Convergent evolution of plant prickles by repeated gene co-option over deep time

Repeated losses of prickles in cultivated eggplants are caused by LOG gene mutations

Previous mapping studies in Brinjal eggplant showed that the loss of prickles is inherited as a single Mendelian locus designated prickleless (pl) and localized to a genomic interval on chromosome 6 (14). Using a recurrent backcross-derived mapping population between Brinjal eggplant and its prickled wild progenitor S. insanum, we confirmed this result and further fine-mapped pl to a ~100 kb interval containing 10 annotated genes (Fig. 2A). Just outside this interval is the previously proposed pl candidate gene SmelARF18, a putative auxin hormone response transcription factor (15). However, we did not find conspicuous coding region loss-of-function mutations in this gene or in any other gene in the interval. Instead, we identified a probable splice-site mutation in a gene encoding a LONELY GUY (LOG)-family cytokinin biosynthetic enzyme. LOG family members catalyze the final step in the biosynthesis of bioactive cytokinin, a hormone with roles in plant cell proliferation and differentiation (16). In a collection of 23 re-sequenced eggplant accessions (17), we found that this splice-site mutation was consistently associated with the prickleless phenotype, except in one accession, which harbored a 474 bp deletion in exon 6 of the LOG gene (Fig. S2A and Table S3).

The discovery of two independent mutations in the LOG candidate gene suggested that the loss of prickles occurred at least twice in Brinjal eggplant or its wild relatives. It also raised the possibility that mutations in orthologous genes may have caused parallel prickle losses in two other independently domesticated African eggplant species, the Scarlet eggplant (S. aethiopicum) and the Gboma eggplant (S. macrocarpon). Genomic resources for these indigenous crop species are limited. We therefore sequenced and assembled high-quality (QV ≥ 51, completeness > 99) chromosome-scale genomes and generated gene annotations for both species (Fig. 1C, Tables S4 and S5). Using these resources, we found that synteny within the pl locus was retained across all three cultivated eggplant species (Fig. 2B), and that prickleless Scarlet eggplant and Gboma eggplant each harbored different loss-of-function mutations in their respective LOG orthologs (Fig. 2C). Scarlet eggplant carries an indel mutation leading to a frameshift in the coding sequence and a prematurely terminated protein product, while Gboma eggplant carries a splice-site mutation. Reverse-transcriptase polymerase chain reaction (RT-PCR) on cDNA revealed lower expression and multiple mis-spliced transcripts in Brinjal eggplant and a mis-spliced isoform with a retained intron in Gboma eggplant (Fig. 2D). PCR sequencing revealed these transcripts were non-functional (Fig. S2B–D).

To further validate that these independent mutations explain the prickleless phenotypes, we next performed co-segregation analysis in F2 populations derived from intraspecific crosses between prickled and prickleless parents (Fig. 1C and S2). In Scarlet eggplant, homozygosity of the LOG mutant allele co-segregated with the prickleless phenotype in a Mendelian recessive fashion in all examined individuals (χ² = 0.52, df = 1, p = 0.47). In Gboma eggplant we observed segregation patterns that indicated the presence of another unlinked recessive variant independently contributing to prickle loss (χ² = 14.8, df = 1, p < 0.001). Leveraging our newly developed genomic resources, we used a mapping-by-sequencing approach to identify a second large interval associated with the loss of prickles on chromosome 4, which we designated pl2 (Fig. 2E). Importantly, all segregating homozygous mutant individuals at pl on chromosome 6 carried the LOG gene splice-site mutation, although this genotype class was represented at lower-than-expected frequency, likely owing to segregation distortion (Fig. S2G). Finally, we modified existing plant regeneration and transformation protocols to engineer loss-of-function PL alleles using CRISPR-Cas9 genome editing in a prickled accession of S. aethiopicum. Analysis of three independently edited multiallelic transformants revealed suppression of prickle development due to numerous frameshift mutations resulting in PL loss-of-function (Fig. 2F and S2F). Transformants lacking mutations retained prickles. Taken together, these results indicate that PL is the LOG candidate gene, and that at least four independent mutations in this gene enabled repeated selection for losses of prickles in cultivated eggplant species.

Mutations in PL are found in prickleless wild and cultivated species across the Solanum

The clade encompassing all three of the cultivated eggplants diverged ~2 Mya, but prickles in Solanum are more ancient, having emerged over ~6 Mya, and 31 independent losses of prickles have been documented, including in additional domesticated and wild species (11). We tested whether mutations in PL underlie these repeated instances of prickle loss across this broader evolutionary timescale by sampling DNA from additional prickleless species and their prickled close relatives. Because many wild Solanum species are too rare or geographically inaccessible for live-tissue sampling, we used a combination of PCR-amplified exon sequencing from herbarium tissue samples and whole-gene sequencing from available live tissue samples to detect PL mutations (Fig. S3 and Table S6).

Along with the four PL mutations identified in our analysis of prickleless eggplants, we identified an additional 12 allelic mutations predicted to deleteriously affect PL function at the pan-genus level across the spiny Solanum (Fig. 3). These mutations, together with those detected by mapping, were associated with 14 out of the 31 recorded losses of prickles across the genus at the species level (Fig. 1B and Table S7). We then confirmed that these mutations were not found in prickled species from closely-related lineages (Fig. S3). In some cases, we detected the same, although not necessarily ancestrally derived, alleles in separate species. For example, the same splice-site mutation found in prickleless Gboma eggplant, native to and cultivated almost exclusively in Africa, was also identified in the wild species S. donianum, whose native range is in Central America and the Caribbean. Likewise, an identical splice-site mutation was found in both the wild species S. lanzae, from western Africa, and the foraged and sometimes cultivated species S. stramoniifolium, native to northern South America. Such genetic convergence at the allelic level may reflect the high penetrance of PL splicing defects, which can be conferred by mutationally accessible single nucleotide variants (18). Together, our results suggest that PL had an important and repeated genetic role in the convergent losses of prickles across Solanum in the wild and in cultivation. However, loci other than PL may explain prickles losses in lineages for which mutations were not identified.

Repeated co-option of LOG homologs underlies prickle convergent evolution

The finding of recurrent mutations in PL orthologs across the spiny Solanums suggested that co-option of cytokinin biosynthetic gene function was critical to prickle evolution. This spurred us to ask whether genetic convergence through LOG gene co-option extends to other prickled species across flowering plants. We searched the literature for studies associating instances of loss or suppression of sharp outgrowths with specific genomic loci or genes. Strikingly, we found that in the grass family (Poaceae) independent mutant alleles in a LOG homolog from rice (Oryza sativa) and barley (Hordeum vulgare) conferred near complete suppression of epidermally-derived sharp projections commonly called “barbs” but botanically classified as prickles (19, 20). In contrast to the conspicuous, multi-cellular, and lignified prickles found in the Solanum (Fig. S1), grass prickles are homologous structures made of silicified single-cells that develop on awns (Fig. 4A), an outer-whorl structure of the grass flower involved in seed dispersal, along with leaves and spikelets.

Mining additional genomic data for LOG mutations co-occurring with losses of prickles in other eudicot lineages, we found that the fruit-bearing tree crop jujube, commonly known as Chinese date (Ziziphus jujuba), in the Rhamnaceae family, carried two independent mutations (a 1 bp deletion and an exonic insertion) in a LOG homolog in two cultivars with suppressed prickles (also known as stipular spines) (Fig. 4B) (21–23). Importantly, neither mutation was found in Sour jujube (Z. jujuba var. spinosa), the prickled wild progenitor. We also detected an exonic insertion in a LOG homolog of the prickle suppressed ‘Purple Queen’ cultivar of giant spider flower (Teranaya hassleriana), a widely cultivated ornamental plant in the Cleomaceae, a small family within the Brassicales closely related to Arabidopsis (Fig. 4C and Table S6) (24). Finally, in rose, which is a commercially important cultivated cut flower, previous mapping for loci conferring “thornlessness” identified two major effect loci (9), as we found in S. macrocarpon. One of these was a ~2.5 Mb interval containing 156 annotated genes (Fig. 4D), which we found includes a LOG homolog. Though there were no obvious coding or splicing mutations in this LOG, we found that its expression was substantially reduced in the leaves of the mapping parent cultivar ‘Bayse’s thornless’ (Rosa wichuraiana) compared to the prickled parent Rosa chinensis (Fig. 4D). To determine whether this candidate LOG has a role in rose prickle development we used a virus induced gene silencing (VIGS) approach to reduce LOG function (25). In 2/14 rose plants infected with the tobacco rattle virus (TRV) expressing an inverted repeat of LOG RNA, strong suppression of prickle development was observed, while wild-type plants of the same background showed normal prickle development (Fig. 4D).

Taken together, these findings suggested that LOG gene reuse was critical in the independent acquisition of prickles in numerous plant lineages that last shared a common ancestor ~150 Mya. Most sequenced seed plants (angiosperms and gymnosperms) retain multiple LOG gene copies within their genomes. In these taxa, the mean number of annotated LOG genes is 15, inflated by recent polyploid lineages, while the median and mode copy numbers are 12 and 10, respectively (N = 160). To understand the phylogenetic context of LOG co-option and to ask whether repeated co-option occurs in a specific clade of LOG gene family members, we conducted an analysis of LOG family proteins from prickled and prickleless species across the angiosperms (Fig. 4E). Most of the prickle co-option associated LOGs occurred within a specific subclade of the LOG family, suggesting that co-option was more favorable in certain LOG family subclades, particularly those with lineage-specific duplications. However, the LOG homolog co-opted in barley is derived from an earlier diverging subclade (20), indicating that despite a subclade bias, co-option of other LOG family members in different clades may also be associated with prickle evolution.

LOG gene diversification preceded PL co-option in Solanum

Given the recurrent co-option of LOG genes against a backdrop of paralogous gene family members, we sought to better understand the phylogenetic and genomic context that facilitated LOG co-option in Solanum. We examined the conservation of the PL locus, comparing the region across Solanum, including Brinjal eggplant and two additional spiny Solanum species, with tomato (S. lycopersicum), an ancestrally prickleless species that diverged prior to the evolution of the spiny Solanums. We first constructed high-quality chromosome-scale genome assemblies for S. prinophyllum (Forest nightshade; QV = 51.6, completeness > 99) and S. cleistogamum [Desert raisin; QV = 49.8, completeness > 99 (Tables S4 and S5)]. Forest nightshade is endemic to southeastern Australia whereas Desert raisin is native to the arid center of Australia and has been foraged by First Nations people for thousands of years for their sweet, dried berries (26) (Fig. 5A). In our screen for PL mutations across the spiny Solanum we did not identify any naturally occurring PL mutations in the Australian Solanum lineages to which these species belong (Fig. S3). Indeed, neither species has been domesticated and are distinct lineages from the cultivated eggplants.

Leveraging these newly developed genomic resources, we found that synteny at the PL locus was conserved across the Solanum, suggesting that PL was co-opted from a standing syntenic ortholog that existed at least since the divergence of tomato and the spiny Solanums ~14 Mya (Fig. 5B). To better understand ancestral PL function across eudicots, we performed a meta-analysis of gene expression data from Arabidopsis (3154 samples) and tomato (5491 samples), reasoning that shared expression profiles reflect the degree of shared inter-species function (27). We assessed each member of the LOG family for its ability to predict co-expression in every other member of the LOG family in the other species. An area under the receiver operating characteristic (AUROC) curve statistic of 0.93 indicated that SlycPL in tomato is co-expressed with nearly identical genes to that of AthaLOG1 in Arabidopsis, pointing to a conserved function. Likewise, three other tomato LOG gene family members also exhibited strongly conserved co-expression with AthaLOG1 (Fig. 5C). The LOG1 clade, to which PL belongs, has therefore maintained signatures of functional conservation across ~120 My.

Tissue-specific knockdown of AthaLOG1 in the Arabidopsis floral meristem has been shown to impair floral organ initiation, suggesting that AthaLOG1 has critical roles in meristem maintenance, similar to the canonical developmental role for LOGs first reported in rice (16, 28). Therefore, duplication and diversification of the LOG1 subclade in the Solanum may have facilitated PL functional co-option. To explore this hypothesis, we generated an expression atlas for prickled forest nightshade and compared it to matched-tissue gene expression data from tomato and Arabidopsis (Fig. 5D). In Arabidopsis, AthaLOG1 possesses a broad expression pattern across tissues, while Solanum PL and LOG1a have evolved more tissue-biased expression patterns. Compared to its ortholog in tomato, forest nightshade SpriPL has evolved enriched expression in flowers and, consistent with its co-opted function, in developing prickles. Therefore, paralog diversification in the Solanum likely enabled functional co-option and redeployment of ancestral LOG1 clade function in prickle development.

Non-pleiotropic removal of prickles with gene editing

We reasoned that the co-option of PL could facilitate the engineering of agriculturally desirable loss-of-function prickleless mutants, even in the Australian spiny Solanum taxa in which we did not detect naturally occurring PL mutations. The duplication leading to PL and its subsequent expression divergence from its ancestral copy would prevent undesirable pleiotropic effects on other traits. Alternatively, cryptic background modifiers in prickleless lineages may have been required to specifically suppress prickle development, and thus eliminating PL would leave prickles intact, or result in pleiotropy. To distinguish between these two possibilities, we devised a pan-genus CRISPR-Cas9 editing strategy to target PL in Forest nightshade, Desert raisin, and tomato, the latter of which harbors a PL ortholog, likely performing an ancestral function outside of prickle development. Adapting techniques previously established in tomato (29), we developed plant regeneration, transformation, and genome editing for Forest nightshade and Desert raisin, thereby elevating these two species into new Solanum genetic systems. We engineered multiple loss-of-function mutations in PL (pl^CR) in all three species and compared their phenotypes (Fig. 5E). In both Forest nightshade and Desert raisin, pl^CR individuals showed strong suppression of prickle development in all tissues and organs where prickles normally develop in wild type plants, though we observed small sporadic prickles (Figs. 5F,G and S4). Meanwhile, in tomato, Slycpl^CR plants resembled the wild type, likely due to genetic redundancy with SlycLOG1a and possibly other LOG family members prior to the PL co-option event ~6 Mya (Fig. 5H). Fruit morphology and sweetness remained unchanged (Brix sugar content ~30% compared to ~50% in grape raisins and ~9% in cherry tomato) and trichome density and morphology appeared unaffected in WT and pl^CR Desert raisin lines. These results suggest that PL targeting is an effective strategy for first line improvement of harvestability in wild or partially domesticated prickled species bearing edible fruits, including additional locally-important cultivated indigenous Solanum such as vila-vila (S. sisymbriifolium) and naranjilla (S. quitoense) (Fig. 5I, S5, and Table S8).

Discussion

Here we showed multiple, phylogenetically independent reuses of LOG family members in prickle development across 150 My of plant evolution. Studies addressing convergent trait evolution at these timescales have hinted that similar and divergent genetic programs can underpin phenotypic convergence (3, 30). For example, the convergent evolution of echolocation in bats and cetaceans is associated with positive selection on variation in shared orthologous genes (31). In plants, convergent evolution of floral asymmetry has been shown in numerous species to occur by modified expression of the transcription factor encoding gene CYCLOIDEA (32). On the other hand, different loci were reported to underlie convergent adaptation to marine habitats in mammals (33). The repeated use of the same genetic program seen in some traits such as prickles may in part be due to their relative simplicity. Unlike composite traits (34), where selection has the potential to act on many different loci affecting many different organismal systems, convergent traits that arise from selection on fewer potentially relevant loci may exhibit greater genetic convergence by virtue of sheer probability. However, even traits of modest complexity, such as animal eye lenses composed of homomeric crystallins (35), can have many distinct genetic origins, indicating that trait complexity alone cannot fully account for observed patterns of convergent evolution.

Genotype-phenotype convergence may also rely on developmental constraints imposed on morphological innovation, which often depends on the re-purposing of ancestral genetic mechanisms (36, 37). Gene co-option may allow key developmental regulators to take on new roles via non-pleiotropically partitioning gene function, particularly when standing paralog diversity exists. This has been suggested as an explanation for the repeated evolution of limbs, for example, by co-option of Hox genes. We suggest that functionally redundant LOG paralogs that arose through lineage-specific or shared ancestral duplication events may acquire specialized functions, as we found with prickles. The lack of an apparent pl mutant phenotype in tomato, coupled with the strong suppression of prickles in pl mutants in prickled lineages without obvious effects on other traits in consistent with PL functional co-option. Even after co-option in prickle development, LOGs may retain some functional redundancy, as engineered and natural (i.e. rice and barley) LOG mutants still produce sporadic small prickles (Figs. 4A and S4). Even partial paralog redundancy may increase the odds of phenotype-genotype convergence by allowing selection for gains and losses of prickles while avoiding developmental pleiotropy.

Perhaps most importantly, as an essential plant hormone with key developmental functions, cytokinin is well-suited to serve a recurrent role in morphological adaptation. Like the plant hormones auxin and florigen, cytokinins have cell-type and stage-specific effects. For example, beyond its role in promoting cell proliferation in shoot meristems (16), cytokinin contributes to microtubule reorientation in maturing root epidermal cells (38) and promotes growth cessation associated with cell wall stiffening in the root differentiation zone (39). The results presented here endow cytokinin activation by LOGs with a central and repeated role in morphological innovation. This could occur by canonical cytokinin activation of cell proliferation but could also involve cytokinin promotion of the differentiation program leading to the hard, lignified structure of the prickle. Other plant morphological innovations are also controlled by cytokinin-related gene activity. Overexpression of a LOG gene is sufficient to induce the ectopic formation of shoot-borne tubers in axillary meristems in tomato (40), while a dominant mutation in a gene encoding a cytokinin receptor protein induces the ectopic formation of root nodules in the legume Lotus japonicus (41), both of which depend on localized cell proliferation. Unlike typical “master” regulators that often coordinate complex programs, such as floral homeotic genes (42), the repeated loss of prickles reported here relies on an enzymatic gene family involved in the activation of several types of cytokinins. Whether redeployment of such hormone activation genes in new developmental contexts is sufficient to generate morphological novelty warrants further study.

Finally, we propose that targeted gene editing of cytokinin biosynthesis and signaling components, as demonstrated here, is likely a predictable and efficient strategy for eliminating prickles in various flowering plant lineages. This approach is particularly promising for roses, where the labor intensive, manual removal of prickles is a common practice for most cut varieties. Though roses have variable ploidy (43) and genome editing in elite germplasm can be challenging (44), we demonstrated this potential using VIGS to suppress prickle development (Fig. 4D). Beyond the species presented here, the observed subclade bias in LOG homolog co-option will likely aid in selection of LOG genes for site-directed mutagenesis in other taxa. However, in principle, the general role of LOG proteins in cytokinin activation could allow more distantly related LOGs to carry out their role in prickle development, as occurs in barley (Fig. 4A,E). This necessitates consideration of both LOG gene expression and phylogenetic context for targeting prioritization. Overall, continued efforts to unite genetics, genomics, and genome editing across diverse plants, as illustrated in this study, will both advance our ability to track evolutionary changes over a broad range of time scales and empower the engineering of novel phenotypes to expand our use of plant diversity in agriculture.

Methods Summary

For the mapping of pl in S. melongena, previously generated introgressions of prickled S. insanum into the prickleless S. melongena background were screened for small prickle-associated genomic intervals on chromosome 6 (45). An individual with the narrowest identified interval was then selfed and a total of 622 resulting progeny were used for fine mapping of pl by PCR-based marker genotyping.

For genome assembly, high molecular weight DNA was extracted from flash-frozen, dark-treated 4-week-old seedlings. A combination of long-read sequencing (Pacific Biosciences, CA, USA) and optical mapping (Bionano Genomics, CA, USA) data were used for assembly. Sequencing reads from each sample were assembled with hifiasm (46) exact parameters and software version varied between samples based on the level of estimated heterozygosity and are reported in Table S4. In addition, high-throughput chromosome conformation capture (Arima Genomics, CA, USA) was performed for one sample, S. prinophyllum, to finalize scaffolding. Using merqury (47), the final consensus quality (QV) of the assemblies was 51.1, on average.

For genome annotation, orthologs with coverage above 50% and 75% identity were lifted from Heinz v4.0 Heinz v4.0 (48) and Eggplant v4.1 (17) via Liftoff (49) and refined using protein and gene microsynteny support. The completeness of the gene models was determined by assessing single-copy orthologs using BUSCO5 (50).

Plant regeneration and Agrobacterium tumefaciens-mediated transformation of S. prinophyllum and tomato were performed according to (51). The same methods were also used for S. aethiopicum and S. cleistogamum with two modifications. For S. cleistogamum, plant regeneration, the medium was supplemented with 0.5 mg/L zeatin instead of 2 mg/L and for the selection medium, 75 mg/L kanamycin was used instead of 200 mg/L. For S. aethiopicum, the protocol was the same as for S. cleistogamum, except the fourth transfer of transformed plantlets was performed onto media supplemented with 50 mg/L kanamycin.

Supplementary Material

Data and materials availability:

Short-read sequencing data and genome assemblies reported in this paper are deposited at the NCBI Short Read Archive under BioProject ID PRJNA1073673. All other data are available in the main text or the supplementary materials. Plant materials used in this study are available upon request.