Background
Over the past decades, the management of patients with metastatic colorectal cancer (mCRC) has only slightly improved since the approval of chemotherapy-based regimens combining 5-fluorouracil, oxaliplatin, irinotecan, epidermal growth factor receptor inhibitors, and antiangiogenic agents [
1]. Most patients benefits for these therapies but disease control remains timely limited due to the emergence of drug resistance leaving the patients with few therapeutic options. Moreover, the difference among patients’ response to approved-standard CRC chemotherapies exemplifies inter-patient tumor heterogeneity. This hallmark of cancer requires adapting the therapeutic strategy for each specific patient.
Molecular Personalized Medicine (MPM) approaches try to overcome inter-patient’s tumor heterogeneity by detecting tumor-specific characteristic using protein-, RNA- and genome-based approaches [
2]. However, pan-solid tumor clinical trials have shown that MPM benefits only to a small subset (10–15%) of the patients [
3‐
6]. MPM is even more limited at predicting the response to chemotherapies whose effectiveness does not correlate to a single molecular alteration. For these reasons, it is crucial to develop alternative strategies to predict treatment response for each patient. In this regards, Functional Precision Medicine (FPM) is emerging as a promising technology to fulfil this gap. FPM is based upon an
ex-vivo drug-test in which living tumor cells from a specific patient are exposed to a panel of anti-cancer drugs. This assay aims at identifying the drug sensitivity and resistance profile of each individual tumor to orient patient treatment efficiently [
7]. Recently, Malani et al
. and Kornauth et al
. showed excellent results in FPM-based strategies for patient with hematological malignancies [
8,
9]. These studies pave the way toward the implementation of FPM in these pathologies which provide a fast and easy access to large quantities of tumor cells. Nevertheless, applying FPM to patients with solid tumors has been challenged by the lack of adequate technology to amplify tumor cells ex vivo rapidly and faithfully. Cell lines are poor cancer surrogates, and patient-derived xenograft models are too costly and time-consuming for therapeutic strategies [
10]. Recent studies suggest that patient derived organoids (PDO) could hold the promise of FPM for solid cancers. PDO are tridimensional, multicellular structures, expanded in vitro, which retain morphological, histological and genetic properties of their tumor of origin [
11]. Furthermore, they are stable over time, self-organized and self-renewing structures. PDO collections have been established for a large variety of tumors, mainly carcinomas, including breast [
12], colorectal [
13,
14], pancreatic [
15,
16] and ovarian [
17,
18] cancers. PDOs are currently emerging as powerful pre-clinical models and are suitable for drug screening strategies [
11]. In a pooled analysis, the specificity and sensitivity of PDO-based predictive scores exceeded 70% [
19]. However, these studies included a very small number of patients and were performed in very favourable settings; including large tumor specimens, no time constraints, single drug tests and no predefined cut-off of sensitivity scores, raising concerns regarding the feasibility to implement PDO technology in routine practice [
19]. Moreover, while these studies suggest the clinical validity of PDOs to identify drug sensitivities, they did not assess clinical significance, i.e. the benefit for patients.
To implement PDO-based drug tests in the clinical path of patients with solid tumors requires solving three challenges: 1) establishing PDOs from a limited amount of tumor material obtained by needle biopsy; 2) testing a large panel of anti-cancer drugs (or drug combinations) to increase the probability to identify efficient therapies; and 3) providing the results to clinicians in a timely manner to avoid any interruption in patient therapeutic management. Here, we report a feasibility study of PDO-based FPM in patients with CRC. This includes robust methodology for generating PDOs from core needle biopsies, a test based on a 25-drug panel (named chemogram) and a scoring strategy whose execution is compatible with the constraints of clinical practice.
Experimental procedures
Human primary specimens
The human study protocols followed all relevant ethical regulations in accordance with the Declaration of Helsinki principles. Fresh tumor tissues were obtained after surgery or core needle biopsy within two institutional review board-approved, ethics committee-approved precision medicine studies (STING, NCT0493252; MATCH-R, NCT02517892). All patients signed informed consent.
Tumor digestion
Tumors were minced into pieces and carefully homogenized mechanically. The tumor mix was split into 3 samples of 150 mg carefully weighted with a high precision scale. The first sample was processed according to manufacturer protocol (Tumor Dissociation Kit (human), GentleMACS Disssociator, Milteny Biotec). The second and third samples were incubated in basal medium for 1 h while shaking with Collagenase A (type IV from Clostridium histolyticum, Sigma-Aldrich) for sample 2 or with 50 µg/ml liberase TH (Roche) for sample 3. After digestion, the three samples were processed the same way to avoid any technical bias. Briefly, digestion was stopped with fetal bovine serum (FBS) and cells washed and spun three times. Viability assay was performed in triplicate with 20 μl of the cell solution according to CellTiter-Glo manufacturer protocol (Promega). Number of viable cells was assessed using Kova Glasstic Slide (Fischer Scientific) after trypan blue staining. Two different tumor samples originating from two different patients were used for the test. Three counts were performed by two different researchers. Viability and number of viable cells were normalized to 1 mg of tumor.
Tumor cell isolation and organoid culture
Tumors were minced into pieces and incubated in basal medium with 50 µg/ml liberase TH (Roche) and 10 µM Y-27632 (Selleckchem) for 1 h while shaking. After incubation, mechanical forces (pipetting) were applied to improve the digestion process. FBS (10%) medium was added, and the mixture filtered through a 100 µm cell strainer. Cells were spun at 350 g for 5 min and the pellet resuspended in red blood cell lysis solution (Milteny Biotec) according to manufacturer procedure. The cell solution was spun at 350 g for 5 min (twice) and the pellet resuspended in basement membrane extract (BME, Matrigel, Corning) and plated. Once embedded in BME, cells were incubated at 37 °C in culture media modified from Fujii et al. [
20] and supplemented by Intesticult™ Organoid Growth Medium (Stemcell™ Technologies; 06010), renewed 3 times a week. PDOs were passaged every 7 to 14 days. PDOs were incubated for 5 to 20 min at 37 °C in TrypLE 1X (Thermo Fischer Scientific) and dissociated into single cells and small clusters (< 10 cells) by applying mechanical force (pipetting) every 5 min. After incubation, 10% FBS medium was added, and the cells were spun at 350 g for 5 min (twice). The pellet was resuspended in BME at appropriate ratio (500 cells/μl of BME) and plated. After BME polymerization, PDO culture media containing 10 µM Y-27632 was added and the culture plates incubated at 37 °C.
The 25 PDOs were cryopreserved in FBS containing 10% DMSO (Sigma- Aldrich) and all of them have been tested successfully for culture after thawing. PDOs nomenclature: CGR for Colon Gustave Roussy. The number assigned to each PDO such as 0001, 0002, etc. corresponds to the order in which the surgical or biopsy specimens were processed in the laboratory.
Histology procedures
PDOs embedded in BME were incubated for 1 h with 4% paraformaldehyde (PFA) at room temperature. After the incubation, the mixture was spun 5 min at 100 g. The PDO pellet was resuspended in PBS and spun 5 min at 100 g. After one more washing step, the PDO pellet was dehydrated with ethanol and embedded in paraffin. Sections were subjected to H&E staining. For detection of CDX2, paraffin sections were processed in a Ventana Benchmark Ultra automated immunostainer instrument for heat-induced antigen retrieval (CC1 Buffer equivalent pH8) for 64 min at 95 °C. Sections were incubated with rabbit monoclonal anti-CDX2 (Roche; #760–4380, clone EPR2764Y, pre-diluted) for 32 min at 36 °C. The signal was revealed with UltraView universal DAB detection kit (Roche #760–500). Finally, the sections were counterstained with hematoxylin and bluing reagent. For detection of CK20, paraffin sections were processed in a Bond Leica automated immunostainer instrument for heat-induced antigen retrieval (ER2 corresponding EDTA buffer pH9). Slides were incubated with CK20 antibody (rabbit monoclonal: LS Bio # LS-C210303, clone SP33, 1:25) for 60 min at room temperature and detected by Bond Polymer Refine Detection kit. The signal was revealed with DAB and slides were counterstained with hematoxylin.
Whole-Exome sequencing and molecular alteration analysis
Read sequences were evaluated for their quality using FastQC v0.11.9. FastqScreen v0.15.1 was also used to assess for any DNA contamination by other species. Sequences were trimmed for their lower quality (BaseQ < 20) and Illumina adapter sequences using Fastp v0.23.2 [
21]. All QC results were compiled to a user-friendly report using multiqc v1.14. Mapping was performed against the human hg19 genome sequence using bwa mem v0.7.17 [
22]. Duplicate reads marking and base quality recalibration were performed using GATK v3.8–1-0 [
23]. Germline variant/indel calling was performed with Varscan mpileup2cns v2.4.3 [
24], using the default parameters. Called variants were then filtered in using bcftools v1.9 according to the following criteria: 1) AD > = 10; 2) Freq > 5; 3) Func_refGene = = ‘exonic,splicing’; 4) gnomAD_exome < 1e-03; 5) ExonicFunc_refGene ! = ‘synonymous_SNV’; 6) ExonicFunc_refGene ! = ‘unknown’. Filtered variants quality was assessed using bcftools stats.
The original tumor molecular alterations were extracted from the medical record of the patients and detailed in Fig.
2D and Supplementary Table
2.
Copy number alteration analysis
Identification of copy-number anomalies were performed using EaCoN v0.3.6 [
25] under R v3.6.2, for all the steps described hereafter. All organoid profiles were compared to a sex-matched in-house reference patient profile (MRA1012_N for females, MRA1144_N for males). GATK base-recalibrated BAM files were internally transformed to the mpileup format using Rsamtools v2.8.0 [
26] ignoring replicates and secondary alignments. To generate the log2ratio data, the test and reference mpileup profiles were binned to windows of 50 nt in median (depending on the capture BED information), and bins with a total depth < 20 were discarded. Using a pre-generated track of GC% content in bins, those with a value < 20% or > 80% were flagged as outliers. The log2ratio (L2R) of test / reference depths was computed for each bin, and linearly imputed for GC% outliers. The L2R was then normalized for GC% using a lowess regression. To generate the BAF data, any non-reference sequences in the mpileups were identified and their depth quantified. SNP variants supported by less than 3 reads and/or for which the total depth was below 20 were discarded. To filter for noisy, low frequency variants, all SNP variants in the test sample with a reference frequency below 33% were discarded. The bivariate (L2R and BAF) data were then segmented, evaluated for their allele-specific and absolute copy-number, as well as ploidy and tumor cellularity, using ASCAT v2.5.2 [
27]. CGR0009 WES data could not be exploited for the CNA analysis.
Drug tests (chemograms)
All the drugs tested were purchased at Clinisciences, except oxaliplatin and carboplatin, which were provided by Gustave Roussy hospital pharmacy. The stock concentration was 10 mM. The solvent was DMSO apart from oxaliplatin, carboplatin and trifluridine-tipiracil for which PBS-tween20 (0.3%) was used. 96-wells drug source plates were prepared with a D-300e digital dispenser (Tecan). The drug concentration in the source plates was 10 times higher than the desired final concentration and the drugs were dissolved in 100 μl of PDO culture media. PDOs tested were incubated for 20 min at 37 °C in TrypLE 1X and dissociated to single cells by applying mechanical force (pipetting) every 5 min. Cells were counted using Kova Glasstic Slide, embedded in BME in a 250 cells /μl ratio and plated (3 μl per well) in the 60 center wells of a 96 wells plates with a pipetting robot (Assist Plus, Integra). The BME domes were overlaid with 125 μl of PDO culture media with a pipetting robot (Viaflo, Integra). Two days post seeding, media was removed and replaced by 112.5 μl of fresh PDO culture media. 12.5 μl of the drug source plate was added using a pipetting robot (Viaflo, Integra). Media and drugs are renewed at day 6. At day 8, media was removed and 50 μl of Cell Titer Glo 3.0 (Promega) diluted by 2 in basal culture media was added to each well. Culture plates were agitated for 5 min on an orbital shaker and luminescence was recorded after 20 min of incubation at room temperature.
The intracellular LDH level was measured using the LDH-Glo assay (Promega) according to manufacturer protocol after PDO lysis with 100 µl of PBS/Triton 1% per well. Lysate was diluted by 600 in PBS. 50 µl of the diluted lysate were incubated with 50 µl of LDH-Glo reagent for 1 h.
Readings of bioluminescence were obtained using Biotek Synergy H1 plate reader. Each condition was tested in triplicate (3 wells). Control wells were containing solvent but no drug.
PDO imaging assay
The microcavity 96 wells Eplasia plate system (Corning) was used according to manufacturer protocol after PDO dissociation into single cells using TrypLe 1X. Drug were added 2 days after cell loading using a D-300e digital dispenser (Tecan). PDOs were imaged with a LionHeart FX (BioTeK Agilent) and segmented with Gen5 software.
Chemogram scoring system
For each condition, the average value and the standard error of the three wells (ATP-bioluminescence signal) were calculated. The average value of each condition was normalized to the average value of the control wells providing the percentage of relative viability. When the standard error was over 12% the data was excluded from the analysis or the test redone. The Area Under the Curve (AUC) for each of the 25 PDO and each of the 25 drugs was calculated using excel analysis. For each drug, the AUC of the average response of the 25 PDOs was divided by the AUC of the PDO of interest. The result was named AUC score. The direct response of the PDO to each drug to the 3 concentrations tested was also calculated and called the sensitivity score. This score was defined by calculating the ratio between the area over the curve and the total area. The AUC score was summed to the sensitivity score to give rise to the final score used in this study.
Discussion
We presented an assay to establish PDOs from core needle biopsies and generate high-quality drug sensitivity data to identify treatment options for patients with advanced CRC (Fig.
7C). We demonstrate that the procedure for PDO generation and amplification fits with deep and fast pharmacological profiling compatible with patient management. As a result, PDOs could be used for precision medicine and increase the chances of identifying effective treatment options.
To the best of our knowledge, two FPM clinical trials on solid tumors have been completed and published: SENSOR [
32] and Tumorspheres Colrec [
33]. The SENSOR trial was negative with only six patients treated without clinical benefit. The phase 2 Tumorspheres Colrec trial was positive. Nonetheless, among the 34 patients treated based on the results of ex vivo drug tests, only half showed progression-free survival at two months. These two studies highlight that functional precision medicine strategies have still to be improved to benefit cancer patients. Together with previous reports and our own observations, this points to four essential criteria toward clinical implementation [
34].
First, the generation of tumor avatars should be delivered for the largest number of patients, regardless of the nature of their tumor (histology, molecular profile). In this regard, the 25 PDO cohort we generated from CRC patients displays molecular heterogeneity. The oncogenic mutations were 94% identical to the ones portrayed in the patients, and consistent with the Cosmic database [
29]. The success rate at generating PDO is commonly high when large tumor samples are used as starting material (surgeries or effusions). For instance, Narasimhan et al. reported a PDO take-on rate of 68% in the APOLLO study [
35]. Althougt, to be deployed at large and impact patient care, PDO establishment must predominantly start from core needle biopsies. This procedure was used in the SENSOR and Tumorspheres Colrec studies with a PDO take-on rate of respectively 57% and 53.6%. Here, starting with core needle biopsies for 13 patients in the constraint of clinical setting, we report a PDO take-on rate of 61.5%. In addition to the quantity of material, the quality is crucial for PDO derivation success. We found strong heterogeneity in the percentage of tumor cells at the inter- and intra-patient levels that can be explained by random sampling or necrosis induced by previous line(s) of therapy. Correlations between cellularity and take-on rate were also drawn from TUMOROID [
36] and SENSOR studies. Mastering the digestion procedure, culture protocols and reducing time from biopsy to the laboratory are key factors to improve the success rate of PDO establishment. Reaching an acceptable take-on rate for clinical implementation can be achieved by multiplying the number of samples retrieved during the biopsy procedure.
As a second criterion, the turnaround time to deliver the results to the clinician is key to inform clinical decisions in a timely manner. This parameter depends on multiple variables including PDO growth rate and the number of drugs tested. This study demonstrates the feasibility of testing as many as 25 drugs in less than 10 weeks (median, 6 weeks). These results are in line with those of SENSOR and Tumorspheres Colrec studies with median times to generate the drug sensitivity data of 10 weeks and 4.8 weeks, respectively. To avoid leaving patients without treatment thoughout the duration of the chemogram, the biopsies could be performed at the onset of the last line of SOC therapy, around 8–16 weeks before receiving the chemogram-guided treatment. In support to this strategy, the SENSOR study used biopsies sampled before and after SOC therapy to show that SOC exposure did not alter the PDO response to the experimental drug tested. Nevertheless, FPM would benefit from shortening assay turnaround time to provide guided treatment as fast as possible to the patients. In this regard, microfluidic technologies, by downscaling the number of cells per condition, are currently emerging as great options to reduce turnaround time [
37,
38].
Third, the largest the panel of drugs tested is, the more likely it will identify a therapeutic option (‘hit’) for the patient. Of course, testing a comprehensive number of molecules requires a substantial amount of PDOs and must be balanced with turnaround time and PDO amplification rate. While 8 and 9 drugs or drug combinations (on average) were tested in the SENSOR and Tumorspheres Colrec studies, respectively, our panel included 25 drugs. If we report the number of drugs tested to the number of weeks necessary to obtain the results, our drug screening throughput (25 drugs in 6 weeks) is 6.8 and 2.1 times higher than those in SENSOR and Tumorspheres Colrec studies, respectively. Importantly, guided by the idea of making FPM available for a maximum of patients, we included chemotherapies and targeted therapies that could be afforded by hospitals in academic clinical trials. Also, most of the molecules included in our panel have been tested in metastatic CRC in the frame of clinical trials [
39‐
43]. Despite being negative at the population scale, a large majority of these trials showed a low but significant percentage of clinical response at individual patient scale. This is concordant with our results showing that 84% of hits do not belong to SOC in this indication. This suggests it is relevant to test non-CRC drugs as additional therapeutic options for refractory CRC patients and positions FPM as a strategy to identify therapies that may not have proven benefit at the population scale but still represent relevant therapeutic opportunities at the single patient level.
Setting a robust method to identify patient tumor vulnerabilities based on ex vivo drug tests is an essential fourth criterion. To date, there are two main strategies to identify drug sensitivity ex vivo. On one hand, score and hit identification systems can be built up by comparing FPM results for an individual drug with clinical outcome after treatment with that drug [
10,
34]. The scoring system developed can then be used prospectively to orient patient treatment based on FPM results. This approach is suitable for standard of care drugs. For example, in CRC, it is possible to compare the clinical outcome and effect on PDOs of 5-FU, oxaliplatin and irinotecan. However, when the drugs to be tested are not prescribed to the patients in routine, the "PDO-patient" comparison is impossible. On the other hand, to fulfill the fundamental principle of FPM to test a wide range of drugs, we thought, in line with other laboratories [
32,
33], to develop an alternative scoring system. This is based on building up a pilot PDO cohort to calculate average responses for each drug at the population scale and being able to to prospectively identify remarkable sensitivities for specific PDOs. Identifying outlier responses defined as “hits” aims at identifying the best therapeutic option for a specific patient across a large panel of anti-cancer drugs that may not be given in their indication.
To do so, we first chose the drug concentration tested to be consistent with the endogenous level that may irrigate the tumor. Second, we adapted the concentrations tested to capture the average IC50 of each drug. Thus, it became possible to observe important drug sensitivity differences between the PDOs. In this way, PDOs can be ranked from best to worse responder for a single drug. Also, our scoring system allows for one PDO to rank the 25 drugs for their in vitro efficiency. In addition to high technical and biological reproducibility (R > 0.92), we also observed a strong correlation between drugs of the same family or sharing the same mechanism of action, validating the chosen concentrations and the scoring system. We believe FPM will benefit from testing related drugs to strengthen the choice of the PDO guided treatment. Interestingly, we observed an inverse correlation between the numbers of hits found and the number of lines of treatment received by the patient before the biopsy. This result is in line with clinical data showing that overall tumor drugs sensitivities decrease along the course of the disease [
44]. With time, the pilot PDO cohort may be enriched with new PDOs and drug tests to report the heterogeneity of the disease and refine hits identifications.
In this study, we could not orientate patient treatment based on the chemogram. Even so, we were able to monitor eight of the patients who received SOC treatment after we performed the chemogram on their PDOs. Various studies have investigated the predictive value of using PDOs to select optimal CRC treatment at different stages of disease in patients undergoing systemic chemotherapy [
19]. The use of PDOs correctly predicted drug response with irinotecan-based chemotherapy in patients with mCRC in the TUMOROID study [
36] and in patients with locally advanced rectal cancer in the ClinCare study [
30]. However, for oxaliplatin-based therapy, results are more mixed: Ganesh et al
. found satisfactory results for rectal cancer patients [
45] while in the TUMOROID [
36] and in APPOLLO [
35] studies, drug screen results for mCRC patients were not satisfactory. In this study, among the eight patients we could follow, six of them showed clinical concordance with the chemogram. Although the number of evaluable patients is relatively small, we did not observe any bias for oxaliplatin-based regimens in our cohort.
The full implementation of ex vivo drug tests for functional precision medicine strategies will require to tackle remaining challenges. These include testing combination therapies, assessing the contribution of intra-tumoral heterogeneity and incorporating the tumor microenvironment to test for immune-oncology therapies. Yet, along with other studies, our work proved that the PDO technology is mature to be tested in clinic to identify unexpected therapeutic options for refractory patients. We also identified clear parameters to improve the standardization and throughput of the ex vivo drug test that will likely improve the implementation of FPM. As a follow-up study, we opened the prospective multicenter ORGANOTREAT clinical trial (NCT05267912). The first segment, ORGANOTREAT-01 is a phase I/II study evaluating the feasibility and efficacy of PDO-based precision medicine in patients with refractory metastatic CRC. Together with other trials, this study will contribute to determine the patients who will benefit from functional precision oncology.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.