Unravelling genome-wide mosaic microsatellite mutations at single-cell resolution

The authors developed the BayesMonSTR algorithm to detect mosaic microsatellite mutations at single-cell resolution, revealing that aging human neurons accumulate a higher burden of longer STR deletions enriched at regulatory regions compared to other tissues.

The Gist

Imagine your body is a massive, bustling city made up of billions of individual houses (cells). For the most part, every house has the exact same blueprint (your DNA). However, over time, tiny errors creep into the blueprints of individual houses. This is called mosaicism—you become a patchwork quilt of slightly different cells.

One specific type of error happens in the "repetitive sections" of the blueprint. Think of these sections like a long string of beads where the pattern is "Red-Blue-Red-Blue-Red-Blue." Sometimes, the string gets tangled, a bead falls off, or an extra bead gets stuck in the middle. In genetics, these are called Short Tandem Repeats (STRs). They are notoriously messy and prone to mistakes, but until now, we've been blind to them when looking at single cells.

Here is a simple breakdown of what this paper achieved:

1. The Problem: The "Blurry Camera"

For years, scientists trying to find these repetitive errors in single cells were like photographers using a blurry camera. The technology was too noisy. The repetitive "Red-Blue" patterns confused the machines, making it impossible to tell if a mistake was a real biological event or just a glitch in the camera lens (technical error).

2. The Solution: The "Smart Detective" (BayesMonSTR)

The researchers built a new tool called BayesMonSTR. Think of this as a super-smart detective that doesn't just look at the photo; it investigates the crime scene from every angle.

• It reads the clues: It looks at the DNA sequence, the quality of the data, and how the DNA strands are linked together.
• It cross-checks: It compares the single cell's DNA against the person's "master blueprint" (bulk tissue) and even checks if the error shows up in the cell's RNA (the instructions the cell is currently using).
• It filters out noise: It uses a "lie detector" (a machine learning model) trained on gold-standard data to ignore the camera glitches and focus only on the real mutations.

3. The Discovery: The "Aging City"

Using this new detective tool, the team looked at three different types of cells in the human body:

• B Cells: The immune system's soldiers (they divide and reproduce often).
• Lung Cells: The air-exchange workers (they divide occasionally).
• Neurons: The brain's messengers (they stop dividing early in life and just sit there for decades).

The Big Surprise:
They expected the dividing cells (B cells and lung cells) to have the most errors because they copy their DNA so often. Instead, they found that neurons had the highest burden of these repetitive errors, especially as people got older.

• The "Long Deletion" Mystery: In aging neurons, the errors weren't just tiny single-bead mistakes. They were long deletions—entire chunks of the "Red-Blue" string were missing. It's as if the brain cells, over a lifetime of sitting still, started losing large sections of their instruction manuals.
• The "Repair Crew" Failure: They found that neurons with the most errors often had broken "repair crews" (mutations in DNA repair genes like PALB2). When the repair crew is tired or broken, the repetitive sections fall apart faster.

4. The Consequence: "Glitchy Instructions"

Why does this matter? These repetitive sections aren't just junk; they are often located in the control switches of the genome (promoters and enhancers) that tell genes when to turn on or off.

• The Analogy: Imagine a light switch that says "Turn on the lights." If a mutation deletes a few letters in the switch's code, the light might flicker, stay on too bright, or never turn on.
• The Alzheimer's Link: The researchers found that in patients with Alzheimer's, these long deletions were piling up specifically in the control switches of genes known to cause the disease (like APP).
• The Proof: They took these mutated switches and tested them in a lab dish. The mutated switches actually made the genes work differently (sometimes turning them up too high), proving that these tiny DNA errors can directly mess up how our brain cells function.

The Takeaway

This paper is a breakthrough because it finally gave us a clear, high-definition view of how our "repetitive" DNA breaks down as we age.

It tells us that aging isn't just about the cells dividing and making mistakes; it's also about the cells that stop dividing (like neurons) slowly accumulating long, damaging deletions in their control switches. This accumulation might be a hidden driver of neurodegenerative diseases like Alzheimer's, opening up new ways to understand and potentially treat these conditions.

In short: We built a better microscope, found that our brain cells are losing their "instruction manuals" as we age, and realized this loss might be a key reason why our brains get sick.

Technical Summary

1. Problem Statement

Short Tandem Repeats (STRs), or microsatellites, are highly mutable genomic regions (mutating ~100x faster than non-repeat regions) that play critical roles in gene regulation and are linked to numerous human diseases. However, detecting somatic mosaic STR mutations (mutations occurring post-zygotically in specific cells) at single-cell resolution has been a major bottleneck.

Existing challenges include:

• Technical Artifacts: PCR stutter (slippage during amplification), sequencing errors, and alignment ambiguities in repetitive regions.
• Biological Complexity: Interruptions within repeat motifs and the difficulty of distinguishing true somatic mutations from germline polymorphisms or amplification errors.
• Lack of Tools: Current genotyping methods (e.g., HipSTR, ExpansionHunter) are designed for germline variants or bulk sequencing and lack the sensitivity and specificity to resolve mosaic STR indels in single cells.

2. Methodology: BayesMonSTR

The authors developed BayesMonSTR, a robust computational framework designed to detect mosaic STR insertions, deletions (indels), and mismatches from single-cell whole-genome sequencing (scWGS) and single-nucleus ATAC-seq (snATAC-seq) data.

Key Algorithmic Components:

• Probabilistic Read Segmentation (HMM): Uses a Hidden Markov Model to segment sequencing reads into "STR," "flanking," and "interruption" states. Unlike traditional alignment, this incorporates base quality scores and reference genome context to accurately define STR boundaries despite sequencing errors.
• Haplotype Phasing: Statistically links STR alleles to neighboring heterozygous SNPs (hSNPs) to assign alleles to specific chromosomal haplotypes, increasing confidence in mutation calling.
• Hierarchical Bayesian Network: Integrates three layers of data:
  1. Population-level sequencing.
  2. Individual-level bulk sequencing.
  3. Single-cell sequencing.
     This allows for the calculation of genotype posteriors while accounting for stutter error models (geometric distribution) and amplification bias.
• Machine Learning Classification: A Random Forest (RF) model trained on over 50 read-level and genome-level features (including phasing discordance, stutter rates, and VAF) to distinguish true mosaic mutations from artifacts.
• Validation Strategy: The tool was validated using a "gold-standard" dataset generated from 29 single cells (breast tissue) with:
  • Duplicate single-cell libraries (DNA-level validation).
  • Single-cell RNA-seq from the same cells (RNA-level validation).
  • Deep bulk DNA sequencing from adjacent tissues.

3. Key Contributions

• First Genome-Wide Tool: BayesMonSTR is the first tool capable of hypothesis-free, genome-wide detection of somatic STR mutations at single-cell resolution.
• Superior Accuracy: It achieves significantly higher validation rates (~60–80%) compared to adapted germline callers like HipSTR and ExpansionHunter, which suffer from high false-positive rates in single-cell data.
• Multi-Modal Support: The framework extends to BayesMonSTR-ATAC (for snATAC-seq) and BayesMonSTR-BulkMonSTR (for bulk WGS), providing a unified toolkit for studying STR instability across different data types.
• Lineage Tracing: Demonstrated that mosaic STR mutations can be used for cost-effective in vivo lineage tracing, as phylogenies built from STR mutations recapitulate those built from single-nucleotide variants (SNVs).

4. Key Results

The authors applied BayesMonSTR to analyze 342 single cells from three tissue types: B cells (blood), lung epithelial cells, and prefrontal cortex (PFC) neurons from donors aged 0–106 years.

A. Mutation Burden and Aging Dynamics:

• Neurons Accumulate More Mutations: Neurons (post-mitotic) harbored significantly more mosaic STR mutations than mitotic B cells or lung epithelium.
• Age-Dependent Dichotomy:
  • 1-bp Indels: Accumulated with age in dividing cells (B cells, lung) but remained stable in neurons, suggesting 1-bp indels in neurons are acquired primarily during early development.
  • Long Indels (>5 bp): All tissues, especially neurons, showed a significant accumulation of longer indels with age. By age 80, ~84% of neurons carried at least one long STR indel.
• Deletion Bias: Neurons exhibited a strong bias toward deletions over insertions for long indels, unlike other tissues where frequencies were more balanced.

B. Mechanisms of Mutagenesis:

• DNA Repair Defects: A specific neuron with a massive burden of long indels (~50) carried a mosaic nonsynonymous mutation in the DNA repair gene PALB2. Neurons with mutations in DNA repair genes showed a marginally significant increase in STR indel burden.
• Repair Pathways: Long deletions were frequently flanked by microhomology, indicating Microhomology-Mediated End Joining (MMEJ) as a primary repair mechanism for double-strand breaks in aging tissues.

C. Functional Impact:

• Regulatory Enrichment: Mosaic STR indels were enriched in Transcription Start Sites (TSSs) and active enhancers of highly expressed genes.
• Alzheimer's Disease (AD) Link: In AD patients, STR indels were significantly enriched in the regulatory regions of AD-related genes (e.g., APP, CD2AP, PICALM).
• Functional Validation: Luciferase assays confirmed that specific mosaic indels in the TSS of APP (6-bp deletion) and MAP2K1 (4-bp deletion) significantly altered gene expression levels.

5. Significance

• New Landscape of Variation: The study reveals a previously unexplored layer of genomic instability where long STR deletions accumulate in aging neurons, potentially driving neurodegeneration.
• Disease Mechanism: It provides evidence that somatic STR mutations in regulatory regions can directly dysregulate gene expression, offering a novel mechanism for the etiology of Alzheimer's disease and other aging-related disorders.
• Tool Availability: By releasing BayesMonSTR as open-source software (Python/R, MIT License), the authors enable the broader scientific community to investigate the role of microsatellite instability in development, aging, and cancer without the previous technical barriers.

In summary, this work bridges the gap between repetitive genomic elements and single-cell biology, establishing that mosaic STR mutations are a pervasive, age-dependent, and functionally significant source of genomic variation in human tissues.