Modeling lung adenocarcinoma metastases using patient-derived organoids

Summary

Approximately 50% of patients with surgically resected early-stage lung cancer develop distant metastasis. At present, there is no in vivo metastasis model to investigate the biology of human lung cancer metastases. Using well-characterized lung adenocarcinoma (LUAD) patient-derived organoids (PDOs), we establish an in vivo metastasis model that preserves the biologic features of human metastases. Results of whole-genome and RNA sequencing establish that our in vivo PDO metastasis model can be used to study clonality and tumor evolution and to identify biomarkers related to organotropism. Investigation of the response of KRAS^G12C PDOs to sotorasib demonstrates that the model can examine the efficacy of treatments to suppress metastasis and identify mechanisms of drug resistance. Finally, our PDO model cocultured with autologous peripheral blood mononuclear cells can potentially be used to determine the optimal immune-priming strategy for individual patients with LUAD.

Graphical abstract

Highlights

• •Specific clinicopathologic features of the primary LUAD tumor affect PDO generation
• •Our in vivo PDO metastasis model is a tool for the study of tumor evolution
• •Our in vivo PDO metastasis model reveals mechanisms of resistance to targeted therapy
• •Coculture of PDOs and PBMCs can identify personalized immune-priming strategies

Liu et al. illustrate multiple applications of an in vivo lung adenocarcinoma metastasis model using patient-derived organoids. Applications include the elucidation of the biology of metastases at an individual patient level, discernment of mechanisms of resistance to targeted therapies, and dissection of immune-priming strategies in patient-derived organoids treated with immunotherapy.

Introduction

Non-small cell lung cancer (NSCLC) remains the leading cause of cancer-related death worldwide.¹ The high mortality of NSCLC is primarily related to the development of metastatic disease.² Lung adenocarcinoma (LUAD), the most common histologic type of NSCLC, has a high tumor mutation burden and well-annotated oncogenic drivers, some of which are targetable.^3,4 An important advance in modeling human cancer during the last decade has been the development of patient-derived organoids (PDOs). Compared with immortalized cancer cells, PDOs are superior in recapitulating the histologic, genetic, and transcriptomic characteristics of the original tumors. These features make PDOs a promising tool for drug development, drug screening, and precision medicine.^5,6 Recently, investigators have used organoids from human bronchial cells and NSCLC to study lung development and cancer phenotypes and to assess in vitro therapeutic vulnerabilities and resistance.^7,8,9 Although lung cancer patient-derived xenografts (PDXs) have been shown to predict response to therapy and to help discern mechanisms of drug resistance, LUAD PDX tumors rarely metastasize; thus, such models afford only limited assessment of the biology of metastases. There remains a gap in our knowledge of whether an in vivo PDO platform could potentially help to elucidate the biology of metastases and dissect mechanisms of drug resistance and immune-priming strategies in LUAD.

We initiated an LUAD PDO platform (Figure S1A) using surgically resected primary tumors, malignant pleural effusions (MPEs), and metastatic LUAD biospecimens. By assessing clinicopathologic features of the primary tumors, we uncovered factors that improve upon the previously reported low success rates of LUAD PDO generation.¹⁰ We then performed RNA sequencing (RNA-seq) and whole-genome sequencing (WGS) on PDOs, their parental tumors, and mouse metastasis-derived organoids (MDOs) developed in our in vivo PDO metastasis model to assess clonality and tumor evolution and to explore potential biomarkers of metastatic organotropism. Finally, we assessed response to therapy and mechanisms of resistance to targeted therapies in vitro and in vivo and examined the immune-priming effects of either chemotherapy or radiation with immune checkpoint inhibitor (ICI) therapy in LUAD PDOs.

Results

Generation and characterization of LUAD PDOs

We initiated a PDO program using human LUAD specimens, including surgically resected primary tumors (n = 32), metastatic tumors (n = 12), and MPEs (n = 5) (Table S1). We chose a bespoke Sanger sequencing screen (Figure S1B) on the basis of the results of MSK-IMPACT¹¹ to identify PDOs. The success rate of generation of PDOs from metastatic tumors and MPEs was 92% and 100%, respectively. However, similar to others,^10,12 the success rate of generation of PDOs from primary LUAD was only 31% (10/32) (Figure 1A). We identified clinicopathologic features positively associated with the successful establishment of PDOs from primary LUAD, including tumor content >30%, predominant solid (SOL) histologic subtype, poor differentiation, tumor maximum standardized uptake value ≥5 on fluorodeoxyglucose-positron emission tomography, and SOL tumor morphologic appearance on computed tomography (Figures 1B and S1C–S1E).

Figure 1.

Establishment and characterization of PDOs

(A) Bars showing the success rate of generation of PDOs from different LUAD biospecimens. “Yes,” preservation of the major mutations of the parental tumor; “No,” major mutations in parental tumor are not preserved in the PDOs.

(B) Bars indicating the success of generation of PDOs from the primary LUAD tumor with indicated clinicopathologic features. SUVmax, maximum standardized uptake value; “Yes” and “No,” as in (A).

(C) H&E (rows 1 and 3), phase contrast (row 2), and Ki-67 (row 4) immunostaining of the primary tumor, noncancerous tissue, and matched PDOs.

(D) Immunostaining of Sox2 in PDOs and their parental tumors.

(E) Correlation heatmap of variant allele fractions across PDOs and their parental primary tumors.

(F) Proportion of substitution context across PDOs and their parental primary tumors.

(G) Correlation heatmap of the transcriptomic profiles across PDOs and their parental primary tumors.

The epithelial origin of PDOs was confirmed by use of immunofluorescence for epithelial cell adhesion molecule (EpCAM) (Figure S2A). PDOs exhibited similar tumor morphologic appearances, proliferation rates (Ki-67)¹³ (Figure 1C), and stemness features (Sox2 [Figure 1D] and ALDH1A1 [Figure S2B]) as their parental tumors.

To confirm that PDOs retain the genomic landscapes of their parental tumors, WGS was performed in PDO_6 and PDO_23 sample sets (Table S2), including (1) PDOs with short-term (early, <2 months) and long-term (late, >6 months) culturing, (2) their parental tumors, and (3) their matched adjacent noncancerous tissues. Comparisons of variant allele fractions (Figure 1E), the distributional profile of single-nucleotide variants (SNVs) (Figure 1F), and junction copy-number¹⁴ changes (Figure S2C) showed strong concordances between PDOs and their parental tumors. RNA-seq revealed that the transcriptomic profile of the PDO is highly correlated with that of its parental tumor (Figure 1G). Additional gene set enrichment analyses identified a signature of genes distinctly expressed in PDOs, compared with their parental tumors, including negative enrichment for gene sets corresponding to cell adhesion and extracellular matrix-receptor interaction and positive enrichment for gene sets corresponding to cell cycle and multiple metabolic-related pathways (Figures S2D–S2E). These differences may be related to different microenvironments between PDOs in culture and their parental tumors.¹⁵ Collectively, these data demonstrate that our PDO system faithfully recapitulates and maintains the histologic, genomic, and transcriptomic profiles present in parental tumors.

Establishment of our in vivo PDO model of metastasis

To mimic LUAD systemic metastases,¹⁶ we used an intracardiac injection in vivo metastatic model¹⁷ to examine the metastatic capacity of PDOs (Figure 2A). PDO_6 and PDO_23 were derived from pathologic stage IA primary LUADs with predominant SOL histologic subtype (Figure S3A). Surveillance in vivo imaging system (IVIS) demonstrated the development of metastasis in 7 of 11 mice injected with PDO_6 (64%) and in 4 of 8 mice injected with PDO_23 (50%) (Figures 2B–2D). Interestingly, we observed distinct organotropism between 2 PDOs. Liver metastases were observed in almost all PDO_6 mice that developed metastases. However, metastases did not appear to exhibit an organotropic preference in PDO_23 mice that developed metastases (Figure 2C). Results of H&E and immunohistochemistry (IHC) staining showed that MDOs recapitulated the features of their parental PDOs, including morphologic appearance, proliferation, and stemness (Figure 2E). In summary, the aforementioned findings serve as strong evidence that PDOs form metastases and can be used to study the biology of LUAD metastasis.

Figure 2.

Establishment of our in vivo PDO metastasis model

(A) Schematic representation of the generation of our in vivo PDO metastasis model: (1) collection and dissociation of surgical LUAD specimens, (2) generation and characterization of PDOs, (3) labeling with luciferase-green fluorescence protein (GFP) (PDO_GL), (4) introduction into mice by intracardiac injection, (5) formation of metastatic site(s), (6) isolation of metastatic tumor and regeneration of organoids (MDOs), and (7) characterization of MDOs.

(B) Bioluminescent images of mice injected with indicated PDOs.

(C) Table showing organs in which metastases formed in mice after intracardiac injection of PDOs at 5 months. GB, gallbladder; LN, lymph node.

(D) Representative H&E staining from healthy mice or mice with intracardiac injection of PDOs.

(E) H&E and immunohistochemical staining of PDO_GL and MDOs.

Using our in vivo PDO metastasis model to explore tumor evolution

To explore tumor evolution, we performed WGS on the sample sets of PDO_6 and PDO_23 (Table S2), and phylogenetic trees were generated. The majority of alterations and structural variations (SVs) were shared among tumors, PDOs, and MDOs, indicating that the tumor genotypes were largely preserved through the metastatic process in mice (Figure 3A). We also observed that some alterations and SVs were uniquely present in PDOs and/or MDOs: ARID1A and MET mutations^18,19 and SVs (including rigma and chromoplexy) in the PDO_23 sample set and PI3KCA mutation²⁰ and CDKN2A homozygous deletion²¹ in the PDO_6 sample set (Figures 3A and S3B). Interestingly, upregulation of the PI3K-AKT pathway was found in PDO_6s, compared with PDO_23s (Figure S3C). Alteration of CDKN2A and/or PI3KCA, as well as upregulation of the PI3K-AKT pathway, may contribute to the organotropic liver metastases of PDO_6 (Figure 2C).^22,23 In addition, the Catalogue of Somatic Mutations in Cancer (COSMIC) mutational signature analyses indicated an agreement of mutational signature events at the genome-wide level in both sample sets (Figure 3A). Tobacco and aging (https://cancer.sanger.ac.uk/cosmic/signatures_v2.tt) were present in both sample sets and were correlated with age and heavy smoking history. We observed that the tobacco signature was uniquely present in selected PDOs and MDOs (Figure 3A). These findings suggest that these alterations could be (1) gained during PDO culturing, (2) gained during in vivo metastasis, or (3) attributable to a growth advantage of a select group of tumor cells.

Figure 3.

Characterization of our in vivo PDO metastasis model

(A) Phylogenetic trees of the sample sets of PDO_6 and PDO_23. Pie charts show the mutational signature of aggregate samples (along a branch) or an individual sample (at the end of a branch). SV, structure variation; TMB, tumor mutation burden.

(B) Schematic of brain metastasis markers. Venn diagram overlapping the differentially expressed genes in (1) PDOs derived from human brain metastases (n = 3), compared with PDOs from human primary tumors (n = 8), and (2) MDOs from mouse brain metastases (n = 2), compared with their parental PDOs (n = 3).

(C) Expressional heatmap of 10 overlapping genes of the putative “Brain Met. Markers” (shown in B). Green brackets, altered in all PDOs and MDOs from brain metastases, compared with PDOs from the primary tumor.

(D) Table summarizing the prediction of outcomes of 10 overlapping genes. The table details the p values obtained (OS and PFS, log rank test). OS, overall survival; PFS, progression-free survival.

(E) Bars indicating BTG2 mRNA in LUAD that metastasized to the brain (n = 6), compared with LUAD that metastasized to other organs (n = 4). Data are represented as mean ± SEM.

(F–I) LUAD H460 cells were transfected with short hairpin RNA (shRNA) BTG2 or scramble (Sc.). (F) Immunoblots showing BTG2 and actin in H460 cells. (G) Bioluminescent images at 11 days after injection with H460 cells. (H) Curves indicating the average total bioluminescence (log region of interest [ROI]) after injection. Data are represented as mean ± SEM. (I) Bioluminescent brain images at 11 days (left) after injection. Curves indicate average bioluminescence (log ROI) in the brain (right). Data are represented as mean ± SEM.

A model to identify potential markers associated with brain metastases

RNA-seq showed a high concordance of transcriptomic profiles among PDOs and MDOs, illustrating that MDOs largely preserve the transcriptomic features of their parental PDOs (Figure S3D). To explore whether our PDO metastatic model could serve as a tool for identification of genes related to LUAD metastases, we compared the RNA-seq data of 2 pools: (1) PDO_6 and PDO_23 versus their MDO_brains (Table S2) and (2) PDOs from human primary tumors versus PDOs from human brain metastases. We posited that the analysis of MDOs or PDOs would help identify cancer cell-intrinsic drivers of LUAD metastases, unlike tumor tissue comparisons that are also driven by stromal differences. These analyses identified 10 “brain metastasis marker” genes that overlapped in both pools (Figure 3B).

Subsequent heatmap analysis indicated that 6 of the 10 genes were consistently overexpressed (TFPI2, IL1B, SAA1, and SAA2) or underexpressed (BTG2 and FGFBP1) in metastases, relative to primaries, in both in vivo PDO metastasis models and PDOs directly derived from human tumors (Figure 3C). We next performed in-depth analyses using 2 LUAD datasets: The Cancer Genome Atlas (TCGA)^24,25 and the National Cancer Center Research Institute (Japan; NCCRI).^26,27 BTG2 was the only gene with consistent prognostic value (Figures 3D, S3E, and S3F). Using another LUAD dataset (GSE18549),²⁸ we validated that only BTG2 mRNA is decreased in LUAD that has metastasized to the brain, compared with LUAD that has metastasized to other organs (Figure 3E). BTG2 is a putative cell-cycle regulator with broad expression across most epithelial but few neuronal tissues. This observation suggests that downregulation of BTG2 may contribute to brain metastasis. To examine this hypothesis, we knocked down BTG2 in LUAD H460 cells (Figure 3F), which rarely develop early (<2 weeks after injection) brain metastasis in our model. After intracardiac injection of control and BTG2^KD cells, we observed that, although the total tumor burden was similar between the groups (Figures 3G, 3H, and S3G), mice injected with BTG2^KD cells had higher rates of brain metastasis, compared with mice injected with control cells (Figure 3I). Similar results were observed in an experiment in which we reinjected H460 BTG2^KD cells isolated from mouse brain metastatic sites (Figures S3H and S3I), which suggests that BTG2^KD promotes brain metastases. These findings illustrate another use of our in vivo PDO metastasis model: identification and further characterization of genes related to LUAD metastasis in specific organs.

Tumor response to targeted therapies

To test whether our PDO system could also serve as a platform for identification of tumor-specific responses to targeted treatment, proof-of-concept sensitivity assays with the tyrosine kinase inhibitor osimertinib were performed. All epidermal growth factor receptor (EGFR) mutant PDOs were highly sensitive to osimertinib, compared with EGFR wild-type PDOs (Figure S4A).

To model additional clinically relevant targeted therapies, we next investigated sotorasib (AMG-510), the first Food and Drug Administration-approved inhibitor of the Kirsten rat sarcoma viral oncogene homolog (KRAS) oncoprotein.²⁹ Sotorasib consistently reduced cell viability in KRAS^G12C PDOs (12 and 23) and reduced phospho-extracellular signal-regulated kinase (p-ERK) (Figure 4A). We then assessed the efficacy of sotorasib for suppression of metastasis using our in vivo PDO metastasis model. Metastases developed in 85% of mice injected with PDO_12 (6/7) and in 37% of mice injected with PDO_23 (3/8) in vehicle treatment groups at 5 months after injection. No metastases were identified in mice injected with either PDO_12 or PDO_23 and treated with sotorasib (Figure 4B).

Figure 4.

PDOs as a platform for assessment of therapeutic efficacy in vitro and in vivo

(A) Drug-response curves (top) showing the percentage of cell death and IC₅₀ of PDOs with KRAS^G12C treated with sotorasib in two independent runs (n = 3). Immunoblots (bottom) indicating p-ERK and total ERK in each PDO.

(B) Representative pictures (top) showing metastases in mice with PDO_12 treated with or without sotorasib. The table (bottom) summarizes the locations of metastases formed.

(C) Drug-response curves showing the percentage of cell death and IC₅₀ of the regenerated organoids from the metastatic sites of mice (MDOs) with PDO_23 treated with sotorasib in two independent runs (n = 3).

(D) Heatmap showing genes differentially expressed in RNA sequencing.

(E) Drug-response curves (left) showing the percentage of cell death and the IC₅₀ of PDO_23 and MDO_liver and MDO_brain treated with sotorasib in two independent runs (n = 3). Immunoblots (right) indicating FGFR1 in MDOs.

(F) Schematic illustration of the immune-priming strategy. PDOs were pretreated with cisplatin plus vinorelbine (chemotherapy), X-ray, or DMSO 24 h before coculture with autologous PBMCs. After 2 weeks of coculture, T cell effector functions were evaluated using flow cytometry. Activated CD8⁺ T cell-mediated killing was evaluated using a PDO-killing assay.

(G) Representative plots gated on CD8⁺ T cells (top) tested for reactivity against autologous PDOs. Bars (bottom) showing quantification of CD107a+CD8⁺ and CD137+CD8⁺ T cells obtained after 2-week coculture.

(H) Microphotographs of untreated PDOs (left) after culture with activated CD8⁺ T cells. PDOs were labeled with CellTrace Yellow (magenta) before coculture; apoptotic cells appeared green. Bars show quantification of PDO killing (right). Data are represented as mean ± SEM (n = 5 fields/sample).

Not all metastases have the same sensitivities to treatments as their primaries.^30,31 To explore whether metastases developed from the KRAS^G12C PDO in our in vivo metastasis model respond similarly to sotorasib, we used MDOs generated from our in vivo treatment-naive PDO_23 metastasis (Figure 2; Table S2). We found that MDO_liver and MDO_brain had a higher half-maximal inhibitory concentration (IC₅₀) and MDO_soft tissue, MDO_GB, and MDO_DIAPH had a relatively lower IC₅₀ to sotorasib, compared with PDO_23 (Figure 4C). Interestingly, these differences in response to sotorasib were not fully explained by restored p-ERK signaling (Figure S4B). To explore alternative mechanisms through which MDOs differently responded to sotorasib, we scrutinized WGS and RNA-seq data (Figure 3). Although we did not find genomic alterations previously associated with resistance to sotorasib,³² RNA-seq data indicated that FGFR1 and CDKN1B were differentially expressed in MDO_liver and MDO_brain, compared with the parental PDO_23 and other MDOs (Figure 4D). Given that upregulation of FGFR1 enhances innate resistance to KRAS^G12C inhibitors in NSCLC,³³ we confirmed, by use of CRISPR knockout, that downregulation of FGFR1 resensitized both MDO_liver and MDO_brain to sotorasib in vitro (Figure 4E). In addition, downregulation of the Notch signaling pathway may contribute to the sensitivity to sotorasib observed in MDO_GB and MDO_DIAPH^34,35 (Figures S4C and S4D). Collectively, these data highlight that our in vivo PDO metastasis model can serve as a powerful tool to evaluate the efficacy of therapies for suppression of LUAD metastasis. Furthermore, we can also explore plausible mechanisms of drug resistance using organoids derived from metastatic sites (MDOs).

Coculture of PDOs and patient-matched PBMCs can inform immune-priming strategies

Chemotherapy or radiation can serve as an immune-priming strategy for patients with lung cancer to enhance the efficacy of ICI therapies.^36,37 It remains unclear which approach—chemotherapy or radiation—when combined with ICI therapy offers the best tumor immune-priming strategy. To begin to address this knowledge gap, we performed proof-of-concept experiments by coculturing chemotherapy- or radiation-pretreated PDOs with autologous peripheral blood mononuclear cells (PBMCs) in the presence of the PD-1 inhibitor nivolumab (Figure 4F). Before the coculturing step, we assessed the response of PDO_36 (PD-L1 expression = 70%) and PDO_37 (PD-L1 expression = 50%) to the treatments by assessing mRNA levels of key immune factors (Figure S4E). The highest levels of CD137+CD8⁺, CD107a+CD8⁺, interferon (IFN)γ+CD8⁺, and CD45RO + CD8 T cells, and production of IFNγ, were observed in both sets of PBMCs cocultured with PDOs pretreated with radiation (Figures 4G, S4F, and S4G). However, tumor necrosis factor alpha production was not significantly different between pretreatments (Figure S4G). To examine whether activated CD8⁺ T cells can kill tumor cells, we cocultured autologous treatment-naive PDOs with pretreated PDO-reactive CD8⁺ T cells. Pretreatment with radiation was a more efficient immune-priming strategy for both tested PDOs, compared with pretreatment with DMSO or chemotherapy. Immune priming with radiation resulted in more tumor-specific reactive CD8⁺ T cells and led to higher levels of tumor cell death (Figure 4H).

Discussion

There is an increasing appreciation of the need to identify optimal patient-derived tissue models to study cancer metastasis, treatment effects, and mechanisms of acquired resistance to therapy. Unfortunately, PDX models of LUAD rarely form metastases. To address the limitations of immortalized cell and PDX in vivo metastatic models, we created an in vivo PDO model to study the biology of metastatic disease. The initial enthusiasm for the use of PDOs to inform personalized medicine has been tempered by findings of alveolar cell overgrowth.¹⁰ Our findings support these observations but also identify specific clinicopathologic features associated with successful generation of PDOs from primary LUAD specimens. Given that organoid culturing is labor, time, and capital intensive, the ability to predict which specimens have a high likelihood for successful PDO generation holds substantial value.

SOL—and micropapillary (MIP)—predominant histologic subtypes are categorized as high architecture grades of LUAD and are associated with poor prognosis; SOL subtype is strongly associated with the development of distant metastatic disease.^38,39 Interestingly, compared with the 100% success rate for generation of PDOs from SOL-predominant primary tumors, no PDOs were generated from MIP-predominant tumors (Figure 1B). This may be related to the very small sample size (n = 3) of MIP-predominant tumors. In addition, we found that MIP-predominant tumors have fewer alterations in the oncogenic Hippo and TP53 pathways and fewer mutations in the TP53 gene, compared with SOL-predominant tumors, in our LUAD cohort.⁴⁰ Both of these pathways confer stemness in cancer cells^41,42 and may contribute to the success of PDO generation.

The median interval between diagnosis of LUAD and emergence of distant metastasis after surgical resection in early-stage disease is 28 months⁴³ This long latency precludes studies of clonality and tumor evolution. Several important features—including intraorganoid heterogeneity, relatively shorter in vivo latency (∼5 months), real-time tractable metastatic cells, and preservation of the histologic, genomic, and transcriptomic profiles of the parental tumors (Figure 2)—make our in vivo PDO metastasis model a powerful preclinical tool for the study of tumor evolution. Similar to tumor evolution patterns in patients with LUAD,⁴⁴ our in vivo PDO model displays linear and branched patterns of clonal evolution (Figure 3). In addition, our in vivo PDO metastasis model represents a valuable resource for identification of genes related to brain metastasis, an application that has been minimally investigated in research for cancer-related organoids. We identified a potential brain metastasis-related candidate gene, BTG2, by combining the transcriptomic profiles of PDOs from human brain metastases and MDOs_brain. Importantly, using LUAD cells, we validated that reduction of BTG2 promotes brain metastasis (Figure 3).

Given the high degree of phenotypic and genotypic similarity between PDOs and their parental tumors, PDOs were thought to be able to play a role in drug screening for personalized medicine. Unfortunately, the current success rates for PDO generation preclude this application. In our in vivo PDO metastasis model, we chose to examine mechanisms of resistance in KRAS^G12C PDOs treated with sotorasib, highlighting our model’s potential utility for in vitro and in vivo drug resistance studies of targeted therapies for LUAD with driver gene alterations. The inevitable emergence of acquired drug resistance, particularly to targeted therapies, remains the biggest obstacle to the longer-term success of these agents. In a collection of primarily treatment-resistant breast tumors, Guillen et al. used PDX and matched PDO models to identify responses to a variety of drug combinations and genomic alterations associated with treatment response.⁴⁵ Similarly, Mertens et al. investigated drug sensitivity in PDOs of KRAS-mutant colorectal cancer tumors and identified a viable triple therapy that is currently being tested in a phase 1/2 trial.⁴⁶ As demonstrated by the observation of increased FGFR1 and its association with resistance to sotorasib, our in vivo PDO metastasis model can be used to study mechanisms of resistance to targeted therapies, which has implications for the development of future therapies.

Recent clinical trials have shown that neoadjuvant ICI therapy plus platinum-based chemotherapy is associated with higher pathologic complete response rates and longer event-free survival in patients with stage IB-IIIA resectable NSCLC.^37,47 In addition, a phase 2 trial of neoadjuvant ICI therapy plus stereotactic radiation therapy in patients with early-stage NSCLC demonstrated that neoadjuvant ICI therapy plus radiotherapy is associated with a high major pathologic response rate.³⁶ Unfortunately, for most patients in these trials, their cancer did not respond to neoadjuvant immunotherapy, suggesting that a singular approach is not appropriate for all patients. PDOs have been previously used to study changes in the tumor immune microenvironment in response to ICI therapy.^48,49,50 Neal et al. treated PDOs with nivolumab alone to identify response to treatment, which indicated that PDOs could be used to predict response to ICI therapy in patients.⁴⁸ Air-liquid interface culture models of PDOs have been shown to predict response to ICI therapy in patients.⁴⁸ However, given our interest in immune priming to enhance tumor cell immunogenicity, we used a PDO composed of pure tumor cells and a PBMC coculture system⁵¹ to focus on the tumor cells exclusively and identify the most appropriate immune-priming strategy (chemotherapy or radiation). We showed that PDOs pretreated with radiation followed by ICI therapy activated more tumor-specific cytotoxic CD8⁺ T cells (leading to higher levels of tumor cell death), compared with PDOs pretreated with ICI therapy alone or with ICI therapy plus chemotherapy. Our PDO and PBMC coculture system is a promising tool to identify the most efficient immune-priming strategy for individual patients.

Compared with most lung cancer PDO studies that focus on drug screening and the efficacy of personalized therapies, we used PDOs to successfully establish an in vivo LUAD metastasis model that functions as a preclinical tool for studying the biology of metastasis and tumor evolution and for assessing the efficacy of drugs for metastasis suppression and mechanisms of drug resistance. In addition, by use of a PDO-PBMC coculture system, our study provides a proof-of-concept model for identifying the most efficient personalized immune-priming strategy.

In summary, our fully characterized PDO system and in vivo PDO metastasis model offer multiple applications to study the biology of metastasis, assess drug efficacy and resistance in vitro and in vivo, and characterize immune-priming strategies for individual patients with LUAD.

Limitations of the study

The approach used to generate and characterize PDOs (Figure S1A) may not fully represent tumor heterogeneity. Given the limited availability of tumor tissue, available tissue was divided into four pieces in an effort to have enough tissue for PDO generation and quality controls. Our studies lack an immune microenvironment. Alternative approaches using humanized mouse models may offer a more representative platform for studying the biology of metastasis and developing efficient therapeutic strategies. Given the challenge of PDO generation from primary LUAD with <30% tumor purity, it is premature to claim that PDOs can be used as a tool for personalized medicine in patients. Alternatively, with the high success rate (>90%) of PDO generation for metastatic tumors or MPEs, PDOs may be used in personalized medicine for these patients. Given the importance of both pathways of cell adhesion and extracellular matrix-receptor interaction in tumor metastasis,^52,53 the negative enrichment of these pathways in PDOs (Figure S2D) may influence their metastatic capability. We appreciate the complexity of acquired drug resistance and acknowledge other contributing mechanisms to resistance to sotorasib, in addition to FGFR1. However, uncovering all potential mechanisms related to sotorasib is beyond the scope of our study. To assess the intact metastatic process, establishment of an in vivo orthotopic PDO metastasis model will be necessary in future studies. Finally, the translational significance of our current in vivo PDO metastasis model necessitates validation using larger patient cohorts, perhaps focusing on those derived from metastatic sites and MPEs.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, David R. Jones (jonesd2@mskcc.org).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Raw data of deidentified RNA-seq are available in the NCBI Gene Expression Omnibus (GSE276387). Raw sequencing data for WGS of patients’ specimens are protected and are not broadly available because of privacy laws. Raw data may be requested from Jonesd2@mskcc.org with appropriate institutional approvals.
• •The software and algorithms for data analyses used in this study are published and referenced throughout the STAR Methods section.
• •This study does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This study was supported by the National Cancer Institute (R01CA217169, R01CA240472, and R01CA285856 to D.R.J.; R01CA192399 to M.W.M.; R37CA229861 to M.I.), Hamilton Family Foundation (to D.R.J.), Department of Defense (LC160212 to P.S.A.), and National Institutes of Health (P30CA008748 to MSK). We acknowledge the Molecular Diagnostics Service at MSK. David B. Sewell, of the MSK Department of Surgery, provided superb editorial assistance.

Author contributions

Conceptualization, Y.L., D.R.J., and M.W.M.; methodology, Y.L., D.R.J., P.L., and M.I.; investigation, Y.L., M.L., H.T., J.Z., S.B., L.-F.L.C.-T., and D.H.; formal analyses, B.M., J.R., and K.E.J.; data curation, B.M., J.G.C., H.B.L., R.C., E.G.D., and C.N.F.; resources, G.R., S.S., J.M.I., M.J.B., C.W.B., P.S.A., N.R., and M.H.B.; writing, Y.L., D.R.J., M.W.M., M.I., and B.T.L.; funding acquisition, M.W.M., P.S.A., D.R.J., and M.I.

Declaration of interests

G.R. has financial relationships with Scanlan, AstraZeneca, and Medtronic. S.S. is a member of the AstraZeneca Advisory Board. J.M.I. has stock ownership in LumaCyte and is a consultant/advisory board member for Roche Genentech. M.J.B. is a consultant for AstraZeneca, Iovance Biotherapeutics, and Intuitive Surgical and receives research support from Obsidian Therapeutics. B.T.L. has served as an uncompensated advisor and consultant to Amgen, AstraZeneca, Boehringer Ingelheim, Bolt Biotherapeutics, Daiichi Sankyo, Genentech, and Lilly; has received research grants (institutional) from Amgen, AstraZeneca, Bolt Biotherapeutics, Daiichi Sankyo, Genentech, Hengrui USA, and Lilly; has received academic travel support from Amgen, Jiangsu Hengrui Medicine, and MORE Health; and has intellectual property rights as a book author at Karger Publishers and Shanghai Jiao Tong University Press. M.H.B. receives royalties from Globus Medical and DePuy Synthes. P.S.A. declares research funding from Atara Biotherapeutics; is a scientific advisory board member and consultant for ATARA Biotherapeutics, Bayer, Bio4T2, Carisma Therapeutics, Imugene, ImmPACT Bio, Johnson & Johnson, Orion, and Outpace Bio; has patents, royalties, and intellectual property on mesothelin-targeted chimeric antigen receptor and other T cell therapies, which have been licensed to Atara Biotherapeutics; and has an issued patent method for detection of cancer cells using virus and pending patent applications on PD-1 dominant negative receptor, a wireless pulse-oximetry device, and an ex vivo malignant pleural effusion culture system. MSK has licensed intellectual property related to mesothelin-targeted chimeric antigen receptors and T cell therapies to Atara Biotherapeutics and has associated financial interests. D.R.J. serves on a clinical trial steering committee for AstraZeneca and has research grant support from Merck.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
CD107a-PE	BioLegend	Cat# 328607 RRID:AB-1186062
IFNγ-APC	BioLegend	Cat# 506510 RRID:AB_315443
CD137-PerCP/Cy5.5	BioLegend	Cat# 309814 RRID:AB_2205686
CD3-AF594	BioLegend	Cat# 300446 RRID:AB_2563236
CD8-PE-Cy7	BioLegend	Cat# 300914 RRID:AB_314118
CD45RO-BV650	BioLegend	Cat# 304232 RRID:AB_2563462
SOX2	Abcam	Cat# ab92494; RRID:AB_10585428
Ki-67	Abcam	Cat# ab15580 RRID:AB_443209
ALDH1A1	Abcam	Cat# ab52492 RRID:AB_867566
BTG2	Abcam	Cat# ab85051 RRID: AB_1861407
FGFR1	Proteintech	Cat# 60325-1-IG RRID:AB_2881435
p-ERK	Cell Signaling	Cat# 4370 RRID:AB_2315112
ERK	Cell Signaling	Cat# 4695 RRID:AB_390779
EpCAM	Thermo Fisher	Cat# MA5-12153 RRID:AB_10982998
ACTIN	Santa Cruz	Cat# sc-8432 RRID:AB_626630
Donkey anti-Mouse IgG (H + L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor™ 594	Thermo Fisher	Cat# A-21203 RRID: AB_2535789
Bacterial and virus strains
Lentivirus, CAG-Luciferase (firefly) (GFP-Puro)	GenTarget	Cat# LVP570
Biological samples
Human tumor tissue specimens	This study	N/A
Human non-cancerous adjacent tissue	This study	N/A
Human lung cancer metastatic samples	This study	N/A
Human pleural effusion samples	This study	N/A
Human peripheral blood monocytes (PBMCs)	This study	N/A
Chemicals, peptides, and recombinant proteins
10% paraffin	Sigma-Aldrich	Cat# HT501128
Collagenase B	Roche	Cat# 11088815001
DNase I	Millipore	Cat# 691823
Matrigel	Corning	Cat# 354234
TrypLE	Thermo Fisher	Cat# 12605010
Ficoll-Paque	Millipore	Cat# GE17-1440-02
Cisplatin	Selleckchem	Cat# S1166
Vinorelbine	Selleckchem	Cat# S4259
Nivolumab	Selleckchem	Cat# A2002
GolgiSTOP	BD	Cat# 554724
GolgiPLUG	BD	Cat# 555029
NucView488 Dye	Biotium	Cat# 30029-T
Cell Recovery Solution	Corning	Cat# 354253
VivoGlo Luciferin	Promega	Cat# P1041
AMG-510 (Sotoracib)	Selleckchem	Cat# S8830
Advanced DMEM/F12	Thermo Fisher	Cat# 12634028
HEPES	Thermo Fisher	Cat# 15630080
Glutamax	Thermo Fisher	Cat# 35050061
N-acetyl-L-cysteine	Millipore Sigma	Cat# A9165-5G
Primocin	InvivoGen	Cat# ant-pm-1
N2	ThermoFisher	Cat# 17502001
B-27 supplement (50X), minus vitamin A	ThermoFisher	Cat# 12587001
Noggin	R & D	Cat# 6057-NG
FGF	R & D	Cat# 233-FB
CHIR 99021	Tocris	Cat# 4423/50
EGF	PeproTech	Cat# AF-100-15
Y-27632 dihydrochloride	Abmole Bioscience	Cat# M1817
SB 431542	R & D	Cat# 1614/10
RPMI 1640	Thermo Fisher	Cat# 42402016
Mounting medium	Thermo Fisher	Cat# P36934
Penicillin/streptomycin	Thermo Fisher	Cat# 10378016
Human serum	Sigma-Aldrich	Cat# H4522
IL-2	PeproTech	Cat# 200-02
UltraGlutamine I	Lonza	Cat# BE17-605E/U1
Fetal Bovine Serum	Thermo Fisher	Cat# 26140079
Critical commercial assays
HRP/DAB IHC Detection Kit	Abcam	Cat# ab236466
DNeasy Blood & Tissue Kit	Qiagen	Cat# 69504
NEBNext Ultra II End Repair/dA-Tailing Module	New England Biolab	Cat# E7595
RNeasy Plus Mini kit	Qiagen	Cat# 74136
CellTiter-Glo Luminescent Cell Viability assay	Promega	Cat# G7571
Human CD8+ T cell Isolation Kit	Miltenyi Biotec	Cat# 130-096-495
Deposited data
RNA-seq	This study	Gene Expression Omnibus (GSE276387)
TCGA lung adenocarcinoma cohort	The cBioPortal for Cancer Genomics	Cbioportal.org
NCCRI lung adenocarcinoma cohort	Gene Expression Omnibus	GSE31210
BTG2 mRNA in lung adenocarcinoma metastasizing to brain	Gene Expression Omnibus	GSE18549
Experimental models: Cell lines
NCI-H460	ATCC	Cat# HTB-177 RRID:CVCL_0459
Experimental models: Organisms/strains
NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG) mouse	The Jackson Laboratory	Strain #:005557 RRID:IMSR_JAX:005557
Oligonucleotides
FGFR1 sgRNA: CTGGTCTTAGGCAAACCCCT	This study	N/A
HLA-A FW: 5′-ggccctgacccagacct-3′	This study	N/A
HLA-A RV: 5′-gcccctcctgctctatcca-3′	This study	N/A
CD137 FW: 5′-cactctgttgctggtcctcaa-3′	This study	N/A
CD137 RV: 5′-ggacagggactgcaaatctga-3′	This study	N/A
CD47 FW: 5′-tggtagcggcgctgttg-3′	This study	N/A
CD47 RV: 5′-tcagtagtgttttgtgcctcca-3′	This study	N/A
CD274 FW: 5′-ttcatgacctactggcatttgct-3′	This study	N/A
CD274 RV: 5′-agtgcagccaggtctaattgtt-3′	This study	N/A
Primers for Sanger sequencing (See Table S5)	This study	N/A
Recombinant DNA
LcV2Hygro	Golden et al.54	Addgene plasmid #91977; RRID:Addgene_91977
Predesigned shRNA BTG2	Millipore Sigma	TRCN0000231595
Software and algorithms
Prism version 10.2.3 (347)	Graphpad	Graphpad.com
R 3.6.1	R Core Development Team	r-project.org
Burrows-Wheeler Aligner, version 0.7.8	BWA-MEM	bio-bwa.sourceforge.net
Living Image Software, version 4.2	PerkinElmer	resources.perkinelmer.com

Experimental model and study participant details

Human specimens

Fresh resected human LUAD specimens, including primary tumors, brain or bone metastases, and MPE samples, were obtained from April 2017 to August 2021. All patients gave informed consent, and the study was approved by the institutional review board (IRB 16–1514, IRB 16–107, and IRB 12–245) at Memorial Sloan Kettering Cancer Center (MSK). Clinical information is available in Tables S1 and S3.

Cell culture

NCI-H460 cells were purchased from ATCC and maintained in RPMI supplemented with 10% FBS in a humidity incubator at 37°C and 5% CO₂.

In vivo LUAD PDO metastasis model

Six-week-old NSG mice were used in this study. All animal experiments were approved by the Animal Care and Use Committee at MSK (protocol #13-10-017). Dissociated PDO_GL or H460 cells were resuspended in PBS at a concentration of 1×10⁵ cells/mL and were injected into the left ventricle of mice. For the experiments with sotorasib treatment, sotorasib (100 mg/kg intraperitoneally daily) or vehicle was administered. Tumor progression was monitored using an IVIS bioluminescence imaging system⁵⁵ monthly for mice injected with PDOs and every 3 days for mice injected with H460 cells. The metastatic burden was quantified in terms of the value of total bioluminescence in each mouse by use of Living Image Software (version 4.2). For experiments with cell lines, an equal number of male and female mice were included in each group. Mice injected with PDOs had the same sex as the patient source of the parental tumor. Mice injected with LUAD cells were euthanized at 2 weeks after injection, and mice injected with PDOs were euthanized at 5 months after injection; necropsy was then performed.

Method details

Tissue processing and organoid generation

Fresh tissues were transported from the operating room to the laboratory and processed immediately. Each tissue sample was cut into 4 pieces: 1 piece was used for generation of organoids, 1 piece was fixed in 10% paraffin for histologic analyses, and the remaining 2 pieces were directly frozen for later RNA and DNA extraction (Figure S1A). Organoids were generated as described previously.¹⁷ In brief, the tissue was minced and incubated in collagenase B and DNase I solution. After filtration, the cells were embedded in Matrigel solution and seeded into a 48-multiwell plate. For MPEs, the cell pellets were collected by centrifugation at 350× g for 5 min at 4°C. After red blood cells were removed using RBC lysis buffer, the cells were embedded in Matrigel solution as described above. Culture medium¹⁷ was added into each well and changed twice per week.

For passaging, the organoids were dissociated into single cells using TrypLE and then replated with Matrigel. Passage was typically performed every 3 to 5 weeks at a 1:5 ratio, depending on the growth speed of the organoids.

For preparation of frozen stocks, the organoids were dissociated into single cells as described above and mixed with organoid-freezing medium (90% FBS +10% DMSO). The suspension was frozen to −80°C using a freezing container overnight and then transferred into liquid nitrogen for long-term storage.

H&E staining and IHC analysis

Tissue was fixed in 10% neutral-buffered formalin solution for 24 h at room temperature. Paraffin embedding, sectioning of slides, and H&E staining were performed by the Pathology Core at MSK, and the percentage of tumor cell content in each sample was determined. Each section was immunostained using the HRP/DAB IHC detection kit. Primary antibodies (Table S4) were used overnight at 4°C.

DNA extraction, sanger sequencing, and WGS

Genomic DNA was extracted using the DNeasy Blood & Tissue Kit. Sanger sequencing analyses were performed as described previously⁵⁴; the primers used are listed in Table S5. For WGS, library construction was performed using the NEBNext Ultra II End Repair/dA-Tailing Module and sequenced on a NovaSeq 6000 at the New York Genome Center. Tumor tissues, PDOs, and MDOs were sequenced at a target coverage of 80×, and normal adjacent tissue was sequenced at a target coverage of 40×.

All collected standard short-read WGS reads (tumor organoids, ∼3 billion reads; normal organoids, ∼2.1 billion reads) were aligned using the Burrows-Wheeler Aligner (BWA-MEM, version 0.7.8; http://bio-bwa.sourceforge.net) to the GRCh37/hg19 ref. ⁵⁶ Harmonized analytic pipelines for somatic SNV or indel calling and loss-of-function mutation calling were applied to the full dataset of WGS profiles. To obtain somatic SNV or indel calls, Strelka2⁵⁷ was run under paired (i.e., tumor with matched normal) mode with default parameters using hg19-based references.

To obtain the contributions of the 67 COSMIC single-base substitution signatures (release 3; https://cancer.sanger.ac.uk/cosmic/signatures/SBS/),⁵⁸ “deconstructSigs” was used with the high-quality somatic SNV calls as input.⁵⁹

Across all samples, a unified list of variants was generated, and the numbers on the branches of each tree represent the somatic variants shared by downstream samples or variants unique to a sample. The normal sample was generated by inputting a null (zero) value for all variants in the matrix.

RNA-seq and data analysis

Total RNA was isolated using the RNeasy Plus Mini Kit. Total RNA was isolated using the RNeasy Plus Mini Kit. For RNA-seq, library preparation and RNA-seq were performed in three batches. Batch A (LUAD23T, PDO23_early, PDO23_late) and Batch B (PDO23_GL, PDO23_MDO2_Brain, PDO23_MDO1_DIAPH, PDO23_MDO1_Liver, PDO23_MDO3_GB): library preparation and RNA sequencing were performed by the Integrated Genomics Operation at MSK using Illumina HiSeq with 50 paired-end reads and approximately 40 million reads/sample (Batch A) and 100 paired-end reads and approximately 20 million reads/sample (Batch B). CutAdapt version cutadapt-1.6 was used to trim and filter reads. Adapters were trimmed with minimum overlap of 10 bases, and reads of less than half original read length were filtered out. STAR (v2.5.0a) was run using the 2-pass method. A genome database was then generated using the minimum read length of half desired read length, and second pass was run against the new genome file with the same parameters as pass 1. Batch C (LUAD1T, 18T, 16T, PDO 1–18, PDO24): library preparation and RNA sequencing were performed by Novogene using Illumina Novaseq 6000 with 150 paired-end reads, with approximately 30 million reads/sample. Reference genome and gene model annotation files were downloaded from the genome website browser (NCBI/UCSC/Ensembl) directly. Indexes of the reference genome were built using STAR (v2.5). Expression count matrixes of all 3 batches were computed using HTSeq (v0.13.5). Parameters set were as follows: -m intersection-strict -s no -t exon.

To account for tissue-specific expression patterns between primary and metastatic samples, differential expression analysis was performed on RNA-seq data from normal tissues in the Human Protein Atlas. Genes that were differentially expressed in brain, lung, liver, or gallbladder were removed from downstream analysis. Genes with low expression in all samples (total counts, <10) were filtered out, and batch correction was applied using the removeBatchEffect function in the limma package. Differential expression was performed using the “DESeq2” R package with default settings (adjusted p < 0.01 and log₂ FC > 1).

Drug sensitivity assays

Dissociated PDO or MDO cells (5000 cells/well) were seeded into 48-well plates with Matrigel in culture medium. After 5 days of culture, fresh medium containing serial dilutions of sotorasib or osimertinib (2.5 nM) was added. For sotorasib treatment, drug-containing medium was changed every 3 days. Cell viability was determined at day 9 after the start of sotorasib treatment and at day 4 after the start of osimertinib treatment using the CellTiter-Glo Luminescent Cell Viability assay. DMSO treatment was used as a control.

Immunofluorescence

PFA-fixed organoids were blocked, washed, and stained in Eppendorf tubes and collected by centrifuging at 250 rpm for 2 min. Primary antibody for detection of EpCAM was used at a 1:100 dilution overnight at 4°C. Secondary antibody labeled with red-fluorescent Alexa Fluor 594 dye was used at 1:2000 dilution for an additional 1 h. Stained organoids were suspended in mounting medium and transferred onto glass slides for microscope photos.

Reinjection of H460 shBTG2 cells isolated from mouse brain metastatic sites

Mice were euthanized at 12 days postinjection of H460 shBTG2 cells, and brains were dissected. Single cells were dissociated using collagenase B and DNase I solution, as described above. After sorting by FACS, GFP-positive cells were suspended into PBS at a concentration of 1×10⁵ cells/mL and were reinjected into the left ventricle of mice. An equal number of male and female mice were included in each group. Tumor progression was monitored using an IVIS bioluminescence imaging system.

Analyses of the TCGA and NCCRI LUAD cohorts

We used two publicly available LUAD datasets: TCGA (cbioportal.org)^24,25 and NCCRI (NCBI Gene Expression Omnibus GSE31210).^26,27 For survival analyses, we examined the expression of 12 genes of the putative brain metastasis markers. Survival was assessed using the Kaplan-Meier method and compared between high and low mRNA groups of individual genes using the log rank test. Cutoffs for each gene were determined by maximally selected rank statistics using the package maxstat in R (version 3.6.1, R Core Development Team, Vienna, Austria).

Isolation of PBMCs

Ten milliliters of peripheral blood were collected at the same time that tumor specimens were collected. PBMCs were isolated using Ficoll-Paque density gradient separation and were cryopreserved until later use.

Immune priming and PDO-PBMC coculture

Dissociated PDO cells were seeded at a density of 5000 cells/well with Matrigel into 48-multiwell plates on day 1. Pretreatment with radiation was started on day 5, and pretreatment with chemotherapy was started on day 7. For radiation, PDOs were irradiated with 20-mA X-rays using the XRad320 irradiator to acquire a dose of 6 Gy per day for 3 days. For chemotherapy, doublet chemotherapy reagents of cisplatin and vinorelbine at concentrations of 17 μM and 2.5 μM, respectively, or 1% DMSO as control were administered for 24 h.

PDO-PBMC coculture was performed on day 8 for a total of 2 weeks, as described previously.⁵¹ In brief, on day 7, matched PBMCs were thawed and resuspended in T cell medium (RPMI 1640 supplemented with 2 mM UltraGlutamine I, 10% human serum, 150 U/mL IL-2, and 1% penicillin/streptomycin) at a concentration of 1×10⁶ cells/mL. On day 8, PBMCs at a density of 1×10⁵ cells/well and dissociated pretreated PDO cells were mixed at a ratio of 20:1 in T cell medium supplemented with 20 μg/mL nivolumab and were seeded into CD28 antibody–coated 96-well U-bottom plates. Fresh T cell medium containing nivolumab was added every 2 to 3 days during coculture. Every 4 days for radiation and every week for chemotherapy and controls, cocultured PBMCs were collected and replated with fresh pretreated PDO cells as described above.

CD8⁺ T cell activation and PDO-killing assays

CD8⁺ T cell activation was evaluated using flow cytometry. After 14 days of coculture, the enriched T cells were collected and re-cocultured with fresh radiation- or chemotherapy-pretreated PDO cells at a ratio of 2:1 in T cell medium supplemented with 1:20 anti-CD107a-PE. GolgiSTOP and GolgiPLUG were added 1 h after coculture commenced. After a total of 5 h of coculture, cells were stained with antibodies (Table S4) for 30 min on ice. Cells were then washed twice in FACS buffer and stained with 1:20 anti-IFNγ-APC for an additional 30 min. FACS was conducted at the Flow Cytometry Core at MSK, and the gating strategy was as described in Figure S4H. In the case of anti-CD137-PerCP/Cy5.5 staining, T cell stimulation with fresh pretreated PDOs was performed for 24 h before staining. PBMCs cultured without PDO stimulation served as a negative control.

For the PDO-killing assay, the CD8⁺ T cells were isolated after 2 weeks of coculture with pretreated PDOs using the human CD8⁺ T cell isolation kit. Fresh treatment-naive autologous PDOs were plated at a density of 5000 cells/well before 5 days of assay. On the day of the experiment, intact PDOs were isolated using cell recovery solution and labeled with NucView488 dye. One in 10 PDOs was dissociated to single cells for counting. Labeled PDOs were cocultured with isolated activated CD8⁺ T cells at a ratio of 1:5 in T cell medium with nivolumab and Y-27632 for 72 h. Caspase activity was detected by quantifying green fluorescence using ImageJ software (https://imagej.nih.gov/ij/).

Quantification and statistical analysis

The results of all experiments represent the mean ± SEM of at least three independent experiments performed in triplicate. Statistical analysis was performed using R or GraphPad Prism 9. Tests for differences were performed using the Mann-Whitney test. Significance was set at a false-discovery rate of ≤0.25 for gene set enrichment analyses and p < 0.05 for all other experiments.

Published: October 15, 2024