3D organoid growth of PDX origin and creation of a PDXO biobank. We set out to build a PDXO biobank by converting our existing PDX library 3, using methodology previously described to create organoids from adult stem cells16, 29, 32, 33. The basic workflow for establishing the PDXO biobank is depicted in Fig. 1a, where PDX tumors were freshly harvested from mice, or from cryopreserved PDX tumor fragments, minced, cells isolated by enzymatic digestion and mechanical disruption and then seeded into Matrigel domes to create an organoid culture which is expanded and banked once growth is established33. We created a large panel of ~ 200 PDXOs from 16 different types of carcinoma with varying take-rate, including non-small cell lung carcinoma (NSCLC), colorectal carcinoma (CRC), gastric carcinoma (GA), pancreatic carcinoma (PA), breast carcinoma (BC), ovarian carcinoma (OV), liver carcinoma (LI), etc., (Supplementary Table 1). The established organoids generally had varying growth kinetics, with culturing time ranging from 7–14 days (or longer) before requiring passaging, similar to those described for PDOs by others. In addition, PDXOs could be cryopreserved with nearly 100% recovery rate, which was critical for establishing a living biobank as well as the PDXO identity authenticated by single-nucleotide polymorphism (SNP) as previously reported34. Using bright-field microscopy, the morphology of PDXOs look similar to those typically observed for PDOs21, 32, 35, inclusive of cystic, compact33, grape-like and budding morphologies (representative morphology phenotypes are shown in Supplementary Fig. 1a, with selected models shown in Fig. 1b), supporting the hypothesis that PDXO structure and morphology is generally similar to those seen for PDOs. Following establishment of organoid cultures and the associated biobank we next set out to characterize the organoids in comparison to their originating PDX and to curate the data in an online organoid database (https://www.crownbio.com/ob-registration) similar to our PDX online database (https://www.crownbio.com/oncology/oncology-databases/hubase).
PDXOs share similar histo- and molecular pathology as their parental PDXs. PDXs and PDOs independently have been reported to share similar histo-/molecular pathology to the original patient tumors1, 3, 17, 21, 22, 26, 36. Here we show that PDXOs share similar histo-/molecular pathology as the parental PDX. First, PDXO cultures were processed for FFPE as described previously33, following which histopathology analysis of H&E stained organoid sections was compared to H&E staining of parental PDX tissue. This histopathology analysis indicated that each PDXO preserved the histopathology of the parental PDX (Fig. 1b and supplement Fig. 1b) across the cancer types. The theory for creating PDXOs is that PDXs are CSC-derived tumors, similar to the original human carcinomas, and that CSCs of PDXs could be cultured in vitro using similar conditions as for PDOs. The presence of CSCs were indeed confirmed in PDX and PDXO CRC model pairs for the expression of common stem cell markers (e.g. CD44, CD133, SOX2, and pan-CK, a basal cell marker37 (Supplementary Fig. 1c).
Next, we set out to profile PDXOs for their genomics, including transcriptome sequencing (RNAseq), as already reported for the original PDX models3, and whole exome sequencing (WES)33 in order to generate a well-annotated PDXO collection. Genomic DNA from PDXOs and corresponding PDXs were analyzed by WES for changes at the DNA level, including single nucleotide variants (SNVs). Across a panel of 59 matching pairs of PDXO/PDX data sets from 9 different cancer types, high mutational concordance (median mutation correlation 98.7%) was observed, demonstrating genetically near-identical between the pairs (summarized by Fig. 1c, individual representatives shown in Supplementary Table 2). Similarly, the gene expression of the transcriptome for 83 corresponding models across 11 cancer types also demonstrated high correlation as shown in Fig. 1d (median R value 0.951) and Supplementary Table 2. These observations confirm that PDXOs largely preserve the genomic and transcriptomic features of the originating tumors.
PDXO 384-well format in vitro cytotoxicity assay. An important characteristic of cell lines is the robust reliability in HTS, therefore we next examined the reliability of our PDXO models in drug response assay in a 384-well format using CellTiter-Glo® (CTG) as an endpoint readout of cell viability6, 33. Our results demonstrated that the quality of PDXO screening was high, with a median signal to noise ratio of 577 across 129 PDXO models (Fig. 2a). Furthermore, the z’ factor for the PDXO screen, reflective of intraplate precision38, reached relative high values with a median of 0.63 (Fig. 2b). Both parameters are significantly above the acceptable robust HTS standards. In addition, we have also tested interplate variability by comparing the IC50 values generated for a range of compounds across several models over 2 passages (Fig. 2c). The correlation of IC50 values across experiments for each agent was analyzed by linear regression and generated a R2 value of 0.94 (p < 0.0001, Pearson’s correlation), similar to that of cell lines (R2 value 0.96, p < 0.0001 Supplementary Fig. 2a).
We have also established 274 PDXs as ex vivo cultures (Supplementary Table 1) for in vitro assays. However, the ex vivo cultures could not be maintained in continuous culture nor cryopreserved and resuscitated, therefore hindering the generation of a living biobank or HTS. PDX material could only be cryopreserved and successfully reanimated into cytotoxic assay. However, the R2 value of 0.63 was observed when the correlation of IC50 values across experiments was assessed (Supplementary Fig. 2b), which was significantly lower than that observed for PDXOs (Fig. 2c) and cell lines (Supplementary Fig. 2c), thus demonstrating high interplate variation. Taken together, the PDXO 3D screen demonstrated to be a robust assay similar to other in vitro assays, but with greater reliability and reproducibility over 3D ex vivo PDX assays.
Rapid identification of drug-sensitivity and optimal combination strategies. With the establishment of robust HTS workflow, we could rapidly screen large panels of PDXOs. We tested commonly used standard of care (SOC) chemotherapies on a large panel of PDXOs from different cancer types (Fig. 2d). The majority of models showed an insensitive phenotype to cisplatin and 5-FU, whereas a large proportion of models were relatively more sensitive to paclitaxel and gemcitabine, which could be used to select models for in vivo testing or combination evaluation.
Combination treatment has become commonplace in cancer therapy to maximize therapeutic effects and/or widen the therapeutic window of available drugs, particularly those with complementing mechanisms of action (MOAs). Testing combinations in vivo is challenging since a matrix design, for considering different combinations including dose levels of each compound, would not be feasible. In vitro assays have intrinsic advantages in characterizing drug-drug pharmacological coordination, such as synergistic, additive or antagonistic effects, as compared to the in vivo setting as multiple numbers of drugs and potential combinations can be tested if HTS is used, including a matrix of different drug concentrations for each drug and without the complexity of pharmacokinetic/drug metabolism. To this end, we have used a 384-well HTS format to assess the synergistic effects of different drugs in PDXOs.
Synergies between MK-1775 (WEE1 inhibitor) and MK-8776 (CHK inhibitor), and each with gemcitabine have been observed in tumor cells, including melanoma, colon cancer and p53 mutant pancreatic cancers 39–43. To verify these synergies in PDXOs, we selected p53 mutated pancreatic PDXOs, PA5389 and PA1252 with a pR282W and pR273H p53 mutation, respectively, and PA2847 with a nonsense mutation (pW91Ter) and examined the response to gemcitabine, MK1775 and MK8876 (Fig. 2e). All models were sensitive to gemcitabine and MK1775 with IC50 values < 0.2 µM and < 1 µM, respectively, whereas MK8776 generated IC50 values of > 10 mM. Following this, each PDXO was treated with a 6-point titration of each inhibitor and combined in a 6 × 6 matrix format for 5 days and cell viability measured. Two independent mathematical models of synergy, Bliss and Loewe,44–46 were used to assess the combination effect, with synergy score > 5 indicating synergy. The combination of MK-1775 and MK-8776 shows very strong synergistic effect in all of the organoid models tested (Supplementary Fig. 3a). The average synergy score was greater than 10 for 2 of the PDXOs (Table 1) using either of the synergy models with the highest score greater than 50, suggesting very strong synergistic interaction between the two drugs.
Table 1
Summary of synergy of combination drug treatments in pancreatic PDXOs
Drug combination | PDXO | Highest Synergy Score | Average Synergy Score |
Bliss | Loewe | Bliss | Loewe |
MK1775 + MK8776 | PA5389 | 50.7 | 71.6 | 12.822 | 13.619 |
PA2847 | 52.92 | 51.71 | 10.556 | 10.594 |
PA1252 | 54.81 | 8.82 | 38.62 | -5.29 |
MK1775 + gemcitabine | PA5389 | 48.8 | 44.57 | 8.402 | 14.947 |
PA2847 | 32.19 | 15.34 | -4 | -5.043 |
PA1252 | 58.72 | 44.75 | 14.234 | 12.065 |
MK8776 + gemcitabine | PA5389 | 64.08 | 56.78 | 19.925 | 21.492 |
PA2847 | 60.1 | 42.64 | 11.552 | -4.805 |
PA1252 | 79.31 | 32.273 | 70.89 | 27.614 |
In combination with gemcitabine, both MK-1775 and MK-8776 shows very strong synergistic effect in all the organoid models (Supplementary Fig. 3a) with average synergy score was greater than 10 in the p53 mutated models consistent with previous reports 39, 40. The highest synergy scores > 30 were seen with MK8776 and gemcitabine for both mathematical models (Table 1). However, some low average values were also observed suggesting antagonism with certain dose combinations. The highest synergy scores across similar concentrations for gemcitabine + MK1775 and gemcitabine + MK8776 were compared for each PDXO (Supplementary Fig. 3d). The peak of the synergy score was significantly greater for gemcitabine + MK8776 than gemcitabine + MK1775 in PA2847-PDXO (containing a nonsense p53 mutation) identified by the Bliss mathematical model (Fig. 2f), which is more compatible for non-interacting drugs that elicit responses independently (e.g. by targeting separate but complement pathways). Overall, the rapid screening of relevant organoids in a combination matrix approach enabled synergistic profiling to be easily conducted providing valuable insight into combination strategies.
In vitro PDXO pharmacology correlates with in vivo PDX pharmacology. PDXs have been shown to be particularly useful in assessing targeted therapies on subjects with specific oncogenic driver mutations5, 6, 8, 47. To test targeted treatments, we selected PDXOs with similar relevant driver mutations. PDXO-LU1235 which harbors a common EGFR activation exon-19 deletion5, 6 was treated with the first-generation EGFR inhibitor, erlotinib and a third-generation inhibitor AZD9291 at various concentrations in 384-well plate to establish an IC50 value. PDXO-LU1235 showed sensitivity to both erlotinib and AZD9291 (IC50 < 1 µM, Fig. 3a & Table 2), which was consistent with the complete tumor regression that was observed in PDX in vivo testing for erlotinib5 and AZD9291 (unpublished data). In comparison, PDXO-LU2512, which does not carry an EGFR activation mutation, showed poor response to erlotinib (IC50 > 20µM, Fig. 3c) similar to its parental PDX model (TGI < 20%). In addition, SOC cisplatin induced an IC50 of 8.8µ m in PDXO-LU2512 and 4.9µM in LU1235 which translated to an insensitive/partially sensitive response in the corresponding PDXs (TGI 55% and 60%, respectively).
Table 2
Correlation of NSCLC LU1235 PDXO and PDX pharmacology.
PDXO in vitro LU12350 | PDX LU1235 |
Drug | IC50 (µM) | Max Inhibition (%) | Drug Sensitivity | Treatment Regimen | Median TGI (%) | Drug Sensitivity |
Erlotinib | 0.1133 | 77.99 | Sensitive | 50 mg/kg, p.o., q.d. | 107 | Sensitive |
AZD9291 | 0.3998 | 84.43 | Sensitive | 10 mg/kg p.o. q.d | 110 | Sensitive |
We next investigated the pharmacological correlation of PDXO in vitro response with in vivo PDX response in a large panel of paired PDXO and PDX models in order to assess the overall potential of PDXO to predict in vivo outcome to drug treatment. To this end, we examined a panel of 5 SOC chemotherapies and 7 targeted agents, across 4 cancer types and 13 PDXO/PDX matched pairs where drug effects were categorized as either sensitive or insensitive based on specific criterion for the resulting IC50 value and TGI (Fig. 3b). Statistical analysis of 30 data points indicated that in vitro and in vivo pharmacology characterizations were not independent, and that in vitro response was predictive of the in vivo outcome with overall ~ 86% accuracy (p = 0.001134, Fisher’s exact test, Fig. 3b), with positive predictions of 75% and negative prediction of 91%. In comparison, PDXs in 3D ex vivo assay had a predictive power of 68%, with positive prediction of 27% and negative prediction of 92% when 40 different drug treatments were compared to the corresponding in vivo PDX response (Supplementary Fig. 2c). Our data suggest that PDXO in vitro pharmacology has overall good predictive power for the corresponding PDXs in vivo in comparison to 3D ex vivo PDX assays in addition to higher reproducibility (Fig. 2c).
PDXO are tumorigenic and similar to their ancestry PDX. If PDXOs are “biologically equivalent” to their corresponding parental PDX, as shown above for the similar genomic, pathological and pharmacological properties, PDXOs should grow in vivo to form xenografts similar to the original PDX. To test this hypothesis, we inoculated PDXOs into immune-compromised Balb/c nu/nu mice to test their tumorigenicity and tumor growth kinetics. We found tumor growth kinetics to be similar to those of the parental PDX (Fig. 3c). The response to erlotinib in NSCLC LU2512 (EGFR wildtype) PDX in vivo (TGI 19.87% with 5 mg/kg IP weekly dose), PDXO in vitro (IC50 value of 21.04µM and 65.19% maximum inhibition) and PDXO xenograft in vivo (53% TGI with 5 mg/kg IP weekly dose) was comparable with poor sensitivity to erlotinib reflected in all three scenarios (Fig. 3c). The H&E staining of PDXO, PDX, and PDXO-derived tumors across various cancer types demonstrated consistent histopathology (Fig. 3d). This further supports the hypothesis that PDXOs are “biologically equivalent” to the parental PDXs, adding value to PDX/PDXO paired libraries as “patient-derived disease models” for future drug discovery and development.
Engineering reporter-PDXO and reporter-PDX to enable imaging analysis. One of the application limitations of PDXs is the challenge associated with engineering the models, e.g. introducing GFP or luciferase reporters and/or overexpression of human tumor associated antigens (TAA), creating drug resistant mutations, etc., due to the limited ability to culture PDX in vitro or to engineer directly in vivo. We therefore engineered the stable in vitro PDXO cultures, for both in vitro and in vivo applications. Specific constructs were designed for lentiviral transduction of PDXO cultures. For orthotopic implantation with longitudinal real-time optical imaging, LI6664-PDXO was efficiently transduced with a luciferase gene under ubiquitous promoter via lentiviral transduction. The resulting LI6664-luc PDXO was implanted subcutaneously to firstly establish tumorigenicity and bioluminescent signal (Fig. 4a). Similar to the parental PDX, LI6664-PDXO is c-met amplified (copy number variant determined by WES > 20 for both PDX and PDXO), therefore the response to crizotinib was tested in the PDXO xenograft and compared back to the original PDX (Fig. 4b & c respectively). Both the response to crizotinib (measured by tumor volume) and the histology (by H&E staining) were equivalent to the original PDX. Implantation of LI6664-luc PDXO into the liver was also successfully established with longitudinal growth measured by optical imaging enabling quantification of the liver tumor (Fig. 4d). LI6664-PDXO was also transduced with GFP reporter gene (Supplementary Fig. 4a). FACS was used to isolate GFP expressing organoids to generate mixed cultures or single clones (Supplementary Fig. 4b), which could be used in vitro with potential value in future in vitro HTS imaging. The efficient transduction of PDXOs, retention of biological features and quantification by imaging will facilitate the use of patient relevant tissue for both in vivo and in vitro applications as well as the generation of new models.
Organoid co-culture systems to investigate immune-oncology modality. IO is an area of intensive research due to the successful development of new immuno-therapeutics and the rapid growth of knowledge on the role of TME interactions, particularly tumor-infiltrating leukocytes (TILs), including T-cells. The limitations of current IO animal modelling calls for alternative model systems4. Recapitulating the TME by co-culturing tumor organoids with immune cells becomes highly attractive, providing an efficient and defined approach to assess immune modulatory and tumor killing effects of investigational IO therapeutics, including monoclonal antibodies, CAR-T cells, CAR-NK and small molecules48, 49.
Here, we co-cultured a PDXO (GA0091) with allogeneic peripheral blood monocyte cells (PBMCs) to assess the effects of allogeneic T-cell killing of PDXOs. CFSE-labeled GA0091-PDXO were co-cultured with activated (72hr-anti-CD3/CD28 antibodies) allogeneic PBMCs, where organoid killing by allogeneic T-cells was assayed by flow cytometry using a live/dead dye (Supplementary Fig. 5a). The results demonstrated allogeneic T-cell mediated organoid killing as shown in Supplementary Fig. 5b. Organoid killing can also be assayed by other types of readout, including automatic monitoring through the killing of CMTPX Red-labeled PDXO using Incucyte™ (data not shown). Our results demonstrate the feasibility of PDXO co-culture to monitor immune cell-mediated PDXO killing.
Antibody dependent cellular cytotoxicity (ADCC) is one of the important MOA of many monoclonal antibody cancer therapies, including Herceptin™ (against HER2) and Erbitux™ (against EGFR). We next tested the ADCC by Herceptin™ against a HER2+ ovarian PDXO, PDXO-OV0250, which was confirmed to express surface HER2 from a panel of PDXOs (Supplementary Fig. 6a). In a co-culture of both PDXO-OV0250 and PBMC in the presence or absence of Herceptin, similar to a standard ADCC assay, we demonstrated effective specific killing of PDXO-OV0250 through ADCC MOA as monitored by flow cytometry (Fig. 5a). Thus, PDXO co-culture could also be a good candidate system to evaluate antibody drug targeting tumor associated antigens.
Next, we tested the killing of PDXO by CAR-T cells in a co-culture system with different pairs of CAR-T cells and PDXOs using different endpoint readout methods (Fig. 5b). First, we engineered a luciferase expressing liver tumor PDXO, PDXO-LI6677-luc to over-express CD19 (Supplementary Fig. 6b). We then co-cultured PDXO-LI6677-CD19-luc with CD19-CAR-T cells and quantified the luciferase activity following co-culture. A significant reduction of luciferase activity was observed in the CD19+ PDXO co-culture only, suggesting the specific killing of CD19+ PDXO mediated by the CD19-CAR-T cells, as shown in Fig. 5b (p < 0.001, unpaired t-test). The result also suggests that luciferase reporter could be a convenient readout for CAR-T cell mediated killing.
In the second experiment, we tested EpCAM-CAR-T cells co-cultured with EpCAM+ gastric PDXO-GA0091 and EpCAM− melanoma PDXO-ME1154, and measured the interferon-gamma and Granzyme B levels in the culture. Our result demonstrated CAR-T cell-specific activity reflected by the elevation of both interferon-gamma (Fig. 5d) and Granzyme B (Fig. 5e) in EpCAM+ gastric-PDXO-GA0091 co-culture as compared to the EpCAM− melanoma PDXO-ME1154 co-culture. In addition to the above CAR-T cell co-culture experiments where the CART-mediated activities were measured by surrogate parameters, the CART-mediated specific killing can also directly be measured by other readouts, including flow cytometry (e.g. Du et al, in submission, where CAR-T-B7H3 were co-cultured with B7H3+ lung LU6438 PDXO with B7H3-targeting CAR-T cell). Collectively these results demonstrate the utility of an organoid co-culture systems for CAR-T efficacy and specificity as proof of concept (POC).