Complex patterns of dopamine-related gene expression in the ventral tegmental area of male zebra finches relate to dyadic interactions with long-term female partners

Abstract

Dopaminergic projections from the ventral tegmental area (VTA) to multiple efferent targets are implicated in pair bonding, yet the role of the VTA in the maintenance of long-term pair bonds is not well characterized. Complex interactions between numerous neuromodulators modify activity in the VTA, suggesting that individual differences in patterns of gene expression in this region may explain individual differences in long-term social interactions in bonded pairs. To test this hypothesis we used RNA-seq to measure expression of over 8000 annotated genes in male zebra finches in established male-female pairs. Weighted gene co-expression network analysis identified a gene module that contained numerous dopamine-related genes with TH found to be the most connected gene of the module. Genes in this module related to male agonistic behaviors as well as bonding-related behaviors produced by female partners. Unsupervised learning approaches identified two groups of males that differed with respect to expression of numerous genes. Enrichment analyses revealed that many dopamine-related genes and modulators differed between these groups, including dopamine receptors, synthetic and degradative enzymes, the avian dopamine transporter, and several GABA- and glutamate-related genes. Many of the bonding-related behaviors closely associated with VTA gene expression in the two male groups were produced by the male’s partner, rather than the male himself. Collectively, results highlight numerous candidate genes in the VTA that can be explored in future studies and raise the possibility that the molecular/genetic organization of the VTA may be strongly shaped by a social partner and/or the strength of the pair bond.

1. Introduction

Dyadic social interactions between males and females are essential for survival and reproductive success. Although unusual across vertebrates, a few mammals (e.g., prairie voles, Microtus ochrogaster) and many bird species (including zebra finches, Taeniopygia guttata) establish and maintain enduring, monogamous pair bonds that are characterized by ongoing mate-directed, affiliative social interactions ^1–3. A large body of research reveals much about the neural mechanisms that underlie the formation of pair bonds ^4–8, yet much less is known about neural mechanisms that function to maintain long-term, affiliative bonds ^9–11. Furthermore, most of the studies on pair bond maintenance focus on the production of selective aggression towards non-mates as an index of bond maintenance, rather than on ongoing affiliative interactions observed between long-term partners (e.g., ^9,11–14).

To maintain a pair bond, individuals must be motivated to approach and engage with mates in positive, affiliative interactions. Dopaminergic projections from the ventral tegmental area (VTA) to both the nucleus accumbens and ventral pallidum are central to the regulation of motivated, reward-directed behaviors ^15,16, and both of these regions are strongly implicated in pair bonding ^4,11,17,18. In addition to its projections to the nucleus accumbens and ventral pallidum, the VTA is a source of dopamine for multiple additional brain regions that are essential to the regulation of affiliative social relationships, including a network of reciprocally connected brain regions often referred to as a “social behavior network” ^19,20. Many of these regions, including lateral septum, bed nucleus of the stria terminalis and medial amygdala have also been implicated in pair bonding in birds and mammals (e.g., ^10,21–24).

Most studies of pair bonding and related social behaviors focus on efferent targets of the VTA (many of which also send projections back to VTA) rather than VTA itself. This may be in part because patterns of activity in VTA emerge from the activity of multiple neuromodulators acting on multiple excitatory and inhibitory afferent and efferent projections as well as inhibitory neurons (reviewed in ²⁵). For example, the modification of dopamine release from VTA projection neurons via GABA or glutamate receptor antagonist administration directly into VTA induces pair bonding in male prairie voles ²⁶. The complex interplay among neuromodulators is highlighted further by studies in songbirds, which show that multiple neuronal markers in VTA (e.g., cells labeled for immediate early genes, dopamine-beta-hydroxylase, vasopressin V1a receptors, tyrosine hydroxylase, neurotensin, opioid neuropeptides, and other measures of catecholamine activity) correlate with pairing-related behaviors (e.g., pairing status, courtship behaviors, clumping, allopreening, and nesting ^24,27–34). It is also noteworthy that markers in VTA can be associated closely with the behavior that an individual receives from a partner ^24,35.

Because the VTA is a complex, critical source of input to numerous areas that are essential for bonding and ongoing affiliative social interactions, individual differences in complex patterns of gene expression in VTA should help to explain individual differences in long-term social interactions in bonded pairs. Here we use RNA-seq to test this hypothesis in established male-female zebra finch pairs with a focus on dopamine and additional related neuromodulators. To initiate this line of research, we limit our focus to the male partner. We used machine learning and other bioinformatics approaches to identify relationships between patterns of gene expression in VTA and both the focal individual’s own behavior as well as the sensory and social input from his mate.

2. Materials and Methods

2.1. Animals

Twenty adult (7–13 months of age) zebra finches (10 male focal subjects and 10 female pair mates) were selected from the breeding colony at the University of Wisconsin-Madison. Prior to the experiment, all birds were housed in indoor stainless steel cages in single sex groups with visual, but not acoustic, isolation from the opposite sex. Room conditions were set to a light cycle of 16 h light:8 h dark, humidity from 30–60% and a temperature from 20–24 °C. Animals had ad libitum access to water and a pellet and seed mix as well as vegetables and egg mixture twice a week. All experiments were approved by the University of Wisconsin Institutional Animal Care and Use Committee and in accordance with the Guidelines of the National Institutes of Health.

2.2. Behavioral testing

Each focal male was assigned a random, non-related, adult female partner. Pairs were each introduced into a separate cage (56 X 58 X 57 cm³) that included a nest box, perches, a water bottle, and cuttlebone. They were provided food, water and nesting material ad libitum. Prior to observations, animals were housed in these pairs for 16 days, a time period sufficient for pair bond formation in this species ³⁶. Birds were habituated daily to a researcher in an observation chair 1.5 m away. To prevent initiation of parental behavior, any eggs laid were counted and removed daily.

A single observer conducted daily 20 min observations of each zebra finch pair in a random order on each of 5 days (days 17–21 after pairing). Each observation began between 30min and 5h 30min after the lights turned on. For each bird in the pair, we recorded multiple variables. These include a measure of courtship behavior (sum of the number of female-directed songs [performed by males only], crouch and quivers, exaggerated greetings, and mounting), a measure of pair-contact behavior (sum of allopreening, the times a bird invited a partner to preen it, and clumping [i.e., birds physically contacted one another with bodies, which is a measure that is the same for both the male and female]), nesting behaviors (sum of nest material gathering, traveling with nest material, looking in the nest box, and entering the nest box), and partner-directed agonistic behavior (sum of displacing and beak fencing). We defined beak fencing as two birds wiping beaks together. Although, this can be observed as part of greeting, we observed it often in association with displacement so considered it agonistic in this study. We also measured synchronized behaviors (sum of times pair members fed, drank, or flew together). Undirected songs (male only) were songs that were produced while facing away from the cage mate. Calls included all non-song vocalizations (we did not distinguish between call types of note the location from which birds were calling). Beak wipes can be part of courtship or performed after feeding. We did not distinguish between the two contexts. Approach behavior consisted of a bird approaching to within approximately 4 cm of another bird. Bouts of drinking and feeding when alone (feeding and drinking when with a partner were included in the measure of synchronous behavior), and autopreening were recorded with bouts separated by at least 2 secs. Beak hues were estimated using the Munsell color system categorical scale ³⁷. We only used the hue categories of the upper mandible in analysis due to lack of hue variation in the lower mandible. Eggs were counted daily beginning on day 1 and nests were weighed at the end of the experiment. We assume that the nest weight is determined more by the male because in zebra finches males tend to deliver material to the nest site, although females generally accept and arrange nest material ^38–41. The sums of each behavior for the 5 days of testing were used for analyses. During brain tissue collection (below), both the length and width of the largest gonad were measured for each bird.

2.3. Tissue collection and processing

All animals were euthanized by rapid decapitation on day 22 (the day after completion of behavioral observations). Brains were immediately removed, frozen on dry ice and stored at −80 °C. Whole brains were sectioned at 200 μm thickness in the coronal plane on a cryostat at −15 ^oC and mounted onto glass slides. A 2 mm punch over the midline containing the ventral tegmental area (VTA; Figure 1) was dissected using a Brain Punch Set (FST 18035–02, Foster City, CA, USA) under a dissection microscope over dry ice and subsequently stored at −80 °C. Although the punches included areas outside of VTA, based on the tyrosine hydroxylase labeling distribution in zebra finches⁴², punches likely did not include tyrosine hydroxylase positive neurons located outside of VTA. Anatomically equivalent sections were taken from all animals. For consistency, the same pair of researchers collected and processed all tissue samples and sample order was randomized. All samples included in this study were similar in RNA concentration and purity.

Figure 1.

Illustration of A) the location of VTA and B) photomicrograph of Nissl stained tissue showing the approximate location of the 2mm diameter tissue punch taken from VTA in a coronal section of brain. A = arcopallium, Cb = cerebellum, DM = dorsomedial portion of the nucleus intercollicularus, HVC = used as a proper name, ICo = nucleus intercollicularus, NIII = 3^rd cranial nerve, NC = caudal nidopallium, PAG = periaqueductal gray, TnA = nucleus taeniae of the amygdala. Scale bar = ~600 μm.

Microdissected tissue was homogenized and total RNA was extracted with the Bio-Rad Aurum Total RNA Fatty and Fibrous Tissue Kit (Catalog No. 732–6830; Bio-Rd, Hercules, CA, USA) according to manufacturer’s instructions. Total RNA was measured using a Nanodrop system (Thermo Scientific, Wilmington, DE, USA). Additionally, the 28S:18S rRNA was measured with an Agilent RNA 6000 Pico kit (Eukaryote Total RNA Pico, Agilent Technologies, Santa Clara, CA). Analysis of these data demonstrated satisfactory marker and ribosomal peaks as well as RNA concentrations and RNA Integrity Numbers. All samples had an A260/A280 ratio greater than 1.8. DNase treated RNA (100 ng/uL per cDNA reaction) was converted into single-stranded cDNA using the Invitrogen SuperScript III First-Strand Synthesis System (Catalog No. 18080–05; Invitrogen, Carlsbad, CA, USA).

2.4. RNA sequencing and quantitative real time PCR

RNA-seq was run to identify patterns of gene expression in VTA for each male. Illumina® Total RNA-Seq TruSeq Stranded Total RNA Library Prep Kit was used to remove cytoplasmic and mitochondrial rRNA, and a sequencing library was generated. The amplified library was purified, quantified, and applied for template preparation using the HiSeq2000 platform. All samples were run within a single NovaSeq run. RNA-seq procedures conformed to best practices following guidelines by ENCODE. All sequencing procedures were performed by the University of Wisconsin-Madison Biotechnology Center’s Next Generation Sequencing Facility. Expression was normalized using RSEM ⁴³, which is an approach designed to help with RNA-seq assembly and quantification in the absence of a fully sequenced genome. Reads were aligned with the zebra finch library in Ensembl. The average for primary reads was 23 914 055 with a range from 19 350 765 to 29 570 056. The average primary mapping rate was 87.16%.

Technical quality was determined using several parameters. The trimming software skewer ⁴⁴ was used to preprocess raw fastq files. Skewer implements an efficient dynamic programming algorithm designed to remove adapters from Illumina-generated sequence reads. Combined cycle base quality, per cycle base frequencies and average base quality, relative 3-Kmer diversity, Phred quality distribution, mean quality distribution, average read length and read occurrence distribution were used as additional quality control measures on the read data.

Quantitative real time PCR (qPCR) was run to verify RNA-seq results for genes of interest (TH, SLC6A2, and DDC), selected based on WGCNA results below, and two reference genes (PGK1 and PPIA). qPCR was run on cDNA synthesized from the same RNA used for the RNA-seq using methods identical to those reported in our past research, including the same reference genes ⁴⁵ using the following primers: TH forward -- GCCATGCTGAACCTCTTCTT, reverse – GATGGTGGCACTTGTCCAAT; SLC6A2 forward – TGTGGAAAGGCGTGAAGACT, reverse – TAGTCAGCAAAGCATCCCTGT; DDC forward – ACAGATTATCGGCACTGGCA, reverse – GGAAGCAGACCAGTCCCAAA.

2.5. Unsupervised approaches to identify male groups based on gene expression

We started by using bioinformatics approaches to determine the degree to which birds naturally sorted into groups based on patterns of gene expression alone. As described in section 2.7, we then explored the degree to which males found to have distinct patterns of gene expression differed with respect to physical and behavioral traits.

We used multiple approaches to evaluate VTA gene expression to determine whether there were natural groupings among the ten paired males. For analysis, we only used genes for which there was an official, protein coding annotation (N = 8386). Using multiple classification tools (see Results) within Weka ⁴⁶, we performed unsupervised cluster analysis. We also used a newer tool for unsupervised learning, Uniform Manifold Approximation and Projection (UMAP) that runs in Python and is proposed to outperform other approaches for looking at complex topography when analyzing individuals with multiple features ⁴⁷.

2.6. Differential gene expression analysis

Using the two male groups identified via unsupervised classification (N=5 per group as detailed in results), we conducted a differential gene expression analysis of RNA-seq data using the EdgeR Bioconductor Package, v. 3.9 ⁴⁸. Statistics, including false discovery rates, are provided in Supporting information Table S1.

2.7. Weka analysis to connect gene expression with traits

We used SVMAttributeEvaluator (within Weka) to rank the male and partner traits that best allowed for identification of the two male groups we had previously identified via unsupervised clustering approaches. We then used the top five ranked traits along with the classifier tool to determine whether these traits were sufficient to accurately predict class (i.e., membership in what we refer to as the Low or High Dopamine group, see Results) using the leave one out cross validation approach, repeated 20 times using different seeds. Classification approaches used were: DeepLearning 4J, LibLinear, and Logistic Regression.

2.8. Weighted gene co-expression network analysis (WGCNA) analysis

WGCNA is a systems biology approach that allows for construction of gene co-expression networks and gene modules from gene expression datasets ⁴⁹. Prior to analysis, data were log 2 transformed. Genes with sparse expression were removed and analysis with WGCNA was run on 7864 genes using R software. To generate a weighted network of genes (nodes) and their expression correlations (edges), correlations were raised to a soft thresholding power β of 12. We used unsupervised hierarchical clustering, a minimum module size of 30 genes, the signed mode, the deepSplit parameter set to 2, the mergeCutHeight parameter set to 0.15, and a threshold setting for merging modules of 0.25. Module eigengene values were also evaluated in terms of their relationships with the male groups (identified based on patterns of VTA gene expression, see above) as well as multiple male and partner (i.e., female) traits. The modules were exported as a Cytoscape network file, which was manually trimmed to consist of genes of interest and their gene-to-gene correlations. Genes from one module were visualized with Cytoscape v3.7.1 ⁵⁰.

2.9. Enrichment analysis

Enrichment analysis was run on the gene set produced by differential analysis and on the gene modules produced by WGCNA. This method identifies genes that are over-represented in the larger set and that have been associated with particular functions, drug actions or other phenotypes. Enrichment tools included ToppCluster ⁵¹, GeneOntology ⁵², and EnrichR ⁵³.

3. Results

3.1. Unsupervised approaches identify male groups based on gene expression

Multiple data mining unsupervised approaches reliably and consistently identified two natural male groupings with distinct patterns of gene expression in VTA, which we refer to as the Low and High Dopamine groups (based on results detailed below that indicate differential expression of genes involved in dopamine synthesis and activity as well as survival and differentiation of dopaminergic neurons, see Table 1), with distinct patterns of gene expression in VTA. For four different classification tools within Weka, namely Canopy, Cascade, EM, and Farthest First, the same identical groupings were found. For EM, the first run only provided one large group, but when the setting was modified to find more than one group, it identified the same groupings as the others. LVQ identified the same two groups, except that it placed one member of the Low Dopamine group in the High Dopamine group. Other Weka tools, such as MTree did not find clear separate groups. UMAP identified the same Low and High Dopamine groups using five different settings within the program, namely, correlation, manhattan, braycurtis, canberra, and cosine (see Figure 2). Thus, the same two groups were identified as having distinct patterns of gene expression using multiple clustering tools.

Table 1.

Dopamine-, glutamate-, and GABA-related genes that differ in VTA in the Low and High Dopamine groups.

Gene ID	Gene	Function	Group diff.	p value
Dopamine-related
TH	Tyrosine hydroxylase	Enzyme; catecholamine synthesis	High > Low	0.006
LMX1A	LIM Homeobox Transcription Factor 1 Alpha	Transcription factor; survival of adult dopaminergic neurons	High > Low	0.021
SLC6A2	NE transporter (proposed dopamine transporter in birds)	Dopamine transporter in birds	High > Low	0.029
COMTD1	Catechol-O-Methyltransferase Domain Containing 1	Enzyme; catecholamine degradation	High > Low	0.033
DDC	Dopa Decarboxylase	Enzyme; dopamine, serotonin, histamine synthesis	High > Low	0.036
GDNF	Glial Cell Derived Neurotrophic Factor	Neurotrophic factor; survival and differentiation of dopaminergic neurons, increases dopamine uptake	High > Low	0.045
ARPP21	CAMP Regulated Phosphoprotein 21	Phosphoprotein, enriched in dopaminergic brain regions, mediates dopamine actions	Low > High	0.011
DRD5	Dopamine Receptor D5	Dopamine receptor	Low > High	0.044
Glutamate-related
GRIN2C	Glutamate Ionotropic Receptor NMDA Type Subunit 2C	Glutamate ionotropic NMDA receptor	High > Low	0.027
CACNG2	Calcium voltage-gated channel auxiliary subunit gamma 2	Regulates the trafficking and gating properties of AMPA-selective glutamate receptors	Low > High	0.001
GRM3	Glutamate metabotropic receptor 3	G-protein coupled glutamate receptor	Low > High	0.005
GRIN1	Glutamate ionotropic receptor NMDA type subunit 1	Glutamate ionotropic NMDA receptor	Low > High	0.027
NAALAD2	N-Acetylated Alpha-Linked Acidic Dipeptidase 2	Enzyme, cleaves N-Acetylaspartylglutamic acid, which activates GRM3	Low > High	0.030
LRRTM2	Leucine Rich Repeat Transmembrane Neuronal 2	Regulates expression of AMPA receptors and development of glutamate release sites	Low > High	0.045
GABA-related
GABRP	Gamma-Aminobutyric Acid Type A Receptor Pi Subunit	GABA ionotropic receptor	High > Low	0.022
LAMP5	Lysosomal Associated Membrane Protein Family Member 5	Regulates synaptic plasticity in a subset of GABA neurons	Low > High	0.023
GAD2	Glutamate Decarboxylase 2	Enzyme, catalyzes GABA production	Low > High	0.047

Figure 2.

Multiple clustering tools in UMAP identified the same two groups that we named the Low Dopamine group (blue, lower left) and the High Dopamine group (red, upper right) as having distinct patterns of gene expression in VTA. Plot represents higher dimension data plotted into 2 dimensions. The x and y axes are arbitrary embedding dimensions generated by UMAP. For more details, see ⁴⁷.

3.2. Differential gene expression between groups

We evaluated differential gene expression between the Low and High Dopamine groups and found 338 genes with a p-value less than 0.05, including several genes for neuromodulators underlying motivated behaviors. These include dopamine-related genes and genes related to classically studied modulators of VTA dopamine neuron activity (i.e., glutamate and GABA) (Table 1) as well as other candidate modulators of VTA dopamine neuron activity (Table 2). (Functions for genes in each table were taken from GeneCards, NCBI or literature cited in the discussion section). While none of these reached significance using a False Discovery Rate correction procedure, in prior studies (e.g., ⁵⁴) we have found qPCR validation of top differential expressing genes at this p value level and thus we expect the differences may be biologically meaningful. Enrichment analysis tools allow genes of interest (e.g., genes with significantly altered expression) to be evaluated against preexisting datasets to determine whether an over-representation occurs within a range of biological traits. Here, enrichment analyses revealed that these genes showed an enrichment for neuron development, behavior, and the synapse, suggesting alterations in CNS function. Further, ToppCluster-enrichment analyses revealed that the top drugs that have actions that affect expression of these genes include dopamine, haloperidol, and cocaine, highlighting the presence of a high number of dopamine-related and dopamine-responsive genes. For all expression results, see Supporting information Table S1.

Table 2.

Genes for other neuromodulators of motivated behaviors that differ in VTA in the Low and High Dopamine groups.

Gene ID	Gene	Function	Group diff.	p value
HDC	Histidine decarboxylase	Enzyme, catalyzes histamine production	High > Low	0.004
HNMT	Histamine N-methyltransferase	Enzyme, inactivates histamine	High > Low	0.005
VIP	Vasoactive intestinal peptide	Neuropeptide	High > Low	0.006
MC4R	Melanocortin 4 Receptor	Melanocortin receptor	High > Low	0.016
MTNR1B	Melatonin Receptor 1B	Melatonin receptor	High > Low	0.036
NPY5R	Neuropeptide Y Receptor Y5	NPY and peptide YY receptor	High > Low	0.040
MC3R	Melanocortin 3 Receptor	Melanocortin receptor	High > Low	0.042
HRH1	Histamine Receptor H1	Histamine receptor	Low > High	0.007
HTR1B	5-Hydroxytryptamine Receptor 1B	Serotonin receptor	Low > High	0.011

3.3. Connections of brain expression profiles with behavior and partner traits

We then used machine learning tools to identify which of the 30 behavioral and physical traits of the males and partner females best predict the Low and High Dopamine groupings of males based on VTA gene expression. The SVM Attribute Evaluator within Weka provided a rank order of the 30 traits (Table 3). The top five traits all relate to female behavior. Further, when only these five features are used, classification tools (Dl4j, Liblinear, Logistic) predict the two groups with 100% accuracy. We provide figures for the top 10 traits to illustrate the tendency for male groups to differ most with respect to traits characteristic of their female partners (Figure 3). Summary statistics for all traits are available in Supporting information Table S2.

Table 3.

Ranking of 30 behavioral and physical traits in order of their ability to differentially predict VTA gene expression in the two distinct male groups (Low and High Dopamine) identified using machine learning approaches. The top 5 traits (shaded) predict the two groups with 100% accuracy. Pink indicates the trait or behavior was produced by the female partner. Blue indicates it was produced by the male subject. White indicates that both the male and female contributed.

Ranking	Trait	Details
1	# of courtship behaviors	performed by the female
2	# of nesting behaviors	performed by the female
3	Final nest weight (g)	likely determined by male (see text)
4	# drinking bouts (alone)	performed by the female
5	# of beak wipes	performed by the female
6	# of agonistic behaviors	performed by the female
7	Largest gonad (mm2)	for the male
8	# of approach behaviors	performed by the female
9	# of undirected songs	performed by the male
10	# of calls	performed by the male
11	# of courtship behaviors	performed by the male
12	Weight (g)	for the female
13	# of contact behaviors	performed by the male
14	# of contact behaviors	performed by the female
15	Largest gonad (mm2)	for the female
16	Weight (g)	for the male
17	# of calls	performed by the female
18	Beak hue (upper mandible)	for the female
19	Beak hue (upper mandible)	for the male
20	# autopreen bouts	performed by the female
21	# of synchronous behaviors	performed by both
22	# of nesting behaviors	performed by the male
23	# feeding bouts (alone)	performed by the male
24	# of beakwipes	performed by the male
25	# of approach behaviors	performed by the male
26	# feeding bouts (alone)	performed by the female
27	# of eggs	performed by the female
28	# drinking bouts (alone)	performed by the male
29	# of agonistic behaviors	performed by the male
30	# autopreen bouts	performed by the male

Figure 3.

Bar plots for each individual male (N=10) illustrating the top 10 traits found to differentially predict gene expression in the 2 distinct male Low and High Dopamine groups identified using machine learning approaches. Plots are organized to show A) behaviors related to reproduction that were performed by the male’s female partner, B) non-specific behaviors performed by the male’s female partner, and C) the male’s own traits. Each bar represents the sum of trait values (counts for behaviors, grams for nest weight) for an individual male or his partner. The 5 bars on the left = Low Dopamine Group. The 5 bars on the right = High Dopamine Group. Summary statistics for all traits are available in Supporting information Table S2.

Although elevated accuracy is expected after feature reduction, high accuracy does not always occur with only a few features. This approach is similar to training in a machine learning model and does not reflect accuracy on an independent test set.

3.4. WGCNA results

WGCNA identified 27 modules of genes (Supporting information Table S3, “modules” tab). The creation of the modules in WGCNA is unsupervised and independent of any group identification. Relationships between these modules were then calculated. A full listing of the modules, the member genes, and how those modules relate to traits is provided in Supporting information Table S3. The darkred module of 53 genes was significantly enriched for dopamine biosynthetic process (in ToppCluster). The module also includes multiple genes related to dopamine activity (i.e., TH, DDC, GCH1, and SLC6A2) with TH found to be the central hub (the most connected gene) of the module. A plot of the relationship of a subset of genes in this module is provided in Figure 4. The module was significantly related to number of eggs, the male’s agonistic behavior, the female partner’s body weight, the female partner’s beak wiping and the female partner’s contact behavior. To gain insight into relationships between gene expression in VTA and bond-related behaviors, we plotted relationships between gene modules and a subset of traits associated with reproduction and affiliative behaviors (Figure 5). We also ran correlation analyses for each of the four dopamine-related genes highlighted above and the pair-bond related behaviors above that related significantly to the darkred module (i.e., a male’s agonistic behavior, female contact behavior, and number of eggs laid). Significant positive correlations were found between male agonistic behaviors and both TH and SLC6A2 expression. In contrast, negative correlations were found between contact behaviors from the female and TH, SLC6A2, DDC, and GCH1 expression. Negative correlations were also found between the number of eggs laid and DDC and GCH1 (statistical details presented in Figure 6). The modules, turquoise, magenta, black, and orange, were significantly related to the Low and High Dopamine groups identified above. Turquoise was the largest of all the identified gene modules (5404 genes) and it included most of our genes of interest that relate to neuromodulation (Supporting information Table S3, “modules” tab) and was significantly related to the female partner’s courtship behavior (Figure 5). The magenta module (92 genes) included histamine receptor (HRH1) and was significantly related to the male’s courtship behavior (Figure 5). The black (114 genes) and orange (41 genes) modules did not include any of our genes identified to be involved in neuromodulation.

Figure 4.

WGCNA network of dopamine-related genes with highly correlated expression. Within the darkred module identified by WGCNA as having highly correlated expression, multiple dopamine related genes are found, including TH, DDC, GCH1, and SLC6A2 (highlighted in yellow). Lines represent significant correlations between two genes. The top genes with the most significant connections to other genes are plotted in Cytoscape using weight (level of significance) as a factor.

Figure 5.

Statistical associations between expression profiles for each of the WGCNA identified modules and a subset of traits related to reproduction and affiliation for both males and their female partners. Partner traits are indicated by a “p”. Boxes contain correlation coefficients (p values). The genes comprising each module are listed in Supporting information Table S3. Note that the grey module is not a true module, but rather includes only those genes not assigned to other modules.

Figure 6.

Correlations between four dopamine-related genes highlighted in Figure 4 and the pair-bond related behaviors above that related significantly to the darkred module (i.e., a male’s agonistic behavior, female contact behavior, number of eggs laid). Totals = Sum of behaviors across all 5, 20 min observation periods and total number of eggs during the entire study. Red line = significant correlation.

3.5. qPCR results

Correlation analyses revealed significant positive correlations between qPCR and RNA-seq measures for TH (r = 0.83, p = 0.0028) and SLC6A2 (r = 0.77, p = 0.0099). For DDC the relationship was positive, but not significant (r = 0.24, p = 0.50). We were unable to run qPCR for GCH1 because we did not have remaining tissue samples. (We had already used tissue to measure other genes as part of a separate project.) We note that the effectiveness of using qPCR to verify RNA-seq results is not clear ^55,56 and discrepencies may relate to bias in the qPCR based on which region of the cDNA is amplified (for additional discussion see ⁵⁷). Thus we do not suggest that the lack of a correlation between qPCR and RNA-seq for DDC indicates a lack of validity for the RNA-seq results.

4. Discussion

This study on male zebra finches provides insight into complex associations between gene expression in the VTA and multiple traits associated with the maintenance of long-term, affiliative pair bonds. Using multiple approaches, we identified numerous candidate genes associated with pair-related traits and in particular found that dopamine-related genes in VTA differed in association with individual differences in pair-related behaviors. This was expected given that multiple studies across species demonstrate that mesolimbic dopamine neurons that originate in VTA underlie an individual’s motivational state and associated behavioral responses to rewarding stimuli (e.g., mates, food, and drugs ^58–62). However, many of the behaviors that related most closely to patterns of VTA gene expression were produced by the male’s partner, rather than the male himself. This finding is not unprecedented, with similar correlations reported between protein markers in VTA and the receipt of social stimuli in prior studies, including studies of songbirds ^24,35. These findings raise the possibility that the molecular/genetic organization of the VTA may be shaped by external social/sensory input and/or pair bond strength.

4.1. Males could be categorized based on distinct patterns of VTA gene expression

Multiple unsupervised learning approaches revealed that individual males could be categorized reliably into two distinct groups based solely upon patterns of gene expression in VTA. Three hundred thirty-eight genes differed significantly between the two groups. Enrichment analyses revealed high numbers of the genes that differed in the two groups to be dopamine-related and dopamine-responsive. These differences allowed us to assign males to what we refer to as the Low and High Dopamine groups, as detailed below. We will focus here on genes related to the modulation of motivated behaviors, but note that genes that differed between these two groups were also enriched for neuron development and function, gene regulation, and the synapse. We note that this categorization reduces the sample size to two groups of 5, which reduces the power of the analysis of the Low and High Dopamine groups.

Differences between the Low and High Dopamine groups were found for at least eight dopamine-related genes, with the High Dopamine group showing greater expression levels than the Low Dopamine group for the majority of genes identified. These include genes for dopamine synthetic enzymes (i.e., tyrosine hydroxylase [TH] and dopa decarboxylase [DDC]), a dopamine degradative enzyme (COMTD1), a dopamine receptor (dopamine receptor D5 [DRD5]), as well as the avian dopamine reuptake transporter (solute carrier family member 2 [SLC6A2]), which acts as a norepinephrine transporter in mammals ⁶³. Thus, the High Dopamine group appears to have a VTA that is primed for enhanced dopaminergic activity relative to the Low Dopamine group.

Numerous genes related to known modulators of dopaminergic neuronal activity also differ between the Low and High Dopamine groups, suggesting that mechanisms influencing dopamine neuronal activity in VTA also differ between the two groups. These include genes for the well-studied modulators of VTA dopaminergic neuronal activity, glutamate and GABA ²⁵, as well as numerous candidate genes related to neuromodulators that are reported to modify VTA dopaminergic neuronal activity either directly or indirectly (e.g., by influencing GABAergic neuronal activity). These candidates include genes related to histamine, melanocortin, melatonin, and serotonin ^64–71.

4.2. Distinct patterns of VTA gene expression in males related to female partner behaviors

Past studies show correlations between a variety of neuronal markers in VTA and bond-related behaviors observed in males, including the production of courtship and nesting behaviors ^24,27–34. Based on these studies, we expected to find that the distinct patterns of gene expression identified between males in the Low and High Dopamine groups would be associated with differences in the numbers of bond-related behaviors produced by the males. However, machine learning approaches revealed that pair-related behaviors performed by the male’s female partner most accurately predicted whether a male was in the Low or High Dopamine group. Specifically, males in the High Dopamine group had female partners that performed much higher rates of nesting and courtship than the females paired with males in the Low Dopamine group. In contrast, males in the Low Dopamine group received a higher level of agonistic behavior from their female partners than males in the High Dopamine group. Zebra finches in the present study were observed at a time when pair bonding would have already occurred, and incubation would be taking place in pairs that had successfully established sufficiently strong pair bonds. However, it is not expected that all randomly assigned pairs would be equally compatible mates. Variations in female behavior two weeks after force pairing may be more indicative of the strength of the pair bond than variations in male behavior. Although causal relationships must be tested in future studies, these associations between female partner behavior and gene expression in the male’s VTA raise the possibility that the genetic architecture of a male’s VTA, and consequent dopaminergic signaling, is altered by the degree to which he is solicited (the High Dopamine group) or rejected (the Low Dopamine group) by his partner and/or the strength of their pair bond.

4.3. WGCNA reveals associations between gene modules in VTA and behavior

As a second approach, we used WGCNA to examine relationships between modules of genes and the traits of both the male and his female partner. WGCNA clustered the VTA transcriptome into 27 modules. The darkred module of 53 genes stood out for a few reasons. One is that it was significantly enriched for dopamine biosynthetic process (in ToppCluster). The module includes TH, a key enzyme in the synthesis of dopamine, as the central hub (the most connected gene) of the module. Further, the module contains DDC, another key enzyme in dopamine synthesis, as well as GTP cyclohydrolase 1 (GCH1), which is involved in the early stage of amino acid modification prior to dopamine synthesis. The avian dopamine reuptake transporter SLC6A2 was also in this module. Thus, in the relatively small darkred module we identified multiple genes related to dopamine availability. With respect to bond-related behaviors, the darkred module related significantly to the male’s own agonistic behavior, the number of times his female partner engaged in contact behaviors (i.e., clumping, allopreening, soliciting allopreening), and the number of eggs that she produced.

Correlation analyses focused selectively on the four dopamine-related genes and behaviors highlighted in the darkred module revealed positive correlations between male agonistic behavior and both TH and SLC6A2 expression in VTA. In contrast, negative correlations were observed between the expression of each dopamine-related gene (TH, SLC6A2, DDC, and GCH1) and the amount of contact a male received from his female partner. The number of eggs females laid also related negatively to both DDC and GCH1 expression. Thus both the WGCNA approach and the unsupervised learning approaches reveal that dopamine-related gene expression in the male VTA is closely associated with the behavioral traits of his partner. The functional relevance of these correlations must now be tested.

Two modules, the turquoise and magenta modules, significantly correlated with both group (Low or High Dopamine) and courtship behavior and included genes related to known dopaminergic modulators. The turquoise module was the largest of all 27 modules, consisting of 5404 genes, including multiple genes related to dopamine, glutamate, GABA, histamine, serotonin, VIP, and melanocortin. It was also significantly related to the female partner’s courtship behavior. The size of this module suggests highly complex and interconnected neural processes that occur in the VTA and relate to both the Low or High Dopamine grouping and to the amount of courtship behavior received from the partner. The magenta module included a histamine receptor gene (HRH1) and correlated with the Low or High Dopamine grouping and the male’s own courtship behavior. These two modules highlight neuromodulators that are known to influence both dopaminergic activity and behavior in their own right and their roles in pair bonding should be investigated further.

4.4. Conclusions and future directions

Despite the established importance of the VTA in motivated, reward-directed behaviors, few studies focus on the role of VTA in pair bonding. This may in part be because of difficulties in understanding the complex interactions between multiple neuromodulators that fine tune VTA activity and consequent motivated behaviors and the fact that different neurons have different projections (reviewed in ²⁵), which may also have distinct molecular profiles that should be examined in future studies. The bioinformatics approaches used here help to fill this gap by providing insights into relationships between over 8000 genes expressed in the VTA and specific aspects of pair bonding-related behaviors. We focus our discussion on what our approaches reveal about candidate genes that are likely to play a role in fine tuning dopamine-neuronal activity that may modify dyadic interactions in male-female pairs. However, this paper provides researchers with data mining opportunities to identify relationships between traits and any subset of these 8000+ genes.

Social relationships are critical for mental health, and multiple psychiatric disorders are associated with a decrease in the motivation to pursue social relationships and/or the ability to maintain these relationships. In humans and in non-human animals, affiliative behaviors are important for establishing social relationships, including the formation of pair bonds. The function and neurochemical regulation of the VTA in motivated behaviors is considered well-conserved across vertebrate species ⁷². Thus by identifying candidate genes in the VTA involved in the maintenance of long-term social relationships in birds we may provide insights into critical neurogenetic modulators that are central to positive social interactions across vertebrates.

Supplementary Material

Supplementary Table 2

Supporting Table 2. Summary statistics for the 30 trait variables for the Low and High Dopamine groups.

Supplementary Table 1

Supporting Table 1. Tab 1) Expression 8K+ genes: RNA-Seq data used to analyze differential gene expression using EdgeR. Tab 2) EdgeR results: Results of a differential gene expression analysis of RNA-seq data using EdgeR.

Supplementary Table 3

Supporting Table 3. Tab 1) modules: Gene modules identified by WGCNA, Tab 2) modules x traits 1: Relationships between WGCNA identified modules and both a male’s own traits and those of his female partner. Tab 3) modules x traits 2: Relationships between WGCNA identified modules and both a male’s own traits and those of his female partner.