Genetic Liability for Internalizing Versus Externalizing Behavior Manifests in the Developing and Adult Hippocampus: Insight From a Meta-analysis of Transcriptional Profiling Studies in a Selectively Bred Rat Model

Abstract

BACKGROUND:

For more than 16 years, we have selectively bred rats for either high or low levels of exploratory activity within a novel environment. These bred high-responder (bHR) and bred low-responder (bLR) rats model temperamental extremes, exhibiting large differences in internalizing and externalizing behaviors relevant to mood and substance use disorders.

METHODS:

We characterized persistent differences in gene expression related to bHR/bLR phenotype across development and adulthood in the hippocampus, a region critical for emotional regulation, by meta-analyzing 8 transcriptional profiling datasets (microarray and RNA sequencing) spanning 43 generations of selective breeding (postnatal day 7: n = 22; postnatal day 14: n = 49; postnatal day 21: n = 21; adult: n = 46; all male). We cross-referenced expression differences with exome sequencing within our colony to pinpoint candidates likely to mediate the effect of selective breeding on behavioral phenotype. The results were compared with hippocampal profiling from other bred rat models.

RESULTS:

Genetic and transcriptional profiling results converged to implicate multiple candidate genes, including two previously associated with metabolism and mood: Trhr and Ucp2. Results also highlighted bHR/bLR functional differences in the hippocampus, including a network essential for neurodevelopmental programming, proliferation, and differentiation, centering on Bmp4 and Mki67. Finally, we observed differential expression related to microglial activation, which is important for synaptic pruning, including 2 genes within implicated chromosomal regions: C1qa and Mfge8.

CONCLUSIONS:

These candidate genes and functional pathways may direct bHR/bLR rats along divergent developmental trajectories and promote a widely different reactivity to the environment.

The strong pattern of comorbidity among psychiatric disorders is believed to be generated by a spectrum of latent liability (1), arising from a complex interplay of genetic risk and environmental factors such as stress and childhood adversity (1–3). At one end of this spectrum are internalizing disorders, which are associated with neuroticism, anxiety, and depression. At the other end are externalizing disorders, which are associated with risk taking and novelty seeking, as seen in mania, substance abuse, and impulse control disorders (1).

We model the genetic contributions underlying both extremes of this spectrum by selectively breeding rats that react differently to a novel environment. Bred high-responder (bHR) rats are highly exploratory with a disinhibited, novelty-seeking temperament, including hyperactivity, aggression, and drug seeking. Bred low-responder (bLR) rats are highly inhibited, exhibiting reduced locomotor activity and anxious and depressive-like behavior (4–13). These behavioral propensities are robust and stable, beginning early in development (14,15) similar to temperament in humans (16).

This highly differentiated phenotype makes bHR/bLR rats ideal for observing the developmental programming and adult manifestation of neurological factors underlying internalizing/externalizing tendencies (8,11,17). This study focused on the hippocampus, a region important for emotional regulation, behavioral inhibition (18–20), and environmental reactivity (18), including stress-related glucocorticoid release (21,22). We previously observed bHR/bLR differences in hippocampal volume, glucocorticoid receptor and growth factor expression, histone methylation, and cell proliferation and survival (5,9,14,23,24).

Our current study characterized hippocampal gene expression in bHR/bLR rats across development and adulthood using a meta-analysis of 8 transcriptional profiling datasets spanning 43 generations of selective breeding. Concurrently, we discovered chromosomal regions containing bHR/bLR segregating variants that are likely to contribute to exploratory locomotor phenotype (25). By comparing across these studies, we identified differentially expressed (DE) genes situated within implicated chromosomal regions and hippocampal functional pathways essential for mood and development (Figure 1). These genes are promising candidates for mediating the influence of selective breeding on behavioral phenotype.

Figure 1.

An overview of the experimental and analytical workflow used to identify top candidate genes for mediating the effects of selective breeding on bHR/bLR phenotype. Left: Many generations of selective breeding based on exploratory locomotor behavior drove segregation of genetic variants that contribute to internalizing and externalizing behavior within our bHR and bLR rats. The effect of these variants on behavior is mediated by alterations in gene expression and cellular function, which produce local changes in cell type balance and structure within brain regions responsible for affective behavior such as the hippocampus. Right: Our concurrent genetic study used exome sequencing to identify genetic variants that segregated bHR/bLR rats in our colony and then used a sampling of those variants to locate regions of the genome (QTL) that might contribute to exploratory locomotor activity (25). Our current study used meta-analyses of hippocampal transcriptional profiling studies to identify bHR/bLR DE genes, pathways, cell types, and networks in development and adulthood. In our results, we emphasize DE genes that were 1) consistently DE across multiple developmental stages, 2) central to differential expression pathways, cell types, and networks, and 3) located near genetic variants that segregated bHR/bLR rats in our colony and/or within QTL for exploratory locomotion. These genes are the top candidates for mediating the effects of selective breeding on bHR/bLR phenotype. bHR, bred high-responder; bLR, bred low-responder; DE, differentially expressed; GSEA, gene set enrichment analysis; PPI, protein–protein interaction; qPCR quantitative polymerase chain reaction; QTL, quantitative trait loci.

METHODS AND MATERIALS

Full methods for the individual experiments and analyses are provided in Supplement 1. The datasets have been released on GEO (Gene Expression Omnibus) (Table 1) and FigShare (https://doi.org/10.1101/774034). Analysis code (R Studio version 1.0.153; R version 3.2.2) is available at https://github.com/isabellie4/PhenotypeProject and https://github.com/hagenaue/HRLR_MetaAnalysisProject.

Table 1.

An Overview of the 8 Transcriptional Profiling Studies Included in Our Current Meta-analysis of Differential Gene Expression in the bHR and bLR Hippocampus at 4 Developmental Time Points: P7, P14, P21, and Adulthood

	MBNI_AffymetrixRae230_F4	MBNI_AffymetrixRgU34A_F6	MBNI_AffymetrixRae230_F15	MBNI_IlluminaRatRef12v1_F15	MBNI_RNASeq_F29	Alabama_Nimblegen_F34	MBNI_RNASeq_F37	MBNI_RNASeq_F43
Generation	4	6	15	15	29	34	37	43
Laboratory	MBNI	MBNI	MBNI	MBNI	MBNI	Alabama	MBNI	MBNI
Lead Scientist	Dr. John Stead	Dr. Sarah Clinton	Dr. Sarah Clinton	Dr. Sarah Clinton	Dr. Sarah Clinton	Dr. Sarah Clinton	Dr. Peter Blandino	Dr. Cigdem Aydin
Age	Adult	P7, P14, P21	P14	P14	P14, adult	P7, P14, P21, adult	Adult	Adult
n per Group	6	6	6	6	2	5	6	5
Platform	Affymetrix Rat Expression Set 230 A	Affymetrix Rat Genome U34A GeneChips	Affymetrix Rat Expression Set 230 A	Illumina RatRef-12 v1 BeadChip	RNA-Seq	NimbleGen Rat Gene Expression 12x135	RNA-Seq	RNA-Seq
Tissue Extraction	Whole hippocampus	Whole hippocampus	Whole hippocampus	Whole hippocampus	Whole hippocampus	Dorsal hippocampus tissue punch	Whole hippocampus	Whole hippocampus
Treatment	Basal	Basal	Basal	Basal	Basal	Basal	Basal	Vehicle injections
Behavioral Testing	Basal	Basal	Basal	Basal	Basal	Basal	Locomotor, anxiety	Social interaction after mild stress
GEO Data Release	GSE140596	GSE29552	GSE140595	GSE140594	GSE140597	GSE88874	GSE140598	GSE140287
Citation		Clinton et al. (14)				Cohen et al. (28)

See (14,28). Note that the MBNI_RNASeq_F37 dataset was collected from a subset of bHR and bLR rats included in the exome sequencing study (25).

bHR, bred high-responder; bLR, bred low-responder; GEO, Gene Expression Omnibus; MBNI, Molecular Behavioral Neuroscience Institute; P, postnatal day; RNA-Seq, RNA sequencing.

The bHR/bLR Rat Colony

All experiments were approved by the local university committee on the use and care of animals following the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

Selective Breeding.

We began selectively breeding bHR/bLR rats in the Molecular Behavioral Neuroscience Institute (MBNI) at the University of Michigan in 2003 [protocol: (11)]. Later, a second colony was begun at the University of Alabama at Birmingham using generation F30 bHR/bLR rats from MBNI. Our meta-analyses use datasets derived from male bHR/bLR rats spanning generations F4 to F43. We refer to these datasets according to their respective institution, transcriptional profiling platform, and generation (Table 1). MBNI_RNA-Seq_F37 also included a bHRxbLR cross called bred intermediate-responder rats.

Behavioral Testing.

For each generation, locomotor response to a novel environment was assessed between postnatal day 50 (P50) and P75 (11). For MBNI_RNASeq_F37, we measured anxiety-like behavior in adulthood using the percentage time spent in the open arms of an elevated plus maze (5-min test) [protocol: (26)]. For MBNI_RNASeq_F43, we measured social interaction in adulthood [protocol: (4)] after 15 minutes of exposure to the anxiogenic open arms of the elevated plus maze [protocol: (27)].

Hippocampal Gene Expression Analyses

Broad Overview of the Datasets.

Our meta-analyses included 8 datasets from bHR/bLR rats aged P7, P14, P21, and adult (Table 1). The rats were housed in standard conditions with minimal intervention besides behavioral testing or saline injections, and they were sacrificed by rapid decapitation without anesthesia. The whole hippocampus was dissected except in Alabama_Nimblegen_F34, which used dorsal hippocampal tissue punches (28). The extracted RNA was profiled using microarray (earlier generations: F4, F6, F15, and F34) or RNA sequencing (RNA-Seq) (later generations: F29, F37, and F43).

Broad Overview of the Data Preprocessing.

The preprocessing steps for each study varied with platform but involved common steps, including reannotation, normalization to reduce technical variation, and quality control. Microarray data were typically summarized into log(2)-transformed expression sets using robust multiarray average (29). Genelevel RNA-Seq read count summaries were converted to log(2) fragments per million. When applicable, transcript data were averaged by gene symbol to obtain a single expression value per sample per gene.

Meta-analyses.

Within each dataset, we calculated the effect size (Cohen’s d and associated variance) of bHR/bLR phenotype on the expression data for each gene within each age group. This output was aligned across datasets using official gene symbols. The meta-analysis for each age group was performed using rma.mv() (metafor) (30) and was corrected for false discovery rate (FDR) (Benjamini–Hochberg method; multtest) (31). Including generation as a covariate provided little additional insight (generation: all genes FDR > .30).

Overlap With Hippocampal DE Genes in Other Bred Rat Models.

We cross-referenced our top findings with published results (32–39) or reanalyzed data (40) (GEO Accession No. GSE20388) from 9 publications profiling hippocampal expression in other bred rat models targeting behavioral traits that we considered extremes on the internalizing/externalizing spectrum.

Overlap With Previously Identified Genetic Variants.

Our exome sequencing study identified bHR/bLR segregating variants (single nucleotide variants) within the F37 generation (n = 12/group) and used a sampling of those variants to pinpoint quantitative trait loci (QTLs) for exploratory locomotion using a bHRxbLR F2 intercross (25). In our current study, we identified all DE genes nearby (±1 MB) segregating variants and QTL peaks (logarithm of the odds score [LOD] > 3) and determined overlap with additional QTLs relevant to bHR/bLR behavioral phenotype from the Rat Genome Database (41) (accessed August 8, 2019, using the following keywords: anxiety, stress, and despair).

Positional Gene Enrichment Analysis.

To explore which genes might be either co-regulated or in linkage disequilibrium with a causal genetic variant, we evaluated the clustering of top bHR/bLR DE genes (nominal p < .01 in P14 and adult meta-analyses; duplicated Entrez IDs removed) within chromosomal regions using positional gene enrichment (http://silico.biotoul.fr/pge) (42).

Gene Set Enrichment Analysis and Protein–Protein Interaction Networks.

To elucidate functional trends within the largest sets of results (P14 and adult), we used gene set enrichment analysis (GSEA) [(43,44); fgsea (45)] and gene set matrix files (.gmt) containing standard gene ontology for rats (http://www.bioinformatics.org/go2msig/) (46) or customized hippocampal-specific gene sets (Table S1 in Supplement 2), including previously identified coexpression modules (47,48) and expression specific to hippocampal neuronal subtypes or subregions (Hipposeq) (49). We further explored top DE genes (adult meta-analysis FDR ˂ .10) and implicated hippocampal gene sets using predicted protein–protein interaction (PPI) networks (https://string-db.org) (50).

Cell Type Data Deconvolution.

To interrogate the relative cell type composition of our samples, we used BrainInABlender methodology (51). Data for genes previously identified as having cell type–specific expression were extracted, normalized, and averaged to produce a cell type index. For this analysis, we excluded the small MBNI_RNASeq_F29 dataset (n = 2/group). We meta-analyzed the effects of bHR/bLR phenotype on these cell type indices using aforementioned methods.

Quantitative Polymerase Chain Reaction Validation.

Hippocampal tissue from bHR and bLR male rats (n = 6/group) was collected at P14 (generation F55) and P90 (generation F51) (Figure S1 in Supplement 1). Following complementary DNA synthesis, Bmp4 (bone morphogenetic protein 4) was quantified with quantitative polymerase chain reaction (qPCR) and custom-designed primers using the Livak method (52) and Gapdh as reference. Group differences in ‒ΔC_q at each age were assessed using Welch’s two-sample t test (53).

RESULTS

Selective Breeding Amplifies the Propensity for Internalizing Versus Externalizing Behavior

The divergence of bHR/bLR exploratory activity in response to selective breeding happened rapidly (Figure 2A), implying oligogenic inheritance (25). This divergence was accompanied by an amplification of internalizing and externalizing tendencies (11,14,54,55). For example, in behavioral data accompanying our transcriptional profiling datasets, bLRs showed more anxiety-like behavior than bHRs (Figure 2B) and spent less time interacting socially following a stressful challenge (Figure 2C). Therefore, we expected that examining gene expression across bHR/bLR generations would reveal a convergence of effects within implicated chromosomal regions and pathways essential to affective behavior and reactivity to the environment.

Figure 2.

Selectively bred bHR and bLR rats model an extreme propensity for internalizing vs. externalizing behavior. (A) Over the course of 56 generations of selective breeding (F1–F56), the bHR rats (green) have developed increasingly elevated exploratory activity in a novel field (y-axis: average total locomotor score), whereas the bLR rats (red) have become less exploratory. These trends plateaued after F42, when our breeding strategy changed to deaccelerate divergence. Arrows indicate the generations during which hippocampal transcriptomic profiling datasets were collected along with a name indicating the respective laboratory, platform, and generation for each dataset. (B) bLR rats have been highly anxious since the initiation of our breeding colony. The example above is from the behavioral data accompanying the MBNI_RNASeq_F37 transcriptomic dataset showing bLRs spending a smaller percentage of time in the anxiogenic open arms of the elevated plus maze than cross-bred (bHRxbLR) bIR rats or bHR rats (effect of phenotype: F_2,15 = 6.72, p = 8.25 × 10⁻³; boxes = first quartile, median, and third quartile; whiskers = range or 1.5× the interquartile range; dot = outlier datapoint falling beyond the whiskers of the boxplot). (C) bLR rats are more reactive to stressors. This example is from the behavioral data accompanying the MBNI_RNASeq_F43 transcriptomic dataset showing bLR rats spending a smaller percentage of time interacting socially following exposure to a single mild stressor in the form of a brief exposure to the anxiogenic open arms of the EPM (F_1,8 = 5.86, p = .0418, boxplot follows the conventions of panel B). bHR, bred high-responder; bIR, bred intermediate-responder; bLR, bred low-responder; EPM, elevated plus maze.

Selective Breeding for Exploratory Locomotion Alters Hippocampal Gene Expression

Between generations F4 and F43, we conducted 8 exploratory studies transcriptionally profiling the hippocampus of bHR/bLR rats at 4 ages (P7, P14, P21, and adult). These small studies individually produced few reliable results (Figures S2 and S3 in Supplement 1). Nevertheless, a formal meta-analysis revealed multiple genes with consistent differential expression across generations (Figure 3 and Table S2 in Supplement 2). These results can be explored interactively at https://mbni.org/dashboard/huzefak/hrlrMetaAnalysis/.

Figure 3.

The top DE genes within the bHR/bLR hippocampal gene expression meta-analyses and their convergence with DE genes identified in other bred models targeting behavioral traits that represent extremes on the internalizing/externalizing spectrum. “Estimate” shows the estimated effect size (i.e., the difference in expression between bHR and bLR rats in units of standard deviation, where green/positive = higher expression in bHRs and red/negative = higher expression in bLRs), and “# of datasets” shows the number of datasets included in the meta-analysis for that gene. Bold gene symbols show genes identified as DE in previous hippocampal transcriptional profiling studies using other bred rat models targeting behavioral traits that could be considered extremes on the internalizing/externalizing spectrum, with the specific citations provided in the “Other Rat Models” column: (32) Wistar Kyoto vs. Fischer 344; (33) Wistar more immobile vs. Wistar less immobile; (37) congenitally helpless vs. helpless resistant; (39) Flinders sensitive vs. Sprague-Dawley; (40) Flinders sensitive vs. Flinders resistant. Not included due to lack of overlapping findings with our model: (34) Roman low avoidance vs. Roman high avoidance; (35) NIH-HS high anxiety vs. NIH-HS low anxiety; (38) Syracuse low avoidance vs. Syracuse high avoidance. bHR, bred high-responder; bLR, bred low-responder; DE, differentially expressed; FDR, false discovery rate; NIH-HS, National Institutes of Health heterogeneous stock; P-value, nominal p value.

Adulthood.

The effect of bHR/bLR phenotype on gene expression was significant for 74 of 16,269 genes (FDR ˂ .05) (Figure 3). About one third of these genes (25/74) had been previously identified as DE in the hippocampus of other bred rat models related to internalizing/externalizing behaviors (32–40) (Figure 3, Table S4 in Supplement 2, and Figure S11 in Supplement 1), and 11% (8/74) had been identified by more than one study, suggesting that our findings represent a rich collection of novel and previously identified candidates that could provide broad insight into internalizing/externalizing behavior.

The estimated effect sizes (βs) were often more extreme for genes exclusively represented in RNA-Seq data from later generations (Figure S4 in Supplement 1), but owing to smaller sample size, these effects were not overrepresented in the top results. To rule out bias due to platform, we ran a meta-analysis using only recent RNA-Seq data (F37/F43) and confirmed that similar DE genes and pathways were identified (Figure S5 in Supplement 1). In the following discussion, we emphasize candidate genes that have evidence that expression diverged during the earliest generations.

Development.

The developmental meta-analyses depended on data from earlier generations and produced less robust results. However, the top genes had consistent effects across age groups. Of 3257 genes included in the P7 meta-analysis, only Ncan (neurocan) was DE, with higher expression in bLRs than in bHRs since an early generation (Figure 4B, C) in a manner that nominally persisted at P14 (Figure 4D). Within the P14 and P21 meta-analyses, none of the effects survived multiple comparison correction (for 15,682 and 3257 genes, respectively). However, the top gene at P14, Bmp4, was consistently expressed at higher levels in bLRs than in bHRs within both the P14 (Figure 5A) and adult datasets since the F4 generation (Figure 5B). We confirmed differential expression of Bmp4 at these ages using qPCR (Figure 5C and Figure S1 in Supplement 1).

Figure 4.

The extracellular matrix constituent Ncan (neurocan) has elevated expression in bLR rats at P7. (A) A genetic variant on chromosome 16 nearby Ncan (˂100 kb) segregated bHR and bLR rats (Fisher’s exact test: p = 1.63 × 10⁻⁷). (B) Box-plots illustrating the effect of age and bHR/bLR phenotype on Ncan expression [log(2) signal] in 2 microarray studies (boxes = median and interquartile ranges; whiskers = range; red = bLR; green = bHR). The effect of phenotype is most obvious at an age when Ncan is elevated in development (P7). (C, D) Forest plots showing that Ncan was more expressed in bLRs than in bHRs (boxes = Cohen’s d from each study ± 95% CIs; Model = estimated effect size ± 95% CIs provided by the meta-analysis) in the P7 meta-analysis (β = 24.16, p = 1.38 × 10⁻⁷, FDR = 4.86 × 10⁻⁴) (C) and nominally in the P14 meta-analysis (β = 0.90, p = 1.01 × 10⁻², FDR = 7.24 × 10⁻¹) (D). bHR, bred high-responder; bLR, bred low-responder; CI, confidence interval; FDR, false discovery rate; P, postnatal day; SNV, single nucleotide variant.

Figure 5.

A regulator of proliferation and differentiation, Bmp4, is more highly expressed in bLR rats than in bHR rats at P14 and adulthood. (A, B) Two forest plots showing that Bmp4 appeared to be consistently elevated in bLR rats (boxes = Cohen’s d from each study ± 95% CIs; Model = estimated effect size ± 95% CIs provided by the meta-analysis) at P14 (β = –1.49, p = 9.40 × 10⁻⁶, FDR = 1.47 × 10⁻¹) (A) and adulthood (β = –1.04, p = 1.01 × 10⁻³, FDR = 9.38 × 10⁻²) (B). This direction of effect mirrors findings in the literature showing that blocking the expression of Bmp4 in mice reduces anxiety and depressive-like behavior (108). (C) Using qPCR, we confirmed that bLRs showed greater Bmp4 expression than bHRs at P14 and adulthood (P90) using hippocampal tissue from later generations (F51 and F55). Log(2) FC in Bmp4 expression was calculated with the Livak method [–ΔΔCq (52)] using Gapdh as the reference housekeeping gene and bLRs as the reference group (therefore, the bLR mean is set to 0 in all panels). **p ˂ .005. (Data release: https://doi.org/10.6084/m9.figshare.10321658; P14: log(2) FC = –3.74, t_5.60 = –6.10, p = .00115; P90: t_8.74 = –6.87, p = 8.44 3 10^-5.) Note that for both ages, the magnitude of effect in the qPCR data appears to be greater than what was observed in the meta-analysis, which may reflect differences in technology or generation. Similarly, in the qPCR data, the log(2) FC at P14 appears to be larger than the log(2) FC in adults in a manner that seems to replicate the larger effects observed in the P14 vs. adult meta-analysis, but these differences should be interpreted with caution owing to the technical factors that differed between batches. (D) Within the behavioral data accompanying the MBNI_RNASeq_F37 dataset, we found that Bmp4 showed a negative relationship with percentage time in the open arms (β = –0.034, R² = .28, p = .024) and a positive relationship with the number of fecal boli produced on the EPM (β = 0.32, R² = .29, p = .020; data not shown). bHR, bred high-responder; bIR, bred intermediate-responder; bLR, bred low-responder; CI, confidence interval; EPM, elevated plus maze; FC, fold change; FDR, false discovery rate; FPM, fragments per million; P, postnatal day; P-value, nominal p value; qPCR, quantitative polymerase chain reaction.

Many bHR/bLR DE Genes Are Located Within Implicated Chromosomal Regions

Our exome sequencing study identified bHR/bLR segregating genetic variants and then used a sampling of those variants to identify chromosomal regions that are likely to contribute to exploratory locomotor phenotype (QTL peaks). These implicated regions overlapped extensively with QTLs relevant to internalizing/externalizing behavior (41), including anxiety (56–61), stress response (62–64), and behavioral despair (65). In our current study, 68% of the DE genes in adulthood (FDR ˂ .05) were within ±1 MB of a bHR/bLR segregating variant, and 21% were within ±100 kB of a highly segregating variant (Bonferroni-corrected α = 5.00 × 10⁻⁵), a 4.7-times enrichment compared with non-DE genes (Figure 6A) (Fisher’s exact test: p = 3.50 × 10⁻⁶). DE genes were also 4.5 times more likely to be located within a QTL for exploratory locomotion (LOD > 4) (Figure 6B, C). These results fit our expectation that the influence of bHR/bLR segregating genetic variants on exploratory locomotion is at least partially mediated by effects on gene expression within the hippocampus.

Figure 6.

Many of the top DE genes are located near genetic variants that segregate bHR/bLR rats within QTLs for exploratory locomotor activity. (A) Our concurrent genetic study used exome sequencing to identify SNVs that segregated bHR/bLR rats in our colony (25). A high percentage of bHR/bLR DE genes (adult meta-analysis: FDR ˂ .05) were found within ±100, 250, or 500 kb or 1 MB of these segregating variants, using either traditional (Bonferroni-corrected p ˂ .05) or more stringent (Bonferroni-corrected p ˂ 5.00 × 10⁻⁵) criteria to define segregation. An asterisk (*) designates enrichment (Fisher’s exact test: p ˂ .0001) in comparison with the non-DE genes in our meta-analysis (FDR > .05). (B) Our concurrent genetic study used a sampling of bHR/bLR segregating variants to identify QTLs for exploratory locomotor activity in a novel field using a bHRxbLR F2 intercross (25). bHR/bLR DE genes (adult meta-analysis: FDR ˂ .05) were 4.5 times more likely to overlap (±1 MB) QTLs for exploratory locomotion than other genes included in our meta-analysis (Fisher’s exact test: p = .0035). (C) A table illustrating the top genes from our meta-analyses (FDR ˂ .10; bold + italic = FDR ˂ .05) that overlap (±1 MB) significant (LOD > 4) and putative (LOD > 3) QTLs for exploratory locomotion identified by our concurrent genetic study (25). Also depicted is overlap with QTLs identified in the Rat Genome Database (41) for the following behaviors relevant to the bHR/bLR phenotype: anxiety (56–61), stress-related responses (62–64), and behavioral despair (65). (D) The top chromosomal loci enriched for bHR/bLR DE genes overlap previously identified QTLs relevant to externalizing and internalizing behaviors. The top chromosomal loci enriched for bHR/bLR DE genes were identified using PGE analysis: location (full coordinates, defined to include the start coordinates for all genes with p ˂ .01 in the P14 or adult meta-analysis in the region), p value, FDR, and enrichment score. Also depicted is the overlap (±1 MB) of these enriched chromosomal loci with QTLs for exploratory locomotor activity (25) as well as with QTLs identified in the Rat Genome Database (41) for anxiety (56–59), stress-related responses (62–64,73–75), and behavioral despair (65). Enrichment Score: ratio of genes with p ˂ .01 out of all genes in the region. bHR, bred high-responder; bIR, bred intermediate-responder; bLR, bred low-responder; Chr, chromosome; DE, differentially expressed; FDR, false discovery rate; LOD, logarithm of the odds score; MB, megabase; PGE, positional gene enrichment; P-value, nominal p value; QTLs, quantitative trait loci; SNV, single nucleotide variant.

Positional Gene Enrichment Analysis Specifies Narrower Chromosomal Regions Contributing to bHR/bLR Phenotype

Positional gene enrichment identified 132 chromosomal regions with a significant enrichment (FDR ˂ .05) of DE genes within the P14 and adult meta-analyses. We focused on the top regions (FDR ˂ .001) (Figure 6D), many of which (10/13) could be identified using recent RNA-Seq data (F37/F43), ruling out bias toward regions overrepresented on older microarray platforms. These top loci were narrow chromosomal regions (measured in kilobases) but overlapped a high percentage of QTLs for exploratory activity [37.5%: 3/8 QTLs with LOD > 4 in (25)], encompassing three DE genes previously associated with brain function, development, and internalizing/externalizing behavior: Ucp2 (uncoupling protein 2), Trhr (thyrotropin-releasing hormone receptor), and C2cd3 (C2 calcium-dependent domain containing 3) (Figures 7 and 8) (66–72). These top loci also overlapped a surprising percentage of QTLs in the Rat Genome Database (41) for anxiety [13%: 6/45 QTLs (56–59)], stress-related responses [21%: 8/38 QTLs (62–64,73–75)], and behavioral despair [23%: 3/13 QTLs (65)]. Therefore, these enriched loci could contain genetic variants contributing to internalizing/externalizing aspects of the bHR/bLR behavioral phenotype beyond exploratory locomotion.

Figure 7.

A region on Chr 1 implicated in the bHR/bLR phenotype contains two genes important for brain function and development: Ucp2 and C2cd3. (A) Genetic variants on Chr 1 within Ucp2 and C2cd3 segregate bHR and bLR rats in our colony (Rnor5 coordinates, Fisher’s exact test: SNV 1_171712284: p = 1.66 × 10⁻⁹; SNV 1_171672484: p = 1.27 × 10⁻¹³). (B) Ucp2 and C2cd3 are located on Chr 1 within a QTL for exploratory locomotor activity. An example of the correlation between genetic variation in this region and behavior is illustrated using the sequencing results from a nearby SNV and exploratory locomotor activity measured in a bHR×bLR F2 intercross (n = 310; adjusted R² = .049, p = 2.00 × 10⁻⁴, FDR = .0048). (C) C2cd3 and Ucp2 were the top DE genes (FDR < .05) within a segment of Chr 1 enriched for DE genes and containing a QTL for exploratory locomotor activity (25). The table illustrates the DE genes within this region. Estimate shows the estimated effect size (green/positive = higher expression in bHRs). Bold = p < .05, and bold + italic = FDR < .05. (D) A forest plot showing that Ucp2 had higher expression in bHRs in 4 of the 5 adult datasets included in the adult meta-analysis (boxes = Cohen’s d from each study ± 95% CIs; Model = estimated effect size ± 95% CIs provided by the meta-analysis; effect of phenotype: β = 1.29, p = 1.27 × 10⁻⁴, FDR = 3.83 × 10⁻²). This direction of effect mirrors findings in the literature showing that Ucp2 knockout mice have higher anxiety-like behavior and lower locomotor activity as well as greater sensitivity to stress (66,67,69,70), much like our bLR rats. (E) A forest plot showing that C2cd3 had higher expression in bHRs in 3 adult datasets included in the adult meta-analysis (effect of phenotype: β = 2.06, p = 2.84 × 10⁻⁵, FDR = 1.73 × 10⁻²). (F) In the behavioral data accompanying the MBNI_RNASeq_F37 dataset, C2cd3 [units: log(2) FPM] showed a positive relationship with exploratory locomotor activity (β = 0.000109, R² = .26, p = 3.20 × 10⁻²). (G) In the behavioral data accompanying the MBNI_RNASeq_F37 dataset, Ucp2 showed a trend toward a positive relationship with percentage of time spent in the anxiogenic open arms of the EPM (β = 0.03, R² = .22, p = 5.16 × 10⁻²). bHR, bred high-responder; bIR, bred intermediate-responder; bLR, bred low-responder; Chr, chromosome; CI, confidence interval; DE, differentially expressed; EPM, elevated plus maze; FDR, false discovery rate; FPM, fragments per million; QTL, quantitative trait locus; SNV, single nucleotide variant.

Figure 8.

Trhr was the top gene within 2 hippocampal-specific gene sets and within a region of Chr 7 implicated in the bHR/bLR phenotype. (A) A genetic variant on Chr 7 nearby Trhr (<200 kb distant) segregates bHR and bLR rats in our colony (Fisher’s exact test: p = 6.20 × 10⁻¹⁴). (B) Trhr is located on Chr 7 within a QTL for exploratory locomotor activity. An example of the correlation between genetic variation in this region and behavior is illustrated using the sequencing results from an SNV nearby Trhr (discussed above) and exploratory locomotor activity measured in a bHR×bLR F2 intercross (n = 315, adjusted R² = .061, p = 5.79 × 10⁻⁵, FDR = .00178). (C) A forest plot showing that Trhr expression was consistently elevated in bHR rats since generation F4 (boxes = Cohen’s d from each study ± 95% CIs; Model = estimated effect size ± 95% CIs provided by the meta-analysis; β = 1.19, p = 2.16 × 10⁻⁴, FDR = 4.94 × 10⁻²). This direction of effect mirrors findings in the literature showing that Trhr knockout mice exhibit greater anxiety and depressive-like behavior (71). (D) Trhr was the strongest result within a segment of Chr 7 enriched for DE genes and overlapping a QTL for exploratory locomotor activity. The table illustrates the DE genes within this region (estimate = estimated effect size [green/positive = greater expression in bHRs]; bold = p < .05; bold + italic = FDR < .05). (E) Trhr was a leading gene in 2 of the top hippocampal-specific gene sets identified as enriched for bHR/bLR DE genes by GSEA (bold = p < .05 in GSEA results; bold + italic = FDR < .05 in GSEA results; NES: positive scores [green] indicate higher expression in bHRs and negative scores [red] indicate higher expression in bLRs). The top 10 leading edge genes for each gene set are shown (bold + italic = FDR < .05 in meta-analysis). These genes have large estimated effect sizes within the meta-analysis and help to drive the enrichment of effects within these gene sets. Regional marker gene sets use the following abbreviations: dg, dentate gyrus; v, ventral; d, dorsal; ca, cornu ammonis subregion. (F) In the behavioral data accompanying the MBNI_RNA-Seq_F37 dataset, Trhr [units: log(2) FPM] showed a trend toward a positive relationship with exploratory locomotor activity (β = 4.48 × 10⁻⁴, R² = .22, p = 5.08 × 10⁻²). (G) Trhr and its ligand, Trh, are hub genes within a hippocampal-specific coexpression network that is enriched for bHR-upregulated genes. Genes within this network with known protein–protein interactions are illustrated above (STRINGdb: confidence setting = .15 due to hippocampal coexpression already suggesting potential interaction). Many of these genes have documented associations with reward behavior. bHR, bred high-responder; bIR, bred intermediate-responder; bLR, bred low-responder; Chr, chromosome; CI, confidence interval; DE, differentially expressed; FDR, false discovery rate; FPM, fragments per million; GSEA, gene set enrichment analysis; NA, not applicable; NES, normalized enrichment score; QTL, quantitative trait locus; SNV, single nucleotide variant.

bHR/bLR Differential Expression Is Enriched Within Hippocampal Functional Pathways

bHR/bLR Phenotype Is Associated With Proliferation and Differentiation.

In total, 8 of 2761 functional ontology gene sets were enriched with differential expression within the P14 meta-analysis (FDR ˂ .05). All were upregulated in bLRs (Figure 9A, Table S3 in Supplement 2, and Figure S6 in Supplement 1) and predominantly related to neurogenesis, differentiation, and brain development. Within the adult meta-analysis, 2 of the 4 top gene sets enriched with differential expression (FDR ˂ .10) were related to proliferation and development but were upregulated in bHRs. This pattern was confirmed using recent RNA-Seq data (F37/F43) (Table S3 in Supplement 2), ruling out bias toward gene families overrepresented on microarray platforms.

Figure 9.

The top bHR vs. bLR DE results are enriched with genes related to cell proliferation, differentiation, and development, including the canonical Mki67. (A) A table of the top functional ontology gene sets identified as enriched for bHR/bLR DE genes by GSEA (bold = p < .05 in GSEA results; bold + italic = FDR < .05 in GSEA results; NES: positive scores [green] indicate higher expression in bHRs and negative scores [red] indicate higher expression in bLRs). The top 10 leading edge genes for each gene set are shown (bold + italic = FDR < .05 in meta-analysis). These genes have large estimated effect sizes within the meta-analysis and help to drive the enrichment of effects within these gene sets. (B) A PPI network constructed using the top genes from the adult meta-analysis (192 genes with FDR < .10; STRINGdb: confidence setting = .40) had a dominant subnetwork that included Bmp4 and Mki67 as hub genes. Many of these genes are related to cell proliferation and differentiation within the brain. (C) Two genetic variants on Chr 1 within Mki67 fully segregated bHR and bLR rats in our colony (Rnor5 coordinates, Fisher’s exact test: SNV 1_214939984: p = 2.02 × 10⁻¹¹; SNV 1_214940284: p = 2.02 × 10⁻¹¹). (D) A forest plot showing that Mki67 was consistently elevated in bLR rats in adulthood (boxes = Cohen’s d from each study ± 95% CIs; Model = estimated effect size ± 95% CIs provided by the meta-analysis; adult: β = 22.01, p = 4.03 × 10⁻⁵, FDR = 1.99 × 10⁻²). (E) Within the behavioral data accompanying the MBNI_RNA-Seq_F37 dataset, Mki67 [units = log(2) FPM] showed a negative relationship with locomotor score (β = 20.000249, R² = .41, p = .0042). bHR, bred high-responder; bLR, bred low-responder; Chr, chromosome; CI, confidence interval; DE, differentially expressed; FDR, false discovery rate; FPM, fragments per million; GSEA, gene set enrichment analysis; NES, normalized enrichment score; P, postnatal day; PPI, protein–protein interaction; SNV, single nucleotide variant.

A PPI network constructed using the top DE genes (adult meta-analysis; FDR ˂ .10; 192 genes) had a dominant subnetwork highlighting many of these same genes (Figure 9B), including hubs Bmp4 and the canonical Mki67 (marker of proliferation ki-67) (Figure 9C–E). Literature review confirmed the PPI interactions within this subnetwork and their role in proliferation and differentiation in the brain (76–88).

bHR/bLR Phenotype Is Associated With the Dentate Gyrus.

Similarly, when performing GSEA using 69 gene sets custom designed to reflect hippocampal-specific cell types and networks (Table S1 in Supplement 2), we observed an enrichment of differential expression related to the dentate gyrus (DG) (Figure 8E, Figure S7 in Supplement 1, and Table S3 in Supplement 2), the location of neural proliferation within the hippocampus. At P14, bLRs showed an upregulation of genes with enriched expression in the DG versus the cornu ammonis regions (49) (FDR ˂ .05). In adulthood, bHRs showed an upregulation of genes with enriched expression in the ventral DG versus dorsal DG (49) (FDR ˂ .05), including Trhr (FDR ˂ .05).

bHR/bLR Phenotype Is Associated With Hippocampal Coexpression Networks Related to Synaptic Signaling.

Coexpression modules can capture regionally important cell types and functions that remain undocumented in traditional ontology databases (89). We observed an enrichment of bHR/bLR effects within six previously identified hippocampal coexpression modules within the P14 meta-analysis (FDR ˂ .05) and five such modules within the adult meta-analysis (FDR ˂ .05) (Figure 8E, Figure S7 in Supplement 1, and Table S3 in Supplement 2), two of which included genes within implicated chromosomal regions.

The first was a large coexpression module (695 genes) previously identified in the mouse hippocampus (48) (named “lightcyan”), with elevated expression in bLRs relative to bHRs at P14 (FDR ˂ .05) and adulthood (FDR ˂ .10). One of the top DE genes in this module, Etv4 (ETS variant 4) (FDR ˂ .05; bHR > bLR), is a transcription factor required for proper hippocampal dendrite development (90) located within an implicated chromosomal region (Figure S8 in Supplement 1). A PPI network constructed using the DE genes in this module (n = 74; adult p ˂ .05) was enriched with genes related to cell projections, neurons, synapses, and cation binding (FDR ˂ .05).

The second was a small coexpression module (39 genes) previously identified in the mouse hippocampus (48) (named “sienna3”), with elevated expression in bHRs in adulthood (FDR ˂ .05). A PPI network constructed using all 39 genes in this module centered on Trhr and its ligand, Trh (thyrotropin-releasing hormone) (Figure 8G), and included many reward-related signaling molecules, including Cart (CART prepropeptide) (91), Oxt (oxytocin/neurophysin I prepropeptide) (92), and Drd1 (dopamine receptor D1a) (93,94).

bHR/bLR Phenotype Is Associated With Microglial- and Endothelial-Specific Gene Expression.

To interrogate possible bHR/bLR differences in hippocampal cell type composition, we performed a cell type deconvolution analysis focused on well-characterized nonneuronal categories. bLRs had greater microglial-specific expression than bHRs at P14 and adulthood (Figure 10A, C, D). At P14, bLRs also showed greater endothelial-specific expression (Figure 10B). These effects could reflect differences in cell type composition or activation state. Two of the top DE genes located within implicated chromosomal regions are regulators of microglial state; Mfge8 (milk fat globule-EGF factor 8) (Figure 10E–H) promotes alternative (M2) activation (95), and C1qa (complement component C1q A chain) (Figure 10I–L) promotes classical activation (96). To interrogate less well characterized hippocampal cell types, we compared our meta-analysis results to the new mousebrain.org database (97) and found that the top DE genes (FDR ˂ .10) were highly expressed in a variety of cell types, including neuronal subcategories (Figure S9 in Supplement 1), mirroring the diversity of hippocampal functions implicated in bHR/bLR phenotype.

Figure 10.

Microglial-related gene expression differentiates bHR and bLR rats. (A) A heatmap illustrating the effect of bHR/bLR phenotype on cell type–specific gene expression, which can reflect overall cell type balance (51) or activation (green = upregulated in bHRs; red = upregulated in bLRs; an asterisk (*) indicates p < .05). Note that only well-characterized nonneuronal cell type categories were included in this analysis. (B–D) Forest plots (boxes = Cohen’s d from each study ± 95% CIs; Model = estimated effect size ± 95% CIs provided by the meta-analysis) showing an upregulation in bLRs of endothelial-specific gene expression at P14 (β = –0.89, p = 5.75 × 10⁻³) (B) and microglial-specific gene expression at P14 (β = –0.69, p = 2.90 × 10⁻²) (C) and adulthood (β = –0.65, p = 4.47 × 10⁻²) (D). (E) A genetic variant on Chr 1 in Mfge8 segregates bHR and bLR rats in our colony (Fisher’s exact test: p = 7.53 × 10⁻⁸). Mfge8 promotes alternative (M2) activation of microglia. (F) A forest plot illustrating that Mfge8 is more highly expressed in bHRs than in bLRs in the adult meta-analysis in all five adult datasets (β = 1.50, p = 2.70 × 10⁻⁵, FDR = 1.73 × 10⁻²). (G) Mfge8 is located on Chr 1 within a QTL for exploratory locomotor activity. An example of the correlation between genetic variation in this region and behavior is illustrated using the sequencing results from a nearby SNV (Rnor5 coordinates 1_141117448) and exploratory locomotor activity measured in a bHRxbLR F2 intercross (n = 317, adjusted R² = .061, p = 1.81 × 10⁻⁵, FDR = .002). (H) Within the behavioral data accompanying the MBNI_RNASeq_F37 dataset, Mfge8 [units = log(2) FPM] showed a positive relationship with total locomotor score (β = 0.000146, R² = .62, p = 1.10 × 10⁻⁴) as well as the percentage time in the open arms of the EPM (β = 0.00603, R² = .28, p = 2.30 ×10⁻²). (I) A genetic variant on Chr 5 near C1qa (500 kb distant) segregates bHR and bLR rats in our colony (Rnor5 coordinates 5_159525078; Fisher’s exact test: p = 1.09 × 10⁻⁷). C1qa promotes classical activation of microglia via the complement cascade. (J) A forest plot illustrating that C1qa is more highly expressed in bLRs than in bHRs in all five datasets in the adult meta-analysis (β = –1.40, p = 6.57 × 10⁻⁵, FDR = 2.61 × 10⁻²). (K) C1qa is the top DE gene (FDR < .05) within a segment of Chr 5 enriched for DE genes. The table illustrates the DE genes within this region (estimate = estimated effect size [red/negative = higher expression in bLRs]; bold = p < .05; bold + italic = FDR < .05). (L) Within the behavioral data accompanying the MBNI_RNASeq_F37 dataset, C1qa showed a negative relationship with total locomotor score (C1qa: β = –5.17 × 10⁻⁴, R² = .29, p = 2.09 × 10⁻²). bHR, bred high-responder; bLR, bred low-responder; Chr, chromosome; CI, confidence interval; EPM, elevated plus maze; FDR, false discovery rate; FPM, fragments per million; P, postnatal day; QTL, quantitative trait locus; RBC, red blood cells; SNV, single nucleotide variant.

DISCUSSION

By selectively breeding rats for more than 16 years, we produced a robust genetic model of the co-occurrence of common internalizing and externalizing behaviors. Such large differences in behavior would be expected to be accompanied by similarly strong differences in gene expression in affective circuitry. By performing a formal meta-analysis across small exploratory datasets, we provide insight into bHR/bLR differences in hippocampal gene expression across development and adulthood. Furthermore, by cross-referencing these results with our concurrent genetic study (25) and previous studies in other bred rats (32–40), we pinpoint strong candidates for mediating the influence of selective breeding on hippocampal function and internalizing/externalizing behavior.

Transcriptional Profiling Converges With Genetic Results to Identify Several Strong Candidates for Contributing to bHR/bLR Behavioral Phenotype

Our exome sequencing study offered a glimpse at the genetic factors contributing to our selectively bred phenotypes (25). However, the implicated chromosomal regions encompass several hundred genes, making their specific effects on gene expression and function difficult to predict. By cross-referencing these findings with our differential expression results, we discovered multiple strong candidates.

Two of these candidates are a compelling proof of principle; Trhr and Ucp2 are located near bHR/bLR fully segregating genetic variants (25) within narrow chromosomal regions highly enriched for DE genes, overlapping QTLs for exploratory activity (25), anxiety (56–58), and stress response (62,63). Trhr was also the top gene within a bHR-upregulated gene set associated with the ventral DG (49), a region important for proliferation and emotional regulation (18), and a hub in a bHR-upregulated hippocampal network containing reward-related signaling molecules. Trhr and Ucp2 have been extensively linked to metabolism and internalizing/externalizing behavior (98–100). Knocking out Ucp2 produces a bLR-like phenotype: higher anxiety-like behavior, lower locomotor activity, and reduced stress resilience (66,67,69,70). Trhr is an important component of the hypothalamic-pituitary-thyroid axis and regulates anxiety and depressive-like behavior (68,71,101).

In our exome sequencing study, the variants associated with Trhr and Ucp2 explained a moderate portion of exploratory locomotor behavior [˂10%, w200 locomotor counts (25)]—a magnitude akin to the bHR/bLR difference in locomotor score present in the F1 generation. Altogether, this evidence suggests that bHR/bLR segregating genetic variants drive differential expression of Ucp2 and Trhr in a manner that could meaningfully contribute to bHR/bLR differences in hippocampal function and internalizing/externalizing behavior. Because other novel candidates were associated with variants explaining a similarly promising portion of exploratory behavior in our exome sequencing study, our results offer new avenues for investigation.

Genes Highlighted in the Developmental Meta-analyses Suggest That bHR/bLR Differences in Hippocampal Structure Arise Early in Development

The different propensity of bHR/bLR rats toward externalizing or internalizing behavior is evident at a young age (14,15), paralleling hippocampal development (14). Our meta-analyses encompassed three postnatal ages (P7, P14, and P21) to provide insight into this neurodevelopmental trajectory. These meta-analyses depended on data from earlier generations and older platforms, yet they identified two strong candidates with clear associations with internalizing/externalizing behavior and hippocampal development.

The top P7 result was Ncan, exhibiting a strikingly large effect size (bHR < bLR) as early as generation F6. Ncan is located adjacent to a bHR/bLR segregating variant (25), overlapping a QTL for despair-related behavior (65). As part of the extracellular matrix, Ncan is upregulated during early brain development (102,103) and modulates cell adhesion, migration, and growth factor binding (104). Ncan has been linked to bipolar disorder (102), and knocking out Ncan enhances locomotor activity, risk taking, hedonia, and amphetamine hypersensitivity (103). Therefore, lower levels of Ncan in early development in bHRs could promote externalizing behavior as well as divergent hippocampal development.

The top P14 result was Bmp4, which was elevated in bLRs since the earliest generations in a manner that appeared to persist into adulthood. As a regulator of development, Bmp4 is important for neural induction (105,106), but later it suppresses neurogenesis (107–111) and promotes other cell fates (88,105,112). Bmp4 was an important driver in bHR/bLR-enriched gene sets related to proliferation and differentiation at P14 and was central within a related PPI network constructed from top adult DE genes. In the hippocampus, Bmp signaling promotes dorsal cell type identity and is essential for DG formation (113), matching our results indicating bLR enrichment of DG-related gene expression at P14 and adulthood.

Blocking Bmp signaling produces a bHR-like phenotype, reducing anxiety, fear conditioning, and depressive-like behavior (108,113). Therefore, Bmp4 is a strong candidate for driving long-term hippocampal structural differences capable of producing stable divergence in temperament. However, Bmp4 was not located near a bHR/bLR segregating variant in our exome sequencing study (25), implying that impactful variation is located in a nearby noncoding region or within a gene upstream in Bmp4’s signaling pathway.

Functional Analyses Implicate Hippocampal Proliferation and Differentiation in bHR/bLR Phenotype

A prominent theme in our results were functions related to cell proliferation and differentiation. Indeed, Mki67 itself contained two bHR/bLR segregating genetic variants and was more highly expressed in bLRs in adulthood (FDR < .05), matching the upregulation in bLRs observed histologically in development (14) and maybe adulthood (9,114). These findings confirm that the relationship between internalizing behavior and cell proliferation in our model is not a general stunting of growth-related processes, contrasting with the neurotrophic model of stress-related mood disorders (115).

Many of our top DE genes were also important regulators of cell fate. Bmp4, Sox9 (SRY-box 9), Sox2 (SRY-box 2), Hes5 (Hes family BHLH transcription factor 5), Cd24 (CD24 molecule), and Tek (TEK receptor tyrosine kinase) regulate functions including the developmental progression of neural differentiation, gliogenesis, and endothelial proliferation (76,84,85,87,116–118). Their role in adulthood includes growth and plasticity in response to neural activity and injury (79,119–121). Therefore, these results could explain previous morphological findings indicating that cell differentiation progressed differently in the adult hippocampus in bHRs and bLRs under mild stress (9). Together, these findings raise the interesting possibility that differential expression within neurodevelopmental programming pathways could provide a general mechanism by which environmental stimuli, such as stress and drugs, produce divergent changes in hippocampal structure in bHR/bLR animals.

Functional Analyses Implicate Microglial Activation in bHR/bLR Phenotype

Microglial-specific gene expression was upregulated at both P14 and adulthood in bLRs, suggesting either an increased proportion of microglia cells or microglial activation. Several top candidates were important regulators of microglial function. bLRs had greater expression of C1qa than bHRs since the earliest generations. The C1q genes promote classical microglial activation (96) and mediate phagocytosis-driven synaptic pruning (96,122,123). In contrast, Mfge8 had greater expression in bHRs. Mfge8 is associated with reduced proinflammatory factors (124) as well as alternative (M2) microglial activation (95), playing an important role in phagocytosis (95,125,126). Ucp2 also has anti-inflammatory functions (66,67,69,127,128) and was more highly expressed in bHRs than in bLRs. Both Mfge8 and Ucp2 contain bHR/bLR segregating genetic variants within probable QTLs for exploratory activity (25), suggesting that genetic variation could contribute to their differential expression in the hippocampus.

Together, these results seem to fit proinflammatory theories of internalizing disorders (129). However, we found little evidence of bHR/bLR differences in traditional inflammatory markers, although protein or serological measures could reveal otherwise. Given our other findings, we speculate that bHR/bLR differences in microglial-related expression are likely tied to nonimmune microglial functions, including the regulation of neurogenesis, cell survival (130), and synaptic pruning (123,131) in response to neuronal activity (123). If true, microglial phagocytosis could serve as a multifaceted tool to tailor plasticity during development or in response to environmental stimuli such as stress and drugs of abuse.

Conclusions and Future Directions

By comparing exome sequencing findings with hippocampal differential expression patterns during development and adulthood across many generations of selective breeding in our bHR/bLR colony, we implicate a diverse and compelling array of genes whose effects may converge to promote internalizing versus externalizing behavior. Owing to our dependence on older platforms and exclusively male rats, we cannot claim to have identified all relevant candidates, nor have we highlighted all promising results in our text. Similarly, by combining results from diverse experimental designs, we may have decreased our statistical power while increasing the generalizability of our results. That said, our results converge with other bred rat models to identify promising candidate genes and implicate two functional pathways that could guide bHRs and bLRs along a divergent developmental trajectory, setting the stage for a widely different reactivity to the environment. These findings inspire new avenues of research (132–134), including cell type–specific morphological analyses and the manipulation of candidate pathways to assess relevance to internalizing/externalizing behavioral and neurological phenotypes.

KEY RESOURCES TABLE

Resource Type	Specific Reagent or Resource	Source or Reference	Identifiers	Additional Information
Add additional rows as needed for each resource type	Include species and sex when applicable.	Include name of manufacturer, company, repository, individual, or research lab. Include PMID or DOI for references; use “this paper” if new.	Include catalog numbers, stock numbers, database IDs or accession numbers, and/or RRIDs. RRIDs are highly encouraged; search for RRIDs at https://scicrunch.org/resources.	Include any additional information or notes if necessary.
Commercial Assay Or Kit	Affymetrix Rat Expression Set 230 A Microarray Chip	Affymetrix (now ThermoFisher, Waltham, MA)
Commercial Assay Or Kit	Illumina RatRef- 12v1 Beadchip	Illumina, San Diego, CA
Commercial Assay Or Kit	TRIzol reagent	Invitrogen, Calsbad, CA	Catalog #15596018
Commercial Assay Or Kit	RNeasy RNA purification columns	Qiagen (sold via ThermoFisher, Waltham, MA), Valencia, CA
Commercial Assay Or Kit	TruSeq library preparation kit	Illumina, San Diego, CA
Commercial Assay Or Kit	AllPrep DNA RNA miRNA Universal Kit 50	Qiagen, Valencia, CA	Catalog #: 80224
Commercial Assay Or Kit	Illumina TruSeq Stranded mRNA Library Preparation kit	Illumina, San Diego, CA	Catalog #s RS-122-2101, RS-122-2102
Commercial Assay Or Kit	Kapa’s library quantification kit for Illumina Sequencing platforms	Kapa Biosystems,Wilmington MA	Catalog # KK4835
Commercial Assay Or Kit	QIAshredder kit	Qiagen, Valencia, CA	Catalog #79654
Commercial Assay Or Kit	RNeasy Mini Kit	Qiagen, Valencia, CA	Catalog #74104
Commercial Assay Or Kit	iScript cDNA Synthesis Kit	Biorad, Hercules, CA	Catalog #1708891
Commercial Assay Or Kit	iQTM SYBR® Green Supermix	Biorad, Hercules, CA
Deposited Data; Public Database	MBNI_AffymetrixRae230_F4 Dataset	Gene Expression Omnibus, ref: this paper	GSE140596
Deposited Data; Public Database	MBNI_AffymetrixRgU34A_F6 Dataset	Gene Expression Omnibus, ref: PMID: 21864320	GSE29552
Deposited Data; Public Database	MBNI_AffymetrixRae230_F15 Dataset	Gene Expression Omnibus, ref: this paper	GSE140595
Deposited Data; Public Database	MBNI_IlluminaRatRef12v1_F15 Dataset	Gene Expression Omnibus, ref: this paper	GSE140594
Deposited Data; Public Database	MBNI_RNASeq_F29 Dataset	Gene Expression Omnibus, ref: this paper	GSE140597
Deposited Data; Public Database	Alabama_Nimblegen_F34 Dataset	Gene Expression Omnibus, ref: PMID: 25791846	GSE88874
Deposited Data; Public Database	MBNI_RNASeq_F37 Dataset	Gene Expression Omnibus, ref: this paper	GSE140598
Deposited Data; Public Database	MBNI_RNASeq_F43 Dataset	Gene Expression Omnibus, ref: this paper	GSE140287
Deposited Data; Public Database	Flinders Sensitive Line vs. Flinders Resistant Line Affymetrix Rat Expression Set 230 Microarray Dataset	Gene Expression Omnibus, ref: PMID: 20830301	GSE20388
Deposited Data; Public Database	bHR vs. bLR Bmp4 qPCR Dataset	FigShare, ref: this paper	https://doi.org/10.1101/774034
Deposited Data; Public Database	bHR vs. bLR Exome Sequencing Dataset	NCBI Bioproject, https://www.ncbi.nlm.nih.gov/bioproject/, ref: PMID: 31182603	PRJNA521139
Deposited Data; Public Database	Rat Genome Database	https://rgd.mcw.edu, ref: PMID: 25355511	accessed 08/08/2019, RRID:SCR_006444
Deposited Data; Public Database	GO2MSIG	http://www.bioinformatics.org/go2msig/, ref: PMID: 24884810	accessed 6/2017, RRID:SCR_018359
Deposited Data; Public Database	Hipposeq	https://hipposeq.janelia.org, ref: PMID: 27113915	accessed 05/11/2018, RRID:SCR_015730
Deposited Data; Public Database	string-db	https://string-db.org/cgi/input.pl, ref: PMID: 25352553	accessed 04/2018, RRID:SCR_005223
Deposited Data; Public Database	Interactive web dashboard for viewing bHR vs. bLR differential expression	https://mbni.org/dashboard/huzefak/hrlrMetaAnalysis/, ref: this paper
Deposited Data; Public Database	Custom chip definition files (.cdf)	http://nmg-r.bioinformatics.nl/NuGO_R.html, http://mbni.org/customcdf/22.0.0/entrezg.download/ , ref: PMID: 16284200	rae230arnentrezgcdf_19.0.0, rgu34arnentrezgcdr_19.0.0, rat2302rnentrezgcdf_22.0.0
Deposited Data; Public Database	R package org.Rn.eg.db	http://bioconductor.org/packages/org.Rn.eg.db/	version 3.4.1, RRID:SCR_018358
Deposited Data; Public Database	NCBI Genome Remapping Service	https://www.ncbi.nlm.nih.gov/genome/tools/remap	accessed 8/8/2019
Deposited Data; Public Database	mousebrain.org	mousebrain.org, ref: PMID: 30096314	accessed 04/09/2018, RRID:SCR_018356
Deposited Data; Public Database	Pubmed	https://www.ncbi.nlm.nih.gov/pubmed/
Organism/Strain	Selectively-bred High Responder rats, species: Rattus Norvegicus, sex: male	Laboratory of Dr. Huda Akil & Dr. Stanley Watson Jr., ref: PMID: 16502134, this paper
Organism/Strain	Selectively-bred Low Responder rats, species: Rattus Norvegicus, sex: male	Laboratory of Dr. Huda Akil & Dr. Stanley Watson Jr., ref: PMID: 16502134, this paper
Sequence-Based Reagent	Custom-designed qPCR primers targeting Bmp4 (Rattus Norvegicus)	Laboratory of Dr. Huda Akil & Dr. Stanley Watson Jr., ref: this paper	ACC# NM_012827.2; forward primer: 5’-CCCTGGTCAACTCCGTTAAT-3’, start = 1214; reverse primer: 5’-AACACCACCTTGTCGTACTC-3’, start = 1319
Sequence-Based Reagent	Custom-designed qPCR primers targeting Gapdh (Rattus Norvegicus)	Laboratory of Dr. Huda Akil & Dr. Stanley Watson Jr., ref: this paper	ACC# NM_017008.4; forward primer: 5’- GTTTGTGATGGGTGTGAACC-3’, start = 459; reverse primer: 5’-TCTTCTGAGTGGCAGTGATG-3’, start = 628
Software; Algorithm	Noldus Ethovision	Noldus, Leesburg, VA
Software; Algorithm	R Studio	CRAN: The Comprehensive R Archive Network, https://cran.r-project.org	v.1.0.153
Software; Algorithm	R	CRAN: The Comprehensive R Archive Network, https://cran.r-project.org	v.3.2.2
Software; Algorithm	All code used to analyze the genomics data in this paper	Github Repository: https://github.com/isabellie4/PhenotypeProject
Software; Algorithm	All code used to analyze the genomics data in this paper	Github Repository: https://github.com/hagenaue/bHRbLR_MetaAnalysisProject
Software; Algorithm	Robust multi-array average (RMA)	ref: PMID: 12582260
Software; Algorithm	R package affy	https://www.bioconductor.org/packages/release/bioc/html/affy.html, ref: PMID: 14960456	version 1.54.0, RRID:SCR_012835
Software; Algorithm	R package metafor	https://cran.r-project.org/web/packages/metafor/index.html, ref: DOI: 10.18637/jss.v036.i03	version 2.0-0, RRID:SCR_003450
Software; Algorithm	R package multtest	https://www.bioconductor.org/packages/release/bioc/html/multtest.html, ref: DOI: 10.1007/0-387-29362-0_15	version 2.32.0, RRID:SCR_001255
Software; Algorithm	Positional Gene Enrichment (PGE) Analysis	http://silico.biotoul.fr/pge/ , ref: PMID: 18346969	RRID:SCR_018360
Software; Algorithm	R package fgsea	https://bioconductor.org/packages/release/bioc/html/fgsea.html, ref: DOI: 10.1101/060012	version 1.2.1
Software; Algorithm	R package BrainInABlender	https://github.com/hagenaue/BrainInABlender, ref: PMID: 30016334	version 0.0.0.9000
Software; Algorithm	R package limma	https://bioconductor.org/packages/release/bioc/html/limma.html, ref: PMID: 25605792	version 3.28.21, RRID:SCR_010943
Software; Algorithm	R package preprocessCore	https://www.bioconductor.org/packages/release/bioc/html/preprocessCore.html, ref: S8	version 1.34.0
Software; Algorithm	Tophat2	https://ccb.jhu.edu/software/tophat/index.shtml, ref: PMID: 23618408	RRID:SCR_013035
Software; Algorithm	Bowtie2	http://bowtie-bio.sourceforge.net/bowtie2/index.shtml, ref: PMID: 22388286	RRID:SCR_005476
Software; Algorithm	Cufflinks	http://cole-trapnell-lab.github.io/cufflinks/, ref: PMID: 22383036	RRID:SCR_014597
Software; Algorithm	CuffDiff	http://cole-trapnell-lab.github.io/cufflinks/, ref: PMID: 22383036	RRID:SCR_001647
Software; Algorithm	Cuffmerge	https://www.genepattern.org/modules/docs/Cuffmerge, ref: PMID: 22383036	RRID:SCR_015688
Software; Algorithm	SubRead aligner	http://subread.sourceforge.net, ref: PMID: 23558742	RRID:SCR_009803
Software; Algorithm	featureCounts	http://subread.sourceforge.net, ref: PMID: 24227677	RRID:SCR_012919
Software; Algorithm	R package compute.es	https://cran.r-project.org/web/packages/compute.es/index.html	version 0.2-2
Software; Algorithm	R package GEOQuery	https://www.bioconductor.org/packages/release/bioc/html/GEOquery.html, ref: PMID: 17496320	version 2.42.0, RRID:SCR_000146
Other	The Genomic Services Laboratory at HudsonAlpha	(now called HudsonAlpha Discovery), Huntsville, AL, https://gsl.hudsonalpha.org/index
Other	University of Michigan DNA Sequencing Core	Ann Arbor, MI, https://seqcore.brcf.med.umich.edu