Darwin was vexed. While natural selection could explain gradual evolutionary transitions, the apparent sudden appearance of diverse flowering plants in the Cretaceous fossil record of approximately 130 million years ago (Mya) was his ‘abominable mystery’ ¹ . Fast-forward to today’s exciting era of high-throughput genome sequencing, and phylogenomic maps assembled from multiple whole-genome sequences tell the evolutionary story with revised timelines:

From the Carboniferous to early Cretaceous periods (approximately 360 to 130 Mya), the land was dominated by gymnosperms (literally ‘naked seeds’) including the cycads, Ginkgo, and conifers that still flourish in subarctic forests ² . The nuclear genomes of several gymnosperms have been sequenced recently, a heroic undertaking, given their exceptional size (10 to 40 gigabases) and high density of long terminal repeat (LTR)-retrotransposon repeats ^{3– 5} . Within these genomes many non-overlapping duplicated chromosomal regions were identified that display gene synteny, meaning that their gene contents are similar to those of other chromosomal blocks within the same genome and across gymnosperm genomes. These gene synteny patterns and complementary transcriptome data support the hypothesis that the gymnosperms emerged from their common ancestor via a WGD named ζ that occurred an estimated 390 Mya during the Devonian period ^{6– 9} . Tell-tale traces of further lineage-specific WGDs have also been discovered in the genomes of Norway spruce, Sequoia and Ginkgo, and in the unusual two-leaved Namibian Welwitschia, suggesting that multiple WGDs contributed to the diversity of these gymnosperms, which include the longest-living and largest organisms on Earth ^{3– 5, 7} .

Flowering plants (angiosperms, ‘seed born in a vessel’) are the most abundant plant group today, having a rich diversity of some 400,000 species from bananas to water lilies, grasses to beech trees ¹⁰ . Many angiosperm genomes have been selected for sequencing to discover genes for special agronomic traits. The resulting genome assemblies have been used collectively to extrapolate back in time, tracking the genome evolution of the monocot and eudicot angiosperms, as well as more primitive flowering plants, and converging on a reconstructed genome of the most recent common ancestor of all angiosperms ⁶ . These new phylogenomic maps solve part of Darwin’s dilemma by confirming that the first flowering plants evolved between 140 and 250 Mya after an unknown gymnosperm went through a WGD (named the ε event) an estimated 300 Mya during the Carboniferous period ^{6, 8, 9} . The clearest support for the ε WGD comes from multiple gene synteny blocks in the genome of the primitive angiosperm Amborella trichopoda, whose ancestral lineage diverged early on from other flowering plants, experiencing no further post-ε WGDs ¹¹ .

The antiquity of the ε WGD means that angiosperms had the entire Triassic and Jurassic periods to evolve and diversify into the species richness reflected in the fossils of approximately 130 Mya that were known to Darwin. Aligning with the recalibrated timelines, older angiosperm microfossils have been discovered, though claims that the earliest ones date to the Triassic are controversial ^{12, 13} . In any case, angiosperm evolution was not as sudden an explosion as Darwin thought, and now a comprehensive phylogenomic framework exists to mine for answers about how angiosperm complexity and variety evolved after the ε WGD.

In the 1970s, Susumo Ohno had the prescience to propose that evolutionary leaps could occur by WGDs because one of each gene pair may continue to do what it was doing before, giving freedom for the other to either be lost from the genome or to evolve new characteristics (neofunctionalise) ¹⁴ . Members of gene families generated via WGDs are named ohnologues in his honour ¹⁵ . Another scenario is for the duplicates to each retain subsets of the functions of the ancestral gene (subfunctionalisation) ¹⁶ . Moreover, the en masse diversification of many gene duplicates after a WGD would be expected to create selective pressures between gene families and opportunities for new interactions among them, so that large duplicated gene and protein sets co-evolve as complex systems.

Floral organs provide a canonical example of how interacting sets of diversified ohnologues can make variant structures. While many gene pairs lost one duplicate after the ε WGD, the retained ohnologue pairs include MADS-box transcription factors. Some of these were characterised more than 20 years ago for their ability to interact with each other in different combinations to specify the floral organs of Antirrhinum (snapdragon) and Arabidopsis. In these seminal studies, homeotic floral mutants were used to deduce an elegant ‘ABC model’, later elaborated into an ‘ABCDE model’ that explains how the four concentric whorls of sepals, petals, male stamens and female carpels develop in these flowers ^{17– 19} . A-function genes specify sepals; A, B and E are needed to make petals; B, C and E male stamens; C female carpels; and D for ovules. In homeotic mutants lacking B-function, for example, sepals replace petals and carpels replace stamens ¹⁸ . Most of the A, B, C, D and E functions are performed by combinations of MADS-box transcription factors that operate as homodimers and heterodimers and tetramers with different selectivities for binding to variants of a common motif in the promoters of target genes ^{19, 20} . Hence, they target overlapping but distinct sets of floral identity genes many of which are themselves ohnologues ²¹ .

The recent genome comparisons indicate that male ‘BC’ and female ‘C’ systems already existed to specify reproductive cells in gymnosperm cones, and they were duplicated via the ε WGD, after which the C duplicates diversified into angiosperm C and D genes. An A/E gymnosperm pro-orthologue gave rise to angiosperm A and E genes, and further duplicated A genes were also retained after the ε WGD ^{19, 21– 24} . These duplicated and diversified gene sets organised to generate the first now-extinct flowers, and recent reconstructions suggest that these were bisexual with petal-like tepals and pollen-bearing stamens arranged in multiple concentric whorls, and female carpels in a central spiral ²⁵ . Among living angiosperms, the Amborella lineage evolved a ‘fading borders’ programme, such that the whole flower is a spiral that gradually transitions from bracts to outer then inner tepals (specified by ABc combinations), from inner tepals to stamens (aBC) then carpels (abC), in which upper case indicates functions of greatest influence in the respective organs ²⁶ . Only in later-evolving flowers such as Arabidopsis and Antirrhinum did the tepals subdivide into sepals and petals, by restricting the boundaries of expression of floral identity genes. For example, the transcription of A and C genes became mutually exclusive ²⁶ . Further evolutionary diversity in flower form occurred by mechanisms that include shifts in the spatial expression of ABC functions across flowers, and by further WGDs that elaborated and extended the ABC regulatory network ²⁷ . For example, in stylised orchid flowers, subfunctionalisation of duplicated B genes underpins the development of three types of petals: three outer tepals, two inner tepals and a modified lip ²⁸ . Recent case studies implicate additional ancient WGDs – including one at the base of the eudicots, the γ genome triplication in the Pentapetalae (five-parted, the largest flower clade), and ρ, σ and τ polyploidisations for monocots – in evolution of the phenomenal variety of architectural form and size in pollen, fruits and seeds, in diversification of plant defence metabolites, and in the co-evolution of angiosperms with pollinators and symbiotic bacteria ^{6, 29– 34} .

Evolutionarily recent polyploidy events are also prevalent in flowering plants. Many crops including coffee, bananas, peanuts, tobacco, kiwifruit and strawberries were unwittingly selected as polyploids for their exaggerated traits such as large fruits, seeds and leaves ³⁵ . For example, the durum wheat used to make pasta is a tetraploid resulting from hybrid doubling of the genomes of two diploid wild-grass ancestors approximately 0.5 Mya, and was selected for domestication much later by Neolithic farmers, during which time hexaploid bread wheat emerged by hybridisation of the tetraploid with a diploid followed by another WGD ³⁵ . Like wheat, many well-characterised crop polyploids are allopolyploid ³⁶ , which means that the genome became polyploid after a hybrid was formed between species, in which case the WGD resolved problems with meiotic pairing by providing each chromosome with a homeologous partner ^{37, 38} . However, autopolyploidy events, self-duplication within a species, are suspected in the ancestry of potatoes, bananas, poplar and soybean ³⁹ .

Recent statistical comparisons suggest that an individual polyploid plant has a higher risk of extinction than its still-diploid relatives ^{40– 43} . It makes sense that the odds are stacked against a newly tetraploid plant. Breeding with still-diploid relatives results in triploid progeny that cannot separate evenly into two gametes during meiosis, most often resulting in sterile offspring, and unless self-pollination can occur, chances may be low of finding compatible polyploid mates. Farmers get around such problems by cloning: grafting apple trees or propagating potato tubers instead of seeds, whilst sterility (for example in seedless bananas) is sometimes even preferred ⁴⁴ . However, the resulting monocultures may be susceptible to pathogens, as when the Fusarium oxysporum fungal pandemic brought the popular triploid Gros Michel banana to the brink of extinction in the 1960s ⁴⁵ .

Nevertheless, their prevalence suggests that once polyploids have beaten the early survival odds, with or without human intervention, their polyploid traits such as larger organs, stress tolerance and altered flowering time may improve fitness or allow them to adapt to new ecological niches ⁴⁶ . Experimental polyploidies show that having extra DNA can produce an immediate phenotypic change, attributable to gene dosage effects. For example, dwarfism in apple plants with colchicine-induced autotetraploidy correlates with increased expression of a microRNA that acts via a gene regulatory network to downregulate synthesis of auxin and brassinosteroid growth regulators ⁴⁷ . More generally, polyploid plants are notable for their increased cell and organ size, which is more than a passive consequence of increased nuclear DNA content: in experiments with Arabidopsis, increases in cell volume upon tetraploidisation were found to vary in different mutants and according to cell type, indicating a genetic contribution ⁴⁸ .

For allopolyploids, the relative contributions of the species hybridisations and the WGDs to subsequent evolution are interwoven. In a recent molecular dissection of the circadian clock in allotetraploids formed between diploids Arabidopsis thaliana (At) and Arabidopsis arenosa (Aa), biases in heterologous combinations of components were discovered: the Aa-derived CCA1 hiking expedition (CHE) transcription factor preferentially binds to the promoter of the At circadian clock associated 1 ( CCA1) gene, elevating its expression over that of the AaCCA1. Such biased patterns of expression, and of protein–protein and protein–DNA interactions in the circadian regulatory network make the rhythm of the allotetraploid distinct from that of either parent ⁴⁹ .

Successful polyploidy is said to be less common in animals than in plants. Based on incidences of chromosomal anomalies in embryos that fail to develop, it appears that when two sperm fertilise one egg or when meiotic cell division fails the result is usually lethal in humans and birds ^{50, 51} . However, polyploidy is relatively common in ectothermic vertebrates. Also, synthetic fish and shellfish polyploids, generally sterile, have been created to increase food production ⁵² . Moreover, helped by technical advances in deep sequencing, genome assembly and pattern-matching software, ancient WGDs have been identified in invertebrate and vertebrate animal lineages, including mammals. The most recent discovery was that the house spider Parasteatoda tepidariorum and bark scorpion Centruroides sculpturatus are common descendants of a WGD that occurred over 450 Mya, and was distinct from an ancestral WGD of horseshoe crabs ^{53, 54} .

Although few ancient WGDs have been identified thus far in unicellular eukaryotes, the diploid baker’s yeast Saccharomyces cerevisiae and five other fungal genera all stem from the same well-characterised ancestral allopolyploid WGD approximately 100 Mya ⁵⁵ , and further fungal WGDs have been identified recently. For example, the opportunistic honeybee fungal pathogen Nosema ceranae, which is spreading to beehives worldwide, is a suspected tetraploid ⁵⁶ .

In summary, ancient WGDs that were successful in leaving modern descendants have occurred in diverse eukaryotes across eons of time, in terrestrial and aquatic environments. Remarkably, despite their radically different contexts and a sparsity of data on ancient WGDs outside of laboratory models and domesticated species ⁵⁷ , common principles are emerging that tie disparate WGD events together:

Studies in S. cerevisiae, plants and mammals have shown that regulatory proteins that form oligomers, that interact transiently with multiprotein complexes, and catalyse consecutive steps in metabolic and regulatory pathways are enriched among duplicate pairs that are retained following a WGD ^{93, 94} . These findings are interpreted by the gene balance hypothesis, which states that copy numbers of genes encoding multi-protein structures and pathways must be kept in a constant ratio to avoid architectural disruption or metabolic imbalances ⁹⁵ , although stoichiometry can also be achieved by other mechanisms such as differential degradation of protein components ⁹⁵ .

Interestingly, the architectures of sister ohnologue proteins are generally conserved with respect to content and order of their domains, at least in vertebrates. Instead, functional divergence between sisters occurs via small-scale mutations that lead to differences in temporal and spatial patterns of expression, altered sites of regulatory post-translational modifications, and changes in specificities and affinities of catalytic domains and interaction interfaces ^{88, 92, 96, 97} .

Examples abound to illustrate how the resulting families of differentiated sister ohnologues contribute to phenotypic robustness, plasticity and complexity. For instance:

These few examples—and the aforementioned ABCDE model of floral development—indicate how the parallelised signalling networks of ohnologues generated via WGDs act as multiple-input multiple-output systems. Collectively, these systems generate different cell phenotypes via differential expression and post-translational switching among sets of ohnologues with different kinetic and regulatory properties.

Deeper understanding of how post-WGD signalling networks operate will underpin advances in understanding polygenic disorders. For instance, mutations in many ohnologue genes are associated with neurological and psychiatric diseases and developmental disorders such as RASopathies ^{107, 108} , and there are many examples of heterogeneous patterns of overexpressions and mutations across ohnologue gene families in cancers ^{109– 111} . For example, overexpression of insulin receptor substrate 4 (IRS4) drives a subset of breast cancers, while IRS1 and IRS2 are not oncogenic in these cancers, even though all three IRS proteins activate PI 3-kinase–Akt growth signalling. The critical difference is that IRS4 lacks a negative feedback mechanism by which its sisters IRS1 and IRS2 can switch off the pathway via the tyrosine phosphatase SHP2 ¹⁰⁹ . This example illustrates how the parallel signalling pathways generated via WGDs can evolve specific regulatory interconnections, which in principle enable these systems to integrate inputs from multiple sensory stimuli, buffer signal noise via responsive feedback loops, and generate a wider repertoire of phenotypic outcomes than would be possible with the original simple circuit ^{112, 113} .

Finally, it should be noted that WGDs do not underpin every evolutionary leap: A WGD was hypothesised to explain why cephalopods (squids, cuttlefish and octopuses) are behaviourally more sophisticated than other molluscs. However, the octopus genome shows no evidence of a WGD. Rather, sensory intelligence in cephalopods is likely to be underpinned by the massively expanded gene families of C2H2 zinc-finger transcription factors and protocadherins that are expressed in their neuronal and sensitive tissues ¹¹⁴ . It will be fascinating to compare the neural network topologies that have been built from gene families generated via WGDs versus those made of multiple small-scale duplications in humans and octopuses.