HOX gene expression in the developing human spine

Lawrence, John E. G.; Roberts, Kenny; Tuck, Elizabeth; Li, Tong; Mamanova, Lira; Balogh, Petra; Usher, Inga; Piapi, Alice; Mazin, Pavel; Anderson, Nathaniel D.; Bolt, Liam; Richardson, Laura; Prigmore, Elena; He, Xiaoling; Barker, Roger A.; Flanagan, Adrienne; Young, Matthew D.; Teichmann, Sarah A.; Bayraktar, Omer; Behjati, Sam

doi:10.1038/s41467-024-54187-0

Download PDF

Article
Open access
Published: 20 November 2024

HOX gene expression in the developing human spine

Nature Communications volume 15, Article number: 10023 (2024) Cite this article

16k Accesses
20 Citations
2 Altmetric
Metrics details

Subjects

Abstract

Positional coding along the anterior-posterior axis is regulated by HOX genes, whose 3’ to 5’ expression correlates with location along this axis. The precise utilisation of HOX genes in different human cell types is not fully understood. Here, we use single-cell and spatial-transcriptomics, along with in-situ sequencing, to create a developmental atlas of the human fetal spine. We analyse HOX gene expression across cell types during development, finding that neural-crest derivatives unexpectedly retain the anatomical HOX code of their origin while also adopting the code of their destination. This trend is confirmed across multiple organs. In the axial plane of the spinal cord, we find distinct patterns in the ventral and dorsal domains, providing insights into motor pool organisation and loss of collinearity in HOXB genes. Our findings shed new light on HOX gene expression in the developing spine, highlighting a HOX gene ‘source code’ in neural-crest cell derivatives.

Single-cell transcriptomic landscape of the developing human spinal cord

Article 24 April 2023

A harmonized atlas of mouse spinal cord cell types and their spatial organization

Article Open access 29 September 2021

Decoding the spatiotemporal regulation of transcription factors during human spinal cord development

Article Open access 05 January 2024

Introduction

The first sign of anatomical patterning in human embryos is the emergence of the primitive streak, which establishes the anteroposterior (rostrocaudal) and left-right axes¹. Molecularly, positional coding of this axis is underpinned by segment-specific expression of HOX genes, a group of transcription factors that contain a common DNA-binding homeodomain sequence². Spread over four chromosomes, HOX genes are grouped into four clusters (A, B, C, and D). During gastrulation, these genes are expressed in a manner dictated by their position within the cluster, such that 3’ genes contribute to rostral identity, and 5’ to caudal; a pattern first described in Drosophila³. In vertebrates, HOX genes are activated sequentially in a temporally-restricted manner, beginning with 3’ transcription, with the timing of progression through the cluster governed by CTCF binding sites^4,5,6. Although the number of HOX genes and clusters varies across species, this collinearity is remarkably conserved^7,8,9,10. In addition to their ability to dictate cell fate based on its position in the embryo, HOX genes also function during development to govern differentiation irrespective of a cell’s rostrocaudal position^11,12.

The positional identity conferred by HOX genes at this early stage is thought to be highly conserved through development, as exemplified by somite transplantation experiments in the chick and retinoic-acid-induced posteriorisation of the mouse vertebral column^13,14. Decades of work with such model organisms has thoroughly characterised HOX gene expression patterns in vertebrates, however subtle differences between these organisms’ HOX genes and those of humans do exist¹⁵. Consequently, the precise utilisation of the rostrocaudal HOX code across human fetal cell types remains incompletely described.

Here, we set out to examine HOX gene expression directly in the human developing spine by forming a highly detailed tissue atlas along the rostrocaudal (antero-porterior) axis using three complementary high-resolution mRNA assays; single-cell RNA sequencing (scRNAseq), Visium spatial transcriptomics (ST) and Caratana in-situ sequencing (ISS). Our aim was to delineate the expression of HOX genes along the rostro-caudal axis across all captured cell types at high resolution, whilst disentangling their roles of patterning and differentiation within the context of the developing spine. We show varying HOX gene expression across cell types through development, highlighting that cells derived from the neural crest express HOX genes that correspond to their point of origin along the rostrocaudal axis. We validate this finding in the fetal limb, gut and adrenal gland using scRNAseq and use ISS to show neurons fall into distinct categories with regards HOX expression. Using ST and ISS, we shed new light on HOX expression across the dorsoventral axis in the developing spinal cord along its rostrocaudal length.

Results

A developing human spine atlas resolved in time and space

We first defined the cellular landscape of the human developing spine. We collected 7 spines from fetuses aged between 5 and 13 weeks post-conception. From post-conception week (PCW) 9 onwards (n = 5), we dissected each spine into precise anatomical segments along the rostrocaudal axis using anatomical landmarks and processed these segments separately (Fig. 1A; see Methods). This enabled us to delineate the inherent rostrocaudal maturation gradient of the fetal spine – the temporal maturation difference of approximately 6 hours that exists between each vertebral level during development¹⁶. We processed fresh tissues using standard techniques to generate single-cell suspensions, enriched for viable cells, from which we derived single-cell mRNA libraries using a droplet-based method (Chromium 10X; see Methods). Applying widely adopted quality filters (see Methods), we obtained count tables of transcripts of approximately 174,000 cells. These segregated into 61 distinct cell clusters (Supplementary Fig. 1A–D; Fig. 2A & B; Supplementary Data 1 for QC), broadly representing neuro-glial (n = 30,376), mesenchymal progenitors (n = 15,756), osteochondral (n = 43,171), muscle (n = 27,128), fibrous (n = 19,080), tendon (n = 20,445), meningeal (n = 3521), dermal (n = 552), haematopoetic (n = 9214), and endothelial (n = 4763) cells. We also captured rare cell types, such as hypertrophic chondrocytes (n = 16), notochordal cells (n = 33), cord floor (n = 38) and roof (n = 83) plate cells (Figs. 1B; 2A, B; Supplementary Data 2 for marker genes).

**Fig. 1: An atlas of the developing fetal spine.**

**Fig. 2: Cellular landscape of the developing spinal cord.**

To validate and spatially resolve this census of cell types at PCW7 and PCW9, we utilised the Visium assay (Chromium), which provides 50μm resolution readouts of gene expression. We sectioned spines axially at different anatomical levels and examined each slide, applying the cell2location algorithm to obtain estimated cell type abundancy values for each voxel (Supplementary Fig. 1E for reconstruction accuracy)¹⁷. At PCW7, we also applied a 123-gene Cartana in-situ sequencing protocol (Chromium) to axial sections (see Methods). Consecutive axial sections were used for the two assays in order to maximise biological similarity. We were thus able to place populations of cells into their anatomical context, validated by the expression of classical marker genes (Fig. 1C–E). At PCW7, we were able to map both neuroprogenitors (NP) and post-mitotic cord neurons, together with marker genes, along the dorsoventral axis of the nascent cord at single-cell resolution (Fig. 2A–E). In doing so, we support the previous finding that PAX7 is expressed by the floor plate in humans, and capture a bridging population (Differentiating Neuroprogenitors) between progenitors and neurons that express NEUROD4; a molecule implicated in neurogenesis following cord injury (Fig. 2B, D)^18,19. At PCW9, mapping cells of the peripheral nervous system revealed clear segregation of transcriptionally similar clusters of glia to either the nerve root emerging from the cord (Schwann cell 1), the dorsal root ganglion (satellite cell) or the exiting spinal nerve tract (Schwann cell 2), whilst mapping sensory and sympathetic neuronal cells to anatomically appropriate locations (Figs. 1E; 3A–C). Examining heterogeneity over time, the fraction of cell states in each compartment changed through development (Fig. 3D). In the mesenchymal compartment, multipotent progenitors dominated early samples, and were gradually replaced by more differentiated cell types such as chondrocytes, osteoblasts, tenocytes and fibroblasts(Fig. 3D & E). Similarly, the neural crest compartment at PCW5 and 7 was dominated by sensory neuron progenitors and immature schwann cells, with more differentiated neurons and glia emerging thereafter (Fig. 3D, E). Similarly, progenitors accounted for the majority of cell types between PCW5 and 9, with CNS glia dominating thereafter (Fig. 3D & E). Our developmental atlas can be freely accessed at https://spinal-development.cellgeni.sanger.ac.uk/.

**Fig. 3: Cellular landscape of the peripheral nervous system.**

Rostrocaudal HOX expression in stationary cell types

Turning to our main question, we examined the pattern of HOX gene expression across different cell types. For this, we first excluded the early-stage fetuses, which could not reliably be dissected into precise anatomical segments. For the initial analyses, we categorised cell types into anatomically mobile (haematopoetic cells), neural crest-derived cells or stationary (all other cell types; Fig. 1B) We first aimed to identify HOX genes that reliably represented positional information across all stationary cells, regardless of type. We grouped stationary cell types and performed differential expression testing by region using the Willcoxon rank-sum test corrected for multiple comparisons to determine statistically significant trends in rostrocaudal HOX gene expression. In order to refine this positional code, we next examined the expression of the genes which reached significance in each of the anatomically stationary cell types separately and excluded any genes that exhibited expression patterns peculiar to a single cell type, presumably due to their playing a tissue-specific (and hence position agnostic) role in development during the period sampled. For example, HOXA6, HOXD3, HOXD4 and HOXD8 were expressed ubiquitously by tendon cells across the rostrocaudal axis (Supplementary fig. 2A). We next removed genes expressed by <10% of cells in their section (lowly expressed genes) and a log2-fold change value between sections below 0.2 (those that exhibited poor segment-specificity). This included group 13 genes, the expression of which triggers axial growth arrest, expressed at very low levels exclusively in the sacral samples, which included the coccyx⁴ (Supplementary fig. 2B). With this approach, we derived a rostro-caudal HOX code comprising 18 genes that exhibited the most position-specific expression patterns across stationary cell types in the spine (Fig. 4A; Supplementary Data 3). Unexpectedly, this included the antisense gene HOXB-AS3, which exhibited strong sensitivity for positional coding of the cervical region amongst HOX genes (p < 10⁻³⁰⁰). We validated our finding with two spatial sequencing technologies; Visium spatial transcriptomics (50μm resolution, whole transcriptome) and Caratana in-situ sequencing (single-cell resolution, 123-gene panel) (Fig. 4B–D; Supplementary fig. 2C; see methods).

Having identified this broadly applicable set of genes, we next investigated HOX expression in individual cell types (Supplementary Data 4). Osteochondral cells exhibited the broader HOX code, with the addition of HOXB6, which was specific to the cervical region. Similarly, in meningeal cells, HOXC11 was specific to the sacrum, HOXC5 the thorax and HOXA5 the cervical region, with HOXC4 demarcating cervical, rather than thoracic, tissue. HOXB2 was not specific to cervical tissue. Tendon cells exhibited the greatest HOX specificity by region. In addition to the broader HOX code, HOXC11 was specific to sacral cells, with HOXB6, HOXA5, HOXB3 and HOXA3 specific to the cervical region. Muscle by contrast had a weaker HOX code, with HOXD10, A11 & D11 not reaching significance in the sacrum, HOXA4 & B5 in the cervical region (which gained HOXA3), and only HOXC4 and HOXC6 maintained in the thoracic region. In fibrous tissue, HOXD11 did not meet the logFC threshold in the sacrum, but instead HOXC11 showed sacral specificity. HOXD9 served as an additional lumbar region marker, and HOXB3, A5 & C4 as cervical markers. With the exception of HOXB-AS3 & HOXB8, all other genes from the broader code remained. Finally, a few HOX genes were significantly differentially expressed in vascular tissue within different regions(Fig. 4A; Supplementary Data 4). This perhaps reflects their partially migratory origins in early development¹⁶. Examining these cell type expression trends through the developmental period sampled, we noted a pattern of weakening of the HOX code in later developmental stages, with PCW13 consistently exhibiting fewer region-specific genes (Supplementary Data 4).

This analysis allowed us to resolve HOX gene expression in cell types with poorly characterised (tendon cells) or controversial (spinal skeletal muscle²⁰; spinal meningeal cells) HOX expression, charting their expression through development²¹.

Rostrocaudal HOX expression in neural-crest derivatives

Having established a core HOX code in stationary, mesenchyme-derived cell types, we assessed its utilisation in anatomically mobile and neural crest-derived (and therefore migratory) cell types. (Fig. 5A; Supplementary Fig. 3). As expected, anatomically mobile haematopoetic cells did not follow the rostro-caudal HOX code. Examining cells expressing the neural crest markers MSX1, FOXD3, SNAI1, SNAI2, SOX5, SOX8, SOX10 and ETS1, we observed expression of the thoracic HOX code, in addition to the cervical gene HOXB5, across all anatomical locations, with no HOX genes differentially expressed in the thorax (Fig. 5A; Supplementary Data 3). In addition, these cells also expressed some HOX genes expressed by co-located stationary mesenchyme-derived cells. This included HOXB2 in NCC-derived cells in the cervical region and HOXC9, A9 & C10 to the lumbar region, together with HOXD11 in the sacral region (Fig. 5A; Supplementary Data 3). However, NCC-derived cells also expressed HOX genes not differentially expressed by local mesenchyme-derived cells, including HOXC4 in cervical cells, HOXC8 in lumbar cells and HOXA9 & C10 in sacral cells (Fig. 5A; Supplementary Data 3). In other words, these cells exhibited thoracic HOX expression regardless of their final destination (perhaps consistent with their truncal neural crest origin), whilst also expressing their own local HOX code, which partially overlapped with the expression pattern identified in stationary, mesenchyme-derived cells. This idiosyncratic HOX expression pattern seemed to remain consistent across the developmental stages sampled (Supplementary Fig. 3).

**Fig. 5: HOX gene expression in neural crest-derived cells.**

We validated this finding of thoracic HOX expression using both spatial transcriptomics (PCW9) and in-situ sequencing (PCW7) of axial sections at cervical, thoracic and lumbar levels (Fig. 5B). In the 50μm resolution spatial transcriptomics data, we found that voxels where the dominant cell type was NCC-derived expressed thoracic HOX genes, again with an expression of a local set of genes (Fig. 5B & D). For single-cell resolution in-situ sequencing data, we again showed the same expression pattern through analysis of NTRK1-expressing neurons within the dorsal root ganglia (DRG), which was first isolated anatomically using the WebAtlas browser’s lasso tool, before subsetting for NTRK1-positive cells only (Fig. 5B–D; See Methods)²². In order to eliminate any potential batch / technical effect, we further validated this finding by performing RNA in-situ hybridisation on a whole spine sagittal section at PCW8, again showing expression of a thoracic HOX gene (HOXB8) in cervical and lumbar DRGs, alongside expression of a local HOX gene, though HOXB4 did exhibit less restricted expression than HOXC10 (Fig. 6A). Exploring this finding further within the in-situ sequencing data, we found that individual NTRK1-expressing NC-derived cells in the cervical and lumbar DRGs fell into four categories with regards their HOX expression pattern: 1) Truncal source code only; 2) local code only; 3) co-expressing or 4) no HOX expression (Fig. 6B–D). Thoracic neurons were categorised as source code or non-expressing (Fig. 6E).

Fig. 6: Categorising dorsal root ganglion (DRG) neurons by HOX gene expression. — **Fig. 6: Categorising dorsal root ganglion (DRG) neurons by *HOX* gene expression.**

The expression of a region-specific set of HOX genes in NCC-derived cells may commence before their migration begins, during their migration, or be adopted after they arrive in their new location. Our data may provide some clues in this regard by assessing, in individual cells, the correlation of the expression of the classical NCC marker SOX10 versus local HOX code genes. If, for example, we found that more immature cells (i.e. those with higher expression of SOX10) expressed local HOX code genes less strongly, it may suggest that at least some of the local HOX code expression is acquired after neural crest settled in their new location and adopt their final identity. To pursue this analysis, we compared this ratio of SOX10 versus local HOX code expression of the most immature and mature cells that we captured in sufficient numbers, i.e. lumbar cells at PCW9 and PCW12 (with the local HOX code HOXC9, HOXA9, and HOXC10). Comparing PCW9 to PCW12 lumbar cells, we found a significant difference in the ratio for two HOX genes (HOXA9 and HOXC10); more immature cells expressed relatively more SOX10 than local HOX code (Fig. 6F). Although this observation would be consistent with a dynamic acquisition of the local HOX code upon arrival of neural crest cells, ultimately this question can only be answered through lineage tracing experiments. As these cannot be performed in human tissue, it would have to be pursued in model systems.

To further examine the expression of this thoracic HOX code through development, we performed scRNAseq on the lumbar spine from a PCW17 fetus. We found the expression of thoracic HOX genes, again with the expression of a set of lumbar genes that partially overlapped with that of other tissues, in neural crest-derived cells (Supplementary Fig. 4A-C). Finally, we examined HOX expression in a published adult lumbar (L4 & L5) DRG single nuclear sequencing dataset, and found consistent high expression of HOXB8, but not of other truncal HOX genes expressed by NCC in all regions of the fetal spine, in several clusters of ELAVL3⁺ neurons that differentially expressed the sensory neuron genes NEFM & NEFH^23,24 (Supplementary Fig. 4D & E). Other neural crest HOX genes were expressed at much lower levels in neuronal cells (Supplementary Fig. 4E).

Validation in other developing organs

To extend and validate our observations about HOX expression in neural crest-derived cells, we examined neural crest derivatives in other fetal organs: the adrenal medulla, the hindlimb and the gut. In the adrenal medulla²⁵, where neural crest-derived Schwann Cell Precursors (SCP) generate sympathoblasts and chromaffin cells via bridge cells, we observed that these differentially expressed thoracic HOX genes, even after differentiation into mature cell types (Supplementary fig. 4F; Supplementary Data 5). By contrast, in the non-neural crest-derived adrenal cortex, HOX gene expression followed that of the adrenal portion of the adrenogonadal primordium, with a more 3’ expression pattern²⁶ (Supplementary Fig. 4F). In the first-trimester fetal hindlimb, mesenchymal tissues (muscle, cartilage and tendon) of the stylopod and zeugopod expressed an anatomically appropriate code of groups 9-11, whereas SCPs expressed thoracic/truncal HOX code in addition to expressing this local code²⁷ (Supplementary Fig. 4G; Supplementary Data 5). In the fetal gut, cells of the neural-crest-derived enteric nervous system exhibited the 3’ HOX code of their origin – that of the vagal neural crest, regardless of anatomical location²⁸ (Fig. 6G). By contrast, mesenchymal cells expressed a regionally-appropriate HOX code, with colonic cells differentially expressing lumbosacral genes when compared to cells of the ileum, which expressed a strong 3’ signal^28,29 (Fig. 6G; Supplementary Data 5). Taken together, these findings delineated a tissue overarching principle of thoracic HOX gene expression in neural crest derivatives; a significant portion of these cells seem to express HOX genes that correspond to their anatomical region of origin in the neural crest.

Axial HOX gene expression in the developing spinal cord

Next, we searched for patterning genes along other body axes – left versus right and dorsoventrally. We first compared in each Visium section transcription of the same cell type in different anatomical side (eg, skeletal muscle-dominated voxels on left versus skeletal muscle voxels on right), and asked whether there were cell type independent transcripts that distinguished these planes. Whilst these analyses did not yield candidate genes that pervade all cell types, they did highlight other complex patterns of spatially variable gene expression in the spinal cord³⁰. We found that the expression of HOXB genes in the human embryonic cord did not exhibit collinearity, with members 3-9 of the cluster to the dorsal cord regardless of anatomical level (Fig. 7A & B; Supplementary Fig. 5A). Intriguingly, this included the antisense transcript HOXB-AS3, which was expressed in the medial aspect of the dorsal horn and the periaqueductal region (Fig. 7A; Supplementary Fig. 5A). Whilst group B genes were all dorsally expressed, their precise expression patterns varied within this domain (Fig. 7A). These findings build on patterns of HOXB expression identified in earlier developmental stages in animals, whereby HOXB genes are initially ubiquitously expressed throughout the neural tube, then restricted to the intermediate zone, which traverses the dorsoventral axis of the neural tube^31,32. Analysis of a single publicly available PCW5 spatial transcriptomics axial section suggested that at this earlier stage HOXB8 expression did not exhibit such clear dorsal restriction (Supplementary fig. 5B)³³. At PCW12, HOXB8 expression was very similar to PCW9 with clear restriction to the dorsal cord (Supplementary fig. 5B).

**Fig. 7: Patterning of the fetal spinal cord in the axial planes.**

The identity taken up by ventral progenitors in the cord is thought to be highly dependent on HOX expression. This was recently highlighted in vitro by Mouilleau et al. who showed that stalling FGF-induced temporal activation of HOX genes in human pluripotent stem cell (hPSC)-derived progenitors resulted in cervical motor neuron types, whereas accelerating HOX progression produced lumbar neuronal subtypes³⁴. We therefore aimed to explore these trends in vivo by examining HOX expression in the fetal ventral spinal cord along the rostro-caudal axis. In contrast to the dorsal horns, in the ventral spinal cord HOX expression varied with rostrocaudal position, highlighting the role of these genes in establishing anatomically appropriate motor neuron (MN) pools (Fig. 7B)^35,36. HOXC5 & 6 were expressed in the ventral cervical cord, with the latter extending somewhat into the thoracic cord (Fig. 7B). HOXC8 & 9 were expressed in the ventral thoracic cord, with the former exhibiting some ventral cervical expression (Fig. 7B). This overlap of HOXC6 & 8 is most likely due to their combined role in patterning digit-innervating neurons as the caudal limit of the cervical cord^36,37. HOXC10 & HOXD10 were exclusive to the ventral lumbar cord (Fig. 7B).

In order to explore these patterns further, we performed in-situ sequencing on a full-length midline sagittal section through the spine at PCW 8 (Fig. 8A). This again confirmed restriction of HOXB8 to the dorsal cord, with minimal ventral expression (Fig. 8A). HOXC6 exhibited diffuse expression throughout the ventral cord, extending more rostrally than other genes (Fig. 8A). Together with HOXC9, it was also expressed throughout the dorsal cord, at comparable levels (Fig. 8A). As in the ST data, HOXC8 exhibited a more rostral superior limit than HOXC9, in keeping with its dual role in patterning the caudal half of the lateral motor columns (LMC), including ETV4⁺ MNs, through co-expression with HOXC6³⁸, which was detected only at the inferior limit of the cervical cord (Fig. 8A). Interestingly, HOXC6 was expressed in the same territory as HOXC9, and HOXC8/9 expression extended into the territory of HOXC10/D10 expression (Fig. 7B, Fig. 8A). Whilst spatial transcriptomics data confirmed that HOXC6 expression dominates the ventral cervical cord, and group 10 genes dominate the ventral lumbar cord, both datasets revealed the surprising maintenance of lower level expression of other motor-column HOX genes in these regions (Figs. 7B and 8A). In the axial ST sections, at PCW7 & 9, multiple voxels covering the ventral cervical and thoracic cord coexpressed HOXC6 and HOXC9, and several voxels covering the ventral lumbar cord coexpressed HOXC9 and HOXC10/D10 (Fig. 7B, white asterisks). Further analysing this at single-cell resolution using ISS at PCW7 (from sections 12 µm adjacent to the Visium samples), we confirmed that some ventral cells at each level did indeed express, in combination, motor column-defining HOX genes previously reported as mutually exclusive (Fig. 7C). The same was true for the sagittal sections at PCW8 (Fig. 8B).

**Fig. 8: Patterning of the fetal spinal cord in the sagittal plane.**

Finally, we also examined expression of these genes by region in the single cell data (almost all of which originated from week 5 & 7 samples), again finding HOXB genes were exclusive to dorsal neurons in the developing cord with a high percentage of cells expressing them (Fig. 8C). The ventral HOX genes showed mixed expression (Fig. 8C). HOXC5, HOXC8, HOXC10 & HOXD10 were exclusive to ventral neurons, expressed by small percentages of the total cells (presumably reflecting their restriction along the rostro-caudal axis to particular regions), but HOXC6 and HOXC9 were more dorsally expressed (Fig. 8C). Together, These findings suggest that the mutual cross-repression of these groups of motor-defining HOX genes in the ventral cord is incomplete during the period studied (PCW7-9).

In a final analysis we searched for other rostrocaudal patterning genes that follow the expression of the core HOX code, that is, genes with rostrocaudal expression gradients across all stationary cell types that correlated with the core HOX code. Although our analyses did not reveal such genes, they enabled us to identify genes that varied with rostrocaudal position in some cell types. This included LINC01497, enriched in cervical perichondrial cells, which is associated with both short stature and decreased bone mineralisation, and potential negative regulators of osteoblastgenesis enriched in lumbar cells, such as the sulphating enzyme SULT1E1, which plays a key role in oestrogen metabolism, and the cell adhesion molecule NEGR1^39,40,41 (Supplementary fig. 7D & Supplementary Data 6). CHST9, which catalyses sulfation of proteoglycans, was enriched in cervical resting chondrocytes, along with HHIP, a regulator of hedgehog signalling⁴² (Supplementary fig. 7D). Interestingly, lumbar chondrocytes differentially expressed TMSB4X, a molecule implicated in load-adaptation (Supplementary fig. 7B)⁴³. Examining the HOX regulons using the DoRothEA R package, we found that none of these genes had direct evidence of being HOX targets⁴⁴.

Discussion

Here, we built a developmental cell atlas of the human spine based on high-resolution, quantitative transcriptional readouts. We were thus able to study the expression of HOX genes in human cell types along the rostrocaudal axis, as well as in the axial plane, which confirms and extends, in humans, the canon of HOX biology derived over decades from model organisms. In addition to serving as positional codes, HOX genes perform additional lineage and cell-type specific functions in development^11,12,45. As a consequence, some HOX genes lack specificity to serve as positional anchors for analyses. Our work distilled from orthogonal methods and multiple independent specimens a core HOX code in mesenchyme-derived tissues in the human spine between the 9th and 13th weeks postconception. Interestingly, we found that during this developmental window, NCC-derived cells seemed to express a thoracic HOX code, perhaps in keeping with their truncal origin within the neural crest, whilst also expressing a set of region-specific HOX genes at their end destination. Whilst there was some overlap between these region-specific genes and the anatomical code derived from mesenchyme-derived cells, differences were also evident. This suggests that NCC-derived cells utilise a different HOX code to mesenchyme for positional information. This idiosyncratic expression pattern is in keeping with other human tissues, such as the mesenchyme of the limb bud (which has its own HOX expression patterns) and the ventral and dorsal cord as discussed here⁴⁶. We showed this pattern of HOX expression corresponding to neural crest origin held true in the fetal limb and adrenal gland (containing truncal neural crest-derived cells), and the fetal gut (containing vagal neural crest-derived cells). Together, these findings shed further light on the utilisation of this enigmatic set of genes by different tissues during development.

The functional consequences of this neural crest HOX ‘source code’ remain unclear. Using in-situ sequencing, we demonstrated that NTRK1-expressing DRG neurons fall into several categories in terms of HOX expression: non-expressing, local code expressing, source code expressing and co-expressing. This may reflect functional subsets of neurons, with particular HOX combinations dictating subsequent biological processes, as is the case for the proprioceptive subset of sensory neurons in mice, in which HOX expression dictates sensory-motor synaptic matching⁴⁷. It may be that similar HOX-driven developmental processes also take place in the non-neuronal NCC-derived tissue studied here, such as the sympathoblasts of the adrenal gland.

Our study also investigated the expression of HOX genes along the dorsoventral axis of the spinal cord. Previous work with model organisms has uncovered the evolving role of the HOXB genes in neurogenesis. In the neural tube of the mouse and chick, HoxB genes are expressed before other Hox clusters, initially in a non-specific manner throughout the tube, before becoming somewhat restricted to the intermediate zone; an area bridging the ventricular zone (containing progenitors) and the mantle zone (containing post-mitotic cells)^31,32. Our study continued the study of this evolving expression pattern, showing with both single cell and spatial transcriptomics that at later time points (in the human) HOXB genes show restriction to the dorsal cord, and continue to lack any rostrocaudal collinearity. Perhaps after regulating neuronal delamination via LZTS1 at the intermediate zone in the neural tube, HOXB genes take on a further role in specifying sensory / dorsal identity in the cord. Further experiments in model systems examining the effect of HOXB inactivation at different developmental stages would shed light on their role at the stages studied here.

A further interesting finding in our study was the co-expression of certain motor-pool determining HOX genes in the ventral cord, which had previously been reported as mutually exclusive in animal models. As discussed above regarding HOXB gene expression, this may again be a reflection of evolving patterns of HOX expression during development; our study window being comparatively late compared to the animal models. However it remains possible that this is a species-specific feature. A previous in vitro study by Mouilleau et al. used the timing of retinoic acid application to culture to recapitulate the temporal nature of motor neuron development; cervical neurons develop earlier than lumbar neurons³⁴. Adding retinoic acid at day 3 onwards resulted in strong HOXC6 expression, whereas addition at day 6 onwards resulted in a clear shift towards HOXC9, recapitulating expression along the rostrocaudal axis³⁴. However, addition at day 5 did seem to result in some overlap in the expression of these two genes, though co-localisation analysis of these two supposedly exclusive genes was not performed. Further studies of their expression dynamics using similar methods would be enlightening. Finally, whilst RNA assays provide high resolution and sensitivity, they do not necessarily reflect the presence of protein; it may be that at the protein level HOX distribution in the human fetal cord is identical to that of animal models.

When examining HOX gene expression along the rostrocaudal axis, we utilised data generated from fetuses dissected into anatomical sections. This analysis was limited by the relatively narrow sample window (PCW9-13), as the earliest samples (PCW5 & 7) could not be reliably dissected into precise anatomical sections. This meant that, whilst robust across these relatively late developmental stages, it remains unclear how HOX gene expression across cell types changes early in human development. Model organism work has shown, in detail, the dynamic nature of HOX expression in the neural tube and early cord, and investigation of these periods using the techniques we employed here would be enlightening, if logistically challenging due to the process of fetal tissue collection from termination of pregnancy^31,32. In the stationary cell types captured by our experiments, it would be similarly interesting to examine earlier developmental stages for any transient HOX expression patterns, such as the waves of expression in the nascent limb bud²⁷.

A further limitation of our study is the breadth of cell capture per fetal spinal section in the scRNAseq data. In particular, neuroglial cells were poorly represented across ages and sections, preventing in-depth analyses of these cell types. Further studies with similar methodology but enriched for neuroglial cells would allow further investigation of HOX patterns in this tissue. Similarly, when investigating the dynamics of HOX expression in NCC-derived cells, we were limited to two samples (lumbar PCW9 & 12) as no other two samples from different fetuses at the same anatomical level captured sufficient NCC-derived cells to allow adequately powered analysis. Furthermore, PCW9 is relatively advanced in terms of neural crest specification; studying earlier time points would shed further light on HOX expression dynamics. Conversely, whilst spatial transcriptomics experiments capture RNA from a whole tissue section, the resolution of the technique in its current form (50μm, with dead space between voxels) is limiting. In our study, this prevented detailed analysis of PCW7 HOX expression at the DRG, as very few voxels covered the DRG, and many that did had significant overlap with non-neuronal tissue. This was particularly the case in the lumbar sample, which had the smallest DRGs. We were thus dependent on the much larger PCW9 samples for single transcriptomics-based validation of the HOX source code and unable to directly compare spatial transcriptomics and in-situ-sequencing results at PCW7. Similarly, the much smaller size of PCW7 samples meant individual voxels were more likely to cover multiple tissue types, meaning PCW9 samples were preferable for mapping sequenced single cells to their predicted location on the ST slides.

High-resolution transcriptional assays of human fetal cells represent a powerful tool for assessing differentiation states. However, transcription is by definition plastic and therefore does not usually generate fixed developmental barcodes (e.g. somatic mutations), which lineage tracing relies upon. Our finding of HOX gene expression in truncal and vagal neural crest-derived cells that corresponds to their point of origin may lend itself to retrospective lineage tracing in development from mRNA assays. As the developmental Human Cell Atlases will be extended in coming years, identification of idiosyncratic gene expression patterns in cells of particular developmental origins may provide a powerful tool to obtain a fuller picture of developmental patterning in humans, not confined to the stages and organ systems we studied here.

Methods

Ethics statement

Human embryonic tissue was collected from elective termination of pregnancy procedures at Addenbrookes Hospital, Cambridge, UK under full ethical approval from the East of England–Cambridge Central Research Ethics Committee (REC-96/085) and from the joint MRC/Wellcome Trust Human Developmental Biology Resource (htttp://hdbr.org; grant number MR/R006237/1) at the Institute of Child Health, London (Registration number 200532; REC 18/LO/0822) under full ethical approval from the London-Fulham Research Ethics Committee. Informed, written consent was obtained from patients after the decision was made in the clinic to terminate pregnancy, in advance of the termination of pregnancy procedure. Experiments followed the 2021 International Society for Stem Cell Research (ISSCR) guidelines in working on human embryos. No developmental abnormalities were visible or known in any of the embryos collected. No participant compensation occurred. In both cases, fetal age (postconception weeks, PCW) was estimated using the independent measurement of the crown rump length (CRL), using the formula PCW (days) = 0.9022 × CRL (mm) + 27.372. Genotyping for embryo sex was not performed.

Tissue dissection and dissociation

Fetal spines were dissected under a microscope using sterile microsurgical instruments. Where anatomy was reliably preserved (PCW9 onwards), the spine was dissected into separate anatomical sections (cervical, thoracic, lumbar and sacral) by an orthopaedic surgeon (JEGL) prior to digestion. The cervico-thoracic junction was identified by the junction of the 1st rib to the body of the first thoracic vertebra. A cut was then made at the C7/T1 disc to isolate the cervical section. Vertebrae were then counted rostrocaudally to identify the first lumbar vertebra, sense-checked by the presence of the 12th rib joining the 12th thoracic vertebral body immediately superior to it. A cut was then made at the T12/L5 disc to isolate the lumbar and thoracic sections. Finally, a cut was made at the L5/S1 disc to isolate the sacro-coccygeal section. Each sample was digested in a 5 µg/ml Liberase TH working solution prepared from Liberase TH powder (Sigma 5401135001) and 1X Phosphaste Buffered Saline (PBS) on a shaking platform (750 rpm) at 37 °C for 30 min. The tissue was gently agitated using a P1000 pipette after 15 min. 5 ml 2% Fetal bovine serum (FBS) in PBS was then added to stop the dissociation, prior to second stage digestion with 0.25% trypsin solution for a further 30 min at 37 °C, with pipette-agitation every 5 min. Cells were then spun down at 750 g at 4 °C for 5 min and resuspended 50-200ul of 2% FBS in PBS. Fetal cells were loaded for scRNAseq directly following sample processing.

Visium spatial transcriptomics experiments

Embryonic spine samples from the cervical, thoracic and lumbar regions at PCW7-9 were embedded in OCT within cryo wells and flash-frozen using an isopentane & dry ice slurry. Ten-micron thick cryosections were then cut in the axial plane and transferred onto Visium slides prior to haematoxylin and eosin staining and imaged at 20X magnification on a Hamamatsu Nanozoomer 2.0 HT Brightfield. These slides were then further processed according to the 10X Genomics Visium protocol, using a permeabilisation time of 12 min for all samples. Images were exported as tiled tiffs for analysis. Dual-indexed libraries were prepared as in the 10X Genomics protocol, pooled at 2.25 nM and sequenced 4 capture areas per Illumina Novaseq SP flow cell with read lengths 28 bp R1, 10 bp i7 index, 10 bp i5 index, 90 bp R2. For PCW7 samples, in order to reduce technical batch effects, two anatomical regions were included on each capture areas (cervical & lumbar; cervical & thoracic; thoracic & lumbar) totalling 3 capture areas. For PCW9, a single capture area was used per anatomical region, with two thoracic sections taken (upper and lower), totalling 4 capture regions. A single lumbar PCW8 section was sequenced on the final capture area.

Cartana in-situ sequencing experiments

In situ sequencing was performed using the 10X Genomics CARTANA HS Library Preparation Kit (1110-02, following protocol D025) and In Situ Sequencing Kit (3110-02, following protocol D100). Embryonic spine samples from the cervical, thoracic and lumbar regions at PCW7 were embedded in OCT and axially sectioned as described above. Sections immediately adjacent to those used for the Visium assay

ISS was performed using the 10X Genomics CARTANA HS Library Preparation Kit (1110-02, following protocol D025) and In Situ Sequencing Kit (3110-02, following protocol D100), which comprise a commercialised version of HybISS⁴⁸.

Briefly: cryosections of developing spine samples were fixed in 3.7% formaldehyde (Merck 252549) in PBS for 30 min and washed twice in PBS for 1 min each prior to permeabilization: sections were briefly digested with 0.5 mg/ml pepsin (Merck P7012) in 0.1 M HCl (Fisher 10325710) at 37 °C for 1 min, then washed twice again in PBS, all at room temperature. Following dehydration in 70% and 100% ethanol for 2 min each, an appropriate SecureSeal hybridisation chamber (either 9 mm diameter, 50 μl volume; 13 mm diameter, 100 μl; or 21.5 × 34.5 mm, 400 μl; all Grace Bio-Labs) was adhered to each slide and used to hold subsequent reaction mixtures. Following rehydration in buffer WB3, probe hybridisation in buffer RM1 was conducted for 16 h at 37 °C. The 123-plex probe panel included 5 padlock probes per gene, the sequences of which are proprietary (10X Genomics CARTANA). The section was washed with PBS-T (PBS with 0.05% Tween-20) twice, then with buffer WB4 for 30 min at 37 °C, and thrice again with PBS-T. Probe ligation in RM2 was conducted for 2 h at 37 °C and the section washed thrice with PBS-T, then rolling circle amplification in RM3 was conducted for 18 h at 30 °C. Following PBS-T washes, all rolling circle products (RCPs) were hybridised with LM (Cy5 labelling mix with DAPI) for 30 min at room temperature, the section was washed with PBS-T and dehydrated with 70% and 100% ethanol. The hybridisation chamber was removed and the slide mounted with SlowFade Gold Antifade Mountant (Thermo S36937). Imaging of Cy5-labelled RCPs at this stage acted as a QC step to confirm RCP (‘anchor’) generation and served to identify spots during decoding. Imaging was conducted using a Perkin Elmer Opera Phenix Plus High-Content Screening System in confocal mode with 1 μm z-step size, using a 63X (NA 1.15, 0.097 μm/pixel) water-immersion objective. Channels: DAPI (excitation 375 nm, emission 435-480 nm), Atto 425 (ex. 425 nm, em. 463–501 nm), Alexa Fluor 488 (ex. 488 nm, em. 500–550 nm), Cy3 (ex. 561 nm, em. 570–630 nm), Cy5 (ex. 640 nm, em. 650–760 nm).

Following imaging, each slide was de-coverslipped vertically in PBS (gently, with minimal agitation such that the coverslip ‘fell’ off to prevent damage to the tissue). The section was dehydrated with 70% and 100% ethanol, and a new hybridisation chamber secured to the slide. The previous cycle was stripped using 100% formamide (Thermo AM9342), which was applied fresh each minute for 5 min, then washed with PBS-T. Barcode labelling was conducted using two rounds of hybridisation, first an adapter probe pool (AP mixes AP1-AP6, in subsequent cycles), then a sequencing pool (SP mix, customised with Atto 425), each for 1 h at 37 °C with PBS-T washes in between and after. The section was dehydrated, the chamber removed, and the slide mounted and imaged as previously. This was repeated another five times to generate the full dataset of 7 cycles (anchor and 6 barcode bits).

Image-based in situ sequencing (ISS) decoding

We employed the ISS decoding pipeline outlined in Li et al.⁴⁹. This pipeline consists of five distinct steps. Firstly, we performed image stitching using Acapella scripts provided by Perkin Elmer, which generated two-dimensional maximum intensity projections of all channels for each cycle. Next, we employed Microaligner version 1.0.0⁵⁰ to register all cycles based on DAPI signals using the default parameters. For cell segmentation, we utilised a scalable algorithm that leverages CellPose version 2.0⁵¹ as the segmentation method. The expected cell size is set to 70 pixels in diameter and further expanded 10 pixels to mimic the cytoplasm. To decode the RNA molecules, we employed the PoSTcode version 1.0 algorithm⁵² with the following parameters: rna_spot_size=7, prob_threshold=0.95, trackpy_percentile=95, trackpy_separation=3. Furthermore, we assigned the decoded RNA molecules to segmented cells using STRtree and subsequently generated AnnData objects, using the AnnData version 0.11 python package⁵³.

Multiplexed smFISH

Cryosections were processed using a Leica BOND RX to automate staining with the RNAscope Multiplex Fluorescent Reagent Kit v2 Assay (Advanced Cell Diagnostics, Bio-Techne), according to the manufacturers’ instructions. Probes may be found in Supplementary Data 7. Prior to staining, fresh frozen sections were post-fixed in 4% paraformaldehyde in PBS for 15 min at 4 °C, then dehydrated through a series of 50%, 70%, 100%, and 100% ethanol, for 5 min each. Following manual pre-treatment, automated processing included digestion with Protease IV for 30 min prior to probe hybridisation. Tyramide signal amplification with Opal 520, Opal 570, and Opal 650 (Akoya Biosciences) and TSA-biotin (TSA Plus Biotin Kit, Perkin Elmer) and streptavidin-conjugated Atto 425 (Sigma Aldrich) was used to develop RNAscope probe channels. Stained sections were imaged as for ISS above.

Single-cell RNA sequencing

The single-cell suspensions derived from each sample were loaded onto separate channels of a Chromium 10x Genomics single-cell 3’version 2 library chip as per manufacturer’s protocol (10x Genomics; PN-120233). cDNA sequencing libraries were prepared as per the manufacturer’s protocol and sequenced using an Illumina Hi-seq 4000 with 2x150bp paired-end reads.

Alignment, quantification and quality control of single-cell data

Raw sequence reads in FASTQ format were processed and aligned to the GRCh38 version 1.2.0 human reference transcriptome through the Cellranger version v3.0.2 pipeline (10x Genomics) using default parameters.

The resulting expression matrices were processed with the SoupX package version 1.4.8 for R to estimate and remove cell-free mRNA contamination prior to analysis with the Seurat version 4.0.1 package for R^54,55.

Cells with fewer than 750 genes and greater than 10,000 genes were filtered, as well as those in which mitochondrial genes represented 10% or greater of total gene expression.

Alignment, quantification and quality control of Visium

Raw FASTQ files and histology images were processed, aligned and quantified by sample using the Space Ranger software v.1.0.0. which uses STAR v.2.5.1b52 for genome alignment, against the Cell Ranger hg38 reference genome refdata-cellranger-GRCh38-3.0.2, available at: http://cf.10xgenomics.com/supp/cell-exp/refdata-cellranger-GRCh38-3.0.2.tar.gz.

Spots were automatically aligned to the paired H&E images by Space Ranger software. All spots under tissue detected by Space Ranger were included in downstream analysis.

Doublet removal

Each 10x run was processed with the Scrublet version 0.2.2 pipeline using default parameters⁵⁵. The louvain algorithm for clustering was then re-run at resolution 10 to provide an over-clustered manifold, and any clusters where greater than 20% of cells were called as doublets were removed. Subsequently, any remaining cells called as doublets were removed from analysis.

Dimensional Reduction, clustering and annotation of single-cell data

Single-cell data were processed using the Seurat package v4.0.1 for R⁵⁶. To account for variations in cell-cycle stage, Seurat’s ‘CellCycleScoring’ function was performed on the remaining cells to produce a quantitative estimation of cell cycle stage. Log normalisation was then performed using the “NormalizeData” function with default parameters such that total number of counts per cell were normalized to 10,000 prior to data scaling, which used cell cycle score, mitochondrial gene expression level and the total unique molecular identifiers (UMIs) per cell as regression variables. Variable features were identified using the “FindVariableFeatures” function to select for 2000 most variably expressed genes. Principal-component (PC) analysis was then performed on log-transformed data using the “RunPCA” function, and the optimum number of PCs (n = 67) for downstream analysis was identified using molecular cross-validation (MCVR) in R⁵⁷. The neighbourhood graph was then computed using these and other default parameters and the graph embedded in two dimensions using uniform manifold approximation and projection. Clustering of single-cell data has been performed by Louvain community detection on the neighbourhood graph with default resolution set to 1.

Differential Gene Expression Analysis for single cell and visium data

Annotation of single cell clusters was performed by computing differentially expressed genes using the Wilcoxon rank sum test via Seurat’s “FindMarkers” function, with genes requiring a minimum of 10% expression within a cluster to be returned, and a log2FC threshold of 0.1. Adjusted p-value (post bonferroni correction) cut off was set to 0.05. Cells were annotated based on differential expression of classical genes previously identified in literature (Supplementary Data 1). Regional (cervical, thoracic, lumbar, sacral) differential gene expression was performed using the same method; dermal and cord/neuroglial cells were excluded from this analysis, as almost all of the former derived from the sacral region of a single fetus and the latter from PCW5 and 7, which could not be reliably dissected into precise anatomical sections. To analyse expression in stationary and neural crest-derived cell types in the Visium data, we classified voxels based on whether their dominant imputed cell types was NCC-derived or not before plotting gene expression using Seurat’s ‘DotPlot’ function and the Visutils version 0.2.1 R package’s “plotVisiumMultyColours” function (for visium data).

For detecting genes that exhibit variable expression in the axial plane, spatially variable gene expression was computed on Visium data by applying the “FindSpatiallyVariableFeatures” function to regions of interest within the slide, using the top 3000 variable genes as input. Features were then extracted using the “SVFInfo” function.

Co-expression of motor column HOX genes in the ventral cord was then evaluated using the “WhichCells” function, specifying the combination of relevant HOX genes (HOXC6 & HOXC9 for the thoracic region; HOXC9 & HOXC10 for the lumbar region) requiring non-zero expression for a voxel to be annotated as co-expressing.

To explore gene expression that varied with rostro-caudal position, single-cell clusters (resting chondrocytes and perichondrium) were collapsed into pseudo bulk RNA data by summing the raw expression for each gene in the cluster in cervical and lumbar regions, at each time point. These data were then analysed using DESeq2, following the standard pre-processing workflow prior to constructing a negative binomial model using anatomical location and batch as input variables⁵⁸. To elucidate genes that varied with location, a likelihood ratio test (LRT) was performed, with batch removed in the reduced model.

Calculating per-cell HOX:SOX10 ratio

To calculate the ratio of lumbar HOX genes to SOX10 mRNA counts, per-cell log-normalised expression values of HOX genes expressed in NCC-derived cells were divided by the value for SOX10, producing a per-cell ratio value. Results were visualised using the ggplot2 version 3.3.3 package in R. Significance was calculated using the Wilcoxon rank-sum test, two-sided.

Examining HOX regulons—DoRothEA

In order to test if genes that varied with rostro-caudal position were targets of HOX proteins, we accessed the R package DoRothEA version 3.19 regulon database, selecting confidence levels A-D; that is transcription factor targets with direct evidence of TF-target interaction. We then searched the database for the differentially expressed genes identified.

Deconvolution of Visium data—cell2location

To map cell types identified by scRNA-seq in the profiled spatial transcriptomics slides, we used the cell2location version 0.1.3 method¹⁷. In brief, this involved first training a negative binomial regression model to estimate reference transcriptomic profiles for all the cell types identified with scRNA-seq. Next, lowly expressed genes are excluded as per recommendations for use of cell2location. Next, we estimated the abundance of each cell type in the spatial transcriptomics slides using the reference transcriptomic profiles of different cell types. Results were visualised using the Visutils package for R.

Deconvolution of ISS data

To map cells types identified by scRNA-seq in the profiled ISS slides, we applied a nearest neighbour approach^59,60. The single-cell data was then subsetted to include only genes in the ISS panel. An annoy index was next constructed from the single cell data using the annoy version 1.0.3 package for python, with the metric specified as euclidean⁵⁹. This index was then used to calculate the K-nearest neighbour of each ISS cell in the single-cell dataset, and the identity of this neighbour applied to the ISS cell.

Gene expression analysis of ISS data

ISS data was analysed as per single-cell RNA sequencing data, using the following workflow: AnnData objects were first converted for use with the Seurat version 4.0.1 package for R, using the “convertFormat” function of the sceasy R package⁶¹. Any cells expressing fewer than two-panel genes were removed, and the data normalised and scaled using the “NormalizeData” and “ScaleData” Seurat functions, respectively. Gene expression was visualised using the “FeaturePlot” Seurat function. To identify muscle cells, tendon cells and neurons using panel genes, the “WhichCells” function was used, specifying scaled expression of the marker genes MYOD1, TNMD and NTRK1 as >0.1 to identify each of these cell types respectively. To exclude any NTRK1-expressing mesenchymal cells, DRG cells were first isolated anatomically using the lasso function in the webatlas version 0.2.1 browser and their barcodes exported, prior to subsetting the Seurat objects for these cells²². To examine co-expression of motor column HOX genes in the ISS samples, the ventral half of the cord was selected with the lasso function, and the Seurat object subsetted based on these cell barcodes. Co-expression was then evaluated using the “WhichCells” function, specifying the combination of relevant HOX genes requiring non-zero expression for a cell to be annotated as co-expressing (HOXC6 & HOXC9 for the thoracic region; HOXC9 & HOXC10 for the lumbar region). HOX gene modules (Fig. 2) were computed using the “AddModuleScore” seurat function. Data were visualised using the “DotPlot” and “FeaturePlot” Seurat functions.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

The new raw transcriptomic data generated in this study have been deposited in the EGA database under accession code EGAD00001009801 [https://ega-archive.org/studies/EGAS00001005090] and are accessible to researchers by application. The in situ sequencing data generated in this study have been deposited in the Bioimage archive database under accession code S-BIAD1117. Processed data can be downloaded and visualised at our data portal (https://spinal-development.cellgeni.sanger.ac.uk). Source data are provided with this paper.

Code availability

No previously unreported custom computer code or algorithms were used to generate results that are reported in this paper.

References

Sheng, G., Martinez, A. A. & Sutherland, A. The primitive streak and cellular principles of building an amniote body through gastrulation. Science 374, abg1727 (2021).
Article PubMed Google Scholar
Lewis, E. B. A gene complex controlling segmentation in Drosophila. Nature 276, 565–570 (1978).
Article ADS PubMed CAS Google Scholar
Harding, K., Wedeen, C., McGinnis, W. & Levine, M. Spatially regulated expression of homeotic genes in Drosophila. Science 229, 1236–1242 (1985).
Article ADS PubMed CAS Google Scholar
Soshnikova, N. & Duboule, D. Epigenetic temporal control of mouse Hox genes in vivo. Science 324, 1320–1323 (2009).
Article ADS PubMed CAS Google Scholar
Neijts, R. et al. Polarized regulatory landscape and Wnt responsiveness underlie Hox activation in embryos. Genes Dev. 30, 1937–1942 (2016).
Article PubMed PubMed Central CAS Google Scholar
Rekaik, H. et al. Sequential and directional insulation by conserved CTCF sites underlies the Hox timer in stembryos. Nat. Genet. 55, 1164–1175 (2023).
Article PubMed PubMed Central CAS Google Scholar
Garcia-Fernàndez, J. The genesis and evolution of homeobox gene clusters. Nat. Rev. Genet. 6, 881–892 (2005).
Article PubMed Google Scholar
Duboule, D. & Dollé, P. The structural and functional organization of the murine HOX gene family resembles that of Drosophila homeotic genes. EMBO J. 8, 1497–1505 (1989).
Article PubMed PubMed Central CAS Google Scholar
Pearson, J. C., Lemons, D. & McGinnis, W. Modulating Hox gene functions during animal body patterning. Nat. Rev. Genet. 6, 893–904 (2005).
Article PubMed CAS Google Scholar
Duboule, D. Temporal colinearity and the phylotypic progression: a basis for the stability of a vertebrate Bauplan and the evolution of morphologies through heterochrony. Dev. Suppl. 1994, 135–142 (1994).
Google Scholar
Morgan, R. & Whiting, K. Differential expression of HOX genes upon activation of leukocyte sub-populations. Int. J. Hematol. 87, 246–249 (2008).
Article PubMed PubMed Central Google Scholar
Kumar, V., Sekar, M., Sarkar, P., Acharya, K. K. & Thirumurugan, K. Dynamics of HOX gene expression and regulation in adipocyte development. Gene 768, 145308 (2021).
Article PubMed CAS Google Scholar
Aoyama, H. & Asamoto, K. Determination of somite cells: independence of cell differentiation and morphogenesis. Development 104, 15–28 (1988).
Article PubMed CAS Google Scholar
Kessel, M. & Gruss, P. Homeotic transformations of murine vertebrae and concomitant alteration of Hox codes induced by retinoic acid. Cell 67, 89–104 (1991).
Article PubMed CAS Google Scholar
Wilming, L. G., Boychenko, V. & Harrow, J. L. Comprehensive comparative homeobox gene annotation in human and mouse. Database (Oxford). bav091 https://doi.org/10.1093/database/bav091 (2015).
Sadler, T. W. Langman’s Medical Embryology. (Lippincott Williams & Wilkins, 2018).
Kleshchevnikov, V. et al. Cell2location maps fine-grained cell types in spatial transcriptomics. Nat. Biotechnol. 40, 661–671 (2022).
Article PubMed CAS Google Scholar
Rayon, T., Maizels, R. J., Barrington, C. & Briscoe, J. Single-cell transcriptome profiling of the human developing spinal cord reveals a conserved genetic programme with human-specific features. Development 148, dev199711 (2021).
Article PubMed PubMed Central CAS Google Scholar
Fukuoka, T. et al. Neurod4 converts endogenous neural stem cells to neurons with synaptic formation after spinal cord injury. iScience 24, 102074 (2021).
Article ADS PubMed PubMed Central CAS Google Scholar
Poliacikova, G., Maurel-Zaffran, C., Graba, Y. & Saurin, A. J. Hox Proteins in the Regulation of Muscle Development. Front. Cell Dev. Biol. 9, 731996 (2021).
Article PubMed PubMed Central Google Scholar
Sakka, L., Gabrillargues, J. & Coll, G. Anatomy of the Spinal Meninges. Oper. Neurosurg. 12, 168–188 (2016).
Li, T. et al. WebAtlas pipeline for integrated single cell and spatial transcriptomic data. Nat. Methods. 1–3 (2024) online ahead of print.
Nguyen, M. Q., von Buchholtz, L. J., Reker, A. N., Ryba, N. J. P. & Davidson, S. Single-nucleus transcriptomic analysis of human dorsal root ganglion neurons. Elife 10, e71752 (2021).
Article PubMed PubMed Central CAS Google Scholar
Rostock, C., Schrenk-Siemens, K., Pohle, J. & Siemens, J. Human vs. Mouse Nociceptors - Similarities and Differences. Neuroscience 387, 13–27 (2018).
Article PubMed CAS Google Scholar
Kildisiute, G. et al. Tumor to normal single-cell mRNA comparisons reveal a pan-neuroblastoma cancer cell. Sci. Adv. 7, eabd3311 (2021).
Article ADS PubMed PubMed Central CAS Google Scholar
Cheng, K. et al. The developmental origin and the specification of the adrenal cortex in humans and cynomolgus monkeys. Sci. Adv. 8, eabn8485 (2022).
Article ADS PubMed PubMed Central CAS Google Scholar
Zakany, J. & Duboule, D. The role of Hox genes during vertebrate limb development. Curr. Opin. Genet. Dev. 17, 359–366 (2007).
Article PubMed CAS Google Scholar
Elmentaite, R. et al. Single-cell sequencing of developing human gut reveals transcriptional links to childhood Crohn’s disease. Dev. Cell 55, 771–783 (2020).
Article PubMed PubMed Central CAS Google Scholar
Kawazoe, Y. et al. Region‐specific gastrointestinal Hox code during murine embryonal gut development. Dev. Growth Differ. 44, 77–84 (2002).
Article PubMed CAS Google Scholar
Sagner, A. & Briscoe, J. Establishing neuronal diversity in the spinal cord: a time and a place. Development 146, dev182154 (2019).
Article PubMed CAS Google Scholar
Graham, A., Maden, M. & Krumlauf, R. The murine Hox-2 genes display dynamic dorsoventral patterns of expression during central nervous system development. Development 112, 255–264 (1991).
Article PubMed CAS Google Scholar
Wilmerding, A. et al. HoxB genes regulate neuronal delamination in the trunk neural tube by controlling the expression of Lzts1. Development 148, dev195404 (2021).
Article PubMed CAS Google Scholar
Li, X. et al. Profiling spatiotemporal gene expression of the developing human spinal cord and implications for ependymoma origin. Nat. Neurosci. 26, 891–901 (2023).
Article PubMed PubMed Central CAS Google Scholar
Mouilleau, V. et al. Dynamic extrinsic pacing of the HOX clock in human axial progenitors controls motor neuron subtype specification. Development 148, dev194514 (2021).
Article PubMed PubMed Central CAS Google Scholar
Philippidou, P., Walsh, C. M., Aubin, J., Jeannotte, L. & Dasen, J. S. Sustained Hox5 gene activity is required for respiratory motor neuron development. Nat. Neurosci. 15, 1636–1644 (2012).
Article PubMed PubMed Central CAS Google Scholar
Wu, Y., Wang, G., Scott, S. A. & Capecchi, M. R. Hoxc10 and Hoxd10 regulate mouse columnar, divisional and motor pool identity of lumbar motoneurons. Development 135, 171–182 (2008).
Article PubMed CAS Google Scholar
Mendelsohn, A. I., Dasen, J. S. & Jessell, T. M. Divergent Hox Coding and Evasion of Retinoid Signaling Specifies Motor Neurons Innervating Digit Muscles. Neuron 93, 792–805 (2017).
Article PubMed PubMed Central CAS Google Scholar
Lacombe, J. et al. Genetic and functional modularity of Hox activities in the specification of limb-innervating motor neurons. PLoS Genet. 9, e1003184 (2013).
Article PubMed PubMed Central CAS Google Scholar
Kichaev, G. et al. Leveraging polygenic functional enrichment to improve GWAS power. Am. J. Hum. Genet. 104, 65–75 (2019).
Article PubMed CAS Google Scholar
Yi, M., Negishi, M. & Lee, S.-J. Estrogen Sulfotransferase (SULT1E1): Its Molecular Regulation, Polymorphisms, and Clinical Perspectives. J. Pers. Med. 11, 194 (2021).
Article PubMed PubMed Central Google Scholar
Xu, X. et al. Bioinformatics analysis on the differentiation of bone mesenchymal stem cells into osteoblasts and adipocytes. Mol. Med. Rep. 15, 1571–1576 (2017).
Article PubMed PubMed Central CAS Google Scholar
Chuang, P. T., Kawcak, T. & McMahon, A. P. Feedback control of mammalian Hedgehog signaling by the Hedgehog-binding protein, Hip1, modulates Fgf signaling during branching morphogenesis of the lung. Genes Dev. 17, 342–347 (2003).
Article PubMed PubMed Central CAS Google Scholar
Blain, E. J., Mason, D. J. & Duance, V. C. The effect of thymosin beta4 on articular cartilage chondrocyte matrix metalloproteinase expression. Biochem. Soc. Trans. 30, 879–882 (2002).
Article PubMed CAS Google Scholar
Garcia-Alonso, L., Holland, C. H., Ibrahim, M. M., Turei, D. & Saez-Rodriguez, J. Benchmark and integration of resources for the estimation of human transcription factor activities. Genome Res. 29, 1363–1375 (2019).
Article PubMed PubMed Central CAS Google Scholar
Vieille-Grosjean, I., Roullot, V. & Courtois, G. Lineage and stage specific expression of HOX 1 genes in the human hematopoietic system. Biochem. Biophys. Res. Commun. 183, 1124–1130 (1992).
Article PubMed CAS Google Scholar
Zhang, B. et al. A human embryonic limb cell atlas resolved in space and time. Nature 1–11 https://doi.org/10.1038/s41586-023-06806-x (2023).
Shin, M. M., Catela, C. & Dasen, J. Intrinsic control of neuronal diversity and synaptic specificity in a proprioceptive circuit. Elife 9, e56374 (2020).
Article PubMed PubMed Central CAS Google Scholar
Hilscher, M. M., Gyllborg, D., Yokota, C. & Nilsson, M. In Situ Sequencing: A High-Throughput, Multi-Targeted Gene Expression Profiling Technique for Cell Typing in Tissue Sections. Situ Hybrid. Protoc. 2148, 313–329 (2020).
Article CAS Google Scholar
GitHub - VasylVaskivskyi/microaligner: Image registration (alignment) software for large microscopy images. GitHub https://github.com/VasylVaskivskyi/microaligner.
Pachitariu, M. & Stringer, C. Cellpose 2.0: how to train your own model. Nat. Methods 19, 1634–1641 (2022).
Article PubMed PubMed Central CAS Google Scholar
Gataric, M. et al. PoSTcode: Probabilistic image-based spatial transcriptomics decoder. bioRxiv https://doi.org/10.1101/2021.10.12.464086 (2021).
Virshup, I., Rybakov, S., Theis, F. J., Angerer, P. & Wolf, F. A. anndata: Annotated data. bioRxiv https://doi.org/10.1101/2021.12.16.473007 (2021).
Stuart, T. et al. Comprehensive Integration of Single-Cell Data. Cell 177, 1888–1902 (2019).
Article PubMed PubMed Central CAS Google Scholar
Young, M. D. & Behjati, S. SoupX removes ambient RNA contamination from droplet-based single-cell RNA sequencing data. Gigascience 9, giaa151 (2020).
Article PubMed PubMed Central Google Scholar
Wolock, S. L., Lopez, R. & Klein, A. M. Scrublet: Computational Identification of Cell Doublets in Single-Cell Transcriptomic Data. Cell Syst. 8, 281–291 (2019).
Article PubMed PubMed Central CAS Google Scholar
Butler, A., Hoffman, P., Smibert, P., Papalexi, E. & Satija, R. Integrating single-cell transcriptomic data across different conditions, technologies, and species. Nat. Biotechnol. 36, 411–420 (2018).
Article PubMed PubMed Central CAS Google Scholar
Batson, J., Royer, L. & Webber, J. Molecular Cross-Validation for Single-Cell RNA-seq. bioRxiv https://doi.org/10.1101/786269 (2019).
Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 1–21 (2014).
Article Google Scholar
GitHub - spotify/annoy: Approximate Nearest Neighbors in C++/Python optimized for memory usage and loading/saving to disk. GitHub https://github.com/spotify/annoy.
Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 1–5 (2018).
Article Google Scholar
GitHub - cellgeni/sceasy: A package to help convert different single-cell data formats to each other. GitHub https://github.com/cellgeni/sceasy.

Download references

Acknowledgements

We thank Professor Denis Duboule and Professor James Briscoe for their invaluable advice for this study. This study was supported by the Wellcome Trust through institutional / programme grants and personal Fellowships (WT211276/Z/18/Z- institutional grant, 108413/A/15/D (S.B. & O.B.), 222902/Z/21/Z (S.B.) and 223135/Z/21/Z (J.E.L.).

Author information

These authors jointly supervised this work: Omer Bayraktar, Sam Behjati.

Authors and Affiliations

Wellcome Sanger Institute, Wellcome Genome Campus, Hinxton, CB10 1SA, UK
John E. G. Lawrence, Kenny Roberts, Elizabeth Tuck, Tong Li, Lira Mamanova, Pavel Mazin, Nathaniel D. Anderson, Liam Bolt, Laura Richardson, Elena Prigmore, Matthew D. Young, Sarah A. Teichmann, Omer Bayraktar & Sam Behjati
Department of Trauma and Orthopaedics, Cambridge University Hospitals NHS Foundation Trust, Addenbrooke’s Hospital, Box 37, Hills Road, Cambridge, CB2 0QQ, UK
John E. G. Lawrence
Department of Cellular and Molecular Pathology, Royal National Orthopaedic Hospital, Brockley Hill, Stanmore, HA7 4LP, UK
Petra Balogh
University College London Great Ormond Street Institute of Child Health, London, UK
Inga Usher & Alice Piapi
Department of Clinical Neurosciences, University of Cambridge, CB2 0QQ, Cambridge, UK
Xiaoling He & Roger A. Barker
Wellcome-MRC Cambridge Stem Cell Institute, Jeffrey Cheah Biomedical Centre Cambridge Biomedical Campus, Puddicombe Way, Cambridge, CB2 0AW, UK
Xiaoling He & Roger A. Barker
Research, Department of Pathology, University College London (UCL) Cancer Institute, London, WC1E 6DD, UK
Adrienne Flanagan
Department of Paediatrics, University of Cambridge, Cambridge, CB2 0QQ, UK
Sam Behjati

Authors

John E. G. Lawrence
View author publications
Search author on:PubMed Google Scholar
Kenny Roberts
View author publications
Search author on:PubMed Google Scholar
Elizabeth Tuck
View author publications
Search author on:PubMed Google Scholar
Tong Li
View author publications
Search author on:PubMed Google Scholar
Lira Mamanova
View author publications
Search author on:PubMed Google Scholar
Petra Balogh
View author publications
Search author on:PubMed Google Scholar
Inga Usher
View author publications
Search author on:PubMed Google Scholar
Alice Piapi
View author publications
Search author on:PubMed Google Scholar
Pavel Mazin
View author publications
Search author on:PubMed Google Scholar
Nathaniel D. Anderson
View author publications
Search author on:PubMed Google Scholar
Liam Bolt
View author publications
Search author on:PubMed Google Scholar
Laura Richardson
View author publications
Search author on:PubMed Google Scholar
Elena Prigmore
View author publications
Search author on:PubMed Google Scholar
Xiaoling He
View author publications
Search author on:PubMed Google Scholar
Roger A. Barker
View author publications
Search author on:PubMed Google Scholar
Adrienne Flanagan
View author publications
Search author on:PubMed Google Scholar
Matthew D. Young
View author publications
Search author on:PubMed Google Scholar
Sarah A. Teichmann
View author publications
Search author on:PubMed Google Scholar
Omer Bayraktar
View author publications
Search author on:PubMed Google Scholar
Sam Behjati
View author publications
Search author on:PubMed Google Scholar

Contributions

Study conception: J.E.L. and S.B. Data analysis and interpretation: J.E.L. with contributions from M.D.Y., S.B., O.B., N.D.A, P.M., S.A.T., A.F., T.L. and R.A.B. Experiments: K.R., E.T., J.E.L., P.B, I.U., A.P., L.M., L.B., L.R. and E.P. Tissue curation: X.H. Manuscript writing: J.E.L and S.B. with contributions from all authors. S.B. and O.B. co-directed the study.

Corresponding authors

Correspondence to Omer Bayraktar or Sam Behjati.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary Information (download PDF )

Peer Review file (download PDF )

Description of Additional Supplementary Files (download PDF )

Supplementary Data 1-7 (download XLSX )

Reporting Summary (download PDF )

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Lawrence, J.E.G., Roberts, K., Tuck, E. et al. HOX gene expression in the developing human spine. Nat Commun 15, 10023 (2024). https://doi.org/10.1038/s41467-024-54187-0

Download citation

Received: 05 January 2024
Accepted: 04 November 2024
Published: 20 November 2024
Version of record: 20 November 2024
DOI: https://doi.org/10.1038/s41467-024-54187-0

This article is cited by

Profiling the epigenomic landscape of late embryonic and adult mouse hind limb muscles
- Samantha R. Queeno
- Alexander S. Okamoto
- Kirstin N. Sterner
Scientific Reports (2026)
Advances in Spatial Transcriptomics in Bone
- Nico Valerio Giger
- Esther Wehrle
Current Osteoporosis Reports (2026)
When imaging technology meets single-cell omics: new paradigm in developmental biology
- Chunling Wang
- Xuejing Zhang
- Feng Liu
Cell Regeneration (2025)
Subgroup of meningiomas involving FOS and FOSB gene fusions
- Kanat Yalcin
- Hasan Alanya
- E. Zeynep Erson-Omay
Nature Communications (2025)