Unified framework for patient-derived, tumor-organoid-based predictive testing of standard-of-care therapies in metastatic colorectal cancer Authors: Tao Tan, Dmitri Mouradov, Margaret Lee, Grace Gard, Yumiko Hirokawa, Shan Li, Cong Lin, Fuqiang Li, Huijuan Luo, Kui Wu, Michelle Palmieri, Evelyn Leong, Jordan Clarke, Anuratha Sakthianandeswaren, Helen Brasier, Jeanne Tie, Niall C. Tebbutt, Azim Jalali, Rachel Wong, Antony W. Burgess, Peter Gibbs, Oliver M. Sieber Published: 2023 ABSTRACT SummaryPredictive drug testing of patient-derived tumor organoids (PDTOs) holds promise for personalizing treatment of metastatic colorectal cancer (mCRC), but prospective data are limited to chemotherapy regimens with conflicting results. We describe a unified framework for PDTO-based predictive testing across standard-of-care chemotherapy and biologic and targeted therapy options. In an Australian community cohort, PDTO predictions based on treatment-naive patients (n = 56) and response rates from first-line mCRC clinical trials achieve 83% accuracy for forecasting responses in patients receiving palliative treatments (18 patients, 29 treatments). Similar assay accuracy is achieved in a prospective study of third-line or later mCRC treatment, AGITG FORECAST-1 (n = 30 patients). “Resistant” predictions are associated with inferior progression-free survival; misclassification rates are similar by regimen. Liver metastases are the optimal site for sampling, with testing achievable within 7 weeks for 68.8% cases. Our findings indicate that PDTO drug panel testing can provide predictive information for multifarious standard-of-care therapies for mCRC. Introduction Metastatic colorectal cancer (mCRC) remains the second leading cause of cancer-related mortality worldwide despite increasing therapy options.1 Heterogeneous treatment responses remain a principal challenge to improving outcomes, with limited clinical means to predict treatment efficacy. Precision oncology for directing therapy use is routinely limited to RAS (KRAS and NRAS) mutation status for anti-EGFR antibody therapy, DNA mismatch repair (dMMR) status (microsatellite instability-high [MSI-H]/dMMR) for immunotherapy, and more recently BRAFV600E mutation status for BRAF-targeted therapies and HER2/ERBB2 amplification status for HER2-targeted therapies.2,3,4,5,6,7,8,9,10 For most treatments, there are no markers to predict responses, and even for genomically selected patients, response rates are often low. Consequently, many patients receive ineffective, costly drugs and suffer unnecessary toxicities. A promising ex vivo approach to predict treatment benefit is personalized tumor response testing using patient-derived tumor organoids (PDTOs).11 PDTOs are self-organizing three-dimensional structures that can be readily derived from fresh patient tumor specimens, mirror the histologic and genomic features of the original tumor, and are eminently suitable for high-throughput drug testing.12,13,14,15,16,17 Post hoc studies for multiple cancer types indicate that PDTO response can emulate the therapeutic response of the matched patient.12,13,14,18,19,20,21 However, few prospective studies have attempted to define classification standards for PDTO assay predictions,13,14 and fewer still have validated predictive PDTO assays in independent patient cohorts.21,22 Currently, two prospective observational studies have reported on PDTO assay accuracies for mCRC treatment. The single-cohort TUMOROID study, fitting prediction cutoffs based on Response Evaluation Criteria in Solid Tumors (RECIST) response of the metastatic lesion from which the PDTO was derived, achieved an accuracy of 86.4% for irinotecan-based, second-line treatment but found poor performance for first-line treatment with 5-fluorouracil plus oxaliplatin (FOLFOX).13 Wang et al., defining a joint prediction cutoff for FOLFOX and 5-fluorouracil (5FU) plus irinotecan (FOLFIRI) neoadjuvant or adjuvant chemotherapy based on best RECIST response, American Joint Committee on Cancer/College of American Pathologists (AJCC/CAP) regression grade, or carcinoembryonic antigen (CEA) levels, achieved a 79.7% accuracy in an independent validation cohort, although data were not detailed by treatment setting or regimen.21 While previous prospective studies have focused on 5FU-, oxaliplatin-, and irinotecan-based chemotherapies, PDTO predictive assays encompassing the spectrum of available standard-of-care chemotherapy and biologic and targeted therapy options remain to be explored. Moreover, while previous studies indicate that PDTO assay turnaround times for small numbers of treatments in defined clinical scenarios are achievable with time frames for patient management,12,13,14,21 whether this can be realized for routine panel testing of standard-of-care therapeutics is uncertain. Other challenges to diagnostic translation include success rate of PDTO generation from biopsy samples, assay standardization, and automation and processes for clinical reporting. Here, we embarked on establishing a unified framework for PDTO standard-of-care drug testing and clinical reporting. To develop a PDTO assay applicable across mainstay therapies compatible with PDTO monoculture (5FU, oxaliplatin, irinotecan, FOLFOX, FOLFIRI, trifluridine/tipiracil, regorafenib, and anti-EGFR antibodies), we prospectively implemented a standardized robotic assay platform in a multicenter CRC community cohort where treatment and outcome data were prospectively collected in a clinical registry. Therapy-specific cutoff values for PDTO assay predictions were derived based on treatment-naive patients with CRC and response rates from first-line mCRC clinical trials, with evaluation of predictive accuracy in independent patients with mCRC receiving palliative treatment. We then conducted a prospective multicenter, observational clinical study to evaluate our PDTO drug testing framework in the third-line or later mCRC treatment setting (FORECAST-1, sponsored by AGITG, ACTRN12620001353987). Primary study aims were determination of clinical feasibility of PDTO drug panel testing and development of a process for clinical reporting, including validation of PDTO assay accuracy and assessment of success rate and turnaround time. Results Initial community cohort implementation of a standardized PDTO drug testing platform Prospective collection of tumor tissues from patients with newly diagnosed or recurrent CRC for PDTO generation was implemented across four hospitals in Melbourne, Australia, with standardized capture of treatment and outcome data (community cohort, Figure 1A).23,24 A total of 169 CRC tissue samples were obtained from 167 patients with CRC at resection of primary tumor (21 stage I, 56 stage II, 54 stage III, and 18 stage IV) or metastatic disease (n = 20), including one patient with primary tumor and liver metastasis and one patient with liver and lung metastasis samples. Patient characteristics are shown in Table S1.Figure 1PDTO generation and histopathologic validation for patients with stage I–IV CRC enrolled in the ACCORD-CRC and TRACC registries(A) Flow chart indicating the number of patients with stage I–IV CRC included, the success rate of establishing cultures from patients, and the number of PDTO lines progressed to drug sensitivity testing.(B) Representative bright-field images and H&E-stained sections of PDTOs grown in LVM suspension culture medium illustrating heterogeneous morphologies ranging from solid/compact structures to thin-walled cystic structures. Scale bars, 200 μm.(C) Representative H&E staining and immunohistochemistry results for PDTOs from patients stained for Ki67, p53, and MUC2. Scale bars, 200 μm.(D) Correlation of tumor Ki67 expression and PDTO Ki67 expression, PDTO growth rate and tumor Ki67 expression, and PDTO growth rate and PDTO Ki67 expression (n = 21 tissues and n = 20 PDTOs). Statistical significance was attributed to values of p < 0.050 as determined by the correlation test. ∗p < 0.050.See also Table S1. PDTO generation and histopathologic validation for patients with stage I–IV CRC enrolled in the ACCORD-CRC and TRACC registries (A) Flow chart indicating the number of patients with stage I–IV CRC included, the success rate of establishing cultures from patients, and the number of PDTO lines progressed to drug sensitivity testing. (B) Representative bright-field images and H&E-stained sections of PDTOs grown in LVM suspension culture medium illustrating heterogeneous morphologies ranging from solid/compact structures to thin-walled cystic structures. Scale bars, 200 μm. (C) Representative H&E staining and immunohistochemistry results for PDTOs from patients stained for Ki67, p53, and MUC2. Scale bars, 200 μm. (D) Correlation of tumor Ki67 expression and PDTO Ki67 expression, PDTO growth rate and tumor Ki67 expression, and PDTO growth rate and PDTO Ki67 expression (n = 21 tissues and n = 20 PDTOs). Statistical significance was attributed to values of p < 0.050 as determined by the correlation test. ∗p < 0.050. See also Table S1. PDTOs were grown in low-viscosity matrix (LVM) suspension culture medium to facilitate passaging and dispensing as described previously25; Wnt ligands were omitted to prevent the outgrowth of normal colorectal epithelium. Overall, 81.1% (137 of 169) PDTO LVM cultures were successfully established and biobanked, with similar establishment rates by patient sex, age at diagnosis, tumor stage, primary tumor site, grade, and RAS status but lower success rates for stage II, MMR-deficient, and RAF mutant tumors (Table S1). PDTO lines exhibited heterogeneous morphologies ranging from predominantly compact to cystic structures (Figure 1B). PDTOs grown in LVM culture conditions were confirmed to resemble the histologic and genomic features of the original tumor. Marker expression analysis demonstrated similar staining patterns for Ki67, p53, and Muc2 (Figure 1C), with proportions of Ki67-positive cells further correlating with PDTO growth rates as determined by time-lapse imaging (Figure 1D).25 Whole-genome sequencing (WGS) of PDTOs at a median passage of 7 (range 1–22) and matched tumor and germline DNA from 21 patients demonstrated similar total mutation loads and point mutation types, including for two hypermutated cases with dMMR deficiency (WCB046) and DNA polymerase epsilon (POLE) mutation (WCB268) (Figure 2A). Genome-wide somatic single-nucleotide variant (SNV) analysis revealed a median SNV overlap of 72.2% (range, 49.4%–87.8%) between PDTO and matched tumor samples (Figure 2A), with analysis of the mutational spectrum in the most frequently mutated CRC genes showing a median overlap of 97.5% (range 81.0%–100%) (Figure 2B). PDTOs also recapitulated the genome-wide DNA copy-number variation (CNV) patterns of paired tumors (Figure 2C), with both showing frequent deletions of chromosome arms 8p, 17p (including TP53), and 18q (including SMAD4) and gains of chromosomes 7, 8q (including MYC), 13, and 20q (Figure 2D). Assessment of matched PDTO and tumor samples for systematic differences in genomic alterations identified no recurrent changes in >2 sample pairs for gene mutations and no bias in gene copy-number gain or loss (Table S2). There was no relationship between PDTO passage number and mutation burden for non-hypermutated cases (r = 0.11; p = 0.667; correlation test).Figure 2Genomic features of PDTOs and matched original tumor tissues from the community cohort as determined by whole-genome sequencing(A) Mutation profiles of 21 PDTOs and matched tumor tissues, including counts of SNVs and insertions or deletions (indels), proportions of nucleotide transitions and transversions, and overlap for genome-wide somatic SNVs.(B–D) Somatic mutation spectra for the most frequently mutated CRC genes (B), genome-wide relative DNA copy-number aberrations (C), and frequencies of DNA copy-number gains and losses for PDTOs and matched tumor tissues (D). T, tumor; O, patient-derived tumor organoid.See also Table S2. Genomic features of PDTOs and matched original tumor tissues from the community cohort as determined by whole-genome sequencing (A) Mutation profiles of 21 PDTOs and matched tumor tissues, including counts of SNVs and insertions or deletions (indels), proportions of nucleotide transitions and transversions, and overlap for genome-wide somatic SNVs. (B–D) Somatic mutation spectra for the most frequently mutated CRC genes (B), genome-wide relative DNA copy-number aberrations (C), and frequencies of DNA copy-number gains and losses for PDTOs and matched tumor tissues (D). T, tumor; O, patient-derived tumor organoid. See also Table S2. 84 PDTO lines from 82 patients with available treatment and follow-up data were progressed to drug sensitivity testing on a previously validated high-throughput imaging platform in a central laboratory (Figure 3A; Table S1).25 PDTOs were exposed to standard-of-care single-agent and combination therapies compatible with PDTO monoculture (5FU, oxaliplatin, SN38 [active metabolite of irinotecan], regorafenib, TAS-102 [trifluridine/tipiracil], FOLFOX, and FOLFIRI), and clinically accessible agents with reported activity in single-arm phase 2 studies (gemcitabine,26,27 pemetrexed,28,29,30 and temozolomide31). Erlotinib was used as a surrogate for anti-EGFR antibodies, as the latter required a different vehicle from the other compounds. Single agents were assayed in duplicate in 9-step, 4-fold dilution series; FOLFOX and FOLFIRI treatments were assayed in duplicate in 7 × 7 drug matrices of 5FU and oxaliplatin or SN38, respectively, with 4-fold dilution series of treatment components. Standard-of-care drug titration ranges encompassed patient-relevant drug concentrations (5FU [Cmax ≈ 7.5 μM]32; oxaliplatin [Cmax ≈ 5.0 μM]33; SN38 [Cmax ≈ 0.14 μM]33; regorafenib [Cmax ≈ 7.3 μM] [ClinicalTrials.gov: NCT01853046]; TAS-102 [Cmax ≈ 18.2 μM] [AusPAR 2018]; and erlotinib [Cmax ≈ 4.6 μM] [ClinicalTrials.gov: NCT00965731]); highest concentrations for investigational agents were 15 μM for gemcitabine, 10 μM for pemetrexed, and 750 μM for temozolomide. PDTOs were grown from single cells over a 3-day period followed by compound addition from preprepared drug dilution plates. PDTOs were then imaged daily for 7 days for PDTO-size-based quantification of viability as validated previously (Figures 3B and 3C).25 Raw and processed PDTO viability data for the community cohort are provided in Tables S3 and S4. Drug responses were quantified as area under the dose-response curve (AUC) for single agents and as mean “pan-matrix” viability score for drug combinations.Figure 3PDTO drug sensitivity for patients with stage I–IV CRC enrolled in the ACCORD-CRC and TRACC registries(A) Schematic of the standardized workflow for semiautomated PDTO drug testing. Established tumoroids were dissociated and seeded as single cells, grown into small PDTOs over 3 days, and treated with drug for 7 days with daily bright-field imaging; viability was quantified by determination of mean PDTO size.(B) Representative images of PDTOs treated with increasing 5FU concentrations on day 10 of the assay.(C) Representative PDTO dose-response curve for 5FU single-agent treatment and viability matrix for 5FU-oxaliplatin combination treatment.(D) Drug sensitivity profiles for 56 PDTO lines from 56 chemo-naive patients with CRC for single-agent 5FU, oxaliplatin, SN38, regorafenib, TAS-102, erlotinib, gemcitabine, pemetrexed, and temozolomide treatments and combination 5FU-oxaliplatin and 5FU-SN38 treatments; indicated cutoffs are based on treatment-naive patients and response rates from clinical trials of first-line mCRC treatment; blue, resistant; red, sensitive.(E) Correlation of PDTO erlotinib versus cetuximab sensitivity (n = 8 PDTOs). Statistical significance was attributed to values of p < 0.050 as determined by the correlation test.(F) Comparison of erlotinib sensitivity between RASMut (n = 28) and RASWT (n = 38) PDTOs. Statistical significance was attributed to values of p < 0.050 as determined by the Student’s t test. ∗p < 0.050.See also Figures S1 and S5 and Tables S1, S3, and S4. PDTO drug sensitivity for patients with stage I–IV CRC enrolled in the ACCORD-CRC and TRACC registries (A) Schematic of the standardized workflow for semiautomated PDTO drug testing. Established tumoroids were dissociated and seeded as single cells, grown into small PDTOs over 3 days, and treated with drug for 7 days with daily bright-field imaging; viability was quantified by determination of mean PDTO size. (B) Representative images of PDTOs treated with increasing 5FU concentrations on day 10 of the assay. (C) Representative PDTO dose-response curve for 5FU single-agent treatment and viability matrix for 5FU-oxaliplatin combination treatment. (D) Drug sensitivity profiles for 56 PDTO lines from 56 chemo-naive patients with CRC for single-agent 5FU, oxaliplatin, SN38, regorafenib, TAS-102, erlotinib, gemcitabine, pemetrexed, and temozolomide treatments and combination 5FU-oxaliplatin and 5FU-SN38 treatments; indicated cutoffs are based on treatment-naive patients and response rates from clinical trials of first-line mCRC treatment; blue, resistant; red, sensitive. (E) Correlation of PDTO erlotinib versus cetuximab sensitivity (n = 8 PDTOs). Statistical significance was attributed to values of p < 0.050 as determined by the correlation test. (F) Comparison of erlotinib sensitivity between RASMut (n = 28) and RASWT (n = 38) PDTOs. Statistical significance was attributed to values of p < 0.050 as determined by the Student’s t test. ∗p < 0.050. See also Figures S1 and S5 and Tables S1, S3, and S4. Coefficients of variation (CVs) for vehicle control wells were less than 20% for all assay plates (n = 168), and PDTOs screened on two separate occasions showed concordant results (Figures S1A and S1B). As anticipated, drug sensitivity profiles varied considerably between individual PDTOs for all single-agent and combination treatments (Figure 3D). Erlotinib sensitivity correlated with cetuximab sensitivity and showed the expected association with RAS mutation status, with resistance demonstrated in RASMut PDTOs (Figures 3E and 3F). For the two patients with PDTOs available from two different lesions, drug testing results showed strong correlations (primary tumor and liver metastasis: r = 0.89; p = 0.0013; liver and lung metastases: r = 0.70; p = 0.036, correlation test). PDTO-based prediction model for standard-of-care drug testing in mCRC To develop a predictive PDTO assay applicable across standard-of-care therapy options for mCRC, we defined therapy-specific cutoff values based on drug sensitivity profiles of treatment-naive patients from our initial community cohort (n = 56) and response rates reported in clinical trials of first-line therapy (Figure 4A): 25% for 5FU and TAS-102, 20% for irinotecan, oxaliplatin, and anti-EGFR antibody therapy (for RASWT tumors), 10% for regorafenib, and 50% for FOLFIRI and FOLFOX (clinical trial details in Table S5). Prediction cutoff values were applied to community patients treated for metastatic disease post-PDTO generation (18 patients, 29 evaluable treatments). One patient had received neoadjuvant treatment for a primary rectal cancer; however, the PDTO for this patient was derived from a recurrent liver metastasis. In vitro drug response predictions were correlated with clinical outcomes, defining patients with complete response (CR), partial response (PR), or stable disease (SD) for more than 24 weeks by clinician assessment as experiencing clinical benefit. For patients receiving both chemotherapy and biologic therapy in a line of treatment, PDTOs were considered “sensitive” if sensitivity was observed to at least one treatment component; similarly, response to FOLFOXIRI therapy was scored based on both FOLFOX and FOLFIRI matrices.Figure 4PDTO-based prediction model for standard-of-care drug testing in mCRC(A) Flow chart indicating the number of PDTOs from treatment-naive community patients and response rate cutoffs from first-line mCRC clinical trials used for predictive model development and the number of PDTOs from an independent validation cohort of community patients used for prospective model evaluation.(B) Details of mCRC patient treatments, clinical outcomes, PDTO predictions, and matching status of predictions and responses (18 patients, 29 treatments). Pie charts detail model accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).See also Table S5. PDTO-based prediction model for standard-of-care drug testing in mCRC (A) Flow chart indicating the number of PDTOs from treatment-naive community patients and response rate cutoffs from first-line mCRC clinical trials used for predictive model development and the number of PDTOs from an independent validation cohort of community patients used for prospective model evaluation. (B) Details of mCRC patient treatments, clinical outcomes, PDTO predictions, and matching status of predictions and responses (18 patients, 29 treatments). Pie charts detail model accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). See also Table S5. PDTO-based drug response predictions were highly correlated with patient clinical outcomes with concordance for 24 of 29 treatments (Figure 4B). Among 11 treatments with matched PDTOs predicting sensitivity, 10 were associated with clinical benefit, and 1 was not; among 18 treatments with PDTOs predicting resistance, 14 did not result in clinical benefit, and 4 did. Thus, classification achieved an accuracy of 82.8% (95% confidence interval [CI] = 64.2%–94.2%) with 71.4% sensitivity (true positive rate) and 93.3% specificity (true negative rate) (p < 0.0005 for accuracy (Acc) > non-informative rate (NIR); binomial test). The median time from PDTO harvest to treatment was 9 weeks (range, 2–53 weeks), with no association between this time window and assay accuracy (p = 0.766; Student’s t test). Feasibility of organoid response assessment to define effective treatments for patients with CRC after failure of standard therapy (AGITG FORECAST-1) To evaluate clinical feasibility of PDTO drug panel testing for patients with mCRC and develop a process for clinical reporting of PDTO results, taking into account other relevant clinical information, we conducted a prospective, multicenter observational study in the third-line or later mCRC treatment setting (AGITG FORECAST-1). Between September 2020 and December 2021, 30 patients were enrolled, with a median age of 59 years (range, 36–71). Fresh tumor biopsies were obtained from liver (n = 16), peritoneum (n = 6), lung (n = 3), lymph node (n = 2), primary tumor (n = 1), brain (n = 1), and bone (n = 1). Patient characteristics are shown in Table S6. Overall, we achieved a 63.3% (19 of 30) success rate of PDTO establishment and drug sensitivity testing from biopsy specimens (Figure 5A). Six tissue cultures did not form PDTOs, and five PDTO cultures failed to expand. Examination of culture success rates by site of biopsy sampling identified CRC liver metastases as the optimal location, achieving a 75.0% (12 of 16) success rate as compared to 50.0% (7 of 14) for metastases from other sites. For all assay plates, CVs for vehicle control wells were less than 20% (Figure S1C). Time from PDTO biopsy to drug screen result for our 9 single-agent and 2 combination treatments based on duplicate 9-step dose-response curves and 7 × 7 drug matrices ranged from 25 to 144 days (median, 72 days). Raw and processed PDTO viability data for the FORECAST-1 cohort are provided in Tables S7 and S4. WGS for 17 PDTOs with available paired blood samples confirmed tumor origin (Figure S2).Figure 5AGITG FORECAST-1 study to determine the feasibility of PDTO-based response prediction in the third-line or later mCRC treatment setting(A) Flow chart indicating the number of patients recruited, the success rate of establishing and drug testing PDTO cultures, and the number of evaluable patients and clinical responses.(B) Representative report indicating PDTO assay results for single-agent treatments grouped into quintiles from low to high sensitivity and summary of respective moderated clinical interpretations; future versions of the report will indicate PDTO assay cutoff-based predictions of sensitivity and resistance.(C) Details of patient treatments, clinical outcomes, PDTO predictions, and matching status of predictions and responses (9 patients, 10 treatments).See also Figures S1 and S2 and Tables S4, S6, S7, and S8. AGITG FORECAST-1 study to determine the feasibility of PDTO-based response prediction in the third-line or later mCRC treatment setting (A) Flow chart indicating the number of patients recruited, the success rate of establishing and drug testing PDTO cultures, and the number of evaluable patients and clinical responses. (B) Representative report indicating PDTO assay results for single-agent treatments grouped into quintiles from low to high sensitivity and summary of respective moderated clinical interpretations; future versions of the report will indicate PDTO assay cutoff-based predictions of sensitivity and resistance. (C) Details of patient treatments, clinical outcomes, PDTO predictions, and matching status of predictions and responses (9 patients, 10 treatments). See also Figures S1 and S2 and Tables S4, S6, S7, and S8. A process for standardized reporting of PDTO-guided treatment recommendations was developed involving assessment of PDTO assay results by a team of study medical oncologists and platform specialists. To facilitate communication of overall PDTO sensitivity patterns to clinicians, and after input from a broader medical oncology panel including all trial site investigators, data were summarized in graphical format with treatments grouped into quintiles from low to high sensitivity based on our reference treatment-naive community cohort (Figure 5B); PDTO assay cutoff-based predictions of sensitivity and resistance will be formally indicated in future versions of the report. Clinical interpretation considered assay predictions and the level of clinical trial evidence for each drug; a higher sensitivity score was given to agents with proven activity in randomized clinical trials versus those with limited prospective data supporting rechallenge strategies versus agents with activity indicated in non-randomized phase 2 studies. To summarize clinical recommendations, drugs were grouped into categories of predicted highest, intermediate, or lowest clinical benefit (Figure 5B). Study team meetings were convened within 1–8 days (mean, 4 days) of each PDTO assay completion, with all reports prepared within 24 h of this meeting, demonstrating process feasibility. Given the observational nature of this study to determine PDTO assay accuracy in the context of current standard treatment, PDTO assay results were not used to direct treatment, with patients receiving palliative therapy as per clinician’s discretion. Of the 19 AGITG FORECAST-1 patients with PDTO assay results, 11 commenced palliative treatment post-biopsy, with 4 patients ultimately receiving multiple post-biopsy treatment lines. A total of 10 treatment responses of 9 patients were evaluable (TAS-102, n = 4; FOLFOX/CAPOX, n = 3; FOLFOX + bevacizumab, n = 1; and FOLFIRI + cetuximab/panitumumab, n = 2). All 10 PDTO assay predictions indicated “resistance” and correctly identified 9 treatments that did not yield clinical benefit, thus achieving a 90% accuracy (Figure 5C). PDTO assay results were obtained prior to commencement of evaluable treatments in 20% (2 of 10) cases. Notably, PDTO testing identified treatments potentially efficacious for 9 of 19 AGITG FORECAST-1 patients (Table S8). One patient received off-label treatment with gemcitabine after exhausting standard-of-care therapy options; respective PDTO sensitivity to gemcitabine corresponded to the 61st percentile of responses as observed in our treatment-naive community reference cohort, with the patient developing PD after 126 days on therapy. PDTO assay predictions by treatment regimen and successive lines of therapy Combining our AGITG FORECAST-1 cohort and community patients receiving palliative treatment (27 patients, 39 evaluable treatments), PDTO assays achieved a prediction accuracy of 84.6% (95% CI = 69.5%–94.1%; p value [Acc > NIR] < 0.001) with 66.7% sensitivity, 95.8% specificity, 82.1% negative predictive value (NPV), and 90.9% positive predictive value (PPV) (Figure 6A). Incorrectly classified cases were not associated with any particular treatment regimen (5FU/capecitabine ± bevacizumab, 1 of 8, 12.5%; FOLFIRI/XELIRI [capecitabine and irinotecan] ± bevacizumab, 0 of 3, 0%; FOLFIRI/XELIRI + cetuximab/panitumumab, 1 of 4, 25.0%; FOLFOX/CAPOX [capecitabine and oxaliplatin] ± bevacizumab, 1 of 10, 10.0%; FOLFOX/CAPOX + cetuximab/panitumumab, 1 of 2, 50.0%; FOLFOXIRI, 0 of 3, 0%; TAS-102, 1 of 8, 12.5%). Patients receiving treatments associated with “resistant” PDTO predictions exhibited inferior progression-free survival (PFS) as compared to those receiving treatments associated with “sensitive” PDTO predictions with a hazard ratio (HR) of 3.97 (95% CI = 1.36–11.57; p = 0.007, log-rank test) (Figure 6B). This survival trend was maintained when adjusting analysis for treatment with single versus combination chemotherapy (HR = 3.63, 95% CI = 1.23–10.76, p = 0.020). Previous PDTO studies of drug combination treatments indicate that analyzing combination matrices, rather than each drug separately, more accurately discriminates patient clinical responses.11 Accordingly, for FOLFOX and FOLFIRI treatments, our combined cohort analysis showed higher concordance of drug combination matrices with treatment responses (84.2%, 16 of 19) as compared to single-agent components (63.2%, 12 of 19) (Figure S3).Figure 6PDTO assay predictions for the combined AGITG FORECAST-1 and community mCRC patient cohorts(A) Pie charts detailing PDTO assay accuracy, sensitivity, specificity, PPV, and NPV for 27 patients with mCRC with 39 evaluable treatments.(B) Kaplan-Meier plot of progression-free survival divided into sensitive and resistant groups according to PDTO assay predictions. Statistical significance was attributed to values of p < 0.050 as determined by the log-rank test.(C) Comparison of drug sensitivity between PDTOs from chemo-naive patients (n = 56) and PDTOs from patients refractory to 5FU-based treatment (5FU, FOLFIRI, and FOLFOX; n = 18) for 5FU, FOLFIRI, FOLFOX, TAS-102, regorafenib, and erlotinib. Statistical significance was attributed to values of p < 0.050 as determined by Student’s t test. ∗p < 0.050.See also Figures S3 and S4. PDTO assay predictions for the combined AGITG FORECAST-1 and community mCRC patient cohorts (A) Pie charts detailing PDTO assay accuracy, sensitivity, specificity, PPV, and NPV for 27 patients with mCRC with 39 evaluable treatments. (B) Kaplan-Meier plot of progression-free survival divided into sensitive and resistant groups according to PDTO assay predictions. Statistical significance was attributed to values of p < 0.050 as determined by the log-rank test. (C) Comparison of drug sensitivity between PDTOs from chemo-naive patients (n = 56) and PDTOs from patients refractory to 5FU-based treatment (5FU, FOLFIRI, and FOLFOX; n = 18) for 5FU, FOLFIRI, FOLFOX, TAS-102, regorafenib, and erlotinib. Statistical significance was attributed to values of p < 0.050 as determined by Student’s t test. ∗p < 0.050. See also Figures S3 and S4. Finally, we probed for impact of treatments in modulating responses to further therapy in exploratory analysis. PDTOs generated from patients who were refractory to 5FU-based regimens in previous lines of therapy (5FU/FOLFIRI/FOLFOX, n = 33) were less sensitive to 5FU-based regimens as compared to PDTOs from treatment-naive patients (n = 56), but this did not appear to influence sensitivity to unrelated treatment regimens (Figure 6C). With guidelines for mCRC treatment favoring successive use of unrelated therapy options, we did not observe increased misclassification rates with successive patient treatments post-PDTO assay, although sample size was limited (18.5%, 5 of 27 for immediate lines versus 8.3%, 1 of 12 for subsequent lines) (Figure S4). Accordingly, the overall trend to inferior PFS for patients with “resistant” PDTO predictions was maintained when adjusting analysis for immediate versus subsequent treatment lines (HR = 3.96, 95% CI = 1.36–11.56, p = 0.012). PDTO assay miniaturization to reduce time from biopsy to drug screen result A limitation of our comprehensive PDTO assay setup highlighted in AGITG FORECAST-1 was the time taken from tissue biopsy to drug screen results. With input cell number being the main limiting factor, we explored reduction of dose titration points by selecting concentration windows for single agents and combinations at which the variance of in vitro drug sensitivity profiles was the largest for our treatment-naive reference cohort. We then performed response prediction for our independent AGITG FORECAST-1 cohort and community patients receiving palliative treatments (27 patients, 39 evaluable treatments). Prediction accuracies are detailed for decreasing titration window sizes from 5 to 2 single points in Table S9. Across all window sizes, PDTO assay performance and association with PFS improved as compared to the original full dose titrations (Figures 7A and 7B), indicating that a 2-point window was adequate for miniaturization of standard-of-care drug testing (requiring ∼144,000 cells). Based on PDTO establishment rates and growth trajectories observed for AGITG FORECAST-1 biopsy samples, this would enable assay results to be generated within a 7-week time frame for 68.8% (11 of 16) patients with CRC liver metastasis but only 28.6% (4 of 14) patients with metastases from other sites (Figure 7C).Figure 7PDTO assay miniaturization to reduce time from biopsy to drug screen result(A) PDTO assay sensitivity, specificity, PPV, NPV, and accuracy for decreasing drug titration window sizes from 5 to 2 single points for 27 patients with mCRC with 39 evaluable treatments from the combined AGITG FORECAST-1 and community mCRC patient cohorts.(B) Kaplan-Meier plot of progression-free survival divided into sensitive and resistant groups according to PDTO assay predictions for decreasing window sizes. Statistical significance was attributed to values of p < 0.050 as determined by the log rank test. ∗p < 0.050.(C) Growth curves of successfully assayed PDTOs from AGITG FORECAST-1 biopsy samples. Horizontal dashed lines indicate the number of cells required to test PDTOs for standard-of-care treatments for decreasing assay window sizes. Vertical dashed lines indicate time to PDTO assay commencement with reporting at 7 weeks. CRC liver metastases are shown in blue, and metastases from other sites are in gray. Percentages of biopsy samples estimated to be tested and reported within 7 weeks are shown for liver and other metastatic sites.See also Table S9. PDTO assay miniaturization to reduce time from biopsy to drug screen result (A) PDTO assay sensitivity, specificity, PPV, NPV, and accuracy for decreasing drug titration window sizes from 5 to 2 single points for 27 patients with mCRC with 39 evaluable treatments from the combined AGITG FORECAST-1 and community mCRC patient cohorts. (B) Kaplan-Meier plot of progression-free survival divided into sensitive and resistant groups according to PDTO assay predictions for decreasing window sizes. Statistical significance was attributed to values of p < 0.050 as determined by the log rank test. ∗p < 0.050. (C) Growth curves of successfully assayed PDTOs from AGITG FORECAST-1 biopsy samples. Horizontal dashed lines indicate the number of cells required to test PDTOs for standard-of-care treatments for decreasing assay window sizes. Vertical dashed lines indicate time to PDTO assay commencement with reporting at 7 weeks. CRC liver metastases are shown in blue, and metastases from other sites are in gray. Percentages of biopsy samples estimated to be tested and reported within 7 weeks are shown for liver and other metastatic sites. See also Table S9. Discussion In this study, we have developed a unified framework for predictive PDTO standard-of-care drug testing and clinical reporting of drug sensitivity for patients with mCRC. Previous prospective studies have investigated PDTO assays for limited chemotherapy options with training based on observed clinical responses, including first-line 5FU-oxaliplatin and second-line irinotecan-based treatment (TUMOROID study13) or joint prediction for neoadjuvant and adjuvant FOLFOX/FOLFIRI/FOLFOXIRI therapy.21 In contrast, we developed a predictive PDTO assay approach applicable across standard-of-care chemotherapy and biologic and targeted therapy options derived from a combination of PDTO assay results from an initial cohort of treatment-naive CRC community patients and response rates reported in first-line mCRC clinical trials. Successful application in an independent community-based cohort of patients with mCRC receiving routine palliative treatment achieved an overall PDTO assay accuracy of 83%. Further prospective validation of our PDTO assay framework in the AGITG FORECAST-1 study of third-line or later-line mCRC treatment also achieved 90% accuracy, although the number of evaluable patients was limited, and expanded validation studies will be required to further validate assay accuracy statistics. While AGITG FORECAST-1 was an observational study, alternative, potentially efficacious treatment options were identified for 9 of 19 patients with PDTO assay results. Consistent with studies propagating PDTOs in Matrigel domes,14,18,34 PDTOs grown in LVM culture conditions exhibited similar marker expression as the original tumor, including Ki67 proliferative index and staining patterns for p53 and Muc2. Our genome-wide sequencing analysis showed a median SNV overlap of 72.2% (range, 49.4%–87.8%) between PDTO and matched tumor samples. A previous study by van de Wetering et al. assessing exome-wide SNV overlap between PDTOs and original tumors had reported a median concordance of 88% (range, 62%–100%) but with manual curation of mutation data34; re-examination of deposited mutation calls using the same in silico filtering approach as for our analysis yielded a median SNV overlap of 85.7% (range, 0.0%–97.6%) (Figure S5A). Moreover, analysis of deposited exome sequencing data on PDTOs and matched tumors from Yao et al. yielded a median SNV overlap of 76.3% (range, 26.8%–92.8%), similar to our findings (Figure S5B).14 Consistent with other studies,12,14,16,34 we observed a high concordance for the mutational spectrum in major CRC driver genes (97.5%, range, 81.0–100%). In our study, two cases (WCB266 and WCB088) had more discordant SNV profiles between the PDTO and the tumor; both cases were non-hypermutated, with discordance likely related to clonal molecular heterogeneity as previously reported by others.35,36 Previous studies sampling CRC biopsies from metastatic sites have reported culture success rates of 54%–63% using culture methods with PDTO embedment into solid support matrices.13,37,38 In the AGITG FORECAST-1 study, we achieved a similar 63% overall success rate for PDTO LVM cultures25 and identified CRC liver metastases as the optimal biopsy site for PDTO generation with a 75% success rate (compared to 50% for other sites). Nonetheless, PDTO culture success rate remains a major barrier to eventual clinical translation with biopsy size, tumor cell content, and viability established as critical factors for success18,38; further improvements in PDTO success rate may be achieved by increasing biopsy number and/or size and by further optimization of culture conditions such as replacement of classic niche factors with emerging biomimetic compounds or introduction of new biopsy methods.35,39,40,41 The duration of PDTO expansion prior to in vitro treatments depends on tumor cell proliferation rate and the required cell input and number of assay points.13,37,38 For AGITG FORECAST-1, we required 4–21 weeks to examine 9 single agents in duplicate 9-step dose titrations and 2 drug combinations in duplicate 7 × 7 matrices. However, PDTO assay miniaturization to a 2-point window was found to maintain and potentially improve prediction accuracy and association with PFS. Based on PDTO establishment and growth rates observed in AGITG FORECAST-1, a 7-week turnaround time for PDTO standard-of-care drug testing thus appears feasible for an estimated 69% patients with CRC liver metastases. This time frame is in line with that of a recent phase 2, single-arm study of PDTO-guided use of investigational agents in refractory mCRC (ClinicalTrials.gov: NCT03251612), which achieved treatment assignment within a median of 51 days from enrollment for 38% study participants and reported improved PFS as compared to benchmark clinical trial data.37 Additional efforts in assay miniaturization, including exploration of emerging organoid-on-a-chip approaches,42,43 will be required for PDTO drug sensitivity testing to meet timelines required for clinical implementation. Although the PDTO assay achieved an overall accuracy of 84.6% with a PPV of 90.9% and an NPV of 82.1% for the combined community and FORECAST- 1 cohorts, sensitivity was limited to 66.7%, with a total of five false negative cases where PDTOs showed resistance while the patients did show clinical benefit. In 2 out of 5 cases, PDTO assay results were close to the prediction cutoff (within 5% viability), indicating bona fide misclassification based on the current model; further cutoff tuning with an expanded patient training cohort may improve model performance. In 3 out of 5 cases, assay results demonstrated definitive PDTO resistance, with discordance with patient outcome likely related to clonal heterogeneity as similarly reported in previous studies assessing multiple tumor biopsies or metastatic lesions13,18,34; addressing this deficiency will require multiple-lesion sampling and assessment. Among non-standard-of-care agents for mCRC explored in our study, limited PDTO responsiveness was observed for temozolomide and pemetrexed. However, a subset of PDTOs was sensitive to gemcitabine treatment, consistent with previous data from the APOLLO and ClinicalTrials.gov: NCT03251612 studies.19,37 One patient treated with gemcitabine in AGITG FORECAST-1 after failure of standard-of-care therapy exhibited PDTO sensitivity corresponding to the 61st percentile of responses in our treatment-naive reference cohort, and this patient demonstrated some treatment benefit, developing PD after 126 days on therapy. Similarly, both the APOLLO and ClinicalTrials.gov: NCT03251612 studies reported evidence of patient benefit for PDTO-guided gemcitabine treatment.19,37 While current prospective data indicate that PDTO drug screen results for patients with mCRC are predictive for irinotecan-based treatment,13,21 findings for oxaliplatin-based treatment remain contentious, with the TUMOROID study indicating poor performance.13 In combined cohort analysis, we found no evidence of differential misclassification rates between FOLFIRI and FOLFOX treatments; misclassification rates were also largely similar for other treatment regimens, although patient numbers were small, precluding definitive analysis. Future work will be required to more robustly evaluate the clinical utility of our PDTO assay framework across the various treatment lines and regimens for mCRC. Earlier lines of treatment may influence responses for later treatment lines. Our data demonstrate acquisition of tumor resistance for 5FU-based therapy, but this did not appear to impact sensitivity to other unrelated treatment regimens in our small study cohort. Nonetheless, given this limitation, retesting of PDTOs for later treatment lines remains warranted, as any therapy post-PDTO establishment has the potential to impact response to future post-PDTO treatments. Current clinical and molecular biomarkers for personalization of mCRC therapy are limited to targeted therapies including left-sided primary cancer location and KRAS/NRAS mutation status for anti-EGFR antibody therapy, BRAFV600E mutation status for BRAF-targeted therapies, and HER2/ERBB2 amplification status for HER2-targeted therapies.2,3,4,5,6,7,8,9,10 Even among patients treated with genomically guided therapies, response rates are limited, and matching of alternative investigational agents by tumor genomic sequencing alone has proven challenging.44 Advances in PDTO technology hold promise for guiding use of routine cancer treatments, enabling prediction for chemotherapy agents for which no biomarkers are currently available and complementing genomic testing for targeted therapies. Our framework for predictive PDTO drug sensitivity testing and clinical reporting is adaptable across the range of standard-of-care therapy options and treatment lines for patients with mCRC, but biopsy culture success rate and time to assay results require further improvement for clinical implementation. Although we observed high assay accuracy in AGITG FORECAST-1, patient deterioration and low response rates constrain the benefit of PDTO standard-of-care drug testing to predict response in the typically treatment refractory third-line setting. While there remains opportunity for clinical gains in the late-line setting, particularly for investigational agents, our findings suggest that the greatest potential for standard-of-care PDTO testing is first- and second-line treatment selection, where equipoise exists for multiple therapy options. Implementation in early lines of treatment would provide the greatest opportunity for benefit of PDTO-guided therapy while avoiding the use of ineffective, toxic, and costly treatments. Early exploration of potential alternative investigational approaches would further provide a larger time frame to source off-label agents for patients and to enable accelerated targeted conduct of confirmatory PDTO assays upon failure of standard-of-care therapy options. Ultimately, the clinical benefit and cost effectiveness of PDTO drug testing compared to standard physician-guided treatment will need to be demonstrated in a randomized clinical trial. Limitations of the study Limitations of this study include that PDTO growth in culture medium lacking Wnt ligands to prevent the outgrowth of normal epithelium resulted in lower success rates for MMR-deficient tumors, which tend to maintain dependence on Wnt-ligand-mediated Wnt/β-catenin signaling due to the acquisition of loss-of-function mutations in RNF43.45 The determination of PDTO assay cutoff values on treatment-naive patients included stage I–IV tumors and may be improved in future models solely trained on metastatic samples. Prediction of anti-EGFR antibody response was based on erlotinib as a surrogate, which cannot model antibody-dependent cellular cytotoxicity, although we verified concordance with clinical cetuximab sensitivity and resistance associated with RAS mutation status. Currently, there are no data on TAS-102 as a single agent for first-line treatment of patients with mCRC, and objective response rate was estimated based on results of the TASCO-1 trial of TAS-102 plus bevacizumab and first-line data for bevacizumab used in combination with other chemotherapy agents.46,47 The BRAF inhibitor encorafenib and the HER2-targeted therapies tucatinib and trastuzumab were not assessed in the current study as these were not available to our patients at the time of study conduct. Anti-angiogenic and immune therapies could not be assessed in a monoculture PDTO assay, and assessing sensitivity to these agents will require development of appropriate co-culture assays. While our current data suggest assay utility for diverse standard-of-care regimens based on limited patient numbers, expanded studies will be required to formally demonstrate accuracy for each treatment option. Similarly, our analyses of PFS need to be considered exploratory, with types and sequence of treatments as potential confounders that could only be partly addressed in adjusted analyses. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesMouse anti-Ki67 antibodyDAKOCat#M7240, RRID: AB_2142367Mouse anti-MUC2 antibodySanta Cruz BiotechnologyCat#sc-7314, RRID: AB_627970Mouse anti-p53 antibodySanta Cruz BiotechnologyCat#sc-126, RRID: AB_628082Biological samplesHuman colorectal cancer tissuesThis paperN/AHuman normal colon tissuesThis paperN/ABuffy coat from patient blood samplesThis paperN/AChemicals, peptides, and recombinant proteinsGlutaMax supplementGibcoCat#35050061HEPESSigma-AldrichCat#H3375-1KGB-27 supplementLife TechnologiesCat#17504001N2 supplementLife TechnologiesCat#17502001NicotinamideSigma-AldrichCat#72340-1KGN-acetyl-L-cysteineSigma-AldrichCat#A7250-1KGRecombinant human basic FGFLife TechnologiesCat#PHG0263Recombinant human EGFLife TechnologiesCat#PHG0313A83-01Tocris BioscienceCat#2939PrimocinInvivogenCat#ant-pm-2Y27632Stemcell TechnologiesCat#72308MatrigelCorningCat#356234Penicillin-streptomycinLife TechnologiesCat#15140122NormocinInvivogenCat#ant-nr-2Dispase IILife TechnologiesCat#17105-041Collagenase IVLife TechnologiesCat#17104019TrypLE Express enzymeThermo Fisher ScientificCat#12604021Tissue-Tek O.C.T CompoundEmgridCat#4583HistoGelThemo Fisher ScientificCat#22-110-678Critical commercial assaysISOLATE II Genomic DNA KitBiolineCat#BIO-52067DIPSEQ platform (BGI)BGIN/AMGIEasy Universal DNA Library Prep SetMGICat#1000006986LookOut Mycoplasma PCR Detection KitSigma-AldrichMP0035-1KTDako Omnis platformAgilentN/ADeposited dataWhole genome sequencing dataThis paper; The CNGB Sequence Archive (CNSA) of China National GeneBank DataBase (CNGBdb)Accession numbers of whole-genome sequencing data in CNGBdb: CNP0001424 and CNP0003751 (https://db.cngb.org/cnsa/)hg38GCA_000001405.15_GRCh38_no_alt_analysis_set.fna downloaded from HMFTools-Resourceshttps://nextcloud.hartwigmedicalfoundation.nl/s/LTiKTd8XxBqwaiC?path = %2FHMFTools-ResourcesExperimental models: Organisms/strainsPatient-derived colorectal cancer organoidsThis paperN/ASoftware and algorithmsSOAPnuke (v2.1.7)Chen et al.48https://github.com/BGI-flexlab/SOAPnuke/releases/tag/SOAPnuke2.1.7Sentieon Genomics software (version: sentieon-genomics-202112)Freed et al.49https://support.sentieon.com/versions/202112/manual/DNAseq_usage/dnaseq/BWA MEMLi et al.50https://arxiv.org/abs/1303.3997SamtoolsLi et al.51https://samtools.sourceforge.net/Picard“Picard Toolkit.” 2019. Broad Institute, GitHub Repository.http://broadinstitute.github.io/picard/; RRID:SCR_006525GATKVan der Auwera et al.52https://currentprotocols.onlinelibrary.wiley.com/doi/10.1002/0471250953.bi1110s43Hartwig toolsPriestley et al.53https://github.com/hartwigmedical/hmftoolsHMFTools-ResourcesHartwig Medical Foundationhttps://nextcloud.hartwigmedicalfoundation.nl/s/LTiKTd8XxBqwaiC?path = %2FHMFTools-ResourcesSAGE (v3.0.1)Priestley et al.53https://github.com/hartwigmedical/hmftools/tree/master/sagePAVE (v1.2.1)Hartwig Medical Foundationhttps://github.com/hartwigmedical/hmftools/releases/tag/pave-v1.2GRIDSS (v2.13.2-1)Cameron et al.54https://github.com/PapenfussLab/gridssGRIPSS (v2.1)Hartwig Medical Foundationhttps://github.com/hartwigmedical/hmftools/releases/tag/gripss-v2.1PURPLE (v3.4)Priestley et al.53https://github.com/hartwigmedical/hmftools/releases/tag/purple-v3.4AMBER (v3.9)Priestley et al.53https://github.com/hartwigmedical/hmftools/releases/tag/amber-v3.9COBALT (v1.13)Priestley et al.53https://github.com/hartwigmedical/hmftools/releases/tag/cobalt-v1.13LINX (v1.19)Shale et al.55https://github.com/hartwigmedical/hmftools/releases/tag/linx-v1.19ImageJ software (NIH Image)Schindelin et al.56https://ImageJ.nih.gov/ij/R (version 4.2.0)R Development Core Team57https://www.R-project.org/NIS-Elements AR software (version 5.21.03, 64-bit)NikonN/AOtherMANTIS Microfluidic Liquid DispenserFormulatrixCat#0695JANUS G3 Liquid Handler WorkstationPerkinElmerG3384 PinTool addition accessoryPerkinElmerN/ANikon Eclipse Ti2 Inverted Microscope SystemNikonEclipse Ti2 REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesMouse anti-Ki67 antibodyDAKOCat#M7240, RRID: AB_2142367Mouse anti-MUC2 antibodySanta Cruz BiotechnologyCat#sc-7314, RRID: AB_627970Mouse anti-p53 antibodySanta Cruz BiotechnologyCat#sc-126, RRID: AB_628082Biological samplesHuman colorectal cancer tissuesThis paperN/AHuman normal colon tissuesThis paperN/ABuffy coat from patient blood samplesThis paperN/AChemicals, peptides, and recombinant proteinsGlutaMax supplementGibcoCat#35050061HEPESSigma-AldrichCat#H3375-1KGB-27 supplementLife TechnologiesCat#17504001N2 supplementLife TechnologiesCat#17502001NicotinamideSigma-AldrichCat#72340-1KGN-acetyl-L-cysteineSigma-AldrichCat#A7250-1KGRecombinant human basic FGFLife TechnologiesCat#PHG0263Recombinant human EGFLife TechnologiesCat#PHG0313A83-01Tocris BioscienceCat#2939PrimocinInvivogenCat#ant-pm-2Y27632Stemcell TechnologiesCat#72308MatrigelCorningCat#356234Penicillin-streptomycinLife TechnologiesCat#15140122NormocinInvivogenCat#ant-nr-2Dispase IILife TechnologiesCat#17105-041Collagenase IVLife TechnologiesCat#17104019TrypLE Express enzymeThermo Fisher ScientificCat#12604021Tissue-Tek O.C.T CompoundEmgridCat#4583HistoGelThemo Fisher ScientificCat#22-110-678Critical commercial assaysISOLATE II Genomic DNA KitBiolineCat#BIO-52067DIPSEQ platform (BGI)BGIN/AMGIEasy Universal DNA Library Prep SetMGICat#1000006986LookOut Mycoplasma PCR Detection KitSigma-AldrichMP0035-1KTDako Omnis platformAgilentN/ADeposited dataWhole genome sequencing dataThis paper; The CNGB Sequence Archive (CNSA) of China National GeneBank DataBase (CNGBdb)Accession numbers of whole-genome sequencing data in CNGBdb: CNP0001424 and CNP0003751 (https://db.cngb.org/cnsa/)hg38GCA_000001405.15_GRCh38_no_alt_analysis_set.fna downloaded from HMFTools-Resourceshttps://nextcloud.hartwigmedicalfoundation.nl/s/LTiKTd8XxBqwaiC?path = %2FHMFTools-ResourcesExperimental models: Organisms/strainsPatient-derived colorectal cancer organoidsThis paperN/ASoftware and algorithmsSOAPnuke (v2.1.7)Chen et al.48https://github.com/BGI-flexlab/SOAPnuke/releases/tag/SOAPnuke2.1.7Sentieon Genomics software (version: sentieon-genomics-202112)Freed et al.49https://support.sentieon.com/versions/202112/manual/DNAseq_usage/dnaseq/BWA MEMLi et al.50https://arxiv.org/abs/1303.3997SamtoolsLi et al.51https://samtools.sourceforge.net/Picard“Picard Toolkit.” 2019. Broad Institute, GitHub Repository.http://broadinstitute.github.io/picard/; RRID:SCR_006525GATKVan der Auwera et al.52https://currentprotocols.onlinelibrary.wiley.com/doi/10.1002/0471250953.bi1110s43Hartwig toolsPriestley et al.53https://github.com/hartwigmedical/hmftoolsHMFTools-ResourcesHartwig Medical Foundationhttps://nextcloud.hartwigmedicalfoundation.nl/s/LTiKTd8XxBqwaiC?path = %2FHMFTools-ResourcesSAGE (v3.0.1)Priestley et al.53https://github.com/hartwigmedical/hmftools/tree/master/sagePAVE (v1.2.1)Hartwig Medical Foundationhttps://github.com/hartwigmedical/hmftools/releases/tag/pave-v1.2GRIDSS (v2.13.2-1)Cameron et al.54https://github.com/PapenfussLab/gridssGRIPSS (v2.1)Hartwig Medical Foundationhttps://github.com/hartwigmedical/hmftools/releases/tag/gripss-v2.1PURPLE (v3.4)Priestley et al.53https://github.com/hartwigmedical/hmftools/releases/tag/purple-v3.4AMBER (v3.9)Priestley et al.53https://github.com/hartwigmedical/hmftools/releases/tag/amber-v3.9COBALT (v1.13)Priestley et al.53https://github.com/hartwigmedical/hmftools/releases/tag/cobalt-v1.13LINX (v1.19)Shale et al.55https://github.com/hartwigmedical/hmftools/releases/tag/linx-v1.19ImageJ software (NIH Image)Schindelin et al.56https://ImageJ.nih.gov/ij/R (version 4.2.0)R Development Core Team57https://www.R-project.org/NIS-Elements AR software (version 5.21.03, 64-bit)NikonN/AOtherMANTIS Microfluidic Liquid DispenserFormulatrixCat#0695JANUS G3 Liquid Handler WorkstationPerkinElmerG3384 PinTool addition accessoryPerkinElmerN/ANikon Eclipse Ti2 Inverted Microscope SystemNikonEclipse Ti2 Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Oliver Sieber (sieber.o@wehi.edu.au). Materials availability This study did not generate new unique reagents. Data and code availability •Whole-genome sequencing data have been deposited in the CNGB Sequence Archive (CNSA) of CNGBdb (https://db.cngb.org/cnsa/)58,59 and are publicly available. Accession numbers are listed in the key resources table. Drug sensitivity test results have been provided as Table S4.•This paper does not report the original code.•Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. •Whole-genome sequencing data have been deposited in the CNGB Sequence Archive (CNSA) of CNGBdb (https://db.cngb.org/cnsa/)58,59 and are publicly available. Accession numbers are listed in the key resources table. Drug sensitivity test results have been provided as Table S4. Whole-genome sequencing data have been deposited in the CNGB Sequence Archive (CNSA) of CNGBdb (https://db.cngb.org/cnsa/)58,59 and are publicly available. Accession numbers are listed in the key resources table. Drug sensitivity test results have been provided as Table S4. •This paper does not report the original code. This paper does not report the original code. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Experimental model and study participant details The community cohort included 167 patients with stage I to IV CRC treated at the Royal Melbourne Hospital, Western Health, Eastern Health and the Northern Hospital in Melbourne, Australia, between May 1, 2017 and August 31, 2022; primary or metastatic tumor tissues and adjacent normal tissues were harvested at surgery or routine biopsy. All patients received standard treatment with clinical and outcome data routinely collected by the Australian Comprehensive Cancer Outcomes and Research Database for Colorectal Cancer (ACCORD-CRC) and Treatment of Recurrent and Advanced Colorectal Cancer (TRACC) Registry.23,24 All patients provided written informed consent, and the study was approved by medical ethics committees at all sites (HREC 2016.249, WEHI). AGITG FORECAST-1 was a prospective, observational clinical study comprising 30 patients with mCRC who had failed ≥ two lines of therapy (Clinical Trial Registration Number: ACTRN12620001353987). AGITG FORECAST-1 included the patient recruiting sites above as well as LaTrobe Regional Health service. Inclusion criteria included patient age >18 years, ability to provide informed consent, ability to safely undergo biopsy to provide fresh tumor samples, ECOG performance status of 0–2, adequate major organ function, fitness for further systemic treatment, failure of or intolerance to standard therapies (prior exposure to trifluridine/tipiracil was not required), discontinuation of prior systemic treatment at least 2 weeks prior to study biopsy, a life expectancy of >3 months, and accessibility for follow up and de-identified coded data collection. All patients were treated with palliative therapy at clinician discretion. Patient characteristics for the community and FOREACST-1 cohorts including sex and age are provided in Tables S1 and S6. PDTOs were authenticated using whole genome sequencing data, with matched normal colon tissues or matched buffy coat from patient blood samples as reference samples. All PDTO cultures were confirmed to be mycoplasma negative using the LookOut Mycoplasma PCR Detection Kit (Sigma-Aldrich, MP0035-1KT). Method details PDTO generation PDTOs were generated in low-viscosity matrix (LVM) suspension culture medium as described previously.25,60 Tumor samples obtained at resection or biopsy were washed 3–5 times with PBS, minced with scissors into pieces that can pass through 1-mL tips, digested at 37°C for 30–60 min with 0.1 mg/mL Dispase II (Life Technologies, 17105-041) and 200 U/ml Collagenase IV (Life Technologies, 17104019) in DMEM/F12 (Life Technologies, 11320082), and then washed with PBS for another 5–10 times at 200 g for 3 min before being plated into 6-well plates (Interpath, 657185) in PDTO LVM culture medium consisting of advanced DMEM/F12 (Thermo Fisher Scientific, 12634028), 1x GlutaMax supplement (Gibco, 35050061), 10 mM HEPES (Sigma-Aldrich, H3375-1KG), 1x B-27 supplement (Life Technologies, 17504001), 1x N2 supplement (Life Technologies, catalog number: 17502001), 10 mM nicotinamide (Sigma-Aldrich, 72340-1KG), 1 mM N-acetyl-L-cysteine (Sigma-Aldrich, A7250-1KG), 20 ng/mL recombinant human basic FGF (Life Technologies, PHG0263), 50 ng/mL recombinant human EGF (Life Technologies, PHG0313) and 500 nM A83-01 (Tocris Bioscience, 2939), with the addition of 100 μg/mL primocin, 10 μM Y27632 (Stemcell Technologies, 72308) and 5% Matrigel (Corning, 356234). PDTOs were cultured at 37°C in 5% CO2. Culture medium was changed every 2–3 days and PDTO cultures were agitated by pipetting to prevent adhesion of organoids to the wells. For passaging, PDTOs were collected, washed with PBS at 200 g for 3 min, digested with TrypLE Express enzyme (Thermo Fisher Scientific, 12604021) at 37°C for 10–15 min with mechanical agitation every 5 min, washed with advanced DMEM/F12 at 200 g for 3 min, and plated in PDTO culture medium containing 100 U/ml penicillin-streptomycin (Life technologies, 15140122), 100 μg/mL normocin (Invitrogen, ant-nr-2). For cryopreservation in liquid nitrogen, PDTOs cultured for 3–7 days were harvested, washed with PBS at 200 g for 3 min, resuspended in CryoStor CS10 cell freezing medium (STEMCELL Technologies, 7930) and frozen at the cooling rate of 1°C per minute in a Corning CoolCell LX Cell Freezing Container (Biocision, BCS-405) at −80°C. For long-term storage, frozen cells were transferred to liquid nitrogen. Histological assessment To prepare samples for histology, pieces of fresh CRC tissue samples were embedded in Tissue-Tek O.C.T Compound (Emgrid, 4583) and rapidly frozen by isopropyl bath on dry ice. Formalin-fixed paraffin-embedded CRC tissue samples were retrieved from pathology archives. PDTOs grown for 2–4 weeks since last passaging were collected at 200 g for 3 min, washed once with PBS at 200 g for 3 min, resuspended in pre-warmed HistoGel (thawed at 95°C and keep at 55°C before usage) (Themo Fisher Scientific, 22-110-678), fixed in 10% formalin at 4°C overnight and then preserved in PBS at 4°C until sent for paraffin embedment by WEHI histology facility. All histology samples were sectioned and stained with hematoxylin and eosin (H&E). Immunohistochemistry (IHC) was performed on a Dako Omnis platform by WEHI histology facility. For Ki67 staining, anti-Ki67 antibody (1:100, clone MIB-1, DAKO, M7240) was used; for p53 staining, anti-p53 antibody (1:100, clone DO-1, Santa Cruz Biotechnology, sc-126) was used after antigen retrieval for section slides in low PH solution (DAKO, K800521-2) at 97°C for 30 min; for Mucin 2 staining, anti-MUC2 antibody (1:100, clone CCP58, Santa Cruz Biotechnology, sc-7314) was used after antigen retrieval for section slides in low PH solution at 97°C for 30 min. Slides were scanned on a 3D Histech Panoramic Scan II histology scanner (3DHISTECH Ltd.) and image data were stored and examined using CaseViewer Software (3DHISTECH Ltd.). Whole genome sequencing and variant calling DNA was extracted from fresh-frozen tissue or blood samples with the ISOLATE II Genomic DNA Kit (Bioline, BIO-52067); tumor samples were confirmed to comprise greater than 60% neoplastic cell content by inspection of H&E-stained sections. A total of 97 samples, 21 trios (adjacent normal tissue, tumor tissue and tumor organoid) from the community cohort and 17 pairs (blood and tumor organoid) from the AGITG FORECAST-1 study, were sequenced on a DIPSEQ platform (BGI). Sequencing libraries were constructed using the MGIEasy Universal DNA Library Prep Set (MGI, 1000006986) and 2 x 100-bp paired-end sequencing was performed to yield data of estimated ≥30 x coverage.61 Data pre-processing, including removal of low-quality reads and adaptor sequences, was carried out using SOAPnuke (v2.1.7).48 High quality reads were mapped and processed for downstream analysis using Sentieon Genomics software (version: sentieon-genomics-202112)49 which includes the following optimized steps: 1) aligned hg38 (GCA_000001405.15_GRCh38_no_alt_analysis_set.fna downloaded from HMFTools-Resources) with BWA MEM50; 2) sorted alignment reads by Samtools51; 3) marked duplicate reads by Picard (http://broadinstitute.github.io/picard/); 4) indel realignment and base quality score recalibration for alignment reads by GATK52; and 5) alignment QC by Picard. Variants of paired samples (adjacent normal tissue vs. tumor tissue and adjacent normal tissue vs. tumor organoid) were called by series of Hartwig tools (https://github.com/hartwigmedical/hmftools).53 Resource files of GRCh38 for Hartwig tools were downloaded from HMFTools-Resources (https://nextcloud.hartwigmedicalfoundation.nl/s/LTiKTd8XxBqwaiC?path=%2FHMFTools-Resources) on March 10, 2022. SAGE (v3.0.1)53 and PAVE (v1.2.1) were used to call and annotate somatic and germline short variants (SNVs, MNVs and INDELs). GRIDSS54 (https://github.com/PapenfussLab/gridss, v2.13.2-1) and GRIPSS (v2.1) were used to call and filter simple and complex somatic SVs. PURPLE (v3.4)53 combines B-allele frequency (BAF) from AMBER (v3.9),53 read depth ratios from COBALT (v1.13),53 somatic SVs from GRIDSS and GRIPSS, and somatic and germline short variants from SAGE and PAVE to estimate variant allele frequency (VAF) and DNA copy number profiles. LINX (v1.19)55 was used to chain, classify, and visualize somatic SVs from PURPLE. To compare the VAF distribution of somatic short variants between tumor tissue and tumor organoid from the same patient, SAGE and PAVE were used to annotate somatic short variants (SNVs, MNVs and INDELs) in multiple tumor samples module, adjacent normal tissue was input as reference sample, tumor tissue and tumor organoid were both input as tumor sample. For SNV overlap analysis high confidence somatic mutations were retained that (a) had a minimum of 20 reads present across tumor, organoid and adjacent normal tissue at the SNV location, and (b) had an alternate allele frequency of at least 20% in at least one of the tumor tissue or tumor organoid samples. RAS, RAF mutation and DNA mismatch repair status RAS (KRAS, NRAS; exons 2, 3 and 4 in both genes) and RAF (BRAFV600E) mutation status were assessed by routine clinical sequencing assays as per Australian Medicare Benefits Schedule Item 73338. Tumor DNA mismatch repair status was determined based on standard diagnostic assays for mismatch repair proteins (MLH1, MSH2, PMS2 and MSH6) or the Bethesda consensus panel of microsatellite markers (BAT25, BAT26, D2S123, D5S346, and D17S250).62,63 For IHC, mismatch repair-deficient was defined as absence of nuclear staining in tumor cells but positive nuclear staining in stromal cells and lymphocytes for at least one mismatch repair protein. For the Bethesda consensus panel, mismatch repair-deficient status was diagnosed if instability was detected at two or more markers. Semiautomated PDTO drug assay PDTO drug profiling was performed on a previously validated, standardized drug testing platform in 384-well format with seeding of 2,000 to 3,000 single cells per well using a MANTIS Microfluidic Liquid Dispenser (Formulatrix, 0695).25 PDTOs were collected at 200 g for 3 min, washed with PBS at 200 g for 3 min, digested with TrypLE Express enzyme (Thermo Fisher Scientific, 12604021) at 37°C for 10 min, washed with DMEM/F12 at 200 g for 3 min, resuspended in PDTO LVM culture medium culture medium, and then dissociated with 26G needles and filtered through 70-μm cell strainers (Bio-Strategy, FAL 352350) to produce a single-cell suspension. Single cells were then suspended in PDTO LVM culture medium with the addition of 1x penicillin-streptomycin (Life Technologies, 15140122), 100 μg/mL normocin (Invivogen, ant-nr-2), 10 μM Y27632 and 3% Matrigel, seeded into 384-well plates (Thermo Fisher Scientific, 242764) and cultured for 3 days at 37°C in 5% CO2 before drug treatment. Using a JANUS Liquid Handler Workstation (PerkinElmer) with PinTool addition accessory (PerkinElmer), single agents were assayed in duplicate in 9 point, 4-fold dilution series with starting concentrations of 50 μM for 5-fluorouracil, 75 μM for oxaliplatin, 0.5 μM for SN38, 50 μM for regorafenib, 50 μM for TAS-102, 15 μM for gemcitabine, 10 μM for pemetrexed, 750 μM for temozolomide and 50 μM for erlotinib. FOLFOX and FOLFIRI treatments were assayed in duplicate in 7x7 drug matrices with 4-fold dilution series with starting concentrations of 50 μM for 5-fluorouracil, 75 μM for oxaliplatin and 0.5 μM for SN38 treatment components. DMSO 0.5% (v/v) and 1 μM bortezomib served as negative (vehicle) and positive (killing) controls, respectively. PDTO cultures were imaged daily for 7 days by bright-field microscopy on an automated Nikon Eclipse Ti2 Inverted Microscope System (NIS-Elements AR software version 5.21.03, 64-bit) with a 4× objective and a 25 μm z stack over a 500 μm range. Image analysis to determine PDTO sizes was performed using ImageJ software (NIH Image).56 PDTO viability normalized to negative and positive controls was calculated as (Mean PDTO sizeTreatment - Mean PDTO sizeKilling Control)/(Mean PDTO sizeVehicle Control - Mean PDTO sizeKilling Control). Dose-response curves were fitted to determine effective concentration metrics (EC). In vitro responses were quantified as area under the drug response curve (AUC) for single agents and mean ‘pan-matrix’ viability score for drug combinations. For drug response measurements using the window design, the average of the respective PDTO viability scores was calculated. Quantification and statistical analysis Study data were collected and managed using REDCap electronic data capture tools hosted by the Clinical Translation Center at WEHI.64 Statistical analyses were performed using the statistical computing software R (version 4.2.0).57 Specific analysis methods for each datasets are specified in the figure and table legends. Differences between groups were assessed using the Fisher’s exact test for categorical variables and the Student’s t test for continuous variables. Correlations between continuous variables were assessed using Pearson correlation coefficient and correlation test. Assay prediction accuracy as compared to non-informative rate was assessed using the binomial test. Data were presented as Mean ± standard deviation in bar charts. Progression-free survival (PFS) was defined as the time from treatment start date to clinician assessed disease progression with censoring done when a patient died or was alive without disease progression at last contact. Associations between PDTO drug predictions and PFS were assessed using Kaplan Meier survival curves and the log rank test; hazard ratios and 95% confidence intervals were calculated using Cox proportional hazard models. All statistical analyses were two-sided and considered significant if p < 0.050. Additional resources •AGITG FORECAST-1 was a prospective, observational clinical study comprising 30 patients with mCRC who had failed ≥ two lines of therapy (Clinical Trial Registration Number: ACTRN12620001353987).•Link to AGITG FORECAST-1 study: https://gicancer.org.au/clinical-trial/forecast-1/. •AGITG FORECAST-1 was a prospective, observational clinical study comprising 30 patients with mCRC who had failed ≥ two lines of therapy (Clinical Trial Registration Number: ACTRN12620001353987). AGITG FORECAST-1 was a prospective, observational clinical study comprising 30 patients with mCRC who had failed ≥ two lines of therapy (Clinical Trial Registration Number: ACTRN12620001353987). •Link to AGITG FORECAST-1 study: https://gicancer.org.au/clinical-trial/forecast-1/. Link to AGITG FORECAST-1 study: https://gicancer.org.au/clinical-trial/forecast-1/.