The Meselson-Stahl Experiment

DNA replication is the process by which a cell copies its genome so that each daughter cell receives a complete set of genetic instructions. Because the two strands of the double helix are complementary and antiparallel (see Base Pairing and Antiparallel Complementarity), each strand can in principle specify the sequence of a new partner strand. The unresolved question in 1953 was not whether copying was possible, but which physical arrangement of old and new material actually occurs when a cell divides. Three models were on the table. In the conservative model, the original double helix remains completely intact, and an entirely new double helix is synthesized from scratch; after one round of replication you would have one all-old molecule and one all-new molecule. In the semiconservative model, the two parental strands separate, and each serves as a template for a new complementary strand; every daughter duplex consists of one old strand paired with one new strand. In the dispersive model, the parental duplex is broken into fragments at multiple points, and new DNA is synthesized in patches; both strands of both daughter molecules would end up as a mosaic of old and new segments. These models make different, testable predictions about the distribution of parental material across generations — not about sequence, but about the physical continuity of strands. Testing them required a way to distinguish "old" DNA from "new" DNA by some physical property other than base sequence, and a way to measure that property with enough precision to separate molecules that differ by only a few percent in mass. That combination — isotopic labeling plus equilibrium density centrifugation — is exactly what Meselson and Stahl assembled, building on physical chemistry developed by Jerome Vinograd for separating macromolecules by buoyant density.

Isotopic labeling as a physical tag

The experiment's cleverness lies in using an isotope of nitrogen as an invisible physical label rather than a radioactive or chemical tracer. Nitrogen exists naturally as ¹⁴N (the common isotope) and as the rare, stable, non-radioactive isotope ¹⁵N. Because DNA's nitrogenous bases contain nitrogen atoms, DNA synthesized in a medium where ¹⁵N is the only nitrogen source becomes uniformly "heavy" — every base, and therefore every strand, is denser than ordinary DNA, but the chemistry and base-pairing behavior are unchanged. This makes ¹⁵N an ideal tracer: it changes mass without changing structure or function. Meselson and Stahl grew E. coli for many generations in a medium containing ¹⁵NH₄Cl as the sole nitrogen source, so that essentially all cellular DNA became fully heavy (¹⁵N/¹⁵N) DNA. They then abruptly shifted the culture to ordinary ¹⁴N medium and harvested samples at successive generation times, extracting DNA from each sample for analysis.

Equilibrium density-gradient centrifugation

To resolve DNA molecules that differ only slightly in isotopic composition, they used a technique invented by Meselson, Stahl, and Vinograd: equilibrium sedimentation in a cesium chloride (CsCl) density gradient. A concentrated CsCl solution is spun at very high speed — roughly

140,000 × g for 20+ hours — in an ultracentrifuge. Under this force, the heavy cesium and chloride ions themselves begin to sediment, but diffusion opposes this, and the two processes reach a steady state that establishes a smooth, stable gradient of solution density along the length of the tube, from lighter near the top to denser near the bottom. Any DNA present in the tube migrates until it reaches the position in the gradient where the solution density exactly equals its own buoyant density — it "floats" at its own weight, neither sinking nor rising further. Because DNA composed of ¹⁵N atoms is measurably denser than ordinary ¹⁴N DNA (roughly 1% denser overall, enough to shift its equilibrium position by a resolvable distance), fully heavy DNA bands lower in the tube (higher density), fully light DNA bands higher (lower density), and a hybrid molecule with one heavy strand and one light strand bands at an intermediate position — essentially the arithmetic average of the two densities. The bands are visualized using ultraviolet absorption photography, since DNA absorbs UV light strongly at 260 nm, producing sharp dark bands against the gradient. This is the crucial resolving power of the method: it does not just distinguish "heavy" from "light" DNA, it precisely resolves a third, intermediate density corresponding to hybrid duplexes — exactly the species predicted uniquely by the semiconservative model after one round of replication.

Predicted banding patterns, generation by generation

Starting entirely ¹⁵N/¹⁵N (fully heavy) and shifting to ¹⁴N medium, each model predicts a distinctive sequence of bands:

Conservative model — Generation 1 would show two bands: one fully heavy (the untouched parental duplexes) and one fully light (entirely new duplexes), in a 1:1 ratio, with no intermediate band ever appearing at any generation.

Semiconservative model — Generation 1 would show a single band at hybrid (intermediate) density, since every daughter duplex has exactly one old heavy strand and one new light strand. Generation 2 would show two bands — hybrid and fully light — in a 1:1 ratio, because the heavy parental strands only exist in fixed number and get diluted across an increasing pool of light strands, while the ratio of hybrid-to-light stabilizes at 1:1 and stays there in later generations (only the light band grows).

Dispersive model — Generation 1 would show a single hybrid band similar to semiconservative, but at every subsequent generation this band would shift progressively toward lighter density (never resolving into two discrete bands), because each strand becomes an ever-finer patchwork mosaic of old and new material rather than a clean single-strand unit.

What the gradient actually showed

Meselson and Stahl grew ¹⁵N-labeled E. coli, shifted to ¹⁴N medium, and sampled DNA at intervals corresponding to successive generations (E. coli doubling time under their conditions was about 30–40 minutes generously spaced to allow clean sampling). The observed pattern:

This is exactly, precisely, the semiconservative prediction: a single hybrid band after one generation (ruling out conservative replication immediately, since conservative predicts two bands with no intermediate), followed by a clean 1:1 split into hybrid and light bands after two generations that does not shift position at later generations (ruling out dispersive replication, since dispersive predicts continued smearing toward lighter density). The hybrid band's position stayed fixed generation after generation — only its relative amount shrank as a fraction of total DNA — which is the signature that individual strands remain physically intact units rather than being fragmented and reshuffled. As a final confirmatory step, Meselson and Stahl took the hybrid-density DNA isolated from generation 1 and heated it to denature the duplex into single strands. The single strands then banded at two distinct positions — one fully heavy, one fully light — rather than at a single intermediate position. This proved that the hybrid duplex was genuinely composed of one intact heavy strand and one intact light strand, not some uniformly intermediate chemical species, directly confirming that whole strands, not fragments, are the conserved unit.

The Meselson-Stahl experiment (1958, PNAS) is itself the classic evidence for semiconservative replication — it is unusual among landmark experiments in that the "evidence" section and the "mechanism" section are the same event. But it's worth being explicit about exactly what was measured, and why the result is airtight rather than merely suggestive.

What was done. E. coli was cultured for at least 14 generations in a minimal medium in which ¹⁵NH₄Cl was the sole nitrogen source, ensuring essentially complete labeling of cellular DNA with heavy nitrogen. The culture was then rapidly diluted into an excess of ordinary ¹⁴N medium (a large excess so that virtually all newly synthesized DNA would incorporate ¹⁴N), and growth continued. Samples were withdrawn at time points corresponding to zero, one, two, and further generations. DNA was extracted from each sample and subjected to equilibrium sedimentation in CsCl in an analytical ultracentrifuge, with the resulting bands recorded by UV absorption photography and quantified by densitometer tracing.

What was observed. At generation 0, a single band at the heavy position. After one generation (DNA content of the culture had doubled), a single band at a position intermediate between heavy and light — not two bands, not a smear, one sharp band at the hybrid density. After two generations, two bands appeared: one at the same hybrid position as before, one at the fully light position, in an approximately equal (1:1) amount. In subsequent generations, the hybrid band remained fixed in position and became a progressively smaller fraction of the total, while the light band grew — precisely as predicted by fixed-number heavy parental strands being diluted across an expanding population of new light strands.

What it proved. The single intermediate band at generation 1 is incompatible with the conservative model, which requires two separate bands (heavy and light) with no intermediate species at any point — since in that model original duplexes never mix material with new ones. The subsequent clean resolution into a stable hybrid band plus a growing light band, with the hybrid band never shifting position or smearing toward lighter density over time, is incompatible with the dispersive model, which requires the hybrid band to progressively lighten each generation as old material gets diluted within increasingly fragmented strands. Only the semiconservative model predicts exactly this two-stage signature: one hybrid generation, then a permanent 1:1 split between hybrid and light that holds indefinitely. The confirmatory denaturation experiment — showing that heat-denatured hybrid DNA separates into one fully heavy and one fully light single strand, rather than a single intermediate species — closed the last logical gap by proving that whole strands, not fragments, are the physically continuous unit of inheritance. The experiment's persuasive power comes from its combination of quantitative precision (density differences smaller than 1% resolved cleanly) and the fact that it tested three sharply distinguishable, mutually exclusive predictions rather than a vague expectation. This is why Meselson and Stahl were confident enough to write, with characteristic understatement, that their results "are in exact accord with the expectation" of the semiconservative model.

Example 1 — Predicting the gradient pattern at generation 3. Suppose an E. coli culture, fully ¹⁵N-labeled at time zero, is shifted into ¹⁴N medium and allowed to double exactly three times (generation 3). Assume semiconservative replication and that every daughter cell's DNA replicates exactly once per generation, and that the original culture contained a normalized total of 1 unit of heavy (¹⁵N) DNA strands (so 2 strands' worth, since each duplex has two strands — track strand count, not duplex count, since strands are the conserved unit). Start: 2 heavy strands (1 heavy duplex).

• After replication 1: the 2 heavy strands each template one new light strand → 2 duplexes, each with 1 heavy + 1 light strand (2 hybrid duplexes, 2 heavy strands + 2 light strands total).
• After replication 2: each of the 4 strands (2 heavy, 2 light) templates a new light strand → 4 duplexes total. The 2 heavy strands now each pair with a new light strand (2 hybrid duplexes); the 2 light strands from before each pair with a new light strand (2 fully light duplexes). Total: 2 hybrid + 2 light = 4 duplexes.
• After replication 3: 8 strands total, always exactly 2 heavy strands (the original ones never disappear or get diluted in number, only in proportion) paired with 2 new light partners → 2 hybrid duplexes; the remaining 6 strands are all light, pairing among themselves → 6 light duplexes (3 duplexes... wait, strand count must be tracked carefully).

Careful strand bookkeeping: total strands double each generation: 2 → 4 → 8 → 16. The number of heavy strands is always exactly 2 (the original two parental strands, never destroyed, never diluted in copy number — only diluted as a fraction of the total). Each heavy strand always pairs with a newly made light strand, so heavy strands always yield exactly 2 hybrid duplexes, regardless of generation number. All remaining duplexes are fully light. At generation n, total duplexes = 2ⁿ. Hybrid duplexes = 2 (constant). Light duplexes = 2ⁿ − 2.

This is exactly what was observed: the hybrid band never disappears (it can't — there are only 2 original heavy strands, and they must go somewhere), but it becomes an ever-smaller fraction of the total DNA, exactly halving its relative proportion each generation — the pattern Meselson and Stahl tracked out to the fourth generation in their actual gradients.

Example 2 — Distinguishing the models with an asymmetric labeling scheme. Suppose a (hypothetical, non-historical) variant experiment labels only one of the two starting strands heavy — imagine an artificially constructed duplex with one ¹⁵N strand and one ¹⁴N strand at generation 0 — then replicates it once in ¹⁴N medium. What would each model predict for generation 1? Conservative model: The original duplex (already mixed, 1 heavy + 1 light strand) would remain physically intact as a unit and a wholly new light-light duplex would be synthesized alongside it. Result: one band at the original hybrid density (unchanged, since that duplex is "conserved") and one band at fully light density. Two bands, 1:1. Semiconservative model: The heavy strand separates and templates a new light strand → one hybrid duplex. The light strand separates and templates a new light strand → one fully light duplex. Result: one hybrid band, one fully light band, 1:1 — indistinguishable in this special case from the conservative prediction! This thought experiment reveals something important pedagogically: starting from a mixed duplex, generation 1 alone cannot distinguish conservative from semiconservative replication — both give a 1:1 hybrid:light split. This is precisely why Meselson and Stahl started from a fully homogeneous heavy population (all-¹⁵N duplexes), so that generation 1 gives a maximally diagnostic single hybrid band under semiconservative replication versus two separate bands under conservative replication. The choice of starting from uniformly labeled DNA, rather than any mixed population, was not incidental — it was the design feature that made the first generation alone decisive.

Matthew Meselson was a graduate student and later professor at Caltech and Harvard when he began, with Franklin Stahl, to design an experiment capable of physically distinguishing parental from newly synthesized DNA. Meselson had trained under Linus Pauling, and brought a strong physical-chemistry sensibility to a biological problem — precisely the mindset needed to invent a density-separation technique refined enough to resolve a roughly 1% mass difference between labeled and unlabeled DNA. Franklin Stahl, a phage geneticist, contributed the microbiological expertise in growing and sampling bacterial cultures with the precision the experiment demanded. The two met at a summer course at Cold Spring Harbor and began the collaboration that would produce, within about two years of intensive work at Caltech, the single most cited demonstration of a mechanism predicted by molecular structure. The work also depended critically on Jerome Vinograd, a Caltech physical chemist who had been developing cesium chloride equilibrium density-gradient centrifugation as a general technique for separating macromolecules by buoyant density; Meselson, Stahl, and Vinograd co-developed the method that made the experiment possible before Meselson and Stahl applied it to the replication question specifically.

The semiconservative principle established by Meselson and Stahl underlies essentially every downstream tool and concept in molecular biology and medicine that depends on accurate genome duplication. DNA replication fidelity and cancer: because each new duplex retains one intact parental template strand, errors introduced during synthesis of the new strand can be detected and corrected by mismatch repair systems that recognize which strand is the (methylated, older) template versus the newly synthesized strand — a discrimination that only makes sense because replication is semiconservative and strand age is a meaningful, trackable property. Failures in this system cause hereditary cancer syndromes such as Lynch syndrome.

PCR and genome sequencing exploit the same semiconservative logic at the heart of every replication cycle — each round of polymerase chain reaction produces new strands templated directly off old ones, doubling target DNA exponentially, a principle directly descended from understanding that a single strand is sufficient information to specify a full duplex.

Density-gradient methods descended from Meselson and Stahl's CsCl technique are still used to separate plasmid DNA from genomic DNA, to isolate GC-rich versus AT-rich genome fractions, and in chromatin immunoprecipitation variants that separate DNA by associated protein content.

Cell biology and aging research use variants of strand-labeling logic (e.g., BrdU or EdU pulse-labeling, conceptually descended from the ¹⁵N approach) to track which cells have divided, how many times, and whether stem cells asymmetrically retain older "immortal strand" template DNA — a still-debated hypothesis that traces its logic directly back to the semiconservative framework Meselson and Stahl established. Understanding strand continuity across generations also underlies modern work in Genome Organization Across Domains of Life, since replication origin number and fork architecture differ dramatically between bacterial and eukaryotic genomes while the semiconservative mechanism itself remains universal.

Ask the tutor: Why does starting with a fully heavy (not mixed) DNA population make generation 1 the single most diagnostic data point in the whole experiment?