Clonal expansion of T memory stem cells determines early anti-leukemic responses and long-term CAR T cell persistence in patients

Abstract

Low-affinity CD19 chimeric antigen receptor (CAR) T cells display enhanced expansion and persistence, enabling fate tracking through integration site analysis. Here we show that integration sites from early (1 month) and late (>3yr) timepoints cluster separately, suggesting different clonal contribution to early responses and prolonged anti-leukemic surveillance. CAR T central and effector memory cells in patients with long-term persistence remained highly polyclonal, whereas diversity dropped rapidly in patients with limited CAR T persistence. Analysis of shared integrants between the CAR T cell product and post-infusion demonstrated that, despite their low frequency, T memory stem cell clones in the product contributed substantially to the circulating CAR T cell pools, during both early expansion and long-term persistence. Our data may help identify patients at risk of early loss of CAR T cells and highlight the critical role of T memory stem cells both in mediating early anti-leukemic responses and in long-term surveillance by CAR T cells.

Adoptive T cell transfer shows great promise for the treatment of both viral infections and cancer but much remains to be learnt about the optimal phenotype of T cell products, their clonal dynamics once infused and the origin of both early responding T cells that mediate initial responses as well as long-term-persisting T cells that provide immunological surveillance. According to the hierarchical model of T cell differentiation, upon antigenic priming naïve T (T_N) cells differentiate into central memory (T_CM) cells that express lymph node homing molecules but have limited effector function, which in turn give rise to effector memory (T_EM) cells that preferentially traffic to the tissues and differentiate to terminally differentiated effector cells (T effector memory CD45RA+ (T_EMRA) cells) that mediate rapid effector functions¹. Subsequently, Gattinoni et al.² demonstrated the existence of the stem cell memory (T_SCM) compartment, with a naïve surface phenotype and high proliferative capacity but also attributes of memory T cells, which appears endowed with both the potential for multipotent differentiation and long-term self-renewal. For durable anti-viral/anti-tumor responses long-term immune surveillance is critical, but the ontogeny of T cells mediating such long-term responses remains a matter of active debate. Early data from adoptive transfer of gene-marked Epstein-Barr virus (EBV)-specific T cells, which had a predominantly T_EM phenotype at infusion, showed these could be detected up to 9 yr post-infusion³. In a macaque model, cytomegalovirus (CMV)-specific CD8 T cell clones derived from T_CM (but not T_EM) cells persisted long term after adoptive transfer⁴. Tracking of T cells retrovirally transduced to express adenosine deaminase (ADA) using integration site (IS) analysis has demonstrated that genetically engineered T_SCM clones persist and preserve their differentiation potential for up to 12 yr post-infusion⁵. On the other hand, similar analyses of marked circulating T cells 2–14 yr after infusion of T cells retrovirally transduced with the herpes simplex virus (HSV) thymidine kinase suicide gene suggested that long-term-persisting marked cells arose predominantly from both infused T_SCM and T_CM compartments⁶.

CD19 CAR T cells show unprecedented responses in relapsed/refractory acute lymphoblastic leukemia (ALL)^7–10. Anti-leukemic responses correlate with both the level of early CAR T cell expansion¹¹ and the duration of CAR T cell engraftment, with long-term persistence critical for durable remission if used as a stand-alone therapy. The origins of early expanding and long-term-persisting CAR T cells may differ and have yet to be defined. This will be critical in designing manufacturing protocols to optimize outcomes. Initial data in a xenogeneic mouse model of mesothelioma where mesothelin CAR-transduced human CD8+ T_SCM/T_CM/T_EM cells were infused with CD4+ CAR T cells demonstrated superior anti-tumor responses with T_SCM2. In a xenogeneic model of lymphoma, CD19 CAR T cells derived from selected CD4+ T_N and CD8+ T_CM populations gave superior anti-tumor activity, and a 1:1 combination of bulk CD4⁺ and CD8⁺ T_CM CD19 CAR T cells showed optimal activity¹², leading the investigators to test this approach clinically in adult ALL¹³. Xu et al.¹⁴ found that the expansion of CD19 CAR T cells in patients correlated with the frequency of CD8+CD45RA+CCR7+ within the infused product whose phenotype approximated T_SCM (it should be noted that in this study T_N cells were not distinguished from T_SCM by evaluation of CD95 expression), although CAR T cell expansion and persistence in this study were poor, possibly due to the lack of lymphodepletion and the use of a CD28 costimulatory domain, respectively. Little is known about the how the phenotype and clonal composition of CAR T cells change following adoptive transfer. IS analysis offers us a unique tool to track the clonal dynamics of genetically modified CAR T cells in patients, as each transduced cell is stably marked by a vector IS acting as a genetic barcode. Using T cell receptor (TCR)-β and IS analysis, Sheih et al.¹⁵ have shown that CAR T cells at early timepoints post-infusion are polyclonal but that individual clones display different clonal kinetics after infusion. However, because of limited CAR T cell persistence in this study, the analysis was restricted to 30 d post-infusion. Moreover, it is not clear whether these data are skewed by the use of the infused products derived from selected CD4+ and CD8+ T_CM cells. To date, it has not been possible to isolate long-term (>1 yr) persisting CAR T cells in patients. In the immunotherapy with CD 19 CAR T cells for high-risk relapsed/refractory pediatric CD19+ acute lymphoblastic leukemia (CARPALL) phase 1 study, the use of an improved low-affinity CD19 CAR resulted in enhanced expansion and persistence of CAR T cells post-infusion¹⁶, enabling us for the first time, to our knowledge, to detect and select long-term-persisting CAR T cell flow cytometrically. By combining phenotypic analysis after treatment with high-resolution IS analysis of selected CAR T cells, we have analyzed the fate of infused CAR T cells in two patients with long-term persistent CAR T cells in peripheral blood and compared this with two patients who had early loss of CAR T cells (Fig. 1a).

Fig. 1. Immunophenotypic characterization of patients with long-lasting CAR T cells in the product and after infusion over time.

a, Schematic summarizing the manufacture of CAR T cells and key aspects of CARPALL clinical study design. The days reported in this schematic are reflective of the days of in vitro expansion since PBMC isolation. b, Schematic representation of survival and expansion potential of the main T cell subpopulations. c, Gating scheme used for the identification of T cell subtypes used for following up CAR T cells and total CD3+ cells. d, FACS plot showing CAR cells within CD3+ cells as well as T cell composition inside the CAR+CD3+ CD45RA+CD62L+ population for Pt4 (top panels) and Pt6 (bottom panels). Percentages of CAR cells and CD62L/CD45RA compartments are showed inside each gate. Timepoints of analysis are shown on top of each set of panels (d, days after CAR cell infusion). e, Percentage of CAR cells in the CD3+ cell compartment over time in Pt4 (top panel) and Pt6 (bottom panel). f, Summary of data shown in panel c for CAR T cell composition over time for Pt4 (left panel) and Pt6 (right panel). Pie charts show composition of CAR T cells at each timepoint. Longitudinal contribution of each subpopulation to the CAR cell compartment is shown in the relative plot below. aCD3, anti-CD3; aCD28, anti-CD28; CAT, CAR molecule CAT-19; FSC-A, forward scatter; SSC-A, side scatter.

Results

CAR T cell kinetics and phenotype post-infusion

To study the composition of CAR T cells in our patients over time, we performed a comprehensive immunophenotypic analysis for the detection of CAR cells and for the characterization of different T cell subtypes (Fig. 1b,c). All four patients analyzed showed similar kinetics of CAR T cell engraftment, showing relative contribution in the CD3⁺ compartment as high as 14–80% at the time of peak expansion (day 14) followed by contraction to 6–30% by day 30 (reflecting elimination of CD19+ targets), and then low-level long-term persistence in patient 4 (Pt4) and Pt6 (0.1% at 36 and 28 months) (Fig. 1d,e). We then assessed the naïve/memory phenotype of CAR T cells in both the infused product and the peripheral blood post-infusion flow cytometrically, using CD45RA, CD62L and CD95 as markers (Extended Data Fig. 1). In the patients with long-term CAR T cell persistence, the infused CAR T cell product was primarily composed of T_CM (78–81% of CAR+ T cells) with a lower contribution of T_EM (17–21%), but also a small but detectable pool (1–2%) of T_SCM (Fig. 1f). Following infusion, in both of these patients the composition of CAR T cells switched substantially towards the T_EM phenotype, which became the predominant fraction at both early and late timepoints (Fig. 2e). There was a small increase in the percentage of terminally differentiated effector CAR T cells (T_EMRA cells) in both patients up to the 6-month timepoint, but this declined thereafter. Interestingly, the small fraction of T_SCM cells present upon infusion became increasingly represented over time, implying that the CAR T_SCM cells detected in the cell product are indeed endowed with high self-renewing potential after infusion.

Fig. 2. Immunophenotypic characterization of patients with short-living CAR T cells in the product and after infusion over time.

a, FACS plot showing CAR cells within CD³⁺ cells as well as T cell composition inside the CARCD3+CD45RA+CD62L+ population for Pt10 (top panels) and Pt17 (bottom panels). Percentages of CAR cells and CD62L/CD45RA compartments are shown inside each gate. Timepoints of analysis are shown on top of each set of panels (d, days after CAR cells infusion). b, Pie charts showing the composition of the drug product in Pt10 and Pt17. c, Longitudinal contribution of each subpopulation to the CAR cell compartment in Pt10 and Pt17 versus Pt4 and Pt6. d, Bar plot showing the relative fraction of T_SCM cells detected in the drug product in Pt4 and Pt6 versus Pt10 and Pt17.

For comparison, we performed a similar phenotyping analysis on CAR T cells from two patients (Pt10 and Pt17) who lost circulating CAR T cells at 2- and 4-months post-infusion to investigate whether differences in product composition or relative engraftment of T cell subsets might correlate with lack of persistence in patients. As shown in Fig. 2, the CAR T cell products (particularly in Pt17) had a more differentiated phenotype with a lower proportion of T_SCM and higher T_EM than in the patients with long-term persistence (Fig. 2a,b,d). Similarly, by day 30 post-infusion, CAR T cells from these patients showed a predominantly T_EM phenotype, whereas in Pt4 and Pt6 we observed a substantially higher preservation of the T_SCM and T_CM compartments (Fig. 2a,c).

IS analysis of CAR T cells

Next, we characterized CAR T cells of these patients at the single-clone level to address the following questions: (1) What are the CAR T cell clonal dynamics occurring during the acute anti-tumor response phase versus steady state? (2) What is the clonal composition of the long-surviving CAR T cells? And, (3) which fraction of the cell product contributes the most to the long-term maintenance of CAR cells in these patients? To this aim, we collected, through our well-established protocol that combines FACS of blood cell subpopulations, whole-genome amplification, linear amplification-mediated PCR and high-throughput sequencing (Illumina)5, a total of 9,881 and 3,667 ISs from T_SCM, T_CM and T_EM isolated from the cell products of Pt4 and Pt6, respectively. Similarly, we were also able to collect 4,417 and 6,646 ISs in Pt4 and Pt6, respectively, from FACS-sorted CAR T_SCM and T_CM/T_EM cells at 14 and 30 d after infusion, as well as 1,676 and 1,930 unique ISs from the total circulating CAR cells of both individuals, respectively, at 180 and 720/840 d after treatment (Extended Data Fig. 2). It should be noted that it was not possible to select subpopulations at these later timepoints due to the low fraction of CAR cells in circulation. The number of ISs collected at long-term follow-ups is well above that previously reported at similar timepoints after CAR T cell treatment¹⁷, and suggests that while clonal diversity decreases somewhat post-infusion, CAR T cells remain highly polyclonal even at late timepoints post-infusion. We then analyzed the distribution of our ISs in the genomes of CAR T cells from the infused products and isolated post-infusion at both early and late time-points. As shown in Fig. 3 (top panels) and Extended Data Fig. 3, in the cell products the lentiviral vector integrated into (on average 69% of ISs in both patients) or in proximity to a variety of different genes, most of which are classical targets of lentiviral integrations in gene therapy¹⁸, with a prevalence of genes related to biological processes such as DNA metabolism and repair (Extended Data Fig. 4). Notably, there was no evidence of clonal selection after infusion with respect with the general insertional profile (Fig. 3, bottom panels, and Extended Data Fig. 3), or of clones with lentiviral genomic insertions in proximity to genes exerting a specific cellular function (Extended Data Fig. 4).

Fig. 3. Distribution and abundance of ISs collected in the product and after infusion over time in Pt4 and Pt6.

Circos plot showing the genomic distribution of ISs in the product (top panels) and after infusion (bottom panels) in Pt4 (left panels) and Pt6 (right panels). Descriptions of each concentric data track and timepoint (d, days after CAR cells infusion) are shown in the legends on the left of the circus plots. Descriptions of the contents of these plots are reported in the gray legend on the top right. Word clouds showing top hit genes in the product and after infusion are shown at the center of each plot (the bigger the gene name the higher the number of ISs detected in its locus). Chr, chromosome.

Clonal diversity is preserved in long-lasting CAR T cells

We next assessed the diversity of CAR T cells in the CAR T products by means of number and relative contribution of ISs collected starting from FACS-sorted T_SCM, T_CM or T_EM cells. Because the frequency of these populations differed in the cell product, to normalize for this we used equal amounts of genomic DNA from each subpopulation for performing IS analysis. All subpopulations showed high diversity in both patients 4 and 6, with the biggest clones ranging from 2.1% to 14.7% of each population (with the exception of one clone with ISs contributing to 30.4% of the reads collected from the T_EM of Pt6) (Figs. 4 and 5, top panels, and Extended Data Fig. 5a). Of note, the T_CM fraction of both individuals showed a higher number of integrants and higher diversity as compared with the other two populations. This suggests that T_CM may be more permissive to transduction as compared with T_SCM and T_EM, possibly as a result of a higher activation/expansion during in vitro transduction. Based on our initial observation on the increasing relative contribution of T_SCM in circulation as compared with that infused in the cell product, we investigated the clonal dynamics of this population after infusion in more detail. To do this, we isolated the T_SCM from the rest of the memory/effector cell fractions in the early response phase (14 and 30 d post-infusion) when the number of circulating CAR cells in circulation permitted cell sorting of CAR T cell subfractions. In Figs. 4 and 5 (bottom panels), we compared the IS profile of sorted CAR T_SCM cells with the rest of the CAR population at the same timepoints. We observed that in both patients the clonal diversity of T_SCM (measured either as Shannon or as Gini–Simpson diversity and shown in Figs. 4 and 5, bottom panels, and Extended Data Fig. 5d, respectively) dropped substantially during the early response phase compared with the infused product. Most of the T_SCM cells isolated from patients early after treatment were in fact composed of individual clones bearing ISs whose reads ranged from 89.0% to 96.9% of the total read count (Extended Data Fig. 5b). It is important to note that neither the number of sorted cells nor the amount of ISs collected per sample would alone explain this finding (Extended Data Fig. 6), and that the same experimental conditions have been applied to the study of small subfractions of progenitor cells in gene therapy patients with different results19. This observation is instead likely a reflection of bona fide clonal bursts and oligoclonal expansion occurring in T_SCM cells during the early response phase (Figs. 4 and 5, bottom panels). At the same timepoints, the clonal composition of the remaining CAR cells (T_CM/T_EM) remained highly polyclonal despite the major shift from T_CM to T_EM phenotype occurring upon infusion, as described above. This would suggest that the CAR T_CM and T_EM cells infused in these patients have expanded in a more ‘homeostatic’ fashion as compared with their T_SCM counterparts.

Fig. 4. Diversity of ISs in different T cell subtypes over time from Pt4.

Top panels, for each product subpopulation analyzed, colored circles (T_SCM, orange; T_CM, light blue; T_EM, blue) contain bubble plots of clones contributing >0.01% to the total population. The dimension of the bubble is proportional to the size of the clone. The name of the gene closest to the relative IS is reported inside the bubble. The plots on the right show Shannon diversity index for each subpopulation. Bottom panels, plots showing Shannon diversity index (left y axis) of ISs over time in TSCM cells (orange lines) and in TCM/TEM cells (dark blue) and in all T cells (green lines). The gray lines show the percentage of CAR cells over time (right y axis). For each timepoint analyzed, colored circles contain bubble plots of clones contributing >0.01% to the total population at each timepoint. The dimension of the bubble is proportional to the size of the clone. The name of the gene closest to the relative IS is reported inside each bubble.

Fig. 5. Diversity of ISs in different T cell subtypes over time from Pt6.

Top panels, for each product subpopulation analyzed, colored circles (T_SCM, orange; T_CM, light blue; T_EM, blue) contain bubble plots of clones contributing >0.01% to the total population. The dimension of the bubble is proportional to the size of the clone. The name of the gene closest to the relative IS is reported inside the bubble. The plots on the right show Shannon diversity index for each subpopulation. Bottom panels, plots showing Shannon diversity index (left y axis) of ISs over time in T_SCM cells (orange lines) and in T_CM/T_EM cells (dark blue) and in all T cells (green lines). The gray lines show the percentage of CAR cells over time (right y axis). For each timepoint analyzed, colored circles contain bubble plots of clones contributing >0.01% to the total population at each timepoint. The dimension of the bubble is proportional to the size of the clone. The name of the gene closest to the relative IS is reported inside each bubble. Analysis performed as in Fig. 4.

We then analyzed in both patients the diversity of the total CAR cell composition at later timepoints (180 and 720/840 d after infusion). While the frequency of CAR cells dropped substantially from the initial phase, we observed that the clonal diversity of long-term-persisting CAR T cells remained high (Figs. 4 and 5, bottom panels, and Extended Data Fig. 5c). This suggests that CAR T cells detected at late timepoints have persisted through ‘homeostatic’ survival and that long-term maintenance of CAR cells in our patients was not dependent on the expansion of a few large clones carrying nonphysiological/aberrant insertional profiles.

To study the clonal dynamics of CAR T cells over time, we first compared all datasets from early or late timepoints and performed unsupervised clustering according to IS similarities (Fig. 6a). We found that ISs isolated at 14 and 30 d after infusion showed the highest similarities and clustered separately from the clonal datasets of 180 and 720/840 d, suggesting that clones active in the early response phases were less likely to be recaptured at later times after treatment. Because we have multiple capture timepoints, we could estimate the lower bound number of clones composing the CAR population by means of mark–recapture statistics, as previously described20. The results suggest that around 34,000 and 55,000 CAR clones might still be in circulation in Pt4 and Pt6, respectively, up to 24/28 months after infusion (Fig. 6b). We then assessed the recapture probability of the expanded T_SCM clones observed at early timepoints (Fig. 6c). We found that these clones were different at each follow-up and that some of them were also detected in the early CAR T_CM/T_EM population, suggesting that this expansion was transient and accompanied by differentiation of the T_SCM cells to memory and effector cells. In contrast, these large clones do not seem to have contributed substantially to the pool of long-term-surviving CAR T cells (Fig. 6c).

Fig. 6. Tracking of IS in different T cell subtypes over time in Pt4 and Pt6.

a, Unsupervised clustering based on IS sharing among subpopulations over time in Pt4 (left panel) and Pt6 (right panel). b, Estimation of the number of clones composing the CAR^{+ cell} population in each patient based on the M_th Chao lower bound (LB) model applied to IS recapture probability. Shown are individual values and relative standard errors from the abundance model fit estimation (for Pt4, abundance = 34,083 and s.e. = 1,699; for Pt6, abundance = 55,888 and s.e. = 2,795). c, Heatmaps showing tracking and relative contribution of TSCM clones (IS, rows) at d14 and d30 recaptured in other T subtypes and timepoints (columns) in Pt4 (top panel) and Pt6 (bottom panel). The intensity of blue is proportional to the relative clone size (IS abundance). d, Distribution of IS in proximity to TET2 in both patients combined. The x axis displays chromosomal coordinates (base pairs) while the y axis displays the IS abundance (log10 of percentage) calculated within the population and timepoint where it has been observed. The dark blue bars show IS detected in the patients while red bars show IS detected in the cell product. The transcription start site (TSS), exons (gray boxes) and introns (gray lines) of the TET2 gene are shown on top of the plot in their respective chromosomal locations. e, Heatmap showing tracking and relative contribution of the IS shown in panel d for each patient in the product (in red) or after infusion (in blue). Relative IS abundance (percentage) calculated within the population and timepoint where it has been observed is reported inside each colored field (white field, not detected).

Importantly, with regard to oncogenic potential of integrated viral vector sequences, we did not observe any major skewing in the distribution of ISs towards gene categories related to T cell growth, activation or survival. In this regard, we identified and tracked 12 clones in our patients bearing an integration in the TET2 gene (Fig. 6d), a locus previously associated with clonal dominance in a single patient with chronic lymphocytic leukemia (CLL) treated with CD19 CAR T cells¹⁷. Of note, none of these clones showed signs of expansion either in the product or after infusion, contributing to a maximum of 0.044% of sequencing reads out of the total collected in each sample/timepoint (Fig. 6e). Collectively, these data show no evidence of insertional mutagenesis resulting in clonal selection of CAR T cells in these patients, either during the early response phase or in long-term-persisting CAR T cells.

Lastly, we compared these data with results from Pt10 and Pt17. The numbers of unique ISs retrieved in these individuals were broadly similar to those seen in Pt4 and Pt6. In Pt10 and Pt17, we collected 4,633 and 2,823 ISs, respectively, in T_SCM, T_CM and T_EM cells belonging to the cell products, as well as 2,295 and 1,020 ISs, respectively, in T_SCM and T_CM/T_EM cells during the first month of follow-up (Extended Data Fig. 7). We first measured the relative abundance of each IS (Fig. 7a) and analyzed the clonal diversity of the sorted CAR populations from the CAR T cell product and after treatment. As shown in Fig. 7b (top panels), these individuals displayed an overall lower level of clonal diversity in all populations composing the drug product as compared with the patients with long-surviving CAR T cells, and this was particularly evident when analyzing the T_SCM compartment. After infusion, we could again observe a clonal burst of CAR T_SCM cells during the first month. However, in contrast to the patients with long-term persistence of CAR T cells, the diversity of the T_CM/T_EM compartments rapidly dropped over the first month in these patients before CAR T cell disappearance (Fig. 7b, bottom panels).

Fig. 7. Diversity of ISs in different T cell subtypes over time from Pt10 and Pt17.

a, Scattered dot plots showing relative abundance of ISs in the product (top section) and at early timepoints (bottom section) after treatment in Pt10 (plots on the top) and Pt17 (plots on the bottom) (d, days after treatment; m, months after treatment). Mean percentage abundance is shown as a dotted line for each sample. Number of events in each dataset is equal to that reported in Extended Data Fig. 7. b, Diversity of drug product and CAR cells after infusion in Pt10 (top panels) and Pt17 (bottom panels). Top section, for each product subpopulation analyzed, colored circles (T_SCM, orange; T_CM, light blue; T_EM, blue) contain bubble plots of clones contributing >0.01% to the total population. The dimension of the bubble is proportional to the size of the clone. The name of the gene closest to the relative IS is reported inside the bubble. The plots on the right show Shannon diversity index for each subpopulation. Bottom section, plot showing Shannon diversity index (left y axis) of IS over time in T_SCM cells (orange lines) and in T_CM/T_EM cells (dark blue). The gray lines show the percentage of CAR cells over time (right y axis). For each timepoint analyzed, colored circles contain bubble plots of clones contributing >0.01% to the total population at each timepoint. The dimension of the bubble is proportional to the size of the clone. The name of the gene closest to the relative IS is reported inside each bubble.

Origin and maintenance of long-term-persisting CAR T cells

Having observed that the long-term-surviving CAR population was composed of a large and stable number of clones of equivalent sizes, we investigated whether we could establish a link between the ISs of the cell product and the ones detected after infusion. Our goal was to understand which of the infused CAR T cell subtypes gave rise to the early expanding and long-term-persisting CAR T cells post-infusion. The depth of our analysis and the number of ISs collected in the product and from the patients allowed us to detect a substantial number of identical ISs in samples collected before and after infusion, for the first time to our knowledge. The sharing of ISs between the CAR T cell product subsets and samples obtained at different timepoints post-infusion is represented in the correlation maps of Fig. 8a. These plots quantify the IS sharing between all samples analyzed, showing in blue dots the paired comparisons where ISs sharing was more substantial (positive correlation). In both patients with long-term persistence, we observed a relatively high frequency of ISs shared among the T subpopulations composing the cell product, suggesting that differentiation has occurred during in vitro expansion at the time of manufacture. Interestingly, we also found a relatively high proportion of ISs shared between the T_SCM cells of the product and the T_SCM cells isolated at 14 and 30 d after infusion in both patients. This would imply that CAR T_SCM cells active during the early CAR T cell response phase are predominantly derived from T_SCM cells in the product that have expanded during in vitro manufacture. Notably, T_SCM ISs sharing with T_CM/T_EM cells up to day 30 after treatment ranged between 0.3% and 23%, averaging 8.3%, suggesting active differentiation in the circulation of T_SCM cells originally contained in the drug product. In contrast, T_CM and T_EM cells in the product shared on average a maximum of 2.4% of ISs with the same subtypes up to early timepoints after therapy (range 0.09–7.1), a result in line with the more limited differentiation potential of these memory cell types as compared with T_SCM cells.

Fig. 8. Comparison of IS distribution pre- and post-infusion.

a, Correlograms showing upper triangle of the correlation matrix among IS datasets from Pt4 (left panel) and Pt6 (right panel). Positive correlations are shown in blue and negative in red. The size of circles is proportional to the correlation value. b, Ring plots showing relative contributions from each subtype (colored section of the ring) to the pool of ISs detected in the product and at early (left plots) or late (right plots) timepoints in Pt4 (top panels) and Pt6 (bottom panel). The total numbers of ISs captured both in the product and after infusion are shown inside each plot. The relative percentages of ISs belonging to each T cell subtype of the product that were shared with samples after infusion are shown in white inside each section of each ring plot. c, Model summarizing our working hypothesis for CAR T cell dynamics occurring in the patients of this study.

Despite the detection of identical ISs between the cell product and late timepoints, the IS numbers were not sufficient to reach significance, possibly due to inherent sampling limitations combined with the paucity of CAR T cells still left in circulation in these patients at 6 to 24/28 months after treatment. Nonetheless, the numbers of shared ISs allowed us to investigate the relative contributions of the product subsets to the early and late phases after infusion. As shown in Fig. 8b and Extended Fig. 8, regardless of long-term persistence, in all four patients the majority of the shared integrants at early timepoints were derived from the T_SCM CAR T cells in the product, suggesting that this compartment has a key role in early expansion. Similarly, in the patients with long-term persistence, the product subpopulation contributing the most to the clonal pool at late timepoints was the T_SCM, from which we collected 46.6% to 60.5% of ISs out of the group of insertions sites detected both in vitro and after infusion. Such a contribution is particularly relevant when one considers the extremely low frequency of T_SCM cells observed at the time of infusion (1–2% of total CAR T cells) as compared with TCM or TEM cells.

Discussion

In summary, by combining immunophenotyping and molecular tracking we have gained important insights into the nature and dynamics of both early expanding CAR T cells that mediate initial responses as well as long-term-surviving CAR T cells providing immune surveillance against leukemic relapse in two patients up to 3 yr after infusion. The IS retrieval methodology has been widely used by our group and others as a reliable method for tracking clones, showing added value compared with high-throughput TCR sequencing. This is because (1) it allows a specific tracking of engineered T cell clones even when they are highly diluted in a population of vector-negative cells and therefore in conditions that prevent FACS of CAR cells (such as at late timepoints); and, (2) differently from high-throughput TCR sequencing, it provides a set of stable genomic markers whose relative contribution is independent from the effect of thymic involution, immune tolerance or extrathymic rearrangements. Nonetheless, it should be noted that it is currently impossible to precisely discriminate individual clones based on the assumption 1 IS = 1 clone, since each genetically engineered cell may carry multiple copies of an integrated vector so that interpretation must be based on data trends rather than on the tracking of specific individual ISs or clones.

We show that CAR T cells in the infused product display a polyclonal IS profile and that CAR T cells remain clonally diverse at early and late timepoints post-infusion, with no evidence of clonal dominance. Individual clones showed different patterns of expansion and contraction with time, with different clones responsible for early expansion and long-term persistence. Interestingly, despite having observed several ISs in the TET2 locus, we could not find any sign of clonal expansion of the relevant CAR T cell clones. It should be noted that the TET2 insertional mutagenesis event described by Fraietta et al.¹⁷ was accompanied by a hypomorphic mutation in the second TET2 allele, an extremely rare occurrence which is unlikely to have taken place in the patients analyzed in the present study. It would seem therefore that bi-allelic disruption of the TET2 locus is needed for clonal expansion.

Our results, summarized in the scheme of Fig. 8c, suggest that following infusion, CAR T cells undergo a rapid switch from a predominantly T_CM to a T_EM phenotype during the early response phases without major clonal imbalance in the memory compartment but with a self-limiting clonal burst phase occurring in the T_SCM population. T_SCM cells from the infused product can potentially differentiate into other subsets post-infusion. Analysis of shared integrants between the CAR T cell product and after treatment at early (<1 month) timepoints demonstrated that, despite the low frequency of T_SCM cells in the products (0.5–2%), a substantial fraction of these clones (10–73%) were responsible for the circulating CAR T cells detected during this early expansion phase in all patients regardless of long-term persistence, implying that differentiation of the oligoclonally expanded T_SCM cells into T_EM cells may be responsible for much of the early anti-leukemic response. The patients with early loss of CAR T cells appeared to have a more differentiated phenotype, with a lower proportion of T_SCM cells and higher T_EM cells as well as reduced clonal diversity in their products. Patients with poor CAR T cell persistence showed markedly reduced clonal diversity in CAR T_CM and T_EM cells compared with those with long-term persistence. We speculate that when the drug product is more differentiated and less polyclonal, CAR T cells may not be able to preserve a clonal repertoire with enough diversity to avoid exhaustion during the clonal burst associated with the early anti-leukemic response. Further studies in larger patient cohorts are needed, but, if these findings are confirmed, these data have important clinical implications suggesting that assessment of the phenotype and clonal diversity of the product and at early timepoints may identify patients at high risk of early CAR T cell loss who may benefit from consolidative transplant.

In the patients with long-term CAR T cell persistence, CAR T cells did not terminally differentiate into T_EMRA cells upon exposure to CD19 antigen, but instead the T_SCM phenotype became more prevalent over time concomitantly with long-term maintenance of clonal diversity. We can therefore speculate that long-term maintenance of CAR T cells is supported by self-renewing cells endowed with homeostatic proliferative capacity such as T_SCM cells. Critically, we could estimate that clones belonging to the small fraction of T_SCM cells present in the cell product at the time of infusion had the highest recapture probability within the pool of long-lived CAR+ T cells in circulation, proving for the first time, to our knowledge, the crucial role of T_SCM cells in long-term anti-leukemic surveillance by CAR T cells.

Our working hypothesis is that the remarkable early expansion and persistence of our CAR T cell product achieved in Pt4 and Pt6 reflects a combination of two key factors. Firstly, our previous preclinical data16 both in vitro and in a xenogeneic mouse model suggested that signaling through our low-affinity CAR molecule CAT-19 (CAT) CAR results in enhanced CAR T cell expansion which may in part be mediated through higher expression of IL-7R and Bcl-2, promoting homeostatic proliferation and preventing apoptosis. Moreover, the faster off rate with this CAR could potentially have reduced differentiation of CAR T cells during the early anti-leukemic response, preventing terminal differentiation and allowing self-renewal of CAR T_SCM cells capable of long-term survival. Our CAR construct incorporated a 4–1BB costimulatory domain, which may confer prolonged persistence compared with CD28 by preventing exhaustion through tonic signaling^21,22. Secondly, our manufacturing protocol was relatively short (7–8 d) and did not utilize exogenous cytokines in the medium, which may favor less in vitro differentiation and preservation of high levels of T_CM cells whose expansion potential could have contributed to an enhanced early response. These culture conditions allowed a small fraction of T_SCM cells to be infused in those patients in large enough doses to survive the initial clonal burst phases and to maintain a pool of CAR+ cells long term.

There are clearly some unavoidable technical constraints to the data reported in this study, including the limited number of patients in whom material was available, the stability of the CD62L marker upon cell culture which could introduce a bias in the estimation of T_SCM/T_CM cells in the drug product, the low frequency of circulating CAR T cells at late timepoints, the potential sampling biases and the surrogate nature of clonal size quantification based on sequencing abundance of each IS, as a consequence of uneven amplifications of vector–genome junctions23.

Despite these limitations, our data provide evidence for the first time, to our knowledge, of the critical role of T_SCM cells in mediating both early anti-leukemic responses and long-term persistence of CAR T cells after infusion. Future efforts will focus on validating these findings in larger patient cohorts and improvement of CAR design and manufacturing methodology to preserve T_SCM and T_CM cells in the infused cell product.

Methods

Patients

The CARPALL study (NCT02443831) was approved by the UK Medicines and Healthcare Products Regulatory Agency (clinical trial authorization no. 20363/0361/001) and ethical approval (including the research in this manuscript) was obtained from the London West London & Gene Therapy Advisory Committee (GTAC) Research Ethics Committee (REC ref. no. 16/LΘ/0283). Written, informed consent was obtained from patients or their parents/guardians before study entry. All patients had relapsed CD19+ ALL and three had relapsed post-bone marrow transplant (BMT). Pt4 and Pt6 had long-term persistence of CAR T cells associated with ongoing durable complete remissions, whereas Pt10 and Pt17 had early loss of CAR T cells. Pt4 is a 21-yr-old man who had a combined bone marrow and central nervous system (CNS) relapse 1 yr post-transplant of matched unrelated donor peripheral blood stem cells and was treated with 10⁶ CAR T cells per kg in the third complete remission (CR) (molecular minimal residual disease (MRD) 2.6 × 10^-5) on 23 November 2016. He is currently well and in molecular CR with detectable circulating CAR T cells 3.5 yr post-infusion. Pt6 is an 11-yr-old boy who had a combined bone marrow and CNS relapse 14 months after a matched unrelated donor BMT for ALL and relapsed again in the cerebrospinal fluid (CSF) with molecular disease in the bone marrow 7 months after the second BMT. He was treated with 10⁶ CAR T cells per kg (molecular MRD 5 × 10^-4) on 5 October 2016 and remains well and in molecular CR with detectable circulating CAR T cells 3.5 yr post-infusion. Pt10 was a 10-yr-old boy with an isolated bone marrow relapse 17 months after matched family donor BMT who was treated with 10⁶ CAR T cells per kg in the third CR (molecular MRD 3 × 10^-5) on 30 August 2017 and had molecular CR and excellent initial CAR T cell expansion at 1 month, but CAR T cells were undetectable by flow and quantitative PCR from 2 months post-infusion. He relapsed with CD19 disease 7 months post CAR T cell therapy and subsequently died of a fungal infection. Pt17 is a 15-yr-old boy with late isolated CNS relapse who relapsed again in the CNS despite cranial irradiation and was treated with 10⁶ CAR T cells per kg in the third CR (molecular MRD 5 × 10^-5 negative) on 25 July 2018 and had continuing molecular CR and excellent initial CAR T cell expansion at 1 month, but CAR T cells were undetectable by flow and quantitative PCR from 4 months. He was consolidated with a mismatched unrelated donor BMT and remains in CR at 2 yr post-transplant. The average viral copy number per cell for the infused CAR T cell product for Pt4, Pt6, Pt10 and Pt17 was 8, 3.8, 4.9 and 5.6, respectively.

Characterization and isolation of lymphoid and myeloid cells

We performed immunophenotyping on whole-blood EDTA samples while cell sorting was performed on peripheral blood mononuclear cells (PBMCs) isolated from the whole blood by density gradient centrifugation using LymphoPrep (Sigma) after dextran sedimentation. For sorting, PBMCs were thawed in RPMI (10% FCS) pre-warmed with DNase 200 μl ml^-1 (also used TexMACS 5% human AB serum). Cells were then washed, counted and stained on ice in 100 μl of PBS for every 5 × 10⁶ cells. First, Fc block (BD 564220, 5 μl) was applied for 10 min and, without washing, an anti-idiotype CD19 CAT19 B12RIgG2a (Evitria) (2.5 μl or 3.5 μg) was added for 20 min. Then cells were washed and stained for 30 min with secondary antibody mix containing anti-rat IgG PE for CAR detection (Biolegend 405406, 0.5 μl), CD3 APC-Cy7 (Biolegend 300426, 5 μl), 7-AAD (BD 555816, 5 μl), CD95 BV711 (BD 563132, 5 μl), CD45RA v450 (BD 8053598, 2 μl) and CD62L APC (Biolegend 304810, 3 μl). Cells were washed, resuspended in sorting buffer (PBS (1% FBS) + 1 mM EDTA), passed through a cell strainer (30–40 μM) and then sorted using a FASCAriaIII cell sorter. Raw FACS data were collected using DIVA software (BD Biosciences) and analyzed with FlowJo (TreeStar). When feasible, an aliquot of the sorted cells was re-run through the cell sorter to check fraction purity.

DNA extraction and genome amplification

We performed DNA extraction using QIAamp DNA Blood Mini Kit (Qiagen) for samples of 10,000 to 5 × 10⁶ cells and MN NucleoSpin Tissue XS Kit (MN) for up to 10,000 cells, according to the manufacturers’ instructions. Genomic DNA was quantified using a Nanodrop One Spectrophotometer (ThermoFisher). For IS analysis, a whole-genome amplification was performed if the DNA yield after extraction was lower than 300 ng using the Qiagen Repli-G Mini Kit (Manufacturer), according to the manufacturer’s instructions.

Is collection and library preparation

For vector IS analysis, we used linear amplification-mediated PCR combined with high-throughput sequencing as previously described^5,20. Briefly, two rounds of linear PCR were performed (50 cycles each) to enrich for vector–genome junctions using biotinylated primers specific for vector long terminal repeats (LTRs). Streptavidin-coupled magnetic beads (Invitrogen Dynabeads Kilobase Binder Kit) were added to each sample, to capture linearly amplified fragments, followed by complementary strand synthesis. Samples then underwent restriction enzyme digestion using three different enzymes, MluCI, AciI or HpyCH₄IV, to minimize bias and improve genome coverage. Linker cassettes were ligated to these fragments, then samples underwent two rounds of exponential amplification PCR to amplify fragments containing vector LTR and linker cassette sequences. PCR fragment size depends on the distance between the known vector sequence and the closest enzyme recognition site. To visualize amplified fragments, we used high-resolution gel electrophoresis (ElchromScientific). Final PCR products were purified with QIAQUICK PCR Purification Kit (QIAGEN). Fusion PCR was used to add customized sequence-specific Illumina adaptors to the final PCR products. Different combinations of sample barcodes included in both LTR adaptors and linker cassette (LC) adaptors were used to differentiate between different samples.

Computational analyses of IS data

IS identification and analysis were performed through a custom analytical pipeline extensively described in previous publications^5,18–20. Briefly, raw IS datasets underwent a series of different bioinformatics filtering procedures according to the type of analysis to be performed. All datasets were processed with a ‘collision detection filter’ to univocally assign each IS to a patient and to one or more T cell subpopulations by applying a tenfold rule for contamination identification as previously reported5. A final matrix M was generated where each row r represented an individual IS while each column c represented an individual cell type/sample and timepoint. Each entry of M contained the abundance of each r for each c in terms of sequencing reads. The data shown in Figs. 3–7 and Extended Data Figs. 2–8 were generated on the basis of the IS databases attached as Source data. Graphical representations of IS analyses were generated with Prism8 (GraphPad Software) unless otherwise specified below. Panels in Fig. 3 were generated using the Circos software (http://circos.ca). Word clouds in Fig. 3 and Extended Data Fig. 4 were generated on the basis of the relative incidence of the single closest gene to each IS and plotted using the online suite WordClouds (https://www.wordclouds.com). Panels with lines in Figs. 4, 5 and 7b and Extended Data Fig. 1 were created by plotting IS diversity over time calculated as Shannon diversity or Gini–Simpson index through the R package BiodiversityR (https://cran.r-project.org/web/packages/BiodiversityR/index.html). The ‘bubble’ plots in Figs. 4, 5 and 7b were created on the basis of the IS with abundance >0.01% relative to each subpopulation and timepoint using the R package packcircles (https://cran.r-project.org/web/packages/packcircles/index.html). The dendrograms in Fig. 6a were generated based on the similarity among IS datasets through the hclust function (https://www.rdocumentation.org/packages/fastcluster/versions/1.1.25/topics/hclust). Estimations of clonal abundance and standard errors shown in Fig. 6b were calculated by the conversion of M to an M(0,1) matrix of incidence and by the application to M(0,1) of log-linear models for closed populations through the R package Rcapture (https://cran.r-project.org/web/packages/Rcapture/index.html, function = closedp.t). The M _th (model for abundance estimate with time and heterogeneity effects) Chao LB method was selected for visualization in Fig. 6b, being the most conservative estimation among the ones with the lowest Bayesian information content. The heatmaps in Fig. 6c,e were generated through the use of the R package gplots (https://cran.r-project.org/web/packages/gplots/index.html). The correlograms in Fig. 8a were generated using the R package corrplot (https://cran.r-project.org/web/packages/corrplot/corrplot.pdf). Correlation values for both plots were generated through the function Mcorr. The IS distribution around transcription start sites, the gene content of IS loci shown in Extended Data Fig. 3 and the Gene Ontology analysis of hit genes shown in Extended Data Fig. 3 (right panels) were generated through the online suite GREAT (Genomic Regions Enrichment of Annotations Tool) (http://great.stanford.edu/public/html/) using BED data from IS coordinates, associating genomic regions with the rule of single nearest gene within 1,000 kilobases to each IS and applying the gene annotations to these regions.

Statistics and reproducibility

No statistical method was used to predetermine sample size. The experiments were not randomized, and no data were excluded from the analyses.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Extended Data

Extended Data Fig. 1. Immunophenotyping of CD3+ cells in Pt4 and Pt6.

Additional FACS plot on CD62L+CD45RA+CD95 expression in CAR+ T cells, subtype frequencies within total CD3+ cells and relative summary of data on the CD3+ T cell composition overtime for Pt4 (top panels) and Pt6 (bottom panels). Percentages of CD62L/CD45RA compartments are showed inside each gate. Longitudinal contribution of each subpopulations to the CD3+ cell compartment is shown below for each patient according to the color code legend on the right.

Extended Data Fig. 2. Integration sites collected from Pt4 and Pt6.

Summary of number of IS (grey bars) and relative sequencing reads (white bars) in Pt4 (top panel) and Pt6 (bottom panel) (d = days after treatment).

Extended Data Fig. 3. Distribution of integration sites collected from Pt4 and Pt6.

Distribution of IS with respect to TSS (left plots) or gene content of the loci (right plots) in the cell product (red) or after infusion at early or late timepoints after treatment (blue) for Pt4 (top panels) and Pt6 (bottom panels).

Extended Data Fig. 4. Gene categories relative to integration sites collected from Pt4 and Pt6.

Plots on the left show word clouds of hit genes in the cell product (in red) and after infusion at early or late timepoints after treatment (in blue) for Pt4 (top panels) and Pt6 (bottom panels). Relative gene enrichment analysis for top biological processes of hit genes relative to each word cloud in the product (red bars) or after infusion (blue bars) is shown on the right plots (significance reported as -log10 binomial p-value from one-tailed tests decreasing from top to bottom).

Extended Data Fig. 5. Relative abundance and diversity of integration sites collected from Pt4 and Pt6.

Scattered dot plots showing relative abundance of IS in the product (a) and at early (b) or late (c) timepoints after treatment in Pt4 (plots on the left) and Pt6 (plots on the right) (d = days after treatment, m = months after treatment). Mean percent abundance is shown as a dotted line for each sample. Number of events in each dataset is equal to what reported in Extended Data Fig. 2. d) Longitudinal plots showing Gini/Simpson Diversity Index (left y-axis) of IS overtime in T_SCM (orange lines) and in T_CM/T_EM (dark blue) and in all T cells (green lines). The grey lines show the percentage of CAR cells overtime (right y-axis).

Extended Data Fig. 6. Correlation of integration sites and number of cells collected from Pt4 and Pt6.

Longitudinal plots showing number of IS collected (a) and number of cells collected (b) in T_SCM (orange lines) and in T_CM/T_EM (dark blue) at early timepoints for Pt4 (left panels) and Pt6 (right panels). The plot in (c) shows the correlation observed between number of cells collected (y-axis) and clonal diversity for all samples and both patients (x-axis, Shannon Diversity Index) shown as blue dots. Interpolation with best fit curve and R squared value are shown in black.

Extended Data Fig. 7. Integration sites collected from Pt10 and Pt17.

Summary of number of IS (grey bars) and relative sequencing reads (white bars) in Pt10 (top panel) and Pt17 (bottom panel) (d = days after treatment).

Extended Data Fig. 8. Sharing of integration sites in the product with the CAR T populations in Pt10 and Pt17 after infusion.

Ring plots showing relative contribution from each subtype (coloured section of the ring) to the pool of IS detected in the product and at early timepoints in Pt10 (left) and Pt6 (right). The total number of IS captured both in the product and after infusion is shown inside each plot. The relative percentage of IS belonging to each T cell subtype of the product that were shared with samples after infusion is shown in white inside each section of each ring plot.

Data availability

The fastq files relative to sequencing of IS amplicons generated for this study are available through the NCBI repository (https://submit.ncbi.nlm.nih.gov/), BioProject accession number PRJNA718947. Source data are provided with this paper. All other data supporting the findings of this study are available from the corresponding author on reasonable request.

Code availability

No new code was generated or used in this manuscript.