Study design. Patients with recurrent GBM and other HGGs were enrolled and treated on this phase I study between May 2015 and February 2021 (NCT02208362). Subjects were eligible for enrollment if they were 12 to 75 years of age with progressive/recurrent grade 3 or 4 malignant glioma expressing IL13Rα2 (≥20% 1+) by immunohistochemistry on the most recent tumor resection tissue available prior to enrollment. Life expectancy > 4 weeks, Karnofsky Performance Score (KPS) ≥ 60% with adequate renal, liver and bone marrow function were required. Subjects with pulmonary, cardiac, or neurologic disease were excluded. This study was conducted in accordance with the Institutional Review Board and Independent Ethics Committee at The City of Hope National Medical Center as well as the U.S. Food and Drug Administration (FDA). All subjects provided written informed consent.
Upon enrollment, research subjects were assigned a unique patient number (UPN), and their PBMC were collected at the COH Donor Apheresis Center. They then underwent a stereotactic biopsy or resection followed by placement of a Rickham catheter(s) – those in Arms 1 and 2 had the catheter placed in or proximal to the MRI-defined tumor site (ICT); those in Arm 3 had the catheter placed in the lateral cerebral ventricle (ICV); and those in Arms 4 and 5 had catheters placed both ICT and ICV. Note that the Supplementary Protocol uses Stratum and Arm interchangeably. On the morning of each T cell infusion, the IL13BBζ-CAR T cells were thawed and reformulated in preservative-free normal saline (PFNS)/2% human serum albumin (HSA) and delivered manually using a syringe to inject into the Rickham catheter. This was followed by a 0.5 mL PFNS flush over approximately 5 minutes through the Rickham catheter. Note that CAR T cells were administered without prior lymphodepleting chemotherapy.
Dose escalation within each arm followed a 3+3 design. Participants were deemed evaluable for dose escalation if they received 80% of each of their assigned 3 weekly infusions and were followed for one additional week, or experienced a dose limiting toxicity (DLT) and did not receive disallowed therapy, or experienced a delay between doses of more than 21 days. Adverse events were graded according to the CTCAE v4.0, as well as the revised CRS grading system and the modified neurological grading system. Full details on toxicity evaluation and DLTs are provided in the Supplementary Protocol.
Toxicities reported here are limited to those that occurred on protocol therapy. Toxicities were followed at least twice a week during the 28-day DLT period. After the DLT period participants were followed for worst grade toxicities monthly, then every 3 months for the first year and yearly thereafter. Participants were evaluable for response or survival if they received 3 doses of CAR T cells in the DLT period. UPN230 was not evaluable for survival due to the extended wait between surgery and the first CAR T cell infusion. Four participants were not evaluable for dose escalation due to either receipt of < 80% of the CAR T cell dose at cycle 3 (UPN201) or cycle 2 (UPN131), receipt of disallowed therapy on day 24 (UPN260), or a longer than allowable delay between cycles 1 and 2 (UPN409). Participants were imaged to assess disease response using modified RANO criteria at the end of the DLT period and every 2-3 months thereafter as clinically indicated.
Participants could continue receiving CAR T cell infusions at a rate no more frequent than once a week and at ≤ the highest tolerated cell dose in the initial dose schedule, provided that the participant continued to meet eligibility criteria and there were cell doses available from the already manufactured cell product. If a research participant on Arms 1 or 2 (ICT) progressed after the first three CAR T cell infusion cycles, they were allowed to receive their optional CAR T cell infusions ICV. For research participants on Arms 4 or 5 (dual ICT+ICV), based on clinical response after the first three infusions, optional infusions could occur at either one or both sites (instead of requiring injections at both ICT and ICV sites) at ≤ the highest dose deemed safe for that delivery site.
MRI acquisition and analysis. Magnetic resonance imaging (MRI) of the brain and spine were acquired on a Siemens MAGNETOM Verio 3.0 Tesla scanner. Pre, post, and dynamic contrast enhanced (DCE), and dynamic susceptibility contrast (DSC) gadolinium T1-weighted, diffusion, and T2-weighted sequences were acquired. Tumor foci were measured on axial T1 MPR weighted images obtained after the administration of MultiHance (gadobenate dimeglumine). Response assessment was recorded using the modified RANO criteria for GBM. Imaging with 18F-fluorodeoxyglucose was performed using a GE Discovery DST HP60 PET-CT scanner (70 cm axial field of view, slice thickness 3.75mm). Maximal standardized uptake values (SUVs) were obtained utilizing Vital Images Vitrea version 6.7.2 software. For the calculation of contrast-enhancing tumor volumes, T2-weighted, T2-weighted Fluid Attenuated Inversion Recovery (FLAIR), T1-weighted pre- and post- contrast images were co-registered and resampled to 1x1x3mm voxel sizes using BraTumIA software41. The registered, non-skull-stripped images were then imported into ITK-SNAP (v3.8.0) for segmentation42. Trained readers generated initial masks of the tumor volumes with final review and volume selection by a radiologist with over 10 years of experience in neuroradiology.
Clinical vector and IL13BBζ-CAR T cell manufacturing. The codon optimized CAR sequence contains a membrane-tethered human IL-13 ligand mutated at a single site (E13Y) to reduce potential binding to IL13Rα1 43,44, a human IgG4 Fc spacer containing two mutations (L235E; N297Q) that prevent Fc receptor-mediated recognition 45, a human CD4 transmembrane domain, a human costimulatory 4-1BB cytoplasmic signaling domain, and a human CD3ζ cytoplasmic signaling domain. A T2A ribosome skip sequence 46 then separates this IL13Rα2-targeting CAR sequence from a truncated human CD19 sequence (CD19t), an inert, nonimmunogenic cell surface marker. Details for the generation of the lentiviral vector encoding the IL13Rα2-CAR and the CD19t transgene are available upon request.
For IL13Rα2-CAR T cell manufacturing, on the day of leukapheresis, PBMC were isolated by density gradient centrifugation over Ficoll-Paque (GE Healthcare) followed by two washes in PBS/EDTA. PBMC were then washed once in PBS, resuspended in X Vivo15 media (Bio Whittaker) containing 10% fetal calf serum (FCS) (Hyclone), and stored on a 3-D rotator overnight at room temperature (RT). The following day, PBMC were incubated with clinical grade anti-CD14 and anti-CD25 microbeads with (for Tcm) or without (for Tn/mem) anti-CD45RA microbeads (Miltenyi Biotec). CD14/CD25+ (for Tn/mem) or CD14/CD25/CD45RA+ (for Tcm) cells were then depleted using the CliniMACS™ depletion mode according to the manufacturer’s instructions (Miltenyi Biotec). After centrifugation, the unlabeled negative fraction of cells was resuspended in CliniMACS™ PBS/EDTA buffer (Miltenyi Biotec) containing 0.5% human serum albumin (HSA) (CSL Behring) and then labeled with clinical grade biotinylated-DREG56 mAb (COHNMC CBG). The cells were then washed and resuspended in CliniMACS™ PBS/EDTA containing 0.5% HSA and then incubation with anti-biotin microbeads (Miltenyi Biotec). The CD62L+ fraction was purified with positive selection on CliniMACS™ according to the manufacturer’s instructions, and resuspended in X Vivo15 containing 10% FCS.
Following enrichment, Tcm (CD14/CD25/CD45RA-, CD62L+) or Tn/mem (CD14/CD25-, CD62L+) were stimulated with GMP Dynabeads® Human T expander CD3/CD28 (Invitrogen) at a 1:3 ratio (T cell:bead), and transduced with clinical grade IL13BBζ-T2A-CD19t_epHIV7 at an MOI of 0.3 in X Vivo15 containing 10% FCS with 5μg/mL protamine sulfate (APP Pharmaceutical), 50 U/mL rhIL-2, and 0.5 ng/mL rhIL-15 . Cultures were then maintained at 37°C, 5% CO2 with addition of X-Vivo15 10% FCS as required to keep cell density between 4x105 and 2x106 viable cells/mL, with cytokine supplementation (final concentration of 50 U/mL rhIL-2 and 0.5 ng/mL rhIL-15) three times per week. CD3/CD28 Dynabeads were removed approximately 7 days after transduction using the Dynal ClinEx Vivo Magnetic Particle Concentrator bag magnet. Cultures were propagated until sufficient cell numbers were generated as determined by Guava PCA, at which time cultures were harvested, washed in Isolyte (Braun) with 2% human serum albumin, then resuspended in Cryostor CS5 (BioLife Solutions) for cryopreservation. Overall, this manufacturing process was completed in 10+ days – schema depicted in Extended Fig. 3a. Quality control tests on freshly thawed cells included viability, potency (CD19t expression), identity (CD3 expression), transgene copy number (WPRE qPCR), replication competent virus testing (VSV-G qPCR and formal RCL testing at the University of Indiana), residual bead count, and sterility. Cell products were further analyzed by flow cytometry as described below.
Patient sample processing. Tumor resection material was collected through the COH Department of Pathology according to the clinical protocol. Peripheral blood samples were collected in vacutainer tubes ±EDTA. Samples with EDTA were ficolled immediately upon receipt and peripheral blood mononuclear cells (PBMC) were frozen in Crystor CS5 at -80°C, followed by transfer to liquid nitrogen for long term storage. Samples without EDTA were allowed to coagulate for 2-3 hours at room temperature; serum was collected by centrifugation, aliquoted in single use 100-200 µL aliquots and stored at -80°C. Tumor fluid (TF) was collected from the ICT reservoir, and cerebral spinal fluid (CSF) was collected from the ICV reservoir in a 3cc syringe, spun down, and cell-free supernatants were aliquoted and stored at -80°C. The CSF cells were resuspended in HBSS-/- (Corning CellGro) with 2% FCS and sodium azide for immediate flow cytometric analysis as described below, with the remaining cells resuspended and frozen in Cryostor CS4 at -80°C and then transferred to liquid nitrogen for long term storage.
Immunohistochemistry. IL13Rα2 immunohistochemistry (IHC) was performed on 5 µm-sections of formalin-fixed paraffin-embedded specimens. Slides were loaded on a Ventana DISCOVERY ULTRA IHC automated stainer, where deparaffinization, rehydration, endogenous peroxidase activity inhibition and antigen retrieval (using TRIS buffer pH8) were first performed. The slides were then incubated with a monoclonal rabbit anti-human IL13Rα2 (E7U7B, Cell Signaling Technology, diluted 1:100) for 32 minutes followed by incubation with reagents from the OptiView DAB IHC Detection Kit. The stains were counterstained with hematoxylin and coverslipped. The stained slides were scanned using a NanoZoomer 2.0-HT digital slide scanner, a NanoZoomer S360 Digital Slide Scanner (Hamamatsu Corporation), or directly acquired on an Olympus BX46 transmitted light microscope with an SC-180 Olympus camera. IL13Rα2 immunoreactivity was scored by a clinical neuropathologist and quantified based on the percentage of tumor cells exhibiting weak (1+), moderate (2+), or strong (3+) intensity of cytoplasmic and golgi-like staining. The H score is obtained by the formula: (3 x percentage of strongly staining cells) + (2 x percentage of moderately staining cells) + (percentage of weakly staining cells), giving a range of 0 to 300 (modified from 47). The H score can be translated into the intensity scoring system described in the enrollment criteria as follows: 0 representing negative (H score 0), 1+ low (H score 1-100), 2+ moderate (H score 101-200) and 3+ high (H score 201-300). Appropriate positive (testicular) and negative (prostate) controls were employed for IL-13Rα2 IHC staining. While the criteria for patient inclusion on the trial was at least 20% of the cells scoring 1+ staining intensity (an H score of 20) at the time of enrollment, Fig. 1c is reporting highest H score from multiple blocks at time of surgery, with the "+" sign reflecting the presence of membranous staining. This test was performed at the Department of Pathology, City of Hope National Medical Center and is regarded as investigational for research. This Laboratory is certified under the Clinical Laboratory Improvement Amendments of 1988 (CLIA) as qualified to perform high complexity clinical laboratory testing.
CD3 IHC was performed similar to IL13Rα2 IHC with the exceptions that slides were loaded on a Leica BOND autostainer, and the BOND™ Ready-To-Use Primary Antibody CD3 (LN10) was used according to manufacturer recommendations. Qualitative scoring for CD3 was performed by a clinical neuropathologist as follows: tumors with few detectable T cells were given a CD3 IHC score of 1; those with scattered CD3+ cells throughout, with possible occasional perivascular aggregates or hot spots were given a score of 2; those with many CD3+ cells throughout and dense hot spots were given a score of 3; and those with many dense CD3+ cell infiltrates were given a score of 4.
Flow cytometry. Cells were washed and immunophenotyped by flow cytometry using fluorochrome-conjugated antibodies specific for either CCR7, CD3, CD4, CD8, CD19, CD27, CD45RA, CD57, CD62L, LAG-3, or PD-1, reference Supplementary Reporting Summary. Gating strategies are depicted in Supplementary Fig. 1. For flow-cytometry based recursive killing assays, patient CAR T cell products were cocultured with patient-derived glioma tumor cells and absolute numbers of viable tumor cells and CAR T cells were assessed as previously described 48. All samples were acquired on MACSQuant Analyzer 10 (Miltenyi Biotec) and analyzed with FlowJo software (v10.1, TreeStar) and GraphPad Prism Software (v9).
Cytokine profiling. Serum, CSF and TF were collected and analyzed by cytokine bead array as previously described 6. Assays were performed by the Clinical Immunobiology Correlative Studies Laboratory (CICSL) at City of Hope using the Human Cytokine 30-Plex Panel kit (Invitrogen) and a FLEXMAP 3D® (Luminex).
qPCR for CAR T cell persistence. Assessment of CAR T cell persistence in peripheral blood was determined by quantification of the WPRE region of the lentiviral transgene by qPCR. gDNA was extracted from frozen 0.3 mL aliquots of whole blood and tested for the WPRE copy number by TaqMan qPCR. Average copy numbers are presented if ≥2 of 3 replicates generated a cycle threshold (Ct) value. Participants were measured for WPRE before and at least once after every CAR T cell infusion.
Orthotopic xenograft models
Raji-ffluc, and patient-derived glioma line (PBT030-2-ffluc-IL13Rα2+) were maintained as previously described 16,49. All tumor lines were authenticated for the desired antigen/marker expression by flow cytometry, tested for mycoplasma using the MycoAlert™ PLUS Mycoplasma Detection Kit (Lonza), and maintained in culture for less than 1-2 months. Tcm or Tn/mem products were enriched from healthy donors and transduced with lentiviral vectors encoding the CD19- or IL13Rα2-targeting CAR as described above and previously described 16,50,51. The resulting CAR T cells were cultured for 18-21 days and analyzed by flow cytometry as described above.
All animal studies were approved by the City of Hope Institutional Animal Care and Use Committee (IACUC). For the lymphoma studies, NOD/Scid IL2RγCnull (NSG) mice (9-10 week-old) were injected with 0.5×106 Raji-ffluc cells (CD19+) intravenously (i.v.) on day 0. Three days after tumor inoculation, mice were treated i.v. with Tcm- or Tn/mem-derived CD19-CAR T cells or mock-transduced controls (1×106). For GBM studies, NSG mice (6-10 week-old) were injected with 0.1×106 patient-derived glioma cell line (IL13Rα2+) intracranially (i.c.) as described before 16. Tcm- or Tn/mem-derived IL13BBζ-CAR T cells or mock-transduced controls (0.1×106 cells) were administered intratumorally on day 8. Overall survival was assessed using Kaplan-Meier methods (GraphPad Prism Software v9).
Single cell RNA sequencing
RNA Library preparation and single-cell sequencing: Single-cell sequencing of cryopreserved samples was carried out using the 10x Chromium platform. Single-cell RNA sequencing (sc-RNAseq) was carried out on excess available CAR T cell product samples from 62 of the 65 treated patients (40 Tcm, 22 Tn/mem). Additionally, expression levels of cell surface proteins of 27 CAR T product samples (14 Tcm, 13 Tn/mem) were quantified using CITEseq.
For Batch 1, 51 patient PBMC exomes were collected via Illumina exome panel and sequenced at 20M read pairs per patient for sample deconvolution. For batches 2-19, Biolegend Total Seq-C hashtag antibodies were used to allow sample deconvolution after pooled processing. 8 barcoded samples were sorted at equal proportions into a single collection tube and the 8 batches (batches 2-6, 14, 17, 18) were treated with Biolegend Total Seq-C Human Universal Cocktail V1.0 for cell surface protein expression. Batch 1 was treated with TS-C #99328 with 198 antibodies, batches 2–6 with TS-C #99814 with 192 antibodies, and batches 14, 17, and 18 with TS-C #399905 with 137 antibodies (Supplementary Table 3). 60,000 cells were loaded to a single Gel Bead-in-Emulsion (GEM) reaction onto the Chromium instrument. Sc-RNA-seq and feature barcode libraries were prepared according to manufacturer protocols and sequenced on Illumina Iseq100 for cell count validation and NovaSeq6000 at the recommended depth for relevant library type. Sequence data were processed using 10x Genomics Cell Ranger V5.0 and Ensemble 98.
sc-RNA Seq bioinformatics: Single-cell sequencing data were analyzed using Seurat v4 52. CellRanger objects for each batch were imported to create a Seurat object for each of the 19 batches. For Batch 1, sample identities were deconvoluted and multiplets were identified using Demuxlet 53. Exome sequencing FASTQ reads were chunked to 40M reads, aligned to GRCh38 using BWA 54, and processed with Samtools Fixmates and Samtools Sort 55. Individual chunks were merged, and PCR and optical duplicates were marked with Samtools. Genotypes were called with DeepVariant (https://github.com/google/deepvariant). The single-cell data were filtered to retain singlets with >500 unique RNA features detected, >1,000 RNA feature counts, and less than 10% of reads mapping to mitochondrial genes. Outliers with >10,000 protein feature counts were excluded.
Batches with both sc-RNAseq and CITEseq gene expression data were normalized with SCT, protein data log-transformed, and each data type was integrated using rPCA. The number of significant principal components (PCs) to be included in dimensionality reduction and unsupervised clustering were determined by calculating the difference between the proportion of variation associated with each PC and their subsequent PC and selecting the last point where the difference is more than 0.1%. For each data type, Uniform Manifold Approximation and Projection (UMAP) was used for dimensionality reduction and 2D visualization of cell clusters. After clustering based on surface protein expression, data were filtered to exclude cells expressing high levels of control antibodies. Clustering based on both gene expression and surface protein expression levels was carried out using the Weighted Nearest Neighbor (WNN) analysis. For the batches with only scRNAseq, gene expression data were similarly SCT normalized and rPCA integrated. After dimensionality reduction with PCA, cluster labels and protein data were transferred from the reference object (RNAseq+CITEseq) to the query (only RNAseq) by projecting the query data onto the UMAP structure of the reference. The two objects were merged using the WNN UMAP. The SCT assay was replaced with data integrated across all 62 samples. Three clusters with low numbers of cells (353, 257, and 2) were excluded from downstream analyses.
Highly expressed marker features for each cluster were identified using the presto implementation of the Wilcoxon rank test (https://github.com/immunogenomics/presto). Clusters were annotated based on cluster markers and canonical T cell marker expression. To further annotate the clusters based on the expression of previously defined T cell expression signatures 56-58 (reference Supplementary Table 4) we conducted a single-cell gene-set variation analysis (scGSVA) using the scGSVA R package (https://github.com/guokai8/scGSVA).
Statistical analysis
Survival calculations and estimates were generally performed using Kaplan-Meier methods, with post hoc tests of survival curve differences being from the Harrington-Fleming G\rho family as indicated in figure legends. An exception is the estimate of effects of CD3 score and product on expected survival times for rGBM patients (fig. 5f), which was performed using a linear model for log survival time with covariates of CD3 score (high vs. low), product (Tcm vs. Tn/mem) and interaction. Within the rGBM group, there is only one censored observation, which was accounted for by imputation, noting that in general log-survival times amongst the rGBM group are well-modelled through a normal distribution. Robustness of the imputation results were checked by observing that this single censored observation would need to be an extreme outlier before it would affect the significance of the 95% CIs obtained.
Worst grade toxicities are tabulated by CAR T product (combined Tcm Arms 1-4 vs. Tn/mem Arm 5) and presented based on attribution (CAR T or surgery and rickham catheter). Baseline characteristics are presented overall and by CAR T product. QLQ-C30 was used to assess QOL., and descriptive statistics were estimated for baseline score and slopes were determined over a 38-day period which included the 3 infusions in the DLT period. Post hoc comparisons were made between the Tcm Arms 1-4 and Tn/mem Arm 5 for QOL using a two-sample t-test. No corrections were made for multiple testing.
Area under the curve of CAR T cells in CSF was calculated using linear interpolation between time points, with the linear x-axis scale in units of days.
Low dimensional representation of the 30-dimensional cytokine data was obtained by the application of multidimensional scaling (MDS) to the proximity matrix generated by a random forest trained on the data.
Density plots were constructed using kernel density estimators with guassian kernels and bandwidth estimated through cross-validation.
Statistical significance tests between groups with a single factor were generally performed using two-sided t-tests after appropriate checks on normality assumptions. No corrections were made for multiple testing. Significance tests between mutifactorial groups (cytokine changes in the CSF over both cycles and arms, fig. 4d) were performed with ANOVA to obtain significance levels for the factor of interest (arm).
Shown linear regressions were performed with standard least squares methods and significance of associations found by standard t-test.
We defined the absolute IFNγ pathway score of available C1D1 CSF samples (Fig.s 4e,f, 6d,e) as the sum of the log-transformed measured values of IFNγ, CXCL9 and CXCL10. Fig 4d shows the fold change in this score from baseline (C1D0).