Low-heteroplasmy mitochondrial DNA mutations improve clonal reconstruction of human cells

Technical Summary: Low-heteroplasmy mitochondrial DNA mutations improve clonal reconstruction of human cells

Problem Statement
Reconstructing clonal relationships in human tissues is essential for understanding development, aging, and disease. While somatic mitochondrial DNA (mtDNA) mutations serve as endogenous single-cell barcodes, traditional lineage tracing has been limited by a reliance on high-heteroplasmy variants (typically >10%). Although these high-frequency variants are easier to detect, they are numerically scarce and may be shaped by selection, thereby capturing only a fraction of the available lineage information. Conversely, the more abundant low-heteroplasmy variants (<10%) have been difficult to distinguish from technical noise in single-cell sequencing, leading to their exclusion from lineage analyses. It remains untested whether these low-heteroplasmy observations encode genuine lineage history or are merely artifacts.

Methodology
The authors utilized a ground-truth benchmarking approach to evaluate the utility of low-heteroplasmy mtDNA mutations.

• Experimental System: Primary human hematopoietic stem and progenitor cells (HSPCs) from a healthy donor were transduced with a lentiviral lineage and RNA-recovery barcode (LARRY v2) library. These cells were expanded ex vivo and profiled using single-cell multiome sequencing with targeted capture of mtDNA and lentiviral barcodes. The lentiviral barcodes provided an independent, ground-truth definition of clonal identity ("lenti-clones").
• Variant Calling and Filtering: The study employed the ReDeeM pipeline for mtDNA mutation profiling. The authors introduced redeemR2.0, an updated post-consensus filtering workflow (termed filter2) that builds upon the original strategy (filter1). Key components of filter2 include:
  • 9-bp fragment-end trimming to remove edge-enriched artifacts.
  • A per-variant binomial goodness-of-fit test to retain variants whose cell-to-cell distributions deviate from random technical noise (FDR 0.05).
  • Removal of germline-like variants based on population frequency (MITOMAP) and exclusion of homoplasmic or blacklisted regions.
  • Retention of variants supported by as few as one endogenous unique molecular identifier (eUMI) per cell, provided they pass statistical filters.
• Lineage Inference: The authors used MitoDrift, a confidence-aware framework that models mitochondrial inheritance and heteroplasmy drift, to reconstruct lineage trees from the filtered mtDNA variant profiles.
• Benchmarking Metrics: Performance was assessed by comparing inferred clades against the ground-truth lenti-clones. Metrics included:
  • Recall: The fraction of ground-truth clones recovered by an inferred clade.
  • Precision: The fraction of inferred clades that matched a ground-truth clone.
  • Jaccard Similarity: Used as a threshold (≥0.5 and ≥0.75) to define a successful match between an inferred clade and a ground-truth clone.
  • Area Under the Precision–Recall Curve (AUPRC): Calculated across varying MitoDrift support thresholds.

Key Results

• Validation of Low-Heteroplasmy Calls: After applying the stringent filter2 workflow, mutation calls supported by single eUMIs or falling below 10% heteroplasmy retained the molecular hallmarks of genuine mtDNA mutations. These included a transition-dominated signature (transversion ratio ~0.03) and uniform distribution along read fragments, distinguishing them from background noise.
• Improved Clonal Recovery: Using the LARRY ground truth, the study demonstrated that retaining the full heteroplasmy spectrum significantly outperformed restrictive cutoffs.
  • Compared to a >10% heteroplasmy cutoff, using the full filter2 set increased the AUPRC from 0.322 to 0.640 at a Jaccard threshold of 0.5 (a 2-fold gain) and from 0.203 to 0.433 at a Jaccard threshold of 0.75 (a 2.1-fold gain).
  • Including 1-eUMI variants further improved performance, raising the AUPRC from 0.573 to 0.640 at the 0.5 threshold.
• Distribution of Lineage Information: Analysis of individual mutation calls revealed that low-heteroplasmy variants are not negligible noise but a substantial source of signal.
  • Among lineage-informative calls (those enriched within specific clones), 52.9% fell below 10% per-cell heteroplasmy, and 42.0% were supported by only 1–3 eUMIs per cell.
  • While the 1-eUMI and lowest-heteroplasmy groups were noisier than higher-support groups, they were significantly enriched for lineage-informative calls compared to a randomized background, confirming their collective value in clonal recovery.
• Filtering Strategy: The redeemR2.0/filter2 strategy dominated all other tested approaches (including filter1 and mgatk-like settings lacking UMI-based error correction) in precision-recall performance.

Significance and Claims
The paper establishes that low-heteroplasmy mtDNA mutations constitute an abundant, bona fide record of clonal history. The authors conclude that:

1. Expanding the Signal: Restrictive heteroplasmy thresholds (e.g., >10%) incur a substantial sensitivity cost. Including variants below this threshold, even those with low molecular support, substantially expands the pool of usable lineage markers.
2. Collective Utility: While individual low-molecule calls should be interpreted with caution, their collective inclusion significantly improves the recovery of true clonal relationships.
3. Best Practice: The authors recommend redeemR2.0/filter2 as the default preprocessing strategy for mtDNA-based single-cell lineage tracing. This approach pairs rigorous molecule-level filtering with the inclusion of the full heteroplasmy spectrum, enabling high-resolution clonal reconstruction in human tissues without the need for genetic engineering.
4. Framework: The study provides a ground-truth-anchored framework for best practices, demonstrating that with appropriate error correction (eUMI consensus and statistical filtering), the "noise" of low-heteroplasmy variants can be converted into high-value lineage data.

The authors note that while the ground truth (lentiviral barcoding) is imperfect due to potential barcode silencing or dropout, the observed gains in lineage recovery are robust. They emphasize that while the specific parameters may vary across cell types, the principle of leveraging low-heteroplasmy variants with rigorous filtering is broadly applicable to expanding the resolution of natural mitochondrial variation in human cell history.