A spatial atlas of the healthy human liver redefines mammalian liver zonation

Yakubovsky et al., Nature, 2026

A new spatial transcriptomics atlas of the healthy human liver, published in Nature by Yakubovsky and colleagues at the Weizmann Institute of Science with collaborators at Sheba Medical Center and Mayo Clinic, demonstrates that key hepatocyte functions previously assumed to be conserved across mammals are zonated differently in humans than in mice. Tissue was collected from live healthy donors and analyzed using a multimodal workflow in which MERFISH on the MERSCOPE Platform provided single-molecule, single-cell validation of the spatial expression patterns inferred from sequencing-based methods. The result is the first reference atlas of human liver biology free from the transcriptomic distortions associated with deceased-donor and adjacent-normal tissue sources, with implications for translational disease modelling, liver biology, and the design of human cell atlases more broadly.

Sample sourcing as a determinant of atlas quality

Prior human liver atlases have been assembled from one of two tissue sources: organs procured from neurologically deceased donors, or histologically normal tissue collected adjacent to diseased liver. Both sources introduce systematic confounders. Deceased-donor livers can be influenced by non-physiological ventilation, nutrition, drug treatments, and ischaemic injury, factors that have been shown to account for as much as 40% of variance in RNA-sequencing data. Adjacent-normal samples, although histologically unremarkable, are taken from organs subject to local or systemic disease and can exhibit altered transcriptional programs.

To establish a cleaner reference, Yakubovsky and colleagues collected liver tissue from live healthy donors who had undergone rigorous medical screening prior to partial hepatectomy for transplantation. The study cohort comprised 16 samples in total: 8 from live healthy donors and 8 adjacent-normal controls. Linear mixed-regression analysis identified sample type as the dominant driver of gene expression variability across the cohort, exceeding the contributions of donor age, sex, or histology. Adjacent-normal livers had reduced expression of the expected functional genes and elevated expression of immune-related genes relative to live healthy donor tissue. This finding supports the hypothesis that proximity to liver disease compromises the use of adjacent-normal tissue as a healthy reference sample.

Figure 1. The hepatic lobule. Liver tissue is composed of hexagonal lobules. Blood flows from the portal vessels at the lobule periphery toward a central vein at the lobule center, generating gradients of oxygen and nutrients. Hepatocytes are organized into three zones along this porto-central axis: periportal (zone 1), mid-lobular (zone 2), and pericentral (zone 3). Source: Cunningham RP and Porat-Shliom N (2021) Liver Zonation – Revisiting Old Questions With New Technologies. Front. Physiol. 12:732929. doi: 10.3389/fphys.2021.732929Human-specific zonation of hepatocyte function

Zonation along the porto-central axis (see Fig 1) was reconstructed from sequencing-based spatial transcriptomics and validated with four complementary approaches: high-resolution spatial sequencing, single-nucleus RNA-seq, PhenoCycler imaging, and MERFISH on the MERSCOPE Platform.

The sequencing-based spatial methods provided transcriptome-wide coverage but at spot resolution, where multiple cells contributed to each measurement. Single-nucleus RNA-seq resolved cell-type heterogeneity but lost spatial context. MERFISH on the MERSCOPE Platform filled a distinct gap: single-molecule, single-cell quantification of a targeted gene panel directly in tissue, allowing the atlas’s zonation calls to be verified at the resolution at which the underlying biology actually operates.

66% of highly expressed hepatocyte-specific genes showed significant spatial zonation. Gene-set enrichment analysis of the differences between species identified pericentral enrichment in humans of multiple gene programs that are periportal or weakly zonated in mice, including glycolysis, gluconeogenesis, fatty acid metabolism, bile acid metabolism, and xenobiotic metabolism. At the gene level, several functions categorized as periportal in mouse exhibited clear pericentral expression in human tissue:

• Urea cycle. Key urea cycle enzymes NAGS, CPS1, OTC, and ASL were pericentrally zonated in human liver. Only ASS1 retained the periportal expression pattern characteristic of mouse liver.
• Gluconeogenesis. The mitochondrial gluconeogenesis gene PCK2 exhibited pericentral zonation, contrary to the established model of gluconeogenesis as an exclusively periportal program. The major hepatic glucose transporter GLUT2, encoded by SLC2A2, was likewise pericentrally biased.
• Lipid metabolism. Multiple lipogenic genes, including FASN, APOA5, GPAM, ACSL5, and DPP4, were strongly pericentral in human tissue but periportal or weakly zonated in mouse liver.
• Transcriptional regulation. HNF4A, a master hepatic transcription factor that is periportally zonated in mice, displayed pronounced pericentral expression in humans.
• Kupffer cell positioning. Resident liver macrophages, periportally enriched in mice, were pericentrally biased in humans. The authors propose that this localisation may relate to clearance of dying hepatocytes in zones of higher metabolic turnover.

Figure 2. Representative mRNA zonation profiles for humans (blue) and mice (red). Lines represent mean expression and shaded regions s.e.m. over individual repeats within each species, normalized to maximum across zones. PC, pericentral; PP, periportal. Source: Yakubovsky, O., Bahar Halpern, K., Shir, S. et al. A spatial atlas of the healthy human liver from live donors. Nature (2026). https://doi.org/10.1038/s41586-026-10377-y

Mouse and human livers differ in lobule size and metabolic rate, raising the possibility that the pericentral shifts observed in human tissue reflect those physiological differences rather than human-specific biology. To test this, the authors generated new 10x Visium spatial transcriptomics datasets from wild boar, domestic pig, and cow — large mammals whose lobule architecture and metabolic rates are closer to those of humans. Across all four non-human species, including these closer physiological matches, hepatocyte gene expression showed significantly less pericentral bias than in humans (one-tailed rank-sum P < 1.45 × 10⁻⁶ for each comparison). The pericentral shift therefore appears to be a human-specific feature, not a general consequence of larger lobules or slower metabolism.

Zonal organization of non-parenchymal cells and intercellular signaling

Hepatocytes account for approximately 60% of liver cells; the remaining cellular fractions play essential roles in coordinating liver function. These non-parenchymal populations have historically been challenging to resolve at zonal resolution because of their relative sparsity, lower transcript abundance, and tendency to be indistinguishable with neighboring hepatocytes in lower resolution methods.

Mapping zonal programs of non-parenchymal cells revealed a novel pericentral signaling axis: fibroblast-expressed BMP9 (encoded by GDF2) and BMP10 are pericentrally zonated morphogens, with matching pericentral BMP receptors on hepatocytes and LSECs. Together with the previously known pericentral WNT2 and RSPO3 signaling from LSECs and fibroblasts, this places stromal populations at the center of a coordinated WNT and BMP circuit that maintains pericentral hepatocyte identity. Resolving these rare-cell programs is exactly the kind of analysis where single-molecule methods like MERFISH on the MERSCOPE Platform have a clear advantage: low-abundance morphogens such as BMP9 and BMP10 in sparsely distributed fibroblasts are precisely the signals that get lost when sequencing data are averaged across tissue spots.

Early metabolic-associated steatotic liver disease

Transplantable livers generally contain ≤20% abnormal lipid accumulation (steatosis), and the study cohort therefore included a range of mild lipid accumulation in otherwise healthy organs. This provided a rare opportunity to examine early-stage metabolic-associated steatotic liver disease (MASLD) in samples uncomplicated by later-stage inflammation or fibrosis. Hepatocytes in early steatosis exhibited a coordinated decline in nuclear-encoded mitochondrial proteins alongside a compensatory increase in mitochondria-encoded transcripts, suggesting a mitochondrial imbalance that may contribute to subsequent metabolic deterioration. Lipogenic genes that are periportally zonated in mice were found to be pericentrally zonated in humans, with direct implications for the interpretation of mouse models of fatty liver disease.

Implications for human cell atlasing

The Yakubovsky atlas is significant on three levels. First, as a clinical reference, it provides the most rigorous baseline of healthy human liver biology assembled to date, against which disease-state findings can be interpreted. Second, as a comparative-biology resource, it requires re-evaluation of how mouse liver findings translate to human physiology, particularly for metabolic disease modelling. Third, as a methodological demonstration, it illustrates the value of combining careful sample sourcing, high-resolution spatial multiomics, and single-molecule sensitivity to resolve zonation patterns and intercellular signalling that are inaccessible to less complete approaches.

The full atlas is publicly accessible through the Itzkovitz laboratory web application at shalevapps.weizmann.ac.il/liver_app. For investigators working on liver biology, MASLD, hepatocellular carcinoma, or translational disease modelling, the atlas serves as a default human reference, and for the broader human cell atlas effort it establishes a methodological standard combining appropriate sample sourcing, complementary spatial modalities, and the resolution required to recover sparse and lowly abundant cell populations.

Interested in spatial atlasing of your tissue using the MERSCOPE Ultra™ Platform?

Our expanded MERSCOPE Pre-designed Panel with Add-on capabilities provides researchers a direct path: existing instruments remain compatible, and labs can apply the enhanced chemistry to projects already in progress.

Panel selection is no longer a bottleneck either. We offer more than 20 validated Pre-designed Panels across human biology, with focused content for major biological networks, alongside an oncology portfolio and dedicated mouse panels. For projects requiring additional targets, researchers can also add up to 100 Add-on genes on top of any existing Pre-designed Panel, extending coverage to specific pathways or cell populations of interest without designing a panel from scratch.

Learn more about MERSCOPE Pre-designed Panels and Add-on capabilities →