Colorectal Cancer Develops Inherent Radiosensitivity That Can Be Predicted Using Patient-Derived Organoids

The Fearon and Vogelstein model of colorectal cancer tumorigenesis defines key genetic changes associated with multistep progression from normal colorectal tissue to early and late adenoma, primary carcinoma and metastatic dissemination (1). While ensuing research [e.g., The Cancer Genome Atlas (TCGA) project; ref. 2] expanded the repertoire of genetic and epigenetic alterations driving colorectal cancer progression, systematic stage-by-stage assessment of changes governing colorectal cancer response to clinical treatment is limited. Before surgical resection, patients with locally-advanced rectal cancer receive neoadjuvant therapy that includes chemotherapy and radiation (nCRT). Patients with a clinical complete response (cCR) to nCRT at our cancer center [Memorial Sloan Kettering Cancer Center (MSKCC), New York, NY] have been successfully managed with a nonoperative “Watch & Wait” approach. This “Watch & Wait” approach allows for preservation of rectum, avoidance of radical surgical resection and associated potential morbidity, and maintenance of bowel function and quality of life by avoidance of a temporary or permanent colostomy (3). Extending benefits of “Watch & Wait” to additional patients will require better patient selection and refinement of neoadjuvant protocols, both of which are hindered by lack of experimental platforms that predict outcome of nCRT therapy.

Recent emergence of organoid technology has bridged the gap between cancer cell lines and xenografts in preclinical studies. A large body of literature now documents patient-derived organoids (PDO) as faithfully recapitulating functional and mutational spectra of parent colon and rectal tumors in vivo (4–7). Importantly, use of the organoid system allows for enrichment of the cancer cell population apart from the neighboring niche environment, including stroma, microvasculature, and the immune system. As such, PDOs provide a platform to study inherent properties of tumor cells.

Several groups have documented the establishment of colon and rectal cancer PDO biobanks as tools for therapeutic response prediction to chemotherapy and radiation in a retrospective fashion (6–10). While these studies mainly focus on evaluation of treatment response of individual patient organoids to fulfill the promise of personalized cancer medicine, none of these studies provides a systematic comparison along the multistage progression of colorectal cancer from the original normal tissue to benign adenoma, malignant adenocarcinoma, and advanced cancer. Here, we investigate radiation responses of normal human large intestine, adenoma and primary colon and rectal cancer PDOs generated from endoscopic biopsies at time of pathologic diagnosis, and post-neoadjuvant PDOs isolated at time of definitive surgical resection.

A problem in the development of predictive assays for radiation effects on human tissues is that tissue survival responses to ionizing radiation are nonlinear with dose, but rather yield tissue-specific descending curves that hinder straightforward analysis of biologic effects. Cutting edge radiobiology has addressed this problem by adopting full dose survival profiling and devised mathematical algorithms to fit curves to observed data to generate highly reproducible numerical data that accurately define clinically relevant inherent radiosensitivities. Recently, we established protocols for use of the single hit multi-target (SHMT) algorithm, which generates a single value, the D₀, a measure of inherent organ radiosensitivity, for evaluating irradiated murine organoids, and showed highly similar radiation dose survival curves of small and large intestinal crypts in vivo and their cognate organoids (11, 12). Note the higher the D₀ value, the greater the radioresistance.

Here, we adapt the SHMT analytic approach for human use with PDOs. Further we apply ionizing radiation–induced foci (IRIF) technology to provide evidence that efficiency of resolution of γH2AX foci, the “gold standard” for assessing repair of DNA double-strand breaks (DSB), correlates closely with D₀ values generated by SHMT fitting of organoid radiation dose survival curves (i.e., radiosensitive tissues repair poorly while radioresistant tumors repair effectively). Using these technologies, we make the following discoveries: During progression from normal human colonic tissue to adenoma, which results in logarithmic expansion of the Lgr5⁺ crypt base columnar stem cell population as we published (13), inherent radioresistance is maintained. However, upon transformation to adenocarcinoma broad radiation sensitivity occurs in a large subset of patients, an observation that correlates with mutational signature 3 [defective in homologous recombination (HR)] upon genomic analysis. To our knowledge, this is the first report of development of inherent radiosensitivity upon human adenoma to carcinoma progression in any tumor type. On the basis of these observations, we performed a pilot clinical trial and show that rectal cancer patient responses to neoadjuvant chemoradiation, including patients that exhibit complete responses, correlate closely with D₀.

The source of all reagents, chemicals, and biological samples used in this study are listed in the Supplementary Materials Key Resource Table.

The study was conducted in accordance with recognized ethical guidelines under MSKCC Institutional Review Board protocol 15–191 and 16–1071. All patients involved in this study provided written informed consent before the derivation of their normal colonic, adenoma, or tumor tissues at time of endoscopic examination or surgical resection at MSKCC. The diagnosis was confirmed by the Pathology Core Facility. Please see Supplementary Table S1 for clinical details of organoids lines used in this study.

This method has been described earlier in (7). In brief, surgically or endoscopically resected tumor tissues, as well as adjacent normal mucosa, were washed with cold PBS-Abs (PBS containing 100 U/mL of antibiotics penicillin and streptomycin). Tissues were stripped of underlying muscle layers and submucosa with surgical scissors and chopped into ∼1-mm pieces in cold PBS-DTT (PBS with 10 mmol/L dithiothreitol). The supernatant was removed after centrifuging at 300 × g for 5 minutes, and then, the normal mucosa and tumor tissues were processed differently as follows: For normal mucosa, samples were resuspended in PBS-EDTA-DTT (PBS with 8 mmol/L EDTA and 0.5 mmol/L DTT) and incubated on a rotator at 4°C for 1 hour. After spinning at 300 × g for 5 minutes, EDTA solution was removed, and pelleted tissues were resuspended in cold ADF-FBS medium [ADF basal medium (see below) and 2% FBS] and vigorously shaken several times to release crypts from mucosal tissues, followed by sedimentation under gravity for 1 minute. Supernatant medium containing released crypts was transferred to a new tube for microscopic inspection. This resuspension–sedimentation cycle was repeated until no crypts were detected in supernatant medium. Crypt-containing supernatant fractions were pooled, filtered through a 100-μm cell strainer and centrifuged at 300 × g for 5 minutes; for tumor, samples were incubated with digestion buffer (Advanced DMEM/F12 medium with 2% FBS, 100 U/mL penicillin/streptomycin, 500 U/mL collagenase, and 125 μg/mL Dispase type II) for 30 minutes at 37°C, shaking every 5 minutes. Tumor samples were washed with 10 mL of ADF-FBS medium and spun at 300 × g for 5 minutes. Normal crypt pellet and tumor cell pellet were resuspended in Matrigel and plated into 24-well culture plates (40 μL droplet per well). After the gel solidified, 500 μL of WENR or EN medium (see below) was added to wells and incubated at 37°C.

ADF basal medium was prepared as follows: Advanced DMEM/F12 medium plus 1 mmol/L GlutaMAX, 1 mmol/L HEPES, and 100 U/mL penicillin/streptomycin. Complete “WENR” medium for normal PDOs was freshly prepared every 1 to 2 weeks as follows: ADF medium was supplemented with 50% Wnt-3A conditioned medium, 10% R-spondrin-1 conditioned medium, 10% Noggin conditioned medium (all three conditioned media were collected from cultures of HEK293 cells expressing recombinant Wnt-3A, R-spondin-1, or Noggin proteins, respectively), 10 nmol/L gastrin I, 500 nmol/L A83–01, 10 mmol/L SB202190, 10 mmol/L nicotinamide, 1X B27 supplement, 1X N2 supplement, 1 mmol/L N-acetyl cysteine, and 50 ng/mL human recombinant EGF. For tumor PDOs, “EN” medium lacking Wnt-3A and R-spondrin-1 was used. During the first three days of culture establishment, 10 μmol/L of the ROCK inhibitor Y27632 was added into the medium to increase organoid survival.

Organoid cultures were routinely passaged every 7 to 10 days, depending on culture density. Organoid-Matrigel mixtures were collected into 15-mL tubes and incubated with Cell Recovery Solution on ice for 40 to 60 minutes to dissolve Matrigel. Released organoids were spun down at 60 × g for 5 minutes, resuspended in 1 to 2 mL of PBS, and broken down into small fragments by physical pipetting 30X with a p1000 pipette. In some cases, organoids were further dissociated with TrypLE Express Solution at 37°C for 3 minutes to obtain smaller fragments or homogeneous single cells. Resulting fragments and cells were washed with 10 mL of ADF medium, pelleted at 300 × g for 5 minutes, resuspended with Matrigel and plated into new 24-well plates with a typical split ratio = 1:3–1:4 v/v. Culture medium was routinely refreshed every 2 to 3 days. For long-term storage, organoid fragments or dissociated single cells were resuspended in cold Recovery Cell Culture Freezing Medium, gradually frozen down to –80°C, and stored in liquid nitrogen.

All animal studies were approved by MSKCC Institutional Animal Care and Use Committee. The azoxymethane/dextran sulfate sodium (DSS) murine model is ideal for studying colorectal tumorigenesis as it recapitulates the aberrant crypt foci observed in the adenoma to carcinoma sequence that occurs in human colorectal cancer (14). In this two-stage model, azoxymethane, an extrinsic carcinogen, induces activating mutations in downstream elements of the Wnt signaling program (e.g., β-catenin), while the sulfated polysaccharide DSS initiates a chronic colitis that promotes tumor development (15). Here, we slightly modify the standard azoxymethane/DSS protocol (16, 17), electing a schedule of decreasing DSS dosing to mitigate toxicity. Six- to eight-week-old male Lgr5-lacZ mice were injected intraperitoneally with azoxymethane (Sigma) at 12.5 mg/kg body weight. After 5 days, mice were fed 2% DSS (MP Biochemicals; molecular mass 36–50 KDa) in drinking water for 5 days, followed by 14 days of regular water. This cycle was repeated twice with 1.5% DSS and 1.0% DSS, respectively. After 90 days, mice were euthanized, and crypt isolation and culture were performed as published (11). Briefly, distal colon was removed and flushed with cold PBS-Abs. Colonic adenomas were sliced into 1- to 3-mm pieces, resuspended in 10 mL digestion buffer (DMEM medium with 500 U/mL collagenase and 1% FBS), and incubated at 37°C for 30 minutes, shaking every 5 minutes. The resulting solution was vigorously shaken to liberate crypts, allowed to settle under gravity for 1 minute and the supernatant inspected for crypts under brightfield microscopy. After 3 cycles of the resuspension/sedimentation procedure, supernatants containing liberated crypts/crypt-like structures (from adenomas) were pooled, passed through a 70-μm cell strainer, and spun at 4°C at 300 × g for 5 minutes. Pelleted crypts/crypt-like structures were washed three times with cold DMEM containing 1% FBS, and centrifuged at 60 × g for 3 minutes. The resulting pellet was mixed with Matrigel and plated into a 24-well plate. 500 μL of mouse EN medium was added into each well and incubated at 37°C. Mouse organoid cultures, like human PDOs, were passaged every 7 to 10 days and medium was changed every 2 to 3 days. Mouse EN medium containing the following components was freshly prepared every 1 to 2 weeks: ADF medium supplemented with 10% R-spondin-1 conditioned medium, 5% Noggin conditioned medium, 1X B27 supplement, 1X N2 supplement, 1 mmol/L N-acetyl cysteine, and 50 ng/mL mouse recombinant EGF.

Detailed methodology and the mathematical algorithm are as described in our recent publication (11). In short, ionizing radiation was delivered with a Shepherd Mark-I unit (Model 68, SN643, J. L. Shepherd & Associates) operating ¹³⁷Cs sources at 1.72 Gy/minute. Organoids were passaged 1 day prior to the experiment, plated at a density of 100 to 150 organoids/well in at least triplicate wells per dose (day 0), and exposed to single fraction radiation ranging 4 to 20 or 2 to 10 Gy (day 1). Manual counting under a brightfield microscope of surviving organoids at day 6 post radiation (day 7) was used to generate classic radiation dose survival curves. Survival fraction was calculated as (number of surviving organoids)/(number of organoids in the unirradiated sample). Radiation dose survival curves were fitted to the SHMT model (18) using GraphPad Prism. This model allows to calculate D₀ (the dose required to reduce the fraction of surviving cells to 37% of its previous value) and Dq (a threshold dose below which there is no effect), as described (18). D₀ values were used in the current study to compare inherent radiosensitivities of PDOs. For the olaparib experiment, DMSO or 10 μmol/L of olaparib was added in the culture medium at the day of plating (day 0). The medium was refreshed at day 3 postirradiation (day 4), and the survival organoids were counted at day 6 postirradiation (day 7).

Genomic DNAs were extracted from liberated organoids or patient bloods using Qiagen AllPrep DNA Universal Kit, and subject to whole-exome sequencing (WES) using the Illumina HiSeq platform at MSK in-house core facility (Integrated Genomics Operation) with 150X coverage. The sequencing data were analyzed using pipelines as previously described (19). Briefly, raw sequencing data were aligned to the hg37 genome build using the Burrows-Wheeler Aligner version 0.7.17 (20). SnpEffect and SnpSift version 4.3 program were used for annotating and predicting the effects of SNPs (21). Further indel realignment, base-quality score recalibration and duplicate-read removal were performed using the Genome Analysis Toolkit (GATK) version 3.8 (22) following raw reads alignments guidelines (23). Single-nucleotide variant and indel-insertion (INDEL) mutations were called using VarScan v2.4.3 (24), Strelka v2.9.10 (25), Platypus 0.8.1 (26), Mutect2 – part of GATK 4.1.4.1 (23), and Somatic Sniper version 1.0.5.0 (27). Variants identified by at least 2 of 5 callers were reported as per TCGA Research Network recommendations (28). The criteria of variant filtering were as followed: 10× or greater coverage with ≥ 4% variant allelic fraction in tumor, and 7× or greater coverage with ≥ 99% normal allelic fraction in normal tissue, were considered high-confidence variants. Variants that did not meet these criteria but that had <20× coverage or <4% variant allelic fraction in tumor were considered low-confidence variants. Common SNPs were eliminated by comparison with snp142.vcf; whereas SNPs that were rare in dbSNP were kept when presenting in tumor and had a normal allelic fraction of zero. Common variables gnomAD v 2.1.1 (incorporated with Exome Aggregation Consortium; https://storage.googleapis.com/gnomad-public/release/2.1.1/README.txt) were excluded. Additional optimization and filtering were applied for INDELS. INDELS in blacklisted regions (https://www.encodeproject.org/annotations/ENCSR636HFF/) and low mappability regions (such as repeat maskers) were excluded as recommended by (29). Previously described mutational signatures were determined using nonnegative least-squares regression as provided by the R package deconstructSigs v1.8.0 (30) using the COSMIC signatures as the mutational signature matrix.

The day before irradiation, liberated organoids were seeded onto 8-well Lab-Tek Chambers (NUNC) precoated with 20 μL of 100% Matrigel (100–150 organoids/well), overlaid with growth medium supplemented with Matrigel (final 4%), and incubated at 37°C. After irradiation at 8 Gy, organoids were incubated for the indicated times. Immunostaining was carried out using a standard protocol and all washes/incubations were with gentle shaking. In short, organoid cultures were quickly washed with PBS and fixed with 4% paraformaldehyde for 30 minutes. After 3X washes with PBS (10 minutes each), samples were quenched with 50 mmol/L NH₄Cl for 30 minutes, followed by 2X quick PBS washes. Samples were then incubated with blocking buffer (PBS with 0.5% BSA, 0.02% Na-azide, 0.36 μmol/L DAPI, and 0.25% Triton X-100) for 3 hours at room temperature. After removing blocking buffer, organoid samples were incubated with primary antibodies (mouse antiphospho-H2AX-Ser139, rabbit antiphospho-BRCA1-Ser1387, mouse antiphospho-DNA-PKcs-T2609, or mouse anti-MDC1) in fresh blocking buffer overnight at 4°C. The unbound primary antibodies were removed by 3X PBS washes (30 minutes each), and organoids were incubated with specie-matched, Alexa-dye–conjugated secondary antibodies in fresh blocking solution for 3 hours at room temperature. After 3X PBS washes (30 minutes each), samples were mounted with ProLong Dimond Antifade reagent. Fluorescent images were acquired with a 63x/1.4NA objective oil lens on a Zeiss widefield microscope (Axio Observer.Z1) equipped with an AxioCam HRc camera. Nine consecutive sections spanning 2 μm in total were imaged, deconvoluted with a theoretical PSF using AutoQuant X software, and max-projected into a single image using Image J (FIJI). For each time point, at least 100 to 500 nuclei/sample were randomly selected from at least 7 different images for quantitation of focus numbers. Foci were scored within a nucleus whose boundary was defined automatically from a DAPI image using FIJI. The focus threshold was determined manually at 8.28 pixels as average focus size based on discreet γH2AX foci generated at a low irradiation dose (2 Gy at 30 minutes) consistent with published results (31). Due to extensive and overlapping foci, focus numbers at 30 minutes after high-dose irradiation were estimated by measuring mean γH2AX fluorescence per nucleus normalized to the standard mean pixel/focus calculated before.

Matrigel-embedded organoids were stained as per (32) with slight modification. Briefly, Matrigel-embedded organoids were washed twice with PBS for 5 minutes each and then fixed in 4% paraformaldehyde for 30 minutes. After washing with PBS 3X for 5 minutes each, organoids were permeabilized with PBST (PBS with 0.1% Triton X-100) for 30 minutes, incubated with blocking buffer (PBST with 2% BSA, 10% goat serum) for 1 hour, and stained with anti–β-catenin primary antibodies in PBS/2% BSA overnight at 4°C. After 3X washes with PBST for 5 minutes each, samples were incubated with the species-matched anti-FITC secondary antibody for 1 hour at room temperature. Samples were washed 2X with PBST and further washed 2X with PBS for 5 minutes each, stained with DAPI to visualize nuclei and mounted with VectaShield (Vector Laboratories) prior to analyses using Cytation 5 Cell Imaging Multi-Mode Reader (BioTek).

The evaluation of tumor response to treatment was assessed by experienced senior colorectal surgeons at MSKCC. Tumor response is multifactorial and based on the pathologic report at the time of tumor resection, the endoscopic imaging, the MRI before and after the treatment, and/or the CT scan (for metastatic tumor), following the principle of Response Evaluation Criteria in Solid Tumors (RECIST) and defining the response into four groups: (i) cCR—no residual tumor is found on examination with endoscopy or MRI; (ii) partial response (PR)—tumor reaches >50% decrease in the product of the area (longest diameter multiplied by greatest perpendicular diameter) is considered as major response; tumor reaches 30–50% decrease is considered as minor response; (iii) stable disease (SD)—neither PR or progressive disease (PD); (iv) PD—growth of tumor on treatment causing rectal obstruction and the measurement not applicable. The disease and survival status are based on the measure from the first day of diagnosis until the last follow up visit in our cancer center. While most patients are alive and have still been closely monitored, 2 patients were deceased before the manuscript preparation and their overall survival are noted with an exact time frame.

Values represent mean ± SD or SEM. Student t test, multiple t test, F test, Spearman correlation, Wilcoxon rank sum test, and Kruskal–Wallis rank sum test were performed using the GraphPad software or by the biostatistics at MSKCC, with P ≤ 0.05 considered statistically significant. Correlations between D₀ and clinical patient response and tumor size reduction, as well as the cut-off point for good or poor responders, were analyzed by Zhigang Zhang, Associate Professor of Biostatistics at MSKCC. Data was plotted using GraphPad Prism 9.

The data generated in this study are available upon request from the corresponding author.

Employing organoid culture technology in mouse small and large intestines, we recently established a quantitative in vitro radiation sensitivity assay that delivers radiation profiles simulating the organ of origin, as analyzed by the SHMT algorithm (11). Here, we adapt principles of this system to evaluate organoids derived from Clinical Colorectal Cancer patients. Large intestinal PDOs derived from normal human colon crypts adjacent to tumor at time of surgical resection, like murine large intestinal organoids, require WNT, EGF, Noggin, and R-Spondin for initiation and propagation in culture. Human large intestinal PDOs display a radiation-resistant phenotype similar to that observed in murine large intestinal organoids (11). In this context, death of large intestinal PDOs was minimal even at doses as high as 20 Gy (Fig. 1A). Radiation dose survival curves from 15 different large intestinal PDO lines treated with escalating doses of radiation ranging from 2 Gy to 19 Gy when analyzed using the SHMT algorithm yielded a mean D₀ (Gy) = 27.2 ± 5.0, with no significant differences between lines (Fig. 1B and Supplementary Fig. S1). Note the D₀ value is a measure of inherent radiation sensitivity, and the higher the D₀ the greater the radiation resistance.

Adenomas constitute a critical intermediary lesion in the transition from normal colonic tissue to colorectal cancer (1). We recently reported murine and human colon adenomas are characterized by logarithmic expansion of the Lgr5⁺ stem cell compartment (13), consistent with prior semiquantitative data (33). This normal to adenoma to carcinoma paradigm posits that sequential acquisition of specific genetic lesions underlies colorectal cancer progression, beginning with constitutive activation of the WNT pathway that initiates adenoma formation (34). As with murine colorectal adenoma-derived organoids, we find that the requirement for WNT and R-spondin for initiation and maintenance of patient-derived adenoma organoids from crypt-like structures is obviated (Fig. 1C), similar to what was reported in adenoma organoids from APC^fl/fl mice (35). Consistent with constitutive activation of the WNT program, organoids from human adenomas display significant nuclear enrichment of β-catenin, whereas normal large intestinal PDOs show only scattered β-catenin localization to the nucleus (Supplementary Fig. S1).

A common approach to study colorectal tumor formation in mice employs the two-stage azoxymethane/DSS model of tumorigenesis (14). In this model, mice are initially exposed to the DNA-damaging agent azoxymethane, followed by repeated exposure to the tumor promoter DSS. Here, we generated organoids from crypt-like structures derived from murine colon azoxymethane/DSS adenomas after 90 days, and compared radiation responses of murine organoids with patient-derived human adenoma organoids. Figure 1C shows that both mouse and human adenomas are highly radiation-resistant, displaying many live organoids at 8 Gy and 20 Gy. We analyzed radiation dose survival curves of 3 independent mouse adenoma lines and 5 human adenoma lines, using escalating doses of ionizing radiation ranging from 4 Gy–20 Gy. Adenomas from both species displayed significant radiation resistance, reminiscent of tissue of origin (colon), with a mean D₀ = 40.4 ± 13.0 for mouse adenomas and a mean D₀ = 32.0 ± 7.6 for human adenomas and no significant differences between mouse and human lines (Fig. 1D and Supplementary Fig. S1).

In contrast to adenoma organoids and normal large intestinal PDOs, a spectrum of radiosensitivities was observed in culture of 18 treatment-naïve colon and rectal cancer organoids. Among them, 4 were from primary colon cancers (HT-2, HT-6, HT-10, HT-15) with 2 from metastasized lesions (HM-10, HM-15) and 12 from primary rectal cancers, derived from patients who had not received treatment. These treatment-naïve colorectal cancer PDOs display wide-ranging responses when exposed to escalating radiation doses from 2 Gy to 10 Gy, yielding values from D₀ = 1.7 ± 0.3 to D₀ = 21.8 ± 11.2 (Fig. 2A and B, and Supplementary Fig. S2A) with a mean D₀ = 7.5 ± 6.3. Consistently, paired colorectal cancer and normal colon organoids isolated from the same patient displayed the same relative responses, i.e., tumor organoids were sensitive compared with adjacent normal colonic tissue (Supplementary Fig. S2A), discarding sampling bias. Of note, human PDOs maintain their phenotype for at least 12 passages and survive freeze thawing with no detectable change in radiation sensitivity (Supplementary Fig. S2B), consistent with published data indicating genomic stability of normal and tumor colonic tissue in organoid culture (4). Together, these data show that while normal human large intestine and adenoma PDOs are radioresistant, transition to colorectal cancer yields as much as 1-log radiosensitivity in a subset of patients. To our knowledge, this is the first report of development of radiosensitivity accompanying adenoma to carcinoma transition in colorectal cancer, an observation that could not currently have been made without availability of organoid technology.

While organoids from treatment-naïve patients display a spectrum of radiosensitivities, PDOs from patients with rectal cancer that had incomplete response to neoadjuvant therapy were uniformly radioresistant. Figure 2C and D show that all 6 patients with rectal cancer who partially responded to neoadjuvant therapy and ultimately required surgical resection displayed radiation resistance with mean D₀ = 21.5 ± 11.4 (Supplementary Fig. S2A), suggesting chemoradiation likely deletes the most radiosensitive cancer population.

While mutations leading to WNT gain-of-function occur during transition from normal mucosa to adenoma, the stage when logarithmic expansion of the stem cell signature-containing population is observed (13), our data reported here show maintenance of the radioresistant phenotype during adenoma formation. The surprising observation that a subset of colorectal cancers develop inherent vulnerability to radiation at the stage of adenoma to adenocarcinoma transition prompted examination of the mutational fingerprint of these PDOs. We sequenced genomic DNA from 15 adenoma or colorectal cancer PDOs (from 13 patients) and paired normal large intestinal PDOs from the same patient, or when normal PDOs were unavailable, from blood DNA from the same patient by performing WES (Supplementary Fig. S3A). To validate our mutation analysis, we first compared mutation load with TCGA cohorts (36). As expected, the tumor mutational burden of our PDOs matched most closely with colon adenocarcinoma among 33 major human cancer types (Supplementary Fig. S3B). The mutational variants include nonsynonymous mutations in coding sequences (i.e., missense, nonsense, insertion, or deletion) and in noncoding sequences (i.e., splice site; Supplementary Fig. S3C). Consistent with a large body of literature (4, 5, 7, 8), we identified key recurrent mutations in APC (93%), TP53 (80%), and KRAS (53%) among those top mutated genes in our colorectal cancer PDOs (Fig. 3A). In addition, we found 1 (RC-MSK-012) of 15 colorectal cancer PDOs was hypermutated, with an overall mutational variant median = 132 (Supplementary Fig. S3D), in line with TCGA data (2). When we sub-grouped those nonhypermutated PDOs by different progression stages, albeit the small sample size, we found no significant difference in total number of somatic mutations between adenomas (n = 3), treatment-naïve cancers (n = 9), and post-neoadjuvant cancers (n = 3; Fig. 3B). Correlation study further showed poor association between tumor mutation burden and D₀ (Spearman r = –0.0679; P = 0.8124), indicating mutation number per se is not associated with radiation sensitivity (Fig. 3C).

Mutations accumulating during cancer development give rise to characteristic patterns in the cancer genome, and each mutational pattern forms a unique signature (37). While the root causes of several mutational signatures remain unclear, other signatures have been associated with exogenous carcinogens or defects in certain DNA repair pathways. In Fig. 3D, we analyzed mutational signatures and find that all colorectal cancer PDOs exhibit signature 1, a signature associated with aging and prevalent in many different cancer types (38). Four signatures (i.e., signature 6, 15, 20, and 26; ref. 37) known to be associated with defective DNA mismatch repair appear in 10 different PDOs that display a broad spectrum of radiation sensitivities (HT-15, HT-9, RC-MSK-022, RC-MSK-002, HM-15, RC-MSK-023, RC-MSK-012, HA-12, RC-MSK-003, HA-11), with D₀ values ranging from 1.7 ± 0.3 Gy to 31.6 ± 4.4 Gy (See Supplementary Fig. S1C, S2A, and S2D). Alternately, four PDOs (HT-9, RC-MSK-022, RC-MSK-002, and RC-MSK-023) display Signature 3, known to associate with defects in HR repair machinery. Interestingly, these PDOs had low or moderate D₀ values (D₀ = 2.6 ± 0.9 Gy, 4.3 ± 1.5 Gy, 4.5 ± 1.6 Gy, and 15.0 ± 2.9 Gy, respectively), potentially indicative of diminished DNA DSB repair capacity.

Radiation sensitivity in mammalian cells is considered to reflect efficiency and fidelity of postradiation DNA DSB repair. Within minutes of irradiation of mammalian cells, recruitment of γH2AX (phosphorylated histone protein H2AX on Ser¹³⁹) forms a discrete nuclear focus at the site of the break, where the downstream DNA damage response (DDR) machinery proteins hierarchically assemble and orchestrate repair of the damage in situ. In this context, acquisition of γH2AX IRIF, visualized by indirect immunofluorescence, within 30 minutes of irradiation quantitatively reflects number of DSBs (39), while kinetics of resolution of γH2AX IRIF provides, by consensus, the most reliable tool to assess DNA repair capabilities of individual cells (40). Using this technology, we recently published that murine colonic Lgr5⁺ stem cells repair DNA damage in vivo more efficiently than small intestinal Lgr5⁺ stem cells, rendering relatively large differences in radiosensitivity of these stem cell populations (41).

Traditionally, in order to obtain the best antibody penetration during immunolabeling, the organoids need to be extracted from Matrigel before fixation, a step that easily takes up to an hour, diminishing the opportunity to faithfully depict the DDR within this time frame. Here, we adapted a methodology that allows us to fix the organoid cultures in situ, maintaining the 3D configuration, enabling 3D whole mount staining at any desired time postradiation without the need of Matrigel extraction (Supplementary Movies S1 and Movie S2; see Materials and Methods for more detail). Consistent with a large body of literature on radioresistance of colonic tissue, and with our PDO clonogenic survival data above, PDOs derived from normal human mucosa resolve IRIF rapidly, resulting in 5-fold reduction in detectable foci at 4 hours postirradiation (Fig. 4A, left; 4B, black line), with almost complete resolution by 6 hours postirradiation. These results correlate radiation dose survival with DNA repair capacity in human PDO culture.

We further evaluated kinetics of γH2AX IRIF resolution in one radioresistant (HT-14) and three radiosensitive (HT-9, HT-15, and RC-MSK-022) treatment-naïve tumor lines, selected from the radiation dose survival experiments above. For these studies, tumor PDOs were irradiated with a dose of 8 Gy, sufficient to cause lethality in sensitive lines, while being sublethal for resistant lines. While γH2AX staining was virtually undetectable prior to irradiation, at 0.5 hour post 8 Gy all PDO lines accrue equivalent numbers of foci (Fig. 4A and B), consistent with DNA DSBs representing biophysical damage to DNA, which thereafter resolve at different rates. Whereas the majority of foci resolve by 6 h in highly radioresistant organoids (Fig. 4B; blue line), γH2AX IRIF resolve much more slowly in highly radiosensitive organoids (Fig. 4B; red lines). In fact, even at 24 hours postirradiation ∼25% of γH2AX IRIF (∼18 of 72 foci) remain unresolved per nucleus in organoids deemed radiosensitive in dose survival clonogenic assays. That these unresolved γH2AX IRIF represent unrepaired DSBs is confirmed by MDC1 focus staining of a subgroup of the same specimens (Supplementary Fig. S4A and S4B), which, by consensus, when correlated with γH2AX IRIF resolution data are together considered to reflect DSB repair (42, 43). These studies indicate that the radiation sensitivity profile of colorectal cancer PDOs, as defined by clonogenic assay, faithfully reflects an inherent DNA repair incapacity rather than an unrelated mechanism.

Ionizing radiation-induced DSBs are repaired by two major complementary DDR pathways: HR and nonhomologous end joining (NHEJ). Following γH2AX signaling, several DDR proteins accumulate at IRIF at DSBs (44). To explore the defective repair pathway involved in colorectal cancer PDO radiosensitivity, we immunostained for two key signaling mediators specific for HR and NHEJ, the BRCA1 and DNA-PKcs proteins, respectively. As shown in Fig. 4C and D, we find normal large intestinal PDO and radioresistant tumor PDOs have almost identical kinetics of BRCA1 focus resolution post 8 Gy, resolving ∼75% of BRCA1 foci per nucleus at 4 hours and >85% of foci per nucleus at 24 hours (Fig. 4C and D; black and blue lines). However, under the same experimental conditions, the three sensitive tumor PDOs clearly had much slower rates of BRCA1 focus resolution (Fig. 4C and D; red lines), leaving ∼40% to 50% of BRCA1 foci per nucleus unresolved at 24 hours post irradiation, indicating defective HR in DSB repair in these sensitive tumor lines. This result is in line with the genomic analysis above that HT-9 and RC-MSK-022 bear mutational signature 3. In contrast, no significant difference in kinetics of DNA-PKcs focus resolution was observed in all normal large intestinal PDO and colorectal cancer PDO lines (Fig. 4E and F). Collectively, these experiments indicate that impairment of the HR machinery, not the NHEJ pathway, is likely a contributor to the radiosensitive colorectal cancer PDO phenotype.

In the current clinical setting, chemoradiation is the standard of care for patients with rectal cancer. We next examined whether the differential radiosensitivity of pre-neoadjuvant PDOs correlates with clinical outcome of patients with rectal cancer receiving nCRT therapy by following patients longitudinally during their course of treatment and obtaining tissue post-nCRT therapy to generate PDOs, if tissue was available. PDO HT-14 collected at time of diagnosis is highly radioresistant (D₀ = 28.2 ± 13.1 Gy; Fig. 5A). At 6 months post chemoradiation, this patient responded poorly to nCRT, presenting with SD evidenced by persistent local tumor at endoscopy and by MRI (Fig. 5B), and returned for surgical tumor resection. The post-nCRT PDO HT-19 from this patient shows, as expected, high radioresistance in our clonogenic assay (D₀ = 32.1 ± 17.1 Gy). A second pre- and post-nCRT PDO pair (RC-MSK-054 and RC-MSK-054TNT) from another patient shows a similar result as the treatment-naïve PDO is moderately radioresistant (D₀ = 9.7 ± 1.6 Gy) and the post-nCRT PDO from the persistent tumor shows high radioresistance (D₀ = 23.1 ± 11.6 Gy; Fig. 5C and D). In the case of HT-21, this treatment-naïve PDO demonstrates moderate radiation sensitivity with D₀ = 6.5 ± 0.6 Gy (Fig. 5E). This patient partially responded to neoadjuvant therapy with tumor size reduced >50% (major response; Fig. 5F). The PDO (HT-22) from this posttreatment tumor remnant exhibits high radioresistance (D₀ = 31.9 ± 12.7 Gy), indicating eradication of sensitive populations within the primary tumor by neoadjuvant treatment. Critically, 2 other patients whose primary tumor PDOs (HT-20 and HT-17) are highly radiosensitive (D₀ = 3.6 ± 0.41 Gy and 2.1 ± 0.0.4 Gy, respectively) achieved cCR with the whole tumor eliminated after chemoradiation (Fig. 5G–J).

We extended our analysis to all rectal cancer patients who received neoadjuvant treatment post primary tumor PDO derivation. As listed in Supplementary Table S2, 13 patients with rectal cancer are available for response evaluation. Among them, 2 patients achieved cCR, 6 patients achieved partial response with tumor reduction >50% (PR, major), 2 patients had partial response with tumor reduction <50% (PR, minor), 2 patients did not respond presenting with SD, while the other patient developed PD after treatment. Importantly, a striking reciprocal correlation is seen between the D₀ of primary tumor PDO and clinical response to neoadjuvant therapy, that is, the smaller the D₀, the better the patient outcome (Fig. 5K). Consistently, the D₀ also inversely correlates with size reduction of primary tumors (Spearman r = –0.7916; P = 0.002; Fig. 5L). It is important to note that while 3 patient PDOs were assayed in a retrospective manner (Fig. 5K and L; pink dot), the other 10 patient PDOs were investigated prospectively without knowing their clinical response at time of the clonogenic survival assay (Fig. 5K and L; black dot). These data indicate our PDO clonogenic radiation assay may have the potential to predict treatment outcome in patients with rectal cancer undergoing nCRT. From these studies, albeit a small sample size, a cut-off point that distinguishes clinical outcome seems to emerge, with D₀ > 7.6 (dashed line in Fig. 5K) distinguishing poor responders (including PD, SD, and PR minor) from good responders (PR major and cCR), with 100% specificity (5/5) and 87.5% sensitivity (7/8).

While ∼50% to 70% of patients withrectal cancer have incomplete response to nCRT (3, 45–47), alternative agents capable of enhancing treatment response are of great interest. To address whether our organoid technology can be used to evaluate potential radiosensitizers, we elected to use the PARP inhibitor olaparib, which is under active investigation in colorectal cancer preclinical and clinical settings (48–50), to manipulate the PDO DDR. First, we surveyed the effect of olaparib alone on colorectal cancer PDOs. Among 11 PDOs tested 4 lines displayed significantly reduced survival (20%–40%) upon 10 μmol/L olaparib treatment (Fig. 6A). Interestingly, 3 of these 4 responding PDOs contain mutational signature 3 indicative of defective HR (HT-9, RC-MSK-022, and RC-MSK-002). While olaparib alone does not cause toxicity to other PDO lines, this reagent also successfully radiosensitizes two PDOs (RC-MSK-023 and RC-MSK-004) from patients that developed local recurrent tumor post nCRT, reducing the D₀ value from 16.3 Gy to 7.4 Gy and from 31.3 Gy to 14.5 Gy, respectively (Fig. 6B and C). Of note, one of these two lines bears mutational signature 3 (RC-MSK-023). This result is in agreement with earlier studies that PARP inhibitors enable radiosensitization in tumors including colorectal cancer (51–53). While further systematic studies are needed, the data shown here provides proof-of-concept that use of colorectal cancer PDOs coupled with SHMT analysis for radiosensitizer screening is possible.

Cancer as a disease entity, with some exceptions, is considered radioresistant (54, 55). Tumor radioresistance has been ascribed to numerous genetic and epigenetic causes including chronic hypoxia, upregulation of antiapoptotic gene programs, predominance of quiescent stem cells, and enhanced scavenging of reactive oxygen species, to list a few. It is presumed, but has never been tested, that progression from normal stroma to adenoma and transition to carcinoma is accompanied by development of radioresistance. However, it has not been possible to accurately define this progression to radioresistance at the level of the tumor cell due to inadequate specimen availability, cumbersome ex vivo tumor model systems such as patient-derived xenografts, complexity of the stroma and systemic influence on tumor cell response. Use of organoid culture has alleviated some of these constraints allowing for the first time examination of inherent radiation sensitivity of the tumor cell in the current study. The most important finding of the current study is that while progression from normal colonic mucosa to benign tumor maintains the inherent radioresistance of the colon, conversion of adenoma to carcinoma yields a spectrum of responses, with many tumors displaying extreme radiosensitivity. This unprecedented finding is somewhat surprising because it is widely accepted that developing cancer cells acquire various advantages to fuel proliferation, halt cell death machinery, and promote diverse cellular fitness attributes to enhance cancer survival. In contradistinction, our study reveals acquisition of properties during adenoma to carcinoma transition that render colorectal cancer cells more vulnerable to radiation-induced DNA damage lethality, a property that is durable over many passages in organoid culture and hence must be considered intrinsic to tumor cells. Whether development of such vulnerability is unique to colorectal cancer progression or is generalizable to other cancers is of interest to pursue in future studies.

Another finding of the current study is the loss of rectal cancer radiosensitivity post nCRT. Whereas tumors comprised primarily of radiosensitive clones as determined by low D₀ values in organoid culture appear to develop cCR after neoadjuvant therapy, tumors with intermediate radiosensitivity before neoadjuvant therapy appear to recur as increasingly radioresistant. We posit this likely reflects eradication of sensitive cancer clones within a tumor, leaving residual tumor cells with innate or acquired resistance ultimately nonresponsive to chemoradiation. This notion is supported by the comparison of the pre- and post-nCRT PDO pairs derived from the same patient that display enhanced posttreatment radioresistance (i.e., RC-MSK054 and RC-MSK054TNT; HT-21 and HT-22). In clinic, it is in fact not uncommon that the tumor remnant post nCRT consists of radioresistant adenoma mainly, with the adenocarcinoma or carcinoma population being mostly eliminated (e.g., the case of patient HA-7; see Supplementary Table S1 for details), consistent with this notion. In Supplementary Fig. S5 we place these new discoveries, and our recent findings on the protective role of the stem cell niche in early onset of colorectal cancer tumorigenesis (13, 56), in the context of the well-known “Vogelgram” paradigm of multistage colorectal cancer progression. In the healthy normal colonic crypt, the intact stem cell niche forms a physical barrier to constrain a sporadic mutant stem cell from errant propagation. Niche injury caused by aging or other insults results in separation of Lgr5⁺ colonic epithelial stem cells from their mesenchymal Wnt source (basal lamina), which in turn provides a fitness advantage for mutated stem cell clones harboring WNT pathway mutations (e.g., APC, CTNNB1 that encodes β-catenin) to outgrow and form early adenomas that consist of a high percentage of Lgr5⁺ stem cell–like cells, a transition that retains radioresistance. Eventually these benign lesions evolve by accumulating additional key driver mutations (e.g., KRAS, TP53) and progress to malignant adenocarcinoma or carcinoma, a transition concomitant with development of inherent radiation sensitivity in cancer cells. Diagnosed rectal tumors are then often challenged with nCRT, eliminating the sensitive subpopulation, which if dominant may lead to cCR and potentially cure, however if sufficient numbers of resistant clones and/or stem cells survive, cCR may not be possible, rendering surgery as the only option for tumor cure.

Recent large-scale genomic studies suggest prevalent alterations of DDR genes across various human cancers (57). It remains largely unclear if and when such DDR alterations occur in colorectal cancer, and how such DDR maladaptions alone, or together with other non-DDR genes, might affect colorectal cancer response to radiotherapy. In essence, colorectal cancer bears on average ∼130 somatic mutations, and such a high number discourages us from molecularly dissecting which mutation(s) plays a determinant role(s) in radiosensitivity, not to mention the probability that different mutations contribute in a multifactorial manner. Cancer mutational signatures, on the other hand, illustrate total changes in genomic landscape and have been implicated in many aspects of modern oncology (58). Here, we demonstrate for the first time that mutational signatures known to be linked with MMR deficiency (i.e., signature 6, 15, 20, 26) appear in all classes of tumors including benign adenoma, treatment-naïve, and post-neoadjuvant cancers, regardless of radiosensitivity. In contrast, mutational signature 3 that has been linked to HR deficiency (also supported by our foci staining data) tends to be associated with strong or moderate radiosensitivity in colorectal cancer. Extensive work documents synthetic lethality by using PARP inhibitors against breast and ovarian tumors that harbor defects in HR repair pathway, a manifestation of “BRCAness” (59). Consistent with this notion, our data shows three of four signature 3–containing PDOs respond to the PARP inhibitor, olaparib, while another signature 3–containing post nCR PDO can be radiosensitized by the same agent, confirming the predictive implication of the BRCAness mutational signature in PARPi sensitivity (60). Although validity of this conclusion is limited by the small number of PDOs with each signature in the current study, we posit that mutational signature 3 will be of importance as a prognostic prediction marker to identify a subset of patients who might benefit from radiation and combination treatment.

Lastly, result from our dose-dependent PDO clonogenic radiation assay correlates with outcome of patients with rectal cancer receiving nCRT consisting of different regimens both in retrospective and prospective settings, albeit the relatively small sample size in our study, in agreement with other retrospective data that indicate in vitro response of single treatment may be associated with response to combination therapy (7, 8). Strikingly, in our cohort, two of the most radiosensitive PDOs pre-nCRT represent the two “Watch & Wait” patients who had a cCR after neoadjuvant treatment. In contrast, those most radioresistant PDOs pre-nCR represent patients responding poorly to treatment, indicating a potential for our assay to select “excellent” as well as the “worst” responders from the patient pool. A cut-off point of D₀ (>7.6) seems to distinguish the worst responders from good responders in the patient pool. In the future with appropriate testing and validation measures met, preclinical stratification could prove useful for more individualized clinical management.

Overall, our current study establishes a framework that PDOs can be used to investigate the inherent property of cancer radiosensitivity and translates the cutting-edge radiobiologic SHMT algorithm into a standardized in vitro assay system with a simple readout (D₀) that in a small cohort correlates well with patient outcome, and with advanced testing may prove predictive of outcome. While powerful, organoid culture lacks microenvironmental regulation of tumor response, which we acknowledge may be a limitation for using PDOs in some settings. Nevertheless, our PDO assay system may serve as an experimental platform for testing/optimizing radiation modulators and combination therapies both in preclinical pilot studies and during future phase I–II trials of novel systemic (i.e., immunotherapy for MSI rectal CA) or radiosensitizer (i.e., PARP, ATR, or NHEJ inhibitors to impede DNA repair) therapies targeting biologic vulnerabilities of resistant tumors in high-risk patients.

V. Makarov reports a patent for EP3090066A2 issued. J. Smith served as a clinical advisor for Guardant Health Inc. (2019) and received travel support from Intuitive Surgical for Fellow Education (2015). Z. Fuks reports patents unrelated to this work (US 10413533B2, US20170333413A1, and US20180015183A1) and is a cofounder of Ceramedix Holding LLC. N. Riaz reports grants from Repare Therapuetics, Repertoire Immune Medicines, and grants from Pfizer outside the submitted work. T.A. Chan is a cofounder of Gritstone bio and holds equity; holds equity in An2H; acknowledges grant funding from Bristol-Myers Squibb, AstraZeneca, Illumina, Pfizer, An2H, and Eisai; has served as an advisor for Bristol-Myers Squibb, MedImmune, Illumina, Eisai, AstraZeneca, and An2H; and is an inventor on intellectual property held by MSKCC on using tumor mutation burden to predict immunotherapy response, with pending patent, which has been licensed to PGDx. M. Nishimura reports personal fees from Boston Scientific, Olympus America, and personal fees from Lumendi outside the submitted work. R. Kolesnick reports patents unrelated to this work: US7195775B1, US7850984B2, US10052387B2, US8562993B2, US9592238B2, US20150216971A1, and US20170335014A1, US20170333413A1, US20180015183A1, US10414533B2, US10450385B2; and is one of the cofounders of Ceramedix Holding LLC. No disclosures were reported by the other authors.

K.-S. Hsu: Conceptualization, resources, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. M. Adileh: Conceptualization, resources, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft. M.L. Martin: Formal analysis, investigation. V. Makarov: Software, formal analysis. J. Chen: Software, formal analysis. C. Wu: Resources. S. Bodo: Methodology. S. Klingler: Methodology. C.-E.G. Sauvé: Data curation. B.C. Szeglin: Data curation. J.J. Smith: Resources, data curation. Z. Fuks: Supervision, funding acquisition, validation, writing–original draft, writing–review and editing. N. Riaz: Data curation, formal analysis. T.A. Chan: Supervision, writing–original draft. M. Nishimura: Resources. P.B. Paty: Conceptualization, resources, data curation, supervision, funding acquisition, project administration. R. Kolesnick: Conceptualization, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

This work was supported by the generous donations of Corinne Berezuk, Michael Stieber, and Patrick A. Gerschel (to P.B. Paty), and the NCI grant R21-CA245636 (to R. Kolesnick). J.J. Smith and C. Wu are supported by R37CA248289–02.

The authors sincerely thank all the patients and their families for participation in this study. They are grateful for reagents provided by colleagues: L-Wnt-3A HEK293 cell line was a kind gift from Robert Vries (Hubrecht Institute), the R-spondrin-1 HEK293 cell line (293T-HA-Rspo1-Fc) from Calvin J. Kuo (Stanford), and the Noggin HEK293 cell line and Lgr5-LacZ mice from Hans Clevers (Hubrecht Institute). The authors acknowledge use of the Integrated Genomics Operation Core Facility in Memorial Sloan Kettering Institute. They thank all authors for critical comments on this manuscript.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.