|Home | About | Journals | Submit | Contact Us | Français|
Although both genetic factors and features of the social environment are important predictors of posttraumatic stress disorder (PTSD), there are few data examining gene-social environment interactions in studies of PTSD. The authors examined whether features of the social environment (county-level crime rate and unemployment) modified the association between the serotonin protein gene (SLC6A4) promoter variant (5-HTTLPR) and risk of current PTSD in a sample of 590 participants from the 2004 Florida Hurricane Study. Interviews conducted in 2005 were used to obtain individual-level risk factor measures and Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, PTSD diagnoses. DNA was extracted from salivary samples. County-level crime and unemployment rates were assessed from Federal Bureau of Investigation and US Census data, respectively. There was a significant interaction between 5-HTTLPR genotype and both crime rate (odds ratio=2.68, 95% confidence interval: 1.09, 6.57) and unemployment rate (odds ratio=3.67, 95% confidence interval: 1.42, 9.50) in logistic regression models predicting PTSD risk, after adjustment for individual-level determinants of PTSD. Stratified analyses indicated that the “s” allele of the 5-HTTLPR polymorphism was associated with decreased risk of PTSD in low-risk environments (low crime/unemployment rates) but increased risk of PTSD in high-risk environments. These results suggest that social environment modifies the effect of 5-HTTLPR genotype on PTSD risk.
Posttraumatic stress disorder (PTSD) occurs following exposure to a potentially traumatic event (1). The majority of Americans are exposed to a potentially traumatic event during their lifetimes, although only a minority develop PTSD (2). Still, the disorder is common: At least 1 in 14 Americans meets criteria for the diagnosis of lifetime PTSD (2). Despite substantial research, our understanding of the factors that determine vulnerability to PTSD remains incomplete. Two recent meta-analyses found that extant risk factors explained only about 20% of the interpersonal variance in PTSD (3, 4). Clearly, new variables including genetic factors and features of the social environment need to be incorporated into models aimed at understanding PTSD risk.
Genetic influences account for approximately one-third of the variance in PTSD risk among persons exposed to trauma (5, 6). Two studies have shown an association between a common variable number of tandem repeats polymorphism in the promoter region of the serotonin transporter gene (SLC6A4), designated 5-HTTLPR, and PTSD. In a Korean sample, the short (“s”) 5-HTTLPR allele, which is less transcriptionally efficient than the long (“l”) allele (7), was more prevalent among PTSD patients than among controls (8). In a sample of persons exposed to the 2004 Florida hurricanes (also used in the current study), the low expression variant of the 5-HTTLPR polymorphism modified risk of posthurricane PTSD, but only under conditions of high hurricane exposure and low social support, a gene-environment interaction (9).
Features of the social environment influence both risk of exposure to potentially traumatic events and risk of PTSD among the exposed (10–12). For example, persons living in areas characterized by greater social disadvantage may be exposed to greater violence (10, 13–16) and may also be more susceptible to PTSD because of the fragmented and disorganized social support systems in such areas (3, 4, 12).
Genetic factors and the social environment may jointly shape risk of PTSD. Although investigators in 2 studies have reported gene-environment interaction in PTSD (9, 17), no studies, to our knowledge, have assessed how features of the social environment, beyond individual-level trauma exposure and social support, influence genetic determination of PTSD. Thus, we examined whether the social environment (county-level crime rate and unemployment rate) modified the association between the 5-HTTLPR polymorphism and risk of current PTSD in a sample of hurricane-exposed adults.
Data for this study were collected in a random digit dialing study of residents of 33 Florida counties in 2004, when Florida was hit by hurricanes Jeanne, Ivan, Frances, and Charley in rapid succession. We used the Waksberg (18) random digit dialing method to select households with telephones to be screened for potential participation. Telephone numbers were called 5 times at different times of day and were replaced if there was no answer after 5 calls.
Inclusion criteria were: 1) speaking English or Spanish; 2) having lived in Florida during at least 1 of the 2004 hurricanes; 3) currently living in a household in 1 of 33 hurricane-exposed counties; and 4) having sufficient physical and mental ability to participate in a telephone interview. Because the original study investigated predictors of postdisaster resilience among older adults, persons aged 60 years or more were oversampled. The response rate was 81%.
Verbal consent was obtained from all participants. Those who completed the diagnostic interview and returned saliva samples received $20. The institutional review boards at the relevant institutions approved all procedures. Interviews were conducted between April 5 and June 12, 2005. More details about the sampling procedure and methods of the 2004 Florida Hurricane Study are provided elsewhere (9, 19–21).
A national survey research firm with considerable experience in conducting diagnostic interviews by telephone performed the highly structured assessment interviews using the computer-assisted telephone interview format (22). Respondents were randomly selected using the most-recent-birthday method when multiple eligible adults were present within a household. Interviews averaged 26.5 minutes in length. A detailed description of the assessment procedure has been published elsewhere (20).
Current PTSD (past 6 months) was assessed using the National Women's Study PTSD module, a widely used measure in population-based epidemiologic research (23). The National Women's Study PTSD module was validated in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (1), PTSD field trial against the Structured Clinical Interview for DSM Disorders; the interrater kappa coefficient was 0.85 for the diagnosis of PTSD, and comparisons with the Structured Clinical Interview for DSM Disorders yielded a kappa coefficient of 0.71 (24). A high correspondence between telephone and in-person administration of the National Women's Study PTSD module has also been documented (25). We operationalized the diagnosis of PTSD on the basis of Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, symptom requirements (i.e., 3 avoidance symptoms, 1 intrusion symptom, and 2 arousal symptoms), including functional impairment.
Social support 6 months before the hurricanes was assessed with a modified 5-item version of the Medical Outcomes Study module (26). Low social support was operationalized as a score less than or equal to 15 based on the cutoff score derived from prior work (22). This scale had good reliability (α=0.86).
Hurricane exposure was assessed with 5 indicators identified as being related to posthurricane mental health functioning in previous research (27): 1) exposure to hurricane-force winds or major flooding; 2) lack of adequate access to food, water, electricity, telephone, or clothing for at least 1 week; 3) losses or significant damage in 2 or more of the following categories: furniture; sentimental possessions; automobiles; pets; and crops, trees, or garden plants; 4) displacement from home for at least 1 week; and 5) unreimbursed losses of $1,000 or more. Participants with 2 or more of these indicators were coded as having high hurricane exposure.
Exposure to other potentially traumatic events was measured by asking participants whether they had been exposed to any of 5 potentially traumatic events and whether they had feared death or serious injury during exposure. The events queried about included: 1) experiencing a natural disaster; 2) having a serious accident at work; 3) being attacked with a weapon; 4) being attacked without a weapon; and 5) being in a war zone. The number of other potentially traumatic events was summed. Participants were categorized as experiencing 0, 1, or ≥2 events.
Data on crime were taken from the Federal Bureau of Investigation's Uniform Crime Reporting Program (Serious Crimes Known to Police, 1999 data (28)). The crime rate was defined as the number of crimes committed per 100,000 residents. High-crime counties were defined as those with a crime rate above the mean rate of the sampled counties.
Data on unemployment were taken from 2000 US Census Summary File 3 (29). High-unemployment counties were defined as those with an unemployment percentage above the mean percentage of the sampled counties.
Saliva samples were provided by 651 participants; valid ancestry data were available for 623 (95.7%), and valid 5-HTTLPR data were available for 590 cases (90.6%). The likelihood of submitting a saliva sample was unrelated to sex, hurricane exposure, social support, or PTSD. Details on the response rate for saliva sampling and correlates of participation are summarized elsewhere (19).
DNA was extracted from saliva using PUREGENE kits (Gentra Systems, Minneapolis, Minnesota). We examined the functional variable number of tandem repeats polymorphism in the 5′-flanking promoter region of the gene (SLC6A4) encoding the serotonin transporter protein. This polymorphism (5-HTTLPR) has 2 common alleles that have been characterized according to their length as “long” (“l”) (16 repeats) and “short” (“s”) (14 repeats). Genotyping was performed via polymerase chain reaction followed by size fractionation as described elsewhere (30) with prior MspI restriction endonuclease digestion for triallelic classification, which allowed classification of “l” alleles into “lA” and “lG” variants (lG has lower reuptake efficiency, similar to the “s” allele). The accuracy of s/l genotyping was confirmed via reanalysis of 100% of the specimens. The uncommon “lG” alleles were classified as “short” for analysis (both denoted below as “s′”) (31). In addition, 36 markers were genotyped to obtain information on ancestry (32–35). We added 1 additional highly informative single nucleotide polymorphism marker, SLC24A5 (36), to the panel described previously.
Ancestry proportion scores were generated to control for population stratification; population differences in allele frequency are known to occur for the SLC6A4 locus (37). Participants’ ancestries were estimated with the set of 37 unlinked ancestry-informative markers through Bayesian cluster analysis, using the STRUCTURE software developed by Pritchard and colleagues (38, 39). For the STRUCTURE analysis, we specified the “admixture” and “allele frequencies correlated” models and used 100,000 burn-in iterations and 100,000 Markov chain Monte Carlo iterations.
We used generalized estimating equations logistic regression models in all analyses to calculate parameter estimates with robust standard errors and to account for potential clustering by county (40, 41). First, we examined the association between PTSD and each demographic characteristic (sex, age <60 years, race/ethnicity) and risk factor (employment status, low social support, high hurricane exposure, other potentially traumatic events, 5-HTTLPR genotype, high crime rate, and high unemployment) individually, without adjusting for other covariates. The selected covariates were factors associated with PTSD in previous analyses of these data (20). Next, we fitted 4 logistic regression models. The first model regressed PTSD on high crime rate, with adjustment for all covariates. The second model included all of the above and added an interaction term for high crime rate × 5-HTTLPR genotype. The third and fourth models paralleled those for crime rate, except that high unemployment replaced crime rate as the county-level variable.
Table 1 presents the characteristics of the sample participants and their association with current PTSD. Among adults with genotype data for 5-HTTLPR (n=590), the prevalence of current PTSD was 3.2% (n=19). Low social support, high hurricane exposure, and exposure to other potentially traumatic events predicted significantly increased risk of current PTSD.
5-HTTLPR genotype frequencies were consistent with previous work (37). The frequency of the “l′/l′” genotype was higher among African-American adults (l′/l′: 34.8%; s′/l′: 47.8%; s′/s′: 17.4%) than among European-American adults (l′/l′: 25.1%; s′/l′: 54.9%; s′/s′: 20.0%), as shown elsewhere (37). However, this difference was not statistically significant (n=553; χ2 (2 df)=1.09; P=0.58). There was no difference in ancestral proportion score between persons with PTSD (mean=0.12 (standard deviation, 0.27)) and persons without PTSD (mean=0.06 (standard deviation, 0.20); t test: t(18.63)=−0.98; P= 0.34). Thus, even without correction based on ancestry coefficients, population stratification was unlikely to be a potential cause of false-positive findings.
Table 2 presents results from the multivariable logistic regression models for crime rate and PTSD. In the main-effects model, low social support, high hurricane exposure, and exposure to other traumatic events were associated with significantly increased risk of PTSD. There was no significant main effect for 5-HTTLPR genotype or county-level crime rate. In the interaction model, the interaction term for 5-HTTLPR genotype × crime rate was significant (P =0.03). Figure 1 presents the prevalence of PTSD for persons from high-crime-rate counties versus persons from low-crime-rate counties by 5-HTTLPR genotype. Stratified analyses showed that the s′ allele of 5-HTTLPR predicted decreased risk of PTSD among persons living in low-crime counties (odds ratio (OR)=0.61, 95% confidence interval (CI): 0.34, 1.10) but increased risk in high-crime counties (OR=1.54, 95% CI: 0.72, 3.30).
Table 3 presents results from the multivariable logistic regression models for unemployment rate and PTSD. In the main-effects model, low social support and high hurricane exposure were associated with significantly increased risk of PTSD. There was no significant main effect for 5-HTTLPR genotype or county-level unemployment rate. In the interaction model, the interaction term for 5-HTTLPR genotype × unemployment rate was significant (P = 0.007). Figure 2 presents the prevalence of PTSD for persons from counties with high unemployment versus those with low unemployment by 5-HTTLPR genotype. Stratified analyses showed that the s′ allele of 5-HTTLPR predicted decreased risk of PTSD among persons living in low-unemployment counties (OR=0.35, 95% CI: 0.14, 0.87) but increased risk in high-unemployment counties (OR=1.46, 95% CI: 0.82, 2.61). Note that because of the loss of power in stratified analyses, only the effect of 5-HTTLPR genotype in low-unemployment counties was statistically significant.
To our knowledge, this study was the first to document a significant interaction between a specific gene and features of the group-level social environment in the risk of a major disorder. We found that county-level crime rate and employment rate modified the association between 5-HTTLPR genotype and risk of PTSD. Although our interpretation of results from the stratified analysis is constrained by low statistical power due to the relatively few PTSD cases in this sample, results suggest that the s′ allele of the 5-HTTLPR polymorphism was associated with decreased risk of PTSD in the low-risk environments (low crime/unemployment rates) but increased risk of PTSD in the high-risk environments. If replicated, this finding is an important extension of the literature, since previous investigations of gene-environment interaction in PTSD have considered only individual-level variables.
The interaction effect for 5-HTTLPR genotype and crime/unemployment rate was significant after controlling for individual-level determinants of PTSD, including sex, hurricane exposure, other potentially traumatic events, and social support. This suggests not merely that the findings are a function of people's living in areas with greater exposure to potentially traumatic events or low social support, but rather that something else about the high-risk social environment influences susceptibility to PTSD among carriers of the s′ allele. Our finding that the s′ allele was associated with decreased risk of PTSD in the low-risk social environments is similar to that of a study which found the s/s genotype to be protective against depressive symptoms among adults raised in a supportive family environment but to increase risk among those raised in a harsh environment (42). Animal studies have also found the s/s genotype to be protective under positive conditions of maternal rearing; effects were reversed under adverse conditions of peer rearing (43). These findings suggest that the s/s genotype is less a risk factor for PTSD or depression per se than a reflection of heightened sensitivity to environmental influences. Human brain imaging studies suggest that carriers of the “s” allele are more attuned to negative emotional stimuli (44, 45). From an evolutionary perspective, such attunement may have been adaptive for survival, thus maintaining the frequency of the “s” allele in the population.
There are at least 4 potential limitations of the data presented in this article. First, our sample was not representative. Although our response rate for the interview portion of the study was better than is typical (46), the rate of return of saliva samples was lower than optimal. However, nonparticipation in the genetic study was not significantly related to the major study variables of hurricane exposure, social support, or PTSD, suggesting that participation bias is unlikely to have influenced our findings (19). Second, it is possible that exposure to potentially traumatic events in the counties with adverse social environments was worse than that experienced in the low-crime/-unemployment counties. Although we controlled for hurricane exposure and other potentially traumatic events in our analyses, it is possible that some unmeasured aspect of these events is driving the association between social environment and PTSD. Third, these findings may not be generalizable to PTSD that follows other types of traumatic events. Finally, the number of affected persons was small, in the context of a reasonably large sample. Thus, we had very limited statistical power for stratified analyses.
This area of research is novel, and we can only offer early conjecture about the mechanisms through which the social environment might influence the risk of PTSD. Much work is needed to either replicate or refute the findings documented here and to understand the mechanisms that may explain these observations. These data argue for extending gene-environment interaction studies to include features of the social environment in future research on population-representative samples. Although twin studies have shown that the heritability of a wide range of individual characteristics and behaviors is modified by environmental factors such as socioeconomic status in childhood (47, 48), early family adversity (49), and school environment (50), no studies to our knowledge have examined whether aspects of the social environment modify the association between specific genes and disorder. The relative risk of disease conferred by the social environment is likely to be lower than that conferred by individual-level risk factors. However, the ubiquity of exposure to social environmental variables suggests that their role in determining the population distribution of PTSD and other major mental disorders will be substantial.
Author affiliations: Department of Society, Human Development, and Health and Department of Epidemiology, School of Public Health, Harvard University, Boston, Massachusetts (Karestan C. Koenen); Department of Epidemiology, School of Public Health, University of Michigan, Ann Arbor, Michigan (Allison E. Aiello, Erin Bakshis, Sandro Galea); Department of Epidemiology, Mailman School of Public Health, Columbia University, New York, New York (Sandro Galea); Department of Psychiatry and Behavioral Sciences, Medical University of South Carolina, Charleston, South Carolina (Ananda B. Amstadter, Kenneth J. Ruggiero, Ron Acierno, Dean G. Kilpatrick); and Departments of Psychiatry, Genetics, and Neurobiology, School of Medicine, Yale University, New Haven, Connecticut (Joel Gelernter).
This research was supported by National Institutes of Health grants MH05220, MH07055, and DA15105. Dr. Koenen was also supported by grants MH070627 and MH07828, Dr. Amstadter by grant MH083469, and Dr. Galea by grants MH078152, MH082729, and DA022720.
Conflict of interest: none declared.