We dissected mouse VNOs from postnatal day 14 (P14) juveniles and P56 adults. Cells were dissociated in the presence of actinomycin D to prevent procedure-induced transcription. From four adult (P56) and four juvenile (P14) mice (equal representation of sexes) we obtained sequence reads from 34,519 cells. The samples and replicates were integrated for cell clustering. In two-dimensional UMAP space, 18 cell clusters can be clearly identified (Figure 1A). These clusters were curated using known cell markers (Figure 1B). There was no obvious difference in the presence of cell clusters between juvenile and adult VNOs (Figure 1C) or between male and female sexes (Figure 1D). Although there were differences in gene expression profiles between the ages (Figure 1—figure supplement 1) and sexes (Figure 1—figure supplement 2), the list of significantly differentially expressed genes did not appear to be influential for the neuronal lineage and cell type specification, or related to cell adhesion molecules, which were the main focuses of this study.

Clustering confirmed all previously identified cell types but also revealed some surprises. The largest portion of cells belonged to the neuronal lineage, including the globose basal cells (GBC), immediate neuronal progenitors (INPs), the immature and mature VSNs, and a population of cells that do not co-cluster with either VSN type (see below). There were substantial numbers of sustentacular cells (SCs), horizontal basal cells (HBCs), and ms4-expressing microvillus cells (MVs). Cells engaged in adaptive immune responses, including microglia and T-cells, were also detected. A population of Fpr-1 expressing cells that were distinctive from the VSNs expressing the Fpr family of genes formed a separate class. These were likely resident cells mediating innate immune responses. We also identified the olfactory ensheathing cells (OECs) and a population of lamina propria (LP) cells, which share molecular characteristics with what we have found in the MOE (Wu et al., 2022).

To obtain the spatial location of the various cell types, we selected 100 target genes based on the scRNA-seq results. Using probes for these genes, we used the Molecular Cartography platform to perform spatial molecular imaging (Figure 1E). Based on molecular clouds and DAPI nuclei staining, we segmented the cells and quantified gene expression profiles to cluster the cells. We then map individual cell clusters onto their spatial locations in VNO slices. Unlike previous studies that relied on a few markers to identify cell types, our approach relied on the spatial transcriptome to calculate the probability that a cell belongs to a specific class. This analysis revealed that the VSNs and supporting cells are located in the pseudostratified neuroepithelium (Figure 1F). The LP cells are located along the LP underlying the neuroepithelium as found in the MOE (Figure 1F and G). Surprisingly, however, there are few HBCs located along the basal lamina, in direct contrast to the MOE (Figure 1F and G). Most of the HBCs are found in the non-neuronal epithelium surrounding the blood vessel, with few near the marginal zone. The marginal zone is thought to be the neurogenic region (Takami and Graziadei, 1991; Mombaerts et al., 1996; Gronowitz et al., 2021). We quantified the distribution of cell types in various regions of the VNO neuroepithelia (Figure 1—figure supplement 3A) and found significantly more GBCs, INPs, and immature VSNs in the marginal zone than in the main zone (Figure 1—figure supplement 3B and Figure 1—figure supplement 3C). A previous study suggested neurogenic activity in the medial zone (Brann and Firestein, 2010), but we did not find evidence in support of that conclusion, which was largely based on BrdU staining of mitotic cells without lineage-specific information. Based on our transcriptomic analysis, we conclude that neurogenic activity is restricted to the marginal zone.

To better understand the developmental trajectory of the VSNs, we segregated the cells in the neuronal lineage from the whole dataset (Figure 2A). The neuronal lineage consists of the GBCs, INPs, immature neurons as determined by the expression of Gap43 and Stmn2 genes (Figure 2—figure supplement 1A and B), and the mature VSNs. Cells expressing Xist, which is expressed in female cells, were intermingled with those from males (Figure 2—figure supplement 1C). This observation is consistent with our previous studies using bulk sequencing indicating that the cell types are not sexually dimorphic (Duyck et al., 2017). It is also consistent with physiological responses of the VSNs to various stimuli (He et al., 2008; He et al., 2010; Holy et al., 2000; Tolokh et al., 2013; Arnson et al., 2010). For further analyses, therefore, we did not segregate the samples according to sex.

The VSNs are clearly separated into the V1R and V2R clusters as distinguished by the expression of Gnai2 and Gnao1, respectively (Figure 2B). Surprisingly, we observed that a portion of the V2R cells also expressed Gnai2. These cells were primarily from adult, but not juvenile male mice (Figure 2—figure supplement 1D). Past studies based on Gnai2 and Gnao1 expression would have identified this group of cells as V1R VSNs, but the transcriptome-based classification places them as V2R VSNs. The signaling mechanism of this group of cells may be different from the canonical V2R VSNs.

We found a major group of cells that expressed the prototypical VSN markers but formed a cluster distinct from the mature V1R and V2R cells (Figures 1A and 2A). Within the cluster, there was an apparent segregation among the cells into V1R and V2R groups based on marker gene expression (Figure 2B). A small group of OR-expressing neurons also belonged to this cluster. These cells expressed fewer overall genes and with lower total counts when compared with the mature V1R and V2R cells (Figure 2—figure supplement 1E and F). The lower ribosomal gene expression suggests that these neurons are less active in protein translation (Figure 2—figure supplement 1G). This group of cells are scattered within the epithelium (Figure 2E).

There are 503 differentially expressed genes in this cluster when compared with other mature neurons (padj <0.001; FC >1.5; Figure 2F, Figure 2—figure supplement 1H). Gene ontology (GO) analysis reveals that differentially expressed genes enriched in this group are associated with odorant binding proteins (Figure 2G). Correspondingly, mucin 2 (Muc2), odorant binding protein 2 a (Obp2a), Obp2b, and lipocalin 3 (Lcn3) genes, usually enriched in non-neuronal supporting or secretory gland cells, are expressed at higher levels in these neurons than the canonical VSNs (Figure 2H and Figure 2—figure supplement 2A–D). They also have higher expression of Wnk1, which is involved in ion transport, the lncRNA Neat1, the purinergic receptor P2ry14, the zinc finger protein Zfp738, and the centrosome and spindle pole associated Cspp1 (Figure 2—figure supplement 2E–I). On the other hand, these cells exhibit lower expression of Omp and JunD (Figure 2F and Figure 2—figure supplement 2M–N). The lower level of Omp suggests that these cells do not have the full characteristics of mature VSNs (mVSNs), but they are also distinct from the immature VSNs (iVSNs) as they do not express higher levels of immature markers such as Gap43 and Stmn2 (Figure 2—figure supplement 1A and B). The lower expression of Ndua2, Ndua7, Cox6a1, and Cox7a2, which are involved in mitochondrial activities, suggest that these cells are less metabolically active than the canonical VSNs (Figure 2F and Figure 2—figure supplement 2O–R). To rule out the possibility that these cells were immature or senescent, we performed a pseudotime analysis on the neuronal lineage and found the cluster to have similar pseudotime values as the mature V1R and V2R linages (Figure 2—figure supplement 3A and B). Taken together, the gene expression profile suggests that this new class of neurons may not only sense environmental stimulus, but also may provide proteins to facilitate the clearing of the chemicals upon stimulation. We, therefore, name these cells as putative secretory VSNs (sVSNs).

We also identify a distinct set of cells expressing the odorant receptors (ORs) that comprise ~2.3% of the total neurons (Figure 2A). A previous study has found the expression of ORs in neurons that display some typical molecular features of VSNs, while another study detecting ORs in the VNO suggested the presence of non-canonical OSNs Lévai et al., 2006; Nakahara et al., 2016. The OR expressing cells were shown to project to the AOB, but it was not clear how prevalent OR expression was in the VNO, nor whether the cells were VSNs or OSNs. While some OR expressing cells cluster with the mature V1R or V2R neuronal lineages, a majority of these cells lacks V1R or V2R markers and forms a cluster distinct from the V1R and V2R VSNs (Figure 2C). These cells express Gnal and Cnga2, which are the canonical markers of OSNs in the MOE.

Differential gene expression analysis (Figure 2—figure supplement 3C) reveals an enrichment for multiple GO terms related to cilium (Figure 2—figure supplement 3D), consistent with the ciliated nature of OSNs. We thus mark the cells as canonical OSNs. Spatial mapping reveals that the OSNs are mainly in the neuroepithelium, with some cells concentrated in the marginal zone (Figure 2D).

The VNO develops from the olfactory pit during the embryonic period and continues to develop into postnatal stages (Garrosa et al., 1998; Suárez, 2011; Katreddi and Forni, 2021). Neurons regenerate in adult animals (Giacobini et al., 2000; Martínez-Marcos et al., 2000). Cell types in the vomeronasal lineage have been shown to be specified by BMP and Notch signaling (Katreddi et al., 2022; Naik et al., 2020), and coordinated by transcription factors (TFs) including Bcl11b/Ctip2, C/EBPγ, ATF5, Gli3, Meis2, and Tfap2e (Enomoto et al., 2011; Lin et al., 2018; Nakano et al., 2019; Taroc et al., 2020; Rawson et al., 2010; Chang and Parrilla, 2016). Recent scRNA-seq studies of the VNO have helped identify Notch signaling as a specifier of the apical and basal lineages and have provided insight into the distinct transcriptional profiles of the basal and apical program (Katreddi et al., 2022; Lin et al., 2022). Despite these advances in understanding the role of individual genes in VNO development, the transcriptional program that specifies the lineages is not known.

To explore VSN development, we performed pseudotime inference analysis of the V1R and V2R lineages for P14 and P56 mice using Slingshot (Figure 3A; Street et al., 2018). We set GBCs as the starting cluster and mVSNs as terminal clusters. A minimum spanning tree through the centroids of each cluster was calculated using the first fifty principal components and fit with a smooth representation to assign pseudotime values along the principal curve of each lineage. Cell density plots for both the V1R and V2R lineage reveal a higher portion of immature VSNs at P14 than at P56. For the mature VSNs, the P14 samples have peaks at an earlier pseudotime than the P56 mice (Figure 3B). This result indicates the VSNs in juvenile mice are developmentally less mature than their counterparts in adults, but these differences do not distinguish them in obvious ways (Figure 1C).

On the other hand, we have identified dynamic changes in transcriptomes associated with the V1R and V2R lineages. There were 2037 significantly differentially expressed genes between V1R and V2R lineages (Figure 3C). While V1R and V2R lineages shared some of the genes in the early stages of development, most show distinct expression patterns between the two. To determine the transcriptional program associated with the lineages, we examined the expression sequence of individual transcription factors and signaling molecules from the gene list.

The transcription factor Ascl1 is expressed by a subset of the GBCs that appears to be the earliest in developmental stage, whereas Sox2 is expressed by a broader set of GBCs (Figure 3D–F). The early INPs are defined by the expression of Sp8, Neurog1, and Neurod1. Neurog1 is mostly restricted in the early INPs, whereas Neurod1 and Sp8 are also expressed by the late INPs (Figure 3G–I and Figure 3—figure supplement 1A–B). Neurod1 is expressed by iVSNs and iOSNs, but Sp8 is only expressed by the iVSNs.

To obtain a refined view of the transition from INPs to the immature sensory neurons, we took the subset and re-clustered them (Figure 3J). The late INPs are separated into two clusters that share marker genes with the V1R and V2R iVSNs, respectively, indicating that commitment to the two lineages begins at the late INP stage. Consistent with previous findings, the homeobox protein Meis2 was specific to the V1R lineage (Figure 3K). Concomitant with Meis2, there are a number of other genes expressed by the late INPs committed to the V1R fate, including secretin (Sct), Foxj1 target gene Fam183b, the microRNA Mir100hg, acyl-CoA dehydrogenase Acad10, and Keratin7 (Krt7; Figure 3L–O and Figure 3—figure supplement 1C). Notably, Bcl11b is exclusively absent from INPs committed to the V1R fate. This observation is consistent with the observation by Katreddi and colleagues that Bcl11b is lost in basal INPs in Notch knockout (Katreddi et al., 2022). Together with the observation that loss of Bcl11b results in increased number of V1R VSNs (Enomoto et al., 2011), these results indicate that transient downregulation of Bcl11b is required for the V1R lineage (Figure 3U).

The transcription factor Tfap2e, which is required to maintain the V2R VSNs (Lin et al., 2018), is expressed by the iVSNs but not by the late INPs committing to the V2R fate (Figure 3P). Sp8 is the only transcription factor specifically expressed in the late INPs committed to the V2R but not the V1R fate (Figure 3I). Emx2 is expressed by all neuronal lineage cells (Figure 3—figure supplement 1E). Krt8 is also found throughout the V2R lineage, but its expression is diminished in the V1R cells (Figure 3—figure supplement 1D).

The OSN lineage is marked by the expression of Olig2, Fezf1, Tshz2, and Nfib (Figure 3Q–T). The expression of Olig2 and Fezf1 is exclusive to the OSN fate. Tshz2 is expressed at a late stage of the iVSNs, but not in the INPs. There are multiple genes that may not directly be engaged in cell fate determination but are clearly markers of the cell types (Figure 3—figure supplement 1F–R). Although Notch1 and Dll4 are not identified as significantly differentially expressed, feature plots show clear distinction in their expression in the V2R and V1R lineage, respectively, as an earlier study showed (Katreddi et al., 2022).

Based on these patterns, we propose a model of molecular cascades that specify the neuronal lineages in the VNO (Figure 3V). Ascl1 and Sox2 specify the GBCs. The down regulation of Ascl1, Sox2, and subsequent upregulation of Neurog1, Neurod1, and Sox8 commits the cells to the early INPs. The VSN and OSN lineages diverge after the early INP stage. The downregulation of Sp8, Nfib, and Bcl11b and the expression of Meis2 promote the V1R fate.

The downregulation of Sp8 and the expression of Fezf1, Tshz2, and Olig2 are required for commitment to the OSN lineage. Downregulation of Sp8 is not required for the V2R lineage, which begin expressing Tfap2e. Neurod1 is expressed in all INPs. These patterns of expression suggest that both the OSN and V1R lineages required the expression of specific transcription factors. The V2R lineage, on the other hand, appears to rely on factors inherited from the early INPs. This suggests the possibility that the V2R lineage is a default path for the VSNs.

We quantified the expression of V1Rs, V2Rs, and ORs to gain insights into how chemosensory cues may be encoded by the VNO. For comparison, we also included OR expression from the MOE (Wu et al., 2022). Within each class of receptors, the probability of expression of a gene follows a power law distribution except for the lower ranked genes (Figure 4—figure supplement 1A). The sharp deviation from the power law curve for the lower ranked receptors is likely from technical dropout of genes expressed at low levels as they cannot be effectively captured by the scRNA-seq platforms. We plotted the relationship between total reads and the number of cells expressing a given receptor, and the average reads per cell for the receptors (Figure 4A–H). We observed weak correlations between the total reads and the cell number expressing a given V1R or a V2R (Figure 4A and B). This non-uniform distribution of VR genes is consistent with our observation from bulk sequencing results (Duyck et al., 2017).

Most V1Rs and V2Rs are expressed by less than 1500 cells (out of 34,519). Most VRs are expressed at less than 100 counts per cell (Figure 4E and F). Several V1Rs, including Vmn1r184, Vmn1r89, Vmn1r196, Vmn1r43, and Vmn1r37, are highly expressed in individual cells (Figure 4A). Vmn1r196, Vmn1r43, and Vmn1r37 are also expressed at the highest level per cell. Some others, including Vmn1r183, Vmn1r81, and Vmn1r13, are expressed in large numbers of cells but have low expression in individual cells. Notably, the two highest expressed receptors recognize female pheromone cues. Vmn1r89, also known as V1rj2, is one of the receptors that detects female estrus signals. Vmn1r184 is one of the receptors for the female identity pheromone (Isogai et al., 2011; Haga-Yamanaka et al., 2015; Haga-Yamanaka et al., 2014). The functions of Vmn1r196, Vmn1r43, and Vmn1r37, however, are unknown. We did not detect the expression of 45 V1Rs, which may be the result of technical dropout (Figure 4E).

We do not observe obvious correlations between expression level and chromosomal locations. Both Vmn1r89 and Vmn1r184 are located on Chromosome 7, in a region enriched in V1R genes. Similarly, Vmn1r37 and Vmn1r43 are in a V1R-rich region on Chr. 6. However, Vmn1r183, Vmn1r13, and Vmn1r81, which are in the same clusters, are expressed by many cells but at some of the lowest levels.

All V2R genes are detected in the VNO (Figure 4B and F). Vmn2r53, which has been shown to mediate intermale aggression through a dedicated circuit, has the highest level of expression and is expressed by the second most cells (Itakura et al., 2022). Among the highly expressed V2Rs, Vmn2r1 and Vmn2r7 are co-receptors for other V2Rs. Vmn2r59 has been shown to detect predator signals (Isogai et al., 2011). Vmn2r115, a receptor for ESP22 that is secreted by juveniles (Ferrero et al., 2013), is expressed by the highest number of cells. However, Vmn2r116 (V2rp5), which recognizes ESP1 and induces lordosis behavior in females, is a close homolog of Vmn2r115 but not among the highly expressed genes (Ferrero et al., 2013; Haga et al., 2010). Notably, Vmn2r114, close homolog of Vmn2r115 and Vmn2r116, is also expressed by large numbers of cells. Vmn2r88, a hemoglobin receptor (Osakada et al., 2022), was not identified as a highly expressed gene. The functions of other highly expressed receptors are not known.

In contrast to the VR genes, total counts for ORs in the VNO exhibit a tight relationship with the number of cells (Figure 4C and G). Except for Olfr124, most of the OR genes align well with the linear regression curve. This relationship is different from the VR genes and is also distinct from the single cell data from the MOE, which exhibits a similar distribution as the VRs (Wu et al., 2022; Figure 4D and H). Out of the 686 OR genes detected in the VNO, only 80 are expressed by more than 10 copies per cells, indicating that a majority of the ORs do not contribute to meaningful signaling of chemosensory cues.

To comprehensively survey receptor expression, we also included VR and OR pseudogenes in our analysis (Figure 4—figure supplement 1B–G). We did not detect a significant expression of pseudogene V1Rs (Figure 4—figure supplement 1E), but Vmn2r-ps87 has the highest count/cell in the V2R population (Figure 4—figure supplement 1F).Olfr709-ps1, and Olfr1372-ps1 are the two highest expressed genes in terms of total count and total number of cells in the VNO (Figure 4—figure supplement 1D).

We next examined transcription factors associated with individual receptor types. In the MOE, ORs are monoallelically expressed (Chess et al., 1994). The unique expression of an OR gene is mediated by chromosomal repression and de-repression, coordinated by transcription factors and genes involved in epigenetic modification (Monahan et al., 2019; Dalton et al., 2013; Clowney et al., 2012; Lyons et al., 2013; Monahan and Lomvardas, 2015; Lomvardas et al., 2006). Monoallelic V1R gene expression is also observed (Rodriguez et al., 1999). Epigenetic modification takes place at V1R gene clusters (Rodriguez, 2013), but the repression of receptor genes appears to permit transcriptional stability rather than receptor choice (Dietschi et al., 2022). How VR genes are selected is not known. To explore our dataset for clues of transcriptional activities associated with VR expression, we plotted the cross-correlation between VRs and their TF profiles. We observed correlations among receptors (Figure 4I and J). We did not find an obvious association between TF profiles and VR sequence similarity of pairs (Figure 4—figure supplement 2A).

There is a high correlation among a large portion of V1Rs (Figure 4I). By analyzing the correlation between individual TFs with the VRs, we found that V1R expression is overwhelmingly associated with Meis2 (Figure 4—figure supplement 2B). The few receptor types that do not show high correlation with Meis2 are associated with Egr1 or Fos. It is not clear whether these prototypical immediate early genes are involved in specifying receptor choice. This result indicates that once the V1R lineage is specified by Meis2, receptor choice is not determined by specific combinations of transcription factors. This scenario is similar to receptor choice by the OSNs in the MOE.

Different from the V1Rs, we observed islands of high similarity of TF expression among the V2Rs, indicating that receptors in these islands share the same set of TFs (Figure 4J). We identify correlation between individual TFs with V2Rs (Figure 4—figure supplement 2C). Whereas Tfap2e is involved in specifying the V2R fate, it is only associated with the expression of a subset of V2Rs. Pou2f1 and Atf5 are more strongly associated with other subsets of receptors. Unlike the V1Rs, there is a disparate set of TFs associated with the V2Rs, suggesting that the V2R choice may be determined by combinations of TFs. Based on the prevalence of individual transcription factors in association with the VRs, we propose a model of transcriptional cascade that may be involved in receptor choice (Figure 4K). For the V2Rs, Tfap2e+ cells can be further specified by Ikzf4, Tcf4, and Trps1. In Ikzf4-negative cells, Batf3, Atf5, Pbx2, and Pou2f1 can specify receptor types, respectively. In the Tfap2e- cells, receptors can be specified by Pou2f1, Rlf, and Batf3.

We next investigated co-expression of receptors in individual cells. Since we applied SoupX to limit ambient RNA contamination and Scrublet to remove doublet cells, we set a stringent criterion in counting receptors expressed by single cells (Young and Behjati, 2020). V1R and V2R genes on average constitute ~2% of total reads per cell, and the ORs constitute less than 1% of the total reads per cell (Figure 5—figure supplement 1A–C). On average, the V2Rs have significantly more than one receptor per cell (Figure 5—figure supplement 1B). Using Shannon Index to measure uniqueness of receptor expression in individual cells, we found that the mature VSN and OSNs have relatively high specificity, but many cells show significantly higher index values, indicating that they expressed multiple receptors (Figure 5A). In support of this observation, there are significant representations of the second and third highest expressed receptors in all three types of neurons (Figure 5—figure supplement 1D–G).

To further reduce contributions of spurious, random low-level expression, we only consider those receptors with at least 10 raw counts per cell and are found in at least five cells to evaluate co- expression among receptors. We split cells into groups according to cell-type, age, and neuronal lineage and plot the percentage of cells expressing zero, one, two, and three or more species (Figure 5B–C). This analysis shows that receptor expression specificity increases as the lineages progressed from progenitors into mature neurons. Immature neurons have more cells co- expressing receptors than mature cells. More co-expressions are observed in the younger animals than the older ones. This line of evidence indicates that the co-expression we have found is not from experimental artifacts but reflects real biological events. Contamination would not be selectively enriched in immature cells and not present in the INP cells.

We performed Fisher’s exact test using contingency tables for every pairing of expressed receptor genes. For the pairs that pass the test, we generated Circos plots to show the genomic loci for all significant V1R, V2R, and interclass pairs (Figure 5D–F). This analysis showed that 47.7% of co-expressed receptors are co-localized on the same chromosome.

We found a few sets of co-expressed VRs that would have strong implications for how pheromone signals are detected. Vmn1r85 (V1rj3) and Vmn1r86 are two receptors located next to each other on Chr. 7, sharing a high level of homology, and belonging to the V1rj clade. They have a high level of co-expression but are not co-expressed with Vmn1r89 (V1rj2), located ~100 Kb away, even though both Vmn1r85 and Vmn1r89 receptors are activated by sulfated estrogen and carry information about the estrus status of mature female mice (Haga-Yamanaka et al., 2015; Nodari et al., 2008; Haga-Yamanaka et al., 2014; Figure 5D). Another set of receptors that recognize female-specific pheromone cues are the V1re clade receptors. We found that Vmn1r185 (V1re12), which recognizes female identity pheromones, was co-expressed with its close homolog Vmn1r184 gene, which is about 350 kb away on Chr 7. They are not co-expressed with Vmn1r69 (V1re9), which also recognizes female pheromones, but is located 16 Mb apart on Chr. 7 (Haga-Yamanaka et al., 2015; Haga-Yamanaka et al., 2014; Fu et al., 2015; Lee et al., 2019; Figure 5D).

On Chr. 7 there are two other major clusters of V1R genes that show co-expression. One cluster includes Vmn1r55-Vmn1r64, 10 genes belonging to the V1rd clade and located within a 600 Kb region. Another one includes Vmn1r167, Vmn1r168, and Vmn1r169, which appear to be specifically paired with Vmn1r175, Vmn1r177, and Vmn1r176, respectively. These receptor pairs are arranged in a 300 Kb region with a head-head orientation. Outside of Chr. 7, several small clusters on Chr. 6 and one large cluster on Chr. 3 exhibit significant co-expression of V1Rs.

The V2R neurons coordinately express one common V2R and one specific receptor (Akiyoshi et al., 2018; Ishii and Mombaerts, 2011). Our co-expression analysis confirms the broad association of Vmn2r1-7, which are located on Chr. 3, with other receptors across various chromosomal locations (Figure 5E). In addition, we also identified co-expression patterns among the specifically expressed V2Rs. Among these receptor genes, intra-chromosomal co-expressions are observed for receptors residing on Chr. 7, Chr. 17, and Chr. 5. There is also inter-chromosomal co-expression between one locus on Chr. 17 with a cluster on Chr. 7 (Figure 5E).

Lastly, we have observed co-expression of receptors across different classes of receptors (Figure 5F). Notably, Vmn2r1-3 are co-expressed with several Vmn1r receptors on Chr. 3. Vmn2r7 is co-expressed with a group of Vmn1r genes on Chr. 13. The odorant receptor Olfr344 is co- expressed with several V1R and V2R genes.

In past studies, VR expression was examined using in situ hybridization, immunostaining, or genetic labeling. The traditional histological methods were not sensitive enough to quantitatively measure signal strength. Moreover, pairwise double in situ is too laborious to capture co- expression of two or more receptors. To verify co-expression of VR genes by individual VSNs, we selected 30 VR genes based on scRNA-seq data and used Molecular Cartography to examine their expression patterns in situ.

Given the high incidents of co-expression between receptors in the Vmn1r55-64 cluster on Chr. 7 (Figure 5D), we included five probes for this set of genes and confirmed colocalization of pairs in the VSNs (Figure 5G), Interestingly, we did not find cells expressing more than two receptor genes for this set. We also confirmed the co-expression between Vmn1r85 and Vmn1r86 as indicated by single cell data (Figure 5D and G).

We confirmed the co-expression among genes from the V2R genes (Figure 5H; Silvotti et al., 2011; Francia et al., 2015). Consistent with previous reports (Silvotti et al., 2011; Silvotti et al., 2007), Vmn2r2, Vmn2r3, Vmn2r6, and Vmn2r7 are comingled in several cells, but Vmn2r1 is not co-expressed with these four broadly expressed V2Rs (Figure 5H). Outside of the Vmn2r1-7 group, we find that Vmn2r81 is co-expressed with Vmn2r20 or Vmn2r24 but without any of the Vmn2r1-7 transcripts (Figure 5H). We detected more incidents of co-expression between Fpr3 and Fpr-rs4, and between Fpr3, Fpr-rs3, Fpr-rs4 with V1Rs than with V2Rs (Figure 5I and J). We also found co-expressions that are not predicted by the single cell analysis (Figure 5K). The discrepancy likely can be attributed to the relatively low-level expression of one of the receptor genes. This type of co-expression may not pass the strict criteria we set for the single cell analysis.

VSNs expressing a given receptor type project to the AOB to innervate glomeruli distributed in quasi-stereotypical positions (Rodriguez et al., 1999; Belluscio et al., 1999; Haga et al., 2010). The high number of glomeruli innervated by a given VSN type raises the question about mechanisms that specify the projection patterns and the connection between the VSNs and the mitral cells. For a genetically specified circuit that transmits pheromone and other information to trigger innate behavioral and endocrine responses, there must be molecules that instruct specific connections among neurons. Several studies have revealed the requirement of Kirrel2, Kirrel3, Neuropilin2 (Nrp2), Epha5, and Robo/Slit in vomeronasal axon targeting to the AOB (Brignall et al., 2018; Prince et al., 2013; Knöll et al., 2001; Walz et al., 2002; Cloutier et al., 2002; Knöll et al., 2003; Prince et al., 2009). However, how individual guidance molecules or their combinations specify connectivity of individual VSN types is completely unknown. Here, we leverage the unbiased dataset to identify surface molecules that may serve as code for circuit specification.

We identified 307 putative axon guidance (AG) molecules, including known cell surface molecules involved in transcellular interactions and some involved in modulating axon growth. Using this panel, we calculated pairwise similarity between two VR genes, and the similarity in their guidance molecule expression. Analysis of the relationship indicates the there is a general increase in guidance molecule similarity associated with VR similarity (Figure 6A). Consistent with this observation, when we plotted the similarity of surface molecule expression among cells expressing different receptors, we found islands of similarity among the receptor pairs (Figure 6B). We then conducted correlation analysis between the guidance molecules with the V1Rs (Figure 6C) and V2Rs (Figure 6D). Consistent with previous reporting, we found that Kirrel2 was associated with nearly half of the V1Rs and Kirrel3 was mainly associated with V2Rs that project to the caudal AOB (Brignall et al., 2018; Prince et al., 2013). Robo2 is associated with nearly all V2Rs. We found that Teneurin (Tenm2 and Tenm4) and protocadherin (Pcdh9, Pcdh10, and Pcdh17) genes were associated with specific receptors (Lee et al., 2008; Alkelai et al., 2016). Epha5, Pdch10, Tenm2, Nrp2 are also strongly associated with V1Rs with partial overlap with each other. Pcdh9, Tenm4, Cntn4, EphrinA3, Pchd17, as well as Kirrel2 and Kirrel3 all show association with specific V2Rs. Numerous guidance molecules that are not broadly expressed are also associated with individual VRs.

Based on these associations, and supported by existing literature, we propose a model of the specification of separate groups of VRs. In this model, Robo2 separates the rostral vs. caudal AOB. Robo2 +V2 R neurons project to the caudal AOB, whereas Robo2- V1R cells project to the rostral AOB (Knöll et al., 2003; Prince et al., 2009; Cloutier et al., 2004). Among V1Rs, Kirrel2 expression distinguishes between two groups of cells (Brignall et al., 2018; Wang et al., 2021). The Kirrel2 +population can be further separated into Epha5 +and Epha5- population (Prince et al., 2013; Wang et al., 2021). The Epha5 +population can be separated further by Pcdh10 expression. In the Kirrel2- cells, Epha5, Pcdh10, Tenm4, and Tenm2 mark separate groups. Ncam1, Epha5, and Cntn4 may contribute to specifying small sets of cells. For the V2Rs, Pcdh9, Cntn4, Tenm4, Tenm2, and Pcdh17 have a decreasing range of expression, which may be used to specify increasingly refined connection. Notably, even though Kirrel2 and Pcdh10 are mostly detected in the V1Rs, they are also expressed by small sets of the V2Rs.

The specification of a vomeronasal circuit needs to be tied to receptor expression. We next address whether a transcriptional code is associated with both VR and AG molecule expression. We calculated pairwise similarity among the receptors according to their expression of TF and AG genes (Figure 7A). The result shows that V1R and V2R are distinctively separated. Besides a small group of broadly expressed V2Rs, all VR types are unique in their gene expression. Fpr types are more similar in their expression profile with the V1Rs, but the OR types are intermingled with both VR types.

To further determine the TF/AG associations that are specific to individual receptor types, we applied stringent criteria to calculate the Jaccard Index for each pair of TF/AG (Figure 7B), which reflect the statistical probability of co-expression within the same cell. The analysis reveals unique combinations for every receptor type (Figure 7C–G). Even though the only TF associated with V1R expression is Meis2, other TFs are involved in specifying AG gene expression. For example, cells expressing V1Rs with high sequence homology and in chromosomal proximity share a similar set of TF and AG genes, but the expression patterns are distinct from each other (Figure 7D). For broadly expressed V2Rs, Vmn2r3 and Vmn2r7, which are co-expressed by the same set of cells, share nearly identical TF/AGs (Figure 7E). In contrast, Vmn2r1, which does not co-express with either Vmn2r3 or Vmn2r7, lacks the expression of Pou2f1 and Tenm2 despite sharing all other genes. Other V2R types also show distinct TF/AG associations (Figure 7F–G).

Wildtype CD1 postnatal day 14 (P14) pups and adults (P56) were used for the experiment. Both sexes were randomly assigned to the experiment. All animals were maintained in Stowers LASF with a 14:10 light cycle and provided with food and water ad libitum. Experimental protocols were approved by the Institutional Animal Care and Use Committee (IACUC) (#2022–151) at Stowers Institute and in compliance with the NIH Guide for Care and Use of Animals.

To test the hypothesis that neurogenesis occurs in the marginal zone of the VNO we first set a minimum cell-type prediction-score threshold of ≥ 0.3; all cells below the threshold were labeled ‘unknown’. Then we used the simple features R package, sf (v1.0.16) to delineate regions of interest in the VNO. We excluded the non-neuronal region of the VNO from the analysis. Using the intersectional boundary between neural and non-neuronal epithelia as the center, we quantified the cells falling within a 750-pixel radius as in the marginal zones. Those fall out of the 750-pixel but within a 1500-pixel radius were quantified as in the intermediate zones. All remaining cells were classified as occurring in the ‘main zone’. Plots displaying the results were created using the function, ggplot, from the R package ggplot2 (Wickham et al., 2016).