The solute carrier superfamily interactome

Abstract

Solute carrier (SLC) transporters form a protein superfamily that enables transmembrane transport of diverse substrates including nutrients, ions and drugs. There are about 450 different SLCs, residing in a variety of subcellular membranes. Loss-of-function of an unusually high proportion of SLC transporters is genetically associated with a plethora of human diseases, making SLCs a rapidly emerging but challenging drug target class. Knowledge of their protein environment may elucidate the molecular basis for their functional integration with metabolic and cellular pathways and help conceive pharmacological interventions based on modulating proteostatic regulation. We aimed at obtaining a global survey of the SLC-protein interaction landscape and mapped the protein-protein interactions of 396 SLCs by interaction proteomics. We employed a functional assessment based on RNA interference of interactors in combination with measurement of protein stability and localization. As an example, we detail the role of a SLC16A6 phospho-degron and the contributions of PDZ-domain proteins LIN7C and MPP1 to the trafficking of SLC43A2. Overall, our work offers a resource for SLC-protein interactions for the scientific community.

Keywords: AP-MS; Protein–protein Interactions; Proteostasis; SLC Superfamily; Trafficking.

Figure 1. A systematic AP-MS approach to define the human solute carrier superfamily interactome.

(A) Overview of the -omics layers to characterize SLCs on a molecular level generated within the RESOLUTE project. (B) Experimental workflow for native purification of SLC-containing protein complexes from HEK 293 Jump In T-Rex cell lines. (C) Modular data analysis pipeline for processing of MS data and scoring of PPIs. (D) SLC baits grouped by protocol and MS-platform. (E) A set of 18 features (Table EV2) and a curated list of labeled PPIs served as input for the scoring. Distribution of scored interactions and scored SLC baits as preys (dark blue) versus background proteins (gray). For visualization of both distributions, the y axis was cut at a density of 10. (F) AP-MS coverage across the SLC superfamily. The left bar chart shows the coverage of SLC baits (396 of 447 SLCs investigated in RESOLUTE) and families (68 of 70 SLC families) included in the SLC interactome. The upper part of the split graph reports the counts of SLC baits used within the study separated per SLC-family. The count of SLCs (y axis) is plotted against the SLC families (x axis). SLCs used as bait in the SLC interactome are marked in blue, SLCs not included in this study or filtered after scoring are colored in yellow. The lower part of the bar chart represents the percentage coverage per SLC-family, indicating in blue the percentage of family members reported in the SLC interactome and in yellow the percent of family members which are not included. The y axis shows percent coverage by family and the x axis indicates SLC families.

Figure 2. Assessment of SLC-protein interaction network and data quality.

(A) Reported protein interactions for SLCs (n = 349, olive) versus TM-domain-containing proteins (n = 3404, teal) and proteins without annotated TM-domain (n = 12,467, gray). Lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Lower and upper whiskers extend from the hinge to the smallest or largest value no further than the 1.5× interquartile range from the hinge, respectively. The black line represents the median. Outliers were removed and to increase readability the y axis was cut at 150 PPIs. (B) PPIs per SLC reported in the PPI library. The upper part shows the associated studies per PPI, and the lower part the number of reported PPIs (gray dots) and PPIs identified within the SLC interactome (blue dots). SLCs were grouped by the associated publication into a group of SLC which were studied by interaction proteomics (dark blue, >10 referenced studies) and a group of poorly characterized SLCs (yellow, <10 associated studies or none). (C) Comparison of PPIs reported for the poorly studied (yellow) against more often studied SLCs (dark blue). For the SLC interactome no bias was observed (in literature poorly characterized, n = 332, average PPIs = 45.4, >10 associated studies, n = 64, average PPIs = 61.1, Wilcox rank-sum test P value = 0.3318, indicated as “n.s.” in the figure panel) in comparison to the literature reported database for which a statistically significant bias between SLCs was found (poorly characterized, n = 375, average PPIs = 21.8, >10 associated studies, n = 72, average PPIs = 111, Wilcox rank-sum test P value < 2.2e-16, indicated with “**” in the figure panel). Lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Lower and upper whiskers extend from the hinge to the smallest or largest value no further than the 1.5× interquartile range from the hinge, respectively. The black line represents the median and the black dots represent outliers. (D) Overlap of SLCs used as baits in the SLC interactome study (blue), in the BioPlex (yellow) and HuRI (red). (E) PPIs reported for SLCs in the BioPlex (yellow), HuRI and the SLC interactome (blue). (F) Fraction of protein interactions reported by BioPlex (yellow), HuRI (red), and the SLC interactome (blue) that were reported by additional studies. (G) PPIs reported in literature and the two large-scale reference studies, and the SLC interactome. The color indicates associated studies for reported PPIs. (H) CORUM-derived interaction pairs are enriched within the SLC interactome in comparison to 10,000 permutated networks with conserved topology and composition. The blue line indicates the overlapping PPIs with reported PPIs of deconvoluted CORUM complex found in the SLC interactome (significantly more PPIs found; P value < 1e⁻⁰⁴ compared to the permutated PPI-networks; as none of the permutated networks recovered more CORUM PPI pairs than the SLC interactome, we estimated the P value by 1/10,000 to be below 1e⁻⁰⁴). (I) Distribution of BSG and EMB, two chaperones of SLC16-family members, across the SLC interactome. The left panel shows the log2 transformed SPCs separated by scored interactions (BSG n = 11 and EMB n = 78, blue) and background (BSG n = 378 and EMB n = 4, gray) within the SLC interactome. Lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Lower and upper whiskers extend from the hinge to the smallest or largest value no further than the 1.5× interquartile range from the hinge, respectively. The black line represents the median and the black dots represent single measurements. Bars on the right side indicate how often the chaperones were scored or found as background across the SLC interactome. (J) Upper part shows for SLC-BSG (yellow) and EMB (green) interactions the log2FC against GFP (dotted black line log2FC > 1), and the lower part shows the pDockQ (dotted black line pDockQ >0.5, high-confidence structures). Complexes for which the structure was experimentally solved are marked with an asterisk and are in bold (*). Predicted SLC-chaperone structures (n = 11 BSG and n = 4 for EMB complexes) were compared against a set of predicted structures of SLC-chaperones for which the chaperones were classified as background (n = 11 BSG and n = 4 for EMB complexes; unpaired Student t test for BSG P value = 0.001589 and EMB with a P value = 0.02762; independent control sets and tests). In the figure, P values below 0.01 are marked with “**”, and P values between 0.01 and 0.05 are marked with “*”. Lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Lower and upper whiskers extend from the hinge to the smallest or largest value no further than the 1.5× interquartile range from the hinge, respectively. Black line represents the median and the black dots represent scores per complex. On the right side predicted structures for SLC16A7-BSG and SLC16A1-EMB complexes are shown.

Figure 3. Clustering SLCs by their interactome similarity.

(A) Dendrogram of hierarchical clustering based on the Jaccard similarity matrix derived from the interactome profiles of 396 SLCs. The heatmap displays the log ratio of the top-level biological process-enriched pathways. Clusters with a significantly enriched SLC functional property (Fisher’s test P < 0.2) are shown in bold and the respective cluster color. The outer ring shows representative parental GO terms significantly enriched within the cluster (P value < 0.01). (B) Distribution of GO semantic similarities between SLC similarity clusters. A similarity threshold of 0.65 (dashed black line) between the clusters was chosen to filter the data. (C) Overlap of protein interactors identified among the SLCs from cluster 31 and cluster 35.

Figure 4. Proteostatic regulation of SLCs.

(A) Generation of cell lines conditionally expressing GFP-tagged SLCs together with an RFP reporter modified with membrane-specific tags or proteins that allow localization to biologically relevant membranes. (B) Protein stability and subcellular localization assay to validate SLC interactions. Selected SLCs were conditionally expressed and monitored using flow cytometry and microscopy. GFP:RFP ratio and GFP localization and intensity were measured after perturbation of interactor abundance or drug treatments. (C) SLC-protein stability after drug treatment as measured by flow cytometry showed distinct susceptibility to degradation pathways. Heatmap of log2 transformed GFP:RFP ratios measured after 24 h of induction and drug treatment in the last 6 h (normalized to DMSO control; n = 3 wells per sample). (D) Median GFP:RFP ratios for 37 out of 135 tested SLC-protein interactions that were found to be significantly different from control treatment (n = 4 wells per condition, P value < 0.01, unpaired independent t test; GFP:RFP and GFP:GFP ratios compared to control treatment changed at least 10%; see “Methods” for details and Dataset EV7 for exact P values). Bars and error bars represent the mean of the median GFP:RFP ratios and the 95% confidence intervals. (E) Results for 9 SLC-protein interactions after interactor cDNA overexpression out of 18 tested interactions analyzed as in (D). (F) KLHL36 and GOLGA5 were strongly enriched in SLC6A15 purifications compared to other SLCs or GFP (n = 2 biologically independent replicates). In flow cytometry experiments, SLC6A15-GFP was destabilized by overexpression of the ubiquitin ligase adapter KLHL36 (n > 4600 events per sample). (G) MTCH1 was destabilized by E3 ligase MUL1 and stabilized by deubiquitinase USP30. In flow cytometry experiments, MTCH1-GFP protein stability directly correlated with depletion by RNAi or cDNA overexpression of its interactors and their biological functions in the ubiquitin-dependent degradation pathway (n > 2500 events per sample). Asterisk and gray figure label of cDNA USP30 condition indicates an effect that was below the set threshold of 10% change in GFP:RFP ratio. See also Fig. EV4.

Figure 5. SLC16A6 stability is regulated by a phospho-degron.

(A) SLC16A6 interacts with subunits of E3 ubiquitin ligase SCF. SLC16A6 colored in beige, interactors in green. SKP1 (gray) was quantified but not scored and added to the network. Novel edges depicted by solid black lines and edges obtained from literature in black dotted lines. (B) Protein stability of SLC16A6 after RNAi treatment of adapter proteins showing strong stabilization after co-depletion of BTRC and FBXW11 (mean ± SD; P values were calculated using one-way ANOVA and Dunnett test; n = 4 replicates). (C) Alphafold predicted structure of SLC16A6 with highlighted cytosolic loop containing the phospho-degron. The βTrCP consensus motif and SLC16A6 phospho-degron motif are visualized on the right side. (D) Protein stability assay of SLC16A6 phosphomutants compared to SLC16A6^WT (mean ± SD; P values were calculated using one-way ANOVA and Dunnett test; n = 4 replicates). (E) Quantitative comparison of SCF complex subunits across SLC16A6 phosphomutants (x axis) against SLC16A6^WT. The log2FC (y axis) were derived for the 31 interaction partners against their abundance in SLC16A6^WT. The dashed black line indicates a ± 1 log2FC threshold. Labeled dots represent SCF complex members (n = 4). (F) PPI-Network of SLC16A6 phospho-mutant interactions with SCF complex subunits. The color and thickness of edges indicate the mean log2FC against SLC16A6^WT. For all AP-MS data shown in the figure panels, n = 2 biologically independent replicates with n = 2 technical injections were used.

Figure 6. Proteins affecting trafficking and subcellular localization of SLCs.

(A) Relative GFP intensity changes at reference regions after interactor RNAi compared to control RNAi are plotted as a percentage (n = 160 images per condition; error bars denote 95% confidence intervals). The graph includes SLCs residing at single subcellular locations and SLC30A7 (since vesicles and Golgi locations are both covered by expression of cell body RFP). GFP intensity changes at the respective reference signal were thresholded at 10% increase or decrease over control treatment with P value < 0.01 (independent t test). Subcellular compartment location of SLC-GFP and reference RFP as well as exact P values are indicated in Dataset EV7. (B) Representative images of plasma membrane-localized SLC1A2-GFP after depletion of three interactors (scale bar, 10 µm). GFP signal intensity was quantified using myr-mRuby3 to identify plasma membrane pixels. (C) Quantification of cellular impedance of HEK 293 SLC1A2-SH cells treated with increasing concentrations of glutamate after GDPD1 RNAi (n = 3 independent experiments). Data were corrected against samples without doxycycline due to significant l-glutamine transport by endogenous SLC1A3 in HEK 293 cells and are shown as background corrected mean area under the curve (AUC) quantification relative to control RNAi ±SEM. P value was calculated using unpaired t test. (D) Relative GFP intensity changes at the reference region (purple) and outside the reference region (teal) for 34 SLC-interactor RNAi combinations for SLC-GFP signals covering more than one subcellular compartment (as indicated in Dataset EV7; n = 160 images per condition; bars denote mean ± 95% confidence intervals). The graph includes interactions with GFP intensity changes exceeding 10% at the reference region or 25% outside the reference region with P values < 0.01 compared to control RNAi (independent t test). For 13 interactions highlighted in bold, GFP intensity changes at reference and non-reference regions differed significantly (P values as indicated, independent t test with Benjamini–Hochberg correction for multiple tests; relative change at least 33.3% of absolute higher value), indicating location-specific changes. GFP intensity changes for other interactions were similar for reference and non-reference regions, indicating general changes in SLC abundance after interactor depletion. (E) Relative fluorescence unit (RFU) curve to measure cadmium uptake of SLC39A8 after RNAi against GOLM1. Traces represent mean min-max normalized RFU values across replicates ±SD (curve shades; n = 3 biologically independent replicates with 8 technical replicates). Results are compared to control RNAi using mean AUC calculated from normalized RFU traces considering all replicates (Appendix Fig. S15). (F) SLC43A2 interactions with trafficking proteins CASK, DLG1, LIN7C, and MPP1 (teal; n = 2 biological replicates with n = 2 technical replicates). Novel edges are depicted by solid lines and edges supplemented from BioGRID are shown with dotted line. (G) AlphaFold model of SLC43A2–LIN7C complex with section of the interaction of the LIN7C PDZ domain and the intracellular C-terminal tail of SLC43A2 (Weighted Score 0.57). (H) AlphaFold model of SLC43A2–MPP1 interaction between the MPP1 PDZ domain and the intracellular C-terminal tail of SLC43A2 (Weighted Score 0.51). (I) Representative images of plasma membrane and vesicle localized SLC43A2 after depletion of three interactors (scale bar, 10 µm). Depletion of CASK or LIN7C increased plasma membrane-associated GFP signal whereas RNAi of MPP1 increased the intensity and relative distribution of SLC43A2-GFP away from the plasma membrane (quantified in (D)). (J) Schematic of SLC43A2 trafficking regulation.

Figure EV1. Characterization of SLC-protein interactions, SLCs as background and SLCs in complexes.

(A) SLCs and PPIs reported across protein interaction databases. (B) Overlap of SLCs and PPIs across protein interaction databases. (C) Reported PPIs across the PPI library were plotted against frequency of identification in the CRAPome database. 127 SLCs were reported in the PPI library and were also part of CRAPome (CRAPome frequency >20% in violet, <20% in gray). (D) Comparison of the share of PPIs reported in the prey role for SLCs in BioGRID with a CRAPome frequency >20% (n = 11) against the share of PPIs reported in the prey role for SLCs with a CRAPome frequency <20% (n = 75). Data are presented as mean ± SD. The mean of the two groups showed a significant difference (two-sample t test, P value: 0.005725). In the figure panel, P values below 0.01 are indicated with “**”. (E) Count of protein complexes (upper part) and protein complex subunits (lower part) included in CORUM. Subunits/ complexes were grouped into SLC, TM protein and no-TM protein and were counted across all reported protein complexes, taking into consideration multiple occurrences of subunits. (F) GALNT2 signal distribution across the SLC interactome. The left panel displays log2 transformed spectral counts for each SLC AP-MS experiment, grouped by scored interactions (n = 21, blue) and background/non-interacting (n = 230, gray). Lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Lower and upper whiskers extend from the hinge to the smallest or largest value no further than the 1.5× interquartile range from the hinge, respectively. Black line represents the median log2 spectral count signal, and the black dots represent the signal per measurement. The right side shows the frequency of GALNT2 identification, scoring, and background in percentages. (G) The upper bar chart shows all SLCs for which GALNT2 was scored within the SLC interactome (log2FC against GFP, threshold of log2FC > 1). The lower section of the bar chart reports the confidence scores of predicted GALNT2-SLC complexes (high confidence, pDockQ threshold of >0.5 dashed black line, medium confidence, pDockQ threshold of >0.25 represented by a dashed gray line). A comparison against randomly sampled SLC-GALNT2 interaction (see “Methods” for details) showed, a significant difference between models of the interactions covered in the SLC interactome (n = 21) and the control set (n = 21; unpaired Student t test, P value = 0.00695). In the figure panel, P values below 0.01 are indicated with “**”. Lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Lower and upper whiskers extend from the hinge to the smallest or largest value no further than the 1.5× interquartile range from the hinge, respectively. Black line represents the median and the black dots represent scores per complex. (H) The AlphaFold model of SLC30A1-GALNT2 is shown on the right side.

Figure EV2. Novel versus known protein interactions for each clade of the structural and evolutionary-based SLCome.

For the construction of the phyologenetic tree the distance matrix of a previous classification of the SLC superfamily based on structural models was used (Ferrada and Superti‑Furga, 2022). Each of the 25 distinct structural clades are represented in a different color. For each clade, the count of novel PPIs (light blue), literature mined PPIs (olive), and the shared PPIs (dark red) are reported. The 405 SLCs included in the study are colored light blue and all SLCs which were not included in the SLC interactome are colored gray.

Figure EV3. Functional property enrichment analysis showed SLC interactome similarity for SLC12 and SLC6 family members.

(A) Significantly enriched SLC functional properties identified in the interactome SLC clustering analysis (Fisher’s test P < 0.2). Colors indicate different SLC functional properties. (B) Sum of enriched SLC functional properties separated by SLC properties. (C) Significantly enriched SLC functional properties for PPI-interactome profile cluster 30 (Fisher’s test P < 0.2). (D) Cluster-specific PPI-network obtained by the SLC baits (yellow octagons) grouped to cluster 30. Interactors are shown in green, and interactions between SLC baits are highlighted with a thick black line. (E) GO biological processing terms significantly enriched (Gene ontology overrepresentation test with a hypergeometric model, Benjamini–Hochberg corrected, P-adjusted <0.01) obtained by GSEA for interactors present in the specific PPI-network. The obtained terms were converted to a similarity network with GO semantic similarity and further filtered to showcase the least overlapping significant terms (similarity cutoff of 0.7). (F) The 25 most connected interactors (hub score) in the cluster 30-specific PPI-network. (G) PPI-network retrieved from literature (STRING confidence score >0.4, physical interactions only) for the 25 most connected interactors within cluster 30. Interactors were grouped as HSP90 chaperones (green), HSP90 co-chaperones/interactors (teal), HSP70 chaperones (blue), HSP70 co-chaperones/interactors (cyan), HSP1 interactors (red) or trafficking associated proteins (yellow). The four interactors not connected were removed from the PPI-network. (H) Distribution of protein length of SLCs interacting with HSP90AA1 (n = 6) and SLCs for which HSP90AA1 was found in the background (n = 379). Comparison showed a significant difference in the protein length (Student t test P value = 1.285e⁻⁰⁸). In the figure panel, P values below 0.01 are marked with “**”. Lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Lower and upper whiskers extend from the hinge to the smallest or largest value no further than the 1.5× interquartile range from the hinge, respectively. Black line represents the median SLC-protein sequence length, and the black dots represent the sequence length per SLC for which HSP90AA1 was identified. (I) Correlation of the chaperone/ chaperone interacting protein abundances with the summed SLC tail length (N-terminal and C-terminal). Significant correlations with a P value < 0.01 are indicated with black ring.

Figure EV4. SLC levels are regulated by protein stability.

(A) Venn diagram of drug treatment effects on SLC-protein stability. SLCs were considered to be stabilized by the respective drug if the GFP:RFP ratio was increased more than 10% compared to mock treated cells (P value < 0.05, independent t test). (B) SLC-protein interactions of KLHL36. The protein was not quantified in the background of any other SLC purification. (C) Western blot result of endogenous SLC6A15 in HEK 293T cells showing SLC6A15 degradation after overexpression of FLAG-tagged KLHL36. The samples shown in the blot were derived from the same experiment and gels/blots were processed in parallel. Uncropped images are provided as source data. (D) MTCH2 interactome showing 7 distinct interactors that were assessed in detail. (E) HAP1 cells expressing endogenously HA-tagged MTCH2 were transfected with RNAi against MTCH2 interactors. Depletion of the E3 ligase proteins MARCHF5 and MUL1 stabilizes MTCH2 levels as strongly as ubiquitination inhibition by TAK-243. (F) Immunofluorescence example images of HA-MTCH2 and quantification at mitochondria as identified by staining for AIF (apoptosis-inducing factor). Over 10,000 cells were imaged per condition; unpaired t test was used to compare treatments (n = 112 images per sample). Lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Lower and upper whiskers extend from the hinge to the smallest or largest value no further than the 1.5× interquartile range from the hinge, respectively. Black line represents the mean and the black dots represent outliers. Source data are available online for this figure.

Figure EV5. Proteins affecting trafficking and subcellular localization of SLCs.

(A) GSEA of cellular compartments was performed for each SLC interactome and the resulting GO terms were hierarchically assigned to 10 main compartments (P value cutoff 0.05; see “Methods” for details). (B) A majority of SLC interactomes contain interaction partners from several subcellular compartments (mean: 4.5 compartments per SLC). (C) This notion also holds true for a subset of SLCs described previously as uniquely plasma membrane-associated (mean: 4.3 compartments). (D) The results from the interactome–compartment enrichment analysis were used to cluster SLCs. Strong clusters can be identified for some mitochondria baits (e.g., MTCH2), lysosomal baits (e.g., MFSD12) or Golgi baits (e.g., SLC30A5). (E) Representative images of plasma membrane- and Golgi-associated SLC35F2 after RNAi (see Fig. 6D for a quantification of the effect; size bar, 10 µm for all images). (F) SNX2 depletion attenuates SLC35F2-mediated import of YM-155. After 48 h RNAi, SLC35F2-GFP expression was induced using doxycycline and cells were subjected to YM-155 at the indicated concentrations for 24 h. Cell viability was measured using CellTiter-Glo luminescence (n = 3, mean ± SEM). Induced conditions at 1 µM YM-155 were compared using unpaired t test. (G) Representative images of plasma membrane- and Golgi-associated SLC22A11-GFP. Depletion of endopeptidase FURIN increased the plasma membrane-associated GFP intensity but not the Golgi-associated GFP intensity. (H) Representative images of plasma membrane- and lysosome-associated SLC23A1-GFP. Whereas depletion of GalNAc transferase GALNT2 increased GFP signals overall, depletion of SYVN1 led to a pronounced increase of GFP overlapping with the lysosomal RFP reference (see Fig. 6D for quantifications). (I) Representative images of plasma membrane- and vesicle-associated SLC31A1-GFP. Depletion of syntaxin-7 increased overall GFP signal while depletion of WDR6 increased GFP intensity at the plasma membrane (see Fig. 6D for quantifications).