Drought response transcriptomes are altered in poplar with reduced tonoplast sucrose transporter expression

Transgenic Populus tremula x alba (717-1B4) plants with reduced expression of a tonoplast sucrose efflux transporter, PtaSUT4, exhibit reduced shoot growth compared to wild type (WT) under sustained mild drought. The present study was undertaken to determine whether SUT4-RNAi directly or indirectly altered poplar predisposition and/or response to changes in soil water availability. While sucrose and hexose levels were constitutively elevated in shoot organs, expression responses to drought were most altered in the root tips of SUT4-RNAi plants. Prior to any drought treatment, constitutively elevated transcript levels of abscisic acid biosynthetic genes and bark/vegetative storage proteins suggested altered metabolism in root tips of RNAi plants. Stronger drought-stimulation of stress-inducible genes encoding late-embryogenesis-abundant proteins in transgenic roots was consistent with increased vulnerability to soil drying. Transcript evidence suggested an RNAi effect on intercellular water trafficking by aquaporins in stem xylem during soil drying and recovery. Co-expression network analysis predicted altered integration of abscisic acid sensing/signaling with ethylene and jasmonate sensing/signaling in RNAi compared to WT roots. The overall conclusion is that steepened shoot-root sugar gradient in RNAi plants increased sensitivity of root tips to decreasing soil water availability.


Shoot-root sucrose partitioning differed between WT and RNAi plants. As in our previous
reports, transcript levels of the single-copy tonoplast PtaSUT4 were reduced by ~75% throughout the plant while transcript levels of other SUT genes were not reduced in WW plants 4 ( Supplementary Fig. S1). DR treatment resulted in a trend of PtaSUT4 increases, followed by a decrease to pre-stress (WW) levels at RC in both genotypes, but with greatest significance in RNAi plants (Fig. 1). Transcript levels of the predominant plasma membrane SUT, PtaSUT3, were high in the sink organs examined ( Supplementary Fig. S1), and expression changes were regarded as indicative of changes in sink demand for sucrose 6,8 . PtaSUT3 exhibited a strong DR-promoted increase in stem (bark and xylem) of WT, and in roots of RNAi plants (Fig. 1A). To further assess whether sucrose partitioning within or between shoot and root was sensitive to the RNAi, sucrose and its constituent hexoses were quantified ( Fig. 2A). Prior to DR, sucrose and total soluble sugars (sum of sucrose and hexose) were sharply elevated in all shoot organs of RNAi plants, and those high levels were sustained throughout the treatment cycle ( Fig. 2A). Sucrose and hexose levels of WT source leaves changed little at DR, but increased considerably at RC. Hexose levels trended downward in bark and xylem of both genotypes at DR, but recovered upon RC. The DR hexose decrease was significant in transgenic xylem, and resulted in a significant decrease of total sugars. While total sugar levels clearly oscillated with DR and RC in transgenic xylem, total sugar levels remained more constant in WT xylem throughout the treatment cycle ( Fig. 2A). Due to the large promotive effects of SUT4-RNAi on sugar concentrations in xylem and bark, starch was also evaluated in those tissues. Xylem starch levels did not differ between genotypes or change with treatment ( Fig. 2B). By contrast, starch was generally more abundant in transgenic bark relative to WT throughout the treatment cycle (Fig. 2B). Sucrose levels trended slightly lower in WT than transgenic roots before stress (P = 0.11), and exhibited a tendency to increase in WT but not RNAi plants under DR. In a follow-up experiment using a repeated measures root sampling approach, sucrose levels decreased significantly in RNAi, but not in WT root tips upon DR (Fig. 2C). Data from both experiments were consistent with reduced sucrose supply to, unloading, or retention within, transgenic compared to WT root tips during soil drying.
A candidate gene approach was used to mine the RNA-Seq data for transcript abundance of genes (other than SUTs) with direct relevance to sugar homeostasis. No transgenic effects on the expression of sucrose cleaving enzymes, sucrose synthases and invertases, were observed across all tissues under WW ( Supplementary Fig. S2). During DR, vacuolar invertase transcript abundance increased in all organs but xylem (Fig. 1B), and the increases were generally larger in WT than RNAi plants. To summarize, the RNAi effects on sugar homeostasis and its drought response dynamics were strongest in xylem ( Fig. 2A) where high transcript levels of PtaSUT3 ( Supplementary Fig. S1) and sucrose metabolism genes ( Supplementary Fig. S2) were commensurate with heavy sucrose trafficking. SUT4-RNAi affected drought response trajectories of aquaporin genes. In light of differential SUT4-RNAi effects on organ water content, e.g., wood versus bark 14 , and potential effects of that on osmotic gradients, we examined transcript abundances in the aquaporin (AQP) gene family. AQPs, both plasma membrane (PIPs) and tonoplast intrinsic proteins (TIPs), permit the passive and selective transport of water across membranes, thereby facilitating water uptake and redistribution in response to osmotic and turgor changes [15][16][17] . AQPs are expressed plant-wide in Populus and their expression is sensitive to osmotic or hydrostatic pressure gradients, Scientific RepoRts | 6:33655 | DOI: 10.1038/srep33655 water availability, embolisms, and transpiration demand [18][19][20][21] . No transgenic effects on total AQP transcript abundance (sum of all AQP transcripts) were observed in WW plants (Fig. 3). However, trends consistent with an overall negative SUT4-RNAi effect on AQP expression in DR leaves and roots, and a promotive effect in xylem were observed (Fig. 3A). During the RC phase, AQP transcript levels trended lower in transgenics, especially in xylem and roots, organs where AQP expression is normally strongest (Fig. 3A).
Examining xylem and root expression more closely, a greater number of AQP genes, primarily PIPs, exhibited significant increases in SUT4-RNAi xylem, while a greater number, primarily TIPs, exhibited significant decreases in WT xylem (DR/WW in Fig. 3B). This resulted in the tendency toward higher transcript abundance of most of the well-expressed PIPs and TIPs in transgenic than WT xylem at DR (RNAi/WT in Fig. 3B). Root transcript levels of several well-expressed PIPs and TIPs were down-regulated by DR both in WT and RNAi plants (Fig. 3C). In the transgenic plants then, opposing xylem vs. root AQP transcript responses to DR were evident (Fig. 3). Following DR, there was a consistent trend toward higher AQP transcript levels in WT than RNAi roots (Fig. 3C). A correlation matrix of PIP and TIP transcript levels between organs was constructed in order to compare systemic coordination of AQP expression in WT and transgenic plants ( Supplementary Fig. S3). This analysis was carried out in recognition of the potential for altered root-shoot sugar concentration gradients (Fig. 2) to osmotically affect water trafficking. Interestingly, strong inter-organ correlations of AQP transcript level were frequently observed. The analysis revealed stronger plant-wide coordination of PIP2.7, PIP1.5 and PIP1.2 than of other PIPs in both WT and transgenics. The overall AQP correlation was strongest between xylem and bark. However, correlations between leaves and other organs were frequently less robust in transgenics, especially for PIP2.5 and PIP2.6. Correlations that included roots were generally very poor in the transgenic plants ( Supplementary Fig. S3).

RNA-Seq revealed latent stresses in root tips of SUT4-RNAi plants. In light of the evidence for
RNAi effects on water as well as sucrose trafficking within and between organs, the potential for there to be effects on gene families known to be associated with perceived shifts in plant or organ osmotic or turgor status was considered. Approximately 100-300 genes, depending on the organ, exhibited differential expression (DE, Q ≤ 0.05 and fold-change [FC] ≥ 1.5) between pre-stress WT and RNAi plants (Fig. 4, Supplementary Dataset S1). Expression evidence for abscisic acid (ABA) involvement was anticipated, and transcript levels of NCED, which encodes nine-cis-epoxycarotenoid dioxygenase for the rate-limiting step in ABA biosynthesis 22 , were perturbed in RNAi plants (Fig. 4B). Under WW conditions, NCED transcript levels were more abundant in bark, xylem and roots of RNAi than WT plants. In accordance with the known drought responsiveness of NCED 23 , the two expressed poplar homologs were highly up-regulated by DR in all four organs of both WT and RNAi plants (Fig. 4B). DR treatment affected the transcript abundance of several thousand genes in each organ of WT and RNAi plants (Fig. 4C), with the greatest response occurring in transgenic roots (Fig. 4C). Many of the genes that encode late embryogenesis abundant (LEA) proteins are known to be induced by desiccation and ABA 24,25 , and a large subset were among the most strongly up-regulated in DR roots, with the response being strongest in the transgenics ( Supplementary Fig. S4A). Just as striking was the magnitude of the DR induction of small heat shock proteins (HSP20s) in transgenic compared to WT roots ( Supplementary Fig. S4B). Transcript levels of several nucleoside phosphorylase-like vegetative/bark storage proteins (VSPs/BSPs) were also constitutively elevated in transgenic roots, and/or exhibited stronger DR-stimulated increases in transgenic than WT roots ( Supplementary  Fig. S4C). Together, the results were consistent with a constitutive RNAi effect on the ABA biosynthetic pathway followed by intensified DR responses in the transgenics.
Self-organizing map (SOM) clustering of DE genes based on their response patterns to the three treatment phases enabled a more comprehensive assessment of RNAi-sensitive pathways than the selective approach above (Fig. 5). The DR and RC response patterns were similar in WT and RNAi plants (small genotype differentials) in six of the clusters (1, 2, 4, 8, 9 and 10). In the other six clusters (3, 5, 6, 7, 11 and 12), the genotype differentials were larger. The differential response clusters were dominated by root DE genes (~60%). About 47% of all root DE genes, compared to 23-32% of DE genes from the other organs, resided in those clusters. Gene ontology (GO) functional enrichment analysis showed that ABA, salicylic acid, jasmonic acid (JA) and ethylene biosynthesis and signaling GO terms were especially well-represented among bark, xylem and root transcripts in cluster 6 ( Fig. 5B), with DR-induced increases in RNAi plants (Fig. 5A). The pattern was different for leaves, where salicylic acid and JA functions were comparatively well-represented in clusters 3 and 7, either suppressed by DR in RNAi, or stimulated by DR in WT. GO categories for cell division, monosaccharide and phenylpropanoid metabolism, response to sucrose stimulus and root development were enriched among DR-suppressed genes in transgenic roots (clusters 3 and 11). Overall, the results were consistent with comparatively strong drought stress sensing as well as reduced growth in transgenic roots, but not in leaves. Bars are means ± SD of n = 3 biological replicates. Significance testing was conducted using the two-sample t-test (*, **, *** represent P ≤ 0.05, 0.01, 0.001, respectively). Transgenic effects are indicated by asterisks: asterisks inside the color bars are for sucrose and hexoses, above the bars are for total sugars (sum of sucrose and hexoses). Treatment effects within each genotype are indicated by dashed lines with P values in the xylem panel. (C) Root sucrose changes in response to DR in an independent experiment. Data are represented by means ± SE (n = 5 WT or 6 RNAi plants). Significant treatment effects were determined using repeated measures 2-way ANOVA. Analysis of major transcription factor (TF) families present in the various SOM clusters revealed their differential sensitivity to RNAi effects (Fig. 6). Overall, there was strong TF representation in cluster 10 ( Fig. 6A), which contained DR-responsive genes that were relatively insensitive to RNAi effects (Fig. 5A). TFs were also well-represented in clusters 6 and 12 that exhibited RNAi-enhanced DR responses. The enrichment of ethylene response factors in these two clusters indicated a magnified sensitivity of ethylene-modulated components of the DR response in bark, xylem and especially roots of RNAi plants (Fig. 6A). Promoter cis element analysis of the SOM gene clusters was also carried out and found to corroborate the above TF observations (Fig. 6B). In general, the strongest representation of stress-related cis-elements was in cluster 10 (increased expression at DR in all organs of both genotypes). However, the classic drought-responsive element as well as W-box and ethylene-responsive element were better represented among cluster 6 than cluster 10 gene promoters, consistent with greater drought sensitivity in RNAi roots. Interestingly, several sugar-responsive elements were well-represented in cluster 10 but not in cluster 6, and one of them (SURE2) was also enriched in clusters 3 and 11 that were down-regulated by DR in transgenic roots. This suggested reduced, or otherwise altered, sucrose-involvement during the DR response of SUT4-RNAi roots.
Gene network connectivity was altered by SUT4 suppression. Based on the high number of DE genes and on SOM clustering and TF analysis, the RNAi effect was most severe in the root tips. Weighted gene correlation networks were therefore constructed for the two genotypes to assess root transcriptome reprogramming. Changes in gene connectivity (RNAi-WT) between the two networks were plotted against changes in expression (RNAi/WT) at DR, and genes were color-coded by co-expression module as determined independently for the two networks (Fig. 7, Supplementary Dataset S2). In the transgenic network, the turquoise module was the largest, with many of its members exhibiting higher connectivity as well as higher expression relative to WT at DR (Fig. 7A, top right quadrant). The most strongly enriched GO categories in the turquoise module pertained to responses to water deprivation, ABA, ethylene and JA stimuli, and oxygen-containing compound   7C). In the WT network, the response to ABA stimulus category was enriched in the comparatively small brown module, along with fatty acid metabolic process and response to water deprivation (Fig. 7C). Fatty acid metabolic process was also co-enriched with phenylpropanoid biosynthetic processes in the WT orange module. The data are consistent with the interpretation that the ABA component of the DR response occurred against a backdrop of greater stress and stronger participation of JA and ethylene components in RNAi roots.
Besides the above-mentioned differences in NCED expression (Fig. 4B), the most striking ABA-related finding was the attenuated DR response of an ABA-stress-ripening homolog in RNAi than WT roots (Supplementary Fig. S5A). This TF is thought to partition between the nucleus and the cytosol for the orchestration of ABA-mediated stress responses in a sugar-dependent manner 26,27 . While DR response of other known effectors of ABA signaling appeared unchanged in RNAi roots ( Supplementary Fig. S5A), expression of ethylene and JA response-related TFs was much stronger in the RNAi roots ( Supplementary Fig. S5B,C).

Discussion
One underlying driver of the comprehensive RNA-Seq experiment was prior observation that SUT4-RNAi poplars exhibited greater reductions of leaf area and height growth during chronic mild drought than WT poplars 14 . Roots were not analyzed in the previous study, so it was unclear whether SUT4-RNAi effects on drought growth originated in the shoots, the roots, or both. The present work aimed to dissect the SUT4-RNAi effects transcriptionally, with supporting data on sugar changes. In addition, the present work utilized an acute drought treatment intended to obtain a change in plant water status without irreversibly damaging the plant, and in a time frame that precluded morphological adaptations that would complicate analysis of the response to changes in water status.

Reduced shoot-root trafficking of sucrose was sensed in SUT4-RNAi plants during DR.
Transcript levels of plasma membrane SUTs can increase both in source and sink organs when demand for sucrose is high, or decrease when it is low 6,8,28 . We therefore examined plasma membrane PtaSUT3 expression in order to determine whether SUT4-RNAi effects on sucrose concentrations perturbed sink-source physiology in a way that affected other SUT gene expression. However, despite large organ-specific increases in sucrose levels in WW RNAi plants, plasma membrane SUT expression was not affected (Supplementary Fig. S1). ). One plausible explanation for this observation is that plasma membrane SUTs are not regarded to mediate sucrose export from leaves of Populus 29,30 . The absence of a negative effect in xylem, a strong sucrose sink, was more surprising. However, because xylem is not a sucrose storage organ, plasma membrane SUT expression in xylem may not be under the same feedback constraints as in sucrose storing tissues such as seed embryos or certain root crops 6,8,28 .
As plants experienced DR, plasma membrane PtaSUT3 expression changed (Fig. 1). The DR-promoted increases were smaller in RNAi than WT stem (bark and xylem), and larger in RNAi than WT roots (Fig. 1). To adopt the feedback regulation model, the ~2-fold SUT3 expression increase in DR-stressed WT stem and root were consistent with increased sucrose demand in both organs during the onset of a water deficit. The comparatively strong DR induction in RNAi roots (~4-fold) may suggest a combination of greater root demand for sucrose and/or impaired delivery of sucrose in RNAi plants. Xylem vacuolar invertase transcripts decreased sharply in both genotypes consistent with DR-promoted changes in sucrose fluxes during transit through the stem toward roots (Fig. 2). Although decreased vacuolar invertase expression would be expected to increase xylem sucrose/hexose ratio, and perhaps vacuolar sucrose storage, sucrose only increased in WT xylem at DR (Fig. 2). A vacuolar sucrose glut apparently preceded DR in RNAi xylem despite reduced photosynthesis in WW RNAi plants 14 . Thus, constitutively high vacuolar sequestration of sucrose in the xylem may have been one contributing factor to reduced root provisioning and increased root demand and SUT3 expression in RNAi plants with increasingly restricted water uptake during soil drying.
AQP expression was consistent with a negative RNAi effect on water trafficking in root tips during DR and RC. The expression of various AQP genes for efficient water distribution throughout the plant is sensitive to ABA, and to osmotic or turgor changes 15,21,31 . AQP expression changes in roots in response to transpiration demand, in leaves for water redistribution associated with specific drought tolerance strategies, and in xylem embolism repair have all been reported in Populus 18,19,21,32 . One interesting observation from the present work was that xylem AQP gene expression increased in the RNAi plants at DR. The normal (WT) DR response for xylem AQP expression appears to be a transcript level decrease (Fig. 3). We might speculate that DR-increased AQP expression caused or reflected an abnormal redistribution of water in RNAi xylem due to inefficient mobilization of vacuolar sucrose. The general effect on cell turgor or intercellular space may be analogous to volume control of intercellular spaces through cell expansion mediated by concerted changes in the expression of PIPs and TIPs in legume nodules to regulate oxygen diffusion 33 .
AQP expression was also altered in roots of the RNAi plants, collectively trending downward in RNAi relative to WT roots at DR and remaining depressed at RC (Fig. 3A). For the plant as a whole, the AQP expression and sugars data from xylem and roots are consistent with the possibility that sugar sequestration in RNAi stems compromised root provisioning and AQP regulation of water trafficking. Compared to WT, correlations of AQP transcript levels between roots and other organs in SUT4-RNAi plants across the WW-DR-RC cycle were reduced. The reduction is consistent with compromised systemic control of water distribution due to localized sugar abundance differentials. Transpiration presumably recovered in WT and RNAi plants during RC as evidenced by restored turgidity of upper sink-source transitional leaves initially used as visual markers of DR onset, so it is not clear why root AQP expression remained low in RNAi plants once watering was resumed. Root expression of many AQPs normally correlates with transpirational demand in Populus 21 . Whether embolism repair, which involves both sucrose trafficking and AQP activity in Populus 18,19 was compromised in the RNAi plants was not investigated here. Our results suggest that sucrose partitioning and spatial accumulation patterns affected AQP expression autonomously in multiple organs, with prolonged consequences to the roots during post-drought recovery.

Interaction of sucrose with hormone-regulated stress responses in poplar root tips.
A possibility suggested by the comparatively strong HSP and LEA transcriptional responses of RNAi roots to DR ( Supplementary Fig. S4) is that RNAi roots were somehow sensitized or more vulnerable to soil drying than WT roots were. Related to this, the enhanced NCED expression in RNAi roots of WW plants (Fig. 4) lent support to the idea that ABA homeostasis was altered prior to DR stress due to reduced tonoplast trafficking of sucrose and/ or reduced root: shoot sucrose concentration ratios. The strong LEA response in RNAi roots is consistent with the notion of a stronger or otherwise perturbed ABA response during changes in soil moisture availability. All 27 differentially expressed LEA genes identified in our study ( Supplementary Fig. S4A) contain multiple ABA response elements in their promoters. At the same time, sugar-response cis elements were under-represented compared to drought-response elements in promoters of genes that were up-regulated by DR specifically in RNAi roots (Figs 5 and 6, cluster 6). ARS1 which is thought to facilitate cross-talk between ABA and glucose 26,27 exhibited a severely compromised DR response in RNAi roots. We suggest that reduced tonoplast trafficking of sucrose stimulated NCED expression regardless of stress level, while restricting long-distance transport of sucrose to roots. As soil dried, intercellular water in root tips would presumably be increasingly drawn toward tissues of greater osmolyte concentration either in shoots or in more vascularized roots, with shoot-to-root tip osmolyte gradients potentially steeper and more damaging in RNAi than WT plants.
Interestingly, transcript levels of several VSP/BSP genes were elevated in RNAi compared to WT root tips prior to, and then in an enhanced way, during DR stress (Supplementary Fig. S4). In Populus, BSP/VSP proteins accrue as storage reserves in response to stress and to seasonal changes in temperature and light 34 . However, BSP/ VSP is also an important target of the wound signal JA in Populus and other species 35 . In addition, mechanical strain caused by root bending has been reported to elevate the abundance of BSPs and HSPs in poplar roots 36 . Other causes of mechanical strain include changes in turgor and osmotic status 37 . The pre-stress abundance of BSP/VSP transcripts, and their sustained high levels in RNAi roots during DR and RC may therefore be consistent with more active re-allocation of nitrogen and carbon reserves in response to multiple signaling pathways including those associated with latent mechanical stress.
During DR, TFs associated with ethylene and JA signaling increased more in root tips of RNAi than WT plants ( Supplementary Fig. S5B,C). The interplay of sugars, ABA, ethylene and JA signaling during drought is extremely complex, and ethylene and JA frequently are antagonistic with ABA [38][39][40] . The relationship between abiotic stress, phytohormones and SUTs has also recently been explored using SUT mutants and exogenous supplements in Arabidopsis 13 . In citrus roots, extreme soil drying leads to a JA response which subsequently gives way to an ABA-mediated response 41 . In the case of the SUT4-RNAi roots, therefore, up-regulation of JAZ transcripts associated with feedback mitigation of the JA response 42 may reflect the residual of a stronger DR-induced JA response than the one generated in WT roots. The apparently heightened response may also reflect altered ABA homeostasis prior to DR treatment in RNAi root tips. Together, the cis-element analysis and the root gene network data suggest that sucrose played a larger role in the WT drought response, perhaps supporting a biosynthetic response commensurate with adaptation. In contrast, there was a stronger integration of the drought response in RNAi plants with generation of oxygen-containing compounds including reactive oxygen species, and with ethylene and JA mediation

Conclusions
The data as a whole are consistent with the idea that a latent strain existed in SUT4-RNAi root tips due to altered homeostatic control of sugar osmolyte distribution. Based on altered leaf-to-root transcript level gradients of NCED (Fig. 4), ABA homeostasis may have been perturbed with possible effects on the magnitude of ethylene and JA responses during soil drying (Supplementary Fig. S5). In previous work, height growth of two independent SUT4-RNAi lines was reduced under chronic mild drought conditions more so than the WT plants 14 . Leaf area was also reduced in both SUT4-RNAi lines compared to WT. In light of those findings, the transcriptome results from the present study lend further support to the idea that tonoplast trafficking of sucrose is consequential for the coupling of growth with differences in soil water availability. A normal adaptation to chronic water limitation in the WT poplars appears to be a substantial reduction in leaf SUT4 expression 14 . In addition to the above-posited idea of latent root strain, a second point may be that SUT4 expression is necessarily dynamic in order to couple carbon use with changes in plant water status, and that constitutively reduced expression may have mitigated SUT4 function by limiting the magnitude of such expression changes.

Methods
Plant growth and experimental treatments. Single-node cuttings were grown in 4-gal Treepots ™ , containing Fafard 3B potting mix supplemented with Osmocote (15-9-12 NPK 4-month release), to approximately 1.5 m in a greenhouse (August to October) under evaporative cooling and supplemental lighting (14 hr photoperiod). The acute drought treatment was conducted as described 14 . Briefly, water was withheld about three days until visible-yet-reversible turgor-loss of sink-source transitional leaves, leaf plastochron index 43 3-5. Source leaves exhibited normal turgor at DR. WW controls and DR plants were harvested at the same time, approximately 60 hrs after the commencement of soil drying. Unharvested, DR-treated plants were then watered and harvested 48 hr later (RC). Sampling occurred under sunny conditions (10 AM to 2 PM). Source leaves (leaf plastochron index 15), phloem with bark, xylem shavings (internodes 20-30), and distal 2 cm of root tips were snap-frozen in liquid nitrogen and stored at − 80 °C.
Analysis of sucrose, glucose and fructose. Freeze-dried powders (10 mg) were extracted twice with 700 μ l methanol: water: chloroform (40:27:33, v/v) containing internal standards (adonitol and 2-methoxybenzoic acid). The aqueous phase was derivatized with N-methyl-N-(trimethylsilyl) trifluoroacetamide and analyzed on an Agilent 7890A GC coupled to an Agilent 5975 MS as detailed previously 14 . Spectra were collected in the scanning ion mode (m/z 50-500) in ChemStation (Agilent Technologies) and deconvoluted using AnalyzerPro (SpectralWorks). Retention times and mass spectra of the abundant sugars sucrose, glucose and fructose were confirmed using authentic standards. Run-to-run uniformity was monitored using a standard mixture of sucrose, glucose and fructose. Relative peak areas of each metabolite (normalized against internal standard adonitol) were used for all statistical analyses by SigmaStat 3.1.
High-throughput sequencing and data analysis. Genomic DNA was extracted from young leaves of three SUT4-RNAi transgenic lines using the E.Z.N.A. Plant DNA kit (Omega Bio-tek), quality-checked by agarose electrophoresis and quantified using a Qubit fluorometer (Life Technologies). Genomic libraries were prepared as described 44 and sequenced on an Illumina NextSeq-500 at the Georgia Genomics Facility. Approximately 12-32 million (M) paired-end 75-bp reads were obtained per sample. To identify T-DNA insertion sites, individual reads were mapped to the vector as well as the Populus tremula x alba custom genome v1.1 45 using BLAT. Custom Perl scripts were used to filter for paired reads flanking an insertion site, e.g., one read of the pair mapped to the T-DNA and the other to the genome.
RNA was isolated using the CTAB method 46 and quality-checked using a BioAnalyzer (Agilent). RNA-Seq library preparation and Illumina HiSeq-2000 sequencing were performed at BGI America. Approximately 9-17M paired-end 50-bp reads were obtained for three biological replicates (two genotypes, three treatments and four tissues). The RNA-Seq data have been submitted to NCBI Sequence Read Archive (accession number SRP041959). Filtered read pairs with adapter and rRNA sequences removed were mapped onto the Populus genome v3.0 using Tophat2 v2.0.6, and processed by Cufflinks (v2.0.2) for transcript abundance estimation and pairwise DE analysis 47 . The significance threshold was set at fold-change ≥ 1.5 and Q value ≤ 0.05. Genes with high coefficient of variation (> 1.3) among replicates were excluded from further analysis. For self-organizing-map clustering, DE genes with FPKM ≥ 3 in at least three individual samples per tissue (two genotypes x three treatment phases x three replicates) were extracted for each tissue and then combined into one dataset for the analysis using MEV v4.9 48 . Functional annotation of signaling and metabolic pathways was obtained from KEGG (http://www. genome.jp/kegg/pathway.html) and PoplarCyc v3.0 (http://plantcyc.org), respectively, and transcription factors from the Plant Transcription Factor Database 49 . Gene ontology (GO) enrichment analysis was performed using topGO 50 with Fisher's exact test, and the negative log10 transformed P values were visualized using heatmaps as described 51 .
Scientific RepoRts | 6:33655 | DOI: 10.1038/srep33655 Cis-element enrichment analysis. Cis-element analysis was performed as described 52 . Briefly, 1,032 plant cis-elements were collected from PlantCare 53 , Transfac 54 , JASPAR 55 and PLACE 56 . Redundant cis-elements identified by MotifComparison 57 with a distance score smaller than 0.3 were removed, leaving 483 cis-elements for analysis. MotifScanner 57 was used to map the cis-elements onto the 2-kb upstream promoter sequences extracted from the JGI Populus genome v3. Distribution of cis-elements was summarized for all genes as a baseline for Chi-square test of cis-element enrichment in the various gene clusters. The resulting P values were corrected for false discovery rate 58 , and the negative log10 transformed values were visualized using heatmaps.
Co-expression gene network construction and visualization for root tips. The WGCNA (v1.29) package in R 59 was used to construct a co-expression network for all root DE genes, and then for two sub-networks (WT and SUT4-RNAi) using an established pipeline 51 . Briefly, 8,655 DE genes across all root samples were obtained with ANOVA P ≤ 0.05 and fold-change ≥ 2. Pearson correlation coefficients were calculated for all gene pairs and an adjacency matrix was constructed using a power of 14. The degree distributions of the networks all followed the power law, indicating that they were scale-free. A hierarchical tree based on a topological overlap matrix was then constructed to identify discrete co-expression modules using the dynamic tree cut method 59 , with the minimum module size set to 140 genes. GO enrichment analysis of co-expression modules was conducted as described above.