Prior to analysis, the duckweed was dried to a constant weight at 60°C and milled. Structural carbohydrates, including glucan, xylan, galactan, arabinan, mannan, lignin, and ash, were determined according to the methods recommended by the National Renewable Energy Laboratory, USA (Sluiter et al., 2012). The starch content was determined via hydrolysis of duckweed with HCl, as described previously (Liu et al., 2015b). The cellulose content was calculated by subtracting starch from glucan. Xylan, galactan, arabinan, and mannan together are considered hemicellulose. Lipid was extracted using diethyl ether with a Soxtec system with reference to AOAC 920.39 B (http://down.foodmate.net/standard/sort/10/25070.html). Total Kjeldahl nitrogen (TKN) was measured using a FOSS KJ2200 System (FOSS Corp., Denmark). The protein content was calculated as the TKN content multiplied by the conventional factor (6.25). Pectin was extracted according to the methods described by (Luo et al., 2016) and (Yang et al., 2011) and determined according to the methods described by (Blumenkrantz and Asboe-Hansen, 1973), with GalUA (Sigma−Aldrich) used as a standard. The contents of carbon and nitrogen were determined using an elemental analyser (Vario EL Cube; Elementar Analysensysteme GmbH, Germany).

Fresh duckweed (0.5 g) was homogenized in 5 ml of precooled enzyme extraction solution (100 mM tricine-NaOH (pH 8.0), 8 mM MgCl₂, 2 mM EDTA, 50 mM 2-mercaptoethanol, 12.5% (v/v) glycerol, and 5% (w/v) insoluble polyvinylpyrrolidone-40) (Nakamura et al., 1989). The homogenate was subsequently centrifuged at 13,400 × g for 10 min at 4°C. The resulting supernatant was subsequently used to measure the enzyme activities. The activities of AGPase (EC 2.7.7.27) and starch synthase (SSS, EC 2.4.1.21; GBSS, EC 2.4.1.242) were analyzed using the methods described by Nakamura et al. (1989). The activities of the two enzymes were tested by measuring the change in NADH at 340 nm using a microplate reader (Thermo Scientific Varioskan Flash, Thermo Fisher Scientific Inc., USA). The activities of α-amylase (EC 3.2.1.1) and β-amylase (EC 3.2.1.2) were estimated following previously described methods (Liu et al., 2015b). Rubisco (EC 4.1.1.39) activity was tested as described by Sharkey et al. (1991) using a spectrophotometric diagnostic kit (Suzhou Comin Biotechnology Co., Ltd., Suzhou, China). PEPC (EC 4.1.1.31), NADP-MDH (EC 1.1.1.82), and malic enzyme (ME, EC 1.1.1.39) activities were assayed according to Gonzalez et al. (1984) and Johnson and Hatch (1970) using a spectrophotometric diagnostic kit (Suzhou Comin Biotechnology Co., Ltd., Suzhou, China). Enzyme activity definitions: AGPase and SS activity: One unit is defined as the production of 1 nmol NADPH per mg protein per minute. Amylase activity: One unit is defined as the hydrolysis of starch to produce 1 mg of maltose per minute. Rubisco activity: One unit is defined as the oxidation of 1 nmol NADH per mg protein per minute at 25°C. PEPC activity: One unit is defined as the consumption of 1 nmol NADH per mg protein per minute. NADP-MDH activity: One unit is defined as the consumption of 1 nmol NADPH per mg protein per minute. NADP-ME activity: One unit is defined as the production of 1 nmol NADPH per mg protein per minute.

Sucrose was extracted according to the method described by (Zhu et al., 2018). The duckweed was dried to a constant weight at 60°C and then powdered using a pulverizer. Then, 50 mg of duckweed powder was suspended in 1 ml of deionized water and sonicated for 30 min. The mixture was incubated at 80°C for 1 h with intermittent shaking every 5 min. The sample was then centrifuged at 13,400 × g for 20 min at 4°C. The supernatant was stored at -20°C. The residue was resuspended in 1 ml of deionized water for a second round of extraction. The supernatants from two rounds of extraction were pooled and filtered through a 0.45-μm-pore size filter.

The sucrose content was determined using a high-performance liquid chromatography (HPLC) system (Thermo 2795, Thermo Corp.) equipped with an evaporative light scattering detector (ELSD) (All-Tech ELSD6000, All-tech., Corp.). The samples were separated on an Aminex HPX-87P column (300 × 7.8 mm) at 79°C using ultrapure water as the mobile phase at 0.6 ml min^-1. Analytical pure sucrose (AR) was used as a standard.

The 3-PGA content was measured via an enzymatic assay described by Flores-Tornero et al. (2017). First, 3-PGA was extracted using precooled methanol/chloroform (1:1, v/v). Then, the endogenous enzymes in the samples were heat-inactivated at 70°C for 10 min. Next, 20 μL of the resulting extract was added to 980 μL of reagent (0.1 M Tris-HCl (pH 7.6), 5 mM MgCl₂, 40 μM NADH, 2 mM ATP, 6 units of PGK and 3 units of GAPDH). The mixture was incubated at 25°C for 20 min. Finally, the samples were measured at 340 nm using a spectrophotometer. In the control, 20 μL of methanol/chloroform (1:1, v/v) instead of the extract was added to the mixture.

The transcript levels of all the key starch biosynthesis genes in the chloroplast were upregulated under the LC treatment ( Figure 3A ). Plastidial ADP−glucose pyrophosphorylase (AGPase) ( Supplementary Figure S6 ), the first rate-limiting enzyme in starch biosynthesis, determines carbon flux into starch to a large extent. AGPase is precisely regulated at the transcriptional and posttranslational levels, including allosteric regulation and redox modulation (Geigenberger, 2011; Streb and Zeeman, 2012). Previous studies revealed that AGPase activity is induced by increased 3-phosphoglycerate (3-PGA) and decreased phosphate through allosteric regulation (Sokolov et al., 1998; Tiessen et al., 2002) and suppressed by phosphate (Nielsen et al., 1998) and nitrate (Scheible et al., 1997) through transcriptional regulation. In our study, elevated CO₂ significantly increased the 3-PGA content from 162.4 μg g^-1 fresh weight (FW) to 292.2 μg g^-1 FW ( Supplementary Figure S10A ). Nutrient limitation caused phosphate and nitrate deficiencies. The combination of activation (mediated by increased 3-PGA and reduced phosphate) and inhibition (mediated by the release of phosphate and nitrate) improved AGPase gene expression and enzyme activity ( Figures 3B ; Supplementary Figure S10B , and 3E ). AGPase expression and activity under LC treatment were 3.0× and 6.5× higher, respectively ( Supplementary Data S2 ; Figures 3A, B ). Most previous attempts to increase starch accumulation have focused only on enhancing AGPase gene expression instead of regulating its enzyme activity, which is possibly why these attempts were not highly successful (Cakir et al., 2019; Kang et al., 2013; Oiestad et al., 2016).

Furthermore, LC treatment increased the expression and activity of other important genes/enzymes, including granule-bound starch synthase (GBSS) and soluble starch synthase (SSS), and the expression of 1,4-alpha-glucan branching enzyme (GBE) ( Figures 3A, C ). Previous studies have shown that the genetic manipulation of any one of these genes usually has limited effects on starch content and does not increase the net starch yield (Ihemere et al., 2006; Stark et al., 1992; Sweetlove et al., 1996; Wang et al., 2007; Zeeman et al., 2010). Starch biosynthesis is a complex system that interconnects a wide variety of cellular processes and metabolic pathways (Geigenberger, 2011). Thus, it is very difficult to develop a comprehensive method to regulate this process. In our case, when the biomass increased by 6×, the starch content increased by 11×, and the starch yield increased by 76× within 10 days compared with that in the beginning stage ( Figures 1 , 2 ). This is presumably because LC treatment significantly improved the activities of AGPase by transcriptional regulation and allosteric regulation, increased the gene expression level and activity of other key enzymes, and ultimately regulated the complex system.

The carbon skeletons of lignocellulose and protein are derived from photoassimilates, and their biosynthesis is strongly affected by carbon partitioning. The lignocellulose contents in duckweed are relatively low and are further reduced under LC treatment ( Figure 2A ). Initially, the cellulose, hemicellulose, and pectin contents were 8.0 ± 0.1%, 7.8 ± 0.1%, and 5.6 ± 0.3%, respectively (d.b.). After ten days of treatment, these values decreased to 1.1 ± 0.2%, 4.1 ± 0.1%, and 1.0 ± 0.0% (d.b.) ( Figure 2A ), corresponding to reductions of 86.3%, 47.4%, and 82.1%, respectively. The expression level of sucrose synthase (SUSY), an enzyme involved in the formation of UDP-glucose from sucrose to cellulose and hemicellulose, was significantly downregulated ( Supplementary Figure S12 ; Supplementary Data S7 ). With respect to cellulose degradation, the expression of genes in glycoside hydrolase family 9 (cellulase) was significantly upregulated more than 5x ( Supplementary Data S8 ). These changes might contribute to the reduction in cellulose and hemicellulose contents.

Lignin, a complex phenol polymer that is difficult to degrade, is the main obstacle for biomass utilization. The lignin content in duckweed decreased by 81.0% ( Figure 2A ), from 5.8 ± 0.1 to 1.1 ± 0.1%, after 10 days of LC treatment. Expression of the gene encoding laccase, the key enzyme for monolignol polymerization and crosslinking, was extremely low (FPKM values < 20) ( Supplementary Figure S12 ; Supplementary Data S9 ), providing a possible explanation for the low lignin content. Additionally, the lignocellulose composition was reduced to a very low level (6.3%), indicating a higher quality of the whole biomass ( Figure 2A ).

Notably, the protein content of duckweed rapidly decreased from 30% to 4% (d.b.) under this treatment ( Figure 2A ). The total amount of protein remained essentially stable over 10 days due to the lack of a nitrogen supply ( Figure 2C ), whereas biomass, especially the starch content, still accumulated rapidly ( Figures 1B, C ). Nitrogen glutamine synthetase (GS), the key gene involved in nitrogen assimilation and recycling, plays an important role in increasing the nitrogen use efficiency (NUE, kg grain yield per kg N application) of crops and increasing cereal yield (Bernard and Habash, 2009; Xu et al., 2012). The expression level of GS was upregulated by 6.7×, and its enzyme activity was 7.4× greater than that of the control ( Supplementary Data S12 and Supplementary Figure S11 ). Previous studies have shown that nitrogen limitation significantly reduces the mRNA expression of GS and its enzyme activity and increases the starch content without increasing plant biomass or starch yield (Balotf et al., 2016; Hermans et al., 2006). However, in this study, both the plant biomass and starch yield increased, possibly by increasing the expression of GS, which strongly promoted the redistribution of ammonium and therefore increased protein reuse ( Supplementary Figure S13 ). Thus, future research could focus on duckweed GS, especially its role in improving NUE.

The reduction in protein and lignocellulose contents in duckweed was consistent with the downregulated expression of relevant genes. This increase in starch content indicated that a large amount of photoassimilate flowed to starch synthesis under the LC treatment. Therefore, duckweed is an ideal model for studies of the regulation of nitrogen and carbon metabolism.

We analyzed the genes of the carbon assimilation pathway in duckweed because of its strong starch accumulation ability. According to the results of whole-genome sequencing of Landoltia punctata 0202 (GenBank accession number: PRJNA546087) and the corresponding gene family analysis ( Supplementary Data S1 ), the total number of genes involved in carbon assimilation, including starch metabolism, the Calvin cycle, and the Hatch−Slack cycle, was only 50. These values are significantly lower than those in Arabidopsis (64), rice (82), and maize (86) ( Table 1 ). Under the LC treatment, in the carbon assimilation pathways, only the transcript levels of the key starch biosynthesis genes and the corresponding enzyme activities were upregulated ( Figures 3A-C and Supplementary Figure S3 ). These starch biosynthesis genes work in a synergistically efficient way, hence enhancing starch formation. There was almost no change in gene expression in the Calvin cycle and no significant change in the activity of its key enzyme Rubisco ( Figures 3A and Supplementary Figure S8C ). In the Hatch-Slack cycle, some genes, such as NADP-MDH, NADP-ME, and PEPC, were also upregulated ( Figure 3A ), similar to the corresponding enzyme activities ( Supplementary Figures S8D,F ). LC treatment mainly increased starch biosynthesis and had a certain effect on the CO₂ concentration but had no obvious effect on the Calvin cycle.

Therefore, what is observed in duckweed contradicts the common knowledge that the more genes an organism possesses, the greater its function. Our results revealed that the changes in starch accumulation may have resulted from the regulation of the starch biosynthetic pathway rather than the number of gene copies.

DNA methylation, a conserved epigenetic modification, plays an important role in assisting in gene regulation and genome stability (Zhang et al., 2018). We studied the epigenome of DNA methylation in the same samples via transcriptome analysis. LC treatment decreased the DNA methylation level in the whole genome of duckweed from 12.9% to 11.2% (mC) at 24 h ( Supplementary Table S7 ). More importantly, the DNA methylation levels of the AGPase, SSS, and GBE promoters were significantly reduced by 46.6, 32.2, and 63.6% (mCG), respectively, while their expression was significantly upregulated. Thus, in the starch biosynthetic pathway, DNA methylation in promoter regions is negatively correlated with gene expression ( Figures 3A, D ). In the Calvin cycle, although the DNA methylation level of the key genes’ promoters decreased, their expression levels did not change significantly. In the Hatch-Slack cycle, the DNA methylation level of the promoters of the same key genes was significantly increased, whereas their expression levels did not decrease. Notably, the expression level, DNA methylation level, and enzyme activity of NADP-MDH, one of the key genes in the Hatch-Slack cycle, increased significantly ( Figures 3A, D ; Supplementary Figure S8E ). Previous studies have indicated that CO₂ elevation can significantly increase the activity of NADP-MDH (Seth and Misra, 2014; Vu et al., 2006). It could be speculated that the increase in the NADP-MDH expression level under the LC treatment is due to the effect of elevated CO₂.

Under LC treatment, DNA methylation did not affect the expression of key enzymes in the Calvin cycle or Hatch−Slack cycle and only played an essential role in the coordinated expression of key genes in the starch biosynthetic pathway.

TFs also play important roles in regulating gene expression (Lai et al., 2019). By analyzing coexpression networks with key genes of the starch biosynthetic pathway, we predicted that some TFs were positively correlated. We found that multiple OBF1 genes, which are TFs of the bZIP family in duckweed, were positively correlated with all key genes involved in starch biosynthesis (AGPase, SSS, GBSS, and GBE) and presented high expression levels (FPKM increased from 140 to 400) ( Supplementary Data S16 ). The bZIP TFs are key regulators of starch biosynthesis genes in rice, maize, and wheat and determine starch quality and quantity in the endosperm (Chen et al., 2016; Kumar et al., 2018; Singh et al., 2015; Wang et al., 2013).

The genes associated with carbon assimilation were all contracted ( Table 1 ), but under the LC treatment, only starch biosynthesis ability was significantly enhanced, which was mutually confirmed by the changing trends at multiple levels ( Figures 1A, C , 3 ). Therefore, high starch yield can be obtained merely through the regulation of expression levels. This discovery deserves further consideration and research.

Sucrose biosynthesis and transportation are crucial for starch accumulation in plants (Julius et al., 2017). This process primarily includes photoassimilate synthesis in chloroplasts, transmembrane transport into the cytoplasm, sucrose biosynthesis, and long-distance transport of sucrose for subsequent conversion into starch in storage organs. The “source-flow-sink” relationship is highly related to crop yield. Our results demonstrated that the duckweed frond, a tissue similar to leaves, acts as a “sink” organ that accumulates stored starch ( Figure 1E ).

The transportation of photoassimilates from chloroplasts relies on the triose phosphate/phosphate translocator (TPT), glucose transporter (PGT), and maltose transporter (MEX). Correspondingly, deletions and mutations of TPT, PGT, and MEX lead to starch accumulation in the chloroplast (Cho et al., 2011; Jang et al., 2015; Walters et al., 2004). Duckweed has a markedly contracted number of genes involved in the transportation of photoassimilates. TPT, PGT, and MEX were reduced to only one copy each ( Figure 4 ; Supplementary Data S3 ). Furthermore, the expression levels of both TPT and PGT were significantly downregulated under the LC treatment. Notably, TPT, the major export transporter of photoassimilates from chloroplasts, decreased by 42.6% ( Figure 4 ). In duckweed mesophyll cells, LC treatment significantly increased the expression of AGPase by 3.0× in the chloroplast but suppressed the export of plastidial triose phosphorate and glucose to the cytosol, resulting in the hyperaccumulation of starch in the chloroplast ( Figures 1E , 3E , 4 ). The subcellular localization of starch granules, AGPase, TPT, PGT, and MEX also confirmed the transition of chloroplast function from being the “source” to the “sink” in duckweed ( Figures 1 ; Supplementary Figure S6 ). The source and sink are thus spatially organized together in the chloroplast, in stark contrast to other crops. This treatment allows duckweed chloroplasts to be highly efficient at forming and storing starch.

The volume of the “flow” in duckweed is affected by the quantity of sucrose and the efficiency of the transporter. The sucrose content in duckweed is normally < 1.1 mg g^-1 FW, which is much lower than that in corn and rice ( Supplementary Figure S18 ). Sucrose synthesis is regulated by two main enzymes, sucrose phosphate synthase (SPS) and sucrose phosphate phosphatase (SPP). SPS is a reversible rate-limiting enzyme that catalyzes the synthesis of sucrose-6P using UDP-glucose and fructose-6P and reversely catalyzes the degradation of sucrose-6P (Huber and Huber, 1996). The number of SPS genes (4) in duckweed was lower than that in rice, corn, and cassava ( Supplementary Data S16A ). SPP, more importantly, is present in only one copy, with a very low expression level (FPKM values <5), resulting in the accumulation of the substrate sucrose-6P and subsequently promoting the reverse catalysis of sucrose-6P degradation. Under LC treatment, the expression level of SPS, whose FPKM value was already greater than 84 at the beginning (0 h), increased significantly (Log₂FC =1.26 for 240 h vs. 0 h), leading to further enhancement of the degradation of sucrose-6P and ultimately resulting in a very low sucrose concentration in the cytoplasm ( Supplementary Data S16B ; Supplementary Figure S18 ). Thus, duckweed has an extremely weak ability to synthesize sucrose and has a low sucrose content in its cell cytoplasm, that is, a low volume of “flow”.

Sucrose transporters (SUTs) and hexose and sucrose transporters (SWEETs) are responsible for the long-distance transport of sucrose to nonphotosynthetic organs. Among them, SUTs are the most important transporters. Duckweed possesses only one SUT gene (LpSUT) and 6 types of SWEET genes ( Figure 4 ; Supplementary Data S4 and Supplementary Data S6 ). Compared with Arabidopsis, which possesses 9 SUT genes, 4 of which have high affinity, the SUT protein in duckweed might have low affinity. The extended N-terminus of LpSUT has a lower affinity for sucrose, which is highly similar to SUT2 in Arabidopsis, the sucrose transporter with the lowest affinity (Schulze et al., 2000) ( Supplementary Figure S17 ). We also observed a very low expression level of LpSUT (FPKM values 9.9–19.6) ( Supplementary Data S6 ). On the other hand, duckweed contains many fewer copies of SWEET genes than do Arabidopsis and rice. Duckweed also lacks homologues of SWEET11 and SWEET12, the key sucrose efflux transporters (Chen et al., 2012) ( Supplementary Data S4 ). At the transcriptional level, the expression of the SWEET genes in duckweed was also very low (FPKM values <30) ( Supplementary Data S5 ). Therefore, the number of SUT and SWEET genes and their expression levels showed a weak “flow” ability in the plants. Impressively, the number of genes regulating sucrose transportation was reduced, and only those with low affinity remained. Owing to the reduced gene number and weakened protein activities, sucrose metabolism was markedly suppressed in terms of synthesis and transportation.

The low sucrose concentration, low sucrose synthesis, and low transport capacity resulted in a weak “flow” in duckweed. LC treatment strongly stimulated starch accumulation in the chloroplast and further reduced its sugar flow ability, turning the “source” frond into a “sink” organ ( Figure 4 ). Therefore, duckweed is an unusual and interesting system in which sources and sinks are spatially organized together, unlike the interdependent compartmentation of sinks and sources in other higher plants.