MARK2/MARK3 Kinases Are Catalytic Codependencies of YAP/TAZ in Human Cancer

The Hippo signaling pathway is a conserved regulator of cell identity and proliferation during metazoan development, with additional roles in tissue regeneration and cancer progression (1). In mammals, the core of the Hippo pathway includes kinases LATS1/2, which catalyze inhibitory phosphorylation of the YAP/TAZ transcriptional coactivators (2, 3). LATS1/2 activity is, in turn, activated by MST1/2 and MAP4K kinases and by the scaffolding protein NF2, which are themselves regulated by signals from the tissue microenvironment (4–10). Once released from LATS1/2-mediated inhibition, YAP/TAZ can enter the nucleus and bind to TEA domain (TEAD) transcription factors to activate a transcriptional program of cell proliferation and lineage plasticity (11–13).

YAP/TAZ and its upstream Hippo pathway are commonly dysregulated in human carcinomas and sarcomas to promote tumor development (14, 15). This can occur via genetic (e.g., YAP/TAZ amplifications; ref. 14) or nongenetic (e.g., perturbations of the extracellular matrix, metabolism, or cell polarity; refs. 16–20) mechanisms, with a consequence being that many human cancers possess a powerful dependency on the function of YAP/TAZ to sustain tumor growth. Because YAP/TAZ activity is dispensable for the homeostasis of several tissues (21–23), the aberrant functioning of this pathway has motivated efforts for developing drugs that interfere with YAP/TAZ functioning, such as small molecules that block the interaction between YAP/TAZ and TEAD proteins (24–27). However, a major obstacle in this effort has been in identifying “druggable” targets that allow for the restoration of Hippo-mediated tumor suppression in YAP/TAZ-dependent cancers.

MARK kinases (also known as PAR1) are conserved regulators of cellular polarity and microtubule dynamics. Genetic screens performed in Caenorhabditis elegans first implicated Par1 as essential for asymmetric cell divisions during early embryonic development through regulation of the mitotic spindle (28). Studies in several mammalian systems implicate MARK kinases (MARK1–4) as regulators of cell polarity through the phosphorylation of microtubule-associated proteins (29–31). In addition, evidence exists that MARK kinases regulate the output of other signaling pathways linked to cell proliferation, such as AMPK (32), Hippo (33), and MAPK (34). However, it is unclear from single-gene knockout studies whether any of the human MARK kinases perform vital functions that support cancer growth, an issue that might be confounded by the genetic redundancy among the four MARK kinases.

Here, we developed a CRISPR screening strategy to identify redundant paralog dependencies in cancer cell line models. Using this approach, we identified kinases MARK2/3 as essential for the growth of cancer cell line models harboring YAP/TAZ activation. Using genetic, transcriptomic, and biochemical assays, we show that MARK2/3 support YAP/TAZ function in cancer by inhibiting the Hippo signaling pathway. We define multiple substrates of MARK2/3 catalytic activity that account for their role in this pathway. Finally, we use an inducible peptide-based inhibitor of MARK2/3 to demonstrate the anticancer activity of targeting these kinases in vivo. Collectively, this work reveals MARK2/3 as druggable targets that allow for the upstream control of YAP/TAZ oncoproteins in human cancer.

Here, we developed a dual single-guide RNA (sgRNA) CRISPR vector system for performing double knockout screens of gene paralogs in search of redundant cancer cell dependencies (Fig. 1A). Using this system, we cloned a pooled library of 64,697 double-guide RNAs (dgRNA) designed to generate 1,719 single gene knockouts and 2,529 paralog double knockouts, focusing on genes involved in signal transduction and epigenetic regulation (Fig. 1A; Supplementary Tables S1 and S2). For each gene, we designed sgRNAs targeting exons that encode conserved protein domains to maximize the efficiency of generating loss-of-function alleles (35). Although prior studies have described CRISPR screening strategies for revealing epistatic gene interactions (36–42), we sought to apply our method to a larger cancer cell line panel to uncover context-specific dependencies that might have been overlooked previously. To this end, we performed negative-selection screens in 22 cancer cell lines grown under standard two-dimensional culture conditions, representing a diverse set of tumor lineages and genotypes (Supplementary Table S3). The performance of control sgRNAs within this library supported the accuracy of these screening datasets (Supplementary Fig. S1A–S1C). For each double knockout, we quantified the degree of genetic redundancy using the Gemini algorithm (43), which validated paralogs that are known to support cancer growth in a redundant manner, such as HDAC1/HDAC2, ESCO1/ESCO2, and EP300/CREBBP (Fig. 1B; Supplementary Tables S4–S6; refs. 44–46). By excluding pan-essential paralog pairs required for all cancer cell lines tested, we nominated the kinase paralogs MARK2 and MARK3 as outliers showing robust redundancy and cell line selectivity as cancer dependencies (Fig. 1B; Supplementary Fig. S1B and S1D). Although prior studies have identified functions for specific MARK kinases in cancer (47–49), the essential redundant functioning of MARK2/3 in human cancer cells has, to our knowledge, not been previously defined.

To validate these screening results, we performed arrayed format competition-based proliferation experiments in a panel of 31 cancer cell lines (Fig. 1C and D; Supplementary Fig. S1E; Supplementary Table S3). These assays validated the redundancy and essentiality of MARK2/3 in 19 cancer lines, whereas 12 cancer lines proliferated normally despite effective MARK2/3 double knockout, confirmed by Western blotting (Fig. 1E; Supplementary Fig. S1F). In these experiments, we noticed that MARK2/3 dependency was biased toward carcinomas and sarcomas, whereas most hematopoietic and neuroendocrine lineage cancers proliferated independently of MARK2/3 (Fig. 1C). Knockout of MARK2/3 led to a G₀/G₁ cell-cycle arrest and apoptosis in pancreatic (YAPC) and breast (MDA-MB-231) cancer lines, with a potency that resembled the effects of inactivating the mutant KRAS oncogene present in these models (Fig. 1F and G; Supplementary Fig. S1G–S1I). MARK2/3 knockout in YAPC xenografts led to robust tumor growth inhibition in vivo (Supplementary Fig. S2A and S2B). Expression of a CRISPR-resistant MARK2 or MARK3 cDNA alleviated cell fitness defect caused by the double knockout, indicating on-target effects (Fig. 1H; Supplementary Fig. S2C–S2F). Using this cDNA rescue assay, we found that mutational inactivation of kinase activity (MARK2^K82H) compromised cancer cell proliferation (Supplementary Fig. S2G). We further validated the importance of MARK2/3 catalytic function using a bump-and-hole strategy (50), in which the replacement of endogenous MARK2/3 with MARK2^M129G rendered the proliferation of YAPC cells sensitive to the bulky kinase inhibitor 1-NM-PP1 (Fig. 1I; Supplementary Fig. S2G). Collectively, these experiments validated MARK2/3 as catalytic dependencies in specific carcinoma and sarcoma cell line models.

We next sought to understand why MARK2/3 are essential in some cancer contexts but dispensable in others. The top mutational correlate of MARK2/3 dependency in our cell line panel was of KRAS (Supplementary Fig. S3A and S3B). However, our validation experiments failed to establish a mechanistic link between MARK2/3 and RAS (Supplementary Fig. S3C and S3D). Using transcriptome analysis, we found that MARK2/3 essentiality across the 31 cancer lines was correlated with the expression of YAP and TAZ and with the expression of canonical YAP/TAZ target genes MYOF, CYR61, DKK1, and CAV1 (Fig. 2A and B; refs. 51–53). Using dual sgRNA vectors, we confirmed that YAP and TAZ function redundantly as dependencies in this cell line panel in a manner that closely correlated with MARK2/3 essentiality (Fig. 2B and C; Supplementary Fig. S3E and S3F). This observation led us to hypothesize that MARK2/3 are critical for maintaining YAP/TAZ function in diverse human cancer contexts. In support of this, we found that the inactivation of MARK2/3 led to reduced expression of a fluorescence-based TEAD:YAP/TAZ reporter in MDA-MB-231 cells (Fig. 2D; ref. 19). In addition, RNA sequencing (RNA-seq) analysis performed in 20 different cancer cell line models after MARK2/3 knockout demonstrated reduced expression of a YAP/TAZ transcriptional signature in MARK2/3-dependent lines (Fig. 2E–G; Supplementary Table S7). We extended this analysis by performing genome-wide profiling of active chromatin (H3K27 acetylation), which revealed that MARK2/3 and YAP/TAZ are each critical to activate TEAD4:YAP−bound enhancer elements (Fig. 2H and I; Supplementary Fig. S3G–S3I). Together, these results suggest that MARK2/3 are required to maintain the essential function of YAP/TAZ in human cancer.

Upon inactivating MARK2/3, we observed a striking increase in LATS1/2 T1079/T1041 phosphorylation (Fig. 3A and B; Supplementary Fig. S4A). This activation mark is known to be catalyzed redundantly by MST1/2 and MAP4K kinases, an activity that is further enhanced by NF2 (Fig. 3A; ref. 4). Knockout of MARK2/3 triggered reduced nuclear levels of YAP/TAZ, which is an expected outcome of strengthening LATS1/2 function (Fig. 3C; Supplementary Fig. S4B and S4C). Although prior studies have shown that MARK2/3 inhibit the function of MST1/2 (33, 49, 54), we reasoned that this substrate would be insufficient to account for MARK2/3 dependency in cancer because MST1/2 function redundantly with MAP4Ks to regulate YAP/TAZ in human cells (Fig. 3A; refs. 4, 6). This prompted us to perform a broader exploration of MARK2/3 substrates in the Hippo pathway using a chemical–genetic strategy (Fig. 3D; ref. 55). Our approach exploited gatekeeper substitutions of MARK2 (M129G) and MARK3 (M132G), which can accommodate bulky ATP-γ-S analogs (e.g., 6-Fu-ATP-γ-S). We coexpressed MARK2^M129G or MARK3^M132G with 18 different epitope-tagged Hippo pathway components in HEK293T cells, followed by treatment with 6-Fu-ATP-γ-S and immunoprecipitation–Western blotting with a phosphothioester–specific antibody. This approach validated the known ability of MARK2/3 to phosphorylate CDC25C and MST1/2, in accord with prior findings (Fig. 3E; Supplementary Fig. S4D and S4E; refs. 49, 56). In addition, we identified NF2, YAP, and, to a lesser extent, TAZ as MARK2/3 substrates in this system (Fig. 3E; Supplementary Fig. S4D and S4E). We did not detect MARK2/3-dependent phosphorylation of LATS1/2 but did identify robust phosphorylation of several MAP4K kinases (Fig. 3E; Supplementary Fig. S4D and S4E). To map the exact sites of phosphorylation, we performed in vitro kinase assays with purified MARK2 and each substrate, followed by mass spectrometric peptide quantification (Supplementary Fig. S4F and S4G). In these assays, MARK2 catalyzed phosphorylation on serine or threonine residues of NF2 (four sites), YAP (five sites), and TAZ (four sites; Fig. 3F–H; Supplementary Fig. S5A–S5K; Supplementary Table S8). By introducing alanine substitutions of these phosphosites into cDNA constructs, we confirmed the importance of these specific serine/threonine residues for MARK2-dependent phosphorylation in human cells (Supplementary Fig. S6A–S6E). Using mass spectrometry (MS) analysis, we also identified sites of MARK2-dependent phosphorylation on MAP4K proteins and MST1/2 (Supplementary Fig. S6F); however, the known redundancy among these kinases (4) led us to prioritize NF2 and YAP/TAZ for further functional investigation.

Two of the sites of MARK2/3-dependent phosphorylation on NF2 were T230 and S315, which have been reported to inhibit NF2 function (57). To further evaluate this, we used a transfection-based assay in HEK293T cells (4, 58), in which NF2 overexpression stimulates p-LATS1/2. We found that the coexpression of wild-type MARK2/3, but not a catalytically dead mutant, negated NF2-stimulated LATS1/2 phosphorylation (Fig. 4A; Supplementary Fig. S6G). In addition, a phosphomimetic allele of NF2, in which all four sites of MARK2-dependent phosphorylation are substituted with aspartate, was incapable of triggering LATS1/2 phosphorylation (Fig. 4B; Supplementary Fig. S6H and S6I). We also found that MARK2 expression was able to disrupt the physical interaction between NF2 and MAP4K kinases (Supplementary Fig. S6J) and block MAP4K4/6- and MST1/2-dependent LATS1 phosphorylation (Supplementary Fig. S6K–S6O; ref. 4). Together, our findings suggest that MARK2/3 can indirectly suppress LATS1/2 activity by phosphorylating upstream components of the Hippo pathway.

We next evaluated the functional importance of YAP/TAZ phosphorylation by MARK2/3. LATS1/2 can sequester YAP/TAZ in the cytoplasm by installing phosphorylation that is recognized by 14-3-3 proteins (59). Owing to the adjacent locations of several MARK2/3 and LATS1/2 substrates on YAP/TAZ (Fig. 3G and H; refs. 60, 61), we hypothesized that MARK2/3-dependent phosphorylation might release YAP/TAZ from 14-3-3–mediated inhibition. To evaluate this, we reconstituted LATS1/2-dependent YAP/TAZ phosphorylation using purified proteins (Fig. 4C; Supplementary Fig. S6P), which was sufficient to trigger interactions with recombinant 14-3-3ε (Fig. 4D and E). However, preincubation of recombinant YAP or TAZ with MARK2 or MARK3 and ATP eliminated the formation of 14-3-3ε complexes despite the presence of LATS1/2-dependent phosphorylation (Fig. 4D and E; Supplementary Fig. S6Q and S6R). In accord with these in vitro findings, the expression of a phosphomimetic allele of YAP or TAZ, in which serine or threonine substrates of MARK2/3 are mutated to aspartic acid, eliminated the 14-3-3ε interaction in cellular lysates (Fig. 4F and G; Supplementary Fig. S7A). Consistent with these findings, this phosphomimetic allele of YAP is present in the nucleus and is more active than the wild-type protein (Supplementary Fig. S7B–S7F). Collectively, these functional experiments support that MARK2/3-dependent phosphorylation of YAP/TAZ can interfere with LATS1/2-dependent formation of 14-3-3 complexes.

The biochemical findings above prompted us to perform epistasis experiments evaluating whether dual regulation of NF2 and YAP/TAZ underlies the essential function of MARK2/3 in cancer identified in our paralog screen. As expected, we found that the pharmacological inhibition or double knockout of MST1/2, or its adapter SAV1, failed to alleviate MARK2/3 dependency (Fig. 5A–C; Supplementary Fig. S8A–S8D). By contrast, inhibition or double knockout of LATS1/2 resulted in a bypass of MARK2/3 essentiality in four different cancer cell line models (Fig. 5A–C; Supplementary Fig. S8C and S8D). In these same models, we found that NF2 knockout or expression of a phosphomimetic allele of YAP (YAP^5D) partially alleviated MARK2/3 dependency (Fig. 5D and E; Supplementary Fig. S8E). Moreover, combining the NF2^KO/YAP^5D genetic alterations led to a nearly complete bypass of MARK2/3 dependency in these contexts, which resembles the effects of inactivating LATS1/2 (Fig. 5E). Unexpectedly, the knockout of NF2 in MDA-MB-231 cells, a breast cancer cell line harboring a biallelic truncating mutation of NF2, alleviated MARK2/3 dependency (Supplementary Fig. S9A–S9C). However, we found that the truncated NF2 protein present in this cell line (amino acids 1–231) is expressed and retains partial functionality (Supplementary Fig. S9D and S9E). Collectively, these results suggest that an essential function of MARK2/3 in cancer is to regulate NF2 and YAP/TAZ, which allows for potent indirect control over the output of LATS1/2.

The Hippo pathway activity is known to be modulated by cell culture conditions (19), which motivated us to validate MARK2/3 dependency in tumor models with more physiological extracellular environments. Because selective small-molecule inhibitors of MARK kinases are not available, we developed a catalytic inhibitor of MARK kinase activity that could be expressed in an inducible manner in various tumor models. The EPIYA-repeat region of the CagA protein of Helicobacter pylori was reported to potently and selectively inhibit MARK kinase activity by competing with substrate binding (62, 63), a peptide we refer to here as MARK kinase inhibitor (MKI; Fig. 6A). We observed that lentiviral expression of MKI, but not an MKI peptide harboring point mutations that abrogate MARK binding (63), reduced the nuclear levels of YAP/TAZ, reduced NF2/YAP/TAZ phosphorylation, and suppressed the expression of YAP/TAZ and MARK2/3 transcriptional signatures (Fig. 6B–E; Supplementary Fig. S10A–S10E). In addition, the proliferation arrest induced by MKI correlated with the overall sensitivity to MARK2/3 double knockout in a cell line panel and the sensitivity to MKI could be alleviated by overexpressing MARK2 (Fig. 6C; Supplementary Fig. S10F and S10G). Our epistasis experiments further indicated that engineering of NF2^KO/YAP^5D alleviated the sensitivity to MKI-mediated growth (Fig. 6F), thus validating MKI as a tool catalytic inhibitor that mimics the biological effects of MARK2/3 double knockout when expressed in cancer cells.

We next engineered a vector that expresses MKI under the control of a doxycycline (dox)-inducible promoter, which was introduced into a panel of YAP- or TAZ-amplified human triple-negative breast cancer or pancreatic cancer organoid cultures (Fig. 6G). We next introduced the dox-inducible MKI constructs into T3M4 cells, a basal-like model of pancreatic cancer, followed by orthotopic transplantation (Fig. 6H–J). Upon tumors reaching >200 mm³, tumor-bearing mice were treated with dox or vehicle control to induce MKI expression (Fig. 6H). Using two different imaging modalities, we found that MKI expression led to a marked inhibition of tumor growth in vivo (Fig. 6K and L; Supplementary Fig. S10H). Similar results were observed in a subcutaneous YAPC xenograft model (Supplementary Fig. S10I and S10J). The findings validate the potent antitumor effects of catalytic MARK2/3 inhibition in YAP/TAZ-dependent cancer models.

It has been observed that human cancers can be broadly classified based on the status of YAP/TAZ (64). YAP/TAZ^OFF tumors tend to be of hematopoietic or neural/neuroendocrine lineages, and in this context, transcriptional silencing of YAP/TAZ is required for tumor development (64–66). By contrast, YAP/TAZ is activated in human carcinomas and sarcomas, which is essential for tumorigenesis (64, 67). This binary classification has important clinical implications, as YAP/TAZ have powerful effects on several tumor cell characteristics, including epigenetic plasticity and drug sensitivity (15, 68). Here, we have exploited the ON versus OFF status of this pathway in human cancer cell line models to reveal a strict requirement for MARK2/3 catalytic activity to support YAP/TAZ function across a diverse array of human carcinomas and sarcomas. Targeting of MARK2/3 leads to potent inhibition of YAP/TAZ and a severe compromise of tumor cell fitness; phenotypes that can be accounted for by phosphorylation of NF2 and YAP as direct MARK2/3 substrates. Our study positions MARK2/3 as dominant regulators of the human Hippo pathway and as a “druggable” target in YAP/TAZ-dependent tumors.

Early genetic studies in model organisms implicated the MARK1–4 ortholog Par1 as a regulator of cell polarity (28, 69). Subsequent work in Drosophila identified Par1 as a negative regulator of the Hippo pathway, which influences cell growth phenotypes (33). Despite this early observation, the connection between MARKs and Hippo in human cells has been controversial, with some studies suggesting MARKs can activate (33, 49, 54) or inhibit (48, 70) YAP/TAZ function. Because these prior studies focused on the manipulation of individual MARK kinase genes, we suspect that genetic redundancy between MARK2 and MARK3 concealed the powerful inhibitory influence of human MARK kinases over the Hippo pathway. Although our findings are generally consistent with earlier work in Drosophila (33), the mechanism by which MARK/Par1 regulates YAP/TAZ might be distinct in each organism, with an expansion of upstream and downstream substrates of MARK2/3 in human cells that allow for multilevel control over the output of LATS1/2. Nevertheless, our collective work suggests an ancient linkage between MARK and Hippo during metazoan evolution, which may have emerged to integrate cellular polarity with organ growth and regeneration. Although our experiments highlight an important role of NF2 and YAP/TAZ as MARK2/3 substrates, it is likely that other substrates in the Hippo pathway contribute to this regulation as well (e.g., MAP4Ks and MST1/2; refs. 33, 49, 54).

Prior studies have described small molecules that block the interaction between YAP/TAZ and TEAD transcription factors (24–27, 71), which are currently the most developed therapeutic strategy for targeting Hippo-dysregulated cancers (72). Although the efficacy of such an approach in human patients has only recently begun to be evaluated in clinical trials (clinicaltrials.gov; NCT04665206 and NCT05228015), our work reveals chemical inhibition of MARK2/3 kinase activity as an alternative strategy for eliminating YAP/TAZ-addicted tumor cells. As kinases, there may exist unique opportunities to develop potent and selective MARK2/3 inhibitors by leveraging decades of experience in the pharmaceutical industry in targeting this class of enzymes (73), which would differ from the challenges of modulating a protein–protein interaction (74, 75). In addition, by functioning upstream to regulate LATS1/2-mediated control over YAP/TAZ, targeting of MARK2/3 would likely select for distinct resistance mechanisms from drugs targeting TEAD:YAP/TAZ interaction (76). Although the liabilities of each targeting strategy await further description in preclinical models and ongoing clinical studies, our study justifies consideration of MARK2/3 as a prominent cancer target in a diverse collection of human carcinomas and sarcomas harboring hyperactive YAP/TAZ function.

One clear clinical opportunity for YAP/TAZ modulation in cancer is in the setting of RAS-mutant cancers. Preclinical studies have identified YAP/TAZ hyperactivation as a common strategy used by tumor cells to evade the anticancer effects of targeted therapy (27, 77–80), a finding well supported by experiments performed in animal models (77, 81, 82). Thus, an opportunity may exist to combine inhibitors of MARK2/3 and RAS to prolong the survival of patients with cancer and prevent acquired drug resistance.

All cell lines were authenticated using short tandem repeat profiling. The HPAF-II, AsPC1, PANC1, MIA PaCa2, NCI-H1299, A549, NCI-H23, RD, MDA-MB-231, NCI-H1048, NCI-H211, NCI-H209, NCI-H1836, NCI-H1436, NCI-H2023, NCI-H1975, NCI-H1703, CHL1, OCI-AML3, THP1, HEK293T, and K562 were purchased from the ATCC. The YAPC, PATU8902, PATU8988T, NOMO1, HEL, SET2, RH30, OCI-AML3, and MOLM13 cell lines were purchased from the “Deutsche Sammlung von Mikroorganismen und Zellkulturen.” The KP2, T3M4, SUIT2, and KLM1 cell lines were purchased from the “Japanese Collection of Research Bioresources Cell Bank.” The COR-L311 cell line was purchased from the “European Collection of Authenticated Cell Cultures.”

All human cell lines were grown in RPMI medium supplemented with 10% FBS and 1% penicillin/streptomycin (Gibco), if not otherwise indicated. HEK293T and MDA-MB-231 cells were grown in DMEM. NCI-H209, NCI-H1836, NCI-H1436, and NCI-H1048 were grown in HITES medium [DMEM supplemented with 5% FBS, 1% penicillin/streptomycin, insulin–transferrin–selenium (Gibco), 10 nmol/L hydrocortisone (Sigma-Aldrich), 10 nmol/L β-estradiol (Sigma-Aldrich), 10 mmol/L 4-(2-hydroxyethyl)piperazine-1-ethane-sulfonic acid (HEPES) (Gibco), and 2 mmol/L L-glutamine (Gibco)]. All lentiviral packaging with HEK293T cells and cancer cell line transduction were performed following standard procedures similar to those previously described (35). For organoid culture transduction, single cells were infected using a spin infection strategy (800 g for 2–4 hours), before virus removal and replating in Matrigel (Corning). All organoids were grown in growth factor–reduced Matrigel. Human patient–derived pancreas and breast cancer organoids were cultured in specific organoid media as described before (83, 84). All cell lines were cultured at 37°C with 5% CO₂ and were periodically tested negative for Mycoplasma contamination. All experiments were performed within 20 passages of thawing of cells.

Cells were lysed directly with 2× Laemmli Sample Buffer (Bio-Rad), supplemented with β-mercaptoethanol (Sigma-Aldrich) or in RIPA buffer supplemented with protease inhibitor cocktail (Roche) and Halt Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific). The same total protein amounts or extracts from the same number of cells were loaded into each lane of SDS-PAGE gel (NuPAGE 4%–12% Bis-Tris Protein gels, Thermo Fisher Scientific) followed by transfer to a nitrocellulose membrane. Membranes were blocked using 5% nonfat dry milk and washed using Tris-buffered saline with 0.1% Tween 20 (TBST) after incubation in both primary and secondary antibodies. After, membranes are developed with chemiluminescent horseradish peroxidase (HRP) substrate (Pierce).

Antibodies used in this study are rabbit HRP-conjugated secondary antibodies (Cytivia, NA934-1ML, 1:5,000–1:20,000), HRP-conjugated β-actin (Sigma-Aldrich, A3854, 1:5,000), HA (Roche, 3F10, 1:10,000), FLAG (Sigma-Aldrich, A8592, 1:5,000), V5 (Invitrogen, R961-25, 1:5,000), myc (Abcam, ab62928, 1:3,000), GAPDH (Cell Signaling Technology, D16H11, 1:3,000), H3 (Cell Signaling Technology, D1H2, 1:5,000), GST-tag (Cell Signaling Technology, 91G1, 1:3,000), and primary antibodies MARK2 (Abcam, ab133724, 1:1,000), MARK3 (Abcam, ab264285, 1:1,000), YAP (Cell Signaling Technology, D8H1X, 1:1,000), p-YAP/TAZ (S127/S89; Cell Signaling Technology, D9W2I, 4911, 1:3,000), TAZ (Cell Signaling Technology, E8E9G, D3I6D, 1:1,000), NF2 (Cell Signaling Technology, D1D8, 1:1,000), MST1 (Cell Signaling Technology, 3682T, 1:1,000), MST2 (Cell Signaling Technology, 3952T, 1:1,000), LATS1/2 (GeneTex, GTX87014, 1:1,000), Phospho-LATS1/2 (T1079/T1041; Cell Signaling Technology, D57D3, Abcam, ab305029, 1:1,000–1:3,000), SAV1 (Cell Signaling Technology, D6M6X, 1:1,000), CDC25C (Cell Signaling Technology, 5H9, 1:1,000), Phospho-CDC25C (S216; Cell Signaling Technology, 63F9, 1:1,000), Thiophosphate ester (Abcam, ab92570, 1:5,000–1:20,000), 14-3-3 (Cell Signaling Technology, 8312, 1:1,000), Erk (Cell Signaling Technology, 137F5, 1:1,000), Phospho-Erk (Cell Signaling Technology, D13.14.4E, 1:1,000). IHC: GFP (Abcam, ab6673, 1:500). Immunofluorescence: anti-HA (Cell Signaling Technology, C29F4, 1:400) and anti–α-Tubulin−FITC (Sigma-Aldrich, F2168, 1:400).

Tumor tissue was fixed in paraformaldehyde (PFA) and embedded in paraffin. Paraffin was removed, and the tissue was dehydrated after which antigen was retrieved by boiling in Tris (10 mmol/L, pH: 9.0)/EDTA (1 mmol/L) buffer. Tissue was blocked with 5% BSA-TBST and incubated with GFP antibody (1:500) overnight, washed with TBST, and stained with HRP-conjugated secondary antibody (1:400) before development with 3,3′-diaminobenzidine and counterstained using hematoxylin.

Six days after double knockout A549 (25,000 cells/well) were seeded on poly-L-lysine–coated four-well µ-slide (ibidi). Then, after 24 hours, it was fixed (10% neutralized formalin) for 10 minutes. The cells were permeabilized with 0.5% Triton X-100/PBS for 10 minutes and washed two times with wash buffer (0.1% BSA/1× PBS). After, cells were incubated with blocking buffer [10% normal goat serum (Invitrogen) + 0.3% Triton X-100] for 45 minutes. Primary antibodies were diluted with dilution buffer (1× PBS, 1% BSA, 1% normal goat serum, 0.3% Triton X-100, and 0.01% sodium azide) and added after blocking. Slides were incubated at 4°C overnight. Cells were washed two times with wash buffer and probed with Alexa Fluor 647–conjugated secondary antibody [goat anti–rabbit IgG (H+L) highly cross-adsorbed secondary antibody, Alexa Fluor 647, 1:1,000, Invitrogen, A21245] for 1 hour at room temperature. Cells were washed two times with wash buffer, and nuclear counterstained by adding mounting media with DAPI (ibidi). For starvation assay, 25 k per well of transfected HEK293T cells were seeded on Collagen-I–coated culture slide (Corning). After 24 hours of incubation with serum-free media, the cells were treated with FBS-containing media for 2 hours. Images were collected with 40×/1.10 HC PL APO water immersion objective lens using Leica TCS SP8 laser scanning confocal microscope and were processed with Leica LAS X software. For three-dimensional (3D) quantification, z-stacks of the specimen were acquired with a slice interval of 0.5 µm. Three z-stacks were taken in three biological replicates each and were analyzed by Imaris software (10.1.1, Oxford Instruments). Nuclear and cytoplasmic volumes and mean YAP intensity were calculated by the Surfaces function in Imaris. Then, 3D reconstructed surfaces were created around blue (DAPI) or red (YAP) signals, and the thresholding cutoff was optimized on the basis of the fluorescence intensity of each fluorescence channel. After 3D surface reconstructions were optimized, we imported DAPI and YAP surfaces to the Imaris Cell Biologists module to compartmentalize the YAP signal to the associated cytoplasm or nuclei. To ensure consistency in the intensity of data measurements within and between, consistent imaging and analysis parameters were applied to all the images.

For apoptosis analysis, cancer cells transduced with sgRNA constructs were stained using conjugated annexin-V proteins (Thermo Fisher Scientific) and DAPI according to the manufacturer's instructions. In brief, 6 days after infection with lentivirus containing dgRNAs linked to GFP, the cells were detached and resuspended in staining buffer followed by incubation with annexin-V and DAPI. The stained cells were analyzed by flow cytometry, and data analysis was performed with FlowJo software. Early apoptotic (annexin-V⁺/DAPI⁻), late apoptotic (annexin-V⁺/DAPI⁺), necrotic (annexin-V⁻/DAPI⁺), and viable cells (annexin-V⁻/DAPI⁻) were identified.

For cell-cycle analysis, cancer cells transduced with dgRNA constructs (day 5) were treated with 10 µmol/L 5-Ethynyl-2′-deoxyuridine (EdU) 4 hours prior to sampling. EdU incorporated into cells was stained according to the manufacturer's instructions (Thermo Fisher Scientific). In brief, cells were detached and fixed in 4% PFA, permeabilized, and EdU conjugated using click chemistry. Stained cells were analyzed by flow cytometry, and data analysis was performed with FlowJo software. Cells were identified on the basis of EdU signal and DNA content (DAPI).

The paralog cotargeting CRISPR library was optimized for the use of SpCas9, a system we recently published (85). Oligonucleotide pools (n = 64,697), dgRNAs targeting 1,719 single gene, and 2,529 gene combinations were synthesized (Twist Bioscience) with BsmBI cutting sites in between overhung sequences for the dual CRISPR RNA (crRNA) fragment. Primers matching the overhang for the lentiviral backbone were used to amplify the oligonucleotide pools. PCR products were purified and cloned using Gibson Assembly Master Mix (New England Biolabs) into LRG3.0, a lentiviral vector with human U6 and bovine U6 promoters expressing the two sgRNAs in inverse orientation. To incorporate the dual Trans-activating CRISPR RNA (tracrRNA), the purified tracrRNA fragment was cloned in between the dual crRNAs by a second round of Gibson assembly.

To generate stable cell lines, cells were first transduced with a Cas9 vector (Addgene, 108100). Next, cell lines were transduced with the paralog cotargeting CRISPR library virus aiming for a representation of 1,000 cells per sgRNA at a low multiplicity of infection (∼0.3). Briefly, the cell lines were transduced by spin infection for 45 minutes at 600g. On day 3, an initial sample was taken, and cells were replated maintaining representation. Once 10 cell doublings were reached, samples for genomic DNA extraction were again taken.

Cells lysed in extraction buffer [10 mmol/L Tris, 150 mmol/L NaCl, 10 mmol/L EDTA, Proteinase K (0.02 mg/mL), and SDS (0.1%)]. Lysates were incubated at 56°C for 48 hours and genomic DNA was extracted using Tris-saturated phenol (Thermo Fisher Scientific).

DNA was PCR-amplified and barcoded with P5/P7 primers (Integrated DNA Technologies) using AmpliTaq Gold DNA polymerase (Thermo Fisher Scientific) according to the manufacturer’s instructions. Briefly, Taq polymerase, reaction buffer, magnesium chloride, primers, and 1 µg of genomic DNA were mixed and used for each reaction (round 1: PCR for 11 cycles). Amplified DNA was size selected (200–300 bp) and barcoded in a second round PCR using stacked P5/P7 primers (round 2: PCR for nine cycles). The PCR product was sequenced using a paired-end 75-bp reads protocol (Illumina).

Reads were counted by mapping dgRNA sequences to the reference file of the library, and a pseudo count of 16 was added. To systematically calculate the degree of phenotypic synergy or epistasis between two paralog knockouts, we used Gemini, a variational Bayesian approach to identify genetic interactions in CRISPR screening data (43). The Gemini R (v.1.4.0) package was used to calculate log₂(fold changes; LFC), and synergy scores and statistics with their corresponding P and FDR values (Supplementary Tables S2 and S4–S6). Gemini calculates the LFC of dgRNA abundance between the initial time point (average abundance of dgRNAs, day 3, n = 10) and the 10 doubling-time endpoint. Gemini has been used to compute the synergy score by comparing the LFCs of each gene pair with the most lethal individual gene of the pair. Nonsynergistic pairs were used to calculate FDR and P values. Bayesian analysis and the prior choice were performed as described previously (43).

Paralog pairs were identified by aligning human proteome (>100,000 amino acid sequences) using the Basic Local Alignment Search Tool. Matches originating from the same gene were removed. Each top-scored paralog-pair identified (E value < 0.01), which shared the same functional domain of interest, was included in the Paralog library. In addition, high-scoring paralogs (E value < 10⁻¹⁰⁰) were included. Functional domains were mapped using Reverse Position-Specific Basic Local Alignment Search Tool and the conserved domain database (86).

Domain annotation and sgRNA positions were compared, and sgRNAs cutting in functional domain regions were included in the sgRNA selection pool. sgRNAs with off-targets in paralog genes were removed from the selection pool. Additionally, sgRNAs incompatible with the cloning strategy were removed from the selection pool. sgRNAs were picked based on their off-target score (calculated on the basis of the number of off-target locations in the human genome factored by the fall-off in cutting-efficiency of SpCas9 in case of crRNA sequence mismatch). For each gene, three to four selective domain-focused sgRNA were picked and combined. A set of sgRNAs targeting known essential genes as positive controls (n = 28) and a set of nontargeting (n = 97), as well as noncoding, regions as negative controls (n = 54) were included in the library. To construct cell line–specific negative controls (nonsynergistic pairs), we selected genes that were not expressed in a cell line according to the RNA-seq data [log₂(TPM + 1) < 0.1, in which TPM is transcripts per million].

For validation, two sgRNAs were synthesized together with bovine U6 promoter as gene blocks (Integrated DNA Technologies) and cloned using Gibson assembly into LRG2.1T (Addgene, 65656). All inserts were verified by Sanger sequencing (Eurofins Genomics). To generate LATS1/2 and MST1/2 double knockout pools, two sgRNAs cotargeting LATS1/2 or MST1/2 were combined and two sgRNAs targeting SAV1 and NF2 were combined on one vector. For lentivirus packaging, HEK293T cells were transfected with sgRNA, pVSV-G, and psPAX2 (Addgene, 138479, 12260) plasmids using polyethyleneimine (PEI) reagent (PEI 25000). Percent GFP⁺ populations were followed over time after infection using the Guava easyCyte HT flow cytometer instrument (Millipore). Complete sgRNA sequences are given in Supplementary Table S9.

All cDNAs were either cloned from Addgene plasmids or synthesized as indicated below. CRISPR-resistant cDNAs were generated either by mutating the protospacer adjacent motif (PAM) sequence or sgRNA binding sites into synonymous codons. All cDNAs were cloned into lentiviral constructs derived from LentiV_Cas9_puro (Addgene, 108100), altered to contain internal ribosome entry site elements and selection marker resistance genes. For doxycycline induction of cDNA expression, genes were cloned into Doxi-LentiV (derived from Addgene, 80921, 89180, and 71782 vectors), and expression was induced using 2 µg/mL doxycycline. MARK1, MARK2, MARK3, and MARK4 (Addgene, 170582, 23404, and 23716 vectors; OriGene: SC107258) were cloned into the LentiV internal ribosome entry site vector after the addition of a FLAG-tag at the N terminus. Hippo pathway genes LATS1, LATS2, NF2, SAV1, TAZ, MOB1A, MOB1B, MST1, MST2, TEAD1, YAP, YAP mutants, TAZ mutants, MST1 mutants, GFP, CDC25C, and YWHAE (14-3-3ε) encoding V5, HA, or myc-tagged cDNAs were from Addgene (66851, 66852, 32834, 32836, and 32839) or synthesized (Integrated DNA Technologies). cDNA encoding for MAP4K1, MAPK4K2, MAPK4K3, MAPK4K4, MAPK4K5, and MAPK4K6 were from Addgene (23484, 23644, 23664, 23486, 23611, and 23522), which were 3xHA-tagged and cloned into LentiV. The MAPK4K7, MAP4K4^10A, and MAP4K4^10D expression vectors were generated by VectorBuilder. All mutations were introduced by gBlock synthesis or PCR. MKI^WT was derived from the coding sequence of CagA (H. pylori strain 26695). The sequence containing the EPIYA-repeat regions amino acid positions 885 to 1,105 was codon optimized. cDNA was synthesized and cloned into LentiVi-P2A-GFP (derived from LentiV_Cas9_puro) or Doxi-LentiV after the addition of a 3xHA or FLAG-tag at the N terminus. To generate a mutant of MKI with impaired MARK binding capacity (MKI^MUT) the leucine-109/143 in the two MARK binding motifs of MKI^WT were mutated to glycine.

To generate a TEAD-driven GFP reporter, the promoter of the established TEAD binding reporter (8xGTIIC; ref. 19; Addgene, 34615) was fused into a construct containing destabilized GFP (Addgene, 138152).

MARK2 analog-sensitive mutants were generated by mutating the gatekeeper amino acid methionine-129 to glycine. The functionality of this mutant was confirmed using rescue assays. YAPC cells were infected with cDNA CRISPR resistant to sgMARK2+3 and three single-cell clones were picked. Mutation of endogenous MARK2 and MARK3 loci for all clones was confirmed using genotyping methods (PCR and nanopore sequencing).

Open reading frame encoding human MARK2 (Addgene, 23404) was cloned into the pFL system with an N-terminal Strep2-SUMO tag. Bacmid was generated using the pFL vector using DH10MultiBac cells (Geneva Biotech). Sf9 cells were transfected with purified bacmids. Cells were lysed and rMARK2 was purified using Strep-Tactin Superflow resin. Protein was aliquoted and snap-frozen at −80°C. Protein concentration was estimated by measuring Abs_280nm, and samples were assessed by Coomassie staining and MS analysis, confirming the absence of other protein kinases. Recombinant LATS1, LATS2, MARK3, and 14-3-3ε were purchased (Active Motif, 81209, SignalChem, L02-11G, M45-10G, and Y75-30H), and purity and correct protein size were confirmed by Coomassie staining.

Human open reading frames encoding YAP and TAZ were cloned into pGEX4T1 vector with N-terminal GST-tag. BL21-CodonPlus (DE3)-RIPL competent cells (Agilent, 230280) were transformed with sequence-validated vectors. Protein expression was induced with IPTG (GoldBio, I2481C) at 16°C for 18 hours. Bacteria were sedimented, lysed, and sonicated and cleared lysates were loaded, washed, and followed by elution using 50 mmol/L Tris pH 8, 300 mmol/L NaCl, 10% glycerol, and 20 mmol/L reduced L-glutathione. Purified proteins were aliquoted and flash-frozen at −80°C. The purity of the proteins was assessed by Coomassie staining. Protein concentration was estimated through Abs_280nm measurements.

Gatekeeper mutant MARK2^M129G or MARK3^M132G cDNA was cotransfected together with cDNAs of individual genes into HEK293T using PEI. After 24 hours, cells were harvested and incubated for 30 minutes at 30°C in bulky-ATP-analog (N⁶-furfuryl-ATP-γ-S) containing kinase-labeling buffer (protease inhibitor, 20 mmol/L HEPES, 100 mmol/L potassium acetate, 5 mmol/L sodium acetate, 2 mmol/L magnesium acetate, 10 mmol/L magnesium chloride, 1 mmol/L ethylene glycol-bis(2-aminoethylether)-N,N,N′,N′-tetraacetic acid (EGTA), 45 µg/mL Digitonin, 0.5 mmol/L tris(2-carboxyethyl)phosphine (TCEP), 5 mmol/L GTP, 600 µmol/L ATP, and 75 µmol/L N⁶-furfuryl-ATP-γ-S). Cells were lysed using RIPA buffer (with the addition of 0.1% SDS and 250 U/mL Benzonase). Thiophosphorylated substrates were alkylated using 2.5 mmol/L para-nitrobenzyl mesylate for 10 minutes at room temperature. Target proteins were affinity purified and analyzed using Western blot and anti-thiophosphate ester-specific antibodies.

Substrate cDNAs were transfected into HEK293T as described above and sampled 24 hours after transfection. Samples were affinity purified using HA-agarose beads (Sigma-Aldrich) and treated with 800 U of Lambda Protein Phosphates (New England Biolabs) for 30 minutes at 30°C. Beads were washed with RIPA buffer (with protease and phosphatase inhibitor cocktails). Next, beads-bound proteins were incubated for 30 minutes at 30°C with 3 µg rMARK2 in kinase-buffer (Tris-HCl pH = 7.5, 5 mmol/L MgCl₂, 2 mmol/L EGTA, 0.5 mmol/L DTT, 100 µmol/L ATP, and protease and phosphatase inhibitor cocktail). Phosphorylated substrates and negative controls were resolved by SDS-PAGE, and proteins were stained with Coomassie blue. The bands corresponding to each putative substrate were excised, and gel bands were destained. After irreversible alkylation of cysteine residues, proteins were digested with trypsin, and peptides were analyzed by LC/MS-MS. Peptides were resolved by nanoscale reversed-phase chromatography and ionized by electrospray (2,200 V) into a quadrupole Orbitrap Exploris 480 mass spectrometer (Thermo Fisher Scientific). The MS was set to collect 120,000 resolution precursor scans before data-dependent Higher-energy collisional dissociation (HCD) fragmentation and collection of MS-MS spectra. The area under the curve (AUC) for chromatographic peaks of precursor peptide ions was used as quantitative metrics for label-free quantification.

Raw files were analyzed using the Proteome Discoverer environment. For peptide identification, the spectra were matched against the UniProt human sequence database, supplemented with common contaminants from the cRAP database and with the sequences of the recombinant proteins expressed as substrates. S/T/Y phosphorylation, N/Q deamidation, and M oxidations were set as variable modifications. Alkylation of C residues with 2-Carboxyethyl Methanethiosulfonate (CEMTS) was set as static modification. Up to three missed trypsin cleavages were allowed. Peptide-spectrum matches were filtered using Percolator to maintain 1% FDR using the target-decoy method. The AUC defined by peptide ion XIC was integrated and used as a quantitative metric for label-free quantification. To evaluate differential phosphorylation in MARK2-treated samples compared with controls, peptides from each putative substrate were parsed out, and label-free quantification (LFQ) AUC values were used as metrics for relative chemical isoform abundance across conditions. Peptides with no LFQ value in any of the samples were disregarded. For peptides only quantified in one experimental arm, the missing value was imputed using a value smaller than the smallest empirical LFQ in the dataset (value chosen as a proxy for LFQ at detection limit). Relative amounts of phosphorylated peptides in MARK2-treated and control samples were assessed for each chemical isoform independently. Phosphopeptides that were either specifically detected in the MARK2-treated samples or showed differential abundance across conditions (more than twofold change in MARK2-treated vs. untreated samples) and whose identity could be confirmed by manual spectral interpretation were prioritized for further validation using in-cell kinase-substrate identification strategy described above. The fragmentation spectra supporting peptide identity and phosphorylation localization together with the extracted precursor ion chromatogram (XIC) can be found in the Supplementary Material.

Cas9-expressing cancer cells were infected with lentivirus. After 3 days, GFP percentage was determined using flow cytometry. GFP⁺ cells were seeded into 24-well plates at a density of 5,000/well. Cells were selected and grown for 10 to 12 days in the presence of 10 µg/mL Blasticidin for controls to reach near confluency. Media were changed every 3 days. Cells were fixed using 4% PFA for 15 minutes, followed by staining with Crystal violet (1 mg/mL in 90/10% water/ethanol) for 5 minutes. Wells were washed four times with water, and plates were imaged.

After perturbation, cancer cells were treated with 500 µmol/L cytosolic extraction buffer (10 mmol/L HEPES, 10 mmol/L KCl, 1 mmol/L DTT, 0.1 mmol/L EDTA, and 0.1 mmol/L EGTA) for 10 minutes on ice. Cells were vortexed for 10 seconds after the addition of NP40 (final 0.65%) to allow hypotonic cell membrane lysis, followed by 5 minutes of 1,500g centrifugation at 4°C. The cytosolic fraction was removed, and pelleted nuclei were lysed in RIPA buffer supplemented with 250 U/mL Benzonase and protease and phosphatase inhibitor cocktail.

HEK293T cells were transfected with vectors expressing myc-LATS1, myc-LATS2, V5-14-3-3, or V5-NF2 together with FLAG-MARK2, FLAG-MARK2^K82H, or FLAG-MARK3 and wild-type or mutant HA-tagged substrate cDNAs. For immunoprecipitation, the cells were lysed in NP40 buffer (20 mmol/L of Tris-HCl, 100 mmol/L of NaCl, 1% NP40, 2 mmol/L of EDTA, protease and phosphatase inhibitor cocktail) or RIPA buffer (Thermo Fisher Scientific) for 10 minutes at 4°C. Protein lysates were then centrifuged at 13,000g for 15 minutes at 4°C. The supernatant was then transferred to new collection tubes and incubated with 30 µL of prewashed anti-myc or anti-V5 beads (ChromoTek) and equilibrated to a final volume of 1,000 µL by adding lysis buffer. Precipitation was performed at 4°C overnight and followed by washing four to five times with lysis buffer. Samples were eluted by boiling for 10 minutes in 2× Laemmli Sample Buffer supplemented with β-mercaptoethanol.

Bacterial purified recombinant GST-YAP or GST-TAZ were preincubated for 30 minutes at 30°C with recombinant MARK2 or MARK3 in Kinase buffer followed by incubation with either recombinant LATS1 or LATS2 for an additional 30 minutes. Phosphorylated YAP or TAZ was then incubated with 6xHis-14-3-3 bound to Ni-NTA affinity resin for 4 to 16 hours followed by washing and samples elution.

A peptide sequence around S127/S128 of YAP was used and charge-stabilized by adding arginine at the carboxy-terminal end (ALTPQHVRAHSSPASLQLGAVR). Peptide was synthesized with or without phosphorylation at S128, S127/S128. Then, 400 ng of LATS2 and 1 µg of peptide were incubated in Kinase buffer (Tris-HCl pH = 7.5, 5 mmol/L MgCl₂, 2 mmol/L EGTA, 0.5 mmol/L Dithiothreitol (DTT), 100 µmol/L ATP, protease and phosphatase inhibitor cocktail), and peptides were prepared and analyzed using MS as described above.

Total RNA was extracted using TRIzol (Thermo Fisher Scientific) followed by cDNA synthesis (Quantabio) following the manufacturer's instructions. Briefly, cells were lysed using 1 mL TRIzol followed by chloroform addition and isopropanol/ethanol precipitation. cDNA synthesis of 1 µg of total RNA using the recommended thermocycler program (5 minutes, 25°C; 30 minutes, 42°C; and 5 minutes, 85°C). For qPCR Power SYBR Green PCR Master mix (Thermo Fisher Scientific) was used. Then, 1 µL of cDNA template per reaction was added to 5 µL of master mix and 0.5 µmol/L of primer, followed by triplicate PCR. Ct values were calculated for each primer pair, normalized to GAPDH and HPRT, and differential MARK2 expression (ΔΔCt) in sgMARK2 samples was quantified compared with sgControl.

For RNA-seq libraries, total RNA was prepared using TRIzol reagent according to the manufacturer’s protocol (Thermo Fisher Scientific). Libraries were constructed with the TruSeq Sample Prep Kit v2 (Illumina) following the manufacturer’s protocol. Briefly, 2 μg of total RNA was used for poly-A enrichment, fragmentation, cDNA synthesis, end repairing, A tailing, adapter ligation, and library amplification. For CUT&RUN, antibody-guided DNA cleavage was performed using the CUTANA CUT&RUN Kit (EpiCypher) according to the manufacturer’s instructions. Briefly, 500,000 knockout cells were cross-linked for 1 minute using 1% PFA and quenched using glycine for minutes. Prewashing buffer was used with detergents (0.05% SDS and 0.2% Triton X-100). Antibodies used were H3K27ac and IgG (EpiCypher, 13-0045;13-0042). Libraries were constructed with the NEBNext Ultra II DNA Library Prep Kit (New England Biolabs) following the manufacturer’s low DNA protocol. Briefly, complete CUT&RUN DNA extracts were spiked-in with Escherichia coli DNA fragments and subjected to end repair, A tailing, and adapter ligation (at 1/25 dilution) followed by PCR amplification. Libraries were purified using AMPure XP beads before and after PCR. Barcoded libraries were sequenced using an Illumina NextSeq.

For cell lines, basal expression data (TPM) and copy number variation (CNV) absolute values from the Cancer Cell Line Encyclopedia (CCLE; ref. 87) were used. RNA-seq data for KLM1 was obtained from GSE140484. Mutational information from both the CCLE and COSMIC databases was used (88). Triple-negative breast cancer and pancreatic ductal adenocarcinoma organoid CNV data were previously published (83, 84). Normalized “The Genotype-Tissue Expression” data were downloaded from “The Human Protein Atlas” (89). Mutational analysis was performed using the custom analysis tool of the DepMap portal (https://depmap.org/portal/), using both hotspot and damaging mutations. Effect sizes and P values were calculated using the linear association model with shrinkage correction (https://github.com/broadinstitute/cdsr_models/blob/master/R/linear_association.R).

Raw reads were pseudoaligned to the transcriptome of the human genome (hg38) using Kallisto (90) with bootstrap 100. For differential gene expression analysis, pseudoalignment counts were read into DESeq2, comparing samples versus control (Ctrl^KO) with two replicates for each sample. The differential expression gene analysis was performed using a gene expression cutoff of >0.5 TPM. Results from multiple sequencing runs were batch-corrected using the R package (sva), before count normalization, transformation, and z-score calculation. For heatmap, z-scores of normalized counts from significantly (adjusted P value < 10⁻⁴) downregulated or upregulated (LFC < −1 or > 2) genes in MARK2+3^dKO condition were used and plotted using R package (ComplexHeatmap).

The differential gene expression gene lists of YAP+TAZ^dKO compared with Ctrl^KO were ranked, and the top 200 downregulated genes in YAP+TAZ^dKO condition were combined. Gene counts were ranked, and the genes found in at least one-third of models were used to generate a general cancer cell line YAP/TAZ target gene set (n = 43; Supplementary Table S7). Differentially expressed gene lists were further analyzed using gene set enrichment analysis with a weighted gene set enrichment analysis preranked tool. Then, 1,000 gene set permutations were applied (91) and the common cancer YAP/TAZ target gene set was used to analyze the effects of sgMARK2/3 dgRNAs on gene expression. All fold changes are provided in Supplementary Table S10A and S10B.

Raw reads were aligned to the human genome (hg19) and E. coli genome (K12) using Bowtie2 software in sensitive mode (92). Duplicate reads were removed before peak calling. deepTools was used to normalize samples to E. coli DNA spike-in controls. Peaks were identified using MACS2 software (93) using 5% FDR cutoff and broad peak option for histone or narrow peak option for transcription factor ChIP-seq datasets. H3K27ac peaks identified from Ctrl^KO and MARK2+3^dKO, YAP+TAZ^dKO samples were merged, and overlapping peaks were combined. Normalized tag counts were calculated using the bamliquidator package (https://github.com/BradnerLab/pipeline) without read extension, and LFC between the control and dKO samples was calculated for each peak. YAP/TAZ sensitive enhancers were defined by bound H3K27ac signal reduction (−1.5 > LFC) and the binding of YAP and TEAD4 in ChIP-seq (only enhancers with relative tag count >3 in Ctrl samples were used; n = 7,896; Supplementary Table S11).

ChIP-seq datasets of TEAD4 and YAP from MDA-MB-231 cells were obtained from public Gene Expression Omnibus data sets TEAD4 and YAP (GSE66081). Sequencing depth normalized ChIP-seq and CUT&RUN pileup tracks were generated using the UCSC Genome Browser.

For tumor growth models, cells were injected into the left or right flank of NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice. For inducible MARK inhibition experiments in vivo, 1 × 10⁵ TRE3G-MKI^WT/MUT-PGK-rtTA3 cancer cells in 100 µL growth factor–reduced Matrigel were transplanted subcutaneously into the left or right flank of NSG mice. Animals were treated with doxycycline in drinking water (2 mg/mL with 2% sucrose; Sigma-Aldrich) to induce MKI protein expression. For stable knockout experiments in vivo, YAPC cells were transduced with (hU6-sgRNA-bU6-sgRNA)-EFS-GFP-2A-BlastR lentivirus, followed by selection with Blasticidin for 3 days. After, 1 × 10⁵ GFP⁺ viable cells were transplanted subcutaneously in 100 µL growth factor–reduced Matrigel into the right flank of NSG mice. For all subcutaneous xenograft experiments, tumor growth was monitored using caliper measurements. Volume was calculated as described previously (94).

For orthotopic transplantation, 5 × 10⁴ TRE3G-MKI^WT-PGK-rtTA3, EFS-LUC-2A-NIS T3M4 cancer cells were injected into the pancreas of NSG mice. Tumor growth was followed by monitored bioluminescence imaging. Animals were treated with doxycycline in drinking water (2 mg/mL with 2% sucrose) and food (625 mg/kg) to induce MKI protein expression. In vivo bioluminescence images were acquired with an IVIS Spectrum scanner and Living Image Software (Revvity). Images were acquired with an open filter on platform D, 12 minutes after a 200 μL i.p. injection of 15 mg/mL D-Luciferin (GoldBio). Mice were anesthetized with 1.5% to 2% isoflurane in air for the duration of the scan. Due to the motile nature (and variable tissue depth) of the pancreas, four views of each mouse were acquired at each imaging time point: dorsal, left flank, ventral, and right flank. The amount of emitted light was then quantified as “radiance (photons)”, as p/seconds/cm²/sr from all four views, again using Living Image Software. The sum of these four views was then made, before graphing as a single data point. In vivo Sodium/Iodide Symporter (NIS) single-photon emission computed tomography (SPECT) images were acquired on a Mediso nanoScan SPECT/CT scanner (Mediso USA). Mice were injected s.c. with a nominal activity of 45 MBq [^99mTc]sodium-pertechnetate, diluted in saline to a volume of 100 µL. After 50 minutes of conscious uptake, the mice were anesthetized with 3% isoflurane in oxygen, weighed, and placed on a Mediso dual imaging cradle that monitored respiration rate and maintained body temperature with circulating warm air. Lubricating ophthalmic ointment was applied, and 1% to 2% isoflurane anesthesia was maintained for the duration of the scan. A CT scan (360 projections at 50 kVp and 112 µAs exposure) was first acquired for anatomical reference and attenuation correction. A whole-body SPECT scan with pinhole collimators was then acquired 60 minutes after [^99mTc]sodium-pertechnetate injection (total SPECT scan time, 23 minutes). CT images were reconstructed using filtered back projection with a cosine filter to a voxel size of 250 µm isotropic. SPECT images were reconstructed using a 3D iterative algorithm optimized for high dynamic range with 48 iterations and three subsets to a 128 × 128 matrix with a 390 µm isotropic voxel size. Attenuation, decay, and scatter corrections were applied. SPECT/CT images were processed and analyzed with VivoQuant 4.0 software (Invicro) as described previously (10). Briefly, SPECT images were converted to units of Standardized uptake value (SUV), and an arbitrary threshold of five SUVs was applied to highlight regions of increased uptake. The stomach and other sites of endogenous pertechnetate uptake were manually removed from the threshold region, and the remaining viable tumor volume was quantified. Additionally, the tumor volume was calculated by manually segmenting a tumor region of interest off the CT images. The humane study endpoint was determined as the control group’s average tumor size reaching >600 mm³ or individual tumor size reaching >2 cm in diameter.

For the proliferation assays, cells were seeded at a density of 500 cells per well into 96-well plates. Cells were treated 24 hours after seeding and cell viability was assessed 5 days after treatment using the CellTiter-Glo luminescent cell viability assay (Promega). Cells were treated with vehicle control DMSO (0.1%) or killing control 10 µmol/L proteasome inhibitor (MG132). Percent viability was calculated by normalizing relative luminescence units to DMSO (0.1%) after subtraction of the killing control MG132 (10 µmol/L) signal.

For organoids, 5,000 or 10,000 cells were seeded in a 10% Matrigel/90% organoid media mix and grown for 10 days in the presence or absence of 2 µg/mL doxycycline, before assessment of viability using the CellTiter-Glo luminescent cell viability assay (Promega).

All mouse experiments were approved by the Cold Spring Harbor Laboratory Animal Care and Use Committee. Animals were treated with doxycycline in drinking water (2 mg/mL with 1% sucrose; Sigma-Aldrich) to induce cDNA expression.

Genomic datasets are available from the Gene Expression Omnibus database under accession code GSE242517. The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD044829. KLM1 dataset was obtained from GSE140484. ChIP-seq data (YAP and TEAD4) were obtained from GSE66083. Cell line datasets from CCLE (expression as well as mutations and CNV data) were obtained online (https://depmap.org/portal/download/). Other data supporting these findings are available in the manuscript and Supplementary Data. Codes used in generating figures and analyzing the data will be available at https://doi.org/10.5281/zenodo.10042504.

D.L. Spector reports grants from NIH during the conduct of the study. D.A. Tuveson reports other support from Leap Therapeutics, Xilis, Mestag Therapeutics, Dunad Therapeutics, and Sonata, as well as grants from ONO outside the submitted work. C.R. Vakoc reports grants from Treeline Biosciences during the conduct of the study, as well as personal fees from Treeline Biosciences and KSQ Therapeutics outside the submitted work. No disclosures were reported by the other authors.

O. Klingbeil and C.R. Vakoc: conceived this project and wrote the manuscript with input from all of the authors. O. Klingbeil and C.R. Vakoc designed the experiments. O. Klingbeil performed experiments with help from D. Skopelitis, C. Tonelli, A. Alpsoy, F. Minicozzi, D. Aggarwal, T. Yoshimoto and S. Russo. O. Klingbeil and T. Ha performed statistical analysis. T.-L. Wee and T. Yoshimoto performed Immunofluorescence imaging and analysis. O. Klingbeil and O.E. Demerdash designed CRISPR sgRNAs. O. Klingbeil designed and cloned paralog cotargeting CRISPR libraries. O. Klingbeil and D. Skopelitis screened libraries in cancer cell lines. O. Klingbeil, C. Tonelli, S.K. Lyons, A.M. Cafiero, J. Merrill, conducted all xenograft imaging experiments. A. Alpsoy generated recombinant MARK2 proteins. F. Minicozzi and O. Klingbeil performed mass spectrometry sample preparations. M.C. Panepinto, F. Minicozzi and P. Cifani performed all mass spectrometry measurements. D. Aggarwal, S. Russo and O. Klingbeil performed organoid experiments. C.R. Vakoc, D.A. Tuveson, P. Cifani, S.K. Lyons, and D.L. Spector supervised the studies and acquired funding.

This work was supported by Cold Spring Harbor Laboratory NCI Cancer Center Support grant P30-CA045508; additional funding by the Pershing Square Sohn Cancer Research Alliance, Simons Foundation, Thompson Foundation, NIH grants CA013106, CA245859, and CA229699; as well as support from the Christina Renna Foundation, The Mary Ruchalski Foundation, Friends of T.J. Foundation, Michelle Paternoster Foundation, Daniela Conte Foundation, and Maddie’s Promise Foundation (to C.R. Vakoc); support from NCI 2P01CA013106 and CSHL/Northwell Health (to D.L. Spector); support from Deutsche Forschungsgemeinschaft (DFG) Research Fellowship KL 3228/1-1 (to O. Klingbeil); support from the Lustgarten Foundation, Simons Foundation, Thompson Foundation, the Pershing Square Foundation, the Cold Spring Harbor Laboratory and Northwell Health Affiliation, the Northwell Health Tissue Donation Program, and the Cold Spring Harbor Laboratory Association; as well as NIH grants CA45508, CA210240, CA224013, CA188134, and CA190092 (to D.A. Tuveson).