Density-Dependent Cladogenesis in Birds

Evolutionary biologists kept finding the same pattern in bird phylogenies: lineages diversify quickly at first and then slow down. The standard interpretation was adaptive radiation — ecological niches fill up, geographic space gets claimed, and the brakes come on. It was a compelling, intuitive story. However, Phillimore and Price showed that this pattern can appear even when there is nothing biological slowing it down at all. The slowdown might be a statistical ghost. So how do you tell the real signal from the illusion? The intuitive story starts with opportunity. When a bird lineage first encounters unexploited ecological niches and open geographic territory, species formation accelerates. Ernst Mayr noted that speciation becomes harder in a crowded environment because populations struggle to persist in new locations — and persistence is what gives populations time to diverge into full species. Schluter and others added that divergent selection is stronger when niche space is wide open, so reproductive isolation evolves faster early in a radiation than later. The fossil record shows episodic pulses of morphological change that fit this picture. But morphological change and speciation do not always track each other, so the cleanest way to assess how speciation rate changes through time is with molecular phylogenies calibrated to geological time. The standard visualization is a lineage-through-time plot. Verbally, you graph the logarithm of the number of lineages against time. Under the simplest model — each lineage has a constant probability of splitting, with no extinction — that graph is a straight line. What researchers kept finding instead was a curve that flattens toward the present. Fewer lineages accumulate near the present than expected. That flattening is the slowdown, and for decades it was read as the hallmark of adaptive radiation: a burst of early diversification followed by ecological saturation putting the brakes on. Here is where Phillimore and Price introduce the twist. They ran simulations under a constant-rate pure-birth model — a birth rate of zero point two, ten thousand replicate trees, with no ecological braking whatsoever. Those simulations produced a consistent negative correlation between clade size and the slowdown statistic. Large clades showed stronger slowdowns than small ones, even though the underlying speciation probability never changed. The mechanism is statistical, and once you see it, it's hard to unsee. Exponential growth amplifies early chance differences. A lineage that, purely by luck, produces more speciation events than average early on will become much larger over time. But as the process continues, its realized rate will tend to drift back toward the overall average — that's regression to the mean, the statistical force that pulls outliers back toward the center. That drift shows up in the phylogeny as a decline in diversification rate toward the present. Meanwhile, researchers have historically focused their attention on large, speciose clades — precisely the ones most likely to have experienced this statistical artifact. That sampling bias inflates the apparent frequency of slowdowns even when underlying rates are constant. The tool Phillimore and Price use to measure this is the c statistic — sometimes called the gamma statistic. It measures how internode times are distributed from root to tip. Under a constant-rate pure-birth model, c follows a standard normal distribution, with a mean of zero and standard deviation of one. A negative c means nodes are concentrated toward the root — early branching, slower recent accumulation. A one-tailed test rejects the constant-rate model when c falls below negative one point six four five at the five percent significance level. That's the threshold the paper uses throughout. So the simulations establish a baseline expectation: some slowdowns will appear by chance, and large clades will show more of them. The critical question is then empirical — does the bird data exceed even that elevated baseline? To answer this, Phillimore and Price assembled a meta-analysis of sequence-based molecular phylogenies for forty-five bird clades, representing roughly one thousand three hundred fifty species. Rather than studying any single radiation in depth, they examined the distribution of c values across all forty-five clades. Phylogenies were reconstructed under a relaxed-clock Bayesian framework, and c was calculated across posterior tree samples, with the median used as each clade's summary value. Lineage counts were restricted to bifurcations dated prior to two million years ago, to avoid conflating recent population structure with genuine speciation. The focus on large clades is deliberate. Small clades simply lack the statistical power to detect realistic slowdowns — a clade needs roughly fifteen or more lineages before c becomes reliably sensitive. So Phillimore and Price concentrated their inference on the twenty-three clades with more than twenty species. Among those, thirteen — that's fifty-seven percent — showed significant slowdowns. Across all forty-five clades, the mean c was negative zero point ninety-eight, plus or minus zero point twenty, producing a t-statistic of four point eighty-nine and a p-value below zero point zero zero one. Among the twenty-two clades with fifteen or more lineages at two million years ago, the median c was negative one point seventy-seven and the median clade size was twenty-nine. The paper interprets that magnitude as consistent with a decline in speciation rates to roughly ten to fifty percent of the initial rate by the later stages of a radiation. Now comes the key comparison. The simulations said: expect some slowdowns in large clades from pure chance. The empirical data said: fifty-seven percent of large clades show significant slowdowns. Are those compatible? Phillimore and Price tested this directly by running ten thousand constant-rate simulations across a range of birth and death parameter values — birth rate zero point two, death rates at zero, zero point zero five, and zero point two — and comparing the observed frequency of significant slowdowns in large clades to the null distribution using chi-square goodness-of-fit tests. Across every parameter combination they examined, the observed excess of slowdowns was unlikely under the constant-rate model, with p-values below zero point zero one. They also ruled out the obvious alternative explanations. Phylogenetic saturation — where substitution models fail to correct for multiple mutations on long branches, artificially compressing early nodes — is unlikely here because trees were reconstructed with a generalized time-reversible model plus invariant sites plus gamma model, and if saturation were driving things, older clades should show stronger slowdowns. They don't. Gene tree error should make slowdowns harder to detect, not easier — so it cannot explain a spurious excess. Tree imbalance, where a few long-lived unproductive lineages distort the shape of the tree, explains only eight percent of the variance in c across the forty-five clades. Not enough to carry the pattern. Then there's extinction, and this is a subtle but important point. Extinction in reconstructed phylogenies tends to make recent branching look more frequent than it actually was — because extinct lineages leave no trace, the survivors are disproportionately recent. That bias pushes c toward zero or positive values, making slowdowns harder to detect. This means the slowdowns Phillimore and Price observed in bird clades are conservative — the true signal is likely stronger than the data show. If extinction rates over the past twenty million years had approached speciation rates, the probability of observing the frequency of slowdowns they found under a constant-rate birth-death model becomes vanishingly small. What density-dependent speciation means, in plain terms, is this: as more lineages fill the available ecological and geographic space within a clade, the per-lineage probability of producing a new species declines. Ecology and geography act as a ceiling on diversification. The pattern Phillimore and Price documented across forty-five bird clades — consistent, statistically robust, and more extreme than any constant-rate simulation predicts — suggests this ceiling is not a feature of any single spectacular radiation. It appears to be a general property of how birds diversify. The methodological lesson is just as important as the biological one. A pattern that looks like an ecological signal can be a statistical artifact of exponential growth and regression to the mean. The only way to separate them is to simulate the null expectation and ask whether the data exceed it. Phillimore and Price built exactly that test, applied it across forty-five clades, and found that the bird data cannot be explained by chance alone. Ecology really is putting the brakes on speciation — just not in every clade and not always as strongly as earlier single-clade studies implied. The ghost was real. But so was the signal hiding behind it. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.