Ret deficiency decreases neural crest progenitor proliferation and restricts fate potential during enteric nervous system development

Significance

Hirschsprung disease (HSCR) is a developmental disorder arising from the inability of enteric neural crest–derived cells (ENCDCs) to properly proliferate and migrate to form a functional enteric nervous system (ENS). Approximately 50% of HSCR patients harbor LoF variants at the receptor tyrosine kinase gene RET. To decipher the cell type–specific effects underlying Ret LoF, we examined the cellular heterogeneity of the ENS during mouse gastrointestinal tract development in the presence and absence of Ret, using single-cell RNA sequencing. Consistent with prior observations, we demonstrate that Ret deficiency leads to precocious differentiation and reduction in the number of proliferating ENS precursors. Our analysis further demonstrates that Ret expression is required for neuronal commitment to a specific postmitotic developmental trajectory.

Abstract

The receptor tyrosine kinase RET plays a critical role in the fate specification of enteric neural crest–derived cells (ENCDCs) during enteric nervous system (ENS) development. RET loss of function (LoF) is associated with Hirschsprung disease (HSCR), which is marked by aganglionosis of the gastrointestinal (GI) tract. Although the major phenotypic consequences and the underlying transcriptional changes from Ret LoF in the developing ENS have been described, cell type- and state-specific effects are unknown. We performed single-cell RNA sequencing on an enriched population of ENCDCs from the developing GI tract of Ret null heterozygous and homozygous mice at embryonic day (E)12.5 and E14.5. We demonstrate four significant findings: 1) Ret-expressing ENCDCs are a heterogeneous population comprising ENS progenitors as well as glial- and neuronal-committed cells; 2) neurons committed to a predominantly inhibitory motor neuron developmental trajectory are not produced under Ret LoF, leaving behind a mostly excitatory motor neuron developmental program; 3) expression patterns of HSCR-associated and Ret gene regulatory network genes are impacted by Ret LoF; and 4) Ret deficiency leads to precocious differentiation and reduction in the number of proliferating ENS precursors. Our results support a model in which Ret contributes to multiple distinct cellular phenotypes during development of the ENS, including the specification of inhibitory neuron subtypes, cell cycle dynamics of ENS progenitors, and the developmental timing of neuronal and glial commitment.

The ENS is derived from migrating vagal and sacral neural crest cells (1–5) that colonize the developing gut between embryonic day (E)9.5 to E14.5 in mice (6) and Carnegie stage (CS) 14 to CS22 in humans (7), leading to the neuronal networks of the myenteric and submucosal plexuses. Beyond the neural crest origins of ENS, lineage tracing studies have demonstrated that a small portion of both neuronal and glial cells in the ENS are also derived from Schwann cell precursors (SCPs) (8).

The ENS has multiple essential roles in motility, absorption, and immunity (9), and disruption of proper ENS development or organization underlies several disorders of the gastrointestinal system (10). Previous studies of ENS development have demonstrated that invading enteric neural crest–derived cells (ENCDCs) are predominantly neuronal precursors early in mouse development (E10.5), but as development proceeds, they differentiate into either glia, expressing classical glial markers including Gfap, Plp1, and S100b (11, 12), or neurons (11). Single-cell transcriptional profiling of mouse ENCDCs during mid and late (E12.5, E15.5, and E18.5) developmental time points reflects this heterogeneity and temporal change in cell fate (13–16).

One of the key factors involved in the migration and differentiation of ENCDCs into the developing gut is the receptor tyrosine kinase RET (13, 17). RET expression in ENCDCs promotes proliferation and cell survival via the MAPK and PI3K/AKT pathways (18). RET LoF leads to HSCR, an uncommon (~1/5,000 live births) developmental disorder of the ENS characterized by the absence of enteric ganglia along variable lengths of the distal bowel (19–23). In the developing mouse gut, Ret deficiency leads to significant transcriptional changes in transcription factors, signaling molecules, and genes involved in specific transport and biosynthesis processes, leading to aganglionosis (24, 25). Tissue-level bulk transcriptomic studies in Ret-deficient developing mouse gut (24, 25) have uncovered significant expression changes in genes harboring mutations in HSCR patients such as Ednrb,Gdnf, Sox10, Pax3, and Phox2b (24–28).

Development and fate specification in the ENS occurs in conjunction with migration along the forming gut tube (12, 29). As such, temporal and spatial influences affect ENCDC competence and can modulate fate decisions during development (12, 30). While Ret is known to play a role in the proper establishment of the ENS, the cell type–specific effects of Ret LoF on normal ENS development have not yet been elucidated and could contribute to our understanding of normal ENS development.

To identify the cell type–specific contributions of Ret to the proper organization and development of the ENS, we examined the cellular heterogeneity of the ENS during mouse gut development in the presence and absence of Ret. We profiled the transcriptomes of 1,003 Ret-expressing single cells in heterozygous and homozygous Ret-null mice at E12.5 and E14.5. Ret heterozygous mice do not exhibit aganglionosis in the developing gut or any other known phenotypes associated with Ret LOF and hence serve as a suitable control for our study. ENCDCs and their progeny were isolated using a cyan fluorescent protein (CFP) reporter driven by the endogenous Ret promoter for enrichment (31). Using a combination of differential gene expression analysis, gene coexpression pattern analysis, RNA velocity, and cell cycle state predictions, we identify several key developmental phenotypes associated with loss of Ret. The specific effects of Ret LoF vary across different cellular states in ENCDCs, including effects on progenitor proliferation, migration, differentiation, and fate specification. We confirm a previously described bifurcation in enteric neuronal developmental fate specification (14, 16) at an earlier time point. We further demonstrate that Ret expression is required for commitment to one of these two neuronal fates and, subsequently, the establishment of cell types from a predominantly inhibitory neuron developmental trajectory. Additionally, we show that homozygous Ret LoF leads to precocious differentiation of ENCDCs and a reduction in ENCDC progenitor proliferation rate. Each of these observed phenotypes may independently contribute to the deficits in proper ENS development, organization, and function observed in HSCR. These results highlight the diverse cell type–specific roles for Ret in the proper development of the ENS, with implications for understanding HSCR-related phenotypes.

Results

To assess cell type–specific roles for Ret in the developing ENS, we used a previously described mouse line harboring a Ret-null fluorescent reporter allele (Ret^CFP) (31). All ENCDCs express Ret during early development (17), enabling isolation via FACS in this reporter system. Ret-null (Ret^CFP/CFP) mouse embryos exhibit total colonic aganglionosis and kidney agenesis and are perinatal lethal (31). Importantly, no defects in ENS formation have been observed in Ret-heterozygous (Ret^CFP/+) mice (31), nor any significant difference in genome-wide expression compared to wild-type (WT) GI tract (32). Comparison of Ret^CFP/CFP to Ret^CFP/+ mice can therefore elucidate the mechanisms by which Ret LoF affects the proper formation of the ENS.

To investigate transcriptional changes of ENCDCs, we performed single-cell RNA sequencing (scRNA-seq) on FACS-enriched populations of ENCDCs in the developing murine gut. Briefly, CFP-positive single cells were collected from dissociated stomach and intestine of male and female embryos from Ret^CFP/+ and Ret^CFP/CFP at E12.5 and E14.5, and subjected to a modified Smart-seq2 scRNA-Seq library preparation (33, 34) (Methods). Libraries from 95 cells for each of two biological replicates were sequenced from all age, sex, and genotype combinations, with the exception of E14.5 Ret^CFP/CFP embryos. Consistent with a previously described reduction in CFP⁺ cells in a Ret hypomorphic mouse model (31), we observed reduction in the cell capture rate in the E14.5 Ret^CFP/CFP mice as compared to other ages and genotypes. We also observed a pronounced sex bias in the number of CFP⁺ cells recovered from E14.5 Ret^CFP/CFP embryos, with fewer viable cells collected from males, consistent with the established sex bias observed in HSCR (35) and described in other mouse models of HSCR (31, 36). As such, only 22 cells from one male E14.5 Ret^CFP/CFP embryo and 190 cells from four female E14.5 Ret^CFP/CFP embryos were recovered (Dataset S1). Full-length cDNA scRNA-Seq libraries from 1,341 individual cells were sequenced to an average depth of 9.91 × 10⁵ (SD 7.43 × 10⁵) paired-end reads per cell. Low-quality cells were removed using standard metrics (Methods) leaving 1,003 cells (Fig. 1A and Dataset S1) with an average read depth of 1.16 × 10⁶ (SD 6.83 × 10⁵) paired-end reads per cell. Relative FPKM values were converted to estimated RNA copies per cell (CPC) using the CENSUS algorithm (37). We identified a set of 14,319 expressed genes with greater than zero CPC in at least 20 cells and a SD greater than zero across all cells. To visualize the transcriptional relationships between cells, we identified a set of 419 high-variance genes (Methods) to use as input for uniform manifold approximation and projection (UMAP) dimensionality reduction to establish a biologically relevant, two-dimensional embedding of the cells (Fig. 1B and SI Appendix, Fig. S1).

Fig. 1.

A cellular map of Ret-expressing enteric nervous system cells. (A) Bar plot of post-quality-control count of sorted Ret^CFP/+ and Ret^CFP/CFP expressing cells from E12.5 and E14.5 embryonic gut in male and female mice. (B and C) UMAP of 1,003 cells colored by (B) genotype and (C) cluster. (D) Stacked bar plot of genotype proportions by cluster. Asterisk denotes SD from naive proportion (chi-squared test for equal proportions, corrected P < 0.05). (E and F) UMAP colored by predicted cell type (F) with RNA velocity vector field superimposed. The dotted line indicates neurogenic commitment, and the solid line indicates gliogenic commitment.

Annotation of Cell Types and States of the Developing ENS.

To quantify cell type and cell state differences with respect to genotype, we first used a Gaussian-based clustering algorithm (38) to identify nine clusters (Fig. 1C) and calculated gene-wise specificity scores to identify gene expression profiles for each cluster (SI Appendix, Fig. S2A). High specificity scores relative to mean expression identified clusters of actively cycling progenitors (cluster 2, Hist1h family genes; cluster 3, Aurka/b and Cdk1), a neurogenic commitment window (cluster 4, Neurod4; cluster 5, Pcp2 and Hes5; cluster 6, Hes6 and Btg2) (39, 40), maturing neurons (cluster 7, Etv1, Vip, Gal, and Cartpt; cluster 8, Slc18a3 (VAChT), Npy, Snap25, and Gfra2; cluster 9, Th, Dbh, and Ddc) (41–43), and a population of postmitotic glial-committed cells (cluster 1, Plp1 and Myl9) (44) (SI Appendix, Fig. S2B and Dataset S2). The neural progenitor cell marker p75 (Ngfr) and the pan-neuronal marker TuJ1 (Tubb3), used to label mature enteric neurons along with Ret (11, 45), have distinct, largely nonoverlapping expression profiles in Ret-expressing cells at these time points (SI Appendix, Fig. S2C).

To annotate cells as either progenitors, developing glia, or developing neurons, we evaluated canonical marker gene expression for mature ENS cell types. However, during development, the majority of cells may not yet express markers of mature, terminally differentiated cell types. To address this, we assigned cell types using a combination of transfer learning and marker gene expression to predict the most likely annotation based on an existing cellular atlas of adult mouse enteric neurons and glia (46). To uncover transcriptional signatures of mature ENS cell types, we learned patterns of gene coexpression on the annotated set of enteric neurons and glia from the public dataset using single-cell Coordinated Gene Activity in Pattern Sets (scCoGAPS) (47) and trained a random forest (RF) model on those learned patterns to predict cell type (Methods). We next projected our single-cell data into the latent spaces defined by these patterns using projectR (47, 48) (SI Appendix, Fig. S3) and employed the RF model to classify cells based on their projected pattern weights (SI Appendix, Fig. S4 A–C). We next calculated cell type proportions by cluster (SI Appendix, Fig. S4D). For clusters in which at least 90% of cells shared an annotation, we transferred the majority annotation to all members of the cluster. As a result, clusters 1 to 5 (1 = 99% glia, 2 = 98%, 3 = 92%, 4 = 95%, and 5 = 94%) were annotated as glia, and clusters 7 and 8 (7 = 98% neuron and 8 = 90%) were annotated as neurons. Annotations for clusters 6 (58% glia) and 9 (63% neuron) were not altered.

Our single-cell gene expression profiling is a snapshot in development for cells that have not yet committed to mature identities. To infer the direction of state transitions for our ENCDC population, we calculated RNA velocity estimates (49) and observed that the velocity field models diverging trajectories from a common origin (Fig. 1F). Cells with high projection weights in patterns associated with proliferating ENCDCs (pattern 40: Brca1/2, Cdc6, and Pola1/2; pattern 41: Ccna1/b1, Cenpe, and Kif24; pattern 42: Aurka/b, Bub1, and Top2a; SI Appendix, Fig. S3 and Datasets S3–S5) (50, 51) exhibit an ellipsoid pattern of velocities, consistent with transition through the cell cycle. Extending from this progenitor population, we observed a cluster of cells expressing canonical markers consistent with progression toward a neuronal fate (cluster 6: Btg2 and Hes6, SI Appendix, Fig. S2B) (39, 40) that further branches into two distinct trajectories of neuronal specification (Fig. 1F), consistent with previous observations of postmitotic branching of cellular fate specification for neural crest–derived neurons in later gut development (14). We infer through shared marker gene expression that cluster 7 in our data corresponds to developmental branch (branch A) (Etv1⁺, Nos1⁺, Vip⁺, and Gal  ⁺), and cluster 8 corresponds to developmental branch B (Bnc2⁺, Dlx5⁺, Mgat4c⁺, and Ndufa4l2⁺) (SI Appendix, Fig. S6A), as described previously in E15.5-E18.5 small intestine from WT mice (14). These data highlight that Ret-expressing cells isolated from the GI tract encompass two broad fates during development (glial or neuronal commitment) and confirm that neuronally committed ENCDCs diverge postmitotically into two distinct developmental trajectories, outlining the diversity of cellular fate trajectories accessible to ENCDCs within the developing mouse gut.

Ret LoF Induces Distinct Transcriptional Responses in Different ENS Cell States.

We asked whether loss of Ret changes the proportion of ENCDC subtypes in our dataset. To assess this, we tested for genotype bias in cluster composition by comparing the proportion of Ret^CFP/CFP cells against the naive (overall) proportion of 34%. Using a chi-squared test for equal proportions, we observed significant deviations in seven of the nine identified clusters: clusters 1, 8, and 9 are enriched for Ret^CFP/CFP cells (cluster 1: 83%, Bonferroni-adjusted P = 9.86 × 10⁻¹⁸; cluster 8: 57%, P = 5.64 × 10⁻⁴; cluster 9: 97%, P = 1.24 × 10⁻¹¹), clusters 4 to 7 are depleted for Ret^CFP/CFP cells (cluster 4: 9%, Bonferroni-adjusted P = 3.48 × 10⁻⁵; cluster 5: 20%, P = 1.00 × 10⁻³; cluster 6: 33%, P = 3.44 × 10⁻³; cluster 7: 2%, P = 6.33 × 10⁻⁵), while clusters 2 and 3 are relatively invariant (cluster 2: 42% Ret^CFP/CFP, Bonferroni-adjusted P = 0.196; cluster 3: 35%, P = 1.0) (Fig. 1D). Clusters 6, 7, and 8 are neuronal-committed clusters and differ significantly from the naive proportion. Clusters 6 and 7 are enriched for Ret^CFP/+ cells and 8 is enriched for Ret^CFP/CFP cells, indicating strong genotypic effects associated with neuronal fate specification. Three glial clusters (1, 4, and 5) differ significantly from the naive proportion as well, with cluster 1 enriched for and clusters 4 and 5 depleted of Ret^CFP/CFP cells. Cluster-specific enrichment of genotype within the neuronal and glial populations suggests that genotypic effects of Ret LoF modulate cell fate commitment with divergent results across ENCDC cell states.

Consistent with previous studies (8, 52, 53), the majority of Ret-null cells captured in our assay correspond to cells associated with vagal nerve fibers in the proximal stomach and are therefore parasympatho-enteric neural crest–derived, and hence derived from SCPs (Fig. 2E). The only cluster separate from the primary manifold is cluster 9, which is enriched for Ret^CFP/CFP cells (29/30, 97%, P = 1.24 × 10⁻¹¹, Fig. 1D). This cluster represents a transcriptionally distinct population that expresses markers of sacrally derived SCPs (Th+, Dbh+) as well as differential Hox gene expression consistent with more caudal development. Specifically, cluster 9 cells exhibit decreased expression of rostral Hox genes and increased expression of caudal Hox genes (SI Appendix, Fig. S5), indicating that these cells may derive from the CFP+ cells detected in the distal hindgut at E14.5 (Fig. 2E). This population likely corresponds to sacral neural crest cells that colonize the hindgut in a caudorostral direction around E14.5 (29, 54–56). While cluster 9 is enriched for Ret^CFP/CFP cells, the majority of cells in cluster 9 (28/30) derive from only one of the nine Ret^CFP/CFP embryos sampled, thereby limiting our ability to make definitive interpretations with respect to genotype.

Fig. 2.

Cellular and molecular changes driven by Ret loss of function. (A) Heatmap of genes DE with respect to genotype. (B) GO gene sets enriched for DE genes with respect to genotype (hypergeometric test, P < 0.05). (C) Correlation of scCoGAPS patterns with annotated cellular features. (D) Violin plots of DE genes associated with the anterior-posterior patterning GO term. (E) Immunofluorescence staining of mouse GI tract at E12.5 and E14.5 showing organization of CFP-expressing cells. Columns one and two show the stomach, and column three shows the hindgut. All IF images 10× magnification. (Scale bar, 200 μm.)

To identify the transcriptional effects of Ret LoF on the developing ENS, we tested 14,319 genes and identified 520 as differentially expressed (DE) with respect to genotype across all cells (monocle likelihood ratio test; Benjamini–Hochberg corrected q value < 0.05, Fig. 2A and Dataset S6A). Hierarchical clustering based on the expression of these 520 genes revealed that cells cluster first by cell type and then by genotype, again suggesting that the transcriptional changes observed are largely cell type specific. Gene ontology (GO) analysis revealed enrichment for several relevant processes, including gliogenesis, cell fate commitment, and axon development (hypergeometric test q value < 0.05, Fig. 2B and Dataset S7A). Significantly up-regulated genes expressed in Ret^CFP/CFP progenitor/glia include several associated with glial maturation (Fabp7, Plp1, and Myl9) (44, 57), indicating that these progenitor/glia cells may be more mature than their Ret^CFP/+ counterparts. A second cluster of genes is more highly expressed in Ret^CFP/+ neurons and includes genes with specific expression in the neurogenic branch A (Cartpt, Vip, Nos1, and Etv1), consistent with our observation that Ret^CFP/+ cells are significantly overrepresented in branch A (cluster 7).

We next sought to map the transcriptional signatures learned from our ENCDC dataset to specific cellular features by using coordinated sets of coregulated genes. To this end, we applied scCoGAPS to learn 25 patterns of gene coexpression (SI Appendix, Fig. S6 and Dataset S8). We calculated the Pearson correlation for pattern weights by cell against both known and learned cellular annotations (Fig. 2C and Dataset S9). This revealed patterns that correlated with technical (read depth and alignment rates), experimental (genotype, age, and sex), as well as biological (cell type and cluster) variables. We identified patterns correlated with the acquisition of different neuronal identities (pattern 20: Elavl4 (HuD), Nrxn3, Tubb3, Nnat, Syt1; pattern 21: Nos1, Tubb2a, Mapt, Sncg, and Ntng1; pattern 22: Onecut2, Gabrg2, Gabra3, Slc18a3 (VAChT), Gfra2, Elavl2, and Elavl3) (SI Appendix, Fig. S6 and Dataset S10) (41), a transient neurogenic window (pattern 19: Pou4f1, Gadd45g, Neurod4, Hes6, Btg1/2, Sox11, and Phox2b) (24, 39), as well as glial identity (pattern 12: S100b, Plp1, and Fabp7) (24, 44, 57). We also identified patterns well correlated with specific genotype-biased clusters of cells including pattern 18 (Ret, Gal, Ass1, and Etv1) and pattern 21, specifically used by branch A neurons in cluster 7, and pattern 12, specifically used by cells in cluster 1 undergoing glial maturation and significantly enriched for Ret^CFP/CFP cells.

Previous studies in mice have shown a severe migratory delay beginning as early as E9.5 in the absence of Ret (8, 17, 31). Consistent with this, we identified four gene expression patterns (patterns 2, 18, 19, and 25), well correlated with genotype, for which Hox family members were identified as pattern markers (58) (Dataset S10). Pattern 25 markers include the more distally expressed Hoxa/c9 and Hoxa/c10. The separation of cluster 9 from the main manifold of ENCDCs, combined with the specific expression of marker genes such as Maob and Pcdh10 (59–61), and specific usage of pattern 25, suggests that these cells are likely caudal SCPs. Enrichment for Hox genes within the remaining three pattern marker lists with strong correlation to genotype, suggests that Hox gene expression in ENCDCs may be a discriminating feature between the two genotypes. This was confirmed by the identification of seven Hox genes, (Hoxa2, b2, d3, and the more caudally expressed Hoxa4, a5, b5, and b6), as well as several anterior-posterior pattern specification genes (Dll1, Gpc3, Hes1, Lef1, Nrarp, Sema3c, and Tbx3), significantly DE (monocle likelihood ratio test; q < 0.05) with respect to genotype (Fig. 2 A and B). With the exception of Gpc3, Lef1, and Sema3c, all of the DE genes associated with anterior-posterior positioning show increased expression in Ret^CFP/+ cells (Fig. 2D). Consistent with this and other previous work at E12.5 (62), we observe that ENCDCs in the absence of Ret have not left the extrinsically innervating nerve fibers of the stomach by E12.5 nor by E14.5 (Fig. 2E), in stark contrast to Ret^CFP/+ cells which have formed a dense network throughout the stomach and all but the end of the hindgut by E12.5, as observed in the Ret heterozygous mouse model (31). As the Hox gene family is known to contribute to axial patterning, this differential expression of Hox genes may therefore be a consequence of significant rostrocaudal positional differences in ENCDCs in the presence and absence of Ret.

Ret Is Required for Commitment to the Branch A Neuronal Developmental Trajectory.

Recent work in E15.5 mouse gut shows that after neuronal commitment cells of the developing ENS branch into two main trajectories, referred to here as branch A (Etv1+, Gal+, Nos1+, and Vip+) and branch B (Bnc2+, Dlx5+, Mgat4c+, and Ndufa4l2+) (14). Our UMAP embedding reveals this branching behavior can be observed as early as E12.5 and that the choice of branch is Ret-dependent (Fig. 3A and SI Appendix, Fig. S7A). As observed above, cluster 7 (branch A), is almost exclusively Ret^CFP/+ cells. Conversely, while cluster 8 (branch B) is significantly enriched for Ret^CFP/CFP cells (Fig. 1D), both genotypes are represented along the branch B trajectory. This suggests a significant bias in the choice of developmental trajectory mediated in part by Ret expression in ENCDCs.

Fig. 3.

Ret expression is required for access to the branch A developmental trajectory. (A) UMAP of neuronal cells comprising branches A and B colored by genotype (Top) and cluster (Bottom). (B) Heatmap of significantly DE genes with respect to genotype for branch B (cluster 8). (C) Heatmap of significantly DE genes between clusters 7 and 8, irrespective of genotype. Row cluster segmentation represents the top six divisions of the hierarchical clustering of gene expression values. (D) UMAPs demonstrating persistent expression of Ret, CFP, and Gfra1 along the branch A developmental trajectory, induction of the noncanonical Ret coreceptor Gfra2 in branch B, and branch A marker gene Cartpt. (E and F) Immunofluorescence assay of CART at (E) E12.5 in Ret^CFP/+ and Ret^CFP/CFP stomach and (F) E14.5 in Ret^CFP/+ hindgut and Ret^CFP/CFP stomach. (G and H) Immunofluorescence assay of GFRA2 at (G) E12.5 in Ret^CFP/+ and Ret^CFP/CFP stomach and (H) E14.5 in Ret^CFP/+ hindgut and Ret^CFP/CFP stomach. All IF images 60× magnification. (Scale bars, 40 μm.)

We next sought to identify branch-specific differences in gene expression with respect to genotype. When we tested cluster 8 (branch B), we identified 12 genes DE with respect to genotype (monocle likelihood ratio test, q < 0.05, Fig. 3B and Dataset S6B), four of which are more highly expressed in Ret^CFP/+ and eight more highly expressed in Ret^CFP/CFP. GO analysis indicates that genes up-regulated in Ret^CFP/CFP are enriched (hypergeometric test, q < 0.05, Dataset S7B) for dopamine biosynthetic processes (q < 4.52 × 10⁻⁶, Agtr2, Ddc, and Th), catecholamine biosynthetic processes (q < 6.44 × 10⁻⁶, Agtr2, Ddc, and Th), and regulation of corticosteroid hormone secretion (q < 3.35 × 10⁻⁴, Agtr2 and Tac1). This suggests that while both genotypes exhibit largely similar transcriptional profiles within branch B, we cannot rule out functional consequences of Ret LoF within this trajectory.

To examine differences between branch A and B trajectories and identify branch-specific markers, we tested for differential expression irrespective of genotype and identified 111 genes (monocle likelihood ratio test, q < 0.05, Fig. 3C and Dataset S6C). GO enrichment analysis (Dataset S7C) revealed that genes with higher expression in cluster 7 are enriched for regulation of neurological system processes (hypergeometric test, q = 1.91 × 10⁻⁵, Cartpt, Ednrb, Fgfr1, Nos1, Nrg1, and Vip) and axonogenesis (q = 6.11 × 10⁻⁵, Ablim1, Auts2, Cdh11, Etv1, Lrrc4c, Lrrn3, Nrg1, Ntng1, and Ret), while genes more highly expressed in cluster 8 are enriched for catecholamine biosynthetic processes (q = 1.15 × 10⁻⁸, Agtr2, Dbh, Ddc, Gata3, Gch1, and Th), regulation of neurotransmitter levels (q = 2.00 × 10⁻⁶, Chrna7, Dbh, Ddc, Gch1, Nrxn1, Prima1, Slc18a1, Sytl4, and Th), and dopamine biosynthetic processes (q = 2.36 × 10⁻⁶, Agtr2, Dbh, Ddc, Gch1, and Th). The increased expression of dopamine biosynthetic process genes in Ret^CFP/CFP cells compared to Ret^CFP/+ cells within cluster 8 (Fig. 3B) suggests that differentiation and/or maturation rate of cells in branch B may be affected by loss of Ret. Additionally, several established markers of branches A and B (14) are DE in accordance with expectations: Etv1, Gal, Nos1, and Vip are higher in cluster 7/branch A; Bnc2, Dlx5, and Ndufa4l2 are higher in cluster 8/branch B.

Both Ret and CFP are down-regulated in branch B, in addition to Gfra1 (Fig. 3D), which encodes the canonical RET coreceptor, and Ednrb, which is epistatic with Ret (63–65). Interestingly, expression of these same genes is maintained in Ret^CFP/+ cells in branch A, and each exhibits a biased expression for cells along this trajectory (Fig. 3D). The exclusion of Ret^CFP/CFP cells and persistent expression of Ret and Gfra1 indicates that commitment to the branch A developmental trajectory, and subsequent establishment of the enteric neuronal subtypes derived from cells adopting this trajectory, requires continued Ret expression. The downregulation of both Ret and CFP expression in branch B, which contains numerous cells of both genotypes (Fig. 3 A and D), suggests that this trajectory does not require persistent Ret expression. While prior immunohistochemical studies found RET protein levels decrease in emerging glia and are maintained in emerging neurons (45), our scRNA-Seq data indicate that Ret expression is only maintained in the branch A neuronal trajectory and is down-regulated in the branch B trajectory, consistent with other recent single-cell studies (14, 16). Together these results suggest that a greater proportion of neuronally committed ENCDCs may engage the branch B trajectory in the absence of Ret, as branch A is an inaccessible developmental trajectory.

To validate branch A inaccessibility in Ret^CFP/CFP ENCDCs, we selected the DE genes Cartpt (branch A/cluster 7) and Gfra2 (branch B/cluster 8) as markers for in vivo validations due to their high expression and high specificity in each branch (Fig. 3D). We performed whole-mount immunohistochemistry (IHC) on the isolated gut (esophagus, stomach, and intestine) of Ret^CFP/+ and Ret^CFP/CFPembryos at E12.5 and E14.5, to identify and localize cells from each branch. We probed for CART (encoded by Cartpt) and CFP to localize branch A cells. In Ret^CFP/+ samples at E12.5, we observe coexpression of CART and CFP in the stomach (Fig. 3E) but not in the esophagus. By E14.5, we observe coexpression of CART and CFP through to the hindgut of Ret^CFP/+ samples (Fig. 3F). Consistent with our scRNA-seq data, we do not observe CART expression in the stomach of Ret^CFP/CFP samples at either age (Fig. 3 E and F), nor in the esophagus. We further validated the absence of branch A cells in Ret^CFP/CFP intestine using single-molecule fluorescent in situ hybridization (RNAscope) to assess mRNA expression of the branch A marker genes Cartpt and Gal (SI Appendix, Fig. S7 B and C). These results demonstrate that cells associated with the branch A developmental trajectory are not produced within the GI tract in the absence of Ret.

To determine the spatial distribution of branch B cells, we assayed expression of Gfra2, which encodes a noncanonical RET coreceptor and shows high, specific expression in branch B neurons. When we probed for GFRA2 and CFP via IHC, we did not observe GFRA2 expression in E12.5 Ret^CFP/+ stomach (Fig. 3G). We did, however, observe coexpression of GFRA2 and CFP restricted to cells along the extrinsically innervating nerve fibers of E12.5 Ret^CFP/CFP stomach (Fig. 3G), confirming that branch B neuronal development trajectory is indeed accessible in the absence of Ret. At E12.5, we did not observe GFRA2 outside the extrinsically innervating nerve fibers of the Ret^CFP/CFP stomach, nor in the intestines of either genotype, consistent with the previously described migratory defect and failure to effectively colonize the ENS in Ret^CFP/CFP mice. We were able to detect putative branch B cells (CFP⁺;GFRA2⁺) in the esophagus in both genotypes, as well as non-neuronal ENCDCs (CFP⁺;GFRA2⁻) and maturing branch B cells (CFP⁻;GFRA2⁺) (SI Appendix, Fig. S7D), indicating that branch B cells are spatially restricted to the esophagus in E12.5 Ret^CFP/+ samples and to the esophagus and extrinsic nerve fibers along the stomach in E12.5 Ret^CFP/CFP. By E14.5, CFP⁺/GFRA2⁺ branch B maturing neurons are observed throughout the Ret^CFP/+ intestine; however, CFP⁺/GFRA2⁺ cells remain restricted to the extrinsically innervating nerve fibers along the stomach in Ret^CFP/CFP embryos (Fig. 3H). Consistent with this difference, at E12.5 we see proportionally more Ret^CFP/CFP cells in cluster 8 (branch B) than Ret^CFP/+ within our scRNA-Seq dataset (28/37 Ret^CFP/CFP) (Fig. 1D). By E14.5, however, Ret^CFP/+ are the predominant cells in cluster 8 (21/33 Ret^CFP/+) (Fig. 1D). This could be related to an overall reduced number of E14.5 Ret^CFP/CFP cells in the intestine, or reduced capture efficiency as fluorescence decreases over biological time along this trajectory due to the downregulation of Ret, and by proxy CFP, as cells progress along branch B (Fig. 3D).

These data indicate that branch A and branch B represent two transcriptionally distinct developmental trajectories of maturing ENS neuronal subtypes that colonize the length of the intestines in the Ret^CFP/+ mice. Branch A neurons are never produced in Ret^CFP/CFP embryos. Furthermore, branch B neurons are produced at a potentially faster developmental rate, but remain restricted to the extrinsically innervating nerve fibers throughout development in Ret^CFP/CFP embryos. As Ret is required for cells to migrate into the developing ENS (Fig. 2E) (62) it follows that branch A, which appears to require Ret, would be inaccessible while branch B, which is Ret-independent, remains accessible in the absence of Ret.

HSCR-Associated Genes Are Expressed across Different Cellular States in the Developing ENS.

While coding and enhancer LoF variants in RET contribute to ~50% of cases of HSCR, rare pathogenic alleles (PAs) in 23 additional genes are also associated with HSCR (19–22). Many of these genes are in the RET-EDNRB GRN that controls ENS development (20). To assess the profiles of HSCR-associated and RET-EDNRB GRN genes in our data, we investigated the expression pattern of 28 genes to determine their coexpression patterns and specific effects due to Ret LoF (Fig. 4A). Three HSCR-associated genes (Gdnf, Edn3, and Nrtn) are not expressed in our ENCDC dataset, as they are predominantly expressed by the surrounding gut mesenchyme. Of the remaining 21 expressed genes (SI Appendix, Fig. S8 A–C), six are significantly DE (monocle likelihood ratio test, q < 0.1, Fig. 4 B and C and Dataset S6A) between Ret^CFP/+ and Ret^CFP/CFP cells (Gfra1, Nrg1, Phox2b, Ednrb, Sema3c, and Sox10). To identify cell type–specific effects of Ret LoF, we next tested the combinatorial effect of genotype and cell type. Three HSCR-associated genes (Gfra1, Sema3d, and Ednrb) exhibit significant cell type–specific differential expression (monocle likelihood ratio test, q < 0.05, Fig. 4 B and C and Dataset S6D). Gfra1 and Ednrb are significantly DE in both tests, suggesting a cell-autonomous effect of Ret LoF on the expression of these two genes. Conversely, four HSCR-associated genes (Nrg1, Phox2b, Sema3c, and Sox10) are globally DE but are not DE with respect to genotype:cell type interactions (monocle likelihood ratio test, q < 0.05; Dataset S6 A and D).

Fig. 4.

Cellular diversity of Hirschsprung disease genes. (A) Heatmap 28 HSCR-associated and RET-EDNRB GRN genes across all cells. Significantly DE genes are highlighted in red. (B) UMAPs showing expression of 11 genes associated with HSCR and/or in the RET-EDNRB GRN, and DE with respect to genotype and/or genotype-cell type interactions (monocle likelihood ratio test q < 0.05). (A and B) # indicates genes both associated with HSCR and in the RET-EDNRB GRN, ¶ indicates genes in the RET-EDNRB GRN not previously associated with HSCR, and no notation indicates HSCR-associated genes not in the RET-EDNRB GRN. (C) Violin plots of these 11 genes show expression with respect to genotype, with mean expression value indicated by a dot. (D) Bar plot showing the proportion of neurons and progenitors/glia by genotype at E12.5 and E14.5. Dashed lines indicate the naive genotype proportion at E12.5 and E14.5. The asterisk indicates a significant deviance from the naive proportion (chi-squared test for equal proportions, P < 0.05).

In addition to deviations in genotype proportion by cluster (Fig. 1D), we also observe that genotype proportions deviate significantly from expectation in E14.5 neurons and approach statistical significance in E14.5 glia (chi-squared test for equal proportions; neurons/P = 0.0158, glia/P = 0.0702) (Fig. 4D). Several HSCR-associated genes (Ednrb, Gfra1, Elp1, Sox10, and Zeb2) are expressed predominantly in progenitor/glia cells (SI Appendix, Fig. S8 A and C). Phox2b, Tcf4, and Ubr4 are ubiquitously expressed across all clusters in our embedding (SI Appendix, Fig. S8 A and C). L1cam is enriched in both maturing neurons and maturing glia, but exhibits reduced expression within the cycling progenitor population (SI Appendix, Fig. S8 A and C), while Ece1 appears predominantly expressed in maturing glial cells. These results suggest that global differential expression of HSCR-associated genes may in some cases be due to the confounding effects of differences in cell type proportions across genotypes. Finally, similar to the observed persistence of Ret expression along branch A, a subset of HSCR-associated genes are DE (monocle likelihood ratio test; q < 0.05, Dataset S6C) between the branches, with higher expression in branch A relative to branch B (Ednrb, Gfra1, Nrg1, and Ret), suggesting a potential role in the establishment of branch A identity in conjunction with Ret.

Despite significant genetic heterogeneity, ~67% of HSCR patients have mutations in RET-EDNRB GRN genes (19). The GRN includes bidirectional transcriptional feedback between RET and EDNRB, and the transcription factors SOX10, GATA2, RARB, and NKX2-5, as well as GFRA1 and its ligand, GDNF, and the ubiquitin ligase CBL, which breaks down phosphorylated RET (20, 25, 64). To ascertain whether this feedback is cell-autonomous or nonautonomous, we examined the expression of these genes in our dataset. Nkx2-5 and Gdnf are not expressed in our ENCDC dataset. Of the remaining six genes, four (Gata2, Gfra1, Rarb, and Sox10) are significantly DE between Ret^CFP/+ and Ret^CFP/CFP ENCDCs (Dataset S6A), with Gfra1 and Rarb more highly expressed in Ret^CFP/+ cells and Sox10 and Gata2 more highly expressed in Ret^CFP/CFP cells, although these moderate differences in expression may also reflect a disproportionate distribution of ENS cell types within our dataset with respect to genotype (Fig. 4 C and D and SI Appendix, Fig. S8C). Finally, with the exception of Sox10, which rapidly decreases expression during neuronal commitment, and Gata2, which is specific to cluster 9, Ednrb, Rarb, Gfra1, and Cbl all exhibit persistent expression along the branch A neuronal developmental trajectory (Fig. 4B and SI Appendix, Fig. S8C), consistent with a role for the RET-EDNRB GRN in establishing branch A identity. The expression of Sox10 exclusively within the progenitor population explains a previous observation that Sox10 haploinsufficiency only affects the maintenance of progenitor cells and not neurons (66). These results highlight the diverse cellular states in which HSCR-associated genes are expressed, suggesting that their roles in HSCR disease etiopathology may involve modulation of different aspects of ENCDC proliferation and development. Collectively, however, RET-EDNRB GRN genes are affected by Ret LoF and are persistently expressed along the branch A trajectory, which is inaccessible in the Ret^CFP/CFP mouse.

Ret LoF Reduces the Number of Actively Proliferating ENCDCs in the Developing Gut.

Reduction in proliferation capacity and precocious differentiation as a result of LoF of other genes within the developing ENS have been proposed as significant drivers of colonic aganglionosis (67, 68). Given the previously established role for RET as a proto-oncogene (18, 69) and its association with increased proliferation in other contexts (70–72), we asked whether Ret LoF was associated with significant differences in the proliferation rate of ENCDC progenitors. Differential gene expression analysis with respect to both genotype and cell type identified 331 genes (monocle likelihood ratio test, q < 0.05; SI Appendix, Fig. S9A and Dataset S6D), consisting of several genes associated with the mitotic cell cycle (Cdkn1c, Cdk1, Top2a, Ccnd1, Cenpk, Cenpe, Cenpf, Ccna2, Ccnb1, and Ccnb2), DNA replication (Pcna, Mcm3, Mcm5, Mcm6, Mcm10, and Gmnn), and cell cycle regulation (Rgcc, Mad2l1, Chek2, and Rad51ap1). Gene set enrichment analysis (GSEA) identified several cell cycle-associated gene sets including mitotic cell cycle processes, cell division, DNA replication, and regulation of cell cycle as top-scoring GO Biological Processes (BP) (hypergeometric test, q < 0.05; Dataset S7D). To focus on cells that are potentially engaging the cell cycle, we next performed a differential gene expression analysis on the subset of ENCDCs identified as progenitors or glia within the contiguous manifold (i.e., excluding ENCDCs committed to a postmitotic neuronal fate and excluding cluster 9). We identified 364 genes with significantly DE with respect to genotype (monocle likelihood ratio test, q < 0.05; Fig. 5A and Dataset S6E), and again GSEA identified gene sets associated with cell cycle regulation in addition to gliogenesis, cell fate commitment, and neural crest cell differentiation (hypergeometric test, q < 0.05; Fig. 5B and Dataset S7E). Thus, both cell cycle occupancy and differentiation state, two processes that are intrinsically linked during developmental fate specification, are disrupted in the Ret^CFP/CFP progenitor population relative to Ret^CFP/+.

Fig. 5.

Ret loss of function results in reduced cell cycle occupancy and a reduction in the number of actively proliferating ENCDC progenitors. (A) Heatmap of significantly DE genes with respect to genotype within the non-neuronal subset of ENCDCs. (B) Bar plots indicating the number of DE genes from (A) associated with each enriched GO:BP, along with P value for enrichment. (C) Tricycle embedding of ENCDCs to estimate cell cycle position. (D) Kernel density estimates for cell cycle position (θ) for each genotype at each age. Proliferative phase of the cell cycle (non-G0/G1) is between 0.25π and 1.75π. (E) Representative immunofluorescence staining of the proliferation marker pH3 and CFP in the stomach of E14.5 Ret^CFP/+ and Ret^CFP/CFP mice. Yellow arrowheads indicate CFP⁺ and pH3⁺ double positive cells. All IF images 40× magnification. (Scale bars, 20 μm.) (F) Box and whiskers plot showing quantification of CFP⁺ and pH3⁺ double-positive cells per embryo, normalized by total microscope fields of view analyzed per embryo. (G) Box and whiskers plot showing the quantification of the % of CFP⁺ cells with pH3⁺ staining in the GI tract of E14.5 Ret^CFP/+ or Ret^CFP/CFP mice.

We next asked whether we could identify any significant differences in cell cycle state assignments for ENCDCs in the absence of Ret. We used the tricycle package (73) to project the 1,003 single ENCDCs into the default reference cell cycle space provided (Fig. 5C). This analysis allows us to estimate the continuous position of each cell along the cell cycle (Θ) (Dataset S11), comparisons of which by genotype and age (Fig. 5D) suggest that at E12.5 there are minimal differences in cell cycle occupancy between genotypes (Kolmogorov–Smirnov test; P-value = 0.660), but by E14.5, there are consistently fewer Ret^CFP/CFP cells engaging the cell cycle than Ret^CFP/+ (Kolmogorov–Smirnov test; P value = 8.28 × 10⁻³).

To further assess whether Ret^CFP/CFP ENCDCs exhibit differential cell cycle occupancy relative to Ret^CFP/+, we next sought to infer cell cycle stage annotations for cells in our dataset using a projection-based approach with an independent, biologically relevant data source. Using the scCoGAPS patterns previously learned from publicly available data, and on which we trained the RF classification model, we observed high usage of three patterns within a population of cycling glia. GO enrichment analysis of pattern markers revealed these patterns correspond to different proliferative phases of the cell cycle (hypergeometric test, q < 0.05, Dataset S4 A and B). Pattern 40 is associated with DNA replication, corresponding to S phase; pattern 41 is associated with cilium assembly, organization, and segregation, and attachment of spindle microtubules to kinetochores, corresponding to prophase or metaphase (early M phase); and pattern 42 is associated with chromatid segregation and nuclear division, corresponding to anaphase or telophase (late M phase). To predict the cell cycle stage for individual cells in our ENCDC data, we scaled projected pattern weights linearly for each pattern to [0, 1] and interpreted these values as cell cycle stage occupancy probabilities (SI Appendix, Fig. S9 C and D). We tested all six possible comparisons of the four interaction conditions of genotype and age per pattern for differences in probabilities using two-sample Kolmogorov–Smirnov tests on the empirical cumulative density functions (SI Appendix, Fig. S9E) and used the Bonferroni method for multiple testing correction. Across the three patterns, E14.5 cells have significantly lower proliferative cell cycle stage occupancy probabilities than E12.5 cells of the same genotype (S phase: Ret^CFP/+ Bonferroni-adjusted P = 1.32 × 10⁻¹⁵, Ret^CFP/CFP P = 2.00 × 10⁻¹⁵; early M: Ret^CFP/+ P = 1.32 × 10⁻¹⁵, Ret^CFP/CFP P = 3.84 × 10⁻¹³; late M: Ret^CFP/+ P = 1.38 × 10⁻⁸, Ret^CFP/CFP P = 1.94 × 10⁻⁹), consistent with the expectation that as progenitor cells differentiate, the proportion of actively cycling cells will decrease over biological time. Again, cell cycle stage occupancy probabilities for ENCDCs did not differ significantly with respect to genotype at E12.5. By E14.5, however, occupancy scores for Ret^CFP/CFP cells are significantly lower across all three patterns (S phase: Bonferroni-adjusted P = 9.66 × 10⁻⁵, early M: P = 1.97 × 10⁻⁴; late M: P = 0.0193), suggesting overall fewer cycling ENCDCs in the Ret^CFP/CFP. This may be due to an increased differentiation rate, and consequently decreased proliferation rate, in Ret^CFP/CFP cells compared to Ret^CFP/+ cells between E12.5 and E14.5.

To validate this predicted decrease in cell cycle occupancy in Ret^CFP/CFP ENCDCs, we next used IHC to assess the number of CFP⁺ ENCDCs coexpressing the mitotic marker phosphohistone H3 (pH3) within the developing gut at E14.5 (Fig. 5E). We confirmed a significant overall reduction in the number of CFP⁺ cells in the Ret^CFP/CFP mouse model at E14.5 (Student’s t test, P < 1.0 × 10⁻⁴) (SI Appendix, Fig. S9G), consistent with the reduced number of CFP⁺ cells observed in our FACS-collected population (Fig. 1A) and previously published studies (31). We also identified a slight difference in the mean number of pH3⁺ cells at E14.5 (Student’s t test, P < 0.05), though modest enough to suggest that the overall number of mitotic cells within the developing gut remains unaffected (SI Appendix, Fig. S9H). This is consistent with a cell-autonomous effect of Ret on the cell cycle, as ENCDCs make up a small portion of all proliferating cells in the developing gut and would therefore not contribute significantly to the overall proliferative capacity of the gut as a whole. The overall number of double positive pH3⁺/CFP⁺ cells was significantly reduced in Ret^CFP/CFP embryos (Fig. 5F and SI Appendix, Fig. S7I) (Student’s t test, P = 0.029), and while not statistically significant, we observed a difference in the mean percentage of pH3⁺ cells within the subset of CFP⁺ ENCDCs in the Ret^CFP/CFP ENS at E14.5 indicating that a smaller fraction of developing ENCDCs are actively engaging the cell cycle in the absence of Ret (Student’s t test, P = 0.11) (Fig. 5G and SI Appendix, Fig. S9J). In aggregate, these results are consistent with the previously established role of Ret in enhancing proliferation and indicate that a loss of Ret in the developing ENS may contribute to an early reduction in the progenitor pool, possibly through induction of precocious differentiation, leading to a failure to colonize the more distal region of the developing gut.

Discussion

The ENS, initially derived entirely from the caudally migrating vagal and cranially migrating sacral neural crest cells, contains more than 100 million neurons of a variety of functionally distinct subtypes (74, 75). One of the critical genes that enables this transition of ENCDCs to diverse types of neurons is RET, the loss of function of which leads to HSCR. Despite the critical role Ret plays in ENS development, our understanding of the diversity of Ret-expressing cells in the developing GI tract, and the precise nature of the cell type–specific effects mediated by Ret LoF is limited. By performing scRNA-seq on enriched Ret-expressing cells during ENS development, we have generated a Ret-dependent fate transition map of ENCDCs to decipher the specific cellular phenotypes mediated by Ret LoF.

Our data indicate that the transcriptionally defined neuronal developmental state of ENCDCs bifurcates into excitatory and inhibitory motor neuron trajectories as early as E12.5, sooner than previously reported (14). Despite the understanding that ENCDCs may differentiate directly into neurons or via SCP (8), we observe no evidence to suggest that SCP- and non-SCP-derived neurons are transcriptionally distinct, either prior to or during this bifurcation. While our study captures major genetic and cellular changes due to Ret LoF at early developmental time points in the ENS, distinguishing transcriptional differences due to cell origin (e.g., SCP) versus Ret LoF exclusively would require lineage-tracing mouse models (8) at earlier time points. Furthermore, at this resolution, gross position within the developing gut does not appear to significantly alter the transcriptional state of branch B cells. While we observe branch B cells both in the intestines of Ret^+/CFP animals and in the stomach of both genotypes, we do not observe position-dependent transcriptional signatures within branch B cells. Therefore, we propose that development of SCP-derived cells and non-SCP-derived cells follow the same general transcriptional program for ENC neuron differentiation.

Ret LoF Affects the Establishment of a Specific Developmental Neuronal Trajectory in the Developing GI Tract.

Our data demonstrate that Ret-expressing cells broadly acquire neuronal and glial fates early during development of the GI tract and help explain why Ret LoF leads to total aganglionosis caused by impaired migration, defects in proliferation, and loss of immature ENCDCs (76). Our data highlight that Ret is required for the induction of an entire neuronal specification trajectory within the developing ENS which has previously been associated with the specific development of at least two distinct subtypes of inhibitory motor neurons: an interneuron subtype and a population of intrinsic primary afferent neurons (IPANs) (14). Although previous studies have posited that branch A corresponds to a predominantly inhibitory motor neuron (MN) trajectory and branch B corresponds to a predominantly excitatory MN trajectory (14), the developmental time points in our study prevent us from detecting this functional distinction given the lack of mature neurons.

Previous studies that characterized SCP-derived neurons in the adult gut as 80 to 90% excitatory cholinergic neurons defined this subclass based on the activity of the endogenous Ret promoter (8). Here, we have shown that the branch B developmental trajectory down-regulates Ret and CFP (Fig. 2E), indicating that the Ret promoter is likely deactivated during branch B cell maturation. While we are able to capture early branch B cells due to residual CFP, it is likely that residual CFP protein in branch B–derived cells will be undetectable by immunofluorescence by adolescence, when the proportion of primary excitatory neurons was described (8). Therefore, branch B cells would likely be specifically excluded from any study of the postnatal ENS that relies on the activity of the Ret promoter for detection. While the absence of branch A–derived cells may be the most relevant for HSCR disease etiology, the importance of the ability of branch B–derived cells to persist in the absence of RET should not be overlooked. At E12.5, we observe putative branch B cells (GFRA2⁺/CFP⁺) in the esophagus of the Ret^CFP/+ and in the esophagus and stomach of the Ret^CFP/CFP mice (Fig. 3F and SI Appendix, Fig. S6C). We also observe branch A cells (CART⁺/CFP⁺) in the stomach of Ret^CFP/+ mice but not the esophagus, and we do not observe branch A cells anywhere in the Ret^CFP/CFP GI tract at E12.5. By E14.5 we observe putative branch B cells in the stomach of the Ret^CFP/CFP, and throughout the gut in Ret^CFP/+ mice, consistent with published observations of both cell types in the small intestines of E15.5-E18.5 Ret^+/+ mice (14) The earlier establishment of branch B cells within the stomach of the Ret^CFP/CFP is consistent with a model of precocious differentiation we identify in this study. The specific failure to produce cells along branch A in the Ret^CFP/CFP suggests that the subset of mature ENS neuronal subtypes derived from this developmental trajectory may be the most acutely affected in disorders such as HSCR We provide evidence that aganglionosis observed in HSCR may arise from a combination of an early exit from the cell cycle for ENS progenitors, leading to precocious differentiation, and a failure to produce the full repertoire of mature ENS cell types. We do not detect any transcriptional signatures of cell death, suggesting that at this early time point, apoptosis or necrosis may not be a significant contributing factor in Ret-deficient aganglionosis. Our data suggest that Ret expression is not required for the establishment of a normal ENS glial identity; however, the observed effects on proliferation and differentiation potential suggest that enteric glia may be a preferred fate in Ret LoF models. Our data demonstrate that although the majority of Ret null cells in this particular model are associated with vagal nerve fibers in the proximal stomach, and likely vagal SCP-derived, a small proportion of these cells are actually located more caudally in the hindgut expressing markers of sacral SCP (Th+ and Dbh+). Since LoF variants in RET account for ~50% of HSCR patients (19), the presence of these cells in the hindgut (future colon) in a Ret-deficient background suggests that certain types of neurons might be still viable in these patients even in the aganglionic segments.

Neuronal Fates That Use Noncanonical RET Coreceptors Remain Accessible in the Absence of Ret.

Compared to a previous study (14), we observe the divergence of the neuronal trajectories as early as E12.5, when ENCDCs are still migrating and undergoing neuronal maturation. While we observe a near complete absence of Ret^CFP/CFP cells in branch A (cluster 7), we note that branch B (cluster 8) contains maturing neurons from both genotypes and is not grossly affected at the transcription level by loss of Ret. In the branch B trajectory, we observe a general reduction in the expression of Ret, CFP, and several RET-EDNRB GRN genes, including Gfra1, followed by upregulation of Gfra2, which encodes a noncanonical RET coreceptor. Although GFRA1 and GFRA2 are both in the GFRA family, GFRA1 preferentially binds GDNF as its ligand, while GFRA2 preferentially binds the GDNF family ligand NRTN. It is possible that GFRA2 binds to residual RET after downregulation of transcription; however, it raises the possibility that GFRA2 complexes with another coreceptor to activate an alternate signaling pathway responding to a different ligand in this Ret-independent trajectory. Differential expression of Gfra family genes within the aganglionic segment of HSCR patients has previously been observed (77), including a specific lack of GFRA1 expression in neuronal cells within the associated hypertrophied nerve fibers. Additionally, Ntrn and Gfra2 null mice exhibit a reduction in neuron fiber density and neuronal size in the ENS and in each case result in the specific loss of cholinergic excitatory myenteric innervation (presumably branch B derived) within the mouse gut, leaving the nitrergic neuronal population (branch A derived) largely unaffected (78, 79), consistent with the segregation of fates and corresponding gene expression differences described here. In aggregate, these results suggest that the canonical RET signaling pathway is responsible for the proper establishment of ENS subtypes derived from the branch A trajectory and that the activity of NRTN-GFRA2 signaling may play a greater role in the establishment of neuron identities along branch B, potentially independent from Ret.

Expression Patterns of HSCR-Associated Genes Highlight Necessity of Disease-Relevant Cellular Context.

The specific expression patterns of many genes that harbor pathogenic alleles in HSCR patients within the ENCDCs of the ENS should lead us to reassess the nature of aganglionosis in HSCR. Our data highlight the need to better understand the specific cellular states and contexts in which these genes and GRNs operate, as well as the cell-autonomous versus cell nonautonomous effects in HSCR. Genes such as Acss2, Eno2, and Adamts17 express very lowly in the ENS population (Fig. 4), yet are robustly expressed in both the mouse and human developing GI tract (19, 25). Furthermore, the design of our study allows us to only capture cell-autonomous effects; additional non-cell-autonomous effects have been observed in Ret LoF studies as well (80). Hence, patients carrying pathogenic alleles in the genes expressed in other cells may contribute to other GI anomalies, and aganglionosis may then be a secondary effect. Although verification of this would require detailed patient phenotyping, the shift between neuronal trajectories we have identified suggests developmental mechanisms that may guide future analysis of phenotypes.

Ret LoF Leads to Proliferation Defects in ENS Progenitors.

During formation of the ENS, the gut itself is increasing dramatically in both size and length, creating a huge proliferative burden on ENCDCs, the progenitors of the ENS. Our in silico estimates of cell cycle stage occupancy in conjunction with an in vivo assay of proliferation suggest a decrease in proliferation of ENCDCs in the absence of Ret. Even a small reduction in proliferative capacity could produce a significant decrease in population size over the course of the several days it takes to colonize the ENS, such as the absence of CFP+ cells we observe in the Ret^CFP/CFP mouse intestines.

Ret deletion in the mouse leads to significant changes in fate acquisition, progenitor proliferation, and ENCDC migration within the developing ENS. It is likely that RET deficiency alleles in humans lead to similar phenotypes in distinct subsets of ENS cell types and states in HSCR patients and that each of these phenotypes may play some combinatorial role in establishing the aganglionosis that is the hallmark of this disease. Furthermore, different RET-deficiency alleles and epistatic interactions may affect these phenotypes in different ways, potentially explaining some of the observed variable expressivity and incomplete penetrance of HSCR. Our study highlights the specific roles that Ret plays within different cellular contexts of the developing ENS and demonstrates that appropriate cellular context is critical for the interpretation of multiple disease-related cellular phenotypes that may emerge from a single causal insult.

Methods

Mouse Strains.

Ret CFP knock-in mice (Ret^CFP/+; MGI:3777556) were generated by crossing Ret^fl/+ mice to β-actin-Cre mice to remove the RET9 cDNA and maintained on a C57BL/6 background. All mouse experiments were conducted in accordance with the NIH Guidance for the Care and Use of Laboratory Animals. All procedures were approved by the Johns Hopkins Animal Care and Use Committee (Protocol number: MO21M56). All mice embryos used were either E12.5 or E14.5, and both male and female mice were used. Details of mouse breeding and genotyping are provided in SI Appendix.

Single-Cell RNA-Seq Library Preparation.

Libraries were prepared from two embryos for each of the eight conditions (Ret^CFP/+ and Ret^CFP/CFP, female and male, E12.5 and E14.5) with two exceptions: libraries were prepared from 1 Ret^CFP/CFP male E14.5 embryo and 4 Ret^CFP/CFP female E14.5 embryos owing to differential viability. Single-cell RNA-seq libraries were prepared following a modified Smart-seq2 protocol (33, 34). Libraries were sequenced (paired end, 50 bp read length) on an Illumina HiSeq 2500 to an average depth of 9.91 × 10⁵ (SD 7.43 × 10⁵) paired-end reads per cell.

Data Preprocessing.

Single-cell RNA-seq libraries were aligned with hisat2 version 2.0.1-beta to a mouse mm10 hisat2 index with the inserted CFP cDNA sequence included in the index. Using samtools 1.2, sam files were compressed to bam files and then sorted and indexed. Using Cufflinks v2.2.1 indexed bam files were quantified against the Gencode mouse vM10 assembly and normalized across all 1,369 samples (1,351 cells, 18 negative control wells). A total of 1,003 cells and 0 negative control wells passed quality control filters. The age, sex, and genotype of these 1,003 cells are as follows: 183 embryonic day (E)12.5 female Ret^CFP/+, 164 E12.5 male Ret^CFP/+, 173 E12.5 female Ret^CFP/CFP, 125 E12.5 male Ret^CFP/CFP, 161 E14.5 female Ret^CFP/+, 156 E14.5 male Ret^CFP/+, 37 E14.5 female Ret^CFP/CFP, and 4 E14.5 male Ret^CFP/CFP. FPKM values were transformed to CPC using the Census algorithm (37). Raw and processed data have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE192676. Details of filtering constraints are provided in SI Appendix.

scRNA-Seq Analysis.

To reduce the computational burden, analyses of gene expression were limited to expressed genes, defined as genes with nonzero CPC values in at least 20 cells and SD >1 × 10⁻²⁰. Differential gene expression tests were performed using the Monocle R package (81). UMAP was performed on 419 high-dispersion genes using the Umapr R package (82, 83). RNAVelocity estimates were obtained using the Velocyto R package (49). Cluster assignment was performed using the mClust package (38). Specificity scores for each gene were calculated using the Jensen–Shannon (JS) distance from the cummeRbund package (84). Details of the selection of high-dispersion genes are provided in SI Appendix.

Seventy gene expression patterns were learned on 11,624 cells from publicly available scRNA-seq data from the ENS of adolescent mice (mousebrain.org) using scCoGAPS (47). We projected our data into these 70 patterns using projectR (47, 48). Twenty-five patterns were learned from our data, and we projected the publicly available data into these 25 patterns. Details of pattern discovery with scCoGAPS are provided in SI Appendix.

The random forest model was trained on all 1,105 neurons and a subset of 1,660 glia from the publicly available data described above (mousebrain.org) to create a 60:40 glia:neuron training dataset to avoid model bias. Glia were subsampled such that the distribution of glial cells across clusters in the subset matched the distribution across clusters in the whole glial population. The model was trained on 90% of cells to predict cluster assignment from 70 pattern weights with 10-fold cross-validation resulting in an accuracy of 89.8%. As pattern weights are constrained to [0,1], projected pattern weights were normalized to [0,1] before being used as input to predict cluster assignment. The parent cell type of each cluster was taken as the predicted cell type for our data.

Immunofluorescence Assay on the Whole Mouse Gut.

Whole gut tissue (stomach, fore, and hindgut) at E12.5 and E14.5 was dissected and washed in ice-cold PBS followed by fixation in 4% PFA at 4 °C for 1 h. The fixed tissues were permeabilized and blocked in 5% normal donkey serum + 0.3% Triton-X for 1 h at room temperature (RT). The tissues were incubated with appropriate primary antibodies at 4 °C overnight, washed with PBS, and then incubated in appropriate secondary antibodies at 4 °C overnight (see Dataset S12 for antibodies and dilutions) followed by DAPI staining, mounting, and visualization. Details of immunofluorescence assays are provided in SI Appendix.

RNAscope.

Fluorescence in situ hybridization was performed following the manufacturer’s protocol for the RNAscope Fluorescent Multiplex V2 Assay using the RNAscope Multiplex Fluorescent Reagent Kit v2 (ACDBio #323100). Sections were deparaffinized and incubated in hydrogen peroxide for 10 m at RT and incubated in 1X Target Retrieval for 15 m at RT, then incubated in Protease Plus for 30 m at 40 °C. Sections were incubated in a 50:1:1 mixture of RNAscope Probes CFP (ACDBio #529381), Mm-Cartpt-C2 (ACDBio #432001-C2), and Mm-Gal-C3 (ACDBio #400961-C3) for 2 h at 40 °C. This was followed by standard steps per the manufacturer’s instructions for staining and quenching. Sections were then incubated in RNAscope DAPI for 30 s at RT and mounted with ProLong Gold Antifade mounting media and imaged using a Nikon Eclipse Ti confocal microscope. A detailed protocol is provided in SI Appendix.

Proliferation Assay and Quantification.

To measure cell cycle occupancy, we stained replicate Ret^CFP/+ and Ret^CFP/CFP whole GI tracts from both male and female mice at E14.5 for phosphohistone H3 (Anti-H3S10p; Millipore Sigma #06-570), an immunomarker specific to cells undergoing mitosis (85), and CFP protein (Anti-GFP; Aves labs #GFP-1010) using the immunofluorescence assay described above. Confocal stack images were acquired on the Nikon Eclipse Ti confocal microscope with a 40× oil objective and type A immersion oil and quantified using a custom CellProfiler (86, 87) workflow. A detailed protocol is provided in SI Appendix.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (XLSX)

Dataset S06 (XLSX)

Dataset S07 (XLSX)

Dataset S08 (XLSX)

Dataset S09 (XLSX)

Dataset S10 (XLSX)

Dataset S11 (XLSX)

Dataset S12 (XLSX)

Data, Materials, and Software Availability

Raw and processed sequencing reads data have been deposited in NCBI Gene Expression Omnibus (GEO) (GSE192676) (88).