 2. We refined the classification through a series of consultations with public health experts and practitioners across Australia. The project consultation process sought to achieve agreement on a version of the classification that was \'good enough\', recognising that the classification of public health is a complex and technically difficult problem, with no definitive formulation or solution. Early consultations were informal, and designed to seek the views of content experts in particular domains (e.g. environmental health, health promotion). Later, more formal consultations were organised through reference group members representing various jurisdictions. Prior to each consultation meeting, we sent material introducing the classification project to participants. All consultations were conducted face to face. The number of participants varied from one or two, to larger groups of up to 15, and the duration varied from one to three hours. In each consultation, an introduction and background to the classification project were given with the aid of a slide presentation, after which an early version of the public health classification, rendered through a Web browser, was demonstrated. This was followed by a \'live\' session using Protégé, which allowed participants to explore the structure of the classification, to suggest additions and changes, and to immediately see their effect on the overall classification. Last, participants were asked to identify practical uses for a multidimensional public health classification. Following consultation meetings, the project reference group debated proposed changes to the classification before deciding to adopt or reject them. The Australian National Public Health Information Working Group -- a committee with representation from all states and territories as well as relevant national bodies -- discussed and provided feedback on an early draft version of the classification. Results ======= Principles of development ------------------------- Development of the public health classification was guided by the following principles: 1\. The classification should be multidimensional. 2\. A range of the most important dimensions need to be considered and developed concurrently. 3\. Existing classification systems (including international and Australian standards) should be used wherever possible. 4\. The classification should be inclusive and deliberately broad at the top levels. Specific boundaries and restrictions to the scope of the classification should be defined in practical applications, rather than be arbitrarily imposed during the development of the classification. The public health classification -------------------------------- Version one of the public health classification consists of six top-level classes: (public health) \'Functions\'; \'Health Issues\'; \'Determinants of Health\'; \'Settings\'; \'Methods\' (of intervention); and \'Resources and Infrastructure\', which are shown as circles in Figure [1](#F1){ref-type="fig"}, together with a hierarchy of subclasses, and, at the lowest level, instances. Each subclass and instance should have a subsumption (\"is-a\") relationship with its parent class. {#F1} Existing classifications (such as the international classifications of diseases \[[@B25]\], functioning and disability \[[@B26]\], and external causes of injuries \[[@B27]\], can be used as subclasses of the classes \'Health Issues\', \'Settings\' and \'Resources and Infrastructure\', while new subclass structures are proposed for the classes of public health \'Functions\', \'Determinants of Health\' and \'Methods\' (see Figure [1](#F1){ref-type="fig"}). The working definitions for the six top-level classes are shown in Table [1](#T1){ref-type="table"}, and their immediate subclasses are given in Table [2](#T2){ref-type="table"}. ###### Classification of public health: six top-level classes and their working definitions **Class** **Working definition** ------------------------------ ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Functions Public health functions. The purpose of public health interventions, actions, activities and programs. Health Issues Health, and well-being issues that affect health (\'issues\' includes: concerns, topics, problems). Health is defined (by the WHO) as \'a state of complete physical, mental and social well-being and not merely the absence of disease or infirmity\'. Determinants of Health Factors that influence health status and determine health differentials or health inequalities. They include, for example, natural, biological factors, such as age, sex and ethnicity; behaviour and lifestyles, such as smoking, alcohol consumption, diet and physical activity; physical and social factors, including employment and education, housing quality, the workplace and the wider urban and rural environment; and access to health care \[a\]. Methods The methods used by organised public health interventions (actions, activities, programs, services) to protect and promote health and prevent illness, injury and disability, that are designed to change population exposure, behavioural or health status. Settings Settings in which public health activities and interventions take place, institutional and social environments, partnerships, and locations (e.g. schools, local government, hospitals, workplaces). Resources and Infrastructure Resources and infrastructure, the means available for the operation of health systems, including human resources, facilities, equipment and supplies, financial funds and knowledge \[b\]. It includes both person-time and calendar time. Source: Gruszin S, Jorm L, Churches T, Straton J: *Public Health Classifications Project Phase One: Final Report: Report to the National Public Health Partnership Group*. Melbourne: National Public Health Partnership; 2005. \[a\] World Health Organisation: Health Impact Assessment (HIA): Glossary of terms used. Geneva; 2006. <http://www.who.int/hia/about/glos/en/index.html> \[b\] World Health Organization: Health Promotion Glossary. Geneva; 1998. \[WHO/HPR/HEP/98.1\] ###### Classification of public health: top two levels of all classes **Top-level class** **Level 2 subclasses** ---------------------------------- ----------------------------------------------------------- ------------------------------------ ------------------------------------------------------------------------------ **Functions** Assess health of populations ***Primary:*** Promote health and prevent disease, disability and injury Protect from threats to health ***Instrumental*** Ensure public health capability Build the evidence base for public health **Health issues** Health and well-being Injury Diseases and conditions Disability and functioning **Determinants of Health** Environmental Socioeconomic External causes of injury Person-level Health system **Methods** Advocacy and lobbying Health impact assessment Research and evaluation Communicable disease control specific Immunisation Road safety methods Community action Infection control Screening to detect disease/risk factors Community development Legislation and regulation Social action Counselling Lifestyle advice Social marketing Diagnosis Management of biological risk Training and workforce development methods Directed investment Monitoring and surveillance Treatment methods Environmental monitoring Personal skills development Urban planning methods Epidemiologic methods Political action Vector control methods Exercise of capabilities Public policy development Waste management methods Food safety methods Radiation safety methods Other methods of intervention Health education Remediation of environment methods **Settings** Educational settings Home settings Other settings Healthcare settings Workplace settings Includes LOCATIONS -- classification of geographical areas (e.g. postcodes). Local government and communities settings Transport settings **Resources and infrastructure** Administrative infrastructure Organisational systems Technical infrastructure Funds Partnerships Time Information systems Physical infrastructure Workforce Legislative infrastructure Policies Workforce development capacity Source: Gruszin S, Jorm L, Churches T, Straton J: *Public Health Classifications Project Phase One: Final Report: Report to the National Public Health Partnership Group*. Melbourne: National Public Health Partnership; 2005. The \'Functions\' class is currently the most highly developed. Both primary and instrumental functions were considered to be important in conceptualising public health (See Additional file [2](#S2){ref-type="supplementary-material"}). Primary functions are *ends*in themselves, while instrumental functions are *means*to those ends. Public health practitioners also described instrumental functions as \'supporting\', \'underpinning\' or \'cross-cutting\' functions because all primary functions rely on them -- they do not belong solely to any one of the primary functions. While there was reasonable agreement among public health experts regarding the subclasses for the top-level classes \'Health Issues\', \'Determinants of Health\', and \'Settings\', the remaining classes are in earlier stages of development. The \'Methods\' class was the subject of some disagreement, with some experts preferring to narrow its scope to those methods that are *peculiar*to -- or only used by -- public health (e.g. population-based epidemiology, health promotion, environmental risk assessment). Others favoured an inclusive approach that would capture *all methods*used by public health, including those that, while not specific to it, are employed by public health workers in the normal course of their work (e.g. administration, management, policy development). An inclusive approach was adopted, in line with development principle (4) above. Public health experts expressed diametrically opposed views as to whether \'infrastructure\' represented aggregates of \'resources\', or whether \'resources\' were in fact a subclass of \'infrastructure\'. An inclusive approach was adopted, in line with development principle (4) above, with the relevant class termed \'Resources and Infrastructure\', and its subclasses capturing both compound (e.g. administrative infrastructure, information systems) and unitary elements (e.g. funds, workforce). Further work is needed to disentangle the mixture of partitive (holonym-meronym) and subsumption (hypernym-hyponym) relationships among the current subclasses of this top-level class. A copy of the full report on this initial phase of the project has been included as an additional file to this paper (Additional file [3](#S3){ref-type="supplementary-material"}), together with copies of the underlying classification ontology as it stood at the time of writing. Two sets of ontology files are provided: a set of Protégé project files (Additional file [4](#S4){ref-type="supplementary-material"}), and a set of interlinked HTML files that can be explored using a Web browser (Additional file [5](#S5){ref-type="supplementary-material"}). Potential uses for the classification ------------------------------------- The classification is used by assigning zero or more attributes chosen from each of the top-level class hierarchies to the \"thing\" being classified. A wide range of \"things\" can potentially be organised according to the classification, including (but not restricted to) public health policies, programs and interventions, the population groups they target, and their outcomes. Figure [1](#F1){ref-type="fig"} depicts some examples of things that could be classified, shown as heptagons. Many participants in the development process expressed a view that the classification would assist them in describing what public health is, and what its characteristics are. They also identified a range of questions that, currently, are difficult or impossible to answer, but which potentially could be answered if the multidimensional classification were used to facilitate aggregated reporting on public health activity. Examples of such questions are given in Figure [2](#F2){ref-type="fig"}. {#F2} Participants also suggested a range of potential practical applications for the classification. Examples of these are given in Figure [3](#F3){ref-type="fig"}. A knowledge base to support communicable disease surveillance, also constructed using the Protégé software suite, has been described \[[@B39]\]; it should be possible to use the classification ontology to create analogous databases for other specific or general areas of public health practice. {#F3} Discussion ========== The process of developing the classification brought to light several areas of basic disagreement among Australian public health practitioners regarding the nature of public health practice. The inclusion, or otherwise, of preventive services delivered on a one-to-one basis to individuals was particularly contentious. Such preventive services include screening and detection, immunisation, and counselling and lifestyle advice to support healthy behaviour, as well as management (through lifestyle changes or pharmacological means) of disease risk factors such as high blood pressure and high cholesterol. Many participants argued that those individual preventive services related to communicable disease (immunisation, contact tracing and treatment for sexually transmitted infections) are public health activities because they help to protect the health of the whole population, through herd immunity and controlling the spread of infection. Others felt that immunisation is only a legitimate part of public health activity when it is delivered as part of an organised government-funded program, such as through local government or school health services. Most public health practitioners agreed that early detection of disease through screening is a public health activity when it is delivered through an organised government-funded program (such as national breast and cervical cancer screening programs). Less clear was whether screening that is not part of an organised program, such as opportunistic bone density screening for osteoporosis, should be seen as part of public health. There was substantial disagreement among public health practitioners regarding the inclusion, or otherwise, of activities relating to prevention and management of non-communicable disease through individual counselling or other interventions directed at lifestyle risk factors (smoking, poor nutrition, risky alcohol use and lack of physical activity), and the early detection and management of biological risk factors such as high blood pressure and high cholesterol. Many contended that the diagnosis of a pathological condition (such as high blood pressure) or disease marked the boundary of public health practice. Others regarded this boundary as spurious, because the pathophysiological processes which underlie the development of chronic disease are continuous, and because interventions such as anti-hypertensive and cholesterol-lowering drugs, or even the \'Polypill\' \[[@B40]\], may have dramatic benefits in terms of morbidity and mortality at the population level. By adopting an inclusive approach, the public health classification allows decisions about boundaries, inclusions and exclusions, to be made at the level of individual applications of the classification. This is especially useful for those boundary issues -- such as where public health ends and clinical practice begins -- about which opinion may evolve with knowledge about preventive interventions, and how and by whom they are best delivered. The issue of whether a public health classification should be restricted to a domain solely within the health sector or whether it should it include the health-related activities of other sectors was also frequently raised. Most public health practitioners agreed, when pressed, that accounting for public health should include the activities of, and investments by, the non-health portfolios (such as education and transport) of national and state governments, as well as the relevant activities of local governments and non-government organisations (NGOs). This is consistent with the definition of public health used by the US Institute of Medicine as: \"what we, as a society, do collectively to assure the conditions for people to be healthy\" \[[@B9]\]. However, in practice there are major difficulties in capturing information on public health-related activities and expenditure by non-health sectors. In Australia, current public health expenditure reporting is limited to the health portfolio expenditures of the state and national governments \[[@B1]\]. One view was that the activities of non-health sectors should only be considered \'in scope\' when public health is their primary purpose (e.g. immunisation organised by local government). The \'boundary\' issues we encountered reflect the way that public health activity is conceptualised, and organised, in Australia. Similar exercises conducted in other countries would doubtless highlight different issues. For example, public health services in the US (which has a strongly privatised approach to health care) are seen as having a major role in filling \'gaps\' in health service provision (such as maternal and child health services) for those without access to health insurance, as well as in evaluating the accessibility and quality of personal health care services \[[@B10]\]. In several European countries, provision of many public health services is fully devolved to the level of local municipalities, which also have responsibility for issues such as air and water quality, noise diminution and waste management. It is likely that the organisation of services in these countries influences the way their public health practitioners conceptualise the boundaries between primary health care, public health and environmental protection. A comparison of the published public health functions of other nations (see Additional file [1](#S1){ref-type="supplementary-material"}) shows that some (e.g. those of Canada \[[@B16]\] and the Americas \[[@B14],[@B15]\]) are limited to primary functions, while others include both primary and instrumental functions. In the UK \[[@B11]\], both primary (e.g. health promotion and disease prevention programs) and instrumental (e.g. development and maintenance of a public health workforce) functions are prominent. Our classification captures the majority of functions that other nations have described, including the (instrumental) \'partnership\' and \'research\' functions that are present in both the UK core functions \[[@B11]\] and the 10 essential public health services of the USA \[[@B10]\]. A \'quality assurance\' function does not currently feature in our classification, although it is specified in the published functions for public health in the USA \[[@B10]\], the Americas \[[@B14],[@B15]\], the UK \[[@B11]\] and in WHO\'s most recent work \[[@B13],[@B15]\]. Public health practitioners and experts in Australia at no stage suggested that such a function was a critical part of their work. It is possible that their views may subsequently have changed, particularly in light of several recent scandals \[[@B41]-[@B43]\] relating to the safety and quality of the care provided in Australian public hospitals. The approach we adopted in developing the classification should maximise its flexibility for application in other settings. The ontology-building process offered particular advantages in dealing with divergent (and often strongly held) views regarding what was and was not \'in scope\'. Although defining and specifying classes was central to the process, the emphasis was on modelling the relationships among classes, rather than on the within-class hierarchies. We were able to adopt an inclusive approach, leaving scope for decisions about rules and exclusions to be made at the level of specific practical applications of the classification. It is to be hoped that such practical applications will make the areas of contention explicit, encourage debate, and offer a way to move towards a common language to describe public health activity in Australia and elsewhere. Conclusion ========== The public health classification is an initial attempt to describe the important dimensions of public health activity. It will facilitate the organisation of information about public health activity so that it can be used to address questions relating to any of these dimensions, either singly or in combination. The authors encourage readers to use the classification, and to suggest further refinements. It is our intention to further refine and extend several of the class hierarchies, and to convert the ontology into RDF and OWL format to make it suitable for use in Semantic Web applications. We welcome potential collaborators in this endeavour. Abbreviations ============= HIV/AIDS: Human Immunodeficiency Virus/Acquired ImmunoDeficiency Syndrome; OECD: Organisation for Economic Co-operation and Development; UK: United Kingdom; US: United States of America; WHO: World Health Organization; RDF: Resource Description Framework; OWL: Web Ontology Language (sic). Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= LJ compiled the argument and drafted the background, discussion and conclusions of the manuscript. SG reported the results, contributed to the background and discussion, and prepared the figures and tables. TC informed the directions of the project, provided key conceptual and practical input to its results, and reviewed and edited the manuscript. All authors worked on the project, and have co-authored, with Judy Straton (see Acknowledgements), the report of phase one of the project. All authors read and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1 **Comparison of published public health functions and roles**. Table presenting public health roles and functions published in international reports. ###### Click here for file ###### Additional file 2 **Classification of public health: detail of the functions class**. Table presenting the \"Functions\" class of the public health classification, and its subclasses. ###### Click here for file ###### Additional file 3 **Gruszin S, Jorm L, Churches T, Straton J: *Public Health Classifications Project Phase One: Final Report: Report to the National Public Health Partnership Group*. Melbourne: National Public Health Partnership; 2005**. Full report of Phase One of the Australian Public Health Classifications Project. ###### Click here for file ###### Additional file 4 **Public Health ontology (PHont) Version 1.0**. Protégé version 3.1 project files for Public Health Ontology (PHont) Version 1.0. ###### Click here for file ###### Additional file 5 **Public Health ontology (PHont) Version 1.0**. HTML web page version of Public Health Ontology (PHont) Version 1.0 ###### Click here for file Acknowledgements ================ The following people contributed to the development of the classification through participating in the Reference Group for the Public Health Classifications Project: Louisa Jorm, Chair (New South Wales Department of Health); Michael Ackland (Department of Human Services, Victoria); Andrea Casasola (Queensland Health); Jenny Cleary, Pam Gollow (Department of Health and Community Services, Northern Territory); Charles Guest (ACT Health, Australian Capital Territory); Paul Jelfs (South Australian Department of Health); Paul Magnus, Catherine Sykes, Robert Van der Hoek (Australian Institute of Health and Welfare); Dean Martin (Australian Bureau of Statistics); Karen Roger (National Public Health Partnership Secretariat); Colin Sindall (Australian Government Department of Health and Ageing); Judy Straton (Department of Health Western Australia); Tony Woollacott (South Australian Department of Health). Professor Richard Madden (National Centre for Classification in Health, University of Sydney) made important conceptual contributions to the classification. The (Australian) National Public Health Partnership funded the Public Health Classifications Project, some results of which are reported in this article. Additional sources of funding for the authors were the New South Wales Department of Health (L Jorm and T Churches) and the Public Health Information Development Unit, The University of Adelaide (S Gruszin). The funding body was represented on the Reference Group, and as such had a role in the project design, operation, interpretation and reporting of results.

Introduction {#Sec1} ============ Malignant brain tumors are considered as the most debilitating tumor types, mostly due to a bad quality of life, poor prognosis and limited therapeutic success \[[@CR1]\]. In 2010, in the USA alone, more than 130,000 patients with primary malignant brain tumors were identified \[[@CR1]\]. The updated World Health Organization (WHO) classification of tumors of the central nervous system not only considers histological but as of 2016 also molecular criteria to guarantee for a more refined diagnosis of the tumors with the main goal to optimize the treatment strategies for each individual patient \[[@CR2]\]. Depending on the tumor type and localization, treatment recommendations comprise surgical removal, adjuvant radiotherapy and/or chemotherapy (e.g., temozolomide) \[[@CR1], [@CR3]\]. However, current treatment approaches have shown unsatisfactory outcomes. In particular, chemotherapy has failed to improve survival leading to a lethal course of the disease within 12--18 months in particular in high-grade brain tumors. Sequencing studies have identified molecular alterations in brain tumors which may constitute promising targets for their treatment with molecularly targeted anticancer drugs \[[@CR4]\]. However, small-molecule inhibitors of these pathways have not demonstrated significant therapeutic efficacy in the clinic \[[@CR5]\]. This has been attributed to the inability of most of these drugs to cross the blood--brain barrier (BBB) and achieve therapeutically effective concentrations inside the brain. The BBB consists of brain capillary endothelial cells linked by tight junctions, which limit paracellular diffusion of drugs into the brain \[[@CR6]\]. The protective function of the BBB is further enhanced by efflux transport proteins in the luminal membrane of brain capillary endothelial cells, i.e., P-glycoprotein (ABC subfamily B member 1, ABCB1) and breast cancer resistance protein (ABC subfamily G member 2, ABCG2), which were shown to work together in restricting the brain entry of dual ABCB1/ABCG2 substrate drugs \[[@CR7]--[@CR9]\]. The majority of currently known molecularly targeted anticancer drugs are ABCB1/ABCG2 substrates and show very limited brain distribution \[[@CR8], [@CR9]\]. ABCB1 and ABCG2 may also be overexpressed in the membrane of brain tumor cells, which may thus form a second barrier to the effective treatment of brain tumors \[[@CR10]--[@CR14]\]. There is strong evidence that significant BBB disruption, as evidenced by contrast enhancement in T1-weighted magnetic resonance imaging (MRI) sequences following intravenous administration of gadolinium-based contrast agents, occurs in primary high-grade brain tumors \[[@CR14], [@CR15]\]. However, this BBB disruption is mostly restricted to the central, necrotic areas of brain tumors and does not extend to regions distant from the tumor core, where infiltrative tumor cells remain protected by an intact BBB \[[@CR14], [@CR15]\]. Moreover, gadolinium-based contrast agents possess distinct physicochemical properties as compared with small-molecule anticancer drugs and may, therefore, not be representative of brain tumor delivery of such drugs. In contrast to high-grade brain tumors, considerably less information is available with respect to BBB permeability in lower grade brain tumors, which encompass a heterogeneous group of tumors that are clinically, histologically and molecularly diverse and often progress to high-grade tumors \[[@CR14]--[@CR16]\]. These types of brain tumors usually show no or only little contrast enhancement on MRI. We have developed \[^11^C\]tariquidar as a prototypical dual ABCB1/ABCG2 substrate radiotracer for positron emission tomography (PET) imaging \[[@CR17]\]. \[^11^C\]Tariquidar is derived from the third-generation ABCB1 inhibitor tariquidar \[[@CR18]\]. It is a metabolically stable substrate of mouse and human ABCB1 and ABCG2 \[[@CR17], [@CR19]\] and shows very low brain distribution when ABCB1 and ABCG2 are functional and markedly increased brain uptake when ABCB1 and ABCG2 are pharmacologically inhibited or genetically knocked out \[[@CR17]\]. As such, tariquidar closely resembles many molecularly targeted anticancer agents, which may be of interest for the treatment of brain tumors (e.g., gefitinib, erlotinib, dasatinib, imatinib and pictilisib) \[[@CR20]\]. The aim of the present exploratory study was to investigate regional brain distribution of \[^11^C\]tariquidar in patients suffering from non-contrast-enhancing brain tumors (WHO grades I--III) by means of PET imaging. In addition, quantitative targeted absolute proteomics (QTAP) was applied to determine ABCB1 and ABCG2 levels in surgically resected tumor tissue. Materials and methods {#Sec2} ===================== All study-related procedures were performed at the Medical University of Vienna in accordance with the International Conference on Harmonization-Good Clinical Practice guidelines and the Declaration of Helsinki. Standard protocol approval was obtained from the Ethics Committee of the Medical University of Vienna and the national competent authority. The study was registered under the European Union Drug Regulating Authorities Clinical Trials Database (EUDRACT) number 2011-004189-13 (date of registration: February 23, 2012; <https://www.clinicaltrialsregister.eu/ctr-search/search?query=2011-004189-13>). Prior to study participation, all subjects gave oral and written consent. Study population {#Sec3} ---------------- A total of seven patients (p01--p07, four female and three male, mean age of 44 years at the time of the PET scan, range 37--57 years) with an intrinsic, non-contrast-enhancing brain tumor and elected for neurosurgery were enrolled in this study. However, one patient denied surgical intervention after study inclusion. PET scans were performed between June 2014 and October 2015. Tumor entities were based on histopathological and molecular analyses and ranged from low-grade to high-grade brain tumors according to the 2016 WHO Classification of Tumors of the Central Nervous System \[[@CR2]\]. Patient recruitment was performed in close cooperation with the Department of Neurosurgery at the Medical University Vienna. Radiotracer synthesis {#Sec4} --------------------- \[^11^C\]Tariquidar was synthesized and formulated in sterile 0.9% (w/v) aqueous saline solution/ethanol (9/1, v/v) containing 0.7% (v/v) polysorbate-80 for intravenous injection into study participants as described previously \[[@CR21]\]. Molar activity at the time of injection was 31 ± 8 GBq/μmol and radiochemical purity was 97 ± 1%. PET imaging protocol and general study procedures {#Sec5} ------------------------------------------------- Prior to surgical brain tissue removal, all patients included in this study were scheduled for a 60-min dynamic \[^11^C\]tariquidar PET scan acquired in three-dimensional mode using an Advance PET scanner (General Electric Medical System, Milwaukee, WI). On the study day, a venous catheter was placed in a cubital vein for radiotracer injection and an additional arterial catheter was applied for arterial blood sampling. Subjects were placed in supine position on the imaging bed with the head stabilized in a fixing mold attached to the imaging bed. A brief transmission scan (duration: 5 min) was performed for tissue attenuation of photons prior to the brain PET scan. Subsequently, \[^11^C\]tariquidar was injected intravenously over 20 s (injected activity: 371 ± 26 MBq, corresponding to 11 ± 3 μg of unlabeled tariquidar) while simultaneous dynamic PET imaging was accompanied by arterial blood sampling. Arterial blood samples were collected initially at intervals of 7 s during the first 3 min after radiotracer injection and further on at 3.5, 5, 10, 20, 30, 40 and 60 min after injection. Whole blood was centrifuged to obtain plasma. Selected plasma samples were analyzed with a previously described solid-phase extraction protocol for radiolabeled metabolites of \[^11^C\]tariquidar \[[@CR21]\]. In brief, plasma (2--4 mL) was diluted with water (2 mL), spiked with unlabeled tariquidar (10 μL, 20 mg/mL in DMSO), acidified with 5 M aqueous hydrochloric acid (40 μL), and loaded on a Sep-Pak vac tC18 cartridge (Waters Corp.), which had been pre-activated with methanol (3 mL) and water (5 mL). The cartridge was first washed with water (5 mL) and then eluted with methanol (2 mL) followed by aqueous ammonium acetate buffer (0.2 M, pH 5.0, 1.5 mL). Radioactivity in all three fractions (plasma, water, and methanol/buffer) was quantified in a gamma counter. Radioactivity in the plasma and water fractions contained polar radiolabeled metabolites, whereas unchanged \[^11^C\]tariquidar was recovered in the methanol/buffer fraction. PET data analysis {#Sec6} ----------------- T1-weighted (pre- and post-gadolinium enhanced) as well as T2-weighted FLAIR MRI data (Siemens Magnetom Trio, Tim System, Siemens Healthcare Diagnostics GmbH, Austria) were available from routine clinical examinations. Summation PET images and T2-weighted FLAIR MRIs were co-registered. The tumor and a contralateral tumor-free brain area, which were both of comparable size (Additional file [1](#MOESM1){ref-type="media"}: Table S1), were manually defined as volumes of interest (VOIs) using PMOD 3.6 imaging software (PMOD Technologies LLC, Zürich, Switzerland), whereby adjacent major blood vessels or ventricular structures were avoided. In one patient (p03), an additional VOI was outlined on the PET images, which corresponded visually to enhanced radioactivity uptake within the tumor (tumor PET enhanced). In addition, a gray matter VOI for normal brain tissue covering the brain hemisphere contralateral to the tumor was extracted for all patients using the Hammersmith n30r83 3D maximum probability atlas of the human brain \[[@CR22]\] as described previously \[[@CR19]\]. Time--activity curves (TACs) in units of kBq/mL were generated for the entire 60-min scanning period. As the percentage of radiolabeled metabolites in arterial plasma samples was \< 10% at all studied time points and as a previous study has shown that modeling outcome parameters were very similar in healthy volunteers for metabolite corrected and uncorrected input functions \[[@CR21]\], no metabolite correction was applied to the arterial plasma input function in the present study. The area under the brain and plasma TACs (AUC) was calculated using Prism 8.0 software (Software, La Jolla, CA, USA). The ratio between the brain AUC and plasma AUC, designated as AUCR, was calculated as a parameter of radiotracer brain distribution \[[@CR19]\]. In addition, Logan graphical analysis \[[@CR19], [@CR23]\] was performed in Microsoft Excel using the arterial plasma input function (not corrected for metabolites) to determine total distribution volume (*V*~T~), which equals the brain-to-plasma radioactivity ratio at steady state. Preparation of plasma membrane fraction from brain tumor samples {#Sec7} ---------------------------------------------------------------- Plasma membrane fractions were isolated as described previously with minor modifications \[[@CR24]\]. Frozen tumor tissues collected during neurosurgery of four patients (p01, p02, p03, p05) were thawed at +4°C, washed at least twice with isotonic buffer solution A (10 mM phosphate buffer pH 7.4, 0.1 M KCl) containing a protease inhibitor cocktail, minced into 1-mm pieces and homogenized using an Ultra-Turrax® (IKA®-Werke GmbH & Co. KG, Staufen, Germany) for 5 min at +4°C. The homogenates obtained were centrifuged at 10,800*g* for 15 min at +4°C and the supernatants were collected and ultracentrifuged at 100,000*g* for 60 min at +4°C. The plasma membrane fraction was obtained from the resulting pellet which was suspended in buffer B (20 mM Tris, pH 7.4, 0.25 M sucrose, 5.4 mM EDTA) containing protease inhibitor cocktail. The BCA protein assay kit (ThermoFisher Scientific, Villebon sur Yvette, France) was used for the total protein quantification. Protein digestion {#Sec8} ----------------- Plasma membrane fractions were digested as described previously without modifications \[[@CR25], [@CR26]\]. Briefly, proteins were solubilized in denaturing buffer (7 M guanidine hydrochloride, 10 mM EDTA, 500 mM Tris, pH 8.5), reduced by DTT and alkylated by iodoacetamide. The alkylated proteins were precipitated with methanol--chloroform--water, resolubilized in 1.2 M urea and 0.1 M Tris, pH 8.5. Samples were first digested using rLysC endoprotease (enzyme:protein ratio = 1:50) for 3 h at room temperature. Then trypsin (enzyme:protein ratio = 1:100) and 0.05% (W/W) ProteaseMAX were added and samples were incubated at 37°C overnight. The stable isotope-labeled peptide mixture (750 fmol of each labeled peptide/50 μg of total protein) was added in trypsic digest before ultrahigh-performance liquid chromatography--tandem mass spectrometry (UHPLC--MS/MS) analysis. Protein quantification by UHPLC--MS/MS {#Sec9} -------------------------------------- ABCB1, ABCG2 and Na^+^/K^+^-ATPase proteins were quantified by the determination of the peptide concentration using UHPLC--MS/MS in multiplexed selected reaction monitoring (SRM) method. Each peptide analyzed was specific to each protein and was released after protein digestion by trypsin. The selected peptides were FYDPLAGK (human specific), VGTQFIR (human and mouse specific) \[[@CR27]\], and AAVPDAVGK \[[@CR28]\] for ABCB1, ABCG2 and Na^+^/K^+^-ATPase, respectively. Samples were injected into an Acquity UPLC® system (Waters, Manchester, UK), equipped with an Acquity UPLC BEH® C18 column (Peptide BEH® C18 Column, 300Å, 1.7 μm, 2.1 mm × 100 mm) supplied by Waters (Guyancourt, France). The mobile phase consisted of mixture of water (formic acid 0.1% (v/v)) and acetonitrile. It was operated with a flow rate of 0.3 mL/min in gradient mode. The total duration of analysis was 34 min. Data were recorded with a Waters Xevo® TQ-S mass spectrometer (Waters, Manchester, UK). Measurements were performed using positive electrospray ionization (ESI) with ion spray capillary voltage at 2.80 kV. Drying gas temperature was set to 650°C at a flow rate of 800 L/h. Detection was performed in multiplexed SRM mode using three or four transitions per native or labeled peptide and the quantification CV% between transitions was lower than 5%. Skyline® software \[[@CR29]\] was used for the optimization of the specific transition parameters (i.e., collision energy (CE) and peak integration). The area ratios of light to labeled peptide were exported from Skyline® and quantification was performed from calibration curves using Microsoft Excel®. Statistical analysis {#Sec10} -------------------- This study was exploratory; sample size was based on feasibility and not on power to test a statistical hypothesis. Differences in PET imaging outcome parameters between tumor and tumor-free brain tissue were assessed with a Wilcoxon matched-pairs signed rank test. To assess correlations, the Spearman\'s rank correlation coefficient (*ρ*) was calculated. A *p* value \< 0.05 was considered statistically significant. Results {#Sec11} ======= Patient population {#Sec12} ------------------ Table [1](#Tab1){ref-type="table"} summarizes demographic data of the enrolled brain tumor patients. Out of the seven included patients, three were diagnosed with diffuse astrocytic and oligodendroglial tumors, grade II or III (p02, p03 and p06, Table [1](#Tab1){ref-type="table"}). Further, p03 and p06 had already undergone previous brain surgery in the past and were re-evaluated towards brain tumor progression by their responsible physicians. P07 refused neurosurgical intervention after inclusion and PET imaging. Therefore, histopathological entity was unknown. At the time of the PET scan, five out of seven patients received antiepileptic therapy for seizure prophylaxis (Additional file [1](#MOESM1){ref-type="media"}: Table S2). P01, p03 and p04 were treated with antidepressants. P02 received a statin and p07 was under treatment for hypertension. Further, p06 received thyroid hormone replacement therapy. One subject (p05) was free of any medication. Table 1Demographic data of enrolled brain tumor patientsp01p02p03p04p05p06p07Weight (kg)7865110628360108Age at time of PET (years)43574337384242SexFFFMMFMTime difference between PET and surgery (days)7102248396No surgeryNeuropathological diagnosisDiffuse glioma, IDH-wildtype and 1p/19q-codeleted, NEC (not elsewhere classified)Oligodendroglioma, IDH-mutant and 1p/19q-codeleted(Focally) anaplastic oligodendroglioma, IDH-mutant and 1p/19q-codeletedDysembryoplastic neuroepithelial tumorSubependymoma(Focally) anaplastic astrocytoma, IDH-mutantN.a.LocalizationLeft frontal/centralRight fronto-temporalRight frontalLeft mesiotemporalRight mesiotemporalRight fronto-temporo-insularLeft mesiobasal temporalWHO classification (grade)IIIIIIIaIIIIIaN.a.M, male; F, female; N.a., not available^a^Pre-operated Imaging data {#Sec13} ------------ Following intravenous injection of \[^11^C\]tariquidar, only a very low amount of radiolabeled metabolites was detected in plasma for the duration of the PET scan. At 60 min after radiotracer injection, 91.1 ± 3.1% of total radioactivity in plasma was in the form of unchanged \[^11^C\]tariquidar. Brain uptake of radioactivity was very low. Tumors were best visualized by T2-weighted FLAIR MRI with a median tumor volume of 12.0 cm^3^ (range 4.2--81.8 cm^3^) (Additional file [1](#MOESM1){ref-type="media"}: Table S1). None of the tumors showed appreciable gadolinium-enhanced areas on T1-weighted MRIs (not shown). In Fig. [1](#Fig1){ref-type="fig"}, T2-weighted FLAIR MR images, co-registered PET/T2-weighted FLAIR MR images and PET images are shown for all patients. In the PET images, no major visual differences in radioactivity distribution between tumor and tumor-free brain areas were evident (Fig. [1](#Fig1){ref-type="fig"}). However, in one patient (p03), a small area of enhanced radioactivity uptake (1.6 cm^3^) was observed within the tumor volume (44.9 cm^3^) (Fig. [1](#Fig1){ref-type="fig"}). As outcome parameters of \[^11^C\]tariquidar brain distribution, we determined AUCR and *V*~T~ \[[@CR19]\], which are shown in Fig. [2](#Fig2){ref-type="fig"} (based on the VOIs defined with PMOD). Normal brain gray matter *V*~T~ values from the entire brain hemisphere contralateral to the tumor (not shown in Fig. [2](#Fig2){ref-type="fig"}) were higher than those in contralateral tumor-free brain area containing both gray and white matter (0.38 ± 0.26 versus 0.16 ± 0.09). There was a good correlation between *V*~T~ values and AUCR values (ρ = 0.8762; *p* \< 0.0001, not shown). AUCR and *V*~T~ values were very low and rather variable among patients, both in tumor and contralateral tumor-free brain areas. No significant differences in AUCR and *V*~T~ values between tumor and tumor-free areas were found (Fig. [2](#Fig2){ref-type="fig"}). In p03, the small area of enhanced radioactivity uptake within the tumor had higher AUCR and *V*~T~ values than the entire tumor and normal brain tissue. Fig. 1Transversal T2-weighted FLAIR MR images (left image), co-registered PET/T2-weighted FLAIR MR images (middle image) and \[^11^C\]tariquidar PET average images (0--60 min) (right image) in all patients. The contours for the tumor and contralateral tumor-free brain area are shown in pink and green colors, respectively. For p03, a tumor area with enhanced radioactivity uptake as compared with the rest of the tumor was visible (indicated by white arrow) Fig. 2Outcome parameters for \[^11^C\]tariquidar brain distribution (**a** AUCR, **b** *V*~T~) in the tumor and contralateral tumor-free brain area (based on PMOD analysis). For p03, a tumor area with enhanced radioactivity uptake as compared with the rest of the tumor was visible (tumor PET enhanced) Tissue levels of ABCB1 and ABCG2 {#Sec14} -------------------------------- In four subjects (p01, p02, p03 and p05), surgically resected tumor tissue was analyzed for ABCB1 and ABCG2 levels with QTAP (Table [2](#Tab2){ref-type="table"}). For the other two patients, not enough material was available for this analysis. As surgical specimens were only available as frozen tissue samples, the isolation of brain capillary micro-vessels was not feasible, so that the measured ABCB1 and ABCG2 levels represent an average value of membrane fractions of all cell types present in the sample. Mean ABCB1 and ABCG2 levels were 0.8 ± 0.1 and 1.2 ± 0.3 fmol/μg protein, respectively. Table 2Absolute ABCB1, ABCG2 and ATPase levels in plasma membrane fractions of surgically removed tumor tissue determined with quantitatively targeted absolute proteomicsTotal protein used for digestion (μg) (***n***)ABCB1\ (fmol/μg) (CV%)ABCG2\ (fmol/μg) (CV%)ATPase\ (fmol/μg) (CV%)p0150 (1)0.97 (\<5%)1.53 (\<5%)178.3 (\<5%)p0250 (4)0.79 (2.7 %)0.85 (4.4 %)257.1 (0.6%)p03a25 (1)0.69 (\<5%)0.94 (\<5%)60.1 (\<5%)p03a50 (3)0.94 (6.4 %)1.4 (8.2 %)164.6 (7.1 %)p0540 (1)0.64 (\<5%)1.10 (\<5%)83.3 (\<5%)For lower limits of quantification, see \[[@CR27]\]%CV, % coefficient of variation: *n* = digestion replicate, each digested sample was injected three times. Quantification CV% \<5% is the variation between transitions and results from three injections (for the three samples digested once and injected three times)^a^Two different samples collected during surgery Discussion {#Sec15} ========== In this exploratory study, we used PET imaging to assess regional brain delivery of \[^11^C\]tariquidar as a small-molecule model ABCB1/ABCG2 substrate in patients with non-contrast-enhancing brain tumors. The main finding of our study was that brain delivery of \[^11^C\]tariquidar was comparably low in tumor and tumor-free brain tissue, which suggested that ABCB1/ABCG2 transport activity was sufficiently intact in tumor tissue to restrict brain entry of anticancer drugs which are dual ABCB1/ABCG2 substrates. The blood--brain tumor barrier (BBTB) is formed by the capillaries supplying brain tumors. Depending on the tumor type and size, the BBTB may substantially differ from the BBB \[[@CR30]\]. While low-grade brain tumors possess continuous non-fenestrated capillaries, which resemble normal brain capillaries, high-grade brain tumors often possess leaky, fenestrated vessels \[[@CR31]\], as manifested by increased permeability to MRI contrast agents \[[@CR32]\]. There is evidence that efflux transporters localized in the BBB can also be found in endothelial cells forming the BBTB \[[@CR33], [@CR34]\]. MRI contrast agents are hydrophilic, gadolinium-containing complexes which are believed to cross the BBTB via the paracellular route through fenestrated capillaries. On the other hand, molecularly targeted anticancer drugs are small, lipophilic molecules which mainly cross the BBB via the transcellular route and which are, in most cases, subject to efflux transport by ABCB1/ABCG2. Even in the presence of a disrupted BBTB with fenestrated capillaries, endothelial cells may have sufficient ABCB1/ABCG2 transport capacity to limit tumor distribution of such drugs \[[@CR34], [@CR35]\]. A limited number of previous studies have determined brain tumor concentrations of anticancer drugs \[[@CR36]--[@CR40]\]. Three studies assessed intratumoral concentrations of protein kinase inhibitors (gefitinib, imatinib and lapatinib) in surgically resected tumor tissue of glioblastoma patients and found very variable tumor concentrations of these agents, which in part exceeded the corresponding plasma concentrations, which pointed to an increased permeability relative to normal brain \[[@CR36], [@CR39], [@CR40]\]. Blakely et al used intraoperative microdialysis to measure the intratumoral pharmacokinetics of methotrexate in patients with recurrent gliomas and found considerably higher drug concentrations in contrast-enhancing regions of the tumor as compared with non-enhancing tissue \[[@CR38]\]. Finally, Brown et al. performed PET with the radiolabeled focal adhesion kinase inhibitor \[^11^C\]GSK2256098 in eight patients with recurrent glioblastoma \[[@CR37]\]. Brain uptake (*V*~T~) of \[^11^C\]GSK2256098 was found to be very low, but approximately two times higher in tumor tissue as compared with surrounding T2 enhancing areas and normal brain. All these data supported a focal disruption of the BBTB in high-grade gliomas, which led to enhanced brain distribution of small-molecule drugs, which were subject to ABCB1 and/or ABCG2 efflux transport. As opposed to these previous studies, we examined in the present study patients with non-contrast enhancing, low- to high-grade brain tumors and used a radiolabeled model ABCB1/ABCG2 substrate instead of a drug which is used for treatment of tumors. Previous experiments showed that \[^11^C\]tariquidar had very low brain uptake in wild-type, *Abcb1a/b*^*(-/-)*^ and *Abcg2*^*(-/-)*^ mice, but approximately sixfold higher brain uptake in triple knockout *Abcb1a/b*^*(-/-)*^*Abcg2*^*(-/-)*^ mice \[[@CR17]\], which was in line with the typical behavior of a dual ABCB1/ABCG2 substrate \[[@CR7]\]. In healthy human volunteers, brain uptake of \[^11^C\]tariquidar was very low, but significantly increased in carriers of the *ABCG2* single-nucleotide polymorphism c.421C\>A following ABCB1 inhibition \[[@CR19]\]. These data suggested that brain distribution of \[^11^C\]tariquidar is dependent on ABCB1/ABCG2 transport activity in rodents and humans. This is also true for the majority of currently known molecularly targeted kinase inhibitors, which have been proposed for the treatment of brain tumors \[[@CR20], [@CR41]\]. In fact, \[^11^C\]tariquidar behaved very similar in rodents and humans in terms of its brain distribution as \[^11^C\]erlotinib \[[@CR42], [@CR43]\], a radiolabeled epidermal growth factor receptor (EGFR) targeted tyrosine kinase inhibitor, which is also a dual ABCB1/ABCG2 substrate and which failed in a phase II trial in patients with glioblastoma multiforme \[[@CR44]\]. In our study, brain distribution of \[^11^C\]tariquidar was found to be very low throughout the brain including tumor tissue, except for one patient with a grade III brain tumor, who had undergone previous surgery and in whom a small area of increased radiotracer uptake was observed within the tumor (Fig. [1](#Fig1){ref-type="fig"}). This suggested that the investigated tumors received their blood supply through intact capillaries, which efficiently restrict brain distribution of small-molecule ABCB1/ABCG2 substrates. The standard treatment for malignant brain tumors is temozolomide. While temozolomide is believed to penetrate the BBB relatively well, a recent study has shown that brain entry of temozolomide is increased in the absence of ABCB1 and ABCG2 activity in mice which translated into an improved antitumor response in experimental intracranial tumor models \[[@CR45]\]. It can, therefore, be expected that brain delivery of temozolomide in tumor patients is also restricted, at least to some extent, by ABCB1/ABCG2 activity. In comparison to our previous study in healthy volunteers \[[@CR19]\], mean *V*~T~ of \[^11^C\]tariquidar in normal brain tissue was approximately threefold lower (0.16 ± 0.09 in this study versus 0.43 ± 0.10 in healthy volunteers). The percentage of unmetabolized \[^11^C\]tariquidar in plasma at the end of the PET scan was comparable in tumor patients and in healthy volunteers (91.1 ± 3.1% in tumor patients versus 89.7 ± 3.6% in healthy volunteers), which rules out an effect of the concomitant medication taken by the tumor patients (Additional file 2: Table S2) on radiotracer metabolism as an explanation for the observed differences in radiotracer brain distribution. One possible explanation for the observed differences may be differences in the applied VOIs. In healthy volunteers, whole brain gray matter was analyzed \[[@CR19]\], while in tumor patients, the position of the applied (contralateral) VOI for normal brain tissue depended on the localization of the brain tumor and contained both gray and white matter. This assumption is supported by additional analysis of the data in brain tumor patients with the same methodology as employed in reference \[[@CR19]\], which provided a normal brain gray matter *V*~T~ value of 0.38 ± 0.26 in patients. It is noteworthy that distribution of \[^11^C\]tariquidar to normal brain tissue displayed a considerably higher inter-individual variability in tumor patients as compared with healthy volunteers. Tariquidar is a third-generation ABCB1 inhibitor which has undergone clinical development as a multidrug resistance reversal agent in patients with systemic tumors \[[@CR18]\]. However, its clinical development has been stopped due to lack of efficacy in tumor patients. In the past decade, tariquidar has been investigated as a potential inhibitor of ABCB1-mediated efflux transport at the human BBB. PET imaging studies in healthy volunteers revealed up to fivefold increases in brain distribution of the radiolabeled model ABCB1 substrates (*R*)-\[^11^C\]verapamil and \[^11^C\]*N*-desmethyl-loperamide following tariquidar administration \[[@CR46], [@CR47]\]. Pharmacological inhibition of efflux transporters at the BBB has also been proposed for a more effective treatment of brain tumors with anticancer drugs, for which brain distribution is limited by ABCB1/ABCG2-mediated efflux transport \[[@CR14], [@CR30]\]. In this context, it has been suggested that ABCB1 inhibition may additionally improve access of anticancer drugs to tumor cells which overexpress ABCB1 in their cell membranes \[[@CR13], [@CR14]\]. However, to achieve effective ABCB1 inhibition in brain tumor cells, an ABCB1 inhibitor would first need to cross the BBTB. The data presented in this work show that tariquidar very poorly penetrates the BBTB and may, therefore, not be effective to overcome ABCB1-mediated multidrug resistance of brain tumors. To examine ABCB1 and ABCG2 levels, we performed QTAP on surgically resected brain tissue samples of four patients included in this study (Table [2](#Tab2){ref-type="table"}). QTAP allows for obtaining absolute levels of proteins in the brain and has been applied to measure ABCB1 and ABCG2 levels in human brain micro-vessels \[[@CR48]\]. In our study, micro-vessels could not be isolated; therefore, the measured ABCB1 and ABCG2 levels represented membrane-bound transporters of all cell types present in the sample (e.g., micro-vessels, glia cells, neurons and tumor cells). Accordingly, mean ABCB1 levels (0.8 ± 0.1 fmol/μg protein) and ABCG2 levels (1.2 ± 0.3 fmol/μg protein) were approximately seven times lower than those previously reported in isolated human brain capillary micro-vessels (ABCB1: 6.1 ± 1.7 fmol/μg protein, ABCG2: 8.1 ± 2.3 fmol/μg protein) \[[@CR48]\]. However, ABCG2/ABCB1 ratios in our samples were comparable (1.4 ± 0.2, range: 1.1-1.7) to previously reported values from isolated human brain micro-vessels (1.3) \[[@CR48]\], which suggested that no major ABCB1 or ABCG2 overexpression occurred in the investigated tumors. Conclusion {#Sec16} ========== We found very low brain delivery of the model ABCB1/ABCG2 substrate \[^11^C\]tariquidar in patients with non-contrast-enhancing brain tumors without significant differences between tumor and tumor-free brain tissue. This supports the presence of an intact BBTB, which is impermeable to small-molecule ABCB1/ABCG2 substrates. This potentially applies to a range of small-molecule kinase inhibitors, which are dual ABCB1/ABCG2 substrates and which are discussed as potential treatment for brain tumors. The best strategy for an effective treatment of brain tumors may thus be the development of drugs with good passive permeability which are not subject to ABCB1/ABCG2-mediated efflux transport at the BBTB \[[@CR41], [@CR49], [@CR50]\]. Supplementary information ========================= {#Sec17} **Additional file 1: Table S1.** Volumes of interest (cm^3^) for analyzed brain tissue. **Table S2.** List of continuous medication at the time of the PET scan. ABCB1 : ABC subfamily B member 1 also known as P-glycoprotein ABCG2 : ABC subfamily G member 2 also known as breast cancer resistance protein AUC : Area under the curve AUCR : AUC ratio BBB : Blood--brain barrier BBTB : Blood--brain tumor barrier MRI : magnetic resonance imaging PET : Positron emission tomography QTAP : Quantitative targeted absolute proteomics SUV : Standardized uptake value TAC : Time--activity curve *V*~T~ : Total distribution volume VOI : Volume of interest WHO : World Health Organization **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= **Supplementary information** accompanies this paper at 10.1186/s13550-019-0581-y. The authors wish to thank their colleagues from the Department of Anesthesiology (Medical University of Vienna) for arterial cannulation and Johann Stanek for assistance during the study days. Furthermore, Meryam Taghi and Lucien Tchara are acknowledged for tumor sample treatment and digestion, and Cerina Chhuon and Chiara Guerrera for proteomic analysis (Université Paris Descartes). BW, MB, TC, MP, MH, MZ, MM and OL were responsible for the study design. BW, HS and TC recruited the patients. BW, MB, CP, MW and OL conducted the study. CP performed the radiotracer synthesis. BW, MB, RK, M-CM, XD, JAH and OL analyzed the data. JAH performed histopathological analysis. M-CM and XD performed quantitative targeted absolute proteomics. OL and BW wrote the manuscript. All authors read and approved the final manuscript. This study was funded by the Austrian Science Fund (FWF) (KLI 139-B00, to M. Müller). The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee (Ethics Committee of the Medical University of Vienna, reference number: 1031/2011) and with the 1964 Helsinki declaration and its amendments or comparable ethical standards. Informed consent was obtained from all individual participants included in the study. Not applicable. Dr. Preusser reports personal fees from Bayer, Bristol-Myers Squibb, Novartis, Gerson Lehrman Group (GLG), CMC Contrast, GlaxoSmithKline, Mundipharma, Roche, BMJ Journals, MedMedia, Astra Zeneca, AbbVie, Lilly, Medahead, Daiichi Sankyo, Sanofi, Merck Sharp & Dome., grants from Böhringer-Ingelheim, Bristol-Myers Squibb, Roche, Daiichi Sankyo, Merck Sharp & Dome, Novocure, GlaxoSmithKline, AbbVie., outside the submitted work. Dr. Müller reports grants from Austrian Science Fund during the conduct of the study. The other authors declare that they have no competing interests.

1. Introduction {#sec1-jcm-08-00208} =============== More than 1.8 million people in England are known to have chronic kidney disease (CKD), while another million of the population are thought to have the condition but are yet to be diagnosed \[[@B1-jcm-08-00208],[@B2-jcm-08-00208]\]. CKD can significantly reduce quality and quantity of life and may lead to the need for renal replacement therapy. Around 45,000 premature deaths are attributed to CKD annually in the United Kingdom \[[@B3-jcm-08-00208]\]. Furthermore, the management of CKD places a substantial economic burden on healthcare systems across the globe, and healthcare expenditure has been noted to be inversely proportional to renal function \[[@B4-jcm-08-00208]\]. The National Health Service (NHS) in England spent approximately £1.45 billion on CKD in 2009--2010 \[[@B1-jcm-08-00208]\]. Given the impact of CKD on quality of life and healthcare expenditure, interventions that may slow the progression of CKD are of great importance. Metabolic acidosis is frequently found in CKD patients, and becomes more common with worsening renal function \[[@B4-jcm-08-00208]\]. Metabolic acidosis has been operationalised as serum bicarbonate levels that are consistently below 22 mmol/L \[[@B4-jcm-08-00208],[@B5-jcm-08-00208],[@B6-jcm-08-00208]\]. It is associated with other pathophysiological changes including systemic inflammation, upregulation of parathyroid hormone receptors in osteoblasts and increased activity of osteoclasts; which leads to accelerated bone resorption and osteopenia, reduction of Na^+^-K^+^-ATPase activity in erythrocytes---resulting in impairment of myocardial contractility and heart failure, and reduced respiratory reserve and exhaustion of body buffer systems, which increases severity of acute intercurrent illnesses \[[@B4-jcm-08-00208],[@B5-jcm-08-00208],[@B6-jcm-08-00208],[@B7-jcm-08-00208],[@B8-jcm-08-00208],[@B9-jcm-08-00208],[@B10-jcm-08-00208]\]. The degree of metabolic acidosis is directly associated with glomerular filtration rate GFR, and is due to the failure of the kidneys to produce ammonia, regenerate bicarbonate and excrete hydrogen ions \[[@B11-jcm-08-00208]\]. Importantly, metabolic acidosis is associated with an accelerated decline in renal function, an association that may be causal in nature \[[@B12-jcm-08-00208]\]. As part of CKD patients' care, oral alkali such as bicarbonates and citrates, base-producing fruits and vegetables or low protein diets are commonly prescribed to avoid or correct metabolic acidosis. Oral bicarbonate therapy has been used to counteract metabolic acidosis in CKD for decades. However, until recently, there have been very few intervention studies testing the effectiveness of bicarbonate therapy at improving metabolic acidosis, or its consequences in patients with CKD, and there are no systematic reviews evaluating the effectiveness and safety of this approach. This is reflected by the current guidelines, which are still unable to make an evidence-based recommendation regarding the correction of mild-to-moderate acidosis in CKD \[[@B13-jcm-08-00208]\]. To date, there is also no consensus on the optimal range for serum bicarbonate in CKD patients. The paucity of clear guidelines results in variability in clinical practice when prescribing oral bicarbonate therapy for patients with CKD and metabolic acidosis \[[@B13-jcm-08-00208]\]. In this systematic review and meta-analysis, we aimed to examine the outcomes of all published randomised controlled trials (RCTs) that investigated the effect of oral bicarbonate therapy in non-haemodialysis dependent adults with CKD. 2. Methods {#sec2-jcm-08-00208} ========== 2.1. Literature Search Strategy {#sec2dot1-jcm-08-00208} ------------------------------- The review protocol was registered in the PROSPERO database (Ref: CRD42018112908) \[[@B14-jcm-08-00208]\]. Three databases (Ovid MEDLINE^®^, EMBASE^®^ and Cochrane Library) were searched in mid-October 2018 for English literature, with no restrictions applied to the publication status or date. Grey literature was not searched. The full search strategy is presented in [Appendix A](#app2-jcm-08-00208){ref-type="app"}. 2.2. Inclusion and Exclusion Criteria {#sec2dot2-jcm-08-00208} ------------------------------------- Our inclusion criteria were randomised controlled trials, availability of full text in English language, all aetiologies and severities of CKD, comparison of bicarbonate therapy with placebo or standard care, and any health-related outcome measures. Our exclusion criteria were children under the age of 18, people receiving haemodialysis and the comparison of bicarbonate therapy with active treatment. 2.3. Study Selection {#sec2dot3-jcm-08-00208} -------------------- Two reviewers independently screened all titles and abstracts. The bibliographies of selected articles were hand-searched for any other potentially relevant studies. Any uncertainties about study eligibility were discussed between reviewers, and if necessary, with a third reviewer. 2.4. Data Extraction {#sec2dot4-jcm-08-00208} -------------------- A data extraction form was designed by adapting and customising the Cochrane collaboration's data collection form for intervention review---RCTs and non-RCTs \[[@B15-jcm-08-00208]\]. Two independent assessors performed the data extraction. Data discrepancies were resolved by discussion and consensus between the two assessors. 2.5. Risk of Bias {#sec2dot5-jcm-08-00208} ----------------- Risk of bias was evaluated using the Cochrane collaboration's risk of bias tool \[[@B15-jcm-08-00208]\]. Criteria assessed included random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting and other biases. Risk was reported as low, high or unclear. Results from the assessment were subsequently tabulated using RevMan 5.3 (The Cochrane Collaboration, Copenhagen, Denmark) to generate a risk of bias graphic and summary table \[[@B16-jcm-08-00208]\]. 2.6. Outcome Measures and Data Synthesis {#sec2dot6-jcm-08-00208} ---------------------------------------- The primary outcome measure of interest was the rate of change in estimated glomerular filtration rate (eGFR). Secondary outcome measures were eGFR at the end of follow-up, mortality, blood pressure, number of patients proceeding to renal replacement therapy and quality of life. Meta-analyses of these and any other health-related outcome measures were performed when there were at least three trials reporting the same outcome measure. The random effects model was used for continuous data and forest plots generated using RevMan 5.3. Statistical heterogeneity in treatment effects was determined using the *I*^2^ test. 2.7. Subgroup Analysis {#sec2dot7-jcm-08-00208} ---------------------- Subgroup analyses were planned by mean study age, mean proportion of men and the duration of the study. 3. Results {#sec3-jcm-08-00208} ========== 3.1. Study Selection {#sec3dot1-jcm-08-00208} -------------------- After deduplication, the search identified 307 potentially relevant articles, and 32 trials were shortlisted after reading titles and abstracts (see [Figure 1](#jcm-08-00208-f001){ref-type="fig"}). Seven trials that recruited 815 participants met the inclusion criteria after a full text review (see [Table 1](#jcm-08-00208-t001){ref-type="table"}). Two trials were set in the USA and India, respectively, with one trial each from Italy, South Korea and the UK. 3.2. Risk of Bias Analysis {#sec3dot2-jcm-08-00208} -------------------------- Three RCTs described utilisation of adequate randomisation processes \[[@B18-jcm-08-00208],[@B21-jcm-08-00208],[@B22-jcm-08-00208]\] to minimise selection bias, while one study used patient's identifying number \[[@B19-jcm-08-00208]\], which was likely to have increased the risk of bias (see [Figure 2](#jcm-08-00208-f002){ref-type="fig"} and [Figure 3](#jcm-08-00208-f003){ref-type="fig"}). Two RCTs applied allocation concealment by employing a central randomisation process \[[@B22-jcm-08-00208]\] or using opaque sequenced envelopes \[[@B23-jcm-08-00208]\], but there was no mention of allocation concealment in the other studies. All studies had a high risk of performance bias due to the nature of the intervention, in which oral bicarbonate therapy was titrated and monitored to achieve desired serum bicarbonate levels. Two studies used placebo as a comparator \[[@B17-jcm-08-00208],[@B19-jcm-08-00208]\], which might potentially reduce the risk of performance bias, but did not completely abolish it. Two RCTs reported blinding of their outcome assessors \[[@B18-jcm-08-00208],[@B23-jcm-08-00208]\] while it was unclear in the rest if this risk of bias was minimised or eliminated. All studies were deemed to have low attrition bias as the dropout rate was less than 10% and accounted for in each study. The missing outcomes were also thought to be insufficient to have a clinically relevant impact on the observed effect size. Two studies recorded all expected outcomes using measurements and analysis methods that were pre-specified \[[@B19-jcm-08-00208],[@B21-jcm-08-00208]\], and therefore had a low reporting bias. Two RCTs were judged to have a high risk of bias because outcome measures were detailed with an unconventional outcome measure (decline in creatinine clearance) \[[@B18-jcm-08-00208]\], and measurements (mean and 95% confidence interval (CI)) \[[@B23-jcm-08-00208]\]. The remaining studies did not offer sufficient information to permit judgement about this criterion. 3.3. Outcomes {#sec3dot3-jcm-08-00208} ------------- ### 3.3.1. Serum Bicarbonate {#sec3dot3dot1-jcm-08-00208} All seven RCTs studied serum bicarbonate levels following the randomisation of study participants (see [Figure 4](#jcm-08-00208-f004){ref-type="fig"}) \[[@B17-jcm-08-00208],[@B18-jcm-08-00208],[@B19-jcm-08-00208],[@B20-jcm-08-00208],[@B21-jcm-08-00208],[@B22-jcm-08-00208],[@B23-jcm-08-00208]\]. Data from 707 patients were analysed; 57.9% were males and their mean age ranged from 37.5 ± 17 years to 65.5 ± 11.4 years. Serum bicarbonate levels were higher after oral bicarbonate therapy (mean difference 3.4 mmol/L; 95% CI 1.9--4.9) but the results were highly heterogeneous (*I*^2^ = 97%). ### 3.3.2. eGFR and Serum Creatinine {#sec3dot3dot2-jcm-08-00208} Six RCTs \[[@B18-jcm-08-00208],[@B19-jcm-08-00208],[@B20-jcm-08-00208],[@B21-jcm-08-00208],[@B22-jcm-08-00208],[@B23-jcm-08-00208]\] investigated eGFR after randomising patients to oral bicarbonate therapy or placebo (see [Figure 5](#jcm-08-00208-f005){ref-type="fig"}). Data were not presented in a format that allowed the rate of change of eGFR (the primary outcome measure) to be analysed, so eGFR at the end of follow-up was used. A total of 667 patients were analysed; 57.7% were males and their mean age ranged from 50.1 ± 11.6 years to 65.5 ± 11.4 years. eGFR favoured bicarbonate therapy (mean difference 3.1 mL/min per 1.73 m^2^; 95% CI 1.3--4.9) but the analysis revealed moderately high heterogeneity (*I*^2^ = 68%). Four studies measured serum creatinine at the end of their follow-up period (see [Supplementary Figure S1](#app1-jcm-08-00208){ref-type="app"}) \[[@B17-jcm-08-00208],[@B19-jcm-08-00208],[@B20-jcm-08-00208],[@B23-jcm-08-00208]\]. Data from 361 patients were included; 61% were males and their mean age ranged from 37.5 ± 17 years to 55.8 ± 12.7 years. Compared with placebo and standard care for CKD, oral bicarbonate supplementation had non-significant effects on serum creatinine (*p* = 0.09). ### 3.3.3. Systolic Blood Pressure {#sec3dot3dot3-jcm-08-00208} Six studies recorded systolic blood pressure (BP) as an outcome measure (see [Figure 6](#jcm-08-00208-f006){ref-type="fig"}) \[[@B17-jcm-08-00208],[@B18-jcm-08-00208],[@B19-jcm-08-00208],[@B21-jcm-08-00208],[@B22-jcm-08-00208],[@B23-jcm-08-00208]\]. A total of 635 patients were included; 57.3% were males and their mean age ranged from 37.5 ± 17 years to 65.5 ± 11.4 years. Oral bicarbonate therapy had uncertain effects on systolic BP when compared to placebo or conventional treatment for CKD (*p* = 0.19). ### 3.3.4. Weight {#sec3dot3dot4-jcm-08-00208} Five RCTs reported the weight of their patients at the conclusion of their studies (see [Figure 7](#jcm-08-00208-f007){ref-type="fig"}) \[[@B17-jcm-08-00208],[@B20-jcm-08-00208],[@B21-jcm-08-00208],[@B22-jcm-08-00208],[@B23-jcm-08-00208]\]. 507 patients were analysed; 69.3% were males and their mean age ranged from 37.5 ± 17 years to 65.5 ± 11.4 years. The effects of oral bicarbonate therapy on weight of CKD patients are uncertain (*p* = 0.3) and the results are highly heterogeneous (*I*^2^ = 87%). ### 3.3.5. Other Outcomes {#sec3dot3dot5-jcm-08-00208} The rate of change of eGFR was intended as a primary outcome, but data from multiple trials were not presented in a format that enabled this parameter to be analysed, so eGFR at the end of follow-up was used. Other listed outcomes that were not available included mortality rate, number of patients proceeding to renal replacement therapy and quality of life. 3.4. Subgroup Analyses {#sec3dot4-jcm-08-00208} ---------------------- Subgroup analyses were performed by the duration of the study, thereby eliminating one source of heterogeneity. These analyses were only possible for eGFR and serum bicarbonate at one year follow-up (see [Supplementary Figures S2 and S3](#app1-jcm-08-00208){ref-type="app"}), as there were too few studies or insufficient data in other outcome measures to allow for meaningful analysis. Three studies investigated eGFR and serum bicarbonate levels at one year after the randomisation of patients to oral bicarbonate therapy or placebo treatment \[[@B18-jcm-08-00208],[@B20-jcm-08-00208],[@B22-jcm-08-00208]\]. Of the 347 patients that were analysed; 55.4% were males and their mean age ranged from 53.3 ± 13.5 years to 65.5 ± 11.4 years. The effects of bicarbonate therapy on eGFR were non-significant when the duration of the study was standardised at one year (*p* = 0.19). The heterogeneity of treatment results was lower than other analyses (*I*^2^ = 46%). Serum bicarbonate at one year was higher after oral bicarbonate therapy (mean difference 3.2 mmol/L, 95% CI 2.0--4.3), but there was still significant heterogeneity (*I*^2^ = 66%). 4. Discussion {#sec4-jcm-08-00208} ============= 4.1. Outcomes {#sec4dot1-jcm-08-00208} ------------- Oral bicarbonate supplementation resulted in a slightly higher eGFR at the end of follow up (three months to five years) compared to those given placebo or conventional CKD treatment. Bicarbonate therapy also improved serum bicarbonate levels by an average of 3.2 mmol/L compared to the control arm, but we did not find any significant treatment effects in other parameters such as systolic BP and weight. These findings should be interpreted with caution due to a high level of heterogeneity between studies. The heterogeneity probably reflects the marked differences across trials, ranging from population demographics to dosing regimen and length of follow-up. Furthermore, the mean age of trial participants was remarkably low despite the increased prevalence of CKD in older people \[[@B1-jcm-08-00208],[@B2-jcm-08-00208]\]. All but one trial recruited patients with a mean age below 56 years. Although most epidemiological studies observed that CKD was more common in women, all seven RCTs in this review enrolled more men than women \[[@B24-jcm-08-00208]\]. This suggests that there may be a degree of selection bias in trial recruitment that may limit the generalisation of results in clinical practice. Some trials recruited atypical populations, e.g., those with stage 2 CKD with albuminuria due to hypertension \[[@B19-jcm-08-00208]\] or those with CKD of unknown aetiology in India \[[@B17-jcm-08-00208]\]. It is not clear that the observed treatment effects would be seen in more typical patients with CKD. Patients' dietary intakes were not clearly accounted for, and this might have influenced the serum bicarbonate levels, as acid- and base-producing diets are known to skew bicarbonate levels. Patients also had various co-morbidities such as cardiovascular diseases and type 2 diabetes mellitus, but this confounding factor might be difficult to avoid when recruiting CKD patients, as most patients have at least one, if not more, co-morbidities. Almost all trials included in this analysis were open-label, and it is known that unmasked trial designs tend to overestimate the treatment effect size compared to placebo. Several trials adopted a 'treat to target' approach for bicarbonate, in which doses were escalated in the intervention arm to try and reach a pre-defined bicarbonate level. Such approaches again tend to magnify treatment effects compared to the comparison of bicarbonate at a fixed dose with controls, although it can be argued that such an approach better approximates clinical practice. Another likely confounding factor is the duration of study. When limited to studies reporting outcomes at one year of follow-up, the positive effect of oral bicarbonate therapy on eGFR was attenuated and eGFR did not differ significantly between those on bicarbonate supplementation and placebo. All trials also recruited patients with a wide range of baseline bicarbonate levels (16 to 24 mmol/L), and this may have had an influence on the observed response to treatment, such that studies that began with lower baseline bicarbonate levels might have witnessed a larger treatment effect size. Our results revealed that a modest total of 815 participants worldwide had been recruited into seven RCTs of bicarbonate therapy. It is highly likely that future studies will influence the results of the meta-analysis. The authors are aware of four further relevant RCTs that have published their protocols but are yet to report their results---BiCARB \[[@B13-jcm-08-00208]\], UBI \[[@B25-jcm-08-00208]\], SoBic \[[@B26-jcm-08-00208]\] and BASE \[[@B27-jcm-08-00208]\]. Their planned recruitment figures should more than double the number of participants in trials of bicarbonate treatment in CKD to date, as well as address current weaknesses in the evidence base, such as a lack of older people and data on physical function and quality of life. 4.2. Potential Adverse Effects of Bicarbonate Oral Therapy {#sec4dot2-jcm-08-00208} ---------------------------------------------------------- As with most medications, oral bicarbonate therapy comes with its own adverse effects and cautions. Sodium bicarbonate tablets are awkward to take, especially for older people with impaired swallowing and polypharmacy; the tablets are large and several tablets are usually required to be taken at one time or in a day \[[@B13-jcm-08-00208]\]. The British National Formulary (BNF) listed abdominal discomfort and bloating as recognised side effects, which may lead to compliance issues in patients \[[@B13-jcm-08-00208]\]. 600 mg sodium bicarbonate tablets taken three times daily cost the NHS £190 per patient per year, despite a lack of robust evidence supporting the efficacy of this treatment. Sodium bicarbonate tablets also contain 6 mmol of sodium in every 500 mg, which could contribute to hypertension and fluid overload if not monitored closely, especially in a population with impaired renal clearance \[[@B24-jcm-08-00208]\]. This finding is corroborated by observational studies, which suggested that elevated serum bicarbonate levels were associated with an increased risk of heart failure in CKD patients \[[@B28-jcm-08-00208]\]. It is therefore reassuring that this meta-analysis did not demonstrate any worsening of systolic blood pressure with bicarbonate therapy. The targeted serum bicarbonate level, dose and time of initiation are yet to be determined. Despite new evidence suggesting the range of potential benefits, overtreatment with oral alkali therapy may result in metabolic alkalosis, which is also associated with poor outcomes in patients with CKD \[[@B11-jcm-08-00208]\]. An alkaline pH has also been shown to augment vascular calcification in animal models \[[@B11-jcm-08-00208]\]. As arterial elasticity declines with age and older people are more likely to have CKD, caution has to be exercised when considering oral bicarbonate therapy. 4.3. Strengths and Limitations {#sec4dot3-jcm-08-00208} ------------------------------ To our knowledge, this is the first synthesis of trials investigating the effects of bicarbonate therapy. All fully published trials were included and we undertook meta-analysis where possible. However, there was substantial heterogeneity in all included studies. The seven RCTs had important differences in various parameters including population demographics, intervention regimes, outcome measures and duration of study. Additionally, studies were often at high risk of bias but poorer quality studies could not be excluded due to the limited number of published RCTs that suited our inclusion criteria. 5. Conclusions {#sec5-jcm-08-00208} ============== Bicarbonate supplementation modestly improved eGFR and serum bicarbonate levels compared to placebo or conventional CKD management. Evidence of improvement in other health-related outcome measures was lacking. These findings should be interpreted with caution due to high heterogeneity and risk of bias in studies. Further trial evidence is needed to establish the net overall benefit or harm of oral bicarbonate therapy in CKD, and to define the target groups most likely to benefit from treatment. The following are available online at <http://www.mdpi.com/2077-0383/8/2/208/s1>, Supplementary Figure S1. Forest plot comparing the effects of oral bicarbonate therapy and control on serum creatinine levels. Supplementary Figure S2. Forest plot comparing the effects of oral bicarbonate therapy and control on eGFR at one year. Supplementary Figure S3. Forest plot comparing the effects of oral bicarbonate therapy and control on serum bicarbonate levels at one year. ###### Click here for additional data file. M.K.H., M.D.W. and R.L.S. designed the study; M.K.H. and R.L.S. carried out literature search; M.K.H. and R.L.S. collected and analysed the data; M.K.H., M.D.W. and R.L.S. made the figures; M.K.H., M.D.W. and R.L.S. drafted and revised the paper; all authors approved the final version of the manuscript. This work was supported by a National Institute of Health Research Health Technology Assessment (NIHR HTA) project grant (Ref: 10/71/10). The authors declare no conflict of interest. jcm-08-00208-t0A1_Table A1 ###### Search strategy in MEDLINE. --- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------- 1 exp Bicarbonates \[MeSH\] or exp Sodium Bicarbonate \[MeSH\] or bicarbonates \[tiab\] 24,367 2 Chronic Kidney Disease or CKD \[mp\] or exp Renal Insufficiency, Chronic \[MESH\] or or Chronic Renal Insufficiency \[tiab\] 127,737 3 (exp Bicarbonates \[MeSH\] or exp Sodium Bicarbonate \[MESH\] or bicarbonates \[tiab\]) and (Chronic Kidney Disease or CKD \[mp\] or exp Renal Insufficiency, Chronic \[MESH\] or or Chronic Renal Insufficiency \[tiab\]) 812 4 Limit 3 to RCT and English language 76 --- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------- exp: Explode; MeSH: Medical Subject Headings; tiab: Title or abstract; mp: Keyword; RCT: Randomised controlled trials. jcm-08-00208-t0A2_Table A2 ###### Search strategy in EMBASE. --- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ --------- 1 exp Bicarbonate \[MeSH\] or Bicarbonate.mp 69,535 2 exp Chronic Kidney Failure \[MeSH\] or (chronic kidney disease or chronic kidney failure or chronic kidney insufficiency or chronic renal disease or chronic renal failure or chronic renal insufficiency).mp 116,647 3 (exp Bicarbonate \[MeSH\] or Bicarbonate.mp) and (exp Chronic Kidney Failure \[MeSH\] or (chronic kidney disease or chronic kidney failure or chronic kidney insufficiency or chronic renal disease or chronic renal failure or chronic renal insufficiency).mp) 2323 4 Limit 3 to English language and exclude Medline journals and RCT 3 --- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ --------- exp: Explode; MeSH: Medical Subject Headings; mp: Keyword; RCT: Randomised controlled trials. jcm-08-00208-t0A3_Table A3 ###### Search strategy in Cochrane. --- ------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------- 1 MeSH descriptor. \[Bicarbonates\] explode all trees 1201 2 MeSH descriptor: \[Sodium Bicarbonate\] explode all trees 610 3 \#1 or \#2 or (Bicarbonate \*) 3123 4 MeSH descriptor: \[Renal Insufficiency, Chronic\] explode all trees 5705 5 Chronic kidney disease or chronic kidney failure or chronic kidney insufficiency or chronic renal disease or chronic renal failure or chronic renal insufficiency 13,858 6 \#4 or \#5 13,524 7 \#3 and \#6 in Trials 304 --- ------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------- MeSH: Medical Subject Headings; \* (truncation symbol) is used to search for multiple variants of a word all at once. {#jcm-08-00208-f001} {#jcm-08-00208-f002} {#jcm-08-00208-f003} {#jcm-08-00208-f004} {#jcm-08-00208-f005} {#jcm-08-00208-f006} {#jcm-08-00208-f007} jcm-08-00208-t001_Table 1 ###### Characteristics of included randomised controlled trial. Study Location *n* Mean Age (Years) CKD Stage Bicarbonate Level Entry Criterion Intervention Comparator Duration Primary Outcome ------------------------------------------------------ ------------- ----- ------------------ ---------------------------------------------------------------------------- --------------------------------------- --------------------------------------------------------------------------------------------------------------- ------------ ----------- ---------------------------------------------------- Mathur et al., 2006 \[[@B17-jcm-08-00208]\] India 40 41 "Mild to moderate" CKD (creatinine \< 442 μmol/L). CKD stage not specified Not specified Oral bicarbonate 1.2 mEq/kg in 3 divided doses, titrated to maintain serum bicarbonate in range 22--26 mmol/L Placebo 3 months Not specified De Brito-Ashurt et al., 2009 \[[@B18-jcm-08-00208]\] UK 134 55 4 or 5 16 mmol/L \< Bicarbonate \< 19 mmol/L Oral bicarbonate 600 mg 3×/day, increased as needed to maintain serum bicarbonate \> 23 mmol/L Usual care 2 years Decline in creatinine clearance of \>3 mL/min/year Mahajan et al., 2010 \[[@B19-jcm-08-00208]\] USA 120 51 2 with hypertension and microalbuminuria Total CO~2~ \> 24.5 mmol/L Oral bicarbonate 0.5 mEq/kg lean body weight Placebo 5 years eGFR decline rate Jeong et al., 2014 \[[@B20-jcm-08-00208]\] South Korea 80 55 4 or 5 Total CO~2~ \< 22 mmol/L Oral bicarbonate 1 g 3×/day, titrated to maintain serum bicarbonate \> 22 mmol/L Usual care 12 months eGFR Goraya et al., 2014 \[[@B21-jcm-08-00208]\] USA 108 54 3 22 mmol/L \< Total CO~2~ \< 24 mmol/L Oral bicarbonate 0.3 mEq/Kg lean body weight in three divided doses Usual care 3 years eGFR Bellasi et al., 2016 \[[@B22-jcm-08-00208]\] Italy 145 65 3b or 4; in patients with T2DM Bicarbonate \< 24mmol/L Oral bicarbonate 0.5 mEq/kg twice daily, until serum bicarbonate 24--28 mmol/L Usual care 12 months Insulin resistance Dubey et al., 2018 \[[@B23-jcm-08-00208]\] India 188 50 3 and 4 Bicarbonate \< 22mmol/L Oral bicarbonate titrated with weekly monitoring Usual care 6 months Mid-arm muscle circumference CKD: Chronic kidney disease; eGFR: Estimated glomerular filtration rate; CO~2~: Carbon dioxide; T2DM: Type 2 diabetes mellitus.

Protein-protein interaction networks (PINs) have become a key measure of cellular organization[@b1]. Surprisingly, only few networks have been elucidated to date, and most suffer from being incomplete. Although tens of thousands of completely sequenced genomes exist, less than a dozen bacterial interactomes have been tackled[@b2][@b3][@b4][@b5][@b6][@b7]. While thousands of interactions between human host and human virus proteins have been detected over the last decades, protein interaction interfaces between bacteriophages and hosts have been studied in detail only for a few phages such as lambda and T7 in *E. coli*[@b8]. For the first time, we present the interactome of the bacterium *S. pneumoniae* and its interactions with two phages, Dp-1 and Cp-1. Although similar studies have been carried out in human viruses, no such comparisons between bacterial systems have been presented for phages to date. The direct comparison of host-virus interactome data is difficult, since such interactions were determined by different and independent studies by methodologies that detect a different subset of interactions[@b9][@b10][@b11]. To combat this issue we analyzed the interactomes under identical conditions to provide a uniquely standardized network evaluation. We have previously investigated the interactomes of both Cp-1 and Dp-1 without considering interactions with their host[@b12][@b13]. Furthermore, most of Cp-1 and Dp-1's genes remain poorly characterized. To provide a benchmark for *Streptococcus*-phage interactions, we compiled reference interactome datasets from the well-characterized *E. coli* bacteriophages T7 and lambda. Lambda and T7 differ significantly in their host interaction patterns, reflecting their different biology[@b8]. T7 is a lytic phage while lambda is lysogenic. In addition, lambda uses an unusually high number of protein modifications such as proteolytic cleavages. As a consequence, T7 and lambda serve as a model for other phages infecting the same host while using different propagationstrategies. A comparison of the two *E. coli* phages (T7 and lambda) with two phages of *Streptococcus* (Cp-1 and Dp-1) allowed us for the first time to extract general interaction patterns of phages with different bacterial hosts. Notably, we show that these interactions are surprisingly specific for each phage even if we use standardized methods to detect host-virus interactions. Our data demonstrate that each phage has evolved species-specific adaptations that manipulate varied facets of host machinery, reflecting the underlying host-phage coevolution. Results ======= Interactions among *S. pneumoniae* and its phages ------------------------------------------------- Using a yeast-two hybrid approach, we screened a collection of 1,704 prey clones derived from *S. pneumoniae* with all 28 open reading frames of the Cp-1 genomes as baits. Similarly, all 72 deduced proteins of phage Dp-1 were screened against the same *S. pneumoniae* prey collection (see Methods for details). The Cp-1 screens allowed us to find 11 interactions between 7 phage and 10 host proteins. While the significance of these interactions remains unknown, tail protein N was found to interact with oligoendopeptidase F, possibly indicating proteolytic cleavage of N. Uridine kinase (Cpl1) interacts with lysozyme, a critical enzyme for bacterial lysis, prompting us to test whether the kinase can affect lysozyme function or *vice versa*. Despite the fairly strong interaction (as measured by 3-AT titration) we did not detect an effect of uridine kinase on lysozyme activity or *vice versa*. All other Cp-1-host interactions involved phage proteins of unknown function whose biological role remains uncertain. All interactions that involved proteins of Cp-1 are listed in [Table 1](#t1){ref-type="table"}. With 72 ORFs, phage Dp-1 is considerably larger than Cp-1 (28 ORFs). When we screened the Dp-1 ORFeome against our *S. pneumoniae* prey library we found 38 interactions between phage and host proteins ([Table 2](#t2){ref-type="table"}). Notably, we observed that RuvB was strongly targeted by the highest number of phage host proteins, indicating that the phage interferes with DNA repair and recombination functions. Furthermore, this protein also weakly interacts with a hypothetical protein of Cp-1 ([Fig. 1A](#f1){ref-type="fig"}). All interactions were verified using a LuMPIS assay ([Tables 1](#t1){ref-type="table"} and [2](#t2){ref-type="table"}), confirming 35 out of the 38 Dp-1 and 8 out of the 11 Cp-1 interactions when we used a cut-off of ≥3 LIR units even though a total of 12 PPIs were borderline positive at LIR values of ≈3 (see Materials and Methods for details). Note, however, that we used all Y2H interactions for the network analysis described below, given that this cut-off is somewhat arbitrary. Phage-host interactions in bacteria are highly species-specific --------------------------------------------------------------- As a well-investigated benchmark of host-bacteriophage interfaces, we comprehensively surveyed the literature and curated 36 interactions between 16 lambda and 23 *E. coli* proteins in *E. coli*[@b8]. Similarly, we compiled 19 interactions between 8 T7 and 14 *E. coli* proteins[@b8]. Mapping such interactions ([Fig. 1A](#f1){ref-type="fig"}), we observed that the majority of host proteins are targeted by one phage protein, while overlaps of phage-specific sets of targeted proteins are limited in both host organisms. Furthermore, we found that essential proteins appeared enriched in such interaction interfaces. While numerous targeted proteins had orthologs in the opposite organism, *Ssb* was the only evolutionarily conserved protein that was targeted in both host bacteria ([Fig. 1A](#f1){ref-type="fig"}). In [Fig. 1B](#f1){ref-type="fig"} we grouped targeted protein sets according to broad functional classes that were defined by clusters of orthologous groups (COGs)[@b15][@b16]. Determining the occurrence of functions in the host-phage interfaces of both organisms, we found that targeted proteins mostly carried transcription, replication, recombination, and repair functions. To investigate the location of targeted proteins in *E. coli*, we assembled a network of 11,463 interactions between 2,765 proteins[@b2][@b17][@b18]. As for *S. pneumoniae*, we experimentally determined a network of 2,036 interactions between 836 proteins using a two-hybrid approach (see Methods for details). Furthermore, we accounted for 197 interactions that were previously determined by a microfluidic high-throughput assay[@b19]. Specifically, we calculated the enrichment of targeted proteins as a function of their degree ([Fig. 1C](#f1){ref-type="fig"}), suggesting that host proteins with an increased number of interaction partners are prime targets for phages in *E. coli*. We found a similar, yet weaker trend for genes in *S. pneumoniae* that were targeted by Dp-1 as well as Cp-1 ([Fig. 1C](#f1){ref-type="fig"}). Similar observations have been previously reported for human host-viral[@b20][@b21] as well as host-parasite interactions[@b22][@b23]. As a corollary to the observed phages' preference to target central positions in the protein interaction network of *E. coli* and *S. pneumoniae*, we hypothesized that targeted proteins allow the pathogen to reach other proteins efficiently. In particular, we calculated shortest paths from targeted proteins to other proteins in the underlying interaction networks. As for bacteriophages of *E. coli,* we observed that lengths of shortest paths from proteins that are targeted by lambda and T7 respectively were significantly shorter than paths from non-targeted proteins (P \< 10^−9^, Student's t-test) ([Fig. 1D](#f1){ref-type="fig"}). Notably, we found similar results when we considered shortest paths from proteins that were targeted by phages Dp-1 and Cp-1 through an interaction network to other proteins in *S. pneumoniae*. Protein complexes are targeted by phages ---------------------------------------- Protein complexes present another level of cellular organization. To obtain protein clusters in the interaction network of proteins in *S. pneumoniae*, we applied the Markov cluster (MCL) algorithm with varying values of its inflation parameter to modulate the granularity (i.e. size) of clusters. Utilizing COG[@b15][@b16] annotations, we calculated the functional coherence (see Methods) of each cluster. Such a measure tends to decrease with large clusters and *vice versa*. To balance such a trend we calculated the modularity efficiency *E*~*M*~ of a given set of clusters[@b24]. We obtained a maximum value of *E*~*M*~, with inflation parameter of 1.6 in *S. pneumoniae*, providing 148 clusters. As for *E. coli*, we utilized a set of 517 protein complexes from a co-affinity purification study that was followed by mass spectrometry analyses[@b17]. We calculated a complex participation coefficient of each *E. coli* protein that indicates if a protein mainly interacts with proteins in the same or different complexes. In particular, a protein's complex participation coefficient tends toward 1 if it predominantly interacts with proteins in the same complex. In turn, such a value tends to 0 if the given protein mainly interacts with proteins in other complexes. Binning proteins according to their corresponding complex participation coefficient, we calculated the fraction of targeted proteins in each group. As a null model we randomly sampled sets of targeted proteins, allowing us to determine the enrichment of targeted proteins as the ratio of observed and expected fractions of targeted proteins in each bin. [Figure 1E](#f1){ref-type="fig"} clearly indicates that *E. coli* proteins that were targeted by bacteriophage lambda and T7, respectively, were enriched in groups of proteins that reached numerous complexes. Calculating their enrichment in bins of increasing complex participation, we confirmed our initial observation that proteins in *S. pneumoniae* targeted by phages Dp-1 and Cp-1 mostly connect different clusters through their interactions ([Fig. 1E](#f1){ref-type="fig"}). Phage proteins target clusters of host proteins as well as, essential and orthologous proteins in bacteria ---------------------------------------------------------------------------------------------------------- As for further clustering characteristics, we mapped all interactions between *E. coli* proteins that were targeted by bacteriophages lambda or T7 ([Fig. 2A](#f2){ref-type="fig"}). Notably, we found a subnetwork that captured 21 out of 27 (77.8%) lambda targets and 11 out of 16 (68.8%) T7 targets. Qualitatively, such a network appeared to significantly pool essential *E. coli* genes and genes that have orthologs in *S. pneumoniae*. Applying Fisher's exact test, we indeed found that targets of bacteriophages lambda and T7 are significantly enriched with essential (P \< 10^−4^) and orthologous genes (P \< 10^−7^). As for bacteriophages of *S. pneumoniae*, their targets failed to form a large connected component (data not shown) but seem to significantly accumulate proteins that have an ortholog in *E. coli* (P \< 0.05) and essential genes (P \< 0.15). To determine their tendency to cluster in the vicinity of targeted genes of the same bacteriophage, we grouped host proteins that were placed a given distance away from the nearest targeted proteins in the underlying protein-protein interaction network of *E. coli*. In each distance bin, we calculated the enrichment of targeted proteins compared to a null model where we randomly sampled sets of targeted proteins. The inset of [Fig. 2B](#f2){ref-type="fig"} indicates that proteins that were targeted by lambda or T7 are placed in the network neighborhood of other proteins that were targeted by the same phage. Analogously, we determined the enrichment of bacteriophage targets of host proteins in *S. pneumoniae*, showing that Dp-1 and Cp-1 targets failed to cluster in close network vicinity of their corresponding targets. Although overlaps of target sets of organism-specific bacteriophages are limited, we investigated if such clustering characteristics can be extended when we considered the shortest distance to targets of the opposite phage. Surprisingly, the main plot of [Fig. 2B](#f2){ref-type="fig"} suggests that targets of bacteriophage T7 were located in surprisingly close proximity to targets of lambda and *vice versa*, a result that held for targets of Dp-1 and Cp-1 as well (main plot, [Fig. 2B](#f2){ref-type="fig"}). Previous analyses indicated that interactions between essential genes in *E. coli* were mostly organized in a large connected component[@b25]. As for *E. coli*, we found a connected component of 398 essential proteins in the underlying protein interactions network. Randomly sampling sets of essential genes we confirmed this result as statistically significant (P \< 10^−4^). Analogously, we observed a significant giant component that was composed of 78 essential genes in *S. pneumoniae* (P \< 10^−4^), generally suggesting that essential proteins cluster tightly. Determining their enrichment in bins to the nearest targeted proteins in *E. coli*, we indeed found that essential proteins tend to cluster in close proximity to proteins that were targeted by lambda and T7 ([Fig. 2C](#f2){ref-type="fig"}). We obtained similar results when we considered essential proteins in the vicinity of Cp-1 targets in the underlying protein-protein interaction network of *S. pneumoniae*. Utilizing 781 ortholog pairs of proteins in *E. coli* and *S. pneumoniae* we found that such proteins formed a large connected component that was composed of 489 proteins in the interaction network of *E. coli* (P \< 10^−4^). Furthermore, orthologous proteins in *S. pneumoniae* formed a large connected component with 156 proteins as well (P \< 10^−4^). To investigate their clustering tendency, we found that *E. coli* proteins with orthologs in *S. pneumoniae* tend to cluster in the immediate vicinity of proteins targeted by lambda or T7. While we observed similar results when we considered targets of Cp-1 in *S. pneumoniae* ([Fig. 2C](#f2){ref-type="fig"}), targets of Dp-1 failed to show such a trend. Such clustering characteristics strongly suggest that targeted, essential, and orthologous genes in host organisms may form large, tightly connected subnetworks. In [Fig. 2D](#f2){ref-type="fig"}, we mapped all interactions between proteins of *S. pneumoniae* that were targeted by bacteriophages Dp-1 and Cp-1. We also accounted for interactions of essential and orthologous proteins that connected phage targets. Notably, we obtained a network that featured a large connected component with 19 out of 28 (67.8%) Dp-1 targets and 6 out of 10 (60.0%) Cp-1 targets ([Fig. 2D](#f2){ref-type="fig"}). Members of these connected components mediated processes that determined success of viral integration. For example, the *E. coli* subnetwork contains proteases (ClpX/A/B, etc.), endonucleases (HsdR/M/S, RecA, etc.), and transcriptional regulators (RpoA/B/C, IhfA/B, etc.). Such proteins reflect the machinery lambda uses for protein processing and phage assembly as well as gene regulation, given that it is a lysogenic phage. *S. pneumoniae*'s phages did not appear to use similar host activities, as patterns in target function are currently difficult to recognize. The observed tendency of targeted, essential, and orthologous proteins to cluster in close proximity of other phage targets led us to hypothesize that proteins in the immediate vicinity of phage targets may carry global impact. Calculating the betweeness centrality of proteins in the underlying interaction networks of *E. coli* and *S. pneumoniae,* we defined the top 20% most central proteins as "bottleneck" proteins. Focusing on proteins that were targeted by bacteriophages in each host, we observed that such sets of central proteins were enriched with targeted proteins, compared to a null-model where we randomly sampled sets of bottleneck proteins ([Fig. 3A](#f3){ref-type="fig"}). Focusing on the immediate neighbors of targeted proteins we observed that such proteins were enriched in sets of bottleneck proteins as well ([Fig. 3A](#f3){ref-type="fig"}). In turn, we also considered remaining proteins that we found diluted in sets of bottleneck proteins ([Fig. 3A](#f3){ref-type="fig"}). To measure a protein's impact on an interaction network's resilience, we performed a robustness analysis. We sorted all targeted proteins of bacteriophages Dp-1 and Cp-1 according to their degree in the underlying interaction network. Starting with the most connected protein we gradually deleted proteins and calculated the mean path length of the remaining protein interaction network after each deletion step. In comparison, we considered sets of equal size of proteins that interact with targeted proteins. [Figure 3B](#f3){ref-type="fig"} indicates that the successive deletion of neighboring proteins had a higher impact on network topology by removing more edges that resulted in a higher mean path length. Notably, such observations held for *E. coli* phages as well. In [Fig. 3C](#f3){ref-type="fig"}, we focused on protein complexes that involved proteins that were targeted, neighboring, and remaining proteins. Considering the functional classes of proteins, we determined the functional heterogeneity of each protein complex defined as the Simpson diversity index[@b26]. Specifically, such a measure tends towards 1 if functions of proteins are similar and *vice versa*. In both organisms we observed that the distribution of complexes that involved targets and their neighbors were shifted to lower values, suggesting that targets and their neighbors secure a broad reach into different functions. In [Fig. 3D](#f3){ref-type="fig"}, we determined the frequency of functional classes of proteins that are targeted and occur in their immediate vicinity. Compared to the distribution [Fig. 1C](#f1){ref-type="fig"} indicates that such sets of proteins enforce the presence of transcription, replication, recombination, and repair functions while broadening the spectrum to other functions. Discussion ========== Characteristics of bacteriophage-host interfaces ------------------------------------------------ Determining interactions between proteins of bacteriophages Dp-1 and Cp-1 and their host *S. pneumoniae*, we compared their interaction patterns to corresponding observations in the interaction interface of bacteriophages lambda and T7 and their host *E. coli*. Although the phages are biologically different, we found that Cp-1 and Dp-1 share similarities with lambda and T7. In particular, we observed that all phages tend to target highly connected host proteins, have shorter paths to other non-targeted proteins, and connect protein complexes through the interactions of their targets. Furthermore, we observed that targets are enriched in bottleneck proteins, reiterating observations that hold true for human viruses[@b20][@b21][@b22][@b27][@b28][@b29][@b30][@b31][@b32]. In turn, we observed that *E. coli* targets of bacteriophages lambda and T7 tend to cluster in close proximity to each other. Furthermore, we found that *E. coli* proteins with orthologs in *S. pneumoniae* and essential genes appear to cluster around phage-targeted proteins as well. Targets of bacteriophages in *E. coli* appeared to be strongly interconnected based on their network path lengths. In comparison, we found mixed clustering patterns characteristics when we considered targets of bacteriophages Dp-1 and Cp-1 and orthologous and essential proteins in *S. pneumoniae*. Our result may reflect the different ways that interactions between phages and host proteins have been determined. In particular, we collected interactions of phages lambda and T7 from many different sources that focused on the experimental determination of single interactions. In turn, we determined interactions of Dp-1 and Cp-1 on a large scale. Assuming that high-throughput approaches suffer from increased false negative rates, potential targets in the immediate neighborhood of proteins that interact with phages may have been missed. As a consequence, experimental focus on potential interactions that involve neighboring proteins may provide similar characteristics compared to the host-phage interactome of lambda and T7. Although their targets hardly interconnect, we observed that orthologous proteins and essential genes of *S. pneumoniae* in the immediate vicinity of Dp-1 and Cp-1 targets allowed these proteins to organize in a large subnetwork. Such observations suggest that essential and conserved proteins may represent (in)direct gateways to take control of the underlying host cell. The role of immediate neighbors of bacteriophage targets in both hosts is further emphasized by their enrichment with bottleneck nodes and functional classes that are similar to phage targets and their functional heterogeneity. Such observations suggest that phages in general not only target responsive candidate genes to influence, but create a host-pathogen interface that appears confined to immediate network neighbors of targets in the underlying host protein interactions networks. Although such 'extended' host-phage interaction interfaces appear to have limited topological reach, phages manage to achieve global impact that permits the pathogens to quickly take control of the underlying host cell by reaching into various cellular functions. Phage biology and evolution --------------------------- Differences between interaction patterns reflect differences in biology. Phages are exquisitely adapted to their hosts and exploit the resources their hosts provide. As a consequence, all phages are adapted to their host's proteomes and interactomes. At this point hardly any detailed comparisons between phage and their relationships to their hosts exist, a surprise given that phages are among the fastest evolving species on earth. Furthermore, they represent an excellent model for genome, proteome, and interactome evolution. More data will be required to understand the dynamic processes involved in phage-host co-evolution. Protein function ---------------- Poorly annotated phage genomes pose another confounding problem in understanding phage biology. While lambda and T7 are well understood, Cp-1 and Dp-1 have only 12/28 (42%) and 44/77 (61%) of their proteins functionally annotated[@b12][@b13]. Furthermore, estimates suggest that there are tens if not hundreds of different phages per bacterial species[@b33], implying that hundreds of unknown phage proteins interact with their hosts. The recent discovery of the phage CRISPR-Cas9 system[@b34] has impressively shown that a large number of useful activities in phage proteomes exists that may also be used for phage therapy or other applications. We are convinced that PPIs will be a useful tool to investigate and illuminate these functions. Materials and Methods ===================== Molecular interactions data of *E. coli* and its phage ------------------------------------------------------ We collected 2,186 binary-Y2H interactions between 1,264 proteins in *E. coli* that were experimentally determined using a yeast-two-hybrid approach (Y2H) by Rajagopala *et al*.[@b2]. Furthermore, we utilized a total of 9,399 co-complex interactions between 2,044 proteins that were experimentally derived from large-scale tandem affinity purification approaches followed by mass spectrometry (AP/MS)[@b17][@b18]. Finally, we obtained 1,929 literature-curated binary interactions between 1,399 proteins[@b2] that were largely curated from small-scale studies obtained by a multitude of methods. We collected 36 protein-protein interactions between 16 lambda and 23 *E. coli* proteins as well as 19 interactions between 8 T7 and 14 *E. coli* proteins from the literature[@b8]. Essential Genes --------------- We used 712 essential proteins in *E. coli* as well as 436 essential genes in *S. pneumoniae* from the database of essential genes DEG10, an update of the database of essential genes (DEG) that collects data about essential genes from the literature[@b35]. Yeast two-hybrid screens of phage-host interactions --------------------------------------------------- Proteins of Cp-1 (Uniprot proteome: UP000009089) and Dp-1 (UP000008920) were derived from previous studies[@b12][@b13]. These baits were cloned into pDEST32 and screened against a yeast two-hybrid (Y2H) array of 1,704 *S. pneumoniae* TIGR4 ORFs cloned into prey vector pDEST22 as described[@b7]. The strength of Y2H interactions was determined by increasing 3-amino-triazole (3-AT) concentrations up to 50 mM[@b36]. [Tables 1](#t1){ref-type="table"} and [2](#t2){ref-type="table"} indicate 3AT scores, defined as 3AT~max~-3AT~background~. Specifically, 3AT~max~ is the maximal 3-AT concentration where a positive signal was found while 3AT~background~ is the 3-AT concentration where self-activation was suppressed. LuMPIS assays of phage-host PPIs -------------------------------- LuMPIS (Luminescence-based MBP pull-down Interaction screening system) assays were used to verify phage-host PPIs detected in the Y2H screens. Specifically, we used Gateway-compatible LuMPIS vectors with N-terminally MBP-tagged baits (in pCR3.1-N-MBP) to co-purify N-terminally eGFP-luciferase-tagged preys (in pCR3.1-N-eGFPLuc) in a pulled down assay via amylose beads. Proteins were expressed in human embryonic kidney cells (HEK) and raw protein extracts were used for the assay as described[@b37][@b38]. The pulled down preys were detected by measurement of the luciferase activity. Each PPI was measured as quadruplicates and compared to a quadruplicate negative control. The empty bait plasmid (MBP w/o ORF) was used in combination with the GFPluc preys to determine prey binding to MBP. PPIs with a luminescence intensity ratio (LIR) \>3 were considered as positive. Protein complexes in *E. coli* and *S. pneumoniae* -------------------------------------------------- We utilized a set of 517 protein complexes from a co-affinity purification study that was followed by mass spectrometry analyses[@b17] in *E. coli*. We determined network clusters in the underlying protein interaction network of *S. pneumoniae* by utilizing the MCL algorithm[@b39]. In particular, we determined sets of clusters with a gradually increasing inflation parameter. Utilizing COG[@b15][@b16] annotations, we calculated the functional coherence *fc* of cluster *i* as *fc*~*i*~* = fp*~*i*~*/p*~*i*~ where *fp*~*i*~ is the number of protein pairs that share a functional annotation, and *p*~*i*~ is the total number of annotated pairs in cluster *i*. Such a measure tends to increase with small clusters but decreases when more proteins are added. To balance such a trend, one maximizes the size of the given clusters by defining the modularity efficiency *E*~*M*~ as , where *n* is the number of clusters, *N* is the total number of proteins while *N*~*i*~ is the number of proteins in the *i*th cluster[@b24]. In particular, we find a maximum of *E*~*M*~ when we used an inflation parameter of 1.6 in the case of *S. pneumoniae*, allowing us to obtain 148 clusters. Functional heterogeneity of protein complexes --------------------------------------------- Utilizing *N* classes of proteins that appear in a protein complex *i*, we calculated its functional heterogeneity as a Simpson diversity[@b26] index defined as , where *p*~*i*~ is the fraction of proteins of function *i*. Such a measure tends to 1 if proteins functions are similar. Protein Complex Participation Coefficient ----------------------------------------- For each protein that is part of at least one protein complex, we defined the protein complex participation coefficient of a protein *i* as where *n*~*i,s*~ is the number of links protein *i* has to proteins in complex *s* out of *N* total complexes. If a protein predominantly interacts with partners of the same complex, *P* tends to 1[@b20]. Functional classes of proteins ------------------------------ *E. coli* and *S. pneumoniae* proteins were grouped according to broad functional classes that were defined by clusters of orthologous groups (COGs)[@b15][@b16] since COGs provide a consistent classification of bacterial and eukaryotic species based on orthologous groups. Enrichment Analysis ------------------- Binning proteins with a certain characteristic *d* (e.g. with a given number of interactions) we calculated the fraction of proteins that had a feature *i* in each group *d, f*~*i*~*(d)*. As a null model we randomly sampled protein sets with feature *i* of the same size 10,000 times and calculated the corresponding random fraction, *f*~*i,r*~ (*d*). The enrichment/depletion of proteins with feature *i* in a group *d* is then defined as . Enrichment analysis as a function of degree ------------------------------------------- We grouped phage proteins according to their number of interactions in an underlying bacterial protein interaction network. We represented each group by *N*~≥*k*~ proteins that had at least *k* interactions and calculated the number of targeted proteins *i, N*~*i*,≥*k*~ in each group. Randomly picking targeted genes we defined as their enrichment where was the corresponding random number of targeted proteins among all *N*~i,≥*k*~ proteins. After averaging *E*~*i*~ over 10,000 randomizations *E*~*i*~ \> *1* pointed to an enrichment and *vice versa*, while *E*~*i*~ \~ *1* indicated a random process[@b40]. Orthologous proteins -------------------- Utilizing all-versus-all BLASTP searches determined by the InParanoid script[@b41] in protein sets of *E. coli* and *S. pneumoniae*, sequence pairs with mutually best scores were selected as central orthologous pairs. To enhance quality, we only accepted BLAST matches with a score \>40 bits, covering at least 50% of the longer sequence. Proteins of both species that showed such an elevated degree of homology were clustered around these central pairs, forming orthologous groups. The quality of the clustering was further assessed by a standard bootstrap procedure. We only considered the central orthologous sequence pair that provided a confidence level of 100% as the real orthologous relationship, allowing us to obtain 781 orthologous protein pairs between *E. coli* and *S. pneumoniae.* Bottleneck proteins ------------------- As a global measure of its centrality, we defined betweeness centrality *c*~*B*~ of a protein *v* as where *σ*~*st*~ was the number of shortest paths between proteins *s* and *t* while *σ*~*st*~ (*v*) was the number of shortest paths running through protein *v*. As a set of bottleneck proteins we defined the top 20% of proteins with highest betweeness centrality. Additional Information ====================== **Data availability**: The protein interactions from this publication have been submitted to the IMEx ([http://www. imexconsortium.org](http://www.imexconsortium.org)) consortium through IntAct (ref. [@b42]) and assigned the identifier IM-25020. **How to cite this article**: Mariano, R. *et al*. The interactome of *Streptococcus pneumoniae* and its bacteriophages show highly specific patterns of interactions among bacteria and their phages. *Sci. Rep.* **6**, 24597; doi: 10.1038/srep24597 (2016). This work was supported by a start-up package from the Univ. of Miami., by the U.S. National Institutes of Health grant U54 DE023786-01, and the VCU Libraries Open Access Publishing Fund. **Author Contributions** S.W. and P.U. designed the study. R.H. and M.V.-P. determined protein interactions, including LuMPIS experiments. R.M. and S.W. analyzed results. R.M., S.W. and P.U. interpreted results and wrote the paper. All authors reviewed the manuscript. {#f1} {#f2} {#f3} ###### Interactions between proteins of *S. pneumoniae* and its phage Cp-1. **Phage protein** **Host protein** **3AT** **LIR** ------------------- ---------------------- --------- ---------------------------------------- ------ ----- Cpl1 Lysozyme SP_1208 uridine kinase 47.5 4 orf6 Hypothetical protein SP_1713 transcriptional regulator, NrdR family 50 6 orf10 Connector protein SP_1354 ribosomal protein L7/L12 2.5 12 orf10 Connector protein SP_1881 glutamate racemase 50 6 orf16 Hypothetical protein SP_0259 Holliday junction DNA helicase RuvB 1 12 orf17 Tail protein N SP_0979 oligoendopeptidase F 50 14 orfb Hypothetical protein SP_2168 putative fucose operon repressor 2.5 1 orfb Hypothetical protein SP_0859 membrane protein 25 1 orfb Hypothetical protein SP_1980 cmp-binding-factor 1 25 2 orfb Hypothetical protein SP_1213 conserved domain protein 50 240 orfc Hypothetical protein SP_0979 oligoendopeptidase F 2 3 Host proteins are given as locus numbers. 3AT is the highest 3-AT concentration at which this interaction was detected. LIR are luminescence intensity ratios from LuMPIS assays rounded to the nearest integer (see text for details). Combinations of high 3AT score and LIR values are most reliable. ###### Interactions between proteins of *S. pneumoniae* and its phage Dp-1. **Phage protein** **Host protein** **3AT** **LIR** ------------------- ----------------------------------------- --------- ------------------------------------------------------------- ------ ----- orf4 Queuosine biosynth. protein QueE SP_2036 PTS system, IIA component 25 454 orf9 No similarity SP_1504 TPR domain protein 0.5 19 orf9 No similarity SP_0259 Holliday junction DNA helicase RuvB 50 11 orf9 No similarity SP_1395 putative phosphate transport system regulatory protein PhoU 50 23 orf9 No similarity SP_2168 putative fucose operon repressor 50 15 orf12 Holliday junction resolvase RecU SP_2168 putative fucose operon repressor 0.1 2 orf14 dUTPase SP_2125 conserved hypothetical protein 50 45 orf16 NAD-dependent DNA ligase SP_0259 Holliday junction DNA helicase RuvB 50 5 orf18 DNA polymerase III, delta' subunit HolB SP_1584 GTP-sensing transcriptional pleiotropic repressor CodY 24.5 6 orf29 Hypothetical protein SP_2012 glyceraldehyde 3-phosphate dehydrogenase 0.25 3 orf31 Hypothetical protein SP_2168 putative fucose operon repressor 0.2 5 orf31 Hypothetical protein SP_1153 hypothetical protein 0.25 6 orf31 Hypothetical protein SP_0259 Holliday junction DNA helicase RuvB 0.5 4 orf32 Hypothetical protein SP_0194 conserved hypothetical protein 47.5 5 orf32 Hypothetical protein SP_0259 Holliday junction DNA helicase RuvB 47.5 3 orf32 Hypothetical protein SP_1540 single-strand binding protein *Ssb* 47.5 3 orf32 Hypothetical protein SP_1669 MutT/nudix family protein 47.5 10 orf32 Hypothetical protein SP_1915 hypothetical protein 47.5 5 orf33 Hypothetical protein SP_1088 DNA repair protein RadC 0.5 3 orf34 Hypothetical protein SP_2157 alcohol dehydrogenase, iron-containing 2 35 orf34 Hypothetical protein SP_0446 acetolactate synthase, small subunit 9 12 orf39 Zinc finger protein SP_0259 Holliday junction DNA helicase RuvB 50 10 orf44 Rho-like domain lipoprotein SP_1725 sucrose operon repressor 2.25 3 orf44 Rho-like domain lipoprotein SP_2157 alcohol dehydrogenase, iron-containing 2.25 13 orf44 Rho-like domain lipoprotein SP_1050 putative transcriptional regulator 4.75 2 orf44 Rho-like domain lipoprotein SP_1536 conserved hypothetical protein 4.75 10 orf44 Rho-like domain lipoprotein SP_0446 acetolactate synthase, small subunit 9.75 6 orf44 Rho-like domain lipoprotein SP_1575 conserved hypothetical protein 9.75 14 orf47 Hypothetical protein SP_0687 ABC transporter, ATP-binding protein 49 4 orf48 Hypothetical protein SP_2168 putative fucose operon repressor 0.5 3 orf48 Hypothetical protein SP_1746 conserved hypothetical protein 2.5 3 orf51 Hypothetical protein SP_1672 recombination protein RecR 0.5 6 orf58 Holin SP_1505 membrane protein 1 1 orf58 Holin SP_1731 conserved hypothetical protein 2.5 3 orf58 Holin SP_1606 glycosyl transferase, family 2 25 3 orf58 Holin SP_1751 putative transporter, CorA family 25 5 orf60 Hypothetical protein SP_2024 PTS system, IIA component 10 5 orf72 Membrane protein SP_1606 glycosyl transferase, family 2 50 50 Host proteins are given as locus numbers. 3AT is the highest 3-AT concentration at which this interaction was detected. LIR are luminescence intensity ratios from LuMPIS assays rounded to the nearest integer (see text for details). Combinations of high 3AT score and LIR values are most reliable. [^1]: These authors contributed equally to this work. [^2]: Present address: Harvard University, Cambridge, MA 02238, USA.

All relevant data are within the paper and its Supporting Information files. Introduction {#sec001} ============ Crew rowing is often quoted as the prime example of real-life joint action, interpersonal coordination dynamics and synchronisation (e.g., \[[@pone.0133527.ref001]--[@pone.0133527.ref005]\]), and for group processes in general as well (e.g., \[[@pone.0133527.ref006], [@pone.0133527.ref007]\]). The degree of synchronisation between the rowers in a boat is generally regarded as an important determinant for optimal crew performance (e.g., \[[@pone.0133527.ref008], [@pone.0133527.ref009]\]). Even a team of individually strong and technically skilled rowers will probably not win the race if they do not properly coordinate their movements together. Researchers, coaches and rowers agree that to achieve optimal performance of the crew, rowers must row in perfect synchrony \[[@pone.0133527.ref010]\]. Despite the well-known tendency of humans to synchronize their movements (e.g., \[[@pone.0133527.ref011]\]), though, rowing in sync does not come naturally. Inexperienced crews often have difficulties row in a common rhythm, and it often takes many years to perfect the synchrony of the crew. In the current study we examine such crew synchronisation processes. Before testing on water, we chose to first study the coordination of rowing dyads in a controlled laboratory setting. More specifically, using a setup of coupled rowing ergometers (see below), the current study investigated the effect of movement frequency on crew coordination. With regard to crew coordination it is important to note that a stroke cycle consists of two phases, namely the *drive*, during which force is applied to the blade in order to move the boat relative to the water, and the *recover*, during which the rowers return to their former position. Hence, there is discontinuous propulsion. Moreover, the rowers are seated on sliding seats, which allows for the essential contribution of the strong leg extensor muscles to the propulsion. In each stroke cycle the relatively heavy crew concurrently pushes off against the foot stretchers, thereby causing the relatively light boat to accelerate in the opposite direction of the crew. As a result, the velocity of the boat fluctuates within each rowing cycle, which implies a 5--6% power loss \[[@pone.0133527.ref012], [@pone.0133527.ref013], [@pone.0133527.ref014]\]. Interestingly, theoretically, an out-of-phase crew coordination pattern may minimise this power lost to velocity fluctuations of the boat \[[@pone.0133527.ref015]\]. By perfectly alternating their strokes, the produced forces and associated movements of the rowers perfectly counteract each other, yielding the net centre of mass of the crew to stay close to the centre of mass of the boat. As a result, the shell velocity remains near to constant over the whole rowing cycle. A recent off-water study in which dyads rowed on mechanically linked ergometers at a rate of 36 strokes per minute (*spm*) confirmed that, compared to in-phase rowing, antiphase crew coordination drastically decreased ergometer displacement, and demonstrated that the 5--6% power lost to velocity fluctuations was indeed regained in the antiphase rowing condition \[[@pone.0133527.ref013]\]. Although antiphase rowing might be beneficial due to reduction of velocity fluctuations of the boat, the question is how stable antiphase patterns can be performed at the higher stroke rates as typically seen in racing. From coordination dynamics literature \[[@pone.0133527.ref016]--[@pone.0133527.ref020]\] it is known that an increase in movement rate may destabilise the coordinative performance, especially for antiphase coordination \[[@pone.0133527.ref017]--[@pone.0133527.ref019]\], which at a critical frequency yields transitions towards in-phase coordination \[[@pone.0133527.ref016]--[@pone.0133527.ref018]\]. As stroke rates during a race can reach values over 40 *spm*, it is interesting to examine whether antiphase crew synchronisation can be maintained with sufficient stability at the highest movement frequencies. For these reasons, the present study set out to systematically examine both in-phase and antiphase crew rowing at increasing stroke rates, from a coordination dynamics perspective. Coordination dynamics {#sec002} --------------------- The coordination between people has been studied for a range of cyclic movement tasks, such as moving hand-held pendulums \[[@pone.0133527.ref021]--[@pone.0133527.ref022]\], rocking chairs \[[@pone.0133527.ref023]\], joint stepping \[[@pone.0133527.ref024]\] and side-by-side walking \[[@pone.0133527.ref025]\] (for an overview, see e.g., \[[@pone.0133527.ref026]\]). Given the cyclical nature of the rowing act, along with the fact that it has to be performed in coordination with other rowers, crew rowing movements can be modelled as coupled oscillations. As such, coordination dynamics offers a pertinent theoretical approach to study crew synchronisation. From this perspective, the coordination between two oscillatory components can be described in terms of the relative phase (*ϕ*), formulated as: $$\phi = \theta_{1} - \theta_{2}$$ with *θ* ~1~ and *θ* ~2~ depicting the phase angle of each oscillator (i.e. were it resides in its cycle from 0° to 360°). From studies on bimanual interlimb coordination, it is known that people are generally able to perform two stable coordinative patterns, namely in-phase coordination (*ϕ* = 0°), wherein the rhythmic hand movements perfectly coincide, and antiphase coordination (*ϕ* = 180°), in which the rhythmic movements perfectly alternate with half-a-cycle difference. Other coordinative patterns are unstable without training \[[@pone.0133527.ref027]\]. For both within- \[[@pone.0133527.ref016]--[@pone.0133527.ref017]\] as between-person coordination \[[@pone.0133527.ref004], [@pone.0133527.ref018]--[@pone.0133527.ref020]\] it has been shown that when people start coordinating in an antiphase pattern, an increase in movement frequency typically results in a transition into in-phase coordination. Importantly, when people start in in-phase, no transition into another pattern (e.g., antiphase) occurs. The fact that this phenomenon is observed in both within- and between-person coordination indicates that the source of coupling (neural, perceptual, or otherwise) is immaterial for the synchronisation process \[[@pone.0133527.ref028], [@pone.0133527.ref029]\]. As an account for these behavioural phenomena, Haken, Kelso and Bunz \[[@pone.0133527.ref017]\] formulated a model of two non-linearly coupled limit cycle oscillators from which an equation of motion could be derived that captured the rate of change in relative phase angle between the two oscillating components, following: $$\overset{˙}{\phi} = - a\mspace{1mu}\text{sin}\mspace{1mu}\phi - 2b\mspace{1mu}\text{sin}\mspace{1mu} 2\phi$$ with *a* affecting the attractor strength of in-phase coordination, and *b* affecting the attractor strength of both in- and antiphase coordination. The ratio *b/a* is directly related to the movement frequency. Given [Eq 2](#pone.0133527.e002){ref-type="disp-formula"}, at low frequencies (i.e., *b/a* \> .25) the model has two stable attractors, namely in-phase and antiphase coordination, while at higher movement frequencies (i.e., *b/a* \< .25) the attractor strength of both in- and antiphase decreases with increasing frequency, although this decrease is stronger for antiphase than for in-phase coordination. This frequency effect for in- and antiphase has been supported in numerous bimanual studies (see \[[@pone.0133527.ref030]\] for an overview) and between-persons studies \[[@pone.0133527.ref018]--[@pone.0133527.ref019]\]. The coupled oscillator model shows that, due to coupling forces, at a certain critical frequency the antiphase pattern is not stable anymore and the coordination is compelled towards a transition to the remaining stable in-phase pattern. Would this also show up in crew rowing at increasing stroke rates? In-phase vs. antiphase crew coordination {#sec003} ---------------------------------------- De Brouwer et al. \[[@pone.0133527.ref013]\] showed that, consistent with previous studies on coordination dynamics, antiphase crew coordination (as measured by the phase relation between the rowers' centres of mass) was less accurate and less consistent compared to conventional in-phase crew coordination. Still, although the dyads in that study performed an antiphase pattern for the first time ever, they demonstrated strikingly little difficulty rowing in antiphase, even at the rather high stroke rate of 36 *spm*. This clearly confirmed the intrinsic stability of the interpersonal antiphase pattern and the ease to perform this pattern. However, in their steady state trials, one dyad showed a brief breakdown of the antiphase pattern towards in-phase crew coordination \[[@pone.0133527.ref013]\]. This suggests that despite the apparent ease with which antiphase rowing was performed, this pattern is still less stable and therefore more prone to perturbations of any kind, which may break down the coordination pattern. For obvious reasons, most studies regarding interpersonal coordination emphasise perceptual coupling (mainly visual) in mediating synchronisation processes \[[@pone.0133527.ref011], [@pone.0133527.ref029]--[@pone.0133527.ref032]\]. In crew rowing, though, there is also a direct *physical* coupling between the rowers, namely through the boat that they share. To date not many studies on interpersonal coordination have addressed this type of coupling. Crew rowing may therefore serve as an interesting experimental task to provide insights for coordinative behaviour of mechanically coupled individuals (see also \[[@pone.0133527.ref033]\]). To create the physical coupling and address the effect of crew coordination on 'boat' movements, similar to De Brouwer et al. \[[@pone.0133527.ref013]\], we used two rowing ergometers serially placed on slides to allow these to move back and forth as one 'boat' (see [Fig 1](#pone.0133527.g001){ref-type="fig"} and *[Experimental Setup](#sec006){ref-type="sec"}*). (Note that with 'boat' (between apostrophes) we refer to the mechanically linked ergometers.) Obviously, rowing on water differs from rowing on linked ergometers (see *[Discussion](#sec017){ref-type="sec"}*). Yet, the kinematics of ergometer rowing are largely similar to those of on-water rowing \[[@pone.0133527.ref034]\], particularly for free-floating ergometers as used in the present study \[[@pone.0133527.ref035]--[@pone.0133527.ref036]\]. In sum, given the above, we expect faster stroke rates to involve poorer crew synchronisation for both in- and antiphase rowing, with a stronger effect for antiphase. Most importantly, antiphase crew coordination is expected to break down with increasing stroke rate. {ref-type="supplementary-material"} and [S2 Movie](#pone.0133527.s003){ref-type="supplementary-material"} for exemplary movies.](pone.0133527.g001){#pone.0133527.g001} Method {#sec004} ====== Participants {#sec005} ------------ Eleven male rowing dyads participated in the experiment (age 24 ± 5 years; length 1.86 ± 0.05 m; mass 76.86 ± 8.27 kg; rowing experience at club level 5 ± 3 years). Nine dyads signed up for the experiment as actual teammates, while the other two dyads were composed based on availability for the experiment and matched for body mass. All participants provided written informed consent prior to participation. The individuals in the movies in the supplementary material ([S1 Movie](#pone.0133527.s002){ref-type="supplementary-material"} and [S2 Movie](#pone.0133527.s003){ref-type="supplementary-material"}) have given additional written informed consent (as outlined in the PLOS consent form) to publish video material of their experimental trials. The Ethics Committee of the Center for Human Movement Sciences, University Medical Center Groningen approved the study that was conducted according to the principles expressed in the Declaration of Helsinki. Experimental setup {#sec006} ------------------ Trials were performed using two mechanically coupled rowing ergometers (Type D, Concept2, USA) serially placed on slides (see [Fig 1](#pone.0133527.g001){ref-type="fig"}). This setup allows the ergometers to move with respect to the floor as a single 'boat', reminiscent to on water rowing. The resistance of the ergometer flywheels was set at a level that reflected on-water blade resistance. Dyads determined and set the resistance at an aerobic constant between 0.96∙10^−4^ kg∙m^2^ and 1.20∙10^−4^ kg∙m^2^ (i.e., drag factor 96--120). Rowers within a dyad rowed at the same drag factor. The kinematics of the rowers and ergometer system were recorded with a Vicon Motion Analysis System (Vicon Motion Systems, Inc., Centennial, CO), using 17 retro-reflective markers. Three markers were placed on the ergometer of the stroke rower, the remaining markers were placed on both sides- and on the middle of each handle and on both left and right shoulder (acromion) and hip (greater trochanter) of each rower, as depicted in [Fig 1](#pone.0133527.g001){ref-type="fig"}. Using 8 infrared cameras placed around the measurement volume, the markers' 3D-trajectories were collected at a sample rate of 200 Hz. Protocol {#sec007} -------- To warm up and get familiar with the experimental set up, each dyad started with rowing five minutes in in-phase coordination and five minutes in antiphase coordination at a self-chosen stroke rate (about 20--24 *spm*), including about 30 *s* of rowing at a higher stroke rate (\>30 *spm*) for each condition. After a short break, each dyad performed a two-minute in-phase and a two-minute antiphase trial at 30 *spm* in counterbalanced order with at least five minutes of rest in between. Next, for both in- and antiphase crew coordination the dyads performed a ramp trial in which they were instructed to increase stroke rate every 30 *s* with 2 *spm*, starting from 30 *spm* until the dyad could not increase the stroke rate any further. The rowers received feedback about stroke rate on a monitor (PM4, Concept 2, USA; see [Fig 1](#pone.0133527.g001){ref-type="fig"}). Exemplary movies of in- and antiphase rowing are provided in [S1](#pone.0133527.s002){ref-type="supplementary-material"} and [S2](#pone.0133527.s003){ref-type="supplementary-material"} Movies. Data analysis {#sec008} ------------- Since in the current study we were specifically interested in the occurrence of coordination breakdowns and the stability of crew coordination over increasing stroke rates, we only report the analysis of the ramp trials. Movement data were analysed in the sagittal plane using customized procedures (Matlab, MathWorks, USA). From these movement data three kinematic time series were used: 1) forward-backward trunk movement of both rowers (estimated as the mean position of hip and shoulder markers) with respect to the 'boat' movement; 2) handle positions (estimated from the three markers placed on the handles) relative to the 'boat' movement; and 3) 'boat' position estimated from the three markers placed on the ergometer system). These data were filtered using a bidirectional second order low-pass Butterworth filter with a 15 Hz cut-off frequency. ### Relative phase {#sec009} The instantaneous spatio-temporal relation between the rowers' trunk movements was expressed by the continuous relative phase (*ϕ*, see [Eq 1](#pone.0133527.e001){ref-type="disp-formula"}). First, the continuous phase angles *θ* ~*i*~, with subscript *i* depicting the stroke rower (1) and bow rower (2) respectively, were determined using $$\theta_{i}(t) = \text{tan}^{- 1}\left( \frac{v_{i}(t)}{x_{i}(t)} \right)$$ with *x* ~i~ indicating horizontal trunk position and *v* ~*i*~ the trunk velocity normalized by the angular frequency for each half cycle separately, following Varlet & Richardson \[[@pone.0133527.ref037]\] (for similar half-cycle normalization procedures, see \[[@pone.0133527.ref038], [@pone.0133527.ref039]\]). The normalization was done because in rowing the drive (cf. backward movement) and recovery (cf. forward movement) are typically not equal in duration \[[@pone.0133527.ref008], [@pone.0133527.ref040]\], hence the backward movements had a slight but systematically higher movement frequency than the forward movements (see also \[[@pone.0133527.ref013], [@pone.0133527.ref041]\]). Start and end of each half cycle were based on the instances of the minimum and maximum excursions of the signal, which were determined using a custom made peak-picking algorithm. From the thus obtained phase angles, the continuous relative phase *ϕ*(*t*) was determined according [Eq 1](#pone.0133527.e001){ref-type="disp-formula"}. Apart from the drive and recovery differing in duration, a rower spends more time around the finish than around the catch of the stroke \[[@pone.0133527.ref013], [@pone.0133527.ref040]\], which implies that in rowing the movement cycle deviates from perfect harmonicity. Therefore, we also calculated a discrete measure of relative phase that is not sensitive to such inharmonicities. As such, for both handle and trunk movements, we determined the discrete relative phase based on point-estimates of peak excursions near the catch of the stroke, which was calculated for each full cycle as: $$\phi_{PE} = \frac{t_{2,j} - t_{1,j}}{t_{1,j + 1} - t_{1,j}}360{^\circ}$$ in which *t* ~1,*j*~ and *t* ~2,*j*~ indicate the time of the *j* ^*th*^ peak of the trunk- and handle position of rower 1 and 2. ### Dependent measures {#sec010} The data were analysed over steady state bins for each stroke rate. These bins were selected based on inspection of stroke rate time series. Evidently, each 2 *spm* change in stroke rate involved a transient stage, which in some cases lasted longer than in other cases. Hence, the bins could only be determined a posteriori. Due to the variation in transient stages, selected bins ranged between 6 and 12 strokes/bin. For each bin, the following measures were determined. Means and standard deviations of relative phase measures were determined using directional (i.e., circular) statistics \[[@pone.0133527.ref042]\]. As a measure of coordinative variability, we calculated the standard deviation of the discrete relative phase based on trunk positions (*SDϕ* ~*PE*~). We used relative phase base on trunk position data, because displacement of the body is directly related to the displacement of the ergometers. Next to that, coordinative variability was also determined from handle movements (*SDϕ* ~*PE*−*H*~), as the handle movements serve as end-effector of the rowing movement (i.e., force is applied to the water/flywheel via the handles/oar(s)). In order to minimize boat velocity fluctuations, antiphase crew coordination not only needs to be stable, but also accurate. Therefore, the absolute deviation from the intended pattern (0° or 180°) was calculated for each time sample of the continuous relative phase *ϕ*, and then averaged in each bin to obtain the absolute error of relative phase (*AEϕ*). To provide an indication of the above-described inharmonicity within the rowing cycle, for each individual rower, for each stroke rate bin, the mean ratio of the backward to forward trunk movement duration (*ratio*) was calculated. Finally, to provide insight in the effect of in- and antiphase crew coordination on 'boat' movements, variability of the ergometer velocity (*SDv* ~*E*~) was calculated for each bin. Statistical analysis {#sec011} -------------------- Each of the above mentioned dependent measures was subjected to a 2 Pattern × 4 Tempo repeated-measures ANOVA. For this analysis we selected the first four levels of stroke rate (i.e., around 30-32-34-36 *spm*) because, as we will see, dyads achieved different maximal stroke rates in each condition, which would preclude proper comparison of all stroke rates. Furthermore, because three dyads did not achieve at least four stroke rate levels in each condition, these could not be included in this analysis. An α of .05 was adopted for all tests of significance. Results {#sec012} ======= All dyads but one (see below) easily managed rowing in antiphase coordination within the five-minute warm-up. In the ramp trials, a first important observation was that dyads achieved higher stroke rate levels in in-phase compared to antiphase rowing ([Table 1](#pone.0133527.t001){ref-type="table"}). Still, each dyad was able to maintain antiphase at stroke rates that are similar to rates achieved in competition. 10.1371/journal.pone.0133527.t001 ###### Performed mean stroke rate at final bin for each dyad in in-phase and antiphase crew coordination. Dyads 4 and 8 showed a transition from antiphase to in-phase. {#pone.0133527.t001g} Dyad nr ----------- --------- ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ In-phase 39.3 37.8 42.2 35.9 46.2 48.3 41.5 39.7 39.9 44.6 44.0 Antiphase 37.6 36.8 34.6 31.3 37.0 40.9 35.9 32.9 35.8 36.7 41.9 Coordination breakdowns {#sec013} ----------------------- Two of the eleven dyads showed a breakdown of the antiphase coordination (for an example, see [Fig 2](#pone.0133527.g002){ref-type="fig"}). No transitions from in- to antiphase crew coordination occurred. Notably, both cases of antiphase breakdown already occurred at the very beginning of the trial at a stroke rate between 30--33 *spm*. After the breakdown occurred, one dyad tried to restore the antiphase coordination pattern but they did not succeed, as illustrated in [Fig 2](#pone.0133527.g002){ref-type="fig"}. The other dyad did not make an attempt to restore coordination and they expressed verbally that they did not feel comfortable rowing in antiphase. This dyad had already shown difficulties maintaining antiphase coordination during the steady state trial and warm-up, suggesting that the coordination breakdown occurred because they were not able (or willing) to row in antiphase at any stroke rate, rather than as a result of an increase in stroke rate. {#pone.0133527.g002} Coordinative performance {#sec014} ------------------------ The effect of stroke rate on steady state crew coordination is depicted in Figs [3](#pone.0133527.g003){ref-type="fig"} and [4](#pone.0133527.g004){ref-type="fig"}. For *SDϕ* ~*PE*~ and *AEϕ* no significant main effect of Tempo, nor a significant Pattern × Tempo interaction was present. For *SDϕ* ~*PE*−*H*~ (i.e., coordination based on the handles), a significant main effect of Tempo was found, *F*(3,21) = 3.102, *p* \< .05, *η* ~*p*~ ^2^ = .307. Post hoc analysis (Bonferroni-corrected *t*-tests) of this main effect revealed no significant differences between the four levels of stroke rate. Looking at the mean values of *SDϕ* ~*PE*−*H*~ for each tempo condition (30 *spm*: 4.13°, 32 *spm*: 3.21°, 34 *spm*: 4.24°, 36 *spm*: 4.81°) the main effect suggests a slight increase in variability towards the higher stroke rate (see also [Fig 3](#pone.0133527.g003){ref-type="fig"}, lower panels). {#pone.0133527.g003} {#pone.0133527.g004} Regarding the difference between in- and antiphase, *AEϕ* was higher for antiphase than for in-phase crew coordination ([Table 2](#pone.0133527.t002){ref-type="table"}, see also Figs [3](#pone.0133527.g003){ref-type="fig"} and [4](#pone.0133527.g004){ref-type="fig"}), which was confirmed by a significant main effect for Pattern (see [Table 2](#pone.0133527.t002){ref-type="table"}). Quite remarkably, the coordinative variability (both *SDϕ* ~*PE*~ and *SDϕ* ~*PE*−*H*~) did *not* differ significantly between the in- and antiphase pattern ([Table 2](#pone.0133527.t002){ref-type="table"}, see also [Fig 3](#pone.0133527.g003){ref-type="fig"} lower panels) as there was neither a significant effect of Pattern nor a Pattern × Tempo interaction. 10.1371/journal.pone.0133527.t002 ###### Mean values for in-phase and antiphase rowing in terms of interpersonal coordination (*SDϕ* ~*PE*~, *SDϕ* ~*PE*−*H*~, and *AEϕ*), ergometer velocity fluctuations (*SDv* ~*E*~) and drive-recovery ratio (*ratio*), and RM ANOVA statistics. {#pone.0133527.t002g} Mean values RM ANOVA results ---------------------- ------------- ------------------ --------- ------- --------- ------ *SDϕ* ~*PE*~ (°) 2.235 4.183 4.753 1, 7 .066 .404 *SDϕ* ~*PE*−*H*~ (°) 4.813 3.438 2.366 1, 7 .168 .253 *AEϕ* (°) 4.125 17.189 21.919 1, 7 \< .005 .758 *SDv* ~*E*~ (m/s^2^) 0.667 0.221 931.383 1, 7 \< .001 .993 *ratio* .814 .884 44.861 1, 15 \< .001 .749 Backward-forward movement ratio {#sec015} ------------------------------- Similar to De Brouwer et al. \[[@pone.0133527.ref013]\], trunk movements of the rowers were slightly faster during the drive than in the recover, for both in- and antiphase crew coordination. This was reflected by the *ratio* values that were generally lower than 1. From [Fig 5A](#pone.0133527.g005){ref-type="fig"} it seems that the *ratio* increased with stroke rate, especially for in-phase rowing, indicating that the rowing cycle became more harmonic. This was supported by the results from a 2 Pattern × 4 Tempo repeated-measures ANOVA over the first four tempo levels (see *'[Statistical analysis'](#sec011){ref-type="sec"}*) for the *ratio* of both rowers of these 8 dyads (n = 16). This analysis revealed that the *ratio* was significantly higher (i.e. closer to 1) for the antiphase pattern (see [Table 2](#pone.0133527.t002){ref-type="table"}) and that for both in-phase and antiphase crew coordination the *ratio* increased with tempo for 30--36 *spm*, given the main effect of Tempo *F*(1.50,22.55) = 19.89, *p* \< .001, *η* ~*p*~ ^2^ = .570. The Pattern x Tempo interaction did not reach significance. {#pone.0133527.g005} Movements of the ergometer system ('boat') {#sec016} ------------------------------------------ [Fig 6](#pone.0133527.g006){ref-type="fig"} shows that the fluctuations in ergometer velocity, as indicated by *SDv* ~*E*~, were much lower for antiphase than for in-phase crew coordination, yielding a significant main effect of Pattern (see [Table 2](#pone.0133527.t002){ref-type="table"}). Furthermore, whereas for in-phase *SDv* ~*E*~ clearly increased with stroke rate, [Fig 6](#pone.0133527.g006){ref-type="fig"} suggests that for antiphase crew coordination this was not the case. This was corroborated by a significant Pattern × Tempo interaction, *F*(3,21) = 74.93, *p* \< .001, *η* ~*p*~ ^2^ = .915. Analysis of the simple effects confirmed a significant effect of Tempo for in-phase, *F*(2.04,11.36) = 152.38, *p* \< .001, *η* ~*p*~ ^2^ = .956, but not for antiphase, *F*(1.75,12.24) = 1.59, *p* = .243, *η* ~*p*~ ^2^ = .185. These effects of pattern and tempo on ergometer movement can also be observed in [Fig 7](#pone.0133527.g007){ref-type="fig"}, which provides examples of the rowers' trunk movements and the ergometer movement during a few strokes of in-phase and antiphase rowing at 32 *spm* and at 40 *spm*. {#pone.0133527.g006} {#pone.0133527.g007} Discussion {#sec017} ========== This study examined in- and antiphase crew coordination at increasing stroke rates on mechanically linked ergometers. The main aim was to investigate whether with an increase in stroke rate antiphase crew coordination could still be maintained stably or that it would break down into the more stable in-phase pattern, as may be expected from coupled oscillator dynamics. Antiphase breakdowns were observed for two of the eleven dyads, whereas no coordinative breakdowns were seen for in-phase crew coordination. In antiphase, dyads did not achieve stroke rates as high as in in-phase crew coordination. Still, the study shows that most dyads were able to sufficiently maintain antiphase crew coordination on linked ergometers at stroke rates as high as used in on-water racing. Coordination breakdowns {#sec018} ----------------------- Two dyads showed a breakdown from anti- towards in-phase crew coordination. These transitions already took place in the very beginning of the ramp trial at 30--33 *spm*, which might suggest that the breakdowns were not necessarily induced by an increase in movement rate. In line with this reasoning, one of these dyads expressed trouble with antiphase crew coordination in general (see '*[Results'](#sec012){ref-type="sec"}*), while the other dyad tried to restore the antiphase pattern, but failed to do so. As can be observed in [Fig 2](#pone.0133527.g002){ref-type="fig"}, once antiphase crew coordination is lost, the 'boat' starts moving at larger amplitudes and it is difficult to counter these boat movements to regain antiphase (see below for more detailed discussion: '*[On mechanical coupling](#sec021){ref-type="sec"}'*). Interestingly, after the experiment the latter dyad mentioned that their initial antiphase breakdown (starting around the 6^th^ s of the trial, see [Fig 2](#pone.0133527.g002){ref-type="fig"}) might have occurred because they temporary lost their attention/concentration. This suggestion fits with earlier findings that antiphase coordination requires more attention than in-phase coordination, also in interpersonal tasks \[[@pone.0133527.ref043]\]. It could well be that the observed breakdowns were instigated by perturbing sources that were not related to an increase in tempo. According to coupled oscillator dynamics, at higher movement frequencies the coordination is less resistant to (either external or internal) perturbations, especially for antiphase coordination \[[@pone.0133527.ref044]\]. As in the present experiment the ramp trials already started at a reasonably high stroke rate (30 *spm*), at this tempo, the coordination might already have been too sensitive to (e.g., attentional) perturbations for these two dyads. In contrast to the two dyads that showed breakdowns of their antiphase rowing patterns, the remaining nine dyads were able to achieve antiphase crew coordination at rates maxing out at 34 up to 42 *spm* (see [Table 1](#pone.0133527.t001){ref-type="table"}). In antiphase crew coordination these nine dyads did not achieve stroke rates as high as in in-phase rowing. In principle, this might indicate that if they would have been able to reach higher rates, antiphase breakdowns might have occurred. It is worth mentioning that in previous experiments on transitions from anti- to in-phase coordination, participants often were instructed 'not to resist if they felt that the pattern tended to change' (e.g., \[[@pone.0133527.ref016], [@pone.0133527.ref018]\]). In the present experiment we simply instructed dyads to perform the instructed patterns up to the fastest as possible rate. In other words, dyads may have decided to stop when they felt they could not maintain the antiphase pattern anymore at a certain tempo. Achieved stroke rates {#sec019} --------------------- An additional explanation for the higher stroke rates attained in in-phase as compared to antiphase crew coordination might be found in a difference in 'boat' movement in these two situations. The difference between the maximum tempos in both patterns may (at least) partly be due to the ergometer setup. As the slides allow the ergometers to float free with respect to the floor, in in-phase the crew concurrently moves the 'boat' backward and forward because the weight of the 'boat' is much lower than that of the crew (see also *[Introduction](#sec001){ref-type="sec"}*and Figs [6](#pone.0133527.g006){ref-type="fig"} and [7](#pone.0133527.g007){ref-type="fig"}). Previous research indicated that moving the light ergometer rather than the heavy body makes it easier to achieve higher stroke rates compared to rowing on an ergometer that is fixed to the floor \[[@pone.0133527.ref045]\]. For antiphase coordination, however, the forces and, hence, movements of the rowers counteract each other and the system of ergometers stays approximately at the same place (Figs [6](#pone.0133527.g006){ref-type="fig"} and [7](#pone.0133527.g007){ref-type="fig"}). Thus, during antiphase rowing the free-floating ergometers behave as fixed ergometers, which can partly explain why achieved stroke rates were lower for antiphase rowing. In the study of De Brouwer et al. \[[@pone.0133527.ref013]\] a servomotor was used to resist the movement of the ergometers, that is, to induce the velocity dependent resistance that is present on the water. In the current experimental set up, however, we did not use such an 'extra' resistance. Therefore, for in-phase rowing the 'boat' could be moved with more ease compared to De Brouwer et al. \[[@pone.0133527.ref013]\] and compared to on-water rowing, which indicates that the highest tempos reached in in-phase in the present experiment likely overestimate on-water stroke rates. For antiphase rowing this was less of an issue because the 'boat' stayed approximately still. Together, the higher stroke rates achieved in in-phase rowing are at least in part due to 1) increased movements of the ergometer system, and 2) the fact that there was no resistance applied to the moving ergometers. Steady state coordinative performance {#sec020} ------------------------------------- As mentioned before, we expected that for both patterns the crew coordination would deteriorate with increasing stroke rate, with a stronger effect for antiphase than for in-phase. Although we found a significant main effect of stroke rate on the consistency of the coordination between handle movements (as indexed by *SDϕ* ~*PE*−*H*~), the effect was rather small. If any, crew coordination variability slightly increased towards the higher frequencies (see [Fig 3](#pone.0133527.g003){ref-type="fig"}). This interpretation would be in line with expectations based on the coupled oscillator model (see '[Introduction](#sec001){ref-type="sec"}'). That said, interpersonal pendulum swinging experiments demonstrated that at high movement rates coordinative variability indeed increased with tempo, while at very low movement rates (i.e. rates far below the eigenfrequency of the pendulum) coordination *also* deteriorated \[[@pone.0133527.ref019]\]. This may be taken to suggest the existence of a stroke rate at which crew coordination is optimal and that at lower stroke rates crew coordination might also become less consistent. Note that in the present study we compared ergometer rowing at rather high stroke rates (namely \> 30 *spm*). Regarding on-water rowing, there are indeed some indications that (in-phase) crew synchronization is poorer for stroke rates \< 30 *spm*. A study by Hill \[[@pone.0133527.ref008]\] reported that the mean synchronization offset of the catch was 14.2 *ms* for endurance rowing (23--25 *spm*) and 11.2 *ms* for intensive rowing (31--41 *spm*). Thus, for lower stroke rates, it remains to be examined whether there is a negative (as predicted by the coupled oscillator model \[[@pone.0133527.ref017]\]) or positive relation between stroke rate and coordinative performance. In sum, however, there was no compelling support that steady-state coordinative performance (for both in- and antiphase) substantially changed over different stroke rates, given that the tempo effect on handle coordination was rather marginal and also given the lack of significant tempo effects in *SDϕ* ~*PE*~ and *AEϕ*. For most dyads antiphase crew coordination seemed sufficiently stable to maintain at the highest possible movement rates, which is a promising message for potential on-water application. In line with previous findings \[[@pone.0133527.ref013]\], in-phase crew coordination was more accurate than antiphase crew coordination. Notably, for rates around 36 *spm* (i.e., the rate used in \[[@pone.0133527.ref013]\]) both the variability and accuracy values (see Figs [3](#pone.0133527.g003){ref-type="fig"} and [4](#pone.0133527.g004){ref-type="fig"}) were reasonably smaller than in De Brouwer et al. \[[@pone.0133527.ref013]\], especially for antiphase rowing. A surprising finding was that coordinative variability (as indexed by *SDϕ* ~*PE*~ and *SDϕ* ~*PE*−*H*~) was *not* significantly higher for antiphase coordination, which is counter to previous findings in rowing \[[@pone.0133527.ref013]\] and ample evidence from interpersonal coordination dynamics \[[@pone.0133527.ref004], [@pone.0133527.ref018]--[@pone.0133527.ref020]\]. This finding indicates that the dyads in the present off-water study coordinated the catch in in- and antiphase rowing with comparable consistency. On water, the oars serve as the end-effectors of the rowing movement, as forces are transferred to the water via the blades. Similar to this, on the ergometer, forces are transferred to the flywheel through the handle, which motivated us to analyse the coordination between the handle movements as well. Note, however, that oar(s) are much heavier and involve more inertia than the ergometer handles. As such, it is uncertain whether on-water oar coordination would yield similar results. In the present study, rowers adapted their individual rowing strokes to the antiphase pattern, given that their backward-forward movement ratios suggested more harmonic rowing cycles in the antiphase than in the in-phase condition. Notably, De Brouwer et al. \[[@pone.0133527.ref013]\] found no difference between in- and antiphase crew coordination in this respect. The backward trunk movement (drive phase) was faster than the forward trunk movement (recovery phase), as reflected by the *ratio* values that were generally lower than 1 ([Fig 5](#pone.0133527.g005){ref-type="fig"}). Because for antiphase crew coordination the drive of one rower coincides with the recovery of the other rower, the inherent backward-forward asymmetry (i.e., *ratio* \< 1) in the cycle of each of the two individual components introduces extra fluctuations in their continuous relative phase (see [Fig 7](#pone.0133527.g007){ref-type="fig"}), which is captured in the lower accuracy of coordination in antiphase (as indicated by *AEϕ*, see [Table 2](#pone.0133527.t002){ref-type="table"} and [Fig 4](#pone.0133527.g004){ref-type="fig"}). Hypothetically, if the movements would be perfectly harmonic (although given the nature of the rowing stroke this is practically impossible) all boat velocity fluctuations may cancel out. In this context it is noteworthy that, on average, in antiphase the movements were indeed more harmonic (e.g. indicated by *ratio;* see [Fig 5](#pone.0133527.g005){ref-type="fig"}) than for in-phase crew coordination, indicating that the rowers adapted their rowing strokes in favour of the crew's coordination pattern. For both in- and antiphase crew coordination the trunk movement became more harmonic (indicated by *ratio*) with an increase in stroke rate. For in-phase crew coordination this is in line with indications from on-water rowing which showed that at increasing stroke rates the recovery phase shortens relative to the drive phase (e.g., \[[@pone.0133527.ref008]\]). Because the less harmonic movements at lower stroke rates introduce more boat fluctuations (especially around the catch and finish, see also \[[@pone.0133527.ref013]\]), this may counteract efficiency-related benefits of antiphase crew coordination \[[@pone.0133527.ref013], [@pone.0133527.ref015]\]. In fact, it can be expected that at the lower stroke rates, for instance, during endurance practice (18--24 *spm*) such benefits of antiphase rowing diminish. Because of these and other reasons (see above), it might be interesting to investigate in- and antiphase crew coordination at strokes rates \< 30 *spm*. On mechanical coupling {#sec021} ---------------------- Next to being perceptually coupled, the rowers in the experiment were physically connected via the 'boat'. Mechanical coupling differs from perceptual coupling (e.g., haptic, auditory and visual coupling) in that it is not possible to escape from: the body of each agent gets passively shaken by the movement of the other agent \[[@pone.0133527.ref028]\], whereas perceptual coupling is mediated by the degree to which an agent is sensitive to, or able to detect the pertinent information \[[@pone.0133527.ref046]\], for instance by means of attention devoted to the information source \[[@pone.0133527.ref004], [@pone.0133527.ref046]\]. As an extreme example of the latter, one could simply close the eyes to escape from visual coupling \[[@pone.0133527.ref032]\]. This implies that the mechanical coupling might be more stringent than perceptual coupling. Although many natural tasks involve mechanically coupled coordination, for instance shaking hands, dancing the tango, or jointly moving furniture, (cyclic) coordination of mechanically coupled humans has been relatively unexplored in scientific literature (however, see, e.g., \[[@pone.0133527.ref033], [@pone.0133527.ref047]\]). Note also that a direct physical link between humans always implies kinaesthetic (i.e., haptic and proprioceptive) coupling, but that kinaesthetic coupling can also be present without mechanical coupling. When dyads coordinate their actions on the basis of haptic information see also \[[@pone.0133527.ref048]\], it has been shown that they amplify their forces to generate a haptic information channel \[[@pone.0133527.ref049], [@pone.0133527.ref050]\]. A similar principle may hold true for crew rowing, as Hill \[[@pone.0133527.ref008]\] suggested that an increase in force output in crew rowing provides a better kinaesthetic perception via the boat, which may facilitate the mutual adaption of force patterns. As noted, when antiphase coordination is lost it is challenging to return to this pattern, because once the 'boat' starts oscillating at larger amplitudes it is difficult to counter these movements ([Fig 2](#pone.0133527.g002){ref-type="fig"}). In fact, we might consider this mechanical link to share similarities with the wall vibrations that underlay the synchronisation of two pendulum clocks as Christiaan Huygens observed in 1665. As a result of the exchange of energy through mechanical vibrations via the wall, the initially uncoordinated clocks became coordinated over time in either in-phase or antiphase coordination \[[@pone.0133527.ref051]\]. Relatedly, in experiments with metronomes placed on a freely moving base, transitions from antiphase to in-phase occurred (e.g., \[[@pone.0133527.ref052]\]---many movies are available on-line in which synchronizing of metronomes is demonstrated, of which arguably the most illustrative can be found here [www.youtube.com/watch?v=yysnkY4WHyM](http://www.youtube.com/watch?v=yysnkY4WHyM) and here [www.youtube.com/watch?v=kqFc4wriBvE](http://www.youtube.com/watch?v=kqFc4wriBvE)). In this non-living system, this is due to the mechanical coupling via a third part that can move, namely the freely moving base. It is important to recognize that while in perfect antiphase the metronomes' movements completely cancel out the movement of the base, this does *not* mean that there is no coupling, or less coupling than in in-phase. In fact, in antiphase the *effect* of the coupling is smaller, in that the base does not move, but once it starts moving (due to some internal or external perturbation) the base movements influence the metronomes' movements, causing them to reside into in-phase synchronization with the base and, thus, with each other. Hence, a similar mechanical interpersonal link may form a strong source for attraction to in-phase that is arguably more stringent than for perceptual coupling (cf. \[[@pone.0133527.ref028]\]). In line with this, our study demonstrated that transitions in between-person coordination occur not only in purely visually coupled humans \[[@pone.0133527.ref018]\], but also when they are mechanically linked. Similar to the metronomes, in the present experiment the movements of each rower (cf. metronome) set the 'boat' (cf. base) in motion (see [Fig 2](#pone.0133527.g002){ref-type="fig"}), thereby mechanically influencing the other crew member. In that sense, both in intra- and interpersonal coordination, mechanical coupling may be considered as a source of perturbation that requires anticipatory movements \[[@pone.0133527.ref053]\]. Alternatively, it may also stabilise coordination by constraining the movements of the coupled agents \[[@pone.0133527.ref033]\]. Movements of the ergometer system ('boat') {#sec022} ------------------------------------------ The effect of crew coordination pattern on velocity fluctuations of the 'boat' was evident (see e.g. [Fig 7](#pone.0133527.g007){ref-type="fig"}). Antiphase crew coordination resulted in a reduction of about 60% in velocity fluctuations of the 'boat', consistent with earlier findings \[[@pone.0133527.ref013]\]. Moreover, for in-phase rowing an increase in stroke rate resulted in an increase in velocity fluctuations of the 'boat' (cf. \[[@pone.0133527.ref012]\]), while this was not the case for antiphase rowing. These results suggest that the effect of antiphase crew coordination on boat velocity fluctuations becomes even more beneficial at higher stroke rates. Generalisation to on-water rowing {#sec023} --------------------------------- Although ergometer rowing is not exactly the same as on-water rowing, the results of the present study do allow to be generalised to on-water rowing. Nevertheless, it is important to take into account that, for instance, there are no lateral and vertical (angular) movements of the boat, handles are used rather than oars (which have a certain length and weight), and oar handling technique and hydrodynamics around the blades are not present in an ergometer task. Furthermore, regarding the coupling underlying the synchronisation of the crew, in interviews rowers have reported that they use several information sources to coordinate their movements, such as the looming of the rower in front of them, the haptic channel and physical connection between rowers via the boat, and the flow of water running past the boat \[[@pone.0133527.ref054]\]. Evidently, the latter was not present in the ergometer setup. Nonetheless, although these issues were beyond the scope of the present study, testing crew coordination on-water is the next important step. Conclusion {#sec024} ========== Although the experiment was initially set up to invoke breakdowns of antiphase coordination, only two transitions from anti- to in-phase occurred in at the very beginning of the ramp trial, whereas the other dyads easily maintained the antiphase pattern at increasing stroke rates. A striking finding was that antiphase crew coordination was not significantly more variable than in-phase crew coordination. Together, the results suggested that although antiphase crew coordination is less stable (i.e. less resistant to perturbations) than in-phase crew coordination, it is certainly possible to maintain antiphase rowing at a sufficiently consistent and accurate level at high movement rates. In fact, we only found minor indications that crew coordination may deteriorate at increasing tempos. In addition, antiphase rowing implies a significant reduce in velocity fluctuations of the 'boat', an effect that grows even stronger with increased stroke rate. This might suggest that benefits of antiphase rowing may actually *in*crease with stroke rate. Supporting Information {#sec025} ====================== ###### Data of analysed variables of all dyads. (XLSX) ###### Click here for additional data file. ###### Exemplary movie of a dyad rowing in-phase at 30 and 36 *spm*. (MOV) ###### Click here for additional data file. ###### Exemplary movie of a dyad rowing antiphase at 30 and 36 *spm*. (MOV) ###### Click here for additional data file. We thank Johan van Cuijk, Jacco Verduijn and Dianne de Vette for their help with the data collection. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: HP. Performed the experiments: HP. Analyzed the data: LC HP FZ. Contributed reagents/materials/analysis tools: HP. Wrote the paper: LS FZ HP.

Background ========== Non-response to postal questionnaires is a well known problem that can introduce bias in surveys and in epidemiological studies. Several ways of increasing response rate have been identified, and research has shown that the odds of response can double using monetary incentives \[[@B1]\]. Even small financial incentives are found to be effective in improving physician response \[[@B2]\]. Non-monetary incentives can also be effective, though should be handed out together with questionnaires rather than afterwards \[[@B1]\]. Interventions that trigger positive emotions, such as candy, have also been shown to have an effect on trial participants\' willingness to solve tasks and to increase response rate among physicians \[[@B3]-[@B5]\]. Problems with non-response have been demonstrated in surveys of practice performance in health care \[[@B6],[@B7]\]. The quality of research within this field could be improved by identifying ways to increase response. Therefore, while planning a prospective study of physiotherapy performance in Norway we decided to test the effect of a non-monetary incentive on response. To our knowledge no study has evaluated the effect of chocolate. Thus, the aim of this study was to assess the effect of a bar of dark chocolate on response rate in a study of physiotherapy performance in patients with knee osteoarthritis. Methods ======= In May 2006 all Norwegian physiotherapists in private practice (n = 2798) were invited to participate in a prospective study measuring physiotherapy performance for knee osteoarthritis \[[@B8]\]. Based on feedback from the first invitation, 744 were considered not eligible, mainly because they did not treat patients with osteoarthritis. The remaining physiotherapists were randomly assigned to an intervention group (n = 1027) that received a bar of chocolate together with the data-collection form, and a control group (n = 1027) that received the data-collection form only. The physiotherapists were randomised in blocks of six by a computer generated table. We distributed the forms and chocolates by postal mail including a pre-paid return envelope. The chocolate bar consisted of 36 grams 70% cacao, wrapped in a specially designed sticker bearing survey logo and the text \"Thank you for helping us to document physiotherapy practice\", Figure [1](#F1){ref-type="fig"}. {#F1} The six-page long data-collection form was developed through several steps involving clinicians and experts. It was designed to prospectively report treatments provided to one patient with knee osteoarthritis through 12 treatment sessions. There were three sections including questions about physiotherapist and patient characteristics. After the first mailing all physiotherapists were sent one follow-up reminder by mail and one by e-mail. All practices with more than five physiotherapists were also contacted by telephone. The study period spanned over nine months. The proportion of completed data-collection forms (response rate) was the primary outcome. By assuming a worst case response rate of 20% and with 1094 participants in each arm, the study had 80% power to detect a 5% increase in response rate in the chocolate group. Results ======= We received a response from 510 physiotherapists (236 in the chocolate group and 257 in the no-chocolate group). Some stated that they did not treat patients with knee osteoarthritis or they reported other reasons for not participating, such as not working in clinical practice or focusing on areas like neurology, child or mental health. Among the responders 280 had completed the data-collection form (Figure [2](#F2){ref-type="fig"}). Before the first reminder was sent out we had received 73 completed forms, 39 (3.8%) from the chocolate group and 34 (3.3%) from the no-chocolate group. By end of the study there was no difference between the chocolate and no-chocolate group in the number of completed forms, 142 (13.8%) in the chocolate group and 138 (13.4%) in the control group, ARR = 0.4 (95%-CI: -3.4 -- 2.6). {#F2} Discussion ========== In this study we evaluated the effect of a bar of dark chocolate on response rate in a prospective study of physiotherapy performance. The overall response rate was very low and the chocolate bar did not improve the number of completed data-collection forms. The findings are similar to the study by Halpern et al which found that mints did not influence response rate in a mailed questionnaire among physicians \[[@B5]\], and support findings from a systematic review on effects of incentives to improve response rates to physician surveys that concluded that token nonmonetary incentives were much less effective than even small financial incentives \[[@B2]\]. One explanation to our findings may be the time lapse between receiving the chocolate and performing the requested tasks, or the amount of work requested. The study required subjects to document treatment over a period of several weeks, and chocolate did not seem to have had a strong enough influence or one that lasted long enough to produce the desired effect. All physiotherapists were sent two reminders. These reminders may have prompted both groups to respond equally, cancelling out any effect of the chocolate. The overall response rate was very low in this study although we tried to prevent non-response in different ways. We contacted participants before they received the questionnaire, the questionnaires were sent by first class post and stamped-return envelopes were provided and we sent two reminders \[[@B1],[@B2]\]. It was professionally designed and kept as short as possible. However, if the use of short questionnaires reduces the accuracy of the measurement process, there are trade-offs between non-response and less precise measurement. Conclusion ========== There are many barriers for health professionals in reporting their practice behaviour. Adding one bar of chocolate did not seem to be a sufficiently strong incentive to increase the response rate. Competing interests =================== We enjoy eating chocolate, but still all authors declare that they have no competing interests. Authors\' contributions ======================= SR had the idea about the chocolate incentive. GJ wrote the protocol and designed the study performed the analysis and drafted the first version of the manuscript. KTD contributed to designing the study, entered data into SPSS and revised drafts of the manuscript. SR and SF contributed to the idea of the project and to design and analysis and revised drafts of the manuscript. All authors approved the final manuscript Acknowledgements ================ Thanks to Doris Tove Kristoffersen who generated the randomisation numbers, and to Ola Saatvedt and Saga Høgheim who helped with the mailing. The study was partly funded by The Norwegian Fund for Post-Graduate Training in Physiotherapy.

{.262} {.263}

Background ========== Preclinical studies indicate that most solid tumors require angiogenesis, the formation of new blood vessels from existing vessels, for growth, survival, and metastasis. While many factors regulate tumor angiogenesis, vascular endothelial growth factor (VEGF) appears to have a dominant role, inducing vascular permeability, endothelial cell proliferation and migration, and new blood vessel growth. Numerous drugs have been developed to target the VEGF pathway with receptor tyrosine kinase inhibitors, with soluble decoy receptors, or with antibodies targeting the VEGF ligand or receptor. Inhibition of VEGF signaling reduces tumor growth in many preclinical models \[[@B1],[@B2]\], however, the benefits of targeting VEGF in mouse models have not completely translated to the clinic. While the FDA has approved multiple VEGF pathway inhibitors for clinical treatment of certain cancers, not all patients benefit from these treatments. Some tumors may initially respond but eventually become refractory, while others show no clinical benefit of inhibiting the VEGF pathway \[[@B3]\]. Some preclinical models have even shown resistance and increased metastatic spread associated with VEGF inhibition \[[@B4]-[@B6]\]. Both preclinical and clinical studies have shown that despite significant reductions in tumor blood vessels with VEGF signaling blockade, some tumor blood vessels remain \[[@B7],[@B8]\]. The blood vessels that remain have a distinct phenotype typically associated with more pericyte coverage \[[@B8]-[@B13]\]. There are a number of possible explanations for this effect. First, the initial reduction in tumor blood vessels leads to tumor cell hypoxia, which, in turn, can cause tumor cells to either secrete more VEGF to overcome the anti-VEGF therapy or stimulate the release of other pro-angiogenic cytokines \[[@B4],[@B14]-[@B16]\]. "Vascular normalization" may also play a role in VEGF resistant tumor vessels. VEGF inhibitors can transiently improve pericyte and basement membrane coverage, decrease tumor vessel tortuosity and hyperpermeability, and increase oxygen and drug delivery \[[@B10],[@B17],[@B18]\]. These vessels may be formed via normalization of the atypical phenotype associated with tumor vessels or a pruning of the abnormal vessels leaving behind pre-existing vessels that have a more normal phenotype. Studies by Hal Dvorak's group indicate that tumor blood vessels are heterogeneous consisting of at least six distinctly different blood vessel types: (1) "mother" vessels, (2) glomeruloid microvascular proliferations, (3) vascular malformations, (4) capillaries, (5) feeder arteries (6) and draining veins \[[@B19],[@B20]\]. Interestingly, only subpopulations of these vessels are sensitive to VEGF inhibition. Immunodeficient mice expressing VEGF-A~164~ initially form vessel subtypes such as "mother" vessels and GMPs that are sensitive to VEGF inhibitors while later stage vessels are VEGF-independent \[[@B20],[@B21]\]. Thus, the developmental stage of tumor vasculature is critical to anti-VEGF therapy sensitivity and the lack of good *in vitro* resistance models has slowed the development of non-VEGF anti-angiogenic therapies. In particular, studies should be developed to identify novel ways of targeting the tumor blood vessels that remain or are insensitive to VEGF inhibition. Many *in vitro* assays have been developed that examine multiple steps in the angiogenic process. These assays interrogate sprouting and tip formation, migration and proliferation, lumen formation, and tube or cord formation. *In vivo* assays also look at many of these similar processes. The majority of these assays, however, are driven by the addition of VEGF or other growth factors to the system and remain sensitive to VEGF inhibition \[[@B22]-[@B25]\]. Disrupting established vessels, cords, or tubes which may be insensitive to VEGF inhibitors, however, has not been a major focus of *in vitro* or *in vivo* approaches. Here, we describe an *in vitro* cord formation assay that demonstrates insensitivity to VEGF inhibition. Similar to what is seen *in vivo*, resistance to VEGF inhibition is associated with cord maturity and pericyte association. The advantage of this approach is its increased throughput and ability to identify novel anti-angiogenic agents that can inhibit VEGF-independent vessels. Finally, we show the translatability of this *in vitro* approach using an *in vivo* model of vasculogenesis to validate the effectiveness of novel treatments on the ability to decrease blood vessels that are insensitive to VEGF inhibition. Results ======= Characterization of multiple *in vitro* angiogenesis models ----------------------------------------------------------- Multiple *in vitro* models of angiogenesis or cord formation were examined (Figure [1](#F1){ref-type="fig"}). Traditionally, co-cultures of HUVECs and NHDFs have been used to analyze and quantify growth factor and drug effects on angiogenesis \[[@B26]\]. Recently, a co-culture model of ECFCs and ADSCs, which has a shorter experimental duration and presence of pericyte biology, has been described \[[@B22]\]. In all of the models examined, cord formation occurred in the controls with increased cord formation induced by 20 ng/mL VEGF (Figure [1](#F1){ref-type="fig"}a). We observed a 44% increase in cords in the NHDF/HUVEC co-culture model while there was a 76% increase in cords in the ADSC/ECFC co-culture model at this VEGF concentration (Figure [1](#F1){ref-type="fig"}a). The optimized media used for these assays, however, contain serum and angiogenesis related growth factors such as epidermal growth factor (EGF) and basic fibroblast growth factor (FGF). In order to reduce background cord formation and increase responsiveness to exogenously added angiogenic growth factors, a basal media (BM) was developed which lacks serum and any additional growth factors. When the ADSC/ECFC co-culture was run in BM, the background cord formation decreased by 68% and there was a 194% increase in cord formation with the addition of VEGF (Figure [1](#F1){ref-type="fig"}a). Immunocytochemical characterization showed that cords formed in the ADSC/ECFC co-cultures express multiple markers common to the *in vivo* vasculature \[[@B27]-[@B29]\] (Figure [1](#F1){ref-type="fig"}b). CD31 (PECAM-1), VEGFR-2, and VE-cadherin were expressed by the endothelial cells forming the cords (Figure [1](#F1){ref-type="fig"}b). In addition, only ADSCs that were in close proximity with endothelial cells differentiated into cells expressing SMA and PDGFR-β, indicative of a pericyte-like phenotype \[[@B28]\] (Figure [1](#F1){ref-type="fig"}b, arrows). These pericyte markers were not expressed in the ADSC feeder layer found away from the cords. Finally, vascular basement membrane markers, such as nidogen and type IV collagen, were expressed and associated with the cords in this co-culture system (Figure [1](#F1){ref-type="fig"}b). In contrast, in the NHDF/HUVEC co-culture model, the cords expressed endothelial and basement membrane markers, but pericyte markers were not expressed (data not shown). {#F1} Time course of ADSC/ECFC cord formation --------------------------------------- To further characterize the development of basal and VEGF-induced cords and its associated SMA cells, ADSC/ECFC co-cultures were examined from 0--7 days (Figure [2](#F2){ref-type="fig"}a and b). After 1 day, many of the basal and VEGF-induced cords have already formed. The basal cords were only stable for a few days, before reductions in total tube area were seen beginning around day 3. VEGF-induced cords were stable over the 7 day time course with slight increases in total tube area after the first 24 hours (Figure [2](#F2){ref-type="fig"}a and b). The increase in SMA index was not observed until day 3, and increased dramatically over the next 48 hours (Figure [2](#F2){ref-type="fig"}a and b). These results indicate that VEGF-induced cords are initially formed within the first day. After the first day, however, the cords appear to remodel and may become more stable by the differentiation of the ADSCs into SMA expressing pericyte-like cells. {#F2} Continuous monitoring of cord formation using GFP-expressing ECFCs cultured with ADSCs demonstrated concentration dependent increases in VEGF induced cord formation (Figure [2](#F2){ref-type="fig"}c). Similar to the fixed endpoint studies, the increased VEGF induced cord formation was found in the first 24--36 hours. After 36 hours, the higher concentrations of VEGF-induced cords were stable over the next 3 days (Figure [2](#F2){ref-type="fig"}c). Targeting the components of the cord formation system ----------------------------------------------------- To determine whether VEGF-induced cords can be targeted with anti-VEGF therapy, cords were treated with a receptor tyrosine kinase inhibitor targeting the VEGF receptors among others (sunitinib), an antibody targeting the VEGF-A ligand (avastin; bevacizumab), or an antibody targeting VEGFR-2 (IMC-1121B; ramucirumab) (Figure [3](#F3){ref-type="fig"}a). Blocking VEGF signaling with sunitinib, bevacizumab, or ramucirumab all concentration-dependently reduced VEGF-driven cord formation in the ADSC/ECFC co-culture system. Sunitinib (0.2 μM) maximally decreased VEGF-induced total tube area by 89% (EC~50~ = 0.027 μM), bevacizumab (20 μg/mL) by 65% (EC~50~ = 0.174 μM), and ramucirumab (20 μg/mL) by 80% (EC~50~ = 0.623 μM) (Figure [3](#F3){ref-type="fig"}a and b). In addition to VEGF-induced cords, sunitinib and ramucirumab decreased basal total tube area by 75% and 72% respectively, while bevacizumab only decreased basal cords by 10% (Figure [3](#F3){ref-type="fig"}a and b). While the inhibitors of VEGF signaling decreased cord formation in a concentration-dependent manner, even at the highest concentrations, cords were not completely eliminated. This may indicate that the remaining cords are dependent on other growth factors secreted by the feeder layer of ADSCs. In fact, examination of cords indicate that the basal cords formed in the ADSC/ECFC co-culture system are also dependent on HGF \[[@B22]\] (Additional file [1](#S1){ref-type="supplementary-material"}: Figure S1). {#F3} The ability to target SMA positive pericyte differentiation was also examined. Previous studies indicate that PDGF expression and stimulation of its receptor, PDGFR-β, are important for pericyte recruitment \[[@B30]-[@B32]\]. Here, cords were allowed to form for 4 days to allow for some pericyte differentiation. After 4 days, the cords were treated with 20 μg/mL IgG~1~ or a PDGFR-β blocking antibody (IMC-2C5; \[[@B33]\]). Inhibition of PDGFR-β signaling with IMC-2C5 decreased the SMA index induced by VEGF by 79%, indicating a reduction in pericyte coverage, but the total tube area was not affected (Figure [3](#F3){ref-type="fig"}c and b). Development of VEGF insensitive cords ------------------------------------- To determine whether cords remain sensitive to inhibitors of VEGF signaling, VEGF-induced cords were allowed to establish for 0, 1, 2, or 4 days prior to addition of bevacizumab or ramucirumab. Bevacizumab (20 μg/mL) decreased total tube area by 64% when given on day 0, by 25% on day 1, 13% on day 2, and 16% on day 4 (Figure [4](#F4){ref-type="fig"}a). Cord formation was reduced by 75% when ramucirumab (20 μg/mL) was given on day 0, 33% on day 1, 21% on day 2, and 18% on day 4 (Figure [4](#F4){ref-type="fig"}a). These results indicate that once established, cords become increasingly resistant to VEGF inhibition. To further characterize this insensitivity to VEGF inhibition, continuous live-cell monitoring of the cords after ramucirumab or bevacizumab treatment at day 0 (neoangiogenic mode) or day 4 (established mode) was compared (Figure [4](#F4){ref-type="fig"}b and c). Bevacizumab (25 μg/mL) decreased neoangiogenic cord formation by 70%, but only reduced established cord formation by 20% (Figure [4](#F4){ref-type="fig"}b). Likewise, ramucirumab treatment (10 μg/mL) decreased neoangiogenic cord formation by 90% and only decreased established cord formation by 30% (Figure [4](#F4){ref-type="fig"}c). {#F4} *In vivo* model of VEGF insensitive vessels ------------------------------------------- A limitation of any *in vitro* model is its translatability to *in vivo* biology. Here, to investigate whether similar VEGF-independent vessels can be established *in vivo*, a co-implant model of *in vivo* vasculogenesis with ADSCs and ECFCs was performed. Injection of a mixture of ADSCs and ECFCs in Matrigel develop blood vessels within 3 days. Evidence of blood perfusion (identified with an erythrocyte marker, TER119) in the vessels was seen beginning at 4 days and increased with additional time (Figure [5](#F5){ref-type="fig"}a and b). At 6 days, however, evidence of hemorrhage, or TER119 not associated with blood vessels became evident (Figure [5](#F5){ref-type="fig"}a, arrows). Four-day treatment with ramucirumab beginning on day 0 decreased the percent area of CD31 by 81%. However, when ramucirumab treatment was given for 4 days beginning on day 4, the percent area of CD31 was only reduced by 24% (Figure [5](#F5){ref-type="fig"}c and d); recapitulating the *in vitro* observations. {#F5} Targeting VEGF-independent cords -------------------------------- A number of different mechanisms, including the upregulation of other angiogenic pathways such as the Notch pathway, have been described to play a role in the development of tumor vessels that are insensitive to anti-VEGF therapy \[[@B4],[@B14],[@B15],[@B34]\]. To determine whether other classes of anti-vascular therapy can reduce cords that are insensitive to VEGF inhibition, a broad spectrum anti-angiogenic antagonist (suramin), a vascular disrupting agent (combretastatin), and a combination therapy of a VEGFR-2 inhibitor (ramucirumab) and a gamma secretase inhibitor (GSI; LY411575 \[[@B35],[@B36]\]) were tested on established cords (Figure [6](#F6){ref-type="fig"}). VEGF-induced cords were allowed to establish for 4-days prior to addition of suramin, combretastatin, or the ramucirumab/GSI (LY411575) combination. Suramin treatment (100 μM) decreased VEGF-established cords by 90% (Figure [6](#F6){ref-type="fig"}a), combretastatin (11 nM) decreased cords by 100% (Figure [6](#F6){ref-type="fig"}b), and the combination of ramucirumab (10 μg/mL) and a GSI (LY411575; 10 nM) decreased established cords by 50% (Figure [6](#F6){ref-type="fig"}c). The reduction in established cords with the combination of ramucirumab and a GSI (LY411575) was greater than either of the drugs alone (Figure [6](#F6){ref-type="fig"}c). Similarly, in the co-implant model of *in vivo* vasculogenesis, ramucirumab or the GSI (LY411575) alone led to slight reductions in vessels that were allowed to establish for 4 days prior to treatment. However, the combination of ramucirumab and the GSI (LY411575) almost completely eliminated the vessels (Figure [6](#F6){ref-type="fig"}d). {#F6} Discussion ========== Results from this study indicate that a co-culture system of progenitor cells and endothelial cells can create a cord network with components found in the vasculature: endothelial cells, pericytes, and basement membrane. Exogenously added VEGF can stimulate cord formation and the cords that develop become insensitive to VEGF inhibition as the cords mature. An *in vivo* co-implant model of vasculogenesis, using the same progenitor and endothelial cells as the *in vitro* approach, develops functional blood vessels that anastamose with the host vasculature. These vessels also become insensitive to VEGF pathway inhibition with time. Together, these studies indicate that the combined use of an *in vitro* high-throughput established cord formation assay and an established *in vivo* co-implant model of vasculogenesis can be used to identify novel drugs that can target VEGF-independent blood vessels. Here, we investigate two different *in vitro* models of angiogenesis. The first model uses co-cultures of human umbilical vein endothelial cells (HUVEC) with normal human dermal fibroblasts (NHDF) and the other uses co-cultures of endothelial colony forming cells (ECFC) with adipose-derived stem cells (ADSC). When seeded with NHDF, the HUVECs recapitulate the major phases of the angiogenic process, initially proliferating and migrating into endothelial clusters followed by differentiation and branching into complex networks over the 7--10 day assay. Under basal conditions, little tube formation occurs, while VEGF addition on day 3--4 stimulates tube formation in a concentration-dependent manner \[[@B37]-[@B39]\]. Further characterization of the HUVEC/NHDF approach shows that the pharmacological and physiological effects on endothelial cell biology are highly translatable to previous *in vivo* characterizations, exemplified by DLL4/Notch inhibition resulting in increased branch point formation late in the angiogenic process \[[@B39]-[@B41]\]. One of the major advantages of the ADSC and ECFC co-culture model is that the process of developing cords occurs quickly and incorporates a pericyte-like biology associated with the cords. Unlike the NHDF/HUVEC model and other similar models, in the ADSC/ECFC model, the majority of VEGF-driven cords are formed within the first 24 hours \[[@B23],[@B24],[@B38],[@B42]\]. Additional stimulation allowed for further remodeling of the vessels and differentiation of the ADSCs into SMA or PDGFR-β expressing pericyte-like cells. Characterization of these SMA associated cords indicates that drugs targeting VEGF or PDGF can inhibit the cords or the pericytes, respectively, similar to what has previously been shown *in vivo*\[[@B8],[@B28]\]. Further, VEGF is not the only growth factor that induces rapid cord formation. Other pro-angiogenic growth factors, including FGF and EGF, also induce cord formation in a concentration-dependent manner and exhibit different phenotypes and kinetics (manuscript under preparation). In addition to the assay duration and the cord similarities to *in vivo* vascular structure, the use of basal media in the ADSC/ECFC co-culture assay has dramatically increased the response window for drug screening purposes. Interestingly, we found that inhibition of the VEGF receptor with sunitinib or ramucirumab leads to decreases in basal and VEGF driven cord formation, but inhibition of the ligand with bevacizumab only affected the VEGF driven cords. There may be several explanations for this effect. Endothelial cells can make VEGF and signal in an autocrine fashion. In fact, previous studies using endothelial cell specific knockout of VEGF indicates that autocrine VEGF signaling is required for the homeostasis of blood vessels \[[@B43]\]. It is also feasible that ligand-independent mechanisms or signaling through heterodimerization of VEGFR2 with other receptors may play an important role in basal cord formation \[[@B44]\]. Internalization of the receptor with ramucirumab or the multi-targeted nature of Sunitinib may play a role as well. Finally, VEGF secreted in the co-culture system may get bound to the extracellular matrix, where it may not be accessible to VEGF antibodies, but may still be able to be affected by receptor inhibition. These possibilities require further exploration as it is unclear what mechanism or mechanisms are involved in the ECFC/ADSC co-culture assay. Unlike most tube and cord formation assays, established cords in the ADSC and ECFC co-culture system lose their dependence on VEGF once the cords are developed. Even after 1 day of establishment, the cords are less sensitive to multiple inhibitors of the VEGF signaling pathway. The mechanism of this VEGF-independence is not clear. Pericyte coverage is thought to make vessels insensitive to VEGF inhibition, but in this assay, we see increased VEGF independence after 1 day even though the SMA differentiation does not occur until day 3. In addition, inhibition of the PDGFR after 4 days of establishment decreased the SMA index but did not significantly alter total tube area. While it is still possible that pericytes play an important role in maintenance of established vessels, in this assay system other growth factors secreted by the ADSC feeder layer likely play a major role in maintaining the cords once they have been created. Clearly, there are some cords that can form without the addition of VEGF. Previous studies and our data indicate that HGF is highly expressed by the ADSCs and contributes to basal cord formation \[[@B22]\]. *In vivo* studies show that inhibition of VEGF and the HGF receptor, c-Met, decrease tumor vessels more than VEGF inhibition alone \[[@B29]\]. Together, these results indicate that the HGF secreted by the feeder layer may have an important role in maintenance of the established cords. In addition, we show that suramin, a broad-spectrum antagonist that inhibits various angiogenesis-related growth factors such as insulin-like growth factor, epidermal growth factor, platelet-derived growth factor, VEGF, and basic fibroblast growth factor, is able to reduce the VEGF-independent cords. Together, these results indicate that other growth factors secreted from the ECFCs or ADSC feeder layer may maintain cords following the initial VEGF stimulation. Using an *in vivo* co-implant model of vasculogenesis with ADSCs and ECFCs, functional vessels can form after anastamosing with the host vasculature. These vessels form over 3 days and blood cells labeled with TER119 can be seen beginning 4 days after the cells are injected into the flank. Treatment of the vasculogenic plugs with a VEGF inhibitor dramatically decreases blood vessel formation if given at the beginning of the assay. If, however, VEGFR signaling was not inhibited until the vessels have established and have blood flow (at day 4), there is little effect. While the mechanism of this insensitivity is not know, it would be interesting to characterize our VEGF independent vessels to determine whether they have similar phenotypes as those described by the Dvorak laboratory \[[@B20],[@B21]\]. Nonetheless, these results are consistent with our high-throughput *in vitro* assay and provide a unique *in vivo* model to examine the effects of novel drugs on vessels that are insensitive to VEGF inhibition. As proof of principle, a broad-spectrum anti-angiogenic inhibitor (suramin), a vascular disrupting agent (combretastatin), and a combination of a gamma secretase and VEGF inhibitor were tested on VEGF established cords. Suramin blocks a variety of growth factors including many angiogenesis-related factors. One of the proposed mechanisms of VEGF-independent tumor vessels is that inhibition of VEGF leads to induction of other proangiogenic factors \[[@B4],[@B14],[@B15],[@B34]\]. Suramin is likely able to block many of these other proangiogenic factors to reduce the cords in our established cord assay system. The vascular disrupting agent combretastatin is a microtubule-depolymerizing agent which binds to tubulin dimers to prevent microtubule polymerization. This results in mitotic arrest and apoptosis of endothelial cells. In addition, combretastatin disrupts the endothelial cell junction molecule (VE-cadherin) leading to vascular collapse \[[@B45]\] and *in vivo* studies show that combretastatin is able to reduce immature vessels \[[@B45],[@B46]\]. We observed reductions in established cords with combretastatin treatment. Clearly, while combretastatin may not reduce all mature vessels *in vivo*, it is able to target a unique population of vessels or cords that are insensitive to VEGF inhibition. In fact, preclinical and clinical studies indicate that combining combretastatin with bevacizumab is more efficacious than either inhibitor alone \[[@B47],[@B48]\]. A recent *in vivo* study indicates that VEGF-independent vessels are driven by DLL4-Notch signaling and are sensitive to gamma secretase inhibition \[[@B34]\]. Consistent with this novel strategy to overcome anti-angiogenic resistance, a gamma secretase inhibitor was tested in our *in vitro* and *in vivo* models alone or in combination with inhibition of VEGF signaling. In the *in vitro* system, treatment with either compound alone prevented a slight increase in cords associated with feeding the cells with fresh VEGF, but did not disrupt established networks. However, when VEGF and gamma secretase inhibitors were combined, there was a reduction in the number of cords. Similarly, in the in vivo co-implant model, ramucirumab or the gamma secretase inhibitor alone elicited a slight reduction in the vessels, but the combination reduced the vessels significantly more. These results indicate that our established cord assays may be used to identify new pathways involved in anti-VEGF/VEGFR directed therapy resistance and potential combinatorial strategies. Many current angiogenesis assays used to screen anti-angiogenic agents are highly VEGF dependent. However, from preclinical and clinical analysis, there clearly exists a population of tumor vessels that are insensitive to VEGF inhibition. Thus, angiogenic assays are needed in which novel agents can be tested for their effectiveness on vessels which are not dependent on VEGF. The ECFC/ADSC assay is high throughput and relatively quick. Results can be obtained in approximately a week and can be run in 96-well and 384-well formats and similar co-culture approaches have previously been used in high-throughput drug discovery \[[@B24],[@B49]\]. In addition, labeling the ECFCs with GFP is a feasible approach to monitor cord formation and effects on established cords using continuous live-cell monitoring. Together, these results indicate that a co-culture cord formation system with ADSCs and ECFCs is a useful method to identify and characterize novel drugs on VEGF-independent cords. It would be interesting to identify selective markers on tumor vessels that remain after VEGF therapy and determine if the same markers exist in this co-culture system. If so, these in vitro and in vivo systems would be conducive to interrogate the mechanisms by which vessels become insensitive to VEGF inhibition though use of shRNA/siRNA knockdowns. With more and more studies being published regarding mechanisms of VEGF resistance, additional targets should be tested in this *in vitro* co-culture system. Conclusions =========== Despite in vivo evidence that VEGF independent vessels exist, the majority of the in vitro assays used are dependent on VEGF. We described an in vitro cord formation assay that shows insensitivities to inhibition of the VEGF pathway. In addition, we were able to show the translatability of this assay using an in vivo model of vasculogenesis. Together, the combined use of this *in vitro* high-throughput established cord formation assay and an established *in vivo* co-implant model of vasculogenesis can be used to identify novel drugs that can target VEGF-independent blood vessels. Methods ======= Cell lines and media -------------------- Human adipose derived stem cells (ADSCs) isolated from lipoaspirates collected during surgical liposuction procedures were purchased from Lonza (Allendale, NJ). Cells were grown in EGM2-MV media (Cambrex; Walkersville, MD) and used at passage 4--6. Endothelial colony forming cells (ECFCs) isolated from cord-blood derived endothelial cells were grown on Collagen I coated flasks in EGM2-MV media supplemented with an additional 5% FBS and used at passage 7--10 (Lonza). For studies examining cord formation over time with continuous live-cell monitoring, ECFCs were lentivirally transduced to express CytoLight Green, a soluble variant of GFP, and optimized for imaging in the IncuCyte™ imaging system. Human umbilical vein endothelial cells (HUVECs) and normal human dermal fibroblast (NHDF) cells and media were purchased from Cambrex. HUVECs were grown in EGM media with 10% FBS and NHDF cells were maintained in EGM-2 media. Co-culture assay of endothelial cells and fibroblasts ----------------------------------------------------- HUVEC and NHDF co-culture cord formation assays were performed with AngioKit optimized media (TCS Cellworks, Birmingham, UK) as previously described \[[@B26],[@B37]-[@B39]\]. Briefly, 20K NHDF cells in 100 μL of media were plated in each well of a 96-well plate and incubated overnight at 37°C, 5% CO~2.~ The next day, HUVECs were added on top of the NHDF cells at 1800 cells/well in 100 μL and incubated overnight at 37°C, 5% CO~2.~ On the third day and every subsequent third day, the media was changed to optimized media containing 20 ng/mL VEGF (R&D Systems). On day 10, the co-culture was fixed, stained, and imaged as described below. Neoangiogenic ADSC and ECFC co-culture cord formation assay ----------------------------------------------------------- ADSC and ECFC co-culture assays were performed with AngioKit optimized media, basal media (MCDB-131 medium with 30 μg/mL L-ascorbic acid 2-phosphate, 1 μM dexamethasone, 50 μg/mL tobramycin, 10 μg/mL r-transferrin AF, and 10 μg/mL insulin) or basal media plus (MCDB-131 medium with 0.3% FBS, 30 μg/mL L-ascorbic acid 2-phosphate, 50 μg/mL tobramycin, 10 μg/mL r-transferrin AF, and 10 μg/mL insulin). ADSCs were plated in 96-well plates at 40--50K cells per well in 100 μL and incubated overnight at 37°C, 5% CO~2.~ The next day, the media was removed and 4--5K ECFCs per well in 50--100 μL of media was plated on top of the ADSC monolayer and incubated at 37°C, 5% CO~2~ for 3--6 hours before the addition of growth factors and inhibitors. After the ECFCs attach, growth factors and test agents were added to the 50--100 μL of media at 2--5× to achieve the final concentrations as indicated. Co-cultures were grown for 0--7 days at which time the cells were fixed, stained, and imaged as described below. Established ADSC and ECFC co-culture cord formation assay --------------------------------------------------------- Established ADSC and ECFC co-culture assays were plated as described above for the neoangiogenesis assay. After the ECFCs were allowed to attach, 20 ng/mL VEGF was use to stimulate and establish the cord network. After 1--4 days the media was changed to contain fresh VEGF in the presence or absence of inhibitors at the indicated concentrations. After addition of the inhibitors, cultures were allowed to grow an additional 3--4 days before the cells were fixed, stained, and imaged as described below to investigate network disruption or cord regression. Fixation and staining of fixed endpoint cords --------------------------------------------- At the completion of the assay, ADSC/ECFC cords were fixed and permeabilized with either 70% ice cold ethanol for 20--30 minutes or 3% paraformaldehyde for 10 minutes followed by 70% ice cold ethanol for 20 minutes. Cells were blocked with PBS + 1% bovine serum albumin (BSA) for 30 minutes at room temperature. Primary antibodies were diluted in PBS + 1% BSA and stained either sequentially or in combination for \>90 minutes at 37°C. Endothelial cells were identified with sheep anti-CD31 (PECAM-1; Sigma; 1:200), rabbit anti-VEGFR-2 (55B11; Cell Signaling; 1:50), or goat anti-VE-cadherin (Santa Cruz; 1:50) antibodies. Cy3 conjugated mouse anti-smooth muscle actin (SMA; Sigma; 1:200) and rabbit anti-platelet-derived growth factor receptor beta (PDGFR-β; Y92; LifeSpan; 1:50) antibodies identified pericytes associated with the cords. Vascular basement membranes were identified with goat anti-Nidogen (R&D Systems; 1:50) and goat anti-type IV collagen (Millipore; 1:50) antibodies. After a brief wash, secondary AlexaFluor 488- and 555-conjugated donkey anti-sheep, donkey anti-rabbit, donkey anti-goat secondary antibodies (Invitrogen; 1:400) were incubated for \~60 minutes at room temperature. Nuclei were identified with Hoechst 33342 (Invitrogen; 1:1000) for 5 minutes at room temperature. After Hoechst staining, the cells were washed and imaged as described below. Fixed endpoint imaging and quantification ----------------------------------------- Cord formation images were captured using a Cellomics Arrayscan VTI and analyzed with the Tube Formation bio-application reading at a magnification of 5×. Objects were identified using an algorithm to detect CD31 staining of cords. Total tube area was calculated from 9 fields for each well with 3--4 wells for each treatment. SMA index was calculated from the intensity of the SMA staining and related to the number of cords/tubes. Continuous monitoring of cord formation --------------------------------------- ADSCs and ECFCs transduced with CytoLight Green were seeded for the assay as described above. After 3--4 hours at 37°C, the cells were treated with test reagents (growth factors ± compounds or antibodies), placed into the IncuCyte FLR for imaging, and allowed to form networks over the course of the 4 day experiment. If running the assay in neoangiogenic mode, looking at the inhibition of tube formation, the assay was terminated at the 96 hour time point. If studying tube regression was desired, the assay was run in established mode. To do this, growth factor-driven networks were formed over the first 96 hours of the assay. At this point, a full media replacement occurred including fresh growth factor in the presence or absence of test agent. The assay plate was then placed back in the IncuCyte FLR and imaged over the desired time frame to quantify regression of established networks. For imaging and quantification, phase-contrast and fluorescent images were automatically collected every 6 hours in the IncuCyte FLR to detect network formation using the Tiled Field of View (FOV) mosaic imaging mode. The integrated Angiogenesis Analysis Module was used to identify the fluorescent signal from background in order to quantify multiple assay metrics, such as tube length and branch formation, for each time point. In the first step of the process, the angiogenesis algorithm analyzed each fluorescent image and assigned a segmentation mask that closely resembles the *in vitr*o network. From here, the mask was refined and filtered to exclude non-tube forming events, specifically measuring angiogenesis over time. Kinetic plots of the angiogenesis metrics was generated using the IncuCyte software, allowing for a direct comparison of test agent treatments to validated control conditions (Figure [2](#F2){ref-type="fig"}b). *In vivo* vasculogenesis assay ------------------------------ ADSCs and ECFCs were mixed in a ratio of 1:4 (0.5 × 10^6^/2 × 10^6^ cells/mL) in Matrigel (BD Biosciences) and injected (0.2 mL/implant) subcutaneously into the flank of female athymic nude mice as previously described \[[@B50],[@B51]\]. Three to six days following implantation, implants were collected and placed into zinc-tris fixative. Treatments with IgG or IMC-1121B (ramucirumab; 10 mg/kg, ip) began on day 0 or day 4. Treatments with the gamma secretase inhibitor (GSI, LY411575; 3 mg/kg QD, ip) alone or in combination with ramucirumab began on day 4. The concentration of drugs used was determined from dose response studies (data not shown). Implants were collected and fixed 4 days post treatment and analyzed using multiplexed immunohistochemistry of sections stained for endothelial cells with a CD31 antibody (PECAM; Bethyl; 1:50), erythrocytes with a TER-119 antibody (BD Biosciences; 1:50), and nuclei with Hoechst 33342 (Invitrogen; 1:1000). Quantifications were made using an iCys research imaging cytometer as previously described \[[@B52]\]. Statistical analysis -------------------- All experiments had an n ≥ 3 for each treatment and similar results were seen in at least two experiments. Results are expressed as means ± SEM. Statistical differences were measured by ANOVA with a Tukey posthoc test using JMP software. Abbreviations ============= VEGF: Vascular endothelial growth factor; ADSC: Adipose derived stromal cells; ECFC: Endothelial colony forming cells, HUVEC, Human umbilical vein endothelial cells; NHDF: Normal human dermal fibroblasts; EGF: Epidermal growth factor; FGF: Fibroblast growth factor; BM: Basal media; SMA: Smooth muscle actin; PDGFR-β: Platelet derived growth factor receptor beta; HGF: Hepatocyte growth factor, Bev, Bevacizumab; Ram: Ramucirumab; GSI: Gamma secretase inhibitor. Competing interests =================== BLF, DM, GFE, JS, MLS, YC, LNL, KN, MU, and SC are employees of Eli Lilly and Co. BOC, KA, and DPM are employees of Essen Biosciences. Authors' contributions ====================== BLF played a role in conception and design of the experiments, acquisition, analysis, and interpretation of the data, and writing the manuscript. BOC and KA participated in optimization of live cell imaging and in the acquisition and analysis of the data. DM, MLS, YC, LNL, KN played a role in the conception and optimization of the cord formation assay. GFE developed, optimized, and performed the in vivo vasculogenesis assays and was involved in the design and acquisition of the data. JS was involved in the acquisition, analysis and interpretation of the in vivo vasculogenesis assay. DPM and MTU were involved in the conception and design of the experiments, analysis and interpretation of the data, and in writing the manuscript. SC played a role in conception, design and optimization of the in vitro experiments and in vivo vasculogenesis assays, analysis and interpretation of the data, and in writing the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1: Figure S1 Role of HGF in basal cords CD31 stained basal cords at 3 days treated with 10 μg/mL hIgG or anti-HGF antibody. ###### Click here for file Acknowledgements ================ The authors would like to thank Jonathan Yingling for helpful scientific discussions, Jonathan Lee and Sarah Oliver for assay characterization and helpful discussions and Laura Benjamin and Bronek Pytowski for critical review of the manuscript.

INTRODUCTION ============ Coral snakes constitute a large taxonomic group of more than 120 species and subspecies divided into three genera, *Leptomicrurus, Micruroides* and *Micrurus*, with a distribution ranging from the United States to Argentina ([@CIT0040]; [@CIT0039]; [@CIT0014]; [@CIT0015]). *Micrurus* venoms are highly-neurotoxic, with clinical manifestations of palpebral ptosis, ophthalmoplegia and respiratory paralysis (in severe cases), indicating neuromuscular blockade ([@CIT0048]; [@CIT0016]; [@CIT0049]; [@CIT0007]; [@CIT0032]). Experimental studies have shown that the neuromuscular blockade is caused by pre- and postsynaptic neurotoxins from these venoms ([@CIT0046]; [@CIT0046]; [@CIT0045]; [@CIT0025]; [@CIT0047]; [@CIT0041]; [@CIT0001]). However, only a few of these toxins have actually been purified and studied *in vivo* and *in vitro* ([@CIT0005]; [@CIT0022]; [@CIT0017]), primarily because of the difficulty in maintaining these snakes in captivity ([@CIT0042]; [@CIT0036]) and their low venom yields ([@CIT0018]). Although bites by *Micrurus* spp. in South America are relatively rare, severe cases of respiratory paralysis can be life-threatening if adequate therapeutic interventions are not implemented. Antivenom administration is the only specific treatment for coral snake bites, although ancillary measures, such as, mechanical ventilation and the administration of cholinesterase inhibitors (the latter for postsynaptically-active venoms), can also be useful ([@CIT0013]; [@CIT0048]; [@CIT0007]). Experimentally, the neutralizing capacity of antivenoms against neurotoxins is frequently studied in neuromuscular preparations *in vitro,* following pre-incubation of venom with antivenom ([@CIT0006]; [@CIT0004]; [@CIT0003]; [@CIT0030]; [@CIT0001]). *M. pyrrhocryptus*, which occurs in central Argentina, Bolivia, Paraguay and Brazil (in the states of Mato Grosso and Mato Grosso do Sul), was originally a subspecies of *Micrurus frontalis*, but was elevated to species status based on morphological features that distinguished it from *M. frontalis* (da Silva and Sites, 1999; [@CIT0033]). Little is known about the composition of *M. pyrrhocryptus* venom ([@CIT0031]; de Roodt 2002; [@CIT0021]) and its neutralization by antivenom. In this work, we studied the effects of *M. pyrrhocryptus* venom on neuromuscular transmission in avian and mammalian neuromuscular preparations, and examined the neutralization of neurotoxicity by commercial antivenom and specific antiserum. MATERIAL AND METHODS ==================== Animals ------- Adult male Swiss white mice (28-35gm) were supplied by the Multidisciplinary Center for Biological investigation (Cemib) at Unicamp. HY Line chicks (4-10 days old) were obtained from Globo Aves (Jaguariúna, SP, Brazil). The animals were housed at 24°C with free access to food and water. The experiments were done in accordance with the guidelines of the Brazilian College for Animal Experimentation (Cobea) and were approved by the institutional Ethics Committee on Animal Use (CEUA/Unicamp, Protocol No. 1550-1). Venom, antivenoms and venom neutralization ------------------------------------------ *M. pyrrhocryptus* venom, obtained from snakes captured in Santiago del Estero, Argentina, and specific antivenom were gifts from Dr Alejandro U Vogt (The Centro Zootoxicológico de Misiones, Argentina). The specific antivenom (SAV, batch 115) was raised in horses by the Instituto Nacional de Producción de Biológicos of the Administración Nacional de Laboratorios e Institutos de Salud (A.N.L.I.S.) "Dr Carlos G. Malbrán", Ministerio de Salud (Buenos Aires, Argentina). The venom of *M.* *pyrrhocryptus* is the main immunogen of this antivenom. Commercial *Micrurus* antivenom (Batch No. 03.2184), produced by immunizing horses with a pool of *M. frontalis* and *M. corallinus* venoms, was obtained from the Instituto Butantan (São Paulo, SP, Brazil). Both antivenoms consisted of F(ab')~2~ equine immunoglobulin fragments. The neutralizing capacity of the antivenoms was studied by pre-incubating venom (10μg/ml) with each antivenom for 30min at 37°C, at a venom:antivenom ratio of 1.5mg of venom:1.0ml of antivenom, before adding to the organ bath. Chick biventer cervicis preparation ----------------------------------- Biventer cervicis muscles obtained from chicks previously anesthetized with halothane were mounted as previously described ([@CIT0024]). The preparations were suspended under a resting tension of 1gm in 5ml of Krebs solution of the following composition: 136mM NaCl, 5mM KCl, 2.5mM CaCl~2~, 23.8mM NaHCO~3~, 1.2mM MgSO~4~, 1.2mM KH~2~PO~4~ and 11mM glucose), maintained at 37°C or 22°C (the latter used to verify the involvement of venom PLA~2~ activity in neuromuscular blockade) and aerated with a mixture of 95%, v/v, O~2~ + 5%, v/v, CO~2~. The preparations were stimulated indirectly with supramaximal pulses (6V, 0.2ms and 0.1Hz) delivered by a Grass S4 electronic stimulator (Grass Instrument Co, Quincy, MA, USA) and were allowed to stabilize for at least 15min before the addition of drugs, venom or venom:antivenom mixtures. In some experiments, the muscle contractures to exogenous carbachol (carbamylcholine - CCh, 8μM), acetylcholine (ACh, 110μM) and KCl (20mM) were obtained before and after incubation of the tissues with venom. In order to determine the kinetics of ACh contracture inhibition, preparations under indirect stimulation and incubated with a single venom dose (0.5μg/ml) were assayed in different time points (5, 10, 15 and 30min) (n=5-10 preparations per time interval). Mouse phrenic-nerve diaphragm muscle preparation ------------------------------------------------ Whole diaphragms along with the phrenic nerves were removed from mice anesthetized with isoflurane and sacrificed by exsanguination. The left diaphragm was mounted essentially as described for rats ([@CIT0008]). The preparation was suspended under a constant tension of 5gm in a 5ml organ bath containing aerated (95%, v/v, O~2~ + 5%, v/v, CO~2~) Tyrode solution (pH 7.4, 37°C) of the following composition: 137mM NaCl, 2.7mM KCl, 1.8mM CaCl~2~, 0.49mM MgCl~2~, 0.42mM NaHPO~4~, 11.9mM NaHCO~3~, and 11.1mM glucose. In some experiments, CaCl~2~ was replaced by 4mM SrCl~2~ to assess the influence of venom PLA~2~ activity on the venom-induced neuromuscular blockade. Supramaximal pulses (0.1Hz, 0.2ms, 3-6V) and tetanic stimuli (50Hz, 0.2ms) delivered by a Grass S4 stimulator were applied by electrodes placed on the motor nerve. Isometric muscle tension was recorded using a Load Cell BG 50gm force-displacement transducer (Kulite Semiconductor Products Inc., Leonia, NJ, USA) coupled to a physiograph (Gould RS 3400, Cleveland, OH, USA). The preparations were allowed to stabilize for at least 20min before the addition of drugs or venom. Reversal of neuromuscular blockade by neostigmine, 3.4-diaminopyridine and washing ---------------------------------------------------------------------------------- The reversibility of the venom-induced blockade was assessed by incubating the preparations with neostigmine (10μg/ml) or 3.4-diaminopyridine (10μg/ml), or by extensive washing, after 50% neuromuscular blockade had been achieved. PLA~2~ activity --------------- Venom PLA~2~ activity was assayed in 10mM Tris-HCl, pH 8.0, containing 10mM CaCl~2~, essentially as described elsewhere ([@CIT0001]). The assays were done in triplicate, and the activity expressed as the increase in absorbance at 425nm measured in a multiwell plate reader (SpectraMax 340, Molecular Devices, Sunnyvale, CA, USA). Venom from the South American rattlesnake, *Crotalus durissus terrificus*, was used as a positive control in this assay. *C. d. terrificus* venom was a gift from JC Cogo (Universidade do Vale do Paraíba, São José dos Campos, SP, Brazil). Creatine kinase (CK) activity ----------------------------- Samples of organ bath solution (100μl) were collected before and after a 120min incubation with venom (5μg/ml in BC preparations; CK release by PND preparations was not examined); the initial 100μl aliquot was replaced by fresh solution. The samples were stored at 4°C, and CK activity was assayed within 4hrs after the experiment, using a commercial kit (Sigma Chemical Co, St Louis, MO, USA). CK activity was also assayed in control experiments without venom. Enzyme activity was expressed in international units per liter (IU/l), with one unit of activity corresponding to the phosphorylation of 1nmol of creatine/min at 25°C. Light microscopy ---------------- At the end of the experiments, when complete blockade had been achieved at venom concentrations of 5, 10 and 50μg/ml, chick BC and mouse PND preparations were immediately fixed in Bouin\'s solution and processed for embedding in historesin. Sections 3-5 μm thick were stained with hematoxylin-eosin and examined by light microscopy, using an Olympus microscope (Olympus Optical Co Ltd, Tokyo, Japan) prior to photographing. Muscle damage was quantified by counting 50 fibers (normal or damaged) in four randomly chosen, non-overlapping fields (200 fibers/section) in one section from each of five tissues (experiments) per preparation (total of 1000 fibers each for BC and PND preparations). Similarly, 1000 fibers were counted from five control experiments for each preparation. The percentage of damaged fibers was calculated as (number of damaged fibers ÷ total number of fibers) ×100. Statistical analysis -------------------- The results were expressed as the mean ±SEM and were compared statistically using Student\'s unpaired *t*-test or ANOVA for repeated measures. A value of p ≤ 0.05 indicated significance. RESULTS ======= Blockade of contractile responses --------------------------------- *M. pyrrhocryptus* venom produced time- and concentration-dependent blockade of contractile responses in BC and PND preparations ([Figure 1A](#F1){ref-type="fig"} and [1B](#F1){ref-type="fig"}), with the former preparations being more sensitive to blockade: At venom concentrations of 5μg/ml and 10μg/ml, the time for 50% neuromuscular blockade was 52±0.9min and 22±3min (n=6 each), respectively, in BC preparations; while in PND preparations, it was 110±2min and 62±4min (n=6 each), respectively. In contrast, there was no difference in sensitivity at the highest venom concentration (50μg/ml: 15.1±1.6min *vs* 18.8±2.3min for BC and PND preparations, respectively). Treatment with neostigmine (a cholinesterase inhibitor) and 3-4-diaminopyridine (a potassium channel blocker) produced a transient reversal after 50% of venom-induced neuromuscular blockade (n=6 each; data not shown), followed by complete irreversible blockade after 60min. This finding suggests that when only half of the endplates are affected by venom, some reversibility of blockade is viable. {#F1} The venom (1, 5, 10 and 50μg/ml) inhibited muscle contractures to exogenous acetylcholine (ACh, 110μM) and carbachol (CCh, 8μM) in BC preparations, but did not affect the responses to KCl (20mM) ([Figure 1C](#F1){ref-type="fig"} and [1D](#F1){ref-type="fig"}). Tetanic responses in mouse phrenic-nerve diaphragm preparations --------------------------------------------------------------- In PND preparations exposed to tetanic stimuli (70Hz, 0.2ms), followed by incubation with venom (10μg/ml, n=6), there was a progressive, time-dependent decrease in the amplitude of the responses ([Figure 2A](#F2){ref-type="fig"} and [2B](#F2){ref-type="fig"}) more similar to that seen with α-bungarotoxin (a non-depolarizing toxin, 10μg/ml, n=6) than with succinylcholine (a depolarizing agent, 10μg/ml, n=6; data not shown). {#F2} Kinetics of inhibition of responses to exogenous acetylcholine -------------------------------------------------------------- The incubation of BC preparations with a low venom concentration (0.5μg/ml) for up to 30min resulted in progressive inhibition of the contractile response to exogenous ACh ([Figure 3](#F3){ref-type="fig"}). This finding suggested that the venom contained components that interacted with postsynaptic nicotinic receptors. {#F3} Effect of venom on directly stimulated preparations --------------------------------------------------- Incubation with venom (10μg/ml) did not significantly affect muscle twitches of curarized (d-tubocurarine, 3μM), directly stimulated PND preparations ([Figure 4](#F4){ref-type="fig"}). {#F4} PLA~2~ activity --------------- The PLA~2~ activity of *M. pyrrhocryptus* venom was 0.09±0.04U/mg, approximately one-third that of *C. d. terrificus* (South American rattlesnake) venom (0.30±0.07U/mg; n=6 each, p\<0.05). To examine whether this PLA~2~ activity could contribute to the venom-induced neuromuscular blockade, experiments were done at 22°C (BC preparations) or Ca^2+^ (1.8mM) was replaced by Sr^2+^ (4.0mM) (PND preparations) in order to inhibit PLA~2~ activity. These interventions did not significantly affect the venom potency and time for neuromuscular blockade, indicating that PLA~2~ activity was not involved or had only a minor role in the venom-induced blockade. Creatine kinase (CK) release ---------------------------- Incubation with *M. pyrrhocryptus* venom (5μg/ml) for up to 120min did not significantly alter the release of CK by BC preparations when comparing with the corresponding controls ([Figure 5](#F5){ref-type="fig"}). {#F5} Light microscopy ---------------- Histological analysis of control BC and PND preparations showed normal muscle morphology with little fiber damage (1.2±0.5% in BC and 1.1±0.7% in PND). Incubation with the venom (50μg/ml) did not significantly alter this basal damage (4.3±1.5% and 3.5±0.8% for BC and PND preparations, respectively; n=5 each in all cases). Neutralization by commercial *Micrurus* antivenom and specific antivenom ------------------------------------------------------------------------ Pre-incubation (30min at 37°C) of *M. pyrrhocryptus* venom (10μg/ml) with commercial *Micrurus* antivenom in the proportion recommended by the manufacturer, or with specific antivenom at a venom:antivenom ratio of 1.5mg of venom to 1.0ml of antivenom, totally abolished the venom-induced neuromuscular blockade in both preparations ([Figure 6](#F6){ref-type="fig"}). {#F6} DISCUSSION ========== *M. pyrrhocryptus* venom caused irreversible, time- and concentration-dependent neuromuscular blockade of muscle twitches in BC and PND preparations, with the former preparations being more sensitive than the latter. The greater sensitivity of BC preparations was probably related to differences in the innervation of these two preparations, with avian muscle having both focally- and multiply-innervated fibers that can respond to electrical stimulation or exogenous nicotinic agonists ([@CIT0045]; [@CIT0030]). These findings agree with reports for other *Micrurus* venoms, such as *M. altirostris* ([@CIT0001]), *M. dumerilii carinicauda* ([@CIT0041]), *M. frontalis* ([@CIT0046]; [@CIT0046]; [@CIT0048]), *M. lemniscatus carvalhoi* ([@CIT0009]), *M. nigrocinctus* ([@CIT0025]) and *M. spixii* ([@CIT0047]). The venom inhibited contractures to exogenous ACh and CCh, indicating a predominantly postsynaptic action through the blockade of cholinergic nicotinic receptors, as also suggested for other *Micrurus* venoms ([@CIT0025]; [@CIT0041], [@CIT0001]). The time-dependent blockade of the responses to exogenous ACh shown in [Figure 3](#F3){ref-type="fig"} can only be properly understood when compared with [Figure 1A](#F1){ref-type="fig"}, which shows venom-induced neuromuscular blockade in chick BC preparations. In [Figure 3](#F3){ref-type="fig"}, a venom concentration of 0.5μg/ml produced complete blockade of the responses to exogenous ACh within 30min. In contrast, within a similar time frame of 30min, there was \<10% blockade in indirectly stimulated preparations incubated with the double venom concentration (1μg/ml). These findings agree with the well-known existence of two populations of nicotinic receptors in BC preparations ([@CIT0012]; [@CIT0030]), *i.e*., one population that is extrajunctional and responds well to exogenous ACh but is rapidly blocked by venom neurotoxins, and another located in the motor endplate that responds to nerve stimulation and is less susceptible to rapid blockade by neurotoxins (possibly because of difficulties related to toxin diffusion into the synaptic cleft). They also agree with studies for other neurotoxins that discriminate between these two populations ([@CIT0010]; [@CIT0011]). The ability of toxins to distinguish between these two receptor populations could be exploited to provide a simple, sensitive assay for screening neurotoxins and nicotinic cholinergic agonists. The postsynaptic action of *M. pyrrhocryptus* venom was also indicated by the absence of fade in the tetanic response to indirect stimulation at 70Hz. This tetanic pattern was more similar to that produced by α-bungarotoxin (a non-depolarizing toxin) than by succinylcholine (a depolarizing agent) ([@CIT0023]; [@CIT0043]). *Micrurus* venoms are rich in PLA~2~ ([@CIT0002]; [@CIT0014]; [@CIT0009]; [@CIT0044]) that may contribute to the biological activities of these venoms ([@CIT0005]; [@CIT0022]; [@CIT0035]). As shown here, *M. pyrrhocryptus* venom had low activity when compared with that of *C. d. terrificus*. The finding that reducing the temperature of the experiment from 37°C to 22°C ([@CIT0025]; [@CIT0038]) or the substitution of Ca^2+^ by Sr^2+^ (Rodrigues-Simioni et al, 1995; [@CIT0037]) to attenuate PLA~2~ activity did not significantly alter the neuromuscular blockade indicated that this enzymatic activity was not a major contributor to venom-induced blockade. Previous work ([@CIT0021]) has shown that *M.* *pyrrhocryptus* venom is neither hemorrhagic nor myotoxic. This finding was confirmed here by the lack of effect on muscle responses to exogenous KCl and to direct stimulation in curarized preparations, the unaltered CK release, and the absence of tissue damage seen in histological analysis. In contrast, the intramuscular injection of *Micrurus* venoms in mice and rats results in myotoxicity, seen morphologically and through an increase in serum CK levels (de Roodt AR, 2002). Myonecrosis has been also reported for several coral species ([@CIT0027]; [@CIT0026], [@CIT0028]; [@CIT0025]). These discrepant results may be related to the different animal models (*in vitro* *vs* *in vivo*) and the venom concentrations used. Commercial Brazilian *Micrurus* antivenom raised against *M. corallinus* and *M. frontalis* venoms and specific antivenom effectively neutralized the neurotoxicity of *M. pyrrhocryptus* venom *in vitro*. This finding suggests that it is not necessary to include *M. pyrrhocryptus* venom in the pool of *Micrurus* venoms used for antivenom production. This conclusion differs from findings for *M. altirostris*, another species originally classified as a subspecies of *M. frontalis*, for which commercial Brazilian antivenom showed little neutralization of the lethality ([@CIT0034]) and neuromuscular activity ([@CIT0001]) of the venom. In fact, various studies have suggested the need to include additional *Micrurus* species in the pool of venoms used in immunization protocols in order to improve the neutralization capacity of commercial antivenoms ([@CIT0029]; [@CIT0020]; [@CIT0044]). CONCLUSIONS =========== *M. pyrrhocryptus* venom produced neuromuscular blockade through a predominantly postsynaptic action that was effectively neutralized by commercial and specific antivenoms. The neutralization of neurotoxicity observed here suggests that these antivenoms may be useful in treating humans envenomed by this species. The authors thank Gildo B Leite and Marta B Leonardo for their excellent technical assistance, Miguel Silva and Marcos Silva for caring for the animals, and Dr Stephen Hyslop for suggestions to the text and language revision. This work was supported by grants from the State of São Paulo Research Foundation (Fapesp) and Teaching, Research and Extension Support Fund (Faepex-Unicamp). TMC was supported by n MSc scholarship from the National Council for Scientific and Technological Development (CNPq). MACH and LRS were supported by research fellowships from CNPq. None declared.

Introduction {#S0001} ============ Dendritic cells (DCs) are the only professional antigen-presenting cells (APC) which are both a component of the innate response and an essential element to induce adaptive immunity \[[@CIT0001]\]. They are an important component of the body\'s line of defense, because they are the first to make contact with foreign antigens and then stimulate other cells of the immune system to combat the invaders \[[@CIT0001]--[@CIT0003]\]. Dendritic cells are a heterogeneous population which can be divided into several subtypes which differ according to immunophenotype and function \[[@CIT0001]\]. Conventional DCs (formerly termed myeloid) are characterized by very long cytoplasmic processes which are found on the surface of the cell body and which mainly specialize in absorbing, processing and presenting captured antigens \[[@CIT0002]\]. They play an important role in stimulating lymphocytes, especially naive T lymphocytes \[[@CIT0003]\], which are necessary for immune memory formation and the selective recognition of foreign antigens. Conventional DCs are derived from lymphoid and myeloid progenitors \[[@CIT0004], [@CIT0005]\]. Plasmacytoid DCs are the body\'s main producers of type I interferon, which is related to the anti-viral response \[[@CIT0002]\]. Unlike the conventional type, plasmacytoid DCs demonstrate a reduced ability to present antigens \[[@CIT0001], [@CIT0003]\]. The other three subpopulations of DCs found in human skin are epidermal Langerhans cells, which are characterized by a longer lifespan (weeks) than other DC types (3--10 days) \[[@CIT0001], [@CIT0006]\], dermal myeloid DCs and dermal plasmacytoid DCs \[[@CIT0007]\]. Dendritic cells derive from multipotent CD34+ bone marrow hematopoietic stem cells and, depending on their environment, they may differentiate into distinct subpopulations. However, pathways leading to the generation of DCs have not been fully elucidated \[[@CIT0008]\]. In a steady-state condition, most DCs are generated from progenitors (especially from the myeloid rather than the lymphoid lineage) \[[@CIT0009]--[@CIT0011]\]. Previous studies indicate that during the inflammation process, they may also develop from monocytes and create a subset of inflammatory DCs \[[@CIT0008], [@CIT0012]\]. However, recent findings indicate that DCs can be derived from monocytes also during steady-state conditions \[[@CIT0013], [@CIT0014]\]. Monocytes are most likely not the direct precursor of DCs, but presumably they differentiate into specialized DC subsets under particular conditions \[[@CIT0008]\]. The process of DC maturation is very important to initiate an appropriate response to an infectious agent, because only mature DCs are able to activate other immune cells. Immature DCs are characterized by high phagocytic ability and the presence of multiple receptors that recognize pathogen-associated molecular patterns (PAMPs) \[[@CIT0001]\]. As a result of contact with foreign antigens, DCs mature and migrate to the lymphoid organs, where they present captured antigens and undergo apoptosis \[[@CIT0015], [@CIT0016]\]. Programmed cell death is an important element of the lifespan of DCs, because it regulates the spread of the immune response to invading pathogens by limiting the access of antigens for T cells \[[@CIT0017]\]. Defects in the genes that regulate this process may result in autoimmunity \[[@CIT0017], [@CIT0018]\], whereas if that process is of a significant intensity, it may contribute to a state of immunosuppression, making the body prone to infections \[[@CIT0017], [@CIT0019]\]. There are many mechanisms that initiate this process, but most of them proceed on two main pathways: extrinsic and intrinsic (mitochondrial pathway) \[[@CIT0017], [@CIT0020]\]. Extrinsic pathway of dendritic cell apoptosis {#S0002} ============================================= The extrinsic pathway of apoptosis is mainly based on the interaction of membrane receptors located on the cell surface, called "death receptors", with corresponding ligands which trigger mechanisms leading to programmed cell death (PCD) ([Figure 1](#F0001){ref-type="fig"}). Samples of such proteins are receptors of the tumor necrosis factor superfamily (TNF) such as TNFR1/DR1, TNFR2, Fas/CD95/DR2/Apo1, DR3/Apo3, TRAILR1/DR4/Apo2 or TRAILR2/DR5 and their ligands: TNF-α, FasL/CD95L/Apo1L, Apo3L and TRAIL/Apo2L, respectively \[[@CIT0020]--[@CIT0022]\]. Each death receptor is composed of three parts: the extracellular and transmembrane parts, as well as the cytoplasmic part, which contains the death domain (DD). Connection of the appropriate ligand to the death receptor leads to changes in the DD and binds various proteins, e.g. adapter protein FADD, as well as inactive zymogens of initiator caspases: procaspase-8 or -10 \[[@CIT0020]\]. This process leads to the creation of an active complex, death-inducing signaling complex (DISC), which is capable of performing proteolysis of the mentioned zymogens into their active form. Activation of these enzymes induces an executive caspase cascade, directing the cell to the programmed cell death pathway \[[@CIT0023]--[@CIT0025]\]. {#F0001} The extrinsic pathway may also be linked with the mitochondrial pathway via Bid protein, which may be proteolyzed (as a result of caspase-8 action) to a shorter form, tBid. The shortened form of this peptide moves to the mitochondrial surface and influences release of cytochrome c from these organelles. It irreversibly triggers initiation of the intrinsic apoptosis pathway \[[@CIT0020], [@CIT0024], [@CIT0026]\]. Intrinsic (mitochondrial) pathway of dendritic cell apoptosis {#S0003} ============================================================= The intrinsic apoptotic pathway is dependent on mitochondria and can be activated as a consequence of direct DNA damage, oxidative stress, disturbances in the hydrogen electron transport system or a lack of growth factors \[[@CIT0020], [@CIT0024]\]. As a result of these stimuli, cytochrome c (Apaf 2), an important component of the mitochondrial electron chain, is released into the cytoplasm by megachannels located at the junction of two mitochondrial membranes. The released cytochrome c connects with pro-apoptotic apoptosis protease-activating factor 1 (Apaf 1) and inactive procaspase-9 in a complex called the apoptosome, which is capable of activating caspase-9 ([Figure 1](#F0001){ref-type="fig"}) \[[@CIT0027]\]. This active cysteine protease affects the activity of other executive caspases such as caspase-3, -6 and -7, whose activity results in characteristic morphological changes in apoptotic cells \[[@CIT0024]\]. The intrinsic pathway of apoptosis is also regulated by proteins belonging to the family of Bcl-2 proteins. Some of them, such as Bax and Bak, are pro-apoptotic proteins, whereas Bcl-2 and Bcl-XL have apoptosis-inhibitory effects \[[@CIT0028], [@CIT0029]\]. The purpose of the anti-apoptotic protein is to reduce the semipermeable properties of the mitochondrial membrane by inhibiting the pro-apoptotic proteins that induce its depolarization. Immature DCs have a high level of Bcl-2 protein expression; however, this decreases significantly during cell maturation, which is related to the natural process of DC apoptosis after antigen presentation. This process regulates the lifespan of DCs and prevents intensive and prolonged stimulation of T and B cells \[[@CIT0017], [@CIT0030]\]. Viral factors versus dendritic cells apoptosis {#S0004} ============================================== Viral infections enhance DC apoptosis in many cases, thereby weakening the body\'s major line of defense. Measles infections are often associated with secondary infections, which are a result of the impaired ability of the immune response. Immunosuppression induced by the measles virus can also in many cases lead to death \[[@CIT0031], [@CIT0032]\]. This virus infects both mature and immature DCs, resulting in their apoptosis, which is dependent on Fas receptors and possibly TRAIL receptors \[[@CIT0019], [@CIT0033]\]. During the infection, intensive virus replication has been observed and, at the same time, apoptosis of both DCs and T cells has been seen to increase dramatically. The measles-infected cells may undergo the Fas-dependent pathway of apoptosis or be activated by T cells and become cytotoxic DCs, which are able to inhibit lymphocyte proliferation \[[@CIT0019]\] and even to direct them to the path of TRAIL-dependent programmed cell death \[[@CIT0033]\]. Such cytotoxic measles-infected DCs also undergo apoptosis induced by FasL. Thus, apoptosis facilitates the release of new virions, which replicate inside the DCs and infect other host cells \[[@CIT0019]\]. Another virus that inhibits DC maturation and induces their apoptosis is the Epstein-Barr virus (EBV). The strategy of the EBV is to eliminate DCs, in order to prevent the response of the host, to induce a state of immunosuppression and maintain chronic infection. This virus induces both the extrinsic and intrinsic apoptosis pathways. In a study of apoptosis of DCs derived from cord blood monocytes, Wang *et al*. \[[@CIT0034]\] observed a significant decrease in membrane potential in the mitochondrial inner membrane of EBV-infected cultures. The researchers also observed that the proteins released from the mitochondria, such as cytochrome c, are involved in formation of the apoptosome and consequently increase the activity of caspase-9. Other caspases, whose activities were also increased were caspase-8, which initiates the extrinsic apoptosis pathway, and caspase-3, which is a major executive caspase responsible for chromatin condensation, degradation of DNA and many nuclear substrates, and nuclear envelope budding. The observed changes were confirmed by a significant decrease in expression of the anti-apoptotic proteins Bcl-2 and Bcl-XL, which are responsible for regulation of inner mitochondrial membrane semi-permeability and the strong inhibition of XIAP (X-linked inhibitor of apoptosis protein) protein expression, which stops the apoptosis process by inhibition of caspase-3, -7, and -9 \[[@CIT0035]\]. Li *et al*. \[[@CIT0036]\] also observed that the virus inhibits the development of DCs by promoting apoptosis of their monocyte precursors, thereby preventing initiation and maintenance of the virus-specific immune response. Many studies of the HIV virus show that numbers of both DC subsets (myeloid and plasmacytoid) were dramatically reduced in HIV-infected patients \[[@CIT0037]--[@CIT0041]\]. A study by Laforge *et al*. \[[@CIT0042]\] indicates that rapid DC death occurs in both pathways -- extrinsic (Fas-dependent) and intrinsic. A progressive process of apoptosis in this case is associated with an imbalance between the levels of pro- and anti-apoptotic proteins. The researchers observe that under the influence of HIV, DCs demonstrate increased expression of Bax and Bak proteins, which initiate the apoptosis process, by affecting mitochondrial pore formation, and decrease the expression of proteins that counter this process -- Mcl-1 (a Bak inhibitor) and FLIP (which counteracts DISC complex formation and the activation of caspases). Moreover, it was demonstrated \[[@CIT0042], [@CIT0043]\] that HIV slows down the DC maturation process and impairs the ability of the cells to produce inflammatory cytokines. Both infected DCs and those not attacked by the virus are involved in the process \[[@CIT0042]\]. Studies of herpes simplex virus (HSV) infections demonstrate that the virus induces rapid cell death of DCs \[[@CIT0044]--[@CIT0046]\]. However, DC apoptosis is not always associated with a state of immunosuppression. The ability of DCs to perform this process appears to be one of the anti-viral defense mechanisms of the body. Bosnjak *et al*. \[[@CIT0046]\], in their work about interaction of HSV with DCs, note that rapid apoptosis of these cells under the influence of the virus has a positive effect on activation of defense mechanisms. HSV-infected immature DCs have significantly lower expression of costimulatory and adhesion molecules, reduced capacity to secrete cytokines and an inhibited maturation process. Such cells are incapable of effective presentation and T cell activation, especially cytotoxic T lymphocytes with CD8+ immunophenotype (CTLs). However, research shows that viral antigens derived from the infected DCs, which undergo apoptosis, activate CTLs much more effectively. Due to the ability of DCs to carry out cross-presentation \[[@CIT0047]\] and cooperation with different cell subpopulations, it is possible to defend the body against the infection. During HSV infection, Langerhans cells are the first to be infected. Then they undergo apoptosis and release apoptotic bodies which are engulfed by another DC subset located in the mucosa. Owning to the cross-presentation process, those DCs stimulate mucosal CTLs, which become able to recognize and eliminate the virus \[[@CIT0046]\]. Chronic hepatitis C virus (HCV) infection is characterized by a very weak or unstable antiviral response \[[@CIT0048]\]. Zhao *et al*. \[[@CIT0049]\] found that the ability of myeloid DCs to stimulate T cell responses during the infection was impaired. The researchers observed that myeloid DCs from chronic hepatitis C patients undergo apoptosis more frequently than myeloid DCs from healthy individuals. Furthermore they noted diminished activity of the NF-κB nuclear factor, which is essential for the antigen-presenting function, comprising the expression of human leukocyte antigen (HLA) class II molecules and co-stimulatory molecules, and, in addition, acts as an antiapoptotic factor for myeloid DCs \[[@CIT0049], [@CIT0050]\]. It was also observed that during HCV infection, the DC maturation process was blocked and was manifested by lower expression of maturation surface markers (CD80, CD86 and HLA-DR) \[[@CIT0048], [@CIT0049]\]. According to Sarobe *et al*. \[[@CIT0048]\], expression of HCV E1 protein (CE1) in the DCs of patients with chronic hepatitis C may play a critical role in the inhibition of the maturation of these cells, which is mediated by TNF-α and CD40L. Another study by Zhao and Tyrrell \[[@CIT0051]\] indicates that myeloid DCs from chronic hepatitis C patients have cytotoxic activity which is upregulated to kill T cells during HCV infection. Decreased expression of co-stimulatory molecules together with the cytotoxicity of myeloid DCs and their increased apoptosis were associated with their reduced ability to activate the T cell response and thereby facilitate HCV persistence. However, not all viral infections promote DC apoptosis. Some studies indicate a reduction of this process in cell cultures infected with human respiratory syncytial virus (HRSV), human metapneumovirus (HMPV) and human parainfluenza virus type-3 (hPIV3), which was probably a result of the activation of Bcl-2 family proteins during the process of maturation and a minor cell infection with the virus \[[@CIT0052]\]. Lundqvist *et al*. \[[@CIT0053]\] suggest that two proteins, Bcl-XL and Bcl-2, may be responsible for the inhibition of apoptosis. At the same time they emphasize that Bcl-XL plays the more important role in this process. Because of its increased range of function, Bcl-XL (which is not only associated with the cell membrane, like Bcl-2, but is also present in cytoplasm) can more efficiently connect to the Bax protein (an apoptosis initiator), making it inactive. Bacterial factors versus dendritic cells apoptosis {#S0005} ================================================== As with viruses, bacteria also contribute to increased DC apoptosis in most cases. *Legionella pneumophila*, causing a serious disease called legionellosis manifested by pneumonia, is a dangerous Gram-negative type of bacteria capable of infecting phagocytic cells \[[@CIT0054]\]. The metabolic processes of infected DCs are imbalanced; hence, the infected cells become a place of intracellular multiplication of the bacteria and thus lose the ability to undergo apoptosis. Dendritic cells, unlike other phagocytes, have mechanisms that enable this process to proceed on two different routes. Studies on DCs \[[@CIT0055]\] derived from mouse bone marrow have shown that the first pathway is associated with Naip5 receptor belonging to the group of NOD-like receptors (NLR). Stimulation of this receptor by flagellin (a bacterial protein) leads to the formation of a complex called the inflammasome \[[@CIT0056]\]. At first, the complex activates caspase-1, then it directly activates caspase-7 (belonging to executive caspases), which consequently and irreversibly leads to cell death. As a result of this process, not only is bacterial replication inhibited, but also some cytokines such as IL-1β and IL-18 are secreted outside the cells. The main function of the mentioned cytokines is to attract other immune cells to the site of infection, as well as to stimulate a response directly against the bacteria \[[@CIT0057]\]. The described pathway is called the pyroptosis, and it is a form of apoptosis closely related to the antibacterial response \[[@CIT0055], [@CIT0058]\]. The second apoptotic pathway activated during *L. pneumophila* infection is the intrinsic pathway. A study conducted by Nogueira *et al*. \[[@CIT0055]\] also implicates significant participation of pro-apoptotic proteins Bax and Bak as well as caspase-3 in this process. Fast DC apoptosis in infections plays another very important function: it prevents migration of the infected cells to lymphoid organs and the spread of pathogens in the body. It should also be noted that only the virulent strains of *L. pneumophila* are able to induce apoptosis of DCs within a few hours following the infection, whereas the non-virulent strains only contribute to activation and maturation of these cells, and do not significantly affect the life of DCs \[[@CIT0059]\]. Similar mechanisms of cell death, as described in *L. pneumophila*, also occur in *Salmonella typhimurium* and *Pseudomonas aeruginosa* \[[@CIT0060]--[@CIT0062]\]. Caspase-1-dependent pyroptosis leads to rapid cell death. Moreover, pro-inflammatory cytokines released under the influence of caspase-1 contribute to recruitment of other immune cells to the site of infection \[[@CIT0058]\]. On the one hand, this process is positive, because it mobilizes the body to fight the pathogen, but on the other hand, it allows the bacteria to infect other cells and transports them to the spleen and liver, where it triggers a secondary infection \[[@CIT0061]\]. Another bacterium which very quickly leads to rapid apoptosis of DCs is the Gram-negative *Streptococcus pneumoniae* \[[@CIT0063]--[@CIT0065]\]. Due to this strategy, the bacteria multiply rapidly in mucous membranes during initial stages of colonization, which results in weakening of the immune system. *S. pneumoniae* induces the programmed cell death of DCs via two mechanisms \[[@CIT0063]\]. The first is caspase-independent and occurs with the participation of pneumolysin (a bacterial virulence factor), a protein capable of inducing hemolysis of red blood cells. The second mechanism takes place with the presence of caspases and TLR2 -- the new "death receptor". The pneumolysin-dependent mechanism causes DC death within a short period of time and is associated with changes in cell membrane permeability, which are related to the ability of this protein to form pores in the cell membrane \[[@CIT0063], [@CIT0064]\]. Researchers investigating this subject \[[@CIT0063]\] observed DNA fragmentation in DCs derived from the bone marrow of mice after 3--6 h following the infection, which is typical of the late stages of apoptosis. The pneumolysin derived from more and less virulent bacteria induces DC apoptosis at a similar level of intensity and does not require bacterial internalization into the cell. The second mechanism, unlike the first one described above, acts with a delay of about 24 h after infection, and occurs probably in mature DCs. A study \[[@CIT0063]\] has shown that this process is induced by neither TNF-α nor FasL. It also does not require the direct influence of the bacteria on a cell, but only cell wall components such as peptidoglycan, lipoteichoic acid or the lipoprotein ligands of the TLR2 receptor. The receptor is not only involved in cell signaling activation, but can also initiate the apoptosis process. After binding a suitable ligand to the TLR2 receptor, the apoptosis signal is transmitted via the MyD88 adapter protein and subsequently by another adapter protein, FADD (which connects to the Fas death receptor), then caspase-8 is activated, which results in apoptosis induction \[[@CIT0063], [@CIT0066]\]. *Yersinia enterocolitica* is another example of a Gram-negative bacterium which induces DC apoptosis with a similar mechanism as *S. pneumoniae*. This apoptosis pathway is also TNF-α-independent, although it contains some elements of the TNF-α pathway. According to the researchers studying this problem \[[@CIT0067], [@CIT0068]\], DC death induced by *Y. enterocolitica* is dependent on caspase-8, which is activated by the atypical DISC complex, whereas the TLR4 receptor is involved in bacterial component recognition. Besides apoptosis, which is caspase-dependent, cell death might occur without the presence of caspases, and this is called "necroptosis" or "programmed necrosis". This process, similar in effect to necrosis, is induced by connecting a suitable ligand to a death receptor, and is dependent on RIP1 protein (receptor-interacting protein 1). In the presence of caspase-8, the protein launches either the transcription of NF-κB factor, which helps cells to adapt to environmental stress, or apoptosis, whereas the inactivation of caspases leads to necroptosis \[[@CIT0069]\]. There are many examples of bacteria that lead to increased DC apoptosis, but the mechanisms that induce this process have not been fully investigated. In studies on *Shigella flexneri* \[[@CIT0070]\] it was observed that the OspF protein, which was secreted by the bacteria during an infection, significantly reduced the level of phosphorylation of Erk1/2 kinases (responsible for growth and differentiation), which may prevent transcription of pro-inflammatory genes and the influx of polymorphonuclear leukocytes into the site of infection, and finally lead to the apoptotic death of DCs, which is likely to have adverse consequences for the generation of adaptive immunity \[[@CIT0070], [@CIT0071]\]. However, the interaction of DCs with bacteria does not always lead to increased apoptosis. Studies on the influence of *Listeria monocytogenes* on DCs show that these cells, unlike macrophages, do not undergo programmed cell death after phagocytosis of the bacteria \[[@CIT0072], [@CIT0073]\]. It has been found that a higher pH is present in the phagosomes of DCs than in other cells, and it causes inhibition of the activity of listeriolysin O (a bacterial virulence factor), which under other conditions allows replicating bacteria to escape from the phagosome. Thereby, DCs are able to survive intracellular bacterial infection and receive appropriate signals for cell maturation as well as antigen processing and presentation to T and B lymphocytes \[[@CIT0073]\]. Summary {#S0006} ======= Apoptosis is a very important element which regulates the lifespan of DCs. Therefore, all immunological processes occurring in the body are in a state of homeostasis. However, pathogens have developed a number of strategies that allow the rapid induction of apoptosis in DCs, which prevents antigen presentation to T cells and, thereby, the development of an acquired immunological response. By eliminating DCs, they improve the chances of effective colonization of the host body. However, the rapid apoptosis of DCs does not imply the weakness of the host immune system in the fight with pathogens; it is one of the body\'s defense strategies. The quick death of DCs by programmed cell death effectively inhibits replication of the microorganisms which invaded them. At the same time it prevents the release of a large number of pathogens into the extracellular space, since they are enclosed in apoptotic bodies, which are then ingested by phagocytes. As mentioned previously, DCs present antigens (and also stimulate CLTs) much more effectively when they are absorbed by phagocytosing the apoptotic bodies, rather than when they are in direct contact with pathogens and are infected. In addition, the apoptosis that occurs rapidly in an area of infection prevents the spread of pathogens in the body via infected DCs which migrate to lymph nodes. Otherwise, they could be a true "Trojan horse". Conflict of interest {#S0007} ==================== The authors declare no conflict of interest.

Uterine leiomyomas, or fibroids, are very common benign tumours among reproductive-aged women; nearly 70% are affected by these lesions before age 50 ([@bib4]). Approximately one quarter of the patients suffer from symptoms such as abdominal pain, abnormal uterine bleeding, subfertility, and complications during pregnancy ([@bib23]). Uterine leiomyomas originate from the smooth muscle layer of the uterus and most are formed sporadically. Several independent studies on various populations have shown that ∼70% of uterine leiomyomas harbour specific mutations in mediator complex subunit 12 (*MED12)* ([@bib13]; [@bib7]; [@bib17]). All observed changes have been missense mutations or small in-frame insertions and deletions in exons 1 and 2 affecting a highly conserved area of the gene ([@bib13]; [@bib7]; [@bib17]; [@bib8]). Other recurrently, albeit less frequently, observed alterations include genomic rearrangements affecting *HMGA2* in chromosome 12q15 and *RAD51B* in 14q23-24 ([@bib6]; [@bib22]). Recurrent non-random translocations and deletions have also been observed in other chromosomal areas, such as *COL4A5* and *COL4A6* locus in X chromosome, 6p21 (*HMGA1*), and 7q22 ([@bib25]; [@bib22]; [@bib19]). In addition to sporadic lesions, uterine leiomyomas are a feature of hereditary leiomyomatosis and renal cell cancer syndrome (HLRCC, OMIM \#150800). HLRCC is a tumour predisposition syndrome caused by heterozygous germ-line mutations in fumarate hydratase (*FH*; [@bib10]; [@bib24]). The gene encodes fumarase enzyme, which catalyses the hydration of fumarate to [L]{.smallcaps}-malate in tricarboxylic acid cycle. *FH* is a tumour suppressor, and syndrome-associated lesions display biallelic *FH* inactivation through loss of heterozygosity (LOH) or inactivating mutation in the wild-type allele. In addition to uterine leiomyomas, *FH* mutations predispose individuals to cutaneous leiomyomas, and more rarely, to aggressive renal cell cancer. Based on our recent preliminary results, somatic *MED12* mutations and biallelic *FH* inactivation might exclude each other as drivers of leiomyomagenesis ([@bib15]; [@bib19]; [@bib14]). Here, our aim was to systematically analyse whether these phenomena are truly mutually exclusive and to clarify the role of *MED12* in the tumourigenesis of HLRCC patients\' uterine leiomyomas. We collected a comprehensive series of HLRCC patients\' samples totalling 122 uterine leiomyomas from 27 individuals. Sixty-six sporadic conventional uterine leiomyomas were also included in the study. *MED12* mutation status was analysed by direct sequencing and biallelic *FH* inactivation was determined by 2SC immunohistochemistry, which has proven to be an accurate and unambiguous method to detect FH deficiency ([@bib1]). The effects of *MED12* and *FH* alterations on global expression profiles and clustering of the tumours were also studied. Materials and methods ===================== Subjects and ethical approval of the study ------------------------------------------ Twenty-seven Finnish HLRCC patients, representing 11 families, and 66 individuals with sporadic leiomyomas were included in the study. Altogether 188 uterine leiomyoma samples were collected, 122 tumours from HLRCC patients (89 formalin-fixed paraffin-embedded (FFPE) and 33 fresh frozen samples) and 66 tumours (FFPE samples) from individuals with sporadic uterine leiomyomas. Samples were collected either after signed informed consent or after the approval by the National Supervisory Authority for Welfare and Health. Sporadic samples were collected and anonymised after the approval by the director of the health care unit. The study was approved by the Ethics Review Board of the Hospital District of Helsinki and Uusimaa (HUS), Helsinki, Finland. DNA/RNA extraction and sequencing --------------------------------- Genomic DNA was extracted from the FFPE samples with a standard phenol--chloroform extraction method or with NucleoSpin DNA FFPE XS Kit (Macherey-Nagel GmbH & Co KG, Düren, Germany), and from fresh frozen tissue samples with FastDNA Kit (MP Biomedicals LLC, Solon, OH, USA). Total RNA was extracted with TRI Reagent (Molecular Research Center Incorporated, Cincinnati, OH, USA). *MED12* and *FH* mutation screenings were performed by direct sequencing. Sanger sequencing was carried out at the Institute of Molecular Medicine Finland, Technology Center, Helsinki, Finland, using an Applied Biosystems ABI3730 Automatic DNA Sequencer (Life technologies, Thermo Fisher Scientific, Waltham, MA, USA). Sequences were analysed both manually and with the Mutation Surveyor software (Softgenetics, State College, PA, USA). *MED12* NM_005120.2 and *FH* NM_000143.3 were used as reference sequences. Primers used in the sequencing are listed in the [Supplementary Table 1](#sup1){ref-type="supplementary-material"}. Loss of heterozygosity analysis ------------------------------- LOH at the *FH* locus was assessed from fresh frozen tumour samples harbouring both a germ-line *FH* mutation and a somatic *MED12* mutation. LOH was analysed by sequencing five independent PCR reactions and comparing the heights of wild-type and mutant peaks in the chromatograms. LOH was scored when the height of wild-type allele peak was repeatedly lower than the mutant allele peak. Tissue microarray construction ------------------------------ To determine the most suitable areas of the samples for tissue-microarray (TMA), haematoxylin and eosin staining was performed to sections cut from all FFPE samples. Tissue samples were analysed and representative tumour areas were marked by a gynaecological pathologist (RB). Four 0.8-mm tumour cores were punched from the original sample block with a manual tissue arrayer (MTA-I, Beecher Instruments, Sun Prairie, WI, USA) and inserted in TMA paraffin block. Normal myometrium samples were included in the TMA as internal controls. Immunohistochemistry -------------------- Biallellic inactivation of *FH* was assessed with 2SC immunohistochemical staining performed with EnVision + kit (Dako, Carpinteria, CA, USA). Staining is based on the recognition of S-(2-succinyl) cysteine modified (succinated) proteins which are formed in FH deficient cells as a result of the accumulation of fumarate ([@bib21]; [@bib1]). Tissue sections (5 *μ*m) were incubated with the anti-2SC antibody (1 : 2000) at +4 °C, o/n (FFPE samples) or at room temperature for an hour (fresh frozen samples). Antibody binding was detected with anti-rabbit horseradish peroxidase polymer. Stainings were assessed by a pathologist (RB) and the sample was scored as positive (+) when the great majority of cells displayed both nuclear and cytoplasmic staining indicating accumulation of fumarate and succinated proteins and thus biallelic inactivation of *FH*. Samples displaying no staining or rare samples showing low cytoplasmic positivity in single cells were scored as negative (−), indicating the cells to retain sufficient FH activity. Gene expression profiling and hierarchical clustering analysis -------------------------------------------------------------- Gene expression profiles of 14 leiomyomas from four HLRCC patients and three FH-deficient sporadic tumours and their corresponding normal myometrial tissue samples were analysed together with 58 uterine leiomyomas and corresponding normal samples from which we had both expression and whole-genome sequencing data in-house (19 *MED12* mutation-positive and 39 *MED12* and *FH* wild-type tumours) ([@bib19], [@bib18]). Gene expression data was constructed using Affymetrix GeneChip Human Exon 1.0 ST Arrays (Affymetrix) at the Biomedicum Functional Genomics Unit, Helsinki, Finland. The expression data was analysed with Partek Genomic Suite v. 6.5 (Partek Incorporated, St Louis, MO, USA) using re-mapped Brainarray Custom CDF files (HuEx10stv2_Hs_ENSG, Version 16). All samples were quantile-normalised by the Robust Multichip Average method and adjusted for probe sequence and GC-content. Unsupervised hierarchical clustering analysis (Cosine dissimilarity) was performed with 1% most variable genes (*n*=372), defined by the coefficient of variation calculated across all tumour samples. Statistical analyses -------------------- R software, version 3.0.2 (R Foundation for Statistical Computing, Vienna, Austria, [www.r-project.org](https://www.r-project.org)), and Python, version 2.7 (Python Software Foundation, Wilmington, DE, USA, [www.python.org](https://www.python.org)), were utilised for statistical analyses. Differences in the frequencies of *MED12* mutations and biallelic *FH* inactivation between uterine leiomyomas from HLRCC patients and sporadic conventional uterine leiomyomas were calculated using Fisher\'s exact test (two-sided *P*-value). Mutual exclusiveness of *MED12* mutations and FH deficiency was evaluated utilising a permutation test. A total of *n*=1 000 000 permutations were performed and mutations were randomly redistributed between samples in each permutation. Empirical *P*-value was computed as *P=*(1+*k*)/*n*, where *k* is the number of permutations where at least one sample presented both *MED12* mutation and FH deficiency. The statistical significance of many *MED12* mutation-positive tumours co-occurring in HLRCC patient My31 was evaluated similarly as above by randomly redistributing tumours to patients (*n*=10 000 000 permutations). Empirical *P*-value was then calculated as *P=*(1+*k*)/*n*, where *k* is the number of permutations where at least one patient with six or more *MED12* mutation-positive tumours and no FH-deficient tumours was observed. Results ======= Mutation screening ------------------ One hundred and twenty-two uterine leiomyomas from 27 HLRCC patients were screened for somatic mutations in exons 1 and 2 of *MED12*. Of these, 34 tumours had been previously screened for exon 2 mutations ([@bib15]). Both exons were successfully sequenced in 83 FFPE samples and in all 33 fresh frozen samples (116/122; 95%). The quality of DNA obtained from six FFPE blocks of patient D6 (dating back to year 1979) was not sufficient for sequencing, and these samples were excluded from mutation screening. Altogether nine tumours from four different HLRCC patients had *MED12* mutations (9/116; 7.8%) ([Supplementary Table 2](#sup1){ref-type="supplementary-material"}). Patients B3, B7, and E1 had one *MED12* mutation-positive tumour each, while the majority of their tumours were *MED12* wild type (1/3, 1/6, and 1/6, respectively) ([Table 1](#tbl1){ref-type="table"}). Surprisingly, all six uterine leiomyomas from patient My31 harboured different *MED12* mutation. Coding exons of *FH* were sequenced from all these six tumours, but no second hit was observed ([Table 1](#tbl1){ref-type="table"} and [Figure 1](#fig1){ref-type="fig"}). Occurrence of this many sporadic *MED12* mutation-positive lesions in HLRCC patient is highly unlikely (*P=*2 × 10^−7^, permutation test with 10 000 000 permutations). Sixty-four sporadic conventional uterine leiomyomas (64/66; 97%) were successfully sequenced for *MED12* exons 1 and 2 (45 reported previously in ([@bib15]). Of these, 35/64 (55%) harboured a somatic *MED12* mutation ([Supplementary Table 3](#sup1){ref-type="supplementary-material"}). Immunohistochemistry -------------------- All 122 HLRCC-associated uterine leiomyomas and 66 sporadic tumours were successfully analysed for FH deficiency utilising 2SC immunohistochemistry. Positive staining indicating biallelic *FH* inactivation was detected in 113 uterine leiomyomas from HLRCC patients (113/122; 92.6%, including six tumours from patient D6 excluded from *MED12* mutation screening). Nine uterine leiomyomas (one tumour from patients B3, B7, and E1 each, and all six tumours from My31) displayed negative 2SC immunostaining implicating the other *FH* allele to be intact and tumours to be proficient for FH activity (9/122; 7.4% [Figure 2](#fig2){ref-type="fig"} and [Supplementary Table 2](#sup1){ref-type="supplementary-material"}). These same tumours were observed to harbour somatic *MED12* mutations in the mutation screening ([Table 1](#tbl1){ref-type="table"}). All 66 sporadic tumours showed negative 2SC staining ([Supplementary Table 3](#sup1){ref-type="supplementary-material"}). Based on mutation screening and results from 2SC immunohistochemistry, the difference in *MED12* mutation and biallelic *FH* inactivation frequencies between uterine leiomyomas from HLRCC patients and sporadic conventional uterine leiomyomas is highly significant (*P*\<2.2 × 10^−16^). Data also shows these events to be mutually exclusive (*P=*1 × 10^-6^, permutation test with 1 000 000 permutations). Gene expression profiling ------------------------- Gene expression profiling of 26 *MED12* mutation-positive, 10 FH-deficient, and 39 *MED12* and *FH* wild-type tumours and their corresponding normal myometrium tissue samples revealed that the majority of leiomyomas clustered according to the mutation status of *FH* and *MED12* ([Figure 3](#fig3){ref-type="fig"}). Leiomyomas harbouring both a germ-line *FH* mutation and a somatic *MED12* mutation displayed expression signatures similar to those with only a *MED12* mutation, instead of clustering with the FH-deficient leiomyomas. All myometrium tissue samples clustered together regardless of their germ-line mutation status. Discussion ========== Most uterine leiomyomas develop sporadically and in the majority (∼70%) tumourigenesis is driven by very specific mutations in exons 1 or 2 of *MED12*. Simple chromosomal changes and chromothripsis-like, more complex chromosomal rearrangements affecting specific loci such as *HMGA2* (chromosome 12) or *COL4A5* and *COL4A6* (chromosome X), have also been suggested as driver aberrations in these tumours ([@bib25]; [@bib22]; [@bib19]). Thus far, none of these alterations have been reported to occur simultaneously in the same lesion ([@bib16]; [@bib19]; [@bib3]; [@bib18]). HLRCC syndrome, caused by heterozygous germ-line mutations in *FH*, is one of the few known hereditary conditions that predispose women to uterine leiomyomas. Loss of the wild-type allele leads to *FH* inactivation and subsequent tumourigenesis. To determine the frequency of *MED12* mutation-positive tumours among HLRCC patients, and to analyse whether somatic *MED12* mutations and biallelic *FH* inactivation are mutually exclusive, we performed *MED12* exons 1 and 2 mutation screening and 2SC immunohistochemical staining indicative for FH deficiency, in both HLRCC-associated and sporadic uterine leiomyomas. In a comprehensive series of 122 HLRCC-associated uterine leiomyomas, 116 tumours were successfully analysed with both methods. The great majority of tumours (107/116; 92%) were wild type for *MED12* and showed 2SC positivity indicating them as syndrome-associated FH-deficient lesions. Twenty-three out of 27 HLRCC patients (23/27; 85%) had only FH-deficient uterine leiomyomas, with number of tumours ranging from 1 to 15 per patient. Nine uterine leiomyomas (9/116; 7.8%) from four different HLRCC patients had somatic *MED12* mutations. All these tumours displayed negative 2SC immunohistochemical staining indicating FH proficiency; these tumours are most likely sporadic. Three patients had only one *MED12* mutation-positive uterine leiomyoma in addition to several syndrome-associated tumours, while all six tumours of patient My31 were *MED12* mutation positive. Overall, these results confirm previous observations that most uterine leiomyomas in HLRCC patients are syndrome-associated and result from biallelic *FH* inactivation. Incidental sporadic tumours do, however, occur in HLRCC patients, yet these lesions account for only a small proportion of tumours. HLRCC patients tend to develop uterine leiomyomas at younger age than women with sporadic tumours ([@bib11]). Syndrome-associated tumours are also suggested to be more numerous and symptomatic, leading to hysterectomies at an earlier age, among women with HLRCC ([@bib11]). This might, to some extent, explain the rarity of sporadic lesions, which are known to reach the incidence peak towards the end of women\'s reproductive age. Another possible explanation is purely stochastic: as the remaining *FH* allele can be inactivated through several distinct mechanisms (missense or nonsense mutation, small insertion, or deletion leading to a shift in the reading frame, larger chromosomal aberration), it may be that one of these events happens more likely than a single, highly specific *MED12* mutation. Altogether 66 sporadic conventional uterine leiomyomas were included in the study. Sixty-four were successfully analysed for *MED12* mutations and the majority (35/64; 55%) were mutation positive. This is in agreement with previous studies, where mutation frequencies have varied between 48 and 92% ([@bib20]). Also the mutation spectrum with 26 highly specific missense mutations and 9 small in-frame deletions resembles previous observations. All sporadic tumours showed negative 2SC immunostaining and were thus FH proficient. This does not exclude the possibility of an underlying heterozygous germ-line *FH* mutation or the presence of one somatic alteration in some of the samples, but it shows that biallelic *FH* inactivation is a rare molecular driver event in sporadic uterine leiomyomas. This is in line with previous analyses, which have shown that the contribution of *FH* alterations on sporadic uterine leiomyomas and sporadic counterparts of other HLRCC-associated tumours is in a minor role ([@bib9]; [@bib12]; [@bib19]; [@bib5]). Taken together, our results on an extensive HLRCC patients\' sample series and sporadic tumours confirm the previous observations that biallelic *FH* inactivation and *MED12* mutations are mutually exclusive. When sporadic *MED12* mutation-positive tumours were observed in HLRCC patients, these were typically individual lesions among several FH-deficient tumours. Interestingly, all six tumours from one HLRCC patient, My31, showed FH proficiency and instead of losing the wild-type *FH* allele, each tumour harboured a *MED12* mutation. All tumours displayed different *MED12* mutations thus excluding the possibility of tumour dissemination. These six tumours had been obtained at hysterectomy, which was performed when the patient was 49. She had been under surveillance due to uterine leiomyomas since the age of 28, and several lesions had been removed in two myomectomies at ages 36 and 40. We were able to collect 24 FFPE uterine leiomyoma specimens from these operations, and we analysed them with direct sequencing and 2SC immunohistochemistry. Twenty-three out of 24 specimens (96%) were considered to represent sporadic tumours due to both *MED12* mutation and negative 2SC staining. Only one tumour was wild type for *MED12* and showed positive 2SC staining, and was thus considered syndrome-associated. The germ-line *FH* mutation in patient My31 is a nonsense mutation affecting the last exon of the gene (c.1439C\>G; Ser480X). The same mutation has been previously observed in another Finnish HLRCC patient with both uterine and cutaneous leiomyomas (unpublished data). This patient had a hysterectomy due to multiple leiomyomas in 1968 at the age of 31. Uterine leiomyoma samples were no longer available, but we were able to collect nine cutaneous leiomyoma samples from the pathology department\'s archives. All these tumours were FH deficient and thus considered syndrome-associated. In addition, other syndrome-associated nonsense and frameshift mutations have been reported in the last exon of *FH*, including one affecting Ser480 and others affecting nearby amino acids Lys477, Gly490, and Trp500 ([@bib2]). Intriguing and highly unlikely (*P=*2 × 10^−7^) observation of several sporadic *MED12* mutation-positive lesions in HLRCC patient My31 implicates the presence of an additional genetic factor that either protects the patient from the loss of the *FH* wild-type allele or predisposes her to somatic *MED12* mutations. Unsupervised hierarchical clustering analysis of gene expression data revealed that FH-deficient and *MED12* mutation-positive leiomyomas display distinct expression profiles. This is not surprising considering the severe metabolic stress *FH* inactivation exposes the cells to, and similar results have been reported also previously ([@bib26]; [@bib19], [@bib18]). We included both HLRCC syndrome-associated and sporadic FH-deficient tumours in the analysis, and all of these tumours clustered together irrespective of whether the inactivation had occurred through one germ line and one somatic or through two somatic mutational events. Interestingly, all six tumours of patient My31, which harbour both one germ-line *FH* mutation and a somatic *MED12* mutation, clustered together with *MED12* mutation-positive tumours and displayed expression signatures typically seen in sporadic *MED12* mutation-positive leiomyomas. Also all normal tissue samples clustered together irrespective of germ-line *FH* mutation status. These results confirm that one active *FH* allele is sufficient to maintain the metabolic balance within the cell. To conclude, HLRCC patients\' uterine leiomyomas can be formed through at least two distinct molecular mechanisms, biallelic *FH* inactivation and somatic *MED12* mutations. The great majority of HLRCC patients\' tumours use the former pathway, while *MED12* mutations are the most frequent alterations in sporadic tumours. Distinct molecular mechanisms within HLRCC patients\' uterine leiomyomas and sporadic tumours may affect treatment options of these patients in the future. Our results also confirm that biallelic *FH* inactivation and *MED12* mutations are mutually exclusive both within HLRCC syndrome associated and sporadic uterine leiomyomas. We thank Sini Nieminen, Inga-Lill Svedberg, Iina Vuoristo, Alison Ollikainen, and Elina Pörsti for their excellent technical assistance. This study was supported by the Academy of Finland (Finnish Center of Excellence in Cancer Genetics Research, grant number 250345 for LAA, and Academy Research Fellow grants 260370 and 292769 for PV), the Sigrid Jusélius Foundation, the Cancer Society of Finland, and the Biomedicum Helsinki Foundation. [Supplementary Information](#sup1){ref-type="supplementary-material"} accompanies this paper on British Journal of Cancer website (http://www.nature.com/bjc) This work is published under the standard license to publish agreement. After 12 months the work will become freely available and the license terms will switch to a Creative Commons Attribution-NonCommercial-Share Alike 4.0 Unported License. The authors declare no conflict of interest. Supplementary Material {#sup1} ====================== ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. {#fig1} {#fig2} {#fig3} ###### *FH* and *MED12* mutation status of 21 uterine leiomyomas from four HLRCC patients with at least one *MED12* mutation-positive tumour **Patient** **Tissue material** **Germ-line** ***FH*** **mutation** ***MED12*** **mutation status** **2SC IHC** ------------- --------------------- ------------------------------------- --------------------------------- ------------- B3 FFPE c.671_672delAG, p.E224fs Wt \+   FFPE c.671_672delAG, p.E224fs c.131G\>A, p.G44D **−**   FFPE c.671_672delAG, p.E224fs Wt \+ B7 Fresh frozen c.671_672delAG, p.E224fs c.113-160del48, p.A38_G53del **−**   Fresh frozen c.671_672delAG, p.E224fs Wt \+   Fresh frozen c.671_672delAG, p.E224fs Wt \+   Fresh frozen c.671_672delAG, p.E224fs Wt \+   Fresh frozen c.671_672delAG, p.E224fs Wt \+   Fresh frozen c.671_672delAG, p.E224fs Wt \+ E1 FFPE c.587A\>G, p.H196R Wt \+   FFPE c.587A\>G, p.H196R Wt \+   FFPE c.587A\>G, p.H196R Wt \+   FFPE c.587A\>G, p.H196R Wt \+   FFPE c.587A\>G, p.H196R Wt \+   FFPE c.587A\>G, p.H196R c.130G\>A, p.G44S **−** My31 Fresh frozen c.1439C\>G, p. S480X c.130G\>A, p.G44S **−**   Fresh frozen c.1439C\>G, p. S480X c.131G\>T, p.G44V **−**   Fresh frozen c.1439C\>G, p. S480X IVS1-1_139del41 **−**   Fresh frozen c.1439C\>G, p. S480X IVS1-8T\>A, p.E33_D34insPQ **−**   Fresh frozen c.1439C\>G, p. S480X c.131G\>A, p.G44D **−**   Fresh frozen c.1439C\>G, p. S480X c.130G\>C, p.G44R **−** Abbreviations: *FH*=fumarate hydratase; FFPE=formalin-fixed paraffin-embedded; HLRCC=hereditary leiomyomatosis and renal cell cancer; IHC = immunohistochemistry; *MED12*=mediator complex subunit 12; wt=wild type.

Introduction {#Sec1} ============ Plants employ different types of defences to resist herbivores. Such defences can be displayed constitutively or can be induced. In general, constitutive defences are present all the time, and induced plant defences are reinforced or activated by herbivory, involving the production of signalling molecules, which results in the upregulation of biosynthesis of specific compounds such as toxins and digestion inhibitors (Karban and Baldwin [@CR33]; Kant et al. [@CR31]). These compounds act directly against herbivores, reducing their growth or survival or reproductive rate (Howe and Jander [@CR25]; Kant et al. [@CR31]). Plants can also defend themselves indirectly by involving the natural enemies of the herbivores (Price et al. [@CR55]), for example, by providing them with food or shelter, or by attracting them with volatiles induced by herbivore damage (Sabelis et al. [@CR62], [@CR63]). Induced plant defences can be local, i.e. expressed at the damaged site, and systemic, i.e. expressed in plant parts that were not damaged by the herbivores. At the damaged sites of the plant, signals may be produced and transmitted systemically to undamaged distal sites, resulting in priming or induction of plant defences in those sites, which provides resistance to future herbivore attacks (Karban and Baldwin [@CR33]; Howe and Jander [@CR25]; Heil and Ton [@CR23]; Conrath et al. [@CR13]). Phytohormones play important roles in regulating induced defences (Pieterse et al. [@CR53]). The most important hormones involved in induced defences are jasmonic acid (JA), salicylic acid (SA) and the hormone ethylene (Erb et al. [@CR18]; Pieterse et al. [@CR54]). Plant defences can differ qualitatively and quantitatively with the herbivore species attacking the plants (de Vos et al. [@CR17]; Rodriguez-Saona et al. [@CR61]) and time since attack (Kant et al. [@CR29]). Moreover, different defence pathways can affect each other either positively or negatively (Koornneef and Pieterse [@CR37]). Therefore, herbivores sharing the same plant can affect each other through the defences they induce (Rodriguez-Saona et al. [@CR60]; Kaplan et al. [@CR32]). Much research has been conducted into such indirect herbivore interactions mediated via induced plant defences, but little information is available on the induction of plant defences by omnivorous predators and their effects on herbivore performance on the same plant. Omnivorous predators are important in natural ecosystems (Coll and Guershon [@CR12]) and are increasingly used as biological control agents because they can feed on plant material and their populations can thus persist in the crop when pest densities are low, making the system more resilient (Messelink et al. [@CR40]). However, some omnivores can also damage the plants (Puentes and Björkman [@CR56]), hence these omnivores are less useful for biological control. Plant feeding by some omnivores is known to activate plant defence mechanisms. For example, the predatory bug *Orius laevigatus* was shown to increase resistance of tomato plants against thrips and whiteflies (De Puysseleyr et al. [@CR16]). In addition, Pérez-Hedo et al. ([@CR52]) and Naselli et al. ([@CR44]) have shown that exposing tomato plants to the mirid bug *Nesidiocoris tenuis* resulted in the activation of the abscisic acid (ABA) and jasmonic acid (JA) signalling pathways involved in plant defences, but not the SA signalling pathway, whereas *Macrolophus pygmaeus* activated only the JA pathway and not the ABA pathway (Pérez-Hedo et al. [@CR51]). Pappas et al. ([@CR47]) found that exposing tomato plants to the omnivorous predator *Macrolophus pygmaeus* reduced the performance of a subsequently infesting herbivore, the two-spotted spider mite *Tetranychus urticae*, but not of the greenhouse whitefly *Trialeurodes vaporariorum*. Here, we further investigated the effects of plant feeding by *M. pygmaeus* on herbivore performance and plant hormone levels in another plant species, sweet pepper (*Capsicum annuum* L.). We specifically tested whether this effect was local or systemic. *Macrolophus pygmaeus* is an important omnivorous predator of several agricultural pests such as whiteflies (Montserrat et al. [@CR42]), thrips (Riudavets and Castañé [@CR59]), aphids (Alvarado et al. [@CR3]), spider mites (Hansen et al. [@CR22]), leaf miners (Arnó et al. [@CR5]) and Lepidoptera species, including *Tuta absoluta* (Urbaneja et al. [@CR67]). It attacks a wide range of arthropod pests and is commercially used for the biological control of spider mites and whiteflies. However, it can also feed on plant tissue (Perdikis and Lykouressis [@CR49]). Plant damage has been observed with high predator densities on tomato, zucchini, and gerbera under experimental conditions, but no such damage has been observed under cropping conditions (Castañé et al. [@CR11]). In this paper, we investigated the effects of phytophagy of *M. pygmaeus* on the performance of three species of its herbivorous prey, the two-spotted spider mite *Tetranychus urticae* Koch (Acari, Tetranychidae), the green peach aphid *Myzus persicae* (Hemiptera, Aphididae), and the western flower thrips *Frankliniella occidentalis* (Pergande) (Thysanoptera, Thripidae). These herbivores species employ different feeding strategies. Spider mites and thrips are cell-content feeders, but spider mites have relatively long stylets and feed on parenchymal cells and avoid damaging the epidermis cells (Bensoussan et al. [@CR8]), and thrips use their mouthparts to punch holes in both the epidermal cells and mesophyll cells and subsequently empty the punctured cells with their stylets, resulting in strong plasmolysis and collapse of cells (Hunter and Ullman [@CR26]; van de Wetering et al. [@CR68]). Aphids use their long flexible stylets to feed only from phloem, causing hardly any damage to mesophyll tissue (Walling [@CR72]; Schwarzkopf et al. [@CR65]). These different feeding strategies are reflected in the differential defence responses they induce in plants. Spider mites induce both the JA and the SA pathway (Kant et al. [@CR29]), thrips predominantly induce the JA pathway (Kawazu et al. [@CR35]), whereas phloem-feeding insects such as whiteflies and aphids generally activate the SA pathway (Walling [@CR71]). JA-related plant defences decrease the performance of spider mites and thrips (Omer et al. [@CR45]; Kant et al. [@CR30]). Aphids and whiteflies are also sensitive to the JA-related defences, but they mainly induce SA-related defences, which can suppress JA-related defences (Omer et al. [@CR45]; Zarate et al. [@CR74]; Walling [@CR72]; Puthoff et al. [@CR57]). Owing to these different sensitivities to plant defences, we expected that the herbivores would be differentially affected by plant feeding by the omnivore, and this would help to evaluate which types of defences are induced by the omnivore. We therefore tested the performance of herbivores on plants previously infested by *M. pygmaeus* and on uninfested plants. To test the induction of different defensive systems by the omnivore, we quantified defence-related plant hormones of both defensive pathways in plants infested by *M. pygmaeus* and in uninfested plants. Specifically, we quantified concentrations of the two main hormones of the two defensive pathways, JA and SA. We furthermore quantified the hormone12-oxo-phytodienoic acid (OPDA), which is the precursor of JA (Wasternack and Hause [@CR73]), JA-isoleucine (JA-Ile), which is the main bioactive form of JA and plays a key role in regulating defence gene expression (Fonseca et al. [@CR20]), and abscisic acid (ABA) (Bodenhausen and Reymond [@CR9]; Pieterse et al. [@CR53]), which plays an important ancillary role in fine-tuning plant defences (Kessler and Baldwin [@CR36]; Vos et al. [@CR70]; Kant et al. [@CR31]). Materials and methods {#Sec2} ===================== Plant material {#Sec3} -------------- Sweet pepper plants (*Capsicum annuum* L. Spider F1 Enza Zaden Beheer B.V., The Netherlands) were grown from seeds in pots (Ø = 14 cm) with soil in a climate room dedicated to growing uninfested plants (25 ± 1 °C, 60--70% RH, 16:8 L:D). Water was supplied twice a week. Four-week-old plants with 6--8 true leaves (about 20 cm high) were used for experiments. Plants of 5--8 weeks old were used for the rearing of spider mites, thrips and aphids. Cultures {#Sec4} -------- A culture of *M. pygmaeus* was established with fifth instar nymphs obtained from a commercial biological control company (Koppert Biological Systems BV, Berkel en Rodenrijs, The Netherlands). It was reared in plastic containers (height = 45 cm, Ø = 35 cm) in a climate room (conditions as above) with *Ephestia kuehniella* eggs as food and green bean pods, which served as both food supply and oviposition substrate. New *E. kuehniella* eggs and beans were added twice a week. Old beans with *M. pygmaeus* eggs were transferred to new containers and kept until the eggs hatched, whereupon *E. kuehniella* eggs and beans were supplied twice a week. Beans were removed from the containers when no new nymphs hatched from them. Adults of 4--8 days old were used for experiments. A culture of *T. urticae* was started with individuals that were obtained from a cucumber culture in our lab (see Janssen [@CR27] for details), and reared on intact sweet pepper plants in a climate room (conditions as above). New plants were provided twice a week. The mites were cultivated for 10 months on sweet pepper plants before being used for experiments. Thus, *T. urticae* females used in the experiments were adapted to sweet pepper plants. Nevertheless, the oviposition rate of these spider mites remained low. A red phenotype of *M. persicae* was obtained from a culture on sweet pepper plants at Wageningen UR Greenhouse Horticulture (Bleiswijk, The Netherlands). The culture was established by placing all individuals on intact sweet pepper plants in insect-proof cages (BugDorm-44545F, 47.5 × 47.5 × 47.5 cm, mesh size: 160 µm) in a climate chamber (conditions as above). New plants were supplied every 2 weeks. A culture of *F. occidentalis* originated from the stock culture of Koppert Biological Systems and was maintained on bean pods and cattail pollen (*Typha latifolia* L.) These thrips were subsequently reared on sweet pepper plants supplemented with cattail pollen in fine mesh cages (as above) in a climate chamber (conditions as above). New pollen was applied on sweet pepper leaves with a fine brush three times per week. New plants were supplied twice per week. To obtain cohorts of thrips larvae, adult thrips were collected with an aspirator connected to a vacuum pump and placed on uninfested sweet pepper leaves on soaked cotton wool in Petri-dishes sealed with Parafilm. The lids had ventilation holes covered with a fine mesh for ventilation. Adults were removed after 24 h and leaves were kept until new larvae hatched. Thereafter, new leaves and pollen were added. In this way, cohorts of similar-aged adults were obtained for experiments. Infestation of sweet pepper plants with *M. pygmaeus* {#Sec5} ----------------------------------------------------- Four-week-old plants with 6--8 true leaves were transferred to a climate room (conditions as above). The fourth leaves from below were treated with omnivorous predators to observe the systemic effects on untreated leaves above and below the treated leaves. To restrict the feeding of the omnivores, bags made of fine mesh (30 µm, size: 15 × 13 cm) were used to enclose them on the leaf to be treated. Five adult males and five adult females of *M. pygmaeus*, starved for 2 h, were released in a bag of half the plants, haphazardly chosen. The other half of the plants also received bags over the fourth leaf, but no omnivores, and served as controls. All bags were closed with stretchable ropes around the stems of the leaves to prevent *M. pygmaeus* from escaping. After 4 days, all adults and bags were removed from all plants. No food or prey was supplied for *M. pygmaeus* during these 4 days, thus preventing the females from ovipositing (Perdikis and Lykouressis [@CR50]). An average of 60% of the females of *M. pygmaeus* was alive 4 days later. Effect of plant infestation by *M. pygmaeus* on herbivore reproduction {#Sec6} ---------------------------------------------------------------------- The second to sixth leaf (from below) of sweet pepper plants treated as above (uninfested or previously exposed to *M. pygmaeus*) were used to measure the performance of spider mites, thrips and aphids. Lanolin (Sigma-Aldrich) was pasted around the stem of each leaf of each plant to prevent spider mites from escaping. For reproduction of the spider mites, adult female *T. urticae* were collected from the culture. To reduce the effect of the previous diet on reproduction, all females were starved for 2 h before being used for the experiment. Ten starved females were carefully introduced on each leaf with a fine brush. Subsequently, the number of females was recorded on a daily basis for 4 days, and the number of eggs was recorded on the 4th day under a stereomicroscope. The experiments were conducted in 3 blocks in time and there were 5 plants of each treatment per block. We calculated the number of eggs per mite-day as follows: First, we observed that almost all mites disappeared from the 2nd leaf from below of all plants, suggesting that the quality of this leaf was low, independent of treatment. Because we could not rule out the possibility that the mites escaped by dropping from the plant immediately after being introduced, we excluded oviposition on this leaf for further analysis. Second, we summed the number of alive, ovipositing females on the other leaves per day over a 4-day period, yielding the total number of mite-days per leaf. Subsequently, the total number of eggs produced during 4 days was divided by this total number of mite-days, thus correcting for mortality or escapes of adult females. Each plant was considered as a replicate but because we were interested in local and systemic effects of the feeding by *M. pygmaeus*, we averaged oviposition per leaf per plant. These oviposition rates on different leaves of each treatment were compared using linear mixed-effects models (LME), with treatment, leaf and their interaction as fixed factors and individual plant and block as random factors. The distribution of the residuals was checked for normality. Non-significant interactions and factors were removed until a minimal adequate model was reached (Crawley [@CR14]). Contrasts were assessed with the Tukey method \[package lsmeans in R, Lenth ([@CR38])\]. For survival of the female spider mites, we again excluded the data from the 2nd leaf. The proportion of females surviving was analysed with a generalized linear mixed-effects model (GLMM) using the function glmer of the lme4 package in R (Bates et al. [@CR7]), with treatment, leaf and their interaction as fixed factors and individual plant and block as random factors. The distribution of the residuals was checked for normality. All statistical analyses were performed with R Development Core Team ([@CR58]). For the assessment of performance of aphids, a single 2- to 4-day-old apterous adult was introduced with a fine brush on leaf 3--6 of plants treated as above. Each leaf was enclosed with a bag (as above) to prevent the aphids from escaping. Four days later, all the bags were removed carefully. All adult aphids survived, except for one female, which died due to handling and was therefore excluded from the analysis. Thereafter, the number of nymphs produced by each female aphid was counted under a stereo microscope. The experiments were conducted in 2 blocks in time with 5 plants of each treatment per block. The numbers of nymphs per female per day were compared using LME as explained above with treatment, leaf and their interaction as fixed factors and individual plant and block as random factors. Given that thrips females are able to fly and also escaped from the bags used for aphids, leaf discs were used in this experiment. Based on the results of the previous experiment, leaves 3--5 (from below) of infested and uninfested plants were used. Five leaf discs (Ø = 15 mm) were made from each leaf, avoiding areas including the midrib. Each leaf disc was placed upside down on soaked cotton wool inside a small plastic cup (Ø = 20 mm, height = 3 cm). A single 4--6 days old female thrips was released inside the cup. All cups were closed with lids with a ventilation hole covered with fine mesh (80 µm). Females were removed 3 days later and all leaf discs were kept for another 4 days until all larvae had hatched from the leaf discs. The total number of larvae that hatched from the eggs produced by each female was recorded. The experiments were conducted in 2 blocks in time and there were 8 replicates (plants) per treatment in total. Thrips that were missing or dead at the end of the experiment (around 16%) were excluded from the analysis of reproduction, but there were always surviving thrips for each leaf of each plant. We calculated the average number of thrips larvae per female of each leaf and this average was used for further analysis. As above, these data were compared using LME with leaf and treatment and their interaction as fixed factors and individual plant and block as random factors. Below, we do not report non-significant interactions. The distribution of the residuals was checked for normality. The number of thrips larvae was log (*x* + 10) transformed for the MANOVA and LME. Contrasts were assessed with the Tukey method \[package lsmeans in R, Lenth ([@CR38])\]. For adult thrips, the proportions of females that survived per leaf of each plant were analysed with a GLMM as described above, with treatment, leaf and their interaction as fixed factors and individual plant and block as random factors. Effect of plant infestation by *M. pygmaeus* on immature herbivore development {#Sec7} ------------------------------------------------------------------------------ Plants that received the same treatments (uninfested plants and plants infested by *M. pygmaeus*) as for the above experiment were used. Based on the previous experiment, leaves 3--5 of each plant were chosen for this experiment. Five leaf discs (Ø = 15 mm) were made from each leaf and placed upside down in small plastic cups on top of soaked cotton wool. Three leaves (leaves 3, 4 and 5) of five plants per treatment were used for the larval development of spider mites, and three leaves of 10 plants per treatment were used for larval development of thrips and nymphal development of aphids. A single newly hatched two-spotted spider mite larva, a newly born aphid nymph, or a first-instar thrips larva was transferred carefully to each leaf disc with a fine brush under a stereomicroscope. The cup was closed with a lid with a ventilation hole covered with fine mesh (80 µm). Juvenile survival and their developmental stages were checked under the stereomicroscope and recorded on a daily basis until the adult stage for spider mites and aphids. Thrips larvae largely stop feeding when they become prepupae; we therefore, analysed the development and survival from larva to prepupa. Discs were replaced by new ones that had received the same treatments every 3 days. Because five spider mite larvae and seven thrips larvae died due to handling, some of the averages per leaf were based four individuals instead of five. Survival of immatures per leaf was compared with a GLMM as above. The developmental times were log (*x* + 0.1) transformed and were analysed with LME following the same procedure as explained above. Phytohormone accumulation {#Sec8} ------------------------- To test whether plant feeding of *M. pygmaeus* induced plant defences, phytohormones were quantified. Sweet pepper plants were treated for 4 days as above, six plants with *M. pygmaeus* inside the bags, the other six plants with only the bags as control. Leaves 2--6 of each plant were frozen in liquid nitrogen and stored at − 80 °C. The metabolites OPDA, JA, JA-Ile, SA and ABA were analysed with liquid chromatography--mass spectrometry/mass spectrometry (LC--MS/MS) using the procedure of Alba et al. ([@CR2]) with some minor modifications. In short, c. 200 mg of frozen leaf material was homogenized (Precellys 24, Bertin Technologies, Aix-en-Provence, France) in 1 ml of ethyl acetate solution with 100 ng/ml of the internal standards D6-SA, D5-JA and D6-ABA (C/D/N Isotopes Inc, Canada). Samples were centrifuged at 15,000 rpm for 20 min at 4 °C and the ethyl acetate phase was transferred to new tubes. Pellets were washed with 0.5 ml of ethyl acetate without internal standards. Subsequently, the supernatants were combined with the previous extraction after centrifugation (20 min at 4 °C at 15,000 rpm). Samples were dried on a vacuum concentrator (CentriVap Centrifugal Concentrator, Labconco, Kansas City, MO, USA) at room temperature. The residue was re-suspended in 0.25 ml of 70% methanol (v/v). Samples were transferred to glass tubes and then analysed with LC--MS/MS. To calculate the amount of each compound, a serial dilution of pure standards of OPDA, JA, JA-Ile, SA, SA and traumatic acid was included. Measurements were conducted on a liquid chromatography tandem mass spectrometry system (Varian 320-MS LC/MS, Agilent Technologies, Amstelveen, the Netherlands). Twenty microlitre of each sample was injected into a Pursuit XRs 5 column (C18; 50 × 2.0 mm, Agilent Technologies, Amstelveen, The Netherlands). The mobile phase comprised solvent A (0.05% formic acid in water; Sigma-Aldrich, Zwijndrecht, the Netherlands) and solvent B (0.05% formic acid in methanol; Sigma-Aldrich). The program was set as follows: 95% solvent A for 1 min 30 s, followed by 6 min in which solvent B increased until 98% (0.2 ml min^−1^) which continued for 5 min with the same flow rate, subsequently returning to 95% solvent A for 1 min until the end of the run. Metabolites were detected using the negative electrospray ionization mode. The mother ions, daughter ions and collision energies used in these analyses are listed in Supplementary Table S1. For all oxylipins, we used D5-JA to calculate the recovery rate, and their *in planta* concentrations were subsequently quantified using the external standard series. For SA and ABA we used D6-SA and D6-ABA, respectively, to calculate the recovery rate and they were quantified using the external standard. Values were expressed as ng per gram fresh leaf material. First, we carried out a multivariate analysis of variance (MANOVA) to check the overall effects of treatments on the concentrations of all plant hormones. Because this MANOVA showed a significant effect, we proceeded with an analysis of the individual hormones. The hormone concentrations in different leaves from each plant in each treatment were compared with LMEs with treatment, leaf and their interactions as fixed factors and individual plant as a random factor. The distribution of the residuals was checked for normality. The concentrations of OPDA and JA-Ile were square root transformed and the JA log (*x* + 1) transformed. Non-significant interactions and factors were removed until a minimal adequate model was reached (Crawley [@CR14]). Vascular connectivity of different leaves of sweet pepper plants {#Sec9} ---------------------------------------------------------------- Rhodamine-B (Sigma-Aldrich, St Louis, MO, USA) was used to track the vascular connectivity of different leaves to leaf four of sweet pepper plants following the protocol described in Orians et al. ([@CR46]) (Supp Mat Methods). Briefly, leaf tissue of leaf four of six plants was removed, and the main vein and part of the petiole was inserted in a 15 ml polypropylene tube (Greiner Bio-One GmbH, Germany) filled with a solution of Rhodamine-B 0.1% (w/v). After 4 h, 24 h, 48 h, leaves 3, 5 and 6 were excised and the distribution of Rhodamine-B in each leaf was tracked using an UV transilluminator (Syngene, UK) with an exposure time of 1.84 s for all pictures. Results {#Sec10} ======= Effect of plant infestation by *M. pygmaeus* on herbivore reproduction {#Sec11} ---------------------------------------------------------------------- Overall, the oviposition rate of *T. urticae* on *M. pygmaeus*-infested plants was lower than on the uninfested plants (Fig. [1](#Fig1){ref-type="fig"}a, LME: *χ* ^2^ = 5.83, *df* = 1, *P* = 0.016), and it varied among leaves (Fig. [1](#Fig1){ref-type="fig"}a, LME: *χ* ^2^ = 10.9, *df* = 3, *P* = 0.012). Female *T. urticae* feeding on the *M. pygmaeus*-damaged leaf (leaf 4) and the 5th leaf of treated plants laid significantly fewer eggs than females on corresponding leaves of uninfested plants (Fig. [1](#Fig1){ref-type="fig"}a, contrasts with glht function of package lsmeans). The same trend was observed on the 3rd and 6th leaves.Fig. 1Average reproduction rate (+ SE) of *T. urticae* (**a**; *n* = 15 plants), *M. persicae* (**b**; *n* = 10) and *F. occidentalis* (**c**; *n* = 8) when feeding on leaves from plants previously exposed to *M. pygmaeus* for 4 days (black bars) and uninfested plants (control, white bars). Significant differences between corresponding leaves from infested and uninfested plants are indicated by asterisks (contrasts after LME, \**P* \< 0.05). Different letters inside bars indicate significant differences among different leaves of uninfested plants (small letters) and of infested plants (capital letters) (contrasts after LME: *P* \< 0.05) Female *M. persicae* produced similar numbers of nymphs on uninfested and treated plants (Fig. [1](#Fig1){ref-type="fig"}b, LME: *χ* ^2^ = 0.12, *df* = 1, *P* = 0.73), and numbers of nymphs did not differ among leaves (Fig. [1](#Fig1){ref-type="fig"}b, LME: *χ* ^2^ = 0.87, *df* = 3, *P* = 0.83). Overall, lower numbers of *F. occidentalis* larvae were found on treated plants than on uninfested plants (Fig. [1](#Fig1){ref-type="fig"}c, LME: *χ* ^2^ = 4.52, *df* = 1, *P* = 0.034). Lower numbers of larvae were found on the leaf infested by *M. pygmaeus* (leaf 4) and on leaf 5 than on the corresponding leaves of uninfested plants (Fig. [1](#Fig1){ref-type="fig"}c, contrast as above). The numbers of larvae differed among leaves (Fig. [1](#Fig1){ref-type="fig"}c, LME: *χ* ^2^ = 10.9, *df* = 2, *P* = 0.004). The survival of adult *T. urticae* females was differentially affected by feeding of *M. pygmaeus* on different leaves (Fig. [2](#Fig2){ref-type="fig"}a, GLMM, interaction between treatment and leaf: *χ* ^2^ = 7.82, *df* = 3, *P* = 0.0498). Survival on the leaf previously exposed to *M. pygmaeus* (leaf 4) was only half as high as on the corresponding leaf of uninfested plants, but survival on the other leaves did not differ significantly between treatments (Fig. [2](#Fig2){ref-type="fig"}a). All aphid females survived the entire experiment. The survival of adult *F. occidentalis* females did not differ between uninfested and *M. pygmaeus*-infested plants (Fig. [2](#Fig2){ref-type="fig"}b, GLMM: *χ* ^2^ = 0.08, *df* = 1, *P* = 0.78), or among different leaves per treatment (GLMM: *χ* ^2^ = 2.36, *df* = 2, *P* = 0.31).Fig. 2Average proportions (+ SE) of **a** *T. urticae* females surviving on leaves of plants previously exposed to *M. pygmaeus* (black bars) and leaves of uninfested plants (control, white bars) on day 4 (*n* = 15 plants), and **b** average proportions (+ SE) of *F. occidentalis* females surviving on leaves of treated and uninfested plants after 3 days (*n* = 8). Significant differences between infested and uninfested leaves are indicated by asterisks (contrasts after LME, \**P* \< 0.05) Effect of plant infestation by *M. pygmaeus* on immature herbivore development {#Sec12} ------------------------------------------------------------------------------ No significant effect of treatment or leaf was found on the survival of *T. urticae* from larva to adult (Fig. [3](#Fig3){ref-type="fig"}a, GLMM: treatment: *χ* ^2^ = 0.0008, *df* = 1, *P* = 0.98; leaf: *χ* ^2^ = 1.32, *df* = 2, *P* = 0.52). Furthermore, the developmental times of spider mite immatures feeding on uninfested and treated plants did not differ significantly (Fig. [4](#Fig4){ref-type="fig"}a, LME: *χ* ^2^ = 0.86, *df* = 1, *P* = 0.35). Developmental times of immatures feeding on different leaves of the same treatment also did not differ significantly (Fig. [4](#Fig4){ref-type="fig"}a, LME: *χ* ^2^ = 4.04, *df* = 2, *P* = 0.13).Fig. 3Average proportions (+ SE) of immature *T. urticae* (**a**; *n* = 5 plants) and *M. persicae* (**b**; *n* = 10) surviving to adult, and of immature *F. occidentalis* (**c**; *n* = 10) surviving to prepupa when feeding on leaves from plants previously exposed to *M. pygmaeus* for 4 days (black bars) and uninfested plants (control, white bars) Fig. 4Average developmental time (+ SE) of immature *T. urticae* (**a**; *n* = 5 plants) and *M. persicae* (**b**; *n* = 10) to adult, and of *F. occidentalis* (**c**; *n* = 10) to prepupa when feeding on leaves from plants previously exposed to *M. pygmaeus* for 4 days (black bars) and uninfested plants (control, white bars). Significant differences between corresponding leaves from infested and uninfested plants are indicated by asterisks (contrasts after LME, \**P* \< 0.05) For aphid survival, no significant effect of treatment or of leaf was found (Fig. [3](#Fig3){ref-type="fig"}b, GLMM: treatment: *χ* ^2^ = 0.82, *df* = 1, *P* = 0.78; leaf: *χ* ^2^ = 2.13, *df* = 2, *P* = 0.34). Nymphs feeding on uninfested and treated plants required similar times to develop into adults (Fig. [4](#Fig4){ref-type="fig"}b, LME: *χ* ^2^ = 2.95, *df* = 1, *P* = 0.086). The developmental period did not differ significantly on different leaves of plants of the same treatment (Fig. [4](#Fig4){ref-type="fig"}b, LME: *χ* ^2^ = 2.48, *df* = 2, *P* = 0.29). The survival of thrips larva to prepupa was not affected by the plant treatment or the plant leaf (Fig. [3](#Fig3){ref-type="fig"}c, GLMM, treatment: *χ* ^2^ = 1.61, *df* = 1, *P* = 0.20; leaf: *χ* ^2^ = 1.24, *df* = 2, *P* = 0.54). A significant effect of the interaction between treatment and leaf was found on the developmental time from larva to prepupa (Fig. [4](#Fig4){ref-type="fig"}c, LME: *χ* ^2^ = 8.93, *df* = 2, *P* = 0.0115). Larvae feeding on treated leaves (leaf 4) required a longer time to develop into prepupae than those feeding on the corresponding leaves of uninfested plants (Fig. [4](#Fig4){ref-type="fig"}c). Phytohormone accumulation {#Sec13} ------------------------- Overall, there was a significant effect of the interaction between treatment and leaf on the concentrations of hormones accumulated (MANOVA: treatment: leaf: *F* ~4,\ 24~ = 2.31, *P* = 0.009). We, therefore, analysed each plant hormone separately. Feeding by *M. pygmaeus* resulted in significantly higher concentrations of OPDA in the attacked leaf (leaf 4) than in the 4th leaf of the uninfested plants and than in the other leaves of the treated plants (Fig. [5](#Fig5){ref-type="fig"}a). This resulted in a significant interaction between treatment and leaf (Fig. [5](#Fig5){ref-type="fig"}a, LME: *χ* ^2^ = 11.83, *df* = 4, *P* = 0.019). Similar concentrations of OPDA were produced in all leaves of uninfested plants.Fig. 5Average concentrations of plant hormones (in ng/g fresh weight), of OPDA (**a**), JA (**b**), JA-Ile (**c**), SA (**d**) and ABA (**e**) (mean + SE; *n* = 6 plants) in different leaves of plants previously exposed to *M. pygmaeus* for 4 days (black bars) and uninfested plants (control, white bars). Significant differences between corresponding leaves from infested and uninfested plants are indicated by asterisks (contrasts after LME, \**P* \< 0.05). Different letters inside the bars indicate significant differences among different leaves of uninfested plants (small letters) and of infested plants (capital letters, contrasts after LME: *P* \< 0.05) The concentrations of JA in leaves of treated and uninfested plants did not differ significantly (Fig. [5](#Fig5){ref-type="fig"}b, LME: *χ* ^2^ = 0.30, *df* = 1, *P* = 0.58), and different leaves of plants with the same treatment produced similar concentrations of JA (Fig. [5](#Fig5){ref-type="fig"}b, LME: *χ* ^2^ = 8.49, *df* = 4, *P* = 0.075). The concentration of JA in the attacked leaf (leaf 4) was higher than that in the corresponding leaf of the uninfested plants, but the difference was not statistically significant due to high variation in the JA concentration in the attacked leaves (Fig. [5](#Fig5){ref-type="fig"}b). Leaves attacked by the omnivore (leaf 4) contained a significantly higher concentration of JA-Ile than the corresponding leaf of the uninfested plants and the other leaves of the treated plants (Fig. [5](#Fig5){ref-type="fig"}c). This resulted in a significant interaction between treatment and leaf (Fig. [5](#Fig5){ref-type="fig"}c, LME: *χ* ^2^ = 21.51, *df* = 4, *P* = 0.0003). JA-Ile concentrations in the other leaves of infested plants did not differ significantly from the corresponding leaves of uninfested plants (Fig. [5](#Fig5){ref-type="fig"}c). Different leaves of uninfested plants accumulated similar amounts of JA-Ile (Fig. [5](#Fig5){ref-type="fig"}c). There was also a significant effect of the interaction between treatment and leaf for the amount of SA in different leaves (LME: *χ* ^2^ = 10, *df* = 4, *P* = 0.040), but there was no significant effect of treatment per leaf (Fig. [5](#Fig5){ref-type="fig"}d). Within each treatment, there was no significant difference in the concentrations of SA among leaves (Fig. [5](#Fig5){ref-type="fig"}d, contrasts as above). Feeding by *M. pygmaeus* also resulted in significantly higher concentrations of ABA in the attacked leaf (leaf 4) and in the uninfested leaves 3 and 6 than in the other leaves of the treated plants and than in the corresponding leaves of the uninfested plants (Fig. [5](#Fig5){ref-type="fig"}e, LME, interaction between treatment and leaf: *χ* ^2^ = 16.76, *df* = 4, *P* = 0.002). Different leaves within the same treatment accumulated different amounts of ABA (Fig. [5](#Fig5){ref-type="fig"}e). Vascular connectivity of different leaves of sweet pepper plants {#Sec14} ---------------------------------------------------------------- After 4 h, Rhodamine-B was observed in half of each leaf 5 and 6, but not in leaf 3 (Supp Mat Figure S1). Subsequently, the Rhodamine-B also accumulated in the other halves of leaf 5 and 6 after 24 h (Supp Mat Figure S1). After 48 h, it was visible in leaf 3 and both sides of the leaf 5 and 6 (Supp Mat Figure S1). Discussion {#Sec15} ========== We show that the feeding of *M. pygmaeus* on sweet pepper plants lowers the performance of two of the three herbivore species feeding on the same plants through induced defences. The reproduction of *T. urticae* and *F. occidentalis* on *M. pygmaeus*-infested plants was significantly lower than on uninfested plants, not only on the leaves that had been exposed to *M. pygmaeus*, but also on other leaves of the same plants, showing that the effect was systemic. Furthermore, *F. occidentalis* larvae developed slower on leaves previously exposed to *M. pygmaeus* than on uninfested leaves. In contrast, there was no effect on the reproduction of female *M. persicae*. The developmental rate and juvenile survival of *T. urticae* and *M. persicae* and larval survival of *F. occidentalis* were not affected by the previous infestation of *M. pygmaeus*. Similarly, Pappas et al. ([@CR47]) have shown that the performance of *T. urticae* was lower on local and systemic leaves of tomato plants that were previously exposed to *M. pygmaeus* than on leaves of uninfested plants, but they found no effect on the greenhouse whitefly. However, these authors found lower survival and oviposition of spider mites, but it is unclear whether the lower oviposition was mainly caused by decreased survival of the ovipositing female, by surviving females producing fewer eggs, or by both. Here, we corrected for mortality of the adult females and show that both the survival and oviposition rate were negatively affected by previous exposure of plants to *M. pygmaeus*. Oviposition rates of spider mites on uninfested plants were relatively low (1--1.5 eggs per day), confirming that sweet pepper is a less suitable host plant for this herbivore (van den Boom et al. [@CR69]). Another study showed that the predatory bug *O. laevigatus* also increased tomato resistance against the thrips *F. occidentalis* (De Puysseleyr et al. [@CR16]). Thus, omnivorous predators can decrease the performance of herbivores sharing the same plants both directly, through predation, as well as indirectly through induced plant defences. The number of spider mite females on the *M. pygmaeus*-infested plants decreased during the experiments compared to those on the uninfested plants. Because there were no predators present, the females either escaped or died because of poor plant quality. Further experiments are needed to confirm whether spider mites disperse more from plants previously exposed to *M. pygmaeus*. Notice that reproduction was corrected for this dispersal and mortality (see "[Materials and methods](#Sec2){ref-type="sec"}"). Because thrips and aphid reproduction was tested with leaf discs and in bags, respectively, they were prevented from escaping. Besides reacting to reduced plant quality due to induction of defences, the herbivores may have responded to the cues left behind by the omnivore on the exposed leaves, as was found for several prey in response to predator cues (Kats and Dill [@CR34]). Possibly the presence of omnivore faeces or other products may have affected the behaviour of the herbivores. It is also known that egg deposition by herbivores (Hilker and Meiners [@CR24]) or even herbivore walking (Bown et al. [@CR10]; Peiffer et al. [@CR48]) can induce changes in plants. Perhaps plants respond in a similar manner to omnivores and this may have contributed to modulation of the plant's local and systemic defence responses. However, the reproduction of spider mites and thrips was also reduced on unexposed leaves of exposed plants, suggesting that plant quality affected the performance of the herbivores. In contrast to reproduction, the juvenile survival and developmental period of *T. urticae* and the juvenile survival of *F. occidentalis* were not affected by the infestation of *M. pygmaeu*s, but the juvenile developmental time of *F. occidentalis* was delayed on leaves previously exposed to *M. pygmaeu*s. This may point at a difference in response of juveniles versus adults to cues associated with previous exposure of the plants. Possibly, adult females refrain from reproducing in the presence of such cues. This could be caused by egg retention by the females (Montserrat et al. [@CR43]), or because the females attempted to escape, thus spending less time and energy on reproduction. The observation that many spider mites did escape or die partly confirms this, but even females that did not escape reproduced less. Although thrips and aphid females were confined, this did not prevent them from attempting to escape and feeding and reproducing less as a consequence. Another explanation would be that juvenile spider mites and thrips feed much less than adult females, hence, suffer less from the decreased plant quality. Furthermore, we used leaf discs for the experiments on larval development experiments instead of intact plants for practical reasons, which may also have affected the results. Although both methods are amply used in the literature, the effects of induced plant defences may differ between these two approaches. Elsewhere (Dias et al., in prep.) we address this issue, showing that the reproduction rates of spider mites on leaf discs and on entire plants showed the same trend. Therefore, it is likely that the effect of infestation by *M. pygmaeus* on thrips reproduction will be comparable on leaves from intact plants. In the current experiments, five pairs of *M. pygmaeus* were released on the 4th leaf of each plant, and no leaf damage by *M. pygmaeus* was observed during the experiments, yet these low numbers were sufficient to activate plant defences. In our experiments, *M. pygmaeus* were released on one leaf, however, in practice, they are free to disperse to other plant parts, and hence more leaves may become exposed to these omnivores, resulting in larger overall effects on herbivore performance. However, pest individuals may actively avoid feeding and reproducing on leaves that were previously exposed to *M. pygmaeus*. This will be the subject of further research. Oviposition of female *M. pygmaeus* and plant feeding of nymphs can also induce plant defences (Pappas et al. [@CR47]). In this study, no prey or food was supplied to *M. pygmaeus.* An earlier study showed that female *M. pygmaeus* did not oviposit on pepper plants in the absence of prey (Perdikis and Lykouressis [@CR50]), and indeed, no nymphs were observed on the plants during our experiments. Thus, the effect on the performance of herbivores was mainly due to the response induced by the feeding of *M. pygmaeus* and not due to the response to oviposition or to the presence of nymphs. The three herbivores species used in our study have different feeding modes and induce and are sensitive to different plant defences. Aphids are sensitive to the JA-related defences, but they mainly induce SA-related defences, which suppress JA-related defences (Omer et al. [@CR45]; Zarate et al. [@CR74]; Walling [@CR72]; Puthoff et al. [@CR57]). Thrips are sensitive to JA-related defences and spider mites are sensitive to both JA- and SA-related defences. Our observation that the performance of spider mites and thrips were affected by plant feeding by *M. pygmaeus* and that of the aphids not, suggests that the omnivore mainly induces JA-related defences. To further confirm whether plant defences were involved in the decrease of performance of some of the herbivores tested here, we quantified the plant hormones accumulated in leaves of uninfested plants and *M. pygmaeus*-infested plants. We found that the concentrations of the plant hormones OPDA and JA-Ile in the exposed leaves was significantly higher than in the corresponding leaves of uninfested plants, but no such effect was found for the uninfested leaves of the exposed plants. The accumulation of JA showed the same trend as that of JA-Ile, but this was not significant. Concentrations of JA and JA-Ile in induced plants were comparable to those found in induced tomato plants (Alba et al. [@CR2]; Schimmel et al. [@CR64]). The amount of SA in the 3rd, 4th, 5th leaves of treated plants was not significantly higher than in corresponding leaves of the uninfested plants. ABA levels were significantly higher in leaves exposed to *M. pygmaeus* and in some of the other leaves of the same plants than in the respective leaves of the uninfested plants. Similarly, Pérez-Hedo et al. ([@CR52]) found that the JA signalling pathway was activated and the amount of ABA was higher in tomato leaves attacked by the omnivore *N. tenuis*, but the SA pathway was not activated. Attack by *M. pygmaeus* only activated the JA signalling pathway in tomato leaves, not the ABA pathway (Pérez-Hedo et al. [@CR51]). It is likely that plants of different species respond differently to omnivore damage. It is known that ABA plays an important role in response to abiotic stress and its role in biotic stress is becoming evident in *Arabidopsis* (Mauch-Mani and Mauch [@CR39]; Asselbergh et al. [@CR6]; Pieterse et al. [@CR53]). In *Arabidopsis*, ABA enhances the JA-related response (Anderson et al. [@CR4]) and antagonizes SA-dependent responses (Mohr and Cahill [@CR41]). All these results suggest that several of the plant hormones investigated here are involved in the local defence induced by the omnivore, but are not decisive in the systemic defences experienced by the herbivores. Perhaps other, unidentified metabolites (e.g. OPC4; Gasperini et al. [@CR21]) are involved in the systemic effects. Alternatively, plant defences may have been primed in undamaged distal leaves of *M. pygmaeus*-infested plants, and JA-regulated defences may not have been activated by *M. pygmaeus* feeding, but could have been rapidly activated upon subsequent attacks by the herbivores, which would explain the decreased performance of herbivores on these leaves. The systemic response by ABA suggests that it may play a role in modulating JA-defences induced by omnivorous mirid bugs. Indeed, ABA plays an important role in herbivore-induced defences by activating primed JA-regulated defences upon secondary herbivore attack (Vos et al. [@CR70]). Concluding, our results suggest that plant feeding by the omnivorous predator *M. pygmaeus* induces the JA-related defensive pathway in sweet pepper plants, and this coincides with reduced performance of the herbivore species *T. urticae* and *F. occidentalis*. The phytohormones accumulated in the *M. pygmaeus*-infested plants confirm that mainly the JA-related pathway was induced. Perhaps aphids can decrease these defences by inducing SA-related defences. Pappas et al. ([@CR47]) found no effect of previous exposure of tomato plants on the performance of whitefly, which is also considered to be able to suppress induced JA-related defences (Zarate et al. [@CR74]; Walling [@CR72]). Together, this suggests that spider mites and thrips are more sensitive to the defences induced by *M. pygmaeus* than are whiteflies and aphids, perhaps because the latter can decrease these defences through cross-talk with the SA-related defences they induce. *Macrolophus pygmaeus* attacked the 4th leaf of the plants in our experiment. This leaf has higher vascular connectivity with leaves 5 and 6 than with leaf 3 (Supp Mat Figure S1). However, the dye used in this experiment was also detected in leaf 3 after 48 h (Supp Mat Figure S1). Because our plants were treated with *M. pygmaeus* for 4 days, defence-related compounds were likely transferred to all untreated leaves during this time, enabling systemic effects on herbivores feeding on the tested leaves. Although the increased concentrations of OPDA, JA and JA-Ile point at a local, non-systemic effect, the concentrations of ABA point at a systemic effect. Perhaps longer exposure of plants to the omnivores, or exposure to higher densities, would result in more pronounced differences in phytohormone concentrations in distal leaves. Alternatively, the systemic response may have been primed rather than induced (Conrath et al. [@CR13]) which implies that systemic effects could only have been seen when comparing primed and unprimed leaves infested with herbivores. This will be the subject of further research. Overall, we conclude that plant feeding by the omnivorous predator *M. pygmaeus* can decrease the performance of herbivores feeding on the same plants through the induction of defences. Possibly plants simply responded to the feeding of the omnivores similar to their response to herbivores. However, plants are known to respond differently to different herbivore species, suggesting that there is specificity in their response (Turlings et al. [@CR66]; De Moraes et al. [@CR15]; Alba et al. [@CR2]). Hence, plants can adapt their defensive response to the agent causing the plant damage on an evolutionary time scale, so induction of defences by *M. pygmaeus* may not be a simple side effect of plant damage. A remaining question is then why plants would mount direct defences when they are already defended by omnivorous predators. Possibly, plants affect the omnivore's diet through induced defences, thus manipulating them to feed more on herbivores and less on the plant tissue in which defences are induced. It is known that omnivores may change their diet according to plant quality (Agrawal et al. [@CR1]; Janssen et al. [@CR28]), and that a decrease in plant quality may increase the predation rate of omnivores (Eubanks and Denno [@CR19]). Further research is, therefore, needed to investigate the diet choice of *M. pygmaeus* on induced and uninduced plants. Another possibility is that in nature, the presence of omnivorous predators is strongly correlated with the presence of herbivores, and thus, plant defences are likely to be induced anyway, so further induction by omnivores has little effect on plant defences and plant fitness. Clearly, further research is needed to investigate the ecological role of omnivorous predators in plant--herbivore--predator interactions. Electronic supplementary material ================================= {#Sec16} Below is the link to the electronic supplementary material. Supplementary material 1 (DOCX 1404 kb) The original version of this article was revised: The citation of one of the papers was published erroneously in the original version and corrected here. **Electronic supplementary material** The online version of this article (10.1007/s00442-017-4000-7) contains supplementary material, which is available to authorized users. A correction to this article is available online at <https://doi.org/10.1007/s00442-017-4015-0>. We thank Enza Zaden Beheer B.V. for the supply of sweet pepper seeds, Koppert Biological Systems for the supply of *Macrolophus*, Alexandra Revynthi for spider mites, Karen Muñoz Cárdenas for thrips, Bart Schimmel for comments on a previous version of the ms and the population biology group for discussions. Maurice Sabelis was involved in the initial planning of the research but sadly passed away without seeing the end result. The comments of two reviewers and the handling editor substantially improved the ms. NXZ received a scholarship from the China Scholarship Council, JMA and MRK were supported by NWO (STW-VIDI/13492, STW-GAP/13550). NXZ and JMA performed the experiments, NXZ and AJ analysed the data and wrote the ms, all authors commented on the ms and corrected it. [^1]: Communicated by Caroline Müller.

Introduction {#Sec1} ============ Sustainable fuel and chemical production by utilization of biomass-derived feedstocks is a very attractive approach compared to conventional petroleum-based fuel and chemical production. Petroleum-based fuel causes serious ecological damage by greenhouse gas emissions and eventually depletes the world's natural resources^[@CR1]^. Biodiesel is one such renewable, eco-friendly fuel produced from biomass, that can directly replace conventional petroleum diesel, and has attained great industrial expansion world-wide. For every ten quintals of biodiesel synthesized by base catalyzed transesterification of plant triglycerides, about one quintal of waste glycerol is coproduced, nearly 10 wt% of the complete product (Supplementary Fig. [S1](#MOESM1){ref-type="media"}). With the rising biodiesel production, its major co-product, crude glycerol, is set to reach a global production of six million tons by 2025^[@CR2]^. The large volume of accumulated glycerol devalues the industrial biodiesel production process, dropping the market price of glycerol. Moreover, discarding surplus glycerol involves costly procedure and is one of the main difficulties provoked by biodiesel industries. Hence, there is an urgent need to manage the glycerol by-product in biodiesel production. This includes finding new applications for glycerol such as chemical production, or energy generation, to improve the long-term viability of the biodiesel production industry. Due to the biodegradable, non-toxic as well as highly functionalized nature, glycerol is considered one of the top twelve bio-based platform chemicals by US Department of Energy^[@CR3]^. Significant global interest has been focused on the utilization of biodiesel waste glycerol as a versatile building block and its transformation into high value-added specialty chemicals. Heterogeneous catalysis is an important approach to establish sustainable chemical reactions and develop new methods for glycerol valorization. Several catalytic processes including hydrogenolysis, oxidation, dehydration, aqueous phase reforming, esterification, and acetalization have been established directing at upgrading glycerol to value-added chemicals and fuels^[@CR4]--[@CR7]^. Amongst various processes, hydrogenolysis of glycerol to propanediols (1,2-propanediol & 1,3-propanediol) is a promising approach for value-added chemical production. 1,3-propanediol (1,3-PDO) is an important industrial commodity and a building block for wide-ranging products including polymers, paints, cosmetics, cleaning products, adhesives, carpets, textiles, coolant, and personal care. The foremost application of 1,3-PDO is the manufacture of polytrimethylene terephthalate (PTT), a commercial polyester used to produce carpet fibers. The expanding use of 1,3-PDO creates high demand for the product, and the global market value is expected to reach USD 776.3 million by 2022^[@CR8]^. Conventionally, 1,3-PDO is produced by petrochemical methods involving hydration of acrolein or hydroformylation of ethylene oxide followed by hydrogenation^[@CR9]^. Owing to the commercial importance and rising consumption of 1,3-PDO in various end-use industries, there is a necessity to produce 1,3-propanediol via sustainable and economically viable processes. 1,3-PDO produced directly from bio-glycerol catalytic hydrogenolysis will provide a sustainable route with a greater emphasis in current chemical processes. The catalytic reaction of glycerol hydrogenolysis favoring 1,2-PDO by now is well established and has been demonstrated to be a relatively feasible process. However, selective production of 1,3-PDO which involves the removal of sterically hindered secondary hydroxyl group from glycerol has turned out to be far more intricate than that of 1,2-PDO, and the reported systems are limited. Supported noble metal (Pt, Ir) catalysts with a co-catalyst (WO~3~, ReOx) especially high in Brønsted acid sites have been proven to be the more efficient catalytic materials for the selective production of 1,3-PDO from glycerol hydrogenolysis^[@CR10]--[@CR17]^ (Fig. [1](#Fig1){ref-type="fig"}). Although some attempts^[@CR11],[@CR17]^ for innovative catalysts favoring 1,3-PDO have been made by researchers, the quest for new highly active and selective catalysts for 1,3-PDO from glycerol hydrogenolysis has been a longstanding challenge. Hence, with a fundamental understanding of the mechanism of glycerol hydrogenolysis reaction, there is a need to design better and more efficient catalysts with high selectivity.Figure 1Catalysts investigated for glycerol hydrogenolysis to 1,3-PDO in literature. Montmorillonite K10 (MMT) is a non-corrosive, low-cost, environmentally benign solid acid catalyst that can exist in both normal and ion-exchanged forms. MMT has been extensively explored as a highly active catalyst and a support material for various chemical transformations and possess unique properties such as, easy handling, recovery and probable tuning of acid sites^[@CR18]^. The interlayer space of MMT provides a good platform to accommodate various guest species or metal nanoparticles^[@CR19]^. The modification of montmorillonite by acid treatment can significantly increase the acid catalytic performance of montmorillonite. Its combination with noble metal Platinum which has been demonstrated to hold high hydrogenation activity and selectivity in C--O bond hydrogenolysis^[@CR20]--[@CR22]^, will therefore be a prospective promising formulation to obtain high yields of 1,3-PDO. To achieve an improved catalytic activity and selectivity to 1,3-PDO in hydrogenolysis of glycerol, the present work focused on the development of highly efficient, eco-friendly and recyclable sulfuric acid-activated MMT supported platinum nanoparticles as a potential catalyst promoting selective production of 1,3-PDO. Different loadings of platinum nanoparticles immobilized on MMT catalysts were prepared, thoroughly characterized by different characterization techniques and evaluated in vapor phase glycerol hydrogenolysis using a fixed-bed reactor under mild reaction conditions (moderate temperature and atmospheric pressure). The effect of process variables on the catalytic performance was carefully investigated. The catalyst stability and reusability has also been evaluated. The research work reported herein is unique and to the best of our understanding, platinum supported on sulfuric acid-activated montmorillonite has not been explored so far as a catalyst for glycerol conversion to 1,3-PDO by hydrogenolysis. Results {#Sec2} ======= Catalyst characterization {#Sec3} ------------------------- ### Powder X-ray Diffraction analysis {#Sec4} The diffraction profiles of montmorillonite, sulfuric acid-treated montmorillonite and various loadings (0.5--3 wt%) of Pt/S-MMT catalysts are shown in Figs [2](#Fig2){ref-type="fig"} and [S2](#MOESM1){ref-type="media"} (Supplementary). The XRD pattern of parent montmorillonite (MMT) (Fig. [2](#Fig2){ref-type="fig"}) showed a characteristic (001) reflection of layered structure at approximately 2θ of 5.98°. The (001) reflection of sulfuric acid-treated montmorillonite (S-MMT) retained with a slight shift to 2θ of 5.61° and appeared less intense than that of parent MMT^[@CR23]^.Figure 2XRD patterns of MMT and S-MMT catalysts. The other characteristic reflections at 2θ 12.5°, 17.7°, 19.7°, 22.08°, 23.7°, 26.54°, 29.94°, 36.56°, 50.01°, 62.0° assigned to the crystal lattice planes of d 060 (0.502 nm), 080 (0.451 nm), 020 (0.378 nm), 060 (0.303 nm), 030 (0.171 nm), 050 (0.151 nm) and 050 (0.148 nm) respectively^[@CR24]^, remained clearly identified in both MMT and S-MMT. However, a significant increase in the intensity of peaks at 2θ 12.5°, 17.7° & 26.54° was noted in the XRD pattern of sulfuric acid-activated montmorillonite. The persistence of all the above diffraction peaks in S-MMT clearly documents that the parent MMT crystalline structure was intact and sulfuric acid treatment did not considerably alter the layered structure of montmorillonite. For the samples of 0.5Pt/S-MMT, 1Pt/S-MMT, 2Pt/S-MMT & 3Pt/S-MMT, all the crystal lattice planes appeared almost unchanged (Supplementary Fig. [S2](#MOESM1){ref-type="media"}). The intercalation of Pt metal nanoparticles onto the layered structure of MMT material showed no impact on its crystallinity, however, significant increase in peak intensity was observed at 2θ 17.7° & 26.54° as the Pt loading was increased. The characteristic peaks corresponding to the formation of metallic Pt (0) phase that generally appears at 2θ 40.5°, 47.0°, and 68.6°, allocated to (111), (200), and (220) reflections of face-centered cubic (fcc) platinum lattice, were not seen in the XRD pattern (12). This could be due to the high dispersion of Pt on surface of MMT and the crystallite size may possibly be very low (\<4 nm), which is below the detection limit of XRD. Thus, XRD results infer that the general crystalline structure of montmorillonite was well retained with no change in d-spacing and intercalation structure, even after the sulfuric acid treatment and the introduction of Pt particles into the inter lamellar structure of montmorillonite. ### Fourier transform-infrared spectral analysis {#Sec5} The FTIR spectra of parent MMT, S-MMT and 0.5--3 wt% Pt incorporated samples are displayed in Fig. [S3](#MOESM1){ref-type="media"} (Supplementary). All the samples exhibited the absorbance bands representative of montmorillonite^[@CR25]^. The absorbance bands at 526 cm^−1^ and 462 cm^−1^ were attributed to the Al-O stretching and Si-O bending vibrations of Si-O-Al groups, respectively. While those peaks around 818 cm^−1^ and 916 cm^−1^ corresponds to Mg-OH-Al and Al-O-Al bending vibrations respectively, the bands at about 1050 cm^−1^ and 1164 cm^−1^ were attributed to Si-O stretching vibration. The OH vibrations arising from hydrated cations in the interlayer appeared at about 1474 and 1378 cm^−1^. The absorbance bands at 3648 cm^−1^ and 1638 cm^−1^ resulted from --OH stretching and bending vibrations respectively. Therefore, the acid activation and Pt incorporation did not destruct the layered structure of montmorillonite and no significant change in the structure, as evident from FTIR spectra. ### BET surface area and pore volumes {#Sec6} As summarized in Table [S1](#MOESM1){ref-type="media"} (Supplementary), the specific surface area of MMT catalyst calculated from BET equation was found to be 188 m^2^/g. Sulphuric acid-activated MMT catalyst was found to exhibit a higher surface area of about 206 m^2^/g which might be ascribed to slight alteration in the layered structure of montmorillonite caused by leaching of Al^3+^ ions or any other impurities by acid treatment. In addition, S-MMT was found to possess larger pore volumes of 0.93 cc/g as compared to MMT which had a pore volume of 0.26 cc/g. The increase in both specific surface area and total pore volume is a crucial element to confirm the process of acid activation^[@CR24]^. This result corroborates with decrease in (001) reflection found from XRD analysis. However, for platinum loaded catalysts (Pt/S-MMT), the surface area and pore volume were found to be reduced to 198 m^2^/g and 0.79 cc/g when compared to S-MMT catalyst, and a gradual decrease was observed with increase in Pt loading. This attributes to the pore blockage of MMT with Pt or extra framework species. ### CO-Chemisorption {#Sec7} The platinum dispersion, average particle size and metal surface area has been determined by chemisorption of CO on a series of 0.5--3 wt% Pt/S-MMT catalysts. The results (Supplementary Table [S2](#MOESM1){ref-type="media"}) demonstrate that as the Pt content in the catalyst increases, the percentage of platinum dispersion was lowered with obvious rise in CO uptake volume. However, the particle size of platinum increased with loading and was found to be in the range of 1.9--4.5 nm. The decrease in dispersion and increase in particle size with Pt loading was probably due to the agglomeration of platinum, promoted by excessive deposition of platinum particles on the surface of MMT, thereby, decreasing the distance between platinum particles. Chary *et al*.^[@CR20]^ in their study on glycerol hydrogenolysis over Pt-WO~3~/SBA-15 catalysts observed that the addition of tungstate species on to SBA-15 has significantly lowered the Pt dispersion. The results of CO-chemisorption are well correlated with those acquired from XRD and TEM analysis. The loading of Pt in a series of 0.5--3 wt% Pt/S-MMT catalysts was confirmed by Inductive Coupled Plasma-Atomic Emission Spectrometer (ICP-AES) and the results are presented in Table [S2](#MOESM1){ref-type="media"} (Supplementary). ### Transmission Electron Microscopic Analysis {#Sec8} Figure [3](#Fig3){ref-type="fig"} illustrate the TEM micrographs and particle size distribution of the Pt incorporated sulphuric acid-activated montmorillonite catalysts. The images clearly show that the Pt nanoparticles are spherical, fine and well dispersed on the exterior of montmorillonite.The average particle size of platinum in 0.5 wt% and 1 wt% Pt/S-MMT catalysts is found to be around 2.4 nm while platinum size in 2 Pt/S-MMT and 3 Pt/S-MMT catalysts is 3.9 nm and 4.5 nm respectively. At lower loadings, platinum nanoparticles seems to be well dispersed on the support and at higher loadings platinum nanoparticles tend to aggregate forming larger particles. The TEM image of PVP capped platinum nanoparticles is presented in Fig. [S4](#MOESM1){ref-type="media"} (Supplementary) and the particle size of platinum nanoparticles is in the range of 2.3--5 nm which is consistent with the sizes observed in supported platinum catalysts.Figure 3Transmission electron micrographs of different Pt/S-MMT catalysts. ### Acidity of catalysts {#Sec9} The acid strength of MMT, S-MMT and 0.5--3 wt% Pt/S-MMT catalysts was determined by ammonia - temperature programmed desorption (NH~3~-TPD) analysis. Based on the temperature regions of ammonia desorption, the acid sites in the catalysts were distinguished as weak (100--200 °C), medium (200--400 °C) and strong (400--700 °C) acid sites. The sum of all the three acid sites measured by the total ammonia desorbed gives the total acidity of the catalysts. The results are compiled in Table [1](#Tab1){ref-type="table"} and the TPD profile is displayed as Fig. [S5](#MOESM1){ref-type="media"} (Supplementary). As shown in the profile, the raw MMT displayed two peaks - one in the low temperature region and the other in high temperature region corresponding to weak and strong acid sites respectively^[@CR26]^. The desorption of NH~3~ in the low temperature region might be due to the physical adsorption of NH~3~ in the interlayer space of MMT, whereas the NH~3~ desorbed in the high temperature region could be ascribed to the framework Al sites of MMT.Table 1Acidities of MMT, S-MMT and various Pt/S-MMT catalysts.CatalystNH~3~ uptake (μmol/g)Total acidities\ (μmol/g)WeakModerateStrongMMT159---203362S-MMT224641454330.5Pt/S-MMT143981393801Pt/S-MMT15443692662Pt/S-MMT16728502453Pt/S-MMT1221535172 Interestingly, the NH~3~-TPD profile of S-MMT catalyst showed the presence of an additional peak at moderate temperature region attributing to medium acid sites. This indicates that acid activation of MMT has induced the generation of moderate acid sites in the catalysts. Similarly, three types of acid sites have been identified in Pt/S-MMT catalysts of varying Pt loadings (0.5--3 wt%). It is notable that with increase in the Pt loading, the extent of desorption was shifted more toward lower temperature which means that there was a loss in both moderate and strong acid sites with substantial increase in the weak acid sites. This infers that addition of platinum has facilitated the formation of weak acid sites maximizing at 2 wt% Pt. Furthermore, the total acidity of catalysts was found to decrease with increase in Pt loading. This is because the addition of platinum on sulphuric acid-treated montmorillonite blocks the Brønsted acid sites although marginally increases the Lewis acidity and also due to the possible agglomeration or bulk nature of platinum on support at higher loadings. The presence of Lewis (L) and Brønsted (B) acid sites in MMT, S-MMT and 0.5--3 wt% Pt/S-MMT catalysts was confirmed by FTIR analysis of corresponding pyridine adsorbed species. The vibration frequencies associated to the chemical properties of acid sites are shown in the spectra (Fig. [4](#Fig4){ref-type="fig"}). The FTIR spectrum of pyridine adsorbed MMT catalyst did not reveal the occurrence of prominent acid sites which indicates the inherent weak acidity of MMT. Whereas the spectrum of pyridine adsorbed S-MMT catalyst shows well resolved significant bands at 1428 cm^−1^ for pyridine associated to Lewis acid sites; 1480 cm^−1^ due to both Lewis and Brønsted acid sites (total acid sites) and 1540 cm^−1^ due to pyridinium linked to Brønsted acid sites^[@CR27]^. A similar pattern was observed in the spectra of pyridine adsorbed 0.5--3 wt%) Pt/S-MMT catalysts. The peak intensity corresponding to Brønsted acid sites (at 1540 cm^−1^) in all the catalysts is relatively higher compared to Lewis acid sites indicating that the concentration of Brønsted acid sites is high in the catalysts. The NH3-TPD and Pyr IR results therefore clearly show that the acid activation of MMT has enhanced the surface acidity of the catalysts, and essentially more of Brønsted acid type.Figure 4FTIR spectra of pyridine adsorbed MMT, S-MMT and various Pt/S-MMT catalysts. ### SEM-EDX {#Sec10} The topographical features and a semi quantitative elemental analysis of the catalysts has been studied by Scanning Electron Microscopy (SEM) coupled with Energy Dispersive X-ray spectroscopy (EDX). The SEM images and EDX profiles of MMT, S-MMT and 2Pt/S-MMT catalysts are presented in Fig. [S6](#MOESM1){ref-type="media"} (Supplementary). The typical SEM image of MMT catalyst displayed larger aggregates (Supplementary Fig. [S6a](#MOESM1){ref-type="media"}) whereas that of S-MMT showed the development of pores on the surface with slight variation in the crystal morphology^[@CR27]^. In case of 2Pt/S-MMT catalyst, the aggregates appeared smaller in size compared to MMT and S-MMT catalysts. The elemental analysis data for all the catalysts is listed in Table [S3](#MOESM1){ref-type="media"} (Supplementary). For all the catalysts, SiO2 and Al2O3 are the major components, consistent with the main structure of montmorillonite. The element sulpur (S) was identified in the chemical spot analysis of sulphuric acid-activated MMT catalyst. The Pt/S-MMT catalysts revealed the presence of Pt and proves the incorporation of Pt in to the catalysts. As evident from the results, the atomic percentage of Pt in the catalysts corresponded to the Pt loading that was used during the catalyst preparation procedure. ### ^27^Al NMR spectroscopy {#Sec11} The ^27^Al nuclear magnetic resonance (NMR) spectroscopy has been used to detect the coordination of aluminium within the interlayer structure of montmorillonite. The ^27^Al NMR spectra (Fig. [5](#Fig5){ref-type="fig"}) of pure MMT exhibited an intense peak at 3.56 ppm corresponding to octahedral coordinated Al and a small peak at 67.39 ppm due to tetrahedral coordinated Al in the framework of montmorillonite structure. However, acid activation of montmorillonite has resulted in shift of octahedral Al peak to 1.81 ppm with a decreased intensity while the tetrahedral Al peak remained intact^[@CR28],[@CR29]^. This indicates leaching of octahedral Al during acid activation and stable tetrahedral structure of aluminium. The spectra of 2Pt/S-MMT catalyst appeared similar to that of S-MMT catalyst with no considerable change in the chemical shifts.Figure 5^27^Al NMR spectra of MMT, S-MMT and 2Pt/S-MMT catalysts. Catalytic performance {#Sec12} --------------------- The hydrogenolysis of glycerol is a multifaceted process in which acid sites and metal sites of the catalyst are indeed the crucial elements for the selective formation of 1,2-PDO and 1,3-PDO. Glycerol conversion to 1,3-PDO was proposed to initiate via Brønsted acid catalyzed dehydration to 3-hydroxypropionaldehyde (3-HPA) followed by metal catalyzed hydrogenation to 1,3-PDO^[@CR10],[@CR21],[@CR30],[@CR31]^. To establish the validity of this hypothesis, several experiments of glycerol hydrogenolysis were performed using various catalysts including the parent montmorillonite, sulphuric acid-activated montmorillonite and a series of 0.5--3 wt% Platinum on sulphuric acid-activated montmorillonite catalysts to investigate the glycerol conversion and product selectivity. The hydrogenolysis of glycerol was performed in a continuous flow fixed-bed reactor set-up. The reactions were typically carried out at a temperature of 200 °C and a hydrogen pressure of 1 bar using 10 wt% aqueous glycerol and 0.5 g of catalyst. The catalytic activity results for various catalysts are summarized in Table [2](#Tab2){ref-type="table"}. The products from glycerol hydrogenolysis included 1,3-PDO as the major while 1,2-PDO, acrolein, and hydroxyacetone are the secondary products (Supplementary Fig. [S7](#MOESM1){ref-type="media"}). Other minor products such as acetaldehyde, 1-propanol, 2-propanol and acetone were also detected.Table 2Catalytic performance of MMT, S-MMT and 0.5--3 wt% Pt/S-MMT catalysts in glycerol hydrogenolysis^a^.CatalystConversion (%)Selectivity (%)1,3-PDO1,2-PDOAcroleinHAOthersMMT36---16.0352920S-MMT701.22.8848.04.00.5Pt/S-MMT795118.9206.04.11Pt/S-MMT865716214.31.72Pt/S-MMT946212143.18.93Pt/S-MMT945813173.09.0^a^Reaction conditions: 10 wt% glycerol aqueous solution; 0.5 g of catalyst; reaction temperature of 200 °C, 1 bar H~2,~ H~2~ flow rate of 70 mL min^−1^; 1,3-PDO: 1,3-propanediol, 1,2-PDO: 1,2-propanediol, HA: Hydroxyacteone, Others include acetaldehyde, 1-propanol, acetone, 2-propanol. When the parent MMT was tested in glycerol hydrogenolysis, the glycerol conversion was fairly low with acrolein being the major product and substantial amounts of degradation products. Sulphuric acid-activated MMT catalyst resulted in increase in glycerol conversion of 70% with a high selectivity to acrolein at around 84% and less selectivity to propanediols (\~3%). This infers that the acid treatment of MMT resulted in generation of high strength Brønsted acidic sites in the catalyst, evident from Pyr IR, and has enabled the double dehydration of glycerol to produce acrolein. A similar finding has been reported by Zhou *et al*^[@CR24]^. The different platinum catalysts were then tested for their activity for glycerol transformation to 1,3-PDO. It is worth noting that the glycerol conversion and selectivity to 1,3-PDO was greatly improved over sulphuric acid-activated montmorillonite supported platinum catalysts compared to the parent MMT and S-MMT. ### Influence of Pt loading {#Sec13} To gain an insight into the 1,3-PDO selectivity, the effect of platinum loading (0.5--3 wt%) was studied under the standard reaction conditions. As the Pt loading is increased from 0.5--2 wt%, there was a gradual increase in the glycerol conversion from 79% to 94% and selectivity to 1,3-PDO from 51% to 62% (Table [2](#Tab2){ref-type="table"}). Over 3 wt% Pt/S-MMT catalyst, the selectivity to 1,3-PDO marginally reduced while the glycerol conversion remained the same. The maximum glycerol conversion and selectivity to 1,3-PDO was achieved over 2Pt/S-MMT catalyst, meaning that 2 wt% Pt was adequate to hydrogenate 3-HPA formed over S-MMT, to 1,3-PDO. Therefore, among all the other catalysts screened, 2 wt% Pt/S-MMT was found to be the optimal and most active catalyst for glycerol transformation to 1,3-PDO, under the standard conditions, and hence chosen for further reaction optimization experiments. Similar findings on the predominant effect of 2 wt% Pt in glycerol hydrogenolysis have been demonstrated earlier^[@CR32],[@CR33]^. ### Reaction parametric study over 2Pt/S-MMT catalyst {#Sec14} The effect of reaction temperature: Figure [6a](#Fig6){ref-type="fig"} represents the results of hydrogenolysis of glycerol over 2Pt/S-MMT catalyst conducted at different reaction temperatures ranging from 160--220 °C and atmospheric pressure. As the reaction temperature elevated from 160 °C to 220 °C, there was a steady increase in the glycerol conversion from 65% to 96%. The selectivity to 1,3-PDO gradually increased until a reaction temperature of 200 °C beyond which it slightly lowered. On contrary, a steady decrease in 1,2-PDO selectivity has been observed when the reaction temperature accelerated. This is ascribed to the fact that higher reaction temperatures would stimulate additional carbon-oxygen and carbon-carbon bond scission in glycerol leading to other byproducts such as propanols, hydroxyacetone, acrolein and acetaldehyde with decreased 1,3-PDO selectivity. This is in accordance with the previous reports^[@CR31],[@CR34]^. 2Pt/S-MMT catalyst presented almost 94% conversion of glycerol and a maximum of 62% selectivity to 1,3-PDO at 200 °C, considering it to be the appropriate and optimum reaction temperature to promote glycerol hydrogenolysis.Figure 6Effect of reaction temperature (**a**), hydrogen flow rate (**b**), weight hourly space velocity (**c**) and contact time (**d**) on glycerol conversion and product selectivity over 2Pt/S-MMT catalyst. Reaction conditions: 10 wt% glycerol aqueous solution; 0.5 g of catalyst;, 1 bar H~2~; 1,3-PDO: 1,3-propanediol, 1,2-PDO: 1,2-propanediol, HA: Hydroxyacteone, Others include acetaldehyde, 1-propanol, acetone, 2-propanol. The effect of of hydrogen flow rate: Several experiments were carried out to check the effect of hydrogen flow rate on hydrogenolysis of glycerol at 200 °C and different H~2~ flow rates of 30, 50, 70 & 90 mL min^−1^ over 2Pt/S-MMT catalyst. The results are summarized in Fig. [6b](#Fig6){ref-type="fig"}. As the hydrogen facilitates glycerol hydrogenolysis reaction, it is obvious that, with rise in H~2~ flow rate, there happened a steep rise in the glycerol conversion. While 1,3-PDO selectivity has gradually increased with an increase in H~2~ flow rate from 30 to 70 mL min^−1^, it must be noted that, increasing H~2~ flow rates suppressed the formation of 1,2-PDO and acrolein favoring 1,3-PDO from glycerol. However, further rise to 90 mL min^−1^ H~2~ flow rate, the selectivity to 1,3-PDO dropped a bit though the glycerol conversion increased. A possible explanation is that high H~2~ flow rate probably promoted the formation of corresponding excessive glycerol hydrogenolysis products such as ethanol, 1-propanol, 2-propanol and acetaldehyde. Thus, for both glycerol conversion and 1,3-PDO selectivity, the optimal H~2~ flow rate was 70 mL min^−1^. This trend of results is in perfect agreement with the past reports^[@CR31],[@CR35]^. The effect of weight hourly space velocity (WHSV): The flow rate of glycerol is expected to have a significant effect on the glycerol hydrogenolysis and to examine this, reactions over the most active catalyst were carried out at 200 °C by changing the WHSV of 10 wt% glycerol from 1.02--4.08 h^−1^. Figure [6c](#Fig6){ref-type="fig"} showed that with rising WHSV, the glycerol conversion decreased significantly from 94% to 79%, together with a steady decline of 1,3-PDO selectivity from 62% to 41%. This suggests that as the glycerol feed flow increases, the number of active sites to react with glycerol decreases and most likely not adequate to convert excess glycerol^[@CR32]^. The other product distribution showed that the selectivity towards acrolein and 1,2-PDO showed uniform increase with the continuously increasing WHSV. Thus, the results showed that the lower WHSV (1.02 h^−1^) was indeed efficient to promote the selective glycerol hydrogenolysis to 1,3-PDO with a maximum selectivity of 62% at 94% glycerol conversion. The effect of contact time (W/F): The ratio of catalyst weight to the feed flow rate is important to optimize the reaction process and to achieve maximum product selectivity and reactant conversion. To understand the effect of contact time (W/F) on the activity and selectivity, experiments were run by varying the weight of the catalyst at constant glycerol feed flow rate. As shown in the Fig. [6d](#Fig6){ref-type="fig"}, the selectivity to 1,3-PDO and glycerol conversion and enhanced steadily with increase in the contact time. This is because, higher the contact time, greater is the interaction between the catalyst active sites and the glycerol that would favor the selective production of 1,3-PDO and simultaneously increase glycerol conversion. However, the opposite trend was observed in case of 1,2-PDO and acrolein selectivity. The results conclude that a higher contact time of 1.0 g mL^−1^ h was found to be ideal for attaining the maximum glycerol conversion and 1,3-PDO selectivity. The observation is consistent with the literature^[@CR20]^. ### Space Time Yield (STY) {#Sec15} The quantity of 1,3-PDO produced per 0.5 g of 2Pt/S-MMT catalyst per one hour of reaction time refers to the space time yield (STY) of 1,3-PDO. The STY of 1,3-PDO has been studied by varying the glycerol concentration (5--20 wt%) of the experiments performed at a reaction temperature of 200 °C, 70 mL/min H~2~ flow rate and atmospheric pressure. The following equation was employed to calculate the space time yield (STY) of 1,3-PDO:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${STY}={{M}}_{1{,}3 \mbox{-} {PDO}}/{{M}}_{{cat}}\times {t}$$\end{document}$$where M~1,3-PDO~ is the mass of 1,3-PDO produced (g); M~cat~ is the mass of catalyst (g); and t is the reaction time (h). The space time yield towards 1,3-PDO is plotted as a function of glycerol concentration and the results are presented in Fig. [S8](#MOESM1){ref-type="media"} (Supplementary). As can be seen from the plot, as the glycerol concentration rises from 5 to 20 wt%, the space time yield towards 1,3-PDO and glycerol conversion increased up until 10 wt% after which a fall has been observed. With increase in the glycerol content (beyond 10 wt%), the viscosity of the feed increases which might reduce the mass transfer performance of the glycerol and 1,3-PDO on the catalyst surface^[@CR36]^, thereby lowering 1,3-PDO selectivity and the glycerol conversion. The optimal glycerol concentration for the reaction was found to be 10 wt% when the glycerol conversion and the 1,3-PDO selectivity were considered together. ### Glycerol hydrogenolysis reaction mechanism over Pt/S-MMT catalyst {#Sec16} The structure-activity correlation indicated that the glycerol hydrogenolysis to 1,3-PDO reaction over a bifunctional Pt/S-MMT catalyst proceeds through a dehydration-hydrogenation mechanism in two different routes. As presented in Fig. [7](#Fig7){ref-type="fig"}, Route 1 is a three-step process in which glycerol undergoes double dehydration over Brønsted acidic sites of S-MMT to produce acrolein and the rehydration of acrolein gives 3-hydroxypropionaldehyde (3-HPA) which then hydrogenates over platinum sites to produce 1,3-PDO. In Route 2, 1,3-PDO is formed via two-step dehydration of glycerol to 3-HPA and its subsequent hydrogenation to 1,3-PDO. A similar mechanism has been proposed in recent work published by Mahesh *et al*^[@CR37]^.Figure 7The possible routes of 1,3-PDO formation in vapour phase glycerol hydrogenolysis over Pt/S-MMT catalyst in a fixed bed reactor under atmospheric pressure. ### Catalyst stability and reusability {#Sec17} The stability of 2Pt/S-MMT catalyst with time-on-stream over was studied for a period of 1--8 h reaction under the optimized reaction conditions as follows: reaction temperature: 200 °C; glycerol concentration = 10 wt.%, WHSV = 1.02 h^−1^, and H~2~ flow rate = 70 mL min^−1^. It is observed that there is a gradual rise in glycerol conversion from 86% to 94% during 1--3 h and then remained unchanged until 5 h; after which it declined slowly to 80% (Supplementary Fig. [S9](#MOESM1){ref-type="media"}[a](#MOESM1){ref-type="media"}). A similar trend was observed in case of 1,3-PDO selectivity, during 8 h period with a maximum selectivity of 62% being achieved at 3 h. The decline in the activity during the latter hours could be probably due to the deactivation of the catalyst which was probably the result of glycerol polymerization or further hydrogenolysis of 1,3-PDO on the catalyst. To confirm this, the spent catalyst was re-activated at 500 °C for 3 h in an air flow (100 mL min^−1^), to remove the carbon species deposited on the catalysts. The regenerated/reactivated catalyst (labelled as 2Pt/S-MMT R) was then tested in a second time-on-stream studies (Supplementary Fig. [S9](#MOESM1){ref-type="media"}[b](#MOESM1){ref-type="media"}). The results showed a better recovery of the catalytic performance, with a similar deactivation trend as that of the fresh catalyst. However, there was a little drop in the glycerol conversion (90%) and 1,3-PDO selectivity (60%) over the regenerated catalyst. The drop in the activity could be due to the change in spent catalyst morphology^[@CR38]^ including decrease in the total acidity and BET surface area, verified by NH~3~-TPD and BET SA analyses; pore blockage by coke deposition, as shown from CHNS analysis (Supplementary Table S4). The acidity of spent (deactivated) catalyst was found to be considerably lower than that of fresh catalyst which is due to the blockage of acid sites with carbonaceous species. However, it is notable that the reactivated catalyst showed a marginal decrease in the acidity to that of fresh catalyst and has resulted in the better catalytic performance. Therefore, NH~3~-TPD results suggests that the acidity of a catalyst plays a crucial role in determining the catalytic activity for hydrogenolysis of glycerol^[@CR21]^. The thermogravimetric analysis (TGA) of fresh and spent 2Pt/S-MMT catalysts was performed at 10 °C/min over a temperature range of 50 °C to 750 °C under a constant flow of nitrogen (20 mL/min) to further understand the mode of catalyst deactivation. The degradation pattern of fresh and spent 2Pt/S-MMT catalysts (Supplemenatry Figure [S10](#MOESM1){ref-type="media"}) showed two well-defined mass loss regions; the first one between 50 °C and 200 °C is attributed to the loss of physisorbed water from the interlayer spacing of montmorillonite and the second mass loss above 400 °C. is due to the dehydration of coordinated water molecules and dehydroxylation of silicate structure. However, a total of about 23 wt% mass loss was observed for the spent catalyst over the range of 50 °C to 750 °C, compared with only \<5 wt% mass loss for the fresh catalyst. This result is a clear evidence of the coke formation and deposition of undesired materials due to glycerol polymerization on the spent catalyst, which contributed to the catalyst deactivation by blocking the catalyst active sites. Discussion {#Sec18} ========== This research work demonstrated a novel approach of catalyst design comprising of 0.5--3 wt% Pt nanoparticles immobilized on sulphuric acid-activated montmorillonite, investigated first time for the selective production of 1,3-PDO by hydrogenolysis of biomass derived glycerol. The activation of montmorillonite by sulphuric acid did not obviously alter the inter layer structure of montmorillonite, as evidenced from XRD, FTIR and ^27^Al NMR but enhanced the acid catalytic performance by generation of new acid sites preferentially of Brønsted acid type, revealed from NH~3~-TPD and Pyr IR analyses. Among the various tested catalysts, 2Pt/S-MMT catalyst exhibited superior performance with 62% selectivity to 1,3-PDO at 94% glycerol conversion under reasonably mild reaction conditions. The results from efficient characterization and the catalytic experiments clarify that Brønsted acid sites generated by acid activation of the catalyst promoted selective abstraction of secondary --OH group of glycerol favoring the formation of 1,3-PDO and the highly dispersed nanosize Pt particles, confirmed from CO-chemissorption and TEM studies, enabled the hydrogenation of 3-hydroxypropanal. The catalytic performance varied based on the reaction conditions and upon optimization, the favorable reaction conditions to accomplish the best result over the most active 2Pt/S-MMT catalyst, were found to be a reaction temperature of 200 °C, 0.1 MPa H~2~, 70 mL min^−1^ H~2~ flow rate, a WHSV of 1.02 h^−1^ and a glycerol concentration of 10 wt%. The catalyst offered good stability and was successfully reused in the consecutive cycle without a larger decline in the activity. Methods {#Sec19} ======= Catalyst Preparation {#Sec20} -------------------- ### Preparation of sulphuric acid-activated montmorillonite catalyst {#Sec21} The montmorillonite clay (henceforth designated as MMT) obtained from Bentonite deposit, China, was used as received without further purification (BET-surface area \~183 m^2 ^gm^−1^). Concentrated H~2~SO~4~ (98.0%, Sigma Aldrich Co., Ltd) was used to activate montmorillonite clay. The dried montmorillonite powder was activated in an aqueous solution of 10 wt% sulfuric acid at 80 °C with constant stirring for 4 h. After reaction, the acid-activated solid was separated by filtration, then washed three times with distilled water, and air-dried at 100 °C for 48 h^[@CR39]^. The solid was then ground to fine powder (30 µm) for further use and labeled as S-MMT. ### Preparation of platinum nanoparticles {#Sec22} Pt nanoparticles (NPs) with an average size of 2.5--5 nm were synthesized following the procedure reported elsewhere^[@CR40]^. In a typical synthesis, tetraammine platinum (II) nitrate \[Pt(NH~3~)~4~(NO~3~)~2~\] was used as a Pt precursor and a polymer poly(vinylpyrrolidone) (PVP) was employed as a capping agent. Various loadings (0.5, 1, 2 & 3 wt %) of Pt NPs were prepared by dissolving required amount of Pt in ethylene glycol and PVP. The solution could boil at solvent temperatures. The as-synthesized PVP-capped Pt NPs were then washed and dispersed in ethanol to give a colloidal solution of Pt NPs. ### Preparation of sulphuric acid-activated montmorillonite supported platinum catalysts {#Sec23} For immobilizing Pt NPs on S-MMT, the colloidal solution of 0.5--3 wt % Pt NPs was added to the S-MMT under stirring. The colloidal suspension was then sonicated for 5 h at room temperature using a UC25-12TPA Ultrasonic cleaner. The resultant solution is centrifuged to obtain brown precipitates, washed with ethanol. Later, the precipitates were dried in an oven at 80 °C overnight, followed by calcination in a muffle furnace at 500 °C for 3 h. All the prepared catalytic materials are hereafter labelled as 0.5Pt/S-MMT, 1Pt/S-MMT, 2Pt/S-MMT, and 3Pt/S-MMT where the number represents the weight percentage (wt%) of Pt. Catalyst characterization {#Sec24} ------------------------- The crystallinity and identification of catalyst phases was studied by powder X-ray diffraction (XRD) analysis performed on Rigaku's Miniflex X-ray diffractometer by means of Ni filtered Cu Kβ radiation (λ = 1.392 Å). The measurements were recorded in the 2θ range of 2 to 80°, scanning at 2° min^−1^. A beam current of 15 mA and a beam voltage of 40 kV were used for the measurements. The FTIR spectra of catalysts were recorded in transmission mode on a UATR Perkin Elmer Spectrum Two instrument. Spectra were acquired in the spectral range of 4000--400 cm^−1^ at a resolution of 4 cm^−1^ and accumulation of 32 scans. For the analysis of Brønsted and Lewis acid sites in the catalysts, infrared spectra of pyridine adsorbed catalysts (Pyr IR) was acquired in the range of 1600--1400 cm^−1^. For the pyridine adsorption, the catalysts undergo an hour pretreatment under N~2~ at 300 °C for moisture removal followed by adsorption of pyridine at 120 °C for further hour until saturation. The catalysts were then cooled down to 30 °C and pelletized with KBr to record the spectra. The amount of Pt in all Pt/S-MMT catalysts was quantitatively analyzed by Inductively coupled plasma atomic emission spectrometry (Agilent Technologies-4200MP-AES). The sample preparation includes acid digestion of catalyst (\~10 mg in 2 mL aquaregia) at 60 °C and dilution to desired concentration. Nitrogen adsorption-desorption isotherms were acquired on Quantachrome Autosorb 1 instrument by degassing the samples at 250 °C for 6 h under vacuum and carrying out the analysis at −196 °C under liquid nitrogen. The specific surface areas of catalysts were calculated using the multi-point Brunauer--Emmet--Teller (BET) method. The average pore diameter and pore volumes were measured by adsorption curve analysis using the Barrett--Joyner--Halenda model (BJH) method. The strength of acid sites in raw MMT, S-MMT and Pt/S-MMT catalysts were determined by NH~3~-TPD measurement using Autochem 2910 (Quantachrome) fitted with a thermal conductivity detector. The samples were prepared by outgassing at 200 °C for 1 h under He flow (50 mL min^−1^). NH~3~ gas was passed through the catalyst until complete saturation for about 30 min and then physisorbed NH~3~ was removed by passing He (50 mL min^−1^) for another 30 min. NH~3~-TPD analysis for continuous monitoring of desorbed NH~3~ was carried out by heating the samples in the temperature range from 100--700 °C at 10 °C min^−1^ under He flow (50 mL min^−1^). The volume of desorbed NH~3~ was measured using GRAMS/32 software. The morphology and elemental investigation of the catalysts were performed using a Phenom XL Scanning Electron Microscopy coupled with an Energy Dispersive X-Ray Analyzer (SEM-EDX). Each catalyst sample was deposited on a carbon tape adhered to an aluminum stub prior to analysis by SEM. The micrographs of the catalysts were captured at 10 kV beam voltage and 5,000× magnification using a backscatter electron detector (BSD). The elemental analysis was performed at high resolution (15 kV of beam voltage) and high vacuum pressure (1 Pa) with a secondary electron detector (SED), where the point and mapping analysis for element identification was performed using a Phenom Pro Suite software. The percentage of platinum dispersion was measured by CO-chemisorption experiments typically carried out in an AutoChem 2910 instrument (Micromeritics) associated with a thermal conductivity detector. Average particle size was also verified during the CO-chemisorption measurements. The experiment begins with reduction of 100 mg of the sample at 300 °C under H~2~ flow (50 mL min^−1^) for 3 h, followed by He treatment at 300 °C for 1 h after which the sample is brought to 30 °C. A number of pulses of 9.96% CO balanced helium were injected in regular intervals over the reduced catalyst to measure the volume of CO uptake by the catalyst. The skeletal aluminium framework structure of montmorillonite catalysts was examined using ^27^Al MAS NMR technique performed on DD2 Oxford Magnet AS-500MHz spectrometer (Agilent Technologies) using aluminium oxide (Aldrich) as a reference material. The chemical shifts (δ) are mentioned in ppm. The microstructure of Pt catalysts was determined by Transmission electron microscopy (TEM) performed on JEOL 2010 electron microscope operating at 200 KV. A thin catalyst suspension is made by ultra-sonicating tiny amount of powder sample in ethanol and a drop of it is dispersed on copper grids. Samples were positioned in the microscope column, at evacuation of less than 1 × 10^−6^ Torr. The thermogravimetric analysis (TGA) of the fresh and spent catalysts was performed on a Perkin Elmer TGA 7 unit. The measurements were performed over a temperature range of 50 °C to 750 °C at 10 °C/min under a constant flow of nitrogen (20 mL/min). Reaction experiment {#Sec25} ------------------- The evaluation of catalysts in the hydrogenolysis of glycerol was performed in an upright continuous-fed fixed bed quartz reactor with dimensions of 9 mm internal diameter and 40 cm length. All reactions were carried out in vapour phase under atmospheric pressure. For each experiment, 0.5 g of catalyst mixed with quartz beads was loaded into the middle portion of the reactor and the temperature at the catalyst bed was monitored by a thermocouple. The catalyst bed was heated to 350 °C at an incremental rate of 10 °C min^−1^ in a flow of pure H~2~ (50 mL min^−1^) to perform catalyst reduction for about 2 h. The temperature of the reactor was then lowered to the required reaction temperature and an aqueous solution with a ratio of glycerol to water at 10:90 (w/w) was pumped into the reactor at a flow rate of 0.5 mL h^−1^ using a B-Braun syringe pump together with a stream of H~2~ at a flow rate of 70 mL min^−1^. The unreacted glycerol and the reaction products were condensed and collected every hour in a cold trap (0 °C). The collected samples were analyzed using a gas chromatography (Shimadzu GC 2014) equipped with an FID and DB-wax capillary column (Agilent, 0.32 mm × 32 m). The conversion of glycerol and selectivity to products were calculated using the following equations:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{Conversion}}\,( \% )=({\rm{moles}}\,{\rm{of}}\,{\rm{glycerol}}\,{\rm{reacted}})/({\rm{moles}}\,{\rm{of}}\,glycerol\,{\rm{fed}})\times {\rm{100}}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{Selectivity}}\,( \% )=({\rm{moles}}\,{\rm{of}}\,{\rm{one}}\,{\rm{product}})/({\rm{moles}}\,{\rm{of}}\,{\rm{all}}\,{\rm{products}})\times {\rm{100}}$$\end{document}$$ Data Availability {#Sec26} ----------------- The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. Electronic supplementary material ================================= {#Sec27} Supplementary Information **Electronic supplementary material** **Supplementary information** accompanies this paper at 10.1038/s41598-018-25787-w. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The authors would like to acknowledge Industry Connect Seed Fund support from the Monash Energy Materials and Systems Institute and Discovery Seed Fund Scheme from Monash Engineering. S.S.P. planned the research, carried out the experimental work and drafted the manuscript. S.K. formatted the manuscript and prepared supplementary document. S.B. reviewed the manuscript. Competing Interests {#FPar1} =================== The authors declare no competing interests.

1. Introduction {#sec1} =============== The exact prevalence of pediatric heart failure is largely unknown but it is increasing. Recent advances in diagnostic methods, new surgical techniques, and improved perioperative management have contributed to an increased survival for patients with complex congenital heart disease who today often survive into adulthood \[[@B1]\]. Even after successful repair, however, right ventricle (RV) pressure overload remains in some patients and eventually impairs RV function and influences long-term mortality and morbidity \[[@B2]--[@B4]\]. Although compensated hypertrophy develops initially, ultimately RV failure will occur. The mechanisms underlying the progression from compensated to decompensated RV hypertrophy have not been well defined \[[@B5]\]. Clinically, the relationship between progressive fibrosis and RV function must be addressed \[[@B6], [@B7]\]. But studies of the mechanisms underlying the transition from a compensated state of hypertrophy to a decompensated state are difficult in patients, because invasive data cannot be easily obtained. For this purpose, animal models may be beneficial. Small experimental animals, such as rats, are widely used in cardiovascular research since they can provide a range of disease models, including cardiac hypertrophy and failure. A major advantage of these disease models is that cardiac material can be easily sampled to study the pathology of the disease in question. The partial pulmonary artery ligation procedure has been widely used to produce right heart failure caused by pressure overload, but this procedure might have important drawbacks including a high surgical mortality caused by bleeding, sudden cardiac arrest, or pulmonary thrombus. Pressure overload may also be induced using half-closed paper clips easily applied with a stopper. We could adjust the closing size by moving the stopper. The objective of this study was to establish rats model of RV failure using the pulmonary artery half-closed clip procedure comparing with partial ligation procedure. 2. Materials and Methods {#sec2} ======================== 2.1. Animal Care {#sec2.1} ---------------- All experimental procedures and protocols used in this investigation were reviewed and approved by the institutional animal care and use committee and were in accordance with the National Institutes of Health "guide for the care and use of laboratory animals" (National Institutes of Health publication number 85-23, revised 1996). 2.2. Animal Model {#sec2.2} ----------------- A rat model of the pulmonary artery banding (PAB) was established to create chronic RV pressure overload. Eight-week-old male Sprague-Dawley rats (240--260 g) were anesthetized with intraperitoneal pentobarbital (50 mg/kg body weight) and xylazine (5 mg/kg) and ventilated with 100% O~2~ by using a volume controlled respirator (2 mL, 60 cycles/min). A left thoracotomy was performed at the fourth intercostal space, and the main pulmonary artery (mPA) was carefully exposed ([Figure 1](#fig1){ref-type="fig"}). ### 2.2.1. Partial Pulmonary Artery Ligation Model (PAL) (*n* = 28) {#sec2.2.1} A 7-0 prolene suture was tied tightly around an 18-gauge needle alongside the mPA. After subsequent rapid removal of the needle, a fixed constricted opening was created in the lumen equal to the diameter of the needle \[[@B5], [@B8], [@B9]\]. ### 2.2.2. Half-Closed Clip Model (PAC) (*n* = 28) {#sec2.2.2} A small clip (LT100 ETHICON) was half-closed around the mPA using a clip applier (LX107 ETHICON) with a stopper ([Figure 1](#fig1){ref-type="fig"}). The blood flow through the mPA was restricted to equal the inner segment of the half-closed clip. We selected type (B) clip because this inner size was almost equal to outer size of a 18 G needle ([Figure 2](#fig2){ref-type="fig"}). Thereafter, the thorax was closed in layers, and the ventilator setting was changed (90 cycles/min) for half an hour to reduce the respiratory load. ### 2.2.3. Sham Operation (*n* = 12) {#sec2.2.3} A left thoracotomy was performed at the fourth intercostal space, and the mPA was carefully exposed. The thorax was closed in layers, and the ventilator setting was changed (90 cycles min^−1^) for half an hour to reduce the respiratory load. 2.3. Echocardiographic Measurements {#sec2.3} ----------------------------------- Echocardiograms were recorded on the preoperative day and on the 4th week and 8th week after the procedure. The rats were anaesthetized for echos (pentobarbital 50 mg kg^−1^ body weight) but were breathing spontaneously and were positioned on their left side. Measuring echocardiograms, the dose of anesthetics was reduced to 50% only on the day when a rat had developed clinical signs, loss of activity, body edema, and pleural effusion, of RV failure. Transthoracic 2-dimensional, M-mode (according to the standards of the American Society of Echocardiography) and Doppler imaging were performed with a 6.5 MHz transducer (Xario TOSHIBA, JAPAN). [Figure 3](#fig3){ref-type="fig"} showed schemata of the measurements of echocardiograms, respectively. The RV morphology was assessed as RV free wall thickness (RVWT) and RV end-diastolic diameter (RVDd) and RV outflow tract dimension (RVOTD). RVWT was measured by M-mode either in the 2-dimensional parasternal short-axis view below the tricuspid valve or in the parasternal long-axis view, depending on the quality of RV free wall visualization. RVDd was measured as the maximal distance from the RV free wall to the septum from the apical four-chamber view. RVOTD and aortic dimension (AoD) were measured at the aortic valve level in the short-axis view. The ratio of RVOTD to AoD (RVOTD/AoD) was calculated as representative of RV dilatation. The position of the transducer was aligned to visualize the RV apex. The maximal RV cavity size preceding the frame with the onset of systolic closure of tricuspid valve leaflets was used to measure RVDd. To assess RV function, the tricuspid annular plane systolic excursion (TAPSE) of the lateral portion of the tricuspid annular plane was measured by the base-to-apex shortening during systole. This was recorded in the M-mode format under two-dimensional echocardiographic guidance from the apical four-chamber view. Furthermore, PA velocity (PAv) was recorded at the aortic valve level in the short-axis view. This was used as an echocardiographic indicator of RV pressure overload \[[@B10], [@B11]\]. The left ventricle (LV) M-mode echocardiograms were recorded to evaluate LV function, as previously described. LV end-diastolic dimension (LVDd), LV end-systolic dimension (LVDs), interventricular septal wall thickness (IVST), and LV posterior wall thickness (LVPW) were measured at the papillary muscles level, and LV fractional shortening (LVFS) was calculated as follows: $$\begin{matrix} {LVFS = \left\lbrack {\frac{\left\lbrack {LVDd - LVDs} \right\rbrack}{LVDd}} \right\rbrack \times 100\left( {\%} \right).} \\ \end{matrix}$$ 2.4. Histopathological Analysis {#sec2.4} ------------------------------- In the PAL group, the remaining 4 rats were killed 4 weeks after surgery and all the remaining rats (*n* = 8) were killed 8 weeks after surgery. In the PAC group, 4 rats were killed 4 weeks after surgery and all the remaining rats (*n* = 6) were killed 8 weeks after surgery. In the sham group, 6 rats were killed 4 or 8 weeks after surgery. The hearts were quickly removed, and the ventricles were dissected free of atrial tissue and large blood vessels. The right ventricle was carefully separated from the left ventricle and interventricular septum (IVS). The fresh ventricular tissues were immediately blotted dry and weighted separately to determine the degree of RV hypertrophy based on 2 parameters: RV wall weight/body weight (RV/BW) and RV wall weight/LV and IVS wall weight (RV/LV + IVS). Tissue specimens for pathological analysis were obtained from whole hearts in cross sections, cut into 5 mm thick sections, and stained with hematoxylin and eosin for morphologic analysis, including measurement of the short-axis dimension of the RV myocardial cell and Masson\'s trichrome staining to determine the amount of interstitial and myocardial fibrosis. Digital images of cross sections were taken by a CCD camera (OLYMPUS DP-72) with a light microscope (OLYMPUS SZX12). The number of pixels of the blue-stained collagen area was calculated with Adobe Photoshop CS5 soft and then divided by the total number of pixels in the RV wall per slide, analyzed, and averaged. 2.5. Statistical Analysis {#sec2.5} ------------------------- All data were expressed as mean ± SEM and range. Student\'s unpaired *t*-test or analysis of variance for parametric values was used to compare group means. Probability of 0.05 or less was considered to be statistically significant. 3. Results {#sec3} ========== 3.1. Procedure {#sec3.1} -------------- Surgical procedure time was defined as the interval from starting to incise the skin to skin closure. The surgical procedure time was 27.6 ± 1.4 min for PAL and 17.3 ± 2.1 min for PAC (*p* \< 0.01). For the PAL procedure, 5 major hemorrhages occurred during trimming the mPA compared to only one major hemorrhage for the PAC procedure. To evaluate the two procedures for RV pressure overload damage, we recorded the postsurgery recovery time. This was defined as the interval from time of extubation until the animal was fully conscious and walking freely. The recovery time was 30.4 ± 5.3 min for the sham procedure and 33.5 ± 5.2 min for PAC which was significantly shorter than the PAL procedure which was 64.9 ± 7.3 min. 3.2. Surgical Risk {#sec3.2} ------------------ Sixty-eight rats underwent PAL (*n* = 28) or PAC (*n* = 28) or sham (*n* = 12) in the surgical risk study. During the perisurgical period, which was the time from the beginning of surgery to 6 hours after surgery, 1 rat in the PAC group out of 28 died due to bleeding (3.6%). In contrast, 7 of 28 perisurgical deaths were seen in the PAL group (25%). Of those 7 rats, 1 died from an intubation accident, 2 died from bleeding, 2 died from pneumothorax, and 2 died from sudden cardiac arrest. All of the animals in the sham group survived regardless of the surgical procedure. After the perisurgical period, there were 3 and 12 deaths in the PAL and PAC groups, respectively. The most common reasons for post-PAB death in both groups were pleural and/or pericardial effusion (pleural effusion: 66.7% in PAL and 83.3% in PAC, pericardial effusion: 33.3% in PAL and 41.7% in PAC) occurring between weeks 4 and 8 after surgery. Some of the rats died of unknown causes although we assumed these were due to either cardiac arrhythmia or heart failure as autopsy of these rats showed no blood in the thoracic cavity. 3.3. Echocardiographic Study {#sec3.3} ---------------------------- [Table 1](#tab1){ref-type="table"} shows the trend for LVDd and AoD to decrease but LV contractility (LVFS) showed no significant difference. Both PAL and PAC groups showed a significantly increased pulsed Doppler Peak PAv, RVOTD, RVDd, and RVWT. In the PAC group, PAv and RVOTD tended to increase more significantly than in the PAL group. The dilatation of the RV associated with the RVDd/LVDd ratio and the RVOTD/AoD ratio. In the PAC group, the RVDd/LVDd ratio was increased and became significantly greater than in the sham and PAL group. The RVOTD/AoD ratio was about 1.0 throughout the observation in the sham group. In contrast, the ratio increased progressively in the PAL and PAC groups, especially in the PAC group. The signs of RV failure associated with TAPSE were more severe in the PAC group than in the PAL group. The signs of moderate or severe tricuspid regurgitation (TR) or pleural effusion or IVC dilation were detected in the PAC group 5 weeks after the PAC procedure, but we could seldom detect these signs in the PAL group throughout the observation period (PAC versus PAL: moderate or more TR 21/27 (77.8%) versus 3/21 (14.3%), pleural effusion 18/27 (66.7%) versus 2/21 (10.0%), and IVC dilation 20/27 (74.1%) versus 0/21 (0%)). Furthermore, the standard deviation in the PAC group was greater than the PAL group. In other words, the data of the PAL group varied more widely than that of the PAC group. More stable data might be obtained with the PAC procedure. 3.4. Morphometric and Histological Analysis {#sec3.4} ------------------------------------------- ### 3.4.1. RV Hypertrophy {#sec3.4.1} A weight analysis showed the heart weight/BW, RV/BW, and RV/(LV + IVS) weight ratios in the PAC and PAL groups to be similar and significantly higher than in the sham group ([Table 2](#tab2){ref-type="table"}). Whole heart findings revealed a thickened RV wall, with an enlarged cavity, and the IVS shifted toward the left side in the PAC and PAL groups ([Figure 5](#fig5){ref-type="fig"}). Myocardial cell size in the PAC and PAL groups was similar and significantly higher than in the sham group (4th week: sham versus PAL versus PAC: 13.7 ± 1.64 *μ*m versus 21.3 ± 3.47 *μ*m versus 22.8 ± 2.38 *μ*m, 8th week: sham versus PAL versus PAC: 14.3 ± 1.40 *μ*m versus 34.7 ± 7.22 *μ*m versus 35.1 ± 8.19 *μ*m) ([Table 1](#tab1){ref-type="table"}, [Figure 4](#fig4){ref-type="fig"}). ### 3.4.2. Fibrosis {#sec3.4.2} Masson\'s trichrome staining showed fibrosis in the RV free walls of the sham, PAL, and PAC groups ([Figure 6](#fig6){ref-type="fig"}). At the 4th week, the percentage of fibrosis in the PAL and PAC groups was similar and significantly higher than in the sham group (sham versus PAL versus PAC: 0.49 ± 0.06% versus 3.12 ± 1.09% versus 4.68 ± 1.63%) ([Figure 6(b)](#fig6){ref-type="fig"}). But at 8 weeks, the percentage fibrosis in the PAC group was significantly higher than in the sham and PAL groups (sham versus PAL versus PAC: 1.51 ± 0.86% versus 9.73 ± 6.05% versus 29.2 ± 6.13%) ([Figure 6(b)](#fig6){ref-type="fig"}). 4. Discussion {#sec4} ============= Definition of heart failure was challenging because there were neither objective cutoff values of cardiac or ventricular dysfunction nor changes in pressure, dimension, or volume that could be reliably used to identify patients with heart failure. According to The European Society of Cardiology task force for the diagnosis and treatment of heart failure, both symptoms at rest or during exertion and objective evidence of cardiac dysfunction at rest should be present to diagnose heart failure \[[@B12]\]. In animal models, however, heart failure symptoms like fatigue or breathlessness are difficult to detect and quantify. Then evaluation of the models depends on objective findings such as reduced cardiac output, increased filling pressure, and progressive fibrosis \[[@B6], [@B7], [@B13]\]. In addition, the valuable signs of progressive heart failure are pericardial or pleural effusion \[[@B14]\]. In this paper, we demonstrated right heart failure by the objective findings of cardiac dysfunction. Most attention is given to LV function, whereas RV function and disease have seldom been focused on. It is an established fact that there is a relationship between LV and RV function. Impairment of the RV might influence LV function \[[@B15], [@B16]\]. RV function is one independent predictor of mortality and the development of heart failure in patients with LV dysfunction \[[@B17]\]. Five to ten percent of patients with advanced chronic obstructive pulmonary disease may suffer from severe pulmonary hypertension and present with a progressively downhill clinical course because of RV dysfunction \[[@B18]\]. Not only in the view point of congenital heart disease, well-established animal models of RV function may be necessary for further investigation of RV function and pathology. Some animal models of RV dysfunction exist (monocrotaline treatment, partial ligation method) \[[@B5], [@B8]--[@B10], [@B19]\]. Monocrotaline treatment has been used to induce pulmonary hypertension resulting in RV hypertrophy and eventually heart failure \[[@B20], [@B21]\]. Monocrotaline treatment might have disadvantages in the form of disease manifestations not usually associated with human heart failure and changes in hormones such as endothelin \[[@B21], [@B22]\]. PAB does not have such side effects and is a promising procedure for inducing symptomatic RV dysfunction. PAL procedure has been used in several hypertrophy experiments; however, slippage of the band and the obvious difficulties ensuring the same degree of constriction among the banded animals when using a surgical nylon (usually prolene) led us to believe that banding with tantalum clips would be a more reliable procedure to create PAB model. In another PAB study using surgical nylon thread, PAL procedure, about 25% of the banded rats showed no hypertrophy \[[@B23]\], suggesting that this was a less reliable way. Also in this study, in the PAL group, the data had greater variance than in the PAC group. In addition, the objective findings of RV dysfunction were present less in the PAL group. To the best of our knowledge this is the first reported evaluation of a newly devised means of stressing the right ventricle---the half-closed clip procedure---and its comparison with the pulmonary artery banding procedure. Using this new technique we could produce a right heart failure model in significantly less time and with much faster recovery and fewer surgical deaths than the PAL procedure. Most importantly, this novel procedure more fully developed subsequent myocardial damage and postsurgery cardiac dysfunction than did the PAL procedure. Furthermore, if we adjust the position of the stopper, we can produce different PAB models using this procedure. A recent study in rats who underwent banding of the mPA by the existing partial ligation procedure showed that pressure overload alone was insufficient to explain right heart failure; the rats showed no effect on cardiac output or fibrosis \[[@B24]\]. Using the PAC procedure, however, we could produce RV dysfunction models that represent not only RV cavity dilation and RV hypertrophy but also moderate or greater tricuspid regurgitation or pleural and/or pericardial effusion. Hardziyenka and colleagues reported LV atrophy in pulmonary hypertension \[[@B25]\]. However, we could not detect significant left-sided effects, but we could find a trend towards LV atrophy ([Table 1](#tab1){ref-type="table"} LVPWT, PAC versus sham, *p* = 0.051). 5. Conclusion {#sec5} ============= This study demonstrated that the application of a tantalum half-closed clip around the pulmonary artery induced right ventricle dysfunction in a reproducible manner. This model should prove valuable in the investigation and treatment of right heart failure caused by pressure overload. Conflict of Interests ===================== The authors declare that there is no conflict of interests regarding the publication of this paper. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} ###### Ultrasound findings in the sham rats, at 4th or 8th week after the PAB procedure.   Preoperative day 4th week 8th week ---------------- ------------------ -------------- -------------- -------------- --------------------- ------------------------- --------------- ---------------------- ------------------------- *n* 12 28 28 12 16 22 6 8 6 LVDd (mm) 4.66 ± 0.43 4.75 ± 0.43 4.67 ± 0.39 5.36 ± 0.36 5.10 ± 0.42 5.07 ± 0.51 5.78 ± 0.19 5.40 ± 0.29 5.42 ± 0.38 LVDs (mm) 2.10 ± 0.5 2.15 ± 0.47 2.06 ± 0.47 2.51 ± 0.38 2.88 ± 0.58 2.55 ± 0.31 2.90 ± 0.52 2.38 ± 0.38 2.68 ± 0.45 IVST (mm) 1.06 ± 0.27 1.11 ± 0.29 1.14 ± 0.31 1.57 ± 0.10 1.56 ± 0.16 1.65 ± 0.27 1.93 ± 0.21 1.90 ± 0.10 2.13 ± 0.22 PWT (mm) 1.01 ± 0.24 1.05 ± 0.27 1.04 ± 0.27 1.71 ± 0.29 1.51 ± 0.08 1.62 ± 0.20 2.23 ± 0.22 2.04 ± 0.09 1.97 ± 0.14 FS (%) 55.17 ± 8.74 55.02 ± 8.17 56.15 ± 8.68 53.27 ± 4.42 42.81 ± 15.09 49.78 ± 4.02 49.62 ± 10.14 56.07 ± 5.10 50.64 ± 5.51 RVDd (mm) 2.46 ± 0.31 2.49 ± 0.29 2.49 ± 0.26 2.43 ± 0.28 **5.48 ± 1.47^∗^** **6.26 ± 1.28^∗^** 2.80 ± 0.58 **5.26 ± 0.96^∗^** **6.95 ± 0.63^∗^** RVWT (mm) 0.63 ± 0.19 0.65 ± 0.18 0.70 ± 0.17 0.54 ± 0.08 **1.25 ± 0.11^∗^** **1.41 ± 0.28^∗^** 0.98 ± 0.1 **1.72 ± 0.15^∗^** **1.80 ± 0.43^∗^** TAPSE (mm) 2.71 ± 0.22 2.69 ± 0.21 2.69 ± 0.23 2.53 ± 0.14 2.40 ± 0.11 **1.67 ± 0.20^∗^** ^\$^ 2.58 ± 0.1 2.32 ± 0.20 **1.42 ± 0.25^∗^** ^\$^ HR (bpm) 438 ± 43 441 ± 39 455 ± 24 447 ± 21 436 ± 26 422 ± 31 451 ± 22 419 ± 20 411 ± 30 Peak PAV (m/s) 0.76 ± 0.11 0.73 ± 0.12 0.71 ± 0.12 0.97 ± 0.14 **2.93 ± 0.54^∗^** **3.91 ± 0.38^∗^** ^\$^ 1.05 ± 0.13 **3.28 ± 0.37^∗^** **4.03 ± 0.27^∗^** ^\$^ RVOTD (mm) 1.99 ± 0.20 2.00 ± 0.24 1.97 ± 0.25 2.64 ± 0.17 **2.98 ± 0.37^\#^** **3.22 ± 0.50^∗^** 2.20 ± 0.16 **2.78 ± 0.37** ^\#^ **3.62 ± 0.16^∗^** ^\$^ AoD (mm) 2.23 ± 0.25 2.25 ± 0.23 2.30 ± 0.21 2.80 ± 0.18 2.65 ± 0.11 **2.55 ± 0.22^\#^** 2.78 ± 0.05 2.44 ± 0.35 2.65 ± 0.10 RVDd/LVDd 0.53 ± 0.08 0.53 ± 0.07 0.53 ± 0.05 0.46 ± 0.07 **1.08 ± 0.28^∗^** **1.25 ± 0.32^∗^** 0.48 ± 0.09 **0.971 ± 0.15^∗^** **1.28 ± 0.08^∗^** ^\$^ RVOTD/AoD 0.90 ± 0.16 0.90 ± 0.16 0.86 ± 0.15 0.95 ± 0.07 **1.12 ± 0.14^\#^** **1.27 ± 0.20^∗^** 0.79 ± 0.07 **1.157 ± 0.21^\#^** **1.37 ± 0.10^∗†^** RVPW/LVPW 0.65 ± 0.23 0.65 ± 0.21 0.69 ± 0.20 0.32 ± 0.05 **0.83 ± 0.08^∗^** **0.87 ± 0.15^∗^** 0.44 ± 0.06 **0.844 ± 0.07^∗^** **0.92 ± 0.24^∗^** *n*, number of rats. \#: *p* \< 0.05 versus sham, ∗: *p* \< 0.01 versus sham, †: *p* \< 0.05 versus PAL, and \$: *p* \< 0.01 versus PAL. ###### Weight analysis at the 4th and 8th week after the PAB procedure. ###### \(a\) Weight analysis at the 4th week after the procedure   BW HW/BW RV/BW RV/(LV + IVS) ------ ---------- -------------------- -------------------- -------------------- Sham 435 ± 39 3.57 ± 0.16 0.56 ± 0.07 0.21 ± 0.02 PAL 403 ± 21 **4.65 ± 0.17^∗^** **1.40 ± 0.13^∗^** **0.50 ± 0.04^∗^** PAC 398 ± 33 **4.88 ± 0.17^∗^** **1.39 ± 0.03^∗^** **0.49 ± 0.04^∗^** ###### \(b\) Weight analysis at the 8th week after the procedure   BW HW/BW RV/BW RV/(LV + IVS) ------ ---------- -------------------- -------------------- -------------------- Sham 508 ± 41 3.30 ± 0.23 0.48 ± 0.04 0.21 ± 0.01 PAL 505 ± 47 **4.34 ± 0.30^∗^** **1.27 ± 0.12^∗^** **0.54 ± 0.02^∗^** PAC 450 ± 50 **4.81 ± 0.32^∗^** **1.40 ± 0.07^∗^** **0.58 ± 0.03^∗^** There was no significant difference between the PAL and PAC groups. BW: body weight at the sacrifice time, HW: heart weight, and ∗: *p* \< 0.05 versus sham. [^1]: Academic Editor: Bodh I. Jugdutt

1. Introduction {#sec1-nutrients-11-02075} =============== Obesogenic diets are associated with a host of chronic diseases. Calorie-dense diets, including those that are high in fat and sucrose, have been shown to blunt insulin sensitivity \[[@B1-nutrients-11-02075],[@B2-nutrients-11-02075]\], increase the risk for cardiovascular disease \[[@B3-nutrients-11-02075],[@B4-nutrients-11-02075]\], and increase the incidence of certain types of cancer \[[@B5-nutrients-11-02075],[@B6-nutrients-11-02075]\]. While consumption of a high-fat (HF) diet is known to produce undesirable outcomes, it has been suggested that maternal diet may play an important role in preventing the negative consequences of poor postnatal diet. For instance, compared to control-fed mice, mice fed an obesogenic diet after weaning had larger adipocytes, higher fasting glucose and insulin levels, and reduced expression of insulin signaling proteins \[[@B7-nutrients-11-02075],[@B8-nutrients-11-02075]\]. However, a maternal low-fat (LF) diet significantly reduced adipocyte size, lowered fasting glucose and insulin, and ameliorated the protein expression changes. Similarly, an obesogenic postnatal diet only resulted in cardiac hypertrophy and elevated fibrosis if it was preceded by an obesogenic prenatal diet, but not if it was preceded by a prenatal LF diet \[[@B9-nutrients-11-02075]\]. It has been hypothesized that perinatal nutrition acts via epigenetic mechanisms to mediate long-term health outcomes. Early life is marked by a highly dynamic epigenetic state. In particular, DNA methylation that is established during gestation is thought to persist into adulthood. Previously, maternal intake of micronutrients such as folate and choline, as well as protein restriction has been associated with altered DNA methylation in adult offspring \[[@B10-nutrients-11-02075],[@B11-nutrients-11-02075],[@B12-nutrients-11-02075]\]. Macronutrient consumption, particularly a HF diet, has been shown to produce methylation differences in energy homeostasis genes, peroxisome proliferator-activated receptor α (*Ppara*) \[[@B13-nutrients-11-02075]\], inflammatory genes, toll-like receptors 1 and 2 (*Tlr1* and *Tlr2*) \[[@B14-nutrients-11-02075]\], and the hepatic cell cycle inhibitor, cyclin-dependent kinase inhibitor (*Cdkn1a*) \[[@B15-nutrients-11-02075]\]. In this experiment, we examined the role of maternal diet on epigenetic programming. Pregnant Sprague-Dawley rats were fed either a HF or a LF diet during gestation and lactation. After weaning, male pups from both groups were fed a HF diet until 12 weeks of age. DNA methylation was measured in hepatic tissue using complementary methylated DNA immunoprecipitation sequencing (MeDIP-seq) and methylation-sensitive restriction enzyme sequencing (MRE-seq). Differentially methylated regions (DMRs) were characterized based on association with CpG islands and genes. Pathway analysis was performed, and gene expression was measured. Finally, we attempted to functionally interpret intergenic DMRs by examining the chromatin structure around each conserved locus. 2. Materials and Methods {#sec2-nutrients-11-02075} ======================== 2.1. Animals and Diets {#sec2dot1-nutrients-11-02075} ---------------------- Timed-pregnant Sprague-Dawley rat dams (Charles River Laboratories, Wilmington, MA) were randomized into two groups for dietary treatment during gestation and lactation. The first group of 12 rats received a high-fat diet (HF; Research Diets, Inc.; 45% calories from fat), and the second group of 12 rats received a low-fat diet (LF; AIN93G Research Diets, Inc.; 16% calories from fat). Dams were individually housed with their pups in standard polycarbonate cages in a humidity- and temperature-controlled room on a 12-h light-dark cycle with ad libitum access to food and drinking water \[[@B16-nutrients-11-02075]\]. On postnatal day 21, male offspring (*n* = 10 rats per group from 10 different dams) were all given ad libitum access to only a HF diet until postnatal week 12. Animals were then sacrificed, and the median lobe of the liver was frozen in liquid nitrogen and stored at −70 °C. It has been shown that lobes differ in their capacity to store minerals \[[@B17-nutrients-11-02075]\], susceptibility to certain diseases \[[@B18-nutrients-11-02075]\], and transcriptomic profiles \[[@B19-nutrients-11-02075]\]. In rodents, the left lobe is developmentally distinct from the right, median, and caudate lobes. By selecting the median lobe, not only do we reduce variation between tissue samples, but we also choose a representative region that is developmentally similar to the majority of the liver. Institutional and governmental regulations regarding the ethical use of animals were followed during the study. The protocol for the ethical use of animals was approved by the Institutional Animal Care and Use Committee (IACUC protocol no. 09112). 2.2. Methylated DNA Immunoprecipitation (MeDIP) and Methylation-Sensitive Restriction Enzyme (MRE) Sequencing {#sec2dot2-nutrients-11-02075} ------------------------------------------------------------------------------------------------------------- Genomic DNA was isolated using previously published methods \[[@B20-nutrients-11-02075]\]. Animals were chosen through an extensive screening process in which gene expression and histology were measured, and the best representatives from each group were used for sequencing. Complementary MeDIP-seq and MRE-seq were then performed using previously published protocols \[[@B20-nutrients-11-02075]\]. Briefly, MeDIP utilizes antibodies against 5-methylcytidine to quantify methylated DNA sequences, while MRE-seq uses restriction enzymes that cut at unmethylated CpG sites. MeDIP-seq provides better coverage and MRE-seq offers superior resolution, so that when combined the methylome can be quantified with high accuracy \[[@B21-nutrients-11-02075],[@B22-nutrients-11-02075]\]. Antibodies, restriction enzymes, DNA fragmentation, and library preparation procedures have been detailed by Li et al. \[[@B23-nutrients-11-02075]\]. 2.3. DMR Identification {#sec2dot3-nutrients-11-02075} ----------------------- MeDIP-seq and MRE-seq data analysis were performed using the methylMnM package in R. A detailed procedure is presented by Zhang et al. \[[@B24-nutrients-11-02075]\]. In brief, the rat genome (Rn4) was partitioned into 500 bp bins, and MeDIP-seq and MRE-seq data were modeled as a function of CpG content, MRE site content, and methylation level within each bin. We used the methylMnM algorithm to test the null hypothesis that methylation level was the same between the two samples. The normalized MeDIP and MRE reads were treated as mutually independent Poisson random variables and their expected values were calculated for each sample within each bin. A test statistic and *p*-value were calculated assuming that the joint distribution of the random variables followed a multinomial distribution. Bins with a Benjamini--Hochberg false discovery rate (FDR) *p*-value \<0.05 were considered significant and were called differentially methylated regions (DMRs). Further information regarding methylMnM can be found on the Bioconductor website: <http://www.bioconductor.org/packages/release/bioc/html/methylMnM.html>) \[[@B23-nutrients-11-02075],[@B24-nutrients-11-02075]\]. 2.4. Annotation and Pathway Analysis {#sec2dot4-nutrients-11-02075} ------------------------------------ Next, we examined the association between DMRs and CpG islands. CpG islands were defined based on three criteria: (i) sequence length greater than 200 bp, (ii) GC content greater than 50%, and (iii) an observed-to-expected CpG ratio greater than 0.6. Shores were the 2,000 bp regions upstream and downstream of each island \[[@B25-nutrients-11-02075]\]. We also annotated DMRs based on location relative to genes. DMRs were classified as either intergenic, intragenic, downstream, or in the promoter. Promoter regions were defined as the 1500 bp upstream of the transcription start site (TSS), while downstream regions were defined as the 1500 bp downstream of the transcription end site (TES) \[[@B16-nutrients-11-02075]\]. Intragenic regions included both exonic and intronic sequences of the gene body. Intergenic DMRs fell outside any gene body or 1500 bp flanking region. Differentially methylated genes (DMGs) were annotated with Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) Pathway terms. DAVID Bioinformatics Resources version 6.7 was used to identify enriched annotation clusters and pathways with a high degree of differential methylation (<http://david.abcc.ncifcrf.gov/>) \[[@B26-nutrients-11-02075]\]. Functional clusters were required to have at least 2 GO and/or KEGG Pathway terms, and the majority of terms within each cluster were required to have a Benjamini--Hochberg FDR *p*-value \< 0.05. Similarly, we report KEGG pathways that contained at least 2 DMGs and had a fold enrichment ≥ 1.5 (based on the proportion of a specific pathway's genes that were DMGs). 2.5. Methylation Specific PCR {#sec2dot5-nutrients-11-02075} ----------------------------- To validate sequencing results, methylation specific PCR (MSP) was used to quantify DNA methylation in all animals. Primer design, genomic DNA isolation, bisulfite conversion, and qPCR were performed using published methods \[[@B20-nutrients-11-02075]\]. The relative amount of methylated DNA was calculated as a ratio using the following equation: % methylated DNA = (quantity of methylated DNA)/(quantity of methylated DNA + quantity of unmethylated DNA) × 100%. All MSP primer information can be found in [Table 1](#nutrients-11-02075-t001){ref-type="table"}. 2.6. Gene Expression {#sec2dot6-nutrients-11-02075} -------------------- Total RNA was extracted using previously published methods \[[@B20-nutrients-11-02075]\]. Briefly, frozen liver tissue was ground in liquid nitrogen and the Direct-zol™ RNA MiniPrep kit (Zymo Research) was used for RNA isolation. RT-PCR was performed using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems) and incubated in a 2720 Thermal Cycler (Applied Biosystems). A serially diluted standard curve was created, and qPCR was carried out using Power SYBR^®^ Green Master Mix (Life Technologies) run in a StepOnePlus™ Real-Time PCR System. All primers were designed using Vector NTI (Life Technologies) and manufactured by Integrated DNA Technologies. Information regarding primers for gene expression is detailed in [Table 2](#nutrients-11-02075-t002){ref-type="table"}. 2.7. Analysis of Intergenic DMRs {#sec2dot7-nutrients-11-02075} -------------------------------- In previous analyses, DMRs that are not located within close proximity of genes are often ignored. We attempted to uncover functions for these DMRs by using previously published high throughput chromatin contact data (Hi-C) \[[@B27-nutrients-11-02075]\]. Such chromatin contact maps are not publicly available for rat tissue, thus we examined only DMRs that were conserved between species using the phastCons9way track from Genome Browser. Conservation scores range from 0 to 1, where 1 denotes perfect sequence alignment and high conservation. For our analysis, we only considered DMRs that had a mean alignment score \>0.5 over the 500 bp bin. Next, the analogous region from the Rn4 genome build was identified in the Hg38 genome. We used the 3D Genome Browser to visualize Hi-C data from human liver tissue \[[@B27-nutrients-11-02075],[@B28-nutrients-11-02075]\]. In particular, we examined each of the 500 bp DMRs and the 1 mb region flanking either side of the region. We then located the topologically associated domain (TAD) which contained the DMR. Genes within the TAD were considered to have contact with the nearby DMR. Those genes were selected for gene ontology and pathway analyses. 2.8. Statistical Analysis {#sec2dot8-nutrients-11-02075} ------------------------- For body weight, food intake, MSP, and gene expression, all pairwise comparisons between groups were made using two-tailed t-tests. For pathway analysis, significant enrichment was determined using the Benjamini--Hochberg FDR *p*-value. All statistical analysis was performed in R (version 3.1.2). 3. Results {#sec3-nutrients-11-02075} ========== 3.1. Offspring Phenotype {#sec3dot1-nutrients-11-02075} ------------------------ Male Sprague-Dawley rats were divided into two groups and exposed to either a LF or a HF diet for seven weeks during gestation and lactation (*n* = 10/group; [Figure 1](#nutrients-11-02075-f001){ref-type="fig"}A). After weaning (postnatal week three), all animals were given a HF dietary challenge that mimicked an obesogenic western diet. Animals were fed the HF diet for nine weeks and sacrificed at 12 weeks of age. Across the nine weeks of post-weaning feeding, there was no difference in food intake between the groups ([Figure 1](#nutrients-11-02075-f001){ref-type="fig"}B). Additionally, body weights were consistent between groups, suggesting that maternal diet was insufficient to compensate for HF-induced postnatal weight gain ([Figure 1](#nutrients-11-02075-f001){ref-type="fig"}C,D). 3.2. DNA Methylation {#sec3dot2-nutrients-11-02075} -------------------- Despite the lack of phenotypic differences, we hypothesized that epigenetic differences might still exist. Methylated DNA was measured using MeDIP-seq and unmethylated DNA was measured with MRE-seq. Combining the two methods, 1,419 differentially methylated regions (DMRs) were identified between groups ([Figure 2](#nutrients-11-02075-f002){ref-type="fig"}A). Of these, 534 (37.6%) were more highly methylated in the HF group, while 885 (62.4%) were more highly methylated in the LF group ([Figure 2](#nutrients-11-02075-f002){ref-type="fig"}B). Next, analysis of the genomic location of the DMRs revealed that 827 (58.3%) were located in intergenic regions (not within 1,500 bp of a gene), 524 (36.9%) were within the gene body (including intronic and exonic sequences), 48 (3.4%) were in a promoter (within 1,500 bp upstream of the TSS), and 45 (3.2%) were downstream of a gene (within 1500 bp downstream of the TES) ([Figure 2](#nutrients-11-02075-f002){ref-type="fig"}C). We next examined DMR location relative to CpG islands and saw that more than twice as many DMRs were located within CpG shores (*n* = 435, 30.7%) than in islands themselves (*n* = 177, 12.5%, [Figure 2](#nutrients-11-02075-f002){ref-type="fig"}D). DNA methylation was validated using MSP. First, we tested one DMR located within the *Myh7b* gene that was identified to be more highly methylated in the LF group by MeDIP-seq and MRE-seq ([Figure 3](#nutrients-11-02075-f003){ref-type="fig"}A--D). MSP was performed on six CpG sites within the DMR. Although one site was not differentially methylated, two of the sites were higher but not significantly hypermethylated in the LF group (*p* \< 0.1) and three were significantly more methylated in the LF group (*p* \< 0.05, [Figure 3](#nutrients-11-02075-f003){ref-type="fig"}E). Furthermore, the average DNA methylation across all six sites was significantly higher in the LF group (*p* = 0.0089, [Figure 3](#nutrients-11-02075-f003){ref-type="fig"}F). Negative control was also performed on four CpG sites within the *Gpam* gene ([Figure 3](#nutrients-11-02075-f003){ref-type="fig"}G). Four CpGs that were not computationally identified by MeDIP and MRE analysis were examined with MSP ([Figure 3](#nutrients-11-02075-f003){ref-type="fig"}H,I). Neither the individual sites nor the average over the region showed differential methylation ([Figure 3](#nutrients-11-02075-f003){ref-type="fig"}J,K). Findings here and in our previous work demonstrate the validity of combined MeDIP-seq and MRE-seq analysis \[[@B16-nutrients-11-02075],[@B20-nutrients-11-02075]\]. In order to understand the functional relevance of the gene-associated DMRs, gene ontology (GO) and KEGG pathway analyses were performed. We found that DMGs tended to cluster in processes involved in ion binding, cell morphogenesis, ion channels, and neuronal development ([Table 3](#nutrients-11-02075-t003){ref-type="table"}). Additionally, we found that the biosynthesis of unsaturated fatty acids pathway was most highly enriched for differential methylation, while pathways in cancer contained the greatest number of differentially methylated genes ([Figure 4](#nutrients-11-02075-f004){ref-type="fig"}). Interestingly, we found that other metabolic and cancer pathways were also enriched, including the insulin signaling, colorectal cancer, and mitogen-activated protein kinase (MAPK) signaling pathways. 3.3. Gene Expression {#sec3dot3-nutrients-11-02075} -------------------- Given the enrichment of differential methylation in cancer and metabolic genes, qPCR was utilized to measure gene expression of 31 differentially methylated genes in those pathways. Details regarding the genomic location and methylation levels of the gene-associated DMRs can be found in [Table 4](#nutrients-11-02075-t004){ref-type="table"}. We found two differentially expressed genes, including *Map3k5* and *Igf1r* (*p* = 0.030 and *p* = 0.026, respectively; [Figure 5](#nutrients-11-02075-f005){ref-type="fig"}). Both *Igf1r* and *Map3k5* were related to cancer and metabolism. Both *Map3k5* and *Igf1r* were more highly expressed in the LF group compared to the HF group. 3.4. Intergenic CpGs {#sec3dot4-nutrients-11-02075} -------------------- Finally, we attempted to understand the impact of DNA methylation in regions that were not associated with particular genes. The majority of the identified DMRs were located in intergenic regions. Previously, such CpGs have been ignored. We hypothesized that even DMRs without obvious gene contact would impact expression via altering chromatic confirmation. Three-dimensional DNA dynamics have been studied in the human genome using chromatic capture techniques; however, these methods have not been performed in rat samples. Thus, we were only able to address the 42 CpG sites that had a conservation score \>0.5 ([Figure 6](#nutrients-11-02075-f006){ref-type="fig"}A). From the Rn4 genome build, the analogous region was identified in the Hg38 genome. The DMR and the 1 mb region flanking either side were queried for topologically associated domains (TADs) using Hi-C data from liver tissue ([Figure 6](#nutrients-11-02075-f006){ref-type="fig"}B) \[[@B27-nutrients-11-02075]\]. KEGG Pathway and Gene Ontology analyses were performed on all genes that fell within a TAD containing a DMR ([Figure 6](#nutrients-11-02075-f006){ref-type="fig"}C,D). Interestingly, these genes also had functions in metabolism, replicating the findings in the gene-associated DMRs. We then measured the expression of ten metabolic genes associated with intergenic DMRs ([Figure 7](#nutrients-11-02075-f007){ref-type="fig"}A). We found that *Pik3c3* and *Enox1* were more highly expressed in the HF group (*p* = 0.0028 and *p* = 0.025, respectively) while *Adh5* was more highly expressed in the LF group (*p* = 0.029). We then examined the DMRs within 1 mb of each gene. Two DMRs were located upstream of the *Pik3c3* TSS and were more methylated in the LF group as identified by MeDIP-seq and MRE-seq ([Figure 7](#nutrients-11-02075-f007){ref-type="fig"}A). Four DMRs were located upstream of the *Enox1* TSS. The two more distant DMRs (−889,247 and −888,747 bp upstream) were hypomethylated in the LF group, while the two more proximal DMRs (−472,747 and −373,747 bp upstream) were hypermethylated in the LF group ([Figure 7](#nutrients-11-02075-f007){ref-type="fig"}C). Two DMRs were located upstream of the *Adh5* TSS. MeDIP-seq and MRE-seq showed that the more distant DMR (−391,108 bp upstream) was hypermethylated in the LF group while the more proximal DMR (−170,108 bp upstream) was hypomethylated in the LF group ([Figure 7](#nutrients-11-02075-f007){ref-type="fig"}D). 4. Discussion {#sec4-nutrients-11-02075} ============= In this study, we investigated the role of maternal diet in hepatic epigenomic programming. Specifically, male Sprague-Dawley rats were fed either a LF or a HF diet during gestation and lactation. After weaning, all animals were given a HF diet challenge. After nine weeks on the post-weaning diet, there was no difference in body weight between groups; however, hepatic DNA methylation was changed at 1419 loci. Closer examination of the DMRs revealed enrichment for metabolic and cancer pathways. Gene expression analysis showed that only *Map3k5* and *Igf1r* were differentially expressed. Finally, we looked at the conserved intergenic DMRs and found that they were also located nearby differentially expressed metabolic genes. There was no difference in body weight and there were very few changes in gene expression between perinatal LF- and HF-exposed animals. Previous experiments have reported inconsistent results regarding body weight of animals that were given an obesogenic post-weaning diet preceded by different perinatal diets. After 14 weeks on a post-weaning HF diet, mice that had been exposed to a maternal HF diet had higher body weights than those exposed to a maternal LF diet \[[@B29-nutrients-11-02075],[@B30-nutrients-11-02075]\]. Conversely, another study found that maternal dietary fat had no impact on body weight when male mouse offspring were fed a post-weaning HF diet for 17 weeks \[[@B31-nutrients-11-02075]\]. Finally, others have suggested that the impact of maternal diet is time dependent. In Sprague-Dawley rats, it was shown that perinatal diet did not change body weight after eight weeks of post-weaning HF feeding but did produce body weight differences after 16 weeks of post-weaning HF feeding \[[@B32-nutrients-11-02075]\]. In our case, we saw that the perinatal diet was not protective against HF-induced postnatal weight gain after nine weeks of HF feeding. Additionally, our study is limited because we did not consider any other metabolic parameters. Previous work has shown that a prenatal HF diet impacts glucose tolerance, lipid profile, and cardiovascular health \[[@B31-nutrients-11-02075],[@B33-nutrients-11-02075],[@B34-nutrients-11-02075],[@B35-nutrients-11-02075]\]. Such physiological characteristics have also been associated with DNA methylation \[[@B36-nutrients-11-02075],[@B37-nutrients-11-02075],[@B38-nutrients-11-02075]\], so it is possible that the perinatal diet acts through epigenetic mechanisms to affect metabolic outcomes. The changes in DNA methylation but lack of body weight difference suggest that either perinatally programmed DNA methylation is slowly washed out by a new dietary challenge, or that DNA methylation programming is robust and foreshadows distinct future metabolic outcomes. Given previous findings, we hypothesize that our rats may have been sacrificed before weight differences could be observed. Because metabolic parameters such as resting metabolic rate, macronutrient oxidation, and body composition change during aging \[[@B39-nutrients-11-02075]\], it is possible that younger rats were able to compensate for perinatal perturbations. A similar principle might also explain the small number of differentially expressed genes. Younger animals might be better equipped to combat disturbances in the methylome. Conversely, it may be the case that DNA methylation established during the perinatal period is susceptible to change by post-weaning diet and thus yields no observable difference in the future metabolic outcome. Indeed, exposure to a post-weaning HF diet was shown to mitigate the epigenomic effects of the early-life diet \[[@B31-nutrients-11-02075]\]. Perinatal diet impacted 1,505 DMRs in male offspring given a post-weaning LF, but only 258 DMRs in offspring fed a post-weaning HF diet. We have also highlighted the important role of the post-weaning diet in establishing DNA methylation patterns \[[@B20-nutrients-11-02075]\]. While we previously found that a post-weaning HF diet increased body weight and impacted 3,966 DMRs, we show here that perinatal diet did not alter body weight and only affected 1419 DMRs. This suggests that post-weaning diet is a strong predictor of body weight and perhaps a more powerful determinant of DNA methylation than the perinatal diet. Future investigation should quantify the contributions of gestation, lactation, and post-weaning diet in determining DNA methylation levels in order to understand the dynamic nature of the methylome. Further work should also explore whether epigenetic alterations have the same impact on gene expression and weight gain in older individuals. Altogether, we measured five differentially expressed genes. Two genes, *Map3k5* and *Igf1r*, contained intragenic DMRs. Previous studies suggested that the two genes respond to metabolic stressors. *Map3k5* is activated by oxidative stress and inflammation \[[@B40-nutrients-11-02075]\], and *Igf1r* participates in insulin signaling and fatty acid uptake \[[@B41-nutrients-11-02075],[@B42-nutrients-11-02075]\]. Moreover, *Map3k5* has been shown to be upregulated in the adipose tissue of obese individuals while *Igf1r* is upregulated in lymphocytes of obese children \[[@B43-nutrients-11-02075],[@B44-nutrients-11-02075]\]. Because a HF diet can induce oxidative stress and inflammation, it is not surprising that we observed lower expression of *Map3k5* in LF-fed animals. Furthermore, the decrease in *Igf1r* expression in the LF group might be attributed to a reduced need for IGF1-mediated fatty acid uptake. Interestingly, expression of both genes was reduced under perinatal LF conditions, but while *Igf1r* had a hypomethylated DMR, *Map3k5* contained a hypermethylated DMR. This might have been due to the position of the DMRs relative to the TSS of each gene. In *Igf1r*, the DMR was located in the first intron, where it could impact transcription in a canonical manner (i.e., increased DNA methylation resulting in decreased gene expression). The DMR associated with *Map3k5* was located much further downstream of the TSS in intron 8. Several studies have suggested that gene body methylation might have non-traditional effects on transcription, which could account for the discrepancy we observed \[[@B45-nutrients-11-02075],[@B46-nutrients-11-02075],[@B47-nutrients-11-02075]\]. Amongst the genes associated with intergenic DMRs, *Adh5*, *Enox1*, and *Pik3c3* were differentially expressed. Given their role in metabolic pathways, it is not surprising that these three genes were differentially expressed as a result of dietary treatments. The reactions catalyzed by *Adh5* and *Enox1* both involve nicotinamine adenine dinucleotide (NAD). The balance between the oxidized and reduced form of the NAD cofactor is important for driving metabolic reactions. *Pik3c3* plays a role in autophagy, which has been shown to be altered by different exposures to a HF diet \[[@B48-nutrients-11-02075],[@B49-nutrients-11-02075],[@B50-nutrients-11-02075]\]. We hypothesize that DNA methylation in more distant areas could be indicative of chromatin state, such that high methylation is correlated with closed chromatin \[[@B51-nutrients-11-02075]\]. This is reflected in the DNA methylation and gene expression patterns that we observed. *Pik3c3* and *Enox1* were expressed at lower levels in the LF group and the proximal DMRs were hypermethylated. Conversely, *Adh5* was highly expressed in the LF group and the proximal DMR was hypomethylated. This study provides insight into early life nutritional programming; however, limitations should be addressed in further experimentation. Here, we only explored dietary effects in male offspring. In previous studies, we have demonstrated sex-specific physiological and molecular changes after HF diet exposure \[[@B52-nutrients-11-02075],[@B53-nutrients-11-02075],[@B54-nutrients-11-02075]\]. We have also found particularly robust changes in hepatic gene expression and DNA methylation in male rats \[[@B16-nutrients-11-02075],[@B20-nutrients-11-02075],[@B55-nutrients-11-02075],[@B56-nutrients-11-02075]\]. Although this study expands upon our prior findings in males, we cannot assume that these results would broadly apply to females. Further investigation is necessary to test whether diet-induced DNA methylation patterns are observed in both sexes. Another limitation of the current study is that epigenomic profiling was performed on whole liver tissue. While bulk tissue measurements have been previously used to uncover differences in hepatic gene expression and epigenetic profiles \[[@B57-nutrients-11-02075],[@B58-nutrients-11-02075],[@B59-nutrients-11-02075]\], future studies should consider isolating hepatocytes in order to better predict the functional role of differences in DNA methylation. Our findings represent a novel contribution to the field of nutritional epigenetics. Whereas several other studies have taken gene-targeted approaches to study DNA methylation, we interrogated the entire epigenome to uncover novel differences in DNA methylation. This allowed us to better investigate intergenic DMRs. Previous analysis has focused on gene-associated DMRs located within 1,500 bp upstream and downstream of a gene body. Indeed, certain technologies such as reduced representation bisulfite sequencing are designed to specifically identify DNA methylation differences in regions near gene promoters. Using these methods, the majority of identified DMRs have the advantage of being functionally interpretable. However, intergenic regions are often overlooked. Interestingly, we found intergenic DMRs to account for more than half of all differential methylation. Intergenic DNA methylation is thought to play an important in cellular function and disease. Intergenic DMRs can impact transcription via interaction with enhancer elements \[[@B60-nutrients-11-02075]\], and intergenic demethylation is observed in various cancers \[[@B61-nutrients-11-02075]\]. Currently, there are no bioinformatics approaches to identify specific functions for large numbers of intergenic DMRs. We hypothesized that proximal genes with significant chromatin contact were most likely to be affected by differential methylation. Although this hypothesis was supported by three differentially expressed genes, understanding the role of intergenic methylation should still be a priority in order to facilitate the development of new computational tools to annotate these regions. 5. Conclusions {#sec5-nutrients-11-02075} ============== Overall, we conclude that perinatal diet impacts hepatic DNA methylation, especially in metabolic and cancer-related pathways. On the other hand, early-life LF diet is not adequate to prevent postnatal weight gain induced by an HF diet. Although minimal, a prenatal LF diet produced changes in gene expression, including increased the expression of *Map3k5*, *Igf1r*, and *Adh5*, and decreased the expression of *Enox1* and *Pik3c3*. Our findings suggest that diet-mediated epigenetic marks established during early life persist despite a HF diet challenge during the postnatal period. We would like to thank Dan Zhou, Huan Wang, and Guanying Bianca Xu who contributed to the success of this study. Conceptualization, L.M., H.C., and Y.-X.P.; Formal analysis, L.M. and J.S.; Funding acquisition, H.C. and Y.-X.P.; Investigation, L.M. and J.S.; Methodology, L.M., J.S., and H.C.; Project administration, Y.-X.P.; Resources, H.C. and Y.-X.P.; Software, L.M. and H.C.; Supervision, Y.-X.P.; Visualization, L.M.; Writing--original draft, L.M.; Writing--review & editing, L.M., J.S., H.C., and Y.-X.P. This project was supported by UIUC Research Board grant \#12192, the USDA Cooperative State Research, Education and Extension Service, Hatch project numbers \# ILLU-971-344 and ILLU-698-391, and the Data Purchase Program from University Library. The authors declare no conflicts of interest. {#nutrients-11-02075-f001} {#nutrients-11-02075-f002} {#nutrients-11-02075-f003} {#nutrients-11-02075-f004} {#nutrients-11-02075-f005} {#nutrients-11-02075-f006} {#nutrients-11-02075-f007} nutrients-11-02075-t001_Table 1 ###### Methylation specific PCR (MSP) primers. Gene CpG Site Position Methylation Primers (5′ → 3′) Efficiency \* ----------------- ----------------- ------------------------------------- ------------------------------------ --------------------------- --------------- Gpam CpG 1 Forward +52,479 U GTAGTGGAATAAGAAGTTTTCGGAG 98.20% Reverse +52,587 U ACCTTCAAATAACAATCACGCTAC Forward +52,485 M AGAGTAGTAGTGGAATAAGAAGTTTTTGG 99.10% Reverse +52,586 M CACCTTCAAATAACAATCACACTAC CpG 2 Forward +44,337 U GGTGGAGAGGTATTTGTTATGGA 108.20% Reverse +44,416 U CCACACAATCTACAAACTTCACAA Forward +44,332 M TTAGAGGTGGAGAGGTATTTGTTAC 101.19% Reverse +44,414 M ACACAATCTACAAACTTCACGAAAA CpG 3 Forward +7619 U TGTAATTTTTTATTTTAATTTATGTGATTTTTGA 95.56% Reverse +7736 U TTTCTACTTCACAATTACTAATCAACCCA Forward +7633 M TTAATTTACGTGATTTTTGATTGTTATTATTTT 98.50% Reverse +7736 M CTACTTCACGATTACTAATCAACCCG CpG 4 Forward +7288 U AAGTTAAGTTGTAGTGGTTGGGTAATTG 93.47% Reverse +7360 U CCCACTTATTTTAAACAACATCAAACC Forward +7289 M AGTCGTAGTGGTCGGGTAATCG 107.81% Reverse +7357 M CCGCTTATTTTAAACAACATCGAA Myh7b CpG 1 Forward +9089 U TGGGTTCGTGTGGGAAATG 99.33% Reverse +9181 U CCACCTCAACTCTCCCTAAACAA Forward +9089 M CGGGTTCGTGTGGGAAAC 92.99% Reverse +9181 M CCGCCTCAACTCTCCCTAAA CpG 2 Forward +9607 U AGGAGTATAAATGGGAGGGTATTGATT 91.86% Reverse +9757 U CATACACAACTTCCAACACCATCC Forward +9621 M CGTGTTGGAGTAGGAGGAGTATAAAC 102.11% Reverse +9753 M CGCAACTTCCGACACCATC CpG 3 Forward +10,000 U TGATTTGAGGATTATGTGTATTGGATTT 99.04% Reverse +10,069 U CCAATTTCTTTTTCCATTCTCCATAC Forward +9995 M CGAGGATTACGTGTATTGGATTTTAA 103.13% Reverse +10,065 M TTTCTTTTTCCATTCTCCGTACAATA CpG 4 Forward +11,606 U AGGAGTGTATGTTTTTTAAGGTTTTAGATG 96.42% Reverse +11,701 U AACAAAACTACTAAAAATTAAATAACTTCCCA Forward +11,606 M AGGAGTGTATGTTTTTTAAGGTTTTAGACG 98.73% Reverse +11,701 M ACGAAACTACTAAAAATTAAATAACTTCCCA CpG 5 Forward +12,857 U GATTTGGATTTGTTGTTAAGGGTTTT 102.17% Reverse +12,924 U AACCAACACCCACCACTACCTAA Forward +12,859 M TTCGGATTTGTCGTTAAGGGTT 106.16% Reverse 12,927 M CCAACACCCACCGCTACC CpG 6 Forward +13,120 U TTATTTGGATATGGGATAAGAGAGGG 98.27% Reverse +13,216 U CACCATCTAAAATAATACTACTTTCTTTCACTTAT Forward +13,120 M TTATTTGGATACGGGATAAGAGAGG 102.64% Reverse +13,219 M CGTCTAAAATAATACTACTTTCTTTCGCTT \* The qPCR amplification efficiency is calculated based on the slope of the standard curve <https://www.lifetechnologies.com/us/en/home/brands/thermo-scientific/molecular-biology/molecular-biology-learning-center/molecular-biology-resource-library/thermo-scientific-web-tools/qpcr-efficiency-calculator.html>. Slopes between −3.1 and −3.6 giving reaction efficiencies between 90 and 110% are typically acceptable. nutrients-11-02075-t002_Table 2 ###### Gene expression primers. Gene Ensembl ID Common Name Position Primers (5′ → 3′) Efficiency \* --------------- -------------------------- --------------------------------------------------------- --------------- --------------------------- --------------- Acacb ENSRNOT00000078868.1 Acetyl-CoA Carboxylase Beta Forward +1081 ACCCCAAACTTCCAGAGC 105.70% Reverse +1189 TGGGCTACAATGGTGGAG Acox3 ENSRNOT00000049798.3 Acyl-CoA Oxidase 3, Pristanoyl Forward +1553 TGACTGGTTGGACTCAGA 94.56% Reverse +1629 TCTGATGACTCTCTCGGA Actb ENSRNOT00000042459.4 Actin Beta Forward +451 GAGACCTTCAACACCCCAGC 104.67% Reverse +526 CAGTGGTACGACCAGAGGCA Acvr1c ENSRNOT00000059280.4 Activin A Receptor Type 1C Forward +511 TGATTTATGATGCCACTGCC 100.24% Reverse +586 ATTGTCCTTGCGATGGTTCT Adh1 ENSRNOT00000036993.4 Alcohol Dehydrogenase 1 Forward +257 ATGAAGGAGTTGGGATAG 100.37% Reverse +318 ATCACCTGTTCTTACGCT Adh5 ENSRNOT00000016891.6 Alcohol Dehydrogenase 5 Forward +637 GTGTGTCTGATTGGATGTGG 117.98% Reverse +700 TGACCTTGGCAGTGTTGA Adh7 ENSRNOT00000015870.4 Alcohol Dehydrogenase 7 Forward +262 GAAGCAGTTGGGATTGTGGAGA 108.18% Reverse +328 TCACTTTGTCACCTGGTCTCACTG Cacng8 ENSRNOT00000078444.1 Calcium Voltage-Gated Channel Auxiliary Subunit Gamma 8 Forward +297 CTGCGTGAAGATCAACCACT 108.34% Reverse +395 ATAGGAAAGATGCTGGAGGC Cpt2 ENSRNOT00000016954.3 Carnitine Palmitoyltransferase 2 Forward +310 GACACCATGAAGAGATACCT 107.28% Reverse +387 ACACAACGCTTCTGTTCT Dapk3 ENSRNOT00000027634.4 Death Associated Protein Kinase 3 Forward +624 TTCGTCGCCCCTGAGATTGTAA 103.21% Reverse +685 ATGACGCCGATGCTCCACATAT Dcc ENSRNOT00000064947.3 DCC Netrin 1 Receptor Forward +1144 GTGGCTGAAAATGAGGCTGGC 100.76% Reverse +1208 ATGGCAGGCTTGGGGACAA Enox1 ENSRNOT00000074868.2 Ecto-NOX Disulfide-Thiol Exchanger 1 Forward +167 TTGAGAGCATCGCACAGTGT 93.82% Reverse +239 ATGCTCCCCAAACCATCA Fads2 ENSRNOT00000059280.4 Fatty Acid Desaturase 2 Forward +711 CGTGTTTGTCCTTGGAGAGTGGC 108.89% Reverse +790 CATGCTGGTGGTTGTAGGGCA Flnb ENSRNOT00000066546.1 Filamin B Forward +353 GCTGGAGAATGTGTCTGT 108.69% Reverse +422 ACTGTCAATGGACACGAG Foxo1 ENSRNOT00000018244.5 Forkhead Box O1 Forward +994 AGGATAAGGGCGACAGCAACAG 102.12% Reverse +1056 GGGACAGATTGTGGCGAATTG Fzd2 ENSRNOT00000032944.2 Frizzled Class Receptor 2 Forward +818 TTTTGCCCGTCTCTGGAT 93.13% Reverse +889 TAGGTGGTGACCGTGAAGAA Gapdh ENSRNOT00000050443.4 Glyceraldehyde-3-Phosphate Dehydrogenase Forward +220 CTCTACCCACGGCAAGTTCAACG 100.39% Reverse +311 CTCGCTCCTGGAAGATGGTGATG Hk3 ENSRNOT00000031935.2 Hexokinase 3 Forward +973 CCCTGGTTCCTGGTGCTCAG 119.40% Reverse +1050 CCAGCACCAGCCTTACCAGC Igfr1 ENSRNOT00000019267.6 Insulin Like Growth Factor 1 Receptor Forward +2239 CTGAGAGGAGGCGGAGAGATG 109.39% Reverse +2304 TGTTCCTGCTTCGGCTGG Il1r1 ENSRNOT00000019673.4 Interleukin 1 Receptor Type 1 Forward +393 GGGTTCATTTGTCTCATTGTGC 101.20% Reverse +465 TGACCTCATTTGGATACTCCGT Irs2 ENSRNOT00000032918.6 Insulin Receptor Substrate 2 Forward +3373 CTTGAAGCGGCTAAGTCT 109.86% Reverse +3435 TGGCTGACTTGAAGGAAG Lpin3 ENSRNOT00000022403.5 Lipin 3 Forward +578 CCCTCATCGCAGCCTAAAGACAT 108.38% Reverse +657 AGGTCAGCAGATGAAAGGTTGGC Map3k5 ENSRNOT00000051496.6 Mitogen-Activated Protein Kinase Kinase Kinase 5 Forward +455 GTTTTTACAACGCTGACATCGC 105.77% Reverse +525 ATGATAAAACAGGGAAGGCTGC Mapk3 ENSRNOT00000087625.1 Mitogen-Activated Protein Kinase 3 Forward +618 CACTGGCTTTCTTACCGAGT 111.47% Reverse +696 GGTGTAGCCCTTGGAGTTAA Mknk2 ENSRNOT00000041106.5 MAP Kinase Interacting Serine/Threonine Kinase 2 Forward +177 TTCAGGGCTTCCACCGTTCG 107.33% Reverse +246 TGGGCGGGGTCTAAGGTGAA Msh3 ENSRNOT00000018449.7 MutS Homolog 3 Forward +166 TGTCCCCCACAGAACCAGCA 109.37% Reverse +229 TTCCCCAGTGACCTCTTCCTGC Mttp ENSRNOT00000014631.6 Microsomal Triglyceride Transfer Protein Forward +1042 TAGAACCTGAGAACCTGTCCAACGC 107.54% Reverse +1113 AAGTGCGGAGGTGCTGAATGAAG Ncoa4 ENSRNOT00000066062.3 Nuclear Receptor Coactivator 4 Forward +611 CCTAGTTCTTCAAGTGTCAGGC 108.75% Reverse +686 TGGATGCTGACTTCTGCTCT Nfatc2 ENSRNOT00000065615.1 Nuclear Factor Of Activated T Cells 2 Forward +1599 GGAGCCAAAGAACAACATGCGGG 100.47% Reverse +1674 CAGCTCGATGTCAGCGTTTCGGA Ntrk2 ENSRNOT00000042145.2 Neurotrophic Receptor Tyrosine Kinase 2 Forward +979 TCCTGGACAAACTCGTCA 99.47% Reverse +1058 GGCTTACAAGGCGTTTCT Pik3c3 ENSRNOT00000066816.2 Phosphatidylinositol 3-Kinase Catalytic Subunit Type 3 Forward +60 CTGTGACCTGGACATCAA 93.26% Reverse +119 TGTTCTCTCTTCCCTTCC Phkb ENSRNOT00000049624.4 Phosphorylase Kinase Regulatory Subunit Beta Forward +369 GCCATAAAGTGTATGAGAGGAG 106.60% Reverse +435 TGAACTGCTGGACCTTATCA Ppp1r3b ENSRNOT00000051720.2 Protein Phosphatase 1 Regulatory Subunit 3B Forward +838 TATGAAAGAATGGAGTTCGCCGTG 107.84% Reverse +909 TTTGCCTTTGTTGCTGTCCCAGTA Rasgrp2 ENSRNOT00000028646.6 RAS Guanyl Releasing Protein 2 Forward +490 CAAGAAGGAAACCGCAGGCAC 108.06% Reverse +565 TCACCTGCCGCTTCCACTTGT Rpl7a ENSRNOT00000044551.4 Ribosomal Protein L7a Forward +64 GAGGCCAAAAAGGTGGTCAATCC 105.33% Reverse +127 CCTGCCCAATGCCGAAGTTCT Runx1t1 ENSRNOT00000066191.1 RUNX1 Translocation Partner 1 Forward +796 TCCCACTGAGACCTTTTG 109.60% Reverse +894 CAGGGTTCTGTTTGGCTA Scd ENSRNOT00000018447.5 Stearoyl-CoA Desaturase Forward +942 TCAATCTCGGGAGAACATCCTG 109.77% Reverse +1013 AAGGCGTGATGGTAGTTGTGGA Scp2 ENSRNOT00000015420.5 Sterol Carrier Protein 2 Forward +203 GGCTATGTGTACGGTGAATCCA 105.56% Reverse +280 AATGATAGGGATGCCAGTCAGC Skp2 ENSRNOT00000089178.1 S-Phase Kinase Associated Protein 2 Forward +809 CTGGATTTTCTGAGTCTGCC 100.52% Reverse +882 CCAGGAGAGGTTCAGTTCAT Taok3 ENSRNOT00000089043.1 TAO Kinase 3 Forward +403 GCTGAAGCACCCGAACACCAT 103.35% Reverse +476 ACTCCATCACCAACCAAGCGG Tgfb2 ENSRNOT00000003313.5 Transforming Growth Factor Beta 2 Forward +1664 ACAATGCTAACTTCTGTGCTGG 91.24% Reverse +1735 TGAGGACTTTGGTGTGTTGTGT Traf3 ENSRNOT00000010906.6 TNF Receptor Associated Factor 3 Forward +870 CTCTTCTGAGGAGTGAGTTGA 108.60% Reverse +942 CGCTTAAAACTACAGGTGC Wnt6 ENSRNOT00000023439.6 Wnt Family Member 6 Forward +452 GGGGGTTCGAGAATGTCAGTTCC 103.79% Reverse +517 GCCTTGCTGTGACTGGAGCAGTT Zbtb16 ENSRNOT00000045356.3 Zinc Finger and BTB Domain Containing 16 Forward +1695 GCATTTACTGGCTCATTCAG 100.89% Reverse +1770 ATCTTCCTTTGAGAACTGGG \* The qPCR amplification efficiency is calculated based on the slope of the standard curve <https://www.lifetechnologies.com/us/en/home/brands/thermo-scientific/molecular-biology/molecular-biology-learning-center/molecular-biology-resource-library/thermo-scientific-web-tools/qpcr-efficiency-calculator.html>. Slopes between −3.1 and −3.6 giving reaction efficiencies between 90 and 110% are typically acceptable. nutrients-11-02075-t003_Table 3 ###### Gene ontology clustering. GO Number GO Annotation Count Fold Enrichment FDR *p*-Value ----------------------------------------------- ------------------------------------------------------- ------- ----------------- --------------- Annotation Cluster 1 (Enrichment Score: 6.24) GO:0046872 Metal ion binding 111 1.54 2.37 × 10^−4^ GO:0043167 Ion binding 113 1.52 1.70 × 10^−4^ GO:0043169 Cation binding 111 1.52 1.61 × 10^−4^ Annotation Cluster 2 (Enrichment Score: 5.24) GO:0000902 Cell morphogenesis 26 3.02 0.0013 GO:0032989 Cellular component morphogenesis 26 2.73 0.0045 GO:0000904 Cell morphogenesis involved in differentiation 20 3.27 0.0046 Annotation Cluster 3 (Enrichment Score: 4.57) GO:0005216 Ion channel activity 24 2.83 0.0016 GO:0005261 Cation channel activity 20 3.18 0.0017 GO:0022838 Substrate specific channel activity 24 2.75 0.0018 GO:0015267 Channel activity 24 2.65 0.0028 GO:0022803 Passive transmembrane transporter activity 24 2.65 0.0028 GO:0022836 Gated channel activity 20 2.95 0.0033 Annotation Cluster 4 (Enrichment Score: 4.18) GO:0048667 Cell morphogenesis involved in neuron differentiation 19 3.63 0.0029 GO:0000904 Cell morphogenesis involved in differentiation 20 3.27 0.0046 GO:0007409 Axonogenesis 17 3.61 0.0056 GO:0048666 Neuron development 23 2.62 0.014 GO:0048812 Neuron projection morphogenesis 17 3.12 0.021 GO:0031175 Neuron projection development 19 2.75 0.034 GO:0048858 Cell projection morphogenesis 17 2.82 0.045 GO:0032990 Cell part morphogenesis 17 2.71 0.066 Annotation Cluster 5 (Enrichment Score: 4.08) GO:0022836 Gated channel activity 20 2.95 0.0033 GO:0005244 Voltage-gated ion channel activity 15 3.47 0.0062 GO:0022832 Voltage-gated channel activity 15 3.47 0.0062 Annotation Cluster 6 (Enrichment Score: 3.45) GO:0051960 Regulation of nervous system development 17 2.95 0.036 GO:0060284 Regulation of cell development 17 2.87 0.044 GO:0045664 Regulation of neuron differentiation 14 3.31 0.042 GO:0050767 Regulation of neurogenesis 15 2.86 0.086 nutrients-11-02075-t004_Table 4 ###### DMR description. Gene Function Chromosome Location Genomic Feature HF MeDIP LF MeDIP HF MRE LF MRE Conservation --------- ------------------- ------------ ---------------------- ----------------- -------------- --------------- -------------- --------------- ------------------ ACVR1C Cancer 3 40041000--40041500 Intron 3/8 2.33 (0.89) 2.58 (1.02) 1.64 (0) 0 0.0040 (0.016) CACNG2 Cancer 7 116000000--116000500 Intron 1/3 8.27 (3.47) 12.26 (4.78) 0.74 (0.42) 0.31 (0.24) 0.77 (0.36) CACNG8 Cancer 1 64069000--64069500 Exon 4/4 2.56 (1.15) 2.00 (0.82) 0.27 (0) 6.07 (0) 0.18 (0.34) DAPK3 Cancer 7 10009000--10009500 Exon 7/9 12.89 (6.18) 18.09 (10.40) 1.10 (0.68) 0.14 (0) 0.38 (0.45) DCC Cancer 18 69043500--69044000 Intron 1/28 8.56 (3.39) 21.00 (9.64) 3.68 (1.51) 2.16 (1.46) 0.0054 (0.0087) FLNB Cancer 15 19052000--19052500 Intron 1/44 4.89 (2.41) 5.98 (2.59) 2.06 (2.03) 0.51 (0.58) 0.014 (0.017) FZD2 Cancer 10 91707500--91708000 Promoter 6.27 (3.41) 1.40 (0.55) 0.27 (0.21) 0.95 (0.71) 0.061 (0.094) IGF1R Cancer 1 122713500--122714000 Intron 1/20 4.31 (1.20) 3.10 (1.81) 0.71 (0) 1.39 (0.99) 0.0057 (0.011) IL1R1 Cancer 9 39589500--39590000 Intron 1/10 7.39 (4.28) 4.33 (1.62) 0.12 (0) 1.00 (0.93) 0.0053 (0.0099) MAP3K5 Cancer 1 15307000--15307500 Intron 8/12 1.88 (0.83) 4.86 (2.15) 7.37 (0) 1.13 (0) 0.018 (0.050) MSH3 Cancer 2 22480500--22481000 Intron 20/23 10.93 (4.44) 4.08 (2.34) 0.42 (0.42) 1.60 (0.93) 0.0050 (0.014) NCOA4 Cancer 16 7641000--7641500 Promoter 12.59 (3.16) 5.25 (2.38) 0.27 (0.085) 1.10 (0.84) 0.0026 (0.0036) NFATC2 Cancer 3 159682000--159682500 Intron 8/9 10.58 (5.74) 3.06 (1.24) 1.07 (0) 1.73 (2.32) 0.015 (0.0038) NTRK2 Cancer 17 11700500--11701500 Intron 12/13 7.81 (3.75) 17.72 (8.57) 4.31 (3.07) 1.21 (1.92) 0.28 (0.41) RASGRP2 Cancer 1 209198000--209198500 Intron 15/16 4.29 (2.91) 6.10 (4.07) 0.85 (0.14) 0.19 (0) 0.22 (0.33) RUNX1T1 Cancer 5 28230500--28231000 Exon 1/11 2.63 (1.09) 1.00 (0) 1.70 (1.81) 8.70 (11.88) 0.31 (0.43) SKP2 Cancer 2 58774500--58775000 Exon 8/10 8.21 (4.43) 17.98 (10.69) 0.93 (1.08) 0 0.13 (0.30) TAOK3 Cancer 12 40571000--40571500 Intron 5/19 12.23 (7.54) 7.64 (3.88) 1.51 (2.00) 4.16 (5.87) 0.23 (0.41) TGFB2 Cancer 13 102723000--102723500 Intron 6/7 10.45 (6.36) 4.61 (2.68) 0.24 (0) 3.47 (0.13) 0.39 (0.47) TRAF3 Cancer 6 135924000--135924500 Promoter 5.27 (2.45) 1.50 (0.71) 0.47 (0.57) 10.45 (21.93) 0.036 (0.073) WNT6 Cancer 9 74115500--74117500 Intron 3/3 8.30 (6.35) 15.22 (7.88) 3.40 (4.52) 3.54 (5.89) 0.26 (0.43) ZBTB16 Cancer 8 52035000--52036000 Intron 2/6 5.24 (2.69) 7.58 (2.93) 4.00 (0.63) 0.70 (0.30) 0.45 (0.49) FOXO1 Cancer/Metabolism 2 141195500--141196000 Intron 1/2 4.84 (2.03) 1.33 (0.58) 0.18 (0.084) 1.31 (0.27) 0.013 (0.026) MAPK3 Cancer/Metabolism 1 185936000--185936500 Exon 2/10 7.00 (3.22) 13.88 (6.77) 8.62 (0) 2.72 (0) 0.74 (0.39) MKNK2 Cancer/Metabolism 7 10559000--10559500 Exon 14/14 8.85 (3.70) 14.22 (6.72) 3.20 (2.54) 0.98 (0.97) 0.32 (0.42) ACACB Metabolism 12 43424500--43425000 Intron 30/58 4.67 (2.22) 5.63 (2.80) 9.21 (0.59) 3.01 (1.59) 0.0020 (0.0041) ACOX3 Metabolism 14 80781000--80781500 Intron 13/19 15.48 (9.69) 3.00 (1.58) 0 0 0.0047 (0.016) FADS2 Metabolism 1 212532500--212533000 Intron 4/10 7.83 (3.78) 12.44 (4.55) 1.67 (1.44) 0.87 (0.92) 0.27 (0.44) IRS2 Metabolism 16 83384000--83384500 Intron 1/1 9.39 (4.22) 1.67 (0.71) 0.24 (0.17) 2.82 (0.27) 0.00077 (0.0032) PHKB Metabolism 19 22453500--22454000 Intron 13/29 4.43 (2.08) 11.92 (5.76) 0.91 (0.18) 0.19 (0) 0.038 (0.12) PPP1R3B Metabolism 16 60562000--60562500 Intron 1/1 4.08 (2.56) 1.00 (0) 0.49 (0) 6.85 (0) 0.026 (0.076) SCD1 Metabolism 1 249463500--249464000 Intron 5/5 10.82 (4.11) 16.68 (5.85) 1.89 (1.58) 0.47 (0.40) 0.012 (0.037) MeDIP and MRE values are given as average reads (standard deviation). Conservation denotes conservation score (standard deviation).

1. INTRODUCTION {#sec1} =============== Reactive oxygen species (ROS) have been implicated in the pathogenesis of various human diseases, particular cardiac disorders \[[@B1], [@B2]\]. ROS include a number of oxygen-containing reactive species, among which superoxide and hydrogen peroxide (H~2~O~2~) are most widely studied in biomedical field \[[@B3], [@B4]\]. ROS attack cellular constitutes, including lipids, proteins, and nucleic acids, leading to oxidative stress and tissue injury. Due to the detrimental effects of ROS, mammalian cells have evolved a number of antioxidants to protect cells from oxidative injury. However, under pathophysiological conditions, ROS formation may overwhelm the normal cellular antioxidant defenses, and as such oxidative tissue injury and disease process occur \[[@B3], [@B4]\]. There are a number of cellular sources of ROS formation, which include mitochondrial electron transport chain (ETC), NAD(P)H oxidase, uncoupled nitric oxide synthases, and cytochrome P450 enzymes \[[@B3], [@B4]\]. The above cellular sources of ROS formation are implicated in oxidative stress of various tissues, including vasculature and myocardium \[[@B1], [@B2]\]. Because of the multiple cellular component nature of mammalian tissues, detection of cell-specific formation of ROS is of critical importance for understanding the oxidative mechanisms of disease process and for developing antioxidative stress-based strategies for intervention of oxidative tissue degeneration. In this context, over the last several decades various methods have been developed for detecting ROS formation in biological systems. These include the electron paramagnetic resonance (EPR) spin trapping, luminescent methods, as well as a number of biochemical assays \[[@B1], [@B5], [@B6]\]. While EPR spin trapping technique is of the highest specificity for detecting oxygen radicals, its sensitivity is limited \[[@B5], [@B6]\]. Moreover, EPR technique and most of the biochemical assays for detecting ROS are unable to directly visualize the ROS formation in a highly sensitive real-time manner. Direct sensitive in situ detection of ROS formation from cultured cells provides important insight into the molecular and biochemical mechanisms by which ROS mediate pathophysiological processes. Accordingly, in this study, we have applied the highly sensitive chemiluminescence (CL) imaging technique to investigate the in situ ROS formation in cultured monolayers of rat H9c2 cardiomyocytes and the factors that modulate the ROS generation in this widely used in vitro model for cardiac cell physiology and pathophysiology. 2. MATERIALS AND METHODS {#sec2} ======================== 2.1. Chemical and reagents {#subsec2.1} -------------------------- D3T with a purity of 99.8% was generously provided by Dr. Mary Tanga at SRI International (Menlo Park, Calif, USA) and Dr. Linda Brady at the National Institute of Mental Health (Bethesda, Md, USA). Benzo(a)pyrene-1,6-quinone (BPQ) was from Midwest Research Institute (Kansas City, Mo, USA). Dulbecco\'s modified Eagle\'s medium (DMEM), penicillin, streptomycin, and fetal bovine serum (FBS) were from Gibco-Invitrogen (Carlsbad, Calif, USA). All other chemicals and reagents were from Sigma-Aldrich (St. Louis, Mo, USA). 2.2. Cell culture and treatment {#subsec2.2} ------------------------------- Rat H9c2 cardiomyocytes (ATCC, Manassas, Va, USA) were cultured in DMEM supplemented with 10% FBS, 100 U/mL of penicillin, and 100 *μ*g/mL of streptomycin in tissue culture flasks at 37°C in a humidified atmosphere of 5% CO~2~. The cells were fed every 2-3 days, and subcultured once they reached 70--80% confluence. For CL imaging experiments, 2 × 10^5^ cells were plated in each well of 6-well plates. The CL images were acquired directly with the confluent cells (2.8 × 10^5^ cells/well or 0.3 × 10^5^/cm^2^) in culture at 37°C after washing of the cells with phosphate buffered saline (PBS). To initiate the CL response, 10 *μ*M luminol and 5 *μ*g/mL horseradish peroxidase (HRP) were added to the cell monolayers in 6-well plates, with each well containing 2 mL PBS. For experiments on induction of antioxidants, H9c2 cells were incubated with 100 *μ*M D3T dissolved in dimethyl sulfoxide (DMSO; 0.1% final concentration) in culture medium for 48 hours. Control cells received 0.1% DMSO only. 2.3. Cell extract preparation {#subsec2.3} ----------------------------- H9c2 cells were collected and resuspended in ice-cold 50 mM potassium phosphate buffer, pH 7.4, containing 2 mM EDTA and 0.1% Triton X-100. The cells were sonicated, followed by centrifugation at 13 000 g for 10 minutes at 4°C. The resulting supernatants were collected, and the protein concentrations were quantified with Bio-Rad protein assay dye (Hercules, Calif, USA) using bovine serum albumin as the standard. The samples were kept on ice for measurement of the antioxidants as described below. 2.4. Measurement of cellular antioxidants {#subsec2.4} ----------------------------------------- Cellular superoxide dismutase (SOD) activity was determined by the method of Spitz and Oberley, as described before \[[@B8], [@B9]\]. The cellular SOD activity was calculated using a concurrently run Cu,ZnSOD (Sigma-Aldrich) standard curve, and expressed as units per mg of cellular protein. Cellular NAD(P)H:quinone oxidoreductase 1 (NQO1) activity was determined using dichloroindophenol (DCIP) as the two-electron acceptor and NADPH as the electron donor in the presence or absence of dicumarol, as described before \[[@B8]\]. The dicumarol-sensitive NQO1 activity was calculated using the extinction coefficient of 21.0 mM^−1^cm^−1^, and expressed as nmol of DCIP reduced per minute per mg of cellular protein. Cellular catalase activity was determined by measuring the decomposition of H~2~O~2~ at 240 nm, according to the method of Aebi \[[@B10]\], and expressed as *μ*mol of H~2~O~2~ consumed per minute per mg of cellular protein. Cellular GSH content was determined by measuring the formation of the fluorescent conjugate from reaction of GSH with *o*-phthalaldehyde according to the procedure described previously \[[@B8]\]. Cellular GSH content was calculated using a concurrently run GSH (Sigma-Aldrich) standard curve, and expressed as nmol of GSH per mg of cellular protein. Cellular glutathione reductase (GR) activity was measured according to the method initially reported by Wheeler et al. \[[@B11]\] with modifications, as described previously \[[@B10]\]. GR activity was calculated using the extinction coefficient of 6.22 mM^−1^cm^−1^, and expressed as nmol of NADPH consumed per minute per mg of cellular protein. Cellular glutathione peroxidase (GPx) activity was determined by the method of Flohé and Günzler \[[@B12]\] with slight modifications, as described before \[[@B13]\]. This assay is based on the formation of GSSG from GPx-catalyzed oxidation of GSH by H~2~O~2~, coupled with NADPH consumption in the presence of exogenously added GR. GPx activity was calculated using the extinction coefficient of 6.22 mM^−1^cm^−1^, and expressed as nmol of NADPH consumed per minute per mg of cellular protein. 2.5. In situ real-time CL imaging {#subsec2.5} --------------------------------- As shown in [Figure 1](#fig1){ref-type="fig"}, a liquid nitrogen cooled back-illuminated charge-coupled device (CCD) camera (VersArray 2048B, Princeton Instruments, Trenton, NJ, USA) with a Nikon 50 mm f/1.2 lens was used to image the 6-well cell culture plates. The CCD sensor has a 27.6 × 27.6 mm imaging area with 2048 × 2048 pixels. It has a \>80% quantum efficiency for visible spectrum range. To capture the ultra low light image, a hardware binning of 8 × 8 was selected to increase the sensitivity by 64 times. To reduce the noise, the CCD sensor was cooled to −110°C, and a 50 kHz read out frequency was used to minimize the dark current and read out noise. The culture plates were placed on a heated plate (37°C). An enhanced aluminum front mirror was placed above the culture plates to redirect the photons to the camera. To ensure that the imaging experiment environment is totally dark, the imaging system is placed in a light tight enclosure. After placing the culture plates in the imaging system, we immediately took 6 CL images, each with ten minutes of exposure time. Before we extracted out the data, we calibrated the intensity of the CCD sensor by using dark frames to remove the background and thermal noise in a totally dark environment. Each dark frame was collected with the same integral time at the same cooling temperature in the experiment. A master dark frame was then computed by averaging 20 dark frames. For each CL image, we subtracted the master dark frame from it to get the final image. For each well in the culture plates, we summed the pixel value inside the well and divided it by the number of pixels in the well to get an average intensity for each well. 2.6. Statistical analysis {#subsec2.6} ------------------------- Data are expressed as means ±SEM from at least 3 separate experiments unless indicated otherwise. Differences between 2 groups were analyzed by Student\'s *t*-test. Statistical significance was considered at *P* \< .05. 3. RESULTS {#sec3} ========== 3.1. Detection of basal and BPQ-stimulated ROS formation by CL imaging in cultured monolayers of H9c2 cells {#subsec3.1} ----------------------------------------------------------------------------------------------------------- As shown in [Figure 2](#fig2){ref-type="fig"}, incubation of cultured monolayers of H9c2 cells with luminol/HRP led to CL responses as detected by the highly sensitive imaging system (see [Figure 1](#fig1){ref-type="fig"}), indicating that H9c2 cells in culture could constitutively release ROS. Notably, no CL responses were elicited by adding luminol/HRP to the plate wells containing PBS alone (data not shown). In addition, under our experimental conditions, viability of H9c2 cells in cultures was \>99% based on trypan blue exclusion assay (data not shown). Dramatically augmented CL responses were observed after addition of 1 *μ*M BPQ, a mitochondrial redox cycler, to the cells in culture, and the CL responses remained elevated for up to 1 hour. The intensity of the BPQ-stimulated CL responses was 4-5 times that of basal CL responses (see [Figure 2](#fig2){ref-type="fig"}). Both constitutive and BPQ-induced CL responses were completely abolished in the presence of exogenously added SOD (250 units/mL) and catalase (250 units/mL) (see [Figure 3](#fig3){ref-type="fig"}). 3.2. Detection by CL imaging of the effects of mitochondrial ETC inhibitors on ROS formation in cultured monolayers of H9c2 cells {#subsec3.2} --------------------------------------------------------------------------------------------------------------------------------- To further investigate the involvement of mitochondria in cellular ROS release, H9c2 cells in culture were exposed to two commonly used mitochondrial inhibitors, antimycin A (AA) and rotenone (ROT). As shown in [Figure 4](#fig4){ref-type="fig"}, AA treatment slightly decreased the CL responses, whereas exposure to ROT led to small increases in the CL responses. The marginal effects by either AA or ROT on the CL responses were only obvious during the first 30 minutes of treatment. From 30 to 60 minutes, the CL responses among the 3 treatment groups were essentially the same (see [Figure 4](#fig4){ref-type="fig"}). AA and ROT were dissolved in DMSO (0.1%). Control cells received 0.1% DMSO only. 3.3. Induction of cellular antioxidants by D3T and the effects on basal and BPQ-stimulated ROS formation in cultured monolayers of H9c2 cells {#subsec3.3} --------------------------------------------------------------------------------------------------------------------------------------------- As shown in [Figure 5](#fig5){ref-type="fig"}, incubation of H9c2 cells with 100 *μ*M D3T for 48 hours resulted in significant induction of cellular NQO1, CAT, GSH, GR, and GPx. These antioxidants are crucially involved in detoxification of ROS, including superoxide and H~2~O~2~ \[[@B3]\]. To determine if upregulation of cellular antioxidants by D3T could blunt ROS formation by H9c2 cells, the cardiomyocytes in culture were first treated with D3T and then exposed to BPQ. As shown in [Figure 6](#fig6){ref-type="fig"}, D3T pretreatment resulted in marked inhibition of BPQ-stimulated ROS formation. Notably, D3T pretreatment also dramatically decreased the constitutive formation of ROS by cultured H9c2 cells. The inhibitory effects of D3T pretreatment on both constitutive and BPQ-stimulated ROS formation were sustained during the 1-hour period of experiment (see [Figure 6](#fig6){ref-type="fig"}). Also as shown in [Figure 6](#fig6){ref-type="fig"}, in contrast to 1 *μ*M BPQ, 0.2 *μ*M BPQ-elicited CL responses decreased over 60 minutes in untreated cell cultures. 4. DISCUSSION {#sec4} ============= Although ROS have been extensively implicated in the pathogenesis of cardiac disorders, studies on direct in situ CL imaging of ROS formation in cultured cardiac cells are lacking in the literature. In this study, we have applied a highly sensitive CL imaging system (see [Figure 1](#fig1){ref-type="fig"}) to investigate the in situ real-time ROS formation in cultured monolayers of rat H9c2 cardiomyocytes, a widely used in vitro cell model for studying cardiac cell biology and cytoprotection \[[@B14], [@B15]\]. By using this innovative CL imaging system, we have demonstrated that significant amounts of ROS could be released from H9c2 cells in culture under unstimulated conditions (see [Figure 2](#fig2){ref-type="fig"}). While the exact cellular sources for this constitutive ROS formation in H9c2 cells remain to be elucidated, one possible site could be the mitochondrial ETC. In this regard, mitochondria account for roughly 40--50% of the total mass of cardiomyocytes. Considering the high mitochondrial activity in cardiomyocytes, we next determined if BPQ could stimulate the ROS formation from H9c2 cells in culture. Indeed, dramatically augmented CL responses were observed after adding 1 *μ*M, and even 0.2 *μ*M BPQ to the cultured monolayers of H9c2 cells (see Figures [2](#fig2){ref-type="fig"} and [6](#fig6){ref-type="fig"}), suggesting that BPQ is a potent ROS generator in cardiomyocytes. It remains unclear why the CL responses elicited by 0.2 *μ*M BPQ decreased over 60 minutes in untreated cell cultures (see [Figure 6](#fig6){ref-type="fig"}). It might be due to the significant detoxification of the small amount of BPQ (0.4 nmol/2.8 × 10^5^ cells) in H9c2 cells, leading to decreased availability of the free BPQ molecules that undergo redox cycling to produce ROS. BPQ is a quinone metabolite derived from benzo(a)pyrene, an environmental pollutant implicated in cardiovascular diseases \[[@B16], [@B17]\]. Benzo(a)pyrene is also present in cigarette smoking, which is a major risk factor for human cardiovascular diseases \[[@B16], [@B18]\]. Metabolism of benzo(a)pyrene by mammalian tissues, including cardiovascular tissue, can form significant amounts of BPQ \[[@B19]\]. We have previously demonstrated that BPQ preferentially undergoes redox cycling in mitochondria, leading to ROS formation \[[@B6], unpublished observation\]. Thus, the mitochondrial ETC could also be the site where BPQ underwent redox cycling to generate ROS in the cultured monolayers of H9c2 cells. To further investigate the involvement of mitochondrial ETC in ROS formation by H9c2 cells, AA and ROT were used to selectively block the electron transport at complexes I and III, respectively \[[@B20]\]. The marginal effects of either AA or ROS on ROS formation by H9c2 cells (see [Figure 4](#fig4){ref-type="fig"}) suggested that blocking the mitochondrial ETC was not an effective mechanism for altering ROS formation by H9c2 cells. The redox cycling of BPQ thus appears to be a much more effective pathway for augmentation of ROS formation by cardiomyocytes. This is an important observation considering the involvement of benzo(a)pyrene in cardiovascular diseases. The potent and sustained stimulation of ROS formation by BPQ in cardiomyocytes would create significant oxidative stress in myocardium. In this context, augmented ROS formation and the subsequent oxidative cell injury in myocardium are crucial events underlying cardiac disorders \[[@B1], [@B2]\]. Likewise, inhibition or scavenging of ROS in myocardium has been shown to be an effective strategy for intervention of oxidative cardiac injury \[[@B21], [@B22]\]. In this study, we also demonstrated that pretreatment with D3T significantly upregulated cellular antioxidants and markedly inhibited both constitutive and BPQ-stimulated ROS formation in cultured monolayers of H9c2 cells (see [Figure 6](#fig6){ref-type="fig"}). D3T is a cruciferous sulfur-containing compound, which potently induces antioxidants in mammalian cells/tissues \[[@B23]\]. In H9c2 cells, D3T potently upregulated NQO1, CAT, GSH, GR, and GPX (see [Figure 5](#fig5){ref-type="fig"}). NOQ1 is recently found to scavenge superoxide \[[@B24]\]. CAT, GSH, GR, and GPx are key cellular factors primarily for detoxification of H~2~O~2~ \[[@B3]\]. As illustrated in [Figure 5(b)](#fig5){ref-type="fig"}, coordinated induction of the above cellular antioxidants in H9c2 cells would enhance detoxification of both superoxide and H~2~O~2~, and thereby leading to the decreases in luminol/HRP-amplified CL responses. This is the first study demonstrating that upregulation of cellular antioxidants by D3T was accompanied by markedly decreased oxidative stress induced by BPQ in cultured cardiomyocytes. The above finding also implicated D3T as a promising chemoprotective nutraceutical for intervention of oxidative cardiac injury associated with exposure to benzo(a)pyrene as well as other pollutants that give rise to redox-cycling quinones. In this regard, consumption of cruciferous vegetables rich in organosulfur compounds, including dithiolethiones, is associated with decreased incidence of cardiovascular events \[[@B25], [@B26]\]. In summary, by using the highly sensitive CL imaging system, we demonstrate that cultured monolayers of H9c2 cardiomyocytes constitutively release ROS under physiological conditions, and the redox cycler BPQ potently stimulates the ROS formation in the cultured cardiomyocytes. With this innovative imaging system, we also demonstrate that upregulation of cellular antioxidants by cruciferous D3T appears to be a highly effective strategy for mitigating oxidative stress induced by BPQ in the cultured cardiomyocytes. These observations lay a basis for investigation of myocardial oxidative stress as well as cardioprotection in animal models. Although most of the data presented in this work were from 2 independent measurements, our ongoing research on luminol/HRP-based CL imaging further confirms the ability of this innovative technique to reproducibly detect ROS in both cellular systems and in vivo animals. This work was supported by NIH R01HL71190 and a grant from Harvey Peters Research Center Foundation to Y. Li. Dr. G. Wang was partially supported by the NIH under Grants EB001685 and CA127189. AA: : Antimycin A BPQ: : Benzo(a)pyrene-1,6-quinone CAT: : Catalase CL: : Chemiluminescence D3T: : 3*H*-1,2-dithiole-3-thione DCIP: : Dichloroindophenol EPR: : Electron paramagnetic resonance ETC: : Electron transport chain FBS: : Fetal bovine serum GPx: : Glutathione peroxidase GR: : Glutathione reductase GSH: : Reduced form of glutathione H~2~O~2~: : Hydrogen peroxide HRP: : Horseradish peroxidase NQO1: : NAD(P)H:quinone oxidoreductase 1 PBS: : Phosphate buffered saline ROS: : Reactive oxygen species ROT: : Rotenone SOD: : Superoxide dismutase. {#fig1} {#fig2} {ref-type="fig"} except that SOD (250 units/mL) and CAT (250 units/mL) were added to the top 3 wells. (c) Shows the quantitative data of the images in (a).](IJBI2008-941729.003){#fig3} {#fig4} {#fig5} {#fig6} [^1]: Recommended by Guowei Wei

All relevant data are available as Supporting Information files. All data are also archived alongside the full suite of code used to perform analyses and generate plots, at <https://zenodo.org/badge/latestdoi/160883450>. Introduction {#sec001} ============ Childhood influenza exposures leave an immunological imprint, which has reverberating, lifelong impacts on immune memory. Foundational work on original antigenic sin \[[@ppat.1008109.ref001]\] and antigenic seniority \[[@ppat.1008109.ref002]\] shows that individuals maintain the highest antibody titers against influenza strains encountered in childhood. But how these serological patterns map to functional immune protection, and shape birth year-specific risk during outbreaks, remains an active area of inquiry. One open question is the breadth of cross-protection provided by immune memory imprinted in childhood. We define immune imprinting as a lifelong bias in immune memory of, and protection against, the strains encountered in childhood. Such biases most likely become entrenched as subsequent exposures back-boost existing memory responses, rather than stimulating de novo responses \[[@ppat.1008109.ref003]\]. By providing particularly robust protection against certain antigenic subtypes, or clades, imprinting can provide immunological benefits, but perhaps at the cost of equally strong protection against variants encountered later in life. For example, every modern influenza pandemic has spared certain birth cohorts, presumably due to cross-protective memory primed in childhood \[[@ppat.1008109.ref004]--[@ppat.1008109.ref010]\]. Recently, we showed that imprinting also protects against novel, emerging avian influenza viruses of the same phylogenetic group of hemagglutinin (HA) as the first childhood exposure \[[@ppat.1008109.ref009],[@ppat.1008109.ref011]\]. Imprinting may additionally shape birth year-specific risk from seasonal influenza \[[@ppat.1008109.ref012]--[@ppat.1008109.ref014]\], but the importance of broadly protective immunity in this context is still being evaluated \[[@ppat.1008109.ref015]--[@ppat.1008109.ref017]\]. Until recently, narrow cross-protective immunity specific to variants of a single HA subtype has been considered the primary mode of defense against seasonal influenza. Lymphocyte memory of variable epitopes on the HA head (i.e. sites at which hemagglutinin antigens of different subtypes show limited homology) drives this narrow, within-subtype protection, which is the main mechanism of protection from the inactivated influenza vaccine. But a growing body of evidence shows protection may also be driven by memory of other influenza antigens (e.g. neuraminidase, NA) \[[@ppat.1008109.ref018]--[@ppat.1008109.ref020]\], or by immune response to conserved epitopes, many of which are found on the HA stalk \[[@ppat.1008109.ref011],[@ppat.1008109.ref015],[@ppat.1008109.ref021]--[@ppat.1008109.ref023]\]. Antibodies that target conserved HA epitopes can provide broad protection across multiple HA subtypes in the same phylogenetic group \[[@ppat.1008109.ref021],[@ppat.1008109.ref023],[@ppat.1008109.ref024]\], where HA group 1 contains hemagglutinin subtypes H1 and H2, while group 2 contains H3. Only three HA subtypes have circulated seasonally in humans since 1918, H1, H2 and H3; H1 and H2 belong to phylogenetic group 1, while H3 is in group 2 \[[@ppat.1008109.ref011],[@ppat.1008109.ref022],[@ppat.1008109.ref025]\]. Current thinking stipulates that within a single host, imprinting probably induces multiple levels of bias in immune memory, to both conserved (broadly protective) and variable (narrowly protective) sites on various influenza antigens. The functional role of any single layer of imprinted immune memory depends on both immunodominance hierarchies and epidemic context. Here, we examine which layers of imprinted memory impact risk from seasonal influenza. Different breadths of immunity are expected to act differently on influenza epidemiology. Within-subtype immunity to HA is known to shape seasonal influenza's epidemiology and evolution \[[@ppat.1008109.ref026]\], but due to rapid decay in the face of antigenic drift, it would not be expected to shape cohort-specific imprinting protection across an entire human lifetime \[[@ppat.1008109.ref027],[@ppat.1008109.ref028]\]. Conversely, broad, HA group-level immune memory arises when lymphocytes target conserved HA epitopes. Responses to these conserved epitopes should be more stable over time, and can play a strong role in defense against unfamiliar influenza strains (e.g. novel, avian or pandemic subtypes \[[@ppat.1008109.ref011],[@ppat.1008109.ref021],[@ppat.1008109.ref023],[@ppat.1008109.ref024],[@ppat.1008109.ref029],[@ppat.1008109.ref030]\], but are not traditionally though to act strongly against familiar, seasonal influenza subtypes. However, responses to the conserved HA stem have recently been identified as an independent correlate of protection against seasonal influenza \[[@ppat.1008109.ref015]\], and might play a particularly strong role against drifted seasonal strains whose variable HA epitopes have become unrecognizable. Thus, childhood immune imprinting may determine which birth cohorts are primed for effective defense against seasonal strains with conserved HA epitopes characteristic of group 1 or group 2, or with variable HA epitopes characteristic of a particular subtype (H1, H2, etc.). A similar line of reasoning may apply to immunity against NA, although much less attention has been paid to this antigen. Since 1977, two distinct subtypes of influenza A, H1N1 and H3N2, have circulated seasonally in humans, with striking but poorly understood differences in their age-specific impact \[[@ppat.1008109.ref009],[@ppat.1008109.ref012]--[@ppat.1008109.ref014],[@ppat.1008109.ref031]\]. These differences could be associated with childhood imprinting: older cohorts were almost certainly exposed to H1N1 in childhood (since it was the only subtype circulating in humans from 1918--1957), and now seem to be preferentially protected against modern seasonal H1N1 variants \[[@ppat.1008109.ref009],[@ppat.1008109.ref012]--[@ppat.1008109.ref014]\]. Likewise, younger adults have the highest probabilities of childhood imprinting to H3N2 (**[Fig 1](#ppat.1008109.g001){ref-type="fig"}**), which is consistent with relatively low numbers of clinically attended H3N2 cases in these cohorts. Alternatively, differences in the evolutionary dynamics of H1N1 and H3N2 could explain the observed age profiles. Subtype H3N2 exhibits slightly faster drift in its antigenic phenotype than H1N1, and as a result, H3N2 may be better able to escape pre-existing immunity in immunologically experienced adults, whereas H1N1 may be relatively restricted to causing disease in immunologically inexperienced children \[[@ppat.1008109.ref032]\]. {#ppat.1008109.g001} We analyzed a large surveillance data set of relatively severe, clinically attended influenza cases to test whether cohort effects from childhood imprinting primarily act against variable epitopes, only providing narrow cross-protection against closely related HA or NA variants of the same subtype, or against more conserved epitopes, providing broad cross-protection across HA subtypes in the same phylogenetic group (**[Fig 1A and 1B](#ppat.1008109.g001){ref-type="fig"}**). We fitted a suite of models to data using maximum likelihood and compared models using AIC. In a separate analysis, we considered the hypothesis that differences in evolutionary rate of H1N1 and H3N2, rather than imprinting effects, shape differences in age distribution. Our results have implications for long-term projections of seasonal influenza risk in elderly cohorts \[[@ppat.1008109.ref013]\], who suffer the heaviest burdens of influenza-related morbidity and mortality, and whose imprinting status will shift through time as cohorts born during different inter-pandemic eras grow older. Results {#sec002} ======= Data {#sec003} ---- The Arizona Department of Health Services (ADHS) provided a dataset containing 9,510 seasonal H1N1 and H3N2 cases from their statewide surveillance system (**[S1 Data](#ppat.1008109.s004){ref-type="supplementary-material"}**). Cases of all ages were confirmed to subtype by PCR and/or culture, primarily from virologic testing at the Arizona State Public Health Laboratory (ASPHL). The ADHS surveillance system aims to characterize circulating strains from patients across the state with medically attended influenza. Although surveillance does not target specific at-risk groups, relatively severe cases (especially those tested at hospital labs) are overrepresented in our data, as these cases are most likely to be medically attended and confirmed to subtype, thus meeting our study inclusion criteria. This is because confirmation to subtype requires a second-line test (PCR or culture); rapid tests are more common, but do not indicate subtype. Although the exact collection setting of individual specimens was not always recorded, ADHS staff internally reviewed reporting source and provider organization of cases in our data to estimate that roughly 76% were reported and/or submitted by hospital labs or may have originated at hospital-associated outpatient clinics. In 2016, an ADHS analysis matched PCR-confirmed cases to hospital discharge data, and found that nearly half of the cases reported from hospital labs were severe enough to warrant hospital admission (if extrapolated to other seasons, roughly 38% of the overall data). The rest (also roughly 38% of the overall data, if extrapolated) were discharged without admission. To obtain a broader representation of clinically attended cases from across the state, ADHS collaborates with county health departments, commercial laboratories, and outpatient clinics to receive specimens. We estimate that roughly 8% of the overall data originated in outpatient settings. The remaining 17% of cases were either tested at commercial labs, or were tested at ASPHL, but with unknown origin. Ultimately, these data allow us to analyze drivers of relatively severe, clinically attended cases, but our results cannot be assumed to generalize to mild or asymptomatic cases. Cases were observed across 22 years of influenza surveillance, from the 1993--1994 influenza season through the 2014--2015 season, although sample sizes increased dramatically after the 2009 pandemic (**[Table 1](#ppat.1008109.t001){ref-type="table"}**). Sampling changed slightly starting in 2004, when commercial labs were first mandated to report positive tests to the state \[[@ppat.1008109.ref033]\], but the vast majority of cases analyzed (9150/9451) were observed from the 2004--2005 season onwards, after this change had been implemented. 10.1371/journal.ppat.1008109.t001 ###### PCR-confirmed cases. {#ppat.1008109.t001g} Season Confirmed H1N1 Confirmed H3N2 ----------- ---------------- ---------------- 1993--94 0 101 1994--95 12 38 2002--03 71 8 2003--04 0 71 2004--05 0 131 2005--06 1 321 2006--07 212 28 2007--08 196 244 2010--11 472 1204 2011--12 595 348 2012--13 80 1578 2013--14 1475 151 2014--15 5 2109 **Total** **3119** **6332** Data representing the first and second waves of the 2009 H1N1 pandemic (2008--2009 and 2009--2010 seasons) were excluded. Following CDC standards, ADHS defines the influenza season as epidemiological week 40 (around early October) through week 39 of the following year \[[@ppat.1008109.ref034]\]. The 2008--2009 and 2009--2010 influenza seasons spanned the first and second wave, respectively, of the 2009 H1N1 pandemic. We did not analyze cases observed during this time period, because age distributions of cases and immune memory differed during the 2009 pandemic from the normal drivers of seasonal influenza's immuno-epidemiology of interest to this study \[[@ppat.1008109.ref014],[@ppat.1008109.ref021],[@ppat.1008109.ref024]\]. From the dataset of 9,510 seasonal cases (defined as any case observed outside the 2008--2009 or 2009--2010 season), we excluded 58 cases with birth years before 1918 (whose imprinting status could not be inferred unambiguously), and one case whose year of birth was recorded in error. Ultimately, we analyzed 9,541 cases. The model {#sec004} --------- ### Reconstructed imprinting patterns {#sec005} We reconstructed birth year-specific probabilities of childhood imprinting to H1N1, H2N2 or H3N2 using methods described previously \[[@ppat.1008109.ref011]\]. These probabilities are based on patterns of first childhood exposure to influenza A and reflect historical circulation (**[Fig 1A](#ppat.1008109.g001){ref-type="fig"}**). Most individuals born between pandemics in 1918 and 1957 experienced a first influenza A virus (IAV) infection by H1N1, and middle-aged cohorts born between pandemics in 1957 and 1968 almost all were first infected by H2N2 (note that because the first influenza exposure may occur after the first year of life, individuals born in the years leading up to a pandemic have some probability of first infection by the new pandemic subtype, **[Fig 1A](#ppat.1008109.g001){ref-type="fig"}**). Ever since its emergence in 1968, H3N2 has dominated seasonal circulation in humans, and caused the majority of first infections in younger cohorts. However, H1N1 has also caused some seasonal circulation since 1977, and has imprinted a fraction of all cohorts born since the mid-1970s (**[Fig 1A](#ppat.1008109.g001){ref-type="fig"}**). Reconstructions assumed children age 0--12 in the year of case observation might not yet have been exposed to any influenza virus. Interactions between imprinting and vaccination of naïve infants are plausible, but poorly understood \[[@ppat.1008109.ref011],[@ppat.1008109.ref035]\]. We did not consider childhood vaccination effects here; only a small percentage of individuals in the ADHS data were born at a time when healthy infants were routinely vaccinated against influenza. ### Expected age distributions under alternate imprinting models {#sec006} If HA subtype-level imprinting protection shapes seasonal influenza risk, primary exposure to HA subtype H1 or H3 in childhood should provide lifelong protection against modern variants of the same HA subtype. If imprinting protection acts primarily against specific NA subtypes, lifelong protection will be specific to N1 or to N2 (**[Fig 1B](#ppat.1008109.g001){ref-type="fig"}**). Alternatively, if broad HA group-level imprinting shapes seasonal influenza risk, then cohorts imprinted to HA subtype H1 or H2 (both group 1) should be protected against modern, seasonal H1N1 (also group 1), while only cohorts imprinted to H3 (group 2) would be protected against modern, seasonal H3N2 (also group 2) (**[Fig 1B](#ppat.1008109.g001){ref-type="fig"}**). Collinearities between the predictions of different imprinting models (**[Fig 1D--1I](#ppat.1008109.g001){ref-type="fig"}**) were inevitable, given the limited diversity of influenza antigenic subtypes circulating in humans over the past century (reflected in **[Fig 1A](#ppat.1008109.g001){ref-type="fig"}**). Note that middle-aged cohorts, which were first infected by H2N2, are crucial, because they provide the only leverage to differentiate between imprinting at the HA subtype, NA subtype or HA group-level level (**[Fig 1B](#ppat.1008109.g001){ref-type="fig"}**). Our approach distinguishes between age-specific risk factors related to health and social behavior, and birth year-specific effects related to imprinting. Specifically, age-specific risk could be influenced by medical factors like age-specific vaccine coverage, age-specific risk of severe disease, age-related changes in endocrinology and immunosenescence, or by behavioral factors like age-assorted social mixing, and age-specific healthcare seeking behavior. These factors should have similar impacts on any influenza subtype. In contrast, imprinting effects are subtype- (or group-) specific. Thus, we fit a step function to characterize the shape of age-specific risk of any confirmed influenza case. Simultaneously, we modeled residual, subtype-specific differences in risk as a function of birth year, to focus on the possible role of childhood imprinting. Each tested model used a linear combination of age-specific risk (**[Fig 1C](#ppat.1008109.g001){ref-type="fig"}**) and birth year-specific risk (**[Fig 1D--1F](#ppat.1008109.g001){ref-type="fig"}**) to generate an expected distribution of H1N1 or H3N2 cases (**[Fig 1G--1I](#ppat.1008109.g001){ref-type="fig"}**). Note that for a given birth cohort, age-specific risk changed across progressive years of case observation (as the cohort got older), whereas birth year-specific risk was constant over time. To test quantitatively whether observed subtype-specific differences in incidence were most consistent with imprinting at the HA subtype, NA subtype or HA group level, or with no contribution of imprinting, we fitted a suite of models to each data set using a multinomial likelihood and then performed model selection using AIC. AIC is used to compare the relative strength of statistical support for a set of candidate models, each fitted to the same data, and favors parsimonious models that fit the data well \[[@ppat.1008109.ref036],[@ppat.1008109.ref037]\]. Technical details are provided in the **Methods**. ### Tested models {#sec007} We fit a set of four models to the ADHS data set. The simplest model contained only age-specific risk, and more complex models added effects from imprinting at the HA subtype level, at the HA group level, or at the NA subtype level. The age-specific risk curve took the form of a step function, in which relative risk was fixed to 1 in age bin 0--4, and one free parameter was fit to represent relative risk in each of the following 12 age bins: {5--10, 11--17, 18--24, 25--31, 32--38, 39--45, 46--52, 53--59, 60--66, 67--73, 74--80, 81+}. Within models that contained imprinting effects, the fraction of individuals in each single year of birth with protective childhood imprinting was assumed proportional to reductions in risk. Two additional free parameters quantified the relative risk of a confirmed H1N1 or H3N2 case, given imprinting protection against that seasonal subtype. ### Effect of influenza evolutionary rate on age profiles {#sec008} We used publicly available data from Nextstrain \[[@ppat.1008109.ref038],[@ppat.1008109.ref039]\], and from one previously published study \[[@ppat.1008109.ref040]\], to calculate annual antigenic advance, which we defined as the mean antigenic distance between strains of a given lineage (pre-2009 H1N1, post-2009 H1N1 or H3N2) that circulated in consecutive seasons (**Methods**). The "antigenic distance" between two influenza strains is used as a proxy for similarity in antigenic phenotype, and potential for immune cross-protection. A variety of methods have been developed to estimate antigenic distance using serological data, genetic data, or both \[[@ppat.1008109.ref039]--[@ppat.1008109.ref041]\]. To assess the impact of antigenic evolution on the epidemic age distribution, we tested whether the proportion of cases in children increased in seasons associated with large antigenic changes. If the rate of antigenic drift is a strong driver of age-specific influenza risk, then the fraction of influenza cases observed in children should be negatively related to annual antigenic advance \[[@ppat.1008109.ref032]\]. In other words, strains that have not changed much antigenically since the previous season should be unable to escape pre-existing immunity in immunologically experienced adults, and more restricted to causing cases in immunologically inexperienced children; strains that have changed substantially will be less restricted to children. Subtype-specific differences in age distribution {#sec009} ------------------------------------------------ Seasonal H3N2 epidemics consistently caused more clinically attended cases in older cohorts, while H1N1 caused a greater proportion of cases in young and middle-aged adults (**Figs [2](#ppat.1008109.g002){ref-type="fig"}, [S1](#ppat.1008109.s001){ref-type="supplementary-material"} and [S2](#ppat.1008109.s002){ref-type="supplementary-material"}**). These patterns were apparent whether we compared H3N2 epidemic age distributions with those caused by the pre-2009 seasonal H1N1 lineage, or with the post-2009 lineage. Observed patterns are consistent with the predicted effects of cohort-specific imprinting (**[Fig 1](#ppat.1008109.g001){ref-type="fig"}**), and with previously reported differences in age distribution of seasonal H1N1 and H3N2 incidence \[[@ppat.1008109.ref012]--[@ppat.1008109.ref014],[@ppat.1008109.ref031]\]. See **[Fig 2](#ppat.1008109.g002){ref-type="fig"}** for seasons where H1N1 and H3N2 co-circulated in substantial numbers, and **[S1](#ppat.1008109.s001){ref-type="supplementary-material"} and [S2](#ppat.1008109.s002){ref-type="supplementary-material"} Figs.** for the entire dataset and alternate smoothing parameters. {ref-type="supplementary-material"}. for all seasons).](ppat.1008109.g002){#ppat.1008109.g002} Imprinting model selection {#sec010} -------------------------- The data showed a strong preference for NA subtype-level imprinting over HA subtype-level imprinting (ΔAIC = 34.54), and effectively no statistical support for broad, HA group-level imprinting (ΔAIC = 249.06), or for an absence of imprinting effects (ΔAIC = 385.42) **([Fig 3](#ppat.1008109.g003){ref-type="fig"}, [Table 2](#ppat.1008109.t002){ref-type="table"}**). Visual assessment of model fits (**[Fig 3C and 3D](#ppat.1008109.g003){ref-type="fig"}**) confirmed that models containing imprinting effects at the narrow, NA or HA subtype levels provided the best fits to data. The lack of support for the no-imprinting model suggests that imprinting is important, and shapes lifelong seasonal influenza risk, just as it does avian-origin influenza (10, 12). However, imprinting appears to act more narrowly against seasonal influenza than against avian influenza, providing cross protection only to a specific NA or HA subtype, instead of broader, HA group-level protection. This result is consistent with the idea that immunodominance of variable HA epitopes limits the breadth of immune cross protection deployed against familiar, seasonal influenza subtypes \[[@ppat.1008109.ref023],[@ppat.1008109.ref024]\]. {#ppat.1008109.g003} 10.1371/journal.ppat.1008109.t002 ###### Maximum likelihood parameter estimates and 95% profile confidence intervals. {#ppat.1008109.t002g} Model NA subtype-level imprinting HA subtype-level imprinting HA group-level imprinting No imprinting ----------------------- ----------------------------------- ----------------------------- --------------------------- ------------------- ΔAIC 0.00 34.54 249.06 385.42 H1N1 impr. protection 0.36 (0.30--0.44) 0.29 (0.24--0.35) 0.65 (0.56--0.76) H3N2 impr. protection 0.66 (0.58--0.76) 0.90 (0.78--1.04) 0.70 (0.62--0.82) Ages 0--4 Reference group: Value fixed to 1 Ages 5--10 0.67 (0.62--0.73) 0.65 (0.60--0.70) 0.65 (0.60--0.71) 0.61 (0.56--0.66) Ages 11--17 0.33 (0.30--0.37) 0.30 (0.28--0.34) 0.32 (0.30--0.36) 0.29 (0.27--0.33) Ages 18--24 0.37 (0.34--0.42) 0.34 (0.32--0.38) 0.37 (0.34--0.42) 0.34 (0.31--0.38) Ages 25--31 0.35 (0.32--0.40) 0.33 (0.30--0.38) 0.34 (0.32--0.38) 0.32 (0.29--0.36) Ages 32--38 0.3 (0.28--0.35) 0.28 (0.26--0.32) 0.3 (0.27--0.34) 0.27 (0.25--0.31) Ages 39--45 0.25 (0.22--0.30) 0.22 (0.20--0.26) 0.25 (0.22--0.29) 0.23 (0.21--0.26) Ages 46--52 0.27 (0.24--0.30) 0.22 (0.20--0.26) 0.26 (0.23--0.29) 0.24 (0.22--0.28) Ages 53--59 0.25 (0.23--0.30) 0.22 (0.20--0.26) 0.23 (0.21--0.27) 0.23 (0.20--0.26) Ages 60--66 0.27 (0.24--0.30) 0.29 (0.26--0.33) 0.24 (0.22--0.28) 0.23 (0.21--0.27) Ages 67--73 0.37 (0.33--0.43) 0.42 (0.37--0.48) 0.34 (0.30--0.38) 0.33 (0.30--0.38) Ages 74--80 0.57 (0.50--0.64) 0.64 (0.57--0.74) 0.52 (0.46--0.59) 0.5 (0.46--0.57) Ages 81+ 0.99 (0.88--1.11) 1.12 (1.00--1.26) 0.9 (0.81--1.01) 0.87 (0.80--0.96) All estimated parameters represent the relative risk of a confirmed case, given the factors listed in the left-hand column. Age-specific risk parameters could take any positive value. Imprinting parameters could take values in \[0,1\], consistent with reductions in risk from the imprinted subtype or group. All tested models included age-specific risk and demographic age distribution. As expected (see **[Fig 1G--1I](#ppat.1008109.g001){ref-type="fig"}**), predictions from the two best models were highly collinear, except in their risk predictions among middle-aged, H2N2-imprinted cohorts (birth years 1957--1968), and some other minor differences arising from normalization across birth-years. Fitted risk patterns {#sec011} -------------------- Fitted age-specific risk curves took similar forms in all tested models. After controlling for demographic age distribution, estimated age-specific risk was highest in children and the elderly, consistent with the buildup of immune memory across childhood, and waning immune function in the aged (**[Fig 3A](#ppat.1008109.g003){ref-type="fig"}** shows the fitted curve from the best model). Estimates of imprinting parameters were less than one, indicating some reduction in relative risk (**[Table 2](#ppat.1008109.t002){ref-type="table"}**). Within the best model, estimated reductions in relative risk from childhood imprinting were stronger for H1N1 (0.34, 95% CI 0.29--0.42) than for H3N2 (0.71, 95% CI 0.62--0.82). In the second-best model, HA subtype-specific imprinting, estimated reductions in H3N2 risk were particularly weak, and the confidence interval overlapped the null value of 1. **[Table 2](#ppat.1008109.t002){ref-type="table"}** shows parameter estimates and 95% profile confidence intervals from all models fitted. Effect of evolutionary rate {#sec012} --------------------------- To test for effects of evolutionary rate on epidemic age distribution, we searched for decreases in the proportion of cases among children in seasons associated with antigenic novelty, when highly drifted strains might be more able to infect immunologically experienced adults. We defined children as ages 0--10, and verified internally that our analysis of evolutionary rate was insensitive to our exact choice of age range for children. Consistent with this expectation, the data showed a slight negative but not significant association between annual antigenic advance and the fraction of H3N2 cases observed in children (**[Fig 4A](#ppat.1008109.g004){ref-type="fig"}**). However, note that no clear relationship emerged between antigenic novelty and the fraction of cases observed in older children (\>10) and adults (**[Fig 4A](#ppat.1008109.g004){ref-type="fig"}**). These are the cohorts in which epidemiological data show the clearest differences between H1N1 and H3N2's age-specific impacts (**[Fig 2](#ppat.1008109.g002){ref-type="fig"}**); if rate of antigenic evolution is a dominant driver of age-specific differences in incidence, we would have expected to see clearer evidence of evolutionary rate effects among adults cohorts, not just between adults and the youngest children. The data contained too few influenza seasons with sufficient numbers of confirmed H1N1 cases to support meaningful Spearman correlation coefficients for either pre-2009 or post-2009 seasonal H1N1 lineages. {#ppat.1008109.g004} Furthermore, if evolutionary rate is the dominant driver of subtype-specific differences in epidemic age distribution, then when subtypes H1N1 and H3N2 show similar degrees of annual antigenic advance, their age distributions of cases should appear more similar. However, the data showed that differences in H1N1 and H3N2's age-specific impacts did not converge when lineages showed similar annual advance. When comparing the fraction of cases observed in specific age classes, H1N1 data consistently clustered separately from H3N2, with H1N1 consistently causing fewer cases at the extremes of age (children 0--10 and elderly adults 71--85), but more cases in middle-aged adults, regardless of antigenic novelty (**[Fig 4A](#ppat.1008109.g004){ref-type="fig"}**). Smoothed density plots showed no clear relationship between annual antigenic advance and age distribution (**[Fig 4B](#ppat.1008109.g004){ref-type="fig"}**). Overall, the data showed a weak, but not significant signal that relatively severe, clinically attended cases may be more restricted to young children when antigenic novelty is low, but the data did not show strong evidence that the magnitude of annual antigenic drift is a systematic driver of epidemic age distribution across the entire population. Discussion {#sec013} ========== We analyzed a large epidemiological surveillance dataset and found that seasonal influenza subtypes H1N1 and H3N2 cause different age distributions of relatively severe, clinically attended cases, confirming previously reported patterns \[[@ppat.1008109.ref012]--[@ppat.1008109.ref014]\]. We analyzed several possible drivers of these differences systematically, and found the greatest support for imprinting protection against seasonal influenza viruses of the same NA or HA subtype as the first influenza strain encountered in childhood \[[@ppat.1008109.ref012],[@ppat.1008109.ref013]\]. The data did not support strong effects from broader HA group-level imprinting, as recently detected for novel zoonotic or pandemic viruses \[[@ppat.1008109.ref009],[@ppat.1008109.ref011]\], or from differences in rates of antigenic evolution \[[@ppat.1008109.ref032]\]. Our results suggest individuals retain a lifelong bias in immune memory, and that this imprint is not erased even after decades of exposure to or vaccination against dissimilar influenza subtypes. External evidence corroborates the idea that birth year, rather than age, drives subtype-specific differences in seasonal influenza risk. When H3N2 first emerged in 1968, it caused little or no excess mortality in the elderly, who had putatively been exposed, as children or young adults, to an H3 virus that had circulated in the late 1800s \[[@ppat.1008109.ref007],[@ppat.1008109.ref009]\]. Meanwhile, H1N1-imprinted cohorts (those \~10--50 years old at the time) experienced considerable excess mortality in the 1968 pandemic \[[@ppat.1008109.ref007]\]. Now, fifty years later, the same H1N1-imprinted cohorts continue to experience excess H3N2 morbidity and mortality as older adults \[[@ppat.1008109.ref012]--[@ppat.1008109.ref014],[@ppat.1008109.ref031]\] (**[Fig 2](#ppat.1008109.g002){ref-type="fig"}**). In model comparison, the data supported childhood imprinting to NA at the subtype level. Although NA is not as intensively studied as HA, these results emphasize the increasingly recognized importance of both antigens as drivers of protection against seasonal influenza \[[@ppat.1008109.ref018]--[@ppat.1008109.ref020]\]. Realistically, some combination of effects from both HA and NA subtype-level imprinting probably shapes seasonal influenza risk; both models of imprinting produced similar fits to data, and far outperformed other models in terms of AIC ([Fig 3](#ppat.1008109.g003){ref-type="fig"}). Unfortunately, due to the limited diversity of seasonal influenza subtypes that have circulated in humans over the past century, collinearities between even the relatively simple models tested here prevented us from testing more complicated models of combined effects from imprinting to multiple antigens. Deeper insights into the respective roles of HA and NA will most likely need to come from focused immunological cohort studies, in which individual histories of influenza infection are recorded and can be studied alongside changes in serology, PBMCs, and/or the B cell repertoire \[[@ppat.1008109.ref035]\]. Alternatively, the development of immunological biomarkers for diagnosis of imprinting status in individual patients could substantially increase the power of epidemiological inference. We did not detect a clear relationship between annual antigenic advance and epidemic age distribution, although small sample sizes may have limited our statistical power. We did detect a weak trend, consistent with the idea that influenza cases are more restricted to immunologically inexperienced children in seasons of low antigenic advance, as previously proposed \[[@ppat.1008109.ref032]\]. But the data did not reveal a clear relationship between antigenic advance and the fraction of cases occurring in adult age groups, where epidemiological data reveal distinct subtype-specific differences in impact. Perhaps antigenic advance shapes how cases are distributed between children and adults, but has small or inconsistent impacts within the adult population. We speculate that clearer relationships between antigenic advance and epidemic age distribution might emerge if methods to estimate antigenic distance were able to incorporate effects such as immune history \[[@ppat.1008109.ref042]\], glycosylation \[[@ppat.1008109.ref042],[@ppat.1008109.ref043]\], and immunity to antigens other than HA \[[@ppat.1008109.ref019],[@ppat.1008109.ref020],[@ppat.1008109.ref044]\]. The exact immunological drivers of imprinting protection against seasonal influenza remain unclear, but our results provide some new clues. Traditionally, within-subtype cross-protection is thought to decay quickly with antigenic drift. Strains that circulated more than 14 years apart rarely show measurable cross-protective titers by the hemagglutination inhibition (HI) assay \[[@ppat.1008109.ref040]\]. The short timescale of immune memory to variable HA head epitopes stands in contrast to patterns observed in our study and others \[[@ppat.1008109.ref012]--[@ppat.1008109.ref014]\], where within-subtype immune memory imprinted in childhood appears to persist for an entire human lifetime, remaining evident even in the oldest cohorts. We speculate that within-subtype imprinting protection may involve epitopes that are more conserved, and stable over time, than those typically measured in HI assays. These inform most existing estimates of antigenic distance, but disproportionately measure antibodies to variable, immunodominant epitopes on the HA head \[[@ppat.1008109.ref015],[@ppat.1008109.ref024]\]. Across a lifetime of exposures to diverse H1N1 and H3N2 variants, repeated back-boosting of antibodies to intermediately conserved sites on HA or NA (i.e. sites conserved within but not across HA and NA subtypes), could explain the longevity of subtype-level imprinting protection. This is consistent with recent evidence that the immune repertoire shifts to focus on more conserved influenza epitopes as we age \[[@ppat.1008109.ref027],[@ppat.1008109.ref028]\]. Another possibility is that memory B cell clones developed during the first childhood influenza infection may later adapt via somatic hypermutation to follow antigenic targets as they drift over time. However, this would be inconsistent with new evidence suggesting memory B cells are relatively fixed in phenotype, and have little potential for ongoing affinity maturation \[[@ppat.1008109.ref045],[@ppat.1008109.ref046]\], or that somatic hypermutation decreases with age \[[@ppat.1008109.ref027]\]. Finally, the role of CD4+ T cells in imprinting is unclear, but T cell memory and T cell help to B cells within germinal centers both play at least some role in the development of the immune repertoire \[[@ppat.1008109.ref047]\]. Signals of imprinting protection are anomalously strong in the current cohort of elderly adults, as reflected by higher estimates of imprinting protection to H1N1 than H3N2. The oldest subjects in our data, born slightly after 1918, and would not have encountered an influenza virus of any subtype other than H1N1 until roughly age 30. Repeated early-life exposures to diverse H1N1 variants may have reinforced and expanded the breadth of H1N1-specific immune memory \[[@ppat.1008109.ref005],[@ppat.1008109.ref048]\]. But this strong H1N1 protection seems to come at a cost; even after decades of seasonal H3N2 exposure, and vaccination, older cohorts have evidently failed to develop equally strong protection against H3N2. HA group 1 antigens (e.g. H1) appear to induce narrower immune responses, and less cross-group protection than structurally distinct HA group 2 antigens (e.g. H3) \[[@ppat.1008109.ref025]\]. Perhaps elderly cohorts imprinted to group 1 antigens have been trapped in narrower responses that offer exceptional protection against strains similar to that of first exposure but relatively poor adaptability to other subtypes. We speculate that imprinting protection, which currently limits the number of severe, clinically-attended H1N1 cases in the elderly, also limits the mortality impact of H1N1 viruses. Although pre- and post- 2009 H1N1 lineages have caused slightly different profiles of age-specific mortality \[[@ppat.1008109.ref013]\], neither H1N1 lineage causes nearly as many deaths as H3N2 in high-risk elderly cohorts \[[@ppat.1008109.ref013],[@ppat.1008109.ref031],[@ppat.1008109.ref049]\]. On the one hand, if strong subtype-specific biases from imprinting remain in future cohorts of elderly adults, our results would corroborate the idea that mortality from H1N1 may increase as protection in the elderly shifts from H1N1 toward other subtypes \[[@ppat.1008109.ref009],[@ppat.1008109.ref013]\]. On the other hand, given that cohorts born after 1968 have had much more varied early life exposures to both H1N1 and H3N2, these cohorts may show a greater ability to act as immunological generalists as they become elderly, capable of effective defense against multiple subtypes. Our study has several limitations. Relatively severe, clinically attended cases are much more likely to be detected, confirmed to subtype, and included in our data than mild cases. Thus, while our results show a clear relationship between subtype-level imprinting and risk of relatively severe, clinically attended influenza, the relationship between imprinting and mild or asymptomatic cases could not be determined from available data. Given the limited number of variables recorded in the data, we could not model explicitly the impact of individual risk factors such as the presence of comorbidities, patient sex, or vaccination status. All these factors are known to shape immunity and influenza risk \[[@ppat.1008109.ref050]\], and all may cause individual imprinting outcomes to vary from the average, population trends measured by our study. Understanding how these patient-level covariates modulate imprinting and other aspects of immunity is the next frontier in this line of research. For now, working within the constraints of the available data, we designed the age-specific risk component of the model to capture empirically the combined effects of several risk factors that could not be modeled individually. Additionally, we analyzed the relative count of H1N1 to H3N2 cases within each single year of birth, not absolute incidence, to control for minor age-specific biases in sampling, which are almost inevitably present in any large surveillance data set. Another limitation was the low number of confirmed cases available in the pre-2009 era. Large, detailed data sets collected continuously over decades provide the greatest power to separate the effects of age from birth year. We emphatically echo earlier calls \[[@ppat.1008109.ref051]\] for more systematic sharing of single year-of-age influenza surveillance data, standardization of sampling effort, and reporting of age-specific denominators, which could substantially boost the scientific community's ability to link influenza\'s genetic and antigenic properties with epidemiological outcomes. Additionally, collection and reporting of covariates such as sex, vaccination status and the presence of comorbidities in surveillance data would help us understand how patient-level variables modulate imprinting, and immunity in general \[[@ppat.1008109.ref052],[@ppat.1008109.ref053]\]. Altogether, this analysis confirms that the epidemiological burden of H1N1 and H3N2 is shaped by cohort-specific differences in childhood imprinting \[[@ppat.1008109.ref009],[@ppat.1008109.ref012],[@ppat.1008109.ref013],[@ppat.1008109.ref016],[@ppat.1008109.ref054]\], and that this imprinting acts at the HA or NA subtype level against seasonal influenza \[[@ppat.1008109.ref016],[@ppat.1008109.ref017]\]. The lack of support for broader, HA group-level imprinting effects emphasizes the consequences of immunodominance of influenza's most variable epitopes, and the difficulty of deploying broadly protective memory B cell responses against familiar, seasonal strains. Overall, these findings advance our understanding of how antigenic seniority shapes cohort-specific risk during epidemics. The fact that elderly cohorts show relatively weak immune protection against H3N2, even after living through decades of seasonal exposure to or vaccination against H3N2, suggests that antibody responses acquired in adulthood do not provide the same strength or durability of immune protection as responses primed in childhood. Immunological experiments that consider multiple viral exposures, and cohort studies in which individual histories of influenza infection are tracked from birth, promise to illuminate how B cell and T cell memory develop across a series of early life exposures. In particular, these studies may provide clearer insights than epidemiological data into which influenza antigens, epitopes and immune effectors play the greatest role in immune imprinting, and how quickly subtype-specific biases become entrenched across the first or the first few exposures. Materials and methods {#sec014} ===================== Ethics statement {#sec015} ---------------- This study analyzed only existing epidemiological data, which was completely anonymized. Estimation of age from birth year in ADHS data {#sec016} ---------------------------------------------- The data contained three variables, influenza season, birth year and confirmed subtype. For most cases, birth year was extracted directly from the reported date of birth in patient medical records, but age was not known. We estimated patient age at the time case observation using the formula \[year of observation\]-\[birth year\]. To ensure that the minimum estimated age was 0, the second year in the influenza season of case observation was considered the calendar year of observation (e.g. 2013 for the 2012--2013 season). Splines {#sec017} ------- In **[Fig 2](#ppat.1008109.g002){ref-type="fig"}**, smoothing splines were fit to aid visual interpretation of noisy data. We fit splines using the command smooth.spline(x = AGE, y = FRACTIONS, spar = 0.8) in R version 3.5.0. Variables AGE and FRACTIONS were vectors whose entries represented single years of age, and the fraction of cases observed in the corresponding age group. The smoothing parameter 0.8 was chosen to provide a visually smooth fit. Alternative smoothing parameter choices (0.6 & 1.0) are shown in **[S1](#ppat.1008109.s001){ref-type="supplementary-material"} and [S2](#ppat.1008109.s002){ref-type="supplementary-material"} Figs**. Although the choice of smoothing parameter changed the shape of each fitted spline, qualitative differences between splines fitted to H1N1 or H3N2 were insensitive. Model formulation {#sec018} ----------------- For each unique season in which cases were observed, define p as a vector whose entries represent the expected probability that a randomly drawn H1N1 or a randomly drawn H3N2 case was observed in an individual born in year b. Each model defined p as a linear combination of age-specific risk, birth year-specific risk (i.e. imprinting effects) and demographic age distribution. All tested models were nested within the equation: $$p = DA*1_{H1N1}\left( I_{H1N1} \right)*1_{H3N2}\left( I_{H3N2} \right)$$ To include risk factors that only modulated risk from one subtype, we included indicator functions **1**~**H1N1**~ and **1**~**H3N2**~, which took value 1 if p described the expected age distribution of H1N1 or H3N2 cases, respectively, and 0 otherwise. ### Demographic age distribution (D) {#sec019} The population of Arizona aged slightly across the study period, so we controlled for shifting demography in all tested models. Demographic age distribution was obtained from intercensal estimates of total population (both sexes) for the state of Arizona, based on the 2000 and 2010 census \[[@ppat.1008109.ref055]\]. The US Census Bureau reports population estimates for ages 0--84, but only provides an aggregate estimate for ages 85+. We impute the number of individuals in each single year of age over 85 using a linear model fit to data on age 75--84, with a minimum threshold of 1000 individuals per single year of age. State-specific population estimates were not available prior to the 2000 census, so we substituted estimates from the year 2000 for cases observed in the 1993--94, and the 1994--95 seasons. Vector D represented the fraction of the total population at the time of case observation that fell in a given birth year. ### Age-specific risk (A) {#sec020} Age-specific risk was defined as a step function, in which relative risk was fixed to value 1 in an arbitrarily chosen age bin, and then z-1 free parameters, denoted r~2~ to r~z~, were fit to describe relative risk in all other age bins. Below, **1**~**i**~ are indicator functions specifying whether each vector entry is a member of age bin i. {#ppat.1008109.e002g} A = 1 1 \+ 1 2 r 2 \+ ⋯ 1 z r z To obtain the predicted fraction of cases observed in each single year of birth, we normalized so that the product of vectors representing demographic age distribution, and age-specific risk, (DA in [Eq 1](#ppat.1008109.e001){ref-type="disp-formula"}) summed to 1. Thus, vector DA can be interpreted as the expected distribution of cases of any influenza case (either subtype), in the absence of birth year-specific biases from imprinting. ### Imprinting (I) {#sec021} An indicator function defined whether a given prediction vector described risk of confirmed H1N1 or H3N2. Let f~IHxNy~ be vectors describing the fraction of cases of each birth year that were protected against strain HxNy by their childhood imprinting. We defined r~IHxNy~ as free parameters describing the risk of confirmed HxNy, given imprinting protection. Finally, the factor describing the effect of imprinting (I) was defined as: $$I_{HxNy} = 1_{HxNy}*\left\lbrack f \right._{IHxNy}r_{IHxNy} + \left( 1 - f_{IHxNy} \right)\rbrack$$ Likelihood {#sec022} ---------- We used Eqs [1](#ppat.1008109.e001){ref-type="disp-formula"}--[3](#ppat.1008109.e003){ref-type="disp-formula"} to generate predicted case age distributions (p) for each influenza season (s) in which cases were observed in the data. Then, the likelihood was obtained as a product of multinomial densities across all seasons. If n~s~ represents the total number of cases observed in a given season, x~0cs~,...x~mcs~ each represent the number of cases observed in each single year of birth, and if p~0cs~...p~mcs~ each represent entries in the model's predicted birth year-distribution of cases, then the likelihood is given by: $$\mathcal{L} = {\prod_{s}{\frac{n_{s}!}{x_{0s}!\ldots x_{ms!}}p_{0s}^{x_{0s}}\ldots p_{ms}^{x_{ms}}}}$$ Model fitting and model comparison {#sec023} ---------------------------------- We fit models containing all possible combinations of the above factors to the surveillance data. We simultaneously estimated all free parameter values using the optim() function in R, with method L-BFGS-B. Imprinting parameters could take values in \[0,1\], representing the possibility of a reduction in risk. Age-specific risk parameters could take any value greater than 0. We calculated likelihood profiles and 95% profile confidence intervals for each free parameter. Confidence intervals were defined using the method of likelihood ratios \[[@ppat.1008109.ref036]\]. Antigenic advance {#sec024} ----------------- We obtained antigenic distance estimates from Nextstrain ([nextstrain.org](http://nextstrain.org)) \[[@ppat.1008109.ref038],[@ppat.1008109.ref056]\], and from source data from Fig 3 in Bedford et al. \[[@ppat.1008109.ref040]\]. Nextstrain calculates antigenic distance using genetic data from GISAID \[[@ppat.1008109.ref057]\], and using methods described by Neher et al. \[[@ppat.1008109.ref039]\]. We analyzed "CTiter" estimates from Nextstrain, which correspond to Neher et al.'s tree model method, and are most directly comparable to pre-2009 H1N1 estimates from \[[@ppat.1008109.ref040]\]. We repeated analyses using estimates from Neher et al.'s substitution model method and verified that our choice of antigenic distance metric did not meaningfully impact our results. The negative Spearman correlation between antigenic advance and proportion of cases in children was lower, but still not statistically significant when using the substitution model (p = 0.06); all other differences were unremarkable. Datasets from Nextstrain and Bedford et al. both contained redundant antigenic distance estimates for the H3N2 subtype, but for subtype H1N1, only Bedford et al. analyzed the pre-2009 lineage, and only Nextstrain data analyzed the post-2009 lineage. The antigenic distance estimates reported by Bedford et al. were roughly proportional to those reported on Nextstrain, but greater in absolute magnitude \[[@ppat.1008109.ref039]\]. To enable visualization of all lineages of H1N1 and H3N2 on the same plot axes, we rescaled pre-2009 H1N1 estimates from Bedford et al. using the formula d~Nextstrain~ = 0.47d~Bedford~. The scaling factor was chosen so that directly-comparable H3N2 distance estimates obtained using each method spanned the same range (**[S3 Fig](#ppat.1008109.s003){ref-type="supplementary-material"}**). The Nextstrain data files used in this analysis are archived within our analysis code. Supporting information {#sec025} ====================== ###### ADHS age distributions, all seasons. Supplement to **[Fig 2](#ppat.1008109.g002){ref-type="fig"}** showing observed age distributions from all influenza seasons. Observed case fractions (points) were only plotted if 10 or more cases of a given subtype were confirmed, to avoid extreme stretching of the y axis. Smoothing splines were only plotted if 50 or more cases of a given subtype were observed, as fits to fewer data points would not have been meaningful. (TIFF) ###### Click here for additional data file. ###### Alternate smoothing parameters, AZDHS data. Supplement to **[Fig 2](#ppat.1008109.g002){ref-type="fig"}**, with smoothing parameters chosen to fit splines that are less (**A-F**), or more (**G-L**) smooth than the splines shown in the main text. Differences between H1N1 and H3N2's age-specific impacts remain evident, especially in the oldest cohorts, regardless of smoothness. (PDF) ###### Click here for additional data file. ###### Comparison of rescaled antigenic distance estimates from the Bedford et al., and Nextstrain datasets. Points represent average antigenic position of all isolates from a given calendar year. (TIFF) ###### Click here for additional data file. ###### PCR-confirmed cases from ADHS surveillance, and state-level census data. (ZIP) ###### Click here for additional data file. We are grateful to Ken Komatsu and Kristen Herrick for their assistance with data access, and to Trevor Bedford for assistance accessing and interpreting antigenic distance data from Nextstrain. We thank Lone Simonsen for helpful discussions. Code and data availability {#sec026} ========================== Code to perform all reported analyses and construct all plots, and all relevant data (Arizona surveillance data and relevant antigenic advance data) is archived at <https://zenodo.org/badge/latestdoi/160883450>. Disclaimer {#sec027} ========== This work does not necessarily represent the views of the US government or the NIH. 10.1371/journal.ppat.1008109.r001 Decision Letter 0 Klein Sabra L. Guest Editor Pekosz Andrew Section Editor © 2019 Klein, Pekosz 2019 Klein, Pekosz This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 5 Aug 2019 Dear Ms. Gostic, Thank you very much for submitting your manuscript \"Childhood immune imprinting to influenza A shapes birth year-specific risk during seasonal H1N1 and H3N2 epidemics\" (PPATHOGENS-D-19-01238) for review by PLOS Pathogens. Your manuscript was fully evaluated at the editorial level and by independent peer reviewers. The reviewers appreciated the attention to an important problem, but raised some substantial concerns about the manuscript as it currently stands. These issues must be addressed before we would be willing to consider a revised version of your study. We cannot, of course, promise publication at that time. We therefore ask you to modify the manuscript according to the review recommendations before we can consider your manuscript for acceptance. Your revisions should address the specific points made by each reviewer. In addition, when you are ready to resubmit, please be prepared to provide the following: \(1\) A letter containing a detailed list of your responses to the review comments and a description of the changes you have made in the manuscript. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. \(2\) Two versions of the manuscript: one with either highlights or tracked changes denoting where the text has been changed; the other a clean version (uploaded as the manuscript file). While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Additionally, to enhance the reproducibility of your results, PLOS recommends that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see <http://journals.plos.org/plospathogens/s/submission-guidelines#loc-materials-and-methods> We hope to receive your revised manuscript within 60 days. If you anticipate any delay in its return, we ask that you let us know the expected resubmission date by replying to this email. Revised manuscripts received beyond 60 days may require evaluation and peer review similar to that applied to newly submitted manuscripts. \[LINK\] We are sorry that we cannot be more positive about your manuscript at this stage, but if you have any concerns or questions, please do not hesitate to contact us. Sincerely, Sabra L. Klein Guest Editor PLOS Pathogens Andrew Pekosz Section Editor PLOS Pathogens Kasturi Haldar Editor-in-Chief PLOS Pathogens ​[orcid.org/0000-0001-5065-158X](http://orcid.org/0000-0001-5065-158X) Grant McFadden Editor-in-Chief PLOS Pathogens [orcid.org/0000-0002-2556-3526](http://orcid.org/0000-0002-2556-3526) \*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\* All reviewers commented on the importance of addressing early life exposure in immune imprinting to influenza A viruses in humans. While the reviewers found the question important a few major concerns were raised that must be addressed. First, concerns were raised about biases in surveillance data collection, which can differ by age (Reviewer 3). Also, how host-related factors, including sex, in addition to age, could impact early life imprinting (Reviewer 1) is not considered in the the analyses to ensure that the conclusions drawn are valid across individuals. Reviewer\'s Responses to Questions **Part I - Summary** Please use this section to discuss strengths/weaknesses of study, novelty/significance, general execution and scholarship. Reviewer \#1: The goal of this paper was to assess how childhood immune imprinting shapes seasonal influenza epidemiology. The authors use large epidemiological surveillance data to test whether this immune imprinting acts primarily by immune memory of a particular influenza subtype or via broader immunity that protects again various subtypes. The authors show that within subtype imprinting has the strongest impact on seasonal influenza risk and that antibody responses acquired later in life do not have the same impact on long-term immunity as early-life influenza responses. This is a well-written manuscript with important implications. My specific comments are below: Reviewer \#2: Overall this is a very insightful manuscript that describes the role of early life influenza exposures in protection against currently circulating seasonal influenza strains. This is a topic of great importance and the manuscript itself is well written with clearly presented data. The authors have published a previous paper examining how early life influenza exposures provide some degree of protection against potentially pandemic influenza strains. However, as humans are repeatedly exposed to seasonal influenza via both vaccination and infection, the protection afforded against circulating strains via imprinting could be vastly different. Indeed, the authors do find early life influenza exposures are able to imprint immunity and provide significant but only subtype-specific protection against future infections into adulthood and beyond. Further, a novel role for the NA protein in imprinting is described. Overall this paper represents an important contribution to the field and the data appear to support the conclusions being drawn. Reviewer \#3: This is a well-written manuscript by Gostic and colleagues examining imprinting, a critical question in the influenza field. The modeling is a real strength, but I have significant concerns about the data that is modeled. \*\*\*\*\*\*\*\*\*\* **Part II -- Major Issues: Key Experiments Required for Acceptance** Please use this section to detail the key new experiments or modifications of existing experiments that should be [absolutely]{.ul} required to validate study conclusions. Generally, there should be no more than 3 such required experiments or major modifications for a \"Major Revision\" recommendation. If more than 3 experiments are necessary to validate the study conclusions, then you are encouraged to recommend \"Reject\". Reviewer \#1: Assessment of host factors Although the authors investigate how age can impact imprinting responses, they do not provide the sex of the individuals in their study, nor do they account for sex in their analyses. Previous work has shown that sex can impact influenza responses in human and animal models, and it is important that in this type of epidemiological study, host factors, including sex, are included. Further, although I recognize it is difficult to access human data over the life course, it would be important to know more about the patients included in the analysis and if any patients had co-morbidities that may have influenced their responses both early on and later in life. With increasing age, susceptibility to infection increases; antibody responses and vaccine efficacy decrease; and sex hormones decrease. These immunological and endocrinological changes are important to account for when assessing how age and year of birth may impact responses to influenza. Reviewer \#2: None Reviewer \#3: 1. This manuscript relies completely on influenza surveillance data, however, no detail is provided on the surveillance system. In particular, typically state surveillance systems are designed in such a way that they target specific at-risk populations. Thus, the data is collected in an intentionally biased way and does not represent what is happening at a population level, but rather maximizes the cases identified. For instance, sentinel surveillance systems tend to have over-representation of pediatric offices to catch non-severe cases, outbreaks tend to be investigated (and reported on) in old age homes, and hospital surveillance may be more heavily done in specific adult hospitals. In addition, the design of the surveillance system may change over time. If severity patterns differ by virus, which is likely, and surveillance setting differs by age (which is also likely, but I am unfamiliar with surveillance in Arizona), then bias could be introduced into the models. The authors should report on the design of the surveillance system in the methods section including any changes over the study period and, most importantly, on how the design of the surveillance system may have impacted their results. \*\*\*\*\*\*\*\*\*\* **Part III -- Minor Issues: Editorial and Data Presentation Modifications** Please use this section for editorial suggestions as well as relatively minor modifications of existing data that would enhance clarity. Reviewer \#1: Childhood age range The authors include "children" from 0 to 10 years of age. Have the authors thought about the fact that individuals within that large range may have varying degrees of imprinting and also exhibit vastly different immune responses (e.g., a 6-month old child will elicit different responses from a 10 year old). It would be helpful if the authors address this idea and suggest how they controlled for this in the current study. Reviewer \#2: While the discussion raises many important points, it is very long and would benefit from being compressed while still retaining the major points being made. Although outside of the scope of the current paper (and analyzed dataset), a future study performing similar analysis on a cohort on which serum/PBMCs could also be analyzed would provide significant insight on the immunologic mechanisms underlying the described findings. Reviewer \#3: 1. The authors should define imprinting more thoroughly in the introduction as non-influenza audiences may not be familiar with the term and in the influenza field imprinting has been used to describe several phenomena that may have different mechanistic bases (and indeed the authors themselves do this when using imprinting vs. within-subtype imprinting). 2\. The authors state that HA group-level responses, which I assume includes the stem, are not thought to play a strong role in defense against familiar strains, however, a recent article by Ng et. al demonstrated that anti-HA stem antibodies are an independent correlate of protection. 3\. Throughout the manuscript the authors refer to "infections" or "infecting" but they are looking at reported cases, not infections. 4\. The authors suggest that imprinting may be a result of CD4 T-cells (as one possibility). Is there any evidence for this? 5\. The authors use a case series. Did the population structure of Arizona change substantially over the time period? If so, they should adjust for the changes in the demographics. \*\*\*\*\*\*\*\*\*\* PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plospathogens/s/editorial-and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose "no", your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No 10.1371/journal.ppat.1008109.r002 Decision Letter 1 Klein Sabra L. Guest Editor Pekosz Andrew Section Editor © 2019 Klein, Pekosz 2019 Klein, Pekosz This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 25 Sep 2019 Dear Ms. Gostic, We are pleased to inform that your manuscript, \"Childhood immune imprinting to influenza A shapes birth year-specific risk during seasonal H1N1 and H3N2 epidemics\", has been editorially accepted for publication at PLOS Pathogens.  Before your manuscript can be formally accepted and sent to production, you will need to complete our formatting changes, which you will receive by email within a week. Please note that your manuscript will not be scheduled for publication until you have made the required changes. IMPORTANT NOTES \(1\) Please note, once your paper is accepted, an uncorrected proof of your manuscript will be published online ahead of the final version, unless you've already opted out via the online submission form. If, for any reason, you do not want an earlier version of your manuscript published online or are unsure if you have already indicated as such, please let the journal staff know immediately at <plospathogens@plos.org>. \(2\) Copyediting and Proofreading: The corresponding author will receive a typeset proof for review, to ensure errors have not been introduced during production. Please review the PDF proof of your manuscript carefully, as this is the last chance to correct any errors. Please note that major changes, or those which affect the scientific understanding of the work, will likely cause delays to the publication date of your manuscript.  \(3\) Appropriate Figure Files: Please remove all name and figure \# text from your figure files. Please also take this time to check that your figures are of high resolution, which will improve the readbility of your figures and help expedite your manuscript\'s publication. Please note that figures must have been originally created at 300dpi or higher. Do not manually increase the resolution of your files. For instructions on how to properly obtain high quality images, please review our Figure Guidelines, with examples at: <http://journals.plos.org/plospathogens/s/figures>. While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. \(4\) Striking Image: Please upload a striking still image to accompany your article if one is available (you can include a new image or an existing one from within your manuscript). Should your paper be accepted, this image will be considered for our monthly issue image and may also appear on our website to feature your article. Please upload this as a separate file, selecting \"striking image\" as the file type upon upload. Please also include a separate \"Other\" file with a caption, including credits and any potential copyright information. Please do not include the caption in the main article file. If your image is from someone other than yourself, please ensure that the artist has read and agreed to the terms and conditions of the Creative Commons Attribution License at <http://journals.plos.org/plospathogens/s/content-license>. Please note that PLOS cannot publish copyrighted images. \(5\) Press Release or Related Media: If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team in advance at <plospathogens@plos.org> as soon as possible. We ask that you contact us within one week to plan ahead of our fast Production schedule. If you need to know your paper\'s publication date for related media purposes, you must coordinate with our press team, and your manuscript will remain under a strict press embargo until the publication date and time. This means an early version of your manuscript will not be published ahead of your final version.  (6)  PLOS requires an ORCID iD for all corresponding authors on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager.  To do this, go to 'Update my Information' (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field.  This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager \(7\) Update your Profile Information: Now that your manuscript has been provisionally accepted, please log into Editorial Manager and update your profile, if needed. Go to <https://www.editorialmanager.com/ppathogens>, log in, and click on the \"Update My Information\" link at the top of the page. Please update your user information to ensure an efficient production and billing process.  \(8\) LaTeX users only: Our staff will ask you to upload a TEX file in addition to the PDF before the paper can be sent to typesetting, so please carefully review our Latex Guidelines <http://journals.plos.org/plospathogens/s/latex> in the meantime. (9) If you have associated protocols in protocols.io, please ensure that you make them public before publication to guarantee immediate access to the methodological details. Thank you again for supporting Open Access publishing; we are looking forward to publishing your work in PLOS Pathogens.  Best regards, Sabra L. Klein Guest Editor PLOS Pathogens Andrew Pekosz Section Editor PLOS Pathogens Kasturi Haldar Editor-in-Chief PLOS Pathogens ​[orcid.org/0000-0001-5065-158X](http://orcid.org/0000-0001-5065-158X) Grant McFadden Editor-in-Chief PLOS Pathogens [orcid.org/0000-0002-2556-3526](http://orcid.org/0000-0002-2556-3526) \*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\* The authors carefully considered the reviews and incorporated as many of the requested modifications as possible, given some limitations of the data set. This is a very important study that will make a significant impact on the field. There are no additional concerns or requests for modifications. Reviewer Comments (if any, and for reference): 10.1371/journal.ppat.1008109.r003 Acceptance letter Klein Sabra L. Guest Editor Pekosz Andrew Section Editor © 2019 Klein, Pekosz 2019 Klein, Pekosz This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 8 Nov 2019 Dear Ms. Gostic, We are delighted to inform you that your manuscript, \"Childhood immune imprinting to influenza A shapes birth year-specific risk during seasonal H1N1 and H3N2 epidemics,\" has been formally accepted for publication in PLOS Pathogens. We have now passed your article onto the PLOS Production Department who will complete the rest of the pre-publication process. All authors will receive a confirmation email upon publication. The corresponding author will soon be receiving a typeset proof for review, to ensure errors have not been introduced during production. Please review the PDF proof of your manuscript carefully, as this is the last chance to correct any scientific or type-setting errors. Please note that major changes, or those which affect the scientific understanding of the work, will likely cause delays to the publication date of your manuscript. Note: Proofs for Front Matter articles (Pearls, Reviews, Opinions, etc\...) are generated on a different schedule and may not be made available as quickly. Soon after your final files are uploaded, the early version of your manuscript, if you opted to have an early version of your article, will be published online. The date of the early version will be your article\'s publication date. The final article will be published to the same URL, and all versions of the paper will be accessible to readers. Thank you again for supporting open-access publishing; we are looking forward to publishing your work in PLOS Pathogens. Best regards, Kasturi Haldar Editor-in-Chief PLOS Pathogens ​[orcid.org/0000-0001-5065-158X](http://orcid.org/0000-0001-5065-158X) Grant McFadden Editor-in-Chief PLOS Pathogens [orcid.org/0000-0002-2556-3526](http://orcid.org/0000-0002-2556-3526) [^1]: The authors have declared that no competing interests exist. [^2]: Current address: Dept. of Ecology and Evolution, University of Chicago, Chicago, Illinois, United States of America

In the article that appeared on page 404 of July-August 2014 issue\[[@ref1]\], authors affiliation was incorrectly displayed. Authors affiliation should have appeared as given below: ^1^Associate Professor, Infection Nosocomial Research Center, Department of Infectious Diseases, Medical School, Isfahan University of Medical Sciences, Isfahan, Iran. ^2^Department of Biostatistics, School of Health, Isfahan University of Medical Sciences, Isfahan, Iran. ^3^Student Research PhD, Infectious Diseases and Tropical Medicine Research Center, Isfahan University of Medical Sciences, Isfahan, Iran. ^4^Department of Midwifery, School of Nursing and Midwifery, Isfahan University of Medical Sciences, Isfahan, Iran.

Introduction {#s1} ============ Pilomatrixoma (calcifying epithelioma of Malherbe) is a benign tumour of hair follicular matrix cells based on histochemical and electron microscopic studies[@R01]. It was first described by Malherbe and Chenantais in 1880 as a calcified tumour from the sebaceous glands[@R02]. It commonly affecs children and adolescents; however, there is a smaller bell graph in the elderly[@R03]. It is slightly more common in females. Its occurrence in upper limb is rare and may mimic other cutaneous lesion ranging from benign to malignant tumours, at its different stages of presentation[@R04]. Case Report {#s2} =========== A 69-year-old lady presented with left shoulder swelling for past 20 years. The swelling was gradually enlarging and associated with pain in the last two years. One month prior to consultation, the swelling ulcerated with seropurulent discharge. She had no fever nor constitutional sign and symptom of malignancy. Clinical examination revealed an 8cm x 5cm x 3cm firm exophytic swelling with central ulceration over the left deltoid region ([Figure 1a](#F1a){ref-type="fig"}). It appeared not adherent to underlying muscle. Range of motion of left shoulder was full. No cervical or axillary lymph nodes were palpable. Neurovascular status of left upper limb was normal. Plain radiograph revealed a heterogeneous soft tissue lesion with foci of opacity. Magnetic Resonance Imaging ([Figure 1b](#F1b){ref-type="fig"}) showed a well defined soft tissue mass involving the cutaneous and subcutaneous layers. It displayed iso/hypointense T1 signal to muscles, heterogeneous T2 signal and heterogeneous enhancement in post intravenous gadolinium study. Medially, there was no clear demarcation with the deltoid fascia, suggestive of fascia involvement, but the deltoid muscle appeared not involved. The radiological differential diagnoses were dermatofibrosarcoma protuberans, squamous cell carcinoma, malignant fibrous histiocytoma or nodular fasciitis. Wide surgical excision of the left shoulder tumour was performed and wound was closed primarily ([Figure 2a](#F2a){ref-type="fig"}). Histopathological examination of the excised tumour confirmed the diagnosis of pilomatrixoma, with no evidence of malignancy ([Figure 2b](#F2b){ref-type="fig"}). On follow up review of the patient at three months, the wound had healed without complication; there was no early local recurrence and left shoulder range of motion was full. On follow up at six months, one year and one and a half years, patient showed no local recurrence of the tumour and scar was clinically quiescent, with full range of motion of the left shoulder and motor power comparable to right shoulder ([Figure 3](#F3){ref-type="fig"}). She was able to perform daily activities independently including driving. Discussion {#s3} ========== The presentation of this lady with a rapidly enlarging swelling and ulceration with seropurulent discharge associated with pain, foci calcification on plain radiograph, as well as heterogenous signal on MRI alerted on the high possibility of a malignant tumour. The clinical appearance mimicked the malignant picture of squamous cell carcinoma. Hence, without pre- excision cytological examination or intralesional biopsy, wide surgical excision of the tumour was performed. This more extensive surgical approach was performed as if it is a malignant tumour. Adjuvant chemotherapy or radiotherapy might be indicated if malignancy was confirmed. The histopathological and microscopic examination however, showed a benign partly ulcerated circumscribed lesion formed by solid, irregular islands of epithelial cells which were embedded within a fibrocollagenous stroma. The epithelial cells were of two basic types, with the predominant cell type comprising of eosinophilic shadow cells (Ghost cell) with distinct cytoplasmic borders and a central unstained area due to lost nucleus ([Figure 2b](#F2b){ref-type="fig"}). The second cell type was composed of mostly squamous cells, with occasional basophilic cells resembling hair matrix cells. Mitotic activity was present focally but no cytological atypia was observed. The attached stroma showed foreign-body giant cell reaction and infiltration by mainly chronic mononuclear inflammatory cells. The typical 'Ghost Cell' appearance is suggestive of pilomatrixoma[@R04]. The clinical picture of the ulcerating lesion suggested a possible suspicion of a malignant tumour and a wide surgical excision was therefore carried out promptly. A fine needle aspiration for cytological examination or intralesional biopsy (in this reported case, intralesional would have been preferred from fine needle aspiration in view of its clinical appearance) would be most helpful to reach a diagnosis and assist in landscaping the excision margin, apart from the radiological findings. A malignant lesion could have been ruled out and a marginal surgical excision would have been adequate[@R04],[@R05]. The surgical excision of the lesion for this case has brought about the cessation of a ulcerating lesion and its subsequent complications such as superimposed infection. The complaint of pain was being solved with the lesion being excised and the cosmesis of the shoulder preserved. The lesion may be left untreated if it is at early stage without producing significant impairment functionally or cosmetically (ie, ulceration, pain or enlarging). Although the reported local recurrence rate after complete surgical excision is rare[@R05], the patient should be followed up for rapid or local recurrence as it can be a red flag for malignant transformation (pilomatrix carcinoma). {#F1a} {#F1b} {#F2a} {#F2b} {#F3}

Introduction ============ Modified nucleosides are one of the most important classes of antivirals.[@cit1],[@cit2] They mostly work through intracellular phosphorylation to the corresponding nucleoside triphosphates (NTPs), which inhibit the viral DNA or RNA polymerase and/or terminate the DNA or RNA chain. Modification on the nucleobase can bring higher metabolic stability and increased affinity to the enzyme, whereas sugar modifications cause termination of the chain and often bring selectivity toward viral polymerases. On the other hand, chemical modifications of nucleosides often lead to inefficient phosphorylation by nucleoside kinases and, therefore, many of the nucleoside drugs are use in form of 5′-phosphate prodrugs.[@cit2],[@cit3] These effects can be demonstrated on blockbuster drug Sofosbuvir (2′-α-fluoro-2′-β-methyluridine phosphoramidate prodrug),[@cit4] which is used for treatment of Hepatitis C (HCV), and on GS-5734 (phosphoramidate of 1′-cyano-ribo-C-nucleoside bearing 4-aminopyrrolo\[2,1-*f*\]triazine base),[@cit5] which is in clinical trials for treatment of Ebola virus. Despite recent progress in treatment of HCV and other viruses, there are many other, so far neglected emerging viruses,[@cit6] for which there is no treatment available, and hence are the challenge for the current medicinal chemistry. Our long-term research of biological activities of 7-deazapurine nucleosides resulted in discovery of two main groups of cytostatics (6-hetaryl-7-deazapurines **1**[@cit7] and 7-hetaryl 7-deazapurines **2**[@cit8]) with nanomolar activities against broad panel of cancer cell lines. These compounds also showed potent anti-HCV effects, which were unfortunately accompanied by cytotoxicity. These results showed the space for modification in the "major groove" part of the molecule and inspired us to design of fused-7-deazapurine nucleosides with the aim of possible selectivity modulating of antiviral and cytostatic activities. First generation of such fused nucleosides, pyrimidoindole ribonucleosides **3a** bearing various hetaryl groups in position 4,[@cit9] displayed negligible cytostatic activity, however, several derivatives bearing 2-hetaryl groups exerted interesting micromolar activity against dengue virus.[@cit9] Benzo-fused 7-deazaadenine analogues **3b** showed[@cit10] similarly potent anti-dengue effect and anti-HCV activity with 4-methyl derivative being the most active compound with sub-micromolar anti-HCV activity (replicon 1B) and no cytotoxicity. Second generation of fused nucleosides, thienopyrrolopyrimidine ribonucleosides **4**,[@cit11] were again cytostatic at nanomolar concentrations with potent anti-HCV activity accompanied by cytotoxicity and no effect against dengue virus. In order to complete the SAR of this class of compounds and to gain selectivity to RNA viruses without cytotoxicity, we designed sugar-modified nucleosides derived from 4-substituted 6-chloropyrimido\[4,5-*b*\]indole ribonucleosides ([Fig. 1](#fig1){ref-type="fig"}). We focused on 2′-deoxy-2′-fluororibo-, 2′-deoxy-2′-fluoroarabino- and arabinonucleosides, because related sugar modified derivatives of 7-hetaryl-7-deazapurine nucleosides were previously shown to be significantly less cytotoxic than corresponding ribonucleosides.[@cit12] Moreover, arabino- or 2′-fluoroarabino sugars occur in clinically used cytostatics Clofarabine[@cit13] and Fludarabine[@cit14] and also some 2′-fluororibonucleotides have displayed biological effects.[@cit15] ![Previously reported 7-deazapurine nucleosides and fused-7-deazapurine nucleosides with cytostatic and antiviral activities. Custom purine numbering (red) is shown in structure **2**, systematic numbering (black) of pyrrolo\[2,3-*d*\]pyrimidines and pyrimido\[4,5-*b*\]indoles is shown in structures **1** and **3**, respectively.](c7md00319f-f1){#fig1} Chemistry ========= Our synthetic plan toward sugar modified pyrimidoindole nucleosides was based on the preparation of the key-intermediate 4,6-dichloropyrimidoindole nucleosides either by glycosylation of known heterocyclic base or by manipulation of functional group in 2′-position of the sugar moiety. The substituents could be introduced to the position 4 on pyrimidine ring either by nucleophilic substitution or by cross-coupling reaction in the final steps. A nucleobase anion glycosylation of the previously reported 4,6-dichloropyrimido\[4,5-*b*\]indole (**5**)[@cit9] with the known *α*-bromo-2-fluoroarabinose **6**[@cit16] furnished the desired key-intermediate fluoroarabinonucleoside **7** in 51% yield ([Scheme 1](#sch1){ref-type="fig"}) from which a series of final 4-substituted 2′-deoxy-2′-fluoroarabinonucleosides **9a--i** was then synthesized. The selection of substituents and reaction conditions was based on our previous experience with fused-deazapurine nucleosides.[@cit9] First, we attempted to deprotect nucleoside **7** to get free 4-chloro 2′-deoxy-2′-fluoroarabinonucleoside, however, the position 4 on pyrimidoindole base was found so reactive, that nucleophilic substitution was easier than debenzoylation and proceeded simultaneously. With the aim to introduce substituents selectively into the position 4 and keep chlorine in position 6 untouched, we applied previously optimized conditions for Suzuki coupling (catalysis by Pd(PPh~3~)~4~ in combination with potassium carbonate as a base in toluene) to synthesize 4-phenyl-, 4-(3-thienyl)- and 4-(3-furyl)-derivatives. Isomeric 2-furyl- and 2-thienyl-derivatives were obtained by Stille coupling with 2-(tributylstannyl)furan or 2-(tributylstannyl)thiophene catalyzed by PdCl~2~(PPh~3~)~2~ in DMF. Methyl group was introduced by Pd-catalyzed methylation with trimethylaluminium. All these reactions were performed starting from the benzoylated nucleoside **7** and intermediates **8d--i** were then deprotected to desired final free nucleosides **9d--i** using the standard Zemplén method -- sodium methoxide in methanol. The amino-, methoxy- and methylsulfanyl-derivatives **9a**, **9b**, and **9c** were obtained by nucleophilic substitution with aqueous ammonia in dioxane at 100 °C, sodium methoxide in MeOH or sodium methanethiolate in EtOH, respectively. Benzoyl groups were simultaneously removed under reaction conditions and the final free nucleosides **9** were isolated in good yields ([Scheme 1](#sch1){ref-type="fig"}, [Table 1](#tab1){ref-type="table"}). {#sch1} ###### Synthesis of 2′-deoxy-2′-fluoroarabinonucleosides **8** and **9** Entry R Conditions Protected nucleoside Yield \[%\] Final nucleoside Yield \[%\] ------- ----------- ------------ ---------------------- ------------- ------------------ ------------- 1 NH~2~ b --- --- **9a** 78 2 OMe c --- --- **9b** 22 3 SMe d --- --- **9c** 32 4 Me e **8d** 46 **9d** 78 5 2-Furyl g **8e** 79 **9e** 69 6 3-Furyl f **8f** 77 **9f** 33 7 2-Thienyl g **8g** 51 **9g** 78 8 3-Thienyl f **8h** 53 **9h** 65 9 Phenyl f **8i** 55 **9i** 65 The synthesis of arabinonucleosides and 2′-deoxy-2′-fluororibonucleosides was envisaged by modification of 2′-position of the corresponding 4,6-dichloropyrimidoindole ribonucleoside intermediate **12**. It was prepared by stereoselective glycosylation of the pyrimidoindole nucleobase **5** with the protected 1-chlororibose **10**[@cit17] followed by sugar deprotection. The desired nucleoside **12** was obtained in overall 29% yield as the pure β-anomer ([Scheme 2](#sch2){ref-type="fig"}). {#sch2} The key 4,6-dichloropyrimido\[4,5-*b*\]indole arabinonucleoside intermediate **16** was then prepared by inversion of configuration at the 2′-carbon of the 3′,5′-protected ribonucleoside **13** using a sequence of redox reactions. Nucleoside **13** was first oxidized by Dess--Martin periodinane to oxo-derivative **14** in excellent 91% yield. Then a well known stereoselective reduction of **14** using NaBH~4~ in ethanol[@cit12a],[@cit12b] furnished the desired silylated arabinonucleoside **15**, which was deprotected to the free arabinonucleoside **16** in very good 72% yield over 4 steps ([Scheme 3](#sch3){ref-type="fig"}). {#sch3} A series of 4-substituted arabinonucleosides **17a--i** was then prepared in good yields by aromatic nucleophilic substitution, Pd-catalyzed cross-coupling reaction with trimethylaluminium or aqueous-phase Suzuki cross-coupling reaction catalyzed by palladium acetate in combination with TPPTS ([Scheme 3](#sch3){ref-type="fig"}, [Table 2](#tab2){ref-type="table"}). The only low yielding reaction was the Suzuki coupling with 2-furylboronic acid probably due to limited stability of the reagent. ###### Synthesis of arabinonucleosides **17** Entry R Conditions Product Yield \[%\] ------- ----------- ------------ --------- ------------- 1 NH~2~ e **17a** 85 2 OMe f **17b** 77 3 SMe g **17c** 71 4 Me h **17d** 68 5 2-Furyl i **17e** 33 6 3-Furyl i **17f** 62 7 2-Thienyl i **17g** 70 8 3-Thienyl i **17h** 75 9 Phenyl i **17i** 58 4,6-Dichloropyrimido\[4,5-*b*\]indole 2′-deoxy-2′-fluororibonucleoside **22** was selected as the key intermediate for the synthesis of a series of 2′-deoxy-2′-fluororibo derivatives. It was obtained in good 35% overall yield by a 6-step synthesis concluded by stereoselective S~N~2 fluorination of the bis-THP-protected arabinoside **21** followed by acidic deprotection ([Scheme 4](#sch4){ref-type="fig"}). A series of 4-substituted 2′-deoxy-2′-fluororibonucleosides **23a--i** was prepared analogously to arabinonucleosides **17** by nucleophilic substitutions or by Pd-catalyzed cross-coupling reactions ([Scheme 4](#sch4){ref-type="fig"}, [Table 3](#tab3){ref-type="table"}). Again, the Suzuki reaction with 2-furylboronic acid gave low yield of desired nucleoside **23e**. {#sch4} ###### Synthesis of fluororibonucleosides **23** Entry R Conditions Product Yield \[%\] ------- ----------- ------------ --------- ------------- 1 NH~2~ g **23a** 82 2 OMe h **23b** 80 3 SMe i **23c** 87 4 Me j **23d** 65 5 2-Furyl k **23e** 12 6 3-Furyl k **23f** 50 7 2-Thienyl k **23g** 70 8 3-Thienyl k **23h** 40 9 Phenyl k **23i** 41 Biological activity profiling ============================= Antiviral activity ------------------ All the title nucleosides were subjected to screening of their antiviral activities against hepatitis C virus (HCV, genus: *Hepacivirus*, family: *Flaviridae*), respiratory syncytial virus (RSV, genus: *Respirovirus*, family: *Paramyxoviridae*), dengue virus type 2 (strain 16681, genus: *Flavivirus*, family: *Flaviviridae*), influenza (H1N1 A/Mexico/4108/2009, genus: *Influenzavirus A*, family: *Orthomyxoviridae*), human coxsackie B3 virus (strain Nancy, genus: *Enterovirus*, family: *Picornaviridae*) and human herpesvirus 1 (strain HF, genus: *Simplexvirus*, subfamily: *Alphaherpesvirinae*, family: *Herpesviridae*). The anti-coxsackie, anti-herpes, anti-influenza activity was measured by determining the extent to which the test compounds inhibited virus-induced cytopathic effect in HeLa cells, Vero cells, and MDCK cells, respectively, as previously described.[@cit18] None of the nucleosides showed any activity against influenza, coxsackie and human herpesvirus. The anti-RSV activity was tested based on methods published previously.[@cit19] All the title arabinonucleosides and fluororibonucleosides were completely inactive against RSV. Few fluoroarabino derivatives (**9c**, **9e**, **9f**, **9i**) showed moderate micromolar activity (13.2, 34.6, 25.2 and 11.3 μM, respectively) against RSV. 2-Thienyl derivative **9g** was the most active compound with EC~50~ = 5.2 μM. The anti-dengue activity was measured by determining the extent to which the test compounds inhibited replication in Vero cells as previously described.[@cit10] Fluoroarabinonucleosides **9a**, **9e**, **9g**, **9h** and arabino derivative **16** inhibited dengue virus with EC~50~ = 10--33 μM, however, their selectivity index was rather low ([Table 4](#tab4){ref-type="table"}). ###### Anti-HCV and anti-dengue activities of nucleosides Compd HCV (1B) HCV (2A) Dengue type 2 -------------- ---------- ---------- --------------- -------- -------- ------- ------ ------ ----- **9a** 6.7 \>44.4 \>6.6 \>44.4 33.1 0.75 10.8 12.7 1.2 **9b** 3.1 \>44.4 \>14.3 10.8 \>44.4 \>4.1 \>50 \>50 --- **9c** 1.6 22.9 14.3 6.9 20.2 2.9 \>50 \>50 --- **9d** 6.3 \>44.4 \>7.0 23.2 \>44.4 \>1.9 \>50 \>50 --- **9e** 4.6 \>44.4 \>9.7 14.7 34.6 2.4 10.5 39.0 3.7 **9f** 23.0 \>44.4 \>1.9 \>44.4 \>44.4 --- \>50 \>50 --- **9g** 2.5 \>44.4 \>17.8 13.9 34.3 2.5 27.9 39.1 1.4 **9h** 4.1 26.5 6.5 16.2 30.3 1.9 33.3 \>50 1.5 **9i** 5.2 \>44.4 \>8.5 15.7 34.4 2.2 \>50 \>50 --- **16** 13.7 32.5 2.4 17.4 21.4 1.2 17.4 40.9 2.4 **17a** 22.3 \>44.4 \>2.0 \>44.4 \>44.4 --- \>50 \>50 --- **17c** 17.4 \>44.4 \>2.6 38.5 \>44.4 \>1.2 \>50 \>50 --- **17d** 18.5 40.2 2.2 \>44.4 \>44.4 --- \>50 \>50 --- **17e** 3.0 \>44.4 \>14.8 24.0 \>44.4 \>1.9 \>50 \>50 --- **22** 4.7 9.1 1.9 9.5 10.3 1.1 \>50 \>50 --- **23a** 4.7 \>44.4 \>9.4 \>44.4 \>44.4 --- \>50 \>50 --- **23b** 8.7 \>44.4 \>5.1 \>44.4 \>44.4 --- \>50 \>50 --- **23c** 6.6 29.0 4.5 17.0 \>44.4 \>2.6 \>50 \>50 --- **23d** 5.2 \>44.4 \>8.5 16.1 \>44.4 \>2.8 \>50 \>50 --- **23e** 2.3 \>44.4 \>19.3 15.7 \>44.4 \>2.8 \>50 \>50 --- **23f** 5.2 \>44.4 \>8.5 21.8 \>44.4 \>2.0 \>50 \>50 --- **23g** 18.4 \>44.4 \>2.4 19.7 \>44.4 \>2.3 \>50 \>50 --- **23h** 25.1 \>44.4 \>1.8 \>44.4 \>44.4 --- \>50 \>50 --- **23i** 8.2 \>44.4 \>5.4 20.3 \>44.4 \>2.2 \>50 \>50 --- Mericitabine 1.2 \>44.4 \>37 0.99 \>44.4 \>44 nt nt --- Screening of anti-HCV activities was performed as previously described[@cit20] and activities compared to standard Mericitabine.[@cit21] The series of arabinonuclesides **17**, methoxy and hetaryl derivatives was inactive, whilst amino, methylsulfanyl and methyl derivatives showed moderate anti-HCV effect (EC~50~ = 17--38 μM). On the other hand, fluoroarabino- and fluororibonucleosides **9a--i** and **23a--i** were all active against both 1B and 2A genotypes of HCV in replicon assay at (mostly) single digit micromolar concentrations ([Table 4](#tab4){ref-type="table"}) and, more importantly, they were not cytotoxic (in contrast to the previously reported corresponding ribonucleosides[@cit9],[@cit10]). We assume that, similarly to most antiviral nucleosides,[@cit1],[@cit2] the mechanism of antiviral activity is intracellular phosphorylation of the nucleosides to NTPs and inhibition of the viral RNA polymerase. To elucidate whether the NTPs of our modified nucleosides would even fit into the active site of the polymerase, we performed a very simple docking and modelling of selected nucleotides into the known crystal structure (PDB code ; 4WTJ)[@cit22] of viral RNA-dependent RNA polymerase HCV NS5B genotype 2A in complex with RNA template 5′-AUCC, RNA primer 5′-PGG, Mn^2+^ and ADP, which binds to polymerase in catalytically relevant conformation but stalls the primer extension. The modelling was performed using program Moloc and the all-atom MAB force field.[@cit23] We selected disphosphates of three amino-substituted nucleosides **9a**, **17a** and **23a** and we used the implemented MAB force field to energy minimize them in the active site to mimic ADP in the original crystal structure. For optimization, the protein and RNA coordinates were kept fixed. The modelling clearly showed that there is enough space to accommodate the fused chlorobenzene ring in the nucleobase binding site and it could have even increased cation--π stacking with Arg158 ([Fig. 2a](#fig2){ref-type="fig"}). Also the sugar moieties in all three derivatives could adopt similar conformation as in ADP while the orientation of 2′-substituent does not seem to have a significant influence on the binding as it can form hydrogen bonds with the enzyme in ribo-configuration as well as in arabino-configuration ([Fig. 2b](#fig2){ref-type="fig"}). The differences in antiviral activities are probably mostly caused by the different efficiency of the intracellular phosphorylation of the nucleosides. {#fig2} Cytotostatic activity --------------------- The *in vitro* cytostatic activities of the title nucleosides were also evaluated against eight cancer cell lines derived from human solid tumors including lung (A549) and colon (HCT116 and HCT116p53--/--) carcinomas and leukemia cell lines (CCRF-CEM, CEM-DNR, K562 and K562-TAX). Toxicity was evaluated using nonmalignant BJ and MRC-5 fibroblasts. Concentrations inhibiting the cell growth by 50% (IC~50~) were determined as described previously,[@cit11] using a quantitative metabolic staining with 3-(4,5-dimethylthiazol-2yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2*H*-tetrazolium (MTS)[@cit24] following a 3 day treatment. Results are summarized in [Table 5](#tab5){ref-type="table"} (only compounds with IC~50~ \<50 are shown) and compared to Gemcitabine.[@cit25] ###### Cytostatic activities of nucleosides Compd. MTS, IC~50~ (μM) ------------- ------------------ ------ ------ ------ ------ ------ ------ ------ ------ ------ **9a** 37.0 43.9 41.4 15.1 32.6 30.4 30.4 27.5 37.2 25.8 **9b** 47.1 \>50 \>50 34.2 45.9 \>50 \>50 \>50 47.2 \>50 **9c** 21.3 22.6 46.8 11.6 18.7 \>50 \>50 17.6 27.4 29.0 **9d** 20.9 \>50 \>50 5.1 7.5 \>50 44.9 44.2 12.4 29.2 **9e** 26.8 28.7 44.6 13.8 25.0 \>50 \>50 21.7 26.3 39.9 **9f** 46.2 49.6 \>50 7.2 \>50 49.9 49.9 \>50 39.9 37.4 **9g** 32.0 27.0 \>50 18.4 25.8 \>50 \>50 28.1 28.1 \>50 **9h** 41.8 41.8 \>50 21.1 27.6 \>50 \>50 33.0 28.9 \>50 **9i** 27.0 30.9 \>50 19.7 25.4 49.5 49.5 24.8 26.3 33.4 **16** \>50 49.0 32.6 5.1 25.1 34.0 28.5 44.6 17.7 27.4 **17a** \>50 \>50 \>50 28.7 \>50 \>50 49.1 \>50 \>50 \>50 **17c** \>50 \>50 \>50 11.7 24.0 36.1 34.0 26.5 25.7 26.4 **17d** \>50 \>50 \>50 16.1 37.2 \>50 \>50 \>50 \>50 \>50 **17e** \>50 \>50 \>50 27.1 36.6 \>50 \>50 \>50 42.7 \>50 **17i** \>50 \>50 \>50 24.7 \>50 \>50 \>50 \>50 \>50 \>50 **22** \>50 \>50 \>50 3.2 36.3 44.6 46.6 \>50 \>50 32.3 **23b** \>50 \>50 \>50 34.2 37.9 \>50 \>50 49.4 34.0 45.2 **23c** \>50 \>50 \>50 22.1 24.1 \>50 \>50 25.6 26.1 27.2 **23d** \>50 \>50 \>50 10.9 10.0 44.7 48.9 \>50 9.5 \>50 **23e** 39.6 \>50 \>50 16.7 27.5 48.5 48.5 30.7 27.2 37.8 **23f** \>50 \>50 \>50 45.6 39.7 \>50 \>50 \>50 \>50 \>50 **23g** \>50 \>50 \>50 16.2 31.3 \>50 41.6 34.7 28.1 35.2 **23h** \>50 \>50 \>50 37.4 40.8 \>50 \>50 \>50 \>50 \>50 Gemcitabine \>50 \>50 0.05 0.02 0.10 0.03 0.41 0.10 0.05 0.18 Fluoroarabinonucleosides **9** showed only moderate (\>10 μM) cytostatic activity and very poor selectivity against fibroblasts. On the other hand, fluororibonucleosides **23** showed similar activity against CEM cell lines and are not toxic to fibroblasts. Arabinonucleosides **17** bearing methoxy and hetaryl groups in position 4 are inactive against most of the cell lines, they displayed only moderate effect against CEM lines. The most cytotoxic compounds were chloro derivatives **16** and **23** with single digit micromolar activity against CCRF-CEM. In general, most of the arabinonucleosides **17**, fluoroarabinonucleosides **9** and fluororibonucleosides **23** are much less cytotoxic than corresponding ribonucleosides. Conclusions =========== We synthesized 3 sets of sugar modified pyrimidoindole nucleosides -- arabino-, fluororibo- and fluoroarabino nucleosides bearing various substituents (amino, methoxy, methylsulfanyl, methyl, 2- and 3-furyl, 2- and 3-thienyl and phenyl) of the heterocyclic base 4-position. The synthesis started with the preparation of key-intermediate 4,6-dichloropyrimidoindole nucleosides, followed by the introduction of the substituent in the last step either by aromatic nucleophilic substitution or palladium catalyzed cross-coupling reactions. In cytostatic activity screening, the sugar-modified derivatives displayed only low activity (compared to ribonucleosides). Some fluoroarabino nucleosides displayed double-digit micromolar anti-dengue and anti-RSV activity. The most interesting result came from anti-HCV screening, where all fluoroarabino- and fluororibonucleosides showed single-digit micromolar activity and no cytotoxicity at maximum tested concentration (44 μM). These compounds are not only more potent than most of the corresponding ribonucleosides,[@cit9],[@cit10] but they are also much more selective. Abbreviations ============= TPPTS : Triphenylphosphine-3,3′,3′′-trisulfonic acid trisodium salt TDA-1 : Tris\[2-(2-methoxyethoxy)ethyl\]amine DAST : (Diethylamino)sulfur trifluoride Conflicts of interest ===================== The authors declare no competing interests. Supplementary Material ====================== Supplementary information ###### Click here for additional data file. This work was supported by the Academy of Sciences of the Czech Republic (RVO 61388963 and the Praemium Academiae award to M. Hocek), by the Czech Science Foundation (16-0011785), and by Gilead Sciences, Inc. The authors thank Dr. Gina Bahador and Dr. Joy Feng (Gilead Sciences) for the anti-HCV and anti-RSV screening. [^1]: †Electronic supplementary information (ESI) available: Experimental part and characterization data for all new compounds, table with HPLC purities of final compounds, details of biological assays and copies of NMR spectra. See DOI: [10.1039/c7md00319f](10.1039/c7md00319f)

1. Introduction {#sec1-genes-11-00676} =============== Cardiovascular diseases (CVDs) are the leading cause of mortality in the world, accounting for almost one-third of deaths worldwide \[[@B1-genes-11-00676]\]. Among the deaths caused by CVDs, ischemic heart disease accounted for 42.5%, and the primary cause is coronary atherosclerosis \[[@B1-genes-11-00676],[@B2-genes-11-00676]\]. Coronary atherosclerosis begins with intimal hyperplasia near the bifurcation of the coronary artery, and then causes stenosis or obstruction of the vascular lumen, eventually leading to fatal cardiovascular events such as myocardial infarction \[[@B3-genes-11-00676],[@B4-genes-11-00676]\]. Clinically, visual stenosis of 50--70% by coronary angiography was defined as the critical lesion of coronary atherosclerosis, which is a common pathological phenotype of the disease \[[@B5-genes-11-00676],[@B6-genes-11-00676]\]. Although the degree of coronary stenosis in patients with critical lesions is similar, the prognosis varies greatly. Some patients remained in a stable state for a long time, while others deteriorated rapidly, leading to major adverse cardiovascular events. Therefore, it is extremely important to develop novel methods and biomarkers to make risk stratification more accurate and identify patients at high risk of adverse events as early as possible. In recent years, systems biology approaches have been widely applied in the studies of cardiovascular-related diseases \[[@B7-genes-11-00676],[@B8-genes-11-00676],[@B9-genes-11-00676],[@B10-genes-11-00676]\]. The progression of coronary atherosclerosis is a chronic and nonlinear process, which involves complicated dynamic regulations in biomolecular networks. For human coronary atherosclerosis, the progression process can be generally divided into three stages---normal state, predisease state (i.e., tipping point) and disease state (major adverse cardiovascular events, e.g., myocardial infarction). The system will rapidly deteriorate into an irreversible disease state just after the tipping point. Therefore, for early prediction and prevention of adverse cardiovascular events, it is critical to identify the predisease state of the coronary atherosclerosis, especially accurate to the individual level. Based on the dynamic evolution characteristics of complex disease, we developed the theory of dynamic network biomarkers (DNBs) to identify the tipping point and leading molecular networks during the progression of a complex disease \[[@B11-genes-11-00676]\], which has been successfully applied in many diseases \[[@B12-genes-11-00676],[@B13-genes-11-00676]\]. However, to detect the tipping point of each patient, the original DNB method requires multiple samples, which are not available generally for individual patients in clinical practices. In addition, it is not trivial to computationally determine DNB members and DNB module size. To solve those problems, the landscape DNB (l-DNB) method was developed to identify the tipping point of diseases from a single sample \[[@B14-genes-11-00676]\]. Based on the three criteria of traditional DNB theory and single-sample network, the l-DNB method can evaluate the local DNB score for molecule by molecule in a sample, and then compile all of the local DNB scores into a landscape of this sample. Therefore, the l-DNB method can be applied to identify the tipping point of human coronary atherosclerosis at the individual level. In this study, we applied the local DNB score of molecules in each sample by using metabolomics data from different stages of coronary atherosclerosis patients. We not only identified the tipping point of the disease, but also predicted the criticality (i.e., early warning signals) of each patient and exactly identified the patients at high risk of adverse cardiovascular events verified by the follow-up information of these patients over the years. Furthermore, we also discovered a group of molecular network biomarkers. Evaluating them as a network by second-order statistics (e.g., correlation and deviation of these molecules) but not by the traditional first-order statistics (e.g., their average values) to assess the disease status, will provide new insights into the discovery of novel network biomarkers. 2. Methods {#sec2-genes-11-00676} ========== 2.1. Patient Information and Study Design {#sec2dot1-genes-11-00676} ----------------------------------------- Patients in this study received selective coronary angiography between June 2011 and March 2015 in Fuwai Hospital (Chinese Academy of Medical Sciences, Beijing, China). We excluded the patients with rheumatic heart disease, cardiomyopathy and other organic heart diseases, and also excluded the patients with severe liver and renal dysfunction, severe infectious diseases, malignant tumors, immune system diseases, connective tissue diseases, hyperthyroidism, Cushing syndrome and other metabolic diseases. Informed consent was obtained from all study participants. This study was performed complying with the Declaration of Helsinki and was approved by the Ethics Committee of Fuwai Hospital. Forty-eight patients were divided into three groups according to their coronary angiographic results. Nineteen patients with stenosis of coronary arteries \< 20% were regarded as the control group (i.e., reference samples). Among them, about 90% of people had no coronary stenosis diagnosed by coronary angiography. Fifteen age- and sex-matched patients with stenosis of the coronary arteries between 50% and 70% were sorted into the group of Stage A. In group of Stage B, 14 patients were diagnosed with acute myocardial infarction (AMI). The follow-up information of the patients was collected in August 2019. One patient (ID: XJ608) in Stage A group received revascularization in September 2018 due to cardiovascular disease. The study design is shown in [Figure 1](#genes-11-00676-f001){ref-type="fig"}a. 2.2. Serum Collection and Preparation {#sec2dot2-genes-11-00676} ------------------------------------- Fasting plasma samples (4 mL) were collected before coronary angiography in heparinized tubes and centrifuged within one hour of collection (4 °C, 10 min at 2300 rpm). The plasma sample was then separated into aliquots and immediately frozen at −80 °C for metabolomics analysis. A total of 400 μL of extraction solvent (V methanol: V acetonitrile = 1:1) was added to 100 μL of plasma thawed at 4 °C, followed by incubation for 1 h at −20 °C to precipitate proteins. The mixture was then centrifuged at 12,000 rpm for 15 min at 4 °C, and the supernatant (425 μL) was transferred into a new eppendorf tube. After drying and re-dissolution in acetonitrile:water (1:1), 60 μL of supernatant was transferred into a 2 mL glass vial for liquid chromatograph-mass spectrometer (LC-MS). Quality control samples were prepared by pooling 10 μL supernatant from each sample. 2.3. Metabolomics Study {#sec2dot3-genes-11-00676} ----------------------- The metabolomics of plasma was performed on a UHPLC system (1290, Agilent Technologies, Santa Clara, CA, USA) with a UPLC BEH Amide column (1.7 μm 2.1 ×\* 100 mm, Waters) coupled to TripleTOF 6600 (Q-TOF, AB Sciex, Foster City, CA, USA). MS raw data files (wiff) were converted to the mzXML format by using ProteoWizard software (AB Sciex, Foster City, CA, USA). Retention time alignment, peak discrimination, data filtering, alignment and matching were carried out by using R package XCMS (version 3.2), which generated a data matrix that consisted of the retention time (RT), mass-to-charge ratio (*m*/*z*) values, and peak intensity. Then, CAMERA R package was used for peak annotation. An in-house MS2 database was applied to identify the metabolites. We obtained 1918 annotated metabolites and 1016 unknown analytes in the positive and negative ion model, and chose the annotated metabolites for subsequent analysis. The partial least square discriminant analysis (PLSDA) of the metabolomics data was conducted by the mixOmics R package. 2.4. l-DNB Analysis {#sec2dot4-genes-11-00676} ------------------- The l-DNB method we used for detecting the individual early-warning signals in complex disease was reported previously \[[@B14-genes-11-00676]\] with slight modification in this study. Briefly, the DNB identification using the l-DNB method was based on the three criteria of previous DNB theory \[[@B11-genes-11-00676]\]. As shown in [Figure 2](#genes-11-00676-f002){ref-type="fig"}a, when a biological system approaches the predisease state from a normal state, a group of biomolecules (i.e., DNB module) satisfies the following three statistical conditions \[[@B13-genes-11-00676],[@B15-genes-11-00676],[@B16-genes-11-00676],[@B17-genes-11-00676]\]: The deviation of each molecule inside the module ($SD_{in}$, standard deviation) fluctuate strongly;The correlation among molecules inside the module ($PCC_{in}$, Pearson correlation coefficients in absolute values) dramatically increases; andThe correlation of the molecules between the inside and outside of this module ($PCC_{out}$, Pearson correlation coefficients in absolute values) dramatically decreases. \(1\) Construction of SSNs In order to calculate the DNB at individual level, we need to construct a single-sample network (SSN) for the given samples first. The theoretical principle of the SSN method has been reported before \[[@B18-genes-11-00676]\]. Briefly, based on a group of reference samples (*n* samples), a reference network can be constructed by the correlations between molecules using the abundance data of this group. The PCC between molecules *x* and *y* in the data of the *n* reference samples (i.e., samples in control group) can be calculated as $$PCC_{n}{({x,\ y})} = \frac{\sum_{i}^{n}{({x_{i} - {\overline{x}}_{n}})}{({y_{i} - {\overline{y}}_{n}})}}{\sqrt{\sum_{i = 1}^{n}{({x_{i} - {\overline{x}}_{n}})}^{2}}\sqrt{\sum_{i = 1}^{n}{({y_{i} - {\overline{y}}_{n}})}^{2}}},$$ where $x_{i}$ and $y_{i}$ are the values of molecules $x$ and $y$ in the *i*th reference sample among the *n* reference samples, respectively. And ${\overline{x}}_{n}$ and ${\overline{y}}_{n}$ are the respective average values for molecules *x* and *y* in the *n* reference samples. When one new sample d is added to the *n* reference samples, the PCC between molecules *x* and *y* in the data of (*n*+1) samples (one new sample + the original *n* reference samples) can be calculated as $$PCC_{n + 1}{({x,\ y})} = \frac{\sum_{i}^{n + 1}{({x_{i} - {\overline{x}}_{n + 1}})}{({y_{i} - {\overline{y}}_{n + 1}})}}{\sqrt{\sum_{i = 1}^{n + 1}{({x_{i} - {\overline{x}}_{n + 1}})}^{2}}\sqrt{\sum_{i = 1}^{n + 1}{({y_{i} - {\overline{y}}_{n + 1}})}^{2}}},$$ where $x_{i}$ and $y_{i}$ are the values of molecules $x$ and $y$ in the *i*th sample among the (*n* + 1) samples, respectively. And ${\overline{x}}_{n + 1}$ and ${\overline{y}}_{n + 1}$ are the respective average values for molecules *x* and *y* in the (*n* + 1) samples. The influence of the new sample d is mainly reflected in the changes of PCC. Therefore, the differential PCC between molecules *x* and *y* for sample d against the *n* reference samples, is defined as $$sPCC_{n}{({x,\ y})} = PCC_{n + 1}{({x,\ y})} - PCC_{n}{({x,\ y})}.$$ The $sPCC_{n}{({x,\ y})}$ follows the volcano distribution \[[@B18-genes-11-00676]\], which approximates a normal distribution when *n* is large enough. Thus, we can use a statistical hypothesis test (*Z*-test or *U*-test) to evaluate whether the molecules *x* and *y* are significantly correlated at the single-sample level. If the *p*-value \< 0.05, *x* and *y* are considered to have significant correlation. \(2\) Calculating the local DNB score for each molecule in a single sample Based on the sample-specific network, we defined the target molecule and its first-order neighbors as a local module. According to the three statistical conditions of DNB theory, the local DNB score for each molecule in sample d can be defined as follows:$$I_{s}{(x)} = sAD_{in}\frac{sPCC_{in}}{sPCC_{out}},$$ where $I_{s}{(x)}$ is the score of the local module of molecule *x* in the single sample *d*. The variables in the equation are calculated as follows: Corresponding to the $SD_{in}$ in DNB theory, we define the single-sample Abundance Deviation ($sAD$) as follows:$$sAD{(x_{d})} = |x_{d} - \overline{x}|,$$ where $x_{d}$ is the abundance of molecule *x* in the new sample *d* and $\overline{x}$ is the average abundance of molecule *x* in the *n* reference samples. Then, we infer the following relational expression: $$sAD_{in} = \frac{1}{1 + n_{x_{d}}}\left\lbrack {sAD{(x_{d})} + \sum\limits_{y_{d}\epsilon N_{x_{d}}}sAD{(y_{d})}} \right\rbrack,$$ which represents the average differential deviation in abundance of all $({1 + n_{x_{d}}})$ molecules in the local module of molecule *x* for sample d against the n reference samples. Furthermore, $N_{x_{d}}$ is a molecule set with $n_{x_{d}}$ molecules, which is the first-order neighbors of the molecule x. Next, we calculated the correlation between molecules within the module and the correlations between inside and outside of the module, which were named as $sPCC_{in}$ and $sPCC_{out}$, respectively. The $PCC_{in}$ is defined as $$sPCC_{in} = \frac{1}{n_{x_{d}}}\sum\limits_{y_{d}\epsilon N_{x_{d}}}sPCC_{n}{({x_{d},\ y_{d}})},$$ where $sPCC_{in}$ is the average value of $sPCC_{n}$ between molecule $x_{d}$ and all its first-order neighbors, $N_{xd}$. Furthermore, $$sPCC_{out} = \frac{1}{n_{x_{d}}m_{x_{d}}}\sum\limits_{\substack{y_{d}\epsilon N_{x_{d}} \\ z_{d}\epsilon M_{x_{d}}}}sPCC_{n}{({y_{d},\ z_{d}})},$$ where $sPCC_{out}$ is the average of $sPCC_{n}$ between the first-order neighbors and second-order neighbors of molecule $x_{d}$. $M_{x_{d}}$ is a molecule set with $m_{x_{d}}$ molecules, which is the number of second-order neighbors of molecule $x_{d}$. All the above modules and molecules are based on SSN. It is worth noting that we only considered the molecules that have at least three first-order neighbors and one second-order neighbor in the network topology. The first-order neighbors and the second-order neighbors are disjoint molecular sets. Additionally, when calculating the DNB scores of the control group, we used the same reference network as the Stage A and B groups. 2.5. Pathway Enrichment of l-DNB Molecules and Related Genes {#sec2dot5-genes-11-00676} ------------------------------------------------------------ The pathway enrichment analysis for l-DNB molecules was performed by MetaboAnalystR 3.0 R package. The canonical pathway analysis of the l-DNB molecule-related genes was carried out by Ingenuity Pathway Analysis (IPA) software (QIAGEN^®^ Bioinformatics, Germantown, MD, USA). 2.6. Heatmap of Local DNB Molecule Related Genes {#sec2dot6-genes-11-00676} ------------------------------------------------ The heatmap of local DNB molecule-related genes was conducted by the gplots R package. The gene expression data used in the heatmap were RNA-Seq data of peripheral blood mononuclear cells from the patients that received selective coronary angiography between June 2011 and March 2015 in Fuwai Hospital. Twenty-four samples (8 samples per group by random sampling) were sequenced on an Illumina platform with paired end 150 bp. The clean data filtered from raw data were aligned to the Human reference genome ENSEMBL GRCh38.p13 using HISAT2 program, and then assembled and quantified by using StingTie software. The average log2 fold change value between groups was analyzed by DESeq2 R package. 2.7. Statistical Analysis {#sec2dot7-genes-11-00676} ------------------------- Continuous data were expressed as mean ± SD and compared using the ANOVA test. Categorical data were expressed as count (percentage) and compared using the chi-square or Fisher exact test. *p*-Value \< 0.05 was considered statistically significant. The statistical result of global DNB scores from all samples is presented as mean ± SEM using GraphPad Prism 5.0 software (GraphPad Software Inc., San Diego, CA, USA) ([Figure 3](#genes-11-00676-f003){ref-type="fig"}a). The bar plots and line chart of statistical results in this study are displayed by GraphPad Prism 5.0 software as well ([Figure 3](#genes-11-00676-f003){ref-type="fig"}b,c). 3. Results {#sec3-genes-11-00676} ========== 3.1. Global Plasma Metabolomics Profile of Coronary Atherosclerosis Patients {#sec3dot1-genes-11-00676} ---------------------------------------------------------------------------- To detect the early-warning signals during the progression of human coronary atherosclerosis, the plasma was collected from patients with different pathological states. As shown in [Figure 1](#genes-11-00676-f001){ref-type="fig"}a, according to the coronary angiography results of patients, the samples were generally divided into three groups---control group (the stenosis of coronary artery is less than 20%), Stage A group (the stenosis of coronary artery is between 50% and 70%) and Stage B group (diagnosed as AMI). The baseline characteristics are shown in [Table 1](#genes-11-00676-t001){ref-type="table"}. There was no significant difference among the three groups in baseline characteristics. To get the global metabolomics profile during the progression of coronary atherosclerosis, these 48 plasma samples were detected on Ultra High-Performance Liquid Tandem Chromatography Quadrupole Time of Flight Mass Spectrometry (UHPLC-QTOFMS). After a series of data preprocessing ([Figure 1](#genes-11-00676-f001){ref-type="fig"}a), we finally identified about 1900 annotated metabolites by relative quantification, including both positive and negative ion models. Through the partial least squares discriminant analysis (PLSDA) result of metabolomics data ([Figure 1](#genes-11-00676-f001){ref-type="fig"}b), we found that Stage A group was quite different from the other two groups in terms of metabolites, which indicated that there might exist a tipping point during the nonlinear process of coronary atherosclerosis progression. 3.2. The Tipping Point of Each Individual during the Progression of Human Coronary Atherosclerosis by l-DNB {#sec3dot2-genes-11-00676} ----------------------------------------------------------------------------------------------------------- The metabolomics profile and previous evidence suggested that the progression of human coronary atherosclerosis is a nonlinear process, and there exists a tipping point during the disease development. The traditional methods based on differentially expressed genes are limited by their static characteristics, which fails to distinguish the predisease state from the normal state. Thus, we turn to the DNB theory ([Figure 2](#genes-11-00676-f002){ref-type="fig"}a), which was developed for identifying the tipping point of complex diseases and discovering the critical networks \[[@B11-genes-11-00676],[@B15-genes-11-00676],[@B19-genes-11-00676]\]. When the system approaches the tipping point, a dominant group of molecules, named DNBs, satisfies the three statistical conditions (i.e., strong fluctuation, high correlation of internal molecules and low correlation with external molecules) (See Methods---l-DNB analysis). Based on these, it is available for us to identify the tipping point of human coronary atherosclerosis and discover the potential network biomarkers. However, even with the same pathological phenotype, there are great differences among individuals. Moreover, the DNB method identifies the tipping point of the whole system by using multiple samples, which is unavailable for personal prediction. To solve these problems, the new method, l-DNB, was developed ([Figure 2](#genes-11-00676-f002){ref-type="fig"}b). First, by using the metabolomics data, we constructed a single-sample network (SSN) for given samples based on the reference samples (samples in the control group). Next, according to the three criteria of DNB theory, we calculated the local DNB score of each molecule in each sample. Finally, the molecules in each sample were sorted by their local DNB score, and the global DNB score for each sample was the mean value of all local DNB scores in this sample. The detailed information of the calculation can be found in Methods. According to the above calculation, we obtained a landscape of local DNB molecule scores for each sample. As shown in [Figure 3](#genes-11-00676-f003){ref-type="fig"}a, the average value of global DNB scores in the Stage A group was significantly higher than that in other two groups, which means that the tipping point of human coronary atherosclerosis is at Stage A. This result coincides with the critical stage defined in clinical medicine. Then, we defined the person with a global DNB score greater than 1.0 as the patient at high risk of adverse cardiovascular events. We found that most of the patients (13 out of 15 persons) in the Stage A group were accurately predicted to be the high-risk patients, which means the true positive rate (TPR) is 86.76% ([Figure 3](#genes-11-00676-f003){ref-type="fig"}b). Meanwhile, the predicted high-risk patients in the Control group and Stage B group were only five and two persons, respectively. Thus, the false positive rate (FPR) is only 21.21%. The global DNB score of each sample can be found in [Table S1](#app1-genes-11-00676){ref-type="app"}. These results demonstrated that the l-DNB method can accurately identify the patient who is at the critical state (tipping point). Although the patients in the Stage A group received the corresponding treatment after the coronary angiography, the patient (ID: XJ608) still underwent major cardiovascular events according to the follow-up information collected in August 2019. Therefore, taking the XJ608 patient for example to test the effectiveness and accuracy of our analysis results, we evaluated the composition of local DNB molecules and the robustness of prediction. First, we respectively selected the top five (or top 10, top 15, top 20, ..., top 500, i.e., total 100 times) local DNB molecules to calculate the global DNB score of each sample. In this way, every sample had 100 global DNB scores corresponding to different molecular sample sizes. Then, we compared the global DNB score of "XJ608" patient with the corresponding global DNB scores of other samples, and finally got the ranking distribution of "XJ608" in 100 times of such sampling. The statistic result of the sample "XJ608" ranking showed that the probability of sample "XJ608" ranking in the top three relative to all samples is 0.52 and the probability of ranking in top five is 0.91 ([Figure 3](#genes-11-00676-f003){ref-type="fig"}c), which implied that XJ608 patient had a high risk of morbidity. This result further confirms that the l-DNB method can personalize the tipping point of coronary atherosclerosis accurately. To further investigate the mechanism of molecule networking that drives the system as it approaches the tipping point, we selected the molecules with local DNB scores in the top 50 from each sample of the Stage A group, and these molecules were also included in at least three samples of the Stage A group. Through the pathway enrichment analysis of these DNB molecules ([Figure 3](#genes-11-00676-f003){ref-type="fig"}d), we found that these molecules were mainly enriched in pathways related to lipid metabolism, fatty acid metabolism and amino acid metabolism, which implied that these enriched metabolic pathways played important roles at the tipping point of human coronary atherosclerosis. Among the DNB molecules mentioned above, 16 l-DNB molecules were candidate molecules (top 50) in at least half of the Stage A samples ([Figure 4](#genes-11-00676-f004){ref-type="fig"}a). In addition, these molecules include many phospholipids and fatty acids. Therefore, we extracted the related genes involved in the metabolism of these molecules using the KEGG database, and obtained their dynamic expression in the three stages using the RNA-Seq data of peripheral blood mononuclear cells in coronary atherosclerotic patients ([Table S2](#app1-genes-11-00676){ref-type="app"}). Through the dynamic expression of the l-DNB molecule-related genes, we found that most of the genes reversed their expression trend after the tipping point ([Figure 4](#genes-11-00676-f004){ref-type="fig"}b). Moreover, the canonical pathway enrichment results of these genes showed that most of the enriched pathways were activated before the tipping point (Stage A group compared with the Control group), except one pathway (antioxidant Action of Vitamin C) that was suppressed ([Figure 4](#genes-11-00676-f004){ref-type="fig"}c). After the tipping point (Stage A group compared with Stage B group), the states of these pathways were reversed. These results suggested that lipid metabolism-related genes and molecules played the key roles in driving the coronary atherosclerosis approaching the tipping point. 4. Discussion {#sec4-genes-11-00676} ============= The progression of human coronary atherosclerosis is a nonlinear dynamic process, which is influenced by many factors, such as heredity, environment and living habits, etc. \[[@B20-genes-11-00676],[@B21-genes-11-00676]\]. Due to the complexity of the disease, patients at a high risk of adverse cardiovascular events cannot be effectively identified by the degree of vascular stenosis \[[@B22-genes-11-00676]\]. Therefore, we must develop new methods for the prediction and diagnosis of an individual patient. The DNB method is able to detect the critical states during disease progression or biological processes, and its effectiveness has been studied in many papers from both experimental and computational viewpoints \[[@B12-genes-11-00676],[@B13-genes-11-00676],[@B23-genes-11-00676],[@B24-genes-11-00676],[@B25-genes-11-00676]\], e.g., cell fate decision \[[@B23-genes-11-00676],[@B24-genes-11-00676]\], immune checkpoint blockade \[[@B25-genes-11-00676]\], pulmonary metastasis \[[@B13-genes-11-00676]\] and steatohepatitis transition \[[@B12-genes-11-00676]\]. In this study, using an untargeted metabolomics approach, we not only accurately identified the tipping point during the progression of human coronary atherosclerosis, but also precisely assessed the risk of adverse events on individual patients by using the l-DNB method. The metabolomics profiles of plasma enable us to systematically study the mechanism of molecular dynamics in pathogenesis under noninvasive conditions, which are widely used to identify novel biomarkers in the cardiovascular field \[[@B26-genes-11-00676],[@B27-genes-11-00676]\]. Using the plasma metabolomics data at different stages, we identified that the tipping point during the progression of human coronary atherosclerosis is at Stage A, which corresponds to the critical stage defined in clinical medicine \[[@B5-genes-11-00676],[@B6-genes-11-00676]\]. More importantly, we accurately identified the early-warning signals for individual patients by their global DNB scores with the TPR of prediction 86.67%, and the FPR only 21.21%. Due to limited sample size, there was only one patient (ID: XJ608) who underwent a major cardiovascular event after medication in the Stage A group. However, in the ranking of patients that were predicted to be at high risk of adverse events, the probability that the XJ608 patient ranked in the top five in 100 times of sampling was 0.91. This result also strongly proves that the l-DNB method can accurately personalize the early-warning signals of coronary atherosclerotic patients. Meanwhile, we discovered a group of DNB molecules that may play driving roles in the molecular regulatory networks during the progression of coronary atherosclerosis. Many of them have been reported to be involved in lipid metabolism and fatty acid metabolism, which validates our results. Phosphatidylglycerol (PG) and phosphatidylethanolamine (PE), the subclasses of glycerophospholipids, have been reported to be observed in plasma lipoproteins and atherosclerotic plaques \[[@B28-genes-11-00676]\]. Depending on the species of the bound fatty acids, different kinds of glycerophospholipids play different roles in biological processes \[[@B28-genes-11-00676],[@B29-genes-11-00676]\]. Although some kinds of PEs have been reported to be closely involved in the progression of atherosclerosis \[[@B30-genes-11-00676],[@B31-genes-11-00676]\], the functions of other kinds of PEs remain concealed. Current studies demonstrate that PGs are the important participants in the signal transduction and stress response \[[@B32-genes-11-00676]\], but their roles in atherosclerotic progression are still unclear. In our result, PG (16:1/16:1) and PG (18:3/20:4) are DNB molecule candidates in 14 samples of the Stage A group (15 samples in total), which reveals the importance of PGs in the progression of coronary atherosclerosis, and their specific molecular mechanisms need to be further studied. Palmitic acid, a molecule involved in fatty acid metabolism, has been reported to be intimately related to cardiovascular disease \[[@B29-genes-11-00676],[@B33-genes-11-00676],[@B34-genes-11-00676]\]. The metabolic dynamic changes caused by palmitic acid intake can influence the levels of total cholesterol and low-density lipoprotein (LDL) cholesterol, which show a strong positive correlation with the development of coronary heart disease \[[@B35-genes-11-00676],[@B36-genes-11-00676],[@B37-genes-11-00676]\]. Our study revealed that palmitic acid is an important DNB molecule, and its dynamic fluctuation as well as the participating regulatory network can drive the progression of coronary atherosclerosis. However, the underlying molecular mechanism needs to be further investigated. More importantly, for the first time, we took these metabolites as a network biomarker (by using their second-order statistics for disease prediction, e.g., correlation and deviation of these molecules). Although these metabolites have been studied separately, taking them as a network (evaluating them as a network, e.g., not by the traditional first-order statistics but by the second-order statistics) to assess the disease status will provide new insights into the discovery of novel network biomarkers. In conclusion, based on the l-DNB method, our study identified the tipping point of human coronary atherosclerosis at Stage A, which was consistent with the clinically defined critical stage. More importantly, we detected the early-warning signal of each patient and accurately identified the patients at high-risk of adverse events. These not only provide a new noninvasive method for the early prediction of adverse cardiovascular events, but also contribute to the realization of personalized diagnosis. Furthermore, we also discovered a group of DNB molecules that play driving roles in the development of the disease, which provide a new insight into the molecular network mechanisms of human coronary atherosclerosis. This research is supported by The Natural Science Foundation of China (Nos. 31930022, 31771476, 81870277). The following are available online at <https://www.mdpi.com/2073-4425/11/6/676/s1>, Table S1: Global_DNB_score_of_each_sample.xlsx. Table S2: Foldchange_of_DNB_molecules_related_genes.xlsx. ###### Click here for additional data file. Conceptualization, L.C., K.D., J.C., X.L., J.G. and C.S.; data collection, analysis and interpretation, J.G., C.S. and C.Z.; writing---original draft preparation, J.G., C.Z. and C.S.; writing---review and editing, X.L. and L.C.; funding acquisition, L.C. and K.D. All authors have read and agreed to the published version of the manuscript. This research was funded by the National Key R&D Program of China (No. 2017YFA0505500), the Strategic Priority Research Program of the Chinese Academy of Sciences (CAS) (No. XDB38000000), Shanghai Municipal Science and Technology Major Project (Grant No. 2017SHZDZX01) and CAMS Innovation Fund for Medical Sciences (CIFMS) (2016-I2M-1-009). The authors declare no conflict of interest. {#genes-11-00676-f001} {#genes-11-00676-f002} {#genes-11-00676-f003} {#genes-11-00676-f004} genes-11-00676-t001_Table 1 ###### Baseline Characteristics. -------------------------------------------------------------------------------------------- Control\ Stage A\ Stage B\ *p*-Value *n* = 19 *n* = 15 *n* = 14 ----------------------------- ---------------- ---------------- ---------------- ----------- Age (years) 58.53 ± 8.64 61.38 ± 9.68 61.50 ± 9.35 0.563 Female 9 (47.4%) 7 (43.8%) 7 (50%) 0.942 BMI (kg/m^2^) 27.22 ± 7.95 25.12 ± 4.46 25.99 ± 2.49 0.558 SBP (mmHg) 126.32 ± 12.11 126.13 ± 13.29 123.86 ± 16.43 0.863 DBP (mmHg) 78.95 ± 11.97 73.44 ± 7.69 75.50 ± 11.33 0.305 Diabetes mellitus 3 (15.8%) 2 (12.5%) 5 (35.7%) 0.236 Hypertension 11 (57.9%) 12 (75%) 10 (71.4%) 0.521 Dyslipidemia 10 (52.6%) 12 (75%) 12 (85.7%) 0.207 Family history of CAD 3 (15.8%) 6 (37.5%) 4 (28.6%) 0.343 Premature CAD 2 (10.5%) 4 (25%) 1 (7.1%) 0.316 Cerebrovascular disease 0 (0.0%) 1 (6.3%) 4 (28.6%) 0.061 Peripheral vascular disease 3 (15.8%) 3 (18.8%) 1 (7.1%) 0.638 -------------------------------------------------------------------------------------------- BMI: body mass index; SBP: systolic blood pressure; DBP: diastolic blood pressure; CAD: coronary artery disease. [^1]: Equally contributed to this work.

Sir, The neural integrity monitor (NIM) electromyogram (EMG) tracheal tube is frequently being utilized, for the head and neck procedures, when laryngeal nerves (LNs) could be injured during the process of surgical dissection. It is particularly useful for identifying the recurrent LN.\[[@ref1]\] The anesthesiologist should be familiar with the proper use and functioning of this device. Initially, this unique tracheal tube must be positioned so that its color-coded contact band is appropriately placed between the vocal cords.\[[@ref2]\] When attempting to identify LNs, a stimulating electrical current of 0.5-2.0 mA is used by the surgeon. This current is administered via a sterile probe, which is placed directly on the anatomical site in question. Additionally, return electrodes are positioned in the skin above the sternum. When a LN is located, an electrical signal is subsequently generated by the motion of the vocal cords. An audibly recognizable "machine gun click" is then produced from the device\'s associated monitor. This sound has a set frequency of 4 times/s (4 Hz). Simultaneously, an oscilloscope-like screen displays an identifiable sinusoidal response. Depending on clinical conditions, either direct laryngoscopy, video laryngoscopy, or fiber-optic intubation can be utilized for airway management. However, tracheal intubation in many of these patients may be difficult; given their concomitant head and neck disease. Thus, the glottis may not be midline or the trachea may be compressed. The presence of head and neck tumors, or prior radiation treatment, could also produce limited neck extension. Inadequate subluxation of the mandible, obesity, or reduced thyro-mental distance may also be present. Accordingly, the anesthesiologist may wish to use fiber-optic intubation. It should be noted that the use of certain associated techniques, such as nebulized or trans-tracheal lidocaine, or superior LN blocks, may interfere with the performance of this tracheal tube with respect to the surgical localization of the LNs. As these tracheal tubes have a minimum outer diameter of at least 8.8 mm, oral intubation is necessary. Furthermore, the conical connector, located on the proximal aspect of this tracheal tube, is non-removable. For these reasons, the use of the Patil-Syracuse fiber-optic-compatible oral airway (FCOA) may be advantageous. As shown in [Figure 1](#F1){ref-type="fig"}, this FCOA is mechanically compatible with the NIM EMG tracheal tube. It also has an anterior channel which greatly facilitates intubations which may be difficult secondary to an anterior-oriented glottis. Moreover, this channel is relatively shallow and will allow the fiberscope to readily move off-midline in the event that the trachea is deviated. Once, the glottis is identified, and the fiberscope advanced into the trachea, the Patil-Syracuse airway is readily removed from the patient\'s oropharynx before advancement of the NIM EMG tracheal tube.\[[@ref3]\] {#F1} Furthermore, the use of non-depolarizing neuromuscular blocking agents is contraindicated when using this tracheal tube. To facilitate tracheal intubation, a depolarizing agent may be required. Pre-paralytic, as well as post-paralytic, assessment of train-of-four monitoring should be carried out. This documents the return of neuro-muscular function to that of the patient\'s baseline. Prior to placement, the cuff of the NIM EMG tracheal tube should be covered with an aqueous lubricant rather than a local anesthetic gel. It is also advisable to use a low FiO~2~ given the associated potential for an electrical fire. Lastly, it should also be noted that this particular tracheal tube is not magnetic resonance imaging compatible. Consideration should also be made to use a cook airway exchange catheter (CAEC) when extubating a patient with a "difficult airway" requiring a NIM EMG tracheal tube. CAECs are also mechanically compatible with these tracheal tubes and can function as a stent; facilitating re-intubation. Furthermore, they also allow for jet ventilation in the event that re-intubation cannot be readily accomplished.\[[@ref4]\] In conclusion, the NIM EMG tube allows for intraoperative localization of the LNs during the surgical dissection. Proper understanding of its structure and function is an essential for the anesthesiologist.

Introduction ============ The presence of thyroglobulin autoantibodies (TgAb) in serum samples of patients with differentiated thyroid carcinoma (DTC) is a serious technical problem affecting the use of thyroglobulin (Tg) as a tumour marker, because TgAb interference may result in an underestimation of serum Tg concentrations. The prevalence of TgAb in patients with DTC is much higher (20--30%) than in the general population ([@b1-biochem-med-22-3-365-11]--[@b5-biochem-med-22-3-365-11]). This high prevalence, together with studies showing that even low levels of TgAb can cause falsely low Tg values may explain why all specimens need sensitive TgAb screening. However, current TgAb assays are qualitatively and quantitatively variable and TgAb concentrations obtained with different methods may show 100-fold differences in the same specimen ([@b5-biochem-med-22-3-365-11],[@b6-biochem-med-22-3-365-11]). The manufacturers of current TgAb assays claim that the secondary standards included in their kits are calibrated against IRP-65/93 reference preparation as primary standard. The assay specific use of different secondary standards is one reason why different methods may provide different TgAb concentrations. These differences reflect not only suboptimal sensitivity and specificity but also different interactions between the patient-specific isoforms and the assay reagents (standards, tracer, antibody). These important variables explain why high TgAb concentrations do not always interfere, while in other cases low TgAb levels result in significant interference with the Tg assay ([@b3-biochem-med-22-3-365-11],[@b5-biochem-med-22-3-365-11],[@b6-biochem-med-22-3-365-11]). Because there is no reliable TgAb reference method, it is necessary to use an independent parameter for seeking the presence of interfering TgAb. The Tg recovery test, which is based on the addition of a known quantity of Tg, proved to detect less reliably TgAb in some studies, and some, although not all studies discourage its use for the follow-up of patients with DTC ([@b3-biochem-med-22-3-365-11]--[@b8-biochem-med-22-3-365-11]). It has been also shown that radioimmunoassay (RIA) is less prone to the interfering effect of TgAb than other methods such as immunometric assays (IMA). In general, immunometric assays give lower serum Tg values in the presence of TgAb as compared to those obtained with RIA ([@b3-biochem-med-22-3-365-11]--[@b6-biochem-med-22-3-365-11]). In some studies the interference of TgAb with Tg measurement was estimated on the basis of discordant results of Tg measurements obtained with IMA and RIA ([@b3-biochem-med-22-3-365-11]--[@b6-biochem-med-22-3-365-11]). Although recent studies confirmed that low TgAb concentrations may have an impact on the validity of Tg measurements ([@b9-biochem-med-22-3-365-11],[@b10-biochem-med-22-3-365-11]), the potential significance of TgAb concentrations within the reference range has not been elucidated or has been often neglected. In the present study we examined whether TgAb levels within or near to the reference range could interfere with the Tg assay. Materials and Methods ===================== Subjects -------- For our *in vitro* experiments we collected during a period of one year 9 serum pools with TgAb concentrations below the functional sensitivity of the TgAb assay. These serum samples were sent to the laboratory for TgAb measurement, and leftover samples were used for the experiments. All blood samples were taken under fasting condition between 08.00 and 10.00 AM. in 2 mL native vacoutainer tubes, then serum samples were aliquoted, stored at −80 °C and processed monthly. The Tg concentrations in these serum pools ranged from 7.8 to 125 ng/mL ([Table 1](#t1-biochem-med-22-3-365-11){ref-type="table"}). Serum samples (N = 134) were also collected from 27 DTC patients after thyroidectomy and radioiodine ablation (22 women and 5 men; median age, 47 years; lower and upper age quartiles, 42 and 58 years, respectively). Median follow-up time was 2.5 years (lower and upper follow-up time quartiles, 0.5 and 8.0, respectively). The number of serum samples obtained from the same patient during follow-up was between 2 and 6. None of the DTC patients had previously diagnosed Graves' disease or Hashimoto thyroiditis. All but 28 samples were obtained from patients receiving thyroxin (L-T4) treatment. Two of the 27 patients developed metastatic DTC during the follow-up period. The study was performed in the Central Laboratory, Markusovszky Teaching Hospital of County Vas, Szombathely, and was approved by the Research Ethics Committee of Markusovszky Hospital. Methods ------- Increasing concentrations of TgAb (ECLIA kit calibrator containing 3250 IU/l polyclonal TgAb produced in sheep against human Tg and solved in human serum matrix; Roche GmbH, Germany, Mannheim) were added to aliquots of the 9 TgAb-free serum pools with varying concentrations of Tg to reach TgAb concentrations within or near to the reference range ([Table 1](#t1-biochem-med-22-3-365-11){ref-type="table"}). Samples were then incubated for one hour at room temperature and assayed for Tg and TgAb. Tg was undetectable in the TgAb solution. Serum Tg and TSH concentrations were measured using electrochemiluminometric assays (ECLMA, Roche) and TgAb by an electrochemiluminescence immunoassay (ECLIA, Roche). Each method was carried out using an Elecsys 2010 automated immunochemical analyser (Roche). The cutoff value for increased TgAb provided by the manufacturer was 115 IU/mL. The within-run coefficients of variations (CV) were 8.6%, 2.1% and 1.8% using samples containing 0.034, 0.91 and 3.96 IU/mL TgAb, respectively. The inter-assay CVs measured in serum samples with target TgAb levels of 115 ± 8.3 and 62.8 ± 5.4 IU/mL were 7.2 % and 8.7%, respectively (N = 21). The analytical sensitivity provided by the manufacturer was 10 IU/mL. We determined the functional sensitivity of the method in 8 serum pools with TgAb concentrations ranging from 10 to 110 IU/mL. The CV of \> 19% was reached at 24 IU/mL and, therefore, values below this limit were considered as undetectable TgAb concentrations. The intra-assay CVs for Tg measurements were 1.8% and 1.4% in serum samples containing 4.1 and 26.9 ng/mL Tg, respectively. The inter-assay CVs were 1.8% and 3.6% for samples containing 3 and 4.2 ng/mL Tg, respectively (N = 21). The analytical sensitivity was 0.1 ng/mL. We determined the functional sensitivity of the method in 7 serum pools with Tg concentrations ranging from 0.46 to 4.6 ng/mL. The CV of \> 19% was reached at 0.46 ng/mL Tg concentration and, therefore, values below this limit were considered as undetectable Tg concentrations. The reference range for the TSH assay was between 0.27 and 4.2 mU/L, and the functional sensitivity provided by the manufacturer was 0.014 mU/L. Statistical analysis -------------------- The values are presented as median and lower and upper quartiles (25% and 75% percentiles) except for *in vitro* experiments where the results are given as means ± SD. TgAb (as independent variable) and Tg concentrations obtained from *in vitro* experiments were logarithmically (Ln) transformed then their relationship was analyzed by linear regression. Changes in Tg concentrations expressed as percent decrease from baseline after the addition of increasing concentrations of TgAb were also analyzed. The association between loss of Tg% and TgAb was examined using logarithmic regression. An equation derived from the curve that could be ordered with the best r-squared value on chart was established and used for correction of Tg levels obtained from patients with DTC who had measurable Tg (\> 0.46 ng/mL) and TgAb (\> 24 IU/L) levels with the assumption that the same TgAb interference exists in the presence of naturally occurring human TgAb. Differences between measured and calculated Tg levels and between L-T4 on and off Tg and TSH levels were assessed using non parametric paired test (Wilcoxon matched pairs test). A P value less than 0.05 was considered statistically significant. For statistical analysis Statistica for Windows '95 4.0 software package and Excel 5.0 (Microsoft) as well as SPSS for Windows (version PASW Statistics 18, USA) with 95% confidence interval were used. Results ======= As shown in [Table 1](#t1-biochem-med-22-3-365-11){ref-type="table"}, there was a substantial decrease of Tg values in each of the 9 serum pools in the presence of the lowest added TgAb which increased measured TgAb levels to 39.1 ± 5.5 IU/mL. In serum pools with baseline Tg concentrations below 11 ng/mL, the addition of TgAb that increased TgAb concentration to 75.6 ± 2.5 IU/mL resulted in undetectable Tg levels, whereas Tg values became undetectable in each of the 9 serum pools in the presence of the highest concentration of TgAb. [Table 2](#t2-biochem-med-22-3-365-11){ref-type="table"} summarises the results of regression analysis for decreasing Tg levels and loss of Tg% as depending variables. The strong relationship between TgAb and measured Tg concentrations allowed us to calculate the loss of Tg in serum pools with detectable TgAb levels ([Figure 1](#f1-biochem-med-22-3-365-11){ref-type="fig"}). Changes in Tg after the addition of increasing concentrations of TgAb were described as loss of Tg% = −0.2408 × Ln(TgAb IU/mL) + 0.1944 (r = 0.83; P \< 0.001). Of the 134 serum samples obtained from patients with DTC, TgAb concentrations were undetectable (below 24 IU/mL) in 13 sera (10%), within the reference range (between 24 and 115 IU/mL) in 106 (79%) and above the upper limit of the reference range (\> 115 IU/mL) in 15 samples (11%). There was no difference in follow-up times between TgAb positive and TgAb negative patients. The percent occurrence of undetectable Tg was similar in samples with TgAb levels within the reference range (56/106 samples, 52.7%) and in those with elevated TgAb levels (8/15 samples, 53%). There was a significant increase of both TSH from a median of 0.10 (0.05 1.27) to 28.3 (16.0 44.0) mIU/L and Tg concentrations from 0.09 (0.09 3.20) to 7.15 (0.09 80.00) ng/mL in serum samples obtained from DTC patients after L-T4 withdrawal (P \< 0.001 for both TSH and Tg). Of the 134 serum samples, 26 samples had detectable amounts of both Tg and TgAb (22 samples obtained during L-T4 treatment and 4 samples after L-T4 withdrawal). Using the equation defined in our *in vitro* experiment, calculated Tg concentrations proved to be significantly (P \< 0.01) higher than measured Tg values in samples obtained during L-T4 treatment \[N = 22; 5.53 (2.17--11.32) *vs*. 3.3 (1.6--6.9) ng/mL\], but in those after L-T4 withdrawal the difference between calculated and measured Tg concentrations was not statistically significant \[N = 4; 32.32 (10.1--258.1) *vs*. 20.4 (6.4--111.6) ng/mL\]. In 2 patients calculated but not measured serum Tg exceeded 2 ng/mL during L-T4 treatment, and a marked increase of measured serum Tg after L-T4 withdrawal (16.8 and 6.1 ng/mL) in these 2 patients indicated persistence of the disease. Discussion ========== Our study shows that serum TgAb concentrations within the reference range exert a significant effect on the Tg assay. We found that addition of increasing amounts of TgAb to serum pools to reach TgAb levels within or near the reference range produced a dose-dependent decrease of Tg levels. The majority of literature data propose that TgAb needs to be taken into consideration only in cases with significantly elevated TgAb levels ([@b2-biochem-med-22-3-365-11],[@b4-biochem-med-22-3-365-11]). The potential significance of TgAb within or near to the reference range on the Tg assay has not been considered or has been neglected, although a few studies indicated that they may have an impact on Tg measurements ([@b5-biochem-med-22-3-365-11]--[@b10-biochem-med-22-3-365-11]). Our study confirmed the frequent occurrence of TgAb in serum samples of DTC patients who underwent thyroid ablation. With the use of data obtained from our *in vitro* experiments with added TgAb, we were able to calculate the loss of Tg due to TgAb interference in serum samples with detectable Tg and TgAb obtained from these patients. We found that calculated Tg concentrations were significantly higher than measured values in serum samples containing detectable Tg and TgAb. In 2 patients the difference between measured and calculated Tg values was clinically remarkable and, indeed, serum Tg measured after L-T4 withdrawal indicated persisting disease in both cases. We believe that these observations may prove to be useful for the follow-up of patients with DTC. However, further studies are needed to explore the clinical relevance of our method of calculated Tg in monitoring these patients. An obvious drawback of our study is that the equation used for calculated serum Tg can be applied only for the assays studied. Regarding the TgAb antibody used for calibration of the TgAb assay, it is not known whether sheep TgAb raised against an excess of heterologous Tg is comparable to naturally occurring human TgAb. Rosario *et al*. used a chemiluminescence TgAb assay calibrated against human TgAb (reference value 1 IU/mL) ([@b7-biochem-med-22-3-365-11]). Görges *et al*. also used a competitive assay and human TgAb standard (cutoff value \< 50 IU/mL) ([@b8-biochem-med-22-3-365-11]). In contrast, our assay was calibrated using sheep TgAb and the reference range was higher than those used in these previous studies. In addition, a recent study showed marked discordance between the 4 TgAb methods studied suggesting that TgAb assays failed to predict precisely the Tg-TgAb interference ([@b11-biochem-med-22-3-365-11]). Inter-individual variations among TgAbs, such as distribution of immunoglobulin classes, epitope specificities and antibody avidities may substantially affect kinetics of Tg-TgAb interaction, clearance rates and biological half-lives of Tg or Tg complexed with Tg\*Ab. Such differences may result in a more complex kinetics of the Tg-TgAb interaction not reflected by the equation used for calculated serum Tg in the present study. Thus, our method fails to solve the problem of Tg-TgAb interactions, but we believe that it increases the value of cohesive biochemical findings and their interpretation in routine practice. It has been proposed that the way out of the analytical trap of determining Tg as tumour marker will probably be provided by the development of more reliable assays. In addition, several other assays have been developed to distinguish Tg produced by malignant and benign thyroid tissues, but they have not been applied for routine use ([@b12-biochem-med-22-3-365-11]). Until more reliable methods become available, we recommend that every laboratory should monitor TgAb interference of the used Tg method. The authors thank Zsuzsanna Papp and technicians of the Radioimmunoassay Division for their technical support. **Potential conflict of interest** None declared. {#f1-biochem-med-22-3-365-11} ###### Measured Tg and TgAb concentrations in aliquots of the 9 serum pools after the addition of 0, 30, 60, 120 and 180 IU/mL TgAb. **Added TgAb (IU/mL)** **Measured TgAb (IU/mL) Mean ± SD (95% CI)** **Measured Tg concentrations in serum pools (ng/mL)** ------------------------ ---------------------------------------------- ------------------------------------------------------- ----- ------ ------ ------ ------ ------ ------ ------ **0** 16.1 ± 3.8 (13.1--18.9) 7.8 11 11.9 14.5 29 50 53.3 73.8 125 **30** 39.1 ± 5.5 (34.9--43.4) 3.9 2.7 4.3 7.7 10.6 16.1 17.2 26.4 37.9 **60** 75.6 ± 2.5 (65.9--85.1) ND ND 0.8 2.1 2.6 5.7 4.5 6.7 8 **120** 131.4 ± 3.8 (121.3--138.5) ND ND ND ND ND 1.9 2.9 1.2 1.9 **180** 200.6 ± 6.2 (197.5--208) ND ND ND ND ND ND ND ND ND ND - not detectable. ###### Summary of regression analysis for decreasing Tg levels and loss of Tg% as depending variables. **Coefficients[^a^](#tfn2-biochem-med-22-3-365-11){ref-type="table-fn"}** --------------------------------------------------------------------------- -------------- ------ ------- ------- ---------- ---------- ------- ------ **1** **Constant** 8.49 0.74 11.45 \< 0.001 6.98 10.0 **LnTgAb** −1.79 0.19 −0.86 −9.58 \< 0.001 −2.16 −1.41 R^2^ = 0.74; ANOVA: F(1;33)=91.7; P\<0.001 **2** **Constant** 3.61 0.38 9.52 \< 0.001 2.84 4.38 **Loss of Tg%** 3.32 0.52 0.75 6.43 \< 0.001 2.27 4.37 R^2^ = 0.56; ANOVA: F(1;33) = 41.3; P \< 0.001 Dependent Variable: LnTg; SE, standard error; ANOVA, Analysis of variance for regression by 95% confidence interval.

1. Introduction {#s0005} =============== Oral administration is the preferred route by patients due to its convenience, price, comfort, and handling[@bib1]. The orally-administered drugs may undergo extensive first-pass metabolism in the gastro-intestinal tract, and this may result in limited systematic bioavailability, and decreased therapeutic effects[@bib2], [@bib3]. Absorption of orally administered drugs takes place primarily in the small intestine, followed by delivery to the liver *via* the portal blood. The small intestine is efficient in the absorption of a wide-spectrum of chemicals due to the high concentration of villi and microvilli in the order of duodenum, jejunum, and ileum[@bib4], abundant epithelial transporters, optimal pH for absorption, high peristalsis, high blood flow, as well as contact for a long time. Alteration or failure to maintain one of these conditions may result in lower bioavailability of the drug[@bib3], [@bib5], [@bib6]. In addition to the small intestine, the large intestine may also be important for the absorption of xenobiotics, especially oral drugs formulated for sustained release[@bib6]. The bacteria in the large intestine contain various enzymes that metabolize xenobioitcs as well as endogenous chemicals such as bile acids and dietary constituents[@bib7], [@bib8]. In addition, colon-specific oral drug-delivery systems have been utilized recently to administer a variety of therapeutic agents[@bib9]. Therefore, it is important to determine the regulation of xenobiotic biotransformation in the intestine. The drug-processing genes involved in the xenobiotic biotransformation include various phase-I and phase-II drug metabolizing enzymes, as well as uptake and efflux transporters. In general, the content of DPGs is lower in intestine than that in liver[@bib10]. DPGs play a critical role in the absorption, metabolism, disposition, elimination and detoxification of xenobiotics and other drugs[@bib11]. Phase-I enzymes catalyze hydrolysis, reduction, and oxidation of drugs. The cytochrome P450s (CYPs) in the first 3 families are among the most important phase-I enzymes that contribute to the biotransformation of the majority of xenobiotics, whereas the CYP4 family members are important for fatty acid metabolism. The NAD(P)H dehydrogenase, quinone 1 (*Nqo1*) is involved in reduction reactions, and it is a prototypical target gene of the oxidative stress sensor nuclear factor erythroid 2-related factor 2 (NRF2). Aldehyde dehydrogenases (ALDHs) are important phase-I enzymes for the detoxification of aldehydes, which are the metabolites of alcohols. Phase-II metabolism refers to the conjugation reactions that generally increase the water-solubility of substrates to form more polar compounds with exceptions. The three major classes of phase-II enzymes include UDP-glucuronosyltransferases (UGTs), sulfotransferases (SULTs), as well as glutathione *S*-transferases[@bib12]. Whereas some drugs diffuse into the intestinal cells, there are two major classes of transporters, namely the solute carriers (SLC) and ATP-binding cassette (ABC) transporters, that are important in the disposition of many large and/or polar chemicals. Intestinal transporters mediate the translocation of chemicals in and out of enterocytes, and this process is important for drug disposition in the body[@bib13]. DPGs can be trans-activated by various nuclear receptors following xenobiotic exposure. The constitutive androstane receptor (CAR/Nr1i3) is one of the important xenobiotic-sensing nuclear receptors that regulate the transcription of DPGs. CAR is activated by various chemicals including steroid hormones, bile acids, pharmaceuticals, as well as environmental, dietary, and occupational chemicals[@bib14], *via* direct or indirect mechanisms. Direct activation of CAR refers to ligand-binding to the CAR protein, and the prototypical CAR ligands include TCPOBOP for the mouse CAR and CITCO for the human CAR. The indirect activation of CAR by chemicals such as phenobarbital disassociates CAR from its cytosolic repressor protein. CAR activation leads to its nuclear translocation and binding to the targeted response elements of genes, and this usually leads to the transcriptional up-regulation of DPGs. Chronic activation of CAR is known to cause liver tumor in rodents but to a much lesser extent in humans[@bib15], [@bib16]. Pharmacological activation of CAR by TCPOBOP has also been shown to reduce obesity in mice[@bib17]. *Car*^*−/−*^ mice have been engineered to determine the necessity of CAR in xenobiotic metabolism and liver physiology[@bib18]. Phenobarbital-mediated up-regulation of the prototypical CAR-target gene *Cyp2b10* does not occur in livers of *Car*^*−/−*^ mice, and there is also decreased metabolism of the classic CYP substrate zoxazolamine, as well as a complete loss of the liver hypertrophic and hyperplastic responses to CAR-inducers. CAR is highly expressed in liver, but it can also be detected at high amounts in the small intestine, and a lower amount in the large intestine[@bib19], [@bib20], [@bib21]. Extensive studies have been done regarding the effect of CAR-activation on the hepatic DPG expression[@bib18], [@bib22], [@bib23], [@bib24]. Despite the important role of the intestine in xenobiotic biotransformation, relatively less is known regarding the effect of genetic depletion of *Car* and the pharmacological activation of CAR on the basal and inducibility of DPGs in different sections of intestine. Therefore, the goal of this study was to determine whether the well-known CAR-targeted DPGs in liver are also regulated by CAR in duodenum, jejunum, ileum, and colon. 2. Materials and methods {#s0010} ======================== 2.1. Chemicals and reagents {#s0015} --------------------------- The mouse CAR ligand 1,4-bis-\[2-(3,5-dichloropyridyloxy)\]benzene (3,3′,5,5′-tetrachloro-1,4-bis(pyridyloxy)benzene, TCPOBOP) and corn oil were purchased from Sigma--Aldrich (St. Louis, MO, USA). 2.2. Animal procedures {#s0020} ---------------------- C57BL/6 wild-type (WT) mice were purchased from the Jackson Laboratory (Bar Harbor, ME, USA). Breeder pairs of the *Car*^*−/−*^ mice in the C57BL/6 background were obtained from Amgen (Thousand Oaks, CA, USA). Mice were housed according to the American Animal Association Laboratory Animal Care Guidelines, and were bred under standard conditions at the University of Washington (WA, USA). All animals were given *ad libitum* access to water and irradiated Picolab Rodent Diet 20 number 5053 (PMI Nutrition International, Brentwood, MO, USA). Sixty-day-old wild type and *Car*^*−/−*^ male mice were administered the CAR-ligand TCPOBOP (3 mg/kg, i.p.), or vehicle, once daily for 4 days (*n*=4--5 per group). Various sections of intestine (duodenum, jejunum, ileum, and colon) were collected on the 5th day. The tissues were frozen immediately in liquid nitrogen and then stored in a --80 °C freezer prior further analysis. All animal procedures were approved by the Institutional Animal Care and Use Committee at the University of Washington. 2.3. RNA isolation {#s0025} ------------------ Total RNA was isolated from each section of frozen intestine tissues using RNA zol Bee reagent (Tel-Test Inc., Friendswood, TX, USA) directed by the manufacturer׳s protocol. The total RNA concentration of each sample was quantified spectrophotometrically at 260 nm using a NanoDrop 1000 Sectrophotometer (Thermo Scientific, Waltham, MA, USA). The RNA integrity of each sample was evaluated by formaldehyde agarose gel electrophoresis, and assured by appearance of 18S and 28S rRNA bands. 2.4. Reverse transcription and quantitative polymerase chain reaction (RT-qPCR) {#s0030} ------------------------------------------------------------------------------- Total RNA was reverse-transcribed to cDNA by High-Capacity cDNA Reverse Transcription Kits 1001073 (Applied Biosystems, Foster, CA, USA) in a final volume 10 µL containing 5 µL of RNA sample and 5 µL of 2×RT master mix directed by the manufacturer׳s protocol. The cDNAs were diluted 10 times and amplified by PCR, using the SsoAdvanced Universal SYBR Green Supermix in a Bio-Rad CFX384 Real-Time PCR Detection System (Bio-rad, Hercules, CA, USA). Two µL of cDNA were added to 8 µL of PCR mix, containing SsoAdvanced Universal SYBR Green Supermix (2×), forward and reverse primers, and nuclear-free water. The primers for PCR were designed using the NCBI Primer Design Tool as shown in [Table 1](#t0005){ref-type="table"}, and were purchased from the Integrated DNA Technologies (Coralville, IA, USA). Data were expressed as percentage of the housekeeping gene *β*-actin. 2.5. Western blotting analysis {#s0035} ------------------------------ Each section of intestinal homogenate was prepared using 250 mL ST buffer (250 mmol/L sucrose, 10 mmol/L Tris base, pH 7.5) with protease inhibitors. The crude membranes were prepared from each section of frozen intestine samples as described previously[@bib25]. The protein concentrations in each section of intestines were determined by a Qubit Protein Assay Kit (Thermo Fisher Scientific, Grand Island, NY, USA) as directed by the manufacturer׳s instructions. The samples were subjected to polyacrylamide gel electrophoresis and transferred onto a polyvinylidene difluoride (PVDF) membrane for 3 h on the ice. After transfer, the membranes were blocked by 5% nonfat dry milk in phosphate-buffered saline with 0.05% Tween 20 (PBST) for 1 h and incubated overnight with the following primary antibodies diluted in 1% dry milk in PBST: rabbit anti-mouse CYP2B10 polyclonal antibody (AB9916, 1:5000, EMD Millipore); or goat anti-mouse H3 polyclonal antibody (ab12079, 1:500, Abcam). After washing with 1% dry milk in PBST, the membranes were incubated for 1 h with a 1:2000 HRP-linked species-appropriate secondary antibody (Sigma Aldrich, St. Louis, MO, USA) diluted in 1% dry milk in PBST. After incubation, the membranes were washed again with 1% dry milk in PBST and then with 1% PBS, followed by incubation in Novex ECL Chemiluminescent Substrate Reagent Kit (Life Technologies, Carlsbad, CA, USA). 2.6. Statistical analysis {#s0040} ------------------------- For RT-qPCR analysis, data among multiple groups were analyzed using analysis of variance (ANOVA) followed by the Duncan׳s *post hoc* test (*P*\<0.05) using the SPSS software (IBM Cooperation, Armonk, North Castle, NY, USA). Asterisk (\*) indicates statistically significant differences between control and TCPOBOP-treated wild-type mice. Pound sign (\#) indicates statistically significant differences (*P*\<0.05) between control and TCPOBOP-treated *Car*^*−/−*^ mice. Dollar sign (\$) indicates significant differences (*P*\<0.05) in the basal mRNA expression of DPGs between control WT and control *Car*^*−/−*^ mice. Protein density was quantified using Image J Software (National Institutes of Health, Bethesda, MD, USA). 3. Results {#s0045} ========== 3.1. DPG expressions in intestine {#s0050} --------------------------------- BioGPS[@bib26] data were examined to determine which DPSs are highly expressed in the small and large intestine ([Fig. 1](#f0005){ref-type="fig"}). Those DPGs that were expressed in the intestine were selected ([Table 2](#t0010){ref-type="table"}) to examine their mRNA expression in response to the CAR-ligand TCPOBOP in intestine, based on the following selection criteria: (1) the DPGs have been shown to be *bona fide* CAR-target genes in mouse liver[@bib24], or (2) the DPGs that were not examined in liver[@bib24] but are highly expressed in small or large intestine (http://biogps.org/)[@bib26], [@bib27]. The relative mRNA abundance of the major DPG families in the small and large intestine are shown in [Fig. 1](#f0005){ref-type="fig"}. Data were retrieved from BioGPS as described above. For the *Cyp1* family, *Cyp1a1* appeared to be the major intestinal *Cyp1* isoform and its mRNA was much higher in small intestine than that in large intestine. *Cyp1a2* mRNA was minimally expressed in intestine ([Fig. 1](#f0005){ref-type="fig"}a), although it was shown to be the major *Cyp1* isoform in liver[@bib28]. *Cyp1b1* was expressed at low levels and it was also small intestine-enriched. For the *Cyp2b* family, *Cyp2b10* was the only isoform that was expressed in intestine, and its mRNA was predominantly found in the small intestine but not in large intestine ([Fig. 1](#f0005){ref-type="fig"}b). For the *Cyp3a* family, the mRNAs of *Cyp3a11*, *3a13*, and *3a25* were highly expressed in small intestine; *Cyp3a13* mRNA was also detected at low levels in the large intestine, whereas other *Cyp3a* isoforms (including *Cyp3a16*, *3a41a*, and *3a44*) were not expressed in intestine ([Fig. 1](#f0005){ref-type="fig"}c). For the *Cyp4* family, *Cyp4b1* was the major intestinal isoform in both small and large intestine. *Cyp4a10* and *4a31* mRNAs were also detected at low levels in the large intestine, whereas they were minimally expressed in the small intestine. Other *Cyp4* isoforms (including *Cyp4a12a* and *4a14*) were minimally expressed in intestine ([Fig. 1](#f0005){ref-type="fig"}d). The *Nqo1* mRNA appeared to be higher than *Nqo2* mRNA in both small and large intestines ([Fig. 1](#f0005){ref-type="fig"}e). For the *Aldh* family, *Aldh1a1*, *1b1*, and *9a1* were the predominant isoforms and in general, they were more highly expressed in small intestine than in large intestine. *Aldh1a7*, *1l1*, *3a2*, and *6a1* were also expressed at low levels in both the small and large intestine, *Aldh16a1* and *18a1* were lowly expressed in small intestine but minimally expressed in large intestine, whereas other *Aldh* isoforms (including *Aldh1a2*, *1a3*, *1l2*, *2*, *3a1*, *3b1*, *3b2*, *4a1*, *5a1*, *7a1*, and *8a1*) were minimally expressed in small and large intestine ([Fig. 1](#f0005){ref-type="fig"}f). For the *Gsta* family, *Gsta1* appeared to be the major intestinal isoform and its mRNA was much higher in small intestine than in large intestine. *Gsta4* mRNA was also detected at low levels in both small and large intestine. *Gsta2* and *Gsta3* were minimally expressed in intestine ([Fig. 1](#f0005){ref-type="fig"}g). For the *Gstm* family, *Gstm1, 2, and 5* were highly expressed in large intestine*; Gstm1, 2, 3,* and *5* were also detected at low levels in small intestine, whereas other *Gstm* isoforms (including *Gstm4*, *m6*, and *m7*) were minimally expressed in small and large intestine ([Fig. 1](#f0005){ref-type="fig"}h). For the *Gstt* family, *Gstt1*, *2*, and *3* mRNAs were the predominant isoforms in both small and large intestine, whereas *Gstt4* was minimally expressed in small and large intestine ([Fig. 1](#f0005){ref-type="fig"}i). For the *Sult* family, *Sult1b1*, *1d1*, and *2b1* appeared to be highly expressed in small intestine with the mRNA abundance in descending order; *Sult1b1* and *1d1* were also expressed at high levels in the large intestine. *Sult1a1* was only expressed in large intestine but not in small intestine, and its mRNA was highly enriched in the large intestine. *Sult1c2* and *2b1* were also lowly expressed in the large intestine. Other *Sult* isoforms (including *Sult1e1*, *2a2*, *3a1*, *4a1*, *5a1*, and *6b1*) were minimally expressed in small and large intestine ([Fig. 1](#f0005){ref-type="fig"}j). For the *Ugt* family, *Ugt1a* transcripts (note: probes were not specific to differentiate various *Ugt1a* isoforms) and *Ugt2b34* mRNA appeared to be the major *Ugt* genes expressed in small and large intestine. *Ugt1a* transcripts were more highly expressed in large intestine than small intestine. *Ugt2b5* and *2b35* were lowly expressed in both small and large intestines. *Ugt2a3* was detected at low levels only in the small intestine, whereas *Ugt2b38* and *3a2* were expressed at low levels only in large intestine. Other *Ugt* isoforms were minimally expressed in intestine ([Fig. 1](#f0005){ref-type="fig"}k). For the Solute carrier organic anion (*Slco*) transporter family (also known as the *Oatp*s), *Slco2a1*, *2b1*, *3a1*, and *4a1* were the intestine-enriched *Slco* isoforms and they were more highly expressed in small than in large intestine. Other *Slco* isoforms (including *Slco1a1, 1a4, 1a5, 1a6, 1b2, 1c1, 5a1, 6b1, 6c1*, and *6d1*) were not expressed in intestine ([Fig. 1](#f0005){ref-type="fig"}I). For the *Abcc* family (*Mrp*), *Abcc3* appeared to be the major intestinal isoform and its mRNA was approximately twice in large intestine as that in small intestine. *Abcc1* was detected at low levels in large intestine, whereas *Abcc2* was detected at low levels in small intestine. *Abcc4*, *5, 9,* and *10* were also expressed at low levels in both small and large intestine. Other *Abcc* isoforms (including *Abcc6, c8*, and *c12*) were not detected in the intestine ([Fig. 1](#f0005){ref-type="fig"}m). In summary, BioGPS has identified distinct DPG isoforms that are expressed in the intestine. Based on this information, as well as the previous findings regarding known CAR-targeted DPGs in liver[@bib24], DPGs listed in [Table 2](#t0010){ref-type="table"} were selected for the induction studies in WT and *Car*^*−/−*^ mice. Previously, it has been shown that CAR is highly expressed in liver[@bib24], and the basal expression of CAR was also found at relatively high levels in various sections of intestine[@bib19], [@bib21]. To further determine the distribution of CAR in intestine as compared to liver in control and TCPOBOP-treated conditions, and to confirm the depletion of *Car* in intestine of the *Car*^*−/−*^ mice, the *Car* mRNA was quantified in various sections of intestine and liver from WT and *Car*^*−/−*^ mice treated with corn oil or TCPOBOP ([Fig. 2](#f0010){ref-type="fig"}). CAR was most highly expressed in liver of the WT mice, followed by duodenum and jejunum, whereas ileum and colon had very low levels of *Car* expression. TCPOBOP down-regulated the *Car* mRNA in liver and duodenum of WT mice, and tended to decrease in jejunum, although a statistically significant difference was not achieved. As expected, the *Car* mRNA was not expressed in any of the tissues examined in the *Car*^*−/−*^ mice. 3.2. Regulation of phase-I drug metabolizing enzymes (Cyps and Aldh) in intestine by CAR {#s0055} ---------------------------------------------------------------------------------------- The expressions of selected phase-I enzymes are shown in [Figure 3](#f0015){ref-type="fig"}, [Figure 4](#f0020){ref-type="fig"}. For *Cyp*s ([Fig. 3](#f0015){ref-type="fig"}a--g), the basal expression of *Cyp1a1* was highest in duodenum, followed by jejunum and ileum, and was minimal in colon. Interestingly, in control *Car*^*−/−*^ mice, the basal *Cyp1a1* mRNA increased 11.97-fold in duodenum and 6.02-fold in jejunum, suggesting that CAR suppresses the basal expression of *Cyp1a1* in duodenum and jejunum of WT mice. TCPOBOP did not alter *Cyp1a1* mRNA in any portions of the intestine in WT mice; however, it down-regulated *Cyp1a1* 89.3% in duodenum of the *Car*^*−/−*^ mice, suggesting an off-target effect of TCPOBOP independent of CAR ([Fig. 3](#f0015){ref-type="fig"}a). Regarding the prototypical CAR-target gene *Cyp2b10*, as shown in [Fig. 3](#f0015){ref-type="fig"}b, the basal expression of *Cyp2b10* was the highest in duodenum followed by jejunum, but was minimally expressed in ileum and colon. In *Car*^*−/−*^ mice, the basal *Cyp2b10* mRNA increased in colon (22.1-fold) but the expression was still minimal. TCPOBOP up-regulated *Cyp2b10* mRNA 4.98-fold in duodenum, 5.72-fold in jejunum, and 20.2-fold in ileum of WT mice in a CAR-dependent manner. As shown in [Fig. 3](#f0015){ref-type="fig"}c, the basal expression of *Cyp3a11* was highest in duodenum, followed by jejunum, ileum, and colon. In *Car*^*−/−*^ mice, basal *Cyp3a11* mRNA decreased 55% in jejunum and 75% in ileum, suggesting that CAR is necessary for the basal expression of *Cyp3a11* in these sections. TCPOBOP up-regulated *Cyp3a11* expression 70% in duodenum in a CAR-dependent manner. However, it did not alter the *Cyp3a11* mRNAs in other sections of the intestine. As shown in [Fig. 3](#f0015){ref-type="fig"}d, the basal expression of *Cyp3a13* was highest in duodenum, followed by jejunum, and was much lower in ileum and colon. In *Car*^*−/−*^ mice, the basal *Cyp3a13* mRNA expression decreased 60% in duodenum, suggesting that CAR is necessary in maintaining the constitutive expression of *Cyp3a13* in the duodenum. TCPOBOP up-regulated *Cyp3a13* mRNA expression 38% in duodenum and 44% in jejunum in a CAR-dependent manner; however, it did not alter *Cyp3a13* mRNA in ileum or colon. As shown in [Fig. 3](#f0015){ref-type="fig"}e, the basal expression of *Cyp3a25* was highest in duodenum, followed by jejunum and ileum but was minimal in colon. In *Car*^*−/−*^ mice, the basal *Cyp3a25* mRNA decreased in duodenum (77.8%), jejunum (58.4%), and ileum (59.6%), suggesting that CAR is necessary for the basal expression of *Cyp3a25* in the small intestine. TCPOBOP in general did not alter the *Cyp3a25* mRNA expression in any sections of intestine. As shown in [Fig. 3](#f0015){ref-type="fig"}f, the basal expression of *Cyp4a10* was the highest in duodenum, followed by colon, jejunum, and ileum. TCPOBOP in general did not alter the expression of *Cyp4a10*, except for a down-regulation (63.5%) in ileum in a CAR-dependent manner. As shown in [Fig. 3](#f0015){ref-type="fig"}g, the basal expression of *Cyp4b1* was similarly expressed in all portions of the intestine. In *Car*^*−/−*^ mice, the basal *Cyp4b1* mRNA decreased 34% in duodenum, 71% in ileum and 81% in colon, suggesting that CAR is necessary in maintaining the basal expression of *Cyp4b1* in these sections. TCPOBOP down-regulated *Cyp4b1* mRNA expression 27.4% in ileum of WT mice. As shown in [Fig. 3](#f0015){ref-type="fig"}h, the basal expression of *Nqo1* was highest in colon, followed by duodenum, ileum, and jejunum. In *Car*^*−/−*^ mice, the basal *Nqo1* mRNA increased 1.09-fold in jejunum, suggesting that CAR suppresses the basal expression of *Nqo1* in WT mice intestine. TCPOBOP moderately down-regulated *Nqo1* 27.1% in ileum of the *Car*^*−/−*^ mice. In summary, CAR suppresses the basal expression of *Cyp1a1* and *Nqo1*, but maintains the constitutive expression of *Cyp3a11*, *Cyp3a13*, *Cyp3a25*, and *Cyp4b1*, in distinct sections of intestine; whereas pharmacological activation of CAR by TCPOBOP up-regulates *Cyp2b10*, *Cyp3a13*, *Cyp3a13*, but down-regulates *Cyp4a10* and *Cyp4b1* in distinct sections of intestine, in a CAR-dependent manner. Regarding the regulation of the aldehyde dehydrogenases (*Aldhs*), as shown in [Fig. 4](#f0020){ref-type="fig"}a, the basal expression of *Aldh1a1* was highest in colon and duodenum, followed by jejunum, and ileum. In *Car*^*−/−*^ mice, the basal *Aldh1a1* mRNA moderately decreased 54% in ileum. TCPOBOP up-regulated *Aldh1a1* mRNA 5.37-fold in duodenum and 87% in jejunum in a CAR-dependent manner. However, TCPOBOP did not alter the *Aldh1a1* mRNA expression in ileum and colon. As shown in [Fig. 4](#f0020){ref-type="fig"}b, the basal expression of *Aldh1a7* was highest in ileum, followed by colon, jejunum, and minimal in duodenum. The absence of CAR leads to an apparent increase in *Aldh1a7* in duodenum and jejunum, although a statistical significance was not achieved. TCPOBOP up-regulated *Aldh1a7* mRNA 51.4-fold in duodenum and 5.00-fold in jejunum in a CAR-dependent manner, but it slightly decreased *Aldh1a7* mRNA in colon. As shown in [Fig. 4](#f0020){ref-type="fig"}c, the basal expression of *Aldh1b1* was relatively similar in all portions of the intestine. In *Car*^*−/−*^ mice, the basal expression of *Aldh1b1* mRNA was 70.9% higher in jejunum and 57.4% in colon than in WT mice, and TCPOBOP did not alter *Aldh1b1* mRNA in any intestinal sections. As shown in [Fig. 4](#f0020){ref-type="fig"}d, the basal expression of *Aldh3a2* was relatively similar in all portions of intestine. In *Car*^*−/−*^ mice, basal *Aldh3a2* mRNA decreased 41.6% in duodenum and 65.6% in ileum. TCPOBOP did not have any effect on the *Aldh3a2* mRNA expression in intestine. As shown in [Fig. 4](#f0020){ref-type="fig"}e, the basal expression of *Aldh3b1* expression was highest in colon and ileum, and was minimal in duodenum and jejunum. Interestingly, In *Car*^*−/−*^ mice, the basal *Aldh3b1* mRNA markedly increased 124.7-fold in duodenum and 26.8-fold in jejunum, which suggest that CAR suppresses the basal expression of *Aldh3b1* in WT mice intestine. TCPOBOP up-regulated *Aldh3b1* expression 160.4-fold in duodenum, 20.9-fold in jejunum, and 41.9% in ileum in a CAR-dependent manner. As shown in [Fig. 4](#f0020){ref-type="fig"}f, the basal expression of *Aldh9a1* decreased from duodenum to colon. TCPOBOP did not alter the *Aldh9a1* mRNA expression in any portions of intestine. In summary, CAR suppresses the basal expression of *Aldh3b1* but maintains the constitutive expression of *Aldh1a1*, *Aldh1b1*, and *Aldh3a2* in distinct sections of intestine, whereas pharmacological activation of CAR by TCOBOP up-regulates *Aldh1a7* and *Aldh3b1* in a CAR-dependent manner in distinct sections of small intestine. 3.3. Regulation of phase-II drug metabolizing enzymes (Sult, Gsta, Gstt, Gstm, Ugt) in intestine by CAR {#s0060} ------------------------------------------------------------------------------------------------------- [Figure 5](#f0025){ref-type="fig"}, [Figure 6](#f0030){ref-type="fig"}, [Figure 7](#f0035){ref-type="fig"}, [Figure 8](#f0040){ref-type="fig"} illustrate the regulation of various phase-II enzymes by CAR in intestine. Regarding the regulation of *Sult*, as shown in [Fig. 5](#f0025){ref-type="fig"}, the basal *Sult1a1* mRNA expression was predominantly expressed in colon, and was very lowly expressed in all portions of the small intestine. TCPOBOP did not alter the expression of *Sult1a1* mRNA in any portions of intestine ([Fig. 5](#f0025){ref-type="fig"}a). As shown in [Fig. 5](#f0025){ref-type="fig"}b, the basal expression of *Sult1b1* mRNA was highest in duodenum, followed by jejunum, colon, and ileum. In *Car*^*−/−*^ mice, there was a decrease in the basal *Sultb1* mRNA in duodenum (47.1%), but an increase in the basal *Sult1b1* mRNA in colon (61.2%). TCPOBOP did not alter the *Sult1b1* mRNA expression in any portions of intestine. As shown in [Fig. 5](#f0025){ref-type="fig"}c, the basal mRNA expression of *Sult1c2* is highest in colon, followed by ileum and jejunum, and was minimally expressed in duodenum. In *Car*^*−/−*^ mice, the basal *Sult1c2* mRNA expression was 60% lower in ileum, suggesting that CAR is necessary in maintaining constitutive expression of *Sult1c2* in ileum of WT mice. TCPOBOP up-regulated *Sult1c2* mRNA 73.6-fold in duodenum and 4.8-fold in jejunum in a CAR-dependent manner. However, TCPOBOP did not alter *Sult1c2* mRNA expression in ileum or colon. As shown in [Fig. 5](#f0025){ref-type="fig"}d, the basal mRNA expression of *Sult1d1* increased from duodenum to colon. In *Car*^*−/−*^ mice, the basal *Sult1d1* mRNA increased 23.9-fold and 6.21-fold in duodenum and jejunum; however, it decreased 75.8% in ileum and 74.4% in colon. TCPOBOP up-regulated *Sult1d1* mRNA 52.4-fold in duodenum and 12.5-fold in jejunum in a CAR-dependent manner, whereas it had no effect on the *Sult1d1* mRNA in ileum, and down-regulated 47.5% in colon of WT mice. As shown in [Fig. 5](#f0025){ref-type="fig"}e, the basal expression of *Sult2b1* mRNA decreased from duodenum, to colon. In *Car*^*−/−*^ mice, the basal *Sult2b1* mRNA decreased in duodenum (53.1%) but increased in colon (1.31-fold). TCPOBOP down-regulated *Sult2b1* mRNA 61.4% in duodenum, but had no effect in other portions of the intestine. As shown in [Fig. 5](#f0025){ref-type="fig"}f, the basal expression of *Sult5a1* was highest in colon and duodenum, and lower in ileum, and jejunum. TCPOBOP had minimal effect on *Sult5a1* mRNA expression, except for a moderate increase (57.4%) in jejunum in a CAR-dependent manner. As shown in [Fig. 5](#f0025){ref-type="fig"}g, the basal mRNA expression of *Papss2* was highest in duodenum and colon, but was much lower in jejunum and ileum. TCPOBOP did not have any effect on *Papss2* mRNA expression in any portions of intestine of either genotype. Regarding the *Gsta*, as shown in [Fig. 6](#f0030){ref-type="fig"}a--c, the basal mRNAs of multiple *Gsta* family members (*Gsta1*, *2*, and *4*) followed a similar expression pattern, which was highest expression in duodenum, followed by jejunum, and minimal expression in ileum and colon. In duodenum, TCPOBOP up-regulated the mRNAs of *Gsta1* (1.46-fold), *Gsta2* (4.56-fold), and *Gsta4* (2.45-fold) in a CAR-dependent manner ([Fig. 6](#f0030){ref-type="fig"}a--c). In jejunum, TCPOBOP also up-regulated *Gsta2* mRNA 1.72-fold in a CAR-dependent manner. However, TCPOBOP had no effect on the *Gsta* expression in other portions of intestine in either WT or *Car*^*−/−*^ mice ([Fig. 6](#f0030){ref-type="fig"}a--c). Regarding the *Gstt*, as shown in [Fig. 6](#f0030){ref-type="fig"}d, *Gstt1* mRNA basal expression was highest in duodenum, followed by colon, jejunum, and ileum, and was not readily altered by TCPOBOP in any portions of intestine, except for a moderate increase by TCPOBOP in ileum of WT mice. CAR deficiency moderately decreased the basal *Gstt1* mRNA in ileum (27.8%). As shown in [Fig. 6](#f0030){ref-type="fig"}e, the basal expression of *Gstt2* mRNA was highest in colon, followed by duodenum, ileum, and jejunum, and was not altered by TCPOBOP in any intestinal sections. As shown in [Fig. 6](#f0030){ref-type="fig"}f, the basal expression of *Gstt3* was low in duodenum and jejunum, but was much higher in ileum and highest in colon. CAR deficiency resulted in a 34.5-fold up-regulation of *Gstt3* mRNA in duodenum and a 10.2-fold in jejunum of WT mice. Interestingly, TCPOBOP also up-regulated *Gstt3* mRNA in these two sections (47.4-fold and 9.34-fold, respectively) in a CAR-dependent manner, whereas it did not have any effect in ileum and colon. Regarding the *Gstm* family members ([Fig. 7](#f0035){ref-type="fig"}a--e), the basal mRNA expression of *Gstm1* ([Fig. 7](#f0035){ref-type="fig"}a), *Gstm4* ([Fig. 7](#f0035){ref-type="fig"}d), and *Gstm5* ([Fig. 7](#f0035){ref-type="fig"}e) was highest in duodenum, followed by colon, jejunum, and ileum; the basal mRNA expression of *Gstm2* was highest in colon, followed by duodenum, ileum, and jejunum; whereas the basal mRNA expression of *Gstm3* was highest in duodenum, followed by jejunum, colon, and ileum ([Fig. 6](#f0030){ref-type="fig"}d--g). In *Car*^*−/−*^ mice, the basal *Gstm3* mRNA decreased 67.0% in ileum, but *Gstm5* mRNA increased 58.4% in colon. TCPOBOP up-regulated all *Gstm* genes, except for *Gstm5*, in duodenum (15.8-fold, 6.28-fold, 5.27-fold, and 5.07-fold, respectively) and jejunum (6.20-fold, 3.65-fold, 3.17-fold, and 3.16-fold, respectively), in a CAR-dependent manner. In ileum, TCPOBOP also moderately up-regulated certain *Gstm* genes, such as *Gstm1* (58%) and *Gstm3* (74.7%) in a CAR dependent manner. TCPOBOP in general had minimal effect on the *Gstm* genes in colon of either genotype ([Fig. 7](#f0035){ref-type="fig"}a--e). TCPOBOP moderately down-regulated the *Gstm1* and *Gstm2* mRNAs in ileum, and up-regulated *Gstm4* in colon of *Car*^*−/−*^ mice, likely due to off-target effect of the chemical. Regarding the regulation of the *Ugt* family, as shown in [Fig. 8](#f0040){ref-type="fig"}a, the basal mRNA expression of *Ugt1a1* was expressed at comparable levels in various sections of the intestine. In *Car*^*−/−*^ mice, there was a moderate decrease in the basal *Ugt1a1* mRNA only in ileum (46.3%). TCPOBOP up-regulated *Ugt1a1* mRNA 3.95-fold in duodenum and 1.46-fold in jejunum in a CAR-dependent manner. However, TCPOBOP had no effect on the *Ugt1a1* mRNA in ileum or colon. As shown in [Fig. 8](#f0040){ref-type="fig"}b, the basal expression of *Ugt1a9* was highest in duodenum, and was lowly expressed in the other sections of the intestine. The basal *Ugt1a9* mRNA was further decreased in colon of the *Car*^*−/−*^ mice. TCPOBOP in general did not alter the *Ugt1a9* mRNA expression in any sections of intestine, although it tended to increase the *Ugt1a9* mRNA in duodenum (a statistical significance was not achieved). As shown in [Fig. 8](#f0040){ref-type="fig"}c and d, the basal mRNAs of *Ugt2b34* and *Ugt2b35* were both highest in duodenum, followed by colon, jejunum, and ileum. In *Car*^*−/−*^ mice, the basal *Ugt2b34* and *2b35* mRNAs decreased in ileum (50.5% and 40.7%, respectively), suggesting that CAR is necessary in maintaining constitutive expression of both *Ugt2b34* and *2b35*. TCPOBOP up-regulated *Ugt2b34* mRNA 5.09-fold in duodenum in a CAR-dependent manner; however, it did not alter the *Ugt2b34* mRNA in other sections of intestine. TCPOBOP had a similar effect on the *Ugt2b35*, except that the mRNA increased in duodenum was not statistically significant. As shown in [Fig. 8](#f0040){ref-type="fig"}e, the basal mRNA expression of *Ugt2b36* was comparable in various sections of intestine. In *Car*^*−/−*^ mice, the basal *Ugt2b36* mRNA was down-regulated 61.4% in ileum. TCPOBOP up-regulated *Ugt2b36* mRNA 1.65-fold in duodenum and 73.1% in jejunum in a CAR-dependent manner, however, it did not alter the *Ugt2b36* mRNA in ileum or colon. In summary, CAR regulates the basal expression of many phase-II enzymes in distinct sections of intestine, whereas pharmacological activation of CAR up-regulates *Sult1c2*, *Sult1d1*, *Sult5a1*, *Gsta1*, *Gsta2*, *Gsta4*, *Gstt3*, *Gstm1-4*, *Ugt1a1*, *Ugt2b34*, *Ugt2b36*, and tends to up-regulate *Ugt1a9* and *Ugt2b35*, in distinct sections of intestine in a CAR-dependent manner. In contrast, pharmacological activation of CAR down-regulates *Sult1d1* in colon and *Sult2b1* in duodenum in a CAR-dependent manner; however, because *Car* is lowly expressed in colon ([Fig. 2](#f0010){ref-type="fig"}), the TCPOBOP-mediated effects in colon may be due to the involvement of other regulatory factors. 3.4. Regulation of the transporters in intestine by CAR {#s0065} ------------------------------------------------------- [Fig. 9](#f0045){ref-type="fig"} shows the mRNA expression of uptake and efflux transporters in various sections of intestine of WT and *Car*^*−/−*^ mice treated with corn oil or TCPOBOP. Regarding the uptake transporters ([Fig. 9](#f0045){ref-type="fig"}a--c), the basal mRNA expression of *Oatp2a1* was higher in colon than in the three sections of the small intestine, and TCPOBOP did not alter the *Oatp2a1* mRNA expression in any sections of intestine, except for a moderate increase in ileum of *Car*^*−/−*^ mice (46.5%, [Fig. 9](#f0045){ref-type="fig"}a). The basal mRNA expression of *Asbt* was highest in ileum and colon, but was minimally expressed in duodenum and jejunum, and TCPOBOP had no effect on the *Asbt* mRNA expression in any sections of intestine ([Fig. 9](#f0045){ref-type="fig"}b). Regarding the efflux transporters *Mrp2--4* ([Fig. 9](#f0045){ref-type="fig"}c--e), the basal mRNA expression of *Mrp2* was highest in duodenum, followed by jejunum, and ileum, but was minimally expressed in colon. TCPOBOP up-regulated *Mrp2* 1.93-fold in duodenum, 1.16-fold in jejunum, and 40.3% in ileum in a CAR-dependent manner. In *Car*^*−/−*^ mice, TCPOBOP down-regulated *Mrp2* mRNA 96.5% in ileum, but up-regulated *Mrp2* mRNA 9.03-fold in colon, which may be due to off-target effect of the chemical ([Fig. 9](#f0045){ref-type="fig"}c). The basal mRNA expression of *Mrp3* was highest in colon, followed by duodenum, jejunum, and ileum. In *Car*^*−/−*^ mice, the basal *Mrp3* mRNA expression was up-regulated 1.18-fold in jejunum and 1.05-fold in ileum, but was down-regulated 48.4% in colon. TCPOBOP up-regulated *Mrp3* mRNA 2.11-fold in duodenum and 71.9% in ileum in a CAR-dependent manner (it also tended to increase *Mrp3* mRNA in jejunum although a statistically significant difference was not achieved). In contrast, TCPOBOP down-regulated *Mrp3* mRNA 50.7% in colon of WT mice. Considering that *Car* is minimally expressed in colon ([Fig. 2](#f0010){ref-type="fig"}), the TCPOBOP-mediated down-regulation of *Mrp3* may be due to off-target effect of the chemical ([Fig. 9](#f0045){ref-type="fig"}d). The basal mRNA expression of *Mrp4* was highest in colon followed by ileum, but was minimally expressed in duodenum and jejunum. Interestingly, in *Car*^*−/−*^ mice, there was a marked increase in the basal *Mrp4* mRNA in both duodenum and jejunum (81.1-fold and 15.6-fold, respectively), suggesting CAR suppresses *Mrp4* basal expression in these two sections. Conversely, pharmacological activation of CAR by TCPOBOP also increased *Mrp4* mRNA in duodenum and jejunum (74.2-fold and 15.4-fold, respectively), in a CAR-dependent manner. However, TCPOBOP did not alter the *Mrp4* mRNA in ileum and colon ([Fig. 9](#f0045){ref-type="fig"}e). In summary, CAR suppresses the basal expression of *Mrp3* in jejunum and ileum, as well as *Mrp4* in duodenum and jejunum, whereas pharmacological activation of CAR by TCPOBOP has minimal effect on the uptake transporters but markedly increases the efflux transporters *Mrp2--4* in small intestine, but decreases the *Mrp3* mRNA in colon, in a CAR-dependent manner. Even though *Mrp3* and *Mrp4* are well known CAR-target genes in liver[@bib24] and intestine ([Fig. 9](#f0045){ref-type="fig"}c--e), their basal expressions are highest in colon where *Car* is lowly expressed, suggesting that other regulatory factors are involved in the basal expression of these transporters. 3.5. Regulation of CYP2B10 protein in duodenum by CAR {#s0070} ----------------------------------------------------- Because duodenum has the highest *Car* and *Cyp2b10* mRNA expression ([Figs. 2](#f0010){ref-type="fig"} and [3](#f0015){ref-type="fig"}b), the protein for the prototypical CAR-target gene *Cyp2b10* was quantified in duodenum of WT and *Car*^*−/−*^ mice by Western blotting analysis ([Fig. 10](#f0050){ref-type="fig"}). Following TCPOBOP treatment, consistent with the mRNA data, CYP2B10 protein was also increased (5.11-fold) in the duodenum of WT mice ([Fig. 10](#f0050){ref-type="fig"}b). However, such TCPOBOP-mediated induction in the CYP2B10 protein expression was completely abolished in the duodenum of *Car*^*−/−*^ mice, suggesting that TCPOBOP-mediated up-regulation of CYP2B10 protein in duodenum is CAR-dependent. 4. Discussion {#s0075} ============= In conclusion, the present study has demonstrated that in addition to its important roles in liver[@bib24], CAR is also critical in both maintaining the basal expression of certain DPGs and the pharmacological regulation of certain DPGs in a section-specific manner of the intestine. A systematic comparison between liver (previous studies) and intestine (present study) has shown that CAR activation in liver and intestine produces overlapping but not identical results. The present study has also compared the section-specific CAR-mediated effect on the DPG expression, and has demonstrated that in general, duodenum appears to be the most responsive section following exposure to the CAR-ligand TCPOBOP, likely because CAR is highest expressed in duodenum as compared to other sections of the intestine. TCPOBOP not only has inducible but also suppressive effect on the DPG expression in intestine. In addition, the CAR-independent off-target effect of TCPOBOP has also been observed in the present study, evidenced by TCPOBOP-mediated changes in DPG expression in *Car*^*−/−*^ mice, and TCPOBOP-mediated changes in DPG expression in WT colon where CAR is minimally expressed. Many *bona fide* CAR-target genes in small intestine were highest expressed in colon where CAR is minimally expressed, suggesting that additional regulatory factors are involved in the basal expression of these genes. A systematic comparison of the CAR effect on DPG expression in between liver and various sections of intestine is shown in [Table 2](#t0010){ref-type="table"} (CAR-mediated basal expression of DPGs) and [Table 3](#t0015){ref-type="table"} (effect of pharmacological activation of CAR on the expression of DPGs). In general, the basal CAR expression is more important for the constitutive expression of DPGs in intestine rather than in liver[@bib24] ([Table 2](#t0010){ref-type="table"}). Regarding the effect of the pharmacological activation of CAR, consistent with liver data[@bib24], TCPOBOP up-regulated many DPGs (*Cyp2b10*, *Cyp3a11*, *Aldh1a1*, *Aldh1a7*, *Gsta1*, *Gsta4*, *Gstm1-m4*, *Ugt1a1*, *Ugt2b34*, *Ugt2b36*, and *Mrp2--4*) in certain portions of the small intestine in a CAR-dependent manner, with duodenum generally being the most inducible section. In contrast, *Nqo1*, *Papss2*, *Ugt1a9*, and *Ugt2b35* were up-regulated by TCPOBOP in liver but were not changed in intestine, therefore the pharmacological activation of CAR in liver and intestine are not identical. Such tissue-specific effects may be due to tissue-specific chromatin epigenetic environment, such as different signatures for DNA methylation (suppressive signal for gene transcription) and/or histone modification patterns, which prevent the CAR-mediated trans-activation of certain DPGs in intestine. The epigenetic signatures within the enhancers and promoters of certain DPGs will need to be examined in future studies. Using the epithelial cells scraped from the whole small intestine, another study in the literature has also shown liver- *vs.* small intestine--specific regulation of some DPGs by TCPOBOP[@bib29]. Our finding is consistent with that study regarding the regulation of *Cyp2b10*, *Gsta1*, *Gstm2*, *Mrp2*, and *Mrp3*, and in addition, the present study has investigated the intestinal-specific CAR-mediated regulation of many other genes, including *Cyp3a13*, *3a25*, *Cyp4a10*, *Cyp4b1*, *Nqo1*, *Gsta2*, *Gsta4*, *Gstt2*, *Gstt3*, *Gstm3*, *Gstm4*, *Mrp4*, and this has added new information to the existing knowledge. Certain discrepancies are also observed between the present study and the Maglich et al.[@bib29] study, in that *Aldh1a1*, *Aldh1a7*, and *Cyp3a11* were not changed in the whole small intestine by TCPOBOP in the previous study, but were up-regulated by TCPOBOP in the present study in a CAR-dependent manner. In addition, the basal *Cyp1a1* expression was decreased in the previous study but increased in the present study. Such discrepancy may be due to difference doses of TCPOBOP (a single dose of TCPOBOP at 0.3 mg/kg in corn oil with 5% DMSO of the previous study *vs.* 3 mg/kg of TCPOBOP in corn oil once daily for 4-days in the present study), or different sample preparation procedures (epithelial cells scraped from whole small intestine in the previous study *vs*. various sections of small intestine in the present study). Previous studies have demonstrated that the basal expression of *Car* is high in liver and small but is lower in the large intestine[@bib19], [@bib21], [@bib30], whereas the present study has confirmed the basal tissue distribution of CAR, and is among the first to show that pharmacological activation of CAR by TCPOBOP actually down-regulates the *Car* expression in liver, duodenum, and jejunum ([Fig. 2](#f0010){ref-type="fig"}), and this is likely due to a negative feedback mechanism to prevent excessive CAR-signaling through decreasing the CAR synthesis. Regarding the regulation of DPGs by CAR, a previous study in the literature has demonstrated the CAR-dependent up-regulation of *Cyp2b10* in duodenum[@bib30]. The present study on *Cyp2b10* in duodenum is consistent with that study, and our study has also examined the expression of *Cyp2b10* and other DPGs in other sections of intestine. Another previous study has performed a preliminary survey in WT male mice regarding the regulation of a few *Ugt*s by TCPOBOP in duodenum, jejunum, ileum, and colon using pooled samples (*i.e*., 1 pooled sample from *n*=5 biological replicates)[@bib31]. The apparent TCPOBOP-mediated increase in the mRNAs of *Ugt1a1* and *Ugt2b35* in that previous study is consistent with the present study, but *Ugt2b34* mRNA is only up-regulated by TCPOBOP in the present study. Such differences are likely due to different method of detection (branched DNA amplification technology *vs*. RT-qPCR), and/or pooled *vs*. individual samples. Our finding on CAR-dependent up-regulation of *Ugt1a1* mRNA by TCPOBOP in duodenum is also consistent with a previous study using Northern blot of *Ugt1a1* in duodenum of WT and *Car*^*−/−*^ mice[@bib29]. The expression of *Car* is gender-divergent (*i.e*. higher in females than in males) in liver but not in any sections of intestine[@bib19]. Therefore, the present study has only tested the effect of CAR activation in intestines of male mice. Although many orally administered drug absorption and delivery are known to take place mostly in small intestine and liver, the present study has shown that many *bona fide* CAR-target DPGs in liver and small intestine are highest expressed in colon, where *Car* is lowly expressed. Examples of these genes include *Aldh3b1*, *Sult1c2*, *Sult1d1*, *Gstt3*, *Mrp3* and *Mrp4*. Functionally speaking, the high expression of *Aldh3b1* may be important in detoxifying the microbial aldehyde produced from ethanol by the intestinal bacteria, and this may be critical in reducing the risk of colon cancer derived from microbial aldehyde[@bib32]. The high expression of certain phase-II enzymes such as *Sults* and *Gst* in colon correlate with its critical function to conjugate and thus detoxify various substances in large intestine. GSTs are also involved in the metabolism of endogenous and exogenous carcinogenic substances, which are implicated in the risks of colorectal cancer[@bib33], [@bib34]. The colon-specific efflux transporters Mrp3 and Mrp4 may also favor the elimination of various potentially toxic chemicals into feces. Another functional significant of colon-specific expression of certain DPGs is that it is associated with the critical roles of DPGs in metabolizing colon-targeted drugs or prodrugs. In addition to the microbial enzymes capability to bio-activate and/or detoxify xenobiotics, the colon tissue derived host enzymes may also contribute to the biotransformation of certain chemicals. One critical question that arises is in regard of the potential species differences in the CAR-mediated regulation of intestinal DPGs. CAR is highly expressed in liver and small intestine of both mice and humans[@bib21], [@bib35]. The species differences of CAR and its tumorigenesis potential have been well characterized in liver, in that pharmacological activation of mouse CAR leads hepatomegaly followed by hepatocarcinogenesis in a CAR-dependent manner[@bib36]. Although human CAR activation is not a risk to cause liver tumor in human, it may cause liver hypertrophy without hyperplasia in response to the human CAR activators phenobarbital and chlordane, suggesting that hCAR is able to induce hypertrophic responses in response to xenobiotic stress[@bib37]. However, both the mouse and human CAR proteins appear to share high similarities in regulating the genes involved in xenobiotic biotransformation in liver. For example, the mouse CAR activation by TCPOBOP up-regulates the expression of *Cyp1a2*, *Cyp2b10*, *Cyp3a11*, and *Ugt1a1* in a CAR-dependent manner in liver[@bib24], whereas the human CAR activation by the human CAR activators also up-regulates the expression of the human orthologs *CYP1A2*, *CYP2B6*, and *UGT1A1*[@bib29]. In human intestine-derived Caco2 cells phenobarbital up-regulates *CYP2B6* and *CYP3A4*[@bib38], [@bib39], and our finding regarding the CAR-mediated up-regulation of the mouse orthologs *Cyp2b10* and *Cyp3a11* in duodenum is consistent with the previous studies. However, relatively less is known regarding the intestinal effect of pharmacological activation or genetic depletion of *Car* in the regulation of many other DPGs *in vivo*, thus the present study has filled this critical knowledge gap. Identification of the xenobiotic responses to CAR activation in mouse and human intestines is critical for understanding certain adverse drug reactions for orally exposed chemicals. The authors would like to thank all members in Dr. Cui׳s laboratory for help in tissue collection, sample preparation, laboratory procedures, and bioinformatics. We also would like to thank Dr. Curtis Klaassen for proof-reading this manuscript. This study is supported by U. S. National Institute of Health R-01 grants ES019487, ES025708, and GM11138, as well as start-up funds from University of Washington Center for Ecogenetics and Environmental Health (P30ES007033). Peer review under responsibility of Institute of Materia Medica, Chinese Academy of Medical Sciences and Chinese Pharmaceutical Association. ![Expression of drug processing genes (DPGs) in small and large intestine of mice. The mRNA abundance of DPGs was retrieved from BioGPS (Biogps.org, Su et al.[@bib26]). The abundance of DPGs was expressed as average probe intensity from the microarray data in BioGPS. DPGs that were highly expressed in at least one section of intestine and/or are known as CAR-target genes (Aleksunes and Klaassen[@bib24]) in liver were selected for the analysis (*n*=2 per tissue). For each graph, black bar represents DPG expression in the small intestine whereas red bar represents DPG expression in the large intestine.](gr1){#f0005} {ref-type="sec"} (WT mice only). Data are expressed as percentage of the housekeeping gene *β*-actin (*n*=4--5 per group). Data between control and TCPOBOP-treated groups were analyzed using a Student׳s *t*-test. Asterisks (\*) indicate statistically significant differences (*P*\<0.05) between control and TCPOBOP-treated WT mice in the same tissue.](gr2){#f0010} {ref-type="sec"}. Data are expressed as percentage of the housekeeping gene *β*-actin (*n*=4--5 per group). Data among multiple groups were analyzed using ANOVA followed by Duncan׳s *post hoc* test. Asterisks (\*) indicate statistically significant differences (*P*\<0.05) between control WT and TCPOBOP-treated WT mice at the same section of intestines. Pound signs (\#) indicate statistically significant differences (*P*\<0.05) between control *Car*^*−/−*^ and with TCPOBOP-treated *Car*^*−/−*^ mice. Dollar signs (\$) indicate statistically significant differences (*P*\<0.05) of the basal mRNA expression between control WT and control *Car*^*−/−*^ mice.](gr3){#f0015} {ref-type="sec"}. Data are expressed as percentage of the housekeeping gene *β*-actin (*n*=4--5 per group). Data among multiple groups were analyzed using ANOVA followed by Duncan׳s *post hoc* test. Asterisks (\*) indicate statistically significant differences (*P*\<0.05) between control WT and TCPOBOP-treated WT mice at the same section of intestines. Pound signs (\#) indicate statistically significant differences (*P*\<0.05) between control *Car*^*−/−*^ and with TCPOBOP-treated *Car*^*−/−*^ mice. Dollar signs (\$) indicate statistically significant differences (*P*\<0.05) of the basal mRNA expression between control WT and control *Car*^*−/−*^ mice.](gr4){#f0020} {ref-type="sec"}. Data are expressed as percentage of the housekeeping gene *β*-actin (*n*=4--5 per group). Data among multiple groups were analyzed using ANOVA followed by Duncan׳s *post hoc* test. Asterisks (\*) indicate statistically significant differences (*P*\<0.05) between control WT and TCPOBOP-treated WT mice at the same section of intestines. Pound signs (\#) indicate statistically significant differences (*P*\<0.05) between control *Car*^*−/−*^ and with TCPOBOP-treated *Car*^*−/−*^ mice. Dollar signs (\$) indicate statistically significant differences (*P*\<0.05) of the basal mRNA expression between control WT and control *Car*^*−/−*^ mice.](gr5){#f0025} {ref-type="sec"}. Data are expressed as percentage of the housekeeping gene *β*-actin (*n*=4--5 per group). Data among multiple groups were analyzed using ANOVA followed by Duncan׳s *post hoc* test. Asterisks (\*) indicate statistically significant differences (*P*\<0.05) between control WT and TCPOBOP-treated WT mice at the same section of intestines. Pound signs (\#) indicate statistically significant differences (*P*\<0.05) between control *Car*^*−/−*^ and with TCPOBOP-treated *Car*^*−/−*^ mice. Dollar signs (\$) indicate statistically significant differences (*P*\<0.05) of the basal mRNA expression between control WT and control *Car*^*−/−*^ mice.](gr6){#f0030} {ref-type="sec"}. Data are expressed as percentage of the housekeeping gene *β*-actin (*n*=4--5 per group). Data among multiple groups were analyzed using ANOVA followed by Duncan׳s *post hoc* test. Asterisks (\*) indicate statistically significant differences (*P*\<0.05) between control WT and TCPOBOP-treated WT mice at the same section of intestines. Pound signs (\#) indicate statistically significant differences (*P*\<0.05) between control *Car*^*−/−*^ and with TCPOBOP-treated *Car*^*−/−*^ mice. Dollar signs (\$) indicate statistically significant differences (*P*\<0.05) of the basal mRNA expression between control WT and control *Car*^*−/−*^ mice.](gr7){#f0035} {ref-type="sec"}. Data are expressed as percentage of the housekeeping gene *β*-actin (*n*=4--5 per group). Data among multiple groups were analyzed using ANOVA followed by Duncan׳s *post hoc* test. Asterisks (\*) indicate statistically significant differences (*P*\<0.05) between control WT and TCPOBOP-treated WT mice at the same section of intestines. Pound signs (\#) indicate statistically significant differences (*P*\<0.05) between control *Car*^*−/−*^ and with TCPOBOP-treated *Car*^*−/−*^ mice. Dollar signs (\$) indicate statistically significant differences (*P*\<0.05) of the basal mRNA expression between control WT and control *Car*^*−/−*^ mice.](gr8){#f0040} {#f0045} {#f0050} ###### RT-qPCR primer sequences Table 1 --------------  -------------- ###### Liver and Intestine regulation difference in *Car*-null control mice compared to WT control mice. Table 2 --------------  -------------- Basal expression of genes is shown as increased or decreased relative to that in *Car*-null mice. Up-regulation suggests CAR suppresses the basal expression of the gene and down-regulation suggests CAR is necessary in maintaining the constitutive expression of the gene. ^⁎^ Note: The liver data were obtained from Aleksunes and Klaassen[@bib24], 2012. ###### Liver and Intestine regulation difference in WT TCPOBOP-treated mice compared to WT control mice. Table 3 --------------  -------------- Basal expression of genes is shown as increased or decreased relative to that in wild type mice. (--) denotes none change. N/A: not available. ^⁎^ Note: The liver data were obtained from Aleksunes and Klaassen[@bib24], 2012. [^1]: These authors made equal contributions to this work.

{#sp1 .57} {#sp2 .58} {#sp3 .59} {#sp4 .60} {#sp5 .60-1} {#sp6 .61} {#sp7 .62} {#sp8 .62-1} {#sp9 .62-2} {#sp10 .63} {#sp11 .64} {#sp12 .65} {#sp13 .66} {#sp14 .67} {#f1 .60-1} {#f2 .62-1} {#f3 .62-2} {#f4 .62-2} {#f5 .62-2}

Background ========== Among the major groups of the phylum Arthropoda -- Chelicerata, Myriapoda, Crustacea, Hexapoda (insects and relatives) -- the crustaceans and insects are allies \[[@B1]\]. They are together classified as members of the Pancrustacea, although their reciprocal monophyly is currently disputed \[[@B2]-[@B4]\]. Despite this phylogenetic uncertainty for taxa that have likely diverged some 600 million years ago \[[@B5]\] the model crustacean *Daphnia*is expected to share genes that are central to arthropod biology and development with well studied insects, such as *Drosophila*, *Anopheles*, *Bombyx*and *Apis*. Indeed, gene-by-gene investigations have already demonstrated the functional conservation of selected loci involved in germline formation and embryonic patterning between representative crustaceans and insects \[[@B6]-[@B8]\]. Yet, these two classes of animals have also evolved in radically different environments; branchiopod crustaceans are adapted to aquatic habitats, while the insects are predominantly adapted to terrestrial habitats. It is therefore expected that proteins required for life in these particular environments will reflect the biotic and abiotic challenges faced by these particular taxonomic groups. Furthermore, model crustaceans like *Daphnia*(order Cladocera) have a highly specialized mode of reproduction called cyclical parthenogenesis \[[@B9]\], which involves environmental sex determination and is derived from obligate sex \[[@B10]\]. Thus, the genetic control of cyclical parthenogenesis may have arisen from modifications in the structure and/or the regulation of arthropod reproductive genes. Similar mechanisms may apply for a variety of other adaptations, including *Daphnia*\'s morphological transmutations in response to predator kairmones (called cyclomorphosis), their ability to shift from direct development into diapause within ephemeral habitats, and mechanisms for acclimating to both natural and anthropogenic stressors such as hypoxia or metal contamination. The evolution of these traits is expected to involve species-specific modifications of gene regulation, the restructuring of genes common to arthropods \[[@B11]\] and innovations unique to their aquatic habitats. Additionally, transitions in breeding systems and the origins of other adaptive traits probably also involve novel genes or lineage specific gene family expansions \[[@B12]\]. Comparative studies into the functional conservation of genes and the genetic basis of adaptation are made easier by the rapid development of genomic data and technologies. For example, cross-species comparisons within the Nematoda, based on over 265,000 expressed sequence tags (ESTs) from 30 species, indicate that roughly 40% of the 93,000 characterized genes have no known homologues within the phylum \[[@B13]\] while 23% of genes are unique to each species \[[@B14]\]. These large differences in gene content reflect (in part) the ecological diversity of the sampled nematodes, including free-living species and others that are plant or animal parasites. Not surprising, genetic novelty can be linked to an organism\'s specialized lifestyle. For instance, unique sequences of the parasitoid nematode *Nippostrongylus brasiliensis*are nearly 10 times enriched with signal peptides compared to conserved sequences, suggesting that the proliferation of these genes is accelerated because of their defensive role against host immunity \[[@B15]\]. Aside from the deeply divergent nematode comparisons \[[@B16]\], studies have thus far been restricted to the eukaryotic crown group \[[@B17]\], contrasts among species from the same order (i.e., primates, rodents) or species belonging to a similar class (insects). This situation is a consequence of the currently sparse coverage of genome sequencing projects along the metazoan phylogenic tree. Therefore, the addition of a crustacean to the growing list of sequenced insect genomes will expand the analysis of gene content among the ecologically diverse arthropod assemblage and provide information on the degree of protein family expansions by appropriately rooting the insect phylogeny. Given the diversity of crustacean body plans, their fascinating biology and their key phylogenetic relationship to model invertebrates with sequenced genomes, the paucity of crustacean molecular data is striking. Indeed, protein sequences from *all*crustaceans represent only 0.1% of 6.9 million records in the NCBI taxonomic database. Among crustaceans, the freshwater zooplankton *Daphnia pulex*has a rich history of attracting attention from biologists -- which now involves researchers in the fields of ecology and evolution, development, toxicology and genetics. Here, we present the first systematic study of transcribed sequences in *D. pulex*. The results of our survey highlight the diversity of crustacean genes that are shared with insects, and also uncovers gene family expansions that likely reflect the demands of aquatic existence, particularly homeostasis, defense/immunity, oxyregulation, and chemical sensing. In companion papers, we describe the development of the first *D. pulex*microarray used to investigate sex-biased transcriptional regulation of these genes (Eads et al. submitted) and the genomic response of this sentinel species to toxic metals commonly found in the environment (Shaw et al. submitted, and in prep). These studies are the initial steps in a coordinated effort by the *Daphnia*Genomics Consortium \[[@B18]\] to build the necessary data banks and reagents needed to discover genomic changes responsible for the phenotypic diversity within the genus and to gain new insights into crustacean biology. This effort will soon include the first crustacean genome sequence. We report on the construction of *D. pulex*cDNA libraries and the sequences and analyses of 1,546 ESTs, of which 1,414 represent approximately 787 nuclear genes. We analyze these transcribed sequences against those of sequenced model invertebrates. Comparing gene inventories by assigning homology among distantly related genomes is not trivial \[[@B19]\]. The first challenge is to discriminate between genetic gains or losses and genes whose sequences are sufficiently divergent to escape detection. The problem is exacerbated by lineage specific gene or genome duplications, by varying rates of molecular evolution and by the sometimes fragile association between sequence similarity and the preservation of gene functions \[[@B20]\]. The second challenge is to recognize that reference genome annotations and data banks are fluid, even those for premier model systems. Therefore, this study uses sequence similarity searches for *Daphnia*genes against a number of different genomic databases for five reference species while intentionally setting low statistical cut-off values. By comparing *Daphnia*sequences to genes from four insect species and using *Caenorhabditis*for an outgroup, we point to functional classes that are shared with the insects. The related microarray data of Eads et al. and Shaw et al. in companion papers demonstrate that most of the sequenced *Daphnia*genes are differentially transcribed in a manner consistent with their putative functions, thus reinforcing their provisional annotations based on sequence alignments to genes from model insects. Our study details the comparative and functional characterization of *Daphnia*transcripts using well studied insects and a phylogenetic approach. Results ======= Production and quality assessment of cDNA libraries --------------------------------------------------- Equivalent non-normalized cDNA libraries were constructed from a genetically clonal *Daphnia*isolate sampled from a natural pond along the Oregon coast. The clone was cultured under growing conditions favoring parthenogenetic reproduction. Consequently, the animals were predominantly juvenile females, adult females and brood-carrying females with a small proportion of males. The strength of conclusions derived from the comparative analysis of expressed gene sequences rests, in large part, on the quality of cDNA libraries. Therefore, we performed quality control tests on 768 randomly chosen cDNA isolates. The cDNA size distribution was determined by agarose gel electrophoresis of PCR amplified inserts. The average molecular weight of inserts sampled from the libraries was 825 bp. To assess the cDNA diversity within the libraries, we sequenced single pass 5\' reads from the cDNA inserts. Of the 768 sequence reads, 619 were informative. Only four plasmids were void of inserts and the failed reads (19%) were the result of capillary failures of the sequencer. Following an assembly of the ESTs, unique sequences comprised 68% of the total, reflecting the relative abundances of specific cDNA within the non-normalized libraries. This number diminished to 43% with over twice as much sequencing effort (see below). To assess potential contamination of cDNA clones derived from prokaryotes and mitochondria, and to measure the distribution of full length ORFs, we aligned the translated sequences to proteins in Genbank. Of the unique sequences, 50% matched Genbank entries with e-values \< 1 × 10^-10^. A separate query of the NCBI non-redundant protein database identified 204 sequences with e-value scores \< 1 × 10^-27^. A total of 34 ESTs (6%) were identified as mitochondrial gene transcripts. No ESTs were identified as non-*Daphnia*. Thus, the cDNA libraries are high quality with a high level of diversity and low levels of contaminant sequences. To investigate whether the libraries contained full-length or nearly full-length inserts, sequences from 170 clones with high similarity to known proteins (Blastx \< 1 × 10^-27^) were investigated for the presence of a translational start site. Of these sequences, 109 ESTs (64%) contained unambiguous open reading frames with an annotated ATG translational start site at their 5\' end, and 44 ESTs (26%) did not contain an ATG that aligned with the start sites of corresponding database sequences. Of the remaining sequences, 7 ESTs (4%) were likely full-length because gapped alignments of the amino acids suggested poor evolutionary conservation at the N-terminus of the proteins, and 10 ESTs (6%) were unresolved because alignments failed altogether. We therefore estimate that 64--68% of the cDNAs are full-length, or close to full-length. This result may be an overestimate since many conserved genes within our non-normalized libraries encode for ribosomal proteins (34% of 170) which seldom have long transcripts. Indeed, the maximum length of investigated cDNA for open reading frames was \< 2 kb, whereas the maximum length of PCR amplified inserts was nearly 3.5 kb. However, when the number of cDNA with and without annotated start sites were compared and sorted by their molecular weights, no association was found between the proportion of full-length transcripts and the size of cDNA, neither by including ribosomal genes (t = 0.39, df = 86; p = 0.70) nor by excluding these genes in the comparison (t = 0.19; df = 60; p = 0.85). A separate investigation of the consistency in our production of full-length or near full-length cDNA was conducted by calculating the proportions of sequences that shared nucleotides within the first 50 bases of the longest EST within contigs (see below for assembly of contigs). Of 233 ESTs forming 81 separate contigs, 202 (87%) shared the first 50 bp of the longest EST from each contig. These data suggest that the majority of the cDNA clones are near full-length. Analysis of EST sequences ------------------------- In total, we produced 5\' sequence reads from 1,648 cDNA isolates. In addition to the 768 randomly selected clones, 880 were selected on the basis of their transcription profiles in microarray experiments. After the removal of vector, poly-A tails and poor quality reads (Table [1](#T1){ref-type="table"}), 1,546 high quality ESTs with an average size of 540 bp (SD = 188, min = 107, max = 852 bp) remained to be clustered (Genbank accession numbers [EE681877](EE681877)-[EE683416](EE683416)). The ESTs were assembled into 804 clusters (including 568 singletons) with an average of 1.93 sequences/cluster (SD = 3.95, min = 1, max = 95). After excluding clusters identified as mitochondrial DNA sequences, 787 nuclear genes remained. These non-redundant sequences are hereafter referred to as assembled sequences. We expect that some pairs of assembled sequences will be found to derive from the same locus, either due to excessive polymorphisms between alleles, or because of the alternative use of 5\'-exons, or due to sequences from truncated cDNA clones that failed to overlap. However, given the high proportion of estimated full-length clones in the libraries, we anticipate the latter class to be small. ----------------------------------------------------- ------- Number of sequenced cDNA isolates 1,529 Number of sequences obtained: 1,648 Number of low quality sequences removed 82 Number of plasmids containing inserts \<100 bp 20 Number of cDNA isolates with ESTs 1,435 Number of ESTs left to assemble 1,546 Number of assembled sequences (contigs + singlets) 804 Number of cDNA isolates represented by a single EST 612 Number of mtDNA gene clusters 17 or Number of mtDNA ESTs 132 Number of assembled sequences from nuclear genes 787 ----------------------------------------------------- ------- Sequencing and clustering statistics for cDNA isolates printed on microarrays. Based on these numbers, 43% of the elements are unique. We investigated the proportion of assembled sequences that may be composed of alternative transcripts of the same genes by further clustering these sequences using more relaxed parameters (see methods). Forty-seven additional clusters were discovered. Eleven were composed of 2 assembled sequences that are likely allelic variants of the same genes, based on their matches to a single location in a preliminary draft assembly of the *Daphnia*genome sequence at the mid-point of the genome sequencing project (4 × coverage; deposited at wFleaBase). These allelic sequences were 84% to 97% similar to each other over a range from 77 to 665 overlapping nucleotides. By contrast, 21 of the additional clusters were of transcripts derived from duplicated genes or from conserved gene families, based on their matches to different locations in the draft genome sequence. In 14 cases, the clusters were composed of 2 sequences. Six clusters consisted of 3 sequences and a single cluster contained 4 similar sequences that shared 86--93% of their nucleotides in pair-wise comparisons. Overall similarities between sequences from closely related genes ranged from 62% to 93% among 220 to 807 overlapping bases. As expected for sequences originating from separate loci, their average similarity (85.5%) was significantly lower than that of allelic sequences (91%)(t = 2.02; df = 44; p = 0.01). Finally, 4 of the additional clusters were composed of paired splice variants from unique loci, while 8 clusters contained from 2 to 4 alternatively spliced transcripts from multiple loci. Therefore, the ESTs from this survey provided sequence tags for up to 787 new *Daphnia*genes, where some genes represent alternative transcripts or are closely related transcripts from duplicated genes. Functional annotation of assembled sequences -------------------------------------------- Confident over the quality of the *D. pulex*cDNA libraries and EST sequences, we explored the range of likely biological or biochemical functions of the genes represented by the ESTs sequences by querying the NCBI non-redundant protein databank (NR) using Blastx \[[@B21]\]. Of the 787 assembled sequences, 452 (58%) matched at least one known protein with an e-value threshold of 1 × 10^-3^and a minimal value of 33 aligned amino acids (Additional file [1](#S1){ref-type="supplementary-material"}). The distribution of their e-value scores showed that 26% of matched sequences have scores \< 1 × 10^-50^, while 79% have scores \< 1 × 10^-10^(Figure [1a](#F1){ref-type="fig"}). Therefore, searches for putative homologues in the protein database gave strong to suggestive information regarding possible biological and biochemical functions. As expected, a survey of the distribution of best Blastx matches against the NCBI taxonomic domains showed that the majority (72%) of assembled *Daphnia*sequences matched best with those derived from other invertebrates (Figure [1b](#F1){ref-type="fig"}), whereas 25% of the highest scoring hits matched best with vertebrate sequences (including those from rodents, primates and other mammals). The cDNA libraries are free of contaminants, as only 6 assembled sequences (1%) matched bacterial proteins. A closer examination of the distribution of the best Blastx hits within the classes of invertebrates showed that 79% of 323 assembled sequences matched annotated proteins from insects (Figure [1c](#F1){ref-type="fig"}): 23% from *Drosophila*, 16% from *Anopheles*, 15% from *Apis*, 4% from *Bombyx*and 21% from other insects. This large insect constituency within the best Blastx matches is clearly a consequence of the limited representation of sequences from Crustacea in the databanks. A survey of the NCBI protein database revealed that out of 6,897,314 archived sequences, only 10,485 (or 0.1%) were from Crustacea. Only 19 assembled sequences (6%) best matched proteins from Branchiopoda, the class that includes *Daphnia*, while an additional 10 assembled sequences best matched proteins from other classes of Crustacea (Malacostraca, Ostracoda). {#F1} The assembled *Daphnia*sequences that matched annotated proteins from genetic model species were assigned Gene Ontology (GO) terms using Blast2GO \[[@B22]\]. Their putative functions spanned a spectrum of biological and biochemical processes (Figure [2a](#F2){ref-type="fig"}). A total of 227 assembled sequences were assigned 799 biological process terms from the fourth level of the GO. The predominant terms were for metabolic processes ascribed to 190 assembled sequences. These terms included cellular metabolism (22%), primary metabolism (21%), macromolecule metabolism (18%), biosynthesis (10%), biopolymer metabolism (7%), catabolism (5%) and regulation of metabolism (\< 1%). From among these processes, 72 assembled sequences were annotated to involve protein biosynthesis, while 30 sequences involved the catabolism of proteins. Sixteen sequences were attributed roles in chitin metabolism, including chitinases and peritrophins. The next most predominant biological process terms were related to the localization of cellular components (establishment of localization, transport, protein localization), which were ascribed to 33 assembled sequences. Four of these sequences coded for genes involved in oxygen transport (hemoglobin). Thirteen of these sequences encoded genes with putative homologues in insects that specifically transport charged atoms like metals, of which seven were also ascribed the functions of cell and ion homeostasis (Figure [2a](#F2){ref-type="fig"}). Finally, the remaining assembled sequences with GO biological process terms were likely involved in cell communication (such as signal transduction, cell signaling and adhesion), development, and physiological processes that specify a response to external stimuli, stress and cell death. ![The distribution of gene annotations for the list of 787 *Daphnia pulex*ESTs based on results from Blastx searches against the NCBI non-redundant protein database. (A) The assignment of 799 annotations of biological process to 227 EST clusters from level 4 of the Gene Ontology. (B) The assignment of 371 annotations of molecular function to 288 EST clusters from level 3 of the Gene Ontology. Blastx queries recorded the best 5 matches with an E-value threshold of 1 × 10^-3^and a minimal value of 33 aligned amino acids. Gene Ontology (GO) terms were assigned to ESTs using Blast2GO \[22\] with the following configurations: Pre-eValue-hit filter 1 × 10^-3^; Pre-similarity-hit filter 2; Annotation cut-off 35; GO weight 5.](1471-2164-8-217-2){#F2} A total of 288 assembled sequences were additionally assigned 371 GO molecular function terms (Figure [2b](#F2){ref-type="fig"}). One hundred and thirteen sequences were suggested to have catalytic activities that included hydrolase (17%), transferase (6%), oxidorectuctase (4%), ligase (1%) and isomerase activities, among others. Another 105 assembled sequences were likely involved in structural activities. Their GO terms included structural constituent of the ribosome (18%), structural constituent of the cuticle (9%) and structural constituent of the cytoskeleton (1%). Indeed, *Daphnia*genes matched to 67 of the total number of 194 listed *Drosophila*ribosomal components. The next major functional class was represented by 86 sequences putatively involved in binding, which included 27 sequences coding for nucleic acid binding proteins, 18 carbohydrate (also listed as pattern and chitin) binding proteins, 17 nucleotide binding proteins, and 15 proteins that bind ions (calcium, zinc, iron). The remaining 11 sequences within this class were protein or lipid binding. The final major functional class contained 26 assembled sequences assigned to have transporter or carrier activities; twelve sequences were annotated as ion transporters. A number of assembled sequences likely encode conserved proteins involved in gene regulatory functions. *Daphnia*genes with such potential functions based on sequence homologies included 19 sequences involved in transcription regulation and 14 sequences with translational regulator activity (Additional file [2](#S2){ref-type="supplementary-material"}). Examples of regulatory genes involved in arthropod development included a putative homologue to *maf-S*, which is a basic-leucine zipper (bZIP) transcription factor in *Drosophila*that is required for the development of pharyngeal structures \[[@B23]\]. A *Daphnia*sequence also matched closely to the *Dorsal switch protein 1*(*Dsp1*) gene that regulates a number of homeotic genes in *Drosophila*\[[@B24]\] and is therefore involved in many developmental pathways. The putative homologue to the fly gene *shaggy*(*sgg*) was identified, which is part of the Notch, Wnt and Smoothened signaling pathways. Interestingly, two other regulators of the Notch signaling pathway called *Cdc42*\[[@B25]\] and *neurotic*(*nti*or *O-fut1*) were also identified (Additional file [3](#S3){ref-type="supplementary-material"}). The fly gene *nti*is specifically required for the proper localization of *Notch*at the cell surface \[[@B26]\], is essential for the physical interaction of *Notch*with its ligand *Delta*, and is an essential component for neurogenesis \[[@B27]\]. A second gene involved in neurogenesis was identified as homologous to *similar to Deadpan*(*Side*), which is a basic helix-loop-helix (bHLH) transcription factor. A *Daphnia*gene was also matched with the fly zinc-finger-C4 transcription factor (ZnFC4) called *ftz transcription factor 1*(*ftz-f1*), which coordinates stage-specific responses to the steroid hormone ecdysone during metamorphosis \[[@B28]\] and directs key developmental events at the transition between prepupal and pupal stages of *Drosophila*development \[[@B29]\]. The putative functions for the other identified transcription regulators (Additional file [2](#S2){ref-type="supplementary-material"}) included the regulation of mitotic progression (the TFIIH transcription factor *Cdk7*) and important roles during gametogenesis (*Rab11, bic*, and the C2H2-zinc finger transcription factor *Meics*). Other putative transcription factors included genes matching *CG1876I*that contains a helix-turn-helix (HTH) DNA-binding motif, *CG18619*that contains a bZIP DNA-binding motif and *CG3224*that contains a putative zinc-finger DNA-binding motif. Finally, among the regulators of translation (Additional file [2](#S2){ref-type="supplementary-material"}), *Daphnia*genes matched to 7 of the total number of 24 listed *Drosophila*translational elongation genes, yet matched only to 2 of the 58 listed translational initiation genes in flies. Their putative functions include DNA repair (*RpLP0*), autophagic cell death (*eIF-5A*, *Ef1gamma*), immune response (*RpS6*, *Thor*), regulation of cell growth (*Thor*) and germ-line stem cell division (*piwi*). To discover *Daphnia*genes that may jointly participate in conserved biological processes or within gene interaction networks, we investigated GO classes that were highly represented within our list of putative homologues to fly proteins. Significant functional groupings of *Daphnia*genes were expected, because more than half of the sequenced ESTs were chosen based on their differential expression patterns in separate microarray experiments examining developmental differences among males and females, juveniles and adults, and toxicological responses to metals (Eads et al. submitted; Shaw et al. submitted). Seventeen genes were identified as candidates for gametogenesis (Additional file [4](#S4){ref-type="supplementary-material"}); the majority of these loci are involved in oocyte development in flies. *Daphnia*loci matching the fly genes *mago*, *Rab11*, *chic*, *Tm1*, *tsu*and *bic*may play conserved evolutionary roles in specifying the anterior-posterior axis of the oocyte. All 6 genes save *bicaudal*(*bic*) coordinate to assemble the pole plasm at the posterior end of the *Drosophila*oocyte by localizing maternally derived transcripts for *oskar*. Moreover, *mago*, *Tm1*and *Rab11*are known to interact in genetic screens \[[@B30]\], while a two-hybrid-based fly protein interaction study \[[@B31]\] implicates *tsu*and possibly a translational elongation *RpLP1*homologue (Additional file [3](#S3){ref-type="supplementary-material"}) within this network. One other *Daphnia*gene with weak sequence similarity to *Sop2*may have transport functions during oogenesis \[[@B32]\] and two genes similar to the signaling gene *Cdc42*and to *tsr*were respectively identified, which are involved in follicle cell development \[[@B33],[@B34]\]. Other genes similar to *snf*, *RpS3A*and *sgg*in flies are also candidate for oogenesis. Only two genes from our survey have known homologues in flies that function in spermatogenesis. The gene *Act5C*has a role in sperm individualization \[[@B35]\] and *Meics*is associated with central spindle and mid-body microtubules during meiosis \[[@B36]\]. The *chic*gene in flies is important in gametogenesis for both sexes \[[@B37]\]. These *Daphnia*genes sharing sequence similarities with *Drosophila*loci, which are known to coordinate conserved developmental processes, are prime candidates for functional investigations of early crustacean development. In contrast to genes that participate in biological processes that are shared between Crustacea and Insecta, gene families that have expanded in *Daphnia*compared to insects may be indicative of new gene functions linked to their specific biology and ecological setting. Therefore, we identified GO classes that were overrepresented within our dataset compared to the Gene Ontologies for the *D. melanogaster*proteome using Fisher\'s exact test, and counted multiple assembled sequences that matched to unique *Drosophila*proteins (Additional file [1](#S1){ref-type="supplementary-material"}). Some lineage expansions seemed to occur primarily by tandem duplication, while other radiations implied interesting functional specializations or innovations. Among the 53 *Daphnia*sequences that were provisionally annotated as cuticle proteins and genes involved in chitin metabolism and molting, only 13 singularly matched to a fly gene (Additional file [5](#S5){ref-type="supplementary-material"}). In most cases, 2--4 assembled sequences matched to the same protein in the fly genome. Yet in another case, 15 sequences matched with the *D. melanogaster*gene CG6305, which contains an insect cuticle protein domain. Because we could not produce a reliable sequence alignment for all 15 assembled sequences, we conclude that the observed gene expansion was not an artificial result from inadequate clustering of redundant ESTs. Yet, from the pairwise comparisons of these 15 sequences, alternative splice variants for three genes were identified: Contigs 20 and 180 showed \>90% sequence similarity, Contigs 23, 241 and 257 were \>94% similar, and Contigs 19 and 24 were \>85% identical over shared exons. Additional cDNA sequence data aligned to a completed genome sequence assembly for *Daphnia*is needed to confirm that cuticle proteins are expanded gene families compared to insects. However, our study also uncovered clearer examples of gene expansions. Unlike insects, which have three ferritin genes that play important roles in iron homeostasis of cells (Fer1HCH, Fer2LCH) and of organelles (Fer3HCH), the annotation of *Daphnia*sequences revealed seven assembled sequences with strong matches to *Drosophila*ferritin proteins (Table [2](#T2){ref-type="table"}). Singlet 73 showed a strong match to the *Drosophila*Fer1HCH protein via a Blastx alignment (bit score = 137), but was poorly matched to crustacean sequences, even to a *D. pulex*ferritin sequence within Genbank (AJ245734; bit score = 71). The remaining six *Daphnia*sequences, plus the Genbank entry, all aligned best to other crustacean sequences and to the single Fer3HCH locus of *Drosophila*. Therefore, Singlet 73 represents the first sequence of an orthologous crustacean Fer1HCH gene. This result was verified by constructing a phylogeny using representative insect and crustacean protein sequences and by including additional *Daphnia*ferritin-like sequences that were extracted from an ongoing *D. pulex*cDNA sequencing project (Colbourne et al. in preparation). *Daphnia*ID *Drosophila*gene ID *Drosophila*gene name FlyBase ID \% similarity E-value Bit score ------------- --------------------- ----------------------- ------------- --------------- ---------- ----------- Singlet 73 CG2216 *Fer1HCH* FBgn0015222 39.81 4.00E-33 137 Contig 91 CG4349 *Fer3HCH* FBgn0030449 39.66 6.00E-27 117 Contig 26 CG4349 *Fer3HCH* FBgn0030449 38.22 8.00E-23 103 Contig 138 CG4349 *Fer3HCH* FBgn0030449 40.61 1.00E-22 103 Contig 217 CG4349 *Fer3HCH* FBgn0030449 40.16 4.00E-14 74.7 Contig 40 CG4349 *Fer3HCH* FBgn0030449 33.75 2.00E-08 54.7 Contig 42 CG4349 *Fer3HCH* FBgn0030449 26.25 2.00E-06 47.8 *Daphnia*genes annotated as candidates for iron ion homeostasis based on sequence conservation with *Drosophila*genes with known functions. Contig 26 and 138 are two alleles from the same locus. Contigs 40, 42 and 91 are also sequence variants from the same locus. The Neighbor-Joining tree of 35 aligned amino acid sequences clustered the pancrustacean ferritins into three main groups (Figure [3](#F3){ref-type="fig"}). Ferritin 1 contained insect genes plus two cDNA from *Daphnia*libraries; the Singlet 73 amino acid sequence was identical to a sequence extracted from other cDNA libraries (branch G, Figure [3](#F3){ref-type="fig"}), while branch F was a unique gene that stemmed at the base of the group. The ferritin 2 group was solely composed of insect genes. However, both crustacean and insect sequences clustered into the third group containing the insect ferritin 3 genes. Although this group contained single copies of the insect genes, the *Daphnia*genes were further subdivided among five branches representing distinct ferritin 3 loci within the *D. pulex*genome. Branches D and E were gene sequences derived from the other libraries and showed 45% sequence divergence from each other, while branches A, B, C each contained at least one sequence from this present study. Further investigations indicated that the multiple sequences clustering within branches A, B and C are different alleles of the same locus. Given that insects and more distant outgroups have only three ferritin genes, *Daphnia*ferritins clearly expanded to include *possibly*one additional ferritin 1 locus and *minimally*four additional ferritin 3 genes. ![Lineage specific expansion of the *Daphnia pulex*ferritin genes. Neighbor-Joining (NJ) tree inferred from the deduced amino acid sequences of the *Daphnia*ferritin genes, including loci from *Drosophila melanogaster*plus other representative insect and crustacean amino acid sequences obtained from the NCBI and FlyBase protein sequence repositories. Ferritin 1 group contains insect and *Daphnia*Fer1HCH gene(s). Ferritin 2 group only contains insect Fer2LCH loci and the Ferritin 3 group contains insect and crustacean Fer3HCH genes. The amino acid sequence alignment was obtained by using t-coffee \[72\] and is available by request. The NJ tree was constructed using MEGA3 \[73\] using the Poisson correction for calculating the distance matrix. The bootstrap support values are shown at the main branch nodes of the tree, which are derived from 1000 pseudo-replication of the data. *D. pulex*sequences denoted by \* were obtained from an ongoing cDNA sequencing project by the Joint Genome Institute and the *Daphnia*Genomics Consortium (Colbourne et al. in prep) and are deposited in Genbank under accession numbers [DQ983425](DQ983425)-[DQ983438](DQ983438). GenInfo (GI) accessions for all other sequences: 6946692; 61744051; 26006755; 46561742; 91081285; 87083910; 66504201; 1807496; 13195275; 55242312; 66524157; 91077442; 24651358; 95702694; 18031707; 62722854; 6409191; 91077446; 66524161; 7272336; 62722856.](1471-2164-8-217-3){#F3} Gene conservation between the crustacean *Daphnia*and the true insects ---------------------------------------------------------------------- To examine the conservation of genes represented by the EST sequences across Pancrustacea, we used tBlastx to match the *Daphnia*assembled sequences to the NCBI UniGene sets for *Drosophila melanogaster*, *Anopheles gambiae*, *Bombyx mori*, *Apis mellifera*and included *Caenorhabditis elegans*for an outgroup. The assembled sequences were clustered into 27 groups, based on the strength of these sequence matches across these taxonomic data sets. This arrangement identified a variety of gene classes that share patterns of sequence conservation (Figure [4](#F4){ref-type="fig"}). The first class of interest was composed of 124 genes (16%) that are conserved equally among all species included in this study. This class was mostly enriched by genes that participate in protein metabolism including protein modifications (67 genes to GO:0044267; p = 4.8 × 10^-7^) plus 32 genes that were likely involved in cellular metabolism (total of 99 genes to GO:0044237; p = 3.2 × 10^-8^). Other genes enriched within this class included transcriptional regulators (12 genes to GO:0045449; p = 2.4 × 10^-3^). The second class of interest was composed of *Daphnia*sequences that matched a nematode protein plus at least one insect locus (183 genes) and others that had no matches to nematode proteins yet matched to at least one insect proteome (167 genes). Therefore, 21% of the sequences were derived within the Pancrustacea -- thus shared by *Daphnia*plus at least one insect in our set -- or lost within the nematode. Especially noticeable were 43 assembled sequences that had no detectable homologues in worms and were uniformly conserved across the four insects (Figure [4](#F4){ref-type="fig"}). Genes that were absent in nematodes were enriched with structural constituents of the cuticle (26 genes to GO:0042302; p = 1.1 × 10^-12^) and loci having serine-type endopeptidase activity (10 genes to GO:0004252; p = 3.6 × 10^-4^), of which 8 genes were annotated as also having chymotrypsin activity (GO:0004263; p = 0.003). The third class of interest consisted of 309 genes (39%) that had no matches to insect proteomes. At face value, this result suggests that these orphaned genes are unique to *Daphnia*or Crustacea; either they have been lost in the insects -- as suggested by 17 *Daphnia*genes (2%) showing sequence similarity to proteins in the nematode database -- or acquired by *Daphnia*or Crustacea since diverging from their last common ancestor. {#F4} There are many potential sources of errors that can inflate our estimate of the fraction of unique *Daphnia*genes compared to the selected pancrustaceans. For instance, sequences may fail to align for technical reasons. This may occur if the *Daphnia*sequences included untranslated regions (UTR) of the cDNA and not the coding regions. Indeed, the mean size of predicted open reading frames (ORFs) within this class differed significantly from that of genes having sequence matches to insect proteomes (t = 12; p \< 0.0001; df = 785). For example, over half of the assembled sequences with no matches had ORFs smaller than 225 bases compared to 16% of matched sequences (Figure [5](#F5){ref-type="fig"}). Therefore, the trivial explanation that these sequences were mostly UTR cannot be dismissed for a large fraction of these genes. Other technical explanations for the absence of matches include genes that had not been annotated or included in the insect UniGene sets. Further analysis by aligning the non-matching sequences to all predicted *Drosophila*gene translations uncovered 9 additional matches with e-values ranging from 4 × 10^-3^to 2 × 10^-10^. Another 4 *Daphnia*sequences were found to have matches to *Drosophila*proteins, based on tBlastx searches against the full genome sequence (e-values ranging from 4 × 10^-3^to 9 × 10^-29^). Finally, given the tremendous evolutionary divergence between *Daphnia*and insects, matches may not have been detected from loci that are not under similar evolutionary constraints. We are unable to investigate this last point with the current data. {#F5} It was previously shown that gene preservation is correlated with gene function. In particular, correlations have been found between the level of gene conservation and sex-biased gene expression among insects \[[@B38],[@B39]\]. There is reason to believe that such correlations are extended to other biological functions. In comparing 787 *Daphnia*assembled sequences to those of insects, 39% of the genes were characterized as orphans because no sequence matches were detected. Interestingly, the orphan genes were not randomly distributed among the gene expression classes. Three specific observations were made by incorporating the gene expression datasets (Table [3](#T3){ref-type="table"}). First, 47% of male-biased genes did not match insect proteins compared to only 22% of female-biased genes. The two fold difference in sequence similarity among the sex-biased genes in *Daphnia*is consistent with differences seen among the insects, reflecting the overall accelerated evolution of male reproductive genes \[[@B40]\]. Second, 46% of metal responsive genes did not match insect proteins compared to only 34% of non metal responsive genes. Third, genes that were responsive to metals and not sex-biased included the greatest proportion of orphans (50%), whereas genes that were female biased and not responsive to metals included the fewest (12%). These results suggest that lineage specific genes are correlated with certain biological functions associated with an organism\'s ecological challenges. Response to metals Up in males Up in females No change between sexes Total -------------------------------------- --------------- --------------- ------------------------- --------------- Responsive to cadmium and/or arsenic 45% (85/188) 42% (25/59) 49% (47/96) 46% (157/343) No change in both metals 48% (95/197) 12% (13/112) 33% (44/135) 34% (152/444) Total 47% (180/385) 22% (38/171) 39% (91/231) 39% (309/787) The percentage of *Daphnia pulex*assembled EST sequences with no matches to insect proteins, partitioned by their differential expression patterns in experiments designed to detect transcriptional differences between the sexes (Eads et al. submitted) and genes responding to toxic metal exposure to cadmium and arsenic (Shaw et al. submitted, and in prep). A 5% false discovery rate is applied to all three experimental results. The number of orphan genes over the total number of genes within the partition is indicated in parentheses. Discussion ========== Diversity in the gene complement of species arises from the expansion of shared ancestral gene families, the loss of existing genes, or the acquisition of newly invented genes \[[@B12],[@B41]-[@B43]\] and can account for lineage specific innovations. It is estimated that nearly half of the paralogous gene families within eukaryotic genomes originated by lineage specific gene expansions; many are related to an organism\'s unique mode of life \[[@B12]\]. For example, the evolution of disease resistance and of self-incompatibility in plant mating systems can partly be attributed to the radiation of novel receptor-like kinases within the plant genome \[[@B44]\]. In *Drosophila*, the trypsin-like serine proteases have expanded to 178 genes \[[@B45]\], suggesting important novel defenses by the fly immune system. Odorant receptors form the largest recorded nematode-specific gene family expansion -- numbering \~800 genes compared to 60 genes in flies \[[@B46]\] -- which suggests the importance of chemosensing in the soil environment. A similar genomic inventory for a branchiopod crustacean genome will soon be made available by the *Daphnia*Genomics Consortium to ultimately contrast the evolutionary diversification of arthropod genes in relation to the aquatic and terrestrial habits of these animals. Yet, crustaceans and insects also share many key biological features due to their common ancestry as members of the Pancrustacea. This study presents the results of a first investigation into the sequence conservation and putative function of *D. pulex*genes, which are identified by sequencing a set of 1,648 cDNA isolates that were interrogated in three microarray studies. From clustering the ESTs, we characterize 787 *Daphnia*loci based on their sequence similarity to genes within a variety of databases, including those for the insects *Bombyx*, *Apis*, *Anopheles, Drosophila*, and for the nematode *Caenorhabditis*. Shared genes ------------ In this study, we characterize two non-normalized cDNA libraries from a clonal population reared under standard laboratory conditions. As a result, the diversity of biological processes and molecular functions represented among the sequenced *Daphnia*genes is relatively modest. Almost one quarter of the genes are likely involved in metabolic processes. More than one quarter of the genes are predicted to have catalytic or structural activities. Although a large fraction of arthropod genomes is composed of genes having these basic cellular functions (35--40% of *Drosophila*genes for instance), the diversity of transcribed genes discovered in *Daphnia*would be augmented by creating libraries from animals under a variety of environmental conditions. Yet, our libraries do contain cDNA from daphniids of mixed life stages, including gravid females, embryos, juveniles and a small number of males. Therefore, some *Daphnia*genes have sequence similarity to insect proteins associated with reproduction, development and growth. These include regulatory genes like *Dsp1*, which operates in patterning the developing fly embryo by acting as a corepressor of the transcriptional regulator Dorsal protein \[[@B47]\]. Under different circumstances, *Dsp1*can also act as an activator or repressor of thorax-group and polycomb-group homeotic genes in *Drosophila*\[[@B24]\]. Because all known homeotic targets of *Dsp1*are conserved in sequence and function in metazoans, *Daphnia*\'s putative orthologue is likely to share this regulatory function. As expected, this gene transcript is enriched in pregnant females compared to males in *Daphnia*microarray experiments (Additional file [2](#S2){ref-type="supplementary-material"}). Putative homologues to three regulatory genes (*sgg*, *Cdc42*, *O-fut1*) within the Notch signaling pathway are identified, which is one of a small number of signal transduction pathways that are highly conserved in insects and throughout animal evolution. The gene *sgg*is a point of convergence between the Notch and Wnt/wingless signaling pathways \[[@B48]\]. We predict that further sequencing of *Daphnia*cDNA will uncover more genes operating within these and other conserved signaling mechanisms: nuclear receptors, Sonic Hedgehog, receptor tyrosine kinases, JAK/STAT, and BMP/TGF-beta. An example is a homologue to the *ftz-f1*nuclear hormone receptor that is also found on the microarray. Like all arthropods, *Daphnia*growth is synchronized with molting and the regeneration of the cuticle, which is governed by pulses of ecdysteroid hormones. Although one isoform of *ftz-f1*is a transcriptional regulator of the embryonic segmentation gene *fushi tarazu*, a second isoform is necessary for larval molting in *Drosophila*and its premature expression results in the disruption of the epicuticle, suggesting that targets for this transcription factor in flies include genes involved in cuticle formation \[[@B49]\]. The function of this gene is conserved in *Caenorhabditis*, where it is also required for epidermal development and regulates molting \[[@B50]\]. No differential expression is observed on the microarray for the signaling genes discussed above (Additional file [3](#S3){ref-type="supplementary-material"}). Overall, our survey of *Daphnia*ESTs uncovers genes expected to be present in crustacean genomes based on their important regulatory roles in conserved cellular and developmental processes. Roughly half of the cDNA isolates that are sequenced for this study are chosen based on their differential expression patterns between males and females and on their responses to toxic metals. These experimental conditions reflect two research interests of our labs involving *Daphnia*: the genetic basis of environmental sex determination and cyclical parthogenesis, and understanding how populations adapt to environmental change in aquatic habitats including industrial pollutants. Therefore, we expect that the provisional annotations of 787 assembled sequences include a fraction of genes sharing functional attributes that would in part be shared with other arthropods and others that are more reflective of *Daphnia*\'s unique biology. Certainly, 17 genes are candidates for gametogenesis. The majority of these genes play significant roles in oogenesis. In particular, six sequences have strong matches to conserved genes that specify the oocyte polarity and four loci are known to genetically interact in flies. By contrast, only three of these 17 *Daphnia*sequences match genes that are known to function during spermatogenesis in *Drosophila*. This discrepancy between the numbers of sex specific transcripts for genes involved in reproduction is likely caused by the small representation of males within the *Daphnia*cultures used to create the cDNA libraries. Equally impressive are the large number of cuticle proteins identified. Expanded protein families ------------------------- In the comparative study of the *Anopheles*and *Drosophila*proteomes \[[@B51]\], cuticular proteins were noted to be particularly active in their lineage specific expansions and deletions. This present study identifies 53 sequences that are either structural components of the cuticle or involved in chitin metabolism. Their abundance within our dataset is a consequence of their transcriptional responses to the microarray experiments; all but four of the assembled sequences were differentially regulated on the arrays. Yet as noted in the comparisons between the two dipteran insects, many *Daphnia*sequences share similarities to single loci within insect genomes. Among these cuticle loci are 15 assembled sequences that are best aligned to a single *Drosophila*gene when compared to the rest of the fly proteome. Alternative transcripts account for only 4 sequences. Thus, no less than 11 loci remain as possible representatives of a large lineage specific gene expansion of *Daphnia*cuticle genes. Further investigations are obviously required to verify this finding, including more thorough sampling of the *Daphnia*transcriptome and functional data such as *in situ*hybridizations to support the notion that these genes may have contributed to biological innovations. Regarding the 15 transcripts on the current array, all but one is enriched in males compared to females, and with the addition of differential expression patterns under metal stress, these transcripts can be grouped into five separate expression profiles. A first compelling case for lineage specific gene expansion is made by investigating the diversity of ferritin genes within the *Daphnia*ESTs. Except for *Aedes aegypti*, their number of ferritin genes are evolutionarily conserved \[[@B52]\]. Ferritins are the principle iron storage proteins for nearly all animals, and their abundance within cells is controlled in part by iron-regulatory proteins that interact with iron-regulatory elements (IREs) within alternatively spliced 5\' UTRs of certain mRNAs \[[@B53],[@B54]\]. In insects, ferritins consist of a heavy-chain homolog (HCH) and a light-chain homolog (LCH) forming heterodimers that function in the secretory pathways of cells, and which also appear to act as iron transporters \[[@B52]\]. The genes encoding the subunits (Fer1HCH, Fer2LCH) are positioned in the *Drosophila*genome in a back-to-back orientation, enabling coordinated regulation of their transcription \[[@B55]\]. This feature is conserved in all insects studied thus far \[[@B52]\]. Except in *Bombyx*, which has IREs within the UTRs of both subunits, insect IREs are predominantly localized to the Fer1HCH locus. Recently, a third *Drosophila*ferritin (Fer3HCH) has been described that controls iron homeostasis of the mitochondria, yet its transcription is not responsive to iron treatment \[[@B56]\]. As in humans and mice, the gene is predominantly expressed in adult testis. Our phylogeny of *Daphnia*ferritin gene transcripts uncovers six or seven distinct *Daphnia*loci (Figure [3](#F3){ref-type="fig"}). The branch F locus cannot be unequivocally included as part of the ferritin expansion until a description of the gene is available based on its alignment to the genome sequence. However, all of the other loci are defined based on their sequence alignments to distinct genome scaffolds assembled at this point in the *Daphnia*genome sequencing project (not shown). Branch G of the ferritin phylogeny represents the first characterized crustacean orthologue to the insect Fer1HCH genes. Following naming conventions, we designate this gene as Dpu_Fer1HCH. The gene has four introns and is the only *Daphnia*ferritin on the array showing differential expression for all three experimental conditions; its transcripts are enriched in males, depleted when exposed to cadmium and enriched when challenged by arsenic (Eads et al. submitted; Shaw et al submitted and in prep). Therefore like the insect subunit, this locus responds to metal ion treatments. The other five *Daphnia*ferritin genes are homologous to the insect Fer3HCH loci and are arranged within a monophyletic cluster, suggesting that they originate from a series of gene duplications. Indeed, each has retained two introns, despite showing amino acid sequence divergences from 14 to 56%. These genes are designated Dpu_Fer3HCH-1 to Dpu_Fer3HCH-5. Regrettably, expression data is not available for genes representing branches D and E. However, the microarray results show that the remaining three loci differ in their transcriptional responses to the experimental treatments. Like the *Drosophila*Fer3HCH gene, elements on the array whose sequences cluster within branch C do not respond to metals, yet unlike the fly gene, Contig 26 transcripts are enriched in females. By contrast, a single element on the array representing branch B is enriched in males, and all 4 cDNA elements from this gene respond to arsenic treatment. These differences are likely the result of undetected splice variants. Lastly, all elements representing branch A show elevated expression patterns when treated with arsenic, yet no sex specific expression is detected. These additional observations strongly support the existence of a Fer3HCH gene expansion that diversified within a crustacean lineage leading to *D. pulex*. A homologue to the insect Fer2LCH genes has yet to be discovered in Crustacea. Orphan genes ------------ The present study is a preliminary annotation of the emerging *D. pulex*genome using comparative and functional data to predict the fraction of genes unique to *Daphnia*. Aside from gene expansions that are suggestive of adaptations specific to aquatic environments, accounts of orphan genes (defined here has having no matches to insects) can offer equally important insights into crustacean biology from the perspective of a class of genes that usually has the shortest average lengths and are most rapidly evolving \[[@B51],[@B57]\]. Our searches using Blastx of 787 assembled sequences against the proteome of four insects suggest that \~68% of the *Daphnia*genes are shared with at least one insect. Less than half of these show matches to the individual protein databases of *Bombyx*, *Apis*, *Anopheles*and *Drosophila*. This discrepancy is likely a combined effect of incomplete datasets and of lineage specific gene losses among the insects. Taking into account associated matches to proteins from our chosen outgroup (*Caenorhabditis*), we discover that 21% of the *Daphnia*genes are either derived within Pancrustacea or lost within the nematodes. Those genes that are uniformly conserved across all four insect species are primarily cuticle proteins and serine proteases having trypsin activity. These gene families are also listed as two of the top 20 most significant expansions or reductions between the *Anopheles*and *Drosophila*proteomes \[[@B51]\], which diverged some 250 million years ago. It is tempting to speculate that, in both crustaceans and insects, a fraction of gene families are equally active in their evolutionary diversification. Such gene families would be candidates for detailed investigations leading to a better understanding of how Pancrustacea succeeded in exploiting its range of ecological settings. A careful evaluation of assembled sequences showing no matches to the insect proteomes suggests that \~1/3 of the genes are either derived in crustaceans or lost within insects. This estimation is admittedly from a very limited sampling of the total number of *Daphnia*genes and is derived from sequencing non-normalized cDNA libraries that were created under standard laboratory conditions and interrogated by microarrays. Although this fraction cannot be extrapolated to the genome, it is comparable to findings from other taxa. In *D. melanogaster*, 10% of the genes have homologous best hits in non-insect species plus 19% have no homologous hits to other species, while the combined estimate for *A. gambiae*is 21% \[[@B51]\]. Within the nematodes, which diverged ca. 600 mya, 23% of the genes are estimated to be unique to species \[[@B14]\]. Of course, the fraction of species specific genes declines dramatically when evaluating close allies; comparing two *Caenorhabditis*species reveals that 4% of their genes are unique \[[@B58]\], and the mouse gene set differs from the human set by only 1% \[[@B59]\]. A future investigation of a larger *D. pulex*gene collection against an equivalent dataset for the congener *D. magna*\[[@B60]\] will help define the true estimate of species specific genes in *Daphnia*. Additional contributions of EST data for non-branchiopod crustaceans will further define the crustacean proteome and shed light on the biological factors that led to the group\'s divergence from insects. However, the sequences presented by our present study are accompanied by expression data from three microarray experiments, which authenticate the orphan sequences as genes and support the notion that ecological factors are more likely to contribute to sequence and functional divergences among genomes. Combining the gene expression data obtained by Eads et al. (submitted) and Shaw et al. (submitted, and in prep) for the 787 assembled sequences reveals that the majority of genes with sex biased expression, including developmental and regulatory loci, do not respond to the cadmium and arsenic metal toxicity. A clear example is provided by genes predicted to regulate translation. All 15 genes save two are differentially expressed in males versus females and only three genes also show transcriptional responses to metal toxicity (Additional file [2](#S2){ref-type="supplementary-material"}). We find that this class of sex biased genes proportionally contains the fewest orphans. The relatively larger number of sequences with matches to the insect proteome suggests that genes functioning during development and reproduction are generally well conserved between crustaceans and insects. Further work is required to elucidate crustacean and *Daphnia*specific components of these central processes. By contrast, nearly half of the *Daphnia*genes that respond to metals, but show no differences between the sexes, are likely absent in insects. This is explicable in light of the fact that metal exposure is an ecological stressor that varies between aquatic and terrestrial environments \[[@B61]\], which has catalyzed the evolution of certain protein types (cuticles, iron metabolism, defense) to increasingly specialized functions. The extent to which ecology has shaped the genome organization of pancrustaceans is an important future direction for research. For example, the mosquito *A. gambiae*spends part of its larval stage in water; by comparing genes differentially expressed during this stage to expression patterns in *D. pulex*, it may be possible to examine the effects of an aquatic lifestyle on the expression of particular protein families. Conclusion ========== This work investigates the sequence preservation and expansion of genes from the crustacean *D. pulex*compared to insect proteomes, based on the analysis of 1,546 ESTs that represent 787 unique transcripts. Our sampling of cDNA from this emerging genomic model species reveals sequences that have largely been conserved in both groups representing arthropods evolving in water or on land. Genes that function for reproduction, regulation of cellular processes and development are identified; some are known to genetically interact in the model insect species *Drosophila*. This provisional annotation of *Daphnia*sequences is further verified by companion studies using cDNA microarrays to examine transcription in males, females and embryos (Eads et al. submitted) and under toxic metal stress (Shaw et al. submitted, and in prep). Here we identify cases of lineage specific gene family expansions by a series of gene duplications. For instance, there are as many as seven distinct ferritin loci indicated by cDNA and genome data, including a crustacean orthologue to the insect Ferritin 1 locus and a monophyletic grouping of five Ferritin 3 genes. Finally, our results suggest that, as we study the genomes of organisms distantly related to the classic model laboratory organisms, the majority of unknown genes will be functionally linked to the organisms\' ecology. Compared to the gene sets showing differential expression among developmental stages, we observe that sets responding to ecological stress contain a greater proportion of loci with no sequencing similarity to previously characterized arthropod genes. A comprehensive inventory of putative orthologs, orphan genes, and lineage specific gene expansions coupled with functional genomics data will provide important insights into genomic changes that led to the adaptive radiation of crustaceans. Methods ======= cDNA library construction and quality assurance ----------------------------------------------- For the purpose of creating a collection of cDNA for printing onto microarrays, a clonal isolate of a *D. pulex*/*D. pulicaria*hybrid (called log52) was cultured under standard laboratory conditions by Jim Haney (University of New Hampshire) within a large, aerated, 200 liter container of filtered lake water by feeding a concentrated monoculture of green algae (*Scenedesmus acutus*). Animals at all life stages were harvested and immediately processed. Total RNA was isolated using Trizol reagent (Invitrogen Life Sciences) and was subsequently purified using the RNeasy protocol (Qiagen). The cDNA libraries were constructed by Darren Bauer and Kelley Thomas (University of New Hampshire) using the Creator SMART (Clontech) system by following the manufacture\'s instructions. The cDNA was ligated into the pDNR-LIB vector supplied by Clontech. To control for bias towards smaller fragments inserting during the ligation of cDNA into plasmids, reaction were performed on four cDNA size fractions. Size fractionation was performed as per the SMART cDNA protocol using the CHROMA SPIN-400 column. The column was prepared for drip procedure by inverting several times to completely resuspend the gel matrix and storage buffer was drained by gravity flow. Seven hundred microliters of column buffer were added to the column and allowed to drain out, then 100 μl of a mixture of *Sfi*I-digested cDNA and xylene cyanol dye were applied to the matrix and allowed to fully absorb. One hundred microliters of column buffer were added to the matrix and allowed to fully absorb, then 600 μl of column buffer were added and single-drop fractions were collected in 16 tubes. The profile of the fractions was verified by running 3 μl of each fraction on a 1.1% agarose/EtBr gel at 150v for 10 minutes. The samples were then pooled into two size classes; fractions 7 and 8 were pooled into the \"large size\" and fractions 9 & 10 were pooled into the \"small size\". From the libraries, 768 colonies were chosen for quality assurance tests. The bacterial transformants were amplified in selective 2xYT media, plasmids were purified by an alkaline lysis protocol according to the manufacturer\'s instructions (PerfectPrep, Eppendorf) and quantified by spectrophotometry. The molecular weights of cDNA inserts were measured by PCR amplification of cDNA inserts using the M13 vector primers M13fw (GTG TAA AAC GAC GGC CAG TAG) and M13rev (AAA CAG CTA TGA CCA TGT TCA C) followed by agarose gel electrophoresis against standards and visualized using a Kodak 440cf imaging station. Sequencing reactions were performed by priming at the 5\' end of cDNA using vector primer pDNRlib30-50 (TAT ACG AAG TTA TCA GTC GAC G), ABI BigDye chemistry and the 3730 sequencer. Vector and poor quality sequences were trimmed from the sequencing reads and ESTs were assembled into contigs using the SeqManII software (DNASTAR package). Homologies with Genbank entries were discovered using Blastx against the non-redundant (nr) protein database. Those sequences with expectation-values better than 1 × 10^-27^were further examined for the presence of an annotated ATG start codon at the 5\' end of the open reading frame (ORF). This last step was accomplished using NCBI\'s ORF finder tool \[[@B62]\]. Only those sequences whose Methionine aligned (including gaps) with the first amino acid of complete sequences were considered full-length transcripts. Characterization of the ESTs ---------------------------- On thousand twenty-eight additional cDNA samples were chosen for sequencing based on the microarray results obtained by Eads et al. (submitted) and Shaw et al. (submitted, and in prep). The sequencing reactions were carried out as outlined above. All 1,648 sequence reads from this study, with their quality scores, were obtained from ABI sequencer data files using phred \[[@B63]\] with default parameter values. The reads were then processed by discarding low quality and vector sequences using Lucy v1.19p \[[@B64]\] with default parameter values, by removing poly-A tails using EMBOSS trimest \[[@B65]\] and by discarding sequences with lengths under 100 bases. The remaining high quality EST set was reduced to a non-redundant set of unique gene transcripts by clustering with phrap \[[@B66]\] using the following parameters: mismatch penalty = -5; minimum match = 50; minimum score = 100. The resulting contigs and singlets that matched to mitochondrial gene transcripts (Genbank accession [NC 000844](NC 000844)) using Blastn were removed from subsequent analyses. To investigate whether the set of assembled sequences contain alternative transcripts of the same loci, the contigs and singlets were further clustered using the SeqManII software with the following relaxed parameters: match size = 12; maximum added gap length = 70; minimum percent match = 80; no gap penalty; gap length penalty = 0.70. The putative open reading frames (ORFs) for the assembled sequences were determined in three steps using Prot4EST v2.2 \[[@B67]\] with DECODER having been disabled, ESTScan -- which is an integral component of Prot4EST -- and getorf from EMBOSS. The ORFs were selected during the first step when the assembled sequence translations aligned to proteins within the NCBI NR database with a Blastx e-value better than 1 × 10^-8^. Failing this first step, the ORFs where selected during the second step using ESTScan or simply by recording the longest uninterrupted ORFs when they were located on the positive stands of the sequences. Otherwise, the longest ORFs were selected during step three, based on the results obtained by using the EMBOSS program that restricted sequence translations from the negative strand. This restriction was justified by observing only 3 ORFs on the negative strand from among 376 predictions from step one. Numerous sequence similarity searches were done for both the high quality EST set and the assembled sequences. First, queries were performed against the NCBI NR protein database (Genbank release 148) using a local installation of the WU-BLAST program \[[@B68]\]. The taxonomic domains were added to the results by parsing the taxa ID from the top match for each query and by retrieving the associated information from the NCBI \[[@B69]\]. Second, for a more confident assessment of whether the assembled sequences were shared with insects, they were compared to protein sequences archived in the NCBI UniGene sets for *Bombyx mori*(build \#7), *Apis mellifera*(build \#5), *Anopheles gambiae*(build \#29) and *Drosophila melanogaster*(build \#37) using tBlastx with an expectation threshold set at E \< 0.005. The same search was performed against the *Caenorhabditis elegans*(build \#23) UniGene database to judge whether the differences can be attributed to gains or losses within the representative insects or crustacean. The assembled sequences were clustered based on the distribution of bit scores across the databases using self organizing maps followed by k-means clustering within 28 nodes in Cluster v2.11 \[[@B70]\]. Third, to further ascertain whether the assembled sequences can be aligned to known proteins, queries were made against all *Drosophila melanogaster*gene translations that are predicted by the annotation v4.2.1 of the genome sequence assembly and against the genome nucleotide sequences of *Drosophila melanogaster*and *Caenorhabditis elegans*using tBlastx. The assembled sequences were classified into gene ontology (GO)-defined functional classes using the program Blast2GO \[[@B22]\] and by extracting the GO annotations from FlyBase for sequences that strongly matched *D. melanogaster*gene transcripts. The putative gene annotations were examined for functional classes that are enriched within our lists of *Daphnia*genes compared to the total set of GO terms for all *Drosophila*genes using Gostat \[[@B71]\] and by testing for the enrichment of GO terms within subsets of the assembled sequences using Fisher\'s Exact Test executed within Blast2GO. Authors\' contributions ======================= JKC, BDE and JA conceived the study, designed and implemented the comparative analyses, and drafted the manuscript. JS and BDE performed the microarray experiments that guided the sequencing efforts, contributed the expression data and interpreted the results in light of this study. DB created the cDNA libraries. EB and JKC characterized the cDNA libraries and contributed EST sequences with BDE. All authors read and improved the final manuscript. Note added to proof =================== The recently released Draft *Daphnia pulex* genome sequence (July 7, 2007) suggests that Daphnia possess a single copy of the Ferritin 1 gene, represented by Singlet 73 on the phylogenetic tree (Figure [3](#F3){ref-type="fig"}). Supplementary Material ====================== ###### Additional file 1 Characterization of the *Daphnia pulex*EST sequences. This file contains the following information about the analysis of EST sequences obtained for the study. \(1\) EST information. \(2\) Top match from Blastx searches of clustered *Daphnia*ESTs against the NCBI non-redundant (nr) protein database. \(3\) Results using Blast2GO (<http://www.blast2go.de/>). \(4\) Top match from Blastx searches of clustered *Daphnia*ESTs against all *Drosophila melanogaster*predicted gene translations from annotation 4.2.1 -- 16 columns describe the results. • Cluster id (*Daphnia*, this study). • Subject id (*D. melanogaster*dmel-all-translation-r4.2.1 data). \(5\) Differential expression results in 3 microarray experiments. \(6\) Top match from Blastx searches of *Daphnia*ESTs against NCBI non-redundant (nr) protein database -- 12 columns describe the results. • Subject id of the best match in the nr database. ###### Click here for file ###### Additional file 2 Supplemental Table 1. *Daphnia*genes annotated as regulators of transcription and translation based on sequence conservation with *Drosophila*genes with known functions. Scores are reported from results obtained by Blastx against all predicted translations from version 4.2.1 of the *D. melanogaster*genome annotation. First and second columns under DE show genes that are differentially expressed (+ = yes) in microarray experiments comparing male versus female transcripts and metals versus no metals exposure, respectively. TF = transcription factor; TR = transcriptional regulation; TE = transcript elongation; E = translation elongation; R = translation regulation. ###### Click here for file ###### Additional file 3 Supplemental Table 2. *Daphnia*genes annotated as signaling proteins and other regulators based on sequence conservation with *Drosophila*genes with known functions. Scores are reported from results obtained by Blastx against all predicted translations from version 4.2.1 of the *D. melanogaster*genome annotation. First and second columns under DE show genes that are differentially expressed (+ = yes) in microarray experiments comparing male versus female transcripts and metals versus no metals exposure, respectively. ###### Click here for file ###### Additional file 4 Supplemental Table 4. *Daphnia*genes annotated as candidates for gametogenesis based on sequence conservation with *Drosophila*genes with known functions. Processes include: SP = spermatogenesis; OO = oogenesis; FCD = follicle cell development; GCD = germ cell development; GT = gametogenesis. Two assembled sequences matched CG4027 and two other sequences matched CG2168. ###### Click here for file ###### Additional file 5 Supplemental Table 4. *Daphnia*genes annotated as genes associated with exoskeletal function and molting. These include: CP = structural cuticle proteins; PM = peritrophic membrane; CM = cuticle metabolism; CA = chitinase; M = molting; CB = cuticle binding. Assignments are to proteins based on sequence conservation with *Drosophila*genes with known functions. ###### Click here for file Acknowledgements ================ This project was financed by collaborative grants from the National Science Foundation (DEB 0221837 and FIBR 0328516) and by seed funds from The Center for Genomics and Bioinformatics, supported in part by the Indiana Genomics Initiative (INGEN) under the Lilly Endowment and by Shared University Research grants from IBM, Inc. to Indiana University. Computer support was provided by Phillip Steinbachs and Sumit Middha at The Center for Genomics and Bioinformatics and by Dick Repasky at the Indiana University Information Technology Services. We thank Kelley Thomas and Jim Haney (University of New Hamphire) for their contributions towards the construction of cDNA libraries. Don Gilbert (Indiana University) and Sumit Middha provided important bioinformatics support. This work benefits from, and contributes to the *Daphnia*Genomic Consortium.

Introduction ============ Groupers (family Serranidae) belong to 109 species and 11 genus ^[@ref-1],\ [@ref-2]^. Groupers are commercial marine fishes that have been harvested intensively from the wild, resulting in decreasing the population worldwide ^[@ref-3],\ [@ref-4]^. The white spot grouper, *Epinephelus coeruleopunctatus,* is one of the most popular groupers and has a high economic value among groupers in Asia-Pacific regions ^[@ref-5],\ [@ref-6]^. However, this species is rare and difficult to catch. According to local fishermen of Padang City, Indonesia, the population of *E. coeruleopunctatus* has been declining sharply over the last two decades ^[@ref-7]^. According to Teixeira *et al*. ^[@ref-8]^ and Mariskha and Abdulgani ^[@ref-9]^ the decreasing fish population is caused by overfishing, habitat perturbation ^[@ref-10]^ and unfriendly fishing practices ^[@ref-11]^. The International Union for Conservation of Nature ^[@ref-12]^ reports this species on the Red List as a threatened species. Culturing of white spot grouper has been initiated in Indonesia; however, the fry (juveniles) are strongly dependent from the wild supply ^[@ref-13]^. Therefore, it is very crucial to develop breeding technology of the white spot grouper. One of the problems in the development of breeding technology is hermaphroditism sex development, which is observed in this species ^[@ref-14]^. Therefore, it is difficult to determine the sex differentiation between male and female. Hermaphroditism has also been reported in several other groupers, such as *E. tauvina* ^[@ref-15]^, *E. aeneus* ^[@ref-16]^, *E. rivulatus* ^[@ref-17]^, *E. striatus* ^[@ref-18]^, and *Plectropomus laevis* ^[@ref-19]^. Hence, this paper reports on the size (length and body weight) of sex transformation in white spot grouper. This information is crucial to plan a better management strategy of fishery resources ^[@ref-20]^ and to develop breeding technology for the white spot grouper. Methods ======= All procedures involving animals were conducted in compliance with Bung Hatta University Research and Ethics Guidelines, Section on Animal Care and Use in Research. Fish were caught from Padang City waters, at GPS coordinates 0 ^0^ 54' 55.34" S, 100 ^0^ 10' 15.49" E ( [Figure 1](#f1){ref-type="fig"}), between April and October 2013. The fish were caught using hooks and hand line at the depth of 30--50 m. Fishing operations were carried out from 6.00 am to 16.00 pm. The sampled fish were anesthetized with MS222, prepared by dissolving 4g of MS222 in 5L tap water ^[@ref-21]^ and then transported to the Laboratory of Fisheries Resources of Bung Hatta University for further analysis. In the laboratory, the fish samples were measured for total length (mm) and body weight (g). The abdomen was dissected and the gonad was removed carefully and cleaned using a tissue paper and then weighed nearest to mg using a digital balance (ACIS: AD300; errors 0.01g). Sex differentiation by gonad was examined microscopically (100x magnification) and determined based on Muchlisin *et al*. ^[@ref-22]^. The data were analyzed descriptively. ![A map of Padang City waters showing the sampling location (red circle) ^[@ref-7]^.](f1000research-6-11962-g0000){#f1} Results ======= A total of 56 fish were recorded during the study, where 50 fish were recognized regarding sex differentiation by gonad, of which 22 were males and 28 females. A total of 6 samples were not recognized regarding their sex, due to being still in the early gonadal development stage. The sex ratio was 2:3 (male:female). The total length of the male fish ranged from 302--537 mm, while females ranged from 183--537 mm. The body weight ranged between 374--2107 g and 85--373 g for male and female fish, respectively. The total length of fish with undetermined sex ranged from 125--242 mm and 85--373 g body weight ( [Table 1](#T1){ref-type="table"} and [Table 2](#T2){ref-type="table"}). The study showed that the first sex differentiation of *E. coeruleopunctatus* occurred at a size above 183 mm; fish of this size were recognized as female and no male fish were detected in this size group. First sex differentiation is species dependent; for example, *E. bleekeri* occurrs at 170 mm ^[@ref-23]^ and *Plectropomus laevis* at 280 mm ^[@ref-19]^. ###### Length frequency distribution of fish *Epinephelus coeruleopunctatus.* --------------------------------------------- Length\ N (%) class (mm) ------------ ---------- ---------- ---------- 125--182 \- \- 3 (50.0) 183--242 \- 4 (14.3) 3 (50.0) 243--301 \- 4 (14.3) \- 302--360 2 (9.1) 5 (17.9) \- 361--419 6 (27.3) 7 (25.0) \- 420--478 5 (22.7) 5 (17.9) \- 479--537 9 (40.9) 3 (10.7) \- Total 22 (100) 28 (100) 6 (100) --------------------------------------------- ###### Weight frequency distribution of fish *Epinephelus coeruleopunctatus*. -------------------------------------------- Weight\ N (%) class (g) ------------ ---------- ---------- --------- 85--372 \- 7 (25.0) 6 (100) 374--662 1 (4.5) 7 (25.0) \- 663--951 6 (27.3) 5 (17.9) \- 952--1240 2 (9.1) 2 (7.1) \- 1241--1529 4 (18.2) 2 (7.1) \- 1530--1818 3 (13.6) 5 (17.9) \- 1819--2107 6 (27.3) \- \- Total 22 (100) 28 (100) 6 (100) -------------------------------------------- The results revealed that the female white spot grouper begun to transform to male at 302 mm in length, indicating a protogynous hermaphroditism. However, the size at which all fish transform to male fish was unknown, since there were no fish sample more than 537 mm in length. But, the existing data show that the ratio of male fish was increased as total length increased; hence, we suspect that all fish have changed sex to male at sizes above 600 mm. For comparison, Renones *et al*. ^[@ref-24]^ reported that the female dusky grouper *E. marginatus* transforms its sex initially from female to a male at a size of 680 mm and all males were detected at size 800 mm. In addition, Tan and Tan ^[@ref-25]^ reported that *E. tauvina* begins to transform their sex from female to male at the size of 650 mm, while at the size of 700 mm all fish are recognized as male. According to Burhanuddin and Fami ^[@ref-26]^ the occurrence of sex transformation in hermaphroditic fish is species dependent and strongly influenced by environmental factors. ###### Click here for additional data file. Copyright: © 2017 Bulanin U et al. 2017 Data associated with the article are available under the terms of the Creative Commons Zero \"No rights reserved\" data waiver (CC0 1.0 Public domain dedication). Conclusions =========== The white spot grouper *Epinephelus coeruleopunctatus* is a protogynous hermaphroditism. Sex differentiation was detected at the total length of 183 mm and at this size the fish are female. The sex transformation began to occur at 302 mm total length. Data availability ================= The data referenced by this article are under copyright with the following copyright statement: Copyright: © 2017 Bulanin U et al. Data associated with the article are available under the terms of the Creative Commons Zero \"No rights reserved\" data waiver (CC0 1.0 Public domain dedication). <http://creativecommons.org/publicdomain/zero/1.0/> Dataset 1: The total length, body weight and sexes of the 56 individual fish sampled. doi, [10.5256/f1000research.11090.d155119](http://dx.doi.org/10.5256/f1000research.11090.d155119) ^[@ref-27]^ The authors thank the Ministry of Research, Technology and Higher Education for providing the financial support to this study. Appreciation goes to our colleagues who helped the authors during field sampling and laboratory analysis. 10.5256/f1000research.11962.r21903 Referee response for version 1 Nugroho Rudy Agung 1 Referee Department of Biology, Faculty of Mathematics and Natural Sciences, Mulawarman University, Samarinda, Indonesia **Competing interests:**No competing interests were disclosed. 18 4 2017 Version 1 Review report for article entitle "Hermaphroditism in the white spot grouper *Epinephelus coeruleopunctatus*(Pisces: Serranidae) harvested from Padang City waters, Indonesia". Overall, this article has displayed originality in the work and the outcome of this work adds benefit to the area of the research. This article is presented well with cohesiveness. However, editorial suggestions should be addressed by the author. Below, the basis for suggestion followed by some special editorial concern by section.   Title:The title is short, informative and well constructed.Abstract:Suggestion:  Information regarding on how sex differentiation was performed should be written.Introduction:The introduction has provided quality relevant information particularly with regard to hermaphroditism. Suggestion: The introduction required to include relevant information specific to sex differentiation by using gonad identification followed by relevant references.Methods:a) Please provide specific instrument that used to measure the total length and body weight.b) Gonad measurement was performed but there is no gonad weight data in the result section.c) Microscopic observations of the gonad is not clearly describe.d) \"The data were analyzed descriptively\" the statement is not clear which data referring to in this article. Please be specificResults:a) Length and weight frequency distribution class in the table 1 and 2 is not clearly defined.b) Set data or the number of the sample is too small.c) It is stated that the occurrence of sex transformation in hermaphroditic fish is species dependent and strongly influenced by environmental factors. The author should include the environment report in the study area. This environment report will give another perspective on this study.Conclusion:To conclude the finding that the female differentiated at 183mm and sex changed to male occurred at 302 mm is too "early" because is based on the length and weight. It is better to include histological, endocrinological or event molecular study in this article.References:Suggestion: delete ref \#27 I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard. 10.5256/f1000research.11962.r21782 Referee response for version 1 Pusey Bradley J. 1 Referee Centre of Excellence in Natural Resource Management, The University of Western Australia, Perth, WA, Australia **Competing interests:**No competing interests were disclosed. 11 4 2017 Version 1 Title and abstract both appropriate. Article content -- design, methods and analysis all appropriate Conclusions are sensible and balanced.  The authors might consider expressing maturation in terms of the length at which 50% of the sample are one sex or another, in addition as say minimum length.  Perhaps more needs to be said of the fact that not all fish in the very largest size class had changed sex into males. I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard. 10.5256/f1000research.11962.r21264 Referee response for version 1 Abol-Munafi Ambok Bolong 1 Referee Institute of Tropical Aquaculture, Universiti Malaysia Terengganu, 21030 Kuala Nerus, Terengganu, Malaysia **Competing interests:**No competing interests were disclosed. 29 3 2017 Version 1 1\. Title     The title is suitable. Only the word \'harvested\' should be deleted. 2\. Abstract    Suggestion:  A total of 56 white spot groupers were sampled; of these 22 were male, 28 were female and 6 were undifferentiated. 3\. Introduction     Accepted but the English Language needs to be revised. 4\. Methods     i)   Please provide the instrument used to measure the total length and body weight.     ii)  Method for gonad measurement can be removed since the data is not tabled in this article.     iii) Must have explanation on gonad structure/conditions of functional male and functional female based on microscopic observations that were used for sex determination. I cannot find the explanation in Muchlisin *et al*. (2010) ^[@rep-ref-21264-1]^. Please elaborate on \"The data were analyzed descriptively\" or the statement is not relevant in this article.     iv) Please explain on what basis the length class and weight class were decided. Why the differences are not equal. Please recalculate. 5\. Results   i) The first sentence need to be rephrased.  ii) Regarding my comment on what basis the length class was decided, the article suggested that the female differentiated at 183mm. If you refer to your Dataset 1, the smallest female is 22.3cm (or 223 mm). Based on the length class, the article suggested that the sex change from female to male occurred at 302 mm. Your Dataset 1 showed that the smallest male caught was 35.0cm (or 350 mm). Please elaborate your data. 6\. Conclusions \- The sample size is too small and the duration of the study is too short to conclude the finding that the female differentiated at 183mm and sex changed to male occurred at 302 mm. 7\. References \- No 27 is considered as one of the references? General remark: \- The English language must be revised. Standard terminology should be used, eg: sex change instead of sex transform; differentiated instead of recognized. I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above. [^1]: UB was responsible for developing research proposal and study design and approved the final draft of the paper. MM was responsible for sample collection and processing, and data analysis. ZAM is responsible for manuscript preparation and proofreading of the draft. [^2]: **Competing interests:**No competing interests were disclosed.