Modeling mitochondrial inheritance enables high-precision single-cell lineage tracing in humans

The paper introduces MitoDrift, a probabilistic framework that leverages somatic mitochondrial DNA mutations to achieve high-precision, engineering-free single-cell lineage tracing in humans, thereby revealing critical insights into clonal dynamics during aging and disease that were previously undetectable.

The Gist

The Big Idea: Reading the "Family Tree" Written in Your Cells

Imagine your body is a massive, bustling city. Every cell in that city is a citizen. As the city grows and changes over time, these citizens divide and multiply. Sometimes, they make tiny mistakes when copying their instruction manuals (DNA). These mistakes are like typos.

Usually, we think of typos as bad. But in this study, scientists realized these typos are actually natural barcodes. If a cell makes a typo, all its children inherit that typo. If a grandchild makes a new typo, it carries both the old one and the new one. By reading these typos, we can trace who is related to whom, effectively reconstructing a family tree for every cell in your body.

The problem? Mitochondria (the tiny power plants inside our cells) are messy. They have hundreds of copies of their own DNA, and they shuffle them around randomly when cells divide. It's like trying to trace a family tree when the family keeps losing pages of the history book or swapping pages with neighbors. This makes the family tree look blurry and confusing.

Enter MitoDrift. This is a new computer program (a mathematical framework) that acts like a super-smart detective. It doesn't just look at the typos; it understands the rules of how these typos get shuffled and lost. It cleans up the blurry picture to give us a crystal-clear family tree.

---

How It Works: The "Weather Forecast" Analogy

Think of trying to predict the weather. You have some data (clouds, wind speed), but it's noisy. You also know the laws of physics (air pressure, temperature).

• Old Methods: Just looked at the clouds and guessed the weather. They often got it wrong because they didn't account for the physics of how air moves.
• MitoDrift: It uses a "Hidden Markov Model." Imagine it's a weather forecaster who knows the laws of physics perfectly. It looks at the messy data (the clouds) and asks, "Given the laws of physics, what is the most likely weather pattern that created this?"

In the paper, MitoDrift treats the shuffling of mitochondrial DNA like a game of chance (called Wright-Fisher drift). It calculates the probability of how a mutation could have drifted from a parent cell to a child cell. By doing this, it can say, "I am 90% sure these two cells are siblings," or "I am only 10% sure, so let's not guess."

It then builds a tree and collapses the weak branches. Think of it like a sculptor chipping away the parts of the statue that are too shaky to be sure about, leaving only the solid, confident parts of the family tree.

---

What They Found: Three Big Discoveries

The researchers tested this tool on human blood cells and cancer cells. Here is what they found:

1. Aging Makes the Blood "Oligoclonal" (The "One-Shop Town" Effect)

As we get older, our blood stem cells (the factories that make blood) tend to lose diversity.

• The Analogy: Imagine a town with 100 different bakeries making bread. When the town is young, everyone buys from different bakeries. As the town ages, one or two bakeries get huge and dominate the market, while the others close down.
• The Finding: In older people, the "blood town" is dominated by a few super-sized clones of cells. Interestingly, this happened differently for different types of blood cells. The "myeloid" cells (which fight infection) became very dominated by a few clones, but T-cells (immune memory) stayed diverse. It's like the bakery town changed, but the library (T-cells) kept all its old books.

2. Stress Makes Cells "Grow Big" (The "Stressed-Out Boss")

They found that cells with a specific "stress response" program (controlled by a group of proteins called AP-1) were the ones that ended up taking over the blood supply in older people.

• The Analogy: Imagine a factory worker who is constantly stressed and working overtime. Eventually, that worker gets promoted and takes over the whole factory floor, pushing out the relaxed workers.
• The Finding: Cells that are good at handling stress seem to have a survival advantage as we age, leading them to expand and dominate the system.

3. Cancer Therapy: The "Whac-A-Mole" Game

They looked at patients with Multiple Myeloma (a blood cancer) before and after treatment.

• The Analogy: Doctors hit the cancer with chemotherapy (a giant hammer). The big, obvious cancer clones get smashed. But sometimes, a few tiny, hidden "mole" clones survive.
• The Finding: Old methods (looking at big genetic changes) couldn't see these tiny survivors. But MitoDrift saw them! It showed that the cancer didn't just shrink; it remodeled. The survivors were often cells that had a specific "adhesion" program (like having sticky feet) that helped them hide in the bone marrow and resist the drugs.
• The Takeaway: MitoDrift can spot the "bad actors" that survive treatment, helping doctors understand why cancer comes back.

---

Why This Matters

Before this, trying to trace cell family trees in humans was like trying to read a book written in a language you don't speak, with half the pages missing.

MitoDrift is like a translator that also fills in the missing pages based on logic. It allows scientists to:

1. See the invisible: Find tiny cell groups that other methods miss.
2. Connect the dots: Link a cell's history (its family tree) to what it is doing right now (its behavior).
3. Understand aging and disease: See exactly how our bodies change over time and how cancer learns to hide from medicine.

In short, this tool gives us a high-definition map of our cellular history, helping us understand how we age and how to fight diseases like cancer more effectively.

Technical Summary

1. The Problem

Somatic mutations in mitochondrial DNA (mtDNA) serve as natural, engineering-free barcodes for tracing cell lineages in human tissues. However, reconstructing accurate lineage trees from mtDNA data faces significant biological and technical challenges:

• Complex Inheritance Dynamics: Unlike nuclear DNA, mtDNA exists in multiple copies per cell. During cell division, these copies segregate stochastically, leading to heteroplasmy drift (random fluctuations in mutation frequency) where variants can drift toward fixation or extinction.
• Incomplete Sampling & Noise: Single-cell sequencing often suffers from sparse sampling, dropout events, and technical artifacts (e.g., false positives/negatives), which confound the detection of true clonal relationships.
• Limitations of Existing Methods: Current phylogenetic reconstruction methods often treat mtDNA variants as static markers or ignore the population-genetic dynamics of heteroplasmy. This results in lineage trees with low precision, spurious splits, and an inability to quantitatively link clonal history to cell states (transcriptional/epigenetic programs).

2. Methodology: MitoDrift

The authors present MitoDrift, a probabilistic framework designed to model mtDNA inheritance as an intracellular population-genetic process observed through noisy single-cell measurements.

• Core Model (WF-HMT): MitoDrift utilizes a Wright–Fisher Hidden Markov Tree (WF-HMT).
  • Latent States: The unobserved heteroplasmy state at every internal node of the lineage tree is modeled as a latent random variable on a discretized Variant Allele Frequency (VAF) grid.
  • Transition Model: Along tree edges, heteroplasmy evolves via a Wright–Fisher drift process, modeled as a Markov chain with transition matrices derived from effective mitochondrial population size ($N$) and generations ($g$).
  • Observation Model: The observed allele counts at leaf cells (single cells) are modeled as a binomial distribution with an added error term ($\epsilon$) to account for sequencing noise, contamination, and mapping artifacts.
• Inference Workflow:
  1. Likelihood Computation: Computes the tree likelihood via sum-product message passing (belief propagation) on the tree structure.
  2. Parameter Learning: Uses Expectation-Maximization (EM) to learn drift parameters ($g, N$) and error rates ($\epsilon$).
  3. Topology Search: Employs Metropolis–Hastings Markov Chain Monte Carlo (MCMC) to sample tree topologies. It uses local Nearest Neighbor Interchange (NNI) proposals and caches dynamic programming quantities to efficiently compute likelihood ratios without re-running full belief propagation for every proposal.
  4. Confidence Refinement: Calculates posterior support for each branch. Low-confidence branches (below a threshold $\tau$) are collapsed, yielding a confidence-refined tree that prioritizes accurate clades while acknowledging unresolved regions.
• Downstream Analysis: The refined trees are integrated with multi-omics data (scRNA-seq, scATAC-seq) using tools like PATH (for lineage-state coupling) and SCENIC+ (for regulon heritability) to link ancestry with cell state.

3. Key Contributions

• First Probabilistic Framework for mtDNA Drift: MitoDrift is the first method to explicitly model the stochastic drift of heteroplasmy and measurement noise within a unified probabilistic framework for single-cell lineage tracing.
• High-Precision Benchmarking: Demonstrated superior performance over existing methods (Neighbor-Joining, UPGMA, MERLIN, Signac) by benchmarking against orthogonal ground truths:
  • Lentiviral Barcoding (LARRY): In vitro expansion of human HSCs with matched LARRY barcodes.
  • Whole-Genome Sequencing (WGS): Single-colony WGS of human hematopoietic stem cells (HSCs) providing nuclear SNV-based ground truth.
• Quantitative Lineage–State Analysis: Established a scalable platform to quantitatively link clonal history to transcriptional and epigenetic programs, moving beyond qualitative associations.

4. Key Results

A. Validation and Benchmarking

• Precision vs. Recall: In the LARRY benchmark, MitoDrift achieved 77% overall clade precision (vs. 55% for NJ and 28% for UPGMA) while maintaining high clone recall (75% of clones recovered with Jaccard similarity $\ge$ 0.5).
• Confidence Calibration: Higher posterior branch confidence strongly correlated with empirical precision, allowing researchers to tune the resolution of the tree by adjusting the confidence threshold ($\tau$).
• Robustness: The method maintained high accuracy across varying clone sizes (strongly, moderately, and weakly expanded clones) and outperformed other methods in precision-recall curves.

B. Application to Human Hematopoiesis (Aging)

• Age-Associated Oligoclonality: Applied to healthy donors (ages 29–81), MitoDrift revealed a significant decline in mtDNA-based clonal diversity (mtSDI) with age, consistent with WGS-based findings.
• Cell-Type Specificity: The decline in diversity was not uniform. Myeloid cells, B cells, and erythroid progenitors showed marked reductions in clonal diversity with age, whereas T cells maintained diversity. This suggests that aged HSCs exhibit lineage-biased output (myeloid-biased), while T-cell diversity is preserved due to the longevity of memory T cells.
• Heritable Regulatory Programs: Identified stable, heritable gene regulatory programs in HSCs, specifically AP-1/stress-response and stemness/lymphoid priming modules.
• Clonal Expansion Drivers: Found a significant positive association between AP-1 transcriptional activity and clone size in aged donors, suggesting that inflammatory/stress-associated programs may drive clonal dominance in aging.

C. Application to Multiple Myeloma (Cancer)

• Fine-Grained Clonal Remodeling: In longitudinal samples from three multiple myeloma patients (pre- and post-therapy), MitoDrift resolved clonal structures and therapy-induced remodeling that were undetectable by Copy Number Alteration (CNA) analysis alone.
• Subclonal Heterogeneity: Within a CNA-defined subclone (1q-gain), MitoDrift identified further subclades with distinct temporal distributions, revealing subclonal heterogeneity in drug sensitivity.
• Phylogeny–State Analysis:
  • Distinguished heritable cell states (e.g., stress-tolerant, signaling-dampened) from plastic states (e.g., proliferative).
  • Identified that pre-treatment states characterized by CD44+ adhesion/migratory signatures (P1B) were phylogenetically closest to post-treatment resistant cells, linking specific gene regulatory programs to therapy resistance.

5. Significance

• Paradigm Shift: Moves single-cell lineage tracing from qualitative, descriptive comparisons to quantitative, high-precision inference by accounting for the biological reality of mitochondrial inheritance.
• Clinical Insights: Provides a new lens to study tissue homeostasis, aging, and cancer evolution. It reveals how specific regulatory programs (like AP-1) drive clonal expansion in aging and how subclonal plasticity contributes to therapy resistance in cancer.
• Scalability: The framework is applicable to large-scale single-cell multi-omics datasets, enabling the reconstruction of lineage trees in primary human tissues without the need for exogenous barcoding.
• Future Directions: The authors suggest that integrating mtDNA with nuclear mutations and epigenetic markers (methylation) could further enhance temporal resolution, particularly for resolving deep ancestral splits.

In summary, MitoDrift establishes a rigorous mathematical foundation for interpreting mtDNA mutations as lineage recorders, enabling researchers to decode the "clonal history" of human cells and its direct impact on cell fate, disease progression, and therapeutic response.