Abstract

   Bladder cancer (BC) is a heterogeneous disease with high recurrence
   rates and variable treatment responses. To address these clinical
   challenges, the world's first bladder cancer patient‐derived organoids
   (PDOs) biobank is established based on an Asian population. Thirty‐six
   BC‐PDOs are generated from 56 patients and demonstrated that the
   BC‐PDOs can replicate the histological and genomic features of parental
   tumors. Drug screening tests are conducted with a broad spectrum of
   conventional chemotherapeutic and targeted therapy drugs and identified
   differential drug sensitivities among the BC‐PDOs. These in vitro
   results are consistently supported by the PDO xenograft animal studies
   and patients’ clinical treatment outcomes, thereby verifying the
   predictive power of PDOs for drug responses in BC patients. By
   analyzing the genetic profiles of the PDOs, specific driver genes that
   correlate with drug sensitivity to two stand‐of‐care chemotherapeutics,
   cisplatin, and gemcitabine, are identified. Additionally, the
   practicality of PDOs in investigating the tumor microenvironment has
   been demonstrated. This study underscores the utility of PDOs in
   advancing the understanding of bladder cancer and the development of
   personalized therapeutic strategies. The BC‐PDOs biobank provides an
   ideal preclinical platform for supporting the development of
   personalized treatment strategies for BC patients. This study also
   provides insights into the potential mechanisms of drug resistance,
   paves the way for subsequent region‐specific research, and demonstrates
   the possibility of using PDO‐related models to direct future research
   in developing drugs targeting tumor microenvironments.

   Keywords: bladder cancer, drug screening, organoids, personalized
   medicine, tumor microenvironment
     __________________________________________________________________

   We have established the first Asian‐based bladder cancer PDOs biobank,
   demonstrating their accuracy in mirroring parental tumors’ properties.
   Drug screening reveals diverse sensitivities to chemotherapies and
   targeted therapies, validated by PDO xenografts and patient outcomes.
   Genetic profiling identifies key genes associated with response to
   standard chemotherapies. Co‐culture model highlights the potential of
   PDOs in exploring the tumor microenvironment.

   graphic file with name ADVS-12-2414558-g002.jpg

1. Introduction

   Bladder cancer includes two main disease conditions, non‐muscle
   invasive bladder cancer (NMIBC) and muscle‐invasive bladder cancer
   (MIBC). For patients with NMIBC, the standard treatment approaches
   include transurethral resection of bladder tumor (TURBT) operation,
   single‐dose intravesical chemotherapy, and Bacillus Calmette‐Guérin
   (BCG) treatment. However, the oncological outcome is not
   satisfactory.^[ [60]^1 ^] For patients with MIBC, radical cystectomy is
   the traditional treatment method, but it is associated with significant
   complications. Some recent genetic studies have highlighted substantial
   molecular differences between NMIBC and MIBC.^[ [61]^2 ^] In addition,
   these studies indicated that bladder cancer is a heterogeneous cancer
   with various subtypes and different genomic mutations.^[ [62]^3 ^]
   These analyses suggest a correlation between the genomic profile and
   the clinical behavior of bladder cancer. For example, patients with
   TP53 and ERBB2 (HER2) mutations have a higher risk of tumor recurrence
   and progression.^[ [63]^4 ^] Furthermore, significant disparities are
   observed among different racial groups, such as the variations in
   survival outcomes and genomic profiles.^[ [64]^5 ^] To comprehensively
   understand the biological implications of common mutations, conduct
   region‐specific research, and facilitate the development of precise,
   personalized treatment methods, it is imperative to establish
   dependable personalized models that closely resemble the original
   tumor.

   Recently, the scientific community has focused on using PDOs, a 3D in
   vitro model, as an experimental approach for developing precision
   medicine. The tumor cells were isolated from parental tissues and
   effectively cultivated in the form of PDOs, accurately reflecting their
   structure and heterogeneity while preserving genetic integrity in
   vitro.^[ [65]^6 ^] Although initial work has established the
   feasibility of culturing bladder cancer‐derived PDOs, a thorough
   description of their biological characteristics, including genomic
   mutations and transcriptional profiles, is yet to be fully
   elucidated.^[ [66]^7 ^] In addition, reports on the application of BC
   PDOs remain scarce, withlimited data on drug screening, in vivo
   experimentation, and the exploration of tumor microenvironment. To
   advance the field, there is a clear need for more extensive research to
   assess the potential of PDOs in personalized treatment.

   In this study, we have successfully established the world's first PDOs
   biobank for bladder cancer based on an Asian population, which
   effectively replicates the histological and genomic features of
   parental tumors. Utilizing these PDOs, we conducted a drug screening
   process to evaluate the sensitivity to various chemotherapeutic
   targeted drugs. The responses observed were subsequently validated by
   in vivo experiments. By analyzing the molecular profiles of PDOs and
   their parental tumors, we identified specific driver genes that
   correlate with the sensitivity to certain chemotherapy drugs.
   Furthermore, we innovatively used PDOs to investigate the interaction
   between tumor cells and tumor‐reactive T cells. This study underscores
   the importance of PDOs in enhancing our understanding of tumor biology,
   facilitating region‐specific research, improving clinical practice, and
   developing personalized therapeutic strategies. The establishment of
   PDO biobank provides a robust platform for future research.

2. Results

2.1. Establishment of a Living BC‐PDOs Biobank

   To establish a living BC PDOs biobank, we recruited 56 patients with
   bladder tumor and collected fresh tumor specimens to develop organoids
   (Figure [67]1A). Among these cases, 35 (62.5%) cases were NMIBC, and 15
   (26.8%) cases were MIBC, 2 cases were metastatic adenocarcinoma from
   colon, 2 cases were adnocarcinoma from prostate, and 2 cases were
   inverted papilloma, demonstrating that our biobank includes known
   population‐level variety of histology (Table [68]S1, Supporting
   Information) Patients’ clinical and pathological characteristics are
   described in Table [69]S1 (Supporting Information). According to
   previous publications, BC organoid was defined as a cellular cluster
   that can self‐organize and self‐renew.^[ [70]^7 , [71]^8 ^] Using the
   microscope, we found 36 (64%) cases have successfully formed organoid
   structures in Matrigel (Figure [72]1B), including 28 NMIBCs, 5 MIBCs, 1
   inverted papilloma, and 2 colon cancer metastases. Compared to the MIBC
   group (5 out of 15, 33.3%), the NMIBC group had a higher success rate
   (28 out of 35, 80%), possibly due to the tumor purity and neoadjuvant
   chemotherapy.^[ [73]^7c ^] From our experience, MIBC samples exhibit
   inferior quality compared to NMIBC, with a higher contamination of
   stroma or adipose tissue, which compromises tumor purity. This impurity
   affects the material exchange between tumor cells and culture
   environment, thereby negatively impacting the growth of tumor cells. In
   addition, some patients with MIBC would receive preoperative
   chemotherapy, which could reduce the viability of tumor cells.

Figure 1.

   Figure 1
   [74]Open in a new tab

   Establishment of BC‐PDOs biobank. A) Schematic of experimental design.
   Fresh tumor tissues were collected from patients with bladder cancer
   and processed as described in the methods. Red lines and arrows
   indicate experiments done on tumor tissues, while blue lines and arrows
   indicate experiments done with PDOs. (Created in BioRender. Ho, V.
   (2024) [75]https://BioRender.com/p29z766). B) Summary of PDO
   formations. (i) Summary of PDO formation among all collected samples.
   (ii) Tumor types and stages of the cases forming organoids. C) PDO
   morphology subtypes. (i) Representative brightfield images of
   morphology subtypes: Solid, Hollow, and mixed (Scale bars, 100 µm).
   (ii) Percentage of different subtypes across all PDOs. D)
   Representative micrographs on parental tumors and matched PDOs. H&E and
   IHC staining on parental tumors for indicated markers, alongside H&E
   and whole‐mount IF staining of matched PDOs for indicated markers (i)
   BCORG1 (ii) BCORG23 (iii) BCORG45. Scale bars: 50 µm, 100 µm.

2.2. PDOs Preserve Morphological and Histological Features of Parental Tumors

   To record the morphological features of BC PDOs, we analyzed
   bright‐field images of PDOs and classified them into three groups based
   on previous report^[ [76]^7c ^]: solid, hollow, or mixed
   (Figure [77]1C). Solid organoids appeared as cellular clusters without
   a central cavity. Hollow organoids displayed a luminal structure.
   Organoids with a mix of solid and hollow features or diverse
   morphologies were classified as mixed group. Among 36 successful cases,
   50% were solid, 47% were mixed, and only 3% were hollow. Representative
   bright field images and subtypes of PDOs are detailed in Figure [78]S1
   and Table [79]S2 (Supporting Information).

   To compare the histological features between PDOs and parental tumors,
   we performed IHC staining and IF staining. We evaluated the expression
   of basal markers (cytokeratin (Ck) 5, p63), luminal markers (Ck8,
   Ck20), a tumor initiating marker (CD44), a proliferating marker (Ki67),
   and urothelial carcinoma markers (GATA3, uroplakin II (UPKII)
   (Figure [80]1D; Figure [81]S2, Supporting Information). H&E staining
   showed good concordance in histological characteristics between PDOs
   and parental tumors. The positive results of GATA3 and UPKII staining
   confirmed that our PDOs originated from urothelial carcinoma. The
   expression of CD44 and Ki67 further validated the strong stemness and
   cell viability of our PDOs. In addition, the protein expression level
   of different markers were mostly consistent between PDOs and parental
   tumors (Table [82]1 ). However, some differences were noted. For
   example, the parental tumor of BCORG45 exhibited positive expression of
   basal markers CK5 and p63, while the PDOs were only positive for p63,
   suggesting that some cell groups may have been lost during the primary
   culture process. For the luminal or basal subtype, the phenotype of
   parental tumors characterized by the predominant expression of either
   luminal or basal markers was mostly observed in PDOs (Table [83]1).
   While most PDOs only expressed markers for one subtype, some cases
   still expressed both basal and luminal markers, indicating the
   existence of intra‐tumoral heterogeneity.

Table 1.

   Summary of marker expression for indicated markers on different
   organoid lines.
 Line    Marker expression in tissue Marker expression in organoid
   Phenotypic similarity
 BCORG1            luminal                      luminal                 Yes
 BCORG2             basal                        basal                  Yes
 BCORG3             mixed                        mixed                  Yes
 BCORG8             basal                        basal                  Yes
 BCORG9             mixed                        mixed                  Yes
 BCORG14            basal                        basal                  Yes
 BCORG16            basal                        mixed             Yes (partial)
 BCORG23            basal                        basal                  Yes
 BCORG38           luminal                      luminal                 Yes
 BCORG43            basal                        mixed             Yes (partial)
 BCORG45            mixed                        mixed                  Yes
   [84]Open in a new tab

2.3. PDOs Have Successfully Recapitulated Genomic and Transcriptional
Features of Parental Tumors

   To compare the genomic features between PDOs and matched parental
   tumors, we analyzed the tumor purity of PDOs and parental tumors. Most
   PDOs (12 out of 15) had higher purity than their corresponding parental
   tumors (Figure [85]2A). Using the methods reported by Martina et al,^[
   [86]^7c ^] we calculated a copy‐number‐based similarity score between
   PDOs and paired parental tumors (Figure [87]2B). All cases showed good
   similarity between matched PDOs and parental tumors (copy number
   variant similarity >50%). To further explore the genomic consistency
   between PDOs and parental tumors, we analysed the proportion of
   deleterious single nucleotide variants (SNVs) in each sample. On
   average, shared mutation accounted for 74.7% (±18.0%), whereas
   PDOs‐speicfic and parental tumor‐specific mutations account for 10.3%
   (±8.8%) and 14.9% (±13.7%), respectively (Figure [88]2C). By comparing
   the allelic fraction of shared and private SNVs, we found that shared
   SNVs had a higher value than specific SNVs in each sample
   (Figure [89]2D; Figure [90]S3, Supporting Information), indicating that
   most tumor cells shared mutations. Copy‐number and point mutations
   profiles of each case showed overall concordance between paired samples
   (Figure [91]2E; Figure [92]S4, Supporting Information).

Figure 2.

   Figure 2
   [93]Open in a new tab

   Genomic analysis and transcriptome analysis between 15 PDOs and
   parental tumors. A) Comparison of tumor purity between PDOs and
   parental tumors. B) Allele specific CNV similarity of matched tumors
   and PDOs. C) Proportion of shared and private deleterious SNVs between
   PDOs and parental tumors. CM: colon cancer metastasis. D) Corrected
   allelic fraction of all shared and private SNVs in PDOs and parental
   tumors (BCORG1 left, BCORG45 right). The count of SNVs within each
   classification is specified. Box plots indicate median, interquartile
   range, and percentiles. Two‐sided Wilcoxon test. E) Copy‐number and
   point mutations profiles. Comparison between PDOs and parental tumors
   for two representative samples (BCORG1 left, BCORG45 right). F) Genomic
   profiling confirms that PDOs have successfully recapitulate common
   mutational alterations from parental tumors. G) Analysis of (i)
   different expression genes and (ii) pathway enrichment between
   organoids and parental tumors. The colored bar illustrates
   log2‐transformed values. Different expression genes: p < 0.05, |log2
   (fold change)| > 1.

   Subsequently, a comprehensive summary of somatic mutations in the PDOs
   and parental tumors showed that PDOs successfully recapitulated
   prevalent genomic alterations observed in the matched parental tumors
   (Figure [94]2F). Among genes crucial for cell cycle regulation,^[
   [95]^9 ^] ELF3 had the highest mutation rate (11 out of 15 cases), and
   a previous paper also reported that ELF3 is one of the most frequently
   mutated genes in bladder cancer.^[ [96]^10 ^] The co‐deletion of
   CDKN2A/B occurred in 40% of cases, which is associated with tumor
   recurrence in certain cancer types.^[ [97]^11 ^] In addition, some
   commonly mutated genes in bladder cancer, such as RB1, ZFHX3, CCND1,
   E2F3, and MGA, also exhibited notable mutation frequencies in our
   sample. Among chromatin modifiers, we found that some genomic
   alterations occurred in both NMIBC and MIBC samples, such as ARID1A and
   KMT2D. Specifically, ARID1A amplifications occurred in 67% of cases,
   making it one of the most common mutations in patients with bladder
   cancer.^[ [98]^12 ^] Our samples also exhibited some common genomic
   mutations reported in bladder cancer,^[ [99]^13 ^] such as SLX4, MTAP,
   SPEN, and SDHC. These genes play essential roles in distinct
   tumor‐related pathways. According to previous reports, some mutations
   of PI3K‐AKT pathway genes are key drivers of dysregulated growth in
   NMIBC,^[ [100]^13 ^] especially PIK3CA and FGFR3 genes. In our cohort,
   missense mutations of PIK3CA and FGFR3 genes were identified in 45% and
   54% of NMIBC cases, respectively. Conversely, no point mutations of
   PIK3CA and FGFR3 genes were detected in MIBC cases.

   We compared gene expression profiles between parental tumors and PDOs
   to assess the maintenance of gene expression during culture process,
   especially oncogenic genes. Quantitative analysis showed a high degree
   of concordance between matched PDOs and parental tumors in terms of
   gene expression (Figure [101]S5, Supporting Information). Furthermore,
   the results revealed that most gene expression profiles were conserved
   in PDOs, except for immune and metabolic pathways (Figure [102]2G).
   Additionally, pathway enrichment analysis revealed significant
   upregulation of certain oncogenic pathways, including PI3K/AKT, Notch,
   p53, and mTOR signaling pathways in PDOs.^[ [103]^14 ^] These findings
   suggest that our PDOs possess pronounced tumor characteristics,
   indicating their potential to be valuable models for bladder cancer
   research (Figure [104]2G).

2.4. BC‐PDOs Show Different Sensitivity to Chemotherapeutics and Targeted
Drugs

   To assess drug response profiles of BC‐PDOs, we conducted drug
   screening to test their sensitivity to different chemotherapeutic and
   targeted drugs. We used the maximum blood concentration (Cmax) of each
   drug as the testing concentration (Table [105]S3, Supporting
   Information), as it better simulates the in vivo drug effects and
   allows us to test more drugs.^[ [106]^7c ^] PDOs were considered
   sensitive when the drug significantly reduced cell viability (z‐score ≤
   −1.5, adjusted p‐value < 0.05, and 1‐FC > 0.5). Among the seven tested
   cases, PDOs showed heterogeneous responses to chemotherapeutic or
   targeted drugs (Figure [107]3A). The effects of some drugs were also
   confirmed by organoid‐formation assay (Figure [108]S6, Supporting
   Information).

Figure 3.

   Figure 3
   [109]Open in a new tab

   Drug response to chemotherapy or targeted drugs observed in ex vivo
   bladder cancer organoids. A) The Heatmap shows the results of PDOs drug
   screening (n = 7). Mean z‐scores, normalized to the vehicle values,
   were obtained from cell viability assays after 72‐h drug exposure for
   each PDO, are calculated for a single experiment corresponding to each
   PDOs. Relevant clinical information and PDO morphology are annotated
   with distinct colors at the bottom of the heatmap. Statistical
   significance between treatments and vehicle was calculated by one‐way
   ANOVA test with Dunnet's multiple comparison, a response is indicated
   by *(z‐score ≤ −1.5, adjusted p‐value < 0.05 and 1‐FC > 0.5). B)
   Association between genomic mutations and sensitivity/resistance to
   cisplatin or gemcitabine.

   For chemotherapeutic drugs, some drugs showed no significant effect in
   tested PDOs, such as cabazitaxel and docetaxel (Figure [110]3A).
   However, doxorubicin and epirubicin were effective in all PDOs. In
   addition, drug response statuses observed in PDOs were validated in
   corresponding follow‐up patients. For instance, one cisplatin‐resistant
   case (BCORG1) came from a patient who received cisplatin + gemcitabine
   combination treatment and experienced tumor progression
   (Figure [111]5D). Notably, BCORG1 showed limited sensitivity to most
   chemotherapeutic drugs, except doxorubicin and epirubicin. In contrast,
   BCORG38 was responsive to most chemotherapeutic drugs, except docetaxel
   and cabazitaxel. As first‐line treatments, cisplatin and gemcitabine
   single treatment showed unsatisfactory results, with only 2 out of 7
   samples and 3 out of 7 samples showing sensitivity, respectively. Most
   targeted drugs did not significantly affect the tested PDOs. Despite
   certain PDOs having mutations in relevant genes, the efficacy of
   corresponding inhibitors remains poor. Specifically, FGFR inhibitor
   erdafitinib was ineffective, as also reported by Martina et al.^[
   [112]^7c ^] However, BCORG2, which did not respond well to most
   chemotherapeutic drugs, was significantly sensitive to EGFR/HER2
   inhibitors. We analyzed the drug response and genomic profiles to
   explore the association between genomic mutations and drug
   sensitivity(Figure [113]3B). According to our analysis, PDOs with
   deleterious mutations in PIK3CA gene would be more sensitive to
   gemcitabine than other PDOs. Conversely, PDOs with deleterious
   mutations in ELF3 gene would be more resistant to gemcitabine
   treatment. PDOs showing copy‐gain in ARID1A, SDHC, or SPEN genes would
   have higher resistance to cisplatin. These findings have illuminated
   the potential mechanisms involved in developing drug resistance in
   bladder cancer, thus paving the way for discovering new therapeutic
   methods.

Figure 5.

   Figure 5
   [114]Open in a new tab

   PDOs have the potential to predict the drug sensitivity of
   corresponding patients. A) Timeline schematic of patient BCORG1
   treatment stages.G/C = gemcitabine/cisplatin;Pemb = Pembrolizumab;T/C =
   docetaxel/cyclophosphamide; WBRT = whole brain radiotherapy. B) CT scan
   at different stages of patient BCORG1 treatment. Tumor lesions are
   marked with red arrows. C) Summary of drug sensitivity for 10 PDOs to
   combined therapy (gemcitabine + cisplatin, G/C). Results are displayed
   as dose‐response curves. The violin plot showed the AUC of each case (n
   = 10). D) Effects on the viability of treatment with G/C using an
   organoid formation assay. Scale bar: 100 µm. IC50 and AUC values were
   analysed by GraphPad Prism 9 software with nonlinear regression (curve
   fit) and the equation log(inhibitor) versus normalized response. Error
   bars represent ± SEM.

2.5. PDOs and PDO xenograft (PDOX) Are Useful Models to Predict the Drug
Response of Corresponding Patients

   To assess the tumorigenic ability of our PDOs, we used them to form
   ectopic xenografts in vivo. Five cases successfully formed tumor (Table
   [115]S2, Supporting Information). We conducted IHC staining analysis
   with a panel of markers (GATA3, Ck5, Ck8, Ck20, CD44, p63, UPKII, Ki67)
   on the xenografts to compare their characteristics with their
   corresponding PDOs and found that the immnunoexpression patterns of all
   PDOXs kept a high concordance with their corresponding PDOs (Figure
   [116]4A; Figure [117]S7, Supporting Information). However, we also
   found that the BCORG1 xenograft was positive to Ck5, which differed
   from its corresponding PDOs.

Figure 4.

   Figure 4
   [118]Open in a new tab

   Tumorigenesis in organoid‐derived xenografts and clinically relevant
   responses to chemotherapy of PDOs in vivo. A) H&E and IHC staining of
   representative xenograft generated from organoids. Scale bar: 50µm. B)
   Sensitivity profile of BCORG1 to different chemotherapeutic drugs. Raw
   data were obtained from cell viability assays after a 72‐h exposure of
   PDOs to drugs (n = 7). The violin plot showed the AUC of each case. C)
   Effects on the viability of treatment with cyclophosphamide (resistant)
   or doxorubicin (sensitive) using an organoid formation assay. Scale
   bar: 100 mm. D) Schematic of in vivo experiments. PDOs were
   subcutaneously transplanted into NSG mice. These mice would be treated
   with specific drugs or vehicles every 6 days once the tumor size
   reached 50–100 mm^3. E) Tumor growth curves in NSG mice. Mice were
   treated with doxorubicin, or vehicle every 6 days for 12 days (n = 5).
   Results are shown as the tumor volume (mean ± SD). F) Representative
   image of BCORG1 xenografts after 12 days of treatment (doxorubicin or
   vehicle, n = 5). G) Comparison of (i) tumor volume and (ii) weight
   between control and doxorubicin‐treated groups. The statistical
   differences between groups were analyzed using unpaired two‐tailed
   t‐test. IC50 and AUC values were analyzed by GraphPad Prism 9 software
   with nonlinear regression (curve fit) and the equation log(inhibitor)
   versus normalized response. Error bars represent ± SEM.

   Then, we used the PDOX mice model to validate the in vitro drug
   sensitivity results in PDOs. We chose doxorubicin as the testing drug
   based on the in vitro drug sensitivity results (Figure [119]4B,C).
   BCORG1 cells were injected subcutaneously into NSG mice. Once the
   tumors reached an appropriate size, the mice were injected
   intraperitoneally with either drugs or a control vehicle
   (Figure [120]4D). As anticipated, there was a significant difference in
   tumor growth was noted between the treatment and control groups in
   BCORG 1 (Figure [121]4E,F). tumor burdens, reflected by tumor weight
   and volume, were significantly reduced in the doxorubicin treatment
   group (Figure [122]4G). These results indicated that the drug screening
   findings in PDOs were replicated in PDOX mice.

   In addition, we compared the drug screening results with actual
   clinical outcomes. According to our findings, PDOs showed heterogeneous
   sensitivity to the combined therapy (gemcitabine + cisplatin, G/C),
   which is also observed in bladder cancer patients. Among these
   patients, patient BCORG1 experienced adjuvant chemotherapy (G/C) after
   surgery and developed colon metastasis, indicating clinical resistance
   to G/C treatment (Figure [123]5A,B). This outcome was consistent with
   the drug response in corresponding PDOs (Figure [124]5C,D). These
   results further demonstrate that PDOs could be a valuable tool for
   predicting drug response in BC patients.

2.6. Gene expression Profiles of BC PDOs in Relationship with Cisplatin or
Gemcitabine Response

   To investigate the underlying mechanism of drug resistance to cisplatin
   and gemcitabine, we did RNA sequencing of 12 PDOs and analyzed their
   gene expression profiles. Based on the dose‐response curves, the 12
   PDOs were stratified into sensitive or resistant groups according to
   their sensitivity to cisplatin (sensitive: normalized AUC <0.5,
   resistance: normalized AUC ≥ 0.5) (Figure [125]6A). By comparing
   the cisplatin‐sensitive group (n = 3) with cisplatin‐resistant group (n
   = 9), we found that genes associated with cell apoptosis or cisplatin
   sensitivity, such as SPINK1, KCNF1, PSCA, ALOX12, MYCN, were
   significantly up‐regulated in the sensitive group (Figure [126]6B).
   Conversely, genes associated with cisplatin resistance, such as NLRP2,
   ALDH1A1, CD47, and CLCA2, were down‐regulated in the sensitive group.
   Pathway enrichment analysis revealed that certain cell
   apoptosis‐related pathways, such as TRIF‐mediated programmed cell death
   and caspase‐mediated cleavage of cytoskeletal proteins were activated
   in cisplatin‐sensitive group (Figure [127]6C).

Figure 6.

   Figure 6
   [128]Open in a new tab

   Transcriptional profiling of different response cohorts to cisplatin or
   gemcitabine treatment. A) Ex vivo drug sensitivity of PDOs to cisplatin
   shown as dose‐response curves. The violin plot showed the AUC of each
   case. B) (i) Heatmap and (ii) volcano analysis showed the most
   differentially expressed genes in the cisplatin sensitive group (n = 3)
   compared to cisplatin‐resistant group (n = 9). Sensitive group
   (AUC < 0.5); Resistant group (AUC ≥ 0.5). Differentially expressed
   genes: pvalue < 0.05, log2FoldChange > 1. C) Gene set enrichment
   enrichment analysis of cisplatin‐sensitive group (AUC < 0.5) compared
   to cisplatin‐resistant group (AUC ≥ 0.5). The colored bar represents
   the ‐log10(p‐value). D) Ex vivo drug sensitivity of PDOs to gemcitabine
   in the form of dose‐response curves. The violin plot showed the AUC of
   each case. E) (i) Heatmap and (ii) volcano plot analysis showed the
   most differentially expressed genes in the gemcitabine sensitive group
   (n = 5) compared to gemcitabine‐resistant group (n = 7). Sensitive
   group (AUC < 0.5); Resistant group (AUC ≥ 0.5). Differentially
   expressed genes: p‐value < 0.05, log2FoldChange > 1. F) Gene set
   enrichment analysis of gemcitabine sensitive group (AUC < 0.5) compared
   to gemcitabine‐resistant group (AUC ≥ 0.5). The colored bar represents
   the ‐log10(p‐value). G) Differentially activated pathways between
   sensitive and resistant groups for cisplatin and gemcitabine. The
   pathway score was the ssGSEA score which was measured from
   Wikipathways. The odds ratio (OR) was the beta value from the linear
   model of (pathway score ∼ drug AUC). IC50 and AUC values were analyzed
   by GraphPad Prism 9 software with nonlinear regression (curve fit) and
   the equation log(inhibitor) versus normalized response. Error bars
   represent ± SEM.

   We also classified 12 PDOs into two groups based on their sensitivity
   to gemcitabine (sensitive: normalized AUC < 0.5, resistance: normalized
   AUC ≥ 0.5) (Figure [129]6D). By comparing the gemcitabine‐sensitive
   group (n = 5) with the resistant group (n = 7), we found that genes
   associated with cell apoptosis or gemcitabine sensitivity, such as CA2,
   EREG, PTGS1, TBXAS1, MMP1, were significantly upregulated in
   the sensitive group (Figure [130]6E). These genes were predominantly
   enriched in well‐known tumor‐related pathways, such as the EGFR
   signaling pathway and the PI3K/AKT signaling pathway (Figure [131]6F).
   Conversely, genes associated with chemotherapy resistance, such as
   ADRA2C, and TFF2, were significantly down‐regulated in the sensitive
   group (Figure [132]6E). Moreover, WikiPathways analysis indicated a
   strong connection between TP53 network and cisplatin resistance, the
   FABP4‐related pathway and gemcitabine resistance (Figure [133]6G).

2.7. BC‐PDOs Could be a Useful Platform to Investigate Tumor Microenvironment

   Tumor‐reactive T cells play an important role in tumor
   microenvironment. To generate tumor‐reactive T cells, we selected four
   PDOs with autologous PBMCs for co‐culture following the protocol by
   Krijn K et al.^[ [134]^15 ^] (Figure [135]7A). Before co‐culture, the
   organoids were treated with 200 ng mL^−1 IFN‐γ overnight. The treated
   PDOs were then used to stimulate autologous PBMCs. After two weeks of
   co‐culture, we isolated the T cells from the mixture using a human T
   cell isolation kit. Flow cytometry results showed that the activity
   (CD25+) of isolated T cells was greatly enhanced, and their reactivity
   (CD107a+) against autologous PDOs was also increased (Figure [136]7B),
   confirming we have successfully generated tumor‐reactive T cells using
   that PDOs co‐culture platform.

Figure 7.

   Figure 7
   [137]Open in a new tab

   Exploring interactions between tumor‐reactive T cells and tumor cells
   using the organoid model. A) Experimental workflow of generating
   autologous tumor‐reactive T cells. PDOs were used to stimulate
   autologous PBMCs. After two weeks of co‐culture, we successfully
   isolated the tumor‐reactive T cells from the mixture. (Created in
   BioRender. Ho, V. (2024) [138]https://BioRender.com/u12w586). B)
   Representative flow cytometry plots gated on CD8+ T cells tested for
   (i) activity and (ii) reactivity against autologous organoids after 2
   weeks of co‐culture with autologous PDOs. C,D) Cell surface expression
   level of MHC‐I in EZH2i‐treated organoid cells (EPZ‐011989 5 µm) were
   analyzed by (C) immunoblotting and (D) flow cytometry after 10 days of
   treatment. E) Organoids: autologous tumor‐reactive T cells co‐culture
   assay with EPZ‐011989 inhibition. For pharmacologic inhibition, PDOs
   were treated with EPZ‐011989(5 µm) or DMSO for 10 days and mixed with
   autologous tumor‐reactive T cells at an O: T ration of 1: 5. After 24 h
   of co‐culture, surviving tumor cells were counted by flow cytometry.
   The statistical differences between groups were analyzed using unpaired
   two‐tailed t‐test. Error bars represent ± SEM. *p < 0.05; **p < 0.01;
   ***p < 0.001.

   To validate the utility of our PDOs co‐culture model, we employed it to
   investigate the role of EZH2 in the microenvironment of bladder cancer.
   In the tumor microenvironment, the ability of CD8^+ T cells to
   recognize and kill tumor cells was influenced by the interaction
   between the MHC‐I complex and the T cell receptors (TCRs). According to
   previous publications, pharmacological inhibition of EZH2 could
   influence the cell surface expression of MHC‐I (HLA‐A/B/C).^[ [139]^16
   ^] However, existing studies have only utilized murine‐derived bladder
   tumors to investigate the relationship between EZH2 and MHC‐II.^[
   [140]^17 ^] To better reflect the relationship between EZH2 and MHC‐I
   in bladder cancer, we used PDOs co‐culture model to explore the role of
   EZH2 in bladder cancer. Based on the growing curve, we selected
   suitable working concentration which would not directly influence the
   viability of PDOs (Figure [141]S8, Supporting Information). After that,
   the results of immunoblotting, flow cytometry, and immunofluorescence
   confirmed that pharmacological inhibition of EZH2 restored cell surface
   MHC‐I (HLA‐B or HLA‐A/B/C) in PDOs (Figure [142]7C,D; Figure [143]S9,
   Supporting Information), indicating the occurrence of enhanced antigen
   presentation on the EZH2i‐treated PDOs. Furthermore, we co‐cultured
   treated PDOs with autologous tumor‐reactive T cells and tested the
   T‐cell cytotoxicity by flow cytometry, organoid formation assay, and
   ELISA assay. The results showed that pre‐treatment of EZH2i
   significantly enhanced the cytotoxicity of tumor‐reactive T cells
   (Figure [144]7E; Figure [145]S10, Supporting Information). In addition,
   the images under fluorescent field or bright field confirmed that
   pre‐treated EZH2 inhibition help reactive T cells to reduce organoid
   size and broken organoid structure (Figures [146]S11 and [147]S12,
   Supporting Information). These results proved that PDOs serve as a
   valuable model for investigating the interactions among various cells
   within the tumor microenvironment, offering insights that can aid in
   the discovering of novel treatment methods.

3. Discussion

   The treatment methods are different across various clinical stages of
   bladder cancer.^[ [148]^18 ^] However, the oncological outcomes for
   both NMIBC and MIBC remain unsatisfactory, with a high recurrence rate
   and various complications.^[ [149]^19 ^] In recent years, targeted
   therapy and immunotherapy, such as erdafitinib and pembrolizumab, have
   been incorporated into clinical practice. Despite this, many patients
   do not benefit from these treatment methods, even with relevant genomic
   mutations.^[ [150]^20 ^] These years, researchers have used genomic
   approaches to explore the inter‐patient heterogeneity of bladder
   cancer. Some studies have revealed the genomic difference between NMIBC
   and MIBC and categorized genomic subtypes of bladder cancer.^[ [151]^21
   ^] However, these findings need to explain the poor treatment outcome
   fully. To understand the biological function of common somatic
   mutations, and develop precise and personalized treatment methods, we
   must establish a pre‐clinical model that includes the known
   population‐level subtypes of bladder cancer.

   In this study, we successfully established a PDOs biobank encompassing
   a broad spectrum of bladder cancer subtypes. We explored the utility of
   PDOs in personalized treatment and immunotherapy for bladder cancer
   patients. Notably, 64% of cases formed organoid structure during the
   culture process, and most of them were derived from the NMIBC group. We
   demonstrated that PDOs can recapitulate the morphological and
   histological features of parental tumors, and our markers analysis also
   revealed intra‐tumoral heterogeneity. We have also compared the genomic
   profiles between matched PDOs and parental tumors. We found that the
   mutational profiles of organoids were highly consistent with not only
   the parental tumors, but also with the overall molecular signature of
   human bladder cancer. In addition, our transcriptional analysis
   indicated that most gene expression profiles were conserved in PDOs.
   However, WES and RNA‐seq found some dissimilar features between PDOs
   and parental tumors, potentially due to specific tumor subclonal
   events, limitations inherent in the culturing process, or technical
   sequencing constraints.^[ [152]^7 , [153]^22 ^] Overall, the ability of
   PDOs to mimic parental tumors validates their potential as a robust
   model for cancer research.

   To assess the drug response profiles of BC‐PDOs, we developed a drug
   screening system including a range of chemotherapeutic and targeted
   drugs. We found significant variability in drug responses across PDOs,
   highlighting the intertumor heterogeneity of bladder cancer. Consistent
   with earlier findings,^[ [154]^7c ^] targeted drugs did not show
   substantial antitumor effects, even for cases with relevant genomic
   mutations. Conversely, some chemotherapeutic drugs performed better
   with broader tumor suppression capability. Then, we used NSG mice to
   affirm PDOs’ tumorigenic ability and verify in vivo drug sensitivities,
   confirming the concordance between in vivo and in vitro drug response
   results. Furthermore, our assessments of drug sensitivity were
   reflected by the actual clinical outcomes, demonstrating the potential
   of PDOs in evaluating and predicting tumor responses to different
   drugs.

   Moreover, we integrated transcriptional profiles of PDOs with their
   drug response data to investigate the potential mechanisms behind
   varying responses to chemotherapy. This analysis revealed some
   upregulated genes linked to cellular apoptosis or drug transport in the
   cisplatin‐sensitive group, including SPINK1 and MYCN.^[ [155]^23 ^]
   Pathways associated with cell apoptosis were activated in this group,
   including TRIF‐mediated programmed cell death.^[ [156]^24 ^]
   Furthermore, WikiPathways analysis also highlighted a robust
   correlation between the TP53 pathway and cisplatin response. These
   findings elucidate the underlying mechanisms of drug resistance,
   guiding our subsequent research into tumor resistance.

   To explore the applicability of PDOs in immunotherapy, we established a
   PDO‐PBMC co‐culture system and generated autologous tumor‐reactive T
   cells.^[ [157]^7a ^] To test the practicality of our PDOs co‐culture
   model, we used it to investigate the influence of EZH2 inhibition, and
   we demonstrated that it could support autologous tumor‐reactive T cells
   in killing tumor cells. For this co‐culture system, it may have
   significant utility in two primary areas.^[ [158]^7a ^] First, it will
   help researchers to investigate the mechanisms that govern the
   sensitivity and resistance of bladder cancer to immunotherapy, just
   like what we have shown. Second, it provides the possibility to achieve
   patient‐specific immunotherapy. The capacity to derive BC‐PDOs from
   even the smallest tumor samples, such as those obtained through needle
   biopsies, coupled with the ability to proliferate circulating
   tumor‐reactive T cells from peripheral blood, offers a minimally
   invasive approach to assess the immunotherapy responsiveness of tumors
   in individual patients at different time points of treatment.

   In recent years, some studies have also reported the establishment of
   PDOs for bladder cancer.^[ [159]^7 ^] However, our study reported the
   world's first PDOs biobank for bladder cancer on an Asian population,
   and it is the most comprehensive to date and the first to investigate
   potential mechanisms of cisplatin and gemcitabine response using BC
   PDOs. According to previous studies, there exists several differences
   between Asian and Caucasian patient cohorts. Notably, both African
   American and Caucasian groups have been reported to exhibit lower
   survival probabilities compared to their Asian counterparts.^[ [160]^5a
   ^] Furthermore, considering the variations in environmental factors and
   clinical practice among different regions, the pursuit of truly
   personalized medicine necessitates the development of PDO biobanks that
   reflect the genetic and clinical profiles of local patient populations.
   Several studies have also reported genomic difference between Asian and
   Caucasian patients.^[ [161]^5b,c ^] For instance, the well‐known driver
   gene TP53 shows a lower mutation rate in the Asian group, which is also
   found in our PDOs biobank. Furthermore, our biobank showed a higher
   success rate in the NMIBC group.^[ [162]^7c ^] As we known, the high
   recurrence is a significant challenge in bladder cancer, with many
   patients experiencing frequent relapses after TURBT, ultimately leading
   to the need for radical cystectomy. The characteristics of our PDO
   biobank offers hope for rapid postoperative drug screening and genetic
   testing, which may pave the way for personalized and bladder‐preserving
   treatment. It may have chance to reduce recurrence rate and enhance the
   quality of life for patients in the future. Additionally, we are the
   first to use PDOs to explore the tumor microenvironment of bladder
   cancer. Our PDOs have some common characteristics with PDOs from other
   ethnicities, such as the observation that targeted therapies are less
   effective than traditional chemotherapeutic drugs. At the genomic
   level, both sets of PDOs harbor common mutations associated with
   bladder cancer, such as PIK3CA, FGFR3, and RB1. This further
   demonstrates the strengths of the PDO model. Above all, our study
   indicates that PDOs could be a valuable tool for region‐specific
   research, clinical decision support, and improving patient prognoses.
   PDOs can contribute significantly to preclinical studies, particularly
   in disease modeling and therapeutic compound identification. Compared
   to previous studies, we utilized the drug‐response curve and AUC to
   explore the reasons for different sensitivity to cisplatin or
   gemcitabine among various PDOs. However, due to the limited number of
   tested PDOs, further mechanistic studies and validation with a larger
   cohort of PDOs are necessary. Based on our analysis, we observed a
   higher success rate in PDO establishment among patients who received
   TURBT. Further investigation is warranted to explain the underlying
   mechanisms responsible for this observation. Furthermore, a
   comprehensive prospective study involving a large cohort of bladder
   cancer PDOs need to be considered to assess their clinical relevance
   and inform their potential integration into personalized cancer care
   strategies. Moreover, as the first PDOs biobank for bladder cancer on
   an Asian population, our analysis of the difference between Asian and
   Caucasian populations’ PDOs is still insufficient. Future research
   should further investigate the disparities among PDOs derived from
   different populations.

   A pressing challenge in bladder cancer management is the development of
   precise methods to predict treatment efficacy for individual patient.
   Our team have established the world's first PDOs biobank for bladder
   cancer based on an Asian population, which paves the way for subsequent
   research on bladder cancer treatments tailored to Asian populations.
   This biobank leverages the potential of PDOs to model patient‐specific
   tumor biology and therapeutic responses, offering a valuable resource
   for personalized medicine and precision oncology in the context of
   bladder cancer. For future studies, we need to establish a larger
   cohort of PDOs to refine and substantiate our findings. This expanded
   collection will enable more comprehensive analysis and bolster the
   reliability of our results.

4. Experimental Section

Sample Collection

   The study was approved by the institutional ethical committee (Joint
   Chinese University of Hong Kong‐New Territories East Cluster Clinical
   Research Ethics Committee, No. 2016.396). Informed consent was taken
   from the patients. Parental tumor tissue was obtained through TURBT or
   radical cystectomy procedures. Part of the parental tumor tissues were
   saved to establish PDOs, and a cold basal medium supplemented with 10%
   fetal bovine serum (FBS, Gibco) and 10 µm ROCK inhibitor was used as
   the transport medium. The remaining tissue samples were examined by a
   dedicated uropathologsts.

Tissue Dissociation and Organoid Culture

   The procedure of organoid primary culture followed the protocol
   described by Jasper et al.^[ [163]^22 ^] Some modifications were made
   to improve the overall viability of the tumor cells. After washing, the
   tissues were finely minced and digested at 37 °C for 30–60 min. Then
   the digestion process was stopped using Adv DMEM/F12 containing 10%
   FBS. Then, the tissue mixture was further digested by TrypLE Express
   supplemented with ROCK inhibitor at 37 °C for 5 min. Afterward, the
   mixture was passed through a 70 µm cell strainer (Corning). After that,
   the cell pellet was mixed with the Matrigel (Corning) and seeded into
   the well. Once the Matrigel solidified, organoid culture medium was
   added. The culture medium was changed twice weekly, and organoids were
   passaged every 10 days.

Immunohistochemistry Staining, Immunofluorescence Staining, and
Immunoblotting

   Both organoids and tissues were fixed in 4% paraformaldehyde for at
   least 1 h, dehydrated, and paraffin‐embedded using standard histology
   procedures. The sections were stained with Hematoxylin and Eosin (H&E).
   Immunohistochemistry (IHC) staining and immunofluorescence (IF)
   staining were performed with primary antibodies, including Anti‐CK5
   (1:250, Abcam), Anti‐CK20 (1:250, Abcam), Anti‐Ki67 (1:250, Abcam),
   Anti‐GATA3 (1:250, Abcam), Anti‐UPKII (1:250, Abcam), Anti‐CK8 (1:250,
   Abcam), Anti‐p63 (1:250, Abcam), Anti‐CD44 (1:250, Abcam). For
   immunoblotting, total protein was extracted from organoids using RIPA
   buffer (supplemented with phosphatase inhibitor and protease
   inhibitor). The primary antibodies include Anti‐GAPDH (1:2000, CST),
   Anti‐HLA‐I (1:1000, Abcam).

Drug Screening and Cell Viability Assay

   Organoids at passages P3‐P10 from available BC PDOs were used for drug
   screening. They were dissociated into single cells or small clusters
   using TrypLE Express, filtered through a 70µm cell strainer to
   eliminate large organoids. Cells were then seeded in at least 3
   replicates for each drug condition, with positive (Staurosporine) and
   negative (DMSO or H[2]O) controls in a ULA 384‐well plate (Corning) in
   organoid culture medium containing 2% Matrigel (10 000 orgnaoids
   mL^−1). Each well would be seeded 1500–3000 organoids. After 48 h of
   culture, each drug was diluted in organoid culture medium and added to
   each well at a final concentration of 1xCmax reported in previous
   publications. The final concentration of DMSO or H[2]O solvent in the
   treatment condition was 0.5%. After 72 h of drug treatment, cell
   viability was measured using the CellTiter‐Glo 3D assay (Promega)
   according to the manufacture's instructions. Raw values were
   individually standardized for each specimen using the equation: (X[s] –
   X[v])/SD[v], where X[s] represents the technical replicates for each
   drug; X[v] denotes the mean of technical replicates from corresponding
   vehicles and SD[v] is the standard deviation of the technical
   replicates from matching vehicles. The Z‐factor was used as a
   statistical indicator of drug screening assay quality and calculated as
   1‐ [3x(SD of sample + SD of control)/(Mean of sample – Mean of
   control)]. The Z‐scores for cisplatin, carboplatin, and 5‐FU were
   calculated using H[2]O as the negative control, while all other drugs
   used DMSO as the negative control. Additionally, raw values were used
   to calculate fold‐changes relative to the mean of raw values from the
   vehicles.

   To evaluate chemotherapy drug responses, organoids were digested and
   seeded in ULA 384‐well plate (Corning) at a density of 2500 cells per
   well in 2% Matrigel/culture medium. After 48 h, fresh organoid culture
   medium with varying drug concentrations was added. The maximal vehicle
   concentration used was <5%. After 72 h, cell viability was assessed
   using the CellTiter‐Glo 3D assay (Promega). IC50 and AUC values were
   analyzed by GraphPad Prism 9 software with nonlinear regression (curve
   fit) and the equation log(inhibitor) versus normalized response.^[
   [164]^25 ^] Each concentration was tested in at least 3 replicates.

Whole‐Exome Sequencing and Analysis

   DNA was extracted from the buffy coat, tumor tissue, and organoid
   samples using either the QIAamp DNA/RNA mini kit (Qiagen) or QIAamp DNA
   blood mini kit (Qiagen) as per the manufacturer's instructions. The
   extracted DNA was then quantified using Qubit dsDNA HS Assay Kits
   (Invitrogen) and a Qubit 4 Fluorometer. The Whole‐exome sequencing
   libraries were prepared using the SureSelect Human All Exome V6 (58M)
   (Agilent Technologies) and VAHTS Universal Pro DNA Library Prep Kit for
   illumina V2 (Vazyme). The libraries were sequenced on an Illumina
   platform using the NovaSeq 6000 instrument at a read length of 2 × 150
   bases.

DNA Sequencing Analysis

   The raw DNA sequencing reads were quality checked by FastQC (v0.11.4)
   and processed to filter the low‐quality reads using FastP (v0.23.4).^[
   [165]^26 ^] Then, clean reads were mapped against the human genome
   assembly (GRCh37/hg19) by BWA mem algorithm (v0.7.17)^[ [166]^27 ^]
   with default settings. The bam files were sorted by samtools (v1.19),^[
   [167]^28 ^] marked the duplicates by picard tools (v1.14) and did the
   indel realignment and base recalibration by the Genome Analysis
   Toolkits (GATK v3.4).^[ [168]^29 ^] Somatic SNVs and indels were called
   by Mutect (v1.1.7),^[ [169]^30 ^] Platypus (v1.0),^[ [170]^31 ^]
   Strelka (v1.0)^[ [171]^32 ^] and Scalpel (v0.5.2).^[ [172]^33 ^]

   The confident mutations were retained by the following criteria: 1) not
   detected in normal sample panel and dbSNP135;^[ [173]^34 ^] 2) variant
   allele frequency more than 0.05.

   Allele specific copy number variations (CNVs) assessment based on
   segmentation by CNVkit (v0.9.10) and CLONET v2 (v2.2.1). The CNV
   similarity of PDOs (o) and parental tumors (t) were calculated by:
    1. d = ||(cnAt, cnBt) – (cnAo, cnBo)|| ((cnA, cnB) being the major and
       minor alleles of a sample for each gene;d is the Euclidean distance
       of PDOs and parental tumors)
    2. d’ = (d‐min(d))/(max(d)‐min(d)) (d’ is the normalized distance);
    3. s = 1 – d’ (s is the similarity)

   The final mutations were annotated by SnpEff,^[ [174]^35 ^] SnpSift
   (v3.6a)^[ [175]^36 ^] and VEP (v105).^[ [176]^37 ^]

   Tumor purity of PDOs and parental tumors were measured by the TPES
   (v1.0.0).^[ [177]^38 ^]

   The association between genomic mutations and drugs activity was tested
   by random intercept Liner Mixed Model (LMM): z ∼ β[1]x + β[0]m, where z
   is vector of z‐scores for a drug across all replicates of samples, x is
   the vector which encoding the mutation status (0 for wild type and 1
   for mutated), m is a vector that including the sample info for each
   replicate. The effect size was obtained from the estimated β[1] of
   association result and p‐values was measured by the Likelihood Ratio
   Test that comparing the above model to the null model. The LMM analysis
   was implemented by R package lme4 (v1.1.35.5).

RNA‐Sequencing and Analysis

   Total RNA was extracted from fresh tumor tissue, FFPE samples, and
   organoids using QIAamp DNA/RNA mini kit (Qiagen) or HiPure FFPE Plus
   kit (Magen) as per the manufacturer's instructions. The extracted RNA
   was then quantified using Qubit RNA HS Assay Kits (Invitrogen) and a
   Qubit 4 Fluorometer. The RNA was first subjected to ribosomal RNA
   depletion using Ribo‐off rRNA depletion kit (Human/Mouse/Rat) (Vazyme),
   followed by library preparation by VAHTS Universal V8 RNA‐seq Library
   Prep kit for Illumina (Vayzme). The sequencing was performed using
   Novaseq 6000 instrument (Illumina) at a read length of 2 × 150 bases.

   The raw RNA reads were quality checked by FastQC (v0.11.4) and
   processed to filter the low‐quality reads using FastP (v0.23.4).^[
   [178]^27 ^] The remaining reads were aligned to human reference genome
   GRCh38/hg38 using STAR (v.2.7.11a).^[ [179]^39 ^] Gene expression level
   were quantified using RSEM (v1.3.3).^[ [180]^40 ^] The spearman
   correlation coefficient and R‐squared were used to assess the
   concordance of PDOs and parental tumors at the gene expression level.
   Then, the gene counts matrix was input into DEseq2 (v1.40.2)^[ [181]^41
   ^] for differential expression analysis. The GSVA scores were
   calculated by the GSVA package (v1.46.0)^[ [182]^42 ^] with the method
   set to “gsva”. Gene set enrichment analysis was implemented in
   clusterProfiler (v4.6.2).^[ [183]^43 ^] The odds ratio (OR) was from
   beta value of linear model: pathway score ∼ drug AUC, which uses lm()
   function from the R base package. The pathway score was the ssGSEA
   score which calculated by GSVA (v1.46.0) with the ssGSEA method and
   pathways were sourced from the WikiPathways database.^[ [184]^44 ^]

PDO Xenograft Models and Chemotherapy in NSG Mice

   All mouse experiments were conducted following the guidelines approved
   by the institutional animal ethics committee (Animal Experimentation
   Ethics Committee in The Chinese University of Hong Kong, No.
   22‐018‐MIS). For PDOX, each 6‐8‐week‐old male NOD.Cg‐Prkdc^scid
   Il2rg^tm1Wjl/SzJ (NSG) mice were subcutaneously injected with 2 × 10^6
   BC PDOs mixed with 75ul Matrigel and 25ul Adv DMEM/F12. tumor
   dimensions were measured every third day, and tumor volume was
   calculated as per previous studies.^[ [185]^8 ^] Once the tumor reached
   50 mm^3, mice were randomly assigned to treatment (doxorubicin) or
   control groups (physiological saline). Doxorubicin (4mg kg^−1) or
   saline was administered intraperitoneally every 6 days. After 12 days,
   mice were sacrificed, and fresh tumors were harvested for subsequent
   experiments.

Generation of Tumor‐Reactive T Cells

   The patient's blood was obtained from the patients after obtaining
   informed consent. After blood collection, peripheral blood mononuclear
   cells (PBMCs) were isolated using Fico‐Paque, following the protocols
   by Bain B et al.^[ [186]^45 ^] The isolated PBMCs were mixed with
   cryogenic medium (10% DMSO + 90% FBS) and transferred into a cryovial.
   The tubes were placed in Mr. Frosty tube racks and immediately stored
   in a −80 °C freezer overnight.

   To generate the tumor‐reactive T cells, the protocol by Krijin K
   et al., was followed.^[ [187]^15 ^] to establish a co‐culture system.
   The day before co‐culture, the culture medium for PBMCs was prepared,
   consisting of RPMI 1640 (GIBCO), 1:100 PSN, and 10% FBS (GIBCO).
   Cryopreserved PBMCs were defrosted in 37 °C T cells culture medium and
   treated with 25 U mL^−1 benzonase (Merck) to prevent cells from
   clumping. After that, the PBMCs were resuspended in the culture medium
   with 150 U mL^−1 interleukin‐2 at 2–3*10^6 cells mL^−1, and incubated
   at 37 °C. Meanwhile, 200 ng mL^−1 human recombinant interferon‐gamma
   (IFN‐γ) (Peprotech) was used to stimulate the PDOs, and used 5 mg mL^−1
   anti‐CD28 (clone CD28.2, eBioscience) to coat the 96‐well U‐bottom
   plates. The next day, stimulated organoids were digested and
   resuspended in T cell culture medium. PBMCs, prepared the previous day,
   were collected, counted, and mixed with organoids at a 20: 1
   PBMCs‐to‐organoids ratio. After washing the anti‐CD28‐coated U‐bottom
   plates twice with PBS, the mixture was seeded at a density of 10^5
   PBMCs/well. To enhance organoid stimulation of T cells, the co‐culture
   system was supplemented with 150 U mL^−1 IL‐2 and 20 mg mL^−1
   anti‐PD‐1‐blocking antibody (kindly donated by Invitrocue, HK). The
   culture medium (supplemented with IL‐2 and anti‐PD‐1) was renewed 2–3
   times weekly. After 7 days, PBMCs were collected, counted, and
   co‐cultured with fresh organoids. The entire co‐culture process lasted
   14 days, after which the mixture would be collected and washed with
   PBS. For organoid killing assay, T cells were isolated from co‐culture
   PBMC using the human T cell isolation kit (EasySep) and cultured in T
   cells culture medium supplemented with IL‐2.

Flow Cytometry

   To evaluate the expression levels of HLA‐I and PD‐L1 on PDOs, TrypLE
   Express was used to dissociate organoids following the digestion
   protocol. Before collecting organoid cells, they were pre‐incubated
   with specific concentrations of IFN‐γ or an EZH2 inhibitor. After
   digesting the organoids into single cells, it was washed twice with
   FACS buffer (PBS + 1% Bovine serum albumin) and stained them with
   anti‐human HLA‐A, B, C‐PE (eBioscience), anti‐human HLA‐B‐PE (Miltenyi
   Biotec) or anti‐CD274‐APC (eBioscience) antibodies for 30 min on ice.
   The mixture was then washed and stained with Fixable Viability Dye to
   exclude dead cells. Samples were analyzed using a BD LSRFortessa cell
   analyzer.

   To evaluate tumor reactivity, 5 × 10^4 single organoids were mixed with
   co‐cultured PBMCs at a 1:2 target‐to‐ effector ratio and seeded in an
   anti‐CD28‐coated U‐bottom plate with 20 mg mL^−1 anti‐PD‐L1 and
   anti‐human CD107a‐PE antibodies (BD). Organoids were pre‐treated with
   IFN‐γ for 24 h. After co‐culturing for 5 h, cells were collected and
   washed twice in FACS buffer, and then blocked with an Fc receptor
   binding inhibitor for 20 min on ice. Cells were then stained with the
   following antibodies: anti‐CD3‐Pacific blue (Invitrogen), anti‐CD4‐FITC
   (BD), anti‐CD8‐APC (BD), and anti‐CD25‐Alexa Fluor 700 (Invitrogen).
   The mixture was subsequently stained with Fixable Viability Dye to
   exclude dead cells and analyzed using a BD LSRFortessa cell analyzer.

Organoid Killing Assay

   For the organoid killing assay, organoids were pre‐treated with EZH2i
   EPZ‐011989 (5µm) for 10 days and dissociated into single cells. The
   organoids and autologous tumor‐reactive T cells were mixed at 1:5
   organoids‐to‐ T cells ratio and seeded in an anti‐CD28‐coated U‐bottom
   plate. After 24 h, cells were collected and stained with
   anti‐CD3‐Pacific blue (Invitrogen) and Fixable Viability Dye. The
   surviving tumor cells were counted using BD LSRFortessa cell analyzer.
   The media of the co‐culture system was collected to test for IFN‐γ
   levels using the human IFN‐γ ELISA kit (Invitrogen).

Statistical Analysis

   Sample size (n) for each statistical analysis is provided in the
   relevant figure legends. The pre‐processing and analysis of sequence
   data were described in the Experimental Section. IC50 and AUC values
   were analyzed by GraphPad Prism 9 software with nonlinear regression
   (curve fit) and the equation log(inhibitor) versus normalized
   response.^[ [188]^25 ^] Each concentration was tested in at least 3
   replicates. In Figure [189]3, statistical significance between
   treatments and vehicle was calculated by one‐way ANOVA test with
   Dunnet's multiple comparison, a response is indicated by *(z‐score ≤
   −1.5, adjusted p‐value < 0.05 and 1‐FC > 0.5). Experiments in
   Figure [190]7B–E were performed in triplicates. In this study, data
   were plotted as mean ± standard deviation, with the student's t‐test
   and one‐way ANOVA were applied to compare two groups and multiple
   groups, respectively. Statistical analyses were conducted by GraphPad
   Prism 9 software. *p < 0.05; **p < 0.01; ***p < 0.001; ns, not
   significant.

Conflict of Interest

   The authors declare no conflict of interest.

Author Contributions

   Z.H. conducted experiments and wrote the manuscript. L.N. conducted
   bioinformatic analysis. H.V.W.S., L.K., C.X., W.H., L.H., Z.D., K.L.W.,
   and H.F.C. helped to do primary culture and mice experiments.
   P.K.‐F.C., I.C.‐H.K., C.H.‐M.W., D.K.‐W.L., and S.K.‐K.Y. prepared the
   clinical samples. H.V.W.S. and W.D. helped write the manuscript. D.X,
   C.FN, J. Y.‐C.T. supervised the process of the study and revised the
   manuscript.

Supporting information

   Supporting Information
   [191]ADVS-12-2414558-s001.docx^ (42.5MB, docx)

Acknowledgements