Introduction

Brain oxytocin receptors (OXTR) regulate a wide range of social behaviors including social recognition, maternal care, social bonding, empathy related behaviors, and aggression (Froemke & Young, 2021; Jurek & Neumann, 2018; Rigney, de Vries, Petrulis, & Young, 2022). In contrast to sex steroid receptors, which have highly conserved brain expression patterns across species, OXTR shows remarkable inter- and intra-species variation in brain distribution (Froemke & Young, 2021; Larry J. Young & Zhang, 2021). For example, monogamous prairie voles have a brain OXTR distribution in cortex, striatum and amygdala that is significantly distinct from promiscuous montane voles and laboratory rats and mice(Larry J. Young & Zhang, 2021). Furthermore, there is robust individual variation in brain OXTR expression among prairie voles that is associated with single nucleotide polymorphisms (SNPs) in the OXTR locus (Oxtr), and this variation predicts pair bonding behavior and resilience to early life social neglect (Ahern, Olsen, Tudino, & Beery, 2021; Barrett, Arambula, & Young, 2015; King, Walum, Inoue, Eyrich, & Young, 2016; Ross et al., 2009). Different species of primates also differ in brain OXTR distribution (Rogers Flattery et al., 2022), and SNPs in the human OXTR have been linked to variation in social function (Skuse et al., 2014; Theofanopoulou, Andirkó, Boeckx, & Jarvis, 2022). Thus, variation in OXTR distribution in brain likely constitutes a mechanism for the emergence of diverse social traits, potentially including psychiatric endophenotypes.

The evolution of gene expression patterns could be mediated by cis (via linked polymorphisms) or trans (through diffusible products of distal genes, e.g., transcription factors) changes, as a result of adaptation to the environment (Mack, Campbell, & Nachman, 2016; Osada, Miyagi, & Takahashi, 2017; Shi et al., 2012). Cis-regulatory differences are more commonly responsible for adaptive evolution and interspecific divergence (Signor & Nuzhdin, 2018). To explore the transcriptional mechanisms giving rise to species-specific Oxtr expression and social behavior, we created transgenic mice using a prairie vole Oxtr (pvOxtr) bacterial artificial chromosome (BAC). The BAC construct covers the full coding sequence, introns and the entire intergenic region of the pvOxtr with its neighboring genes. We expected that our BAC construct accommodates all the critical regulatory sequences responsible for species- and tissue-specific expression, so that pvOxtr transgenic mice would express brain pvOxtr in a prairie vole-like pattern, largely independent of integration sites. Then we would examine if our transgenic mouse could develop social behaviors similar to that of prairie vole.

We generated eight pvOxtr mouse lines with successful germline transmission. Each line, defined by a distinct BAC integration site, exhibited a unique brain-specific expression pattern of Oxtr, distinct from both wildtype mice and prairie voles. Notably, Oxtr expression in the mammary gland was conserved across all lines. This suggests that brain, but not mammary gland, expression of Oxtr is shaped by distal regulatory elements beyond the BAC insert. These findings imply that evolution has placed brain Oxtr expression under more complex regulatory control, likely to accommodate the diversification of social behaviors, whereas mammary gland expression is maintained by simpler mechanisms supporting conserved lactation functions. Furthermore, behavioral assays revealed line-specific differences in partner preference and maternal behaviors.

The 3D architecture of the genome, segregated into TADs demarcated by CCCTC-binding factor (CTCF), crucially influences gene expression. Through the loop extrusion model, CTCF enables interactions between distant regulatory elements, facilitating precise gene regulation, which ultimately shapes an organism’s phenotype and behavior(Wendt et al., 2008). Hi-C data provide a genome-wide view of chromosome conformation in multiple layers, such as compartments, TADs, and loops(Gibcus & Dekker, 2013; Lieberman-Aiden et al., 2009). Our TAD analysis with Hi-C database showed a clear difference exists between the 3D structures of Oxt and Oxtr gene, and a clear difference exists between the 3D structures of brain tissue Oxtr and peripheral tissue Oxtr, suggesting that Oxtr expression patterns in brain depends on contributions from very distal sequences. These results provide important clues to understand the genetic mechanism underlying species differences in brain gene expression.

Together, our data demonstrate that transcriptional redistribution of brain Oxtr can drive variation in social and parental behaviors—variation that natural selection could act upon in specific ecological contexts. Our results also offer insights into the genetic mechanisms underlying species-specific brain gene expression.

Results

Creation of the pvOxtr-P2A-Cre BAC transgenic mice

BAC vectors can accommodate dispersed cis-regulatory elements across large regions of genome. Transgene expression from BAC constructs are generally resistant to insertion position effects (Suster, Abe, Schouw, & Kawakami, 2011) due to both the large spans of insulating genetic material which protect the transgene cassette from the influence of the chromosomal environment, and through the inclusion of necessary regulatory elements such as enhancers, silencers, locus control regions, and matrix attachment regions (Beil, Fairbairn, Pelczar, & Buch, 2012; Bian & Belmont, 2010). Large-scale screening of CNS gene expression in BAC transgenic mice by the GENSAT (Gene Expression Nervous System Atlas) Project found that more than 85% of BACs express reproducibly in multiple independent transgenic lines and reporter gene expression faithfully recapitulates endogenous gene expression patterns (Gerfen, Paletzki, & Heintz, 2013; Heintz, 2004; Schmidt, Kus, Gong, & Heintz, 2013). Indeed, reproducible expression from mouse BAC vectors has been achieved for more than 500 genes (Gerfen et al., 2013; Heintz, 2004; Schmidt et al., 2013). Few that don’t express reproducibly are influenced by variation in BAC copy number and insertion sites (S. Gong et al., 2007). BAC engineering has also been used to mediate cross-species transgene expression. A large number of mouse models of human dominant neurodegenerative disorders have been created using human BAC transgene approaches. These model mice successfully express human transgenes in human-like patterns and recapitulate disease-like phenotypes (Crook & Housman, 2011; Ferrante, 2009; Shenoy et al., 2022; Taguchi et al., 2019; Yang & Lu, 2008). Previously we constructed Pkcd-BAC-Cre transgenic mice and obtained independent lines with reproducible expression patterns as well (Zhang et al., 2016)(Supplementary Fig. 1).We therefore decided to use BAC approaches to create “volized” Oxtr mouse lines.

To create pvOxtr-P2A-Cre BAC transgenic mouse line, the prairie vole BAC clone we used includes the entire pvOxtr locus (the coding sequence and all the introns) with the full intergenic sequence upstream (150kb) and downstream of the pvOxtr coding region (9kb) (Fig.1 B). The BAC clone contains a portion of Rad18, minus the first 11 exons, located upstream of pvOxtr, and Cav3, minus the first exon including the start codon, downstream of pvOxtr. Therefore, there will be no interference of these two flanking transgenes on behavior analysis, and the proximal topological structure around pvOxtr locus is preserved to the maximum extent. A P2A-NLS-Cre cassette was inserted in-frame just before the Oxtr stop codon to ensure co-expression of Cre and Oxtr (Fig.1 C). Transgenic mice were generated by pronuclear injection using C57BL/6J zygotes. Proper integration of the reporter into the BAC clone was confirmed using gene specific PCR assays and sequencing (Fig.1 C, D and data not shown). We obtained 11 founder mice carrying the transgenes, which was verified by PCR with multiple primer sets (Fig.1 C, E). One founder line was sterile, and two founder lines did not show germ-line transmission. Eight founder lines successfully produced offspring and stably transmitted the transgene to at least 3 generations. We named these lines “Koi”, meaning “love” in Japanese.

Fig. 1

Each Koi line showed a unique expression pattern of transgene in brain, yet conserved expression in mammary gland

We systemically evaluated 8 Koi lines by crossing them with ROSA-26-NLS-LacZ mice, a Cre reporter mouse line expressing nuclear-localized beta-galactosidase (Kandasamy et al., 2021; Zhang et al., 2016). Since Lac-Z is only expressed in the nucleus of double positive (CRE+, LacZ+) cells, but not in the neuronal processes, we could easily assess the expression pattern of our transgene. We found that each Koi founder line displayed a unique pattern of expression in brain distinct from that of wildtype mice or prairie voles (Fig.2 A-O, and supplementary Fig.2). Almost all Koi lines showed transgene expression in olfactory bulb, lateral septum (Fig.2 F-I and Supplementary Fig.2 B,E,H,K) and ventromedial nucleus of the hypothalamus (Fig.2 K-N and Supplementary Fig.2 C,F,I,L), though with different intensities, suggesting that these regions constitute a stable core Oxtr expression network, as these regions also express OXTR in mice and voles (Inoue, Ford, Horie, & Young, 2022; Newmaster et al., 2020). Several Koi lines expressed the transgene in some brain regions specific for vole Oxtr expression, including nucleus accumbens (NAc) (Fig.2 A,B,D and Supplementary Fig.2 A,G,J), prefrontal cortex (PFC) (Fig.2 A,B,C and Supplementary Fig.2 J), lateral amygdala (Fig.2 K,L,M and Supplementary Fig.2 C,I,L) and deep layers of cingulate cortex (Fig.2A,B,C and Supplementary Fig.2 C,L), which suggested that cis regulatory elements in the BAC are capable of mediating the expression of Oxtr in the reward and reinforcement circuitry of brain, albeit depending on the sequence/structure of distal sequences >150 kb upstream and >21 kb downstream of the transcription start site. None of the Koi lines exactly “mirrored” the vole-specific expression patten. Six of the Koi lines expressed the transgene in broad areas of thalamus (Fig.2 K,L,N and Supplementary Fig.2 C,I,L), reminiscent of the strong expression of V1a vasopressin receptor (Avpr1a) in the thalamus (Larry J. Young & Zhang, 2021). Intriguingly, the expression of OXTR in mammary glands showed similar pattern among all the Koi lines and mouseOxtr(mOxtr)-Ires-Cre knock-in line (Fig.2 P-T and Supplementary Fig.3).

Fig. 2

Transgenes that are introduced by pronuclear injection typically integrate into a single site of the genome as tandem concatemers (Smirnov et al., 2019). BAC copy number variation and different insertional loci can lead to distinct patterns of ectopic expression, and increased BAC transgene copy numbers often correlate with increased BAC gene expression (S. Gong et al., 2007). We then examined the relationship between transgene copy number and LacZ signal using quantitative genomic PCR (Chandler et al., 2007). As expected, the copy number of WT, heterozygous and homozygous mOxtr-Ires-Cre knock-in mice was 0, 1 and 2 respectively (Figure.2 Z). The copy number of transgene in Koi lines are: Koi-1 (12), Koi-2 (2), Koi-3 (2), Koi-4 (1), Koi-5 (2), Koi-6 (4), Koi-7 (9), Koi-8 (3). Koi-6 and Koi-7 showed the most limited expression of reporter gene, though the copy number of transgene were 4 and 9 respectively. Koi-4 showed strong expression of reporter gene in NAc, septum and thalamus, though there was only 1 copy of transgene. Therefore, integration site rather than copy number is the reason of the variant BAC transgene expression patterns across our Koi lines.

Adult brain OXTR binding patterns differ across Koi lines

To detect the distribution of OXTR protein in adult Koi lines without the interference of endogenous mouse OXTR (mOXTR) signal, Koi lines were crossed with mOxtr^-/- line twice sequentially to produce Koi: mOxtr^-/- mice, and pvOXTR binding was detected using receptor autoradiography. We focused on analyzing 4 lines: Koi-1: mOxtr^-/-, Koi-2: mOxtr^-/-, Koi-3: mOxtr^-/- and Koi-4: mOxtr^-/- based on the strong expression of reporter gene in behaviorally relevant neuronal populations. Autoradiography demonstrated functional pvOXTR in brain resembling to some extent Cre-induced reporter gene expression (Fig.3). Brain OXTR binding of Koi-3: mOxtr^-/- was consistent with Lac-Z expression in the reporter line (Fig.2 C, H, M and Fig.3 C, H, M). Interestingly, the OXTR binding in PFC of Koi-3: mOxtr^-/- was similar to that of prairie vole (Figure.3 C). In contrast, Koi-4: mOxtr^-/- showed low OXTR binding in NAc when compared with the high Lac-Z staining (Fig.2 D and Fig.3 D). Koi-2: mOxtr^-/-, the line which displayed the expression of reporter genes in the broadest regions, did not show strong signal in autoradiography. It is interesting that Layer IV of cortex in Koi-4: mOxtr^-/- showed strong OXTR binding (Figure.3 I and N), which is highly likely to originate from the axon projections from the thalamus (Fig.2 N), as was previously reported for NAc projections in voles (Inoue et al., 2022). Consistent with Lac-Z expression, we do not see a clear relationship between OXTR binding and transgene copy number. For example, Koi-1: mOxtr^-/- showed rather low OXTR signal (Fig.3 A, F, K) with the highest copy number, while Koi-4: mOxtr^-/- showed very strong OXTR signal at lateral septum and thalamus (Fig.3 I, N), with the lowest copy number.

Fig. 3

Oxtr mRNA level in Koi-3: mOxtr^-/- and Koi-4: mOxtr^-/- was further investigated with Fluorescent RNAscope in situ hybridization (Fig.4). Consistent with Lac-Z expression in the reporter line, Koi-3: mOxtr^-/- showed strong signal in PFC and Ctx (Fig.4 A, C), while Koi-4: mOxtr^-/- showed prominent signal in NAc and Thalamus (Fig.4 B, D). qRT-PCR analysis further revealed that adult Koi-4: mOxtr^-/- mice had the highest Oxtr mRNA signal in the ventral striatum, several fold higher than WT, consistent with the Lac-Z staining in the reporter line (Fig.5 A, B). The discrepancy of OXTR signal in NAc of Koi-4: mOxtr^-/- between autoradiography and Lac-Z staining/RNAscope/qRT-PCR may, therefore, be due to the translational regulation or differences in the sensitivity of the detecting methods.

Fig. 4

Fig. 5

Although none of our Koi lines mirrored the endogenous OXTR expression in prairie voles, we confirmed strong expression of pvOxtr in olfactory bulb of Koi-1line, PFC and amygdala of Koi-3line, striatum and thalamus of Koi-4 lines, and the broad expression of pvOxtr in Koi-2 line. As OXTR signaling in some of these regions are similar to that in voles, albeit in separate lines, and postulated to be involved in vole social behavior (H. Walum & L. J. Young, 2018), we used these lines to address the question of whether variation in Oxtr transcription in brain can lead to variation in social behaviors.

Different Koi lines showed diverse behaviors in a Partner Preference Test

The partner preference test is widely used as a laboratory proxy for pair bonding in prairie voles. In a pilot study, we found that Kio-4 female mice on a mOxt^+/+ background showed a preference for their mate relative to a novel male. We then focused on the females of 4 lines (Koi-1, 2, 3, 4) based on transgene expression in PFC, BLA or NAc. Again, we bred the Koi lines with mOxtr^-/- mice (both on C57/B6 background) to generate Koi: mOxtr^-/- offspring expressing only pvOxtr. Partner preference tests (PPT) were performed in ovariectomized females after 21 days of cohabitation with 3 mating bouts elicited by estrogen and progesterone injections on days 2, 9 and 16. We divided our analysis into 3 phases (Fig.6 A, B). In the habituation phase, the experimental subject was habituated to the chamber for 5 minutes with both pen-boxes empty. In the recognition phase, the partner stimulus animal was restricted in one pen-box, and a novel “stranger” male was restricted in another pen-box, and the experimental female was allowed to explore the arena for 10 minutes to test social novelty preference. In the preference phase, the positions of the partner stimulus animal and the “stranger” stimulus animal were switched to avoid position bias, and the female was allowed to explore the chamber for an additional 20 minutes.

Fig. 6

Mixed ANOVA was conducted to examine the effect of genotype and the stimulus box on the area stay time during all the phases. In the habituation phase (Fig.6 C), there was no statistically significant interaction between the effects of genotype and position of empty pen-box on stay time, F (4, 60) =0.69, p=0.6. There was no difference in position preference for all the groups, F (1,60) =0.03, p=0.87. In the recognition phase (Fig.6 D), there was a main effect of stimulus animal (i.e., partner or stranger) on stay time, the time staying in close proximity to the partner was significantly shorter than the time stay with stranger animal (F (1,60) =15.38, p=0.0003). There was a main effect of genotype on social stay time, F (4,60) =2.62, p=0.04, and a significant interaction between the effects of genotype and stimulus animal on area stay time, F (4, 60) =2.61, p=0.04. Bonferroni adjusted pairwise comparisons were applied to analyze the simple main effects. Koi-4: mOxtr^-/- and WT mice spent significantly longer time in “stranger” area than in “partner” area (p=0.0001 for Koi-4: mOxtr^-/-; p=0.028 for WT), demonstrating a typical novelty preference. There was no significant difference between the stay time with “stranger” and “partner” for Koi-1: mOxtr^-/- (p=0.96), Koi-2: mOxtr^-/- (p=0.49) and Koi-3: mOxtr^-/- (p=0.09) mice. In the preference phase (Fig.6 E), conducted after both stimulus mice were presumably familiar following the initial 10 min recognition phase, there was no main effect of stimulus animal on stay time, F (1,60) =0.04, p=0.84. There was no main effect of genotype on stay time either, F (4,60) =1.36, p=0.26. There was a statistically significant interaction between the effects of genotype and stimulus animal on area stay time, F (4, 60) = 4.64, p=0.003. Bonferroni adjusted pairwise comparisons revealed that Koi-2: mOxtr^-/- mice spent significantly longer time in “stranger” area than in “partner” area (p=0.024). In contrast, Koi-4: mOxtr^-/- mice spent significantly longer time in “partner” area than in “stranger” area (p=0.002). There was no significant difference between the stay time with “stranger” and “partner” for WT (p=0.99), Koi-1: mOxtr^-/- p=0.16) or Koi-3: mOxtr^-/- (p=0.45) mice.

Different Koi lines showed diverse behaviors in Pup Retrieval Test

Based on the abundant pvOXTR of Koi-3and Koi-4 in reward-related brain regions (Fig.3 C,D,H,I,M,N), and the role of OXTR in mediated maternal care in mice and voles (Carcea et al., 2021; Marlin, Mitre, D’Amour J, Chao, & Froemke, 2015; Olazábal & Young, 2006), we further tested the parental behaviors of these two lines in virgins (Fig.6 F, G). Mixed ANOVA was conducted to examine the effect of genotype and pup number (1-3) on the latency of pup retrieval. Mauchly’s test showed that the sphericity was violated. Therefore, the Greenhouse-Geisser correction was used for the repeated measures ANOVA. There was a significant main effect of genotype on latency to retrieve, F (2,48) =29.58, p<0.0001. Koi-4: mOxtr^-/- took a significantly longer time to retrieve all the pups than WT and Koi-3: mOxtr^-/- (p<0.0001 for both). There was also a significant interaction between the effects of genotype and retrieved pup number on the retrieval latency, F (2.40, 57.49) = 4.45, p =0.011. Post Hoc multiple comparisons with Bonferroni correction revealed that Koi-4: mOxtr^-/- took a significantly longer time to accomplish the 1st retrieval (p=0.0003 and p=0.0005), the 2nd retrieval (p<0.0001), and the 3rd retrieval (p<0.0001) when compare to WT and Koi-3: mOxtr^-/-. And Koi-3: mOxtr^-/- took a significantly shorter time to accomplish the 1st retrieval (p=0.01), the 2nd retrieval (p=0.006), and the 3rd retrieval (p=0.003) when compare to WT.

Crouching time was analyzed as well (Fig.6 F, H). There was a statistically significant difference between groups as determined by one-way ANOVA (F (2,48) = 27.56, p <0.0001). A Bonferroni post hoc test revealed that the crouching time of Koi-3: mOxtr^-/- was statistically significantly longer than WT (p=0.008) and Koi-4: mOxtr^-/- (p<0.0001). Additionally, the crouching time of Koi-4: mOxtr^-/- was statistically significantly shorter than WT (p=0.001). Therefore, two Koi lines with different OXTR expression pattern demonstrated quantitative differences in parental behavior tests.

The TAD structure of brain Oxtr is markedly distinct from that of both Oxt and peripheral Oxtr

The divergent brain expression patterns across our Koi lines suggest that distal elements external to our BAC cassette contribute to the variation in Oxtr expression. TADs represent a key feature of hierarchical genome organization comprising regions of high chromatin inter-connectivity, and TAD formation appears to be linked to transcription regulation (Beagan & Phillips-Cremins, 2020; Dekker, Rippe, Dekker, & Kleckner, 2002; Dixon et al., 2012; Rajderkar et al., 2023). TADs are generally conserved within the same tissue type across different species. We therefore utilized publicly available 3D chromatin datasets from mouse (Extended data Table7-1) and analyzed the TAD structure of Oxtr and Oxt, both in brain related tissues/cells and peripheral tissues/cells. We discovered that Oxtr is consistently located at the boundary region of TAD structures across several tissues and cell lines (Fig.7 A). Specifically, only in brain related samples (whole brain tissue, neural progenitor cells, and cortex tissue), Oxtr is situated at the boundary of a large inter-TAD interaction between two neighboring TADs (Fig.7 A). In contrast, the oxytocin peptide gene, Oxt, which is expressed in a highly conserved brain pattern across vertebrates, lacks organized TAD structures (Fig.7 B). These findings demonstrate that the TAD landscape differ dramatically between Oxtr and Oxt, and highlight that the 3D genomic architecture surrounding the Oxtr gene varies between brain tissues/cells and peripheral tissue/cells. There exists clear long-distance interaction in brain Oxtr gene TAD structure. The presence of extensive long-distance chromatin interactions around Oxtr in brain tissues may at least partially explain why its expression is hypersensitive to distal regulatory sequences beyond the range of a BAC construct. Moreover, such large inter-TAD interactions likely expose Oxtr regulation to a broader set of cis- and trans-acting influences, rendering brain Oxtr expression inherently more variable across individuals and species.

Fig. 7

Discussion

Diversity of Oxtr expression patterns causes diversity of social behaviors

The diverse OXTR expression of our Koi lines is reminiscent of the diverse OXTR expression across different species. The unique expression pattern of Oxtr in monogamous prairie vole was mainly observed in PFC, NAc and BLA. We obtained expression in all of these specific regions, but in 3 separate lines. These Koi lines with distinct brain OXTR expression pattern provided an opportunity to test the effect of variation of OXTR expression on diverse social behaviors.

It is intriguing that Koi-4 line showed preference to their mating partners but not strangers, which is a behavior specific in prairie voles but not in mice. Multiple studies suggested that NAc and PFC play important roles in mediating partner preference behavior in prairie voles (Amadei et al., 2017; Hasse Walum & Larry J. Young, 2018). Consistently, the Koi-4 line showed robust transgene expression in these brain regions. Notably Koi-4 line also demonstrated strong Oxtr expression in the entire thalamus region, which reminds us the distribution of Avpr1a (Phelps & Young, 2003). Given the fact that OXTR and AVPR1A genes were derived from a duplication of a single ancestral gene (Grinevich, Knobloch-Bollmann, Eliava, Busnelli, & Chini, 2016), it is not surprising that the similar expression regulation machinery exists. Recent studies indeed revealed the critical functions of different thalamic nuclei in social memory and cognition (Chen et al., 2024; Kappel et al., 2022; Rangel, Baldo, & Canteras, 2018; Wolff & Vann, 2019). Partner preference formation and maintenance needs to recruit multi- and cross-modality sensation, social recognition and social memory, selective attention, arousal, and reward systems. We propose a model that the sufficient upregulation of the OXTR expression in one or multiple of these systems can tune the network and boost partner preference behavior. In addition, the maternal behavior in Koi-3 and Koi-4 lines was enhanced and reduced, respectively. This result suggested that OXTR signaling facilitates social partner preference and maternal care by coordinating activity across different neural networks. A main conclusion of our study is that simply by redistributing OXTR binding in space, variation in social preference and parental behaviors emerged, which conceivably could impart advantageous behaviors that natural selection could act upon and amplify. Future study will determine how the variation in OXTR distribution in our Koi lines affects neural circuits mediating social behaviors.

A potential genetic mechanism contributing to diversity of brain OXTR expression

The sex steroid and oxytocin systems play important roles in modulating reproductive and related social behaviors in vertebrates. Sexual behavior, e.g., the motivation to mate and motor patterns, is highly conserved across vertebrate species. By contrast, social behaviors, e.g., sociality, mating strategy (monogamy vs polygamy), parental and alloparental behaviors are more nuanced and can vary dramatically across species in the same genus or even between individuals of a single species. It is interesting, therefore, that the distribution of steroid receptors is quite conserved among species (L. J. Young & Crews, 1995), while that of the OXTR is extremely diverse across and even within species (Froemke & Young, 2021; Larry J. Young & Zhang, 2021). It is even more interesting that although the distribution of OXTR is diverse, its ligand, oxytocin, is expressed in a highly conserved neuroanatomical pattern in vertebrates (Ross & Young, 2009). Previous work showed that three independent rat lines with puffer fish oxytocin BAC transgene (containing 16 kb of 5’ flanking regulatory sequence) expressed the Fugu oxytocin transcript specifically in oxytocin neurons in rat (Venkatesh, Si-Hoe, Murphy, & Brenner, 1997). Moreover, a similar transgenic experiment transferring approximately 5 kb of the Fugu oxytocin already resulted in a faithful expression of the fish genes in mouse oxytocin neurons (Gilligan, Brenner, & Venkatesh, 2003). These studies suggest there is a remarkably conserved transcriptional regulatory machinery across species, even between the Fugu and rodent oxytocin genes, despite being separated by 400 million years of evolution.

The GENSAT (Gene Expression Nervous System Atlas) Project demonstrated that most BAC transgenes reliably recapitulate endogenous expression across independent lines, provided that extensive flanking intergenic regions are included (Gerfen et al., 2013; S. Gong et al., 2007; Shiaoching Gong et al., 2003; Heintz, 2004; Schmidt et al., 2013). Our pvOxtr clone contained the entire locus with full upstream and downstream intergenic sequences, yet the transgenic lines we generated displayed strikingly divergent brain expression patterns. These differences were not explained by transgene copy number, but instead reflected strong insertion-site effects. This paradox contrasts with the general reproducibility of BAC transgenics and echoes the observation that virtually every rodent or primate species has a unique distribution of OXTR binding in the brain.

The 3D genomic structure analysis showed that the Oxtr loci sit within large TAD structures, increasing its exposure to distal regulatory elements through an expanded interaction landscape, whereas Oxt lacks organized TAD structures. This difference might explain why short promoter sequences suffice for conserved Oxt expression across species, while even a large BAC is insufficient for precise, reproducible transcription regulation of Oxtr. The large TAD and extensive long-range interactions render Oxtr especially sensitive to chromosomal context, accounting for the position effects we observed.

It is also noteworthy that pvOxtr expression in mammary gland (which is responsible for essential lactation in mammals) was conserved across all Koi lines, indicating that the essential cis-regulatory elements for peripheral expression are captured within the BAC and are resistant to position effects. This conclusion is consistent with the chromatin architecture we observed: the chromatin landscape differs considerably between brain-related tissues and peripheral tissues. There exists an obvious increase of long-distance interactions between neighboring TADs, forming a larger TAD structure in brain-related tissues but not peripheral tissues. Moreover, critical peripheral regulatory sequences may be clustered proximally around the locus, so that they fall entirely within the BAC. By comparison, in brain tissues, regulatory elements are dispersed across a much broader genomic interval, extending beyond the BAC. Together, the larger interaction landscape and distributed regulatory architecture in brain tissues likely explain why BAC constructs are sufficient for peripheral but not brain expression of Oxtr.

This is well aligned with broader principles of genome organization: subtle changes within 3D chromatin structure may result in substantial changes in local gene expression (Xiao, Hafner, & Boettiger, 2021). TADs also display increased SNPs and structure variations (SVs) density and higher recombination rate compared to inter-TAD regions(Beagan & Phillips-Cremins, 2020; Dekker et al., 2002; Dixon et al., 2012; Rajderkar et al., 2023). In fact, SNPs in the Oxtr intron explains 74% of the variance in striatal Oxtr expression and social attachment in prairie voles (King et al., 2016) and it is possible that these SNPs near Oxtr are linked to variations of long-range chromosomal interaction. Collectively, these evidences support the notion that brain Oxtr expression is hyper-sensitive to genetic variation.

Such an organization may endow Oxtr with enhanced evolvability. Oxytocin is a secreted neuromodulator that can diffuse along axons and through the extracellular space(Leng & Ludwig, 2008; Mitre et al., 2016; Parmaksiz & Kim, 2025). Its functional effects depend critically on the precise localization of its receptors, as the spatial distribution of oxytocin receptors determines the neural circuits engaged and the behavioral outcomes elicited. By making brain expression more reliant on distal regulatory sequences, Oxtr becomes subject to a wider range of cis- and trans-acting influences, rendering its expression in the brain inherently more variable across individuals and species. This variability allows oxytocin signaling to recruit novel neural circuits and modulate diverse aspects of social behavior, thereby contributing to the evolutionary diversification of sociality. Future studies should aim to resolve the precise mechanisms underlying this variability—for example, by identifying which distal regulatory elements physically interact with Oxtr within brain-specific chromatin architectures.

Ideas and Speculation

In summary, our results revealed that region-specific expression of Oxtr in the brain requires long-distance interactions between proximal and distal cis-regulatory elements. In contrast, many other genes remained consistent across BAC-transgenic lines, indicating that proximal regulatory sequences within the BAC were sufficient to drive faithful expression in these tissues, and that this expression was resistant to positional effects.

Our findings support a broader hypothesis: genes vary in their regulatory robustness and evolutionary flexibility. We propose a framework dividing genes into two broad classes based on their transcriptional architecture—rigid and flexible genes. Rigid genes— perhaps comprising the majority of the genome—maintain core life-sustaining functions (e.g., metabolism, motor function, development, reproduction) and are typically regulated by compact proximal elements. Their expression is stable across species, and across individuals within a same species. For such genes, BAC constructs often recapitulate endogenous expression reliably—examples include CamKII, Protein Kinase C Delta(Pkcd), and other genes characterized in large-scale projects like GENSAT.

In contrast, a subset of genes are flexible genes—such as neuromodulatory receptors like Oxtr and potentially Avpr1a and Drd4 (Dopamine receptor D4)— possess inherently more fragile or flexible transcriptional regulation. Their expression is highly sensitive to multiple factors including the chromatin landscape and complex long-range regulatory interactions. We hypothesize that this regulatory sensitivity is not a flaw but an evolutionary feature—what we term transcriptional evolvability. Subtle genetic or epigenetic changes, such as SNPs, structural variants, or chromosomal rearrangements, may lead to variation in neuromodulator receptor expression, thereby altering the modulation of preexisting neural circuits. Such changes could unmask cryptic behavioral phenotypes or give rise to novel social traits without requiring changes to core neural architecture. For example, caregiving circuits might evolve into pair bonding, empathy, or romantic attachment, while defensive or aggressive circuits might become recontextualized into behaviors like jealousy or mate guarding.

This model of transcriptional evolvability helps explain why Oxtr and Avpr1a expression varies so markedly across and within species, while their ligands (Oxt and Avp) remain stably expressed. The functional consequence of receptor variability is amplified by the secreted and diffusive nature of these neuromodulators: even small shifts in receptor localization can profoundly alter which neural circuits are engaged and which behaviors are modulated. It also raises the possibility that the capacity for behavioral innovation is, at least in part, encoded in the regulatory fragility of a small set of genes. Our model carries profound implications. It suggests that the evolution of complex social behavior is driven not only by neural circuitry or protein function, but by the regulatory architecture of a small set of flexible genes. These genes serve as behavioral tuning knobs—sensitive to the chromosomal environment, and capable of generating diverse phenotypes in response to evolutionary or environmental pressures. This model also offers insight into variability in human social phenotypes, including those observed in psychiatric conditions such as autism or social anxiety disorders, where subtle changes in receptor gene regulation may lead to major phenotypic shifts. Further research into the regulatory logic of Oxtr, Avpr1a, and similar genes may reveal a generalizable mechanism by which genome structure enables behavioral diversity and adaptation across the animal kingdom.

Materials and Methods

pvOxtr-P2A-Cre BAC vector construction

The vole BAC clone (GeneBank: DP001214.1) containing the Oxtr locus was obtained from a prairie vole BAC library (CHORI-232) (Lisa A. McGraw et al., 2010; L. A. McGraw, Davis, Thomas, Young, & Thomas, 2012). The NLS (nuclear localization signal)-Cre-Frt-Amp-Frt cassette was obtained from Dr. Takuji Iwasato, and was slightly modified to introduce a P2A sequence (Kim et al., 2011). The P2A sequence was introduced using PCR with the following primers:

fw5’-ATGGGAAGCGGAGCTACTAACTTCAGCCTGCTGAAGCAGGCTGGAGA

CGTGGAGGAGAACCCTGGACCTCTCGAAACTGACAGGAGAACCACC-3’;

rv5’-AGACTGGAGTCCGCATAGCCCCCCTCCCCCGCCCCAGGCGCGGTGGGCC

AGGCAGGTGGCTCACCTTGACCAAGTTGCTGAAGTTCCTATTCC-3’. The P2A sequence is underlined. And the resulting amplified cassette had the P2A upstream of the NLS-Cre. Then the P2A-NLS-Cre-Frt-Amp-Frt cassette was inserted in-frame right before the stop codon of Oxtr in the BAC clone using the Red/ET Recombineering kit (Gene Bridges GmbH, Heidelberg Germany) after adding homologous arms by PCR with the following primers: fw5’-CACCTTCGTCCTGAGTCGCCGCAGCTCCAGCCAGAGGAGCTG

CTCTCAACCATCTTCAGCAATGGGAAGCGGAGCTACTAAC-3’;

rv5’-GAGGAGAGGGATACACACCAATAGGCACCTTATACAACTCCACGCAC

GGCCACCAGGGGCAGACTGGAGTCCGCATAGCC-3’. The ampicillin selection marker was deleted with the 706-Flpe-induced recombination method (Gene Bridges Heidelberg Germany). A correctly modified BAC clone was verified by using PCR at both the 5’ and 3’ junctions of the targeted insertion with the following primer pair: fr5’-GCCTTCATCATCGCCATGCTCTT-3’(p2) and rv5’-GATGGCTGAGTG ACTGGCATCT-3’(p8);

The construct was further verified by sequencing using the following 7 primers specific either for BAC vector or the p2A-NLS-Cre fragment:

5’-CCTGCAGCCAACTGGAGCTTC-3’(p1); 5’-CTTCCTTGGGCGCATTGACGTC-3’(p5);

5’-TACCTGTTTTGCCGGGTCAG-3(p4); 5’-GCCTTCATCATCGCCATGCTCTT-3’(p2);

5’-CTGACCCGGCAAAACAGGTA-3’(p7); 5’-TCCGGTTATTCAACTTGCACCATGC-3’(p6);

5’-GATGGCTGAGTGACTGGCATCT-3’(p8).

Animals

pvOxtr-P2A-Cre BAC transgenic mouse lines (Koi lines)

pvOxtr-P2A-Cre BAC vector was digested with NotI for linearization and to remove the pTARBAC vector backbone. The Vole-Oxtr-P2A-Cre BAC linearized fragment was purified using CL-4B sepharose (Sigma-Aldrich), and injected into pronuclei of C57BL/6J zygotes. Mice carrying the BAC transgene were identified and confirmed using PCR with 5 pairs of primers: fr5’-TCGACCAGGTTCGTTCACTC-3’(p3) + p7; p2+ p6; p2 +p7; p4 +p8; p2+p8. 8 lines (Koi-1∼Koi-8) were confirmed with successful transgene inheritance. These lines were maintained on a congenic C57BL/6J background by backcrossing with wildtype C57BL/6J for assessments of BAC DNA copy numbers and initial transgene expression analysis. The ROSA-26-NLS-LacZ mouse line (Zhang et al., 2016) was obtained from Itohara Lab at RIKEN.

The mOxtr-Ires-Cre knock-in (Hidema et al., 2016) and mOxtr^-/- mutant (Takayanagi et al., 2005) lines were obtained from Dr. Katsuhiko Nishimori at Tohoku University, and both were backcrossed with C57BL/6J for more than 10 generations. Each Koi line and mOxtr-Ires-Cre knock-in mice were crossed with ROSA-26-NLS-LacZ mice to generate Cre⁺, LacZ⁺ double positive mice for further gene expression analysis. Female Koi mice were crossed with male mOxtr^-/- to produce offspring with the genotype of pvOxtr-P2A-Cre(Koi): mOxtr^+/-. These offspring were then bred to produce Koi: mOxtr^-/- mice expressing pvOxtr while lacking endogenous mOxtr, and these mice were used for autoradiography, qRT-PCR and behavioral tests. All mice were generated using continuously housed breeder pairs and P21 as the standard weaning date. All the animal procedures were approved by University of Tsukuba Animal Care and Use Committee. Mice were housed under constant temperature and light condition (12 h light and 12 h dark cycle), and received food and water ad libitum.

Determination of transgene copy number

Custom Taqman® MGB probes were synthesized by oligoJp (Thermo Fisher, Japan) for detecting the transgenic Cre and the mouse Jun gene (internal control). The following primer pairs and probes were used: for the Cre assay, fr5’-ATGACTGGGCACAACAGACAAT-3’; rv5’-CGCTGACAGCCGGAACAC-3’; probe: 5’-FAM-AAACATGCTTCATCGTCGGTC CGG-MGB-3’; for Jun assay, fr5’-GAGTGCTAGCGGAGTCTTAACC-3’; rv5’-CTCCAGACGGCAGTGCTT-3’; probe:

5’-VIC-CTGAGCCCTCCTCCCC-MGB-3’. Real-time PCR was performed using an Applied Biosystem7500. Transgene copy number was determined through absolute quantification (standard curve method) following the protocol described in (Chandler et al., 2007). Briefly, 2 microliters (20 ng) of genomic DNA samples or copy number standards were analyzed in a 20μL reaction volume with two primer-probe sets (Cre, Jun). The data from transgenic samples were then compared to a standard curve of calibrator samples that are generated by diluting purified BAC DNA (linearized) over a range of known concentrations into wild-type mouse genomic DNA. In addition, no-template controls were included in each experiment. All reactions were performed in triplicate.

Histology

Adult mice (3-4month old), 4 male and 4 female from each line/generation were analyzed for 3 generations. Mice were perfused intracardially with 4% formalin in 0.1M sodium phosphate buffer. Brains were embedded in 2% agarose in 0.1M PB and cut into 100-m-thick sections with a Micro-slicer (Dosaka, Kyoto, Japan). Mammary glands were isolated from virgin female mice (2month old), 2 mice/each line were analyzed. The brain slices and whole mount mammary glands preparation were stained in X-gal solution (5 mM K3FeCN6, 5mMK4FeCN6,2mMMgCl2, 0.02% NP-40, 0.01% Na-deoxycholate, 1 mg/ml X-gal in 0.1M PB) at 37°C for 6 hours and then were stained with hematoxylin.

All experiments were done with positive and negative controls (Cre⁺, LacZ^-) to monitor the reliability of the X-gal staining. Images were caught by Nanozoomer 2.0-HT slide scanner (Hamamatsu Photonics).

Receptor Autoradiography

OXTR autoradiography was performed as previously described (Inoue et al., 2022). Briefly, freshly frozen brains were stored at -80°C. Coronal sections were cut in a cryostat and 20 µm sections were collected and then stored at -80°C until use in autoradiography. Brain sections were removed from -80°C storage and air dried, and then fixed for two minutes with 0.1% paraformaldehyde in PBS at room temperature, and rinsed twice in 50 mM Tris buffer, pH 7.4, to remove endogenous OT. They were then incubated in 50 pM 125I-OVTA (2200 Ci/mmol; PerkinElmer; Boston, MA), a selective, radiolabeled OXTR ligand, for one hour. Unbound 125I-OVTA was then washed away with Tris-MgCl₂ buffer (50 mM Tris plus 2% MgCl2, pH 7.4) and sections were air dried. Sections were exposed to BioMax MR film (Kodak; Rochester, New York) for five days. Digital images were obtained with a light box and a Cannon camera (Cannon 6D MarkII, Japan). The brightness and contrast of representative images were equally adjusted for all autoradiography images within a panel using Adobe Photoshop.

Fluorescent RNAscope in situ hybridization for Oxtr mRNA localization

RNAscope Multiplex Fluorescent Reagent Kit v2 (Advanced Cell Diagnostics, Newark, California; Cat. No. 323270) was used according to the manufacturer’s instructions.

Adjacent sections to those utilized for I-125 OVTA receptor binding autoradiography were used. Briefly, sections were thawed and fixed in 10% neutral buffered formalin for 15 minutes followed by treatment with hydrogen peroxide for 10 minutes and RNAscope Protease IV for 30 minutes. Probes specific to prairie vole Oxtr mRNA (Cat. No. 500721) were then applied and allowed to hybridize for two hours at 40 °C. After three amplification steps, hybridized probes were labelled with horseradish peroxidase (HRP), and TSA Vivid Dye 570 fluorophore. Sections were counterstained with DAPI for nuclei labeling and coverslipped with ProLong Gold Antifade Mountant (ThermoFisher Scientific; Waltham, Massachusetts). Negative controls were performed using a bacterial dapB probe (Cat. No. 310043), which did not exhibit any nonspecific labeling.

Fluorescent RNAscope images were acquired using a Keyence BZ-X710 microscope. Whole sections were captured with 10x objective lens in two channels (red and blue), and images were stitched together using Keyence BZX Analyzer Software. Brightness and contrast of the images were adjusted using Photoshop (Adobe; San Jose, California).

Quantification of Oxtr mRNA expression by qRT-PCR

Real-time quantitative polymerase chain reaction (RT-qPCR) using an Applied Biosystems 7500 Real-Time PCR system was used to quantify Oxtr mRNA expression levels in different brain regions of Koi lines. Total RNA of different brain regions was isolated using the RNeasy extraction kit (Qiagen, Germantown, US), and cDNA was synthesized after Deoxyribonuclease I (Invitrogen, Waltham, US) treatment by using EvoScript Universal cDNA Master (Roche, Penzberg, Germany), and then quantitative RT-PCR was performed with FastStart Universal SYBR Green Master (Roche, Penzberg, Germany). The relative standard curve method was used to obtain the relative quantities of Oxtr expression following the manual of Applied Biosystems. To allow the comparation between Koi lines and WT mice, the following primer pair (which can detect the expression of both mOxtr and pvOxtr) was used: fr5’-GCCTTTCTTCTTCGTGCAGATG-3’ and rv5′-ATGTAGATCCAGGGGTTGCAG-3′ In addition, to specifically detect the expression of pvOxtr, the primer pair fr5′-GCCTTTCTTCTTCGTGCAGATG-3 and rv5’-AAAGAGGTGGCCCGTGAAC-3′ was used in the 2^nd qRT-PCR experiment. GAPDH was used as the endogenous control for both experiments and was amplified by the following primer pair: fr5’-GGGTTCCTATAAATACGGACTGC-3’ and rv5’-CCATTTTGTCTACGGGACGA-3’.

Samples were analyzed in triplicates. A non-template control was performed to ensure that there was no amplification of genomic DNA. All of the experimenters were blind to the genotype of the subjects.

Partner preference test (PPT)

Subjects were housed with 4 age-matched and same-sex littermates until testing at adulthood (2-5 months old). 12-14 homozygous? mice were tested for each group. Ovariectomized females were paired and cohoused with a sexually experienced adult wildtype male for 21 days before PPT test. The female was injected with estradiol benzoate (10 μg and 5 μg at 48 h and 24 h before induced mating) and progesterone (500 μg at 4–7 h before induced mating) to ensure high sexual receptivity before mating.

Mating was induced on Day 2,9 and16. PPT data was analyzed using TimeSSI1 for social interaction test system (O’ Hara & Co., Ltd.). Briefly, the experimental subject was placed in a chamber, in which two pen-boxes (diameter 8cm, height10cm) were located at two diagonally opposite corners. The experimental animal was free to move throughout the chamber and the time spent in close proximity to each pen-box (stay time) is recorded using an automated mouse tracking system (Fig.4 A, B). In the habituation phase, the experimental subject was allowed to explore the chamber for 5 minutes with both pen-boxes empty. In the recognition phase, the partner stimulus animal was restricted in one pen-box, and a novel “stranger” stimulus animal was restricted in another pen-box, and the experimental subject was allowed to explore the chamber for 10 minutes. In the preference phase, the positions of the partner stimulus animal and the “stranger” stimulus animal were switched, and the experimental subject was allowed to explore the chamber for 20 minutes. To assure unbiased design, pen-box assignments were counterbalanced for the diagonal positions. All of the experimenters were blind to the genotype of the subjects.

Parental behavior test

To examine maternal behavior, we performed pup retrieval test with virgin female mice (2- to 4-months old, 15-18mice/line), following the method described in (Stolzenberg, Stevens, & Rissman, 2012) with some modifications. Briefly, 5 days prior to the behavioral testing, each subject was isolated and housed in separate home cages in the test room. Nesting material (about 1.5 g of cotton wool) was provided for nest making. The tests were conducted in the dark phase of light/dark cycle (12 hours/12 hours). On the day of the test, each subject was placed in the recording area and allowed 15 minutes for habituation. The stimulus pups (age: P2-P3) were collected from a group of donor mothers immediately before the start of the experiment. A retrieval test began with the placement of 3 stimulus pups on the side farthest form the nest, and the female’s behavior was recorded by an ARNAN 4 channel security system for 15 minutes. Pup retrieval was defined as picking up the pup and bringing it to the nest. If the subject did not complete the retrieval for 3 pups in 15 minutes, the video recording time was extended to 30 minutes. The latency to retrieve each pup (1st, 2nd, 3rd) and the time spent crouching on the pups were time-stamped and calculated manually. A latency of 1800 seconds was assigned if the pup-retrieval was not completed in 30 minutes. Crouching was defined as the mouse supporting itself in a lactation-position over the pups in the nest. Crouching time on a single pup, two pups, and three pups were summed up as total crouching time. All of the experimenters were blind to the genotype of the subjects.

3D chromatin verification on Hi-C datasets of humans and mice

The 3D genome structure tracks were obtained from 4DNucleome Consortium (https://www.4dnucleome.org/) where datasets were systematically reanalyzed following recommended standard protocols. The datasets IDs and sample information are provided in the supplementary material (Supplementary Table 1,2). 3D Genome Browser(Wang et al., 2018) and WashU Epigenome Browser(Li et al., 2022) were employed for visualizations.

Experimental design and statistical analysis

The sample sizes were determined by G*Power 3 program (Faul, Erdfelder, Lang, & Buchner, 2007) based on pilot experiments. They are also within the range that is generally accepted in the field. All experiments were performed and analyzed blind to the genotypes and no data point was excluded. All reported sample numbers (n) represent independent biological replicates that are the numbers of tested mice or tissue. All the statistical analyses were performed by SPSS21(IBM). Mixed ANOVA was used for analyzing PPT data and pup retrieval data. Mauchly’s test of sphericity was used to test whether or not the assumption of sphericity was met in repeated measures. Greenhouse-Geisser correction was applied when the sphericity was violated. Two-way ANOVA was used for analyzing qRT-PCR region specific data across different mouse lines. One-way ANOVA was used for analyzing crouching time in the parental behavioral test. If there was a significant main effect of an independent factor, post hoc test was used to do multiple comparisons. If there was a significant interaction between within-subject factor and between-subject factor, post hoc pairwise comparisons were applied to analyze the simple main effects. Bonferroni correction was used for all the post hoc tests.

Supplementary Information

Supplementary Fig. 1

Supplementary Fig. 2

Supplementary Fig. 3

Supplementary Table 1

Data availability

Original data created for the study are available in a persistent repository as follows: Repository Name and DOI/accession number(s): zhang, qi, "Dataset - Hi-c analysis for Oxtr and Oxt", Mendeley Data, V1, doi: 10.17632/jdn6z5m6zs.1 zhang, qi, "Dataset for the analysis of koi mice", Mendeley Data, V1, doi: 10.17632/fjztcwtj2m.1

Acknowledgements

This work was supported by the RIKEN Incentive Research Project (100226201701100443) to Q.Z., the Brain Science Project, Center for novel science initiatives, National institutes of natural sciences (BS291003) to Q.Z., the RIKEN Aging Project (10026-201701100263-340120) to Q.Z., the JSPS Kakenhi Grant-in-Aid for Young Scientists (17K18362) to Q.Z., the JSPS Kakenhi Grant-in-Aid for Challenging Research (19K21807) to Q.Z., the JSPS Kakenhi Grant-in-Aid for Challenging Research (22K19478)to Q.Z., the JSPS Kakenhi Grant-in-Aid for Scientific Research (25K10198) to Q.Z., the Takeda Medical Research Grant to Q.Z., the International Education and Research Laboratory Program of University of Tsukuba to L.Y. and Q.Z.. We thank Dr. Katsuhiko Nishimori (Tohoku University and Fukushima Medical University, Japan) for providing mOxtr-Ires-Cre knock-in line and mOxtr-/- mutant line; Dr. Cary Lai (Indiana University, USA), Dr. Takuji Iwasato(National Institute of Genetics, Japan) and Dr. Seiya Mizuno(University of Tsukuba, Japan) for their advice on BAC vector construction; Dr. Volkhard Maeckel (CERN, Switzerland) for helping with image capturing; Dr. Yong Zhang (Tongji University, China) for the stimulating discussion about 3D chromatin structure, and we thank the Laboratory of Animal Resource Center of University of Tsukuba for pronuclear injection.

Additional information

Author Contributions

Q.Z. and L.Y. conceived and designed the study. Q.Z., M.T., L.N., K.I., L.K., M.P., Y.N., M.S., and T.S. performed experiments. Q.Z., M.T., L.N., K.I., L.K., M.P., and K.N. analyzed data. S.I. provided guidance and facilities for transgene vector construction. Q.Z. and L.Y. supervised the project and wrote the manuscript. All authors contributed to discussion and approved the final version.

Funding

MEXT | RIKEN (100226201701100443)

• Qi Zhang

MEXT | RIKEN (10026-201701100263-340120)

• Qi Zhang

MEXT | Japan Society for the Promotion of Science (JSPS) (25K10198)

• Qi Zhang

MEXT | Japan Society for the Promotion of Science (JSPS) (22K19478)

• Qi Zhang

MEXT | Japan Society for the Promotion of Science (JSPS) (19K21807)

• Qi Zhang

MEXT | Japan Society for the Promotion of Science (JSPS) (17K18362)

• Qi Zhang

Takeda Medical Research Foundation (Medical Research Grant)

• Qi Zhang

University of Tsukuba (International Program for Research and Education)

• Larry J Young