Targeting alternative splicing as a new cancer immunotherapy-phosphorylation of serine arginine-rich splicing factor (SRSF1) by SR protein kinase 1 (SRPK1) regulates alternative splicing of PD1 to generate a soluble antagonistic isoform that prevents T cell exhaustion

In cancers, widespread alteration in alternative splicing of multiple genes has been identified. In myeloid leukaemia evidence has been produced for a therapeutic approach targeting components of the spliceosome such as SF3B1 [1], and of controlling regulators of alternative splicing such as the SR protein kinase SRPK1 [2, 3]. These are moving into the clinic with phase I trials of SF3B1. However, these have been targets in the cancer cells themselves. The tumour microenvironment, and in particular the immune system, is a key regulator of tumour growth and metastasis, but also could be a target for alternative splicing therapy.

In many cancers tumour progression and metastasis occur because the immune system of the patient cannot target cancer cells. This is in part because the cancer cells develop the ability to control the host immune system by repressing immune cells that would normally target the cancer as being a foreign object – and in particular the T cell. There are multiple pathways through which cancer cells can repress the immune system, including validated targets that now form the bedrock of immunotherapy, including the PD1-PDL1 and the CTLA4-CD80/86 pathways. Programmed cell death ligand-1 (PD-L1), the target of atezolizumab, durvalumab and others acts on the programmed cell death receptor -1, (PDCD1 or PD-1), the target of nivolumab, pembrolizumab and others, on T cells and induces T cell exhaustion, preventing the T cell from mounting an immune response against the cancer. The cytotoxic T-lymphocyte-associated-4 (CTLA4) protein binds to CD80 and CD86 ligands on antigen presenting cells and prevents the stimulatory CD28 from activating T cells. CTLA4 is the target of ipilimumab, the first checkpoint inhibitor to be licensed, and tremelimumab. Alternative splicing of these molecules, particularly resulting in exclusion of the transmembrane domain, could lead to soluble endogenous inhibitors and such alternative splicing has been shown for CD28 [4], CD80 [5], CD86 [6], CTLA4 [7] and for PD-1 [8].

The pdcd1 gene consists of 5 exons with each exon encoding different components of the protein; Exon 1 the signal sequence, Exons 1 and 2 the extracellular ligand binding domains, exon 3 the transmembrane domain and exons 4 and 5 the intracellular signalling domains. (Fig. 1A). The full-length functional PD-1 receptor (262 amino acids, 32 kDa, but with glycosylation, approximately 50 kDa) requires all five exons but alternative splicing of exon 3 (∆Ex3) could result in a protein lacking the 52 amino acids encoded by exon 3, which includes the entire transmembrane domain, and hence produce a soluble version of PD-1 protein (25 kDa) that would be secreted from the T cell, (∆Ex3PD1, a soluble PD-1, or sPD1) which could bind to PD-L1 and PD-L2, and prevent activation of the endogenous full-length receptor. This PD-1 splice variant was originally shown to be present by Nielsen et al. in 2015 [8] and has now been shown in coeliac disease [9], but its role has not been explored, nor the control of splicing found. As a therapeutic strategy, switching the splicing from flPD1 to ∆Ex3PD1 could prevent cancer cell-mediated T cell exhaustion and enable an anti-cancer immune response. We therefore aimed to test the hypothesis that alternative splicing of PD-1 could result in checkpoint inhibition.

PBMCs were isolated from healthy donors. Unstimulated cells did not show PD-1 expression by PCR, but treatment with 1.3 µM ionomycin (Ion) and 10 nM phorbol myristate acetate (PMA) – both T cell activators—resulted in expression of both flPD-1 and ∆Ex3PD1 (Fig. 1B), with flPD1 being expressed more highly (Fig. 1C). In a T cell line (Jurkat cells) isoforms expressing both the full length and the short form were detected, and this expression was increased after treatment with ionomycin and PMA (Fig. 1D) with the balance of isoforms maintaining predominance of flPD-1 after treatment (Fig. 1E). Cloning and sequencing of the PCR products from PBMC and Jurkat cells confirmed that these products were flPD-1 containing the exon 2/3 junction and sPD1 containing the exon 2/4 junction (∆Ex3PD1 Fig. 1F).

To investigate if ∆Ex3PD1 could be an inhibitor of the PD-1/PD-L1 pathway the short isoform of PD-1 was cloned into an expression vector and transfected into HEK cells or Jurkat cells. Figure 1G shows expression of ∆Ex3PD-1 mRNA in both cell types. In HEK cells, which did not express endogenous ∆Ex3PD1, transfection with recombinant ∆Ex3PD1 expression plasmid (∆Ex3) resulted in high expression of ∆Ex3PD1. In Jurkat cells, expressing both isoforms, transfection of ∆Ex3 resulted in high mRNA expression (Fig. 1G) of ∆Ex3PD-1. Jurkat cells were then transfected with either a control plasmid (GFP), one containing the full-length PD-1 (flPD1), or a plasmid containing the ∆Ex3PD1 cDNA. Immunoblotting showed enhanced expression of PD-1 in the flPD1 transfected cell lysate but enhanced expression of PD-1 in the media of the ∆Ex3PD1 transfected cells, indicating that ∆Ex3PD1 cDNA but not flPD1 resulted in enhanced secretion of soluble PD-1 (Fig. 1H). This was confirmed using an ELISA for soluble PD-1 of the media, where transfection with flPD1 did not show enhanced amounts of soluble PD-1 in the media, whereas transfection with ∆Ex3PD1 cDNA did enhance the amount of soluble PD-1 in the media (Fig. 1I).

We then went on to determine whether this could inhibit T cell exhaustion induced by tumour cells. To determine whether ∆Ex3PD1 could reduce T cell exhaustion, we measured IL-2 mRNA in T cells expressing ∆Ex3PD1 after co-incubation of the T cells with IFN-γ-stimulated MM418 melanoma cells (expressing PD-L1, Fig. 1J). When seeded with Jurkat cells at a ratio of 6.4:1, co-incubation of Jurkat cells with these cells resulted in a significant inhibition of IL-2 production (Fig. 2A). Jurkat cells were then transfected with either a control vector expressing GFP or recombinant ∆Ex3PD1 plasmid (∆Ex3). When co-culturing the transfected Jurkat cells with MM418 cells the reduction in IL-2 production seen with the GFP transfected cells was significantly attenuated by transfection of ∆Ex3PD1 vector (Fig. 2B), indicating that ∆Ex3PD1 is capable of reducing T cell exhaustion by cancer cells. To determine whether ∆Ex3PD1 was able to enhance survival of T cells in the presence of cancer cells, we undertook flow cytometry with a live dead stain and CD3 staining for T cells in a Jurkat-cholangiocarcinoma co-culture (Fig. 2C). The proportion of live, activated (PMA and ionomycin treated) Jurkat cells was significantly decreased by the presence of the cholangiocarcinoma cell line KKU-M055 (Fig. 2D), and this was partially reversed when ∆Ex3PD1 was overexpressed. To determine whether ∆Ex3PD1 was able to enhance killing of cancer cells by T cells, KKU-M055 were co-incubated with activated Jurkat cells. This resulted in a visible killing of cholangiocarcinoma cells (Fig. 2E), which was enhanced by transfection of the Jurkat cells with the ∆Ex3 construct and confirmed by cell counts of the KKU-M055 (p < 0.001, Fig. 2F).

To determine the likely control of PD-1 splicing we ran the sequences around the splice sites for exon 3, 4 and 5 (Fig. 3A) through ESE finder, a bioinformatic tool that predicts possible splice factor recognition sequences. It was striking that the exon 3 5’ splice site was highly enriched for SRSF1 sequences compared with exon 4 and 5 3’splice sites (Fig. 3B). To determine whether SRSF1 could bind to PD-1 mRNA we immunoprecipitated SRSF1 from Jurkat cells and subjected the precipitate to RNA extraction and RT-PCR, whereas both isoforms were present in control cells and in the input RNA flPD1 was only found in the SRSF1-immunoprecipitated RNA, and no ∆Ex3PD1 or flPD1 RNA in the IgG control-precipitated RNA (Fig. 3C) indicating that SRSF1 was preferentially associated with the full length over the short PD-1 isoform RNA. A constitutively active SRSF1 construct has previously been used to determine the role of SRSF1 in splicing [11]. A plasmid encoding this phosphomimetic SRSF1 was transfected into Jurkat cells and expression of PD-1 was examined. Whereas control vector transfected cells (GFP) had expression of both isoforms, ∆Ex3PD1 expression was not seen in the phosphomimetic transfected cells (Fig. 3D). Quantification of the isoforms indicated that there was an increase in flPD1 and reduction in ∆Ex3PD1 after transfection with phosphomimetic SRSF1 (Fig. 3E) resulting in an increase in the ratio of flPD1 to ∆Ex3PD1 (Fig. 3F). To be able to identify the sequences required for flPD1 expression, we generated a minigene containing the coding region of exon1 fused to exon 2, intron 2 (266nt), exon 3, intron 3 (119nt), exon 4, intron 4 (649nt) and the coding sequence of exon 5 under control of the CMV promoter. When transfected into HEK cells this resulted in transcription of both isoforms (Fig. 3G). When transfected into Jurkat cells this resulted in over-expression of both isoforms at a ratio that was similar to endogenous expression (Fig. 3G). When co-transfected with the phosphomimic SRSF1 this increased splicing to the flPD1 (Fig. 3H). To determine key splicing regulatory sequences required for the generation of flPD1 a C > A mutation of the three cytosines in the polypyrimidine tract, was made (Fig. 3I). The C > A mutation had a small effect on splicing of PD-1 (Fig. 3J), enhancing the production of the ∆Ex3 isoform (Fig. 3K), but the phosphomimic SRSF1 did not increase splicing to flPD1 in the mutant minigene as it did in the wild type minigene (Fig. 3H) or the endogenous gene as shown in Figs. 3D–F, suggesting that the C–A mutation prevents phosphorylated SRSF1 from substantively enhancing exon 3 inclusion. These results are consistent with phosphoSRSF1 binding to the intronic region upstream of the splice site and acting as an intronic splicing enhancer.

SRSF1 is known to be phosphorylated by the splicing factor kinase SRPK1 [12]. We therefore investigated whether inhibition of SRPK1 could prevent splicing to flPD1 and result in a reversal of cancer cell-mediated T cell exhaustion. SRPK1 knockdown in Jurkat cells was achieved by lentiviral shRNAi transduction, resulting in a highly significant knockdown (Fig. 4A) and a switch in splicing from the flPD1 to ∆Ex3PD1 isoform (Fig. 4B, C). To determine whether this switch was amenable to pharmacological inhibition we used the SRPK1 selective inhibitor SPHINX31 [13]. Figure 4D shows that increasing concentrations of SPHINX31 resulted in increased ∆Ex3PD1 and a concentration-dependent increase in the ratio of ∆Ex3PD1:flPD1 (Fig. 4E). Nonlinear curve fitting to the normalised response showed an EC50 of 550 nM (95% CI 257–1065 nM, Fig. 4F). To determine whether this could result in inhibition of T cell exhaustion, Jurkat cells were co-cultured with MM418 melanoma cells in the presence or absence of pre-treatment with 1 µM SPHINX31 for 24 h. This resulted in a complete amelioration of the suppression of IL-2 production by activated Jurkat cells (Fig. 5A). To determine whether this was a generalised cancer response, we measured the effect on KKU-M055 cholangiocarcinoma cells and compared it with the effect of a PD-1 inhibitor pembrolizumab. KKU-M055 cells inhibited IL-2 mRNA expression in Jurkat cells when combined at a ratio of 6.4:1 cancer:T cells (Fig. 5B). This repression was dose dependently inhibited by pembrolizumab (Fig. 5B), and SPHINX31 (Fig. 5C). To determine whether this IL-2 response was sufficient to protect T cells, tumour cells and T cells were co-cultured with increasing concentrations of SPHINX31, after which cells were stained with live-dead and CD3 stains, and analysed by flow cytometry (Fig. 5D). Figure 5E shows that SPHINX31 partially reversed the reduction in live T cells induced by co-culture in a concentration-dependent manner with an EC50 of 2.7 µM (Fig. 5F).

Here we show that PD-1, the major target for immune checkpoint inhibition, has an endogenously expressed splice variant that acts as an endogenous inhibitor, and potentiates the immune response of T cells, and that this alternative splicing process is under control of the SR protein SRSF1 and can be switched by inhibition of its phosphorylation by SRPK1. Five different splice variants of PD-1 have been described, ∆Ex2,3,4, ∆Ex2,3 ∆Ex3 ∆Ex2 and flPD1, and the first three of these are likely to be soluble. However, only ∆Ex3 is likely to have the PD-L1 binding IgG domains. Our findings indicate that in stimulated T cells ∆Ex3 is a major isoform, although quantitatively the relative amounts remain to be determined, and RT-PCR can favour amplification of smaller isoforms, but it is clear that ∆Ex3 isoform can be a substantive component of the total PD-1 mRNA. Soluble PD-1 can be found circulating in plasma in both normal and cancer patients, but the origin of this circulating protein could come from either ectoprotein domain shedding (e.g. by proteolytic cleavage) [14] or alternative splicing. Blocking of PD-1/PD-L1 pathway using this ectodomain sPD1 has been shown to inhibit tumour growth and enhance immune response. The blockade of PD-L1 by this sPD1 also increases the transcription activities of IL-2 and IFN-γ genes [15]. Increased sPD1 levels in serum are associated with immune activation in chronic HBV infection suggesting sPD1 as a biomarker to immune activation and hepatocellular carcinoma development [16]. A study on patients with non-small cell lung cancer investigated the change in sPD1 concentration in the blood over the course of treatment with erlotinib, up until the development of clinical resistance, showed that patients with increased sPD1 levels during erlotinib treatment were associated with prolonged progression free survival and overall survival [17]. Another study on patients with cystic echinococcosis (CE, a chronic helminthic disease) showed an elevated concentration of sPD1 in the serum of CE patients as compared with the healthy control subjects [18]. The clinical relevance of sPD1 in rheumatoid arthritis (RA) patients is also indicated by significantly elevated plasma and synovial levels of sPD1, suggesting that it could be a useful marker for rheumatoid arthritis [19]. Similarly, elevated serum levels of sPD‐1 were observed in patients with systemic sclerosis and correlated with the extent of fibrosis and immunologic abnormalities [20]. However, current detection methodologies do not distinguish between proteolytically cleaved sPD1 and the ∆Ex3 splice variant, and the existing antibodies to sPD1 are targeted to the extracellular domain. However, in the ∆Ex3 splice variant the amino acids encoded by exons 4 and 5 are still present in the secreted protein, meaning that antibodies that specifically detect the exon 2-exon 4 splice junction could be generated to explore whether circulating sPD1 is from a splice or from a proteolytic cleavage event.

We showed here that exogenous expression of ∆Ex3PD1 was sufficient to partially reverse inhibition of IL-2 by melanoma and enhance killing of cholangiocarcinoma cells. Several studies have used recombinant soluble PD1 (rsPD1) to investigate the role of the extracellular domain of PD-1 in inhibiting ligand activation of the receptor. rsPD1 may functionally block the regulatory effect of flPD1 on T cells and may lead to alteration in T cell differentiation and regulation [21, 22]. rsPD1 can enhance T cell immunity [23] and rescue virus-specific CD4 and CD8 T cells proliferation during chronic infection [24, 25] and can facilitate anti-tumour immunity [15, 26]. This suggests a paracrine mechanism of action, although we have not shown that directly by using conditioned media, so these all support the concept that switching splicing to the ∆Ex3 splice variant could result in paracrine inhibition of PD-L1 and enhancement of T cell immunity.

Both the cancer cell types (melanoma [27] and cholangiocarcinoma [28]) used here have previously been shown to release PD-L1 to stimulate PD-1 and cause T cell exhaustion, but PD-L1 has also been shown to be released from tumour associated macrophages [29]. The mechanism of sPD1 action was assumed to be through binding to and preventing tumour cell PD-L1 from activating the endogenous PD-1 receptor, acting as a competitive inhibitor, but it could also function in vivo by preventing tumour associated macrophage-secreted PD-L1 or PD-L2. The reduction in IL-2 was noticeably less of an effect than the heightened killing of the tumour cells by the Jurkat cells. This could be due to synergy of activity between IL-2 and other cytotoxic cytokines (TNFa, IL-1, etc.), which we did not measure. Thus, the IL-2 output may be only a part of the cytotoxic effect enhanced by ∆Ex3PD1.

We then went on to show that PD-1 splicing was under control of SRSF1 binding and phosphorylation. SRSF1 has been proposed to act as either an exon splicing enhancer or intronic splicing repressor. The region of the RNA that mapped to SRSF1 was the start of the exon, and expression of phosphomimetic SRSF1 resulted in preferential inclusion of exon 3, indicating that SRSF1 was acting as an intronic splicing enhancer by binding in that region and so preventing exon skipping. It also indicates that the exon skip was enhanced by the mutation in the intron, again suggesting that in this circumstance SRSF1 might be acting as an intronic splicing enhancer. We did not investigate whether other RNA binding proteins could also interact, and this would need to be the subject of further study, so while it is clear that SRSF1 is involved, we cannot say that it is the only SR protein (or RNA binding protein) involved in splicing of PD-1.

Finally, we showed that control of SRSF1 phosphorylation by the kinase SRPK1 was able to enhance the splicing of the ∆Ex3 isoform. This is particularly interesting because it demonstrates that the balance of isoforms can be controlled. This contrasts with the results shown in Fig. 1 where exogenous ∆Ex3PD1 was able to slightly ameliorate the reduction in IL-2 production by Jurkat cells, as switching endogenous splicing reduces the amount of full-length PD-1 expressed by the same amount as increasing the ∆Ex3PD1 expressed, thus having a synergistic effect on inhibition of PD-L1 –there is more inhibitor and less receptor. Thus switching alternative splicing offers a potentially more attractive therapeutic approach than using monoclonal antibodies that can only act on the receptor, as it makes the T cell less able to respond, and produces its own endogenous inhibitor that can prevent PD-L1 from acting on other T cells as well. The range through which this can act has not yet been determined – by for instance investigating the effect of bulk media from ΔEx3PD1 expressing cells on co-cultured flPD1 expressing T cells with PDL1 expressing cancer cells.

It remains to be seen whether SRPK1 inhibitors or knockdown could be used in in vivo models—∆Ex3PD1 mRNA has not yet been described in mice, so more complex models might be necessary. The more muted effect of SRPK1 inhibition compared with knockdown may be due to drug efficacy, effect of protein expression rather than activation, or off target effects. (Although SPHINX31 is highly selective for SRPK1, it may also affect phosphorylation of other SR proteins.) Interestingly, SRPK1 inhibition also increased expression of a band between ∆Ex3PD1 and flPD1. This product has not been identified but could be another splice variant of PD-1. Although we have not shown here direct inhibition of SRSF1 phosphorylation by SPHINX31 in T cells, the concentrations of SPHINX31 used are consistent with previous studies showing inhibition of SRSF1 phosphorylation in epithelial cells [13]. We have also not shown directly that the effect of SRPK1 inhibition is a direct consequence of the switch in splicing of PD-1, for instance by showing that it is blocked by excess soluble PD-L1, but given the change in splicing of PD-1 induced by SRPK1 inhibition shown here, and the effect of the ∆Ex3PD1 isoform, the evidence points to this acting through a PD-1-mediated splicing pathway. The results shown here therefore suggest that SRPK1 could be a target in T cells in which PD-1 expression contributes to immune suppression in cancers.

Immune checkpoint inhibitors are a key breakthrough in cancer therapy and are currently used as antibodies inhibitnfi PD-1 or its ligand PD-L1. We show here that PD1 is alternatively spliced to form an antagonistic, soluble version ∆Ex3PD1 in human T cells, which is under control of the splicing factor SRSF1 and its phosphorylation by SRPK1. These results go on to demonstrate that SRPK1 inhibition switches splicing of PD-1 to generate the antagonistic isoform, which enhances T cell killing of tumour cells, opening up the possibility of small molecule SRPK1 inhibitors as novel pharmacological immunotherapies.