Robustness and reliability of single-cell regulatory multi-omics with deep mitochondrial mutation profiling

Abstract

The detection of mitochondrial DNA (mtDNA) mutations in single cells holds considerable potential to define clonal relationships at scale, coupled with information on cell state in humans. Previous methods focused on higher heteroplasmy mutations, which, while informative, are few in number and may be shaped by functional selection, providing limited lineage information and potentially introducing biases for tracing. Although more challenging to detect, intermediate- to low-heteroplasmy mtDNA mutations are valuable due to their high diversity, abundance, and lower propensity to selection. To enhance mtDNA mutation detection and facilitate fine-scale lineage tracing, we developed the single-cell Regulatory multi-omics with Deep Mitochondrial mutation profiling (ReDeeM) approach, an integrated experimental and computational framework. Here, we specifically address two analytical challenges central to single-cell mtDNA-based lineage analysis: the fidelity of variant-calling workflows and the reliability of phylogenetic inference. We demonstrate that, by leveraging consensus-based error correction, ReDeeMs mtDNA mutation calls achieve high fidelity, aligning with bona fide mutational signatures even for mutations supported by a single molecule per cell. We also developed an improved post-consensus filtering approach, termed "filter2" that systematically identifies and filters residual edge-enriched artifacts, even though these affect only a minority of mutation calls. To systematically validate ReDeeM, we recently conducted a lentiviral barcoding experiment in human HSCs, uniquely labeling each cell prior to expansion and differentiation to provide a ground truth for assessing lineage tracing accuracy1. Such validation demonstrate that the original ReDeeM analytic approach (filter1)2 robustly recovers true clonal structure at high resolution and recall. Including intermediate to low-heteroplasmy variants (<10% per cell) strongly improves lineage inference, whereas excluding mutations supported by a single molecule per cell removes true clonal signal and degrades recovery of true clones. Both filter1 and filter2 accurately recover true clones and outperform prior mtDNA-based lineage tracing approaches in ground-truth precision and recall, with filter2 providing additional gains in performance. Finally, while ReDeeM advances mtDNA-based lineage tracing by recovering true clones at high resolution, yet quantifying the phylogenetic uncertainty from mitochondrial inheritance remains unmet. To address this, we recently developed and validated MitoDrift1 as a next-generation, drift-aware framework for mtDNA lineage tracing that further strengthens phylogenetic inference and provides interpretable uncertainty estimates.