Potential translational targets revealed by linking mouse grooming behavioral phenotypes to gene expression using public databases

Abstract

Rodent self-grooming is an important, evolutionarily conserved behavior, highly sensitive to pharmacological and genetic manipulations. Mice with aberrant grooming phenotypes are currently used to model various human disorders. Therefore, it is critical to understand the biology of grooming behavior, and to assess its translational validity to humans. The present in-silico study used publicly available gene expression and behavioral data obtained from several inbred mouse strains in the open-field, light-dark box, elevated plus- and elevated zero-maze tests. As grooming duration differed between strains, our analysis revealed several candidate genes with significant correlations between gene expression in the brain and grooming duration. The Allen Brain Atlas, STRING, GoMiner and Mouse Genome Informatics databases were used to functionally map and analyze these candidate mouse genes against their human orthologs, assessing the strain ranking of their expression and the regional distribution of expression in the mouse brain. This allowed us to identify an interconnected network of candidate genes (which have expression levels that correlate with grooming behavior), display altered patterns of expression in key brain areas related to grooming, and underlie important functions in the brain. Collectively, our results demonstrate the utility of large-scale, high-throughput data-mining and in-silico modeling for linking genomic and behavioral data, as well as their potential to identify novel neural targets for complex neurobehavioral phenotypes, including grooming.

1. Introduction

Large scale, high-throughput data-mining and data integration are rapidly becoming key methods for scientific discovery (Tabakoff et al., 2009, Xuan et al., 2010), emphasizing the importance of sharing biological data (Akil et al., 2011, Sears et al., 2006). Integration of behavioral phenotypes with neural and genomic data, such as phenomics and ‘genetical genomics’, is emerging as a promising strategy for the dissection of complex gene-behavior interactions (Bennett et al., 2011, Bhave et al., 2007, Tabakoff et al., 2007). Among the behavioral phenotypes, self-grooming is especially important, because it represents an evolutionarily ancient behavior with multiple biological functions (from hygiene to stress reduction) and a complex, patterned nature (Chen et al., 2010, Fentress, 1988, File et al., 1988, Sachs, 1988, Spruijt et al., 1992). In rodents, grooming is one of the most frequently occurring behaviors, often correlating with the levels of arousal (Fentress, 1968, 1977, 1988) and various anxiety-like behaviors (Denmark et al., 2010, Kalueff and Tuohimaa, 2005a, c, Kyzar et al., 2011). Mounting evidence shows the value of analyzing grooming as a behavioral endpoint following genetic or pharmacological manipulations in experimental models of various brain disorders (Audet et al., 2006, Chen, Tvrdik, 2010, Estanislau, 2012, Greer and Capecchi, 2002, Kalueff et al., 2004, Kalueff and Tuohimaa, 2005c).

While mouse self-grooming is an important behavioral domain, little is known about its genetic architectonics or genomic correlates (Bergner et al., 2010). Established in 2000, the Mouse Phenome Database (MPD) is a publicly available platform, providing phenotypic data on different mouse strains (Grubb et al., 2009). While the MPD initially lacked mouse grooming data, it now contains reports on grooming frequency in A/J, C57BL/6J, consomic (Lake et al., 2005) and wild-derived strains (Koide and Takahashi, 2006), as well as grooming duration from multiple inbred strains in several anxiety tests (Brown et al., 2004). Comparison of these data with other behaviors using the MPD online tools has revealed correlations with anxiety-sensitive behaviors, reflecting the importance of measuring grooming in animal anxiety paradigms (Crawley, 2007, Hart et al., 2010, Kalueff and Tuohimaa, 2005b).

The present study aimed to examine the potential link between mouse grooming behavior and the expression of selected genes within the brain. This study also demonstrates the utility of large-scale data-mining and in-silico (computer-based) modeling for linking genomic and behavioral data, and its potential to identify new neural targets for specific phenotypes of interest. Using mouse grooming as a representative phenotype, this proof-of-concept study can be applied in future research to other mammalian behavioral and physiological phenotypes.

2. Methods and Results

2.1. General overview

To achieve the goals of this study, we utilized the MPD data containing behavioral phenotypes (Brown, Gunn, 2004) and whole-brain genomic microarray results (Tabakoff, Bhave, 2007); see (Bennett, Saba, 2011, Stewart et al., 2011) for the conceptual framework. We first used the MPD tools to identify significant correlations for grooming and gene expression data across four widely used behavioral paradigms (open-field test, elevated plus-maze, elevated zero-maze and light-dark box); Fig. 1. We next ranked these candidate genes in terms of the strength of their correlations with grooming behaviors, identifying a sub-group of genes whose expression within the brain most strongly correlated with grooming phenotypes. We then used the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) (Szklarczyk et al., 2011) and Cytoscape tools to create interactome networks for both the selected genes and their human orthologs. Based on the overlap between these networks, we identified 31 candidate genes with translational potential, and determined their associated biological roles within the brain using GoMiner and Mouse Genome Informatics tools (Shaw, 2009). Following identification of candidate genes, we completed in-silico validation of this approach by comparing patterns of expression in the candidate genes to neutral ‘control’ genes (chosen by selecting probes at random). Compared to the control genes, the candidate genes were more strongly correlated with grooming behavior, and showed altered expression levels in key brain regions involved in grooming using the Allen Brain Atlas (ABA) (Lein et al., 2007, Ng et al., 2009). These candidate genes also produced more nodes per gene within interactome networks, thus demonstrating the usefulness of this in-silico phenotype-genomic methodology.

Figure 1. Flowchart summarizing the methodology of selecting and analyzing candidate grooming genes.

During Phase I, the Mouse Phenome Database (MPD) brain expression microarray data and behavioral data were selected from several murine anxiety paradigms and a wide spectrum of mouse strains. Step 1 used the MPD toolbox to generate Pearson correlation coefficients between these two data sets yielding 844 mouse genes that correlated with self-grooming (P<0.005). In step 2, this list of genes was prioritized by 1) how many sets of grooming data from different behavioral paradigms a gene was correlated with, and 2) the strength of their Pearson correlations, yielding a total of 40 candidate genes, 31 of which shared human orthologs (step 3) and were selected for interactome analyses (step 4). During Phase II, we ranked the strains based on their grooming duration data (step 5). We then selected 31 random control genes (step 6) and correlated the grooming data with gene expression of the 31 candidate and randomly selected control genes (step 7). See text for details on the selection of control genes for this study. Because the expression data from this study were from the whole brain, the Allen Brain Atlas (ABA) provided further data for the regional expression patterns of both control and candidate genes (step 8). The STRING database of protein-protein interactions was used to create an interaction network (interaction confidence ≥0.8) of the protein products of these genes in mice, visualized using Cytoscape. A similar network was generated using orthologous human proteins and Cytoscape, to calculate the intersection between the mouse and human interactomes, resulting in a shared interactome containing only the nodes and edges conserved for both mice and humans (step 9).

Figure 1 outlines the overall methodological approach used in this study, summarizing its phases and steps. The rationale of Phase I is using correlational analyses from two MPD projects for a step-by-step dissection and identification of potential candidate genes and creating an integrated ‘translational’ molecular network for these genes. Phase II of this projects aims to provide an in-silico validation for Phase I, assessing known functions of the selected candidate genes, their expression in relation to regional distribution in the brain, and correlation with grooming behavior. Collectively, this approach allowed us to identify an interconnected network of candidate genes (which have expression levels that correlate with grooming behavior, display altered patterns of expression in key brain areas related to grooming and underlie important functions in the brain) which are therefore likely to represent potential neural targets for mouse grooming behavior.

2.2. Phase I. The search for candidate genes

General approach and generation of candidate genes (Step 1)

The Brown laboratory’s 2004 study (Brown1 project in MPD) contains grooming duration data for male and female mice of the 10 weeks-old 129S1/SvImJ, A/J, AKR/J, BALB/cByJ, BALB/cJ, C3H/HeJ, C57BL/6J, CAST/EiJ, DBA/2J, FVB/NJ, MOLF/EiJ and SJL/J strains (Brown, Gunn, 2004). Only male mouse data were used in the present experiment, to eliminate potential confounds associated with using mixed-sex cohorts (Table 1). There were marked strain differences in grooming duration in the four behavioral tests, with C57BL/6J mice showing the longest, and MOLF/EiJ, BALB/cJ, BALBc/ByJ and FVB/NJ strains showing lower grooming. Since strain differences in grooming behavior were not the main focus of this analysis, this aspect will not be discussed here. Note, however, that the MPD enables a fast comparison of all grooming scores in the Brown et al. 2004 study (Brown, Gunn, 2004), and is publicly available for further evaluation. The reliability of these grooming data was first assessed using the MPD toolbox, to determine whether the strain means for duration of grooming in the open field, elevated plus-maze, elevated zero-maze and light-dark box tests of anxiety were significantly correlated. Briefly, from grooming phenotypical data (Brown1 MPD project) for each of the four tests we chose “Other tools/toolbox”, and used “Correlations and relationships between phenotypes” option to “Search all MPD for correlated phenotypes”, selecting the Brown1 project from the drop-down menu. Tables 1 and 2 show the correlations for male mouse grooming duration in 12 inbred strains in the four different behavioral tests. Completed globally for all strains and separately within each strain and using strain means, these results generally show positive correlations between grooming duration data in different novelty-based anxiety tests.

Table 1.

Mouse Phenome Database-derived correlational analyses (Pearson R) of Brown et al. (2004) grooming duration data in male mice of multiple inbred strains tested in the open field test (OFT), light-dark box (LDB), elevated plus-maze (EPM) and elevated zero-maze (EZM) tests.

Tests	Strains	LDB	EPM	EZM
OFT	129S1/SvImJ	0.51# N = 12	− 0.08 N = 12	0.34 N = 10
	A/J	0.54* N = 20	−0.11 N = 19	−0.08 N = 6
	AKR/J	0.59** N = 19	0.43# N = 19	@
	BALB/cByJ	0.24 N = 12	−0.06 N = 12	−0.03 N = 12
	BALB/cJ	−0.22 N = 12	−0.23 N = 14	−0.18 N = 10
	C3H/HeJ	0.35 N = 16	−0.13 N = 16	$
	C57BL/6J	0.49# N = 13	−0.03 N = 13	0.49 N = 10
	CAST/EiJ	−0.04 N = 10	0.10 N = 10	@
	DBA/2J	0.11 N = 13	0.07 N = 13	0.88*** N = 10
	FVB/NJ	−0.18 N = 12	0.14 N = 12	0.22 N = 12
	MOLF/EiJ	−0.10 N = 11	0.94*** N = 11	@
	SJL/J	−0.38 N = 12	0.35 N = 12	@
LDB	129S1/SvImJ		−0.18 N = 12	0.74* N = 10
	A/J	-	−0.28 N = 20	−0.21 N = 6
	AKR/J		0.45 # N = 19	@
	BALB/cByJ		−0.09 N = 12	0.31 N = 12
	BALB/cJ		0.70* N = 12	0.20 N = 10
	C3H/HeJ		−0.21 N = 18	$
	C57BL/6J		0.38 N = 13	0.018 N = 10
	CAST/EiJ		0.25 N = 12	@
	DBA/2J		0.22 N = 13	−0.22 N = 10
	FVB/NJ		−0.23 N = 12	−0.23 N = 12
	MOLF/EiJ		−0.18 N = 11	@
	SJL/J		−0.54# N = 12	@
EPM	129S1/SvImJ			−0.08 N = 10
	A/J		-	−0.25 N = 6
	AKR/J			@
	BALB/cByJ			−0.23 N = 12
	BALB/cJ			0.05 N = 10
	C3H/HeJ			$
	C57BL/6J			−0.34 N = 10
	CAST/EiJ			@
	DBA/2J			0.06 N = 10
	FVB/NJ			0.58* N = 12
	MOLF/EiJ			@
	SJL/J			@

Correlations were not performed for some strains due to an insufficient sample size (n<3; $) or very low grooming activity (resulting in a lack of data variability; @).

^*P<0.05;

^**P<0.01;

^***P<0.001;

^****P<0.0001,

^#P = 0.05–0.1 (trend);

NS – not significant (p>0.05).

Table 2.

Correlation (Pearson R) of mean grooming duration across 12 strains of mice, listed in Table 1, between the elevated plus-maze (EPM) and zero-maze (EZM), light-dark box (LDB) and open field (OFT) tests.

	EPM	LDB	EZM
OFT	0.40	0.37	−0.01
EPM		0.61*	0.28
LDB			0.40

Correlations with the EZM used 11 strains as C3H/HeJ were not tested on the EZM in the Brown et al. project;

^*P<0.05.

Gene expression data from microarray studies from the Tabakoff laboratory, also available on the MPD (Tabakoff1 project), used an Affymetrix GeneChip Mouse Genome 430 2.0 containing 39,985 probesets to analyze whole-brain mRNA expression in multiple inbred strains of male mice (Tabakoff, Bhave, 2007) of the same age (10 weeks) as in Brown et al. (2004) study. For each strain, 4–6 replications were performed in order to minimize random variation between subjects, and the Robust Microchip Average expression measure was used to normalize the values for each gene for a given mouse strain (see (Irizarry et al., 2003) for details). To parallel grooming behavior with gene expression, the strain means for grooming data (Brown, Gunn, 2004) on each behavioral test were correlated with the mean whole-brain mRNA expression data (Tabakoff, Bhave, 2007) for male mice of the same strains using the MPD correlational toolbox. Briefly, from grooming phenotypical data (Brown 1 project) we chose “Other tools/toolbox”, and used its “Correlations and relationships between phenotypes and genotype or gene expression” option to “Find correlated gene expression probesets” in the “brain_Tabakoff1” project (selected from the drop-down menu). Note, however, that the MPD user interface undergoes regular modifications, and its future online versions and menu options may differ from those used in this 2012 study. A stringent level of significance (P<0.005) was used for this procedure, yielding significant correlations between 1028 mRNA probesets and grooming duration, of which 881 were located in known gene areas, in total accounting for 844 different mouse genes.

Analysis of candidate genes (Steps 2–5)

After identifying genes whose expression strongly correlated with grooming duration (P<0.005), we ranked these genes based on the number of behavioral paradigms in which they significantly correlated with grooming. All genes significantly correlating with grooming in more than one behavioral test (e.g., Tubgcp4, Ttl, Ptger3, Hoxb4, Pdgfb, Ptpra, Faah) were first included in our analysis as independently reconfirmed in several different behavioral models. Next, we ranked the remaining genes as potential candidate genes based on the absolute size of the Pearson correlation coefficient between grooming duration and mRNA expression (R nearest 1 or −1 in one of the four tests). In order to obtain a manageable number of genes for network analysis, we limited our search to the first 40 genes, allowing us to ensure adequate statistical power and avoid false positives. Due to the translational nature of our study, we further focused on the 31 candidate genes which were present in both mice and humans (Fig. 1).

GoMiner software (Zeeberg et al., 2003) was used to analyze the function of the candidate genes and determine any known role of these genes in brain function and neurobehavioral disorders. To complement these analyses, the Mouse Genome Informatics (MGI) database (Shaw, 2009) provided aberrant phenotypes of genetically modified mice (relevant for each of the candidate genes), allowing additional insight into gene-behavior interactions for the group of candidate genes identified in our study (Table 3). Protein-specific BLAST searches enabled further comparison of the homology between candidate mouse gene products and their human orthologs (Table 3), and the Ontological Discovery Environment (ODE) (Baker et al., 2009) and Drug Related Gene Database (DRG) (Gardner et al., 2008) were used to further search published gene expression studies linking candidate genes to mouse neural phenotypes. As shown in Table 3, bioinformatics-based analysis of the 31 genes present in both mice and humans revealed several interesting patterns, including 4 genes that encode tubulin-associated proteins (Tubgcp4, Ttl, Racgap and Mapre1), and 5 genes related to either actin or myosin (Pdgfb, Myo1b, Cdh1, Myo7a and Parvg). Finally, the Protein database (Pruitt et al., 2007) and sequence analysis using Hum-mPLoc (Shen and Chou, 2007) were also used in this study to characterize cellular location of protein products of the selected candidate gene (Fig. 3E).

Table 3.

A list of potential candidate genes selected for further analysis based on high correlation of grooming duration in the open field test (OFT), light-dark box (LDB), elevated plus-maze (EPM) and elevated zero-maze (EZM) tests, with brain expression microarray data.

Tests and Pearson correlation, P	Gene name	Selected gene ontology	Selected mammalian phenotypes from genetically modified mice(Shaw, 2009)	Human orthologs (and % homology)	Selected phenotypes from the Drug Related Gene Database and the Ontological Discovery Environment
Genes selected based on high correlation with grooming activity observed independently in two different tests
EZM R=0.89 P<0.0005 OFT R=−0.83 P<0.001	Tubgcp4 (tubulin, γ-complex associated protein 4)	Cytoplasm, microtubule, cytoskeleton organization	NA	TUBGCP 4 (99 %)	Ethanol preference in mice (Rodriguez et al., 1994); expression in cerebellum linked with activity in open field test (Philip et al., 2010) and varies across early development in mice (Kagami and Furuichi, 2001) (may have potential implications for motor phenotypes, reward pathways and neurodevelopmental disorders)
EZM R =0.87 P<0.0005 OFT R=−0.75 P<0.005	Ttl (Tubulin tyrosine ligase)	ATP binding, ligase activity, microtubule cytoskeleton organization, regulation of axon extension	Abnormal telencephalon development, abnormal neuron differentiation	TTL (96%)	Increased expression in nucleus accumbens following chronic cocaine in rats (Renthal et al., 2009); expression in striatum correlates with distance traveled in rats (Philip, Duvvuru, 2010); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications for motor phenotypes, reward pathways and neurodevelopmental disorders)
EZM R=−0.83 P<0.005 OFT R=−0.77 P<0.005	Ptger3 (prostaglandin E receptor 3 subtype EP3)	Activation of phospholipase C via G-protein-coupled receptor signaling	Abnormal body temperature regulation and pain threshold	PTGER3 (86%)	Expression in cerebellum linked with activity in the open field test after cocaine in mice (Philip, Duvvuru, 2010); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications for motor phenotypes, reward pathways and neurodevelopmental disorders)
EPM R=−0.82 P<0.005 EZM R=−0.75 P<0.005	Hoxb4 (homeobox B4)	Stem cell division, transcription and regulation, DNA-dependent, sequence specific DNA binding	Decreased body size and lower survivor rate	HOXB4 (90%)	Associated with ethanol consumption in mice (Mulligan et al., 2006); cerebellum expression associated with increased vocalization threshold (Philip, Duvvuru, 2010) and expression across early development in mice (Kagami and Furuichi, 2001) (may have potential implications for motor phenotypes, reward pathways and neurodevelopmental disorders)
EZM R=0.82 P<0.005 OFT R=0.75 P<0.005	Pdgfb (platelet derived growth factor, B polypeptide)	Cytoskeleton organization, cell projection assembly, activation of fibroblast growth factor receptor signaling pathway	Capillary aneurisms, edema endothelial hyperplasia	PDGFB (90%)	Associated with ethanol consumption in mice (Mulligan, Ponomarev, 2006); expression in neocortex associated with hyperactivity after ethanol consumption in mice (Philip, Duvvuru, 2010); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications for motor phenotypes, reward pathways and neurodevelopmental disorders)
LDB R=0.76 P<0.005 OFT R=0.76 P<0.005	Ptpra (protein tyrosine phosphatase receptor type A)	Positive regulation of oligodendrocyte differentiation, integral to membrane, phosphatase activity	Decreased anxiety-related response, hypoactivity, abnormal spatial learning	PTPRA (96%)	Associated with ethanol consumption in mice (Mulligan, Ponomarev, 2006); decreased expression in striatum following acute morphine and increased expression following chronic morphine in mice (Korostynski et al., 2007); decreased expression in dorsolateral prefrontal cortex following chronic crack cocaine in humans (Lehrmann et al., 2003); expression in hippocampus correlates with base-line activity during fear conditioning in mice (Philip, Duvvuru, 2010); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications for motor phenotypes, reward pathways and neurodevelopmental disorders)
LDB R=−0.76 P<0.005 OFT R=0.75 P<0.005	Faah (fatty acid amide hydrolase)	Degradation of bioactive fatty acid amides	Hypoactivity, increased alcohol consumption, analgesia	FAAH (84%)	Decreased expression in nucleus accumbens following chronic cocaine in mice (Renthal, Kumar, 2009) (may have potential implications for reward pathways)
Genes selected based on significant very high correlation with grooming activity observed in a specific single behavioral model
EZM R=0.96 P<0.000005	Prune2 (prune homologue 2-Drosophila)	Induction of apoptosis, phosphatase activity	NA	PRUNE2 (68%)	NA
EZM R=0.96 P<0.000005	Myo1b (myosin 1B)	Actin binding, motor activity, myosin complex, ATP binding	NA	MYO1B (96%)	Decreased expression in nucleus accumbens following chronic cocaine in mice (Renthal, Kumar, 2009); brain expression correlated with activity in open field test in mice (Philip, Duvvuru, 2010); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001)(may have potential implications for motor phenotypes, reward pathways and neurodevelopmental disorders)
EZM R=0.95 P<0.000005	Trp53bp1 (transformation related protein 53 binding protein 1	Response to DNA damage stimulus, regulation of transcription	Decreased body weight, postnatal growth retardation	TP53BP1 (81%)	NA
EZM R=0.95 P<0.00001	Gnb5 (guanine nucleotide binding protein beta 5)	G-protein coupled receptor protein signaling pathway, signal transducer activity	Seizures, small body size, ataxia, impaired motor coordination	GNB5 (99%)	Associated with activation of mesolimbic dopamine reward pathway after acute ethanol in mice (Kerns et al., 2005); increased expression in nucleus accumbens following chronic cocaine in mice (Renthal, Kumar, 2009); decreased expression in striatum following acute cocaine in rats (Paletzki et al., 2008); expression correlates with novel environment exploration in mice (Philip, Duvvuru, 2010)(may have potential implications for motor phenotypes and reward pathways)
OFT R=−0.95 P<0.000005	Cdh1 (cadherin 1)	Calcium ion binding, cell-cell junction, regulation of caspase, actin cytoskeleton	Decreased hair follicle number, abnormal skin	CDH1 (82%)	Decreased expression following chronic cocaine (Lehrmann, Oyler, 2003) and increased expression following chronic crack cocaine in dorsolateral prefrontal cortex in humans (Lehrmann, Oyler, 2003); expression modulated by nicotine in several brain regions in mice (Wang et al., 2008); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications for reward pathways and neurodevelopmental disorders)
EZM R=0.94 P<0.00005	Tceb3 (transcription elongation factor B (SIII), polypeptide 3)	Regulation of transcription from RNA polymerase II promoter	Delayed brain development	TCEB3 (81%)	Associated with ethanol consumption in mice (Mulligan, Ponomarev, 2006); differentially expressed in nucleus accumbens 24 h following ethanol consumption in mice (Mulligan, Ponomarev, 2006); expression modulated by nicotine in several brain regions in mice (Wang, Gutala, 2008); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications in reward pathways and neurodevelopmental disorders)
EZM R=0.94 P<0.00005	Arhgap26 (rho GTPase activating protein 26	NA	NA	ARHGAP26 (97%)	Associated with nicotine dependence in humans (Drgon et al., 2009); expression in neocortex correlates with distance traveled in mice (Philip, Duvvuru, 2010) (may have potential implications in motor phenotypes and reward pathways)
EZM R=0.94 P<0.00005	Gosr1 (golgi SNAP receptor complex member 1)	Golgi membrane, protein transport, SNAP receptor activity	NA	GOSR1 (98%)	Associated with ethanol consumption in mice (Mulligan, Ponomarev, 2006); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications in reward pathways and neurodevelopmental disorders)
EZM R=−0.89 P<0.0005	Wdfy1 (WD repeat and FYVE domain containing 1)	Metal ion binding, early endosome	NA	WDFY1 (97%)	Increased expression in nucleus accumbens following chronic cocaine in rats (Renthal, Kumar, 2009); ethanol preference in mice (Rodriguez, Plomin, 1994); variable expression in nucleus accumbens following ethanol consumption in rats (Bell et al., 2009) (may have potential implications in reward pathways)
EZM R=0.93 P<0.00005	Myo7a (myosin VIIA)	Melanosome, synapse, binding, cytoskeleton, nucleotide binding, auditory receptor cell differentiation	Abnormal hair cell morphology, small body size, circling, hyperactivity, altered anxiety	MYO7A (96%)	Decreased expression in the nucleus accumbens with chronic cocaine in rats (Renthal, Kumar, 2009); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications in reward pathways and neurodevelopmental disorders)
EZM R=0.93 P<0.00005	Ccnd2 (cyclin D2)	Cell cycle and division, G1 S transition of mitotic cell cycle, up-regulation of cell proliferation	Abnormal cerebellar morphology, absent Purkinje cells, motor incoordination, increased ethanol intake	CCND2 (90%)	Associated with ethanol consumption in mice (Mulligan, Ponomarev, 2006); ethanol preference in mice (Rodriguez, Plomin, 1994); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications in reward pathways and neurodevelopmental disorders)
OFT R=−0.93 P<0.00005	Racgap1 (Rac GTPase-activating protein 1)	α,β,γ-tubulin binding, cytokinesis, actomyosin contractile ring assembly	Premature death	RACGAP1 (84%)	Hippocampus expression correlates with Dowel test time in mice (Philip, Duvvuru, 2010); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications in motor phenotypes and neurodevelopmental disorders)
EZM R=0.93 P<0.00005	Mboat2 (membrane O-acetyltransferase domain containing 2)	Endoplasmic reticulum, phospholipid biosynthesis	NA	MBOAT2 (87%)	Associated with ethanol consumption in mice (Mulligan, Ponomarev, 2006) (may have potential implications in reward pathways)
OFT R=−0.92 P<0.00005	Mapre1 (microtubule-associated protein, RP/EB family member 1)	Cell projection, microtubule plus end binding, mitosis, centrosome	NA	MAPRE1 (97%)	Associated with ethanol consumption in mice (Mulligan, Ponomarev, 2006); expression altered in hippocampus after experimental brain injury (Matzilevich et al., 2002) and correlates with hippocampal neurogenesis in mice (Philip, Duvvuru, 2010); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications in reward pathways and neurodevelopmental disorders)
EZM R=0.92 P<0.00005	Tulp4 (tubby like protein 4)	Intracellular signal transduction	NA	TULP4 (95%)	Increased expression in nucleus accumbens after chronic cocaine in mice (Renthal, Kumar, 2009); altered brain expression after ethanol treatment in rats (Kerns, Ravindranathan, 2005); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications in reward pathways and neurodevelopmental disorders)
OFT R=−0.91 P<0.00005	Serf2 (small EDRK-rich factor 2)	NA	NA	SERF2 (100%)	Expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications in reward pathways)
EZM R=−0.91 P<0.0005	Parvg (parvin, gamma)	Actin binding, cell adhesion, cell junction, cytoskeleton	Abnormal retinal and astrocyte morphology, gliosis, aberrant blood brain barrier	PARVG (80%)	Increased expression in nucleus accumbens during chronic cocaine in mice (Renthal, Kumar, 2009) (may have potential implications in reward pathways)
EZM R=0.90 P<0.0005	Tm2d2 (TM2 domain containing 2)	Integral to membrane	NA	TM2D2 (90%)	Cerebellar expression correlates with activity in the open field in mice (Philip, Duvvuru, 2010); associated with consumption of ethanol in mice (Mulligan, Ponomarev, 2006); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications in motor phenotypes, reward pathways and neurodevelopmental disorders)
EZM R=−0.89 P<0.0005	Nr6a1 (nuclear receptor subfamily 6, group A member 1)	Down-regulated transcription from RNA polymerase II promoter, protein homodimerization, sequence-specific DNA binding	Abnormal brain development and cranial nerve morphology, lower testosterone	NR6A1 (96%)	Cerebellar expression correlates with activity in the light-dark box and time spent in zero maze quadrants in mice (Philip, Duvvuru, 2010) (may have potential implications in motor phenotypes)
EZM R=−0.90 P<0.0005	Fam163a (family with sequence similarity 163, member A)	Integral to membrane	NA	FAM163A (85%)	NA
OFT R=−0.89 P<0.0001	Aqp7 (aquaporin 7)	Porin activity, glycerol transport, water transport	Abnormal kidney physiology, hypoglycemia	AQP7 (77%)	Associated with ethanol consumption in mice (Mulligan, Ponomarev, 2006); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications in reward pathways and neurodevelopmental disorders)
EZM R=−0.89 P<0.0005	Gadd45gip1 (growth arrest DNA-damage-inducible, gamma interacting protein 1)	Negative regulator of cell cycle progression	Embryonic growth retardation, lethality	GADD45GIP1 (78%)	Increased expression in nucleus accumbens during chronic cocaine in mice (Renthal, Kumar, 2009), differential expression in striatum in several inbred mouse strains (Korostynski et al., 2006) (may have potential implications for motor phenotype and reward pathways)
EZM R=0.89 P<0.0005	Tnfaip2 (tumor necrosis factor, alpha-induced protein 2)	Dentritic cell marker, modulation of inflammation and angiogenesis	Aberrant angiogenesis	TNFAIP2 (71%)	NA
OFT R=−0.89 P<0.0005	Snai1 (snail homolog 1 Drosophila)	Hair follicle morphogenesis, down-regulation of cell differentiation	Open neural tube, abnormal cell migration	SNAI1 (88%)	Increased expression in D1 mutant mice in the caudoputamen (Zhang et al., 2005); expression varies in cerebellum across early development in mice (Kagami and Furuichi, 2001) (may have potential implications in reward pathways)

Over 800 genes had significant correlations with grooming duration using P<0.005. Potential candidate genes were then ranked based on the number of behavioral tests in which gene expression correlated significantly with grooming behavior. The remaining genes were then ranked based on the strength of their Pearson correlation between grooming duration and expression in one test. Gene ontology information was gathered from GoMiner, mouse phenotypes were selected from Mouse Genome Informatics (MGI) database (Shaw, 2009), and protein-specific BLAST provided the degree of homology between candidate mouse proteins and their human orthologs. The final column includes relevant neurological information from the Drug Related Gene Database (DRG) (Gardner, Akil, 2008) and the Ontological Discovery Environment (ODE) (Baker, Jay, 2009). Genes which highly correlated with grooming activity in the EPM, OFT and or EZM, but do not have human orthologs, include A230056J06Rik, A930015D, Cml3, A630012P03Rik, 4930402C16Rik, Gm11738, 1110005A03Rik, AI853106 and 1700023L04Rik (|R|=0.76–0.95, P<0.0005–0.00001), and were not assessed here. NA – information is not available.

Figure 3. Network-based analysis of mouse candidate genes, their human orthologs and shared networks (see Fig. 1 for general rationale).

A: Generating mouse grooming interactome using Cytoscape with data from grooming duration (Brown, Gunn, 2004) and whole brain expression microarray (Tabakoff, Bhave, 2007). Mouse grooming duration data from 12 strains in four behavioral tests were correlated using the MPD Pearson R toolkit with whole brain expression microarray data, generating 844 genes (step 1 in Fig. 1). The 40 most promising genes were selected for further analysis (step 2). Using the 31 genes present both in mice and humans, the STRING database generated a list of all proteins known to interact (Interaction Confidence ≥0.8). Cytoscape software then allowed these proteins to be displayed graphically in a user-friendly network (step 3). The 31 candidate gene products are displayed in red, while interacting proteins are displayed in green and the edges are in black. Twenty-two genes had interaction data available at an Interaction Confidence ≥0.8 (while 9 did not), and 15 of the candidate genes maintained connectivity within a single network.

B: Generating human interactome based on human orthologs of mouse candidate genes identified in panel A. Using the 31 genes present in both mice and humans, the STRING database generated a list of all proteins known to interact (Interaction Confidence ≥0.8). Cytoscape software then allowed these proteins to be displayed graphically in a user-friendly network (Step 3). Interaction data were unavailable for 7 genes at an Interaction Confidence ≥ 0.8; the remaining 20 human orthologs remained connected within a single network (step 3).

C: Generating shared network of candidate genes (Step 4). Using mouse (A) and human (B) interactomes generated previously in Step 3, Cytoscape calculated the intersection between them to identify the candidate genes and interactors that were conserved between the two species (Step 4). This network displays only the nodes and edges that are present in both networks. Of the 31 genes conserved in both species, 10 remained connected within a single network. The genes in this network are highly correlated with grooming behavior, interact within a small network, and are prominent in both mice and humans. Further analysis for these genes can be found in Table 3 (also see steps 5–8 in Fig. 1).

D: Generating the control gene network. In order to assess the candidate gene network, the control genes network was generated using the same approach (step 9), showing little connectivity between the genes, as can be assessed visually by comparing ‘candidate’ mouse interactome (A) with ‘control’ interactome (this panel). Further analyses of the average number of interactors per network showed that the mouse interactome of candidate genes tended to have more nodes per gene, compared to the mouse interactome of candidate genes (#P<0.08, trend, U-test), suggesting a generally higher functional interconnectedness compared to the randomly selected control genes.

E: Diagram showing cellular location of protein products of 31 candidate genes (A). Cellular location of these proteins was established based on protein sequence from Protein database (Pruitt, Tatusova, 2007) and sequence analysis using Hum-mPLoc (Shen and Chou, 2007).

2.3. Phase II. In-silico validation

Selection of control genes (Step 6)

To examine the validity of the procedure used to select candidate genes in Phase I, we used a random approach to select a group of control genes (see similar methods of selecting control genes to link gene activity to behavioral phenotype used in published literature (Mignogna and Viggiano, 2010)). For the present study, using the list of probes contained on the Affymetrix GeneChip Mouse Genome 430 2.0 microarray, we selected every 500^th probe (e.g., 500, 1000, 1500), which resulted in 31 control probes that targeted a gene-coding area, which were present in both mice and humans, and were not part of the 844 ‘putative’ candidate genes correlated with grooming duration in the previous step (Fig. 1). The control genes selected for this study included Cdkn2d, Trpm7, Sult2b1, Tnfaip1, N6amt2, Bmp7, Tbc1d1, Tspan8, Chrna4, Rps6kb2, Lipe, Csnk1g2, Rhbdd1, Slc27a4, Lpxn, Map2k7, Srek1, Fmn1, Txndc1, Nfam1, Syt11, Alkbh4, Ppp1r14c, Wwox, Sf3a3, Ppm1l, Cotl1, Gpr183, Erbb2ip, Lpp and Zfp879 (based on ABA data, all these genes are expressed in the mouse brain, and therefore were appropriate to use as control for this study).

Correlation of strain rankings of grooming duration and gene expression (Steps 5 and 7)

In order to validate the selection criteria used to generate candidate genes, the grooming duration measurements for each strain were compared with the gene expression of each strain. The Brown et al. data provided grooming duration for 12 strains which we ranked from 1 to 12, based on their results in 4 separate behavioral paradigms (the C3H/HeJ mice were not tested in the elevated zero maze in the Brown et al. study, and their strain rank for grooming was calculated based on 3 other behavioral tests). The four ranks for each strain/test were then averaged across all tests, enabling us to organize the 12 mouse strains according to their overall grooming duration rank ranging from 1 to 12 (Fig. 2A).

Figure 2. Analysis of strain ranking and regional gene expression.

A: Comparison of strain grooming duration with strain gene expression. Raw behavioral data were used to rank the 12 mouse strains based on their grooming duration in 4 behavioral tests. C57BL/6J mice groomed the most while the MOLF/EiJ strain groomed the least. Genes were divided into three groups: positively correlating with grooming (17 genes; top panel), negatively correlating with grooming (14 genes; middle panel), and randomly selected control genes (31 genes; bottom panel). Microarray data allowed the expression of the genes in these three groups to be summed for each strain, to be used for correlation analyses (using Spearman correlation) with the strain ranking of grooming duration.

B: Regional expression analysis of data obtained using the Allen Brain Atlas (ABA) for C57BL/6J mouse strain. The following brain areas (selected based on ABA pre-defined brain sectioning) were included in this analysis: HP = hippocampal formation; OC = olfactory cortex; MD = medulla; IC = isocortex; CB = cerebellum; CS = cortical subplate; MB = midbrain; PN = pons; TM = thalamus; ST = striatum; HT = hypothalamus; PD = pallidus. To investigate the expression patterns of candidate genes in the C57BL/6J mouse strain (showing robust grooming responses in Brown et al. 2004 behavioral study), we used the ABA data to establish the raw expression scores for 31 candidate and 31 control genes across 12 regions of the mouse brain (see Methods for details of selecting control genes). Based on their Pearson correlations with grooming duration, the candidate genes were again divided into two groups, positively and negatively correlating with grooming, in order to investigate whether different brain regions differentially affect expression data. The expression values for each gene were numbered 1–12 (with 12 indicating the region of highest expression for a given gene and 1 the lowest expression) and averaged for each brain region. Control genes expression ranking score differed significantly from both the positively correlated and negatively correlated genes in 6 brain regions (P<0.05, U-test vs. the respective ranking scores of control genes) – medulla, cerebellum, midbrain, thalamus, striatum and hypothalamus, all strongly implicated in the regulation of mouse grooming behavior.

The Tabakoff laboratory’s microarray data provided expression values for our candidate and control genes in each of the 12 mouse strains (genes with multiple probes targeting the same gene were averaged to obtain a single value per gene). The expression values for each gene were ranked from 1 (lowest) to 12 (highest), to match with the number of mouse strains used in this study. At this point, based on their initial ‘strain’ Pearson correlation coefficients, as explained in Phase I, the candidate genes were divided into positively correlating with mouse grooming (17 genes) or negatively correlating with mouse grooming (14 genes), in addition to 31 control genes. For each mouse strain, we then calculated the total rank of expression of genes from each group separately, i.e., positively correlating (range: 17–204), negatively correlating (range: 14–168) and control genes (range = 31–372). Once the strain ranking for grooming and gene expression were calculated, we applied Spearman correlation coefficient to further analyze these data. As the strains were ranked from highest to lowest grooming duration, we graphed the gene expression data for each strain (Fig. 2A). The genes positively correlating with grooming trended downward (i.e., the lower the grooming duration of a given strain, the weaker the gene expression; Spearman R = 0.53; P<0.05). The genes which negatively correlated with grooming showed the opposite pattern (Fig. 2A; Spearman R = −0.92; P<0.00001). In contrast, as shown in Fig. 2A, the control genes showed no significant correlation between strain gene expression and strain grooming duration (Spearman R = 0.19; P<0.5, NS).

Notably, the C57BL/6J mice had the highest grooming duration in the behavioral tests, consistent with earlier observation of robust grooming behavior in this common inbred mouse strain (Kalueff and Tuohimaa, 2004). In this strain, the genes which positively correlated with grooming were highly expressed in the brain while the genes with negative correlation were expressed at low levels, thereby supporting the genes’ selection criteria described above, and the suitability of this strain for further analyses and validation.

Regional expression analysis using the Allen Brain Atlas (Step 8)

Since the microarray data used here (Tabakoff, Bhave, 2007) provided only whole-brain expression data, the regional expression of candidate and control genes was assessed using the ABA expression data for the C57BL/6J strain (Lein, Hawrylycz, 2007). Notably, of the 12 strains investigated in this analysis, the C57BL/6J mice displayed the longest grooming duration, supporting the use of this strain in dissecting the expression patterns of our candidate genes. The ABA contains RNA expression values from 12 different regions and multiple genes across the entire genome (Lein, Hawrylycz, 2007). If the ABA gene expression data contained multiple experiments for the same gene, data were averaged across experiments to obtain a single value per gene for each brain region. Expression data were unavailable for 3 candidate genes (Gosr1, Tm2d2, Racgap1) and 2 control genes (Srek1, Tnfaip1). Because some genes are expressed at high levels across the brain while other genes have uniformly lower expression, we converted each raw expression score into a rank from 1 to 12, giving each gene equal weighting, regardless of their raw expression levels. This strategy was first applied to a cohort of randomly selected control genes, reflecting the expression patterns of the entire genome. Next, the candidate genes were divided into two groups (as described previously), including genes positively or negatively correlating with grooming duration. The brain structures where candidate and control genes were expressed differently provided us with potential regions of importance for mouse grooming. Analyzing the average expression for each brain area, the highest deviation in expression between control and candidate genes occurred in the medulla, but not in the areas usually not implicated in the grooming phenotypes of mice, such as the olfactory cortex and pons (Fig. 2B). Overall, the candidate genes that positively correlated with grooming differed significantly (by U-test) from the control cohort in 6 regions (medulla, cerebellum, midbrain, thalamus, striatum and hypothalamus), while the negatively correlated candidate genes differed from the control genes in these same 6 regions (Fig. 2B). Interestingly, the positively and negatively correlated candidate genes showed similar trends in expression across different brain regions, implicating these genes in grooming in a striking contrast to randomly selected control genes (Fig. 2B).

Network analysis using STRING database (Step 9)

Examining the functionality of the genes in our study, we used the STRING database (Szklarczyk, Franceschini, 2011) containing known and predicted protein-protein interactions to analyze the protein products of the 31 candidate genes present in both humans and mice. Several other studies have already utilized protein-protein interaction networks to make predictions about the role of a gene and its potential phenotypes (Lage et al., 2007, Wang and Marcotte, 2010). The STRING database calculated all direct interactions between these 31 candidate protein products and the rest of the proteome, generating a network of protein-protein interactions. The confidence that a given protein-protein interaction represents a functional relationship is reported by the STRING database as an Interaction Confidence ranging from 0 to 1. To increase the predictive power of this network, we generated a protein-protein interactome for the 31 candidate protein products using a stringent Interaction Confidence of at least 0.8, a threshold that is high enough to manage false positives and is commonly used in the literature (Kim et al., 2010, Rybarczyk-Filho et al., 2011). Cytoscape software (Cline et al., 2007) was used to visualize these interactions in a web of nodes and edges, organized for visualization using a layout algorithm (Fig. 3A). Of the 31 candidate mouse gene-products, 9 did not have known interactors exceeding the interaction confidence threshold, while 15 remained connected within a single network. For comparative purposes, a similar network was also generated for the control genes (Fig. 3B).

The same procedure was next applied to the respective human orthologs of these proteins, using an interaction confidence of ≥ 0.8 (Fig. 3A). Cytoscape generated and visualized the interactome of 31 human gene candidates, where 7 did not have known interactors at the chosen interaction confidence, and 20 remained connected in a single network. We then used Cytoscape to assess the overlap between the mouse and human interactomes, yielding a final network of interacting proteins present in both species (Fig. 3A). Ten of the 31 gene candidates remained interconnected within a single network, representing promising translational targets to study grooming based on correlation between behavioral phenotypes, brain gene expression and integration within the larger cross-species protein interactome (Fig. 3A, Table 3). Finally, in order to assess the differences in connectivity between candidate and control genes, we constructed a mouse interactome for our control genes and performed an unpaired Wilcoxon-Mann-Whitney U-test comparing the number of connectors per node between the networks of mouse candidate and control genes. Overall, the networks differed both qualitatively (with control genes appearing less interconnected in the graphic form) and quantitatively, as the candidate gene networked showed a trend (P<0.08) to more connectors per node, compared to the respective control gene network (Fig. 3B).

3. Discussion

The present study is the first comprehensive in-silico analysis combining behavioral and genomic data to examine mouse grooming behavior. An increased understanding of this important phenotype is likely to lead to insights into complex neurobehavioral disorders, such as autism and obsessive-compulsive disorder, OCD) (Berridge et al., 2005, Bienvenu et al., 2009, Crawley, 2007, Feusner et al., 2009, Rapoport, 1991, Shmelkov et al., 2010, Silverman et al., 2010, Swedo et al., 1989, Welch et al., 2007, Yang and Lu, 2011). In addition, this ‘proof-of-concept’ approach can easily be adapted to other complex traits in mice, as well as can be applied to grooming and other complex behaviors in various model organisms and humans.

As already mentioned, brain expression microarray results initially provided 844 genes with the expression significantly correlating with mouse grooming behavior. Since these genes have been selected with a high stringency (P<0.005), we first chose genes with high significance demonstrated independently in several behavioral tests, then selecting the remaining candidates based on the strength of the correlation in a single test. By selecting the top 40 genes, we were able to generate a highly integrated web of candidate genes and their interactors, revealing easily visualized, potentially novel interactions for mouse grooming behavior (Fig. 3A). Selecting genes with highly homologous and similarly interconnected human orthologs further supported the translational potential of the candidate genes identified in this study (Fig. 3A). To ensure that our candidate genes yielded a robust and meaningful network, the protein-protein interactions in the mouse and human interactomes were generated with a stringent Interaction Confidence (Fig. 3A). Since human and mouse genomes share a high degree of homology (Boguski, 2002), many of the pathways and interactions in humans are expected to be present in mice.

Representing a prominent phenotype sensitive to various genetic, behavioral and pharmacological manipulations (Angrini et al., 1998, Greer and Capecchi, 2002, Kalueff et al., 2005, Kalueff and Tuohimaa, 2005c), rodent grooming is a complex, highly organized behavior that can be further dissected for an in-depth analysis of centrally-controlled neurophenotypes (Kyzar, Gaikwad, 2011). The current study has generated a list of putative genes for the further study of mouse self-grooming behavior, representing a promising step in understanding of the genetic control of multifaceted behavioral domains. This information may help elucidate the relatively unknown neural and molecular mechanisms of self-grooming and other patterned motor responses, including pathological stereotypic behavior in OCD, attention deficit/hyperactivity disorder, schizophrenia and autism spectrum disorder (Chao et al., 2010, Mahone et al., 2004, Nayate et al., 2011).

In general, as genetic contributions to baseline anxiety/activity levels, motor coordination and other factors may modulate mouse grooming behavior through multiple mechanisms, each of the candidate genes may influence baseline grooming activity, or self-grooming in response to novelty stress. For example, the mammalian phenotype for prostaglandin E receptor 3 (Ptger3) mutation includes an abnormal body temperature and impaired pain threshold (Ushikubi et al., 1998), which may produce variation in thermoregulation and subsequently affect baseline grooming behavior. While myosin VIIA (Myo7a) mutation primarily leads to vestibular dysfunction, numerous reports reveal comorbidity between balance disorders and anxiety in both rodents (Kalueff et al., 2008, Shefer et al., 2010) and humans (Alvord, 1991, Balaban and Thayer, 2001), consistent with altered anxiety phenotypes in Myo7a mice (Shefer, Gordon, 2010).

Importantly, a number of cytoskeletal genes were associated with grooming behavior in this study. While cytoskeletal proteins are well-known for their role in cellular organization (Kellogg et al., 1994, Misteli, 2001), recent evidence has implicated actin- and myosin-related proteins in more complex phenomena, such as receptor trafficking, dendritic plasticity and sensorimotor gating (Bosch and Hayashi, 2011, Fradley et al., 2005, Yuen and Yan, 2009). Certain cytoskeletal genes are likely to be differentially regulated in various brain areas, leading to increased divergence and specialized functions in neurons. Therefore, variation in synaptic receptor expression, driven by cytoskeletal mechanisms, may contribute to the observed strain differences in grooming activity. This mechanism is only beginning to be recognized by the field, as very few studies have focused on the multifaceted role of cytoskeletal genes in complex behavioral and physiological domains. The importance of actin in chromatin remodeling has been well documented (Ferrai et al., 2009, Obrdlik et al., 2007, Percipalle and Visa, 2006), possibly explaining why the actin-associated proteins Pdgfb, Myo1b, Cdh1, Myo7a and Parvg were implicated by this study (Table 3). The presence and interconnectedness of the cytoskeletal proteins Tubgcp4, Pdgfb, Cdh1, Racgap1, Myo7a, Mapre1 and Parvg in the shared interactome (Fig. 3A) further suggest their role in various processes related to grooming behavior.

Our analysis also produced some unexpected results, as several notable genes implicated in compulsive grooming and OCD-like behavior (Slitrk5 and Sapap3) in rodents (Shmelkov, Hormigo, 2010, Ting and Feng, 2011, Welch, Lu, 2007, Yang and Lu, 2011) and humans (Bienvenu, Wang, 2009, Boardman et al., 2011, Zuchner et al., 2009) were not identified here. This may be due to the detrimental effects arising from the genes’ mutation or knockout, leading to the disruption of striatal neuronal differentiation and neurotransmission (Shmelkov, Hormigo, 2010, Welch, Lu, 2007, Yang and Lu, 2011), whereas we focused on the gene expression of wild-type inbred mice whose brain function has not been disrupted through genetic modification. While genes identified in genetically modified animals may not be involved in the normal self-grooming behavior, their disruption affects corticostriatal circuitry, which can indirectly evoke aberrant grooming. Specifically, the selective over-activation of the orbitofrontal cortex, abnormalities in striatal anatomy/cell morphology, and alterations in glutamate receptor composition that accompany a mutation (Shmelkov, Hormigo, 2010), may disrupt key neural pathways involved in normal grooming behavior.

Likewise, while our analysis revealed the role of the transcription factor homeobox B4 (Hoxb4) in grooming, we did not observe a correlation of the more widely reported Hoxb8 gene, linked to compulsive grooming and other OCD-like behaviors (Chen, Tvrdik, 2010, Greer and Capecchi, 2002, Yang and Lu, 2011). The Hox genes are arranged in the genome in a collinear fashion, with the Hoxb genes clustered together on chromosome 17 (Chambeyron and Bickmore, 2004). Hoxb4 and Hoxb8 are both DNA sequence-specific transcription factors responsible for various developmental processes, including hematopoesis. Interestingly, the aberrant grooming in Hoxb8 mutant mice has recently been linked to defective hemopoietic-derived microglia (Chen, Tvrdik, 2010). Thus, the possible role of a hematopoietic gene Hoxb4 in mouse grooming in this study is congruent with the hematopoitic hypothesis of aberrant grooming in mice (Chen, Tvrdik, 2010). Several of our candidate genes have previously been linked to psychiatric disorders, some of which are closely related to pathological grooming. For example, Ptpra knockout mice display defects in neuronal migration, sensorimotor gating and habituation to startle response, thereby linking Ptpra to schizophrenia (Takahashi et al., 2011). They also show altered anxiety phenotypes (Skelton et al., 2003), whereas its human ortholog PTPRA resides in the 20p13 region which has repeatedly been linked to psychosis (Fanous et al., 2008, Teltsh et al., 2008). As already suggested in the literature (Audet, Goulet, 2006, Isingrini et al., 2011, Papaleo et al., 2011), grooming responses in mice may represent traits highly relevant to schizophrenia, anxiety and depression, and our results are in line with this notion.

Previous work has implicated abnormal brain development in several complex neuropsychiatric disorders. For example, aberrant neuronal migration and callosal hypoplasticity are commonly reported in schizophrenia (Connor et al., 2011, Knochel et al., 2012), whereas autistic patients show underdevelopment of the cerebellum and migration defects (Verhoeven et al., 2010). Analysis of the 31 shared genes generated by this study in mice and humans implicates these genes in brain development (Table 3), including abnormal cerebellar morphology (Ccnd2), abnormal neuronal migration (Snai1), abnormal brain development (Tceb3, Nr6a1) and impaired neuronal differentiation (Ttl). Moreover, genes selected for the analyses have different numbers of known interactors (Fig. 3A), many of which form key nodes in our shared interactome (e.g., Tubgcp4, Pdgfb, Cdh1, Gnb5, Racgap1, Myo7a, Ccnd2, Mapre1, Parvg). The genes Tubgcp4, Racgap1 and Mapre1 were mentioned previously because of their involvement with tubulin, while Cdh1, Myo7a and Parvg interact with actin and may play a role in chromatin remodeling. Examining public databases for Ccnd2 revealed its importance in the development of the cerebellum (Table 3), where Gnb5 is also important, since Gnb5 knockout mice have abnormal cerebellar development and motor incoordination (Zhang et al., 2011).

Furthermore, there were several limitations in our study. First, while we used correlational in-silico analyses, more specific studies are needed to investigate the functionality of identified genes in mouse grooming behaviors. Also, since our study utilized whole-brain microarray data, this limitation may be further resolved using region-specific gene expression analyses, empowered by sophisticated databases, such as the ABA (Lein, Hawrylycz, 2007). Far from providing an expansive and complete list of genes associated with mouse grooming, the approach described in this study offers a rapid, cost-effective and promising way to find new targets for important neurobiological functions (see (Stewart, Gaikwad, 2011) for review). Furthermore, while our study focused on quantity (duration) data, mouse grooming is a complex behavior with an important sequential (patterning) microstructure (Kalueff et al., 2007). Therefore, future studies may elucidate the correlation between gene expression and sequencing of mouse grooming. Again, because correlations in biological systems do not necessarily represent functional interrelationships between different phenomena or processes, future integrative research (currently underway in our laboratory) will have to assess in-depth the exact causal pathways of aberrant grooming examined here. Finally, epigenetic factors play an important role in the regulation of activity of various genes (Fish et al., 2004, Meaney and Szyf, 2005, Rothbart and Posner, 2005, Sheese et al., 2007, Voelker et al., 2009, Weaver et al., 2007). Therefore, further characterization of the genes generated by our method, as well as analysis of their epigenetic regulation and gene x environment interactions, may provide important clues in understanding the neurobiology of grooming behavior and identifying targets for modulating complex patterned behavior across a number of model species. While the link between genes and behavior remains a major challenge in modern biological psychiatry, our study may offer one of potential large-scale, data-mining approaches to address these aspects.

Highlights.

• Mouse grooming is an important, evolutionarily conserved behavioral phenotype

• This in-silico study uses open gene expression and behavioral databases

• We identify an interconnected network of candidate genes implicated in grooming

• Our results show the utility of large-scale data-mining and in-silico modeling

Abbreviations

ABA

Allen Brain Atlas

DRG

Drug Related Gene Database

EPM

The elevated plus-maze test

EZM

The elevated zero-maze test

GO

Gene Ontology

LDB

The light-dark box test

MPD

Mouse Phenome Database

MGI

Mouse Genome Informatics

OCD

Obsessive-compulsive disorder

ODE

Ontological Discovery Environment

OFT

The open field test

STRING

Search Tool for the Retrieval of Interacting Genes/Proteins

URI

Uniform Resource Identifier