A reference-grade genome of the xerophyte Ammopiptanthus mongolicus sheds light on its evolution history in legumes and drought-tolerance mechanisms

Abstract

Plants that grow in extreme environments represent unique sources of stress-resistance genes and mechanisms. Ammopiptanthus mongolicus (Leguminosae) is a xerophytic evergreen broadleaf shrub native to semi-arid and desert regions; however, its drought-tolerance mechanisms remain poorly understood. Here, we report the assembly of a reference-grade genome for A. mongolicus, describe its evolutionary history within the legume family, and examine its drought-tolerance mechanisms. The assembled genome is 843.07 Mb in length, with 98.7% of the sequences successfully anchored to the nine chromosomes of A. mongolicus. The genome is predicted to contain 47 611 protein-coding genes, and 70.71% of the genome is composed of repetitive sequences; these are dominated by transposable elements, particularly long-terminal-repeat retrotransposons. Evolutionary analyses revealed two whole-genome duplication (WGD) events at 130 and 58 million years ago (mya) that are shared by the genus Ammopiptanthus and other legumes, but no species-specific WGDs were found within this genus. Ancestral genome reconstruction revealed that the A. mongolicus genome has undergone fewer rearrangements than other genomes in the legume family, confirming its status as a “relict plant”. Transcriptomic analyses demonstrated that genes involved in cuticular wax biosynthesis and transport are highly expressed, both under normal conditions and in response to polyethylene glycol-induced dehydration. Significant induction of genes related to ethylene biosynthesis and signaling was also observed in leaves under dehydration stress, suggesting that enhanced ethylene response and formation of thick waxy cuticles are two major mechanisms of drought tolerance in A. mongolicus. Ectopic expression of AmERF2, an ethylene response factor unique to A. mongolicus, can markedly increase the drought tolerance of transgenic Arabidopsis thaliana plants, demonstrating the potential for application of A. mongolicus genes in crop improvement.

This study reports a high-quality chromosome-level genome assembly for the xerophytic legume shrub Ammopiptanthus mongolicus. Comparative genomic analyses provide insights into the genome expansion and chromosome evolution of this relict plant. Analyses of its cuticular wax biosynthesis and phytohormone signaling pathways shed light on its mechanisms of adaptation to desert environments.

Introduction

The genus Ammopiptanthus (family Leguminosae or Fabaceae, subfamily Papilionoideae) comprises only two species, Ammopiptanthus mongolicus and Ammopiptanthus nanus (Cheng, 1959). The model species, A. mongolicus (Figure 1A), is distributed throughout the southern Gobi Desert and adjacent areas in central Asia, whereas A. nanus is restricted to Kashgar, Xinjiang Province, China (Zhao and Zhu, 2003). Geographical research suggests that deserts have existed in eastern central Asia for approximately 22 million years (Xie and Yang, 2012). The genus Ammopiptanthus, an endemic plant, is considered a relict descendant of the ancient Tethyan flora that evolved to survive in desert environments (Xie and Yang, 2012). Despite growing in extremely arid and cold environments, Ammopiptanthus spp. are evergreen broadleaf shrubs (Zhao et al., 2016a). These evergreen shrubs play an essential role in maintaining the semi-desert ecosystem, as they can obstruct drifting sand and prevent further desertification (Liu and Qiu, 1982). However, the harsh environment of the Gobi Desert leads to low seed germination rates and increased growth pressure. Therefore, both Ammopiptanthus species have been listed as endangered and second-class state-protected wild plants in China (Ge et al., 2005).

Figure 1.

Morphology, karyotyping, Hi-C interaction heatmap, and genomic landscape of A. mongolicus.

(A) A population of A. mongolicus grown in a field on Alxa Plateau, Inner Mongolia, China (July 2015).

(B) Karyotyping of A. mongolicus by fluorescence in situ hybridization (FISH) using probes from repeat sequences of rDNAs (green, 18S rDNA; red, 5S rDNA). Scale bar, 5 μm.

(C) Assembly of A. mongolicus chromosomes using Hi-C data. Bin size was set to 200 kb when plotting the Hi-C interaction heatmap.

(D) Genomic landscape of A. mongolicus. i, intragenomic collinear blocks between or within chromosomes, with the most dramatic one highlighted in pink; ii, nine pseudo-chromosomes (unit, Mb); iii, GC content in each 500-kb bin of the genome; iv, gene density along the chromosomes, expressed as the number of genes per bin; v, repeat ratio of each bin; vi, number of simple short repeats (SSRs) in each bin; vii, percentage of LTR_Copia among TEs; viii, percentage of LTR_Gypsy among TEs. All density or ratio information was determined in non-overlapping 500-kb windows.

Phylogenetic analyses based on the nuclear ribosomal internal transcribed spacer (nrITS) marker and the plastid matK gene have shown that Ammopiptanthus is monophyletic and has a sibling genetic relationship with a clade including Sophora and Ammodendron (Xie and Yang, 2012). A study of chloroplast intergenic spacers and ITS markers estimated the divergence time between A. mongolicus and A. nanus at approximately 0.77 million years ago (mya) (Su et al., 2016). Our recent chloroplast and mitochondrial sequence analyses revealed that these two species have almost the same organelle gene content, with only a few insertions and deletions in non-coding regions, suggesting that they diverged very recently (Feng et al., 2017, 2019).

Mechanisms of abiotic stress tolerance in this genus are receiving increasing attention. Morphological studies have shown that its leaf surfaces are covered with an exceptionally thick waxy cuticle (up to 18 μm) that helps to limit transpiration and maintain plant water status (Han and Li, 1992; Li et al., 2023). There are dense silvery trichomes on both leaf surfaces that protect against heat and cold (Liu et al., 2004). A. mongolicus leaves also have a lower stomatal density than leaves of Arabidopsis (Li et al., 2015), further reducing the rate of transpiration. Notably, the palisade mesophyll of A. mongolicus has nine or 10 cell layers, resulting in succulent leaves and thus an increased water storage capacity. Gene expression profiling of this genus under drought, cold, and salinity stress has identified hundreds of abiotic-stress-inducible genes (Zhou et al., 2012; Liu et al., 2013; Pang et al., 2013, 2015; Wu et al., 2014; Gao et al., 2015; Yang et al., 2022b). However, the exact stress-tolerance mechanisms of this unique desert flora remain unclear.

Legumes (family Fabaceae) form the third largest angiosperm family in the plant kingdom, with 751 genera and approximately 19 500 species (Bruneau et al., 2013). With recent advances in genome sequencing technologies, genomes of more than 20 legume species have been sequenced (Wang et al., 2017; Bertioli et al., 2019; Xie et al., 2019; Zhuang et al., 2019; Xu et al., 2020; Yang et al., 2022a; Chang et al., 2022). These sequenced genomes are invaluable resources for functional studies of individual plant species and comparative and evolutionary studies of eudicots. However, the sequenced genomes represent only a small portion of the large legume family. An 823-Mb draft genome of A. nanus was recently assembled using only Pacific Biosciences (PacBio) sequencing data (Gao et al., 2018); it provides basic information on the genome but is insufficient for a detailed understanding of its genomic features, downstream functional studies, and comparative genomic analyses.

Here, we assembled a high-quality reference-grade genome sequence for A. mongolicus by combining Illumina shotgun sequencing, PacBio long-read sequencing, Bionano optical mapping, and high-throughput chromosome conformation capture (Hi-C) sequencing. The final assembly is 843.07 Mb in length, with ∼98.7% anchored to nine chromosomes. Transcriptome sequencing and gene expression analyses revealed that ethylene signaling and cuticular wax biosynthesis play important roles in the drought tolerance of this desert plant. Our results provide new insights into the genomic characteristics, evolutionary history, and drought-tolerance mechanisms of A. mongolicus.

Results

Genome sequencing and assembly

Previous karyotyping studies have revealed that the genus Ammopiptanthus is diploid (2n = 2X = 18) (Feng and Song, 1988), and this was confirmed by our fluorescence in situ hybridization (FISH) analysis, which clearly showed nine pairs of chromosomes in A. mongolicus (Figure 1B). Two approaches were used to predict the genome size of A. mongolicus. The size predicted by flow cytometry was ∼979 Mb (Supplemental Figure 1), and that estimated from 60 Gb of next-generation sequencing (NGS) data (Illumina HiSeq 2000) from a 180-bp-insert-size DNA library was ∼926 Mb, with a GC content of 37% and a repetitive sequence content of 66% (Supplemental Figure 2). Deep sequencing of the A. mongolicus genome was performed using both Illumina NGS technology and single-molecule real-time (SMRT) sequencing on the PacBio Sequel II platform. Illumina sequencing based on multiple paired-end DNA libraries generated ∼395 Gb of data, representing a sequencing depth of >400× (Supplemental Table 1). PacBio sequencing generated 214 Gb of long reads (average length = 16.9 kb) (Supplemental Table 2), representing a sequencing coverage of >100×. All PacBio long reads were self-corrected and used for assembly with Canu v2.1 (Koren et al., 2017), and 114 Gb of shotgun reads from NGS libraries (insert sizes: 200, 500, and 800 bp) were used for error correction with Pilon v1.22 (Walker et al., 2014). To further increase genome quality, we performed Bionano optical mapping sequencing to enable better scaffolding. In total, 130.35 Gb of data were generated by the Bionano Saphyr system (N50 = 328.5 kb for reads >150 kb, and N50 = 214.5 kb for reads >20 kb; Supplemental Tables 3 and 4). After scaffolding, the contig N50 length was increased to 3.13 Mb. Super-scaffolding (or chromosome-level scaffolding) was achieved using Hi-C sequencing, which generated 316.42 million paired-end reads (94.93 Gb of clean data; Supplemental Table 5). Approximately half (49.5%) of the Hi-C reads (156.70 million paired-end reads) were uniquely mapped to the genome, and 30.7% (97.18 million paired-end reads) turned out to be valid interaction pairs (Supplemental Table 6). The 3D-DNA Hi-C assembly pipeline was then used to cluster scaffolds with the agglomerative hierarchical clustering method (Dudchenko et al., 2017). After clustering and orientation, the scaffolds were clustered into nine chromosomes or superscaffolds (Figure 1C). The final assembly was 843.07 Mb in length, including 498 scaffolds and 1374 contigs. The N50 length of the scaffolds was 92.57 Mb, and that of the contigs was 2.90 Mb. The lengths of the nine chromosomes ranged from 70.9 to 108.8 Mb (Supplemental Table 7), with a total length of 831.99 Mb, accounting for ∼98.7% of the overall assembly (Table 1). Notably, all telomere sequences were found at both ends of the nine chromosomes, and the identified telomere repeat motif in A. mongolicus (TTTAGGG)n is also conserved among most terrestrial plants, including Arabidopsis thaliana. The chromosome sizes of the final assembly (Supplemental Table 7) were consistent with the karyotyping results obtained using FISH (Figure 1B).

Table 1.

Global statistics of the A. mongolicus genome and gene predictions.

	Number or ratio	Size
Assembly feature
Estimated genome size	–	926 Mbp
Total chromosomes	9	831.99 Mbp
Chromosome N50	–	92.57 Mbp
Longest chromosome	–	108.85 Mbp
Total scaffolds (>1000 bp)	498	843.07 Mbp
Scaffold N50	–	92.57 Mbp
Longest scaffold	–	108.85 Mbp
Total contigs	1374	842.10 Mbp
Contig N50	–	2.90 Mbp
Longest contig	–	14.38 Mbp
GC content	36.77%	–
Genome annotation
Total repetitive sequences	70.71%	596.16 Mbp
Transposable elements	69.55%	586.35 Mbp
Gene models	47 611	222.31 Mbp
Non-coding RNAs	5701	1.17 Mbp

The quality of the assembled genome was assessed using Benchmarking Universal Single-Copy Orthologs (BUSCOs). The BUSCO database provides 255 single-copy orthologous eukaryotic genes for evaluation of genome quality (https://busco.ezlab.org, eukaryota_odb10); 250 of these BUSCOs (98.1%) were completely mapped to the A. mongolicus genome, and two additional BUSCOs (0.8%) were partially mapped. BUSCO also provided 5366 single-copy orthologous genes from legumes (fabales_odb10); 4937 of these single-copy genes (92.0%) were completely represented in the A. mongolicus genome, and 92 more BUSCOs (1.7%) were fragmented in the genome (Supplemental Table 8).

Genome annotation

Repetitive sequences were annotated using homology-based and ab initio methods. The genome of A. mongolicus is highly repetitive, with repetitive sequences accounting for 70.71% of its total genome sequence (Supplemental Table 9). Approximately half of the repeats (35.39% of the genome) matched well with known transposable elements (TEs). As reported for many land-plant genomes, long-terminal-repeat retrotransposons (LTR-RTs) are the dominant TE type in A. mongolicus (∼84.64% of total TEs). Most LTR-RTs were from the Ty1/Copia and Ty3/Gypsy superfamilies, which accounted for 9.35% and 20.14% of the genome, respectively. Non-TE repeats (e.g., simple short repeats, transfer RNA [tRNAs], small nuclear RNAs [snRNAs], and rRNAs) constituted 1.16% of the genome (Supplemental Table 9).

Protein-coding genes were predicted using a combination of homology-based, transcriptome-based, and ab initio gene prediction strategies, which yielded 47 611 protein-coding genes (Supplemental Table 10) and 3542 non-coding RNAs (Supplemental Table 11). The average coding sequence (CDS) length of protein-coding genes was 1488 bp, with an average of 5.5 exons per gene. The total length of the genic regions was 222.31 Mb (26.37% of the genome) (Supplemental Table 12). Functional annotation assigned putative functions to 91.44% of the protein-coding genes (43 536) on the basis of sequence similarity searches against public databases (Supplemental Table 13). The 3542 non-coding RNAs included 1212 tRNAs, 380 rRNAs, 1779 snRNAs, and 171 microRNAs (Supplemental Table 11). The GC content and repeat ratio were higher in centromeric and pericentromeric regions than in the chromosome arms, whereas protein-coding genes were distributed mainly in the chromosome arms (Figure 1D). Among the TEs, simple sequence repeats (SSRs) had distribution patterns similar to those of protein-coding genes, whereas LTR-Gypsy sequences were enriched in centromeric and pericentromeric regions, and LTR-Copia sequences were more dispersed throughout the genome (Figure 1D). These distribution characteristics of the A. mongolicus genome are consistent with those of the archetypal eukaryotic genome, which is segregated into A/B compartments, with A compartments (i.e., active) located in chromosome arms and B compartments (i.e., inactive) in centromeric and pericentromeric regions (Wang et al., 2021).

Comparative genomic analyses revealed ancient polyploidy of the common legume ancestor but no species-specific genome duplications in Ammopiptanthus

Legumes evolved from a common tetraploid ancestor approximately 58 mya (Young et al., 2011). The ancestor underwent two polyploidization events at 130 and 58 mya, followed by more recent events in individual legume lineages, such as the soybean-specific tetraploidy event at ∼13 mya (Schmutz et al., 2010; Wang et al., 2017). As a result, numerous syntenic blocks were detected between the A. mongolicus genome and those of other closely related legumes such as soybean (Glycine max) and barrel medic (Medicago truncatula) (Figure 2A). For each syntenic block associated with barrel medic, there is usually one corresponding block in A. mongolicus and two in soybean (Figure 2A). The larger number of intragenomic syntenic blocks in soybean is indicative of its more recent specific polyploidization event at 13 mya. In addition, 325 collinear blocks were identified when the A. mongolicus genome was compared with itself (Supplemental Table 13), reflecting the ancient polyploidizations of its legume ancestors. On average, each collinear block contained 13.81 paralogous gene pairs; the most dramatic collinear block was found in Chr01, with 236 paralogous gene pairs (Supplemental Table 14; Figure 1D). The divergence times between homologous gene pairs were predicted from the four-fold synonymous third-codon transversion (4DTv) rates. The 4DTv distribution of paralogous gene pairs from intragenomic collinear blocks in the soybean genome showed peaks at 0.4, 0.18, and 0.04, corresponding to three whole-genome duplication (WGD) events at 130, 58, and 13 mya, respectively (Figure 2B). However, paralogous gene pairs in intragenomic collinear blocks of the A. mongolicus genome showed only two significant peaks around 0.4 and 0.18, indicating that no WGD events had occurred since 58 mya, and similar results were obtained for A. nanus (Figure 2B). The 4DTv distributions of paralogous genes in A. mongolicus and A. nanus almost completely overlapped (Figure 2B). When considering orthologous gene pairs from intergenomic collinear blocks between A. mongolicus and other legumes, the 4DTv distribution of orthologous gene pairs between A. mongolicus and A. nanus showed a peak at 0.01, suggesting a divergence time for these two species of ∼3.25 mya (Figure 2C). The ratio of the WGD time/4DTv peak values of paralogous gene pairs in G. max was used to calibrate the divergence time. Synonymous substitution rates (Ks) were also calculated for syntenic gene pairs in both intragenomic (Supplemental Figure 3A) and intergenomic regions (Supplemental Figure 3B), revealing patterns consistent with those of the 4DTv distributions (Figure 2B and 2C). The Ks analysis thus supports the conclusion that no species-specific genome duplications have occurred in the genus Ammopiptanthus since 58 mya.

Figure 2.

Synteny, orthology, and polyploidization analyses of A. mongolicus and closely related legume species.

(A) Intergenomic syntenic blocks identified among A. mongolicus, M. truncatula, and G. max. The connections that are specific to chromosome 1 of M. truncatula are highlighted. The purple ribbons indicate syntenic blocks shared by M. truncatula and A. mongolicus, and the red and green ribbons denote those shared by M. truncatula (barrel medic) and G. max (soybean); each syntenic block in M. truncatula usually corresponds to two blocks in G. max.

(B) Four-fold synonymous third-codon transversion (4DTv) rates of paralogous gene pairs in intragenomic collinear regions of selected plant genomes. The numbers I–III indicate WGD or triplication events during evolution. I, γ event (whole-genome triplication) shared by eudicots ∼130 million years ago (mya); II, WGD event shared by legumes 58 mya; III, WGD event specific to the genus Glycine 13 mya. Amo, A. mongolicus; Ana, A. nanus; Cca, C. cajan; Gma, G. max; Mtr, M. truncatula.

(C) 4DTv distributions calculated from orthologous gene pairs in intergenomic collinear regions between different species.

(D) Numbers of orthologous gene groups among selected plants.

The phylogenetic relationships among A. mongolicus and other legumes

To better understand the evolutionary history of A. mongolicus, we performed phylogenetic analyses of 15 closely related legumes along with A. mongolicus (Supplemental Table 15). Orthologous gene groups were identified on the basis of sequence similarity using the OrthoFinder pipeline and were clustered using the Markov Cluster Algorithm (Emms and Kelly, 2018). A Venn diagram highlighted the overlaps of orthologous groups among A. mongolicus, three other legumes, and A. thaliana (Figure 2D). The enriched biological process Gene Ontology (GO) terms in the A. mongolicus-specific gene groups included DNA/RNA-mediated transposition, regulation of root development, meristem growth, and aldonic acid metabolic process (Supplemental Table 16), which may enhance survival of the species or adaptation to severe abiotic stress.

A phylogenetic analysis based on 322 single-copy orthologous genes from these 15 legume species confirmed their phylogenetic relationships (Figure 3A). The topology of the phylogenetic tree aligns with previous analyses of legume species (Wang et al., 2006, 2018). The estimated divergence time between A. mongolicus and A. nanus was 2.4 mya (Figure 3A), which was slightly more recent but similar to the estimate from 4DTv analysis (3.25 mya) (Figure 2C). The divergence time between A. mongolicus and Lupinus angustifolius was calculated to be 27.79 mya, predating the whole-genome triplication event in the genus Lupinus at around 25 mya (Hane et al., 2017).

Figure 3.

Inference of divergence times and MRCAs of A. mongolicus.

(A) A phylogenetic tree inferred from single-copy orthologous genes of legume species. Numbers on the nodes represent the divergence times from the present (mya). The divergence time between G. max and P. vulgaris (20.1 [19.0–21.0] mya) was used as the standard for calibration (Zheng et al., 2016). Polyploidization events are marked by red stars on the tree. The background colors of different lineages represent different legume clades. A. thaliana was used as the outgroup taxon.

(B) Reconstructed ancestral genomes of six legume species. The phylogenetic tree and timescale were redrawn from (A). WGD, whole-genome duplication (2×, tetraploidy; 3×, hexaploidy); CRE ratio, completely rearranged endpoint; EA, estimated accuracy; AEA, accumulated estimated accuracy. The numbers of shuffling events (fissions and fusions) are labeled beside each branch. The shuffling events marked by an asterisk (∗) in peanut were calculated against ancestor 1.

We also examined the expansion and contraction of gene families. The orthologous genes were classified into two groups: group I, which included genes that expanded/contracted in both A. mongolicus and A. nanus, and group II, which included genes that expanded/contracted only in A. mongolicus. Group I contained 155 expanded gene families and 31 contracted families, representing 3749 expanded genes in A. mongolicus and 1311 expanded genes in A. nanus. Group II comprised 219 expanded orthologous groups and 128 contracted groups (Supplemental Figure 4). Functional annotation of the expanded genes indicated that most were TE related, such as reverse transcriptases, the DNA/RNA polymerase superfamily, and helicases. Other gene groups included senescence-associated proteins, UDP-glycosyltransferases, the NAD(P)-binding Rossmann-fold superfamily, the cytochrome P450 2C family, the pectin lyase-like superfamily, disease-resistance proteins, and osmotins (Supplemental Tables 17 and 18). Members of the cytochrome P450 family may be involved in cuticular wax biosynthesis, which helps to limit transpiration and reduce water loss (Pandian et al., 2020; Zhang et al., 2020). Osmotins, members of the pathogenesis-related protein family, may be critical for adaptation to osmotic stress (Bashir et al., 2020). The expanded gene groups were enriched in GO terms associated with transpositional recombination, anatomical structure homeostasis, ATP biosynthetic pathways, proton transmembrane transport, and abiotic stimuli (Supplemental Figure 5; Supplemental Tables 19 and 20).

Further analyses of single-copy orthologous genes revealed that 719 genes were under positive selection in A. mongolicus. Functional annotations indicated that some were related to stress responses, such as osmotin-like proteins, the ATP-binding cassette transporter family, ATPases, membrane transporter proteins, cadmium-transporting ATPases, sugar transporters, peroxidases, and genes related to long-chain fatty acid biosynthesis (Supplemental Table 21), and they were enriched in GO terms such as ion transmembrane transport, peroxisomal transport, hormone biosynthesis, and membrane lipid biosynthetic processes (Supplemental Table 22; Supplemental Figure 6). These positively selected genes may have played important roles in the evolution of abiotic stress resistance and adaptation in A. mongolicus.

Reconstruction of ancestral legume genomes

The chromosome-level genome assembly enabled us to characterize the evolution of the A. mongolicus genome from those of its legume ancestor(s). To this end, we selected A. mongolicus and five other legumes—soybean (G. max), common bean (Phaseolus vulgaris), barrel medic (M. truncatula), white lupin (Lupinus albus), and a wild diploid peanut species (Arachis ipaensis)—and reconstructed the most recent common ancestors (MRCAs) of these legume species. Protein-coding genes that could be anchored to the chromosomes of these six species were aligned and clustered using OrthoFinder. A total of 26 038 ortho-groups were identified, each containing an average of 8.6 orthologous genes (Supplemental Table 23), and DRIMM-synteny analyses revealed 220 syntenic blocks shared by these plants (Supplemental Tables 24 and 25). Six intermediate ancestral genomes (Figure 3B) were computed using the Inferring Ancestor Genome Structure (IAGS) package (Gao et al., 2022). Overall, a large number of rearrangements (including both fusions and fissions) between a species and its ancestor, as in the cases of white lupin and soybean, indicates the occurrence of paleopolyploidy and post-polyploid diploidization. Compared with the other legumes in the analysis, A. mongolicus had the fewest rearrangements, with only 22 fissions and 22 fusions between it and ancestor 1. This finding provides strong molecular support for the claim that A. mongolicus is a “relict plant” or “living fossil” (Xu et al., 2002).

LTR-RT bursts caused the expansion of the A. mongolicus genome

TEs are important driving forces in genome evolution, as they can induce genomic rearrangements such as translocations, inversions, and duplications through recombination (Pereira and Ryan, 2019). We therefore performed a comprehensive analysis of repetitive sequences in 17 legume species together with A. thaliana (Supplemental Table 15). In general, legume genomes had a much higher proportion of repetitive sequences than A. thaliana (Figure 4A), and the A. mongolicus genome had the second highest proportion (70.71%), only slightly lower than that of A. ipaensis (71.65%). Most repetitive sequences in the legume genomes were TEs. In A. mongolicus, 35.39% of the genome consisted of intact TEs, with the dominant type being LTR-RTs (∼84.64% of total TEs); 9.35% of the genome was occupied by Ty1/Copia LTR-RTs and 20.14% by Ty3/Gypsy LTR-RTs (Figure 4A). The actual proportions of Ty1/Copia and Ty3/Gypsy LTR-RTs may be even higher, because many older LTR-RTs have experienced severe deletions or fragmentation during genome evolution through processes such as unequal homologous recombination and illegitimate recombination (Du et al., 2010). Intact LTR-RTs were identified on the basis of their structural features using LTR_retriever; they included 3054 Ty1/Copia, 3019 Ty3/Gypsy, and 1076 non-autonomous LTR-RTs, with an average length of 7.65 kb (Supplemental Figure 7). Insertion time analysis indicated that LTR-RTs have greatly expanded in A. mongolicus during the last 2.5 million years (Figure 4B). A recent systematic survey of LTR-RTs classified Ty1/Copia and Ty3/Gypsy into 16 and 14 lineages, respectively. The 16 Ty1/Copia lineages include Ale, Alesia, Angela, Bianca, Bryco, Lyco, Gymco-I–IV, Ikeros, Ivana, Osser, SIRE, TAR, and Tork, and the 14 Ty3/Gypsy lineages include chromoviruses (CRM, Chlamyvir, Galadriel, Tcn1, Reina, and Tekay) and non-chromovirus elements (Athila, Tat-I,II,III, Ogre, Retand, Phygy, and Selgy) (Neumann et al., 2019). In A. mongolicus, the Ty1/Copia type included eight lineages (Ale, Angela, Bianca, Ikeros, Ivana, SIRE, TAR, and Tork) (Figure 4C), although the Bianca lineage was represented by only three LTR-RTs, echoing the case of soybean, in which the entire Bianca lineage has been lost (Du et al., 2010). The Ty3/Gypsy LTR-RTs of A. mongolicus comprised six lineages: Athila, CRM, Ogre, Reina, Retand, and Tekay (Figure 4D). The Ty1/Copia LTR-RTs were further sorted into 458 clusters based on their reverse transcriptase (RT) domain sequences using the cd-hit package, and the Ty3/Gypsy LTR-RTs were sorted into 336 clusters (Figure 4E and 4F). The results of phylogenetic analyses of these two LTR-RT groups were consistent with the classification using TEsorter and provide detailed features of TE expansion during the evolution of the A. mongolicus genome.

Figure 4.

Classification of repetitive sequences in legumes and evolution of TEs in A. mongolicus.

(A) Percentages of transposable elements (TEs): Ty1/Copia, Ty3/Gypsy, other TEs (unknown types), non-TE repeats (microsatellites, low-complexity sequences, rRNAs, snRNAs, etc.), and non-repeat sequences in the genomes of the respective species.

(B) Insertion times of long-terminal-repeat retrotransposons (LTR-RTs), calculated as T = K/(2r), where T is insertion time, K is the number of base substitutions per site between the two LTRs of an LTR-RT, and r is the synonymous nucleotide substitution rate, expressed as the number of synonymous mutations per site per year. For legumes, r is 7e−9 (Jing et al., 2005).

(C and D) Clade-level classification of LTR-RTs using TEsorter (https://github.com/zhangrengang/TEsorter). (C) Ty1/Copia. (D) Ty1/Gypsy.

(E) Classification of Ty1/Copia into sub-lineages.

(F) Classification of Ty3/Gypsy into sub-lineages. The phylogenetic trees in (E) and (F) were constructed from the RT domains of the LTR-RTs.

To investigate the potential contribution of the LTR-RT expansion to the drought tolerance of A. mongolicus, we performed a comprehensive GO analysis. Genes that overlapped with or were in close proximity to LTR-RTs were identified, resulting in the discovery of 7125 LTR-RT-related genes (Supplemental Table 26). GO enrichment analysis revealed that these genes were significantly enriched in various biological processes, including retrotransposition, DNA transposition, and metabolic processes related to D-gluconate, sphingolipids, and aldonic acid (Supplemental Figure 8A). In terms of gene expression, 75% of the LTR-RT-related genes showed no transcriptional activity, whereas 50% of non-LTR-RT-related genes were unexpressed (Supplemental Figure 8B). However, there were no distinct patterns between LTR-RT-related and non-LTR-RT-related genes in most leaf samples (Supplemental Figure 8C). Further studies are needed to fully understand the involvement of LTR-RTs and their associated genes in the drought tolerance of A. mongolicus.

The involvement of cuticular wax in the drought-stress tolerance of A. mongolicus

Leaves of A. mongolicus are rhombic-elliptic or broad-lanceolate in shape, measuring 2.0–3.8 cm in length and 0.6–2.7 cm in width (Han and Li, 1992). Both surfaces of the leaves are coated with a waxy cuticle that can reach a thickness of 18 μm (Han and Li, 1992). This cuticle thickness exceeds that observed in other desert plants and may therefore provide significant protection from excessive transpiration. Cuticular waxes are mixtures of very-long-chain aliphatic compounds such as fatty acids, primary alcohols, acyl esters, alkanes, aldehydes, secondary alcohols, and ketones (Dimopoulos et al., 2020). Very-long-chain-fatty-acids (VLCFAs) are produced from fatty acyl-coenzyme A (CoA)-thioesters through a process catalyzed by the multienzyme fatty acid elongase (FAE) complex. To produce aliphatic compounds, VLCFA-CoAs are modified by enzymes in the endoplasmic reticulum (Figure 5). By combining the genome sequence and transcriptomic data, we identified genes related to cuticular wax biosynthesis and transport in A. mongolicus and analyzed changes in their expression during polyethylene glycol (PEG)-induced dehydration stress. Overall, these genes had higher expression levels in leaves than in roots, and most were induced by PEG treatment, including KCR (encoding beta-ketoacyl reductase, a member of the FAE family), WSD1 (encoding wax ester synthase and diacylglycerol acyltransferase), CER3 (encoding ECERIFERUM 3, a transmembrane protein with similarity to the sterol desaturase family), and LTPs (encoding lipid transfer proteins). LTPs, together with ATP-binding cassette transporters, transport cuticular wax mixtures (Figure 5; Supplemental Table 27). Interestingly, many of the cuticular wax biosynthetic genes had high copy numbers and were tandemly duplicated in the A. mongolicus genome, including the KCR genes (Chr03.g17075, Chr03.g17076, Chr04.g22540, Chr04.g22542) and CER4 genes (Chr04.g18365, Chr04.g18366, Chr07.g34606, Chr07.g34607, and Chr07.g34608). Most notably, the MAH1 genes (encoding midchain alkane hydroxylases) had more than 100 copies in the A. mongolicus genome, and some were also tandemly duplicated (Figure 5; Supplemental Tables 27 and 28).

Figure 5.

Genes involved in cuticular wax biosynthesis and transport in A. mongolicus and their responses to simulated drought stress.

Heatmaps show the expression levels of genes involved in the cuticular wax biosynthesis and transport pathways upon drought treatment simulated by polyethylene glycol (PEG) at 0, 2, 24, and 48 h in leaves and roots of 2-month-old A. mongolicus plants. Gene expression values were normalized as log10(FPKM + 1). LACS, long-chain fatty acid acyl-CoA synthetase; FAE, fatty acid elongase; VLFCA, very-long-chain fatty acids; FAR, fatty acyl-CoA reductase; CER, eceriferum; WSD, wax ester synthase and diacylglycerol acyltransferase; MAH, midchain alkane hydroxylase.

Ethylene and other stress phytohormones play important roles in the stress tolerance of A. mongolicus

In the drought-responsive signaling pathway, messenger molecules such as stress hormones and/or second messengers (e.g., inositol phosphates and reactive oxygen species) that regulate intracellular Ca²⁺ levels are released upon stress signal perception. Changes in Ca²⁺ levels then initiate phosphorylation cascades, activating diverse transcription factors that act upon downstream stress-tolerance effector genes (Xiong et al., 2002). In our transcriptomic study, we observed significant induction of genes related to biosynthesis or signaling of several stress hormones, including ethylene, abscisic acid (ABA), and jasmonic acid (JA) (Supplemental Table 29). Among these hormone-related genes, two ethylene biosynthetic genes, Chr06.g29235 and Chr06.g29237, which are homologous to Arabidopsis 1-Aminocyclopropane-1-Carboxylic Acid Oxidase (ACO), caught our attention because their expression was induced by more than 39.8-fold and 4-fold, respectively, after 24 h of PEG treatment (Figure 6A). In addition to the two ACO genes, several other ethylene biosynthesis and signaling genes were also significantly upregulated by PEG-induced dehydration stress, particularly at 24 h (Figure 6B). For example, Chr01.g00583, Chr06.g30417, and Chr01.g01008, three genes homologous to Arabidopsis 1-Aminocyclopropane-1-Carboxylic Acid Synthase (ACS), and genes encoding ethylene receptors (ETR1 and ETR2), key mediator Ethylene Insensitive 2 (EIN2), key transcription factors (EIN3, EIN3-like 3 [EIL3], and EIN4), and downstream ethylene response factors (ERFs) were all activated (Figure 6B).

Figure 6.

Ethylene-related genes play important roles in stress resistance of A. mongolicus.

(A) Both transcriptomic and qRT–PCR analyses indicated marked induction of the ethylene biosynthetic genes EFE1/ACO and EFE2/ACO. Data are from 2-month-old A. mongolicus plants treated with 15% PEG for 24 h. The housekeeping gene AmEIF was used as an internal control in qRT–PCR analysis, and the significance of changes was evaluated by Student’s t-test (∗∗p < 0.01).

(B) Expression profiles of genes related to ethylene biosynthesis and signaling in leaves of 2-month-old A. mongolicus plants during PEG treatment. The expression levels of each gene at different treatment time points are displayed in the heatmaps by converting FPKM values into Z scores. ACS, ACC synthase; ACO, ACC oxidase; ETR, ethylene receptor; EIN, ethylene insensitive; EIL, EIN3-like; ERF, ethylene response factors; ERT, ethylene-responsive transcriptional coactivator-like protein; EGY, ethylene-dependent gravitropism-deficient and yellow-green-like; AP2, ethylene-responsive element-binding protein 2; ERS, ethylene response sensor 1.

(C) Microsynteny analysis of collinear regions of the ERF2 gene and its flanking regions among A. mongolicus, A. nanus, M. truncatula, C. cajan, G. max, and V. vinifera. Blue and green boxes represent gene loci in the chromosome regions. Gray ribbons represent orthologous gene relationships among the six species. A red ribbon highlights the ERF2 locus in A. mongolicus and A. nanus. ALA2, Aminophospholipid ATPase 2 (a homolog of Arabidopsis AT5G44240); TAT2, Tyrosine aminotransferase 2 (a homolog of AT5G36160).

(D) Western blot validation of 35S::AmERF2-6Myc transgenic Arabidopsis plants using an anti-Myc antibody. Ponceau S staining of the Rubisco large subunit band was used as a loading control. WT, wild-type (Col-0); OE1, OE2, and OE3, three independent AmERF2-overexpressing lines.

(E) Drought stress assay of 35S::AmERF2-6Myc transgenic A. thaliana plants. Ten-day-old seedlings were transplanted to soil and subjected to drought treatment for 16 days without watering in a 22°C growth chamber with a 16-h light/8-h dark photoperiod. After the drought treatment, the plants were rewatered for 2 days to allow for recovery.

Among the PEG-induced ERF genes, ERF2 was found to be unique to the genus Ammopiptanthus. In a phylogenetic tree of ERF2 and its homologs in legume species, as well as the outgroup taxa grape and Arabidopsis, AmERF2 (Chr07.g34904) of A. mongolicus and AnERF2 (EVM0018613) of A. nanus formed a distinct clade, separate from homologous genes in other legume species (Supplemental Figure 9). Despite conservation of the AP2 domain among these plants, ERF2 displayed unique sequence patterns in its C terminus (Supplemental Figure 10). Microsynteny analysis revealed that ERF2 was positioned between the Aminophospholipid ATPase 2 (ALA2) and Tyrosine aminotransferase 2 (TAT2) genes in grape (Vitis vinifera), A. mongolicus, and A. nanus. By contrast, no genes were found between ALA2 and TAT2 in syntenic regions of typical legume plants such as G. max, M. truncatula, and Cajanus cajan (Figure 6B). Grape, which has not undergone WGD since the formation of eudicots, is often considered to have retained ancestral traits and genetic organization (Jaillon et al., 2007). By contrast, legumes have experienced at least one WGD since 58 mya, resulting in the presence of two copies of ancestral genes. Phylogenetic and synteny analyses suggested that ERF2 is most likely an ancestral gene that has lost one copy during the evolution of most legume species, whereas both copies (Chr07.g34904 and Chr04.g18700) were retained during the evolution of the genus Ammopiptanthus. We therefore heterologously expressed a Myc-tagged AmERF2 (Chr07.g34904) gene in Arabidopsis (Col-0) under the control of the CaMV 35S promoter and evaluated the drought-tolerance of the resulting transgenic plants. Three successful homozygous transgenic lines (OE1, OE2, and OE3) were confirmed by western blot analysis of transgene expression (Figure 6D). After 10-day-old soil-grown seedlings of the wild type (WT, Col-0) and the AmERF2-overexpressing lines were exposed to drought for 16 days and rewatered for 2 days for recovery, the AmERF2-6Myc transgenic plants all survived, whereas the WT and empty-vector-transformed control plants (35S::6Myc) all died (Figure 6E). To determine whether the high drought tolerance of the AmERF2-overexpressing lines was due to high levels of ethylene, we determined ethylene content by measuring the ethylene precursor 1-aminocyclopropane-1-carboxylic acid (ACC) by gas chromatography–tandem mass spectrometry under normal and drought-stress conditions. The results showed that ACC levels were much higher in the AmERF2-overexpressing lines than in the WT under normal conditions (Supplemental Figure 11). Under drought conditions, ACC content increased further in the WT plants, but no further increase was observed in the AmERF2-overexpressing lines (Supplemental Figure 11). These results suggested that overexpression of AmERF2 resulted in a high level of ethylene, which contributed to the better drought tolerance of AmERF2-overexpressing lines.

ABA is known to be involved in the regulation of plant water status. Indeed, PEG treatment of A. mongolicus resulted in significant upregulation of two ABA biosynthetic genes, NCED3 (Chr09.g44316) and AAO3 (Chr07.g34300), and an ABA-inducible gene, HAI3 (Chr06.g29412), as well as downregulation of the ABA catabolic gene CYP707A2 (Chr06.g31923) (Supplemental Table 29). Similarly, expression of the JA biosynthesis gene AOS (Chr06.g31606) was also significantly upregulated during osmotic stress. By contrast, expression of genes related to the biosynthesis of growth-promoting hormones such as brassinosteroid (BR) and auxin was suppressed by osmotic-stress treatment. These included two BR biosynthetic genes, DWF4 (Chr01.g05194) and CPD (Chr02.g10433), and two auxin biosynthetic genes, TAA1 (Chr08.g41747) and YUC (Chr07.g38214) (Supplemental Table 29). To confirm the changes in expression of these hormone-related genes during drought treatment, 15 genes involved in phytohormone biosynthesis, catabolism, and signaling were selected for validation by real-time qRT–PCR (Supplemental Tables 29 and 30). The expression patterns of these genes were consistent in the qRT–PCR and transcriptomic analyses (Supplemental Figure 12), indicating the reliability of our transcriptomic data (Supplemental Table 29).

Discussion

By combining multiple sequencing strategies, we successfully assembled a chromosome-level genome of A. mongolicus. Initial Illumina shotgun reads did not yield satisfactory assembly results, with a contig N50 of only 138.6 kb (Supplemental Table 31), and PacBio long-read sequencing data were therefore used to improve the contig N50 to 2.9 Mb (Table 1). Hi-C sequencing was performed to generate chromosome-scale scaffolds, and 98.63% of the assembly was anchored onto the nine chromosomes of A. mongolicus (Supplemental Table 6). This approach effectively captured chromosomal interactions and led to a high rate of valid interaction pairs. The resulting high-level chromosomal assembly highlights the importance of capturing interacting chromatin in Hi-C-based genome assemblies.

Repetitive sequences dominate the A. mongolicus genome, and retrotransposons are the dominant type of repetitive element, among which Ty3/Gypsy elements are twice as abundant as Ty1/Copia elements (Figure 4A). By contrast, a higher Ty1/Copia-to-Ty3/Gypsy ratio was observed in the L. angustifolius genome (Hane et al., 2017). Hence, the burst of LTR-RTs in A. mongolicus must have occurred after its divergence from L. angustifolius at ∼28 mya. Indeed, insertion time analyses of LTR-RTs indicated that they have expanded mainly within the last 2.5 million years. We also observed that A. mongolicus has a larger average chromosome size than other legumes such as soybean and barrel medic (Figure 2A). The larger genome size of A. mongolicus may be due to the expansion of retrotransposons, similar to the LTR-RT expansion in Pisum sativum (Kreplak et al., 2019). However, the two species have different LTR-RT compositions. The major LTR-RT lineage in P. sativum is Ty3/Gypsy Ogre, whereas the dominant LTR-RT lineages in A. mongolicus are Ty3/Gypsy Tekay and Reina. Genome sequencing of Ginkgo biloba, another representative relict plant, revealed that over 80% of its genome consists of repetitive sequences (Liu et al., 2021). By contrast, a recent study on the relict plant Cyclocarya paliurus revealed only 48.1% repetitive sequences in its genome (Qu et al., 2023). These findings indicate that a high percentage of repetitive sequences is not a universal characteristic of relict plants. Functional enrichment and expression analyses were performed for LTR-RT-related genes to explore their potential relevance to stress tolerance in A. mongolicus. However, no clear patterns were observed between LTR-RT-related genes and non-LTR-RT-related genes in the majority of leaf samples (Supplemental Figure 8C). Further studies are necessary to gain a comprehensive understanding of the role of LTR-RTs and their associated genes in enhancing the drought tolerance of A. mongolicus.

Several transcriptomic studies have been performed on A. mongolicus under different abiotic stresses (Zhou et al., 2012; Liu et al., 2013; Pang et al., 2013, 2015; Wu et al., 2014; Gao et al., 2015, 2016). Significant changes in gene expression have been observed under single or combined stress conditions, and some biological pathways have been identified. Nonetheless, RNA-sequencing-based transcript assemblies have inevitably led to incorrect gene-model predictions owing to the abundance of paralogous genes in plant genomes, resulting in large discrepancies in reported gene numbers. For example, 11 357 unigenes have been reported for A. mongolicus (Pang et al., 2015), which is far fewer than the average gene number (40 200) of legume species (Supplemental Table 15). The reference-grade genome for A. mongolicus thus enables more accurate gene structure annotations and forms a reliable basis for the study of epigenetic regulation, which is another important mechanism of plant adaptation to harsh environments.

As evergreen xerophytes, Ammopiptanthus species have unique morphological features compared with other legumes. A thick waxy cuticle and sparse stomata are excellent transpiration barriers, and the succulent leaves formed by multiple mesophyll cell layers increase water storage (Han and Li, 1992). Interestingly, Rhazya stricta, another evergreen shrub in the Dogbane family that is mainly found in arid regions of the Middle East, has a leaf structure similar to that of Ammopiptanthus (Schuster, 2016). This morphological similarity could be the result of convergent evolution and implies that cuticular wax is a key factor in plant adaptation to arid environments. In this study, most genes involved in cuticular wax biosynthesis or transport in A. mongolicus were specifically expressed in leaves and were significantly upregulated by PEG-induced dehydration stress. Some of these genes, such as KCR, CER4, and MAH1, were expanded as tandem gene duplicates (Figure 5). LTP genes were also highly expressed in leaves, even under normal growth conditions, and were further upregulated by PEG treatment. LTPs are required for the assembly of a waterproof lipid barrier, as they can transfer and deposit cutin and cuticular waxes (Edqvist et al., 2018). Cuticular wax consists of complex mixtures of hydrophobic VLCFAs and their derivatives. Triterpenoids are the major components of cuticular wax in plants such as Sorghum bicolor (Busta et al., 2021). Recently, Ma et al. (2024) analyzed genes related to cuticular wax biosynthesis and transport in several xerophytic plants, including Zygophyllum xanthoxylum and A. nanus. They observed a slight expansion of LACS and cytochrome b5 (CYBT5) gene copy number compared with that in the non-xerophyte plants Arabidopsis and grape. The expansion of these wax-related genes may contribute to the water-retention capacity of these xerophytes in desert environments (Ma et al., 2024). Further studies on the metabolome or chemical analyses of A. mongolicus should provide more details about the development of its cuticular wax.

Ethylene is a gaseous phytohormone released in response to various biotic and abiotic stresses/stimuli (Pei et al., 2017). In higher plants, ethylene is synthesized from S-adenosylmethionine (SAM), an activated form of methionine (Met), via ACC. Two key enzymes, ACC synthase (ACS) and ACC oxidase (ACO), are responsible for the synthesis of ACC and its conversion into ethylene, respectively (Yang and Hoffman, 1984). The expression of three ACS genes (Chr01.g00583, Chr01.g01008, and Chr06.g030417) and three ACO genes (Chr01.g00435, Chr06.g29235, and Chr06.g29237) was significantly induced by PEG treatment in A. mongolicus (Figure 6A and 6B), indicating that ethylene biosynthesis was induced by dehydration stress. Drought-induced ethylene accumulation due to upregulation of ACS genes has been reported previously in Arabidopsis (Urano et al., 2017). Ethylene action is triggered by signal perception by membrane-localized receptors (Hall et al., 2012), which initiate a signal transduction pathway involving the key transcription factors Ethylene Insensitive 3 (EIN3) and EIN3-like1 (EIL1), as well as downstream ERFs, resulting in the expression of ethylene-responsive genes (Merchante et al., 2013). Members of the ERF subfamily have been shown to regulate plant resistance to various abiotic stresses, including drought, salinity, and chilling stress (Quan et al., 2010; Rong et al., 2014; Wu et al., 2022). In the present study, we found that the AmERF2 gene was ancestral and unique to A. mongolicus (Figure 6C, Supplemental Figure 9) and that its expression was significantly upregulated under PEG-induced dehydration stress (Figure 6B). We therefore speculate that this unique gene may play an important role in the drought tolerance of A. mongolicus. Indeed, heterologous overexpression of AmERF2 in Arabidopsis significantly enhanced the drought tolerance of the transgenic plants (Figure 6D and 6E). These results demonstrate the important role of ethylene in inducing plant drought resistance.

Significant induction under PEG treatment was also observed for genes related to other stress hormones such as ABA and JA (Supplemental Figure 12). All these results indicate that the drought tolerance of A. mongolicus is controlled by sophisticated mechanisms, likely involving the action of multiple stress hormones. The complete mechanisms of A. mongolicus drought tolerance await more detailed analyses of all the drought-responsive genes and pathways.

Methods

Plant materials and growth conditions

A. mongolicus seeds were collected in Bayanhot, Inner Mongolia, China, from July 2015 to July 2019. They were surface sterilized with 70% ethanol for 10 min, washed three times with distilled water, placed in bleach for 30 min, and washed five times with distilled water. Sterilized seeds were germinated on agar plates supplemented with half-strength Murashige and Skoog (½ MS) medium (20 seeds/Petri dish, 10 cm × 25 mm) in the dark at 28°C for 2 days. Germinated seedlings were transferred into Erlenmeyer flasks with ½ MS agar and grown in the dark for 7 days before being flash-frozen in liquid nitrogen and stored at −80°C.

Illumina shotgun sequencing

Total DNA was extracted from etiolated seedlings using a modified Cetyltrimethylammonium bromide (CTAB) method (Feng et al., 2019). Five grams of tissue powder ground in liquid nitrogen were mixed with 15 mL of extraction buffer (100 mM Tris–HCl [pH 8.0], 150 mM NaCl, 1% [w/v] SDS, 100 mM EDTA, and 4% [w/v] polyvinylpyrrolidone) and incubated in a 65°C water bath for 1 h. An equal volume (15 mL) of phenol:chloroform:isoamyl alcohol (25:24:1) was added to the tissue mixture for DNA isolation. After centrifugation at 12 000 g for 10 min, the supernatant was transferred into a new centrifuge tube and mixed with half the volume of precooled isopropanol. Crude DNA was pelleted by centrifugation at 12 000 g for 10 min and re-dissolved in 1 mL of TE buffer (10 mM Tris, 1 mM EDTA, pH 8.0). RNA contamination was removed by addition of 50 μL RNase A (Omega Bio-Tek, catalog no. AC117, 25 mg/mL) to 1 mL DNA solution. DNA quantity and quality were assessed using a Qubit 3.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA) and agarose gel electrophoresis, respectively. DNA was then purified using the Aurora DNA Clean-up System (Boreal Genomics, Vancouver, BC, Canada) following the manufacturer’s protocol. The purified DNA was dissolved in 100 μL of 0.25× TBE buffer (89 mM Tris, 89 mM boric acid, and 2 mM EDTA) for later use.

Eight laddered DNA libraries were constructed, with insert sizes of 180 bp, 260 bp, 500 bp, 800 bp, 2 kb, 5 kb, 10 kb, and 20 kb. Small-insert libraries (180–800 bp) were constructed and sequenced by Macrogen (South Korea). For these libraries, 10 μg of DNA was sheared into fragments of 180–800 bp using a Bioruptor Pico. After end repair, A-tailing, and adapter ligation, the ligated fragments were size selected at 180, 260, 500, and 800 bp on an agarose gel and amplified by ligation-mediated PCR to produce the corresponding short-insert libraries for sequencing on the HiSeq 2500 system (Illumina, San Diego, CA, USA) in PE150 mode.

Mate-pair libraries were constructed by BGI Shenzhen (Beijing Genomics Institute, Shenzhen, China). In brief, 20–40 μg of genomic DNA was sheared to the desired insert size using nebulization (for 2 kb) or HydroShear (Covaris, Woburn, MA, USA) (for 5, 10, and 20 kb). Next, the DNA fragments were end repaired using biotinylated nucleotide analogs, size selected, and circularized via intramolecular ligation. Circular DNA molecules were then sheared with Adaptive Focused Acoustics (Covaris, Woburn, MA, USA) to an average size of 500 bp. Biotinylated fragments were purified with magnetic beads (Invitrogen, CA, USA), end repaired, A-tailed, ligated to Illumina paired-end adapters, size selected again, and purified by ligation-mediated PCR. Sequencing was performed on the HiSeq 2500 system in PE100 mode by BGI Shenzhen.

PacBio SMRT sequencing

For PacBio long-read sequencing, cell nuclei were first extracted from young leaves (1.5 g) of a single 1-year-old plant as described previously (Weng et al., 2018). Nuclear DNA was isolated using CTAB extraction buffer (100 mM Tris–HCl, 25 mM EDTA, 1.5 M NaCl, and 3% CTAB). DNA libraries for SMRT PacBio genome sequencing were constructed following standard protocols (Pacific Biosciences, Menlo Park, CA, USA). In brief, genomic DNA was sheared to ∼20-kb targeted size, followed by damage repair and end repair, blunt-end adaptor ligation, and size selection. Finally, the libraries were sequenced on the PacBio Sequel II platform (Beijing Berry Genomics, China).

Bionano optical mapping

To identify a suitable nicking enzyme, the draft genome sequence was scanned using LabelDensityCalculator v1.3.0 (https://bionano.com/software-downloads/), and the enzyme Nt.BspQl (New England Biolabs, Ipswich, MA, USA) was chosen for labeling high-molecular-weight DNA. A label density of 9–15 labels per 100 kb of DNA was preferred. After loading onto the IrysChip, nicked and fluorescently labeled long DNA strands were linearized and imaged using the Irys single-molecule genome mapping system (Bionano Genomics, San Diego, CA, USA) to generate single-molecule physical maps. Nb.BssSI (New England Biolabs) was also used for labeling genomic DNA, which was then read using the Bionano Saphyr system at Berry Genomics.

Hi-C sequencing

For Hi-C sequencing, 2 g of young leaves from 2-month-old A. mongolicus was cross-linked with 40 mL of 2% formaldehyde at room temperature for 15 min and then quenched by adding 4.324 mL of 2.5 M glycine. A Hi-C library was constructed according to the method described in Shi et al. (2019). DNA was digested overnight with a 4-bp cutter restriction enzyme (400 units of MboI) at 37°C on a rocking platform. Hi-C libraries were amplified by 10–12 cycles of PCR and sequenced using the Illumina HiSeq instrument in 2 × 150-bp read mode. Sequencing was performed by ANNOROAD Gene Technology Beijing (China).

Genome assembly

All PacBio long-read data were self-corrected and used for assembly using Canu v2.1 (Koren et al., 2017) with the parameters genomeSize = 930m, useGrid = false, errorRate = 0.02, corOutCoverage = 30, corMhapSensitivity = normal, minReadLength = 4000, and corMinEvidenceLength = 4000. Because the PacBio reads had an error rate of ∼15%, 114 Gb of Illumina shotgun reads from paired-end libraries were used to correct assembly errors from the PacBio reads with Pilon v1.22 (Walker et al., 2014). Bionano Solve v3.3 was used for scaffold assembly with the optical mapping data. To anchor scaffolds onto superscaffolds (or chromosomes), the Hi-C reads were mapped to scaffolds using Juicer v1.6 (https://github.com/aidenlab/juicer) with default parameters. The Hi-C assembly pipeline 3D-DNA v180922 (https://github.com/theaidenlab/3d-dna) was applied to cluster scaffolds using the agglomerative hierarchical clustering method with an interaction matrix between sequences (Dudchenko et al., 2017). The Hi-C interaction map was imported into Juicebox (https://github.com/aidenlab/Juicebox/wiki/Download) for manual checking and adjustment as described in https://www.youtube.com/watch?v=Nj7RhQZHM18. The Hi-C data were then re-aligned to the nine pseudo-chromosomes with HiC-Pro v2.11.4 (Servant et al., 2015). Finally, an interaction heatmap was generated using Juicebox_1.11.08 (https://github.com/aidenlab/juicer/wiki/Download).

Identification of repetitive sequences

Repetitive DNA regions include TEs (or interspersed repeats) and tandem repeats. RepeatModeler v2.0.1 (www.repeatmasker.org) was used with default parameters for ab initio prediction of repetitive DNA regions in the genome (Price et al., 2005). These DNA repeats were then mapped to Repbase (Dfam v3.2, https://www.dfam.org/releases/Dfam_3.2/families/Dfam.h5.gz) to determine whether they had homologous sequences in other species using RepeatMasker (Tarailo-Graovac and Chen, 2009).

Gene prediction

Gene prediction was performed on the repeat-masked genome using the EVidenceModeler pipeline (Haas et al., 2008), integrating protein sequences from closely related or remote species and RNA sequencing (RNA-seq) data from A. mongolicus for homology-based gene prediction using GeMoMa-1.6.1 (Keilwagen et al., 2016). Several well-annotated legume genome and protein datasets from six closely related species, G. max (Gmax_508_Wm82.42.v1.protein.fasta, Phytozome v13), Glycine soja (v1, http://www.wildsoydb.org/Gsoja_W05/), M. truncatula (Mtruncatula_285_Mt4.0v1.protein.fasta, Phytozome v13), Lotus japonicus (v3.0, http://www.plantgdb.org/XGDB/phplib/download.php?GDB=Lj), L. angustifolius (LupAngTanjil_v1.0.pep.all.fa.gz, https://plants.ensembl.org), and A. nanus (http://gigadb.org/dataset/100466). The PacBio isoform sequencing data from A. mongolicus were processed using SMRT analysis and clustered to reduce redundancy (Supplemental Table 32). Transcriptomic shotgun reads from roots and leaves were aligned to the genome using HISAT v2.0.4 (Pertea et al., 2016), and transcripts were reconstructed with Cufflinks v2.2.1 (Trapnell et al., 2012). Open reading frames (ORFs) were predicted with PASA v2.0.1 (Haas et al., 2003) using isoform sequences and RNA-seq transcripts as input and with the criteria for each ORF set to contain 100–1000 amino acids and at least two exons. The cDNA sequences of ORFs were used as a training set for downstream gene prediction. Ab initio gene prediction was performed with Augustus v3.0.3, SNAP, GlimmerHMM, and GeneMark-ET v4.57. All homology-based and ab initio-based gene predictions were integrated using EVidenceModeler (Haas et al., 2008).

Prediction of non-coding RNAs

snRNAs and microRNAs in the genome were identified by searching against the Rfam database (v14.1) with INFERNAL v1.1.2 (Nawrocki and Eddy, 2013) using the following parameters: -Z 1686 --cut_ga --rfam --notrunc –nohmmonly. tRNAs were detected with tRNAscan v2.0.9 using the general tRNA model (corresponding parameter: -G) (Chan and Lowe, 2019). Because rRNA genes (5S, 5.8S, 18S, and 28S) are highly conserved among plants, they were predicted in the genome by searching for the rRNA genes of G. max (GenBank: X15199.1, X02623.1, and XR_003264281.1) and A. thaliana (AT2G01020.1) using BLASTn v2.7.1 (Camacho et al., 2009).

Identification of syntenic blocks

Syntenic blocks between genomes were identified using the JCVI pipeline (https://github.com/tanghaibao/jcvi). All-versus-all BLASTP was performed to identify paralogous or orthologous gene pairs with an E-value cutoff of 1e−5. Collinear blocks containing at least five genes were identified using MCScanX with default parameters (Wang et al., 2012).

Identification of orthologous genes

Sixteen plant species were used for comparative analyses, including the legumes A. ipaensis, L. angustifolius, A. mongolicus, A. nanus, M. truncatula, Cicer arietinum, P. sativum, Trifolium pratense, L. japonicus, Vigna angularis, V. unguiculata, V. radiata, G. max, G. soja, and P. vulgaris and the outgroup A. thaliana. OrthoFinder v2.5.2 (Emms and Kelly, 2018) was used with default parameters to detect orthologous groups among the 16 plant species. All-versus-all protein alignment with a default E-value cutoff of 1e−3 was performed using Diamond v0.9.22 (Buchfink et al., 2015). Orthologous protein groups were clustered using Markov Cluster Algorithm v14-137 (Enright et al., 2002).

Calculation of 4DTv rates and Ks rates

The protein sequences of each orthologous/paralogous gene pair were aligned using MAFFT v7.427 with the following parameters: –quiet –clustalout (Katoh and Standley, 2013). The corresponding codon alignment was obtained using PAL2NAL v14 (Suyama et al., 2006), and columns with gaps were removed using the parameter –nogap. 4DTv rates between gene pairs located in syntenic blocks were calculated using custom Perl scripts (https://github.com/JinfengChen/Scripts/blob/master/FFgenome/03.evolution/distance_kaks_4dtv/bin/calculate_4DTV_correction.pl). The 4DTv value of each orthologous/paralogous gene pair was calculated as the ratio of transversion numbers in four-fold third codons to the total number of four-fold third codons. After obtaining the 4DTv values of all gene pairs, their distribution curve was visualized using the geom_density function of the ggplot2 package in RStudio. The Ks value of each paralogous/orthologous gene pair was calculated using KaKs Calculator v3.0 (Zhang, 2022).

Phylogenetic tree construction and divergence time estimation

The protein sequences of all 322 single-copy orthologous genes inferred by OrthoFinder were concatenated into a supergene for each species using custom shell scripts. The supergenes of the 16 species were aligned using MAFFT v7.427 with default parameters (output format: fasta). After alignment and gap removal, 74 865 amino acids remained for each species. RAxML (Randomized Axelerated Maximum Likelihood) v8.2.9 was used to build a phylogenetic tree with the following parameters: raxmlHPC-PTHREADS -s proteins.alignment.fasta -n raxml.out -m PROTGAMMAAUTO -p 12345 -x 12345 -# 100 -f ad -T 20. Divergence times were estimated using the mcmctree function in PAML with the following parameters: burn-in = 5 000 000, sample-number = 1 000 000, sample-frequency = 50. The divergence time between soybean (G. max) and common bean (P. vulgaris) in the Milletoid clade at 20.1 (19.0–21.0) mya was used to calibrate the systematic analysis (Zheng et al., 2016).

Identification of positively selected genes

Single-copy genes from eight legumes (A. ipaensis, A. mongolicus, P. sativum, M. truncatula, L. japonicus, V. angularis, P. vulgaris, and C. cajan) were identified using OrthoFinder with default parameters. Codon alignment of the resulting 3195 single-copy genes was performed using Clustal W2 and PAL2NAL. An unrooted phylogenetic tree for the eight legumes was used to detect positively selected genes, with the A. mongolicus lineage set as the foreground branch. Positive selection was detected using the branch-site model in the Codeml package of PAML (Liu et al., 2014).

Reconstruction of ancestral legume genomes

Ancestral legume genomes were reconstructed according to the method of Yang et al. (2021). In addition to A. mongolicus, five other legume species were also used for ancestral genome reconstruction: soybean (G. max), common bean (P. vulgaris), barrel medic (M. truncatula), white lupin (L. albus), and the wild peanut ancestor (A. ipaensis) (Supplemental Table 15). Only proteins that were anchored to chromosomes were retained for identification of orthologous genes using OrthoFinder v2.5.2 (https://github.com/davidemms/OrthoFinder/releases) with the following parameters: -t 8 -S diamond. Next, syntenic blocks were identified using DRIMM-synteny (Pham and Pevzner, 2010) with a cycle length threshold of 20 and a dust threshold of 19. The output file from DRIMM-synteny was converted into a GRIMM-like format (Tesler, 2002) using processDRIMM (https://github.com/xjtu-omics/processDrimm). Intermediate ancestor genomes were inferred using IAGS (https://github.com/xjtu-omics/IAGS). On the basis of the phylogeny and WGD events in legumes, the multicopy GHHP model was used for ancestor 4 and ancestor 6, whereas the other four ancestors were inferred using the multicopy GMP model (Supplemental Table 33).

Identification of intact LTR-RTs

LTR-RT libraries were constructed using LTR_FINDER_parallel v1.1 (https://github.com/oushujun/LTR_FINDER_parallel) and LTRharvest v1.6.1 (Ellinghaus et al., 2008). The parameters of LTR_FINDER_parallel were: -threads 10 -harvest_out -size 1 000 000 -time 300. The parameters of LTRharvest were: -minlenltr 100 -maxlenltr 7000 -mintsd 4 -maxtsd 6 -motif TGCA -motifmis 1 -similar 85 -vic 10 -seed 20 -seqids yes. Intact LTR-RT sequences were then identified and isolated using LTR_retriever v2.9.0 (Ou and Jiang, 2018) with the following parameters: -inharvest -threads 10.

Estimation of LTR-RT insertion time

For each intact LTR-RT, the DNA sequences of the two LTRs were extracted using bedtools v2.30.0 (https://github.com/arq5x/bedtools2) on the basis of the LTR_retriever results. These two DNA sequences were then aligned using Muscle v3.8.31 (https://www.drive5.com/muscle/) with default parameters. The distance between the two sequences was calculated using the Kimura two-parameter model. The insertion time of the LTR-RT was calculated following the manual at https://github.com/SIWLab/Lab_Info/wiki/Ageing-LTR-insertions using the formula T = K/(2r), where T is the insertion time, K is the divergence distance between two LTRs in an LTR-RT, and r is the synonymous nucleotide substitution rate, or synonymous mutations/site/year (Jedlicka et al., 2020). In legumes, r has been determined to be 7e−9 (Jing et al., 2005). The distribution of insertion times was plotted using the geom_histogram function of the ggplot2 package.

Classification and phylogenetic tree construction of LTR-RTs

The clade-level classification of LTR-RTs was performed using TEsorter v1.3 (https://github.com/zhangrengang/TEsorter; Zhang et al., 2019) with the following parameters: -db rexdb-plant. TEsorter was also used to identify functional domains in LTR-RTs, such as the three most conserved polyprotein domains: RT, ribonuclease H, and integrase. The RT domains of LTR-RTs were collected and clustered using cd-hit v4.8.1 (Li and Godzik, 2006) with the parameter -sc. Evolutionary analyses were performed using MEGA X (Kumar et al., 2018), and the phylogenetic tree was inferred by the maximum likelihood method with the Jones-Taylor-Thornton (JTT) matrix-based model. Phylogenetic trees were polished using FigTree v1.4.4 (https://github.com/rambaut/figtree/).

Transcriptome sequencing and data analysis

A. mongolicus seeds were surface sterilized with bleach for 30 min and germinated in ½ MS for 3 days at 28°C. Young seedlings were transplanted into pots filled with vermiculite and watered with half-strength Hoagland’s solution. The pots were placed in a greenhouse at 25°C under long-day conditions (16-h light/8-h dark). Two months after germination, the seedlings were divided into two groups: one group was transferred to fresh half-strength Hoagland’s solution (as a control), and the treatment group was transferred to half-strength Hoagland’s solution supplemented with 15% PEG6000 to mimic drought stress. During treatment, the leaves and roots of the seedlings were harvested at 0, 2, 24, and 48 h, with three independent biological replicates per treatment and time point. Samples were collected for RNA-seq, flash frozen in liquid nitrogen, and stored at −80°C.

Total RNA was extracted from frozen plant tissues using TRIzol reagent (Takara Bio, Beijing, China) according to the manufacturer’s protocol, dissolved in Diethyl pyrocarbonate (DEPC)-treated water, and stored at −80°C. RNA quality was measured using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA) and an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Strand-specific RNA-seq libraries were constructed according to the method described by Horvath et al. (2015). In brief, cDNA libraries were constructed using the Illumina TruSeq RNA preparation kit (Illumina, San Diego, CA, USA). After second-strand cDNA synthesis, the cDNAs were fragmented, ligated with adapters, and amplified by PCR. Quantitated cDNA libraries were mixed, loaded into two lanes of a flow cell, and sequenced on an Illumina HiSeq 4000 system in paired-end 100-bp mode (BGI, Shenzhen, China).

Transcriptome data were analyzed as described by Feng et al. (2023). In brief, high-quality reads were mapped to the assembled reference genome using HISAT2 v2.0.5 (--rf --trim5 10) (Kim et al., 2015). The fragments per kilobase per million mapped reads (FPKM) value of each gene was calculated using StringTie v1.3.3 (-e -B) (https://ccb.jhu.edu/software/stringtie/). Comparisons between groups were performed with the Ballgown package in Bioconductor. Differentially expressed genes (DEGs) were defined as |Fold Change| > 1.5 with p < 0.05.

Gene sequences were mapped to the SwissProt database (ftp://ftp.ncbi.nih.gov/blast/db/FASTA/swissprot.gz) using the blastx function in Diamond (Buchfink et al., 2015). Each gene annotated by SwissProt was then associated with GO terms using idmapping datasets (ftp://ftp.pir.georgetown.edu/databases/idmapping/idmapping.tb.gz). The GO annotation of A. mongolicus was constructed using the AnnotationForge package in Bioconductor. GO enrichment analyses of the DEGs were performed with the clusterProfiler package in Bioconductor.

Identification of putative genes involved in ethylene biosynthesis and signaling

Ethylene biosynthesis and signaling pathways have been well studied in model plants such as A. thaliana (Dubois et al., 2018). A specific set of ethylene-related genes from Arabidopsis, including SAM (AT1G02500), ACS (AT4G11280), ACO (AT1G62380), ETR1/EIN1 (AT1G66340), CTR1 (AT5G03730), EIN2 (AT5G03280), and EIN3 (AT3G20770), was chosen as the reference dataset. To identify ethylene-related genes in A. mongolicus, a BLASTp search (E value <1e−10) was conducted against the A. mongolicus proteome database using these reference genes from A. thaliana. The ERF gene family was identified by searching for the AP2 domain (Pfam accession: PF00847) within the A. mongolicus proteome database using the hmmscan function in HMMER version 3.1b2.

Identification of cuticular wax biosynthesis genes

Wax-related genes in A. mongolicus were identified by a BLASTp search (E value <1e−10) against the A. mongolicus proteome database using previously reported wax-related genes from A. thaliana (Dimopoulos et al., 2020). The biosynthetic pathway shown in Figure 5 was redrawn from Bernard and Joubès (2013), Zhao et al. (2016b), and Xue et al. (2017). A heatmap of gene expression levels in PEG-treated 2-month-old A. mongolicus leaves and roots was plotted using the pheatmap function in R 4.0.3. Gene expression values in all samples were normalized using the formula log₁₀(FPKM + 1).

Validation of DEGs by real-time qRT–PCR

Reverse transcription and qRT–PCR were performed according to the methods described by Feng et al. (2023). Total RNA was extracted from plant tissues using the EZNA Total RNA Kit II (Omega BIO-TEK). After quantification with the NanoDrop 2000 instrument, 1 μg of total RNA from each sample was reverse transcribed with random hexamer primers using the RevertAid First Strand cDNA Synthesis Kit (Thermo Scientific) according to the manufacturer’s instructions. The cDNA products were diluted 10-fold with sterilized Milli-Q water and used as templates. Each reaction mixture (20 μL) contained 2 μL of cDNA template, 4 μL of forward primer (1 μM), 4 μL of reverse primer (1 μM), and 10 μL of 2× premixed reagent (SYBR Premix Ex Taq II, Tli RNase H Plus, TaKaRa). A two-step PCR protocol was used in the CFX96 Touch Real-Time PCR Detection System (Bio-Rad) with the following procedure: 95°C for 3 min, followed by 39 cycles of 95°C for 10 s and 60°C for 30 s. The housekeeping gene AmEIF (eukaryotic translation initiation factor, Chr02.g10509) was used as the internal standard. Primer sequences for qRT–PCR are listed in Supplemental Table 30. Three independent biological replicates were prepared for each sample. The relative expression of each gene of interest was calculated using the 2^−ΔΔCt method (Livak and Schmittgen, 2001). Student’s t-test was used to evaluate the significance of changes.

Construction of transgenic plants and assessment of drought resistance

The cDNA sequence of AmERF2 (Chr07.g34904) was amplified and cloned into the pCanG-MYC vector to construct 35S::AmERF2-6Myc plasmids, which were then introduced into Agrobacterium strain GV1301 for transformation of A. thaliana ecotype Col-0 using the floral dip method (Zhang et al., 2006). The primers used for PCR amplification of the gene sequences are listed in Supplemental Table 34. Successful transgenic plants were validated by western blotting using mouse anti-Myc primary antibody (catalog no. 631206, Clontech) and anti-mouse IgG (H + L) secondary antibody (catalog no. 32430, Thermo Scientific). The Ponceau S-stained Rubisco large subunit band on the western blot membrane was used as a loading control.

Ten-day-old Arabidopsis seedlings (WT [Col-0], empty vector-transformed [35S::6Myc], and AmERF2 overexpressors) were transplanted to soil and cultivated in a 22°C growth chamber (16-h light/8-h dark) without watering for 16 days. Watering was resumed for 2 days, and the survival rates and phenotypic responses were recorded.

Detection of ACC content

One-week-old seedlings of WT (Col-0) and AmERF2-overexpressing lines were planted in potted soil and grown at 22°C with a 16-h light/8-h dark cycle. No additional water was provided until all the WT plants had become dehydrated. The leaves were collected and frozen in liquid nitrogen, and 50 mg of tissue was used for ACC detection by gas chromatography–tandem mass spectrometry according to previously described methods (Peng et al., 2022). A stable isotope analog of ACC ([2H4]ACC) was used as an internal standard in each sample. A series of gradient-diluted ACC standards were used as references for quantification. The significance of changes was evaluated using Student’s t-test. The experiments were repeated three times.

Data and code availability

The data supporting the findings of this study are provided in the supplemental information files. The genome assembly and annotations have been deposited at the Chinese National Genomics Data Center (https://ngdc.cncb.ac.cn/) under BioProject accession number PRJCA024714. Raw sequencing data have been deposited in the NCBI Sequence Read Archive database (http://www.ncbi.nlm.nih.gov/sra) under accession number PRJNA1085185. The genome sequences and annotations have also been posted on our website https://ammopiptanthus.github.io/ for data download.

Funding

This work was supported by the National Natural Science Foundation of China (NSFC) (no. 91125027); GRF grants (CUHK codes 14148916 and 14104521); AoE grants (AoE/M-05/12 and AoE/M-403/16) from the Research Grants Council (RGC) of Hong Kong; the NSFC-RGC Joint Scheme (N_CUHK452/17); the National Key Research and Development Program, Key Innovative and Collaborative Science and Technology Scheme for Hong Kong, Macau, and Taiwan (2017YFE0191100); direct grants from the Chinese University of Hong Kong; and the China Postdoctoral Science Foundation (2023M741234).

Author contributions

J.-X.H., H.-M.L., and L.-Z.A. conceived the project. L.-Z.A., J.-X.H., L.F., X.-L.Y., L.-F.G., G.-H.M., C.-M.W., S.-W.T., S.-N.T., J.-C.Z., and Y.L. collected the seed resources and prepared samples for sequencing. B.-J.Z. and W.-C.Y. performed flow cytometry and FISH experiments. L.F. and S.-W.T. performed the high-molecular-weight DNA purification and part of the Bionano optical mapping experiments. N.L. and L.F. performed gene overexpression experiments in Arabidopsis. L.F., F.T., and T.-Q.D. assembled the draft genomes. L.F. and J.-C.Z. performed chromosome assembly, genome annotation, evolutionary analyses, transcriptome data analyses, and qRT–PCR assays and organized all data visualization. W.S. provided logistics and coordination between the project team and the sequencing company (BGI). L.F. and J.-X.H. wrote the manuscript. L.-Z.A., S.-M.N., Y.-M.J., and Y.-L.Z. provided critical discussions and comments on the manuscript. J.-X.H. and H.-M.L. supervised the study and approved the final manuscript.

Published: April 1, 2024