Genetic Architecture of Addiction‐Relevant Behaviors in Outbred Sprague–Dawley Rats Reveals Loci for Anxiety‐Like and Nociceptive Traits

ABSTRACT

Studies have shown that substance use liability is associated with novelty seeking, anxiety‐like behavior, and pain sensitivity. We examined whether common genetic variation in outbred Sprague–Dawley rats explained variation in behavioral measures from three assays with established links to substance use: locomotor response to a novel environment, elevated plus maze, and tail flick. We estimated single‐nucleotide polymorphism heritability and performed genome‐wide association analyses using permutation‐derived significance thresholds (N = 534–654 rats across traits). Heritability estimates ranged from 0.14 to 0.38 across 11 traits. Three independent loci were identified: chromosome 1 for elevated plus maze open‐arm behavior (α = 0.05), chromosome 14 for elevated plus maze immobility (α = 0.10), and chromosome 17 for tail flick latency (α = 0.05). Candidate genes included Slc18a2, Gfra1, and Pdzd8 (chromosome 1); Rel and Bcl11a (chromosome 14); and Eci2 and Eci3 (chromosome 17). We compared these loci with our genome‐wide association study of a F₂ intercross of selectively bred high‐ and low‐responder rats, originally derived from Sprague–Dawleys, that model individual differences in externalizing and internalizing behavior. The current loci are distinct from the ones identified in the bred lines. This difference likely reflects selection history in the high‐ and low‐responder F₂s, which focused on facets of exploratory locomotion, while loci for anxiety and pain sensitivity traits were identified in the outbreds. This highlights the benefit of using both outbred and selectively bred rats to probe causal variants contributing to individual differences in substance use liability. The current outbred findings implicate monoaminergic signaling, transcriptional control, and lipid metabolism as testable mechanisms for addiction‐relevant behaviors.

Genome‐wide association analyses in outbred Sprague–Dawleys identified loci for anxiety‐like and nociceptive behavior on chromosomes 1, 14, and 17, which are distinct from facets of exploratory locomotion significant in our GWAS of a F₂ intercross of selectively bred high‐ and low‐responder (to a novel environment) Sprague–Dawleys.

1. Introduction

Substance use disorders place a very large economic and population‐health burden. For example, the U.S. economic cost of opioid use disorder plus fatal opioid overdose was estimated at $1.02 trillion in 2017 [1]. These outcomes co‐vary with dimensional liabilities that cut across diagnoses, including both negative emotionality and anxiety and high reactivity to novelty and sensation seeking. Human genetic work shows substantial shared heritable liability between externalizing traits and substance use and related disorders [2] and links impulsivity facets to drug experimentation and use [3]. On the other end of the temperament spectrum, neuroticism, an internalizing trait, is also associated with several loci identified via genome‐wide association studies (GWAS) and strong positive genetic correlations with major depressive and anxiety disorders, which share co‐morbidities with substance use disorders [4]. Contemporary frameworks further organize these domains and validate them psychometrically in clinical samples [5]. These combined findings suggest that there are multiple paths to substance use disorders, arising from both externalizing phenotypes that can promote drug experimentation and internalizing phenotypes that can promote the use of drugs to cope with stress, anxiety, and mood dysregulation.

To interrogate these cross‐cutting behavioral liabilities mechanistically and genetically, we turned to rats as model organisms. Rats support rich ethology, pharmacology, and causal circuit manipulations, and outbred populations provide levels of linkage disequilibrium (LD) that are amenable to GWAS [6]. Sprague–Dawley rats are widely used in behavioral neuroscience; although they are not extensively used for genetic studies, they have been genetically characterized. Sprague–Dawley rats obtained from Charles River have shown diversity and LD features favorable for GWAS of complex behaviors [7].

We previously performed a GWAS on a F₂ intercross of our selectively bred high responder/low responder (bHR/bLR) rat lines, which were originally created using Sprague–Dawley founders [8]. The bHR/bLR lines differ in their high or low propensity to explore a novel environment and provide a good model for individual differences in temperament and the extremes in human behavioral liabilities. As they diverged with selective breeding for exploratory locomotion, the bHR/bLR lines came to represent the polar extremes of emotionality exhibited by the outbred Sprague–Dawley rats from which they are derived, thus modeling two distinct paths to substance use [9, 10, 11, 12, 13, 14]. However, given the specific focus of this selective breeding on exploratory locomotion in a novel environment, it is not expected that the bHR/bLR model would capture the whole range of factors that contribute to substance use liability. For example, sensitivity to pain plays a special role in opioid use disorders, and anxiety and negative affect may be especially relevant to the propensity for continued drug use and relapse, sometimes referred to as the “dark side of addiction” [15]. Thus, distinct genetic, neural, and behavioral features are likely to play a role in these differing components of substance use disorders and are likely to be differentially detectable based on the population being studied.

To approach this question broadly, we tested whether common genetic variation in commercially outbred Charles River Sprague–Dawley rats contributed to individual differences in response to novelty (novelty‐induced locomotion), approach‐avoidance anxiety (elevated plus maze, EPM), and noxious thermal nociception (tail flick test), as each of these traits are associated with addiction‐relevant behaviors. Specifically, novelty‐induced locomotion and anxiety‐like behavior are predictors of drug self‐administration and relapse‐like responding [12, 16, 17], and thermal nociception and pain states modulate opioid seeking and consumption [18, 19, 20]. The temperament trait harm avoidance in humans [21, 22, 23], which constitutes a higher tendency to feel fear and anxiety accompanied by an elevated emotional response to aversive stimuli, is linked to decreased mu‐opioid binding in brain. This reflects not only a biological association between temperament and pain sensitivity but also corresponds with decreased mu‐opioid receptor levels in our high‐anxiety bLR rats [24]. Recently, several genes found linked to either basal pain sensitivity or risk for chronic pain conditions [25] are also widely known to influence the development, maintenance, or severity of addiction [26, 27, 28, 29]. We have found differences in basal tail‐flick latency among outbred Sprague–Dawley rats behaviorally classified as high versus low responders to novelty [30] and have replicated this observation in our selectively bred HR/LR lines (unpublished finding). These models consistently exhibit different behavioral and biological responses to drugs of abuse, with HRs having a higher basal vulnerability to addiction‐like behavior. This higher nociceptive responsiveness (lower latency) of HRs in tail flick coupled with the finding that basal pain sensitivity prior to initial exposure to drug (i.e., independently of withdrawal‐induced hyperalgesia) predicts some aspects of addiction susceptibility [31] led us to hypothesize that the reflexive nociceptive response, as measured by tail flick test, may be associated with an addiction vulnerability phenotype. However, this phenotypic observation does not by itself imply a shared causal locus or mechanism linking novelty reactivity and thermal nociception.

Our goals were to identify allelic associations with these traits across independent Sprague–Dawley‐derived populations and to discover additional variants not captured by selective breeding. To this end, we performed GWAS in outbred Sprague–Dawley rats and situated these analyses alongside prior mapping in an F₂ intercross of bHR/bLR rats in the same broader background, which identified loci for novelty‐induced exploratory locomotion and anxiety‐like traits [8, 32]. Our findings shed light on genetic differences associated with comparable exploratory locomotion and anxiety‐like behaviors in outbred Sprague–Dawley rats vs. selectively bred rats on the same background.

2. Methods

2.1. Phenotyping

Sprague–Dawley outbred rats were purchased from Charles River Laboratories, Raleigh, North Carolina, USA. Non‐cross‐fostered, non‐littermates arrived at the Michigan Neuroscience Institute, University of Michigan, at 10 weeks of age, approximately 70 days old. They were housed two per cage, with males and females housed in separate rooms. Table S1 includes demographic information about the 654 rats used in this study. Animals were provided ad lib water and standard rat chow (5LOD, PicoLab Laboratory Rodent Diet, LabDiet) in a temperature (70°F–78°F) and humidity (30%–50%)‐controlled room with a 12/12 light/dark cycle (lights on at 06:00, lights off at 18:00). Behavioral testing took place between 07:30–13:00 (Zeitgeber Time (ZT) ZT1.5 to ZT7). All procedures were conducted in accordance with the guidelines outlined in the National Institutes of Health Guide for the Care and Use of Animals and were approved by the Institutional Animal Care and Use Committee at the University of Michigan. The IACUC protocol number was PRO00008487.

2.1.1. Testing Schedule

After a one‐week acclimation period, rats completed locomotor activity testing during week 2 and a 5‐min elevated plus maze (EPM) session during week three. During week 4, a subset of animals received tail‐flick testing. Animals were euthanized for tissue collection (to obtain DNA for genotyping) during week 5.

2.1.2. Locomotor Response to a Novel Environment

Locomotor response to a novel environment (hereafter, “locomotor”) was measured by placing each rat individually into a clean cage (23 cm × 43 cm) located in a different room with novel cues. The number of beam breaks (counts) was measured over 60 min for both lateral and rearing movements, defined as the lateral and rearing locomotor scores, respectively. The cumulative lateral and rearing locomotor scores over the 60‐min test were combined to generate the total locomotor score. More details on the testing apparatus can be found in Stead et al. [13].

2.1.3. Elevated Plus Maze

Each rat was tested for 5 min on a standard EPM with two open arms and two closed arms. Video tracking quantified time allocation and movement (Ethovision, Noldus Information Technology). Details of the procedure are described in [8]. We analyzed the following EPM traits: time in the open arms in seconds (EPM open arm time), number of entries into the open arms (EPM open arm entries), percent time in open arms across 5 min expressed as 0%–100% (EPM percent time in open arms), total distance traveled in centimeters (EPM distance traveled), mean velocity in centimeters per second (EPM velocity), and immobile time in seconds calculated as 300 minus moving time based on a center point criterion (EPM time immobile). The EPM open arm time ratio is the percent time in the open arms divided by the time in the open arms plus the closed arms. This ratio takes into account potential differences in the amount of time that animals spend in the center of the EPM. If center time is similar among animals, then the EPM open arm time ratio is comparable to the EPM percent time in open arms.

2.1.4. Tail Flick

Tail flick latency was recorded in week 4. To assess individual differences in baseline pain thresholds to a noxious thermal stimulus, we used a modified tail flick (thermal nociception) test. Rats were gently restrained, the distal 1.5 cm of the tail was exposed to thermal nociceptive input and the latency to flick out of the beam was recorded, with a maximum cutoff time of 15 s to prevent tissue damage. While we calibrated the test to produce an average latency of approximately 8 s, we reduced the thermal intensity to increase the range of responses produced, allowing for better resolution of individual differences. Under lower thermal intensities, high reactivity rats demonstrate higher sensitivity to thermal stimulation [33], a difference that disappears at higher intensity stimulation. Each animal underwent two trials, separated by 90–120 s, but due to inconsistent appearance of acute endogenous analgesia, we chose to only use the first trial in the analysis.

2.2. Statistical Analysis

We first quantile normalized each trait. As a sensitivity analysis for tail flick latency, we repeated the tail‐flick GWAS after excluding animals at the 15 s maximum cutoff time before quantile normalization. We then tested a set of candidate covariates that reflected the study design, including sex, age at the relevant behavioral test, test batch and apparatus identifiers for locomotor activity, experimenter and within‐day test order for EPM and tail flick, and colony and housing room. For each trait, covariates that individually accounted for more than 2% of the phenotypic variance were retained, and the resulting covariate‐adjusted values were used for all downstream analyses. Pairwise phenotypic correlations were estimated on the covariate‐adjusted traits using Spearman's ρ with two‐sided p‐values and all available data for each trait pair (pairwise complete observations); p ≤ 0.05 was considered significant. Sex was included as a candidate covariate for all traits and was retained (and regressed out) for several EPM measures but not for locomotor or tail flick traits; the retained covariate set for each trait is summarized in Table S2.

2.3. Genotyping and Variant Processing

2.3.1. DNA Extraction, Sequencing, and Alignment

High molecular weight (HMW) genomic DNA from 654 outbred Sprague–Dawley spleen samples was isolated via the MagAttract HMW DNA Kit (Qiagen) at the University of Michigan and transferred to the University of California San Diego for further processing. All DNA samples were sequenced in low coverage using the Twist 96‐Plex Library Preparation Kit (Twist Bioscience). Reads were demultiplexed with fgbio (http://fulcrumgenomics.github.io/fgbio/), trimmed with cutadapt [34], aligned to the GRCr8 reference genome with BWA‐MEM [35], and duplicate reads were removed with the Picard Toolkit [36]. Mean coverage was approximately 0.312×. In addition, 8 of these 654 rats were deeply sequenced with KAPA HyperPre (Roche) followed by Illumina PE150 (Illumina) to approximately 19×; these eight outbred Sprague–Dawley samples formed a “truth set” to assess genotyping accuracy. Across these eight controls, the discordance between the final genotype data and the deep sequencing data was 0.99%, indicating high genotyping accuracy.

2.3.2. Reference Panel for Imputation

We used inbred strains from the hybrid rat diversity panel (HRDP) as characterized in De Jong et al. [37], to build the reference panel for STITCH‐assisted imputation [38]. Sequencing was available for 181 strains, of which 153 with the greatest depth were retained for the reference panel. Variants and genotypes for the reference panel were called with bcftools [39]. Genotype posteriors below 0.99 were set to missing, only biallelic single‐nucleotide polymorphisms (SNPs) with QUAL ≥ 20 and genotype missing rate ≤ 0.10 were retained. The resulting reference panel was then imputed and phased with Beagle 4.1 (−gtgl option) [40].

2.3.3. Genotype Calling and Imputation

Genotypes for the 654 Sprague–Dawley rats were inferred with STITCH (reference‐panel assisted). Sites supplied to STITCH included all loci present in the HRDP‐derived reference panel and in the eight deeply sequenced Sprague–Dawley controls. We set K = 16 mosaic haplotypes and n iterations = 3. STITCH labeled genotypes with posterior probability < 0.90 were set as missing; missing genotypes were subsequently imputed with Beagle 4.1.

2.3.4. Variant‐Level Quality Control

We began with 14,874,570 autosomal SNPs from the Beagle‐imputed dataset. Variant filters were applied jointly on per‐SNP genotype missingness (F‐MISS), minor allele frequency (MAF), and Hardy–Weinberg equilibrium (HWE). We applied a slightly higher missingness threshold at loci that were present in either the HRDP‐derived reference panel or the eight deeply sequenced Sprague–Dawley controls (STITCH “observed” sites; F‐MISS ≤ 0.07) and a stricter threshold at loci supported only by imputation (imputed‐only sites; F‐MISS ≤ 0.04). In total, 1,643,596 autosomal SNPs passed all criteria and were retained for genetic analyses, and 13,230,974 SNPs were excluded by at least one criterion. Of the 9,088,280 SNPs with MAF < 0.05, 8,003,804 (88.1%) were monomorphic (MAF = 0) in this dataset. The distribution of variant failures across MAF, HWE, and missingness filters is summarized in Table S3.

2.4. Genetic Analysis

2.4.1. Genetic Relatedness, SNP Heritability, and Genetic Correlation

We computed a genomic relatedness matrix (GRM) from the filtered autosomal SNP set using GCTA (−make‐grm). SNP‐based heritability for each trait was estimated via REML in GCTA. Pairwise genetic correlations (r _g) were estimated using bivariate GREML. We performed exploratory bivariate GREML and, because many r _g estimates were boundary‐constrained or had large standard errors, we did not report r _g values [41].

2.4.2. GWAS And Genome‐Wide Thresholds

GWAS used GCTA linear mixed models [41] with a leave‐one‐chromosome‐out GRM (LOCO) to reduce proximal contamination [42]. Genome‐wide significance thresholds were derived using permutation tests. The resulting thresholds were −log₁₀ P = 5.98 (α = 0.05) and −log₁₀ P = 5.60 (α = 0.10). Because all traits were quantile normalized, the same thresholds were used for all traits.

2.4.3. QTL Calling, Conditional Analysis, and Variant Annotation

For each locus, we defined the index SNP as the SNP with the smallest p‐value (largest −log₁₀ P) within the region. We called a QTL when at least one SNP (the index SNP) exceeded the permutation‐derived genome‐wide threshold and a second SNP within 0.5 Mb had a p‐value within 2 − log₁₀ P units of the index SNP. To identify additional independent loci on the same chromosome, we included the lead SNP(s) as covariates and iterated chromosome‐specific scans until no further SNP exceeded the threshold. QTL intervals were defined as all markers in LD r ² ≥ 0.6 with the peak SNP [43]. All variants within each QTL interval were annotated using SnpEff [44]. As supportive annotation, we queried gene‐level hippocampal cis‐eQTL results from the bHR/bLR F₂ intercross described by Hebda‐Bauer et al. [32] for genes within the outbred GWAS intervals. Because the present GWAS used GRCr8 whereas the bHR/bLR F₂ dataset used RN6, this comparison was restricted to gene‐level lookups.

2.4.4. Cross‐Study Coordinate Mapping and Replication Queries

The bHR/bLR F₂ analyses used the Rnor_6.0 (RN6) reference, whereas the outbred Sprague–Dawley GWAS in this study used GRCr8. For phenotype‐matched comparisons, we converted QTL peak SNPs and their LD‐defined intervals (r ² ≥ 0.6) between builds using UCSC liftOver [45]: outbred GRCr8 loci were lifted to RN6 for querying in the F₂ summary statistics, and F₂ RN6 loci were lifted to GRCr8 for querying in the outbred summary statistics. For each lifted locus we recorded the −log₁₀ P at the exact lifted top SNP when present and the maximum −log₁₀ P among all variants within the lifted LD‐defined QTL interval. Because this procedure searches many correlated variants inside each interval, it is anti‐conservative. Therefore, we treat these lookups as descriptive and do not regard them as confirmatory replication. We also ran an exploratory cross‐phenotype interval‐overlap screen.

3. Results

3.1. Phenotyping

All Sprague–Dawley rats were tested in three behavioral assays in the following order: locomotor activity in a novel environment, EPM, and tail flick. Measures from the first two assays contain exploratory locomotion traits, with the EPM assay also measuring spontaneous anxiety‐like behavior. Acute nociception was assessed in the tail flick test. Figure 1 contains frequency distributions of these behaviors prior to quantile normalization of the data. Summary statistics for the trait data are reported in Table S4. Apparent sex differences in the raw trait distributions (Figure 1) were modeled by including sex as a candidate covariate and regressing it out when it explained more than 2% of the phenotypic variance; the full set of covariates regressed out for each trait is reported in Table S2.

FIGURE 1.

Frequency distributions of behavioral phenotypes by sex. Panels A–C (locomotor response to a novel environment): total locomotor score, lateral locomotor score, rearing locomotor score. Panels D–J (EPM): EPM open arm time, EPM open arm entries, EPM percent time in open arms, EPM open arm time ratio, EPM distance traveled, EPM velocity, EPM time immobile. Panel K (tail flick): tail flick latency. Data shown prior to quantile normalization. Summary statistics are in Table S4; assay details and trait definitions are in Section 2.1 (phenotyping).

3.2. Phenotypic Correlations

Several very high correlations were expected by design: EPM distance traveled and EPM velocity are algebraically linked because velocity is distance over a fixed 5‐min session (ρ ≈ 1.00); EPM open arm time and EPM percent time in open arms are the same quantity rescaled by total time (ρ ≈ 1.00); and total locomotor score aggregates lateral locomotor score and rearing locomotor score, explaining their high correlations (ρ ≈ 0.84–0.98). Within the EPM, EPM percent time in open arms and EPM open arm time ratio were highly correlated (ρ ≈ 0.96, p ≤ 0.05), and EPM open arm entries correlated with both (ρ ≈ 0.71, p ≤ 0.05). EPM distance traveled and EPM velocity are each related inversely to EPM time immobile (ρ ≈ −0.84, p ≤ 0.05). Tail Flick Latency showed limited associations, significant with one of 11 traits (ρ ≈ 0.14) and otherwise not significant. Results are shown in Figure 2 and detailed values are provided in Table S5.

FIGURE 2.

Phenotypic correlations among behavioral traits. Lower‐triangular heatmap of Spearman rank correlations (Rho). Cells show rho; entries labeled “n.s.” indicate p ≥ 0.05 (two‐sided). Full rho and p values are in Table S5. Trait definitions are in Section 2.1. Thick black lines delineate locomotor, EPM, and tail flick trait groups.

3.3. SNP Heritability

Across 11 traits, SNP heritability (h ²) ranged from 0.136 to 0.375. Locomotor traits showed the highest heritability estimates, exploratory locomotion and immobility measures in the EPM were moderate, and open‐arm anxiety measures and tail flick latency were lower, with several open‐arm measures not reaching statistical significance. Full values and p‐values are provided in Table S6.

3.4. GWAS

We identified two independent quantitative trait loci (QTLs) at the 5% genome‐wide threshold and one additional QTL at the 10% threshold (−log₁₀ P = 5.98 at α = 0.05; −log₁₀ P = 5.60 at α = 0.10).

The strongest association was for EPM open‐arm behavior on chromosome 1. The three open‐arm EPM traits (EPM percent time in open arms, EPM open arm time, EPM open arm time ratio) index the same construct and are simple rescalings, which is also reflected by their phenotypic correlations; accordingly, we treat them as a single locus on chromosome 1. At this locus, the top SNP for EPM percent time in open arms and EPM open arm time was Chr1:267.773 Mb (−log₁₀ P = 7.28), whereas the top SNP for EPM open arm time ratio was Chr1:268.059 Mb (−log₁₀ P = 6.74); the LD interval at r ² ≥ 0.6 was 0.55 Mb. A moderate‐impact missense variant in Eno4 was identified within this interval (Table S7). Figure 3B shows the regional association plot for EPM percent time in open arms at Chr1:267.773 Mb.

FIGURE 3.

Genome‐wide significant associations across behavioral traits, with representative loci at Chr1 (EPM percent time in open arms, 267.773 Mb) and Chr17 (tail flick latency, 30.015 Mb). (A) Combined Manhattan plot showing genome‐wide significant hits. SNPs are plotted by genomic position (x) and −log₁₀(p) (y) for traits with genome‐wide significant associations. Red (genome‐wide; −log₁₀(p) = 5.98) and blue (suggestive; −log₁₀(p) = 5.60) lines mark the significance thresholds. Large, filled markers highlight curated lead SNPs (colors = traits); overlapping peaks are slightly separated horizontally for visibility. (B) Regional association plot for EPM percent time in open arms centered at the lead SNP (Chr1:267.773 Mb). Position (Mb) vs. −log₁₀(p); point color = LD (r ²) to the lead SNP. Window 266.0–269.5 Mb; significance lines at 5.98 (genome‐wide) and 5.60 (suggestive). Gene track displays genes across the region with arrows indicating transcription direction and labels by gene symbol. (C) Regional association plot for tail flick latency centered at the lead SNP (Chr17:30.015 Mb). Position (Mb) vs. −log₁₀(p); point color = LD (r ² ) to the lead SNP. Window 29.5–31.0 Mb; significance lines at 5.98 (genome‐wide) and 5.60 (suggestive). Gene track displays genes across the region with arrows indicating transcription direction and labels by gene symbol.

Tail flick latency mapped to chromosome 17 with top SNP Chr17:30.015 Mb (−log₁₀ P = 6.27) and an LD interval of 0.41 Mb at r ² ≥ 0.6. A moderate‐impact missense variant in Tex56p was identified within this interval (Table S7). Figure 3C shows the regional association plot for tail flick latency at Chr17:30.015 Mb. In a sensitivity analysis excluding animals at the 15 s maximum cutoff time before quantile normalization, no loci reached genome‐wide significance. At the original chromosome 17 lead SNP (Chr17:30.015 Mb), the signal decreased from −log₁₀ P = 6.27 to 3.35 (Figure S1).

EPM time immobile reached the 10% threshold on chromosome 14 with top SNP Chr14:102.124 Mb (−log₁₀ P = 5.84) and an LD interval of 0.13 Mb at r ² ≥ 0.6. No moderate‐impact missense variants were observed in this interval.

Figure 3A shows the combined Manhattan plot for all genome‐wide significant associations. Full statistics are provided in Table S8. Manhattan plots for all traits are in Figure S1. Additional regional association plots are in Figure S2.

3.5. Comparison With bHR/bLR F₂

3.5.1. Phenotype Set and Age at Testing

Here we compared the current GWAS in outbred Sprague–Dawley rats with our previous GWAS of an F₂ intercross of selectively bred high‐responder and low‐responder (bHR/bLR) rats [8]. The studies used the same vivarium, apparatus, protocols, scoring rules, and preprocessing, and they assessed overlapping locomotor and EPM phenotypes. Age at testing differed between studies, with the F₂ rats tested at older ages than the outbred Sprague–Dawley rats (Table S9). Pavlovian conditioned approach (PavCA) was collected only in Chitre et al. [8], whereas tail flick latency was collected only in the current study.

3.5.2. Direction of Phenotypic Correlations

In the bHR/bLR F₂ rats, EPM distance traveled and EPM percent time in open arms were positively correlated with lateral, rearing, and total locomotor scores, whereas EPM time immobile showed negative or weak correlations. In outbred Sprague–Dawley rats, the corresponding correlations were smaller and several were not significant despite the higher sample size in the present study. Side‐by‐side values for all overlapping pairs are provided in Table S10.

3.5.3. Cross‐Study QTL Evaluation

In the bHR/bLR F₂ rats, genome‐wide significant loci were detected for lateral locomotor score, total locomotor score, rearing locomotor score, and EPM distance traveled [8]. For the same phenotypes measured in both studies, we lifted outbred Sprague–Dawley GRCr8 QTLs to RN6 and queried them in the F₂ summary statistics, and reciprocally lifted F₂ RN6 QTLs to GRCr8 and queried them in the outbred Sprague–Dawley summary statistics. We then performed the same liftover‐based comparison for cross‐phenotype pairs (e.g., F₂ locomotor intervals vs. outbred EPM intervals). In all cases, there were no overlapping LD‐defined QTL intervals between the two studies; when signals occurred on the same chromosome, peaks were separated by more than 100 Mb. Several lifted intervals contained variants with nominal p < 0.05 in the reciprocal dataset, but these represent the minimum p selected from many correlated tests within each interval and are therefore anti‐conservative. We report them for completeness in Tables S11 and S12 and do not interpret them as evidence of replication.

3.5.4. Gene‐Level eQTL Lookup

Gene‐level lookup in the hippocampal cis‐eQTL dataset from the bHR/bLR F₂ intercross described by Hebda‐Bauer et al. [32] provided supportive annotation for a subset of interval genes, including Eno4 and Eci3 (Table S13).

4. Discussion

In this study we performed a GWAS in commercially outbred Charles River Sprague–Dawley rats tested in young adulthood across three paradigms: locomotor response to a novel environment, EPM, and tail flick test. We analyzed 11 phenotypes. We chose these paradigms for their relevance to substance use liability and maintenance, with novelty‐related traits forecasting acquisition, escalation, and reinstatement of psychostimulant self‐administration [12, 16, 17], approach versus avoidance on the EPM relating to motivation for cocaine intake and supported by mixed‐reality human analogs [46, 47, 48], and thermal nociception and pain states modulating opioid seeking and consumption [18, 19, 20, 31].

In a separate study, we previously performed a GWAS in a bHR x bLR F₂ intercross derived from Sprague–Dawley rats [8]. However, we did not find any significant overlap between the findings across the two studies, as the two populations appear to have captured different facets of addiction‐relevant traits. While most of the traits for which loci were identified in the selectively bred F₂ rats reflect facets of exploratory locomotion in a novel environment, most of the traits identified in the outbred Sprague–Dawley reflect loci for anxiety‐related traits and not exploration. This difference in significant loci may reflect insufficient power, different alleles or allele frequencies between the two populations (perhaps due to selection for the HR and LR phenotypes), or age differences between the two cohorts or other factors. Our findings shed light on genetic differences associated with comparable exploratory locomotion and anxiety‐like behaviors in outbred Sprague–Dawley rats vs. selectively bred rats on the same background.

4.1. Phenotypic Correlations Align With Assay Expectations

Several very high correlations were expected by design: EPM distance traveled and EPM velocity, indexes of general activity, are algebraically linked because velocity is distance over a fixed five‐minute session, EPM open arm time and EPM percent time in open arms are the same quantity rescaled by total time, and total locomotor score aggregates lateral locomotor score and rearing locomotor score. Among anxiety‐related measures, EPM open arm time and EPM open arm entries were highly positively correlated and were each inversely correlated with EPM time immobile. EPM time immobile was also inversely correlated with the EPM general activity measures (i.e., distance traveled and EPM velocity). This pattern matches standard interpretations of the EPM for anxiety‐like behavior and locomotor activity within the apparatus [49].

Cross‐assay correlations for distance were weak (e.g., EPM distance traveled vs. lateral locomotor score ρ ≈ 0.11). This indicates that EPM movement is recorded under risk and immobility tendencies in outbred Sprague–Dawley rats, so it shares little variance with free locomotion in a walled novel cage [50, 51, 52]. In contrast, moderate cross‐assay correlations for distance were found in the bHR/bLR F₂ rats [8], highlighting differences in the outbred versus selectively bred (for a propensity to explore a novel environment) rat lines. Tail flick latency also showed minimal correlations with EPM and locomotor measures in the outbreds aside from a small positive association with EPM time immobile. Given the available pairwise sample sizes (EPM ≈ 654, locomotor ≈ 650, tail flick = 534), these weak cross‐assay associations likely reflect limited shared variance and comparatively greater measurement noise in tail‐flick latency rather than simple under‐sampling.

4.2. SNP Heritability of Behavioral Traits

Across 11 traits, SNP heritability estimates ranged from 0.136 to 0.375. Estimates were highest for the locomotor traits, moderate for EPM movement indices, modest for tail flick latency, and lowest for EPM open‐arm metrics; three open‐arm estimates did not reach p < 0.05. Although lower overall, this range of SNP heritability estimates from higher for exploratory locomotion traits to lower for anxiety‐related traits is consistent with that found in our previous study examining bHR/bLR F₂ rats [8]. The current results indicate that common variants account for a measurable share of trait variance in this outbred Sprague–Dawley population, supporting its use for GWAS [7].

4.3. GWAS Results

We studied young adult outbred Sprague Dawley rats across three paradigms capturing facets of exploratory locomotion, anxiety‐like behavior, and nociception and conducted GWAS on 11 behavioral phenotypes. Although locomotor traits showed the highest SNP heritability, no significant QTLs were identified for them. We identified three independent QTLs for other traits, two for anxiety‐related traits and one for nociception. The three EPM open arm measures, which quantify the same underlying construct (see Section 2.1), mapped to a single chromosome 1 locus, EPM time immobile mapped to chromosome 14, and tail flick latency mapped to chromosome 17. Across traits, sample sizes ranged from 534 to 654 rats, which reduces statistical power and is a limitation of the study. An approximate power analysis using QUANTO and the permutation‐derived genome‐wide significance threshold indicated that the present study had approximately 80%, 50%, and 30% power to detect loci explaining about 4.9%, 3.6%, and 2.9% of phenotypic variance at N = 654, and about 6.0%, 4.4%, and 3.5% at N = 534. These calculations indicate adequate power for moderate‐effect loci but limited power for smaller‐effect loci. We also acknowledge that effect‐size estimates for loci surpassing a stringent discovery threshold may be overestimated due to winner's curse. In the subsections below we summarize each interval and nominate positional candidates.

4.3.1. Chromosome 1 Locus for Open‐Arm Exploration in the EPM

Multiple genes in the chromosome 1 interval provided plausible mechanisms for individual differences in anxiety‐related open‐arm behavior. Slc18a2 encodes VMAT2, the vesicular monoamine transporter that sets vesicular loading of dopamine, norepinephrine, and serotonin. VMAT2 gain‐of‐function increases dopamine release and improves emotional behavior in mice, consistent with a monoaminergic pathway for EPM variation [53, 54]. Gfra1, the co‐receptor for GDNF, is required for mature synaptic connectivity in the medial habenula, a node that links limbic forebrain inputs to monoaminergic midbrain outputs and shapes anxiety and fear behaviors [55, 56]. Pdzd8 regulates endoplasmic reticulum‐mitochondria tethering and dendritic calcium handling, and its disruption in mice produces abnormalities in exploration and anxiety‐relevant behaviors, providing a non‐monoaminergic route to altered approach‐avoidance [57, 58]. We identified a moderate‐impact missense variant in Eno4. Although Eno4 has been described as testis‐enriched [59], gene‐level lookup in the hippocampal cis‐eQTL dataset from the bHR/bLR F₂ intercross described by Hebda‐Bauer et al. [32] also provided supportive annotation for this gene (Table S13). We therefore retain Eno4 as a possible positional candidate, while prioritizing Slc18a2, Gfra1, and Pdzd8 based on clearer established neurobiological roles. Taken together, Slc18a2, Gfra1, and Pdzd8 still provide the most coherent mechanistic framework for this locus, spanning monoamine handling, habenular circuitry, and neuronal calcium signaling.

4.3.2. Chromosome 14 Locus for EPM Immobility

Immobility in the EPM indexes anxiety state and coping within a risk context. Within this interval on chromosome 14, transcriptional regulators emerge as candidates. Rel encodes c‐Rel, an NF‐κB subunit. NF‐κB signaling influences affective behavior and hippocampal plasticity, and mouse work shows reduced anxiety‐like behavior when the pathway is perturbed, supporting a role for Rel in exploratory suppression and immobility [60, 61]. Bcl11a (CTIP1) controls cortical circuit development. Haploinsufficiency in humans and mice produces cognitive and behavioral alterations, consistent with forebrain mechanisms that could modulate immobility on the EPM [62]. Rel and Bcl11a represent plausible positional candidates for this locus.

4.3.3. Chromosome 17 Locus for Tail Flick Latency

Tail flick latency reflects spinal and supraspinal processing of thermal nociception. The interval on chromosome 17 contains Eci2 and Eci3, which encode enzymes involved in the β‐oxidation of unsaturated fatty acids. Gene‐level lookup in the hippocampal cis‐eQTL dataset from the bHR/bLR F₂ intercross described by Hebda‐Bauer et al. [32] provided supportive annotation for Eci3 but not Eci2 (Table S13). Lipid‐derived mediators are increasingly implicated in nociceptive processing, suggesting that regulatory variation in fatty‐acid handling, particularly their oxidation, could alter pain sensitivity [63, 64]. We also observed a moderate‐impact missense variant in Tex56p, a testis‐biased gene, but there are no known roles in pain. Given its expression pattern and lack of established links to nociception or pain, this missense variant is an unlikely causal driver. Overall, this locus appears novel and reinforces the established role of lipid metabolism in pain sensitivity. In a sensitivity analysis excluding animals at the 15 s ceiling before quantile normalization, the chromosome 17 association was no longer genome‐wide significant. This analysis removed a substantial group of animals at the high‐latency end of the phenotype and reduced the analyzed sample size from n = 524 to n = 456. We therefore retain the full‐sample tail‐flick GWAS as the primary analysis, while acknowledging the 15 s ceiling effect as an important limitation of the tail‐flick phenotype. A more affective‐relevant measure of pain, as opposed to a nociceptive reflex as measured in the tail flick test, may have revealed more significant QTL(s) that are also co‐localized with anxiety‐like traits. Future studies investigating the genetic variation of behavioral traits accompanying affective pain assays are worth pursuing.

4.4. Comparison With bHR/bLR F₂

We sought to identify allelic variants associated with the same behavioral traits by performing a GWAS in outbred Sprague Dawley rats from Charles River and comparing the results with the bHR/bLR F₂ intercross rats [8]. In this comparison, no phenotype‐matched genome‐wide significant QTL intervals overlapped between the studies, and we did not observe cross‐phenotype interval overlaps when F₂ locomotor loci were compared with outbred EPM loci. After lifting discovery intervals between builds and querying the reciprocal summary statistics within the lifted LD‐defined QTL intervals (r ² ≥ 0.6), several intervals showed nominal support (p < 0.05); however, these interval queries are anti‐conservative because they select the minimum p‐value across many correlated variants within each interval. We therefore treat these nominal signals as descriptive checks rather than evidence of replication.

Several factors could contribute to the lack of phenotype‐matched replication. Limited statistical power is likely an important factor in both studies, given sample sizes and the polygenic architecture of these behavioral traits. In addition, allele frequencies and local LD likely differed between the bHR/bLR F₂ and the outbred Sprague–Dawley cohort due to the history of selection in the bHR/bLR lineage and colony over time, which can influence which alleles are common enough to detect in GWAS [7]. Epistatic interactions could further limit replication if single‐locus effects depend on genetic background [65, 66, 67]. Finally, the F₂ rats were older at testing than the outbred rats (123 ± 5 days vs. 77.7 ± 0.9 days for locomotor and 85.2 ± 1.6 days for EPM), and adult‐age differences can shift baseline novelty and anxiety‐like behavior and alter apparent effect sizes [68, 69]. Nevertheless, both the outbred Sprague–Dawley and the bHR/bLR F₂ GWASs provide valuable genetic insight into addiction‐relevant behaviors observed in a normal distribution of the population compared to those generated through selective breeding based on a specific behavioral criterion for modeling individual differences and the extremes in human behavioral liabilities.

5. Conclusion

In conclusion, we found three independent QTLs in outbred Charles River Sprague–Dawley rats for anxiety‐related traits in the EPM and nociception in the tail flick test. Candidate genes and annotations point to monoaminergic signaling, transcriptional regulation, and lipid metabolism. We compared these loci with those from an F₂ intercross of selectively bred high‐responder and low‐responder rats [8]. We did not observe phenotype‐matched replication. This comparison highlights genetic differences associated with comparable exploratory locomotion and anxiety‐like behaviors in outbred Sprague–Dawley rats vs. selectively bred rats on the same background. Genetic characterization of both natural behavioral variation and that created by selective breeding provides insight into determining causal variants tied to individual differences in behavioral liabilities relevant to psychiatric and substance use disorders. Taken together, these results extend the genetic characterization of addiction‐relevant behaviors in outbred Sprague–Dawley rats and motivate replication and targeted experimental tests of mechanism.

Funding

This work was supported by the NIDA Center for GWAS in Outbred Rats (P50DA037844), the Center for Genetics, Genomics, and Epigenetics of Substance Use Disorders in Outbred Rats (P30DA060810), NIDA U01DA043098 (H.A., J.L., A.A.P.), ONR 00014‐19‐1‐2149 (H.A.), the Pritzker Neuropsychiatric Research Consortium (H.A., S.J.W.), and the Hope for Depression Research Foundation (HDRF) (H.A.).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Demographic information. Behavioral assay, trait, sample size by sex, and testing age. Age values are mean ± SD in days at the time the assay was run. “M/F” gives the male and female counts. Within an assay block, all listed traits share the same sample size and age as the first row of that block unless otherwise noted. For trait definitions and procedures, see Section 2.1.

Table S2: Covariates regressed out for each trait after the > 2% variance screen. For each behavioral trait, the table lists covariates that individually explained more than 2% of the phenotypic variance and were therefore regressed out prior to downstream analyses. For each covariate, R ² denotes the proportion of trait variance explained by that covariate alone, and percent variance explained reports 100×R ². Sex met the > 2% criterion for several EPM measures but not for locomotor or tail flick traits.

Table S3: Variant‐level quality control summary. Filter category indicates which variant‐level quality control filter(s) each SNP failed: minor allele frequency (MAF < 0.05), Hardy–Weinberg equilibrium (HWE p < 1 × 10⁻⁷), and/or per‐SNP genotype missingness (F‐MISS above the specified thresholds). Rows labeled MAF only, HWE only, and F‐MISS only correspond to SNPs that failed only that single filter, whereas combination rows correspond to SNPs that failed multiple filters. Total SNPs gives the total number of autosomal SNPs before QC. As noted in the main text, of the 9,088,280 SNPs with MAF < 0.05, 8,003,804 (88.1%) were monomorphic (MAF = 0).

Table S4: Summary statistics for behavioral traits by sex and overall. For each trait within each behavioral assay (locomotor response to a novel environment, elevated plus maze, tail flick), columns report mean ± SD, median, IQR, Min, and Max for female, male, and overall. Values are computed from data before quantile normalization. IQR, interquartile range; SD, standard deviation.

Table S5: Phenotypic correlations. Columns: Trait 1, Trait 2, rho, p_value, sig. rho is the Spearman rank correlation coefficient; p_value is two‐sided; sig indicates significance at p ≤ 0.05 (yes/no). Correlations were computed on covariate‐adjusted, quantile‐normalized residuals using pairwise complete observations. Trait definitions are in Section 2.1.

Table S6: SNP‐based heritability of behavioral traits. Heritability (h ²) was estimated with GCTA GREML using a genomic relationship matrix. The p value column reports the GCTA GREML likelihood‐ratio test p‐value for H₀: h = 0. Values that appeared as 0.000 in source files are shown as < 0.001. Traits are ordered by h ² from highest to lowest. Trait definitions are provided in Section 2.1.

Table S7: Moderate‐impact coding variants within QTL intervals. Summary of coding variants in linkage disequilibrium (LD) with trait lead SNPs. Coordinates are on the GRCr8 rat genome build. Candidate gene is the gene harboring the coding variant and does not imply causality. QTL: peak marker of the QTL. SNP in LD with peak marker: moderate‐impact missense variant (per SnpEff) in LD (r ² ≥ 0.6) with the peak marker. LD r ² with trait top SNP: pairwise LD r ² between the listed variant and the trait lead SNP, estimated in the outbred Sprague–Dawley sample. SNP change: HGVS c. annotation. Amino acid change: HGVS p. annotation.

Table S8: Summary of QTLs. Peak markers and linkage‐disequilibrium (LD) intervals for quantitative trait loci. Coordinates are on the GRCr8 rat genome build. −log10P is the test statistic; Significance_Level indicates the 5% or 10% genome‐wide threshold derived by permutation. A1 is the effect allele, A2 is the other allele, and Freq is the effect‐allele frequency. b is the per‐allele effect on the quantile‐normalized trait with standard error (SE). The LD interval is defined as all variants with r ² ≥ 0.6 around the peak, reported as start and stop base‐pair positions and physical size in megabases.

Table S9: Shared phenotypes between the outbred Sprague–Dawley (this study) and the bHR/bLR F₂ rats [8]. This table shows the mean age ± SD at the time of behavioral testing for the rats in the two studies.

Table S10: Phenotypic correlations (Spearman rho) between elevated plus maze (EPM) measures and locomotor traits in outbred Sprague–Dawley (this study) and bHR/bLR F₂ rats [8]. This table reports Spearman rho and two‐sided p for each overlapping EPM × locomotor trait pair measured in both studies. Values are shown as: rho_Outbred, p_Outbred, rho_F2, p_F2. Only the traits present in both studies appear in the rows of this table. Outbred Sprague–Dawley refers to this study; bHR/bLR F₂ refers to the intercross reported in Chitre et al. [8].

Table S11: bHR/bLR F₂ (RN6) QTLs lifted to GRCr8 and Queried in outbred Sprague–Dawley. This table lists QTLs discovered in the bHR/bLR F₂ study on RN6 and their lifted coordinates on GRCr8 used to query the outbred Sprague–Dawley GWAS. Columns: behavioral test, trait (F₂ label); original top SNP: F₂ peak SNP formatted as chr:pos; original top SNP −log₁₀ P: F₂ peak strength; original LD start: F₂ LD window start in bp; original LD stop: F₂ LD window stop in bp; original interval size: F₂ LD window size in Mb computed as (stop − start)/1e6; queried trait: queried outbred trait; lifted top SNP: lifted peak SNP on GRCr8 formatted as chr:pos, NA if absent; lifted LD start: lifted LD window start in bp on GRCr8; lifted LD stop: lifted LD window stop in bp on GRCr8; lifted interval size: lifted window size in Mb; best −log₁₀ P: best signal in outbred Sprague–Dawleys within the lifted window reported as −log₁₀ P; best SNP: position of that best signal formatted as chr:pos. Sizes are in megabases and rounded.

Table S12: Outbred Sprague–Dawley (GRCr8) loci lifted to RN6 and queried in bHR/bLR F₂. This table lists QTLs discovered in outbred Sprague–Dawleys on GRCr8 and their lifted coordinates on RN6 used to query the bHR/bLR F₂ GWAS. Columns: behavioral test, trait: outbred trait label; original top SNP: outbred peak SNP formatted as chr:pos; original top SNP −log₁₀ P: outbred peak strength; original LD start: outbred LD window start in bp; original LD stop: outbred LD window stop in bp; original interval size (Mb): outbred window size in Mb; queried trait: queried F₂ trait; lifted top SNP: lifted peak SNP on RN6 formatted as chr:pos, NA if absent; lifted LD start: lifted LD window start in bp on RN6; lifted LD stop: lifted LD window stop in bp on RN6; lifted interval size (MB): lifted window size in Mb; best −log₁₀ P: best signal in F₂ within the lifted window reported as −log₁₀ P; best SNP: position of that best signal formatted as chr:pos. Sizes are in megabases and rounded.

Table S13: Gene‐level lookup of hippocampal cis‐eQTL support from the bHR/bLR F₂ intercross described by Hebda‐Bauer et al. [32] for genes within the outbred GWAS intervals. This table lists genes located within the final outbred Sprague–Dawley GWAS intervals and whether each gene showed supportive hippocampal cis‐eQTL evidence in the related dataset described by Hebda‐Bauer et al. [32]. Columns: outbred lead SNP, top SNP for the outbred locus formatted as chr:pos on GRCr8; associated trait(s), outbred behavioral trait(s) mapping to that locus; gene, gene symbol within the outbred interval; gene description, gene name/description; hippocampal cis‐eQTL support, yes if the gene was present in the gene‐level hippocampal cis‐eQTL results from the related dataset and no otherwise. Because the present outbred GWAS used GRCr8 whereas the related eQTL dataset used RN6, this comparison was restricted to gene‐level annotation rather than SNP‐ or LD‐level matching.

Figure S1: Manhattan plots for each behavioral trait. SNPs are plotted by genomic position on GRCr8 (x) and −log₁₀ P from LOCO mixed‐model GWAS (y). Horizontal lines show genome‐wide (5% at −log₁₀ P = 5.98) and suggestive (10% at −log₁₀ P = 5.60) thresholds derived by permutation.

Figure S2: Regional association plots for additional significant loci. (A) EPM open arm time ratio at the chromosome 1 locus. (B) EPM open arm time at the same chromosome 1 locus. (C) EPM time immobile at the chromosome 14 locus. Points are colored by LD r ² to the lead SNP; axes show genomic position in Mb and −log₁₀ P. Horizontal lines mark the genome‐wide and suggestive thresholds (−log₁₀ P = 5.98 and 5.60, respectively). The gene track displays annotated genes with transcription direction. Panels A and B map to the same chromosome 1 interval as the EPM percent time in open arms locus in the main figure.

Data Availability Statement

The phenotype, genotype, and GWAS summary statistic data supporting this study are available through UC San Diego Library Digital Collections: https://doi.org/10.6075/J0222VS5. Raw sequencing reads are available through NCBI's Sequence Read Archive under BioProject accession PRJNA1460565: https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1460565.