Metabolic and transcriptomic study of pennycress natural variation identifies targets for oil improvement

Summary

Pennycress (Thlaspi arvense L.), a member of the Brassicaceae family, produces seed oil high in erucic acid, suitable for biodiesel and aviation fuel. Although pennycress, a winter annual, could be grown as a dedicated bioenergy crop, an increase in its seed oil content is required to improve its economic competitiveness. The success of crop improvement relies upon finding the right combination of biomarkers and targets, and the best genetic engineering and/or breeding strategies. In this work, we combined biomass composition with metabolomic and transcriptomic studies of developing embryos from 22 pennycress natural variants to identify targets for oil improvement. The selected accession collection presented diverse levels of fatty acids at maturity ranging from 29% to 41%. Pearson correlation analyses, weighted gene co‐expression network analysis and biomarker identifications were used as complementary approaches to detect associations between metabolite level or gene expression and oil content at maturity. The results indicated that improving seed oil content can lead to a concomitant increase in the proportion of erucic acid without affecting the weight of embryos. Processes, such as carbon partitioning towards the chloroplast, lipid metabolism, photosynthesis, and a tight control of nitrogen availability, were found to be key for oil improvement in pennycress. Besides identifying specific targets, our results also provide guidance regarding the best timing for their modification, early or middle maturation. Thus, this work lays out promising strategies, specific for pennycress, to accelerate the successful development of lines with increased seed oil content for biofuel applications.

Introduction

Thlaspi arvense, commonly known as pennycress, is an emerging bioenergy cover cash crop that belongs to the Brassicaceae family. It can be grown in rotation with other major crops, such as soybean or corn, without displacing them (Cubins et al., 2019), and decrease soil erosion and nutrient loss on otherwise barren farmland (Johnson et al., 2017; Noland et al., 2018). Pennycress is widely distributed and is highly adapted to a wide variety of climatic conditions, presenting extreme freezing tolerance (Zhou et al., 2007). Wild pennycress seeds contain around 30%–35% oil with a unique fatty acid (FA) composition corresponding to approximately 94% of unsaturated FAs, making the oil suitable for biodiesel production, with better low‐temperature properties than other commodity feedstocks (Moser et al., 2016). The content of erucic acid, the most abundant FA in pennycress seeds, is particularly valuable for jet fuel, lubricants, cosmetics, and plastics industries. While pennycress yield potential is high, studies of natural populations have shown a high degree of yield variability (200–2400 kg/ha, Cubins et al., 2019). A recent techno‐economics analysis of pennycress oil as a source of renewable jet fuel highlighted that yield and oil content were among the highest sensitivity variables for pennycress production (Mousavi‐Avval and Shah, 2020). Thus, despite its numerous suitable characteristics, improving oil content and yield is required for establishing pennycress in the market.

Molecular tools for pennycress genetic optimization are being developed rapidly due to its close similarity to Arabidopsis. Pennycress is amenable to Agrobacterium‐mediated transformation by simple floral dip infiltration, and a predominantly one‐to‐one correspondence with Arabidopsis genes has been found when the diploid genome pennycress inbred line Spring 32–10 was sequenced (McGinn et al., 2019). Because of these attributes, pennycress is a suitable target for seed oil biosynthesis engineering.

Pennycress, as other Brassicaceae, predominantly store oil in the embryo and rely on nutrients (sugars and amino acids) provided by the parent plant. Oil biosynthesis requires coordinated integration between different cellular compartments (Baud and Lepiniec, 2010; Ver Sagun et al., 2023). De novo FA synthesis takes place in the plastids and produces FAs of 16 or 18 carbons. FAs can be further elongated by the FA elongase complex, a membrane‐bound multienzyme complex of the endoplasmic reticulum (ER) that uses cytosolic malonyl‐CoA as a substrate. Polyunsaturated FAs found in oils are the result of the activity of membrane‐associated FA desaturases (Baud and Lepiniec, 2010). Subsequently, the different FAs are incorporated into triacylglycerides (TAGs) in the ER and accumulated in oil bodies (OB) that are released into the cytosol (Pyc et al., 2017). Each part of this process could be independently optimized to improve oil yield (Ver Sagun et al., 2023). However, integrated metabolic engineering approaches showed better outcomes in vegetative tissues (Vanhercke et al., 2014, 2019) and could also be applied to seed oil accumulation. FA production can be ‘pushed’ by overexpression of transcription factors, such as WRINKLED1 that positively regulates the expression of the chloroplast FA synthase genes (Kong et al., 2019). Because of the feedback inhibition of FA biosynthetic enzymes by the accumulation of acyl‐CoA or acyl‐ACP (Andre et al., 2012), additional mechanisms, such as diacylglycerol acyltransferase, that can ‘pull’ the acyl groups from the chloroplast into the ER for TAG assembling, need to be provided (Jako et al., 2001; Roesler et al., 2016). Besides, the overexpression of OB proteins like oleosins has the potential to secure packaging to increase TAG content (Zhang et al., 2019). Finally, to guarantee ‘protection’, lipid degradative processes, normally associated with the onset of germination, need to be reduced or turned off (Ding et al., 2019; Kim et al., 2014). Although some genes involved in oil synthesis and storage are common to many species, a combination of factors must be evaluated, including the identity of the target genes, their expression strength, and the developmental stage.

Analysis of natural variation in pennycress previously reported significant genetic diversity that could explain some of the phenotypic differences observed in the germplasm (Frels et al., 2019). Understanding the strategies that characterize different pennycress accessions with different oil levels at maturity will be a crucial step in reducing the number of potential targets. Although comparative studies considering the metabolic variations between two accessions with contrasting oil levels were useful to identify potential key components of metabolism that could boost oil production (Johnston et al., 2022), the target discovery was limited to the specific differences between those two lines. To have a more extensive coverage of the strategies that pennycress put in place to achieve higher levels of oil, we investigated here 22 geographically dispersed natural accessions. We analysed biomass differences, and the underlying metabolome and transcriptome of their developing embryos to pinpoint promising candidates for oil improvement. This study identified metabolites and genes whose level or expression, respectively, correlated with oil content at maturity. Their role in fatty acid synthesis and their potential in terms of targets for metabolically engineering pennycress to improve oil content are discussed.

Results

Geographic dispersion and workflow

Twenty‐two accessions, originally collected from diverse geographic regions around the world and whose genetic diversity was previously studied (García Navarrete et al., 2022), were used in this study to identify potential targets associated with pennycress seed oil accumulation (Table S1). The different lines were grown under greenhouse conditions and reached a similar reproductive stage, with comparable flowers and siliques structures (Figure 1). Embryo samples were collected at two developmental stages and at maturity (Figure 1e). The time points will be further referred to as 10 days after pollination (DAP) or early stage, 16DAP or middle stage, and mature stage. The selected points correspond to the beginning of the reserve accumulation and a middle point in the process before maturity (Tsogtbaatar et al., 2015). We performed biomass analysis in embryos from all three stages, as well as targeted metabolomics and transcriptomics studies at both developmental points (Figure 1).

Figure 1.

Pennycress vegetative and reproductive organs, and experimental workflow. Typical pennycress plant at reproductive stage (a), flower at 1 day after pollination (DAP, b), immature pod at 16DAP (c), and mature pod at approximately 28DAP (d) are shown. (e) Sample collection and analyses.

Broad biomass diversity in the set of accessions reveals correlations between biomass components and final fatty acid content

To determine the extent of natural biomass variability in the mature embryos of the selected line collection, the embryo dry weight (DW), FA and protein content for each accession were measured at maturity (Figure 2; Figure S1). It is worth mentioning that starch content in mature embryos was measured to be <0.5% for all accessions under investigation (data not shown). The DW of the embryos ranged from 0.50 mg in PC19 to 0.86 mg in PC1 (Figure 2a). The levels of protein and FAs per embryo followed a similar pattern to the DW (Figure 2a), indicating the high influence of the embryo size in the final content in FA/embryo or protein/embryo. To circumvent the embryo DW impact, FA and protein content were expressed as a percentage of the DW (Figure 2b). The percentage of total FA at maturity (FAMat) ranged from 29% to 41% across the accessions. The line PC22, (aka TAMN106, Table S1), not only presented the highest total FA percentage (41%, Figure 2b), but also the highest level of oil per embryo (325 μg FA/embryo, Figure 2a). PC22, PC8, PC15 and PC10 showed the highest oil percentages while PC12, PC2, PC6 and PC19 presented the lowest levels (Figure 2b; Figure S1); herein referred to as high oil (HO) and low oil (LO) lines, respectively. In addition, the protein content varied between 33% in PC19 and 46% in PC17 (Figure 2b; Figure S1). Therefore, our collection presented enough natural variation to investigate metabolic strategies to achieve higher oil levels. Then, the correlation between oil and protein, and between oil and DW, previously reported to be negative for other species, was investigated in this set of pennycress accessions. Our results showed no significant correlation between those variables (Figure S2a,b).

Figure 2.

Natural variation in seed biomass composition at maturity. (a) Biomass expressed as micrograms per embryo, where the total height of the bar corresponds to the embryo weight (μg) and the lines are ordered from the highest to the lowest μg oil per embryo. (b) Biomass expressed as a percentage (mg/100 mg DW) and the lines are ordered from the highest to the lowest percentage of oil. (c) Each fatty acid is expressed as a percentage of total FA content and the lines are ordered from the highest to the lowest percentage of erucic acid content. FA: Fatty acids, Prot: Proteins, and Others: refers to the remaining biomass content regarding the embryo weight or to 100% in A and B respectively. C16:0, palmitic acid; C18:0, stearic acid; C18:1, oleic acid; C18:2, linoleic acid; C18:3, α‐linolenic acid; C20:0, arachidonic acid; C20:1, gondoic acid; C20:2, eicosadienoic acid; C22:1, erucic acid C22:2, docosadienoic acid; and 24:1, nervonic acid. Four biological replicates were used in each case. Error bars represent the standard deviation.

To investigate if the natural variation in oil content was associated with a divergent FA composition, the content of the individual FA was determined. In general, pennycress embryos contained palmitic (C16:0), stearic (C18:0), oleic (C18:1), linoleic (C18:2); α‐linolenic (C18:3), arachidonic (C20:0), gondoic (C20:1), eicosadienoic (C20:2), erucic (C22:1), docosadienoic (C22:2), and nervonic (24:1) acids. The most abundant FA in all the lines was erucic acid, followed by the FAs with eighteen carbons and one to three double bonds (Figure 2c). While stearic and palmitic acid proportions showed the strongest negative correlations with FAMat (Figure S2c), erucic acid levels had the strongest positive correlation (Figure S2d).

The individual analysis of each biomass component at early, middle, and end maturation, showed a rather uniform behaviour among pennycress accessions (Figure S3). HO lines, like PC22 and PC8, shared a similar pattern to other lines, such as PC14 or PC20 (Figure S3). Along the developmental progression, the percentage of oil content increased while the protein level was more constant, with only a small decrease at 16DAP for several lines (Figure S3). Although starch levels at maturity were extremely low, its content reached a maximum at 16DAP (Figure S3). Thus, despite differences in absolute levels, pennycress lines accumulated biomass components with similar patterns. FA composition varied across the developmental stages and lines. Indeed, the most abundant FA was found to be linoleic acid at 10DAP, while erucic acid held the majority at 16DAP and maturity (Figure S4; Figure 2c). We carried out a correlation analysis to determine if the proportion of each FA during development was associated with FAMat. At 10DAP, only the percentage of oleic acid showed a significant positive correlation with FAMat (r = 0.27, Table S2). At 16DAP, the proportion of linolenic acid had the strongest negative correlation with FAMat (r = −0.39, Table S2). The total FA and protein contents at 16DAP presented a positive and negative correlation with FAMat, respectively (r = 0.30 and r = −0.33). These associations are important to determine how the biomass components during development may influence the final seed composition at maturity.

Metabolic divergence among pennycress accessions highlights distinctive biochemical characteristics in high oil lines

To detect early metabolic differences that could influence the final oil content in pennycress seeds, 78 intermediates of central metabolism were quantified in developing embryos at 10 and 16DAP (Data S1). Interestingly, 14 out of the top 15 most variable metabolites were the same at 10 and 16DAP, namely sucrose, glucose, malate, citrate, inositol, alanine, arginine, asparagine, aspartate, serine, glutamate, glutamine, proline, and histidine (Table S3). This result highlights the contribution of these compounds to the variance within the set of accessions under investigation.

Two complementary approaches were used to search for variables associated with FAMat: (i) Pearson correlation analysis evaluated the presence of a dose‐dependent association between each compound and FAMat, using the metabolite profiling data from the 22 lines; and (ii) biomarkers identification determined the fold change and the cut‐off value for a given metabolite that allows the separation of the HO lines from the LO ones, only taking the data from these lines into consideration for the analysis. Following the first strategy, the identification of possible associations between the absolute level of each metabolite during development and the FAMat was analysed by Pearson correlations. Out of the 78 metabolites quantified, only seven and six showed significant correlations at the early and middle stages, respectively (Table S4). Histidine, homoserine, arginine, and citrulline were negatively correlated with FAMat at 10DAP with a Pearson correlation coefficient (r) of −0.27 to −0.26, while at 16DAP citrulline and ornithine were found to be negatively correlated (−0.37 < r < −0.32). On the other hand, positive correlations (0.26 < r < 0.28) were observed for glucose, 2‐ketoglutarate, and quinic acid at 10DAP, and for ribulose 1,5‐bisphosphate at 16DAP. Finally, two metabolites showed negative correlations at 16DAP: nicotinic acid (r = −0.27) and UDP‐xylose (r = −0.26). Pearson correlation analysis identified that the quantities of these metabolites during embryo development are associated with the capability of pennycress accessions to produce and store oil.

For the second strategy, the focus was on identifying metabolites whose quantities were discriminant between the HO and LO lines at both developmental points. The analysis included the best twenty ratios between two metabolite concentrations, as some studies have suggested that metabolite ratios may provide more information than analysing metabolite quantities individually (Chong et al., 2019). The ratios and individual metabolites at the top of the list, with areas under the Receiver Operating Characteristic Curves (AUC values) closer to one, were in accordance with the correlation results (Table S4). The top four biomarkers identified at 10DAP were arginine/pentose5P (AUC = 0.95), alanine/hydroxyproline (AUC = 0.93), hydroxyproline/GTP (AUC = 0.93), and arginine/6‐phosphogluconate (AUC = 0.92), while at 16DAP, malate/UDP‐glucose (AUC = 0.98), glutamine/quinic acid (AUC = 0.97), asparagine/UDP‐xylose, and citrulline (AUC = 0.93 for both), highlighting a strong influence of the amino acid levels. For each of the mentioned biomarkers, the distribution of the abundance values and their cut‐off are shown in Figure S5. Therefore, the control of the levels of these specific compounds during embryo development could provide a strategy to modulate oil content at maturity.

Gene expression variability among pennycress accessions highlights distinctive processes in high oil lines

With the aim of identifying genes that potentially contribute to the biomass variations among the lines, especially to their oil production, an analysis of the gene expression in 10 and 16DAP embryos from the 22 accessions was performed. Putative functions for each pennycress transcript were inferred by searching their best matches against the Arabidopsis genome. Only 129 pennycress genes were not assigned Arabidopsis homologues. The top 200 most variable genes were identified at each stage, and 92 of those were found highly variable at 10 and 16DAP. In an enrichment analysis of those 200 most variable genes using the KEGG pathways database, the most significantly enriched process was Protein processing in endoplasmic reticulum for both stages. Other enriched processes included Circadian rhythm, Tryptophan metabolism and Terpenoid backbone biosynthesis at 10DAP, Galactose and Glutathione metabolism at 16DAP (Figures S6a,b). Thus, despite the 12% variation in FAMat among the accessions, the changes in the expression of lipid metabolism genes were not as extensive as other processes taking place in the embryos.

Following the same two approaches previously described to pinpoint metabolic targets for oil improvement, genes whose expression pattern correlates with FAMat were identified. Firstly, the Pearson correlation analysis revealed 1196 (587 positive and 609 negative) and 1159 (556 positive and 603 negative) transcripts that significantly correlated with FAMat at 10 and 16DAP, respectively (Table S5). The level of correlations was between 0.6 and 0.3, meaning moderate to weak associations. A pathway enrichment analysis on each gene set revealed that the processes most represented at both developmental stages were those related to Photosynthesis, especially the Antenna proteins as well as the Porphyrin and chlorophyll metabolism (Figures S6c,d and S7). The expression of the genes within this classification was mainly positively correlated with FAMat (Figure S7). Moreover, other enriched categories with higher pathway impact were associated with FA biosynthesis and degradation, Carbon fixation, and Amino acid metabolisms (Figure S6c,d). To provide an overview of the gene functions in the correlation lists (Table S5), a Gene Ontology (GO) enrichment analysis was performed, considering three types of classifications: Biological Processes (BP), Cellular Component (CC), and Molecular Function (MF, Figure 3). At both stages, the genes that showed an association with FAMat were mainly involved in functions occurring in the plastid, highlighting a potential role of this organelle in controlling oil content (Figure 3). As previously observed in the pathway enrichment analysis (Figure S6), the main BP categories were related to photosynthetic processes, with mainly positive correlations. Additionally, developmental stages were distinct in terms of gene MFs associated with FAMat. Indeed, ATPase activity, Tubulin, RNA, and Protein binding were enriched at 10DAP, whereas Acyl carrying proteins, Acyl transferase, NADH dehydrogenase and Oxidoreductase activities appeared in the top 20 enriched categories at 16DAP (Figure 3). It is noteworthy that all the genes involved in Chlorophyll binding and Glyceraldehyde 3‐phosphate dehydrogenase activity were found to be positively correlated with FAMat (Figure 3). While correlation does not imply causation, all these enriched processes presented a certain level of association with the capability of pennycress embryos to accumulate lipids.

Figure 3.

Gene ontology functional classification of genes whose level of expression correlates with the content of FAMat. The list of genes used corresponds to the Arabidopsis homologues. Top twenty significant processes (P‐value < 0.05) from each of the three categories, Biological Processes, Cellular Compartment and Molecular Function are shown in (a) and (b) for the genes at 10 and 16DAP, respectively. The bars were divided according to the proportion of the genes that showed positive (green) or negative (red) correlation in each individual category, and in parentheses, at the beginning of each bar, the actual number of genes that were positive/negative correlated is shown.

Biomarker genes behaving differently in HO and LO lines were also identified at each developmental stage, and partially overlapped with the ones from the correlation analysis (Table S5). The top four gene biomarkers at 10DAP were Ta1.0_19703 (putative calmodulin protein, AUC = 1), Ta1.0_12878 (putative thioredoxin protein, AUC = 1), Ta1.0_19745 (tetratricopeptide repeat like superfamily protein, AUC = 0.99), and Ta1.0_08187 (membrane trafficking family protein, AUC = 0.99), and at 16DAP, Ta1.0_16738 (possible transcription factor from the C2C2‐YABBY family, AUC = 1), Ta1.0_16768 (unknown protein, AUC = 1), Ta1.0_24457 (putative uroporphyrin methylase, AUC = 0.98), and Ta1.0_10189 (thioesterase superfamily protein, ALT, AUC = 0.98). Several of these genes were found to be biomarkers at both developmental stages, highlighting that the related processes also expand along with the reserve accumulation period. Higher expression levels of the unknown protein, the putative calmodulin, and the membrane trafficking protein were seen in the HO lines at 10 and 16DAP, while for the thioesterase and the tetratricopeptide repeat like protein, the levels were higher in the LO lines (Table S5).

To include indirect associations between genes and obtain a robust measurement of co‐expression networks, the analysis at 16DAP was complemented with a Weighted Gene Co‐expression Network Analysis (WGCNA). After identifying co‐expression modules (labelled by different colours), we determined the association between each module's eigengene and FAMat (Figure S8 and Table S6) and revealed four modules with a significant correlation (adjusted P‐value ≤ 0.05). At 16DAP, 271 genes (Table S5) were simultaneously identified as being significantly associated with FAMat under the gene individual correlation analysis, the biomarker identification, and the correlation as modules. One of the positive correlated modules (r = 0.35), Light‐green, grouped 516 genes involved in biological processes such as Ubiquitin‐dependent ERAD (Endoplasmic Reticulum Associated Protein Degradation) pathway, Polysaccharide biosynthetic process, Protein folding, and Cellular carbohydrate metabolic process, according to the module GO BP enrichment analysis (Figure 4). In addition, the module Honeydew1 was also positively correlated with FAMat (r = 0.34) and the genes of this module (380 genes) were related to all the components of the embryo photosynthetic machinery including the Reductive pentose‐phosphate cycle, Chlorophyll biosynthetic process, Granum assembly, and Photosystem I and II (PSI and PSII, Figure 4). On the other hand, the remaining two significantly correlated modules showed a negative association: Floral‐white (r = −0.38) and Salmon4 (r = −0.31). The first one included 117 genes involved in Carotene metabolic processes, Abscisic acid metabolic process, Cellular response to Hypoxia, and Response to jasmonic acid (Figure 4). The largest module, Salmon4, with 3389 genes, was enriched with genes related to Seed maturation, Carboxylic acid catabolic process, Cellular response to hypoxia, Cellular response to abscisic acid stimulus and Cell wall biogenesis (Figure 4). Lastly, the genes from each FAMat‐related module were ranked by their network features using Maximal Clique Centrality (MCC) topological analysis to infer their importance in the biological network. The top 10 ranked genes (hubs) were identified for each module (Figure 4; Table S7). GO CC analysis showed that the hub genes identified from the two positive correlated modules were most enriched in genes whose products have putative chloroplast localization (Table S8), emphasizing the role of chloroplast metabolism in seed oil accumulation in pennycress.

Figure 4.

Gene ontology classification and hub genes identification for the four modules whose eigengene presented a significant correlation with FAMat at 16DAP. For each module associated with FAMat (Light‐green, Honeydew1, Floral‐white and Salmon4), the Pearson correlation coefficient (r), adjusted P‐value, number of genes in the module, the most enriched Biological Processes, and top 10 genes according to the score on the topological analysis Maximal Clique Centrality (MCC) are presented. These selected nodes are shown with a colour scheme where red indicates the most essential genes. GO, gene ontology.

Finally, to support gene annotation based on Arabidopsis homology, a protein subcellular localization likelihood analysis was performed on the potential list of targets identified by Pearson correlation analysis and crossed with the biomarkers list and the FAMat‐related modules (Table S5). The software allows the assignment of proteins to different cellular compartments, and the differentiation between soluble versus membrane proteins. Most of the subcellular localization predicted was consistent with the one described for the homologues in Arabidopsis, enabling the assignment of the putative targets into individual cellular compartments (Table S5).

Oil‐related genes identified as potential targets to improve FA content in pennycress

To extract the genes directly involved in oil metabolism from the list of transcripts associated with FAMat (Table S5), we cross‐referenced them with Arabidopsis thaliana genes related to lipid synthesis, editing, degradation, export, transport, or storage (Lipid Genes, LG in Table S5). Genes were detected for almost every biochemical step; those are presented in their corresponding pathways in Figure 5. Many putative genes involved in plastidic de novo FA biosynthesis showed positive correlations with FAMat (Figure 5; Table S5), including a subunit of the plastidic acetyl‐coA carboxylase (ACCase), that was also identified as a hub gene in the Light‐green module, and several components of the FA synthase complex. Besides plastidic enzymes, significant correlations were found with the cytosolic ACCase, FA elongation reactions and TAG synthesis steps in the ER membrane (Figure 5). Additionally, putative acyl‐CoA synthetases, thioesterases, thiolases, lipid transfer proteins, lipid transporters, ABC transporters, acyl transferases, OB‐associated proteins, intracellular trafficking and vesicular transport proteins showed positive or negative association with FAMat in at least one of the developmental stages (Table S5). Between them, we can highlight putative long‐chain acyl‐CoA synthetases (Ta1.0_05744, Ta1.0_10536) and acyl‐CoA binding proteins (Ta1.0_22173, Ta1.0_01446, Figure 5). Additionally, specific isoforms of microtubules and actin monomers were identified in the correlation analysis, biomarker identification, and MCC analysis as positively associated with FAMat (Figure 5; Tables S5 and S7). Besides OB mobility, no connection between cell cytoskeleton, OB dispersion and oil accumulation has been reported in plants yet.

Figure 5.

Lipid metabolism in developing pennycress embryos. Transcript levels correlated with FAMat are shown with squares. Positive, negative, and no correlation are highlighted by filling the symbols with red, blue, and white, respectively. For each transcript level, the left symbol corresponds to 10DAP, and the right one to 16DAP. The * symbol was used to highlight the hub genes identified by MCC analysis at 16DAP. Scheme adapted from Li‐Beisson et al. (2013). AACT, acetoacetyl‐CoA thiolase (*Ta1.0_14178); AAE acyl activating enzyme (Ta1.0_01787 and Ta1.0_10844); ABMM, ATP binding microtubule motor family protein (Ta1.0_24750); ACBP, acyl‐CoA binding protein (Ta1.0_22173, *Ta1.0_01446); ACCc, cytosolic acetyl‐CoA carboxylase (Ta1.0_04259); ACT2, actin 2 (Ta1.0_16104 and Ta1.0_01238); ADF, actin depolymerizing factor (ADF4: Ta1.0_26108 and ADF6: Ta1.0_25986); ALT, acyl‐lipid thioesterase (Ta1.0_10189); ATK5, kinesin 5 (Ta1.0_21808); DAG, diacylglycerol; DAGL, diacylglycerol lipase; DGAT, acyl‐CoA: diacylglycerol acyltransferase (Ta1.0_22867 and Ta1.0_05088); DHAP, dihydroxyacetone phosphate; ECI, Enoyl‐CoA isomerase (Ta1.0_22882); ER, enoyl‐ACP reductase; erLACS, ER long‐chain acyl‐CoA synthetase (Ta1.0_10536); FAE, fatty acid elongase; FAS, fatty acid synthase; FATA (B), fatty acyl thioesterase A (B); FFA, free fatty acid; G3PDH, glycerol 3‐phosphate dehydrogenase (Ta1.0_18945); GK, glycerol kinase (Ta1.0_08946); GPAT, glycerol 3‐phosphate acyltransferase (Ta1.0_09794); HAD, hydroxyacyl‐ACP dehydrase (Ta1.0_00919 and Ta1.0_20755); KAR, ketoacyl‐ACP reductase (Ta1.0_12150); KAS, ketoacyl‐ACP synthase (KASII: Ta1.0_03905, KASIII: Ta1.0_11965); KAT, ketoacyl‐CoA thiolase (Ta1.0_05404); KCR, ketoacyl‐CoA reductase (Ta1.0_12143); KCS, ketoacyl‐CoA synthase (Ta1.0_08400); KIS, KIESEL gene tubulin folding cofactor (Ta1.0_14925); LDAP, LD‐associated protein (LDAP2:Ta1.0_10486 and LDAP3:Ta1.0_01437); LPAAT, 2‐lysophosphatidic acid acyltransferase; LPCAT, 2‐lysophosphatidylcholine acyltransferase; MAG, monoacylglycerol; MAGL, monoacylglycerol lipase (Ta1.0_09516); MCMT, malonyl‐CoA:ACP malonyltransferase; MHCR, myosin heavy chain related (Ta1.0_25384); OBPA1A, oil body‐associated protein 1A (*Ta1.0_05325); Oleosin (Ta1.0_19034); pACC, plastidic acetyl‐CoA carboxylase (biotin carboxylase subunit: *Ta1.0_09801); pACP, plastidic acyl carrier protein; PAKRP1, phragmoplast‐associated kinesin‐related protein 1 (Ta1.0_14606); PAP, phosphatidate phosphatase; PC, phosphatidylcholine; PDAT, phospholipid:diacylglycerol acyltransferase; PDH, pyruvate dehydrogenase complex; peLACS, peroxisomal long‐chain acyl‐CoA synthetase (Ta1.0_08099); PLA, phospholipase A; pLACS, plastidic long‐chain acyl‐CoA synthetase (*Ta1.0_05744); Pyr, pyruvate; SAD, stearoyl‐ACP desaturase (Ta1.0_04402); SEIPIN1 (Ta1.0_04422); TAG, triacylglycerol; TAGL, triacylglycerol lipase (Ta1.0_26944); TPXL8, cell cycle regulated microtubule associated protein (Ta1.0_01935); TUB1, tubulin 1 (Ta1.0_06454); XIE, myosin family protein with Dil domain (Ta1.0_21316); ZWI, kinesin‐like calmodulin‐binding protein (Ta1.0_21309).

Regarding lipid degradation, correlations between FAMat and phospholipases (PL), triacylglycerol lipases (TAGL), monoacylglycerol lipases (MAGL), GDSL lipases, and proteins with an α/β hydrolase motif were identified (Figure 5; Table S5). Interestingly, two putative α/β hydrolase (Ta1.0_10076 and Ta1.0_01745) whose homologues have not yet been characterized in Arabidopsis, showed a moderate positive correlation, and were identified as oil biomarkers at both developmental stages (Table S5). Another putative α/β hydrolase was identified as a hub gene in the negatively correlated Salmon4 module (Ta1.0_01274). Finally, during embryo development, FAMat was found to be negatively correlated with a peroxisomal long acyl‐CoA synthetase, an acyl‐activating enzyme, an enoyl‐CoA isomerase, a ketoacyl‐CoA thiolase, and an acetoacetyl‐CoA thiolase related to the catabolism of FA (Figure 5). Therefore, lipid anabolic and catabolic pathways should be considered to improve oil content in pennycress.

High oil lines presented higher expression of glycolytic genes as well variation in cell wall metabolism enzymes

Oil production also requires precursors provided by central metabolism in the form of carbon sources, reductive power, and energy. To identify potential targets associated with the provision of supplies for FA biosynthesis in pennycress embryos, we searched for genes involved in central metabolism that presented significant correlations with FAMat (Central Metabolism Genes, CMG in Table S5). The expression of many genes encoding for glycolytic enzymes (cytosolic and plastidic) was positively correlated (Figure 6); and a subunit of a putative ATP citrate lyase was identified as a hub gene in the Light‐green module (Table S7); a higher expression of these genes in HO lines may provide more acetyl‐CoA, carbon source for FA synthesis and elongation. In addition, enzymes involved in competing pathways like cell wall metabolism (e.g. cellulose synthases like, glucan synthases, pectinesterases, rhamnogalacturonan endolyases, galacturonosyltransferases) also showed correlations with FAMat (Table S5). For example, a positive association with FAMat was observed for four putative pennycress pectinacetylesterases (Ta1.0_10773, Ta1.0_19650, Ta1.0_07433 and Ta1.0_01500, 0.49 < r < 0.30, Table S5). On the other hand, the module Salmon4, negatively correlated with FAMat, was enriched in genes that belong to the BP category Cell wall biogenesis, suggesting a redirection of part of the cell wall carbon towards FA in HO lines.

Figure 6.

Pennycress embryo central metabolism highlighting transcripts and metabolites correlated with FAMat or identified as biomarkers. Positively and negatively correlated metabolites are shown with red and blue circles, respectively. Enzymes and proteins whose transcript levels were positively and negatively correlated with FAMat are shown with red and blue squares, respectively. In both cases, the left symbol corresponds to 10DAP, and the right one to 16DAP. The * symbol was used to highlight the hub genes identified by MCC analysis at 16DAP. A red (positively correlated module) or blue (negatively correlated module) asterisk was used when the individual correlation did not reach significance. Nitrate negative correlation was determined at maturity. The names of metabolites identified in the top four strongest biomarkers at both stages were marked in the scheme with red or blue according to the expected levels in high oil lines. Scheme adapted from (Baud and Lepiniec, 2010). The cytosolic branch of the oxidative pentose phosphate pathway was included according to (Tsogtbaatar et al., 2020). 1,3PG, 1,3‐bisphosphoglycerate; 2A2HB, 2‐aceto 2‐hydroxy butanoate; 2AL, 2‐acetolactate; 2KG, 2‐ketoglutarate; 2PG, 2‐phosphoglycerate; 3PG, 3‐phosphoglycerate; 6PG, 6‐phosphogluconate; 6PGL, 6‐phosphogluconolactone; AcCoA, acetyl‐CoA; ADP‐GLC, ADP‐glucose; Arg, arginine; ArgSuc, argininosuccinate; Asn, asparagine; Chl, chlorophyll; CIT, citrate; Citru, citrulline; DHAP, dihydroxyacetone phosphate; E4P, erythrose 4‐phosphate; F1,6P, fructose 1,6‐bisphosphate; FRU, fructose; FUM, fumarate; G1P, glucose 1‐phosphate; G6P, glucose 6‐phosphate; GAP, glyceraldehyde 3‐phosphate; GLC, glucose; Gln, glutamine; Glu, glutamate; ICT, isocitrate; LHCI, light harvesting complex I; LHCII, light harvesting complex II; MaCoA, malonyl‐CoA; MAL, malate; OAA, oxaloacetate; Orn, ornithine; PEP, phosphoenolpyruvate; PSI, photosystem I; PSII, photosystem II; PYR, pyruvate; R5P, ribose 5‐phosphate; Ru1,5BP, ribulose 1,5‐bisphosphate; Ru5P, ribulose 5‐phosphate; S1,7BP, sedoheptulose 1,7‐bisphosphate; S7P, sedoheptulose 7‐phosphate; SUC, succinate; SUC‐CoA, succinyl‐CoA; UDP‐GLC, UDP‐glucose; UDP‐GlcA, UDP‐glucuronic acid; UDP‐Rha, UDP‐rhamnose; UDP‐Xyl, UDP‐xylose; Xu5P, xylulose 5‐phosphate. Enzymes or protein correlated to FAMat: 1 – phosphofructokinase (Ta1.0_21185), 2 – glyceraldehyde 3‐phosphate dehydrogenase (Ta1.0_14741), 3 – phosphoglycerate mutase (Ta1.0_03325), 4 – plastidic glucose translocator (Ta1.0_04394), 5 – hexokinase (Ta1.0_17071), 6 – fructokinase (Ta1.0_22569), 7 – phosphofructokinase (Ta1.0_00222), 8 – glyceraldehyde 3‐phosphate dehydrogenase (Ta1.0_03731), 9 – phosphoglycerate kinase (Ta1.0_12557), 10 – phosphoglycerate mutase (Ta1.0_03326), 11‐ pyruvate dehydrogenase plastidic E2 subunit (* Ta1.0_04980), 12 – acetyl‐CoA carboxylase‐biotin carboxylase subunit (*Ta1.0_09801), 13 – acetohydroxyacid synthase (Ta1.0_24015), 14 – transaldolase (Ta1.0_03091), 15 – ribulose 5‐phosphate epimerase (Ta1.0_00722), 16 – phosphoribulokinase (Ta1.0_12604), 17 – glyceraldehyde 3‐phosphate dehydrogenase (Ta1.0_04919 and Ta1.0_03025), 18 – sedoheptulose bisphosphatase (Ta1.0_15274), 19 – glutamate 1‐semialdehyde aminomutase (Ta1.0_13342), 20 – aldolase (Ta1.0_19789), 21 – hydroxymethylbilane synthase (Ta1.0_00770), 22 – uroporphyrinogen decarboxylase (*Ta1.0_07323), 23 – coproporphyrinogen III oxidase (*Ta1.0_08336), 24 – magnesium protoporphyrin IX methyltransferase (Ta1.0_18070), 25 – uroporphyrin methylase (Ta1.0_24457), 26 – nitrate transporter (Ta1.0_24305 and Ta1.0_25278), 27 – nitrate transporter (Ta1.0_04706), 28 – nitrate reductase (Ta1.0_05727), 29 – nitrite reductase (Ta1.0_26293), 30‐ glutamine synthetase (Ta1.0_26919), 31 – amino acid permease (Ta1.0_02089), 32‐ glutamine‐dependent asparagine synthase (Ta1.0_23820), 33 – pyruvate dehydrogenase E1 component alpha subunit (Ta1.0_20470) and E2 subunit (Ta1.0_11266), 34 – citrate synthase (Ta1.0_03365), 35 – 2‐ketoglutarate dehydrogenase E2 subunit (Ta1.0_03129), 36 – succinate dehydrogenase 2–3 (*Ta1.0_02582), 37 – ATP citrate lyase subunit B2 (*Ta1.0_16776), 38 – cytosolic acetyl‐CoA carboxylase (Ta1.0_04259) and 39 – phosphoenolpyruvate carboxylase (Ta1.0_17122).

The photosynthetic capacity of pennycress embryo strongly contributes to its FA accumulation

To evaluate the weight of photosynthesis in FA accumulation in pennycress, we searched among the potential target list, genes related to light harvest and carbon fixation. As shown by the enrichment analysis of correlated genes and members of the Honeydew1 module, a high number of pennycress genes with potential photosynthetic functions positively correlated with FAMat (Photosynthesis related Genes, PSG in Table S5). Among these, six pennycress homologues to Arabidopsis Calvin cycle enzymes were identified: phosphoribulokinase (AtPRK), phosphoglycerate kinase (AtPGK1), both type A subunits of the plastidic glyceraldehyde 3‐phosphate dehydrogenase (AtGAPA1 and AtGAPA2), sedoheptulose‐bisphosphatase (AtSBPASE), and ribulose 5‐phosphate 3‐epimerase (AtRPE, Figure 6). Additionally, putative structural proteins of light‐harvesting antenna complexes (LHCI and LHCII), PSII, PSI, and ATPase were pinpointed (Table S5; Figure 6). Particularly, the MCC analysis highlighted two putative subunits G and one P of the PSI (Ta1.0_06300, Ta1.0_11108 and Ta1.0_10549) as being in the top rank of connectivity in the module Honeydew1 (Figure 3; Tables S5 and S6). In addition to the higher expression of the proteinaceous components of the PSI and PSII in the HO embryos, the chlorophyll biosynthesis pathway was also tightly controlled in the HO lines (Figure 6). Specifically, the branching step where the uroporphyrinogen III could be utilized by a uroporphyrinogen decarboxylase to continue towards the biosynthesis of chlorophyll, or by the uroporphyrin methylase to synthesize siroheme, the prosthetic group of sulphite and nitrite reductases (Figure 6). On one side, the expression levels of the putative genes encoding for the uroporphyrinogen decarboxylase (HEME2) and coproporphyrinogen III oxidase (LIN2) were not only positively correlated but also part of the identified hub genes (Figure 6; Tables S5 and S7). On the other side, the expression of a pennycress gene homologue to the uroporphyrin methylase 1 (AtUPM1) was found to be one of the best negative oil biomarkers and strongest negative correlation at 16DAP (Figure 6; Table S5). These results suggest that each component of the photosynthetic machinery strongly contributes to the requirements for FA synthesis in pennycress embryos. Then, to evaluate the impact of photosynthesis in the accumulation of FA, PC22 embryos at early stage were grown in culture for 6 days under normal (20 μmol/m⁻²/s⁻¹) or high (70 μmol/m⁻²/s⁻¹) light intensities. A 33% increase in the total FA content was observed at higher light intensity (Figure S9), confirming that improving the photosynthetic machinery of pennycress embryos is a key target to improve oil content.

High oil lines expressed lower levels of nitrogen assimilation and transport‐related genes

The high divergence of amino acid levels and the identification of amino acid metabolic reactions in the gene enrichment analysis, led us to evaluate the relation between the expression of nitrogen‐related genes and FAMat (Nitrogen‐related Genes, NG in Table S5). Negative correlations with different steps of nitrogen transport, assimilation, and metabolism in the embryo were observed, especially at 16DAP, mainly grouped in the Floral‐white and Salmon4 modules (Figure 6; Table S6). Among these, putative nitrate transporters (homologues to AtNRT1.7, AtNRT3.1 and AtNRT2.7, Figure 6), enzymes involved in nitrogen assimilation (homologues to the nitrate reductase 1, cofactor of nitrate reductase and xanthine dehydrogenase 2, nitrite reductase 1, glutamine synthetase, Figure 6), and transcription factors (TFs) known to induce nitrate‐dependent gene expression (homologues to NIN‐like protein 8, Ta1.0_18491, and NIN‐like protein 2, Ta1.0_04609, Table S5), were identified. In addition, another regulatory factor related to nitrogen metabolism was found to be three pennycress homologues of an Arabidopsis nitric oxide‐responsive protein (Ta1.0_22305, Ta1.0_22306 and Ta1.0_22307, Table S5). To evaluate a potential association between nitrate and oil levels, nitrate content was measured in mature seeds of 10 accessions. Interestingly, the results showed a negative correlation between the two variables (Figure S10), suggesting a competition between carbon and nitrogen metabolism.

Besides inorganic nitrogen, the transport and metabolism of amino acids seemed to have a negative relationship with FAMat. Indeed, a negative association with a putative amino acid permease (AtAAP4 homologue) was detected, in addition to negative and positive correlations with the expression of genes related to specific amino acid synthesis, degradation or interconversion (Table S5). The metabolism of histidine, branched and aromatic amino acids were highly represented, as previously shown in the pathway enrichment (Figures S6c,d). Thus, the HO lines showed a tight transcriptional control of inorganic and organic nitrogen availability, which was associated with their capacity to accumulate oil.

Discussion

Exploiting pennycress natural divergence in seed oil content to identify promising targets

The utilization of alternative crops to meet the increasing demands for biofuels production is restricted by seed oil productivity. There is no universal engineering approach that can effectively improve seed oil accumulation across species. Indeed, a combination of factors must be evaluated to pinpoint the best strategy, including the identification of potential targets and the determination of the location, timing, and amplitude of the changes. To address these questions in pennycress, 22 natural accessions were evaluated for their embryo biomass composition, metabolite content, and mRNA steady‐state levels at two critical points of seed development, and the potential association of each one of these features with FAMat was assessed (Figure 1).

The selected collection of pennycress accessions presented a 29–41% divergence in FAMat in the embryos (Figure 2), making it suitable to identify targets for oil improvement. Previous reports of mature seed oil content were in the ranges of 13.5%–38.7% for 80 wild accessions (Sedbrook et al., 2014), 24.7%–38.7% for 34 lines from USDA NPGS, and 27.5%–35.6% for 41 accessions grown in three locations (Altendorf et al., 2019). The higher oil levels measured in our set of 22 pennycress accessions is likely due to the present study focusing on the embryos, site of synthesis and storage of TAGs, rather than the entire seed. It is important to note that no significant correlation was observed between FAMat levels and embryo DW at maturity (Figure S2), which corroborates previous studies using whole seeds (Altendorf et al., 2019). Although we did not find a negative correlation between the level of oil and protein in mature embryos, a recent study with a larger population (174 pennycress lines) revealed a significant negative correlation in entire seeds harvested across different growing environments (Tandukar et al., 2022). Besides, FAMat was found to be positively correlated with erucic acid, and negatively with palmitic and stearic acids (Figure S2), which indicates that efforts towards increasing total oil content will potentially improve erucic acid levels and decrease those of the saturated precursors as well.

To obtain a systematic overview of embryo metabolism and identify candidate metabolites and genes effectively, two complementary approaches were undertaken. The first one considered a dose‐dependent response between the variable analysed and the oil content (Pearson correlation analysis and WGCNA), while the second determined a cut‐off separating of HO and LO lines (biomarker identification). The combination of the approaches led us to identify mechanisms that contributed to the increased oil content in the HO lines and key targets to achieve it (Figure 7). Although the common characteristics to HO lines were the focus of this study, mechanisms specific to individual HO accessions cannot be discarded.

Figure 7.

Integrative approach for oil improvement in pennycress embryos. Organelles are shown in black, biomass components in green, promising candidate genes to up‐regulate in red, and genes to down‐regulate in blue. PAE, pectinacetylesterase; PEPC, phosphoenolpyruvate carboxylase; pHK, plastidic hexokinase; pGlcT, plastidic glucose transporter; pGAP, plastidic glyceraldehyde 3‐phosphate dehydrogenase; PRK; phosphoribulokinase; LHC, light harvesting complex genes; PSAG, photosystem I subunit G; PTAC8, photosystem I subunit P; UPM1, urophorphyrin methylase; HEME2, uroporphyrinogen decarboxylase; LIN2, coproporphyrinogen III oxidase; ACC, plastidic acetyl‐CoA carboxylase; SAD, stearoyl‐ACP desaturase; ALT, acyl‐lipid thioesterase; pLACS, plastidic long‐chain acyl‐CoA synthetase; ACBP, acyl‐CoA binding protein; KCS, ketoacyl‐CoA synthase; DGAT, acyl‐CoA: diacylglycerol acyltransferase; LDAP3, LD‐ASSOCIATED PROTEIN 3; TUB, tubulin; ACT, actin; TAGL, triacylglycerol lipase; peLACS, peroxisomal long‐chain acyl‐CoA synthetase; AACT, acetoacetyl‐CoA thiolase; NTR1.7, nitrate transporter; WR3, nitrate transporter; AAP4, amino acid transporter; NIA, nitrate reductase; NIR, nitrite reductase; NLP8 and NLP2, NIN‐like protein 8 and 2.

Enhancing carbon partitioning towards the chloroplast to increase its availability for FA synthesis

Many of the transcriptional changes that correlated with changes in FAMat were of transcripts encoding for chloroplast‐localized proteins (Figure 3), indicating adjustments of plastidic processes in the HO lines. In addition, higher levels of oil were associated with higher content of glucose, enhanced expression of a putative plastidic glucose transporter (pGlcT) and a plastidic hexokinase, as well as reduced levels of cell wall biosynthesis intermediates UDP‐xylose and UDP‐glucose (Table S4, Figure 6; Figure S5). Transcripts encoding potential cell wall biosynthetic enzymes were also identified in the negatively correlated Salmon4 module (Figure 4). The increased glucose availability and pGlcT expression in HO lines, guarantee higher carbon partitioning towards the chloroplast, reducing its availability for competing pathways. This transporter has been associated with the export of glucose produced by starch degradation during night metabolism (Facchinelli and Weber, 2011). However, pGlcT catalyses glucose‐facilitated diffusion between the cytosol and the stroma, and passive glucose influx into the chloroplast has been observed in vitro (Li et al., 2017; Servaites and Geiger, 2002; Weber et al., 2000). Therefore, despite the presence of multiple translocators responsible for transferring carbon from the cytosol to the plastid, the results suggest a non‐conventional carbon entry point in pennycress HO lines through the direct transport of glucose and its sequestration by a plastidic hexokinase (Figure 6).

Another alternative entry point in pennycress HO lines is the translocation of malate into the plastid. This study identified a positive correlation between FAMat and the expression of a putative phosphoenolpyruvate carboxylase (PEPC) at 16DAP, as well as malate/UDP‐glucose as the best oil biomarker at the same stage (Figure 6; Figure S5). Malate can be produced in the mitochondria through the TCA cycle or in the cytosol by the consecutive reactions of PEPC and malate dehydrogenase (Figure 6). Plastidic malate provides carbon and reductants necessary for FA synthesis through the action of the plastidic malic enzyme (Figure 6). ¹³C‐labelling experiments in developing embryos recently determined that this malic enzyme produces 17% of the pyruvate in the plastids in pennycress (Tsogtbaatar et al., 2020), and limits FA synthesis in maize (Cocuron et al., 2019). Therefore, increasing malate levels by an up‐regulation of PEPC, especially at 16DAP, was a successful strategy in the HO lines to contribute to plastidic FA synthesis.

Asparagine and glutamine were identified to be positively associated with FAMat as part of the Gln/quinic acid and Asn/UDP‐xylose biomarkers (Figure S5). Maternal asparagine and glutamine can be directed to the mitochondria to be metabolized and converted into malate that can be translocated to the plastid for further FA synthesis (Figure 6). Indeed, labelling developing pennycress embryos with ¹³C‐glutamine demonstrated that the carbons from this amino acid produce ¹³C‐labelled malate that enters the plastid to produce ¹³C‐FAs (Tsogtbaatar et al., 2020). Overall, to improve oil accumulation in pennycress, carbon commitment towards the chloroplast should be prioritized at the expense of competitive pathways. Thus, exploiting pennycress natural divergence in seed oil content instructs metabolic engineering efforts aimed at redirecting the incoming carbon into the FA biosynthesis (Figure 7).

Enhancing the expression of specific oil‐related genes to improve FA content

Another complementary strategy followed by the HO lines was to optimize lipid metabolism itself. Indeed, this study distinguished correlations of FAMat with genes that have been experimentally proven or suggested to regulate oil production in Arabidopsis, such as ACCase that catalyses the first committed and rate‐limiting step of FA biosynthesis (Sasaki and Nagano, 2004); ACBP, an acyl‐CoA binding protein recently associated with the control of unsaturated TAG content in Arabidopsis seeds (Guo et al., 2019); pPLAIIIδ, a patatin‐related phospholipase A that mediates phospholipid turnover (Li et al., 2013); ALT1, an acyl‐lipid thioesterase that, when overexpressed, lead to the accumulation of 12–14 carbon‐length FAs and 35% reduction in total oil content (Kalinger et al., 2021); and DGAT3, a diacylglycerol acyltransferase involved in TAG biosynthesis (Aymé et al., 2018, Figure 5; Table S5). Despite being annotated as lipolytic enzymes, genes identified as a putative α/β hydrolase and a GDSL lipase, whose homologues have not been characterized in Arabidopsis yet showed positive correlation with FAMat. The genes mentioned above, highly expressed in HO lines, are promising targets to increase carbon commitment into lipid biosynthesis. Unexpectedly, at 16DAP a pennycress homologue of the Oil Body Associated Protein1 (AtOBAP1), necessary to maintain the structure of the oil bodies in Arabidopsis (López‐Ribera et al., 2014), presented a negative correlation and was identified as a hub gene of the Salmon4 module (Figure 5; Table S7), suggesting a distinctive role for this protein in pennycress.

In addition, at the early stage, pennycress accessions showed positive correlations between FAMat and the proportion of oleic acid, and with the expression of a plastidic stearoyl‐ACP desaturase, enzyme responsible for its production (Figure 5 and Table S2). At 16DAP, a ketoacyl‐CoA synthase (FAE1/KCS2, involved in FA elongation) was positively correlated with FAMat while the percentage of linolenic acid was negatively correlated (Figure 5; Table S2). These results indicate that in this set of accessions, higher mature seed oil is associated with early increases in oleic acid synthesis and accumulation, and a posterior stimulation of its elongation (increasing erucic acid levels) over its desaturation (reducing linolenic acid synthesis). These findings support the notion that increasing the levels of erucic acid results in higher total oil content. Indeed, a recent study that abolished erucic content in pennycress seeds by mutating FAE1, led to a significant lower oil content (Sedbrook and Durrett, 2020). Therefore, besides a direct improvement of FA synthesis and degradation, adjustments in FA composition also turned out to be a successful strategy for the HO lines to increase total oil.

Enhancing photosynthetic capacity to support FA synthesis

The high number of pennycress genes with potential photosynthetic functions that positively correlated with FAMat (Table S5) suggests that photosynthesis within pennycress embryos contributes to FA synthesis. The present study underlined that higher oil accumulation in pennycress was associated with overexpression of genes encoding for many putative structural proteins of LHCI, LHCII, PSI, PSII, enzymes involved in chlorophyll biosynthesis and Calvin cycle enzymes, mostly grouped in the Honeydew1 module, and with increased levels of ribulose 1,5‐bisphosphate (Figure 6; Tables S4 and S6). In addition, incubating PC22 embryos at higher light intensity, increased total FAs by 33% (Figure S9). Light could stimulate oil accumulation at several levels, providing ATP and NADPH, regulating key enzymes activity through the ferredoxin‐thioredoxin system, generating O₂ to avoid anoxic conditions that could impair oxidative ATP production, and potentially improving carbon conversion efficiency through RUBISCO (Baud and Lepiniec, 2010; Puthur et al., 2013; Simkin et al., 2020; Ver Sagun et al., 2023). Then, by the increase of the light intensity in the cultures or by a better‐equipped thylakoid membrane in the HO lines, pennycress embryos are able to improve the process of oil accumulation. Furthermore, a more active linear photosynthesis would imply higher production of O₂ and, therefore, a decreased cellular response to hypoxia, a process found to be enriched in both negative correlated modules (Floral‐white and Salmon4, Figure 4). Previous reports in Arabidopsis (Liu et al., 2017), soybean (Allen et al., 2009; Ruuska et al., 2004), rapeseed (Goffman et al., 2005; Ruuska et al., 2004; Schwender et al., 2004), false flax (Carey et al., 2020), and pennycress (Tsogtbaatar et al., 2020) provided evidence that embryonic photosynthesis participates to some degree in seed reserve accumulation. More specifically, in Brassicaceae false flax (Carey et al., 2020) and rapeseed (Goffman et al., 2005), an increased light intensity—above the physiological conditions—to the developing embryos did not improve oil content (Ver Sagun et al., 2023). Our study shows that the photosynthetic capacity of pennycress embryo strongly contributes to oil accumulation, which was not only validated in culture conditions at higher light intensity (Figure S9) but also stands out from previous reports in other Brassicaceae. Indeed, ¹³C‐labeling experiments in developing pennycress embryos showed that re‐fixation of CO₂ by Rubisco was responsible for 25% of the total phosphoglycerate produced in the plastid, increasing carbon available for FA synthesis (Tsogtbaatar et al., 2020). Optimizing this process in pennycress represents a promising avenue for oil increase. Finally, it is interesting to note that a putative nicotinate phosphoribosyltransferase, involved in the biosynthesis of NAD, presented a positive correlation with FAMat at both developmental stages (Table S5), whereas the content of nicotinic acid, its substrate, showed an inverse correlation (Table S4). Thus, enhancing the synthesis of the cofactors (NAD and NADP) could also benefit oil synthesis.

Reducing embryo nitrogen availability at 16DAP to increase seed oil at maturity

Negative correlations between FAMat and different steps of nitrogen transport, accumulation, assimilation, and metabolism at 16DAP, highlight the importance of a tight control of nitrogen availability at this stage to achieve higher levels of oil at maturity (Figure 6). All accessions received the same standard nitrogen fertilization; thus, the differences must rely on specific capacities of the lines: absorption, transportation, assimilation and/or utilization. Analogously, pennycress grown using different nitrogen fertilization treatments in chambers and in the field, showed no significant differences in seed oil content (Rukavina et al., 2011). These data indicate that the nitrogen's external supply does not impact final seed oil content.

Nitrogen supplied to seeds under development depends on both inorganic and organic uptake from the vascular system (Tegeder and Masclaux‐Daubresse, 2018). One strategy followed by the HO lines to restrict nitrogen availability at 16DAP was to specifically reduce the expression of nitrate transporters and assimilating enzymes in the embryos (Figure 7). According to the correlation analysis, the expression of three putative nitrate transporters homologues to AtNRT1.7, AtNRT3.1 and AtNRT2.7 was reduced in HO lines (Table S5). AtNRT1.7 and AtNRT3.1 encode for low‐ and high‐affinity nitrate transporters on the cell membrane, respectively (Fan et al., 2009; Okamoto et al., 2006), and AtNRT2.7 is a tonoplast transporter, mainly active during late embryogenesis, that controls long term nitrate storage in seeds (Chopin et al., 2007; Tegeder and Masclaux‐Daubresse, 2018). The reduction of up to 64% in the expression of these transporters in HO lines explains the lower levels of nitrate at maturity (Figure S10). Besides the transporters, HO lines also presented reduced pennycress homologues of the nitrate reductase 1 (AtNR1), nitrite reductase 1 (AtNiR1), and uroporphyrin methylase 1 (AtUPM1), contributing to the control of the nitrogen availability (Figure 6). Lastly, the down‐regulated TF homologues to NIN‐like protein 8 (AtNLP8) and AtNLP2 were previously shown to induce nitrate‐dependent gene expression (Konishi and Yanagisawa, 2013), contributing to lowering the expression of nitrate‐dependent genes.

A second strategy limiting nitrogen availability to embryos at 16DAP in the HO lines relies on the control of amino acid transport and metabolism. A negative correlation was observed for a putative amino acid permease in the middle stage (AtAAP4 homologue). AtAAP4 is a plasma membrane protein able to transport with high‐affinity glutamine, asparagine, alanine, valine and tryptophan (Fischer et al., 2002). The fact that a knockout mutant in AtAAP2, an amino acid permease that shares AtAAP4 amino acid specificity, showed an increase in seed oil content (Zhang et al., 2010) corroborates our results. In addition, amino acid levels were found to be highly divergent (Table S3). For instance, arginine and both intermediates, citrulline and ornithine, were inversely correlated with FAMat at 10 and/or 16DAP (Table S4). Arginine and nitrate, both negatively correlated with oil, are precursors of nitric oxide (NO) synthesis (Chamizo‐Ampudia et al., 2016; Crawford, 2006; Tejada‐Jimenez et al., 2019). NO, a signalling molecule, has been recently reported to reduce seed oil accumulation (Liu et al., 2020). Indeed, the Arabidopsis nia1nia2 double mutant, carrying mutations in both nitrate reductases, presented significantly lower NO amounts, but higher oil content (Liu et al., 2020). Besides, a strong negative correlation with pennycress homologues of an Arabidopsis nitric oxide‐responsive protein was observed at 10 and/or 16DAP. This protein is a member of the Light Response Bric‐a‐Brac/Tramtrack/Broad Complex (BTB) family of Arabidopsis, AtLRB3 (Zarban et al., 2019). Therefore, NO levels and responses are promising targets to optimize pennycress seed oil.

Although the present study revealed a negative correlation at maturity between oil content and nitrate levels (Figure S10), no significant correlation was found between oil and proteins at that final stage (Figure S2). As we previously discussed, nitrate's nitrogen in the embryo could have multiple fates such as being stored in the vacuole, converted into nitric oxide—a gaseous reactive oxygen species with a high diffusion rate across biological membranes (Beligni and Lamattina, 2001)—or serving as a nitrogen source for many molecules, including proteins. In addition, protein synthesis is also affected by the transport rate of amino acids into the embryo. The multiple and different variables affecting nitrate and protein levels and their control along development, even after 16DAP, could explain why they showed a divergent correlation with oil content at maturity.

Finally, the reduction of nitrogen availability to possibly benefit oil accumulation was subtle, as many parts of the active metabolism require nitrogen, such as the photosynthetic apparatus for example. Although in this set of accessions the total protein content did not show a negative correlation with FA in mature embryos (Figure S2), 16 DAP proteins did. All together these findings reinforce the fact that a tight control of nitrogen metabolism at 16DAP contributes to achieving higher levels of seed oil in pennycress.

Overall, this work presented the mechanisms by which natural pennycress HO lines achieved that valuable trait. An enhancement of chloroplast carbon uptake, an optimization of FA synthesis, degradation and proportion, an active photosynthesis, and a tight control of nitrogen availability, were identified as the main processes contributing to higher oil production in these accessions. For each process, key candidate targets were proposed to orient engineering efforts aiming to enhance and optimize pennycress seed oil.

Experimental procedures

Plant growth and sample collection

Twenty‐one pennycress accessions used in this study were obtained from the U.S. National Plant Germplasm System (Table S1). The line TAMN106 (PC22) was kindly provided by Dr. Marks from the University of Minnesota. Each accession was grown under greenhouse conditions. Immature embryos were collected simultaneously for biomass and metabolomic at approximately 9–12 days and 6 days later after (15–18 DAP), depending on the line. Four biological replicates were evaluated for each assay, for each pennycress accession, and at both developmental stages. The simultaneous collection of embryos for RNAseq analysis is described in García Navarrete et al., 2022. The accession PI 650284 (PC17) was an exception: the number of seeds produced by this line was insufficient to collect embryos at the earlier stage. Finally, mature embryos were also collected and analysed for their final biomass composition. For a detailed description of the plant growth and sample collection see Appendix S1.

Biomass quantification and targeted metabolomics

The extraction of the biomass components and metabolites from the pooled embryos and their quantification was performed as previously described (Cocuron et al., 2014) with small modifications. Briefly, total fatty acid from grounded dried embryos, supplemented with 50 μg of glyceryl triheptadecanoate (internal standard), were extracted by three consecutive rounds using 1 mL hexane:isopropanol (2:1, v:v). The pooled FA supernatant was dried and then methylated using methanol:sulphuric acid 2.5% for 120 min at 80 °C. FAMEs were analysed by GC–MS using an Agilent 5975B series Gas Chromatograph/Mass Spectrometer (Santa Clara, CA). The GC and MS conditions used were as previously described (Tsogtbaatar et al., 2015). GC–MS data were acquired and processed using Agilent MSD Productivity ChemStation software (Agilent). FAME derivatives were identified using FAME standards purchased from (St. Louis, MO). Proteins were extracted from the residual pellet by three rounds of 1 mL of 20 mM Tris–HCl, pH 7.5, 1% (w/v) SDS and 150 mM NaCl. After combining the supernatants, total protein content was determined using the DC Protein Assay kit (Bio‐Rad: Hercules, CA, USA). Finally, the remaining pellets containing the non‐soluble starch were digested with 10 μL of amyloglucosidase (Megazyme International Ireland Ltd Total Starch Assay Kit, Wicklow, Ireland) for 1,5 h at 55 °C. Glucose content was then quantified by LC–MS/MS (Cocuron et al., 2014).

RNAseq data analysis

Paired‐end libraries obtained from García Navarrete et al. (2022) were evaluated with FastQC v.0.11.9 (Andrews, 2010). The trimmed process was carried with Trimmomatic v.0.39 (Bolger et al., 2014), and the clean reads were used as input in HISAT v.2.0.4 (Kim et al., 2015) to map and align the reference pennycress genome (Dorn et al., 2015). Reads were counted using feature Counts v.1.5 (Liao et al., 2014); only uniquely aligned reads were considered. Transcripts per million (TPM) was used as the unit for calculating the transcript expression. Arabidopsis homologue genes with pennycress were determined through the module GenomesAligner from NGSEP v.4.0.0 (with default parameter, Tello et al., 2019). To identify the 200 more variables transcripts at each developmental stage, we entered the raw count matrix previously made with feature Counts v.1.5 into the DESeq2 v.1.24 package, where the vst function was applied. Gene Ontology (GO) term enrichment analysis was carried on through the package of Bioconductor topGO v.2.36.097 (Alexa et al., 2006) using a Fisher exact test. Protein subcellular localization likelihood analysis was performed with DeepLoc v.1 (Almagro Armenteros et al., 2017) by a prediction algorithm using deep neural networks (Appendix S1).

Batch correction, principal component analysis, Pearson correlation, biomarkers analysis and pathway enrichment analysis

Tools of the online available software MetaboAnalyst 5.0 (www.metaboanalyst.ca) were used for the different studies in this work (Pang et al., 2021): (1) ‘Statistical analysis’, for Pearson correlation and Principal component analysis in log transformed and autoscaled data; (2) ‘ComBat’, for correction of variance between batches for the metabolomics analysis; (3) ‘Joint pathway analysis’ for pathway enrichment analysis and (4) ‘Biomarkers analysis’ for oil biomarker identification (Appendix S1).

Weighted gene co‐expression network analysis

Modules of highly correlated transcripts expressed along the 22 lines at 16DAP were constructed using the R package WGCNA (Langfelder and Horvath, 2008; Appendix S1). Enrichment analysis for each module was performed by Panther Gene List Analysis (Mi et al., 2019). For further analysis, the WGCNA network was exported using the exportNetworktoCytoscape function. Network visualization was performed using Cytoscape (Shannon et al., 2003) and the plugin cytoHubba (Chin et al., 2014) was utilized for hub gene analysis. Genes were ranked using scores from Maximal Clique Centrality (MCC) topological analysis.

Pennycress cultures

Embryos of PC22 (TAMN106) were cultured as described in (Tsogtbaatar et al., 2020), using two light intensity 20 (normal) or 70 (high) μmol/m²/s (Appendix S1).

Nitrate content measurement

Nitrate quantification was done by a colorimetric assay, as previously described (Miranda et al., 2001) with minor adaptations described in the Appendix S1.

Accession numbers

The different accessions of pennycress that were used in this work, were ordered from the U.S. National Plant Germplasm System (https://npgsweb.ars-grin.gov/gringlobal): Ames 32 908, Ames 32 872, Ames 31 499, Ames 31 497, Ames 29 512, Ames 31 026, Ames 31 501, Ames 31 500, Ames 31 488, Ames 30 933, Ames 30 985, Ames 24 499, Ames 29 531, Ames 22 461, PI 650287, PI 633415, PI 650284, Ames 30 982, Ames 31 012, Ames 31 498, PI 650285. The line TAMN106 (PC22) was kindly provided by Dr. Marks from the University of Minnesota.

Conflict of interest

The authors declare no conflict of interest.

Author contributions

A.P.A. and E.G. conceived and designed the project. C.L.A., T.S. and F.Y. collected the embryos, planned, and performed the experiments. L.T.G.N. and E.M. processed the RNAseq data. C.L.A., L.T.G.N., E.M., J.H., E.G. and A.P.A. contributed to data analysis and interpretation. C.L.A. and A.P.A. drafted the manuscript, C.L.A., L.T.G.N., E.M., E.G. and A.P.A. reviewed and edited the document. All authors critically revised and agreed to the published version of the manuscript.

Supporting information

Figure S1 Variation in seed composition at maturity in the pennycress accession collection selected for this study.

Figure S2 Correlation analysis between biomass components at maturity.

Figure S3 Biomass accumulation in early, middle and maturity stages for each pennycress accession.

Figure S4 FA composition during development.

Figure S5 Distribution of the levels of the top four metabolic biomarkers in the high and low oil lines for each developmental stage.

Figure S6 KEGG pathway enrichment at 10 and 16DAP.

Figure S7 Pathway enrichment of genes whose level of expression correlates with the content of FAMat.

Figure S8 Module‐trait relationships at 16DAP.

Figure S9 PC22 embryos cultured at normal and high light intensities.

Figure S10 Negative correlation between nitrate and oil content at maturity.

Data S1 Raw data of metabolite levels in pennycress embryos at 10 and 16DAP after ComBat batch correction.

Appendix S1 Supplemental methods and references.

Table S1 Pennycress accessions and original locations.

Table S2 Correlation analysis between FA composition and FAMat during development.

Table S3 Standard deviation and median absolute deviation for each metabolite.

Table S4 Metabolites correlated with FAMat.

Table S5 Transcripts correlated with FAMat and biomarkers of oil content.

Table S6 FAMat‐related modules.

Table S7 Top 10 ranked genes according to the Maximal Clique Centrality topological analysis (Hubs) for each FAMat‐related module.

Table S8 Gene Ontology of the top 10 ranked genes according to the Maximal Clique Centrality topological analysis (Hubs) for each FAMat‐related module, Cellular Component category.