A 3D tumor spheroid model with robust T cell infiltration for evaluating immune cell engagers

Summary

A strong interest in drugs targeting the tumor microenvironment (TME) necessitates new experimental systems that incorporate key TME components. Compared to traditional 2D cell lines, 3D ex vivo spheroids from patient-derived xenograft (PDX) materials may better capture patient tumor characteristics. We developed and validated a 3D tumor spheroid model from non-small cell lung cancer (NSCLC) PDXs to enable T cell infiltration. Histologic and transcriptomic analysis suggested that tumor spheroids closely recapitulate the source PDX tumor tissues. Consistent T cell infiltration into tumor spheroids was achieved using a well-established magnetic nanoparticle technology, which maintained T cell function and tumor-killing activity. Drug treatment studies with immunotherapy agents also demonstrated the potential scalability of 3D tumor-T cell spheroids in assessing drug activity, including tumor viability and cytokine secretion. This platform provides a useful tool for evaluating drug candidates that can be translated to patient tumor responses related to both tumor intrinsic and TME factors.

Graphical abstract

Highlights

• •We developed a 3D tumor spheroid model from lung cancer patient-derived xenografts
• •The model enabled robust T cell infiltration and preserved T cell cytotoxic functions
• •Histology and RNA-seq showed that tumor spheroids closely resembled source tumors
• •Proof-of-concept experiments showed this platform’s utility in preclinical drug testing

Biological sciences; Biotechnology; Natural sciences; Tissue Engineering

Introduction

Recent scientific advances have revealed that the tumor microenvironment (TME) plays a major role in determining tumor progression and therapeutic response.^1,2 As a result, there is immense interest in moving toward a more holistic understanding of tumor-TME interactions as opposed to studying tumor cells separately from other cell types. This has led to an exponential increase in the number of drug candidates with mechanisms of action that target TME components such as immune cells and fibroblasts.³ There has also been some therapeutic success in the clinic, with the most prominent examples being immune checkpoint inhibitors such as anti-PD1 and anti-CTLA4, but there is still much ongoing research focused on improving response rates and overcoming resistance. Despite this paradigm shift, recapitulating the TME in a laboratory setting remains challenging. Most preclinical experiments today still make use of tumor cell focused models that were established many decades ago, suggesting that there is a critical need to develop new models that can effectively reproduce the complex cellular composition and architecture of a human tumor.

Both basic cancer research and pre-clinical drug discovery rely heavily on immortalized cancer cell lines in two-dimensional (2D) culture. In efforts to simulate tumor-TME interactions, tumor cell lines are intermixed with other cell types, such as T cells, in 2D co-culture assays.^4,5 These systems have offered useful insights, are easy to set up and are amenable to high throughput applications. However, they do not adequately capture the intra- and inter-patient tumor heterogeneity and lack many tumor features such as 3D cellular organization, hypoxia, and rewired metabolism.⁶ In comparison, in vivo animal tumor models can reconstitute a physiologically relevant TME but are limited by their low throughput and poor translatability to humans.⁷ The emergence of patient-derived xenografts (PDX) has enabled the maintenance of human tumors as a clinically relevant research tool. Still, their propagation requires immunodeficient animal hosts that lack key immune cell populations.

New developments in 3D humanized tumor models may offer an attractive intermediate between traditional 2D cell culture and animal models. 3D models recapitulate important features such as physiological cell-cell interactions and nutrient diffusion gradients.^8,9 Studies have shown that 3D culture of cancer cell lines led to significant differences in cell signaling,¹⁰ metabolism,^11,12 and response to drug treatment compared to 2D culture.¹³ For example, in side-by-side CRISPR knockout screens performed on tumor cells cultured as 3D spheroids versus 2D monolayers, 3D spheroids identified more cancer-relevant hits from driver pathways known to be essential for tumor growth, including p53, TGF-b, and hypoxia signaling.¹⁴ In other side-by-side experiments, 3D tumor spheroids also display responses to genetic and pharmacologic perturbation that more closely recapitulate in vivo responses.^13,15

Furthermore, 3D tumor models have also found use in studying the role of the tumor immune microenvironment.^{5,16,17,18,19} For example, promising studies provided proof-of-concept evidence that 3D models derived from patient tumors can predict patient response to immunotherapy and provide additional mechanistic insights such as cytokine secretion.^20,21,22 In other mechanistic studies, colorectal cancer cell line-derived spheroids were co-cultured with T cells and NK cells to evaluate TME relevant targets such as MICA/B as therapeutic antibody targets.⁴ MICA/B bind to activating NKG2D receptors on NK cells and T cell subsets. The addition of an anti-MICA/B antibody reduced NKG2D engagement but stimulated antibody-dependent cell cytotoxicity (ADCC), resulting in a net increase in NK cell infiltration and cytotoxicity.⁴ This demonstrates the potential of 3D models for understanding nuanced therapeutic mechanisms. Despite the promise offered by these tumor-immune cell 3D models, technical limitations still exist, including the tendency for most immune cells to interact only with the edge of tumor spheroids rather than penetrate the spheroid mass.^23,24 Focusing on the problem statement, we sought to develop a robust 3D tumor-immune cell co-culture model where immune cells can be consistently incorporated into human tumor spheroids, which may better recapitulate tumor-immune cell interactions in a 3D context.

In this study, we describe a 3D tumor spheroid system that combines PDX-derived tumor cells and PMBC-derived T cells (from unrelated donors). Using commercially available magnetic nanoparticles designed for 3D spheroid formation,^25,26,27 we generated tumor spheroids with or without T cell incorporation. We characterized the spheroids by flow cytometry, histology, and RNAseq in comparison with the source PDX tumors. We also demonstrate that the model shows significant infiltration by functional T cells that display tumor killing activity. Finally, to explore the applications of this model in drug testing, we performed proof-of-concept drug studies using a T cell-engaging bispecific antibody and checkpoint inhibitor. In summary, we have established a flexible, scalable, and reproducible tumor-T cell spheroid model with potential applications in the pre-clinical testing of cancer therapeutics.

Results

Establishment of a non-small cell lung cancer patient-derived xenograft derived spheroid system

Our overall aim is to develop an ex vivo tumor spheroid model from non-small cell lung cancer (NSCLC) PDXs with opportunities for incorporating other non-tumor cell types. As a proof-of-concept, we chose to focus on T cells in order to build a model for evaluating drug response to T cell-targeting immunotherapies. We first established a protocol to dissociate and isolate human tumor cells from freshly harvested NSCLC PDXs from mice (Figure 1A). To ensure the quality and purity of tumor cells, we included specific steps in the protocol to remove dead cells and deplete mouse stromal cells, which consistently resulted in a human tumor cell purity of >97%.

Figure 1.

The spheroid system recapitulates the histology of neoplastic tissues

(A) Schematic diagram for the formation of ex vivo spheroids from PDX tumors and healthy donor PBMC-derived T cells. Dissociated PDX tumor cells and T cells are incubated with magnetic nanoparticles, which bind to the cell surface non-specifically. Cells are seeded, and a magnetic field is applied to induce spheroid formation. This method may be used to form tumor-only or tumor-T cell spheroids.

(B) H&E-stained images of tumor-only spheroids and their original PDX tissues. Far left panel shows lower magnification images of spheroids. Scale bar, 50μm.

(C) Time-lapse live microscopy images (left) and growth curves as measured by CellTiter Glo-3D (right) of tumor-only spheroids at different time points after seeding (Normalized to Day 1 signal). Please note the different scales of the y axes between tumor models due to variations in survival/growth duration. Scale bar, 300um. Data are presented as mean ± SEM, n = 3 spheroids.

(D) Representative immunofluorescence microscopy images of spheroids labeled for DAPI and Ki67. Spheroids were collected on Day 7 post-seeding. Scale bars, 250 μm. See also Figure S1.

To obtain T cells for incorporation into spheroids, we isolated T cells from unmatched healthy donor peripheral blood mononuclear cells (PBMCs) followed by 7 days of in vitro activation and expansion. In contrast to previously reported spheroid-T cell co-cultures, where T cells were added exogenously after tumor spheroids had formed,^24,28,29,30 where the pre-formed tumor spheroid may structurally impede T cell infiltration, we pursued a method that enables consistent T cell incorporation from the early stages of spheroid formation. Therefore, we adapted an established, magnetic-directed spheroid formation method,^25,26,27 where tumor cells and T cells were bound with specialized, biologically inert magnetic nanoparticles on their cell surface, and then seeded onto low attachment multi-well plates (Figure 1A). The plates were then inserted onto a magnetic device that placed a small magnet underneath each well, drawing both tumor and T cells to come together. Importantly, once a spheroid was formed after 24 h, the magnet was removed. Using this method, we successfully generated both tumor-only spheroids and tumor-T cell spheroids, showing that this method offers the flexibility to form spheroids with single or multiple cell types (Figure 1A).

Tumor spheroids resemble parent patient-derived xenograft tumors in their tissue morphology

As pre-clinical tumor models, it is important for the tumor spheroids to recapitulate the diverse disease features of different NSCLC tumors. Thus, we generated tumor-only spheroids from five NSCLC PDX models with different clinical and molecular features (Figure S1A). No T cells were added to enable the characterization of tumor cell features without any confounding effects introduced by T cells.

We first tested whether our spheroid culture could maintain the histological features and heterogeneity seen in their corresponding human NSCLC PDX tumors. To do so, we performed H&E and immunohistochemistry (IHC) staining on tumor spheroids and their source PDXs. H&E staining revealed that cellular morphology and other histological features are well-preserved across PDXs and their associated tumor spheroids, as reviewed by a pathologist (Figures 1B and S1A). Additionally, we performed IHC staining for human pan-cytokeratin (panCK) as a human tumor cell marker on PDXs and spheroids. Results showed that all PDX tumors show large areas of pan-CK positive areas interspersed with areas of mouse stoma that are negative for pan-CK. On the other hand, corresponding ex vivo spheroids stained strongly and uniformly for human pan-CK (Figure S1B), confirming that there was high purity of human epithelial tumor cells with no or negligible presence of mouse stromal cells within spheroids.

Patient-derived xenograft-derived tumor spheroids retain growth capacity in ex vivo culture

To understand PDX tumor cell survival and growth kinetics under our ex vivo culture conditions, we monitored the growth of tumor-only spheroids at different time points using live cell imaging and CellTiter-Glo3D assay (Figures 1C, S1C, and S1D). At a starting cell number of 10,000 tumor cells, the lung adenocarcinoma (LU-Ad) spheroid models, LU450, LU481, and LU554, were able to proliferate and grow for over one week (Figure 1C, LU554 shown in Figure S1D in blue curve). LU211, a lung squamous cell carcinoma (LU-Sq) model, showed growth in the first five days of culture, followed by the stabilization of cell number until Day 7. After Day 7, LU211 cell viability decreases sharply (Figure 1C). Similarly, another LU-Sq model, LU120, only maintained viability for up to 5 days (Figure S1C). This suggests that ex vivo growth and survival differences might be present between different NSCLC subtypes. LU-Ad has a preferential growth advantage under the current ex vivo growth conditions over LU-Sq. Additional optimization is ongoing to evaluate the growth potential, specifically for LU-Sq. Ki67 staining further showed that 20% and 45% of tumor cells within LU450 and LU481 are Ki67 positive, confirming that LU-Ad PDXs maintain proliferative capacity when cultured ex vivo as spheroids (Figure 1D).

Presence of magnetic nanoparticles does not affect tumor spheroid growth rate

The use of magnetic nanoparticles in the magnetic assembly of spheroids is important for our later experiments with T cell incorporation. While magnetic nanoparticles are not necessary for spheroid formation for most tumor models, they also help to maintain more consistent spheroid sizes and growth rates. The magnetic nanoparticles have been previously reported to have no impact on cellular processes.²⁵ In our system, we assessed whether the presence of magnetic nanoparticles led to any changes in spheroid growth. Using two of the PDX models, LU554 and LU481, we saw no significant difference in growth kinetics in the presence or absence of magnetic nanoparticles (Figure S1D).

Together, the above experiments showed that our spheroid culture conditions preserve the proliferative capacity and overall histological structure of PDX models ex vivo.

Patient-derived xenograft-derived tumor spheroids closely resemble parent patient-derived xenografts in gene expression analysis

Next, we asked whether spheroids recapitulate in vivo tumors in their gene expression profile. We performed RNA-seq on isolated human tumor cells from five PDX tumor models and their matched ex vivo spheroids (Figure 2A). To gain biological insights from the sequencing results, we first performed principal component analysis (PCA), which revealed that samples strongly cluster by patient of origin, regardless of whether samples came from freshly dissociated PDXs or PDX-derived spheroids (Figure 2B). In PDX models that had RNA collected from multiple passages (LU450, LU481, LU554, and LU211), samples of different passages still clustered closely together by PDX lineage. This indicates that our PDX-derived spheroids preserved strong lineage-dependent characteristics from their parental PDX tumors. Despite our limited number of evaluable PDXs, this supports that the interpatient tumor heterogeneity found in NSCLC can also be seen in our transcript-level analysis.

Figure 2.

Transcriptomics evaluation (RNA-seq) confirms overlap between PDX-derived tumor spheroids with their respective source PDXs

(A) Schematic diagram of RNA-seq experimental design. Human tumor cells were freshly isolated from PDXs and prepared for sequencing or cultured as spheroids for 7 days before RNA-seq.

(B) Principal component analysis of RNAseq profile from tumor spheroids and matched PDX tumor cells. Each PDX line may have multiple passages included in the experiment.

(C)Heatmap showing gene expression levels of 24 lung adenocarcinoma genes and 24 lung squamous cell carcinoma genes in PDXs and PDX-derived tumor spheroids.

(D) Volcano plots showing differentially expressed genes between tumor spheroids and matched fresh PDXs. DUSP1 and KLF2 are highlighted as differentially expressed genes that were common between all models.

(E) Summary table of the number of differentially expressed genes within each pair of PDX and tumor spheroids.

See also Figure S2.

Since the spheroid models included both LU-Ad and LU-Sq subtypes of NSCLC, we further assessed whether spheroids recapitulated distinct transcriptomic features associated with each subtype by examining their expression levels of a 24-gene LU-Ad signature and a 24-gene LU-Sq signature established from a human patient dataset.³¹ Indeed, LU-Ad spheroids show high expression of LU-Ad-enriched genes (e.g., NKX2-1 which encodes for TTF-1) and low expression for LU-Sq-enriched genes (e.g., KRT5 and TP63), and vice versa for LU-Sq spheroids (Figure 2C). Moreover, spheroids and their parental PDXs are highly concordant in their expression patterns of these LU-Ad and LU-Sq gene sets. Together, these results indicate that NSCLC spheroids maintain the transcriptional characteristics that are distinctive of their tumor subtypes.

Although the overall gene expression profiles of tumor spheroids and parental PDXs were very similar, we further explored whether specific genes or pathways were altered by ex vivo spheroid culture. To this end, we found a small but significant subset of genes that were differentially expressed within each pair of matched PDXs and spheroids (Figures 2D, 2E, S2A; Tables S1 and S2). Based on these differentially expressed genes, we also performed gene ontology analysis to explore potential biological differences at the pathway level. For example, compared to LU450 PDX tumors, the matched spheroids showed an up-regulation of genes related to “alcohol metabolic process” and a down-regulation of genes associated with “response to reactive oxygen species” (Figure S2B). Overall, there was no overlap in gene ontology or pathway terms between multiple PDX models, suggesting that these were specific to individual PDX models.

Differentially expressed genes that showed overlap across multiple parental PDX and PDX-derived spheroid pairs were of particular interest as they likely reflect more general changes that were associated with ex vivo culture. We found a small number of genes that were consistently down-regulated in all PDX-derived tumor spheroids when compared to their parental PDXs, including DUSP1 and KLF2. DUSP1 belongs to the dual-specificity phosphatases (DUSPs) family that inactivate various mitogen-activated protein kinase (MAPK) proteins, playing a role as a negative regulator of growth factor signaling.³² On the other hand, KLF2 is a transcription factor that has been extensively studied for its role in cell growth, differentiation, and apoptosis.³³ It also negatively regulates growth factor signaling in tumor cells. Based on previous reports of these genes’ role in suppressing growth factor signaling pathways,³⁴ we speculated that the expression changes of these genes at least partially reflect the adaptive response of spheroids after exposure to the growth factor-enriched ex vivo culture conditions.

Magnetic nanoparticles for spheroid assembly had no impact on the gene expression profile of tumor spheroids

To determine if the magnetic-directed spheroid assembly method impacted the transcriptomic profile of PDX-derived tumor spheroids, we seeded spheroids with or without magnetic nanoparticles and performed RNA-seq analysis. Subsequent differential gene expression analysis identified minimal numbers of differentially expressed genes caused by the presence of magnetic nanoparticles. There were also no overlapping genes that were significantly modulated by magnetic nanoparticles across different PDX models, suggesting that the nanoparticles did not alter or interfere with biological processes in tumor cells (Figure S2C).

Magnetic-directed spheroid assembly enabled consistent T cell incorporation into spheroids

Upon validating that tumor spheroids maintain key biological features of their parental tumors, we next incorporated T cells into the spheroids (Schematic diagram in Figure 1A). T cells are one of the major immune components in the NSCLC tumor microenvironment (Figure S3A). T cell-targeting immunotherapies have also demonstrated robust therapeutic responses in the clinic, with many new drug moieties under pre-clinical investigation. Thus, we reasoned that building a spheroid model with T cells will have strong potential applications in drug discovery while enabling model validation using well-studied therapeutic agents.

To obtain T cells for incorporation into spheroids, we isolated T cells from unmatched healthy donor peripheral blood mononuclear cells (PBMCs), followed by 7 days of in vitro activation and expansion. We first compared the magnetic spheroid assembly method (referred to as “Method 1” in Figure 3A) with a commonly used method for generating 3D tumor-T cell co-culture systems, which involved the formation of tumor spheroids first and then adding T cells in suspension (referred to as “Method 2” in Figure 3A). We hypothesized that magnetic spheroid assembly enables higher levels of T cell incorporation. To test this hypothesis, we set up T cell-PDX co-culture systems using both methods side-by-side, cultured them for three days, and then dissociated spheroids to quantify T cell infiltration utilizing flow cytometry (Figure 3A). Interestingly, spheroids formed by magnetic assembly with simultaneous T cell addition showed higher levels of T cell infiltration across multiple donors as compared to the commonly used method (Figure 3B). Consistent with this observation, the tumor viability % is lower in magnetically assembled spheroids, suggesting higher T cell-mediated killing (Figure S3B). Finally, we compared the expression of T cell activation markers (CD69 and CD25) and % viability in T cells in spheroids generated by different methods. These were largely unchanged by the choice of spheroid formation method, suggesting that the presence of magnetic nanoparticles did not significantly impact T cell phenotypes (Figure S3C).

Figure 3.

T cells can be consistently incorporated into ex vivo tumor spheroids

(A) Schematic of experiment to compare T cell infiltration when spheroids were formed via Method 1 (the magnetic spheroid assembly method), Method 2 (by forming tumor spheroids first then adding T cells in suspension as described in other publications) or Method 3 (by aggregating tumor cells and T cells without magnetic nanoparticles).

(B) Flow cytometry quantification of T cell infiltration when spheroids were formed via Method 1 versus Method 2. Error bars indicate SEM, n = 4–5 spheroids.

(C) Representative immunofluorescence images of tumor-T cell spheroids labeled with pan-cytokeratin, CD3 and DAPI. The panels on the left represents entire tumor spheroid at lower magnification (scale bar, 250um) and the panels on the right show greater detail at higher magnification (Scale bar: 50um). Tumor spheroids were seeded at a T cell: tumor ratio of 1:3 and stained on Day 5 post-seeding. Quantification of %CD3+/total nuclei from histological sections is also provided. Data are presented as mean ± SD. LU450, n = 7 spheroids. LU481, n = 3 spheroids.

∗p < 0.05; N.S, not significant. Statistical testing was performed using a two-tailed, unpaired t-test. See also Figure S3.

Additionally, we tested the effectiveness of forming spheroids by mixing tumor and T cells directly without magnetic nanoparticles and compared that against magnetically assembled spheroids (referred to as Method 3 in Figure 3A). We observed significantly higher T cell infiltration across multiple donors in spheroids formed by magnetic assembly (Figure S3C). T cell activation levels remained unchanged between the two methods (Figure S3E). These results suggested that the use of magnetic-directed spheroid assembly promoted the incorporation of T cells into 3D tumor spheroids compared to alternative methods.

We further characterized the spatial distribution of T cells in magnetically assembled spheroids. Multiplex immunofluorescence staining for tumor marker, pan-cytokeratin, and T cell marker, CD3, confirmed that T cells integrated with the tumor cells within tumor spheroids (Figure 3C). LU450 tumor-T cell spheroids (5 days post-seeding) showed a %CD3+/total nuclei of 5.1% and LU481 tumor-T cell spheroids showed a %CD3/total nuclei of 20.9%. Moreover, the T cells also consistently infiltrated into the inner regions of the spheroids. This contrasted with previous studies where the addition of T cells to pre-formed spheroids resulted in most T cells localizing to the outer area of spheroids, which also restricted immune cell-mediated killing to the periphery of the spheroid.^24,30,35 The more consistent infiltration of T cells into the tumor spheroid space may contribute to the consistent and robust T cell infiltration seen in magnetically assembled spheroids (Figures 3B and S3D).

Additionally, flow cytometry analysis on dissociated magnetically assembled spheroids indicated both CD4⁺ and CD8⁺ T cells were incorporated into spheroids and that their respective proportions were similar to the donor T cells prior to co-culture (Figure S3F). Taken together, these results indicate that magnetic-directed spheroid assembly enabled consistent incorporation of T cells into tumor spheroids.

Infiltrating T cells exhibit tumor cell killing properties

We investigated whether tumor spheroid-infiltrating T cells exhibited functional killing activity, with the consideration that T cells and tumor cells are not from the same donors. At a T cell: tumor cell ratio of 1:3, we observed strong T cell mediated cytotoxicity. This cytotoxicity was specifically seen in the presence of T cells (Figure 4A). This effect was seen across both LU450 and LU481 PDX models and multiple T cell donors. Additionally, we asked whether interactions with tumor cells in our ex vivo culture conditions induced T cell activation (Figure 4B). Compared to T cells in monoculture in vitro, T cells that were incorporated into tumor spheroids generally displayed an increase in expression levels of various cell surface activation markers, including CD69, CD25, PD-1, Lag3, and Tim3. This was seen in both CD8⁺ and CD4⁺ T cells (Figures 4C and S4) and was consistent with previous reports that in an allogenic setting, interactions with tumor cells can lead to T cell stimulation⁴ and at least partially mimic the T cell activities that are seen in human cancers.

Figure 4.

Infiltrating T cells show tumor cell killing activity and upregulation of activation markers in tumor spheroid-T cell co-culture

(A) Number of viable tumor cells in LU450 and LU481 spheroids +/− T cells from different T cell donors, quantified by flow cytometry. The number scale has been normalized to spheroids with no T cells. Spheroids were seeded at a T cell: tumor ratio of 1:3 and evaluated on Day 7 post-seeding.

(B and C) Experimental schematic (B) and quantification (C) of CD8⁺ T cell activation marker expression after co-culture within tumor spheroids, compared to T cell-only culture. Samples included LU450 + Donor 1 T cell, LU450 + Donor 2 T cell, and LU481 + Donor 1 T cell. Spheroids were seeded at a T cell: tumor ratio of 1:3 and evaluated on Day 7 post-seeding.

∗, p < 0.05; ∗∗,p < 0.01; and ∗∗∗p < 0.001. Statistical testing was performed using a two-tailed, unpaired t-test. Data are presented as mean ± SEM, n = 3 spheroids. See also Figure S4.

Patient-derived xenograft-derived tumor spheroids as models to assess drug response to T cell-targeting immunotherapies

One of the primary opportunities of 3D tumor models is to model tumor response to drug treatments in a robust and high throughput manner. Having successfully validated the consistent incorporation of T cells into tumor spheroids, we investigated whether this system could be used to evaluate drug response to T cell immunomodulatory agents.

CD3 bispecific antibodies are a tumor immunotherapy modality that has been clinically approved for the treatment of hematological and solid tumors.^36,37 This class of molecules work by binding to an antigen expressed on tumor cells while simultaneously binding to CD3 receptors on T cells, thereby enabling the T cell recognition of cancer cells and the formation of an immunological synapse that triggers tumor cell killing.³⁸ An EpCAM-CD3 bispecific antibody is currently under clinical investigation for the treatment of solid tumors, and has demonstrated activity in enhancing T cell killing and T cell activation in in vitro experiments.^39,40

As a proof-of-concept study, we treated LU450 and LU481 spheroids with a commercially available EpCAM-CD3 bispecific antibody as both tumor models express high levels of EpCAM on the cell surface (Figure 5A). To evaluate drug response, we devised a multi-color flow cytometry panel, which included tumor and T cell lineage markers, a cell viability dye to evaluate tumor cell killing, and various T cell activation markers to evaluate T cell activation phenotype (a schematic of the method is shown in Figure 5B). In this experiment, the spheroid T cell: tumor ratio was adjusted to 1:12 (decrease in T cells) after optimizing to provide better dynamic range for evaluating activity of the therapeutic antibody, with baseline tumor killing by T cells at under 10%.

Figure 5.

EpCAM-CD3 bispecific antibody enhances T cell cytotoxicity and activation in tumor spheroids

(A) Quantification of EpCAM expression on the cell surface of LU450 and LU481 tumor cells by flow cytometry.

(B) Schematic diagram of the EpCAM-CD3 drug treatment experiment.

(C) Dose-response curve measuring the number of live tumor cells in LU450 + Donor 1 T cell and LU450 + Donor 2 T cell spheroids at increasing concentrations of EpCAM-CD3. Tumor cell number was normalized to the Ctrl-CD3 treated group. Spheroids were seeded at a T cell: tumor ratio of 1:12.

(D) Mean fluorescence intensity of T cell activation markers expressed on CD8⁺ T cells from LU450 + Donor 1 T cell spheroids treated with increasing concentrations EpCAM-CD3.

(E and F) same as (C and D) but for LU481 + Donor 1 tumor-T cell spheroids. Data are presented as mean ± SEM, n = 3–4 spheroids.

See also Figure S5.

Following treatment with EpCAM-CD3, T cell-infiltrating LU450 and LU481 spheroids displayed a dose-dependent reduction in viable tumor cells (Figures 5C and 5E). This was accompanied by a dose-dependent increase in several T cell activation markers in both CD8⁺ and CD4⁺ T cells, including CD25, 41BB, and ICOS, PD1, Tim3, and Lag3 (Figures 5D, 5F, and S5A). These trends were consistently seen in LU450 tumor spheroids infiltrated with T cells from a second donor (Figures 5C and 5D). We did not observe any statistically significant expansion of T cells or changes in the CD8+/CD4+ T cell ratio (Figures S5B and S5C). As a control experiment, the treatment of spheroids with a non-tumor targeting bispecific CD3 antibody did not lead to any increase in T cell cytotoxic activity from baseline (Figure S5D).

Apart from the evaluation of CD3 bispecific antibodies as single agents, there is ongoing interest in testing combinatorial treatments with other therapeutics. Based on previous reports that anti-PDL1 checkpoint inhibitor enhances the treatment efficacy of CD3 bispecific antibodies,⁴¹ we selected a dose of EpCAM-CD3 close to IC50, then concurrently added the anti-PD-L1 antibody, Atezolizumab, to assess if the combination treatment has stronger activity than each drug as a single agent. Notably, LU450 expressed much higher PD-L1 levels than LU481 (Figures 6A and 6B). Treatment with anti-PD-L1 as a single agent did not lead to a significant response in LU450 or LU481 tumor-T cell spheroids when compared to control treatment (Figures 6C and 6D). However, in LU450 tumor-T cell spheroids, dual treatment with EpCAM-CD3 and anti-PD-L1 led to additional tumor lysis compared to EpCAM-CD3 alone (Figure 6C). On the other hand, LU481 tumor-T cell spheroids did not exhibit any additional tumor killing when treated with the combination as compared to EpCAM-CD3 alone (Figure 6D). Based on the known mechanism and clinical biomarkers of response to Atezolizumab, we speculate that this may at least in part be due to the difference in PD-L1 expression levels by the two PDX models.

Figure 6.

Tumor spheroids as ex vivo platform to evaluate phenotype and cytokine profiles associated with T cell-mediated killing in response to immunotherapy combinations

(A and B) Quantification of PD-L1 expression on LU450 and LU481 tumor cells by flow cytometry (A) or immunohistochemistry (B).

(C) Normalized live tumor cell number (left) and mean fluorescence intensity of activation markers on CD8⁺ T cells (right) from LU450 + Donor 1 T cell spheroids treated with EpCAM-CD3 (0.16 ng/mL) and/or anti-PD-L1 (10 μg/mL) as single agents or in combination. Spheroids were seeded at a T cell: tumor ratio of 1:12 and collected at 48h post-treatment.

(D) Same as (C) but for LU481 + Donor 1 T cells.

(E and F) Cytokine heatmaps for LU450 (E) and LU481 spheroids (F) treated with EpCAM-CD3 and/or anti-PD-L1. Red color indicates an increase in cytokine levels in the treatment group compared to control (∗, p < 0.05).

(G) Measured levels of IL-4 in conditioned medium after treatment with EpCAM-CD3 and/or anti-PD-L1.

∗, p < 0.05. Statistical testing was performed using a two-tailed, unpaired t-test. Data are presented as mean ± SEM, n = 3–4 spheroids. See also Figure S6.

To characterize molecular mechanisms associated with drug treatment, we next asked if other biological markers were altered following treatment. T cell activation cell surface markers were strongly up-regulated by EpCAM-CD3 treatment alone, and the addition of anti-PD-L1 did not further induce the expression of activation markers (Figure 6C right panel; Figure S6A). Combination treatment also did not further impact T cell expansion (Figure S6B). This could be partly due to strong T cell activation with CD3-bispecifics and the response from PD-L1 blockade being shielded in a concurrent treatment setting.

Since bispecific antibodies and checkpoint inhibitors could also alter cytokine production, we measured secreted cytokine levels in spheroid conditioned media following treatment. In both LU450 and LU481 spheroids, EpCAM-CD3 treatment induced a strong increase of multiple inflammatory cytokines including Granzyme B and interferon gamma, as compared to Ctrl-CD3 treatment (Figures 6E, 6F, and S6C). Interestingly, comparison between the EpCAM-CD3 only group and the dual treatment group revealed that IL-4 was significantly up-regulated in LU450, which showed an enhanced response to the combination treatment (Figures 6E and 6G). This change in IL-4 was not seen in LU481 spheroids, where anti-PD-L1 did not enhance response to EpCAM-CD3 (Figures 6F and 6G). IL-4 is generally associated with tumor-promoting functions but has been reported to be positively correlated with response to immunotherapy in patients with NSCLC.⁴² In our experiment, the increase in secreted IL-4 levels in the LU450 model may be associated with an observed response to the EpCAM-CD3 and anti-PD-L1 combination treatment.

The above results suggested that our tumor-T cell spheroid model could be used to monitor drug-induced changes to cytokine levels, with potential applications in studying cell-type specific molecular mechanisms associated with drug response.

Altogether, our study demonstrated that we have been successful in establishing a tumor spheroid model with robust T cell infiltration, which present potential applications for evaluating tumor cytotoxicity, T cell phenotypes and cytokine profiles for diverse therapeutic modalities in a physiologically relevant setting.

Discussion

Compelling evidence suggests that traditional 2D tumor cell cultures do not accurately recapitulate the complexity of patient tumors. Emerging 3D models provide a more accurate environment to recapitulate interactions between the tumor and other cell types. In this work, we described a robust and efficient approach to generate PDX-derived 3D tumor spheroids that recapitulate original tumor characteristics and can be reconstituted with cells from the tumor microenvironment such as T cells for tumor immunology studies.

PDXs are well-established as an effective source of human tumor material that is clinically relevant by preserving inter- and intra-tumor heterogeneity. We developed a workflow for isolating human tumor cells from NSCLC PDXs for ex vivo tumor spheroid formation and validated that tumor spheroids maintain original tumor characteristics by utilizing histological staining and transcriptomic analysis. Notably, the expression levels of the majority of genes are unchanged in tumor spheroids. Tumor spheroids also strongly maintained gene signatures characteristic of their originating lung tumor subtype, suggesting that our ex vivo culture conditions preserved the overall gene expression profile of the original tissues. When examining differentially expressed genes that were shared between multiple PDX models, we found a small number of genes that were altered in all tumor spheroid models independent of their tumor subtype, including known negative regulators of growth factor signaling, KLF2 and DUSP1. We speculated that this reflects a feedback response toward the growth factor-enriched environment needed to maintain tumor cell viability ex vivo, as it is consistent with previous studies showing the transcriptomic down-regulation of KLF2 and DUSP1 in cancer cells following 2–24 h of growth factor stimulation.^34,43 This observation also suggests that in cases where ex vivo systems are used to study growth factor-related signaling pathways, careful considerations on the growth factor composition of the culture media are warranted.

To adapt the tumor spheroid system for co-culture with T cells, we used a previously published magnetic-based spheroid formation method. Compared to more conventional 3D co-culture approaches where tumor spheroids are first formed and then T cells are added in suspension in a separate step, our magnetic nanoparticle-based approach enables T cells to be incorporated from the beginning of spheroid formation. The main benefits offered by this system are 1) generally higher and more consistent level of T cell incorporation and 2) the enhanced ability for T cells to infiltrate more deeply inside the spheroid rather than remain on the edge of the spheroid, thereby better recapitulating the interactions between tumor and T cells within a 3D context. We further demonstrated that the T cells incorporated into tumor spheroids are functional in their cytotoxic and cytokine production activities and that the infiltrating T cells show an up-regulation of activation markers in response to interactions with tumor cells. Notably, the magnetic nanoparticles are unlikely to limit T cell movement as they are of much smaller size (50nm) than human T cells. We also observed that the nanoparticles naturally detach from cells 24–48 h post-spheroid formation, after their initial effect to promote consistent T cell infiltration. Although our study focused primarily on T cells, we expect the magnetic spheroid assembly methodology to be applicable to incorporating other tumor microenvironment cell types, such as NK or dendritic cells.

On the other hand, 3D co-culture tumor spheroids also have advantages and limitations compared to tumor organoid or explant models, which are typically generated by mincing the tumor material into small fragments for culture. Notably, tumor organoids have been shown to preserve tumor-infiltrating immune cells for a short term,²¹ but the immune cell population decreases significantly after 2–3 days in culture. Compared to these systems, our spheroid co-culture model lacks tumor-native immune cells but allows for 7 or more days of tumor-immune cell co-culture, as well as flexibility in cell types and numbers to be included in the spheroids. Additionally, spheroid cultures have the benefits of being simple, matrix-free, and generally more automatable in their set-up. Therefore, we expect that tumor-T cell co-culture spheroids may be particularly well-suited to applications that require greater throughput or more precise control over cellular composition.

There is particular interest in the use of patient or PDX-derived models for pre-clinical drug testing and biomarker discovery due to their strong clinical relevance. In this study, we demonstrated the application of PDX-T cell spheroids for the pre-clinical evaluation of a commercially available EpCAM-CD3 bispecific antibody as a single agent or in combination with the anti-PD-L1 antibody, Atezolizumab. In these experiments, both PDX models (LU450 and LU481) evaluated had high expression of EpCAM and responded to EpCAM-CD3 treatment. However, only LU450-T cell spheroids (higher %PD-L1+) exhibited additional tumor lysis when treated with the EpCAM-CD3/Atezolizumab combination as compared to EpCAM-CD3 alone, whereas LU481-T cell spheroids (lower %PD-L1+) did not show any additional efficacy. Cytokine analysis of conditioned media suggested that LU450-T cell spheroids increased secretion of IL-4 following combination treatment, which was not seen in LU481-T cell spheroids. IL-4 is largely considered to be an immune-suppressive cytokine in the tumor microenvironment by favoring the generation of immunosuppressive M2 macrophages and myeloid-derived suppressor cells, as well as suppressing T cell cytotoxic activity.⁴⁴ Intriguingly, it has been previously reported that IL-4 levels in peripheral blood is associated with increased response to immune checkpoint inhibitors in patients with NSCLC.⁴² More extensive clinical data will be needed to further understand the complex relationship between the different cytokines and response to immunotherapies. Even though our finding is preliminary, the experiment provides a proof-of-concept demonstration that ex vivo tumor cultures may provide molecular insights or generate new hypotheses on how tumor and T cells respond to different immune modulating therapies on an individual patient level. Overall, we established a flow cytometry-based approach to interrogate 3D co-culture spheroids for multiple experimental read-outs simultaneously (tumor cytotoxicity, T cell phenotypes and cytokine profiles) and highlighted their versatile applications for assessing drug activity or potential biomarkers in different cell types.

Nonetheless, there are several challenges in our current approach. First, it requires the successful organization and survival of PDX cells as ex vivo spheroids. In our experience, tumor cells isolated from approximately 50% of NSCLC PDX models are capable of forming and maintaining viability in 3D culture. While the magnetic spheroid assembly system enhanced the quality of ex vivo PDX spheroids (e.g., more consistent sizes and growth rates), it did not increase the overall success rate in terms of numbers of PDX models that can survive as ex vivo spheroids. Other advances in ex vivo culture techniques such as microfluidic systems, extracellular matrix scaffolds and more sophisticated culture media may be key for improving the success rate of tumor spheroid cultures. An additional caveat of the study is our use of allogeneic donor T cells for generating tumor-T cell spheroids. Although the testing of CD3 bispecific antibodies is not limited by this set-up (their mechanism of action involves the redirection of T cell recognition of tumor cells and is not limited by mismatched T cell donors), the lack of tumor reactive T cells in the allogeneic setting may hamper the testing of other T cell modulating therapies. Interestingly, a recent publication described a technique for the ex vivo expansion of tumor-reactive T cells from PBMCs of patients with NSCLC and demonstrated that the expanded T cells showed reactivity and successful killing of patient-matched tumor organoids.¹⁸ This finding highlights matched patient blood as an accessible source for tumor reactive immune cells. Additional studies are needed to understand the predictive value of PBMC-derived immune cells for tumor response to immunotherapies, but we anticipate that in future experiments, the incorporation of patient-matched T cells and other immune cells into patient or PDX-derived spheroids will provide unique opportunities to model and predict responsiveness to drug treatments in a patient-specific manner.

In conclusion, we established a PDX-derived 3D tumor spheroid system that recapitulates tumor tissue characteristics and enables the consistent incorporation of T cells. Our functional studies demonstrated the utility of spheroids for preclinical drug testing in a flexible and scalable format, with relevant translational applications in evaluating drug activity, identifying novel drug combinations, and assessing associated molecular mechanisms.

Limitations of the study

Although we have established a 3D tumor spheroid model with robust T cell infiltration, some of the limitations of the current approach are as follows: First, our study was not focused on an in-depth characterization of ex vivo conditions that would enable optimal survival and growth of all NSCLC PDX models. For example, we reported that lung adenocarcinoma PDXs showed a preferential growth advantage over lung squamous cell carcinoma PDXs under the current ex vivo growth conditions. Enhancements in culture media or the use of techniques, such as extracellular matrix scaffolds or microfluidics, may enable more effective ex vivo culture of a broader range of PDX models. Another caveat is the use of allogeneic donor T cells for generating tumor-T cell spheroids, due to the scarcity of donor-matched tumor and T cell/PBMC materials. This is a common limitation of similar tumor-T cell co-culture studies and highlights a need for more resources for patient-matched tumor and immune cells. Finally, while our study has demonstrated some proof-of-concept use cases for this 3D tumor spheroid model, there is further potential to scale up experimental data generation by integrating the model into high-throughput or high-content screening systems.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Hin Ching Lo (hinching.lo@abbvie.com).

Material availability

This study did not generate new unique reagents.

Data and code availability

• •All data reported in this article will be shared by the lead contact upon request.
• •This article does not report original code.
• •The PDX and spheroid RNAseq data generated in this article are deposited at GEO: GSE295993 and are publicly available as of the date of publication.
• •Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.

Acknowledgments

The authors would like to thank the following AbbVie employees: Trusha Sondkar, Juan Blanco, and Elmer Payson for their help with tumor collection and animal experiments. This work is supported by AbbVie.

Author contributions

HCL and SR conceptualized the study. HCL and HC performed the experiments and analyzed the data. KK, SR, LG, FC, and AS performed histological staining and analysis. All authors helped with the discussion and interpretation of results and the editing of the manuscript.

Declaration of interests

HCL, HC, KK, SR, LG, FC, AS, XZ, AD, RF, and SR are employees of AbbVie. The design, study conduct, and financial support for this research were provided by AbbVie. AbbVie participated in the interpretation of data, review, and approval of the publication.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
PanCK	Agilent	Cat# GA05361-2
PD-L1	Agilent	Cat# SK0006
PanCK	Leica	Cat# NCL-L-AE1/AE3-601
CD3	Thermo	Cat# RM-9017
Anti-Ki-67	Abcam	Ab16667
CD3-BUV737	BD Biosciences	612750
EPCAM-PE-Cy7	Biolegend	324222
CD4-AF488	Biolegend	317414
CD8-BUV805	BD Biosciences	612890
PD1-APC	Biolegend	329908
LAG3-PE-CF594	BD Biosciences	565718
TIM3-BV711	BD Biosciences	565566
CD69-PE	Biolegend	310906
CD25-BV605	Biolegend	302632
ICOS-APC-CY7	Biolegend	313530
41BB-PERCP-CY5.5	Biolegend	309814
Biological samples
Healthy donor PBMC	Discovery Life Sciences	N/A
Chemicals, peptides, and recombinant proteins
DMEM/F12 media	Gibco	# 11320033
bFGF	ThermoFisher	Cat# 13256-029
EGF	ThermoFisher	Cat# PHG0315
N2	ThermoFisher	Cat# 17502048
B27	ThermoFisher	Cat# 17504044
ROCK inhibitor	Sigma	Cat# Y0503
100X penicillin/streptomycin	Gibco	Cat# 15240062
M199 media	Gibco	Cat# 11150067
Liberase DL	Roche	Cat# 12352200
DNase I	Invitrogen	Cat# 18068015
RPMI-1640 media	Gibco	Cat# 11875093
IL2	ThermoFisher	Cat# PHC0021
Pembrolizumab	Selleck Chem	Cat# A2005
EpCAM-CD3 bispecific antibody	Creative Biolabs	Cat# BSSG-079
DAPI	Akoya	Cat# NEL801001KT
Ficoll-Paque Plus solution	Cytiva	17144003
Critical commercial assays
NanoShuttle kit	Greiner Bio-One	Cat# 657841
96-well ULA plates	Greiner Bio-One	Cat# 655970
EasySep Human T Cell Isolation Kit	Stemcell Technologies	Cat# 100-0695
Human CD3/CD28 T Cell Activator	Stemcell Technologies	Cat# 10971
Opal fluorescent secondary detection reagents	Akoya	Cat# NEL801001KT
EnVision FLEX Detection Kit	Agilent	Cat# K8000
CellTiter-Glo-3D	Promega	Cat# G9681
LEGENDplex Human CD8/NK Panel	Biolegend	Cat# 741186
LIVE/DEAD Fixable Violet Dead Cell Stain Kit	ThermoFisher	Cat# L34955
Mouse cell depletion kit	Miltenyi Biotec	Cat# 130-104-694
Deposited data
PDX/spheroid bulk RNAseq	This paper	GEO: GSE295993
Experimental models: Organisms/strains
Mouse: NOD.CB17-Prkdcscid/NCrCrl	Charles River Laboratories	394
Software and algorithms
GraphPad Prism version 9.0	GraphPad	N/A
FlowJo	FlowJo LLC	N/A
R software (v4.2.1)	Open source software	N/A
Incucyte S3 Live-Cell Analysis System	Essen Bioscience	N/A

Experimental model and study participant details

Mouse models for propagation of patient-derived xenografts

Patient-derived xenograft (PDX) tumor cells were implanted and propagated in female NOD/SCID mice (5 to 10 weeks of age; Charles River Laboratories). Effects of sex on PDX growth were not assessed. No randomization was needed as all animals in each experiment were injected with the same PDX cell suspension. All in vivo studies were conducted in accordance with protocols approved by the AbbVie’s Institutional Animal Care and Use Committee. Animal health was monitored daily, and humane euthanasia was performed at protocol clinical endpoints per the American Association for Laboratory Animal Science guidelines.

Ex vivo culture of patient-derived xenograft spheroids

Following PDX harvest, cell preparation and spheroid formation as described below in “method details”, PDX-derived tumor spheroids were cultured at a 37°C humidified incubator containing 5% CO2. Sex of PDX models was provided in Figure S1.

Method details

Propagation of patient-derived xenografts in mice

Patient-derived xenograft (PDX) tumor cells were implanted and propagated in female NOD/SCID mice (5 to 10 weeks of age; Charles River Laboratories). Dissociated single cells from PDXs were suspended in 1:1 mixture of medium (Hank’s balanced salt solution with 2% FBS and 2.5% HEPES) and Matrigel, and 50,000 live human cells in 100μL volume was implanted subcutaneously on the flank of each mouse. Cells were counted using flow cytometry. Measurements of the length (L) and width (W) of the tumor were taken weekly, and the volume was calculated according to the following equation: V = (L × W²)/2. Once tumors reached approximately 1000 mm³, mice were humanely euthanized, and tumors were harvested for spheroid culture and/or further propagation in mice. PDX cells were routinely tested to ensure there was no mycoplasma contamination. At every passage, SNP fingerprinting was used to confirm PDX lineage identity and flow cytometry staining was performed on dissociated cells to screen for mouse or human lymphoma outgrowth. Antibodies used included: human CD45-PE (clone 2D1, Biolegend), mouse CD45-FITC (clone 30-F11, Biolegend), mouse H2Kd-APC (clone SF1-1.1, Biolegend), human CD326-PerCP (clone 9C4, Biolegend), mouse Ly-6G/Ly-6C-APC-Cy7 (clone RB6-8C5, Biolegend), human CD46-PECY7 (clone TRA-2-10, Biolegend).

PDX-derived tumor cell preparation

PDX tumors were minced into small pieces with sterile blades and placed in digestion media (M199 media (Gibco) with 0.5mg/mL Liberase DL (Roche) and 1mg/mL DNase). The mixture was incubated on a shaker at 37°C for 30 min. After incubation, the mixture was sequentially passed through 100- and 40-μm cell strainers to generate a single cell suspension. To remove dead cells and debris, the strained cells were gently layered on top of Ficoll-Paque Plus solution (Cytiva) and centrifuged at 112 × g for 15 min without brakes. Next, the mouse cell depletion kit (Miltenyi Biotec) was used to remove mouse cells from the cell suspension according to the manufacturer’s protocol. The purified human cells were centrifuged at 112 × g for 5 min and resuspended in Minimum Basal Media¹⁵ (MBM; serum-free medium DMEM/F12 (Gibco) supplemented with 20 ng/mL of bFGF (Invitrogen), 50 ng/mL human EGF (Invitrogen), N2 (Invitrogen), B27 (Invitrogen), 10 μM ROCK inhibitor (Selleck Chem), and 1% penicillin/streptomycin (Gibco)). Cells were subsequently used for spheroid formation.

Tumor spheroid formation

The NanoShuttle kit (Greiner Bio-One, Catalog # 657841) was used to form PDX tumor cell spheroids according to the manufacturer’s instructions. Briefly, the dissociated PDX-derived tumor cells were resuspended in MBM culture medium. Nanoshuttle magnetic nanoparticles was added to achieve the desired concentration of 1uL/1 × 10⁵ tumor cells. The cell suspension with Nanoshuttle particles was then centrifuged at 112 × g for 5 min. The pellet was resuspended by gentle pipetting, and centrifuged twice more at 112 × g for 5 min each to facilitate magnetic nanoparticle binding to tumor cell surface. The cell suspension was seeded into 96-well ULA plates (Greiner Bio-One, Catalog # 655970) at a density of 1 × 10⁴ tumor cells per well.

Next, the ULA plates were placed in the supplied magnetic drive and incubated at a 37°C humidified incubator containing 5% CO2 for 24 h to enable spheroid formation. After 24 h, the magnetic drive was removed from ULA plates, and the ULA plates containing formed spheroids were maintained in culture in 37°C for up to 14 days.

T cell isolation, activation, and spheroid incorporation

Healthy donor peripheral blood mononuclear cells (PBMC) were purchased from Discovery Life Sciences. PBMC vials were rapidly thawed in a 37 C water bath and washed twice with pre-warmed RPMI-1640 medium supplemented with 10% FBS. T cells were then isolated from the thawed PBMCs using the EasySep Human T Cell Isolation Kit (Stemcell Technologies, Catalog #100-0695) according to the manufacturer’s instructions.

The isolated T cells were resuspended in complete RPMI-1640 medium supplemented with 10% FBS, 1% penicillin-streptomycin, and 50 U/mL IL-2. T cells were seeded at a density of 1 × 10⁶ cells/mL in a sterile tissue culture plate. To activate the T cells, ImmunoCult™ Human CD3/CD28 T Cell Activator (Stemcell Technologies, Catalog #10971) was added to the culture at a final concentration of 25 μL/mL. The T cells were incubated at 37°C in a humidified incubator containing 5% CO2 for 48 h, then resuspended with fresh complete media to a density of 1–2 x10⁶ cells/mL. T cells were maintained in culture and expanded for another 5 days as per the manufacturer’s instructions. On Day 7 after initial activation, T cells were washed with fresh media and mixed with PDX-derived tumor cells for 3D spheroid formation as described above. A cellular ratio of 1 T cell: 3 tumor cells was used unless otherwise specified.

Live imaging of tumor spheroids

To track spheroid growth over time, live time-lapse imaging of 3D cell spheroids was performed using the Incucyte S3 Live-Cell Analysis System (Essen Bioscience). Briefly, after cell seeding on ultra-low attachment plates and incubation on magnetic drive for 15 min, the plates were then transferred to the Incucyte system and placed inside the incubation chamber. For incubation, the system was set to maintain a temperature of 37°C, humidity of 95%, and 5% CO2 concentration. Images were acquired using a 10× objective lens at 24-h intervals over a total duration of 5 days.

Histology

PDX tissues and spheroids were fixed with 4% paraformaldehyde (PFA), paraffin-embedded (FFPE), and sectioned at 4-micron thickness. H&E staining was performed using a Tissue-Tek Prisma Autostainer (Sakura). Immunohistochemistry for PanCK (AE1/AE3; Agilent cat. GA05361-2) was performed using a DAKO Link 48 Autostainer (Agilent) and the EnVision FLEX Detection Kit (Agilent cat. K8000), and PD-L1 staining using the PharmDx Kit (Clone 22C3, Agilent, Cat. SK0006). All slides were hematoxylin counterstained before being digitally scanned using a Leica AT2 scanner (Leica) and viewed using the Concentriq platform (Proscia). Histological slides of spheroids and corresponding parental PDXs were evaluated by a board-certified pathologist to compare tumor histological and morphological features such as differentiation status, nucleus structure and mitotic activity. Specific descriptions for each PDX model were provided in Figure S1.

Immunofluorescence

Immunofluorescent staining of FFPE sections was performed using PanCK (AE1/AE3 1:50; Leica cat. NCL-L-AE1/AE3-601), CD3 (SP7 1:125; Thermo cat. RM-9017), Ki-67 (SP6 1:100; Abcam cat. Ab16667) and Opal fluorescent secondary detection reagents (Akoya, Cat. NEL801001KT) on the Leica Bond autostainer (Leica). Sections were counterstained using DAPI (Akoya, cat. NEL801001KT) and coverslips mounted with Prolong Diamond (ThermoFisher), before being digitally scanned using a Vectra Polaris (Akoya) with a 20× Objective. Whole slide fluorescent images were analyzed and quantified using HALO (Indica Labs).

Transcriptomic analysis

RNA was extracted from tumor spheroids and matched PDX tumors (purified for human cells as described above) using the RNeasy mini Kit (Qiagen). RNA library preparation (NEBNext Ultra RNA Library Prep Kit for Illumina, NEBNext Poly(A) mRNA Magnetic Isolation Module) and sequencing (Illumina NovaSeq S4, 150 nucleotide paired-end reads) were performed by Admera Health (South Plainfield, NJ). FastQC (version v0.11.8) was applied to check the quality of raw reads. Trimmomatic (version v0.38) was applied to cut adaptors and trim low-quality bases with default setting. STAR Aligner version 2.7.1a was used to align the reads. Picard tools (version 2.20.4) was applied to mark duplicates of mapping. The StringTie version 2.0.4 was used to assemble the RNA-Seq alignments into potential transcripts. The featureCounts (version 1.6.0)/HTSeq was used to count mapped reads for genomic features such as genes, exons, promoter, gene bodies, genomic bins and chromosomal locations. Sequenced reads were aligned to the human reference genome (GRCh38). The DESeq2 R package was used to normalize the gene/isoform expression matrix.

R software (v4.2.1) with limma (v3.52.3) and ggplot2 packages was used for RNAseq differential gene expression analysis and data visualization. Differentially expressed genes were considered significant if the FDR-adjusted p-value was smaller than 0.01 and log2 fold change was greater than 2 (Table S1). ClusterProfiler package (v4.4.4) was used to perform Gene Ontology Enrichment Analysis on the “Biological Processes” ontology. For comparison of lung adenocarcinoma and lung squamous cell carcinoma gene expression, log2 TPM values were plotted in a heatmap using pheatmap package (v1.0.12) and the gene lists were based on a previous publication.³¹ Gene names are also provided in Figure 2C.

Flow cytometry

Tumor spheroids were washed once with PBS and dissociated using TrypLE Express. The cell suspension was passed through 40 μm cell strainers and stained with the desired antibodies for 30 min. The antibodies used included: CD3-BUV737 (clone UCHT1, BD Biosciences), EPCAM-PE-Cy7 (clone 9C4, Biolegend), CD4-AF488 (clone 317414, Biolegend), CD8-BUV805 (SK1, BD Biosciences), PD1-APC (clone EH12.2H7, Biolegend), LAG3-PE-CF594 (clone T47-530, BD), TIM3-BV711 (clone 7D3, BD Biosciences), CD69-PE (clone FN50, Biolegend), CD25-BV605 (clone BC96, Biolegend), ICOS-APC-CY7 (clone C398.4A, Biolegend), 41BB-PERCP-CY5.5 (clone 4B4-1, Biolegend). Cells were also stained with LIVE/Dead Fixable Violet cell dye to determine cell viability. Flow cytometry analysis was performed on an LSRFortessa system (BD Biosciencies) equipped with a high throughput sampler and results were analyzed with FlowJo software (BD Biosciences).

Drug treatment and evaluation of drug response

PDX-derived tumor cells and donor T cells were seeded on 96-well ULA plates (Greiner Bio-One) as described above. Once spheroids have formed and stabilized after 3 days, 10 different concentrations of EpCAM-CD3 bispecific antibody (Creative Biolabs, Catalog # BSSG-079) was added with or without the addition of Pembrolizumab (Selleck Chem, Catalog # A2005; 10 μg/mL). The isotype controls for the two drugs are BSA-CD3 bispecific antibody (Catalog# BSFV-c001) and human IgG4 isotype control (Catalog #A2052), respectively. 6 days after initiation of drug treatment, CellTiter-Glo-3D (Promega) was added to each well according to manufacturer’s protocol. The plates were agitated for 30 min at RT prior to luminescence reading. Dose-response curves were generated using Graph Pad Prism. Cell culture supernatant was also collected and LEGENDplex Human CD8/NK Panel (BioLegend) was used to quantify secreted cytokine levels according to manufacturer instructions.

Quantification and statistical analysis

Statistical analysis was performed using GraphPad Prism software (Version 9.0). Data are presented as mean ± standard error of the mean (SEM) and n numbers denote the number of tumor spheroids, unless otherwise specified in figure legends. Further details of statistical tests used, n numbers, and the entities represented by n can be found in figure legends. Statistical significance was determined using the Student’s t-test (two-tailed, unpaired) with p < 0.05 considered statistically significant.

Published: June 25, 2025