DNA Sequencing: Sanger and Next-Generation Methods

A single missing 3′-hydroxyl group is enough to stop a polymerase cold. That one fact built the entire genomic era. In 1977, Frederick Sanger added 2′,3′-dideoxynucleotides to a DNA polymerization reaction. He was exploiting a subtle piece of nucleotide chemistry. Without a 3′-OH, the growing chain has no nucleophile left. It cannot attack the next incoming triphosphate. Synthesis terminates irreversibly at that exact position. Read out over four lanes or four colors, the terminated fragments spell out the template sequence. This happens one base at a time. This matters far beyond a historical curiosity. Sanger sequencing decoded the first genomes. It underwrote the Human Genome Project's finishing phase. It remains the gold standard for clinical variant confirmation. This is because of its long, unambiguous reads. Next-generation sequencing (NGS) then inverted the economics. Instead of one long reaction read serially, millions to billions of short fragments are sequenced in parallel. This happens on a solid surface. It uses cycles of reversible termination, fluorescence imaging, and computational reassembly. The shift from "one reaction, one read" to "billions of reactions, one image per cycle" is the central engineering idea of modern genomics. It rests on exactly the same polymerase chemistry that Sanger used. That chemistry is enzymatic template-directed synthesis. It is governed by Watson-Crick pairing and the geometry of the deoxyribose sugar. The big picture is that sequencing is fundamentally a controlled-termination experiment. It is layered on top of The Polymerase Chain Reaction (PCR)-style enzymatic amplification and priming. Every sequencing chemistry must solve the same three sub-problems. This includes Sanger dideoxy termination, Illumina sequencing-by-synthesis, Ion Torrent proton detection, or nanopore strand-threading. First, how to prime synthesis at a defined start site. Second, how to make each incorporation event distinguishable and stoppable. Alternatively, how to make it otherwise individually detectable. Third, how to convert a physical/chemical signal into a base call. Such signals include radioactivity, fluorescence, pH change, or ionic current. The base call must carry a quantifiable error rate. This article works through the chain-termination mechanism in full atomic detail. It derives why dideoxynucleotides truncate synthesis. It also explains why their molar ratio to normal dNTPs controls fragment-length distribution. The same logic is then extended to the sequencing-by-synthesis cycle used in Illumina NGS. This includes how phasing and dephasing set a hard limit on read length. Quantitative treatments include the Sakharkar-style termination probability derivation. They also include per-cycle signal decay in NGS. Worked examples span a textbook question, a clinical Sanger trace interpretation, and a competition-level phasing-noise calculation.

Key takeaways

Dideoxynucleotides (ddNTPs) lack a 3′-OH, so DNA polymerase cannot form the next phosphodiester bond after incorporation — chain termination is irreversible and sequence-specific.
Sanger sequencing uses four ddNTP/dNTP mixes (or one pooled, dye-labeled mix) to generate a ladder of fragments, each ending at a specific base, resolved by capillary electrophoresis.
The ddNTP:dNTP molar ratio sets the mean fragment length via a geometric-distribution argument: low ratios give long reads, high ratios give short reads.
Illumina NGS replaces one long reaction with millions of clonal clusters on a flow cell, using reversible terminators cleaved after each fluorescence read to sequence one base per cycle.
Read length in NGS is limited by phasing/pre-phasing (asynchronous clusters) and signal decay, which accumulate roughly linearly with cycle number and cap useful reads around 150–300 nt.
Long-read platforms (PacBio, Oxford Nanopore) trade per-base accuracy for read length by monitoring real-time polymerase kinetics or ionic current through a pore rather than terminating synthesis at all.
Key idea: every sequencing chemistry converts an enzymatic, base-specific event into a measurable, discrete signal — the whole engineering problem is maximizing the ratio of true signal to positional/chemical noise per read.

Where this lives today

Clinical diagnostic laboratories still rely on Sanger sequencing for definitive confirmation of specific variants flagged by NGS panels — its long, per-molecule-averaged reads and extremely low false-positive rate make it the accepted gold standard for regulatory-grade variant confirmation in disorders like cystic fibrosis (CFTR sequencing), hereditary cancer syndromes (BRCA1/2), and pharmacogenomic testing. Illumina's sequencing-by-synthesis chemistry, built directly on the reversible-terminator phasing model derived above, now underlies the overwhelming majority of whole-genome and whole-exome sequencing performed worldwide, including population biobank projects (UK Biobank, All of Us) and virtually all clinical oncology tumor-profiling panels that scan hundreds of cancer-associated genes simultaneously. During the COVID-19 pandemic, both Sanger confirmation and Illumina/Oxford Nanopore high-throughput sequencing were deployed together for SARS-CoV-2 variant surveillance — nanopore's real-time, portable long-read chemistry allowed genomic surveillance directly in the field without shipping samples to a centralized facility. Long-read single-molecule platforms (PacBio HiFi, Oxford Nanopore) now resolve structural variants, repetitive regions, and full haplotypes that short-read Illumina data cannot span, and were essential to completing the first truly gapless human genome assembly (the Telomere-to-Telomere consortium's T2T-CHM13 assembly, 2022). The chain-termination and phasing principles derived here also directly inform CRISPR-based diagnostic assay design, single-cell RNA sequencing library preparation, and ancient-DNA sequencing protocols, where extremely degraded, short template fragments demand chemistries tuned for very short expected read lengths.

Ask the tutor: How would you redesign the ddNTP:dNTP ratio if you wanted even band intensity across a 1,200-nucleotide Sanger read instead of the usual ~800?

The physical picture — sugar geometry, primed synthesis, and the read-out chain

DNA polymerases synthesize in the 5′→3′ direction by nucleophilic attack of the primer's 3′-hydroxyl oxygen on the α-phosphate of the incoming deoxynucleoside triphosphate (dNTP), displacing pyrophosphate. This is the same two-metal-ion mechanism used throughout The Polymerase Chain Reaction (PCR): two Mg²⁺ ions in the polymerase active site position the 3′-O⁻ nucleophile and stabilize the pentacovalent transition state at the α-phosphate, then stabilize the pyrophosphate leaving group. The entire chain-termination strategy hinges on one structural detail of the incoming nucleotide: whether carbon 3′ of the deoxyribose ring bears a hydroxyl group or not. A normal dNTP has an OH at C3′. A dideoxynucleotide triphosphate (ddNTP) has H at both C2′ and C3′ — it is "dideoxy" because both ring oxygens that would normally carry an -OH (as in ribose) are missing; in dNTPs only the 2′-OH is already absent relative to ribonucleotides, and ddNTPs remove the 3′-OH as well. When a ddNTP is incorporated at the growing 3′ end of a DNA strand, that strand terminus becomes a dead end: there is no 3′-oxygen nucleophile to attack the next triphosphate's α-phosphorus. The polymerase releases the template, and that particular strand can never elongate again. Because incorporation is stochastic and competes with normal dNTP incorporation at every position, a population of millions of primer molecules produces a full ladder of fragments truncated at every possible base position, each fragment's length exactly reporting where a ddNTP happened to be incorporated. Sanger's original protocol ran four separate reactions, each spiked with a small amount of one ddNTP (ddATP, ddCTP, ddGTP, or ddTTP) alongside all four normal dNTPs and a radiolabeled primer. Products were resolved on four adjacent lanes of a denaturing polyacrylamide gel, and the sequence was read directly from the pattern of bands moving up the gel (shortest fragments migrate fastest). Modern automated Sanger sequencing uses four differently colored fluorescent ddNTPs in a single tube, resolved by capillary electrophoresis, with a laser detector reading colors as bands pass a fixed point — producing the now-iconic four-color chromatogram trace. Next-generation sequencing keeps the same underlying chemistry (base-specific enzymatic events) but changes the geometry of the experiment entirely: instead of one reaction generating a length ladder, millions of spatially separated clonal DNA clusters are synthesized one base at a time in lockstep, using reversible terminators. After each base is added and imaged, a chemical cleavage step removes the terminating and fluorescent groups, regenerating a free 3′-OH so synthesis can continue for one more cycle. The read length ceiling in NGS comes not from a single termination event but from the gradual loss of synchrony among millions of copies within a cluster — a purely statistical/kinetic limitation rather than a chemical dead-end.

Figure

Dideoxynucleotide vs deoxynucleotide: the missing 3′-OH

Figure

Sanger ladder read-out: four ddNTP reactions resolved by size

The core ideas — termination probability and read-length statistics

The central quantitative relationship in Sanger sequencing is the competition between ddNTP and dNTP incorporation at every templated position, treated as a per-step Bernoulli trial repeated along the template.

p = \frac{k_{cat}^{dd}[ddNTP]}{k_{cat}^{dd}[ddNTP] + k_{cat}^{d}[dNTP]}

where:

$p$ = probability that termination (ddNTP incorporation) occurs at any given templated position,
$k_{cat}^{dd}, k_{cat}^{d}$ = catalytic efficiency ( $k_{cat}/K_M$ ) of the polymerase for ddNTP versus dNTP incorporation,
$[ddNTP], [dNTP]$ = molar concentrations of the dideoxy and normal triphosphate for that base in the reaction mix.

Because $p$ is the same at every position (concentrations are constant throughout the reaction and the polymerase has no memory), the length of a terminated fragment is a geometrically distributed random variable:

P(L = n) = (1-p)^{n-1} p

where:

$P(L=n)$ = probability that the strand terminates exactly at the $n$ th nucleotide after the primer,
$(1-p)^{n-1}$ = probability of $n-1$ successive normal (non-terminating) incorporations,
$p$ = probability the $n$ th incorporation is the terminating event.

The mean fragment length follows directly from the geometric distribution's expectation value:

\langle L \rangle = \frac{1}{p}

Reading: if $p$ is small (low ddNTP:dNTP ratio), the average fragment is long — few terminations happen per template, so you build a ladder covering large fragment sizes, ideal for reading distant bases. If $p$ is large, fragments are short on average, giving dense, well-resolved reads near the primer. Real protocols use different ddNTP:dNTP ratios or read strategies (primer walking) to cover a full 700–900 nucleotide read with even band spacing throughout. For next-generation sequencing-by-synthesis, the governing relationship is not termination probability but phasing noise accumulation across synchronized cycles. In each cycle, a small fraction of strands within a cluster either fail to incorporate a nucleotide (lagging, "phasing") or incorporate two bases in one cycle due to incomplete terminator/dye removal (leading, "pre-phasing"). If $\phi$ is the per-cycle phasing rate and $\rho$ the per-cycle pre-phasing rate, the fraction of a cluster's strands still perfectly in-register after $c$ cycles is:

f(c) = (1 - \phi - \rho)^{c}

where:

$f(c)$ = fraction of DNA copies within a cluster synchronized with the intended cycle after $c$ cycles,
$\phi$ = probability per cycle that a given strand fails to add a nucleotide (falls behind),
$\rho$ = probability per cycle that a strand adds an extra, uncontrolled nucleotide (gets ahead).

Reading: because $f(c)$ decays geometrically, signal purity (and thus base-call confidence) degrades roughly exponentially with cycle number, which is why Illumina reads historically plateaued near 100–300 bases even though the chemistry could in principle run indefinitely — the noise floor, not the chemistry, sets the practical read length.

Derivation AChain-termination fragment-length distribution and the ddNTP:dNTP design rule

Goal: derive the geometric distribution of fragment lengths from single-step incorporation kinetics, and show how the ddNTP:dNTP ratio must be chosen to give useful sequencing coverage across a target read window.

Step 1 — Model each templated position as an independent Bernoulli trial. At every position along the template, the polymerase's active site is presented with a mixture of the correct dNTP and the corresponding ddNTP (all four ddNTPs and all four dNTPs are present in modern one-tube reactions, distinguished later by fluorescence). Because $K_M$ and $k_{cat}$ for a given polymerase (e.g., a modified T7 DNA polymerase, "Sequenase," engineered to have reduced 3′→5′ exonuclease activity and near-equal affinity for dd- and d-NTPs) are fixed molecular constants, and reaction volumes are well-mixed, the probability that the next incorporation event uses a ddNTP rather than a dNTP is the same constant $p$ at every position — this is the key independence assumption that makes the process memoryless.

Step 2 — Write the survival function. Let $S(n)$ be the probability that the strand has not yet terminated after $n$ successive incorporation events (i.e., all $n$ were normal dNTPs):

S(n) = (1-p)^n

Step 3 — Differentiate the survival function to get the termination-at-exactly- $n$ probability. The event "terminates at position $n{+}1$ " is "survived through $n$ " AND "terminated on trial $n{+}1$ ":

P(L = n+1) = S(n)\cdot p = (1-p)^n p

Relabeling $n+1 \to n$ recovers the geometric PMF quoted above:

P(L=n) = (1-p)^{n-1}p, \qquad n = 1,2,3,\dots

Step 4 — Compute the mean via the standard geometric-series identity.

\langle L\rangle = \sum_{n=1}^{\infty} n(1-p)^{n-1}p = p\sum_{n=1}^\infty n(1-p)^{n-1}

Using $\sum_{n=1}^\infty n x^{n-1} = 1/(1-x)^2$ with $x = 1-p$ :

\langle L\rangle = p \cdot \frac{1}{p^2} = \frac{1}{p}

Step 5 — Translate into a design rule. If a reaction is to give useful, well-spaced bands out to a target read length $L_{max}$ (say 800 nt), we want $\langle L\rangle$ comparable to or somewhat beyond $L_{max}$ so that a substantial fraction of molecules terminate near the far end of the read window while still populating shorter fragments densely enough to resolve every intervening base. Since $\langle L \rangle = 1/p$ and

p \approx \frac{[ddNTP]}{[dNTP]} \quad \text{(when } k_{cat}^{dd}\approx k_{cat}^{d}\text{, a design goal of engineered polymerases)}

the practical ratio is:

\frac{[ddNTP]}{[dNTP]} \approx \frac{1}{L_{max}}

For $L_{max} = 800$ , this gives a target molar ratio near 1:800 — consistent with real Sanger sequencing protocols, which use ddNTP:dNTP ratios in the range of roughly 1:50 to 1:1000 depending on the desired read window and specific dye chemistry (dye-labeled ddNTPs also have altered incorporation kinetics relative to unlabeled ones, requiring empirical tuning around this theoretical estimate).

Go deeper · Derivation B · 2 minNGS phasing/pre-phasing decay and the effective read-length ceiling

Derivation BNGS phasing/pre-phasing decay and the effective read-length ceiling

Goal: derive how synchronization loss across a clonal cluster limits usable read length in cyclic reversible-termination sequencing, and obtain a quantitative cycle number at which signal purity drops below a usable threshold.

Step 1 — Define the three possible per-cycle states for a single strand in a cluster. In an ideal cycle, every one of the $\sim$ 1000 identical DNA copies within a cluster incorporates exactly one labeled reversible-terminator nucleotide, is imaged, then is chemically deprotected to expose a fresh 3′-OH for the next cycle. In practice, at each cycle a given strand either: (i) incorporates correctly and stays in phase (probability $1-\phi-\rho$ ), (ii) fails to incorporate (probability $\phi$ , "phasing," strand falls one cycle behind), or (iii) incorporates an extra, uncontrolled base due to incomplete terminator removal (probability $\rho$ , "pre-phasing," strand jumps one cycle ahead).

Step 2 — Track the fraction remaining perfectly in-register after $c$ cycles. Because these are independent per-cycle Bernoulli-type events (occurring with the same small probabilities at every cycle, to good approximation, since terminator/dye chemistry and enzyme conditions are held constant), the probability a given strand has stayed exactly in phase through all $c$ cycles is the product of $c$ independent "stay in phase" events:

f(c) = (1-\phi-\rho)^c

Step 3 — Approximate for small per-cycle error rates. Typical Illumina chemistries achieve $\phi + \rho \sim 0.001$ – $0.005$ per cycle (0.1–0.5%). For $x = \phi+\rho \ll 1$ , use $\ln(1-x)\approx -x$ :

\ln f(c) = c\ln(1-\phi-\rho) \approx -c(\phi+\rho)

f(c) \approx e^{-c(\phi+\rho)}

This shows the in-phase fraction decays exponentially with cycle number, with decay constant $\phi+\rho$ — a compounding-error problem structurally identical to any repeated multiplicative-loss process.

Step 4 — Define a usability threshold and solve for the critical cycle number. Define $c^*$ as the cycle at which the in-phase fraction falls to some threshold $f_{thresh}$ below which base calls become unreliable (mixed signal from out-of-phase strands swamps the true signal). Solving $f(c^*) = f_{thresh}$ :

e^{-c^*(\phi+\rho)} = f_{thresh}

c^* = \frac{-\ln f_{thresh}}{\phi+\rho} = \frac{\ln(1/f_{thresh})}{\phi+\rho}

Step 5 — Plug in representative numbers. Take $\phi+\rho = 0.002$ per cycle (0.2% combined error) and require the in-phase fraction to stay above $f_{thresh}=0.7$ (70% of cluster signal still correctly phased — a rough but standard rule of thumb before quality scores degrade too far for confident calling):

c^* = \frac{\ln(1/0.7)}{0.002} = \frac{0.357}{0.002} \approx 178 \text{ cycles}

This lands squarely in the observed range of standard Illumina read lengths (100–300 bp), confirming that phasing/pre-phasing error accumulation, not the underlying polymerase or optics, is the dominant hard limit on short-read length. Reducing $\phi+\rho$ by improving terminator cleavage chemistry or enzyme fidelity is precisely the lever that has pushed practical read lengths from ~35 bp (early Illumina/Solexa) to ~300 bp (current paired-end chemistries).

Worked examples

Example 1 (foundational): computing termination probability and expected fragment length. A Sanger reaction is prepared with $[dATP] = 200\ \mu M$ and $[ddATP] = 2\ \mu M$ , and the modified polymerase has essentially equal catalytic efficiency for both ( $k_{cat}^{dd}/K_M^{dd} \approx k_{cat}^{d}/K_M^{d}$ ). Find the probability of termination at any A position and the expected fragment length. Using $p \approx [ddATP]/([ddATP]+[dATP])$ :

p = \frac{2}{2+200} = \frac{2}{202} = 0.0099

Mean fragment length:

\langle L \rangle = \frac{1}{p} = \frac{1}{0.0099} \approx 101\ \text{nucleotides}

So on average, a strand runs about 101 nucleotides before hitting a terminating ddATP — comfortably inside a typical 700–900 nt capillary read window, giving good coverage of A-terminated fragments throughout the readable range. As a limiting-case check: if $[ddATP]\to 0$ , then $p\to 0$ and $\langle L\rangle \to \infty$ — no termination ever occurs and the reaction just produces full-length, non-terminated product, exactly as expected.

Example 2 (applied): reading a real four-color Sanger chromatogram fragment ambiguity. A clinical laboratory sequences a PCR-amplified exon to confirm a suspected heterozygous point mutation. At position 245 in the forward-strand trace, the base-caller reports overlapping green (A) and black (G) peaks of roughly equal height, while all flanking positions show clean single-color peaks. Interpret this signal chemically and decide what follow-up is needed. Because the four-color Sanger reaction pools all four fluorescent ddNTPs in a single tube and sequences a clonal PCR product (in this case from a heterozygous individual, so the template itself is a 50:50 mixture of two alleles), a true heterozygous SNP produces two independent, overlapping termination ladders — one built on each allele's template. At a variant position, roughly half the terminated fragments in the pooled reaction end in the reference base and half in the variant base, so the capillary electrophoresis detector reads two co-migrating, similarly sized fragments carrying different dyes, appearing as a double peak (here, A and G superimposed, consistent with a heterozygous A/G — i.e., a purine transition, commonly annotated R in IUPAC ambiguity code). This is mechanistically distinct from a sequencing artifact such as a polymerase stutter through a homopolymer run, which would show a shifted, not co-located, double peak downstream. Because the flanking bases are clean, the most likely explanation is a genuine heterozygous variant rather than a technical artifact; standard practice is to confirm with an independent reverse-strand read (variant should reappear at the complementary position, e.g., T/C on the reverse strand) before reporting.

Example 3 (competition-level): required cluster synchrony for confident SNP calling at cycle 150 given a tightened chemistry. An engineering team improves an Illumina-style reversible-terminator chemistry so that the pre-phasing rate is reduced to $\rho = 0.0005$ per cycle but the phasing rate remains $\phi = 0.0015$ per cycle. A downstream variant caller requires at least 60% of a cluster's fluorescence signal to originate from correctly phased strands at cycle 150 for a base call to be flagged "high confidence." Determine whether this chemistry meets the requirement, and find the maximum $\phi+\rho$ that would exactly meet a 60% threshold at cycle 150. First compute the actual in-phase fraction at $c=150$ with the improved chemistry, using the exact geometric expression (not just the exponential approximation, to be careful near larger $c$ ):

\phi + \rho = 0.0015 + 0.0005 = 0.0020

f(150) = (1-0.0020)^{150} = (0.998)^{150}

Take logarithms:

\ln f(150) = 150\ln(0.998) = 150\times(-0.0020020) = -0.30030

f(150) = e^{-0.30030} \approx 0.7407

Since $0.7407 > 0.60$ , the improved chemistry comfortably clears the 60% threshold at cycle 150 — about 74% of cluster signal is still correctly phased. Now find the maximum combined error rate $(\phi+\rho)_{max}$ that would put the cluster exactly at the 60% threshold at $c=150$ , solving $(1-x)^{150} = 0.60$ for $x$ :

150\ln(1-x) = \ln(0.60) = -0.5108

\ln(1-x) = -0.003406

1-x = e^{-0.003406} \approx 0.99660

x = (\phi+\rho)_{max} \approx 0.00340\ (0.34\%\text{ per cycle})

So the design margin between the achieved chemistry (0.20%/cycle) and the failure threshold (0.34%/cycle) is roughly 1.7-fold — a reasonable but not enormous safety factor, illustrating why NGS chemistry developers treat phasing/pre-phasing rates as tightly controlled, first-order design parameters rather than incidental noise. As a limiting-case sanity check: if $\phi+\rho \to 0$ exactly, $f(c)\to 1$ for all $c$ , meaning a perfectly synchronized cluster could in principle be read indefinitely — confirming that phasing error, not any fundamental optical or chemical limit, is what caps read length in this model.

Limiting cases, scope & common traps

Regime / question	What actually happens	Common mistake
$[ddNTP] \to 0$	$p\to 0$ , $\langle L\rangle \to \infty$ ; no useful termination ladder forms, only full-length extension product	Assuming lower ddNTP always "just gives longer reads" without noting the ladder becomes too sparse near the primer to resolve early bases
$[ddNTP]/[dNTP]$ too high	$p$ large, $\langle L\rangle$ small; ladder is dense near primer but nothing terminates far out — far bases unread	Forgetting that a single ratio cannot cover both near and far positions well; real protocols may need primer walking or multiple reactions
Homopolymer runs (e.g., AAAAAA) in Sanger	No inherent chemical problem — each A position is an independent Bernoulli trial, dye/ladder resolves fine	Confusing this with the NGS homopolymer problem, where synchronized multi-base incorporation (e.g., in ion semiconductor or pyrosequencing) makes counting run-length harder
Homopolymer runs in Ion Torrent / pyrosequencing	Signal is proportional to the number of identical bases incorporated in one flow; long runs give proportionally larger but harder-to-quantize signal, causing indel errors	Assuming all NGS platforms have the same error profile as Illumina; homopolymer errors are platform-specific
$\phi+\rho \to 0$ in NGS	$f(c)\to 1$ for all cycles; read length limited only by reagent chemistry/optics, not synchrony loss	Treating phasing as a fixed, immutable NGS property rather than a chemistry-dependent parameter under active engineering improvement
Heterozygous SNP in Sanger trace	Two co-migrating, differently colored peaks at the same position	Misreading it as sequencing noise or as a homopolymer slippage artifact (which shifts position, not color-overlap)
Long-read platforms (PacBio, nanopore)	No termination step at all; signal is real-time kinetic (fluorescence pulse duration/order) or ionic-current blockade as ssDNA threads a pore	Assuming "next-generation" is a single fixed technology; long-read and short-read NGS use fundamentally different physical read-out principles

The chemist behind it

Frederick Sanger (1918–2013) is the only person to win two Nobel Prizes in Chemistry for two entirely separate breakthroughs in the chemistry of information-bearing biomolecules: the 1958 prize for determining the complete amino acid sequence of insulin (establishing that proteins have a exact, genetically specified sequence rather than being amorphous colloids), and the 1980 prize (shared with Walter Gilbert and Paul Berg) for developing the dideoxy chain-termination method of DNA sequencing. Sanger's insulin work already displayed the pattern that defined his career: patient, methodical fragmentation-and-reassembly logic, first cleaving a macromolecule into pieces small enough to characterize completely, then working out how the pieces must have connected. He carried the same intellectual habit into nucleic acids, developing the "plus and minus" sequencing method in the mid-1970s before refining it into the chain-termination approach using ddNTPs, first published in 1977 alongside the complete sequence of the bacteriophage ΦX174 genome — at 5,386 nucleotides, the first fully sequenced DNA genome. Notably self-effacing and famously uninterested in titles or administrative power, Sanger worked at the MRC Laboratory of Molecular Biology in Cambridge for his entire career and reportedly turned down a knighthood (though he did accept the Order of Merit), preferring to remain simply "in the lab."

Origins · 1 minFrederick Sanger and how this was discovered

Frederick Sanger

1918–2013 · MRC Laboratory of Molecular Biology, Cambridge, UK

Sanger trained originally in biochemistry at Cambridge and spent his career solving sequence-determination problems methodically — first for proteins (insulin, 1955), then for RNA fragments, and finally for DNA itself via the dideoxy method (1977). His approach exemplifies a recurring theme in biochemistry: complex macromolecular information can be read out by controlled, quantifiable chemical termination or cleavage events, provided the resulting fragments can be separated and detected with enough resolution.

History

- 1955 — Sanger completes the full amino acid sequence of bovine insulin, the first protein ever fully sequenced.

- 1975 — Sanger and Coulson publish the "plus and minus" DNA sequencing method, a precursor using differential nucleotide incorporation.

- 1977 — Sanger, Nicklen, and Coulson publish the dideoxy chain-termination method; the ΦX174 bacteriophage genome (5,386 nt) is fully sequenced, the first complete DNA genome.

- 1980 — Sanger shares the Nobel Prize in Chemistry with Walter Gilbert and Paul Berg for nucleic acid sequencing methods.

- 1986 — Applied Biosystems commercializes the first automated fluorescent capillary Sanger sequencer, enabling the Human Genome Project's scale-up.

- 2005–2008 — 454 (pyrosequencing), Illumina/Solexa (reversible terminator sequencing-by-synthesis), and SOLiD platforms launch, inaugurating the "next-generation" high-throughput era.

- 2011–present — Oxford Nanopore and Pacific Biosciences commercialize long-read, single-molecule real-time sequencing technologies without any chain-termination step at all.

Why can't Sanger sequencing just use 100% ddNTPs to guarantee termination at the very first base?

Using pure ddNTP with no normal dNTP would indeed terminate every strand at the first incorporation, but that defeats the purpose: sequencing requires a distribution of fragment lengths spanning the whole read window, since each length reports the identity of one base position. The reaction needs $p$ small enough that most strands extend well past the primer before terminating, so that the resulting population of fragments forms an unbroken ladder covering every position from 1 to several hundred nucleotides. A vanishingly small ddNTP fraction (large dNTP excess) achieves this; pure ddNTP would only ever report the identity of position 1.

How do fluorescent dye-labeled ddNTPs allow a single-tube, four-color Sanger reaction instead of four separate tubes?

Each of the four ddNTPs (ddATP, ddCTP, ddGTP, ddTTP) is covalently attached to a spectrally distinct fluorophore (commonly via a linker off a modified base position that does not interfere with Watson-Crick pairing or 5′→3′ synthesis). All four labeled ddNTPs plus all four unlabeled dNTPs are combined in one reaction with template and primer. Because each terminated fragment carries the color corresponding to whichever ddNTP stopped it, a single capillary electrophoresis run with a four-channel laser detector can resolve both fragment size (by migration time) and terminal base identity (by color) simultaneously, eliminating the need for four parallel lanes.

Why does Illumina sequencing need reversible terminators instead of the permanent dideoxy terminators used in Sanger sequencing?

Sanger sequencing deliberately wants permanent termination, because it is reading a population of independently terminated fragments by size — one reaction, one read, many different stopping points. NGS sequencing-by-synthesis instead wants every cluster to add exactly one base per cycle in lockstep, then be imaged, then continue; a permanent dideoxy terminator would stop the reaction after the very first base and prevent any further cycles. Reversible terminators carry a removable blocking group at the 3′ position (chemically cleaved after imaging, regenerating a free 3′-OH) so the same strand can be extended one controlled base at a time across hundreds of cycles.

What causes phasing versus pre-phasing errors mechanistically, and can they ever be fully eliminated?

Phasing occurs when a strand within a cluster fails to incorporate a nucleotide during a given cycle — often because the reversible terminator or fluorescent dye was not efficiently delivered or because of transient polymerase stalling, so that strand falls one cycle behind its neighbors. Pre-phasing occurs when incomplete removal of the terminating/blocking group during the cleavage step leaves a free 3′-OH prematurely, allowing that strand to incorporate an extra, unintended base ahead of schedule. Both are fundamentally incomplete-reaction-efficiency problems in a multi-step enzymatic/chemical cycle repeated on millions of molecules, and since no enzymatic or chemical step is ever 100.000% efficient, some nonzero rate is thermodynamically and kinetically unavoidable, though it can be engineered smaller.

Why do long-read platforms like Oxford Nanopore not use chain termination at all?

Nanopore sequencing reads a native or minimally modified single strand of DNA (or RNA) as it is threaded through a protein nanopore embedded in a membrane, under an applied voltage; each passing base (or short k-mer window of bases) perturbs the ionic current in a sequence-characteristic way that is decoded computationally, in real time, without any synthesis step at all. Because there is no polymerization, dideoxy termination, or fluorescent-cycle chemistry involved, there is no termination-probability or phasing-decay limitation on read length — reads are limited instead by the physical integrity of the DNA molecule itself, which can span tens of thousands to over a million bases if handled gently during extraction.

Is Sanger sequencing now obsolete given the throughput of NGS?

No — Sanger sequencing remains standard for targeted, low-throughput, high-confidence confirmation, particularly in clinical diagnostics where a single, unambiguous read of a specific exon or variant is needed and regulatory guidelines often specifically require or strongly favor it for final reporting. Its per-reaction cost is far higher than NGS on a per-base basis, but for confirming one known variant in one sample, Sanger is faster, cheaper, and simpler than running a full NGS library preparation and sequencing run, so the two technologies occupy complementary rather than competing niches in most modern laboratories.

Why does the geometric distribution, rather than some other distribution, correctly describe Sanger fragment lengths?

The geometric distribution is the unique discrete probability distribution describing the number of independent, identically distributed Bernoulli trials needed to observe the first "success" (here, termination), which is exactly the physical situation: each templated position offers an independent, constant-probability chance of ddNTP incorporation, with no memory of prior positions influencing later ones (memorylessness is the defining property of the geometric distribution). Any deviation from true memorylessness — for example, if a nearby paused secondary structure temporarily changed local nucleotide accessibility, or if polymerase stalling depended on sequence context — would make the distribution only an approximation, but under standard reaction conditions with well-mixed constant reagent concentrations, the geometric model matches experimental band-intensity profiles closely.

DNA Sequencing: Sanger and Next-Generation Methods

Where this lives today

The physical picture — sugar geometry, primed synthesis, and the read-out chain

The core ideas — termination probability and read-length statistics

Worked examples

Limiting cases, scope & common traps

The chemist behind it

Before this