Technologies to Assess Drug Response and Heterogeneity in Patient-Derived Cancer Organoids

Abstract

Patient-derived cancer organoids (PDCOs) are organotypic 3D cultures grown from patient tumor samples. PDCOs provide an exciting opportunity to study drug response and heterogeneity within and between patients and this research can guide new drug development and inform clinical treatment planning. We review technologies to assess PDCO drug response and heterogeneity, discuss best practices for clinically relevant drug screens, and assert the importance of quantifying single-cell and organoid heterogeneity to characterize response. Autofluorescence imaging of PDCO growth and metabolic activity is highlighted as a compelling method to monitor single-cell and single-organoid response robustly and reproducibly. We also speculate on the future of PDCOs in clinical practice and drug discovery. Future development will require standardization of assessment methods for both morphology and function in PDCOs, increased throughput for new drug development, prospective validation with patient outcomes, and robust classification algorithms.

1. INTRODUCTION

Patient-derived cancer organoids (PDCOs) are 3D cultures grown from clinically derived patient tumor samples (e.g., from biopsy or surgical resection). These cultures retain the cell–cell communication, 3D architecture, and molecular features of the original tumor, making them attractive for modeling human cancer. PDCOs offer several advantages over traditional models of cancer, such as immortalized cell lines grown in 2D and patient-derived xenografts (i.e., patient tumors grown in mice). For example, PDCOs are successfully established quickly, at lower cost, and from a greater proportion of patients than patient-derived xenografts (1–3). Immortalized cell lines are established from selective advantages of growth rate, background genetics, and prior therapeutic resistance, which do not represent the variability of clinical disease (4). Recently, protocols have been established for PDCO culture and media conditions across cancer types (5–7); these protocols have contributed to the growing popularity of PDCOs in cancer research, with the potential to one day guide clinical decision-making.

PDCOs matched to individual patients have accurately predicted patient response to treatment across multiple tumor types and treatment regimens (8–14). This outcome provides an exciting opportunity to rationally design treatments for individual patients on the basis of measured response from their own tumor cells. This is one key advantage of PDCOs compared to current clinical tools to manage cancer treatment, including circulating biomarkers and whole-body imaging (Table 1). PDCOs have also accelerated cancer drug development by providing multiple patient-matched samples to test novel combination treatments that are difficult to screen in clinical trials and by identifying patient cohorts that are mostly likely to benefit from these treatments (8–14). To date, PDCOs have not been integrated into prospective clinical decision-making, partially due to a lack of technologies to measure PDCO drug response quickly and accurately in a clinical setting. Another bottleneck is the time and expertise required to culture PDCOs, although these protocols are becoming standardized (15–17).

Table 1.

Advantages and disadvantages of PDCOs and current methods to manage cancer treatment in patients

Technique	Advantages	Disadvantages
PDCOs	Scalable for testing across multiple therapies Assesses subclonal heterogeneity Identifies resistance mechanisms for clinical treatment planning and new drug development	Requires additional tissue sampling Time and training required to generate cultures Emerging technique that requires standardization to guide clinical decisions
Circulating biomarkers (e.g., carcinoembryonic antigen, ctDNA)	Serial sampling over treatment course (i.e., dynamic biomarker) Early detection of response for effective therapies Clinically available with commercial CLIA-approved assays	Cannot reliably predict response Uncertain predictions if there is subclonal heterogeneity Varied sensitivity during active treatment
Cross-sectional imaging (e.g., MRI, CT)	Established gold standard for clinical decisions Response standard used for clinical trial design	Cannot predict success of alternative treatments Inconsistency when assessing response in heterogeneous tumors No guidance on molecular basis of therapeutic resistance

Abbreviations: CLIA, clinical laboratory improvement amendments; CT, computed tomography; ctDNA, circulating tumor DNA; MRI, magnetic resonance imaging; PDCO, patient-derived cancer organoid.

Several prior reviews have described PDCOs for modeling human cancer (18–20). Here, we evaluate PDCOs from an engineering perspective, with a focus on improved technologies to study PDCOs, quantitative analysis of drug response and heterogeneity, and clinically relevant drug screens for 3D cultures. The implications of these engineering considerations are discussed for applications in patient treatment planning and drug discovery. We also provide our opinions on the future of PDCOs, their use in the clinic and research labs, and new assessment techniques. Overall, we argue that PDCOs offer a wealth of patient-specific data that are not fully exploited with existing assessment technologies and that deeper insights into cell-level and organoid-level behavior within PDCOs can reveal more effective cancer treatments (Figure 1).

Figure 1.

Overview of improved methods to assess drug response and heterogeneity in PDCOs. Abbreviations: PDCO, patient-derived cancer organoid; OMI, optical metabolic imaging.

2. PATIENT-DERIVED TUMOR ORGANOIDS MODEL THE ORIGINAL PATIENT TUMOR

For context, we provide an overview of our own PDCO generation protocol that achieves high success rates for viable PDCOs across multiple patients and tumor types (21). Briefly, PDCOs are generated from human tumor samples obtained from needle biopsy or surgical resection. Tumor samples are placed immediately into a chilled chelation buffer and placed on ice to maintain tissue viability and integrity during transportation to the tissue culture laboratory. Samples are washed and then placed in a digestion buffer (collagenase and dispase) and shaken vigorously to dissociate cells. After digestion, the samples are pipetted vigorously to break them up further and are spun, and the pellets produced are washed to remove any residual digestion buffer. Next, the pellets are resuspended in media and mixed 1:1 with cold Matrigel (a gelatinous protein mixture that mimics the extracellular environment of tissue), then pipetted into each well of a prewarmed 24-well plate. The dishes are incubated for 2–5 min at 37°C to allow the Matrigel to solidify, then the plates are flipped upside down and incubated for at least 20 min at 37°C. This allows the Matrigel to form a hanging drop in which the PDCOs can grow three-dimensionally. After incubation, the plates are flipped right side up, the droplets are covered with media, and the plates are returned to the incubator. The PDCOs are fed with fresh media every 2–3 days. Details on the media conditions for each cancer type are overviewed in prior work (21).

2.1. Comparison of PDCOs with their Original Tumor

PDCOs maintain the key histology and 3D structures of the original tumors, including crypts and epithelial structure (Figure 2) (21). Additionally, surface markers specific to tumor types (e.g., synaptophysin for neuroendocrine tumors), along with functional features (e.g., mucous formation in mucinous tumors) of the original tumors, are faithfully recapitulated in PDCOs (Figure 2). These structural consistencies coupled with pairwise molecular profiling provide confidence in the organ-specific modeling capabilities of PDCOs. This validation is important to avoid contamination from healthy tissue or the tissue microenvironment, as has been previously reported [e.g., in >50% of primary lung cancer organotypic cultures (22)]. PDCOs have also successfully modeled some tumor types that were otherwise difficult to grow in culture, such as neuroendocrine tumors that have exceptionally low proliferation rates that cannot be recapitulated with cell lines or animal models. Neuroendocrine tumor PDCOs mimic the low proliferation rates of the original patient tumors, enabling drug testing on patient samples for this cancer type (Figure 2) (23).

Figure 2.

PDCOs maintain key histology of original patient tumors. (a) H&E-stained tumor sections and whole mounts of PDCOs generated from the tissue that was adjacent to that shown in the tumor section (colorectal cancer). (b) Colorectal cancer PDCOs develop cryptlike structures reminiscent of malignant glands within the tumor. Outlined areas in panels a and b are enlarged in panels to the right. (c) PDCOs generated from mucinous adenocarcinomas also produce mucin. (d) GEP-NET–specific stain using synaptophysin for a tumor slice (top) and PDCO (bottom) from a single patient. (e) Percentage of Ki67-positive cells assessed from GEP-NET PDCOs and the patient tissue from which they were derived. Abbreviations: GEP-NET, gastroenteropancreatic neuroendocrine tumor; H&E, hematoxylin and eosin; PDCO, patient-derived cancer organoid. Figure adapted with permission from Reference 21 (panels a, b, and c) and Reference 23 (panels d and e).

Importantly, PDCOs maintain the driver mutations and subclones of the original tumors; these features are often found with higher purity in PDCOs due to the selection of tumor cells (and elimination of many stromal cells) in the culture process (Figure 3). Prior studies have shown that driver oncogenic alterations are shared in ~90% of patient-derived organotypic models with similar copy-number variants, and nonshared mutations are entirely passenger mutations (9, 13, 21, 24–27). In colorectal cancer, the major molecular subtypes are represented within a so-called living biobank that can be used to model a diverse population set (24). Subsequent work established PDCO biobanks that capture the molecular diversity and culture-specific heterogeneity of esophageal (28), ovarian (29), hepatic (30), pancreas (31), breast (32), endometrial (33), and renal cancers (34).

Figure 3.

Cancer hotspot next-generation sequencing confirms consistency between each patient’s bulk tumor sample and the associated PDCOs. (a) The nonsynonymous mutations were similar between the tumor and PDCOs for the microsatellite-stable cancers. Alterations in known driver genes were identical between the PDCOs and the adjacent cancer, except for DC47 in which a subclonal PIK3CA E545K mutation was found that was unique to the PDCOs and the mismatch repair-deficient DC46 cancer that had additional alterations in APC (*) and KRAS (**) that were not found in the bulk tumor sample. (b) To examine the prevalence of subclonal populations with the PDCOs, first the allele frequency of known alterations was examined. The allele frequencies were approximately 50% for the founder driver oncogenes and 100% for founding tumor suppressor genes across the samples queried (black bars). For each sample, those alterations with an allele frequency of 10–35% were identified (gray bars); these indicate the presence of subclonal populations within the PDCOs. Abbreviations: D, deficient; DC, colorectal cancer; DP, pancreatic adenocarcinoma; MMR, mismatch repair; P, proficient; PDCO, patient-derived cancer organoid. Figure adapted with permission from Reference 21.

2.2. Outlook for PDCO Models

Successful protocols for PDCO generation were iteratively determined through multiple, independent studies that focused on different tumor types (24, 28–34). Many of these protocols have now been standardized, but access to PDCOs is still a bottleneck for their widespread use across cancer research labs. Given the critical role of PDCOs in the future of cancer research and patient care, there is a need for a biobank of PDCOs with matched clinical findings and patient history. To address this need, the National Cancer Institute has curated a patient-derived models repository resulting in >180 PDCO lines that scientists can access (https://pdmr.cancer.gov/). These banked samples are well annotated with patient history, treatment outcomes, and histological findings. Well-defined protocols for culture and media conditions are also provided for each sample. As this and other biobanks (35) build, more diversity in cancer types, treatment outcomes, and driver mutations will be important for robust PDCO studies that motivate new clinical trials and improve cancer treatments.

Although tumor cell purity in PDCOs is advantageous for identifying subclones and driver mutations, loss of the surrounding nontumor cells (e.g., endothelial, stromal, and immune) limits studies of cell–cell interactions and their effect on tumor progression and drug response. Alternatively, PDCOs can be generated from macrosuspensions of the tumor instead of single tumor cell isolations, which maintains some of the native matrix and nontumor cells within the PDCO (36).

Additionally, PDCO generation is challenged by the limited availability of viable tumor tissue from research sampling, and cultures must be propagated to microliter scales to assess therapeutic response with standard techniques, resulting in excessive reagent waste with culture expansion (37). To overcome these issues, a recent report integrated rapid culture expansion with plating on superhydrophobic microwell array chips in lung cancer to perform response assessments at a nanoliter scale (38). Encouraging growth was confirmed in 95% of cultures. Other microscale models have shown consistent therapeutic response in pancreatic cancer for rapid drug screening (39). These emerging methods could address the challenges of tumor sampling and reagent use but require standardization and normalized growth characteristics for widespread use.

3. METHODS OF RESPONSE ASSESSMENT

There are no predefined standards for assessing therapeutic response in organoids, so multiple assays are used for different contexts. Techniques for high-throughput drug screening in PDCOs have been recently been reviewed with details of prior drug screens, plating techniques, experimental timing, and technical components of screen preparation (5). Repeatability and reproducibility are key considerations for drug response assays in PDCOs, including standardization of biologic and technical replicates. Here, we focus on quantitative metrics of PDCO response, relevant drug doses and scheduling for PDCO screens, and considerations for clinical translation of successful drug screens.

The current standard for response assessment in PDCOs is ATP-based viability assays (i.e., CellTiter-Glo), which rely on a luminescent reporter (9, 10, 30, 40). CellTiter-Glo was developed for 2D samples and has been refined with 3D-specific detergents to maximize cell lysis. The advantages of CellTiter-Glo are its robust signal and compatibility with plate readers for high-throughput applications. However, consistent platting between wells is required, creating challenges with matrix-based cultures due to difficulties in transfer using automated liquid handling systems (5). Additionally, response is assessed through dose response measurements that are often outside the clinically relevant dose range.

Growth is of particular importance when interpreting inhibitory concentration (IC₅₀) and area under the curve at the end of a cell viability experiment. For this reason, a baseline assessment of activity including the day 0 viability signal is warranted to track metrics of growth rate inhibition (41). Responses should be scored for each PDCO line individually to account for the signal-to-noise ratio from solvent positive controls to standardized negative controls (i.e., staurosporine or cycloheximide).

Well-level assessments, including CellTiter-Glo, fail to capture the contributions of inter- and intraorganoid heterogeneity and therefore cannot detect subclonal resistance across organoids. Additionally, within a patient or cancer type there is a distribution of baseline PDCO sizes that is ignored with well-level response assessments. To address these issues, wide-field microscopy of autofluorescence from metabolic coenzymes can be used to segment individual PDCOs for heterogeneity measurements without the use of dyes or labeling (Figure 4a). Individual PDCO segmentation also provides improved sensitivity to drug response, by normalizing growth to the pretreatment diameter for each organoid (Figure 4b).

Figure 4.

Drug response assessed by PDCO diameter 96 h posttreatment with and without organoid-level baseline normalization.

(a) Representative autofluorescence intensity images using wide-field microscopy of colorectal cancer PDCOs with control and FOLFOX treatment over 96 h. Autofluorescence is from the reduced form of NAD(P)H. (b) Pooled analysis of nine independent colorectal cancer PDCO lines for control (n = 806 PDCOs, black) and 48-h FOLFOX treatment (n = 1,282 PDCOs, red) assessed at 96 h. (c) Growth of individual PDCOs benchmarked from day 0 for each organoid; the effect size (Glass’s delta, GΔ) of FOLFOX treatment is increased compared with pooled analysis. The region of interest is defined using triangle thresholding (42) of NAD(P)H intensity with absolute Feret’s diameter (43). Distributions are plotted as fits to multiple Gaussians using GraphPad. Sensitivity is assessed using population effect size (GΔ). Abbreviations: NAD(P)H, nicotinamide adenine dinucleotide (phosphate); PDCO, patient-derived cancer organoid.

Organoid growth alone also has limitations as a single marker of therapeutic efficacy. For example, it is known that rare self-renewing cell populations can readily yield clinically significant resistance even in the case of an initial therapeutic response (44); these cell populations are not modeled by the bulk of organoid growth and suggest the need for enhanced sensitivity, including the consideration of single-cell assessments. Sphere size alone may not predict live cells within a culture in the setting of other factors such as necrotic central cores or central mucin deposition. Therefore, new functional assessment techniques are needed that can monitor single-cell and single-organoid heterogeneity, normalize to baseline culture characteristics, and test clinically relevant drug doses and schedules.

3.1. Optical Metabolic Imaging of Drug Response in PDCOs

Multiphoton microscopy of the metabolic coenzymes NAD(P)H [the reduced form of nicotinamide adenine dinucleotide (phosphate)] and flavin adenine dinucleotide (FAD), or optical metabolic imaging (OMI), provides label-free 3D imaging of single-cell metabolism within PDCOs (11, 45–51). NAD(P)H is an electron donor involved in hundreds of metabolic reactions within the cytosol and mitochondria, while FAD is an electron acceptor in the mitochondria. Therefore, the ratio of NAD(P)H to FAD fluorescence intensities (optical redox ratio) provides a label-free measurement of the cell oxidation-reduction state (52, 53) that is sensitive to malignancy and drug response (11, 45–50, 54). The fluorescence lifetimes of NAD(P)H and FAD are both biexponential due to distinct lifetimes in the free and protein-bound states (55), so fluorescence lifetime imaging of NAD(P)H and FAD provides information on the protein-binding activities for each molecule (56). The fluorescence lifetime is independent of total photon counts (above signal-to-noise ratio requirements) and provides highly reproducible results (57). Previous studies have found that a linear combination of the optical redox ratio, mean NAD(P)H lifetime, and mean FAD lifetime (OMI index) provides improved sensitivity to drug response over any OMI variable alone in PDCOs and in vivo animal models of cancer (11, 45).

OMI is particularly attractive for drug response measurements in PDCOs because the sensing depth of multiphoton microscopy is sufficient to image through the entire thickness of the PDCO. The label-free features of OMI enable imaging of cellular-level response to treatment over time, so that the evolution of drug resistance and metabolic heterogeneity can be quantified. OMI fills a niche for noninvasive single-cell imaging of functional changes within PDCOs that cannot be achieved with existing techniques that require sample destruction (e.g., genetic sequencing, flow cytometry, immunohistochemistry) or cannot assess response on a single-cell level (e.g., plate reader assays, diameter changes). OMI has previously monitored metabolic heterogeneity in multiple cancer models including response predictions in prospective clinical studies (11, 21, 56, 58–60) (Figure 5).

Figure 5.

OMI of PDCOs predicts pancreatic cancer patient response to therapy. (a,b) Representative redox ratio, NAD(P)H mean lifetime (τ_m), and FAD τ_m images of pancreatic PDCOs from (a) a patient with RFS greater than 21 months (responder) and (b) a patient with RFS of 5 months (nonresponder). Left columns are control organoids and right columns are organoids treated with the same drugs as the patient adjuvant treatment. (c,d) The effect size, defined by Glass’s delta (GΔ) (61), of the same drugs on the OMI index averaged across all cells in organoids derived from the (c) responder and (d) nonresponder. Error bars indicate mean ± standard error of the mean; *p < 0.0001. (e,f) Single-cell OMI index subpopulation analysis of treatment response in organoids from the (e) responder and (f) nonresponder. Abbreviations: 5-FU, 5-fluorouracil; FAD, flavin adenine dinucleotide; NAD(P)H, nicotinamide adenine dinucleotide (phosphate); O+F, oxaliplatin + 5-fluorouracil; OMI, optical metabolic imaging; PDCO, patient-derived cancer organoid; RFS, recurrence free survival. Figure adapted with permission from Reference 11.

Multiphoton OMI is attractive for cell-level assessment and high-content studies of cancer drug resistance and heterogeneity. However, multiphoton microscopy uses ultrafast lasers and optical elements (e.g., dispersion compensation) that require significant expertise to reproducibly operate and are cost prohibitive for widespread use in research labs and clinics. Therefore, new methods have been developed to reproducibly measure the optical redox ratio from PDCOs using accessible wide-field fluorescence microscopes that are already present in most research and clinical labs. Validation studies confirm that wide-field OMI can track metabolic changes with drug treatment in PDCOs on a single-organoid level (62). Time-course measurements of drug response along with single-organoid tracking were crucial for reproducible results (Figure 6). PDCO drug response measured with wide-field microscopy showed no direct correlation between individual organoid morphology and optical redox ratio, suggesting that these are independent metrics of therapeutic response (62). These studies indicate that OMI could be implemented on a wider scale for high-throughput drug testing and patient treatment planning, provided that drug response is tracked on the single-organoid level (62).

Figure 6.

Wide-field OMI with comparison of single-organoid tracking and pooled analysis. (a) Representative wide-field OMI of the optical redox ratio in colorectal cancer PDCOs treated with paclitaxel. (b,c) Comparison of single-organoid tracking and pooled analysis (well-level) using time series data from colorectal cancer PDCOs. Normalization refers to dividing the value of each organoid at each time point by either its own pretreatment value (organoid-level normalization) or the well-level mean (well-level normalization). The pretreatment normalized optical redox ratio using (b) organoid-level normalization and (c) well-level normalization is shown for each treatment group. Data plotted are mean (colored lines) ± standard error of the mean (shaded areas). A significant difference between a treatment and control (p < 0.05) is indicated with a circle color-coded for the treatment group (see key). Abbreviations: 2DG, 2-deoxy-D-glucose; FAD, flavin adenine dinucleotide; NAD(P)H, nicotinamide adenine dinucleotide (phosphate); OMI, optical metabolic imaging; PDCO, patient-derived cancer organoid. Figure adapted with permission from Reference 62.

3.2. Outlook for PDCO Response Assessment Technologies

Automation of assessment methods will be key for robust and reproducible measurements in a clinic or cancer research lab setting. Ideally, push-button systems will be developed that assess response on a single-organoid or a single-cell level using individual organoid tracking over time for improved accuracy. These methods must be coupled with well-trained classifiers for treatment response decisions, a process that requires large data sets of diverse patient samples and treatment protocols. It is likely that different classifiers will be required for certain histological cancer types and/or treatment conditions (e.g., antibody-based treatments versus chemotherapies) using the same assessment technologies. Additionally, measures of drug response that change earlier than cell viability (e.g., metabolism) could minimize readout times and ultimately improve the sensitivity and throughput of response assessment technologies.

4. PDCOs AS A MODEL FOR TUMOR HETEROGENEITY

4.1. Tumor Heterogeneity

Cancer heterogeneity is a major obstacle in the development of effective therapeutic strategies for patients (63). This heterogeneity, unfortunately, exists on multiple levels. This includes heterogeneity between patients with the same cancer histologic type, differences between separate tumors within the same patient, distinct cell types within the tumor microenvironment (64), and metabolic (65) and mutational differences between cancer cell clones (66). All these sources of heterogeneity can have important clinical impacts, including significant differences in therapeutic benefit between patients with similar cancers. Heterogeneity within an individual patient becomes even more complex in the setting of therapeutic exposure, as diverse resistance mechanisms are often simultaneously encountered within the same patient (67). To overcome the impact of this multilevel heterogeneity within and between cancers, improved preclinical modeling is required to evaluate the impact of these factors on depth and duration of treatment response.

Molecular or mutational heterogeneity between tumor cells within a patient can be acquired through tumorigenesis, cancer progression, and the development of therapeutic resistance. Many cancers develop as a consequence of DNA mutations that occur due to aging and associated factors (68). Other cancers develop secondary to carcinogens or DNA repair defects, which can lead to greater genetic diversity within the tumors (68). Regardless of how these alterations are acquired, there are two general classes of mutations observed within cancers (69). The more abundant alterations are passenger mutations that do not offer the cancer cells a selective growth advantage but are commonly used to evaluate for clonal evolution and phylogenetic tracing. These alterations contrast with driver mutations that provide the cancer cells a selective growth advantage, such as mutations in the KRAS oncogene or tumor suppressors such as TP53 and APC. The impact of driver mutations is more apparent due to numerous therapies that directly target these mutant proteins or the downstream consequences of their cellular signaling.

When examining driver and passenger mutations within a cancer it is also important to understand which of these variants are founding mutations that were present when the cells transitioned into a cancer (69). For most cancers in the pretreatment setting, the driver mutations identified, usually two to five per cancer, are also founding alterations. These alterations are observed in all cancer cells of a tumor and are concordant between the primary and metastatic sites. As discussed above, multiple groups, including our own, have shown that PDCOs faithfully possess the founding driver alterations observed in cancers across numerous histologic subtypes. As these cancers progress following tumorigenesis, they have the potential to develop rare subclones. If these subclones arise early enough in tumor development, then sufficient subclonal mixing can occur, resulting in the identification of these subclones throughout the tumor (70, 71). However, if subclones acquire additional driver mutations later in the progression of the cancer, these alterations can be sequestered in geographic regions of the cancer, limiting the chance that these alterations will be identified from tissue biopsies. It is these later onset mutations that result in the cancer cell heterogeneity that is of the greatest clinical concern, since these mutations are difficult to identify with bulk sequencing of the tumor yet can lead to clinically significant resistance (67).

4.2. Heterogeneity in PDCOs

Most preclinical studies investigating new therapies, whether using PDCOs or not, completely ignore cancer heterogeneity. This omission is largely related to the lack of clinically relevant heterogeneity in preclinical models. Multiple research groups have now reported on the use of PDCOs to model different features of cancer heterogeneity, including the presence of subclones with distinct molecular features (35, 72–74), heterogeneity in features between a patient’s cancers of a similar histologic type (29, 32, 35, 72–79), and differential sensitivity to therapeutics (29, 32, 35, 72–79).

Heterogeneity in organoid growth is identified within nearly every PDCO culture across subjects. To evaluate whether key culture conditions could create this growth heterogeneity, we evaluated the baseline size, position of the individual organoids within a well, plating density, and the number of times the organoid cultures were passaged (80, 81). We did not find that any of these factors correlated with organoid growth. The growth rate instead varied significantly on the basis of which patient the sample was derived from, the molecular profile of the cancers, the histologic subtype, and the presence of anticancer therapies.

Further studies support heterogeneity in organoid growth and metabolism. Our group and others have demonstrated that almost all PDCO cultures possess multiple subclones (21, 35, 72–74, 80, 81) (Figure 3). It is likely that each individual organoid is either clonal or oligoclonal, but diverse molecular changes can be seen across the individual organoids within a culture. The majority of subclonal mutations identified within PDCOs are passenger mutations, but subclonal driver alterations can be detected in some instances (21, 82). Next-generation sequencing of organoids does allow for enhanced sensitivity to detect these subclonal alterations, as most cultures eliminate the majority of contaminating stromal cells that can limit the sensitivity of bulk tumor sequencing. Whole-genome sequencing of organoids has also been used to identify karyotype alterations and monitor karyotype diversification over time (83).

Beyond molecular changes, differences in morphology have also been observed within single cultures. For example, some PDCOs within the same well will have a more dense phenotype, while others are more hollow in nature (11). It is likely that those organoids with a hollow phenotype possess secretory cells, resulting in fluid accumulation within the center or lumen of these spheres. Beyond morphologic changes, OMI of cellular heterogeneity has been used by our group to characterize heterogeneity in breast cancer, pancreatic cancer, and colorectal cancer PDCOs (11, 21, 45, 48–50). Baseline metabolic heterogeneity was observed with single-cell OMI techniques, demonstrating differences across cells within individual organoids and differences in growth rate across organoids within the same culture (11) (Figure 7). Interestingly, the metabolic heterogeneity within and across organoids is mostly independent of the growth heterogeneity (62), indicating that metabolism and growth provide complimentary information.

Figure 7.

PDCO growth and metabolic heterogeneity. (a,b) Representative brightfield imaging of PDCOs treated with (a) control media or (b) the antiepidermal growth factor receptor inhibitor panitumumab. Images are annotated with the percent change in growth for individual PDCOs. (c,d) Two photon optical redox ratio [NAD(P)H/FAD fluorescence intensity] imaging demonstrates single-cell-level heterogeneity after 48 h of treatment with (c) control media and (d) panitumumab. Abbreviations: FAD, flavin adenine dinucleotide; NAD(P)H, nicotinamide adenine dinucleotide (phosphate); PDCO, patient-derived cancer organoid.

Previous studies have shown that cellular metabolic heterogeneity within an in vivo tumor is recapitulated in organoids, both before and after treatment (58). OMI was performed in vivo in a genetically engineered mouse model of breast cancer and in organoids derived from the same model, before and after treatment with paclitaxel and the phosphatidylinositol-3-kinase inhibitor XL147. Metabolic heterogeneity in both the organoids and in vivo images were quantified with a weighted heterogeneity index defined by single-cell OMI population distributions across all samples. This weighted heterogeneity index showed parallel shifts in vivo and in organoids between control and treated conditions. Interestingly, these previously unappreciated heterogeneous metabolic responses in tumors and organoids could not be attributed to tumor cell fate or varying leukocyte content within the microenvironment, suggesting that heightened metabolic heterogeneity upon treatment is largely due to heterogeneous metabolic shifts within tumor cells (58). Together, these studies show that OMI revealed remarkable heterogeneity in response to treatment, a finding that could provide a novel approach to predict the presence of potentially unresponsive tumor cell subpopulations lurking within a largely responsive bulk tumor population that might otherwise be overlooked by traditional measurements.

4.3. Outlook for Tumor Heterogeneity Modeling with PDCOs

The heterogeneity within tumors is of known clinical importance yet is not routinely measured for research or clinical purposes. This heterogeneity is especially relevant for studies developing new treatment strategies. The use of PDCOs to faithfully model patient tumor heterogeneity will have a significant impact on new drug development but must be coupled with new technologies to assess heterogeneity in growth, metabolism, and molecular changes. Additionally, organoid culture methods are becoming increasingly complex, incorporating additional cell types (e.g., immune cells, cancer-associated fibroblasts, blood vessels) and microfluidic structures to build patient-derived 3D systems to test drugs that target the tumor microenvironment (84–88). Alternatively, heterogeneity can be assessed in living tumor slices from patients to determine spatial distributions of response across distinct tumor and stromal niches (89). Finally, technologies that can segment individual cell types in a complex culture (90) could have a significant translational impact, especially for the development of immunotherapies or stromal targeting agents.

5. THERAPEUTIC DRUG SCREENING USING PDCOs

Given the faithful ways in which PDCOs represent cancers, PDCOs hold great promise as a tool to identify novel therapeutic combinations. New methods are essential to meet the requirements of high-throughput drug screening while overcoming some of the limitations unique to PDCO culture. An ideal technology for PDCO drug screening would assess at least hundreds of drugs or drug combinations simultaneously, allow for physiologic dosing, provide a quick and sensitive readout, allow for diverse samples to be investigated, and accurately identify therapies for further investigation.

Physiologically relevant dosing is critical for successful drug screens. Most recently published drug screens with PDCOs have not considered the clinically achievable dose within the blood or tumor of patients (12, 32, 35). The duration of drug exposure must also be clinically relevant even if the dose is clinically achievable. For example, gemcitabine is administered clinically once a week intravenously and can clinically achieve a plasma concentration of 30 μM, but it is rapidly cleared in hours, with minimal drug remaining at 24 h postinfusion. Many studies using gemcitabine in preclinical studies allow this treatment to remain in the culture for 72 h or more, making these studies challenging to interpret, as the dosing is not achievable in patients (12). The dose of therapeutics should be selected to recapitulate clinical pharmacokinetics of each agent to not exceed the maximum concentration and the duration of exposure.

5.1. Challenges with Low-Volume Wells in High-Throughput Screens

The most common PDCO culture methods use 6- to 24-well plates that are not conducive to medium- or high-throughput applications. To overcome this limitation, multiple groups have now adapted organoid methods for 96- to 384-well plate formats (5, 12, 32, 35, 73, 91). However, organoids suspended in a Matrigel matrix are difficult to plate at these low volumes. Most groups are now plating the Matrigel or other matrix materials into the wells first (91). Then the organoids are plated with the culture media. The organoids then embed into the top of the Matrigel, allowing therapeutic agents to be added to the media using robotic liquid handling devices once the PDCOs have matured.

With the PDCOs plated in a high-throughput format and drugs dosed in a clinically relevant manner, it is important to determine the appropriate readouts of therapeutic response. The most common readout, CellTiter-Glo (92), is sensitive to organoid density per well, and, unfortunately, the plating of organoids is not nearly as uniform as 2D cultures. With the potential for significant differences in baseline number and size of organoids within the wells and the use of increasingly smaller wells for high-throughput processing, the interpretation of the viability readouts is critical. To overcome these limitations, prolonged response assessment time points have been used to increase changes between control and treated groups (12, 32, 73). This leads to drug exposure durations that are not clinically relevant. Our group has used individual organoid tracking, as described above, for diameter and metabolic imaging assessments at earlier time points (21, 62, 80, 81, 91). Organoid tracking improves the sensitivity to drug response compared with well-level assessments and diminishes the impact of variable organoid size and density within a well for high-throughput studies of smaller volumes. It also allows for a greater number of organoids to be imaged more quickly for high-throughput applications.

5.2. Outlook for Drug Screening with PDCOs

Successful drug screens will require increased throughput, accurate dosing, and precise timing, along with reproducible assessment techniques. Once these methods are standardized, PDCOs will likely become a preferred sample for drug screening, with the advantage of patient-relevant diversity. As with any high-throughput drug screen, lead compounds will require further validation in vitro and in vivo, but the use of PDCOs could improve the likelihood of identifying successful therapies.

6. PDCOs FOR CLINICAL RESPONSE PREDICTIONS

Several studies have shown that organotypic cultures can predict clinical response in cancer (Table 2). This includes prior investigations with all standard-of-care chemotherapies for colorectal cancer including 5-fluorouracil (5-FU) (9, 10, 21), oxaliplatin (9, 21), irinotecan (10), cetuximab (9), trifluridine/tipiracil hydrochloride (TAS-102) (13), and regorafenib (13). Our group reported the combination of growth inhibition and single-cell OMI for the prediction of 5-FU and oxaliplatin combination chemotherapy in the refractory colorectal cancer setting (21). Ganesh et al. (9) reported the correlation between cell viability and therapeutic response to 5-FU, oxaliplatin, and radiation prior to surgery in rectal cancer. In advanced colorectal cancer, Ooft et al. (10) reported response prediction to 5-FU and irinotecan chemotherapy using a limited cohort of patients yet failed to predict response to a combination of 5-FU and oxaliplatin. Additionally, Vlachogiannis et al. (13) measured PDCO response from a diverse cohort of gastric and colorectal cases to predict clinical response. These early studies are encouraging but call for larger patient cohorts. Additionally, most of these studies relied on one or two response measurements, which are unlikely to provide robust predictive power for independent samples.

Table 2.

Studies of PDCOs in validating clinical outcomes

Reference	Disease	Dosing	Response assessment	Matrix	Treatment	Number of patients	Clinical outcome	Sensitivity	Specificity
Driehuis et al. (8)	HNSCC	Log-fold	Luminescence from CellTiter-Glo, AUC	BME	Radiation	7	Relapse	0.75	1
Ganesh et al. (9)	Rectal	Log-fold	Luminescence from CellTiter-Glo, AUC	Matrigel	FOLFOX/5-FU	7	PFS	NR	NR
					Radiation	7	Endoscopic response	NR	NR
Kratz et al. (81)	Multiple	Fixed	Organoid diameter, baseline OMI Index	Matrigel	Multiple	19	RECIST	GΔ > 1.25; 100%	GΔ > 1.25; 91.6%
								ROC AUC = 0.97
Ooft et al. (10)	CRC	Log-fold	Luminescence from CellTiter-Glo 3D, GR Score	Geltrex	FOLFIRI	12	RECIST	1	0.83
					Irinotecan	10		1	0.8
					FOLFOX	16		0.89	0.5
Sharick et al. (11)	Pancreatic	Fixed	OMI index	Matrigel	Multiple	7	RFS (> 12 months)	1	1
Tiriac et al. (12)	Pancreatic	Log-fold	Luminescence from CellTiter-Glo, AUC	Matrigel	Multiple	9	PFS	1	0.67
Vlachogiannis et al. (13)	GI (multiple)	Log-fold	Luminescence from CellTiter-Blue, AUC	Matrigel	Multiple	19	NR	1	0.93
Yao et al. (14)	Rectal	Fixed	Organoid size	Matrigel	Chemoradiation	80	Tumor regression grade	0.92	0.79
					Radiation			0.41	0.98
					5-FU			0.59	0.88
					Irinotecan			0.78	0.79

Representative studies of patient-derived cancer organoid (PDCO) response assessment with ≥7 clinical outcomes in specified disease types. Studies performed across histologic types including head and neck squamous cell carcinoma (HNSCC), rectal, colorectal (CRC), pancreatic, and other gastrointestinal (GI) cancers. Dosing of agents reported as single fixed dosing or titration on log-fold scales. Response assessment includes area under the curve (AUC), optical metabolic imaging (OMI) index, and growth inhibition (GR) score. BME denotes basal matrix extract. Clinical outcomes reported include progression-free survival (PFS), response evaluation criteria in solid tumors (RECIST), relapse free survival (RFS), and not reported (NR). Sensitivity measures include Glass’s delta (GΔ) and receiver operator characteristic area under the curve (ROC AUC).

6.1. Multivariate Classifiers

Multivariate classifiers are likely to perform better than any single response threshold measurement. For example, multivariate measurements of PDCO morphology and metabolism parameters could be coupled with classifiers such as logistic regression models, support vector machines, or random forest classifiers (51). This approach requires a large set of well-annotated PDCOs including multiple histological subtypes, treatment protocols, and patient outcomes, along with standardized PDCO response measurements across all samples. When patients are assigned to independent and nonoverlapping training and testing sets, an unbiased assessment can be made for the performance of a prediction model. If performance is sufficiently high, these classifiers can then be used to predict response in patients who were not used in the training or testing sets.

6.2. Outlook for PDCOs in Clinical Treatment Planning

The clinical validity of PDCOs should be comprehensively documented with prospective clinical investigations that ultimately achieve Clinical Laboratory Improvement Amendments certification (93) that allows for routine clinical use. This process requires standardization of methods to assess response in the clinic and prospective validation of unbiased classifiers with respect to progression-free survival. If successful, PDCOs could be used to guide clinical treatment planning and patient selection for clinical trials. Future drug response studies in PDCOs should also consider temporal dynamics and response heterogeneity, as most patients initially respond to treatment before developing adaptive resistance.

SUMMARY POINTS.

1. Patient-derived cancer organoids (PDCOs) recapitulate key features of the original patient tumor and are amenable to drug screens and heterogeneity analyses that are difficult to perform in other models (e.g., mouse models).

2. Current methods to measure drug response in PDCOs often overlook heterogeneity between cells and organoids within a patient, diminishing the sensitivity of drug screens.

3. Optical metabolic imaging of the intrinsic metabolic coenzymes NAD(P)H and FAD provides sensitive, label-free assessment of drug response in PDCOs at the cell and organoid level.

4. Drug screens in PDCOs often require drug doses and exposure times that are not clinically relevant, and new methods have emerged to address these deficiencies including single-organoid segmentation and tracking.

5. Prospective clinical studies show promise for PDCOs in clinical treatment planning.

FUTURE ISSUES.

1. Standardization of assessment methods is required before PDCOs can be used for routine clinical treatment planning.

2. Assessment of heterogeneity between organoids and/or cells is needed to detect subclones that drive treatment resistance in patients.

3. High-throughput drug screens in PDCOs will require single-organoid segmentation and tracking for sufficient sensitivity.

4. Large retrospective studies that include multiple tumor types, treatments, and patient outcomes are required to build robust predictive classifiers for prospective treatment evaluations.

5. Ability to add in key nontumor cells (e.g., stromal, endothelial, and immune cells) that contribute to the tumor microenvironment and drug response