Investigation of product-derived lymphoma following infusion of piggyBac-modified CD19 chimeric antigen receptor T cells

Chimeric antigen receptor (CAR) T cells have shown remarkable efficacy in relapsed and refractory B-cell malignancy.^1,2 The majority of CAR T-cell products in use have been manufactured using retroviral and lentiviral vectors. Genetic modification carries an inherent risk of mutagenesis as demonstrated by development of leukemia in children receiving genetically modified hemopoietic stem cells for severe combined immunodeficiency.^3,4 Although there have been examples of monoclonal CAR T-cell proliferation in patients receiving CD19- and CD22-specific T cells,^5,6 these have not been detrimental to patients, and T cells appear to have a low risk of malignant transformation.⁷ 

Because of the expense and limited transgene capacity of viral vectors, nonviral gene-modification systems are being explored as alternatives for CAR T-cell production.^8-11 We previously devised a simple CAR T-cell production protocol using the piggyBac transposon system^12,13 and have seen robust clinical efficacy in patients receiving donor-derived CD19-specific CAR T cells for relapsed and refractory B-cell malignancy post–allogeneic stem cell transplant in a phase 1 clinical trial (the CARTELL study; see accompanying article by Bishop et al¹⁴).

Here, we identify 2 cases of malignant transformation of piggyBac transposon–manufactured CD19-specific CAR T cells occurring after infusion. The investigations demonstrate transgene-mediated transcriptional readthrough and other genetic changes including point mutations, background genomic copy-number variations, and a high transgene copy number per cell. We did not identify insertion of the transgene into a known oncogene^3-6 and conclude that multiple genetic alterations led to malignant transformation in modified T cells. These results highlight the need for caution in implementing new gene-modification technologies for the manufacture of immune and other therapies.

CAR T cells specific for CD19 (CAR19 T cells) were produced from the HLA-matched sibling donors of patients 2 and 8 using the piggyBac transposon system for gene modification as previously described (supplemental Methods, available on the Blood Web site).^12,13

The CAR T cells were administered in the context of the CARTELL study, registered with the Australian New Zealand Clinical Trials Registry (ACTRN12617001579381), and approved by the local human research ethics committee and the Australian Therapeutic Goods Administration (see accompanying article by Bishop et al¹⁴). The CAR T-cell products met all release criteria including no autonomous proliferation in vitro using an assay previously described.¹⁵ 

Immunophenotypic and T-cell activation pathway phosphoflow analysis of CAR T cells was performed on FACSCanto or Fortessa flow cytometers (BD Biosciences) and analyzed using FCS Express 6 (De Novo Software, Pasadena, CA) (supplemental Methods).

Cells were stained with a cocktail of antibody-oligonucleotide conjugates (BD Biosciences) (supplemental Table 1) and loaded onto a 10x Chromium.

An antibody-oligonucleotide conjugate–specific PCR1 primer was used to amplify the AbSeq library, and was sequenced on NextSeq500 or NovaSeq6000 platforms (Illumina, San Diego, CA); the Cellranger pipeline was applied to obtain the protein expression (AbSeq) matrix (supplemental Methods).

CD4⁺, CD8⁺ and CAR⁺ T cells were selected by flow cytometry sorting on the FACSAria III sorter (BD Biosciences) (supplemental Methods).

For details regarding extraction of DNA and RNA from blood and tissue samples, see supplemental Methods.

T-cell receptor (TCR) γ and β (TCRG and TRB) clonality was assessed using the Invivoscribe Identiclone TCRG Gene Clonality Assay and LymphoTrack TRB, respectively (Invivoscribe, San Diego, CA) (supplemental Methods).

Extracted DNA was analyzed for the presence of mutations in a panel of genes commonly mutated in hematological malignancies as previously described.¹⁶ Genomic copy-number analysis was performed using on- and off-target reads from this hybridization-based next-generation sequencing (NGS) panel (supplemental Methods).^16,17

Integration site analysis in CAR T cells was conducted using muA-mediated integration site recovery as previously described.^18,19

CAR transgene copies per cell were assessed by droplet digital PCR using primer/probes specific for the junction of CD28 transmembrane and 4-1BB intracellular sequences and the Bio-Rad Human RPP30 control gene (Hercules, CA) (supplemental Methods).

Whole-genome sequencing was performed by the Australian Genome Research Facility (AGRF) on the NovaSeq 6000 platform (Illumina) to a minimum of 30-fold base coverage (30X) per sample.

CAR19 activity was detected by aligning reads to a modified human reference (GRCh38 with CAR19 as an additional contig) using Sentieon²⁰ with default settings. Structural variants (SVs) were called using the SURVIVOR pipeline,^21,22 which measured consensus among Delly (v0.8.2),²³ Lumpy (v0.2.13),²⁴ and Manta (v1.6.0)²⁵; SVs were annotated with annotatePeaks.pl from HOMER²⁶ using the hg38 build (supplemental Methods).

RNA-sequencing libraries were sequenced by AGRF at 1 × 100 bp single-end reads on an Illumina NovaSeq 6000 SP 100 lane.

Reads were aligned to a custom genome (hg38 and CART insert) using STAR²⁷ with RefSeq annotations. Alignments and transcriptional shadows were visualized using IGV (v2.72).²⁸ Genome-position analyses were performed using genomic ranges²⁹ based on RefSeq annotations imported with “rtracklayer”³⁰ and gene lists were annotated with Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (adjusted P value <.05) using the clusterProfiler package.³¹ 

Analyses were performed in the R statistical environment (http://www.r-project.org) with tidyverse³² (supplemental Methods).

Patient 2 was a 67-year-old man with diffuse large B-cell lymphoma (DLBCL) transformed from follicular lymphoma resistant to multiple lines of chemotherapy, an autologous hemopoietic stem cell transplant (HSCT; December 2014), an HLA-matched sibling allogenic HSCT (November 2015), donor lymphocyte infusions (June, August, November 2016), and radiotherapy. Progressive DLBCL involving the left lobe of the liver, external iliac nodes, iliacus muscle, and paraspinal muscles was diagnosed in June 2017. He received 3 doses of CAR19 T cells (2 × 10⁷ total nucleated cells [TNCs] per m², 5 × 10⁷ TNCs per m², 10 × 10⁷ TNCs per m²) produced from his HLA-matched sibling donor administered after lymphocyte-depleting fludarabine and cyclophosphamide (first infusion) or cyclophosphamide alone (second and third infusions) in January, March, and May 2018. At the time of the first dose of CAR T cells, persistent disease on positron emission tomography (PET) was present in the lumbar paraspinal muscles and left iliac nodes. The left para-aortic nodal lesion remained the only PET⁺ area on repeated scans, gradually increasing in size over 12 months (Figure 1B). The patient had ongoing B-cell aplasia and stable, low-level CAR T cells detectable in his peripheral blood by flow cytometry (Figure 1A). Human herpes virus 6 and cytomegalovirus were repeatedly detected in the patient’s peripheral blood, and BK virus was detectable in the urine and in the peripheral blood to a level of 82c000 copies per mL (Figure 1A).

A computed tomography–guided core biopsy 16 months after the first dose of CAR T cells (tumor size, 13 cm × 8 cm) showed fibroadipose tissue infiltrated by intermediate to large CD3⁺ lymphocytes. TCRG and TRB complementarity-determining region 3 sequencing confirmed monoclonality with a TRVB2-restricted complementarity-determining region 3 sequence (ASSTQGSGNTIY) not matching known sequences.³³ Donor-recipient chimerism by microsatellite marker analysis confirmed donor origin.

A repeat laparoscopic biopsy confirmed CAR expression on the monoclonal T cells (Figure 1C-D). Cyclophosphamide, vincristine, and prednisolone chemotherapy was administered leading to transient reduction in tumor size, but was complicated by delirium, sepsis, and acute renal and liver failure leading to death 18 months after CAR T cells. The patient’s family declined an autopsy.

Multiparameter AbSeq immunophenotypic analysis was performed on malignant T cells and compared with the CAR T-cell product (Figure 2A). This showed that the pure CD4⁺CAR⁺ T cells with variable CD3 expression were negative for markers of B cells, macrophages, monocytes, dendritic cells, natural killer (NK) cells, and γ-δ T cells (Figure 2B), expressing markers of terminally differentiated effector T-helper type 1 subtype T cells (Figure 2C-D). Cells were positive for activation markers CD38 and CD56 but negative for CD25 and CD137. Cells were positive for T-cell immunoglobulin and mucin domain–containing protein 3 but negative or low for other coinhibitory molecules including programmed death ligand 1, programmed cell death protein 1, and lymphocyte activation gene 3 (Figure 2E).

To determine whether the malignant CAR T-cell proliferation could have been driven by CAR engagement with persistent B-cell malignancy in the patient, we assessed their expansion in response to CD19 antigen and compared this to nonmalignant CAR T cells isolated from the peripheral blood taken at the time of biopsy. Irradiated healthy donor peripheral blood mononuclear cells (PBMCs) were used as a source of CD19 as per original CAR T-cell production.^12,13 Peripheral blood CAR T cells expanded 731-fold over 4 weeks, compared with only threefold expansion of tumor CAR T cells, with reduced viability compared with blood CAR T cells (Figure 3A).

Expanded CD3⁻CD4⁺ CAR T cells from the biopsy expressed high levels of CAR compared with those from the peripheral blood (mean fluorescence intensity, 17c774 in tumor and 771 from blood CAR T cells) (Figure 3B), raising the possibility of tonic CAR signaling. However, there was no increase in phosphorylated CD3ζ, ζ-associated protein of 70 kDa (ZAP70), and AKT³⁴ in unstimulated, directly isolated malignant CAR T cells compared with CD4⁺ CAR T cells from the product or untransduced sibling donor T cells (Figure 3C-D).

To determine whether integration of the transgene into known oncogenes may have led to malignant transformation, insertion site analysis was performed on malignant CAR T cells, the CAR T-cell product, and 7 other products manufactured using the same methodology. As the malignant CAR T cells were monoclonal, whole-genome sequencing was used to verify insertion sites.

The malignancy contained inserts within 2.5 kb of transcriptional start sites and within genes themselves on 3 and 11 occasions, respectively, but no insertions into exons or into traditional oncogenes in the Catalogue Of Somatic Mutations In Cancer (COSMIC) database (Table 1).³⁵ 

The slight enrichment of oncogenes in the product of patient 2 compared with a statistically random distribution throughout the genome (Figure 4A-B) was similar to previous data comparing piggyBac to retroviral and lentiviral vector insertion profiles^36-38 (Figure 4B). None of these traditional oncogenes were seen in the malignant CAR T cells.

Transgene copy number per cell in the malignancy was higher than expected at 24, but was not higher than nonmalignant CD4⁺ CAR T cells from the product (29.8) or peripheral blood (27.6).

The landscape of SVs and their intersections were characterized within genomic DNA extracted from untransduced CD4⁺ donor T cells, and CD4⁺ CAR T cells from the product administered to patient 2, his peripheral blood and malignant CAR T cells (Figure 5A). The largest set of events was observed in all samples, representing 34% of SVs called (2856 of 8366). Each sample contained several hundred unique SVs, and their types, lengths, and genomic annotations reveal general consistency, with no enrichment of any class of variant in the malignant cells (Figure 5B).

NGS targeting genes recurrently mutated in hematological malignancies was performed on DNA extracted from malignant CAR T cells, CAR T cells expanded from the patient’s peripheral blood, the product, and untransduced donor T cells as previously described.¹⁶ An acquired nonsense mutation in PIGA (Glu264*) in malignant CAR T cells alone was detected. In addition, numerous copy-number changes were detected in malignant CAR T cells including copy-number gains involving chromosomal regions on 1q, 4q, 5, 6, 10q, 11q, and 17q and copy-number losses involving regions on 4q and 17p.

To determine whether the insertion sites or somatic mutations identified could be directly related to the development of the CAR T-cell malignancy, examination for global, insertion, and SV-related transcription changes was performed using whole-transcriptome analysis. Directly isolated malignant CAR T cells were compared with flow cytometry–sorted CD4⁺ CAR T cells expanded from peripheral blood from patient 2, the product, and untransduced CD4⁺ T cells from the peripheral blood of 3 healthy donors including the sibling donor of patient 2.

Multidimensional scaling of gene-expression profiles showed distinct clusters of the untransduced donor T cells, and the nonmalignant CAR T cells, and marked global differences with the malignant CAR T cells.

Genes with fourfold altered expression in malignant T cells compared with donor CD4⁺ CAR T cells from the product (584 genes) or untransduced healthy donor CD4⁺ T cells (1414 genes) and an additional 915 genes with fourfold differential expression in donor CD4⁺ CAR T cells compared with untransduced healthy donor CD4⁺ T cells were further analyzed (Figure 6A; supplemental Excel file). Gene ontologies enriched in this 3-way comparison showed distinct clusters of upregulated genes in the CAR T-cell malignancy (Figure 6A-B; supplemental Excel file) associated with primitive embryonic development and cell-cell adhesion (Figure 6B,F), whereas downregulated genes were associated with regulation of cell adhesion, T-cell activation, and activation of phospholipase C (Figure 6B).

Alterations in transcription were analyzed to determine whether the CAR insertions had directly impacted expression of surrounding genes and how that might relate to global changes in transcription. Despite the presence of insulator sequences around the CAR transgene, RNA levels surrounding insertion sites showed transcriptional readthrough extending from the 3′ end of the transgene affecting the forward or reverse strand depending on the direction of insertion (Figure 6C-D). This “transcriptional shadow” extended for up to 1000 kb including noncoding intronic and out-of-frame sequences. Analysis of levels of in-frame, exonic sequences showed increased expression of 4 genes: FAM110D, COL8A1, HIVEP1, and FYN (Table 1). Of these, HIVEP1 and FYN could be assessed for direct impact on the transcriptional changes in the malignant cells. Genes that have the HIVEP1-binding motif within a 3-kb promoter window were not overexpressed in the malignant cells (Wilcoxon rank-sum P > .05) compared with other samples (binomial test > 0.05) (Figure 6E). Pathway analysis confirmed changes in primitive development clusters (Figure 6F) but did not reveal changes in gene expression linked to FYN except for the downregulation of the TCR activation pathway (as seen in KEGG; supplemental Figure 1).

Genome-wide expression changes between malignant CAR T cells and product were associated with copy-number gains and losses (Wilcoxon rank-sum P < 1e-16; Figure 7). Furthermore, overexpressed (more than fourfold) genes in the malignant cells compared with product or untransduced cells were more than twofold enriched at regions with increased copy number (odds ratio = 2.19; P < .0001). In contrast, genomic regional clusters of differentially expressed genes did not correspond to any insertion site.

Seventy-two of the SVs detected on whole-genome sequencing (10%) that were unique to the malignant CAR T cells were near differentially expressed genes; however, all but 2 of these were intergenic, repeats, or intronic. A 1.3-kbp inversion involving the promoter of IFITM2 and a 76-kbp deletion of the transcriptional termination region of LRRN3 were associated with profound downregulation (−8 log-fold change and −12 log-fold change, respectively) but neither were near an insertion site and neither genes are associated with lymphoma.

This indicates that the transcriptional impact of transgene insertion and (in the majority of cases) SVs did not contribute significantly to the changes in genome-wide differential expression observed in malignant cells. The dominant driver of this differential expression is most likely due to copy-number variation and this potentially explains the disconnect between insertion sites, SVs, and global gene expression, obscuring that of potential insertion or mutation-related changes.

All trial patients were notified and screened for lymphoma by PET scan. Patient 8, a 31-year-old man, was found to have an asymptomatic mediastinal tumor and cervical lymphadenopathy 12 months post-CAR T cells. A biopsy revealed a CD3⁻, CD2⁺, CD8⁺ monoclonal CAR T-cell infiltrate. The tumor was chemotherapy responsive, and the patient is in remission after a second allogeneic HSCT.

Experiments were repeated on tumor samples from patient 8, and showed a smaller number of integrants (4), with similar readthrough as seen in the first patient, but lack of increased in silico expression of any gene (supplemental Table 2; supplemental Figure 4). There was transgene-mediated increased expression of a noncoding RNA on chromosome 1 (LOC107985043) of uncertain function and significance. In both malignancies, there was insertion into an intron of the BACH 2 locus on chromosome 6 (intron 3 and 4 in patients 2 and 8, respectively). Unlike in patient 2, whole-genome sequencing revealed a marked increase in SVs compared with untransduced and nonmalignant CAR T cells, however, like patient 2, there was no enrichment of any particular class of variant (supplemental Figure 5). There was reduced expression of clusters of genes associated with T-cell activation and increased expression of clusters of genes associated with cell division, extracellular matrix and motility. However, there was no enrichment for genes controlling embryonic development (supplemental Figure 4). Overall changes in transcription once again correlated most strongly with alteration in genomic copy-number variation (supplemental Figure 4).

We identified monoclonal CAR T-cell malignancies in 2 patients, leading to the death of 1 patient receiving CAR19 T cells manufactured from healthy HSCT donors using the piggyBac transposon system. Although the clinical findings bear a superficial analogy to T-cell malignancies seen in early gene therapy trials with retroviral vectors,^3,4 investigations show no integration of the transgene into known oncogenes and a unifying underlying pathogenic mechanism could not be definitively identified. The malignant cells failed to expand in vitro even after stimulation with CD19 antigen, suggesting that signaling through increased CAR19 copies was not responsible for the cell proliferation. We cannot exclude responsiveness to other antigens such as alloantigens or infectious antigens that may have been present only in vivo. No clear oncogenic driver sequence variant abnormalities were detected using targeted NGS; however, widespread copy-number gains and losses were observed involving oncogenes and tumor-suppressor genes, respectively. These copy-number variations were not detected with whole-genome sequencing due to the insensitivity of SV callers to interspersed duplications.^39-41 

Transgene promoter activity caused increased transcription of surrounding genes, which was not expected as the cassette includes 5′ and 3′ insulator sequences.⁴² The transcriptional shadow identified enhanced expression of 4 genes in patient 2, of which one, FYN, is an Src family tyrosine kinase involved in cell growth, proliferation, and adhesion. FYN is a proximal component of TCR-mediated T-cell activation⁴³ and activating mutations have been seen in adult T-cell leukemia and lymphoma.⁴⁴ None of the other 3 genes, FAM110D, COL8A1, or HIVEP1, have been associated with hematological malignancies, although COL8A1 expression has been seen with progression of solid tumors.^45,46 Although overexpression of FYN is a prime candidate for the initial driver of malignant transformation in patient 2, KEGG analysis of FYN and T-cell receptor activation pathways showed neutral or downregulation of downstream mediators of T-cell activation (supplemental Figure 1) and phosphorylation of CD3z was not increased in the tumor (Figure 3) as would be expected with increased FYN activity. The initiating event may possibly have been FYN overexpression with consequent downregulation of other mediators of activation to maintain cellular homeostasis, with subsequent mutational changes such as copy-number variation occurring due to FYN-driven proliferation; but FYN also plays a role in programmed cell death protein 1–mediated T-cell inhibition,⁴⁷ reducing this as a likely mechanism. Transgene integration into introns and reduced expression of BACH2 occurred in both malignancies, making this another gene for further consideration. BACH2 is a transcription factor that promotes regulatory T-cell development and suppresses T-cell activation and differentiation.⁴⁸ Although not expected to interrupt gene expression, the intronic insertions may have led to reduced BACH2 expression, permitting persistent, excessive activation and proliferation with subsequent malignant changes. However, BACH2 expression is physiologically reduced with differentiation of T cells consistent with the levels seen in both tumors; these alterations may therefore be secondary. Overexpression of BACH2 in DLBCL is associated with poor prognosis,^49,50 but changes in expression have not been described in the development of T-cell malignancies, including those associated with retroviral gene therapy. This is despite BACH2 being a recurrent insertion site for retroviruses,^51-53 suggesting that BACH2 integration is unlikely to be playing a role in malignant behavior.

Until now, mature T cells transduced with viral vectors have been resistant to mutagenesis.^7,54 No cases of secondary malignancies have been reported in hundreds of patients who have received virally transduced CAR T cells for treatment of HIV⁵⁵ or B-cell malignancies over the last 2 decades. Despite transgene-mediated dysregulation of gene expression, long-term engrafted retrovirally gene-marked T cells have stable genotype, phenotype, and function without malignant behavior.⁵⁶ Two prior reports have described clonal expansion in patients receiving CAR T cells produced using lentiviral vectors. In 1 case, clonal expansion was associated with insertion into the TET2 gene, whereas in the other there was a single transgene insertion into the CBL gene. In contrast to our cases, neither clonal population manifested clinically malignant behavior, which in the former case may have been therapeutically beneficial. Examination of clonality of lentiviral CAR19 T cells administered to 10 patients performed by Sheih et al has shown that dominance of oligoclonal populations is common, but independent of particular insertion sites.⁵⁷ 

Although the development of a T-cell malignancy in these 2 patients raises safety concerns, our findings cannot necessarily be extrapolated to all production methodologies utilizing the piggyBac or other transposon systems. Using different production processes, other groups have not seen any suggestion of malignant transformation with piggyBac CAR T cells in preclinical work^{10,12,13,36,37,58-60} and up to 2 years of follow-up in patients receiving a B-cell maturation antigen–specific CAR T cell for multiple myeloma.^61,62

The occurrence of 2 CAR T-cell malignancies suggests that our production methodology is the predominant factor in malignant transformation. Unique aspects of our method include the electroporation device and settings (with a single high-voltage pulse), concentration of transposon and transposase, and expansion with irradiated PBMCs and interleukin 15 (IL-15; described in supplemental Methods). Studies aimed at determining whether the high-electroporation voltages or high nuclear levels of transposase in particular could mediate nonspecific DNA damage will be especially important. Preclinical assessment of safety relied on data demonstrating a favorable integration profile compared with other transposon and viral vectors, normal cytogenetics, and no antigen-independent proliferation over several months of culture in vitro or in vivo.¹³ We were also reassured by others’ data^36,59 and by mature T cells appearing to be resistant to transformation. We did not specifically assess readthrough in our preclinical studies, using the EF1a promoter seen to be safe in lentiviral CAR T-cell trials even in the absence of insulators^1,63,64 and the previously well-characterized core CHS4 insulator.^65-67 Although we hypothesize that the transgene-mediated transcriptional readthrough played a role in the pathogenesis of the CAR T-cell malignancies, readthrough has been described as a secondary feature of de novo malignancy⁶⁸ and further studies are planned including analysis of single-cell clones to assess its frequency in nonmalignant piggyBac CAR T cells.

We will also attempt to recapitulate malignant transformation with potential mechanisms identified, such as overexpression of FYN or targeted integration of a transgene into BACH2 introns. Similar methods will be used to exclude alternative splicing of genes surrounding insertion sites related to a cryptic splice site in the CHS4 insulator as has previously been described with lentiviral vectors, but not demonstrated here.⁶⁶ Given the high transgene copy number in 1 of the 2 malignancies, its contribution to transformation is uncertain. However, ensuring a copy number of 3 to 5 per transduced cell for future studies will reduce this as a risk. As we have observed that transposon-based CAR T cells with less than 5 copies per cell display robust tumor-specific function in vitro and in vivo (G.S., D.J.G., K.G., K.P.M., unpublished data), manipulations to limit transduction efficiency would not be expected to impair efficacy.

Although the mechanism underlying the malignant behavior of the CAR T-cell clone is not fully established, as investigators we felt that releasing these preliminary data was important, to act as a caution for others exploring the translation of new gene-modification methodologies such as piggyBac for CAR T-cell production. The overarching reason for embarking on this work was the need to produce CAR T cells affordably and with capacity for more complex genetic circuits, which remains compelling. Hopefully, once a complete understanding of the mechanisms that have led to this malignancy has been reached and addressed, work using this system can cautiously resume.

This work was supported by the National Health and Medical Research Council of Australia (project grants 1102172 and 1121704) and by a translational program grant from Cancer Council NSW and the Cancer Institute NSW (TPG 19-01). E.B. is a NSW Cancer Institute Early Career Fellow. C.E.M. was supported by the Starr Cancer Consortium (I13-0052), the Pershing Square Sohn Cancer Research Alliance, the National Institutes of Health, National Cancer Institute (R01CA249054, P01CA214274), and Leukemia & Lymphoma Society (LLS) grants MCL7001-18, LLS 9238-16, and LLS-MCL7001-18.

Reagents for AbSeq immunophenotypic analysis were provided by BD Biosciences free of charge. BD Biosciences did not provide any restrictions on this work and did not have a role in the conceptualization, design, data collection, analysis, decision to publish, or preparation of the manuscript.

Contribution: K.P.M. designed and oversaw the trial and experiments and wrote the manuscript; K.G., Z.L., J.A.S., and L.M. performed experiments and data analysis including cell culture, transcriptome, and genome sequencing; B.S.G. performed insertion site, transcriptome, and NGS analysis; M.A.M. assisted with experimental work and produced the figures; G.S. performed the insertion site data acquisition and analysis; L.E.C. oversaw the production of CAR T cells; D.C.B. assisted with patient care and trial oversight; R.H.Y.L., C.C., and F.L. performed the AbSeq analysis; J.F., M.M., F.J.S., and C.E.M. performed the whole-genome-sequencing analysis; P.B. performed the TCR clonality and NGS analysis; and D.J.G. and E.B. designed and oversaw the trial and experiments.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Kenneth P. Micklethwaite, Department of Haematology, Westmead Hospital, Hawkesbury Rd, Westmead, NSW 2145, Australia; e-mail: kenneth.micklethwaite@sydney.edu.au.