Are cereal grasses a single genetic system?

Abstract

In 1993 a passionate and provocative call to arms urged cereal researchers to consider the taxon they study as a single-genetic system and collaborate with each other. Since then that group of scientists has seen their discipline blossom in the following decades. In an attempt to understand what unity of genetic systems means and how the notion was borne out by later research, we survey the progress and prospects of cereal genomics: sequence assemblies, population-scale sequencing, resistance gene cloning and domestication genetics. Gene order may not be as extraordinarily well-conserved in the grasses as once thought. Yet still, several recurring themes have emerged. The same ancestral molecular pathways defining plant architecture have been co-opted in the evolution of different cereal crops. Such genetic convergence as much as cross-fertilization of ideas between cereal geneticists has led to rich harvest of genes that, it is hoped, will lead to improved varieties.

Introduction

Thirty years ago, Bennetzen and Freeling¹ proclaimed the grasses to be a single genetic system. What does this mean? Before we proceed to answer that question, let us recapitulate why we should care about cereal grasses. In fact, we should care about any kind of grass, not only the cereal sort. Grasslands cover between a quarter and two fifth of Earth’s land area^2-4 and 1 in 25 flowering plant species are grasses⁵ (Figure 1). For about ten thousand years, a dozen or so grass species have been the main source of carbohydrates and a substantial one of protein and fat for people and their livestock. The two combined now make up 96 % of terrestrial mammalian biomass⁶. The most nutritious organs of cereals are, for a non-ruminant consumer, their single-seeded fruits, which are commonly known as grains, technically referred to as caryopses and divided into the germ – the “embryo seed” that grows into a plant – and the endosperm, a starch-rich storage organ that feeds the embryo (and people).

Figure 1. Simplified tree of the family Poaceae.

The numbers of genera and species were obtained from Soreng et al.²²⁶. The timescale (million years ago [Mya]) indicates approximate divergence times for subfamilies and the Triticeae tribe according to Gallaher et al.²²⁷. BOP and PACMAD are the main monophyletic clades used to divide the Poaceae family.

Cereals were among the first plant species cultivated by human beings. Since then people have grown reliant on grains for their sustenance, sometimes desperately so: the Romans tried to avert devastating foliar disease in the wheat crop by canine sacrifice⁷. The measures to safeguard crop production have become more effectual over the millennia, owing in part to science. America’s land grant universities and agricultural research stations were established from the 1860s onwards to benefit that country’s mainly cereal-growing farmers. European nations woke up to the challenge of increasing domestic wheat production in the aftermath of the First World War. Later the Green Revolution arose from the newly arisen need for crops that can take advantages of a now unlimited supply of nitrogen thanks to the Haber-Bosch process⁸. Scientific principles in breeding in its various manifestations turned out to be an economic success story throughout the world, especially in low-income countries⁹, although much remains to be done to achieve “zero hunger”, one of the 17 sustainable development goals of the United Nations.

Cereals have played a key role in that story. Agriculture originated independently several times. Civilizations in Europe, East Asia and Mesoamerica persisted to a large degree on, respectively, wheat, rice and maize. For a long time, there were few contacts between these civilizations. From the 1500s onwards, the flow of people, ideas and seeds brought about rapid scientific and economic progress. Genetics, the science of crosses, dates from the 20^th century¹⁰. Crosses within species have done a lot for crop improvement. Hybrid maize stands as one of genetics’ most enduring monuments¹¹. Crosses between different species found practical applications as well: introgressions from wild relatives protect wheat against pathogens^12,13 and genetically homogeneous wheat seeds can be obtained rapidly by applying maize pollen to wheat pistils¹⁴. Wheat genes have been transformed into barley^15,16. But does this make them a single genetic system?

Shifting expectations: from molecular markers to genome sequences

No single definition of “genetic system” exists. Bennetzen and Freeling started from the then brand-new observation that molecular marker maps of different grasses were collinear^17-19. This suggested that genes in, say, maize are more or less in the same order as they are in wheat, corn’s distant cousin to the 80 millionth degree or so²⁰. As per the testimony of those that made the discoveries, such collinearity came as a big surprise²¹. The next generation of geneticists has grown up with different expectations: occasional interchromosomal translocations in rye and oat relative to the Pooideae’s common ancestor were taken as evidence for mosaicity²² and reticulate evolution²³ (Figure 2).

Figure 2. Gene collinearity in the grasses: concept and visualizations.

(a) Early genetics hinted at conserved gene order between the grasses. (b) Genome sequences have borne out that notion at a chromosomal scale. (c) But they have also revealed local rearrangements even between members of the same species. Images were taken from Moore et al.²¹, Lovell et al.²²⁸, and Brunner et al.²²⁹

In 1993, Bennetzen and Freeling drew practical conclusions²⁴. They and others had seen that restriction fragment length polymorphism (RFLP) probes cloned from the rice genome hybridized to the maize or wheat genomes and could be mapped genetically in those species. By exchanging molecular marker probes, they argued, scientists working on different grass species might speed up progress in their respective fields. Up to a certain point^25-27 that worked as advertised and culminated, twenty years later, in so-called “genome zippers” that projected gene order from grasses with small genomes and contiguous sequence assemblies onto dense genetic linkage maps of their cousins with larger genomes^23,28,29. By 2010, the compact genome sequences of rice (390 Mb)^30,31, sorghum (730 Mb)³² and the model grass Brachypodium distachyon (purple false brome; 300 Mb)³³ had been assembled (Table 1, Figure 3). Doing just that with the diploid barley genome (5 Gb) required dividing sequencing duties among the members of an international consortium³⁴ as had been done by the Human Genome Sequencing Consortium. Assembling a wheat genome sequence (15 Gb) was considered an even more daunting task³⁵. Since the formation of the first sequencing consortia, genome sequencing assembly has become both faster and easier^36-39 thanks to technological progress in DNA sequencing. (Figure 3). Genetic and physical maps were constructed with ever greater ease^40-42. Computers became faster and algorithms more sophisticated^43,44. A widely used approach these days is accurate long-read sequencing⁴⁵ combined with a physical linkage map constructed by chromosome conformation capture sequencing^46,47 (Box 1). Most chromosomes of cereal species have yet to be assembled telomere-to-telomere, but the gaps in the most recent barley reference genome sequence assembly^48,49 are sparser than the markers in the first linkage map of that species⁸. To use a turn of phrase of Bennetzen and Freeling, the challenges in genome sequencing have become “less scientific than organizational”. It is now no more difficult to sequence a wheat genome than a rice genome^50-53. That has brought a curious shift in our priorities. Not so long ago it seemed that the genome sequence of rice “zipped” onto a molecular marker map of wheat would go a long way⁵⁴. Now much hope is invested in sequencing ever more individuals of the same species.

Table 1. Selected genome sequences of cereal crops.

Species	Size	Year	Quality	Methods	Ref.
Rice (Oryza sativa)	390 Mb	2005	Pseudomolecules	BAC sequencing.	30
		2013	Pseudomolecules	Short-reads and optical mapping.	215
		2023	T2T	Long-reads (PacBio HiFi and ONT), Hi-C	31
Sorghum (Sorghum bicolor)	730 Mb	2009	Pseudomolecules	BAC sequencing, genetic map	32
		2018	Pseudomolecules	Short-read resequencing of BACs, genetic map, primer walking.	216
Purple false brome (Brachypodium distachyon)	300 Mb	2010	Pseudomolecules	BAC sequencing, genetic map	33
Wheat (Triticum aestivum)	15 Gb	2012	Contigs	Short-reads	172
		2014	Contigs	Isolated chromosome-arm sequencing with short-reads	217
		2018	Pseudomolecules	Short-reads, BAC sequencing with short-reads, optical maps, radiation hybrid maps, genetic maps, Hi-C	103
Barley (Hordeum vulgare)	4.8 Gb	2012	Contigs	BAC sequencing with short-reads	173
		2017	Pseudomolecules	BAC sequencing with short-reads	218
		2019	Pseudomolecules	Short-reads, Hi-C, linked reads, optical map	219
		2021	Pseudomolecules	Long-reads (PacBio HiFi), Hi-C	175
Maize B73 (Zea mays)	2.1 Gb	2009	Pseudomolecules	BAC sequencing, optical map	220
Maize Mo17 (Zea mays)		2023	T2T	Long-reads (ONT and PacBio HiFi).	176
Teff (Eragrostis tef)	577 Mb	2014	Contigs	Short-reads	221
		2020	Pseudomolecules	Long-reads (PacBio), Hi-C	222
Rye (Secale cereale)		2017	Contigs	Short-reads, chromosome arm assingment	223
		2021	Pseudomolecules	Short-reads, Hi-C, linked reads, optical map	224
	7.9 Gb	2021	Pseudomolecules	Long-reads (PacBio), short reads, Hi-C	225
Oat (Avena sativa)	11 Gb	2022	Pseudomolecules	Short-reads, Hi-C, linked reads, optical map	22

Figure 3. Time line of cereal genomics.

The upper part shows achievements in cereal genetics and genomics. The technologies that enabled those achievements are listed in the bottom part.

Box 1. Telomere-to-telomere assemblies: the final frontier.

Schematic representation of hard-to-assemble regions

The recent completion of a human pan-genome²⁰⁷ led many to recount the story how the first human genome came into being. Cereal genomicists would have a similar yarn to spin. Molecular marker maps, bacterial artificial chromosomes and short-read sequencing have all had their heyday (Table 1, Figure 3). Now other methods for whole-genome sequencing and assembly have superseded them. Contigs assembled from accurate long-reads extend into the dozens of megabases and span occasionally entire chromosomes or chromosome arms^208,209. Reads may even get longer, more accurate and more abundant in the future. But progress may relent. We may not see any more such technological and algorithmic sea changes as were wrought in the last decade by successive next generations of sequencing. Assembling the sequence of a cereal genome is now easy and fast. It has yet to become cheap. A chromosome-scale assembly of a 15-Gb bread wheat genome still costs thousands of US dollars. But the tools of molecular and computational biology employed in genome assembly do not require any longer international sequencing consortia, but are accessible to single research labs.

So, what is left to do apart from exercising our newly gained capabilities of assembling the sequences of many more grass genomes wild and domesticated? One frontier is telomere-to-telomere assemblies. One is now available for the human genome, although some leeway must be granted for the estimation of ribosomal repeat numbers in that assembly²¹⁰. Among the exciting biology emanating from that assembly is the epigenetic landscape of human centromeres²¹¹. Such research would also find its practitioners in cereals, but obtaining the requisite sequences is hard. The current barley reference genome breaks off at the centromeres, owing possibly to GA-rich satellite repeats inaccessible to PacBio HiFi sequencing⁴⁸. The organization and expression of ribosomal DNA (rDNA) arrays is a topic no less fascinating than centromeres²¹². Apart from those functional loci, interspersed tri- or nucleotide satellite arrays so large they can be observed cytologically disrupt the barley reference⁴⁸ and longer tandem repeats dot the wheat genome²¹³. Ultra-long nanopore reads have recently helped bridge such regions in maize¹⁷⁶ and rice³¹. The HiFi reads alone were sufficient to assemble the centromeres of einkorn, a diploid wheat²¹⁴.

At the moment putting together a single telomere-to-telomere assembly of a large cereal genome might require more thought and curation than assembling a hundred such genome sequences with a dozen gaps or so per chromosome. The needs of many scientists are better served by the latter resource, but as reads get longer, gapless assembly is well worth trying.

Sequencing many genomes

Convinced that such optimism is not misplaced, we and others have argued for using genome sequencing better to understand and utilize genetic diversity in crops^55,56. These tasks entail the processing of tens of thousands of DNA samples. Sequencing has grown ever cheaper, but not quite as much yet as to scale whole genome sequencing to entire genebank collections. That is why genotyping-by-sequencing is most commonly done through digestion of DNA with restriction enzymes followed by short-read sequencing⁵⁷. The resultant sequence data are, by means of alignment to a reference genome and computational discovery of sequence variants, transformed into a matrix representation with sampled individuals in columns and one row for each genome-referenced marker. With such variant matrices in hand, we can define ancestrally related groups of individuals (germplasm groups), select representative subsets from the universe of sampled individuals (core sets)⁵⁸ and find duplicates (groups of identical samples) and outliers. The latter comprise those samples whose ancestral group assignment does not mesh with their geographic provenance⁵⁵. Think of purportedly Ethiopian barley landraces whose closed genetic relatives were all collected in China. Such “genebank genomics” has been done in maize⁵⁹, sorghum⁶⁰, barley⁶¹, wheat⁶² and rice⁶³.

Augmenting traditional passport data – a table with each sample’s provenance and taxonomy - with those of the molecular sort is no small feat. An equally exciting but vaster undertaking is the prediction of phenotype from genotype. The cereal collections of a single genebank^61,62 may house tens of thousands of seed lots with associated metadata (“accessions”). In out-crossing species such as rye and maize, accessions representing traditional varieties may be made up of hundreds of genotypes⁶⁴. Algorithms premised on Fisher’s infinitesimal model⁶⁵ infer from DNA sequences of many genotypes and phenotypic observations of fewer genotypes in the larger set. Those traits can be either readily observable (e.g. flowering time) or conceptual, such as the likely merit of an individual’s offspring for crop improvement (its breeding value). Either way, quantitative genetic estimates are cheaper than field trials and can be as accurate⁶⁶. As genome sequencing can be scaled up much easier than field work, the genomic prediction of phenotypes may help select the most promising plant genetic resources held in genebanks. First such efforts have been reported in wheat^67,68, barley⁶⁹, maize⁷⁰, sorghum⁷¹ and rice⁷². A caveat to use genomic predictive models is their dependence on the choice of training population and possibly poor prediction accuracies in environments to which the training population is not adapted. Prediction across environments is a field of active research⁷³.

Complexity reduction methods employing restriction enzymes^57,74, capture probes^75,76 or chromosome sorting⁷⁷ are being replaced by whole-genome sequencing unless sample numbers are large. Chromosome-scale sequences of several varieties of a crop have been assembled with qualities on a par or sometimes superior to that of the crops’ respective “reference” genome sequence. Such pangenomics has received much attention of late. A practical definition of a pangenome is a collection of genome sequence assemblies of multiple individuals of a species (or higher taxonomic group) along with ancillary resources such as annotations and genome browsers.⁷⁸ Pangenomes comprising tens of genotypes have been constructed in maize⁷⁹, wheat⁸⁰ and barley⁸¹ (Figure 3).

The population genetics of crop evolution

Early evolutionary biologists put a lot of emphasis on natural selection. As the physical carriers of genetic information became known at ever greater resolution, geneticists came to realize that many processes affecting those carriers – mutagenesis, recombination, allele frequency oscillations – do not affect fitness most of the time and follow predictable patterns not unlike the laws of physics⁸². The statistical analysis of such nearly neutral diversity is among the concerns of population genetics. Sophisticated analysis methods have been developed to infer demographic history from sequence data. The diploid sequence of a single human genome informs about the size trajectory of a considerably larger swathe of the human race⁸³. The analysis method behind such inference, the sequentially Markovian coalescent, has been employed in maize⁸⁴, rice⁸⁵, wheat⁸⁶ and sorghum⁸⁷. A trajectory shared by all of these crops is a steady decline in the number of ancestors whose genetic material went its way into the varieties grown today. Such “bottlenecks” are a cause for concern for those who worry that low diversity might curtail future genetic gains. Comfort might be derived from recent findings that mutation breeding⁸¹ and the directed use of wild germplasm^12,68,80 may have counteracted genetic erosion to a greater extent than previously thought.

No less impressive than the numeric underpinnings of population genetics is the experimental prowess of its practitioners. Our ideas about human demographic history have been repeatedly upended by insights gleaned from minute quantities of DNA extracted from bones^88,89. Osseous tissue is a better protectant from heat, humidity and other disturbances than seed coats. That is one reason why we know so much more about the genetic provenance of the first Middle Eastern farmers^90,91 than about that of their crops. Still ancient DNA studies have illuminated aspects of crop evolution in maize^92,93, barley⁹⁴ and wheat⁹⁵. A common theme emerging from this body of research is the role of geneflow between crops and their wild counterparts. Adaptive introgressions may have propelled the geographic range expansion of maize⁹⁶. Hybrid swarms have confounded barley taxonomists⁹⁷ and, more disconcertingly, all but obliterated pristine wild populations of Asian rice⁹⁸. The geneflow moniker also fits the coalescence of multiple unreduced gametes from different species into stable allopolyploid offspring. Wheat is one such hybrid. Its polyploidy is a topic that has been excellently treated elsewhere^99-101. The consensus is that the polyploidy of wheat was a consequence of domestication and has been a likely prerequisite for improvement of the crop¹⁰². It is hoped that the rather recent completion of reference genome sequences of bread wheat¹⁰³ and its progenitors^104,105 will lead us to discover more about how polyploidy has played out in the crop. However, demographic trajectories over the last several millennia are more likely to enrapture crop evolutionists than breeders and genetic engineers, who look for genetic markers diagnostic of traits or work out links between a gene’s experimentally established function and agronomic performance.

Genes and genomes

A genome is a catalogue of all genes of an organism along with intervening non-coding sequence. But knowing the DNA sequences of all genes does not tell us what any of them actually does. Linking genes to phenotypes is the remit of genetic mapping. One class of genes that has received special attention from researchers are resistance genes. Owing to the need of protecting crops against rapidly evolving pathogens¹⁰⁶, plant pathologists are cloning resistance genes to unravel the molecular basis of plant resistance. Cloning here means the establishment of conclusive links between a given phenotype (how well a plant can withstand a certain pathogen) and the presence of a certain piece of DNA. By one count, 47 wheat resistance genes had been cloned as of 2021 and it is hoped that by 2031 all will have been sequenced¹⁰⁷. Resistance gene mapping has traditionally employed experimental populations so as to harness recombination in crosses between a small number of founders (often two) in the hunt for genes. Alternatively, genome-wide association studies (GWAS) can be conducted. In these, observed traits are associated with genome sequences that capture natural diversity and recombination events that happened long ago. The statistical signals can be fickle and further evidence from experimental populations and, more importantly, induced mutants is required conclusively to link the presence of genetic elements to resistances (or other traits)¹⁰⁸. Those caveats notwithstanding, GWAS scans have led to the isolation of resistance genes in diploid wheat wild relatives^109,110, not least thanks to the genome sequences of those species and the clever crunching of population sequence data¹¹¹.

Disease genetics has been a driving force behind genomics, in medicine¹¹² as in plant science. Early insights into the limitations of cereal comparative genomics were gleaned from the difficulties that were encountered during resistance gene mapping^26,27 and which turned out be the consequence of a more general pattern. Resistance genes are now known to evolve faster and to be more diverse than other many gene families, both in nucleotide sequence and structure^113,114. This is why resistance gene loci in different grasses often lack the micro-collinearity that might have been expected in a single genetic system¹¹⁵. When genome sequencing was expensive and difficult, enriching sequencing libraries for resistance gene homologs was the way to go^51,116,117. These days, assembling the sequence of a whole wheat genome makes sense even if one is only after a single resistance gene^118,119. On top of that, full genomes avoid the risk of losing genes overlooked by capture designs. The most interesting resistance genes are arguably those that do not belong to large gene families^120,121. A single reference genome cannot stand for all members of a species, but even larger collections of genome sequences fall short in that respect. Alleles conferring resistance against recently arisen pathogens may occur at low frequencies in global populations^122,123, reside in clusters⁵¹ evolving so fast that even closely related genomes provide little guidance, or stem from crop-wild introgressions of obscure provenance¹²⁴. Evolutionary studies stand to profit from ever more genome sequences. How resistance genes evolve is a fascinating research question with practical implications. The more structurally correct and complete genome assemblies we have, the better we will be able to answer it.

Another group of genes that has intrigued scientists are those that control the composition, size, shape, number and arrangement of the single-seeded fruits that people derive sustenance from (Table 2). The genetic basis of grain dimensions has been intensely studied in rice. Different people like their rice long or short, round or thin, glutinous or non-sticky. These diversified selection pressures have endowed the rice genome with a plethora of quantitative trait loci (QTL) for grain-related traits¹²⁵. The genes under the QTL have sundry molecular functions. To name but a few, genes have been cloned controlling grain number (Gn1a, a oxidase/dehydrogenase¹²⁶); grain length (GS3, a protein with four domains, one of which is a plant-specific organ size regulation domain¹²⁷); grain width (GW5, a gene with two IQ calmodulin-binding motifs and one unknown function domain¹²⁸); amylose content (Wx, a granule-bound starch synthase¹²⁹) and overall yield (DEP1, a phosphatidylethanolamine-binding protein-like domain protein¹³⁰). Genes affecting the properties of grains have been cloned in other cereals: protein content (Gpc-B1, a NAC transcription factor¹³¹) and threshability (Q, an AP2 transcription factor¹³²) in wheat and the adherence of integuments to barley grains (NUD, an ethylene response factor family transcription factor gene¹³³). It is as reasonable to wonder whether the same genes or the same molecular pathways have been co-opted to accommodate the needs of their domesticators as it is to doubt such convergence given the functional disparity of the genes cloned thus far. Educated guesses about candidate genes have worked sometimes^61,134, sometimes not¹³⁵. Even in the absence of convergent evolution, we might presume to induce it by genetic engineering. Transgenic approaches have yielded promising results with resistance genes shuttled from wheat into barley^15,16. More research into the possibility of achieving like feats at greater phylogenetic distances and with genetically more complex traits are welcome.

Table 2. Genes and QTLs involved in architectural changes in cereal crops.

General function	Gene	Species	Ref.
Grain characteristics	Gn1a (QTL of grain number on chromosome 1)	Rice	126
	GS3 (QTL of grain size on chromosome 3)	Rice	127
	GW5 (QTL of grain widh on chromosome 5)	Rice	128
	Wx (waxy)	Rice	129
	Gpc-B1 (grain protein content)	Wheat	131
	Q	Wheat	132
	NUD (nudum)	Barley	133
Branching and architecture	TB1 (TEOSINTE BRANCHED1)	Maize	144
		Barley (Vrs5 or Int-C)	134
		Wheat	198
	IPA1 (IDEAL PLANT ARCHITECTURE1)	Rice	146
	DEP1 (DENSE AND ERECT PANICLE1)	Rice	130
		Wheat	135
	Vrs1 (SIX-ROWED SPIKE1)	Barley	149
		Wheat	150
	Vrs4, Vrs3 and Vrs2 (SIX-ROWED SPIKE4, 3 and 2)	Barley	151-154
Seed shattering	Btr1 and Btr2 (BRITTLE RACHIS1 and 2)	Barley	156,157
		Wheat	157
		Rye	157
	qSH1 (QTL of seed shattering on chromosome 1)	Rice	160
		Pearl millet	166
	Sh1 (Shattering1)	Rice	164
		Sorghum	164
		Foxtail millet	165
		Maize (possibly)	164
	Sh4 (Shattering4)	Rice	161,167
	qSH3 (QTL of seed shattering on chromosome 3)	Rice	162,167

Genomic research has made important contributions to gene mapping. Genome sequences help develop markers and, once a locus has been delimited to a sufficiently narrow interval, inform about the gene content therein. How far such help goes is best illustrated by the rapid progress of research in the crop whose genome sequence we have enjoyed longest. In each single year after 2006, more rice genes were cloned than in all years before 2000 combined¹³⁶. Such exuberance began with the release in 1995 of a library of bacterial artificial chromosomes and in 2001 of a reference genome sequence¹³⁷.

Convergent architectural changes in cereal crop evolution

It is matter of debate if the earliest farmers selected (possibly inadvertently) for larger grains or whether another set of traits took precedence, temporally and conceptually^138-140. The latter pertain to the arrangement of the grain-bearing organs on the mother plants and, of possibly even greater importance, their separation from it.

Let us begin with plant architecture before moving on to grain dispersal. There is a one-to-one correspondence in the grasses between flower and single-seeded fruit. Flowers are encapsulated in subtending structures called spikelets, each of which supports one or more flowers. Spikelets are arranged in varied compound structures that are properly referred to as synflorescences, but are called inflorescences for convenience¹⁴¹. Several inflorescence-bearing shoots are created in a process called tillering that is distinct from vegetative branching in dicots. We limit ourselves here to a few examples where convergent evolution has affected the same gene actions (Table 2).

The origin of maize has perplexed scientists for decades until George Beadle showed through crosses that five or so genes turn teosinte, a bushy kind of grass from Central America, into low-tillering maize^142,143. One of the five is TEOSINTE BRANCHED1 (TB1)¹⁴⁴, which belongs to a class of transcription factors partly named in its honor: the Teosinte branched1/Cycloidea/Proliferating cell factor class II gene family. Through arduous fine-mapping, which would undoubtedly have been helped by the genome sequences to come, Studer et al. pinpointed a retrotransposons insertion 50 kb upstream as a causal variant¹⁴⁵, furnishing, by way of example, latter-day pangenomicists with a justification for their discipline.

Branching has been important in balancing resource allocation to grains and vegetative organs in other cereals. The above-mentioned DEP1 and the optimistically named IDEAL PLANT ARCHITECTURE1 (IPA1) enhance the potential yield of rice by, respectively, shortening the length of inflorescence branches¹³⁰, and suppressing tillering¹⁴⁶. Higher yields in wheat and barley can be achieved by increasing the number of grain-bearing florets, either by stimulating their initiation or suppressing their subsequent abortion^147,148. If the SIX-ROWED SPIKE1 (Vrs1) gene is knocked out, the ancestrally infertile lateral grains at each inflorescence node bear grains, thereby tripling grain number¹⁴⁹. Alleles of the gene also serve that aim in wheat. The multiflorous spikelets of that species set more grains in those of its varieties that express the gene at low levels¹⁵⁰. The isolation of Vrs1 in barley was followed (although not in order) by a count-down of other genes that modify lateral spikelet fertility: Vrs5¹³⁴, Vrs4¹⁵¹, Vrs3^152,153 and Vrs2¹⁵⁴. Mapped in classical mutants, the only one among them that has played a role in crop evolution is, Vrs5, better known to barley geneticists as INTERMEDIUM-C. After cloning and sequence analysis, Vrs5 was found to be the ortholog of maize TB1¹³⁴.

Evidence for convergent evolution tinkering with the actions of the same genes is seen also in seed shattering. The survival of a wild plant’s offspring is contingent on its separation from the parental organs that have fed them. In crops, by contrast, seed dispersal lowers yield and makes harvest laborious and ineffective¹⁵⁵. That is why grain shattering has been abolished across the cereals by genetic changes selected by ancient farmers. These changes have been effected by modifications to different pathways according to whether a cereal crop’s inflorescence is a spike or panicle. But within these two groups, overlapping sets of genes have been co-opted to respond to new selection pressure exerted by the domesticators.

A spike is a branchless inflorescence along whose main axis (the rachis), spikelets are arranged in opposite vertical rows. The spikes of the progenitors of the domesticated Triticeae wheat, barley and rye break above each grain-bearing node of the rachis. Genes controlling that process were first isolated in barley. Functional knock-out of either BRITTLE RACHIS1 (Btr1) or Btr2 is sufficient to make grains adhere to the rachis. This bolsters long-held beliefs that barley may have been domesticated more than once¹⁵⁶. The molecular functions of Btr1 and Btr2 are as yet unknown, but may have been acquired thanks to subfunctionalisation following a segmental duplication, possibly in the ancestor of the Triticeae¹⁵⁷. Putative loss-of-function variants were found in the Btr1 orthologs of emmer and durum wheats (T. turgidum)^104,158,159 as well as in that of einkorn wheat.

The inflorescences of rice, sorghum and millets are panicles with multifarious branching patterns. The propensity for seed shattering has been mapped in those species as a quantitative trait under the control of several loci. Among the best studied are qSH1, Sh1 and Sh4, respectively. qSH1 and Sh4 of Asian rice were cloned in 2006¹⁶⁰. However, loss-of-function alleles of either of them are insufficient on their own to abolish shattering¹⁶¹. Yet another gene, qSH3, was cloned more recently¹⁶², fueling speculations about a step-wise domestication of rice ¹⁶³. Sh1 was cloned in sorghum¹⁶⁴. Expression of its ortholog in rice was reduced in a shattering resistant mutant¹⁶⁴ and a transposable element (TE) inserted in the second exon of the foxtail millet¹⁶⁵ ortholog. In pearl millet, a region on chromosome 6 responsible for seed shattering and spikelet structure corresponds to qSH1 in rice¹⁶⁶.The three genes – qSH1, Sh1, Sh4 – are involved in abscission layer formation^160,164,167, but rice grains may adhere more firmly to the mother plant also thanks to closed rather than open panicles¹⁶⁸. The ears of maize and its progenitor teosinte are neither spikes nor panicles and less is known about the genes that control their shattering in that species. Sh1 seems to play a role in it¹⁶⁴. The making and breaking of cereal inflorescences has responded to the same selective pressures and orthologous genes have afforded the molecular basis for that response. That makes, in the words of Paterson et al., an “impelling” case for comparative mapping¹⁶⁹. But has this method really “obviated the difficulties” associated with map-based cloning, as posited by those authors almost twenty years ago? Since then, a gene-centric view has been complemented, and complicated, by studies of the intergenic space.

Regulatory diversity and loss of functional orthology

A gene may be the structural ortholog of another gene in a related species in the sense that the gene and its ortholog trace back to a common ancestor and that this ancestral relationship applies also to the genes in their respective surroundings. In a pair-wise alignment of cereal genome sequences, structural orthology is visible as colinear gene blocks (Fig. 2). At the same time, this gene and its structural orthologs may not carry out equivalent biological functions¹⁷⁰. Such loss of functional orthology may arise from changes to the coding sequence. For example, non-synonymous variants might bring about protein conformational changes that alter the binding affinities of transporters and transcription factors. Another avenue of functional divergence is regulatory variation. For example, a TE might insert into regulatory regions and inactivate a gene’s promoter. The large intergenic space of cereal genomes holds ample opportunity for this and other types of gene regulatory evolution to occur.

Molecular biologists have known for a long time that many cereal genomes are large and rich in repetitive DNA. However, technical obstacles stood in the way of cataloguing non-coding elements in the same manner as genes. The most common types of TEs such as BARE1 element of barley, were discovered by cloning and Sanger sequencing and by the mid 2000s, a taxonomy of repetitive elements¹⁷¹ had been compiled that grouped TEs in eukaryotic genomes into classes (retrotransposons and DNA transposons), orders (e.g. long-terminal repeat [LTR] retrotransposons, superfamilies (Gypsy and Copia) and families such as BARE1. But early reference sequences of large cereals genome like those of wheat¹⁷² and barley¹⁷³ did not represent the entire repetitive space because short reads could not resolve long identical stretches of DNA, even if assembly was restricted to small (~100 kb) portions of genomes that were represented in bacterial artificial chromosomes¹⁷⁴. This stumbling block has recently been removed by PacBio HiFi reads. Their length exceeds that of cereal TEs so that they can represent accurately even recent insertions¹⁷⁵. The remaining obstacle to the gapless sequence assembly of cereal genomes are long homogeneous arrays of ribosomal DNA and centromeric satellite arrays^48,176 (Box 1).

Cereal genomicists did not have to wait for the advent of telomere-to-telomere assemblies to glean insights into interplay of mobile DNA and evolutionary innovation. The role of maize TB1 has already been mentioned. Also in maize, Morgante et al. showed that the movement of Helitron DNA transposons resulted in exon reshuffling and gene duplication¹⁷⁷. A method for dating insertion of LTR-TEs by counting variants that have accumulated in their LTR was developed by SanMiguel et al¹⁷⁸. Studies comparing loci there were structurally orthologous in different cereal genomes established the virtual non-conservation of the intergenic spaces owing to the rapid turnover of mobile elements^179,180. The pangenomes of cereal crops and their wild relatives that are being developed at the moment^79-81,181 will extend these findings to the whole-genome level. But genome sequences can provide only hints that need to be followed up by functional studies. Beyond the algorithmic and computational challenge of annotating and classifying transposable elements correctly within and across species^182,183, experimental work will be required to establish conclusive links between putative regulatory polymorphisms – many of them structural variants –, the molecular functions of the genes they influence and the interactions of those genes with other genetic elements^184,185. Epigenomic^186,187 and transcriptomic atlases^188,189 have recently been extended to the level of individual tissues and cell-type thanks to spatially resolved¹⁹⁰ and single-cell¹⁹¹ transcriptomics and epigenomics¹⁸⁷. Functional proof of interactions between genetic elements will require the targeted generation of regulatory diversity, for example by the creation of allelic series of promotor variants¹⁸⁴. The outcomes of these experiments may further challenge the notion that grasses are a single genetic system.

Single genetic systems: unity of purpose

How has the “single genetic system” fared since Bennetzen und Freeling proposed it in 1993? The powers once ascribed to RFLP markers lie now with pangenomes, both of a single species and at higher taxonomic levels. The latter should enable us to assess how colinear grass genomes are in relation to those of taxa like the Solanaceae (nightshades) or mammals that are of similar age. Freeling and Bennetzen thought gene order in cereals exceptionally well conserved because the first molecular marker maps of tomato and pepper bore little resemblance to each other. By the time the genomes of the crops in question were assembled, the argument had collapsed: chromosomes 2 of Capsicum annum and Solanum lycopersicum are about as collinear as chromosomes 2H of Hordeum vulgare and 2A of Triticum aestivum. To bolster such off-the-cuff argumentation, we had better estimate precisely how fast structural genome evolution happens in different taxa. This requires more sequencing. Luckily, tree-of-life genome projects^192,193 are already underway. Ten thousand or so grass genomes along with those of 2,700 Solanaceae¹⁹⁴ and 6,400 mammals¹⁹⁵ will put comparative genomics on a surer footing. Sequence assemblies of all Triticeae will also help identify the donor species of those alien introgressions whose provenance is obscure¹².

In developmental genetics, the single genetic system remains a concept that is frequenctly evoked. McSteen and Kellogg proposed that grasses “function” as single genetic system thanks to the “broad similarity” between their genes¹⁹⁶. To avoid a term as heavily loaded as function¹⁹⁷, one might say that some properties of the grasses’ genomes and their shared evolutionary history have made life easier for scientists. Homology to maize TB1 helped clone¹³⁴ barley INTERMEDIUM-C whose wheat counterpart turned out to play a role in that species’ improvement as well¹⁹⁸. FLOWERING LOCUS T and its various paralogs¹⁹⁹ are important across the board, in the grasses²⁰⁰ as much as in Arabidopsis thaliana²⁰¹. Evolution has been said to tinker with existing molecular machinery²⁰². A recent review in these pages shows as much for land plant evolution²⁰³. Similar co-options of ancestral architectural pathways to meet new adaptive needs have occurred in the grasses. Depending on one’s perspective, this either argues in favor or against the notion of a single genetic system: on the one hand all grasses are descended from a common ancestor, on the other hand there are evolutionary innovations that are private to each species.

The practical merit of the single genetic system is beyond doubt. The assertion in 1993 that cereal genomicists were to face organizational rather than scientific challenges may not square with how the researchers that overcame such obstacles look back on their work. But taken as a rallying cry rather than a prediction, the statement was prescient. Cereal genomics benefited from many collaborative research initiatives, among them those that sequenced genomes^{30,33,103,173} and others that put them into databases^204,205. How much of all this is attributable to the article of Bennetzen and Freeling is hard to know. Regardless, cereal geneticists stand now ready to solve those scientific problems whose solutions need genome sequences in their thousands. When writing their article, Bennetzen and Freeling were motivated in part by their desire to counter the argument that all basic research had better been done in the model plant Arabidopsis thaliana (personal communication, J. Bennetzen). Plant scientists and their funders still struggle to strike a good balance between basic research in docile models, which seem to offer an easier avenue to potentially transformative insights, and research in crops, which are more challenging genetic systems but are obviously closer to translational applications in breeding. The increases in throughput and reduction in the cost of sequencing have partly obliterated that dichotomy by negating the cost advantage of small genomes. We expect genome sequencing will remain at the forefront of research for at least another two decades. Thanks to such things as personalized medicine and GWAS at biobank scale²⁰⁶, the interest in genomics taken by governments and the health care systems they are in charge of will keep up an influx into that discipline of talent and funds for decades to come. Crop genetics, cereal and otherwise, will share in the benefits.