Summary
Many species from across the vascular plant tree-of-life have modified standard plant tissues into tubers, bulbs, corms, and other underground storage organs (USOs). Bomarea multiflora (Alstroemeriaceae) is a tropical climbing monocot with unique underground morphology, including tuberous roots. We take a comparative transcriptomics approach to characterizing the molecular mechanisms of tuberous root formation in B. multiflora and compare these mechanisms to those identified in other underground storage structures across diverse plant lineages.
We sequenced transcriptomes from the growing tip of four tissue types (aerial shoot, rhizome, fibrous root, and root tuber) of three individuals of B. multiflora. We identify differentially expressed isoforms and test the expression of candidate genes that have been implicated in underground storage in other taxa.
We identify 271 genes that are differentially expressed in root tubers versus fibrous roots, including genes implicated in cell wall modification, environmental signaling and flowering time, and starch biosynthesis. We also identify a phosphatidylethanolamine-binding protein (PEBP) that is over-expressed in tuberous roots.
These findings demonstrate that deeply parallel processes underlie the formation of underground storage structures despite long evolutionary distances between taxa and non-homologous morphologies, illustrating that repeated co-option of similar genetic pathways can lead to convergent morphologies.
Introduction
The vast majority of scientific attention in botanical fields focuses exclusively on aboveground biomass. However, a holistic understanding of plant morphology, ecology, and evolution requires that considerable research effort go towards generating a comprehensive understanding of belowground biomass. While most studies of plant form and function focus on aboveground organs, on average 50% of an individual plant’s biomass lies beneath the ground (Niklas, 2005). Often, belowground biomass is thought to consist solely of standard root tissue, but in some cases, plants modify ‘ordinary’ structures for specialized underground functions. Plants called “geophytes” are toward the extreme end of this belowground/aboveground allocation spectrum. These species rely on nutrients stored in belowground organs (underground storage organs or “USOs”), and their ephemeral aboveground parts resprout from buds located on belowground organs (Raunkiaer, 1934; Dafni et al., 1981a,b; Al-Tardeh et al., 2008; Veselý et al., 2011). Geophytes are ecologically and economically important, morphologically diverse, and have evolved independently in all major groups of vascular plants except gymnosperms (Howard et al., 2019a,b). These plants and their associated underground structures are a compelling example of evolutionary convergence; diverse taxa form a variety of structures, often from different tissues, that serve parallel physiological and ecological functions. However, our understanding of the molecular processes that drive this convergence, and the extent to which these processes are themselves convergent, remains remarkably limited, due in part to the lack of molecular studies in geophyte lineages. This is particularly true for monocotyledonous geophytic taxa, which comprise the majority of ecologically and economically important geophyte diversity.
Some of the world’s most important crop plants are geophytes, including potato (stem tuber, Solanum tuberosum), sweet potato (tuberous root, Ipomoea batatas), yam (epicotyl- and hypocotyl-derived tubers, Dioscorea spp.), cassava (tuberous root, Manihot esculenta), radish (swollen hypocotyl and taproot, Raphanus raphanistrum), onion (bulb, Allium cepa), lotus (rhizome, Nelumbo nucifera), and more. While several of these crop plants are well studied and have sequenced genomes or other genetic or genomic data that may inform the molecular mechanisms underlying underground storage organ development, most research has focused on a select few taxa that do not represent the diversity of geophyte morphology, taxonomy, or ecology. In particular, most genetic research on geophytes and their associated underground storage organs has been conducted in eudicotyledons (with the exception of important studies in onion, such as in Navarro et al., 2011), which neglects a huge proportion of geophytic diversity in the monocots. Furthermore, geophytes form important components of many ecosystems, particularly Mediterranean biomes and other seasonal habitats, as their underground nutrient reserves fuel regrowth following periods of seasonal dormancy, prolonged dormancy, or short term resource limitation (Rundel, 1996; Hoffmann et al., 1998; Parsons, 2000; Proches et al., 2006; Cuéllar-Martínez & Sosa, 2016; Sosa & Loera, 2017; Ott et al., 2019; Howard et al., 2019b). The geophytic habit has evolved multiple times across the vascular plant tree of life, and geophyte lineages have been shown to diversify at faster rates than do related non-geophytic taxa, particularly in taxa with bulbs, corms, and tubers (Howard et al., 2019b), indicating that the geophytic habit may be correlated with increased diversification and/or reduced extinction rates.
Underground storage organs originate from all major types of plant vegetative tissue: roots, stems, leaves, and hypocotyls. Bulbs, corms, rhizomes, and tubers are some of the most common underground storage organ morphologies (Pate & Dixon, 1982), but the full breadth of morphological variation in USOs includes various root modifications (tuberous roots, taproots, etc.), swollen hypocotyls that merge with swollen root tissue (e.g. Adenia: Hearn, 2009), and intermediate structures such as rhizomes where the terminal end of the rhizome forms a bulb from which aerial shoots emerge (e.g. Iris: Wilson, 2006). Despite this morphological complexity, USOs all develop through the expansion of standard plant tissue, either derived from the root or shoot, into swollen, discrete storage organs. These storage organs also serve similar functions as belowground nutrient reserves (Veselý et al., 2011), often containing starch or other non-structural carbohydrates (NSCs), storage proteins, and water. The functional and physiological similarities of underground storage organs may be driven by deep molecular homology with parallel evolution in the underlying genetic architecture of storage organ development, despite differences in organismal level morphology and anatomy.
The economic importance of some geophytes and the relevance of understanding the formation of storage organs for crop improvement have motivated studies on the genetic basis for storage organ development in select taxa. Potato has become a model system for understanding the molecular basis of USO development, and numerous studies have demonstrated the complex and interacting roles of plant hormones such as auxin, abscisic acid, cytokinin, and gibberellin on the tuber induction process (reviewed in Hannapel et al., 2017). These hormones have been additionally identified in USO formation in other tuberous root crops including sweet potato (Noh et al., 2010; Dong et al., 2019) and cassava (Melis & van Staden, 1985; Sojikul et al., 2015), in rhizome formation in Panax japonicus var. major (Tang et al., 2019) and Nelumbo nucifera (Cheng et al., 2013b; Yang et al., 2015), and in corm formation in Sagittaria trifolia (Cheng et al., 2013a), suggesting that parallel processes trigger tuberization in both root- and stem-originating USOs. Three primary mobile signaling genes have been implicated in triggering the onset of tuberization in potato: SELF-PRUNING6A (StSP6A), a FLOWERING LOCUS T-like gene (FT gene); StBEL5 in the BEL1-like transcription factor family; and POTH1, a KNOX type transcription factor. FT-like genes have been additionally implicated in USO formation in Dendrobium (Wang et al., 2017), Callerya speciosa (Xu et al., 2016), tropical lotus (Nelumbo nucifera; Yang et al., 2015), and onion (Allium cepa; Lee et al., 2013), indicating either deep homology of FT involvement in USO formation across angiosperms or multiple independent co-option of FT orthologues in geophytic taxa.
The lateral expansion of roots into tuberous roots may be driven by either cellular proliferation concurrent with primary growth (primary thickening growth; Kaplan, unpublished), by cellular proliferation subsequent to primary growth (secondary thickening growth; Kaplan, unpublished), by cellular expansion, where individual cells expand in size, or by a combination of these processes. Expansion in plant cells requires the modification of the rigid cell wall to accommodate increases in cellular volume (Dolan & Davies, 2004; Humphrey et al., 2007), and genes such as expansins have been implicated in cellular expansion during tuberous root development in some taxa such as cassava and Callerya speciosa (Sojikul et al., 2015; Xu et al., 2016). Recent studies of the tuberous roots of sweet potato (Ipomoea batatas) and other members of Convolvulaceae indicate that USO formation in these taxa involves a MADS-box gene implicated in the vascular cambium (SRD1; Noh et al., 2010) and a WUSCHEL-related homeobox gene (WOX4; Eserman et al., 2018), also involved in vascular cambium development. Additional work on cassava (Manihot esculenta) also suggests that tuberous root enlargement is due to secondary thickening growth originating in the vascular cambium (Chaweewan & Taylor, 2015). However, geophytes are especially common in monocotyledonous plants (Howard et al., 2019a,b), which lack a vascular cambium entirely. No previous study has addressed the molecular mechanisms of USO development in this major clade, so the causes of root thickening are particularly enigmatic. Do monocots form tuberous roots through genetic machinery that shares deep homology with the eudicot vascular-cambium-related pathways, or have they evolved an entirely independent mechanism?
Bomarea multiflora (L. f.) Mirb. is a scandent monocotyledonous geophyte that is native to Venezuela, Colombia, and Ecuador (Hofreiter, 2008) and is invasive in New Zealand (National Pest Plant Accord: https://www.mpi.govt.nz/protection-and-response/long-term-pest-management/national-pest-plant-accord). It typically grows in moist cloud forests between 1800 3800 meters elevation (Hofreiter, 2008). Bomarea multiflora is an excellent model in which to study the molecular mechanisms underlying underground storage organ formation in the monocots because it has two types of underground modifications: tuberous roots and rhizomes. However, prior to this study, no genomic or transcriptomic data was available for any species of Bomarea. Comparative transcriptomics permits comprehensive examination of the molecular basis of development, tissue differentiation, and physiology in ecologically relevant taxa by comparing the genes expressed in different organs, developmental stages, or ecological conditions (Ekblom & Galindo, 2011; Oppenheim et al., 2015). Because no prior genomic or transcriptomic data is needed for comparative transcriptomic studies, this method is especially appropriate for studies of non-model organisms and can yield novel insights into the expression profiles of specific tissues. In this study, we investigate the molecular mechanisms underlying the formation of tuberous roots in Bomarea multiflora using a comparative transcriptomics approach and quantify the extent to which these mechanisms are shared across the taxonomic and morphological breadth of geophytic taxa.
Materials and Methods
Greenhouse and laboratory procedures
Seeds were collected from a single inflorescence of Bomarea multiflora in Antioquia, Colombia [vouchered as Tribble 194, deposited at UC; Index Herbariorium http://sweetgum.nybg.org/science/ih/] and germinated in greenhouse conditions at the University of California, Berkeley designed to replicate B. multiflora’s native conditions (70°F – 85°F and ~50% humidity). Six months after germination, three sibling individuals were harvested as biological replicates. The aerial shoot apical meristem (SAM) of a single branch, the rhizome apical meristem (RHI), root apical meristems (RAM) of several fibrous roots, and the growing tip of a tuberous root (TUB; Figure 1) were dissected from each of three individuals, for a total of 12 tissue samples. Samples were immediately frozen in liquid nitrogen and maintained at −80°C until extraction. Total RNA was extracted from all samples using the Agilent Plant RNA Isolation Mini Kit (Agilent, Santa Clara, Ca), optimized for non-standard plant tissues, especially those that may be high in starch. As this protocol did not work well for the vegetative SAM samples, RNA was extracted from two dissected SAMs using an alternative plant tissue RNA extraction protocol (Yockteng et al., 2013). Quality of total RNA was measured with Qubit (ThermoFisher, Waltham MA) and Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA); if needed, a Sera-Mag bead clean-up was used to further clean extracted RNA (Yockteng et al., 2013). Two SAM samples failed to extract at sufficient concentrations, so we harvested the SAMs of two additional individuals, froze, and extracted using the Yockteng et al. 2013 protocol. Samples with an RNA integrity (RIN) score > 7 proceeded directly to library prep. The KAPA Stranded mRNA-Seq Kit (Kapa Biosystems, Waltham MA) protocol was used for library prep. Half reactions were used with an input of at least 500 ng of RNA; however, most samples (all but two) had ~1 ug of RNA. RNA fragmentation time depended on RIN score (7< RIN < 8: 4 min; 8 < RIN < 9: 5 min; 9 < RIN: 6 min). Samples were split in half after the second post-ligation clean up (Step 10 in the Kapa protocol) in order to fine-tune the enrichment step. The first half of the samples were amplified with 12 PCR cycles; this proved too low and was increased to 15 cycles for the second half of the samples. Samples were combined and library quality was assessed with a Bioanalyzer 2100 using the DNA 1000 kit. A bead clean-up was performed on libraries showing significant adaptor peaks (Yockteng et al., 2013). Samples were cleaned, multiplexed, and sequenced on a single lane of HiSeq4000 at the California Institute for Quantitative Biosciences (QB3) Vincent J. Coates Genomics Sequencing Lab.
Bioinformatics Data Processing
Raw reads were cleaned, processed, and assembled using the Trinity RNA-Seq De novo Assembly pipeline (Grabherr et al., 2011) under the default settings unless otherwise stated in associated scripts. All analyses were run using the Savio supercomputing resource from the Berkeley Research Computing program at UC Berkeley. Reads were cleaned with Trim Galore! (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/), keeping unpaired reads and using a minimum fragment length of 36 base pairs. Data from all tissue types and biological replicates were concatenated in order to generate a consensus transcriptome, assembled de novo from the concatenated data using Trinity. Each sample was compared back to the assembled consensus transcriptome, aligned using Bowtie 2 (Langmead & Salzberg, 2012), and quantified using RSEM (RNASeq by Expectation Maximization; Li & Dewey, 2011). The consensus transcriptome was annotated with a standard Trinotate pipeline (https://trinotate.github.io/), comparing assembled transcripts to SWISS-PROT (Boeckmann et al., 2003), RNAmmer (Lagesen et al., 2007), Pfam (Finn et al., 2014), eggNOG (Powell et al., 2014), KEGG (Tanabe & Kanehisa, 2012), and Gene Ontology (Gene Ontology Consortium, 2004) databases. The concordance of biological replicates was tested by looking for significant differences between the total number of fragments per replicate, by comparing the transcript quantities of all replicates to each other, and by checking the correlations between replicates. Transcripts with less than 10 total counts were discarded for all downstream analyses. Transcript counts were transformed using the variance-stabilized transformation (VST) and all 12 samples were compared using a principal components analysis. All scripts for these analyses are available on GitHub (github.com/cmt2/bomTubers).
Data Analysis
Differentially expressed isoforms (hereafter referred to as DEGs) between fibrous (FR) and tuberous (TR) roots were identified with DESeq2 package in R (Love et al., 2014), extracting the results of comparing tuberous roots to fibrous roots and using a p-adjusted cut-off (padj, uses a Benjamini-Hochberg correction for false discovery rate) of 0.01 and a log2-fold change cut-off of 2 to specifically only statistically significant and sufficiently differentially expressed isoforms for downstream analyses. To identify specific candidate genes, gene families, or molecular processes that might be involved in the development of underground storage organs, we surveyed the literature for recent publications on the molecular basis of USO formation. For each group of genes hypothesized to be involved in USO formation (either gene families or molecular/ physiological processes), the annotated transcriptome was queried for transcript annotations matching the associated process or family (See Table S1 for the specific search terms used). The resulting transcripts were 1) compared using the distribution of the group’s log2-fold change values to the distribution of values from all isoform using a non-parametric Mann-Whitney test, and 2) tested for the presence of any statistically significant differentially expressed isoforms. For all targeted candidate genes, the amino acid sequence of the candidate gene was blasted to the assembled consensus transcriptome (see Table S2 for the blasted sequences specifications) using an e-value cut-off of 0.01 to assess if the identified homologs were differentially expressed. All associated scripts for data analysis are available on GitHub (github.com/cmt2/bomTubers).
The evolutionary history of the phosphatidylethanolamine-binding protein (PEBP) gene family was reconstructed by combining amino acid sequences from an extensive previously published alignment (Liu et al., 2016) with the addition of sequences specifically implicated in USO formation in onion and potato or from geophytic taxa such as Narcissus tazetta (accession AFS50164.1), Tulipa gesneriana (accessions MG121853, MG121854, and MG121855), Crocus sativa (saffron, accession ACX53295.1), and Lilium longiflorum (accessions MG121858, MG121857, MG121859) (Navarro et al., 2011; Tsaftaris et al., 2012; Lee et al., 2013; Li et al., 2013; Leeggangers et al., 2017) and with copies identified in our transcriptome. Amino acids were aligned with MAFFT as implemented in AliView v1.18 (Larsson, 2014), using trimAl v1.4.rev15 (Capella-Gutierrez et al., 2009) with the -gappyout option. The best evolutionary model was selected with ModelTest-NG v0.1.5 (Darriba et al., 2016) and unrooted gene trees were reconstructed under a maximum likelihood framework as implemented in IQtree (Nguyen et al., 2014), run on XSEDE using the CIPRES portal (Miller et al., 2010).
Results
Transcriptome Data
A total of 359 M paired-end 100 bp reads were recovered from the single HiSeq 4000 lane for the multiplexed 12 samples, to be made available as SRAs in an NCBI BioProject prior to publication. The assembled consensus transcriptome consists of 370,672 transcripts, corresponding to 224,661 unigenes, also to be made available on Dryad prior to publication. The combined data have a GC content of 45.14%, N50 of 1191 bp, median transcript length of 317 bp, and mean transcript length of 556.95 bp. All four tissue types showed concordance between the three biological replicates with generally 1:1 ratios of transcript quantities to each other (see Figures S1 – S4) so no further replicates were discarded and all differentially expression analyses incorporated data from all three biological replicates. A principal component analysis (PCA) of the VST transcript counts (Figure 2) shows that the first PC axis (45% of the variance in samples) generally explains the variation between tissue types. The shoot tissues (SAM and RHI) cluster separately, while the root tissues (ROO and TUB) cluster together. The underground rhizome samples (RHI) fall out intermediate between the aerial shoot samples (SAM) and the underground root and tuber samples (ROO and TUB) along this axis. The co-clustering of fibrous and tuberous root samples in the PCA indicates that the overwhelming, general pattern of expression between all root samples is similar, especially in contrast to the very distinct shoot samples. This broad pattern of similarity provides an excellent opportunity to identify the particular genes that cause such obvious morphological differences between the tuberous and fibrous roots, given their similar background patterns of expression. The second PC axis (22% of variance) generally explains variance between biological replicates, with Individual A particularly distinct from other individuals.
Differential Expression
A total of 271 differentially expressed isoforms (DEGs) were recovered between fibrous and tuberous roots (FR vs. TR) (Table S3). Of these, 226 correspond to regions of the annotated consensus transcriptome with functional annotations. The most common gene ontology (GO) terms associated with these DEGs are presented in Table 1. The top cellular components include the nucleus, the cell membrane and cell wall, and plastids; the top molecular functions include kinase activity and binding to ATP, nucleic acids, proteins, and various metals; and the top biological functions include cell wall organization, morphological development, environmental sensing, and protein transport. Of the 271 DEGs, 126 (46.5%) were over-expressed in tuberous roots while the remaining 145 (53.5%) were under-expressed. All top ten most differentially expressed isoforms (the ten DEGs with the highest absolute value log2-fold change values between fibrous and tuberous roots) were functionally annotated. These ten DEGs are implicated in various cellular and biological processes (Table 2). All but nine of these top ten DEGs are overexpressed in tuberous roots and are generally implicated in nucleotide and ATP binding, cell wall modification, root morphogenesis, and carbohydrate and fatty acid biosynthesis. The top most differentially expressed isoform with a 40.25 log2-fold change value, TRINITY_DN116220_c0_g1_i4, is a Zinc finger CCCH domain-containing protein 55, a possible transcription factor of unknown function. Other notable top DEGs include TRINITY_DN128685_c1_g3_i4, callose synthase 3, which regulates cell shape, TRINITY_DN121298_c2_g2_i5, a heat shock protein, TRINITY_DN127064_c0_g3_i1, an LRR receptor-like serine implicated in lateral root morphogenesis, and TRINITY_DN121430_c10_g2_i, a carbohydrate metabolism protein. The tenth most differentially expressed DEG, under-expressed in tuberous roots, is implicated in abscisic acid signaling.
Parallel Processes Across Taxa
Eleven gene groups — either gene families, physiological signaling pathways, or biosynthesis pathways — were identified that have been implicated in USO formation by expression of functional analyses across various geophytic organisms: abscisic acid response genes, calcium-dependent protein kinases (CDPK), expansins, lignin biosynthesis, MADS-Box genes, starch biosynthesis, auxin response genes, cytokinin response genes, 14-3-3 genes, gibberellin response genes, and KNOX genes (See Figure 3). For each group, putative homologs were located in Bomarea multiflora and their expression patterns were analyzed. Of these 11 gene groups, the log2-fold change values of six are distributed significantly differently than the overall distribution of log2-fold change values for all isoforms; in two cases (starch biosynthesis and MADS-Box genes) the expression levels overall are significantly greater than expected (generally over-expressed in TR compared to FR) and in the remaining four cases (abscisic acid, CDPK, expansins, and lignin biosynthesis) the groups’ log2-fold change values are under-expressed. Fifteen individual DEGs were identified in these gene groups (Table 3); interestingly, there seems to be no generalizable relationships between the significance and directionality of a particular group’s distribution with the presence and directionality of log2-fold change values for individual DEGs. For example, the expression distribution of gibberellin genes does not deviate significantly from the global pool of isoforms, but there is one significantly under-expressed DEG in the gibberellin group; similarly, the CDPK genes are under-expressed as a group, but the only CDPK-related significant DEG is over-expressed (Table 3).
Five specific candidate genes were identified from the literature: qRT9 has been implicated in root thickening in rice (Li et al., 2015); IDD5 and WOX4 are implicated in starch biosynthesis and TR formation, respectively, in Convolvulaceae (Eserman et al., 2018); sulfite reductase is associated with TR formation in Manihot esculenta (Sojikul et al., 2010); and FLOWERING LOCUS T (FT) has been implicated in signaling the timing of USO formation in a variety of taxa, notably Allium cepa and Solanum tuberosum (Navarro et al., 2011; Hannapel et al., 2017). We recover between nine and 63 putative homologs of these candidates using a BLAST E-value cut-off of 0.01 (Table 4), but only one putative homolog is significantly differentially expressed (padj < 0.01): a putative FT homolog (TRINITY_DN129076_c1_g1_i1), further investigated in PEBP Gene Family Evolution (below). One putative qRT9 homolog is marginally significant (padj = 0.050), and the E-value from the BLAST result to this isoform was 0.09. Given these marginal significance values, it is likely the result is spurious and we do not follow up with further analysis.
PEBP Gene Family Evolution
Thirty-seven Bomarea isoforms were identified as putative FLOWERING LOCUS T (FT) homologs. The longest isoform per gene was selected to include in an alignment of phosphatidylethanolamine-binding protein (PEBP) amino acid sequences. Coding sequences from the Bomarea multiflora transcriptome were translated to amino acid sequences using TransDecoder v5.5.0 (Haas et al., 2013), removing isoforms that failed to align properly; ultimately, we include six sequences, including the significantly differentially expressed copy. Using the JTT+G4 amino acid substitution models (Jones et al., 1992) we recover three major clusters in our unrooted gene tree, all with strong bootstrap support (Figure 4a); these correspond to the FT cluster, TERMINAL FLOWER 1 (TFL1) cluster, and MOTHER OF FT AND TFL1 (MFT) cluster recovered in previous analyses (Liu et al., 2016). Three of the six Bomarea multiflora isoforms fall out with FT genes and three fall out with TFL1 genes. The Bomarea DEG homolog is highly supported in the TFL1 cluster with sequences from other monocot taxa (Figure 4b). FT homologs from Allium cepa and Solanum tuberosum that have been functionally implicated in stem tuber and bulb formation, respectively, are in the FT cluster but do not cluster together; rather all USO-implicated PEBP genes are more closely related to non-USO copies than to each other.
Discussion
How to Make a Tuberous Root
The top 10 most highly represented biological processes of the 271 Differentially expressed isoforms (DEGs) include cell wall organization, responses to environmental signals, growth and development, and carbohydrate biosynthesis (Table 1). Together, these processes describe the various components of development by which the plant modifies fibrous roots into tuberous roots: 1) how expansion occurs, 2) when tuberization is triggered, and 3) what the tuberous roots store.
Root expansion likely occurs due to cellular expansion and primary thickening growth in B. multiflora. Due to the absence of a vascular cambium, secondary growth is not likely to be involved, despite the prevalence of this mechanism in other taxa such as sweet potato (Noh et al., 2010; Eserman et al., 2018) and cassava (Melis & van Staden, 1985). DEGs with cell wall functional annotations, the most common GO biological process, likely contribute to root enlargement through permitting cellular expansion. For example, pectinesterase TRINITY_DN122210_c6_g1_i1 (log2-fold change = 21.91; padj = 1.22E−8) modifies pectin in cell walls leading to cell wall softening, as demonstrated, for example, in Arabidopsis (Braybrook & Peaucelle, 2013). Interestingly, expansins were not over-expressed in B. multiflora, though this is the mechanism by which cell expansion occurs in other taxa (see Expansins discussion below). The DEGs functionally annotated as contributing to cell division (one of the top GO biological processes in the DEG dataset) may contribute to root enlargement through increased cellular proliferation at the growing tip of the tuberous root (primary thickening growth).
Several of the top GO biological processes are responses to environmental stimuli, including cold, water deprivation, flowering, and defense, which may trigger tuberization. Flowering development genes in particular may be co-opted for tuber formation, a hypothesis discussed in more detail below (PEBP gene family evolution). Tuberization signaling may also be mediated by callose production, influencing symplastic signaling pathways through plasmodesmata modification. Callose synthase 3 is one of the most highly differentially expressed DEGs (TRINITY_DN128685_c1_g3_i4, Table 2). Callose is a much less common component of cell walls than is cellulose (Schneider et al., 2016), but it is often implicated in specialized cell walls and in root-specific expression (Vatén et al., 2011; Benitez-Alfonso et al., 2013). Callose synthase has been implicated in the development of other unique root-based structures such as root nodules (Gaudioso-Pedraza et al., 2018) and mutations in callose synthase 3 affect root morphology (Vatén et al., 2011), suggesting that callose synthase 3 plays an integral role in triggering tuberous roots development in B. multiflora through symplastic signaling pathways. Callose signaling-induced USO formation has not previously been reported and be unique to B. multiflora or to monocotylendous taxa.
Finally, starch is thought to be the primary nutrient reserve in Bomarea tubers (Kubitzki, 1998). Many previous studies have found evidence of overexpression of carbohydrate and starch synthesis molecules in USOs (for example in sweet potato; Eserman et al., 2018)). Differentially expressed isoforms implicated in the carbohydrate metabolic process support the presence of active starch synthesis in our data. One of the most differentially expressed isoforms is a homolog of sucrose non-fermenting 4-like protein (Table 2, TRINITY_DN121430_c10_g2_i1) and participates in carbohydrate biosynthesis, demonstrating that B. multiflora tubers were actively synthesizing starch when harvested. Additionally, genes implicated in defense response, such as TRINITY_DN127064_c0_g3_i1 (LRR receptor-like serine/ threonine-protein kinase HSL2, Table 2) may be differentially expressed in tuberous roots to protect starch reserves against potential predation by belowground herbivores.
Similarities in Molecular Mechanisms of USO Formation
We identify six molecular processes, previously implicated in USO formation in other taxa, which are either over- or under-expressed in the tuberous roots of Bomarea multiflora (Figure 3). These processes show parallel function across deeply divergent evolutionary distances and in distinct plant structures.
Abscisic Acid (ABA) has been shown to increase initially and then decrease during USO formation in the rhizomes of Nelumbo nucifera (Yang et al., 2015), the tuberous roots of sweet potato Ipomoea batatas (Dong et al., 2019), and the stem tubers of potato Solanum tuberosum (Xu et al., 1998). We find that ABA signaling genes are under-expressed in tuberous roots (Figure 3), perhaps indicating that our samples were in a later developmental stage, characterized by lower ABA levels. Experimental manipulation of B. multiflora tuberous roots and developmental time series would be needed to confirm the role of ABA in tuberous root formation. An ABA signaling DEG (Table 3) is one of the most differentially expressed isoforms (TRINITY_DN122787_c0_g1_i1: Protein IQ-DOMAIN 32, Table 2). This isoform and other ABA genes may signal the cessation of continued elongation in monocot tuberous roots, similar to their role in deeply divergent taxa and non-homologous USOs.
Calcium-dependent Protein Kinases (CDPKs) play an integral role in tuber formation in cassava Manihot esculenta (Sojikul et al., 2010). Sujikul et al. (2010) propose that CDPKs may signal the initiation of tuberous root development, similarly to the process described in stem tuber formation in potato (Raíces et al., 2003) and rhizome development in Nelumbo nucifera (Cheng et al., 2013b). In these studies, CDPKs are over-expressed and signal the initiation of USO development. Our results show that on average CDPKs as a group are under-expressed in the tuberous roots of B. multiflora. However, the only significant CDPK DEG (TRINITY_DN124121_c3_g1_i11: Calcium-dependent protein kinase 2) is over-expressed, suggesting that CDPK 2 expression plays a role in initiating tuberous root development and that CDPK involvement in tuberous root formation is ubiquitous in all studied taxa.
Expansins are cell wall modifying genes known to loosen cell walls in organ formation (Dolan & Davies, 2004; Humphrey et al., 2007; Braybrook & Peaucelle, 2013). Their involvement in USO formation has been documented in the tuberous roots of cassava (Sojikul et al., 2015), and Callerya speciosa (Xu et al., 2016) the rhizomes of Nelumbo nucifera (Cheng et al., 2013b), the tuberous roots of various Convolvulaceae (Eserman et al., 2018), and the stem tubers of potato (Jung et al.). As a group expansins are under-expressed in tuberous compared to fibrous roots, but none are statistically significant, so it seems unlikely that expansins play an important role in tuberous root formation in Bomarea multiflora. It is possible that expansin involvement in USO formation is unique to eudicots.
Lignin biosynthesis genes are under-expressed in several geophytic taxa with tuberous roots, including cassava (Sojikul et al., 2015), wild sweet potato (Ipomoea trifida; (Li et al., 2019), and Callerya speciosa (Xu et al., 2016). Similarly, we find that lignin biosynthesis overall is under-expressed in tuberous compared to fibrous roots, and one isoform in particular is significantly under-expressed: TRINITY_DN125451_c5_g1_i3: Cinnamoyl-CoA reductase-like SNL6 (Table 3). This gene has been found to significantly decrease lignin content without otherwise affecting development in tobacco (Chabannes et al., 2001). Decreased lignin in tuberous roots may further allow for cell expansion and permit lateral swelling of tuberous roots during development.
MADS-Box genes are implicated in USO formation in the tuberous roots of wild sweet potato (Ipomoea trifida; (Li et al., 2019), and sweet potato (Ipomoea batatas; (Noh et al., 2010; Dong et al., 2019), the rhizomes of Nelumbo nucifera (Cheng et al., 2013b), and the corms of Sagittaria trifolia (Cheng et al., 2013a), indicating widespread parallel use of MADS-Box genes in the formation of USOs. Similarly, we find that MADS-Box genes overall, and one DEG in particular, are over-expressed in Bomarea tuberous roots. MADS-Box genes are implicated widely as important transcription factors regulating plant development (Buylla et al., 2000). It is thus unsurprising that MADS-Box genes are regularly implicated in USO formation. It remains unclear if the MADS-Box genes identified in the aforementioned studies represent independent neofunctionalizations of MADS-Box genes from other aspects of plant development, or if they form a clade of USO-specific copies.
Starch biosynthesis genes are very commonly identified in the formation of USOs, including in cassava (Sojikul et al., 2010; 2015), Nelumbo nucifera (Cheng et al., 2013b; Yang et al., 2015), wild and domesticated sweet potatoes (Eserman et al., 2018; Li et al., 2019; Dong et al., 2019), and potato (Xu et al., 1998). Since starch is so ubiquitous in USOs, this is unsurprising. We also find starch isoforms overall to be over-expressed in Bomarea tuberous roots, and three genes in particular are significantly over-expressed (see Table 3).
The other molecular processes we tested failed to show group-level differences from the global distribution of expression levels. However, the presence of DEGs in some of these groups indicates that the phytohormones in particular may play a role in tuberous root formation. One gibberellin response isoform that is significantly under-expressed in tuberous roots, which aligns with previous research suggesting that decreased gibberellin concentrations in roots can lead to root enlargement (Tanimoto, 2012) and tuber formation (Xu et al., 2016; Li et al., 2019; Dong et al., 2019). The lack of significant auxin-related isoforms as differentially expressed is surprising, as auxin has been implicated in USO formation in several previous studies (Noh et al., 2010; Cheng et al., 2013b; Sojikul et al., 2015; Yang et al., 2015; Xu et al., 2016; Hannapel et al., 2017; Li et al., 2019; Dong et al., 2019; Kolachevskaya et al., 2019), but it is possible this function is not utilized in the tuberous root formation for monocot taxa.
PEBP and FT-Like Gene Evolution in Geophytic Taxa
Gene tree analysis of PEBPs indicates that FT and TFL1 genes have been independently co-opted several times in USO formation in diverse angiosperms (including monocots and eudicots) and in diverse USO morphologies (including tuberous roots, bulbs, and stem tubers). Furthermore, the presence of TFL1 and FT homologs in gymnosperms and other non-flowering plants (Liu et al., 2016) suggests that the origin of these genes predates the evolution of flower as a reproductive structure. Instead, it seems likely that these genes originally evolved as environmental signaling genes with wider involvement in triggering the seasonality of various aspects of plant development. Subsequent specialization of these genes in the timing of shifts to reproductive development as well as USO development occurred. Given our results, in seems likely that USO-specialized FT and TFL1 genes arose at least four times independently, indicating broadly parallel molecular evolution underlies the convergent morphological evolution of USOs. Additional USO-specific PEBP genes would shed more light on this pattern, but the dearth of studies on the molecular basis of USO development impedes such analysis. With increased sampling, follow-up studies could identify unique patterns of convergent molecular evolution on the USO-specific FT genes. Do these copies share independently derived subsequences or motifs that could reflect or cause shared function?
In Bomarea multiflora, TFL1 involvement in USO formation is surprising, as neotropical cloud forests are generally considered relatively aseasonal environments. However, B. multiflora shows two annual peaks in flowering corresponding to two peaks in annual rainfall (Ortiz & Idárraga Piedrahita, 2009), suggesting that important aspects of the plant’s phenology are tied to seasonal fluctuations. While no previous studies have looked at the timing of tuber production in Bomarea, it is possible that tuber development is also tied to seasonal cues.
Conclusion
We provide the first evidence of the molecular mechanisms of tuberous root formation in a monocotyledonous taxon, filling a key gap in understanding the commonalities of storage organ formation across taxa. We demonstrate that many molecular processes are shared across geophytic taxa, suggesting that deep parallel evolution at the molecular level underlies the convergent evolution of an adaptive trait. In particular, we demonstrate that PEBP genes implicated in underground storage organ formation have been recruited multiple times across the gene tree, demonstrating that repeated morphological convergence is matched by repeated molecular convergence. These findings suggest further avenues for research on the molecular mechanisms of how plants retreat underground and evolve strategies enabling adaptation to environmental stresses. More molecular studies on diverse, non-model taxa and more thorough sampling of underground morphological diversity will enhance our understanding of the full extent of these convergences and add to our general understanding of the molecular basis for adaptive, convergent traits.
Author Contribution
FA provided materials for the experiment, CMT and CDS designed the experiment, CMT and JMG executed the experiment, CMT, JMG, and CJR analyzed the data, and CMT, JMG, CJR and CDS contributed to writing the manuscript.
Acknowledgements
We thank Lydia Smith (Evolutionary Genetics Lab, UC Berkeley) for training and sharing her expertise on RNA-Seq, NSF GRFP, SSB, ASPT, Pacific Bulb Society, UC Berkeley’s Integrative Biology Department, and the Tinker Foundation for support to CMT, and UC Berkeley CNR and the University and Jepson Herbarium for supporting sequencing costs.