Spatial Transcriptomics Maps the Cellular Battlefields of Fatal COVID-19 Lungs

When severe COVID-19 turns fatal, the lung undergoes diffuse alveolar damage, a progressive injury process that begins with inflammation and fluid flooding the air sacs and ends, in many cases, with dense scarring that destroys the tissue architecture needed for gas exchange. For years, researchers have tried to reconstruct the molecular choreography of this decline, but conventional methods have a fundamental flaw: they grind tissue to a pulp before reading its genes, erasing the very spatial relationships that define how a lung works. A new study published in Nature Communications by Carlos A. Garcia-Prieto, Manel Esteller and colleagues has now overcome that limitation by applying Visium spatial transcriptomics to post-mortem lung samples, producing the most detailed in situ map yet of the cellular and molecular events that drive lethal SARS-CoV-2 disease.

The team analyzed twenty-three formalin-fixed, paraffin-embedded lung tissue samples from nineteen patients who died with COVID-19-associated diffuse alveolar damage, alongside four control lungs with normal morphology. The samples spanned the two canonical histological stages of the disease: seven cases of acute diffuse alveolar damage, marked by early epithelial lesions, interstitial inflammation and edema, and twelve cases of proliferative damage, characterized by pneumocyte hyperplasia and fibroblast proliferation. Using the 10x Genomics Visium platform, which captures gene expression on a slide dotted with thousands of barcoded spots each covering roughly one to ten cells, the researchers profiled a total of 91,068 tissue spots after quality control. To identify which cell types occupied each spot, they used the cell2location algorithm to deconvolve the spatial data against the integrated Human Lung Cell Atlas, a single-cell reference containing hundreds of thousands of lung cells, resolving forty-five distinct cell populations across four major lineages: epithelial, endothelial, immune and stromal.

The resulting cellular census revealed a striking, directional remodeling of the dying lung. In healthy tissue, endothelial cells are the most abundant lineage, and capillary cell types alone account for roughly thirty percent of all lung cells. As diffuse alveolar damage progressed from acute to proliferative stages, the researchers documented a progressive loss of endothelial populations, including aerocyte capillary, arterial, general capillary and venous pulmonary endothelial cells, a finding that aligns with growing evidence linking SARS-CoV-2 infection to endothelial dysfunction. The only shift that reached statistical credibility by the Bayesian compositional analysis scCODA, at a false discovery rate below 0.05, was the reduction of endothelial cells in proliferative disease compared with controls, but the trend was consistent across the entire endothelial compartment. On the epithelial side, alveolar type 1 cells, which normally form the thin blood-gas barrier, were significantly depleted in proliferative damage, while alveolar type 2 cells, the progenitors that repair injured alveoli, expanded. This inverse relationship captures the essence of the pathology: the fragile gas-exchange surface is destroyed while the repair machinery struggles to compensate.

The immune landscape shifted in parallel. Alveolar macrophages, CCL3-positive macrophages and monocyte-derived macrophages were significantly overrepresented in proliferative disease, and interstitial perivascular macrophages rose steadily across stages. In contrast, the lymphoid compartment contracted: CD4 and CD8 T cells declined, and natural killer cells showed a significant drop from control and acute samples to proliferative cases. Because natural killer cells are innate effectors central to the response against acute viral infection, their progressive disappearance reinforces prior reports tying NK cell dysfunction to COVID-19 severity and lethality. Among stromal cells, the most dramatic change belonged to peribronchial fibroblasts, which increased from control to acute damage and then skyrocketed in the proliferative stage. Other fibroblast subsets behaved differently: subpleural and adventitial fibroblasts expanded, while alveolar fibroblasts and pericytes, which contribute to tissue homeostasis, decreased. The specificity of the peribronchial fibroblast expansion would prove to be one of the study’s most consequential observations.

Beyond counting cells, the study interrogated how they were arranged. Non-negative matrix factorization of cell-type abundance estimates identified eight recurring modules of co-occurring cell types, or spatial compartments. In normal lung, these corresponded to physiological microenvironments, such as a large immune-endothelial compartment, an epithelial compartment, and a smooth-muscle and fibroblast compartment in which peribronchial fibroblasts resided in isolation. With acute damage, this architecture fractured: proliferating alveolar type 2 cells were recruited into the epithelial compartment, macrophage subtypes and type-2 dendritic cells formed a new immune cluster, and plasma cells appeared as an isolated population. In proliferative damage, the compartmentalization deepened further, with mature lymphatic endothelial cells emerging as a unique niche. A graph-based comparison method called GraphCompass quantified these architectural shifts, showing that peribronchial fibroblasts, alveolar fibroblasts, alveolar type 2 cells, aerocyte capillary endothelial cells and monocyte-derived macrophages underwent the most pronounced spatial remodeling. Filtration curve analysis confirmed the picture: peribronchial fibroblasts and alveolar type 2 cells clustered ever more tightly in proliferative disease, while capillary endothelial cells became fragmented and dispersed as they were depleted.

The researchers then turned to intercellular communication, using spatially weighted ligand-receptor interaction scores decomposed into three communication signatures by non-negative matrix factorization. The third signature best discriminated diseased from healthy lung and was significantly enriched for interactions belonging to the transforming growth factor beta pathway, the master driver of fibrosis, while being depleted for the epidermal growth factor receptor pathway. The top ligand-receptor pairs defining this signature included TIMP1 binding CD63, an interaction known to promote lung fibrosis through the TGF-beta1/SMAD3 axis; amyloid precursor protein binding CD74, implicated in capillary barrier defense during infection; CD99 with CD81, both regulators of lymphocyte activity; and lumican binding integrin beta 1, linked to extracellular matrix remodeling. Additional prominent interactions involved S100A8 and S100A9 binding their receptors, known to activate macrophages in COVID-19, SPARC binding endoglin, and vimentin binding CD44, an attachment factor that facilitates SARS-CoV-2 entry into endothelial cells. The three most frequent receptors in the diseased-lung interactome were CD44, integrin beta 1, which associates with ACE2 to mediate viral entry, and LRP1, previously enriched in mortal and long COVID-19 cases.

Inside the cells, transcription factor activity analysis using the decoupleR framework and the CollecTRI regulatory network revealed the same profibrotic logic operating at the level of gene regulation. Proliferative damage was characterized by upregulation of SMAD3, the key mediator of TGF-beta signaling that drives extracellular matrix production and scar formation, coupled with downregulation of SMAD7, the pathway’s natural antagonist. This SMAD3-high, SMAD7-low axis proved remarkably disease-specific: when the researchers validated their findings against the extended Human Lung Cell Atlas, spanning more than 1.4 million cells from healthy lungs, non-COVID-19 pneumonia, idiopathic pulmonary fibrosis and chronic obstructive pulmonary disease, SMAD3 ranked as the second most upregulated and SMAD7 the second most downregulated transcription factor in COVID-19 relative to healthy lung, a pattern absent in the other conditions. Spatially, peribronchial fibroblast abundance co-clustered with SMAD3 activity and segregated from SMAD7 activity, and the TIMP1-CD63 interaction showed the strongest spatial co-localization of any ligand-receptor pair, particularly within proliferative lesions. A predictive model linking communication and transcriptional signals to cell abundance explained a mean of 21 percent of the variance in cell-type composition, with transcription factor activity contributing roughly twice as much as ligand-receptor interactions.

By inferring a putative causal signaling network within peribronchial fibroblasts, the team traced a pathway with direct therapeutic implications. Activation of the integrin ITGB1 receptor unleashed downstream kinases including ILK, AKT1, MAP2K4 and MAPK14, ultimately activating SMAD3, MYB and BRCA1 while repressing SMAD7 and CIITA, the latter repression potentially unleashing collagen expression after injury. Because MAPK14 inhibitors such as ralimetinib and ARRY-797 have shown antiviral activity against SARS-CoV-2, and because the approved antifibrotic drugs pirfenidone, which inhibits TGF-beta, and nintedanib, which blocks multiple kinases, are already used in idiopathic pulmonary fibrosis, the authors argue that their spatially resolved pathway map points to concrete candidates for repurposing against COVID-19-associated fibrosis. They also flag NLRP3 inflammasome inhibitors for the hyperactivated myeloid compartment, MYB and the S100A9-CD68 interaction as myeloid targets, and even speculate that amyloid precursor protein interactions in the injured lung might connect pulmonary damage to the cognitive impairment seen after COVID-19, suggesting dual-action strategies such as lysophosphatidic acid receptor inhibitors.

The study closed with two forward-looking analyses. A graph neural network approach showed that natural killer cells in healthy lung depend on multiple neighboring cell types, including CD8 T cells, aerocyte capillary endothelial cells and alveolar type 1 cells, dependencies that vanished entirely in diseased tissue, while CD4 T cells became prominent communication receivers in damaged lung. And a spatio-temporal trajectory reconstruction using the stLearn pseudo-time-space algorithm captured alveolar type 2 cells actively transitioning into type 1 cells in proliferative lesions, with one trajectory clade expressing the late maturation marker CAV1, evidence that regeneration programs do engage even in fatally injured lungs, though too late or too incompletely to restore gas exchange. The authors caution that their samples are post-mortem and may reflect downstream consequences of prolonged illness, and that longitudinal samples from the same patients were unavailable to confirm temporal progression. Still, by benchmarking their findings against non-COVID-19 diffuse alveolar damage and idiopathic pulmonary fibrosis, and showing that peribronchial fibroblasts form unique spatial niches only in COVID-19, the work establishes a spatially informed blueprint of fatal COVID-19 lung pathology, and a proof of principle for spatial transcriptomics as a bridge between molecular data and personalized medicine.

Subject of Research: Spatial transcriptomic mapping of cellular and molecular changes in fatal COVID-19 lung diffuse alveolar damage

Article Title: Spatial transcriptomics unveils the in situ cellular and molecular hallmarks of the lung in fatal COVID-19

Article References: Spatial transcriptomics unveils the in situ cellular and molecular hallmarks of the lung in fatal COVID-19. (n.d.). https://doi.org/10.1038/s41467-026-78188-3

Image Credits: AI Generated

DOI: 10.1038/s41467-026-78188-3

Keywords: spatial transcriptomics, COVID-19, SARS-CoV-2, diffuse alveolar damage, lung fibrosis, TGF-beta, SMAD3, peribronchial fibroblasts, endothelial cells, natural killer cells, alveolar regeneration, cell-cell communication