Applications of Organoids for Cancer Biology and Precision Medicine.

In general, organoids can be classified not only by their tissue of origin, but also by their generation from adult tissues 4 , 5 versus embryonic stem cells 6 . Organoids can be further distinguished by whether they solely contain epithelium 4 , or also possess stromal cells 5 . These distinct classes of organoids can be reproducibly generated by correspondingly diverse techniques ( Figure 1 ). Many seminal organoid studies utilized intestine as starting material with growth within solid gels of extracellular matrix (laminin-rich Matrigel or equivalents, such as BME-2), submerged beneath tissue culture media. Intestinal organoids can be established from either purified adult small intestinal crypts or sorted single LGR5 + intestinal stem cells (ISCs) in an elegant growth factor-defined culture condition that mimics the stem cell niche 4 . This culture condition generally includes Wnt pathway ligands such as Wnt3a and/or R-spondin, epidermal growth factor (EGF), and the bone morphogenetic (BMP) inhibitor Noggin that allow ISCs to undergo long-term self-renewal and differentiation into all cell lineages 4 . As a result, the ISCs form highly polarized 3D epithelial structures that represent intestinal crypt-villus compartments. Variations of this submerged organoid culture system use specialized growth factor cocktails adapted for many different organs, including small and large intestine 4 , 7 - 9 , stomach 10 , 11 , esophagus 9 , 12 , salivary gland 13 , 14 , taste buds 15 , pancreas 16 - 19 , liver 20 - 23 , airway 24 - 27 , breast 28 , prostate 29 , 30 , kidney 31 , fallopian tube 32 , ovary 33 , 34 , endometrium 35 and skin 36 . As discussed below, this method has also been used extensively for cancer modeling and to grow tumor biopsies. Because of the necessity for cellular disaggregation during processing, engendering deleterious Rho kinase (ROCK)-dependent anoikis as well as stress 37 and injury responses 38 , ROCK inhibitors have greatly increased the efficiency of organoid generation 4 . Typically, submerged organoids are polarized with the apical surface facing a central lumen, although this polarity can be deliberately inverted 39 . A defining characteristic of submerged organoids is their possession of exclusively epithelial cells, and absence of stroma. A parallel method utilizes air-liquid interface (ALI) culture to grow organoids that contain both epithelial cells and surrounding stroma as a cohesive unit directly from tissue fragments. In ALI, mechanically dissociated tissues are grown into a type I collagen matrix on top of an inner transwell insert with culture medium provided through a permeable membrane 5 , 40 , 41 . The top of the transwell directly contacts air, thereby facilitating oxygen diffusion 42 , which may underlie the ability of ALI to grow large multicellular organoids that preserve native tissue architecture, such as epithelium, en bloc with endogenous stromal cells without reconstitution. In ALI, the stromal cells are sufficient to support organoid growth without growth factor supplementation, likely by producing essential endogenous niche factors 5 . The ALI method grows intestine, stomach and pancreas organoids as cystic structures containing both epithelium and mesenchymal stroma, displaying expansile growth, stem cell populations, multilineage differentiation and even intestinal peristalsis 5 , 41 , 42 . These features render ALI a complementary method to submerged organoids, particularly for inclusion of stromal populations, which has been exploited for cancer oncogene engineering 41 and tumor microenvironment modeling 28 as detailed below. In contrast to submerged and ALI methods, the generation of iPSC-derived organoids obligately entails directed differentiation to the target tissue of interest. Diverse iPSC-derived organoids have been described, such as brain 43 - 46 , small and large intestine 6 , 47 , stomach 48 , 49 , esophagus 50 , liver 51 , airway 52 - 55 , thyroid 56 , 57 , kidney 58 , 59 , pituitary gland 60 , 61 , inner ear 62 - 64 , retina 65 , hair follicle 66 and blood vessels 67 . These efforts have been greatly aided by bespoke tissue-specific differentiation protocols, typically requiring weeks to months, to reprogram iPSCs to the desired organ type. Of note, some tissues, like brain, cannot be grown in submerged or ALI culture, making iPSC organoids a unique platform for studying the timing of development of neural tissues. A salient feature of endodermal iPSC models is the frequent co-development of epithelium alongside mesenchymal stroma 6 , 47 - 49 . Compared with adult cell types, molecular characteristics of iPSCs-derived organoids are more similar to immature fetal cells and in vivo transplantation can enhance maturation 68 - 72 . iPSC organoids have been broadly deployed for disease modeling, including cancer, as reviewed elsewhere 73 , 74 .

Human patient-derived tumor organoids (PDOs) can be directly established from clinical cancer biopsies in a tissue-specific fashion. As opposed to the “bottom-up” approach of engineering oncogenic alterations into wild-type tissue organoids, the direct organoid culture of fresh tumor tissue is a “top-down” strategy of studying pre-established malignancies ( Figure 3 ). Attempts to propagate fresh tumor specimens as in vitro conventional 2D cancer cell lines or in vivo patient-derived xenografts (PDX) 134 , 135 , while successful, have been historically fraught with inefficiency. However, success rates for tumor-derived organoids and PDOs have been typically higher than conventional 2D culture and PDX and while varying for different tissues, can be robust for commonly used tumor organoid systems such as pancreatic and colon cancer ( Table 1 ). PDOs have been established from surgical resections, tissue biopsies, circulating tumor cells, and ascitic fluid 34 , 136 - 138 . General features of PDO cultures, and indeed essential quality control metrics, are their accurate preservation of histologic and genetic features of the parent tumor, the latter extending to truncal alterations, such as copy number variations (CNV) and single-nucleotide variations (SNV). Furthermore, PDOs are extremely tractable from an experimental standpoint, undergoing facile lentiviral transduction, CRISPR-based gene editing and in vivo xenografting. These properties are significantly abetted by their ready expansion which renders PDOs potentially useful as preclinical models for drug screening or functional genomics with throughput greater than PDX studies. On the other hand, PDO cultures have until recently generally lacked important in vivo attributes such as stroma or immune components 40 . As such, tumor organoids containing exclusively epithelium generally exhibit transcriptional concordance with the original cancers 86 , 139 - 141 , except for downregulation of blood-, immune- and extracellular matrix-related genes 139 , 140 . Recently, many organoid biobanks have emerged that contain large numbers of PDOs from different cancer types, including colon 136 , 137 , 141 - 145 , rectum 146 , 147 , stomach 86 , 140 , 148 , 149 , pancreas 16 , 88 , 94 , 150 , 151 , bladder 152 , 153 , prostate 138 , 154 , ovary 33 , 34 , 155 , esophagus 156 , 157 , breast 28 , oral mucosa 158 , endometrium 159 , liver 160 - 162 , kidney 163 , airway 164 and brain 139 typically utilizing the submerged Matrigel culture method, although ALI PDOs have also been described 40 . The Human Cancer Models Initiative (HCMI) seeks to biobank thousands of new PDO models, replete with genetic sequencing and clinical annotations 165 . Overall, the breadth of successful PDO generation provides an exciting new bridge between basic and translational cancer research. PDOs are now used to identify new therapeutic strategies via discovery of gene-drug interactions, correlating between drug sensitivity to genomic alterations. One proof-of-concept for small molecular compound screening of PDOs was performed in an early biobank of colon cancer organoids 136 . By connecting drug sensitivity data and genetic information of PDOs, a positive correlation between oncogenic KRAS -mutated organoids and resistance to anti-EGFR inhibitor cetuximab was identified, recapitulating previous clinical observations 166 . Similarly, androgen receptor (AR) amplification in prostate cancer organoids conferred sensitivity to the AR antagonist enzalutamide. In agreement with previously identified synthetic lethality 167 , a recent biobank revealed poly (ADP-ribose) polymerase (PARP) inhibition as a therapeutic vulnerability of BRCA1/2 -deficient breast cancer organoids 28 . A bladder cancer organoid biobank indicated that organoids harboring gain-of-function FGFR3 mutations exhibited significant sensitivity to the MEK inhibitor trametinib and the ERK inhibitor SCH772984 153 . These results illustrate the ability of PDOs to accurately recapitulate clinically observed therapeutic vulnerabilities to archetypal genetic alterations. Organoid cultures can dissect the growth factor “niche” dependency of tumor cells within their microenvironment, examining if cancers exhibit altered or reduced growth factor requirements. Conventionally, tumor organoid cultures have been grown in rich conditions replete with signals for Wnt, R-spondin, EGF, BMP and other pathways, while conversely, omission of such growth factors has enabled functional selection of CRISPR-induced oncogenic mutations 76 , 77 . However, alternation of niche requirements can be crucial for the selective growth advantage of certain subtypes of cancers and for establishing cognate PDOs. The use of diverse conditions allowed successful PDO generation from many rare colon cancer subtypes previously refractory to PDO culture, such as sessile serrated adenoma, tubulovillous adenoma, mucinous adenocarcinoma and neuroendocrine carcinoma, preserving pathological features of the original tumors both in vitro and following in vivo kidney capsule xenotransplantation 144 . Despite differences in genetic background often translating to specific growth dependencies of PDOs, the functional classification of tumor subtypes based on distinct niche dependencies instead of genetic signatures may lead to identification of new molecular mechanisms underlying inter-tumor heterogeneity 144 . Previously unappreciated GATA6 -mediated Wnt signaling dependencies identified in organoids have defined three functional subtypes of PDACs 88 . Similarly, gastric cancer organoids harboring simultaneous TP53 and CDH1 mutations gained R-spondin independency whereas combined CRISPR/Cas9 double knockout of TP53 and CDH1 in wild-type gastric organoids was sufficient to allow culture without R-spondin 86 . Cancers grow within a complex microenvironment where tumor epithelial cells are invested by diverse stromal cellular components, including fibroblasts, endothelial and immune cells 168 - 170 . The crucial roles of endothelial and immune components during tumorigenesis is particularly illustrated by the clinical efficacy of anti-angiogenic and immunotherapies, respectively 171 , 172 . Nowhere has the need for holistic tumor microenvironment culture been more acute than for tumor immunology and the accompanying need to study interactions between cancer cells and their veritable ecosystem of co-habitating immune cells. However, conventional organoid models of cancer have typically only represented tumor epithelium, and holistic culture of cancer cells alongside endogenous stromal elements has been elusive. In the absence of culture systems allowing tumor cells and stroma to be preserved alongside each other, numerous studies have reconstituted heterologous cell types along with organoids containing solely tumor epithelium. Examples include “tumor-only” submerged organoids with addition of distinct cancer-associated fibroblast (CAF) subtypes 173 , 174 , or supplementation with chimeric antigen receptor (CAR) natural killer (NK) cells to study tumor cytotoxicity 175 . Of note, multiple rounds of co-culture of PDOs with matched autologous peripheral blood lymphocytes were sufficient to select for tumor-reactive T cells, which performed specific killing of tumor cells but not matched normal organoids 176 . As typical PDOs only contain tumor epithelium, there has been a singular lack of robust in vitro culture systems that retain cancer cells alongside infiltrating endogenous immune stroma and recapitulate in vivo checkpoint inhibition responses. However, recent ALI tumor organoid studies have robustly preserved the complex cellular diversity and physical architecture of both endogenous tumor and stroma compartments. PDOs in ALI preserved cancer-associated fibroblasts, numerous infiltrating immune cells, such as T cells (cytotoxic, helper, regulatory and exhausted), B cells, NK cells and macrophages over 1-2 months 40 . Importantly, subsets of ALI PDOs from non-small cell lung cancer, melanoma, renal cell carcinoma and bladder contained functional tumor-infiltrating lymphocytes (TILs) that rapidly exhibited clonal expansion, activation and/or cytotoxic responses upon short-term PD-1/PD-L1 checkpoint blockade treatment. Additionally, mouse tumors from syngeneic immunocompetent hosts could be similarly cultured in ALI with anti-PD-1/PD-L1-dependent T cell responses 40 . Although this type of holistic tumor microenvironment culture is in its infancy, potential applications include mechanistic immunooncology studies, screening of novel immunomodulatory agents and precision medicine. Cancer organoid cultures present an opportunity to capture the intratumoral heterogeneity (ITH) of human cancer for in vitro investigation. Generally, tumor organoids retain mutations from the tissue of origin 16 , 34 , 142 , 150 , 162 , as well as copy number status 108 , 140 . However, PDOs can undergo dynamic changes in culture. Many studies only compare early passage organoids with the tumor tissue but following serial passaging subclonal mutations can be gained or lost, although truncal mutations tend to be retained 153 , 164 . Long-term culture of CRC organoids for >6 months revealed that MSI lines with mismatch repair deficiency developed 75 and 82 de novo non-synonymous mutations respectively, whereas a microsatellite stable (MSS) was comparatively unchanged 144 . Liver tumor organoids retained around 92% of the tissue mutations after 80% of mutations after >4 months in culture 162 . Thus, organoid cultures show strong correlation with tissue mutational status at early passages, but genetic drift and selection can occur. In addition, any dynamic PDO genetic or epigenetic changes in culture could be biased by culture media that may not fully reflect in vivo conditions. Since organoids can only represent their originating tumor biopsy, faithful recapitulation of tumor heterogeneity may require culture from multiple sites of a single tumor or from distinct metastatic sites. Several examples of organoid cultures from multiple tumor regions indicate that these “sibling” cultures can manifest distinct drug sensitivity profiles. For example, organoids from different metastases from a PDAC patient showed similar response to three chemotherapy drugs, but different responses to 5-FU 150 . In CRC, five sibling PDO cultures from multiple regions of the same tumor exhibited up to 30 fold differences in drug response, suggesting that a single tumor biopsy may not be enough to capture all ITH 143 . A study of 27 liver cancer organoids from 5 patients indicated that most tested drugs possessed low intra-patient variation, but several drugs, including targeted therapies such as tyrosine kinase inhibitors, displayed high intra-patient variability 160 . The most extensive ITH study performed genomic, transcriptomic and epigenetic analysis of 78 clonally derived CRC PDOs across multiple tumor regions from 3 previously untreated patients 104 . Clonal organoids from the same region shared common driver mutations but could still exhibit substantial differences in overall mutational content (40%) and sensitivity to chemotherapeutics and targeted drugs (3 log variation in IC 50 ). In this study, principal component analysis of methylation and transcriptional state showed that while clones from the same patient clustered together, subclones, (for example TP53 WT vs TP53 mutant) clustered separately. Overall, organoid ITH studies indicate that although concordance exists, even clonal organoids derived from the same tumor region can exhibit highly variable mutations, transcriptomes, epigenomes and drug responses. Additionally, PDOs can model tumor mutational evolution with shared truncal mutations leading to subclonal divergence.

The genetic manipulation of primary wild-type organoids provides a unique opportunity to model human tumor initiation and progression in a tissue-specific fashion that closely recapitulates the oncogenic process ( Figure 2 ). Conventionally, interrogating the function of a candidate oncogene or tumor suppressor in vitro has utilized 2D transformed cell lines, whose underlying genetic complexity both precludes true de novo cancer initiation and may contain significant modifier loci that could mask oncogenic effects. Analogously, in vivo examination of oncogenes and tumor suppressors has required transgenic manipulation, typically in a mouse context. In contrast, systematic introduction of oncogenic loci into wild-type tissue organoids provides an attractive method for “bottom-up” tumor initiation into normal cells possessing a non-mutated genetic background, yielding a hybrid combining the tissue architecture, stem cells and differentiation of in vivo mouse models with the facile genetic manipulation of transformed human cancer cell lines. This tractability in turn renders wild-type organoids amenable to forward genetic approaches such as oncogene engineering and functional genomics. A first proof of concept for bottom-up organoid modeling of cancer exploited ALI organoids for robust in vitro transformation of adult wild-type intestinal organoids to adenocarcinoma 41 Organoids were generated from adult transgenic mice containing floxed alleles, thus allowing combinatorial CreER-mediated deletion of Apc, Tp53 and/or CreER activation of Kras G12D expression, in concert with Smad4 shRNA knockdown. The addition of tamoxifen induced CreER-dependent progressive transformation depending on whether 1, 2, 3 or 4 genetic lesions were present. In vivo growth as tumors required 4 mutations 41 , recapitulating the classical multihit “Vogelgram” model of colon tumorigenesis 75 . Thereafter, two independent studies using submerged Matrigel organoids successfully transformed normal human colonic organoids to adenocarcinoma by CRISPR-based sequential engineering of common alterations in colon cancers, including APC −/− , KRAS G12V/D , SMAD4 −/− , PIK3CA E545K , and TP53 −/− 76 , 77 . Importantly, colon organoids with quadruple mutations grew in vivo when implanted subcutaneously into mice, confirming oncogenic transformation 76 , 77 . To compensate for the inefficiency of CRISPR, these human organoid studies used selective media to enrich for mutants. For example, APC mutations can be selected for by growing organoids in medium without Wnt/R-spondin, whereas selection for TP53 mutations is possible using medium with the MDM2-P53 complex inhibitor Nutlin-3 76 , 77 . In addition, SMAD4-mutated organoids can survive in medium lacking Noggin and the transforming growth factor-β (TGFβ) inhibitor A83-01, but containing TGFβ, whereas oncogenic KRAS G12V/D and PI3KCA E545K CRISPR knock-in allows organoid growth without EGF 76 , 77 . Engineered human colon organoids can also be used to study premalignant colorectal polyposis 78 . CRISPR/Cas9-engineered BRAF V600E human colon organoids in combination with TGFβ signaling manipulation induce a mesenchymal phenotype consistent with sessile serrated adenomas (SSA) 79 , supporting prior hypotheses of BRAF V600E function 80 . Furthermore, CRISPR/Cas9 introduction of an EIF3E-RSPO2 fusion gene in human colon organoids modeled progression of traditional serrated adenomas (TSA) in vitro with histologic recapitulation of SSA following in vivo orthotopic xenotransplantation 81 . Superimposed expression of the mesenchymal bone morphogenetic protein antagonist GREM1 in EIF3E-RSPO2; BRAF V600E organoids induced TAS-like morphological aberrations 81 , consistent with the phenotype of Grem1 transgenic mice 82 . The paradigm of oncogenic conversion of colon organoids has now been expanded to additonal studies in microsatellite instability 83 . Oncogene engineering studies have rapidly progressed from early work in colorectal cancer (CRC) to many prevalent solid tumor types. In vitro conversion of normal mouse gastric and pancreatic organoids to adenocarcinoma has been achieved by deletion of Tp53 and Kras G12D activation in ALI organoids, thereby allowing in vitro dysplasia and in vivo tumorigenicity upon transplantation and recapitulating features such as desmoplasia and epithelial-mesenchymal transition (EMT) 41 . Further, Tgfbr2 deletion enhances tumorigenicity of Cdh1 −/− ; Tp53 −/− ALI gastric organoids 84 . Similar mouse transformation of colon 85 and pancreatic 16 studies have been performed using submerged Matrigel models. CRISPR/Cas9 oncogene editing of wild-type human organoids has allowed oncogenic conversion of human gastric 86 , pancreatic 87 , 88 , esophagus 89 , breast 90 and liver 91 tissue. In human stomach organoids, deletion of TP53 or CDH1 confers transformation with a diffuse gastric cancer phenotype in the latter 86 . Normal pancreatic organoids harboring four oncogenic driver mutations, including KRAS G12V , CDKN2A, TP53 , and SMAD4 , acquired niche factor independency and histologically resembled PanIN-like lesions and invasive pancreatic ductal adenocarcinomas (PDACs) 87 , 88 . Normal esophageal organoids lacking APC quickly developed Barrett esophagus-associated neoplasia 89 . Similarly, engineered breast organoids carrying mutations in four tumor suppressor genes, TP53, PTEN, RB1 and NF1 , recapitulate breast tumorigenesis 90 . Additionally, iPSC-derived organoids model neoplastic progression. Generation of iPSC from PDAC tumors, followed by redifferentiation to pancreatic lineages, results in rare lines recapitulating early PanIN lesions 92 . Early neoplastic changes are observed upon colonic differentiation of iPSC, followed by expression of mutant APC 93 , or upon mutant KRAS and TP53 expression in iPSC induced to pancreatic lineages 94 . Engineered iPSC-derived cerebral organoids carrying oncogenic mutations exhibit many features of tumor malignancy 95 . iPSC-derived organoids may find particular utility in modeling germline cancer predisposition syndromes, as with APC mutation in Familial Adenomatous Polyposis 93 or TP53 mutation in Li-Fraumeni syndrome 96 . However, potential drawbacks to cancer modeling using iPSCs include their fetal context compromising the study of adult cancers. A significant potential application of the bottom-up organoid strategy is the functional validation of novel oncogenic alterations. Numerous examples exist, including confirmation of miR-483 as a driver oncogene amplified at 11p15 in colon cancer through overexpression in colon organoids 41 , as well as demonstration of the oncogenicity of EIF3E-RSPO2 fusions 81 , 97 . Expression of Rhoa mutant alleles in gastric organoids showed that Rhoa Y42C acts as an oncogene through the activation of focal adhesion kinase in Cdh1 -mediated diffuse gastric cancer 98 . CRISPR/Cas9-engineered liver organoids have validated the function of a previously uncharacterized tumor suppressor gene, BAP1 , in hepatic carcinogenesis 91 . In these various cases, mutation of the test oncogene or tumor suppressor generally elicits in vitro and in vivo tumorigenic phenotypes. Engineered human organoids also provide tractable platforms for functionally studying interactions between cancer genotypes and phenotypes. For example, recent landmark human cancer genomic projects, such as The Cancer Genome Atlas (TCGA), have molecularly characterized four distinct subtypes of primary gastric cancer, including Epstein-Barr virus (EBV)-positive, microsatellite instability (MSI), chromosomal instability (CIN), and genomically stable (GS) tumors 99 . Cre-lox-mediated deletion of Cdh1 , recurrently mediated in the GS subtype of gastric cancers, along with Tp53 , induced signet ring cells characteristic of diffuse gastric cancers and enhanced in vivo tumorigenicity in mouse ALI gastric organoids 84 . CRISPR/Cas9 deletion of CDH1 and RHOA in normal human gastric organoids elicited striking morphological changes and vigorous migratory features characteristic of diffuse-like gastric cancers. Interestingly, RHOA inhibition reverses these phenotypes and sustains CDH1 knockout organoid cells, supporting the concurrency of CDH1 and RHOA mutations in gastric cancer 86 . Thus, many comprehensive tumor models with specific mutations have been established using organoid-based “bottom up” genetic approaches, allowing functional assessment of small numbers of genetic alterations commonly seen in patient tumor samples. The 3D properties of organoids may be especially beneficial in modeling tumor suppressor loci. For instance, lung cancer cell lines display stronger growth phenotypes in response to CRISPR-mediated knockout of tumor suppressor loci when grown as 3D spheroids versus 2D monolayer 100 . Such benefits may also extend to 3D organoid systems. Moving forward, substantial potential exists for exploiting organoids for broader-based unbiased oncogene screens leveraging substantial genetic and epigenetic data from genome-scale sequencing studies, for instance using sgRNA or shRNA barcoded approaches. Previous studies have identified mutational signatures and driver mutations by whole genome sequencing in seemingly normal colonic glands 101 , and clonal evolution in breast cancer has been assessed by single cell DNA sequencing 102 . The success in establishing organoids from normal and tumor tissue allows biomass expansion from single cell-derived clones, thus increasing fidelity of whole genome sequencing 103 and extension to multi-omics sequencing 104 , to examine mutational processes in normal 103 , 105 , 106 or tumor tissue 104 . Whole genome sequencing of mouse clonal organoids from single stomach glands, colon crypts or prostate cells revealed distinct mutational signatures amongst individual clones 105 . A similar approach demonstrated that human adult stem cells (ASCs) accumulated approximately 36 mutations per year but exhibited different turnover rates and tissue-specific mutational processes 103 . Analogous studies of forward genetic oncogenic transformation of human wild-type organoids can similarly elaborate the evolutionary dynamics and driving forces underlying tumor initiation and progression. Loss of APC and TP53 in normal colon organoids stimulates abnormal chromosome segregation and extensive aneuploidy, suggesting a key driving force of chromosome instability, a hallmark of tumor progression 77 . Accumulation of total mutational burden was quickly observed after depletion of MLH1 , a key DNA repair gene, in normal colonic organoids 83 . These studies indicated the potential for longitudinal tumor evolution studies in human organoids. For example, the chemotherapeutic drug 5-Fluorouracil (5-FU) was shown to accelerate tumor evolution with rapid T>G transversions after treatment 107 . Additionally, ongoing chromosomal instability (CIN) in colorectal tumor organoids has been documented by single-cell karyotype sequencing, further demonstrating the prevalence of genomic heterogeneity in human cancer 108 . In vivo lineage tracing in engineered mouse models is a powerful tool to study stem cell properties in adult mammalian tissues 109 , as evidenced by inducible Cre recombinase lineage tracing of long-lived adult Lgr5 + ISCs 110 . Similarly, lineage tracing and clonal analysis in organoids represents a parallel strategy to study human cancer stem cells 111 , 112 . Following Apc loss in mice, Lgr5 + ISCs, but not other cell types, give rise to intestinal adenomas 113 reminiscent of the cell hierarchy of normal intestine 114 and consistent with Lgr5 + ISCs representing a cancer stem cell (CSC) population 115 . Analogously, this mouse lineage tracing strategy was subsequently applied to primary human organoids to demonstrate cancer stem cell activity 112 , 116 . Human LGR5 + cells exhibited multipotency and self-renewal in colon organoids harboring inducible CRISPR/Cas9 knock-in lineage tracing alleles of LGR5 116 , which following transplantation self-renewed and differentiated to generate heterogeneous progeny in tumors 112 . Remarkably, specific ablation of LGR5 + CSCs within tumors did not induce long-term regression of primary tumors, which instead exhibited dramatic plasticity where other cell types replenished the LGR5 + CSC population 112 . Surprisingly, in a parallel transgenic mouse study 117 , ablating Lgr5 + CSCs did not induce primary tumor regression, but formation and maintenance of colon cancer-derived liver metastases was reduced, suggesting that targeting Lgr5 + CSCs could selectively eradicate colon cancer metastases. Upon mouse transplantation of oncogene-engineered colon organoids, the majority of metastases were seeded by circulating Lgr5 − colon cancer cells which could regenerate functional Lgr5 + CSCs, indicating potential prevention of metastases by targeting cellular plasticity 118 . In humans, it remains unclear whether LGR5 + cells serve as functional CSCs in human colon cancer metastases, where LGR5 expression is downregulated 38 , 119 . Similar cellular plasticity of Lgr5 + ISCs has also been described previously in normal mouse intestinal epithelium 120 , 121 . Implicit to cancer stem cell analysis in organoids are cell of origin questions for particular cancers. Indeed, successful oncogenic transformation of a particular lineage in organoids may represent prima facie evidence that a given population serves as a cell of origin for the cognate cancer. A lack of appropriate primary culture systems has previously constrained the exploration of numerous human pathogens. Organoid culture technology has filled numerous emergent needs in the field of infectious diseases pathogenesis. Co-cultures of organoids with pathogens, such as parasites 122 , bacteria 11 , 49 , 123 - 129 , and viruses 130 - 132 , have modeled host-pathogen interactions and pathogenic-induced oncogenesis. Microinjection of gastric cancer-associated Helicobacter pylori into the lumen of normal gastric organoids stimulates a robust inflammatory response 11 and epithelial hyperproliferation 49 , 128 , 129 . A causal relationship between chronic Salmonella enterica infection and gallbladder cancer was revealed in a gallbladder organoid model 124 . The oncogenic Salmonella -infected organoids acquired growth factor independency and exhibited histological characteristics of cellular transformation, including loss of cellular polarity and increased nuclear pleomorphism. Various colorectal cancer-associated bacterial species are enriched in patients, and therefore the intestinal microbiome has been suggested to facilitate tumorigenesis 133 . Accordingly, colibactin-producing pks + Escherichia coli directly induced DNA mutagenesis in a long-term organoid co-culture system. Whole exome sequencing data indicated that organoids exposed to pks + Escherichia coli exhibited a characteristic single thymine (T) deletion signature, as well as increased numbers of single base substitution, preferably T>N substitutions. Interestingly, these genotoxic Escherichia coli -induced mutational signatures were identified in a subset of CRC patients, suggesting a possible cause of tumorigenesis 127 .

Organoid culture technologies have rapidly exerted a transformative impact on cancer research, with widespread adoption for both basic and translational applications. However, despite promising studies to date, numerous challenges remain. Although tumor PDOs often preserve the genetic composition of the original tumor at early passages, the extent of genetic drift, clonal selection and continued tumor evolution at later passage and after freeze-thaw cycles has not been well documented and may be tumor-specific 144 . Similar caveats apply to forward oncogene engineering of normal tissue as well. Differences in seemingly normal cells with regard to telomere length, genetic and epigenetic state could potentially influence phenotypic outcome. In addition, the ordering of mutations in serial genetic engineering, the exact location, alteration and duration of CRISPR knock-out modifications, genetic bottlenecks introduced by selection of rare transformed clones, and presumed ongoing evolution may all produce variable results. This may account for differences in invasive behavior between two recent organoid models of colorectal cancer that engineered the inclusion of the same set of mutations 76 , 77 . Another key limitation of conventional organoid methods is a demonstrated lack of endogenous tumor-associated stromal components, particularly immune cells and fibroblasts. Thus, the continued development of organoid systems that more holistically represent the tumor microenvironment is an urgent need. Ultimately, PDOs incorporating immune and other stromal components may open new research directions such as immuno-oncology and help actualize the promise of precision cancer therapies. Although essential culture components for growing organoids from different normal tissues and cancers have been widely investigated, costs are substantial and reagents can vary between laboratories, abetted by use of conditioned medium containing factors such as Wnt3a, R-spondin, and Noggin. Further, the presence of serum in conditioned media may disadvantage long-term culture 88 . Recently developed artificial water-soluble Wnt agonists 187 or afamin-stabilized Wnt3a 188 may facilitate serum-free alternatives to conventional Wnt conditioned media 4 and enable economical, fully defined and recombinant culture reagents. A practical challenge is reproducing the potency of organoid media across different batches and laboratories, which may provide a selection pressure for a subset of cells that might not be representative of original tumors. At the same time, the evaluation of numerous different culture conditions can improve success rates of PDO generation 144 . Current CRISPR/Cas9-mediated oncogene engineering in primary human organoids has been inefficient and largely restricted to selected genes in which mutational pathway activation allows bespoke media-based functional selection strategies to enrich for gene knockouts. Future approaches may bypass such screening requirements, perhaps by allowing phenotypic screening, increasing efficiency or using selectable knock-in cassettes. Opportunities also exist for large-scale genetic and chemical perturbations of organoids. Improved, highly efficient editing methods 185 , 189 , 190 could facilitate screening of a broader range of target loci, extending to high-throughput genome-scale functional screens 191 , 192 . Similarly, small molecule screening in organoids has been restricted to modest panels of therapeutics, which could be enlarged to more extensive compound libraries by ultra-high throughput miniaturized cultures 193 . The scale of both genetic and small molecule screens could be substantially augmented by methods to expand organoid biomass to much larger degrees than currently practiced. Perhaps the most revolutionary potential applications of organoids for oncology reside in therapeutic prediction. Numerous studies suggest the feasibility of PDO-based functional testing of patient drug responses, which now require prospective confirmation in validation cohorts. Optimally, biologic organoid testing will require rapid turnaround in clinically actionable timeframes, evidence of enhanced survival advantage and conceivably, extension to modalities such as immunotherapy. Additionally, the creation of ever-enlarging PDO biobanks, for example with the HCMI initiative 165 , may establish large collections representing diverse cancer subtypes and genotypes, whose systematic drug screening could establish ready-made therapeutic response patterns that are possibly relevant to patients with similar tumor genotypes. Such transformation of prior descriptive metrics into organoid-based biological predictive assays using living cells would be a capstone of the rapid progress with cancer organoids to date, portending additional advances in both basic biology and clinical translation.

Over the past decades, preclinical cancer biology studies have commonly relied on immortalized 2D cancer cell lines in vitro , as well as xenografted or transgenic animal models 1 . Although these approaches have contributed enormous insights, extensively passaged cell lines may not accurately represent the biology and pathophysiology of the original parent tumor, and animal models are costly and time consuming. Alternatively, newly developed 3D organoid methods now facilitate the robust culture of healthy human tissues and their cognate tumors, thus representing an independent in vitro approach to studying cancer. To be classified as an organoid, the culture should retain the identity of the modeled organ, contain multiple cell types, preserve some physiological aspects of the organ and self-organize according to the same principles as the organ 2 , 3 . Furthermore, organoids maintain several properties of primary tissues, such as self-renewal, multilineage differentiation, signaling nodes and histology. Once established, organoids can often be cultured long-term, expanded, cryopreserved, and genetically manipulated similar to traditional 2D cell lines. As such, organoids combine the tractability of in vitro systems with the 3D architecture and differentiation of in vivo model organisms. Accordingly, recent years have witnessed an exponential growth in the use of organoids as a replacement for immortalized cell lines, allowing the use of primary human cells for the in vitro modeling of biological processes, including normal physiology, stem cell biology and diverse pathophysiologic states. As relevant to this review, the singular attributes of organoids have led to their rapid adoption as powerfully exploitable cancer models.

Precision Medicine is a treatment approach that seeks to exploit patient-specific individualized therapeutic strategies 177 . Given their tractability and potential to capture patient and tumor type diversity, organoids are well-suited for the development of personalized therapeutic approaches. The diversity of cancers captured by organoid biobanks has created numerous applications for therapeutics screening. Such drug sensitivity profiling can be particularly powerful when combined with genomic and/or transcriptomic organoid characterization, identifying specific drug response profiles for different tumor subtypes, or alternatively, defining sensitive and resistant PDOs. PDOs can also be used for synthetic lethality analysis, identifying drugs with selective efficacy against organoids with specific mutation(s). Numerous examples include screening a liver cancer organoid biobank of 6 liver PDOs against 29 anti-cancer drugs, which nominated ERK inhibitors as a potential therapeutic agent for a subset of liver cancers 162 . Similarly, exposure of 27 liver cancer organoid lines from multiple tumor regions of 5 patients to 129 FDA-approved anti-cancer drugs revealed 13 drugs with broad cytotoxicity including inhibitors of histone deacetylase (HDAC), proteasome, DNA topoisomerase II, protein translation and RNA synthesis, as well as several novel agents. Comparison of organoid drug response with publicly available drug response databases demonstrated separate clustering of organoids versus traditional liver cancer cell lines, indicating that primary 3D PDOs may respond to drugs differently than conventional 2D-cultured cells 160 . Systematic exposure of a genetically characterized 66 patient pancreatic cancer PDO biobank to five clinically used chemotherapeutic agents such as gemcitabine, revealed that 33% of PDAC samples were resistance to all five drugs, but half of the multi-resistant lines were sensitive to a targeted therapy when tested against a panel of 21 such agents, suggesting alternative treatment strategies. In this case, correlations between PDO gene expression and drug sensitivity revealed a gemcitabine drug sensitivity gene signature that significantly correlated with improved progression-free survival (PFS) upon post-surgical adjuvant gemcitabine treatment in a 55 patient cohort 150 . Large-scale drug screening of PDOs may also identify unexpected treatments or combinations thereof. Evaluation of gastric cancer PDOs against a 37 compound library has indicated genotype-specific vulnerabilities 140 . Further, generation of matched tumor and normal PDOs from the same patient was exploited to discover the gemcitabine as a potentiator of the EZH2 inhibitor UNC1999 combination in PDAC PDOs 94 . Drug sensitivities between sibling pairs of PDOs and PDXs have been compared in CRC models 141 . In a panel of 46 PDOs and 59 PDXs, the 19 tumors captured in both systems displayed fairly concordant responses to 16 tested drugs. However, different patterns were observed for individual drug sensitivity and RNA signatures for hypoxia, EMT, G2/M checkpoints, proliferation, stemness and metabolism. Since both PDOs and PDXs recapitulate many key genetic and phenotypic features of their parental tumors but possess distinct drawbacks 135 , future studies combining the strengths from these preclinical models could provide clinically relevant metrics in precision medicine. The large number of developing organoid biobanks provides a unique opportunity for targeted drug development 178 . Compound screening against large, well defined organoid repositories has the potential to identify drugs efficient against a subsets of tumors with a specific genetic composition. While such screens have been perfomed extensively for cells grown in 2D monolayers 179 , 180 , emerging evidence suggests that cells grown in 3D often respond differently to drugs 160 , including increased chemoresistance 181 . For example, 2D versus 3D drug responses in ovarian cancer found correlation for cytotoxic but divergence for cytostatic drugs 182 . Conceivably, 3D organoid biobanks could lower false positives in large compound screens, which in turn could reduce the extent of expensive confirmatory preclinical mouse studies. In addition, organoids can be grown from corresponding healthy tissue which would give an early indication of compound toxicity, as well inform underlying biology 94 , 183 . Organoid-based models to fully replace mouse experiments for drug discovery have yet to emerge, but there are encouraging developments including en bloc organoid culture with tumor and infiltrating immune cells 40 and integrated multi-organoid body- on-a-chip systems to study drug metabolism and toxicity 184 . Positive compound hits will still need to be verified in xenograft experiments in the foreseeable future. Overall, current nascent efforts with PDO biobanks represent a promising approach to screening therapeutics against different molecular subtypes within distinct solid tumors. The rigorous genetic characterization of such PDOs, combined with small molecule screening, has potential to identify new patient-relevant gene-drug interactions and synthetic lethal vulnerabilities, yielding novel insights versus traditional cancer cell lines and PDX models. Additionally, the biobank paradigm has broad application beyond cancer, such as with recently established cystic fibrosis organoid resources 185 . Precision Oncology efforts have traditionally exploited predictive mutational biomarkers for targeted therapies. However, such biomarkers are often lacking 145 ; for instance, whole exome sequencing on a mixed cohort of tumors only identified actionable mutations with cognate FDA-approved drugs for 3 of 737 patients (0.4%) 177 . Even when targetable biomarkers are discovered, patients do not always respond to targeted therapy, raising a pressing need for strategies to predict efficacy or identify alternative options 177 . PDO technologies represent a tantalizing opportunity for functional therapeutic response prediction in living cells. Several studies have reported that PDOs can forecast patient anticancer drug responses for a limited number of cases (n5 and <10) 146 , 150 , 158 . Generally, published tumor organoid drug responses correlate with patient outcome. For example, in an organoid biobank of well-characterized pancreatic cancer PDOs eight out of nine organoids responded similarly to the cognate patients when exposed to chemotherapy, giving a match rate of 89% 150 . To date, only a handful of studies have measured patient and organoid drug correlations for more than 10 patients. For example, patient-organoid drug response across different gastrointestinal tumor types and drug regimens indicated the matched organoids could correctly predict responders and non-responders in 20 of 21 patients 142 . Similarly, comparison of metastatic colorectal patient versus PDO responses indicated relative success for irinotecan monotherapy (n=10), and the combination of irinotecan and 5-FU (n=12), but not for a combination of 5-FU and oxaliplatin (n=10). The authors speculated that the failure of PDOs to predict oxaliplatin response may be due to the lack of an immune system and stroma in organoids 145 . Two recent publications have concluded that organoids can accurately portend responses to neoadjuvant chemoradiation (NACR) in rectal cancer. A large study established 80 PDOs from biopsies of treatment-naive patients with locally advanced rectal cancer (LARC) from a phase III clinical trial of neoadjuvant radiation combined with 5-FU with or without irinotecan 147 . Organoids were treated with radiation, 5-FU or irinotecan separately but not in combination. Considering the organoids as sensitive to NACR when sensitive to at least one of the three treatments (radiation, 5-FU, irinotecan), the study correctly predicted outcome in 85% of patients. Since many organoids were only sensitive to single or double agents, the authors hypothesized that not all three therapies are necessary in combination, and that organoid screening can identify ineffective treatments and thus avoid overtreatment. Another study, also studying NACR in rectal cancer, indicated varying sensitivity of rectal PDOs to ex vivo radiation, which correlated with clincal response in a cohort of 19 patients 146 . Taken together, functional precision medicine, where individualized predictions are made based on a cellular drug response test instead of identification of a biomarker, is an intriguing prospect for PDOs. The success rate of establishing cultures in current biobanking efforts varies substantially between tumor types, but can be high (>80%), suggesting the feasibility of organoid-based functional precision medicine. From these few initial studies, some preliminary observations can be made. Despite tumor heterogeneity being a significant potential confounding variable for extrapolating therapeutic response from a single biopsy region, functional prediction appears to be feasible at least in the specific tumor types and drugs tested. However, the current lack of stromal and immune compartments in most organoid-based therapy prediction systems may preclude universal applicability. Negative results have not been described, raising the issue of publication bias, and perhaps most significantly, studies to date have trained predictive metrics based on foreknowledge of clinical outcome. Although current studies depict high (80-95%) rates of concurrency, it will be necessary to observe if initial successes persist in larger prospective independent validation cohorts of patient-organoid pairs where response metrics are fixed in advance.