Skip to main content
Back to timeline
bioRxivSource publication:

Liver senescence read beyond marker lists: spatial causal models link stress inputs to a matrix-deposition loop

Synopsis

Across 11 human liver Visium spatial transcriptomics samples, two causal-discovery algorithms with correctness guarantees were extended to spatial data, yielding BOSSsrsi and LGLsrsi models with 224 and 38 direct upstream genes and 659 and 53 direct downstream genes. Overlap with the SenMayo and SenUMN marker lists is limited, the largest being SenMayo covering 13.4% of the BOSSsrsi upstream list. Pathway analysis of the BOSSsrsi lists describes senescence-associated signaling as integrating cellular stress inputs and immune responses into a self-sustaining loop of integrin-mediated TGF-β activation and continuous matrix deposition.

Source-provided article image: Spatial causal discovery reveals novel mechanisms of cellular senescence in human liver
Figures 1A · bioRxiv · Page 14

Interpretation

Senescence marker genes in human liver tissue carry significant spatial autocorrelation, so treating each Visium spot as an independent sample discards tissue-position information. Prior knowledge of the senescence transcriptome came from two sources: membership in known pathways, or genes whose expression differs between senescent and control cells; neither distinguishes whether a gene causes senescence, results from it, or is confounded with it. Among the selected non-M4M genes, 97.8% rejected the Moran's I null of no spatial autocorrelation (p<0.05), and all 10 M4M marker genes rejected it.

Spatial causal discovery adds the average expression of each spot's neighborhood as extra variables, letting the model separate signal within a spot from signal arriving from nearby tissue. BOSS and LGL were built for tabular observational data, with no variables designed for spatial position and no test of their correctness assumptions on real biological data. The study built spatial lag variables Y1 through Ym at hexagonal taxicab distances, and across 110 tests on 11 samples and 10 marker genes, 100 satisfied the spatial causal Markov property at distance 2 and 10 at distance 3.

The upstream and downstream gene lists from the two spatial algorithms overlap only partly with standard senescence marker lists, pointing to many candidate regulatory relationships outside those lists. Lists such as SenMayo and SenUMN are assembled from expression differences or pathway membership and do not label direction. BOSSsrsi produced 224 upstream and 659 downstream genes, LGLsrsi 38 upstream and 53 downstream; SenMayo covered 13.4% of the BOSSsrsi upstream list, the largest overlap among the lists, while LGLsrsi upstream and downstream reached 47.4% and 32.1%.

Pathway enrichment on the BOSSsrsi upstream and downstream gene sets describes the senescent microenvironment as stress inputs, immune activation and extracellular matrix remodeling reinforcing one another. Earlier pathway knowledge assigned genes to known mechanisms but could not show whether those pathways are mobilized together in human liver tissue. Upstream genes were enriched for humoral immune and complement responses, extracellular matrix organization, oxidative stress and hypoxia responses, and wound healing; the top three downstream GO terms were all matrix-related, with enrichment for TNFα response via NFKB and integrin and platelet-responsive pathways.

Perspective

The analysis covers only Visium spatial transcriptomics of human liver tissue: 11 samples from 10 participants aged 32 to 70, with liver health ranging from normal to highly fibrotic; the conclusions apply to that tissue and platform, and the authors propose extending to other animal models, other tissue types and more samples.

The authors note that results come only from human liver tissue and may not apply to other tissues or animals; the analysis centers on 10 human liver senescence marker genes, and experts may prefer different marker sets; only 11 samples were analyzed, so biological variation between tissue samples may not be fully captured; the modeled causal relationships await experimental validation.