Introduction

The enteric nervous system (ENS) controls motility and other gastrointestinal functions via its vast networks of interconnected neural ganglia ^1–3. Enteric neurons and glia are both generated from neural crest cell (NCC)-derived progenitors that colonize the developing bowel ^{4, 5}. Most progenitors originate directly from vagal NCCs, which enter the foregut mesenchyme and then migrate toward the hindgut to eventually colonize the entire gastrointestinal tract before birth ^{6, 7}. When this process fails, ENS ganglia are lacking in the rectum and over varying lengths of the colon, a severe birth defect known as Hirschsprung disease (HSCR) ^{5, 8, 9}. NCC-derived Schwann cell precursors (SCPs) constitute another source of ENS progenitors, which in this case use extrinsic nerves as entry points to colonize the bowel, mainly in colon, and after birth ^{10, 11}. However, although extrinsic innervation of aganglionic colon in HSCR is more abundant than normal ¹², the accompanying increase in SCPs cannot spontaneously compensate for the lack of vagal NCC-derived pool ^{13, 14}.

Current HSCR treatment still involves surgical removal of the bowel segment lacking ENS ganglia (aganglionic) ^{15, 16}. This procedure is generally life-saving but postoperative problems are common ^{15, 17–19}. As alternative to this imperfect treatment, we are developing a non-surgical approach based on post-natal stimulation of tissue-resident ENS progenitors ^{20, 21}. Using mouse models, we demonstrated that daily rectal administration of GDNF (glial cell line-derived neurotrophic factor) for only 5 days, from post-natal day (P)4 to P8, can regenerate the missing ENS in the colon, with long-term benefits on overall colon function and mouse survival ²⁰. Moreover, GDNF promoted de novo neurogenesis in human colonic explants obtained from HSCR surgery and cultured in vitro ²⁰. Our preliminary mechanistic studies suggested that the beneficial effects of GDNF are at least in part due to its ability to promote neuronal differentiation of SCPs from extrinsic nerves ²⁰.

During normal pre-natal ENS development, GDNF greatly influences the migration, proliferation and neuronal differentiation of vagal NCCs via the transmembrane tyrosine kinase RET (Rearranged during transfection), upon binding to co-receptor GFRα1 (GDNF family receptor, alpha 1) ^22–25. This is reflected in HSCR genetics, where RET variants constitute the main risk factor of this multifactorial disease ^{8, 9}, raising reasonable doubts about global applicability of GDNF-based therapy. Yet, in the Schwann cell lineage, GDNF-bound GFRα1 instead signals through NCAM1 (Neural cell adhesion molecule 1) ^{26, 27}, suggesting that RET could in fact be dispensable for mediating postnatal ENS-inducing effects of GDNF in a HSCR context. Clarifying this point will be key to advance GDNF-based therapy closer to clinical application.

A better understanding of cellular heterogeneity among tissue-resident ENS progenitors is also warranted. The majority of GDNF-induced neurons are not derived from SCPs ²⁰. Identifying these non-SCP ENS progenitors would allow to better evaluate the requirement for RET signaling, and perhaps also explain how the GDNF-induced ENS is properly balanced in terms of neuron subtypes ²⁰ despite that SCP progeny is normally biased toward a cholinergic phenotype ¹⁰.

Results

GDNF changes the proportions of neural cell subpopulations in the distal colon of Hol^Tg/Tg mice

To better understand the mechanism of GDNF-induced neurogenesis in the otherwise aganglionic colon of HSCR mice, we turned to single-cell RNA sequencing (scRNA-seq). We continued to focus on the Holstein mouse model for trisomy 21 (collagen VI upregulation)-associated HSCR ^28–30. This model outperforms two other models available in our laboratory in terms of either phenotypic penetrance of megacolon (compared to incomplete male-biased penetrance in TashT line ^{31, 32}) or fertility rate (compared to subfertility in Piebald-Lethal line ³³) – both of which being also responsive to GDNF treatment ²⁰. To facilitate ENS cell recovery, we bred Holstein mice with G4-RFP transgenics ³⁴. The G4-RFP reporter is known to label most NCC derivatives with the red fluorescent protein (RFP) DsRed2 during pre-natal development, including ENS progenitors ^{29, 34, 35}. In the most distal quarter of the colon from P10 Hol^Tg/Tg;G4-RFP double transgenic mice that were previously treated or not with GDNF enemas between P4-P8 (Fig.1), we validated that RFP marks with variable intensity 88-90% of SOX10+ glial cells (including extrinsic nerves-associated SCPs) in a GDNF-independent manner (Fig.1a,b) and 70% of GDNF-induced HuC/D+ enteric neurons (Fig.1b) – which are otherwise absent in untreated controls (distal aganglionosis covers ∼25% of colon length in the Holstein line ²⁹). Of note, unless otherwise indicated, all immunofluorescence stainings in the current study were performed on samples from the last quarter of the colon. Although the percentage of marked neurons was slightly lower than we were hoping, it is noteworthy that similar coverage (73%) did allow to profile a wide variety of postnatal enteric neurons in a recent scRNA-seq study ³⁶. Nonetheless, to maximize their recovery by fluorescence-activated cell sorting (FACS), RFP+ cells were sorted in “yield” mode (allowing co-sorting of RFP-negative cells that are close to RFP+ cells in the flow stream) using a non-stringent gating strategy that also included cells with very low levels of RFP signal (in the range of background autofluorescence). These FACS parameters were applied to dissociated cell suspensions from the distal third of the colon (including part of the hypoganglionic transition zone) of 6 untreated control and 8 GDNF-treated mouse pups, from which recovered cells corresponded to ∼1% and ∼2% of all single live cells, respectively. Individual cells were then processed using the 10x Genomics platform (Fig.2a), and subsequently analyzed with the Trailmaker software. This first allowed to separate recovered cells into 19 cell clusters of variable size, including seven associated with the ENS: three with an SCP identity (expressing specific markers like Dhh, Mpz, Mbp and Egfl8), two with an enteric glial cell (EGC) identity (expressing specific markers like Nell2, Col20a1, Ccn2 and Cpe), and another two with an enteric neuron identity (expressing general markers Elavl4, Tubb3, and Gap43). Consistent with our non-stringent gating strategy for FACS-mediated recovery, the other cell clusters were not directly associated with the ENS, including specific groups of lymphoid, myeloid, vascular, epithelial and smooth muscle cells. Therefore, to refocus on the ENS, we only selected cells that express either the pan-neuronal marker Elavl4 or the pan-glial and ENS progenitor marker Sox10 (Fig.2a), leaving 1354 and 3006 high-quality cells from untreated and GDNF-treated Hol^Tg/Tg;G4-RFP mice, respectively. Unsupervised reclassification of these ENS-associated cells revealed 14 distinct clusters representing subpopulations of SCPs (three clusters, including one with mixed EGC identity), EGCs (eight clusters, including three proliferating) and enteric neurons (three clusters) (Fig.2a-d), which are present in both untreated and GDNF-treated groups (Fig.2e). Analysis of relative proportions revealed a marked enrichment in enteric neurons and proliferating Mki67⁺ EGCs in the GDNF-treated group, which occurred mainly at the expense of SCPs and cells with mixed SCP/EGC identity (Fig.2f). These observations are consistent with our prior discovery that GDNF-induced neurons are at least in part derived from SCPs ²⁰, thereby making this dataset a reliable resource to better understand the underlying mechanism.

Figure 1.

Figure 2.

The neurogenic effect of GDNF in distal colon is mediated by NCAM1/FAK signaling

We first took advantage of our scRNA-seq data to examine the expression profile of GDNF receptor-coding genes, focusing on GDNF-treated Hol^Tg/Tg;G4-RFP dataset. A striking difference was noted for signaling receptors, with Ncam1 highly expressed in all cell clusters, while Ret expression is mostly enriched in a subset of enteric neurons only (Fig.3a). Binding receptor-coding genes Gfra1 and Gfra2 were also found to be widely expressed, with a partly overlapping and complementary distribution (Fig.3a). Gfra3 expression appeared restricted to SCP clusters, whereas weak Gfra4 expression could only be detected in a few neurons (Fig.3a). These observations strongly suggested that GDNF could bind either GFRα1 or GFRα2 ³⁷ and then signals through NCAM1 to make new neurons from SCPs and perhaps other ENS progenitors within the EGC clusters. To further validate this possibility, we evaluated the distribution of the NCAM1 protein and its downstream signaling effector phospho-FAK[Y397] in most distal colon of Hol^Tg/Tg mice over time, before (P4_[0h]), during (P4_[6h] and P6) and after (P10) the end of GDNF treatment. NCAM1 is already present in SOX10+ cells before GDNF treatment begins at P4, both within and outside extrinsic nerve fibers (Fig.3b,c). As soon as 6h after first GDNF administration, NCAM1 levels then increase over time, with high levels being maintained even two days after completion of treatment period (Fig.3b,c). Phospho-FAK[Y397] levels are lower in these SOX10+ cells before the start of GDNF treatment, but then similarly increase over time (Fig.3b,c). These changes in NCAM1 and phospho-FAK[Y397] levels do not appear to be due to a simple temporal effect, as we did not observe overt changes in the ENS-containing mid-colon of untreated Hol^Tg/Tg mice between P4 and P10 (Fig.3d). Hence, these additional observations suggest that NCAM1 could also be subjected to GDNF-induced autoregulation, as we previously showed for RET ²⁰.

Figure 3.

To confirm the relative importance of NCAM1 vs. RET in postnatal colon of Hol^Tg/Tg mice, we next verified GDNF treatment efficacy in presence of chemical inhibitors that prevent phosphorylation of FAK (PF-562271)³⁸ and RET (BLU-667)³⁹, respectively. Each of these inhibitors was administered via daily i.p. injections (30 mg/kg) during the GDNF enema treatment period (P4-P8), and impact on colon tissues was then assessed 6h after last enema/injection at P8 (Fig.4a). Western blot analysis of full-thickness proximal/mid colon confirmed efficacy of each inhibitor (Fig.4b), while further highlighting that RET is poorly activated in this context compared to a control tissue with robust endogenous GDNF-RET activity like the hypothalamus ⁴⁰ (Fig.4b). Accordingly, immunofluorescence staining of most distal colon segment from same animals revealed that phospho-RET[Y1062] inhibition with BLU-667 has no impact on GDNF-induced neurogenesis. In marked contrast, phospho-FAK[Y397] inhibition with PF-562271 led to a significant decrease in the number of GDNF-induced neurons per mm² (Fig.4c), without inducing apoptosis in the ENS (Fig.4d). Rare apoptotic neurons positive for cleaved Caspase 3 were only detected in presence of the phospho-RET[Y1062] inhibitor BLU-667 (Fig.4d). These data confirm that NCAM1/FAK signaling is crucial for GDNF-induced neurogenesis in aganglionic colon of a HSCR mouse model.

Figure 4.

GDNF-induced neurogenesis from distal colon SCPs partly involves an intermediary EGC-like state

We reasoned that our scRNA-seq data would also be helpful to identify additional sources of GDNF-induced neurons, other than SCPs ²⁰. Indeed, as some EGCs behave as ENS progenitors with a neurogenic potential under certain conditions ^{14, 41–45}, we hypothesized that at least one of the many subgroups of EGCs in our analysis of P10 mice would contain such progenitors. To uncover neuronal-fated progenitors, we verified the expression pattern of the late pan-neuronal marker Elavl4, predicting that it should start to be upregulated while the EGC transcriptional program is still active. Again focusing on the GDNF-treated Hol^Tg/Tg;G4-RFP dataset, this analysis confirmed that several Elavl4 expressing cells are among both SCP and EGC clusters, in addition to all enteric neuron clusters (Fig.5a). Reclustering these Elavl4 expressing cells then unveiled an intriguing connection between SCP, EGC and enteric neuron clusters (Fig.5b-d). This linear relationship was corroborated by a pseudotime trajectory analysis (without predicting directionality), which further confirmed it exists in untreated Hol^Tg/Tg;G4-RFP dataset as well (Fig.5e) – most likely reflecting presence of ENS-containing transition zone in our samples. Moreover, we observed a similar linear connection when the reclustering was based on Tubb3 expression instead of Elavl4, with overall inverse expression gradients of Sox10 and these pan-neuronal markers (Fig.5d). Considering all of the above, these analyses suggest the existence of a stepwise differentiation model (Fig.5c-d) whereby a subset of SCPs (expressing Dhh, Mpz, Mbp and Egfl8) progressively loose their Schwann cell lineage identity to differentiate into EGCs expressing Nell2, Col20a1 and Ajap1 (SCP/EGC and EGC #a clusters), then EGCs expressing Ccn2, Col9a2 and Igfbp4 (EGC #b cluster), and finally enteric neurons expressing general markers like Phox2b, Tubb3 and Gap43, as well as different combinations of subtype-specific markers like Chat, Nos1, Vip and Calcb (EN #a-f clusters). Interestingly, the EGC #b cluster also expresses Gfap, Slc18a2, Sox2 and Cpe (Fig.5c-d). These genes collectively characterize EGCs poised for neurogenesis in the small intestine of Plp1-GFP mice at P14 ⁴⁶. Moreover, these poised EGCs were recently assigned to the “EG 1-2” clusters identified in pediatric human ileum and colon biopsies ⁴⁷.

Figure 5.

To validate the stepwise differentiation model described above at both genetic and protein levels, we turned again to a cell lineage tracing system that previously allowed us to follow progeny of Dhh-expressing SCPs in different contexts ^{11, 14, 48, 49}, including Hol^Tg/Tg mice treated with GDNF ²⁰. In these GDNF-treated triple-transgenic Hol^Tg/Tg pups also bearing the Dhh-Cre transgene ⁵⁰ and the R26R-YFP reporter allele ⁵¹ (Hol^Tg/Tg;Dhh-Cre^Tg/+;R26^YFP/+), we now evaluated the presence of SCP-derived cells co-expressing either GFAP or SLC18A2 proteins together with the early enteric neuron marker PHOX2B in the distal colon at P10 (i.e. 2 days after last GDNF enema, as for scRNA-seq). Triple-immunofluorescence staining clearly shows that GDNF-induced ENS ganglia contain several SCP-derived (YFP+) cells that are differentiating into PHOX2B+ enteric neurons (as also evidenced by their typical large, rounded nucleus) while still expressing EGC markers GFAP (Fig.6a) or SLC18A2 (Fig.6b). Yet, quantitative analysis as a function of time revealed that GFAP+ cells in the distal colon are not all derived from Dhh-expressing SCPs on the first day of treatment (Fig.6c). Furthermore, this kinetic analysis highlighted a progressive decline of GFAP protein expression in SCP-derived cells in the distal colon during (P4_[6h] and P6) and after (P10 and P20) GDNF treatment, with a mean of 87% of YFP+ cells per ganglion also positive for GFAP after the first 6h of treatment to a mean of 31% at P20 (Fig.6c). This progressive decline most likely reflects the inverse progression of neuronal differentiation, as GFAP is no longer detectable in SCP-derived cells that stain positive for the late neuronal marker HuC/D at P20 (Fig.6d). Collectively, these results support the notion that some extrinsic nerve-associated SCPs from the distal colon can differentiate into enteric neurons by transitioning through an EGC-like state.

Figure 6.

GDNF-induced neurogenesis in the distal colon is quicker from SCPs than from EGCs and persists beyond the treatment period

To clarify the relationship between distal colonic SCPs and EGCs in response to GDNF treatment of Hol^Tg/Tg mice, we extended our cell lineage tracing studies by comparing the kinetic profile of GDNF-induced neurogenesis from Dhh-expressing SCPs as well as from EGCs expressing either Gfap (with tamoxifen-inducible GFAP-CreERT2 driver) ⁵² or Slc18a2 (with the Slc18a2-Cre driver) ⁵³. Like most of our other kinetic analyses, we analyzed one time point just before treatment begins (P4_[0h]), two time points during the GDNF treatment period (P4_[6h] and P6), and two time points after the end of treatment (P10 and P20) (Fig.7a). Regardless of their origin, HuC/D+ neurons in distal colon became detectable by immunofluorescence remarkably quickly, as early as 6h after the start of treatment (Fig.7b). Otherwise undetectable in this most distal colon region without exposure to GDNF, these HuC/D+ neurons began to appear either isolated or organized into very small ganglia (Fig.7c-e). At this very early stage, they were detected in all distal colon samples of GDNF-treated pups, but not in all fields of view (with 60x objective). After this P4_[6h] time point, the mean density of HuC/D+ neurons per mm² increased in a steady manner and, surprisingly, even well after the 5-day GDNF treatment period (P4-P8) ended, eventually reaching a plateau from P20 onwards (Fig.7b).

Figure 7.

When considering Cre drivers used for cell lineage tracing, our comparative kinetic analysis revealed that Dhh-expressing SCPs are among the first newly generated HuC/D+ neurons, after only 6h of treatment (Fig.7c). GDNF-induced neurogenesis became apparent much later in EGCs expressing either Slc18a2 (Fig.7d) or Gfap (Fig.7e), with first detection of YFP+ HuC/D+ neurons at P6 and P10, respectively. Before these stages, both of these Cre drivers allow to label a subset of SOX10+ EGCs with the YFP reporter (Fig.7d,e), but no neurons. On one hand, this temporal difference in neurogenesis thus supports the stepwise differentiation model (SCP → EGC → enteric neurons) described above (Figs.5-6). On the other hand, as Gfap-expressing EGCs most likely exist independently of the SCP lineage (Fig.6c), it also suggests that neuronal differentiation in distal colon is simply slower from EGCs than from SCPs. After first detection, percentage of YFP+ neurons then increases over time with all three Cre drivers, in a way roughly similar to the overall time-dependent increase in neuronal density (Fig.7b), with final contribution at P20 reaching between 27-39% of all GDNF-induced neurons (Fig.7c-e). Hence, most neurons originating from SCPs and/or EGCs (at least those labeled with our cell lineage tracing tools) are generated after the GDNF treatment period between P4-P8. During the GDNF treatment period, most of the newly induced neurons instead seem to have another origin, distinct from Dhh-expressing SCPs or Slc18a2/Gfap-expressing EGCs.

Direct transdifferentiation is the preferred mode of neurogenesis in response to GDNF, regardless of targeted ENS progenitor subtype

The rapid neuronal differentiation uncovered at P4_[6h] (Fig.7b) strongly suggested that the first GDNF-induced neurons that appear in the distal colon of Hol^Tg/Tg mice result from direct transdifferentiation, without cell division. To validate this hypothesis and more globally evaluate the contribution of such direct neuronal conversion, we conducted EdU incorporation assays in GDNF-treated Hol^Tg/Tg pups, again as a function of time and subtypes of ENS progenitors. Daily EdU treatment began at P3 and continued throughout duration of GDNF treatment period (Fig.8a). As expected, immunofluorescence staining of distal colon showed that EdU incorporation into GDNF-induced HuC/D+ neurons was extremely rare at P4_[6h] (Fig.8b), in contrast to numerous proliferative cells in their vicinity. We found only two neurons in total, from the same mouse, that stained positive for EdU at this stage. Extending this global analysis at later time points revealed a time-dependent increase in EdU+ neurons, but these always remained the minority, resulting in consistently high percentages (>67%) of direct conversion (i.e., without EdU incorporation), both during (P6) and after (P10 and P20) the GDNF treatment period (Fig.8b). Similar trends can be observed when data are analyzed based on Cre drivers used for lineage tracing (Fig.8c-e). Percentages of EdU-negative neurons in distal colon are always the highest at first detection of YFP+ neurons, with relative levels being inversely proportional to the stage at which this occurs in each mouse line: 100% at P4_[6h] for Dhh-expressing SCPs (Fig.8c), 91% at P6 for Sclc18a2-expressing EGCs (Fig.8d), and 77% at P10 for Gfap-expressing EGCs (Fig.8e). These data offer a plausible mechanism to explain the rapid generation of GDNF-induced neurons after just 6h of treatment, while also indicating that direct transdifferentiation remains the preferred mode of neurogenesis irrespective of temporal constraints.

Figure 8.

A neural crest-independent ENS progenitor subtype appears necessary to balance overall cholinergic and nitrergic differentiation of GDNF-induced neurons in the distal colon

Finally, we sought to determine the specific contribution of SCPs and EGCs to WT-like diversity of neuronal subtypes that we know GDNF can generate in the distal colon of Hol^Tg/Tg mice ²⁰. Strikingly, our analyses at P20 showed that the vast majority of YFP+ neurons derived from either Dhh-expressing SCPs or Gfap-expressing EGCs are also positive for the excitatory motor neuron marker ChAT (choline acetyltransferase), with proportions of 83% and 80%, respectively (Fig.9a,b). Much smaller percentages of YFP+ neurons were found to co-express the inhibitory motor neuron marker NOS1 (nitric oxide synthase 1), with averages of 12% for those derived from Dhh-expressing SCPs and 14% from Gfap-expressing EGCs (Fig.9a,b). Similar trends were noted with the Slc18a2-Cre driver (Fig.9c), with a slightly higher percentage of nitrergic neurons that most likely reflects, at least in part, Slc18a2 expression in a subset of these neurons (Fig.2d). When compared to each other, data for Dhh-expressing SCPs and Gfap-expressing EGCs result in cholinergic to nitrergic ratios of 6.3 : 1 on average, which is much greater than the WT-like ratio of 1.2 : 1 previously reported for all GDNF-induced HuC/D+ neurons (52% ChAT+ vs. 42% NOS1+) ²⁰. This ratio difference led us think that the additional subtype of ENS progenitors already predicted from our temporal cell lineage tracing data (Fig.7) must be more prone to differentiate into nitrergic neurons.

Figure 9.

To verify that this presumed additional progenitor subtype could be derived from NCCs but was simply missed by our scRNA-seq strategy based on G4-RFP activity (Figs.1 and 2a), we traced all NCC derivatives in the colon using the Wnt1-Cre2 driver ⁵⁴. When used in conjunction with the R26R-YFP reporter allele (Wnt1-Cre2^Tg/+;R26^YFP/+), this “pan-NCC” Cre driver normally enables YFP expression in virtually all enteric neurons and EGCs of distal colon at P20, as we confirmed ourselves (Fig.10a). However, this was not the case when the same system was used in the Holstein mutant background. To our astonishment, about a quarter of GDNF-induced HuC/D+ neurons were negative for YFP in the distal colon of these Hol^Tg/Tg;Dhh-Cre^Tg/+;R26^YFP/+ mice at P20 (Fig.10a). This ENS contribution from a non-NCC source appeared specific to the distal colon (not observed in already ganglionated proximal colon of the same mice; see insets in Fig.10a) and was strongly oriented toward neurons: less than 3% of SOX10+ EGCs were similarly negative for YFP in the same GDNF-treated animals (Fig.10a). Importantly, this atypical non-NCC-related subtype of ENS progenitors appeared ∼4X more prone to differentiate into nitrergic neurons, now yielding roughly equal proportions of NOS1+ and ChAT+ neurons (Fig.10b).

Figure 10.

Discussion

In this study, we greatly deepened our understanding of GDNF-induced neurogenesis in the context of HSCR. Just a few hours after rectal administration, GDNF triggers a robust neurogenic response in the distal colon, which then persists for several days even when GDNF is no longer administered. This GDNF-induced neurogenic program is mediated by NCAM1-FAK signaling, not by RET. Within the aganglionic distal colon, at least three distinct subtypes of ENS progenitors are likely involved (four, if we consider Gfap-expressing and Slc8a2-expressing EGCs separately), and all need to be stimulated in a coordinated way to ensure enteric neuronal diversity in well-balanced proportions. Most surprisingly, one of these progenitors seems not to derive from NCCs as virtually all ENS-related cells normally do.

RET is dispensable for GDNF-induced neurogenesis in aganglionic distal colon

GDNF-RET signaling is critical for normal ENS formation from vagal NCCs ^{55, 56}, explaining why RET variants are the main genetic risk factor for HSCR ^57–59. However, modeling this association in mice has proved difficult. The impact of Ret deficiency on the formation of the murine ENS is generally either benign at heterozygous state ⁶⁰ or too severe at homozygous state (total intestinal aganglionosis) ⁵⁶. Distal colon aganglionosis mimicking short-segment HSCR can be induced by treating hypomorphic Ret^9/- mice with mycophenolate mofetil ⁶¹, but general toxicity associated with this compound prevented us from evaluating GDNF effect in this context ²⁰. Here, we used a pharmacological approach to circumvent these problems, demonstrating that RET activity is dispensable for mediating the neurogenic effect of GDNF in the aganglionic colon of Holstein mice. This result is also consistent with our scRNA-seq data showing that Ret expression is mostly restricted to a subset of induced neurons only. In marked contrast, Ncam1 is robustly expressed in virtually all ENS progenitors and induced neurons, and pharmacological inhibition of the NCAM1-FAK signaling axis significantly impaired the neurogenic effect of GDNF.

NCAM1 was already known to influence ENS gangliogenesis before birth ⁶², and to mediate GDNF effects in the Schwann cell lineage after birth ²⁶. Our new mouse data now suggest that NCAM1 also mediates GDNF effects in postnatal EGCs as well, at least in an aganglionosis context. There is also cumulative circumstantial evidence that NCAM1 can similarly mediate GDNF signaling in human tissues. In particular, NCAM1 (also known as CD-56) has been recently reported as a reliable marker for FACS-mediated recovery of ENS cells from human gastrointestinal tissues ⁶³, including aganglionic samples obtained from children with HSCR ⁴⁷. ScRNA-seq data generated from these NCAM1+ cells revealed the presence of both SCPs (referred to as Schwann-like enteric glia) and EGCs (referred to as canonical enteric glia) in aganglionic samples ⁴⁷. These observations are also supported by our own immunofluorescence analyses of human aganglionic colon samples, which show robust NCAM1 protein expression both within (presumably in SCPs) and outside (presumably in EGCs) βIII-Tubulin+ extrinsic nerves (Fig.11).

Figure 11.

The reason why RET is present in a subset of GDNF-induced neurons is unclear. RET is normally present in a subset of adult enteric neurons as well, but a recent study based on two distinct conditional mutant mouse lines (Vip^Cre/+;Ret^Flox/Flox and tamoxifen-inducible Ret^{CreER/Flox-GFP}) suggests that it is not critical in these cells for ileal and colonic neuron survival or control of gastrointestinal motility ⁶⁴. Yet, another study previously reported that RET inactivation in Dhh-expressing SCPs can reduce the number of neurons specifically in the distal colon of adult mice ¹⁰. Given that spontaneous SCP-derived neurogenesis is not impaired (being even more robust than normal) in the colon of Ret^{51(C618F)/−} mutant pups at P5 ¹³, this suggests that RET could be necessary for long-term survival of some SCP-derived neurons once induced.

Combined with our previous and current data, these observations are nonetheless reassuring about the global applicability of GDNF-based therapy, even in presence of lower RET activity. Indeed, even if RET is necessary for the long-term survival of some newly induced neurons or for mediating the presumed beneficial effects of GDNF in non-ENS cells (e.g., in epithelial cells ^{64, 65}), it is noteworthy that HSCR-associated RET gene variants are generally predicted to reduce, but not abrogate, RET protein levels/activity ^{58, 59}. In such cases with reduced RET levels/activity, GDNF-based therapy would still be highly valuable since GDNF enemas can trigger a feedback loop that increases protein levels of both RET and endogenous GDNF in the colon wall ²⁰. This increase in endogenous GDNF is also interesting in relation to the persistence of neurogenesis after the end of GDNF treatment. Whether this is the only reason, or whether it implicates another signalling pathway (or a combination of the two) is certainly worthy of investigation in future research.

The aganglionic distal colon is rich in ENS progenitors that can be directly converted into neurons by GDNF

This study provides additional evidence that aganglionosis does not mean absence of ENS progenitors ^{66, 67}. Both mouse (this study) and human ⁴⁷ scRNA-seq data uncovered two main progenitor subtypes that have the transcriptional signature of neural crest-derived SCPs or EGCs, as we further validated using several genetic cell lineage tracing tools. Yet, determining the exact relative contribution of each of these progenitor subtypes to the pool of GDNF-induced neurons is almost impossible to determine with currently available tools. The perfect models do not exist. For instance, we previously reported that ∼10% of SCPs from aganglionic colon are not fluorescently labeled when R26R-YFP reporter allele expression depends on the Dhh-Cre driver ²⁰. Although efficacy of the GFAP-CreERT2 driver for depleting GFAP+ EGCs has been previously demonstrated during the early postnatal period ⁶⁸, its tamoxifen-inducible nature could nonetheless also potentially underestimate the neuronal contribution of these EGCs beyond the tamoxifen administration window. However, our exploratory experiments suggest that this is not the case: extending tamoxifen administration until P18 (at 2-day intervals after P8; necessary to reduce deadly toxicity of chronic tamoxifen administration) resulted in a similar contribution of 33% of GDNF-induced neurons. On the other hand, the contribution of Slc18a2-expressing EGCs is likely overestimated with the Slc18a2-Cre driver, as this transgene is known to have intrinsic activity in various neuron subtypes in the central nervous system ^69–72 and our scRNA-seq data show Slc18a2 gene expression in the neuronal cluster EN #3. Moreover, the portrait is greatly complicated by the SCP-EGC lineage connection that we have discovered. In the end, the best estimate that we can make is based on the broader Wnt1-Cre2 driver, showing that ∼75% of GDNF-induced neurons are collectively derived from NCCs.

This observation that not all GDNF-induced myenteric neurons are fluorescently labeled when using the pan-NCC driver Wnt1-Cre2 was surprising. Indeed, in our hands, the Wnt1-Cre2 driver normally turns on the R26R-YFP reporter allele in virtually all enteric neurons from the colon of WT-like control mice, and this pattern is even conserved in the “normoganglionic” proximal colon of Holstein mutant mice bearing the same tools and genetic background (FVB strain). Therefore, these data suggest that the non-NCC-derived ENS progenitor subtype that generates ∼25% of GDNF-induced neurons is uniquely present and/or responsive to GDNF in the diseased colon segment. The fact that this non-NCC-derived progenitor subtype differentiates preferentially into neurons (with increased capacity for nitrergic differentiation) rather than glial cells is also very intriguing. The quest to identify this unconventional progenitor cell population promises to be long and complex. Four alternative sources of ENS progenitors unrelated to the neural crest lineage have already been proposed over the years, including ventral neural tube ⁷³, endodermal ⁷⁴, mesodermal ⁷⁵ and placodal ⁷⁶ cell types. All of them were reported to be solicited during normal ENS development or aging. However, as mentioned above, our cell lineage tracing data with the Wnt1-Cre2 driver do not support such conclusion for the ENS-containing colon of both WT-like and Holstein mouse pups before weaning, rather indicating full contribution by NCCs. That being said, such discrepancy might be due to diverse reasons related to experimental conditions/tools/settings, and these previously suggested unconventional progenitors could nonetheless represent legitimate candidates in the aganglionic colon soon after birth. Of note, inspiration could also come from a recent preprint reporting the spontaneous generation of enteric neurons in human intestinal organoids without the need to add NCCs ⁷⁷. Clearly, there will be a lot of potential candidates (both known and not yet known) to experimentally test in future work using a framework similar to the current study.

Another intriguing discovery that was made during these studies is that direct transdifferentiation is the main mode of GDNF-induced neurogenesis in the aganglionic distal colon regardless of ENS progenitor subtypes and time point relative to treatment period. This contrasts with prior studies in WT mice in which systemic subcutaneous administration of GDNF between P0 to P30 expanded the submucosal neuron pool (but not the myenteric pool) by increasing the proliferation rate of submucosal neuronal precursors ⁷⁸, a finding also corroborated by the study of Gdnf^+/- mice ⁶⁰. Yet, we also observed an increase of such “indirect” differentiation via a dividing precursor over time, thereby most likely preventing GDNF to rapidly exhaust the pool of ENS progenitors. Such delayed response suggests that the pool of ENS progenitors is somehow protected by a quorum sensing-like mechanism that controls cellular density at tissue level, as observed in the immune system ⁷⁹. Regardless of exact mechanism, the resulting two-step response is highly desirable from a therapeutic standpoint, as it allows to combine both rapid and long-lasting effects of GDNF treatment.

Finally, our discovery that multiple progenitor subtypes are required to obtain a properly balanced ENS also has important implications for all cell transplantation-based approaches aimed at replacing missing neurons. This means that the subtype(s) of ENS stem/progenitor cells to transplant should be selected with care, as a function of the neuronal subtype to replace.

Materials and methods

Sex as biological variable

Both female and male individuals were used in all experiments involving mouse or human samples, and similar findings are reported for both sexes.

Mice

Holstein (Tg[Sox3-GFP,Tyr]HolNpln) and G4-RFP (Gata4p[5kb]-RFP) transgenic lines were as previously described in our prior work ^{29, 34}. Dhh-Cre (FVB[Cg]-Tg[Dhh-cre]1Mejr/J; JAX strain #012929), GFAP-CreERT2 (B6.Cg-Tg[GFAP-cre/ERT2]505Fmv/J; JAX strain #012849) and Wnt1-Cre2 (B6.Cg-E2f1^{Tg(Wnt1-cre)2Sor}/J; JAX strain #022501) mouse lines were purchased from the Jackson Laboratory (Bar Harbor, MM, USA). The Slc18a2-Cre (Tg[Slc18a2-Cre]OZ14Gsat/Mmucd; MMRRC strain #034814-UCD) line was obtained from the Mutant Mouse Resource and Research Centers (MMRRC) at University of California at Davis, an NIH-funded strain repository, and was donated to the MMRRC by Nathaniel Heintz (The Rockefeller University, GENSAT) and Charles Gerfen (National Institutes of Health, National Institute of Mental Health) ⁸⁰. The Cre reporter line Rosa26^FloxSTOP-YFP (B6.129X1-Gt[ROSA]26Sor^tm1(EYFP)Cos/J) was directly provided by Dr. Frank Costantini from Columbia University. All mice used in this study were maintained on the FVB/N genetic background, after at least 5 generations of backcrossing if acquired on a different background. All mice were housed and bred in individually ventilated cages within the conventional animal facility at the Université du Québec à Montréal, under a 12-hour light/dark cycle (7 AM to 7 PM) and with ad libitum access to Charles River Rodent Diet #5075 (Cargill Animal Nutrition). Genotyping was either made by visual inspection of coat color (for Holstein and G4-RFP) or standard PCR (for all Cre driver and reporter lines) using primers listed in Table S1. Euthanasia was performed using either CO₂ inhalation or decapitation (for pups aged of 10 days or less), both following isoflurane anesthesia.

Administration of substances

GDNF enemas (10µl at 1mg/ml in PBS) were administered once daily between P4-P8, using a 24-gauge gavage needle (Fine Science Tools, Canada) connected to a micropipette, as previously described ^{20, 21}. Briefly, gavage needle tip was gently inserted into the rectum just beyond the anus, and enema was carefully injected over a few seconds. Two batches of human recombinant GDNF were used, both standardly produced in bacteria. Research-grade GDNF purchased from PeproTech (cat. #450-10, Cranbury, NJ) was used for single-cell RNA-seq experiments, while clinical-grade GDNF kindly provided by Vivifi Biotech (London, United-Kingdom) was used for all other experiments. When required, tamoxifen (20µl at 4mg/ml in corn oil; Sigma Cat.#10540-29-1) and/or EdU (10 µl at 10mM in PBS; Thermo Fisher Cat.# C10337) were administered via daily intraperitoneal injections in parallel to GDNF enemas, but starting a day earlier (P3-P8). Phospho-FAK[Y397] and phospho-RET[Y1062] inhibitors, PF-562271 (MedChemExpress Cat.# HY-10459) and Pralsetinib/BLU-667 (MedChemExpress Cat.# HY-112301), respectively, were similarly administered via daily intraperitoneal injections using same 30mg/kg concentration (in a vehicle solution of 2% DMSO; 25% PEG 300; 5% Tween 80; 68% Saline), but in this case only during GDNF treatment period (P4-P8).

Human Tissues

Human colon samples were obtained from three patients (2 males, 1 female) diagnosed with short-segment HSCR who underwent surgical resection of their aganglionic colon segment. These patients were recruited at the Centre hospitalier universitaire Sainte-Justine and their ages ranged between 3-6 months at time of surgery. Informed consent for collection and use of these tissues was obtained from all parents or legal guardian. Following surgical resection, full-thickness colon samples of aganglionic segment were immediately placed in ice-cold, oxygenated Krebs solution and promptly transported to UQAM research laboratory for further processing.

Single-cell RNA-seq

Samples of the distal third of the colon (including part of transition zone) were dissected from P10 Hol^Tg/Tg;G4-RFP double-transgenic pups that were previously administered GDNF enemas or not between P4-P8. For each condition, colon samples were collected from 6 to 8 pups and pooled on day of dissection. Once pooled, each piece of distal colon was cut opened along longitudinal axis, washed extensively in ice-cold PBS, and mechanically dissociated during 5 minutes using a scalpel blade in a small volume of pre-warmed (37°C) EMEM (Eagle’s Minimum Essential Medium) containing collagenase (0.4 mg/ml; Sigma-Aldrich C2674), dispase II (1.3 mg/ml; Life Technologies 17105-041), and DNAse I (0.5 mg/ml; Sigma-Aldrich DN25). Resulting grossly dissociated tissue preparation was transferred in an Eppendorf tube, into which the above dissociation medium was added to reach ∼1 ml total volume, and the tube was then placed in an agitating heating block at 37°C for 30 minutes at 300rpm. Final tissue preparation was filtered through a 40µm cell strainer and centrifuged at 1200 rpm for 5 minutes at 4°C. Pelleted cells were resuspended in 500 µl of sterile PBS also containing 0.04% non-acetylated BSA, stained with CellTracker™ Green CMFDA Dye (Invitrogen, Cat.#C2925) diluted at 1/10000, and separated in a BD FACSJazz cell sorter (BD Biosciences) to recover viable CMFDA+ RFP+ cells, under most permissive Yield mode parameters (Yield Mask = 32 and Purity Mask = 0). Single-cell RNA-seq libraries were prepared following the guidelines of the 10x Genomics protocol. Brefly, single-cell Gel Bead-In-Emulsions (GEMs) were generated by combining suspended cells, barcoded gel beads, and master mix onto the Chromium Next GEM Chip G, which was then processed on the Chromium iX controller Instrument. After breaking the GEMs, barcoded cDNAs were purified and amplified according to desired cell recovery target. Subsequently, the cDNA underwent fragmentation, A-tailing, and ligation with adaptors. For sample indexing, additional PCR cycles were conducted based on cDNA input quantified during QC step. Resulting libraries were finally subjected to 100bp paired-end sequencing on an Illumina NovaSeq 6000 sequencer.

Single-cell RNA-seq data analysis

Single-cell RNA-seq data were analyzed using Trailmaker (Parse Biosciences). After alignment to the reference genome (mm10) and transcript quantification, quality filtering was applied to exclude low-quality cells. The initial dataset contained 10,018 to 9,477 barcodes, depending on the experimental conditions, with a total of 22,142 to 22,778 detected genes. A mitochondrial content filter was applied, setting a maximum threshold of 10% mitochondrial reads with a bin step of 0.30. After filtering, the number of retained cells decreased from 10,018 to 3,997 in the untreated Hol^Tg/Tg group (-60.1%) and from 9,477 to 7,091 in the GDNF-treated Hol^Tg/Tg group (-25.2%). Data normalization and batch effect correction were performed using the integrated Trailmaker tools. Dimensionality reduction was conducted via principal component analysis (PCA), and cell clustering was performed with the Leiden algorithm. Cell populations were visualized using uniform manifold approximation and projection (UMAP). Cluster annotation was performed by comparing differentially expressed marker genes with reference databases (PanglaoDB, CellMarker). Differential gene expression analysis was carried out using the Wilcoxon test implemented in Trailmaker. Cell trajectory reconstruction was conducted using Trailmaker’s pseudotime tools.

Tissue processing, labelling and imaging

Mouse colon tissues to be immunostained were cut longitudinally along the mesentery, washed in PBS, pinned onto Sylgard-coated (Thermo Fisher Scientific Cat. #50822180) petri dishes, fixed with 4% PFA at 4°C overnight, and finally further microdissected to separate muscle and mucosa layers. Whole muscle strips of the most distal quarter of the colon were then used for immunofluorescence staining, which was performed as previously described ²⁰. For human colon tissues, immunofluorescence staining was performed on full-thickness aganglionic samples using a previously described protocol for 3-dimensional imaging ⁸¹. All antibodies used for immunostaining are listed in Table S2. EdU detection in mouse samples was performed using Invitrogen Click-iT EdU Imaging Kit (Thermo Fisher Scientific Cat. #C10337), in accordance with manufacturer’s instructions. All images were acquired on a Nikon A1 confocal microscope, using either Plan Fluor 20x/0.75 MImm or Plan Apo λ 60x/1.40 objectives (Nikon A1R, Melville, NY). For each biological replicate, 3 to 11 representative z-stack images of myenteric plexus were acquired, and then analyzed using the ImageJ software (National Institutes of Health, Bethesda, MD).

Western blot analysis

Remaining part of colon samples not used for immunofluorescence (i.e., proximal and mid segments) were homogenized in RIPA buffer containing protease and phosphatase inhibitors and centrifuged (14000rpm for 15min at 4C) to sediment debris. Protein concentration in supernatant was determined using DC protein assay (Bio-Rad Kit I #5000111) according to manufacturer’s guidelines. For western blot analysis, 50 µg of total protein extracts were separated on a 10% SDS-PAGE and then transferred onto an Immun-Blot PVDF membrane (Bio-Rad Cat. #1620177). Immunoreactive bands were revealed using Immobilon Western Chemiluminescent HRP Substrate (MilliporeSigma Cat. #WBKLS0050) and visualized using the Fusion FX imaging system (Vilber, Marne-la-Vallée, France). All antibodies used are detailed in Table S2.

Data statistics

Statistical analyses were performed using GraphPad Prism v9.5.1. As indicated in Figure legends, either Student’s t-tests or one-way ANOVA with post-hoc Tukey’s test were used to calculate statistical significance, with P values < 0.05 considered as indicative of significance. All experiments included a minimum of 3 biological replicates.

Study approval

All experiments involving mice were approved by the Université du Québec à Montréal ethics committee (CIPA #959) and conducted in accordance with the Canadian Council on Animal Care (CCAC) guidelines. Use of human samples in our experiments strictly adhered to guidelines and approval of human research ethics committee of the Centre Hospitalier Universitaire Sainte-Justine (CER protocol #4172).

Data availability

All relevant data are included in the article. Raw files of single-cell transcript profiling can be accessed on the Gene Expression Omnibus database (series GSE313116).

Acknowledgements

The authors thank Gregoire Bonnamour from the Cellular analyses and Imaging core (CERMO-FC, UQAM) for assistance with flow cytometry and confocal imaging, and Farzaneh Ramdani from the Bioinformatics core (CERMO-FC, UQAM) for help with scRNA-seq data processing.

Additional information

Author contributions

Conceived and Supervised the study: RS and NP. Designed the experiments: AG, RS and NP. Contributed study materials: AA and CF. Performed the experiments: AG, MAL, NL and RS. Analyzed the data: AG, RS and NP. Wrote the paper: AG (original draft), RS (original draft and review), and NP (review and editing). All authors read and approved the final manuscript.

Funding

This study was funded by a grant from the Canadian Institutes of Health Research (CIHR grant # PJT-180290) to NP.

Funding

Canadian Institutes of Health Research (CIHR) (PJT-180290)

• Nicolas Pilon

Additional files

Supplemental material. Table S1. Specific primers used for genotyping. Table S2. List of primary antibodies used in this study.