Plant pangenomes for crop improvement, biodiversity and evolution

Abstract

Plant genome sequences catalogue genes and the genetic elements that regulate their expression. Such inventories further research aims as diverse as mapping the molecular basis of trait diversity in domesticated plants or inquiries into the origin of evolutionary innovations in flowering plants millions of years ago. The transformative technological progress of DNA sequencing in the past two decades has enabled researchers to sequence ever more genomes with greater ease. Pangenomes — complete sequences of multiple individuals of a species or higher taxonomic unit — have now entered the geneticists’ toolkit. The genomes of crop plants and their wild relatives are being studied with translational applications in breeding in mind. But pangenomes are applicable also in ecological and evolutionary studies, as they help classify and monitor biodiversity across the tree of life, deepen our understanding of how plant species diverged, and show how plants adapt to changing environments or new selection pressures exerted by human beings.

Introduction

Genomes comprise the entirety of genic and non-genic sequences of an organism. Comparisons between the genome sequences of different individuals of the same species have revealed a high extent of intraspecies variation, which ranges from single nucleotide changes and small insertions or deletions (indels) to large-scale structural variation; some individuals lack entire genes that are present in others, or the linear order of genetic elements can differ between members of the same species¹. By creating genome sequence assemblies for a representative set of individuals from a species of interest, researchers can catalogue and characterize the genetic diversity within and between species; decipher the role of structural changes in evolutionary processes, such as speciation, adaptation, domestication or polyploidization; and investigate the genetic basis of phenotypic variation.

Pangenomics (pan, from the Greek word meaning whole) seeks to capture the full spectrum of genetic variation within a species through the assembly and comparative analysis of genome sequences from multiple individuals. In bacterial genomics, where the term originated, the pangenome is often defined as the full complement of genes present in the members of a species², with the core genome consisting of genes that are present in all or nearly all individuals and the accessory genome including genes that are variably present among different individuals or strains. In eukaryotes such as animals and plants, allowance must be made for non-coding sequences, most of which are derived from repetitive elements. Although ‘pangenome’ in an all-encompassing sense is defined as the full genomic content of a species, terms such as “pangenome reference”³ or simply “pangenome”⁴ often denote a collection of genome sequences belonging to one species or higher taxonomic group and the computational and logistical infrastructure⁵ appertaining to it (Figure 1).

Figure 1. Pangenomics: assembly and comparison of genome sequences.

(a) Sequence reads, these days mostly long (>15 kb) reads, are assembled into contigs, which are arranged into chromosome-scale scaffolds (pseudomolecules) with the help of genetic and physical linkage information (dashed lines). (b) Comparisons of sequence assemblies reveal the full spectrum of sequence variation in the assembled genomes. (c) Pangenome graphs are computational representations of the assemblies and the differences between them. In this example, colour bands represent genomes as paths through the pangenome graph. Graphs with single base pair resolution are still challenging to construct at the whole-genome level. (d) A gene-centric view reduces complexity as do (e) pairwise alignments of genome sequences. (f) Short-read data (red bars), which is used for population-scale resequencing, can be integrated with pangenomes, for example, by aligning them to pangenome graphs.

Genome sequence assemblies are central to pangenomics. The first genome sequence assemblies were ‘drafts’, split into thousands of sequence contigs and often without a linear order⁶. As genome sequencing used to be a costly undertaking, its applications long relied on a single ‘high-quality’ genome sequence per species that was at chromosome scale, contiguous and often constructed with much effort. Many more genomes were only re-sequenced with comparatively cheap short-read sequencing technologies, with short reads then aligned to one or more reference genomes for variant detection^7–9. But short reads cannot accurately represent repetitive sequences, which abound especially in plant genomes, and fail to resolve balanced structural variants (SVs), such as inversions or inter-chromosomal translocations^10–12 (Figure 1). Advances in DNA sequencing technologies, in particular the development of long-read sequencing, has made genome assembly easier and faster and its products more contiguous and complete^13,14,15,16. Gap-free sequences have been generated for the chromosomes of several eukaryotes, from plants¹¹ to humans^17,18. Piecing together such telomere-to-telomere assemblies remains challenging, owing in large part to the presence of homogeneous repeats^19,20. But the routinely applicable techniques that underpin primary contig assembly and subsequent scaffolding yield in silico representations of chromosomes that, despite their occasional lack of completeness or positional assignment, are sufficiently informative to extract most biological information²¹.

In agriculture, pangenomes of crop species promise to advance crop improvement^22–24 by identifying genetic variation underlying the expression of desirable genetic traits, which in turn might help breed improved crop varieties. Over the past decade, crop pangenomes have been constructed for cereals, legumes, vegetables, fruit trees and tuber-bearing crops (Table 1). Although a success story going full circle from the identification of a SV to creating a new crop variety has yet to emerge, crop pangenomes have shown the impact of structural variation on crop evolution. Recently, the application of pangenomes has moved beyond domesticated plants to their wild progenitors and more distant relatives, highlighting their utility for ecological and evolutionary studies. Pangenomes are on their way to replace short-read reference genomes as the preferred inventories of sequence variation, with tree-of-life projects, such as the Earth BioGenome Project, the Darwin Tree of Life project and the 10,000 Plant Genome Project (10KP), aiming to sequence and compare as many genomes as possible to help capture and preserve biodiversity.

Table 1. Plant pangenome studies.

Species	Common name	Genome size (Mbp)	n	Sequencing technology	Year	Ref
Glycine soja	Wild soybean	1,000	7	Illumina	2014	28
Medicago truncatula	Barrel medic	400	15	Illumina	2017	183
Brachypodium distachyon	Purple false brome	250	54	Illumina	2017	184
Oryza spp.	Rice wild relatives	400-500	13	Illumina	2018	82
Oryza sativa and O. rufipogon	Asian and common wild rice	400	66	Illumina	2018	185
Solanum lycopersicum	Tomato	950	72 5	Illumina	2019	4
Solanum lycopersicum	Tomato	950	10 0	Oxford Nanopore	2020	152
Oryza sativa	Asian rice	400	12	PacBio	2020	186
Brassica napus	Rapeseed	1,100	8	PacBio	2020	187
Hordeum vulgare	Barley	5,000	20	Illumina	2020	40
Glycine max and G. soja	Soybean	1,100	29	PacBio	2020	188
Arabidopsis thaliana	Thale cress	135	8	PacBio	2020	189
Triticum aestivum	Bread wheat	15,000	15	Illumina	2020	39
Oryza sativa	Rice	400	33	PacBio	2021	190
Zea mays	Maize	2,100	26	PacBio	2021	49
Sorghum bicolor	Sorghum	800	16	PacBio	2021	191
Raphanus spp.	Radish	500	11	PacBio	2021	192
Cucumis sativus	Cucumber	350	12	PacBio	2022	193
Solanum lycopersicum	Tomato	950	32	PacBio (HiFi)	2022	60
Solanum spp.	Potato wild relatives	800	44	PacBio (HiFi)	2022	83
Solanum tuberosum	Potato	800	6	Illumina Oxford Nanopore	2022	194
Glycine spp.	Soybean wild relatives	1,100	26	PacBio	2022	195
Gossypium spp.	Cotton	750–2,500	7	Oxford Nanopore	2022	196
Vigna unguiculata	Cowpea	650	7	PacBio	2023	197
Arabidopsis thaliana	Thale cress	135	38	PacBio (HiFi)	2022	198
Pennisetum glaucum	Pearl millet	1,700	10	PacBio (HiFi)	2023	199
Zea mays	Maize	2,100	26	PacBio	2023	200
Citrus spp.	Orange	217-419	12	PacBio Oxford Nanopore	2023	201
Solanum spp.	Tomato and relatives	770-1,200	13	PacBio	2023	43
Setaria italica	Foxtail millet	430	11 0	PacBio	2023	202
Capsicum spp.	Chili peppers	3000–4,100	11	PacBio (HiFi)	2023	203
Arabidopsis thaliana	Thale cress	135	72	PacBio (HiFi) Oxford Nanopore	2023	204

In this Review, we take stock of the progress of plant pangenomics over the past decade. We first provide an overview of the application of pangenomes to crop plants, before discussing the role pangenomes can play in the conservation of biodiversity and how they further evolutionary and biodiversity research. Finally, we outline future developments for the field.

Applications of pangenomes in crop plants

The publication of the first genome sequence of a crop, that of rice in 2005²⁵, ushered in a step-change in the speed of genetic research in that species²⁶, largely owing to easier and faster gene mapping²⁷. As we move beyond single reference genomes, complete genome sequences for multiple individuals are used to catalogue sequence diversity in crop plants. One of the first reports that can lay claim to being a plant pangenome paper was published in 2014²⁸. In that study, Li et al. assembled draft genome sequences of seven accessions of Glycine soja, the wild progenitor of domesticated soy bean. Plant genetic resources²⁹, which comprise traditional landraces and wild progenitors, are potential sources of beneficial genes and alleles, for example, those conferring resistance against disease, that are absent from modern elite varieties. Pangenomes can assist in realizing this potential by more effectively linking sequence variation to phenotypes deployed in breeding programmes. In this section, we describe three applications of pangenomes in crop plant genetics: mapping or selecting for beneficial alleles; the generation of inventories of resistance genes (also known as R genes); and the study of crop-wild relatives (Figure 2).

Figure 2. Pangenomics in crop plants.

Most pangenome studies to date have focused on crops. The varieties under investigation are selected based on different criteria.(a) Cultivars of great ‘importance’ include those that are widely grown or used in genetic research. (b) Surveys of population structure enable the selection of core sets that represent with a limited number of samples genetic diversity in a given crop as best as possible. The diversity space of species is often represented in principal component analysis (PCA). Population structure is reflected in clusters (shown in different colours) that correspond to geographic origins or infraspecific taxonomy. (c) Crop-wild relatives (wild progenitors and more distant relatives) are studied because they broaden allelic diversity in cultivated varieties. (d) Pangenomes have diverse applications in crop genetics. Genome sequences of the parents of experimental population assist in mapping traits to single genetic factors (coloured bar). (d). (e) Catalogues of resistance genes enrich the toolkit of plant pathology and may be represented in matrices that record the presence (blue square) or absence (grey square) of genes in the sequenced individuals. (f) Thanks to genome sequences, geneticists can include structural variants in their search for causal polymorphisms under GWAS peaks.

Genetic mapping and selection of variants associated with desired traits

Genetic mapping refers to the process of identifying and understanding the genetic basis of specific traits within a population, often with the aim of improving those traits through selective breeding. In crop genetics, mapping employs molecular markers — predominantly based on single-nucleotide polymorphism (SNPs) or indels — to establish causal relationships between discrete genetic elements and variation in breeding-relevant phenotypes such as seed traits, yield or disease resistance^30,31. Molecular markers targeting SNPs and indels are the most amenable to rapid genotyping and hence effective at delineating genomic regions of interest in experimental or natural populations. But the search for candidate genes needs to take into account SVs, which have also been associated with phenotypes relevant to breeding progress. For example, spring wheats harbour whole-gene deletions at the VERNALIZATION2 locus, which means that, in contrast to winter wheats, they do not require prolonged exposure to low temperatures to flower³² and can thus be planted in spring. Another example is the locus Mla of barley, which confers resistance to the fungal disease powdery mildew and is a hotspot of copy-number variation³³. As a third example, a 13-Mb inversion in the maize genome that originated in a wild relative has a possible role in environmental adaptation, although functional studies are hampered by the fact that it is inherited as a single haplotype block^34,35. These three examples imply that SVs are not only associated with agronomic traits but may exert a direct influence on such traits.

Genome sequences are crucial in identifying and characterizing structural variation. For example, genome sequences of resistant donor varieties of wheat and barley have facilitated mapping of resistance genes^36–38, and pangenomes of tomato have shed light on the role of regulatory variation in this crop (Box 1). However, to date, whole-genome assembly remains a costly undertaking, which is why crop pangenome projects must balance different interests when genotypes are selected for sequencing. Researchers may choose to focus on either an ‘important’ crop variety, that is, a genotype that many farmers grow or researchers work with³⁹; a diversity panel that aims, with a ‘core set’ of predetermined size, to represent as much of the genetic diversity of a specific crop as feasible^40,41; or crop-wild relatives, which are used as a means to broaden genetic diversity but often still lack sequence assemblies^42,43 (Figure 2). As more genomes are assembled, crop pangenomes might in the future turn into ‘haplotype catalogues’, in which researchers can look up the genome sequence of their variety of interest⁵. When choosing which crop varieties to work with, one might select genotypes that are amenable to genetic transformation^44–47, those of parents of experimental populations^38,48,49 or genotypes that carry beneficial genes or haplotypes of interest⁵⁰. Maize serves as a good example, given that genome sequences for many maize varieties have been published in the past 5 years, each with different aims. A pangenome of 25 maize lines represents global diversity⁴⁹, whereas other studies have reported on single varieties: a parent of a widely used mutant population⁵¹; a parent of a mapping population^18,52; or a tropical maize line that proved helpful in the mapping of yield-related traits⁵³.

Box 1. Pangenomics in tomato and potato.

Tomato and potato, two species in the genus Solanum, are the world’s most important vegetable and tuber crop, respectively²⁰⁵. Because of their economic importance and – at least in tomato – tractable genetics, the genomes of both species have been studied intensely. Three tomato pangenomes^4,60,152 and a super-pangenome⁴³ of the crop and its wild relatives have been published to date. Thanks to the depth and breadth of genetic research in tomato, links between classic mutants and structural variants (SVs) became evident as soon as pangenomes were available. Li et al.⁴³ sequenced 13 genomes of cultivated tomato and its wild relatives and compiled a catalogue of structural variants. One of these was a 244-bp deletion in the cytochrome P450 gene Sgal12g015720 that was found in all cultivated tomatoes but only in 22% of wild forms (see the figure, left panel). Transgenic overexpression of the gene had higher yield in a laboratory setting (see the figure, right panel), making the gene a potential target for future breeding efforts. A variant in the sequence inventory of Gao et al.⁴ was a substitution in the promotor of a gene involved in the synthesis of flavour compounds. Alonge et al.¹⁵² observed that SVs have a widespread impact on gene expression and dissected the haplotype structure at loci affecting the weight and aroma of fruits. Zhou et al.⁶⁰ genotyped SVs with the help of a pangenome graph to improve genome-wide association mapping. A recurring theme is the link between structural variation and gene expression, which in turn modulates gene regulatory networks. A systematic inquiry of gene-by-gene through induced regulatory and genic variation in pairs of putatively interacting gene has been proposed as a strategy to investigate dosage-dependent regulatory interactions in crops¹⁵⁵.

The potato haploid genome size is similar to that of tomato, but genetics and genomics approaches in the crop are more difficult to implement because of autotetraploidy and clonal propagation. Haplotype-resolved genome assembly became feasible only with accurate long reads^206,207. An ambitious research programme aims to turn potato into an inbred seed crop to speed up genetic gains and set up hybrid breeding^208,209. Wild relatives play a crucial role in that effort. Several Solanum species are sources of self-compatibility. Single genetic factors involved in the transition from outcrossing to selfing have been isolated^210,211. As of now, one pangenome study of tetraploid cultivars and two on diploid relatives have been published. Hoopes et al.¹⁹⁴ established the technical feasibility of pangenomics in autopolyploids and studied gene expression in an allele-specific manner. Tang et al. compared the resistance gene repertoire of wild potato and identified a tuber identity gene⁸³. Wu et al.⁴² focused on deleterious variants, which are a barrier to an inbred potato: harmful variants are masked by functional alleles at the same locus in heterozygous genomes but may be lethal in inbred lines. Genome-assisted selection can rid the genome of individual loci, but the mutational load across the entire genome requires a phylogenomic approach, as chosen by Wu et al.⁴². They assembled the genomes of 87 Solanaceous species and 5 outgroup taxa and inferred from multiple sequence alignments evolutionarily constrained sites that do not tolerate amino acid exchanges in wild relatives but are affected by putatively deleterious variants in the crop. When included in genomic prediction models, the prediction accuracy for yield grew by an astonishing 25%⁴².

ToC blurb

Plant pangenomes have had a transformative impact on crop enhancement, biodiversity conservation and evolutionary research. This Review delves into the application of pangenomes for understanding trait diversity, aiding breeding, biodiversity classification and monitoring, and illuminating evolutionary innovations.

Inventories of SVs are useful beyond genetic mapping. Genomic selection is a breeding technique that predicts phenotypes from genome-wide marker profiles⁵⁴. Rather than linking any single genetic variant to phenotypic variation, the statistical models underlying genomic selection are premised on evolutionary models that posit that quantitative traits, such as yield or yield components, are controlled by many loci of small effect^55,56. Even when only a few thousand markers are used, the accuracy of genomic selection matches that of phenotypic selection, that is, the evaluation and selection of individuals based on observable traits⁵⁷. Still, the inclusion of SVs can improve prediction accuracy. Linked SNPs have proved to be but incomplete proxies of SVs⁵⁸. Models that take into account pangenome data are better at imputing sparse genotyping data⁵⁹ and predicting phenotypes⁶⁰ (Box 1).

Disease resistance gene atlases

Resistance breeding involves the development of crops that are more resilient to factors that can limit their productivity, such as diseases, pests or environmental stresses. Various methods are used for this approach: crossbreeding different plant varieties relies on natural variation to integrate resistance traits; marker-assisted selection uses genetic markers to efficiently identify desired traits; and genetic engineering directly modifies or inserts genes into a plant genome to confer resistance. Developing crops with inherent resistance reduces the reliance on chemical pesticides and fungicides, thereby promoting more sustainable agricultural practices and food security, as resistant crops are more likely to maintain their yield potential in adverse conditions. However, many plant pathogens, including viruses, bacteria and fungi, evolve rapidly due to large populations and short generation times. Resistance conferred through genes introduced by breeders is often overcome in a matter of years by newly evolved pathogen strains⁶¹. Complete knowledge of the resistance gene repertoire of a crop can help breeders find novel sources of resistance and combine them to achieve durable resistance. ^62–64 For example, a resistance gene atlas has been proposed in wheat⁶², an effort that entails, among other things, assembling the sequences of all wheat resistance genes⁵⁰.

Before whole-genome assembly became economical for a large number of samples, researchers focussed on one class of resistance genes. For example, the nucleotide-binding leucine-rich repeat (NLR) genes are a multifarious class of resistance genes that guard vast swathes of the green kingdom against a host of pathogens⁶⁵ and, for that reason, are especially well-researched⁶⁶. Thanks to conserved gene structure and sequencing, capture approaches using oligonucleotide baits as were employed in wheat are a cost-effective means of sequencing many NLR genes at once. Sequence-wide inventories of NLR genes have been compiled for the model plant Arabidopsis thaliana⁶⁷, tomato and wheat wild relatives⁶⁸. In future, targeted resistance genes enrichment sequencing may be eschewed in favour of whole-genome sequencing. An ‘NLRome’, that is, a pangenome limited to a single gene family, is defined by more than mere sequence content. Resistance genes are found in clusters that evolve rapidly owing to frequent unequal crossing over⁶⁹. What matters beyond the mere presence or absence of genes is the number and arrangement in a given genome of often virtually identical copies of NLR genes³³. The structural and functional annotation of R gene homologues is an active research field⁷⁰, and the availability of more reference-quality genome annotations will help to annotate newly assembled genomes. As computational methods improve, including for the prediction of protein structures such as AlphaFold2 (ref. ⁷¹), it may become possible to model molecular interactions and design targeted interventions to respond to rapidly evolving pathogens⁷². Pangenomes may also tell us about the evolutionary origins and patterns of structural variation also in other types of resistance genes⁷³ and similarly complex loci where duplication is common, such as metabolic gene clusters⁷⁴ or storage proteins⁷⁵.

Crop-wild relatives and ‘super-pangenomes’

Crosses between crops and their wild relatives are sources of variation that breeders are keen to exploit^29,76. Crop-wild introgressions harbour wild-derived (or ‘alien’) chromatin in an otherwise elite background and have been successfully deployed in many crops, among them wheat⁷⁷ and tomato⁷⁸. A classification scheme by Harlan and de Wet⁷⁹ divides the wild relatives of a given crop into primary, secondary and tertiary tiers, or ‘gene pools’, according to how easily they can be crossed with the cultivated form. This gene pool hierarchy affords a natural order — from most to least amenable to inter-specific crosses — in which crop wild relatives may be prioritized for genome sequencing. Extending pangenomics to higher taxonomic levels presents few conceptual hurdles, even though an entire taxonomic group is studied rather than one species. Analysis methods may differ according to whether sequence variants are fixed between reproductively isolated species or segregate in mutually interfertile populations connected by gene flow⁸⁰. Moreover, owing to the rapid turnover of repetitive sequences, sequence alignments are often confined only to genes and conserved regulatory elements. But the bare sequence assembly of the genomes of ten wheat wild relatives is not much harder than a similar feat with ten wheat cultivars.

Hence, crop wild relatives have been first among the targets of “super-pangenomics”⁸¹, a moniker given to the comparative analysis of genome sequences at taxonomic levels above the species. Genome sequence assemblies of 13 wild relatives of rice in the genus Oryza have shed light on the evolutionary dynamics of genes and repetitive elements in that taxon⁸². By scouring variation in 46 genomes of potato and its wild relatives, one study identified a gene involved in the development of tubers, storage organs that have made Solanum species targets for domestication⁸³ (Box 1). As genomes of ever more crop wild relatives^84–86 are being reported, inquiries into the evolutionary origins of crops and translational applications in introgression breeding will benefit.

The role of pangenomes in biodiversity research

The degradation of ecosystems and global warming threaten species richness^87,88 and ecosystem services, such as freshwater availability, temperature regulation and carbon sequestration⁸⁹. Pangenomics, including the sequencing of purposefully chosen genomes^90,91, can help to counteract the environmental fallout of human economic activity, including agriculture⁹², by helping ecologists monitor and mitigate biodiversity losses^93,94, thereby supporting conservation efforts, sustainable agriculture and ecosystem management.

Digitizing living libraries

Conservation genomics is the application of genetic sequencing to understand, catalogue and safeguard biodiversity. Such biodiversity may be sampled in situ, from herbaria⁹⁵ or genebanks of cultivated and wild plants²⁹. Examples of ‘plant genetic resources’, as they are termed by conservationists and breeders, are seeds stored in genebanks around the world and wild plants in the Amazon rainforest awaiting botanists’ collection missions. Genebanks are structured collections of plant materials (‘germplasm’) associated with searchable and curated data records (so-called ‘accessions’). As such, genebanks lend themselves well to systematic sequencing⁹⁶. A non-exhaustive list of crop species that have been the focus of genebank genomics include rice⁹⁷, maize⁹⁸, wheat⁹⁹, barley¹⁰⁰ and chili peppers¹⁰¹. These activities yield “molecular passport records”⁹⁶ that provide information about the structure and representativeness of collections and can help monitor seed identity in the future. DNA sequences can be complemented with other types of data, such as historic field observations and molecular phenotypes, including seed metabolites, transcript abundances or epigenetic profiles, to serve as ‘digital twins’ of genebank accessions. These surveys of genetic diversity, stored in easily accessible biorepositories, support informed choices on which germplasm to sequence in pangenomics projects. In contrast to short-read sequencing, long-read sequencing does not yet scale to thousands of samples stored in genebanks. Moreover, the genomes of some plant species, for example outcrossing and polyploid taxa, remain difficult to assemble with the latest technologies and can require expensive supporting evidence, such as genetic maps^102,103, which are more time-consuming to construct than sequence assemblies. Hence, a judicious mix of short-read and long-read sequencing at appropriate depths of coverage is needed to maximize knowledge gain per unit of currency in genebank genomics efforts.

A tiered strategy for pangenomics

The method of choice for genebank genomics is reduced representation sequencing, also known as genotyping-by-sequencing (GBS) or restriction site associated DNA markers sequencing (RADseq)⁸. In contrast to other marker systems, such as SNP arrays, sequencing does not require prior knowledge on the patterns of diversity¹⁰⁴ in a species or larger group of taxa and works reasonably well in the absence of even reference genome sequences¹⁰⁵. Tens of thousands of accessions can thus be genotyped. The high levels of duplication in genebanks^100,106 mean indiscriminate whole-genome sequencing will become cost-effective only if, and when, future drops in sequencing costs obliterate the gap between reduced representation and whole-genome sequencing. Those employing sequence-based genotyping often restrict their attention to SNPs, and for a good reason. When no alternatives were around, short-reads were used to discover and genotype SVs⁷. Of late, we have come to realize the extent to which anything but the most accurate of long reads compromises our ability fully to grasp the spectrum of structural variation^12,107. Conclusions drawn from short-read data may have been premature. Chromosome-scale sequences are now being assembled on the scale of dozens to hundreds of individuals per species. The resultant catalogues of variants, running the gamut from SNPs, short indels, to genic copy number variants to inversions or translocations of large chunks of chromatin, underpin the genotyping of SVs in short-read data of a wider set of germplasm⁵. Allelic states of SVs can be inferred by either linked SNPs¹⁰⁸ or k-mers, short oligonucleotide sequences whose copy numbers are indicative of those of the underlying sequence variation^68,109. More sophisticated approaches employing alignments of reads to genome graphs to call variants are being developed^3,9,107. A design of a pangenomics projects that strikes a reasonable balance between sequencing depth and broad taxon sampling can be visualized as a pyramid at whose tip sit genome sequences of a select few and whose base is short-read genotyping of many accessions (Figure 3). It is hoped that, thanks to technological progress, long-read sequencing will percolate all the way to the foundation¹⁰.

Figure 3. A tiered strategy for pangenomics.

Different sequence strategies (level of the pyramid) are suitable for different panel size (represented by leaf numbers). Reduced representation sequencing is done on as many genotypes, sampled in situ or from genebank collection, as possible. Representative coresets, sequenced to ever greater depth, are selected for different applications. Low-coverage (1- to 5-fold coverage) short-read whole genome sequencing aided by imputation is useful for genome-wide association scans and for genotyping known SVs. High-coverage (> 10-fold for inbred, > 30-fold for heterozygous genomes) short-read sequencing underpins selection scans, haplotype definition and demographic analyses. Genome assemblies based on long-read sequencing and chromosome-scale mapping catalogue the full spectrum of structural variation. Potentially extraordinary effort will be expended on a small number of genotypes to close gaps in difficult-to-assemble regions such as long tandem repeat arrays and centromeres to obtain telomere-to-telomere (T2T) assemblies. As technology progresses, the pyramid may turn into a cube and long-read sequencing may be employed in the bottom layers as well.

Pangenomes in evolutionary research

Full genome sequences have enabled phylogenomics or “big-data phylogenetics”¹¹⁰ where high-throughput sequencing data and increasingly whole-genome sequences are used to construct and refine phylogenetic trees. Similar to their role in crop plant research, genome sequences are useful tools for evolutionary biologists as they help map discrete genetic factors that underlie evolutionary innovations or are driving speciation.

Pangenomes in taxonomy and phylogenetics

Taxonomists and phylogeneticists name species and represent their evolutionary relationships in phylogenetic trees. But drawing boundaries between species can be difficult. Full genome sequences of entire taxonomic groups can improve the robustness of phylogenetic inference by obtaining consensus trees across many genes¹¹¹, or help explain, when there is no consensus, discrepancies that arise, for example, from hybridization or incomplete lineage sorting (the persistence of segregating variants inherited from a common ancestor)¹¹². Tree-of-life genome projects aim to sequence all forms of life. This ambitious goal requires taxonomic or geographic circumscription to achieve logistical viability in the short term. For example, the Earth Biogenome Project limits itself to eukaryotes¹¹³. Nested therein, the Darwin Tree of Life Project¹¹⁴ focusses on the British Isles. Other geographically circumscribed efforts target other regions of the world, for example, Europe¹¹⁵ or California¹¹⁶. The 1000 Plants Initiative reported assemblies of the vegetative transcriptomes of 1,124 plants species sensu latu, including green plants (Chloroplastida), glaucophytes and red algae¹¹⁷. These data illustrate and confirm hallmarks of land plant evolution, such as repeated whole-genome duplications and expansions of gene families. Even with abundant gene sequences and broad taxon sampling, some discordant phylogenies remain unresolved, possibly because of rapid speciation millions of years ago¹¹⁷. The successor to the 1000 Plants Initiative is the 10KP (10,000 Plants) Genome Sequencing Project, whose aim is to sequence representative genomes of embryophytes and green algae¹¹⁸. Complementary to these taxonomically comprehensive efforts, reduced representation sequencing has resolved several more recent branches of the plant tree of life, such as Hordeum¹¹⁹, Triticum¹²⁰ and Crocus¹²¹. It is hoped that pangenomes will serve the same purpose in the future. Pangenomic studies will aid in defining species boundaries and will be pivotal in assessing the diversity and relatedness of different populations and subspecies.

Revisions to the tree of life springing from genomics may not only move about nodes and redraw edges, but also question the very nature of the tree; that is, the tree of life would be more appropriately named the graph of life, for some of its nodes have more than one parent¹²². Horizontal gene transfer set off the evolution of eukaryotes¹²³ and, as genome sequences have revealed, happened repeatedly in the evolution of land plants^84,124,125. Un-tree-like structures may also arise from polyploidization — the coalescence of two parental species’ genomes in one nucleus — or homoploid hybrid speciation, whereby diverged, but not yet fully reproductively isolated parents mate and their hybrid offspring evolve into a thriving species in their own right. Polyploid plants are common, both on our plates¹²⁶ and in our laboratories¹²⁷. How frequent is homoploid speciation and how can it be reconciled with the homogenizing effects of gene flow counteracting speciation is an open question¹²⁸. Genomics has helped answer research questions on polyploidy using RFLP markers¹²⁹ and has served that purpose ever since¹³⁰. Of later provenance are inferences, from genome assemblies and resequencing data, for example, about the homoploid hybrid origin of mind shade¹³¹ and chestnut trees¹³². Genome sequencing can help map barriers to gene flow and thus illuminate the mechanisms of incipient speciation. Research is underway on the importance (or lack thereof) of islands of speciation¹³³. These discrete genomic regions of elevated differentiation between taxa may be related to reproductive isolation. Alternatively, such patterns may arise from processes other than population divergence such as linked selection or heterogeneous recombination landscapes¹³⁴. The latter pattern may stem from inversions, SVs that flip around large chunks of chromatin and have been for decades known to impede crossovers. Catalogues of polymorphic inversions are a by-product of pangenomes⁴⁰, and such inventories have helped dissect the role of these rearrangements in species such as barley⁴⁰ and sunflower¹³⁵. In summary, the information on the full complement of genes and their arrangements in different species afforded by pangenomes expands our ability to resolve phylogenetic trees and to understand sources of discrepancy arising in tress as a result of gene flow between closely related species.

Pangenomes reveal evolutionary innovations

The same genetic methods¹³⁶ that have been pioneered and are routinely used by crop scientists — mapping of quantitative trait loci, genome-wide association scans population genomic selection scans — are steadfastly entering the evolutionary biologist’s toolkit^137,138 (Figure 4). For example, common garden experiments supplemented by genome sequences have provided cues as to how switchgrass, both a promising bioenergy crop and an important component of the tallgrass prairie, is adapting to climate change¹³⁹. An enticing prospect is offered by the insights afforded by the comparison of 22 mammalian genome sequences into the convergent evolution of echolocation in such animals¹⁴⁰. Similar genomic approaches to study convergent evolution might be adopted to study evolutionarily innovative metabolite profiles in fruit and vegetable crops and their wild relatives, which may be mediated by copy number and presence/absence variation in genes involved in the biosynthesis of such molecules. For example, the sympatric speciation of bee orchids (Ophrys spp.) might be driven by mimicking the scent, through compounds such as alkene hydrocarbons¹⁴¹, and the shape of the pollinating insect^142,143. Traditional model plants may also be useful in evolutionary and ecological research. Arabidopsis thaliana, a plant exceptionally well-adapted to the laboratory, also grows in the wild and has become a study object of ecological genomics¹⁴⁴. Long before population-scale sequencing had been applied to other plants, 1000 Arabidopsis genomes were sequenced¹⁴⁵.

Figure 4. Pangenomics at different taxonomic levels.

Reference sequences can be assembled for the genomes of both wild and domesticated plants. Diversity panels employed in pangenome studies may span different taxonomic levels, from single species to the tree of life. The term ‘super-pangenome’ is a useful shorthand to refer to pangenomics beyond the species level. Analysis methods differ according to whether the observed genomic variants segregate in a population of interfertile individuals or represent fixed differences between reproductively isolated species. Broadly speaking, intraspecific diversity fuels genetic mapping and breeding, whereas super-pangenomes hold answers to taxonomic and evolutionary questions. At higher taxonomic levels, taxon sampling cannot but look beyond crops, as the species that farmers attend to are in a minority.

Pangenomes can help delineate adaptive changes and evolutionary processes by revealing the acquisition or loss of genes during the evolutionary history of a lineage. Beyond speciation, another evolutionary process whose study benefits from genome sequences is domestication, which results from rapid adaptation to new habitats. Owing to the selective breeding of plants (and animals) for desirable attributes, a suite of traits are commonly observed in diverse domesticated species, a phenomenon that is referred to as the domestication syndrome^146–148. Arduous genetic mapping and laboursome sequencing were required to identify the insertion of a transposable element as the causal genetic variant underlying the reduced tillering observed in domesticated maize compared with its wild progenitor^149,150. Arguably, genome sequences would make an analogous task easier these days. As more genomes of crops and their wild relatives are sequenced, more links between crop evolution and specific SVs will emerge. For example, research in tomato has established the role of structural variation in tomato breeding and its link to gene regulatory interactions^151,152 (Box 1).

In addition to genetic factors, which are comprehensively encapsulated in pangenomes, epigenetic information must be considered to understand environmental adaptation^153,154. This is part of a wider research agenda that no longer considers only the actions of genes but also their interactions¹⁵⁵. Gene regulatory networks are influenced by and, in turn, influence the abundance of genic transcripts and various classes of non-coding RNAs, DNA methylation levels, histone modifications and chromatin accessibility¹⁵⁶. Pan-epigenomes, that is, epigenetic profiles of several individuals of a species have been collected in cereal crops^49,157. Epigenomes of long-lived organisms such as forest trees might help us understand how they cope with climate change¹⁵⁸. Epigenetic mechanisms may underlie plasticity in growth and development, which in turn may enable plants better to respond to environmental stress factors¹⁵⁸. Pan-epigenomics is still in its infancy, but similar to genome assembly this field has benefited from faster and cheaper whole-genome sequencing and the development of new protocols such as assay for transposase-accessible chromatin using sequencing (ATAC-seq) to map chromatin accessiblity¹⁵⁹ and DNA affinity purification sequencing (DAP-seq) to map transcription-factor binding sites¹⁶⁰. When applied to diverse species, these techniques are bound to lead to new insights into the molecular facets of biodiversity¹⁶¹.

Future perspectives

Sequencing a human genome cost US$100,000 in 2009 and US$1,000 in 2019. Another drop by three orders of magnitude would make genome sequencing no costlier than genotyping it with a set of PCR markers. Genomicists have generated widely applicable resources for breeders, evolutionary biologists and developmental geneticists. As we widen our gaze to find new uses for our now mature tools, we should be aware of what might constrain pangenomics other than the per gigabase price of sequencing.

Future methodological challenges

Challenges in logistics may overshadow those in the lab, although both aspects are interwoven. Genotyping DNA samples by their thousands is now considerably cheaper, but not much easier and faster, than it was 10 years ago. As long-read sequencing scales to large germplasm collections or the tree-of-life’s foliage¹⁶², taxon and tissue sampling^163,164, isolation and quality control of high-molecular weight DNA^10,165, preparation of multiplexed sequencing libraries^166,167 and data management and archiving¹⁶⁷ will become increasingly challenging. Approaches that can extract nucleic acids from myriad seeds into digital sequences in a matter of weeks are needed, as are improved analysis methods that can compare thousands of whole-genome sequences in a reference-free manner. The assembly and comparison of genome sequences is particularly challenging in plants owing to the large size and repeat-rich nature of their genomes resulting from polyploidy¹⁶⁸. Haplotype phasing, the assignment of sequence to parental haplotypes¹⁶⁹, is required in heterozygous and autopolyploid plants. The development of better alignment algorithms would enable the comparison of many genomes at base-level resolution¹⁷⁰. Pangenome graphs have emerged in the last couple of years as the data structure most suited to storing and analysing multiple genome sequences¹⁷¹. They hold the promise of greater accuracy in various downstream applications such as variant calling, transcript abundance quantification and the resolution of structurally complex loci³. But there is an obstacle to the widespread adoption of pangenome graphs in plants: at the moment, these graphs test the limits of computational infrastructures, even if they operate on only a few dozen human-sized genomes¹⁷². Another open question is how pangenome graphs, which are now geared towards multiples genomes of a single species, can be generalized to genome sequences of entire genera, where divergence is higher and alignment rates are lower.

Pangenomes as community resources

A key aspect of reference genome and later pangenome projects has been the compilation of community resources and the provision of an ancillary infrastructure to facilitate access, such as genome browsers and repositories for bulk download^173,174. With genome sequencing about to become a quotidian effort, genome assembly often occurs in the pursuit of narrowly circumscribed research projects without consideration of long-term resourcing. However, accessibility in the long term is in part what democratizing DNA sequencing¹⁷⁵ is about. Submitting genome assemblies to public sequence archives enables later synthesis and collation of genome sequences to derive knowledge from a comparative outlook¹⁷⁶ or simply curation to host all genome sequences of a species in one community hub¹⁷⁷. As of now, such efforts have been few and far between and are possibly hampered by the diversity of applied sequencing strategies. As the speed of progress relents and best practices settle in, these endeavours will gain in prominence. Procedures for depositing sequence assemblies and their underlying raw data are well established by the repositories that are members of the International Nucleotide Sequence Database Collaboration¹⁷⁸. Standards recommendation for phylogenomic sequencing initiatives, including minimum quality standards for assemblies and annotation, have been proposed by the Earth Biogenome Project¹⁷⁹.

Access to biological diversity

Genome researchers are accustomed to unfettered access to sequence data with no other obligations than to cite research articles written by data depositors. By contrast, property rights to plant genetic material are governed by a legal framework of bewildering complexity (or so it may seem to the uninitiated). Access to plant genetic resources is governed by international covenants, among the best known of which are the Convention on Biological Diversity, the International Treaty on Plant Genetic Resources for Food and Agriculture, and the Nagoya Protocol on access and benefit sharing¹⁸⁰. Today’s debates revolve around Digital Sequence Information (DSI), a concept that defies easy definition¹⁸¹. In any event, genome sequences fall well under the DSI extended purview. Many scientists wish for sequence data to remain publicly accessible, and practical solutions have been proposed to keep international sequence archives accessible while ensuring the equitable distribution of proceeds among stakeholders¹⁸².

Conclusions

Surveys of structural variation have taught us that, to understand the full extent of sequence diversity of a species, we need to compare many individual genome sequences. However, pangenomics remains in its infancy; although sequencing technologies and analysis methods are improving at a rapid pace, and most crop plants have reference genomes, few have pangenomes. Pangenomes of higher taxonomic units may become foundational community resources that help to better appreciate the role of evolutionary processes. After crop genome sequencing, wild relatives are the next frontier in agricultural genomics. A democratization of pangenomes driven by cheaper DNA sequencing and easier-to-use computational methods is underway. We eagerly await the outcomes of sequencing the tree of life.