Concurrent depletion of Vps37 proteins evokes ESCRT-I destabilization and profound cellular stress responses

Genetic alterations induce reprogramming of intracellular signaling, which is a driving force of tumorigenesis. The duration of signal transduction is dependent on endocytosis (Floyd and De Camilli, 1998; Mosesson et al., 2008; Schmid, 2017). Some mechanisms for tumor cell-specific changes in the activity of endocytic components that affect intracellular signaling have already been identified (Barbieri et al., 2016; Di Fiore and von Zastrow, 2014; Mellman and Yarden, 2013). For instance, many tumors exhibit deregulated expression of the ubiquitylation machinery and small GTPases that control the rate of receptor degradation and recycling, respectively (Porther and Barbieri, 2015). However, despite the abundance of publicly available data, such as those deposited in The Cancer Genome Atlas (TCGA; Weinstein et al., 2013), little has been done to systematically analyze the expression of receptor trafficking regulators in tumors and across tumor stages. This knowledge could facilitate patients' stratification for treatment with bioengineered macromolecules delivered through receptor-mediated endocytosis (Tashima, 2018).

An important group of trafficking regulators constitute the four sequentially acting endosomal sorting complexes required for transport (ESCRT-0, ESCRT-I, ESCRT-II and ESCRT-III) and their accessory proteins, including Vps4A and Vps4B. The ESCRT machinery mediates receptor degradation not only by recognition and local clustering of ubiquitylated cargo on endosomes but also through membrane deformation and scission to form intraluminal vesicles (ILVs). Many rounds of ILV formation create multivesicular bodies that fuse with lysosomes leading to cargo degradation. In addition, ESCRTs contribute to cytokinesis, autophagy, virus budding, exovesicle release and repair of plasma and intracellular membranes (Hurley, 2015; Olmos and Carlton, 2016; Szymanska et al., 2018; Vietri et al., 2020). Despite the well-established roles of ESCRT components in maintaining cell homeostasis, much less is known about their contribution to tumorigenesis (Alfred and Vaccari, 2016; Gingras et al., 2017; Mattissek and Teis, 2014), and the underlying molecular mechanism has been clarified only in a couple of cases (Manteghi et al., 2016; Sadler et al., 2018). For instance, we previously demonstrated that the expression of VPS4B, encoding an ESCRT-associated ATPase, is decreased in colorectal cancer (CRC), and VPS4B-deficient cells are critically dependent on the Vps4A protein. This synthetic lethality between VPS4 paralogs triggers stress-associated sterile inflammatory response and immunogenic cell death and thus may be used as a basis for personalized therapy (Szymańska et al., 2020).

ESCRT-I is composed of three core components (Tsg101, Vps28 and one of four Vps37 family members) and a single auxiliary protein (UBAP-1, Mvb12A or Mvb12B) (Stefani et al., 2011; Wunderley et al., 2014). At least under some conditions, Tsg101 and Vps37A may act as tumor suppressors (Li and Cohen, 1996; Moberg et al., 2005; Xu et al., 2003). In parallel, high-throughput screens for cancer vulnerability within the DepMap project (Behan et al., 2019) have demonstrated that multiple cancer cell lines display reduced fitness upon TSG101 and VPS37A knockout, whilst the effects of perturbed expression of VPS28, VPS37C or UBAP1 are cell type-dependent.

Tsg101 and Vps37A are not only regulators of vesicular trafficking (Bache et al., 2004; Bishop et al., 2002) but also other processes, such as transcription and autophagy (Lin et al., 2013; Takahashi et al., 2019). Transcriptomic analysis of Tsg101-depleted cancer cells revealed increased expression of the prototypical nuclear factor-κB (NF-κB)-dependent genes without exogenous stimulation (Brankatschk et al., 2012). We dissected the molecular basis of this phenomenon, showing that the absence of Tsg101 or Vps28 leads to the accumulation of ligand-free cytokine receptors on endosomes because of disturbed sorting into ILVs and degradation of cargo. The proximity of accumulated receptors on endosomes evoked their oligomerization to trigger NF-κB signaling (Banach-Orlowska et al., 2018; Mamińska et al., 2016). However, none of the Vps37 proteins incorporated into ESCRT-I was identified as a genuine regulator of the NF-κB pathway.

NF-κB is a family of ubiquitously expressed transcription factors, whose activation is a hallmark of inflammation often associated with cancer (Taniguchi and Karin, 2018; Zhang et al., 2017). There are two interconnected NF-κB signaling cascades. Whereas the canonical NF-κB pathway culminates in nuclear entry of NF-κB dimers consisting of p65 (encoded by RELA) and p50 (encoded by NFKB1), the non-canonical signaling is marked by proteasomal processing of the p100 NF-κB precursor (encoded by NFKB2) to p52 to form transcriptionally active complexes with RelB (Hayden and Ghosh, 2008). Although endocytic trafficking and NF-κB inflammatory signaling are important in carcinogenesis, the molecular links between them are poorly studied.

Here, we systematically analyzed the expression of endocytosis regulators across stages of CRC using publicly available RNA-Seq data and found decreased expression of VPS37 paralogs. Because no genome-wide expression studies have explored the cellular consequences of individual and concurrent depletion of Vps37 proteins, we investigated their roles focusing on the processes related to cell growth and inflammatory response. Our findings reveal the importance of VPS37 paralogs in orchestrating cell homeostasis through maintaining the stability of ESCRT-I.

CRC is a leading cause of cancer-associated deaths worldwide because it is often diagnosed in advanced stages when patients display clinical symptoms (Siegel et al., 2018). Aberrant endosomal trafficking in CRC has been linked to adverse phenotype and resistance to therapies (Gargalionis et al., 2015). To gain insights into transcriptional changes of genes involved in endosomal trafficking in CRC, we performed a meta-analysis of RNA-Seq datasets, including those deposited in TCGA, against a custom-made list (Table S1) of 445 components acting in endocytic transport (Fig. 1A). Of these, 112 genes were differentially expressed between cancer and healthy colon tissue. We found that 100 and 69 genes were differentially expressed in early and advanced stages of CRC, respectively (Fig. 1B). Differential expression of 57 genes was common across stages of tumorigenesis, whereas 43 and 12 genes were unique for early and advanced stages, respectively (Table S2). In addition to decreased mRNA abundance of VPS4B, which we studied previously (Szymańska et al., 2020), we detected reduced expression of an ESCRT-I component – VPS37B – in advanced stage CRC (Fig. 1C).

Because two out of three core ESCRT-I components, namely Tsg101 and Vps28, restrict NF-κB-dependent transcription (Brankatschk et al., 2012; Mamińska et al., 2016), we analyzed expression of genes encoding the core ESCRT-I subunits in healthy colon tissue samples from the RNA-Seq datasets. Using the transcripts per million (TPM) metrics to normalize expression data with respect to gene length and sequencing depth, we observed that colorectal tissue and its cancer counterpart expressed high levels of TSG101, VPS28 and VPS37B; moderate levels of VPS37A and VPS37C; and negligible levels of VPS37D. Differential expression analysis of CRC samples compared against matching healthy tissue controls revealed that transcription of VPS28, VPS37A and VPS37B tended to be decreased in early stages of CRC. In advanced stages, only expression of VPS37B was downregulated (Fig. 1D; Table S3). Expression of TSG101, VPS37A and VPS37C was stable across tumor stages (Fig. 1E; Table S3).

In summary, VPS37B expression is decreased during progression from early to advanced stages of CRC. Conversely, transcription of VPS28 and VPS37A exhibits a tendency towards downregulation only in early stages of CRC.

Because VPS37B expression was decreased in CRC patients investigated in our meta-analysis and VPS37A mRNA and protein abundance was previously shown to be reduced in CRC patients (Chen et al., 2018; Miller et al., 2018; Vasaikar et al., 2019), we performed reverse transcription qPCR (RT-qPCR) analysis of VPS37 mRNA levels using an independent set of CRC samples from our previous study (Mikula et al., 2011; Skrzypczak et al., 2010). We observed a significant decrease in VPS37B and VPS37A mRNA abundance in adenocarcinoma (Fig. 2A).

To assess whether transcriptional alterations at mRNA level correlate with the diminished abundance of Vps37 proteins in CRC, we performed immunohistochemistry (IHC) staining of Vps37A and Vps37B in an array setup consisting of 100 pairs of treatment-naïve primary CRC samples and non-cancerous colon tissue using specific antibodies (Fig. S1A–G). We evaluated the tissue arrays using a semi-quantitative scoring method based on staining intensity.

Both Vps37A and Vps37B displayed strong cytoplasmic staining in normal colon tissue (Fig. 2B). Out of the 100 investigated samples, protein staining of Vps37A was decreased to the medium intensity level (from a score of 3+ to a score of 2+) in cancerous tissue of all examined patients. Vps37B staining was decreased to the medium intensity level (from 3+ to 2+) in 70% of patients and to weak intensity levels (from 3+ to 1+) in 30% of patients (Fig. 2C). Because the analyzed group of treatment-naïve CRC patients was very homogenous with respect to pathological tumor status, pathological nodes and disease grade (Table S4), we could not correlate Vps37 staining intensity with clinical disease staging.

Overall, these results corroborate the finding of our bioinformatics analysis that VPS37B is downregulated at mRNA and protein levels in CRC. They also suggest reduced mRNA and protein levels of VPS37A in CRC.

Humans have four VPS37 genes whose protein products display distinct domain architecture suggesting partly different cellular functions. They all possess the Mod(r) domain, which mediates interaction with the other ESCRT-I subunits. Whereas only Vps37A has the ubiquitin-binding UEV domain, the other members contain the proline-rich region (PRR) essential for protein–protein interactions (Fig. 3A). To study cellular functions of Vps37 paralogs, we used the DLD1 cell line as an in vitro model of human CRC, wherein expression of VPS37 paralogs reflects that observed in TCGA CRC cohorts.

To understand the molecular consequences of individual and concurrent depletion of Vps37 proteins on cellular homeostasis, we performed RNA-Seq in DLD1 cells. We verified high knockdown efficiency and selectivity of siRNA (two independent sequences per target) by measuring the protein abundance of the three paralogs (Fig. S2A–C). For RNA-Seq, we used one siRNA (designated #1) per target, alone or in double and triple combinations, resulting in nine experimental conditions. Genes were considered differentially expressed when their expression was either ≤0.667-fold or ≥1.5-fold, and with P<0.05 (Wald test), when normalized against the two control conditions – siCTRL#1 (a combination of non-targeting siRNAs) and non-transfected cells (NT) (Fig. S2D). All subsequent validation experiments were performed with two independent siRNA sequences per target.

Co-silencing of VPS37A, VPS37B and VPS37C (referred to hereafter as VPS37ABC) elicited the greatest number of transcriptional changes (1277 genes). Pronounced changes (781 genes) were also detected after co-silencing of VPS37A and VPS37B (VPS37AB), indicating the importance of these two subunits in cell physiology. Conversely, a limited number of genes underwent transcriptional changes upon other silencing combinations of VPS37 paralogs (Fig. S2D). Hierarchical clustering of all investigated conditions on a set of differentially expressed genes after individual, double and triple silencing demonstrated that the branch containing the VPS37ABC and VPS37AB conditions (siVPS37ABC#1 and siVPS37AB#1, respectively) was distinct from the remaining conditions (Fig. S2E). Also, the pools of genes induced upon single depletion of Vps37A, Vps37B and Vps37C were largely non-overlapping (Fig. S2F). Thus, multifaceted transcriptional responses to co-depletion of Vps37 proteins could stem from the accumulation of paralog-specific cellular defects, but highly similar Vps37B and Vps37C proteins could also potentially compensate for each other's absence.

Differentially expressed genes under each silencing condition were subjected to Gene Ontology (GO) enrichment analysis for biological processes. We identified enrichment of biological processes only in siVPS37A#1, siVPS37AB#1, VPS37AC#1 and VPS37ABC#1 conditions. Among the top 15 gene signatures (ordered based on a number of genes in the cluster upon VPS37ABC silencing) were processes related to cell migration, cellular signaling, inflammatory response, cell growth, and adhesion (Fig. 3B). We further focused on the ‘inflammatory response’ (GO:0006954) and ‘regulation of growth’ (GO:0040008) clusters. The inflammatory response heatmap contained genes encoding cytokines (CXCL8), adhesion molecules (ICAM1), and negative regulators of NF-κB signaling (TNFAIP3). In-depth interrogation of the ‘regulation of growth’ cluster showed the presence of cyclin-dependent kinase inhibitors (CDKN1A, CDKN2D) (Fig. 3C,D). To determine the signaling pathways associated with the genes in the inflammatory response and regulation of cell growth clusters, we conducted a pathway network analysis using the Reactome database. It yielded enrichment of annotations related to signaling initiated by cytokines, receptor tyrosine kinases and G-protein-coupled receptors, and involving mitogen-activated protein kinases (MAPK) and phosphoinositide 3-kinase (PI3K)–Akt (Fig. 3E).

Collectively, these data show at least partly non-redundant cellular functions of VPS37 paralogs. The type and magnitude of transcriptional responses after their co-silencing are the cumulative response to perturbations of individual functions executed by Vps37 proteins. Co-silencing of VPS37AB profoundly affects gene expression patterns linked to ‘inflammatory response’ and ‘regulation of cell growth’, and additional silencing of VPS37C on top of VPS37AB knockdown further potentiates perturbations in gene transcription. These data suggest that Vps37 depletion activates multifaceted stress responses in cells.

To validate our RNA-Seq analysis, we selected the most pronouncedly induced genes from the inflammatory response cluster that represented different classes of molecules (Fig. 3C). We performed RT-qPCR analysis after (co-)silencing of VPS37 paralogs in DLD1 and HEK293 cells. We previously discovered the induction of inflammatory responses upon TSG101 and VPS28 silencing in HEK293 cells (Mamińska et al., 2016). Knockdown of VPS37A stimulated transcription of CXCL8 and TNFAIP3 in HEK293 cells (Fig. S3E,H). Co-depletion of Vps37AB promoted transcription of TNFAIP3 in DLD1 cells, as well as promoting CXCL8, NFKBIA and TNFAIP3 expression in HEK293 cells. HEK293 cells with VPS37AC co-silencing had enhanced transcription of CXCL8. Knockdown of VPS37ABC increased CXCL8, NFKBIA and TNFAIP3 transcription in HEK293 cells and TNFAIP3 expression in DLD1 cells. Silencing of VPS37B, VPS37C or VPS37BC did not affect any of the investigated genes (Fig. S3A–H).

Because the magnitude of transcriptional changes in DLD1 and HEK293 cells was modest, we tested expression of the same genes in another CRC cell line, RKO. We found that VPS37A knockdown had a modest, yet insignificant, effect on CXCL8 expression and no effect on ICAM1, TNFAIP3 or NFKBIA transcription (Fig. 4A–D). VPS37AB co-silencing increased CXCL8, ICAM1, TNFAIP3 and NFKBIA expression; however, the increase did not reach statistical significance for CXCL8 (Fig. 4A–D). We observed that Vps37ABC co-depletion had further positive effects on the magnitude of CXCL8, ICAM1, NFKBIA and TNFAIP3 transcription compared with the effects in Vps37AB-depleted cells (Fig. 4A–D). VPS37B, VPS37C, VPS37AC and VPS37BC knockdown did not substantially affect transcription of the investigated targets (Fig. 4A–D).

To understand the differences between effects of the same on-target siRNA combinations in DLD1 and RKO cells, we analyzed the efficiency and duration of silencing by immunoblotting (Fig. S3I,J). By exposing blots close to signal saturation, we found some differences in the efficiency and duration of Vps37 protein depletion upon transfection with on-target siRNA combinations between RKO and DLD1 cells, which may account for the observed discrepancies in some of our assays. Specifically, transfection with siVPS37ABC#1 more efficiently downregulated Vps37A, Vps37B and Vps37C proteins at day 4 in DLD1 and RKO cells than transfection with siVPS37ABC#2. Because the effects of depletion of Vps37 proteins on gene expression were more pronounced in RKO cells (Fig. S3J), we subsequently used these cells as our main model.

In summary, co-depletion of Vps37 proteins, in particular Vps37AB and Vps37ABC, strongly activates transcription of multiple classes of inflammatory genes. However, some effects of siRNA treatment and the rate of gene transcription are cell type-dependent.

Because depletion of the ESCRT-I subunits Tsg101 and Vps28 induces NF-κB-driven inflammatory response (Brankatschk et al., 2012; Mamińska et al., 2016), and because our RNA-Seq analysis yielded the ‘inflammatory response’ gene cluster as being differentially expressed in siVPS37AB#1 and siVPS37ABC#1 conditions (Fig. 3C), we combined genes from each cluster into a single list and subjected them to in silico analysis of transcription factor-binding sites (TFBS) using RcisTarget (Aibar et al., 2017). This revealed enrichment of TFBS for NF-κB, FOS and AP1 transcription factors. The consensus sequence for p65 (RelA) binding had the highest normalized enrichment score and was the top annotated TFBS (Table 1).

We next explored the molecular basis of enhanced expression of NF-κB-dependent genes upon (co-)silencing of VPS37 paralogs (Fig. 4A–D; Fig. S3A–H). We measured p65 phosphorylation and p100-to-p52 processing as hallmarks of canonical and non-canonical NF-κB signaling, respectively. Immunoblotting analysis of lysates from RKO cells with silencing of individual VPS37 paralogs as well as VPS37AC and VPS37BC showed no significant induction of the NF-κB pathway (Fig. 4E–G). VPS37AB and VPS37ABC knockdowns enhanced p100 production and its cleavage to p52, whereas only VPS37ABC knockdown induced p65 phosphorylation (Fig. 4E–G).

MAPKs cooperate with NF-κB in driving inflammation (Hoesel and Schmid, 2013). Because our transcriptomic data revealed increased expression of genes whose products regulate the MAPK cascade, we tested phosphorylation of JNK (JNK1 and JNK2, also known as MAPK8 and MAPK9, respectively), p38 MAPK (p38-α, -β, -γ, and -δ, also known as MAPK14, MAPK11, MAPK12 and MAPK13, respectively) and ERK (ERK1 and ERK2, also known as MAPK3 and MAPK1, respectively). Silencing of individual VPS37 paralogs, VPS37AC and VPS37BC did not affect JNK, p38 and ERK phosphorylation (Fig. 4H–J). VPS37AB knockdown activated JNK but not p38 or ERK (Fig. 4H–J). Vps37ABC co-depletion induced JNK and p38 phosphorylation but did not increase basal activity of ERK (Fig. 4H–J).

Overall, we conclude that Vps37ABC co-depletion activates canonical and non-canonical NF-κB signaling as well as JNK and p38 MAPKs.

Because GO analysis of our RNA-Seq data revealed the regulation of growth gene cluster (Fig. 3D), we examined the effects of depletion of Vps37 proteins on cell growth using a short-term BrdU proliferation assay and a long-term colony formation assay in DLD1 and RKO cells.

Using the proliferation assay, we observed that VPS37A silencing modestly decreased the proliferation rate of DLD1 and RKO cells (Fig. 5A,B). Knockdown of VPS37B, VPS37C, VPS37AC or VPS37BC did not alter growth of DLD1 and RKO cells. VPS37AB silencing significantly inhibited DLD1 and RKO cell proliferation, whereas the strongest inhibition was seen upon VPS37ABC knockdown (Fig. 5A,B). In turn, depletion of Vps37A alone or co-silencing of either VPS37AB or VPS37ABC inhibited the ability of DLD1 cells to form colonies in the clonogenic assay performed 14 days after siRNA transfection (Fig. 5C; Fig. S4A). Analysis of clonogenic growth of RKO cells revealed a trend towards downregulation upon depletion of Vps37A or Vps37AB but this did not reach the level of significance for all comparisons. Nevertheless, knockdown of VPS37ABC significantly inhibited the ability of RKO cells to form colonies (Fig. 5D; Fig. S4B). In parallel, we checked the impact of TSG101 silencing on cell proliferation and colony formation. TSG101 knockdown in RKO cells inhibited both processes, comparably to the inhibition following co-silencing of VPS37 paralogs (Fig. S4C–E).

In conclusion, concurrent depletion of Vps37 proteins has detrimental effects on cancer cell growth in vitro, and the phenotype of Tsg101-depleted cells closely resembles the one observed in VPS37ABC-knockdown cells. Our results also indicate that growth rate is primarily dependent on the expression of VPS37A and VPS37B. Additional co-silencing of VPS37C potentiated proliferation and colony-forming defects of VPS37AB knockdown. In contrast, long-term growth rate appears to be largely dependent on Vps37A, at least in DLD1 cells.

Accelerated division of tumor cells is partly a result of abnormal activity of cyclins and cyclin-dependent kinase inhibitors (CDKNs) (Bonelli et al., 2014). Because our RNA-Seq data revealed increased expression of three CDKN genes (CDKN1A, CDKN2B and CDKN2D; Fig. 3D and GSE152195), we used RT-qPCR to profile their transcription in RKO and DLD1 cells upon silencing of VPS37 paralogs. In RKO cells, knockdown of VPS37A did not affect CDKN1A or CDKN2B transcription (Fig. 5E,F) but modestly induced CDKN2D expression (Fig. 5G). Silencing of either VPS37B, VPS37C, VPS37AC or VPS37BC did not change transcription of any of the analyzed genes (Fig. 5E–G). VPS37AB silencing increased CDKN1A and CDKN2D transcription (Fig. 5E,G). Concurrent depletion of Vps37ABC increased transcription of CDKN1A and CDKN2D, and had a positive effect on CDKN2B expression, but in the latter case without reaching statistical significance for all comparisons (Fig. 5E–G). In line with these results, we found a similar pattern of CDKN expression in differentially transfected DLD1 cells; however, we could not corroborate the enhanced CDKN2D transcription upon co-depletion of Vps37 proteins that we initially identified in our RNA-Seq analysis (Fig. S4F–H). Finally, the transcription pattern of CDKNs after depletion of Tsg101 in RKO cells paralleled those observed for CDKN1A and CDKN2B expression after VPS37ABC co-silencing. Silencing of TSG101 did not induce CDKN2D expression (Fig. S4I).

Increased expression of CDKNs after concurrent Vps37ABC depletion suggests an impact on the proliferation rate and cell cycle progression. CDKN1A encodes p21, which inhibits cell cycle progression in the G1, S and G2 phases, whereas CDKN2B and CDKN2D encode p15^INK4B and p19^INK4D, respectively, which inhibit complexes formed by cyclin D and halt cell cycle in the G1 phase (Bonelli et al., 2014). Thus, we evaluated proliferation of Vps37ABC-depleted RKO cells upon co-silencing of CDKN1A, which was the most pronouncedly induced gene in our RT-qPCR analysis. We observed that concurrent knockdown of VPS37ABC and CDKN1A partly rescued cell proliferation, corroborating the inhibitory impact of p15^INK4B and p19^INK4D on cell division (Fig. 5H). In accordance with this, p21 depletion in cells with TSG101 knockdown improved RKO cell proliferation (Fig. S4J).

To gain further insights into the inhibition of cell growth after differential silencing of VPS37 paralogs, siRNA-transfected RKO cells were stained with propidium iodide and the cell cycle was analyzed by flow cytometry. We observed that knockdown of either VPS37B, VPS37C or VPS37BC did not change cell cycle progression, as indicated by the unaltered percentage of cells in the G0/G1 and S phases (Fig. 5I,J). Knockdown of VPS37A displayed a trend towards increasing the percentage of cells in the G0/G1 phase with concurrent decrease of the number of cells in the S phase (Fig. 5I,J). The impact of Vps37AC depletion closely paralleled that observed after VPS37A silencing but was not statistically significant in all comparisons. VPS37AB knockdown resulted in an increased number of cells in the G0/G1 phase and a drop in the number of cells in S phase (Fig. 5I,J). The proportion of cells in the G0/G1 and S phases after co-silencing of VPS37ABC was comparable to that observed in Vps37AB-depleted cells (Fig. 5I,J). Finally, silencing of TSG101 closely paralleled the effects observed after VPS37ABC knockdown (Fig. S4K–M). None of the analyzed silencing conditions (involving VPS37 paralogs and TSG101) altered the percentage of cells in the G2/M phase (Fig. 5K; Fig. S4M).

In summary, our data revealed that concurrent depletion of Vps37 proteins induces the expression of three CDKNs, which cooperatively halt the cell cycle in the G1 phase. The phenotype of Tsg101-depleted cells closely resembles that observed in VPS37ABC-knockdown cells.

We next investigated the molecular basis for NF-κB induction and p21-mediated growth arrest after Vps37ABC co-depletion in RKO cells. Because CDKN1A, which encodes p21, was the most potently affected gene in our RT-qPCR analysis (Fig. 5E) and its knockdown in Vps37ABC-depleted RKO cells partly rescued their proliferation (Fig. 5H), we used it as readout to assess the relationship between the inflammatory response and cell growth arrest.

We first investigated the timecourse of changes in p21 levels and activation of the NF-κB pathway after VPS37ABC silencing in RKO cells. We observed that depletion of Vps37ABC increased p21 abundance after 24 h and 72 h post-transfection (Fig. 6A). We found rapid phosphorylation of p65 in Vps37ABC-depleted cells, 24 h and 72 h post-transfection; however, the increase in p65 phosphorylation at 24 h was not statistically significant (Fig. 6B). The abundance of p100 and p52 increased from 24 h to 72 h post-transfection in cells with VPS37ABC knockdown (Fig. 6C,D). Throughout 24–72 h post-transfection, abundance of Vps37A, Vps37B and Vps37C gradually decreased (Fig. S5A–C). These data showed that the activation of NF-κB signaling and production of p21 occurred within a similar timeframe after Vps37ABC depletion, thus none of these processes preceded the others.

In certain cell types, the canonical NF-κB pathway is crucial for CDKN1A transcription (Ledoux and Perkins, 2014). We, thus, verified whether increased CDKN1A expression after co-depletion of Vps37 proteins required the canonical NF-κB subunit p65 (encoded by RELA). Knockdown of RELA in Vps37ABC-depleted cells did not affect CDKN1A expression (Fig. 6E), although it effectively blunted transcription of CXCL8 and ICAM1, two prototypical NF-κB target genes (Fig. S5D–I). These results suggest that negative effects on cell growth stemming from co-silencing of VPS37 paralogs are not consequences of induction of canonical NF-κB signaling.

p21 modulates NF-κB signaling in immune cells (Rackov et al., 2016; Trakala et al., 2009), but whether similar mechanisms occur in CRC cells is unknown. Thus, we explored whether silencing of CDKN1A affected phosphorylation of p65 and processing of p100 to p52 upon Vps37ABC depletion. As assessed by immunoblotting, co-silencing of CDKN1A in Vps37ABC-depleted cells did not affect the levels of p65 phosphorylation and p100-to-p52 processing compared with the effects of co-silencing of VPS37 paralogs alone (Fig. 6F–H; Fig. S5J–M). We also established that co-silencing of CDKN1A inhibited transcription of CXCL8 but not ICAM1, TNFAIP3 and NFKBIA in Vps37ABC-depleted cells (Fig. 6I–L). These results showed no modulatory impact of p21 on NF-κB signaling and three out of four investigated target genes.

Overall, we conclude that the induction of NF-κB inflammatory response and p21-mediated cell growth inhibition are two independent processes in CRC cells following co-depletion of Vps37 proteins. Thus, cell growth arrest is not caused by activation of an inflammatory response.

We speculated that the type and magnitude of transcriptional responses after individual and concurrent silencing of VPS37 paralogs might be attributed to distinct ESCRT-I stability. It has previously been shown that knockdown of some ESCRT-I core components induces partial or complete degradation of other complex subunits (Stefani et al., 2011; Wunderley et al., 2014); yet, a detailed characterization of all ESCRT-I subunits after individual and concurrent Vps37 protein depletion has not been performed so far.

First, we checked whether knockdown of individual ESCRT-I components affected the stability of its remaining subunits expressed in CRC cells. Immunoblotting analysis of lysates from RKO cells revealed that depleting one of either Tsg101 or Vps28 destabilized the other (Fig. S6A,B), as well as causing depletion of  Vps37A, Vps37B, Vps37C, Mvb12A and Mvb12B, and lowering UBAP-1 protein abundance (Fig. S6C–H). We observed that silencing of VPS37A diminished UBAP-1 protein abundance, indicating that ESCRT-I complexes containing Vps37A preferentially incorporate UBAP-1 (Fig. S6F). Depletion of Vps37B reduced the abundance of Tsg101 (Fig. S6A) and partially reduced that of Mvb12A (Fig. S6G). Conversely, silencing of MVB12A decreased Vps37B levels (Fig. S6D), indicating partnering preference between these subunits. We did not observe any relationship between the stability of Vps37C and Mvb12 proteins (Fig. S6E,G–H).

We next analyzed the stability of ESCRT-I core and auxiliary subunits upon VPS37 co-silencing. Using different sets of siRNAs, we achieved selective and efficient knockdown of Vps37 proteins alone or in combination (Fig. 7A-C). Vps37AB or Vps37BC depletion decreased Tsg101 and Vps28 abundance, whereas the effects of VPS37AC knockdown were less potent (Fig. 7D,E). Knockdown of all three VPS37 genes led to the complete destabilization of Tsg101 and Vps28 proteins (Fig. 7D,E). Protein abundance of UBAP-1 was decreased in all silencing combinations involving VPS37A (Fig. 7F) corroborating the results of individual VPS37A knockdown (Fig. S6F). Similarly, Mvb12A abundance was reduced whenever cells were depleted of Vps37B (Fig. 7G), whereas such effect was less pronounced for Mvb12B (Fig. 7H). Silencing of all VPS37 genes depleted both Mvb12 proteins (Fig. 7G,H). Notably, the stability of the core and auxiliary ESCRT-I subunits after concurrent VPS37ABC knockdown closely resembled the effects of TSG101 silencing (Fig. S6A).

In summary, these data show an inter-dependability of ESCRT-I subunits for maintaining the complex stability (Table S5). They indicate that the incorporation of auxiliary subunits is selective with respect to their Vps37 partners (Vps37A with UBAP-1 and Vps37B with Mvb12A). Simultaneous interference with VPS37A and VPS37B expression induces pronounced decrease in ESCRT-I stability, and Vps37C depletion only slightly magnifies this effect. Our results further argue that the type and magnitude of transcriptional responses after differential depletion of Vps37 proteins correlate with the abundance of core and accessory ESCRT-I components.

Tumors develop various mechanisms to prolong exposure of plasma membrane receptors to ensure constitutive signaling that is beneficial for their growth. One such mechanism relies on altered expression of endocytic transport regulators (Barbieri et al., 2016; Floyd and De Camilli, 1998; Mellman and Yarden, 2013; Mosesson et al., 2008; Schmid, 2017). The advent of next-generation sequencing technologies has permitted unbiased screening of components orchestrating receptor transport and endolysosomal degradation in distinct pathological settings (Buser et al., 2019; Yoshida et al., 2010). Here, by mining RNA-Seq datasets we revealed differential expression of 112 endocytic machinery components across stages of CRC, several of which were previously shown to be reduced in adenocarcinoma (Kwong et al., 2005; Szymańska et al., 2020; Tanigawa et al., 2019). In-depth validation of our in silico screen uncovered reduced mRNA and protein abundance of VPS37A and VPS37B paralogs in CRC patients. Our in vitro data indicate that Vps37 proteins have partly non-overlapping functions in the cell. We also showed that acute Vps37 co-depletion evokes stress responses manifested, among others, by activation of the NF-κB inflammatory response and p21-mediated impairment of cell growth. We correlated the magnitude of stress responses with the degree of ESCRT-I subunit destabilization after (co-)depletion of Vps37 proteins. Although decreased abundance of VPS37A and/or VPS37B mRNA and proteins appears not to be an oncogenic driver per se, these passenger alterations might represent potential vulnerabilities of cancer cells to therapeutic treatment.

Human VPS37 paralogs have distinct chromosomal localization and protein sequence identity. VPS37A, VPS37B, VPS37C and VPS37D are localized on chromosome 8p, 12q, 11q and 7q, respectively. This different chromosomal localization of VPS37 genes could favor independent regulatory mechanisms of expression in pathophysiological circumstances. Another layer of complexity is added by the incorporation of Vps37 proteins into ESCRT-I to form functionally distinct complexes (Stefani et al., 2011; Wunderley et al., 2014). Changes in mRNA and protein levels of VPS37A were previously documented in various cancer types, including liver, prostate, breast, ovarian, renal, lung, glioma, gastric, oral and colon cancer (Chen et al., 2015, 2020, 2018; Du et al., 2016; Fu et al., 2018; Lai et al., 2009; Perisanidis et al., 2013; Sun et al., 2017; Vasaikar et al., 2019; Wittinger et al., 2011; Wu et al., 2019; Xu et al., 2017a,b, 2014, 2017c, 2003; Yang et al., 2016, 2017; Zhu et al., 2015). Several of these studies suggested that VPS37A acts as a tumor suppressor and that its loss can serve as an adverse prognostic factor. However, the abundance of the remaining VPS37 paralogs at mRNA and protein levels or their contribution to oncogenesis have not been investigated across cancer types and disease stages.

Our meta-analysis of RNA-Seq datasets of CRC patients revealed decreased VPS37B transcription during the transition from early to advanced stages of colorectal adenocarcinoma and a trend towards downregulation of VPS37A in patients of early disease stages. Analysis of an independent cohort of patients corroborated results of our meta-analysis that VPS37B expression is indeed reduced in adenocarcinoma patients. Additionally, this cohort showed small, but significant, decreases in VPS37A expression. We further confirmed the lower abundance of both proteins in treatment-naïve primary CRC samples. The observations we made of VPS37 paralog expression extend previous bioinformatic analysis of colon adenocarcinoma (COAD) and rectum adenocarcinoma (READ) cohorts from TCGA performed without grouping these patients based on disease stages (Miller et al., 2018) and a proteogenomic study of a homogenous cohort of treatment-naïve patients undergoing primary surgery for colon adenocarcinoma (Vasaikar et al., 2019). The VPS37A gene undergoes frequent deletion as a part of chromosome 8p during progression from adenoma to adenocarcinoma, which would explain its decreased abundance at mRNA and protein levels (Meijer et al., 1998). Further studies will need to clarify the molecular origin of diminished VPS37B expression in CRC, because chromosome 12q, where it is located, undergoes frequent amplifications, suggesting an existence of a (post-)transcriptional mechanism (Wood et al., 2007).

Here, we systematically studied the consequences of Vps37 protein (co-)depletion in CRC cells. Analysis of the transcriptome of CRC cells revealed that Vps37 proteins have partly non-overlapping function, as deduced from distinct sets of genes induced upon their individual depletion. It also suggests that the type and magnitude of transcriptional responses upon concurrent VPS37 paralog silencing stem from the cumulative inhibition of cellular processes executed by their protein products. Among genes induced upon either VPS37AB or VPS37ABC knockdown, we did not find prototypical drivers of tumorigenesis but rather cell growth inhibitors, such as CDKN1A/p21, CDKN2B/p14^INK4B and CDKN2D/p19^INK4D. As a consequence, we observed decreased ability of CRC cells to progress through the cell cycle that resulted in the inhibition of proliferation and colony-forming ability. Our results from Vps37ABC-depleted cells reinforce the notion that knockdown of other ESCRT-I components, beyond TSG101, VPS28 and UBAP1, induces cell cycle arrest and halts cell proliferation (Krempler et al., 2002; Miller et al., 2018; Morita et al., 2007). We also demonstrate that the degree of cell growth impairment depends primarily on the perturbed expression of VPS37A and whether it is silenced on its own or in combination with other paralogs. To our knowledge, expression of neither VPS37B nor VPS37C has been related to cancer growth but rather has been linked to virus release and infectivity (Stuchell et al., 2004). A vast majority of CRC cell lines tested within the DepMap project (Behan et al., 2019) showed no or slight changes in cell fitness upon RNAi-mediated depletion of VPS37A or VPS37B (VPS37C was not tested with this technology). Of note, CRISPR-Cas9-mediated knockout of VPS37A markedly decreased RKO and DLD1 cell fitness, whereas the effects of VPS37B and VPS37C knockout were less deleterious. Observations made within the DepMap project support the results of our colony formation and proliferation assays and indicate distinct effects of VPS37 paralog loss on long- and short-term growth. Although our data suggest a negative impact of Vps37 paralog (co-)depletion on cancer cell growth, we postulate that the effect of their CRC-associated reduction is more nuanced. Decreased expression of VPS37A may be beneficial for cells of advanced stage CRC, because its RNAi-mediated silencing promotes resistance of prostate and breast cancer cells to chemotherapeutics (Sun et al., 2017; Yang et al., 2016). Furthermore, loss of chromosome 8p bearing VPS37A promotes tumor growth (Cai et al., 2016; Xue et al., 2012), whereas the contribution of Vps37B to cancer growth remains elusive. Our RNA-Seq meta-analysis did not reveal increased transcription of CDKN1A, CDKN2B and CDKN2D in patients with advanced stage CRC. These patients displayed only 38% and 16% reduction in expression of VPS37B and VPS37A, respectively, which at least in part might explain why no induction of CDKN expression was observed. Analysis of gene expression patterns in patients with early stage CRC revealed 24% and 22% decrease of VPS37A and VPS37B transcription, respectively (Tables S2,S6). This suggests that pronounced co-depletion of Vps37 proteins may be detrimental to CRC cell growth. Thus, the discrepancy in expression of CDKNs between patients and our in vitro model of CRC is probably related to differences in the magnitude of VPS37 paralog downregulation. Yet another important difference is that our in vitro results describe the impact of acute and transient depletion of Vps37 proteins on CRC growth. Future studies using patient-derived xenografts might clarify the impact of moderately decreased expression of VPS37A and VPS37B on CRC growth and progression.

The pronounced inhibition of cell growth in vitro upon Vps37AB and Vps37ABC depletion correlated with activation of inflammatory and stress signaling mediated by NF-κB and MAPK. These cellular responses induced upon VPS37 co-silencing can be viewed as another example of sterile inflammation, which resembles stress reactions caused by intracellular dysfunction of numerous membrane organelles, such as malfunctioning mitochondria, ER or endosomes (Keestra-Gounder et al., 2016; Mamińska et al., 2016; West et al., 2015). Although Vps37B depletion did not promote inflammatory gene transcription in CRC cell lines, the subset of advanced stage CRC patients with decreased VPS37B expression in our meta-analysis showed elevated mRNA abundance of CXCL8 and ICAM1 (Table S6). In our in vitro experimental settings, expression of these genes was increased only upon concurrent knockdown of either VPS37AB or VPS37ABC. If a subgroup of CRC patients with loss of both VPS37A and VPS37B was identified, it would be worth testing whether they display an inflammatory phenotype that could be modulated pharmacologically. However, inflammatory gene expression in advanced stage CRC patients is likely a result of multiple lesions accumulated in the course of disease progression. In addition to the expression of several NF-κB-dependent cytokines that in our GO analysis of biological processes were annotated to terms related to (chemo-)taxis of immune system cells, VPS37AB- and VPS37ABC-knockdown cells produced high levels of the cell cycle inhibitor p21. Although several papers have described the canonical p65–p50 NF-κB dimers as regulators of CDKN1A/p21-dependent cell cycle arrest in normal and cancer cells (Basile et al., 2003; Hinata et al., 2003; Nicolae et al., 2018; Wuerzberger-Davis et al., 2005), our data point to a different mechanism. Plausibly, it involves the release of Tsg101-mediated repression of the CDKN1A promoter (Lin et al., 2013), which in our study correlated with ESCRT-I destabilization after co-depletion of Vps37 proteins. TSG101^−/− knockout mice show accumulation of the p53 (TP53) transcription factor and a proliferative block leading to early embryonic lethality (Ruland et al., 2001; Wagner et al., 2003). Notably, Tsg101 deficiency also causes instability of the ubiquitin E3 ligase MDM2 and accumulation of p53 and its downstream target p21 (Carstens et al., 2004). Thus, it is plausible that ESCRT-I destabilization upon Vps37 co-depletion might induce p53-driven transcription of CDKN1A (El-Deiry, 2003). However, the CDKN1A promoter bears SP1, SP3, BRCA1, E2F1 and E2F3 transcription factor binding sites (Abbas and Dutta, 2009), thus its regulation may be complex. The p21 protein regulates the NF-κB pathway in macrophages (Rackov et al., 2016; Trakala et al., 2009), but we excluded that a similar mechanism occurs in CRC cells. Thus, we concluded that inflammatory response induction and inhibition of cell growth after VPS37 co-silencing are two independent and parallel processes.

The most important finding of this study is that differential depletion of Vps37 proteins elicits distinct effects on ESCRT-I subunit stability that align with the type and magnitude of transcriptional responses. Our data reinforce the notion that Vps37 proteins dictate the incorporation of UBAP-1, Mvb12A, and Mvb12B, leading to the assembly of distinct and functionally non-redundant ESCRT-I complexes. More specifically, our data indicate a partnering preference of Vps37A for UBAP-1 and Vps37B for Mvb12A. This is in line with previous studies on ESCRT-I stability, which contradicted the stochastic association of ESCRT-I components (Wunderley et al., 2014). We extended these studies, showing that co-silencing of all VPS37 paralogs leads to the nearly complete destabilization of remaining ESCRT-I subunits, resembling effects achieved upon either TSG101 or VPS28 knockdown (Table S5). It also explains why we observe similar effects on cell homeostasis, namely induction of an inflammatory response and cell growth inhibition, upon knockdown of TSG101 and co-silencing of VPS37 paralogs and overall similarities in transcriptional responses (Brankatschk et al., 2012; Mamińska et al., 2016; Miller et al., 2018). The degree of ESCRT-I destabilization after (co-)silencing of VPS37 paralogs correlates with the type and magnitude of transcriptional responses; however, the precise mechanism remains to be determined. Destabilization of Vps37 proteins upon Tsg101 depletion has been well documented (Bache et al., 2004; Stefani et al., 2011; Stuchell et al., 2004; Wunderley et al., 2014). Here, we observed mild but distinct transcriptional alterations after either Vps37B or Vps37C depletion that might result from their partly overlapping functions. Both proteins possess the PRR domain and may share similar binding partners. Indeed, using the BioGRID database (Oughtred et al., 2019), we found that both Vps37B and Vps37C have large and partially overlapping interactomes, whereas Vps37A, bereft of PRR, has only a few interacting proteins. Thus, VPS37A as the only paralog encoding the UEV domain might execute functions that cannot be taken over by any other family member. As a consequence, its depletion induces expression of a distinct set of genes compared to those expressed after knockdown of either VPS37B or VPS37C. We note that, at least under some conditions, the cell can very well compensate for the loss of a single VPS37 paralog, as illustrated by our RNA-Seq analysis. On the other hand, co-silencing of either VPS37AB or VPS37ABC evokes profound transcriptional alterations that we believe, by similarity to Tsg101 or Vps28 depletion, arise from adverse effects of non-degraded plasma membrane proteins and alterations in protein networks that may contribute to prolonged oncogenic signaling (Mamińska et al., 2016). In the context of cancer, ESCRT-I destabilization after Vps37 co-depletion could result in a more adverse tumor phenotype. This notion is consistent with our transcriptomic analysis of CRC cells depleted of Vps37 proteins, which identified processes related to cell migration, growth and signaling. Downregulation of ESCRT-I components has been shown in vitro to prolong epidermal growth factor signaling, sensitize cells to low doses of transforming growth factor, as well as promote cell invasion and migration through the process of epithelial to mesenchymal transition (Miller et al., 2018; Yang et al., 2016).

In summary, we have established that ESCRT-I subunit destabilization after co-depletion of Vps37 proteins evokes profound cellular stress manifested by a sterile inflammatory response and cell growth arrest. Our findings reveal potential vulnerabilities of CRC cells with reduced levels of VPS37A and VPS37B that may be more susceptible to chemotherapeutics and pharmacological modulators of inflammatory responses. We identified candidates with known functions in endocytosis, beyond VPS37 paralogs, whose expression is changed in CRC and thus warrant further investigation in the context of cancer cell pathophysiology.

Human HEK293 (CRL-1573), DLD1 (CCL-221) and RKO (CRL-2577) cell lines were obtained from American Type Culture Collection (ATCC). HEK293 and DLD1 were cultured in Dulbecco's modified Eagle's medium (DMEM; Sigma-Aldrich, M2279) supplemented with 10% (v/v) fetal bovine serum (FBS; Sigma-Aldrich, F7524) and 2 mM L-Glutamine (Sigma-Aldrich, G7513). RKO were maintained in Eagle's minimum essential medium (EMEM; ATCC, 30-2003) supplemented with 10% (v/v) FBS. Cell lines were passaged using 0.05% trypsin+EDTA (Sigma-Aldrich, T4049). Cells were cultured in an incubator at 37°C in a humidified atmosphere containing 5% CO₂. During the study, cells were regularly tested for mycoplasma, and the identities of DLD1 and RKO were confirmed by short tandem repeat (STR) profiling performed by the ATCC Cell Authentication Service.

Cells were either forward- or reverse-transfected with siRNAs using Lipofectamine RNAiMAX transfection reagent, according to the manufacturer's instructions (Thermo Fisher Scientific, 13778150). The concentration of each single siRNA duplex used for transfection was 20 nM. In experiments with simultaneous knockdown of two, three and four genes, the total concentration of siRNA was 40 nM, 60 nM and 80 nM, respectively, and the proportions of individual siRNAs duplexes were kept equal. The following PreDesigned or Validated Ambion Silencer Select siRNAs (Thermo Fisher Scientific) were used: negative control no. 1 (siNC#1, 4390843) and negative control no. 2 (siNC#2, 4390846); on-target siVPS37A#1 (s44037), siVPS37A#2 (s44038), siVPS37B#1 (s36177), siVPS37B#2 (s36178), siVPS37C#1 (s30059), siVPS37C#2 (s30060), siRELA (s11916), siCDKN1A#1 (s415), siCDKN1A#2 (s417), siTSG101#1 (s14439), siTSG101#2 (s14440), siVPS28#1 (s27577), siVPS28#2 (s27579), siUBAP1#1 (s27812), siUBAP1#2 (s27813), siMVB12A#1 (s41121), siMVB12A#2 (s41122), siMVB12B#1 (s40157) and siMVB12B#2 (s40158). Additionally, two custom-ordered Silencer Select duplexes were used: negative control no. 3 (NC3; sense strand, 5′-UACGACCGGUCUAUCGUAGtt-3′; antisense strand, 5′-CUACGAUAGACCGGUCGUAtt-3′) and negative control no. 4 (NC4; sense strand, 5′-UUCUCCGAACGUGUCACGUtt-3′; antisense strand, 5′-ACGUGACACGUUCGGAGAAtt-3′). The composition of siRNA mixes in experiments with individual and concurrent gene silencing is listed in Table S7.

Cells were plated in 12-well plate format at a density of 60,000 cells/ml in 1 ml of medium. After 16–24 h, cells were left non-transfected or differentially transfected according to the forward transfection protocol. 72 h later, cells were washed with phosphate-buffered saline (PBS), and the cell pellet was collected. Sequencing libraries were generated using an Ion AmpliSeq Transcriptome Human Gene Expression Panel (Thermo Fisher Scientific). Sequencing was performed using an Ion Proton instrument with seven or eight samples per chip, with an Ion PI Hi-Q Sequencing 200 Kit (Thermo Fisher Scientific). Reads were aligned to the hg19 AmpliSeq Transcriptome ERCC v1 using Torrent Mapping Alignment Program (version 5.0.4, Thermo Fisher Scientific). Transcripts were quantified using HTseq-count (version 0.6.0) run with default settings (Anders et al., 2015).

Gene level differential expression analysis was performed using the R package DESeq2 (version 1.18.1; Love et al., 2014) for genes with at least 100 counts across conditions by taking into the account the batch effect and applying the following contrasts (α=0.05): NT (non-transfected) versus siCTRL#1, NT versus siVPS37A#1, NT versus siVPS37B#1, NT versus siVPS37C#1, NT versus siVPS37AB#1, NT versus siVPS37AC#1, NT versus siVPS37BC#1, NT versus siVPS37ABC#1, siCTRL#1-T versus siVPS37A#1, siCTRL#1-T versus siVPS37B#1, siCTRL#1-T versus siVPS37C#1, siCTRL#1-T versus siVPS37AB#1, siCTRL#1-T versus siVPS37AC#1, siCTRL#1-T versus siVPS37BC#1, siCTRL#1-T versus siVPS37ABC#1. We excluded non-protein-coding genes from downstream analysis.

The overlap for different silencing conditions and normalization contrasts was visualized using the VennDiagram package (version 1.6.20). Genes in the set of overlapping differentially expressed genes in the different on-target siRNA conditions, normalized against either NT or siCTRL#1- transfected patterns, were subjected to GO analysis of biological processes and Reactome pathway analysis using clusterProfiler (version 3.6.0; Yu et al., 2012) and ReactomePA R-packages (version 3.8; Yu and He, 2016) taking advantage of enrichGO and enrichPathway functions, respectively. All enrichment P-values in the GO analysis were corrected for multiple testing using the Benjamini–Hochberg method, and only genes with adjusted P-value <0.05 were considered significant. The minimal and maximal sizes of gene clusters were set to 10 and 500, respectively. Redundant terms were removed by means of the simplify function with cutoff 0.6. Count data were transformed using the transcript per million (TPM) normalization and were scaled across conditions (Z-score). Differentially expressed genes binned in the selected GO processes were used for hierarchical clustering, which was performed on Euclidean distances using Ward's algorithm. Heatmaps of differentially expressed genes were visualized using ComplexHeatmap (version 1.17.1; Gu et al., 2016). All calculations were performed in R version 3.4.4 (https://www.R-project.org).

The code for this analysis is available on GitHub (https://github.com/kkolmus/VPS37_RNA-Seq). RNA-Seq data have been deposited at Gene Expression Omnibus (GEO) under the accession code: GSE152195.

Non-transfected cells or cells subjected to reverse transfection with different siRNAs were seeded at a density of 1000 cells per well in a 6-well plate format and cultured for 14 days to form colonies. For staining, colonies were washed with PBS, fixed for 5 min in a 3:1 (v/v) solution of acetic acid:methanol, and incubated for 15 min in 0.2% Crystal Violet solution in 70% ethanol. The whole procedure was performed at room temperature. Plates with colonies were scanned using an Odyssey Infrared System (LI-COR). Images were acquired using an Odyssey Infrared Imaging System (LI-COR) with the following parameters: resolution 169 μm; quality, high; focus offset 4.0 mm; intensity 700×800, 2.0×2.0, and were exported as ‘show this channel in black shapes on white background grayscale’ using the Image Studio software. Images were next analyzed adhering to the description outlined previously (Guzmán et al., 2014). Data are expressed as a percentage of the staining intensity displayed by non-transfected cells.

1500 cells were left non-transfected or were reverse-transfected with different siRNAs in a 96-well plate format and allowed to proliferate for 120 h. A BrdU Cell Proliferation ELISA assay (Roche, 11647229001) was used to assess the proliferation of RKO and DLD1 cells, according to the manufacturer's instructions with the following modifications: BrdU reagent was added 4 h prior to cell fixation and 100 μl of substrate solution was added for 5 min followed by addition of 50 μl of 1 M HCl. The colorimetric signal was detected at 450 nm using a Tecan Sunrise microplate reader system with Magellan v. 6.6 software. Data are expressed as the percentage of proliferating, non-transfected cells.

Cells were plated in either 6- or 12-well plate format at a density of 60,000 cells/ml in 1 and 2 ml of medium, respectively. After 16–24 h, cells were left non-transfected or differentially transfected according to the forward transfection protocol. 72 h later, cells were lysed in RIPA buffer (1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 50 mM Tris pH 7.4, 150 mM NaCl and 0.5 mM EDTA) supplemented with protease inhibitor cocktail (6 μg/ml chemostatin, 0.5 μg/ml leupeptin, 10 μg/ml antipain, 2 μg/ml aprotinin, 0.7 μg/ml pepstatin A and 10 μg/ml 4-amidinophenylmethanesulfonyl fluoride hydrochloride; Sigma-Aldrich) and phosphatase inhibitor cocktails 2 and 3 (P5726 and P0044, Sigma-Aldrich). Protein concentration was determined using a BCA Protein Assay kit (Thermo Fisher Scientific, 23225). 25–30 μg of total protein per sample was resolved on 12 or 15% SDS-PAGE gels, transferred onto a nitrocellulose membrane (Amersham Hybond, GE Healthcare Life Science, 10600002), blocked with 5% milk in PBS with 1% Tween, probed first with specific primary and then secondary antibodies, and imaged using detection solution (BioRad, 170-5061) and a ChemiDoc imaging system (Bio-Rad). Because of different molecular masses of the analyzed proteins, nitrocellulose membranes were routinely cut into pieces and a single loading control was used for densitometry analysis of several proteins of interest. In Fig. 4, vinculin blots shown in panels F and G, as well as H and I are parts of the same membranes. In Fig. 6, vinculin blots shown in panels A, C and D, as well as G and H are parts of the same membranes. In Fig. 7, vinculin blots shown in panels A, B, F and H, as well as C, D and E are parts of the same membranes. In Fig. S2, vinculin blots shown in panels A and B are parts of the same membranes. In Fig. S5, vinculin blots shown in panels A and B, as well as J, K, and L are parts of the same membranes. In Fig. S6, vinculin blots shown in panels A, B, E and G, as well as C, D, F and H are parts of the same membranes. All primary antibodies are listed in Table S8. Secondary horseradish peroxidase-conjugated anti-mouse (315-005-008), anti-rabbit (111-035-144) and anti-goat (305-035-046) antibodies were from Jackson ImmunoResearch and were used at working dilution of 1:10,000. Densitometry of protein bands was carried out using ImageJ software (Schneider et al., 2012). p65 was used as the loading control for quantification of phosphorylated p65. Vinculin was used as a loading control in all other experiments. Results are presented as fold change compared to non-transfected cells.

Cells were seeded in a 12-well plate format at a density of 60,000 cells/ml in 1 ml of medium. After 16–24 h, cells were left non-transfected or differentially transfected according to the forward transfection protocol. 72 h later, total RNA was isolated using a High Pure Isolation kit (Roche, 11828665001). 500 ng of total RNA was subjected to cDNA synthesis. M-MLV, random nonamers and oligo(dT)₂₃ (Sigma-Aldrich; M1302, R7647 and O4387, respectively) were used for cDNA synthesis according to the manufacturer's instructions. Expression of genes of interest was measured using primers designed with the NCBI Primer designing tool and custom synthesized by Sigma-Aldrich. Primers used in this study are listed in Table S9. Real-time cDNA amplification was performed with the Kapa Sybr Fast qPCR Kit (KapaBiosystems, KK4618). Fluorescence was monitored using a 7900HT Fast real-time PCR thermocycler (Applied Biosystems). Expression of each gene was normalized to either expression of the ACTB (β-actin) or GAPDH (glyceraldehyde 3-phosphate dehydrogenase) reference genes. Results are presented as fold change compared to non-transfected cells. For clarity, the y-axis is interrupted in some cases.

Samples of the normal colon (n=24) and adenocarcinoma (n=26) had been collected for the purpose of previous studies (Mikula et al., 2011; Skrzypczak et al., 2010). In order to determine the abundance of VPS37A and VPS37B transcripts, RT-qPCR was performed as described before (Mikula et al., 2011; Skrzypczak et al., 2010). The sequences of primers for VPS37A and VPS37B are listed in Table S9.

Cells were seeded in a 6-well plate format at a density of 60,000 cells/ml in 2 ml of medium. After 16–24 h cells were left non-transfected or differentially transfected according to the forward transfection protocol. At 96 h post-transfection, cells were briefly washed with PBS, harvested with trypsin+EDTA, washed twice with PBS and fixed for 24 h in ice-cold 70% ethanol. Washed cells were then incubated first with extraction buffer (4 mM citric acid in 0.2 M Na₂HPO₄) for 5 min at room temperature, and next with staining solution [3.8 mM sodium citrate, 50 μg/ml propidium iodide (PI) and 0.5 mg/ml RNase A] for 30 min at room temperature. Analysis of cells was performed on a BD LSRFortessa flow cytometer (Becton Dickinson). A total of 10,000 cells from the single-cell gate were counted for each transfection condition. Flow cytometry data were plotted and analyzed using FlowJo (Becton Dickinson) and ModFit LT (Verity Software House) software. Data is presented as percentage of analyzed cells in the given cell cycle phase.

The study protocol for analysis of protein levels of Vps37A and Vps37B in human normal colon and CRC samples was approved by the Bioethics Committee of the Maria Skłodowska-Curie National Research Institute of Oncology in Warsaw (decision no. 40/2017). Informed consent was obtained from all subjects. The experiment conformed to the principles set out in the WMA Declaration of Helsinki and the Department of Health and Human Services Belmont Report. High-density tissue microarrays were constructed from formalin-fixed, paraffin-embedded diagnostic samples of 100 pairs of treatment-naïve CRC tissues and matched normal colon samples from the collection of the Maria Skłodowska-Curie National Research Institute of Oncology in Warsaw. IHC was performed using automated immunohistochemical stainer (Dako Denmark A/S) and specific anti-VPS37A and anti-VPS37B antibodies listed in Table S8. An EnVision Detection System (Agilent) was used for detection. Samples were reviewed for the abundance of Vps37 proteins in normal and neoplastic tissue by two pathologists who were blinded to the outcome. A semi-quantitative method was applied for IHC evaluation, involving a scoring system based on the staining intensity: 0, no staining; 1+, weak staining; 2+, medium intense staining; 3+, strong intensity staining. Staining homogeneity was above 90%.

Clinicopathological and transcriptional profiles from the two TCGA cohorts: rectum adenocarcinoma (READ) and colon adenocarcinoma (COAD) were retrieved using the TCGAbiolinks package (Colaprico et al., 2016). READ and COAD datasets were analyzed together, because a previous study showed a major overlap in their expression patterns (Weinstein et al., 2013). The presented analysis encompassed only matched tumor–normal tissue samples for 31 patients with available clinicopathological data. 19 and 12 of these patients were assigned to the early stage group (encompassing stages I and II) and the advanced stage group (encompassing stages III and IV), respectively. Matching TCGA sequencing data acquired using an Illumina HiSeq platform were used. Differential gene expression was performed using DESeq2 (version 1.18.1; Love et al., 2014) with the default settings, for genes with at least 50 counts across samples taking into account correction of batch effect relying on the inter-plate variation.

Raw files for the GSE50670 dataset (Kim et al., 2014), which profiled transcriptomes of 18 CRC patients belonging to the advanced stages of carcinogenesis (stage IV), were downloaded using SRA-Tools, trimmed according to quality using Trimmomatic (version 0.39; Bolger et al., 2014) with default parameters, except MINLEN, which was set to 50. Trimmed sequences were mapped to the human reference genome provided by ENSEMBL (version grch38_snp_tran) using Hisat2 (Kim et al., 2015) with defaults parameters. Metadata on tissue diagnosis as well as tumor staging associated with the analyzed sample were retrieved using the R package GEOquery (Davis and Meltzer, 2007).

In order to perform meta-analysis of TCGA and GSE50760 datasets, we averaged fold change ratios recovered after differential gene expression with DEseq2 and calculated new P-values from an independent hypothesis test using the Stouffer's z method implemented in the combinePValues function from the R scran package (Lun et al., 2016).

Volcano plots and boxplots of differentially expressed genes were prepared using ggplot2 (CRAN available package). Heatmaps of differentially expressed genes were visualized using ComplexHeatmap (version 1.17.1; Gu et al., 2016). The overlap between differentially expressed genes was visualized using the VennDigram package (CRAN available package). All calculations were performed in R version 3.6.1 (https://www.R-project.org). The code for this analysis is available on GitHub (https://github.com/kkolmus/CRC_transcriptomics_proj).

Data are shown as mean±s.d. of at least three independent biological experiments where non-transfected cells (NT) were used as reference to calculate fold change ratios. The value for NT cells was set to 1 for RT-qPCR and immunoblotting experiments or to 100% for proliferation and colony formation assays. The only exceptions are Fig. 5I,J and Fig. S4K–M, which show the percentage of cells in the given phase of the cell cycle upon FACS analysis, and Fig. 6A–D, which show fold change ratio calculated against siCTRL#1-transfected cells. For RT-qPCR and the BrdU assay, samples were assayed in technical triplicates. Statistical analysis was performed using GraphPad Prism 6 software. Data with normal distribution were analyzed using either one-sample t-test or one-way ANOVA followed by Dunnett's correction. In the case of non-normal distribution, the Mann–Whitney U-test was applied. Analysis of categorical data was performed using Fisher's exact test. The significance of mean comparison is annotated as follows: ns, not significant (P≥0.05); *P<0.05; **P<0.01; ***P<0.001; ****P<0.0001. Results were considered significant when P<0.05. No statistical methods were used to predetermine sample size.

We thank A. Zeigerer (Helmholtz Zentrum München) and members of the Miączyńska laboratory for critical reading of the manuscript. We are grateful to D. Zdżalik-Bielecka for help with pilot FACS experiments, and A. Paziewska and A. Dąbrowska for their technical support in RNA-Seq analysis.

The peer review history is available online at https://jcs.biologists.org/lookup/doi/10.1242/jcs.250951.reviewer-comments.pdf