Shared expression of crassulacean acid metabolism (CAM) genes pre-dates the origin of CAM in the genus Yucca

Although large differences in metabolism exist between C₃ and CAM species, we find that many CAM genes have similar expression patterns regardless of photosynthetic pathway, suggesting ancestral propensity for CAM.

Abstract

Crassulacean acid metabolism (CAM) is a carbon-concentrating mechanism that has evolved numerous times across flowering plants and is thought to be an adaptation to water-limited environments. CAM has been investigated from physiological and biochemical perspectives, but little is known about how plants evolve from C₃ to CAM at the genetic or metabolic level. Here we take a comparative approach in analyzing time-course data of C₃, CAM, and C₃+CAM intermediate Yucca (Asparagaceae) species. RNA samples were collected over a 24 h period from both well-watered and drought-stressed plants, and were clustered based on time-dependent expression patterns. Metabolomic data reveal differences in carbohydrate metabolism and antioxidant response between the CAM and C₃ species, suggesting that changes to metabolic pathways are important for CAM evolution and function. However, all three species share expression profiles of canonical CAM pathway genes, regardless of photosynthetic pathway. Despite differences in transcript and metabolite profiles between the C₃ and CAM species, shared time-structured expression of CAM genes in both CAM and C₃Yucca species suggests that ancestral expression patterns required for CAM may have pre-dated its origin in Yucca.

Introduction

Crassulacean acid metabolism (CAM) is a carbon-concentrating mechanism that can reduce photorespiration and enhance water use efficiency (WUE) relative to plants that rely solely on the C₃ photosynthetic pathway. In CAM plants, stomata are open for gas exchange at night, when transpiration rates are lower, and incoming CO₂ is initially fixed by phosphoenolpyruvate carboxylase (PEPC) rather than Rubisco. Carbon is temporarily stored as malic acid within the vacuole, and during the day stomata close and the malic acid is decarboxylated in the cytosol, ultimately resulting in high concentrations of CO₂ around Rubisco. The extra steps of CAM—carboxylation of PEP, decarboxylation of malic acid, and transport into and out of the vacuole—come with extra energetic costs relative to C₃ photosynthesis, but CAM plants have the advantage of acquiring carbon with increased WUE. In addition, Rubisco is able to act more efficiently with a high concentration of CO₂, and the risk of photorespiration is thought to be significantly minimized (Cushman and Bohnert, 1997; Schulze et al., 2013). CAM plants are therefore adapted to habitats where water stress is unavoidable and where the energetic cost of CAM is offset by reduced photorespiration under water-limited conditions. CAM has evolved at least 35 independent times in flowering plants (Silvera et al., 2010), thus making it a remarkable example of convergent evolution of a complex trait.

CAM has been studied from a biochemical standpoint for decades, and much is known about the carbohydrate turnover, starch cycling, and enzymatic machinery of CAM plants (Cushman and Bohnert, 1997; Chen et al., 2002; Dodd et al., 2002). Additionally, physiological studies of CAM plants have revealed the importance of succulence and large cells (Kluge and Ting, 1978; Nelson et al., 2005; Nelson and Sage, 2008; Barrera Zambrano et al., 2014). In terms of the basic machinery required for CAM, carbonic anhydrase (CA) aids in the conversion of CO₂ to HCO₃⁻ at night (Fig. 1A). PEPC fixes the carbon from bicarbonate into oxaloacetate (OAA), but its activity is regulated by a dedicated kinase, PEPC kinase (PPCK). Phosphorylated PEPC is able to fix carbon in the presence of its downstream product, malate, whereas the unphosphorylated form is sensitive to malate (Nimmo, 2000; Taybi et al., 2000). As day approaches, PPCK is down-regulated by two mechanisms: circadian regulation (Carter et al., 1991; Hartwell et al., 1996) or through metabolite control of transcription which results from elevation of cytosolic malate (Borland et al., 1999). During the day, the stored malic acid exits the vacuole and is decarboxylated by PEP carboxykinase (PEPCK) and/or NADP/NAD-malic enzyme (ME), depending on CAM species (Holtum and Osmond, 1981). NADP/NAD-ME CAM plants additionally have high levels of orthophosphate dikinase (PPDK), which converts pyruvate to PEP. This final step is important for CAM plants, as PEP can be further used to generate carbohydrates and is also required for CO₂ fixation during the following night (Fig. 1A).

Fig. 1.

Overview of the CAM photosynthetic pathway and physiology of the Yucca hybrid system. (A) CAM pathway diagram. CA, carbonic anhydrase; PEP, phosphoenolpyuvate; PEPC, PEP carboxylase; PPCK, PEPC kinase; OAA, oxaloacetate; ME, malic enzyme; PEPCK, PEPC carboxykinase; PPDK, orthophosphate dikinase. (B) Net CO₂ accumulation on the same samples used for RNA-Seq, with error bars representing 1 SD from the mean. The white and black bars under plots represent day and night, respectively. (C) Delta H⁺ (the total titratable acid accumulated during the night) measured on samples used for RNA-Seq from well-watered (‘W’) and drought-stressed conditions (‘D’), with error bars representing 1 SD from the mean. Both gas exchange and titratable acidity plots are modified from data published in Heyduk et al. (2016). (D) Up/down differential expression between well-watered and drought-stressed plants at each time point based on EdgeR, with counts as a proportion of total transcripts expressed. (This figure is available in color at JXB online.)

A daily turnover of sugars or starch for PEP generation is a defining characteristic of CAM plants. Carbohydrates that are laid down during the day must be broken down to PEP at night to provide substrate for CO₂ fixation via PEPC. The nocturnal demand for PEP represents a significant sink for carbohydrates which CAM plants must balance with partitioning of carbohydrates for growth and maintenance (Borland et al., 2016). The interplay between carbohydrate metabolism and CAM is clearly an important regulatory mechanism; previous work has shown that plants with reduced carbohydrate degradation have decreased CAM function at night (Dodd et al., 2003; Cushman et al., 2008a). The evolution of temporally integrated carbon metabolism in CAM plants presumably involves rewiring of gene regulatory networks to link these processes with the circadian clock. Although timing of photosynthetic gene expression is to some degree circadian controlled in both C₃ and CAM species, the links between CAM metabolism and circadian clock oscillators may be stronger will be quite different than in C₃ species (Hartwell, 2005; Dever et al., 2015).

CAM is typically described as a complex phenotype largely because of its role in central metabolism, and therefore its recurrent evolution is considered a remarkable and textbook example of convergence. Yet the frequent transitions to CAM across plants suggest that the evolutionary trajectory from C₃ to CAM is not insurmountable. For example, recent work has suggested that increasing flux through existing pathways in C₃ species may be enough to trigger low-level CAM under some scenarios (Bräutigam et al., 2017), and comparative genomics suggest that rewiring of pathways, rather than creating them de novo, probably underpins the transition to CAM (Yin et al., 2018). To add to the difficulty of elucidating a mechanistic understanding of CAM evolution, the CAM phenotype is more accurately represented as a continuum, where plants can be C₃, CAM, or a combination of both pathways called ‘weak’ CAM (hereafter, C₃+CAM) (Silvera et al., 2010; Winter et al., 2015). C₃+CAM plants should exhibit mixed phenotypes at both physiological and genomic scales, and are potentially powerful systems for exploring the transition from C₃ to CAM.

Our understanding of the genetics and genome structure of CAM has come predominantly from studies that involve comparisons between C₃ and CAM tissues sampled from evolutionarily distant species, or from samples taken from one species at different ages or under different environmental conditions (Taybi et al., 2004; Cushman et al., 2008b; Gross et al., 2013; Brilhaus et al., 2016; but see Heyduk et al., 2018). Recent studies have profiled gene expression before and after CAM induction in Mesembryanthemum crystallinum (Cushman et al., 2008b) and in Talinum (Brilhaus et al., 2016). These studies, together with comparison of transcript abundance profiles in photosynthetic (green) and non-photosynthetic (white) parts of pineapple (Annanas comusus) leaf blades, have also provided insights into the regulation of canonical CAM genes (Zhang et al., 2014; Ming et al., 2015). However, RNA sequencing (RNA-Seq) data of closely related C₃ and CAM species, as well as intermediate C₃+CAM lineages, are lacking.

In this study, we compared transcript profiles among three closely related Yucca (Agavoideae, Aspargaceae) species with contrasting photosynthetic pathways: Y. aloifolia consistently has night-time uptake of CO₂ with concomitant malic acid accumulation in leaf tissue, as well as anatomical characteristics indicative of CAM function; Y. filamentosa has typical C₃ leaf anatomy and showed no positive net CO₂ uptake or malic acid accumulation at night; Y. gloriosa, a hybrid species derived from Y. aloifolia and Y. filamentosa, acquires most of its CO₂ from the atmosphere through C₃ photosynthesis during the day with low-level CO₂ uptake at night, but, when drought stressed it transitions to 100% night-time carbon uptake (Heyduk et al., 2016). The leaf anatomy of Y. gloriosa is intermediate between that of the two parental species, and to some extent may limit the degree of CAM it can employ (Heyduk et al., 2016). Clones of all three species (Supplementary Table S1 at JXB online) were grown in a common garden setting under both well-watered and drought-stressed conditions, and were sampled for gene expression and metabolomics over a 24 h diel cycle.

Materials and methods

Plant material and RNA sequencing

RNA was collected during experiments described in Heyduk et al. (2016). Briefly, clones of the three species of Yucca were acclimated to growth chambers with a day/night temperature of 30/17 °C and 30% humidity in ~3 liter pots filled with a 60:40 mix of soil:sand. Photoperiod was a 12 h day, with lights on at 08.00 h and a light intensity of ~380 µmol m⁻² s⁻¹ at leaf level. One clone was kept well-watered for 10 d while the second clone was subjected to drought stress via dry down beginning after the end of day 1. Clones of a genotype were randomly assigned to watered and drought treatment. On the seventh day of the experiment, after plants had water withheld for the five previous days, RNA was sampled every 4 h, beginning 1 h after lights were turned on, for a total of six time points; very old and very young leaves were avoided, and samples were taken from the mid-section of the leaf blade from leaves that were not shaded. Sampling per species was biologically replicated via 3–4 genotypes per species, for a final design of 3–4 genotypes×2 treatments×3 species per time point sampled (Supplementary Table S1). Genotypes from the three species were randomly assigned to three different growth chamber experiments conducted in July 2014, October 2014, and February 2015. Well-watered and drought-stressed clonal pairs were measured in the same experimental month. RNA was isolated from a total of 130 samples (n=36, 47, and 47 for Y. aloifolia, Y. filamentosa, and Y. gloriosa, respectively) using Qiagen’s RNeasy mini kit. DNA was removed from RNA samples with Ambion’s Turbo DNase, then RNA quality was assessed on an Agilent Bioanalyzer 2100. RNA libraries were constructed using Kapa Biosystem’s stranded mRNA kit and a dual-index barcoding scheme. Libraries were quantified via quantitative PCR (qPCR) then randomly combined into four pools of 30–36 libraries for PE75 sequencing on the NextSeq 500 at the Georgia Genomics Facility.

Assembly and read mapping

Reads were cleaned using Trimmomatic (Bolger et al., 2014) to remove adaptor sequences, as well as low-quality bases and reads <40 bp. After cleaning and retaining only paired reads, Y. aloifolia had 439 504 093 pairs of reads, Y. filamentosa had 675 702 853 pairs of reads, and Y. gloriosa had 668 870 164 pairs. Due to the sheer number of reads for each species, a subset was used to construct reference assemblies for each species (14% of total reads, or ~83 million read pairs per species) (Haas et al., 2013). Trinity v. 2.0.6 (Grabherr et al., 2011) was used for digital normalization as well as assembly. The full set of reads from each species library were mapped back to that species’ transcriptome assembly with Bowtie v. 2 (Langmead et al., 2009); read mapping information was then used to calculate transcript abundance metrics in RSEM v. 1.2.7 (Li and Dewey, 2011; Haas et al., 2013). Trinity transcripts that had a calculated FPKM (fragments per kilobase of transcript per million mapped reads) <1 were removed, and an isoform from a component was discarded if <25% of the total component reads mapped to it.

To further simplify the assemblies and remove assembly artifacts and incompletely processed RNA reads, the filtered set of transcripts for each species was independently sorted into orthogroups, or inferred gene families, that were circumscribed using OrthoFinder (Emms and Kelly, 2015) clustering of 14 sequenced genomes (Brachypodium distachyon, Phalaenopsis equestris, Oryza sativa, Musa acuminata, Asparagus officinalis, Ananas comusus, Elaeis guiensis, Acorus americanus, Sorghum bicolor, Vitis vinifera, Arabidopsis thaliana, Carica papaya, Solanum lycopersicum, and Amborella tricopoda). First, transcripts were passed through Transdecoder (http://transdecoder.github.io/), which searches for ORFs in assembled RNA-Seq data. Transdecoder reading frame coding sequences for each species were individually matched to a protein database derived from gene models from the monocot genome dataset using BLASTx (Altschul et al., 1990). The best hit for each query sequence was retained and used to sort the Yucca transcript into the same orthogroup as the best hit sequence. Assembled Yucca sequences were further filtered to retain only putatively full-length sequences; Yucca transcripts that were shorter than the minimum length of an orthogroup were removed. Transdecoder produces multiple reading frames per transcript, so only the longest was retained. Scripts for orthogroup sorting and length filtering are available at www.github.com/kheyduk/OrthoSort. Read counts for the final set of orthogrouped transcripts were re-calculated using Bowtie v. 2 and RSEM, and analyzed in EdgeR (Robinson et al., 2010) in R 3.2.3, using the trimmed mean of M-values (TMM) normalization. To ensure that using separate genotypes as biological replicates was appropriate, we calculated correlations across the filtered set of transcripts within each time point×treatment×species combination using the PtR analysis script from Trinity and the corr() function in R 3.5.1. Transcriptome quality—both the transcriptomes filtered for low-abundance transcripts and the final gene family sorted assemblies—were assessed for quality by read mapping individual library reads with Kallisto (Bray et al., 2016) to calculate average coverage per transcript as well as the average percentage of transcripts with no reads mapped.

Expression analysis of differentially expressed genes

In a given Yucca species, all libraries were separated by time point; within each time point, quantile-adjusted conditional maximum likelihood via EdgeR (‘classic mode’) was used to find the number of up- and down-regulated genes in response to drought stress, using a P-value cut-off of 0.05 and adjusting for multiple testing with a Benjamini and Hochberg correction. Gene Ontology (GO) annotations for individual genes were obtained from the TAIR10 ontology of each gene family’s Arabidopsis members. GO enrichment tests were conducted for each time point per species, comparing GO categories of differentially expressed genes in well-watered versus drought-stressed treatments, using a hypergeometric test within the phyper base function in R 3.2.3.

Temporal profile clustering of gene expression

To assess larger patterns in the expression data while taking into account temporal patterns across time points, we employed maSigPro (Conesa et al., 2006), using options for read count data (Nueda et al., 2014). maSigPro is a two-step algorithm for profile clustering; the first step involves finding transcripts with non-flat time series profiles by testing generalized linear models with time and treatment factors (using a negative binomial error distribution) against a model with only an intercept (y~1); the second step involves assessing the goodness of fit for every transcript’s regression model and assessing treatment effects. This two-step method is advantageous in that it rapidly reduces a large number of transcripts to only those that show significant variation across time, and it also readily allows users to select transcripts that have a clear expression profile (as assessed by goodness of fit of the model). For the Yucca data, gene regression models were computed on TMM-normalized read counts with a Benjamini and Hochberg corrected significance level of 0.05. Gene models were then assessed for goodness of fit via the T.fit() function, which produces a list of influential genes whose gene models are being heavily influenced by a few data points (in this case, samples). Those genes were removed, and regression models and fit were re-calculated. Genes with significant treatment effects can have either (i) different regression coefficients for the two treatments and/or (ii) different intercepts (i.e. magnitude of expression) between the two treatments. To cluster transcripts with similar profiles, we employed fuzzy clustering via the ‘mfuzz’ option in maSigPro. The clustering steps require a user-defined value for k number of clusters. We assessed the optimal number of clusters for the data of each species by examining the within-group sum of squares for k=1:20 clusters. A k was chosen where the plot has a bend or elbow, typically just before the group sum of squares levels off for higher values of k. A k of 9, 12, and 15 was used for Y. aloifolia, Y. gloriosa, and Y. filamentosa, respectively. To estimate m, the ‘fuzzification parameter’ for fuzzy clustering, we employed the mestimate() function in the Mfuzz package. m of 1.06, 1.05, and 1.05 was estimated for Y. aloifolia, Y. gloriosa, and Y. filamentosa, respectively.

By default, maSigPro clusters only genes that are significantly differentially expressed between treatments. We modified the code for the see.genes() function to fuzzy cluster all transcripts that had non-flat profiles (i.e., significant fit to a polynomial regression across time points), regardless of whether they showed a significant change in expression as a result of drought stress. Afterwards, we found transcripts that were significantly different between treatments with an R² cut-off of 0.7. The modified code for the see.genes() function, as well as a detailed guide to the steps taken for this analysis, are available at www.github.com/kheyduk/Timecourse-RNA.

Gene annotation and gene tree estimation

All transcripts were first annotated by their membership in gene families; gene family annotations were based on the Arabidopsis sequences that belong to the gene family, using TAIR10 annotations. To address homology of transcripts across species, gene trees were constructed from protein-coding sequences of gene families of interest, and included Yucca transcript sequences as well as the 14 angiosperm sequenced genomes. Nucleotide sequences were first aligned via PASTA (Mirarab et al., 2014). Gene trees were estimated via RAxML (Stamatakis, 2006) using 200 bootstrap replicates and the GTRGAMMA nucleotide model of substitution. Raw newick files for gene trees can be found on www.github.com/kheyduk/RNAseq_Yucca.

For genes of interest, additional tests were done to assess whether species differed significantly in their temporal pattern of expression. Because count data cannot be accurately compared between species, we instead used TPM (transcript per million) values. For each gene family of interest, we selected a single transcript per species based on highest expression, which in nearly all cases resulted in usage of immediate orthologs based on the gene tree. TPM values were scaled within each species’ transcript, separately for well-watered and drought-stressed libraries, by the maximum TPM value. All TPM values for each gene for each treatment had a polynomial model fit with degree=5 without distinguishing species, and a second polynomial model that included species as a factor. Using ANOVA, we compared the fits of the two models and report the t-statistic and P-value.

Metabolomics

Samples for starch and metabolomics were collected from a separate experiment conducted in February 2017 at the University of Georgia greenhouses. Growth conditions in the chamber were identical to conditions used when harvesting tissue for RNA-Seq (above), and plants used were the same genotype, but not the same clone, as for RNA-Seq. As this was only a preliminary analysis, samples for starch and metabolites were collected only for the parental species, and only under well-watered conditions. Samples were collected every 4 h starting 1 h after the lights were turned on from six replicate plants per species; replicates were from different genotypic backgrounds (see Supplementary Table S1). Gas exchange data were collected concurrently to ensure plants were behaving the same as when RNA was collected previously (Supplementary Fig. S1). Tissue for starch was flash-frozen in liquid N₂, then later dried in a forced air oven. Samples were ground and 0.02 g was washed first with room temperature acetone, then with 80% EtOH, and finally heated at 90 °C in 1% hydrochloric acid and centrifuged to pellet any remaining tissue (Hansen and Moller, 1975; Oren et al., 1988). A 1:40 dilution of 5% Lugol’s idodine was added to starch extracts and measured in a spectrophotometer at 580 nm. Values were compared against a standard curve made from corn starch dissolved in 1% hydrochloric acid.

Tissue for metabolic analysis was flash-frozen in liquid N₂ then stored at –80 °C until samples were freeze-dried. A 1:1 mixture of MeOH and chloroform (400 μl) was added to 10 mg of freeze-dried, ball-milled (Mini-beadbeater, Biospec products Bartlesville OK, USA) tissue along with adonitol as an internal standard. Mixtures were sonicated for 30 min at 8–10 °C, equilibrated to room temperature, and polar metabolites recovered by liquid phase partitioning after 200 μl of H₂O was added to the extract. A 10 μl aliquot of the aqueous methanol phase was dried and derivatized for GC-MS by adding 15 μl of methoxyamine hydrochloride and incubating at 30 °C for 30 min, then by adding 30 μl of MSTFA and incubating at 60 °C for 90 min. Derivatized samples were analyzed via GC as in Frost et al. (2012). Chromatograms were deconvoluted using AnalyzerPro (SpectralWorks, Runcom, UK). Peak identities were based on NIST08, Fiehnlib (Agilent Technologies; Kind et al., 2009), and in-house mass spectral libraries. Peak matching between samples was based on the best library match according to AnalyzerPro and retention index (Jeong et al., 2004). Initial metabolite peak calls were filtered first by the confidence level of their best library match (>0.5) and then by raw peak area (>1000). Filtered metabolite peak areas were then normalized based on adonitol peak areas. Standard curves were run for ascorbate, sucrose, malic acid, and citric acid to determine absolute concentrations in μmol g^–1 DW.

Normalized values were imported into R v. 3.3.3 and, where appropriate, multiple metabolite peaks were summed to obtain a single value per metabolite. Time points 3 and 6 (last day time point and last night time point) were removed from analysis due to errors in derivitization steps. Remaining values were filtered for sample presence, retaining only metabolites that were found in at least 25% of all samples. The resulting 217 metabolites were imported into maSigPro, where we tested for time-structured abundance using species as a treatment in the design matrix, allowing for polynomials with degree=3, and using a quasipoisson distribution in the glm model.

Data generated are available on NCBI’s Short Read Archive (RNA-Seq data, BioProject #PRJNA413947), or at github.com/kheyduk/RNAseq_Yucca/ (for count matrices and metabolite data).

Results

Photosynthetic phenotypes

As described in previous work, Y. aloifolia conducts atmospheric CO₂ fixation at night via CAM photosynthesis, while Y. filamentosa relies only on daytime CO₂ fixation and the C₃ cycle. Yucca gloriosa, a C₃+CAM intermediate species, uses mostly daytime CO₂ fixation with low levels of nocturnal gas exchange under well-watered conditions, then relies solely on CAM photosynthesis under drought stress (Fig. 1B, C). Gas exchange and titratable acidity measurements shown in Fig. 1 are from prior work when RNA was sampled, though gas exchange patterns were largely consistent in Y. aloifolia and Y. filamentosa during a second round of sampling for metabolites (Supplementary Fig. S1).

Assembly and differential expression

After filtering to remove low abundance transcripts (FPKM <1) and minor isoforms (<25% total component expression), an average of 55 000 assembled transcripts remained per species. For the filtered transcriptome of each species, mean coverage per transcript (and mean percentage of transcripts with a coverage of 0) was 204.44 (5.2%), 165.75(6.41%), and 117.96 (4.72%) for Y. aloifolia, Y. filamentosa, and Y. gloriosa, respectively. Transcripts were then sorted into gene families (orthogroups) circumscribed by OrthoFinder clustering of protein-coding sequences from 14 sequenced plant genomes. Transcripts were removed if their length was shorter than the minimum length for a gene family. Considering only transcripts that sorted into a gene family and had the proper length, transcriptome sizes were reduced further: 19 399, 23 645, and 22 086 assembled transcripts remained in Y. aloifolia, Y. filamentosa, and Y. gloriosa, respectively. The filtered set of transcripts had improved coverage overall: 443.23 (1.44% transcripts uncovered), 344.95 (2.06%), and 342.08 (1.5%) for Y. aloifolia, Y. filamentosa, and Y. gloriosa, respectively. N50 values of the filtered transcript assemblies for Y. aloifolia, Y. filamentosa, and Y. gloriosa were 1925, 1745, and 1745 bp, respectively. Biological replicates had average (and minimum) correlations of 0.94 (0.86), 0.86 (0.74), and 0.91 (0.80) for Y. aloifolia, Y. filamentosa, and Y. gloriosa, respectively.

Differential expression analysis at each time point between well-watered and drought-stressed samples showed distinct patterns in the three species (Fig. 1D). The effect of drought on expression was greatest 1 h after the start of the light period in the CAM species Y. aloifolia, but just before light in the C₃ species Y. filamentosa. Yucca gloriosa (C₃+CAM intermediate) had near constant levels of differentially expressed transcripts across the entire day/night cycle. GO enrichment tests showed general processes, such as protein ubiquitination, RNA processing, mitochondrion, and photosynthesis (including chloroplast and thylakoid membrane organization), as being commonly enriched in the differentially expressed transcripts (Supplementary Table S2).

Transcripts of each species were classified as time structured if their expression across time under either well-watered or drought conditions could be better described with a polynomial regression, rather than a flat line, with significance-of-fit corrected for multiple tests (Supplementary Table S3). There were 612, 749, and 635 transcription factor annotated transcripts with time-structured expression profiles in Y. aloifolia, Y. filamentosa, and Y. gloriosa, respectively. Of those, 92, 62, and 83 were differentially expressed in Y. aloifolia, Y. filamentosa, and Y. gloriosa, respectively, under drought (Supplementary Table S4). Putative CAM pathway genes (Fig. 1) largely showed the expected expression patterns in Y. aloifolia, and additionally all three species shared time-structured gene expression patterns for some canonical CAM genes regardless of photosynthetic pathway. In all three Yucca species PEPC, its kinase PPCK (Fig. 2A, B), as well as genes coding decarboxylating enzymes NAD/P-ME, PPDK and, PEPCK showed time-structured expression (Supplementary Fig. S2). PEPC and PPCK (Fig. 2A, B) exhibited time-structured expression in all three species, though they were only differentially expressed between well-watered and drought treatments in Y. gloriosa. Expression of PEPC in Y. filamentosa was much lower in terms of TPM, but had the same temporal pattern as both Y. aloifolia and Y. gloriosa [well-watered, F_(12,46)=1.38, P<0.212; drought, F_(12,42)=0.53, P<0.866]. In all three species, PPCK was most highly expressed at night (Fig. 2B), consistent with its role in activating PEPC for dark carboxylation, and showed no difference in temporal expression across species [well-watered, F_(12,46)=0.85, P<0.605; drought, F_(12,42)=1.21, P<0.309]. CA, involved in conversion of CO₂ to HCO₃^–, had only three transcripts that were temporally structured in their expression in Y. aloifolia; two α-CA and one γ. In none of these cases did expression increase at night as might be expected (Supplementary Fig. S3).

Fig. 2.

Gene expression for PEPC (A) and PPCK (B) in all three Yucca species, shown for day (white background) and night (gray background) time points, under both well-watered (blue bar) and drought-stressed (red bar) conditions. Mean TPM ±1 SD is plotted. Transcripts shown represent all copies in each gene family (note, an alternative gene family, which is expressed at a lower level, exists for PEPC and is not shown here), and all transcripts were significantly time structured in all three species.

Metabolomics

Of the 214 metabolites that were present in at least 25% of samples, 87 had a significant fit to a polynomial regression line (Fig. 3), with 16 having significant differences in either abundance or temporal regulation between Y. aloifolia and Y. filamentosa (R²>0.5) (Supplementary Table S5). Starch degradation is one possible route CAM plants can use for the nightly regeneration of PEP. Whilst starch content overall was comparable between the C₃ and CAM species, there was no net dark depletion of starch in the CAM species, suggesting little reliance on starch for nocturnal generation of PEP in the CAM Yucca (Fig. 4A). In contrast, starch is degraded at night in the C₃ species and hybrid, with increased levels of α-glucan phosphorylase (PHS), an enzyme responsible for phosphorolytic degradation of starch (Smith et al., 2005; Borland et al., 2016). Maltose, a starch-derived breakdown product, was substantially elevated in the C₃ species compared with the CAM species (Fig. 4B). The difference in maltose content was reflected by higher expression of the maltose exporter MEX1 gene in Y. filamentosa (Fig. 4B). Malic acid had greater turnover in Y. aloifolia, and transcript abundance of malate dehydrogenase (MDH), responsible for interconversion of malic acid and oxaloacetate (Fig. 4C), was likewise higher in the CAM species. Expression of MDH across the three species was similar under well-watered conditions [F_(12,46)=1.89, P<0.061], but differed by species under drought [F_(12,42)=2.09, P<0.039].

Fig. 3.

Heatmap of abundance of a biologically meaningful subset of metabolites that had time-structured fluctuations in abundance (i.e. could be fit to a polynomial), shown for each species during the day (white bar) and night (black bar).

Fig. 4.

Gene expression and related metabolites, shown over a day (white bar) and night (black bar) period, under both well-watered (blue bar) and drought-stressed (red bar) conditions in RNA-Seq data only. The maximum TPM for each gene is shown below the gene tree; asterisks indicate transcripts that were significantly time structured, with red coloring indicating differential expression between watered and drought. Gene tree circles are color coded by species [dark gray=Y. aloifolia (CAM), white=Y. filamentosa (C₃), light gray=Y. gloriosa (C₃+CAM)]. The colors are carried to the metabolite plots (dark gray bars=Y. aloifolia, white bars=Y. filamentosa). (A) Starch synthase 1 (SS1), involved in the production of starch; glucan phosphorylase (PHS), involved in degradation of starch. (B) Maltose exporter 1 (MEX1), transports maltose out of plastids. (C) Malate dehydrogenase (MDH), responsible for interconversion of oxaloacetate and malic acid.

An alternative source of carbohydrates for PEP can come from soluble sugars. Several soluble sugars had higher abundance in Y. filamentosa, including fructose and glucose (Fig. 5A). Fructose (but not glucose or sucrose) had a significant temporal difference between Y. aloifolia and Y. filamentosa (Supplementary Fig. S4), with concentrations in Y. filamentosa decreasing during the dark period while concentrations in Y. aloifolia remained flat. Both species accumulate similar amounts of sucrose (Fig. 5A; Supplementary Fig. S5), indicating no difference in the amount of hexoses dedicated to sucrose production. There is a slight temporal change across the day–night period in Y. aloifolia, but it was not significant based on polynomial regression analysis. Gene expression also does not implicate conversion of hexose to triose phosphates as a mechanism for generating differences in hexose concentrations: both Y. aloifolia and Y. filamentosa express fructose 1,6-bisphosphate aldolase (FBA) at equal levels, although different gene copies are used in Y. aloifolia versus Y. filamentosa (Fig. 5A), and Y. filamentosa has a significantly different timing of expression relative to Y. aloifolia under both well-watered [F_(12,46)= –4.293, P=8.99e-05] and drought-stressed conditions [F_(12,42)= –6.79, P=3.55e-08]. The different FBA paralogs expressed in Y. aloifoia and Y. gloriosa compared with Y. filamentosa represent alternative localizations; the FBA homolog expressed in Y. filamentosa has an Arabidopsis ortholog which localizes to the chloroplast, while the copy expressed in Y. aloifolia and Y. gloriosa is orthologous to the cytosolic Arabidopsis copy. FBA in the chloroplast acts as an intermediate step in the Calvin cycle. The cytosolic version found in the CAM and C₃+CAM intermediate species is thought to be involved in glycolysis and gluconeogenesis.

Fig. 5.

Gene expression and related metabolites, shown over a day (white bar) and night (black bar) period, under both well-watered (blue bar) and drought-stressed (red bar) conditions in RNA-Seq data only. The maximum TPM for each gene is shown below the gene tree; asterisks indicate transcripts that were significantly time structured, with red coloring indicating differential expression between watered and drought. Gene tree circles are color coded by species [dark gray=Y. aloifolia (CAM), white=Y. filamentosa (C₃), light gray=Y. gloriosa (C₃+CAM)]. The colors are carried to the metabolite plots (dark gray bars=Y. aloifolia, white bars=Y. filamentosa). (A) Fructose bisphosphate aldolase (FBA), responsible for interconversion of fructose-6-P and triose phosphates, and sucrose phosphatase (SPP) produces sucrose from glucose and fructose molecules. (B) Triose phosphate isomerase (TPI) interconverts the two forms of triose phosphates, APE2 is a triose phosphate transporter out of the plastid, and G3PDH is involved in glycolysis.

Triose phosphates are too small to measure through GC-MS metabolomics methods, but genes associated with interconversion of glyceraldehyde 3-phosphate and dihydroxyacetone phosphate (triose phosphate isomerase, TPI) as well as genes involved in transport (triose phosphate transporter, APE2) show higher expression in Y. aloifolia compared with both Y. filamentosa and Y. gloriosa (Fig. 5B), though genes do not significantly differ in temporal expression pattern based on post-hoc linear model tests. G3PDH (glyceraldehyde-3-phosphate dehydrogenase), an enzyme involved in downstream branches of glycolysis, likewise has the highest expression in Y. aloifolia; accounting for species in the linear model of gene expression significantly increases fit of the model under both well-watered [F_(12,46)=2.45, P<0.015] and drought-stressed conditions [F_{(12, 42)}=4.94, P<4.97e-5].

Metabolites involved in reactive oxygen species (ROS) scavenging pathways also showed large differences between Y. aloifolia and Y. filamentosa. Vitamin C, or ascorbic acid, was present at much higher levels in Y. aloifolia (Fig. 6A; Supplementary Fig. S5), as was the oxidized form dehydroascorbic acid (Fig. 6B). Neither had a temporal expression pattern, however, indicating constant levels of both metabolites across the day–night period. Previous work has implicated increases in ascorbic acid as a method for CAM plants to remove ROS produced by high nocturnal respiration (Abraham et al., 2016), but genes involved in mitochondrial respiration (cytochrome c, CYT-c) are similarly expressed in all species (Fig. 6C). The biogenesis of ascorbate through galactono-gamma-lactone dehydrogenase (GLDH) is not significantly different between species (Fig. 6C).

Fig. 6.

Abundance over the day (white bar) and night (black bar) period for (A) ascorbic acid, (B) dehydroascorbic acid, and (C) citric acid. Gene expression for (D) cytochrome c (CYT-c), (E) galactono-gamma-lactone dehydrogenase (GLDH), and (F) phosphoglycolate phosphatase (PGP) over the day (white bar) and night (black bar), under both well-watered (blue bar) and drought-stressed (red bar) conditions. The maximum TPM for each gene is shown below the gene tree; asterisks indicate transcripts that were significantly time structured, with red coloring indicating differential expression between watered and drought. Gene tree circles are color coded by species .

Alternatively ROS might be produced from daytime activities; however, whether or not CAM plants reduce oxidative stress (via reduced photoihibition; Adams and Osmond, 1988; Griffiths et al., 1989; Pieters et al., 2003) or instead produce high levels of O₂ (from increased electron transport behind closed stomata; Niewiadomska and Borland, 2008) remains unclear in the literature. Regardless, the photorespiratory pathway has been proposed as a means of protection from oxidative stress (Kozaki and Takeba, 1996), but reduced photorespiration is thought to be a key benefit of CAM photosynthesis. Kinetic modeling of Rubisco’s oxygenase/carboxylase activity suggested that CAM plants either have equivalent levels of photorespiration to C₃ species or are reduced by as much as 60% (Lüttge, 2010). An increase in antioxidants such as ascorbic acid would therefore be beneficial in CAM plants if photorespiration is reduced relative to C₃, while O₂ and ROS are still produced at the same rate. Metabolites that take part in photorespiration, including glycine and serine, peak in the day period in C₃Y. filamentosa, and are generally much higher in Y. filamentosa than seen in Y. aloifolia (Fig. 3) (Scheible et al., 2000; Novitskaya et al., 2002). However, phosphoglycolate phosphatase (PGP)—involved in the first step in breaking down the photorespiratory product 2-phosphoglycolate—is elevated, but not significantly different, in the C₃ and C₃+CAM species (Fig. 6F). Further work is needed to elucidate functional connections between photorespiratory stress and CAM photosynthesis.

Discussion

CAM pathway genes

Time-structured expression of key CAM genes in a C₃ species of Yucca suggests that ancestral expression patterns required for CAM may have pre-dated its origin in Yucca. This important observation is in line with recent suggestions that the frequent emergences of CAM from C₃ photosynthesis was facilitated by evolution acting directly on a low flux pathway already in place for amino acid metabolism (Bräutigam et al., 2017). Flux analysis using ¹³C-labeled substrates has shown that C₃ plants can use organic acids at night to fuel amino acid synthesis (Gauthier et al., 2010; Szecowka et al., 2013). Such carbon labeling experiments, in addition to shared gene expression shown here, suggest that the evolution of CAM may simply require increasing the flux capacity of existing carboxylation pathways in C₃ plants, without the need for extensive rewiring or diel rescheduling of enzymes (Bräutigam et al., 2017).

Carbohydrate metabolism

To provide the nightly supply of PEP needed as substrate for CO₂ and PEPC, CAM plants break down either soluble sugars (including polymers of fructose in fructans) or starches to regenerate PEP via glycolysis. Work in the closely related genus Agave indicates that soluble sugars are the main pool for nightly PEP regeneration (Abraham et al., 2016). As seen in Agave, the CAM Yucca species Y. aloifolia uses soluble sugars as a carbohydrate reserve for PEP requirements, and probably as a source for other nocturnal metabolism, while C₃Y. filamentosa probably relies on starch pools. Although starch concentrations were largely equal in Y. aloifolia and Y. filamentosa, degradation of starch to form maltose was significantly higher in Y. filamentosa. The low levels of MEX1 and PHS1 expression in Y. aloifolia further suggest that starch degradation has been down-regulated (or up-regulated in Y. filamentosa) as Y. aloifolia and Y. filamentosa diverged only 5–8 million years ago (Good-Avila et al., 2006; Smith et al., 2008). Yucca gloriosa has intermediate expression of MEX1 and PHS relative to its parental species, indicating some reliance on starch for carbohydrates like its C₃ parent.

Soluble sugars, such as glucose, fructose, and sucrose, can serve as an alternative source of carbohydrates for glycolysis. In Agave, fructans (chains of fructose monomers) are the predominant source of nocturnal carbohydrates for PEP (Wang and Nobel, 1998; Arrizon et al., 2010). Agave, relative to Arabidopsis, has temporal regulation of soluble sugar production and a 10-fold increase in abundance (Abraham et al., 2016). In general, there was a lack of diel turnover in soluble sugars in Y. aloifolia, although it is possible that unmeasured fructans constitute the majority of the carbohydrate pool. With one exception, neither species shows temporal fluctuation of abundance of soluble sugars (Y. filamentosa exhibits time-structured variation in fructose concentrations; Supplementary Fig. S4). Sucrose concentrations are largely equal between the two species, while glucose and fructose are elevated in C₃Y. filamentosa. Glucose and fructose are the building blocks of sucrose, but it is unclear from the metabolite and transcript data alone whether these are elevated in Y. filamentosa due to degradation of sucrose, or for some other reason.

Many of the genes involved in glycolytic processes had much higher expression in Y. aloifolia, suggesting that the breakdown of triose phosphates into PEP is occurring at a higher rate in CAM Yucca. Fructose bisphosphate aldolase (FBA) acts as a major control point for glycolysis by converting fructose 1,6-bisphosphate into triose phosphates and is also involved in the reverse reaction in the Calvin cycle (formation of hexose from triose phosphates). FBA expression is initially high in both parental species (Fig. 5A), then rapidly drops in the C₃ species and is sustained throughout the day period in both Y. aloifolia and Y. gloriosa, although alternative copies of this gene are used in CAM and C₃ parental species. FBA is thought to be driven toward triose phosphate production within the cytosol (the gene copy expressed in the CAM species), whereas the chloroplastic copy expressed in the C₃ species is involved in Calvin cycle carbohydrate synthesis. Gene expression patterns therefore suggest that while Y. aloifolia expresses FBA for production of triose phosphates for glycolysis and PEP regeneration, Y. filamentosa uses the reverse reaction to synthesize greater concentrations of soluble sugars.

In total, metabolite data and gene expression suggest that soluble sugar pools are not the critical part of carbon metabolism for CAM in Yucca; instead, it is more likely that flux through the system, particularly through glycolysis, is important for the maintenance of PEP and thus effective CAM function. The apparent variation in which carbohydrate pool is used—starch for C₃, soluble sugars for CAM—is surprising, given the relatively short evolutionary distance between the two species. The functional importance of large glucose and fructose accumulation and retention in Y. filamentosa relative to Y. aloifolia is unclear. Roles for the hexoses (glucose and fructose) in C₃ plants include hormonal signaling (Zhou et al., 1998; Arenas-Huertero et al., 2000; León and Sheen, 2003), plant growth and development (Miller and Chourey, 1992; Weber et al., 1997), and gene expression regulation (Koch, 1996); because CAM plants undergo all of the same metabolic processes, the stark difference in concentrations of these hexoses in C₃ and CAM Yucca species remains to be investigated. Metabolite data presented in this study are preliminary and are meant to highlight differences in metabolite pools between these closely related C₃ and CAM species. Further studies to describe how the parental C₃ and CAM species metabolomes behave under drought stress, as well as the metabolic profile of the C₃+CAM Yucca hybrid, will provide a greater understanding for the links between metabolites, carbon metabolism, and photosynthesis.

Antioxidant response in CAM

Previous work in Agave discovered high levels of ascorbate and NADH activity relative to C₃ Arabidopsis, which was thought to be due to increases in mitochondrial activity at night in CAM species relative to C₃ (Abraham et al., 2016). Similarly, Y. aloifolia has much higher levels of ascorbic acid and dehydroascorbic acid relative to its C₃ sister species and implies different requirements for antioxidant response between the two species. For example, respiration rates might be expected to be higher in CAM species at night to sustain the active metabolism. Although citric acid abundance is nearly identical in C₃ and CAM Yucca species, expression of cytochrome c, a part of the mitochondrial electron transport chain, is higher in the CAM Y. aloifolia. Alternatively, due to inhibited photorespiratory response in the CAM species, an alternative form of ROS scavenging may be needed to regulate oxidation in the cells resulting from either photoinhibition or O₂ accumulation from electron transport behind closed stomata during the day. It is possible that CAM species are using antioxidant metabolites such as ascorbic acid to prevent oxidative stress, rather than relying on photorespiration. Indeed genes (PGP) and metabolites (glycine and serine) involved in photorespiration were expressed at a lower level and found in lower abundance, respectively, relative to C₃Y. filamentosa. Whether or not increased antioxidant response is required for CAM to function efficiently in plants is unknown, and future work discerning ROS production and mitigation—particularly in the hybrid Y. gloriosa—will inform understanding of the role of ROS scavenging and its impact on photosynthetic functions.

Transcriptomics and metabolomics of the parental species Y. aloifolia and Y. filamentosa revealed many changes to regulation, expression, and abundance. The most notable differences included degree of expression of core CAM genes (Fig. 1A) and fundamental differences between the C₃ and CAM species in starch and soluble sugar metabolism. The diploid hybrid species, Y. gloriosa, exhibited gene expression profiles more similar to its CAM parent, Y. aloifolia, than the C₃ parent, Y. filamentosa. Additionally, the CAM species Y. aloifolia had heightened antioxidant response (both in metabolites and in gene expression) relative to Y. filamentosa, indicating that the operation of CAM imposes a significant oxidative burden. Despite these differences, similarities exist in levels of gene expression of a few CAM genes (PEPC, for example) between the C₃ and CAM Yucca species studied here, perhaps indicating shared traits in an ancestral genome that may have facilitated the convergent evolution of CAM photosynthesis within the Agavoideae. Continued comparative research on closely related C₃ and CAM species, as well as intermediate forms, is necessary to understand the genetic underpinnings of CAM, as well as to determine whether ancestral genetic enabling has facilitated the evolution of CAM.

Supplementary data

Supplementary data are available at JXB online.

Table S1. Genotypes sequenced through RNA-Seq.

Table S2. GO term enrichment.

Table S3. Gene family circumscription, cluster assignment, and annotation.

Table S4. Differentially expressed transcription factors.

Table S5. Significantly different metabolites between C₃ and CAM Yucca.

Fig. S1. Gas exchange data for metabolomics samples.

Fig. S2. Decarboxylation gene expression.

Fig. S3. Carbonic anhydrase gene expression.

Fig. S4. Temporally variable metabolite regressions.

Fig. S5. Calibrated concentrations of key metabolites.