Genome-wide identification of alternative splicing events that regulate protein transport across the secretory pathway

After biosynthesis in the endoplasmic reticulum (ER), proteins are transported to their destination via the secretory pathway. Transport processes are highly flexible and adapt to new requirements, for instance during differentiation or after activation (Farhan and Rabouille, 2011; McCaughey and Stephens, 2018). This adaptation has been studied in regard to differential gene expression (Dunne et al., 2002; Coutinho et al., 2004; Schotman et al., 2009), changes in membrane morphology and dynamics (Forster et al., 2006; Guo and Linstedt, 2006; Farhan et al., 2008), and altered activity of kinases and phosphatases (Farhan et al., 2010). Furthermore, expression of different paralogs of the COPII coat has been suggested to be involved in conferring tissue-specific functionality. For example, the COPII inner coat SEC24A, SEC24B, SEC24C and SEC24D proteins have been reported to show differential interaction patterns (Adolf et al., 2016) and knockout of SEC24C leads to tissue-restricted neurological phenotypes in mice (Wang et al., 2018), implying a role for paralog expression in controlling tissue-specific functionality. Nevertheless, the regulation of protein transport and the adaptation to different cellular requirements remain incompletely understood.

Alternative splicing (AS), which strongly increases proteome diversity and often leads to severely different interaction surfaces (Pan et al., 2008; Yang et al., 2016), has been widely overlooked as a regulator of membrane trafficking. AS has been shown to react to various external stimuli in a highly dynamic manner (Heyd and Lynch, 2011; Preußner et al., 2017), and would thus be perfectly suited to control protein secretion in response to changing cellular environments. Although such a connection has been suggested, this was mainly based on in silico predictions (Blue et al., 2018). We have previously described how AS of SEC16A exon 29 increases the efficiency of the early secretory pathway (Wilhelmi et al., 2016) but this remains one isolated example. Here, we identify over 200 candidate AS events that could act to modulate the secretory pathway in a tissue-specific and activation- and differentiation-dependent manner. Our genome-wide approach, together with selected experimental validations provides evidence for a global control of protein secretion by AS.

To identify regulators of the secretory pathway, several RNAi screens have been performed (Farhan, 2015). One of these screens used the VSVG reporter to monitor protein transport upon knockdown (KD) of 22,000 human genes (Simpson et al., 2012). In this screen, four RNA-binding proteins (RBPs) that regulate AS and whose KD led to a strong inhibition of secretion were identified (Table S1): HNRNPA1, PTBP1, RBM27 and SRSF1 (Fig. 1A). As these RBPs have no known direct function in protein secretion, we considered that they could exert an indirect effect, through controlling AS of components of the secretory pathway, which then changes cargo flux (Fig. 1B). We therefore used published KD RNA-Seq datasets for these RBPs (see Table S1 for accession numbers) to globally determine their effect on AS. We propose that the impact of these RBPs on the secretory pathway is mediated by a network of shared AS events, and thus generated an overlap of alternative isoforms that showed differential splicing in the single KDs compared to control datasets. We propose that splice variants that are regulated by at least three of the four RBPs are potential secretion modulators, and henceforth call their formation a secretion-related AS event (Fig. 1C). Meta-analysis of these events, which mostly involve cassette exons, indicates that they have a modulatory effect on protein function instead of regulating total gene expression, as they lie almost exclusively within the coding sequence and consist of fewer nonsense-mediated decay (NMD)-inducing exons than is seen for the the skipped exons that are regulated by any of these RBPs (i.e. the RBP background, defined as all AS events that show differential splicing upon any of the KDs; Fig. S1). When performing gene ontology (GO) analysis, only three terms for biological processes were found to be significantly enriched, all of which fit the hypothesis that the secretion-related AS events play a role in membrane trafficking (Fig. 1D). We then grouped the AS-controlled proteins by function (Fig. 1E). As expected, a large cluster is directly connected to the secretory pathway or to cytoskeletal organizers that can provide scaffolding functions for vesicle transport. Additionally, we find phosphoinositide homeostasis and ARF signalling, which are known to regulate vesicular trafficking (Ooms et al., 2009) and proteins involved in kinase and phosphatase signalling, which can post-translationally control trafficking proteins (Farhan et al., 2010). Apart from these targets, there are several groups that may have an indirect effect on secretion. First, transcriptional, post-transcriptional and translational modulators can change gene and protein expression of various targets, which can be further controlled by degradation pathways. Second, mitochondrial proteins could impact on mitochondria-related traffic. Finally, there are several chaperones, a group of proteins which have been shown to influence protein secretion as well (Roth et al., 2012). Closer inspection of the proteins known to be a direct part of the trafficking machinery revealed that they are localized in all compartments along the secretory pathway (Fig. 1E). While we focussed our analysis on AS events controlled by the four RBPs, it is interesting to note that all four RBPs have been reported to shuttle between the nucleus and the cytoplasm (Kamath et al., 2001; Iervolino et al., 2002; Kavanagh et al., 2005; Twyffels et al., 2011). A further impact of one or several of these RBPs in controlling protein secretion, for example, by controlling stability or translation of mRNAs encoding for secretion-associated proteins, is thus conceivable but remains to be experimentally addressed.

To validate the functionality of the secretion-related AS events, we employed the RUSH system (Boncompain et al., 2012) in which a GFP–GPI reporter is transported from the ER to the plasma membrane (PM) after addition of biotin. Upon arrival of the reporter at the PM, GFP will be located on the extracellular side of the cell where its amount can be quantified by staining against GFP without permeabilizing the cell (Fig. 2A). A time course experiment in HEK293T cells and the quantification of antibody-stained (PM-localized) GFP per cell is shown in Fig. 2B. To verify that expression levels of HNRNPA1, PTBP1, RBM27 and SRSF1 had an effect on protein transport in our assay, we performed KDs of these proteins and of two further RBPs (MBNL3 and SRSF6), as a control (Fig. 2C). The RUSH assay was subsequently performed by staining against GFP 1 h after biotin addition. Although we did not observe a significant difference between control and either the MBNL3, SRSF6 or HNRNPA1 KDs, cells treated with siRNA against PTBP1, RBM27 and SRSF1 (denoted siPTBP1, siRBM27 and siSRSF1, respectively) showed a substantial and highly significant decrease in GFP surface staining (Fig. 2D,E). The lack of an influence for the HNRNPA1 KD (Simpson et al., 2012) may be due to the use of a different reporter in our study, which could point to a role in cargo selectivity. However, we also notice that HNRNPA1 has the lowest number of shared splicing targets among the four RBPs, as the largest overlap in our analysis are the targets controlled by PTBP1, RBM27 and SRSF1 (Fig. 1C). For further validations, we therefore focussed on events that were controlled by these three RBPs.

To validate the influence of individual splicing events on secretion, we selected four targets that are directly involved in different parts of the secretory pathway (Fig. 1E): for the early secretory pathway, we selected SEC31A, which is part of the outer coat of COPII vesicles (Gürkan et al., 2006) and SEC22C, which is a homolog of the SNARE protein Sec22 in yeast (Tang et al., 1998; Yamamoto et al., 2017). As examples for splicing events affecting Golgi and post-Golgi components of the secretory pathway, we selected OCRL, which acts at the trans-Golgi and later compartments as a phosphoinositide phosphatase (De Matteis et al., 2017) and EXOC7 (also known as EXO70), which is part of the exocyst complex involved in targeting vesicles to the PM (He and Guo, 2009). We used splice-site blocking morpholinos (MOs) to manipulate the AS of these targets without altering the endogenous gene expression level (Fig. 3A), thereby recapitulating changes in exon exclusion observed upon knockdown of the RBPs (Table S1). We then performed RUSH assays on these cells and indeed observed a highly significant decrease in GFP reporter transported to the PM for all candidates (Fig. 3B,C).

To independently validate the observed effect, we used CRISPR/Cas9 to generate isoform-specific knockouts (KOs). We selected the alternative microexon 19 in OCRL, as it stands out both because the MO-induced splicing change from ∼25% inclusion to complete exclusion led to a drastic transport defect and as this is achieved by exclusion of only eight amino acids. We generated four independent OCRL exon 19 KO cell lines and validated them both on the genomic and transcriptomic level (Figs S2A, S3D). Cell lines showed expression of only the exclusion isoform with the overall expression level remaining largely unchanged (Fig. S2B). We then performed the RUSH assay and again observed a highly significant reduction in the amount of surface GFP for all clones and that was in the same range as for the OCRL exon 19 MO-treated cells, further validating the MO approach (Fig. 3E). These data together validate our bioinformatics approach, as all tested splicing events indeed controlled protein transport efficiency. This strongly increases the confidence that our group of secretion-related AS events is genuine, thereby adding a new regulatory layer to the secretory pathway and substantially increasing the number of alternative isoforms with known cellular functionality.

To address the physiological significance of the connection between AS and protein transport, we investigated whether the secretion-related AS events act in a tissue-, differentiation- or activation-dependent manner. To this end, we initially used RNA-Seq data from various human organs to calculate inclusion levels for the secretion and the RBP background events. We indeed observed a tissue-specific usage for a larger proportion of the secretion-related AS events in comparison to what was seen for the RBP background (Fig. 4A; Fig. S3A). This visual impression is supported by two measures of variability. First, in comparison to the RBP background, the secretion-related AS events displayed more variable splicing in all two-tissue comparisons (i.e. a higher percentage of differentially spliced events, Fig. 4B). Second, the strongest percentage spliced in difference (dPSI) was on average also larger for the secretion-related AS events (Fig. 4C). This strongly points to a global role of AS in adapting the secretory pathway to tissue-specific requirements. Next, we turned to two cellular systems where cells with basal secretory requirements differentiate into a cell type with higher secretory load (Fig. 4D): T-cell activation and differentiation of pre-adipocytes into adipocytes. We used RNA-Seq datasets from primary human CD4⁺ T-cells and human SGBS adipocytes at pre- and post-differentiation to analyse gene expression and alternative splicing. When analysing differential AS, we found significant overlaps with the secretion-related AS events in both sets of data (Fig. 4E,F), from which we validated three targets each (Fig. S3B,C), implying that AS can adapt protein secretion upon activation and differentiation. Of note, the vast majority of secretion-related AS events act independently of transcription, as ∼80% of the corresponding genes show an expression fold change smaller than two in both model systems (Table S2). Altered AS may be controlled by HNRNPA1, PTBP1, RBM27 and SRSF1, which all show slightly increased mRNA expression upon T-cell activation and a modest reduction during adipocyte differentiation (Table S2). However, the activity of RBPs is extensively regulated at post-transcriptional and post-translational levels, which likely plays a determining role. We observed that while the T-cell overlap events lie within proteins found throughout the secretory pathway, the adipocyte overlap events lie within proteins that mainly locate in post-Golgi compartments, which is also reflected in a GO term analysis (Fig. S3D). Additionally, we found that components of the COPII machinery are upregulated during adipocyte differentiation but not during T-cell activation (Fig. 4G; Table S2). This suggests that adipocytes use transcriptional regulation in the early steps of protein secretion and AS in post-Golgi compartments to adapt their secretory pathway, whereas T-cells rely on AS to regulate secretion capacity during activation. This difference in the control of membrane trafficking may be explained by specific requirements of these cells after differentiation. Activated T-cells produce cytokines and cytotoxins at the ER and then transport them out of the cell, using the whole secretory pathway, which happens in a highly dynamic and temporally controlled manner (Huang et al., 2013). Adipocytes produce and export adipokines and also need to adapt their secretory machinery in this regard (Kuryszko et al., 2016), but this happens during a longer differentiation process that is more suitable for stable transcriptional changes. However, a fundamental task in mature adipocytes is the rapid shuttling of the glucose transporter GLUT4 (also known as SLC2A4) from post-Golgi vesicles to the PM in response to insulin and the recycling of the receptor (Stöckli et al., 2011). A major and more dynamic adaptation is therefore required for the post-Golgi trafficking machinery, which is achieved by AS. These data together strongly argue for a global role of AS in controlling the efficiency of the secretory pathway in a tissue-specific manner, as well as in dynamic settings such as during differentiation and activation (Fig. S3E).

In summary, we combine data from a genome-wide screen for modulators of protein secretion with knockdown RNA-Seq datasets to discover over 200 AS events that, based on our validations using the RUSH system, are high-confidence secretion regulators. Our analysis shows that AS is involved in regulating all stages of the secretory pathway (Fig. 4H). We furthermore show that secretion-regulating AS events are used in diverse biological contexts to adapt the secretory pathway to tissue-specific or differentiation- and activation-dependent requirements (Fig. 4H; Fig. S3E). Although individual AS events have been reported to play a role in membrane trafficking (Wilhelmi et al., 2016; Valladolid-Acebes et al., 2015; Blue et al., 2018), we show, in a system-wide manner, that the secretory pathway is regulated by AS at all stages. Our findings thus add an additional layer of complexity to the regulation of the secretory pathway. In addition, the dynamic nature of these splicing events in various biological contexts provides an important step towards understanding differentiation-specific control of protein secretion. AS might help the cell to adapt the secretory pathway in an intermediate timeframe, supplementing the very fast modulations of kinases and phosphatases and the slower effects of transcription. Despite clear variations in cargo load and specificity in different cells and tissues, the molecular basis for these distinct adaptations remains largely enigmatic. Our data provide evidence that cell-type-specific adaption of protein secretion is, at least partially, controlled by a network of splicing changes. Our finding that several RBPs are involved in controlling this process is consistent with splicing decisions being under combinatorial control of several or many cis- and trans-acting factors, and explains how different cell types can adapt their splicing patterns to the respective requirements. The expression and activity of these RBPs can be individually controlled to result in a variety of activities that is tailored to the secretory requirement of the respective cell type.

Accession numbers and all results from analyses are listed in Tables S1 and S2. RNA-Seq analyses were essentially performed as previously described (Herdt et al., 2017). In short, reads were mapped to the hg38 human genome using STAR version 2.5.3a (Dobin et al., 2013). For AS analyses, the ‘mixture of isoforms’ (MISO) (Katz et al., 2010) version 0.5.3 with a custom-made annotation was used. Differential events are defined by having a minimum ΔPSI of 10% and a Bayes factor of greater than 5. Replicates were merged for analyses after mapping. Gene expression analysis was performed using DESeq2 (Anders et al., 2012). Further analyses were performed using custom Python scripts (available from the corresponding author upon request). GO term analyses were performed using the PANTHER classification system version 13.1 (https://pantherdb.org) (Mi et al., 2013). Network maps were generated using Cytoscape version 3.6.1 (Montojo et al., 2010).

HEK293T cells were cultivated in high-glucose Dulbecco's modified Eagle's medium (DMEM; Biowest, Nuaillé, France) containing 10% fetal calf serum (FCS; Biochrom, Berlin, Germany) and 1% penicillin/streptomycin (Biowest) at 37°C and 5% CO₂. These cells have been extensively used in our laboratory (Herdt et al., 2017; Preußner et al., 2017; Goldammer et al., 2018) and test negative for mycoplasma contamination on a monthly basis. Plasmid transfections were performed using RotiFect (Carl Roth, Karlsruhe, Germany) following the manufacturers' instructions. MOs were obtained from Gene Tools. They were transfected at a final concentration of 3 µM using Endo-Porter following manufacturers' instructions. For KDs, a pool of four siRNAs (Dharmacon, Lafayette, Colorado, US) was used at final concentration of 10 nM. They were transfected using HiPerFect (Qiagen, Venlo, Netherlands) according to the manual. Genome-editing using CRISPR/Cas9 was performed as previously described (Wilhelmi et al., 2016). All sequences for MOs, siRNAs, guide sequences and genotyping primers are listed in the Table S3.

Human SGBS cells were cultivated and differentiated as described previously (Wabitsch et al., 2001; Fischer-Posovszky et al., 2008). In brief, cells were seeded in DMEM/F12 containing 10% FCS at a density of 4×10⁴ cells per well in six-well plates for 3 days to reach confluency. Differentiation was started by application of a specific cocktail (2 μmol/l rosiglitazone, 25 nmol/l dexamethasone, 0.5 mmol/l methyliso-buthylxantine, 0.1 μmol/l cortisol, 0.01 mg/ml transferrin, 0.2 nmol/l triiodotyronin and 20 nmol/l human insulin) in DMEM/F12 without serum and albumin. Medium was changed every 4 days (DMEM/F12 0.1 μmol/l cortisol, 0.01 mg/ml transferrin, 0.2 nmol/l triiodotyronin, and 20 nmol/l human insulin). Cells were harvested in Qiazol 3 days after seeding (pre-adipocytes) and 14 days post differentiation (adipocytes) for RNA extraction using the RNeasy MinElute Kit (Qiagen) according to manufacturers' instructions.

The RUSH plasmid consists of a KDEL ER hook and an EGFP-GPI reporter, a gift from Franck Perez (available from Addgene, #65294).

RNAs from primary human T-cells were prepared as described (Michel et al., 2014). RNA extraction, radioactive RT-PCR and Phosphor imager quantification were performed as previously described (Wilhelmi et al., 2016). RNA was extracted using RNATri (Bio&Sell, Feucht, Germany). For reverse transcription, 1 µg RNA was used with a gene-specific reverse primer. Low-cycle number PCRs were performed with a radioactively labelled forward primer, and products were separated by denaturing PAGE. Imaging was performed using a Phosphoimager and gels quantified using ImageQuantTL version 8.1 (GE, Boston, Massachusetts, US). Quantitative reverse transcription PCR (qRT-PCR) was performed in technical duplicates in a 96-well format using an Absolute qPCR SYBR Green Mix (Thermo Fisher, Waltham, MA) on a Stratagene (San Diego, CA) Mx3000P machine. Mean values of the technical duplicates were used and expression normalized to GAPDH. Non-radioactive genomic PCR products were separated on an agarose gel. See figure legends for number of independent biological replicates. Primer sequences are provided in Table S3.

The day before transfection, cells were seeded on precision coverslips (0.17 mm thickness, Sigma) that had previously been coated with poly-L-lysine (Sigma) for 1 h. For the standard RUSH assay, 1.0×10⁵ cells were seeded. At 24 h post seeding, cells were transfected with the RUSH plasmid, and at 16 h post transfection, protein synthesis was stopped by addition of cycloheximide (10 µg/ml final concentration, Carl Roth) and D-biotin (40 µM final concentration, Sigma) was added to release the reporter. The assay was stopped, usually after 1 h unless indicated otherwise, by washing with ice-cold PBS (Biowest) and fixation of cells using 4% formaldehyde (Carl Roth) solution in PBS for 10 min. After 1 h of blocking using 5% goat serum (Sigma) in PBS, primary rabbit anti-GFP antibody (Jamieson et al., 2015; Invitrogen, Carlsbad, CA; cat. #A11122, batch #1891900, 1:200) was applied for 2 h at room temperature. After washing, the secondary donkey Alexa Fluor 594-conjugated anti-rabbit-IgG antibody (Ram et al., 2017, Invitrogen, cat# A21207, batch# 1107500, 1:200) was applied for 1 h. Coverslips were mounted on microscopy slides (Roth) with ProLong Gold antifade mountant with DAPI (Thermo Fisher). Slides were stored at room temperature for 24 h and either directly imaged or stored at 4°C until imaging. When the RUSH assay was combined with either MO or siRNA treatment, only 0.5×10⁵ cells were seeded to account for the additional day of treatment. MOs or siRNAs were transfected 32 h before the RUSH plasmid.

Fluorescence microscopy was performed on a Leica SP8 confocal microscope. Image analysis was performed using custom Python scripts (available from the corresponding author upon request). To calculate the percentage of GFP surface staining, single cells were initially defined by user input. Pixels were defined as GFP-positive (green value) or GFP staining-positive (red value) if the respective intensity value was above a certain threshold (75 for green, 65 for red fluorescence). The GFP surface staining was calculated as the number of double-positive pixels divided by the number of green-positive pixels multiplied by 100.

If not mentioned otherwise, either one-sample or unpaired t-tests were used to calculate P values. P values are indicated by asterisks and explained in the legend of each figure; n numbers in the legends represent biological replicates. P-value calculations were only performed if at least three biological replicate values were obtained.

The authors thank the HPC Service of ZEDAT, Freie Universität Berlin, for computing time. We thank Franck Perez for generously providing the RUSH plasmid and Rainer Pepperkok for sharing unpublished information on the genome-wide screen performed in his laboratory. We also thank members of the Heyd laboratory for discussions and comments on the manuscript.