A Reference Atlas of the Human Dorsal Root Ganglion

Abstract

Somatosensory perception largely emerges from diverse peripheral sensory neurons whose cell bodies reside in dorsal root ganglia (DRG). Damage or dysfunction of DRG neurons is a major cause of chronic pain and sensory loss. In mice, deep single-cell transcriptomic profiling and genetically defined models have offered important clues into DRG function, but in humans, the cellular and molecular landscape of DRG neurons remains less understood. Here, we constructed a reference cell atlas of the human DRG by profiling transcriptomes of cells and nuclei from 126 donors sampled across cervical, thoracic, and lumbar DRGs. This atlas resolves 22 neuronal subtypes, including known and previously unrecognized subtypes linked to nociception, mechanosensation, thermosensation, and proprioception, as well as 10 types of non-neuronal cells. Cross-species integration, spatial transcriptomics, and microneurography enabled cell-type-specific comparisons of soma size and conduction velocity between species. Human DRG somata are larger across all cell types than their mouse counterparts, and the conduction velocities of human hair follicle innervating A-fibers are faster than in mice, suggesting a functional shift in rapid mechanical detection in humans. This integrated human DRG reference cell atlas provides a resource for exploring new molecular and physiological features of human DRG, which could help identify new strategies for treating chronic pain and other diseases of the peripheral nervous system.

Figure 1.. The human DRG cell atlas

(A) Left: Diagram of DRG anatomy and major cell types. Center: Dissociated human DRG cells and nuclei were profiled using SMART-seq2, 10x Genomics 3’v3, and 10x Genomics FLEX platforms. Right: UMAP plots describing integrated DRG cell atlases comprising 53,596 neurons and 498,444 non-neuronal cells. (B) Top: bar plot displaying the number of DRG donors from which sc/snRNA-seq was performed since 2021, Bottom: line plot displaying the number DRG neurons (blue) and non-neurons (red) profiled by sc/snRNA-seq since 2021. Data from this study is shaded. C) Bar plots displaying the number of donors included in the DRG cell atlas, grouped by the spinal level (Left), sex (Center), or age (Right). Donors without a recorded spinal level or age were excluded from this plot.

Figure 2.. Transcriptional diversity of 22 human DRG neuronal subtypes

(A) Human DRG neuronal atlas (53,596 cells/nuclei) projected into UMAP space. Cells/nuclei are colored by their final cell type annotations. (B) UMAP projection displays the human DRG neuronal atlas, colored by sequencing platform (10x Genomics 3′ RNA: 9,125 nuclei; 10x Genomics FLEX: 40,917 nuclei; Soma-seq: 1,598 cells; and 10x Visium: 1,956 cells). (C) Dendrogram displays the hierarchical relationship of human sensory neurons. This hierarchal relationship was calculated based on the Euclidean distance of the principal components. Branch lengths are scaled to Euclidean distance. Dotted line indicates division between A-fibers and C-fibers. New cell types identified in this resource are colored in red. (D) Dot plot displays cell-type-specific marker gene expression in DRG neuronal subtypes. Dot size indicates the fraction of cells/nuclei expressing each gene, and color indicates average log-normalized scaled expression of each gene. (E) The heatmap displays the percentage of subtype annotations in the human DRG neuronal atlas that correspond to those independently identified in the soma-seq datasets (see Methods). Neuronal classifications differed for three subtypes: Aβ-LTMR.CCKAR was divided into two clusters in the human DRG neuronal atlas, while C-PEP.ADORA2B and C-Thermo.RXFP1 were identified only in the human DRG neuronal atlas. (F) The heatmap displays genes classified as neuropeptides, GPCRs, ion channels, and transcription factors that are enriched in each neuronal subtype relative to all other neuronal subtypes (log₂FC > 0.75, adj. p<0.05). For visualization, only the top 15 genes (or less) are displayed per subtype per classification.

Figure 3.. Non-neuronal cells in the human DRG

(A) Human DRG non-neuronal atlas projected into UMAP space (498,444 nuclei). Nuclei are colored by cell types. (B) Dot plot displays cell-type-specific marker gene expression in DRG non-neuronal subtypes. Dot size indicates the fraction of nuclei expressing each gene, and color indicates average log-normalized scaled expression of each gene. (C) Human DRG immune cells projected into UMAP space (59,469 nuclei). Nuclei are colored by immune subtypes. (D) Dot plot displays the expression of select marker genes for immune subtypes. Dot size indicates the fraction of nuclei expressing each gene, and color indicates average log-normalized scaled expression of each gene. (E) and (G) Bar plots display the GO terms enriched in genes that exhibit significant ligand-receptor interactions between neurons and immune cells (E) or neurons and other non-neuronal cells (G) compared to all other ligand-receptor pairs. (F) Dot plot displays the gene expression of ligand C3 and its receptors NRP1 and CD46 in individual neuronal and immune subtypes. Dot size indicates the fraction of nuclei expressing each gene, and color indicates average log-normalized scaled expression of each gene. The arrows link the top 5 (ranked by aggregated rank) cell type pairs for each ligand-receptor interaction. (H) Dot plot displays the gene expression of ligand NCAM1 and its receptors, including GFRA1 and PTPRA , in individual neuronal subtypes and non-neuronal cell types. Dot size indicates the fraction of nuclei expressing each gene, and color indicates average log-normalized scaled expression of each gene. The arrows link the top 5 (ranked by aggregated rank) cell type pairs for each ligand-receptor interaction. (I) Heatmap displays genes that are differentially expressed (log₂FC>1, FDR<0.05) in each non-neuronal subtype relative to all other non-neuronal subtype. For each gene, cell types with expression in less than 5% of the nuclei are colored dark blue.

Figure 4.. Spatially defined DRG neuronal subtypes

(A) and (B) Representative Xenium (A) or MERSCOPE (B) spatial transcriptomics images of human DRG. Segmented neuronal regions are colored by their transcriptionally defined neuronal subtype. Scale bar: 2 mm (Xenium) and 1 mm (MERSCOPE). (C) Heatmap displays correlations between subtype specific markers from the DRG neuronal atlas and the spatial transcriptomic datasets (merged Xenium and MERSCOPE). Heatmap is colored by Spearman’s r. (D) Bar plot displays percentage of neuronal subtypes detected in each spatial platform. Xenium and MERSCOPE data were merged. Bars are colored by fiber class. Xenium (1186 cells from 1 donor) and MERSCOPE (619 cells from 1 donor) data were merged. (E) Box plot displays soma diameters of each neuronal subtype. Boxes are colored by fiber class. Xenium (1186 cells from 1 donor) and MERSCOPE (619 cells from 1 donor) data were merged.

Figure 5.. Variation in neuronal subtype composition across spinal levels, sex, and age.

(A) Bar plots display proportional distributions of neuronal subtypes across in, thoracic, or lumbar DRGs (± SEM across donors). Bars are colored by neuronal subtype and ordered by median abundance across all spinal regions. (B) Box plots display the proportions of A-Propr.HAPLN4 and A-Propr.EPHA3 subtypes in cervical, thoracic, and lumbar DRGs. Each dot represents one donor. Donors with <1,000 neuronal nuclei profiled by sc/snRNA-seq were excluded. Significance was assessed using a beta regression model (*: p < 0.05; **: p < 0.01; ***: p < 0.005). (C) Violin plots of gene module scores associated with DRG neurons from each spinal level. Modules were defined by genes enriched in all neurons from a given region relative to other regions (log₂FC > 1; adj. p < 0.05; Table S7). (D) Violin plots display log-normalized expression levels of CACNA1A, P2RX7, and KCNK2, separated by spinal region (cervical, thoracic, or lumbar). (E) Bar plot displays the mean difference (female – male) in subtype proportions between sexes. Donors with <1,000 neuronal nuclei were excluded. Bars are colored by sexes. F: female; M: male; *: p < 0.05; **: p < 0.01. (F) Bar plot displays the mean difference in neuronal subtype proportions between donors younger than 45 years and those 45 years or older. Donors with <1,000 neuronal nuclei profiled by sc/snRNA-seq were excluded. Bars are colored by age (<45 or >45). *: p < 0.05.

Figure 6.. Cross-species analysis of mouse and human DRG neurons.

(A) Dendrogram displays cell type similarities of mouse and human DRG neurons after cross-species integration by canonical correlation analysis. This hierarchical relationship was calculated based on the Euclidean distance of the principal components. Branch lengths are scaled to Euclidean distance. (B) Heatmap displays z-scored cell type classification probabilities from the variational autoencoder model between mouse (columns) and human (rows) cell types. (C) Table displays human DRG neuronal subtypes with their corresponding mouse DRG neuronal subtypes. For each human cell type (rows), we report the corresponding mouse cell type based on the nomenclature reported by Krauter et al. (2025), Qi et al. (2024), and Bhuiyan et al. (2024). (D) Box plot displays the distribution of somata sizes per neuronal subtype class for human and mouse. Human somata sizes are calculated from spatial transcriptomics data presented in Figure 4, while mouse data are calculated from the recent Krauter et al., 2025 study. Boxes are colored and split by species, and subtypes are merged into physiological classes. (E) Box plot displays distribution of soma sizes for Aβ-LTMR.NSG2 and Aβ-LTMR.CCKAR, along with their transcriptomic correlate in mouse, Aβ-Field and Aδ-LTMR. Human somata sizes are calculated from spatial transcriptomics data presented in Figure 4, while mouse data are calculated from the recent Krauter et al., 2025 study. Boxes are colored by species. Statistical significance was calculated using a Wilcoxon rank sum test (p-value > 0.05 = “n.s”, p-value < 0.001 = “***”).

Figure 7.. Physiological properties of human A-LTMR subtypes.

(A) Plot displays the recorded conduction velocities of SA1, SA2, A-LTMR(HF), and Field receptors (mean ± SEM). Each point represents an individual neuron; circles, upper limb (n=130 neurons); triangles, lower limb (n=48 neurons). Conduction velocities did not differ among subtypes in the upper limb (Kruskal–allis, P = 0.0922) or lower limb (P = 0.0836). HF: hair follicle. (B) Top left: Traces display representative A-LTMR(HF) spike activity in response to an air puff. Bottom left: Trace displays instantaneous spike frequency. Top right: Trace displays spike activity recorded after shaving the receptive field. Bottom right: Trace displays instantaneous spike frequency (C) Traces display representative A-LTMR(HF) spike activity in response to single-hair deflection (CV = 40.2 m/s). Top: Traces display raw spike trace; Bottom: Trace displays instantaneous spike frequency. (D) Top: Traces display representative A-LTMR(HF) spike activity in response to brushing at 10 cm/s in both directions. Bottom: Trace displays instantaneous spike frequency. (E) Box plot displays mean firing frequencies of A-LTMR(HF)s to brushing at 1, 3, and 10 cm/s, grouped by directional preference. For each recorded unit, brushing directions (proximal-distal and distal-proximal) were reassigned as preferred or non-preferred, with preference defined as a ≥20% higher mean firing frequency than the opposite direction across velocities. Direction preference was stable within each unit. Each circle represents an individual trial (n=3 neurons, multiple trials per neuron). Responses were significantly greater in the preferred direction (mixed-effects model, P < 0.0001). Mean ± SEM., n.s.: not significant. *P < 0.05, ***P < 0.001, ****P < 0.0001. (F) Traces display representative C-LTMR spike activity in response to single-hair deflection. Top: Traces display raw spike trace; Bottom: Trace displays instantaneous spike frequency. (G) Top: Traces display representative C-LTMR spike activity in response to brushing at 1 cm/s in both directions. Bottom: Traces display application of stimuli, in both directions. (H) Bar plot displays mean firing frequencies of C-LTMR to brushing at 1, 3, and 10 cm/s, comparing proximal-distal and distal-proximal strokes. Anatomical directions are shown, as no consistent direction preference was observed across units. Each circle represents an individual trial (n=5 neurons, multiple trials per neuron). C-LTMR responses were not modulated by brushing direction (2-way RM-ANOVA, P = 0.6501). Data are mean ± s.e.m.; n.s.: not significant.