Cumulative Influence of Inflammatory Response Genetic Variation on Long-Term Neurobehavioral Outcomes after Pediatric Traumatic Brain Injury Relative to Orthopedic Injury: An Exploratory Polygenic Risk Score

Abstract

The addition of genetic factors to prognostic models of neurobehavioral recovery following pediatric traumatic brain injury (TBI) may account for unexplained heterogeneity in outcomes. The present study examined the cumulative influence of candidate genes involved in the inflammatory response on long-term neurobehavioral recovery in children with early childhood TBI relative to children with orthopedic injuries (OI). Participants were drawn from a prospective, longitudinal study evaluating outcomes of children who sustained TBI (n = 67) or OI (n = 68) between the ages of 3 and 7 years. Parents completed ratings of child executive function and behavior at an average of 6.8 years after injury. Exploratory unweighted and weighted polygenic risk scores (PRS) were constructed from single nucleotide polymorphisms (SNPs) across candidate inflammatory response genes (i.e., angiotensin converting enzyme [ACE], brain-derived neurotrophic factor [BDNF], interleukin-1 receptor antagonist [IL1RN], and 5’-ectonucleotidase [NT5E]) that showed nominal (p ≤ 0.20) associations with outcomes in the TBI group. Linear regression models tested the PRS × injury group (TBI vs. OI) interaction term and post-hoc analyses examined the effect of PRS within each injury group. Higher inflammatory response PRS were associated with more executive dysfunction and behavior problems in children with TBI but not in children with OI. The cumulative influence of inflammatory response genes as measured by PRS explained additional variance in long-term neurobehavioral outcomes, over and above well-established predictors and single candidate SNPs tested individually. The results suggest that some of the unexplained heterogeneity in long-term neurobehavioral outcomes following pediatric TBI may be attributable to a child's genetic predisposition to a greater or lesser inflammatory response to TBI.

Introduction

The unexplained heterogeneity in outcomes following pediatric traumatic brain injury (TBI) is one of the most critical barriers to the development of effective prognostic tools and therapeutic interventions.^1,2 TBI sustained before age 18 resulted in 2529 deaths, >23,000 hospitalizations, and >812,000 emergency department visits in the United States in 2014, making it one of the leading causes of morbidity and mortality in childhood.³ Although some children with TBI display good recovery, many others have chronic neurobehavioral impairments, especially in executive functioning, behavioral adjustment, and social skills, which hamper their long-term academic, occupational, and social functioning.^4–6 Current prognostic models, which include sociodemographic, injury, and family factors to predict children's recovery, explain only ∼35% of variance in outcomes.⁷ The addition of genetic factors to prognostic models may significantly decrease the unexplained variance in outcomes and advance the field of pediatric TBI toward precision medicine, to improve individual prognostication, predict response to rehabilitation, and identify novel targets for treatment development.

Systematic reviews of the growing literature examining the impact of genetic variation on clinical outcomes following TBI^8–11 highlight a range of candidate single nucleotide polymorphisms (SNPs) that may be associated with outcomes. The most common are apolipoprotein E (APOE)^11,12 and genes involved in expression of monoamine neurotransmitters, brain-derived neurotrophic factor, cytokines, and mitochondrial proteins.⁹ A recent application of a systems-biology-based approach identified two primary biologic processes over-represented with genetic variants that are implicated in clinical outcomes of TBI⁸: (1) response to injury, involving cell proliferation, cell death, inflammatory response, and cellular metabolism; and (2) neurocognitive and behavioral reserve, involving brain development, cognition, and behavior. Guided by a recent surge in research into the complex role of the immune response in neurobehavioral recovery from TBI^13–15 and emerging evidence relating inflammation to psychiatric conditions in individuals with and without TBI,^16,17 the present study examines the cumulative influence of candidate genes involved in the inflammatory response on long-term neurobehavioral recovery following TBI sustained in early childhood.

TBI triggers neuroinflammation through host-mediated central nervous system (CNS) and peripheral inflammatory responses.¹⁸ The neuroimmune response is largely mediated by activated microglia that facilitate inflammation by recruiting pro-inflammatory cytokines, chemokines, and cell-adhesion molecules to clear debris and dead neural tissue.^19,20 In adults with TBI, greater CNS or systemic inflammation during the acute and chronic phases of recovery predicts increased mortality²¹ and poorer global outcomes,^22–25 as well as depression,²⁶ suicidal ideation, and impulsive behavior.²⁷ Although children with TBI also show increases in cerebrospinal fluid (CSF) and serum/plasma inflammation following TBI, associations of inflammation with outcomes have been equivocal, likely because of the small sample sizes.^28–32

Several candidate genes have been previously studied in relation to clinical outcomes following TBI and identified as involved in the inflammatory response by our systems-biology based approach: angiotensin converting enzyme (ACE), adenosine A1 receptor subtype gene (ADORA1), brain-derived neurotrophic factor (BDNF), interleukin (IL)1β, IL1 receptor antagonist (RN), IL6, 5’-ectonucleotidase (NT5E), and tumor necrosis factor (TNF).⁸ The ACE gene plays a significant role in the physiology of blood vessels and the inflammatory process, including the synthesis of proinflammatory cytokines IL-6 and IL-8.³³ In adults with TBI, ACE SNPs are associated with subacute neuropsychological performance³⁴ and global outcome at 6 months post-injury.³⁵ The protein encoded by the ADORA1 gene is an adenosine receptor that belongs to the G-protein coupled receptor 1 family. The adenosinergic system, through effects at A₁, A_2a, and A₃ receptors, is involved in inhibitory neuromodulation, as well as the modulation of glial responses,³⁶ including those related to inflammation following TBI.³⁷ A study of adult TBI showed an association of ADORA1 SNPs with post-traumatic seizure risk.³⁸ One of the mechanisms by which the inflammatory response affects brain function following TBI may involve modulation of BDNF, including reduced BDNF gene expression and function resulting in a detrimental effect on neurogenesis and neural repair.^39,40 BDNF SNPs have shown associations with mortality^41,42 and cognitive function in adults with TBI.^43,44 The NT5E gene influences T-cell immunoactivation and inflammatory cell adhesion,⁴⁵ and is associated with shorter time to first seizure and an increased seizure rate 3 years after TBI in adults.⁴⁶ Genes coding for proinflammatory cytokines IL-6 and IL-1β) have shown associations with mortality,⁴⁷ global outcomes,⁴⁸ and post-traumatic seizure risk after adult TBI,⁴⁹ and a similar association is found between IL1RN and increased risk for cerebral hemorrhages.⁵⁰ To our knowledge, the only study of inflammatory response genes that included children with TBI examined 11 cytokine SNPs in 1096 individuals with TBI ranging in age from 0 to 93 years (mean age 37).⁵¹ Results showed a significant association between the TNF-α gene and global outcome at 6 months post-injury.

The present study aimed to build on existing research in several important ways. No previous published study has to our knowledge examined the influence of genes involved in the inflammatory response on TBI outcomes in an exclusively pediatric sample. In addition, few studies have examined the association of inflammatory response genes with finer-grained neurobehavioral outcomes or outcomes beyond 6 months post-injury. Additionally, previous studies have rarely included a non–brain-injured comparison group for examining the differential association of genetic variants with neurobehavioral recovery following injury to the brain. Without examination of the moderating effect of brain injury on the association between genes and outcomes, previous studies have been limited in their ability to differentiate genetic effects on neurobehavioral recovery after TBI from differences that reflect pre-morbid genetic influences on neurobehavioral function, as found in individuals without brain injuries.^52–54 We included a comparison group of children with orthopedic injuries (OI) and no indication of brain injury for this purpose. Finally, most prior genetic association studies in TBI have examined the individual effects of a single SNP or the individual effects of a small group of SNPs. Because of the complex nature of recovery after TBI, single variants are unlikely to account for significant variability in outcomes. Unbiased genome-wide association study (GWAS) methodology has not yet been employed in TBI because of the very large sample sizes required. Examination of the cumulative effect of a group of related variants simultaneously through the calculation of a polygenic risk score (PRS)^55,56 holds promise as a more powerful method for identifying possible genetic influences on TBI recovery.

The PRS approach tests the cumulative explanatory power of multiple genes on a phenotype.⁵⁵ In its most rigorous implementation, a PRS is constructed based on the most highly associated genetic markers identified by a GWAS first conducted in a “discovery sample,” and then the PRS is tested in an independent replication sample.^55,56 In the exploratory examination of complex phenotypes for which GWAS data may not be available (e.g., TBI), however, precedent exists for constructing PRS based on evidence from candidate gene association studies or groups of genes known to be enriched within biological systems or pathways of relevance to the phenotype. This more exploratory PRS approach has been applied in predicting sensation seeking,⁵⁷ delay discounting,⁵⁸ brain aging,⁵⁹ smoking relapse,⁶⁰ coronary artery disease,⁶¹ and in predicting cognitive recovery⁶² and mortality⁶³ following TBI in adults. Therefore, the primary aim of the present study was to examine the cumulative influence of multiple candidate genes involved in the inflammatory response on long-term neurobehavioral recovery following TBI sustained in early childhood. Building on results of our prior systematic literature review and systems-biology analysis,⁸ we created an exploratory inflammatory response PRS and tested the cumulative influence of the PRS on neurobehavioral recovery in children with TBI relative to children with OI. We hypothesized that variation within genes involved in the inflammatory response would be associated with long-term executive function and behavioral adjustment, and that these associations would be moderated by injury group, with more pronounced effects within the TBI relative to the OI group.

Methods

Participants

Participants were drawn from a prospective, longitudinal study evaluating outcomes of children who were hospitalized overnight for a mild to severe TBI or OI between the ages of 3 and 7 years.^64,65 Recruitment occurred at three tertiary care children's hospitals and one tertiary care, general hospital in Ohio. Participants in both injury groups (TBI and OI) completed a total of six research assessments across multiple visits, including a baseline assessment during the immediate post-acute period (0–3 months after injury); assessments at 6, 12, and 18 months and ∼3.5 years post-injury; and a final long-term follow-up assessment an average of 6.8 (±1.1) years post-injury. Outcome data presented herein are from the final long-term follow-up assessment.

Inclusion criteria were hospitalization overnight for traumatic injury (TBI or OI), no evidence of child abuse as the cause of the injury, no history of documented neurological problems or developmental delays pre-injury, and English as the primary language spoken in the home. TBI severity was determined using the lowest post-resuscitation Glasgow Coma Scale (GCS) score.⁶⁶ Mild TBI was defined as a GCS score ≥13 without abnormal neuroimaging. Moderate TBI was defined as a GCS score of 9–12 with or without abnormal neuroimaging or a higher GCS score with abnormal neuroimaging as defined by an intracranial or parenchymal injury or depressed skull fracture. Severe TBI was defined as a GCS score ≤8. The OI group included children who sustained a bone fracture (not including skull fractures), had an overnight stay in the hospital, and did not exhibit alterations in consciousness or other signs or symptoms of head trauma or brain injury. The study was approved by the institutional review boards at each of the participating medical centers, and informed consent was obtained from participating caregivers.

Outcome measures

Participants in both injury groups and their parents completed a range of outcome measures at each research assessment, including performance-based neuropsychological measures,^4,67 observational measures of the home environment,^65,68,69 parent–child interactions,^70–72 and classroom functioning,^5,69 and parent-report measures of the family environment^65,73,74 and child functioning.^64,75 In the present report, we selected parent-report measures of child behavioral adjustment and executive functioning as the primary measures of interest because these domains are among the most commonly affected by pediatric TBI^76–78 and are persistent^79,80 and associated with significant functional impairment.^4,81,82 In addition, these measures could be completed by parents remotely, thereby providing the largest sample size among the various outcomes measures collected.

Parents completed the age-appropriate forms of the Behavior Rating Inventory of Executive Function (BRIEF) and Child Behavior Checklist (CBCL⁸³) at each timepoint, with the baseline assessment based on the child's function prior to injury and used as a covariate in analyses to control for pre-morbid differences in child functioning. The BRIEF is a parent-report measure of child executive function as evident in everyday behavior.⁸⁴ We analyzed the age-standardized global executive composite T score (BRIEF GEC) to assess global executive function behaviors. Higher scores reflect poorer executive functioning. The CBCL is a parent-report measure of child behavioral adjustment and possesses high test–retest reliability and criterion-related validity. We analyzed the age- and sex-standardized Total Problems T score to assess child behavioral adjustment. Higher scores reflect poorer behavioral adjustment. Both outcome measures are National Institute of Neurological Disorders and Stroke (NINDS)-recommended common data elements for pediatric TBI and are well validated for the pediatric TBI population.⁸⁵

DNA Collection, Genotyping, and Quality Control

DNA was collected from saliva samples and purified using Oragene OG-500 self-collection tubes (DNA Genotek, Ottawa, Canada). Oragene allows the collection, stabilization, and long-term storage of DNA from saliva at ambient temperature. Salivary DNA is a valid and reliable alternative to blood DNA for high-throughput genotyping. The HumanExome v1.1 Bead Chip (Illumina, San Diego, CA) was used to perform genotyping using the Illumina iScan system. SNPs from the sex chromosome, mitochondria, and indels were excluded. Quality of SNP calls from the chip were also evaluated. SNPs that failed Hardy Weinberg Equilibrium (p < 0.0001) or had minor allele frequencies <10% were excluded. Thresholds for quality control for call rates at individual and SNP levels were 99% and 90%, respectively.

Candidate gene selection and SNPs available for analysis

Based on results of our prior systematic literature search,⁸ we identified candidate genes that met three criteria: (1) previously associated with TBI outcomes in clinical studies, (2) available on the HumanExome v1.1 Bead Chip, and (3) passed quality control. Because our interest was in inflammatory-related genes, we focused on genes associated with inflammatory-related biologic pathways based upon a systems-biology analysis.⁸ Inflammatory genes identified from the prior systematic literature search included ACE, ADORA1, BDNF, IL1β, IL1RN, IL6, NT5E, and TNF. Of these, ACE, ADORA1, BDNF, IL1RN, and NT5E were available for analysis on the HumanExome v1.1 Bead Chip. Combined Annotation Dependent Depletion (CADD) was used to annotate SNPs with respect to associated gene and type of variant (https://cadd.gs.washington.edu/).⁸⁶ SNPs located in intergenic regions and not associated with a specific gene according to CADD annotation were excluded prior to analysis. Three SNPs that were in linkage disequilibrium (LD) with r² > 0.9 were excluded from analysis. The exome chip comprised 542,585 variants initially, and 134,527 variants remained after quality control. Thirty-three SNPs located in the five inflammatory genes of interest were available for analysis (Table 1).

Table 1.

SNPs Investigated for Inclusion in PRS

Gene	SNP	Chrom	Pos	Ref allele	Alt allele	MAF	Consequence	CADD
ACE	rs4329	17	61563458	A	Ga	0.45	Intronic	5.19
ACE	rs4331	17	61564052	A	Ga	0.45	Synonymous	5.337
ACE	rs4343	17	61566031	Ga	A	0.48	Synonymous	5.011
ACE	rs4362	17	61573761	T	Ca	0.49	Synonymous	6.02
ACE	rs4611524	17	61591652	Ta	C	0.42	Intronic	12.12
ADORA1	rs17042888	2	113862173	G	Aa	0.10	Upstream	0.141
ADORA1	rs13382561	2	113863536	A	Ga	0.38	Upstream	7.067
ADORA1	rs11677140	2	113865808	A	Ca	0.29	Upstream	0.041
ADORA1	rs1688072	2	113869347	A	Ga	0.16	Intronic	3.568
BDNF	rs988712	11	27563382	G	Ta	0.21	Intronic	14.38
BDNF	rs7481311	11	27583129	Ta	C	0.25	Intronic	3.02
BDNF	rs10835201	11	27618265	A	Ga	0.26	Intronic	12.33
BDNF	rs10734394	11	27628412	Ga	A	0.27	Intronic	12.39
BDNF	rs6416056	11	27646745	Ga	A	0.30	Intronic	0.479
BDNF	rs4074134	11	27647285	C	Ta	0.21	Intronic	0.494
BDNF	rs925946	11	27667202	Ta	G	0.28	Intronic	3.007
BDNF	rs1519480	11	27675712	Ca	T	0.42	Downstream	2.23
BDNF	rs6265	11	27679916	C	Ta	0.17	Non-synonymous	2.099
BDNF	rs10767664	11	27725986	Ta	A	0.19	Upstream	4.935
BDNF	rs2030323	11	27728539	Aa	C	0.19	Intronic	1.449
BDNF	rs7934165	11	27731983	G	Aa	0.47	Intronic	2.987
IL1RN	rs17042888	2	113862173	G	Aa	0.10	Upstream	0.141
IL1RN	rs13382561	2	113863536	A	Ga	0.38	Upstream	7.067
IL1RN	rs11677140	2	113865808	A	Ca	0.29	Upstream	0.041
IL1RN	rs1688072	2	113869347	A	Ga	0.16	Intronic	3.568
IL1RN	rs315931	2	113869843	Ca	A	0.34	Intronic	11.81
IL1RN	rs315919	2	113876213	Ta	G	0.43	Intronic	0.089
IL1RN	rs3213448	2	113879297	G	Aa	0.16	Intronic	4.257
IL1RN	rs423904	2	113887262	C	Ta	0.24	Intronic	3.562
IL1RN	rs315952	2	113890304	T	Ca	0.31	Synonymous	0.388
NT5E	rs6942065	6	86168265	G	Aa	0.16	Intronic	18.54
NT5E	rs10944128	6	86180732	A	Ca	0.50	Intronic	21.4
NT5E	rs2229523	6	86199233	Aa	G	0.29	Non-synonymous	19.93
NT5E	rs2229524	6	86199243	T	Ca	0.11	Non-synonymous	8.24

^aMinor allele.

ACE, angiotensin converting enzyme; ADORA1, adenosine A1 receptor subtype gene; Alt, alternative; BDNF, brain-derived neurotrophic factor; CADD, Combined Annotation Dependent Depletion score;⁵⁷ Chrom, chromosome; IL1RN, interleukin-1 receptor antagonist; MAF, minor allele frequency; NT5E, 5’-ectonucleotidase; Pos, position; SNP, single nucleotide polymorphism; downstream, the DNA variant allele occurs outside the coding region of the gene, after the gene; these regions may affect gene regulation; intronic: the DNA variant allele occurs within a gene, but between the regions coding for protein; these regions may affect gene regulation; non-synonymous, the DNA variant allele alters the sequence of the coding region of a gene and alters the amino acid sequence; the protein structure will be altered; synonymous, the DNA variant allele alters the sequence of the coding region of a gene but the amino acid sequence remains the same; although the protein is not changed, variants in these regions could affect gene regulation; upstream: the DNA variant allele occurs outside the coding region of the gene, before the gene; these regions may affect gene regulation.

Statistical analyses

Statistical analyses were conducted using SAS 9.4 (SAS, Cary, NC). Prior to analyses, child outcome data were reviewed for plausibility according to the following rules to reduce the potential influence of outliers: (1) participants with changes in outcome scores between 6 months post-injury and the long-term time point that exceeded three standard deviations of change were excluded from respective analyses (n = 2 from BRIEF analysis; n = 0 from CBCL analysis); and (2) outcome scores of T > 90 (4 standard deviations above the normative mean) were winsorized to 90 (n = 5 from BRIEF analysis; n = 1 from CBCL analysis).

Cryptic relatedness was checked using Graphical Representation of Relationships (GRR)⁸⁷ (http://csg.sph.umich.edu/abecasis/GRR). Principal component analysis was employed to confirm European and African continental ancestry, which aligns with self-reported white and black race, using 200 validated ancestry informative markers and HapMap genotypic data from individuals of known ancestry as referent groups. Concordance with self-reported race was >95%. The first principal component was used as a covariate in regression models to adjust for racial variability.

To select SNPs for inclusion in our PRS, we employed general linear regression models to examine associations between available individual SNPs in each of the five inflammatory response genes and each outcome (executive function and behavioral adjustment) in the TBI group only. In all association tests, we used an additive genetic model in which major homozygotes were coded as 0, heterozygotes were coded as 1, and minor homozygotes were coded as 2. We chose a nominal threshold of p ≤ 0.20 for inclusion of SNPs into our PRS. A less stringent threshold for inclusion into PRS, relative to thresholds used in testing individual SNP associations, is considered appropriate to achieve a balance between the number of false-positive and true-positive risk alleles.^56,88 Consistent with prior PRS studies, we considered two PRS approaches: unweighted and weighted. Unweighted PRS for each outcome were computed as a summation of risk alleles meeting the inclusion threshold. Weighted PRS for each outcome were computed as a summation of risk alleles meeting the inclusion threshold multiplied by their respective β values extracted from general linear regression models. Finally, to test the differential effect of the inflammatory response PRS on neurobehavioral recovery in children with TBI relative to children with OI, we employed linear regression models using data for all participants for each outcome and tested the PRS × injury group (TBI vs. OI) interaction term. Covariates initially included in regression models and then trimmed if non-significant were continental ancestry principal component 1, the child's pre-morbid functioning on the measure of interest, and socioeconomic status (SES), defined by averaging sample z scores for maternal education and median income.⁶⁵ Post-hoc exploration of significant PRS × injury group interaction terms used linear regression to examine the effect of PRS over and above significant covariates within each injury group.

Results

Sample description

Of the 221 participants enrolled in the original study, 141 provided DNA samples. Participants with genetic data did not differ significantly from those without genetic data in demographic characteristics or measures of pre-morbid functioning (see Table S1). Of those with genetic data for whom covariates were also available, executive function was assessed in 135 participants and behavioral adjustment in 134. The TBI (n = 67) and OI (n = 68) groups did not differ significantly in race, sex, age at injury, age at assessment, or SES (Table 2). Children who sustained TBI had marginally poorer parent-reported pre-morbid behavioral and executive functioning than the OI group, although all means were solidly in the average range. Consistent with our prior reports from this study,^79,89,90 children with TBI had poorer executive functioning and behavioral adjustment than children with OI at long-term follow-up after controlling for pre-morbid ratings (Table 2).

Table 2.

Demographic and Outcome Characteristics by Injury Group

	OI n = 68	TBI n = 67	p
Gender, n (%)			0.432
Male	35 (47.3)	39 (52.7)
Female	33 (54.1)	28 (45.9)
Race, n (%)			0.406
White	52 (52.5)	47 (47.5)
Non-white	16 (44.4)	20 (55.6)
Age at injury in years, mean (SD)	5.12 (1.08)	5.11 (1.13)	0.940
zSES, mean (SD)	0.1 (0.94)	-0.11 (0.99)	0.223
CBCL Total Problems, pre-morbid, mean (SD)	45.07 (10.97)	50.33 (12.87)	0.012
CBCL Total Problems, long-term, mean (SD)	45.12 (10.87)	55.79 (12.96)	<0.001
BRIEF, pre-morbid, mean (SD)	47.19 (11.69)	51.72 (15.09)	0.053
BRIEF long-term, mean (SD)	48.88 (10.29)	59.18 (12.74)	<0.001

BRIEF, Behavior Rating Inventory of Executive Function; CBCL, Child Behavior Checklist; OI, orthopedic injury; SD, standard deviation; TBI, traumatic brain injury; zSES, socioeconomic status z. score

Injury characteristics are presented by injury subgroup in Table 3 for descriptive purposes. In the TBI group, 10 children had mild TBI, 41 had moderate TBI, and 16 had severe TBI. In addition to differences in GCS, the subgroups differed in their mean New Injury Severity Score⁹¹ (NISS), defined as the sum of the squares of the Abbreviated Injury Scale (AIS) scores for each child's three most severely injured body regions, and their mean “non-head injury” NISS, computed as the NISS minus the AIS for the head region. The severity of injuries to regions other than the head was higher in the OI than in the TBI groups by virtue of the inclusion/exclusion criteria. Neuroimaging abnormalities were classified based on radiological reports available for all but two children. The categorization of neuroimaging was based on previous research relating outcomes of TBI to the presence versus absence and type of brain lesions^67,92–94: no lesion; mild abnormalities, defined as a single subdural, subarachnoid, or epidural hemorrhage, or a single intraparenchymal lesion, contusion, or hemorrhage; moderate abnormalities, defined as multifocal lesions without diffuse abnormality (i.e., no edema, mass effect, swelling, midline shift, volume loss, or diffuse axonal injury); and severe abnormalities, defined as any diffuse abnormality, with or without focal lesions. The severe and moderate TBI groups were more likely to demonstrate neuroimaging abnormalities than the mild TBI group, which had none by definition. The severe and moderate TBI groups did not significantly differ in the proportion of neuroimaging abnormalities.

Table 3.

Injury Characteristics by Subgroup

	Group
	Severe TBI (n = 16)	Moderate TBI (n = 41)	Mild TBI (n = 10)	OI (n = 68)	p
Lowest GCS score, mean (SD)	4.00 (1.90)	13.46 (2.17)	13.60 (0.52)	—	<0.001
NISS total, mean (SD)	12.29 (9.08)	14.20 (7.26)	8.8 (6.58)	6.56 (2.83)	<0.001
NISS non-head-related, mean (SD)	1.33 (2.23)	2.18 (3.63)	2.30 (2.87)	6.56 (2.83)	<0.001

Neuroimaging abnormalities, n (%)a
Absent	7/15 (47%)	8/40 (20%)	10/10 (100%)	68/68 (100%)	<0.001
Mild	2/15 (13%)	8/40 (20%)	0/10 (0%)	0/68 (0%)
Moderate	1/15 (7%)	7/40 (18%)	0/10 (0%)	0/68 (0%)
Severe	5/15 (33%)	17/40 (43%)	0/10 (0%)	0/68 (0%)

Imaging results were missing from two participants (one severe TBI, one moderate TBI).

^aAbnormality was absent in the mild TBI and OI groups by definition.

GCS, Glasgow Coma Scale; NISS, New Injury Severity Score; OI, orthopedic injury; SD, standard deviation; TBI, traumatic brain injury.

PRS composition

General linear regression models in the TBI group identified 10 SNPs with p ≤ 0.2 for inclusion in the PRS for executive function across ACE, BDNF, IL1RN, and NT5E genes. Four SNPs had p ≤ 0.2 for inclusion in the PRS for behavioral adjustment across BDNF and IL1RN genes (Table 4). Retained covariates included the continental ancestry principal component 1 and pre-morbid executive function for the executive function models and continental ancestry principal component 1, SES, and pre-morbid behavioral adjustment for the behavioral adjustment models. These single candidate SNP models each explained between 27% and 38% of variance in outcomes.

Table 4.

SNPs Identified for Inclusion in PRS Based on Results of General Linear Regression Models in the TBI Group

Gene	SNPs	EF β	EF p	Adj R2	BA β	BA p	Adj R2
ACE	rs4343	-2.4	0.266	0.282	0.06	0.978	0.335
ACE	rs4362	2.6	0.187	0.288	0.13	0.946	0.335
ACE	rs4329	3.9	0.067	0.306	0.8	0.702	0.337
ACE	rs4611524	1.3	0.48	0.274	1.67	0.369	0.344
ADORA1	rs10920576	0.2	0.951	0.268	1.48	0.563	0.339
ADORA1	rs11590405	1.3	0.563	0.272	-0.74	0.732	0.337
ADORA1	rs17465037	-0.3	0.899	0.268	-0.18	0.930	0.335
ADORA1	rs6686206	2.3	0.39	0.277	2.18	0.407	0.343
ADORA1	rs6701725	-0.1	0.969	0.268	-1.42	0.492	0.340
BDNF	rs7934165	-3.6	0.057	0.309	-2.9	0.127	0.360
BDNF	rs10734394	1.2	0.544	0.272	-0.15	0.937	0.335
BDNF	rs1519480	0.5	0.799	0.269	-0.13	0.949	0.335
BDNF	rs925946	0.8	0.691	0.270	0.56	0.779	0.336
BDNF	rs10835201	0.2	0.93	0.268	1.05	0.61	0.338
BDNF	rs7481311	0.2	0.929	0.268	2.16	0.312	0.346
BDNF	rs988712	2.6	0.292	0.281	2.17	0.363	0.344
BDNF	rs6416056	4.5	0.053	0.311	2.48	0.275	0.348
BDNF	rs4074134	4.1	0.094	0.300	2.65	0.264	0.349
BDNF	rs2030323	4.1	0.103	0.298	3.27	0.176	0.355
IL1RN	rs315931	-3.2	0.09	0.301	-3.12	0.097	0.364
IL1RN	rs11677140	-1.7	0.402	0.276	-2	0.331	0.345
IL1RN	rs423904	-1.1	0.66	0.270	-1.25	0.606	0.338
IL1RN	rs315952	-0.3	0.907	0.268	-0.92	0.676	0.337
IL1RN	rs315919	-3.3	0.078	0.304	-0.82	0.663	0.337
IL1RN	rs1688072	0.9	0.702	0.270	-0.57	0.814	0.336
IL1RN	rs3213448	-3.9	0.155	0.291	0.06	0.984	0.335
IL1RN	rs13382561	2.1	0.249	0.283	0.33	0.858	0.336
IL1RN	rs17042888	-0.3	0.924	0.268	5.69	0.038	0.380
NT5E	rs6942065	-5.9	0.028	0.323	-2.65	0.34	0.345
NT5E	rs10944128	1.9	0.361	0.278	-0.46	0.822	0.336
NT5E	rs2229524	-0.9	0.774	0.269	0.05	0.988	0.335
NT5E	rs2229523	-0.1	0.961	0.268	0.27	0.907	0.335

SNPs meeting PRS inclusion threshold of p ≤ 0.2 in bold; positive β indicates that the minor allele is associated with poorer outcome; negative β indicates that the minor allele is associated with better outcome.

ACE, angiotensin converting enzyme; ADORA1, adenosine A1 receptor subtype gene; BA, behavioral adjustment; BDNF, brain-derived neurotrophic factor; EF, executive function; IL1RN, interleukin-1 receptor antagonist; NT5E, 5’-ectonucleotidase; PRS, polygenic risk scores; SNP, single nucleotide polymorphism; TBI, traumatic brain injury

Differential effect of the inflammatory response PRS on neurobehavioral recovery in TBI relative to OI

Linear regression models showed significant unweighted and weighted PRS × injury group interactions for both the executive function and behavioral adjustment models (Table 5; p = 0.03–0.004). The interaction terms explained 2–3% of variance over and above pre-morbid function, continental ancestry, and SES. Post-hoc linear regression models within each injury group revealed that the unweighted and weighted PRS were significantly associated with executive function and behavioral adjustment in the TBI group, accounting for 8–13% of variance over and above covariates, with a total of 44–46% of variance in outcomes explained by the full models (Table 6). In contrast, neither PRS was significantly associated with either outcome in the OI group (Table 7), suggesting a differential effect of the inflammatory PRS following TBI relative to OI. This finding is illustrated in Figure 1, showing increasing executive dysfunction and behavior problems with higher unweighted and weighted PRS in children with TBI in contrast to no association between PRS and outcome in children with OI.

Table 5.

Linear Regression Results Testing Inflammatory Response PRS × Injury Group Interaction

Dependent variable	PRS	Parameter	Estimate	p	Adj R2	Δ Adj R2
EF	Unweighted
		Pre-morbid EF	0.421	0
		Continental ancestry	-38.455	0.002
		uwPRS	0.723	0.012
		Injury group	-7.795	0	0.429
		uwPRS × injury group	-1.452	.008	0.456	0.027
	Weighted
		Pre-morbid EF	0.423	0
		Continental ancestry	-39.121	0.002
		wPRS	0.186	0.014
		Injury group	-7.874	0	0.428
		wPRS × injury group	-0.406	0.004	0.460	0.032
BA	Unweighted
		Pre-morbid BA	0.527	0
		Continental ancestry	-4.859	0.712
		zSES	-2.049	.040
		uwPRS	1.242	.047
		Injury group	-7.422	0	0.459
		uwPRS × injury group	-2.536	.038	0.473	0.014
	Weighted
		Pre-morbid BA	0.525	0
		Continental ancestry	-3.811	0.770
		zSES	-1.998	0.044
		wPRS	0.393	0.026
		Injury group	-7.400	0	0.463
		wPRS × injury group	-0.744	0.032	0.478	0.015

BA, behavioral adjustment; EF, executive functioning; injury group, traumatic brain injury versus orthopedic injury; PRS, polygenic risk score; uwPRS, unweighted polygenic risk score; wPRS, weighted polygenic risk score; zSES, socioeconomic status z score.

Table 6.

Linear Regression Results Testing Inflammatory Response PRS within the TBI Group

Dependent variable	PRS	Parameter	Estimate	p	Adj R2	Δ Adj R2
EF	Unweighted
		Pre-morbid EF	0.438	0
		Continental ancestry	-30.530	0.081
					0.291
		uwPRS	1.466	<0.001	0.409	0.118
	Weighted
		Premorbid EF	0.438	0
		Continental ancestry	-30.530	0.081
					0.291
		wPRS	0.406	<0.001	0.419	0.128
BA	Unweighted
		Pre-morbid BA	0.506	0
		Continental ancestry	1.847	0.919
		zSES	-3.071	0.039
					0.346
		uwPRS	2.574	0.005	0.417	0.071
	Weighted
		Pre-morbid BA	0.506	0
		Continental ancestry	1.847	0.919
		zSES	-3.071	0.039
					0.346
		wPRS	0.775	0.003	0.427	0.081

BA, behavioral adjustment; EF, executive functioning; PRS, polygenic risk score; TBI, traumatic brain injury; uwPRS, unweighted polygenic risk score; wPRS, weighted polygenic risk score; zSES, socioeconomic status z score.

Table 7.

Linear Regression Results Testing Inflammatory Response PRS within the OI Group

Dependent variable	PRS	Parameter	Estimate	p	Adj R2
EF	Unweighted
		Pre-morbid EF	0.449	0
		Continental ancestry	-28.657	0.093
					0.265
		uwPRS	-0.168	0.674	0.255
	Weighted
		Pre-morbid EF	0.449	0
		Continental ancestry	-28.657	0.093
					0.265
		wPRS	-0.059	0.568	0.257
BA	Unweighted
		Pre-morbid BA	0.567	0
		Continental ancestry	-9.630	0.633
		zSES	-0.512	0.713
					0.313
		uwPRS	-0.255	0.768	0.302
	Weighted
		Pre-morbid BA	0.567	0
		Continental ancestry	-9.630	0.633
		zSES	-0.512	0.713
					0.313
		wPRS	-0.015	0.950	0.301

BA, behavioral adjustment; EF, executive functioning; OI, orthopedic injury; PRS, polygenic risk score; uwPRS, unweighted polygenic risk score; wPRS, weighted polygenic risk score; zSES = socioeconomic status z score.

FIG. 1.

Association of unweighted (A and C) and weighted (B and D) inflammatory response polygenic risk scores (PRS) with long-term executive function (A and B) and behavioral adjustment (C and D) in children with traumatic brain injury (TBI) or orthopedic injury (OI). There are increasing executive dysfunction and behavior problems with increasing unweighted and weighted PRS in children with TBI but no association between PRS and outcome in children with OI.

Discussion

The primary aim of the present study was to examine the differential cumulative influence of candidate genes involved in the inflammatory response on long-term neurobehavioral recovery following TBI sustained in early childhood. We tested the influence of an exploratory inflammatory response PRS on neurobehavioral recovery in children with TBI relative to children with OI. We found that variation within genes involved in the inflammatory response was associated with long-term (∼ 6.8 years post-injury) executive function and behavioral adjustment, and that these associations were moderated by injury group (TBI vs. OI). A higher inflammatory response PRS was associated with more executive dysfunction and behavior problems in children with TBI but not in children with OI. The cumulative influence of inflammatory response genes examined using the PRS explained additional variance in long-term neurobehavioral outcomes following early childhood TBI, over and above well-established predictors and single candidate SNPs tested individually.

The results provide preliminary evidence that some of the unexplained heterogeneity in long-term neurobehavioral outcomes following pediatric TBI may be attributed to a child's genetic predisposition to a greater or lesser inflammatory response to TBI. The inflammatory response PRS accounted for an additional 7–13% of variance in neurobehavioral outcomes after accounting for the effects of continental ancestry, pre-morbid function, and SES, with a total explained variance of 44–46%, more than many previous models. Although the innate inflammatory response is crucial for clearing debris and damaged cells early following TBI, unregulated acute and chronic inflammation appear to be deleterious for recovery. Higher CNS or systemic inflammation during the acute and chronic phases of recovery is associated with heightened risk of depression,²⁶ suicidal ideation, and impulsive behavior,²⁷ as well as poorer global outcomes^22–25 and mortality²¹ in adults with TBI. Although elevated CSF and serum/plasma biomarkers of inflammation have been documented acutely following TBI in children, associations of inflammatory response protein biomarkers with outcomes have been equivocal, likely because of the small sample sizes, ranging from 14 to 29 children in previous studies.^28–32 Additional research into the inflammatory response in pediatric TBI is needed, especially given pre-clinical evidence suggesting a particularly robust inflammatory response in infants and children with severe brain injury relative to adults.^15,95,96

The inflammatory response candidate genes that showed nominal associations with long-term neurobehavioral outcomes in the TBI group and were therefore included in the PRS were ACE, BDNF, IL1RN, and NT5E. These results replicate reported associations between these genes and outcomes exclusively in adults following TBI,^{34,35,41–44} suggesting roles for genetic variation in the synthesis of proinflammatory cytokines, modulation of BDNF gene expression and function, IL1RN and IL1β levels and production, and T-cell immunoactivation and inflammatory cell adhesion in neurobehavioral recovery from TBI across the lifespan. The PRS explained more variance in long-term neurobehavioral outcomes than single candidate SNPs tested individually (PRS: 44–46% vs. individual candidate SNPs: 27–38%), confirming the polygenic nature of genetic influences on recovery from TBI and underscoring the value of the PRS approach for providing insights into the genetic architecture of TBI recovery.⁶²

Strengths and limitations

An important limitation of our PRS approach was that we restricted genes included in the PRS to those candidate genes that were previously associated with TBI outcomes in clinical studies,⁸ precluding the identification of previously unexamined genes that might prove influential. Without the very large sample sizes required for using GWAS to discover new genes important for TBI recovery, our next step will be to employ gene-enrichment analysis⁸ to identify previously unexamined genes for inclusion in an inflammatory response PRS. Gene-enrichment analysis is a pathway-based statistical genetics approach that can be used to identify novel genes over-represented (“enriched”) in biologic processes hypothesized to be associated with the outcome of interest. This approach may provide new insights into genetic influences on the inflammatory response and recovery following pediatric TBI.

A further limitation of the present study is that the PRS was constructed and tested in a single small sample of children with TBI. More rigorous approaches construct the PRS based on a discovery sample (usually a GWAS, which is not available in TBI) and externally validate it in an independent sample.^55,56 Therefore, our results are likely inflated, should be considered preliminary and exploratory, and are in need of replication using independent and larger samples. Confidence in the results is bolstered, however, by the finding that the PRS was differentially associated with outcomes in children with TBI but not OI, lending some preliminary evidence for discriminant validity. This finding suggests that genetic variability associated with inflammatory response genes is associated with executive functioning and behavioral adjustment only in the context of TBI rather than being associated with pre-bmorbid variability in these phenotypes in non–brain-injured children.⁹⁷ As the majority of existing studies of genetic influences on TBI outcomes have not included a non–brain-injured comparison group, our ability to demonstrate significant associations of the PRS in TBI but not OI is a significant strength.

Our sample included children with TBI ranging from mild to severe and with a variety of neuroimaging findings. Although our inclusion of children from a wide range of TBI severity may improve the generalizability of our results, genetic effects on recovery may be moderated by TBI severity or brain lesion type. Because of the small number of children within each TBI severity subgroup and lesion type, however, our study was insufficiently powered to examine the effect of these factors on the results. Along the same lines, orthopedic injuries in both injury groups also likely contribute to the inflammatory response; however, our finding that the PRS was only associated with outcomes in the TBI group suggests that it may be brain injury specific. Our measurement of child outcomes relied entirely on parent report, including reliance on retrospective recall of pre-morbid child functioning, which may be subject to bias. Additionally, the exome genotyping chip did not capture all common variation in our genes of interest, limiting the identification of the most promising variants for future studies. In addition to the inclusion of a non–brain-injured comparison group, further strengths of our study include the exclusively pediatric sample, examination of fine-grained neurobehavioral outcomes, and the characterization of genetic associations with long-term outcomes.

Significance and future directions

Although genetic variation may contribute to heterogeneity in neurobehavioral outcomes following pediatric TBI, future research also needs to consider and integrate moderators of gene expression, especially those relating to the child's environmental context. Decades of research have demonstrated a moderating effect of the family environment on neurobehavioral outcomes following pediatric TBI, such that the effect of TBI is buffered or exacerbated by environmental factors, including SES, parenting style and stress, and other psychosocial stressors.^65,98 In non-injured children, these same environmental factors have been shown to interact with genotype in producing various neurobehavioral phenotypes: the phenomenon known as “gene–environmental interaction.”^99–101 Most recently, epigenetic processes that regulate gene expression without altering the corresponding primary DNA sequence have been identified as a potential mechanism by which the social and biological environment of an individual impacts when and to what extent genes are expressed within each cell type. Of particular relevance to the present report, children who experience greater psychosocial adversity show altered epigenetic profiles in genes involved in the immune response, as well as in genes involved in the hypothalamic–pituitary–adrenal axis and neuronal development and neuroplasticity.^102–104 Thus, integrating the study of genetic and environmental influences on neurobehavioral recovery from pediatric TBI going forward will be essential to developing a comprehensive understanding of the various factors contributing to heterogeneity in outcomes. A handful of reports have tested gene–environmental interactions in pediatric TBI^105–107 and a study of epigenetic influences on neurobehavioral recovery from pediatric TBI is currently underway.¹⁰⁸

The continued examination of genetic factors as a source of unexplained heterogeneity in neurobehavioral recovery following pediatric TBI has potential clinical and research significance. Findings such as ours that identify genetic variants potentially associated with outcomes following TBI can provide a starting point for research into functional mechanisms that might eventually lead to the development of new treatment options. Clinically, the identification of genetic biomarkers that more accurately predict neurobehavioral outcomes could allow for improved prognostic accuracy, earlier identification of children at greatest risk for poor recovery, and early provision of targeted treatment. Pertinent to the present results, several multi-center clinical trials have failed to show any benefit of anti-inflammatory TBI therapies despite pre-clinical and single-center effects.¹⁵ The ability to characterize inflammatory phenotypes and stratify patients based on their genetic predisposition could allow for improved clinical trials targeting specific patients with personalized immunomodulatory treatments.¹⁵ Finally, the integration of genetic biomarkers with other biological and psychosocial predictors using novel machine learning approaches¹⁰⁹ holds promise for moving the field of TBI toward precision medicine.

Funding Information

This study was supported in part by the National Institute for Child Health and Human Development (NICHD) grant T32 HD040686, and the Rehabilitation Medicine Scientist Training Program (RMSTP) grants K12 HD001097-16, K01 HD097030, K23 HD074683, and R01 HD42729. Additional support was provided by the National Institute of Neurological Disorders and Stroke grant 1R01NS096053, and grants 8 UL1 TR000077 and KL2 TR001856 from the National Center for Advancing Translational Sciences (NCATS), the Society for Clinical Neuropsychology Early Career Pilot Award, the UPMC Children's Hospital of Pittsburgh Scientific Program, and the University of Pittsburgh/UPMC Rehabilitation Institute Pilot Award. Keith Yeates is supported by the Ronald and Irene Ward Chair in Pediatric Brain Injury, funded by the Alberta Children's Hospital Foundation. The content is solely the responsibility of the authors and does not necessarily represent the official views of the funders.

Author Disclosure Statement

No competing financial interests exist.

Supplementary Material

Supplementary Table S1