Pooled In Vitro and In Vivo CRISPR-Cas9 Screening Identifies Tumor Suppressors in Human Colon Organoids

•An optimized protocol for pooled CRISPR-Cas9 library screening in human colon organoids•Organoid xenografts enable unbiased identification of in vivo tumor suppressors•gRNA functionality in organoids is less predictable compared to 2D cancer cell lines•Clonal tracing with a UMI library allows adjustment for clonal drift during selection Colorectal cancer (CRC) is characterized by prominent genetic and phenotypic heterogeneity between patients. To facilitate high-throughput genetic testing and functional identification of tumor drivers, we developed a platform for pooled CRISPR-Cas9 screening in human colon organoids. Using transforming growth factor β (TGF-β) resistance as a paradigm to establish sensitivity and scalability in vitro, we identified optimal conditions and strict guide RNA (gRNA) requirements for screening in 3D organoids. We then screened a pan-cancer tumor suppressor gene (TSG) library in pre-malignant organoids with APC−/−;KRASG12D mutations, which were xenografted to study clonal advantages in context of a complex tumor microenvironment. We identified TGFBR2 as the most prevalent TSG, followed by known and previously uncharacterized mediators of CRC growth. gRNAs were validated in a secondary screen using unique molecular identifiers (UMIs) to adjust for clonal drift and to distinguish clone size and abundance. Together, these findings highlight a powerful organoid-based platform for pooled CRISPR-Cas9 screening for patient-specific functional genomics. Colorectal cancer (CRC) is characterized by prominent genetic and phenotypic heterogeneity between patients. To facilitate high-throughput genetic testing and functional identification of tumor drivers, we developed a platform for pooled CRISPR-Cas9 screening in human colon organoids. Using transforming growth factor β (TGF-β) resistance as a paradigm to establish sensitivity and scalability in vitro, we identified optimal conditions and strict guide RNA (gRNA) requirements for screening in 3D organoids. We then screened a pan-cancer tumor suppressor gene (TSG) library in pre-malignant organoids with APC−/−;KRASG12D mutations, which were xenografted to study clonal advantages in context of a complex tumor microenvironment. We identified TGFBR2 as the most prevalent TSG, followed by known and previously uncharacterized mediators of CRC growth. gRNAs were validated in a secondary screen using unique molecular identifiers (UMIs) to adjust for clonal drift and to distinguish clone size and abundance. Together, these findings highlight a powerful organoid-based platform for pooled CRISPR-Cas9 screening for patient-specific functional genomics. Tumorigenesis involves the successive acquisition of somatic alterations that confer unrestricted local growth before progression to metastatic disease. Colorectal cancer (CRC) is the third most frequent type of newly diagnosed cancer in men and women and responsible for 8% of cancer-related deaths in the United States (Siegel et al., 2018Siegel R.L. Miller K.D. Jemal A. Cancer statistics, 2018.CA Cancer J. Clin. 2018; 68: 7-30Crossref PubMed Scopus (6445) Google Scholar). Exome sequencing has led to the identification of recurrent genetic lesions, but the detection of less frequent but potentially actionable drivers remains challenging (Cancer Genome Atlas Network, 2012Cancer Genome Atlas NetworkComprehensive molecular characterization of human colon and rectal cancer.Nature. 2012; 487: 330-337Crossref PubMed Scopus (5904) Google Scholar, Grasso et al., 2018Grasso C.S. Giannakis M. Wells D.K. Hamada T. Mu X.J. Quist M. Nowak J.A. Nishihara R. Qian Z.R. Inamura K. et al.Genetic mechanisms of immune evasion in colorectal cancer.Cancer Discov. 2018; 8: 730-749Crossref PubMed Scopus (274) Google Scholar, Lawrence et al., 2014Lawrence M.S. Stojanov P. Mermel C.H. Robinson J.T. Garraway L.A. Golub T.R. Meyerson M. Gabriel S.B. Lander E.S. Getz G. Discovery and saturation analysis of cancer genes across 21 tumour types.Nature. 2014; 505: 495-501Crossref PubMed Scopus (2087) Google Scholar). Oncogenes and tumor suppressor genes (TSGs) tend to show strong context specificity that depends on the tumor type, the cell of origin, and environmental factors, including metabolism, inflammation, and the microbiome (Berger et al., 2011Berger A.H. Knudson A.G. Pandolfi P.P. A continuum model for tumour suppression.Nature. 2011; 476: 163-169Crossref PubMed Scopus (367) Google Scholar, Schneider et al., 2017Schneider G. Schmidt-Supprian M. Rad R. Saur D. Tissue-specific tumorigenesis: context matters.Nat. Rev. Cancer. 2017; 17: 239-253Crossref PubMed Scopus (180) Google Scholar, Seehawer et al., 2018Seehawer M. Heinzmann F. D’Artista L. Harbig J. Roux P.-F. Hoenicke L. Dang H. Klotz S. Robinson L. Doré G. et al.Necroptosis microenvironment directs lineage commitment in liver cancer.Nature. 2018; 562: 69-75Crossref PubMed Scopus (198) Google Scholar). Clonal selection and expansion result in a considerable genetic and phenotypic heterogeneity (Roerink et al., 2018Roerink S.F. Sasaki N. Lee-Six H. Young M.D. Alexandrov L.B. Behjati S. Mitchell T.J. Grossmann S. Lightfoot H. Egan D.A. et al.Intra-tumour diversification in colorectal cancer at the single-cell level.Nature. 2018; 556: 457-462Crossref PubMed Scopus (294) Google Scholar, Sottoriva et al., 2015Sottoriva A. Kang H. Ma Z. Graham T.A. Salomon M.P. Zhao J. Marjoram P. Siegmund K. Press M.F. Shibata D. Curtis C. A Big Bang model of human colorectal tumor growth.Nat. Genet. 2015; 47: 209-216Crossref PubMed Scopus (644) Google Scholar), creating a demand for advanced experimental models that can recapitulate the complexity of tumor progression. Genetic screening allows assignment of gene function in an unbiased manner. In translational oncology, pooled barcoded lentiviral screens using the CRISPR-Cas9 technology have leveraged the identification of resistance mediators and genotype-specific fitness genes (Shalem et al., 2014Shalem O. Sanjana N.E. Hartenian E. Shi X. Scott D.A. Mikkelson T. Heckl D. Ebert B.L. Root D.E. Doench J.G. Zhang F. Genome-scale CRISPR-Cas9 knockout screening in human cells.Science. 2014; 343: 84-87Crossref PubMed Scopus (3102) Google Scholar, Wang et al., 2014Wang T. Wei J.J. Sabatini D.M. Lander E.S. Genetic screens in human cells using the CRISPR-Cas9 system.Science. 2014; 343: 80-84Crossref PubMed Scopus (1877) Google Scholar, Hart et al., 2015Hart T. Chandrashekhar M. Aregger M. Steinhart Z. Brown K.R. MacLeod G. Mis M. Zimmermann M. Fradet-Turcotte A. Sun S. et al.High-resolution CRISPR screens reveal fitness genes and genotype-specific cancer liabilities.Cell. 2015; 163: 1515-1526Abstract Full Text Full Text PDF PubMed Scopus (862) Google Scholar, Pan et al., 2018Pan D. Kobayashi A. Jiang P. Ferrari de Andrade L. Tay R.E. Luoma A.M. Tsoucas D. Qiu X. Lim K. Rao P. et al.A major chromatin regulator determines resistance of tumor cells to T cell-mediated killing.Science. 2018; 359: 770-775Crossref PubMed Scopus (452) Google Scholar, Zhan et al., 2019Zhan T. Rindtorff N. Betge J. Ebert M.P. Boutros M. CRISPR/Cas9 for cancer research and therapy.Semin. Cancer Biol. 2019; 55: 106-119Crossref PubMed Scopus (153) Google Scholar). After xenotransplantation, new TSGs and metastatic drivers have been uncovered (Chen et al., 2015Chen S. Sanjana N.E. Zheng K. Shalem O. Lee K. Shi X. Scott D.A. Song J. Pan J.Q. Weissleder R. et al.Genome-wide CRISPR screen in a mouse model of tumor growth and metastasis.Cell. 2015; 160: 1246-1260Abstract Full Text Full Text PDF PubMed Scopus (567) Google Scholar, Song et al., 2017Song C.-Q. Li Y. Mou H. Moore J. Park A. Pomyen Y. Hough S. Kennedy Z. Fischer A. Yin H. et al.Genome-wide CRISPR screen identifies regulators of mitogen-activated protein kinase as suppressors of liver tumors in mice.Gastroenterology. 2017; 152: 1161-1173.e1Abstract Full Text Full Text PDF PubMed Scopus (74) Google Scholar). Although cancer cell lines are readily accessible for large-scale experiments, they cannot reflect all aspects of human pathophysiology. Screening in primary tumor cells would be desirable to investigate patient-specific responses, processes involving tumor stem cells, cell differentiation, and interactions with the tumor microenvironment. Yet such screening technologies are currently not accessible. Patient-derived tumor organoids have recently emerged as preclinical models that faithfully recapitulate the molecular and phenotypic characteristics of CRC (Sato et al., 2011Sato T. Stange D.E. Ferrante M. Vries R.G.J. Van Es J.H. Van den Brink S. Van Houdt W.J. Pronk A. Van Gorp J. Siersema P.D. Clevers H. Long-term expansion of epithelial organoids from human colon, adenoma, adenocarcinoma, and Barrett’s epithelium.Gastroenterology. 2011; 141: 1762-1772Abstract Full Text Full Text PDF PubMed Scopus (2093) Google Scholar, van de Wetering et al., 2015van de Wetering M. Francies H.E. Francis J.M. Bounova G. Iorio F. Pronk A. van Houdt W. van Gorp J. Taylor-Weiner A. Kester L. et al.Prospective derivation of a living organoid biobank of colorectal cancer patients.Cell. 2015; 161: 933-945Abstract Full Text Full Text PDF PubMed Scopus (1348) Google Scholar, Vlachogiannis et al., 2018Vlachogiannis G. Hedayat S. Vatsiou A. Jamin Y. Fernández-Mateos J. Khan K. Lampis A. Eason K. Huntingford I. Burke R. et al.Patient-derived organoids model treatment response of metastatic gastrointestinal cancers.Science. 2018; 359: 920-926Crossref PubMed Scopus (855) Google Scholar). In a 3D matrix, tumor and normal epithelial cells can be expanded and remain dependent on exogenic factors that control proliferation and differentiation. Importantly, these exogenous signals (WNT, NOTCH, BMP/transforming growth factor β [TGF-β], and EGFR/RAS) comprise the major oncogenic pathways in CRC, which allows functional enrichment of tumor cells (van de Wetering et al., 2015van de Wetering M. Francies H.E. Francis J.M. Bounova G. Iorio F. Pronk A. van Houdt W. van Gorp J. Taylor-Weiner A. Kester L. et al.Prospective derivation of a living organoid biobank of colorectal cancer patients.Cell. 2015; 161: 933-945Abstract Full Text Full Text PDF PubMed Scopus (1348) Google Scholar, Fujii et al., 2016Fujii M. Shimokawa M. Date S. Takano A. Matano M. Nanki K. Ohta Y. Toshimitsu K. Nakazato Y. Kawasaki K. et al.A colorectal tumor organoid library demonstrates progressive loss of niche factor requirements during tumorigenesis.Cell Stem Cell. 2016; 18: 827-838Abstract Full Text Full Text PDF PubMed Scopus (456) Google Scholar). Using the CRISPR-Cas9 technology, defined mutations can be introduced to transform normal organoids and induce tumorigenic growth upon xenotransplantation (Schwank et al., 2013bSchwank G. Koo B.-K. Sasselli V. Dekkers J.F. Heo I. Demircan T. Sasaki N. Boymans S. Cuppen E. van der Ent C.K. et al.Functional repair of CFTR by CRISPR/Cas9 in intestinal stem cell organoids of cystic fibrosis patients.Cell Stem Cell. 2013; 13: 653-658Abstract Full Text Full Text PDF PubMed Scopus (968) Google Scholar, Drost et al., 2015Drost J. van Jaarsveld R.H. Ponsioen B. Zimberlin C. van Boxtel R. Buijs A. Sachs N. Overmeer R.M. Offerhaus G.J. Begthel H. et al.Sequential cancer mutations in cultured human intestinal stem cells.Nature. 2015; 521: 43-47Crossref PubMed Scopus (661) Google Scholar, Matano et al., 2015Matano M. Date S. Shimokawa M. Takano A. Fujii M. Ohta Y. Watanabe T. Kanai T. Sato T. Modeling colorectal cancer using CRISPR-Cas9-mediated engineering of human intestinal organoids.Nat. Med. 2015; 21: 256-262Crossref PubMed Scopus (719) Google Scholar). Here, we have combined the exploratory power of CRISPR-Cas9 screening with the 3D organoid system. By adaptation of the specific requirements with respect to scale and growth dynamics of organoids, we have devised protocols to study clonal advantages that are caused by human TSGs in vitro and in vivo. In order to establish pooled screens in human colon organoids, we first sought a phenotypic trait that allows robust positive selection. TGF-β sensitivity represents a suitable model because the TGF-β pathway is well characterized and contains several TSGs recurrently mutated in CRC (Jung et al., 2017Jung B. Staudacher J.J. Beauchamp D. Transforming growth factor β superfamily signaling in development of colorectal cancer.Gastroenterology. 2017; 152: 36-52Abstract Full Text Full Text PDF PubMed Scopus (151) Google Scholar). Pharmacologic inhibition is critical for expansion of normal human colon organoids (Sato et al., 2011Sato T. Stange D.E. Ferrante M. Vries R.G.J. Van Es J.H. Van den Brink S. Van Houdt W.J. Pronk A. Van Gorp J. Siersema P.D. Clevers H. Long-term expansion of epithelial organoids from human colon, adenoma, adenocarcinoma, and Barrett’s epithelium.Gastroenterology. 2011; 141: 1762-1772Abstract Full Text Full Text PDF PubMed Scopus (2093) Google Scholar), and addition of recombinant TGF-β to the culture medium allows efficient killing of organoids (Figure 1A; Matano et al., 2015Matano M. Date S. Shimokawa M. Takano A. Fujii M. Ohta Y. Watanabe T. Kanai T. Sato T. Modeling colorectal cancer using CRISPR-Cas9-mediated engineering of human intestinal organoids.Nat. Med. 2015; 21: 256-262Crossref PubMed Scopus (719) Google Scholar). For selection, the TGF-β receptor (TGFBR) inhibitor (A83-01) was withdrawn from the medium and TGF-β was added, which caused fully penetrant death within 3 weeks. In contrast, negative selection for other pathways, such as Wnt loss or TP53 stabilization, led to persistence of organoid cells (Figure S1A), possibly by induction of differentiation or cellular senescence, which could cause background in pooled screens. Next, we induced genetic resistance in normal organoids stably expressing Cas9. Transduction with a lentiviral gRNA against TGFBR2 rescued the toxicity (Figure 1A), offering a positive control to investigate performance and scalability of this model. Pooled lentiviral screening requires a multiplicity of infection (MOI) below 1; however, it is unknown whether single virus transduction is sufficient to induce biallelic loss-of-function mutations in organoids. In addition, the intrinsic cell heterogeneity due to spontaneous differentiation could result in inhomogeneous transduction and/or Cas9 modification. To address these questions, the transduction efficiency was first titrated (Figure S1B) before cells were simultaneously transduced at low titer with two TGFBR2 gRNA lentiviral vectors that contained either a GFP or a DsRed reporter (Figure 1B). Without selection, 4.0% of cells were reporter positive, but after TGF-β selection, we found strongly increased reporter expression in 68.0% of cells (Figure 1C). Clear enrichment of cells with only one color was observed, with only moderate bias toward double integrations (Figure 1C). By fluorescence microscopy, the majority of organoids displayed a single color, indicating their clonal origin (Figure 1D) and demonstrating that transduction at MOI < 1 allows efficient CRISPR-Cas9 modification. As benchmark parameters, we determined that 5.4% ± 0.4% of all organoid colonies were edited, corresponding to 5,000 events per 1 mL Matrigel (Figure S1C). Next, we turned to a pooled-barcoded scenario, in which multiple gRNAs can be studied in parallel by next-generation sequencing (NGS). For this purpose, a “training library” was designed targeting six critical components of the TGF-β pathway (Figure 2A) and 94 control genes. Among the control genes were neutral genes, as well as non-targeting random sequences. gRNAs were designed using the CRISPR library designer software (Heigwer et al., 2014Heigwer F. Kerr G. Boutros M. E-CRISP: fast CRISPR target site identification.Nat. Methods. 2014; 11: 122-123Crossref PubMed Scopus (525) Google Scholar, Heigwer et al., 2016Heigwer F. Zhan T. Breinig M. Winter J. Brügemann D. Leible S. Boutros M. CRISPR library designer (CLD): software for multispecies design of single guide RNA libraries.Genome Biol. 2016; 17: 55Crossref PubMed Scopus (45) Google Scholar; Table S1), and each gene was targeted by 20 independent gRNAs to favor calling of positive hits. The entire library contained 2,200 gRNAs and was transduced in normal human colon organoids stably expressing Cas9 (Figure 2B). 20-fold coverage was envisaged, corresponding to 44,000 clonal events, and the screen was conducted in two experimental replicates. Organoids were either collected 2 days after transduction (T0 control), after 2 weeks of regular culture as “no selection control” (NSC) or following TGF-β selection (as above) after 6.5 weeks, when all non-transfected controls had died and positive clones had reached a discernable size (Figure S2A). In total, we observed 178 TGF-β-resistant colonies in the first and 148 in the second replicate (Figures S2B and S2C). In a pooled fashion, genomic DNA was extracted and lentiviral barcodes were analyzed by NGS. Individual gRNAs against the TGFBR1 and TGFBR2 were most strongly enriched, confirming effective selection. Concomitantly, the median gRNAs distribution was reduced compared to plasmid, T0 control, and NSC (Figure 2C; Table S2). Similar gRNAs were identified in both replicates (Figure 2D), and control gRNAs targeting neutral genes were not recovered, arguing for biologic specificity. Remarkably, however, only a small fraction of positive control gRNAs against the TGFBR2 (4/20), TGFBR1 (2/20), and SMAD2 (1/20) showed enrichment (mean log2 fold change > 2) and SMAD3, SMAD4, and SARA were not recovered (Figures 2D and 2E). Genomic position of gRNA binding sites or scores from various gRNA design tools did not associate with activity (Figures S3A and S3B), indicating limited power to predict efficient gRNAs in organoids. To further study the gRNA characteristics in organoids, we compared the library performance side by side with an immortalized cell line. HepG2 cells were chosen because they are TGF-β sensitive (Senturk et al., 2010Senturk S. Mumcuoglu M. Gursoy-Yuzugullu O. Cingoz B. Akcali K.C. Ozturk M. Transforming growth factor-beta induces senescence in hepatocellular carcinoma cells and inhibits tumor growth.Hepatology. 2010; 52: 966-974Crossref PubMed Scopus (160) Google Scholar) and have previously been used for CRISPR-Cas9 screens (Xia et al., 2016Xia P. Zhang X. Xie Y. Guan M. Villeneuve D.L. Yu H. Functional toxicogenomic assessment of triclosan in human HepG2 cells using genome-wide CRISPR-Cas9 screening.Environ. Sci. Technol. 2016; 50: 10682-10692Crossref PubMed Scopus (38) Google Scholar). HepG2 cells were transduced at 100-fold coverage and collected after 3 weeks of TGF-β selection. The screen was conducted in two replicates, and 2,000 and 4,000 resistant colonies were obtained (Figures 3A and 3B ). Barcode sequencing showed strongly decreased representation of gRNAs after selection (Figure 3C; Table S3), although individual gRNAs against the TGFBR1 (13/20), TGFBR2 (16/20), SMAD3 (12/20), SMAD4 (13/20), and SARA (4/20) were strongly enriched (mean log2 fold change > 2; Figures 3D and S4B). Global comparison showed that gRNAs with high efficiency in organoids generally performed well in HepG2 cells (Figures 2E, 3D, and 3E). The activity in HepG2 cells was not generally associated with functionality in organoids. However, restriction to the ten TGFBR1/2 gRNAs with highest activity in HepG2 cells increased the fraction of effective gRNAs from 15% to 40% (Figure 3F). We conclude that prescreening in heterologous cells allows obtaining smaller, more effective gRNA libraries and thereby improve performance in organoids. To investigate the limiting factors for CRISPR-Cas9 activity in organoids, we directly measured the genome editing frequencies using the “Inference of CRISPR Edits” (ICE) assay (Hsiau et al., 2018Hsiau T. Maures T. Waite K. Yang J. Kelso R. Holden K. Stoner R. Inference of CRISPR edits from Sanger trace data.bioRxiv. 2018; https://doi.org/10.1101/251082Crossref Scopus (0) Google Scholar). Two TGFBR2 gRNAs with low and high enrichment in organoids were compared that showed similar indel frequency in HepG2 cells (Figure S5A). Experiments were performed in the absence of TGF-β to exclude cell-specific phenotypes upon selection. In organoids, however, significantly different editing was found (Figure S5B). Impairment of the weaker gRNA was preserved in organoids with APC and KRAS mutations, arguing that oncogenic transformation alone is not sufficient to improve CRISPR-Cas9 activity (Figure S5C). In contrast, additional mutation of TP53 partially restored gRNA functionality, suggesting that, similar to other untransformed cells, a CRISPR-Cas9-induced, P53-dependent DNA damage response limits the activity of some gRNAs in organoids (Haapaniemi et al., 2018Haapaniemi E. Botla S. Persson J. Schmierer B. Taipale J. CRISPR-Cas9 genome editing induces a p53-mediated DNA damage response.Nat. Med. 2018; 24: 927-930Crossref PubMed Scopus (594) Google Scholar, Ihry et al., 2018Ihry R.J. Worringer K.A. Salick M.R. Frias E. Ho D. Theriault K. Kommineni S. Chen J. Sondey M. Ye C. et al.p53 inhibits CRISPR-Cas9 engineering in human pluripotent stem cells.Nat. Med. 2018; 24: 939-946Crossref PubMed Scopus (508) Google Scholar). Our next aim was to perform screening in transplanted organoids to study clonal selection in a complex microenvironment. As a positive-selection paradigm, we generated a pre-tumorigenic organoid line that can form subcutaneous tumors in NSG mice upon one additional genetic hit. Colon organoids that had been engineered with loss of APC and the oncogenic KRASG12D allele (AK organoids) showed growth only after additional deletion of TGFBR2 (AKT organoids; Figures 4A–4C, S6A, and S6B), providing a sensitive genetic system. To test whether gene ablation can induce a clonal growth advantage, AK organoids were transduced at low titer with a lentiviral gRNA together with a GFP reporter. We found that TGFBR2 gRNA, but not a random control gRNA, strongly increased the fraction of GFP-positive cells from 8.4% before transplantation to 98.5% after tumor growth (Figures 4D–4F). We furthermore confirmed that AK organoids retained a genetically stable diploid phenotype in vitro (Figures S6C and S6D) and could be efficiently modified by CRISPR-Cas9 (Figures S6E and S6F). Comprehensive tumor sequencing programs have only identified a limited number of TSGs that are recurrently mutated in CRC (Cancer Genome Atlas Network, 2012Cancer Genome Atlas NetworkComprehensive molecular characterization of human colon and rectal cancer.Nature. 2012; 487: 330-337Crossref PubMed Scopus (5904) Google Scholar, Grasso et al., 2018Grasso C.S. Giannakis M. Wells D.K. Hamada T. Mu X.J. Quist M. Nowak J.A. Nishihara R. Qian Z.R. Inamura K. et al.Genetic mechanisms of immune evasion in colorectal cancer.Cancer Discov. 2018; 8: 730-749Crossref PubMed Scopus (274) Google Scholar, Lawrence et al., 2014Lawrence M.S. Stojanov P. Mermel C.H. Robinson J.T. Garraway L.A. Golub T.R. Meyerson M. Gabriel S.B. Lander E.S. Getz G. Discovery and saturation analysis of cancer genes across 21 tumour types.Nature. 2014; 505: 495-501Crossref PubMed Scopus (2087) Google Scholar). To investigate the role of drivers that are less frequent in CRC and to cover diverse biological pathways, we designed a pan-cancer TSG gRNA library that included 85 TSGs that were previously identified in 16 solid non-neuronal cancer entities (Davoli et al., 2013Davoli T. Xu A.W. Mengwasser K.E. Sack L.M. Yoon J.C. Park P.J. Elledge S.J. Cumulative haploinsufficiency and triplosensitivity drive aneuploidy patterns and shape the cancer genome.Cell. 2013; 155: 948-962Abstract Full Text Full Text PDF PubMed Scopus (476) Google Scholar; Table S4A). For each target, we designed 20 independent gRNAs, and including controls, the entire library contained 2,600 gRNAs. After transduction, AK organoids were transplanted into NSG mice at 20-fold library coverage (Figure 5A). For this scale, organoids corresponding to 70,000–90,000 independent CRISPR-Cas9 targeting events were injected subcutaneously into 8–10 mice. After 11 to 12 weeks, the genomic DNA of all tumors was pooled and the screen was conducted in 3 experimental replicates that showed heterogeneous results (Figure 5B). Strong enrichment of individual gRNAs was observed, although more homogeneous gRNA distribution was observed in plasmid, T0, and NSC (Figure 5C). After normalization, TGFBR2 was the most highly enriched gRNA in all three replicates, indicating a dominating growth-repressive effect of TGF-β in the tumor microenvironment. Additional hits were prioritized according to the mean enrichment and induction in at least 2 of 3 replicates (Figures 5D and 5E; Table S4F). Within the top 20 TSG barcodes, 3 independent gRNAs targeting TGFBR2 were recovered. Next strongest hits included other known drivers in CRC (TP53, ATM—2 gRNAs, CASP8, and SMAD4) as well as TSGs not somatically mutated (MED23, STK11—2 gRNAs, SMARCA4, and NETO1). However, among the most highly enriched gRNAs were a number of control guides targeting neutral and essential genes, indicating presence of false positives, e.g., due to neutral drift during the prolonged selection. For validation and to obtain insight on the clonal composition of organoid-derived tumors, we generated a library that couples each gRNA to unique molecular identifiers (UMIs). Using a lentiviral vector described previously (Michlits et al., 2017Michlits G. Hubmann M. Wu S.-H. Vainorius G. Budusan E. Zhuk S. Burkard T.R. Novatchkova M. Aichinger M. Lu Y. et al.CRISPR-UMI: single-cell lineage tracing of pooled CRISPR-Cas9 screens.Nat. Methods. 2017; 14: 1191-1197Crossref PubMed Scopus (62) Google Scholar), we introduced the top 281 gRNAs from the TSG screen flanked by a 10-bp degenerated UMI (Figure S7A). In addition, gRNAs targeting neutral and essential genes were included together with positive controls (TGFBR1/2 gRNAs effective in HepG2 cells; Table S5A). By NGS, we first tested the rate of barcode uncoupling, which could compromise clonal detection. Using a pair of predefined gRNA and UMI sequences, we only detected PCR-introduced uncoupling in 0.1% of all reads (Figure S7B), thus permitting accurate clone tracing. The entire library then was transduced into Cas9 expressing AK organoids followed by subcutaneous transplantation in NSG mice and tumor growth for 8 weeks (Figures 6A and S7C). To study biological robustness and the effect of library coverage, tumors were either sequenced individually (n = 5; 15-fold coverage each) or as a pool of 10 tumors (150-fold coverage). Overall gRNA diversity was reduced in both the pool and individual tumors compared to the T0 control (Figure 6B; Tables S5B and S5C). Scoring for UMIs revealed a considerable heterogeneity of tumor clone sizes with Gini coefficients ranging between 0.6 and 0.8 (Figures S7D and S7E). After normalization, an average number of 3.6 × 103 individual clones was measured per tumor (Figure S7F). Yet the majority of all reads originated from a small percentage (1%–3%) of large clones (Figures S7G and S7H), paralleling the heterogeneity observed above (Figure 5B). To score for barcode incidence rather than abundance, we removed the largest three clones for each gRNA. Side-by-side comparison with conventional analysis showed that such removal of “outlier clones” allowed eliminating false positives (Figures 6C and 6D; Table S5D). Gene set enrichment analysis for positive controls and essential genes showed similar enrichment in both datasets, as well as in single tumors. Thus, the sensitivity of detection is neither compromised by outlier removal nor limited by the library coverage (Figure S7I). The biological robustness was confirmed by statistical analysis of enrichment in single tumors (Figures 6E and 6F). Inspection of single tumors revealed that false positives in the conventional analysis are caused by single outlier clones (Figure S7J). Together, our results demonstrate that correction for clonal drift strongly improves the resolution of pooled CRISPR-Cas9 screens in organoids. Our validated set of TSGs (15 gRNAs > 1 log2 fold) contained TGFBR2 and SMAD4 (both with 3 gRNAs each), as well as PIK3R1, ZC3H13, and FBXW7 that are all recurrently mutated in CRC (Cancer Genome Atlas Network, 2012Cancer Genome Atlas NetworkComprehensive molecular characterization of human colon and rectal cancer.Nature. 2012; 487: 330-337Crossref PubMed Scopus (5904) Google Scholar, Davoli et al., 2013Davoli T. Xu A.W. Mengwasser K.E. Sack L.M. Yoon J.C. Park P.J. Elledge S.J. Cumulative haploinsufficiency and triplosensitivity drive aneuploidy patterns and shape the cancer genome.Cell. 2013; 155: 948-962Abstract Full Text Full Text PDF PubMed Scopus (476) Google Scholar) and STK11 that has been described in familial gastrointestinal tumors (OMIM 602216). Although TP53 was not recovered, loss of the cell cycle inhibitor CDKN2A (2 gRNAs) could result in a similar block of cell cycle arrest. Mediators not previously associated with CRC were PBRM1 (recurrent in esophagus, stomach, head and neck, and kidney cancer) and ZFP36L1 (recurrent in bladder and breast cancer; Davoli et al., 2013Davoli T. Xu A.W. Mengwasser K.E. Sack L.M. Yoon J.C. Park P.J. Elledge S.J. Cumulative haploinsufficiency and triplosensitivity drive aneuploidy patterns and shape the cancer genome.Cell. 2013; 155: 948-962Abstract Full Text Full Text PDF PubMed Scopus (476) Google Scholar). The opportunity to quantitatively measure clonal advantages prompted us to compare the effect of single TSGs on clone size and number. For this purpose, gRNAs targeting TGFBR2, TP53, and STK11 or a random control gRNA were individually coupled to a UMI library. In a competitive assay, AK organoids expressing test or control gRNA were then mixed, followed by tumor growth in NSG mice and barcode sequencing (Figure 7A; Tables S