Developmental organization of sensory and sympathetic ganglia

Keng Ioi Vong; Yanina D Alvarez; Qingquan Zhang; Jiaming Weng; Geoffroy Noel; Scott T Barton; Changuk Chung; Robyn Howarth; Naomi Meave; Fiza Jiwani; Sai B Patarlapalli; Fenyong Yao; Fugui Zhu; Chelsea Barrows; Arzoo Patel; Jian Xiong Wang; Neil Chi; Stephen F Kingsmore; Melanie D White; Xiaoxu Yang; Joseph G Gleeson

doi:10.1038/s41586-026-10313-0

. Author manuscript; available in PMC: 2026 Apr 23.

Published in final edited form as: Nature. 2026 Apr 1;654(8119):734–743. doi: 10.1038/s41586-026-10313-0

Developmental organization of sensory and sympathetic ganglia

Keng Ioi Vong ^1,^2,¹⁵, Yanina D Alvarez ³, Qingquan Zhang ⁴, Jiaming Weng ^1,^2,⁵, Geoffroy Noel ⁶, Scott T Barton ⁶, Changuk Chung ^1,^2,⁷, Robyn Howarth ^1,², Naomi Meave ^1,², Fiza Jiwani ^1,², Sai B Patarlapalli ^8,^9,¹⁰, Fenyong Yao ⁴, Fugui Zhu ⁴, Chelsea Barrows ^1,², Arzoo Patel ^1,², Jian Xiong Wang ³, Neil Chi ^4,^11,^12,¹³, Stephen F Kingsmore ², Melanie D White ^3,¹⁴, Xiaoxu Yang ^8,^9,^10,^15,^*, Joseph G Gleeson ^1,^2,^*

PMCID: PMC13102452 NIHMSID: NIHMS2162736 PMID: 41922758

Abstract

The neural crest generates a broad spectrum of cell types that migrate across the body plan to populate multiple tissues¹. However, the relationship between lineages of neural crest derivatives remains unclear, and the extent to which neural crest cells delaminated from the neural tube have specified fates remains debated. Here, leveraging CRISPR barcoding in mice and mosaic variant barcode analysis in humans, we demonstrate robust bilateral progenitor clonal spread of neural crest progenitors along the rostrocaudal axis but limited clonal overlap between sensory and sympathetic lineages. Computational modelling of mosaic variants suggests that most neural crest cells show strong fate restriction before delamination. Real-time imaging of quail embryos further shows a fibroblast-growth-factor-dependent rostrocaudal dispersion of neural crest cells across multiple axial levels. These findings support a model in which neural crest fate bias predominantly emerges within the neural tube, with only a minor subset of delaminated progenitors retaining multipotency to generate both sensory and sympathetic derivatives.

The principle of cell fate decision has been a central question of developmental biology. Waddington’s ‘landscape model’ predicts decisions resulting from largely irreversible processes whereby multipotent progenitors progressively lose potency to commit to fate, akin to a ball rolling down branching valleys. Cell fate decisions often involve two stages: an early phase where progenitors become more likely to adopt certain fate(s), which are at least partially reversible when exposed to non-native environments; and a late phase where cell fate decisions are predominantly irreversible even in non-native environments¹.

The neural crest (NC) gives rise to diverse cell types critical for the vertebrate body plan, including peripheral nervous system ganglia, craniofacial skeleton, and smooth muscle, among others, creating challenges in understanding mechanisms of cell fate decisions². The early emergence, rapid expansion, and mobility of NC cells during gastrulation have presented hurdles for precise fate mapping^3-5. Some have suggested that the NC does not entirely conform to Waddington’s model, because cells retain plasticity during migration, and cell fate decisions can be influenced by extrinsic signals like morphogens^6,7. Conventional NC lineage tracing in avian embryos has led to conflicting models with regard to whether NC cells delaminating from the neural tube are multipotent or are fate-specified^8-19. Several studies, starting with the pioneering work of James Weston and Nicole Le Douarin in chick-quail chimeras, established NC origins, and suggested the majority are fate-restricted^8,20, as have some recent studies^13,21,22. However, others suggested delaminated cells are multipotent, capable of contributing both sensory (dorsal root ganglia, DRG) and sympathetic ganglia (SG)^16-18.

During nearly each cell cycle, cells acquire somatic mosaic variants (MVs) that are faithfully inherited by daughter cells as ‘barcodes’^23,24. Rapid proliferation during embryogenesis creates clones carrying mosaic variants, which, when assessed in cellular pools, reveal varying allelic fractions (AFs) that reflect clonal populations. Analyzing these MVs as tracing barcodes to reconstruct clonal relationships, an approach called Mosaic variant barcode analysis (MVBA), has proven powerful in deconvolving lineages in the brain and hematopoietic system in humans^25-27. Here, leveraging MVBA and live cell imaging to infer clonal dynamics, we seek to resolve two central questions in NC cell fate decision: 1. Are trunk NC progenitors already fate-specified prior to delamination with regard to contributing to DRGs or SGs? 2. How are the different DRG and SG cells across the body axes clonally related?

Applying MVBA to mice using CRISPR/Cas9, and humans using naturally occurring MVs (summarized in Fig. 1a), complemented by live imaging in quail embryos, here we perform an unbiased assessment of clonal relationships among DRGs and SGs, spanning cervical to lumbar levels. In mice, we found that delaminated NC progenitors predominantly give rise to a single DRG or SG lineage, suggesting strong fate bias prior to delamination. Single-cell transcriptomic analysis further suggests emergence of fate bias in pre-delaminated NC as early as embryonic day (E) 8.5. Next, by evaluating three neurotypical human donors, our data indicate that most NC progenitors are fate-specified prior to delamination. Moreover, the DRG and SG from the same or neighboring levels were often phylogenetically distinct, while ganglia belonging to the same type were often clonally related even across multiple spinal levels. Cell movements that conformed to these findings were documented in real-time by imaging quail embryos. Furthermore, we demonstrate that the rostrocaudal clonal spread depends on FGF signaling gradient. Together, our study contrasts the prevailing model positing that delaminated NC cells are largely multipotent.

Delaminating NC progenitors produce a single lineage in mouse

To reconstruct clonal relationships among trunk NC derivatives, we utilized the Homing CRISPR barcoding mouse²⁸ with Cas9 under control of Sox10-Cre, chosen because expression initiates first in NC cells delaminating from the neural tube²⁹. Individual DRG and SG were isolated from 1-month-old mice, followed by targeted amplification of barcode loci. We reconstructed the phylogenetic tree of DRG and SG based on the percentage of identical edits in each ganglion, considering multiple barcode loci. Interestingly, SG and DRG largely clustered independently (Fig. 1b). For example, the largest clade consisted of 27 SG without a single DRG. Analysis of individual loci indicated that Cas9 edits were predominantly distributed to either DRGs or SGs but not shared. This implied that the fate of Sox10-Cre-labelled delaminated NC cells was mostly restricted to a single ganglion type (Extended Data Fig. 1a). To orthogonally verify these findings, we analyzed clonal lineages of DRG and SG using an independent CRISPR barcoding mouse model, DARLIN³⁰, under control of a doxycycline-inducible Cas9 allele. Doxycycline was injected at embryonic day (E) 8.5, followed by ganglia dissection at E14.5. Hierarchical clustering of clonal similarities confirmed shared lineage relationships between the same type of ganglia along the rostrocaudal axis, without clear left-right bias (Extended Data Fig. 1b, c). These results are seemingly at odds with models suggesting that delaminated NC progenitors are largely multipotent and can give rise to cells both within DRG and SG, and instead suggest they are largely clonally independent.

Single-cell analysis confirms mouse NC fate bias

To further examine the possibility of NC fate specification prior to delamination, we analyzed the single-cell transcriptome of Wnt1-Cre marked NC cells isolated from E8.5, E9.0 or E9.5 mouse embryos (Extended Data Fig. 2a). Transcription factors (TFs) suggested to drive sensory fate include Neurog1, Neurog2, Neurod1 and Pou4f1¹¹. By E8.5 and E9.0, many NC progenitors already co-expressed at least 2 of these TFs. Remarkably, cells expressing at least two sensory TFs were more commonly Zic1+, a marker for NC progenitors prior to delamination (pre-EMT NC), rather than Sox10+, a marker for delaminating NC (Extended Data Fig. 2b-d). Likewise, a substantial fraction of pre-EMT NC already expressed sympathetic TFs (Extended Data Fig. 2e-f). Furthermore, RNA velocity analysis suggested distinct TFs showed significant sensory or sympathetic fate-biased expression dynamics between E8.5 and E9.5 (Extended Data Fig. 3). These findings suggest that NC cells exhibit fate bias as early as E8.5, prior to delamination.

Identification of somatic MVs in human DRG and SG

To compare results in humans, we performed MVBA analysis using naturally occurring somatic MVs from two male and one female donors (ID06, ID07, ID08, respectively) (Fig. 1c, d). We isolated 94 DRGs and 93 SGs in total across bilateral cervical, thoracic, and lumbar levels. To distinguish somatic MVs from germline variants, we also profiled peripheral organs including the brain, heart, liver, kidneys, and skin. A subset of ganglia and organ biopsies were subjected to 30x or 300x whole-genome sequencing respectively, followed by state-of-the-art MV calling and filtering to identify candidate MVs (Extended Data Fig. 4a, Methods). We conducted ultra-deep massive parallel amplicon sequencing (MPAS), to both validate all candidate MVs and quantify allele fractions (AF). We identified >1,200 bona fide MVs and measured AF at >500× coverage (Fig. 1e-g, Extended Data Fig. 4b, c).

To estimate the percent of ganglia DNA derived from NC, we performed single-nucleus RNA sequencing. Analysis of ganglia from donor ID08 suggested that trunk NC derivatives accounted for approximately 80% of sequenced cells in DRG or SG (Extended Data Fig. 5). Notably, more than 30% of MVs showed AF differences of at least 1.5-fold (i.e. 50%) between DRGs and SGs of donors ID06 and ID07 (30.8% and 37.4% respectively, Fig. 1e-g, Methods). This proportion was slightly lower in ID08, likely because fewer DRGs were isolated due to variable sample quality resulting from longer post-mortem intervals (see Methods). Nevertheless, these observations confirmed that MVBA can be used to delineate clonal relationships of paraspinal ganglia.

Cells within DRG and SG are clonally distinct in humans

To visualize the distribution of clonal MVs across the body, we mapped the AF of each MV onto a schematic body plan (Fig. 2a-d), referred to as a ‘geoclone’. Consistent with the findings that more than 30% of MVs were enriched in either type of ganglia, there were numerous examples of MVs shared between a single type of ganglion across different axial levels but not between DRG and SG of proximal levels. For example, MV chr3:86467504-TC-T of donor ID06 was enriched in almost all DRGs spanning across C2 to T8 levels (Fig. 2b). Another MV chr7:47802159-T-A of donor ID07 was distributed bilaterally in SG across up to 12 spinal levels apart (Fig. 2c). However, these MVs were often not found in both DRG and SG of the same level, consistent with our findings that approximately one-third of MVs were enriched (i.e. at least 1.5-fold difference) or exclusive in either type of ganglia (Fig. 1e-g).

Next, to evaluate whether these examples reflected the overall spatial organization of clonal MVs, we calculated the individual variance in AF between ganglia along the rostrocaudal axis (i.e. between the ganglia of same type on each side) and the dorsoventral axis (i.e. between all ganglia of a spinal level). We only considered the thoracic levels, where DRG and SG develop in equivalent pairs. Because there are 12 DRGs and SGs on each side but only 4 ganglia at each spinal level, when quantifying variance at the dorsoventral axis we used a “rolling-level average” approach that included ganglia from every three spinal levels (Extended Data Fig. 6a, Methods). In all three donors, the overall variance in MV distributions between DRG and SG along the dorsoventral axis was significantly larger than that between spinal levels along the rostrocaudal axis (Extended Data Fig. 6b-d). These findings suggest that MVs were often more restricted to ganglia of the same type across different axial levels, while DRG and SG from the same or proximal levels were more clonally distinct.

To assess overall lineage relationships between all sampled ganglia, we performed hierarchical clustering displaying AFs of all MVs in a phylogenetic dendrogram (Fig. 3a, c, e). For all three human donors, DRGs were predominantly clustered separately from SGs. Therefore, DRGs across different levels were clonally more similar among themselves, as were SGs. In contrast, DRGs and SGs were often phylogenetically distant from each other. Notably, clonal relationships of ganglia lacked observable constraints along the rostrocaudal axis, nor were they segregated between the left and right sides. In addition, we performed bootstrap analysis to assess the statistical robustness of the dendrogram clustering pattern (Fig. 3a-f, See also Supplementary Fig. 2). Almost all (16/18) were significantly stable, indicating that the closest lineage relationships, consisted of pairs of either DRG or SG that were 2-6 axial levels apart. In contrast, no statistically significant stable branches were formed by any pair of DRGs and SGs from the same level. These results suggested that DRG and SG from the same level represent predominantly clonally independent lineages derived from divergent ancestral populations.

Fig. 3. ∣ — (a, c, e) Hierarchical clustering heatmaps with Manhattan distances using square-root transformed AFs of MVs from ID06 (a), ID07 (c), or ID08 (e). Note that DRGs and SGs predominantly clustered separately, whereas the left and right tend to intermix together, suggesting the lineage relationship between ganglia was not driven by their anatomical position but rather by their identity (DRG or SG). Statistical confidence for the dendrogram was tested with 10,000 bootstrapping analyses. Colored branches indicate pair of statistically stable ganglia with approximated unbiased p-value >95%.

(b, d, f) Ganglia from statistically stable branches mapped to anatomical body plan of ID06 (b), ID07 (d), or ID08 (f) with corresponding colors. Note most branches consist of same type of ganglia at different axial levels, with a few examples of different types of ganglia from adjacent levels. No branches consist of DRG and SG from the same level. C: cervical, T: thoracic, L: lumbar.

Human NC cell fate specification precedes delamination

The phylogenetic relationships implied that DRG and SG are derived from distinct progenitor pools, raising the possibility that NC progenitors have fate bias within the human neural tube prior to delamination. Given that delaminated NC cells rarely revert their decision on migrating to either side³¹, MVs shared bilaterally likely arose prior to delamination. If delaminated NC cells give rise to only one cell type, the MVs acquired before delamination should distribute to both sides but specific to either DRG or SG, while MVs acquired after delamination should distribute to only one (or very few) ganglion of only one side, i.e. to either DRG or SG (Fig. 4a). In contrast, if delaminated NC cells retain multipotency to generate both cell types, MVs acquired before delamination should be shared bilaterally across several ganglia from the same level irrespective of type, while those acquired after delamination should be shared between DRGs and SGs of the same side (Fig. 4b). A third possibility could be that both scenarios exist at the same time in a continuum with progressive fate restriction.

Fig. 4. ∣ — (a-b) Modeling two possible scenarios whereby clones are more similar among Left/Right (L/R) than DRG/SG (scenario 1, a) or the opposite (scenario 2, b). Axes: normalized AF difference of each MV between the DRG/SG (y-axis) vs. L/R (x-axis).

(c-d) Normalized AF difference of each MV observed from ID07 (c) and ID08 (d) between DRG/SG (y-axis) vs. L/R (x-axis). Green dots: individual MVs. Blue contour: 2D Kernel density MV estimation plots highlighted. Grey curves: kernel density estimation along the x- or y-axis. The mostly vertical distribution suggests greater AF difference between DRG/SG than L/R.

(e) Effect of founder population size on AF of MVs. Cells acquire MVs (purple dots) after the L/R split, distributing to one side exclusively; or after DRG/SG fate specification, thus found in one type of ganglion. AFs quantified under three founder population cell numbers scenarios: n=10, n=100, or n>>100 shown. For cells specific to one group, a smaller number of cells immediately after L/R split, or after DRG/SG specification, correlate with higher AFs difference.

(f-g) Estimated minimum size of the founder population at DRG/SG vs. L/R split in cervical (f) or thoracic regions (g), suggesting NC cell fate specification occurs before L/R split. P-values: two-tailed Mann-Whitney U tests. Y-axis: log₁₀-transformed estimated cell population size.

(h) Effect of founder population size AFs. Green: MVs acquired during early embryogenesis before the L/R vs. DRG/SG split but with varying AFs (green dots). A larger hypothetical founder population size correlated with a smaller AF difference.

(i-j) Estimated maximum size of founder population at DRG/DG vs. L/R split in cervical (i) or thoracic regions (j), suggesting NC cell fate specification occurs before L/R split. Y-axis: log₁₀-transformed estimated cell population size.

To test these scenarios, we compared the AF difference for each MV between the left and right side against that between the DRG and SG in corresponding axial levels. Intriguingly, most MVs exhibited fewer AF differences between left and right while simultaneously showing higher AF variation between DRGs and SGs (Fig. 4c, d, Extended Data Fig. 6e, f). While many MVs were distributed bilaterally, they were often exclusive to either DRGs or SGs. These findings suggest NC progenitors are fate-specified before delamination and commitment to either side.

Because cell number increases steadily during embryogenesis, we estimated whether NC fate specification or delamination occurs first by simulating the relative cell population size when NC progenitors are first fate-specified and first delaminated. Considering MVs that were fully lateralized or exclusive to either type of ganglion (‘group-specific’ MV), and assuming each MV was nascently acquired in a single progenitor, the observed AF for any group-specific MV can predict the minimum possible cell number at the time of group separation, i.e., left-right commitment or DRG versus SG cell fate restriction (Fig. 4e). Thus, if the MV arises when the cell population is relatively small, a higher group-specific AF will result. We simulated the minimum possible population size at the point of group separation in all 3 donors. For donors ID07 and ID08, the cell population at the time of DRG or SG fate commitment was significantly smaller than that at delamination to either side, while the size difference for donor ID06 was relatively modest (Fig. 4f). This trend was consistent across cervical and thoracic levels (Fig. 4g). These findings suggest multipotency of delaminated NC cells is not a predominant mechanism of NC development.

In addition, we performed a stepwise simulation to estimate the cell number of common progenitor for different sample groups (i.e. DRG/SG or left/right) based on a previously described hypergeometric distribution model we previously published^26,32 (Methods). The difference in AF for shared MVs should correlate with the maximum possible cell number at the time the groups separate, with larger AF differences estimating smaller population sizes at the time of separation (Fig. 4h, Methods). We performed a stepwise simulation for all the shared MVs using previously published mathematical methods (Extended Data Fig. 6g, h). The maximum possible cell number at NC fate specification was significantly lower than that at left-right commitment (Fig. 4i, j). The results further suggest that most delaminated NC cells are fate-specified and give rise to either DRGs or SGs.

Single-cell analysis confirms human DRG and SG clonality

As we performed MVBA analysis from bulk tissue biopsies, we asked whether similar clonal relationships can be recapitulated at the single cell level. Thus, we performed single-nucleus simultaneous genomic and transcriptomic amplification from the same cell using the ResolveOME workflow^27,33, followed by genotyping for MVs with single-nucleus massive parallel amplicon sequencing (snMPAS) and cell type identification with single-nucleus RNA sequencing (Fig. 5a, Extended Data Fig. 7a-c). We profiled a total of 224 human single nuclei from both types of ganglia at the T2 and T3 levels of the left side. Due to suboptimal RNA quality likely resulting from post-mortem intervals, we could annotate cell type for only 75 out of the 224 nuclei, using expression of pan-neuronal or glial marker genes (55 neurons and 20 glia) (Fig. 5a).

To deconvolve lineage relationships between isolated single cells, we reconstructed a phylogenetic tree using MVs genotyped for all 224 cells, where the terminal branches of the tree indicated the most closely related pair of cells. More than half of the terminal branches contained pairs of cells from different levels (Extended Data Fig. 7c). We performed a permutation test by randomly shuffling the labels of the 224 nuclei and computing the probabilities of different terminal branch combinations. The terminal branches with pairs of cells from the SG of T3 (T3-SG: T3-SG) were significantly more common than random (Extended Data Fig. 7d, e), suggesting frequent local clonal expansion within the ganglion.

Since more than 80% of ganglionic cells were neurons or glial cells derived from NC (Extended Data Fig. 5), we asked whether the clonal relationships were driven by a particular cell type. We performed hierarchical clustering using MV genotyped by snMPAS in the 75 cells assigned as either neurons or glia. Cells from DRG and SG, regardless of cell type, clustered independently in the dendrograms (Fig. 5b, Extended Data Fig. 8a, b). Half of all 28 terminal branches in the phylogenetic tree were shared by nuclei within a single ganglion, again implying local clonal expansion during gangliogenesis (Fig. 5c). Remarkably, 10 of the 14 remaining branches comprised cells from different axial levels, suggesting cells from ganglia of T2 and T3 levels more frequently shared the closest clonal relationships, implying that the clonal relationship between ganglia was organized across different axial levels rather than between DRG and SG (Fig. 5c). We performed a permutation test for the statistical significance of the terminal branch composition. As expected, the occurrence of the T3-SG: T3-SG pair, indicative of local proliferation in the SG, was significantly more frequent than random (p=0.0011) (Extended Data Fig. 8c). Interestingly, the branches shared by DRGs and SGs from the same level (T3-DRG: T3-SG) were significantly less prevalent than random (p=0.0427) (Extended Data Fig. 8d), further suggesting clonal relationships were organized rostrocaudally across different axial levels.

FGF-dependent NC migration along the rostrocaudal axis

The shared clonal relationship between ganglia along the rostrocaudal axis might imply progenitors migrate along the neural tube and disseminate clones to different spinal levels. To directly observe NC progenitor dynamics during embryogenesis, we performed real-time imaging of quail embryos electroporated with NC-specific Pax7 enhancer reporter (Extended Data Fig. 9a), first expressed in cells fated to become NC at the neural plate border⁵. Tracking of cell migratory paths within the 6 hours of imaging revealed 21.2% of NC progenitors migrated along the rostrocaudal axis across more than one somite level (Extended Data Fig. 9b-c and Supplementary video 1). Because numerous clones were distributed bilaterally, we asked if NC progenitors cross the midline to disseminate clones to both sides. To test this possibility, we electroporated the NC-specific reporter to one side of the quail embryo neural tube. We observed around 20% of labeled cells migrated across the midline prior to delamination, taking up contralateral positions (Extended Data Fig. 9d-f).

To identify potential signaling pathways that guide the rostrocaudal migration of NC progenitors, we first examined the expression of receptors for major signaling molecules in mouse NC cells. FGF receptors were enriched in pre-EMT and delaminating NC progenitors (Extended Data Fig. 9g). FGF signaling blockade accelerates the delamination of trunk NC progenitors³⁴. To test whether FGF signaling is required for rostrocaudal migration, we incubated quail embryos with the pan-FGF receptor inhibitor Infigratinib. Cells that migrated along the rostrocaudal axis by more than 1-somite distance dropped from 19.35% in DMSO control to 1.35% following inhibition of FGF signaling (Fig. 5d, Supplementary videos 2 and 3). While total migration distance did not differ, NC cells showed significantly reduced migration along the rostrocaudal axis and significantly increased displacement along the mediolateral axis (Fig. 5e-h). The findings suggest that rostrocaudal migration of NC cells depends on FGF signaling and that the NC cells tend to delaminate from the neural tube in the absence of a FGF signaling gradient.

Clonal spread of NC cells during rostrocaudal migration

Our data from human MV analysis did not exclude the possibility that NC progenitors retain multipotency during rostrocaudal migration. We postulated two models representing extreme ends: 1) NC cells specify to either DRG or SG lineages during times of rostrocaudal migration, predominantly disseminating clones across different levels but not across DRG and SG dorsoventrally (Extended Data Fig. 10a, left); 2) NC cells have the potency to generate both DRG and SG during rostrocaudal migration (Extended Data Fig. 10a, right). While these two scenarios are mutually exclusive for one progenitor, they may co-exist within an organism at the population level. Thus, to evaluate the prevalence of either model, we plotted the normalized AF difference for each MV between the rostral (T1-T6) and caudal (T7-T12), versus between the DRGs and SGs, using contour graphs for the results of all MVs. For the majority of MVs, the AF differed substantially between DRG and SG but only subtly between rostral and caudal levels, resulting in a flat contour shape (Extended Data Fig. 10b, c). Together, the results supported the first model, whereby most NC cells are specified to either DRG or SG lineages during rostrocaudal migration (Extended Data Fig. 10a, b).

Discussion

Here we performed a comprehensive large-scale analysis of trunk NC lineage relationships in mice and humans, complemented by live imaging in quail embryos, revealing previously unrecognized aspects of NC development: i) NC cell fate specification occurs predominantly before delamination; ii) Committed NC cells can traverse the midline before delamination and migrate robustly across multiple spinal levels and; iii) Fate-specified NC cells can migrate rostrocaudally in the neural tube, thereby disseminating clones bilaterally but often specifically to either DRGs or SGs of multiple spinal levels (Extended Data Fig. 10d, e). These observations revise the current concept of NC development.

The pivotal question of whether delaminated NC cells are inherently multipotent or rather represent a heterogeneous pool of progenitors with restricted fates has been debated for decades ^{13,14,16,20,21,35,36}. The methodologies used in the prior studies were mostly limited to dye-labeling or viral tracing, which may be confounded by rapid cell divisions or labeling of cell mixtures in the embryo. Other studies that might be more definitive include multicolor fluorescent Confetti reporters, but may be limited by analysis within a single somite level or may have missed rostrocaudal cell movements we observed herein. Such potential technical challenges in prior studies were at least partly overcome in our study by MVBA in humans leveraging spontaneously arising MVs that spread naturally by progenitor expansion and migration during development, and which are thought to be static throughout life.

Using MVBA, we performed an unbiased analysis of clonal relationships among ganglia from the cervical to lumbar levels. Several independent lines of evidence suggest the majority of NC cells are already fate-specified prior to delamination: i) MVs were seldom shared between proximally located DRG and SG, both in humans and mice; ii) DRG and SGs were more clonally similar within-group than between-group; iii) Cell population size was significantly smaller when cell fate splits between DRG and SG than when NC cells delaminate. Single-cell sequencing experiments in mice further suggest a strong sensory or sympathetic lineage bias, which accounts for a large degree of clonal separation between DRG and SG observed in the CRISPR barcode mouse models.

MVBA circumvents the need for transgenic manipulation and allows reconstruction of clonal dynamics in humans. Analyzing lineage trajectories during human embryonic development is not feasible. MVBA allows reconstruction of early developmental events by analyzing late-stage donor tissues. Due to the postmortem interval of human cadavers, single-nucleus transcriptional profiling was only possible in a minority of nuclei assessed. While our results suggest NC cell fate specification predominantly occurs within the neural tube, single-nucleus sequencing suggests a minor population of delaminated NC progenitors are multipotent in humans. Thus, we confirm their presence as described in previous publications¹⁶.

However, because we leverage naturally occurring somatic MVs, it is almost impossible to pinpoint the exact timing when each MV arises. Similarly, while previous studies demonstrated Sox10-Cre mediates recombination in delaminating NC¹⁶, we cannot rule out the possibility that a fraction of the Cre-mediated Cas9 edits arise later in development. Nonetheless, the bilateral distribution of MVs likely implies their emergence before delamination. Correspondingly, hierarchical clustering using AFs of MVs in both humans and mice shows no significant lateralization, suggesting most analyzed MVs occur before or during delamination.

While the classical Waddington’s model suggests development as a ball rolling down a hill with branching valleys, our findings may imply the landscape for NC development more resemble successive valleys where progenitors undergo developmental decisions toward DRG vs. SG fate, and subsequently delaminate to the left versus right. The model takes stochasticity into account, thus potentially accounting for a fraction of progenitors generating both DRG and SG after delamination.

Rostrocaudal migration of trunk NC cells was thought to be possible only within a narrow temporal and spatial window, primarily reserved for cells that reached the ganglia³⁷. Delaminated trunk NC cells were observed to migrate exclusively through the rostral but not the caudal half of each sclerotome. Only after SG cells reach the ventral edge of the sclerotome, near the dorsal aorta, can they extend interganglionic filopodia, forming a contiguous narrow stream spanning up to two segments rostrally or caudally from their axial origins³⁸. Cells destined for DRG were not expected to migrate ventrally past the sclerotome and thus were not thought to migrate across axial levels^38,39. Our observations do not disagree with these conclusions, but suggest that rostrocaudal clonal spread is extensive, crossing multiple axial levels, likely as a result of movement immediately following delamination but prior to leaving the neural tube. This clonal spread is unlikely a result simply of general extension of the embryonic body, because MVs were often exclusive to either DRGs or SGs. Indeed, upon abolishing the FGF signaling gradient, NC cells failed to migrate across multiple somite levels but instead moved laterally out of the neural tube. Our data support a model in which FGF activity is required to restrain premature emigration from the neural tube to distribute the clones across multiple spinal levels.

In summary, our findings suggest a strong fate bias for sensory or sympathetic lineage in pre-delaminating NC progenitors. These results are clinically relevant as they may provide insights into human neurocristopathies^40-42, for example, in neuroblastoma caused by the uncontrolled proliferation of sympathoadrenal lineage. The revised model of fate specification for sensory versus sympathetic lineage may suggest that tumorigenic driver mutation can be acquired early within the neural tube.

Methods

Human donor recruitment

Organs of ID06, ID07, and ID08 were collected from the UC San Diego Anatomical Material Program. Organs of ID06 were donated by a 70-year-old male. Organs of ID07 were donated by an 85-year-old male. Organs of ID08 were donated by a 74-year-old female. All three donors were documented to be of European ancestry and were neurotypical at the time of demise. Organs were collected within a 13-hour postmortem interval for all donors (ID06: 13 hrs, ID07: 11 hrs, ID08: 6 hrs). Prior medical history showed no signs of neurological, psychiatric, or cancer diseases. Donors tested negative for infection with HIV, Hepatitis B, or COVID-19 prior to dissection. According to existing law 45 CFR 46.102(e)(1), the use of human anatomical cadaver specimens of ID06, ID07, and ID08 are exempt from oversight of the University of California, San Diego Human Research Protections Program (IRB) but are subject to oversight by the University of California, San Diego Anatomical Materials Review Committee (AMRC). This study was overseen and approved by the AMRC with approval number is 106135. Donors met AMRC qualifications: (i) Obtained information or biospecimens through intervention or interaction with the individual, and uses, studies, or analyzes the information or biospecimens; or (ii) Obtained, useed, studied, analyzed, or generated identifiable private information or identifiable biospecimens.

Tissue dissection from human donors

All dissection was performed by an anatomical pathologist (G.N. and S.T.B.). For dissection to capture MVs from dorsal root ganglia (DRG) and sympathetic ganglia (SG) of ID06, ID07, and ID08, laminectomy was first performed, and the entire spinal cord was collected and snap frozen on dry ice. Dissection was then performed to collect DRG and SG from cervical (C), thoracic (T) and lumbar (L) levels. Dissected specimens were snap-frozen on dry ice. Organ samples including the left and right prefrontal cortex, cerebellum, left and right kidneys, left and right heart wall, as well as the liver, were biopsied with an 8mm skin punch and samples were also snap-frozen on dry ice. After dissection, samples were stored at −80°C.

DNA extraction from tissue

Small biopsies were bissected on dry ice, with half stored for future single-nuclei isolation, and the other half homogenized with a Pellet Pestle Motor (Kimble, #749540-0000) and resuspended with 450 μL RLT buffer (Qiagen, #40724) in a 1.5 ml microcentrifuge tube (USA Scientific, 1615-5500). DNA was extracted using the Qiagen DNeasy Blood & Tissue kit following standard protocols from the manufacturer (Qiagen, #69506).

Whole-genome library preparation and sequencing

300x whole genome sequencing (WGS) was performed to identify clonal mosaic variant (MV) detection ²⁶, with the following modifications to improve the MV barcoding strategy by including 30x WGS from clonally distributed variants from each single ganglion. This allowed the detection of locally shared clonal MVs at a significantly reduced sequencing cost, compared with 300x from each sample, while still retaining good sensitivity for MVs. 300x WGS (8 samples from ID06, 8 samples from ID07, and 17 samples from ID08) was performed on 1.0 μg of extracted DNA for PCR-free library construction using the KAPA HyperPrep PCR-Free Library Prep kit (Roche, #KK8505). Mechanical shearing using the Covaris microtube system (Covaris, #SKU 520053) was performed to generate fragments with a peak size of approximately 400bp, and then fragmented DNA sample were used to generate a library using Illumina dual index adapters. Beads-based double-size selection was performed to ensure the fragment sizes were between 300-600 bp as measured by Agilent DNA High Sensitivity NGS Fragment Analysis Kit (Agilent, #DNF-474-0500). 30x WGS (34 ganglia from ID06, 22 ganglia from ID07, and 18 ganglia from ID08), > 1.0 ng of DNA was used as input, and libraries were prepared with the NEBNext Ultra II FS DNA Library Prep Kit for Illumina following manufacturer’s guide, indexed with IDT for Illumina TruSeq UDI, then quantified with the KAPA Library Quantification Kits for Illumina platforms (Roche/KAPA Biosystems, #KK4824) on a Roche LightCycler 480 Instrument (Roche). Libraries with concentrations of more than 3 nM and fragments with peak size of 400 bp were sequenced on an Illumina NovaSeq 6000 S4 Flow Cell (FC), in 6-8 independent pools. For each sequencing run, 24 WGS libraries were normalized to obtain a final concentration of 2 nM using 10 mM Tris-HCl (pH 8 or 8.5; Fisher Scientific, #50-190-8153). 0.5 to 1% PhiX library was spiked in as positive control. The normalized libraries in a pool with a total of 311 μl libraries were denatured with 77 μl of 0.2 N NaOH (VWR, #82023-092) at RT for 8 min, terminated with. 78 μl of 400 mM Tris-HCl, then 400 pM pool was loaded on the S4 FC using Illumina SBS kits (Illumina, #20012866) with the following setting on the NovaSeq 6000: PE150:S4 FC, dual Index, Read 1:151, Index_Read2:8; Index_Read3:8; Read 4:151. The target for whole genome sequencing with high-quality sequencing raw data was 120 GB or greater with a Q30 >90% per library per sequencing run. FASTQ files generated with Picard’s (v 2.20.7) bcl2fastq2 (v 2.20) generated FASTQ files from raw sequence files were used for ID06, ID07, and ID08.

WGS data processing

FASTQ files were then aligned to the human_g1k_v37_decoy genome by BWA’s (v 0.7.17) mem with -K 100000000 -Y parameters, with SAM files to BAM files via SAMtools’s (v 1.7) view command. BAM files were subsequently sorted by SAMBAMBA’s (v 0.7.0) sort command and duplicated reads marked by its markdup command. Reads aligned to the INDEL regions were realigned with GATK’s (v 3.8-1) RealignerTargetCreator and IndelRealigner following the best practice guideline. Base quality scores were recalibrated using GATK’s (v 3.8.1) BaseRecalibrator and PrintReads. Germline heterozygous variants were called by GATK’s (v 3.8.1) HaplotypeCaller. The distribution of library DNA insertion sizes for each sample was summarized by Picard’s (v 2.20.7) CollectInsertSizeMetrics. The depth of coverage of each sample was calculated by BEDTools’s (v2.27.1) coverage command. The code and Snakemake wrapper of the pipeline are freely accessible on GitHub (https://github.com/shishenyxx/Human_DRG_SG).

Mosaic SNV/INDEL detection in WGS data

Mosaic single nucleotide variants/mosaic small (typically below 20 bp) INDELs were called by using a combination of four different computational methods: 1] MosaicHunter (single-mode, v 1.0)⁴³ with a posterior mosaic probability >0.05, 30x and 300x model for samples with different expected depth of coverage; 2] Single-mode of GATK’s (v 4.0.4) Mutect2 (https://doi.org/10.1101/861054) with “PASS” followed by DeepMosaic (v 1.0.1)⁴⁴; 3] Single-mode of GATK’s (v 4.0.4) Mutect2 with “PASS” followed by MosaicForecast (v 8-13-2019)⁴⁵ with models trained at different depth for 30x (30x model) and 300x (250x model), were implemented for sample-specific or tissue-shared variants; 4] The intersection of variants from the paired-mode of Mutect2 and Strelka2 (v 2.9.2) (set on “pass” for all variant filter criteria)⁴⁶ were collected for sample-specific variants. For the panel of normal samples required for the pipeline of DeepMosaic and MosaicForecast, we employed an in-house panel of similarly (300×) sequenced normal tissues (n = 15 sperm and 11 blood samples from 11 individuals)⁴⁷. For ‘tumor’-‘normal’ comparisons, required by Mutect2 and Strelka2 pipelines, we employed left-right combined heart tissues as ‘normal’. Variants were excluded if: 1) residing in segmental duplication regions as annotated in the UCSC genome browser (UCSC SegDup) or RepeatMasker regions⁴⁸; 2) residing within a homopolymer or dinucleotide repeat with more than 3 units; 3) overlapped with annotated germline INDELs. We further removed any variants with a population allele frequency higher than 0.001 in gnomAD (v 2.1.1)⁴⁹. Finally, variants with a lower CI of AF <0.001 were considered noises from reference homozygous and removed. Fractions of mutant alleles (i.e., AF) for variants called in one sample were calculated in all the other samples together with the exact binomial confidence intervals using scripts described below for MPAS analysis. This bioinformatic pipeline yielded a total of 3357 candidate MVs for ID06, 3380 candidate MVs for ID07, and 2415 candidate MVs for ID08 (for each skin sample because of the clonal nature only 100 of the total calls were randomly selected) that were interrogated with MPAS. Scripts for variant filtering are provided on GitHub (https://github.com/shishenyxx/Human_DRG_SG).

Single nucleus transcriptome profiling and primary template-directed amplification (PTA)

After isolation of neclei and sorting with BD Influx cell sorting machine (BD), a total of 224 single nuclei from each of 4 samples (left T2 DRG, left T2 SG, left T3 DRG, and left T3 SG from ID07) were snap-frozen on dry ice. Nuclei underwent the ResolveOME^™ workflow (BioSkryb Genomics Inc.)²⁷ . Briefly, Biotin-dT-primed first strand cDNA was generated. After termination of the reaction and nuclear lysis, whole genome amplification with primary template-directed amplification was performed. The mRNA-derived cDNA was affinity purified with streptavidin beads from the combined pool of cDNA and amplified genome. The remaining cDNA was pre-amplified on beads. Independently, amplified cDNA and single-cell genomic DNA from each cell underwent SPRI (Beckman Coulter, #B23319) cleanup prior to library preparation. Illumina libraries were prepared using the ResolveOME library preparation kit (BioSkryb Genomics Inc.) with NEXTFLEX Unique Dual Index Barcodes (PerkinElmer Applied Genomics, #NOVA-534100). Libraries were sequenced at low-pass (2x 50bp paired-end) targeting 2 million reads on a NextSeq (Illumina) instrument. Libraries of interest were identified based on QC sequencing and were subsequently sequenced at paired-end 150 bp (DNA libraries) and paired-end 100 bp (RNA derived libraries) on a NovaSeq X Plus (Illumina) platform on a 25B flowcell.

Massive parallel amplicon sequencing (MPAS) and single nucleus MPAS (snMPAS)

Three customized AmpliSeq Custom DNA Panel for Illumina (Illumina, #20020495) were used for MPAS for ID06, ID07, and ID08. Apart from MVs detected in ID06, ID07, and ID08 described above, we randomly selected 253 high-confidence heterozygous variants from ID06, 231 from ID07, and 181 from ID08 as positive controls heterozygous variants, which showed measured AFs between 48-52% for all the sequenced bulk tissues, with reading depths between 270-330×, all of which were present in gnomAD at population allele frequencies. We also randomly selected 123 alternative homozygous variants from ID06, 115 from ID07, and 100 from ID08 as negative controls, to exclude potential contamination or amplification bias, which showed measured ~0% AF across all sequenced samples, with average depth 270-330×, and gnomAD (v 2.1.1) allele frequency >0.5. Candidate MVs as well as heterozygous and homozygous variants were subjected to the AmpliSeq design system for each of the donor. Amplicon pools were designed form ID06 (#208852), ID07 (#208853), and ID08 (#212497), respectively. Regions of 2546 mosaic candidates, 212 het variants, and 101 hom variants were designed for ID06, 2607 mosaic candidates, 181 het variants, and 101 hom variants for ID07, as well as 1893 mosaic candidates, 143 het variants, and 91 hom variants for ID08. DNA from extracted tissue (for bulk sample MPAS) or that amplified from single nuclei (for snMPAS) was diluted to 5 ng/μl in low TE provided in AmpliSeq Library PLUS (384 Reactions) kit (Illumina, #20019103). An unrelated control DNA sample was included in individual plate of MPAS or snMPAS. AmpliSeq was carried out following the manufacturer's protocol (Illumina, #document 1000000036408v07). After amplification and FUPA treatment, libraries were barcoded with AmpliSeq CD Indexes (Illumina, #20031676) and pooled with similar molecular numbers based on measurements made with a Qubit dsDNA High Sensitivity kit (Thermo Fisher Scientific, #Q32854). To avoid index hopping, the three MPAS library pools ID06, ID07, and ID08, and the snMPAS library pool for ID07) were sequenced on separate lanes on different NovaSeq X Plus runs. 209.2 GB of FASTQ data were obtained from the ID06 MPAS libraries, 49.1 GB of FASTQ data were obtained from the ID07 MPAS libraries, 143 GB of FASTQ data were obtained from the ID08 MPAS libraries,150 GB of FASTQ data were obtained from the ID07 snMPAS libraries, all aiming for average 10000× for each variant.

Data analysis for MPAS and snMPAS

Raw reads from MPAS and snMPAS were mapped to the human_g1k_v37_decoy genome with BWA’s (v 0.7.17) mem command. BAM files were processed without removing PCR duplicates. Reads near insertion/deletions were re-aligned with GATK’s (v 3.8.1) IndelRealigner and base qualities scores were recalibrated with GATK’s (v 3.8.1) BaseRecalibrator. The final BAM files were parsed by SAMtools’s (v 1.7) mpileup and the 95% confidence intervals (CIs) of the measured AF of all the candidate MVs, together with the homozygous (negative control) and heterozygous (positive control) variants were estimated based on an exact binomial estimation (https://github.com/shishenyxx/Human_DRG_SG). Following depth calculation, regions of 2546 mosaic candidates, 212 heterozygous variants (positive controls), and 101 homozygous variants (negative controls) were detected and subjected to the next genotyping steps for ID06, 2607 mosaic candidates, 181 heterozygous variants, and 101 homozygous variants for ID07, as well as 1893 mosaic candidates, 143 heterozygous variants, and 91 homozygous variants for ID08. The genotypes of candidate MVs from MPAS were determined by comparing them to the AF distribution of the reference homozygous and heterozygous variants. The exact binomial lower bounds of all reference homozygous variants with >30 read depth were estimated and the 95% single-tail confidence threshold for the lower bound was calculated to be 0.0278686750 (ID06), 0.0454769028 (ID07) and 0.0034990564 (ID08). The distribution of the exact binomial upper bound of all heterozygous variants was calculated and 0.4084610 - 0.5799192 (ID06), 0.4166667 - 0.5710154 (ID07) and 0.4644791 - 0.5381781 (ID08) were the threshold for the upper bond based on ~5% false discovery rate (FDR) based on the built-in heterozygous and alternative homozygous genomic positions. Mosaic candidates from WGS were considered positive if variants met the following criteria at the same DNA samples: 1) the 95% exact binomial lower bound was higher than lower bound limit for each donor, 2) the upper CI of the unrelated control sample was lower than the lower bound of the same variants in the donors, 3) the 95% exact binomial upper bound was lower than upper bound threshold, 4) the sequencing depth was >30, and 5) the assessed alternative allele was supported by ≥3 reads. 6) for each donor, variants are detected in the original WGS tissue or adjacent dorsal root ganglia or sympathetic ganglia. These criteria ensured the false positive rate for each variant was under 5%. After the MPAS quantification, 827 MVs from ID06, 252 MVs from ID07, and 511 MVs from ID08 were considered positively validated in the sample where the variant was originally detected. The validation rate was 32.5%(827/2546), 9.7%(252/2607), and 27.0%(511/1893) in ID06, ID07, and ID08, respectively. 511 MVs from ID08 were used for all the analyses presented throughout the manuscript. In snMPAS, mosaic candidates from WGS were considered positive if the lower CI of AF was larger than the upper CI of AF in the normal control sample. To assess the precision of snMPAS data, we first calculated the chance of the failure of detection of true-positive variant calls (dropout), using heterozygous variants. Among AF values of the variants known as heterozygous in ID06, ID07, and ID08 bulk and homozygous in control bulk samples, those were recovered as heterozygous in snMPAS, resulting in an estimated false-negative rate of 19%. Next, we calculated the chance of false-positive discovery using homozygous genomic sites. The known homozygous genomic positions in bulk samples recovered as homozygous were 95%), resulting in an estimated false-positive rate of roughly 5%.

Permutation test for the significance of the snMPAS result

Permutation test for the significance of the snMPAS result were based on the known false-negative rate and false-positive rate described above. Variants in single nuclei were genotyped. Detection events of each MV were randomly re-assigned within a total of 224 nuclei, maintaining the original detection frequency. The observed number of the type of ganglia (SG:SG, DRG:DRG or SG:DRG) and the axial levels (T2:T2, T3:T3 or T2:T3), were used as outcome. A random distribution was generated from 10,000 permutations. The probability of observing terminal clusters was used as the one-tailed permutation p-value. A similar permutation analysis was also performed for the 75 nuclei from which the cell identity was inferred. Codes for the permutation analysis are provided on GitHub (https://github.com/shishenyxx/Human_DRG_SG).

Single-nucleus RNA sequencing with Chromium platform

Nuclei were isolated from dissected ganglia following mincing by scissors and disrupted with a pellet pestle. Resuspended in sorting buffer containing 1% BSA, stained with DAPI and sorted with BD Influx FACS. Sorted nuclei were resuspended in sorting buffer at 800~1000 nuclei/ul. Gel beads emulsion (GEM) generation, cDNA, and sequencing library constructions were performed in accordance with instructions in the Chromium Single Cell 3' Reagent Kits User Guide (v3.1).

Single-nucleus RNA-seq analysis

For the snRNA-seq data made with the Chromium platform, fastq files from single-nucleus libraries were processed through Cell Ranger (v7.0.2) analysis pipeline with –include-introns option and hg19 reference genome. Seurat (v4.0.5) package was used to handle single nuclei data objects. Data from each nucleus passed a control filter (nCount > 400, nFeature_RNA < 2000, percentage of mitochondrial gene < 10%), with a total of 15,896 protein-coding genes used for further downstream analysis. Data were normalized and scaled with the most variable 1000 features using the ‘ScaleData’ functions. Dimensionality reduction by PCA and UMAP embedding was performed using runPCA and runUMAP functions. Clustering was performed by FindNeighbors and FindClusters functions. For the full-length transcript snRNA-seq through the ResolveOME platform, quality control was carried out for raw FASTQ from single nucleus RNA sequencing from ResolveOME files using fastqc (v0.11.8). Preprocessing was carried out with cutadapt (v1.16). Cleaned FASTQ files were aligned to GRCh38 human genome and gencode (v27) gtf annotation using STAR (v2.6.0c). Aligned BAM files were indexed with SAMtools (v1.7). PCR duplicates were marked with Picard (v2.20.7) MarkDuplicates. Post-alignment quality control used Picard (v2.20.7) CollectRnaSeqMetrics, CollectInsertSizeMetrics, CollectGcBiasMetrics as well as qualimap (v2.2.2-dev). Raw read counts were collected with featureCounts (v2.0.0). Transcripts were collected with rsem (v1.3.1) with seed 12345 using the same gtf file. For both snRNA-seq data made with Chromium platform and ResolveOME, cell type identification was performed using known cell type markers expressed Dorsal Root Ganglia and Sympathetic Ganglia as well as using positive markers found by FindAllMarkers function with 1000 most variable features in scaled data. The final visualization of various snRNA-seq data was performed by r-ggplot2 (v3.3.5).

Phylogenetic tree analysis

From the 184 MVs in 224 raw and 75 cell-type-resolved single nuclei from ID07, the alleles at each genomic position were combined into one ‘pseudo-sequence’ for each sample, and a sequenced-based phylogenic tree was reconstructed to deconvolve the clonal relationship between single cells. Multiple sequence alignments were carried out with MUSCLE with manual correction for indels based on the information genotyped on each locus. An ‘evolutionary model’ was selected to infer the phylogeny of sequenced cells. The evolutionary history was inferred using the Minimum Evolution method⁵⁰. Evolutionary distances were computed using the Maximum Composite Likelihood method⁵¹, using base substitutions per site. The Minimum Evolution tree was searched using the Close-Neighbor-Interchange (CNI) algorithm at a search level of one⁵². The Neighbor-joining algorithm⁵³ was used to generate the initial tree. Evolutionary analyses were conducted with Molecular Evolutionary Genetics Analysis (MEGA, v11.0.13)⁵⁴, and the Nearest-neighbor-interchange heuristic method, the bootstrap consensus phylogenetic tree after 1000 bootstrap replications was generated. All ambiguous positions were removed for each sequence pair (pairwise deletion option) with a total of 189 positions in the final dataset. Cell types were based on transcriptomic information from the cell. For bootstrap analysis of bulk tissues. After 1,000 bootstrap replications and using the nearest-neighbor interchange heuristic method, the bootstrap consensus phylogenetic tree was generated. A bootstrap value of >95 (i.e. approximately unbiased p-value >95%) rejects the hypothesis that the cluster does not exist with a significance level <5%.

Estimate the number of starting cell populations

Variants shared by different groups of cells are used to estimate the upper bond starting populations. Based on the corrected value If

$N u m b e r o f P o s i t i v e V a r i a n t_{G r o u p 1} = 0$ or $N u m b e r o f P o s i t i v e V a r i a n t_{G r o u p 2} = 0$ , the variant is defined as “group 1 specific or group 2 specific”; if

0.666667 < \frac{\frac{N u m b e r o f P o s i t i v e V a r i a n t_{G r o u p 1}}{T o t a l s a m p l e_{G r o u p 1}}}{\frac{N u m b e r o f P o s i t i v e V a r i a n t_{G r o u p 2}}{T o t a l s a m p l e_{G r o u p 2}}} < 1.5,

the variant is defined as “shared between group 1 and groups 2”, if

1.5 < \frac{\frac{N u m b e r o f P o s i t i v e V a r i a n t_{G r o u p 1}}{T o t a l s a m p l e_{G r o u p 1}}}{\frac{N u m b e r o f P o s i t i v e V a r i a n t_{G r o u p 2}}{T o t a l s a m p l e_{G r o u p 2}}} < inf,

the variant is defined as “Group 1 enriched”, if

0 < \frac{\frac{N u m b e r o f P o s i t i v e V a r i a n t_{G r o u p 1}}{T o t a l s a m p l e_{G r o u p 1}}}{\frac{N u m b e r o f P o s i t i v e V a r i a n t_{G r o u p 2}}{T o t a l s a m p l e_{G r o u p 2}}} < 0.6667,

the variant is defined as “Group 2 enriched”. For any given variants, if it is shared between different groups (such as between dorsal root ganglia and sympathetic ganglia, or between left and right samples), the different presentations of the cells carrying the variants between each group indicate the number of starting cell population size when they split. Assuming that both groups had a similar number of cells at starting, according to the law of large numbers, the smaller the starting population is, the higher allelic fraction difference is expected to be observed between each population. On the other hand, stepwise simulation helps us obtain the number of cells that could support the observed difference between the two groups. 1000 values from starting cell number n=0 to 10,000 were tested with step 10. For each starting cell number, the hypergeometric distribution CI of the given success (based on average group MAF) is calculated.

Variants that are specific to one sample group (DRG-specific or SG-specific, Left-specific or Right-specific) are used to estimate the lower bond of the starting population before the groups split from the common progenitor. Under the assumption that only one cell was carrying the mutation when the two groups of cells were present, we estimated:

n u m b e r_{S t a r t i n g P o p u l a t i o n_{i}} = \frac{1}{2 * A v e r a g e A l l e l i c F r a c t i o n_{g r o u p_{i}}}

Clustered variant groups from ID06 with known bias resulting from potential local clonal expansion are excluded. The concept of this estimation is similar to what is described previously²⁶.

Comparing the allelic fraction differences between ganglia chains and ganglia at similar levels

To compare the allelic fraction differences within DRG and SG chains against those between DRG and SG at similar levels, we carried out a standard deviation analysis. To ensure a comparable sample size, each of the allelic fractions of variants in ganglia at the thoracic level is used for calculation. “Rostrocaudal” deviation within the ganglia chain is calculated as $\sqrt{\frac{\sum {(A F_{i} - \bar{A F_{g a n g l ı a c h a i n}})}^{2}}{N u m b e r o f g a n g l i a i n t h e c h a i n}}$ for each of the DRG-L, SG-L, DRG-R, and SG-R chains, whereas $i$ is the $i^{t h}$ thoracic level. “Dorsoventral” deviation within the neighbor ${(i - 1, i, i + 1)}^{t h}$ levels were calculated as $\sqrt{\frac{\sum {(A F_{i} - \bar{A F_{a t t h e {(ı - 1, ı, ı + 1)}^{t h}) l e v e l s}})}^{2}}{N u m b e r o f D R G a n d S G S a t t h e {(i - 1, i, i + 1)}^{t h} l e v e l s}}$ , where $A F_{i}$ is from any left or right DRG or SG in the ${(i - 1, i, i + 1)}^{t h}$ levels. We ensure a fair comparison by considering 12 ganglia for both rostrocaudal (DRG or SG from T1-T12) and dorsoventral axes (Left DRG/SG and right DRG/SG for 3 adjacent axial levels). As these groups contain the same number of observations (12 in each group), their allelic fraction standard deviations for each mosaic variant likely reflect the deviation of cellular factions of clones during rostal-caudal and DRG-SG migrations.

Mouse experiment

All experiments were performed in accordance with the Institutional Animal Care and Use Committees (IACUC) protocols at UC San Diego. All mice were kept in 12h light/dark cycles and fed ad libitum.

Mouse lineage tracing with Homing CRISPR

The Sox10-Cre mouse⁵⁵ (Jax #025807) was crossed with ROSA26-LSL-Cas9-GFP mice⁵⁶ (Jax #026175), then with Homing CRISPR MARC1-PB7 mice²⁸ (MMRRC #0654240UCD), and genotyped following standard protocol. Ganglia were dissected from 1-month-old mice, followed by tissue lysis, DNA extraction, targeted loci amplification, sequencing on a miSeq platform and data analysis following the detailed protocol previously published²⁸.

Mouse lineage tracing with DARLIN

DARLIN barcode mouse model and doxycycline-inducible Cas9 (ODInCas9) were described previously^30,57. Doxycycline at 25 mg/g body weight was administered to pregnant females via retro-orbital injection. The rostral and caudal halves of sympathetic chain and DRG were dissected. Bulk RNA was isolated with TRIzol. Targeted reverse transcription and library preparation were performed according to published protocol⁵⁸. Data analysis was performed based on published methods. Detailed parameters and codes are provided on GitHub (https://github.com/XiaoxuYangLab/Human_DRG_SG/tree/main/Analysis/Darline).

Single-cell RNA sequencing and analysis

Trunks of Wnt1-Cre;TdTomato (JAX #022137⁵⁹, #007914⁶⁰) mouse embryos were dissected and pooled according to stages. Cells were pooled from embryos of at least 2 litters. Tissues were dissociated in 0.25% Trypsin/EDTA and sorted using a flow cytometer (For gating strategy, see Supplementary Fig. 1). Gel beads emulsion generation and cDNA and sequencing library construction were performed following the Chromium Single Cell 3′ Reagent Kits User Guide (v4). A library pool was sequenced with using NovaSeq X. Fastq files from single-cells libraries were processed through Cell Ranger (v.9.0.1) analysis pipeline. Seurat (v.4.0.5) package was used to handle single cell data objects. Nuclei passed a control filter (number of genes >500, number of reads >1,000, percentage of mitochondrial gene < 10%) was used for downstream analysis. Protein-coding genes were used for further downstream analysis. Data were normalized and scaled with SCTransform functions. Dimensionality reduction by PCA and UMAP embedding was performed using RunPCA and RunUMAP functions. Clustering was performed by FindNeighbors and FindClusters functions. Cell type identification was performed using known cell type markers.

RNA velocity analysis

RNA velocity was analyzed separately for the E8.5, E9.0 and E9.5 timepoints to infer developmental trajectories using scVelo (v0.3.3) and Scanpy (v1.11.4). Spliced and unspliced transcript abundances were quantified using velocyto (v0.17.17) against the mm10 (GRCm38) reference genome with repeat masking from 10x Genomics. For each time point, loom files from biological replicates were merged into a single stage-specific loom matrix using loompy (v3.0.6), Following integration with Seurat-derived cell annotations and UMAP embeddings, the standard scVelo preprocessing pipeline was applied (filter_and_normalize with min_shared_counts=20 and 2,000 highly variable genes, PCA with 30 components, a k-nearest-neighbour graph with 30 neighbours and 30 PCs, and computation of spliced/unspliced moments). Velocity fields were estimated using the dynamical modeling framework (recover_dynamics). Finally, focusing on a curated set of transcription factors, we applied scVelo’s differential_kinetic_test at each time point, comparing lineage-specific fits (sensory NC vs all NC, or sympathetic NC vs all NC, defined from cell-type labels) to a single global kinetic fit of each gene’s spliced–unspliced phase portrait, and prioritized TFs with statistically significant lineage-biased kinetics as candidate regulators of sensory and sympathetic NC differentiation.

Quail experiment

Wildtype quails (Coturnix japonica) were hosted and bred at the University of Queensland. All experiments were approved by the University of Queensland Health Sciences Animal Ethics Committee (2023/AE000559).

Quail embryo culture

Eggs were kept at 14°C before use and incubated at 37.5°C in a humidified atmosphere until the desired stage. Quail eggs were cooled down for approximately 1 hour at room temperature (RT) before culture. A 5 mL syringe (Nipro) and a 18G sterile needle (BD PrecisionGlide) were used to aspirate 2 mL of thin albumin from the blunt end of the egg. A small window (~1.0 × 0.5 cm) was cut on the top of the eggshell to facilitate visualization and staging. Embryos were staged based on morphological criteria as described previously⁶¹. Embryos were cultured according to a protocol described previously⁶². Briefly, the egg yolk and albumin were slid out of the shell into a 35 mm petri dish with the embryo resting at the top of the yolk. The albumin covering the surface of the embryo was carefully wiped away with Kimwipe tissue. A piece of 1.5 × 1.5 cm square filter paper with a 0.2 × 1.0 cm hole in the center was placed onto the surface of the embryo. The filter paper was then cut along the edge and placed onto a 35 mm 1 well dish pre-coated with 1 mL agar-albumin. The yolk was washed away from the embryo using preheated Hanks’ Balanced Salt solution (HBSS). After washing, the embryo was transferred into another 35 mm 1 well dish pre-coated with 1 mL agar-albumin mixture consisting of 0.3% w/v bacto-agar and 50% v/v albumin.

Quail embryo electroporation

Ex ovo electroporation of PAX7-enh-H2B-Citrine at 2.0 μg/μL or FoxD3-EGFP (2.0 μg/μL) with pH2B-emiRFP670 (0.5 ug/ul) was performed with paddle platinum electrodes and NEPA21 Type II super electroporator (NepaGene). Hamburger and Hamilton (HH4) stage embryos were placed dorsal side down into the HBSS solution directly over the electrode. Approximately 1 μL of the plasmid was injected between the ectoderm and vitelline membrane with a glass capillary. For the unilateral electroporation experiments, the plasmid was injected only on one side and expression of H2B-emiRFP670 was used to confirm unilateral expression. A series of square electrical pulses (three pulses, 3 volts per pulse with 50-millisecond pulse length) was applied followed by 500 ms resting pulses. Following electroporation, embryos were transferred to an agar-albumin-coated dish and incubated at 37.5°C in a humidified environmental chamber until the HH9 stage. PAX7-enh-H2B-Citrine was a kind gift from Ruth Williams, pH2B-emiRFP670 was a gift from Vladislav Verkhusha (Addgene plasmid #136571) and pNC2-EGFP (FoxD3-enhancer driven EGFP plasmid) was a gift from Marianne Bronner.

FGF inhibition

For FGF inhibition, cultured embryos were treated with the pan-FGF receptor inhibitor infigratinib (NVP-BGJ398, Focus Bioscience at 10 μM diluted in HBSS). A 20 μL drop of infigratinib or DMSO control was pipetted onto an agar-albumin coated dish, and the embryo was placed on top, dorsal side down. An additional 20 μL of infigratinib or DMSO control was pipetted onto the ventral side of the embryo. The embryos were incubated at 37.5°C in a humidified environmental chamber for 1.5 hr before imaging began.

Live imaging and image processing

Following culture, embryos were allowed to rest for at least 1h at RT and then transferred dorsal side facing down to a 6-well imaging plate pre-coated with 250 μL of agar-albumin. All live imaging was done at the IMB microscopy facility (University of Queensland) supported by the Australian Cancer Research Foundation. Live imaging was performed on a Zeiss Axiovert 200 Inverted Microscope Stand with LSM 710 Meta Confocal Scanner fitted with dedicated GaAsP 488 nm detectors for increased sensitivity with environmental chambers at 37.5°C while imaging. Z-stacks were taken to encompass the entire dorsal half of the neural tube, the overlying ectoderm and the surrounding mesoderm. The time-lapse imaging was registered in Imaris 10.0.1 (Bitplane) to account for embryo movement or drift over time. PAX7-enh-H2B-Citrine positive cells were then tracked in Imaris, and track positions were exported to MATLAB (MATLAB ver. R2021a) for track displacement calculation. The track displacement was normalized to the displacement of the somite at the same axial level as the track starting point, to account for posterior elongation of the embryo. Somite displacement was measured using the Brightfield channel in FIJI. The centre position for each somite was measured as the centroid of a rectangular selection enclosing the somite. Somite length was measured using a straight line parallel to the rostrocaudal axis in FIJI in all the imaged embryos. Statistical analysis was performed using Prism Graphpad software (version 10.0.2).

Statistical tests and packages for customized plots

Hierarchical clustering with p-value via bootstrap resampling was performed using r-pvclust (v2.2.0) package. Pearson's product-moment correlation is calculated using cor.test() in R. One-way ANOVA with Tukey multiple comparisons of means was performed with aov() and TukeyHSD() function. Various heatmaps with dendrograms and sidebars were generated by ComplexHeatmap (v2.16.0) package. Various plots including violine plots, scatter plots, contour plots, bar plots, upset plots lolliplots were generated using r-ggplot2 (v3.4.3). The oncoplot was generated using maftools (v2.16.0) in R. UMAP analysis was performed with r-umap (v0.2.10.0) package. Single-nucleus RNA-seq data was analyzed and plotted using Seurat4 (v4.2.0) package in R.

Extended Data

Extended Data Fig. 3. — (a) RNA velocity analysis on UMAP at E8.5, E9.0 or E9.5 suggests the direction of NC development.

(b-d) Identification of putative sensory fate bias regulatory genes by differential kinetics test to estimate dynamics of splicing. Latent time and phase plots showing sensory fate-biased expression dynamics of *Onecut1* at (b) E8.5 (P = 3.61 × 10⁻³) and (d) E9.5 (P = 5.93 × 10⁻⁴), but not at E9.0 (c; P > 0.05).

(e-g) Latent time and phase plots showing statistically significant sympathetic fate-biased expression dynamics of *Rfx4* at (e) E8.5 (P = 7.72 × 10⁻⁶⁴), (f) E9.0 (P = 8.06x10⁻⁹⁸) and (g) E9.5 (P =1.02x10⁻⁷³).

Extended Data Fig. 4. — (a) Bioinformatic workflow for the identification of candidate mosaic variants (MVs). See Methods for a detailed description.

(b) Violin plots of log-transformed total read depths (y-axes) of all somatic MVs in 140 samples from ID06 and 57 from either ID07 or ID08 (x-axes). The blue dashed lines indicate 500x read depth.

(c) The distribution of the allelic fraction of MVs categorized into ‘homozygous’, ‘heterozygous’, ‘shared mosaic’, or ‘unique mosaic’ variants in all the samples from the 3 donors. AF, allelic fraction.

Extended Data Fig. 5. — (a) Estimation of cell type proportion in the DRG and SG from donor ID08 by single-nucleus RNA sequencing. Uniform manifold approximation and projection (UMAP) graph from single-nucleus RNA sequencing with nuclei isolated from DRG (n=1,714) and SG (n = 2,652).

(b) Feature plot showing expression of known markers for major cell types in DRG and SG.

(c) Bar graph showing cell type proportion in the DRG and SG following annotation of cluster identity with marker gene expression.

Extended Data Fig. 6. — (a) Schematics showing the comparison in the difference of AF for all MVs along the dorsoventral (D-V) axis (between DRG and SG) and along the rostrocaudal (R-C) axis (between all thoracic levels). Different colors indicate the ganglia used for quantification of standard deviation. Comparison made between every 12 different ganglia either along D-V or R-C axis.

(b-d) Violin plots showing the standard deviation of square-root transformed allelic fraction (s.d. of sqrt-AF) for the variants from ID06 (b), ID07 (c), and ID08 (d) compared between the D-V axis and R-C axis. Each dot represents individual MVs, with the color of the dots indicating the specific ganglia from which standard deviation was quantified, either along the D-V or R-C axis. Lower hinge of box indicates 25^th percentile, middle line within box indicates 50^th percentile, upper hinge of box indicates 75^th percentile. Whiskers extend from the hinges to the largest/smallest values that are within 1.5-time inter-quartile-range of the hinge. Data points beyond whisker ends are plotted individually as outliers. N=13*number of MVs in individual donor, number of MVs in ID06, ID07 and ID08 are 827, 252 and 511 respectively. P values are from two-tailed Mann-Whitney U tests. W statistics are the sum of the ranks.

(e-f) Contour plots showing the normalized difference in the allelic fraction of each mosaic variant observed from ID07 (e) and ID08 (f) between DRG and SG at T7-T12 levels (y-axis) and between left and right (x-axis). Green dots: individual mosaic variants. Blue contour: 2D Kernel density estimation. Grey curves: kernel density estimation along the x- or y-axis.

(g-h) Representative variants from ID07 for simulating the cell number prior to DRG-SG separation (g) and prior to left-right separation (h). Blue and red dashed lines: difference in average AF for the individual variant between DRG and SG or between left and right. Black lines: 95% bands of hypergeometric distribution from each simulated starting population size.

Extended Data Fig. 7. — (a) False positive rate of mosaic variant genotyping by snMPAS. Distribution of the reference homozygous variants from ID07, with the x-axis representing 1 minus the lower bound of 95% binominal confidence interval (95% CI). The blue dashed line indicates the cut-off for a false discovery rate of 5%.

(b) False negative rate of mosaic variant genotyping by snMPAS. Distribution of the upper bound of 95% binominal confidence interval. The blue dashed line indicates the cut-off for a false discovery rate of 5%

(c) Phylogenetic tree after 1,000 bootstrap replications from the 184 MVs in the 224 single nuclei. The numbers at branches of the tree are bootstrap values supporting each edge. Purple indicates a terminal branch pair that consists of both DRG and SG from the same level (either T2:T2 or T3:T3). Pink indicates ganglia pairs from different levels at the terminal branch (T2:T3).

(d-e) Graphs comparing the actual number of terminal branches observed in the phylogeny tree (blue dashed line, observation) and the distribution of expected number after 10,000 permutations (black line, permutation) for T3-DRG: T3-SG pair (p=0.4032, permutation test) (d) and T3-SG: T3-SG pair (p=0.0218, permutation test) (e).

Extended Data Fig. 8. — (a-b) Hierarchical clustering with Manhattan distances from a total of 55 neurons (a) and 20 glia (b) from the DRG and SG at T2 and T3 levels of ID07. Note that DRGs and SGs predominantly clustered separately at both levels, suggesting the lineage relationship between ganglia is not driven by spinal level but rather by identity (DRG or SG).

(c-d) Number of terminal branches observed in the phylogeny tree (blue dashed line, observation) and distribution expected after 10,000 permutations (black line: permutation) for T3-SG: T3-SG pair (p=0.0011, permutation test) (c) and T3-DRG: T3-SG pair (p=0.0427, permutation test) (d).

Extended Data Fig. 9. — (a) Experimental design for real-time imaging of NC progenitor migration in quail embryos. Quail embryos at Hamburger-Hamilton stage 4 (HH4) were electroporated with *Pax7* enhancer-driven citrine reporter and allowed to develop until HH9. Cells were imaged for 6h.

(b) Representative tracks showing migration paths of NC cells expressing *Pax7* reporter. Right panel: magnification of boxed region from left panel. Brackets: somites. Scale bar: 50μm.

(c) Histogram showing the rostrocaudal migration distance of 212 cells (n=6 embryos) tracked for 6h each. Blue shading: cells migrating caudally by more than 1 somite (mean somite length 81 ± 1 μm, n=92 somites from 6 embryos). There were 21.2% of NC cells migrating rostrocaudally across more than 1 somite level.

(d) Schematic of experimental design to evaluate midline-crossing migration of NC progenitors. Quail embryos at HH4 were electroporated with H2B-RFP and *FoxD3* enhancer-driven EGFP on one side. Spatial specificity of electroporation was confirmed by inspecting H2B-RFP expression 2 h later. Live imaging and tracking of *FoxD3+* NC progenitors were then performed at HH11.

(e-f) Quantification of cells migrating on the side of electroporation (ipsilateral) and crossing the midline (contralateral) indicated that 20% of NC migrated to the opposite side during the 6 hours of imaging (n=5 embryos).

(g) Dot plot showing gene expression of receptors for major signaling pathways in the mouse NC cells from single-cell RNA sequencing to identify putative signaling pathways guiding NC cell migration.

Extended Data Fig. 10. — (a) Models and predicted contour plots of possible cases whereby MVs are shared between ganglia along the rostrocaudal axis but restricted dorsoventrally (case 1, left) or shared the dorsoventral axis while restricted rostrocaudally (case 2, right). Axes: normalized AF difference between rostral and caudal against that between DRG and SG. In case 1, most MVs deviate from the center along the x-axis but cluster at the center of the y-axis because MVs are shared among rostral and caudal levels. In case 2, most MVs cluster around the center at the x-axis but not the y-axis since clones are more frequently shared dorsoventrally between DRG and SG, but not between different levels. Clones with genomic similarity are colored similarly.

(b-c) MV contour plots observed from ID07 (b) and ID08 (c) thoracic levels with the normalized difference in AFs between rostral (defined as T1-T6 levels) and caudal (defined as T7-T12 levels) (y-axis) vs. normalized difference between the DRG and SG (x-axis). Green dots: individual MVs. Blue contours: kernel density estimation of MV distributions. Grey curves: kernel density estimation along the respective axes. Data for the left-sided or right-sided ganglia are plotted separately.

(d-e) Comparison of the prevailing model of trunk NC development and the proposed model based on findings from this study. In the prevailing model (d), delaminated NC cells are multipotent, giving rise to both DRG and SG (blue, dark green, and light green cells). The NC cells do not migrate rostrocaudally until reaching the SG, while cells of the DRG do not migrate across different levels. In our proposed model (e), most NC cells are cell fate-specified (red, cyan, and yellow cells) but a minority of NC cells are multipotent (dark blue cells). NC cells migrate rostrocaudally and take up contralateral positions prior to delamination, thereby distributing descendant cells across multiple levels but mostly exclusive to either DRG or SG lineage.

Supplementary Material

Supplementary Video 1

Download video file^{(42.1MB, mp4)}

Supplementary Video 2

Download video file^{(17.3MB, mp4)}

Supplementary Video 3

Download video file^{(13.3MB, mp4)}

Supplementary Information

NIHMS2162736-supplement-Supplementary_Information.docx^{(622.4KB, docx)}

Acknowledgments

We thank all the individuals who donate their bodies and tissues for the advancement of scientific research. We thank J. Wallingford (UT Austin), B. Hamilton (UCSD) and K. Liu (UCL) for feedback and M. Bronner (Caltech) for advice. We thank M. Bronner (Caltech) and R. Williams (U Manchester) for FoxD3enh-GFP and Pax7enh-GFP plasmids, respectively; K. Kennedy (Bioskryb) for support with ResolveOME; and S. Wang and L. Li (Westlake) for data analysis with DARLIN. We thank C. Fine, M. Espinoza and M. Banihassan (UCSD) for FACS technical support, and acknowledge the UCSD Stem Cell Program and CIRM Major Facilities (grant no. FA1-00607) at the Sanford Consortium for Regenerative Medicine, the NIMH (grants no. U01MH108898 and no. R01MH124890 to J.G.G. and no. R21MH134401 to X.Y.), the Larry L. Hillblom Foundation (to J.G.G.), the NICHD (grants no. P01HD104436 to J.G.G. and no. R00HD111686 to X.Y.), the Simons Foundation (grant no. SFI-AN-AR-00012194 to K.I.V.), the Rady Children’s Institute for Genomic Medicine, the San Diego Supercomputer Center (SDSC) (grant no. TG-IBN19002), and SIG grant no. S10OD026929 to UCSD IGM and grant no. S10OD021644 supporting the Center for High Performance Computing (CHPC) and Utah Center for Genetic Discovery (UCGD) at the University of Utah. We acknowledge the Institute for Molecular Bioscience Advanced Imaging Platform, the ARC (grant no. DP220101878) and the NHMRC (grant no. 2013027, 2038843) to M.D.W.

Footnotes

Competing interests

The authors declare no competing interests.

Data availability

Raw whole genome sequencing massive parallel amplicon sequencing (MPAS), single nucleus MPAS (snMPAS), mouse single-cell RNA sequencing and human single-nucleus sequencing data are available through SRA (accession number: PRJNA799597); human_g1k_v37 reference genome: http://ftp.1000genomes.ebi.ac.uk/vol1/ftp/technical/reference/; gnomAD reference panel: https://gnomad.broadinstitute.org/; mm10 (GRCm38) reference genome: https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_000001635.20/.

Code availability

Codes for the data processing and annotation are provided on GitHub (https://github.com/shishenyxx/Human_DRG_SG).

References

1.Le Douarin NM & Dupin E The "beginnings" of the neural crest. Dev Biol 444 Suppl 1, S3–S13 (2018). 10.1016/j.ydbio.2018.07.019 [DOI] [PubMed] [Google Scholar]
2.Le Douarin NM A life in Science with the avian embryo. Int J Dev Biol 62, 19–33 (2018). 10.1387/ijdb.170287NL [DOI] [PubMed] [Google Scholar]
3.Teillet MA, Kalcheim C & Le Douarin NM Formation of the dorsal root ganglia in the avian embryo: segmental origin and migratory behavior of neural crest progenitor cells. Dev Biol 120, 329–347 (1987). 10.1016/0012-1606(87)90236-3 [DOI] [PubMed] [Google Scholar]
4.Lallier TE & Bronner-Fraser M A spatial and temporal analysis of dorsal root and sympathetic ganglion formation in the avian embryo. Dev Biol 127, 99–112 (1988). 10.1016/0012-1606(88)90192-3 [DOI] [PubMed] [Google Scholar]
5.Basch ML, Bronner-Fraser M & Garcia-Castro MI Specification of the neural crest occurs during gastrulation and requires Pax7. Nature 441, 218–222 (2006). 10.1038/nature04684 [DOI] [PubMed] [Google Scholar]
6.Hovland AS et al. Pluripotency factors are repurposed to shape the epigenomic landscape of neural crest cells. Dev Cell 57, 2257–2272 e2255 (2022). 10.1016/j.devcel.2022.09.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Thomas S et al. Human neural crest cells display molecular and phenotypic hallmarks of stem cells. Hum Mol Genet 17, 3411–3425 (2008). 10.1093/hmg/ddn235 [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Weston JA A radioautographic analysis of the migration and localization of trunk neural crest cells in the chick. Dev Biol 6, 279–310 (1963). 10.1016/0012-1606(63)90016-2 [DOI] [PubMed] [Google Scholar]
9.Noden DM An analysis of migratory behavior of avian cephalic neural crest cells. Dev Biol 42, 106–130 (1975). 10.1016/0012-1606(75)90318-8 [DOI] [PubMed] [Google Scholar]
10.Serbedzija GN, Bronner-Fraser M & Fraser SE A vital dye analysis of the timing and pathways of avian trunk neural crest cell migration. Development 106, 809–816 (1989). 10.1242/dev.106.4.809 [DOI] [PubMed] [Google Scholar]
11.Soldatov R et al. Spatiotemporal structure of cell fate decisions in murine neural crest. Science 364 (2019). 10.1126/science.aas9536 [DOI] [Google Scholar]
12.Erickson AG et al. Unbiased profiling of multipotency landscapes reveals spatial modulators of clonal fate biases. bioRxiv, 2024.2011.2015.623687 (2024). 10.1101/2024.11.15.623687 [DOI] [Google Scholar]
13.Krispin S, Nitzan E, Kassem Y & Kalcheim C Evidence for a dynamic spatiotemporal fate map and early fate restrictions of premigratory avian neural crest. Development 137, 585–595 (2010). 10.1242/dev.041509 [DOI] [PubMed] [Google Scholar]
14.McKinney MC et al. Evidence for dynamic rearrangements but lack of fate or position restrictions in premigratory avian trunk neural crest. Development 140, 820–830 (2013). 10.1242/dev.083725 [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Vincent E et al. Ret deficiency decreases neural crest progenitor proliferation and restricts fate potential during enteric nervous system development. Proc Natl Acad Sci U S A 120, e2211986120 (2023). 10.1073/pnas.2211986120 [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Baggiolini A et al. Premigratory and migratory neural crest cells are multipotent in vivo. Cell Stem Cell 16, 314–322 (2015). 10.1016/j.stem.2015.02.017 [DOI] [PubMed] [Google Scholar]
17.Bronner-Fraser M & Fraser S Developmental potential of avian trunk neural crest cells in situ. Neuron 3, 755–766 (1989). 10.1016/0896-6273(89)90244-4 [DOI] [PubMed] [Google Scholar]
18.Bronner-Fraser M & Fraser SE Cell lineage analysis reveals multipotency of some avian neural crest cells. Nature 335, 161–164 (1988). 10.1038/335161a0 [DOI] [PubMed] [Google Scholar]
19.Nitzan E et al. A dynamic code of dorsal neural tube genes regulates the segregation between neurogenic and melanogenic neural crest cells. Development 140, 2269–2279 (2013). 10.1242/dev.093294 [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Henion PD & Weston JA Timing and pattern of cell fate restrictions in the neural crest lineage. Development 124, 4351–4359 (1997). 10.1242/dev.124.21.4351 [DOI] [PubMed] [Google Scholar]
21.Harris ML & Erickson CA Lineage specification in neural crest cell pathfinding. Dev Dyn 236, 1–19 (2007). 10.1002/dvdy.20919 [DOI] [PubMed] [Google Scholar]
22.Luo R, Gao J, Wehrle-Haller B & Henion PD Molecular identification of distinct neurogenic and melanogenic neural crest sublineages. Development 130, 321–330 (2003). 10.1242/dev.00213 [DOI] [PubMed] [Google Scholar]
23.Ju YS et al. Somatic mutations reveal asymmetric cellular dynamics in the early human embryo. Nature 543, 714–718 (2017). 10.1038/nature21703 [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Coorens THH et al. Extensive phylogenies of human development inferred from somatic mutations. Nature 597, 387–392 (2021). 10.1038/s41586-021-03790-y [DOI] [PubMed] [Google Scholar]
25.Lee-Six H et al. Population dynamics of normal human blood inferred from somatic mutations. Nature 561, 473–478 (2018). 10.1038/s41586-018-0497-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Breuss MW et al. Somatic mosaicism reveals clonal distributions of neocortical development. Nature 604, 689–696 (2022). 10.1038/s41586-022-04602-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Chung C et al. Cell-type-resolved mosaicism reveals clonal dynamics of the human forebrain. Nature 629, 384–392 (2024). 10.1038/s41586-024-07292-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Leeper K et al. Lineage barcoding in mice with homing CRISPR. Nat Protoc 16, 2088–2108 (2021). 10.1038/s41596-020-00485-y [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Britsch S et al. The transcription factor Sox10 is a key regulator of peripheral glial development. Genes Dev 15, 66–78 (2001). 10.1101/gad.186601 [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Li L et al. A mouse model with high clonal barcode diversity for joint lineage, transcriptomic, and epigenomic profiling in single cells. Cell 186, 5183–5199 e5122 (2023). 10.1016/j.cell.2023.09.019 [DOI] [PubMed] [Google Scholar]
31.Kulesa P, Bronner-Fraser M & Fraser S In ovo time-lapse analysis after dorsal neural tube ablation shows rerouting of chick hindbrain neural crest. Development 127, 2843–2852 (2000). 10.1242/dev.127.13.2843 [DOI] [PubMed] [Google Scholar]
32.Ye AY et al. A model for postzygotic mosaicisms quantifies the allele fraction drift, mutation rate, and contribution to de novo mutations. Genome Res 28, 943–951 (2018). 10.1101/gr.230003.117 [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Marks JR et al. Unifying comprehensive genomics and transcriptomics in individual cells to illuminate oncogenic and drug resistance mechanisms. bioRxiv, 2022.2004.2029.489440 (2023). 10.1101/2022.04.29.489440 [DOI] [Google Scholar]
34.Martinez-Morales PL et al. FGF and retinoic acid activity gradients control the timing of neural crest cell emigration in the trunk. J Cell Biol 194, 489–503 (2011). 10.1083/jcb.201011077 [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Stanley EF, Ehrenstein G & Russell JT Evidence for anion channels in secretory vesicles. Neuroscience 25, 1035–1039 (1988). 10.1016/0306-4522(88)90056-5 [DOI] [PubMed] [Google Scholar]
36.Serbedzija GN, Fraser SE & Bronner-Fraser M Pathways of trunk neural crest cell migration in the mouse embryo as revealed by vital dye labelling. Development 108, 605–612 (1990). 10.1242/dev.108.4.605 [DOI] [PubMed] [Google Scholar]
37.Krull CE Segmental organization of neural crest migration. Mech Dev 105, 37–45 (2001). 10.1016/s0925-4773(01)00395-1 [DOI] [PubMed] [Google Scholar]
38.Kasemeier-Kulesa JC, Kulesa PM & Lefcort F Imaging neural crest cell dynamics during formation of dorsal root ganglia and sympathetic ganglia. Development 132, 235–245 (2005). 10.1242/dev.01553 [DOI] [PubMed] [Google Scholar]
39.Yip JW Migratory patterns of sympathetic ganglioblasts and other neural crest derivatives in chick embryos. J Neurosci 6, 3465–3473 (1986). 10.1523/JNEUROSCI.06-12-03465.1986 [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Etchevers HC, Dupin E & Le Douarin NM The diverse neural crest: from embryology to human pathology. Development 146 (2019). 10.1242/dev.169821 [DOI] [Google Scholar]
41.Fries LE, Dharma S, Chakravarti A & Chatterjee S Variability in proliferative and migratory defects in Hirschsprung disease-associated RET pathogenic variants. Am J Hum Genet (2025). 10.1016/j.ajhg.2025.02.004 [DOI] [Google Scholar]
42.Kaufman CK et al. A zebrafish melanoma model reveals emergence of neural crest identity during melanoma initiation. Science 351, aad2197 (2016). 10.1126/science.aad2197 [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Huang AY et al. MosaicHunter: accurate detection of postzygotic single-nucleotide mosaicism through next-generation sequencing of unpaired, trio, and paired samples. Nucleic Acids Res 45, e76 (2017). 10.1093/nar/gkx024 [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Yang X et al. Control-independent mosaic single nucleotide variant detection with DeepMosaic. Nat Biotechnol 41, 870–877 (2023). 10.1038/s41587-022-01559-w [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Dou Y et al. Accurate detection of mosaic variants in sequencing data without matched controls. Nat Biotechnol 38, 314–319 (2020). 10.1038/s41587-019-0368-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Kim S et al. Strelka2: fast and accurate calling of germline and somatic variants. Nat Methods 15, 591–594 (2018). 10.1038/s41592-018-0051-x [DOI] [PubMed] [Google Scholar]
47.Gullace S et al. Universal Fabrication of Highly Efficient Plasmonic Thin-Films for Label-Free SERS Detection. Small 17, e2100755 (2021). 10.1002/smll.202100755 [DOI] [PubMed] [Google Scholar]
48.Nassar LR et al. The UCSC Genome Browser database: 2023 update. Nucleic Acids Res 51, D1188–D1195 (2023). 10.1093/nar/gkac1072 [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Karczewski KJ et al. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature 581, 434–443 (2020). 10.1038/s41586-020-2308-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Rzhetsky A & Nei M A Simple Method for Estimating and Testing Minimum-Evolution Trees. Molecular Biology and Evolution 9, 945–967 (1992). [Google Scholar]
51.Tamura K, Nei M & Kumar S Prospects for inferring very large phylogenies by using the neighbor-joining method. Proc Natl Acad Sci U S A 101, 11030–11035 (2004). 10.1073/pnas.0404206101 [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Nei M & Kumar S Molecular evolution and phylogenetics, [by] Masatoshi Nei and Sudhir Kumar. (Oxford University Press, 2000). [Google Scholar]
53.Saitou N & Nei M The neighbor-joining method: a new method for reconstructing phylogenetic trees. Mol Biol Evol 4, 406–425 (1987). 10.1093/oxfordjournals.molbev.a040454 [DOI] [PubMed] [Google Scholar]
54.Tamura K, Stecher G & Kumar S MEGA11: Molecular Evolutionary Genetics Analysis Version 11. Mol Biol Evol 38, 3022–3027 (2021). 10.1093/molbev/msab120 [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Matsuoka T et al. Neural crest origins of the neck and shoulder. Nature 436, 347–355 (2005). 10.1038/nature03837 [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Platt RJ et al. CRISPR-Cas9 knockin mice for genome editing and cancer modeling. Cell 159, 440–455 (2014). 10.1016/j.cell.2014.09.014 [DOI] [PMC free article] [PubMed] [Google Scholar]
57.Lundin A et al. Development of an ObLiGaRe Doxycycline Inducible Cas9 system for pre-clinical cancer drug discovery. Nat Commun 11, 4903 (2020). 10.1038/s41467-020-18548-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Li L et al. DARLIN mouse for in vivo lineage tracing at high efficiency and clonal diversity. Nat Protoc 20, 2319–2344 (2025). 10.1038/s41596-025-01141-z [DOI] [PubMed] [Google Scholar]
59.Lewis AE, Vasudevan HN, O'Neill AK, Soriano P & Bush JO The widely used Wnt1-Cre transgene causes developmental phenotypes by ectopic activation of Wnt signaling. Dev Biol 379, 229–234 (2013). 10.1016/j.ydbio.2013.04.026 [DOI] [PMC free article] [PubMed] [Google Scholar]
60.Madisen L et al. A robust and high-throughput Cre reporting and characterization system for the whole mouse brain. Nat Neurosci 13, 133–140 (2010). 10.1038/nn.2467 [DOI] [PMC free article] [PubMed] [Google Scholar]
61.Ainsworth SJ, Stanley RL & Evans DJ Developmental stages of the Japanese quail. J Anat 216, 3–15 (2010). 10.1111/j.1469-7580.2009.01173.x [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Alvarez YD et al. A Lifeact-EGFP quail for studying actin dynamics in vivo. J Cell Biol 223 (2024). 10.1083/jcb.202404066 [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Video 1

Download video file^{(42.1MB, mp4)}

Supplementary Video 2

Download video file^{(17.3MB, mp4)}

Supplementary Video 3

Download video file^{(13.3MB, mp4)}

Supplementary Information

NIHMS2162736-supplement-Supplementary_Information.docx^{(622.4KB, docx)}

Data Availability Statement

Codes for the data processing and annotation are provided on GitHub (https://github.com/shishenyxx/Human_DRG_SG).

[R1] 1.Le Douarin NM & Dupin E The "beginnings" of the neural crest. Dev Biol 444 Suppl 1, S3–S13 (2018). 10.1016/j.ydbio.2018.07.019 [DOI] [PubMed] [Google Scholar]

[R2] 2.Le Douarin NM A life in Science with the avian embryo. Int J Dev Biol 62, 19–33 (2018). 10.1387/ijdb.170287NL [DOI] [PubMed] [Google Scholar]

[R3] 3.Teillet MA, Kalcheim C & Le Douarin NM Formation of the dorsal root ganglia in the avian embryo: segmental origin and migratory behavior of neural crest progenitor cells. Dev Biol 120, 329–347 (1987). 10.1016/0012-1606(87)90236-3 [DOI] [PubMed] [Google Scholar]

[R4] 4.Lallier TE & Bronner-Fraser M A spatial and temporal analysis of dorsal root and sympathetic ganglion formation in the avian embryo. Dev Biol 127, 99–112 (1988). 10.1016/0012-1606(88)90192-3 [DOI] [PubMed] [Google Scholar]

[R5] 5.Basch ML, Bronner-Fraser M & Garcia-Castro MI Specification of the neural crest occurs during gastrulation and requires Pax7. Nature 441, 218–222 (2006). 10.1038/nature04684 [DOI] [PubMed] [Google Scholar]

[R6] 6.Hovland AS et al. Pluripotency factors are repurposed to shape the epigenomic landscape of neural crest cells. Dev Cell 57, 2257–2272 e2255 (2022). 10.1016/j.devcel.2022.09.006 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R7] 7.Thomas S et al. Human neural crest cells display molecular and phenotypic hallmarks of stem cells. Hum Mol Genet 17, 3411–3425 (2008). 10.1093/hmg/ddn235 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R8] 8.Weston JA A radioautographic analysis of the migration and localization of trunk neural crest cells in the chick. Dev Biol 6, 279–310 (1963). 10.1016/0012-1606(63)90016-2 [DOI] [PubMed] [Google Scholar]

[R9] 9.Noden DM An analysis of migratory behavior of avian cephalic neural crest cells. Dev Biol 42, 106–130 (1975). 10.1016/0012-1606(75)90318-8 [DOI] [PubMed] [Google Scholar]

[R10] 10.Serbedzija GN, Bronner-Fraser M & Fraser SE A vital dye analysis of the timing and pathways of avian trunk neural crest cell migration. Development 106, 809–816 (1989). 10.1242/dev.106.4.809 [DOI] [PubMed] [Google Scholar]

[R11] 11.Soldatov R et al. Spatiotemporal structure of cell fate decisions in murine neural crest. Science 364 (2019). 10.1126/science.aas9536 [DOI] [Google Scholar]

[R12] 12.Erickson AG et al. Unbiased profiling of multipotency landscapes reveals spatial modulators of clonal fate biases. bioRxiv, 2024.2011.2015.623687 (2024). 10.1101/2024.11.15.623687 [DOI] [Google Scholar]

[R13] 13.Krispin S, Nitzan E, Kassem Y & Kalcheim C Evidence for a dynamic spatiotemporal fate map and early fate restrictions of premigratory avian neural crest. Development 137, 585–595 (2010). 10.1242/dev.041509 [DOI] [PubMed] [Google Scholar]

[R14] 14.McKinney MC et al. Evidence for dynamic rearrangements but lack of fate or position restrictions in premigratory avian trunk neural crest. Development 140, 820–830 (2013). 10.1242/dev.083725 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R15] 15.Vincent E et al. Ret deficiency decreases neural crest progenitor proliferation and restricts fate potential during enteric nervous system development. Proc Natl Acad Sci U S A 120, e2211986120 (2023). 10.1073/pnas.2211986120 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R16] 16.Baggiolini A et al. Premigratory and migratory neural crest cells are multipotent in vivo. Cell Stem Cell 16, 314–322 (2015). 10.1016/j.stem.2015.02.017 [DOI] [PubMed] [Google Scholar]

[R17] 17.Bronner-Fraser M & Fraser S Developmental potential of avian trunk neural crest cells in situ. Neuron 3, 755–766 (1989). 10.1016/0896-6273(89)90244-4 [DOI] [PubMed] [Google Scholar]

[R18] 18.Bronner-Fraser M & Fraser SE Cell lineage analysis reveals multipotency of some avian neural crest cells. Nature 335, 161–164 (1988). 10.1038/335161a0 [DOI] [PubMed] [Google Scholar]

[R19] 19.Nitzan E et al. A dynamic code of dorsal neural tube genes regulates the segregation between neurogenic and melanogenic neural crest cells. Development 140, 2269–2279 (2013). 10.1242/dev.093294 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R20] 20.Henion PD & Weston JA Timing and pattern of cell fate restrictions in the neural crest lineage. Development 124, 4351–4359 (1997). 10.1242/dev.124.21.4351 [DOI] [PubMed] [Google Scholar]

[R21] 21.Harris ML & Erickson CA Lineage specification in neural crest cell pathfinding. Dev Dyn 236, 1–19 (2007). 10.1002/dvdy.20919 [DOI] [PubMed] [Google Scholar]

[R22] 22.Luo R, Gao J, Wehrle-Haller B & Henion PD Molecular identification of distinct neurogenic and melanogenic neural crest sublineages. Development 130, 321–330 (2003). 10.1242/dev.00213 [DOI] [PubMed] [Google Scholar]

[R23] 23.Ju YS et al. Somatic mutations reveal asymmetric cellular dynamics in the early human embryo. Nature 543, 714–718 (2017). 10.1038/nature21703 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R24] 24.Coorens THH et al. Extensive phylogenies of human development inferred from somatic mutations. Nature 597, 387–392 (2021). 10.1038/s41586-021-03790-y [DOI] [PubMed] [Google Scholar]

[R25] 25.Lee-Six H et al. Population dynamics of normal human blood inferred from somatic mutations. Nature 561, 473–478 (2018). 10.1038/s41586-018-0497-0 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R26] 26.Breuss MW et al. Somatic mosaicism reveals clonal distributions of neocortical development. Nature 604, 689–696 (2022). 10.1038/s41586-022-04602-7 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R27] 27.Chung C et al. Cell-type-resolved mosaicism reveals clonal dynamics of the human forebrain. Nature 629, 384–392 (2024). 10.1038/s41586-024-07292-5 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R28] 28.Leeper K et al. Lineage barcoding in mice with homing CRISPR. Nat Protoc 16, 2088–2108 (2021). 10.1038/s41596-020-00485-y [DOI] [PMC free article] [PubMed] [Google Scholar]

[R29] 29.Britsch S et al. The transcription factor Sox10 is a key regulator of peripheral glial development. Genes Dev 15, 66–78 (2001). 10.1101/gad.186601 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R30] 30.Li L et al. A mouse model with high clonal barcode diversity for joint lineage, transcriptomic, and epigenomic profiling in single cells. Cell 186, 5183–5199 e5122 (2023). 10.1016/j.cell.2023.09.019 [DOI] [PubMed] [Google Scholar]

[R31] 31.Kulesa P, Bronner-Fraser M & Fraser S In ovo time-lapse analysis after dorsal neural tube ablation shows rerouting of chick hindbrain neural crest. Development 127, 2843–2852 (2000). 10.1242/dev.127.13.2843 [DOI] [PubMed] [Google Scholar]

[R32] 32.Ye AY et al. A model for postzygotic mosaicisms quantifies the allele fraction drift, mutation rate, and contribution to de novo mutations. Genome Res 28, 943–951 (2018). 10.1101/gr.230003.117 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R33] 33.Marks JR et al. Unifying comprehensive genomics and transcriptomics in individual cells to illuminate oncogenic and drug resistance mechanisms. bioRxiv, 2022.2004.2029.489440 (2023). 10.1101/2022.04.29.489440 [DOI] [Google Scholar]

[R34] 34.Martinez-Morales PL et al. FGF and retinoic acid activity gradients control the timing of neural crest cell emigration in the trunk. J Cell Biol 194, 489–503 (2011). 10.1083/jcb.201011077 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R35] 35.Stanley EF, Ehrenstein G & Russell JT Evidence for anion channels in secretory vesicles. Neuroscience 25, 1035–1039 (1988). 10.1016/0306-4522(88)90056-5 [DOI] [PubMed] [Google Scholar]

[R36] 36.Serbedzija GN, Fraser SE & Bronner-Fraser M Pathways of trunk neural crest cell migration in the mouse embryo as revealed by vital dye labelling. Development 108, 605–612 (1990). 10.1242/dev.108.4.605 [DOI] [PubMed] [Google Scholar]

[R37] 37.Krull CE Segmental organization of neural crest migration. Mech Dev 105, 37–45 (2001). 10.1016/s0925-4773(01)00395-1 [DOI] [PubMed] [Google Scholar]

[R38] 38.Kasemeier-Kulesa JC, Kulesa PM & Lefcort F Imaging neural crest cell dynamics during formation of dorsal root ganglia and sympathetic ganglia. Development 132, 235–245 (2005). 10.1242/dev.01553 [DOI] [PubMed] [Google Scholar]

[R39] 39.Yip JW Migratory patterns of sympathetic ganglioblasts and other neural crest derivatives in chick embryos. J Neurosci 6, 3465–3473 (1986). 10.1523/JNEUROSCI.06-12-03465.1986 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R40] 40.Etchevers HC, Dupin E & Le Douarin NM The diverse neural crest: from embryology to human pathology. Development 146 (2019). 10.1242/dev.169821 [DOI] [Google Scholar]

[R41] 41.Fries LE, Dharma S, Chakravarti A & Chatterjee S Variability in proliferative and migratory defects in Hirschsprung disease-associated RET pathogenic variants. Am J Hum Genet (2025). 10.1016/j.ajhg.2025.02.004 [DOI] [Google Scholar]

[R42] 42.Kaufman CK et al. A zebrafish melanoma model reveals emergence of neural crest identity during melanoma initiation. Science 351, aad2197 (2016). 10.1126/science.aad2197 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R43] 43.Huang AY et al. MosaicHunter: accurate detection of postzygotic single-nucleotide mosaicism through next-generation sequencing of unpaired, trio, and paired samples. Nucleic Acids Res 45, e76 (2017). 10.1093/nar/gkx024 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R44] 44.Yang X et al. Control-independent mosaic single nucleotide variant detection with DeepMosaic. Nat Biotechnol 41, 870–877 (2023). 10.1038/s41587-022-01559-w [DOI] [PMC free article] [PubMed] [Google Scholar]

[R45] 45.Dou Y et al. Accurate detection of mosaic variants in sequencing data without matched controls. Nat Biotechnol 38, 314–319 (2020). 10.1038/s41587-019-0368-8 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R46] 46.Kim S et al. Strelka2: fast and accurate calling of germline and somatic variants. Nat Methods 15, 591–594 (2018). 10.1038/s41592-018-0051-x [DOI] [PubMed] [Google Scholar]

[R47] 47.Gullace S et al. Universal Fabrication of Highly Efficient Plasmonic Thin-Films for Label-Free SERS Detection. Small 17, e2100755 (2021). 10.1002/smll.202100755 [DOI] [PubMed] [Google Scholar]

[R48] 48.Nassar LR et al. The UCSC Genome Browser database: 2023 update. Nucleic Acids Res 51, D1188–D1195 (2023). 10.1093/nar/gkac1072 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R49] 49.Karczewski KJ et al. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature 581, 434–443 (2020). 10.1038/s41586-020-2308-7 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R50] 50.Rzhetsky A & Nei M A Simple Method for Estimating and Testing Minimum-Evolution Trees. Molecular Biology and Evolution 9, 945–967 (1992). [Google Scholar]

[R51] 51.Tamura K, Nei M & Kumar S Prospects for inferring very large phylogenies by using the neighbor-joining method. Proc Natl Acad Sci U S A 101, 11030–11035 (2004). 10.1073/pnas.0404206101 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R52] 52.Nei M & Kumar S Molecular evolution and phylogenetics, [by] Masatoshi Nei and Sudhir Kumar. (Oxford University Press, 2000). [Google Scholar]

[R53] 53.Saitou N & Nei M The neighbor-joining method: a new method for reconstructing phylogenetic trees. Mol Biol Evol 4, 406–425 (1987). 10.1093/oxfordjournals.molbev.a040454 [DOI] [PubMed] [Google Scholar]

[R54] 54.Tamura K, Stecher G & Kumar S MEGA11: Molecular Evolutionary Genetics Analysis Version 11. Mol Biol Evol 38, 3022–3027 (2021). 10.1093/molbev/msab120 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R55] 55.Matsuoka T et al. Neural crest origins of the neck and shoulder. Nature 436, 347–355 (2005). 10.1038/nature03837 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R56] 56.Platt RJ et al. CRISPR-Cas9 knockin mice for genome editing and cancer modeling. Cell 159, 440–455 (2014). 10.1016/j.cell.2014.09.014 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R57] 57.Lundin A et al. Development of an ObLiGaRe Doxycycline Inducible Cas9 system for pre-clinical cancer drug discovery. Nat Commun 11, 4903 (2020). 10.1038/s41467-020-18548-9 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R58] 58.Li L et al. DARLIN mouse for in vivo lineage tracing at high efficiency and clonal diversity. Nat Protoc 20, 2319–2344 (2025). 10.1038/s41596-025-01141-z [DOI] [PubMed] [Google Scholar]

[R59] 59.Lewis AE, Vasudevan HN, O'Neill AK, Soriano P & Bush JO The widely used Wnt1-Cre transgene causes developmental phenotypes by ectopic activation of Wnt signaling. Dev Biol 379, 229–234 (2013). 10.1016/j.ydbio.2013.04.026 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R60] 60.Madisen L et al. A robust and high-throughput Cre reporting and characterization system for the whole mouse brain. Nat Neurosci 13, 133–140 (2010). 10.1038/nn.2467 [DOI] [PMC free article] [PubMed] [Google Scholar]

[R61] 61.Ainsworth SJ, Stanley RL & Evans DJ Developmental stages of the Japanese quail. J Anat 216, 3–15 (2010). 10.1111/j.1469-7580.2009.01173.x [DOI] [PMC free article] [PubMed] [Google Scholar]

[R62] 62.Alvarez YD et al. A Lifeact-EGFP quail for studying actin dynamics in vivo. J Cell Biol 223 (2024). 10.1083/jcb.202404066 [DOI] [Google Scholar]

PERMALINK

Developmental organization of sensory and sympathetic ganglia

Keng Ioi Vong

Yanina D Alvarez

Qingquan Zhang

Jiaming Weng

Geoffroy Noel

Scott T Barton

Changuk Chung

Robyn Howarth

Naomi Meave

Fiza Jiwani

Sai B Patarlapalli

Fenyong Yao

Fugui Zhu

Chelsea Barrows

Arzoo Patel

Jian Xiong Wang

Neil Chi

Stephen F Kingsmore

Melanie D White

Xiaoxu Yang

Joseph G Gleeson

Abstract

Fig. 1. ∣. Analysis of clonal relationships between sensory (DRG) and sympathetic ganglia (SG) by mosaic variant barcode analysis (MVBA) in mice and humans.

Delaminating NC progenitors produce a single lineage in mouse

Single-cell analysis confirms mouse NC fate bias

Identification of somatic MVs in human DRG and SG

Cells within DRG and SG are clonally distinct in humans

Fig. 2. ∣. Spatial distribution of clonal MVs in human DRG and SG.

Fig. 3. ∣. DRG and SG are clonally independent tissues in humans.

Human NC cell fate specification precedes delamination

Fig. 4. ∣. NC cell fate specification occurs prior to left-right determination.

Single-cell analysis confirms human DRG and SG clonality

Fig. 5. ∣. Rostrocaudal migration across multiple levels distributes clones specific to DRG or SG.

FGF-dependent NC migration along the rostrocaudal axis

Clonal spread of NC cells during rostrocaudal migration

Discussion

Methods

Human donor recruitment

Tissue dissection from human donors

DNA extraction from tissue

Whole-genome library preparation and sequencing

WGS data processing

Mosaic SNV/INDEL detection in WGS data

Single nucleus transcriptome profiling and primary template-directed amplification (PTA)

Massive parallel amplicon sequencing (MPAS) and single nucleus MPAS (snMPAS)

Data analysis for MPAS and snMPAS

Permutation test for the significance of the snMPAS result

Single-nucleus RNA sequencing with Chromium platform

Single-nucleus RNA-seq analysis

Phylogenetic tree analysis

Estimate the number of starting cell populations

Comparing the allelic fraction differences between ganglia chains and ganglia at similar levels

Mouse experiment

Mouse lineage tracing with Homing CRISPR

Mouse lineage tracing with DARLIN

Single-cell RNA sequencing and analysis

RNA velocity analysis

Quail experiment

Quail embryo culture

Quail embryo electroporation

FGF inhibition

Live imaging and image processing

Statistical tests and packages for customized plots

Extended Data

Extended Data Fig. 1. Analysis of clonal relationships between sensory (DRG) and sympathetic ganglia (SG) by Homing CRISPR barcoding in mice.

Extended Data Fig. 2. Single-cell RNA sequencing of mouse NC cells between E8.5 and E9.5.

Extended Data Fig. 3. RNA velocity analysis of mouse NC development.

Extended Data Fig. 4. Workflow for mosaic variant barcode analysis (MVBA).

Extended Data Fig. 5. Single-nucleus transcriptomic analysis of DRGs and SGs from donors.

Extended Data Fig. 6. Spatial distribution of clonal MVs in human DRGs and SGs.

Extended Data Fig. 7. Clonal analysis of DRGs and SGs at single-nucleus resolution.

Extended Data Fig. 8. Clonal relationship of single nuclei of DRGs and SGs.

Extended Data Fig. 9. Live imaging of NC migration in quail embryos.

Extended Data Fig. 10. NC migration distributes clones bilaterally across axial levels while mostly restricted to either DRG or SG.

Supplementary Material

Acknowledgments

Footnotes

Data availability