Integration of Organoids With CRISPR Screens: A Narrative Review

ABSTRACT

Organoids represent a significant advancement in disease modeling, demonstrated by their capacity to mimic the physiological/pathological structure and functional characteristics of the native tissue. Recently CRISPR/Cas9 technology has emerged as a powerful tool in combination with organoids for the development of novel therapies in preclinical settings. This review explores the current literature on applications of pooled CRISPR screening in organoids and the emerging role of these models in understanding cancer. We highlight the evolution of genome‐wide CRISPR gRNA library screens in organoids, noting their increasing adoption in the field over the past decade. Noteworthy studies utilizing these screens to investigate oncogenic vulnerabilities and developmental pathways in various organoid systems are discussed. Despite the promise organoids hold, challenges such as standardization, reproducibility, and the complexity of data interpretation remain. The review also addresses the ideas of assessing tumor organoids (tumoroids) against established cancer hallmarks and the potential of studying intercellular cooperation within these models. Ultimately, we propose that organoids, particularly when personalized for patient‐specific applications, could revolutionize drug screening and therapeutic approaches, minimizing the reliance on traditional animal models and enhancing the precision of clinical interventions.

A general overview of pooled screening in organoids is depicted, noting alterations compared to cell lines. Clockwise from bottom left: (1–2) Tissue sample from the organism is processed for release of progenitors from the extracellular matrix, to be used for culturing. In comparison to cell lines which have accumulated years of epigenetic alterations for adaptation in monolayer culture, fresh samples are better representatives of their source. (3–4) Processed tissue can be isolated for stem cells by fluorescence‐ or magnetic‐ activated cell sorting (FACS/MACS), laser capture microdissection (LCM); or handpicking, but this step is optional. Protocols have been developed in specific cases for processing tissues which dispense the need of sorting. They work cooperatively with the organoid culture condition which is supplemented by a growth factor cocktail that induces proliferation of the stem cell compartment. In an extracellular matrix, this proliferation results in the stem cell generating the different cell types which are present in the native tissue, all of which ultimately self‐assemble into a functional organoid. (5–6) Pooled guide RNA (gRNA) library is delivered, usually through viral vectors, in the established organoid. Now each transformed cell has a unique gene knocked out, and under a particular selective culture condition the fitter cells are expected to proliferate more. An NGS (next‐generation sequencing) at the end of the experiment would reveal the differential enrichment/loss of all gRNAs now integrated in the cells’ genomes. Due to the 3D culture conditions, the effect of cell–cell and cell–matrix interactions, is captured in the differential fitness for each cell. (7–9) The endpoint can be flexibly chosen according to the experiment—expression of a given protein of interest (like a cluster of differentiation, for instance) or a prespecified reporter can be correlated with the gRNA present in that cell, connecting the expression with that particular gene's knockout. The same can be done for phenotypic changes visible through microscopy. Carrying out full transcriptomics of each cell can also be done to study effect of each knockout on gene expression; but this intensive approach is not frequently adopted. At its simplest, the NGS data is simply plotted in order of the fold‐change in number of copies for each gRNA, from the time of delivery to the endpoint. A list of the top n gRNAs, or the genes they correspond to (the “hits”), are selected for deeper study.

1. Introduction

For the uninitiated, the term “organoid” may hold various indications with equally varying intended applications. The essence, arguably, is an in vitro culture developed from progenitors in a growth environment that facilitates the development of a three‐dimensional (3D) configuration. Such a culture needs to be identical to the original tissue in two crucial aspects: presence of a heterogeneous population with cell types from the native tissue and having mechanical/electrical connections with adjacent (homotypic/heterotypic) cells and with the intercellular matrix.

The requirement of an organoid as a model system reflects the requirement of cells to be part of the tissue they make up in order to perform their full suite of physiological functions. This has been the highlight of organoids cultured using intestinal, breast, hepatic, and renal tissues. Though 2D cell lines originating from each of these organs exist, more complex functions are only realized in the corresponding 3D organoids (Clevers 2016). Setting up these structural aspects allows an organoid to exhibit certain physiological functions of the native organ, which individual cells or disorganized spheroids cannot. Mammary tissue organoids, for instance, are capable of collective contraction by the myoepithelial cell subset on oxytocin stimulation, with even the contraction frequency in agreement with in vivo experiments (Sumbal et al. 2020; Mroue et al. 2015). Or, nephrons generated using induced pluripotent stemcells (PSCs) neatly separate into collecting ducts and the glomerulus connected by the proximal and distal ducts when cultured in three dimensions (Takasato et al. 2015). Recreating these structural and functional aspects of the organ, in a plate, is a hallmark of organoids.

Organoid research has gone through a resurgence with the Clevers’ group's landmark culture of crypt‐villus intestinal organoids in 2009, although structured self‐assembly was already demonstrated by Bissel and colleagues as early as 1987. Their mammary epithelial cell cultures grown in Engelbreth‐Holm‐Swarm (EHS) tumor extract basement membrane were among the earliest examples of true 3D structures in the form of ducts and lamina, with functional milk protein secretion, which was unheard of in cultures on plastic dishes. The EHS extract (trade name Matrigel) has since been a staple requirement as the extracellular matrix in the culture of organoids and a template for artificial substitutes including hydrogels (Jubelin et al. 2022).

Current organoid techniques rely on stem cells as progenitors, using PSCs/induced PSCs (iPSCs), adult stem cells (ASCs) and embryonic stem cells (ESCs) cultured with defined cocktails of growth factors for development into the desired tissue. This universality has therefore also led to the inclusion of stem cell progenitors in a definition for organoids (Lancaster and Knoblich 2014) (though the term stem cell is itself possibly niche‐dependent; see Simian and Bissell [2017] for an interesting discussion on definitions). To note, cells derived from tumor specimens of solid tumors or cancer cell lines when cultured in Matrigel also form stable 3D cultures with significant differences from their plastic dish monolayer counterparts. These tumor organoids, or “tumoroids,” display essential traits of cellular heterogeneity and oxygen/metabolic gradients similar to tumorspheres, further improved by presence of extracellular matrix interactions allowing drug resistance and epithelial–mesenchymal transitions (EMT) to occur. Due to their resemblance with the source tumor, tumoroids have always been looked upon as having great potential to carry out real‐time therapeutic experiments for use in patients; the reader is referred to several excellent reviews on the subject (Drost and Clevers 2018; Tang et al. 2022; Veninga et al. 2021; Thorel et al. 2024; Veninga and Voest 2021). A list of established “tumoroid” biobanks from a spectrum of cancers is presented in the Table 1 below.

Table 1.

List of established human cancer derived tumor organoids (tumoroids).

Origin/Type	No. of patients/tumoroid lines	Investigation	Reference
Pancreatic ductal adenocarcinoma	66 tumoroids	Pharmacotyping, i.e., correlating differential inter‐patient drug response to genomics data, intra‐patient differential drug response	Tiriac et al. (2018)
	49 tumoroids	Growth factor dependence and corresponding subtyping, co‐culture with cancer associated fibroblasts, GATA6 knockdowns/knockouts and driver gene engineering of KRAS, TP53, CDKN2A, SMAD4	Seino et al. (2018)
	17 tumoroids	Tumorigenic capacity in vivo, respiration rate measurement, drug screen	Huang et al. (2015)
Urothelial carcinoma	16 patients	Principal component analysis of gene expression data comparing organoid with patient, drug screening	Lee et al. (2018)
	53 patients	Drug screen	Mullenders et al. (2019)
Colorectal carcinoma	20 patients/ 22 tumoroids	High‐throughput drug screen	Van de Wetering et al. (2015)
	41 patients/ 65 tumoroids	Chemo‐radiation treatment	Ganesh et al. (2019)
	40 tumoroids	Genetic and clonal evolution in long‐term culture, growth factor dependence, tumorigenic capacity in vivo	Fujii et al. (2016)
Gastric cancer	34 patients/ 46 tumoroids	Clonal evolution in long‐term culture, drug screening	Yan et al. (2018)
	37 tumoroids	Growth factor dependency	Nanki et al. (2018)
Breast cancer	95 tumoroids	TP53 CRISPR knockout, drug screen, comparison of drug response with in vivo response	Sachs et al. (2018)
	87 tumoroids	Drug screen	Bhatia et al. (2022)
	60 tumoroids	Drug screen	Wu et al. (2024)
Renal cancer	54 tumoroids	High‐res 3D imaging, TP53 CRISPR knockout, drug screen	Calandrini et al. (2020)
	10 tumoroids	Drug screen, in vivo transplantation	Grassi et al. (2019)
Glioma	20 tumoroids	LC‐MS of metabolites, culturing under differential (5% vs. 21%) oxygen conditions	Abdullah et al. (2022)
	53 patients/ 70 tumoroids	Single‐cell RNAseq of organoids and origin tumors, in vivo transplantation and imaging, drug screen, CAR‐T co‐culture	Jacob et al. (2020)
Ovarian cancer	32 patients/ 56 tumoroids	TP53 CRISPR knockout, drug screen, comparison of drug response with in vivo response	Kopper et al. (2019)
	23 patients/ 34 tumoroids	Analysis of factors determining response to DNA damage repair inhibitors	Hill et al. (2018)
	23 patients/ 36 tumoroids	Pharmacotyping, intra‐patient differential drug response	de Witte et al. (2020)
Head and neck cancers	26 tumoroids	Drug screen, radiation therapy	Driehuis et al. (2019)
	13 tumoroids	Drug screen	Tanaka et al. (2018)
	24 tumoroids	Subtype‐specific biomarker discovery	Wang et al. (2022)
Hepatic cancers	48 tumoroids	—	van Tienderen et al. (2022)
	11 tumoroids	in vivo transplant, sorafenib response	Nuciforo et al. (2018)
	5 patients	Intra‐patient differential drug response	Li et al. (2019)
Lung cancers	43 tumoroids	Mechanistic studies of growth factor independence and targeted therapy	Ebisudani et al. (2023)
	16	Drug screen, in vivo transplantation, RSV infection	Sachs et al. (2019)
		Drug screen	Shin et al. (2019)
Cervical cancer	12 tumoroids	Drug screen, in vivo transplantation	Lõhmussaar et al. (2021)
Endometrial cancer	8 tumoroids	Drug screen, differential ion‐channel expression analysis	Boretto et al. (2019)
Prostate cancer	7 patients	Principal component analysis of gene expression data, drug screen	Gao et al. (2014)
Neuroendocrine neoplasia; various organs of origin	20 patients	Growth factor dependence, TP53 and RB1 CRISPR double knockout, overexpression of the transcription factors ASCL1, NEUROD1, POU3F2, NKX2‐5, SOX2, and TP73	Kawasaki et al. (2020)
Mesothelioma	2 patients	Drug screen	Mazzocchi et al. (2018)
Melanoma	9 tumoroids	Immune cells co‐culture, drug screen (chemo and immunotherapy), demonstration of adaptive immunity	Votanopoulos et al. (2020)
Esophageal adenocarcinoma	10 patients	Long term clonal evolution, drug screen	Li et al. (2018)
Chordoma	5 patients	High‐throughput drug screening	Al Shihabi et al. (2022)

2. Pooled (Library) CRISPR Screens: Applied To Organoids

CRISPR/Cas technology is currently the most precise and versatile gene‐editing tool in extensive use today. Originally named after phage‐derived repeat sequences in bacterial and archaeal genomes, contemporary use of the term refers to a suite of ribonucleoprotein enzyme complexes (CRISPR associated proteins, or Cas) having nucleotide sequence recognition ability coupled with effector function. The naturally occurring enzymes and their engineered counterparts cover a wide range of effector functions, including dsDNA cleavage (both blunt or staggered), ssDNA cleavage, RNA cleavage, point mutation of all 12 types, indel generation; and non‐genetic ones like nucleotide methylation/demethylation, histone methylation/acetylation, transcriptional repression/activation, and fluorescence tagging (Nakamura et al. 2021). All aforementioned actions take place on precisely those locations in the genome which carry a sequence complementary to a single‐stranded RNA oligonucleotide fragment (guide RNAs, or gRNAs) complexed with the Cas enzyme, itself having a sequence of the experimenter's choice, giving unprecedented power in genetic engineering.

Library screens were among the first applications of the CRISPR/Cas canonical dsDNA cleavage function. Originally based on small interfering RNA (siRNA) constructs, library screens allowed knockdown of a large number of genes for carrying out unbiased forward screens. Attempts to use TALEN‐based knockouts for the same have also been tried, which come with a modest success rate, particularly for methylated DNA regions (Kim et al. 2013). CRISPR/Cas9‐based knockout screens, combining the best of both siRNA and TALENs, have since captured the field. By introducing dsDNA cleavage, which is repaired by the cell using the non‐homologous end joining (NHEJ) machinery, indels are introduced at the break sites, causing the particular gene to be knocked out. Using a set of gRNAs complementary to sites present in all known protein‐coding genes, whole‐genome level knockouts can be obtained.

The final methodology of a CRISPR library screen is an aggregate of the experimenter's choices at each of the important steps:

1. Library screens begin with defining a set of genes to be targeted, followed by preparation of individual gRNAs, which can knockout their sequence‐specific gene via the CRISPR/Cas9 mechanism. Anywhere from 3 to –20 gRNA sequences can be designed per gene, complementary to differing exonic regions. The final set of all gRNAs, capable of targeting a given list of genes with multifold (usually 4) redundancy, is what’s termed the library. This can be the set of all ~18000 known genes (a genome‐wide library) or a targeted subset (focused library).

2. An arrayed library consists of individual gRNA fragments packaged in separate compartments, like a well plate. This allows handling each gRNA one at a time for delivery into cells of choice, thus allowing to conduct as many separate tests as there are genes to interrogate. However, the very same reasons make the arrayed library both expensive and labor intensive. A simpler version, the pooled library, has the entire collection of gRNA fragments pooled in a single vial, to be delivered into a culture of cells at once.

3. A pooled library works on the observation that a sufficiently dilute library preparation will result in only one gRNA transfected/transduced per cell, while the majority of cells remain untransformed. This is achieved by starting with a cell population 2–5 times higher than the number of gRNAs, and using a selection step (like via antibiotic resistance, or surface markers). This number, the ratio of transforming agents to cells being transformed, is termed the multiplicity of infection or MOI.

4. The transformed cells are allowed to proliferate in a specific selective condition that would confer a differential fitness advantage to each cell depending on the gene knockout they underwent, that is, the gRNA that they received. The selection criterion can be effected in two ways: when all gRNAs that survive the selective process are of interest (termed positive selection) or when all gRNAs that are lost during the process are of interest.

5. After a few population doublings, change in abundance of each gRNA at endpoint (usually measured by sequencing) correlates with the effect on fitness of a cell due to knockout of the corresponding gene. The final readout thus gives an excellent idea of how crucial or redundant (subject to the conditions of cell culture) the function of each interrogated gene is. However, counting differential abundance is only one way of assessing the effect of a gene knockout, and many cellular phenotypes of experimenter’s choice have been used as endpoints. They include single‐cell transcriptomics, cellular differentiation, expression/secretion of a specific protein (Parnas et al. 2015), migration and invasion ability (Prolo et al. 2019), and microscopy‐based selection (Walton et al. 2022), to name a few.

6. Lastly, advances in Cas engineering have enabled multispectral perturbation beyond just knockouts, including gene expression enhancement, interference, and epigenetic activation/silencing (Quintero‐Ruiz et al. 2024; Roth et al. 2020). Broadly, the screen is classified as loss‐of‐function or gain‐of‐function based on the effect of perturbation on the genes.

Genome‐wide CRISPR gRNA library screens have become a standard practice since their arrival a decade ago (Iyer et al. 2020), with the first ones reported in December 2013 (Koike‐Yusa et al. 2014; Shalem et al. 2014; Wang et al. 2014). In organoids, however, the use of library screens is only picking up. Still, this emerging set of studies covers many of the variations in methodology, as described above: whether it is genome‐wide or focused libraries, varying selection conditions and type, and the endpoints chosen. The common elements include pooled screening and using only the double‐stranded break knockout action of Cas9 for editing. But as briefly noted above, the CRISPR/Cas toolbox has been significantly expanded with the development of Cas9 variants. Once the seminal 2012 report led by the collaboration of Charpentier and Doudna brought the decades‐old field of CRISPR from microbiology into engineering, innovations to broaden the range of this technique have exploded (Jinek et al. 2012; Jore et al. 2012). Within a year, two synthetically altered variants of Cas9 were introduced with inactivating mutations in one or both of its two nuclease domains (Qi et al. 2013; Gasuinas et al. 2013). Named nCas9 and dCas9, they were shown to be capable of causing single‐stranded breaks, or nicks, and passive binding without cleaving activity, respectively. Deactivated, or dead, Cas9 (dCas9) was soon fused to transcriptional activators/repressors to allowthe modulation of eukaryotic transcription and thereafter utilized in a screen (Gilbert et al. 2013; 2014). But more importantly, the dCas9 was recognized as a modular component capable of carrying out an assortment of effector tasks, limited only by its binding components. It is the rational design and fusion of binding partners, from biotin ligases to GFP to FLAG‐tags (Anton et al. 2018) that has majorly contributed to the present repertoire of the CRISPR/Cas toolbox.

An important step in the gene editing respect of this toolbox came in the form of base editors, that is, fusion of cytidine/adenine deaminases to dCas9, for carrying out nucleotide transitions without causing double‐stranded breaks (Gaudelli et al. 2017; Komor et al. 2016). The same lab later perfected the idea by introducing prime editing, by fusion of reverse transcriptase to Cas9 nickase. By using a modified guide called prime editing gRNA (pegRNA) that carries both the complementary target sequence and the desired edit, this complex can carry out nucleotide transitions and transversions or even small indels at a given genomic site (Anzalone et al. 2019). Today, both approaches have been used in forward screens for generating site saturation variants of proteins, analyzing VUS (variants of unknown significance) in clinical genomics, studying regulatory regions in the genome, and so on; including, of course, knockouts via nonsense mutations (Cirincione et al. 2025; Hanna et al. 2021). We see that the landscape for CRISPR screening applications in organoids is far more vast than the covered ground.

Existing investigations applying pooled CRISPR screens in organoids as the model system have investigated questions broadly falling in the categories of cancer and development. Developmental biology in general and embryonic/pluripotent stem cell‐derived organoids in particular are not the focus of the current article; hence oncology applications are given slightly more emphasis. An elegant collection of original research utilizing CRISPR‐engineered organoids was summarized in Driehuis’ (Driehuis and Clevers 2017) review, which at the time had no examples of whole‐genome screens to describe (Driehuis and Clevers 2017). Eight years later, we have the authors’ predictions realized with a small but emerging number of articles utilizing a blend of these novel technologies.

2.1. Intestinal Organoids

Ringel et al. worked on the backdrop of the knowledge that TGF‐β signaling, which acts as a tumor suppressor in the colorectal cancer context, becomes ineffective in checking tumor growth even when no mutations in known TGF‐β pathway components are present. In order to find mutations in genes not part of the TGF‐β pathway that confer resistance to TGF‐β inhibition, the Brunello library (19,114 genes targeted with 76,441 gRNAs) was delivered into an organoid single‐cell suspension derived from patient biopsies. Following up with a bottom‐up approach, screens were repeated in organoids engineered with APC knocked out, knowing it as an early mutated gene in the genetic progression (“Vogelgram”) of colorectal cancer. The results were able to generate a list of validated hits (ARID1A, SMARCA4, CNIH4, KEAP1, NBAS) fitting the context criteria: genes whose dysregulation conferred TGF‐β resistance, genes requiring tandem APC mutation for conferring resistance, and genes not themselves a part of the TGF‐β pathway (Ringel et al. 2020).

Resistance to TGF‐β agonists as the selection condition post gRNA delivery was also used by Michels et al. in a focused library screen on sigmoid colon organoids. While the number of genes interrogated in the training library was much smaller (100), 20 gRNAs were used per gene (the usual practice being 4–5 per gene) (Bock et al. 2022; Hart et al. 2017). In the discovery screen, a library of 85 tumor suppressors commonly mutated in solid neoplasia was delivered in organoids engineered with both APC knockout and KRAS G12D lesion. However, this screen was selected in the in vivo setting via xenotransplantation. In addition to known players in CRC, the group identified CDKN2A, PBRM1, and ZFP36L1 as novel components (Michels et al. 2020). In the above experiments, organoids set up from normal colon/intestinal tissues provided a tabula rasa milieu for selection of TGF‐β resistant clones in both wild type and APC knockouts. This represents an important merit for organoids compared to cell lines when investigating early‐stage neoplasia. This bottom‐up approach is being used for exploring pathways implicated in tumorigenesis, using normal organoids mutated with commonly observed drivers in early stage cancers, as described. A clinical extension of these studies that use early stage tumoroids instead follows naturally.

An interesting selection criteria was used by Hansen et al. when screening organoids grown from fetal intestinal progenitors, deserving mention while in a review with ASC and tumor organoids as the primary focus. Using a comparison of single cell RNA‐seq datasets from fetal enterospheres and adult organoids of mouse origin, a custom library of 167 transcription factors and 59 epigenetic modifiers, which were found to be upregulated in the fetal organoids, was prepared for CRISPR screening. The authors had hypothesized that this set should contain factors that actively inhibit the maturation of these cells and whose knockout would thus lead to their differentiation into assuming adult‐organoid like characteristics. The screen successfully identified validated hits Smarca4 and Smarcc1 with the most prevalent gRNAs in cells with reduced fetal markers SCA1 and cKIT. Smarca4/Smarcc1 knockout fetal enterospheres also differentiated into a phenotype leaning more towards small intestinal epithelia compared to non‐targeting control (Hansen et al. 2023). The authors further speculate based on their results that Smarca4 and Smarcc1 might not be the only factors inhibiting maturation into adult cell types, hoping that conducting a wider screen would shed light on the additional regulators (Hansen et al. 2023).

Finally, a focused library consisting of 7210 gRNAs targeting 1800 transcription factors (plus 100 negative controls) was employed by Lin et al. in colon and intestinal organoids derived from patient tissue (Lin et al. 2023). Aiming to delineate transcriptional control of the expression of NEUROG3, a master regulator of enteroendocrine development, the authors found the less‐known transcription factor ZNF800 as a potent repressor of enteroendocrine cell differentiation. Post‐validation by ChIP‐seq also established NEUROG3, along with PAX4, INSM1, and SOX4 (themselves present as hits in the CRISPR screen) as the downstream binding partners of ZNF800 (Lin and Clevers 2024).

2.2. Gastric Organoids

In a similar approach to using TGF‐β agonist nonresponsiveness as selection criterion, Murakami et al. used Wnt activation independence as the selection criterion post a whole‐genome CRISPR screen on murine gastric epithelial organoids. On transferring normal organoids to “low‐Wnt” culture conditions after transduction by the GeCKO A library (20,611 genes with 3 gRNAs each), a list of hits was obtained having no prior annotation to Wnt‐dependent epithelial turnover. In the list, Bclaf3 and Prkra, whose expression in mouse gastric glands around the stem cell niche was checked by immunohistochemistry, were validated to reveal Prkra as mediating apoptosis through phosphorylation of Eif2s1 proteins. The authors end with a note on the potential role of the newly uncovered factors in gastric cancer (Murakami et al. 2021).

All above studies used normal organoids sourced from murine or patient tissues. One example of a true “tumoroid” application, sourced from surgical resection of gastric cancer patients, is reported by the group of Stange and Weitz (Mircetic et al. 2023). The group had previously set up cultures using resection specimens from 20 gastric cancer (GC) patients, of which the most prevalent subtype (chromosomally instable or CIN) was chosen for the screen. Their focused library included epigenetic modifiers in cancer stem cells, with added gRNAs relating to important pathways implicated in GC. A total of 95 genes, excluding controls, with 4 gRNAs per gene, out of the 18 gRNAs significantly depleted after 2 weeks, four were judged to be not commonly essential, and thus subjected to deeper validation. Comparison of knockouts for these four genes (KDM1, UHRF1, CHD8, CARM1) in GC tumoroids versus normal gastric tissue organoids revealed KDM1A and CARM1 knockouts to have tumor‐specific effects. Following up with a series of experiments led to the elucidation of NRDG1 as a major tumor suppressor activated in KDM1A knockouts, a mechanism for its transcriptional repression by KDM1A, and downregulation of Wnt signaling and cell cycle progression that NRDG1 induces. Encouraging though the results are, the authors did stress on delineating the downstream factors of NRDG1 that would explain their observed cases of its inhibition failing to reduce tumor viability.

2.3. Hepatic Organoids

The liver holds a unique spot not only as the master gland involved in metabolism and detoxification, but also its striking regenerative capability. An early screen in organoids sourced from murine livers was reported by Planas‐Paz et al. back in 2019, which planned to investigate an aspect of liver regeneration called ductular reaction. Since expansion of cholangiocytes in the liver is central to ductular reaction, organoids composed of cholangiocytes provided a natural system to probe via CRISPR screen with the growth (or lack thereof) of the organoid as the measured endpoint. Screening 192 genes (5 gRNA per gene) holding relevance in liver regeneration and correcting for essential genes resulted in Yap1 showing up as a significant hit. Genes in the mTORC1 (Mtor, Raptor) and the Hippo complex (Sav1, Lats1) were also found essential for growth. Though Wnt/βcatenin was one of the pathways that emerged to be significant in the screen, follow‐up validation experiments showed its dispensability for ductular reaction (Planas‐Paz et al. 2019).

Zhao and Wei's group, also working on cholangiocyte organoids derived from murine livers, used cholangiocyte‐to‐hepatocyte differentiation instead of growth as the endpoint. Interrogating 79 transcription factors/chromatin remodelers using a library of 236 gRNAs, the authors employed single‐cell RNA‐seq as the readout for gRNA enrichment. This single‐cell CRISPR (scCRISPR) screen was able to identify c‐Fos and UBR5 as inducers of hepatocyte differentiation/maturation in intrahepatic cholangiocyte organoids derived from both murine and human origins. The findings hold relevance for advances in liver regeneration without the requirement of healthy donors (Liang et al. 2023).

2.4. Breast Tumor Organoids

Rowdo's recent report on a kinome‐wide CRISPR screen in tumoroids derived from triple‐negative breast cancer patients is a rare, possibly only, study aimed at exclusively tumoroid application of both small molecule inhibitor drugs and broad‐range CRISPR screens. The authors utilized two tumoroids from their organoid biobank for a kinome‐wide screen followed by validation using small molecule inhibitors of the obtained kinase vulnerabilities. Its straightforward translation potential of suggesting combination kinase inhibitor therapies, in patient tissue sampled from an underrepresented community, makes for a notable highlight in organoid research.

The screen itself covered 482 kinases with around 6 gRNAs per gene in conditions of EGFR or MEK1 inhibition using either gefitinib or trametinib, respectively. Some of the kinases identified in this screen were tested in 10 other tumoroids from their biobank, yielding four (ILK, CDK2, GAK, and GSG2) of them as sensitive targets. A point of interest is always the identification of characteristics missed by experiments in 2D; here too, the authors report obtaining kinases including EGFR, CDK13, GSG2, and S6K1, which were not identified in screens on cell lines. The authors also carried out a high‐throughput drug screen of 156 compounds, plotting synergy scores for the more efficacious combinations across a range of concentrations. In summary, FGFR‐EGFR and RAF1‐MEK1 combined inhibitions, as informed by CRISPR screening in the presence of EGFR and MEK1 inhibitors, were validated as synergistic blockades.

The design provides an unbiased method for finding combination therapies in kinase inhibitors, which (to the best of the authors’ knowledge) has not been leveraged in tumoroids before. Even setting aside important caveats of reproducibility and relevance of the model system used (Trembath and Spanheimer 2025), using cellular states with mutational and epigenetic landscapes that are wholly distinct from the “canonical” ones represented by cell lines is an important step towards gauging the diversity of tumors responding to particular combination therapies, which might not be observed in cell lines. In light of its limitations, however, using kinome screens in tumoroids to inform therapy directly is still uncertain.

2.5. ESC and PSC/iPSC Derived Organoids

Though similar from the technician's perspective, embryonic/pluripotent stem cell‐derived organoids show important differences in their fundamental biology compared to their ASC counterparts, making them suited for nonoverlapping applications. These organoids take months to set up, using a multistep protocol for sequential reprogramming of cells. With the last step being differentiation into the desired tissue, the organoid assumes a finite size. While ideal for studying embryonic/postembryonic development, these characteristics make ESC/PSC derived organoids less suited for modeling cancer and quite inappropriate for personalized oncotherapy applications (Schutgens and Clevers 2020). On the other hand, modeling organs with inappreciable regenerative capacity has ESC/PSCs as perhaps the only source for organoids. Brain and kidney organoids created this way have been probed using pooled CRISPR screens. While not directly in this review's focus due to the wholly distinct nature of their inquiry, the genome‐wide screen in kidney organoids by Ungricht et al. and the targeted screens on brain organoids by Knoblich and colleagues deserve mention for the technical milestone.

This review has attempted to collect all existing reports on library screens in organoids, but excludes a number of articles fulfilling the broad inclusion without lining with the narrative's focus. Studies on murine trophoblasts (Mao et al. 2023), cerebral and kidney organoids generated from human embryonic cell lines and pluripotent stem cells (Esk et al. 2020; Li et al. 2023; Ungricht et al. 2022), assembloids (Meng et al. 2023; Onesto et al. 2024), and so forth; fall in this group. Smaller screens, like by Hendriks et al. (2021), were also filtered out. Adopting a technical definition for an organoid based on structure and including all cultures falling under the criteria (even ones not using the term “organoid”) would be the ideal method, which we have not utilized. Additionally, sources other than peer reviewed journal articles were ruled out, including conference abstracts and preprints (Buckhaults et al. 2023; Cramer et al. 2023; Wangmo et al. 2022; Sun et al. 2022). Method articles not on the trail of specific biological questions were too, excluded (Inglebert et al. 2022; Price et al. 2022).

3. Challenges and Mitigations

A number of the above publications report important challenges faced during organoid screening, as compared to simpler 2D models. Heterogeneous growth rate is a primary difficulty inherent to organoids, which skews results of a pooled screen obtained by bulk sequencing. This happens because of the fittest clones expanding disproportionately, resulting in artefactual enrichment of the gRNA they carry. A figure beautifully demonstrating the comparison of growth rates, homogenous in 2D and heterogeneous in 3D, is included in the Knoblich and colleagues’ report. The comparison is made possible by lineage tracing, which involves use of unique oligo sequences downstream of the gRNA to be delivered that act as unique barcodes. While Esk et al. used a double barcoding approach (termed CRISPR‐LICHT), one each for every gRNA and every cell, Michels et al. used a single (gRNA) barcode coupled with manual elimination of the largest outlier clones. Ringel et al. solved the problem by individual organoid sequencing, instead of bulk sequencing, at the experiment endpoint. The order in which the three methods are listed roughly follows in decreasing order, the robustness they provide, but also the cost of increased sophistication and analysis requirements.

A second limitation not necessarily exclusive to organoids is the introduction of multiple gRNAs into a single cell. It is expected that over time, as protocols for this young technique become standardized, the issue will fade. However, today it leads to the exclusion of (potentially misleading) data, as in the study by Liang et al., who omitted all such cells containing functional plus passenger gRNAs from downstream analysis. Recently, a technique that combines Cre‐LoxP recombination with CRISPR has enabled expression of a single gRNA per transformed cell, which was utilized to carry out a small in vivo screen. Including all gRNAs of the library in a single open‐reading frame on the plasmid, successively sandwiched between loxP sites, restricts Cre‐mediated recombinations to a single productive event where only one of the gRNAs ends up adjacent to and downstream of the promoter (Tang et al. 2022).

Technical limitations and neutral drift make data from organoid screens noisier than in 2D cultures, a complication that requires more than twice the coverage to ameliorate. In other words, more than twice the number of cells are required as input for one replicate when they are part of organoids when robust CRISPR screening is desired. Using a positive‐selection design for the pooled screen reduces, to some extent, false‐negative hits obtained on account of drift. To note, a positive‐selection design was not chosen for some of the aforementioned screens.

Lastly, computational tools that are used to predict the activity of a given gRNA sequence do not provide accurate results when organoids are the target system, as shown by Michels et al. In their specific case, a hands‐on check of all gRNAs for TGFβ‐receptor II revealed better agreement of efficiencies between the colon organoids and the HepG2 cell line, prompting their recommendation of prescreening in heterologous cells for selecting the best‐performing gRNAs for each gene. Now, the activity of a gRNA can differ based on the chromosomal location it is designed to target and the probability of the double‐stranded break it generates being repaired by the NHEJ mechanism. Adopting a base/prime editing approach can circumvent the issue but is currently restricted by efficiency limits of its own (Richardson et al. 2023). Efforts to improve the tool on this aspect to allow robust applications, are underway (Chen et al. 2021).

Being a developing technique, factors like lack of standard 3D culture models, lower consensus in practices concerning comparison of studies (Duval et al. 2017), unsuitability of most analysis and validation techniques (light microscopy, for example) used in 2D cultures for adoption in 3D (Jubelin et al. 2022), problems on reproducibility (Lehmann et al. 2019) are part and parcel of the field. It is practical to use simple systems for a tool like library screening that yields highly context‐specific results. Concurrently, the information and predictive power gained for obtained hits stays limited due to differential fitness being the only quantity as the final readout. However this practical recommendation of carrying out broad screens in simpler 2D models and saving organoids for the validation experiments is yet to be rigorously tested. For example, the series of papers Kinase Requirements in Humans I–V represent a noteworthy study elucidating the reproducibility and precautions concerning kinome‐wide siRNA screening on cell lines and primary cultures. Their results provide yet another concrete meaning to the oft‐cited “relevance” of cell lines as model systems: a lack of overlap between hits obtained on screening identical cancer cell lines from a different origin. Meta‐analyses have been done on CRISPR screen hits also with the broad motivation of mapping out the essential genes in human cells, with results analyzed taking into account factors like coverage, gRNAs per gene, and number of replicates (Hart et al. 2017). The Kinase Requirements I article observes three situations resulting in highly overlapping hits: (i) in replicates of same cells, (ii) isogenic cell lines differing by one gene, and (iii) fresh primary cultures originating from the same epithelial tissue (Grueneberg et al. 2008). Thus, what is sorely lacking is a study conducted with the same objectives and spirit aimed at gauging the relevance of hits obtained on a CRISPR/Cas9 screen performed on tumoroids.

4. Hallmarks and Cooperation

While it is relatively straightforward to compare a normal organoid to its native organ, measuring concordance of tumoroids to the tumor of origin lacks the baseline of default tumor “function” in the human body. Here the cancer hallmarks concept provides a useful template for gauging the tumor resemblance of an organoid, especially in comparison with other model systems. For example, glioblastoma organoids were shown to display immunosuppressive characteristics by Riemenschneider and Proescholdt's group, in two separate ways (Braun et al. 2023; Schuster et al. 2024). Also, co‐cultures were adopted by Biffi et al. (2019) and Trumpi et al. (2018), to demonstrate inflammatory signatures in pancreatic ductal adenocarcinoma tumoroids and tumor budding by invasion mechanisms in colorectal tumors, respectively. Currently, the only universally standard characterization techniques performed on tumoroids established from diverse tissues and research labs are microscopy based (Zhao et al. 2022). Sequencing techniques including whole genome, whole exome, transcriptome, and targeted panels, are a close second. Some articles report profiling of the secretome to demonstrate a tumorigenic microenvironment (Atanasova et al. 2023; Lorenzo‐Martín et al. 2024); to our knowledge this method is not prevalent for measuring source fidelity of tumoroids. Here, methods to validate tumoroids by demonstration of functional cancer hallmarks capabilities may provide for key “rule‐in” tests of resemblance. It has long been appreciated that many hallmarks of cancer could be exhibited by, or even necessarily require, intercellular cooperation, especially in cases like immune evasion and drug resistance (Aktipis et al. 2015; Axelrod et al. 2006; Tabassum and Polyak 2015). Emergent behavior from the tumor collective, including stromal, immune, and endothelial components cannot be investigated in homogeneous monolayer models, necessitating the adoption of tumoroids.

Library screens in patient‐derived cells performed before the establishment of organoids with added cell populations have the potential to analyze hallmarks of cancer resulting from cooperation at the tumor level as opposed to individually at the cellular level. Axelrod and Pienta's theoretical report provided, way back in 2006, a beautiful compilation of the predictive domain, the definitive experiments, and novel questions pertaining to intercellular cooperation in tumors. Somewhat unsurprisingly, the experiments they suggest to falsify the cooperation hypothesis mention only cell lines. A number of studies (not all utilizing true organoids) investigating heterotypic interactions do exist, reporting effects on drug resistance (Lovitt et al. 2018), invasion (Trumpi et al. 2018), or inflammation (Biffi et al. 2019), to name a few. Yet this (to the best of the authors’ knowledge) largely unexplored field of employing organoids for studying cooperative interactions in tumors has much to offer.

The formation of the Human Tumor Atlas Network (HTAN) in 2018 has begun work on the long‐uncharted problem of connecting the histological with the molecular features of solid cancers. Institutes part of the HTAN network have started generating multimodal data capturing the 3D structure of tumor samples (Johnson et al. 2022; Lin et al. 2023; Mo et al. 2024), which most of the above described studies have not had the benefit of consulting. The quality of this up‐and‐coming data, however, makes it clear that the next phase of tumoroid research will be grounded in well‐formed principles of model fidelity, reproducibility, and relevance. Concurrently, CRISPR/Cas systems have too since evolved to incorporate diverse effector functions using both modifications of the original Cas9 enzyme and other enzymes of the Cas family, with ever‐improving techniques. Together, these technologies promise a hitherto unknown understanding of the solid tumor system in the days to come.

5. Conclusion

The advancement of organoid technology paired with CRISPR/Cas9 technology has opened new avenues for research, enabling explorations of more detailed tumor biology and therapeutic responses in in vitro models. Organoids provide a more physiologically relevant model compared to conventional 2D cell lines by replicating the complex structure and cellular interactions found in native tissues. This is particularly important in oncology, where the tumor heterogeneity and its microenvironment play a crucial role in disease progression and treatment response. The ideal utilization of tumoroids has been a personalized drug library screen on cultures propagated from the samples for each patient in need. The two immediate hindrances are low success rate and long durations needed for culture establishment. Despite these advantages, several challenges remain in using CRISPR/Cas9 screening in organoids, as standardized protocols and characterization techniques are required to ensure reproducibility and comparability across studies. Notwithstanding the status of organoids as “intermediate” to cell lines and in vivo models, rational use can skip the need for validation in animal models when their human origin is the decisive factor, as highlighted by the FDA Modernization Act 2.0 (Wadman 2023). It is more pragmatic than optimistic to predict a future where facts based on research on organoid models will be sufficient, both for informing clinical trials or personalized therapy and for in itself. As research continues to evolve, it is essential to collaborate with interdisciplinary teams, combining expertise in cancer biology, bioengineering, and computational analysis. Such efforts will not only enhance our understanding of cancer dynamics but also facilitate the translation of findings into clinical applications. The potential of organoid models, especially when utilized as models to resemble patient biology in drug screening, could revolutionize personalized medicine, ultimately improving outcomes for patients with cancer.

In conclusion, organoids stand at the forefront of a transformative era in cancer research, linking basic science and clinical practice. By addressing current challenges and leveraging the strengths of this innovative technology, we can open up new therapeutic possibilities and advance our understanding of cancer biology in unprecedented ways.

Conflicts of Interest

The authors declare no conflicts of interest.