No Replication of Alzheimer’s Disease Genetics as a Moderator of Combat Exposure’s Association with PTSD risk in 138,592 Combat Veterans

Abstract

Large-scale cohort and epidemiological studies suggest that posttraumatic stress disorder (PTSD) confers risk for late-onset Alzheimer’s disease (AD) and related dementias (ADRD); however, the basis for this association remains unclear. Several prior studies of military Veterans have reported that carriers of the apolipoprotein E (APOE) ε4 gene variant are at heightened risk for the development of PTSD following combat exposure, suggesting that PTSD and ADRD may share some genetic risk. This cohort study was designed to further examine the hypothesis that ADRD genetic risk also confers risk for PTSD. To do so, we examined APOE ε4 and ε2 genotypes, an AD polygenic risk score (PRS), and other Veteran-relevant risk factors for PTSD in age-stratified groups of individuals of European (n = 123,372) and African (n = 15,220) ancestry in the US Department of Veterans Affairs’ Million Veteran Program. Analyses revealed no significant main effect associations between the APOE ε4 (or ε2) genotype or the AD PRS on PTSD severity or diagnosis. There were also no significant interactions between measures of AD genetic risk and either combat exposure severity or history of head injury in association with PTSD in any age group. We conclude that the association between PTSD and the primary ADRD genetic risk factor, APOE ε4, that was reported previously was not replicable in the largest relevant dataset in the world. Thus, the epidemiological association between PTSD and ADRD is not likely to be driven by the major genetic factors underlying ADRD risk.

Individuals with posttraumatic stress disorder (PTSD) are at heightened risk for age-related cognitive decline and dementia^1-10, with some population estimates suggesting that the risk for dementia is increased three- to four-fold among those with PTSD compared to controls⁴. Though the mechanisms linking PTSD to dementia in later life remain unclear, various hypotheses for the association have been advanced (see ¹¹), including the possibility that it reflects shared genetic risk. The strongest and most well-known genetic risk factor for dementia is the ε4 variant of the apolipoprotein E (APOE) gene. Inheriting two copies of this variant is associated with a greater than 14-fold increased risk for AD among those with European ancestry (EUR)^12,13 and a three- to four-fold increase among those with African ancestry (AA)^14,15. Risk is also substantially increased in both populations in the presence of just one copy of the ε4 variant^12,16. Apart from APOE, other common genetic risk variants have a more modest effect on AD risk¹⁷. APOE ε4 has an age-related risk profile relative to the more common ε3 genotype¹², as the ε4 risk variant is associated with earlier onset of AD^18,19, consistent with an accelerated aging phenomenon. In contrast, the APOE ε2 variant, which is less common than ε4, is associated with reduced risk of AD relative to the ε3 variant. Homozygosity for ε2 occurs rarely in the population and is associated with substantially reduced AD risk²⁰, but increased risk for type III hyperlipoproteinemia (characterized by high total cholesterol and triglycerides)²¹ and peripheral vascular disease²². APOE codes for apolipoprotein E, which regulates cholesterol metabolism. Its primary influence on risk for AD is via the accumulation and insufficient clearance of amyloid beta (Aβ) in the brain²³. However, the gene also impacts several other biological mechanisms relevant to both AD and PTSD, potentially leading to their comorbidity, including synaptic plasticity, hippocampal neurogenesis, neuroinflammation, and glucose metabolism²³. The ε4 variant also increases risk for cardiovascular and peripheral metabolic disorders²⁴. Studies have also shown that APOE ε4 may interact with PTSD to increase dementia risk, as PTSD is associated with a larger shift in dementia risk in APOE ε4 carriers¹⁰, and is associated with a higher impact of PTSD on cognitive performance in the presence of ε4²⁵. There is also some evidence for polygenic overlap between PTSD and AD²⁶.

Two studies have suggested that APOE ε4 may be associated with increased PTSD severity and/or diagnostic prevalence in the context of adverse environments that contribute to PTSD²⁷, such as combat exposure^28,29. In the first, Lyons et al.²⁸ reported an interaction between the APOE ε4 allele and the number of combat experiences such that the association between combat exposure and both PTSD diagnosis and symptom count was greater among ε4 carriers than non-carriers given a similar combat exposure level. This was observed in a small sample of 170 primarily White, middle-aged (mean age: 55 years) male combat-exposed US Veterans in the Vietnam Era Twin Study of Aging (VETSA). In a larger study of 765 young non-Hispanic White and 859 young non-Hispanic Black Veterans of the post 9/11 conflicts, Kimbrel et al.²⁹ reported an interaction between APOE ε4 and combat exposure on PTSD that was evident only among non-Hispanic Black participants. Thus, there is mixed evidence for an ε4 X combat exposure interaction in association with PTSD. Only one small study (n = 59) reported some limited evidence of an ε4 main effect in association with PTSD reexperiencing symptoms, after adjustment for lifetime trauma exposure³⁰. In contrast, other candidate gene studies, large-scale GWASs of PTSD, and a large scale GWAS of reexperiencing symptoms have not identified associations between PTSD and APOE ε4 or the APOE region^31-35.

Main effects of the APOE ε2 on PTSD diagnosis³⁶ and symptom severity³⁷ have been reported in two small studies, though a meta-analysis of just four studies reported no association³⁸. Mice with the ε2 allele show impaired extinction to conditioned fear^37,39, suggesting that the variant may be associated with alterations in fear-related learning and fear circuitry. No prior studies have evaluated these ε4 and ε2 main or interactive associations across groups differing in both age and ancestry, and none have examined this in a cohort of individuals over the age of 65.

This study addressed these limitations using quantitative data from combat-deployed US military Veteran participants in the U.S. Department of Veterans Affairs’ Million Veteran Program (MVP), representing three age groups (spanning young, middle, and old age) and two genetic ancestry groups (EUR and AA). In addition to evaluating APOE ε4 interactions with combat exposure (and main effects) on PTSD severity and risk, we also examined the hypothesis that a history of head injury could be a confounding factor in the combat exposure x APOE ε4 interaction, as head and traumatic brain injury (TBI) are correlated with combat exposure and also associated with greater PTSD severity^40-42. TBI has also been shown to interact with APOE ε4 in association with self-reported PTSD symptoms⁴³. Based on this, we also performed a parallel investigation of a head injury X ε4 interaction. Though our primary aim was to test for replication of the APOE ε4 X combat interaction association with PTSD, this study also provided the opportunity to conduct a parallel set of analyses evaluating the APOE ε2 isoform. Finally, we evaluated the possible effects of an AD polygenic risk score (PRS) that excluded the APOE region to further evaluate broad genetic AD risk in association with PTSD.

Results

The descriptive statistics for the EUR and AA MVP participants reporting a combat deployment who also had genotype data are presented in Table 1, separated by age group. The demographic factors in Table 1 varied significantly by age group for almost all examined factors (p < 0.05) except for head injury in the AA group (p = .094) and APOE ε2 carrier proportion in the EUR cohort (p = .095). The ε2 carrier proportion did not approach a significant difference by age group in the AA cohort (p = .93). However, the lower significance of associations in the AA cohort might be due to the lower sample sizes relative to the EUR cohort across all age categories. The number of subjects in each age group varied, with a larger number of subjects in the older age categories in both the EUR and AA cohorts. Notably, the cohorts examined were all much larger than the young and middle-aged cohorts of prior publications reporting combat x APOE ε4 interactions.

Table 1.

Demographic Characteristics of the Cohorts

	EUR Combat cohort (N = 123,372)				AA Combat cohort (N = 15,220)
	Young Age n = 7,242 M (SD) or n (%)	Middle Age n = 28,658 M (SD) or n (%)	Old Age n = 87,472 M (SD) or n (%)	p-valuea	Young Age n = 1,316 M (SD) or n (%)	Middle Age n = 6,466 M (SD) or n (%)	Old Age n = 7,438 M (SD) or n (%)	p-valuea
Age	36.61 (5.47)	58.17 (6.04)	72.63 (7.01)	<2.2e-16	38.24 (5.26)	56.49 (5.95)	71.11 (5.64)	<2.2e-16
Sex (male)	5795 (80.02%)	26771 (93.42%)	87027 (99.49%)	<2.2e-16	914 (69.45%)	5626 (87.01%)	7364 (99.01%)	<2.2e-16
Sex (female)	1447 (19.98%)	1887 (6.58%)	445 (0.51%)	<2.2e-16	402 (30.55%)	840 (12.99%)	74 (0.99%)	<2.2e-16
War Erab
Post-9/11	4911 (67.81%)	7999 (27.91%)	1377 (1.57%)	<2.2e-16	719 (54.64%)	1683 (26.03%)	138 (1.86%)	<2.2e-16
Gulf War	2967 (40.97%)	13145 (45.87%)	4829 (5.52%)	<2.2e-16	623 (47.34%)	3132 (48.44%)	484 (6.51%)	<2.2e-16
Post-VN Cold War	461 (6.37%)	13578 (47.38%)	11813 (13.50%)	<2.2e-16	104 (7.90%)	2827 (43.72%)	1074 (14.44%)	<2.2e-16
Vietnam War	0 (0.000%)	13135 (45.83%)	66212 (75.70%)	<2.2e-16	1 (0.076%)	1821 (28.16%)	5464 (73.46%)	<2.2e-16
Pre-Vietnam	0 (0.000%)	42 (0.15%)	29799 (34.07%)	<2.2e-16	0	14 (0.22%)	1728 (23.23%)	<2.2e-16
Head Injury	1105 (15.26%)	3070 (10.71%)	7083 (8.10%)	<2.2e-16	137 (10.41%)	603 (9.33%)	650 (8.74%)	0.094
Combat Sev	26.49 (11.40)	24.32 (11.11)	25.35 (12.00)	<2.2e-16	25.12 (12.10)	24.31 (12.19)	27.96 (14.48)	<2.2e-16
Smoker	1257 (17.36%)	4931 (17.21%)	8266 (9.45%)	<2.2e-16	212 (16.11%)	1381 (21.36%)	1274 (17.13%)	<2.2e-16
Prob. Alch Use	2016 (27.84%)	7213 (25.17%)	20683 (23.65%)	<2.2e-16	294 (22.34%)	1243 (19.22%)	1062 (14.28%)	<2.2e-16
Education	4.21 (1.28)	3.89 (1.46)	3.81 (1.57)	<2.2e-16	4.04 (1.29)	3.71 (1.36)	3.41 (1.43)	<2.2e-16
PTSD DX	2343 (32.35%)	7617 (26.58%)	12760 (14.59%)	<2.2e-16	587 (44.60%)	2588 (40.02%)	2223 (29.89%)	<2.2e-16
PTSD Sev	40.97 (18.13)	37.88 (17.42)	31.58 (14.90)	<2.2e-16	45.85 (20.13)	43.90 (19.42)	38.96 (17.99)	<2.2e-16
APOE: ε4 Carrier	25.64%	25.53%	23.97%	1.07e-07	35.26%	34.41%	32.99%	0.037
APOE: ε2 Carrier	11.71%	12.55%	12.88%	0.095	10.94%	11.34%	11.64%	0.93

Note. Education was coded on a 1 to 7 scale where 1 = less than high school and 7 = professional or doctorate degree; categories associated with mean values: 3 = some college credit; 4 = associate’s degree.

^aTwo-sided test of a difference of means or difference of proportions by age category (i.e., ANOVA or chi-square) without any covariates or interaction terms included.

^bWar era categories were not mutually exclusive. Dates for each era were as follows: Post-9/11 – September 2001 and later; Gulf War – August, 1990 to August, 2001; Post-VN (Vietnam) Cold War – May, 1975 to July, 1990; Vietnam War – August, 1964 to April, 1975; Pre-Vietnam – July, 1964 or earlier. We excluded four individuals from the War Era reporting who self-reported non possible eras based on their age.

Sev = severity; prob = problematic; alch =alcohol; PTSD = posttraumatic stress disorder; dx = diagnosis.

Analyses of PTSD Symptom Severity

Combat X APOE ε4.

Table 2 shows the associations between PTSD symptom severity and combat exposure and covariates (age and sex; ancestry substructure PCs were also included in the model but are not shown in the table for simplicity) in each age group and ancestry cohort. As expected, PTSD severity was strongly associated with combat exposure. It was also negatively associated with age in the middle and old age cohorts, but age was positively associated with PTSD in the young age groups. Sex was not significantly related to PTSD severity, but this may be due to low power as the sample was primarily male. Table 2 also shows the main and interactive effects of combat and APOE ε4 on PTSD severity. There were no main effects of APOE ε4 on current PTSD severity in any of the cohorts (p > 0.07). There were no combat X APOE ε4 interactions in any of the age/ancestry cohorts (p ≥ 0.18).

Table 2.

Combat, Head Injury, APOE ε4, and AD PRS as Predictors of PTSD Severity: Results of Linear Regression Models in Age Cohorts

Young Age	EUR				AA
	Combat Model		HI Model		Combat Model		HI Model
	B	p	B	p	B	p	B	p
Covariate Blocka
Age	0.20	2.42E-08	0.13	4.00E-04	0.26	0.0086	0.21	0.038
Sex (M)	−2.89	9.07E-09	−0.076	0.88	−0.59	0.61	1.17	0.32
Combat or HI	0.65	6.89E-271	14.76	7.26E-142	0.59	7.99E-39	13.00	6.03E-13
Genotype Blockb
APOE: ε4	−0.019	0.96	0.16	0.70	1.08	0.26	1.30	0.19
AD PRS	0.087	0.66	−0.024	0.91
Interaction Blockc
Combat or HI X APOE: ε4	0.017	0.63	−0.12	0.92	0.042	0.58	−4.11	0.21
Combat or HI X AD PRS	0.036	0.033	−0.10	0.86
Middle Age	Combat Model		HI Model		Combat Model		HI Model
	B	p	B	p	B	p	B	p
Covariate Blocka
Age	−0.23	2.65E-45	−0.15	4.99E-18	−0.24	9.21E-10	−0.21	4.07E-07
Sex (M)	−1.79	5.85E-06	0.36	0.38	−2.24	0.0014	−0.43	0.55
Combat or HI	0.56	<E-280	12.92	<E-280	0.50	2.HE-145	10.20	3.98E-35
Genotype Blockb
APOE: ε4	−0.12	0.55	−0.095	0.65	0.61	0.14	0.82	0.060
AD PRS	0.056	0.56	0.12	0.22
Interaction Blockc
Combat or HI X APOE: ε4	−0.015	0.40	1.16	0.091	0.0076	0.83	−0.086	0.95
Combat or HI X AD PRS	0.012	0.16	0.39	0.22
Old Age	Combat Model		HI Model		Combat Model		HI Model
	B	p	B	p	B	p	B	p
Covariate Blocka
Age	−0.35	<E-280	−0.42	<E-280	−0.48	3.59E-42	−0.61	2.63E-64
Sex (M)	−0.71	0.28	2.24	0.001	−2.74	0.16	1.15	0.57
Combat or HI	0.41	<E-280	9.66	<E-280	0.38	3.71E-166	10.84	2.77E-51
Genotype Blockb
APOE: ε4	0.098	0.33	0.077	0.46	0.32	0.37	0.069	0.85
AD PRS	−0.029	0.53	−0.044	0.36
Interaction Blockc
Combat or HI X APOE: ε4	0.011	0.18	0.20	0.59	−0.0059	0.81	1.40	0.28
Combat or HI X AD PRS	9.64E-05	0.98	−0.26	0.13

Note. Results from multiple linear regression models. Sample size for EUR young/middle/old age covariates blocks were n = 7,242/28,658/87,472, respectively. When genetic variables were added in the second and third steps of the models, the sample sizes for APOE ε4 analyses were n = 7,089/28,102/85,766 and for AD PRS were n = 7,242/28,658/87472 for the young/middle/old age models, respectively. Sample size for AA young/middle/old age covariate blocks were n = 1316/6,466/7,438, respectively. When genetic variables were added in the second and third steps of the models, the sample sizes for the APOE ε4 analyses were n = 1171/5,857/6,726 for the young/middle/old age models, respectively. All p-values are unadjusted two-sided tests. P-value adjustments for multiple testing are reported in the main text.

^aCombat and HI were evaluated in separate models but we also examined a model with age, sex, and both combat and HI included at the same time. Both combat and HI were significantly associated with PTSD when included in the same model. All estimates are from the block of the model that did not include the genotypes or the terms added in the interaction block.

^bAncestry substructure PCs were also included in this model. APOE ε4 and AD PRS were evaluated in separate models. All genotype main effects are from the 2^nd block of the model that included the covariates but did not include the interaction term.

^cInteraction terms were each evaluated separately in the model with the variables contributing to the interaction term also included as main effects in the model.

HI = head injury; AD PRS = Alzheimer’s disease polygenic score; PTSD = posttraumatic stress disorder; EUR = European ancestry; AA = African ancestry

Head Injury X APOE ε4.

We next examined the potential role of head injury in place of combat in the APOE ε4 interaction models. Head injury was associated with current PTSD severity in every age group in both the AA and EUR cohorts (p ≤ 6.03E-13; Table 2). There was one nominally significant interaction for Head Injury X ε4 in the young AA cohort (p = .042), however, this did not survive a multiple testing correction.

Main and Interactive Effects of APOE ε2.

There were no main effects of APOE ε2 on PTSD severity in any cohort. There were also no interactive effects of APOE ε2 with either combat exposure or head injury on PTSD severity (Table S1).

AD PRS as a Moderator.

We next examined the possibility of a non-APOE related interaction between genetic risk for AD and combat (and separately, head injury) using an AD PRS that excluded the APOE region. As shown in Table 2, there were no significant PRS main effects on PTSD severity (p ≥ 0.5). There was a nominally significant interaction for Combat X AD PRS in the young EUR cohort, p = .033, but this did not survive correction for multiple testing. There were no significant Head Injury X AD PRS associations with PTSD severity.

Analyses of PTSD Case status

We also examined APOE ε4, ε2, and AD PRS (alone and in interaction with combat and head injury) as predictors of probable current PTSD diagnosis, using a cut-point on the PCL-C. There was a nominally significant (p = .040; Table S2) main effect of AD PRS on probable PTSD diagnosis among the middle age EUR group in the head injury model, but this effect was not significant after p-value threshold adjustment for multiple testing. There were no significant interaction effects in any model or cohort and no significant effects involving ε2 (Table S3).

Discussion

This is the largest study to date to address the question of whether the impact of combat exposure on PTSD symptom severity and diagnostic status is moderated by the APOE ε4 genotype. Prior studies have yielded intriguing, though mixed, results concerning possible G X E interactions between ε4 and combat (and TBI) in risk for PTSD^28,29,38,43. In a comprehensive analysis that spanned young, middle, and old-age cohorts and both European and African ancestries, we found no significant G X E effects associated with either PTSD severity or probable diagnosis. As with APOE ε4, there were no interaction effects of cumulative genetic AD risk on PTSD severity or probable diagnosis, as indexed by a polygenic AD score that excluded the APOE region^50,51, nor were their any main or interactive effects of APOE ε2 on PTSD. Given the large sample size employed for this study, particularly in the EUR cohorts, we suspect prior reports of these interactions may have been sample-specific associations that do not generalize to a more representative sample of U.S. military Veterans.

We also found no evidence of a main effect of ε4 or ε2 on PTSD severity or probable diagnosis in any of our age-based cohorts, which is perhaps unsurprising given the lack of association in this region in large-scale GWAS studies of PTSD (e.g., ³⁵). Based on these results, it seems unlikely that the increased risk for late-life AD/dementia given PTSD that has been reported in prior studies^1,2 is due to the pleiotropic effects of APOE ε4 increasing risk for both conditions. Although a null result can never be proven, our conclusion is strengthened by similar null findings using alternate methodologies which collectively suggest that while PTSD is associated with increased ADRD risk, this does not appear to be due to ADRD genotypes or associated AD biological processes impacting both conditions. For example, a recent study by Lutz et al.²⁶ examined the genetic overlap between PTSD and late-onset AD in a pleiotropy analysis of summary data from PTSD and AD GWASs. The distribution of p-values for SNPs associated with PTSD at multiple significance thresholds from the PTSD GWAS were more significant in the AD-GWAS than would be expected by chance. However, the effect was less evident in the reverse direction (e.g., AD-associated SNPs showed relatively less enrichment in the PTSD GWAS), though the authors noted this could be due to a disparity in the sample sizes between the two GWASs, as many more cases were included in the ADRD versus PTSD GWAS. Alternatively, the lack of evidence for AD-associated variants as predictors of PTSD may be because AD risk variants simply do not confer increased risk for PTSD. Further, if the association between PTSD and ADRD were due to ADRD-related pathophysiology, we would expect to see established ADRD biomarkers, like Aβ-40 and Aβ-42, associated with PTSD. However, studies evaluating this question have failed to find evidence of association between PTSD and peripheral or central nervous system measures of Aβ-40 and Aβ-42^54-57, or have provided only limited support for this association in select subgroups of individuals⁵⁸.

Rather, the relationship between PTSD and increased dementia risk may be due to other shared risk factors or the adverse biological consequences of chronic PTSD on later brain health. For example, both conditions may be related to genetic risk for mutually associated traits such as cardiometabolic dysfunction⁵⁹. Alternately, PTSD-related cardiometabolic dysregulation, inflammation, and/or oxidative stress^60-68 could increase risk for ADRD. This is supported by some animal studies which identified mechanisms linking PTSD to AD that related to neuronal death, oxidative stress, and glucocorticoid signaling^69,70. Confounding environmental factors could also increase the odds for both conditions (e.g., combat exposure, head injury, toxin exposure) which could contribute to the association between PTSD and ADRD. Several behavioral factors (e.g., insufficient sleep, substance misuse) that often cooccur with PTSD could also increase dementia risk^71,72. Finally, it is possible that the lack of evidence for AD-related genotypes and biomarkers on risk for PTSD is because associations between PTSD and dementia are not specific to AD per se. It could be that PTSD increases risk primarily for other types of dementia (e.g., vascular dementias) and that we have not yet been able to observe this specific association, given the challenges of differentiating vascular dementia from AD among living participants.

Study Limitations

Several limitations of our study should be considered when evaluating these results. While we have information about current PTSD symptomatology in young, middle, and old age, we are unable to model the course of PTSD symptoms since deployment so we cannot discriminate between newly developed symptoms and chronic symptoms. The sample sizes in the age groups also varied, and therefore, it is unclear whether effects in the young and middle-age groups may have been missed due to reduced statistical power in these smaller subgroups. However, we note that our sample size is much larger than those of the original studies reporting these associations. Specifically, our young AA cohort was 50% larger, and our young EUR cohort was 9 times the size, of the corresponding younger AA and EUR cohorts examined by Kimbrell et al.²⁹. Our middle age AA and EUR cohorts were 38 and 169 times the size of the primary EUR cohort examined in Lyons et al.²⁸, respectively. Therefore, our study was well powered to replicate these associations if they were apparent. This study examined a large cohort consisting primarily of male Veterans, similar to the original reports, and it is unclear whether these results would generalize to cohorts of women or to civilians exposed to war-related (or other) trauma. Further, Veterans who participated in MVP are not fully representative of the broader U.S. Veteran population as they are treatment-seeking Veterans cared for at VA hospitals nationwide. The prevalence of current probable PTSD in this cohort (approximately 15-44% per subgroup) was greater than that reported for the broader Veteran population (approximately 5%; RAND, 2015). Thus, this sample was enriched for PTSD and this is likely due to several reasons including that we included only combat-exposed Veterans, that Veterans receiving VA care are approximately twice as likely to have a PTSD diagnosis relative to Veterans not treated by VA⁷³, and that we employed a self-report measure of PTSD which is prone to item over endorsement⁷⁴. We note that the estimated prevalence of PTSD in this cohort is similar to that obtained by Stein et al. in their MVP GWAS³⁵ which defined lifetime PTSD cases using a probabilistic scoring algorithm derived from EMR data⁷⁵. Enrichment for PTSD would be expected to increase the chances of observing significant effects if they were present. Given this, while the MVP cohort may not be representative of the broader US Veteran population, we suspect that this feature of the cohort allows for a more powerful test of the associations of interest. We investigated two types of Veteran-related environmental stressors that have been associated with PTSD and AD (i.e., combat exposure and head injury), but there are undoubtedly other measures of trauma exposure (e.g., lifetime trauma exposure, childhood trauma) and other environmental risk factors in military and civilian cohorts that might increase risk for PTSD and that may interact with AD genetic risk. As well, as we used a self-report measure of PTSD symptoms, this assessment wasn’t explicitly linked to a specific traumatic experience in the same way that a structured diagnostic interview for PTSD would be administered, which is another difference between our study and prior APOE X combat studies^28,29. Finally, we note that while we examined APOE ε4, the strongest AD genetic risk variant, APOE ε2, and an AD genetic risk score and did not observe any main effects or interactions associated with PTSD symptom severity, there might be other individual AD risk loci that impact PTSD symptom severity either directly or in interaction with high levels of trauma exposure. For example, Lutz et al. reported evidence of several AD risk loci in the MS4A6A region which might be associated with both AD and PTSD risk²⁶.

Conclusions

In this largest study to date of interactions between AD genetic risk and military-related environmental vulnerability factors for PTSD, we found no evidence that the largest AD genetic risk factor, APOE ε4, was related to PTSD symptom severity or diagnosis among young, middle, or old-age EUR or AA Veterans. Thus, while PTSD appears to increase risk for AD and/or related dementias^1-3, these results, in combination with the lack of APOE-region associations from large-scale PTSD GWAS studies, suggest that the largest AD genetic risk factor, APOE ε4, does not increase risk for PTSD, either alone or in interaction with combat stress or head injury. This highlights the importance of identifying alternative behavioral and biological mechanisms for the purported association between PTSD and dementia so that interventions targeting these pathways can be developed. Future work could employ Mendelian randomization using summary statistics from the most robust AD and PTSD GWASs to test predicted causal associations between the two disorders. Likewise, future work could strive to distinguish associations between PTSD and specific dementia diagnoses to determine if such associations are unique to a particular type of dementia. As the ability to accurately diagnose AD and distinguish it from other dementias among living individuals improves, we may have similarly improved specificity in understanding the relationship, and shared genetic and environmental etiology, between PTSD and specific types of dementias.

Methods

Participants

MVP is a VA-funded biobank which includes genetic data on 658,582 Veterans (MVP Genotype Release Version 4) linked with the VA electronic medical record (EMR). Nearly half of the MVP cohort was age 65 or older at enrollment. Veterans are representative of the broader VA Healthcare users’ population. This study was based on a total sample of 138,592 participants comprising 123,372 EUR and 15,220 AA MVP male and female participants; demographics are listed in Table 1 and Figure 1 provides the STROBE flowchart of enrollment. MVP participants provided a blood sample for genetic analysis and consent for electronic medical record (EMR) access. In addition to EMR data, there were two self-report surveys completed by subsets of MVP participants: the “MVP Baseline Survey” and the “MVP Lifestyle Survey”⁴⁴. These surveys assessed a broad range of topics including employment, social support, personality, psychiatric disorders, health risks, exercise, family history of disease, and military-related trauma. We focused our attention on MVP participants who completed the Lifestyle Survey and who reported having been deployed or serving in a combat or war zone (Lifestyle Survey items 36 and 37). As we were interested in age-dependent effects, analyses were performed within three age group categories: young age (19 to 44, n = 7,242 EUR and n = 1,316 AA), middle age (45-64, n = 28,658 EUR and n = 6,466 AA) and old age (≥ 65; n = 87,472 EUR and n = 7,438 AA), with age determined at the completion of the Lifestyle Survey, which corresponds roughly to age at MVP enrollment. All participants provided written informed consent to take part in MVP, and this study was performed with VA research committee oversight (local review board number 1613685, VA Central IRB number 18-17). Veterans were not compensated for their participation in MVP.

Figure 1.

is a STROBE flowchart that shows how we arrived at our final sample relative to the full MVP cohort. It also shows the cell sizes for the various analytic groups.

Procedures and Measures

Current PTSD symptom severity was determined based on the PTSD Checklist for DSM-IV – Civilian Version (PCL-C)⁴⁵, which was administered as part of the MVP Lifestyle Survey. Total symptom severity scores (17 items, possible scores range from 17 to 85) were the primary dependent variable. Combat exposure was assessed using the Deployment Risk and Resilience Inventory (DRRI) Combat Experiences subscale⁴⁶, which was included in the Lifestyle Survey. Items on this measure assessed exposure to several putative traumatic combat experiences, such as being under friendly fire or being part of a unit that received casualties, on a scale ranging from “never” to “daily or almost daily.” Items were summed to index total combat severity (15 items, possible total score ranged from 15 to 75), consistent with the standard scoring for the DRRI. Head injury determinations were derived from responses to item 39 on the Lifestyle Survey. Veterans who endorsed a history of a deployment-related “head injury” or an injury which resulted in “having symptoms of concussion afterwards (such as headache, dizziness, irritability, etc.)” were considered positive for head injury. We examined this variable rather than EMR-based data because we wanted to examine the long-term impact of deployment-related head injuries, and these would not be present in the EMR for Veterans of this service era. In the sensitivity analyses (described below) we examined current probable PTSD diagnosis, as defined by a cut-point score ≥ 50 on the PCL-C⁴⁵.

Genotyping and Genetic Data Statistical Processing

The chip design and genotype cleaning pipeline have been described in detail elsewhere⁴⁷ and will be described briefly here. The genotype data were generated using the MVP 1.0 custom Axiom array which assesses 668,418 markers. Several quality control (QC) measures were implemented. These included checking genotype and self-reported sex concordance, correction for genotyping batch, and checking for genetic relatedness among participants. The MVP Phase 4 genotype data (October 2020) includes imputed genotype data for 658,582 subjects. The imputation was performed using the African Genome Resources (AGR) panel from the Sanger Institute which includes all of the 1000 Genomes Phase 3 reference panel plus an additional cohort of unrelated African-descent individuals⁴⁸. Ancestry was determined by the Harmonized Ancestry and Race/Ethnicity (HARE) method⁴⁹, which is based on self-report ancestry and genome-wide genotype data. It is similar to other genetic-based algorithms for determining ancestry but excludes subjects whose genetic ancestry differs substantially from the mean of principal components of ancestry for their self-report racial group. Racial and ethnic ancestry reported in Table 1 were determined via the genotype data using the HARE method. All participants were of non-Hispanic ancestry as we did not have sufficient sample size to examine these interactions in each age group among the smaller cohort with Hispanic ancestry. Sex was determined via the medical record. APOE ε4 and ε2 isoform genotypes were determined based on the imputed genotypes for rs7412 and rs429358 using an 80% probability call threshold.

Data Analyses

Multiple linear regression models were conducted to predict PTSD symptom severity in each age and ancestry cohort. We first entered the covariates (age, sex) and combat exposure total score in a “covariate block.” Next, we added APOE ε4 as a genetic risk factor along with the top 3 ancestry substructure principal components (PCs) in the second step of the model in a “genotype block.” The interaction between the genetic variable and combat exposure was subsequently added in a third step, the “interaction block.” Results for each term are reported from the block where a given variable was initially entered (e.g., the APOE main effects reported in block 2 are estimated with the covariates but without the interaction term in the model). All terms entered into initial blocks were included in subsequent blocks regardless of their statistical significance from the initial block. This series of models was evaluated separately in the EUR and AA cohorts. We used the Bonferroni method to adjust the p-value threshold for statistical significance by correcting the threshold for 6 tests (i.e., 2 ancestries X 3 age groups) at p < .008. A separate parallel set of analyses was conducted to evaluate the main and interactive effects of the APOE ε2 variant. Based on calculations with R’s pwr function, we were well powered to detect the effects noted in Lyons et al.²⁸ and Kimbrell et al.²⁹. Specifically, we estimated 80% power to detect an interaction effect explaining approximately 1/10^th to 6/10ths of a percent of the variability in PTSD across the African American MVP age cohorts and 1/100^th and 1/10^th of a percent across the European Ancestry MVP age cohorts.

To explore the potential role of confounding caused by the correlation of combat exposure severity and head injury, we performed a similar series of models replacing combat exposure severity with the presence/absence of head injury in the models described above. We also evaluated a full model which included combat exposure and head injury simultaneously.

Next, in the EUR cohort, we examined a similar series of models evaluating the AD PRS for potential main and interactive effects on PTSD severity to examine the possibility that any observed association was due to PTSD interactions with general AD genetic risk rather than APOE specifically. We did not examine the PRS in the AA cohort, as it was based on EUR AD GWAS weights, and performance is likely to be much lower in non-EUR cohorts. For EUR MVP participants, we computed an AD PRS based on the results of the AD GWAS of EUR subjects^50,51. The PRS was calculated using the program PRSice⁵². As we evaluated APOE separately from the AD PRS, we computed the PRS excluding the APOE region (Chr 19 44,409,039 to 46,412,650 according to GRCH37/hg19). The PRS was computed using SNPs with p <5x10⁻⁸ (GWAS significant) associations in the original Kunkle et al. study, as recent evidence indicates that AD genetic risk is oligogenic⁵³.

Finally, an identical series of models was conducted to predict probable PTSD diagnosis using logistic regression (see supplementary materials) to test if the pattern of results for the dichotomous PTSD determinations mirrored those for PTSD symptom severity.

Data Availability

The phenotypes and genetic data analyzed here are available to all researchers with MVP access. Approval to access the individual-level MVP data is only available to VA researchers for an approved and funded MVP project, either through a VA Merit Award or a Career Development Award. See https://www.research.va.gov/funding/Guidance-MVP-Data-Access-Merit-Award.pdf for details. All GWAS results for MVP are made available on dbGAP.

Code Availability

Analytic code is available for those with approved MVP access by writing to the corresponding author. Our group generated this code.