Sequence effects on patterns of variation and DNA strand asymmetries observed from whole-genome sequenced UK Biobank participants

This study analyzes whole-genome sequence data from 500,000 UK Biobank participants to reveal how pentanucleotide contexts and mutational signatures influence variant frequencies and to identify unexplained, chromosome-specific DNA strand asymmetries that suggest the existence of currently unknown strand-specific mechanisms affecting mutation and selection.

The Gist

Imagine the human genome as a massive, ancient library containing 3 billion books (our DNA). Each book is written in a language of just four letters: A, C, G, and T. Over millions of years, typos (mutations) have crept into these books. Most of these typos are harmless, but some are so bad that the "editors" of evolution (natural selection) delete them immediately. Others are so rare they haven't been seen before.

This paper is like a detective story where the author, David Curtis, uses a massive new dataset from the UK Biobank (500,000 people's DNA) to figure out why certain typos happen more often than others, and why some typos survive while others vanish.

Here is the breakdown of the findings using simple analogies:

1. The "Neighborhood" Matters (Context is King)

Imagine you are writing a story. If you accidentally write a typo, the likelihood of it being noticed depends on the words around it.

• The Finding: The paper shows that a mutation isn't just about the letter that changed; it's about the "neighborhood" of letters surrounding it.
• The Analogy: Think of a typo like a typo in a sentence. If you write "The cat sat on the mat," changing the 't' to a 'k' (The cak) is obvious. But if you change a letter in a word that looks like "k," it might blend in. The study found that knowing the five-letter neighborhood (pentanucleotide) around a mutation helps predict how often that mutation happens with incredible accuracy (96% correlation). It's like knowing that a specific type of typo is 10 times more likely to happen if it's surrounded by the letters "C" and "G" than if it's surrounded by "A" and "T."

2. The "Singleton" vs. The "Popular Kid" (Mutation vs. Selection)

The study looked at two groups of typos:

• Singletons: Typos seen in only one person. These are like brand-new errors that just happened. They tell us about the mutation machine (how DNA copying goes wrong).
• Common Variants (SNPs): Typos seen in many people. These are the "survivors." They tell us about selection (which errors are allowed to stay).

The Big Surprise:

• The "CG" Trap: There is a specific pair of letters, C and G, that are like a "trap." When a C changes to a T in this specific neighborhood, it happens less often as a new mutation (it's rare). However, if it does happen and survives, it becomes very common in the population.
• The Analogy: Imagine a factory that makes toys. The "CG" machine is very careful and rarely makes a red toy (C>T mutation). But, if a red toy does get made, it turns out to be a very popular, durable toy that everyone wants to keep. Conversely, other types of mutations happen frequently but are often "defective" and thrown away by the quality control team (natural selection).

3. The "Left-Handed" vs. "Right-Handed" Bias (Strand Asymmetry)

DNA is a double helix, like a zipper. It has two sides: a "plus" strand and a "minus" strand. Usually, we assume the zipper works the same on both sides.

• The Finding: The study found that the DNA zipper is not symmetrical. Some typos happen much more often on the "plus" side, while others happen on the "minus" side.
• The Chromosome Split: Here is where it gets weird. The author found that most chromosomes (like 1, 2, 3...) all agree on which side is "plus" and which is "minus." But, a specific group of five chromosomes (10, 14, 19, 21, 22) are doing the opposite.
• The Analogy: Imagine a city where everyone drives on the right side of the road. Suddenly, you find five specific neighborhoods where everyone drives on the left. The study found this "driving on the left" pattern in those five chromosomes. The researchers checked if this was because those neighborhoods had more schools or hospitals (genes), but it wasn't. The reason is still a mystery, like a secret traffic rule we haven't discovered yet.

4. The Reference Book Itself is Biased

The study didn't just look at people's DNA; it looked at the "Reference Genome" (the master copy of the human book used by scientists).

• The Finding: Even the master copy has a bias. Certain five-letter sequences appear way more often on the "plus" side of the master book than the "minus" side.
• The Analogy: Imagine if you found that the word "TTCGT" appeared 670,000 times on the left page of a dictionary, but only 460,000 times on the right page. This suggests that the process of writing the master dictionary itself had a bias, or that nature prefers these words on one side over the other.

Why Does This Matter?

• Understanding Cancer: Cancer is essentially a book full of typos. By understanding the "neighborhoods" where typos happen most, we can better understand how cancer starts.
• Predicting Disease: If we know that certain typos are "survivors" (common) and others are "defects" (rare), we can better predict if a new genetic change found in a patient is dangerous or harmless.
• The Mystery: The biggest takeaway is that we still don't fully understand the "molecular machinery" that copies our DNA. There are hidden rules (like the five-chromosome split) that scientists haven't figured out yet.

In a nutshell: This paper is a massive census of genetic typos. It reveals that the "neighborhood" of DNA letters dictates how often mistakes happen, that some mistakes are surprisingly resilient, and that our DNA has a mysterious "left-right" bias that changes depending on which chromosome you are looking at. It's a reminder that even in the most basic building blocks of life, there are still deep, unsolved mysteries.

Technical Summary

1. Problem Statement

The study addresses the complex interplay between local DNA sequence context, mutation rates, natural selection, and DNA strand asymmetry in the human genome. While mutational signatures in cancer and de novo mutations in trios are well-studied, there is a need to understand how short DNA sequences (pentanucleotides) influence:

• The probability of specific mutations occurring (mutation rates).
• The likelihood of these mutations being retained in the population versus being purged by negative selection.
• The existence of strand-specific biases (asymmetries between the plus and minus DNA strands) in both mutation processes and the reference genome itself.

The author utilizes the massive scale of the UK Biobank's 500,000 whole-genome sequenced (WGS) participants to move beyond simple trinucleotide contexts and investigate the effects of full pentanucleotide backgrounds on variant frequencies and strand biases.

2. Methodology

• Data Source: Whole-genome sequencing data from 500,000 UK Biobank participants (510 million singletons, 240 million doubletons, and 411 million SNPs).
• Variant Classification: Variants were categorized into singletons (likely recent mutations) and SNPs (allele count > 2, representing standing variation).
• Contextual Analysis: Every single-base substitution (SBS) was analyzed within its pentanucleotide context (two flanking bases upstream and downstream).
  • Frequency Calculation: Variant counts were normalized by the frequency of the background pentanucleotide in the GRCh38 reference genome to calculate mutation probabilities independent of background sequence abundance.
• Statistical Modeling:
  • Logistic Regression: Used to model the odds of a specific variant occurring (for singletons) or being retained (SNP/Singleton ratio) based on the central base and flanking nucleotides (from core to full pentanucleotide).
  • Mutational Signature Decomposition: Observed singleton counts (96 trinucleotide types) were fitted against 86 COSMIC reference signatures using constrained linear regression to identify the best-fitting combination of signatures.
  • Strand Asymmetry Analysis: Log-ratios of variant frequencies (plus strand vs. minus strand) were calculated. Principal Component Analysis (PCA) and correlation matrices were used to compare these asymmetries across chromosomes and against gene burden asymmetry.
• Reference Genome Analysis: Counts of pentanucleotide sequences in the reference genome were compared between strands to identify intrinsic sequence asymmetries.

3. Key Contributions

• Pentanucleotide Resolution: Demonstrated that while trinucleotide contexts explain most variation, the full pentanucleotide background significantly improves the prediction of variant frequencies (correlation $R$ increases from 0.92 to 0.96) and selection ratios ($R$ increases from 0.94 to 0.96).
• Chromosome-Specific Strand Bias: Identified a distinct group of five chromosomes (10, 14, 19, 21, 22) where strand asymmetry patterns for singleton variants are negatively correlated with the rest of the genome, suggesting chromosome-specific mutational drivers.
• Selection vs. Mutation Decoupling: Showed that while mutational strand asymmetries vary by chromosome, selection-related strand asymmetries (retention of variants) and reference genome sequence asymmetries are consistent across all chromosomes.
• Reference Genome Asymmetry: Discovered that the reference genome itself contains significant strand asymmetries for specific pentanucleotides (e.g., TTCGT appears ~673k times on the plus strand vs. ~465k on the minus), independent of variant calling.

4. Key Results

A. Sequence Context and Mutation Rates

• Context Dependence: The frequency of singleton variants is strongly driven by the central base substitution but significantly modulated by flanking bases.
• C>T in CG Context: Singleton C>T variants are less frequent in the CG context compared to other contexts. However, for common variants (SNPs), C>T in the CG context is more frequent.
  • Interpretation: This suggests that while C>T mutations in CG contexts may be suppressed by repair mechanisms (or the mutation rate is lower), those that do occur are less subject to negative selection and are more likely to be retained in the population compared to C>T mutations in other contexts.
• Mutational Signatures: The distribution of singleton variants could be well-approximated ($R=0.82$) by a linear combination of just five mutational signatures derived from cancer genomes, indicating that somatic mutational processes share mechanistic roots with germline mutation patterns.

B. Selection Pressures

• Retention Ratios: The ratio of SNPs to singletons varies drastically by sequence context.
  • Variants like C[G>A]X (complementary X[C>T]G) have extremely high retention odds (OR > 14), indicating they are highly tolerated.
  • Conversely, C[C>A]X variants have low retention odds (OR ~0.5–0.7), suggesting they are deleterious.
• Pentanucleotide Effects: Flanking bases can double or halve the probability of a variant being retained. For example, TA[C>T]GC has a retention OR of 52.7, while TA[G>T]GG has an OR of 0.237.

C. Strand Asymmetries

• Mutational Asymmetry: Most chromosomes show positive correlation in strand bias patterns. However, chromosomes 10, 14, 19, 21, and 22 form a cluster with negative correlation to the rest. This discrepancy is not explained by gene content asymmetry (gene count or transcript length).
• Selection Asymmetry: Unlike mutational asymmetry, the bias in which variants are retained (SNP/Singleton ratio) shows consistent patterns across all chromosomes.
• Reference Genome Asymmetry: Specific pentanucleotides show massive strand bias in the reference genome (e.g., TTCGT is 1.45x more frequent on the plus strand). This suggests systematic biases in the assembly or evolutionary history of the reference sequence itself.

5. Significance and Implications

• Mechanistic Insights: The findings suggest that cellular mechanisms for mutation prevention and DNA repair are highly sensitive to local sequence context (up to 5 bases) and operate differently across chromosomes.
• Evolutionary Biology: The decoupling of mutation and selection patterns (e.g., C>T in CG contexts) provides a nuanced view of how specific mutations survive in the population.
• Reference Genome Limitations: The discovery of intrinsic strand asymmetry in the reference genome implies that current reference assemblies may harbor systematic biases that could affect variant calling and interpretation, particularly for strand-specific analyses.
• Future Research: The author posits that these sequence-dependent effects could be linked to susceptibility to neoplastic disease, as the "healthy" mutation patterns observed in singletons might reflect the efficiency of DNA repair mechanisms that, when compromised, lead to cancer.

Conclusion: This study leverages the scale of the UK Biobank to reveal that DNA sequence context (specifically pentanucleotides) is a critical determinant of mutation rates, selection pressures, and strand asymmetries. It highlights complex, chromosome-specific mutational drivers and intrinsic biases in the human reference genome that require further molecular investigation.