Abstract
The α-arrestins form a large family of evolutionally conserved modulators that control diverse signaling pathways, including both G-protein-coupled receptor (GPCR-) mediated and non-GPCR mediated pathways, across eukaryotes. However, unlike β-arrestins, only a few α-arrestin targets and functions have been characterized. Here, using affinity purification and mass spectrometry, we constructed interactomes for six human and twelve Drosophila α arrestins. The resulting high-confidence interactomes comprised 307 and 467 prey proteins in human and Drosophila, respectively. A comparative analysis of these interactomes predicted not only conserved binding partners, such as motor proteins, proteases, ubiquitin ligases, RNA splicing factors, and GTPase-activating proteins, but also those specific to mammals, such as histone modifiers and the subunits of V-type ATPase. Given the manifestation of the interaction between the human α-arrestin, TXNIP, and the histone-modifying enzymes, including HDAC2, we undertook a global analysis of transcription signals and chromatin structures that were affected by TXNIP knockdown. We found that TXNIP activated targets by blocking HDAC2 recruitment to targets, a result that was validated by chromatin immunoprecipitation assays. Additionally, the interactome for an uncharacterized human α- arrestin ARRDC5 uncovered multiple components in the V-type ATPase, which plays a key role in bone resorption by osteoclasts. Our study presents conserved and species-specific protein-protein interaction maps for α-arrestins, which provide a valuable resource for interrogating their cellular functions for both basic and clinical research.
eLife assessment
This valuable study provides a comparative interactome analysis of α-arrestin in human and Drosophila. Using convincing methodology that includes affinity purification/mass spectrometry, bioinformatic tools and experimental data in human cells, the authors identify biological roles of protein-protein interactions (PPI) involving α-arrestin in the two species. This study will serve as a broadly relevant resource for understanding the PPI network of human and Drosophila α-arrestins.
Introduction
The discovery of first arrestin protein in retinal rods contributed to a deeper understanding of photoreceptor signaling mediated by rhodopsin, which is one of the G protein-coupled receptor (GPCR) class. As its ability to arrest the GPCR signaling pathway, the protein was first named as “arrestin” (Kuhn, Hall, & Wilden, 1984; Wilden, Wust, Weyand, & Kuhn, 1986; Zuckerman & Cheasty, 1986). Shortly after this discovery of the first arrestin protein in the retina, another arrestin protein that specifically turns off β-adrenergic signaling was identified and named “β-arrestin”. This arrestin-mediated termination of signaling from GPCRs is called “receptor desensitization” (Benovic, DeBlasi, Stone, Caron, & Lefkowitz, 1989; Lohse, 1992; Shenoy & Lefkowitz, 2011), one of crucial cellular process in maintaining cellular homeostasis and preventing overstimulation of signaling pathways. Further studies have revealed that β-arrestins regulate the receptor desensitization of other signaling pathways through ubiquitination and regulation of trafficking of various cargo molecules (Y. M. Kim & Benovic, 2002; Malik & Marchese, 2010; Puca & Brou, 2014).
Another class of arrestin, α-arrestin, was first studied in fungi and yeast (Andoh, Hirata, & Kikuchi, 2002) and subsequently recognized as new class of arrestins (Boase & Kelly, 2004; Herranz et al., 2005). They contain characteristic arrestin domains, arrestin_N and arrestin_C, and PPXY motifs, which are unique to the α-arrestin clan. A phylogenetic study of arrestin proteins showed that α-arrestins are the ancestral class of the arrestin family and conserved from yeast to human (Alvarez, 2008). To date, six α-arrestins, arrestin domain containing protein 1 (ARRDC1), ARRDC2, ARRDC3, ARRDC4, ARRDC5, and thioredoxin-interacting protein (TXNIP), have been found to be in the human genome. These human α-arrestins were first studied in conjunction with β-arrestins in the regulation of the β2-adrenergic receptor (β2AR) in human cells. ARRDC3 and ARRDC4 works as an adaptor protein for the ubiquitination of β2AR by recruiting the NEDD4 protein, an E3 ubiquitin ligase, through its conserved PPXY motifs(S. O. Han, Kommaddi, & Shenoy, 2013; Nabhan, Pan, & Lu, 2010; Shea, Rowell, Li, Chang, & Alvarez, 2012).In addition to their β2AR-associated roles, α- arrestins are involved in trafficking and sorting of other GPCRs and signaling molecules through post-translational modifications, including ubiquitination. For example, ARRDC1 and ARRDC3 were reported to play roles in the degradation of the Notch receptor (Puca & Brou, 2014) and in the ubiquitination of ALG-2-interacting protein X (ALIX) (Dores, Lin, N, Mendez, & Trejo, 2015). ARRDC1 contains a PSPA motif, which binds the tumor susceptibility gene 101 (TSG101) protein, an essential component of an endosomal sorting complex. ARRDC1 also recruits E3 ligases, such as WW domain-containing E3 ubiquitin protein ligase2 (WWP2), inducing ubiquitination of itself and the subsequent release of ARRDC1-associated microvesicles (Nabhan, Hu, Oh, Cohen, & Lu, 2012). Another well-known α-arrestin, TXNIP, was first named as vitamin D3-upregulated protein 1 (VDUP1) after verification that its gene is a vitamin D3 target in cancer cells (K. S. Chen & DeLuca, 1994; Qayyum, Haseeb, Kim, & Choi, 2021). Since then, TXNIP had been reported to directly interact with thioredoxin, which is an essential component of the cellular redox system, to inhibit its activity as an antioxidant (Junn et al., 2000; Nishiyama et al., 1999; Patwari, Higgins, Chutkow, Yoshioka, & Lee, 2006). TXNIP was also reported to inhibit glucose uptake by inducing the internalization of glucose transporter 1 (GLUT1) through clathrin-mediated endocytosis and by indirectly reducing GLUT1 RNA levels (Wu et al., 2013). Although the TXNIP is known to be localized in both cytoplasm and nucleus, biological functions of TXNIP have been mostly explored in cytoplasm but remained poorly characterized in nucleus.
A few α-arrestins appear to have evolutionarily conserved functions in both human and invertebrates. For instance, the Hippo signaling pathway, which impacts a variety of cellular processes such as metabolism, development, and tumor progression (Mo, Park, & Guan, 2014; Pei et al., 2015; Schutte et al., 2014; Y. Wang et al., 2010; Zhi, Zhao, Zhou, Liu, & Chen, 2012), was shown to be regulated by α-arrestin in both Drosophila (Y. Kwon et al., 2013) and human cells (Xiao et al., 2018). In Drosophila, the protein Leash was identified as an α-arrestin and shown to down-regulate Yki by promoting its lysosomal degradation, leading to a restriction in growth (Y. Kwon et al., 2013). In human cells, ARRDC1 and ARRDC3 were shown to induce degradation of the mammalian homolog of Yki, YAP1, by recruiting the E3 ubiquitin ligase ITCH in renal cell carcinoma (Xiao et al., 2018), suggesting functional homology between human and Drosophila. However, because the α-arrestins interact with multiple targets, an unbiased, comparative analysis of interactome is required to determine whether other α-arrestin from human and Drosophila have common and specific interacting partners, which will determine their functional homology and diversification.
A comprehensive understanding of their protein-protein interactions (PPIs) and interactomes will shed light on the underlying molecular mechanisms, reveal novel regulatory axes, and enable the identification of previously unrecognized roles of α-arrestin in cellular processes. Furthermore, extensive characterization of the α-arrestin interactome may help uncover potential therapeutic targets and provide valuable insights into the treatment of diseases associated with dysregulated signaling pathways (Diaz et al., 2005; Lu, Simin, Khan, & Mercurio, 2008; Q. Y. Wang et al., 2018; Zhou, Tardivel, Thorens, Choi, & Tschopp, 2010)
In this study, we conducted affinity purification/mass spectrometry (AP/MS) of six human and twelve Drosophila α-arrestins. A high-confidence PPI network was constructed by selecting a cut-off for receiver operating characteristic (ROC) curves of Significance Analysis of INTeractome express (SAINTexpress) scores (Teo et al., 2014). The constructed interactomes were validated using known affinities between domains of prey proteins and the short linear motifs of α-arrestins. We also investigated orthologous relationships between binding partners from human and Drosophila and found that many proteins with both known and novel functions could be conserved between two species. Finally, we performed experiments to provide new insights into the functions of TXNIP and ARRDC5 that were revealed in our study. Together, our results provide a valuable resource that describes the PPI network for α-arrestins in both human and Drosophila and suggest novel regulatory axes of α-arrestins.
Results
High-confidence α-arrestin interactomes in human and Drosophila
Genome-scale sets of prey proteins interacting with α-arrestins (referred to herein as ‘interactomes’) were compiled by conducting AP/MS for six human and twelve Drosophila α- arrestin proteins (Figure 1A; Table S1). Proteins possibly interacting with α-arrestins were pulled down from total cell lysates of human embryonic kidney 293 (HEK293) and S2R+ cells stably expressing GFP-tagged α-arrestins (Figure 1B; Figure S1). All α-arrestin experiments were replicated twice, and negative control experiments were conducted multiple times. In total, 3,243 and 2,889 prey proteins involved in 9,908 and 13,073 PPIs with human and Drosophila α-arrestins, respectively, were initially detected through AP/MS (Figure 1B; Table S2).
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_fig1.tif/full/max/0/default.jpg)
Identification of high-confidence α-arrestin PPIs
(A) Phylogenetic tree of α-arrestins from human (6, top) and Drosophila (12, bottom) based on protein sequences. The numbers in parentheses indicate the length of each protein. aa, amino acids; Arr_N: Arrestin N domain; Arr_C: Arrestin C domain; PPxY: PPxY motif. (B) Shown is a schematic flow of AP/MS experiments and computational analysis. (C) ROC curves of SAINTexpress scores along with AUC values. The arrows point to the cutoff scores used in subsequent studies in human (left) and Drosophila (right). (D) (Top) The fraction of high-confidence and all raw (unfiltered) PPIs that are supported by known affinities between short linear motifs and protein domains in human (left) and Drosophila (right). One-sided, Fisher’s exact test was performed to test the significance. (Bottom) The sum of log2 spectral counts (log2 spec) of interacting proteins with WW domains observed in the high confidence and all raw PPIs are visualized in the heatmap. (E) The α-arrestins and interacting prey proteins were hierarchically clustered based on the log2 mean spectral counts and summarized for human (top) and Drosophila (bottom) in the heatmaps. The functionally enriched protein groups of preys are indicated at the top. Previously reported proteins interacting with α-arrestins are labeled at the bottom. On the right, the functional composition of prey groups is summarized with the sum of log2 mean spectral counts of each prey group, which are colored to correspond with the labels on the left.
To build high-confidence interactomes of α-arrestin family proteins, a probabilistic score for individual PPIs was estimated using SAINTexpress (Teo et al., 2014) and an optimal cutoff for the scores was set using positive and negative PPIs of α-arrestin from public databases and the literature (Colland et al., 2004; Dotimas et al., 2016; Draheim et al., 2010; Mellacheruvu et al., 2013; Nabhan et al., 2012; Nishinaka et al., 2004; Puca & Brou, 2014; Szklarczyk et al., 2015; Warde-Farley et al., 2010; Wu et al., 2013) (Tables S2). The resulting ROC curves showed high area under curve (AUC) values and the SAINTexpress scores at which the false discovery rate (FDR) was 0.01 were selected as cutoffs (0.85 for human and 0.88 for Drosophila, Figure 1C). Given the cutoffs, 1,306 and 1,732 PPIs involving 902 and 1,732 proteins were selected for human and Drosophila, respectively. Because proteins of low abundance (low spectral counts) are easily affected by a stochastic process (Lundgren, Hwang, Wu, & Han, 2010; Old et al., 2005), the minimum spectral count of PPIs was set at 6, allowing us to select PPIs with higher confidence. In fact, the spectral counts of the filtered PPIs were highly reproducible between replicates (Figure S2A; Pearson’s correlations, 0.91 for human; 0.89 for Drosophila) and principal component analysis based on log2 spectral counts also confirmed a high reproducibility between replicates (Figure S2B). Moreover, we successfully detected many known interaction partners of α-arrestins such as NEDD4, WWP2, WWP1, ITCH and TSG101, which have been previously reported in the literature and PPI databases (Figure S2C) (Colland et al., 2004; Dotimas et al., 2016; Draheim et al., 2010; Mellacheruvu et al., 2013; Nabhan et al., 2012; Nishinaka et al., 2004; Puca & Brou, 2014; Szklarczyk et al., 2015; Warde-Farley et al., 2010; Wu et al., 2013). Finally, our filtered interactomes of human and Drosophila α-arrestins, comprised of 390 PPIs between six α- arrestins and 307 prey proteins in human, and 740 PPIs between twelve α-arrestins and 467 prey proteins in Drosophila, are hereafter referred to as ‘high-confidence PPIs’ (Table S3).
Short-linear motifs and protein domains enriched in α-arrestins and their interactomes
To validify our high-confidence PPIs, we sought to analyze known short-linear motifs in α-arrestins, which are commonly 3-15 stretches of amino acids that are known to participate in interactions with other protein domains (Dinkel et al., 2015). Utilizing the known affinities between short linear motifs in α-arrestins and protein domains in interactomes from eukaryotic linear motif (ELM) database (Dinkel et al., 2015), we evaluated whether our high-confidence PPIs could be explained by the known affinities between them (Table S4). The fractions of our high-confidence PPIs (green, Figure 1D top), supported by the known affinities were significantly greater than those of all raw PPIs (red, Figure 1D top) in both species (P < 9.37 X 10-11 for human and P < 0.0012 for Drosophila, one-sided Fisher’s exact test, Figure 1D top). One of the most well-known short-linear motifs in α-arrestin is PPxY, which is reported to bind with high affinity to the WW domain found in various proteins, including ubiquitin ligases (Macias et al., 1996). Our analysis revealed the specific enrichment of WW domain-containing proteins in the interactomes of α-arrestins with at least one PPxY motif but not in that of the human α-arrestin (ARRDC5) without a PPxY motif (Figure 1D, bottom-left). The interactomes of five out of the eight Drosophila α-arrestins with a PPxY motif were enriched for WW domain containing proteins, but there was no such enrichment for any of the Drosophila α-arrestins without a PPxY motif (Figure 1D, bottom-right). In conclusion, a considerable portion of the high-confidence PPIs identified in this study can be evident by known affinities between short linear motifs and protein domains.
Next, we conducted enrichment analyses of proteins domains among interactome of each arrestin to investigate known and novel protein domains commonly or specifically interacting with α-arrestins (Figure S3A; Table S5). The most prominent interacting domains in both species were the Homologous to E6AP C-terminus (HECT), WW, and C2 domains (FigureS3A; Table S5). HECT and C3 domains are well known to be embedded in the E3 ubiquitin ligases such as NEDD4, HECW2, and ITCH along with WW domains (Weber, Polo, & Maspero, 2019) and as we observed strong preference of WW domains to PPxY containing proteins (Figure 1D), these domains were significantly enriched in binding proteins of α- arrestins with PPxY motif in human and Drosophila (FDR < 0.033 ∼ 1.23 X 10-11 for human; FDR < 0.045 ∼ 4.10 X 10-6 for Drosophila, Figure S3A ; Table S5). Other common protein domains involved in the protein degradation process, such as proteasome domains, were also significantly enriched in the interactomes (of ARRDC4 in human and Leash in Drosophila) in both species (FDR < 6.41 X 10-4 for human and FDR < 1.30 X 10-5 for Drosophila, Figure S3A ; Table S5). Interestingly, some α-arrestins (ARRDC3 in human and Vdup1, Leash, and CG18746 in Drosophila) appeared to interact in common with RNA binding domains, such as DEAD box, helicase, WD40, and RNA recognition motif, but others did not. In addition, the cargo and motor protein domains IBN_N (FDR < 0.0076 for human and FDR < 2.50 X 10-4 ∼ 2.11 X 10-6 for Drosophila) and myosin_head (FDR < 0.033 for human and FDR < 2.11 X 10- 6 for Drosophila) also interacted with several α-arrestins in common (ARRDC4 in human and CG1105, CG18745, and CG18748 in Drosophila, Figure S3A; Table S5). These enriched domains explain the conserved interactomes associated with RNA splicing and protein transport in both species. In addition, human α-arrestins seem to interact with human-specific domains, such as PDZ, Rho-GEF, MCM, laminin, zinc finger, and BAG6 domains, providing an expanded interactome of human α-arrestins (Figure S3A; domains in black), indicating the presence of both conserved and specific protein domains interacting with α-arrestins.
Expanded functional signatures of α-arrestin interactomes
Because the functions of α-arrestins can be inferred based on their binding partners, the prey proteins were grouped based on their interactions with α-arrestins, which revealed specialized functions of the respective α-arrestins with some redundancy as well as both known and novel functions (Figure 1E). The analysis of protein class enrichment by the PANTHER classification system (Thomas et al., 2003) revealed previously reported functions, such as ‘Ubiquitin ligase’ (FDR < 0.0019 and 5.01 X 10-7 for human; Benjamini-Hochberg correction) and ‘Protease’ (FDR < 1.93 X 10-6 for human and 5.02 X 10-6 for Drosophila) (Dores et al., 2015; Y. Kwon et al., 2013; Nabhan et al., 2012; Puca, Chastagner, Meas-Yedid, Israel, & Brou, 2013; Rauch & Martin-Serrano, 2011; Shea et al., 2012; Xiao et al., 2018). In fact, the known binding partners, NEDD4, WWP2, WWP1, and ITCH in human and CG42797, Su(dx), Nedd4, Yki, Smurf, and HERC2 in Drosophila, are related to ubiquitin ligases and protein degradation (Figure 1E; Figure S2C). In addition, novel biological functions of α-arrestins were uncovered. For instance, in human, prey proteins interacting with ARRDC3 displayed enrichment of ‘RNA splicing factor and helicase’ functions as well as ‘GTPase-activating proteins’, those of ARRDC4 were enriched with ‘Apolipoprotein’, and those of ARRDC5 with ‘ATP synthase’ (Figure 1E, up). Motor protein, protease, ubiquitin ligase, RNA splicing factor, and helicase were functions that were also enriched in Drosophila prey proteins (Figure 1E, bottom). Among them, the motor protein and RNA splicing, and helicase functions seemed to be novel conserved functions between human and Drosophila. The functional compositions of the interacting proteins summarized the common or highly specialized functions of α-arrestins well (Figure 1E, right panel). For example, in human, proteins that interacted with TXNIP, ARRDC2, and ARRDC4 showed similar ubiquitination and protease-related functions, whereas ARRDC3 and ARRDC5 displayed unique interactomes associated with other functions. For Drosophila, the interactomes of the [Vdup1, CG10086 and CG18744], [CG18748 and CG18747], and [CG1105 and CG14696] α-arrestin subsets each exhibited similar functional compositions, but the Leash interactome showed a distinct enrichment of ubiquitination-related and protease functions. Taken together, these results suggest that the resulting high-confidence PPIs of α-arrestins expanded the functional interactome maps of α- arrestins in both human and Drosophila.
Subcellular localizations of α-arrestin interactomes
Cellular localizations of proteins often provide valuable information of their functions and activity, but only a small number of α-arrestins are known for their preferential subcellular localization. We thus examined the subcellular localizations of the interacting proteins using the cellular component feature in Gene Ontology (GO) using DAVID (Huang da, Sherman, & Lempicki, 2009a, 2009b) (Figure S3B; Table S6). Prey proteins (246 for human and 245 for Drosophila) that were localized in at least one cellular compartment were examined. We found that prey proteins of ARRDC5 were preferentially localized in the endoplasmic reticulum and at the plasma membrane (PM) but were less often localized in the nucleus, compared to those of other human α-arrestins (Figure S3B, left). Similarly, the prey proteins of ARRDC1 and 4 were less often localized in the nucleus, instead being preferentially localized in the cytoplasm (ARRDC4) or extracellular space (ARRDC1), in agreement with previous reports (Nabhan et al., 2012; Q. Y. Wang et al., 2018). TXNIP seemed to preferentially interact with prey proteins in cytoplasm and nucleus (Figure S3B, left), consistent with a previous report (S. K. Kim, Choe, & Park, 2019; Saxena, Chen, & Shalev, 2010). ARRDC3, which was suggested to be localized in cytoplasm in previous study (Nabhan et al., 2010), appeared to interact with proteins preferentially localized in nucleus in addition to the ones in cytoplasm, implying novel functions of ARRDC3 in the nucleus. In Drosophila, the localization of interacting proteins is often uncharacterized compared to human, but a preference for a localization for part of the interactomes can be observed (Figure S3B, right). Some of them are preferentially localized at the PM (CG18747), mitochondria (CG14696), peroxisome (CG14696), lysosome (CG2641), or cytoskeleton (CG18748), compared to others. However, interactomes of Leash, Vdup1, CG2641, CG18745, CG18746, and CG10086 are preferentially localized in the nucleus. Taken together, these data about the preferential localizations of interacting proteins provide evidence about the functions and activity of α-arrestins in cells.
Functional complexes in α-arrestin interactomes
The fact that protein functions are often realized in complexes (Hartwell, Hopfield, Leibler, & Murray, 1999) urged us to search for functional complexes that extensively interact with α-arrestins. For this analysis, protein complexes that are significantly connected with each α-arrestin were examined using the COMPlex Enrichment Analysis Tool (COMPLEAT) (Vinayagam et al., 2013), resulting in the detection of 99 and 18 protein complexes for human and Drosophila, respectively (Table S7). The complexes were iteratively combined with cellular components from GO (Table S7) based on the overlap coefficients (Vijaymeena & Kavitha, 2016). The significance of the resulting combined complexes was then tested with the connectivity to each α-arrestin using the interquartile means (IQMs) of SAINTexpress scores compared to those from 1000 random cohorts (P < 0.05). This approach showed that 44 clustered complexes comprising 324 protein subunits were significantly interacting with six human α-arrestins (Figure 2A; Table S7) and 21 clustered complexes comprising 192 subunits were significantly interacting with Drosophila α-arrestins (Figure 2B; Table S7).
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_fig2.tif/full/max/0/default.jpg)
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_fig2a.tif/full/max/0/default.jpg)
Network of α-arrestins and their interacting protein complexes
Network of α-arrestins and the functional protein complexes that significantly interact with them in human (A) and Drosophila (B). α-arrestins are colored yellow and prey proteins in protein complexes are colored according to the SAINTexpress scores of the PPIs. The gray edges indicate that evidence supporting the complex was provided by COMPLEAT and/or GO cellular components. The thickness of the green arrows indicates the strength of the interaction between α-arrestins and the indicated protein complexes, which was estimated with -log10 FDR of complex association scores. COMPASS, complex proteins associated with Set1; SMN, survivor of motor neurons; TFIIIC, transcription factor III C; RNA polII, RNA polymerase II; MCM, minichromosome maintenance protein complex; SAC, spindle assembly checkpoint; NSL, non-specific lethal; WASH, Wiskott-Aldrich syndrome protein and scar homolog; Arp2/3, actin related protein 2/3. TEF, transcription elongation factor.
The two largest complexes found to interact with α-arrestins were related to protein degradation (proteasome and ubiquitin-dependent proteolysis) and RNA splicing and processing in both species (Figure 2; Table S7). ARRDC1, 2, and 4 and TXNIP in human and Leash and CG2993 in Drosophila were found to interact with protein degradation complexes. CG2993 and CG18747 appeared to bind to a putative complex comprising NEDD4 family interacting protein 2, recently reported to be a mediator of ubiquitin ligase (Trimpert et al., 2017). On the other hand, ARRDC2, 3, and 4 in human and Leash, CG18746, Vdup1, CG10086, and CG18744 in Drosophila were found to interact with RNA splicing and processing complexes. Although the above-mentioned α-arrestins interacted in common with the two complexes described above, they were also found to bind to distinct complexes. For instance, TXNIP specifically binds to transcriptional and histone deacetylase (HDAC) complexes, ARRDC1 to axon guidance, endosomal sorting, and laminin complexes, ARRDC2 to the Set1C/COMPASS complex, ARRDC3 to transcription elongation factors and spindle assembly checkpoint and cell polarity complexes, and ARRDC4 to clathrin-coated pit and BAT3 complexes in human. In Drosophila, Leash specifically binds to AP-2 adaptor and WASH complexes and CG18746 to the UTP B complex. In addition to the two largest complexes and their associated α-arrestins, ARRDC5 in human and CG2641, CG1105, CG14696, and CG18745 in Drosophila interact in common with protein transport and localization complexes. ARRDC5 is specifically associated with V-type ATPase and vacuolar protein sorting complexes in human. CG18748 and CG18747 are associated with motor protein complexes including actin, myosin, and microtubule-associated complexes in Drosophila. Taken together, the results from this analysis provide a glimpse of underexplored roles for α-arrestins in diverse cellular processes.
Conserved interactomes of α-arrestins
Given that α-arrestins are widely conserved in metazoans (DeWire, Ahn, Lefkowitz, & Shenoy, 2007), we sought to exploit the evolutionally conserved interactomes of human and Drosophila α-arrestins. For this analysis, we searched for orthologous relationships in the α- arrestin interactomes using the DRSC Integrative Ortholog Prediction Tool (DIOPT) (Hu et al., 2011). Among high-confidence prey proteins, 68 in human and 64 in Drosophila were reciprocally predicted to have ortholog relationships, defining 58 orthologous prey groups (DIOPT score ≥ 2; Table S8). α-arrestins were then hierarchically clustered based on the log2 transforemd mean spectral counts of these orthologous interactome, defining seven groups of α-arrestins. Orthologous prey proteins were grouped according to their shared biological function, defining nine functional groups and others of diverse functions (Figure 4). The resulting clusters revealed PPIs that were functionally conserved. For instance, ARRDC3 in human and CG18746 in Drosophila actively interact with proteins in RNA binding and splicing groups. Leash in Drosophila appeared to interact with proteins in similar functional groups as ARRDC3 but, like ARRDC1, it also extensively interacts with members of ubiquitin-dependent proteolysis groups. In addition, ARRDC4 interacts with proteins in the motor protein and trafficking group, similar to CG18748 in Drosophila, and binds to proteins in the ubiquitin dependent proteolysis group, similar to TXNIP. Similarly, CG10086 and Vdup1, CG14696 and ARRDC5, and CG2993 and ARRDC2 appeared to have conserved interactomes between human and Drosophila.
The most prominent functional modules shared across both species were the ubiquitin dependent proteolysis, endosomal trafficking, and small GTPase binding modules, which are in agreement with the well-described functions of α-arrestins in membrane receptor degradation through ubiquitination and vesicle trafficking (Dores et al., 2015; Nabhan et al., 2012; Puca et al., 2013; Xiao et al., 2018) (Figure 4). In contrast, the functional modules involving cyclin and cyclin-dependent kinase, casein kinase complex, and laminin seemed to be conserved between relatively specific sets of α-arrestins, whereas those related to motor proteins and RNA binding and splicing were more generally conserved. Taken together, the comparative analyses led us to identification of detailed, orthologous interactome maps of α- arrestins, which extend beyond the limited insights provided by sequence-based comparative analysis alone (Figure S4). Conserved roles of α-arrestins in both established and previously uncharacterized signaling pathways expand our understanding of the diverse roles of α- arrestins in cellular signaling.
Chromatin accessibility is globally decreased by TXNIP knock down
TXNIP is one of the most well-studied α-arrestins. Previous studies reported that TXNIP interacts with transcriptional repressors, such as FAZF, PLZF, and HDAC1 or HDAC3, to exert antitumor activity (S. H. Han et al., 2003) or repress NF-kB activation (H. J. Kwon et al., 2010). However, although such studies provided information about interactions with a few transcriptional repressors, they barely provided a systematic view of the roles of TXNIP in controlling the chromatin landscape and gene expression. In that sense, our PPI analysis first revealed that TXNIP extensively binds to chromatin remodeling complexes, such as the HDAC and histone H2B ubiquitination complexes, as well as to transcriptional complexes, such as the RNA polymerase II and transcription factor IIIc complexes (Figure 2A). Such PPIs indicate that TXNIP could control transcriptional and epigenetic regulators. To examine how the global epigenetic landscape is remodeled by TXNIP, we knocked down its expression in HeLa cells with a small interfering RNA (siTXNIP) and confirmed a decrease at both the RNA and protein levels (Figure 4A and B). We then produced two biological replicates of ATAC- and RNA-seq experiments in HeLa cells with TXNIP depletion (Table S9) to detect differentially accessible chromatin regions (dACRs) and differentially expressed genes (DEGs) (Figure 4C; Table S10). The replicated samples were well grouped by the siTXNIP condition in principle component spaces (Figure S5A and B). The normalized ATAC-seq signal and the RNA level of expressed genes clearly showed the enrichment of open chromatin signals around the transcription start sites (TSSs) of genes that are actively transcribed (Figure S5C). We detected 70,746 high confidence accessible chromatin regions (ACRs) across all samples, most of which were located in gene bodies (38.74%), followed by intergenic regions (32.03%) and promoter regions (29.23%, Figure S6A). TXNIP knockdown appeared to induce a global decrease in chromatin accessibility in many genomic regions including promoters (Figure 4D). Of the high confidence ACRs, 7.38% were dACRs under TXNIP depletion; most dACRs showed reduced chromatin accessibility under this condition (dACRs(-), Figure 4D; Figure S6B). dACRs(-) were preferentially localized in gene bodies, whereas dACRs(+) were more often observed in promoter regions (Figure S6C).
The global chromatin changes induced by TXNIP knockdown could impact gene expression at corresponding loci. In fact, our gene expression analysis showed that 956 genes were downregulated, and 295 genes were upregulated by TXNIP knockdown compared to the control (Figure 4E), suggesting that the global decrease in chromatin accessibility induced by TXNIP depletion would mediate the repression of gene expression. To confirm this phenomenon, we first selected sets of differentially (“Down genes” and Up genes” in Figure 4E, “Down” and “Up” in Figure 4F) and non-differentially expressed genes (“None” in Figure 4F) with at least one detectable ACR in promoter or gene body. Next, the cumulative distribution function (CDF) of accessibility changes demonstrated that the genes with a decreased RNA level (“Down”) showed significantly reduced chromatin accessibilities at promoters compared to those with no changes in the RNA level (“None”) (Figure 4F; P < 5.81 X 10-28 for max changes; Figure S6D; P < 3.76 X 10-32 for mean changes, Kolmogorov-Smirnov (KS) test). In contrast, genes with increased RNA expression (“Up”) exhibited no changes in chromatin accessibility at the promoter (Figure 4F; P < 0.68 for max changes; Figure S6D; P < 0.49 for mean changes, KS test), indicating that chromatin opening at promoters is necessary but not sufficient to induce gene expression. ACRs located in gene bodies also showed a similar trend: genes with a decreased RNA level (“Down”) showing decreased chromatin accessibility upon TXNIP depletion (Figure S6E; P < 0.002 for max changes and P < 7.68 X 10-7 for mean changes, KS test), suggesting that TXNIP is likely to be a negative regulator of chromatin repressors that induce heterochromatin formation. We then used GO analysis(Raudvere et al., 2019) to examine the biological functions of genes that exhibited decreased chromatin accessibility at their promoter and decreased RNA expression upon TXNIP knockdown (Table S11). In general, genes associated with developmental process, signaling receptor binding, cell adhesion and migration, immune response and extracellular matrix constituents appeared to be repressed upon TXNIP depletion (Figure 4G).
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_fig3.tif/full/max/0/default.jpg)
A substantial fraction of α-arrestin-PPIs are conserved across species
Human and Drosophila α-arrestins are hierarchically clustered based on log2-transformed mean spectral counts of their orthologous interactome. They are then manually grouped according to shared biological functions and assigned distinct colors. The names of orthologous proteins that interact with α-arrestins are displayed on the right side of the heatmap.
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_fig4.tif/full/max/0/default.jpg)
TXNIP knockdown induces a global decrease in chromatin accessibility and gene expression
(A-B) HeLa cells were treated with either siRNA against TXNIP (siTXNIP) or negative control (siCon) for 48 hours (hr) and analyzed of changes in the mRNA (A) and protein levels (B) of TXNIP. (A) Expression levels of RNAs were quantified by RNA-seq (left, log2 counts per million mapped reads (CPM), see “Processing of RNA-seq data” in “Materials and Methods”) and RT-qPCR (right, relative levels of TXNIP in siTXNIP compared to siCon condition, see “Quantitative Reverse-transcription PCR” in Supplementary Information). (B) Protein levels were first visualized by western blot analysis of lysates from HeLa Cells and band intensities of three independent experiments were quantified (right). (A-B) Gray dots depict actual values of each experiment and bar plots indicate mean ± standard deviation (sd). ***FDR < 0.001 (test of differential expression by edgeR (Robinson et al., 2010), see “Processing of RNA-seq data” in “Materials and Methods”) for RNA-seq. *P < 0.05, *** P < 0.001 (two-sided paired Student T test) for RT-qPCR and western blots. (C) A schematic workflow for detecting dACRs and DEGs using ATAC- and RNA-seq analyses, respectively. (D) Volcano plots of differential chromatin accessibility for all ACRs (left) and those associated with promoters (right). (E) Volcano plots of differential gene expression. (D-E) Blue dots denote “dACRs(-)” of significantly decreased chromatin accessibility (D) and “Down” genes of significantly down regulated genes (E) in siTXNIP-treated cells (FDR ≤ 0.05, log2(siTXNIP / siCon) ≤ -1); red dots denote “dACRs(+)” of significantly increased chromatin accessibility (D) and “Up” genes of significantly up-regulated genes (E) in siTXNIP-treated cells (FDR ≤ 0.05, log2(siTXNIP / siCon) ≥ 1). Black dots denote data points with no significant changes. (F) Changes in chromatin accessibility of ACRs located in the promoter region of genes were plotted as CDFs. Genes were categorized into three groups based on changes in RNA levels (“Up”, “Down” as in (E) and “None” indicating genes with -0.5 ≤ log2(siTXNIP / siCon) ≤ 0.5. The number of genes in each group are shown in parentheses and P values in the left upper corner were calculated by one-sided KS test. (G) Top 10 GO terms (biological process and molecular function) enriched in genes that exhibited decreased chromatin accessibility at their promoter and decreased RNA expression upon TXNIP knockdown (Table S11).
TXNIP represses the recruitment of HDAC2 to target loci
Given that TXNIP knockdown led to a global reduction in chromatin accessibility with decreased transcription, we focused on identifying the potential role of the epigenetic silencer HDAC2, one of the strong binding partners of TXNIP in the AP/MS analysis, in mediating the TXNIP-dependent epigenetic and transcriptional modulation. Consistent with the AP/MS data, immunoprecipitation (IP) experiments showed that the two proteins indeed interact with each other. Furthermore, TXNIP knockdown reduced the amount of TXNIP-interacting HDAC2 protein but did not affect the HDAC2 expression level (Figure 5A). To find out how the TXNIP HDAC2 interaction impacts the epigenetic and transcriptional reprogramming of target loci, we first checked whether the TXNIP-HDAC2 interaction causes cytosolic retention of HDAC2 to inhibit nuclear HDAC2-mediated global histone deacetylation. However, both the expression level and subcellular localization of HDAC2 were unaffected by a reduction in TXNIP, as confirmed by Western blot analysis using cytoplasmic and nuclear fractions as well as by an immunofluorescence assay (Figure 5B and Figure S7A), indicating that TXNIP might modulate HDAC2 activity in a different way.
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_fig5.tif/full/max/0/default.jpg)
TXNIP directly represses the recruitment of HDAC2 to target loci
(A) Analysis of co-IP between the TXNIP and HDAC2 proteins. Lysates from HeLa cells that had been treated with either siCon or siTXNIP for 48 hr were subjected to IP and immunoblotting with antibodies recognizing TXNIP and HDAC2, with IgG used as the negative control. (B) Nuclear and cytoplasmic fractions of HeLa cells were analyzed with Western blots following transfection with siCon or siTXNIP for 48 hr (left). Lamin B1 and GAPDH were used as nuclear and cytoplasmic markers, respectively. Western blot results from three independent experiments for TXNIP and HDAC2 were quantified as in Figure 4B. C, cytoplasm; N, nucleus. (C) Genomic regions showing RNA expression and chromatin accessibility at CD22 and L1CAM gene loci (top). Through the ChIP-qPCR analysis, the fold enrichment of HDAC2 and histone H3 acetylation (H3ac) at the CD22 and L1CAM promoter regions in HeLa cells treated with either siCon or siTXNIP for 48 hr were quantified (bottom). Data are presented as the mean ± sd (n=3, biological replicates). Gray dots depict actual values of each experiment. *P < 0.05, **P < 0.01, ns : not significant (two-sided paired Student T test).
We next asked if the transcriptional suppression of TXNIP-target genes was mediated by changes in HDAC2 recruitment to and histone acetylation of chromatin. To address this question, genes that were significantly downregulated by TXNIP knockdown and that contained at least one dACR in the promoter were selected by the following additional criteria: 1) the RNA level in normal HeLa cells is ≥ 10 TPM and 2) the total ATAC-seq read count at the promoter in siTXNIP-treated HeLa cells is reduced ≥ 1.5-fold compared to that in normal cells. Among the four TXNIP-target genes selected by the above-mentioned criteria, the expression levels of CD22 and L1CAM were significantly reduced (P < 0.05, Student’s T test, Figure S7B). The two genes were further examined to determine whether the levels of HDAC2 binding signal and histone acetylation in their promoter regions were changed upon TXNIP knockdown (Figure 5C). We observed that RNA- and ATAC-seq coverages in exonic and promoter region of CD22 and L1CAM genes were clearly reduced upon TXNIP depletion (Figure 5C top) and an analysis of chromatin immunoprecipitation (ChIP) signals for HDAC2 and histone H3 acetylation at each dACR(-) detected in the L1CAM and CD22 promoters revealed that TXNIP knockdown increased the recruitment of HDAC2 to TXNIP-target loci, accompanied by decreased histone H3 acetylation (Figure 5C bottom). Therefore, these results suggest that the TXNIP interaction with HDAC2 inhibits the chromatin occupancy of HDAC2 and subsequently reduces histone deacetylation to facilitate global chromatin accessibility.
ARRDC5 plays a role in osteoclast differentiation and function
Given that various subunits of the V-type ATPase interact with ARRDC5, we speculated that ARRDC5 might be involved in the function of this complex (Figure 6A). V-type ATPase plays an important role in the differentiation and function of osteoclasts, which are multinucleated cells responsible for bone resorption in mammals (Feng et al., 2009; Qin et al., 2012). Therefore, we hypothesized that ARRDC5 might be also important for osteoclast differentiation and function. To determine whether ARRDC5 affects osteoclast function, we prepared osteoclasts by infecting bone marrow-derived macrophages (BMMs) with lentivirus expressing either GFP-GFP or GFP-ARRDC5 and differentiating the cells into mature osteoclasts. After five days of differentiation, ectopic expression of GFP-ARRDC5 had significantly increased the total number of tartrate resistant acid phosphatase (TRAP)-positive multinucleated cells compared to GFP-GFP overexpression (Figure 6B). In particular, the number of TRAP-positive osteoclasts with a diameter larger than 200 μm was significantly increased by GFP-ARRDC5 overexpression (Figure 6B), suggesting that ARRDC5 expression increased osteoclast differentiation. Additionally, the area of resorption pits produced by GFP ARRDC5-expressing osteoclasts in a bone resorption pit assay was approximately 4-fold greater than that of GFP-GFP expressing osteoclasts (Figure 6C). These results imply that the ectopic expression of ARRDC5 promotes osteoclast differentiation and bone resorption activity.
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_fig6.tif/full/max/0/default.jpg)
Interaction of ARRDC5 with the V-type ATPase in osteoclasts
(A) The human ARRDC5-centric PPI network. V-type and P-type ATPases, their related components, and extracellular exosomes are labeled and colored. Other interacting proteins are indicated with gray circles. (B) TRAP staining of osteoclasts. Cell differentiation was visualized with TRAP staining of GFP-GFP or GFP-ARRDC5 overexpressing osteoclasts (scale bar = 500 μm). TRAP-positive multinucleated cells (TRAP+ MNC) were quantified as the total number of cells and the number of cells whose diameters were greater than 200 μm. * P < 0.05. (C) Resorption pit formation on dentin slices. Cell activity was determined by measuring the level of resorption pit formation in GFP-GFP or GFP-Arrdc5 overexpressing osteoclasts (scale bar = 200 μm). Resorption pits were quantified as the percentage of resorbed bone area per the total dentin disc area using ImageJ software. The resorption area is relative to that in dentin discs seeded with GFP-GFP overexpressing osteoclasts, which was set to 100%. ** P < 0.01. (D) Localization of Arrdc5 and the V-type ATPase in osteoclasts. The V-type ATPase was visualized with immunofluorescence (red), GFP-GFP and GFP-ARRDC5 were visualized with GFP fluorescence (green), and nuclei were visualized with DAPI (blue). Representative fluorescence images are shown. Dashed lines were used to outline representative osteoclasts (scale bar = 100 μm).
The V-type ATPase is localized at the osteoclast PM (Toyomura et al., 2003) and its localization is disrupted by bafilomycin A1, which is shown to attenuate transport of the V-type ATPase to the membrane (Matsumoto & Nakanishi-Matsui, 2019). We analyzed changes in V-type ATPase localization in GFP-GFP and GFP-ARRDC5 overexpressing osteoclasts. GFP signals were detected at the cell cortex when GFP-ARRDC5 was overexpressed, indicating that ARRDC5 might also localized to the osteoclast PM (Figure 6D). In addition, we detected more V-type ATPase signals at the cell cortex in the GFP-ARRDC5 overexpressing osteoclasts, and ARRDC5 and V-type ATPase were co-localized at the osteoclast membrane (Figure 6D). Notably, bafilomycin A1 treatment reduced not only the V-type ATPase signals detected at the cortex but also the GFP-ARRDC5 signals (Figure 6D). These results indicate that ARRDC5 might control the membrane localization of the V-type during osteoclast differentiation and function.
Discussion
We constructed high-confidence interactomes of α-arrestins from human and Drosophila, comprising 307 and 467 interacting proteins, respectively. The resulting interactomes greatly expanded previously known PPIs involving α-arrestins and the majority of interactomes were first reported in this study, which needs to be validated experimentally (Tian, Kang, & Benovic, 2014; Zbieralski & Wawrzycka, 2022). However, some known PPIs were missed in our interactomes due to low spectral counts and SAINTexpress scores, probably resulting from different cellular contexts, experimental conditions, or other factors (Figure S2C).
Integrative map of protein complexes that interact with α-arrestins (Figure 2) hint towards many aspects of α-arrestins’s biology that remain uncharacterized. For example, role of α-arrestins in the regulation of β2AR in human remained controversial. One study proposed that α-arrestins might act coordinately with β-arrestins at the early step of endocytosis, promoting ubiquitination, internalization, endosomal sorting and lysosomal degradation of activated GPCRs (Shea et al., 2012). The another study, however, proposed different hypothesis suggesting that α-arrestins might act as secondary adaptor localized at endosomes to mediate endosomal sorting of cargo molecules (S. O. Han et al., 2013). Among the protein complexes that interact with α-arrestins, we identified those related with clathrin coated pit in human and AP-2 adaptor complex in Drosopihila (Figure 2). They are multimeric proteins to induce internalization of cargo molecules to mediate clathrin-mediated endocytosis, which suggests involvement of α-arrestins in early step of endocytosis.
Among the interacting proteins, 58 orthologous interacting groups were observed to be conserved between human and Drosophila, suggesting conserved roles of α-arrestins between two species (Figure 3). Among conserved proteins, proteins known to interact with human α-arrestins, such as NEDD4, WWP2, WWP1, and ITCH, were identified along with its orthologs in Drosophila, which are Su(dx), Nedd4, and Smurf, implying that regulatory pathway of ubiquitination-dependent proteolysis by α-arrestins is also present in invertebrate species. Besides the known conserved functions, the novel conserved functions of α-arrestins interactomes were also identified, such as RNA splicing (Figure 1E; Figure 3). Because our protocol did not include treatment with RNase before the AP/MS, it is possible that RNA binding proteins could co-precipitate with other proteins that directly bind to α-arrestins through RNAs, and thus could be indirect binding partners. Nevertheless, other RNA binding proteins except for RNA splicing and processing factors were not enriched in our interactomes, indicating that this possibility may be not the case. Thus, it might be of interest to explore how α-arrestins are linked to RNA processing in future.
Some protein complexes and functional modules were found to be involved in specific cellular processes discovered in only human, suggesting that some functional roles of α- arrestins have diverged through evolution. As examples of specific cellular functions of α- arrestins, we explored the biological relevance of two interacting protein complexes: 1) the interaction between TXNIP and chromatin remodelers and 2) the interaction between ARRDC5 and the V-type ATPase complex. Given that TXNIP interacts with chromatin remodelers, such as the HDAC and histone H2B ubiquitination complexes, we speculated that chromatin structures could be affected by the interactions. Although we showed that siTXNIP treatment directed a global decrease in chromatin accessibilities and gene expression by inhibiting the binding of HDAC2 to targets, histones themselves could be also controlled by the interaction between TXNIP and the H2B ubiquitination complex. An impact of TXNIP on histone ubiquitination could strengthen the negative regulation of target loci by siTXNIP treatment. In addition, TXNIP interacts with the proteasome, which induces the degradation of binding partners (Figure 2). However, we observed that the cellular level and localization of HDAC2 were not affected by TXNIP reduction (Figure 5A and B; Figure S7A), meaning that the proteasome seems not to be involved in TXNIP’s influence on HDAC2; rather, TXNIP directly hinders HDAC2 recruitment to target loci.
Because the V-type ATPase plays a key role in osteoclast differentiation and physiology (Feng et al., 2009; Qin et al., 2012), we investigated a possible role for the ARRDC5-V-type ATPase interaction in this cell type. We showed that the ectopic expression of ARRDC5 increased both the differentiation of osteoclasts into their mature form and their bone reabsorption activity. Additionally, ARRDC5 co-localized with the V-type ATPase at the PM (Figure 6C). Thus, further characterization of ARRDC5 and its interactome in osteoclasts might clarify how ARRDC5 regulates the V-type ATPase to play a role in osteoclast differentiation and function. With the results, the discovery of new binding partners and their functions of TXNIP and ARRDC5 will facilitate the further investigations to explore the novel PPIs of α-arrestins.
Given the plethora of PPIs uncovered in this study, we also anticipate that our study could provide insight into many disease models. In fact, despite a limited knowledge of their biology, α-arrestins have already been linked to a range of cellular processes and several major health disorders, such as diabetes (Batista et al., 2020; Wondafrash et al., 2020), cardiovascular diseases (Domingues, Jolibois, Marquet de Rouge, & Nivet-Antoine, 2021), neurological disorders (Tsubaki, Tooyama, & Walker, 2020), and tumor progression (Y. Chen et al., 2020; Mohankumar et al., 2015; Oka et al., 2006), making them potential therapeutic targets. For the community, we provide comprehensive α-arrestin interactome maps on our website (human: http://big.hanyang.ac.kr/alphaArrestin_Human and Drosophila: http://big.hanyang.ac.kr/alphaArrestin_Fly). Researchers can search and download their interactomes of interest as well as access information on potential cellular functions associated with these interactomes.
Materials and Methods
Generating Drosophila α-arrestin-GFP fusion DNA constructs
To create Drosophila ARRDC entry clones, we gathered cDNA sequences of twelve Drosophila α-arrestins : CG2993 (#2276, Drosophila Genomics Resource Center, DGRC, Bloomington, IN, USA), CG18744 (#1388606, DGRC), CG18745 (#12871, DGRC), CG18746 #9217, DGRC), CG18747 (#1635366, DGRC), CG18748 (#1387253, DGRC), CG2641 (#1649402, DGRC), CG10086 (#8816, DGRC), CG14696 (#1644977, DGRC), CG1105 (#4234, DGRC), Vdup1 (#1649326, DGRC), and Leash (Y. Kwon et al., 2013). We then subcloned each cDNA sequence of Drosophila α-arrestins into pCR8 entry clone vector using pCR8/GW/TOPO TA cloning kit (#K250020, Thermo Fisher Scientific, Waltham MA, USA), by following manufacturer’s protocol. To generate plasmids with suitable system for protein expression in Drosophila cell culture, we then subcloned these α-arrestins-containing-pCR8 plasmids into pMK33-Gateway-GFP destination vector (Y. Kwon et al., 2013; Kyriakakis, Tipping, Abed, & Veraksa, 2008) using Gateway LR Clonase II enzyme mix (#11791020, Thermo Fisher Scientific), where coding sequences of α-arrestins are inserted before GFP sequence. Final constructs were validated by GENEWIZ Sanger Sequencing.
Establishing Drosophila α-arrestin-GFP stably expressing cell lines
S2R+ cells were maintained in Schneider’s Drosophila Medium (#21720024, Thermo Fisher Scientific) supplemented with 10% heat inactivated FBS (#16140071, Thermo Fisher Scientific) and 1% Penicillin Streptomycin (#15070063, Thermo Fisher Scientific) at 24°C. To establish α-arrestin-GFP stably expressing Drosophila cell lines, 0.4x106 S2R+ cells were seeded in 6-well plates and were transfected with 1 μg of each pMK33-ARRDC-GFP construct using Effectene transfection reagent (#301425, Qiagen, Venlo, Netherlands). pMK33 plasmid is a copper-induced protein expression vector, which carries Hygromycin B-antibiotic-resistant gene. Therefore, we selected for α-arrestin-GFP stable cell lines by maintaining cells in Schneider’s Drosophila Medium supplemented with 200 μM Hygromycin B (#40-005, Fisher Scientific). The stable cells were transferred into T25 cm2 flasks to repopulate. To induce the expression of α-arrestin-GFP fusion proteins, we exposed the stable cells to 500 μM CuSO4 (#C8027, Sigma Aldrich, Burlington, MA, USA) to the media. We confirmed the GFP-tagged α-arrestin protein expressions using fluorescence microscopy.
Synthesizing human α-arrestin coding sequence
Due to the lack of commercially available stock, we utilized GENEWIZ (South Plainfield, NJ, USA) gene synthesis service to synthesize human ARRDC5 coding sequence (NM_001080523).
Generating mammalian GFP- α-arrestin fusion DNA constructs
To create human α-arrestin entry clones, we subcloned ARRDC3 (#38317, Addgene, Watertown, MA, USA) and ARRDC5 (GENEWIZ) into pCR8 entry clone vector using pCR8/GW/TOPO TA cloning kit (#K250020, (Thermo Fisher Scientific), by following manufacturer’s protocol. ARRDC1 (BC032346, GeneBank), ARRDC2 (BC022516, GeneBank), ARRDC4 (BC070100, GeneBank), and TXNIP (BC093702, GeneBank) were cloned into pCR8. To generate plasmids with suitable system for protein expression in mammalian cell culture, we then subcloned these α-arrestin s-containing-pCR8 plasmids into pHAGE-GFP-Gateway destination vector (gift from Dr. Chanhee Kang at Seoul National Univesity) using Gateway LR Clonase II enzyme mix (#11791020, Thermo Fisher Scientific), where coding sequences of α-arrestin are inserted after GFP sequence. Final constructs were validated by GENEWIZ Sanger Sequencing.
Establishing mammalian GFP- α-arrestin stably expressing cell lines
We produced GFP-α-arrestins lentiviral particles by seeding 5 x106 HEK293T cells in 10 cm2 dish with Dulbecco’s Modified Eagle Medium (#11965118, Thermo Fisher Scientific) supplemented with 25 mM HEPES, 10% heat-inactivated fetal bovine serum (#16140071, Thermo Fisher Scientific), and 1% Penicillin Streptomycin (#15070063, Thermo Fisher Scientific) at 37°C in humidified air with 5% CO2. Approximately after 16-24 hours (hr), at 90% cell confluency, we changed the cell media to Opti-MEM medium (#31985070, Thermo Fisher Scientific) and transfected the cells with 10 μg pHAGE-GFP-α-arrestin construct, 10 μg lentivirus packaging plasmid (pCMV-dR8.91), and 10 μg virus envelope plasmid (VSV-G) using PEIPro DNA transfection reagent (#115010, VWR, Radnor, PA, USA). GFP-α-arrestins lentiviral particles were harvested 40 hr-post transfections. To establish GFP-α-arrestins stably expressing mammalian cell lines, HEK293 cells were seeded in 10 cm2 dish with Dulbecco’s Modified Eagle Medium (#11965118, Thermo Fisher Scientific) supplemented with 25 mM HEPES, 10% heat-inactivated fetal bovine serum (#16140071, Thermo Fisher Scientific) and 1% Penicillin Streptomycin (#15070063, Thermo Fisher Scientific) at 37°C in humidified air with 5% CO2. At 90% cell confluency, cells were infected with pHAGE-GFP- ARRDC lentivirus particle, and stable cells were selected by maintaining cells in media supplemented with1.5 μg/mL puromycin (#BP2956100, Thermo Fisher Scientific). We confirmed the GFP-tagged α-arrestin protein expressions using fluorescence microscopy.
Affinity purification of Drosophila and human GFP-tagged α-arrestin complexes
We seeded each of the Drosophila α-arrestin-GFP stable cells in six T-75 cm2 flasks (2.1x 106 cells per flask) and α-arrestin-GFP expression was induced for 48 hr with 500 μM CuSO4. Meanwhile, we seeded each of the human GFP-α-arrestin stable cells in eight T-75 cm2 flasks and grown for 48 hr before collection. The cells were harvested by spinning down cells at 1,000g for 5 minutes (min) and washed once with cold PBS. We lysed the cells by resuspending cells in lysis buffer (10 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 1.5 mM MgCl2, 5% glycerol, 0.5% NP-40, 25 mM NaF, 1mM DTT, and 1x HALT protease and phosphatase inhibitor (#PI78442, Thermo Fisher Scientific)) and incubating them for 40 min. The lysate was separated from the insoluble fraction by centrifugation at 20,000 g for 15 min at 4℃. To capture the α-arrestins and their native interactors, each α-arrestin-containing lysate was incubated with GFP-nanobody-conjugated to Dynabeads M-270 Epoxy magnetic beads (#14301, Thermo Fisher Scientific). The supernatant was separated from the beads using a magnetic rack, and the beads were washed five times with lysis buffer. The protein complexes were eluted from the beads by adding 200 mM glycine pH 2.5 and the pH was neutralized with Tris base pH 10.4. Purified α-arrestin proteins were confirmed by running a fraction of the eluted proteins on SDS-PAGE/Coomassie gel.
Protein sample preparation for mass spectrometry
To digest protein samples into peptides for mass spectrometry analysis, we precipitated the eluted proteins by adding trichloroacetic acid (#T0699, Sigma Aldrich) to 20% final concentration, followed by spinning down samples at maximum speed for 30 min at 4℃. The precipitates were washed with 10% trichloroacetic acid solution and three additional times with Acetone (#A929, Thermo Fisher Scientific), and left to dry in room temperature. Protein precipitations were digested with Trypsin (Promega, #V5113) diluted in Digestion buffer (100 mM Ammonium Bicarbonate and 10% Acetonitrile) in 1:40 ratio. Resulting peptides were purified using ZipTip Pipet tips (#ZTC18M096, Thermo Fisher Scientific).
LC/MS-MS analysis
We used cells stably expressing GFP and wild-type HEK293 or S2R+ cells alone as control baits. AP/MS experiments for all Drosophila and human α-arrestin baits were performed in two biological replicates, with the exception of human ARRDC3 baits (two technical replicates). Samples were resuspended in Mass Spectrometry buffer (5% Formic Acid and 5% Acetonitrile) and were analyzed on an Liquid Chromatography Orbitrap Fusion Lumos Tribrid Mass Spectrometer (#IQLAAEGAAPFADBMBHQ, Thermo Fisher Scientific) equipped with a nano-Acquity UPLC system and an in-house developed nano spray ionization source. Peptides were separated using a linear gradient, from 5-30% solvent B (LC-MS grade 0.1% formic acid (#A117, Thermo Fisher Scientific) and acetonitrile) in a 130 min period at a flow rate of 300 nL/min. The column temperature was maintained at a constant 5°C during all experiments. Peptides were detected using a data dependent method. Survey scans of peptide precursors were performed in the Orbitrap mass analyzer from 380 to 1500 m/z at 120K resolution (at 200 m/z) with a 5 x 105 ion count target and a maximum injection time of 50 milliseconds (ms). The instrument was set to run in top speed mode with 3 seconds (sec) cycles for the survey and the MS/MS scans.
Identification of high-confidence bait-prey PPIs
SAINTexpress analysis
To identify high-confidence bait-prey PPIs, spectral counts of AP/MS data from S2R+ and HEK293 cells were subjected to the SAINTexpress algorithm (version 3.6.1) (Teo et al., 2014), which calculates the probability of authenticity for each bait prey PPI. The program outputs the SAINTexpress scores and the Bayesian false discovery rates (BFDR) based on the spectral count distribution of true and false PPI sets. Before calculating the scores, bait-to-bait self-interactions were removed manually. SAINTexpress was run with the “-R 2” parameter, which specifies the number of replicates, and the “-L 3” parameter, which specifies the number of representative negative control experiments to be considered.
PPI validation datasets
To evaluate the performance of the PPI prediction based on the SAINTexpress score, validation datasets including positive and negative PPIs were precompiled as described in previous studies (Y. Kwon et al., 2013; Vinayagam et al., 2016). Briefly, the positive PPIs were initially collected by searching for known PPIs involving α- arrestins from STRING version 10.5 (Szklarczyk et al., 2015), GeneMANIA version 3.4.1 (Warde-Farley et al., 2010), Bioplex (Huttlin et al., 2015), and DpiM (Guruharsha et al., 2011). For human, additional positive PPIs were curated from the literature (Colland et al., 2004; Dotimas et al., 2016; Nabhan et al., 2012; Nishinaka et al., 2004; Puca & Brou, 2014; Wu et al., 2013). After these steps, 30 PPIs (21 preys) for human and 46 PPIs (17 preys) for Drosophila were considered as positive PPIs (Tables S3A and C). Proteins manually curated from the Contaminant Repository for Affinity Purification (CRAPome) (Mellacheruvu et al., 2013) were compared to those detected in our negative controls and only those that were detected in both were considered as were negative PPIs (Tables S3B and D). As a result of these steps, 1,372 PPIs (268 preys) for human and 1,246 PPIs (122 preys) for Drosophila were compiled as negative PPIs.
Construction of high-confidence PPI networks
The performance of SAINTexpress was evaluated using the positive and negative PPIs. Because there is an imbalance between positive and negative PPIs, 1000 random cohorts of negative PPIs number-matched with that of positive PPIs were generated. The average true positive and false positive rates were plotted as ROC curves over different SAINTexpress scores as a cutoff, and AUC values were calculated using the ROCR R package (version 1.0-11, https://cran.r-project.org/web/packages/ROCR). Based on these results, we chose an optimal cutoff for high confidence PPIs with a BFDR of 0.01, where the false positive rates were less than 3% (∼1.8 % for human and ∼2.7% for Drosophila) in both species, and the true positive rates were substantially higher (∼66.7 % for human and ∼45.7% for Drosophila). The cutoffs correspond to SAINTexpress scores of 0.85 and 0.88 for human and Drosophila, respectively.
Identification of protein complexes associated with α-arrestins
To examine protein complexes significantly enriched in the α-arrestin interactomes, we collected known protein complexes from two databases: COMPLEAT (Vinayagam et al., 2013), which is a comprehensive resource of protein complexes built from information in the literature and predicted by orthologous relationships of proteins across species (human, Drosophila, and yeast), and the DAVID GO analysis of cellular components (Huang da et al., 2009a, 2009b) (Benjamini-Hochberg FDR ≤ 0.05) (Table S7B and E), from which bulk cellular compartments such as the nucleus, cytosol, and so on were excluded. From the COMPLEAT database, we evaluated the association of the resulting protein complexes with each α-arrestin by the complex association score, which is the IQM of SAINTexpress scores (Equation 1)
, where the first quartile is
From the clustered set of complexes, we manually removed those with fewer than three subunits or two PPIs. Subunits in the complexes that have no connection among themselves were also removed. Lastly, the significance of associations of the resulting complexes with each α-arrestin were tested in the same manner as done in COMPLEAT using complex association score. The resulting P values were corrected by the Benjamini-Hochberg procedure and only interactions with statistical significance (FDR < 0.05) were visualized with Cytoscape v3.5.1 (Shannon et al., 2003) (Figure 2).
Orthologous networks of α-arrestin interactomes
DIOPT (version 7.1) was used to search for orthologs of all prey proteins and only those with a DIOPT score ≥ 2 were selected for the identification of orthologous PPIs between Drosophila and human. Next, the orthologs were tested for the enrichment of GO biological process and molecular functions and Kyoto Encyclopedia of Genes and Genomes pathway using the DAVID (Huang da et al., 2009a, 2009b). In addition, manual curation of individual genes was performed through the Uniprot database (UniProt Consortium, 2018). Theorthologs were manually grouped into functional modules based on the results and α-arrestins were modularized into seven groups based on hierarchical clustering of log2-transformed mean spectral counts using the correlation distance and the Ward linkage method. The heatmap was plotted using the pheatmap R package (version 1.0.12).
TXNIP knockdown in HeLa cells
HeLa cells (CCL-2; ATCC, Manassas, VA, USA) were cultured in complete DMEM supplemented with 10% FBS and 1% penicillin-streptomycin . Cells were cultured in an incubator at 37°C in humidified air containing 5% CO2. For siRNA-induced knockdown of TXNIP in HeLa cells, the following siRNA duplex was synthesized (Bioneer, Daejeon, South Korea): sense: 5’-GUCAGUCACUCUCAGCCAUdTdT -3’, anti-sense: 5’-AUGGCUGAGAGU GACUGACdTdT-3’. Random sequence siRNAs (AccuTarget Negative control siRNA; Bioneer), which are non-targeting siRNAs that have low sequence homology with all humans, mouse, and rat genes, were used as negative controls (siCon). 100 nM of each siRNA was transfected into 105 HeLa cells using Lipofectamine RNAiMAX (#13778075, Invitrogen, Carlsbad, CA, USA; Thermo Fisher Scientific) according to the manufacturer’s instructions. Transfected cells were harvested after 48 hr for RNA-seq and ATAC-seq (two biological replicates for each sequencing data).
RNA sequencing
For RNA-seq, total RNA was extracted using TRIzol (#15596018, Invitrogen; Thermo Fisher Scientific) according to the manufacturer’s protocol. Total RNA concentration was calculated by Quant-IT RiboGreen (#R11490, Invitrogen; Thermo Fisher Scientific). To assess the integrity of the total RNA, samples are run on the TapeStation RNA screentape (#5067- 5576, Agilent Technologies, Santa Clara, CA, USA). Only high-quality RNA preparations, with RNA integrity number greater than 7.0, were used for RNA library construction. A library was independently prepared with 1ug of total RNA for each sample by Illumina TruSeq Stranded mRNA Sample Prep Kit (#RS-122-2101, Illumina, Inc., San Diego, CA, USA). The first step in the workflow involves purifying the poly-A containing mRNA molecules using poly-T-attached magnetic beads. Following purification, the mRNA is fragmented into small pieces using divalent cations under elevated temperature. The cleaved RNA fragments are copied into first strand cDNA using SuperScript II reverse transcriptase (#18064014, Invitrogen, Thermo Fisher Scientific) and random primers. This is followed by second strand cDNA synthesis using DNA Polymerase I, RNase H and dUTP. These cDNA fragments then go through an end repair process, the addition of a single ‘A’ base, and then ligation of the adapters. The products are then purified and enriched with PCR to create the final cDNA library. The libraries were quantified using KAPA Library Quantification kits (#KK4854, KAPA BIOSYSTEMS, Wilmington, MA, USA) for Illumina Sequencing platforms according to the qPCR Quantification Protocol Guide and qualified using the TapeStation D1000 ScreenTape (#5067- 5582, Agilent Technologies). Indexed libraries were then submitted to an Illumina NovaSeq 6000 (Illumina, Inc.) as the paired-end (2×100 bp) sequencing. Both library preparation and sequencing were performed by the Macrogen (Macrogen, Inc., Seoul, South Korea).
ATAC sequencing
100,000 cells were prepared using LUNA-FL™ Automated Fluorescence Cell Counter (#L20001, logos biosystems, Gyeonggi-do, South Korea). Cells were lysed using cold lysis buffer, which consist of Nuclease-free water (#10977023, Invitrogen; Thermo Fisher Scientific), IGEPAL CA-630 (#I8896, Sigma Aldrich), 1M Trizma HCI(PH7.4) (#T2194, Sigma Aldrich), 5M NaCl (#59222C, Sigma Aldrich), and 1M MgCl2 (#M1028, Sigma Aldrich). The nuclei concentration was determined using Countess II Automated Cell Counter (#AMQAX1000, Thermo Fisher Scientific) and nuclei morphology was examined using microscopy. Immediately after lysis, resuspend nuclei (50,000 cells) were put in transposition reaction mix 50 μl, which consist of TED1 2.5μl and TD 17.5 μl (#20034197, Illumina, Inc.), nuclease free water 15 μl, and the nuclei resuspension (50,000 nuclei, 15 μl). The transposition reaction was incubated for 30 min at 37°C. Immediately following transposition, the products were purified using a MinElute PCR purification Kit (#28004, Qiagen). Next, transposed DNA fragments were amplified using Nextera DNA Flex kit (#20018704, Illumina, Inc.). To reduce GC and size bias in PCR, the appropriate number of cycles was determined as follows: qPCR side reaction was run, the additional number of cycles needed were calculated, liner Rn versus cycle was plotted and the cycle number that corresponds to 1/4 of maximum fluorescent intensity was determined. The remaining PCR reaction was run to the cycle number determined. Amplified library was purified and then quantified using KAPA library quantification kit (#07960255001, Roche, Basel, Switzerland) and Bioanalyzer (Agilent technologies). The resulting libraries were sequenced using HiSeq X Ten (Illumina, Inc.). Both library preparation and sequencing were performed by the Macrogen (Macrogen, Inc).
Processing of RNA-seq data
For quality checks and read trimming, RNA-seq data were processed by FastQC (version 0.11.8) (Andrews, 2010) and sickle (version 1.33) (Joshi NA, 2011) with default parameters. After the trimming, the reads were aligned to human transcriptomes (GENCODE version 29, GRCH38/hg38) (Frankish et al., 2019) using STAR (version 2.5.3a_modified) (Dobin et al., 2013) with default parameters and read counts were determined using RSEM (version 1.3.1) (Li & Dewey, 2011). The DEG analysis was performed using the edgeR R package (version 3.32.1) (Robinson, McCarthy, & Smyth, 2010). Batch information was added as confounding variables to adjust for batch effects. The DEGs are summarized in Supplementary Table S10.
Processing of ATAC-seq data
Each ATAC-seq dataset was processed using the ENCODE ATAC-seq pipeline implemented with Caper (https://github.com/ENCODE-DCC/atac-seq-pipeline) (Jin Lee, 2016). Briefly, reads were mapped to the human reference genome (GRCH38/hg38) using Bowtie2 (version 2.3.4.3), and unmapped reads, duplicates, and those mapped to the mitochondrial genome were removed. Peaks were called by MACS2 (Zhang et al., 2008) and optimal peaks that were reproducible across pseudo replicates were used in the downstream analysis. The numbers of processed reads and peaks are summarized in Table S9. Plots of ATAC-seq signals around the TSSs of expressed genes were generated by the R genomation package (version 1.22.0) (Akalin, Franke, Vlahovicek, Mason, & Schubeler, 2015). The batch effects of the signals were corrected by the removeBatchEffect function from the limma R package (version 3.46.0) (Ritchie et al., 2015). Of the broad and narrow peaks resulting from the ENCODE ATAC-seq pipeline, the latter were used as an input to obtain consensus ACRs using the diffBind R package (version 3.0.15) (Ross-Innes et al., 2012). The dACRs were detected using the edgeR R package (version 3.32.1) (Robinson et al., 2010). In total, 70,746 ACRs and 5,219 dACRs were detected in HeLa cells and are summarized in Supplementary Table S10. The genomic positions of the ACRs were annotated through the ChIPseeker R package (version 1.26.2) (Yu, Wang, & He, 2015). If the ACRs spanned more than one genomic region, their positions were assigned based on the following priority: promoters > 5’ untranslated regions (UTRs) > 3’UTRs > other exons > introns > downstream > intergenic regions. The promoter of a gene was defined as the region 5 kb upstream and 500 bp downstream of the TSS.
Nuclear–cytoplasmic fractionation
Prior to transfection, HeLa cells were seeded in 100 mm cell culture dishes containing Opti-MEM medium and incubated overnight (reaching a confluency of approximately 30%- 40%). The cells were then transfected with siTXNIP. Cells were harvested after 48 hr of transfection and fractionated according to the manufacturer’s instructions using NE-PER Nuclear and Cytoplasmic Extraction Reagents (#78833, Thermo Fisher Scientific). Protease inhibitor cocktail (P8340; Sigma Aldrich) was added as a supplement to the lysis buffer and the protein concentration was measured using a Pierce BCA Protein Assay Kit (#23225, Thermo Fisher Scientific).
Chromatin Immunoprecipitation (ChIP) Assay
Cells were crosslinked with 1% formaldehyde at 37°C or room temperature for 15 min and the reaction was stopped by the addition of 0.125M glycine. ChIP was then performed using a ChIP-IT High Sensitivity kit (#53040, Active Motif, Carlsbad, CA, USA) according to the manufacturer’s instructions. Enrichment of the ChIP signal was detected by quantitative real-time PCR (qPCR). The data of each biological replicate were normalized with negative control IgG signals and enrichment values were calculated using the ΔΔCt method (Hellemans, Mortier, De Paepe, Speleman, & Vandesompele, 2007). The following antibodies were used: TXNIP (14715, Cell Signaling Technology, Beverly, MA), HDAC2 (57156, Cell Signaling Technology), H3ac antibody (39139; Active Motif, Carlsbad, CA), and normal rabbit IgG antibodies were used. The primers used for ChIP-qPCR are summarized in Table S12.
Osteoclast differentiation and collection of lentiviruses for ARRDC5 expression
BMMs were cultured as previously described (S. Y. Kim et al., 2019). Briefly, bone marrow was obtained from mouse femurs and tibias at 8 weeks of age, and BMMs were isolated from the bone marrow using Histopaque (1077; Sigma Aldrich). BMMs were seeded at a density of 1.2 × 105 cells/well into 24-well culture plates and incubated in α-MEM (SH30265.01; Hyclone, Rockford, IL, USA) containing 20 ng/mL macrophage colony stimulating factor (M-CSF) (300-25; PeproTech, Cranbury, NJ, USA). To induce osteoclast differentiation, BMMs were treated for 24 hr with lentiviral-containing medium that also contained M-CSF, after which the medium was changed to α-MEM containing 20 ng/ml M CSF and 20 ng/ml RANKL (462-TEC; R&D Systems, Minneapolis, MN, USA). The differentiation medium was changed every 24 hr during the 5-day differentiation period.
To obtain the media containing lentivirus, HEK293 cells were cultured in DMEM containing 4.5 g/L glucose (SH30243.01; Hyclone) supplemented with 10% FBS (SH30084.03; Hyclone) and 1% penicillin-streptomycin. After seeding cells at a density of 1 × 105 cells/well into 6-well culture plates, the cells were incubated with lentivirus co-transfected media for 16 hr. Lentivirus co-transfected media was prepared according to the manufacturer’s instructions using the CRISPR & MISSION® Lentiviral Packaging Mix (SHP002; Sigma Aldrich) and the lentiviral transfer vector, pHAGE-GFP-GFP or pHAGE-GFP-ARRDC5. After the incubation, the medium was replaced with fresh α-MEM medium supplemented with 10% FBS and 1% penicillin-streptomycin. The medium was collected twice (after 24 and 48 hr), designated as lentiviral-containing medium, and stored in a deep freezer until used to infect BMMs.
TRAP staining and bone resorption pit assay
Osteoclast differentiation and activity were determined by TRAP staining and a bone resorption pit assay, respectively. TRAP staining was performed using a TRAP staining kit (PMC-AK04F-COS; Cosmo Bio Co., LTD., Tokyo, Japan) following the manufacturer’s instructions. TRAP-positive multinucleated cells with more than three nuclei were counted under a microscope using ImageJ software (NIH, Bethesda, MD, USA). The bone resorption pit assay was performed using dentin discs (IDS AE-8050; Immunodiagnostic Systems, Tyne & Wear, UK). Cells were differentiated to osteoclasts on the discs over a 4-day period, after which the discs were stained with 1% toluidine blue solution and the resorption pit area was quantified using ImageJ software.
Immunofluorescence staining of the V-type ATPase and visualization with GFP ARRDC5
To inhibit V-type ATPase transport to the membrane (Matsumoto et al., 2019), osteoclasts on the fifth day of differentiation were incubated with 100 nM bafilomycin A1 (19- 148; Sigma Aldrich) for 3 hr. Then, immunofluorescence staining was performed to visualize the localization of the V-type ATPase in bafilomycin A1-treated and untreated cells. The cells were fixed using a 4% paraformaldehyde solution (PC2031-100; Biosesang, Gyeonggi-do, Korea) and permeabilized using 0.05% Triton X-100 at room temperature for 5 min. The cells were incubated with anti-V-type ATPase antibody (SAB1402125-100UG; Sigma Aldrich) at room temperature for 1 hr, and then stained with the Alexa Fluor 594-conjugated anti-mouse antibody (A-21044; Invitrogen) at room temperature for 30 min. Finally, cells were mounted using Antifade Mountant with DAPI (P36962; Invitrogen). Fluorescence images were observed under a ZEISS confocal microscope (LSM5; Carl Zeiss, Jena, Germany).
Data availability
All AP/MS raw tables from human and Drosophila α-arrestins are deposited in Source Data. ATAC-seq and RNA-seq data from HeLa cells treated with siCon and siTXNIP, can be downloaded from the Korean Nucleotide Archive (KONA; PRJKA220517, https://www.kobic.re.kr/kona/).
Code availability
All source codes and in-house codes used in the study are available at GitHub (https://github.com/Kyung-TaeLee/alphaArrestins_PPI_network)
Acknowledgements
We thank all of the BIGLab and Kwon Lab members for critical reading and comments. This work was supported by the National Research Foundation (NRF) funded by the Ministry of Science & ICT (2020R1A4A1018398, 2021R1A2C3005835, and 2022M3E5F1018502 to JWN, and 2020R1A6A3A13077354 to KTL).
Supplementary Information
Supplementary Methods
Immunofluorescence imaging of human α-arrestins
Stably α-arrestin-GFP expressing HEK293 cells were cultured in a 12 well-plate with pre-sterilized round glass coverslips in each well. Cells on coverslip were fixed in 4% paraformaldehyde (PFA) (RT15710, Electron Microscopy Sciences, Hatfiled, PA, USA) diluted in PBS for 30 min and then washed three times with PBST (PBS supplemented with 0.2% Triton X-100) with 5 min intervals. To label the nucleus, samples were stained with DAPI (1:5000; D9542, Sigma Aldrich) in PBST supplemented with 1% BSA (A7906, Sigma Aldrich) for 1 hr at room temperature. Stained cells samples were washed three times with PBST and preserved in Vectashield (H-1000, Vector Laboratories, Burlingame, CA, USA). Fluorescence images were acquired using an Olympus FV1200 confocal microscope with 40X oil objective lens and 2X zoom factor. NIH ImageJ software was used for further adjustment and assembly of the acquired images.
Database searching and analysis of mass spectrometry data
MS/MS spectra were queried using the Comet search engine (Eng et al., 2013) to search for corresponding proteins in Flybase (Gramates et al., 2017) and Uniprot (The UniProt, 2017). Common contaminant protein sequences from the Common Repository of Adventitious Proteins (cRAP) Database (ftp://ftp.thegpm.org/fasta/cRAP) were used to filter contaminating sequences. Searching was done with following parameters: tryptic digest, internal decoy peptides, the number of missed cleavages=2, precursor tolerance allowing for isotope offsets=20 ppm, a 1.00 fragment bin tolerance, static modification of 57.02 on cysteine, and variable modification of 16.00 on methionine. The acetylation, phosphorylation, and ubiquitination searches add variable modifications of 42.01 on lysine, 79.97 on serine/threonine/tyrosine, and 114.04 on lysine, respectively. The search results were then processed through the Trans-Proteomic Pipeline suite of tools version 4.8.0 (Keller et al., 2005) where the PeptideProphet tool (Keller et al., 2002) was applied to calculate the probability that each search result is correct and the ProteinProphet tool (Nesvizhskii et al., 2003) was applied to infer protein identifications and their probabilities.
Functional annotations and multiple sequence alignment of α-arrestin sequences
The sequences of twelve Drosophila and six human α-arrestins were retrieved from the Uniprot database (UniProt Consortium, 2018). Domains and motifs including the PPxY motif were annotated based on sequences from Pfam version 31.0 (El-Gebali et al., 2019) and the eukaryotic linear motif (ELM) database (Dinkel et al., 2015). The sequences were subjected to the multiple-sequence alignment tool T-COFFEE (Notredame et al., 2000) using default parameters. The output of T-COFFEE was applied to RAxML (version 8.2.11) (Stamatakis, 2014) to generate a consensus phylogenetic tree with 1,000 rapid bootstrapping using “-m PROTGAMMAWAGF” as the parameter (https://cme.h-its.org/exelixis/resource/download/ NewManual.pdf).
Checking the reproducibility of spectral counts among replicates
If multiple proteins isoforms were detected, they were collapsed into a single gene. To avoid the divide-by-zero error, spectral counts of “0” were converted to a minimum non-zero value, “0.01”. To examine the integrity and quality of spectral counts from the AP/MS, the average correlation coefficients (Pearson) of spectral counts from α-arrestins were calculated and plotted. At each cutoff of spectra counts from 1 to 15, only the PPIs with spectral counts that were the same or higher than the cutoff for all replicates were kept and used to calculate correlation coefficients between replicates. The resulting coefficients from the α-arrestin interactomes were then averaged and plotted. At the cutoff of 6 spectral counts, saturation of average correlation coefficients was observed and chosen as an optimal cutoff to filter the PPIs. Principal component analysis (PCA) of the filtered PPIs was conducted based on spectral counts (with a pseudo count 1 added) transformed into a log2 using the factoextra R package (version 1.0.7).
Hierarchical clustering of high-confidence PPIs
Hierarchical clustering based on log2 spectral counts (pseudo count 1 added) of high confidence PPIs was conducted using the Pearson correlation as the clustering distance and Ward’s method as the clustering method. Heatmaps were visualized through the ComplexHeatmap R package (version 2.6.2) (Gu et al., 2016). Six clusters were identified for each species based on the results of hierarchical clustering; the PANTHER protein class overrepresentation test was performed for the proteins in each cluster (Thomas et al., 2003). False discovery rates (FDRs, Fisher’s exact test) of indicated protein classes were ≤ 0.05 for all classes except for “GTPase-activating protein” in human (FDR < 0.133) and “GEFs” in Drosophila (FDR < 0.109), respectively. Interacting prey proteins from the positive PPIs were selectively labeled.
Domain and motif analysis of bait and prey proteins
For human and Drosophila, respectively, 53 and 65 short linear motifs in α-arrestins were annotated using the ELM database (Dinkel et al., 2015), and 423 and 546 protein domains in prey proteins were annotated using the Uniprot database (UniProt Consortium, 2018) (Table S4). To test for enrichment of protein domains, we implemented the Expression Analysis Systematic Explorer (EASE) score (Hosack et al., 2003), which is calculated by subtracting one gene within the query domain and conducting a one-sided Fisher’s exact test. Protein domains enriched in the interactomes of each α-arrestin (Benjamini-Hochberg FDR ≤ 0.05) were plotted using the ComplexHeatmap R package (version 2.6.2). Next, to see how reliable our filtered PPIs were, we utilized information about known affinities between domains and short linear motifs from the ELM database (Dinkel et al., 2015). Because the arrestin_N (Pfam ID : PF00339) and arrestin_C (Pfam Id : PF02752) domains in α-arrestins do not have known interactions with any of the short linear motifs in the ELM database (Dinkel et al., 2015), only the interactions between the short linear motifs in α-arrestins and protein domains in the interactome (prey proteins) were considered in this analysis. We found that 59 out of the 390 human PPIs and 64 out of the 740 Drosophila PPIs were supported by such known affinities (Table S4). One-sided Fisher’s exact test was used to test the significance of the enrichment of the supported PPIs in the filtered PPI sets versus those in the unfiltered PPI sets (Figure 1D).
Subcellular localizations of bait and prey proteins
To search for annotated subcellular localizations of the proteins in the α-arrestin interactomes, we first obtained annotation files of cellular components (Gene Ontology (GO) : CC) for human and Drosophila from the Gene Ontology Consortium (Ashburner et al., 2000). From the annotations, we only utilized GO terms for 11 subcellular localizations (name of subcellular localization – GO term ID: Cytosol – GO:0005829; Plasma membrane – GO:0005886; Nucleus – GO:0005634; Mitochondrion – GO:0005739; Endoplasmic reticulum – GO:0005783; Golgi apparatus – GO:0005794; Cytoskeleton – GO:0005856; Peroxisome – GO:0005777; Lysosome – GO:0005764; Endosome – GO:0005768; Extracellular space – GO:0005615). If a protein was annotated to be localized in multiple locations, a weighted value (1/the number of multiple localizations) was assigned to each location. Finally, the relative frequencies of the subcellular localizations associated with the interacting proteins in the filtered PPIs were plotted for each α-arrestin (Figure S3B).
Immunoblotting and co-immunoprecipitation Assays
Cells were lysed in radioimmunoprecipitation assay (RIPA) buffer supplemented with protease inhibitor. For immunoblotting, the cell lysates were separated by 10% SDS polyacrylamide gel electrophoresis (PAGE) and transferred to nitrocellulose membranes. After blocking membranes with 5% skim milk in Tris buffered Saline containing 0.1% Tween-20 (TBS T) for 2 hours (hr) at room temperature, the nitrocellulose membranes were incubated with appropriate primary antibodies overnight at 4°C and subsequently reacted with horseradish peroxidase (HRP)-conjugated secondary antibodies for 1 hr at room temperature. Bands were visualized using an enhanced chemiluminescence (ECL) detection system, West-Q Pico ECL Solution (W3652-02, GenDEPOT, Katy, TX, USA). For quantification of immunoblot results, the densities of target protein bands were analyzed with Image J.
For immunoprecipitation, the cell lysates (2 mg) were incubated with appropriate antibodies (1 µg) overnight at 4°C and precipitated with TrueBlot Anti-Rabbit Ig IP agarose beads (Rockland, Philadelphia, PA) for 2 hr at 4°C. The immunocomplexes were washed with chilled PBS three times and heated with 3x sample loading buffer containing ß-mercaptoethanol. The samples were separated by 6-8 % SDS-polyacrylamide gel electrophoresis (PAGE) and immunoblot was performed as described above.
The following antibodies were used for immunoblotting and co-immunoprecipitation assays: anti-TXNIP (#14715), anti-HDAC2 (#57156) and anti-alpha Tubulin (#3873) were obtained from Cell Signaling Technology (Beverly, MA); anti-H3ac (39139) was obtained from Active Motif (Carlsbad, CA); anti-ß-actin (GTX629630) was obtained from GeneTex; normal anti-rabbit IgG (sc-2027) was obtained from Santa Cruz Biotechnology (Dallas, TX); TrueBlot anti-rabbit IgG HRP (18-8816-31) was obtained from Rockland (Philadelphia, PA).
Quantitative Reverse-transcription polymerase chain reaction (PCR)
Total RNA was isolated using TRIzol reagent (#15596018, Invitrogen, Carlsbad, CA, USA; Thermo Fisher Scientific) and subjected to reverse transcription PCR (RT-PCR) with ReverTra Ace qPCR RT kit (#FSQ-101, Toyobo, Osaka, Japan) or GoScript RT-PCR system (#A5001, Promega, Madison, WI, USA) according to the manufacturer’s instructions. The mRNA expression levels of target genes were quantified using the CFX Opus 96 (Biorad, Hercules, CA) or Applied Biosystems QuantStudio 1 (Applied Biosystems, Foster city, CA) real time PCR. AccuPower 2X GreenStar™ qPCR Master Mix (#K6251, Bioneer, Daejeon, Republic of Korea) or SYBR Green Realtime PCR Master Mix (#QPK-201, Toyobo, Osaka, Japan) were applied according to the manufacturer’s protocols. The data normalized by GAPDH or alpha tubulin mRNA levels and calculated using the ΔΔCt method (Hellemans et al., 2007). The primers used for qRT-PCR analysis are summarized in Table S12.
PCA of ATAC- and RNA-seq data
For ATAC-seq, normalized read counts derived from the diffBind R package (version 3.0.15) (Ross-Innes et al., 2012) were transformed into a log2 function. Batch effect corrections were done using the limma R package (version 3.46.0) (Ritchie et al., 2015). For RNA-seq, counts per million mapped reads (CPM) were also processed in the same manner. For PCA, 2,000 features with the highest variance across samples were extracted and utilized. Plots of principal components 1 and 2 were generated by the factoextra R package (version 1.0.7).
Functional signatures of repressed genes upon TXNIP depletion
Genes that exhibited decreased chromatin accessibility at their promoter and decreased RNA expression upon TXNIP knockdown (Table S11) were selected based on the following criteria: 1. log2 (RNA level in siTXNIP-treated cells/RNA level in siCon-treated cells) (hereafter, siTXNIP/siCon) ≤ -1; 2. log2 (siTXNIP/siCon) of ACRs in the promoter region ≤ -1 (If there are multiple ACRs in the promoter region, the one with the highest ATAC-seq signal was selected) or log2 mean (siTXNIP/siCon) of all ACRs in the promoter region ≤ -1. Enrichment analysis of the GO terms in the gene set was performed by g:Profiler (Raudvere et al., 2019). Top 10 enriched terms from the biological process and molecular functions categories were plotted (Figure 4G).
Immunofluorescence of HDAC2 and TXNIP
HeLa cells were cultured in 6-well plates with cover slips in each well (1.5 x104 cells/well). After cells were incubated overnight in Opti-MEM, TXNIP knockdown was induced by transfection of siRNA at a concentration of 100 nM. Following 48 hr of transfection, the cells were washed twice with PBS and then fixed with 100% ice-cold methanol for 10 min at -20°C. After rinsing three with PBSTw (PBS containing 0.1% Tween 20), the cells were blocked with 3% BSA in PBS and incubated for 45 min at room temperature. Next, cells were incubated with the primary antibody for 150 min followed by the secondary antibody for 60 min in the dark. For co staining with a second primary antibody, the blocking step followed by the primary and secondary antibody incubation steps were repeated. All of the antibodies were diluted in antibody dilution buffer (1% BSA in PBS). Information of the antibodies are listed in “antibody” section in STAR Method. The cover slips were rinsed three times with PBSTw and then mounted with VECTASHIELD Antifade Mounting Medium containing DAPI (Vector Laboratories, Newark, CA, USA) according to the manufacturer’s instructions. The fluorescence was visualized with a Nikon C2 Si-plus confocal microscope.
Supplementary figures
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_figs1.tif/full/max/0/default.jpg)
Fluorescence images showing HEK293 and S2R+ cells stably expressing GFP tagged α-arrestins
Representative images of HEK293 (A) and S2R+ (B) cells stably expressing GFP-tagged α- arrestins.
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_figs2.tif/full/max/0/default.jpg)
Affinity purification / mass spectrometry (AP/MS) data are highly reproducible and recapitulate known PPIs
(A) Average Pearson correlation coefficients of log2 spectral counts between replicates of AP/MS of each α-arrestin at varying cutoffs are shown (mean ± standard deviation(sd)). The cutoff used in this study, 6, is shown as a dashed line. (B) PCA plots based on log2 spectral counts of high-confidence PPIs for human (upper) and Drosophila (lower) are shown. (C) SAINTexpress scores and average spectral counts (log2) of the positive PPIs (Table S2A and C) are shown and density plots for each axis are also plotted. The positive PPIs that are included in the filtered set are selectively labeled.
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_figs3.tif/full/max/0/default.jpg)
Protein domains and subcellular localization of α-arrestin interactomes
(A) Protein domains enriched in each α-arrestin interactome for human (top) and Drosophila (bottom) are shown. The significance of the enrichment test (-log10 FDR) is indicated in shades of green, as depicted in the legend. SPOC, spen paralogue and orthologue C-terminal; MCM, minichromosome maintenance protein complex; FDRM, F for 4.1 protein, E for ezrin, R for radixin and M for moesin; TBP, TATA binding protein; GEF, guanine nucleotide exchange factor; THRAP3, thyroid hormone receptor-associated protein 3; BCLAF1, Bcl-2-associated transcription factor1; RMMBL, RNA metabolising metallo beta lactamase; CaMKII, C-terminus of the Calcium/calmodulin dependent protein kinases II; CPSF, cleavage and polyadenylation specificity factor; DCB, dimerization and cyclophilin-binding domain; FRAP, FKBP12-rapamycin complex-associated protein; ATM, ataxia telangiectasia mutant; THRAP, transformation/transcription domain associated proteins; MIF4G, middle domain of eukaryotic initiation factor 4G; AAA, ATPase family associated with various cellular activities; C4, C terminal tandem repeated domain in type 4 procollagen; SMC, structural maintenance of chromosomes. (B) Subcellular localizations of prey proteins of each α-arrestin for human (left) and Drosophila (right).
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_figs4.tif/full/max/0/default.jpg)
Phylogenetic tree showing relationships between α-arrestins from human and Drosophila
Phylogenetic tree of α-arrestins from both species based on protein sequences were drawn as in Figure 1A.
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_figs5.tif/full/max/0/default.jpg)
ATAC- and RNA-seq results are highly reproducible and ATAC-seq results exhibit a pattern typical of strong enrichment around TSSs of expressed genes
(A and B) PCA plots of ATAC- (A) and RNA-seq (B) results based on batch-corrected log2 counts and CPM, respectively. Numbers in parentheses are percentages of explained variance for the corresponding PCs. (C) Heatmaps of ATAC-seq read counts (read counts have been transformed into a log2 function and corrected for batch effects) in regions surrounding TSSs along with log2 (RNA level in siTXNIP-treated cells/RNA level in siCon-treated cells) for genes having the corresponding TSS are plotted for each sample.
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_figs6.tif/full/max/0/default.jpg)
Genomic locations of ACRs and association between chromatin landscapes and transcriptional activity
(A) Genomic locations of 70,746 consensus ACRs identified from ATAC-seq analysis. (B) Composition of dACRs(-), dACRs(+), and other ACRs (“others”, not significantly changed) under the TXNIP knockdown condition compared to the control. (C) Genomic locations of 4,825 dACRs(-) and 394 dACRs(+) are depicted. Colors in the bar plot have the same symbolism as in (A). (D) Cumulative distribution function (CDF) of mean changes in accessibility of all ACRs located in gene promoters. The genes were categorized into three groups (“None”, “Down”, and “Up”) as explained in Figure 4F. P values on the left upper corner were calculated with the one sided Kolmogorov-Smirnov (KS) test comparing “Down” or “Up” groups to the “None” group. (E) CDF of changes in accessibility of ACRs located in gene bodies. Changes in accessibility of ACRs whose intensity is highest among all ACRs located in gene bodies are depicted on the left and mean changes in accessibility of all ACRs located in gene bodies are depicted on the right. P values on the upper left corners are calculated in the same manner as in (D).
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_figs7.tif/full/max/0/default.jpg)
TXNIP depletion does not affect the protein level or subcellular localization of HDAC2 but represses transcription of target genes
(A) Representative immunofluorescence images of TXNIP and HDAC2 after HeLa cells were transfected with either siCon or siTXNIP for 48 hr (magnification ×600); TXNIP (red), HDAC2 (green), and DAPI (blue). (B) RT-qPCR results of four target genes whose RNA expression and chromatin accessibility in their promoters, quantified using high-throughput sequencing data, were observed to be strongly repressed in HeLa cell. Data are presented as the mean ± sd, n=3). Gray dots depict actual values of each experiment. *P < 0.05, ns: not significant (two sided paired Student T test).
Supplementary Table and legends
Table S1. Information about α-arrestin proteins from human (A) and Drosophila (B).
Table S2. Positive and negative PPIs of α-arrestins ffor human (A and B) and Drosophila (C and D).
Table S3. Output tables of SAINTexpress run on AP/MS data for human (A) and Drosophila (B).
Table S4. Protein domains and short linear motifs annotated in the interactome of each α- arrestin of human (A and B) and Drosophila (D and E). Annotated interactions between the short linear motifs in α-arrestins and protein domains in the interactome from the ELM database are also summarized for human (C) and Drosophila (F).
Table S5. Results of enrichment test of Pfam domains in the interactome of each α-arrestin for human (A) and Drosophila (B).
Table S6. Summary tables of subcellular localizations assigned to proteins in the interactomes of α-arrestins for human (A) and Drosophila (B).
Table S7. Results of the protein complex enrichment analysis tool (COMPLEAT) and enrichment test of cellular component GO terms for human (A and B) and Drosophila (D and E). Protein complexes of two databases were iteratively clustered and are summarized in (C) and (F) for human and Drosophila, respectively. Information of clustered protein complexes from (C and F) were utilized to plot network of protein complexes in Figure 2.
Table S8. DIOPT predictions of orthologs of α-arrestins and proteins in their interactomes (A and B). Results of enrichment test of GO terms (biological process, molecular functions and BP, MF, and Kyoto Encyclopeida of Genes and Genomes pathway are also summarized for orthologs in human (C) and Drosophila (D).
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_tbls1.tif/full/max/0/default.jpg)
Summary of ATAC- and RNA-seq read counts before and after processing. For ATAC-seq, the number of properly paired reads, filtered/deduplicated reads, and identified narrow peaks are summarized. For RNA-seq, the number of filtered and alignable reads are summarized. *Filtered/dedup reads, filtered/deduplicated reads
Table S10. Result of analysis of differential accessibility of ACRs (A) and differential gene expression (B).
Table S11. (A) Profiles of ATAC-seq peaks located in promoters of genes. Changes in peak intensities and gene expression levels are also summarized. (B) List of genes that exhibited decreased chromatin accessibility at their promoter and decreased RNA expression upon TXNIP knockdown.
![](https://prod--epp.elifesciences.org/iiif/2/88328%2Fv1%2Fcontent%2F544486v1_tbls2.tif/full/max/0/default.jpg)
List of primer sequences used in this study.
References
- Genomation: a toolkit to summarize, annotate and visualize genomic intervalsBioinformatics 31:1127–1129https://doi.org/10.1093/bioinformatics/btu775
- On the origins of arrestin and rhodopsinBMC Evol Biol 8https://doi.org/10.1186/1471-2148-8-222
- PY motifs of Rod1 are required for binding to Rsp5 and for drug resistanceFEBS Lett 525:131–134https://doi.org/10.1016/s0014-5793(02)03104-6
- Andrews, S. (2010). A Quality Control Tool for High Throughput Sequence Data [Online]. Available online at: http://www.bioinformatics.babraham.ac.uk/projects/fastqc/.
- Arrestin domain-containing 3 (Arrdc3) modulates insulin action and glucose metabolism in liverProc Natl Acad Sci U S A 117:6733–6740https://doi.org/10.1073/pnas.1922370117
- Beta adrenergic receptor kinase: primary structure delineates a multigene familyScience 246:235–240https://doi.org/10.1126/science.2552582
- A role for creD, a carbon catabolite repression gene from Aspergillus nidulans, in ubiquitinationMol Microbiol 53:929–940https://doi.org/10.1111/j.1365-2958.2004.04172.x
- Isolation and characterization of a novel cDNA from HL-60 cells treated with 1,25-dihydroxyvitamin D-3Biochim Biophys Acta 1219:26–32https://doi.org/10.1016/0167-4781(94)90242-9
- Research Progress of TXNIP as a Tumor Suppressor Gene Participating in the Metabolic Reprogramming and Oxidative Stress of Cancer Cells in Various CancersFront Oncol 10https://doi.org/10.3389/fonc.2020.568574
- Functional proteomics mapping of a human signaling pathwayGenome Research 14:1324–1332https://doi.org/10.1101/gr.2334104
- Beta-arrestins and cell signalingAnnu Rev Physiol 69:483–510https://doi.org/10.1146/annurev.physiol.69.022405.154749
- Beta4 integrin subunit gene expression correlates with tumor size and nuclear grade in early breast cancerMod Pathol 18:1165–1175https://doi.org/10.1038/modpathol.3800411
- ELM 2016—data update and new functionality of the eukaryotic linear motif resourceNucleic acids research 44:D294–D300
- STAR: ultrafast universal RNA-seq alignerBioinformatics 29:15–21https://doi.org/10.1093/bioinformatics/bts635
- The Emerging Role of TXNIP in Ischemic and Cardiovascular Diseases; A Novel Marker and Therapeutic TargetInt J Mol Sci 22https://doi.org/10.3390/ijms22041693
- The alpha-arrestin ARRDC3 mediates ALIX ubiquitination and G protein-coupled receptor lysosomal sortingMol Biol Cell 26:4660–4673https://doi.org/10.1091/mbc.E15-05-0284
- Diabetes regulates fructose absorption through thioredoxin-interacting proteinElife 5
- ARRDC3 suppresses breast cancer progression by negatively regulating integrin beta4Oncogene 29:5032–5047https://doi.org/10.1038/onc.2010.250
- Atp6v1c1 is an essential component of the osteoclast proton pump and in F-actin ring formation in osteoclastsBiochemical Journal 417:195–203https://doi.org/10.1042/Bj20081073
- GENCODE reference annotation for the human and mouse genomesNucleic Acids Res 47:D766–D773https://doi.org/10.1093/nar/gky955
- A protein complex network of Drosophila melanogasterCell 147:690–703https://doi.org/10.1016/j.cell.2011.08.047
- VDUP1 upregulated by TGF-beta1 and 1,25-dihydorxyvitamin D3 inhibits tumor cell growth by blocking cell-cycle progressionOncogene 22:4035–4046https://doi.org/10.1038/sj.onc.1206610
- Distinct roles for beta-arrestin2 and arrestin-domain-containing proteins in beta2 adrenergic receptor traffickingEMBO Rep 14:164–171https://doi.org/10.1038/embor.2012.187
- From molecular to modular cell biologyNature 402:C47–52https://doi.org/10.1038/35011540
- qBase relative quantification framework and software for management and automated analysis of real-time quantitative PCR dataGenome Biol 8https://doi.org/10.1186/gb-2007-8-2-r19
- Arrestin-related proteins mediate pH signaling in fungiProc Natl Acad Sci U S A 102:12141–12146https://doi.org/10.1073/pnas.0504776102
- An integrative approach to ortholog prediction for disease-focused and other functional studiesBMC Bioinformatics 12
- Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene listsNucleic Acids Res 37:1–13https://doi.org/10.1093/nar/gkn923
- Systematic and integrative analysis of large gene lists using DAVID bioinformatics resourcesNat Protoc 4:44–57https://doi.org/10.1038/nprot.2008.211
- The BioPlex Network: A Systematic Exploration of the Human InteractomeCell 162:425–440https://doi.org/10.1016/j.cell.2015.06.043
- Jin Lee, G. C., Grey Christoforo, CS Foo, Chris Probert, Anshul Kundaje, Nathan Boley, kohpangwei, Mike Dacre, Daniel Kim. (2016). kundajelab/atac_dnase_pipelines: 0.3.3.
- Joshi NA, F. J. (2011). Sickle: A sliding-window, adaptive, quality-based trimming tool for FastQ files (Version 1.33) [Software]. Available at https://github.com/najoshi/sickle.
- Vitamin D3 up-regulated protein 1 mediates oxidative stress via suppressing the thioredoxin functionJ Immunol 164:6287–6295https://doi.org/10.4049/jimmunol.164.12.6287
- TXNIP-mediated nuclear factor-kappaB signaling pathway and intracellular shifting of TXNIP in uric acid-induced NLRP3 inflammasomeBiochem Biophys Res Commun 511:725–731https://doi.org/10.1016/j.bbrc.2019.02.141
- Ablation of Stabilin-1 Enhances Bone-Resorbing Activity in Osteoclasts In VitroCalcif Tissue Int 105:205–214https://doi.org/10.1007/s00223-019-00552-x
- Differential roles of arrestin-2 interaction with clathrin and adaptor protein 2 in G protein-coupled receptor traffickingJ Biol Chem 277:30760–30768https://doi.org/10.1074/jbc.M204528200
- Light-induced binding of 48-kDa protein to photoreceptor membranes is highly enhanced by phosphorylation of rhodopsinFEBS Lett 176:473–478https://doi.org/10.1016/0014-5793(84)81221-1
- Vitamin D3 upregulated protein 1 suppresses TNF-alpha-induced NF-kappaB activation in hepatocarcinogenesisJ Immunol 185:3980–3989https://doi.org/10.4049/jimmunol.1000990
- The Hippo signaling pathway interactomeScience 342:737–740https://doi.org/10.1126/science.1243971
- Tandem affinity purification in Drosophila: the advantages of the GS-TAP systemFly (Austin 2:229–235https://doi.org/10.4161/fly.6669
- RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genomeBMC Bioinformatics 12https://doi.org/10.1186/1471-2105-12-323
- Stable overexpression of human beta 2-adrenergic receptors in mammalian cellsNaunyn Schmiedebergs Arch Pharmacol 345:444–451https://doi.org/10.1007/BF00176623
- Analysis of integrin beta4 expression in human breast cancer: association with basal-like tumors and prognostic significanceClin Cancer Res 14:1050–1058https://doi.org/10.1158/1078-0432.CCR-07-4116
- Role of spectral counting in quantitative proteomicsExpert review of proteomics 7:39–53
- Structure of the WW domain of a kinase-associated protein complexed with a proline rich peptideNature 382:646–649https://doi.org/10.1038/382646a0
- Arrestin-2 interacts with the endosomal sorting complex required for transport machinery to modulate endosomal sorting of CXCR4Mol Biol Cell 21:2529–2541https://doi.org/10.1091/mbc.E10-02-0169
- Proton pumping V-ATPase inhibitor bafilomycin A1 affects Rab7 lysosomal localization and abolishes anterograde trafficking of osteoclast secretory lysosomesBiochemical and Biophysical Research Communications 510:421–426https://doi.org/10.1016/j.bbrc.2019.01.118
- The CRAPome: a contaminant repository for affinity purification mass spectrometry dataNat Methods 10:730–736https://doi.org/10.1038/nmeth.2557
- The Hippo signaling pathway in stem cell biology and cancerEMBO Rep 15:642–656https://doi.org/10.15252/embr.201438638
- An in vivo screen identifies ependymoma oncogenes and tumor-suppressor genesNat Genet 47:878–887https://doi.org/10.1038/ng.3323
- Formation and release of arrestin domain-containing protein 1-mediated microvesicles (ARMMs) at plasma membrane by recruitment of TSG101 proteinProc Natl Acad Sci U S A 109:4146–4151https://doi.org/10.1073/pnas.1200448109
- Arrestin domain-containing protein 3 recruits the NEDD4 E3 ligase to mediate ubiquitination of the beta2-adrenergic receptorEMBO Rep 11:605–611https://doi.org/10.1038/embor.2010.80
- Importin alpha1 (Rch1) mediates nuclear translocation of thioredoxin-binding protein 2/vitamin D(3)-up-regulated protein 1J Biol Chem 279:37559–37565https://doi.org/10.1074/jbc.M405473200
- Identification of thioredoxin-binding protein-2/vitamin D-3 up-regulated protein 1 as a negative regulator of thioredoxin function and expressionJournal of Biological Chemistry 274:21645–21650https://doi.org/10.1074/jbc.274.31.21645
- Thioredoxin-binding protein-2-like inducible membrane protein is a novel vitamin D3 and peroxisome proliferator-activated receptor (PPAR)gamma ligand target protein that regulates PPARgamma signalingEndocrinology 147:733–743https://doi.org/10.1210/en.2005-0679
- Comparison of label-free methods for quantifying human proteins by shotgun proteomicsMol Cell Proteomics 4:1487–1502https://doi.org/10.1074/mcp.M500084-MCP200
- The interaction of thioredoxin with Txnip. Evidence for formation of a mixed disulfide by disulfide exchangeJ Biol Chem 281:21884–21891https://doi.org/10.1074/jbc.M600427200
- YAP is a critical oncogene in human cholangiocarcinomaOncotarget 6:17206–17220https://doi.org/10.18632/oncotarget.4043
- Alpha-arrestins - new players in Notch and GPCR signaling pathways in mammalsJ Cell Sci 127:1359–1367https://doi.org/10.1242/jcs.142539
- Alpha-arrestin 1 (ARRDC1) and beta-arrestins cooperate to mediate Notch degradation in mammalsJ Cell Sci 126:4457–4468https://doi.org/10.1242/jcs.130500
- Role of Thioredoxin-Interacting Protein in Diseases and Its Therapeutic OutlookInt J Mol Sci 22https://doi.org/10.3390/ijms22052754
- V-ATPases in osteoclasts: structure, function and potential inhibitors of bone resorptionInt J Biochem Cell Biol 44:1422–1435https://doi.org/10.1016/j.biocel.2012.05.014
- Multiple interactions between the ESCRT machinery and arrestin-related proteins: implications for PPXY-dependent buddingJ Virol 85:3546–3556https://doi.org/10.1128/JVI.02045-10
- g:Profiler: a web server for functional enrichment analysis and conversions of gene lists (2019 update)Nucleic Acids Res 47:W191–W198https://doi.org/10.1093/nar/gkz369
- limma powers differential expression analyses for RNA-sequencing and microarray studiesNucleic Acids Res 43https://doi.org/10.1093/nar/gkv007
- edgeR: a Bioconductor package for differential expression analysis of digital gene expression dataBioinformatics 26:139–140https://doi.org/10.1093/bioinformatics/btp616
- Differential oestrogen receptor binding is associated with clinical outcome in breast cancerNature 481:389–393https://doi.org/10.1038/nature10730
- Intracellular shuttling and mitochondrial function of thioredoxin-interacting proteinJ Biol Chem 285:3997–4005https://doi.org/10.1074/jbc.M109.034421
- Hippo Signaling Mediates Proliferation, Invasiveness, and Metastatic Potential of Clear Cell Renal Cell CarcinomaTranslational Oncology 7:309–321https://doi.org/10.1016/j.tranon.2014.02.005
- Cytoscape: a software environment for integrated models of biomolecular interaction networksGenome Research 13:2498–2504https://doi.org/10.1101/gr.1239303
- Mammalian alpha arrestins link activated seven transmembrane receptors to Nedd4 family e3 ubiquitin ligases and interact with beta arrestinsPLoS One 7https://doi.org/10.1371/journal.pone.0050557
- beta-Arrestin-mediated receptor trafficking and signal transductionTrends Pharmacol Sci 32:521–533https://doi.org/10.1016/j.tips.2011.05.002
- STRING v10: protein-protein interaction networks, integrated over the tree of lifeNucleic Acids Res 43:D447–452https://doi.org/10.1093/nar/gku1003
- SAINTexpress: improvements and additional features in Significance Analysis of INTeractome softwareJ Proteomics 100:37–43https://doi.org/10.1016/j.jprot.2013.10.023
- PANTHER: A library of protein families and subfamilies indexed by functionGenome Research 13:2129–2141https://doi.org/10.1101/gr.772403
- beta-arrestins and G protein-coupled receptor traffickingHandb Exp Pharmacol 219:173–186https://doi.org/10.1007/978-3-642-41199-1_9
- From lysosomes to the plasma membrane - Localization of vacuolar type H+-ATPase with the a3 isoform during osteoclast differentiationJournal of Biological Chemistry 278:22023–22030https://doi.org/10.1074/jbc.M302436200
- NDFIP allows NEDD4/NEDD4L-induced AQP2 ubiquitination and degradationPLoS One 12https://doi.org/10.1371/journal.pone.0183774
- Thioredoxin-Interacting Protein (TXNIP) with Focus on Brain and Neurodegenerative DiseasesInt J Mol Sci 21https://doi.org/10.3390/ijms21249357
- UniProt: the universal protein knowledgebaseNucleic Acids Res 46https://doi.org/10.1093/nar/gky092
- A survey on similarity measures in text miningMachine Learning and Applications: An International Journal 3:19–28
- Protein complex-based analysis framework for high-throughput data setsSci Signal 6https://doi.org/10.1126/scisignal.2003629
- An Integrative Analysis of the InR/PI3K/Akt Network Identifies the Dynamic Response to Insulin SignalingCell Rep 16:3062–3074https://doi.org/10.1016/j.celrep.2016.08.029
- ARMMs as a versatile platform for intracellular delivery of macromoleculesNature Communications 9https://doi.org/10.1038/s41467-018-03390-x
- Overexpression of yes-associated protein contributes to progression and poor prognosis of non small-cell lung cancerCancer Science 101:1279–1285https://doi.org/10.1111/j.1349-7006.2010.01511.x
- The GeneMANIA prediction server: biological network integration for gene prioritization and predicting gene functionNucleic Acids Res 38:W214–220https://doi.org/10.1093/nar/gkq537
- HECT E3 Ligases: A Tale With Multiple FacetsFront Physiol 10https://doi.org/10.3389/fphys.2019.00370
- Rapid affinity purification of retinal arrestin (48 kDa protein) via its light-dependent binding to phosphorylated rhodopsinFEBS Lett 207:292–295https://doi.org/10.1016/0014-5793(86)81507-1
- Thioredoxin-Interacting Protein as a Novel Potential Therapeutic Target in Diabetes Mellitus and Its Underlying ComplicationsDiabetes Metab Syndr Obes 13:43–51https://doi.org/10.2147/DMSO.S232221
- AMPK-dependent degradation of TXNIP upon energy stress leads to enhanced glucose uptake via GLUT1Molecular cell 49:1167–1175
- ARRDC1 and ARRDC3 act as tumor suppressors in renal cell carcinoma by facilitating YAP1 degradationAm J Cancer Res 8:132–143
- ChIPseeker: an R/Bioconductor package for ChIP peak annotation, comparison and visualizationBioinformatics 31:2382–2383https://doi.org/10.1093/bioinformatics/btv145
- alpha-Arrestins and Their Functions: From Yeast to Human HealthInt J Mol Sci 23https://doi.org/10.3390/ijms23094988
- Model-based analysis of ChIP-Seq (MACS)Genome Biol 9https://doi.org/10.1186/gb-2008-9-9-r137
- YAP Promotes Breast Cell Proliferation and Survival Partially through Stabilizing the KLF5 Transcription FactorAmerican Journal of Pathology 180:2452–2461https://doi.org/10.1016/j.ajpath.2012.02.025
- Thioredoxin-interacting protein links oxidative stress to inflammasome activationNat Immunol 11:136–140https://doi.org/10.1038/ni.1831
- A 48 kDa protein arrests cGMP phosphodiesterase activation in retinal rod disk membranesFEBS Lett 207:35–41https://doi.org/10.1016/0014-5793(86)80008-4
- Gene ontology: tool for the unification of biologyNature genetics 25
- ELM 2016—data update and new functionality of the eukaryotic linear motif resourceNucleic acids research 44:D294–D300
- The Pfam protein families database in 2019Nucleic Acids Res 47:D427–D432
- Comet: an open-source MS/MS sequence database search toolProteomics 13:22–24
- FlyBase at 25: looking to the futureNucleic Acids Res 45:D663–D671
- Complex heatmaps reveal patterns and correlations in multidimensional genomic dataBioinformatics 32:2847–2849
- qBase relative quantification framework and software for management and automated analysis of real time quantitative PCR dataGenome Biol 8
- Identifying biological themes within lists of genes with EASEGenome Biol 4
- A uniform proteomics MS/MS analysis platform utilizing open XML file formatsMol Syst Biol 1
- Empirical statistical model to estimate the accuracy of peptide identifications made by MS/MS and database searchAnal Chem 74:5383–5392
- A statistical model for identifying proteins by tandem mass spectrometryAnal Chem 75:4646–4658
- T-Coffee: A novel method for fast and accurate multiple sequence alignmentJ Mol Biol 302:205–217
- g:Profiler: a web server for functional enrichment analysis and conversions of gene lists (2019 update)Nucleic Acids Res 47:W191–W198
- limma powers differential expression analyses for RNA-sequencing and microarray studiesNucleic Acids Res 43
- Differential oestrogen receptor binding is associated with clinical outcome in breast cancerNature 481:389–393
- RAxML version 8: a tool for phylogenetic analysis and post-analysis of large phylogeniesBioinformatics 30:1312–1313
- UniProt: the universal protein knowledgebaseNucleic Acids Res 45:D158–D169
- PANTHER: A library of protein families and subfamilies indexed by functionGenome Res 13:2129–2141
- UniProt: the universal protein knowledgebaseNucleic Acids Res 46
Article and author information
Version history
- Sent for peer review:
- Preprint posted:
- Reviewed Preprint version 1:
- Reviewed Preprint version 2:
- Reviewed Preprint version 3:
- Version of Record published:
Copyright
© 2023, Lee et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Metrics
- views
- 925
- downloads
- 78
- citations
- 0
Views, downloads and citations are aggregated across all versions of this paper published by eLife.