High rate of mutation and efficient removal by selection of structural variants from natural populations of Caenorhabditis elegans

Using long-read sequencing on *C. elegans* mutation accumulation lines, this study estimates a structural variant mutation rate of approximately 0.03 per genome per generation and demonstrates that natural selection efficiently removes these variants from both exonic and intergenic regions in wild populations.

The Gist

Imagine your genome (your body's instruction manual) as a massive, intricate library. Usually, when we talk about "typos" in this library, we focus on single-letter mistakes (like changing an 'A' to a 'G'). But this paper is about the big, messy structural changes: entire paragraphs getting deleted, whole chapters being pasted in the wrong place, or pages being ripped out and replaced with gibberish. These are called Structural Variants (SVs).

Here is the story of what the researchers found, told simply:

1. The Problem: The "Blurry Camera"

For a long time, scientists couldn't see these big typos well. They were using "short-read" sequencing, which is like trying to read a book by taking photos of individual words. If a whole paragraph is missing or a new chapter is inserted, a photo of a single word doesn't tell you the story is broken.

To fix this, the researchers used long-read sequencing (PacBio). Think of this as switching from a blurry snapshot to a high-definition, wide-angle drone shot. Now, they could see the whole paragraph and the whole chapter at once.

2. The Experiment: The "Copy-Paste" Game

The scientists set up a massive experiment with C. elegans (tiny, transparent worms).

• The Setup: They took a few "ancestor" worms and started 6 separate family lines.
• The Rule: For about 250 generations, they forced each line to reproduce by picking just one single worm to be the parent of the next generation.
• The Goal: This is like a game of "Telephone" where you only pass the message through one person at a time. It minimizes the "noise" of natural selection (where bad mutations get weeded out) and lets you see exactly what new typos happen naturally.

3. The Findings: How often do the big typos happen?

When they compared the final "great-grandchildren" worms to the original ancestors, they found:

• Frequency: Big structural changes happen about 10 times less often than single-letter typos, but they are still surprisingly common. Roughly 1 in every 30 generations, a worm gets a major structural mutation.
• The "Noise" Factor: Because these big changes are hard to spot, the computer software made a lot of mistakes (false alarms). The researchers had to act like human editors, looking at the data "with their own eyes" (using a tool called IGV) to verify every single change. It was tedious work, like proofreading a 1,000-page manuscript line-by-line.

4. The Twist: Nature is a Strict Editor

Here is the most surprising part. In the "Telephone game" (the lab), these big typos happened frequently. But when the researchers looked at wild worms living in nature, they found very few of these big typos.

The Analogy: Imagine a factory that produces cars.

• The Lab: The factory makes 100 cars a day, and 10 of them have a missing door or a roof glued on backward.
• The Road (Nature): You look at the cars driving on the highway, and you see almost no cars with missing doors.
• The Conclusion: The factory makes the mistakes, but the road (natural selection) immediately crashes the broken cars. They don't survive.

The study found that nature is extremely efficient at removing these big structural errors. It doesn't just remove them from the "important" parts of the genome (like genes that code for proteins); it also removes them from the "junk" areas (intergenic regions) that scientists used to think didn't matter.

5. The "Hotspots" and the "Junk" Myth

• Hotspots: The researchers found that these big typos love to happen in specific "messy" areas of the genome—places full of repeating patterns (like "ATATATAT"). It's like a printer jamming more often when the paper has a weird texture.
• The "Junk" Myth: Because nature removes these typos even from the "junk" DNA, the paper suggests that maybe that DNA isn't junk after all. If changing the distance between two "junk" sections breaks the organism, then that section must be doing something important, like holding a regulatory switch in place.

Summary

This paper tells us that:

1. Big genetic errors happen more often than we thought, but we needed better "cameras" (long-read sequencing) to see them.
2. Nature is a ruthless editor. It quickly deletes these big errors from the wild population, even in areas we thought were unimportant.
3. The "Junk" DNA might actually be useful. If nature fights so hard to keep the "junk" sections the same size, they probably have a job to do.

In short: Our genomes are constantly trying to tear themselves apart with big structural changes, but evolution is constantly gluing them back together, keeping us alive.

Technical Summary

1. Problem Statement

Structural variants (SVs)—including large insertions, deletions, inversions, and translocations—are a major source of genetic diversity and are associated with significant phenotypic effects and human diseases. However, their mutational properties (rates, spectra, and fitness effects) remain poorly characterized compared to single nucleotide variants (SNVs) and short indels.

• Technical Barrier: SVs are difficult to detect using standard short-read sequencing because reads cannot span complex or repetitive regions, leading to ambiguous mapping.
• Knowledge Gap: There is a lack of robust estimates for spontaneous SV mutation rates and the strength of natural selection acting against them, particularly in non-coding regions. This gap biases estimates of the Distribution of Fitness Effects (DFE) and the "omnigenic" model of complex traits.

2. Methodology

The authors employed a Mutation Accumulation (MA) experimental design combined with long-read sequencing to overcome detection limitations.

• Experimental Design:

  • Organism: Caenorhabditis elegans (two strains: N2 and PB306).
  • Protocol: Six MA lines were propagated for approximately 238–250 generations under minimal selection (bottlenecking to a single hermaphrodite every generation).
  • Sequencing: Whole-genome sequencing was performed using Pacific Biosciences (PacBio) long-read technology (Sequel and Sequel 2).
  • Comparators: Ancestral progenitors and four wild isolates (CB4856, QX1211, JU1088, and N2) were also sequenced to assess standing variation.

• Variant Calling Pipeline:

  • Workflow: A modified SMRT-SV2 pipeline was used. Raw PacBio subreads were aligned to the reference genome using minimap2, followed by de novo assembly of 60-kb tiles into "pseudo-reads" (contigs).
  • Variant Calling: The pseudo-reads were re-aligned, and SVs were called using PBSV (Pacific Biosciences SV caller).
  • Validation: Due to high false-positive rates inherent in de novo mutation calling, every candidate SV was manually verified using IGV (Integrative Genomics Viewer) and JBrowse, comparing pseudo-reads, raw reads, and independent Illumina data.
  • Definition: SVs were defined as variants >30 bp (a lower threshold than the standard 50 bp, justified by long-read capabilities).

• Error Estimation:

  • False Positives (FP): Identified via manual inspection.
  • False Negatives (FN): Estimated by simulating SVs (insertions, deletions, inversions) of various sizes (100 bp to 100 kb) into the reference genome and measuring the "recall rate" of the pipeline.

3. Key Contributions

• First Long-Read SV Rate in MA Lines: Provides the first robust estimate of spontaneous SV mutation rates in C. elegans using long-read sequencing, overcoming the biases of short-read technologies.
• Refined Detection Threshold: Demonstrates that long-reads can reliably detect indels between 30–50 bp, a range often missed by short-read studies.
• Selection on Intergenic Regions: Offers evidence that natural selection efficiently removes SVs even from intergenic (non-coding) regions, challenging the "junk DNA" hypothesis.
• Hotspot Identification: Identifies that a significant fraction of SVs occur in pre-existing low-complexity repeat regions (mutational hotspots).

4. Key Results

A. Mutation Rates and Spectrum

• SV Mutation Rate: Estimated at ~0.03 SVs per genome per generation.
  • This is approximately 1/10th the SNV rate and 1/4th the short indel rate.
  • SVs constitute roughly 7.5% of all spontaneous mutations (or ~6% if using the >50 bp threshold).
• Spectrum:
  • Types: 12 insertions, 28 deletions, 0 inversions were confirmed. Deletions outnumbered insertions 2:1.
  • Size: Spontaneous SVs were marginally larger than standing variants (Mean MA size: 808 bp vs. WI: 602 bp).
  • Transposons: 4 of the 40 SVs were transposon-related (CER1 and RTE insertions), suggesting TEs are a source of large insertions but are heavily purged by selection.
• False Positives: The initial signal-to-noise ratio was low (40 true positives vs. 52 false positives). However, after excluding one atypical line (MA517) with low coverage, the ratio improved to ~2:1.

B. Detection Accuracy (Recall Rates)

• Recall Performance:
  • **Small Deletions (<10 kb):** High recall (>90%).
  • Insertions: High recall for small sizes (~90% at 100 bp) but drops sharply as size increases.
  • Inversions: Very low recall for small inversions (100 bp) but better for large ones (100 kb).
  • Reference Bias: Recall rates were lower in wild isolates (especially the divergent QX1211 strain) compared to the N2 reference, indicating reference bias affects detection.

C. Distribution and Selection

• Genomic Distribution:
  • Spontaneous SVs are enriched in chromosomal arms (where recombination is high and low-complexity DNA is overrepresented) compared to centers.
  • Standing variation in wild isolates mirrors this distribution, suggesting the pattern is driven by both mutation bias (higher in arms) and selection.
• Fitness Effects (Selection Strength):
  • Using relative diversity ($w^*$), the study found strong purifying selection against SVs:
    • Exons: $w^* \approx 0.14$ (14% of neutral expectation).
    • Introns: $w^* \approx 0.30$.
    • Intergenic Regions: $w^* \approx 0.52$.
  • Significance: The fact that intergenic SVs are reduced to ~50% of neutral expectations implies that natural selection effectively removes SVs from non-coding regions, suggesting these regions are functionally constrained (e.g., regulatory elements).

D. Mutational Hotspots

• Repeat-Associated SVs: ~22.5% (9/40) of SVs occurred in loci with pre-existing low-complexity repeat variation in the ancestor.
• Manual Correction: Many of these required manual correction because automated callers (PBSV) failed to distinguish between a new mutation and a change in an already polymorphic repeat region. This highlights that loci with existing SVs are prone to recurrent SVs.

5. Significance and Implications

• Revising the DFE: Ignoring SVs in evolutionary models artificially inflates the perceived importance of SNVs and small indels. SVs contribute significantly to the mutational load and the distribution of fitness effects.
• "Junk DNA" Re-evaluation: The efficient removal of SVs from intergenic regions suggests that a larger fraction of the genome is functionally relevant than previously thought based on SNP data alone. Regulatory distances and structural integrity of non-coding DNA appear to be under selection.
• Human Genetics: Given the similarity in effective population size ($N_e$) between C. elegans and humans, these findings suggest that rare, large-effect SVs likely contribute significantly to human complex traits and diseases, but are often missed by GWAS focused on SNPs.
• Methodological Benchmark: The study establishes a rigorous workflow (Long-read + Assembly + Manual Curation) necessary for accurate de novo SV detection, serving as a template for future mutation accumulation studies in other organisms.

In conclusion, the paper demonstrates that while SVs occur at a lower rate than SNVs, they have a disproportionate impact on genomic diversity and are subject to strong purifying selection across the entire genome, including non-coding regions.