The Luria-Delbrück Fluctuation Test Explained
In 1943, a physician-turned-virologist and a theoretical physicist devised an elegant experiment using test tubes, culture plates, a bacterial virus and careful counting. Salvador Luria and Max Delbrück asked a deceptively simple question: do bacteria acquire resistance to a deadly virus before they meet it, or after, in response to the threat? Their answer, published in the journal Genetics under the title "Mutations of Bacteria from Virus Sensitivity to Virus Resistance," settled a long-running argument about how heredity works in microbes. It became one of the founding papers of bacterial genetics, and it was part of the body of work for which Luria, Delbrück and Alfred Hershey shared the 1969 Nobel Prize in Physiology or Medicine.
Why the question mattered: Lamarck vs. Darwin at the bacterial level
The idea of acquired characteristics, that traits developed in response to the environment can be inherited, had largely been abandoned for plants and animals by the 1940s. Bacteria were a different story. Microbiologists routinely watched cultures "adapt": a population exposed to a toxin or a virus would be wiped out, and then a resistant population would grow back. Many bacteriologists interpreted this as a physiological response, the cells somehow being trained or induced by the challenge, and some doubted that bacteria even had genes in the same sense as fruit flies or maize.
If bacterial resistance was induced by the challenge itself, then bacteria would be a standing exception to the genetic principles that applied everywhere else. If, on the other hand, resistance arose by chance and the challenge merely revealed and selected it, then bacteria obeyed the same rules as other organisms, and Darwin's logic of variation followed by natural selection would hold even in the microbial world. The problem was that both hypotheses predicted the same headline result. Expose a large culture to a virus, and a few resistant survivors appear either way. Something cleverer than simply counting survivors was needed.
The researchers: a refugee and a physicist
Salvador Luria was an Italian physician who had trained in radiology and become interested in bacteriophage in Rome. The Fascist racial laws of 1938 forced him, as a Jew, out of Italy. He worked in Paris, fled again when France fell in 1940, and reached the United States, where he eventually took a post at Indiana University.
Max Delbrück was a German theoretical physicist who had studied with Niels Bohr and moved into biology in the 1930s. His 1935 paper with Nikolai Timoféeff-Ressovsky and Karl Zimmer, which treated the gene as a physical, molecular structure that could be altered by radiation, was a major influence on Erwin Schrödinger's book What Is Life?. Delbrück came to the United States in 1937 and was teaching at Vanderbilt University by the early 1940s.
Both men were drawn to phage for the same reason: it was the simplest self-replicating system available, and it could be studied with the quantitative rigor physicists were used to. Together with Alfred Hershey they formed the core of what became known as the "phage group," an informal network that set shared standards for strains and methods. The Nobel Committee's 1969 citation honored the three "for their discoveries concerning the replication mechanism and the genetic structure of viruses."
The slot-machine insight
The key idea came, by Luria's own account, at a faculty dance at a country club near Bloomington, Indiana, in early 1943. Watching a colleague feed coins into a slot machine, Luria reflected that the machine paid out small amounts often and a large jackpot only rarely. The payout distribution was wildly uneven, and that unevenness was a consequence of how the machine worked.
He realized that bacterial mutations would behave the same way if they occurred at random during growth. A mutation that happened early in a culture's growth would be inherited by every descendant of that cell, producing a "jackpot" of resistant cells by the time the culture was full grown. A mutation that happened late would leave only one or a few resistant cells. Since growth is exponential, most of the cells in a culture are produced in the final few generations, so most cultures would contain few mutants while an occasional culture would contain an enormous number. Luria wrote to Delbrück with the idea, and Delbrück worked out the mathematics that turned an intuition into a testable prediction.
The setup: E. coli and phage T1
The experiment used a strain of Escherichia coli known as E. coli B and a phage that is now called T1 (in the paper it went by an earlier name). In an ordinary culture of sensitive E. coli, T1 infects and kills almost every cell. Occasionally a cell survives because it carries a heritable change that makes it resistant; it then grows into a visible colony on an agar plate covered with phage. Resistance in this system is now understood to involve loss or alteration of the receptor on the bacterial outer membrane that the phage uses to attach, so a resistant cell simply cannot be infected.
Bacteriophage: a virus that infects bacteria, often shortened to "phage." Because phage infections are fast, cheap and countable, phage became the main model system for early molecular genetics.
The two hypotheses and their predictions
The heart of the paper is a comparison between two models, each with a precise statistical prediction.
Hypothesis 1: acquired immunity (induction). Every cell has some small, equal chance of becoming resistant at the moment it encounters the phage. Under this model, the number of survivors on each plate depends only on how many cells were plated. If you plate many independent cultures of the same size, the counts should scatter around the average in the way rare independent events always do, following a Poisson distribution. A key property of the Poisson distribution is that the variance equals the mean. If the average is 10 resistant colonies per plate, most plates will show somewhere between about 4 and 16.
Hypothesis 2: random mutation before exposure. Resistance mutations arise spontaneously during growth, at some constant rate per cell division, regardless of whether phage is present. The phage only reveals them. Under this model, the counts across independent cultures should follow a very different, highly skewed distribution, now called the Luria-Delbrück distribution. Most cultures have few or no resistant cells, a minority have many, and a rare culture has a huge jackpot. Its variance is far larger than its mean.
So the test was not "are there survivors?" but "how are the survivors distributed across many independent cultures?" Small variance would point to induction. Huge variance, with jackpots, would point to pre-existing random mutation.
The experiment: parallel cultures and a control
Luria set up a series of small, independent cultures, each started from a very small number of sensitive cells so that no culture was likely to begin with a resistant mutant already in it. Each culture grew in liquid broth until it contained a large number of bacteria. Then the entire contents of each culture, or a fixed fraction, were spread on phage-coated plates, and resistant colonies were counted after incubation.
As a control, he also took a single large culture, divided it into many equal samples, and plated each of those. Because all of these samples came from the same culture, any mutants in it had already been mixed evenly through the liquid. The variation among those samples could only come from ordinary sampling error, so their counts should follow a Poisson distribution whichever hypothesis was true.
The results were clear. The samples from a single culture behaved as expected: their counts clustered around the average, with variance close to the mean. The independent cultures behaved completely differently. Many had zero or only a handful of resistant colonies, while a few had far more than the average. In the independent series, the variance was many times larger than the mean, far beyond anything sampling noise or uniform induction could produce. The pattern matched the jackpot prediction of the random-mutation model.
This is the opposite of a common retelling, which claims the data "matched a Poisson distribution." It was the single-culture control that looked Poisson. The independent cultures departed sharply from Poisson, and that departure was the evidence.
Estimating a mutation rate
The paper did more than reject the induction hypothesis. Delbrück's analysis also showed how to estimate the mutation rate, the probability per cell division that a sensitive cell gives rise to a resistant one, from the fluctuation data. One method used the fraction of cultures that contained no mutants at all. Under the random-mutation model, that fraction depends directly on the expected number of mutation events per culture, so counting the empty cultures gives an estimate of the rate. A second method used the mean number of mutants per culture.
This was a major step in itself. Before 1943 there was no reliable way to measure how often a bacterial gene mutated. Luria and Delbrück's approach turned the problem into an exercise in counting and probability. Their estimates were tiny per cell division, which fits the general picture that has held since: individual genes in bacteria mutate rarely per division, but bacterial populations are so large that rare mutants are almost always present somewhere.
Follow-up experiments: making the invisible visible
The fluctuation test was a statistical argument. It inferred that resistant mutants existed before selection, but it never showed anyone a resistant cell that had never seen a phage. Some microbiologists remained unconvinced, and two later experiments closed the gap.
Newcombe's spreading experiment (1949)
Howard Newcombe, a Canadian geneticist, published a simpler test in Nature in 1949. He spread sensitive bacteria on plates and let them grow for a few hours into microcolonies. Then he took some of the plates and re-spread them, smearing the microcolonies around the plate, while leaving others undisturbed. Both sets were then challenged with phage.
If resistance was induced only on contact with phage, re-spreading should make no difference. If resistant mutants had already arisen and divided into small clusters before phage arrived, re-spreading would scatter the members of each cluster, and each would grow into its own resistant colony. The re-spread plates yielded far more resistant colonies than the undisturbed ones, exactly as the pre-existing-mutation model predicted.
The Lederbergs' replica plating (1952)
The most visually convincing confirmation came from Joshua and Esther Lederberg in 1952, in a paper in the Journal of Bacteriology titled "Replica plating and indirect selection of bacterial mutants." Their technique used a square of sterile velvet pressed onto a master plate of colonies grown without any selection. The velvet fibers picked up cells from every colony and could then stamp an identical pattern onto several fresh plates, some of them containing the selective agent.
By comparing positions, the Lederbergs could identify which colonies on the original, never-exposed master plate gave rise to resistant growth on the selective plates. They then went back to those colonies on the master plate and enriched for resistance through repeated rounds, isolating resistant bacteria that had never been in contact with the selective agent. This "indirect selection" provided direct physical evidence for what the fluctuation test had inferred mathematically.
If you want to see how other mid-century experiments established the molecular basis of heredity, the Hershey-Chase experiment and the Griffith transformation experiment form natural companions to this story.
The adaptive mutation debate
For more than four decades, the fluctuation test and replica plating were treated as a closed case: mutations are random with respect to their usefulness. Then, in 1988, John Cairns, Julie Overbaugh and Stephan Miller published a paper in Nature called "The origin of mutants" that reopened the question.
Cairns and colleagues pointed out a limitation in the classic design. In the Luria-Delbrück experiment, the selective agent kills sensitive cells quickly. Any mutation that might be induced by the phage would have no chance to happen, because the cell would be dead first. The experiment could detect pre-existing mutants, but it could not rule out mutations that arise under non-lethal stress. Cairns and colleagues studied E. coli that could not use lactose, placed on plates where lactose was the only food. The cells were starved but not killed. Over days, mutants able to use lactose kept appearing. The paper raised the provocative possibility that bacteria might in some sense "choose" which mutations to make.
The claim triggered years of intense work. A widely used test system, described by Cairns and Patricia Foster in a 1991 Genetics paper, used a strain called FC40 carrying a leaky frameshift mutation in the lac genes on an F′ plasmid. Several broad lines of explanation emerged:
- Stress-induced mutagenesis. Starving cells activate stress responses that switch on error-prone DNA repair, including the polymerase DinB (Pol IV). This raises the mutation rate broadly, not just in useful genes. Mutations are more frequent under stress but still not directed toward a benefit. Work by Susan Rosenberg, Patricia Foster and others developed this view.
- Gene amplification and selection. John Roth, Dan Andersson and colleagues argued that the leaky lac allele could be duplicated and amplified many times under selection, giving the cell enough enzyme activity to grow slowly. Amplification increases the number of target copies, making a reversion more likely, and the extra mutagenesis seen in the system may partly reflect coamplification of the nearby dinB gene on the same plasmid. In this view, the phenomenon is mostly selection acting on ordinary variation.
- Special features of the test system. The lac allele sat on a conjugative plasmid, and plasmid-specific processes contributed to the elevated mutation rates seen in this one setup.
Most researchers today would summarize the outcome this way: genuinely directed mutation, in which a cell targets the specific change it needs, has not been established. What has been established is that mutation rates are not fixed. Stress can raise them, sometimes in a regulated way, and amplification and selection can make adaptive outcomes appear surprisingly fast. The core result of Luria and Delbrück, that useful mutations are not created on demand by the selective agent, has survived.
Criticism and limitations of the original test
The fluctuation test is a powerful design, but it carries assumptions worth knowing.
Mutants and non-mutants must grow at the same rate. The original model assumes resistant cells divide as fast as sensitive ones. In practice many resistance mutations carry a fitness cost, and slower growth of mutants distorts the distribution.
Plating must be efficient. The model assumes every resistant cell plated becomes a colony. If some fail to grow, counts are biased low.
Phenotypic lag. A newly mutated cell may still carry old, sensitive receptors on its surface for a few generations before the new phenotype appears. This delay affects the counts, especially for the latest mutations.
Only lethal selection was tested. As Cairns pointed out, killing cells quickly means the design cannot by itself test whether non-lethal stress affects mutation.
The mathematics was hard. The Luria-Delbrück distribution has no simple closed form, which made practical estimation of rates awkward for decades. Douglas Lea and Charles Coulson published the expected distribution of mutant numbers in 1949 and offered an estimation method based on the median of the counts, which is less sensitive to jackpots than the mean.
None of these issues overturns the main conclusion, but they matter greatly for anyone using fluctuation analysis to measure precise mutation rates.
Common misconceptions
"The test proved all mutations are random." It showed that resistance to phage T1 in E. coli arose before exposure. It did not show that mutation rates are uniform across the genome or unaffected by the environment. Some sites mutate far more often than others, and stress can raise rates overall. "Random" in this context means random with respect to usefulness, not equally likely everywhere.
"The data fit a Poisson distribution, proving randomness." The reverse is closer to the truth. The Poisson distribution was the prediction of the induction model. The independent cultures showed far more variance than Poisson allows, and that excess was the signature of mutations arising at different times during growth.
"Bacteria are passive victims of mutation." Bacteria regulate their own DNA repair, can transiently increase mutation under stress, and swap genes horizontally through plasmids and phages. The fluctuation test addressed only where one particular kind of variation came from.
"Selection creates the mutants." Selection creates nothing. It changes which pre-existing variants become common. That distinction is the core lesson of the experiment.
Modern applications
Measuring mutation rates
Fluctuation analysis remains the standard laboratory method for measuring mutation rates in bacteria and yeast. Researchers grow many parallel cultures, plate them on a selective medium, and fit the distribution of mutant counts. Modern estimators are far better than those available in 1943. FALCOR, a web-based fluctuation analysis calculator described by Hall and colleagues in 2009, offers the Lea-Coulson method of the median and the Ma-Sandri-Sarkar maximum-likelihood estimator, among others. The R package rSalvador provides maximum-likelihood estimates and confidence intervals, with options to account for practical complications such as plating efficiency.
Antibiotic resistance
The central insight of the fluctuation test is now standard in thinking about antibiotic resistance. In a large bacterial population, mutants resistant to a single drug often already exist before treatment begins. A course of antibiotics does not usually teach bacteria to become resistant. It removes their sensitive competitors and lets pre-existing resistant cells take over. That reasoning underpins the rationale for combination therapy in tuberculosis and HIV: if resistance to each drug arises independently and rarely, the chance that a single cell carries resistance to two or three drugs at once is far smaller. It also underpins stewardship efforts to avoid unnecessary antibiotic use, since every exposure is a chance to select whatever resistant variants are present.
Mutation is only part of the picture. Many clinically important resistance genes spread between bacteria horizontally, on plasmids and other mobile elements, a route Luria and Delbrück did not consider and that complicates simple mutation-rate reasoning.
Cancer
The same logic has been applied well beyond bacteria. In cancer, a tumor may contain rare cells carrying resistance to a drug before treatment starts, which helps explain why many targeted therapies initially shrink tumors only for resistant clones to regrow. This is one reason oncologists combine drugs with different mechanisms.
Why the experiment endures
The Luria-Delbrück fluctuation test is remembered not because it used sophisticated equipment but because it asked the right statistical question. Instead of looking at the average outcome, which both hypotheses predicted equally well, it looked at the variance across independent replicates, where the two hypotheses diverged sharply. That move, using the shape of a distribution rather than its mean to tell competing mechanisms apart, is now a general tool in biology.
It also helped establish bacteria as genetic organisms whose genes mutate like those of any other living thing. That conclusion opened the way to bacterial conjugation, transduction, and the molecular biology that followed, including studies of DNA replication and the genetic code.
Sources
- Luria SE, Delbrück M (1943). Mutations of Bacteria from Virus Sensitivity to Virus Resistance. Genetics 28:491–511
- The Nobel Prize in Physiology or Medicine 1969 (NobelPrize.org)
- Genes to Genomes (Genetics Society of America): Luria-Delbrück, jackpots and epiphanies
- Profiles in Science, National Library of Medicine: Salvador E. Luria papers
- Newcombe HB (1949). Origin of Bacterial Variants. Nature 164:150–151
- Lederberg J, Lederberg EM (1952). Replica plating and indirect selection of bacterial mutants. Journal of Bacteriology 63:399–406
- Cairns J, Foster PL (1991). Adaptive reversion of a frameshift mutation in Escherichia coli. Genetics
- Roth JR et al. (2003). Origin of mutations under selection: the adaptive mutation controversy
- Hall BM et al. (2009). Fluctuation AnaLysis CalculatOR (FALCOR): a web tool for the determination of mutation rate
- Zheng Q (2017). rSalvador: An R Package for the Fluctuation Experiment. G3
- Wikipedia: Luria-Delbrück experiment
FAQ
Q: Why did Luria and Delbrück use bacteriophage and not antibiotics? A: Phage was the central experimental tool of the small group of scientists working on bacterial and viral genetics in the early 1940s, and phage resistance is an all-or-nothing trait: sensitive cells die, resistant cells form colonies. That made counting resistant cells simple and reliable. The same logic applies to antibiotics.
Q: What does the Poisson distribution have to do with the test? A: The Poisson distribution was the prediction of the induction hypothesis. If every cell had an equal chance of becoming resistant on contact with phage, counts across cultures would scatter with variance roughly equal to the mean. The independent cultures instead showed variance far larger than the mean, with occasional jackpots, which is the signature of mutations that arose at different times during growth.
Q: What is a "jackpot" culture? A: A jackpot is a culture in which a resistance mutation happened early in growth, so the mutant cell had many generations to multiply. Such a culture ends up with a very large number of resistant cells. Jackpots are rare, but their existence is what makes the Luria-Delbrück distribution so skewed.
Q: How did replica plating confirm the result? A: In 1952 Joshua and Esther Lederberg used velvet to copy colonies from an unselected master plate onto selective plates. They identified which colonies on the master plate contained resistant cells and isolated resistant bacteria that had never been exposed to the selective agent, physically showing that resistance existed before selection.
Q: Did the adaptive mutation debate overturn Luria and Delbrück? A: No. The 1988 Cairns paper showed that the classic design could not detect mutations arising under non-lethal stress. Later work found that stress can raise mutation rates and that gene amplification plus selection can produce rapid adaptation, but directed mutation toward a specific useful change has not been established.
Q: How are fluctuation tests used today? A: Fluctuation analysis is still the standard way to measure mutation rates in bacteria and yeast. Researchers grow many parallel cultures, count mutants on selective plates, and fit the distribution using tools such as FALCOR or the rSalvador R package to estimate the rate per cell division.
Q: Why did it take until 1969 for them to win the Nobel Prize? A: The prize recognized their broader body of work on the replication and genetic structure of viruses, shared with Alfred Hershey, not the fluctuation test alone. Over the decades after 1943 the phage group's methods and findings became foundational to molecular biology, and the Nobel Committee honored that cumulative contribution.
Read next:
Follow new articles
Add SciHub101 to your feed reader to see new science and technology stories.
About the Author
SciHub101 is an independent science and technology publication founded and edited by Buğra Sözeri, a software engineer and app developer based in Türkiye. AI tools assist with drafting and research. Source documentation in the existing archive is uneven, and articles are being reviewed and corrected. The Editorial Team byline identifies editorial responsibility; it does not certify specialist or independent human review. Where we compare products, anything built by our own team is labelled as such — see the editorial guidelines and About page for how corrections and disclosures work.
Related Articles

Evolution and Natural Selection — How Life Changes
Darwin's theory, modern synthesis, mechanisms, evidence, speciation, and human evolution — explained clearly with common misconceptions corrected.

The Lenski Long-Term Evolution Experiment: Evolution in Real Time
Since 1988, Richard Lenski has grown E. coli for 80,000+ generations. The result: a living record of evolution, including a once-impossible citrate mutation, visible in real time.

The Fox Domestication Experiment: Breeding Tameness in Real Time
Starting in 1959, Soviet geneticist Dmitry Belyaev selectively bred silver foxes for tameness alone. Within a few generations, the foxes developed floppy ears, piebald coats, and dog-like behavior.

Mendel's Pea Plant Experiments: How Genetics Was Born
Between 1856 and 1863, Gregor Mendel cross-bred 30,000 pea plants and found the mathematical rules of heredity. Here is what he did and why it still defines genetics.