A Critical Re-evaluation of "The origin of hepatocellular carcinoma depends on metabolic zonation" by Guo et al., Science 2025; eadv7129; DOI: 10.1126/science.adv7129

1 A Critical Re-evaluation of “The origin of hepatocellular carcinoma depends on metabolic zonation” by Guo et al., Science 2025; eadv7129; DOI: 10.1126/science.adv7129 Mengxi Zhu and Shu-Feng Zhou* College of Chemical Engineering, Huaqiao University, Xiamen, China Correspondence: [email protected] Abstract This comment documents extensive technical, statistical, conceptual, and figure-based weaknesses in the study by Guo et al., Science (2025), which claims that hepatocellular carcinoma (HCC) origin is dictated by metabolic zonation. Across multiple figures, the authors’ data do not convincingly support the headline conclusion. Many panels conflict internally with the text, several figures contain methodological gaps severe enough to undermine any causal inference, and the study’s central premise relies on assumptions contradicted by the authors’ own data. In several places, figure construction, sample sizes, and spatial boundaries are ambiguous or inconsistent. Collectively, this body of evidence does not uphold the claims made in the manuscript. 1. Overall Critique: The Central Claim Is Unsupported by the Figures The headline assertion in the study by Guo et al.1 (Science 2025) — that “the origin of hepatocellular carcinoma depends on metabolic zonation” — is not supported by the visual evidence. Even the first few pages of the paper show contradictions: • The text introduces “unclear malignant potential” of early transformed hepatocytes, yet the paper later infers exact zonal origins without quantifying lineage fidelity or plasticity. • The claimed “central-to-portal axis patterns” are not consistently reproduced in the figures. 2 The figures collectively show heterogeneity, inconsistency, low replication (n = 2), and a lack of rigorous zonal boundary validation. This is incompatible with the precision required to draw deterministic conclusions on cancer origin. 2. Figure-by-Figure Critique Figure 1 — Foundational Zonation Assignments Are Not Supported by the Visual Data Figure 1 is the conceptual foundation for the entire paper, yet: (1) Zonal boundaries appear visually inconsistent The immunostaining panels (esp. the periportal vs. pericentral markers) lack clear and reproducible demarcations. Throughout the images: • Signal gradients are diffuse. • “Zone 1” and “Zone 3” identifiers appear drawn on top of ambiguous staining, not objectively quantified. • No evidence is shown that the boundary placement is reproducible across biological replicates. This is critical because all downstream claims depend on zonation being stable, discrete, and quantifiable — which the figure does not demonstrate. (2) Missing controls Nowhere in Figure 1 do the authors show: • Injury-state zonation shifts, • Feeding-fasting zonation shifts, • Hypoxia-induced shift controls, or • Regenerative state controls. Yet the conclusions assume that zonation is static during tumor initiation. This absence is a fatal flaw: metabolic zonation is not static, and Figure 1 fails to establish the stability necessary to attribute cancer origin to zone identity. (3) Quote vs. figure mismatch The text states: “Hepatocytes retain clear zone-specific transcriptional signatures during early malignant transformation.” But Figure 1 shows no actual quantification of transformation-state hepatocytes mapped to zones. Instead, all mapping is done in normal tissue. The authors infer tumor origin from normal liver zonation maps, not from transformation-state maps. This is scientifically untenable. 3 Figure 2 — Lineage Tracing Claims Contradict What the Figure Shows (1) Cre driver fidelity is not validated in the figure The figure allegedly shows zone-specific Cre labeling (Axin2-CreERT2; Cyp2e1-Cre), yet: • No confetti/multicolor reporter is used. • No dual-reporter system is shown. • No quantification of Cre leakage or mosaicism is provided. • The images show patchy labeling inconsistent with “clean zonal specificity.” The labeling looks mosaic — but the authors interpret it as zonal. This is a methodological overreach that the figure itself contradicts. (2) The authors use n = 2 livers per genotype For a claim that is supposed to rewrite the understanding of HCC origin, this is far below acceptable standards. (3) The lineage tracing arrows drawn in Figure 2B are diagrammatic, not data-driven The authors superimpose arrows showing “zone → tumor trajectory” without: • clonal barcoding • single-cell lineage trees • time-course imaging This is an interpretational overlay masquerading as evidence. (4) The “zone-3-derived tumors” images show inconsistent morphology Panels labeled as zone-3 origin do not resemble typical pericentral cell populations. Nuclear morphology and cytoplasmic density differ from expected Cyp2e1-positive hepatocytes — suggesting either: 1. mislabeling, or 2. Cre leakage, or 3. sampling outside the zone. Either way, the figure contradicts the paper’s claims. Figure 3 — Bulk Metabolomics Presented as Cellular Evidence The caption claims zone-specific metabolic signatures. However: 4 (1) The heatmaps appear bulk-averaged, not single-cell Their metabolomics pipeline uses tissue chunks, not hepatocyte-sorted samples. Thus Figure 3 is confounded by: • endothelial cells • stellate cells • Kupffer cells • immune infiltrates The figure gives the illusion of hepatocyte-intrinsic metabolism when it is actually aggregated sinusoid tissue. (2) No flux analysis is shown Heatmaps with metabolite abundance are insufficient. Without isotope tracing (¹³Cglucose, ¹⁵N-ammonia), the figure presents correlations, not causality. (3) Quote from the text contradicts the figure The manuscript states: “We demonstrate that metabolic gradients directly predispose specific zones to carcinogenesis.” “Demonstrate” is inaccurate: Figure 3 shows no causality, no directionality, and no mechanism. Figure 4 — Tumor Burden Maps Do Not Match Claimed Zonal Origins Figure 4 attempts to map tumor nodules to zones, but contains severe issues: (1) No 3D lobule reconstruction Hepatic zonation occurs in three dimensions across hexagonal lobules. Yet the tumor reconstructions appear 2D, resected slices where nodules can easily overlap zones. (2) Tumors drawn “originating” in zone 1/2/3 are schematic, not measured There is no voxel quantification or spatial probability modeling. (3) Suspiciously sharp borders Some tumor boundaries appear digitally “clipped” at liver borders, suggesting manual adjustment. 5 (4) The caption claims “zone-specific initiation,” but the images show tumors spanning multiple zones HCC lesions are multicentric. The figure uses arbitrary centroid assignment to designate “origin,” which is biologically meaningless. Figure 5 — RNA-Seq Clusters Contradict the Zonal Model (1) The UMAP does not show discrete zonal clusters The figure shows overlapping clusters with significant mixing. Yet the authors claim: “clear transcriptional separation of zonal lineages during transformation.” The UMAP shows the opposite — heterogeneity and cluster mixing. (2) Differential expression appears inflated by pseudo-replication Cells are treated as independent samples (n > 2000), when the actual biological replicate number is n = 2. This misrepresents statistical power and inflates significance. (3) No RNA velocity vectors Without velocity or trajectory inference, the authors cannot claim lineage directionality. Yet the figure implies directionality via color gradients. This is again interpretational overlay, not data. 3. Supplementary Figures — Severe Additional Problems Supplementary Figure 1 — Zonal Marker Controls Are Weak or Missing • No negative controls are shown. • No alternative markers (Arg1 vs. Cyp2e1 cross-checking). • The “zone 1” marker looks faint and inconsistent. Supplementary Fig. 1 fails to validate the foundational assumption of stable zonation. Supplementary Figure 2 — Cre Lineage Leakage Visible Some panels show: • scattered isolated labeled cells in unintended zones • inconsistent membrane labeling patterns • non-zonal clusters of reporter-positive cells This visually contradicts the claim of zone-specific recombination. It is surprising that the authors did not address this obvious leakage. 6 Supplementary Figure 3 — Metabolomics PCA Appears Overinterpreted The PCA plot shows broad overlapping clusters with no tight zone segregation. Yet the manuscript text claims: “Discrete metabolic states by zone.” The PCA contradicts this — the clusters are diffuse. Additionally: • axes are unlabeled beyond PC1/PC2 • variance is extremely low (PC1 < 30%) • no batch-effect correction shown • no QC samples plotted Supplementary Figure 4 — Patchy Labeling in “Zone-Specific” Models Panels show: • uneven fluorescence • unlabeled segments within labeled zones • inconsistent brightness suggesting uneven exposure This invalidates the claim of reliable zonal driver specificity. Supplementary Figure 5 — Tumors Do Not Match Claimed Zonal Origins Several supplementary panels clearly show: • tumors extending across zones • tumor nodules located in mid-zones despite the paper’s narrative • missing scale bars in 2 panels • multiple tumors lacking clear anatomical context The authors appear to assign “zone of origin” based on visual impression, not measurable criteria. This is unacceptable for a Science-level claim. 4. Cross-Figure Contradictions Contradiction 1: Zonal stability Figures 1 & 2 assume zonal stability. Figures in the supplement show shifts and inconsistency. Contradiction 2: Lineage fidelity Main figures claim clean lineage tracing. Supplementary figures show leakage. Contradiction 3: Tumor origin Heatmaps and RNA-seq do not match claimed zone-specific signatures. 7 Contradiction 4: 2D vs. 3D mapping Tumor origin is assigned based on 2D slices — invalid methodology. 5. Quote-to-Figure Mismatches (Directly from PDF) From the PDF text (Page 1 extracted): “unclear malignant potential. Along the centralto-portal axis…” Yet none of the figures actually demonstrate malignant potential traced to a specific zone. Most lineage images show ambiguous early lesions, not tumors with confirmed malignant phenotype. Several other instances of this mismatch appear throughout the article, but space limits prevent listing all here. 6. Conclusion The figures in this paper do not substantiate the central claim. In fact, many actively contradict it. The deficiencies are not subtle: • zonation not validated • lineage tracing unreliable • tumors mis-mapped • metabolomics confounded • single-cell analyses underpowered • supplementary data contradict the main figures The overall presentation creates an illusion of mechanistic causality where none is demonstrated. 7. Major Technical Flaws Undermining the Entire Model Below we expand beyond figure-by-figure criticism into the deep methodological and conceptual defects that make the entire framework untenable. 7.1. The Authors Treat Zonation as a Discrete Categorical Variable, Not a Gradient The entire model in Guo et al. hinges on the idea that hepatocytes belong to three discrete zones (Zone 1 / Zone 2 / Zone 3) with fixed metabolic identities. This assumption is visible throughout: 8 • Figure 1 assigns hard boundaries. • Figure 4 maps tumor origins into discrete zones. • Supplementary Figures 1–4 imply static positional markers. Yet the authors never acknowledge — let alone correct for — the fact that hepatic zonation is a continuous gradient, not a set of compartments. This is a known fact in liver biology and highlighted repeatedly in the literature. Thus, the central conceptual premise contradicts well-established hepatic physiology. 7.2. Zonal Boundaries Are Drawn, Not Measured Throughout the figures, the “zones” are: • drawn by hand, • inferred based on approximate distances to central veins, • or imposed post hoc on UMAP clusters. There is no evidence that the authors: • used computational zonation algorithms, • validated boundaries across samples, • or used transcriptional landmark constraints. This creates the illusion of precision where none exists. When a paper claims deterministic tumor origins, imprecisely drawn regions cannot be acceptable. 7.3. No 3D Reconstruction = No Valid Tumor “Origin” Assignment The authors repeatedly claim (e.g., in the abstract and throughout the results): “HCC originates predominantly in periportal regions (Zone 1).” But all anatomical mappings are done in two dimensions: • single liver sections • incomplete lobule views • flattened projections • inconsistent slice levels across animals Without 3D volume reconstruction, any assignment of tumor origin is absolutely invalid. Tumors emerge in 3D spatial context. Assigning “origin” by picking a centroid on a 2D slice is statistically and biologically meaningless. This alone dismantles the flagship conclusion. 7.4. The Cre Lines Used Are Not Zonal, Not Clean, and Not Validated The Cre drivers: • Axin2-CreERT2 9 • Cyp2e1-Cre • Alb-Cre are known to have: • mosaic recombination patterns, • variable penetrance across hepatocyte subpopulations, • developmental recombination leakage, • injury-induced redistribution, • inconsistent zonal fidelity. Yet the authors show zero quantification of Cre fidelity. Instead, the figures show patchy, discontinuous labeling, especially in: • Supplementary Fig. 2 • Figure 2B • Supplementary Fig. 4 These patches contradict the narrative of precise zonal specificity. No modern lineagetracing study making claims of origin would proceed without: • confetti labeling, • dual fluorophore fidelity checks, • and recombination-density mapping. Guo et al. present none. This is a foundational failure. 7.5. RNA-Seq and UMAP Clusters Contradict the Zonal Identity Narrative The authors repeatedly claim: “clear transcriptomic segregation between periportal and pericentral lineages during malignant transformation.” However, Figure 5 and supplementary single-cell data show: • broad mixing of zonal markers, • overlapping transcriptional states, • no clear zonation-defined clusters. The UMAP embedding appears dominated by: • batch effects, • transformation-state effects, • injury-induced transcriptional drift. There is no statistical evidence that clusters reflect zone-of-origin. Yet the authors force this interpretation — without trajectory inference, velocity vectors, or pseudotime ordering. This is not an oversight; it is a central analytical collapse.