ABSTRACT
The fast dynamics and reversibility of posttranslational modifications by the ubiquitin family pose significant challenges for research. Here we present SUMO-ID, a technology that merges proximity biotinylation by TurboID and protein-fragment complementation to find SUMO-dependent interactors of proteins of interest. We developed an optimized split-TurboID version and show SUMO interaction-dependent labelling of proteins proximal to PML and RANGAP1. SUMO-dependent interactors of PML are involved in transcription, DNA damage, stress response and SUMO modification and are highly enriched in SUMO Interacting Motifs, but may only represent a subset of the total PML proximal proteome. Likewise, SUMO-ID also allowed us to identify novel interactors of SUMOylated SALL1, a less characterized SUMO substrate. Thus, SUMO-ID is a powerful tool that allows to study the consequences of SUMO-dependent interactions, and may further unravel the complexity of the ubiquitin code.
INTRODUCTION
Ubiquitin-like (UbL) proteins belong to a superfamily of small proteins that attach covalently to target substrates in a transient and reversible manner. The UbL family includes Small Ubiquitin-like Modifiers (SUMOs). The mammalian SUMO family consists of at least three major SUMO paralogues (SUMO1,-2,-3). Human SUMO2 and SUMO3 share 97% sequence identity, whereas they share 47% of sequence identity with SUMO11. Protein SUMOylation is a rigorously regulated cycle involving an enzymatic machinery that acts in a stepwise manner. Briefly, the C-terminal di-glycine motif of mature SUMOs binds lysines in substrates through the sequential action of E1 SUMO-activating enzyme SAE1/SAE2, E2 conjugating enzyme UBC9 and SUMO E3 ligases2. If required, SUMO as well as the substrate can be recycled by the action of SENPs that cleave the isopeptide bond. Like Ub, SUMO has internal lysines that can be further modified, extended as SUMO chains, modified by Ub chains to target degradation, or even modified by smaller moieties, like acetyl groups3–5. Together, these constitute the concept of the “SUMO code” and the ongoing challenge is to understand how these modifications drive distinct substrate outcomes and cellular fates.
SUMO plays crucial roles in nuclear processes underlying health and disease such as the DNA damage response, cell cycle regulation, transcription and proteostasis6. SUMO is known to control vital biological processes including development7 and cholesterol homeostasis8. Improvements in mass spectrometry technology have led to the identification to date of more than 40,700 SUMO sites within 6,700 SUMO substrates9. While cell-wide proteomics approaches can help to understand global SUMO signaling10, better tools are needed that allow the study of the cause and consequences of particular SUMOylation events for individual substrates.
SUMO can also interact non-covalently with SUMO interacting motifs (SIMs) found in some proteins. SIMs are β strands composed of an hydrophobic core motif that interacts with the hydrophobic residues of the SIM-binding groove of SUMOs to form an intramolecular β-sheet11. A well characterized role of the SUMO-SIM interaction concerns the SUMO-targeted Ub ligases (STUbL). The two described human STUbLs, RNF4 and Arkadia/RNF111, recognize poly-SUMOylated substrates through their SIMs and ubiquitylate them, leading to their proteasomal degradation12,13. The SUMO-SIM interaction also plays critical roles in assembling protein complexes: interaction of the SIM1 of Ran Binding Protein 2 (RanBP2) with the SUMOylated version of Ran GTPase-activating protein 1 (RanGAP1) is crucial for the RanBP2/RanGAP1*SUMO1/UBC9 E3 ligase complex formation14.
Another intriguing function of SUMO-SIM interaction is the targeting of proteins to Promyelocytic Leukemia Nuclear Bodies (PML NBs). PML NBs are membrane-less ring-like protein structures found in the nucleus. They are bound to the nuclear matrix, make contacts with chromatin fibers15 and associate with transcriptionally active genomic regions16. They consist of a shell composed of PML proteins that surround an inner core in which client proteins localize. Due to the heterogeneity of client proteins, PML NBs have diverse nuclear functions (reviewed in17,18). The PML gene contains 9 exons and numerous splicing variants. All PML isoforms contain the N terminal TRIpartite Motif (TRIM) that is responsible for PML polymerization and NB formation19, binding to Arsenic Trioxide (ATO)20 and may act as an oxidative stress sensor18. PML also contains a phospho-SIM located at its exon 7 and shared by most PML isoforms21. Almost all PML isoforms contain three putative SUMO sites: K65, K160 and K490. PML SUMOylation is a well characterized signal for RNF4-mediated ubiquitylation and degradation22.
Proximity-dependent labelling methods are based on promiscuous labeling enzymes that produce reactive molecules that covalently bind neighbor proteins. Labeled proteins can be then purified and identified using affinity-purification coupled to mass spectrometry methods23. Proximity-dependent biotin identification (BioID)24 uses a promiscuously active Escherichia coli biotin ligase (BirA*) generated by a point mutation (R118G) to biotinylate lysines in nearby proteins within an estimated range of 10 nm25. By fusing BirA* to specific proteins, BioID efficiently identifies interactors at physiological levels in living cells26. It has been extensively used in the Ub field, for instance, to identify substrates of E3 ligases27,28. Recently, a more efficient version of BioID, termed TurboID, has been developed29, being this more suitable for transient protein-protein interaction (PPI) detection. Several studies have developed split-versions and applied “protein fragment complementation” to BioID and TurboID, where proximal biotinylation is dependent on the proximity of the fusion partners, opening new opportunities for spatial and temporal identification of complex-dependent interactomes30,31.
To study how SUMOylation and SUMO-SIM interactions can lead to other roles and fates for particular substrates poses particular challenges. SUMOylation occurs transiently and often in a small percentage of a given substrate. Modified proteins can be readily deSUMOylated and SUMO can be recycled and passed to other substrates. SUMO-SIM interactions are also difficult to analyze due to their weak affinity (Kd 1-100 μM). To overcome those technical issues, we developed SUMO-ID, a new strategy based on Split-TurboID to identify interactors of specific substrates dependent on SUMO conjugation or interaction. Using PML as a model, we demonstrate that SUMO-ID can enrich for factors that depend on PML-SUMO interaction. Importantly, the identified proteins are represented among proximal interactors of PML identified using full-length TurboID. We also applied SUMO-ID to a less-characterized SUMO substrate, Spalt Like Transcription Factor 1 (SALL1), and identified both known and novel interactors that depend on intact SUMOylation sites in SALL1. SUMO-ID is thus a powerful tool to study transient and dynamic SUMO-dependent interaction events. The developed methodology is generic and therefore widely applicable in the Ub and UbL field to identify readers of these modifications for individual target proteins to improve our insight in non-covalent signal transduction by Ub and UbL.
RESULTS
Identification of SUMO-dependent interactions: the SUMO-ID strategy
We posited that Split-TurboID, in which one fragment is fused to SUMO and the complementary fragment to a protein of interest, could identify transient SUMO-dependent interactors (Fig. 1). Upon covalent SUMOylation or non-covalent SUMO-SIM interaction, both fragments are brought together, presumably close enough to allow refolding of the TurboID enzyme. In the presence of biotin, the reconstituted TurboID can then label proximal complexes, which can be purified by streptavidin pull-down and identified by liquid chromatography-mass spectrometry (LC-MS). Due to the high affinity of streptavidin-biotin interaction, harsh cell lysis and stringent washes that significantly reduce unspecific protein binding can be applied. We named this approach “SUMO-ID”.
T194/G195 Split-TurboID enables SUMO-ID studies
We applied the previously described E256/G257 BioID split-point30 to TurboID, but found it unsuitable for SUMO-ID. While SUMO-dependent reconstitution of the E256/G257 was observed in our pilot experiments, the NTurboID256-fusions had a significant background biotinylating activity (Figure S1). This is likely due to residual biotin binding and activation by the intact NTurboID256 biotin-binding pocket. We examined the BirA structure to identify a new TurboID split-point that would yield two completely inactive fragments (see File S1). The biotin binding pocket of BirA is composed of three β-strands (strands 5, 8 and 9), the N-terminus of helix E and the 110-128 loop (Fig. 2A). We split TurboID at T194/G195, so that the resulting NTurboID194 fragment (NTbID) carries the principal 110-128 biotin binding loop and the β-strands 5 and 8, while the CTurboID195 fragment (CTbID) carries the β-strand 9 necessary to the formation of the biotin binding β-sheet.
We tested T194/G195 Split-TurboID for SUMO-ID. CTbID-SUMOs were incorporated into substrates in a very efficient manner (Fig. 2B). NTbID and CTbID alone were catalytically inactive and yielded no biotinylation. Combining NTbID-substrate and CTbID-SUMOs resulted in a high-yield biotinylation activity of TurboID after 16 hours of biotin exposure. Modification of NTbID-substrates by CTbID-SUMOs (Fig. 2B, FLAG blot, black arrowheads) and its corresponding biotinylation activity (Fig. 2B, biotin blot, black arrowheads) were efficiently detected, notably in the case of PML protein. Free biotinylated CTbID-SUMOs, that might come from recycling of previously labeled moieties, were observed (Fig. 2B, biotin blot, white squares). We also examined by immunofluorescence, and confirmed that the streptavidin signal recognizing the biotinylated substrates is dependent on fragment-complementation (Fig. 2C). Thus, T194/G195 Split-TurboID biotinylation activity is dependent on fragment complementation, with reduced or no leaky biotinylation of the two fragments, so it could be useful for SUMO-ID and for studying other protein-protein interactions.
We applied the rapamycin-inducible dimerization system, based on FKBP (12-kDa FK506-binding protein) and FRB (FKBP-rapamycin-binding domain)32, which has been used previously to evaluate the PPI dependency of Split-BioID reconstitution30. We fused NTbID and CTbID to FRB and FKBP, respectively, and stably expressed the constructs in HEK293FT cells. We tested short and long rapamycin treatments, together with short biotin labeling times, to evaluate self-biotinylation activity of the reconstituted TurboID. As previously shown, the NTbID, even though it contains the principal biotin binding 110-128 loop, was catalytically inactive after 24 hours of biotin treatment (Fig. 2D). We observed that biotinylation activity of the reconstituted TurboID correlated well with rapamycin and biotin treatments, showing its dependency on PPI and biotin labeling times. 24 hours of rapamycin treatment led to a 16-fold higher FKBP/FRB PPI dependent biotinylation activity at 2 hours of labeling time. Altogether, these data demonstrate that T194/G195 Split-TurboID fragments have low intrinsic affinity and high biotinylation activity at short biotin labeling times, making them suitable for SUMO-ID.
SUMO-ID detects both covalent and non-covalent SUMO-dependent interactions using short biotin-labeling times
Interaction of a protein with SUMO can be via covalent SUMOylation or non-covalent SUMO-SIM interaction. We used PML, which can both be SUMOylated and has a well-characterized SIM domain, in conjunction with SUMO wild type (WT) or mutants that lack the C-terminal di-glycine (ΔGG) necessary for covalent conjugation. We used a stable HEK293FT cell line expressing NTbID-PML, into which CTbID-SUMOs were transfected, using short biotin-labeling times (0.5-2 hours). We observed that CTbID-SUMO1/2 transfections led to high SUMO-dependent biotinylation activity after only 2 hours of biotin treatment (Fig. 3A, biotin blot, black arrowhead). Additionally, ATO treatment, which induces PML SUMOylation, further enhanced the SUMO-dependent biotinylation. With 30 minutes of biotin treatment, CTbID-SUMO1/2ΔGG induced biotinylation of unmodified NTbID-PML, likely through SUMO-SIM interactions (Fig. 3A, biotin blot, white arrowhead). With longer biotin labeling (2 hours), biotinylation of endogenous SUMO modified NTbID-PML was also detected, more strongly in the case of ATO treatment (Fig. 3A, biotin blot, black arrowhead). Biotinylated free CTbID-SUMO1/2ΔGG were detected, while the WT counterparts were not biotynilated at 2 hours (Fig. 3A, biotin blot, white squares), supporting that recycling of biotinylated SUMOs may be linked to longer labeling times. Indeed, additional experiments showed that appearance of free biotinylated CTbID-SUMOs increased with longer labelling times (Figure S2). Altogether, these results demonstrate that SUMO-dependent biotinylation activity for specific targets, especially at short biotin labeling times, may be a useful strategy for identifying specific SUMO-dependent interactors of those proteins.
Reduced labeling times and lower expression levels reduced SUMO recycling, but still allowed some degree of recycling (and therefore loss-of-specificity) to occur, so we incorporated two further modifications. First, we designed SUMO isopeptidase-resistant versions of CTbID-SUMOs (SUMO non-cleavable, or SUMOnc, Fig. 1B)33. This could also reduce the target identification derived from SUMO-SIM interactions involving free unincorporated CTbID-SUMOs, since most CTbID-SUMOncs would be incorporated into substrates. The same strategy was applied to Ub (CTbID-Ubnc). Secondly, we transferred non-cleavable CTbID-SUMOs into pTRIPZ, an all-in-one doxycycline-inducible (Dox) lentiviral vector (Fig. 1B). Regulated expression would offset any deleterious effects stemming from non-cleavable SUMO isoforms, and provide useful experimental control (i.e. non-induced versus induced). Inducible TRIPZ-CTbID-SUMOnc showed enhanced SUMOylation compared to the constitutive WT SUMO versions (Fig. 3B). Free non-incorporated versions of SUMO1/2nc and Ubnc were not detectable (Fig. 3B, MYC blot, white squares). Stable cell populations were established (HEK293FT, U2OS and RPE-1 cells) for each (SUMO1nc, SUMO2nc, Ubnc). Validation of TRIPZ-CTbID-SUMO2nc by WB and immunofluorescence is shown (Figure S3). We then introduced constitutively-expressed NTbID-PML into TRIPZ-CTbID−SUMO1nc or -SUMO2nc cells and proved that biotinylation occurs in PML NBs as expected, in a doxycycline dependent manner (Figure S4). These data show that the use of regulated SUMOnc versions leads to both high activity and specificity needed for the SUMO-ID approach.
To further validate the specificity of SUMO-dependent biotinylation activity with PML, we generated control cells carrying NTbID-PML3MAS, a mutated version of PML lacking the three principal SUMOylation sites (K65, K160 and K490) and the best-characterized SIM domain. While strong SUMO-ID biotinylation activity was observed with the WT version of PML, this biotinylation activity was completely abrogated in the case of PML3MAS (Fig. 3C). This lack of biotinylation activity was specific to SUMO, as ubiquitylation-dependent biotinylation activity was observed in TRIPZ-CTbID-Ubnc / NTbID-PML3MAS double stable cell line (Figure S5). NTbID-PMLWT forms true NBs, while NTbID-PML3MAS forms NB-like bodies, as reported previously34 (Fig. 3D). Biotinylation driven by SUMO-ID was observed in NBs containing NTbID-PMLWT, and it was enhanced after 2 hours of ATO treatment, but not in the NB-like structures containing NTbID-PML3MAS. Thus, these results show that SUMO-ID biotinylation activity is dependent on substrate-SUMO interaction.
SUMO-ID identifies SUMO-dependent interactors of PML
Since PML NBs are known hubs of SUMO-dependent signaling17,18, we wondered which interactions in NBs via PML are SUMO-dependent, so we performed SUMO-ID using NTbID-PMLWT compared to NTbID-PML3MAS, each combined with TRIPZ-CTbID-SUMO2nc. Biotinylated proteins were purified by streptavidin pull-down and sequenced by LC-MS (Table S1). 59 high-confidence SUMO-dependent interactors of PML were enriched in PMLWT SUMO-ID compared to PML3MAS SUMO-ID (Fig. 4A). Among those, SUMO E3 ligases (PIAS1, PIAS2, PIAS4, TRIM28), transcriptional regulators (TRIM22, TRIM24, TRIM33, GTF2I, IRF2BP2, IFI16, ZNF280B, MED23, MEF2D, SNW1, RPAP3), and DNA repair proteins (RMI1, BLM, SLX4, XAB2) were identified. Of note, PIAS1 is known to induce PML SUMOylation35 and SUMO-SIM interaction of BLM is necessary for its targeting to PML bodies36, which highlights the specificity of the SUMO-ID strategy. Of particular interest, GTF2I and IRF2BP2, identified here by SUMO-ID, form fusion proteins with RARA and cause Acute Promyelocytic Leukaemia (APL, see Discussion)37,38. We validated these two proteins, as well as TRIM33 and UBC9, as SUMO-dependent interactors of PML by WB (Fig. 4B).
STRING networking of SUMO-dependent interactors of PML shows a highly interconnected cluster related to protein SUMOylation, DNA damage response and transcriptional regulation (Fig. 4C), while GO enrichment also highlighted protein SUMOylation and transcriptional regulation, as well as DNA repair and stress response pathways (Fig. 4D; Table S2). Collectively, this data show that the SUMO-ID strategy can efficiently identify SUMO-dependent interactors of PML, and that SUMO interaction with PML reinforces essential processes.
PML SUMO-ID hits localize to PML NBs
We checked whether some of the SUMO-dependent interactors of PML localize to NBs. We generated a YFP-PML cell line by inserting YFP into the endogenous PML locus in U2OS cells (Figure S6), and looked for co-localization of selected SUMO-dependent PML interactors by confocal microscopy. Within individual cells, we observed frequent and multiple co-localization events for PIAS4, TRIM24, TRIM33 and UBC9 in PML NBs (Fig. 5), whereas PIAS2, GTF2I and IRF2BP2 colocalizations were less frequent, suggesting heterogeneity in PML NB composition that may depend on different factors (including, but not limited to SUMOylation density, subnuclear localization, cell cycle stage, other PTMs, or contrastingly, technical limitations with antibodies or fixations).
SUMO-dependent interactions are a subset of PML proximal proteome
PML NBs are membraneless structures thought to behave as phase-separated liquids and with high heterogeneity in composition39. These characteristics make their purification very challenging, and no proteomic data are nowadays available. Therefore, to compare the obtained PML SUMO-ID specific sub-proteome with the regular PML interactome, we decided to characterize a comprehensive PML and PML3MAS proximity interactome using standard full-length TurboID (FLTbID). We generated stable U2OS cell lines for FLTbID-PMLWT, FLTbID-PML3MAS and FLTbID alone, and treated them or not with ATO to induce PML SUMOylation. High confidence PML proximal proteome was composed of 271 proteins that were enriched in FLTbID-PMLWT samples compared to FLTbID alone (Fig. 6A, Table S3). STRING networking showed a main core cluster composed of 73.6% of the identified proteins (Fig. 6B). The most representative subclusters were composed of 1) RNA splicing and mRNA processing proteins, 2) transcription, RNA biosynthesis and DNA damage response proteins and 3) replication and SUMOylation related proteins. This largely aligned with Gene Ontology (GO) enrichment analysis, which revealed that PML proximal interactors participate in replication, transcription, RNA splicing, DNA damage response, cell cycle regulation, SUMOylation and ubiquitylation, and telomere maintenance, consistent with fact that PML in U2OS regulates the ALT mechanism40 (alternative lengthening of telomeres; Fig. 6C, Table S4).
SUMOylation of PML is thought to be a controlling factor for composition and dynamics of NBs. Are all NB interactions linked to PML dependent on SUMO? To answer this question, we subdivided the PML interactome into SUMO-dependent or - independent interactors, by comparing FLTbID-PMLWT and FLTbID-PML3MAS samples. We observed some proteins that likely localize to PML NBs41, such as NCOR-1, STAT3, JUN, BRCA2 and HDAC9, were also enriched in TurboID-PML3MAS, suggesting SUMO-independent targeting to PML NBs (Table S3). Importantly, many of SUMO-dependent interactors identified by SUMO-ID are part of SUMO-dependent PML NBs proteome using standard TurboID, including PIAS2, PIAS4, TRIM24, TRIM33 and IRF2BP2 (Fig. 7A; Table S3), supporting the validity of SUMO-ID to identify SUMO-dependent interactors. Interestingly, scores of some PML interactors decreased after ATO treatment (TRIM24, TRIM33, SENP5), suggesting that those proteins may rapidly undergo dissociation or degradation in response to PML SUMOylation. We confirmed such effect for TRIM24 by WB (Fig. 7B). Altogether, these data confirm that SUMO-ID identified hits are a subset of the SUMO-dependent PML proximal proteome.
SUMO-dependent interactors of PML are enriched in SIMs
We expected that many of the SUMO-dependent PML interactors might do so via SUMO-SIM interactions and, therefore, should contain or be enriched in SIMs. To test this, we designed and executed an in-silico SIM enrichment analysis. We generated 1000 random lists of 59 proteins (the size of the SUMO-ID identified protein list) and evaluated the presence of SIMs (Table S5). The median of single SIM and multiple SIM presence in the random lists were 45.76% and 23.73% respectively (Fig. 7C). SUMO-ID identified proteins showed a much higher content of SIMs, with single SIM and multiple SIM presence values of 96.61% and 89.83% respectively. It is noteworthy that around 83% of the identified SIMs in PML SUMO-ID list were preceded or followed within the first 4 amino-acids by acidic residues (D, E) or a Serine. Since longer proteins are expected to have more SIMs, we then normalized the SIM content with the size of proteins on the lists to obtain the value of “SIMs per thousand of amino acids” (STAA) (Table S5). The values obtained with the random lists showed a Gaussian distribution (d’Agostino and Pearson normality test, K2 value 3.836, p-value 0.15) (Fig. 7D). The median of the values obtained with the random lists was 4.85 (Log2 = 2.28) STAA, while for PML SUMO-ID list was 18.42 (Log2 = 4.20) STAA, which translates to a SIM enrichment value of 3.8 times higher than the random lists (empirical p-value < 0.001). These results show that SUMO-dependent interactors of PML are highly enriched in SIMs.
SUMO-ID identifies interactors of SUMOylated SALL1
To test the sensitivity and discovery potential of the SUMO-ID, we applied this technique to SALL1, a transcriptional repressor that is SUMOylated42,43, but of which nothing is known about the causes or consequences of this modification. Using TRIPZ-CTbID–SUMO1nc or SUMO2nc HEK293FT stable cell lines, we introduced NTbID– SALL1WT or SALL1ΔSUMO (with mutations in 4 major SUMOylation consensus sites) and evaluated SUMO-ID by WB. Efficient SUMO-ID biotinylation activity was observed when using SUMO2nc (Fig. 8A, black arrowhead). NTbID–SALL1WT localizes to the nucleus, forming nuclear bodies with high SUMO-ID activity, and NTbID–SALL1ΔSUMO also forms aggregates in the cytoplasm (Fig. 8B). Specificity of SALL1 SUMO-ID was confirmed in cells, as biotinylation occurs only in SALL1WT upon doxycycline induction and biotin supplementation. SALL1 SUMO-ID identified potential SUMO-dependent interactors of SALL1 such as the transcription factors TLE3, DACH1/2 and ARID3B, as well as NuRD complex proteins GATAD2A/B, MTA1/2 and RBBP4/7 (Fig. 8C; Table S6), already known as SALL1 interactors44. We also identified components of the SUMOylation machinery, such as PIAS1. We confirmed that SUMOylated SALL1 was biotinylated and purified via SUMO-ID (Fig. 8D, black arrowheads) as well as NuRD complex proteins GATAD2B, MTA2 and RBBP4 (Fig. 8D). MCODE subclustering of the STRING interaction network showed a highly interconnected cluster composed of NuRD complex proteins (Fig. 8E) that was also enriched as GO term (p-value 2.40·10−4). Thus, SUMO-ID is sensitive and specific, allowing the study of SUMO-dependent interactors for proteins of interest, opening new avenues of understanding how SUMO can affect their function.
DISCUSSION
The fast dynamics and reversibility of SUMOylation, and the low affinity of SUMO-SIM interactions pose significant challenges not only for SUMO research, but for respective studies of Ub and other UbLs. The use of His-tagged K0-SUMO to isolate substrates and map SUMOylation sites has been instrumental to show the widespread presence of this modification in the human proteome9,45. Direct purification of SUMOylated proteins using immunoprecipitation is a gold standard and can be applied to cells and tissues46, but is also challenging because SUMOylation might affect a small proportion of low abundance proteins, and perhaps only under certain conditions (e.g a discrete cell cycle phase or upon DNA damage). Recently, the NanoBiT-based ubiquitin conjugation assay (NUbiCA) was described that uses a split-luciferase approach to allow a quantitative assessment of Ub-modified proteins47. Bimolecular fluorescence complementation (BiFC) approaches employ a split fluorescent protein that enables the localization of UbL-modified proteins in yeast or human cells to be monitored48–50. If applied to UbLs and substrates, the BiCAP method51, which allows purification of reconstituted GFP using GFP nanobodies, could likely enrich modified substrates and perhaps interactors. However, none of these approaches captures the dynamic environment of specific UbL-modified proteins, often characterized by weak and transient interactions.
Here we describe SUMO-ID, a powerful technique that allows the study of the causes and consequences of SUMO-dependent interactions for specific proteins of interest. The fast biotinylation activity of TurboID and the specificity obtained with “protein fragment complementation” permit SUMO-ID to specifically biotinylate interactors of substrates in a SUMO-dependent manner. Combined with sensitive proteomic methods, SUMO-ID allows the identification of specific interactors, potentially revealing enzymatic machinery responsible for the SUMOylation as well as interactors that may be stabilized or recruited as a consequence of the modification. Like approaches using BiFC, the subcellular localization of SUMO-modified substrates using SUMO-ID can also be inferred, through the use of fluorescent streptavidin. However, caution should be taken with the mentioned factors in order to maintain specificity, such is the use of non-cleavable forms of UbLs or the application of short biotin labelling times. This strategy might compromise the identification of SUMO isopeptidases since their binding to SUMOylated substrate is likely affected.
At the core of SUMO-ID is Split-TurboID, which individual halves should ideally have no activity, as with all split-protein approaches. For SUMO-ID, we initially applied the E256/G257 split point described for Split-BioID30 to the fast-labelling TurboID derivative, but found that the N-terminal half (1-256) retained substantial biotinylation capacity. We speculate that this is because the “biotin pocket” is still intact and might allow leaky release of biotinoyl-AMP. Leaky biotinylation of TurboID 1-256 was also observed by Cho and colleagues in their recently published report on Split-TurboID31. Their final design was based on a L73/G74 split point which showed efficient proximity-dependent reconstitution and biotinylation, but still leaves the biotin pocket intact in C-terminal 74-321 half, opening the possibility of leaky biotinylation during longer labelling times or in stable cell lines. To avoid this, we developed and validated T194/G195 Split-TurboID that separates the β-strands 5 and 8 from the β-strand 9, completely abrogating any residual biotinylation activity of the fragments.
Here we used SUMO-ID to unravel the role of PML SUMOylation in PML NBs function. We identified 59 proteins as SUMO-dependent PML interactors that participate in essential nuclear processes such as protein SUMOylation, transcriptional regulation, DNA repair and stress response. There is growing evidence that PML interaction with SUMO might allow partners to localise into PML NBs through SUMO-SIM interactions36,52. We demonstrated that most of the proteins identified by SUMO-ID are indeed part of the proteome of SUMO-dependent PML NBs and that they are enriched in SIMs, suggesting SUMO-SIM interaction dependency. It has been proposed that, after such partner recruitment, proteins might undergo SUMOylation by the PML NB-localized SUMO machinery that reinforces their sequestration53. In fact, PML NBs are, together with the nuclear rim, the major targets of active SUMOylation54. Our data reinforce this enzyme/substrate co-concentration model as we observed that SUMOylation machinery enzymes (UBC9, PIAS1, PIAS2, PIAS4 and TRIM28) localize to PML NBs in a SUMO-dependent manner and 80% of the SUMO-dependent PML interactors (47 out of 59) are SUMO substrates9,10.
To compare our list of SUMO-dependent versus general interactors of PML, we performed a TurboID assay for PML, with cells alone or treated with ATO, and identified 271 proteins. ATO induces PML NB formation, subsequent PML SUMOylation, partner recruitment and finally PML degradation22,55. It is used to treat APL, a type of Acute Myelocytic Leukaemia (AML), which is mainly caused by the t(15;17) translocation that fuses PML to RARA. Interestingly, two of our SUMO-ID hits, IRF2BP2 and GTF2I, also form fusion proteins with RARA and cause APL, albeit less commonly than PML fusions37,38. We validated that both localize to PML NBs. While many of the SUMO-ID candidates show increased peptide intensity in PML NBs after ATO treatment, we observed that some of them decreased. IRF2BP2 and TRIM24, which has also been linked to AML56,57, showed reduced levels after ATO treatment, suggesting that they might undergo degradation. In line with this idea, the 11S proteasome components are recruited into mature PML NBs and their localization is enhanced with ATO treatment58, suggesting that mature PML NBs may also act as proteolytic sites. In fact, inhibition of ubiquitylation accumulates SUMOylated proteins within PML NBs59, suggesting that many clients that are targeted to PML NBs and that are SUMOylated, might undergo ubiquitylation and degradation. Altogether, these data provide further insight into the role of PML SUMOylation in NB biology and open new ways of looking at the mechanisms of ATO in APL treatment.
The successful application of SUMO-ID to SALL1, a poorly characterized SUMO substrate, illustrates the sensitivity and utility of SUMO-ID. Although SALL1 SUMOylation levels are vanishingly low under physiological conditions, SUMO-ID revealed SUMO-dependent enrichment of the NuRD complex proteins GATAD2A/B, MTA1/2 and RBBP4/7. The association between SALL1, a transcriptional repressor, and the NuRD complex, a repressive histone deacetylase complex, has been previously described44. The interaction is mediated by an N-terminal 12 amino acid motif of SALL144. Once recruited, SUMOylation of SALL1 might serve to stabilize the repressor complex via SUMO-SIM interactions, with predicted SIMs present in multiple NuRD complex subunits. As histone SUMOylation is also linked to transcriptional repression60, SUMO-SIM interactions might further reinforce the SALL1/NuRD complex and drive histone deacetylation at SALL1 targets. In addition, we also found TLE3, DACH1/2 and ARID3B transcription factors as SUMO-dependent interactors of SALL1. TLE3, a transcriptional repressor of the Groucho/TLE family, interacts with HDAC2 (another NuRD complex component) and can regulate acetylation levels61. Both TLE3 and the tumor suppressor DACH1 are negative regulators of Wnt signaling62,63. Interestingly, SALL1 has been shown to enhance Wnt signaling64. Perhaps interaction with SUMOylated SALL1 serves to counteract these negative effects.
In summary, we demonstrate here that SUMO-ID, based on the 194/195 Split-TurboID reconstitution, can facilitate the identification of SUMO-dependent interactions with a protein of interest. It has little or no background, with high biotinylation activity when expressed at low levels and with short biotin incubation time. We believe that this technique improves sensitivity and selectivity when applied to infrequent SUMOylation events and low-affinity of SUMO-SIM interactions. This strategy can be applied to other UbL modifications (e.g Ub-ID shown in Figure S5), and the 194/195 Split-TurboID may be useful for other applications in cell biological studies.
METHOD DETAILS
Cell Culture
U2OS (ATCC HTB-96) and HEK293FT (Invitrogen) were cultured at 37°C and 5% CO2 in Dulbecco’s modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (FBS, Gibco) and 1% penicillin/streptomycin (Gibco). Human telomerase reverse transcriptase immortalized retinal pigment epithelial cells (hTERT-RPE1, ATCC CRL-4000) were cultured in DMEM:F12 (Gibco) supplemented with 10% FBS, 2 mM L-Glutamine and 1% penicillin and streptomycin. Cultured cells were maintained through 20 passages maximum and tested negative for mycoplasma.
Cloning
TurboID was a kind gift of A. Ting (Addgene #107171)29. PMLIVaWT and PMLIVa3MAS were previously described21. SUMO1, SUMO2, Ub, RANGAP1 and UBC9 ORFs were amplified from U2OS cell cDNA by high-fidelity PCR (Platinum SuperFi DNA Polymerase; Invitrogen). All constructs were generated by standard cloning or by Gibson Assembly (NEBuilder HiFi Assembly, NEB) using XL10-Gold bacteria (Agilent). Depending on the construction, plasmid backbones derived from EYFP-N1 (Clontech/Takara), Lenti-Cas9-blast (a kind gift of F. Zhang; Addgene #52962) or TRIPZ (Open Biosystems/Horizon) were used. After assembly, all vectors were validated by sequencing. Additional details for constructs are described in Table S7. Cloning details are available upon request.
Lentiviral transduction
Lentiviral expression constructs were packaged in HEK293FT cells using calcium phosphate transfection of psPAX2 and pMD2.G (kind gifts of D. Trono; Addgene #12260, 12259) and pTAT (kind gift of P. Fortes; for TRIPZ-based vectors). Transfection media was removed after 12-18 hours and replaced with fresh media. Lentiviral supernatants were collected twice (24 hours each), pooled, filtered (0.45 μm), and supplemented with sterile 8.5% PEG6000, 0.3 M NaCl, and incubated 12-18 hours at 4°C. Lentiviral particles were concentrated by centrifugation (1500 x g, 45 minutes, 4°C). Non-concentrated virus (or dilutions thereof) were used to transduce HEK293FT, and 8x concentrated virus was used for U2OS and hTERT-RPE1 cells. Drug selection was performed as follows: 1 μg/ml puromycin (Santa Cruz) for U2OS and HEK293FT cells, 5 μg/ml for hTERT-RPE1 cells; 5 μg/ml blasticidin (Santa Cruz) for U2OS, HEK293FT and hTERT-RPE1 cells.
CRISPR-Cas9 genome editing
Human PML encodes multiple isoforms, but most differ at the 3’ end. To target EYFP into the first coding exon, shared by most PML isoforms, an sgRNA target site was chosen (CTGCACCCGCCCGATCTCCG) using Broad institute GPP sgRNA Designer65. Custom oligos were cloned into px459v2.0 (a kind gift of F. Zhang; Addgene #62988). A targeting vector was made by amplifying 5’ and 3’ homology arms using U2OS genomic DNA, as well as the EYFP ORF (see Table S7 for oligo details). These fragments were assembled by overlap extension using high fidelity PCR and the resulting amplicon was TOPO-cloned and sequence-confirmed. Lipofectamine 2000 (Invitrogen) was used to transfect U2OS with linear targeting vector and px459 encoding SpCas9, puromycin resistance, and the PML-targeting sgRNA. 24 hours post-transfection, cells were selected for additional 24-36 hours with 2 μg/ml puromycin. Cells were plated at low density and clones were examined by fluorescence microscopy. Clones with clear nuclear body signals were manually picked and expanded. YFP-PML insertions and copy number were validated by PCR, sequencing, and Western blotting.
Transient transfections and drug treatments
HEK293FT cells were transiently transfected using calcium phosphate method. U2OS cells were transiently transfected using Effectene Transfection Reagent (Qiagen). After 24 hours of transfection, cells were treated with biotin (50 μM; Sigma-Aldrich) for indicated exposure times. For stably transduced TRIPZ cell lines, induction with doxycycline (1 μg/ml; 24 hours; Sigma-Aldrich) was performed prior to biotin treatment. ATO (1 μM; 2 hours; Sigma-Aldrich) treatments were performed (with or without biotin, depending on experiment) prior to cell lysis or immunostaining.
Western blot analysis
Cells were washed 2x with PBS to remove excess biotin and lysed in highly stringent washing buffer 5 (WB5; 8 M urea, 1% SDS in 1X PBS) supplemented with 1x protease inhibitor cocktail (Roche) and 50 μM NEM. Samples were then sonicated and cleared by centrifugation (25000 x g, 30 minutes, RT). 10-20 μg of protein was loaded for SDS-PAGE and transferred to nitrocellulose membranes. Blocking was performed in 5% milk in PBT (1x PBS, 0.1% Tween-20). Casein-based blocking solution was used for anti-biotin blots (Sigma). Primary antibodies were incubated over-night at 4°C and secondary antibodies 1 hour at room temperature (RT). Antibodies used: anti-biotin-HRP (1/1000), anti-Myc (1/1000), anti-alpha-Actinin (1/5000) (Cell Signaling Technology); anti-Flag (1/1000), anti-GTF2I (1/1000) (Sigma-Aldrich); anti-BirA (1/1000; SinoBiological); anti-IRF2BP2 (1/1000), anti-UBC9 (1/1000), anti-TRIM24 (1/1000), anti-TRIM33 (1/1000), anti-PIAS2 (1/1000), anti-PIAS4 (1/1000), anti-GATAD2B (1/1000), anti-MTA2 (1/1000), anti-RBBP4 (1/1000), anti-PML (1/1000), anti-GAPDH (1/5000), anti-beta-Actin (1/5000), anti-alpha-Tubulin (1/5000) (Proteintech); anti-PML (1/1000) (Bethyl); anti-GFP (1/1000) (SantaCruz); anti-Mouse-HRP, anti-Rabbit-HRP (1:5000) (Jackson ImmunoResearch). Proteins were detected using Clarity ECL (BioRad) or Super Signal West Femto (ThermoFisher). Quantification of bands was performed using ImageJ (v2.0.0-rc-69/1.52n) software and normalized against loading controls (GAPDH, actin, tubulin or alpha-actinin depending on experiments).
Immunostaining and confocal microscopy
U2OS and HEK293FT cells were seeded on 11 mm coverslips (25,000 cells per well; 24 well plate). HEK293FT coverslips were pre-treated with poly-L-lysine. After washing 3 times with 1x PBS, cells were fixed with 4% PFA supplemented with 0.1% Triton X-100 in PBS for 15 minutes at RT. Then, coverslips were washed 3 times with 1x PBS. Blocking was performed for 30 minutes at RT in blocking buffer (2% fetal calf serum, 1% BSA in 1x PBS). Primary antibodies were incubated for 2 hours at 37°C and cells were washed with 1x PBS 3 times. Primary antibodies used: anti-BirA (1/500; SinoBiological); anti-Myc (1/200; Cell Signaling Technology); anti-GTF2I (1/100; Sigma-Aldrich); anti-IRF2BP2, anti-UBC9, anti-TRIM24, anti-TRIM33, anti-PIAS2, anti-PIAS4, anti-CBX4 (all 1/100; Proteintech); anti-B23 (1/100) (Santa Cruz); anti-SC35 (1/100) (BD biosciences); anti-SUMO2/3 (1/100) (DSHB). Then secondary antibodies (together with fluorescent streptavidin) were incubated for 1 hour at 37°C, followed by nuclear staining with DAPI (10 minutes, 300 ng/ml in PBS; Sigma Aldrich). Antibodies used: anti-Rabbit Alexa Fluor 488, anti-Mouse Alexa Fluor 568, anti-Rabbit Alexa Fluor 568 (all 1/200; ThermoFisher). Streptavidin Alexa Fluor 594 (1/200, Jackson ImmunoResearch) was used. Fluorescence imaging was performed using confocal microscopy (Leica SP8 Lightning) with 63x Plan ApoChromat NA1.4. To obtain the signal histograms for co-localization studies in Fig. 5, we used the plot profile tool in ImageJ (v2.0.0-rc-69/1.52n).
Pulldown of biotinylated proteins
Cleared lysates from WB5 lysis buffer were adjusted to the same protein concentration before incubating them with 1/50 (vbeads/vlysate) equilibrated NeutrAvidin-agarose beads (ThermoFisher) over-night at RT. Due to the high-affinity interaction between biotin and streptavidin, beads were subjected to stringent series of washes, using the following WBs (vWB/2vlysate), all made in 1x PBS: 2x WB1 (8 M urea, 0.25% SDS); 3x WB2 (6 M Guanidine-HCl); 1x WB3 (6.4 M urea, 1 M NaCl, 0.2% SDS); 3x WB4 (4 M urea, 1 M NaCl, 10% isopropanol, 10% ethanol and 0.2% SDS); 1x WB1; 1x WB5; and 3x WB6 (2% SDS). Biotinylated proteins were eluted in 80 μl of Elution Buffer (4x Laemmli buffer, 100 mM DTT) through heating at 99°C for 5 minutes and subsequent vortexing. Beads were separated by centrifugation (25000 x g, 2 minutes).
Liquid Chromatography Mass Spectrometry (LC-MS/MS)
A stable TRIPZ–MYC-CTurboID195–SUMO2nc U2OS cell line was transduced with either EFS–FLAG-NTurboID194–PMLWT or EFS–FLAG-NTurboID194–PML3MAS for PML SUMO-ID experiments. Selection in blasticidin (5 μg/ml) and puromycin (1 μg/ml) was performed. Expression was validated by Western blot and immunostaining prior to scale-up for mass spectrometry. The TurboID-PML experiments used U2OS stable cell lines expressing low and equivalent levels of PMLWT–TurboID, PML3MAS–TurboID and TurboID alone, selected by blasticidin (5 μg/ml), and treated or not with ATO for 2 hours. For SALL1 SUMO-ID, a HEK293FT stable cell line expressing low levels of TRIPZ– MYC-CTurboID195–SUMO2nc (selected with puromycin, 1 μg/ml) was transiently transfected with EFS-FLAG-NTurboID194–SALL1WT or EFS-FLAG-NTurboID194– SALL1ΔSUMO (a mutant carrying Lys>Arg mutations at K571, K592, K982, K1086). Three independent pulldown experiments (8 × 107 cells per replicate, 8 ml of lysis) were analyzed by LC-MS/MS.
Samples eluted from the NeutrAvidin beads were separated in SDS-PAGE (50% loaded) and stained with Sypro Ruby (Invitrogen) according to manufacturer’s instructions. Gel lanes were sliced into 3 pieces as accurately as possible to guarantee reproducibility. The slices were subsequently washed in milli-Q water. Reduction and alkylation were performed using dithiothreitol (10 mM DTT in 50 mM ammonium bicarbonate) at 56°C for 20 min, followed by iodoacetamide (50 mM chloroacetamide in 50 mM ammonium bicarbonate) for another 20 minutes in the dark. Gel pieces were dried and incubated with trypsin (12.5 μg/ml in 50 mM ammonium bicarbonate) for 20 minutes on ice. After rehydration, the trypsin supernatant was discarded. Gel pieces were hydrated with 50 mM ammonium bicarbonate, and incubated overnight at 37°C. After digestion, acidic peptides were cleaned with TFA 0.1% and dried out in a RVC2 25 speedvac concentrator (Christ). Peptides were resuspended in 10 μl 0.1% formic acid (FA) and sonicated for 5 min prior to analysis.
PML samples were analyzed in a novel hybrid trapped ion mobility spectrometry – quadrupole time of flight mass spectrometer (timsTOF Pro with PASEF, Bruker Daltonics) coupled online to a nanoElute liquid chromatograph (Bruker). This mass spectrometer takes advantage of a novel scan mode termed parallel accumulation – serial fragmentation (PASEF). Sample (200 ng) was directly loaded in a 15 cm Bruker nanoelute FIFTEEN C18 analytical column (Bruker) and resolved at 400 nl/min with a 100 minutes gradient. Column was heated to 50°C using an oven.
For the analysis of SALL1 samples peptides were eluted from stage-tips in a solvent composed of deionized water/acetonitrile/FA at a ratio of 50/50/0.1 v/v/v. Peptides were lyophilized and dissolved in solvent A composed of deionized water/FA at a ratio of 100/0.1 v/v and subsequently analyzed by on-line C18 nano-HPLC MS/MS with a system consisting of an Ultimate 3000 nano gradient HPLC system (ThermoFisher), and an Exploris 480 mass spectrometer (ThermoFisher). Fractions were loaded onto a cartridge precolumn (5 mm x ID 300 μm, C18 PepMap 100 A, 5 μm particles (ThermoFisher)), using solvent A at a flow of 10 μl/minute for 3 minutes and eluted via a homemade analytical nano-HPLC column (50 cm × ID 75 μm; Reprosil-Pur C18-AQ 1.9 μm, 120 A; Dr. Maisch GmbH). The gradient was run from 2% to 40% solvent B (water/acetonitrile/FA at a ratio of 20/80/0.1 v/v/v) in 40 minutes. The nano-HPLC column was drawn to a tip of ∼10 μm and acted as the electrospray needle of the MS source. The temperature of the nano-HPLC column was set to 50°C (Sonation GmbH). The mass spectrometer was operated in data-dependent MS/MS mode for a cycle time of 3 seconds, with a HCD collision energy at 28 V and recording of the MS2 spectrum in the orbitrap, with a quadrupole isolation width of 1.6 Da. In the master scan (MS1) the resolution was set to 60,000, and the scan range was set to 300-1600, at an Automatic Gain Control (AGC) target of 3 × 106 with automated fill time. A lock mass correction on the background ion m/z=445.12 was used. Precursors were dynamically excluded after n=1 with an exclusion duration of 30 seconds, and with a precursor range of 10 ppm. Charge states 2-6 were included. For MS2 the first mass was set to 120 Da, and the MS2 scan resolution was 30,000 at an AGC target of 75,000 with automated fill time.
Mass Spectrometry data analysis
Raw MS files were analyzed using MaxQuant (version 1.6.17)66 matching to a human proteome (Uniprot filtered reviewed H. sapiens proteome, UP000005640) with a maximum of 4 missed cleavages and with precursor and fragment tolerances of 20 ppm and 0.05 Da. Label-Free Quantification was enabled with default values except for a ratio count set to 1. Slices corresponding to same lanes were considered as fractions. Biotinylation on lysine and on protein N-term was included as a variable modification for SALL1 SUMO-ID data, and biotinylated peptides were set to be included for quantification. Matching between runs and matching unidentified features were enabled. Only proteins identified with at least one peptide at FDR<1% were considered for further analysis. Data were loaded onto the Perseus platform (version 1.6.14)67 and further processed (Log2 transformation, imputation, median normalization when needed). A t-test was applied in order to determine the statistical significance of the differences detected. Proteins detected with at least 2 peptides (except when otherwise specified) and in at least 2 of the 3 replicates were included.
Network analysis was performed using the STRING app version 1.4.268 in Cytoscape version 3.7.269, with a high confidence interaction score (0.7). Transparency and width of the edges were continuously mapped to the String score (text mining, databases, coexpression, experiments, fusion, neighborhood and cooccurrence). The Molecular COmplex DEtection (MCODE) plug-in version 1.5.170 was used to identify highly connected subclusters of proteins (degree cutoff of 2; Cluster finding: Haircut; Node score cutoff of 0.2; K-Core of 2; Max. Depth of 100). Gene ontology analysis was performed using g:Profiler web server71.
The mass spectrometry proteomics data corresponding to PML SUMO-ID, PML-TurboID and SALL1 SUMO-ID experiments have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository72 with the dataset identifiers PXD021770, PXD021809 and PXD021923, respectively.
SIM enrichment analysis
A thousand lists with the same number of proteins as PML SUMO-ID list (59) have been randomly generated from the human proteome (Uniprot filtered reviewed H. sapiens proteome, UP000005640). All those lists have been analyzed by adapting the script from73 and running it on Python version 2.7.5, to obtain the content and number of SIM motifs per protein (ψ-ψ-X-ψ; ψ-X-ψ-ψ; ψ-ψ-ψ; where ψ is either a L, I or V and X is any amino acid) and the number of SIMs per thousand of amino acids (STAA). After removing three outliers (lists 46, 782, 794; ROUT method, Q=1%), STAA values from the random lists were normalized to Log2 and validated for Gaussian distribution (d’Agostino and Pearson normality test). Enrichments were computed using R software v3.6.0 and calculated as the ratio between PML SUMO-ID STAA value and the median of STAA values from the random lists. Empirical p-values have been calculated by counting the number of random lists whose STAA values were as extreme as the PML SUMO-ID STAA value. The raw data from the SIM enrichment analysis and the script can be found in Table S5 and Supplementary Source 1, respectively.
AUTHOR CONTRIBUTIONS
O.B.-G., J.D.S. and R.B designed experiments, analyzed data and wrote the manuscript. O.B.-G., F.T., V.M., I.C., A.R.C., C.P., M.A., I.I. and J.D.S. developed experimental protocols and performed experiments. A.C., A.M.A., F.E., U.M. and A.C.O.V. provided scientific resources.
COMPETING INTERESTS
The authors declare no competing interests.
ACKNOWLEDGEMENTS
We are thankful to Iraide Escobes for her work in the Proteomics Platform at CIC bioGUNE and Arnoud de Ru in the Center for Proteomics and Metabolomics at the LUMC. RB and ACOV acknowledge funding by the grant 765445-EU (UbiCODE Program). RB acknowledges funding by grants BFU2017-84653-P (MINECO/FEDER, EU), SEV-2016-0644 (Severo Ochoa Excellence Program), SAF2017-90900-REDT (UBIRed Program) and IT1165-19 (Basque Country Government). Additional support was provided by the Department of Industry, Tourism, and Trade of the Basque Country Government (Elkartek Research Programs) and by the Innovation Technology Department of the Bizkaia County. FE is at Proteomics Platform, member of ProteoRed-ISCIII (PT13/0001/0027). FE and AMA acknowledge CIBERehd. AC acknowledges the Basque Department of education (IKERTALDE IT1106-16), the MCIU (PID2019-108787RB-I00 (FEDER/EU), Severo Ochoa Excellence Accreditation SEV-2016-0644, Excellence Networks RED2018-102769-T), the AECC (GCTRA18006CARR), La Caixa Foundation (ID 100010434), under the agreement LCF/PR/HR17, and the European Research Council (Starting Grant 336343, PoC 754627, Consolidator grant 819242). CIBERONC was co-funded with FEDER funds. UM acknowledges the Basque Government Department of Education (IT1165-19) and the Spanish MCIU (SAF2016-76898-P (FEDER/EU)).
Footnotes
Lead Contact: Rosa Barrio rbarrio{at}cicbiogune.es