CT2A, CT2A-EGFRvIII, and HEK293 cells were graciously donated by Dr. John Sampson’s lab. CT2A and CT2A-EGFRvIII were modified to express GFP and/or luciferase by transducing lentivirus made by the Duke Viral Core Facility (Addgene #89608 and #105621). Tumor cell lines were referred to as CT2A-GFP-Luc or 2A and CT2A-EGFRvIII-Luc or vIII. The presence of human EGFRvIII on CT2A-EGFRvIII-Luc was confirmed by flow cytometry with anti-human EGFRvIII antibody [L8A4] (Kerafast) (data not shown). Tumor cell lines were cultured in complete DMEM (cDMEM): DMEM, 10% FBS, 1%: Pen/Strep, Non-Essential Amino Acids (NEAA), and L-Glutamine. HEK293 cells, unless otherwise noted, were cultured in D10 media: DMEM with 4.5 mg/ml glucose and 0.11 mg/ml pyruvate (Gibco) + 10% FBS. T cells were cultured in T cell media (TCM) with or without human IL2 (donated from Dr. Sampson’s lab): RPMI 1640 (w/L-glutamine and sodium pyruvate), 10% FBS, 1%: Pen/Strep, NEAA, sodium pyruvate, L-glutamine, and 0.1%: β2-mercaptoethanol (Thermo Fischer), and 50 mg/mL gentamycin (Sigma-Aldrich).

CAR T cells were made using a previously established protocol (11). The pMSGV plasmid backbone with MSCV promoter was used to make third generation CAR T cells. Plasmids contained the human 139 single-chain antibody variable fragment (scFv) specific for EGFRvIII (19) along with transmembrane mouse CD8 and intracellular mouse CD3z, CD28, and 4-1BB signaling domains to enhance proliferation and function (51). When appropriate, plasmids were further modified to express mouse IL7 and/or Flt3L using an N-terminal secretion signal and P2A and/or T2A cleavage peptides. To make retrovirus, HEK293 cells were plated on a poly-l-lysine (Millipore Sigma) coated petri dish in D10 media. The following day, fresh D10 media was added to HEK293 cells and transfected using the previously described helper plasmid, pCL-Eco (Addgene), and lipofectamine 2000 (Invitrogen) in OptiMEM (Gibco). Spleens from 6- to 12-week-old female C57B6/J mice (Jackson Labs) were made into a single cell suspension using a 70 µm filter and washed with TCM without IL2. Cells were resuspended in RBC lysis buffer (BD) for 2 minutes and quenched with TCM. Cells were resuspended at 2x10^6 cells/mL in TCM with 2 µg/mL Concanavalin A (Sigma-Aldrich) and 50 IU/mL IL2 and plated in 24 well plates. The next day, fresh TCM without IL2 was added to HEK293 cells and, separately, a 24 well plate was coated with 25 µm/mL RetroNectin (Takara Bio) overnight at 4C. Concanavalin A activated splenocytes were transduced at 1x10^6 cells/mL with viral supernatant and spun for 1.5 hours at 2000 RPM at 32C. Fresh TCM with IL2 was added to each well. CAR expression on cells was assessed through flow cytometry on day 4 using a tetrameric peptide recognizing the CAR, CD3 Pe/Cy7 (17A2 BioLegend #100220), and CD8 PE (53-6.7 BioLegend #100708) on the NovoCyte 2060 or Cytek NL-3000. CAR T cells were used for in vivo experiments on day 5.

A fluorescently labeled peptide to identify CAR T cells specific to EGFRvIII was made similar to a previous study (11). A custom peptide conjugated with biotin (JPT Peptide Technologies) was added in a 10:1 molar ratio with streptavidin-Alexa 647 (Invitrogen) in PBS to make a 1 mg/mL peptide solution. After incubation for 1 hour at room temperature in the dark, the solution was diluted to 0.5 mg/mL in PBS and incubated for another 2 hours. The solution was further diluted to 0.1 mg/mL and passed through a 30 kDa MWCO polyethersulfone filter (Millipore Sigma) to remove the free conjugated peptide. The tetrameric peptide was aliquoted and stored in the dark at 4C until use.

ELISAs determined protein secretion of CAR T cells cultured alone or with tumor cells. 1x10^4 tumor cells in 200 µL of cDMEM were allowed to adhere for 5 hours in a 96-well plate. For the vIII + 2A condition, tumor cells were mixed in a 1:1 ratio with 0.5x10^4 EGFRvIII-positive and 0.5x10^4 EGFRvIII-negative cells before plating. cDMEM was removed from tumor cells and 1x10^5 CAR T cells were plated in 200 µL TCM without IL2. After 24 hours, the supernatant was harvested and frozen at -80C until assayed with IL7, Flt3L, IFNγ, or Granzyme B ELISA kits (R&D Systems) using the Spectramax i3x plate reader.

We used the bioluminescent signal from tumor cells as a surrogate marker for the tumor killing ability of CAR T cells. 1x10^4 tumor cells in 200 µL cDMEM were plated for 5 hours in an opaque 96-well plate. cDMEM was removed and replaced with 1x10^5 CAR T cells in 200 µL TCM without IL2. After 24 or 72 hours, Xenolight-D-luciferin (Perkin Elmer) was added at 150 µg/mL and incubated for 5 min until acquiring the signal on a Spectramax i3x plate reader (24 hour data not shown).

We assessed CAR T cell proliferation by co-culturing CAR T cells with EGFRvIII-positive tumor cells. 1x10^6 CT2A-EGFRvIII-Luc cells were allowed to adhere to a 24-well plate for 5 hours. Separately, CAR T cells were stained per the manufacturer’s protocol with CellTrace Far Red (Thermo Fischer) for 20 minutes at 37C. Tumor media was removed from tumor cells and 1x10^6 CAR T cells were plated in TCM without IL2. Cells were incubated at 37C for three days and processed for flow cytometry. Samples were stained using Live/Dead Fixable Green reagent (Thermo Fischer) and anti-mouse CD8a PE antibody (53-6.7 BioLegend #100708).

6- to 12-week-old female C57B6/J mice (Jackson labs) were used in accordance with the approved Duke Institutional Animal Care and Use Committee (IACUC) protocol. Mice were given Buprenorphine SR at 1 mg/kg and anesthetized using 3 to 5% isoflurane or 90 mg/kg ketamine with 10 mg/kg xylazine. Mouse skin was shaved and cleaned thrice with chlorhexidine and 70% ethanol. A small incision exposed the cranium and 1 to 2 drops of 0.25% bupivacaine were placed on the skull. Using a stereotaxic microinjector, a Hamilton syringe with a 25-gauge needle injected either 55,000 CT-2A-EGFRvIII-Luc or 75,000 CT2A-GFP-Luc plus 75,000 CT2A-EGFRvIII-Luc cells 1.5 mm posterior and left of bregma and 4 mm deep in 5 µL of 4000 cP methylcellulose (Sigma-Aldrich) with DMEM/F-12 (Thermo Fischer), sodium bicarbonate (Sigma-Aldrich), and HEPES (Thermo Fischer). The injection site was sealed with bone wax and wound clips closed the incision. Animals were monitored every two days for signs of distress or illness. 5 days following tumor injection, animals were imaged with IVIS Kinetic (Caliper Life Sciences) using an injection of 0.2 µm filtered Xenolight D-luciferin (Perkin Elmer) at 150 mg/kg body weight. Animals were randomized into treatment groups based on tumor size to allow for an equal distribution in each group. Animals without tumors based on bioluminescent imaging were excluded. On the indicated day, animals were irradiated with 0.5 Gy or 5 Gy irradiation using a Cesium-irradiator. The following day, animals were injected with 2x10^6 CAR T cells in 24 µL PBS at 60 µL/min at the same injection location.

Animals were euthanized 14 days following tumor inoculation. The brain was harvested and processed for flow cytometry based on previous protocols (52, 53). Briefly, animals were perfused with cold HBSS without Ca/Mg and the tumor-bearing hemisphere was placed in a digestion buffer (TCM without IL2, 50 mg/mL DNase I grade II from bovine pancreas (Roche), and 20,000 units/mL collagenase type IV (Gibco)). The digested tissue suspension was passed through a 100 µm filter and incubated on a tube rotator at 37C for 30 minutes. The suspension was filtered using a 70 µm filter, spun down at 500g, and resuspended in a 25% Percoll density gradient (GE Healthcare). After centrifuging at 521g at 18C for 20 minutes, the suspension went through multiple washes, red blood cell lysis buffer, and cell counting. The cells were blocked in mouse fc-block (BD biosciences), and stained with live/dead stain and antibodies. The following are the antibodies with clone noted and reagents from BioLegend unless otherwise noted: Zombie Aqua fixable viability kit (#423101), CD45 BV711 (30-F11 #103147), CD3 APC/Cy7 (17A2 #100222), CD8 BV650 (53-6.7 #100742), CD4 PE/Cy5 (GK1.5 #100410), CAR tetramer Alexa 647, PD-1 BV421 (29F.1A12 #135221), TIM3 BV605 (RMT3-23 #119721), LAG3 PE (C9B7W #125208), CD25 PE/Cy7 (3C7 #101916), CD69 BV785 (H1.2F3 #104543), CD11b APC/Cy7 (M1/70 #101226), F4/80 BV421 (BM8 #123137), CD11c BV605 (N418 #117334), MHCII PE (M5/114.15.2 #107607), CD103 PE-CF594 (M290 BD Biosciences #565849), CD45R PE/Cy7 (RA3-6B2 #103222), XCR1 APC (ZET #148206), CD86 BV785 (GL-1 #105043), and CD40 PE/Cy5 (3/23 #124618). The stained cells remained at 4C until read immediately on the BD LSRFortessa X-20. Animal studies were replicated twice. Data was analyzed using FlowJo v10.7.2.

RNA was isolated from the tumor-bearing hemisphere and extracted using the RNeasy Lipid Tissue Mini kit (Qiagen). RNA purity was determined using the NanodropOne (Invitrogen) along with integrity and concentration measured with a Fragment Analyzer. The bulk RNA sequencing library was prepared using polyA tail enrichment with the KAPA Stranded mRNA-Seq kit (Roche). Sequencing was performed on the NovaSeq (Illumina) with 151 paired-end reads. For TCR sequencing, RNA was further processed with the mouse T-cell Receptor Panel QIAseq Immune Repertoire RNA Library Kit (Qiagen) using unique molecular indices with gene specific primers. Samples were run on MiSeq Version 3 (Illumina) using 300 paired-end reads. TCRseq data was analyzed using the online Qiagen platform with the IMSEQ algorithm (54). RStudio (R 4.2.1) was utilized for data processing, visualization, and statistics. For bulk RNAseq data, quality control was performed using FastQC/MultiQC. Star Alignment was run using default parameters and soft clipping for the Illumina universal adapter sequence. FeatureCounts was used for expression quantification of alignment output. Downstream analysis of the FeatureCounts raw counts output matrix was performed using DESeq2 (version 1.36.0). For differential gene expression analysis, to find genetic differences from responders and non-responders determined by LUC counts, one sample from each group was excluded based on principal component analysis clustering. The alternative shrink estimator ashr with a benjamini-hochberg correction was used to control for false discovery rates (FDR) (55). Differential gene expression was determined using an FDR < 0.1. The CIBERSORTx tool was used to infer cell fractions based on RNA sequencing data using the wild type samples from Seurat objects created from GSE197879 with S-mode batch correction and 100 permutations (56-58). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis was performed using fast gene set enrichment on genes with a non-adjusted p-value < 0.05 (59). Cytosig analysis was performed using mean-centered log transformed data (60).

When applicable, data has been presented using the mean + or - SEM. ELISA data was normalized using 10^x before statistical analysis. When applicable, a one-way analysis of variance (ANOVA) with Tukey’s posthoc test was used for analysis. When noted in the figure legend, groups were compared within tumor types (vIII, vIII + 2A, or 2A). The Kaplan-Meier plot was used for the survival curve and significant differences were assessed using the log-rank Mantel-Cox test. For TCR sequencing diversity metrics, a Kruskal-Wallis test with Dunn’s multiple comparisons was used to determine significance. A p-value of < 0.05 was considered significant. * = denotes significance to all other groups in the comparison. ns = denotes not significant. Unless otherwise noted, all statistical analysis was conducted using Graphpad Prism version 9.0.2.

Our first goal was to determine the functionality of IL7 Flt3L CAR T cells in vitro. We modified third generation CAR T cells by cloning mouse IL7 and/or Flt3L with an N-terminal secretion signal along with self-cleaving P2A and/or T2A peptides ( Figure 1A ). We confirmed no significant differences in the amount of CAR T cells from transduced splenocytes between CAR (vCAR), Flt3L (vFL), IL7 (vIL7), or IL7 Flt3L (vIL7FL) CAR T cells ( Figure 1B ). Additionally, cells were mostly CD8 CAR T cells due to the culture conditions. We then quantified secreted Flt3L and IL7 when CAR T cells were cultured alone or co-cultured with either CT2A-EGFRvIII-Luc (vIII), CT2A-GFP-Luc (2A) or 50% vIII and 50% 2A (vIII+2A) tumor cells. After 24 hours, we observed secretion of Flt3L and IL7 from CAR T cells programed to express those respective cytokines ( Figure 1C ). Interestingly, the expression of Flt3L remained stable despite the introduction of antigen-negative tumor cells, while IL7 expression seemed to diminish as 2A cells were introduced, although not significant. We further investigated delivering 50% vIL7 and 50% vFL, referred to as 7&3. While not significant, we did find lower levels of Flt3L for 7&3 compared to vFL and vIL7FL. We confirmed that CAR T cells secreted effector proteins through the expression of pro-inflammatory cytokine interferon gamma (IFNγ) and cytotoxic protein granzyme B ( Supplementary Figure 1 ). Next, we measured tumor killing through a surrogate marker of bioluminescence signal due to luciferase expression in tumor cells. When CAR T cells were co-cultured with vIII the signal diminished compared to non-transduced splenocytes demonstrating effective elimination of tumor cells ( Figure 1D ). CAR T cells in the presence of vIII + 2A failed to eliminate luminescence signal. Notably, vCAR and vIL7 both significantly reduced luminescence signal compared to the non-transduced control when cultured with 2A for 72 hours. Since we confirmed IL7 secretion for CAR T cells, we assessed the functional effect of IL7 on T cell proliferation. CAR T cells were co-cultured with vIII at a 1:1 ratio and cellular proliferation was quantified after three days. Low proliferation was defined as the peak with the highest amount of signal, medium proliferation represented the intermediate population, and high proliferation reflected the population with the lowest signal ( Figure 1E ). In comparison to vCAR, vIL7 and vIL7FL showed a significant increase in the percentage of medium and high proliferating CD8 T cells ( Figure 1F ). Thus, IL7 secreted from vIL7 and vIL7FL had the capacity to increase proliferation in vitro.

Since we observed increased proliferation of CAR T cells secreting IL7 in vitro, we assessed the intratumoral presence of CAR T cells 7 days after in vivo delivery. Previous studies utilizing TMZ for lymphodepleting pre-conditioning demonstrated peak CAR T cell abundance in the blood one week post-treatment and showed significant increases in intratumoral CAR T cells 7 days after treatment (11). 100% EGFRvIII positive tumors were inoculated into mice that received 5 Gy total body irradiation (TBI) the day before intracranial injection of CAR T cells ( Figure 2A ). One week after CAR T cell delivery, tumor tissue was processed for flow cytometry. vIL7 and vIL7FL significantly increased the CD8+ CAR T cell percentage of CD45+ cells compared to vCAR and vFL ( Figure 2B ). There was no change in CAR-negative (CAR-) T cells between groups. This is most likely due to the lymphodepletion caused by 5 Gy irradiation (13).

Next, we examined the ability of vIL7FL to alter immune cell populations in a more rigorous animal model. Animals were inoculated with tumors containing 50% EGFRvIII positive and 50% EGFRvIII negative cells. Prior to CAR T cell injection, a non-lymphodepleting dose (0.5 Gy) of irradiation was administered ( Figure 3A ). Compared to 100% EGFRvIII syngeneic mouse tumors, this model better recapitulates antigen heterogeneity while preserving the endogenous immune system and the anti-tumor benefits of irradiation. Using a multi-color flow cytometry panel, the tumor-bearing hemisphere was analyzed one week after CAR T cell delivery for T cells and dendritic cells (DCs) ( Supplementary Figure 2 ). vIL7 and vIL7FL significantly increased intratumoral CAR T cells compared to vCAR ( Figure 3B and Supplementary Figure 3 ). Interestingly, vIL7FL had significantly higher CD8 CAR T cells compared to all groups, including vIL7. This could suggest a synergistic effect from IL7 and Flt3L. There was no significant difference in CAR- CD8 T cells. The introduction of any CAR T cell significantly reduced the proportion of PD1+TIM3+LAG3+ of CAR- CD8 T cells, referred to as phenotypically exhausted T cells, compared to PBS. Notably, vIL7 significantly decreased exhausted T cells compared to vFL. Additionally, we assessed activation of CAR- T cells through expression of CD25 or CD69. While the proportion of CD25+ or CD69+ T cells of CD8 CAR- T cells did not significantly differ, vIL7 significantly increased absolute counts of CD69+ CD8 CAR- T cells compared to PBS, vCAR, and vFL ( Supplementary Figure 3 and Figure 3B ). Next, we investigated intratumoral conventional DCs (cDCs) due to their importance in antigen presentation. cDCs were classified by surface expression of CD45+F480-CD11c+CD11b-CD45R-MHCII+. vIL7FL enhanced intratumoral cDCs as well as the CD103+XCR1+ subset known as migratory antigen-presenting DCs ( Figure 3C ). There was no significant difference in plasmacytoid DCs.

We evaluated the therapeutic outcome of vIL7 and vIL7FL in EGFRvIII heterogeneous tumors treated with non-lymphodepleting irradiation. Animals were subjected to 0.5 Gy TBI 6 days after tumor inoculation with 50% EGFRvIII positive and 50% EGFRvIII negative tumor cells. On day 7, CAR T cells were delivered intracranially, and animals were monitored for survival. vIL7 and vIL7FL significantly enhanced survival compared to PBS, vCAR, and vFL ( Figure 4A ). Interestingly, vCAR and vFL had a survivor which could be due to the bystander effect from using irradiation (61). Bioluminescent imaging of tumors 5 days post-treatment indicated repression of tumor burden in vIL7 and vIL7FL ( Supplementary Figure 4A ). To understand the effect of vCAR, vIL7, or vIL7FL treatment on a transcriptional level, we evaluated tumor gene expression 7 days after CAR T cell delivery using bulk RNA sequencing. First, we assessed inferred cell fractions using CIBERSORTx. We utilized previously published single cell RNAseq data of mouse CD45 pre-sorted CT2A tumors to infer immune cell populations from bulk RNAseq data. The cDC1 population was significantly increased in vIL7FL compared to vIL7, which correlated with our flow cytometry data ( Figure 4B ). Since EGFRvIII positive tumor cells expressed EGFRvIII and luciferase (LUC) and EGFRvIII negative tumor cells expressed GFP and LUC, we used these transgenes as a surrogate markers for antigen-positive and negative tumor cells. GFP and LUC transgenes displayed a responder and non-responder effect ( Figure 4C ). The principal component analysis revealed that one sample from each group did not cluster with the other samples and samples were differentiated by LUC counts ( Figure 4D ). Additionally, the responder status did not necessarily indicate increased CAR reads ( Figure 4D ). Thus, to determine transcriptional differences between responders and non-responders to generate hypotheses for future studies we excluded the following samples that didn’t cluster: vCAR 3, vIL7 2, and vIL7FL 3. LUC and GFP were differentially expressed between vIL7 vs. vCAR and vIL7FL vs vCAR ( Figure 4E ). This was to be expected based on our exclusion criteria. There were no differentially expressed genes between vIL7 vs vIL7FL. vCAR treatment upregulated various differentially expressed immunosuppressive and immunostimulatory cytokines, including IL7 ( Figure 4F ). KEGG pathway analysis found enrichment in the neuroactive ligand receptor interaction pathway ( Figure 4G ). Upregulation of this pathway has been associated with glioblastoma (62, 63). However, one study found GBM patients with deficiencies in the neuroligand receptor interaction pathway have a poor prognosis due to mutations or low expression of Calcr (64). vIL7 and vIL7FL increased differential gene expression of Calcr compared to vCAR. However, the role of Calcr in GBM remains ambiguous. One common adverse event in CAR T cell therapy is cytokine release syndrome (CRS). CRS can occur 1 to 14 days post-CAR T cell therapy resulting in elevated levels of cytokines, including IL2, IL6, IL10, and TNF (65, 66). We did not observe overt weight loss 14 or 15 days post CAR T cell delivery in either experimental cohort ( Supplementary Figure 4B ). Additionally, the Cytosig platform was used to predict cytokine signaling in tumors 7 days post CAR T cell injection using gene expression data from bulk RNA sequencing (60). While there was a slight upregulation of IL6 and CD40L, which is known to induce IL6, TNF and IL2 were down-regulated in the treatment responders ( Supplementary Figure 4C ) (67, 68).

Finally, we investigated the impact of vIL7FL on T cell receptor beta chain (TRBC) diversity. We assessed the T cell receptor (TCR) alpha and beta chains due to their importance in immunotherapy (69). The alpha and beta protein chains are created through recombination of variable (V), joining (J) and, in certain cases, diversity (D) gene segments. The alpha chain utilizes V-J recombination, while the beta chain uses V-(D)J recombination. While both are required together for antigen recognition, the TRBC is recognized as more uniquely expressed in T cells than the alpha chain, due to the additional D recombination and allelic exclusion, and therefore is the focus of this study (70–72). First, we compared the observed number of clonotypes, and asymptotic diversity metrics found by extrapolating the unique molecular indices to infinity ( Figure 5A ). The asymptotic number of clonotypes was found using chao 1 statistics. While not significant, vIL7FL tended to have more clonotypes and higher indexes of diversity. We then visualized V-J gene segment pairings between groups. vIL7FL qualitatively showed more combinations of V-J pairings than vCAR or vIL7 ( Figure 5B ). Finally, we assessed T cell epitope spreading of the top 10 clones. While vIL7 groups showed the highest peak frequency, there was no consistency between groups ( Figure 5C ). When assessing the CD3 receptor nucleotide sequence between groups, qualitatively vIL7FL had a dense representation of sequences ( Figure 5D ). We observed similar trends for the TCR alpha chain ( Supplementary Figure 5 ). While not significant, we did observe the potential for vIL7FL to alter T cell diversity.