Synergy between the anthocyanin and RDR6/SGS3/DCL4 siRNA pathways expose hidden features of Arabidopsis carbon metabolism

Anthocyanin pigments furnish a powerful visual output of the stress and metabolic status of Arabidopsis thaliana plants. Essential for pigment accumulation is TRANSPARENT TESTA19 (TT19), a glutathione S-transferase proposed to bind and stabilize anthocyanins, participating in their vacuolar sequestration, a function conserved across the flowering plants. Here, we report the identification of genetic suppressors that result in anthocyanin accumulation in the absence of TT19. We show that mutations in RDR6, SGS3, or DCL4 suppress the anthocyanin defect of tt19 by pushing carbon towards flavonoid biosynthesis. This effect is not unique to tt19 and extends to at least one other anthocyanin pathway gene mutant. This synergy between mutations in components of the RDR6-SGS3-DCL4 siRNA system and the flavonoid pathway reveals genetic/epigenetic mechanisms regulating metabolic fluxes.

T he conspicuous colors and non-essential nature of anthocyanin pigments have made their biosynthesis one of the best-studied plant metabolic pathways 1 . Anthocyanins are flavonoid pigments synthesized on the cytoplasmic surface of the endoplasmic reticulum 2 and transported to vacuoles, where the acidic pH bestow upon them characteristic red, purple or bluish coloration. Flavonoids are derived from primary metabolism, through the general phenylpropanoid pathway, from the amino acid phenylalanine or tyrosine 3 . Anthocyanin pathway enzymes can be divided into those involved in making the anthocyanidin core, and those mediating modifications, including glycosylation, acylation, and prenylation 4 . The extensive knowledge on anthocyanin biosynthesis enzymes from many plants contrasts with how little is known about the mechanisms by which these pigments are transported from their site of synthesis to the vacuole 5,6 . Key players in this process are a group of conserved glutathione S-transferase (GST) proteins and ABC transporters that interact with anthocyanins [7][8][9] and are believed to facilitate their transport across the tonoplast 10,11 . The first GST described as associated with anthocyanin transport was maize BRONZE2 (BZ2), initially believed to conjugate glutathione to cyanidin-3-Oglucoside (C3G) 12 . Subsequent studies in maize and other plants, however, failed to identify anthocyanin-glutathione conjugates 7 .
In Arabidopsis, TRANSPARENT TESTA19 (TT19) is the functional homolog of BZ2 and is required for the accumulation of anthocyanins as well as proanthocyanidin (PA) pigments in the seed coat 13 . TT19 shows a broad subcellular distribution and can bind cyanidin as well as C3G 8 . There are eight tt19 mutant alleles that have been characterized to some degree (Supplementary Table 1). How TT19 and related GST proteins function to promote pigment accumulation remains unclear, but some possible roles for these GSTs include the stabilization of anthocyanins and/or PA precursors, or cooperation with transporters for vacuolar uptake 5 .
To investigate possible connections between anthocyanin pathway regulation, metabolic flux, and product sequestration, we screened for genetic suppressors that allow anthocyanins to accumulate in seedlings in the absence of TT19 function. Here, we show that all six tt19-8 suppressors identified correspond to components of the RDR6-SGS3-DCL4 small RNA system. We demonstrate that PAP1 over-expression is insufficient to restore anthocyanin accumulation in the tt19 mutant. Indeed, our metabolic analyses show that, contrary to a pulling effect of PAP1 through an increased demand for carbon from the flavonoid pathway, the combination of tt19 and mutations in rdr6, sgs3 or dcl4 results in a significant push into the pathway, with three to five times more carbon devoted to flavonoid production. In seedlings, this carbon is derived from sucrose through central metabolism, and from the down-regulation of other metabolic pathways. By combining rdr6 with tds4 (impairing the expression of leucoanthocyanidin dioxygenase) 28 , we demonstrate that the increased carbon influx to the flavonoid pathway is not unique to the combination of tt19 and rdr6 mutations, but more likely a generalized consequence of the inability to properly accumulate flavonoids. Our findings support a model in which the small RNAs derived from the RDR6-SGS3-DCL4 system provide a regulatory layer over central metabolism that is only exposed when combined with specialized metabolism dysfunctions.
The eight tt19 suppressors were back-crossed with tt19-8 and F2 seeds (BC 1 F 2 ) were grown in AIC. The resulting seedlings showed the expected segregation of~25% pigmented and~75% non-pigmented, confirming that the tt19 suppressing mutations were all recessive. Using bulked segregant analysis by whole-genome re-sequencing, we found that the eight causative mutations were located in two regions. The tt19 suppressors S1, S2, S4, and S8 corresponded to allelic mutations in At3g49500 (RDR6), with the S1 mutation replacing Q with a stop codon at position 293 (RDR6 Q293stop ), S2 replacing Q with a stop codon at position 1,055 (RDR6 Q1055stop ), S4 changing G664 for R (RDR6 G664R ), and S8 changing W to a stop codon at position 299 (RDR6 W299stop ) ( Fig. 1a and Supplementary Fig. 2). In S3, S5, and S6, the same G-to-A mutation at nucleotide position 2283 of At5g23570 (SGS3) caused a splice defect at the 3'-end of intron 3 in which the normal acceptor site was mutated. The alignment of SGS3 transcript reads from S5 suggested an alternative splicing site 15 nt downstream of nucleotide 2,283 that resulted in the deletion of five amino acids (Fig. 1b). In S7, a G-to-A change at nucleotide position 6,509 resulted in a splice defect at the 5'-end of intron 17 of At5g20320 (DCL4), which was confirmed by RNA-seq (Fig. 1c). In summary, an unbiased screen for phenotypic suppressors of tt19-8 resulted in all the mutations mapping to genes encoding components of the RDR6-SGS3-DCL4 small RNA system (Table 1).
To confirm that the identified mutations were responsible for suppressing the anthocyanin pigmentation deficiency of tt19-8, we crossed tt19-8 to characterized reference mutants for RDR6, SGS3, and DCL4 (rdr6-11 and rdr6-15; sgs3-11 and sgs3-14; and dcl4-2e and dcl4-2t). We confirmed that homozygous double mutant seedlings for tt19-8 rdr6/sgs3/dcl4 accumulated anthocyanins in AIC (upper row, Supplementary Fig. 3). Consistent with PAP1 being controlled by TAS4 via the RDR6-SGS3-DCL4 system 24 , the single rdr6, sgs3, and dcl4 mutant seedlings accumulated more anthocyanins than the wildtype (Col-0) control, at levels comparable to PAP1-D seedlings, in which PAP1 is under the control of four copies of CaMV35S promoter 15 (lower row, Supplementary Fig. 3). Similar restoration of the anthocyanin pigmentation phenotypes were observed in tt19 suppressor plants at different developmental stages grown on soil ( Supplementary Fig. 4). In contrast to the restoration of anthocyanin accumulation in vegetative organs, none of the eight tt19 suppressors was able to reinstate the seed coat PA pigmentation characteristic of wild-type and significantly reduced in tt19-8 plants (Supplementary Fig. 5).
PAP1 is insufficient for restoring tt19 pigmentation. A possible explanation for why mutations in either RDR6, SGS3, or DCL4 overcome anthocyanin pigment deficiency in tt19-8 plants is that more anthocyanins are produced by preventing PAP1 transcript degradation via TAS4-siRNA81 24 and the subsequent upregulation of anthocyanin biosynthesis genes 15 15 . PAP1 mRNA levels were about fivefold higher in PAP1-D seedlings in AIC, compared to Col-0 seedlings (Fig. 2a), and PAP1-D seedlings showed~60% increase in anthocyanin accumulation when compared to Col-0 ( Fig. 2b). Although the PAP1 mRNA steady-state level in tt19-8 PAP1-D seedlings was comparable to that of PAP1-D (Fig. 2a), the tt19-8 PAP1-D line failed to accumulate more anthocyanins than the tt19-8 seedlings (Fig. 2b). The tt19 suppressor lines that have mutations in RDR6 resulted in an elevation of PAP1 mRNA levels and an increase in anthocyanin levels relative to tt19-8.
However, compared to tt19-8 PAP1-D seedlings, the tt19-8 S4 and S8 lines accumulated larger quantities of anthocyanins, despite lower PAP1 mRNA levels (Fig. 2). These results demonstrate that, although PAP1 upregulation is likely necessary for enhanced anthocyanin accumulation in the tt19 suppressor lines, it is clearly insufficient to overcome the inability of tt19 to accumulate high anthocyanin levels. Therefore, the mechanisms by which the tt19 suppressors restore anthocyanin accumulation involve other factor(s), besides PAP1.
Flavonoid accumulation changes in suppressor lines. Arabidopsis accumulates at least 15 distinct anthocyanins (A1-A15) 29,30 , most of which are detectable under the metabolic stress conditions imposed by AIC 22,31 . To compare the anthocyanin composition of the suppressor lines with those present in Col-0 and tt19-8, we grew homozygous seedlings in AIC and analyzed their anthocyanin profiles. Under AIC, the prevalent anthocyanins in Col-0 were A3, A5, and A7 to A11, with A11 being the most abundant ( Supplementary Fig. 6a). Despite tt19-8 accumulating only~10% of the anthocyanin content in Col-0, the relative peak abundance remained unaltered ( Supplementary Fig. 6b). The anthocyanin profiles of the suppressor lines were also indistinguishable from those of wild-type and tt19-8, despite accumulating overall different pigment quantities ( Fig. 1; Supplementary Fig. 6b).
As previously reported, tt19 mutants accumulate significantly larger quantities of flavonols than wild-type plants 8 . Thus, we analyzed the flavonol aglycone composition of seedling from the different genotypes grown in AIC. Our results showed that tt19-8 seedlings had an~25% increase in total flavonol content, primarily represented by quercetin, compared to Col-0. Unexpectedly, S2, S4, S5, S7, and S8 showed 4-to 9-fold increases in Fig. 1 Molecular characterization of the tt19-8 phenotypic suppressors S1-S8 and corresponding anthocyanin accumulation. a Structure of the RDR6 gene indicating the position and nature of the EMS-induced mutations for the S1, S8, S4, and S2 suppressors. The images show 4-day-old seedlings grown in AIC and accumulating various levels of anthocyanins evaluated by absorbance at 532 nm (A 532 ), and normalized to the dry weight of the seedlings. The levels present in Col-0 were taken as 100%. b Structure of the SGS3 gene indicating the mutations for the S3, S5, and S6 suppressors. All three suppressors showed similar levels of anthocyanins in AIC and the same mutation, which abolishes a splicing acceptor site. The genome browser image on the top right shows the reads obtained in RNA-seq experiments for S5, and the region demonstrating the use of an alternate splicing acceptor site in the S5 mutant is shown expanded on the lower image. c Structure of the DCL4 gene indicating the mutation for the S7 suppressor, which abolishes a splicing donor site, as evidenced by the genome browser images comparing the RNA-seq reads of Col-0, tt19-8, and S7. Similar anthocyanin quantification results were obtained from at least three independently repeated experiments. Bar, 400 μm.  quercetin and 2-to 3-fold increases in kaempferol compared to Col-0, with S2, S8, and S7 having the highest levels ( Fig. 3). To determine whether this increase in flavonol accumulation in the suppressor lines reflected an overall increase in other flavonoid classes, the samples were subjected to acid hydrolysis and then the absolute amounts of the flavones apigenin and luteolin, the flavanones naringenin and eriodictyol, the flavanonols dihydrokaempferol (DHK) and dihydroquercetin (DHQ), the anthocyandins perlargonidin, cyanidin, and peonidin, and the flavan 3ol epicatechin were measured ( Fig. 3 and Supplementary Fig. 7). The most significant differences were observed for DHQ, which is almost undetectable in Col-0 and tt19-8, but showed a 20-to 100fold increase in the suppressor lines, compared to Col-0 ( Supplementary Fig. 7). Again, providing evidence that these metabolic changes were independent of the activation of PAP1, the tt19-8 PAP1-D line showed levels comparable to Col-0 for all these compounds, with the exception of naringenin, which was 6-fold higher, but still present in much smaller absolute quantities than flavonols ( Supplementary Fig. 7). No other pathway intermediate was found to accumulate in tt19-8 to explain the significantly lower levels of anthocyanins in this mutant. These results demonstrate that the combination of mutations in TT19 and either RDR6, SGS3, or DCL4 results in metabolic changes that are significantly different from those found in the single mutants, suggesting a synergistic cross-talk between cellular processes controlled by the RDR6-SGS3-DCL4 system and those modulated by TT19.
Enhanced sugar metabolism likely source of extra carbon. Our results show that a significantly larger amount of carbon is devoted to flavonoid biosynthesis in the strong suppressor lines, compared to Col-0 or the single mutants (Fig. 3). To determine the possible source of this extra carbon, we performed untargeted metabolic profiling of the extracts obtained from seedlings corresponding to Col-0, tt19-8, and five suppressor lines under AIC, resulting in the identification of~7,000 altered m/z differentially accumulated features (Supplementary Data 1). Pairwise analyses between tt19-8 and each suppressor line permitted to identify the contrasting features that were mapped to predictive pathways and  Fig. 9a, b). The significant decrease in sucrose and glucose levels in the tt19 suppressor lines compared to tt19-8 indicated that more sucrose and glucose was being consumed in the tt19 suppressors, providing the extra carbon necessary for the enhanced flavonoid production.
To determine to what extent the observed increase in sugar (sucrose and glucose) metabolism was a consequence of the highsucrose levels in the media, we grew Arabidopsis seedlings in water in high-light conditions. Anthocyanin accumulation was still evident, but at about half the levels present in high-sucrose media ( Supplementary Fig. 9c). Logically, sucrose levels were considerably lower (>90% reduction) in extracts of water-grown seedlings ( Supplementary Fig. 9a). Interestingly, the suppressor lines presented a similar reduction in endogenous sucrose and glucose levels as observed when the seedlings were grown in sucrose-rich media, and this reduction was not observed in the tt19-8 or rdr6-11 single mutants ( Supplementary Fig. 9a, b). These results indicate that the additional carbon required for the increased flavonoid biosynthesis in the suppressor lines could be provided by sugar catabolism.
Transcript accumulation changes in suppressor lines. To better understand the cross-talk between the RDR6-SGS3-DCL4 system and the TT19 pathway, we carried out RNA-seq on seedlings grown in AIC for Col-0, tt19-8, and tt19 suppressor lines. The tt19-8 mutant showed differences in mRNA accumulation patterns when compared to Col-0, with 2,642 genes displaying statistically significant (FDR < 0.05) differences (Supplementary Data 2). The genes with~2-fold reduced expression included those involved in late anthocyanin biosynthesis, consistent with the lower accumulation of anthocyanins in tt19-8. However, this  reduction in gene expression does not explain why tt19-8 shows only~10% of the anthocyanin content of wild-type plants (Fig. 1a). Gene expression patterns in the tt19 suppressor lines were rather different from both tt19-8 and Col-0. Gene ontology enrichment of K-means clusters showed that a large number of genes (captured in clusters 1, 2, 3, and 5, Fig. 4a) exhibited different expression in tt19-8 S2 (strong suppressor), compared to tt19-8 or Col-0. Cluster 2 included late anthocyanin pathway genes that showed decreased mRNA accumulation in tt19-8 and early anthocyanin pathway genes (e.g., CHS and CHI) that are also required for flavonol accumulation (Fig. 3, Supplementary Data 3).
One of the genes that showed a dramatic increase in mRNA accumulation in the suppressor lines was At1g02940 (AtGSTF5) encoding a phi-type GST ( Supplementary Fig. 10a). The Arabidopsis Next-Gen Sequence DB 32 shows the existence of a single siRNA that potentially targets AtGSTF5 for degradation. AtGSTF5 is a GST similar to TT19 ( Supplementary Fig. 10b), and therefore, we surmised that it might compensate for the flavonoid binding function missing in tt19-8. Constructs for wild-type AtGSTF5 CDS (WT) as well as for an optimized version (OPT) harboring synonymous substitutions across the entire CDS that would render it insensitive to siRNAs were transformed into tt19-8. We verified expression of both versions of AtGSTF5 by RT-qPCR in the transgenic plants, yet observed no complementation of the anthocyanin deficiency phenotype. To determine whether AtGSTF5 is necessary for the restoration of anthocyanin synthesis in the tt19-8 suppressor lines, we designed two guide RNAs to edit target sequences in exons 2 and 4 of AtGSTF5 by CRISPR/ Cas9 and identified six tt19 suppressor lines with mutated AtGSTF5 as homozygous, harboring various indels in AtGSTF5. However, when we compared the anthocyanin levels between the tt19-8 suppressors with wild-type or mutated AtGSTF5, we did not detect statistically significant differences ( Supplementary  Fig. 10). This result demonstrates that AtGSTF5 is unlikely to participate in the stabilization/vacuolar sequestration of anthocyanins, yet its enhanced expression in the suppressor lines, compared to both Col-0 and tt19-8, suggests a possible role in another cellular function normally carried out by TT19. Whereas all tt19 suppressors had similar effects on anthocyanin accumulation, they showed surprisingly different patterns of gene expression (Fig. 4b), indicating that, while they probably function together as part of the RDR6-SGS3-DCL4 system, each one has separate functions that are independent of the others. The tt19 S4 (RDR6 G664R ) suppressor is potentially interesting, since it showed an effect on mRNA accumulation that was dramatically different from the effects observed for S2 and S8, both harboring nonsense mutations in RDR6 (Fig. 4c). Only 244 differential expressed genes were shared by all the five suppressor lines analyzed against tt19-8 (Fig. 4b, Supplementary Data 4). From these 244 genes, 197 show changes in mRNA accumulation with the same trend (up or down). We rationalized that these 197 genes contained those that are most likely responsible for the ability of the suppressor lines to accumulate anthocyanins in the absence of TT19 function.
As anticipated, among these 197 genes, there were 22 upregulated genes related to phenylpropanoid/flavonoid biosynthesis or regulation, and an additional 26 repressed genes associated with various metabolic or regulatory activities (Fig. 4b,  Supplementary Data 4). Among those 26 genes, six have been functionally characterized (or predicted) to be involved in competitive pathways with phenylpropanoid/flavonoid biosynthesis, including CYP79B3 and MYB34 involved in indole glucosinolate biosynthesis 33,34 ; CYP84A4 involved in arabidopyrone biosynthesis 35 ; CYP90A1/CPD involved in a critical step for brassinosteroid biosynthesis 36 ; UGT71C1 involved in the decoration of salicylic acid 37 ; and flavonol-phenylacyltransferase 1 (FPT1) likely participating in phenylalanine consumption 38 . Thus, a shared feature of all the suppressor lines is the increased expression of phenylpropanoid/flavonoid-related genes, and decreased accumulation of gene products likely competing for the early precursors of the pathway (e.g., Phe and Tyr).
To explore the possible link between the increased sugar metabolism providing the extra carbon and those genes differentially expressed in tt19 suppressor lines compared to Col-0 and tt19-8, we analyzed GO term enrichment in the six clusters shown in Fig. 4a. We identified 15 genes under the GO term monosaccharide/hexose/glucose metabolic process only in Cluster 2, corresponding to genes highly expressed in tt19-8 S2 (Supplementary Fig. 11). An analysis of previous published RNAseq data for rdr6-11 and Col-0 from Arabidopsis shoots 39 confirmed that 13 of these genes are highly expressed in tt19-8 S2, but not in rdr6-11 alone. Thus, the upregulation of these 13 carbohydrate metabolic process-related genes in the tt19 suppressors provides one likely explanation for the increased carbon flux towards flavonoids in the suppressor lines.
We investigated whether the synergistic effect of the tt19 and RDR6-SGS3-DCL4 system mutations on increased flux toward the flavonoid pathway involved small RNAs. We performed a global analysis of small RNAs (20 to 24 nt) between Col-0, tt19-8, and tt19-8 S2 and S7 seedlings, all grown in AIC (Supplementary Fig. 12a and Supplementary Data 5). As anticipated, the suppressor mutations abolished the biogenesis of 21 nt tasiRNAs, which was also evidenced by the increased accumulation of the TAS4 precursor RNA in all the suppressor lines, as well as increased mRNA accumulation of tasiRNA gene targets, including the TAS4 targets PAP1 and PAP2 24 , and TAS3 target ARF4 40 . We found that levels of five miRNAs were significantly different between Col-0 and tt19-8 seedlings (Supplementary Fig. 12b). Significantly higher in tt19-8 than in Col-0 were miR158b, miR167a-5p and miR399c-5p, while miR161.2 and miR164b-5p showed the opposite pattern ( Supplementary Fig. 12b). The increased expression of miR158b in tt19-8 did not revert to wildtype levels in the S2 and S7 suppressors. Similarly, the low accumulation of miR161.2 in tt19-8 compared to wild-type, was not restored in the suppressor lines, indicating that the accumulation of miR158b and miR161.2 does not involve the RDR6-SGS3-DCL4 system. In contrast, the effect of the tt19-8 mutation on the accumulation of miR399c-5p, miR167a-5p, and miR164b-5p was abolished in the S2 and S7 suppressor lines. These results suggest that these miRNAs (or their targets) participate in the cross-talk between the flavonoid pathway and the RDR6-SGS3-DCL4 system.
Since RNA-seq analysis identified 13 carbohydrate metabolic process-related genes with higher mRNA steady-state Fig. 4 RNA-seq analysis of Col-0 and Arabidopsis mutant seedlings grown in AIC. a Total differentially expressed genes among Col-0, tt19-8, and tt19-8 S2 were grouped into six clusters. b The UpSet plot summarizes all intersections of significantly differentially expressed genes between tt19 suppressors and tt19-8, sorted by degree. The combinations of intersections are indicated as dark circles in the matrix and the numbers of differential expressed genes across various exclusive intersections are represented as vertical columns with number of genes on the top. The bars plot on the bottom left displays the set number of total differentially expressed genes in each tt19 suppressor line against tt19-8. The clustering plot between Col-0, tt19-8, and tt19-8 S2 on the top right indicates that among the 244 shared genes (highlighted with red column) across all suppressors, 22 flavonoid-related genes (pink lines, including early pathway gene 4CL3; core flavonoid pathway genes CHS, CHI, F3'H, DFR, and LDOX; anthocyanin-modification genes UGT78D2, UGT79B1, 3AT1, 3AT2, UGT75C1, 5MAT, and SAT; flavonol biosynthesis genes BGLU1 and UGT73B2; PA biosynthesis genes AHA10 and LAC15; regulatory genes TT8, TTG2, and MYBL2 as well as ABCC2 and CHIL) are part of Cluster 2 and 26 other metabolic genes (blue lines) are part of Cluster 5 in a. Dark blue lines display expression changes of six genes that are involved in competitive pathways with flavonoid biosynthesis. c Global gene expression clustering analysis on Col-0, tt19-8, and tt19 suppressors (S2, S4, S5, S7, and S8). NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-16289-3 ARTICLE NATURE COMMUNICATIONS | (2020) 11:2456 | https://doi.org/10.1038/s41467-020-16289-3 | www.nature.com/naturecommunications accumulation in tt19-8 S2, we mapped the reads from the small RNA-seq analysis to these differentially expressed genes. One of the genes, At3g43190 (SUS4), encoding SUCROSE SYNTHASE 4, showed increased mRNA levels in tt19-8 S2 ( Supplementary  Fig. 12c). Our analyses identified a small RNA (known as siR6611 41 or AtsRR178278 42 ) derived from the PPR (Pentatricopeptide repeat) gene (At1g62930) whose expression is significantly decreased in the suppressor lines (Supplementary Data 5), and which may target SUS4 mRNA for cleavage 43 (Supplementary Figure 12c and Supplementary Data 6). Indeed, the formation of siR6611 required miR161.2 and tasiRNAs derived from TAS2 44-46 , providing one mechanism by which the RDR6-SGS3-DCL4 system and the TT19 pathway may converge on the control of sugar metabolism (Supplementary Fig. 12d).
Second flavonoid mutant similarly affected by rdr6 mutation. The results so far cannot distinguish whether the synergistic effect of RDR6-SGS3-DCL4 and tt19-8 on the increase of carbon flux into the flavonoid pathway relates to the absence of TT19 function itself, or to the abnormal flavonoid accumulation that characterizes the tt19-8 mutant (Fig. 3). To distinguish between these two possibilities, we generated a double mutant between rdr6-15 and tds4-4. tds4-4 corresponds to a null allele of the anthocyanidin synthase (ANS) gene, which encodes an enzyme that converts leucoanthocyanidins into 3-hydroxyanthocyanidins 47 , pathway intermediates involved in anthocyanin but not flavonol formation. We then conducted flavonoid aglycone composition analyses in single (rdr6-15, tt19-8, tds4-4) and double (tt19-8 rdr6-15 and tds4-4 rdr6-15) mutant seedlings grown in AIC. The total flavonoid levels of Col-0, tt19-8 and tds4-4 were very similar; the main difference being that tds4-4 seedlings accumulated no anthocyanins (Fig. 5). The total flavonoid levels of tds4-4 rdr6-15 were comparable to those in tt19-8 rdr6-15, and approximately three times larger than those in Col-0 (Fig. 5). These results demonstrate that the synergistic effects observed in the suppressor lines are likely to involve an interaction between components of the RDR6-SGS3-DCL4 system and flavonoid compounds affected in both the tt19-8 and tds4-4 mutants.
To further delineate what aspect of flavonoid pathway dysfunction might be responsible for the interaction with the RDR6-SGS3-DCL4 system to increase carbon flux into the pathway, we generated the tt5-4 rdr6-15 double mutant, in which tt5-4 corresponds to a mutation in chalcone isomerase, an early pathway step necessary for both flavonol and anthocyanin accumulation ( Supplementary Fig. 7). Similar to tt5-4, but different from Col-0 or rdr6-15, tt5-4 rdr6-15 seedlings grown in AIC lack the flavonols quercetin and kaempferol, as well as cyanidin, yet show an increase in naringenin chalcone/naringenin (hydrolysis converts naringenin chalcone to naringenin), because of the block in the pathway (Supplementary Fig. 13a-d). When we evaluated the absolute values of glucose and sucrose in the tt5-4 rdr6-15 double mutant, they were not significantly different from those in Col-0 or single mutants ( Supplementary Fig. 13e, f), indicating that it was not the absence of flavonoids that synergized with the RDR6-SGS3-DCL4 system, but an imbalance between different flavonoids caused by the tt19-8 or tds4-4 mutants.
We then took advantage of the possibility to feed seedlings grown in AIC with various flavonoids to determine whether the accumulation of flavonols in the absence of anthocyanins would have an effect. For this, we added quercetin or kaempferol, but neither was sufficient to trigger an enhanced catabolism of glucose or sucrose ( Supplementary Fig. 13e, f), despite flavonols accumulating at high levels in the seedlings ( Supplementary  Fig. 13a, b). The addition of naringenin (the product of chalcone isomerase) resulted in flavonol and cyanidin formation in ratios slightly different from those present in Col-0, but still not sufficient to trigger enhanced sugar catabolism ( Supplementary  Fig. 13). The results suggest that what triggers enhanced sugar catabolism in the presence of a disabled RDR6-SGS3-DCL4 system is the absence of anthocyanins (regardless of whether flavonols are present or not), or a pathway dysfunction, likely characterized by the accumulation of a pathway intermediate specific to the anthocyanin branch of the pathway present in both tds4-4 and tt19-8, but absent in tt5-4.

Discussion
Our quest for phenotypic suppressors of tt19 resulted exclusively in the identification of multiple recessive alleles of components of the RDR6-SGS3-DCL4 siRNA pathway, which is involved in the post-transcriptional gene silencing of a large number of genes 39 . It is notable that the only identified context in which anthocyanins can still accumulate in a tt19 mutant is by disabling a major regulatory mechanism that causes global gene expression changes. Our results demonstrate that a consequence of these gene expression changes is a flux increase into the phenylpropanoid/ flavonoid pathway, resulting in greater levels of anthocyanins and other flavonoids (Fig. 6). Thus, while our results cannot rule out a role of TT19 in the stabilization, solubilization, and/or vacuolar sequestration of anthocyanins, it is evident that if enough anthocyanins are produced, they can accumulate, even in the absence of TT19. A possibility that we considered is that the gene expression changes promoted by the tt19 suppressors might induce a TT19-like protein. We identified AtGSTF5, which is normally expressed at very low levels in both wild-type and tt19-8, as one of such candidates. However, by combining gain-and loss-of-function approaches, we showed that it is neither necessary nor sufficient to overcome the absence of TT19 function (Supplementary Fig. 10). Another possibility that our experiments ruled out is that enhanced expression of PAP1, as a consequence of disabling the RDR6-SGS3-DCL4 siRNA system 24,27 , would be sufficient to overcome the need for TT19 (Fig. 2). Indeed, our results show that the augmented anthocyanin accumulation that characterizes rdr6, sgs3, and dcl4 plants 48 occurs largely through the impaired tasiRNA-mediated degradation of the PAP1 mRNA. Thus, while increased PAP1 expression is necessary, it is clearly insufficient to explain the anthocyanin pigmentation of the tt19 suppressors.
The tt19 suppressors are characterized by a three to fivefold increase in carbon flow into the flavonoid pathway (Fig. 6). Increase in PAP1 mRNA accumulation, either by PAP1-D or when the RDR6-SGS3-DCL4 siRNA system is disabled (Fig. 6), results in a modest (~1.7 fold) increase in overall carbon to the pathway, likely caused by the transcriptional activation of late pathway genes by PAP1 17 , which results in a flavonoid biosynthesis pull. In contrast, we propose that most of the additional carbon dedicated to flavonoid accumulation in the tt19 suppressor lines is a consequence of carbon being pushed into the pathway, by a synergistic interaction between the loss RDR6-SGS3-DCL4-controlled siRNAs and a flavonoid metabolic dysfunction. In addition, we show that the cross-talk between the impaired RDR6-SGS3-DCL4 system and flavonoid pathway dysfunction is not specific to the loss of TT19 as a second anthocyanin pathway mutant (tds4-4) shows a similar carbon flow synergistic effect with rdr6 (Fig. 5).
What is the source of the additional carbon pushed into the flavonoid pathway in the tt19 suppressors? The RNA-seq results demonstrate that part of the carbon is likely coming from pathways that compete with phenylpropanoid/flavonoid biosynthesis tt19 -8 S7 (DCL4 splice defect ) Fig. 6 Flavonoid flux maps displaying total carbon input into the different branches of the pathway. The block arrow (black at the top) displays the estimated total input of carbon for flavonoid biosynthesis. Green and purple arrows show the distribution of flavonols and anthocyanins, respectively. The data used for the graphical representation of carbon distribution is provided in Supplementary Fig. 7.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-16289-3 ARTICLE NATURE COMMUNICATIONS | (2020) 11:2456 | https://doi.org/10.1038/s41467-020-16289-3 | www.nature.com/naturecommunications (Fig. 4b). However, the results of untargeted metabolic profiling indicated that the majority of the carbon is actually derived from alterations in central metabolism, including carbohydrate degradation ( Supplementary Fig. 8). Indeed, sucrose and glucose levels are dramatically decreased in the tt19 suppressor lines, with little or no changes in the tt19-8 or rdr6-11 seedlings ( Supplementary  Fig. 9). Interestingly, the induction of anthocyanin triggered by PAP1-D or rdr6 had minimal, if any effect on sugar metabolism ( Supplementary Fig. 9), indicating that flavonoid increases resulting from pulling can be absorbed by intermediates or by carbon diversion from other pathways, further underscoring the differences between the pull and push mechanisms.
What are possible molecular mechanisms underlying the synergistic interaction between the loss of RDR6-SGS3-DCL4-controlled siRNAs and the flavonoid metabolic dysfunction? The RDR6-SGS3-DCL4 siRNA system controls the expression of a large number of genes 39 , and flavonoids have been shown to play many regulatory roles [49][50][51][52] , thus the opportunity for synergistic interactions are multiple, reflected in the rather striking effects on gene expression (Fig. 4). However, the activation in the tt19 suppressor lines of SUS4 offers insights on some unexpected molecular mechanisms that could lead to major metabolic alterations. Indeed, miR161.2 accumulation is decreased in tt19-8, and does not recover in the suppressor lines (Supplementary Fig. 12b). One of the targets of miR161.2 is a PPR gene, which is also targeted by a product of the RDR6-SGS3-DCL4 system on TAS2 44,45 . This results in the formation of siR6611, which is predicted to target SUS4 for degradation [41][42][43] (Supplementary Fig. 12c). Our results show that, because of the decreased accumulation of miR161.2 and the disabled RDR6-SGS3-DCL4 system in the tt19 suppressors, the expression of siR6611 is decreased resulting in SUS4 mRNA increasing (Supplementary Fig. 12d). Whether miR161.2 or additional miRNAs are affected in other anthocyanin biosynthesis mutants to explain the altered expression of genes involved in central metabolism remains to be determined.
Our findings are consistent with a model in which, under normal conditions, RDR6-SGS3-DCL4-controlled small RNAs maintain the homeostasis of central metabolism. In such conditions, the disabling of the small RNA pathway has modest consequences, reflected in the surprisingly normal phenotype of rdr6/ sgs3/dcl4 mutants 25,53-56 . When a specialized metabolic pathway is perturbed (such as resulting from the tt19-8 mutation or from PAP1 overexpression), small RNAs adjust expression of the enzymes and regulators involved in central metabolism to accommodate to this new condition. This model suggests the existence of a surveillance apparatus that communicates metabolic dysfunctions to the RDR6-SGS3-DCL4 system, and miR161.2 could be part of it. However, the many regulatory effects of flavonoids, particularly through the modulation of hormone function [49][50][51][52] , and the control of small RNAs by plant hormones (and vice versa) 57-59 provide opportunities for the identification of other components of this surveillance system.
In conclusion, the search for tt19 phenotypic suppressors resulted in the identification of a cross-talk mechanism between central metabolism and flavonoid biosynthesis that involves the RDR6-SGS3-DCL4 siRNA system. Whether similar mechanisms are involved in the coordination of central metabolism and other specialized metabolic pathway is not yet known, but both small RNAs and specialized metabolites are rapidly evolving 60,61 , providing ample opportunities for emerging synchronization, and for the metabolic engineering of desirable phytochemicals that might be limited by carbon flow from central metabolism.
EMS mutagenesis to isolate tt19 suppressors. Taking advantage of the very low anthocyanin accumulation of tt19-8 seedlings, we screened 2.5 g (~125,000) of EMS-treated seeds for potential dominant suppressors by scoring restoration of anthocyanin pigmentation of 4 day-old M 1 seedlings grown in AIC. We then transferred~10,000 of the M 1 seedlings to soil, grew to maturity and collected intõ 300 pools of M 2 seeds. The recessive suppressors that restored anthocyanin accumulation were isolated from the M2 seed pools germinated under AIC.
Anthocyanin extraction and composition analysis. Arabidopsis seedlings were grown in AIC for four days. Seedlings were then lyophilized and dry tissue weighted. Total anthocyanins were extracted with 50% methanol containing 3% formic acid at 200 μg of dry weight per μL of extraction buffer, overnight at room temperature. To evaluate total anthocyanin concentration, the extracts were subjected to spectrophotometry and absorbance at 532 nm was normalized to the dry weight of the tissue (and expressed as A 532 /mg of dry tissue). For HPLC, 20 μL of the acidic methanol extracts were analyzed with a 2695 Alliance HPLC system (Waters Corp., Beverley, MA) on a Symmetry C18 column (Waters, 3.5 μm 4.6 × 75 mm) at 35°C with flow rate 1 mL/min. Buffer A corresponds to 5% formic acid in water and buffer B to 5% formic acid in acetonitrile. The gradient running condition was: 0 min, 100% A; 20 min, 75% A and 25% B; 22 min, 20% A and 80% B; 22.1 min, 100% B; 25 min, 100% B; 25.1 min, 100% A; and 30 min, 100% A. Anthocyanin profiles were recorded at 532 nm and individual anthocyanin peaks from the HPLC were collected and applied to the LC-MS/MS to determine anthocyanin composition 52 .
Flavonoid quantification. Aglycone flavonoids were prepared by HCl hydrolysis 62 . Briefly, lyophilized Arabidopsis seedlings (four days in AIC) were soaked with 2 M HCl (1 mg dry weight in 10 μL) and incubated at 100°C for 20 min. Half volume of isoamylalcohol (1 mg dry weight in 5 μL) was then mixed to extract and concentrate aglycone flavonoids (top layer after centrifugation). The amount of flavonoids in each sample was detected by multiple reaction monitoring (MRM) and quantified by comparing with flavonoid standards on high-performance reverse phase liquid chromatography coupled with triple quadruple mass spectrometer 63 using Waters MassLynx v4.1 SCN805. For creating standard curves of flavonoid compounds, the flavonoid standards were dissolved in 80% of methanol with 0.1% formic acid. The UHPLC-MS/MS run was 10 min with a flow of 0.3 ml/min at 40°C. The gradient was as follows with solvent A (0.1% formic acid in water) and solvent B (0.1% formic acid in acetonitrile): 0-0.5 min, 90% A and 10% B; 0.5-6 min, 50% A and 50% B; 6-8 min, 0% A and 100% B; and 8.01-10 min 90% A and 10% B.
Untargeted metabolite profiling and LC-MS/MS data analysis. For extraction of total metabolites, lyophilized Arabidopsis seedlings (four days in AIC or water) were incubated with methanol/water (1/1, v/v) containing 0.1% formic acid and 1 μM telmisartan (internal standard) at the ratio 1 mg dry weight in 10 μL extraction buffer overnight at room temperature. After centrifugation at 10,000 xg for 5 min, the supernatant was then diluted ten times with 80% of methanol with 0.1% formic acid for LC-MS/MS analysis.
All the genotypes were analyzed in triplicate by LC-MS/MS using Acquity UHPLC system with HSS T3 C18 column (100 × 2.1 mm, 1.8 μM, Waters, Milford, MA) linked to the mass spectrometer (Xevo G2-XS QTOF, Waters). At a flow rate of 0.3 mL/min and the column temperature of 40°C, a 10 min gradient run was applied with solvent A (0.1% formic acid in water) and solvent B (acetonitrile): 0-0.5 min, 100% A; 0.5-6 min, 50% A and 50% B; 6.01-6.5 min, 1% A and 99% B; 6.51-8.5 min, 1% A and 99% B; and 8.51-10 min 99% A and 1% B. Mass spectrum acquistion was performed in negative mode with m/z scan range at 50-2000. All MS raw data files (.raw) were converted to mzXML format using ProteoWizard version 3 64 and then uploaded to XCMS online 65,66 (https:// xcmsonline.scripps.edu/) for data processing. For all the samples, peak alignment (feature detection), candidate metabolite identification, MS/MS matching, and statistical analysis were performed. Feature detection and pairwise analysis were performed with the following paramerters: centWave method for feature detection (Δm/z = 15 ppm, minimum peak width = 5 s, and maximun peak width = 50 s); obiwarp method for retention time correction (profStep = 1); and settings for alignment (bw = 5, minfrac = 0.5, mzwid = 0.015) and statistics (Welch t-test). The total numbers of aligned features and pairwise cloud plots were exported from XCMS online and reported in Supplementary Data 1 and Supplementary Fig. 8, respectively.
Quantitative RT-PCR. Total RNA was extracted using QIAGEN RNeasy Plant Mini Kit. First-strand cDNA was synthesized from 1.5 μg of RNA using Super-Script II Reverse Transcriptase (ThermoFisher Scientific) and then diluted 1:5 with nuclease-free water. Quantitative PCR was performed in a 10 μL of mixture including 5 μL of SYBR Green PCR Master Mix (ThermoFisher Scientific), 1.2 μL of gene specific primer pair (0.6 μL of each 5 μM forward and reverse primers), and 2 μL of the cDNA samples. Quantitative PCR was conducted with the ABI QuantStudio 7 Flex PCR system (ThermoFisher Scientific) with the following cycling conditions: 50°C for 2 min, 95°C for 10 min, then followed by 40 cycles of 95°C for 15 s, 58°C for 10 s, and 72°C for 30 s. PCR results were processed and visualized using QuantStudio Real-Time PCR software v1.3 (ThermoFisher Scientific). The experiments were performed in biological replicates (3 or more) and the housekeeping gene Actin2 was used as the internal control for normalization of expression levels. The primer set used in this study is listed in Supplementary Table 2.
Whole-genome re-sequencing. Total genomic DNA from pooled leaves of 50 Arabidopsis BC 1 F 3 pigmented plants and 50 non-pigmented plants were extracted using Wizard Genomic DNA Purification Kit (Promega, WI). The pooled gDNA samples were then sent to BGI genomic services (http://www.genomics.cn/en/ index) for 150 bp paired-end Illumina sequencing. The raw data quality check was done using FastQC (v0.11.5). Reads were aligned to the Arabidopsis reference genome (TAIR10) using Bowtie2 (v2.3.4). In total, approximately 340 million paired end reads were obtained. The alignment files were sorted and converted to BAM files using samtools (v1.6). The Mpileup tool in SAMtools was used for SNP calling and the conversion into VCF matrix was carried out using bcftools (v1.2). Variants were parsed using vcftools (v0.1.15) to include only high-quality single SNPs, which were defined as those with a minimum read depth of 5, a maximum read depth of 200 and a minimum mapping quality greater than 20 (-minDP 5;-MaxDP 200;-MinMQ 20;-remove-indels). SNP frequency was then calculated and used to filter each of the suppressor lines. Filter criteria was based on common SNPs in each suppressor line which had an allele frequency greater than 0.8 in the samples from pigmented plants and less than 0.4 in the corresponding samples from non-pigmented plants.
RNA-seq analysis. Total RNA was extracted from Arabidopsis seedlings in AIC for four days using QIAGEN RNeasy Plant Mini Kit (Qiagen, Germany) according to the manufacturer's instructions. Around 1 μg of RNA was used for construction of the RNA-seq libraries and sequenced with 100 bp paired-ends on an Illumina sequencing platform by BGI.
RNA-seq reads were evaluated for quality using FastQC (v.0.11.5). The reads were then aligned to the Arabidopsis thaliana reference genome (TAIR 10) using Hisat2 (v2.1). Raw read counts for gene features were then quantified from these alignments using FeatureCounts (v1.5). Transcript levels were calculated in transcripts per million (TPM). The R package DESeq2 (v1.24) was employed to test differential expression (DE) by paired contrast between tt19-8 and Col-0 lines, and then between each suppressor lines against tt19-8 and Col-0, and the resulting Pvalues were adjusted using the Benjamini and Hochberg's method. Genes with an adjusted P-value <0.05 found by DESeq2 were assigned as differentially expressed, irrespectively of the fold change value, unless otherwise specified. The set of DE genes shared between different suppressor lines were assessed using a local R script and visualized using UpSetR package 67 .
Soft clustering was applied using Mfuzz (v2.44) 68 on normalized centered gene expression values with an optimal fuzzifier value of 3.6 estimated from the total set of expressed genes (a total expression higher than 100 TPM using all the samples) to obtain a better view of the common behavior between the suppressor lines. The number of clusters was approximated by a built-in function and further adjusted until a suitable number of clusters appeared, i.e., without any empty group and with inter-cluster correlation value lower than 0.85.
Small RNA-seq analysis. Total RNA was used to generate small RNA-seq library and sequenced with 50 bp single ends using the BGISEQ-500 platform. Adapter sequences were trimmed-off from the remaining high-quality sequences using Adapter Trimming for Small RNA-Sequencing tool (v0.3.2, http://www.bcgsc.ca/ platform/bioinfo/software/adapter-trimming-for-small-RNA-sequencing) and sequences smaller than 15 nt were discarded. Unique small RNA tags were mapped to Arabidopsis thaliana reference genome (TAIR 10) using Bowtie (v1.1.2). In order to obtain miRNA abundance, we used the miRBase 22.1 database (http:// www.mirbase.org/) annotations for Arabidopsis thaliana. The siRNA annotation was performed using compiled annotations from Araport11 41 .
In the global quantification analysis of small RNAs derived from miRNAs, TAS loci, transposable elements, and protein-coding genes, only small RNAs unique to single gene categories were included. For the quantification of small RNAs, we used a local Perl script for exact quantification, using loci length and read size restrictions depending of the small RNA category analyzed. For read normalization and differential expression analysis of small RNAs, we used edgeR (v3.26.5) to apply trimmed means of M-values as a normalization method (TMM) with FDR threshold value less than 0.05.
Statistical analyses. Statistical analyses were conducted by GraphPad Prism 6.0c software (San Diego, California).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.