Targeting KRAS4A splicing through the RBM39/DCAF15 pathway inhibits cancer stem cells

The commonly mutated human KRAS oncogene encodes two distinct KRAS4A and KRAS4B proteins generated by differential splicing. We demonstrate here that coordinated regulation of both isoforms through control of splicing is essential for development of Krasmutant tumors. The minor KRAS4A isoform is enriched in cancer stem-like cells, where it responds to hypoxia, while the major KRAS4B is induced by ER stress. KRAS4A splicing is controlled by the DCAF15/RBM39 pathway, and deletion of KRAS4A or pharmacological inhibition of RBM39 using Indisulam leads to inhibition of cancer stem cells. Our data identify existing clinical drugs that target KRAS4A splicing, and suggest that levels of the minor KRAS4A isoform in human tumors can be a biomarker of sensitivity to some existing cancer therapeutics.

The KRAS gene is the most frequently mutated oncogene in human cancers, particularly in tumors of the pancreas, colon and lung, and has consequently been a major focus of cancer drug discovery for decades. However, in spite of these huge efforts, tumors carrying KRASmutations remain among the most difficult to treat, largely because of development of drug resistance due to tumor cell plasticity and/or acquisition of new mutations. A major advance in targeting a specific mutant KRAS G12C protein found in a subset of lung adenocarcinomas was recently described1, but methods for direct targeting of KRAS that would apply to most or all KRAS mutant cancers are presently lacking. In order to develop a more integrative view of the biology of KRAS signaling and identify fresh approaches to inhibiting the KRASpathway in cancers, we used a combination of mouse and human genetic analyses to investigate the functions and cell type-specific expression of the two known proteins produced by the KRAS locus, in normal tissues and during tumor development.

KRAS undergoes alternative splicing of the last exon to generate two isoforms, KRAS4A and KRAS4B, which are identical except for the 22/23 amino acids at the carboxyl terminus required for post-translational modifications and intracellular trafficking2. Germline deletion of the mouse Kras gene results in embryonic lethality3, but specific deletion of only Kras4A, achieved by deletion of exon 4A4 or knock-in of Kras4B cDNA (Kras4BKI mice, henceforth Kras4A−/−)5, had no effect on viability, suggesting that the main developmental functions of Kras are mediated through the Kras4B isoform. It has previously been demonstrated that mice lacking only the Kras4A isoform are resistant to chemically induced lung cancer6,7, in spite of the fact that Kras4A is expressed only in a subpopulation of normal and tumor cells. These data led to the proposal that Kras4A plays an essential role in tumor development, possibly through effects on a minor stem cell population6. Subsequent analysis of human cancer data from TCGA also suggested a more important role for KRAS4Ain human cancer than had previously been appreciated, as some tumors expressed unexpectedly high levels of this “minor” isoform8.

Here, we design a genetic approach to investigate the distinct functions of Kras4A and Kras4B in mouse models and human cancers. This analysis shows that each isoform is individually dispensable for normal mouse development, but that expression of both splice isoforms from the same Kras allele is required for initiation and growth of Kras mutant cancers. Disruption of splicing through the DCAF15/RBM39 pathway represents a novel route to target the KRAS pathway. Since KRAS4A is enriched particularly in a sub-population of cells with stem cell properties, our data suggest a model in which coordination of the stem-progenitor cell transition in cancers may be achieved through regulation of KRASmRNA splicing, identifying a potential vulnerability in KRAS mutant tumors.

Coordinated expression of Kras4A and Kras4B is essential for mouse lung tumorigenesis

In order to compare the roles of Kras4A and Kras4B in lung tumorigenesis, we used the previously published strategy5 to insert a Kras4A cDNA into the Kras locus, thus generating an allele that completely lacks expression of Kras4B (Kras4B−/− mice)(Supplementary Fig. 1A–C). Kras4B−/− homozygous mice were born at normal Mendelian ratios, but were infertile. Expression of Kras4A and Kras4B isoforms in normal tissues from these mice showed the expected patterns (Supplementary Fig. 1), and mouse embryonic fibroblasts (MEFs) isolated from E13.5 embryos showed robust EGF-induced activation of MapK and Akt, indicating that Kras4A and Kras4B are both activated in response to EGF in cells that lack the alternative isoform (Supplementary Fig. 2). Both knockout strains were crossed over 6 generations onto the FVB/N background, and injected with urethane. No tumors were found in treated Kras4A−/− animals, in agreement with previously published results6,7. However, this experiment showed that Kras4B−/− homozygous mice were also highly resistant to lung tumor development (Fig. 1A–D). Of the 18 Kras4B−/− homozygous animals treated with urethane, 8 developed one or two extremely small tumors each (Fig. 1A, B), and sequencing of Kras codons 12, 13, and 61 revealed no mutations. Thus, the effect of eliminating Kras4B, similar to loss of Kras4A, is to confer resistance to chemically-induced lung tumours with Kras mutations. Resistance is not due to changes in gene structure or absence of introns in the introduced cDNA construct, as mice carrying a similar cDNA encoding Hras rather than Kras isoforms developed lung cancers in which the cDNA construct carried the appropriate carcinogen-specific mutation6.

Urethane-induced lung tumor number (A) and average diameter (B) in WT, Kras4B- heterozygous and homozygous mice. Homozygous mice (n = 18) showed a highly significant reduction in tumor number and size, while heterozygous mice (n = 18) developed significantly more tumors than WTmice (n = 12). (C, D) Urethane-induced lung tumor number (C) and average diameter (D) in WT (n = 12), Kras4A- heterozygous (n = 12) and homozygous mice (n = 3) showed very similar patterns as seen in the Kras4B- cross. E Urethane-induced (left panel) and MMU-induced (right panel) lung tumor number in Kras4A+/− heterozygous (n = 7 in left panel and n = 5 in right panel) and double heterozygous Kras4A/4B mice (n = 3). No tumors were found in double heterozygous Kras4A/4Bmice. All lung tumorigenesis experiments were performed on a FVB/N background. n is the individual mouse number. The center line is the median, the bottom of the box is the 25th percentile boundary, the top of the box the 75th, and the whiskers define the bounds of the data that are not considered outliers, with outliers defined as greater/lesser than ± 1.5 x IQR, where IQR = inter quartile range.

Full size image

Any tumors that arose in either Kras4A+/− or Kras4B+/− heterozygous mice carried mutations in the fully functional wild type allele (Supplementary Table 1), suggesting that the coordinated expression of both isoforms from the same Kras allele may be essential for tumorigenesis. To test this possibility, we carried out carcinogenesis studies in the lung using double heterozygous Kras4A/4B mice expressing both isoforms, but in which splice regulation is disrupted. Specifically, 1 double heterozygote was treated with 5 doses of urethane, 2 with 3 doses of urethane, and 3 with 3 doses of the carcinogen N-methyl-N-nitrosourea (MNU) in an attempt to increase the mutation burden and likelihood of generating mutant forms of one or both isoforms. No tumors were detected in the lungs of any of these double heterozygous animals 30 weeks after treatment (Fig. 1E). The resistance of mice expressing both isoforms, but on different alleles, suggests that the initiating Krasmutation has to be in an endogenous Kras gene capable of generating mutant versions of both splice variants.

Both KRAS4A and KRAS4B contribute to the tumorigenicity of KRASmutant human cancers

To assess the functional relevance of the oncogenic KRAS isoforms in human cancer cell lines in vitro, we carried out shRNA-mediated knockdown of each isoform in KRAS mutant cancer cell lines A549, SUIT2, YAPC, and H358. We screened several hybridoma cell lines to identify an antibody (10C11E4) capable of detecting endogenous KRAS4A (see Methods) which is normally expressed at low levels in cultured cells. Specific knock-down by isoform-specific shRNA was confirmed by Westerns with the KRAS4A-specific antibody 10C11E4, and the KRAS4B-specific antibody (Santa Cruz Kras2b antibody) (Supplementary Fig. 3A). Knockdown of either KRAS4A or KRAS4B in SUIT2 cells significantly compromised their capability to form colonies in soft-agar assays (Supplementary Fig. 3B). Reduction of KRAS4A or KRAS4B in A549 and YAPC cells has minor effects on growth in vitro, but significantly inhibited growth in vivo in immunocompromised mice (Supplementary Fig. 3B, C), suggesting, in agreement with the mouse studies, that both KRAS4A and KRAS4B are required for growth.

We next used CRISPR/Cas9 technology to generate complete knockouts of KRAS4A or KRAS4B in human KRAS mutant cells. Single guide RNAs (sgRNAs) were designed to target the Cas9 nuclease to KRAS exons 4A or 4B in the human lung and pancreatic cancer cell lines A549 (G12S mutation) and SUIT2 (G12D mutation). Frameshift insertions and deletions in KRAS exon 4A or 4B were determined by allele-specific cloning and sequencing of selected clones (Supplementary Fig. 4). We were unable to isolate any homozygous KRAS4Bframeshift mutant clones, and noted a greatly reduced number of colonies proliferating beyond the first few cell divisions in this arm of the experiment. Complete loss of KRAS4A and reduced levels of KRAS4B were shown using isoform-specific antibodies (Fig. 2A). To determine the effects of KRAS isoform-specific knockouts on cell growth, cell proliferation on plastic and in soft agar were assessed. Homozygous knockout of KRAS4A and heterozygous knockout of KRAS4B significantly reduced growth in both assays (Fig. 2B–D). Subcutaneously injected knockout cells in nude mice also had impaired tumor forming capacity compared to the parental SUIT2 or A549 cells (Fig. 2E). Taken together, these data indicate that KRAS4A and KRAS4B are both required for optimal growth of KRAS mutant lung and pancreatic tumor cells in vitro and in immunodeficient mice, but the effects of Kras4Ainhibition on tumor growth in the context of an intact immune system remain to be established.

Splicing of KRAS4A is regulated by the RBM39 RNA-binding protein

Since our mouse genetic data suggested that controlled expression of both Kras4A and Kras4B isoforms from the same allele is necessary for tumor formation, we explored the possibility that small molecule inhibitors of different components of the splice site machinery may impact growth of KRAS mutant tumour cells9,10. We tested four inhibitors for effects on KRAS4A/B splicing: Pladienolide, which targets the splicing factor SF3B, disrupting the early stage of splice complex assembly11; Isoginkgetin, a bioflavonoid and general splicing inhibitor12; and two related sulfonamides Indisulam and Tasisulam, recently shown to cause proteasomal degradation of the RBM39 RNA binding protein and inhibition of RBM39-mediated splicing13,14. Isoginkgetin had no obvious effect on levels of KRAS4A or KRAS4B mRNA, while Pladienolide downregulated both KRAS4A and KRAS4B splice variants in SUIT2 cells (Fig. 3A, B) in agreement with its general role in disruption of the splicing machinery. Treatment with Indisulam or Tasisulam, specifically downregulated KRAS4A, but had no obvious effect on KRAS4B (Fig. 3A, B). Similar results were obtained using two additional cell lines A549 (Fig. 3C, D), and AsPC1 (Supplementary Fig. 5A, B), suggesting that RBM39 controls KRAS4A levels. The downregulation of KRAS4A, but not KRAS4B, was also confirmed at the protein level in SUIT2 and A549 cells (Fig. 3E, F and Supplementary Fig. 5C), and in AsPC1 cells (Supplementary Fig. 5D).

To test the involvement of RBM39 in control of KRAS splicing, we first verified that Indisulam treatment led to downregulation of RBM39 in SUIT2 and A549 cells (Fig. 3E, F). In all cases where KRAS4A expression was reduced at RNA and protein levels, the RNA binding protein RBM39 was also reduced (Fig. 3E, F and Supplementary Fig. 5D). This effect was seen in all three cell lines for both Indisulam and Tasisulam, which target RBM39 through interaction with DCAF1513. The general splice inhibitor Pladienolide again caused inhibition of both 4A and 4B isoforms (Fig. 3E, F).

We then generated RBM39 knockdown cells using CRISPRi/dCas9-KRAB15 using specific guide RNAs targeting the RBM39 gene in SUIT2 and A549 cells. Depletion of RBM39 caused a significant reduction in KRAS4A splice variant level but had no effect on KRAS4B (Fig. 3G), attesting to the specificity of RBM39 in control of KRAS pre-mRNA splicing. Although the RBM39 guide RNAs had a significant effect on RBM39 protein in A549 cells (Fig. 3G), the effect at the mRNA level was more modest (Supplementary Fig. 5E). It is possible that this reflects a feedback activation of RBM39 transcription when the protein is depleted, as Indisulam treatment surprisingly caused a significant upregulation in RBM39 mRNA while decreasing the protein levels (e.g. compare Fig. 3G and Supplementary Fig. 5E). Further studies of this apparent feedback control of RBM39, may shed light on the exact mechanisms involved.

Indisulam was originally proposed to act as an inhibitor of CA9 (carbonic anhydrase 9) which is involved in HIF1A regulation and hypoxia16. However CRISPRi/dCas9-KRAB downregulation of CA9 mRNA (Supplementary Fig. 5F) had no effect on levels of KRAS4A, or on the impact of Indisulam on KRAS4A levels (Fig. 3G). While RBM39 influences splicing of many pre-mRNAs genome-wide, these data support the proposal13 that RBM39 has a more limited set of physiological targets than would be expected for a general inhibitor of the splice machinery, and suggest that KRAS4A splicing may be controlled by RBM39. In support of this possibility, analysis of RBM39 binding sites genome-wide identified several PAR-CLIP tags within the KRAS gene17. The most significant of these tags was located within KRAS close to exon 4A (Supplementary Fig. 6), and therefore represents a candidate target site for RBM39 that may control KRAS4A splicing. However further detailed analysis by site-specific mutagenesis would be required to verify that RBM39 influences KRAS4A splicing directly rather than through an intermediate protein complex.

In order to determine whether Indisulam can mimic the effect of KRAS4A deletion on human tumor growth in immunocompromised mice, we injected SUIT2 and A549 cells into NSG immunodeficient mice and treated the animals carrying established tumors with Indisulam by retro-orbital injection (Fig. 3H). Both cell lines showed significant tumor growth inhibition by Indisulam in vivo, although the effect was more pronounced for SUIT2 cells (Fig. 3H). Others have also demonstrated a significant effect of Indisulam on growth of the KRAS mutant HCT116 human colon cancer cell line in vivo13.

Our data therefore identify an existing druggable pathway involving inhibition of RBM39that can target expression of the minor KRAS4A isoform. Interestingly, inspection of the DEPMAP portal shows that Indisulam activity in human tumor cell lines is highest in a subset of tumors derived from tissues such as the bone, haematopoietic system, soft tissues, and central nervous system. Ranking of these tumor cell lines according to the ratio of KRAS4Aand KRAS4B levels in fact shows that these tumor types have the lowest levels of this ratio (Fig. 3I, J and Supplementary Fig. 7A). To establish whether there is a significant effect of KRAS4A on Indisulam sensitivity, we obtained several of the cell lines characterized in DEPMAP and repeated cell growth assays in the presence of Indisulam, and also directly measured the levels of expression of the KRAS isoforms. We chose 2 blood cancer cell lines (Jurkat and HL60, wild-type KRAS), 2 lung cancer cell lines (NCI-H661 and NCI-H1703, wild-type KRAS), and 2 pancreatic cancer cell lines (YAPC and SW1990, mutant KRAS) for these studies. Supplementary Fig. 7B shows that blood cancer cells are more sensitive to Indisulam than lung NCI-H661 and pancreatic cancer cell lines, in agreement with the results from DEPMAP analysis. YAPC showed the highest KRAS4A level and HL60 cells have a lower KRAS4A/KRAS4B ratio than lung and pancreatic cancer cells (Supplementary Fig. 7C), in agreement with the results from the Cancer Cell Line Encyclopedia (CCLE). Treatment with Indisulam downregulated RBM39 and KRAS4A, but not KRAS4B, in Jurkat and HL60 cells (Supplementary Fig. 7D). Compared to the effect of Indisulam in SUIT2, downregulation of KRAS4A by Indisulam was seen at lower concentration in HL60 (Supplementary Fig. 7E). We also looked at the sensitivity of tumor cells with WT KRAS to treatment with Indisulam, and demonstrated that KRAS4A was downregulated in response to both Indisulam and knockdown of RBM39 in H1650 cells (Supplementary Fig. 7F). We therefore propose that the absolute level of KRAS4A, or the ratio of KRAS4A to KRAS4B, may be a biomarker of sensitivity to Indisulam or other drugs that interfere with KRAS4A splicing, but additional clinical studies will be necessary to confirm this possibility.

KRAS4A is enriched in cells with cancer stem cell properties

The above results showing that inhibition of KRAS4A can impact tumor growth in vivo was surprising, in view of the fact that KRAS4A levels are generally much lower than those of the major isoform KRAS4B, and Kras4A is only expressed in a subpopulation of cells in normal tissues or tumors6,18. We tested the hypothesis that KRAS4A may be preferentially expressed in a subpopulation of cancer stem cells19,20 using a well characterized assay for “side population” cells21 that have been shown to be enriched in stem cell properties. We isolated side population cells and analyzed sphere formation efficiency of side population and non-side population cells from 3 different cell lines (Supplementary Fig. 8A, B). Expression of a marker of side population cells, (ABCG2)22 was significantly elevated in side population cells from both A549 and SUIT2 cells, as well as from an additional pancreatic KRAS mutant cell line AsPC1 (Fig. 4A). Importantly, KRAS4A expression was also enriched in the same stem cell populations, as shown by specific TaqMan analysis of both KRAS splice isoforms (Fig. 4B). Homozygous knockout cells that no longer expressed KRAS4A had a reduced proportion of side population cells, which could be rescued by re-introduction of a functional KRAS4Aexpression construct (Fig. 4C). The activity of the side population stem cell marker aldehyde dehydrogenase (ALDH)23, was also reduced in the KRAS4A knockout cells, and restored by the re-introduced KRAS4A isoform (Fig. 4D). In contrast to these observations regarding KRAS4A, there was no significant difference in ALDH expression in cells with reduced levels of KRAS4B in the heterozygous knockout lines (Fig. 4D) indicating that expression of KRAS4A, but not KRAS4B is linked to cells with stem cell properties. In an attempt to increase the specificity for isolation of side population stem cells, we used additional reported markers of stem cells (CD166 and CD133) in combination with ALDH for FACS analysis. CD133 provided some additional enrichment in SUIT2 cells but not in A549 cells, while CD166 did not improve discrimination between side population and non-side population cells in either cell line (Supplementary Fig. 8C, D). Finally, in accordance with the reduction in KRAS4A by guide RNAs targeting RBM39 (Fig. 3), inhibition of RBM39 also reduced the proportion of ALDH positive cells in both SUIT2 and A549 cells (Supplementary Fig. 5G).

Stem cell side populations (SP) were assessed in A549, SUIT2 and AsPC1 cells by Hoechst 33342 staining. A ABCG2 levels were significantly increased in side population cells from all 3 cell lines. Data are presented as mean ± s.e.m from n = 3 independent experiments. *P < 0.05; **P < 0.01 by one-way ANOVA with Tukey’s multiple comparison. B KRAS4A, but not KRAS4B was significantly increased in levels by TaqMan analysis of side population cells. Data are presented as mean ± s.e.m from n = 3 independent experiments. *P < 0.05; **P < 0.01 by one-way ANOVA with Tukey’s multiple comparison. C, D Loss of KRAS4A reduces the proportion of side population cells (C) as well as the expression of the ALDH side population marker (D). Both the proportion of side population cells and ALDH expression levels were restored by transfection of a KRAS4A expression construct using fluorescence-activated cell sorting analysis. Data are presented as mean ± s.e.m from n = 7 independent experiments in left panel of (C) and n = 3 independent experiments in right panel of (C) and (D). *P < 0.05; **P < 0.01; ***P < 0.001 by Unpaired two-tailed t- test.

Full size image

KRAS4A and KRAS4B are induced by different types of cellular stress

The observation that both KRAS splice isoforms are required for tumor growth suggested that they may have different functions in cancer stem and progenitor cells. We therefore investigated the possibility that KRAS4A and KRAS4B may be responsive to different kinds of cellular stress. Treatment of parental A549, SUIT2, and AsPC1 cells with CoCl2 induced an increase in expression of HIF-1α, a known marker of hypoxia (Fig. 5A). KRAS4A, but not KRAS4B (Fig. 5B) showed a significant increase in expression as a consequence of this treatment, and the increase was predominantly in the side population cells rather than in the bulk cell population (Fig. 5C). Rapid tumor growth can lead to a hypoxic state which has been linked to stem cell activation and metabolic reprogramming24. In contrast, Endoplasmic Reticulum (ER) stress, which has been identified as a therapeutic target in KRAS mutant cancers25 caused a significant increase in the ER marker HSPA5 (also known as GRP78) and in KRAS4B, but not KRAS4A, in both A549 and SUIT2 cancer cell lines (Fig. 5D, E). Although both CoCl2 and tunicamycin may not be completely specific for hypoxia and ER stress respectively, these data emphasize the functional differences between the two isoforms in responding to different types of stress in the same human cancer cell lines.

Low levels of KRAS4A are associated with increased expression of cell cycle/mitotic markers

We next investigated the relative expression levels of the two KRAS splice isoforms in KRASmutant human lung cancer data sets26,27. We assessed the isoform-specific expression of KRAS in a cohort of 86 human lung adenocarcinomas26 using TaqMan probes designed to recognize KRAS4A or KRAS4B transcripts. Twenty three of these tumors harbor activating mutations in KRAS codons 12 or 13, while the remaining 63 are WT at KRAS codons 12, 13, and 61. TaqMan analysis revealed that both total KRAS and KRAS4A expression were significantly elevated in the KRAS mutant subset of tumors, while KRAS4B expression showed no significant difference (Fig. 6A–C). As gene expression microarray data were available for these samples, we separated them into two subgroups according to their KRAS4A/KRAS4Btranscript ratios (high (n = 8) or low (n = 7)). Supplementary Table 2 shows the top 100 genes that are highly expressed in samples with a low KRAS4A/KRAS4B transcript ratio. Surprisingly, in view of the data in Fig. 2 indicating that deletion of KRAS4A in human carcinoma cells in vitro causes some growth inhibition, GO enrichment analysis of genes highly expressed in primary lung carcinomas with low KRAS4A levels identified pathways linked to the cell cycle, microtubule organization, mitosis, and DNA damage responses (Supplementary Table 3). Gene set enrichment analysis confirmed enrichment for cell cycle genes in the low KRAS4A/KRAS4B samples (Fig. 6D). The reasons underlying this difference are unclear and may be due to differences in activation of proliferation and DNA damage checkpoint genes in vitro and in vivo, or to fundamental differences in growth control including the role of an active immune system in human patient samples. These questions will be the subject of future investigations.

A Total KRAS; B KRAS4A; and C KRAS4B expression in human lung cancers were determined by isoform-specific TaqMan assays. The number of individual lung cancer patients with wildtype KRAStumors (n = 63) and mutant KRAS tumors (n = 23) were analyzed. The center line is the median, the bottom of the box is the 25th percentile boundary, the top of the box the 75th, and the whiskers define the bounds of the data that are not considered outliers, with outliers defined as greater/lesser than ± 1.5 × IQR, where IQR = inter quartile range. (D–F) GSEA plots for functional gene sets enriched in low KRAS4A/KRAS4B ratio tumors in UCSF (D) and TCGA lung (E) and pancreas (F) data sets. Tumors with a low ratio of KRAS4A/KRAS4B have higher expression of genes linked to cell cycle and mitosis. Treatment with KG5 (G) or Rigosertib (H) significantly impaired growth of the KRAS4A knockout SUIT2 cells compared to the parental or KRAS4B knockdown cells. Cells were treated with inhibitors for 72 h and cell counts measured using the Cyquant assay. Data are presented as mean ± s.d from n = 7 independent experiments in (G) and n = 3 independent experiments in (H). ***P < 0.001; ****P < 0.0001 by two-way ANOVA with Dunnett’s multiple comparison.

Full size image

We also analyzed the isoform-specific expression levels of KRAS from TCGA lung adenocarcinoma (LUAD) samples. Although RNA was not available for isoform-specific TaqMan analysis, RNAseq reads spanning the alternatively spliced exon junctions of KRASwere used to estimate the percentage of total KRAS transcripts made up by KRAS4A (see Methods). Highly significant increases in both KRAS4A and KRAS4B transcripts were found in RNAseq data from KRAS mutant versus WT lung adenocarcinoma (LUAD) tumors, but KRAS4A showed a greater fold increase (1.7) than KRAS4B (1.2) (Supplementary Fig. 9A–C). This analysis confirmed that KRAS mutant lung tumors on average have a significantly higher ratio of KRAS4A to KRAS4B than KRAS WT tumors. Similar results were recently reported by Stephens et al28. To determine if splicing shifts towards higher KRAS4Aexpression in other cancer types, these analyses were extended to the TCGA pancreatic adenocarcinoma (PAAD) and colorectal adenocarcinoma (COADREAD) RNAseq datasets29. Significant increases of KRAS4A and KRAS4B expression in KRAS mutant tumors were observed, with greater fold increase in KRAS4A (Supplementary Fig. 9D–I). The percentage of total KRAS transcripts made up by KRAS4A showed similar increases in KRAS mutant PAAD and COADREAD, although significance was not reached, possibly due to the limited number of samples in these cohorts with known KRAS mutational status. Altogether, these data argue that the increase of KRAS expression observed in KRAS mutant cancers is driven not only by an overall increase in KRAS expression, but by altered splicing favoring an increase in the ratio of KRAS4A to KRAS4B transcripts. In order to verify the observation in Supplementary Table 2 and 3 showing enrichment of mitotic/cell cycle pathways in UCSF lung tumors with a low KRAS4A/KRAS4B ratio, we separated the available TCGA lung carcinoma samples into KRAS4A/KRAS4B high (n = 7) and low (n = 7) groups and pancreatic adenocarcinoma samples into KRAS4A/KRAS4B high (n = 9) and low (n = 9) groups and examined the total RNASeq data for differences in these gene expression pathways. Figure 6E, F and Supplementary Table 4 show the GO enrichment for pathways over-represented in the low KRAS4A/KRAS4B ratio subgroup, again demonstrating that cell cycle/mitotic gene transcripts are enriched in tumors with low KRAS4A mRNA levels.

We further verified the proposed relationship between side population cells, stemness, and KRAS4A levels using microarray data derived from gene expression analysis of human malignant pleural mesothelioma (GSE33734 dataset). In agreement with the observation that side population stem cells are mostly quiescent30, expression of a cell cycle marker (MCM7) was decreased, while genes associated with growth suppression in stem cells (GFI1and NDN) were increased in side population cells (Supplementary Fig. 10A–C)22,31,32,33. Importantly, a probe located within KRAS exon4A (ILMN_1652104) was also significantly increased in side population cells compared to non-side population cells (Supplementary Fig. 10D). ABCG2 expression was also enriched in side population cells (Supplementary Fig. 10E), in agreement with our data in Fig. 4A, B.

These data based on human tumors from TCGA suggested that low KRAS4A levels in vivo may be linked to increased sensitivity to some drugs that target the G2/M checkpoint during cell cycle progression. Since rigorous testing of this hypothesis would require new human clinical studies, we first examined the possible relationship between expression of KRAS4Aand altered sensitivity to some existing therapeutic drugs using in vitro cell culture conditions. Rigosertib34 and KG-535 target microtubules and are active in the G2/M cell cycle checkpoint, but also have other activities36,37. No differences were found in the effects of these drugs on growth of parental and KRAS4B knockdown cells, but KRAS4A knockout SUIT2 cells showed increased sensitivity to both compounds (Fig. 6G, H). These data support the proposal that inhibition of KRAS4A can increase sensitivity to a subset of known cancer therapeutics in vitro, but further in-depth screening using small molecule libraries or CRISPR/Cas9 approaches will be required to identify vulnerabilities that may be dependent on variations in KRAS4A/KRAS4B levels.

Essential functions for both KRAS4A and KRAS4B in cancer development

Targeting the KRAS pathway for cancer therapy has been particularly challenging, in spite of the vast amount of knowledge accumulated on the genetic alterations leading to pathway activation in tumors, and on the signaling networks that are driven by activated RAS oncoproteins. The lack of success in direct inhibition of mutant RAS proteins led to a focus on downstream RAS effectors, and to identification of many targeted drugs that are approved or presently in clinical trials38. A major problem encountered in all targeted approaches to cancer therapy including those targeting KRAS, is development of drug resistance, variously attributed to development of novel mutations that circumvent the effects of drug exposure, or to stem cell plasticity resulting in a new cell fate with loss of dependence on the original driver mutation39. The data presented here help us to address these major issues by taking an integrative approach to understanding the biological functions of both proteins produced by the KRAS locus, rather than focusing only on the more abundantly expressed isoform. A combination of mouse and human genetic approaches allowed us to identify a novel route to targeting of KRAS at the expression level by inhibition of KRAS4A splicing, and by demonstrating that the two splice variant proteins produced by the KRAS locus are primarily expressed in distinct stem and progenitor cells. This cell heterogeneity is a cardinal feature of KRAS mutant tumors19, 20, 38 contributing to the cell plasticity that impacts therapeutic responses and development of drug resistance.

The existence of two distinct isoforms of Kras has been known for many years, but almost all of the research on KRAS has focused on the KRAS4B isoform which is widely and abundantly expressed across a range of tissues. Both Kras4A and Kras4B isoforms are required for initiation and/or progression of carcinogenesis in the mouse lung, as tumors are only induced in animals that have at least one functional Kras allele capable of expressing both proteins. Double heterozygous Kras4A/4B mice were also extremely resistant, even under conditions where the carcinogen dose was significantly increased, suggesting that coordinated splicing to generate mutant versions of both isoforms from the same allele is required for tumorigenesis.

In the mouse, Kras4A is expressed during differentiation of pluripotent embryonic stem cells and in a subset of cells in adult tissues18, raising the possibility that Kras4A has specific functions in a small population of cells with stem cell properties6. In agreement with this possibility, human KRAS4A, but not KRAS4B, is enriched in stem cell-like side population cells derived from human cancer cell lines. Loss of KRAS4A, but not KRAS4B, causes a decrease in the proportion of cells with side population characteristics, as well as decreased activity of ALDH, a known marker of these cells. In addition, hypoxic conditions which are known to lead to reactivation of stem cells, cause an upregulation of KRAS4A, but not KRAS4B. We propose that the splicing of KRAS to generate the 4A and 4B isoforms may be a critical event in controlling stress responses and the proliferative or metabolic requirements in stem and progenitor cells (Fig. 7). Indeed, cancer stem and progenitor cells have previously been shown to exhibit differences in their growth and metabolic propserties20,40 and cancer stem cells have also been linked to development of drug resistance41.

The sulfonamide drug Indisulam has been in various cancer clinical trials for several years but its specific targets were unknown until the recent demonstration that it can induce degradation of the RBM39 RNA-binding protein, by tethering to DCAF1513,42. The mechanisms and targets downstream from RBM39 have been unclear, but we demonstrate here that an important consequence of RBM39 loss is inhibition of the minor KRAS4Aisoform. CRISPRi-mediated inhibition of RBM39 reduces KRAS4A levels, as well as the proportion of side population stem cells in human cancer cell lines. While RBM39 clearly has other splice targets in the transcriptome, these data link a druggable pathway with targeting of KRAS expression.

These results have several possible clinical applications. The levels of KRAS4A, and/or the KRAS4A/KRAS4B ratio, may be a biomarker for predicting responses to known cancer drugs. Evidence in favor of this possibility can be found in existing data sets, for example the DepMap and CCLE resources on cancer drug sensitivity across a large panel of human tumor cell lines. Levels of KRAS4A and the KRAS4A/KRAS4B ratio across this cell line panel are low in tumors derived from bone, neuronal, and haematopoietic/lymphoid tissues, but are significantly higher in epithelial tumors of the lung, pancreas, and intestine (Fig. 3). The haematopoietic, lymphoid, and neuronal tumor cell lines show the highest sensitivity to several cell cancer therapeutics including Rigosertib and Indisulam, although interpretation of these results is complicated by the off-target effects of these drugs, which have been reported to inhibit several targets including microtubules and cyclin-dependent kinases43. The present data, therefore, do not allow us to conclude that drug effects are due to low KRAS4A levels, G2/M checkpoint inhibition, or a combination of both. The relationship between KRAS4A levels, cell proliferation and DNA damage responses in vivo in the presence of an intact immune system also remains to be clarified.

Inhibition of KRAS4A through interference with the RBM39 splice machinery13, or directly by inhibition of splicing using oligonucleotide-based drugs44, may reveal new vulnerabilities to existing drugs, as suggested by the genetic experiments in Fig. 6. Indisulam-mediated inhibition of KRAS4A provides an important step in this direction, although this drug clearly has other targets and may have limited possibilities for combinatorial therapies. Further screening will be required to identify drug combinations that may optimally impact cancer stem-progenitor cell populations, leading to more effective treatments for KRAS mutant tumors.