The dorsal root ganglion (DRG) is a peripheral sensory hub containing primary afferent neuron cell bodies and a tightly interwoven assemblage of neuronal, glial, vascular, and immune elements. In clinical practice the DRG is a frequent target for local pain-management interventions, including nerve root blocks, epidural injections, and neuromodulation, often using corticosteroids such as dexamethasone for their broad anti-inflammatory action. However, the fundamental cellular and spatial organization of immune transcriptional states within human DRG tissue has not been well defined. Whether DRG immune activity aligns with a single dominant inflammatory axis or instead comprises multiple, spatially heterogeneous transcriptional programs remained an open question.
Advances in single-nucleus RNA sequencing (snRNA-seq) and spatial transcriptomics enable combined interrogation of transcriptomic heterogeneity and tissue context. Single-cell approaches allow unsupervised identification of transcriptional programs but lose anatomical localization; spatial transcriptomics preserves architecture but requires defined gene programs for projection-based analysis. Integrating these modalities permits characterization of transcriptional heterogeneity and spatial patterning without presupposing discrete polarization states. This study applied a rank-stable non-negative matrix factorization approach to human DRG snRNA-seq to define immune metaprograms, then projected those programs onto Xenium spatial transcriptomics data to test for spatially localized micro-niche organization. Reproducibility was assessed in an independent human DRG snRNA-seq dataset.
Study design and analytical workflow
The investigation was a mechanistic transcriptomic analysis using only publicly available, de-identified human datasets. The workflow had three major steps: (1) identify immune transcriptional metaprograms from human DRG snRNA-seq (GSE189501) using rank-stable non-negative matrix factorization (NMF); (2) project the resultant metaprograms onto human DRG Xenium spatial transcriptomics data (GSE273557) to evaluate spatial localization; and (3) assess reproducibility by projecting program signatures onto an independent human DRG snRNA-seq dataset (GSE168243). No new human subject research was performed.
Computational environment and reproducibility
All analyses were performed locally in R (version 4.4.3) with standard packages: Seurat for preprocessing and module scoring; NMF for program decomposition and rank surveys; ComplexHeatmap, fgsea and msigdbr for visualization and enrichment; Arrow for Xenium parquet handling; spatstat.geom and FNN for spatial processing. Random number generator seeds were set explicitly for NMF, rank surveys, sampling, and permutation tests to ensure deterministic reproducibility. Intermediate R objects, including Seurat objects after QC and program scoring, were saved as RDS files.
Data selection and program derivation
A curated subset of 388 immune nuclei was extracted from the discovery snRNA-seq dataset to minimize lineage contamination. Rank selection for NMF was surveyed across k = 4–9; consensus stability metrics supported selection of k = 7. Program characterization combined gene loading patterns, differential expression between high- and low-scoring cells, and Hallmark gene set enrichment analyses.
Spatial projection and aggregation testing
Program gene sets were projected onto Xenium spatial transcriptomics data by recalculating program scores from spatial transcript counts. Region-specific program enrichment was evaluated, and spatial aggregation of program-high cells was quantified using nearest-neighbor distance analysis with permutation-based significance testing to assess non-random clustering.
Validation
To evaluate robustness, the program gene signatures were projected onto an independent human DRG snRNA-seq dataset (GSE168243) to assess reproducibility of the inferred metaprogram architecture.
Unsupervised NMF decomposition of the curated immune nuclei identified seven reproducible transcriptional metaprograms (k = 7). These metaprograms did not converge on a single dominant inflammatory signature. Instead, they represented parallel transcriptional states with distinct gene-loading patterns and functional annotations derived from differential expression and Hallmark pathway enrichment.
Among the identified programs, one displayed receptor- and sensing-associated transcriptional features without strong enrichment for classical inflammatory pathways, highlighting transcriptional programs related to detection and response rather than overt inflammation. Projection of program gene signatures onto Xenium spatial transcriptomics data revealed region-specific enrichment of program-high cells within the DRG tissue sections.
Nearest-neighbor distance analysis with permutation testing demonstrated statistically significant spatial aggregation for selected metaprograms in specific tissue regions, supporting localized micro-niche organization. The projection onto the independent snRNA-seq dataset provided a reproducibility assessment; program gene signatures were tested across cohorts to evaluate whether the metaprogram architecture generalizes beyond the discovery cohort.
The integrative analysis indicates that immune organization within the human DRG is composed of multiple parallel transcriptional programs that are spatially localized rather than reflecting a single tissue-wide inflammatory axis. This contrasts with some central nervous system paradigms—where microglial activation and Apoe-associated states are emphasized—and reflects the distinct immune composition and tissue architecture of the peripheral DRG, which lacks microglia and contains macrophages, satellite glia, Schwann cells, endothelium, and stromal components within a relatively permeable vascular environment.
By combining rank-stable NMF on snRNA-seq with projection onto Xenium spatial transcriptomics, the study provides a spatially grounded framework for understanding DRG immune architecture. The observed micro-niche enrichment patterns suggest biological contexts in which DRG-targeted local interventions act, and caution against assuming uniform anti-inflammatory effects across the tissue.
This investigation used exclusively publicly available human datasets and computational inference; it did not include experimental perturbation or direct functional validation of the inferred metaprograms. Details beyond those reported in the analyzed sources—such as cell-type–level functional assays or clinical outcome correlations—were not performed. The study's conclusions rely on the quality and scope of the underlying GEO datasets and on the assumptions inherent to projection-based spatial analyses.
Human DRG immune organization reflects multiple, parallel transcriptional metaprograms with spatially localized enrichment, consistent with a micro-niche model rather than uniform tissue-wide activation. These findings provide a transcriptomic and spatial framework that may inform mechanistic interpretation of DRG-targeted local interventions. Future work should extend these observations with experimental validation, functional assays of specific programs, and correlation with clinical interventions and outcomes.
Data availability and reproducibility
All analyzed datasets are publicly available via GEO (GSE189501, GSE273557, GSE168243). Raw Xenium outputs, including histology images and parquet-formatted tables, were accessed through the corresponding GEO record. Computational analyses were performed in R 4.4.3 with explicit RNG seeds and intermediate objects saved to support reproducibility.