henhaobio
Sign in

NatureCell BiologyPeer-reviewed journal article

A spatial transcriptomics atlas of live donors reveals unique zonation patterns in the healthy human liver

Oran YakubovskyiD, Keren Bahar HalperniD, Sapir ShiriD, Roy NovoselskyiD, Amichay AfriatiD, Adi Egozi, Tal BarkaiiD, Yotam HarnikiD, Rouven HoeffliniD, Yael Korem KohanimiD, Ofra GolaniiD, Inna GoliandiD, Yoseph AddadiiD, Merav KedmiiD, Hadas Keren‐ShauliD, Liat Fellus‐Alyagor, Lena Prichislov, Dana HirschiD, Chen MayeriD, Ron Pery, Niv PencovichiD, Timuçin TaneriD, Ido NachmanyiD, Shalev ItzkovitziD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Reconstructing gene expression atlases for human tissues is challenging due to limited access to healthy samples from live donors. Neurologically deceased donors often show ischemic changes, while tissues near diseased regions may have altered gene expression. The liver, with its unique regenerative capacity, allows analysis from live healthy donors (LHDs). Using spatial transcriptomics (Visum, Visium HD and MERFISH), we analyzed 16 liver samples: eight from young LHDs and eight from patients with liver pathology, sampling adjacent normal tissue. LHD livers displayed significant gene expression differences from adjacent normal tissues. Hepatocytes exhibited marked zonation along the porto-central axis of liver lobules, with key functions pericentrally shifted compared to other mammals. Our atlas identified dynamic programs in early steatotic hepatocytes, showing transitions from lipid uptake in low-lipid regions to insulin hypersensitivity in high-lipid regions. This study presents a spatial gene expression reference for the healthy human liver and insights into hepatocyte adaptations in steatosis.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (23 Feb 2025) · CC BY 4.0

Spatial transcriptomics of live healthy donor livers shows most hepatocyte genes are zonated, with key functions pericentrally shifted versus mouse, pig, boar and cow.

Human liver atlases have relied on neurologically deceased donors, whose livers are altered by ventilation, non-enteral feeding, medications and ischemia-driven RNA degradation, or on tissue next to lesions that disease may perturb. Live healthy donor samples give an unperturbed spatial zonation reference for zonated liver diseases and let early steatosis be studied in otherwise healthy livers.

Key results

  1. ShownLivers from live healthy donors (LHDs) separate transcriptionally from adjacent normal resection tissue. The adjacent normal tissue carries immune and acute-phase signals, while LHDs show higher expression of classic hepatocyte metabolic genes.

    Pseudo-bulk clustering and PCA on 323 genes across 16 Visium samples (44,243 high-quality spots) separated LHDs from adjacent normal; adjacent normal up: S100A8/9, C1QA/B, IGHA1, IGHG3, SAA1/2, CRP, TIMP1; LHD up: ASS1, CYP2E1, CYP2C8, PCK2. A neurologically deceased donor liver showed IFITM2/3, SAA1/2 and hypoxia genes LDHA, GAPDH. · LHD liver tissues are transcriptomically unique; Pseudo-bulk analyses of human samples; Figure legends

  2. ShownMost hepatocyte-specific genes are zonated along the porto-central axis in healthy donors. Visium-derived profiles agree with orthogonal MERFISH imaging and with higher-resolution Visium HD.

    1,141 of 1,724 hepatocyte-specific genes differed significantly across 8 zones (q-value<0.25; 5 non-steatotic LHDs). MERFISH vs Visium portal-bias Spearman R=0.49, p=7.4e-7 (M5) and R=0.57, p<5e-324 (M8); Visium HD on M6 reproduced the standard Visium profiles. Pericentral: CYP2E1, CYP27A1, FASN, MLXIPL; periportal: CRP, SAA1, PCK1, ALDOB. · Zonation patterns in the healthy human liver; section 8; Human portal-central axis spot annotation; Human hepatocyte marker gene Identification

  3. ShownHuman hepatocyte genes are biased toward the pericentral zone relative to mouse, cow, domestic pig and wild boar. Secreted proteins and respiration genes shift pericentrally, as do most urea cycle enzymes versus mouse, and HNF4A is pericentral in human.

    Across 1,344 hepatocyte genes with 1:1 orthologs, human (n=7) genes were pericentrally biased vs each species (one-tailed rank-sum p<3.4e-15; mouse n=2, cow n=2, boar n=2, pig n=3). Secreted proteins p<5.22e-3; complement proteins not significant (p=0.06). Urea cycle enzymes except ASS1 pericentral vs mice; HNF4A pericentral (q-value=8.5e-14). · Zonation patterns in the healthy human liver; Pericentral shifts in human hepatocytes compared…; Cross-species portal-central axis spot…; Cross-species ortholog genes analysis

  4. SuggestedSome zonation features are shared across human, mouse, cow, pig and boar, including glutamine-cycle genes and Wnt target programs. This is consistent with Wnt signaling being an evolutionarily conserved regulator of zonation.

    GLUL and ADH4 were pericentral and ALB and GLS2 periportal across the species examined. Mean profiles of the 20 most Wnt-activated and 20 most Wnt-repressed genes (from Apc knockout mouse liver) were pericentral and periportal, respectively, in every species; human Wnt ligands WNT2 and RSPO3 and receptor LGR5 were pericentral. · Zonation patterns in the healthy human liver; Pericentral shifts in human hepatocytes compared…; Cross-species ortholog genes analysis

  5. SuggestedIn early steatosis, pericentral hepatocytes follow a biphasic program. Lipogenic genes rise and then fall as lipid increases, fatty acid oxidation, ANGPTL4 and insulin/adiponectin receptor genes increase, and bile acid and gluconeogenic genes decline.

    4 steatotic patients (M1, M2, M3, P6); pericentral spots binned into 8 lipid-content percentile bins (0% to 89.6% lipid droplet area, QuPath classifier). FASN, RXRA, MLXIPL rose then declined; CPT1A, PPARA, ANGPTL4, INSR, ADIPOR1, ADIPOR2 increased; ACLY, ELOVL6 elevated irrespective of lipid level; PLIN2 reduced. Kruskal-Wallis, q<0.25, K-means K=4. · Functional changes along the course of steatosis; section 8; Lipid droplet quantification

Methods, data and reproducibility

10x Visium profiled 16 human livers (8 live healthy donors, 8 adjacent normal) and 7 cow, pig and wild boar livers, plus published mouse data. Human spots were zoned via CYP2E1-correlated landmark gene sets (validated by MERFISH and Visium HD); other species used landmark genes or septa distance, and species were compared via 1:1 orthologs. A QuPath pixel classifier on H&E quantified lipid per spot.

  • Live healthy donor liver biopsies (1 cm³, taken at abdominal opening before resection), Mayo Clinic; 10x Visium · human · n = 8 donors (3 with mild steatosis) · Zenodo record 14795740
  • 'Adjacent normal' tissue sampled at least 5 cm from lesions in pathology-driven hepatectomies, Sheba Medical Center; 10x Visium · human · n = 8 patients (5 steatosis, 2 portal fibrosis, 1 ALPPS) · Zenodo record 14795740
  • MERFISH (MERSCOPE, 500-gene pan cancer panel) sections from two live donors for zonation validation · human · n = 2 patients (M5, M8)
  • Visium HD FFPE sections analyzed at 8 µm bins; M2 excluded for low RNA yield · human · n = 2 sections (M2, M6); M6 analyzed
  • 10x Visium of livers from cattle and domestic pigs (butcher-sourced) and wild boars (population-control culls) · cow (Bos taurus), domestic pig, wild boar (Sus scrofa) · n = cow n=2, domestic pig n=3, wild boar n=2 · Zenodo record 14795740
  • Published mouse liver Visium spatial transcriptomics dataset used for cross-species comparison · mouse · n = 2
  • Visium data of a neurologically deceased donor liver from Andrews et al., used for pseudo-bulk comparison · human
  • Single-cell liver atlas of Massalha et al., used to define 1,724 hepatocyte marker genes · human
Methods & tools
10x Visium Spatial Gene Expression, 10x Visium HD with CytAssist, MERFISH (Vizgen MERSCOPE, pan cancer pathways 500 gene panel), Space Ranger 2.0.1, MATLAB R2022a, MERlin image analysis pipeline, Cellpose, QuPath v.4.0 artificial neural network pixel classifier, Ensembl BioMart (1:1 orthologs), GSEA with KEGG and Hallmark gene sets, Illumina NovaSeq 6000, Kruskal-Wallis, Fisher's method, Benjamini-Hochberg FDR
Software & environment
Space Ranger 2.0.1 (human to hg38; pig and boar to Sscrofa11.1; cattle to BTauARS-UCD1.2); MATLAB R2022a; QuPath v.4.0; MERlin and Cellpose for MERFISH processing.
Reproducibility gaps
Only the raw Visium processing code is public; downstream analysis code is available from the authors on request. The Zenodo link was printed with an access token, suggesting restricted preview access at posting.
Limitations (5)
  • The steatosis analysis covered only early stages before hepatic inflammation began; steatohepatitis was not examined in this cohort. · stated by the authors
  • The analysis linking lipid content to expression used only 4 of the 7 steatotic patients, because H&E image quality was too low to extract reliable lipid content for the others. · stated by the authors
  • The Visium HD analysis rested on a single donor (M6), because the second section (M2) had much lower RNA capture (median 83 vs 169 UMIs per 8 µm spot). · stated by the authors
  • Health status is confounded with site, sampling procedure and demographics: LHD biopsies came from Mayo Clinic at abdominal opening and adjacent normal tissue from Sheba resections, with 8 per group. LHD vs adjacent normal differences may partly reflect these factors. · noted in this summary
  • Cross-species comparisons rest on 2 to 3 animals per species and a separately published mouse dataset. Zone assignment also differed (landmark genes in cow and mouse, distance from fibrotic septa in pig and boar; 6 zones vs 8 in human). · noted in this summary

Read this if you need a healthy human liver zonation reference, compare hepatocyte specialization across mammals, or study early steatosis using spatial transcriptomics with H&E-based lipid quantification.

Citations
22
FWCI
38.99
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISHildebrandt '21Hasan '26Ben‐Moshe '19Vandenbon '22Watson '25Zhang '26
70% semantic matchGenomics
Franziska Hildebrandt, Alma Andersson, Sami Saarenpää+10 more·Nature Communications·Dec 2, 2021· 195 citations