Image-based spatial transcriptomics identifies molecular niche dysregulation associated with distal lung remodeling in pulmonary fibrosis

Key results

1. ShownImaging-based profiling recovered 39 cell types, including disease-emergent states plus endothelial and mesenchymal cells that scRNA-seq tends to underrepresent. PF samples had lower proportions of alveolar cell types and higher proportions of activated fibroblasts, lymphoid cells and airway-type epithelium.

   1,126,820 nuclei holding 69,406,577 transcripts were kept, out of 210,370,118 high-quality transcripts across 28 samples. The 39 types split into 6 endothelial, 11 epithelial, 13 immune and 9 mesenchymal. propeller found that 24 of 39 types (61.5%) differed across the unaffected, LF and MF groups at FDR < 0.05. · Diverse cellular landscape of the lung; Dimensionality reduction, clustering, and cell…; Differential cell type composition by disease…

2. ShownSample-level pathology scores tracked both composition and expression: capillary and alveolar cells fell, and activated fibroblasts and lymphoid cells rose, with severity. Annotated lesions showed which cells make up fibroblastic foci, honeycombing and hyperplastic alveolar epithelium.

   In linear models, 19 cell types (48.7%) and 115 genes (33.5%) were associated with pathology score at FDR < 0.01. Authors annotated 21 feature types in 22 samples. Fibroblastic foci held CTHRC1+ and other fibroblast subtypes, smooth muscle cells and macrophages. Hyperplastic regions were mostly transitional AT2 cells. · Cellular and molecular characterization of…; Pathology score assignment; Annotation of histological features; Differential expression, cell type composition…

3. SuggestedTwo complementary niche methods each gave 12 niches: k-means on cell neighborhoods, and GraphSAGE transcript embeddings that need no cell segmentation. Healthy alveolar niches T4/C12 were depleted in PF, and less fibrotic regions resembled more fibrotic ones rather than controls.

   GraphSAGE was trained for ten epochs on a joint graph of 18,594,542 nodes and 421,316,086 edges (training accuracy 0.82). A GMM (k = 12) then clustered 206,449,064 transcripts. Cell niches used each cell's 25 nearest neighbors in Seurat v5. The LF versus MF similarity is only shown descriptively; no statistical test is reported. · Disease-specific cellular interactions are…; Transcript-based niches; Cell-based niches

4. SuggestedNiches T3 and C7 held most KRT5-/KRT17+ aberrant basaloid cells at fibrotic fronts next to CTHRC1+/FAP+ fibroblasts. They marked epithelium detaching from its basement membrane, and they led to detachment sites a clinician had not annotated plus structurally intact regions that may be sub-pathologic.

   In both analyses, annotated epithelial detachment was made up mainly of T3 (transcript-based) and C7 (cell-based). Cells in these niches appeared in LF, MF and intermixed PF samples but were virtually absent from controls. Detachment was also seen in adjacent sections that were not run on Xenium, although an ex vivo artifact cannot be fully excluded. · Disease-specific cellular interactions are…; Discussion

5. SuggestedRanking alveolar airspaces by composition suggested an order of remodeling. First came loss of capillary endothelium, AT1 cells, alveolar fibroblasts and interstitial macrophages, with a transient rise in AT2 cells. Activated fibroblasts appeared at the airway-like transition, then FABP4+ and later SPP1+ macrophages.

   slingshot ordered 1,233 alveolar airspaces on PCA of cell type proportions, and tradeSeq GAMs tested features along pseudotime. In cell type level GAMs across 24 cell types, early-transition hits were 37% epithelial and 15.8% AT2. Late-remodeling hits were 20% epithelial and 37% myeloid. 87 genes changed during the early transition. · A timeline of alveolar dysregulation; Ranking alveoli along a pseudotime by severity…; Assessing changes in cell types, niches, and…; Figure legends