Phylogenomic approaches untangle early divergences and complex diversifications of the olive plant family

An olive tree. Not as food, not as symbol — as a phylogenetic puzzle. The olive family contains twenty-five genera and roughly six hundred species: ash trees, jasmine, forsythia, and lilac. For decades, scientists couldn't agree on how these plants were related to each other. Not for lack of trying — previous analyses produced six competing topologies for just the five major tribes. The signals kept contradicting each other. Dong and colleagues decided to find out why, and the answer turns out to be written into the genome of every living member of the family. The olive family, Oleaceae, is organized into five tribes: Myxopyreae, Fontanesieae, Forsythieae, Jasmineae, and Oleeae — that last one containing the olive itself, along with ash and lilac. The trouble with building a family tree for this group runs deeper than data quality. Two biological processes produce the same confusing pattern. The first is incomplete lineage sorting, or ILS. When two species split from a common ancestor, not all of their ancestral genetic variation gets cleanly sorted. Some polymorphisms persist through multiple speciation events and get fixed differently in different descendant lineages — by chance, not by ancestry. The result is that individual genes tell different evolutionary stories, none of them wrong, none of them the whole truth. The second process is ancient introgression: post-speciation gene flow, where lineages that already diverged come back into contact, hybridize, and exchange genetic material. Introgression also scrambles individual gene histories. The problem is that both processes leave the same fingerprint — gene trees that disagree with the species tree. Distinguishing them is hard under the best circumstances, and Dong and colleagues note that Oleaceae began diversifying around sixty-point-five million years ago, with five major lineages separating within roughly eight million years. That's fast. Fast radiations are exactly where incomplete lineage sorting and introgression are hardest to disentangle. To tackle the problem, Dong and colleagues assembled three complementary data types. First, whole plastid genomes — the circular chloroplast chromosome, maternally inherited, essentially a single non-recombining locus. They analyzed one hundred eighty plastid genomes representing one hundred forty species across all five tribes and twenty-four genera, aligning seventy-seven protein-coding genes into a matrix of fifty-five thousand two hundred ninety-six base pairs containing over ten thousand variable sites. Second, genome-wide nuclear single nucleotide polymorphisms — called SNPs — obtained by mapping genome-skim reads to three reference genomes, including olive and ash. Those matrices ranged from ninety-one thousand to over four hundred thousand SNPs, sampling many independent loci across the nuclear genome. Third, and most powerful for resolving gene-level histories: large sets of single-copy nuclear orthologs — two thousand six hundred eight orthologous genes for inter-tribal comparisons, and one thousand eight hundred sixty-five for within-tribe relationships among the Oleeae subtribes. Each gene tree was inferred independently, then the team compared those thousands of trees using both concatenation and coalescent methods, and quantified conflict using concordance factors. That's when the extent of the problem became clear. Substitution rates — the pace at which DNA changes accumulate — vary dramatically across the family. Dong and colleagues ran branch-model tests and found they had to reject a single global molecular clock. Jasmineae branches evolved about five point fifty-eight times faster than the background rate. Adding a second local clock for the Oleeae subtribe Ligustrinae improved the fit further, with rates of six point ninety-eight and two point twenty-nine times higher, respectively. When different lineages evolve at such different speeds, trees inferred under wrong-rate assumptions get distorted — which is exactly where much of the prior confusion originated. The gene-tree discordance was just as striking. For the node uniting three of the four Oleeae subtribes, the gene concordance factor — the fraction of individual gene trees that actually support that grouping — was only thirty-nine point fifty-seven out of one hundred. The site concordance factor was forty-nine point twenty-nine. Barely half the data agreed. Across one thousand eight hundred sixty-five gene trees for the Oleeae subtribes, all fifteen possible rooted topologies occurred. The single most common topology matched the species tree in only thirty percent of gene trees, with the next two most frequent accounting for eighteen and seventeen percent respectively. The signal was real, but it was fighting massive noise. The plastid and nuclear trees also conflicted — the organellar genome and the nuclear genome were telling different stories at multiple scales. Now came the key question: was that discordance caused by incomplete lineage sorting, or by actual gene flow between lineages? Dong and colleagues ran three independent tests to find out. The first was the D-statistic, also called the ABBA-BABA test. The logic is elegant: if two species are related by a clean tree, you'd expect certain site patterns to appear in equal numbers. When one pattern consistently outnumbers the other, the asymmetry signals gene flow. Using a Z-score threshold above three — corresponding to a p-value below zero point zero zero two — significant D-statistics appeared in all tested triplets, with mean absolute values ranging from zero point zero nine to zero point four one. Six of ten pairwise species comparisons were significant, and the test identified introgression signals at seven locations among the tribes. But the D-statistic has a known blind spot: it can miss complex or multiple admixture events. So the team also used QuIBL, a method that examines the distribution of internal branch lengths in gene trees. Under pure incomplete lineage sorting, those lengths follow one statistical pattern; under incomplete lineage sorting plus introgression, they follow another. QuIBL compared models using the Bayesian information criterion, and twenty-six of thirty triplets met the stringent cutoff — a difference in that criterion below negative ten — supporting the incomplete lineage sorting plus introgression model. QuIBL detected introgression at all possible locations among the five tribes. Where the D-statistic missed a signal between Oleeae and Jasmineae, QuIBL found it. Third, a tool called PhyloNet modeled reticulate evolution explicitly, allowing the researchers to infer network topologies with one, two, or three hybridization events. All three scenarios converged on the same answer: Oleeae appeared as a hybrid lineage, consistent with what the D-statistic and QuIBL had found. The conclusion was clear. Ancient introgression — not incomplete lineage sorting — is the primary driver of discordance among the five tribes of Oleaceae. Within the Oleeae subtribes, however, the story flips: there, incomplete lineage sorting dominates, with post-speciation gene flow very limited. The answer is context-dependent, and scale matters. That brings us to the most remarkable finding in the paper. Tribe Oleeae — the olive's own clade — appears to have originated through ancient hybridization combined with polyploidy, an allopolyploidization event dated to approximately fifty-two point five million years ago. The likely parents were an ancestor related to Forsythieae and a lineage related to Jasmineae — or possibly a ghost lineage, an extinct ancestor of Jasmineae that left no living descendants except its genomic signature inside Oleeae. The evidence converges from multiple directions. Plastid trees consistently placed Jasmineae as sister to Oleeae, consistent with a Jasmineae-like maternal contribution. PhyloNet placed Forsythieae as the paternal parent, with inheritance probabilities of zero point seventy-six, zero point seventy-three, and zero point seventy-three across the one, two, and three reticulation scenarios. Then genome synteny analysis added independent corroboration: olive transcripts mapped to the Forsythia suspensa genome in far greater numbers — thirty-four thousand five hundred forty-two transcripts — than to the Jasminum sambac genome, where only twenty thousand forty mapped. Synteny block counts told the same story: olive shared three hundred three blocks with Forsythia but only one hundred seventy-three with Jasminum. The nuclear genome looks more like Forsythia. The plastid looks more like Jasmineae. That asymmetry — the organellar genome pointing one direction, the nuclear genome pointing another — is exactly what you expect from an ancient allopolyploid. Whether the maternal parent was Jasmineae itself or a ghost lineage that preceded it remains unresolved. The ghost lineage hypothesis isn't speculation for its own sake; it's the most parsimonious explanation for why the plastid and nuclear signals diverge in the way they do. After the allopolyploidization, Oleeae radiated rapidly. The four subtribes diverged between roughly forty-seven and thirty-nine million years ago, with crown ages for individual subtribes ranging from twenty-two to thirty-four million years. By then, the difficult phylogenetic signal within Oleeae was being generated by incomplete lineage sorting during a fast radiation — a different mechanism operating at a different timescale. The broader takeaway is that the olive family's tangled history isn't a failure of genomic analysis. It's a record of real events: ancient hybridizations, ghost lineages, rapid radiations, and rate accelerations that left overlapping marks in the genome. Phylogenomics at this scale — thousands of genes, multiple independent tests, genome synteny — can finally read those marks separately and name what caused them. 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.