{"id":"48f52937-6885-4f87-901e-3750ee0bbc07","arxiv_id":"2510.18589","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Cell-cycle exit in human bone marrow stromal cells is statistically clustered along lineage branches, indicating heritable (epigenetic) regulation, as detected by a new lineage-topology entropy.","lead":"This paper defines an \"inheritance entropy\" that measures whether cells that stop dividing are clustered in particular branches of a cell lineage tree, and tests it on 32 human bone marrow stromal cell colonies. It reports that in 21 of 28 testable clones the observed clustering is far stronger than expected under a scrambled, non-hereditary null, suggesting cell-cycle exit is influenced by heritable, likely epigenetic, traits.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Spatial confounder not excluded: scrambling null destroys physical-position structure, so branch-clustered inactivity could arise from a dish-level gradient; Appendix B's distance correlation does not rule this out.","rationale":"The paper's statistical machinery is carefully built: the permutation test is exact conditional on the exchangeability of inactive cells within generations, and the large number of significant clones (21/28) is strong evidence that the positions are non-random. However, the load-bearing inference to 'heredity' requires that the null be the correct counterfactual for a non-hereditary process. The scrambling null is too broad: it removes all spatial correlations by randomly relocating cells on the dish, so a spatial gradient—a non-hereditary, non-exchangeable factor—would also be erased and would generate a low P-value. Appendix B's rebuttal only considers a localized spot and relies on a global distance correlation that is not specific to the inactive-cell pattern. This is a concrete, addressable gap rather than a fundamental flaw; a spatially constrained permutation test would directly decide whether the signal survives when space is conserved. I therefore keep the reader's CONDITIONAL verdict: the exciting claim needs this additional control before being accepted as a proof of epigenetic inheritance. The lack of code/data, noted by the reader, is secondary and does not change the core concern.","tokens_in":15303,"tokens_out":16246,"duration_ms":144272,"concrete_test":"Using the raw tracking data, build a spatially constrained null: at each generation k, partition cells into spatial bins of width ~250–500 pixels (approximately the cell scale) and randomly permute cell identities only within each bin, carrying their subtrees. This preserves the empirical spatial distribution of inactive cells while severing genealogical links. Recompute the entropy P-values for all 28 clones with 10^5 resamples. If the number of significant clones (P<0.05) falls markedly from 21/28 to near the expected ~1.4 under the global null, the original inheritance signal is an artifact of spatial heterogeneity rather than heredity. A second variant: permute within radial quantiles to break left–right gradients. If results remain significant under both constraints, the heredity claim is substantially strengthened; if not, the central inference in Section III.C is compromised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III.C concludes that the topology 'cannot be attributed to random variations in the positions of inactive cells,' thereby 'proving' a strong hereditary character. The proof rests on the within-generation scrambling null (Appendix A) that randomly permutes cell identities at each generation, carrying entire subtrees. This null severs mother–daughter links, but it also randomizes the physical positions of cells: each cell's recorded spatial trajectory travels with it during the swap. Consequently, the null does not represent a non-hereditary process that includes spatial control. Any dish-level spatial gradient (e.g., nutrients, oxygen, waste) that makes inactivity more likely in one region will generate branch-clustered inactivity because cells are nearly sessile and a founder's descendants tend to stay in the same region. Such a gradient would produce low S in the biological tree, while scrambled trees—with inactive cells spread evenly across space—would have higher S, yielding a small P-value. Thus the test conflates heredity with spatial non-exchangeability. Appendix B attempts to rule out physical proximity by showing a weak overall correlation (ρ=0.24) between physical and topological distance for all mitosis pairs. That does not address the confounder: a large-scale gradient affects whole regions, and cells in the same lineage branch are necessarily in the same region, so branch clustering and spatial clustering are confounded. The appendix only discusses a 'localized spot of size R' and never tests whether the specific inactive cells are spatially biased. The statement that the structure 'cannot be due to physical proximity' is therefore unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a topology-only measure, the 'inheritance entropy' S, to detect heritable control of cell-cycle exit in clonal BMSC colonies. For each mitosis, the imbalance is the absolute difference between the numbers of missing generation-7 leaves in the two daughter branches; S is the Shannon entropy of the normalized imbalances (Eqs. 1-3). To assess significance, the authors randomly permute cells within each generation while carrying subtrees, preserving per-generation inactive counts (Appendix A), generate 10^6 scrambled trees per clone, and compute the fraction with S_scrambled <= S_biological (Section III.C). Of 28 testable clones (4 have at most one inactive cell), 21 give P<0.05. The paper interprets this as proof of hereditary epigenetic regulation, and additionally reports a mutation lag of 3.1 generations between the inferred change in inactivity propensity and the first G0 cell (Section III.E).","tokens_in":15649,"tokens_out":8951,"duration_ms":77765,"significance":"The permutation test is simple, transparent, and parameter-free: it uses only the observed tree topology, preserves a natural null constraint, and is implemented with 10^6 scrambles. The definition of the null is non-circular and the P-values are straightforward. If the spatial confounder can be excluded, the method would be a useful contribution to lineage-tracing analysis. However, the central inference currently rests on an unverified assumption that all non-hereditary sources of branch-clustered inactivity are removed by within-generation scrambling, and the language 'prove' exceeds what a permutation test can establish. These issues are fixable with additional analyses.","major_comments":[{"comment":"The spatial confounder is not excluded by the current null. The scrambling procedure (Appendix A) randomly permutes cell identities within each generation and carries subtrees with them, so it also randomizes the physical dish positions of inactive cells. If a dish-level gradient (nutrients, oxygen, waste) makes G0 entry more probable in one region, cells descended from a founder in that region will tend to be inactive because they are sessile; this produces branch-clustered inactivity and low S in the biological tree, while scrambled trees spread inactive cells over space and have higher S. Appendix B's Spearman rho=0.24 between pairwise physical and topological distances for all mitoses does not test this: a large-scale gradient affects entire regions, and the pairwise scatter plot in Fig. 8 does not restrict the spatial scale of the putative G0-promoting region. Please add a spatial n","section":"Section III.C; Appendix A-B; Eqs. (1)-(3)"},{"comment":"The word 'proving' in Section III.C (and in the abstract) overstates what a permutation test can show: rejecting the scrambling null provides statistical evidence, not a proof. In addition, 28 hypothesis tests are performed without any multiple-comparison correction. Although the very small P-values for many clones suggest the aggregate conclusion would survive correction, the paper should report adjusted P-values (e.g., Benjamini-Hochberg) and state how many clones remain significant. This is load-bearing because the headline result is '21 of 28 colonies are significant'.","section":"Section III.C; Fig. 5"},{"comment":"The mutation-lag analysis appears to be based on visual inspection of the imbalance maps rather than a pre-specified algorithm. The text says 'an inspection of the imbalance map ... shows' and Fig. 6 annotates mutations manually. The reported average lag of 3.1 generations therefore lacks an objective selection rule, uncertainty quantification, and validation on simulated trees. Since this section is presented as a quantitative result, it needs a defined procedure (e.g., threshold on normalized imbalance and a rule for resolving chains) and a sensitivity analysis.","section":"Section III.E; Fig. 6"}],"minor_comments":[{"comment":"The phrase 'we prove' should be softened to 'provide evidence' or 'show'; the analysis is a permutation test, not a proof.","section":"Abstract; Introduction; Section III.C"},{"comment":"If all inactivity imbalances I_m are zero, S is undefined. State a convention for this edge case.","section":"Eq. (3)"},{"comment":"Report exact P-values and effect sizes in a table; violin plots make small counts hard to read. Also state how many of the 28 clones remain significant after multiple-comparison correction.","section":"Fig. 5"},{"comment":"The dataset and the 84-hour G0 threshold are described in the companion preprint [35], not in this manuscript. Please clarify data availability and, if possible, move the threshold robustness analysis into this paper or a supplement.","section":"Section II.A; Reference [35]"},{"comment":"The term 'non-hereditary lineage' is stronger than 'tree generated by the scrambling null'; consider rewording to avoid the impression that the null represents all possible non-hereditary mechanisms.","section":"Abstract; Section II.B"},{"comment":"The 17 apoptotic cells are merged with G0 cells; since apoptosis may have a different mechanism, a sensitivity analysis excluding them would be useful. The inference from intra-colony heritable variation to inter-colony heterogeneity is also speculative and should be marked as a hypothesis.","section":"Section II.A; Section IV"}],"recommendation":"major_revision","confidential_remarks":"The paper has a genuinely interesting idea and the central permutation test is transparent, but the spatial confounder is a serious gap and the 'proof' language is not appropriate. I believe the authors can address this with a spatial null or covariate analysis; if they do, the paper could be acceptable. Please also consider requiring the companion paper [35] or the underlying lineage data to be available for review, since the biological conclusions depend critically on those external details."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a useful paper. It defines a new observable — the inactivity imbalance — and a normalized Shannon entropy, then tests whether observed lineage trees have unusually concentrated imbalance relative to within-generation scrambling. The method is clearly explained, the null preserves per-generation inactive counts, and there are no fitted parameters. On its own terms the test works: most clones show low entropy relative to scrambled trees, and the result is not manufactured by fitting. That is a real contribution.\n\nThe soft spots are real but proportionate. First, 'proving' hereditary character is too strong. The permutation test establishes that inactive cells are not exchangeable across branches conditional on per-generation counts. That is evidence against one particular non-hereditary model, not a proof of inheritance. The authors acknowledge the test is sufficient but not necessary, which helps, but the conclusion section reverts to 'can only be due to inheritance.'\n\nSecond, the null assumes exchangeability of inactive positions across cells within each generation. That assumption fails under any spatial confounder: a dish-level gradient in nutrients or oxygen makes inactivity more likely in one region, and since descendants of a founder stay near it, branch-clustered inactivity follows even with zero heredity. The scrambling destroys spatial structure, so the comparison conflates heredity with spatial non-exchangeability. Appendix B does not fix this. The Spearman correlation between physical and topological distance for all mitosis pairs is about a different question; a large-scale gradient affecting whole regions would not show up as local spot clustering. The appendix never tests whether inactive cells themselves are spatially biased. So the claim that spatial effects are ruled out is not supported.\n\nAlso minor: 28 tests without multiple-comparison correction; but 21/28 and many tiny P-values mean this is probably not load-bearing. Data and code are not shipped, which is addressable.\n\nBottom line: as a descriptive topological statistic with a clean null, this is worth engaging with. As a proof of epigenetic inheritance, it overreaches. The paper deserves a serious referee; with a careful rewrite that softens the causal claim and adds a spatial-control test (or ships data so others can test), it could be solid.\n\nI'd bring it to reading group.","headline":"A clever topology-based entropy test for clonal heterogeneity, with a solid permutation null, but the 'proof' of epigenetic inheritance outruns the null model.","tokens_in":16167,"tokens_out":1596,"would_cite":true,"duration_ms":15717,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims to prove that heritable traits, not random cell-to-cell variation, determine where bone-marrow stromal cells stop dividing, and that the inheritance is epigenetic.","keywords":["cell lineage trees","inheritance entropy","cell-cycle exit","bone marrow stromal cells","mesenchymal stem cells","epigenetic inheritance","lineage tracing","quiescence"],"falsifier":"A decisive test would be to physically separate the two sister cells at a mitosis flagged with a large imbalance and grow their progeny in identical, randomized environments; if the branch difference in inactivity propensity does not persist, the heredity claim fails. Alternatively, a single-cell molecular readout of a candidate inherited mark taken at the flagged ancestral mitosis should show a systematic difference between the two daughter branches before any inactive cell appears; if no such mark exists, the proposed epigenetic mechanism is unsupported.","tokens_in":15228,"feed_emoji":"🧬","tokens_out":6301,"duration_ms":55676,"temperature":0.7,"pith_summary":"The paper claims that the heterogeneous behaviour of clonal human bone marrow stromal cell colonies is not the product of random cell-to-cell variation, but of heritable traits that regulate cell-cycle exit. To make this case, it defines an 'inheritance entropy' that measures whether inactive cells—cells that stop dividing—are clustered within a few branches of the proliferation tree or scattered evenly across it. Randomly rescrambling each colony one million times to erase mother-daughter links while preserving the number of inactive cells per generation, the authors find that 21 of 28 testable colonies have entropy lower than essentially all scrambled versions (P<0.05). A low entropy compared with the scrambled ensemble is interpreted as evidence that the propensity to become inactive is transmitted down lineages, probably through epigenetic marks, since the experiments span too few generations for genetic mutations to accumulate. If true, this would explain why colonies originating from single cells differ systematically, and would offer a path toward controlling colony heterogeneity in bone-regeneration therapies.","feed_headline":"Most bone-marrow clones inherit their cell-cycle exit pattern","feed_subtitle":"A topology-based entropy test on 32 human colonies finds 21 of 28 show branch-clustered inactivity, pointing to epigenetic control.","key_machinery":"The machinery is the pair (inactivity imbalance, inheritance entropy). At a given mitosis m, I_m is the absolute difference between the counts of missing seventh-generation leaves in the left and right sub-branches; it measures how much more destructive to the final tree one daughter lineage is than its sister. Normalising I_m over all mitoses gives weights w_m, and the inheritance entropy S = -Σ w_m log w_m condenses the whole imbalance map into a number: many similarly-sized scattered imbalances give high S, while a few dominant imbalances concentrated at a handful of nodes give low S. The test then compares S to the distribution obtained by randomly permuting all cells within each generat","core_discovery":"The central discovery is that the topology of BMSC lineage trees carries a detectable hereditary signal. For each mitosis, the authors compute an inactivity imbalance: the absolute difference between the number of missing seventh-generation leaves in the two daughter branches. They normalise these imbalances across the tree and define S = -Σ w_m log w_m; low S means that most of the imbalance is concentrated on a few nodes, which is the signature of a heritable change in the probability of producing inactive cells. To test significance, they scramble each biological tree by random permutations within each generation, which destroys mother-daughter relationships while keeping the per-generati","pith_inferences":["The null preserves only per-generation inactive counts; it does not randomize away all non-hereditary correlations that could cluster inactivity in branches, such as local microenvironmental niches shared by adjacent lineages or sister-cell asymmetries unrelated to heritable state. If such factors matter, some of the 21 significant clones could be false positives for inheritance.","The 3.1-generation average lag is directly testable: single-cell measurements of candidate epigenetic marks along a lineage should show a change at the flagged ancestral mitosis before the first inactive cell appears; absence of such a change would challenge the interpretation.","The entropy test's power is limited by tree depth and inactive-cell count; extending lineage tracing beyond seven generations or experimentally increasing the number of cell-cycle exits should sharpen the test and may expose heredity in the 7 currently non-significant clones, which are mostly clones with only 2–3 inactive cells.","The same logic could be inverted to measure non-heritable branch correlation: comparing physical proximity, division-time correlations, and the entropy signal in the same trees could separate inherited propensity from environment-driven clustering, though the paper's Appendix B only begins that separation."],"forward_implications":["If confirmed, the paper implies that colony-to-colony heterogeneity in BMSC transplants is traceable to heritable epigenetic states rather than stochastic noise, so protocols that modulate those states could reduce transplant unpredictability.","The method gives a branch-level map of where inactivity propensity changes; this map can be used upstream to locate the epigenetic event 3–4 generations before inactive cells appear, enabling marker discovery.","Since the test needs only lineage topology, not molecular readouts, it can be applied to any cell type where a fate decision prunes branches of a tree—senescence, quiescence, differentiation, or apoptosis.","A non-significant result is explicitly not evidence against inheritance; the test only detects heritability when there is measurable variation in inactivity propensity along the lineage."],"fun_headline_variants":["Inheritance entropy shows cell-cycle exit is heritable in bone-marrow clones","Hereditary signal in cell-cycle exit detected via lineage topology entropy","Most bone-marrow clones inherit cell-cycle exit pattern, entropy shows","Lineage tree entropy exposes heritable cell cycle control in BMSC"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that after mixing up each generation, the only thing that has been removed is the tendency of daughters to resemble their mothers; if cells that stop dividing also cluster in branches for reasons unrelated to inheritance—such as a shared local environment or asymmetric division effects—then low entropy does not prove heredity.","fun_headline_variants_meta":{"raw":{"variants":["Inheritance entropy shows cell-cycle exit is heritable in bone-marrow clones","Hereditary signal in cell-cycle exit detected via lineage topology entropy","Most bone-marrow clones inherit cell-cycle exit pattern, entropy shows","Lineage tree entropy exposes heritable cell cycle control in BMSC"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001202,"raw_usage":{"total_tokens":4813,"prompt_tokens":787,"completion_tokens":4026,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":531,"completion_tokens_details":{"reasoning_tokens":3948}},"tokens_in":531,"tokens_out":4026,"duration_ms":20659,"temperature":1.0,"reasoning_tokens":3948,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T08:49:51.921511+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A decisive test would be to physically separate the two sister cells at a mitosis flagged with a large imbalance and grow their progeny in identical, randomized environments; if the branch difference in inactivity propensity does not persist, the heredity claim fails. Alternatively, a single-cell molecular readout of a candidate inherited mark taken at the flagged ancestral mitosis should show a systematic difference between the two daughter branches before any inactive cell appears; if no such mark exists, the proposed epigenetic mechanism is unsupported.","supporting_citations":[],"review_version":1}