A human tumor-immune organoid model of glioblastoma

SUMMARY

A major obstacle to identifying effective therapies for the aggressive brain tumor glioblastoma is the lack of human-specific, immunocompetent models that reflect the human tumor microenvironment. To address this, we developed the immune-human organoid tumor transplantation (iHOTT) model, an autologous co-culture platform that integrates patient-derived tumor cells and matched peripheral blood mononuclear cells within human cortical organoids to enable the study of patient-specific immune responses and tumor-immune interactions. This platform preserves tumor and immune populations, immune signaling, and cell-cell interactions observed in patient tumors. Treatment of iHOTT with pembrolizumab, a checkpoint inhibitor, mirrors cell-type shifts and cell-cell interactions observed in patients. T cell receptor (TCR) sequencing further reveals pembrolizumab-driven expansion of stem-like CD4 T cell clonotypes exhibiting patient-specific repertoires. These findings establish iHOTT as a physiologically relevant platform for exploring autologous tumor immune interactions and underscore the need for antigen-targeted strategies to enhance immunotherapy in glioblastoma.

In brief

Baisiwala et al. introduce iHOTT, an autologous organoid co-culture system that preserves human tumor-immune interactions and T cell clonal responses. iHOTT mirrors clinical responses to PD-1 blockade and provides a platform for evaluating personalized immunotherapy strategies in glioblastoma.

Graphical abstract

INTRODUCTION

Glioblastoma (GBM) is the most common malignant adult brain tumor.^1,2 It carries a dismal prognosis and remains largely refractory to immunotherapy.^3,4 Despite the success of checkpoint inhibitors and CAR-T therapies in other cancers, only a minority of GBM patients respond, reflecting profound immune evasion and limited understanding of relevant tumor-immune interactions.^5,6 There is therefore a critical need for human models that preserve and allow manipulation of tumor and immune compartments to investigate mechanisms of response and resistance.³

Existing glioma models rely on immunodeficient mice or genetically engineered systems that fail to replicate human tumor-immune interactions and human cytokine signaling. As a result, they cannot reliably model responses to immunotherapy.^7–10 Human organoid platforms now enable patient-specific modeling of brain development and cancer, providing an opportunity to establish a fully human co-culture system that preserves tumor and immune compartments and allows mechanistic evaluation of resistance and response.^11,12

Recent work has incorporated CAR-T therapies into glioblastoma organoids, providing a platform for interrogating patient-specific responses to these T-cell-based therapies. While these provide valuable insights, they are limited by the scarcity of endogenous immune cells in this immunologically “cold” tumor.^6,13–16 Importantly, these models do not include circulating immune populations or the cortical microenvironment. Two-dimensional (2D) co-cultures provide even less physiological relevance. A 3D human system incorporating tumor cells and matched peripheral immune cells is therefore needed to study patient-specific tumor-immune interactions.

We previously established a human organoid tumor transplantation (HOTT) platform that preserves patient-derived glioblastoma cell heterogeneity within pre-grown cortical organoids.¹⁷ Here, we extend this system by introducing matched peripheral immune cells to create immune-human organoid tumor transplantation (iHOTT), an autologous tumor-immune co-culture model. iHOTT enables controlled interrogation of tumor-immune interactions and provides a human platform to study mechanisms of immune resistance and therapeutic response.

We use matched circulating peripheral blood mononuclear cells (PBMCs) to model patient-specific immune responses, reflecting the continuous entry of circulating immune cells into glioblastoma.^18,19 Unlike tumor-infiltrating lymphocytes, PBMCs have not undergone chronic exhaustion and provide a responsive immune compartment for studying early tumor recognition and clonal dynamics.^20,21 This design enables analysis of physiologically relevant immune responses to tumor antigens and immunotherapy.

iHOTT supports sustained tumor-immune co-culture, preserves major circulating immune populations, and induces cytokine and transcriptional programs associated with immune activation. Benchmarking against pembrolizumab-treated patient tumors demonstrates that iHOTT recapitulates key treatment-associated changes in cell composition, signaling, and T cell receptor (TCR) clonotypes. Together, these findings establish iHOTT as a human, autologous platform for dissecting tumor-immune interactions and evaluating personalized immunotherapy responses in glioblastoma.

RESULTS

iHOTT enables tumor-immune co-culture

iHOTT enables the co-culture of freshly isolated, patient-matched glioblastoma cells and PBMCs within a human cortical microenvironment. Human cortical organoids are generated over 8–12 weeks as previously described (Figure S1A).¹⁷ These organoids contain major cell types present in the human cortex and thus provide a human cortical microenvironment for the subsequent transplant of patient tumor cells.¹⁷ On the day of surgery, the patient’s tumor is dissociated, transduced with GFP-expressing lentivirus, and transplanted together with matched PBMCs at a 1:1 ratio onto pre-grown cortical organoids (Figure 1A).¹⁷

Figure 1. iHOTT enables tumor and immune co-culture and reflects patient tumor cytokine profiles.

(A) Human cortical organoids were cultured for 8–12 weeks, then transplanted with GFP⁺-patient-derived glioblastoma cells and matched PBMCs. Cultures were maintained for 7 days prior to flow cytometry, immunostaining, cytokine profiling, and scRNA-seq. Inset: GFP⁺ tumor cells and CellTrace⁺ PBMCs within the organoid.

(B) Immunofluorescence of sectioned iHOTT organoids stained for GFP, CD3, CD14, and CD19 showed robust CD3⁺ T cell infiltration with smaller CD14⁺ and CD19⁺ populations and GFP⁺ tumor invasion. Scale bars, 125 μm for all panels. Insets show individual channels.

(C) Flow cytometry of input PBMCs showed a large CD3⁺ population. After 7 days, iHOTT cultures maintained distinct tumor (GFP⁺), immune (CD45⁺), and organoid (GFP⁻ CD45⁻) compartments, with immune composition similar to input PBMCs.

(D) Cytokine profiling at day 7 showed increased secretion of tumor-associated cytokines, including G-CSF, IL-6, and IL-10, in iHOTT relative to tumor-only and PBMC-only controls. PBMC-only cultures did not produce these cytokines, indicating tumor-immune interaction-dependent signaling. The heatmap shows log2 fold change relative to tumor-only controls (mean of all replicates). Boxplots show the median (center line) and interquartile range (box).

Culture media were optimized to preserve immune populations, with a 50:50 mixture providing the best recovery despite modestly reduced total cell yield (Figures S1B and S1C). Lentiviral labeling was efficient, with ~70% GFP⁺ tumor cells by flow cytometry (Figure S1D). Co-cultures were maintained for 7 days before scRNA-seq, flow cytometry, immunostaining, and cytokine profiling, using 3–6 patient tumors per assay, with technical replicates performed across flow cytometry and imaging experiments (Figure 1A).

Immunostaining of iHOTT was performed to evaluate the presence of tumor cells (enhanced GFP [eGFP]), T cells (CD3), myeloid cells (CD14), and B cells (CD19) (Figures 1B and S2A–S2C). To assess whether similar interactions occur in vivo, freshly isolated patient tumors were stained, demonstrating CD3⁺ T cell infiltration into Nestin⁺ tumor regions (Figure S2D).

Flow cytometry quantified immune composition and tumor engraftment (Figure 1C). Input PBMCs consisted of ~65% CD3⁺ T cells, 11% CD14⁺ myeloid cells, and 13% CD19⁺ B cells, consistent with reported human peripheral blood distributions.²² iHOTT cultures showed effective tumor infiltration into organoids and recovery of CD3⁺ T cells and CD19⁺ B cells, with reduced CD14⁺ myeloid cells (Figure 1C). Percentages did not total 100%, likely reflecting additional immune subsets not captured in this panel.

To improve myeloid representation, cultures were supplemented with interleukin (IL)-2, IL-15, and granulocyte-macrophage colony-stimulating factor (GM-CSF), but myeloid recovery remained limited (Figure S1E). This is consistent with broader challenges in preserving myeloid cells ex vivo.²³ As such, we leveraged the iHOTT system to dissect tumor-immune dynamics driven predominantly by T cell interactions.

Cytokines characteristic of the tumor microenvironment are secreted from iHOTT

Media were collected from iHOTT as well as tumor-only and PBMC-only organoid controls. Samples were analyzed for an extensive array of 37 cytokines and chemokines. Compared to tumor-only organoid cultures, iHOTT showed significant upregulation of cytokines previously associated with glioblastoma, including IL-6, IL-8, IL-10, and granulocyte colony-stimulating factor (G-CSF) (Figure 1D). IL-6 and IL-8 promote inflammatory activation, IL-10 is linked to immune modulation, and G-CSF supports myeloid recruitment and survival.^24–27

Most cytokines did not increase in PBMC-only cultures, indicating that tumor-immune interactions drive cytokine production in this system (Figures S3A and S3B). An iHOTT control lacking lentiviral GFP labeling showed no major differences in tumor-associated cytokines, confirming that transduction does not account for these signatures (Figure S3C).

scRNA-seq reflects preservation of tumor and immune cell types

Three freshly isolated patient tumors were co-cultured in iHOTT with matched PBMCs. After 7 days, eGFP⁺ tumor cells and CD45⁺ immune cells were sorted and subjected to single-cell RNA sequencing (scRNA-seq). Cell types were annotated using our previously established glioblastoma meta-atlas and a PBMC reference atlas, simplified as detailed in Table S1.^28,29 All major tumor and immune populations typically observed in patients were preserved^28,29 (Figure 2A). Canonical marker visualization confirmed expected tumor and immune subsets, with immune cells showing low expression of exhaustion markers (HAVCR2 and LAG3) and high expression of activation markers (GZMB and CD69), indicating that iHOTT maintains functionally competent immune populations.

Figure 2. scRNA-seq confirms preservation of tumor and immune compartments and recapitulates cell-cell interactions.

(A) UMAPs of three patient samples showing annotation by patient, tumor vs. PBMC origin, co-culture condition, and reference projection onto a GBM meta-atlas and PBMC atlas. Feature plots validate tumor and immune identities, activation, and exhaustion states.

(B) Comparison of immune cell composition between input PBMCs and iHOTT after 7 days shows preservation of major immune subsets, including CD4⁺ and CD8⁺ T cells, B cells, NK cells, and myeloid populations.

(C) GSEA comparing input PBMCs and iHOTT PBMCs shows enrichment of immune activation pathways in co-culture.

(D) CellChat analysis demonstrates mean tumor-immune interaction strength with CD8⁺ T cells as major signal receivers.

(E) Ligand-receptor analysis shows outgoing signaling from oligodendrocyte-like and astrocyte-like tumor populations and increased MHC class I and MIF signaling in co-culture, indicating active tumor-immune crosstalk. Bar graph shows the signaling strength score for each interaction.

Comparison of input PBMCs to post-co-culture immune fractions showed retention of major immune cell types (Figure 2B). Higher-resolution analysis further identified dendritic cells, myeloid subsets, and natural killer (NK) cell populations within the immune compartment (Figure S4A).

We additionally sequenced four tumors in which PBMCs were transplanted onto organoids without tumor cells and compared them to matched PBMC input and tumor-PBMC co-cultures. PBMC-only cultures showed reduced preservation of NK cells and other T cell subsets, consistent with limited immune stimulation in the absence of tumor antigens (Figure S4B and S4C). Differences between flow cytometry and scRNA-seq immune compositions likely reflect the higher resolution of multi-marker single-cell annotation relative to single-marker flow cytometry.

To assess potential allogeneic activation, we examined immune activation signatures in PBMC-organoid cultures. No upregulation of activation markers was detected in these conditions, and organoids lacked major histocompatibility complex (MHC) class I and MHC class II expression, consistent with their embryonic stem cell origin and known hypoimmunogenicity (Figure S4D).^30,31 TOX was the only marker increased in PBMC-only cultures, indicating T cell exhaustion rather than activation (Figure S4D). Flow cytometry further showed no increase in activation marker CD137 expression in PBMC-organoid cultures compared with PBMCs cultured in a dish (Figure S4E). Together, these findings indicate that the organoid does not induce an allogeneic immune response and that immune activation in iHOTT is primarily tumor driven.

A key feature of iHOTT is its ability to determine how tumor populations change in response to immune exposure. Compared with tumor-only cultures, co-culture with PBMCs led to shifts in tumor cell composition, including expansion of progenitor-like cells and reduction of mesenchymal-like populations (Figure S5A). Tumor cells also showed transcriptional changes, with enrichment of metabolic pathways (Figure S5B). Single-cell resolution further enabled assignment of cytokine sources. In co-cultures, VEGFA was predominantly expressed by tumor cells, whereas chemokines such as CCL3, CCL4, and CXCL10 were primarily produced by myeloid and dendritic cell populations (Figure S5C).

Interactions between tumor and immune cell types are preserved in iHOTT

Within the immune compartment, we compared pathway enrichment in co-cultured immune cells versus input PBMCs from the same patients. Immune response programs, including B-cell-mediated immunity, complement activation, and antigen presentation, were significantly upregulated in iHOTT (Figure 2C). This upregulation reflects an activated immune response that emerges only in the presence of tumor cells. Similar immune activation pathways have been observed in the glioblastoma tumor microenvironment and are associated with both endogenous and therapy-induced immune responses.³²

Cell-cell interaction analysis showed that tumor oligodendrocyte-like cells were major signaling senders, while CD8⁺ T cells were the dominant recipients, followed by NK cells and unconventional T cell subsets (γδ, MAIT, and double-negative T cells) (Figures 2D and 2E). The most upregulated signaling pathways included MHC class I and macrophage migration inhibitory factor (MIF) (Figure 2E). MHC class I enables CD8⁺ T cell antigen recognition, while MIF contributes to immune recruitment and tumor immune evasion.^33–39 The prominence of CD8⁺ T cells as signaling targets aligns with their central role in cytotoxic tumor surveillance.⁴⁰ Together, these results suggest that our model not only preserves the diversity of tumor and immune cell populations but also facilitates dynamic and physiologically relevant interactions that reflect the complex tumor-immune crosstalk occurring in patients.

Pembrolizumab treatment in iHOTT results in expansion of B and T cells

Immune checkpoint inhibitors targeting the programmed death 1 (PD-1)/programmed death ligand 1 (PD-L1) axis have transformed cancer therapy by restoring T cell function.^41,42 Pembrolizumab, a PD-1 blocking antibody, has been trialed in glioblastoma but has shown limited efficacy, with modest response rates in recurrent disease.^5,43–46 We therefore evaluated pembrolizumab in iHOTT to benchmark against patient data and investigate mechanisms underlying limited response.

Three patient-matched tumor and PBMC samples were co-cultured with pembrolizumab or immunoglobulin G (IgG) control for 7 days and analyzed by scRNA-seq and cytokine profiling. All major tumor and immune populations were preserved across conditions (Figure 3A). Detailed annotations and marker genes were additionally explored (Figures S6A and S6B).

Figure 3. Pembrolizumab treatment in iHOTT recapitulates cell type and interaction changes observed in treated GBM patients.

(A) UMAPs of iHOTT samples annotated by patient, treatment condition, and cell type showing preservation of all major tumor and immune populations in IgG-and pembrolizumab-treated cultures.

(B) Cytokine heatmap of iHOTT supernatants from three tumors treated with pembrolizumab or IgG control, showing increased IL-15, IL-25, and IL-17F following treatment. Heatmap coloring represents log2 fold change in aPD1 samples relative to IgG controls (mean of all replicates).

(C) UMAPs of recurrent GBM patient tumors treated with placebo or pembrolizumab, annotated by patient, treatment, and immune cell type, demonstrating preservation of major immune subsets comparable to iHOTT.

(D) Cell-type abundance changes after pembrolizumab in iHOTT and patient tumors show parallel increases in CD4⁺ T cells, B cells, innate lymphoid cells (ILCs), and non-canonical T cell subsets. Bar graphs represent the percentage composition across our single-cell dataset.

(E) CellChat interaction analysis shows increased signaling involving ILCs and non-canonical T cells after pembrolizumab in both iHOTT and patient tumors, with persistent interactions to CD8⁺ T cells and NK cells across conditions.

Subclustering confirmed retention of major immune cell types, with modest increases in CD4⁺ T cells and B cells after pembrolizumab treatment (Figure S6C). Differential pathway analysis showed enrichment of immune activation programs, including B-cell-mediated immunity, humoral responses, and MHC class II signaling, in pembrolizumab-treated samples relative to input PBMCs (Figure S6D).

Subclustering of tumor cells demonstrated preservation of neuronal and mesenchymal populations in both conditions, with modest reductions in other tumor subsets after pembrolizumab treatment (Figure S7A). Pathway analysis revealed increased biosynthetic, synaptic, and ribonucleotide processing programs, indicating that PD-1 blockade reshapes tumor transcriptional states in addition to activating immune pathways (Figure S7B).

Flow cytometry showed no increase in tumor cell death after pembrolizumab treatment, and cell-cycle analysis revealed no major changes in G1, S, or G2/M phase distribution, indicating that checkpoint inhibition does not directly alter tumor viability or proliferation (Figures S7C and S7D).

Pembrolizumab treatment in iHOTT alters cytokine signaling profile

Cytokine profiling of iHOTT revealed key changes in response to pembrolizumab treatment (Figures 3B, S8A, and S8B). Compared to IgG control, pembrolizumab increased secretion of IL-15, IL-17F, and IL-25 (IL-17E). IL-15 promotes T and NK cell proliferation and survival, consistent with enhanced cytotoxic lymphocyte activation.^47–49 IL-17F, produced by Th17 cells, is associated with inflammatory activation in the tumor microenvironment.⁵⁰ IL-25 shapes type 2 immune responses and may contribute to lymphocyte recruitment or polarization.⁵¹ Together, these cytokine shifts indicate increased immune activation in response to PD-1 blockade.

iHOTT mirrors immune cell type shifts observed in treated patient samples

Forty-two patients with recurrent glioblastoma were previously enrolled in a randomized trial comparing pembrolizumab versus placebo at first or second recurrence (Table S2). Eighteen patients received placebo and twenty-four received anti-PD-1 therapy, with a sex distribution consistent with known GBM demographics.⁵²

The immune fraction of these tumors was isolated for scRNA-seq, confirming preservation of all major immune cell types with myeloid predominance (Figure 3C). This myeloid cell enrichment is consistent with prior reports and expected from freshly dissociated samples lacking in vitro expansion.⁵³

Immune cell distributions in pembrolizumab-treated versus untreated patients closely mirrored those observed in iHOTT following pembrolizumab exposure, supporting the model’s relevance for recapitulating autologous immune responses to checkpoint blockade (Figure 3D).

iHOTT mirrors shifts in cell-cell interactions observed in treated patient samples

Cell-cell interaction analysis was performed in iHOTT control and pembrolizumab-treated samples and compared with untreated and pembrolizumab-treated patient tumors. As expected, iHOTT demonstrated extensive tumor-immune interactions (Figure S9A). Subclustering of immune cells showed that CD8⁺ T cells were major interaction receivers in all samples, and the increase in signaling to unconventional T cell subsets (γδ T cells, MAIT cells, and double-negative T cells) seen in treated patients was also observed in iHOTT (Figure 3E). While CD8⁺ T cell involvement is established, emerging evidence suggests these noncanonical T cell subsets may also expand under checkpoint blockade.⁵⁴

Pembrolizumab-treated iHOTT samples showed enrichment of resistin, WNT, complement, and CSF pathways (Figure S9B). Immune-only subclustering revealed parallel pathway changes in patients and iHOTT, including decreased CD45 signaling and increased complement, CD226, TIGIT, and CD70 signaling in treated samples, consistent with enhanced co-stimulation and cytotoxic engagement (Figures S10A–S10C).⁵⁵ CD226 and TIGIT belong to the TIGIT-CD226-CD96 checkpoint axis, which regulates activation-inhibition balance through shared ligand competition.^56,57 These concordant pathway shifts underscore that iHOTT captures clinically relevant immunologic features of PD-1 blockade. Cell types contributing to these pathways were likewise consistent between patients and iHOTT, further supporting the model’s physiological relevance (Figure S10C).

T cell receptor sequencing in iHOTT reveals an increase in clonotype diversity due to the production of novel clones

TCRs are highly diverse antigen-recognition molecules generated by somatic recombination of V, D, and J segments, producing a vast repertoire of clonotypes.⁵⁸ TCR sequencing allows for high-resolution profiling of T cell clonotypes (shared TCR sequences in a population), offering insight into the clonal dynamics of an immune response.^59–61 Given that pembrolizumab treatment in iHOTT led to T cell expansion and increased tumor-immune interactions, we performed TCR sequencing across patient tumors and iHOTT samples.

Analysis of input PBMCs, IgG controls, and pembrolizumab-treated co-cultures demonstrated progressive shifts in clonal composition with tumor exposure and PD-1 blockade. The proportion of unique clonotypes increased stepwise, indicating diversification driven by antigenic stimulation and immune modulation. Pembrolizumab-treated samples also contained more rare clonotypes, suggesting expansion or recruitment of low-frequency T cell populations (Figure 4A). These patterns were accompanied by altered TCR chain length distributions and increased diversity indices in pembrolizumab-treated samples, consistent with a more heterogeneous clonal architecture (Figures S11A and S11B).

Figure 4. TCR sequencing reveals pembrolizumab-induced clonal remodeling of T cells in iHOTT and patient samples.

(A) TCR-seq of iHOTT samples showed stepwise increases in clonotype diversity from input PBMCs to IgG controls and pembrolizumab-treated cultures, with greater representation of rare clonotypes and emergence of new clones. Bar graphs represent the percentage composition across our single-cell dataset.

(B) Novel clonotype analysis revealed a higher proportion of new clonotypes in pembrolizumab-treated samples compared to input and control conditions. Bar graphs represent the percentage composition across our single-cell dataset. Scatterplot shows partial repertoire overlap and expansion of newly detected clones.

(C) UMAPs of reclustered iHOTT T cells annotated by treatment condition and TCR detection. Top expanded clonotypes mapped onto the UMAP showed CD4⁺-dominant expansion after pembrolizumab. Bar graphs represent the percentage composition across our single-cell dataset.

(D) Analogous TCR analysis in recurrent GBM patient tumors showed similar emergence of novel clonotypes and CD4⁺ enrichment among top clones in pembrolizumab-treated samples. Bar graphs represent the percentage composition across our single-cell dataset.

(E) GLIPH2 clustering demonstrated predominantly patient-specific (private) clonotypes in both iHOTT and patient datasets, with limited shared (public) clonotypes, consistent with antigenic heterogeneity in GBM.

To assess whether increased clonal diversity reflected expansion of new versus persistent clonotypes, we compared TCR repertoires across treatment conditions. “Novel” clonotypes were defined as those absent from input PBMCs but detected after co-culture or treatment. Approximately half of clonotypes in IgG control samples were novel, indicating expansion of previously rare T cell populations in response to tumor exposure. This effect was further amplified in pembrolizumab-treated samples, which contained a greater number of novel clonotypes, consistent with enhanced activation driven by PD-1 blockade (Figure 4B).

All pembrolizumab-treated samples demonstrated novel clonotype expansion relative to input PBMCs; however, clonotype overlap with input was similar between pembrolizumab and IgG conditions, indicating that increased diversity arises primarily from newly detected clonotypes rather than expansion of existing ones (Figure S11C).

Integration of TCR clonotypes with transcriptional profiles reveals CD4-driven response to pembrolizumab treatment in iHOTT and in patients

To investigate the cellular and transcriptional features of dominant T cell clonotypes, we integrated TCR sequencing with our single-cell dataset. T cells were re-clustered, annotated as previously described, and examined for subtype, progenitor, activation, and exhaustion markers (Figure S11D). Approximately 45% of T cells had TCR information. The five most expanded clonotypes (4–54 cells each) were mapped onto the T cell UMAP, revealing that dominant clones were associated with CD4⁺ T cell expansion (Figure 4C).

We next distinguished novel from persistent clonotypes across treatments. In pembrolizumab-treated co-cultures, novel clonotypes were significantly enriched within CD4⁺ T cells compared to controls (Figure S11E). Marker analysis demonstrated increased TCF7 and LEF1 expression and reduced TOX expression after PD-1 blockade, consistent with a stemlike precursor phenotype associated with effective checkpoint responses (Figure S11F).^20,62 Prior studies show that TCF7⁺ CD4 and CD8 T cells self-renew and generate functional effectors following PD-1 inhibition, suggesting that pembrolizumab promotes maintenance of stem-like CD4⁺ clonotypes in iHOTT.^20,62

Checkpoint inhibition responses in patients mirror iHOTT and reflect personalized clonal landscapes

TCR sequencing from untreated and pembrolizumab-treated patient tumors was integrated with matched single-cell transcriptomes, focusing on the T cell compartment. Assessment of unique clonotype frequency and overall clonal distribution showed heterogeneous patterns, with no consistent differences between treated and untreated samples, highlighting interpatient variability in clonal responses to PD-1 blockade (Figures S12A and S12B).

To examine treatment-associated changes, T cells were reclustered, and ~16% had paired TCR information. The top five clonotypes per condition (10–33 cells each) were identified; as in iHOTT, dominant clonotypes were enriched for CD4⁺ T cells, supporting concordance between patient tumors and our model (Figures 4D and S12C).

Shared antigen specificity was evaluated using GLIPH2, which groups TCRs based on predicted convergent antigen recognition.⁶³ Clusters were classified as private (single patient) or public (shared across individuals). Across both iHOTT and patient datasets, most TCR clusters were private, with only limited public clonotypes detected (Figure 4E). These findings suggest that T cell responses to pembrolizumab in glioblastoma are largely patient-specific, with minimal evidence of convergence onto shared antigens, likely contributing to the limited clinical activity of PD-1 blockade in GBM. iHOTT reproduces this individualized clonal architecture, reinforcing its relevance for modeling personalized immune responses.

DISCUSSION

In this study, we present iHOTT as a human, autologous organoid model that enables detailed evaluation of tumor-immune interactions in glioblastoma. Using multimodal characterization, we show that iHOTT benchmarks closely to patient tumors and recapitulates patient-specific responses to pembrolizumab. The preservation of key tumor and immune compartments, together with treatment-induced features that parallel clinical samples, supports the physiological relevance of this system.

Patient-matched PBMCs were selected to model the early immune response to glioblastoma, reflecting the initial infiltration of circulating immune cells into the tumor microenvironment. Unlike tumor-infiltrating lymphocytes, which often display chronic exhaustion, PBMCs provide a responsive immune population encountering tumor antigens for the first time. The 7-day culture period was intentionally chosen to capture acute tumor-immune interactions while maintaining immune cell viability.^64,65 Although longer cultures may model chronic immune suppression, the current design is optimized to study early activation and checkpoint blockade responses.⁶⁶ Future adaptations could extend the system to model additional immune states by prolonging co-culture duration, incorporating myeloid-driven immunosuppressive signals, or conditioning PBMCs ex vivo to mimic exhausted phenotypes.

Our data indicate that iHOTT is particularly well suited for studying T-cell-mediated and CAR-T-based immunotherapies. The platform supports autologous tumor-T cell interactions and enables tracking of clonal dynamics, antigen specificity, and functional states, including activation, exhaustion, and stem-like phenotypes, at single-cell resolution. This capability is especially relevant for CAR-T studies, where antigen heterogeneity, immunosuppressive signaling, and limited T cell persistence hinder therapeutic efficacy. In iHOTT, engineered T cells can be introduced into a human tumor microenvironment and monitored for proliferation, cytotoxic activity, and target engagement using integrated flow cytometry, cytokine profiling, and transcriptomic readouts. In short, it enables comprehensive mechanistic dissection of T cell behavior in a way that static endpoint assays cannot achieve.

Beyond mechanistic discovery, iHOTT holds promise for translational applications such as preclinical drug testing, immune engineering, and personalized immunotherapy development. Its ability to capture inter-patient variability, including individualized clonotype usage and immune responses, supports its use as a functional precision medicine tool. By modeling key cellular, transcriptional, and clonal features of human tumor-immune interactions, iHOTT provides a versatile platform to accelerate development of patient-tailored immunotherapeutic strategies for glioblastoma.

Limitations of the study

Although major lymphocyte populations were preserved, we noted a relative underrepresentation of myeloid cells in the iHOTT system compared to fresh patient tumor samples. This discrepancy likely reflects limited myeloid cell survival or retention in culture, a challenge commonly observed in in vitro models due to the short lifespan and high plasticity of monocytes and dendritic cells ex vivo. The absence of tissue-derived stromal or vascular cues may also limit myeloid cell recruitment and maintenance. This highlights an important area for future optimization, particularly given the central role of myeloid populations, including monocyte-derived macrophages and microglia, in glioblastoma immunosuppression, antigen presentation, and T cell regulation.

Potential strategies to enhance myeloid representation include media supplementation with cytokines such as GM-CSF, IL-4, or M-CSF to support differentiation and survival or additional chemokine modulation. Extending co-culture duration or including patient-derived microglia and macrophages may also help capture the full spectrum of tumor-associated myeloid diversity. Addressing this limitation will be crucial for leveraging iHOTT to study immunosuppressive circuits and evaluate therapeutic interventions targeting the myeloid compartment in glioblastoma.

RESOURCE AVAILABILITY

Lead contact

Requests for further information should be directed to and will be fulfilled by the lead contact, Aparna Bhaduri (abhaduri@mednet.ucla.edu).

Materials availability

Further information and inquiries for resources and reagents should be directed to the lead contact. This study did not generate new unique reagents.

STAR★METHODS

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell & tissue culture

This study involved the use of multiple human-derived components, including primary tumor tissue, peripheral blood mononuclear cells (PBMCs), cortical organoids, and gliomasphere cultures. Detailed protocols for each of these components are described in the relevant sections below. Briefly, primary glioblastoma tumor tissue and matched peripheral blood were obtained intraoperatively from consented patients under IRB-approved protocols (see patient tissue acquisition). Cortical organoids were generated from human embryonic stem cells (UCLA6 line) and matured for use in co-culture experiments as detailed in cortical organoid generation. All cell cultures utilized in this study were tested regularly for mycoplasma contamination and were found to be negative. All lines were authenticated by short tandem repeat (STR) analysis and karyotyping at acquisition from the Broad Stem Cell Core.

Gliomasphere culture

This study also utilized patient-derived gliomasphere lines (generously provided by the Nathanson Lab at UCLA) for the optimization experiments before the utilization of fresh patient tissue. Gliomaspheres were cultured in Neurosphere Media composed of DMEM/F12 (ThermoFisher cat# 11330057), 5 mL of 100x Glutamax (ThermoFisher cat# 35050061), 5 mL of 100x Pen-Strep (ThermoFisher cat# 15140122), and 10 mL of 50x B27 Supplement without Vitamin A (ThermoFisher cat# 12587010) per 500 mL bottle. Cultures were supplemented every 5 days with 1x HEF (2.5 μL of 400x stock per mL of media), a mixture containing 5 μg/mL heparin, 20 ng/mL FGFb (ThermoFisher cat# PHG0263), and 50 ng/mL EGF (ThermoFisher cat# PHG0313).

Cells were maintained in non-adherent, filter-cap cell culture–treated flasks at 37°C with 5% CO₂. Culture density was kept between 50,000 and 200,000 cells/mL, with an optimal range of 100,000 cells/mL. Spheres were passaged every 4–10 days based on growth rate, sphere size, and media color; a 7-day interval was typically sufficient. Cells were dissociated using TrypLE Express (ThermoFisher cat# 12605028) and frozen in Bambanker freezing media (Bulldog Bio cat# BB01) when needed.

Patient tissue acquisition

Primary glioblastoma specimens and matched peripheral blood were collected from patients undergoing surgery at Ronald Reagan UCLA Medical Center under IRB-approved protocols (IRB #10–000655 and #21–000108), following informed research consent. Samples were processed immediately following surgical resection for either tumor dissociation or PBMC isolation as described below.

The number and clinical characteristics of patients included in each experiment are summarized in Table S2. Overall, the cohort reflects the known sex distribution of glioblastoma, with a male-to-female ratio of approximately 1.6:1. For initial iHOTT validation experiments (without treatment), six tumors were used, with at least three represented in each condition: PBMC-only, tumor-only, and iHOTT (tumor–PBMC co-culture). All flow cytometry, immunofluorescence, scRNA-seq, and cytokine profiling assays were performed across a minimum of three tumors, dependent on clinical sample availability. For iHOTT pembrolizumab experiments, three tumors were used consistently across all assays. For comparison with clinical treatment responses, data from 42 recurrent glioblastoma patients were analyzed, including 18 patients treated with a placebo and 24 patients treated with an anti–PD-1 agent, and all samples were included in downstream analyses.

Tumor cell dissociation

Tumor samples were obtained fresh from the operating room and were processed on the same day. Tissue was first manually dissected into small fragments using a sterile scalpel. The fragments were then transferred to 5 mL tubes containing 2.5 mL of Papain and 125 μL of DNase I prepared per manufacturer’s protocols (Worthington). Samples were incubated at 37°C for 45–60 min, with intermittent manual agitation every 5-10 min to promote tissue breakdown. Following incubation, tissues were gently triturated and centrifuged at 300 × g for 5 min to isolate a pellet of single cells. The papain-containing supernatant was removed, and the cell pellet was resuspended in Sasai3 media (recipe described above in Organoid Preparation).

Tumor cell suspensions were passed through a 40 μm cell strainer to remove debris and counted using a hemocytometer. For GFP labeling, 5 million tumor cells were resuspended in 1.5 mL total volume in Eppendorf tubes containing 1 mL Sasai3 media, 1 μL polybrene (final concentration 8 μg/mL), and 2 μL of CMV-GFP lentivirus (SignaGen) per 1 million cells. Tubes were incubated on a rotator at 37°C for 90 min to allow transduction. After incubation, cells were washed three times with PBS to remove excess virus and polybrene, and the final pellet was resuspended in fresh media for immediate use in organoid co-culture experiments, as described in the iHOTT model generation and culture section.

PBMC isolation

Peripheral blood was collected intraoperatively from patients and processed fresh on the day of surgery. Blood was diluted up to 50 mL with sterile PBS and gently mixed by pipetting. The diluted blood was then carefully layered over 20 mL of Ficoll-Paque PLUS (GE/Cytiva) in two separate 50 mL conical tubes (10 mL per tube), allowing the blood to drip slowly down the side to avoid disturbing the interface. Samples were centrifuged at 400 × g for 30 min at 18°C with no brake on the centrifuge.

The PBMC layer typically appears as a white, cloudy interface after centrifugation. It was extracted using a sterile plastic Pasteur pipette by gently aspirating just above the Ficoll layer. Extracted cells were transferred to a clean 50 mL tube and topped off with PBS. Samples were centrifuged at 400 × g for 10 min at 4°C. The supernatant was poured off, and cells were resuspended in 10 mL of PBS and centrifuged again under the same conditions. Cells were then counted using Trypan Blue exclusion and a hemocytometer.

PBMCs were either used fresh for transplant, and any remaining PBMCs were cryopreserved. For cryopreservation, cells were resuspended at 5–20 million cells/mL in freezing media composed of 90% heat-inactivated fetal bovine serum (FBS; filtered) and 10% DMSO. 1 mL of cell suspension was aliquoted per cryovial and frozen at −80°C.

Cortical organoid generation

Human embryonic stem cells (hESCs, UCLA6 line) were maintained under feeder-free conditions in Matrigel-coated 6-well plates using mTeSR Plus medium (STEMCELL Technologies) supplemented with 10% mTeSR Plus Supplement and 1× Primocin (InvivoGen). Media was replaced every other day, and cells were passaged when cultures reached approximately 75–90% confluence.

For passaging, ReLeSR (STEMCELL Technologies) was added to each well and incubated at room temperature for 1 min before aspiration. Plates were then incubated at 37°C for an additional 5 min to facilitate detachment. Cells were gently disaggregated into small clusters and replated at a 1:4 to 1:6 ratio onto fresh Matrigel-coated plates. For cryopreservation, passaged cells were resuspended in mFreSR freezing media (STEMCELL Technologies), aliquoted into cryovials, and frozen at −80°C for 24–48 h before transfer into liquid nitrogen for long-term storage.

Cortical organoids were generated as previously described.²⁹ Briefly, hESCs at ~80% confluence were incubated with 1 mL Accutase (STEMCELL Technologies) per well for 5 min at 37°C. Detached cells were collected, centrifuged at 300 × g for 5 min, and resuspended in Sasai1 medium, composed of GMEM (Gibco), 20% KnockOut Serum Replacement (ThermoFisher), 0.1 mM β-mercaptoethanol, 1× Non-Essential Amino Acids (NEAA), 1× Sodium Pyruvate, and 1× Primocin.

Cells were seeded at a density of 1 million cells divided evenly across a 96-well low-attachment V-bottom plate in Sasai1 medium. Sasai1 medium consisted of GMEM (ThermoFisher, cat# 11710-035), 20% KnockOut Serum Replacement (ThermoFisher, cat# 10828-028), 0.1 mM β-mercaptoethanol (Sigma, cat# M3148), 1× Non-Essential Amino Acids (CCF), 1× Sodium Pyruvate (CCF, 11 mg/mL stock), and 1× Primocin (CCF). This media was further supplemented with 20 μM Y-27632 (ROCK inhibitor), 5 μM SB431542 (TGF-β receptor inhibitor), and 3 μM IWR-1-endo (Wnt pathway inhibitor) to promote aggregation and neuroectodermal induction. Aggregates were cultured for 72 h undisturbed, after which half the media was replaced with fresh Sasai1 containing the same inhibitors. Media was then changed every other day until Day 7, with continued supplementation with inhibitors, after which Y-27632 was omitted from the medium.

On Day 18, organoids were transferred to ultra-low attachment 6-well plates and cultured in Sasai2 medium, composed of DMEM/F12 with Glutamax (ThermoFisher, cat# 10565-018), 1× N-2 Supplement (ThermoFisher, cat# 17502-048), Lipid Concentrate (ThermoFisher, cat# 11905-031), and 1× Primocin. From Day 35 onward, organoids were cultured in Sasai3 medium consisting of DMEM/F12 with Glutamax (cat# 10565-018), 1× N-2 Supplement (cat# 17502-048), Lipid Concentrate (cat# 11905-031), 10% FBS (ThermoFisher, cat# A5670801), 5 μg/mL heparin (Sigma, cat# H3149), 0.5% vol/vol growth factor–reduced Matrigel (BD Biosciences, cat# 354230), and 1x Primocin. Media changes were performed every other day throughout the culture period.

METHOD DETAILS

iHOTT model generation and culture

iHOTT co-cultures were established using mature cortical organoids (weeks 8–12) generated as described above in cortical organoid generation. Freshly dissociated primary tumor cells and matched PBMCs were prepared on the same day, following protocols out-lined in the tumor cell dissociation and PBMC isolation sections.

For co-culture, 500,000 GFP-tagged tumor cells and 500,000 PBMCs (1:1 ratio) were combined and transplanted onto organoids using a hanging drop method. Organoids were transferred to the inverted lid of a 10 cm dish using wide-bore pipette tips, and excess media was removed. A 10 μL droplet containing tumor cells and PBMCs was gently placed atop each organoid. The lid was then inverted over a 10 cm dish containing 10 mL of base culture media to maintain humidity and prevent evaporation. Hanging drop co-cultures were incubated at 37°C for 12–16 h before transfer to ultra-low attachment 6-well plates in Sasai3 medium. Tumor cells and PBMCs typically surrounded the organoid and began migrating inward over the following days.

Organoid co-culture models were then cultured for 7 days in 50-50 Sasai3:RPMI media without additional supplementation. Cultures were maintained under continuous rotation on an orbital shaker with media changes every other day. On day 7, organoids were harvested for downstream analysis, including immunofluorescence, flow cytometry, single-cell RNA sequencing, cytokine profiling, and TCR sequencing, as detailed in subsequent sections.

Immunofluorescence

Organoids were fixed in 4% paraformaldehyde (PFA) at room temperature for 45 min, followed by three washes with PBS. Fixed organoids were incubated overnight in 30% sucrose at 4°C, embedded in OCT compound over dry ice, and stored at −80°C. Cryo-sections (12 μm thick) were collected on Superfrost Plus slides and air-dried for at least 1 h at room temperature prior to storage and staining.

Sections were washed in PBS. Antigen retrieval was performed with a boiled citrate-based antigen retrieval solution for 20 min. Samples were then permeabilized and blocked with blocking buffer containing 5% donkey serum, 3% bovine serum albumin (BSA), and 0.1% Triton X- in PBS for 1 h at room temperature. Primary antibodies were diluted 1:200 in blocking buffer and incubated overnight at 4°C in a humidified chamber. The following primary antibodies were used.

• CD3 epsilon (rabbit, Abcam, cat# ab16669)

• CD14 (mouse, Abcam, cat# ab181470)

• CD19 (rabbit, Cell Signaling Technology, cat# 3574S)

• GFP (goat, Novus Bio, cat# NB100-1770)

Following primary incubation, sections were washed three times with PBS containing 0.1% Triton X-and incubated with species-specific secondary antibodies for 2 h at room temperature in the dark. The secondary antibody mix also included DAPI (Thermo Fisher) at a 1:1000 dilution. All secondary antibodies were diluted 1:1000 in blocking buffer and included.

• Alexa Fluor 488 anti-goat (donkey host), Invitrogen, cat# A32814

• Alexa Fluor 647 anti-mouse (donkey host), Invitrogen, cat# A32787

• Alexa Fluor 647 anti-rabbit (donkey host), Invitrogen, cat# A31573

Nuclei were counterstained using DAPI (ThermoFisher) for 5 min before a final PBS wash. Slides were mounted with ProLong Gold Antifade Mountant (ThermoFisher) and stored at 4°C prior to imaging.

Imaging was performed on an EVOS cell imaging system. Identical exposure and acquisition settings were applied across samples for any experiments involving quantification. Representative fields were selected for display.

Dissociation, sorting, and flow cytometry

Organoids were harvested on day 7 of co-culture and dissociated into single-cell suspensions using enzymatic digestion. For each sample, tumor-transplanted organoids were transferred to a 1.7 mL Eppendorf tube containing 1 mL of Papain solution and 50 μL of DNase I (Worthington), then incubated at 37°C for approximately 30 min. To enhance tissue breakdown, samples were shaken vigorously by hand for 10 s every 5 min during incubation. After enzymatic digestion, samples were triturated with a P1000 pipette and centrifuged at 300 × g for 5 min. The resulting cell pellet was resuspended in cold FACS buffer (PBS with 2% FBS and 0.5 mM EDTA), filtered through a 40 μm strainer, and stained with DAPI (1 μg/mL) for viability assessment.

For FACS sorting, single-cell suspensions were analyzed on a BD FACSAria machine. Cells were gated to exclude debris, doublets, and dead cells using forward/side scatter and DAPI exclusion. GFP+ tumor cells and CD45⁺ immune cells were sorted for downstream applications, including single-cell RNA sequencing. GFP gating thresholds were set using non-transplanted control organoids as negative controls.

In parallel, flow cytometric profiling of dissociated iHOTT samples was performed using an Attune NxT Flow Cytometer (ThermoFisher). Cells were stained with a panel of fluorophore-conjugated antibodies to define immune subpopulations. The following antibodies were used.

• CD3 (AF700, clone UCHT1, ThermoFisher, cat# 56-0038-80)

• CD14 (APC, clone M5E2, BioLegend, cat# 301807)

• CD19 (APC-eFluor780, clone HIB19, ThermoFisher, cat# 47-0199-41)

• CD45 (PE, clone HI30, ThermoFisher, cat# 12-0459-42)

Staining was performed on ice for 30 min in FACS buffer, followed by two washes and resuspension in DAPI-containing FACS buffer. Fluorescence minus one (FMO) controls were used to define gates. Data were analyzed using FlowJo software (BD Biosciences), and populations were quantified based on consistent gating strategies across experiments.

Cytokine profiling

Conditioned media was collected from organoid co-cultures on day 7 of culture. For each sample, 20 μL of culture media was carefully harvested in triplicate and immediately stored at −20°C until further processing. Samples were submitted to the UCLA Immune Assessment Core for analysis using a 37-plex Luminex cytokine and chemokine panel.

All samples were run in technical triplicate, and cytokine concentrations were quantified against standard curves using the manufacturer’s software. Raw data were exported and analyzed using Microsoft Excel and R/RStudio. Cytokine values were normalized to the relevant control population and used to generate comparative heatmaps and bar plots. Samples falling below the detection threshold for a given cytokine were assigned a value of zero for downstream analysis.

scRNASeq cell capture

Single-cell suspensions were generated from dissociated organoid co-cultures following FACS enrichment (GFP+ tumor or CD45⁺ immune cells). Cell suspensions were processed using the Chromium Controller and the Chromium Next GEM Single Cell 3′ v3.1 Reagent Kit (10x Genomics), following the manufacturer’s protocol. For each sample, up to 30,000 cells were targeted for capture. cDNA synthesis and library construction were performed according to 10x Genomics specifications. Libraries were sequenced on an Illumina NovaSeq 6000 platform at high depth to ensure saturation of the transcriptome complexity.

QUANTIFICATION AND STATISTICAL ANALYSIS

scRNAseq analysis

Raw sequencing reads were aligned and processed using the Cell Ranger pipeline (10x Genomics, v8.0.0) with default parameters. For human samples, reads were aligned to a standard GRCh38 reference genome obtained from 10x Genomics. Cell-by-gene UMI count matrices were generated for downstream analysis.

Initial data preprocessing was performed in R using the Seurat package (v5.0). Cells with fewer than 200 detected genes or greater than 10% mitochondrial gene content were excluded from the dataset. UMI counts were log-normalized and scaled with a size factor of 10,000. The top 2,000 most variable genes were identified, and principal component analysis (PCA) was applied to the scaled expression matrix. The number of principal components (PCs) used for downstream analysis was determined by identifying the maximum value between.

1. the squared standard deviation of each PC and

2. the square root of (number of genes/number of cells +1), squared.

Dimensional reduction and visualization were performed using UMAP based on the selected PCs. Cells were clustered using Seurat’s FindNeighbors and FindClusters functions with a resolution of 2.5.

Cell type annotation

Cell type annotations were performed using Seurat’s reference-based mapping pipeline, following the recommended MapQuery protocol described in the Seurat documentation. Briefly, query datasets were normalized and integrated using SCTransform, and canonical correlation analysis (CCA) was used to identify anchors between the query and reference datasets. These anchors were then used to transfer cell type labels onto the query data.

For tumor compartments, cells were projected onto a previously published GBM single-cell meta-atlas.²⁹ For immune compartments, reference-based mapping was performed using a previously published reference dataset, which includes matched gene expression and surface protein profiles across major immune lineages.²⁸ Annotations were manually reviewed and refined using known marker gene expression if needed. Annotation results were used to define major cell types for all downstream analyses, including interaction modeling, gene set enrichment, and clonotype integration.

GSEA

Gene set enrichment analysis was performed using the clusterProfiler package (v4.6.2) in R.⁶⁷ Differential gene expression was computed using Seurat’s FindMarkers function with default Wilcoxon rank-sum testing. Resulting gene lists were ranked by average log fold change for use in GSEA. Pathway enrichment was assessed using the GSEA() function from clusterProfiler, querying against multiple gene set collections, including the MSigDB Hallmark, Reactome, and GO Biological Process databases. Gene sets with adjusted p-values <0.05 were considered significantly enriched. Visualization of enriched pathways was performed using the dotplot function from the clusterProfiler suite.

Ligand-receptor signaling analysis

Cell-cell communication networks were inferred using the CellChat R package (v1.6.1), a tool for analyzing intercellular signaling based on known ligand-receptor interactions (67). Processed single-cell RNA-seq data were first subsetted by condition, and then normalized expression matrices and metadata were used to initialize CellChat objects.

CellChat’s built-in human ligand-receptor interaction database was used for all analyses. The workflow included identification of overexpressed genes and interactions, computation of communication probabilities, and aggregation of signaling networks. The computeCommunProb and computeCommunProbPathway functions were used to quantify global and pathway-specific communication, respectively. Sender-receiver relationships were visualized using circle plots, heatmaps, and hierarchical network diagrams. Differential pathway usage between conditions was computed using CellChat’s built-in statistical framework.

TCR sequencing and analysis

T cell receptor (TCR) sequencing was performed using the 10x Genomics Chromium Single Cell V(D)J platform. Matched gene expression and TCR libraries were generated using the 10x Genomics Single Cell 5′ V(D)J kit according to the manufacturer’s instructions. Samples were processed in parallel for gene expression (GEX) and TCR enrichment using cellranger multi, which enables simultaneous demultiplexing, alignment, and quantification of both libraries. For V(D)J alignment, reads were mapped to a human TCR reference downloaded from the official 10x Genomics reference repository.

TCR contigs were assembled and annotated by Cell Ranger, including CDR3 sequences, clonotype IDs, and V(D)J gene usage. Productive, paired alpha-beta TCRs were retained for downstream analysis. Resulting filtered_contig_annotations.csv files were imported into R and integrated with single-cell gene expression data using the scRepertoire package (v1.7.1). TCR metadata (e.g., clonotype frequency, clonal expansion, and clone type) was added to Seurat objects for visualization and subgroup analysis. Clonotype dynamics, repertoire diversity, and overlap across samples or treatment conditions were assessed using scRepertoire and immunarch packages in R, adapting developer vignettes to build our analysis pipelines.

To investigate potential shared antigen recognition among clonotypes, GLIPH2 (Grouping of Lymphocyte Interactions by Paratope Hotspots) was used to cluster TCRs by sequence-based similarity.⁶³ CDR3 sequences from productive, paired TCRs were input into GLIPH2, which outputs clusters of TCR CDR3 sequences predicted to recognize common epitopes. Output clusters were used to assess convergence within and across samples in our datasets.

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2025.116790.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
CD3 epsilon	Abcam	ab16669; RRID:AB_443425
CD14	Abcam	ab181470; RRID:AB_2725781
CD19	Cell Signaling Technology	3574; RRID:AB_2275523
GFP	Novus Bio	NB100-1770; RRID:AB_10128178
Alexa Fluor 488 anti-goat (donkey host)	Invitrogen	A32814; RRID:AB_2762838
Alexa Fluor 647 anti-mouse (donkey host)	Invitrogen	A32787; RRID:AB_2762830
Alexa Fluor 647 anti-rabbit (donkey host)	Invitrogen	A31573; AB_2536183
CD3 Monoclonal Antibody (UCHT1), Alexa Fluor™ 700, eBioscience	Thermo Fisher Scientific	56-0038-80; RRID:AB_906222
APC anti-human CD14	BioLegend	301807; RRID:AB_314189
CD19 Monoclonal Antibody (HIB19), APC-eFluor™ 780, eBioscience	Thermo Fisher Scientific	47-0199-41; RRID:AB_1582231
CD45 Monoclonal Antibody (HI30), PE, eBioscience	Thermo Fisher Scientific	12-0459-42; RRID:AB_1724079
Bacterial and virus strains
LV-CMV-GFP-Puro	SignaGen	SL100268
Biological samples
Patient Tumor Samples	UCLA Neurosurgery and Brain Tumor Translational Resource	See “Patient Sample Info” tab in Supplementary Tables
Matched Patient Peripheral Blood Mononuclear Cells	UCLA Neurosurgery and Brain Tumor Translational Resource	See “Patient Sample Info” tab in Supplementary Tables
Donor Peripheral Blood Mononuclear Cells	UCLA/CFAR Virology Core Laboratory	See “Patient Sample Info” tab in Supplementary Tables
Chemicals, peptides, and recombinant proteins
DAPI	Thermo Fisher Scientific	62248
Rock Inhibitor Y-27632	Tocris	1254
IWR-1-endo	Cayman Chemical	13659
SB431542	Tocris	1614
DMEM/F12	Thermo Fisher Scientific	11330057
GlutaMAX	Thermo Fisher Scientific	35050061
Pen-Strep	Thermo Fisher Scientific	15140122
B27 Supplement without Vitamin A	Thermo Fisher Scientific	12587010
Heparin	Sigma-Aldrich	H3149
Human FGF-basic Recombinant Protein	Thermo Fisher Scientific	PHG0263
Human EGF Recombinant Protein	Thermo Fisher Scientific	PHG0313
TrypLE Express	Thermo Fisher Scientific	12605028
Bambanker Cell Freezing Medium	Bulldog Bio	BB01
Papain Dissociation System	Worthington Biochem	PDS
Polybrene	Millipore Sigma	TR-1003-G
PBS, pH 7.4	Thermo Fisher Scientific	10010031
Ficoll-Paque	Cytiva	17544202
Trypan Blue Solution, 0.4%	Thermo Fisher Scientific	15250061
Fetal Bovine Serum, Premium	Thermo Fisher Scientific	A5670801
DMSO	Millipore Sigma	D2650-100ML
mTeSR Plus	STEMCELL Technologies	100–0276
Matrigel	Corning	354230
Primocin	InvivoGen	ant-pm-1
ReLeSR	STEMCELL Technologies	100–0483
mFreSR	STEMCELL Technologies	5855
Accutase	STEMCELL Technologies	07920
GMEM	Thermo Fisher Scientific	11710035
Knockout Serum Replacement	Thermo Fisher Scientific	10828028
2-Mercaptoethanol	Thermo Fisher Scientific	21985023
MEM Non-Essential Amino Acids Solution	Thermo Fisher Scientific	11140050
Sodium Pyruvate	Thermo Fisher Scientific	11360070
DMEM/F12 with Glutamax	Thermo Fisher Scientific	10565018
N-2 Supplement	Thermo Fisher Scientific	17502048
Chemically Defined Lipid Concentrate	Thermo Fisher Scientific	11905031
RPMI 1640 Medium	Thermo Fisher Scientific	11875093
16% Paraformaldehyde Aqueous Solution	Electron Microscopy Sciences	15710
Fisher Healthcare Tissue-Plus™ O.C.T. Compound	Fisher Scientific	23-730-571
Sucrose	Millipore Sigma	84097-1KG
Phosphate Buffered Saline, 10X Powder	Fisher Scientific	BP665-1
Normal Donkey Serum	Millipore Sigma	566460
MACS BSA Stock Solution	Miltenyi Biotec	130-091-376
Triton X-	Millipore Sigma	X100-100ML
ProLong Gold Antifade Mountant	Thermo Fisher Scientific	P36934
Ethylenediaminetetraacetic acid (EDTA) solution	Millipore Sigma	E8008-100ML
Critical commercial assays
Immunoassay and xCELLigence	UCLA Immune Assessment Core	–
Chromium Next GEM Single Cell 3′ v3.1 Reagent Kit	10X Genomics	PN-1000121
Single Cell 5′ GEM-X kit v3	10X Genomics	PN-1000701
Single Cell Human TCR Amplification	10X Genomics	PN-1000252
Deposited data
Single-Cell RNA-sequencing iHOTT dataset	UCSC Cell Browser	https://ihott.cells.ucsc.edu; dbGAP: phs003936.v1.p1
Single-Cell RNA-sequencing iHOTT Pembro dataset	UCSC Cell Browser	https://ihott.cells.ucsc.edu; dbGAP: phs003936.v1.p1
Reference patient Single-Cell RNA-sequencing dataset	Lee et al.57	GEO: GSE154795
Experimental models: Cell lines
UCLA 6 (RRID:CVCL_9957)	University of California, Los Angeles	NIHhESC; NIHhESC-11-0089
Gliomaspheres	Nathanson Lab at UCLA	GS005
Software and algorithms
R	Publicly Available	https://www.r-project.org/
R Studio	Publicly Available	https://posit.co/download/rstudio-desktop/
Seurat v5	Publicly Available	https://satijalab.org/seurat/
Microsoft Excel	Microsoft	https://www.microsoft.com/en-us/microsoft-365/excel
Cell Ranger 8.0	10X Genomics	https://www.10xgenomics.com/support/software/cell-ranger/8.0
clusterProfiler v4.6.2	Yu et al.67	https://github.com/YuLab-SMU/clusterProfiler
CellChat v1.6.1	Jin et al.68	https://github.com/sqjin/CellChat
scRepertoire v1.7.1	Borcherding et al.69	https://github.com/BorchLab/scRepertoire
GLIPH version 2	Huang et al.63	http://50.255.35.37:8080/
immunarch	ImmunoMind Team70	https://immunarch.com/
Other
Ultra-low attachment 6-well plates	Corning	3471
Ultra-low Attachment 96-well plates	SBio	MS-9096VZ
Rectangular Canted Neck Cell Culture Flask with Vent Cap	Corning	3289
40μm Cell Strainer	Corning	431750
100 mm TC-treated Culture Dish	Corning	430167
Superfrost Plus Microscope Slides	Fisher Scientific	1255015
Embedding Molds	Fisher Scientific	22–19

Highlights.

• iHOTT enables the co-culture of matched patient GBM and immune cells in cortical organoids

• iHOTT preserves tumor and immune populations, signaling pathways, and cell-cell interactions

• iHOTT recapitulates patient-specific immune shifts and clonal responses to PD-1 blockade

• TCR profiling in iHOTT reveals pembrolizumab-driven expansion of CD4 T cell clonotypes

Data and code availability

• Original datasets are viewable and downloadable on an interactive public UCSC Cell Browser: https://ihott.cells.ucsc.edu. Raw data have been deposited in dbGAP (Accession: phs003936.v1.p1). The reference patient dataset used for benchmarking our results was previously published.^57,67

• Code will be made available upon request.

• Additional data (images, flow cytometry, etc.) can be made available upon request.