Graphical abstract

   graphic file with name fx1.jpg
   [45]Open in a new tab

Highlights

     * •
       Establishment of a biobank comprising living gastric cancer (GC)
       organoids
     * •
       Identification of a chemosensitivity-associated gene expression
       signature
     * •
       GC organoids reproduce chemotherapeutic drug responses of patients
       in vitro and in vivo
     * •
       GC organoids can be co-cultured with T cells and cancer-associated
       fibroblasts
     __________________________________________________________________

   Yi et al. present the establishment of a biobank of gastric cancer
   organoids derived from patients. These organoids mirror the complex
   dynamics of chemotherapy response. It may offer a valuable tool for the
   refinement of personalized treatments and the prediction of individual
   patient responses to chemotherapy.

Introduction

   Gastric cancer (GC) is the fifth most common cancer and the fourth most
   common cause of cancer death globally, and its incidence rate has
   increased over the last decades.[46]^1 Surgical resection combined with
   perioperative chemotherapy is the first-line curative option for
   patients with locally advanced GC.[47]^2 The most common conventional
   chemotherapeutic drugs include fluorouracil (e.g., 5-fluorouracil
   [5-FU], capecitabine, and S-1) and platinum compounds (e.g., cisplatin
   and oxaliplatin).[48]^3^,[49]^4^,[50]^5 However, the responses of
   patients with GC to these chemotherapies are highly varied without
   clear reasons.[51]^6^,[52]^7 To improve the efficacy of chemotherapy
   and clinical outcome, it is therefore unmet to develop approaches that
   can accurately predict the efficacy of chemotherapy in patients with
   GC.

   Organoid cultures are three-dimensional cell aggregates. Recent studies
   have reported that patient-derived organoids (PDOs) can be predictive
   of patient’s treatment response.[53]^8^,[54]^9 In particular, PDOs
   could accurately preserve and recapitulate both the genetic landscape
   and histopathological features of their originated tumors, implicating
   their potential application in personalized cancer treatment.[55]^10
   Three independent groups reported the establishment of PDOs derived
   from patients with GC and explored the application of PDOs for
   predicting drug responses.[56]^11^,[57]^12^,[58]^13 However, these
   studies lack investigation into gene expression characteristics related
   to chemosensitivity. Gene signatures characterized from organoids or
   original tumor tissues can facilitate the identification of predictive
   biomarkers of treatment response[59]^14^,[60]^15; hence, integrated
   analyses on gene expression and chemosensitivity based on PDOs are
   therefore important. Moreover, the sample sizes of previous studies of
   drug screening on GC PDOs were relatively
   small.[61]^11^,[62]^12^,[63]^13 Performing drug screening on a large
   collection of GC PDOs is necessary to provide accurate and reliable
   personalized treatment prediction for patients with GC. Here, we
   reported the establishment of a patient-derived GC organoid biobank and
   their ability to model the histological features of human GC, as well
   as their responses to chemotherapeutic drugs. We analyzed the molecular
   characteristics of GC PDOs to identify marker genes associated with
   chemotherapeutic drug sensitivity and resistance. We also utilized a
   patient-derived organoids-based xenograft (PDOX) model to verify the
   results of drug sensitivity test from PDOs, and we further testified
   with actual treatment response in individual patients with GC.

Results

Establishing a living patient-derived GC organoid biobank

   We obtained gastric tumor tissues from 73 patients with histologically
   diagnosed gastric adenocarcinoma from surgical resection specimens. The
   clinicopathological characteristics of all patients are shown in
   [64]Table S1. We successfully established 57 organoids derived from
   these GC tumor tissues with an overall success rate of 78% (57/73)
   ([65]Figure 1A). Specifically, 53 organoids were derived from
   treatment-naive patients, and four organoids were derived from patients
   with neoadjuvant chemotherapy ([66]Figure 1B). GC organoids were
   derived from various sites of the stomach, such as the cardia (n = 8),
   fundus (n = 5), body (n = 16), antrum (n = 22), and horn (n = 6). These
   organoids were obtained from patients with different
   tumor-node-metastasis (TNM) stages, including stage I (n = 17), II (n =
   7), III (n = 28), and IV (n = 5) ([67]Figure 1B).

Figure 1.

   [68]Figure 1
   [69]Open in a new tab

   Establishing a living patient-derived GC organoid biobank

   (A) Schematic of GC organoid culture establishment and analysis
   performed.

   (B) Pie chart showing the sample origins, tumor sites, and TNM stages
   of the established 57 GC organoids.

   (C) Representative pictures of 2 GC organoids with high growth rate and
   2 with low growth rate. Scale bar, 200 μm.

   (D and E) Heatmap and volcano plot analysis showing the most
   significantly differential genes in GC organoids with high growth rate
   (n = 4) compared to organoids with low growth rate (n = 13). The
   colored bar represents the log2-transformed values. Individual genes
   are denoted by gene name. Differentially expressed genes (p < 0.05,
   log2 (fold change) > 1) of high-growth-rate organoids and
   low-growth-rate organoids are indicated in red or blue, respectively.

   (F) KOBAS of the differential gene expression profiles of
   high-growth-rate organoids compared to the low-growth-rate organoids.

   (G) Stemness score analysis between high-growth-rate organoids and
   low-growth-rate organoids.

   Distinct GC organoids had different growth rates. Five organoid
   cultures could be serially propagated for up to 17 passages without any
   apparent diminution in their ability to propagate. These organoids
   including G5T, G9T, G27T, G30T, and G72T were therefore defined as
   high-growth-rate organoids, which showed a proliferative progenitor
   phenotype and the formation of a spheroid, cyst-like morphology without
   lumen ([70]Figure 1C). Other organoid cultures were successfully
   established but failed to further propagate after 8–9 passages; hence,
   they were defined as low-growth-rate organoids. These 52
   low-growth-rate organoids had a glandular morphology with the
   appearance of lumens ([71]Figure 1C). Thus, these results indicated
   that we have successfully established a biobank of GC PDOs. These
   organoids had different growth features, such as high-growth-rate or
   low-growth-rate organoids.

Elucidating gene expression profiles in organoids with high growth rate or
low growth rate

   RNA sequencing was conducted to compare GC organoids with high growth
   rate or low growth rate. The significant differential gene expression
   was identified in high-growth-rate organoids compared to
   low-growth-rate organoids ([72]Figure 1D). Proliferation- and
   stemness-related genes REG4, KLF4, ERBB3, HRAS, NOTCH1, and MYC were
   outlier genes and significantly upregulated in high-growth-rate
   organoids ([73]Figure 1E), while genes related to cell growth
   inhibition, including BAX, DKK3, TNFSF12, MCC, BNIP3, and TP53BP1, were
   downregulated in high-growth-rate organoids compared to low-growth-rate
   organoids ([74]Figures 1D and 1E). Consistently, Kyoto Encyclopedia of
   Genes and Genomes (KEGG) Orthology-Based Annotation System (KOBAS) of
   differentially expressed genes confirmed that proliferation- and
   stemness-related pathways were more upregulated in high-growth-rate
   organoids, including mitogen-activated protein kinase (MAPK) family
   signaling cascades and signaling by NOTCH, all of which could
   accelerate cell proliferation and maintain stem function
   ([75]Figure 1F). Cell differentiation and apoptosis-related pathways,
   including the Hippo signaling pathway and the p53 signaling pathway,
   were relatively downregulated in high-growth-rate organoids compared to
   low-growth-rate organoids ([76]Figure 1F). The high-growth-rate
   organoids also had higher stemness score (0.784) in comparison to
   low-growth-rate organoids (0.345) (p < 0.05) ([77]Figure 1G).
   Additionally, we analyzed the correlation between the growth rates of
   organoids and clinic pathological parameters and found that the growth
   rates (high growth rate vs. low growth rate) of organoids did not
   exhibit a significant correlation with clinicopathologic
   characteristics ([78]Table S2). Whole-exome sequencing was conducted to
   examine the mutational profiles of organoids in terms of growth rate.
   There was no significant difference for gene mutations between
   high-growth-rate organoids and low-growth-rate organoids, including the
   well-known driver mutations of TP53, CDH1, ARID1A, MUC6, and RNF43
   ([79]Figure S1). Hence, high-growth organoids and low-growth organoids
   have different gene expression patterns. High-growth organoids showed
   more robust proliferation and stemness properties, which could at least
   in part explain their difference in growth features.

Gene expression analysis of GC organoids and corresponding tumor tissue

   Next, we investigated how stable tumor-specific gene expression is
   maintained in organoid culture. We have now compared the RNA sequencing
   results of organoids and primary tumors. We observed a high degree of
   similarity in gene expression patterns between organoids and their
   corresponding primary tumor tissues by RNA sequencing ([80]Figures 2A
   and 2B). To test the similarity between primary tumors and organoids,
   we quantitatively analyzed and compared gene expression at the level of
   alternative transcripts from RNA sequencing (in log[2] fragments per
   kilobase of exon model per million mapped fragments) in primary tumors
   and organoids. A high level of concordance in gene expression was
   demonstrated between the matched organoids and primary tumors (average
   ρ: 0.785, average R^2:0.64) ([81]Figure S2A).

Figure 2.

   [82]Figure 2
   [83]Open in a new tab

   Transcriptome analysis and histopathologic characteristics between GC
   organoids and their derived primary GCs

   (A) Heatmap of differentially expressed genes and (B) pathways between
   organoids and their corresponding tumor tissue. The colored bar
   represents the log2-transformed values. Differentially expressed genes
   (p < 0.05, log2 (fold change) > 1) of organoids and tissue are
   indicated in red or blue, respectively.

   (C) Representative images of bright-field microscopy, hematoxylin and
   eosin staining of GC organoids, and matched primary GC tumor tissues.

   (D) Immunohistochemistry of GC protein markers CK7 and CEA in organoids
   and their corresponding primary GC tumor tissues. Scale bar, 50 μm.

GC organoids preserve the histopathologic characteristics of their derived
primary GCs

   We assessed whether GC organoids could preserve the histological
   characteristics and protein expression of their originated tumors. As
   expected, GC organoids retained specific glandular features that were
   observed in their corresponding primary tumors, including glandular,
   discohesive, or solid growth pattern and nuclear stratification
   ([84]Figures 2C and [85]S2B). The organoids also exhibited similar
   histological features as the subtypes of their corresponding primary
   tumors including diffuse type, intestinal type, and mixed type of GC.
   GC organoids derived from intestinal-type tumors (G20T and G43T)
   exhibited a large sphere that consisted of a single epithelial layer or
   cribriform glandular morphology with cells forming multiple lumens of
   varying sizes. GC organoids derived from diffuse-type tumors (G27T and
   G72T) formed loosely cohesive solid cell clusters, or cystic structures
   (G24T and G39T). GC organoid derived from mixed-type GC (G36T and G41T)
   displayed either or both histological patterns of diffuse- and
   intestinal-type tumors ([86]Figures 2C and S2B). To confirm that GC
   organoids maintained the characteristics of their originated tumors, we
   performed immunohistochemistry for the protein expression of GC markers
   cytokeratin 7 (CK7) and carcinoembryonal antigen (CEA). As shown in
   [87]Figures 2D and [88]S2C, GC organoids displayed a similar presence
   and intensity of these protein markers compared to their corresponding
   primary tumors. We found that the Allred scores of CK7 and CEA
   intensity from the organoid line and the original tumors were highly
   correlated (Pearson correlation R^2 is 0.87 and 0.91, respectively)
   ([89]Figure S2D). Due to the heterogeneity of the tumor tissue, we also
   checked the inconsistent cases and found that the expression intensity
   of CK7 (in G43T and G56T) and CEA (in G62T and G56T) was different
   between the original tumors and organoids ([90]Figure S2D). These
   results indicated that GC organoids retained the histopathologic
   features of their primary tumors, inferring that GC PDOs could mimic
   primary GCs of individual patients for drug testing.

Drug screening of GC PDOs to chemotherapeutics

   To explore the responses of GC organoids to different drugs, six
   conventional chemotherapeutics were applied to 41 GC organoids. Each
   drug was screened at five different concentrations in triplicate.
   Specifically, according to previous studies,[91]^11 GC organoids were
   exposed to different fixed concentrations of 5-FU (0.2, 1, 5, 10, and
   50 μmol/L), oxaliplatin (0.2, 1, 5, 10, and 50 μmol/L), cisplatin (0.2,
   1, 5, 10, and 50 μmol/L), paclitaxel (0.01, 0.1, 1, 10, and
   100 nmol/L), doxorubicin (0.0001, 0.001, 0.01, 0.1, and 1 μmol/L), or
   SN-38 (0.01, 0.1, 1, 10, and 100 nmol/L), respectively. Two different
   passages of each organoid line were treated with the aforementioned
   drugs in different concentrations to evaluate their responses. We found
   that the area under the dose-response curve (AUC) values of biological
   replicates from two organoid passages of the same organoid line were
   highly correlated (Pearson correlation R^2 > 0.87), indicating that our
   culture system is stable and unaffected by external conditions and the
   disparity in organoid passages ([92]Figures 3A and [93]S3A). To ensure
   the accuracy of half-maximal inhibitory concentration (IC50) estimates,
   we tested the sensitivity of the drugs using an additional
   ten-concentration system. Our analysis revealed a strong correlation
   between the AUC values and IC50 values obtained from two sets of drug
   concentrations (ten concentrations and five concentrations) derived
   from the same organoid line (Pearson correlation R^2 > 0.92)
   ([94]Figures S3B and S3C). We observed that the responses to
   chemotherapeutic drugs varied among organoids generated from different
   patients with GC ([95]Figure 3B; [96]Table S3). Some organoids were
   resistant, and some others were partially sensitive or sensitive to
   5-FU, oxaliplatin, cisplatin, paclitaxel, SN-38, and doxorubicin
   ([97]Figure 3C). The organoids generated from diffuse-type GC were more
   likely to be sensitive to 5-FU and paclitaxel ([98]Figure 3C). These
   results collectively suggested that GC PDOs are potentially effective
   tools for drug screening with varied responses to different
   conventional chemotherapeutics.

Figure 3.

   [99]Figure 3
   [100]Open in a new tab

   Drug screening of GC PDOs to chemotherapeutics

   (A) Scatterplot showing the correlation between the 1-AUC values in two
   independent runs of drug screening (n = 3). Correlation is indicated as
   Pearson R^2 value.

   (B) Dose-response curves showing the chemosensitivity of 41 GC
   organoids to 6 conventional chemotherapeutics (5-FU, oxaliplatin,
   cisplatin, paclitaxel, SN-38, and doxorubicin) (n = 3). AUC from
   dose-response data is displayed as a violin plot; every point indicates
   41 GC organoids, respectively.

   (C) Heatmap of AUC of 6 chemotherapeutics in 41 GC organoids. For each
   chemotherapeutic agent, the PDO library was divided into 2 subgroups:
   resistant (AUC = 50%–100%) and sensitive (AUC = 0%–50%).

Gene expression signatures of GC PDOs in response to chemotherapeutics

   We analyzed the gene expression signatures of 17 PDOs by
   transcriptional sequencing according to their responses to 5-FU or
   oxaliplatin, the most conventional chemotherapeutic drugs for patients
   with GC. The 17 GC organoids were separated into either sensitive or
   resistance groups based on their responses to 5-FU or oxaliplatin
   (sensitive: AUC <50%, resistance: AUC ≥50%). The gene expression
   profiles were compared between sensitive or resistant groups, and
   differential gene expression was demonstrated in GC organoids of the
   sensitive (n = 13) and resistance groups (n = 4) to 5-FU
   ([101]Figure 4A). Significantly upregulated tumor suppression genes
   were identified in the 5-FU-sensitive group compared to 5-FU-resistant
   organoids, including MSMB, S1PR4, PURPL, TP53INP1, and TRIM16
   ([102]Figure 4B). KOBAS demonstrated the upregulation of the
   tumor-suppressing p53 signaling pathway and cellular senescence in
   5-FU-sensitive organoids ([103]Figure 4C), while genes associated with
   proliferation and invasion, such as FKBP10, DARPP32, ASCL2, BCL2,
   ABCC2, and NOTCH1, were significantly downregulated in 5-FU-sensitive
   organoids compared to the resistant organoids ([104]Figures 4A and 4B).
   These downregulated genes were enriched in tumor invasion and
   stemness-related pathways including NOTCH signaling, WNT signaling, Rho
   GTPases signaling, and extracellular matrix organization
   ([105]Figure 4C).

Figure 4.

   [106]Figure 4
   [107]Open in a new tab

   Gene expression signatures of GC PDOs in response to chemotherapeutics

   (A) Heatmap and (B) volcano plot analysis showing the most
   significantly differential genes in 5-FU-sensitive organoids (n = 13)
   compared to organoids with 5-FU resistance (n = 4). The colored bar
   represents the log2-transformed values. Individual genes are denoted by
   gene name. Differentially expressed genes (p < 0.05, log2 (fold
   change) > 1) of 5-FU-sensitive organoids and 5-FU-resistant organoids
   are indicated in red or blue, respectively.

   (C) KOBAS of the differential gene expression profiles of
   5-FU-sensitive organoids compared to 5-FU-resistant organoids.

   (D and E) Heatmaps and volcano plot analysis showing the most
   significantly differential enriched genes in oxaliplatin-sensitive
   organoids (n = 11) compared to organoids with oxaliplatin resistance
   (n = 6).

   (F) KOBAS of the differential gene expression profiles of
   oxaliplatin-sensitive organoids compared to oxaliplatin-resistant
   organoids.

   (G) Knockout of 5-FU-sensitive gene MSMB and S1PR4 in G27T and G72T
   organoids (n = 3).

   (H) Knockout of 5-FU-resistant gene FKBP10 in G9T organoids (n = 3).

   (I) Knockout of oxaliplatin-sensitive gene MYO1A in G27T and G72T
   organoids (n = 3).

   (J) Knockout of oxaliplatin-resistant gene NDUFA4L2 in G9T organoids
   (n = 3).

   Differential gene expression was also observed in GC organoids of
   sensitive (n = 11) and resistant groups (n = 6) to oxaliplatin
   ([108]Figure 4D). Genes related to tumor suppression were significantly
   upregulated in the oxaliplatin-sensitive group as compared to
   oxaliplatin-resistant organoids, including MYO1A, CLDN3, DMBT1,
   TNFRSF1B, and TRIM15 ([109]Figure 4E). These upregulated genes in
   oxaliplatin-sensitive organoids were mainly enriched in
   tumor-suppressing pathways particularly tumor necrosis factor (TNF)
   signaling and FOXO signaling ([110]Figure 4F), while genes associated
   with proliferation and invasion were significantly downregulated in the
   oxaliplatin-sensitive group compared to the resistant group, including
   NDUFA4L2, DKK4, KLK10, FOSL1, STC2, and L1CAM ([111]Figure 4E). These
   downregulated genes were involved in proliferative pathways including
   PI3K signaling and WNT signaling ([112]Figure 4F). Collectively,
   we demonstrated the gene expression signatures of GC PDOs in response
   to chemotherapeutics. Besides, we conducted functional experiments in
   organoids with knocked-out resistance or sensitivity genes identified
   by RNA sequencing. MSMB or S1PR4 (5-FU-sensitive gene) knockout in G27T
   and G72T organoids significantly decreased drug sensitivity to 5-FU
   treatment compared with control cells ([113]Figure 4G). Knockout of
   FKBP10 (5-FU-resistant gene) in G9T organoids increased drug
   sensitivity to 5-FU treatment ([114]Figure 4H). Knockout of MYO1A
   (oxaliplatin-sensitive gene) or NDUFA4L2 (oxaliplatin-resistant gene)
   also decreased or increased drug sensitivity to oxaliplatin treatment
   ([115]Figures 4I and 4J), respectively. These results confirmed the
   functional impact of these genes in organoids on their responses to
   5-FU and oxaliplatin treatment.

Gene mutation features of GC PDOs in response to chemotherapeutics

   Whole-exome sequencing was performed to examine the mutational profiles
   of organoids in response to chemotherapeutic drugs. There was no
   significant difference in gene mutations between 5-FU-sensitive and
   5-FU-resistant organoids, including the well-known mutations of ARID1A,
   TP53, KRAS, CDH1, and PIK3CA in GC ([116]Figure S4A), while more
   mutations were present in the oxaliplatin-sensitive group than those in
   the oxaliplatin-resistant group, with MUC6, ARID1A, and RIMS2 being the
   top three most frequently occurred mutations ([117]Figure S4B).

Gene expression biomarkers to distinguish sensitive and resistant responses
to chemosensitivity in patients with GC

   We then explored gene expression biomarkers of tumor tissues that could
   distinguish sensitive patients from resistant patients to 5-FU
   treatment by the recursive feature elimination algorithm. A total of 9
   genes with significant differential expression were identified
   including 6 upregulated genes (RNASEH2A, MMP3, CXCL1, COL5A2, MXRA5,
   and PTTG1) and 3 downregulated genes (SLC2A1, RRAD, and APLP1) in tumor
   tissues with 5-FU-sensitive patients compared to those with
   5-FU-resistant patients ([118]Figure 5A). We then utilized these 9
   genes to verify accuracy by using logistic and naive Bayes models. The
   results showed that this gene expression biomarker panel could
   discriminate between 5-FU-sensitive patients and resistant patients at
   AUCs of 87% (95% confidence interval (CI): 72.22%–100%) by the logistic
   model and 86% (95% CI: 64.42%–100%) by the naive Bayes model
   ([119]Figure 5B).

Figure 5.

   [120]Figure 5
   [121]Open in a new tab

   Gene expression biomarkers to distinguish sensitive and resistant
   responses to chemosensitivity in patients with GC

   (A) Heatmap showing a total of 9 genes with significant differential
   expression between 5-FU-sensitive patients (n = 18) and 5-FU-resistant
   patients (n = 5). The orange bar represents the log2-transformed fold
   change values.

   (B) Receiver operating characteristic (ROC) analysis to discriminate
   5-FU sensitivity and resistance using the 9 differentially expressed
   genes based on the logistic model (left) or the naive Bayes model
   (right).

   (C) Heatmap showing a total of 6 genes with significant differential
   expression between oxaliplatin-sensitive patients (n = 11) and
   oxaliplatin-resistant patients (n = 12).

   (D) ROC analysis to discriminate oxaliplatin sensitivity and resistance
   using the 6 differentially expressed genes based on the logistic model
   (left) or the naive Bayes model (right). FC, fold change; RFE,
   recursive feature elimination; ROC, receiver operating characteristic.

   A similar analysis was performed to identify the gene expression
   biomarkers of tumor tissues to distinguish sensitive patients from
   resistance patients to oxaliplatin treatment. A total of 6
   differentially expressed genes were identified including 3 upregulated
   genes (TNFSF9, WNT2, and EGR3) and 3 downregulated genes (MT2A,
   SULT2B1, and ALDOB) in tumor tissues with oxaliplatin-sensitive
   patients compared to those with oxaliplatin-resistant patients
   ([122]Figure 5C). This gene expression biomarker panel was further
   validated by using logistic and naive Bayes models, and the results
   showed that this panel could discriminate between sensitive patients
   and resistant patients to oxaliplatin treatment at AUCs of 81% (95% CI:
   59.85%–98.48%) by the logistic model and 83% (95% CI: 60.61%–99.24%) by
   the naive Bayes model ([123]Figure 5D). These findings suggested that
   the gene expression biomarker panel could distinguish sensitive
   patients from resistant patients to chemotherapies of 5-FU and
   oxaliplatin.

   Additionally, we verified this gene signature in another cohort (The
   Cancer Genome Atlas [TCGA]) and analyzed its correlation with survival
   curves. We screened patients treated with 5-FU or oxaliplatin in TCGA
   cohort as our validation sets. Patients in the sets were assigned to
   sensitive-like and resistant-like groups by their gene signature
   expression level. Survival analysis showed that in the 5-FU-treated
   cohort (n = 68), patients assigned to the sensitive group had a better
   prognosis for progression-free survival (p < 0.05) ([124]Figure S5A).
   Due to the limited number of patients receiving oxaliplatin treatment
   in TCGA set (n = 17), we were unable to identify significant
   differences in prognosis between the two groups ([125]Figure S5B). We
   have compared our gene expression biomarker panel with other panels
   published, such as 30-gene signature,[126]^16 53-gene
   signature,[127]^17 and 32-gene signature,[128]^18 in our dataset by
   using logistic and naive Bayes models. Our biomarker panels for 5-FU
   and oxaliplatin have the highest AUC values in both models, thereby
   confirming the effectiveness of the panels in our study
   ([129]Figure S6A and S6B).

Drug-response status observed in GC PDOs is validated in PDOXs in mice

   To validate the drug-response results identified in GC PDOs, we
   established PDOXs in NCG mice. Mice were first implanted with different
   lines of GC organoids. Once the tumors grew larger than 50 mm^3, all
   mice were randomized to receive either drug or vehicle treatment for
   28 days ([130]Figure 6A). G5T-PDOs were sensitive to paclitaxel and
   resistant to 5-FU ([131]Figure 6B). Consistently, mice engrafted with
   G5T-PDOXs demonstrated a significant treatment response to paclitaxel
   (p < 0.0001), as evidenced by significantly reduced tumor growth
   ([132]Figure 6C) and tumor weight ([133]Figure 6D) as compared to
   untreated G5T-PDOXs. Moreover, G5T-PDOX mice showed resistance to 5-FU
   treatment, consistent with the result of G5T-PDOs ([134]Figure 6B). In
   supporting this, cellular proliferation was significantly decreased
   (p < 0.001) and apoptosis was significantly increased (p < 0.01) in
   G5T-PDOXs treated with paclitaxel compared to untreated controls,
   whereas no difference was observed in the 5-FU treatment group
   ([135]Figure 6E). In addition, G9T-PDOs were assessed, which were
   sensitive to 5-FU and resistant to oxaliplatin ([136]Figure 6F). As
   expected, mice engrafted with G9T-PDOXs demonstrated a significant
   treatment response to 5-FU (p < 0.0001) but no response to oxaliplatin
   treatment ([137]Figures 6G and 6H). Following 5-FU treatment, cellular
   proliferation was significantly decreased (p < 0.0001) and apoptosis
   significantly increased (p < 0.01) in G9T-PDOXs compared to untreated
   G9T-PDOXs, whereas no difference was observed in the oxaliplatin
   treatment group ([138]Figure 6I). G27T-PDOs were sensitive to 5-FU,
   oxaliplatin, and paclitaxel ([139]Figure S7A). Consistently, mice
   engrafted with G27T-PDOXs demonstrated a significant treatment response
   to 5-FU, oxaliplatin, and paclitaxel (p < 0.0001), as evidenced by
   significantly reduced tumor growth and tumor weight ([140]Figures S7B
   and S7C) as compared to untreated G27T-PDOXs. G30T-PDOs were assessed
   as sensitive to oxaliplatin and paclitaxel but partially resistant to
   5-FU ([141]Figure S7D). As expected, mice engrafted with G30T-PDOXs
   demonstrated a significantly better response to oxaliplatin and
   paclitaxel (p < 0.0001) than untreated G30T-PDOXs, but a partial
   response to 5-FU treatment ([142]Figures S7E and S7F). Besides,
   G72T-PDOXs were sensitive to 5-FU, oxaliplatin, and SN-38, consistent
   with the drug test result of G72T-PDOs ([143]Figures S7G–S7I). These
   findings suggested that drug responses observed in GC organoid cultures
   were validated in PDOXs in vivo.

Figure 6.

   [144]Figure 6
   [145]Open in a new tab

   Drug-response status observed in GC PDOs is validated in PDOXs in mice
   and corresponding follow-up patients with GC

   (A) Schematic of the mouse experiment.

   (B) Dose-response curves of 5-FU (AUC = 68.61%, IC50 = 39.22 μM) and
   paclitaxel (AUC = 4.32%, IC50 = 0.03 nM) in G5T-PDO in vitro (n = 3).

   (C) Mean tumor size (mm^3) of G5T-PDOX with 5-FU, paclitaxel, or
   vehicle control over time (n = 6 per group) (mean ± SD).

   (D) Images of G5T-PDOX tumors and tumor weight at sacrifice.

   (E) Representative Ki-67 staining and TUNEL staining of mice tumor
   sections with quantitative analysis of Ki-67 index (top) and Allred
   score for apoptosis (bottom) (n = 6), respectively. Scale bar, 50 μm.

   (F) Dose-response curves of 5-FU (AUC = 31.13%, IC50 = 6.136 μM) and
   oxaliplatin (AUC = 47.37%, IC50 = 17.53 μM) in G9T-PDO in vitro (n =
   3).

   (G) Mean tumor size (mm^3) of G9T-PDOX with 5-FU, oxaliplatin, or
   vehicle control over time (n = 6 per group) (mean ± SD).

   (H) Images of G9T-PDOX tumors and tumor weight at sacrifice.

   (I) Representative Ki-67 staining and TUNEL staining with quantitative
   analysis (n = 6). Scale bar, 50 μm.

   (J) Dose-response curves showing the chemosensitivity of 12 GC
   organoids to combined chemotherapeutics (5-FU and oxaliplatin). AUC
   from dose-response data is displayed as a violin plot; every point
   indicates 12 GC organoids, respectively.

   (K) The heatmap summarizes the results for the 12 PDOs and
   corresponding patient drug responses.

Drug-response status observed in GC PDOs is validated in corresponding
follow-up patients with GC

   We further validated the drug-sensitive or resistant results from GC
   PDOs in patients with GC with actual clinical treatment and outcomes.
   Twelve original patients with GC who underwent chemotherapy and had the
   combined drug screening results from their derived organoids were
   followed up ([146]Figures 6J and 6K; [147]Table S4). These patients
   received perioperative or postoperative adjuvant chemotherapy with
   oxaliplatin and 5-FU and were followed up for 1–49 months. According to
   CheckMate 649 trial,[148]^19 which explored chemotherapy of advanced
   GC, the median progression-free survival time in the observation group
   is 6.05 months. Thus, a patient who had a recurrence within 6.05 months
   was defined as a clinically poor responder to chemotherapy. We found
   that 11 out of 12 (91.7%) patients with GC showed consistent results of
   drug response as observed in their derived PDOs ([149]Figure 6K;
   [150]Table S4). For example, patient G32T, who was diagnosed with
   locally advanced GC and received postoperative adjuvant chemotherapy
   with oxaliplatin and S-1, had been followed for 35 months with
   tumor-free survival. In keeping with the actual clinical response,
   organoids derived from G32T were also sensitive to combined therapy.
   Thus, drug responses in GC PDOs could be validated in their
   corresponding follow-up patients with GC, indicating that GC PDOs are
   useful tools for drug screening and prediction of drug response in
   patients.

The organoid model as a platform to investigate interaction between cells of
the tumor microenvironment and tumor cells

   To explore the possibility of co-culturing organoids with fibroblasts
   cells, we co-cultured GC organoids (G83T and G84T) and matched
   cancer-associated fibroblasts (CAFs). CAFs grew longer and gradually
   formed a net-like structure that encircled organoids after 3 days
   ([151]Figure 7A). On the other hand, co-culture with CAFs significantly
   enlarged the size of organoids after 5 days ([152]Figures 7B and
   [153]S8). We then examined the effects of CAFs on tumor organoids in
   response to 5-FU and oxaliplatin. Of note, autologous CAFs had a
   significant effect on enhancing the drug resistance to 5-FU and
   oxaliplatin in both organoids (G83T and G84T) ([154]Figure 7C).
   Consistently, organoids (G27T and G72T) with heterologous CAF
   co-culture also increased drug resistance compared with organoids
   alone. These results suggested the impact of CAFs on the organoids in
   response to chemotherapy ([155]Figure 7D).

Figure 7.

   [156]Figure 7
   [157]Open in a new tab

   The organoid model as a platform to investigate interaction between
   cells of the tumor microenvironment and tumor cells

   (A) Representative images of bright-field microscopy and
   immunofluorescence staining of CAFs, organoids, and co-cultures were
   immunostained using antibodies against human homologs to show alpha
   smooth muscle actin (α-SMA) (red) and CK7 (green), respectively. Nuclei
   are stained with DAPI (blue). Scale bar, 50 μm.

   (B) The effect of CAFs on organoid growth on day 0, 3, and 5.

   (C and D) Dose-response curves showing chemosensitivity of organoid and
   organoid with autologous or heterologous CAF co-culture to 2
   chemotherapeutics (5-FU and oxaliplatin) (n = 3).

   (E) 3D multispectral images of T cells (blue) and GC organoid (yellow)
   co-cultures at the indicated time points. Scale bar, 50 μm.

   Furthermore, to explore the interaction impacts of organoids and immune
   cells, we then co-cultured GC organoids derived from three patients
   with GC (G72T in high microsatellite instability [MSI-H], G83T and G85T
   in microsatellite stable [MSS] status) with autologous T cells. To
   activate and stimulate T cells, peripheral blood mononuclear cells
   (PBMCs) were seeded on an anti-CD28-coated plate together with
   autologous GC organoids dissociated into single cells in 20:1 ratio.
   After 2 weeks of co-culture and stimulation, the exposure of PBMCs to
   autologous tumor organoids led to a significant increase of
   interferon-gamma (IFN-γ) and TNF-α tumor-reactive T cells in G72T
   patients (MSI-H), a slight increase of IFN-γ and a significant increase
   of TNF-α tumor-reactive T cells in G83T patients (MSS), and a slight
   increase of IFN-γ and a decrease of TNF-α tumor-reactive T cells in
   G85T patients (MSS), as compared to control PBMCs ([158]Figure S9A).
   The patient-specific tumor-reactive T cells were therefore applied for
   co-culture assays to evaluate their tumor-killing effects in organoids.
   Collectively, these data suggest the feasibility of inducing
   patient-specific tumor-reactive T cell responses by co-culture of PBMCs
   and autologous tumor organoids. The addition of autologous T cells led
   to damaged structure and increased apoptosis of G72T organoids by
   immunofluorescence staining ([159]Figure 7E). Flow cytometry confirmed
   the increased apoptosis of tumor cells after co-culture with T cells
   ([160]Figure S9B). Such effect was not observed in organoids G83T and
   G85T in MSS statuses co-cultured with T cells ([161]Figures 7E and
   [162]S9B). These experiments allowed us to observe the interactions
   between cells in the tumor microenvironment and tumor cells, thereby
   gaining insights into how these interactions impact the response to
   treatment.

Discussion

   Compared to traditional cancer cell lines and patient-derived tumor
   xenograft models, PDO models preserve the heterogeneity and complexity
   of the original tumor, better reflecting an individual patient’s
   pathological and genetic characteristics.[163]^8 In our study, we
   established an organoid biobank of 57 GC organoid lines from surgically
   excised samples of patients with GC. Our PDOs could greatly preserve
   the genetic landscape of tissues where they were derived from, allowing
   accurate recapitulation of the actual GC condition. High-growth-rate
   PDOs were derived from patients with more advanced GC (TNM stage ≥3);
   hence, it showed significant upregulation of several well-acknowledged
   oncogenic pathways particularly NOTCH and MAPK signaling in these
   organoids.[164]^20 These results collectively indicate the potential to
   activate proliferation- and stemness-related genes and pathways to
   accelerate cell proliferation in GC organoids, thereby facilitating and
   maintaining their long-term cultivation.

   We found that most of the genes were reasonably well conserved in
   organoids, except for immune pathways mainly enriched in tissue and
   metabolic pathways enriched in organoids, which is consistent with
   previous reports.[165]^15^,[166]^21 We then assessed the phenotypic
   characteristics of GC PDOs and confirmed the consistency between
   organoids and their corresponding tumors tissues. In particular, our
   results showed that GC PDOs exhibit similar histological features and
   protein expression as the subtypes of their corresponding primary
   tumors. The diffuse type of GC is known to have high expression of
   CK7,[167]^22 whereas organoids derived from diffuse-type tumors also
   have upregulated CK7 expression. It is well acknowledged that drug
   response is greatly influenced by genomic alterations and tumor
   features. Hence, before utilizing organoids for drug screening, it is
   of importance to ensure that PDOs could accurately mimic the actual
   physiological condition by retaining the specific structure and
   function of their primary tumor tissues.

   RNA sequencing was performed to identify gene signatures in GC PDOs
   associated with chemosensitivity or chemoresistance. In organoids
   sensitive to chemotherapeutics, multiple well-acknowledged tumor
   suppressors were upregulated, including TP53INP1, MYO1A, and DMBT1, of
   which their abilities to inhibit GC and induce cell death were
   previously
   reported.[168]^23^,[169]^24^,[170]^25^,[171]^26^,[172]^27^,[173]^28 In
   organoids resistant to chemotherapeutics, several proliferation-
   and invasion-related genes were enriched (DARPP32 in 5-FU-resistant
   PDOs; L1CAM in oxaliplatin-resistant PDOs). DARPP32 could induce
   tumorigenesis and resistance of GC cells to gefitinib by promoting
   interaction between epidermal growth factor receptor and receptor
   tyrosine protein kinase erbB-3,[174]^29^,[175]^30 while L1CAM is a
   well-established marker associated with poor prognosis in patients with
   GC and it could accelerate GC progression and metastasis.[176]^31 Our
   analysis also identified the upregulation of PI3K and WNT signaling in
   chemoresistant PDOs. The dysregulation of WNT/β-catenin signaling
   pathway is critically associated with GC development and
   chemoresistance.[177]^32^,[178]^33 Collectively, the gene expression
   profiles of chemosensitive or chemoresistant organoids were found to be
   different, and such disparity may at least in part explain the diverse
   responses among GC PDOs toward chemotherapy.

   Given that gene expression signatures are different between
   chemosensitive and chemoresistant patients, we therefore evaluated
   whether these gene biomarkers could be used for the discrimination of
   chemotherapy response. Our results demonstrated that the top
   differentially expressed genes accurately discriminate between
   chemosensitivity and chemoresistance with AUC over 80%, a performance
   considered as clinically satisfactory.[179]^34 In practice, selecting
   appropriate therapeutics is crucial for patients; hence, it is of
   importance to develop a strategy that can accurately and effectively
   predict treatment outcomes in each patient. A previous study reported
   that intratumoral infiltration of immune cells including CD3^+ and
   CD8^+ cells is a signature for predicting chemotherapy response in
   patients with GC.[180]^35 Here, we instead revealed the feasibility of
   utilizing gene signatures to predict chemotherapy outcomes in patients
   with GC. Notably, further validation of these gene signatures on human
   samples and patients is necessary prior to clinical application.

   Furthermore, PDO models more closely mimic the in vivo tumor
   microenvironment, including cell-cell interactions and extracellular
   matrix, and can be co-cultured with immune cells, fibroblasts, and
   other cell types. Hence, it provides a more reliable platform for drug
   screening.[181]^10 In our study, we demonstrated that GC PDOs are
   effective tools for drug screening with varied responses to different
   conventional chemotherapeutics and combination treatment with multiple
   drugs.

   To validate the drug screening results of PDOs, we established a PDOX
   mouse model using organoids derived from five patients with GC. As
   expected, consistent responses to chemotherapeutic drugs were observed
   between PDOX mice and GC PDOs, indicating that in vitro organoid
   cultures could model in vivo drug responses. We then testified the drug
   screening results of PDOs in patients with GC. Over 90% of patients
   with GC showed consistent results of drug response as observed in their
   derived PDOs, suggesting the potential of PDOs as tools for predicting
   drug response in patients with GC. These findings are in line with
   previous studies using PDOs to predict drug responses in patients with
   GC, but with smaller sample sizes.[182]^11^,[183]^12^,[184]^13
   Moreover, each organoid screening process took less than 2 weeks. This
   quick turnaround time undoubtedly enables rapid testing in real
   clinical scenarios. Importantly, we are able to establish organoids
   using GC biopsies by gastrointestinal endoscopy or puncture. Organoids
   can assist in identifying potential and appropriate drugs, thereby
   preventing overtreatment and minimizing treatment-related side effects.
   Currently, it is an effective tool to be instrumental for those
   patients.

   The interaction between cells of the tumor microenvironment and tumor
   cells is crucial to understand (and predict) response to treatment.
   Previous studies have suggested that CAFs secrete a variety of growth
   factors to regulate cancer development, such as transforming growth
   factor β, epidermal growth factors, and fibroblast growth
   factors.[185]^36 Here, we established an organoid-CAF-co-culture
   system, and found that the CAFs promoted the growth of organoids and
   drug resistance. The system, which could provide more predictive
   models, demonstrates that the direct contact between CAFs and cancer
   cells plays a critical role in tumor progression. Furthermore, we also
   performed co-culture assays of autologous T cells and organoids and
   showed that the organoids from patients with MSI-H are more likely to
   be killed by autologous T cells, demonstrating that tumor organoids can
   be used to establish personalized model systems to support T cell-based
   therapies, which is consistent with previous reports.[186]^37

   In conclusion, we established a collection of patient-derived GC
   organoids that retain the structural and functional characteristics of
   their originated tumors. These organoids reflected the actual response
   to chemotherapy in patients, indicating that they could be utilized for
   the prediction of treatment response in individual patients. Assessing
   chemosensitivity in organoids can be used as a robust tool for the
   personalized selection of chemotherapeutic drugs in patients with GC.

Limitations of the study

   A prospective study involving a large cohort of GC organoids is
   necessary before considering implementation in clinical cancer care.
   The clinical implications and translational potential of our findings
   necessitate further mechanistic studies and validation in larger
   patient cohorts.

STAR★Methods

Key resources table

   REAGENT or RESOURCE SOURCE IDENTIFIER
   Antibodies
     __________________________________________________________________

   Anti-Cytokeratin 7 Abcam Cat#ab9021, RRID: [187]AB_306947
   Anti-Carcino Embryonic Antigen CEA antibody Abcam Cat#ab207718, RRID:
   [188]AB_3086740
   Anti-Prostate Secretory Protein/PSP antibody Abcam Cat#ab128897, RRID:
   [189]AB_11150619
   S1P4/EDG-6 Antibody Novus Biologicals Cat# NBP2-24500, RRID:
   [190]AB_3086741
   FKBP10/FKBP65 Polyclonal antibody Proteintech Cat#12172-1-AP, RRID:
   [191]AB_2102550
   Anti-MYO1A Rabbit Polyclonal Antibody Proteintech Cat#17499-1-AP, RRID:
   [192]AB_2148172
   NDUFA4L2 Polyclonal antibody Proteintech Cat#16480-1-AP, RRID:
   [193]AB_2150637
   Anti-Ki67 Cell Signaling Technology Cat#12202S, RRID: [194]AB_2620142
   Anti-alpha smooth muscle Actin antibody Abcam Cat#ab124964, RRID:
   [195]AB_11129103
   Anti-CD28 Mouse Monoclonal Antibody BioLegend Cat#302934, RRID:
   [196]AB_11148949
   BUV395 Mouse Anti-Human CD3 BD Biosciences Cat#563546, RRID:
   [197]AB_2744387
   BV510 Mouse Anti-Human CD45 BD Biosciences Cat#563204, RRID:
   [198]AB_2738067
   APC Mouse Anti-Human CD8 BD Biosciences Cat#566852, RRID:
   [199]AB_2869906
   PE-CY7 Mouse Anti-Human TNF-a BD Biosciences Cat#557647, RRID:
   [200]AB_396764
   BV605 Mouse Anti-Human IFN-γ BD Biosciences Cat#562974, RRID:
   [201]AB_2737926
     __________________________________________________________________

   Biological samples
     __________________________________________________________________

   Human gastric cancer sample First Affiliated Hospital of Sun Yat-sen
   University N/A
     __________________________________________________________________

   Chemicals, peptides, and recombinant proteins
     __________________________________________________________________

   Cisplatin Selleck Cat#S1166
   Oxaliplatin Selleck Cat#S1224
   5-Fluorouracil Selleck Cat#S1209
   SN-38 Selleck Cat#S4908
   Paclitaxel Selleck Cat#S1150
   Doxorubicin Selleck Cat#S1208
   B-27™ Supplement, serum free Thermo Fisher Scientific Cat#17504044
   GlutaMAX Gibco Cat#35050061
   HEPES Gibco Cat#15630080
   Pen/Strep Gibco Cat#15140122
   N-acetyl-L-cysteine Sigma Cat#A9165-5G
   Gastrin-1, human Genscript Cat#PR12740
   Recombinant Human EGF Peprotech Cat#AF-100-15
   Recombinant Human FGF-10 Peprotech Cat#100-26
   A83-01 TOCRIS Cat#2939
   Y-27632 Selleck Cat#S1049
   Recombinant Human Noggin Peprotech Cat#120-10C
   Primocin Invivogen Cat#ant-pm-1
   NucRed Dead 647 Thermo Fisher Scientific Cat#[202]R37113
   Oregon Green™ 488 BAPTA-1 Thermo Fisher Scientific Cat#O6807
   eBioscience Cell Proliferation Dye eFluor 450 Thermo Fisher Scientific
   Cat#65-0842-90
   CellEvent™ Caspase-3/7 Green Thermo Fisher Scientific Cat#[203]C10423
   Matrigel Corning Cat#356231
   Advanced DMEM/F-12 Gibco Cat#12634010
   RPMI 1640 Gibco Cat#11875168
   Collagenase type IV Gibco Cat#17104019
   dispase type II Gibco Cat#17105041
   TrypLE Express Gibco Cat#12605010
   Fibroblast culture medium iCELL Cat#PriMed-iCELL-003
   Stain Buffer BD Pharmingen™ Cat#554656
     __________________________________________________________________

   Critical commercial assays
     __________________________________________________________________

   TUNEL apoptosis detection kit KeyGen BioTECH Cat#KGA700
   CellTiter-Glo assay Promega Cat#G7570
   QIAamp® DNA Mini Kit Qiagen Cat#51304
   RNeasy Mini Kit QIAEN Cat#74104
     __________________________________________________________________

   Deposited data
     __________________________________________________________________

   Raw and analyzed data This paper GSA: HRA006195
     __________________________________________________________________

   Experimental models: Organisms/strains
     __________________________________________________________________

   NCG (NOD-Prkdc^em26Cd52Il2rg^em26Cd22/Nju) mice GemPharmatech, Nanjing,
   China Cat#T001475
     __________________________________________________________________

   Oligonucleotides
     __________________________________________________________________

   MSMB sgRNA-1: AACACCCAATAAACTCGGAG
   sgRNA-2: ATTGCATAAAGTCACGAAGG This paper N/A
   S1PR4 sgRNA-1: CGGCCATCACCAGCCACATG
   sgRNA-2: TGTAGTGCAGAACAATGAGC This paper N/A
   FKBP10 sgRNA-1: GGCACTTATGACACCTACGT
   sgRNA-2:
   TGGAAGATGTGGTCATCGAG This paper N/A
   MYO1A sgRNA-1: ACATTTACTCACAAATCGGG
   sgRNA-2: TGTGGCGTACCAGTCACTGA This paper N/A
   NDUFA4L2 sgRNA-1: GCAGATCAAAAGACATCCGG
   sgRNA-2: TATGGCAGGAGCCAGTCTTG This paper N/A
     __________________________________________________________________

   Software and algorithms
     __________________________________________________________________

   GraphPad 10.0 GraphPad Software N/A
   ImageJ NIH [204]https://imagej.nih.gov/ij/
   [205]Open in a new tab

Resource availability

Lead contact

   Further information and requests for resources and reagents should be
   directed to and will be fulfilled by the lead contact, Prof. Li
   (lixiaox23@mail.sysu.edu.cn).

Materials availability

   This study did not generate new unique reagents.

Data and code availability

   The sequencing data reported in this paper are deposited in Genome
   Sequence Archive (GSA) with project number HRA006195
   ([206]https://ngdc.cncb.ac.cn/gsa-human/browse/HRA006195) and will be
   publicly available as of the date of publication. There are no
   restrictions on data access for academic use. Requests to access data
   should follow the GSA’s "Data Access Request Guidance" available at
   [207]https://ngdc.cncb.ac.cn/gsa-human/document. The software and
   algorithms for data analyses used in this study are published and
   referenced throughout the [208]STAR Methods section. No original code
   is featured in this paper. Any additional information required to
   reanalyze the data reported in this paper is available from the
   [209]lead contact upon request.

Experimental model and study participant details

Human gastric cancer tissues

   Fresh tumor tissues were obtained from 73 patients with GC receiving
   surgical resection ([210]Table S1) at the First Affiliated Hospital of
   Sun Yat-sen University, Guangzhou, China, from 2019 to 2023. This study
   was approved by the Clinical Research Ethics Committee of the First
   Affiliated Hospital of Sun Yat-sen University and conducted in
   accordance with the Declaration of Helsinki (No. [2020]052). Written
   informed consent was obtained from all patients.

PDOX models and chemotherapy in NCG mice

   Male NCG (NOD-Prkdc^em26Cd52Il2rg^em26Cd22/Nju) mice were purchased
   from Gempharmatech Co., Ltd. (Jiangsu, China). For PDOX, digested GC
   organoids (G5T-PDO, G9T-PDO, G27T-PDO, G30T-PDO, G72T-PDO) were
   injected subcutaneously into the right dorsal flank of 5-week-old male
   NCG mice (1× 10^6 cells in 100 μL Matrigel/PBS per mouse). Tumor size
   was measured using a digital caliper every third day, and tumor volume
   (mm^3) was calculated using the formula: volume =
   length × width^2 × 0.5. When the tumor size reached 50 mm^3, mice were
   randomly divided into sensitive, resistant or control groups. Drugs
   were administered by intraperitoneal injection, including 5-FU
   (25 mg/kg in 0.9% NaCl, twice a week), oxaliplatin (6 mg/kg in 5%
   glucose solution, once a week), paclitaxel (10 mg/kg in 0.9% NaCl, once
   a week), and SN-38 (15 mg/kg in 0.9% NaCl, twice a week). Mice were
   sacrificed after treatment, and fresh tumors were harvested for
   subsequent analysis. All experiments were conducted in accordance with
   the guidelines approved by the Animal Ethics Committee of the First
   Affiliated Hospital of Sun Yat-sen University.

Method details

Establishment of organoid cultures

   Freshly collected tissues were immediately placed in tissue storage
   solution (MACS). Digested tumor tissues were filtered through a 70 μm
   filter and centrifugated at 300×g. The cell pellet was resuspended in
   Matrigel after washing, and a volume of 30 μL Matrigel-cell mixture was
   seeded per well on a prewarmed 48-well plate. After Matrigel was
   solidified, 250 μL of complete GC organoid medium was added which
   consisted of advanced DMEM/F12 (Gibco, 12634010), 1x GlutaMax, 1x
   HEPES, 40% Wnt3a, 10% RSPO-1, 100 ng/mL Noggin (PeproTech, 120-10C), 1x
   B27 (Thermo Fisher Scientific, 17504044), 50 ng/mL EGF (PeproTech,
   AF-100-15), and 100 ng/mL FGF10 (PeproTech, 100-26), 1mM
   N-Acetylcystenine, 1nM Gastrin, 1μM A83-01, 10 μM Y-27632 and 1μg/ul
   primocin. The culture medium was replenished every 3–4 days, and
   organoids were passaged every 1–2 weeks using TrypLE Express (Gibco).

Histology, immunohistochemistry and immunofluorescence

   Tissues and organoids were fixed in 4% paraformaldehyde and they were
   sectioned at a thickness of 4 μm after processing. Hematoxylin and
   eosin staining was conducted using the standard histological protocol.
   Immunohistochemistry staining was performed using anti-CK7 antibody
   (1:1000, Abcam, ab9021), anti-CEA antibody (1:20000, Abcam, ab207718),
   and anti-Ki67 antibody (1:400, Cell Signaling Technology, 12202S).
   According to the manufacturer’s instructions, TUNEL (terminal
   deoxynucleotidyl transferase nick-end labeling) staining was performed
   using an apoptosis detection kit (KeyGen BioTECH, KGA700).
   Immunofluorescence staining was performed using anti-CK7 antibody, and
   α-SMA (1:1000, ab124964, Abcam). To quantify staining, five random
   areas per specimen were captured under microscope with 40×
   magnification, and the percentage of positively stained cells was
   determined using ImageJ software.

Drug screening and cell viability assay

   GC organoids were harvested and digested into single cells and counted
   with 0.4% Trypan Blue solution. Organoids were resuspended in advanced
   DMEM/F12 and Matrigel at a ratio of 1:2 and embedded in 96-well plates
   (250 cells per 5 μL suspension per well). After Matrigel was
   solidified, 100 μL of complete medium was added with incubation for 24
   h. The complete medium was then replaced with culture medium with
   different concentrations of chemotherapy drugs, including 5-FU,
   oxaliplatin, cisplatin, paclitaxel, doxorubicin, and irinotecan active
   metabolite SN-38 (Selleck, Washington). 10 μM phenylarsine oxide was
   used as positive control, and dimethyl sulfoxide was used as negative
   control. Each drug concentration was performed in triplicate. For
   combined drug treatments, organoids were treated as follows: 5-FU:
   oxaliplatin, 25:1. Cell viability was assessed using CellTiter-Glo
   Luminescent Cell Viability Assay (Promega). After 6 days of drug
   incubation, the luminescence of each well was measured. There were some
   reports that drug sensitivity was classified by both IC50 and area
   under the curve (AUC), but these studies focused on relative
   sensitivity and relative resistance, without a specific cut-off
   value.[211]^38^,[212]^39 In this study, we mainly used the AUC value to
   classify the organoids as resistant (AUC >50%) or sensitive (AUC <50%).

Whole-exome sequencing and analysis

   DNA was extracted from paired tumor and adjacent non-tumorous gastric
   epithelium mucosa as well as PDOs using QIAamp DNA Mini Kit (Qiagen,
   Germantown, MD). Whole-exome sequencing was performed using HiSeq X TEN
   platform (Illumina), and the sequencing mean depths were 300x.

   Sequencing reads were mapped against human genome build 37 (hg19) by
   Burrows-Wheeler Aligner (version 0.7.10). The BAM files were further
   processed in terms of duplicate marking using samtools (version
   0.1.19). Local realignment and base recalibration were performed by
   Genome Analysis Toolkits (GATK v3.2-2). Single-nucleotide polymorphisms
   and small insertions/deletions were identified by providing the
   peri-tumor tissues as a reference and their corresponding tumor or
   organoid sequencing data to MuTect2 (involved in GATK v3.2-2) with
   default parameters. Filters for Next Generation Sequencing (FiNGS)
   software was used to generate VCF files with high quality by applying
   the filtering criteria of International Cancer Genome Consortium (ICGC)
   to the remaining variants. Annovar was then used to annotate somatic
   variants with the following databases: gnomAD, dbSNP, ICGC, Catalog of
   Somatic Mutations in Cancer (COSMIC), ClinVar, dbNSFP and dbscSNV.
   Mutational signatures were visualized using maftools v2.4.12
   ([213]https://bioconductor.org/packages/release/bioc/vignettes/maftools
   /inst/doc/maftools.html).

RNA-sequencing and analysis

   Total RNA was extracted from tumor organoids and tissue using RNeasy
   Mini Kit (QIAEN, 74104) following the manufacturer’s instructions. The
   amount of extracted RNA was quantified using Agilent 5400 Bioanalyzer.
   RNA-sequencing was performed using HiSeq X TEN platform (Illumina), and
   150-bp paired-end reads were generated. RNA-sequencing data were
   processed to filter out low quality reads and trim all reads in front
   and tail using Fastp. The remaining reads were aligned to the human
   reference genome Hg19 using HISAT2. Gene expression levels were
   quantified by RNASeQC2 to generate fragments per kilobase of exon model
   per million mapped fragments (FPKM) and read counts, which were used
   for pathway enrichment analysis and differential gene expression
   analysis, respectively. Stemness score was generated according to the
   reported method.[214]^40 We use the mRNA-based stemness index (mRNAsi)
   to evaluate the dedifferentiation potential of tumor cells. The
   stemness score is defined as the Spearman correlation coefficient
   between the gene expression profile of our samples and the reported
   expression profile of stem cell features in the article mentioned
   above. The score ranges between 0 and 1, with a higher value indicating
   a closer similarity to stem cell expression features, implying a higher
   degree of stemness. We observed a significant difference between the
   fast-growing group and the slow-growing group (p = 0.023), with an
   average score of 0.784 for the fast-growing group and 0.345 for the
   slow-growing group. We use Spearman correlation coefficient (ρ) and R
   square (R^2) to access the gene expression level (in log2 FPKM) between
   organoids and their corresponding primary tumor tissues. Spearman
   correlation coefficient (ρ) was calculated by "cor" function with
   parameter "method = spearman" in R package stats (v4.2.0) and R square
   (R^2) was computed by "lm" function in R package stats (v4.2.0).
   According to the results of in vitro drug screening, organoids were
   separated into either sensitive group (AUC <50%) or resistant group
   (AUC ≥50%). Significantly differentially expressed genes between the
   two groups were filtered by DESeq2 package in R with read counts
   (p < 0.05, log2-fold change ≥1 or ≤ −1) and visualized as volcano plots
   by R ggplot2. KEGG Orthology Based Annotation System (KOBAS)
   ([215]http://kobas.cbi.pku.edu.cn/anno_iden.php) was applied to FPKM
   expression data. Analysis of enriched pathways was annotated using KEGG
   and Reactome. Recursive feature elimination (RFE) algorithm was then
   performed for feature ranking and selection. Logistic regression model
   was performed with LogisticRegression module in Scikit-learn20 (v
   0.24.2) python package. Naive Bayes model was performed using e1071 (v
   1.7–9, [216]https://cran.r-project.org/package=e1071) R package.

CRISPR-Cas9 knockout of gene

   CRISPR-Cas9-mediated knockout was performed with lentiCRISPRv2 system.
   Briefly, guide RNAs targeting the human genomic locus of MSMB, S1PR4,
   FKBP10, MYO1A, NDUFA4L2 (sg #1 and #2) were obtained from GenScript and
   cloned into the lentiCRISPRv2 vector. Lentivirus were then generated by
   co-transfection of lentiCRISPRv2 vector and packaging plasmids in 293T
   cells. To perform knockout, GC organoids were harvested and digested
   into single cells and counted with 0.4% Trypan Blue solution, 2x10^5
   cells were resuspended in medium containing Rho kinase inhibitor Y27632
   (10 μM) and Polybrene (8 μg/mL) and CRISPR-Cas9-gRNA lentivirus
   spinfected (500g, 1 h, 32°C). After 3–5 h of incubation at 37°C, cells
   were seeded in Matrigel. Puromycin selection was started 3 days after
   transduction. Puromycin-resistant organoids were isolated and confirmed
   by western blot.

Co-culture of tumor organoids and CAFs

   Human CAFs were isolated from gastric cancer tissues. Tumor tissue was
   digested with a digestion solution: collagenase type IV (0.25 mg/mL,
   Gibco), dispase (0.125 mg/mL, Gibco) in Advanced DMEM/F12 (Gibco) for
   60 min. Digested tumor tissues were filtered through a 70 μm filter and
   centrifugated at 300 × g. The pellet was plated in a T25 flask and
   fibroblast was allowed to grow in fibroblast culture medium
   (PriMed-iCELL-003, iCELL). The fibroblasts were checked by using
   immunofluorescence staining of α-SMA (1:1000, ab124964, Abcam).

   GC organoids were harvested and digested into single cells and counted
   with 0.4% Trypan Blue solution. CAFs used for experimental studies at
   passages 4–6. For co-cultures, CAFs and organoids at a ratio of 1:2
   embedded in Matrigel. After Matrigel was solidified, complete medium
   was added. After co-culturing organoids with CAFs for 5 days, the
   organoids image was captured and diameters of was measured using ImageJ
   software on day 0, day 3, day 5, respectively. For drug response,
   organoids with or without CAFs was treated with 5-FU and oxaliplatin as
   described above.

T cells and organoid co-culture assays

   Fresh peripheral blood was collected with informed consent. The
   peripheral blood mononuclear cells (PBMC) fraction was isolated from
   peripheral blood by Ficoll-Paque PLUS density gradient separation,
   counted and cryopreserved for T cell activation assays. To expand and
   activate T cells, PBMCs were seeded on an anti-CD28-coated plate
   together with autologous GC organoids dissociated into single cells in
   20:1 ratio. T cell culture Medium was refreshed every 2 days, which was
   composed of RPMI 1640 medium supplemented with 1%
   penicillin/streptomycin, 1% Ultraglutamine I, 1% HEPES, 50 nM
   β-Mercaptoethanol, 10% fetal bovine serum and 200 U/mL IL-2. After
   2 weeks of expansion, T cells were ready for coculture assays to
   evaluate tumor killing effect.

   Organoids for T cell coculture were released from Matrigel by cold PBS
   and resuspended in GC culture medium. The suspensions were filtered
   with a 100μm strainer to remove organoids in large size. Activated
   T cells were cocultured with GC organoids at a ratio of 5:1 for 18 h.
   Cells were plated in a 96-well, glass-bottom, high content screening
   microplates (Corning) in 200 μL medium containing 100 μL T cell culture
   medium and 100 μL GC organoid culture medium. NucRed Dead 647 (two
   drops per milliliter, Thermo Fisher Scientific) was added for labeling
   of dead cells. To facilitate visualization, GC organoids were
   pre-stained with Oregon Green 488 BAPTA-1 (Thermo Fisher Scientific)
   and T cells with eBioscience Cell Proliferation Dye eFluor 450 (Thermo
   Fisher Scientific) in RPMI 1640 for 20 min at 37°C, then washed with
   completed medium for twice. Living cell imaging were performed at 0h
   and 18h after cell seeding using laser scanning confocal microscope
   (Olympus FV3000).

Flow cytometry

   GC organoids cultured with or without T cells were incubated with
   CellEvent Caspase-3/7 Green (Thermo Fisher Scientific) to label
   apoptotic cells and evaluate killing effect by T cells. Organoids were
   dissociated into single cells using TrypLE Express and washed in stain
   Buffer (BD Pharmingen, 554656). The cells were then filtered with 70 μm
   cell strainers and stained with BUV395 Mouse Anti-Human CD3 (BD
   Biosciences), BV510 Mouse Anti-Human CD45 (BD Biosciences), APC Mouse
   Anti-Human CD8 (BD Biosciences), PE-CY7 Mouse Anti-Human TNF-α (BD
   Biosciences), BV605 Mouse Anti-Human IFN-γ (BD Biosciences). After
   washed twice with stain buffer, cells were detected by BD FACS
   LSRFortessaTM instrument. Data was analyzed using FlowJo V10.

Quantification and statistical analysis

   Data analyses were performed by GraphPad Prism 10 (GraphPad Software
   Inc., San Diego, CA) and R software. Unpaired two-tailed Student’s
   t-tests were used to compare two groups. Differences among three or
   more groups were compared by Analysis of variance (ANOVA). The results
   are presented as mean ± standard deviation, and p value less than 0.05
   was considered significant.

Acknowledgments