Alternative splicing reverses the cell-intrinsic and cell-extrinsic prooncogenic potentials of YAP1

In addition to acting as a transcriptional coactivator, YAP1 directly mediates translocalization of the pro-oncogenic phosphatase SHP2 from the cytoplasm to nucleus. In the cytoplasm, SHP2 potentiates RAS-ERK signaling, which promotes cell proliferation and cell motility, while in the nucleus, it mediates gene regulation. As a result, elucidating the details of trafficking comprises


Introduction
YAP1 (and its paralogue TAZ) is a transcriptional coactivator that interacts with a number of sequence-specific transcription factors, most notably TEADs, and thereby transactivates downstream target genes, which stimulates cell proliferation while preventing cell death (1). Accordingly, YAP1 is considered to be a prooncogenic coactivator, deregulation of which promotes the development of cancer (2). YAP1 shuttles between the cytoplasm and nucleus in response to various signals including the Hippo signal, which plays an important role in restricting organ size by controlling cell proliferation and apoptosis (3). The Hippo pathway contains two core serine/threonine kinases, MST and LATS. When the Hippo signal is off, non-phosphorylated YAP1 enters the nucleus, where it acts as a transcriptional coactivator by binding with TEADs (4,5). When the Hippo signal is on, however, MST-activated LATS phosphorylates cytoplasmic YAP1 to prevent its nuclear translocalization (6). Thus, impaired Hippo signaling also contributes to neoplastic transformation of cells by deregulating YAP1 (7). The YAP1 gene comprises 9 exons and generates at least 8 differentially spliced YAP1 isoforms (YAP1-1α, YAP1-1β, YAP1-1γ, YAP1-1δ, YAP1-2α, YAP1-2β, YAP1-2γ, and YAP1-2δ) (Fig. 1A) (8,9). YAP1-1 and YAP1-2 isoforms are made by exclusion and inclusion, respectively, of exon 4, which encodes the second WW domain. Skipping of exon 6, which encodes the γ-segment comprising a 16-amino-acid sequence, gives rise to α isoforms, which are characterized by the presence of a functional leucine zipper motif that can mediate protein-protein interaction. Conversely, inclusion of exon 6 results in the generation of γ isoforms, which do not possess a leucine zipper due to insertion of the γ segment. Additionally, an alternative splice donor site in intron 5 makes a variant exon 5 transcript encoding a 4-amino-acid (VRPQ) extension, which also disrupts the leucine zipper heptad repeats. Thus, variant exon 5 generates β and δ isoforms in the absence and presence of exon 6, respectively (Fig. 1A) (8,9). YAP isoforms containing two WW domains exhibit stronger coactivator activity than do those having a single WW domain (10). Conversely, YAP1 isoforms lacking the canonical leucine zipper exhibit a dramatically reduced coactivator function (11). Of these YAP1 isoforms, YAP1-2α and YAP1-2γ are two well-studied YAP1 splicing isoforms that correspond to the shorter and longer forms of Yap1 (mouse ortholog of human YAP1), respectively (10,12). In contrast to YAP1, there has been no report showing the presence of multiple splicing isoforms of human TAZ. SHP2, encoded by the PTPN11 gene, is a ubiquitously expressed non-receptor-type tyrosine phosphatase.
Gain-of-function mutations in PTPN11 have been found in a variety of sporadic human malignancies (13). Deregulation of SHP2 by the Helicobacter pylori CagA oncoprotein also plays an important role in the development of gastric cancer (14). Thus, SHP2 is regarded as a prooncogenic phosphatase. In the cytoplasm, SHP2 potentiates the magnitude of RAS-ERK signaling (15,16) and, indeed, recent studies have shown that the growth of cancer cells carrying oncogenic KRAS mutations is still dependent on SHP2 activity (17). Like YAP1, SHP2 is specifically present in the cytoplasm at a high cell density but is abundantly distributed in the nucleus at a low cell density. The cell density-dependent subcellular localization of SHP2 is primarily regulated by the interaction of SHP2 with YAP1, in which YAP1 acts as a carrier and SHP2 serves as a cargo (18). It was also shown that Shp2 (mouse ortholog of human SHP2) interacts with the shorter form of Yap1 (hereinafter referred to as "Yap1-2α") but not with the longer form of Yap1 (hereinafter referred to as "Yap1-2γ"). Since these YAP1/Yap1 isoforms are generated by differential splicing, YAP1-SHP2 interaction is considered to be regulated by alternative splicing of YAP1 pre-mRNA (18), suggesting that differential splicing not only influences the coactivator activity but also affects the SHP2-binding capability of YAP1. In the present study, we focused on YAP1-2α and YAP1-2γ as representative isoforms that harbor an intact leucine zipper and a disrupted leucine zipper, respectively. Contrary to our expectations, cell-intrinsic and cell-extrinsic prooncogenic activities are reversed between the two YAP1 isoforms, caused by the difference in their SHP2-binding capacity. Furthermore, inclusion of exon 6 into YAP1 precursor-mRNA (pre-mRNA) is promoted by the SRSF3 splicing factor, the expression of which is suppressed by oncogenic KRAS. The results reveal a complicated interplay between differentially spliced YAP1 isoforms and SHP2 in the regulation of cell-autonomous and non-cell-autonomous YAP1 activities, perturbation of which contributes to the formation of a differential tumor microenvironment.
As previously reported, SHP2/Shp2 was localized to both the nucleus and cytoplasm at a low cell density (6.25x10 3 cells/cm 2 ), whereas it was excluded from the nucleus at a high cell density (6.25×10 4 cells/cm 2 ) (18). Since YAP1-SHP2 interaction was regulated by differential splicing of YAP1 pre-mRNA, nuclear translocalization of SHP2 was considered to be specifically mediated by YAP1 isoforms capable of binding with SHP2. To test this idea, constitutively active YAP1 mutants, YAP1-2α S127A and YAP1-2γ S127A , both of which are localized to the nucleus independently of cell density (21), were co-expressed in AGS human gastric cancer cells and in nontransformed NIH3T3 mouse fibroblasts in a high cell density condition. As a result, SHP2/Shp2 and YAP1-2α S127A were both localized in the nucleus ( Fig. 1H; white arrows). On the other hand, SHP2/Shp2 was localized to the cytoplasm in cells expressing YAP1-2γ S127A in the nucleus ( Fig. 1H; yellow arrows). Accordingly, subcellular localization of SHP2 was determined by the SHP2-binding activity of the YAP1 isoforms.
Both SHP2 and YAP1 are known to be prooncogenic, positively regulating cell proliferation as well as cell motility (1,15). We therefore compared the prooncogenic potentials of YAP1-2α and YAP1-2γ. To this end, we established two independent YAP1-knockout cell clones from AGS cells (YAP1 KO #1 and YAP1 KO #2 AGS cells) using two distinct YAP1-sgRNAs (Fig. S2A). Whereas the level of SHP2 in AGS cells was not significantly changed by YAP1 knockout, TAZ expression was slightly elevated in YAP1 KO cells as previously reported ( Fig.  S2B) (22). An expression vector for YAP1-2α or YAP1-2γ was then re-introduced into YAP1 KO #1 or YAP1 KO #2 AGS cells to establish AGS-derived stable transfectant clones that exclusively express one of the two YAP1 isoforms, referred to as YAP1 KO #1/1-2α AGS and YAP1 KO #1/1-2γ AGS cells ( Fig. 2A) or YAP1 KO #2/1-2α AGS and YAP1 KO #2/1-2γ AGS cells (Fig. S2C), at comparable levels. In YAP1 KO #1 AGS or YAP1 KO #1/1-2γ cells, SHP2 was localized to the cytoplasm at a low cell density due to the lack of interaction with YAP1 ( Fig. 2B; lower row). On the other hand, SHP2 was accumulated in the nucleus of YAP1 KO #1/1-2α AGS cells even at a low cell density because of the YAP1-SHP2 complex formation ( Fig. 2B; middle row). Consistent with the notion that YAP1 is pro-mitogenic, growth of YAP1 KO #1 or YAP1 KO #2 AGS cells was substantially retarded compared to that of parental AGS cells, and the retarded growth was restored by re-expressing YAP1-2α or YAP1-2γ in YAP1 KO# S2D). Since a luciferase reporter assay using AGS or NIH3T3 mouse fibroblasts revealed that YAP1-2α exhibits stronger transcriptional coactivator activity toward TEAD than YAP1-2γ does as previously described (Fig. S2E) (10,11), the difference in growth stimulation could not be explained by their differential coactivator activities. Because cytoplasmic SHP2 is critically involved in activation of the RAS-ERK signaling pathway (16, 23), a reduction in the level of cytoplasmic SHP2 by YAP1-2α was suspected to be involved in the diminished mitogenic activity of YAP1-2α compared to that of YAP1-2γ. In fact, the level of RAS-ERK signal activation, determined by the phosphorylated/active form of Erk (pErk), was reduced in YAP1 KO #1 or YAP1 KO #1/1-2α AGS cells but was elevated in YAP1 KO #1/1-2γ AGS cells compared to that in YAP1 KO #1 AGS cells (Fig. 2D). As in the case of YAP1, knockdown of Taz expression in NIH3T3 cells by specific siRNA gave rise to an increase in the level of cytoplasmic Shp2 at a low cell density (Fig. S2F), which was concomitantly associated with the increase in the active form of Erk (pErk) (Fig.  S2G). To substantiate this idea, wild-type SHP2 or an SHP2 mutant that cannot bind YAP1 (SHP2-PA) and thus is retained in the cytoplasm (18) was reintroduced into YAP1 KO #1/1-2α AGS or YAP1 KO #2/1-2α AGS cells. The cell growth curve revealed that ectopic expression of SHP2, especially SHP2-PA, gave rise to a substantial increase in cell growth (Figs. 2E and S2H). A wound-healing assay was also conducted to compare the effects of YAP1-2α and YAP1-2γ on cell motility (24,25). Both YAP1-2α expression and YAP1-2γ expression enhanced the motility of YAP1 KO #1 AGS cells and, again, the motogenic activity of YAP1-2γ was stronger than that of YAP1-2α (Fig. 2F). Next, to investigate the oncogenic potential of the YAP1 splicing isoforms in vitro, a colony formation assay was performed to examine if these YAP1 isoforms are capable of conferring anchorage-independent growth, the well-recognized neoplastic trait of cells in vitro (5,26), using non-transformed NIH3T3 fibroblasts. To do so, NIH3T3 cells were transduced with a lentivirus expressing YAP1-2α or YAP1-2γ and subjected to a soft agar assay. No colonies were developed by NIH3T3 cells and a few colonies with small sizes were formed by NIH3T3 cells expressing YAP1-2α. In contrast, numerous colonies with large sizes were formed by NIH3T3 cells expressing YAP1-2γ (Fig. 2G). The weak colony-forming activity of YAP1-2α was substantially potentiated by co-expressing SHP2-PA ( Fig.  S2I), indicating that cytoplasmic SHP2 is critically associated with the colony forming activity of YAP1. From these observations, we concluded that YAP1-2γ exhibited greater in vitro pro-oncogenic activities than YAP1-2α did, most likely being due to the differential SHP2-binding activity between YAP1-2α and YAP1-2γ.
To consolidate the in vitro observations that YAP1-2γ is more oncogenic than YAP1-2α in an in vivo situation, a tumor formation assay was performed in nude mice again using non-transformed NIH3T3 mouse fibroblasts. Parental NIH3T3 cells are non-tumorigenic in nude mice and have long been used to evaluate in vivo oncogenic potential of genes of interest, especially those constituting the RAS signaling pathway. To this end, 5 pairs of stable NIH3T3 transfectants (Pair-1 to Pair-5) that express comparable levels of YAP1-2α or YAP1-2γ, as determined by the HA-tag levels, were established ( Fig. 3A, upper). All of the mice injected with the pairs of NIH3T3 transformants developed palpable tumors, whereas injection of parental NIH3T3 cells did not result in tumor development, indicating that both YAP1-2α and YAP1-2γ conferred in vivo tumorigenicity upon NIH3T3 cells. In striking contrast to the results obtained in in vitro studies, however, the sizes of the tumors induced by YAP1-2α were much larger than those induced by YAP1-2γ in all 5 pairs (Fig. 3A, lower).
Histopathological examination revealed massive infiltration of F4/80-and CD11b-positive macrophages into YAP1-2γ-induced NIH3T3 tumors but not into YAP1-2α-induced NIH3T3 tumors (Fig.  3B). To examine if the macrophage infiltration was responsible for the reduced in vivo tumorigenicity, nude mice were pretreated with clodronate liposome, which specifically depletes macrophages in vivo (27). In mice treated with clodronate, YAP1-2γ-expressing NIH3T3 cells gave rise to tumors with sizes that were comparable to or even larger than those of tumors formed from YAP1-2α-expressing NIH3T3 cells (Fig.  3C, upper and lower left). Reduced macrophage infiltration into YAP1-2γ-induced NIH3T3 tumors in clodronate-treated mice was confirmed by immunohistochemical analysis (Fig. 3C, lower right). On the other hand, YAP1-2α-induced NIH3T3 tumors consistently grew in vivo without inducing macrophage infiltration regardless of clodronate treatment.

Mechanism
Underlying Macrophage Recruitment by NIH3T3 Tumor Cells Expressing YAP1-2γ In Vivo.
To elucidate the mechanism by which YAP1-2γ-expressing NIH3T3 cells recruited macrophages in vivo, total RNAs were extracted from two pairs of NIH3T3 transfectants (Pair-4 and Pair-5) and subjected to RNA-Seq analysis using Mouse Cancer Targeted RNA Panel (28) to identify genes differentially expressed between the cell pairs. Fig. 4A shows the top 4 genes, for which the relative mRNA levels were higher in YAP1-2γ-expressing NIH3T3 cells than in YAP1-2α-expressing NIH3T3 cells (Ccl2, Pdk4, Bbc3, and Nrp1) and top 4 genes, for which the relative mRNA levels were higher in YAP1-2α-expressing NIH3T3 cells than in YAP1-2γ-expressing NIH3T3 cells (Edn1, Ctgf, Cd5a2, and Wnt5a) by RNA-Seq analysis. Among those genes, the greatest difference in expression was observed in mRNA for C-C motif chemokine ligand 2 (CCL2 in humans, Ccl2 in mice), a major chemoattractant of monocytes/macrophages (29). In both Pair-4 and Pair-5, YAP1-2γ-expressing NIH3T3 cells gave rise to an approximately two-fold increase in Ccl2 mRNA compared to that in cells expressing YAP1-2α (Figs. 4A and S3A). CCL2/Ccl2 has been shown to be a direct transcriptional target of TEAD/Tead (30,31). As previously reported (11), YAP1-2α was more potent in activating TEAD/Tead reporter than YAP1-2γ was (Fig. S2E) and indeed YAP1-2α transactivated Cyr61 and Ctgf, the unique Tead target genes, more strongly than YAP1-2γ by qRT-PCR analysis (Fig. S3B). In striking contrast, YAP1-2γ was more potent than YAP1-2α in Ccl2 activation (Fig. 4B). Given this, we hypothesized that TEAD-mediated CCL2 induction was repressed by binding of YAP1 with SHP2. Indeed, disruption of the leucine zipper in YAP1-2α (YAP1-2α LZM ) converted YAP1-2α from a repressor to an activator of Ccl2 expression, indicating that SHP2 binding repressed the coactivator function of YAP1 (Fig. 4B). To corroborate the conclusion, Shp2-knockout NIH3T3 (Shp2 KO #1 NIH3T3) cells were generated by using the CRISPR-Cas9 system. Whereas Shp2 knockout did not alter the levels of Yap1 and Taz protein expression (Fig. S3C), it caused a marked increase in the level of Ccl2 mRNA level (Fig. 4C). Re-expression of SHP2 not only promoted cell growth (data not shown) but also restored the Ccl2 mRNA to the original level (Fig. 4C). While the phosphatase-dead SHP2 C459S mutant retained the ability to repress Ccl2 transactivation, SHP2-PA, which cannot bind YAP1, failed to do so (Fig. 4C). These results indicated that YAP1-2α-SHP2 complex formation but not SHP2 phosphatase activity is required for the repressor function of SHP2. The results provided compelling evidence for the transrepressional role of SHP2/Shp2 in CCL2/Ccl2 expression.
We next examined whether SRSF3/Srsf3 was actually involved in the regulation of YAP1/Yap1-mediated CCL2/Ccl2 expression by measuring Ccl2 mRNA levels in NIH3T3 or Srsf3 KO #1 NIH3T3 cells by qRT-PCR analysis. As a result, Srsf3-knockout NIH3T3 cells displayed a substantially reduced level of Ccl2 mRNA, which was restored and even further boosted by ectopic re-expression of SRSF3 (Note that human SRSF3 and mouse Srsf3 are 100% identical at the amino acid level.) (Fig. 5C). Furthermore, SRSF3 knockout did not substantially influence the growth potentials of YAP1 KO #1 AGS, YAP1 KO #1/1-2α AGS, and YAP1 KO #1/1-2γ AGS cells (Fig. S4F), indicating that SRSF3-mediated splicing regulation of YAP1 was responsible for the differential growth of these cells. Reciprocal regulation of CCL2 expression by YAP1-2α and YAP1-2γ was also shown at the protein level in AGS cells, although the effect was rather weak, possibly because intracellular CCL2 levels were measured by immunoblotting (Fig. S4G). Ectopic expression of SRSF3 also stimulated Ccl2 -1200 bp/+85 bp reporter activity in parental NIH3T3 cells but not in Yap1 KO #1 NIH3T3 cells (Fig. 5D). Collectively, these observations indicated that SRSF3/Srsf3 elevated CCL2/Ccl2 mRNA by promoting incorporation of exon 6 into YAP1/Yap1 mRNA, the product of which exclusively acts as a coactivator of TEADs because of the lack of SHP2/Shp2 binding activity.

Oncogenic KRAS Signal Influences YAP1 Exon 6 Splicing via SRSF3
Oncogenic KRAS has been shown to promote nuclear translocalization of SHP2 (34), raising the possibility that it enhances exon 6 skipping in YAP1/Yap1 pre-mRNA. To test this idea, NIH3T3 cells stably expressing moderate and comparable levels of oncogenic KRAS were established ( Fig.  6A and S5A). qRT-PCR analysis revealed that inclusion of exon 6 in Yap1 mRNA was markedly decreased in NIH3T3 cells upon expression of oncogenic KRAS (Fig. 6B). Consistently, expression of Srsf3 was also reduced in NIH3T3 cells stably expressing oncogenic KRAS (Fig. 6C). As a result, oncogenic KRAS promoted exon 6 skipping in YAP1 pre-mRNA by reducing the SRSF3/Srsf3 protein level. Since treatment with a MEK inhibitor or a proteasome inhibitor partially restored the oncogenic KRAS-mediated reduction in Srsf3 expression in NIH3T3 cells (Fig. S5A), REK/Erk signal activation might be involved at least partly in the reduction of Srsf3 (35).
KRAS is most frequently mutated in pancreatic and colorectal cancers (36). Kaplan-Meier analysis of the TCGA database (37,38) revealed that survival of patients with a low level of SRSF3 expression was inferior to that of patients with high SRSF3 expression for both cancers (Fig. S5B). The survival of patients with a low level of CCL2 expression was also inferior to that of patients with a high level of CCL2 expression for pancreatic cancer (Fig. S5C). A co-expression study showed a significant positive correlation between SRSF3 and CCL2 in colon cancer (Fig. S5D). Reciprocally, a negative correlation was observed between SHP2 expression and CCL2 expression (Fig. S5E). CCL2 expression levels were lower in pancreatic and colon cancer patients carrying KRAS mutations (Fig. 6D). These clinical data also support the notion of interplay between KRAS and SRSF3 in regulating YAP1 mRNA splicing that influences the prognosis of patients with cancers harboring KRAS mutations.

Discussion
YAP1-2α and YAP1-2γ are two representative isoforms of human YAP1 that are generated by alternative splicing of exon 6, which encodes the γ-segment (8). Inclusion of the γ-segment disrupts the leucine zipper motif in YAP1. Thus, YAP1-2α, lacking exon 6, possesses a functional leucine zipper, whereas YAP1-2γ, containing exon 6, does not. In the present study, we found that YAP1-2α interacts with SHP2 in a manner that is dependent on the functional leucine zipper. We also found that YAP1-2α forms a heterodimer with TAZ via functional leucine zippers. Taken together with the results of previous work (18,21), our results suggest that the proline-rich tail of SHP2 may interact with one of two WW domains in YAP1-2α only when YAP1-2α and TAZ are heterodimerized. A reduced YAP1-1α association with SHP2 compared to YAP1-2α is most likely because YAP1-1α has only a single WW domain. Upon complex formation, YAP1-2α/ΤΑΖ and SHP2 act as a carrier and a cargo, respectively, so that SHP2 is efficiently translocalized from the cytoplasm to the nucleus.
The present study revealed that YAP1-2γ stimulates cell proliferation and motility as well as anchorage-independent growth of non-transformed cells more strongly than YAP1-2α does. The difference is due to differential binding of the YAP1 isoforms with SHP2. Since YAP1-2γ does not bind SHP2, YAP1-2γ expression does not diminish cytoplasmic SHP2 below a certain threshold level and thereby potentiates RAS-ERK signaling that stimulates cell proliferation and cell motility. Consequently, YAP1-2γ confers greater cell-intrinsic prooncogenic potential upon non-transformed cells than YAP1-2α does. In contrast to these in vitro observations, cells expressing YAP1-2α are more potent in vivo tumorigenesis than are cells expressing YAP1-2γ. The inverted oncogenic potential between YAP1-2α and YAP1-2γ in in vivo is due to the difference in the magnitude of macrophage recruitment to the tumor tissue, which is explained by the finding that YAP1-2γ-predominant tumor cells secrete CCL2, the chemokine that potently recruits macrophages to the tumor site (39), whereas YAP1-2α-predominant tumor cells do not. Mechanistically, CCL2 is transactivated by the YAP1/TEAD complex (30,31), and we identified the TEAD-binding cis-element in the CCL2 promoter in the present study. YAP1-2γ/TEAD transactivates CCL2 by binding to the identified cis-element and constitutively transactivates CCL2 in YAP1-2γ-predominant tumor cells, allowing infiltration of tumor-inhibitory macrophages into tumors in vivo. On the other hand, YAP1-2α/TEAD is converted from a transactivator to a transrepressor upon binding of YAP1-2α with SHP2. The lack of CCL2 production/secretion in YAP1-2α-predominant tumor cells dampens macrophage accumulation in vivo. Our study therefore revealed that YAP1-2α and YAP1-2γ have opposing cell-extrinsic roles in establishing a tumor microenvironment. It is not unprecedented that cell-autonomous and non-cell-autonomous biological activities of a single molecule can lead to opposing outcomes in tumorigenesis. The Hippo signal kinase LATS acts as a tumor suppressor by preventing nuclear translocalization of YAP1. However, a recent study showed that LATS inhibits acquired tumor immunity in vivo (40). Also notably, whereas CCL2 recruits macrophages, polarization of infiltrated macrophages into a tumor-inhibitory (M1) or a tumor-stimulatory (M2) subtype is context-dependent, determined by combined effects of multiple humoral and cellular factors (41). In fact, several studies have shown that CCL2 recruits M2 macrophages rather than M1 macrophages (31).
Alternative splicing is mediated by the spliceosome, which is composed of small nuclear ribonucleic proteins (snRNPs) and non-snRNPs.
Non-snRNPs include serine/arginine-rich splicing factor (SRSF) family proteins, which bind to pre-mRNA and regulate the interaction between snRNPs and pre-mRNA, thereby stimulating or suppressing the recognition of splicing sites (42). SRSF3, also known as SRp20, is the smallest member of the SRSF family. It is variably expressed in many different cell types and has been shown to play a key role in the regulation of alternative exon skipping/inclusion in a variety of genes, including cancer-associated genes such as CD44, PKM, TP53, and RAC1 (43,44), and the present work adds YAP1 to this list. In the present study, SRSF3-knockout cells showed ~50% reduction in YAP1 mRNA isoforms containing exon 6, which was concomitantly associated with markedly reduced CCL2 mRNA expression, corroborating the role of the SRSF3/YAP1/SHP2 axis in the regulation of CCL2 production. SRSF3 is overexpressed in various cancers and has been considered to be prooncogenic (44). This notion is not inconsistent with the results of the present study showing that YAP1-2γ production, which is promoted by SRSF3, exhibits greater cell-intrinsic prooncogenic potential upon YAP1 by retaining SHP2 in the cytoplasm, which strengthens RAS-ERK signaling. Intriguingly, however, oncogenic KRAS was found to inhibit SRSF3 expression, a finding that is associated with the dominant expression of YAP1 isoforms lacking exon 6 over those containing exon 6 in cells expressing oncogenic KRAS (Fig. 6B). Given that innate immune stimulation is crucial for subsequent activation of acquired immunity, oncogenic KRAS may facilitate the creation of an immunologically "cold" microenvironment in which tumor cells do not stimulate innate immunity by inhibiting CCL2 production via YAP1-2α/SHP2 repressor complex. Consistently, SHP2 negatively regulates proinflammatory genes (45). Mice with intestinal epithelial-specific deletion of Shp2 develop severe colitis with elevated expression of proinflammatory cytokines (46). Hepatocyte-specific deletion of Shp2 induces hepatic inflammation, which predisposes to liver cancer (47). The present study therefore indicates that SHP2 influences inflammation/tumorigenesis through transcriptional regulation of cytokine/chemokine expression via splicing-dependent interaction with YAP1 (Fig. S6).
The relevance of using mouse cells for analyzing YAP1-SHP2 interaction and vice versa should be noted here. In the present study, we showed that YAP1-2α but not YAP1-2γ bound to Taz (Fig. S3D). We also showed that YAP1-2α but not YAP1-2γ bound to Shp2 (Fig. S3E). Furthermore, TAZ bound to YAP1-2α but not YAP1-2γ (Fig.  1D). In our previous study, we demonstrated that NIH3T3 cells predominantly express two endogenous Yap1 isoforms, the short-form (Yap1-2α) and long-form (Yap1-2γ), and that Shp2 selectively interacts with Yap1-2α (18). That study also showed that Yap1-2α interacted with SHP2, whereas Yap1-2γ (the long-form of Yap1) did not (18). Taken together, these observations indicated that the species difference (i.e., human vs. mouse) does not hamper the analysis of physical/functional interaction of YAP1/Yap1 isoforms with TAZ/Taz and SHP2/Shp2. The notion was also supported by the strong amino-acid sequence conservations in these proteins: YAP1 and Yap1 exhibit 88.7% identity and 92.7% similarity, TAZ and Taz exhibit 91.2% identity and 94.8% similarity, and SHP2 and Shp2 exhibit 99.5% identity and 99.8% similarity. Furthermore, the leucine zipper regions display 92.5% identity and 95.0% similarity between YAP1-2α and Yap1-2α, and the leucine zipper regions of TAZ and Taz are 100% identical. In addition, the γ-segments, which disrupt the leucine zipper, are 100% identical in YAP1 and Yap.
Although the present study has a limitation of experiments being primarily performed by overexpression of particular splicing isoforms of YAP1 (approximately 5-fold greater than that of the endogenous one), the results provide the first experimental evidence for the presence of context-dependent alterations in the function of differentially spliced isoforms, which are distinct from those made by simple loss-or gain-of-function splicing variants. The results therefore extend the importance of alternative splicing in further diversifying the biological functions of a single gene product. Notably, mutations in genes encoding splicing regulators in hematological malignancies indicate that abnormal splicing is also associated with oncogenesis (48). In the case of YAP1, the balance between splicing isoforms with and those without exon 6 is important for determining the direction and magnitude of cell-autonomous and non-cell-autonomous YAP1 activities that could act oppositely in terms of oncogenesis. The present work points to the notion that artificial manipulation of YAP1 pre-mRNA splicing will lead to the development of novel cancer therapeutics targeting a tumor microenvironment. In several experiments, cells were cultured at a low density (6.25×103 cells / cm2) or a high cell density (6.25×104 cells/cm 2 ).

Xenograft transplantation experiments
Female athymic nude mice (BALB/c; nu/nu) at 6 weeks of age were used for tumor formation assay. All mice were maintained in micro isolator cages. NIH3T3 cells prepared for injection were culture in log phase at the time of harvest and 1.5×10 6 cells were injected into both flanks of nude mice. The mice were maintained under aseptic barrier conditions until they were sacrificed at 6 weeks.
For macrophage depletion, Clodronate Liposome or control Liposome (Small, Hygieia Bioscience) was injected into nude mice (16.5 mg/kg) before tumor injection. Clodronate Liposome or control Liposome was injected once a week until mice were sacrificed. All the animal experiments were carried out according to the protocol approved by the Ethics Committee for Animal Experiments at the Graduate School of Medicine, The University of Tokyo.

Soft agar assay
NIH3T3 cells were used for soft-agar assay. Four × 10 4 NIH3T3 cells in 1.5 ml of DMEM containing 10% bovine serum and 0.3% Agarose-ME (Iwai Chemicals Company) were seeded onto the solidified bottom layer (2 ml of DMEM containing 10% bovine serum and 0.5% agarose) in a 6-well plate. 500 µl medium was added once a week. The cells were cultured for 4 weeks. The colonies were stained with Crystal violet (nacalai tesque) for 1 h at room temperature and photographed using microscope.

Establishment of the stable transfectants using lentivirus infection
Lentivirus vectors and packaging vectors (psPAX2, pMD2.G) were transfected into Lenti-XTM293T cells (TaKaRa). The supernatant containing the lentivirus was collected and concentrated following the PEG-itTM (SBI) protocol. Lentivirus titer (MOI; Multiplicity of Infection) was measured by following the protocol of Lenti-X qRT-PCR Titration Kit (TaKaRa). Cells were infected with lentivirus in culture medium containing 8 µg/ml hexadimethrine bromide; polybrene (SIGMA) for infection. At 48 h after virus infection, cells were selected with puromycin at a final concentration of 2.5 µl/ml.

Immunofluorescent staining and immunohistochemical staining
Cells were seeded on poly-D-lysine-coated glass cover-slips (IWAKI) and fixed with 4% paraformaldehyde. The cells were then permeabilized with 0.1% TritonX-100 for 10 min and followed by primary antibody treatment. Fluorescent images were obtained as non-compressed 16-bit TIF files by using a FLUOVIEW FV1200 (Olympus) Olympus FV1200 laser-scanning microscope systems. 40x/1.25 (silicone oil immersion) or 60x/1.35 (silicone oil immersion) UPlanSApo Objectives was used. Fluorescence was excited with 405-, 488-and 561-nm lasers. Images were acquired with 8.0 µs/pixel using FLUOVIEW viewer software (ver.4.1a, Olympus). The ImageJ were used for data analysis.
Tumor specimens were fixed in paraformaldehyde, embedded in paraffin, and sectioned. The sections were deparaffinized, rehydrated, antigen-retrieved, and then incubated with antibodies. Staining was performed using the Vectastain ABC-Elite kit (Vector Laboratories) according to the manufacturer's instructions. Samples were observed by Nikon Eclipse 55i (Nikon) microscopy. All the animal experiments were carried out according to the protocol approved by the Ethics Committee for Animal Experiments at the Graduate School of Medicine, The University of Tokyo.
The obtained chemiluminescence was exposed to X-ray film (Fuji Film) or quantified using LAS4000 system (Life Technologies).

Wound healing assay
Cells (6×10 5 cells / 3.5cm-dish) were seeded and incubated for 24 h at 37°C. The cellular layer in each plate was scratched using a plastic pipette tip (P20). The migration of the cells at the edge of the scratch was analyzed at 0 and 8 h when microscopic images of the cells were captured. The images were analyzed by ImageJ software.

RNA isolation and Quantitative PCR
Total RNAs were purified from cells or tumors using Trizol (Invitrogen). 1 µg RNA was used for reverse transcription with SuperScript II (Invitrogen). Quantitative PCR was performed using SYBR Premix Ex Taq II (TaKaRa) and OnePlus Real Time PCR system (Life Technologies).

Dual-luciferase assay
The promoter region of mouse Ccl2 (sequence NM_011333) covering 1.2 kb (position -1200 bp/+85 bp relative to the transcription initiation site, referred to as Ccl2 -1200 bp/+85 bp ) was cloned into pGL3 luciferase reporter vector (Promega AG). With the same approach Ccl2 -600 bp/+85 bp and Ccl2 -300 bp/+85 bp reporters were made. All constructs were verified by sequencing. Cells (1 x 10 5 per well) were plated in 12-well plates and transfected 24 h later using empty vector (pGL3) as a negative control. For normalization, cells were co-transfected with 10 ng of a renilla luciferase plasmid (pRL-TK). Twenty-four h post-transfection they were lysed and assayed for luciferase activity using the Dual Glo luciferase reporter system (Promega AG).

Transcriptome analysis on cancer-related genes using a next generation sequencer
Mouse Cancer Targeted RNA Panel profiles the expression of 395 cancer-associated genes (including pre-defined house-keeping genes as internal control) (Qiagen). Total RNAs were extracted using RNeasy Plus Micro. RNA concentration and quality were analyzed by RNA highly sensitive tape using Tape station 4200 (Agilent). Total RNA (400 ng) was used to construct cDNA library with Qiaseq mouse cancer transcriptome panel (RMM-003Z) and Qiaseq Targeted RNA 12 index I (Qiagen). 10 pM library was used for denaturing and subjected to next generation sequencing using Miseq (Illumina).

Statistical analysis
All samples represent biological replicates. Sample sizes are indicated in the figure legends. No statistical methods were used to predetermine sample size. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment. The variance similar between the groups has been tested before statistically compared. Unpaired two-tailed Student's t-tests were used for the statistical analysis of the data obtained by qRT-PCR and Luciferase assay. Paired two-tailed Student's t-tests were used for the analyses of growth curve, wound healing assay, soft-agar colony formation, in vivo tumor formation, and immunohistochemical staining data. Kaplan-Meier curves were generated using data downloaded from TCGA database [http://www.oncolnc.org]. Log-Rank test was used to compare the prognosis of high expression group and low expression group.

Data availability
All the data and methods necessary to reproduce this study are included in the manuscript and Supporting Information.  and mouse Yap1 gene. The second WW domain encoded by exon 4 (Yellow) presents in hYAP1-2 isoforms and mouse Yap isoforms. The extended transcript of exon 5 encodes additional 4 amino acids, termed β-segment (Purple). And the exon 6 encodes 16 amino acids, termed γ-segment (Green). The presence of these additional amino acids disrupts the leucine zipper motif (Blue). B, Co-precipitation analysis of HEK293T cells expressing Flag-SHP2 and a various splicing isoform of HA-YAP1. Cell lysates were subjected to immunoprecipitation with and an anti-Flag antibody. The total cell lysates (TCL) and the anti-Flag immunoprecipitates (IP) were subjected to immunoblot analysis with the indicated antibodies. The experiment was performed 3 times independently and the results were totally reproducible. C, Co-precipitation analysis of HEK293T cells co-expressing Flag-SHP2 and HA-YAP1-2α (WT) or HA-YAP1-2α LZM (LZM). D, Co-precipitation analysis of HEK293T cells co-transfected with the indicated expression plasmids. E, Co-precipitation analysis of HEK293T cells co-transfected with plasmids expressing Flag-YAP1-2α or Flag-YAP1-2α LZM and Myc-TAZ or Myc-TAZ LZM . F, Co-precipitation analysis of HEK293T cells co-transfected with plasmids expressing Flag-YAP1-2α or Flag-YAP1-2β and Myc-TAZ. G, Co-precipitation analysis of HEK293T cells co-transfected with plasmids expressing Flag-SHP2 and HA-YAP1-2α. TAZ knockdown was performed by transfecting TAZ-specific shRNA#1. H, (upper) AGS (left) and NIH3T3 (right) cells transiently transfected with an HA-YAP1-2α S127A or HA-YAP1-2γ S127A vector were cultured at high cell density and then stained with an anti-SHP2 antibody (Green), anti-HA antibody (YAP, Red) and DAPI (nuclei, Blue). The arrows indicate cells showing nuclear distribution of HA-YAP1-2α S127A or HA-YAP1-2γ S127A . Scale bars = 20 µm. (lower). The nuclear/cytoplasmic ratio of the SHP2 expression level, which was determined by measuring fluorescence intensity using imageJ, is plotted for each sample. The red bar represents the median value. n = 20, *p<0.05, ANOVA post Bonferroni.       indicate mean ± SD, n = 3, **p<0.01. n.s., not significant, Student's t-test. (right) RT-PCR analysis of mRNAs from AGS cells, SRSF3 KO #1 AGS cells, and SRSF3 KO #2 AGS cells. 5'-primer (targeting exon 5) and 3'-primer (targeting exon 7) detected the expected bands from the full-length YAP1-2α and YAP1-2γ cDNAs (lines 1 and 2). C, qRT-PCR analysis of Ccl2 mRNA expression in NIH3T3, Srsf3 KO #1 NIH3T3, and Srsf3 KO #1 NIH3T3 cells expressing ectopic SRSF3. "Empty" indicates wild-type NIH3T3 cells transfected with the CRISPR-Cas9 empty vector (px330). SD and average were calculated from technical triplicate. Error bars indicate mean ± SD, n = 3. **p<0.01, Student's t-test. (lower) The protein expression levels in these cells were shown by immunoblot analysis with the respective antibodies. D, Luciferase reporter assay of the Ccl2 promoter activity in wild-type NIH3T3 cells or Yap1 KO #1 NIH3T3 cells transfected with a Myc-SRSF3 vector or a control empty vector. SD and average were calculated from technical triplicates. Error bars indicate mean ± SD, n = 3, **p<0.01. n.s., not significant, Student's t-test. (lower) The protein expression levels were determined by immunoblot analysis with the respective antibodies. The protein levels expressed in these cells were determined by immunoblot analysis with the respective antibodies. The values shown below the second (Srsf3) row indicate relative densities of the Srsf3 bands. B, Relative ratio of exon 6-containing Yap1 mRNAs vs. total Yap1 mRNAs in NIH3T3 cells, KRAS G12D -expressing NIH3T3 cells, or KRAS G12V -expressing NIH3T3 cells was determined by qRT-PCR. SD and average were calculated from technical triplicates. Error bars indicate mean ± SD, n = 3. **p<0.01, Student's t-test. C, Relative expression of Srsf3 mRNA in NIH3T3 cells, KRAS G12D -expressing NIH3T3 cells, or KRAS G12V -expressing NIH3T3 cells was determined by qRT-PCR. SD and average were calculated from technical triplicates. Error bars indicate mean ± SD, n = 3. *p<0.05, Student's t-test. D, The expression level of CCL2 mRNA is lower in patients with KRAS mutations in both pancreatic cancer (left) and colon (right) cancer patients.