Abstract

   Despite immunotherapy’s promise in cancer treatment, patient responses
   vary substantially because of the individual nature of the immune
   system and the lack of reliable biomarkers. To address this issue, we
   developed a precision ex vivo platform that integrates patient-specific
   tumor and immune cells to study the mechanisms of antitumor immune
   response, predict immunotherapy outcomes, and identify effective
   treatments. This platform revealed unique single-cell immune response
   mechanisms and sensitivities to standard-of-care immunotherapies.
   Furthermore, we were able to identify a synergistic combination of
   anti–programmed cell death protein 1 (anti–PD-1) together with a
   Casitas B lineage lymphoma-b inhibitor that overcame anti–PD-1
   resistance in selected patient samples. Activation of the
   interferon-γ–stimulated cytokines predicted combination efficacy, while
   immunosuppressive cytokines were associated with poor response. Our
   findings underscore the platform’s potential in tailoring
   immunotherapies and advancing drug development, offering avenues for
   personalized cancer treatment.
     __________________________________________________________________

   A patient-derived ex vivo platform enables the detection of unique
   immunotherapy responses in solid tumors.

INTRODUCTION

   Immunotherapies, especially immune checkpoint inhibitors (ICIs), have
   notably improved the management of various cancers by offering durable
   clinical benefits. ICIs targeting programmed cell death protein 1
   (PD-1), programmed death-ligand 1 (PD-L1), or cytotoxic T
   lymphocyte–associated antigen 4 (CTLA-4) have demonstrated remarkable
   treatment efficacy in cancers like melanoma and non–small cell lung
   cancer (NSCLC) ([64]1). Despite their initial success, variable
   efficacy and the need for more reliable biomarkers present a
   substantial challenge in predicting patient treatment responses to
   ICIs, hindering the success of immuno-oncology (IO) treatments. Despite
   the use of tumor PD-L1 expression as a guide for anti–PD-(L)1 therapy
   in NSCLC and other cancers, its dynamic nature and poor predictive
   capabilities have made its use as a biomarker controversial ([65]2).

   In addition, other immune checkpoints such as lymphocyte activation
   gene 3 (LAG-3), T cell immunoglobulin and mucin-domain containing 3
   (TIM-3), as well as immunosuppressive cytokines, contribute to
   antitumor immune evasion and therapy resistance ([66]3, [67]4).
   Combining different ICIs beyond anti–PD-(L)1 and anti–CTLA-4 has
   emerged as a potential strategy to overcome resistance to the current
   IO treatments ([68]5, [69]6). Although several clinical studies are
   investigating IO combination strategies, the pace of development lags
   behind the emergence of new immunotherapeutic agents. Consequently,
   there is an urgent need for sophisticated preclinical platforms that
   could accurately replicate the complexity of human antitumor immunity
   and the tumor immune microenvironment (TIME).

   The existing preclinical models for assessing IO efficacy suffer from
   substantial limitations. Patient-derived tumor organoids typically lack
   immune contexture, while humanized mice are limited by issues of
   reproducibility and cost. Recently, organ-on-chip (OOC) systems using
   tissue fragments have been used to evaluate ICI responses ([70]7).
   Still, their applications are restricted by the small size of tumor
   resections and the absence of systemic immune events. Other OOC-based
   approaches involving inactivated or nonspecifically activated
   peripheral blood mononuclear cells (PBMCs), or commercially available
   nonmatching immune cell lines, also fail to incorporate
   patient-specific antitumor immunity ([71]8–[72]12). Therefore, there is
   an urgent need for a precision platform capable of faithfully mimicking
   personalized responses to IO drugs. Here, we describe a patient-derived
   ex vivo platform that leverages patient-matched tumor organoids and
   tumor-stimulated autologous immune cells to evaluate mechanisms of
   antitumor immune response and resistance mechanisms. We used
   single-cell methods to study the individual mechanisms of antitumor
   immune recognition and activation, as well as to gain better
   understanding of acute immune and IO drug mechanisms. OOC technology
   was further used to recapitulate the immune cell migration toward the
   tumor site and to measure immune cell–mediated tumor killing. By using
   this platform, we aim to accelerate the development of effective IO
   treatments while guiding personalized treatment selection for patients
   with cancer in the future.

RESULTS

Assessment of tumor-specific immune responses using tumor-derived organoids
and circulating immune cells

   To understand individual antitumor immune responses, we first generated
   a living biobank comprising three-dimensional (3D) organoids from lung
   tumor and benign tissue, achieving a successful culture rate of 83% (49
   of 59) of the samples, with 15 selected for further analysis. Patient
   tumor characteristics, including organoid mutation profiling and
   diagnostic PD-L1 expression from the tumor samples, can be found in
   table S1.

   Using patient-matched PBMCs allowed us to investigate the acute and
   personalized interactions between non-primed immune cells and
   tumor-derived organoids, overcoming the limitations associated with the
   direct use of tumor-infiltrating lymphocytes (TILs), which may be
   limited in numbers and have diminished functionality. This approach
   also enabled systemic examination of diverse immune cell types beyond
   just lymphocytes. First, we cocultured the PBMCs with autologous tumor
   organoids, aiming to mimic systemic antitumor immune activation in the
   body [[73]Fig. 1A; the protocol for generating tumor-stimulated immune
   cells was adapted from ([74]13)]. Using flow cytometry–based
   immunoprofiling, we examined changes in viable T lymphocytes (CD3^+),
   helper T cells (CD3^+/CD4^+), and cytotoxic T cells (CD3^+/CD8^+) in
   the baseline PBMCs versus in the immune cells isolated from the
   cocultures (tumor-stimulated immune cells, “ts-immune”). Our findings
   reveal that, while there was no significant overall shift in T cell
   proportions, individual variability was evident ([75]Fig. 1, B and C,
   and fig. S1A). Notably, interferon-γ (IFN-γ) expression was induced by
   the tumor stimulation in 50% (7 of 14) of the samples, whereas 29% (4
   of 14) were not reactive. Twenty-one % (3 of 14) of the patients had
   preexisting production of IFN-γ, which was further increased by tumor
   stimulation in two samples ([76]Fig. 1D). IFN-γ induction in the CD8^+
   T cells was time dependent, and the dynamics varied between patients
   (fig. S1B), describing the individual nature of activation.

Fig. 1. Generation of tumor-stimulated immune cells via coculture of tumor
organoids and autologous PBMCs.

   [77]Fig. 1.
   [78]Open in a new tab

   (A) Experimental workflow: The study cohort included 15 patients with
   lung cancer carrying various mutations. Lung cancer tumor resections
   were used to form 3D tumor organoids. Matching peripheral blood samples
   were collected to isolate PBMCs, which were then cocultured with
   autologous tumor organoids for 2 weeks to generate ts-immune cells.
   These ts-immune cells were subsequently used for scRNA-seq and to
   evaluate immune and drug killing efficacy on a chip. The illustration
   indicates the source of the patient material used in the different
   experiments, with patients color-coded and clinically relevant
   mutations highlighted. (B) Effect of coculture on T cell
   subpopulations. Immune profiling by flow cytometry is presented as a
   violin plot, comparing the bulk percentage of viable CD3^+ cells and
   subpopulations of CD4^+ and CD8^+ T cells before (week 0) and after
   (week 2) stimulation with patient-matched tumor organoids. In the
   violin plot, two dots of the same color represent biological
   replicates, while different colors indicate individual patients. The
   mean is depicted as a black line. The figure also includes the main
   clinically relevant mutations. An unpaired Student’s t test was used to
   compare the proportions of live cells at 0 and 2 weeks, with P values
   displayed in the figure. (C) Activation of circulating T cells by
   stimulation with matching tumor organoids. Representative flow
   cytometry plots show CD8^+/IFN-γ^+ lymphocytes from three patients. (D)
   Percentage of CD8^+/IFN-γ + T cells before and after stimulation.
   Symbols in the graph represent the percentage of anti–IFN-γ in CD8^+ T
   cells from one biological replicate (for ORG-35, ORG-54, and LUNG-19,
   an average of two biological replicates is depicted by each symbol).
   Different colored symbols indicate individual patients.

   Although our study was limited in size, we observed a trend where tumor
   organoids with KRAS mutations provoked a stronger immune response than
   those with epidermal growth factor receptor (EGFR) mutations. This
   observation aligns with the clinically established understanding that
   KRAS- and EGFR-mutant tumors differ in their immunogenic potential,
   with KRAS mutations often triggering more robust immune reaction.

Exploring individual immune responses to tumor stimulation via single-cell
transcriptomics

   To better understand the underlying mechanisms of antitumor immunity
   and individual patient responses to tumor stimuli, we used single-cell
   RNA sequencing (scRNA-seq) complemented by antibody-based cell hashing
   to analyze circulating immune cells before and after ex vivo tumor
   stimulation in three patients with lung adenocarcinoma with distinct
   genetic mutations ([79]Fig. 2A and table S1). The immune cells were
   again stimulated with the matching organoids for a period of 2 weeks,
   after which all cells were collected for single-cell analysis. In
   general, the analysis revealed a diverse array of immune cell subtypes
   and states, undergoing significant shifts upon tumor challenge
   ([80]Fig. 2, B to D, and markers in table S2).

Fig. 2. Heterogeneity of immune cell subtypes revealed by scRNA-seq following
stimulation with tumor organoids.

   [81]Fig. 2.
   [82]Open in a new tab

   (A) Uniform Manifold Approximation and Projection (UMAP) visualization
   of sample conditions. The plot displays the clustering of PBMCs versus
   ts-immune cells from three patients with lung cancer, showcasing the
   cellular heterogeneity in response to tumor organoid stimulation. Total
   n of immune cells = 7666 (LUNG-19: n = 3006, LUNG-25: n = 2999,
   LUNG-35: n = 1661). Data integration was performed before generating
   the UMAP visualization. (B) Immune cell annotation in UMAP plot. The
   UMAP provides a high-level annotation of various immune cell subtypes
   identified in the samples. Note that B cells, originally located
   distantly on the plot, have been repositioned closer for clearer
   visualization and comparison. (C) Low-level immune cell annotation in
   UMAP plot. Twenty different immune cell clusters were identified;
   T[H]2, T[H]1, T[H]17, TEM, central memory T cell (TCM), T regulatory
   cell (T[reg]), plasmacytoid dendritic cell (pDC), and myeloid dendritic
   cell (mDC). (D) Proportional analysis of immune cell subtypes. The
   chart illustrates the relative proportion difference of ts-immune and
   PBMC in the immune cell subtypes identified in (C), a positive value
   indicating higher cell type proportion in ts-immune. (E) Pathway
   enrichment analysis. Bar plots display selected enrichment pathways in
   CD4^+, CD8^+, and NK cells. These pathways are based on integrated
   genes from ts-immune cells of the three patients following stimulation
   with tumor organoids. Pathway analysis was conducted using Metascape,
   incorporating data from KEGG, GO biological processes, Reactome gene
   sets, and the Hallmark gene sets. (F) Gene Expression Profiling. Dot
   plots illustrate the average expression of selected up-regulated
   effector genes in CD4^+T, CD8^+ T, and NK cells of ts-immune, in
   comparison to baseline PBMCs.

   Specific T and natural killer (NK) cell subsets were enriched in
   response to the tumor stimuli, whereas naive T cells, monocytes, B
   cells, and dendritic cells demonstrated a relative decrease. Among the
   patients, most T/NK cell proliferation (indicative of tumor
   recognition) was observed in patient LUNG-35, followed by LUNG-19. In
   the same samples, pronounced T helper cell (T[H]1/T[H]17, and T[H]2)
   expansion together with the loss of naïve CD4^+ population suggested
   increased helper T cell differentiation. LUNG-35 stood out with an
   increased presence of LAIR2^+ gamma delta T cells [γδ T, CD3^+, and T
   cell receptor (TCR) α/β^−] and LAIR2^+ T effector memory (TEM) cells
   following stimulation. γδ T cells are in general recognized for their
   antitumor capabilities, notably their ability to identify
   stress-induced ligands on tumor cells, bypassing the need for
   conventional antigen presentation ([83]14). In addition, LAIR2 acts as
   an antagonistic inhibitor to the immunosuppressive collagen receptor
   LAIR1, which is known to dampen lymphocytic activity through SHP-1
   signaling ([84]15). This suggested an increased antitumor activation
   especially in patient LUNG-35.

   Next, we looked at the immune cell transcriptional landscape after
   tumor stimulation. There were genes that showed a significant increase
   in expression due to stimulation in all three patients (CD4^+ T cells:
   n = 227 genes, CD8^+ T cells: n = 74 genes, NK cells: n = 94 genes) but
   also large number of individual-level changes in both up- and
   down-regulated genes (fig. S2, A to F). Notably, LUNG-35 and LUNG-19
   shared a significant number of differentially expressed genes (DEGs)
   after tumor challenge, whereas LUNG-25 had more unique DEGs,
   emphasizing the diversity of immune responses elicited by individual
   tumors. By using the DEGs, we examined alterations in signaling
   pathways following tumor stimulation ([85]Fig. 2E). In the CD4^+ T
   cells, we found significant up-regulation in pathways related to
   proliferation and activation, immune response regulation, and leukocyte
   migration. This was accompanied by the amplification of the
   interleukin-2 (IL-2)/signal transducers and activators of transcription
   5 (STAT5) signaling pathway, known for its role in the transcriptional
   response of CD4^+ cells to TCR stimulation ([86]16). Similarly, CD8^+ T
   cells exhibited up-regulated pathways related to activation,
   proliferation, and both endocytic and cell-killing activities,
   suggesting a cytotoxic response. In addition, NK cells exhibited
   pathways related to increased activation and migration, indicating that
   both innate and adaptive antitumor immune responses were present.

   In a further analysis of the DEGs, we identified several genes that
   were significantly up-regulated across immune cell subtypes. In CD4^+ T
   cells, we noted an increase in the expression of CD52, a co-stimulatory
   molecule having a role in both T cell activation and suppression
   [[87]Fig. 2F ([88]17)]. The increased expression of CD40LG pointed to
   more robust helper T cell activation and increased T-B cell
   interactions, boosting antibody-mediated responses. Enhanced cell-cell
   communication was further suggested by the up-regulation of integrin
   gene ITGAL. In CD8^+ T cells, the elevated levels of granzyme A (GZMA)
   and IL32 were indicative of increased cytotoxicity and proinflammatory
   modulation of the immune microenvironment ([89]Fig. 2F). Notably, the
   up-regulation of PECAM1 and genes like KLF2 and CX3CR1 hinted at
   changes regarding T cell migration and interactions with endothelial
   cells. CX3CR1 expression in CD8^+ T cells has been previously suggested
   as a blood-based biomarker correlating with response to anti–PD-1
   ([90]18). Among NK cells, significant up-regulation was observed in
   genes including NKG7, KIR2DL4, and TOX, marking enhanced innate immune
   activation ([91]Fig. 2F).

   LUNG-25 exhibited only subtle responses within the three patients,
   whereas LUNG-35 displayed the most significant immune activation,
   followed by LUNG-19. The magnitude of the ex vivo immune response was
   correlating with the tumor mutational burden (TMB) observed in the
   organoids, with LUNG-35 having the highest TMB and LUNG-25 the lowest
   (fig. S2G).

Mechanisms of lung cancer immunotherapy in tumor-immune cocultures

   To further investigate the interactions between the tumor and immune
   cells in relation to treatment effects, we next compared isolated tumor
   cells to those stimulated with ts-immune, with or without anti–PD-1
   treatment (nivolumab, as detailed in [92]Fig. 3A). First, we mapped the
   tumor cells onto the Human Lung Cell Atlas (HLCA) reference, noting
   that they clustered independently, although closely approximating the
   epithelial cell types of the lung ([93]Fig. 3B), similar to the
   observations made earlier by the HLCA project ([94]19). When ts-immune
   cells were introduced, a distinct immune compartment emerged ([95]Fig.
   3, C to E, and fig. S3A). Unexpectedly, in sample LUNG-25, even the
   tumor cell-only condition displayed immune cells, indicative of the
   organoid model retaining some TILs (fig. S3B).

Fig. 3. Single-cell level responses to ex vivo immunotherapy.

   [96]Fig. 3.
   [97]Open in a new tab

   (A) scRNA-seq overview. The schematic depicts the scRNA-seq process of
   various conditions from three patients with lung cancer. Five
   conditions were sequenced. PBMCs, ts-immune following 2 weeks of
   stimulation with autologous organoids, baseline organoids or organoids
   cocultured with ts-immune with or without nivolumab (100 μg/ml) for 48
   hours. (B) Tumor cell clustering in UMAP. The UMAP plot highlights the
   distinct clustering of our tumor organoid cells (encircled in black
   with dotted lines) against healthy lung cells from the HLCA. Total
   number of tumor cells: 6943 (LUNG-19: n = 2084, LUNG-25: n = 1245,
   LUNG-35: n = 3614). (C to E) UMAP cell clustering of tumor and immune
   cell clustering from tumor alone, tumor + ts-immune, and tumor +
   ts-immune + nivolumab: The plots illustrate cell clustering,
   color-coded by the condition (C), patient identity (D), and cell type
   (E). The cell types were predicted using Azimuth. Data integration was
   not performed before generating the UMAP visualization. (F) Pathway
   analysis heatmap reveals differentially enriched pathways in different
   conditions. Each heatmap value denotes the difference in the average
   module scores. A Wilcoxon rank-sum test was performed for the
   single-cell level module scores, and *, **, *** signify different
   levels of statistical significance: FDR ≤ 0.05, FDR ≤ 0.01, and FDR ≤
   0.001, respectively. The analysis uses the Hallmark gene signature
   database. (G) Cytotoxic gene expression. The plot displays the
   expression levels of cytotoxic genes perforin (PRF1) and granzyme B
   (GZMB) in immune cells. (H to J) TCR clonality analysis. The plots show
   the color-coded TCR clones and their proportions in T cells from the
   three patients across the different conditions outlined in (A).

   Upon tumor challenge, coculture led to an up-regulation of pathways
   related to interferon response, inflammation, and apoptotic signaling
   ([98]Fig. 3F). This was accompanied by transcriptional changes,
   indicated by enhanced E2F and MYC pathway, suggesting rewiring of
   transcriptional programs. Immunosuppressive transforming growth
   factor–β (TGF-β) signaling was decreased in LUNG-19 and LUNG-35. Some
   of the observed changes, including the interferon signaling, were more
   pronounced in the presence of nivolumab. Upon addition of nivolumab,
   LUNG-35 demonstrated the most alterations, followed by LUNG-19, while
   in LUNG-25, the changes were not significant.

   In LUNG-35, we also detected considerable activity of perforin (PRF1)
   and granzyme B (GZMB) in the T cells amplified by ex vivo nivolumab,
   indicating enhanced immune cell–mediated cytotoxicity ([99]Fig. 3G).
   LUNG-19 also showed a slight increase in GZMB with nivolumab treatment,
   while changes were not observed in LUNG-25. In the two reactive
   patients, TCR sequencing revealed marked clonal expansion following
   nivolumab treatment ([100]Fig. 3, H to J), indicating the presence of
   specific T cell clones that recognized and responded to
   tumor-associated antigens and treatment. When further analyzing the
   cell populations with the specific TCR clones, we detected a unique
   pattern of prevalent cell type in each clone. LAIR2^+ TEM cells were
   highly significant in several TCR clones in LUNG-35 and less in LUNG-19
   while not being detected in LUNG-25 (fig. S3, E to G). These findings
   further supported our idea that LAIR2^+ T cell subsets could play a
   critical role in mediating antitumor responses also in the presence of
   nivolumab treatment.

Personalized responses to the combination of nivolumab and chemotherapy
revealed by a TOC platform

   Clinical studies in NSCLC suggest that overall survival (OS) benefits
   from single-agent anti–PD-(L)1 mainly in patients with ≥50% PD-L1
   expression, while patients with negative/low PD-L1 expression only
   occasionally benefit ([101]20, [102]21). However, anti–PD-1 blockade
   combined with chemotherapy (pemetrexed/carboplatin, PC) has shown more
   benefits (overall response rate, ORR: 58% in anti–PD-1 + PC and 33% in
   PC alone) and has become a standard-of-care treatment regardless of
   PD-L1 status for patients with no targeted therapy options ([103]22,
   [104]23). However, identifying the patients who will truly benefit from
   the treatment remains a challenge, considering the varied responses and
   potential for increased toxicities.

   To model the individual responses to anti–PD-1 and chemotherapy more
   physiologically, we used a microfluidic tumor-on-a-chip (TOC) system,
   integrating tumor organoids, ts-immune cells, and extracellular matrix
   (ECM). Tumor cells were positioned into the central channel of the chip
   and allowed to form organoids in the ECM to mimic the tumor
   microenvironment (TME). The previously attained ts-immune cells were
   added to the proximal side channel to model immune cell infiltration
   into the tumor from the circulatory. Drugs were added to the distal
   side channel to assess the drug efficacy and distribution into the
   tumor ([105]Fig. 4A). Immune cells were allowed to migrate and interact
   with the tumor cells for 48 hours before further analysis. Immune cell
   infiltration toward tumor organoids was further validated using a
   microwell plate, with infiltration into the organoids observed 5 hours
   after tumor-immune coculture (fig. S4A).

Fig. 4. TOC models primary immunotherapy resistance and reveals differential
responses to nivolumab combined with chemotherapy.

   [106]Fig. 4.
   [107]Open in a new tab

   (A) TOC design for IO testing. The chip contains a middle channel where
   tumor organoids are seeded in Matrigel (ECM) and separated from two
   side channels, where ts-immune and selected drugs were added. A
   bright-field image shows ts-immune infiltration to the tumor site
   surrounding the organoids after 24 hours of seeding. (B) Live (AO,
   green) and dead (PI, red) fluorescence images of LUNG-19 organoids with
   or without ts-immune and subjected to different treatments after 48
   hours of coculture in the TOC. (C) Overview Sankey diagram showing
   tumor cell viability in control condition or in response to
   single-agent nivolumab (100 μg/ml) or its combination with 2.5 μM/375
   μM PC after 48 hours of coculture in the TOC with or without ts-immune
   cells. Values indicated in different colored boxes or linked by
   differently colored lines are the means of two to five technical
   replicates (in LUNG-19, the mean indicates technical replicates of two
   biological replicates) and represent the percentages of viable tumor
   cells in each patient (n = 8). **P < 0.01; ***P < 0.001; ns, not
   significant, unpaired t test. (D) Live tumor cell % quantification of
   LUNG-19 48 hours after addition of immune cells and treatments. The
   median is indicated as a black line. (E) Dot plot showing the
   expression changes of activating or inhibitory checkpoints and ligand
   genes in tumor cells after interaction with ts-immune with or without
   nivolumab (100 μg/ml, 48 hours). IFN-γ expression by cytotoxic immune
   cells is shown. (F) Dot plot showing the expression of immune
   checkpoints and activation genes in T cells upon interaction with tumor
   cells with or without nivolumab (100 μg/ml, 48 hours). In (E) and (F),
   circled dots indicate statistical significance.

   First, we used the platform to evaluate responses to nivolumab alone or
   combined with chemotherapy (PC) across eight patients with NSCLC from
   various mutational subgroups. Notably, all patients exhibited low
   baseline PD-L1 expression in both clinical samples and organoids (table
   S1), clinically associated with poor single-agent nivolumab response.
   Within the TOC, introducing ts-immune cells enhanced the eradication of
   tumor organoids, although baseline immune cell–mediated killing was
   highly variable among patients ([108]Fig. 4, B to D). The addition of
   nivolumab alone together with the ts-immune did not significantly
   enhance organoid killing, which aligns with expectations for patients
   exhibiting low/negative PD-L1 status. However, the introduction of
   nivolumab in combination with PC revealed a pronounced response in 38%
   (3 of 8) of the individuals (ORG-2: 20%, ORG-55: 13%, and LUNG-19: 11%
   increase in killing efficacy compared to PC alone; [109]Fig. 4C).
   LUNG-25, which showed poor immune activation and nivolumab effects in
   the mechanistic studies, demonstrated a lack of treatment response. In
   contrast, LUNG-19, characterized by high-TMB and more pronounced immune
   activation, was responsive. This suggests that the initial capacity for
   antitumor immune recognition and activation may serve as a predictive
   factor for the effectiveness of combining nivolumab with chemotherapy.
   Our findings support the clinical understanding that certain patients,
   including those with low PD-L1 expression, may derive substantial
   benefits from combining PD-1 blockade with chemotherapy. In addition,
   ex vivo testing could prove beneficial in identifying these
   individuals.

Immune checkpoint alterations after tumor stimulation and nivolumab treatment

   While we observed primary resistance to single-agent nivolumab in the
   PD-L1–negative/low patient samples, in the light of recent studies on
   IO combinations, these patients might respond more favorably to IO
   combination therapies involving other immune checkpoints. Consequently,
   we investigated the effects of immune stimulation and nivolumab
   treatment on checkpoints and ligands beyond PD-(L)1 (fig. S4B). In
   tumor cells, immune stimulation revealed an interplay of both immune
   activating and inhibitory signals ([110]Fig. 4E), including
   up-regulation in immune-activating signals such as the FAS death
   receptor gene and the NK-activating ligand MICB. The induction of human
   leukocyte antigen genes suggested enhanced visibility of tumor cells to
   T cells. In addition, several inhibitory molecules, such as the TIGIT
   ligand CD155 (PVR), as well as TIM-3 ligands Galectin-9 (LGALS9) and
   CEACAM1, were also induced. We also noted an up-regulation of the
   IFN-γ–induced transcription factors STAT1 and the immunosuppressive
   STAT3, also important for PD-L1 (CD274) transcription. The effects of
   nivolumab on the tumor cells were minor, with only STAT1 showing
   significant up-regulation.

   In T cells, immune stimulation by the tumor led to the up-regulation of
   many co-stimulatory receptors, including ICOS, TNFRSF18, TNFRSF4, and
   TNFRSF9/4-1BB ([111]Fig. 4F). Nivolumab specifically increased
   co-stimulatory receptor CD27 expression in LUNG-19. Following immune
   cell stimulation, an inhibitory phenotype was observed in LUNG-19 and
   LUNG-25, characterized by the up-regulation of TIM-3 (HAVCR2), with
   TIGIT being uniquely up-regulated in LUNG-25. CTLA-4 and LAG-3 were
   induced in all patients after tumor interaction. Notably, nivolumab
   treatment led to further overexpression of LAG-3 in LUNG-19, suggesting
   the potential for cotargeting strategies with LAG-3–targeting
   antibodies. CTLA-4, TIM-3, and LAG-3 were overexpressed in LUNG-25
   following the exposure to nivolumab, although the up-regulation was not
   significant. Regarding NK cells, the impact of nivolumab on checkpoints
   and their ligands was more variable and less pronounced (fig. S4C). Our
   data suggested, that while many of the immune checkpoints were
   activated upon tumor stimulation, nivolumab only had mild impact on the
   well-known checkpoints.

Up-regulation of Cbl-b E3 ubiquitin ligase induced by nivolumab treatment

   Although nivolumab’s influence on the well-established extracellular
   immune checkpoints and their ligands appeared modest, our further
   analysis of the DEGs identified Casitas B-lineage lymphoma-b (Cbl-b),
   an E3 ubiquitin ligase, as a widely up-regulated gene across a range of
   immune cells. Recognized for its crucial role as an intracellular
   checkpoint regulating both adaptive and innate immune responses, the
   potential of Cbl-b in enhancing antitumor immunity has lately gained
   attention ([112]24). Recent preclinical studies have highlighted the
   possibility that inhibiting Cbl-b could substantially enhance antitumor
   immune activities ([113]25), prompting the initiation of early clinical
   trials to explore its viability as a target for cancer immunotherapy
   ([114]NCT05107674, [115]NCT05662397).

   In our data, CBLB expression was widespread among various immune cell
   subsets, with a distinct and broader expression profile compared to the
   well-known immune checkpoints ([116]Fig. 5A). This included high
   expression in the vital effector subtypes of CD8^+ T cells and NK
   cells, along with the previously identified tumor-activated population
   of LAIR2^+ γδ T cells. Upon nivolumab treatment, widespread
   up-regulation of CBLB was observed in CD8^+ T cells, LAIR2^+ γδ T
   cells, and LAIR2^+ TEMs, suggesting that the most tumor-reactive immune
   cell subtypes were affected ([117]Fig. 5B and fig. S5A). These findings
   suggest that nivolumab was inducing Cbl-b up-regulation in key
   antitumor immune cell subtypes, highlighting potential cotargeting
   strategies in the PD-L1–low/–negative patient cohort.

Fig. 5. CBLB expression in tumor and immune cells.

   [118]Fig. 5.
   [119]Open in a new tab

   (A) Expression profile for CBLB across different immune cell types
   compared to other well-known and targetable immune checkpoints. CBLB
   was found broadly expressed in the relevant subtypes of proliferating
   NK cells, LAIR2^+ cells, and effector T cells. (B) Expression dot plot
   showing CBLB altered by the antitumor immune stimulus and nivolumab
   (100 μg/ml, 48 hours). (C) Kaplan-Meier curve showing the correlation
   between low or high levels of CBLB expression and the survival
   probability of patients with lung squamous carcinoma (LUSC). The data
   were extracted from the Human Protein Atlas database. (D) Kaplan-Meier
   curve showing the correlation between low or high levels of CBLB
   expression and the survival probability of patients with lung
   adenocarcinoma (LUAD). The data were extracted from the Human Protein
   Atlas database. (E) Immunohistochemistry staining of Cbl-b expression
   in lung cancer tissues. Data were retrieved from the Human Protein
   Atlas database.

   The clinical or cellular implications of Cbl-b expression in lung
   cancer remain largely unexplored. Analysis of the Cancer Genome Atlas
   Program expression data revealed a significant correlation between high
   CBLB expression and reduced survival in lung squamous cell carcinoma,
   with a similar yet nonsignificant trend observed also in lung
   adenocarcinoma ([120]Fig. 5, C and D). To enhance our understanding of
   Cbl-b’s role in the TME, we next investigated its expression patterns
   to determine whether Cbl-b is predominantly localized in tumor or
   immune cells. Tissue sections, as well as sc-data, indicated that Cbl-b
   was expressed in both immune and tumor cells but with a higher
   magnitude in immune cells ([121]Fig. 5E, URL:
   [122]https://proteinatlas.org/ENSG00000114423-CBLB/pathology/lung+cance
   r, and fig. S5, A and B). As a summary, Cbl-b expression was associated
   with reduced survival in lung cancer, and its predominant localization
   in immune cells underscored its potential as a target for enhancing
   antitumor immunity.

Synergistic effect of Cbl-b inhibition and nivolumab in selected patients

   Given the limited response to single-agent nivolumab and the prominent
   up-regulation of Cbl-b upon treatment, we investigated whether Cbl-b
   inhibitor could enhance the response to nivolumab in the TOC model. The
   treatment efficacy was tested in samples of five patients with NSCLC
   (ORG-54, ORG-55, LUNG-19, LUNG-24, and LUNG-25). Among them, ORG-55 and
   LUNG-19 had previously shown an increased T cell killing response to
   the combination of nivolumab with chemotherapy, unlike ORG-54, LUNG-24,
   and LUNG-25. We found that neither single-agent nivolumab nor the Cbl-b
   inhibitor alone significantly increased immune cell–mediated tumor
   killing or caused cytotoxicity to the tumor cells in the absence of
   immune cells ([123]Fig. 6, A to D). However, the combination of
   nivolumab and Cbl-b inhibitor demonstrated a synergistic effect,
   enhancing immune cell killing capability in selected patients, notably
   more than in conditions without immune cells. This synergy was observed
   with 40% (2 of 5) of the patients showing a marked benefit from the
   combined treatment ex vivo (LUNG-19 and LUNG-24), while 60% (3 of 5)
   did not respond (ORG-54, ORG-55, and LUNG-25). Notably, LUNG-19 had
   earlier responded positively to the combination of nivolumab with
   chemotherapy, suggesting broader IO benefits. Meanwhile, LUNG-25
   remained unresponsive, whereas LUNG-24 showed a unique response
   exclusively to the combination of nivolumab and Cbl-b inhibitor.

Fig. 6. Nivolumab combined with Cbl-b inhibition improves anti–PD-1 response
in a subset of patients with NSCLC.

   [124]Fig. 6.
   [125]Open in a new tab

   (A) Live (AO, green) and dead (PI, red) example fluorescence images of
   LUNG-19 organoids with or without ts-immune and subjected to nivolumab
   (100 μg/ml) or its combination with 50 nM Cbl-b inhibitor after 48
   hours of coculture in the TOC. Images of control and nivolumab
   conditions with or without ts-immune are the same as in [126]Fig. 4B.
   Scale bar, 100 μm. (B) Overview Sankey diagram showing tumor cell
   viability in control condition or in response to single-agent nivolumab
   (100 μg/ml) or its combination with 50 nM Cbl-b inhibitor after 48
   hours of coculture in the TOC with or without ts-immune cells. Values
   indicated in different colored boxes or linked by differently colored
   lines are the means of two to five technical replicates (in LUNG-19,
   the mean indicates technical replicates of two biological replicates)
   and represent the percentages of viable tumor cells in each patient (n
   = 5). Indicated values of control and nivolumab conditions with or
   without ts-immune in ORG-54, ORG-55, LUNG-19, LUNG-24, and LUNG-25 are
   the same as in [127]Fig. 4C. **P < 0.01; ns, not significant, unpaired
   t test. (C and D) Live tumor cell % quantification of combination
   treatment nonresponder (ORG-55) versus responder patient (LUNG-19). In
   ORG-55, scatter plots represent technical replicates, while in LUNG-19,
   they represent technical replicates of two biological replicates. The
   median is indicated as a black line. ****P < 0.0001; ns, not
   significant, unpaired t test. (E) Heatmap showing the secretion of 12
   different cytokines in nonresponder LUNG-25 compared to the responder
   LUNG-24. Media were collected from the treatment conditions after 48
   hours of coculture in the TOC. (n = 2 replicates) expressed as log[2]
   fold-change (L[2]FC) relative to ts-immune control.

   Cytokines, as critical signaling proteins, orchestrate immune responses
   against tumor cells within the TME/TIME. To investigate the impact of
   nivolumab and Cbl-b inhibition on immune cell activity within the chip,
   we analyzed the secreted cytokines extracted from the TOC, comparing
   the cytokine profiles between a combination responder (LUNG-24) and a
   nonresponder (LUNG-25) model. Intriguingly, in the responder patient
   sample, we detected elevated levels of IFN-γ–induced cytokines, such as
   IP-10, MIP-1a, and MIP-1b, known to promote T cell recruitment and
   activation ([128]Fig. 6E). Conversely, the nonresponder, LUNG-25,
   exhibited high levels of IL-13, recognized for its role in inhibiting
   antitumor immune response ([129]26). Increases in IL-13 levels were
   induced by both nivolumab and Cbl-b inhibitor, but a synergistic
   increase was observed with their combination.

   The contrast in the observed cytokine profiles may provide valuable
   insights into the mechanisms underlying specific antitumor responses in
   certain patients. In summary, our results underscore the potential of
   combining nivolumab with a Cbl-b inhibitor to enhance therapeutic
   responses in a subset of patients with NSCLC. Furthermore, this study
   highlights the utility of ex vivo and cocultures and OOC models in
   unraveling individual patient mechanisms and responses to IO
   strategies.

DISCUSSION

   ICIs have become the standard of care in the treatment of NSCLC.
   However, current ICIs benefit only a fraction of patients and lack
   accurate biomarkers for predicting treatment responses. This challenge
   is compounded by the rapid development of numerous new IO drugs and
   drug combinations, with an insufficient number of patients available to
   test their efficacy in a real-world setting. To navigate these
   complexities, the integration of functional precision medicine
   methodologies, such as personalized TOCs, emerges as a promising
   strategy. These approaches not only offer the potential to
   simultaneously evaluate multiple treatment options but also promise to
   expedite the preclinical phases of IO and biomarker development. While
   ex vivo models with immune cells have previously been established, they
   often lack personalization or throughput, limiting their utility for IO
   drug testing. Previous approaches have heavily relied on unmatched
   immune cells or TILs as sources of immune cells, undermining the
   systemic effects taking place in the cancer immunity cycle.
   Furthermore, recent study on lung and colorectal tumors showed that
   tumors can carry substantial amounts of CD8^+ bystander TILs, whose
   presence might not be directly relevant to the antitumor immune
   response ([130]27). Small clinical samples can also result in a limited
   yield of tumor cells and TILs for functional testing.

   Here, we developed a functional ex vivo IO testing platform, which
   incorporates patient-matched tumor and immune cells, and allows for
   high-throughput testing and future inclusion of personalized TME
   components such as fibroblasts and endothelial cells. One of the unique
   features of our platform is the systemic utilization of immune cells
   stimulated with autologous tumor organoids rather than through
   nonspecific activation using antibodies. Our findings align with those
   of Dijkstra et al. ([131]28), demonstrating the induction of IFN-γ
   production by CD8^+ T cells in around 50% of patients following the
   coculturing of circulating immune cells with matching organoids. By
   combining stimulation assays with single-cell methods, we were able to
   study and identify individual antitumor immune response and resistance
   mechanisms at the cellular level. This approach could be especially
   useful for identifying previously unknown IO targets and biomarkers in
   the future. Specifically, we found that the activation and clonal
   expansion of LAIR2^+ cells appeared to correlate with effective
   antitumor activity, whereas intracellular immune checkpoint Cbl-b was
   identified as an interesting therapeutic target.

   Previous studies have shown that genetically engineered mice with
   CBLB^−/− T cells exhibit hyperactivated T cell and autoimmune responses
   ([132]29, [133]30). In the context of antitumor immunity, the same
   knockout was shown to downmodulate immune suppression and induce
   IFN-γ–dependent antitumor immunity ([134]31). In addition, genetic
   inhibition of Cbl-b has been proven to prevent chimeric antigen
   receptor (CAR) T cell exhaustion. In our study, we observed that CBLB
   expression was widespread in multiple effector immune cell subtypes,
   with a notable increase in the most tumor-reactive immune cell types
   following nivolumab treatment. Although Cbl-b has been loosely linked
   to the regulatory pathways of PD-L1 ([135]32), and the first-in-human
   studies with Cbl-b inhibitors have recently commenced ([136]NCT05107674
   and [137]NCT05662397, the latter in combination with anti–PD-1
   cemiplimab), further research is necessary to clarify Cbl-b’s precise
   role within the PD-1/PD-L1 signaling pathway.

   Last, by using the TOC model, we successfully mimicked primary
   nivolumab resistance and identified potential responders to nivolumab
   in combination with chemotherapy. The TOC system was also used to trial
   the combination IO treatment using nivolumab together with Cbl-b
   inhibitor, which proved effective in a subset of patients. Notably, the
   positive responses observed with either nivolumab and chemotherapy or
   with Cbl-b inhibitor occurred despite the tumors exhibiting
   negative/low PD-L1 expression, underscoring the limitation of PD-L1
   immunohistochemistry as the only pretreatment adjunct currently in
   clinical use. In addition, exploring therapeutic strategies that target
   the observed cytokines could optimize antitumor responses from ICIs,
   offering new tools for personalized cancer therapy. The development of
   “monitoring biomarkers,” derived from cytokine profiles obtained
   through ex vivo IO testing platforms, holds promise for advancing
   personalized, more effective immunotherapy treatments.

   Despite the innovative approach of integrating patient-specific tumor
   and immune cells to predict immunotherapy outcomes, this study
   acknowledges certain limitations. First, the ex vivo platform may not
   fully recapitulate the complexity of the TME and its influence on
   immune cell function. In addition, the study’s findings are derived
   from a limited number of patient samples, which may restrict the
   generalizability of the results. Further co-clinical validation in a
   larger cohort is necessary, both for the platform and for the
   identified IO drug combination, to confirm the platform’s predictive
   value and the identified biomarkers for immunotherapy response. As a
   summary, we describe a patient-derived ex vivo IO testing platform that
   enabled the detection of synergistic combinations of anti–PD-1 therapy
   with Cbl-b inhibitor or chemotherapy, effectively overcoming poor
   responses to single-agent anti–PD-1 treatment.

MATERIALS AND METHODS

Reagents and resources

   All used reagents and resources (including cell culture reagents,
   antibodies, and drugs) can be found from table S3: Reagents and
   resources.

Patient samples

   Patient primary tumor and blood samples were collected in accordance
   with the Declaration of Helsinki from consented patients with lung
   cancer with concurrent BioBank deposit. For the study, Helsinki
   University Hospital (HUS) IRB has been granted (HUS/8/2022) with a
   statement from the institutional ethical board HUS/970/2021. Peripheral
   blood was collected into lithium heparin vacuum tubes according to HUS
   Laboratory’s standard protocols in a laboratory chosen by the patient.
   Patients were recruited systematically from a tertiary hospital’s
   thoracic surgery service. All patients underwent a planned surgery for
   malignant lung tumor following a multidisciplinary team consensus
   meeting. Of 15 patients included in the study, 12 were treatment-naïve
   local primary lung cancers, 1 was primary lung cancer operated after
   neoadjuvant therapy, and 2 were lung metastasis of colorectal cancer
   who underwent metastasectomy. Furthermore, tumors with interesting
   oncogenic mutations (such as EGFR, KRAS, and others) were included.

Peripheral blood lymphocyte isolation

   PBMCs were isolated from peripheral blood by SepMate density gradient
   centrifugation according to the manufacturer’s instructions. PBMCs were
   then assessed for viability and cultured or cryopreserved.

Cryopreservation of PBMCs

   Isolated PBMCs were resuspended in freezing medium at a concentration
   of 5 to 10 × 10^6 cells/ml and placed in cryo freezing container at
   −80°C for 24 hours to allow even cooling. The following day, the
   samples were moved to −150°C for long-term storage.

Tumor organoid establishment and culture

   Primary lung tumor surgical specimen samples were stored in wash medium
   and delivered to the laboratory (samples were given the codes ORG- and
   LUNG- since they were established in two different laboratories) within
   an hour since the tissue was collected. Upon arrival, a small piece of
   the tissue was cut and quickly put into a freezing tube to be stored at
   −80°C for future sequencing if required. The rest of the tissue was
   minced using surgical scalpels to 1 to 2 mm^3 fragments and further
   dissociated with digestion medium on a shaker at 37°C overnight. The
   digestion was stopped by one wash with 20% fetal bovine serum (FBS) in
   cold Dulbecco’s modified Eagle’s medium (DMEM) and followed by two
   washes with phosphate-buffered saline (PBS). To increase the chance of
   successfully establishing tumor organoids from most of the patients and
   be able to isolate tumor cells and fibroblasts, we developed a two-step
   establishment culture system. First, we embedded half of the
   cells/fragments into 3D culture using Matrigel. Thirty minutes after
   embedding, Matrigel solidified and special organoid medium was added.
   Second, the other half of cells/fragments were established as 2D
   cultures by resuspension with standard medium and then plated in two
   wells of a tissue-treated six-well plate. After 4 to 6 days and on the
   basis of cells status following the 2D culture initiation, every well
   was resuspended, and suspension was transferred to the adjacent well,
   while the original wells were supplemented with more media. This step
   was repeated after 2 days with the new well ending up with six or more
   wells per sample. This allows for enrichment of epithelial tumor cells
   or fibroblasts in different wells. The fibroblasts were then collected
   and expanded in standard medium. Meanwhile, confluent epithelial cells
   at passage zero or one were collected and embedded in Matrigel allowing
   organoid formation.

   Organoid growth was assessed with a Nikon Eclipse TS100 microscope. On
   the basis of their growth, the organoids were passaged once every 1 to
   2 weeks at a 1:2 split ratio. Organoids were dissociated with prewarmed
   TrypLE express for 5 to 10 min at 37°C. To stop the TrypLE express
   effect, cold 20% FBS in DMEM/F12 was added and samples were centrifuged
   (300g for 5 min at 4°C), which was followed by a wash with cold
   advanced DMEM/F12. At this step, the organoids were broken into single
   cells or aggregate of three to five cells that were resuspended with
   Matrigel. The Matrigel suspension was domed and solidified at 37°C for
   30 min and then overlaid with special organoids medium. The presence of
   tumor-bearing driver mutations in the organoids was validated using
   whole-exome sequencing.

Organoid-PBMC coculture

   For the generation of tumor-stimulated immune cells (ts-immune), we
   have adapted a protocol that was established by Cattaneo et al.
   ([138]13). Two days before coculture, tumor organoids (1 to 5 × 10^4
   for every 1 × 10^6 PBMC) were isolated from Matrigel by mechanical
   dissociation followed by 15 min of incubation with preheated dispase (2
   mg/ml in PBS) at 37°C. The dispase was deactivated by 0.5 M EDTA,
   washed with PBS, resuspended with complete organoid media, and plated
   in tissue culture–treated six-well plate. Organoids were cultured for
   24 hours at 37°C. One day before coculture, the organoids were
   stimulated overnight with IFN-γ (200 ng/ml) to enhance their antigen
   presentation. Ninety-six–well U-bottom plates were coated with
   anti-CD28 antibody (5 μg/ml; to provide co-stimulatory signals) in PBS
   (50 μl per well). The plate was wrapped with parafilm and incubated for
   24 hours at 4°C. Cryopreserved PBMCs were thawed in T cell thawing
   medium followed by centrifugation. The PBMCs were then incubated for 15
   min at 37°C in 5 ml of T cell thawing medium with 1:1000 benzonase and
   then washed with T cell thawing medium. The cells were resuspended at 2
   × 10^6 per ml in T cell medium with IL-2 (150 U/ml) and incubated
   overnight in a 15-ml falcon tube at 37°C.

   The next day, organoids stimulated with IFN-γ were collected and
   resuspended with 1 ml of TrypLE Express and combined with the remaining
   cells that adhered to the bottom of the six-well plate for 5 min at
   37°C and then washed with PBS. Dissociated organoids were then
   resuspended at 5 × 10^4 cells per ml of T cell medium.
   Overnight-incubated PBMCs were washed with PBS and resuspended at 1 ×
   10^6 cells/ml in T cell culture medium and supplemented with IL-2 (300
   U/ml) and anti–PD-1 (40 μg/ml). Equal volumes of PBMCs and dissociated
   organoids were mixed in a ratio of 20:1. Anti–CD-28–coated 96-well
   plates were washed two times with PBS. Tumor cells and PBMCs were
   cocultured by plating 200 μl of the dissociated organoid-PBMC
   suspension per well. Coculture medium was refreshed three times a week
   including IL-2 (300 U/ml) and anti–PD-1 (40 μg/ml). Seven days after
   coculture, PBMCs were collected and counted and then restimulated for
   another 7 days with freshly isolated organoids (which had been
   stimulated with IFN-γ 24 hours before coculture) by seeding 2 × 10^5
   PBMCs with 10^4 dissociated organoids per well of anti-CD28–precoated
   plate. By the end of coculture, ts-immune were collected and
   cryopreserved or used for downstream analysis and tumor killing assay.
   Ts-immune cells were cryopreserved at 5 × 10^6 cell/ml in cold 10%
   dimethyl sulfoxide in human male serum and then placed in freezing
   container at −80°C and transferred to −150°C the next day.

Evaluation of tumor reactivity using IFN-γ

   For the test condition, 1 × 10^5 of ts-immune were isolated from the 14
   days of coculture, and for the negative control, 1 × 10^5 of
   unstimulated PBMCs were used. Different conditions were pelleted at
   330g for 5 min at 4°C and washed twice with fluorescence-activated cell
   sorting (FACS) buffer. After the washes, cell-surface staining was
   performed by adding 1:20 anti–CD-4–fluorescein isothiocyanate, 1:20
   anti–CD-3–PerCP–Cy5.5, 1:200 anti–CD-8–V450 antibodies, and 1:1000
   near-infrared viability dye in FACS buffer for 30 min. The cells were
   washed with FACS buffer twice, fixed and permeabilized for 20 min on
   ice, washed twice with 1× Perm/Wash buffer, and intracellularly stained
   with 1:40 anti–IFN-γ–APC in Perm/Wash buffer for 30 min at 37°C. The
   cells were then twice washed with Perm/Wash buffer, resuspended in FACS
   buffer, and analyzed by BD FACS VERSE.

Drug testing using TOC

   The AIM idenTx 3 microfluidic device was used. Tumor organoids were
   mechanically isolated from Matrigel and disassociated to single cells
   or aggregates of two to five cells by 10 to 15 min of incubation at
   37°C with TrypLE. Disassociated tumor cells were resuspended at 2 ×
   10^3 cells/μl with 90% cold Matrigel, and 10 μl of tumor cell Matrigel
   suspension was loaded to the middle channels of the AIM device.
   Matrigel was allowed to solidify for 25 min at 37°C followed by
   addition of 100 μl of special medium to the right and left top-bottom
   media channels. Tumor cells in Matrigel within the microfluidic chip
   were then allowed to form organoids for 48 hours at 37°C. To set up the
   Solid-IO cocultures and screen for immunotherapy responses, special
   medium was aspirated from the media inlets at day 2, and 100 μl of T
   cell medium with or without 8 × 10^4 ts-immune cells were seeded to the
   inlet of the right-top channel and another 100 μl of T cell medium to
   the left-bottom channel of the device. This was followed by the
   addition of 100 μl of special medium with or without selected drugs to
   the inlets of the right top and bottom media channels. When all
   components were loaded to the microfluidic chip, the device was placed
   for 48 hours in a 37°C incubator on a rocker with a speed of 1 rpm and
   tilting angle of 14° to create a flow. The used drugs and their
   concentrations are nivolumab (100 μg/ml), Cbl-b inhibitor (50 nM/ml),
   and chemotherapy as combination of carboplatin (375 μM/ml) + pemetrexed
   (2.5 μM/ml).

Coculture killing assay and multiplexed scRNA-seq readout

   Tumor organoids were isolated using dispase (5 U/ml) for 10 to 15 min
   followed by EDTA deactivation and washes. The organoids were then
   resuspended with 2 ml of special medium per well and plated in a
   six-well plate and then incubated at 37°C and 5% CO[2]. After 48 hours,
   the organoids were collected, and an aliquot of the suspension was
   isolated and then treated with TryplE to dissociate the organoids into
   single cells to determine tumor single-cell count within the original
   organoid suspension. Tumor organoids were plated at 50 000 cells per
   well in 200 μl of special medium on a 48-well plate without or with
   ts-immune cells in T cell medium (1:4 target to effector ratio) and
   without or with nivolumab (100 μg/ml). For conditions with tumor
   organoids only, 200 μl of T cell medium was added to reach 400-μl final
   volume. For conditions with ts-immune, IL-2 (10 ng/ml) was added.
   Cocultures were then incubated at 37°C and 5% CO[2] for 48 hours. The
   coculture was performed in duplicates for each condition. Experiments
   were done with tumor organoids, ts-immune, and PBMCs of three patients.

   After 48 hours of coculture, cells from each well were dissociated into
   single cells using TryplE followed by two to three washes in 10 ml of
   PBS. Depending on viability, the cells were resuspended with 100 μl of
   cold washing buffer 1 or washing buffer 2, and then 10 μl of TruStain
   FcX blocking reagent was added, and cells were blocked for 10 min at
   +4°C. A unique TotalSeq-C hashing antibody was added to each sample (2
   μl/2 μg per sample), and cells were incubated for 30 min at +4°C. The
   cells were then washed three to five times with 3.5 ml of washing
   buffer 1 or washing buffer 2, and then the samples were combined in
   cold PBS + 0.04% bovine serum albumin and proceeded to scRNA-seq.

   Single-cell gene expression profiles were studied using 10x Genomics
   Chromium Single Cell 5’ Gene expression with Feature Barcoding
   technology platform. The Chromium Single Cell 5’RNAseq run, and library
   preparation were done using the Chromium Next GEM Single Cell 5’ Immune
   Profiling with Feature Barcoding technology version 2 chemistry. The
   sample libraries were sequenced on the Illumina NovaSeq 6000 system
   using the following read lengths: 26 base pair (bp) (read 1), 10 bp (i7
   index), 10 bp (i5 index), and 90 bp (read 2).

Acknowledgments