A clonally expanded nodal T-cell population diagnosed as T-cell lymphoma after CAR-T therapy

Clinical cases of possible post-CAR-T lymphoma

We systematically assessed our clinical database of patients with non-Hodgkin lymphoma (NHL), MM, and B-cell ALL receiving one of the six FDA-approved CAR-T products to quantify the incidence of secondary malignancy at our center from 2017 to 2023 with a focus on possible T-cell lymphomas. From 626 total patients, we identified 43 cases of secondary malignancy (6.9%, n = 12 myelodysplastic syndrome [MDS]/acute myeloid leukemia [AML], n = 28 solid tumors, n = 3 possible TCL); six cases were in MM patients and 37 in NHL (Fig. 1). Of the possible TCLs, one was a bona-fide case bearing the CAR transgene after Cilta-cel, as described above¹³. Of the remaining two cases, one occurred in a patient treated with Cilta-cel for MM who presented with an 8 mm facial lesion approximately 6 months after CAR-T. Biopsy revealed a population of lymphocytes positive for CD2, CD5, CD3, CD8, MUM-1, TCR-BF1, Granzyme B, and Perforin with Ki-67 of 100%, and negative for CD20, CD138, CD30, CD4, EBV, and TCR-Delta. This was diagnosed as Stage I peripheral TCL. Remarkably, the lesion spontaneously resolved after biopsy and the patient received no TCL therapy.

The second case was an 80-year-old man treated with third-line Axi-cel for transformed follicular lymphoma (FL). The full clinical course leading up to CAR-T is described in Fig. 2a (also see “Methods”). The patient achieved a complete metabolic response on the first restaging PET/CT 30 days after CAR-T, which continued for 2.5 years. Two years after his CAR-T cell infusion, the patient developed COVID-19 infection (despite 6 total vaccinations) with a protracted course, complicated by recurrent superimposed bacterial pneumonias. In the midst of these infections, he presented with weight loss and fatigue, prompting a PET/CT scan, which revealed a highly FDG-avid enlarged right cervical lymph node (LN), concerning for disease relapse or new malignancy (Fig. 2b).

a The patient was originally diagnosed with stage III, grade 1–2 follicular lymphoma (FL) 9 years prior to CAR-T. Upon progression, he was treated with rituximab and achieved a partial response (PR) for 2.5 years before he relapsed with transformed disease. He was then treated with rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP) for 6 cycles with a complete response (CR) lasting 3 years. He was treated on a clinical trial (NCT03636503) with rituximab/avelumab/utomilumab combination immunotherapy, with a PR followed by progression after 6 months. He then received CAR-T therapy as third-line therapy. Yellow shading indicates time prior to CAR-T, blue shading indicates time post-CAR-T. b PET/CT image of the new lymphadenopathy presenting 2.5 years after CAR-T. c Representative immunohistochemistry images of T- and B-cell markers of the from a single core lymph node biopsy. d PET/CT image of spontaneous regression of the lymphadenopathy 6 weeks after initial scan.

Hematoxylin and eosin (H&E) stain of a core needle biopsy demonstrated architecture effaced by an infiltrate of medium to large cells with irregular/folded nuclei, condensed chromatin, variably prominent nucleoli, and scant eosinophilic cytoplasm with small, scattered lymphocytes in the background (Fig. 2c). By immunohistochemical staining, the abnormal appearing cells expressed mature T-cell markers (CD3, CD2, CD5, CD7), with many positive for both CD4 and CD8, and were positive for TCR-BF1. Scattered cells were also positive for MUM-1 as well as BCL-6, Stathmin, PD-1, and BCL-2 suggestive of a TFH subtype. The abnormal lymphoid cells were negative for CD10, CD30, ALK-1, CD20, PAX5, CD79a, CD21, and C-MYC. Only rare cells stained positive for TCR-Delta. By Ki-67 staining, the cellular proliferation rate was 40–50%, and EBV in situ hybridization was negative. Altogether, the morphologic and immunophenotypic profile was thought to be most in keeping with a TFH TCL. Moreover, TCR gene rearrangement studies demonstrated a clonal process with a Vɣ1-8 primer, supporting a diagnosis of malignancy. Clinical flow cytometry demonstrated sample viability of <5%, limiting interpretation. Given the patient’s history, the highest concern was for TCL arising from a CAR-T cell. Although the biopsy was diagnostic for TCL, the clinical scenario of lymphadenopathy arising in the context of COVID-19 infection and bacterial pneumonia raised the possibility that this represented an infectious or inflammatory phenomenon, and the patient had no other signs or symptoms of progressive malignancy. Thus, the decision was made to observe with short-interval repeat imaging. Indeed, PET/CT 6 weeks later demonstrated spontaneous near-complete resolution of the cervical lymphadenopathy with complete resolution by 3 months (Fig. 2d).

Genomic characterization of possible TCL

To determine the cellular etiology and molecular characteristics of this aberrant T-cell population, remaining cervical LN formalin-fixed paraffin embedded (FFPE) sections and genomic DNA derived from peripheral blood mononuclear cells (PBMCs) obtained 12 months post-CAR-T (“normal”) were evaluated by WGS. To assess for the presence of the CAR construct, we aligned to a custom reference genome including the axi-cel sequence^18,19 (“Methods”), and identified zero reads mapping to the axi-cel construct with high mapping confidence (MAPQ ≥ 30). We estimate the sequencing depth provides 95% power to detect reads mapping to the scFv region of the CAR construct (which is distinctly mappable from the human genome) even if the CAR were present in as few as ~3.4% of cells (“Methods”). Thus observing no mapping reads indicates that the clonal T-cell population was not derived from a CAR-T cell. Copy number analyses revealed no large-scale aberrations, but focal deletions at the TCR alpha, beta, and gamma loci were estimated to be present in 67% of cells by ABSOLUTE. MiXCR²⁰ identified two rearranged clonotypes at each of the three loci (Supplementary Data 1).

The spontaneous resolution of lymphadenopathy raised the possibility that this clonal population did not represent TCL, despite the aberrant histologic appearance and phenotype (e.g., CD4+ CD8+ double positive), apparent clonality, and presence of TFH markers, the combination of which are considered diagnostic of TCL²¹. Of 4584 total mutations (19 nonsilent variants, Supplementary Data 2) that were LN-specific (i.e., not in normal PBMCs), one was identified in TET2, which was a frameshift mutation found in exon 3 resulting in a premature stop codon. While not seen in our normal sample, which was collected at 1-year post-CAR-T, this TET2 mutation was found in the clinical next-generation sequencing panel performed on PBMCs collected three months after the new lymphadenopathy presented, raising the possibility of clonal evolution or clonal hematopoiesis of indeterminate potential. Nevertheless, at such a high allele frequency, it is likely that the mutation is present in the atypical T cells in addition to the myeloid compartment and could be a driver of abnormal T cell proliferation. Of the other 18 non-silent mutations, none have been previously linked to the development of TCL or T cell lymphoproliferative neoplasms. A point mutation was observed in HIP3K, a kinase that can bind to Fas, leading to phosphorylation of Fas-associated death domain (FADD), which plays a key role in T cell apoptosis and proliferation^22,23. While HIP3K mutations have not been specifically linked to TCLs, dysregulation of FADD phosphorylation has been implicated in the proliferation capacity of T cell lymphoblastic lymphoma²⁴ and could be a potential mutation of interest for future study.

Spatial transcriptomic interrogation of nodal pathology

To more deeply interrogate the clonality, phenotype, and spatial microenvironment of this hyperproliferative T-cell population, we performed single-nuclei spatial transcriptomics on a fresh-frozen core needle biopsy of the suspicious LN with Slide-tags²⁵. Previously, Slide-tags enabled spatial profiling of nuclei with 3’ capture chemistry of mRNA transcripts. However, 3’ capture chemistry poses challenges for TCR sequencing since the variable region is at the 5’ end of the TCR mRNA transcript. For improved recovery of TCR sequences, we adapted Slide-tags to be compatible with 5’ capture chemistry (“Methods”). Slide-tags 5’ snRNA-seq revealed a LN composition dominated by transcriptionally abnormal CD4+ CD8+ double positive T-cells, making up 51% of profiled nuclei (Fig. 3a). The rest of the LN was composed of 20% macrophages, 10% regulatory T-cells, 5% fibroblasts, 5% CD4+ T-cells, and smaller proportions of CD8+ T-cells, NK cells, endothelial cells, and dendritic cells. No B cells were detected. The CD4+ CD8+ T-cells were distributed across the full area of the profiled LN in a spatially disorganized manner, uncharacteristic of normal LN architecture (Fig. 3b). To examine these T-cell phenotypes, we quantified the usage of T-cell gene expression programs defined from a diverse collection of healthy, COVID-19, cancer, and autoimmunity patient samples (Fig. 3c and Supplementary Data 3)²⁶. The double positive T-cells scored highly for TFH and T peripheral helper cell program usage. Differential gene expression analysis between the double positive T cell population and all other profiled T cells revealed an enrichment for genes involved in proliferation, hypoxia, and stress response, further supporting an abnormal cellular phenotype (Fig. 3d). Concordant with the WGS analysis, we mapped the snRNA-seq results to a custom reference containing the axi-cel construct and did not detect any spatially mapped nuclei containing at least one count of an axi-cel mapped read (“Methods”). Additionally, qPCR to detect the CAR construct sequence in the Slide-tags library from the biopsy was negative for transgene amplification (Supplementary Data 4).

a UMAP embedding of Slide-tags 5’ snRNA-seq transcriptome profiles, colored by cell type assignment. b Spatial mapping of snRNA-seq profiles, split by cell type. c T-cell gene expression program average usage scores. d Volcano plot of differentially expressed genes in the CD4+ CD8+ T cell population compared with all other T cells using a two-sided Wilcoxon rank-sum test with adjustment for multiple testing using the Bonferroni correction. Select genes are highlighted. e Size of beta chain clonotypes by T-cell population. Each color represents a different clonotype. The two beta chain sequences belonging to the most expanded clonotype are written in white text. f Module scoring of viral reactivity transcriptional signature by T-cell population. g Spatial neighborhood enrichment heatmap where Z scores denote the magnitude of enrichment or depletion between a given center cell type and neighbor cell type. Z scores are capped at 9 and −9 for visualization purposes.

The majority of CD4 or CD8 single positive T-cells belonged to single member or small-sized clonotypes (Fig. 3e). In marked contrast, the CD4+ CD8+ T-cells belonged to a highly expanded clonotype, consisting of two alpha chains (CASPGGLTGGGNKLTF, CALSHPFRNSGNTPLVF) and two beta chains (CASSLVVWGRGLNEQFF, CASSQQDSRNTIYF), suggestive of an atypical biallelic rearrangement of both alpha and beta chains (Supplementary Data 5)^27,28,29. Examining public CDR3α and CDR3β TCR clonotypes, we identified 6 alpha chain TCRs and 2 beta chain TCRs with reported specificity to viral antigens (Supplementary Data 6 and “Methods”), but the highly expanded clonotype did not match any known viral-reactive TCRs. Double positive T-cells scored lowly on a transcriptional signature for viral antigen reactivity, despite the lymphadenopathy arising during COVID-19 infection (Fig. 3f). The highly clonal nature of the aberrant CD4+CD8+ double positive population supports the notion of TCL, though may also be consistent with a non-malignant lymphoproliferative process.

Finally, we leveraged the spatial data to assess the immune microenvironment surrounding the expanded double positive T-cell population. While double positive T cells spatially clustered with each other, their neighborhoods were depleted of DCs, CD8+ T cells, CD4+ Treg cells, endothelial cells, and fibroblasts (Fig. 3g). We observed a weak enrichment for macrophages and pDCs in the neighborhoods of double positive T cells. Overall, the aberrant T cell population exists in immune-excluded spatial niches.