Prokaryotic evolution and the tree of life are two different things

Darwin's single illustration in "On the Origin of Species" is a branching tree. Hand-sketched, it is the only diagram in the entire book. That image has governed how biologists think about life's relationships for over 160 years. Now, a consortium of eleven researchers — including philosophers, geneticists, and evolutionary biologists — argues that it simply does not apply to most life on Earth. Their argument is not that Darwin was wrong about evolution. Rather, it is that the tree model, which works beautifully for multicellular animals and plants, has been quietly extrapolated to prokaryotes, which include bacteria and archaea, without considering whether the underlying mechanisms actually support it. Bapteste and colleagues, writing in "Biology Direct," assert that prokaryotic evolution and the tree of life are genuinely different phenomena, and that treating them as one has become more a matter of tradition than evidence. Here is why the tree works for us. In eukaryotes — organisms with nucleated cells, including everything from fungi to vertebrates — sex and meiosis shuffle genes mostly within species boundaries. Lineages diverge and remain separate. Over geological time, that results in nested splits: a branching pattern you can truly represent as a tree. The key phrase here is "remain separate." Once two lineages split, their genes continue independently. The history takes a tree shape because the biology takes a tree shape. In prokaryotes, the biology is different in kind. Bacteria acquire genes sideways — through transformation, transduction, conjugation, and gene transfer agents — not just downward from parent to offspring. This is known as lateral gene transfer, or LGT, and the genomic data reveal that it is not a rare occurrence. It is pervasive, and its scale is staggering. Three individuals of the same nominal Escherichia coli species may share only about 40 percent of their combined gene set. Vibrio strains that appear identical at standard housekeeping loci can differ by up to 800 kilobases — roughly 20 percent of the genome. Frankia strains with more than 97 percent ribosomal RNA identity can differ by as many as 3,500 genes, representing up to 77 percent of the smaller genome. Pathogenicity islands acquired through lateral transfer can represent more than 30 percent of a bacterial genome. These instances are not edge cases; they are the norm. The consequence for phylogenetics — the reconstruction of evolutionary relationships — is direct. When you build a family tree using one gene, you get one answer. When you build it using a different gene from the same organisms, you often get a completely different answer. Shi and Falkowski constructed phylogenetic trees for 682 orthologous protein families from 13 cyanobacterial genomes and found no predominant unanimous topology across all those trees. The single topology shared by the largest number of gene families accounted for only about 2 percent of the datasets. That is not a signal; it is noise. Worse, when Shi and Falkowski tested five candidate species-tree topologies, between 97.5 and 99.6 percent of individual genes failed to statistically discriminate among those five very different trees. In other words, the data could not distinguish between the competing hypotheses. This does not support a consensus tree; rather, it reflects an absence of signal, dressed up as agreement. This is where the paper's epistemological argument hits hardest, and it is worth slowing down for. Many methods used to test whether a tree of life exists already assume one. The standard approach is to concatenate genes — stack them together — build a single tree from the combined data, and then check whether individual genes are compatible with it. But the null hypothesis in these congruence tests is that all genes share the same underlying tree. Failing to reject that null is often treated as support for the tree. As Bapteste and colleagues note, in statistics, failing to reject a null hypothesis does not mean confirming it. Genes with weak phylogenetic signal will almost always fail to reject the consensus tree simply because they lack the power to reject anything. It's as if the courtroom decided on a verdict before hearing the evidence. Data curation tightens the trap. Soria-Carrasco and Castresana progressively filtered their proteobacterial dataset — excluding multi-copy genes, requiring ubiquitous core genes, and trimming alignment length — and still found that 46.5 percent of individual proteobacterial genes were incompatible with the concatenated species tree, compared to 23.4 percent for eukaryotes. After restricting to markers longer than 300 amino acids, only 20 proteobacterial genes survived the curation, versus 88 eukaryotic genes. That represents roughly 0.8 percent of a bacterial genome. The tree of life for prokaryotes is being reconstructed from less than one percent of their genetic information, after filtering that information specifically to make it tree-like. Cicarelli and colleagues' earlier effort to find universally shared, never-transferred genes yielded only 31 genes — about 1 percent of a typical prokaryotic genome. Aggregate confidence metrics do not solve the issue. Bootstrap values and posterior probabilities can strongly support incorrect splits when the underlying evolutionary model is violated. In one supertree analysis of 120 input trees, the Aquifex–Thermotoga pairing was favored by the majority-rule method, even though only 20 of the 120 source trees actually supported it. The arithmetic of aggregation overwhelmed the biology. Ford Doolittle, one of the most respected evolutionary biologists in the field, reviewed this paper and offered a genuine counterproposal rather than a dismissal. He agrees that the universal tree is untenable for prokaryotes as conventionally stated. However, he argues for what he calls the Tree of Cells — the TOC — which traces not gene histories but cell-division events themselves. Even if genes move laterally, the cells that carry them have a tree-like history of fissions and speciations. Doolittle's point is that it remains meaningful and true to say that E. coli K12 and O157:H7 are more closely related to each other than either is to Yersinia pestis, and that a tree showing this relationship "represents some sort of useful historical truth about diverging populations." His closing line in the review is blunt: "The prokaryotic tree of life is dead!" But what he means is the universal, gene-based tree of life, not the cellular lineage record. Doolittle's position is importantly different from the paper's. He wants to recover the tree of vertical inheritance and use it as a framework for annotating lateral transfers. He explicitly says he wants the right to "paint in red in the net of life the bifurcating subgraph which traces back vertical inheritance." That is a pluralist position — trees and networks as complementary tools — but it still centers the tree as a reference. Bapteste and colleagues push further: no single representation should dominate, because different questions about prokaryotic evolution require genuinely different models. The paper proposes four concrete alternatives. The first is a Central Trend, which is a similarity-based backbone that may be phenetic rather than genealogical, capturing the average signal across genes without claiming it reflects common ancestry throughout. The second is the Banyan Tree — a familiar tree shape for conserved information-processing genes, with lateral transfers treated as essential additions rather than noise. The third is an explicit Network: a graph in which tree-like and reticulate phases of evolution are both represented, capturing chromosome dynamics that trees simply cannot illustrate. The fourth is Pattern Pluralism — a phylogenetic forest, reconstructing as many trees and networks as different genes, regions, taxa, or questions require, with no single image declared authoritative. The Thermotoga case illustrates why pluralism is not evasion. Ribosomal components of Thermotogales group with Aquificae. Whole-genome analyses often group them with Clostridia and Bacilli. About 8 percent of Thermotoga genes have their closest homologs in archaea. The same organisms form a single recombination-based species but multiple ecotype-based species. There is no single correct picture here — there are multiple legitimate evolutionary histories layered in the same genome. Bapteste and colleagues clarify that the tree of life was not a failure of imagination. Before genome-scale data, it was the most scientifically productive model available. The problem is that the belief in a universal prokaryotic tree has grown stronger than the evidence supporting it. The plurality of models — not a new universal replacement — is what the data now demand. Prokaryotes represent most of the diversity of life and most of its evolutionary history. Getting their story right means allowing their biology, not our preferred diagram, to determine the shape of the explanation. 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.