*2.8. Functional Genomics Analysis of the CCl4 Signature*

Pathway analysis with PROGENy. Pathway activity scores were calculated with the functional genomics tools PROGENy [26,27]. While classical pathway analysis methods (e.g., Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis) rely on gene sets containing genes of pathway members PROGENy exploits so called "footprint gene sets" containing not the pathway members but the most responsive genes upon corresponding pathway perturbation. PROGENy was applied on contrast using the moderated t-value as a gene-level statistic.

Transcription factor (TF) analysis with DoRothEA. DoRothEA is a high-quality data resource of TF-target interactions (regulons) [26,28]. Coupling DoRothEA regulons with a statistical method allows to infer TF activity from the expression of its transcriptional targets. We used as statistical method the function viper from the R package viper (version: 1.17.0) [29] that computes for each TF a normalized enrichment score that we consider as TF activity. DoRothEA in combination with viper was applied on contrasts using moderated t-values as a gene-level statistic.

Gene Ontology (GO) term enrichment. We used the R package msigdf (https://github.com/ ToledoEM/msigdf) to query GO terms (biological process and molecular functions) from MsigDB. Gene set enrichment analysis was performed on contrasts using the R package fgsea (version 1.10.0) [30] with 100,000 permutations. The moderated t-value was used as a gene-level statistic.

Construction of consensus pericentral and periportal gene sets. Consensus pericentral and periportal gene sets were constructed by integrating pericentral and periportal gene sets from three independent studies [31–33]. Gene sets reported by Braeuning et al. [31] were extracted from their corresponding Supplementary Table S1. Saito et al. [33] made their data available for both male and female mice. Since only male mice were used in the here presented work we focused on common (with respect to male and female mice) and male specific pericentral and periportal genes. Halpern et al. [32] do not provide explicit pericentral and periportal gene sets in their supplement; therefore, we generated them systematically starting from their Supplementary Tables S1 and S2. Supplementary Table S1 contains the UMI counts of 1415 hepatocytes/cells. Supplementary Table S2 reports information about the spatial organization of the 1415 cells across the 9 lobule layer (layer 1 is the most pericentral and layer 9 is the most periportal layer). For each cell and lobule layer combination a probability is given that indicates the likelihood that a cell was originally located in the respective layer. We started our analysis pipeline by removing genes that were expressed in less than 15 cells (out of 1415). Subsequently, we normalized the sub-setted count matrix using the R package scran (version 1.11.27) [34]. To assign cells to a specific lobule layer we selected for each cell the layer with the highest probability yielding in a zonation table reporting the spatial distribution of all cells across the lobule. Given this zonation table we identified genes with significant monotonic expression changes across the lobule layer by applying the exact version of the Jonckheere–Terpstra test from the R package clinfun (version 1.0.15, https://cran.r-project.org/web/packages/clinfun/index.html). Monotonically increasing genes (from layer 1–9) were considered as potential periportal and monotonic decreasing as potential pericentral genes. Only genes with an FDR ≤ 0.001 were considered as pericentral and periportal gene set members. As the three independent studies were published within more than a decade we updated all MGI gene symbols to their current alias using the function alias2SymbolTable from the limma package (version: 3.39.18) [25]. The consensus pericentral and periportal gene sets contain only those genes that are reported in at least two studies.

Characterization of the overlap of pericentral/periportal genes and the most responsive genes of CCl4 treatment. In this analysis, the overlap of pericentral and periportal genes with the differentially expressed genes after CCl4 treatment is characterized with over-representation analysis (ORA). To ensure a reasonable overlap size we relaxed the condition of differentially expressed genes to abs(logFC) ≥ 0.8 and FDR ≤ 0.2. We identified the final overlap gene set (for both zonations: pericentral and periportal) independently of time by taking the union of overlapping genes across all time points (months 2–12). ORA was performed using Fisher's exact test. The number of background genes has been set to 20,000 as this reflects a typical number of genes in a mouse transcriptome experiment. We tested the following gene sets: GO terms (molecular functions and biological processes), DoRothEA regulon's, PROGENy's footprint gene sets, KEGG gene sets. *P*-values were corrected using Benjamini Hochberg correction (false discovery rate, FDR) [35].
