Genome instability and DNA repair

Every human cell suffers tens of thousands of DNA lesions a day: bases lost to spontaneous hydrolysis, bases oxidized by the by-products of metabolism, mismatches left by DNA polymerase, and the occasional double-strand break. A network of repair pathways fixes almost all of them. Cancer cells typically carry thousands to hundreds of thousands of mutations, plus large-scale rearrangements and changes in chromosome number. Lesson 1 called this genome instability an enabling characteristic: it does not make a cell grow, but it speeds the acquisition of changes that do. This lesson looks at the main repair pathways, what happens when they fail, how the scars of past damage can be read as mutational signatures, and how repair defects become therapeutic targets.

Sources of DNA damage

Damage comes from inside and outside the cell. Endogenous sources include depurination, deamination of cytosine to uracil and of 5-methylcytosine to thymine (the most common source of mutations in human tissues, which accumulates steadily with age), reactive oxygen species that produce 8-oxoguanine, and errors and stalled forks during replication. Exogenous sources include ultraviolet light, which links adjacent pyrimidines into cyclobutane dimers; tobacco carcinogens such as benzo[a]pyrene, whose metabolites form bulky adducts on guanine; aflatoxin from molds on stored grain; ionizing radiation, which causes strand breaks; and chemotherapy drugs that are designed to damage DNA. Whether damage becomes a mutation depends on whether it is repaired correctly before the cell replicates.

The main repair pathways

Base excision repair removes small, non-distorting base lesions. A glycosylase cuts out the damaged base, an endonuclease nicks the backbone, and polymerase β and ligase fill the gap. PARP1 detects single-strand breaks and recruits repair proteins, a role central to the story of synthetic lethality below. Nucleotide excision repair removes bulky, helix-distorting lesions such as UV dimers and tobacco adducts, cutting out a stretch of about 30 nucleotides. Inherited defects in it cause xeroderma pigmentosum, with extreme sensitivity to sunlight and skin cancer rates more than a thousand times normal at young ages. Mismatch repair corrects base mismatches and small insertion-deletion loops left by replication, using MSH2–MSH6 and MSH2–MSH3 to recognize errors and MLH1–PMS2 to direct their removal.

Double-strand breaks are the most dangerous lesions, because both strands are cut. Two pathways repair them. Non-homologous end joining ligates the broken ends directly; it works throughout the cell cycle but can delete or add a few bases. Homologous recombination uses the intact sister chromatid as a template and restores the original sequence; it works only in S and G2 phase, after replication. BRCA1 helps commit breaks to homologous recombination by promoting resection of the ends, and BRCA2 loads RAD51 onto the single-stranded DNA so it can invade the sister. A backup, microhomology-mediated end joining, depends on polymerase θ and leaves characteristic deletions flanked by short repeated sequences.

Mismatch repair deficiency and microsatellite instability

Microsatellites are short tandem repeats, such as a run of A bases or a CA dinucleotide repeated 20 times. During replication the polymerase can slip on them, adding or removing a repeat unit. Mismatch repair normally fixes these loops. When it fails, microsatellites change length throughout the genome, a state called microsatellite instability (MSI), and the mutation rate rises 100- to 1,000-fold. Tumors with mismatch repair deficiency (dMMR) or high microsatellite instability (MSI-H) carry very many mutations, especially frameshifts in coding repeats, which inactivate genes such as TGFBR2 and BAX.

About 15 percent of colorectal cancers are MSI-H. A minority result from Lynch syndrome, the inherited form introduced with Daniel's family in lesson 3, caused by germline variants in MLH1, MSH2, MSH6, PMS2 or deletions affecting EPCAM. Most sporadic MSI-H colorectal cancers arise instead from methylation of the MLH1 promoter, often with a BRAF V600E mutation, which helps distinguish sporadic from inherited cases. Testing every colorectal and endometrial cancer for mismatch repair status, by immunohistochemistry for the four proteins or by an MSI test, is now standard in many centers. It identifies families with Lynch syndrome and, as lesson 10 shows, patients likely to respond to immune checkpoint inhibitors, because the many mutations produce abnormal peptides that the immune system can recognize.

Chromosomal instability

Most solid tumors are aneuploid: they have abnormal numbers of chromosomes, often with gains and losses of whole arms and many structural rearrangements. Chromosomal instability (CIN) is the ongoing tendency to gain and lose chromosomes at division, caused by defects in the spindle assembly checkpoint, abnormal centrosome numbers, merotelic kinetochore attachments, or replication stress that leaves under-replicated regions to break in mitosis. CIN generates diversity for selection to act upon, and the loss of chromosome arms is a common route to the second hit of a tumor suppressor. Whole genome doubling, in which a cell duplicates its entire genome without dividing, occurs in roughly a third of advanced tumors and buffers the cell against the harmful effects of later chromosome losses.

Sequencing has revealed even more dramatic events. In chromothripsis, described in 2011, one or a few chromosomes shatter into many pieces in a single catastrophe and are stitched back together in random order, often after being trapped in a micronucleus outside the main nucleus. Extrachromosomal DNA, circular pieces of DNA carrying amplified oncogenes such as MYC or EGFR, is found in a substantial share of tumors and is unequally partitioned at division, allowing copy number to change rapidly from cell to cell.

Telomeres and replicative immortality

Linear chromosomes shorten each time DNA is copied, because the replication machinery cannot complete the lagging strand at the very end. Telomeres, long arrays of the repeat TTAGGG capped by the shelterin protein complex, buffer this loss. In most human somatic cells, telomeres shorten by about 50 to 100 base pairs per division until they become critically short and trigger a DNA damage response, halting division after roughly 40 to 60 doublings in culture, the Hayflick limit. Cells that evade this checkpoint, for instance by losing p53 and RB, keep dividing until telomeres fail completely, chromosomes fuse end to end and break again in mitosis, a crisis that fuels chromosomal instability.

Escaping crisis requires restoring telomeres. About 85 to 90 percent of cancers reactivate telomerase, the reverse transcriptase TERT with its RNA template, often through mutations in the TERT promoter that create new binding sites for transcription factors. TERT promoter mutations are among the most common non-coding mutations in cancer, found in most glioblastomas, many melanomas and bladder cancers. The remaining tumors use alternative lengthening of telomeres (ALT), a recombination-based mechanism frequently associated with loss of ATRX or DAXX. Elizabeth Blackburn, Carol Greider and Jack Szostak received the 2009 Nobel Prize in Physiology or Medicine for discovering how telomeres and telomerase protect chromosomes.

Mutational signatures

Each mutational process leaves a characteristic pattern. Mutations can be classified by the base change (six classes, such as C to T) and the bases immediately before and after, giving 96 trinucleotide contexts. Ludmil Alexandrov, Michael Stratton and colleagues showed in 2013 that the mutation counts of thousands of tumors could be decomposed into a few dozen recurrent signatures, using non-negative matrix factorization. Some signatures have known causes: SBS1, C to T at CpG sites, reflects deamination of methylated cytosine and accumulates with age; SBS4 reflects tobacco smoke; SBS7 reflects ultraviolet light; SBS2 and SBS13 reflect APOBEC cytidine deaminases; SBS3 reflects homologous recombination deficiency; and signatures 6, 15, 26 and 44 reflect mismatch repair deficiency. The COSMIC catalog lists the current set. Signatures turn a tumor's genome into a record of its history and can identify repair defects that matter for treatment.

# Toy mutational signature analysis: factor tumour mutation counts into signatures with NMF.
import numpy as np
rng = np.random.default_rng(4)
contexts = 12                                    # a compressed set of mutation contexts (illustrative)
S = np.array([                                   # three invented signatures, each sums to 1
    [.30, .25, .20, .10, .05, .04, .02, .01, .01, .01, .005, .005],   # 'ageing-like'
    [.01, .02, .03, .05, .30, .30, .15, .08, .03, .01, .01, .01],     # 'tobacco-like'
    [.08, .08, .08, .08, .08, .09, .09, .09, .08, .08, .085, .085]])  # 'flat, HR-deficiency-like'
expo = rng.gamma(2, [[400, 50, 30]] * 20) * (rng.random((20, 3)) < 0.7)  # 20 tumours
counts = rng.poisson(expo @ S)
k = 3
W = rng.random((counts.shape[0], k)) + 0.1; H = rng.random((k, contexts)) + 0.1
for _ in range(3000):                            # Lee-Seung multiplicative updates (KL divergence)
    WH = W @ H + 1e-9
    H *= (W.T @ (counts / WH)) / W.sum(0)[:, None]
    WH = W @ H + 1e-9
    W *= ((counts / WH) @ H.T) / H.sum(1)[None, :]
scale = H.sum(1); H /= scale[:, None]; W *= scale[None, :]
match = [int(np.argmax([np.dot(h, s) / np.linalg.norm(h) / np.linalg.norm(s) for s in S])) for h in H]
for j, m in enumerate(match):
    cos = np.dot(H[j], S[m]) / np.linalg.norm(H[j]) / np.linalg.norm(S[m])
    print(f"recovered signature {j} best matches true signature {m} (cosine similarity {cos:.3f})")

recovered signature 0 best matches true signature 1 (cosine similarity 0.998)
recovered signature 1 best matches true signature 2 (cosine similarity 0.892)
recovered signature 2 best matches true signature 0 (cosine similarity 0.999)

The two peaked signatures are recovered almost perfectly; the flat one, which resembles the featureless signature of homologous recombination deficiency, is recovered less well, because a flat pattern is easy to confuse with mixtures of the others. That is a real problem in practice, and one reason flat signatures are assigned cautiously. Real analyses use the full 96 contexts, hundreds to thousands of genomes and careful statistics to decide how many signatures are present, but the principle is the same: mutation counts are modeled as non-negative mixtures of non-negative signatures, and each tumor's exposure to each process is estimated.

Synthetic lethality and PARP inhibitors

Two genes are synthetically lethal when losing either alone is tolerated but losing both kills the cell. The idea, borrowed from yeast genetics, offers a way to kill cancer cells that have lost a tumor suppressor, something that cannot be targeted directly because the protein is gone. BRCA-deficient cells cannot repair double-strand breaks by homologous recombination. Inhibiting PARP traps PARP1 on DNA at single-strand breaks; when replication forks hit these trapped complexes, they collapse into double-strand breaks. Normal cells repair them by homologous recombination; BRCA-deficient cells cannot and die. Two groups, led by Alan Ashworth and Thomas Helleday, reported this in 2005.

PARP inhibitors such as olaparib, niraparib, rucaparib and talazoparib are now used for BRCA-mutant ovarian, breast, prostate and pancreatic cancers, and for tumors with other homologous recombination defects. Their effectiveness confirms the principle, and their limits illustrate resistance: tumors can restore homologous recombination by secondary mutations that put BRCA2 back in frame, or by losing other proteins such as 53BP1 that block resection. Polymerase θ inhibitors, which target the backup pathway that HR-deficient cells lean on, are in clinical trials.

# Synthetic lethality: viability of cells by genotype and PARP inhibitor (illustrative values).
viability = {   # fraction of colony-forming cells surviving
    ('BRCA2 normal', 'no drug'): 1.00, ('BRCA2 normal', 'PARP inhibitor'): 0.85,
    ('BRCA2 lost', 'no drug'): 0.90,   ('BRCA2 lost', 'PARP inhibitor'): 0.02,
    ('BRCA2 lost + reversion', 'no drug'): 0.92, ('BRCA2 lost + reversion', 'PARP inhibitor'): 0.80}
for (geno, drug), v in viability.items():
    bar = '#' * round(v * 40)
    print(f"{geno:24s} {drug:15s} {v:4.2f} {bar}")
sel = viability[('BRCA2 normal', 'PARP inhibitor')] / viability[('BRCA2 lost', 'PARP inhibitor')]
print(f"selective killing of BRCA2-lost cells: about {sel:.0f}-fold; a reversion mutation removes it")

BRCA2 normal             no drug         1.00 ########################################
BRCA2 normal             PARP inhibitor  0.85 ##################################
BRCA2 lost               no drug         0.90 ####################################
BRCA2 lost               PARP inhibitor  0.02 #
BRCA2 lost + reversion   no drug         0.92 #####################################
BRCA2 lost + reversion   PARP inhibitor  0.80 ################################
selective killing of BRCA2-lost cells: about 42-fold; a reversion mutation removes it