Abstract

   Pancreatic ductal adenocarcinoma (PDAC) has a poor prognosis. Although
   chemotherapy has become increasingly important in recent years, there
   are no practical markers to predict therapeutic efficacy. Here, we have
   aimed to identify novel markers that predict resistance to chemotherapy
   drugs using patient-derived organoids (PDOs) of PDAC. PDOs were
   established using endoscopic ultrasound - guided fine needle aspiration
   (EUS-FNA) specimens. Drug sensitivity tests were performed on 15 PDOs
   and the correlation between drug sensitivity and transcriptome analysis
   were evaluated. BARD1 and RAD50 have been identified as genes
   associated with gemcitabine resistance. Additionally, SLC25A10 and
   MAP3K9 were identified as genes associated with
   gemcitabine + paclitaxel resistance.PDOs derived from EUS-FNA specimens
   can be used to assess individual drug resistance and to identify
   predictive factors for chemotherapy resistance.

Supplementary Information

   The online version contains supplementary material available at
   10.1038/s41598-025-09395-z.

   Keywords: Pancreatic cancer, Organoids, Drug screening, Biomarkers

   Subject terms: Cancer, Molecular biology, Gastroenterology, Oncology

Introduction

   Pancreatic ductal adenocarcinoma (PDAC) has a poor prognosis and it is
   the third leading cause of cancer-related deaths　with a five-year
   survival rate of only 12%^[50]1. Although it is difficult to improve
   outcomes with surgery alone,　neoadjuvant chemotherapy (NAC) has been
   proposed as a new treatment method for PDAC and is now recognized as
   the preferred approach for borderline resectable (BR) PDAC according to
   the 2022 NCCN guidelines^[51]2. Recent cohort studies have indicated
   the effectiveness of NAC in both resectable and BR PDAC^[52]3–[53]5.
   Thus, NAC is becoming increasingly important for improving the
   prognosis of PDAC; however, current treatment regimens are still
   insufficient, and there are no practical markers to predict the
   response to chemotherapy^[54]6. Therefore, there is a clear need to
   develop more effective pharmacological therapies and discover useful
   markers for predicting drug sensitivity.

   “Organoids” are artificially created tissues that resemble
   organs^[55]7. Leading experts in the organoid research field define the
   term organoid as follows: an organoid is a three-dimensional structure
   derived from (pluripotent) stem cells, progenitors, and/or
   differentiated cells that self-organize through cell-cell and
   cell-matrix interactions to recapitulate aspects of the native tissue
   architecture and function in vitro^[56]8. In 2015, the　feasibility of
   culturing tumorous lesions　was demonstrated as human and mouse
   pancreatic cancer organoids. These maintained their histological
   properties and they have already been used in research^[57]9.

   Endoscopic ultrasound-guided fine needle aspiration (EUS-FNA) has used
   for the histological diagnosis of PDAC. Organoids derived from EUS-FNA
   specimens enable the genetic analysis of PDAC^[58]10. In recent years,
   it has been reported that drug sensitivity can be evaluated by
   administering chemotherapeutic drugs to PDAC organoids derived from
   EUS-FNA samples^[59]11and their application in personalized medicine is
   expected.

   This study has aimed to evaluate the availability of patient-derived
   organoids (PDOs) in drug sensitivity tests and identify markers
   associated with drug resistance. We generated PDOs from EUS-FNA
   specimens of patients with PDAC and performed drug sensitivity tests by
   administering chemotherapeutic drugs to these PDOs. Based on the drug
   sensitivity results, we performed transcriptomic analysis to evaluate
   the expression of genes associated with gemcitabine (GEM) or
   gemcitabine + paclitaxel (GEM + PTX) resistance and explored predictive
   markers for GEM or GEM + PTX resistance.

Results

Patient characteristics

   We attempted to create PDOs from the EUS-FNA specimens with a confirmed
   diagnosis of adenocarcinoma in 47 cases; their characteristics are
   shown in Table [60]1. PDOs were successfully established in 32
   cases　and grown to confluence in 17 cases. These 17 cases were
   classified as the well growth group and 30 cases that failed to grow to
   confluence were classified as the poor growth group; the clinical
   characteristics between them were compared, but no clear significant
   difference was observed (P value < 0.05). The group with good growth
   was subjected to a drug sensitivity test; however, 2 cases were
   excluded from the study. One case had poor growth after thawing and the
   other had a bacterial infection. Drug sensitivity tests were performed
   in 15 cases, and their characteristics are shown in Table [61]2.

Table 1.

   Background on all cases attempted establishment of organoids.
   Variables                 Overall(n = 47)
   Age, yr                   69 (49–79)
   Famale sex                20 (43)
   BMI*, kg/m^2              22 (16–30)
   TNM staging by UICC (8th)
   I                         7 (15)
   II                        8 (17)
   III                       7 (15)
   IV                        25 (53)
   Location
   Head                      14 (30)
   Body                      19 (40)
   Tail                      14 (30)
   Tumor size, mm            33 (12–80)
   EUS-FNA
   Punctures, count          3.7 (3–5)
   Cytodiagnosis
   Positive                  29 (62)
   Suspicious                13 (28)
   Negative                  5 (11)
   Histological examination
   Positive                  31 (66)
   Suspicious                6 (13)
   Negative                  10 (21)
   Organoid culture
   Established               32 (68)
   Confluent                 17 (36)
   [62]Open in a new tab

   Data were presented as number (percentage) of patients unless otherwise
   indicated. Abbreviations: EUS-FNA, endoscopic ultrasonic fine needle
   aspiration. *BMI is the weight in kilograms divided by the square of
   the height in meters.

Table 2.

   Background of 15 organoid cases with drug sensitivity testing.
   Variables                 Overall(n = 15)
   Age, yr                   70 (50–79)
   Female sex                8 (53)
   BMI*, kg/m^2              21 (17–30)
   TNM staging by UICC (8th)
   I                         2 (13)
   II                        2 (13)
   III                       3 (20)
   IV                        8 (53)
   Location
   Head                      4 (27)
   Body                      7 (47)
   Tail                      4 (27)
   Tumor size, mm            33 (12–80)
   EUS-FNA
   Punctures, count          3.7 (3–5)
   Cytodiagnosis
   Positive                  10 (67)
   Suspicious                3 (20)
   Negative                  2 (13)
   Histological examination
   Positive                  10 (67)
   Suspicious                3 (20)
   Negative                  2 (13)
   [63]Open in a new tab

   Data were presented as number (percentage) of patients unless otherwise
   indicated. Abbreviations: EUS-FNA, Endoscopic ultrasonic fine needle
   aspiration. * BMI is the weight in kilograms divided by the square of
   the height in meters.

PDOs show pathological and genetic homology with donor tumor

   Pathological evaluation of the PDOs (H8 and H9) was performed. HE
   staining of H8 and H9 cases revealed tumor cells with large nuclei and
   irregular ductal structures, which was consistent with the histology of
   the EUS-FNA specimen (Fig. [64]1A, B).

Fig. 1.

   Fig. 1
   [65]Open in a new tab

   PDOs show pathological and genetic homology with donor tumor. A, B. HE
   stains image of the H8 (A) and H9 (B) case. In both cases, tumor cells
   with large nuclei and irregular ductal structures were observed, the
   findings were consistent with the HE staining image of the EUS-FNA
   specimen. Scale bars are 500 μm for weakly magnified A (top) and B
   (top), and 100 μm for strongly magnified images A (middle) and B
   (middle). C. Mutation analysis was performed using next-generation
   sequencing of approximately 400 genes that were evaluated using CCP of
   15 PDOs. The vertical axis represents the mutated gene name, the
   horizontal axis represents the number of organoid samples; missense
   mutations are shown in green, and nonsense mutations are shown in red.
   Tumor mutation burden (TMB) is shown in the upper part of the graph.

   A comprehensive cancer panel (CCP) analysis showed that all of the 15
   PDOs harbored KRAS mutations. Other mutations were observed in TP53
   (47%), TNFRSF14 (20%), RET (13%), TSC2 (13%), SF3B1 (7%), and SMAD4
   (7%) (Fig. [66]1C). In pancreatic cancer, KRAS and TP53 mutations are
   representative driver mutations^[67]13–[68]15and it has been concluded
   that these PDOs show pathological and genetic homology with pancreatic
   cancer.

Evaluation of drug sensitivity test for PDOs

   Drug sensitivity tests were performed on 15 PDOs derived from EUS-FNA
   specimens to evaluate their response to three chemotherapeutic drugs:
   GEM, PTX, and GEM + PTX (Fig. [69]2A-C). Among the 15 PDOs, there were
   strains with various passage numbers, and the cryopreserved PDOs were
   stably recultured after thawing. The dose-response curve of the PDOs
   differed among the organoid strains, indicating individual differences
   in drug sensitivity. In drug sensitivity test for GEM and GEM + PTX,
   two PDO strains, H32 and H17, maintained approximately 50% cell
   viability even at drug administration with high dosage. This indicates
   that these two PDOs were resistant to GEM and GEM + PTX.

Fig. 2.

   [70]Fig. 2
   [71]Open in a new tab

   Drug sensitivity test results for PDOs. Drug sensitivity tests were
   performed on 15 PDOs derived from EUS-FNA specimens to evaluate their
   response to three chemotherapeutic drugs. The dose-response curves,
   which derived from cell viability, for GEM (A), PTX (B), and GEM + PTX
   (C) are shown. The dose-response curve of the PDOs differed among the
   organoid strains, indicating individual differences in drug
   sensitivity. The vertical axis represents cell viability, the
   horizontal axis represents the logarithmically transformed drug
   concentration. D. Violin plot obtained from Normalized AUC values for
   each drug. The vertical axis represents the normalized AUC value, and
   the horizontal axis represents the drug.

   Normalized the area under the curve (AUC) were generated from each
   dose-response curve, with high normalized AUC values indicating low
   drug sensitivity, and the low values indicating high sensitivity. The
   violin plot shows the normalized AUCs of PDOs treated with GEM, PTX, or
   GEM + PTX (Fig. [72]2D). Drug sensitivity tests were performed at
   similar drug concentration ranges in the GEM and GEM + PTX groups. The
   normalized AUC values were significantly lower in the GEM and GEM + PTX
   groups than in the PTX group (P value < 0.01). There was no
   statistically significant difference between the GEM and GEM + PTX
   groups; however, there was a trend toward lower normalized AUC values
   in the GEM + PTX group.

Evaluation of the transcriptional changes associated with GEM or GEM + PTX
resistance

   We obtained the transcripts of 4,880 genes from 15 PDOs using
   transcriptome analysis. Using the obtained log10(TPM), Pearson
   correlation analysis was performed with the normalized AUC values of
   each drug sensitivity analysis result to identify the genes with a P
   value < 0.001. Among the genes correlated with the normalized AUC, 37
   were identified in the GEM group, 6 in the PTX group, and 12 in the
   GEM + PTX group.

   Non-negative Matrix Factorization (NMF) analysis was performed on the
   identified genes in each group. In the GEM group, 37 genes were
   classified into two groups: basis1 and basis2. The basis1 gene group
   was associated with high normalized AUC values and low drug
   sensitivity, whereas the basis2 gene group had low normalized AUC
   values and high drug sensitivity (Fig. [73]3A). GO enrichment analysis
   and pathway enrichment analysis were performed on characteristic genes
   in the basis1 gene group (framed blue in Fig. [74]3A) associated with
   drug resistance (supplementary Table [75]s1,s2). In the biological
   processes (BP) of the GO terms involved in the regulation of cancer
   cell growth, such as DNA damage repair, cell cycle regulation, and
   replication fork processing, BARD1 and RAD50 were identified as
   significant factors (Fig. [76]3B, C). Regression analysis revealed a
   positive correlation, in which the normalized AUC value increased as
   the expression levels of BARD1 and RAD50 increased (Fig. [77]3D, E).
   For both BARD1 and RAD50, normalized AUC values correlated with gene
   expression levels, indicating that high expression levels of these
   genes correlate with GEM resistance.

Fig. 3.

   [78]Fig. 3
   [79]Open in a new tab

   Drug sensitivity test and transcriptome analysis revealed the genes
   correlated to GEM resistance. A NMF analysis of 37 genes identified in
   the GEM group. The basis components were 37 genes on the vertical axis
   and two on the horizontal axis. The gene group framed blue is related
   to basis1, and the gene group framed red is related to basis2. For the
   mixture coefficients, the vertical axis represents the rank of 2, and
   the horizontal axis represents the PDOs sample. The relationships
   between each basis and normalized AUC values divided into three groups
   are shown. B, C Gene concept network of GO enrichment analysis (B) and
   pathway enrichment analysis (C) of the basis1 gene group. GO enrichment
   analysis identified BARD1, RAD50, PPP1R10, SRSF6, MORC3, and USP37.
   Pathway enrichment analysis identified BARD1, RAD50, RAB12, and ALS2.
   D, E Regression analysis targeting BARD1 (D) and RAD50 (E) identified
   in enrichment analysis. The normalized AUC of each gene is shown on the
   vertical axis, and log10 (TPM) on the horizontal axis. The regression
   line showed a positive slope, indicating that a high expression level
   was correlated with a high normalized AUC value. In the box-and-hide
   diagram, the vertical axis shows the log10(TPM) in each gene, and the
   horizontal axis shows the PDO groups divided into three groups
   according to the normalized AUC values, group1 had low, group2 had
   middle and group3 had high AUC values. It is shown that the group with
   a high normalized AUC value also has a high log10(TPM) value.

   NMF analysis was performed on the 12 genes identified in the GEM + PTX
   group (Fig. [80]4A). GO enrichment analysis and Pathway enrichment
   analysis (supplementary Table s3, s4) were performed on the basis1 gene
   group (framed in blue in Fig. [81]4A). SLC25A10 was identified in both
   GO enrichment analysis and pathway enrichment analysis. MAP3K9, a
   factor involved in the mitogen-activated protein kinase (MAPK) pathway,
   was also observed by GO enrichment analysis (Fig. [82]4B, C).
   Regression analysis revealed a positive correlation, in which the
   normalized AUC value increased as the expression levels of SLC25A10 and
   MAP3K9 increased (Fig. [83]4D, E). NMF and enrichment analyses were
   performed on the six genes identified in the PTX group; however, no
   genes associated with drug resistance were identified.

Fig. 4.

   [84]Fig. 4
   [85]Open in a new tab

   Drug sensitivity test and transcriptome analysis revealed the genes
   correlated to GEM + PTX resistance. A NMF analysis results for 12 genes
   identified in the GEM + PTX group. Basis components are 12 genes on the
   vertical axis and rank of 2 on the horizontal axis. The gene group
   framed blue is related to basis1, and the gene group framed red is
   related to basis2. For the mixture coefficients, the vertical axis is
   rank of 2, and the horizontal axis is PDOs sample. The relationships
   between each basis and normalized AUC values divided into three groups
   are shown. B, C. Gene concept network of GO enrichment analysis (B) and
   pathway enrichment analysis (C) performed on the basis1 gene group,
   which corresponded to a high normalized AUC value in the NMF analysis
   (adjusted P-value < 0.05). GO enrichment analysis identified SLC25A10,
   MAP3K9, TMX1, MANF, and OSGIN2. The pathway enrichment analysis
   revealed that SLC25A10 and MRPL47. D, E. Regression analysis targeting
   SLC25A10 (D) and MAP3K9 (E) as identified by enrichment analysis. The
   normalized AUC of each gene, SLC25A10 or MAP3K9, is shown on the
   vertical axis, and log10 (TPM) on the horizontal axis. The regression
   line showed a positive slope, indicating that a high expression level
   correlated with a high normalized AUC value. In the box-and-hide
   diagram, the vertical axis shows the log10(TPM) in each gene, and the
   horizontal axis shows the PDO groups divided into three groups
   according to the normalized AUC values, group1 had low, group2 had
   middle and group3 had high AUC values. It has been shown that the group
   with a high normalized AUC value also has a high log10(TPM) value.

   We also performed a multiple regression analysis in order to determine
   whether the combination of genes extracted in the NMF analysis was
   associated with stronger drug resistance. First, for the basis1 group
   in the GEM group, we focused on six genes that were significant in the
   GO enrichment analysis. It was confirmed that the combination of BARD1
   and RAD50, PPP1R10 expression (P value = 0.000049, R^2 = 0.9095) or
   BARD1 and RAD50, SRSF6 expression (P value = 0.000056, R^2 = 0.9074),
   which was associated with normalized AUC values, namely, GEM
   resistance. Also, on the basis1 gene group in the GEM + PTX group, we
   focused on the 5 genes that were found to be significant in the GO
   enrichment analysis. The combination of SLC25A10 and MAP3K9 expression
   (P value = 0.000015, R^2 = 0.8926) had a higher association with the
   normalized AUC value, namely, GEM + PTX resistance.

Comparison of the identified genes expression with gemcitabine responsiveness
in clinical practice

   Sixty-five PDAC patients underwent EUS-FNA between April 2023 and June
   2024. Ultimately, the expression levels of 4 identified genes as GEM
   resistance: BARD1, RAD50, MAP3K9 and SLC25A10, were evaluated by
   RT-qPCR using 8 stably established PDOs derived from patients whose
   treatment responses were assessed (Fig. [86]5A, supplementary Table
   s5). These PDOs derived from two groups of PDAC patients: good response
   (GoR, n = 4) and poor response (PoR, n = 4). Both groups were treated
   with GEM-containing regimens, a different cohort from the patients in
   the drug sensitivity analysis (Fig. [87]5A). There were no differences
   in the expression levels of RAD50 and SLC25A10 in 2 groups, GoR and PoR
   (data not shown). MAP3K9 expression was also not different between the
   two groups, while BARD1 expression in the PoR group tended to be higher
   than in the GoR group (Fig. [88]5B, C).

Fig. 5.

   [89]Fig. 5
   [90]Open in a new tab

   BARD1 and MAP3K9 expression in 2 groups PDAC patients with different
   chemotherapeutic effects who received treatment with GEM-containing
   regimens. PDAC Patients selection for BARD1 and MAP3K9 expression assay
   using RT-qPCR (A). Sixty-five PDAC patients underwent EUS-FNA from
   April 2023 to June 2024. Finaly, BARD1 and MAP3K9 expression were
   evaluated using stably established 8 PDOs derived from PDAC patients
   who received GEM-containing regimens and were evaluated treatment
   responses. FOLFIRINOX indicates
   Fluorouracil、Leucovorin、Irinotecan、Oxaliplatin. BARD1 (B) and MAP3K9
   (C) expressions were evaluated with RT-qPCR using PDOs derived from
   2groups PDAC patients, good response (GoR, n = 4) and poor response
   (PoR, n = 4). Both groups received treatment with GEM-containing
   regimens, and they are different cohorts from those who underwent drug
   sensitivity analysis. BARD1 expression in PoR group tended to be higher
   than GoR group. MAP3K9 expression didn’t have differences between 2
   groups. Abbreviations: GoR, good response. MiR, middle response. PoR,
   poor response.

Discussion

   In the present study, we performed drug sensitivity tests and
   transcriptome analyses of PDOs obtained from EUS-FNA specimens to
   identify markers for evaluating drug resistance. We identified some
   markers associated with resistance to GEM and GEM + PTX; among them,
   BARD1 and RAD50 were considered potential predictors of GEM resistance
   in PDAC.

   GEM, 2′,2′-difluoro 2′-deoxycytidine (dFdC), is a deoxycytidine analog
   and utilized for treating several cancers. In PDAC, it are most
   classical chemotherapeutic agents^[91]16. GEM is metabolized within
   cells to form the nucleotides gemcitabine diphosphate (dFdCDP) and
   triphosphate (dFdCTP), which have been shown to exhibit cytotoxic
   effects by inhibiting DNA synthesis^[92]17,[93]18. In addition, dFdCTP
   competes with deoxycytidine triphosphate (dCTP) to induce apoptosis
   after incorporation into the DNA strand by DNA polymerase^[94]19. In
   other words, GEM inhibits DNA synthesis by incorporating into DNA
   strands and exhibits anticancer effects.

   BARD1 forms a heterodimer with BRCA1 via its N-terminal RING
   domains^[95]20. It has been suggested that BRCA1 directly binds to DNA
   and promotes DNA repair^[96]21; however, BARD1 has also been found to
   bind to DNA and contribute to the DNA-binding ability of the
   BRCA1-BARD1 complex^[97]22. BARD1 and BRCA1 are highly expressed in
   tamoxifen-resistant breast cancer cells, suggesting that they resist
   DNA-damaging chemotherapy by enhancing DNA damage repair^[98]23. RAD50
   forms the MRE11-RAD50-NBS1 (MRN) complex and plays an important role in
   DNA damage response. The MRN complex is the core conductor of early and
   sustained responses to DNA double-stranded breaks, replication fork
   stalling, telomere dysfunction, and viral DNA infection^[99]24.
   Additionally, the MRN complex may contain replication fork surveillance
   or DNA repair mechanisms that respond to chemotherapeutic drug-induced
   replication inhibition and may contribute to drug resistance^[100]25.
   Therefore, it is considered that the increased expression of BARD1 and
   RAD50 improves DNA repair ability and also suppresses the induction of
   cell death, possibly resulting in GEM resistance. BARD1 and RAD50 are
   potential predictive markers of GEM resistance.

   Additionally, it was confirmed that the combination of BARD1 and RAD50
   expression added to PPP1R10 or SRSF6 expression showed a higher
   association with normalized AUC values; PPP1R10 is a protein
   phosphatase 1 (PP1) nuclear-targeting subunit that responds to DNA
   damage involved in DNA repair^[101]26,[102]27. SRSF6 activates DNA
   damage response pathways such as DNA repair and double-strand break
   repair by regulating alternative splicing^[103]28. In addition to BARD1
   and RAD50, multiple DNA repair-related genes may synergistically confer
   GEM resistance in PDAC.

   We identified SLC25A10 and MAP3K9 as potential predictors of GEM + PTX
   resistance in PDAC. The SLC25 family of nuclear-encoded transporters
   located in the inner mitochondrial membrane is involved in many
   metabolic pathways and cellular functions^[104]29. SLC25A10 is also a
   dicarboxylate carrier that transports malate and succinate out of
   mitochondria in exchange for phosphate, sulfate, and
   thiosulfate^[105]30,[106]31. Overexpression of SLC25A10 has been
   associated with decreased overall survival in lung cancer and breast
   cancer patients^[107]32. It has also been suggested that SLC25A10 may
   confer cisplatin resistance by protecting tumor cells from oxidative
   stress^[108]33.

   MAP3K9 (mitogen-activated protein kinase kinase kinase 9), also known
   as mixed-lineage kinase 1 (MLK1), is an important upstream component in
   the MAPK pathway^[109]34,[110]35. It has been suggested that MAP3K9
   enhances PDAC cell growth and partially inhibits cell apoptosis through
   activation of the MEK/ERK and NF-kB pathways^[111]36. MLK1 inhibitors
   have been shown to inhibit the growth of pancreatic cancer cell
   lines^[112]37. These results suggest that SLC25A10 is involved in tumor
   protection and MAP3K9 is involved in tumor growth, and the increased
   expression of SLC25A10 and MAP3K9 shown in this study may contribute to
   GEM + PTX resistance.

   We evaluated the expression of the genes identified as associated with
   GEM or GEM + PTX resistance by qPCR in PDAC patients who received
   chemotherapy with GEM-containing regimens. There were no differences of
   the expression levels of RAD50 and SLC25A10 in 2 groups, GoR and PoR
   (data not shown). Similarly, MAP3K9 expression didn’t have significant
   differences between 2 groups (Fig. [113]5C), while BARD1 expression in
   PoR group tended to be higher than GoR group. Based on these results,
   among the four GEM resistance-related genes identified in this study,
   BARD1 showed potential for clinical application as a predictive factor
   for GEM resistance.

   This study had four limitations. First, although 32 PDOs were
   established, a drug sensitivity test was performed on 15 PDOs with good
   growth; PDOs with insufficient growth were excluded. The small sample
   size of 15 PDOs may limit statistical power. Second, the short
   observation period of patients for whom PDOs were established made it
   difficult to assess the correlation between patient outcomes and the
   results of drug sensitivity tests. Third, the study had to be performed
   in the absence of cancer-related fibroblasts (CAFs), which is an
   important requirement for drug sensitivity assessment. Fourth, the
   treatment response assessment was insufficient due to the small number
   of cases (8 cases) and the short observation period of chemotherapy.
   PDAC patients who received chemotherapy with GEM-containing regimens
   included many cases in which treatment was limited to a short period as
   preoperative chemotherapy. We also considered using EUS-FNA specimens
   instead of PDOs, but EUS-FNA specimens are used for cancer gene panel
   analysis and therefore their use was restricted.

   In future, we plan to conduct drug sensitivity tests using more PDOs
   and evaluate their relationships with patient outcomes in sufficient
   observation period. In addition, to reproduce the tumor
   microenvironment, we are planning to use a co-culture model of PDOs and
   CAFs^[114]38,[115]39 in order to regulate the expression of the
   identified genes and evaluate the changes in drug sensitivity.

   In conclusion, we have performed drug sensitivity tests and
   transcriptome analysis of PDOs obtained from EUS-FNA specimens and
   identified BARD1 and RAD50 as predictive markers associated with GEM
   resistance and SLC25A10 and MAP3K9 as predictive markers associated
   with GEM + PTX resistance. PDOs have the potential to evaluate
   chemotherapeutic drug sensitivity and predict markers of drug
   resistance to chemotherapeutic drugs clinically.

Methods

Human pancreatic cancer samples

   The participants were patients with pancreatic cancer who underwent
   EUS-FNA at Yamagata University Hospital between March 2020 and March
   2023. This study was approved by the Ethics Committee of Yamagata
   University Hospital. All experiments were performed in accordance with
   the Declaration of Helsinki and the Commission guidelines and
   regulations. We obtained informed consent from all the participants in
   accordance with Ethical guidelines for life science and medical
   research involving human subjects. Also, the specimens and the data
   provided from the participants were stored, in accordance with the same
   ethical guidelines, immediately converted into a state that does not
   allow for the identification of individuals, and password-protected.

EUS-FNA

   This procedure was performed using a convex ultrasonic endoscope
   (GF-UCT260; Olympus, Japan) equipped with an ultrasound endoscopic
   puncture needle (AcquireTM 22G; Boston Scientific, USA). The puncture
   needle was moved back and forth 20 times per session, and 2–4 sessions
   were performed for each patient. After puncturing, the specimen was
   removed with a stylet, and tumor cells were confirmed by rapid on-site
   evaluation (ROSE). If tumor cells were detected by ROSE, the procedure
   was terminated, and PDOs were created from the surplus specimen.

Organoid culture

   PDOs were cultured with slight modifications to the method described by
   Seino et al.^[116]12. The FNA samples were incubated with Red Blood
   Cell Lysis Buffer (Roche) to remove the red blood cells. The resulting
   pancreatic cancer cells were embedded in Matrigel and cultured in
   Advanced DMEM/F12 supplemented with penicillin/streptomycin, 10 mM
   HEPES, 2 mM GlutaMAX, 1×B27 (Thermo Fisher Scientific), 10 nM gastrin I
   (Sigma-Aldrich), and 1mM N-acetylcysteine (Wako, Japan). The culture
   medium was supplemented with the following niche factors: 100 ng/ml
   mouse recombinant noggin (Peprotech), 10% R-spondin-1 conditioned
   medium, 25% Afamin-Wnt-3 A serum-free conditioned medium, 500 nM A83-01
   (Tocris).

Pathological analysis

   The prepared PDOs were solidified into a jelly according to the
   protocol of iPGell (GenoStaff, Japan). They were then fixed in 10%
   neutral buffered formalin, and tissue blocks were created. Thin
   sections were prepared from tissue blocks, stained with hematoxylin and
   eosin (HE), and pathologically evaluated.

Drug sensitivity test

   PDOs were disassembled into single cells and seeded at 3000 cells per
   well in a 96-well plate. The chemotherapy drugs used were gemcitabine
   (GEM), paclitaxel (PTX), and gemcitabine + paclitaxel (GEM + PTX).
   GEM + PTX was mixed at a GEM: PTX ratio of 8:1. The concentration range
   of each is 1.0 × 10^−8 M to 2.0 × 10^−2 M for GEM, 1.0 × 10^−8 M to
   2.6 × 10^−3 M for PTX, and 1.0 × 10^−8 M to 2.3 × 10^−2 M for
   GEM + PTX. Each drug was administered in duplicates. After incubating
   the plate treated with chemotherapy drugs at 37 °C for 72 h, 20 µl per
   well of CellTiter-Glo Cell Viability Assay (Promega) was added and
   incubated in the dark for 4 h. The number of viable cells was assessed
   by measuring fluorescence using a Varioskan LUX multimode microplate
   reader (Thermo Fisher Scientific). Cell viability was calculated at
   each concentration using the minimum concentration as the reference
   value.

Drug sensitivity analysis

   A dose-response curve (top 1, bottom 0) was derived from cell
   viability, and AUC was calculated using GraphPad Prism 9 (GraphPad
   Software). The normalized AUC values were calculated by dividing the
   AUC by the AUC of the maximum concentration, and the normalized AUC
   values of each drug were divided into three groups and then compared.

Extraction of DNA and RNA from the PDOs

   The PDOs were disrupted with 250 µl per well Accumax (Innovative Cell
   Technologies) and collected in a microtube. DNA was extracted using a
   DNeasy Blood & Tissue Kit (Qiagen), and RNA was extracted using an
   RNeasy Mini Kit (Qiagen). The concentration of the extracted nucleic
   acids was measured using a Qubit Fluorometer (Qiagen).

Panel sequencing

   Using the extracted DNA, we performed mutation analysis using a
   next-generation sequencer on approximately 400 genes that could be
   evaluated using CCP. First, a library was constructed using the Ion
   AmpliSeq Comprehensive Cancer Panel (4477685, Thermo Fisher Scientific)
   and the Ion AmpliSeq Library Kit Plus (4488990, Thermo Fisher
   Scientific), according to the manufacturer’s protocols. Library
   quantification was performed on a 4150 TapeStation system (Agilent
   Technologies) using a D1000 ScreenTape (5067–5582, Agilent
   Technologies). Using the prepared library, sequencing was performed on
   an Ion GeneStudio S5 (Thermo Fisher Scientific) using the Ion 540
   Kit-OT2 and Ion 540 Chip (A27753, A27765, Thermo Fisher Scientific),
   according to the manufacturer’s protocols. The obtained BAM files were
   analyzed using an in-house pipeline. Sequence reads were mapped to the
   human genome reference sequence (GRCh38/hg38) using BWA
   ([117]http://bio-bwa.sourceforge.net/bwa.shtml) and Bowtie2
   ([118]http://bowtie-bio.sourceforge.net/bowtie2/index.shtml). Variant
   calling was performed using Freebayes
   ([119]https://github.com/ekg/freebayes, Ver.1.3.6), and annotation was
   performed using ANNOVAR([120]http://annovar.openbioinformatics.org).
   Sequence data with coverage > 50 were used for the analysis.

Transcriptome analysis

   A library was constructed using the extracted RNA with an Ion AmpliSeq
   Transcriptome Human Gene Expression Kit (A26325; Thermo Fisher
   Scientific) according to the manufacturer’s protocol. Library
   quantification was performed on a 4150 TapeStation system (Agilent
   Technologies) using a D1000 ScreenTape (5067–5582, Agilent
   Technologies). Using the prepared library, sequencing was performed on
   an Ion GeneStudio S5 (Thermo Fisher Scientific) using the Ion 540
   Kit-OT2 and Ion 540 Chip (A27753, A27765, Thermo Fisher Scientific),
   according to the manufacturer’s protocols. RNA sequencing (RNA-seq) was
   analyzed using an in-house pipeline. The obtained sequence reads were
   mapped against the human genome reference sequence (GRCh38/hg38) using
   STAR ([121]https://github.com/alexdobin/STAR, Ver.1.6.0) (quantMode
   TranscriptomeSAM), and gene-level transcript quantification and
   transcripts per million (TPM) calculations were performed using RSEM
   ([122]https://github.com/deweylab/RSEM, Ver.1.3.3)
   (rsem-calculate-expression).

Evaluation of BARD1 and MAP3K9 expression in the PDAC patients who received
GEM-containing regimen

   Additionally, we investigated the relationship between the drug
   resistance genes identified in this study and GEM-containing regimen
   responsiveness in a cohort study. Sixty-five PDAC patients underwent
   EUS-FNA between April 2023 and June 2024. Ultimately, the expression
   levels of 4 identified genes as GEM resistance: BARD1, RAD50, MAP3K9
   and SLC25A10, were evaluated using 8 stably established PDOs derived
   from patients whose treatment responses were assessed (Fig. [123]5A,
   supplementary Table s5). For the evaluation of GEM-containing regimen
   responsiveness, treatment response was classified based on RECIST
   criteria and the number of major tumor markers (CA19-9, CEA, Span-1,
   DUPAN-2) that showed a decrease. Specifically, the response was
   categorized as follows: Good response (GoR): SD + decrease in all 4
   tumor markers, middle response (MiR): SD + decrease in 3 tumor markers,
   poor response (PoR): PD or SD + decrease in 2 or fewer tumor markers.

   Using 4 PDO samples each derived from PDAC patients with GoR and PoR,
   respectively, we evaluated the expression of BARD1 and MAP3K9 by
   RT-qPCR. Both groups came from a different cohort than the patients
   used for the drug sensitivity analysis. The RNA extraction method was
   the same as described above, using the RNeasy Mini Kit (Qiagen). cDNA
   was synthesized according to the protocol of the High-Capacity cDNA
   Reverse Transcription Kit (Thermo Fisher Scientific). Additionally,
   qPCR was performed using TaqMan Fast Advanced Master Mix for qPCR
   (Thermo Fisher Scientific) and TaqMan Gene Expression Assays (Thermo
   Fisher Scientific).

Statistical analysis

   Using the obtained log10(TPM), Pearson correlation analysis was
   performed with the normalized AUC values of each drug sensitivity
   analysis result, and genes were identified under the condition of P
   value < 0.001. NMF analysis, a nonhierarchical clustering method, was
   used to extract genes associated with normalized AUCs representing drug
   sensitivity. Using the R package NMF
   ([124]http://renozao.github.io/NMF/, Ver.0.26), NMF analysis was
   performed on the log10(TPM) of the identified genes (rank = 2,
   options=’t’, nrun = 400). Based on the results of the NMF analysis, GO
   enrichment analysis and pathway enrichment analysis were performed on
   gene groups that were found to be associated with the normalized AUC
   values using ClusterProfiler version 4.8.3, and ReactomePA version
   1.44.0 (pAdjustMethod = ‘fdr’, pvalueCutoff = 0.05). Finally, we
   performed a regression analysis of the genes obtained from the
   enrichment analysis and normalized the AUC values.

Electronic supplementary material

   Below is the link to the electronic supplementary material.
   [125]Supplementary Material 1^ (29.6KB, docx)

Acknowledgements