henhaobio
Sign in

bioRxivImmunologyPreprint

Layer-to-cell-class correspondence is gene-specific for pyroptosis-related transcription in human skin

Daisuke OryojiiD, Goro Doi, Sho FujimotoiD, Naoya NishimuraiD, Kyoko Otsuka, Ayako Kuwahara, Masahiro AyanoiD, Yasutaka KimotoiD, Koichi AkashiiD, Hiroaki NiiroiD, Hiroki MitomaiD

PDFDOI
In-depth full-text digest available · Jump to digest ↓

Abstract

Anatomical-layer enrichment does not establish cellular attribution, so correspondence with major-cell-class expression must be tested gene by gene. Systemic sclerosis (SSc) skin is informative because inflammasome- and pyroptosis-related molecules have been reported across epidermal, fibroblast-associated and vascular contexts. Spatial transcriptomics of SSc and control skin, with same-donor single-nucleus profiles from 10 SSc donors, revealed gene-specific correspondence. Across 13 discovery sections, NLRP1, PYCARD and CASP4 were epidermally biased, whereas GSDMD was dermally biased. An independent 10-section SSc cohort supported epidermal NLRP1 (q = 0.0078) and PYCARD (q = 0.043), with directional, non-significant support for CASP4 and GSDMD. Despite epidermal NLRP1 bias, endothelial class-pseudobulk exceeded keratinocyte class-pseudobulk in all 10 donors. Restricted three-class reconstruction reproduced the NLRP1 direction in 1 and 0 of 10 donors under the two fixed dermal weightings and in 0-8 of 10 across a post hoc sweep of the dermal endothelial share; PYCARD and GSDMD were concordant in most donors at every weighting, and CASP4 in no more than 4 of 10. Thus, tissue layer and major cell class are gene-specifically non-equivalent axes of spatial interpretation.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (23 Sept 2026) · CC BY 4.0

Epidermal NLRP1, PYCARD and CASP4 versus dermal GSDMD polarity in systemic sclerosis skin is reproduced by same-donor cell-class pseudobulk for PYCARD and GSDMD, but rarely NLRP1.

Spatial analyses often treat enrichment in an anatomical compartment as evidence of which cells express a gene, without checking matched cell-class data; in systemic sclerosis (SSc) skin, inflammasome molecules have been linked to epidermal, fibroblast and vascular settings. Pairing Visium and single-nucleus data from the same donors, the study tests this attribution gene by gene and finds it holds for some pyroptosis transcripts but fails for others.

Key results

  1. ShownDiscovery sections showed opposing layer polarity: NLRP1, PYCARD and CASP4 were epidermal and GSDMD was dermal in both healthy and SSc skin, while NLRP3, MEFV and AIM2 leaned dermal, so the pathway is not one uniform spatial program.

    13 discovery sections (4 HC, 9 SSc), exploratory BH m=6 screen: PYCARD, NLRP1 and CASP4 epidermal in 13 of 13; GSDMD dermal in 13 of 13 (Cohen's d, dermis minus epidermis); NLRP3 dermal in 12 of 13; broader inflammasome composite showed 60.8% descriptive post-selection dilution relative to the triad. · Discovery identifies opposing tissue-layer…; Pathway selectivity and composite dilution

  2. ShownIn an independent SSc Visium cohort, direct gene-level tests supported epidermal NLRP1 and PYCARD after multiple-testing correction; CASP4 and GSDMD kept their expected directions without significant adjusted support.

    GSE312129, section-level exact sign tests, BH m=4: NLRP1 epidermal 10 of 10 (q=0.0078), PYCARD 9 of 10 (q=0.043), CASP4 7 of 10 (q=0.34), GSDMD dermal 8 of 10 (q=0.15); 270 endothelium- and 909 other-labelled spots excluded; moving endothelium spots into dermis left every direction count unchanged. · Direct four-gene replication establishes…; Direct gene-level replication deepening

  3. ShownSame-donor class pseudobulk matched layer direction for PYCARD and GSDMD but contradicted it for NLRP1: despite epidermal NLRP1 polarity, endothelial pseudobulk exceeded keratinocyte pseudobulk in every SSc donor. CASP4 followed its layer direction in only a minority of donors.

    10 SSc donors, log2(CPM+1), at least 20 nuclei per class: keratinocyte minus fibroblast/endothelial_strict mean followed layer direction for PYCARD and GSDMD in 10 of 10 (q = 0.0039 each) and CASP4 in 3 of 10 (q = 0.34); NLRP1 ran opposite in 9 of 10 (q = 0.029), median K−E −0.915 (keratinocyte lower in all 10). · Same-donor major-cell-class context is…; Same-donor snMultiome pseudobulk

  4. SuggestedA restricted keratinocyte/fibroblast/endothelial proxy reproduced PYCARD and GSDMD directions in most donors at every dermal weighting and CASP4 in a minority. It reproduced NLRP1 in most donors only at low dermal endothelial shares (0 to 0.10), and in at most 1 of 10 under either fixed weighting.

    Exact sign concordance, M1/M2: NLRP1 1/0 of 10, PYCARD 9/9, CASP4 4/3, GSDMD 8/8. Post hoc endothelial-share sweep: NLRP1 8, 7 and 7 of 10 at 0, 0.05 and 0.10, 4 of 10 at 0.20, none at 0.5 or above; at the discovery-derived share 0.0965: 7, 9, 2 and 8 of 10 for NLRP1, PYCARD, CASP4, GSDMD. · Restricted same-donor reconstruction reveals…; Composition adjustment and restricted mixture…

  5. SuggestedIn discovery sections, dermal GSDMD rose with Cell2location-estimated endothelial context and fell with estimated fibroblast context. A small association between an interferon-gamma-response score and GSDMD remained after endothelial adjustment.

    13 discovery sections, frozen four-test family: GSDMD vs endothelial meta rho=0.181, positive in 13 of 13 (q=2.9×10^−7); vs fibroblast meta rho=−0.166, negative in 12 of 13 (q=9.5×10^−5). Post hoc IFN-γ score vs GSDMD partial rho median 0.065 (95% CI 0.034 to 0.114; positive in 12 of 13; Wilcoxon P=0.00049). · Endothelial and interferon-gamma context; Replication endpoints, deconvolution, and…

Methods, data and reproducibility

Reanalysis of public Visium skin sections using computed epidermis/dermis labels and sectionwise sign tests, Cell2location deconvolution, and same-donor snMultiome pseudobulk for keratinocyte, fibroblast and endothelial nuclei. These were mixed in a restricted three-class proxy to test whether class expression reproduces each gene's layer direction.

  • Discovery Visium FFPE skin sections reanalyzed from Li et al., healthy control and systemic sclerosis · human · n = 13 sections (4 HC, 9 SSc) · Zenodo doi: 10.5281/zenodo.14577696
  • Whitfield replication Visium sections from SSc skin · human · n = 10 sections from 10 SSc donors · GEO GSE312129
  • Whitfield single-nucleus multiome reference from the same SSc donors as the replication Visium; used for Cell2location and class pseudobulk · human · n = 65,066 nuclei; 4 control and 10 SSc participants · GEO GSE312129
  • Released analysis code, selected intermediate values and environment snapshots (version 1.1.0) · Zenodo doi: 10.5281/zenodo.22215741
Methods & tools
Visium spatial transcriptomics, Single-nucleus multiome (snMultiome), Cell2location, CellTypist, Leiden clustering, Multinomial logistic regression (layer label refinement), Cohen's d, Exact two-sided sign test, Wilcoxon signed-rank test, Partial Spearman correlation, Benjamini-Hochberg FDR, Variance inflation factor and condition number diagnostics
Reagents & environment
Environment snapshots are deposited with the Zenodo code release (version 1.1.0); the text does not list individual software versions.
Reproducibility gaps
Code and data are cited only as DOIs or a GEO accession, with no URLs. Some plotting and sensitivity code is only in Supplementary Software 1. Processed count matrices were reused without rerunning FASTQ processing.
Limitations (6)
  • Visium and snMultiome are distinct mRNA-level platforms. Differences in nuclear recovery, tissue dissociation and per-gene capture could shift class-pseudobulk ordering relative to the spatial signal. · stated by the authors
  • The proxy uses only three classes, does not measure Visium spot composition and assumes the dermal fibroblast:endothelial weighting. NLRP1 concordance depended on that assumption, and omitted immune, perivascular or appendage populations could drive discordance. · stated by the authors
  • Both layer labels and abundance estimates come from expression data. The same single-nucleus object served earlier as the deconvolution reference in discovery, so these analyses cannot count as independent biological confirmation. Replication labels were not checked against histology. · stated by the authors
  • Epidermal location and keratinocyte abundance were collinear (maximum VIF 15.5; 5 of 13 sections failed the gate). As a result, simultaneous composition adjustment could not be estimated, and the attenuation-prone two-stage q of 1.0 cannot exclude layer polarity. · stated by the authors
  • No protein-level data: NLRP1/ASC localization, GSDMD cleavage, pore formation and executed pyroptosis were not assessed, so cellular origin and causality remain unresolved. · stated by the authors
  • The replication Visium contains only SSc sections, and pseudobulk for the 4 HC donors was descriptive only. It is therefore untested whether the mismatch between NLRP1 layer direction and cell-class expression also holds in healthy skin. · noted in this summary

Read this if you infer which cell types drive a spatial compartment signal from layer enrichment, or if you study inflammasome and pyroptosis transcription in systemic sclerosis skin.

Citations
0

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISAbel '24Tabib '21Dunlap '21Marella '25Li '25Makinde '22
57% semantic matchGenomics
Tamar R. Abel, Noelle N. Kosarek, Rezvan Parvizi+13 more·bioRxiv·Mar 22, 2024· 3 citations