BDNF Genotype is Associated with Hippocampal Volume in Mild Traumatic Brain Injury

Abstract

The negative long-term effects of mild traumatic brain injury (mTBI) have been a growing concern in recent years, with accumulating evidence suggesting that mTBI combined with additional vulnerability factors may induce neurodegenerative-type changes in the brain. However, the factors instantiating risk for neurodegenerative disease following mTBI are unknown. This study examined the link between mTBI and brain derived neurotrophic factor (BDNF) genotype, which has previously been shown to regulate processes involved in neurodegeneration including synaptic plasticity and facilitation of neural survival through its expression. Specifically, we examined nine BDNF single nucleotide polymorphisms (SNPs; rs908867, rs11030094, rs6265, rs10501087, rs1157659, rs1491850, rs11030107, rs7127507, and rs12273363) previously associated with brain atrophy or memory deficits in mTBI. Participants were 165 white, non-Hispanic Iraq and Afghanistan war veterans between the ages of 19 and 58, 110 who had at least one mTBI in their lifetime. Results showed that the BDNF SNP rs1157659 interacted with mTBI to predict hippocampal volume. Further, exploratory analysis of functional resting state data revealed that rs1157659 minor allele homozygotes with a history of mTBI had reduced functional connectivity in the default mode network compared to major allele homozygotes and heterozygotes. Apolipoprotein E (APOE) was not a significant predictor of hippocampal volume or functional connectivity. These results suggest that rs1157659 minor allele homozygotes may be at greater risk for neurodegeneration after exposure to mTBI and provide further evidence for a potential role for BDNF in regulating neural processes following mTBI.

Introduction

Traumatic brain injury (TBI) is a major cause of disability world-wide, with high prevalence rates in certain populations including athletes and military personnel (Zaloshnja et al., 2008, Sahler & Greenwald, 2012, Gubata et al., 2014). Increasing evidence suggests that TBI heightens the risk for neurodegenerative diseases such as Alzheimer’s disease (AD; Mortimer et al., 1991, Fleminger et al., 2003) and chronic traumatic encephalopathy (McKee et al., 2013). Although most of the evidence comes from studies of moderate and severe head injury, there is growing indication that mild TBI (mTBI), or concussion, when combined with additional vulnerability factors such as advanced age, repetitive injury, and genetic risk, is also a risk factor for neurodegeneration (McKee et al., 2013, Hayes et al., 2017b). As mTBI represents the majority of all head injuries (~75%; Bergman & Bay, 2010), studying the factors linking mTBI to brain atrophy and dementia is worthy of greater attention. Here we examined a potential genetic link that may confer risk for AD following mTBI in a military sample exposed to injuries including blast mTBI, sports injuries, and motor vehicle accidents. Specifically, we tested the hypothesis that the confluence of genetic variation in brain derived neurotrophic factor (BDNF) and environmental exposure (mTBI) is associated with reduced structural integrity of the hippocampus.

The hippocampus is a medial temporal lobe structure involved in episodic memory encoding and retrieval (Eichenbaum, 2000). Hippocampal volume atrophy is a well-established biomarker of AD and is the earliest sign of AD disease pathology using MRI (Frisoni et al., 2010). Further, the hippocampus is an important component of the default mode network, an intrinsic connectivity network enhanced during tasks involving internal mentation such as autobiographical memory, self-referential thinking and mind wandering (Spreng et al., 2009). There is now well-established evidence for altered functional connectivity among the nodes of the default mode network in AD (Brier et al., 2012). Alterations in hippocampal networks have also been noted in mTBI (Yan et al., 2016, Rangaprakash et al., 2017). The evidence for hippocampal atrophy, although established for more severe TBI (Bigler et al., 1997, Tate & Bigler, 2000), has less support in the mTBI literature (but see Little et al., 2014 for decreases in parahippocampal gyrus in mTBI). However, it may be that individual differences in BDNF genotype moderate the relationship between mTBI and hippocampal volume, providing a link between mTBI and neurodegeneration.

The BDNF gene, located on chromosome 11p13, is a member of the neurotrophin family. Its expression has a number of roles in the brain, including promoting synaptic plasticity, dendritic arborization, and axonal sprouting in both animal models and humans (Mamounas et al., 1995, Kaplan et al., 2010). BDNF is highly expressed in the hippocampus and is involved in learning and memory (Bekinschtein et al., 2008), thus making it a target for study in TBI and AD. Recent work has demonstrated that BDNF single-nucleotide polymorphisms (SNPs) are associated with lifetime history of concussion (Dretsch et al., 2017), as well as memory and attention performance in mTBI (McAllister et al., 2012). Following mTBI, BDNF mRNA expression is increased in the hippocampus, facilitating neuronal survival (Hicks et al., 1999). By contrast, insufficient levels of BDNF in the central nervous system may lead to loss of synaptic plasticity (Korte et al., 1995) and are associated with increased risk for dementia (Phillips et al., 1991, Weinstein et al., 2014). Consistent with data from animal models, human longitudinal studies have demonstrated that BDNF polymorphisms are associated with hippocampal volume atrophy in probable AD (Hashimoto et al., 2009, Honea et al., 2013), suggesting that BDNF plays a role in AD-related neurodegeneration.

In the present study, we hypothesized that mTBI interacts with BDNF genetic variants to produce reduced hippocampal volume, potentially one indicator of early neurodegenerative disease processes following neurotrauma. We examined six BDNF SNPs that have previously been associated with hippocampal and cerebral atrophy (rs908867, rs11030094, rs6265, rs10501087, rs1157659, rs1491850; Honea et al., 2013), along with four BDNF SNPs associated with long delay recall deficits in mTBI patients (rs11030102, rs11030107, rs7127507, rs12273363; McAllister et al., 2012). In additional analyses, we considered whether the APOE E4 allele was also a significant predictor of hippocampal volume given its important role in AD.

In addition to examining the association between BDNF genotype and mTBI-and AD-related pathology of hippocampal structure, we also explored the influence of BDNF on resting-state brain function in individuals with and without mTBI. Preliminary evidence suggests that BDNF genotype influences nodes of the default mode network in terms of function (Thomason et al., 2009) and structure (Lin et al., 2016). However, it is unknown whether BDNF genotype modulates default mode functional connectivity within the context of mTBI. We hypothesized that BDNF and mTBI would be associated with reduced default mode network functional connectivity, which may be an additional indicator of prodromal risk of neurodegenerative disease following mTBI.

Finally, we note that potential findings related to mTBI, BDNF genotype, and hippocampal volume may be complicated by the high PTSD comorbidity rates in the military population, as PTSD has also been associated with BDNF genotype (Dretsch et al., 2016) as well as hippocampal volume (Levy-Gigi et al., 2015) and function (Miller et al., 2017). Therefore, we also examined the influence of PTSD and BDNF genotype to disentangle the contribution of this condition from mTBI-related alterations in hippocampal volume.

Materials and Methods

Experimental Subjects

The final sample included 165 white, non-Hispanic U.S. combat veterans of Iraq and Afghanistan conflicts who were consecutively enrolled in the Translational Research Center for TBI and Stress Disorders (TRACTS) National Research Center at VA Boston Healthcare System, Jamaica Plain Campus. Participants were recruited from the Boston Metropolitan area through outreach military community events (e.g., Yellow Ribbon events). Exclusion criteria included history of moderate/severe TBI, history of seizures not related to TBI, vascular disease, heart conditions, current serious mental illness including bipolar disorder and schizophrenia, and current suicidal or homicidal ideation requiring intervention. Participants with implanted ferromagnetic objects incompatible with MRI including shrapnel, clips, pacemaker, or who were pregnant were excluded from image acquisition. All available TRACTS data that met selection criteria for the study were included. 197 white, non-Hispanic subjects were identified for the study as they were the largest ancestral subset in the full TRACTS sample and limiting analyses to this group would reduce the concern of population stratification. Of these, 29 were missing neuroimaging data and three were missing clinical assessment data. Participants ranged in age from 19–58 (M = 31.04, SD = 8.2) and were predominately male (92.7%). One hundred ten participants (67%) reported having at least one mTBI in their lifetime, and of those individuals, 59 (54%) had a military-related mTBI, and 52 (47%) had experienced more than one mTBI. Seventy-one (65%) subjects with mTBI also met criteria for PTSD, suggesting high mTBI/PTSD comorbidity. Table 1 summarizes demographic characteristics stratified by TBI and characteristics of the mTBI sample are shown in Table 2. Fifty-five subjects had no history of mTBI and served as the control group. Of these, 20 (36%) subjects also met criteria for PTSD. There were no significant differences in sex, estimated IQ scores, lifetime substance abuse diagnosis, or education between the mTBI and control groups. The control group was slightly older by an average of two years and had significantly lower PTSD symptom severity scores.

Table 1.

Demographic and clinical characteristics.

Variable	Total (n=165)	Lifetime mTBI (n=110)	No Lifetime mTBI (n=55)	Group Comparison	Cohen’s d	95% CI
						Lower	Upper
Age in years, M (SD)	31.04 (8.2)	30.14 (7.7)	32.85 (8.9)	t(163)=2.04, p=0.043*	0.33	0.08	5.36
Males, no. (%)	153 (92.7)	102 (92.7)	51 (92.7)	χ2(1)=0.00, p=1.00	—	0.29	3.48
Education in years, M (SD)	13.93 (2.0)	13.76 (1.9)	14.25 (2.1)	t(163)=1.51, p=0.133	0.24	−0.15	1.13
WTAR Estimated IQ, M (SD)	103.51 (11.5)	103.55 (11.4)	103.42 (11.9)	t(163)=−0.07, p=0.943	0.01	−3.91	3.64
CAPS total, M (SD)	44.73 (28.1)	50.93 (26.3)	32.35 (27.7)	t(163)=−4.20, p<0.001*	0.69	−27.31	−9.85
Lifetime Substance Use Diagnosis no. (%)	108 (65.5)	76 (69.1)	32 (58.2)	χ2(1)=1.93, p=0.165	—	0.82	3.14

Note: WTAR=Wechsler Test of Adult Reading; CAPS=Clinician-Administered PTSD Scale; CI=Confidence Interval. Confidence intervals for categorical variables (Males and Lifetime Substance Use) reflect odds ratio.

^*= statistically significant.

Table 2.

Injury characteristics of the mTBI sample.

Lifetime mTBI (N, Percent)
Total	110, 100%
Military-related mTBI	59, 54%
More than one mTBI	52, 47%
Most Severe Mild TBI (Minutes, SD)
Altered Mental Status	143.58, 461.98
Posttraumatic Amnesia	43.08, 221.69
Loss of Consciousness	2.50, 5.01
Nature of Injury (Most Severe)
Military mTBI
Blast	27
Fall	5
Motor Vehicle Accident	4
Physical Assault	2
Sports-related Injury	4
Other†	5
Total	47
Pre-deployment mTBI
Fall	8
Motor Vehicle Accident	10
Physical Assault	8
Sports-related Injury	23
Other†	6
Total	55
Post-deployment mTBI
Fall	2
Motor Vehicle Accident	2
Physical Assault	1
Sports-related Injury	2
Other†	1
Total	8
Nature of Injury Totals (N, Percent)
Blast	27, 24.5%
Fall	15, 14%
Motor Vehicle Accident	16, 14.5%
Physical Assault	11, 10%
Sports-related Injury	29, 26%
Other†	12, 11%

Note:

^†Other injuries include: hit by blunt object (e.g., pole, bar, vehicle, ceiling). Most severe nature of injury is reported, but many individuals had multiple types of injuries.

Seven participants had incidental findings as reviewed by a board-certified neuroradiologist who reviewed FLAIR, susceptibility-weighted, and high resolution T1 scans. Three subjects from the control group had incidental findings including scattered hyperintensities (n=1) and diffuse atrophy (n=2). Four subjects from the TBI group had findings that were consistent with TBI injury, including focal hemorrhage (n=1), contusion (n=2), and atrophy (n=1). Analyses removing these seven participants revealed no change in the results and therefore these participants were left in the study. Written informed consent consistent with the Declaration of Helsinki was collected from each participant prior to enrollment. All study procedures were approved by the VA Institutional Review Board.

Clinical Measures

Participants were assessed for lifetime mTBI with the Boston Assessment of TBI-Lifetime (BAT-L; Fortier et al., 2014), a semi-structured interview modeled on Department of Defense TBI diagnostic criteria. MTBI diagnosis was defined as a period of posttraumatic amnesia (PTA) less than 24 hours, and/or loss-of-consciousness (LOC) less than 30 minutes, and/or altered mental status less than 24 hours. The BAT-L interview was designed specifically to assess lifetime TBI in Operation Enduring Freedom/Operation Iraqi Freedom (OEF/OIF) military personnel who were exposed to unique experiences such as blast-related injuries from explosive ordnance and psychological war trauma. For this reason, the BAT-L uses a forensic approach to TBI assessment to collect details of the events leading up to, during, and after the injury. Follow-up questions assess memory gaps, unresponsiveness, psychological reaction, and medical treatment in order to assess altered mental status, LOC, and PTA. As medical charts were unavailable, the interviewer queried the participant regarding eyewitness reports, medic reports, and injuries to other veterans present at the time of the event in order to cross-check self-report of the event. TBI diagnosis was ultimately decided via a diagnostic consensus meeting of at least three doctoral-level psychologists. Meetings were held weekly with discussion regarding whether reactions to the injury event could have resulted from physiological disruption of brain function (i.e., TBI), given the presence of other factors that complicate self-report such as psychological response to the situation, changes in sensory abilities, and possible substance use at the time of injury. Participants diagnosed with at least one lifetime mTBI were placed in the mTBI group; those without a lifetime mTBI were placed in the control group.

Doctoral level psychologists with advanced training in psychological assessment evaluated participants for PTSD symptom severity with the Clinician-Administered PTSD Scale (CAPS; Blake et al., 1995), the gold-standard for assessment of PTSD. Mood and substance disorders were diagnosed with the Structured Clinical Interview for DSM-IV (SCID-IV; First et al., 1997). Scores on the Wechsler Test of Adult Reading (WTAR) were included as an estimate of premorbid IQ (Green et al., 2008, Wechsler, 2001).

MRI Acquisition and Processing

Participant structural scans were acquired on a Siemens 3-Tesla TIM Trio whole-body MRI scanner at the VA Boston Healthcare System, Jamaica Plain division. Two T1 three dimensional magnetization-prepared rapid gradient-echo (MP-RAGE) scans were acquired (voxel size = 1mm³; T1 = 1000ms; repetition time =2530s; echo time = 3.32ms; field of view = 256 × 256, number of slices = 176; flip angle = 7⁰). These two MP-RAGE scans were then averaged into a single image to achieve a high contrast-to-noise ratio. The FreeSurfer image analysis suite (version 5.0, freely available for download online http://surfer.nmr.mgh.harvard.edu) pipeline was used which includes motion correction and averaging (Reuter et al., 2010) of multiple volumetric T1 weighted images, removal of non-brain tissue using a hybrid watershed/surface deformation procedure (Ségonne et al., 2004), automated Talairach transformation, segmentation of the subcortical white matter and deep gray matter volumetric structures (including the hippocampus; Fischl et al., 2002, Fischl et al., 2004a), intensity normalization (Sled et al., 1998), tessellation of the gray matter white matter boundary, automated topology correction (Fischl et al., 2001, Ségonne et al., 2007), and surface deformation following intensity gradients to optimally place the gray/white and gray/cerebrospinal fluid borders at the location where the greatest shift in intensity defines the transition to the other tissue class (Dale et al., 1999, Dale & Sereno, 1993, Fischl & Dale, 2000). Automated FreeSurfer hippocampal volumes have shown to have high overlap and correlation with expert hand drawn volumes of the hippocampus (Morey et al., 2009, Mulder et al., 2014, Fischl et al., 2002). Participant imaging data underwent visual examination for accuracy. Gray matter region of interest (ROI) volumes were adjusted to control for head size using the following steps: (1) total intracranial volume (ICV) and ROI raw volumes were extracted from FreeSurfer and (2) each raw volume ROI was then regressed against the group mean ICV to determine the regression coefficient β, which was then inserted into the formula: adjusted volume = raw volume – β (ICV- group mean ICV), where β is the slope of a regression of a region-of-interest volume on ICV, as recommended in Buckner et al., (2004). This approach yields a more Gaussian distribution of adjusted volumes than a ratio (HC/ICV) approach (Jack Jr et al., 1989).

A subset of 139 participants had available functional resting-state data. Two six-minute resting state scans were acquired in the same session during which participants were required to stay awake with their eyes open (gradient echo echo-planar images, TR/TE: 3000/30ms, flip angle: 90°, 3 × 3 × 3.75mm voxels, 38 slices). Resting state data were processed using the FSFAST processing stream (http://fresurfer.net/fswiki/FsFast) and followed a standard preprocessing pipeline that included motion correction, time shifting, concatenation of scans, regression of motion, the average time courses from the white matter and ventricles, and the global mean, and band pass filtering between 0.01 and 0.1 Hz. Data were sampled and smoothed to the surface and warped to a surface-based template (Fischl et al., 1999a). The seed region was derived from surface parcellations of the cortex (Fischl et al., 2002), of which the bilateral superior third of the isthmus of the cingulate (defined in each participant’s native space via FreeSurfer labeling) was used, which corresponds with the posterior cingulate (PCC) core hub region of the default mode network. Group-level surface-based connectivity maps were assessed and clustered at a vertex-wise threshold of p < 10⁻²⁰. This yielded four regions of interest from each hemisphere that included isthmus of the cingulate, angular gyrus, middle temporal gyrus, and medial prefrontal cortex (mPFC). For the purposes of this paper, we examined the connectivity between the PCC and mPFC, and the PCC-isthmus as these regions encompass core nodes and best represent the default mode network.

DNA Genotyping and SNP Selection

The generation of genome-wide genotypes for this sample has been described in detail elsewhere (Sadeh et al., 2016). Briefly, genome-wide genotypes were generated at Pharmacogenomics Analysis Laboratory (PAL) at the Central Arkansas Veterans Healthcare System in Little Rock, Arkansas. DNA was taken from peripheral blood samples and isolated on a Qiagen AutoPure instrument with Qiagen reagents. Blood samples were normalized using Quant-iT™ PicoGreen assays (Invitrogen) and DNA sample quality was evaluated using TaqMan® RNase P Detection assay (Applied Biosystems Assay, Life Technologies, Carlsbad, CA) with fluorescence detection on a 7900 Fast Real Time PCR System (Applied Biosystems, Life Technologies, Carlsbad, CA). Individual DNA samples were run on Illumina HumanOmni2.5–8 BeadChips according to manufacturer’s protocol (Illumina, San Diego, CA). BeadChip arrays were scanned using an Illumina iScan System. Genome-wide call rates were >99% for non-Hispanic white participants. Self-reported ancestry was confirmed via genetic principal components (PCs) analysis by the program EIGENSTRAT (Price, 2006) based on 100,000 SNPs merged with 1000 genomes phase 1 reference data (The 1000 Genomes Project Consortium, 2012). All subjects in the current analysis were self-reported white, non-Hispanic < 6 SDs of the EUR (European ancestry) sample mean on the first 10 PCs. An additional set of PCs were computed for use in modeling potentially confounding population substructure.

Analyses were limited to six BDNF SNPs that were associated with either hippocampal or whole brain atrophy (rs11030094, rs6265, rs1491850, rs1157659, rs10501087, and rs908867) selected from (Honea et al., 2013), along with four SNPs reported in (McAllister et al., 2012) that were associated with long delay recall deficits in mTBI (rs11030102, rs11030107, rs7127507, and rs12273363). BDNF SNP rs11030102 was subsequently removed from analyses due to it being nearly 100% in linkage disequilibrium (LD) with rs11030107 according to 1000 genomes European sample (D′=1.0, R²=0.9945, p<0.0001), thereby leaving nine SNPs in the analyses. Two SNPs (rs10501087 and rs11030102) were imputed. Imputation of rs10501087 and rs11030102 was performed using the IMPUTE2 program (Howie et al., 2009, Marchini & Howie, 2010, Howie et al., 2011, Howie et al., 2012). Prior to imputation, ambiguous SNPs (C/G and A/T), SNPs with more than 1% missing data, and SNPs failing a Hardy-Weinberg equilibrium test (p < 10–6) were removed.

Apolipoprotein E (APOE) genotyping was conducted by the PAL using the following protocol: DNA was extracted from whole blood using the Qiamp mini kit (Qiagen). Genotypes of the two SNPs (rs429358 and rs7412) required to reconstruct the APOE allelic variants were determined by allelic discrimination on an ABI TaqMan® 7900 Fast Real-Time PCR system, using the TaqMan® SNP Genotyping Assays C-3084793-20 and C-904973-10 and TaqMan® Universal PCR Master Mix, AmpErase® UNG. Assays were carried out on an ABI 7900HT apparatus according to the manufacturer’s recommendations.

Statistical Analysis

Volumetric and functional connectivity data were analyzed using hierarchical linear regression models. Variables met the assumptions of multiple linear regression, including linearity, normality, homoscedasticity, and multicollinearity. In the first step of the analysis, hippocampal volumes (left and right) were regressed onto covariates including age, PTSD symptom severity, lifetime substance use diagnosis, premorbid IQ, and education. MTBI (yes/no variable) and each of the nine BDNF SNPs were evaluated individually in the second block, and each mTBI × SNP interaction was examined in the third block. Both nominal (uncorrected) significance and multiple-testing corrected significance (adjusting for analysis of nine SNPs and hippocampal hemisphere) for main effects and interactions was determined using Monte-Carlo null simulation with 10,000 replicates in which the genetic data were randomly permuted between subjects. This analysis imposed strict multiple-testing control while taking into account the correlations between SNPs and the hippocampal hemispheres. Additional regression models added each of the 10 PCs to examine the influence of population substructure. Follow-up tests, adjusted for all covariates, were performed to examine pairwise comparisons between mTBI cases and controls and genotype. To determine specificity of the BDNF findings on hippocampal volume, linear models examined the main effect of APOE (E2/E3 carriers, n = 19; E3/E3 carriers, n = 109, E4 carriers, n = 37) and the mTBI by APOE interaction. Further, to examine selectivity of hippocampal volume, linear models examined the relationship between mTBI, the nine BDNF genotypes, and two other brain volumes, the posterior cingulate cortex (PCC) and amygdala. PCC volume was selected because, like the hippocampus, this area is associated with early decline in AD (see Jack et al., 2010) and is a core component of the default mode network. The amygdala was selected because of its spatial proximity to the hippocampus.

Because of high mTBI and PTSD comorbidity in this population, we examined the influence of PTSD diagnosis with a regression model replacing mTBI and the mTBI × SNP interaction with PTSD diagnosis and PTSD × SNP interaction on hippocampal volume, with age, lifetime substance use, premorbid IQ, and education as covariates. This model was repeated using PTSD symptom severity score in place of mTBI.

The relationship between functional connectivity of the default mode network (PCC-mPFC, PCC-isthmus), mTBI, and BDNF genotype was examined in a subset of subjects with available data (n = 139). Hierarchical regression analysis using the same covariates as reported above was conducted followed by pairwise comparisons to probe significant interaction effects.

Results

BDNF SNP rs1157659 and mTBI Predict Hippocampal Volume

For the left hippocampus, the final regression model indicated a significant main effect of BDNF SNP rs1157659 (p = 0.043), main effect of mTBI (p = 0.018), and age (p = 0.009). In addition, there was a significant rs1157659 by mTBI interaction that survived multiple testing correction across all nine BDNF SNPs and both hemispheres (p = 0.002, corrected p = 0.021; see Table 3). Similarly in the right hippocampus, there was a main effect of the same SNP rs1157659 (p = 0.005), mTBI (p = 0.011), and age (p = 0.001). The rs1157659 by mTBI interaction was also significant (p = 0.001, corrected p = 0.016). No other BDNF SNP by mTBI interaction term was significantly associated with hippocampal volume (see Supplementary Table 1 for interaction results for all SNPs). Regression models with each of the 10 PCs as covariates revealed that the interaction of rs1157659 and mTBI significantly predicted hippocampal volume when taking population stratification within our white, non-Hispanic sample into account. None of the PCs had a main effect on volume (all p’s > 0.3). Rs1157659 genotype frequencies are reported in Table 4. There were no differences in genotype frequencies (i.e., A/A, A/G, G/G) across the mTBI and control groups [χ²(2) = 2.16, p = 0.34].

Table 3.

Summary of regression analysis for rs1157659 and mTBI with left hippocampal volume.

	Model 1				Model 2				Model 3
Variable	B	SE(B)	β	p	B	SE(B)	β	p	B	SE(B)	β	p
Age	−.89	3.88	−.19	.023*	−9.06	3.93	−.19	.022*	−10.15	3.84	−.22	.009*
LSDX	60.09	65.09	.08	.357	64.48	65.75	.08	.328	90.08	64.47	.11	.164
CAPS	−1.15	1.1	−.09	.297	−1.18	1.16	−.09	.313	−1.2	1.13	−.09	.292
WTAR	−3.14	2.67	−.1	.241	−2.92	2.71	−.09	.284	−1.19	2.69	−.04	.660
Education	10.3	16.85	.05	.542	10.56	16.95	.06	.534	17.33	16.63	.09	.299
rs1157659					−28.12	44.18	−.05	.525	139.21	68.34	.25	.043*
Lifetime mTBI					−11.66	66.86	−.01	.862	253.28	106.34	.31	.018*
rs1157659 × mTBI									−283.98	90.19	−.51	.002+*
R2		0.05				0.05				.11
F for ΔR2		1.56				0.22				9.92**

Note: LSDX=lifetime substance use disorder; CAPS=Clinician-Administered PTSD Scale; WTAR=Wechsler Test of Adult Reading; mTBI=mild traumatic brain injury.

⁺corrected value of this interaction term p = 0.021.

^*= statistically significant

Table 4.

BDNF and APOE allele frequencies stratified by mTBI and PTSD diagnosis.

Genotype	mTBI (n=110)	No mTBI (n=55)	PTSD (n=91)	No PTSD (n=74)
BDNF rs1157659
A/A	29(26.4%)	16(29.1%)	27(29.7%)	18(24.3%)
A/G	62(56.4%)	25(45.5%)	52(57.1%)	35(47.3%)
G/G	19(17.3%)	14(25.5%)	12(13.2%)	21(28.4%)
BDNF rs6265
Val/Val	81(73.6%)	38(69.1%)	65(71.4%)	54(73%)
Met/Val	24(21.8%)	15(27.3%)	23(25.3%)	16(21.6%)
Met/Met	5(4.5%)	2(3.6%)	3(3.3%)	4(5.4%)
APOE
E2/E3	12(10.9%)	7(12.7%)	13(14.3%)	6(8.1%)
E3/E3	71(64.5%)	38(69.1%)	54(49.5%)	55(50.5%)
E4 carrier	27(24.5%)	10(18.2%)	24(17.6%)	13(26.4%)

Note: BDNF=brain-derived neurotrophic factor; APOE=apolipoprotein. There were no differences in allele frequencies among mTBI/no mTBI and PTSD/no PTSD groups.

To decompose the rs1157659 by mTBI interaction effect, we examined the effect of genotype (A/A, A/G, G/G) between mTBI cases and controls, adjusting for all covariates. Within G/G carriers, those with mTBI had lower left hippocampal volume than controls (p = 0.007, Cohen’s f² = 0.31). By contrast, A/A carriers with mTBI had higher left hippocampal volume than control A/A carriers (p = 0.013, Cohen’s f² = 0.31). There were no group differences within A/G carriers (p = 0.89, Cohen’s f² = 0.0002). Results in the right hippocampus paralleled those in the left hippocampus. Within the G/G group, those with mTBI had lower right hippocampal volume than controls (p = 0.012, Cohen’s f² = 0.22). By contrast, A/A carriers with mTBI had greater right hippocampal volume than control A/A carriers (p = 0.01, Cohen’s f² = 0.14). There were no group differences among the A/G group (p = 0.72, Cohen’s f² = 0.002).

Given the high comorbidity of mTBI and PTSD in our sample, we next examined the influence of PTSD diagnosis, PTSD symptom severity, and rs1157659 on hippocampal volume. There was no main effect of PTSD on left hippocampus (p = 0.56) or a PTSD by rs1157659 interaction (p = 0.98), nor was there a main effect of PTSD on right hippocampus (p = 0.84) or a PTSD by rs1157659 interaction (p = 0.80). There was no main effect of PTSD symptom severity on left (p = 0.27) or right (p = 0.35) hippocampus, or a symptom severity by rs1157659 interaction on left (p = 0.86) or right (p = 0.69) hippocampus.

Determining the Specificity of Results

To test the specificity of the BDNF SNP by mTBI effect on hippocampal volume, we examined whether APOE interacts with mTBI to produce a similar profile of hippocampal volume as BDNF. Analyses revealed that there was no significant APOE by mTBI interaction on left (p = 0.17) or right (p = 0.66) hippocampal volume. Further, the rs1157659 interaction with mTBI remained significant after including APOE as a covariate for the left (p = 0.002) and right (p = 0.001) hippocampus.

To test the selectivity of the hippocampus as a key region associated with mTBI and rs1157659, we examined the interaction between mTBI and each of the nine BDNF SNPs on PCC and amygdala volume. Analyses revealed that there were no main effects or interactions that survived multiple comparisons correction for any SNP on left PCC or right PCC, or left or right amygdala. Specifically, there was no mTBI by rs1157659 interaction on left (p = 0.16, corrected p = 0.84) or right PCC (p = 0.12, corrected p = 0.74). There was no mTBI by rs1157659 interaction on right amygdala (p = 0.22, corrected p = 0.91). There was a nominal mTBI by rs1157659 interaction on left amygdala (p = 0.04) although this effect did not survive multiple comparisons correction (corrected p = 0.40).

BDNF SNP rs1157659, mTBI, and Default Mode Network

As rs1157659 was the only SNP that significantly interacted with mTBI to predict hippocampal volume, we limited our analyses to this SNP and left PCC seed for subsequent exploration of functional data. The results of the regression model are reported in Table 5. The analysis revealed a significant main effect of mTBI (p = 0.005), and lifetime substance diagnosis (p = 0.036). In addition, there was a significant rs1157659 by mTBI interaction on PCC-mPFC connectivity that survived multiple comparisons correction (p = 0.015, corrected p = 0.039). There were no significant effects for PCC-isthmus connectivity.

Table 5.

Summary of regression analysis in the left default mode network.

	Model 1				Model 2				Model 3
Variable	B	SE(B)	β	p	B	SE(B)	β	p	B	SE(B)	β	p
Age	.00	.00	−.03	.73	.00	.00	−.01	.919	.00	.00	−.02	.825
LSDX	−.06	.02	−.22	.015*	−.06	.22	−.23	.012*	−.05	.02	−.19	.036*
CAPS	.00	.00	.15	.082	.00	.00	.11	.223	.00	.00	.12	.185
WTAR	.00	.00	.01	.884	−.00	.00	−.00	.963	.00	.00	.05	.598
Education	.01	.01	.16	.099	.01	.01	.16	.097	.01	.01	.18	.050
rs1157659					−.01	.02	−.04	.675	.04	.02	.22	.102
Lifetime mTBI					.03	.02	.13	.138	.1	.04	.4	.005*
rs1157659 × mTBI									−.08	.03	−.43	.015*
R2		0.08				0.1				.14
F for ΔR2		2.43*				1.22				6.14*

Note: LSDX=lifetime substance use disorder; CAPS=Clinician-Administered PTSD Scale; WTAR=Wechsler Test of Adult Reading; mTBI=mild traumatic brain injury.

^*= statistically significant.

To decompose the rs1157659 by mTBI interaction effect on PCC-mPFC connectivity of the left default mode network, we examined the effect of genotype in mTBI cases and controls separately, adjusting for all covariates. In mTBI cases, this yielded a significant effect of genotype (p = 0.043; see Fig. 2). Within the G/G group, subjects with mTBI had reduced PCC-mPFC connectivity than controls (p = 0.016, Cohen’s f² = 0.22). There was no significant between-group difference for A/A (p = 0.127, Cohen’s f² = 0.06) or A/G (p = 0.124, Cohen’s f² = 0.03).

Fig. 2. Left adjusted default mode network connectivity between PCC and mPFC as a function of rs1157659 genotype and history of mild TBI.

Connectivity values are adjusted for age, PTSD symptom severity, lifetime substance use diagnosis, premorbid IQ, and education. PCC = posterior cingulate cortex; mPFC = medial prefrontal cortex; AA = major allele homozygotes; AG = heterozygotes; GG = minor allele homozygotes. *p < 0.05.

APOE, mTBI, and Default Mode Network

Analyses revealed that there was no significant APOE by mTBI interaction on default mode network connectivity in PCC-mPFC (p = 0.24) or PCC-isthmus (p = 0.76) connectivity. Further, the rs1157659 interaction with mTBI remained significant after including APOE as a covariate to the main regression analysis (p = 0.017).

Discussion

The objective of this study was to examine the relationship between mTBI and BDNF SNPs previously associated with neurodegenerative disease and reliable MRI indicators of AD including hippocampal volume and default mode network connectivity. One BDNF SNP, rs1157659, interacted with mTBI to predict altered brain metrics across these two different MRI indicators of brain pathology. Lower hippocampal volume and disrupted PCC-mPFC functional connectivity of the default mode network was observed among individuals homozygous for the minor allele in mTBI in comparison to minor allele homozygotes without mTBI, suggesting that minor allele carriers exposed to mTBI may be at greater risk for AD-related indicators of pathology. Individuals homozygous for the major allele in BDNF SNP rs1157659 showed greater hippocampal volume following mTBI relative to controls. This finding was unexpected, and points to the complexity of the association between rs1157659 genotype and mTBI. Clearly, additional study and replication of these findings is warranted to elucidate the role of the major allele in hippocampal volume following mTBI.

Rs1157659 has previously been linked with progressive atrophy of left hippocampal volume (Honea et al., 2013). We extend these results by demonstrating that the added influence of mTBI exposure may further negatively impact hippocampal volume. In addition, we demonstrated that the observed effects were not influenced by APOE, another genetic risk factor for AD, and that the effects were selective to the hippocampus. To date, few other studies have examined rs1157659 and therefore much is not yet known regarding this SNP and its effect on brain and neurodegenerative disease. By contrast, numerous studies have focused on the Val66Met BDNF polymorphism (rs6265) and its association with hippocampal volume and memory performance (Gatt et al., 2009, Gruber et al., 2012, Sanchez et al., 2011, Molendijk et al., 2012b, Richter-Schmidinger et al., 2011, Schofield et al., 2009, Smith et al., 2012, Soliman et al., 2010, Miyajima et al., 2008, Benjamin et al., 2010, Karnik et al., 2010, Kanellopoulos et al., 2011, Brooks et al., 2014, Lim et al., 2014, Kim et al., 2015), although the results have been inconsistent. As of yet, it is inconclusive whether rs6265 has a reliable effect on hippocampal volume as many of these studies were likely underpowered (Molendijk et al., 2012a, Harrisberger et al., 2014). In the current study, rs6265 was not associated with either left or right hippocampal volume as a main effect or interaction with mTBI. These results suggest that the Val66Met genotype should not be exclusively studied and points to the possibility that other BDNF SNPs such as rs1157659 may influence hippocampal volume.

The factors linking mTBI to neurodegenerative disease are currently unknown. One potential process that occurs in both mTBI and AD is loss of synaptic plasticity in the hippocampus. A leading view suggests that prolonged abnormalities in synaptic plasticity, which refers to the modification of neuronal activity as a result of experience, is associated with toxic pathological markers of AD such as amyloid and neurofibrillary tangles (Mesulam, 1999). Consistent with this view, recent work has suggested that alterations in synaptic plasticity are tightly coupled with the clinical features of neurodegenerative disease (Lu et al., 2013). Synaptic plasticity in the hippocampus is also reduced following mTBI (Aungst et al., 2014). Aungst and colleagues demonstrated that both single and repeated mTBI altered synaptic physiology and histopathology of hippocampal subfields relative to a sham group, providing evidence for loss of synaptic plasticity at even mild levels of injury severity (although greater changes in synaptic plasticity occurred with repetitive mTBI). BDNF is a potent factor in enhancing synaptic transmission, plasticity, and growth and can be influenced by environmental factors to produce morphological changes in the hippocampus (Alonso et al., 2004). Given this framework of the neuropathology of neurodegenerative disease, one interpretation of the results is that mTBI may confer risk for neurodegenerative disease through modification of BDNF-controlled synaptic plasticity in the hippocampus.

In addition to the structural findings, we also found that BDNF genotype and mTBI were associated with disrupted functional connectivity among the core nodes (i.e., posterior cingulate cortex and medial prefrontal cortex) of the default mode network. We observed decreased connectivity between these regions when mTBI was combined with rs1157659, suggesting that individual differences in genetic makeup influence the direction of alterations in functional connectivity. As AD has been associated with reduced default mode connectivity in the core hubs (Agosta et al., 2012), it is possible that individuals with the minor allele are most susceptible to functional changes consistent with neurodegeneration following mTBI.

In our military sample, we observed high mTBI and PTSD diagnosis/symptom severity comorbidity. This is not surprising given that mTBI is a hypothesized risk factor for the development and/or maintenance of PTSD (Yurgil et al., 2014) and that the TBI event (e.g., experiencing blast exposure from an improvised explosive device) is also a traumatic event that may be the basis for development of PTSD. Although we did not find a significant association between PTSD, BDNF genotype and hippocampal volume, our findings should not minimize the importance of posttraumatic stress symptoms and genetic risk factors on brain outcomes. PTSD has been associated with BDNF genotype, particularly the Val66Met polymorphism (Felmingham et al., 2013, Dretsch et al., 2016) although there is currently a lack of evidence of a PTSD by BDNF genotype interaction on hippocampal volume. It may be that other candidate genes are more closely tied to smaller hippocampal volumes in PTSD, as recently reported by Hayes and colleagues (2017a). In this study, PTSD symptom severity interacted with COMT to predict reduced hippocampal volume. That is, individuals who were homozygous for the Val/Val genotype and had high symptom severity showed lower hippocampal volume than Met carriers. These open questions regarding PTSD and BDNF genotype notwithstanding, our findings reinforce the notion that BDNF genotype, particularly rs1157659, is a moderating factor in the association between mTBI and hippocampal volume and function and should be further studied as a potential risk factor for neurodegenerative disease following mTBI.

This study has several limitations that should be considered. First, the cross-sectional nature of the study prohibits assessment of the time course of hippocampal volume reductions, and therefore, we are unable to assess whether the reduced hippocampal volume observed among minor allele carriers with mTBI occurred as a result of injury or represents a pre-existing condition. Further, we are unable to assess whether any individual in this study will develop neurodegenerative disease in the future. However, we are following these individuals longitudinally to determine the association between mTBI-related reduction in white matter integrity, hippocampal volume, and neurodegenerative disease. Self-reported mTBI is another limitation of the current study. We did not have access to corroborating military medical records, which for many military personnel, were often lost or destroyed because of environmental conditions in the field (Benson et al., 2013). However, given that this is a limitation for most other studies of this nature, semi-structured interviews are the acknowledged standard for diagnosing remote mTBI. These findings may not generalize to ethnic groups other than Caucasian, non-Hispanic individuals or non-military samples and additional work is necessary to determine how hippocampal atrophy in military samples compares to civilian mTBI and neurological samples. Finally, we note that the findings from this study should be considered provisional until replication from additional samples is obtained.

Conclusions

In a large sample of recent war veterans, we examined the relationship between mTBI, BDNF genotype, and structural and functional brain metrics related to neurodegenerative disease. We provide evidence that rs1157659, which has previously been associated with progressive hippocampal atrophy, interacts with mTBI to predict reduced hippocampal volume. In addition, we report for the first time the association between this SNP, mTBI, and reduced PCC-mPFC functional connectivity of the midline core of the default mode network. Taken together, the results are the first to provide evidence that mTBI in combination with BDNF is associated with changes in brain metrics that overlap with pathological markers of neurodegenerative disease.

Fig. 1. Genotype by mTBI interaction on hippocampal volume.

(A) Sample sagittal slices of the hippocampus using automated construction by FreeSurfer. (B) Left adjusted hippocampal volume as a function of rs1157659 genotype and history of mild TBI. Volumes are adjusted for age, PTSD symptom severity, lifetime substance use diagnosis, premorbid IQ, and education. AA = major allele homozygotes; AG = heterozygotes; GG = minor allele homozygotes. *p < 0.05.