BGPT: Paper Review: Clonal lineage tracing of innate immune cells in human cancer

Explore by Goal

Quick Explanation Copied

What this paper adds (skeptically)

This preprint introduces mtscATAC-seq + mtDNA lineage tracing (Mitotrek) to reconstruct clonal relationships across innate myeloid cell states in matched human tumor, adjacent non-involved lung, and blood, and then links those clones to epigenetic programs (e.g., DC3-like vs macrophage-biased fates) using the cohort scale of ~218k cells in NSCLC plus ~94k in ovarian cancer samples.

Main strength: explicit effort to reduce false lineage links (strict clone calling, removing high-frequency variants, discarding multi-clone assignments) and a stated accuracy benchmark using an external ground-truth dataset.

Key skepticism point: the biological conclusions (e.g., “fate predetermined before TME infiltration”) rely on cross-sectional sampling (no time series) and mtDNA constraints/technical artifacts that can occur in solid tumor single-cell contexts (the paper itself discusses ambient mtDNA and mitochondrial transfer).

Long Explanation

Paper Review (Visual-First, Skeptical & Evidence-based)

Title: Clonal lineage tracing of innate immune cells in human cancer

Data scale (as stated): NSCLC cohort includes n=218,715 cells from tumor, non-involved lung tissue (NILT), and PBMC, and ovarian cohort includes n=93,757 cells (plus clone-level recovery stats reported for both cancers).

1) Visual Overview of Core Claims

Method: mtDNA mutations are used as endogenous lineage markers, combined with single-cell chromatin (ATAC) to link clonality and cell-state/epigenetic programs in native tissues.
Mapping: TME-resident myeloid subsets (including macrophages and DC3) are clonally linked to circulating and tissue-infiltrating monocytes.
Fate bias: Distinct myeloid clones are DC-biased vs macrophage-biased; monocyte precursors for each fate class show different epigenetic profiles in blood, motivating “peripheral programming” / predisposition before TME infiltration.

2) Data & Clone Recovery (from the provided paper text)

Numbers used above come from the provided paper text: NSCLC profiled 218,715 cells and ovarian profiled 52,154 + 41,603 = 93,757 cells for the ovarian cohort in the abstract-like section, while clone counts are 5,146 (NSCLC) and 4,560 (ovarian) using Mitotrek.

3) Clone Size Structure (QC-passing cells → clones ≥3 and ≥10)

The paper reports: NSCLC 5,146 clones total, with 3,190 clones containing ≥3 cells and 573 containing ≥10 cells; ovarian 4,560 clones total, with 2,389 ≥3 cells and 287 ≥10 cells.

4) Mechanistic Interpretation vs What the Design Can Prove

What is relatively well-supported

Clonal linking across compartments is operationally supported by Mitotrek’s conservative clone assignment rules and internal consistency checks (e.g., higher probability of same-cell-type sharing for clone-mates than for permuted random grouping).
DC3 ↔ monocyte relationship is supported by both epigenetic motif patterns and clonal frequency/correlation analysis that specifically places DC3s closer to monocytes than to cDC2s.
Clone-intrinsic DC-biased vs macrophage-biased differentiation is consistent with the clone-level subtype composition clustering and the reported persistence of epigenetic differences in circulating monocytes for each bias class.

What cannot be strongly proven with the current design

Temporal ordering (e.g., “fate predetermined before TME infiltration”) is inferred from cross-sectional comparisons of PBMC vs tumor/NILT clones, not directly observed longitudinally. The authors explicitly list lack of temporal dynamics / time series as a limitation.
mtDNA artifacts / non-clonal sharing in solid tumor profiling are a known risk; the paper itself describes ambient mtDNA and horizontal mitochondrial transfer as contributors to spurious sharing patterns, hence its conservative filtering. Still, residual confounding is always possible when the lineage marker is not perfectly clonal and can be affected by sampling/capture.
Collapsed subclonal structure limits statements about hierarchical branching vs terminal differentiation because Mitotrek collapses highly correlated variants into single clones and does not attempt to resolve subclonal architecture.

5) Critical Appraisal Checklist (evidence quality & bias resistance)

Dimension	What the paper reports	What could still bias inference
Lineage-marker validity	mtDNA somatic mutations used as endogenous lineage markers; clone calling uses strict thresholds and discards variants too common within a sample.	Residual contamination/transfer could still connect unrelated cells; unknown sensitivity to cohort composition, capture efficiency, and tissue-specific ambient mtDNA levels.
Benchmarking	Mitotrek benchmarked on full-length scRNA-seq from single-colony HSPCs with known clonal identities; reported accuracies 91.3%/85.3% (assignment) with lower clone assignment rates due to conservative settings.	Benchmark tissue/context differs (HSPCs vs tumor solid tissue); accuracy under ambient mtDNA/transfer may deviate from benchmark.
Cell-state inference	Uses scATAC-seq for epigenetic accessibility and TF motif enrichment/activity inference (e.g., chromVAR).	Motif accessibility implies regulatory potential, not necessarily causal TF activity; ATAC sparsity and motif-to-gene mapping limitations.
Confounding & causality	Proposes fate bias may be shaped by systemic tumor signaling; differentiates clone-intrinsic vs tissue effects via clone composition comparisons.	Causality (tumor signals → peripheral reprogramming) not tested directly; correlational inference without perturbation. The authors acknowledge need for longitudinal/temporal data.

6) Useful Background Citations (to ground interpretation)

mtDNA heteroplasmy and its dynamics are foundational to the logic of mtDNA lineage tracing: mtDNA accumulates somatic mutations over time and exhibits heteroplasmy dynamics, which can be leveraged for lineage information.
Conservative clone calling explicitly addresses measurement noise from genetic drift and sequencing depth when binarizing heteroplasmy.
TF motif accessibility inference is linked to widely used computational approaches for scATAC TF activity estimation.

Note: Some “background” citations above are limited because the provided reference list in the prompt excerpt does not include DOI metadata for every foundational mtDNA/ATAC concept; therefore, I only made claims that were explicitly supported by the provided paper text or by citations with DOIs visible in the excerpt.

Recommendation for reading

If you care most about method validity, focus first on Mitotrek’s filtering/assignment design and its benchmark; if you care about biology, focus on the DC3 ↔ monocyte linkage and then examine how the “DC-biased vs macrophage-biased” distinction is established at the clone level (not just at the cell-type level).

Author Reviews (bespoke deep dives)

Feedback:

Updated: April 15, 2026