Identification of candidate genes that underlie the QTL on chromosome 1 that mediates genetic differences in stress-ethanol interactions

Abstract

Alcoholism, stress, and anxiety are strongly interacting heritable, polygenetic traits. In a previous study, we identified a quantitative trait locus (QTL) on murine chromosome (Chr) 1 between 23.0 and 31.5 Mb that modulates genetic differences in the effects of ethanol on anxiety-related phenotypes. The goal of the present study was to extend the analysis of this locus with a focus on identifying candidate genes using newly available data and tools. Anxiety-like behavior was evaluated with an elevated zero maze following saline or ethanol injections (1.8 g/kg) in C57BL/6J, DBA2J, and 72 BXD strains. We detected significant effects of strain and treatment and their interaction on anxiety-related behaviors, although surprisingly, sex was not a significant factor. The Chr1 QTL is specific to the ethanol-treated cohort. Candidate genes in this locus were evaluated using now standard bioinformatic criteria. Collagen 19a1 (Col19a1) and family sequence 135a (Fam135a) met most criteria but have lower expression levels and lacked biological verification and, therefore, were considered less likely candidates. In contrast, two other genes, the prenylated protein tyrosine phosphate family member Ptp4a1 (protein tyrosine phosphate 4a1) and the zinc finger protein Phf3 (plant homeoDomain finger protein 3) met each of our bioinformatic criteria and are thus strong candidates. These findings are also of translational relevance because both Ptp4a1 and Phf3 have been nominated as candidates genes for alcohol dependence in a human genome-wide association study. Our findings support the hypothesis that variants in one or both of these genes modulate heritable differences in the effects of ethanol on anxiety-related behaviors.

in recent years, alcohol abuse and alcoholism have been shown to have a heritable component, although the underlying genetic causes have been difficult to identify. Alcohol addiction is a complex, polygenic disease that involves the interplay between many genetic, physiological, and environmental cofactors. Stress and anxiety are factors that have been shown to contribute to alcohol use and abuse (13, 44), although the nature of these interactions are not clearly defined and the genetic factors that influence these interactions are currently not well known. The self-medication hypothesis theorizes that the relationship between anxiety/stress and alcoholism can be causal, suggesting that the presence of one disorder contributes to the development of the other (6, 22). Additional studies have indeed shown that people with higher stress and anxiety levels are more likely to become alcoholics (50). Stress and anxiety also have strong genetic contributions; therefore, it is important to identify specific genes and pathways that mediate the interactions between these factors.

Due to the complexity of human genetic variation, animal models have been used to further understand and facilitate the identification of relevant candidate genes. One highly used animal model is the panel of BXD recombinant inbred (RI) mouse strains, which was derived from C57BL/6J (B6) and DBA/2J (D2) progenitors. The BXD genetic reference population has been completely genotyped and consists of ∼150 lines, which has made it favorable for mapping quantitative trait loci (QTL) including alcohol- and stress-related phenotypes (4, 11, 14, 24, 29, 30, 34, 45). Although studies have identified numerous loci that mediate genetic differences in alcohol- and stress-related phenotypes, few of the specific genes or genetic pathways have been identified, nor are pathways that mediate interactions between these two factors known.

Recently, we tested mice from BXD RI lines as well as another RI line, the LXS made by crossing the long-sleep and short-sleep lines, in an elevated zero maze following ethanol or saline treatment (51). This previous study identified a number of QTLs across these different populations. One QTL on chromosome (Chr) 1 at 23–31.5 Mb was observed only in ethanol-treated animals and mediated several phenotypes related to activity in the closed arms of the elevated zero maze. In the present analyses, we focused on these specific phenotypes and the QTL that underlies this differential behavior because of 1) the consistency across multiple measures, 2) the strength of the QTL, and 3) the utility in identifying genetic components involved in ethanol-and anxiety-related phenotypes that may aid in understanding the stress/anxiety/alcohol relationship. Specifically, we further evaluated the phenotypic data, including addressing the role of sex on these ethanol-related phenotypes. Subsequently, we concentrated on identifying potential candidate genes within the QTL using sequential steps in systems genetic analyses to isolate the genes that met specific criteria. This study was facilitated by a number of recently developed bioinformatic tools, sequencing results, and expression databases created by us and our colleagues that are publically accessible at GeneNetwork (http://www.genenetwork.org). Similar bioinformatic analyses have successfully identified candidate genes for a number of other phenotypes including hypertension, iron regulation, and central pattern generators (CPGs) by us and other investigators (4, 7, 16, 19, 24, 28, 29, 33, 34, 46, 47).

MATERIALS AND METHODS

Mice.

Male and female mice from the expanded BXD family of strains were used in all studies. As described in the previous study (51), all animals were bred at the University of Tennessee Health Science Center (UTHSC) animal facility. Animals used for behavioral testing were transferred to the University of Memphis, and following a 1 wk acclimation period, behavioral testing was conducted. Separate groups of animals were used for transcriptome studies and remained at UTHSC. At both locations, animals were group-housed and maintained on a 12 h:12 h light/dark cycle and given free access to food and water. All procedures involving animals were approved by the Animal Care and Use review boards of the UTHSC and the University of Memphis.

Behavioral phenotypes.

B6, D2, reciprocal F1s, and 72 BXD strains were used in the behavioral studies. For each treatment condition, there was an average of five or six animals of each sex from every strain. In the ethanol-treatment group, mice were injected with 1.8 g/kg ip ethanol. In the control group, mice were given an isovolumetric injection of saline. Five minutes after injection of saline or ethanol, mice were placed in an elevated zero maze. The elevated zero maze has been described in detail elsewhere (10, 51) and is commonly used to measure anxiety-related behaviors in mice (21). Each animal was only tested in the elevated zero maze once to avoid any influences of prior exposure to the maze on anxiety levels. Furthermore, because animals within an RI strain are genetically identical, values from saline-treated animals can be used as “baseline” measures for ethanol-treated animals. The following measures were obtained for all animals: 1) activity in closed quadrants (ACQ) for the first 5 (0–5) min, 2) ACQ for the last 5 (5–10) min, and 3) ACQ during the total 10 (0–10) min (see Table 1, Table 2).

Table 1.

Ethanol and saline phenotypes in the closed arms of the elevated zero maze, with the number of strains used, and the value and location of the maximal LRS

Trait ID	Description	n	Max LRS	Max LRS Location Chr & Mb
12389	ethanol treated during first 5 min (0–5 min)	74	13.2	Chr 1: 25.489364
12390	ethanol treated during last 5 min (5–10 min)	74	17.6	Chr 1: 25.632282
12391	ethanol treated during total 10 min (0–10 min)	74	17.0	Chr 1: 25.632282
12419	saline treated during first 5 min (0–5 min)	75	20.1	Chr 1: 157.588921
12420	saline treated during last 5 min (5–10 min)	75	21.7	Chr 1: 156.052563
12421	saline treated during total 10 min (0–10 min)	75	24.4	Chr 1: 156.052563

Trait IDs can be used to find data on GeneNetwork (genenetwork.org). LRS, likelihood ratio score; Chr, chromosome.

Table 2.

Each of the 3 ethanol-related phenotypes

Pearson	Spearman Rank Correlation (rho)
Ethanol Group
	12389: 0–5 min	12390: 5–10 min	12391: 0–10 min
12389: 0–5 min		0.809	0.937
12390: 5–10 min	0.809		0.955
12391: 0–10 min	0.948	0.954
Saline Group
	12419: 0–5 min	12420: 5–10 min	12421: 0–10 min
12419: 0–5 min		0.744	0.928
12420: 5–10 min	0.716		0.929
12421: 0–10 min	0.929	0.924

Each of the 3 ethanol-related phenotypes were significantly correlated, and similarly, the 3 saline-related phenotypes are also strongly correlated, as shown by Pearson's (lower left) and Spearman rank (upper right) correlation (see Table 1 for trait ID descriptions). Because of this, the ethanol phenotype activity in closed quadrants (ACQ) 5–10 min (12390) and the saline phenotype ACQ 5–10 min (12420) were chosen to represent the ethanol and saline groups, respectively in subsequent analyses. For each group n = 74.

Statistical analysis.

Data were evaluated with a three-variable (strain, sex, treatment), multifactor (ACQ 0–5, ACQ 5–10, ACQ Tot) ANOVA. Pearson correlation coefficients were used to determine significant correlations between measures.

Database descriptions.

Expression data sets were accessed through GeneNetwork, a public web source used to study relations among markers, genes, and phenotypes. We made use of large transcriptome data sets for the amygdala, hippocampus, ventral tegmental area (VTA), hypothalamus, pituitary, and adrenal glands. These tissues were selected because of their well-known involvement in anxiety/stress-related and ethanol-related phenotypes (3, 5, 12, 36, 40, 41, 49). Specific data sets used were “INIA Adrenal Affy MoGene 1.0ST (Jun12) RMA,” “INIA Amygdala Cohort Affy MoGene 1.0 ST (Mar11) RMA,” “Hippocampus Consortium M430v2 (Jun06) RMA,” “Hypothalamus Affy MoGene 1.0 ST (Nov10),” “INIA Pituitary Affy MoGene 1.0ST (Jun12) RMA,” and “VCU BXD VTA EtOH M430 2.0 (Jun09) RMA.” Expression data were analyzed from 50 to 99 BXD strains for each data set. Detailed information and access to all data is available at http://www.genenetwork.org. Genome coordinates are based on the mm9 UCSC genome assembly.

QTL mapping.

All QTL mapping for the phenotypes was conducted with interval-mapping software and genotypes in GeneNetwork.org. We report loci achieving genome-wide significance (P < 0.05) based on 2,000 or more permutation tests. Linkage is reported in terms of the likelihood ratio statistics (LRS). All strain data (means, SE, and sample sizes) are publicly available at GeneNetwork.org (see Table 1).

Criteria for identification of candidate genes.

We employed a multistep process to identify candidate genes associated with behavioral differences. All analyses, except the literature analyses, were done using tools in GeneNetwork. The steps were as follows: 1) We defined the position of candidate genes as all genes falling within a 1.5 LOD support interval of the QTL. 2) We downgraded genes with low expression levels in both the Allen Brain Atlas and in multiple gene expression data sets in GeneNetwork. 3) Next, we identified those genes/transcripts with expression variation across BXD strains that correlated with the behavioral traits in one of the relevant regions. Correlations coefficients were computed between each of the behavioral traits and expression of each gene in each of the six brain regions and the adrenal gland. 4) We then determined whether these candidates were associated with significant cis-eQTLs in one or more of the tissues. 5) Candidates were evaluated for the presence of polymorphisms between B and D parental haplotypes. 6) Finally, literature reviews were then conducted to determine the biological function(s) of the genes in relation to our traits of interest.

RESULTS

Behavioral phenotypes.

Male and female BXD RI strains, as well the parental strains and F1s, were evaluated in an elevated zero maze following ethanol or saline administration to investigate the effects of ethanol on measures of anxiety. These analyses focused primarily on ACQ, as the QTL identified in our previous study was associated with this activity (51). Strain-specific differences in activity were observed both in the control, saline-treated mice and in the ethanol-exposed mice in each of the 5 min time bins and the total 10 min (Fig. 1). To evaluate the interaction between treatment, strain, and sex, a three-way ANOVA was performed on the data for the total 10 min (ACQ Tot) of the behavioral test. The results show a significant effect for strain, F(75, 1,240) = 4.744 P < 0.0001, and treatment, F(1, 1,240) = 45.096 P < 0.0001. A significant strain by treatment interaction, F(73, 1240) = 3.617 P < 0.0001, was found, as seen in Fig. 1, which shows that the pattern of activity in the two treatment groups is not the same among strains, suggesting differential regulation across strains. The effect of sex was tested as well; however, there was no significant effect; F(1, 1240) = 0.011, P = 0.915. Due to the lack of sex differences, males and females were grouped in subsequent analyses.

Fig. 1.

Strain-specific differences in activity in the closed quadrants (ACQ) of the elevated zero maze. Activity was measured by amount of time spent in the closed quadrants of the elevated zero maze across 72 BXD, parental B6 and D2, and F1 strains. Males and females are combined in each strain. Strains are in order by name on the x-axis, and the average number of beam breaks (±SE) ACQ is found on the y-axis of the histogram. For some strains, variance was so small such error bars are not visible. For several other strains, we were only able to test 1 animal per group. Quantitative trait locus (QTL) analyses were run both with and without the strains with n = 1, and the location of the QTLs remains the same. A: activity for ethanol-exposed mice during the last 5 min (5–10 min). The GeneNetwork ID for this trait is 12390. B: activity for control, saline-treated mice during the last 5 min (5–10 min). The GeneNetwork ID for this trait is 12420.

Strain differences observed after saline treatment could confound QTLs identified following ethanol treatment; therefore, we: 1) examined the QTLs separately under each condition as discussed below and 2) conducted analyses of derived values that took into account basal (saline) differences in activity. Animals within an RI strain are isogenic; therefore, activity means for each strain following saline exposure were subtracted from the activity scores of the ethanol-treated mice of each respective strain. Strain differences persisted following analyses of the derived values (data not shown). The derived activity values were also used in QTL analyses as described below.

QTL mapping.

Interval mapping was performed to identify genetic loci contributing to activity-related measures of anxiety. Data were evaluated from each of the two 5 min bins, as well as for the total 10 min test. As shown in Tables 1 and 2 and Fig. 1, the results within conditions (saline or ethanol) are consistent across these three measures. A significant activity-related QTL in control, saline-treated animals mapped to Chr 1 at 155 ± 5 Mb (see Fig. 2A). This QTL region overlaps those found in a number of other studies using a range of anxiety-related assays including the elevated-plus maze and the open-field test (4, 14, 34, 45, 51). In contrast, a significant activity-related QTL (LRS = 18.7) in ethanol-treated mice mapped to Chr 1 between 23 and 31.5 Mb, as seen in Fig. 2, B–D. These distinct QTL regions suggest that the QTL on proximal Chr 1 specifically mediates responses following ethanol exposure.

Fig. 2.

Ethanol-specific QTLs were identified on proximal chromosome (Chr) 1, while a QTL for saline values was identified on distal Chr 1. QTLs for the control, saline-treated mice mapped to distal Chr1 at 155–163.5 Mb, whereas the QTLs for ethanol-exposed mice mapped on proximal Chr1 at 23- 31.5 Mb. The x-axis shows the megabases (Mb), and the y-axis shows the likelihood ratio score (LRS). For A and B, all 3 phenotypes are shown on the same plot with red lines denoting ACQ during first 5 min (0–5 min), green lines denoting ACQ during last 5 min (5–10 min), and blue lines denoting ACQ for the total 10 min (0–10 min). A: focused examination of the Chr1 QTL for control, saline-treated mice. B: focused examination of the Chr1 QTL for ethanol-treated mice. C: close-up view of the 1.5 LOD region on Chr 1, 23–31.5 Mb, including gene indicators for the top 4 candidate genes. D: QTL of ethanol-exposed mice during the last 5 min (5–10 min) on proximal Chr 1. The lower, gray horizontal line marks the suggestive LRS level (P < 0.63) and the upper, pink horizontal line marks the significant LRS level (P < 0.05). The blue trace denotes the LRS score of this ethanol phenotype, which exceeds the significant line on Chr1 with a score of 17.6.

As discussed above, an additional analysis, using derived values (ethanol − saline), was conducted to determine whether basal locomotor activity impacts QTLs identified following ethanol exposure. Interestingly, derived activity values also mapped to proximal Chr 1 (23–35 Mb), providing further support that this QTL is specific to ethanol-induced activity.

Candidate gene identification.

Using available data within GeneNetwork, we were able to identify candidate genes in the proximal Chr 1 QTL region. In addition, a number of expressed sequence tags (ESTs) also mapped to this region; however, these were not further evaluated due to either the lack of expression data or lack of information on their functional significance.

Strict criteria were used to evaluate the candidate genes. Initially, genes with mean expression levels <7.0, i.e., lower than background, were eliminated from further consideration. We restricted our expression analyses to brain regions relevant to anxiety/stress- and ethanol-related phenotypes, including the amygdala, hippocampus, hypothalamus, pituitary gland, and VTA, as well as the adrenal gland (Fig. 3).

Fig. 3.

The top gene candidate, protein tyrosine phosphate 4a1 (Ptp4a1), correlates with ethanol phenotypes in the closed arms of the elevated zero maze. The ACQ during the last 5 min is shown as an example. Expression of Ptp4a1 in the hippocampus (A; n = 62, r_s = −0.282, P ≤ 0.05), amygdala (B; n = 46, r_s = −0.484, P ≤ 0.001), adrenal gland (C; n = 48, r_s = −0.269, P ≤ 0.05), and ventral tegmented area (VTA) (D; n = 32, r_s = −0.364, P ≤ 0.05) was correlated using a Spearman's correlation coefficient. Brain region expression is on the x-axis, while phenotypic readout is on the y-axis.

Although there is evidence that the aforementioned brain regions are involved in anxiety/stress- and ethanol-related phenotypes, the role of each region in the behavioral phenotypes of interest is not clear. Therefore, to assess whether the candidate genes had similar function(s) in each of these regions, we assessed whether their expression was correlated across these five brain regions and the adrenal gland. However, there was no significant correlation across these regions (data not shown). This lack of correlation across regions was used to define criteria in the subsequent analysis, which assessed the correlation of gene expression for each candidate in each region with each of the three phenotypes of interest. Because the expression levels of the genes were not correlated across regions, the criterion used was that a candidate must correlate with the behavioral phenotypes in at least one region, rather than all six.

Subsequently, we mapped the expression quantitative trait loci (eQTLs) for each gene to determine if each gene is cis- or trans-regulated in each of the relevant brain and body regions. Cis-regulated eQTLs are caused by a sequence variant located within 5 Mb of the gene's location. Genes that are cis-regulated are more likely to control their own expression and are also likely to have downstream effects on the expression of other genes and/or phenotypic readouts (9). Trans-regulated eQTLs, on the other hand, are regulated by sequence variants >5–10 Mb from the gene's location or potentially on a different chromosome. From this analysis, the following six genes remained viable candidates: opioid growth factor receptor-like 1 (Ogfrl1), beta-1, 3-glucuronyltransferase 2 (B3gat2), family with sequence similarity 135, member a (Fam135a), collagen 19a1 (Col19a1), plant homeodomain finger protein 3 (Phf3), and protein tyrosine phosphate 4a1 (Ptp4a1). The genes showed differential regulation across the brain and body regions examined. Ogfrl1 and Ptp4a1 are cis-regulated in the amygdala, hippocampus, and VTA (Fig. 4). B3gat2 is cis-regulated sequences in the amygdala. Fam135a and Col19a1 are cis-regulated in the pituitary gland. Phf3 is cis-regulated in the hippocampus. Two genes were cis-regulated in multiple regions: Ogfrl1 and Ptp4a1, while other genes were only cis-regulated in one region: B3gat2, Fam135a, Col19a1, and Phf3.

Fig. 4.

Ptp4a1 expression quantitative trait loci (eQTLs), demonstrating that this gene is cis-regulated in a subset of the relevant brain regions. Chromosome number can be found across the top of the plot with megabases (Mb) at the bottom on the x-axis. The y-axis contains the LRS. The location of Ptp4a1 is marked by an arrowhead found on the x-axis and is at Chr 1 at 30.997 Mb. A: amygdala (LRS = 26.7), B: hippocampus (LRS = 74.6), C: VTA (LRS = 37.2).

Polymorphism analysis.

Finally, we evaluated whether the remaining genes contained polymorphisms and whether these polymorphisms were in coding regions of the genes or regulatory regions. Of the six remaining candidate genes, two were discarded because they lacked any relevant polymorphisms. Genes lacking polymorphisms are unlikely to be involved in the regulation of differences in our behavioral phenotypes.

Using the aforementioned criteria, we retained only four candidate genes, Ptp4a1, Phf3, Col19a1, and Fam135a, for further analyses (Table 3). Ptp4a1, Phf3, and Fam135a have significantly high mean expression levels in all of the relevant regions, while Col19a1 is moderately high, as shown in Table 3. These four candidate genes also have significant correlations with the relevant phenotypes (Table 3, Fig. 3) and, as discussed below, have missense polymorphisms and therefore are viable candidate genes (Table 4).

Table 3.

Expression mean and range, correlation with behavioral phenotype, and the presence of cis-regulation for the top 4 candidate genes in the amygdala, hypothalamus, pituitary, hippocampus, VTA, and adrenal gland

	Fam135a	Col19a1	Phf3	Ptp4a1
Amygdala
Expression	9.151	7.461	10.322	9.335
(Range)	(8.88–9.35)	(7.26–7.65)	(10.12–10.48)	(8.912–9.78)
Phen correlate1	NS1	0.0510	NS1	0.0005
eQTL location	none2	Chr5: 43.571	none2	cis-eQTL
(LRS)2		(18.9)		(26.7)
Hypothalamus
Expression	9.111	6.872	10.225	9.111
(Range)	(8.73–9.38)	(6.69–7.07)	(9.83–10.37)	(8.134–9.73)
Phen correlate1	NS1	NS1	NS1	0.0005
eQTL location	none2	none2	none2	Chr19: 4.609
(LRS)				(17.5)
Pituitary Gland
Expression	9.431	6.957	10.806	9.436
(Range)	(9.15–9.67)	(6.73–7.21)	(10.64–10.97)	(8.98–10.09)
Phen correlate1	0.0453	NS1	NS1	NS1
eQTL location	cis-eQTL	cis-eQTL	none2	none2
(LRS)2	(20.3)	(23.8)
Hippocampus
Expression	9.639	9.302	11.184	10.700
(Range)	(8.71–10.31)	(8.33–10.00)	(10.94–11.46)	(9.95–11.22)
Phen correlate1	NS1	0.0351	0.0461	0.0104
eQTL location	none2	Chr1: 176: 535	cis-eQTL	cis-eQTL
(LRS) 2		(18.4)	(14.4)	(74.6)
VTA
Expression	10.007	8.120	8.220	11.073
(Range)	(9.61–10.23)	(7.62–8.99)	(7.58–8.72)	(10.75–11.29)
Phen correlate1	NS1	0.0013	0.0741	0.0015
eQTL location	none2	none2	none2	cis-eQTL
(LRS) 2				(37.4)
Adrenal Gland
Expression	9.164	7.029	10.663	10.277
(Range)	(8.778–9.56)	(6.71–7.30)	(10.50–11.07)	(9.55–11.38)
Phen correlate1	NS1	NS1	NS1	0.0317
eQTL location	none2	none2	none2	Chr2: 55.266
(LRS) 2				(14.1)

Expression mean (“Expression”) and range (“Range”), correlation with behavioral phenotype (“Phen correlate”) and the presence of cis-regulation (“cis present”) for the top 4 candidate genes in the amygdala, hypothalamus, pituitary, hippocampus, ventral tegmental area (VTA), and adrenal gland. The phenotype examined was 12390, which is ACQ during the last 5 min (5–10 min). For the phenotype correlations the P value is shown. ¹Shown are all significant P values (< 0.05) as well as those that have a trend toward significance (0.05- 0.1). All correlations that did not reach significance are denoted with NS. ²For the expression quantitative trait locus (eQTL) analyses we list eQTL LRSs that are 14 or above; eQTL LRSs that are <14 are denoted by none. Trans-eQTLs list the location, while the cis-eQTLs is located within 5 Mb of the gene itself.

Table 4.

Location of the genes and type of polymorphism for the top 4 candidate genes

Symbol	Chr: Mb	Type of Polymorphism
Fam135a	1: 24.017934	nonsynonymous
Col19a1	1: 24.264522	nonsynonymous
Phf3	1: 30.859187	codon deletion
Ptp4a1	1: 30.997148	copy number variant

The biological relevance of the four candidate genes were examined on online databases. Ptp4a1, Phf3, and Col19a1 or their families have been shown to function in the central nervous system (CNS), whereas Fam135a does not. In summary, our four candidate genes, Ptp4a1, Phf3, Fam135a, and Col19a1, contain nonsynonymous sequence polymorphisms, copy number variants (CNVs), or codon deletions with two having strong functional significance in the CNS (see discussion).

DISCUSSION

In the present study, we sought to expand our previous work on genetic factors underlying anxiety-related behaviors (51), with a focus on identifying candidate genes mediating ethanol-induced activity in the closed arms of the elevated zero maze. A drawback of QTL research is that it is often an arduous task to go from the QTL to the quantitative trait gene (QTG). Improvements in systems genetic analyses, as well as the development of numerous, well-populated databases, has refined QTG research methods. This methodology has been successfully used in this lab to identify numerous QTLs, as mentioned in the introduction (4, 7, 16, 19, 24, 28, 29, 33, 34, 46, 47). A QTL on Chr 1 at 23–31.5 Mb specific to ethanol-related activity was identified. The present study also used an innovative, systems genetic method to identify QTGs underlying this QTL. Several bioinformatic databases were used, in conjunction with a set of strict criteria, specifying gene expression levels, correlations of expression with behavioral phenotypes, and sequence polymorphisms, to identify candidate genes. We were successful in identifying four candidate genes, with two considered as optimal candidates based on further analyses.

In the present study we examined sex differences, and because we used a large number of strains, we were able to examine the sex-by-strain interaction. Significant sex differences were not found in the present study; however, there was a moderately significant sex-by-strain interaction although this could be due to the strong effects of strain. While sex differences have been found in other measures of anxiety (17, 26, 32, 52), as well as in a wide range of ethanol-related phenotypes (2, 26, 28, 37, 39), to date little is known about how these factors interact. The lack of an effect of sex in the present study is consistent with a previous report that did not show sex-specific differences on the effects of ethanol on anxiety-related behaviors tested in a single strain of rats (48), suggesting that sex may not be an important variable in ethanol-stress interactions.

It is known that the amygdala, hypothalamus, pituitary gland, hippocampus, VTA, and the adrenal gland are involved in anxiety- and alcohol-related traits; however, the specific role(s) each of these regions plays in these phenotypes is/are unclear (2, 5, 12, 36, 40, 41, 49). Therefore, whether expression of a particular candidate gene in one region is a more critical criterion than expression in another region is unknown. To address this, we initially assessed whether expression of each gene was correlated across the different anatomical regions and showed that the correlations were low. As discussed below, expression of three of the four final candidate genes was correlated with the relevant behavioral phenotypes in multiple regions. This suggests that this type of analyses could be useful not only in elucidating the relevant genetic pathways, but also to uncovering clues to the neuronal substrates that underlie specific behavioral phenotypes.

One caveat of the present study is that the expression data were evaluated in untreated, control mice and not in ethanol-exposed mice. However, because strain-specific behavioral differences are apparent only 5 min after ethanol injection, it suggests that baseline expression differences are likely more relevant to the behavioral differences than ethanol-induced expression changes.

Candidate genes.

Two genes lacked polymorphism and therefore are less likely to be candidates mediating alcohol- and anxiety-related phenotypes: Ogfrl1 and B3gat2. Ogfrl1 is involved in neuronal development and has been shown to contribute to synapse formation and axonal growth and guidance in developing neurons (31). B3gat2 codes for a transmembrane protein that catalyzes the transfer of glucuronic acid to a galactose in different glycoproteins or glycolipids and is important in the lymphoid system (42). In the nervous system, B3gat2 functions in adhesion and has been implicated as a risk gene for schizophrenia as analyzed by a genome-wide association study (15, 18).

The genes that met all of our selection criteria and thus are the most promising candidate genes mediating alcohol/anxiety-related phenotypes are Ptp4a1, Phf3, Col19a1, and Fam135a. All four genes are expressed in relevant brain and body regions. Importantly, the genes were also significantly correlated with the behavior and were predominately cis-regulated (Table 3). All genes are polymorphic with nonsynonymous single nucleotide polymorphisms (SNPs), or CNVs or codon deletions.

Previous studies were evaluated to ascertain the biological relevance of Ptp4a1, Phf3, Col19a1, and Fam135a to better understand their role(s) in the effects of ethanol on measures of anxiety. To our knowledge, Fam135a is a gene of unknown function, although it has been shown to be differentially expressed in the hippocampus in mice following exposure to the norepinephrine reuptake inhibitor, nortriptyline (25). Although not much is known about Col19a1, collagen is known for its role as the main structural protein in connective tissue and has been shown to be involved in myelination (8, 38). Myelin deficiency has been documented in the frontal cortex of in the brains of alcoholics, and alcohol may also influence the process of myelin maintenance and repair (20, 23). While these studies show that collagen is present in myelin and that alcohol effects myelination, whether Col19a participates in these processes is unknown. Although we cannot eliminate these genes as possible candidates, the limited information relating these specific genes to alcohol- and anxiety-related phenotypes diminishes their viability.

Ptp4a1 is a member of the prenylated protein tyrosine phosphate family. The Ptp4a1 protein is a cell signaling molecule that is involved in a variety of cellular processes. Overexpression of this gene is associated with diseases such as cancer (43). Ptp4a1's role in alcohol- or anxiety-related phenotypes is not well understood. However, it has been shown, in mice, that Ptp4a1 expression in the prefrontal cortex is significantly regulated by ethanol exposure (20). Phf3 is a member of the zinc finger protein superfamily, which are regulatory proteins in the nucleus and cytoplasm and are associated with chromatin-mediated translational regulation (1, 35, 53). It has been reported that expression of Phf3 is altered in alcoholics although, it could not be determined whether this was due to basal differences in gene expression prior to alcohol exposure (23).

Of particular relevance to our findings, Zuo and colleagues (53, 54, 55) recently identified Ptp4a1 and Phf3 as candidate genes for alcohol dependence in human genome-wide association studies. Zuo and colleagues demonstrated that Ptp4a1 and Phf3 were the only genes that were linked to alcohol dependence across three different human populations with 12 SNPs within this region associated with alcohol dependence across all three populations. Moreover, these authors demonstrated that these genes are cis-regulated and correlated with a number of other genes related to alcohol addiction. Further analyses confirmed that the linkage of these genes is specific to alcohol dependence and not related to other psychiatric disorders (53, 54, 55). The fact that we identified these same genes in a murine model provides cross-species support for Ptp4a1 and Phf3 as possible risk genes for some alcohol- and anxiety-related phenotypes.

In summary, the present study identified Ptp4a1 and Phf3 as optimal QTG candidates underlying the QTL on Chr 1 at 23–31.5 Mb, mediating differences in ethanol-induced anxiety responses. Future studies will look at confirming the role of these genes and their genetic and biological pathways in alcohol- and anxiety-related phenotypes. To date, no connection has been found between Ptp4a1 or Phf3 and anxiety/stress-related pathways. Therefore, it will be of interest to determine if these molecules are members of a novel genetic pathway in which anxiety and ethanol interact.

GRANTS

This project was supported by National Institute on Alcohol Abuse and Alcoholism Grants U01 AA-014425 (L. Lu), INIA U01 AA-13499 and U01 AA-16662 (R. W. Williams), U01 AA-016667 (Michael Miles); by National Natural Science Foundation of China Grant 30971591 (L. Lu), and by UTHSC Bridge funding (L. Lu).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: M.N.C., R.W.W., and L.L. conception and design of research; M.N.C. and L.L. performed experiments; M.N.C., J.A.B., S.A.H., K.M.H., and L.L. analyzed data; M.N.C., J.A.B., S.A.H., K.M.H., and L.L. interpreted results of experiments; M.N.C., J.A.B., S.A.H., R.W.W., K.M.H., and L.L. edited and revised manuscript; M.N.C., J.A.B., S.A.H., R.W.W., K.M.H., and L.L. approved final version of manuscript; J.A.B., K.M.H., and L.L. prepared figures; J.A.B., K.M.H., and L.L. drafted manuscript.