We cultivated T. koksaghyz and N. benthamiana plants under controlled greenhouse conditions (18°C, 16-h photoperiod, 260 PPFD high-pressure sodium lamps with enhanced yellow and red spectrum) as previously described (Unland et al., 2018). T. koksaghyz tissues were harvested separately for expression analysis and immediately flash-frozen in liquid nitrogen. After lyophilization, root tissues were pulverized using a ZM 200 Ultra Centrifugal Mill (Retsch, Germany), and leaf tissues were ground under liquid nitrogen with a pestle and mortar. Latex was transferred from cut root surfaces to rubber extraction buffer (REB) [100 mM Tris-HCl pH 7.8, 350 mM sorbitol, 10 mM NaCl, 5 mM MgCl₂, 5 mM dithiothreitol (DTT)], flash-frozen and used for RNA extraction without further processing.

SRPPs were expressed in Escherichia coli BL21Ai (DE3) cells (Thermo Fisher Scientific, USA) transformed with expression vector pET23a(+) containing codon-optimized sequences of TkSRPP3/4/5 ( Supplementary Table S1 ). Protein expression and purification were carried out as previously described (Laibach et al., 2018).

Latex was harvested from the roots of 12-week-old T. koksaghyz plants (line 203-1-ST) as previously described (Post et al., 2012; Niephaus et al., 2019) with slight modifications ( Supplementary Methods S1 in Supplementary Data Sheet 1 ). For affinity enrichment, 100 µL Ni-NTA agarose (Qiagen, Germany) in a 2-mL tube was washed three times with 500 µL Ni-NTA lysis buffer (50 mM NaH₂PO₄, 300 mM NaCl, 10 mM imidazole, pH 8.0) and centrifuged (500 g, 5 min, 4°C). We dissolved 200 µg recombinant TkSRPP3/4/5 in 1 mL Ni-NTA lysis buffer containing a protease inhibitor cocktail (diluted 1:10), added this to the Ni-NTA agarose beads, and incubated the mixture for 1 h at room temperature (RT), shaking at 200 rpm. The samples were then centrifuged (500 g, 5 min, RT) and the beads were washed three times with Ni-NTA wash buffer (50 mM NaH₂PO₄, 300 mM NaCl, pH 8.0). We diluted 400 µL of the latex or its fractions with 500 µL Ni–NTA wash buffer, added this to the Ni-NTA agarose beads, and repeated the incubation, centrifugation and washing steps as above. After seven further washes with 1 mL Ni-NTA wash buffer, 500 µL of the wash buffer was added to the beads and the samples were analyzed by LC-MS. Beads loaded only with recombinant protein or latex fractions were used as controls. Three replicate samples were prepared.

LC-MS/MS data acquisition and processing steps were carried out as previously described (Lassowskat et al., 2017). Briefly, proteins were extracted and digested using a modified filter-assisted sample preparation protocol (FASP). After reduction and alkylation, the samples were digested with trypsin, followed by LC-MS/MS analysis using an EASY-nLC 1200 device coupled to a Q Exactive HF mass spectrometer (both from Thermo Fisher Scientific). For details see Supplementary Methods S2 in Supplementary Data Sheet 1 .

Raw data were processed using MaxQuant v1.6.9.0 (Cox and Mann, 2008). MS/MS spectra were assigned to the T. koksaghyz proteome (Lin et al., 2018). The sequences of 248 common contaminant proteins and decoy sequences were automatically added. Trypsin specificity was required and a maximum of two missed cleavages was allowed. We set cysteine carbamidomethylation as a fixed modification and methionine oxidation, deamidation of N and Q and protein N-terminal acetylation as variable modifications. We applied a false discovery rate of 1% for peptide spectrum matches and proteins, and enabled matching between runs, label-free quantification (LFQ) and iBAQ.

Proteins were annotated by BLASTP searching against the NCBI non-redundant and UniProt databases (e ≥ 1 × 10^–3). Data were processed using Perseus v1.6.0.7 and v2.0.11 (Tyanova et al., 2016). We removed proteins that were only identified by site, reverse hits, or potential contaminants. LFQ intensities were log₂ transformed and proteins were filtered for those with a mean LFQ intensity in the no-prey control lower than the 10^th percentile of all data (19.45 for TkSRPP3 runs; 19.91 for TkSRPP4/5 runs) or undetected in this control, as well as proteins quantified in at least two replicate AE-MS runs. Missing values were imputed using the quantile regression imputation of left-censored missing data (QRILC) algorithm in the package imputeLCMD v2.1 (Lazar and Burger, 2022; R Core Team, 2021) and normal distribution was confirmed by consulting histograms ( Supplementary Figure S1 ). Data reproducibility was examined by principal component analysis (PCA). Replicates formed clusters separately from control samples except for rubber phase control 1 from the TkSRPP3 run and pellet phase control 3 from the TkSRPP4/5 run, which were therefore excluded from further analysis ( Supplementary Figure S2 ). For enrichment analysis, we applied two sample t-tests between AE-MS and no-bait control samples. A permutation-based FDR (q-value) was calculated to correct for multiple testing. To visualize enriched proteins in volcano plots, q-values equal to zero, representing high confidence, were replaced with the next even value smaller than the smallest calculated q-value within the same approach. Volcano plots and Gene Ontology (GO) heat maps were produced using the ggplot2 v3.5.1 R package (Wickham, 2016). Venn diagrams were generated using InteractiVenn (Heberle et al., 2015). GO terms were assigned using eggNOG-mapper (Cantalapiedra et al., 2021). GO enrichment analysis was carried out using the topGO R package v2.50.0 (Alexa and Rahnenfuhrer, 2022). Protein classes were determined using PANTHER 19.0 (Thomas et al., 2022). Protein abundance data (Benninghaus et al., 2020) were processed and heat maps generated in Perseus v2.0.11 (Tyanova et al., 2016).

Protein domains were predicted using Interpro (Paysan-Lafosse et al., 2023). Phosphorylation and N-glycosylation sites were predicted using CLC Main Workbench v23.0.3 (Qiagen). Phylogenetic trees were constructed using MEGA11 (Tamura et al., 2021) and cis-acting regulatory elements were detected using NSITE-PL (Shahmuradov and Solovyev, 2015). The chromosome map was created using MapChart (Voorrips, 2002). Isoelectric points (pI) and protein charges were determined using Prot Pi v. 2.2.29.152 (https://www.protpi.ch/).

TkSRPP and TkGUT80B1 coding sequences were amplified from T. koksaghyz latex cDNA with flanking primers based on the genomic sequences (Lin et al., 2018) ( Supplementary Table S2 ). Amplified fragments were digested with restriction enzymes indicated in the primer names and ligated into the Gateway pENTR 4 entry vector (Thermo Fisher Scientific).

For the split-ubiquitin membrane yeast two-hybrid (SUY2H) assays, plasmids pRS313 and pRS314 (Sikorski and Hieter, 1989) were modified to form Gateway destination vectors. The Gateway cassette was amplified from pAG304-P_GAL1-ccdB (Addgene, USA) using primers M13 rev and attR BglII fw, and overhangs were prepared by digestion with BglII and KpnI. The pRS314 interim vector was digested with BamHI and KpnI and ligated with the Gateway cassette. For pRS313, the Gateway cassette was amplified from pBatTL (Jach et al., 2006) using primers attR SpeI fw and attR AgeI rev, and ligated into the digested pRS313 vector. TkSRPPs, TkUGT80B1 and mEmerald were introduced into pRS313-ccdB-CRU and pRS314-Nua-ccdB using Gateway LR Clonase II mix (Thermo Fisher Scientific). For co-immunoprecipitation (co–IP), TkSRPP3/4/5 and TkUGT80B1 were transferred to pAG425-P_GPD-ccdB-Cerulean and pAG423-P_GPD-ccdB-HA (Addgene), respectively, by Gateway cloning. For N. benthamiana transient expression experiments, genes were introduced into the Gateway-compatible vector pBatTL-ccdB-Cerulean as previously described (Epping et al., 2015; Unland et al., 2018). To test TkUGT80B1 activity in yeast, the TkUGT80B1 coding sequence was inserted into pAG423-P_GAL1-ccdB (Addgene) by Gateway cloning. All constructs were validated by Sanger sequencing.

RNA extraction, cDNA synthesis and qPCR were carried out as previously described (Niephaus et al., 2019) with slight modifications ( Supplementary Methods S3 in Supplementary Data Sheet 1 ). Briefly, normalized expression was calculated using the ΔC_q method (Equation 1) relative to the mean C_q value of the reference genes. To account for different primer efficiencies, mean Cq values of technical replicates were adjusted by multiplication with an adjustment coefficient based on primer efficiency (Equation 2).

Transient expression in N. benthamiana leaves was carried out using pBatTL constructs in which the protein of interest was fused to the N-terminus of the blue fluorescent protein Cerulean (Müller et al., 2010). Monomeric red fluorescent protein (mRFP) C-terminally fused to the N-terminal sequence of CYP51G1 (CYP51G1-mRFP) was used to mark the cytosolic surface of the ER (Bassard et al., 2012). Two-pore-channel 1 (TPC1) was fused to orange fluorescent protein (OFP) as a tonoplast marker (Batistič et al., 2010). To test LD localization, we co-expressed Arabidopsis LEAFY COTYLEDON 2 (AtLEC2) to induce LD formation, and infiltrated the leaves with Nile red solution (Santos Mendoza et al., 2005). AtLEC2 was amplified from A. thaliana cDNA that was obtained as previously described (Jekat et al., 2013) using flanking primers introducing restriction sites for cloning into pENTR4. This construct was used for Gateway cloning into pBatTL-ccdB using Gateway LR Clonase II mix (Thermo Fisher Scientific) (Müller et al., 2010). Leaf discs were analyzed by confocal laser scanning microscopy (CLSM) using a Stellaris 8 microscope (Leica Microsystems, Germany). Cerulean fluorescence was detected at 445–550 nm (excitation at 440 nm), mRFP fluorescence at 570–648 nm (excitation at 555 nm), and Nile red fluorescence at 571–587 nm (excitation at 541 nm).

The Saccharomyces cerevisiae strain InvSc1 (Thermo Fisher Scientific) was transformed with combinations of pRS313 and pRS314 using the lithium-acetate method (Agatep et al., 1998). Positive clones were identified by colony PCR using gene-specific and vector primers. They were grown for 5–6 h at 30°C in 1 mL synthetic defined (SD) medium containing 50 µM CuSO₄ and lacking histidine, tryptophan and methionine. Cultures were centrifuged (19,000 g, 1 min, RT) and the OD₆₀₀ was adjusted to 1 using 1× TE. We transferred 10 µL of three serial dilutions to SD medium lacking histidine and tryptophan, or to selective media containing (1) 300 µM methionine, 50 µM CuSO₄ and lacking histidine, tryptophan and uracil, or (2) 300 µM methionine, 50 µM CuSO₄ and 1 g/L 5-fluoroorotic acid (5-FOA), and lacking histidine and tryptophan. The plates were incubated at 30°C for 2–3 days.

Yeast strain InvSc1 was transformed with each of the pAG425-P_GPD-TkSRPP-Cerulean constructs and pAG423-P_GPD-TkUGT80B1-HA alone or with the three combinations of TkUGT80B1 and each TkSRPP. Positive transformants were identified by colony PCR using gene-specific and vector primers. To test for the interactions between TkUGT80B1, TkSRPP3 and TkSRPP5, 5 mL of SD medium lacking histidine, leucine or both were inoculated with a single colony of each genotype, incubated at 30°C until the OD₆₀₀ reached 3 and harvested by centrifugation (4,000 g, 10 min, RT). Cells expressing TkSRPP4 and a control expressing only TkUGT80B1 were cultivated at 20°C to enable sufficient recombinant protein synthesis. Overnight cultures were used to inoculate 50 mL SD medium to an OD₆₀₀ of 0.3. The main cultures were then cultivated at 20°C, shaking at 140 rpm. They were harvested by centrifugation after 42 h (4,000 g, 10 min, RT) and washed once with 10 mL 1× TE. Proteins were extracted with 1 mL cell lysis buffer comprising 20 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM Na₂EDTA, 1 mM EGTA, 1% Triton X-100, 2.5 mM Na₄P₂O₇, 1 mM β–glycerophosphate, 1 mM Na₃VO₄ and 1 µg/ml leupeptin (Cell Signaling Technology, USA) supplemented with 1 mM phenylmethylsulfonylfluoride (PMSF) and cOmplete EDTA-free protease inhibitor cocktail (Merck, Germany) for 1–5 min at 30 Hz in an MM400 bead mill (Retsch) followed by centrifugation (11,000 g, 5 min, 4°C). ChromoTek GFP-Trap magnetic agarose (Proteintech Group, USA) was used for immunoprecipitation according to the manufacturer’s instructions. Extracts were incubated with the beads for 1 h at 4°C on a platform rocker. Proteins were eluted in 50 µL 5× SDS loading buffer containing 100 mM DTT. Samples were separated by SDS-PAGE on 10% SDS polyacrylamide gels and analyzed as previously described (Niephaus et al., 2019) with modifications. Briefly, membranes were incubated with either an anti-GFP primary antibody (Clontech Laboratories, USA; #632380) diluted 1:2,000, or with an anti-HA primary antibody (Merck; #H3663) diluted 1:1,000–2,000, followed by washing and incubation with either a goat anti-mouse IgG coupled to alkaline phosphate (Merck; #A3562) diluted 1:10,000, followed by detection using SIGMAFAST BCIP/NBT tablets (Merck), or a goat anti-mouse IgG coupled to horseradish peroxidase (Thermo Fisher Scientific; #32430) diluted 1:1,500, followed by detection using SuperSignal West Dura Extended Duration Substrate (Thermo Fisher Scientific).

We transformed a S. cerevisiae strain, previously engineered for enhanced triterpenoid production and expressing T. koksaghyz lupeol synthase (TkLup) under the control of a galactose-inducible promoter (Bröker et al., 2018), with pAG423-P_GAL1-TkUGT80B1 or the empty vector using the lithium-acetate method (Agatep et al., 1998). As a control, the same strain lacking TkLUP was transformed with TkUGT80B1 or the empty vector. Positive transformants were identified by colony PCR using gene-specific and vector primers (or two vector primers for the empty vector). For yeast cultivation, we inoculated 5 mL SD medium lacking histidine and tryptophan with a single colony of each genotype and incubated it overnight at 30°C on a rolling wheel. We inoculated 50 mL of the same medium, supplemented with 150 µM CuSO₄ to repress sterol synthesis, with the overnight cultures to an OD₆₀₀ of 0.2 and incubated them at 30°C, shaking at 140 rpm. When the cultures reached an OD₆₀₀ of 0.5–0.6, inducible gene expression was activated by switching to SD medium containing 2% galactose instead of glucose. Cells were harvested when cultures reached an OD₆₀₀ of 4. Metabolites were extracted from lyophilized yeast pellets by adding glass beads and 1 mL ethyl acetate, followed by lysis for 30 min at 30 Hz in an MM400 bead mill. After centrifugation (11,000 g, 1 min), the supernatant was transferred to a fresh tube and the extraction was repeated twice with 0.5 mL ethyl acetate, each time vortexing for 15 min. The extracts were analyzed by LC-MS/MS using an UltiMate 3000 Rapid Separation System (Thermo Fisher Scientific) and amazon speed ion trap MS (Bruker Corporation, USA) or using an Acquity Premier LC system (Waters Corporation, UK) coupled to a Synapt XS 4k (Waters Corporation) ion mobility time-of-flight mass spectrometer. Extracts were separated using a Reprosil Pur Basic C18 (5 µm particle size, 4×250 mm) (Analytik Altmann, Germany) by isocratic elution with 10% mobile phase A (90:10 v/v isopropanol:water + 10 mM ammonium formate) and 90% mobile phase B (methanol + 10 mM ammonium formate). The flow rate was 0.4 mL/min and the run time 45 min. The column temperature was set to 23°C. For the identification of lupeol, an authentic standard was analyzed separately.

We identified 17 SRPP-like sequences in the T. koksaghyz genome (Lin et al., 2018, 2022). Twelve of them cluster on pseudo-chromosome 4 ( Figure 1A ). The nomenclature we applied is based on sequence similarity to known orthologs (mainly from T. brevicorniculatum) and the first publicly available annotations (Collins-Silva et al., 2012; Hillebrand et al., 2012; Laibach et al., 2018; Schmidt et al., 2010a). Corresponding sequence IDs and comparisons with other published nomenclatures are summarized in Supplementary Table S3 . We identified one copy each of TkSRPP3/4/5 and TkSRPP6, the latter being the only full TkSRPP gene located outside the cluster, on the other arm of pseudo-chromosome 4. The most recent genome annotation (Lin et al., 2022) lists TkSRPP5 as a pseudogene, but the earlier version (Lin et al., 2018) includes an open reading frame, which we confirmed by amplification from cDNA. Similarly, the 2022 genome assembly contains a premature stop codon for TkSRPP6, but amplification of an open reading frame from cDNA confirmed its integrity. We found that TkSRPP3/4/5 form a contiguous set along with 10 additional paralogs. This indicates an inverted duplication in which the adjacent copies of the first five genes run in the opposite direction. Because the first three contiguous genes are more closely related to each other than other paralogs, we named them TkSRPP1a-c and their copies with sequence identities of 93–100% TkSRPP1a1/b1/c1 respectively ( Table 1 ). However, TkSRPP1c1 contains a stop codon after 30 bp and has only 93% identity to TkSRPP1c, suggesting that functional redundancy has allowed sequence divergence in this case. The next two genes of the cluster were designated TkSRPP7a and TkSRPP2a, and their copies TkSRPP7b and TkSRPP2b. TkSRPP2a is not annotated in the 2022 genome release, but was identified manually by homology searches. Three additional TkSRPP sequences sharing 99.9% identity form another cluster on pseudo-chromosome 3 and were designated TkSRPP8a-c.These sequences appeared to originate from incomplete gene transposition and duplication of TkSRPP2a/b, because they contain only the first 654 bp (exon I, intron I, part of exon II) of TkSRPP2a/b which could encode a 50 aa peptide. The high conservation between TkSRPP8a/b/c suggests a relatively recent duplication event. Because of this truncation resulting in an incomplete REF domain, we excluded TkSRPP8 from further analysis. TkSRPP gene duplications have been proposed before (He et al., 2024), but our classification differs in that we identified five additional genes and established a nomenclature based on previous publications. The TkSRPPs, sequence identities and predicted protein properties are summarized in Table 1 .

We designed primer pairs to quantify TkSRPP gene expression in the tissues of wild-type T. koksaghyz plants. No discriminating primers could be designed for the proposed gene duplications, so the data for TkSRPP1, TkSRPP2 and TkSRPP7 reflect the expression of all duplicates. We observed extremely strong TkSRPP3/4/5 expression in latex ( Figure 1B ), in agreement with published RNA-Seq data and previously reported high protein levels (Collins-Silva et al., 2012; Lin et al., 2018, 2022; Niephaus et al., 2019). TkSRPP3 and TkSRPP4 showed similarly high transcript levels, each about twice the level of TkSRPP5. A comparable profile was observed in the roots, which contain considerable amounts of latex ( Figure 1B ). In leaves, TkSRPP3/4/5 expression was low. TkSRPP1 was also predominantly expressed in latex, albeit at level of one tenth or less of that of TkSRPP3/4/5. TkSRPP2 expression notably differed from the others, with the highest level in leaves. TkSRPP6 was expressed at low levels and TkSRPP7 at moderate levels in all tissues. These diverging sequences and spatial expression profiles indicate specialized, tissue-specific functions. Because latex is of special interest in T. koksaghyz, the remarkably strong expression of TkSRPP3/4/5 and their contribution to NR biosynthesis prompted us to screen for protein interaction partners of these paralogs in latex (Collins-Silva et al., 2012). We also identified several N-glycosylation and phosphorylation sites that could play a role in these interactions and their biological functions ( Figures 1C–E ).

Recombinant TkSRPPs with His₆ tags were produced in E. coli and coupled to agarose beads before mixing with whole latex as well as the separate rubber phase (RP), interphase (IP) and pellet phase (PP) obtained by centrifugation ( Figure 2 ). The latex phases are thought to comprise different cellular fractions, with the RP mostly containing rubber particles (Collins-Silva et al., 2012; Schmidt et al., 2010b), the PP containing most of the membrane and organelle components of the latex, and the IP containing most of the cytosol. This was supported by the enrichment of GO terms of the cellular component (CC) category in one fraction relative to all detected latex proteins ( Supplementary Table S4 ), enabling us to infer interaction sites in the laticifers and potentially detect interactions with less-abundant proteins that are enriched in particular phases. This is interesting because of potential TkSRPP functions other than rubber particle stabilization and TbSRPP3/4/5 were localized in the ER and cytosol in N. benthamiana (Laibach et al., 2018). We used two controls to ensure qualitatively reliable results. First, we used TkSRPP-His₆-coupled beads that were not loaded with latex as no-prey controls for each paralog. Second, uncoupled agarose beads were loaded with each latex fraction as no-bait controls. Proteins were identified by LC-MS/MS and quantified by LFQ using the MaxQuant software suite. The no-bait controls were used to calculate protein enrichments ( Figure 2 ).

Proteins were considered as potential interaction partners when they were significantly enriched compared to the no-bait control (LFQ log₂ fold change (FC) ≥ 1; q < 0.05) ( Figure 3A ). Given the nature of the experiment, the MS data contained a relatively large number of missing values that were mainly assumed to be missing not at random (MNAR). Accordingly, missing values were imputed using QRLIC ( Supplementary Figure S1 in Supplementary Data Sheet 1 ), which performs best for left-censored MNAR data (Wei et al., 2018). Imputation is necessary to enable quantitative statistical analysis, but it can influence data analysis and interpretation. Therefore, we did not strictly exclude proteins from the group of potential interaction partners if they fell outside the defined thresholds for log₂FC and q-value but were subjected to imputation ( Figure 3A , blue dots). Our candidate lists thus contain proteins meeting the defined threshold criteria and represent potential interaction partners with the highest confidence according to our data analysis strategy, but we do not limit TkSRPP interaction partners to these proteins ( Supplementary Tables S7–S9 in Supplementary Data Sheet 1 ).

For each TkSRPP, four independent datasets were obtained representing whole latex and its three fractions, revealing different numbers of interaction partners in partially overlapping sets ( Figures 3B, C ). The volcano plots show that the respective bait TkSRPP was one of the most strongly enriched proteins in all AE-MS experiments ( Figure 3A ). The number of interaction partners identified for TkSRPP5 (83) was much lower than for TkSRPP3 (232) and TkSRPP4 (662), this could have been caused by different affinities of TkSRPP3/4/5 to the Ni-NTA agarose beads or the protein stability on the Ni-NTA agarose. Further, it must be distinguished between the total number of hits and the number of different proteins detected within each latex phase. Some of the interactors (16% for TkSRPP3, 19% for TkSRPP4 and 23% for TkSRPP5) were detected under more than one condition, providing more confidence in their veracity ( Figure 3B ). The large number of interacting proteins enriched exclusively in one phase confirmed that the use of separate latex phases allowed the identification of more specialized interactions and low-abundance interactors. For TkSRPP3, most interacting proteins were enriched from the whole latex (114), followed by the PP (87), IP (50) and RP (25). In contrast, most TkSRPP4 interactors were enriched from the IP (318), followed by the RP (229), PP (185) and whole latex (66, 18 of which were exclusive to whole latex). We found 75 proteins enriched from both the RP and PP, suggesting they are not exclusive to rubber particles but are also found in other organelles. For TkSRPP5, only three interactors were enriched from the PP, with one also enriched from whole latex. Most TkSRPP5 interactors were enriched from whole latex (48) and the IP (39), with 12 enriched from both. The interactomes indicated that TkSRPP4 is a promiscuous hub protein that binds many partners from different compartments, whereas TkSRPP5 interacts more specifically, primarily with proteins present in rubber particles or the cytosol. We also found 10 interactors common to TkSRPP3, 4 and 5 ( Figure 3C ). TkSRPP4 shared the most interactors with the other paralogs, probably reflecting the presence of more interactors overall. More than half of the proteins interacting with TkSRPP5 also interacted with TkSRPP3 or TkSRPP4, leaving only 37 unique to TkSRPP5. In contrast, most TkSRPP3 and TkSRPP4 interactors were exclusive.

Because RP interactors are of particular interest for the elucidation of the role of TkSRPP3/4/5 in NR biosynthesis, compared the TkSRPP3/4/5 RP interactomes and found that four interactors are shared by TkSRPP4 and TkSRPP5 ( Supplementary Table S6 ) whereas all TkSRPP3 RP interactors are exclusive, again highlighting the functional specialization on the rubber particle, as previously shown on LDs in N. benthamiana for TbSRPP4 and TbSRPP5 but not TbSRPP3 (Laibach et al., 2018). The shared TkSRPP4/TkSRPP5 RP interactors comprised Ras-related protein Rab11C, a CBL-interacting protein kinase, α-ketoglutarate-dependent dioxygenase, and a protein similar to an uncharacterized protein from lettuce. Notably, TkSRPP5 was identified as an interaction partner of TkSRPP3 by co-enrichment from whole latex (log₂FC 1.1) and the IP (log₂FC 2.7), but TkSRPP3 was not identified as an interactor when TkSRPP5 was the bait. For the closely related homologs TbSRPP3/4/5, all pairwise interactions have been shown by bimolecular fluorescence complementation (BiFC) (Laibach et al., 2018).

To gain insight into the specific functions of each TkSRPP in latex, we screened for GO terms enriched within each subset of paralog-specific interactors ( Figure 4 ). Interestingly, ‘chloroplast stroma’ (12), ‘thylakoid’ (5) and ‘cytosol’ (30) associated proteins were significantly enriched in the CC category among the exclusive TkSRPP3 interactors (compared to all TkSRPP interactors) whereas the TkSRPP4 interactors were significantly enriched for ‘polysomal ribosome’ (15) and ‘cytosolic large ribosomal subunit’ (19), ‘ER membrane’ (13) and ‘membrane’ (181). Accordingly, in the molecular function (MF) category, ‘structural constituent of ribosome’ (30) was enriched along with ‘inorganic molecular entity transmembrane transporter activity’ (26). For the TkSRPP5 interactome, the enrichment of ‘cytosolic small ribosomal subunit’ (3) in the CC category aligns with the enrichment of ‘ribosomal small subunit biogenesis’ (2) and ‘rRNA processing’ (2) in the biological process (BP) category. GO analysis thus indicated that TkSRPP4 and TkSRPP5 are associated with cytosolic ribosomal processes. It is possible that the enrichment of ribosomal proteins was favored because of the assumed connection between TkSRPPs and the ER, the likely origin of rubber particles. But their detection may also reflect the high affinity of ribosomal proteins for the agarose beads (Keilhauer et al., 2015). For TkSRPP5 interactors, we further observed the enrichment of ‘microtubule cytoskeleton’ (3) in the CC category, as well as proteins related to responses to hormone and external stimuli (3, 4), ‘lipid modification’ (2) and ‘regulation of auxin mediated signaling pathway’ (2) in the BP category. For the TkSRPP3 interactors, the BP category terms ‘cellular macromolecule biosynthetic process’ (12), ‘protein-containing complex assembly’ (8), ‘response to cadmium ion’ (11) and ‘organonitrogen compound biosynthetic process’ (15) were enriched. The TkSRPP4 interactors were enriched for two more general BP terms relating to ‘regulation of post-embryonic development’ (13) and ‘cell differentiation’ (20). In the MF category, ‘kinase activity’ (7) was the only enriched term among the TkSRPP3 interactors, whereas the TkSRPP5 interactors were enriched for terms related to GTP, purine ribonucleoside and nucleoside phosphate binding (3, 3, 4), ‘GTPase activity’ (3) and ‘transferase activity’ (2).

Protein class analysis based on annotated UniProt IDs revealed similar distributions for all three interactomes, partially matching the enriched GO terms ( Figure 4 ; Supplementary Figure S3 ). Prominent protein classes among the interactors included metabolite interconversion enzymes such as oxidoreductases, transferases and hydrolases, protein-modifying enzymes such as proteases, and protein-binding affinity modulators such as protease inhibitors and G-proteins.

We extracted data relating to the abundance of each interactor from our dataset in wild-type T. koksaghyz roots at 8, 12 and 24 weeks (Benninghaus et al., 2020), and clustered the interactors and corresponding bait TkSRPP according to their temporal accumulation profiles ( Supplementary Figure S4 ). This identified interacting proteins with similar abundance profiles as their TkSRPP partners, which supports their status as interaction partners because gene co-expression is more likely for interacting proteins than random protein pairs (Ge et al., 2001; Jansen et al., 2002; von Mering et al., 2002). TkSRPPs 3/4/5 were all assigned to clusters with increasing protein levels over time. All three are already highly abundant after 8 weeks of growth, so that interactions with proteins from the same cluster may represent basal interactions rather than conditional interactions in response to particular stimuli.

To identify protein interactions related to NR synthesis, the interactors were correlated with proteins that are enriched or depleted when TkCPT-like 1 (TkCPTL1) is downregulated in the latex of T. koksaghyz plants by RNA interference (RNAi) (data obtained by Niephaus et al., 2019) ( Table 2 ). These TkCPTL1-RNAi plants produce significantly less NR than wild-type plants because TkCPTL1 is thought to form heterodimers with TkCPT1 and/or TkCPT2 and thus assemble into a cisPT complex that catalyzes the synthesis of poly(cis-1,4-isoprene) on the surface of rubber particles (Niephaus et al., 2019). We identified TkCPT1 as a TkSRPP3 interactor and TkCPT1 was significantly less abundant in the TkCPTL1-RNAi plants. An interaction between TkSRPP3 and TkCPT1 highlights the connection between TkSRPP3 and NR synthesis on the rubber particle surface, and is consistent with the reported interaction between Hevea brasiliensis SRPP and CPT6 (Brown et al., 2017). TkSRPP4 interacted with 10 proteins whose abundance changed in the NR-depleted transgenic plants, including the rate-limiting enzyme of the mevalonate (MVA) pathway: 3–hydroxy-3-methylglutaryl-CoA reductase (HMGR). The MVA pathway provides the C₅ building block isopentenyl diphosphate (IPP) for NR synthesis, and was downregulated in the RNAi lines. Other interactors involved in isoprenoid metabolism, including squalene epoxidase 1 (SQE1) and germacrene oxidase (GAO), were more abundant in the TkCPTL1-RNAi lines. Another downregulated interactor was a homolog of a ricin B-like lectin, and additional lectin homologs were found in the TkSRPP3 and TkSRPP4 interactomes. One TkSRPP5 interactor similar to a lettuce (Lactuca sativa) putative methyltransferase was also downregulated in the RNAi plants.

We selected two of the 10 candidate interactors shared by TkSRPP3/4/5 for confirmation using a second method and characterized them in more detail. The first candidate (annotated as TkSRPP6) was selected because TkSRPP6 is the only complete TkSRPP gene found outside the main cluster, and its function has not been studied thus far. The second candidate, annotated as a sterol 3–β-glucosyltransferase/UDP-glycosyltransferase (UGT) 80B1 family member, was designated TkUGT80B1. It was selected because glycosides are known to be involved in plant defense and stress responses, but the role of UGTs in latex has not been investigated. The proteins were enriched to different levels in different phases in the AE-MS datasets for TkSRPP3, 4 and 5 ( Table 3 ).

For SUY2H, TkSRPP3/4/5 baits were N-terminally fused to a modified N-terminal ubiquitin fragment (N_UbA) with lower affinity for the ubiquitin C-terminus (C_Ub), thus minimizing false positive results (Johnsson and Varshavsky, 1994). The N_UbA fusions were co-expressed with TkSRPP6 C-terminally fused to C_Ub. Interactions reconstitute functional ubiquitin, leading to the cleavage and degradation of the URA3 reporter, thus conferring uracil auxotrophy and resistance to 5-FOA (Johnsson and Varshavsky, 1994; Reichel and Johnsson, 2005). We used the monomeric fluorescent protein mEmerald combined with the TkSRPPs as negative controls. Using this system, we were able to confirm that TkSRPP6 interacts with TkSRPP4 and TkSRPP5. For TkSRPP3 the growth pattern was indistinct but suggested weak interaction ( Figure 5A ). This supports the initial screens, but indicates that protein interactions are dependent on the experimental conditions and highlights the importance of independent confirmation. For interactions between TkSRPP3/4/5 and TkUGT80B1, we were able to pull down TkUGT80B1-3×HA with Cerulean-tagged TkSRPP3 and TkSRPP5 by Co-IP, but we could not confirm the interaction with TkSRPP4 in this experimental setup ( Figures 6A ; Supplementary Figure S5 ). In summary, additional methods confirmed four of six pairwise interactions indicated by AE-MS, two for each candidate, suggesting the unconfirmed interactions are restricted to specific native conditions or part of bigger complexes.

TkSRPP6 in silico analysis identified the REF domain common to all known dandelion SRPPs and REF proteins, as well as three potential phosphorylation sites ( Figure 5B ). Phylogenetic comparisons showed that TkSRPP6 has diverged from other TkSRPPs and is more closely related to other REF proteins (including those involved in stress responses in plants that do not produce NR) than to the tightly clustered TkSRPP1/2/3/4/5 and 7 ( Figure 5C ). This was supported by protein identities of ~60% between TkSRPPs 3/4/5, but only 37–49% when TkSRPP6 was compared to the other paralogs ( Table 1 ). We therefore screened a 1-kb region of the TkSRPP6 promoter for stress-responsive elements ( Figure 5D ; Supplementary Table S5 ), revealing E–box elements (CANNTG) at positions –200 and –500 bp relative to the start codon, and the core sequence of a G-box type E-box (CACGTG) and extended G-box elements at –800 bp ( Figure 5D ) (Galvāo et al., 2019; Nagao et al., 1993; Shahmuradov and Solovyev, 2015). The G-box recruits G-box binding factors (GBFs), which include bZIP and bHLH proteins such as MYC2 (Heim et al., 2003; Menkens et al., 1995; Sibéril et al., 2001; Williams et al., 1992; Zhang et al., 2019). G-box elements mediate the effects of hormones, light and temperature (Eyal et al., 1993; Guiltinan et al., 1990; Hong et al., 1995; Mason et al., 1993; Shaikhali et al., 2012; Toledo-Ortiz et al., 2014), whereas E-box elements regulate temperature-dependent and circadian expression in stress-responsive genes (Liu et al., 2015; Seitz et al., 2010). These findings suggest that TkSRPP6 is transcriptionally regulated by different stress factors, in agreement with other data for SRPP genes (Cao et al., 2017; Dong et al., 2023a; Fricke et al., 2013; He et al., 2024; Hillebrand et al., 2012).

TkUGT80B1 was found to contain a UDP-glucuronosyltransferase/UDP-glucosyltransferase domain, an N-terminal domain similar to glycosyltransferase family 28, and a C-terminal domain resembling that of CIII-like, another glycosyltransferase, including the nucleotide diphosphate sugar binding site ( Figure 6B ) (Moncrieffe et al., 2012). The last amino acid in the so–called UDPGT motif differs between UDP-glucosyltransferases (where it is glutamine) and UDP-galactosyltransferases (histidine), so the presence of glutamine in TkUGT80B1 suggests it has UDP-glucose transferase activity ( Figure 6B , gray box) (Kubo et al., 2004). Sequence analysis also predicted that the C-terminal domain is cytosolic, separated from the N-terminal part by a transmembrane domain of 19 amino acids. The protein contains nine putative phosphorylation sites and two N-glycosylation sites. Phylogenetic analysis supported the relationship between TkUGT80B1 and UGT80B1 enzymes from the family Asteraceae and other plants, as well as more distant relationships with other UGT families ( Figure 6C ). The TkUGT80B1 promoter ( Figure 6D ) contains three elicitor response elements (ERE1-3), which contribute to fungal elicitor-mediated gene expression (Yang et al., 1998). Additionally, we found an E-box, two WRKY11-binding sites (one overlapping with ERE2/3) and one WRKY40-binding site (W-box). WRKY transcription factors are involved in plant defense (Javed and Gao, 2023), suggesting stress-responsive transcriptional regulation, which ties in with the role of glycosylated secondary metabolites in the plant defense system (Hussain et al., 2019; Louveau and Osbourn, 2019).

We had already determined the spatial expression profile of TkSRPP6 when comparing SRPP paralogs ( Figure 1 ). Applying the same approach to TkUGT80B1 in 10-week-old wild-type T. koksaghyz plants, we observed strong expression in the latex (consistent with the AE-MS experiments) but low expression in roots and leaves ( Figure 6E ), similar to the expression profiles of TkSRPP3/4/5. The previous detection of TkUGT80B1 protein in roots may reflect the large amount of latex in this tissue (Benninghaus et al., 2020).

Temporal expression profiling in latex revealed a steady increase in TkSRPP3/4/5 mRNA levels during weeks 6–14 ( Figure 7 ), as shown for root protein levels before (Benninghaus et al., 2020). However, transcript levels stayed constant or decreased slightly between weeks 14 and 16 ( Figure 7A ). TkSRPP3 and TkSRPP4 expression declined after 12 weeks but increased again after 14 weeks. Similarly, TkUGT80B1 expression increased over time, declined slightly after 12 weeks, and stayed constant between weeks 14 and 16 ( Figure 7B ). TkSRPP6 expression was constant at low levels throughout the experiment ( Figure 7B ). The expression data reflected the high level of heterogeneity between individuals reported earlier (McAssey et al., 2016; Nowicki et al., 2019; Panara et al., 2018; Wieghaus et al., 2022). Our data demonstrated comparable temporal expression patterns for TkUGT80B1 and TkSRPP3/4/5, but not TkSRPP6.

The analysis of different latex phases by AE-MS provided crude data concerning the potential localization of TkSRPP6 and TkUGT80B1. For more detailed analysis, we expressed fusion proteins in N. benthamiana along with subcellular markers. We prepared constructs in which TkSRPP6 and TkUGT80B1 were C-terminally fused to the fluorescent reporter Cerulean, and transiently co-expressed them with ER and tonoplast markers. TkSRPP6-Cerulean fluorescence and the ER marker CYP51G1-mRFP (Bassard et al., 2012) overlapped almost completely ( Figure 8A ), whereas TkUGT80B1-Cerulean fluorescence largely coincided with the tonoplast marker TPC1-OFP (Batistič et al., 2010) ( Figure 8C ). Tk/TbSRPPs 3/4/5 were previously shown to be associated with rubber particles (Collins-Silva et al., 2012; Hillebrand et al., 2012), which are related to LDs, thus explaining the LD localization of TbSRPPs in N. benthamiana (Laibach et al., 2018). We therefore determined whether TkSRPP6 and TkUGT80B1 also associate with LDs by co-expressing the Cerulean fusion constructs with AtLEC2, encoding a transcription factor that promotes LD formation in leaves (Santos Mendoza et al., 2005). We then stained the LDs with the lipophilic fluorescent dye Nile red. We found that the Cerulean fluorescence profiles of TkSRPP6 and TkUGT80B1 described above included additional punctuate fluorescence that overlapped with the Nile red signal ( Figures 8B, D ). The affinity of these candidates for LDs, despite the absence of enrichment in the RP fraction in AE-MS experiments, suggests they interact with TkSRPP3/4/5 on the surface of rubber particles but in a conditional manner.

Finally, we tested the predicted UGT activity of TkUGT80B1 in a yeast strain engineered for optimized pentacyclic triterpenoid synthesis and harboring a T. koksaghyz lupeol synthase gene (TkLup) (Bröker et al., 2018). Lupeol is a pentacyclic triterpenoid present in T. koksaghyz roots and NR, and is therefore a potential native substrate for TkUGT80B1 (Benninghaus et al., 2020; Pütter et al., 2019). Isoprenoid metabolites were extracted from yeast cultures and LC-MS chromatograms were compared to control strains either expressing TkLUP together with an empty vector or TkUGT80B1 without TkLup ( Figure 9 ; Supplementary Figure S6 ). We observed an additional peak (m/z +606.5) for yeast cells expressing TkUGT80B1 and TkLup ( Figure 9B ). The mass corresponds to a positively charged lupeol hexose ammonium ion adduct, and thus indicates TkUGT80B1 has lupeol glycosylating activity. Based on the molecular structure of lupeol, we deduce that TkUGT80B1 is a C₃-glycosyltransferase.

Our AE-MS experiments revealed distinct but overlapping interactomes for TkSRPP3/4/5 in whole latex and its three fractions. TkSRPP4 interacted with more proteins than the others and may function as a hub. TkSRPP3 and TkSRPP5 also interacted with each other, although enrichment was only observed from whole latex and IP, not from the RP fraction ( Supplementary Table S7 ). SRPP heterodimers have also been reported for T. brevicorniculatum (Laibach et al., 2018). These findings suggest TkSRPPs can act cooperatively, in agreement with the additive effect of TbSRPPs 3/4/5 on artificial poly(cis-1,4-isoprene) body size and dispersity (Laibach et al., 2018). Lipid–protein interactions influence membrane composition (Harayama and Riezman, 2018) so the TkSRPP3/TkSRPP5 interaction may induce specific rearrangements in the lipid monolayer of rubber particles that promote the most stable lipid distribution, and/or enhance the steric repulsion assumed to be caused by SRPPs on the rubber particle surface ( Figure 10A ) (Hillebrand et al., 2012). TkSRPP3 and TkSRPP5 may also form complexes with their common interactors, including a REF family protein distantly related to a perilipin-4-like protein from the tobacco hawkmoth Manduca sexta, which was significantly less abundant in NR–depleted T. koksaghyz roots (Benninghaus et al., 2020). Perilipins are LD-associated proteins in animals that promote the formation and stability of LDs by regulating lipolysis (Griseti et al., 2024). Although this protein was not enriched from the RP, further analysis to determine its impact on rubber particles would be interesting, especially given its lack of interaction with TkSRPP4.

To understand the role of TkSRPP3/4/5 in NR biosynthesis beyond rubber particle stability, we screened the interactomes for further proteins related to NR and other isoprenoids ( Supplementary Table S7 ). Many candidate TkSRPP interactors were found to differ in abundance when comparing proteins in the latex of TkCPTL1-RNAi plants and wild-type controls, suggesting a contribution to NR biosynthesis (Niephaus et al., 2019) ( Table 2 ). Interestingly, TkSRPP3 interacted with TkCPT1 and TkREF, two well-known components of the NR biosynthesis machinery. The interaction with TkCPT1 supports the hypothesis that TkSRPPs affect TkCPT1 activity, causing the low NR content in Tk/TbSRPP RNAi plants (Collins-Silva et al., 2012; Hillebrand et al., 2012). Based on the proposed role of TbREF in rubber particle biogenesis (Laibach et al., 2015) and the recruitment of HbCPT6 from the cytosol to the ER by HbSRPP (Brown et al., 2017), TkSRPP3 may recruit TkCPT1 to ER sites where TkSRPP3 is in contact with other TkSRPPs and TkREF, thus modifying the lipid composition, helping TkCPT1 to channel nascent poly(cis-1,4-isoprene) chains between ER leaflets and favoring the formation of rubber particles ( Figure 10B ). This is supported by the reported interaction between HbSRPP and HbREF (Yamashita et al., 2016). The TkSRPP3/TkCPT1 interaction was detected from whole latex and the IP. Although NR-producing CPTs have mostly been identified in rubber particles (Dai et al., 2013; Schmidt et al., 2010b), the recruitment of TkCPT1 from the cytosol to the ER by TkSRPP3 may be the mechanism by which TkCPT1 becomes localized to this compartment.

TkSRPP3 and TkSRPP4 interacted with MVA pathway enzymes that provide the C₅ building block IPP for NR polymerization (Pu et al., 2021; Salehi et al., 2021). This group comprised ATP-citrate synthase (TkACL1), TkHMGR, and mevalonate kinase (TkMVAK1) (all interacting with TkSRPP4), as well as TkMVAK10 (interacting with TkSRPP3) and phosphomevalonate kinase 3 (TkPMVK3, interacting with TkSRPPs 3 and 4). TkSRPP4 also interacted with an FPP synthase (TkFPS1), which provides the most likely starter molecule for NR polymerization in vivo (Puskas et al., 2006; Tanaka et al., 1996; Xie et al., 2008). TkSRPP3 and TkSRPP4 may therefore be required to ensure an efficient supply of metabolic precursors to the cisPT complex by forming the structural components of a NR metabolon ( Figure 10C ), although metabolic channeling experiments would be required for confirmation (Srere, 1972, 1987; Zhang and Fernie, 2021). This model is supported by the enrichment of TkMVAK1, a cytosolic enzyme (Cho et al., 2022; Niu et al., 2021; Simkin et al., 2011), from the RP by TkSRPP4 ( Supplementary Data S2 in Supplementary Data Sheet 1 ). TkSRPP3 may also stabilize the complex by interacting with TkSRPP5, which in turn interacts with multiple proteins related to the ‘microtubule cytoskeleton’ ( Figure 4 ). TkHMGR was enriched from the IP, whereas HMGRs are usually found in the ER (Hampton et al., 1996; Leivar et al., 2005). However, the presence of eight HMGR paralogs in the T. koksaghyz genome (Lin et al., 2022) and the differential expression of TbHMGRs (Van Deenen et al., 2012) suggests there is scope for functional specialization, with at least one HMGR associated with NR synthesis (Campos and Boronat, 1995; Chappell, 1995; Leivar et al., 2005; Lin et al., 2022; McCaskill and Croteau, 1998). The involvement of this TkHMGR paralog in the supply of precursors for NR synthesis is supported by its depletion in TkCPTL1-RNAi plants ( Table 2 ) (Niephaus et al., 2019).

Intriguingly, TkSRPP4 and TkSRPP5 also interacted with plastidial methylerythritol (MEP) pathway enzymes, an alternative route to IPP (Vranová et al., 2012). Specifically, TkSRPP4 interacted with 1-deoxy-d-xylulose-5-phosphate synthase (TkDXS8) whereas both TkSRPP4 and TkSRPP5 interacted with 4-hydroxy-3-methylbut-2-en-1-yl diphosphate reductase (TkHDS1). Although latex does not contain genuine chloroplasts, MEP pathway enzymes and low levels of corresponding mRNAs have been detected in T. koksaghyz latex before (Lin et al., 2018; Niephaus et al., 2019). Additionally, specialized plastids known as Frey-Wyssling (F.W.) complexes (Frey-Wyssling, 1929) have been described in T. koksaghyz and H. brasiliensis latex (Abdul Ghaffar, 2017; Dickenson, 1969; Gomez and Hamzah, 1989; Moir, 1959), and these compartments may comprise the MEP pathway in latex. However, TkDXS8 and TkHDS1 were enriched from the RP, IP and whole latex but not the PP where F.W. complexes would be presumed. This may reflect the disruption of F.W. complexes during processing or the liberation of MEP pathway enzymes by another mechanism, although we would also expect interactions with TkSRPP3 in this scenario due to the significant number of chloroplast-related TkSRPP3 interactors. Feeding experiments in H. brasiliensis showed that in this species the MEP pathway contributes to carotenoid rather than NR biosynthesis in latex (Sando et al., 2008). Still, the interactions of TkSRPP3/4/5 with different isoprenoid precursor pathways raise the possibility that they drive IPP flux towards NR synthesis and provide more evidence that TkSRPP4 is a hub protein whereas TkSRPP3 and TkSRPP5 are more specialized.

TkSRPP4 also appears to engage with isoprenoid pathways downstream of FPP by interacting with enzymes involved in sesquiterpene lactone and triterpenoid biosynthesis (González-Coloma et al., 2011; Huber et al., 2015, 2016; Padilla-Gonzalez et al., 2016), the latter including TkSQS1, TkSQE1 and oxidosqualene cyclase 5 (TkOSC5). TkSRPP3 and TkSRPP5 interacted with TkOSC1 and TkOSC5, respectively. Both are latex-specific enzymes and TkOSC1 produces at least four different triterpenoids from 2,3-oxidosqualene, most likely provided by TkSQE1 (Pütter et al., 2019; Unland et al., 2018), whereas TkOSC5 did not produce any triterpenoids in N. benthamiana (Pütter et al., 2019). The transcriptional co–regulation of TkSQS1, TkSQE1 and TkOSC1 facilitates tight metabolic coupling, and TkSQS1 colocalizes with TkSQE1 in the ER of N. benthamiana (Unland et al., 2018). Given the presence of transmembrane domains in both proteins, their interaction with TkSRPP4 in the IP may reflect their translocation caused by the phase separation procedure. TkOSC1 has yet to be detected in the ER following heterologous expression (Pütter, 2017). Therefore, TkSRPP3 and TkSRPP4 may cooperatively mediate the assembly of these three consecutive enzymes in the cytosol, ER or on rubber particles for the efficient synthesis of bioactive triterpenoids, given that other OSCs were shown to localize to LDs in yeast (Milla et al., 2003) and triterpenoids are the most abundant non-polyisoprenoid component in separated NR from T. koksaghyz (Pütter et al., 2019). TkOSC5 may need to interact with TkSRPPs to maintain stability or activity. TkSRPP3/4/5 could thus affect the quality of NR as an industrial raw material because the triterpenoid content is proposed to influence the physical properties of the polymer (Xu et al., 2017).

Common TkSRPP3/4/5 interactors were related to membrane and vesicular trafficking, lipid metabolism and stress responses ( Supplementary Table S6 ). TkSRPP3/4/5 may therefore promote stress tolerance by affecting lipid modification, the proteolytic cleavage of pathogen-derived proteins, the inhibition of pathogen-derived proteases and/or cellular adaptations by membrane trafficking. Rab7 proteins, for example, are involved in vacuolar trafficking and improve abiotic stress tolerance when overexpressed (Mazel et al., 2004; Rodriguez-Furlan et al., 2023; Tripathy et al., 2017).

As the protein TkSRPP7 was enriched from whole latex and the IP by TkSRPP3/4/5, despite the relatively low TkSRPP7 transcript levels in the latex of 10-week-old plants ( Figure 1B ) and the lack of evidence showing its association with rubber particles, this paralog may interact with TkSRPP3/4/5 elsewhere in the laticifers as part of stress-related processes. This may also explain why TkSRPP2 interacted with TkSRPP4 and TkSRPP5 in all latex fractions, and why TkSRPP1 interacted with TkSRPP3 in whole latex ( Supplementary Table S7 ). The roles of these TkSRPPs in stress responses are supported by their transcriptional upregulation following treatment with MeJA or the overexpression of TkMYC2 (He et al., 2024; Wu et al., 2024). TkSRPPs may also form multimers to fulfil their functions, as suggested for HbSRPP (Wititsuwannakul et al., 2008).

The overrepresentation of proteins associated with the chloroplast stroma and thylakoids among exclusive TkSRPP3 interactors was striking ( Figure 4 ). TkSRPP3 and its interactors related to the chloroplast stroma and thylakoids may associate with the plastid-like F.W. complexes, explaining why most of these proteins were enriched from the PP. Notably, rubber particles were also observed within plastid-like structures in T. koksaghyz laticifers (Abdul Ghaffar, 2017), so TkSRPP3 and its chloroplast-related latex interactors may also be connected with such rubber particles. The presence of ferritins, peroxiredoxins, glutathione Ѕ-transferase (GST) and aconitate hydratase among those TkSRPP3 interactors indicate roles in antioxidant defense, redox regulation and detoxification (Briat, 1996; Dietz, 2003; Kumar et al., 2024; Kumar and Trivedi, 2018; Moeder et al., 2007; Pascual et al., 2021). Ferritin inhibits the formation of reactive oxygen species (ROS) by Fe (Halliwell and Gutteridge, 1984; Kroh and Pilon, 2020; Ravet et al., 2009), peroxiredoxins are antioxidants that mediate redox-dependent signaling (Liebthal et al., 2018; Sevilla et al., 2015), and GSTs counter oxidative stress by conjugating glutathione (Cozza et al., 2017; Dixon and Edwards, 2009; Labrou et al., 2015; Nianiou-Obeidat et al., 2017; Wagner et al., 2002). Overexpression of these genes confers abiotic and biotic stress tolerance (Deák et al., 1999; Lo Cicero et al., 2017; Roxas et al., 2000; Wang et al., 2023; Xiao et al., 2023; Zang et al., 2017), and their endogenous promoters are responsive to stress and phytohormones (Dellagi et al., 2005; García Mata et al., 2001; Horling et al., 2002, 2003; Marrs, 1996; Tiwari et al., 2016). GST and aconitate hydratase are also associated with cadmium stress tolerance, a significantly enriched process among TkSRPP3 interactors ( Figure 4 ) that has also been experimentally linked to REF proteins (Dixit et al., 2011; Kim et al., 2011; Liu et al., 2013, 2016; Zhou et al., 2019). These interactors may therefore form a TkSRPP3-dependent network of stress tolerance effectors within F.W. complexes (Cerveau et al., 2016; Manevich et al., 2004), as illustrated in Figure 10C . The large number of kinases among the TkSRPP3 interactors suggests the effector network is regulated by kinase cascades, supported by multiple phosphorylation sites on TkSRPP3. Finally, the glutathione peroxidase activity of GSTs prevents lipid oxidation (Dixon et al., 2009; Dixon and Edwards, 2009; Ohkama-Ohtsu et al., 2011) and could be particularly valuable in latex that contains diverse lipids, many associated with bioactive properties (Bae et al., 2020; González-Coloma et al., 2011; Pütter et al., 2019).

The involvement of TkSRPP5 in stress response is supported by the enrichment of interactors associated with responses to different stimuli ( Figure 4 ). The activation of TkSRPP5 in response to such stimuli could be achieved by phosphorylation at the six predicted phosphorylation sites or by N-glycosylation ( Figure 1E ).

TkSRPP6 was one of two candidate interactors that we chose for further analysis, due to its isolated genomic locus and phylogenetic clustering with stress-related proteins from non-rubber-producing plants rather than other TkSRPPs ( Figure 5C ). The TkSRPP6 gene was also expressed at lower levels than TkSRPP3/4/5 in all tissues (and in latex over time) ( Figures 1 , 7 ). SUY2H results confirmed that TkSRPP6 interacted with TkSRPP4 and TkSRPP5, but not TkSRPP3 ( Figure 5A ). The interaction of TkSRPP6 with the abundant latex proteins TkSRPP4 and TkSRPP5, despite their different molecular characteristics, was striking. The identification of MYC2-binding sites in the TkSRPP6 promoter suggested inducible expression, supported by the transcriptional induction observed after MeJA treatment and TkMYC2 overexpression ( Figures 1 , 5D ) (TkSRPP1 in He et al., 2024; TkSRPP7 in Wu et al., 2024). This induction might be transient, as shown for MeJA (He et al., 2024), and would therefore not show up in our qPCR data for wild-type plants. Stress-induced expression is a common feature of the REF family and is supported by the homology of TkSRPP6 to stress-related proteins ( Figure 5C ). The heterologous expression of TkSRPP6 in N. benthamiana resulted in localization to the ER and LDs, matching its enrichment from the RP by TkSRPP3 ( Figures 8A, B ; Table 3 ). A recent study (He et al., 2024) suggested TkSRPP6 (named TkSRPP1 therein) was localized to the cytosol, plasma membrane and chloroplast, but the authors did not induce LD formation nor did they use plasma membrane and ER markers to confirm their assumptions. We observed no plastid signals for our TkSRPP6-Cerulean fusion protein ( Figures 8A, B ). TbSRRPs were found to be localized to the cytosol (Laibach et al., 2018) and TkSRPP6 enrichment from the IP supports a cytosolic localization. We conclude that TkSRPP6 may have affinities for different cellular compartments that may change depending on specific conditions.

The localization of TbSRPPs 4 and 5 and TkSRPP6 together with indications of inducible gene expression suggest that TkSRPP6 may interact at basal levels with TkSRPP4 and TkSRPP5 on the surface of the ER and rubber particles, but predominantly after its short-term transcriptional induction in response to various stimuli. These interactions most likely play a role in stress tolerance, and TkSRPP6 could engage with established complexes formed by TkSRPP4/TkSRPP5. TkSRPP6 may respond to stress not only in the latex but also in green tissues as its transcript levels were comparable low in all tissues.

TkUGT80B1 was the second candidate selected for in depth characterization because glycosyltransferases have not been described in dandelion latex before and glycosides are involved in stress responses (Augustin et al., 2011; Rogowska and Szakiel, 2020). An independent co-IP assay confirmed that TkUGT80B1 interacts with TkSRPP3 and TkSRPP5 ( Figure 6A ). The interactions were further corroborated by the similar temporal expression patterns of TkUGT80B1 and TkSRPP3/4/5 in latex ( Figure 7 ) (Ge et al., 2001; Jansen et al., 2002; von Mering et al., 2002).

The predicted glycosyltransferase activity of TkUGT80B1 was confirmed for the C₃ position of the triterpenoid lupeol in yeast, with UDP-glucose as the most likely sugar donor ( Figure 9 ). This was supported by the presence of glutamine in the C-terminal UDPGT motif, which is conserved among glucosyltransferases but not galactosyltransferases ( Figure 6D ) (Kubo et al., 2004). Accordingly, we have identified the first enzyme from the latex of T. koksaghyz that produces a triterpenoid saponin and have provided first evidence for the presence of these compounds in dandelions. Given the low substrate specificities of UGTs (Vogt and Jones, 2000), the glycosylation of additional, structurally similar triterpenoids in T. koksaghyz latex is likely. The heterologous expression of TkUGT80B1 did not result in the glycosylation of yeast sterols when sterol synthesis was repressed ( Figure 9 ). It is unclear whether TkUGT80B1 can also utilize phytosterol substrates, as described for its homologs (Stucky et al., 2014), or other lipids in latex. The analysis of Arabidopsis UGTs indicated that AtUGT80B1 is not required for the synthesis of major steryl glucosides but rather for the production of minor glucosides (Stucky et al., 2014). Accordingly, TkUGT80B1 may glycosylate triterpenoids rather than major membrane sterols in accordance with its predominant expression in latex ( Figure 6E ). Adaptation to freezing stress was inhibited in atugt80b1 knockout plants, and was potentially related to low levels of sterol glycosides, the products of AtUGT80B1 (Mishra et al., 2015). It would be interesting to determine whether triterpenoid saponins have a similar positive effect on freezing tolerance in T. koksaghyz because the lipid–rich latex in its roots has already been proposed to act as an anti-freezing protectant during extremely cold winters, which are common in its native habitat.

The presence of several stress-related cis-acting regulatory elements in the TkUGT80B1 promoter, including binding sites for WRKY transcription factors, suggests the gene is transcriptionally regulated in response to biotic stress ( Figure 6D ). WRKY transcription factors can affect defense response positively as well as negatively (Javed and Gao, 2023; Wani et al., 2021), but given the reported positive correlation between SRPPs and stress tolerance (Kim et al., 2016, 2012; Laibach et al., 2018; Seo et al., 2010), TkUGT80B1 is also likely to improve stress tolerance, although this should be investigated in more detail.

Saponins are stored in the vacuole (Kesselmeier and Urban, 1983; Mylona et al., 2008; Urban et al., 1983). TkUGT80B1 expression in N. benthamiana suggested tonoplast localization ( Figure 8C ) and the enrichment of TkUGT80B1 from the PP by TkSRPP3 supports this finding. Enrichment from the IP by TkSRPP4 and TkSRPP5 may reflect the rupture of vacuoles during phase separation. Tonoplast localization may allow the immediate storage of nascent saponins in the vacuole. TkSRPP3 and TkSRPP5 could mediate the transfer of saponins from their biosynthetic enzyme to a transporter by linking both proteins. The transport of saponins to the vacuole may be mediated by ABC-type transporters (Kato et al., 2022; Ramilowski et al., 2013) and two proteins with homology to ABC-type transporters were identified as TkSRPP4 interactors, one of which also interacted with TkSRPP5. Interactions between TkSRPP4/TkSRPP5 and these transporters could also promote saponin efflux from the vacuole to promote stress tolerance. Localization studies in N. benthamiana showed that TkUGT80B1 can also accumulate in LDs, suggesting it might be located on the surface of rubber particles and interact with TkSRPPs there. The interaction between TkSRPPs and TkUGT80B1 may also promote the metabolic flux towards triterpenoid saponin synthesis by linking the triterpenoid synthesizing TkOSC1/5 with the glycosylating TkUGT80B1. TkSRPP3 and TkSRPP5 could further recruit either TkUGT80B1 or specific lipid substrates to direct glycoside synthesis and ultimately modify the composition of membranes in response to environmental conditions.