Pharmaco-genomic characterization of pancreatic and biliary tract cancer tumoroids for drug response

Summary

Pancreatic ductal adenocarcinoma (PDAC) and biliary tract cancer (BTC) are highly aggressive malignancies with limited therapeutic options. In this study, we established eleven tumoroids with paired patient-derived xenograft (PDX) models (five PDAC and six BTC), enabling scalable in vitro drug screening and corresponding in vivo validation. These tumoroids retained the histological and genetic characteristics of their original tumors and exhibited varied responses to chemotherapeutic agents. Drug screening identified PI3Kα inhibitors as promising candidates for both PDAC and BTC tumoroids, which was further validated in matched PDX models. Moreover, we uncovered genetic alterations and transcriptomic signatures associated with different drug sensitivities. Notably, combining a G9a degrader (G9D-4) with the KRAS^G12D inhibitor MRTX1133 elicited synergistic anti-tumor effects in KRAS^G12D-mutant tumoroids. Overall, our study provides preclinical insights from a small PDAC and BTC tumoroid cohort, supporting tumoroid-based platforms for exploratory drug screening and pharmacogenomic analyses and suggesting potential therapeutic directions that warrant further validation.

Graphical abstract

Highlights

• •Establishment of PDAC & BTC tumoroids retaining histological and genetic tumor features
• •Matched tumoroids and PDX models allowing drug screening and in vivo companion validation
• •Identification of PI3K inhibitors as potential molecular therapeutic agents
• •Synergistic effect of MRTX1133 and G9A degrader in KRAS^G12D-mutant PDAC & BTC tumoroids

Health sciences

Introduction

Cancers of hepatic, pancreatic, and biliary tract (HPB), including hepatocellular carcinoma (HCC), pancreatic ductal adenocarcinoma (PDAC), and biliary tract cancer (BTC), account for a high rate of cancer-related mortality worldwide. Due to shortage of reliable detection methods for early diagnosis, HPB cancers are mostly diagnosed at an advanced and unresectable stage. Patients with inoperable cancers generally receive systemic therapies for the purposes of palliative care and prolonging life.¹ Great improvement of molecular target and immune therapy has been made in HCC; in contrast, multi-agent chemotherapies remain the major treatment regimens for patients with advanced or metastatic PDAC and BTC.^2,3,4 However, the effects are limited, and the overall 5-year survival remains dismal, which necessitate identifying more efficient treatment options for PDAC and BTC patients.

Over the last two decades, targeted therapy has served as a backbone of cancer treatment, highlighting the necessity and importance of precision oncology. For a considerable period of time, erlotinib, a selective epidermal growth factor receptor (EGFR) tyrosine kinase inhibitor, was the sole available molecular targeted drug for PDAC treatment; meanwhile, no molecular targeted therapies were available for BTC. Tremendous efforts have been made to expand the targeted therapy options for PDAC and BTC. As a result, the landscape of targeted therapy options has gradually expanded, albeit not extensively. For PDAC, the PARP inhibitor olaparib has been approved for patients harboring BRCA1 or BRCA2 mutations.² In BTC, targeted therapies have emerged for tumors with oncogenic fibroblast growth factor receptor 2 (FGFR2) fusions (pemigatinib and futibatinib) and gain-of-function variants of isocitrate dehydrogenase 1 (IDH1) (ivosidenib).³ In addition, TRK inhibitors larotrectinib and entrectinib, as well as the BRAF inhibitor dabrafenib and the MEK inhibitor trametinib in combination, have been approved for the treatment of solid tumors, including PDAC and BTC, with NTRK gene fusion and BRAF V600E mutation, respectively. Nonetheless, the clinical impact of these targeted therapies remains limited due to the low prevalence of these actionable genetic alterations in PDAC and BTC.^5,6

With advancements of next-generation sequencing (NGS), large-scale genomic studies have identified many potential driver mutations in PDAC and BTC, some of which may serve as therapeutic targets. In contrast to patients with other tumor types, individuals with PDAC and BTC who harbor actionable mutations do not consistently benefit from matched targeted therapies. Several factors may contribute to this discrepancy. First, both PDAC and BTC are heterogenetic cancers with a bunch of gene mutations. The abundance of these mutations and drug-induced selection pressure can promote clonal evolutions, ultimately influencing treatment response.^7,8 Second, PDAC and BTC are characterized by a highly desmoplastic tumor environment, suggesting that genetic alterations alone do not fully determine therapeutic outcomes.⁹ Therefore, drug responses should be evaluated using reliable preclinical models that accurately recapture the heterogeneity of PDAC and BTC in clinical settings.

The application of patient tumor-derived preclinical models, such as patient-derived xenograft (PDX), has brought significant benefit to translational cancer research.¹⁰ In recent years, patient-derived organoids (PDOs), also known as tumor organoids or tumoroids, have emerged as a powerful in vitro model for studying tumor biology and evaluating anti-cancer therapies, as they faithfully preserve the genetic characteristics and heterogeneity of the original tumors.¹¹ This in vitro setting enables efficient genetic manipulation and large-scale applications, which are challenging to achieve in PDX models.

In this study, we established PDAC and BTC tumoroids by using fresh tumor tissues from patients (PDOs) or PDX (PDXOs) to model histological features of these two tumor types and evaluate their responses to several anti-cancer drugs. After confirming their varied responses to chemotherapeutic drugs, we screened a series of targeted therapeutic agents and identified that PI3Kα inhibitors demonstrated promising in vitro efficacy against our established PDAC and BTC tumoroids, which was further validated in the matched PDX models. Genetic alterations and gene expression profiles were analyzed to identify potential molecular signatures predictive of drug response. Furthermore, given that KRAS mutations frequently occur in PDAC and BTC and are always associated with poor prognosis and unfavorable therapeutic outcomes, we also attempted to explore potential combinational therapies. We revealed that combining a G9a degrader (G9D-4) with a KRAS inhibitor (MRTX1133) exerts a significant synergistic inhibitory effect on KRAS^G12D-mutant PDAC and BTC tumoroids, providing a clue for clinical development. Together, our study demonstrated that tumoroids can effectively bridge the gap between molecular profiling and therapeutic development, enabling the discovery of tailored treatment strategies for PDAC and BTC.

Results

Establishment of PDAC and BTC tumoroids with matched PDX models

We successfully established 11 tumoroids with matched PDX models, including 5 PDACs and 6 BTCs. The BTC group comprised 5 intrahepatic cholangiocarcinoma (ICC) and 1 gallbladder cancer (GBC) tumoroids. Among these, two PDAC tumoroids were directly generated from patient tumor tissues. In detail, PDO and the matched PDX model of PC-001 were established by using surgical resection, while PDO PC-002 was generated by using diagnostic needle biopsies, and its matched PDX was established by transplanting PDOs into immunocompromised mice. The other nine PDXOs were generated by using PDX-derived tumor tissues.

Consistent with previous studies, both PDAC and BTC tumoroids displayed diverse morphologies, ranging from cystic hollow structures to cohesive/solid spheres (Figure 1A). Generally speaking, tumoroids formed within one week of seeding and exhibited progressive growth over time (Figure 1B). The morphological feature of the tumoroids was well retained, even after being expanded more than 20 passages or cultured for more than 1 year (Figure 1C). The average passage time for all tumoroids was approximately 14–21 days (Figure 1D). Notably, the passage times across different generations of PDAC tumoroids remained relatively consistent, while BTC tumoroids exhibited longer initial passage times that shortened after 4–5 passages (Figure 1E).

Figure 1.

Establishment of PDAC and BTC tumoroids

(A) Representative brightfield microscopy images of 5 PDAC tumoroids and 6 BTC tumoroids. Scale bars, 100 μm.

(B) Brightfield images of time course showing the growth of the representative tumoroids at the same passage. Scale bars, 100 μm.

(C) Brightfield images of the representative tumoroids at different passages. Scale bars, 100 μm.

(D) The average growth time of each tumoroid.

(E) The time required for the tumoroids to grow to be passaged across different generations.

PDAC, pancreatic ductal adenocarcinoma; BTC, biliary tract cancer.

Histopathological characterization of established tumoroids

We next performed H&E staining to observe the histopathological features of tumoroids and compared them with their original tumors. At the histological level, the tumoroids exhibited more heterogeneous phenotypes. These cystic-like tumoroids consisted of either a single-layered cell (PC-001 and CC-003) or multi-layered cells (PC-002, PC-028, and PC-043). Those solid tumoroids exhibited either glandular structures (CC-078 and CC-097) or cribriform structure (PC-249, CC-001, CC-115, and GBC-676). Of note, the morphological feature of each tumoroid was highly consistent with its corresponding primary tumors in patients or xenografts (Figures 2A and 2B).

Figure 2.

Histopathological characteristics of PDAC and BTC tumoroids and their primary tumors

(A and B) H&E staining. Scale bars, 50 μm (tumoroids) and 100 μm (xenografts).

(C and D) IHC staining of the protein markers CK19, CK7, PDX1, and EpCAM. Scale bars, 50 μm.

(E and F) IHC staining of the proliferative marker Ki67. Scale bars, 50 μm.

In addition to histological characterization, the expression of tumor-specific biomarkers was evaluated by IHC analysis. As expected, tumoroids displayed similar expression levels of biomarkers as the parental tumors. Specifically, PDAC tumoroids expressed PDX1, CK19 and EpCAM, while CC tumoroids expressed CK7, CK19, and EpCAM but did not express the HCC marker HepPar1 or AFP (except for CC-115) (Figures 2C, 2D, and S1). In the case of GBC, tumor tissue had weak expression of CK7, which was maintained precisely in its tumoroids (Figure 2D). Furthermore, Ki67 staining showed moderate-to-high expression in all tumoroids, which reflected their ability to proliferate and propagate (Figures 2E and 2F). Together, the H&E and IHC assay demonstrated that the histological characteristics of the PDX tumor were preserved in tumoroids.

Genetic alteration of tumoroids and their original tumor tissues

To determine whether tumoroids captured major genomic characteristics of their parental tissues, tumoroids and their original tumor tissues were subjected to whole-genome sequencing (WGS) analysis. As shown in Figure 3A, the distribution of single base substitutions in the exon of all samples (including tumor tissues and tumoroids) demonstrated over-representation of the G>A/C>T and T>C/A>G transversions both in PDAC and BTC, as described in previous studies.^12,13 The proportion of exonic variations in the original tumor tissues was well retained in most tumoroids (Figure 3B). Mutational profiles of the genes recurrently mutated in PDAC and BTC were compared between tumoroids and their original tumor tissues. As shown in Figure 3C, most gene mutations identified in tumor tissues were retained in the tumoroids. Consistent with previous studies, KRAS and TP53 mutations had the highest frequency, and CDKN2A, SMAD4, ARID1A, and ARID2 mutations were also identified in the PDAC and BTC samples. Less concordance was observed in PC-001, CC-003, and GBC-676 samples, which might be attributed to intratumor heterogeneity (ITH) and clonal selection during tumoroid culture, and this phenomenon has also been presented in previous studies.^14,15 Besides gene mutation, tumoroids also recapitulated the copy number variation (CNV) patterns of paired tumor tissues in PDAC and BTC driver genes. In 5 of 8 cases, specifically PC-001, PC-043, PC-249, CC-115, and GBC-676, the profile of CNVs found in tumor tissues was well retained in the tumoroids (Figure 3D). Genes including CCND1, FGF3, FGF19, and Myc were frequently amplified, while tumor suppressor genes, including AXIN1 and RB1 were deleted in most samples. Interestingly, certain CNVs undetectable in tumor tissues were identified in the matched tumoroids, including the matched samples of PC-028, CC-001, and CC-003. Considering the high concordance of gene mutations in PC-028 and CC-001, we propose that the differences in CNVs between tumor tissues and tumoroids of these two samples might be due to the enrichment of tumor cells in tumoroids. In addition, the gene expression pattern of tumoroids and their corresponding PDX tumor tissues was analyzed using RNA-seq. Principal component analysis (PCA) showed that samples from the same patient clustered closely along the PC1 component, whereas the PC2 component captured the variance between tissues and tumoroids (Figure S2A). The correlation coefficients for the ten paired tumoroids and PDX samples were high, with a mean value of 0.88 (Figure S2B). These findings indicated that our established PDAC and BTC tumoroids recapitulated genetic and transcriptomic features of the original tumors.

Figure 3.

Genomic profiling in PDAC and BTC tumoroids and corresponding primary tumors

(A) Overall distribution of base substitutions detected in all samples of PDAC and BTC.

(B) Proportion of all variations detected in each tumoroid line and the corresponding tumor tissues.

(C) Overview of the mutations detected in tumoroid lines (O) and the corresponding tumor tissues (T) across common pathways.

(D) Overview of the gene amplification and deletion detected in tumoroids and tumor tissues.

Drug response of tumoroids to chemotherapeutics

To evaluate the drug sensitivity of PDAC and BTC tumoroids to conventional chemotherapeutics, five commonly used chemotherapeutic drugs (paclitaxel, SN38, gemcitabine, 5-FU, and cisplatin) were applied to 11 tumoroids. The tumoroids were treated with a dilution series of each compound for 6 days, and then, cell viability was measured. The assay was conducted with two biological replicates, and the correlation of IC₅₀ values showed an R value of 0.90 (p < 0.0001; 95% CI, 0.82–0.95) (Figure S3A).

Tumoroids exhibited varied responses to these drugs, but generally speaking, all PDAC and BTC tumoroids were sensitive to paclitaxel and SN38 and resistant to cisplatin (Figures 4A and 4B). 5-FU is the most conventional chemotherapeutic drug for patients with PDAC and BTC; thus, we then analyzed the gene expression characteristics of tumoroids by bulk RNA-seq according to their responses to 5-FU. The tumoroids were separated into two groups based on their response to 5-FU (sensitivity, AUC < 60%; resistance, AUC > 60%), and the gene expression profiles were compared between the sensitive (PC-001, CC-003, CC-097, CC-115, and GBC-676) and resistant groups (PC-028, PC-043, PC-249, and CC-001). We identified 1,007 differentially expressed genes (DEGs) between these two groups, with 633 being upregulated and 344 being downregulated (Figure 4C). Significantly upregulated tumor suppression genes (TSGs), including AXIN2, MST1, MSMB, and WNK2, and apoptosis regulation factors, including BBC3, BMF, BNIP3, and EAF2 were identified in the 5-FU-sensitive group compared to the 5-FU-resistant group. Meanwhile, some drug resistance-related genes, such as AKR1C2, CYP2C18, and MAOB, were significantly upregulated in the 5-FU-resistant group (Figure 4D). KEGG analysis demonstrated the upregulation of the TNF signaling pathway and necroptosis in 5-FU-sensitive tumoroids, while genes associated with tumor invasion and metastasis, such as ACTG1, ADCY7, COL4A4, GNB4, ITGA2, ITGA5, ITGB4, and ITGB5 were significantly upregulated in 5-FU-resistant tumoroids. These genes were enriched in focal adhesion, actin cytoskeleton, and PI3K-Akt signaling pathway (Figures 4D and 4E).

Figure 4.

Different responses of tumoroids to chemotherapeutics

(A) Dose-response curves of PDAC and BTC tumoroids to each chemotherapeutic drug. Data represent relative cell viability values, with DMSO-treated tumoroids used as control. Error bars represent the mean ± SD of three triplicate wells. The experiments were repeated twice.

(B) Violin plots showing IC₅₀ values of five chemotherapeutic drugs in 11 tumoroids.

(C) Volcano plots showing differentially expressed genes of the 5-FU-sensitive and -resistant groups.

(D) Heatmap of the expression levels of differentially expressed genes between the 5-FU-sensitive and -resistant groups. The colored bar represents the log₂-fold change values base on TPM.

(E) Top ten pathways enriched in the 5-FU-sensitive and -resistant groups.

(F) Dose-response curves of the PDAC tumoroid PC-002 to five different chemotherapeutic drugs. Data represent relative cell viability values, with DMSO-treated tumoroids used as control. Error bars represent the mean ± SD of three triplicate wells. The experiments were repeated twice.

(G and H) Tumor growth curve (G) and tumor weight (H) of the PC-002 PDX after treatment with 5-FU, gemcitabine, and taxol. Data are representative of 5–12 mice (vehicle group, n = 12; 5-FU and gemcitabine groups, n = 6; taxol group, n = 5). Error bars represent the mean ± SD. ∗p < 0.05 and ∗∗p < 0.01 vs. vehicle.

To validate the drug-response results observed in PDOs, we evaluated the anti-tumor effects of chemotherapeutics in vivo. As shown in Figure 4F, PC-002 tumoroids were sensitive to paclitaxel, gemcitabine, and 5-FU. Consistently, mice engrafted with PC-002-PDOXs demonstrated significant treatment response to all these three drugs. The tumor burdens, as indicated by tumor volumes and tumor weights, were apparently decreased in all the treatment groups compared with those in the control group (Figures 4G and 4H).

Drug screening of molecular target drugs in the established tumoroids

Next, we performed drug sensitivity testing in 11 tumoroids to identify potential molecular target drugs for PDAC and BTC. Thirteen kinase inhibitors were tested, including drugs in clinical use or under development. Among these, two compounds (a multi-target angiogenesis inhibitor, AL3810, and a selective PI3Kα inhibitor, CYH33) were previously developed by our lab in collaboration with chemists and are currently undergoing clinical trials.^16,17

Drug sensitivity was represented by two common summary statistics, the normalized area under the dose-response curve (AUC) and the half-maximal inhibitory concentration (IC₅₀) (Figures 5A and 5B; Table S3). A positive correlation was observed between the AUC and IC₅₀ values for each of the drugs presented in Figure 5C (Spearman’s r > 0.61). In alignment with the results observed in chemotherapy treatments, the molecular target drug-response profiles varied among different tumoroids. Specifically, different tumoroids showed varied sensitivities to the same drug, and the same tumoroid exhibited varied responses to different drugs. Among four multi-target tyrosine kinase inhibitors (TKIs), tumoroids were less sensitive to AL3810 and lenvatinib than to sorafenib and regorafenib (Figures 5A, 5B, and S4A). However, AL3810 at a relative low dose (5 mg/kg) exhibited comparable or even superior efficacy in inhibiting tumor growth across a panel of PDX models compared with sorafenib at a higher dose (60 mg/kg) (Figure S4B). This discrepancy between tumoroid and PDX model responses might be partially attributed to in vivo pharmacokinetic differences and distinct kinase inhibition profiles of these TKIs.^16,18 Specifically, sorafenib and regorafenib possess a broader kinase inhibition spectrum that includes Raf family members in addition to shared targets with AL3810 and lenvatinib,¹⁸ which may contribute to their enhanced activity in tumoroid models. These findings highlight the importance of integrating multiple preclinical models to comprehensively evaluate the efficacy of multi-target TKIs.

Figure 5.

The PI3Kα inhibitor CYH33 demonstrated potent anti-cancer effect against PDAC and BTC

(A) Summary of the different molecular targeted compounds used in drug screening, and the screen results represented as a summary of 1-AUC.

(B) Scatterplots showing IC₅₀ distribution of the screened compounds in 11 tumoroids.

(C) Scatterplot showing a positive correlation between IC50 values and AUC for most screened targeted compounds.

(D) A significant positive correlation of IC₅₀ values between CYH33 and alpelisib.

(E) Tumor growth curves of three PDX models (PC-028, CC-115, and GBC-676) after treatment with CYH33 and ravoxertinib. Data are representative of 6/12 mice (vehicle group, n = 12; drug-treated groups, n = 6). Error bars represent the mean ± SD. ∗p < 0.05 and ∗∗∗p < 0.001 vs. vehicle.

(F) T/C values of seven PDX models after treatment with CYH33 and ravoxertinib.

(G) Copy number variations and mutations of the selected genes.

(H) Top five pathways enriched in the CYH33-sensitive and -resistant groups.

It was noteworthy that most tumoroids demonstrated pronounced sensitivity to the selective PI3Kα inhibitors alpelisib and CYH33, as well as the CDK4/6 inhibitor abemaciclib, whereas their response to fexagratinib (an FGFR inhibitor), ravoxertinib (an ERK inhibitor), and crizotinib (a multi-target TKI against ALK/MET/ROS) was relatively modest (Figures 5A and 5B). Meaningfully, there was a strong correlation between the IC₅₀ values of alpelisib and CYH33, which reflected their similar mechanisms of action (Figure 5D). To validate the drug response results identified in tumoroids, we evaluated the in vivo anti-tumor effect of CYH33 and ravoxertinib, as representative of sensitive and resistant drugs, using the matched PDX models. Consistently, CYH33 demonstrated potent efficacy in PDX models, with 5 out of 7 tested models resulting in a T/C value of approximately 40%. Although ravoxertinib exhibited a certain degree of inhibitory effect on tumor growth, only one PDX model could achieve a T/C ratio of less than 40% (Figures 5E and 5F).

Based on different responses to CYH33 (AUC threshold: 60%), tumoroids were stratified into sensitive (PC-028, PC-043, CC-001, CC-003, CC-097, CC-115, and GBC-676) and resistant categories (PC-001, PC-249, and CC-078). Genetic alteration and gene expression profiles were analyzed to identify potential factors contributing to the differential sensitivities between the two groups. Of note, amplification of CCND1 and EGFR and MET, as well as TP53 mutation were present in ≥50% CYH33-sensitive tumoroids. There was no significant difference in other gene mutations between the sensitive and resistant tumoroids, including KRAS, ARID1A, PIK3CA, CDKN2A, and BRAF (Figure 5G). RNA-seq analysis further demonstrated that several genes involved in cell cycle regulation and DNA replication pathway were significantly upregulated in CYH33-sensitive tumoroids (Figure 5H). Interestingly, there was a positive correlation between the IC₅₀ values of CYH33/alpelisib and abemaciclib, suggesting that genes involved in proliferative pathways might contribute to CYH33 sensitivity in PDAC and BTC (Figures S3B and S3C). In CYH33-resistant tumoroids, a number of autophagy- and mitophagy-related genes were enriched, including ATF4, BCL2L13, CFLAR, and SQSTM. Our analysis also identified the upregulation of FoxO and MAPK signaling in resistant tumoroids. Collectively, we demonstrated the gene expression signatures of PDAC and BTC tumoroids in response to CYH33. Collectively, these findings confirmed the anti-tumor activity of the PI3Kα inhibitor in vitro in our PDAC and BTC tumoroid cohorts, providing a scientific basis for its potential clinical application.

The synergistic combination strategy against KRAS^G12D tumoroids

Previous studies have reported that the most prevalent KRAS allelic variant is G12D in PDAC and BTC, with approximately 40% in PDAC and ICC patients.^19,20,21 By analyzing the KRAS mutation subtypes, we found that three PDAC tumoroids (PC-001, PC-028, and PC-249) and one ICC tumoroid (CC-115) harbor G12D mutation, while the other tumoroids carry either G12V mutation (PC-043) or are wide type (CC-001 and CC-003) (Figures 3C and 6A). MRTX1133 is a selective non-covalent KRAS^G12D inhibitor and has exhibited promising anti-tumor efficacy in G12D-mutant PDAC and CRC cell lines and PDX.^22,23 However, effectiveness of MRTX1133 as a single agent is expected to be compromised due to the occurrence of acquired drug resistance.²⁴ Therefore, potential combination therapies should be developed to negate the resistance and enhance the efficacy of KRAS^G12D inhibitors.

Figure 6.

The synergistic anti-proliferative effect of G9a degrader and MRTX1133 combination against KRAS^G12D tumoroids

(A) Common oncogenic mutations in tumoroid lines (O) and the corresponding tumor tissues (T).

(B) Survival rate of tumoroids after the indicated treatment. Data represent relative cell viability values, with DMSO-treated tumoroids used as control. Error bars represent the mean ± SD of three triplicate wells. The experiments were repeated thrice. ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001.

(C) 3D synergy maps via ZIP mathematical model of G9a degrader and MRTX1133 in CC-115 and CC-001 (representative of two and three independent experiments, respectively).

(D) Western blotting showing the effects of G9D-4 and MRTX1133 on their targeted proteins.

(E and F) Results of RT-qPCR (E) and western blotting (F), showing the effects of G9D-4 and MRTX1133 on Bim expression. ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001.

(G) Western blotting demonstrating the effect of G9D-4 and MRTX1133 combination on the increased expression of cleaved PARP and caspase-3.

G9a, also known as EHMT2 or KMT1C, is a histone methyltransferase that mediates the methylation of histone H3K9 and is overexpressed in various human cancers.²⁵ It was reported that the deletion of G9a inhibits KRAS mutation-driven pancreatic carcinogenesis.²⁶ Taking advantage of the different KRAS mutations in our established tumoroids, here, we compared the anti-tumor effect of MRTX1133 and G9D-4 (a selective G9a degrader) in combination against PDAC and BTC tumoroids with varying KRAS mutations. As expected, in tumoroids with G12D mutation, MRTX1133 exerted concentration-dependent inhibitory effect on tumoroid growth, which was significantly enhanced in the presence of G9D-4. However, in non-G12D-mutant tumoroids, MRTX1133 alone had no effect on tumoroid growth and almost did not influence the inhibitory activity of G9D-4 when used in combination. After calculating the synergy score matrix using the ZIP model, we also found that the peak value of the synergy score in CC-115 (KRAS G12D) was significantly higher than that in CC-001 (KRAS WT) (Figures 6B and 6C). Moreover, the combination effect of G9D-4 and MRTX1133 appeared to be unaffected by the co-occurrence of other mutations commonly existing in PDAC and BTC, such as TP53, CDKN2A, SMAD4, and ARID1A (Figures 6A and 6B).

The effects of these two compounds on the expression of their target proteins were evaluated by western blotting. G9D-4 effectively induced G9a degradation and decreased the expression level of its downstream protein H3K9me2. MRTX1133 significantly suppressed ERK1/2 phosphorylation (Figure 6D), which, in turn, reduced Bim phosphorylation at key regulatory serine residues and promoted the accumulation of the pro-apoptotic Bim protein.²⁷ Moreover, chemical perturbation or silencing of G9a also induced the expression of Bim by decreasing H3K9me2.²⁸ In our study, the combination enhanced both the mRNA and protein levels of Bim (Figures 6E and 6F). Consistent with this mechanism, the combination synergistically enhanced the apoptosis of tumoroids, as evidenced by the increased levels of cleaved PARP and caspase-3 (Figure 6G). KRAS is frequently co-mutated with other genes and is associated with worse drug response.^29,30 Here, we observed that co-mutation of KRAS with TP53, CDKN2A, SMAD4, ARID1A, or ARID2 did not affect the combination effect of G9D-4 and MRTX1133. These results suggested that the combination of G9a degrader and MRTX1133 represents a promising regimen for further evaluation in preclinical models and clinical cohorts of PDAC and BTC harboring KRAS^G12D mutation.

Discussion

Major hurdles in improving the prognosis of patients with PDAC and BTC can partially be attributed to the lack of effective systemic therapies. Compared to conventional 2D cell lines, PDO and PDX models more accurately preserve the differentiation status, histoarchitecture, and genetic heterogeneity of primary tumors, which make them an ideal research model for developing potential therapeutic strategies. In this study, we generated 11 tumoroid lines with matched PDX models. Using these tumoroids, we identified the potential utility of PI3Kα inhibitors for PDAC and BTC treatment by screening a series of molecular targeted drugs on tumoroids, which was further validated in PDX models. Furthermore, we revealed the synergistic anti-tumor effects of G9a degrader and the KRAS^G12D inhibitor MRTX1133 against PDAC and BTC tumoroids harboring KRAS^G12D mutation.

In line with previous studies, our established tumoroid lines also demonstrated long-term culture stability, being maintained for more than 20 passages while retaining their original morphology,³¹ thus overcoming the limitation seen in some other platforms where certain tumoroids can only sustain short-term in vitro growth. This high expandability provided a solid foundation for high-throughput drug screening and genome editing. Interestingly, the passaging intervals of PDAC tumoroids remained relatively consistent across passages, whereas BTC tumoroids showed longer intervals during early passages that gradually shortened over time. This suggested that tumoroid adaption to culture conditions may vary by tumor types. Histological features and biomarker expression analysis revealed that the tumoroids retained key architectural and molecular characteristics of their original tumors. Moreover, the proportion of genetic variations of primary tumors were well retained in tumoroids. Gene mutations and CNV frequently happened in PDAC and BTC tumors were also maintained in tumoroids. Notably, our cohort possessed genetic diversity covering key subtypes. Specifically, KRAS mutations are present in over 90% of PDAC cases, and among these, KRAS^G12D is the most frequent subtype.^32,33 In our cohort, KRAS-mutated samples accounted for a proportional fraction consistent with this clinical feature. For TP53, another core driver mutation, the prevalence ranges from 50% to 70% in PDAC, and in BTC, it varies between 30% and 45%.^34,35,36 The TP53-mutated samples in our collection align with these clinical mutation spectra, ensuring the representativeness of our models. Due to the limitation of sample size, our current cohort has not yet covered some other common mutations in PDAC and BTC. In PDAC, CDKN2A and SMAD4 are important driver mutations, with mutation rates of approximately 25% and 15%, respectively. ICC frequently harbors IDH1/2 mutations and FGFR2 fusions, while HER2 amplification is quite common in GBC.³⁷ We plan to expand our sample collection in future studies to incorporate these subtypes, thereby improving the heterogeneity coverage of our models. Together, these results demonstrated that the established PDAC and BTC tumoroids recapitulated the histological complexity and molecular features of their original tumors.

We next assessed the drug sensitivity of the tumoroids to five commonly used chemotherapeutic agents. As expected, both PDAC and BTC tumoroids showed substantial different responses to a single chemotherapy but were generally sensitive to paclitaxel and SN38 and resistant to cisplatin, which is consistent with the findings of a previous study.³⁸ Because 5-FU is widely used in the treatment of PDAC and BTC, the transcriptional profiles of tumoroids were investigated to reveal the different responses. In the sensitive group, well-acknowledged TSGs (AXIN2, MST1, MSMB, and WNK2) and apoptosis-related factors (BBC3, BMF, BNIP3, and EAF2) were upregulated. KEGG analysis further demonstrated the enrichment of the TNF signaling pathway and necroptosis, indicating that multiple cell death mechanisms might be involved. In contrast, resistant tumoroids exhibited increased expression of drug metabolism genes (AKR1C2, CYP2C18, and MAOB), potentially reducing 5-FU efficacy. Moreover, genes linked to invasion and metastasis were enriched in pathways such as focal adhesion and PI3K-Akt signaling, suggesting a shift toward a more aggressive and drug-resistant phenotype. Thus, the differential gene expression profiles in 5-FU-sensitive and -resistant tumoroids may at least partially explain the varied responses. Patients with high expression of resistance-related genes may benefit less from 5-FU-based chemotherapy and thus require alternative treatment strategies. However, the underlying regulatory mechanisms of these genes in 5-FU resistance necessitate more in-depth investigation, and their clinical validity needs to be verified in larger patient cohorts. Together, these findings demonstrated our established PDAC and BTC tumoroids as potentially effective tools for developing targeted therapies.

We further identified several compounds from a series of multiple kinase inhibitors that had anti-tumor property against a broad range of tumoroids, including PI3Kα inhibitors (CYH33 and alpelisib) and CDK4/6 inhibitor (abemaciclib). The anti-tumor efficacy of CYH33 observed in tumoroids was validated in the matched PDX in vivo. PIK3CA mutation is a known marker for alpelisib treatment in advanced breast cancer (HR-positive and HER2-negative) and may also be predictive in gynecological cancers, as suggested by a preliminary clinical trial.³⁹ Given that only one tumoroid (CC-115_O) harbors PIK3CA mutation, the association between this mutation and the sensitivity to PI3Kα inhibitors in PDAC and BTC requires further studies with a larger sample size. In addition, our previous study revealed that CCND1 amplification was correlated with increased sensitivity to CYH33 in esophageal squamous cell carcinoma (ESCC).⁴⁰ Here, CCND1 amplification was detected in four of the six CYH33-sensitive tumoroids, leading to the hypothesis that CCND1 amplification might be linked to CYH33 sensitivity in this context. This preliminary observation merits further validation in independent tumoroid cohorts or patient-derived datasets. Furthermore, our analysis identified the upregulation of cell cycle and DNA replication pathways in CYH33-sensitive tumoroids and the upregulation of autophagy, mitophagy, FoxO, and MAPK signaling pathway in CYH33-resistant tumoroids. These findings revealed the potent inhibitory activity of PI3Kα inhibitors in our established PDAC and BTC tumoroid models. Although the underlying mechanisms warrant further investigation, they provide valuable preclinical clues for clinical application. CYH33 is currently in phase 1/2 clinical trials for several solid tumors, including ESCC and head and neck squamous cell carcinoma (HNSCC). The correlation of patient response with the genetic and transcriptional signatures identified in our study deserves analysis for better clinical benefit, and the application of CYH33 in PDAC and BTC patients also warrants investigation.

The development of mutant-specific KRAS inhibitors has successfully turned KRAS from an “undruggable” to a “druggable” target. Following the approval of G12C inhibitors, the development of selective G12D inhibitors has been a significant focus. MRTX1133 is the first non-covalent and selective G12D inhibitor and is currently undergoing a multi-center phase 1/2 clinical trial (NCT05737706) for G12D-mutant PDAC, CRC, and NSCLC patients. Selective G12C and G12D inhibitors showed promising anti-tumor efficacy; however, the response was not durable, and acquired resistance has emerged as a big challenge.²⁴ Innovative combination strategies are being developed to overcome drug resistance and improve the efficacy of KRAS inhibition. Several molecular target drugs including the pan-ERBB inhibitor afatinib and the BCL2 inhibitor venetoclax have been demonstrated to enhance the anti-tumor effect of MRTX1133 in PDAC in preclinical models.^41,42 Our previous study also revealed that the G9a degrader G9D-4 could sensitize pancreatic cancer cells with KRAS^G12D mutation to the KRAS^G12D inhibitor MRTX1133.⁴³ Due to inefficient tumor accumulation of G9D-4, the anti-tumor effect of the combination strategy could not be evaluated in vivo. Here, our finding highlighted the synergistic effect and mechanism of G9D-4 and MRTX1133 in KRAS^G12D mutant PDAC and BTC tumoroids, raising the hypothesis that PDAC or BTC patients harboring KRAS^G12D mutation are likely to derive greater benefit from the combination therapy of G9a degrader and MRTX1133. This result provides impetus for further optimization of G9a degrader candidates with favorable drug-like characteristics. To translate this finding into clinical practice, we will further explore the complete molecular mechanism and actively collect more clinical samples to verify the synergistic efficacy of this combination in patient-derived models and clinical specimens.

In conclusion, we established PDAC and BTC tumoroids with matched PDX models to provide a comprehensive platform that enables scalable in vitro drug screening, while also offering insights into tumor biology and drug response in vivo. Our findings demonstrated the potent efficacy of PI3Kα inhibitors against patient-derived preclinical PDAC and BTC models and indicated that the combination of G9a degrader with MRTX1133 might be a potential strategy for inhibiting the viability of PDAC and BTC cells harboring KRAS^G12D mutation.

Limitations of the study

Although the gene mutations and CNVs frequently observed in PDAC and BTC tumors were well recapitulated in our established tumoroid models, some canonical but low-frequency mutations were not represented. Additionally, the current study is limited by a small and mixed cohort of PDAC- and BTC-derived tumoroids. While we identified some interesting findings using this cohort, including potential biomarkers for anti-cancer drugs and potential combination therapy strategies, large-scale validation and systematic investigation remain essential.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Yi Chen (ychen@simm.ac.cn).

Materials availability

Organoids generated in this study will be made available on request, but we may require a payment and/or a completed material transfer agreement if there is potential for their commercial application.

Data and code availability

• •All processed WGS and bulk RNA-seq data generated from this study have been deposited with GSA under accession code PRJCA052646. The variation data reported in this paper have been deposited in the genome variation map (GVM) under accession number GVM: GVM001234. The bulk RNA-seq counts reported in this paper have been deposited in the Open Archive for Miscellaneous Data (OMIX) under accession number OMIX: OMIX013486. R code is available on https://github.com/ChrisFang96/Pharmaco-genomic-characterization-of-pancreatic-and-biliary-tract-cancer-tumoroids-code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon reasonable request.

Acknowledgments

We thank Prof. Xiangyin Kong and Hailong Wang from the Laboratory of Tissue Microenvironment and Tumor, Shanghai Institute of Nutrition and Health, Chinese Academy of Sciences for assistance with WGS and RNA-seq data analyses. This study was supported by Innovative Drug Research and Development National Science and Technology Major Project (no. 2025ZD1804100), National Natural Science Foundation of China (no. U25D9023), the Strategic Priority Research Program of the Chinese Academy of Science (nos. XDB0830000 and XDB1060401), Shandong Laboratory Program (no. SYS202205), Program of Shanghai Academic/Technology Research Leader under the Science and Technology Innovation Action Plan (no. 22XD1404400), and Shanghai Municipal Science and Technology Commission “Shanghai Action Plan for Science, Technology and Innovation” in the field of experimental animal research project (no. 21140902000).

Author contributions

Conceptualization, J.D., Y.F., and Y.C.; methodology, Y.F. and Y.C.; investigation, K.F., W.Z., Q.Z., Y.H., R.W., B.Y., and Y.Z.; writing – original draft, Y.F.; writing – review & editing, K.F. and Y.F.; funding acquisition, J.D., Y.F., and Y.C.; resources, Y.S., B.Z., J.D., Y.F., and Y.C.; supervision, J.D., Y.F., and Y.C.

Declaration of interests

J.D. is the Chief Strategy Advisor of Shanghai HaiHe Biopharma Co., Ltd. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
p44/42 MAPK (Erk1/2) (137F5) Rabbit Monoclonal Antibody	Cell Signaling Technology	Cat# 4695, RRID:AB_390779
Phospho-p44/42 MAPK (Erk1/2) (Thr202/Tyr204) (D13.14.4E) Rabbit Monoclonal Antibody	Cell Signaling Technology	Cat# 4370, RRID:AB_2315112
Di-Methyl-Histone H3 (Lys9) (D85B4) Rabbit Monoclonal Antibody	Cell Signaling Technology	Cat# 4658, RRID:AB_10544405
Anti-Cleaved-Caspase 3 p17 Rabbit mAb	Epizyme	Cat# R013264
Anti-Cleaved-PARP1 Rabbit mAb	Epizyme	Cat#R012387
Bim Rabbit mAb	ABclonal	Cat# A19702, RRID:AB_2862744
alpha Tubulin Monoclonal Antibody	ABclonal	Cat# AC012, RRID:AB_2768341
Histone-H3 antibody	Proteintech	Cat# 17168-1-AP, RRID:AB_2716755
Alpha-1-Fetoprotein, AFP Rabbit Polyclonal Antibody	Genetech	Cat# GA000829
Beta-catenin (EP35) Rabbit Monoclonal Antibody	Genetech	Cat# GT211929
CK7 (GM303)Mouse Monoclonal Antibody	Genetech	Cat# GT244629
CK19 (RCK108)Mouse Monoclonal Antibody	Genetech	Cat# GM088829
Ep-CAM (Ber-EP4) Mouse Monoclonal Antibody	Genetech	Cat# GM080429
Hep Par-1 (OCH1E5) Mouse Monoclonal Antibody	Genetech	Cat# GM715829
Ki-67 (MIB-1) Mouse Monoclonal Antibody	Genetech	Cat# GM724029
PDX1 (EP139) Rabbit Monoclonal Antibody	Genetech	Cat# GT229329
G9A Rabbit mAb	AiFang biological	Cat# AFRM82243
Biological samples
Tumor tissues of PDAC patients	Fudan University Shanghai Cancer Center	N/A
Chemicals, peptides, and recombinant proteins
DMEM/F-12	Gibco	Cat# 11320033
GlutaMAX™ Supplement (100×)	Gibco	Cat# 35050061
HEPES (1M)	Gibco	Cat# 15630080
B-27™ Supplement (50×)	Gibco	Cat# 17504044
N-2™ Supplement (50×)	Gibco	Cat# 17502048
TrypLE™ Express Enzyme (1×), phenol red	Gibco	Cat# 12605010
N-acetyl-L-cysteine	Sigma Aldrich	Cat# A9165
Nicotinamide	Sigma Aldrich	Cat# N0636
Gastrin I (human)	Tocris Bioscience	Cat# 3006
A83-01	Tocris Bioscience	Cat# 2939
Human EGF, Animal-Free Recombinant Protein	PeproTech	Cat# AF-100-15
Human FGF-10 Recombinant Protein	PeproTech	Cat# 100-26
Human HGF Recombinant Protein	PeproTech	Cat# 100-39H
Human Noggin Recombinant Protein	PeproTech	Cat# 120-10C
Human R-Spondin 1 Recombinant Protein	PeproTech	Cat# 120-38
Matrigel® Matrix Basement Membrane Growth Factor Reduced, Phenol Red Free	Corning	Cat# 356231
Forskolin	Selleck	Cat# S2449
Y-27632	Selleck	Cat# S6390
Critical commercial assays
Tumor Dissociation Kit, mouse	Miltenyi	Cat# 130-096-730
QlAamp DNA Mini Kit	QIAGEN	Cat# 56304
RNeasy MinElute Cleanup kit	QIAGEN	Cat# 74204
EZ-press RNA Purification Kit	EZBioscience	Cat# B0004D
Evo M-MLV RT Mix Kit with gDNA Clean for qPCR Ver.2	Accurate Biotechnology	Cat# AG11728
SYBR Green Premix Pro Taq HS qPCR Tracking Kit	Accurate Biotechnology	Cat# AG11735
Deposited data
Human WGS and bulk RNA-seq, processed data	This paper	GSA: PRJCA052646
Oligonucleotides
BIM forward primer: TTGATTCTTGCAGCCACCCT	This paper	N/A
BIM reverse primer: GGAAGCTTGTGGCTCTGTCT	This paper	N/A
Actin forward primer: CCACGAGCGGTTCCGATG	This paper	N/A
Actin reverse primer: GCCACAGGATTCCATACCCA	This paper	N/A
Software and algorithms
GraphPad Prism Software version 10.4	GraphPad Prism	https://www.graphpad.com; RRID:SCR_002798
RStudio Software	RStudio	https://posit.co/download/rstudio-desktop/ RRID:SCR_000432
Maftools Package		https://github.com/PoisonAlien/Maftools; RRID:SCR_024519
ggplot2 Package	ggplot2	https://cran.r-project.org/web/packages/ggplot2/index.html; RRID:SCR_014601
Dplyr Package	dplyr	https://cran.r-project.org/web/packages/dplyr/index.html; RRID:SCR_016708
Tidyr Package	tidyr	https://CRAN.R-project.org/package=tidyr; RRID:SCR_017102
ComplexHeatmap Package	ComplexHeatmap	https://bioconductor.org/packages/release/bioc/html/ComplexHeatmap.html; RRID:SCR_017270
Viridis Package	viridis	https://cran.r-project.org/web/packages/viridis/vignettes/intro-to-viridis.html; RRID:SCR_016696
Pheatmap Package	pheatmap	https://www.rdocumentation.org/packages/pheatmap/versions/0.2/topics/pheatmap; RRID:SCR_016418
DESeq2 Package	DESeq2	https://bioconductor.org/packages/release/bioc/html/DESeq2.html; RRID:SCR_015687
gridExtra Package	gridExtra	https://CRAN.R-project.org/package=gridExtra; RRID:SCR_025249
Other
gentleMACS Octo Dissociator with Heaters	Miltenyi	Cat# 130-096-427
gentleMACS C Tubes	Miltenyi	Cat# 130-096-334
100 μm cell strainer	Falcon	Cat# 431752
24 Well Cell Culture Multiwell Plate	GreinerBio-one	Cat# 662102
Qubit™ 4 Fluorometer	Invitrogen	Cat# Q33226
NanoDrop	Thermo Fisher Scientific	N/A

Experimental model and study participant details

Tumor tissue processing

This study involved the establishment of tumoroids from fresh tumor tissues from two patients undergoing surgical resection or diagnostic needle biopsies at Fudan University Shanghai Cancer Center. Both patients participated with informed consent, with the research protocol approved by the ethical committee (No. 050432-4-1212B). Fresh resected tumor tissues from cancer patients or nude mice were mechanical minced into fragments, and digested with Tumor Dissociation Kit (Miltenyi Biotec) for tissue dissociation. The tissue was incubated at 37 °C for 1-2 h with mixing every 15 min. The digestion was stopped when no large pieces of tissue were left. The suspension was then filtered through a 100 μm cell strainer (Falcon) and spun for 5 min at 300 g. The pellet was washed in cold advanced DMEM/F12 (GIBCO) twice, and then cell numbers were calculated on the Countess II FL Automated Cell Counter (Thermo Fisher Scientific) after trypan blue staining.

Establishment of tumoroid cultures

Cell suspension was spun for 5 min at 300 g and resuspended in Matrigel (Corning). 5,000-10,000 cells were seeded per well in 24-well suspension culture plates (GreinerBio-one). After a 15-min incubation period, Matrigel dome were solidified and culture medium was added to the plates. For pancreatic cancer tumoroids, the culture medium consisted of advanced DMEM/F-12 supplemented with 1% penicillin/streptomycin, 1% Glutamax (GIBCO), 10 mM HEPES (GIBCO), 1:50 B27 (GIBCO), 1:100 N2 (GIBCO), 1.25 mM N-acetyl-L-cysteine (Sigma), 10 mM nicotinamide (Sigma), 10 nM [Leu15]-gastrin I (Tocris Bioscience), 50 ng/ml EGF (PeproTech), 100 ng/ml FGF10 (PeproTech), 25 ng/ml Noggin (PeproTech), 300 ng/ml R-Spondin-1 (PeproTech) and 5 μM A83-01 (Tocris Bioscience). For biliary tract cancer tumoroids, the same medium was further supplemented with 25 ng/ml HGF (PeproTech), 10 μM forskolin (Selleck), and 10 μM Y27632 (Selleck). The culture medium was changed twice a week. Tumoroid pictures were taken using an inverted microscope Leica DMi1 and Leica DFC 450C camera. Once mature, tumoroid cultures were passaged by dissociating in TrypLE Express (GIBCO) following the disruption of Matrigel by a wash step with cold basal medium (advanced DMEM/F12 supplemented with 1% penicillin/streptomycin, 1% Glutamax, and 10 mM HEPES).

Xenograft tumor model

All animal experiments were approved by the Institutional Animal Care and Use Committee of Shanghai Institute of Materia Medica (No. 2018-05-DJ-35, 2019-05-DJ-45 and 2022-06-DJ-68) and performed in accordance with the Association for Assessment and Accreditation of Laboratory Animal Care. Established PDX tumors were cut into 1 mm³ fragments and transplanted subcutaneously into the right flank of BALB/C nude mice (female, 4-6 weeks) using a trocar. When the tumor reached a volume of 100−200 mm³, the mice were randomly assigned into control and treatment groups (the dose and vehicle of each drug was provided in Table S2). Tumors were measured twice weekly and volumes were calculated using the formula TV = length × width² × 1/2. Individual relative tumor volume (RTV) was calculated as (TV on measured day) / (TV at initial treatment). The T/C rate was calculated as (final RTV treatment / final RTV vehicle) × 100%, and T/C value ≤ 40% is considered as effectiveness.⁴⁴

Method details

Immunohistological staining

Tumor tissues and tumoroids were collected and fixed with 4% paraformaldehyde for 24 h, and then embedded in paraffin, cut into 5 μm sections. The paraffin slides were deparaffinized, hydrated and stained with hematoxylin-eoslin (HE). For immunohistochemistry (IHC) staining, the slides were incubated in 3% H₂O₂ to eliminate endogenous peroxidase activity and subjected to antigen retrieval using citrate sodium solution (pH 6). To reduce background nonspecific staining, the slides were blocked with 1% BSA for 1 h at room temperature. After incubation with primary antibodies (anti-AFP, anti-β-Catenin, anti-CK7, anti-CK19, anti-EpCAM, anti-HepPar1, anti-Ki67, and anti-PDX1, purchased from GeneTech, Shanghai) overnight at 4°C and with secondary antibody-HRP conjugate for 1 h at room temperature, slides were finally developed with 3,3′-diaminobenzidine (DAB) for 5 min, counterstained with hematoxylin, and mounted with DPX (Sigma-Aldrich). The slides were scanned into images by using NanoZoomer S210 (Hamamatsu, Japan).

WGS and data analysis

Genomic DNA was extracted from tumoroids and the matched tumor tissues using QlAamp DNA Mini Kit (QIAGEN). DNA integrity was detected by agarose gel electrophoresis. DNA concentration was accurately quantified by Qubit Fluorometer (Invitrogen), and DNA purity (OD260/OD280 ratio) was detected by NanoDrop (Thermo Scientific). WGS libraries with an insert size of 350-450 bp were prepared using TruSeq Nano DNA library Prep Kit (Illumina) according to the protocol. The libraries were sequenced on an Illumina NovaSeq 6000 system following Illumina-provided protocols for 2×150 paired-end sequencing. Quality control of raw sequence data was performed by using FastQC (v0.11.9), and qualified reads were aligned to the human reference genome hg19 using BWA (v0.7.17). Sentieon (202010.01) was run to call single nucleotide variations (SNV) as well as insertion and deletion (InDel), and filter out low quality variations. ANNOVAR (date 20180416) was then used to annotate the variants. Cancer-related variants were further filtered by the following criteria: (i) exonic variants were picked. (ii) synonymous variants were removed. (iii) variants with a population frequency (>1%) in the ExAC database were filtered out. (iv) Cancer-related genes were identified by COSMIC database. Copy number alterations (CNVs) were detected by CNVkit (0.9.9) and annotated with AnnotSV. CNV results were shown using log2-transformed copy number ratio values, or visualized, where segments with log2 values greater than 0.4 or less than -0.4 are defined as amplification or deletion, respectively. Maftools, ComplexHeatmap, and pheatmap were used to visualized the results.

RNA-seq and data analysis

Total RNA was isolated from tumoroids using RNeasy MinElute Cleanup kit (QIAGEN). RNA concentration was quantified by NanoDrop (Thermo Scientific) and the integrity was measured on an Agilent 2100/2200 Bioanalyzer (Agilent Technologies). The cDNA synthesis, end-repair, A-base addition, and ligation of the sequencing adapters were performed according to Illumina’s TruSeq RNA protocol. RNA-sequencing was performed using Illumina NovaSeq 6000 system, and 150 bp paired-end reads were generated. The reads were aligned to the human reference genome hg19 using HISAT. Genes with read counts < 10 in 50% samples were filtered out. Gene-level read counts were transformed into TPM according to previous study.⁴⁵ The principal components for the tumoroids and tumor tissues were calculated from the normalized DESeq2 counts, and the first two (PC1, PC2) were plotted. To assess concordance of gene expression between tumoroid and tumor tissues, the Spearman’s correlation coefficient was calculated pairwise. Significantly differentially expressed genes between two groups were analyzed by DEseq2 package in R with read counts (p<0.05, log2-fold change ≥1 or ≤-1) and visualized as volcano plots by R ggplot2. Analysis of enriched pathways was annotated using KEGGREST.

Drug sensitivity assay

Tumoroids were dissociated into 2–5 cell clumps by enzymatic dissociation with TrypLE. Then 3000 cells per well in 4 μl of matrigel were plated into a 96-well flat clear bottom white polystyrene microplate (Corning) and cultured for 6 days to allow the growth of tumoroids. At day 7, the culture medium is replaced with fresh one and a concentration dilution series of tested drugs (the detailed information was provided in Table S1) was added to the medium. Cell viability was measured after 6 days using CellTiter-Glo 3D Cell Viability Assay (Promega). Luminescence was measured on a Multi-Mode Microplate Reader (BioTek). Cell viability was normalized to vehicle and calculated as ODtreated cells/ODcontrol cells ×100%. IC₅₀ was calculated using Prism (GraphPad) software. Area under curve (AUC) was calculated using R package (nplr).

Western blotting

Experiments were conducted as described previously.⁴⁶ The following antibodies were used for protein detection: anti-ERK1/2 (#4695S), anti-Phospho-ERK1/2 (#4370S) and anti-H3K9me2 (#4658S) from CST, anti-cleaved caspase3 (#R013264) and anti-cleaved PARP (#R012387) from epizyme, anti-Bim (#A19702) and anti-β-tubulin (#AC012) from ABclonal, anti-H3 (#17168−1-AP) from Proteintech, and anti-G9a (#AFRM82243) from AiFang biological.

Real-time quantitative polymerase chain reaction (RT-qPCR)

Total RNA was isolated using the EZ-press RNA Purification Kit (EZBioscience, B0004D). Subsequently, cDNA was synthesized using Evo M-MLV RT Mix Kit with gDNA Clean for qPCR Ver.2 (Accurate Biotechnology, AG11728) according to the manufacturer’s instructions. RT-qPCR was performed in triplicate with SYBR Green Premix Pro Taq HS qPCR Tracking Kit (Accurate Biotechnology, AG11735) on a CFX96 PCR instrument (Bio-Rad, Hercules, CA, USA). Actin mRNA expression was used for normalization and relative quantification (RQ values) was employed to compare gene expression across groups. The primer sequences used for RT-qPCR were listed as follows:

Gene	Forward primer sequence (5′-3′)	Reverse primer sequence (5′-3′)
BIM	TTGATTCTTGCAGCCACCCT	GGAAGCTTGTGGCTCTGTCT
Actin	CCACGAGCGGTTCCGATG	GCCACAGGATTCCATACCCA

Quantification and statistical analysis

Statistical analyses were performed utilizing GraphPad and R software. Sample size (n) used for statistical analyses were provided in the relevant figures and supplementary figures. Spearman correlation was used for correlation analysis. Two-tailed unpaired Student’s t-test was used for comparisons between two groups, and one-way ANOVA was applied for multiple group statistical comparisons. Values of p < 0.05 were considered statistically significant, ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001.

Published: February 18, 2026