The C-value Paradox

Is the paradox resolved? For the question as Thomas posed it, yes, and uncontroversially. There is no contradiction between DNA content and complexity once you know that most DNA does not encode proteins. Every genome sequenced since has confirmed the mechanism, and the composition of large genomes is directly readable.

T. Ryan Gregory argued in 2001 that continuing to call it a paradox is a mistake for exactly this reason — a paradox implies a contradiction, and there is none left. He proposed C-value enigma instead, and the term has been widely adopted. His point was not that the questions are answered. It was that they have changed:

• What determines how much non-coding DNA a lineage accumulates?
• Why do rates of DNA gain and loss differ so much between lineages?
• Genome size correlates tightly with cell size and inversely with cell division rate across many groups. Is this a cause, a consequence, or a coincidence?
• Is there an upper bound, and what enforces it?

Those are open. The mechanism of the variation is what remains unexplained; the existence of the variation is not.

Three competing accounts of why big genomes exist.

Selfish DNA (Doolittle and Sapienza; Orgel and Crick, 1980). Transposable elements accumulate because they replicate faster than they are purged. Genome size is a by-product of an intragenomic arms race, and needs no organism-level explanation at all.

Mutational hazard (Michael Lynch and colleagues, from around 2003). Extra DNA is slightly deleterious, so it is removed when selection is efficient enough to see it. Efficiency depends on effective population size. Species with small populations — large-bodied, long-lived organisms — have weaker purifying selection and therefore accumulate more junk. Genome size on this account is mostly a readout of population genetics, not of biology.

Nucleotypic function (Thomas Cavalier-Smith, and taken up by Gregory). The bulk of DNA has a physical role independent of its sequence: it sets nucleus size, which sets cell size, which affects metabolic rate, developmental rate and body plan. The correlation between genome size and cell size is one of the strongest in comparative biology, and salamanders — which have both enormous genomes and enormous cells, with correspondingly slow development and simplified anatomy — are the standard case.

These are not exclusive, and the current view mixes them. The disagreement is about weighting, and it is live.

The junk DNA argument. Susumu Ohno coined "junk DNA" in 1972 from a mutational load calculation: given observed mutation rates, a mammal cannot maintain more than a certain number of genes under selection, so most of the genome cannot be genes. The term has caused forty years of confusion, because "junk" was taken to mean "does nothing" when Ohno meant something closer to "not under selection as coding sequence".

In 2012 the ENCODE consortium reported that around 80 percent of the human genome is "biochemically functional", which was widely reported as the death of junk DNA. The criticism was immediate and, on the whole, persuasive: the definition of function used was extremely permissive — any reproducible biochemical activity, including transcription that produces nothing used. Dan Graur and others pointed out that this definition would classify most of the genome of an onion as functional too, and that evolutionary conservation, a far stricter test, marks roughly 8–10 percent of the human genome as under selection. The exchange is worth knowing because it is a clean example of a result whose substance depended entirely on an undeclared definition.

1948–1950. André Boivin, Roger and Colette Vendrely, and independently Hewson Swift, established that DNA content per nucleus is constant within a species and halved in gametes. Swift introduced the term "C-value" for that constant amount. At the time this was evidence that DNA, not protein, was the hereditary material.

1950s–1960s. Genome sizes were measured across an expanding range of species using densitometry of stained nuclei. The between-species picture became steadily more confusing. Amphibians, and especially lungfishes and salamanders, turned up with genomes far larger than mammals.

1971. Charles A. Thomas Jr. published "The Genetic Organization of Chromosomes" in the Annual Review of Genetics and gave the problem its name.

1968–1972. Roy Britten and David Kohne's work on DNA reassociation kinetics demonstrated that a large fraction of eukaryotic DNA consists of highly repetitive sequence, present in hundreds of thousands of copies. This is the empirical result that made the resolution possible: the extra DNA was visibly repetitive, and repetitive sequence cannot be a proportional store of information.

1972. Susumu Ohno published "So much 'junk' DNA in our genome".

1977. Introns were discovered independently by Richard Roberts and Phillip Sharp, showing that even protein-coding genes are mostly not coding sequence.

1980. Doolittle and Sapienza, and Orgel and Crick, published the selfish DNA argument in consecutive papers in Nature.

2001. T. Ryan Gregory published his review in Biological Reviews proposing the C-value enigma and setting out the surviving questions, particularly the relationship between genome size and cell size.

2001–2004. The human genome sequence returned a protein-coding gene count of roughly twenty thousand, well below most predictions, and the same puzzle reappeared in the new form of the G-value paradox.

2010. Jaume Pellicer, Michael Fay and Ilia Leitch measured Paris japonica at about 148.9 picograms per haploid genome, then the largest eukaryotic genome recorded.

It broke the equation between DNA and information. Before the resolution, it was natural to treat a genome as a blueprint whose length reflects the complexity of what it builds. The C-value paradox is the observation that killed that picture, and what replaced it — genomes as historical accumulations, shaped by selfish elements, duplication accidents and the limits of purifying selection — is much closer to how genomes actually behave.

It made junk DNA a prediction rather than an excuse. Ohno's argument and the C-value data together predicted, before any genome was sequenced, that most of a large genome would turn out not to be genes. Sequencing confirmed it. That is a successful prediction from a paradox, which is not common.

It underpins the modern view of transposable elements. Once the extra DNA was identified as largely transposon-derived, an entire field followed: how elements are silenced, how they occasionally get co-opted into useful roles — including in the vertebrate adaptive immune system and in placental development — and how they generate disease when they insert in the wrong place. All of that started from the question of why onions have so much DNA.

It is a durable lesson about proxies. Genome size was adopted as a measure of complexity because it was the only quantity anyone could measure. It measured something real; it simply did not measure the thing it was being used for. Fields adopt proxies like this constantly, and the C-value paradox is the cleanest available demonstration of how such a proxy fails and how long it takes anyone to notice.

Sequencing large genomes is still hard, and the reason is repeats. The reason the lungfish, the axolotl and conifer genomes were assembled long after the human genome is that they are dominated by near-identical repeated sequence, which short-read sequencing cannot place. Long-read technologies changed this, and the completion of a truly gap-free human genome sequence in 2022 was largely a matter of finally resolving the repetitive regions the original draft had skipped. Genome size is a direct predictor of assembly difficulty and cost, which makes the C-value a practical planning number for every large sequencing project.

Conservation and comparative genomics. Projects sequencing thousands of vertebrate and plant species have turned genome size from a curiosity into a measured comparative trait with tens of thousands of data points. The Animal Genome Size Database, maintained by Gregory, is the standard reference. The correlations that emerge — with cell size, developmental rate, metabolic rate, and in plants with seed mass and climate — are the live research questions.

Crop breeding. Polyploidy is not merely historical. Bread wheat is hexaploid, with three complete ancestral genomes and a total size around five times the human genome. Understanding which copy of a gene is doing what, in a genome carrying several near-identical copies of everything, is a central practical problem in plant breeding and in genome editing.

What "functional" means. The ENCODE argument is not settled and matters beyond genetics. Every claim that a fraction of the genome is functional depends on a definition — biochemical activity, evolutionary conservation, or measurable phenotypic effect on deletion — and these give answers differing by an order of magnitude. Anyone reading a percentage should ask which definition produced it.

A note on framing. The onion has become the field's standard reality check. The "onion test", proposed by Gregory, asks anyone claiming a universal function for non-coding DNA a simple question: whatever you think all that DNA is doing, why does an onion need five times more of it than a human, and why does one species of onion need twice as much as another? It is not a rigorous argument. It is a very effective one.

Units, and reading the literature

Genome size is quoted two ways. Cytometric measurements give mass in picograms of DNA per haploid nucleus, written 1C. Sequence-based measurements give base pairs. The conversion is

$$1\ \text{pg} \approx 0.978 \times 10^{9}\ \text{bp}$$

so a genome of 1 pg is about 978 Mb. The human 1C value of roughly 3.2 pg corresponds to about 3.1 Gb. Paris japonica at 148.9 pg is about 146 Gb.

Two cautions when comparing figures. First, 1C is the haploid amount and 2C the somatic amount for a diploid; papers are not always explicit. Second, for a polyploid the C-value counts all the chromosome sets present, so a tetraploid's 1C value is roughly double its diploid ancestor's — which is why C-value comparisons across polyploid series measure ploidy as much as anything else.

The composition argument, quantitatively

Write the genome size $G$ as a sum of components:

$$G = G_{\text{coding}} + G_{\text{intron}} + G_{\text{TE}} + G_{\text{repeat}} + G_{\text{other}}$$

Across eukaryotes, $G_{\text{coding}}$ is remarkably stable — most eukaryotes carry between roughly $10^4$ and $4\times10^4$ genes with average coding lengths around 1–2 kb, giving $G_{\text{coding}}$ in the range of tens of megabases. The variation in $G$ across five orders of magnitude therefore lives almost entirely in $G_{\text{TE}}$ and $G_{\text{repeat}}$.

The consequence is that the coding fraction

$$f_{\text{coding}} = \frac{G_{\text{coding}}}{G}$$

falls roughly as $1/G$. In yeast it is around 70 percent; in humans, 1–2 percent; in a 100 Gb salamander genome, below 0.05 percent. Any correlation between $G$ and a biological property is therefore overwhelmingly a statement about repetitive DNA, not about genes. This is the whole resolution expressed as arithmetic.

Why accumulation is the default

Model the copy number $n$ of an active transposable element family as a birth–death process with replication rate $\lambda$ per copy per generation and loss rate $\mu$ per copy per generation, where loss combines deletion and purifying selection. Then

$$\frac{dn}{dt} = (\lambda - \mu),n$$

and copy number grows exponentially whenever $\lambda > \mu$. Because $\lambda$ is a property of the element and $\mu$ depends on how effectively selection sees a slightly deleterious insertion, the balance can tip either way in different lineages with no change in the organism's biology.

Lynch's mutational hazard argument makes $\mu$ explicit. An insertion with selective disadvantage $s$ is effectively invisible to selection when

$$|s| < \frac{1}{2N_e}$$

where $N_e$ is the effective population size. Large-bodied, long-lived species have small $N_e$, so a wider band of mildly deleterious insertions drifts to fixation. This predicts that genome size should correlate inversely with $N_e$ across taxa — a prediction that holds moderately well and is the main empirical support for the mutational hazard account.

The nucleotypic correlation

Across a wide range of taxa, cell volume $V$ scales with genome size approximately as a power law,

$$V \propto G^{\beta}$$

with $\beta$ estimated near 1 in several groups, and nucleus volume scaling similarly. Because cell volume constrains surface-to-volume ratio and therefore metabolic exchange, and because a larger cell takes longer to divide, this produces downstream correlations with developmental rate and, in amphibians, with morphological simplification — salamanders with the largest genomes tend to lose digits and simplify their skulls, plausibly because their cells are too large to build fine structures.

Whether this is a cause of large genomes, a consequence of them, or a shared consequence of something else is precisely what Gregory's title asks — "Coincidence, coevolution, or causation?" — and precisely what remains open. The existence of the correlation is not in doubt; its direction is.