Time-resolved Interaction Proteomics of the Putative Scaffold Protein GIGANTEA in Arabidopsis thaliana

The large, plant-specific protein GIGANTEA (GI) is involved in many physiological processes, mediating rhythmic, post-translational regulation in part through circadian and light regulation of GI RNA expression. GI binds several proteins implicated in the circadian clock, the control of photoperiodic flowering, and abiotic stress responses, and has co-chaperone activity. By extension, further interaction partners might mediate the less well-understood roles of GI but the number and rhythmicity of these interactors is unknown. Here, we seek potential interactors in a time-specific manner, using quantitative proteomics from a time series study of transgenic Arabidopsis thaliana plants constitutively expressing an epitope-tagged GI protein. Previously-identified, direct and indirect interactors of GI were detected but no further F-box proteins related to known GI partners ZTL/FKF1/LKP2. The predominantly non-rhythmic, interacting proteins were implicated in protein folding or degradation, metabolism and chromatin modification, including a small set of partners shared with other clock-related proteins. A transcription factor homologue that we name CYCLING DOF FACTOR 6 (CDF6) was shown to interact both with GI and the ZTL/FKF1/LKP2 proteins and to control photoperiodic flowering. Our results indicate the biochemical pathways, beyond circadian and flowering regulation, that might be affected by GIGANTEA’s rhythmic, post-translational control. Significance Statement Significance statement of up to two sentences of no more than 75 words total; The GIGANTEA protein of Arabidopsis was known for circadian and flowering functions, mediated by the FKF1/LKP2/ZTL family of GI-interacting, F-box proteins, then for a co-chaperone activity of unknown scope. We performed time-resolved, interaction proteomics, identifying CDF6 (At1g26790) as a morning-specific GI interactor that controls flowering time. Unlike FKF1 and CDF proteins, most of the 240 candidate partners were not rhythmically enriched. They link GI to proteostasis and metabolic functions that might mediate GI’s physiological functions.


Introduction
Arabidopsis thaliana has well-documented, 24-hour rhythms in many physiological processes, from the elongation rate of seedling hypocotyls to photosynthetic functions and the defense of mature leaves against herbivore attack (Dodd et al. 2014, Millar 2016. The overt, circadian rhythms are thought to be driven by a transcriptional-translational feedback circuit (Millar 2016). Detailed, dynamic models based mostly upon transcriptional repression recapitulate the rhythmic expression profiles of these so-called clock genes, including manipulations of the system in mutant plants and under changing photoperiods (Flis et al. 2015, Flis et al. 2016, Pokhilko et al. 2012. Gene expression switches can operate on a timescale of minutes, however, which does not obviously explain the slow, 24-hour timescale of the circadian clock. Chromatin modification might extend the timescale of transcriptional regulation (Barneche et al. 2014). A further, critical factor is the slow degradation rate of the transcriptional repressors, which mathematical analysis has long identified and experiments confirm as key to timing (Gerard et al. 2009, Ruoff et al. 2005. Regulated protein degradation is also crucial to plant circadian 7 GI's role in the clock is mediated at the biochemical level by co-chaperone activity, which involves binding to HSP90 and appears to stabilize ZTL (Kim et al, 2011;Cha et al, 2017).
This activity can affect other test substrates but its other native targets, if any, are unknown.
As outlined above, GI's known functions with the ZTL family and SOS2 are mediated by protein-protein interaction. Therefore, GI has been suggested to serve as a scaffold that orchestrates other protein interactions (Imaizumi & Kay, 2006), for example to provide chaperone activity (Cha et al. 2017).
Although such protein interactions are thought to mediate GI's functions, these interactions have not been broadly tested. We therefore conducted interaction proteomics assays and obtained time-resolved data on potential direct and indirect partners of GI, over the daily timecourse. Here we discuss the abundance profiles and functions of candidate interactors and highlight a DOF protein, CDF6, validating its direct interactions and functional importance. 8

Characterisation of a GI:3F6H line
We transformed Arabidopsis thaliana plants carrying the strong gi-2 mutation (a deletion allele predicted to truncate ~90% of GI protein, Fowler et al. 1999) with a construct to express 3xFlag-, 6xHis-tagged GI protein under the control of the CaMV 35S promoter (GI:3F6H). After isolating several positive transformants, we characterised one of the lines that expressed sufficient GI protein for tandem affinity purification (TAP) in a preliminary study ( Figure S1). Analysis of specific gel bands by mass spectrometry (MS) identified GI and known interactors (Data S1). In addition, we tested expression of relevant genes in seedlings of this transgenic line using qPCR. The GI transcript was expressed constitutively, close to wild-type (WT) peak levels in the GI:3F6H line (Figure 1a). During the day, CO and FT mRNAs were higher in the GI:3F6H line than in the WT (Figures 1b, 1c), consistent with activation of these flowering-promoting genes by GI in the light (Sawa & Kay, 2011;Sawa et al, 2007). This was also reflected by slightly early flowering of the GI:3F6H plants relative to the WT, clearly rescuing the late flowering of the gi-2 mutant parent ( Figure 1d).
Simulating the rescued mutant line in a mechanistic, mathematical model that includes photoperiodic flowering (Chew et al. 2017) predicted both molecular and flowering phenotypes (Figures 1d, 1e). The WT and gi-2 simulations closely matched the RNA data, though this data set favoured morning (1-4h after dawn) expression of CO and FT compared to evening (13-16h) expression slightly more than the model, possibly reflecting a reduced GI RNA level at 13h. The model predicted elevated CO mRNA level at 1h in GI:3F6H plants but the observed level was even higher (Figure 1b). This CO peak induced more FT at 4h (Figure 1c) and earlier flowering in the data than the model.

9
The potential to identify further, GI-interacting proteins was tested in a qualitative proteomics study. Affinity purification of GI:3F6H (GI-TAP) was carried out with two different buffers (RIPA or SII) on GI:3F6H extract and a WT control for each buffer, followed by digestion and MS analysis (Figure 2b). 50 Arabidopsis proteins were identified by at least one peptide in each of the GI:3F6H samples and none in the controls (Table 1; full results in data S2).
Detection of known direct interactors FKF1, ZTL and LKP2 validated the methodology, as well as indirect interactors ASK1A and ASK2, which are direct interactors of ZTL and FKF1 (Han et al, 2004;Kevei et al, 2006;Takahashi et al, 2004;Sawa et al, 2007). Among the remaining proteins on the list, half were abundant, ribosomal or cytoskeletal proteins or proteins presumably inaccessible to GI due to their annotated, chloroplast or mitochondrial location (but see Discussion). The remaining 17 proteins indicate the potential to identify previously undiscovered interactions, as discussed below.

Candidate GI-interacting proteins from timeseries data
In order to obtain time-resolved interaction data, we applied the same GI-TAP method to plants grown in short-day conditions, at 6 timepoints in biological quintuplicate, with additional duplicate samples at time point 31h (replicating the 7h timepoint; Figure 2a).
Extraction, TAP and sample preparation for mass spectrometry was carried out as for the qualitative analysis, using SII buffer (Figure 2b). In order to preserve low-abundance and indirect interactors, the experimental protocol did not remove all proteins that were bound non-specifically. Using the Mascot search engine to identify peptides, our choice of peptide score cutoff of 20 resulted in an FDR of 0.023. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository (Vizcaino et al. 2016) with the dataset identifier PXD006859 and DOI 10.6019/ PXD006859. After identification and quantification of proteins (Figure 2c), one technical outlier was excluded from subsequent analysis (see Experimental Procedures; Figure S2). PCA of the remaining GI samples maximally separated the mid/late-night timepoints 19h and 23h from mid-day timepoints 7h and its replicate 31h (Figure 2e). 2336 peptides were detected in the time series study, from which 498 proteins were quantified ( Table 2). The analytical methods also quantified the identified peptide peaks in WT samples, in contrast to the qualitative study. Therefore, the fold-enrichment of each protein in the GI-TAP was calculated relative to the average abundance in the control samples. Potential interacting proteins were identified as significantly enriched by t-test compared to the WT (excluding 8 proteins detected in only 1-2 WT samples, including RD21B, At5g43060), with a significance threshold adjusted for multiple testing (BH adjusted q-value < 0.05). Fold-enrichment threshold values were informed by the results for known interactors. The direct interactors FKF1, ZTL and LKP2 were more than ten-fold enriched over the control in the GI-TAP timeseries. Indirect interactors CUL1/CUL2 and GLN2 (Han et al, 2004;Song et al, 2014) were two-to three-fold more abundant at their peaks than the timeseries control and were not identified in the preliminary or qualitative studies. Hereafter we refer to significantly-enriched proteins with at least four-fold enrichment as highly enriched (136 proteins) and to proteins with two-to four-fold enrichment as weakly enriched (a further 104 proteins).
Among these 240 proteins, 25 changed significantly in abundance within the time series (assessed for any change by ANOVA) or 18 as assessed for rhythmic profiles (by JTK_CYCLE; Table 2 and data S3). Peak abundance for most of the 35 proteins with changing or rhythmic abundance occurred at 7h, or 24h later at 31h, with lowest abundance at 23h (Figure 2d). The abundance of the constitutively-expressed GI changed about two-fold over time (Figure 3a). The observed abundance changes for GI and ZTL closely matched the predicted protein profiles (Figure 3d) from simulation of GI:3F6H (as in Figure 1e). The change in GI abundance was not significant by ANOVA or JTK_cycle. Subsequent analysis used raw abundance rather than normalizing to remove the change in GI abundance; analysis of normalized data gave very similar results (data not shown and data S2). Our enriched protein sets should prioritise candidates for additional, direct or indirect interaction partners of GI, with greater scope than the smaller-scale studies.

Functional categorization of GI-TAP enriched proteins
According to a gene ontology (GO) search using the Panther tool (http://pantherdb.org, Thomas et al, 2003), the dataset contains 166 plastid proteins, of which 101 were weakly-or highly-enriched (data not shown). More detailed GO analysis in the topGO R package (Alexa et al, 2006), using all enriched proteins as foreground and all other identified proteins as background, revealed significant over-representation of plastid related terms (data S4, Figure   S 3, appendix S1). GI protein is thought to be predominantly nuclear and cytosolic, so the analysis was repeated after excluding proteins annotated with predominantly chloroplast or mitochondrial localization. Abundant, cytoskeletal proteins were also excluded as likely nonspecific interactors. The 35 remaining candidates are listed in Table 3, along with 3 related proteins. GO terms related to protein degradation were over-represented among the candidates (data S3), along with remaining, abundant protein classes such as glycolytic enzymes. The most significantly enriched term after 'regulation of circadian rhythm' was 'ubiquitin-dependent protein catabolic process' (GO:0006511; 7 enriched out of 14 detected proteins; p=0.0016).

Rhythmic profiles of known (in)direct interactors
In contrast to the weakly rhythmic trend in abundance of the GI protein, known interactors showed contrasting profiles (Figure 3a). The direct interactors FKF1, ZTL and LKP2 showed temporal profiles consistent with their RNA expression patterns. Only FKF1 was strongly rhythmic, with a peak at 7-11h, resembling previous data (Fornara et al. 2009, Imaizumi et al. 2003. The ZTL profile paralleled GI, consistent with their mutual stabilization (Kim et al, whereas ZTL and LKP2 also interact with the core components of the SCF ubiquitin E3 ligase detected here, ASK1 and CUL1 and/or CUL2 (closely-related proteins that were not distinguished by the peptides detected). Of these, only CDF3 was scored rhythmic, with a peak at 27h and background levels between 11h and 19h. Therefore CDF3 level is in antiphase with FKF1, in line with the degradation of CDFs by FKF1 (Fornara et al. 2009, Imaizumi et al. 2005. These results demonstrate the consistency of our data with published results. CDF3 was quantified using a single, individually-inspected peptide, indicating that such data should not be excluded from analysis.

The predicted functions of interactors include protein degradation and stabilization
In addition to verifying the known indirect interactor CUL1, our analysis enriched other proteins involved in protein stability (Figure 3c). The stress-related proteases RD21A and RD21B were enriched in the timeseries and qualitative studies, as were ubiquitin-specific proteases (UBP) 12 and 13 (AT5G06600 and AT3G11910), AAA-type ATPase family 13 proteins related to components of the 26S proteasome (At1g45000 and/or At4g27680) and a protease inhibitor, cystatin 1 (At5g12140). Several proteins annotated as being involved in protein stabilization were also significantly enriched by GI-TAP, such as Cpn60 chaperonin family proteins (At1g24510, At3g18190, At3g02530 and At3g03960) and at least one Hsp90 heat shock protein (At5g56030, At5g52640 and/or At5g56000, Figure 3c). Hsp70 family proteins were just below the enrichment cutoff (Table 3).
In contrast, neither additional F-box proteins nor other proteins involved in circadian timekeeping were identified as strong candidate interactors. The chloroplast RNA binding (CRB, At1g09340; Figure S3) is known to affect the amplitude of circadian clock and clock output genes (Hassidim et al, 2007) but the relevance of interaction with GI is unclear, given its localization (see Discussion). Multiple, metabolic enzymes and translation elongation or initiation factors were enriched in the GI-TAP timeseries. Among those that were not exclusively plastid localized (Table 3), were GTP-binding translation factors (At1g56070; At1g07920; At1g72730 and At1g54270 and/or At3g19760), two carbonic anhydrases (At1g70410 and At5g14740; Figure 3c and data S3), Phosphoglucomutase 2 and/or 3 (At1g23190, At1g70730; data S3), trehalose-6-phosphatase synthase 8 (TPS8; Figure 3c), a phosphofructokinase family protein (At1g20950), a GMP synthase homologue (At1g63660), a pyruvate kinase family protein (At2g36580) and 3 to 5 proteins of the methionine cycle, involved in regenerating S-adenosyl methionine (At3g03780; At2g36880/At1g02500; At4g13940/At3g23810) for methyltransferase reactions. At3g63410, At4g25080 and At3g63410 were the only enriched methyltransferases but all are plastid or mitochondriallocalised (Data S3). The candidate interactors suggest new clients and mechanisms of GI action related to those in other species (see Discussion), though their physiological significance awaits confirmation.

CDF6 is a GI interactor that contributes to photoperiodic flowering
At1g26790 encodes a predicted DOF transcription factor that was up to 15-fold enriched around dawn (23h and 27h; Figure 4a). This protein was the most significantly rhythmic of the candidate interactors after FKF1 and CDF3 (BH-adjusted p value from JTK_cycle = 8 x 10 -6 ). Its levels were the most anti-correlated with FKF1 levels among the highly-enriched proteins, followed by CDF3 (r = -0.65 and -0.40, respectively). We therefore validated the interaction and function of this protein, hereafter termed CYCLING DOF FACTOR 6 (CDF6). Yeast To test the potential function of CDF6 in the context of other members of the CDF gene family, transgenic plant lines were generated, in which CDF6 was expressed from the SUC2 promoter that is active in phloem companion cells (Song et al, 2012). Two lines, pSUC2_HA:CDF6 #8 and #11, over-expressed the CDF6 transcript at ZT4 and in later timepoints of a qPCR timeseries (Figure 4e), fluctuating around 20-60% of the WT peak level, whereas CDF6 levels in WT were very low except at the ZT1 peak. Both transgenic lines flowered significantly later than WT under long photoperiods, with less effect under short photoperiods (Figure 4d). If CDF6 acts in a similar way to CDF1 (Song et al, 2012), we would expect it to inhibit transcription of both CO and FT. Indeed, CO RNA levels were reduced at 10h, 13h and at night in the transgenic plants compared to WT, and FT expression was reduced more than ten-fold at 4h and in later timepoints ( Figure 4d). These results are consistent with CDF6 participating in the photoperiodic regulation of flowering, where CDF6 protein levels in WT are regulated through interaction with GI and its partners.

Discussion
Proteostasis, the set of protein-metabolic processes, is expected to be critical for diel rhythms in general, because the removal of transcriptional repressor proteins controls the slow timing of circadian feedback circuits (Gerard et al. 2009, Ruoff et al. 2005. GIGANTEA not only indirectly mediates the degradation of transcriptional repressors (Kiba et al. 2007, Mas et al. 2003, Sawa et al. 2007a) but also has protein chaperone functions (Kim et al, 2011;Cha et al, 2017). Our studies identified further proteostatic proteins associated with GI, and suggested links to metabolic sensing, providing candidates for the unknown targets of GI's proteostatic functions (Cha et al, 2017) and recalling previous data linking GI, metabolic inputs and biological timing.

Technical requirements, known interactors and CDF6
Constitutive expression of tagged GI protein (GI:3FH6) in the gi-2 mutant background rescued the mRNA expression and flowering time phenotypes of the mutant. GI protein tended to greater abundance during the day than during the night, in line with its light- Conversely, the large size and proposed proteostasis functions of GI (discussed below) risk falsepositive results. GI has not been found localized in or associated with the chloroplast but rather in the nucleus or cytoplasm (Huq et al, 2000;Kim et al, 2013a;Mizoguchi et al, 2005).
The abundant, plastid-localised proteins enriched as interactors ( Figure S3; Appendix S1) likely reflect unspecific binding, at least in the case of chloroplast-encoded proteins, which was an expected cost of detecting low-abundance and indirect interactors. Conservatively, we excluded cytoskeletal, ribosomal, mitochondrial and chloroplast proteins (see Methods; Data S3) from the candidate interactors (Tables 1 and 3

Metabolic and nuclear functions of GI
The mechanisms that allow GI to link carbon metabolism and timing, via a long-term response of the circadian clock to sucrose (Dalchau et al, 2011) and the photoperiodic adjustment of the rate of starch biosynthesis (Mugford et al, 2014). The trehalose-6phosphate (T6P) pathway mediates several such sugar responses (Ramon et al, 2009;Wahl et al. 2013). TREHALOSE PHOSPHATE SYNTHASE 8 (TPS8) is a paralogue without known enzymatic activity but with diurnally-regulated expression, repressed by sucrose (Usadel et al, 2008). GI interaction with TPS8 was highly enriched and, unusually, peaked at ZT19 (Figure 3c), providing one of several possible mechanisms for GI to mediate between metabolism and biological timing. Strikingly, at least one of each enzyme in the cycle that regenerates the methyl donor S-adenosyl methionine (SAM; Figure S4) was enriched in the timeseries dataset, with 5-7 out of all 9 isoforms identified across our datasets (Table 3). The plastid-localised VTE3 was one of three enriched methyltransferases (Data S3). Interestingly in view of the Paraquat-sensitivity of gi mutants, VTE3 is expected to contribute via tocopherol synthesis to tolerance of reactive oxygen species, such as those generated by

GI interactors involved in protein metabolism
Protein degradation of clock-relevant, transcriptional repressors was the first biochemical function supported for GI, acting as a scaffold for F-box proteins, though GI's co-chaperone function is now also implicated (Cha et al, 2017;Kim et al, 2007). No further F-box proteins or other ubiquitin E3 ligases were identified here, suggesting that GI mediates further physiological roles through different biochemical mechanisms. Involvement in translational regulation was suggested by the enrichment of translation-associated proteins, including the LOS1 protein (Tables 1, 3) implicated in cold stress (Guo et al. 2002). Chaperone proteins are typical, non-specific contaminants of affinity purification studies but direct, physiologically-relevant binding of HSP90 with GI has been demonstrated (Cha et al. 2017).
HSP90 isoform(s) were highly enriched and weakly rhythmic in our timeseries (Figure 3c).
TPR4, which encodes a tetratricopeptide repeat (TPR) protein with potential to interact with Hsp90/Hsp70 as a co-chaperone (Prasad et al. 2010) was also strongly enriched (Table 3).
The Hsp70 family proteins that might function with GI and HSP90 (Cha et al. 2017) were below the significance threshold in the timeseries study (Table 3) but one (At5g02500) was the fourth most-enriched protein in the qualitative study (Table 1).
In contrast, several other proteins involved in proteostasis were highly and reproducibly enriched. TCP-1/cpn60 chaperonin family proteins (At3g03960; At3g20050) that can facilitate intercellular trafficking of transcription factors (Xu et al. 2011) were detected in the timeseries and qualitative studies (Tables 1, 3). Granulin-repeat cysteine proteases RD21A and RD21B (Figure 3c) were also repeatedly detected. These are among the major protease 20 activities of the leaf and are inducible by drought stress and senescence (Koizumi et al, 1993;Gu et al. 2012). Their functional relationship to the other stress-related GI interactors SOS2 and 14-3-3λ is unclear. RD21A was previously identified as an interactor of three transcription factors (Huang et al. 2016a, Huang et al. 2016b, Wang et al. 2013, two of which interact with 14-3-3λ, and also with the final, proteostatic interactors of GI, the ubiquitin-specific proteases UBP12 and UBP13.

Generation of plant materials
To generate plants with epitope-tagged GI protein, gi-2 mutant Arabidopsis thaliana plants were transformed with a construct expressing C-terminally 3xFLAG-6His tagged GIGANTEA protein (GI:3F6H). The full-length GI cDNA without the stop codon was amplified and inserted into pENTR/D-TOPO vector (Invitrogen). After sequence verification, the GI cDNA was transferred into the pB7HFC vector, designed for in-frame epitope fusion (Huang et al., 2016) by a Gateway cloning reaction (Invitrogen). pB7HFC-GI-3F6H was introduced to the gi-2 mutant by Agrobacterium-mediated transformation. Transgenic plants that rescued the gi-2 phenotype were selected, and the expression of the GI:3F6H transgene was verified by Western blotting (as in Figure S1; Methods S1). Samples for the preliminary and qualitative GI-TAP studies were prepared as described in (Song et al., 2014).
After cold-treatment at 4°C for 5 days, seeds were grown on plates (2.15g/l Murashige & Skoog medium Basal Salt Mixture (Duchefa Biochemie), pH 5.8 (adjusted with KOH) and 12g/l agar (Sigma)) in Percival incubators (CLF Climatics) for 17 days at 85µmol m −2 s −1 (full-spectrum white fluorescent bulbs) and 21°C in short day conditions (8h light). Seedlings were transferred to soil, for 20 days in the same conditions with a light intensity of 110 µmol m −2 s −1 . Starting at 7h after dawn, 80 rosettes without roots were harvested for each sample, at timepoints and replicates shown in Figure 2a and flash-frozen in liquid nitrogen. Dim green safelight was used during darkness.

Quantitative PCR (qPCR) analysis
Fresh seedlings were ground into powder with a mortar and pestle with liquid nitrogen, and total RNA was isolated by using an illustra RNAspin Mini kit (GE Healthcare) according to the manufacturer's instructions. Two µg of total RNA was reverse-transcribed using the iScript cDNA synthesis kit (Bio-Rad) according to manufacturer's instructions. cDNA was diluted to 5 times with water, and 2 μl was used as a template for quantitative PCR (qPCR) analysis using primers shown in Table S1. Isopentenyl pyrophosphate/dimethylallyl pyrophosphate isomerase (IPP2) was used as an internal control for normalization. The average value from WT was set to 1.0 to calculate the relative expression of other lines. To amplify CO and CDF6, three-step PCR cycling program was used: 1 min at 95 °C, followed

Protein Extraction and tandem affinity purification (TAP) procedure
All steps in the protein extraction, TAP and preparation for mass spectrometry were carried out in random sample order in order to avoid bias due to order of processing.
Frozen plant tissue was ground to a fine powder in a liquid nitrogen and dry ice-cooled mortar and processed essentially as described (Song et al., 2014). Detailed procedures are described in Supporting Experimental Procedures (Methods S2).

Protein digestion and mass spectrometric analysis
Preparation of samples for mass spectrometry for the qualitative and timeseries studies analysis used an on-bead digest, prior to mass spectrometric analysis. Detailed procedures are described in Supporting Experimental Procedures (Methods S3).

Proteomics data analysis and bioinformatics
For the qualitative study, database searches were performed using Comet (Eng et al. 2013), searching against the Uniprot Arabidopsis protein sequence database, and using Percolator (Matrix Science, Boston, USA) with a q-value cutoff of 0.01. Cysteine residue masses were considered statically modified by iodoacetamide, and methionine dynamically modified by a single oxidation. Precursor mass tolerance was 10 ppm, and product ion tolerance 0.5 Da.
The principle of parsimony was used for protein inference, and at least two unique peptides were required for each identified protein.
The timeseries data were analysed using the commercial Progenesis LC-MS software implemented by an R script (data S5). The average Pearson correlation coefficient of each GI-TAP replicate with the other replicates of the same time point was above 95% for all GI:3F6H samples apart from sample 19E ( Figure S2), which was also clearly separated from all other samples by PCA and was therefore discarded.
A custom R script performed further statistical analysis and plotting (data S7). We used a ttest to determine for each protein, whether the maximum GI-TAP time point (omitting the 31h time point) is significantly different from the WT control average using q-values

Yeast-2-Hybrid assay
Full-length CDF6 coding sequence was PCR-amplified using cDNA as template with primers shown in Table S1, cloned into pENTR D-TOPO (Invitrogen) and sequence-verified. The plasmid cassette was transferred to pAS-GW, a gateway compatible bait vector (Nakayama et al. 2002) using LR clonase II (Invitrogen). The GI-N, GI-C, FKF1, LKP2, and ZTL clones used in this analysis were described previously; GI-N andGI-C (Sawa et al. 2007b), andFKF1, LKP2, andZTL (Imaizumi et al. 2005). Yeast strains Y187 and AH109 were transformed with prey and bait vectors, respectively using the standard yeast transformation protocol (Clontech). After colonies formed on -W or -L containing media, three independent colonies were grown together, and then mated against their corresponding pairs for 3 days on YPDA media. After mating, yeast colonies were transferred onto -WL media. After checking for mating confirmation, yeast sectors were retransferred at the same time onto -WL and -WLH media. The experiments were repeated several times with the same results.
Rural Development Administration, Republic of Korea) to T.I. The authors declare no conflicting interests. Table S1: Oligonucleotide primer sequences. Figure S1: Validation of the GI-TAP procedure in the preliminary study, showing the gel bands selected for mass spectrometric anlaysis.

Supporting Experimental Procedures
Methods S1. Verification of GI:3F6H expression by Western Blotting and Silver-stain gel followed by in-gel digestion.

Methods S2. Protein extraction and enrichment by TAP for mass spectrometric analysis.
Methods S3. Protein digestion, peptide and mass spectrometric analysis.
Overview of supporting data files [Note: .zip files that cannot be uploaded will be provided online upon publication]

Data file name
File contents Data S1.xlsx Proteins identified by LC-MS analysis of bands excised from silver-stained gel after GI-TAP (Preliminary study) Data S2.  Control (WT tissue) and GI-TAP samples were extracted in RIPA or SII buffer. 25 Arabidopsis proteins were identified by at least one peptide in each GI-TAP sample and none in the controls, excluding likely contaminant proteins that are abundant (ribosome, cytoskeleton) or localized to other compartments than GI (chloroplast, mitochondria). The right-hand columns cross-reference the timeseries study (Table 3), with fold-enrichment and significance (q-value) of the maximum GI-TAP time point relative to the WT control. Bold: known direct or indirect interactors and homologues. *: below threshold in timeseries study. n.d., not detected.

Supporting experimental procedures
Methods S1. Verification of GI:3F6H expression by Western Blotting and Silver-stain gel followed by in-gel digestion Western blot analysis shown in Figure S1 was performed as described (Song et al., 2014), using a commercial 4-12% Bis-Tris PAGE (Life Technologies) and the iBlot system according to the manufacturer's instructions. The M2 mouse anti-Flag antibody (Sigma) and the IRDYE® 800 CW goat anti mouse secondary antibody (Licor) were used for detection of anti-Flag signal on the membrane using a Licor Odyssey fluorescence scanner. For silver staining, the SilverQuestTM silver staining kit (Life Technologies) was used according to the manufacturer's instructions. Mass spectrometry results of in-gel digestion of specific bands are reported in data S1.

Methods S2. Protein extraction and enrichment by TAP for mass spectrometric analysis
Frozen plant tissue was ground to a fine powder in a liquid nitrogen and dry ice-cooled mortar and processed essentially as described (Song et al., 2014). For protein extraction, one tissue volume of SII buffer (100mM sodium phosphate pH7.4, 150mM KCl, 5mM EDTA, 5mM EGTA, 0.1% TritonX-100) or RIPA buffer (50mM Tris pH 7.5, 150mM KCl, 1% NP-40, 0.5% Dexoycholate) with 1 Complete protease inhibitor Cocktail EDTA free mini tablet (Roche) per 10ml, PhosStop phosphatase inhibitor mixture (Roche), 50µM MG-132 and 1mM PMSF was added to the tissue and tubes were vortexed vigorously. Crude extracts were sonicated with a sonicating probe at 10µM amplitude for 10s three times, cleared at 3220xg twice and filtered through 0.45µm syringe filters. Protein was quantified with a standard Bradford assay. The remaining TAP procedures and mass spectrometry were as described (Song et al., 2014).

56
For the timeseries, extract containing 28mg of protein was used for TAP. For AP on anti-FLAG M2 magnetic beads, a ratio of 10µl of beads (20µl of supplied 50% slurry, Sigma) per 4mg of protein was used. To equilibrate, beads were washed in ten packed gel volumes of TBS (50 mM Tris HCl, 150 mM NaCl, pH 7.4) twice, followed by one wash with SII or Ripa buffer without inhibitors. Beads were added to appropriate volume of extract as a 50% slurry and the mix was incubated on a rotating wheel at 4°C for 2h for binding of protein to the beads. Beads were washed twice with 20 bead volumes and once with ten bead volumes of SII buffer without inhibitors (1 bead volume being 100% bead gel), followed by two washes with 10 bead volumes of Flag-to-His buffer (0.5M NaPhosphate pH7.4, 150mM KCl, 0.05% TritonX-100). Protein was eluted 3 times with 2 bead volumes each of Flag-to-His buffer containing 250µg/ml 3xFLAG peptide (sigma) for 15min each. For 6xHis tag purification, 50µl of bead slurry were used for each sample. Dynabeads® for His purification (Life Technologies; timeseries) or Dyna1 Protein G magnetic beads (qualitative study) were equilibrated by washing with 20 slurry volumes of Flag to His buffer twice. Combined eluates were incubated with the beads on a rotating wheel for 15min. Beads were washed twice with 1ml of Flag-to-His buffer followed by three washes with 1ml of freshly prepared 25mM Ammonium Bicarbonate (Sigma). After removal of the final wash, beads were stored at -80°C until on-bead digest.

Methods S3. Protein digestion, peptide and mass spectrometric analysis
For the qualitative study, the protein digestion, peptide analysis and mass spectrometry were as described (Song et al., 2014).
For the timeseries, on-bead digest was performed at room temperature as follows: 25µl 10mM DTT was added to the beads and samples were incubated at room temperature for 30min to reduce sulfhydryl groups. For carbamidomethylation, 50µl of 25mM iodoacetamide