DNA replication II: eukaryotes, telomeres and fidelity

Eukaryotic replication follows the same basic logic as bacterial replication: helicase opens the duplex, primers start new strands, polymerases extend them 5′ to 3′, and the lagging strand is made in fragments that are joined. But eukaryotic cells face three extra problems. Their genomes are far larger and are copied far more slowly, so one origin per chromosome would never be enough. Their chromosomes are linear, which creates a problem at the ends. And their DNA is wrapped in nucleosomes that must be taken apart and rebuilt. This lesson examines how cells solve those problems, and then asks how any cell copies billions of base pairs with fewer than one error per billion.

Many origins, once per cycle

A human cell replicates about 6.2 billion base pairs during S phase, which typically lasts around eight hours. Eukaryotic forks are slower than bacterial ones, moving at roughly 1 to 3 kb per minute, about 20 to 50 nucleotides per second, partly because they must deal with chromatin. A single origin per chromosome could not possibly finish in time. Instead, each human cell uses roughly 30,000 to 50,000 origins, spaced every 50 to 300 kb or so, which fire at different times during S phase.

# Replication time for a human genome with different numbers of evenly spaced origins
genome = 6.2e9          # diploid bp
fork_speed = 40         # nt per second (illustrative)
for origins in (100, 1_000, 10_000, 30_000, 50_000):
    segment = genome / origins          # bp served by each origin (two forks)
    t_hours = (segment / 2) / fork_speed / 3600
    print(f"{origins:6,d} origins: segment {segment/1e3:9,.0f} kb -> {t_hours:8.1f} h")

   100 origins: segment    62,000 kb ->    215.3 h
 1,000 origins: segment     6,200 kb ->     21.5 h
10,000 origins: segment       620 kb ->      2.2 h
30,000 origins: segment       207 kb ->      0.7 h
50,000 origins: segment       124 kb ->      0.4 h

Having many origins creates a new danger: if any origin fired twice in one cell cycle, part of the genome would be copied twice, producing extra copies that destabilize chromosomes. Cells prevent this with a two-step licensing system. In G1, when activity of cyclin-dependent kinases (CDKs) is low, the origin recognition complex (ORC), together with the proteins Cdc6 and Cdt1, loads inactive double rings of the MCM helicase onto origins. This is licensing. In S phase, rising CDK and DDK activity activates the loaded helicases, adding partner proteins (Cdc45 and GINS) to form the active CMG helicase, and at the same time prevents any new MCM loading. Because loading and activation can happen only in different phases, each origin can fire at most once per cycle.

Eukaryotic polymerases

Eukaryotes divide the work among several polymerases. DNA polymerase α, bound to primase, makes a short RNA primer followed by a short stretch of DNA. Because polymerase α has low processivity and no proofreading, it hands off quickly. DNA polymerase ε then synthesizes most of the leading strand, and DNA polymerase δ synthesizes most of the lagging strand, both held on DNA by the sliding clamp PCNA (proliferating cell nuclear antigen), the eukaryotic equivalent of the bacterial β clamp. On the lagging strand, primers are removed by the nuclease FEN1 and by RNase H, polymerase δ fills the gaps, and DNA ligase I seals the nicks. Eukaryotes also have many specialized polymerases for repair and for copying past damaged bases.

Chromatin and replication

Nucleosomes ahead of the fork must be disassembled to let the replisome pass, and both daughter duplexes must be repackaged behind it. Histone chaperones, such as FACT ahead of the fork and CAF-1 behind it, carry out this work. Old histones, carrying their chemical modifications, are distributed roughly equally to the two daughters, and newly made histones fill the gaps. Cells make a burst of new histones during S phase for exactly this reason. Because old histones carry marks that indicate whether a region was active or silent, this recycling helps daughter cells inherit patterns of gene expression, part of the epigenetic memory discussed in BIOL-201 lesson 12 and lesson 10 of this course. The size of eukaryotic Okazaki fragments, about the length of DNA in one nucleosome, appears to reflect how lagging-strand synthesis stops when polymerase δ meets a newly deposited nucleosome.

The end-replication problem

Linear chromosomes have a problem at their ends. On the lagging strand, the last Okazaki fragment needs an RNA primer, and when that primer is removed there is no upstream 3′ end from which a polymerase can fill the gap. In addition, the ends are processed by nucleases to leave a single-stranded 3′ overhang. As a result, each round of replication would shorten the chromosome by about 50 to 200 base pairs. Over many divisions, genes near the ends would be lost.

Cells solve this problem in two ways. First, chromosome ends are capped by telomeres, long arrays of a simple repeat that contain no genes, so losing some of the repeat does no immediate harm. In humans and other vertebrates, the repeat is TTAGGG, present in tandem copies stretching 5 to 15 kb at birth. Second, the enzyme telomerase can extend the 3′ end. Telomerase is a reverse transcriptase that carries its own RNA template (in humans, TERC) and a protein catalytic subunit (TERT). It binds the overhang, adds a TTAGGG repeat using its RNA as a template, moves along and adds another. The complementary strand is then filled in by primase and polymerase α. Elizabeth Blackburn, Carol Greider and Jack Szostak shared the 2009 Nobel Prize for discovering telomerase and how telomeres protect chromosomes.

The single-stranded overhang does not dangle free. A protein complex called shelterin binds telomeric DNA and folds the end back into a loop, the t-loop, in which the overhang invades the duplex region. This hides the end from the cell’s DNA damage machinery, which would otherwise treat it as a broken chromosome and try to join it to other ends.

Telomeres, aging and cancer

Most human somatic cells express little or no telomerase. Their telomeres shorten with each division. In the 1960s Leonard Hayflick showed that normal human fibroblasts in culture divide only a limited number of times, roughly 40 to 60, before entering a permanent non-dividing state called replicative senescence. Shortened telomeres are a major trigger of that limit. Germ cells and stem cells express telomerase and maintain their telomeres. Most cancer cells, by contrast, reactivate telomerase, about 85 to 90 percent of human tumors, and the rest use a recombination-based alternative, which is why telomerase is studied as a target for cancer therapy (BIOL-450).

import numpy as np
rng = np.random.default_rng(207)

# 46 chromosomes x 2 ends = 92 telomeres; senescence when the SHORTEST falls below 3 kb.
start = rng.normal(9000, 1500, 92)      # bp, illustrative spread at the start
loss_mean, loss_sd = 100, 30            # bp lost per division per telomere
tel = start.copy()
div = 0
while tel.min() > 3000:
    tel -= np.clip(rng.normal(loss_mean, loss_sd, tel.size), 0, None)
    div += 1
print("starting mean telomere length:", round(start.mean()), "bp")
print("starting shortest telomere:   ", round(start.min()), "bp")
print("divisions until shortest < 3 kb:", div)
print("mean telomere length at senescence:", round(tel.mean()), "bp")

starting mean telomere length: 9164 bp
starting shortest telomere:    5418 bp
divisions until shortest < 3 kb: 25
mean telomere length at senescence: 6656 bp

The simulation makes a subtle point. Senescence comes when the shortest telomere becomes critically short, long before the average reaches that length. This is why measurements of average telomere length in blood cells, sometimes marketed as tests of “biological age”, are only loosely related to an individual’s health. Telomere length varies widely between people of the same age and is influenced by genetics, and claims about lengthening telomeres with supplements are not supported by good evidence.

Fidelity: how accurate is replication?

Replication is astonishingly accurate. In E. coli and in human cells, the final error rate is roughly one mistake per billion to ten billion nucleotides copied (10⁻⁹ to 10⁻¹⁰). This accuracy comes from three layers of quality control, each multiplying the effect of the last.

The first layer is base selection. The polymerase active site fits a correctly paired Watson–Crick base pair snugly, and an incorrect pair has the wrong shape. The polymerase closes around a correct incoming dNTP but closes poorly around a wrong one, so incorporation is much slower. Base selection alone gives an error rate of about 1 in 10⁴ to 10⁵.

The second layer is proofreading. Replicative polymerases (Pol III in E. coli, Pol δ and ε in eukaryotes) have a 3′ to 5′ exonuclease activity. When a wrong nucleotide is added, the mismatched end pairs poorly, slips into the exonuclease site and is removed, and the polymerase tries again. Proofreading improves fidelity about 100-fold.

The third layer acts after the fork has passed. The mismatch repair system scans newly made DNA for mispaired bases, identifies which strand is new, removes a stretch of the new strand containing the error and resynthesizes it. In E. coli, the new strand is recognized because it is not yet methylated on adenines in GATC sequences; in eukaryotes, nicks in the new strand and the orientation of PCNA appear to mark it. Mismatch repair improves fidelity a further 100- to 1,000-fold. BIOL-201 lesson 10 described the mismatch repair proteins and the cancer syndrome (Lynch syndrome) caused by their loss.

genome = 6.2e9
layers = [("base selection only", 1e-5),
          ("+ proofreading (x100)", 1e-7),
          ("+ mismatch repair (x300)", 1e-7 / 300)]
for name, rate in layers:
    print(f"{name:26s} error rate {rate:.1e} -> {genome * rate:12,.1f} errors per division")

base selection only        error rate 1.0e-05 ->     62,000.0 errors per division
+ proofreading (x100)      error rate 1.0e-07 ->        620.0 errors per division
+ mismatch repair (x300)   error rate 3.3e-10 ->          2.1 errors per division

These estimates connect directly to evolution and medicine. Errors that escape all three layers become mutations, the raw material of evolution (BIOL-108). In the human germ line, the per-generation mutation rate is about 1.2 × 10⁻⁸ per base pair, giving each child roughly 60 to 70 new mutations, most of them from the father because sperm precursors divide many more times than egg precursors (BIOL-201 lesson 10).

Replication stress

Forks frequently encounter obstacles: damaged bases, tightly bound proteins, transcribing RNA polymerases, unusual structures such as G-quadruplexes, and regions where both nucleotides of a pair are scarce. When a fork stalls, the cell activates checkpoint kinases (ATR and CHK1) that slow the cell cycle, suppress the firing of distant origins and stabilize the stalled fork until it can restart or be rescued by a fork coming from a adjacent origin. Some regions, called common fragile sites, replicate late and are prone to breakage under stress. Many cancer cells experience chronic replication stress because oncogenes drive them to replicate before they are ready, and drugs that inhibit ATR are being tested in clinical trials as a way to exploit this weakness.