Traits of fear resistance and susceptibility in an advanced intercross line

Abstract

Genetic variability in the strength and precision of fear memory is hypothesized to contribute to the etiology of anxiety disorders, including post-traumatic stress disorder. We generated fear-susceptible (F-S) or fear-resistant (F-R) phenotypes from an F₈ advanced intercross line (AIL) of C57BL/6J and DBA/2J inbred mice by selective breeding. We identified specific traits underlying individual variability in Pavlovian conditioned fear learning and memory. Offspring of selected lines differed in the acquisition of conditioned fear. Furthermore, F-S mice showed greater cued fear memory and generalized fear in response to a novel context than F-R mice. F-S mice showed greater basal corticosterone levels and hypothalamic corticotrophin-releasing hormone (CRH) mRNA levels than F-R mice, consistent with higher hypothalamic–pituitary–adrenal (HPA) axis drive. Hypothalamic mineralocorticoid receptor and CRH receptor 1 mRNA levels were decreased in F-S mice as compared with F-R mice. Manganese-enhanced magnetic resonance imaging (MEMRI) was used to investigate basal levels of brain activity. MEMRI identified a pattern of increased brain activity in F-S mice that was driven primarily by the hippocampus and amygdala, indicating excessive limbic circuit activity in F-S mice as compared with F-R mice. Thus, selection pressure applied to the AIL population leads to the accumulation of heritable trait-relevant characteristics within each line, whereas non-behaviorally relevant traits remain distributed. Selected lines therefore minimize false-positive associations between behavioral phenotypes and physiology. We demonstrate that intrinsic differences in HPA axis function and limbic excitability contribute to phenotypic differences in the acquisition and consolidation of associative fear memory. Identification of system-wide traits predisposing to variability in fear memory may help in the direction of more targeted and efficacious treatments for fear-related pathology.

Introduction

Learning about and remembering fearful experiences is critical to survival. Disturbances in this fundamental behavior are linked to a variety of psychiatric diagnoses. Stronger acquisition of conditioned fear is a common trait among patients with anxiety disorders, including post-traumatic stress disorder (PTSD) and depression (Orr et al., 2000; Lissek et al., 2005; Nissen et al., 2010; Norrholm et al., 2011). PTSD in particular is associated with dysregulation of fear memory circuits (Tryon, 1999; Elzinga & Bremner, 2002; Shin & Handwerger, 2009). Pavlovian fear conditioning is a robust and well-established paradigm for studying emotional memory in both humans and animals (Orr et al., 2000; Amstadter et al., 2009; Johnson et al., 2012; Mahan & Ressler, 2012). Importantly, individual variability in ‘conditionability’, or the strength of memory formation, is heritable (Hettema et al., 2003; Johnson et al., 2012). Thus, identifying biological traits predicting fear memory formation has important clinical implications for both understanding the etiology of, and developing intervention strategies for, disorders of fear regulation (Shin & Handwerger, 2009; Johnson et al., 2012).

Selective breeding has been used in behavioral genetics for decades (Hall, 1938). Genetically heterogeneous populations are necessary for selective breeding. A simple and well-defined outbred population is an advanced intercross line (AIL). AILs are created through pseudo-random breeding over multiple generations, reducing the linkage of nearby alleles (Parker & Palmer, 2011). In this study, we used three generations of bi-directional selective breeding for contextual fear memory to establish two divergent lines, which we call fear-susceptible (F-S) and fear-resistant (F-R). We then examined physiological and neural traits in behaviorally naïve mice whose parents were from the final selected generation. These included parametric study of fear conditioning, baseline hypothalamic–pituitary–adrenal (HPA) axis function, HPA axis gene expression, and measures of brain structure and function in vivo with manganese-enhanced magnetic resonance imaging (MEMRI). MEMRI signal enhancement primarily derives from intracellular accumulation of Mn²⁺ through voltage-gated Ca²⁺ channels, allowing high-resolution visualization of functionally dependent neural processing and brain anatomy (Koretsky, 2004; Silva et al., 2004; Silva & Bock, 2008; Koretsky, 2012).

We are using a carefully designed AIL with short-term selection to segregate trait-relevant allelic combinations with minimal random genetic drift (Belknap et al., 1997). This strategy greatly reduces potential non-specific contributions of the genetic background, and increases the likelihood that divergence between lines relates directly to the selection criterion. A major strength of our behavioral selection strategy is that we were able to evaluate these phenotypes in behaviorally naïve animals, and thus we can identify genetically determined physiological profiles that may contribute to the differences in strength of fear memory encoding. Identification of a physiological profile of fear resistance and susceptibility is a key milestone in identifying endophenotypes of altered fear memory processing and ultimately a biological basis for PTSD and other anxiety disorders.

Materials and methods

Subjects

Mouse lines were derived from the F₈ generation of a C57BL/6J × DBA/2J AIL created by Palmer (Parker et al., 2012). F₈ AIL mice were trained and tested for cued and contextual fear in a Pavlovian fear conditioning protocol in Chicago, as described previously (Ponder et al., 2007; Parker et al., 2012). Mice showing either the highest or lowest contextual fear [selected generation 1 (S1)] were assigned into breeding pairs and shipped to the Uniformed Services University of the Health Sciences (USUHS) laboratory animal facility. Breeding and selection of successive generations [selected generation 2 (S2) and selected generation 3 (S3)] was conducted at USUHS. Mice were housed two to five per cage in a climate-controlled vivarium on a 12 : 12-h light cycle (lights on 06:00 h) with ad libitum access to food and water. Experiments were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Animals. All experimental procedures were reviewed and approved by the appropriate (University of Chicago or USUHS) Institutional Care and Use Committee.

Behavioral phenotyping

F-R and F-S mice, 8–12 weeks old, were tested for contextual and cued conditioned fear response with a fear conditioning protocol described previously and below (Ponder et al., 2007) (Fig. 1A). Freezing was scored with anymaze analysis software (Stoelting, Wood Dale, IL, USA). The selection criterion was the percentage of freezing during the first 300 s or the contextual fear test. F-S mice with the highest freezing and F-R mice with the lowest freezing were retained for breeding, and paired according to their freezing scores. To maintain heterogeneity, related mice were not paired for breeding. We used mass selection, meaning that we did not place limits on how many mice were selected from each family to breed the next generation. Two or three litters per breeding pair were phenotyped, to increase the likelihood that the majority of breeding pairs would be represented in the subsequent generation.

Figure 1.

(A) Diagram of fear conditioning for phenotypic selection. (B and C) Contextual and conditioned fear expression in S2 and S3 males and females selected for breeding. (B) In S2, F-S males froze significantly more than any other group, and F-S males and females froze more than F-R animals of either sex. In the cued fear test, F-S mice froze more than F-R mice. (C) In S3, there were no differences between males and females in either contextual or conditioned fear. S2 mice selected for breeding and spares: F-S males, n = 12; F-S females, n = 14; F-R males, n = 16; F-R females, n = 14. S3 mice selected for breeding and spares: F-S males, n = 16; F-S females, n = 15; F-R males, n = 12; F-R females, n = 14. ***P < 0.0001 between F-S and F-R; #P < 0.05 between males and females.

In S1, 118 mice were phenotyped and 60 mice were selected to create 15 F-R and 15 F-S breeding pairs (~50% of the total S1 cohort). Three hundred and forty-eight mice (150 F-S and 198 F-R) were phenotyped in S2 to establish 11 F-S and 12 F-R breeding pairs (~15% of each cohort in S2). Two hundred and thirty mice were phenotyped in S3 (134 F-S and 98 F-R). Fifteen F-S and 15 F-R breeding pairs were established from the S3 litters (25–30% of each cohort in S3).

Phenotyping was performed over a period of 3 days (Fig. 1A): training on day 1, context fear test on day 2, and cued fear test on day 3. Training consisted of a 3-min exploration period followed by two presentations of a 30-s, 4.5-kHz, 75-dB pure tone conditioned stimulus (CS) co-terminating in a 2-s, 0.5-mA foot shock delivered with a 30-s interval through a conducting floor. On day 2, the mice were returned to the training chamber for 600 s. The first 300 s were scored for context-related fear. On day 3, the chamber was altered for visual, tactile and olfactory cues. House lights in the chamber were turned up and filtered through a yellow filter. The conducting rods were covered with smooth plastic, and patterned inserts covered the chamber walls. The cleaning solution was changed from 70% ethanol to 1% acetic acid. After 3 min of exploration in the new environment, mice were presented with two 30-s, 4.5-kHz, 75-dB tones identical to the training tone with a 30-s interval. Data for context freezing on day 2 and freezing to in response to tone were analyzed by two-way anova with line (F-S/F-R) and sex as between-subject factors.

Fear conditioning

Naïve selected generation 4 (S4) males and 8–12-week-old females from 14 F-R and 11 F-S breeding pairs were used for these experiments. One hundred and twenty-seven male and female mice were fear conditioned with one of three training conditions, all of which utilized a 30-s, 5-kHZ, 75-dB pure tone auditory CS: (i) two presentations of the CS co-terminating in a 2-s, 0.5-mA foot shock (2 × 0.5), with a 30-s interval; (ii) two presentations of the CS co-terminating in a 2-s, 1-mA foot shock with a 30-s interval (2 × 1.0); or (iii) 10 CS presentations co-terminating in a 2-s, 0.5-mA foot shock with a variable inter-trial interval (10 × 0.5). Data for two 2 × 1.0 mice were lost because of camera failure, and seven mice in the 2 × 0.5 and 2 × 1.0 training groups had freezing levels outside two standard deviations from the mean, and were dropped from analysis as outliers. Because the determination of cued fear was based on an auditory stimulus, and early-onset deafness is sometimes found in the DBA/2 parent strain (Zheng et al., 1999), videos from F-S and F-R mice with <10% freezing in response to tone 1 of the cued memory test were reviewed for an orienting response. We wanted to exclude the possibility that the cued fear response in either group, but primarily F-R mice, was artificially reduced because of hearing impairment. In two sibling mice, it could not be determined whether the mouse heard the tone (i.e. pausing briefly or turning towards the sound). These two mice were among the seven mice excluded.

One day after training, cued fear memory was tested in the alternative context (described above), with two 30-s 5-kHz, 75-dB tones with a 30-s interval. Final numbers for each experiment are stated in the figure legends. Freezing was scored using freezeframe (Coulbourn Instruments, Whitehall, PA, USA). Freezing in the training context and novel test context prior to the first tone CS (pre-training and pre-testing baseline freezing) and freezing in response to each CS during training and testing were determined. Mean freezing in response to the two test CSs is presented.

Sample collection for enzyme-linked immunosorbent assay (ELISA)

Urine samples were collected between 07:30 h and 08:30 h from naïve male offspring of five F-S and six F-R breeding pairs (seven F-S and eight F-R). Mice were 8–11 weeks old (mean age: F-S, 10 weeks; F-R, 8.25 weeks). Collection was performed over three consecutive days. Samples were diluted 1 : 10 in the dilution buffer provided with the ELISA kit, and stored at −20 °C.

Corticosterone ELISA

Corticosterone concentrations were determined with an AssayMax corticosterone ELISA kit (AssayPro, St Charles, MO, USA). Absorbance of standards and samples was read on a Tecan microplate reader (Tecan, Research Triangle Park, NC, USA). Samples were run in triplicate, and concentrations were calculated from mean absorbance. Corticosterone concentrations in urine samples were determined from a six-point standard curve. Corticosterone concentrations were quantified in 13 samples from seven F-S males and 14 samples from eight F-R males. For mice with replicate samples, corticosterone concentrations for the 2 days were averaged and plotted as a single data point.

Quantitative polymerase chain reaction (qPCR)

A separate cohort of naïve adult S3 and S4 F-S and F-R mice was used (30 F-S and 51 F-R). The hypothalamus was rapidly dissected from 1.5-mm coronal sections with a 16-gauge needle, and immediately frozen in dry ice. Total RNA was extracted from homogenized tissue with the RNeasy Minikit (Qiagen, Valencia, CA, USA). Oligonucleotide primers were designed with primer 3 software for the genes for corticotrophin-releasing hormone (CRH), CRH receptor 1 (CRHR1), mineralocorticoid receptor (MR), and glucocorticoid receptor (GR), and two reference genes: the β-actin gene and the β₂-microglobulin gene. Reactions were quantified with the cycle threshold method with sds 2.2 software (Applied Biosystems). An average quantity value (Qty mean) for each sample, run in duplicate, was calculated for each gene. The geometric mean of the Qty mean of the β₂-microglobulin and β-actin genes was calculated. The data for each gene were expressed as Qty mean for the gene of interest/geometric mean of Qty mean for the two reference genes.

MEMRI

MEMRI was conducted on male F-S (n = 7) and F-R (n = 7) age-matched and weight-matched S4 mice. Mice (n = 14) were treated with Mn²⁺ in 400 mm bicine buffer (pH 7.4) continuously over 3 days with a micro-osmotic pump (1.0 μL/h; Model 1003D; Alzet, Durect Corp., Cupertino, CA, USA), giving a cumulative final MnCl₂ dose of 180 mg/kg (MnCl₂.4H₂O; Sigma-Aldrich). This Mn²⁺ dose produces significantly enhanced T1-weighted magnetic resonance imaging (MRI) signal intensity without observable toxicity (Mok et al., 2011). Mice (two F-R) not implanted with a micro-pump were scanned to estimate the relative change in T1-weighted MEMRI signal intensity (SI). Mice underwent fear conditioning at the end of Mn²⁺ dosing (day 3; data not shown). We interpreted the functional MEMRI results as representing basal activity, because Mn²⁺ uptake in the brain was slow and occurred over 72 h in the home cage prior to conditioning.

MEMRI acquisition

MRI was conducted at the end of day 3 with a 7-T Bruker BioSpec imaging system (Bruker Biospin, Billerica, MA, USA) with a 20-cm horizontal inner bore diameter and a dedicated mouse phased-array (n = 4) head coil. Images were acquired in anesthetized mice (1.5–2% isoflurane), by use of a three-dimensional (3D) rapid acquisition with refocused echoes sequence: repetition time, 200 ms; echo time, 6.81 ms; rapid acquisition with refocused echoes factor, 2; image matrix, 192 × 168 × 128; isotropic resolution, 100 μm. The total scan time was 72 min. During scanning, mice were monitored for temperature, respiration, and heart rate.

MEMRI functional analysis

The influence of surface coil location on SI for MEMRI images was reduced with the inhomogeneity N3 correction (Sled et al., 1998) (mipav; NIH, Bethesda, MD, USA). Mean SI was measured from several regions of interest (ROIs). ROI alignment was conducted by an experimenter blinded to mouse phenotype. A predefined mouse atlas was used to delineate 14 3D ROIs (vivoquant; InviCRO, Boston, MA, USA). We manually segmented eight additional 3D ROIs (vivoquant) to incorporate a fear memory circuit into the analysis (Maren & Quirk, 2004; Pape & Pare, 2010). Manual 3D ROIs were the dorsal hippocampus, ventral hippocampus, lateral amygdala (LA), basal amygdala, central amygdala, prelimbic cortex, infralimbic cortex, and periaqueductal gray. Finally, we designated seven additional two-dimensional ROIs within the hippocampus, sampling from one rostral and one caudal location. We anatomically segregated subfields of the dorsal (CA1, CA3, and dentate gyrus) and ventral (CA1, CA2, CA3, and dentate gyrus) hippocampus, because of the high T1-weighted contrast in these areas (Fig. 5). All 3D ROIs were bilateral, and all 2D ROIs were unilateral (right ventral hippocampus and left dorsal hippocampus). Manually defined ROIs were delineated according to anatomical reference points and a stereotaxic mouse atlas (Franklin & Paxinos, 2008). For functional comparisons, each ROI was normalized by use of the whole brain SI. In the first step of the normalization procedure, the SI of the whole brain was calculated for each subject by using a mean of the 14 predefined 3D ROIs (described above). Next, the whole brain SI for each subject was normalized to the mean whole brain SI of the group. In the final step, the ROI SI for each subject was divided by its own normalized mean whole brain SI. Relative Mn²⁺ enhancement was determined by calculating the percentage change in SI for each ROI from the corresponding ROI SI of the group without Mn²⁺. Additional measurements were conducted to validate the quality of whole brain normalization procedure described above (Supporting information). Percentage change in SI was the dependent variable in data reduction [principal components analysis (PCA)].

Figure 5.

MEMRI revealed that a unique and coordinated pattern of basal neuronal activity differentiates F-S and F-R mice. (A) Representative T1-weighted whole brain MRI scans (isotropic resolution, 100 μm) of a mouse brain without (−MnCl₂) and after (+MnCl₂) MnCl₂ infusion (180 mg/kg final dose). (B) MEMRI SI for F-S mice was normalized to F-R mice (F-S/mean F-R) for ROIs with the highest loading (>0.5) on PC3 (‘a fear memory network’). Representative MEMRI images depict ROIs that differentiated the phenotype at a (1) rostral (CA1 and LA) and (2) caudal (CA1, CA2, and C3) coronal plane. (C) Volumetric analysis of the hippocampus in vivo. Representative image of the mouse hippocampus reconstructed in three dimensions from 100-μm isotropic MEMRI scans (left panel). There were no differences in hippocampal volume between lines (right panel). F-S males, n = 6; F-R males, n = 6. *P < 0.05 between F-S and F-R.

MEMRI volumetric analysis

Lines were established by use of a hippocampus-dependent behavioral task (contextual fear conditioning). Therefore, we hypothesized that structural alterations related to the phenotype would most likely be located in the hippocampus. The hippocampus was reconstructed in three dimensions from N3-corrected raw MEMRI scans (vivoquant). Volume (mm³) was the dependent variable for statistical analyses.

Statistical analysis

Fear conditioning was analyzed with anova, with line (F-S/F-R), sex and training (2 × 0.5, 2 × 1.0, and 10 × 0.5) as between-subject variables, and day (training/test) and segment (pre-CS baseline, CS1, and CS2) as within-subject variables. Subsequent baseline-subtracted conditioned freezing was compared by use of three-way repeated measures anova, with day as the within-subject variable and line, sex and training as between-subject variables (baseline subtraction detailed below). qPCR data were compared by use of two-way anova, with line and sex as between-subject variables. For anova, interactions were reported first. Where there were no interactions, main effects were reported. Follow-up analyses were conducted with one-way anova, with Bonferroni correction for multiple comparisons or independent samples t-tests for single comparisons. Corticosterone concentrations were compared between lines by use of independent samples t-tests.

PCA

PCA was used to visualize and interpret brain-wide activity patterns. PCA is a data reduction technique used to capture hidden patterns that best explain variance in large datasets. Here, PCA was used to visualize functional brain circuits. The result of PCA is a set of loadings and scores. Loadings are typically used to visualize and interpret patterns. PCA was used in this study not to visualize patterns as such, but rather to extract patterns that reflect underlying phenotypic differences in basal brain-wide network cellular activity. To this end, scores were used to determine which, if any, of the principal components (PCs) could account for a pattern of brain activity associated with the phenotype. Component scores provide an index of where individuals are distributed along a particular component. We grouped scores by phenotype to evaluate differences in the pattern of brain activity associated with each component. The variables for PCA were the mean ROI SIs. All variables were log-transformed prior to PCA. A covariance matrix PCA was chosen to generate a more defined component structure. Loadings were subject to varimax rotation to maximize simplicity, interpretation and orthogonality among components. Unless otherwise noted, all statistical comparisons between phenotype were conducted with one-way anova. Correlations were conducted with Pearson’s r. Statistical analyses were conducted with spss (IBM SPSS, Armonk, New York, USA) and graphpad prism (GraphPad Software, La Jolla, CA, USA). Values in the text represent the mean ± standard error of the mean. Statistical significance was set at <0.05 for all comparisons.

Results

Behavioral phenotyping

In S2, there was a sex × line interaction (F_1,55 = 45.57; P < 0.0001) in the contextual fear test. F-S males froze more than F-S females (P < 0.0001) and F-R mice of either sex (P < 0.0001) (Fig. 1B). F-S females froze more than F-R mice (P < 0.0001) (Fig. 1B). In the cued fear test, there was a main effect of line (F_1,52 = 13.34; P = 0.0006). No other effects or interactions of sex were identified in either S1 or S3, and may be stochastic. Main effects of line were identified in context (F_1,54 = 244.72; P < 0.0001) and cued (F_1,53 = 16.08; P = 0.0002) fear tests in the S3 breeding population (Fig. 1C). These results indicate that the behavioral spectrum of the AIL population was broad, as would be expected in a natural population. Selecting from the extreme tails of the spectrum in this way established a robust phenotype in S4.

Behavioral responding in F-R and F-S lines

Differences between the F-S and F-R lines in acquisition and cued fear were evaluated in S4. Differences between F-S and F-R mice were detectable both between and within training and test days. anova identified line × day (F_1,105 = 17.61; P < 0.0001) and line × segment (F_2,210 = 6.579; P = 0.002) interactions, indicating that, as expected, behavioral differences were detectable both between days and within test or training days. During training, there was a line × segment interaction (F_1,115 = 15.707; P < 0.0001). There was a small but significant difference in freezing between F-S (9.3%) and F-R (5.2%) mice during the first tone, before the shock was applied (P = 0.0003), which may reflect an orienting response or a low level of anxiety, but does not reflect fear, as the first shock was not presented until the end of the tone. At CS2, F-S mice still froze more than F-R mice (F-S, 35.7%; F-R, 21.17%; P < 0.0001) (Fig. 2).

Figure 2.

F-S and F-R fear memory phenotypes are not altered by training strength. (A–F) Freezing in F-S mice was significantly higher than in F-R mice during both training. (A, C, and E) and test for cued fear (B, D, and F). (G–H) Phenotypic differences in conditioned fear depended on consolidation. F-R mice froze to the same extent (G) or more (H) at acquisition (day 1) than F-S mice when a more intense training protocol was applied, but phenotypic differences in conditioned fear memory (day 2) remained. (A–F) 2 × 0.5-mA training condition: F-S males, n = 10; F-R males, n = 16; F-S females, n = 10; F-R femalers, n = 8. 2 × 1.0-mA training condition: F-S males, n = 7; F-R males, n = 11; F-S females, n = 15; F-R females, n = 9. 10 × 0.5-mA training condition: F-S males, n = 6; F-R males, n = 6; F-S females, n = 11; F-R females, n = 8. (G–H) F-S, 2 × 0.5, n = 20; F-R, 2 × 1.0, n = 20; F-R, 10 × 0.5, n = 14. ***P < 0.0001, **P < 0.001 and *P < 0.05 between F-S and F-R.

On day 2 (test), F-S mice always froze more than F-R mice (F_1,115 = 40.472; P < 0.0001) (Fig. 2B, D, and F). Phenotypic differences in cued fear were not disrupted by increasing training intensity (2 × 1.0) or duration (10 × 0.5). However, the more intensely trained F-R mice showed similar (main effect of day: F-S 2 × 0.5 as compared with F-R 2 × 1.0, F_1,39 = 38.29, P < 0.0001) or even greater (segment × day interaction: F-R 2 × 0.5 as compared with F-R 10 × 0.5, F_1,33 = 26.47, P < 0.0001; train, P = 0.006; test, P = 0.0009) levels of freezing on day 1 (training) than F-S mice subjected to the milder 2 × 0.5 training paradigm on day 1 (Fig. 2G and H). However, in spite of the difference in fear load at acquisition, F-S mice froze more than F-R mice when tested for fear memory expression 24 h later, implicating a genetically determined difference in fear memory stabilization and consolidation (Fig. 2G and H).

Two-way anova of day and line on pre-tone baseline freezing identified an interaction (F_1,105 = 18.285; P < 0.0001). Before training, baseline freezing in a novel environment was identical between F-S and F-R mice (P = 0.58) (Fig. 3A). However, after training, F-S mice showed a pronounced freezing response to a novel context (testing chamber) (P < 0.0001) (Fig. 3B). Initial differences in pre-cue freezing on day 2 between F-S and F-R mice complicate the interpretation of the cued fear response. Differences in baseline fear were addressed by subtracting baseline freezing from cued freezing within subjects (Fig. 3). We hypothesized that baseline freezing on day 2 would disproportionally affect CS1 freezing. When baseline was subtracted from CS1, a line × sex interaction emerged (F_1,105 = 5.47; P = 0.021). Cued freezing in response to tone 1 remained statistically significant in males (P = 0.001) (Fig. 3C) but not in females (P = 0.88) (Fig. 3D). Fear generalization at baseline in F-S and F-R females was identical to that in their male counterparts. Potentially, the interactions between generalized fear and cued fear differ in males and females. Further study is required to allow firm conclusions to be drawn regarding sex differences in cued fear in F-S and F-R mice.

Figure 3.

F-S mice show fear generalization after conditioning. (A) Before training, there were no differences in freezing behavior in the novel training chamber. (B) After training, freezing was substantially higher in F-S mice than in F-R mice in a new (test chamber) environment. (C and D) The interaction of fear generalization was different between males and females. (C) After baseline subtraction, significant differences in the cued fear response to CS1 remained between F-S and F-R males. (D) In females, subtracting baseline freezing from the conditioned response eliminated phenotypic differences in the conditioned fear response. (A and B) F-S, n = 59; F-R, n = 58. (C and D) F-S males n = 23; F-R males, n = 33; F-S females, n = 36; F-R females, n = 25. ***P < 0.0001 and **P < 0.01 between F-S and F-R.

Regulation of the HPA axis

Morning secreted corticosterone differed between F-S and F-R mice (F-S > F-R; P = 0.004) (Fig. 4A). CRH mRNA expression was higher in F-S mice than in F-R mice (P = 0.029) (Fig. 4B), whereas expression levels of both MR and CRHR1 were lower (P < 0.001 for each) in F-S mice than in F-R mice (Fig. 4C and D). There was no significant difference in GR expression between F-S mice and F-R mice (P = 0.08; Fig. 4E). The level of GR expression relative to the level of MR expression may be particularly important, given the distinct roles of the two corticosteroid receptors.

Figure 4.

F-S and F-R mice differ in basal HPA axis activity and gene expression. (A) F-S males secreted higher levels of corticosterone in urine in the circadian trough than F-R males. (B–D) qPCR analysis of relative expression of key HPA axis regulators showed an increase in CRH mRNA (B) and decreases in CRHR1 (C) and MR (D) mRNA in the hypothalamus. Data are relative expression values adjusted to β₂-microglobulin and β-actin reference genes. (A) ELISA: F-S males, n = 7; F-R males, n = 8. (B–D) qPCR: F-S males and females, n = 30; F-R males and females, n = 51. ***P < 0.0001, **P < 0.01 and *P < 0.05 between F-S and F-R.

MEMRI

MEMRI was used to study the brain structure and function of F-S and F-S mice in vivo. Activity across a wide range of brain regions was measured, including regions associated with fear and stress. PCA was applied to the dataset to help in the visualization and interpretation of patterns of brain activity. To account for outliers, we omitted cases with PCA scores for PC1 that were ±1.5 times the interquartile range (one F-S and one F-R), and re-ran the PCA (n = 12). PCA was conducted on 12 cases and 29 brain regions, resulting in seven PCs. A statistical comparison (one-way anova) of the scores for each PC (1–7) indicated that only PC3 segregated the phenotype (F_1,10 = 23.1; P = 0.001). None of the remaining PCs (1–2 and 4–7) reached statistical significance (Table 1). These results indicate that the covariance associated with PC3 represents a unique pattern of activity that is directly related to the phenotype. We interpret the remaining PCs (1–2 and 4–7) as reflecting sources of variance unrelated to the phenotype, such as population heterogeneity of the AIL, lingering inhomogeneity in the MEMRI signal (including, perhaps, minor differences in positioning within the head coil), or non-specific uncontrollable events such as occur in a shared housing facility. Loadings associated with PC3 were examined to interpret the structure of the extracted pattern. Loadings represent the substantive importance of a given variable (ROI) to the component. Examination of the rotated component matrix indicated that the amygdala and hippocampus loaded on PC3, suggesting that neuronal activity in these areas cluster together and relate to one another (Table 1). In particular, the LA, CA2, dorsal hippocampus, dorsal CA1, ventral CA1 and ventral CA3 loaded on PC3. Reconstruction of the hippocampus from MEMRI-enhanced images identified no volumetric differences between F-S and F-R mice (Fig. 5).

Table 1.

Rotated component loading matrix

ROI	Component
	1	2	3	4	5	6	7
Central amygdala	0.868
Hypothalamus	0.831
Basal amygdala	0.783
Pallidum	0.763
Prelimbic cortex	–0.741						0.571
Lateral amygdala	0.644		.600
Ventricles	–0.616
Dorsal CA3	–0.568
Cortex	0.561
Corpus callosum		0.907
Pons		–0.847
Dentate gyrus		0.842
Ventral hippocampus		0.774
Striatum		0.611
Periaqueductal gray		0.610				0.532
Ventral CA2			0.904
Dorsal hippocampus			0.790
Ventral CA3			0.789
Ventral CA1		0.508	0.762
Dorsal CA1			0.746
Midbrain				0.919
Thalamus				0.828
White matter				0.809
Hippocampus				–0.647
Medulla				–0.541	.519
Olfactory					–0.916
Cerebellum					0.801
Pituitary						0.932
Infralimbic cortex							0.924
% variance explained = 91.9	29.4	22.6	16.3	9.3	7.9	5.7	3.9
Eigenvalues	0.039	0.030	0.021	0.012	0.010	0.007	0.005

Varimax rotation. Component loadings with an absolute value of >0.5 were suppressed. Loading values with an absolute value of >0.4 were suppressed. The coefficients were sorted according to size.

Discussion

Using bi-directional selection, we generated mouse lines with high (F-S) or low (F-R) fear expression after fear conditioning. First, we explored whether acquisition defects explained the differences between F-S and F-R mice in conditioned fear memory. Importantly, freezing in F-R mice increases with increased training intensity, indicating that the fear memory deficit is probably not caused by deficient learning but is a true fear memory phenotype. Whereas F-R mice could achieve equivalent or even greater levels of freezing during acquisition (day 1), F-R mice froze less on day 2 when the cue was presented (Fig. 2), implicating memory stabilization and consolidation processes. Before fear conditioning, freezing in a novel context was equivalent between lines (Fig. 3). After fear conditioning, however, F-S mice showed greater fear in a novel environment, suggesting either greater fear generalization or threat appraisal in F-S mice (Dunsmoor et al., 2009; Britton et al., 2011). Glucocorticoid signaling clearly impacts on fear memory formation (Thompson et al., 2004; Rodrigues et al., 2009); however, whether genetic selection for fear memory established intrinsic HPA axis differences was unknown. Morning corticosterone levels were higher in F-S mice, and hypothalamic CRH, CRHR1 and MR mRNA expression differed between F-S and F-R mice (Fig. 4). These results support the behavioral phenotypes, as corticosterone strengthens contextual fear memory (Thompson et al., 2004) and increases fear generalization (Kaouane et al., 2012). Furthermore, both the CRH system and corticosteroid receptors are implicated in psychopathology related to fearfulness, hyper-arousal, and anxiety (de Kloet et al., 1998; Gold & Chrousos, 2002; Derijk et al., 2011). Finally, we identified functional differences within the limbic fear network by using in vivo MEMRI (Drapeau & Nachshen, 1984; Silva et al., 2004). Neural activity in the LA and hippocampus was enhanced in F-S mice as compared with F-R mice (Fig. 5). These findings also align with the behavioral phenotype, as the amygdala and hippocampus are critical components of the neural circuitry underlying associative fear memory (Fig. 5) (Johnson et al., 2012).

The present results extend those of previous studies using short-term selection of fear behavior (Ponder et al., 2007), in several ways. Starting from an AIL rather than the F₂ cross means that smaller regions around the selected alleles were influenced by selection. Applying selective pressure to the AIL population changes the frequency of trait-relevant alleles without having a systematic impact on trait-irrelevant alleles, and we can be more confident that correlations are attributable to the selected alleles. Additionally, in an AIL, the frequency of all alleles should be ~0.5, which is optimal for bi-directional selection. Whereas the previous study identified mRNAs that were differentially expressed between high-fear and low-fear mice after fear conditioning, here we examined divergent traits in F-S and F-R mice. Our strategy of short-term, rather than longer-term, selection is advantageous because the frequency of trait-relevant alleles is changed with minimal random genetic drift (Belknap et al., 1997). Generating an AIL for short-term selective breeding is a superior approach, because it increases the likelihood that observed differences between the selected lines result from selection rather than drift of linkage between selected and non-selected alleles. We selected fear phenotypes based on freezing, a validated and quantifiable measure of fear behavior in rodents (Blanchard & Blanchard, 1969b; a; Fanselow, 1980). Pain sensitivity and baseline differences in locomotion potentially influence the degree of freezing after fear conditioning. Previous work in the F₂ intercross of the C57BL/6 and DBA/2 strains found no differences in shock sensitivity or locomotor activity (Ponder et al., 2007), so the behavioral phenotypes most likely relate directly to the anatomical, physiological and molecular mechanisms of fear acquisition and memory (Fig. 6).

Figure 6.

Putative model of relationships between observed physical and behavioral phenotypes. Hormonal, molecular and imaging data in F-S and F-R lines suggest that these systems may independently and/or cooperatively contribute to Pavlovian fear memory phenotypes in F-S and F-R mice. cort, corticosteroid.

The mechanisms and circuitry of fear memory consolidation established in rodent models translate directly to the study of humans and pathological fear (Johnson & Ledoux, 2004; Delgado et al., 2006). Stronger conditioned associations and generalization of fear to non-threatening stimuli are reported in PTSD (Orr et al., 2000; Peri et al., 2000), anxiety disorders (Lissek et al., 2005; Lissek et al., 2010), and depression (Nissen et al., 2010). More importantly, prospective studies in first-responders demonstrate that ‘conditionability’ predicts the psychological impact of later trauma (Guthrie & Bryant, 2006; Orr et al., 2012). F-S and F-R lines therefore provide a translational tool for identifying underlying determinants of fear memory strength as well as susceptibility and resistance to fear-related psychopathology.

Alterations in circadian corticosteroid rhythms are proposed endophenotypes of anxiety disorders, affective disorders, and PTSD (Baker et al., 2005; Wichers et al., 2008; Lok et al., 2012). In non-human primates, cortisol levels correlate with defensive freezing (Kalin et al., 1998), whereas in humans, higher basal cortisol levels are positively correlated with fear learning (Pineles et al., 2012), fear anticipation, amygdala activation (Grillon et al., 2006; Merz et al., 2012), and fear-potentiated startle (Grillon et al., 2006). In human populations, the relationship between HPA axis regulation and psychopathology is not always as straightforward (Klaassens et al., 2012; Lok et al., 2012); however, F-S and F-R lines may provide a tool for identifying traits that could prospectively identify at-risk individuals (Lok et al., 2012).

Basal cortisol levels may directly impact on neural fear networks. Resting prefrontal–amygdala connectivity inversely relates to basal cortisol levels (Veer et al., 2012), and alterations in the resting-state connectivity of the fear and emotion circuitry have been reported in pathological anxiety (Indovina et al., 2011), depression (Veer et al., 2010), and PTSD (Sripada et al., 2012). Using MEMRI, we uncovered differences between F-S and F-R lines in a well-characterized amygdala–hippocampus circuit (LeDoux, 2000; Rodrigues et al., 2009). One caveat to our interpretation is that all mice underwent fear conditioning at the end of MnCl₂ dosing. We interpreted our results as representing basal activity, because the Mn²⁺ infusion dose rate was low (1.0 μL/h) and continuous over a period of 72 h before conditioning. Slow infusion of Mn²⁺ over a period of 72 h reduces the Mn²⁺ concentration gradient at any one point in time (Mok et al., 2011). Therefore, residual Mn²⁺ in the extracellular space after infusion was effectively negligible, and not likely to contribute significantly to the final distribution pattern of Mn²⁺. Mn²⁺ enters cells through multiple Ca²⁺ transporter proteins (Drapeau & Nachshen, 1984; Hankir et al., 2012), including those implicated in fear learning-induced neuroplasticity (Rodrigues et al., 2004; Johansen et al., 2011). Therefore, MEMRI indicated more Ca²⁺-dependent activity in the fear learning circuit of F-S mice. Enhanced activity in F-S mice was predominantly localized to the hippocampus (Fig, 5). This finding is significant because of hippocampal involvement in contextual fear conditioning (Lee & Kesner, 2004; Daumas et al., 2005). There were no corresponding increases in hippocampal volume (Fig. 5), suggesting that functional differences in the hippocampus are independent of structural differences.

Differential expression of HPA axis regulatory genes probably contributes to F-S and F-R memory phenotypes. MR, but not GR, is downregulated in F-S mice relative to F-R mice. MR is primarily expressed in the hypothalamus and hippocampus, and exerts tonic control of the HPA axis through high-affinity binding of glucocorticoids (Kovacs & Mezey, 1987; Han et al., 2005). Importantly, MR is also an important contributor to fear memory formation (Zhou et al., 2010). Inhibition of MR signaling (Ratka et al., 1989; Berger et al., 2006; Brinks et al., 2009) enhances fear learning and generalization, and alters circadian corticosterone levels. Expression of GR is equivalent between F-S and F-R mice. F-S and F-R mice may therefore differ in MR/GR ratio. The relative contributions of MR–GR signaling impacts on aspects of cognition and behavior in rodents (Harris et al., 2012), and may contribute to affective disorders in humans (de Kloet et al., 1998; Qi et al., 2012).

Genes regulating the HPA axis play a role in psychopathology independently of corticosteroids. Upregulation of CRH may contribute to the elevated basal corticosterone levels found in F-S mice. However, brain CRH levels are also associated with fear and anxiety (Kalin & Takahashi, 1990), PTSD (Baker et al., 1999), and depression (Nemeroff et al., 1984; Wang et al., 2009). Conversely, mRNA for CRHR1 is decreased in F-S mice. Evidence from CRHR1 knockout mice suggests that hypothalamic CRHR1 signaling negatively regulates basal corticosterone prior to weaning (Schmidt et al., 2006), but whether this inhibitory regulation is maintained into adulthood is not clear. Alternatively, CRHR1 downregulation may be compensatory in the response to increased corticosteroid signaling (Nemeroff et al., 1988). However, CRHR1 polymorphisms and reduced CRHR1 binding are associated with psychopathology (Nemeroff et al., 1988; Gillespie et al., 2009; Wasserman et al., 2010). How CRHR1 downregulation in the hypothalamus relates to the fear memory phenotype of F-S and F-R mice is unknown, but it represents a marker for future study (Fig. 6).

Rapid selection in AILs for fear phenotypes provides a unique opportunity to isolate a coordinated network of traits establishing fear susceptibility and resistance. Recent studies have identified an increased ‘fear load’ in PTSD patients, suggesting that people with PTSD have an exaggerated memory for and poor discrimination of fearful vs. neutral cues (Felmingham et al., 2003; Jovanovic et al., 2012). In F-S mice, we identified similar behavioral differences corresponding to pre-existing differences in HPA axis regulation and limbic circuit activity between F-S and F-R lines.

Prospective studies of fear pathology in human populations are difficult to implement and interpret, expensive, and therefore infrequent. However, the identification of traits and systems predisposing individuals to strongly encode fear memory is important for developing targeted, efficacious interventions for PTSD and other fear-related disorders. The identification of fear-resistance factors may identify additional treatment strategies. Bi-modal selection from an AIL facilitates enquiry along each of these lines. These studies in naive animals have established that F-S and F-R mouse lines have robust differences in fear learning and memory and distinct physiological traits.

Supplementary Material

Supplementary Material

Doc. S1. Supplemental material.

Fig. S1.

Fig. S2. Principal components analysis (PCA) on 12 cases and 29 regions of interest (ROIs).

Abbreviations

AIL

advanced intercross line

CRH

corticotrophin-releasing hormone

CRHR1

corticotrophin-releasing hormone receptor 1

CS

conditioned stimulus

3D

three-dimensional

ELISA

enzyme-linked immunosorbent assay

F-R

fear-resistant

F-S

fear-susceptible

GR

glucocorticoid receptor

HPA

hypothalamic–pituitary–adrenal

LA

lateral amygdala

MEMRI

manganese-enhanced magnetic resonance imaging

MR

mineralocorticoid receptor

MRI

magnetic resonance imaging

PC

principal component

PCA

principal components analysis

PTSD

post-traumatic stress disorder

qPCR

quantitative polymerase chain reaction

ROI

region of interest

S1

selected generation 1

S2

selected generation 2

S3

selected generation 3

S4

selected generation 4

SI

signal intensity

USUHS

Uniformed Services University of the Health Sciences