Systematics

Linnaeus sorted flowering plants by counting stamens and pistils. He knew this was artificial, and said so; it was a key for finding things, not a claim about nature. The trouble is that a key built on one character scatters obvious relatives. Michel Adanson and then Antoine-Laurent de Jussieu insisted that characters be weighed in bulk. Jussieu's Genera Plantarum of 1789 arranged a hundred families by overall agreement across many features, and most of those families are still recognised today, which is a remarkable survival rate for a pre-evolutionary book.

But why should agreement in many characters be trustworthy? Jussieu had no answer; he had an intuition that worked. The answer arrived in 1859, when Darwin's argument gave the nested hierarchy a cause. Groups within groups is what descent with modification produces, and characters agree in bulk because they were inherited together. Classification stopped being a filing system and became a claim about history — a claim that could be right or wrong.

Making the hierarchy genealogical did not settle its base. Ernst Mayr, cataloguing the birds of New Guinea and the Solomon Islands, found populations on neighbouring islands that were distinguishable, interbred where they met, and in some cases did not. In Systematics and the Origin of Species (1942) he defined a species as a group of populations that interbreed with each other and are reproductively isolated from others, and argued that species usually form when geography separates populations long enough for incompatibility to accumulate. Species became a fact about gene flow.

This is the definition most people learn, and it has a known list of failures. It says nothing about the asexual — a dandelion lineage, an amoeba, a bacterium. It cannot be applied to fossils, which do not breed for inspection. It strains at ring species, where neighbouring populations interbreed all the way round a mountain range except where the ends meet, and at the oaks, which hybridise constantly and remain recognisable. The response has not been one replacement but dozens of concepts in parallel, and by the 1960s the question of which to use had become a question about method.

Robert Sokal and Peter Sneath answered it in 1963 by trying to remove judgement entirely: score hundreds of characters, compute a similarity coefficient for every pair, cluster. Phenetics was the first computational taxonomy, and it made everyone state their procedures explicitly. It also drew the obvious objection — that overall similarity confuses inheritance with convergence, so that a crocodile comes out closer to a lizard than to a bird — and in answering it, phylogenetics took the subject over.

Linnaeus named roughly 7,700 plant and 4,400 animal species in his lifetime. The Catalogue of Life now holds about 2.1 million accepted species names. Describing a species means publishing a diagnosis, designating a type specimen and placing it in a genus, and the world's taxonomists do this about 18,000 times a year.

Take 8.7 million as the number of eukaryotic species, the central estimate of Mora and colleagues in 2011. The work remaining is

$$8{,}700{,}000 - 2{,}100{,}000 = 6{,}600{,}000 \text{ species},$$

and at the current rate that takes

$$\frac{6{,}600{,}000}{18{,}000 \text{ per year}} \approx 370 \text{ years}.$$

Mora's own version of this calculation, which allowed for the cost of specimens, travel and publication, put it at about 1,200 years of work by 300,000 taxonomists and roughly $364 billion. Either figure makes the same point: on present methods the catalogue will not be finished.

Now change the estimate. If the eukaryotic total is 2 million, as some analyses of the slowing discovery of higher taxa suggest, the work is nearly done. If it is 100 million, dominated by insect-associated fungi and nematodes, the answer is

$$\frac{100{,}000{,}000 - 2{,}100{,}000}{18{,}000} \approx 5{,}400 \text{ years},$$

and the project as conceived is hopeless. The three answers — nearly finished, four centuries, never — differ by a factor of fifteen, and the uncertainty is not in the arithmetic. It is in the extrapolation from the groups that have been sampled to the groups that have not.

The microbes are not in these numbers at all. Estimates of bacterial and archaeal diversity run to $10^{12}$ operational units, counted by sequence similarity rather than by breeding, which is an admission that for most of life the Linnaean species is not the unit being counted. That is the practical form of the species problem: not a philosophical puzzle, but the reason the denominator of every biodiversity statistic is unknown to within a factor of ten.

By the late 1960s systematics had three parties. The pheneticists clustered by overall similarity. The evolutionary taxonomists, Mayr and George Gaylord Simpson among them, held that a classification should reflect both ancestry and the amount of subsequent change, which is why they kept Reptilia as a group and birds outside it. The cladists held that only ancestry counts, so that a group must contain all the descendants of its ancestor or it is not a group at all — which makes birds reptiles and leaves Reptilia, as traditionally used, nonexistent.

The quarrel was fierce, personal and productive, and the cladists won it. What followed was not the end of systematics but its conversion into an inferential science: given characters, find the tree. That is the subject of phylogenetics. The fossils that the biological species concept could not reach became the business of palaeontology, which had been assembling its own deep hierarchy from stone since the 1790s.