DNA Repair as a Guardian of the Genome

DNA repair biochemistry underlies major swaths of modern clinical oncology and molecular diagnostics. Microsatellite instability (MSI) testing and immunohistochemistry for MMR proteins (MLH1, MSH2, MSH6, PMS2) are now routine in colorectal and endometrial cancer workups, both to screen for Lynch syndrome and to select patients for immune-checkpoint-inhibitor therapy (pembrolizumab was the first drug approved by tissue-agnostic biomarker — MSI-high/dMMR status — rather than tumor site, directly built on the mechanistic understanding that MMR-deficient tumors accumulate frameshift neoantigens). PARP inhibitors (olaparib, niraparib, rucaparib) exploit BER biology directly: PARP1 senses single-strand breaks generated during BER, and its inhibition causes unrepaired BER intermediates to persist into S phase, collapsing replication forks into double-strand breaks that require homologous recombination (BRCA1/2-dependent) to fix — in BRCA-mutant tumors, this creates synthetic lethality, the chemical basis for one of the most successful rationally designed cancer drug classes of the past two decades. Temozolomide and other alkylating chemotherapeutics deliberately generate BER substrates (O6-methylguanine, N7-methylguanine) to kill rapidly dividing glioma cells, and tumor resistance frequently arises via upregulation of MGMT (a direct-reversal repair enzyme) or altered BER flux. Cisplatin and related platinum chemotherapeutics form bulky intrastrand DNA crosslinks that are substrates for NER; NER capacity (assessed via ERCC1 expression) is an actively studied predictive biomarker for platinum chemotherapy response in lung and ovarian cancers. On the instrumentation side, repair-pathway biochemistry underlies CRISPR-based genome editing outcomes: after Cas9 creates a double-strand break, the balance between non-homologous end joining and homology-directed repair (a cousin pathway to the excision-repair logic described here) determines whether an edit produces a small indel or a precise sequence replacement, and drugs/small molecules that bias this balance (NHEJ inhibitors like SCR7) are actively used to improve editing precision in research and therapeutic contexts.

Ask the tutor: How does the choice between homologous recombination and non-homologous end joining for double-strand break repair compare mechanistically to the excision-and-resynthesis logic of MMR, BER, and NER?

DNA damage recognition is fundamentally a problem of molecular contrast. A repair enzyme patrolling billions of base pairs of essentially uniform double helix must find a single altered nucleotide — a task analogous to finding one mistyped letter in a library's worth of text without knowing where to look. Each repair pathway solves this with a different physical strategy exploiting a different chemical signature of the damage. Mismatch repair confronts the hardest version of the problem: two Watson-Crick-incorrect but still base-paired nucleotides (say, a G·T wobble instead of G·C) barely perturb the helix — the distortion is on the order of a slight bend and altered hydrogen-bonding geometry, not a bulky adduct. MutS-family proteins detect this via a sensing mechanism in which a conserved Phe-X-Glu motif inserts into the minor groove and probes for the reduced stability and altered stacking of the mismatch, kinking the DNA by roughly 45–60°. Critically, MMR must also solve an information problem: given a mismatch, which base is the "wrong" one — the template or the newly made strand? In bacteria this is solved by transient hemimethylation (the parental strand is methylated at GATC sequences by Dam methylase, the new strand lags behind), while in eukaryotes strand discrimination instead exploits pre-existing nicks in the discontinuous lagging strand and the 3′ or 5′ ends generated during replication. Base-excision repair confronts lesions that are chemically wrong but geometrically almost invisible — a deaminated cytosine (uracil), an oxidized guanine (8-oxoG), or an alkylated base still fits inside the double helix, roughly Watson-Crick paired. The BER solution is base flipping: a DNA glycosylase rotates the target nucleotide by ~180° out of the duplex into an extrahelical pocket where it is chemically read out and, if damaged, its N-glycosidic bond is hydrolyzed. This creates an abasic (apurinic/apyrimidinic, AP) site, itself a lesion that is then handled by a dedicated sub-pathway. Nucleotide-excision repair confronts lesions that do distort the helix substantially — cyclobutane pyrimidine dimers and 6-4 photoproducts from UV light, and bulky chemical adducts (e.g., benzo[a]pyrene-guanine adducts from smoke). Here the strategy is not to identify the specific chemical lesion but to sense the resulting helical distortion and thermodynamic destabilization, then excise a whole single-stranded oligonucleotide patch around the lesion, regardless of its exact chemical identity — a strategy sometimes called "damage-nonspecific" excision, in contrast to the enzyme-specific chemical readout of BER.

The chemistry underlying all three repair pathways reduces to a small set of governing relationships: how repair efficiency depends on lesion recognition rates and competing replication, how the free energy of mismatch destabilization drives detection, and how nucleotide-level accounting explains excision-patch sizes and net synthesis costs.

where: $\Delta G_{WC}$ is the stabilization free energy of a correct Watson-Crick pair (~ -1 to -2 kcal/mol per stack, sequence-dependent), and $\Delta G_{\text{distortion}}$ is the destabilization introduced by a mismatch or lesion; a larger destabilization corresponds to a more easily "sensed" defect and correlates with higher repair efficiency for a given surveillance protein.

Reading: MMR proteins detect small $\Delta G_{\text{distortion}}$ (G·T mismatches, ~1–3 kcal/mol destabilization) with high specificity binding, whereas NER factors detect large, cooperative destabilizations from helix-distorting bulky lesions; this explains why NER's sensor (XPC) binds unpaired/flipped-out bases on the undamaged strand opposite a bulky lesion rather than the lesion itself.

where: $N_{\text{excised}}$ is total nucleotides removed in NER, $N_{3'}$ and $N_{5'}$ are nucleotides removed on each side of the lesion by the two structure-specific nucleases (XPG cuts 3′, XPF–ERCC1 cuts 5′ in humans), and $N_{\text{lesion}}$ is the lesion-containing nucleotide(s) themselves.

Reading: in human NER, incisions occur at fixed distances from the lesion (about 20 nt 5′ and about 5 nt 3′), giving a conserved patch size of roughly 24–32 nt regardless of the specific chemical adduct — this is the operational definition of "damage-nonspecific" excision.

where: $k_{\text{repair}}^{\text{net}}$ is the observed rate of correct repair, $k_{\text{recognition}}$ is the rate at which a mismatch or lesion is physically found by a sensor protein, and $f_{\text{strand-correct}}$ is the probability that the pathway removes the erroneous (not the template) strand.

Reading: MMR fidelity is limited less by finding the mismatch (fast, facilitated by processive sliding clamps) than by getting $f_{\text{strand-correct}} \to 1$; this is why strand-discrimination signals (methylation state in bacteria, nicks/PCNA orientation in eukaryotes) are the rate- and fidelity-determining feature of the whole pathway.

Example 1 — Foundational: how many uracils does base-excision repair remove from a human genome per day, and what would happen without UNG? Cytosine deaminates spontaneously to uracil at a rate of roughly $1 \times 10^{-10}$ per cytosine per second at physiological temperature and pH (a well-established estimate for hydrolytic deamination). The human genome contains about $3.2\times10^9$ bp, with cytosine content near 20% (GC content ≈ 40%, so C alone ≈ 20%), giving:

Using one strand's cytosine count per genome copy for a simple estimate, $N_C \approx 6.4\times10^8$. Deamination events per day:

Each of these, if unrepaired, would base-pair with adenine during replication (uracil reads like thymine), converting an original G·C pair into an A·T pair — a transition mutation — with essentially 100% efficiency if it survives to the next round of replication. UNG finds and excises the vast majority of these uracils within minutes (in vitro turnover numbers for UNG exceed $10^3$ min⁻¹ per active site, and cellular UNG copy number is more than sufficient to process ~5500 events/day, i.e., roughly 4 events/minute on average). Without UNG (as in ung⁻ bacterial mutants or UNG-deficient human cell lines), spontaneous C→T transition mutation rates increase by roughly one to two orders of magnitude, directly demonstrating that BER — not MMR or NER — is the dedicated pathway guarding against this specific, extremely common hydrolytic lesion.

Example 2 — Applied: Lynch syndrome and microsatellite instability. Lynch syndrome patients carry one inactivated allele of an MMR gene (commonly MLH1 or MSH2). Because MMR is haplosufficient at the biochemical level (one functional allele produces enough protein for essentially wild-type repair), carriers are phenotypically normal until a somatic "second hit" inactivates the remaining allele in a particular cell lineage (Knudson two-hit logic, echoing tumor-suppressor genetics). Once both alleles are lost, that cell line loses MMR function entirely, and DNA polymerase slippage errors at short tandem repeats (microsatellites) — which normally occur at rates of roughly $10^{-4}$ to $10^{-3}$ per locus per generation before MMR correction, but are almost entirely (>99%) repaired by functional MMR — go uncorrected. Consider a mononucleotide repeat tract of $n=10$ adenines. DNA polymerase slippage (strand misalignment during replication of repetitive sequence) generates insertion/deletion loops at a pre-MMR rate of about $\mu_0 = 5\times10^{-4}$ per replication. In an MMR-proficient cell, post-repair mutation rate:

In an MMR-deficient cell (Lynch tumor), $f_{\text{repair}} \to 0$:

giving a $100$-fold increase in mutation rate at this locus, consistent with the massive microsatellite instability (MSI) observed clinically and used as a diagnostic biomarker (MSI-high tumors) for Lynch syndrome and as a predictive biomarker for immune-checkpoint-inhibitor therapy (MSI-high tumors present a high neoantigen burden due to frameshift mutations accumulating in coding microsatellites, making them unusually immunogenic).

Example 3 — Exam/competition level: quantitative comparison of NER capacity against UV dose. A human keratinocyte is exposed to a UVB dose that generates $N_{\text{CPD}} = 5\times10^4$ cyclobutane pyrimidine dimers (CPDs) per genome (a physiologically realistic dose, roughly equivalent to several minutes of unprotected midday sun). Global-genome NER (GG-NER) removes CPDs relatively slowly, with observed repair half-life $t_{1/2} \approx 24$ hours in normal fibroblasts (CPDs are repaired much more slowly than 6-4 photoproducts, which have $t_{1/2}\approx 2$–4 h, because CPDs cause a smaller helical distortion and are detected less efficiently by XPC). Assume first-order repair kinetics:

With $t_{1/2}=24$ h, $\lambda = 0.693/24 = 0.0289\ \text{h}^{-1}$. After $t=48$ h (two half-lives):

confirming the expected quarter-remaining after two half-lives. Now suppose this is a xeroderma pigmentosum group A (XP-A) cell, completely NER-deficient ($\lambda \to 0$): $N(48\text{h}) = N_0 = 5\times10^4$, unchanged — every CPD persists until either replicated over (by translesion polymerases, itself mutagenic) or the cell dies. Given that unrepaired CPDs block replicative polymerases and are bypassed by error-prone translesion synthesis polymerase η (which, notably, correctly inserts two adenines opposite a T-T CPD in most cases, making it relatively safe for thymine dimers specifically — this is why XP variant, an XPV/Pol η defect rather than a GG-NER defect, still causes dramatically elevated skin cancer, since backup translesion polymerases are far more error-prone), the persistence of $5\times10^4$ unrepaired CPDs per genome per UV exposure event, repeated over years of sun exposure in an NER-null XP patient, compounds into the observed >1000-fold elevated skin cancer incidence — an empirically documented, mechanistically fully explicable clinical outcome.

Paul Modrich's four-decade dissection of Escherichia coli mismatch repair, culminating in the full in vitro reconstitution of the MutHLS pathway from purified components in 1989, established the biochemical template against which all subsequent DNA repair mechanisms — bacterial or human — have been understood. Modrich's laboratory demonstrated, using purified MutS, MutL, MutH, DNA helicase II, single-strand binding protein, exonucleases, DNA polymerase III holoenzyme, and DNA ligase, that a hemimethylated heteroduplex substrate could be repaired with absolute strand specificity: the unmethylated (newly synthesized, mismatch-containing) strand was always the one degraded and resynthesized, never the methylated parental strand. This was a landmark demonstration that a complex, multi-enzyme genetic fidelity system could be entirely explained — no vitalistic "cellular memory," no unexplained selectivity — by ordinary enzyme kinetics and a single physical strand-marking chemical modification (Dam methylation). Aziz Sancar's parallel reconstitution of the E. coli UvrABC nucleotide-excision-repair system, and later the human NER system, established the analogous "cut-and-patch" logic for bulky lesions and revealed the precise, conserved dual-incision geometry (asymmetric distances 5′ and 3′ of the lesion) that generalizes across bacteria and humans despite very different protein machinery. Both Modrich and Sancar, together with Tomas Lindahl (who established base-excision repair and the chemical instability of DNA as a foundational concept — that DNA is not the inert, permanently stable molecule it was long assumed to be), shared the 2015 Nobel Prize in Chemistry for mechanistic studies of DNA repair.