This study presents data from 29 pediatric and five adult patients, who underwent treatment with tisagenlecleucel (Kymriah^®) in three European centers after receiving lymphodepleting chemotherapy with cyclophosphamide and fludarabine.

CAR T cell and B cell monitoring was carried out on EDTA-anticoagulated peripheral blood as part of routine follow-up. Additional flow cytometry and real-time PCR tests for analytical validation or subpopulation analysis were performed on leftovers. Cryopreserved aliquots from rinsed tubing of tisagenlecleucel products after infusion were used for spike-in experiments and subpopulation analyses. Leftover samples from randomly selected, anonymous patients who had no history of CAR T cell treatment served as negative controls to determine the assay sensitivity. Our retrospective analysis of CAR T cell and B cell detectability was based on 272 measurements on 30 patients (Supplementary Table 1). The study was approved by the local Ethics Committee (#20-807). All CAR T cell patients had given written informed consent.

A four-color flow cytometry panel was designed using a commercial CD19 CAR Detection Reagent (Miltenyi Biotec, Bergisch Gladbach, Germany). The reagent consists of a biotinylated CD19 antigen that specifically binds CD19-targeted CARs. In a second incubation step the biotin-labeled CAR T cells are then stained with a fluorochrome-conjugated anti-biotin antibody.

In brief, 200 µl whole blood were treated for 10 min with 2 ml of NH₄Cl-based erythrocyte lysing solution (Beckman Coulter, Krefeld, Germany) and washed with PBS, containing 0.5% HSA. After removal of the supernatant down to 200 µl, cells were resuspended and 100 µl were transferred to a new flow cytometry tube. Following 15 min of incubation with 1 µl CD19 CAR Detection Reagent, cells were washed twice and incubated for 15 min with 1 µl Anti-Biotin-PE (Miltenyi Biotec, Bergisch Gladbach, Germany), 10 µl 7-AAD, 5 µl CD3-APC, and 5 µl CD45-KrO (all purchased from Beckman Coulter Immunotech, Marseille, France). After a final washing step, cells were acquired on a NAVIOS flow cytometer (Beckman Coulter, Krefeld, Germany). Cellular debris was excluded based on light scatter properties and CAR T cells were defined as 7-AAD^-/CD45⁺/mononuclear cells/CD3⁺/CD19 CAR⁺ (Supplementary Figure 1).

B cells, T cells, and other leukocyte subpopulations were regularly assessed on a FC500 flow cytometer (Beckman Coulter, Krefeld, Germany) as described previously (26). In a dual-platform approach, leukocyte counts were measured with the Sysmex XN-1000 hematology analyzer (Sysmex Deutschland, Norderstedt, Germany) and absolute cell counts of leukocyte subpopulations were calculated with the percentages determined by the flow cytometric assay. Relative B cell levels are reported as percentage of the lymphocyte gate.

CAR T cell-specimens were stained with 0.5, 1, 2.5, 5, 7.5, or 10 µl of CD19 CAR Detection Reagent and with the same volume of Anti-Biotin-PE. To ensure comparability, the titration experiments were otherwise conducted concordant with the protocol described above. We determined the optimal antibody volume as the volume with the highest mean fluorescence intensity (MFI) of the positive fraction divided by the MFI of the negative fraction (signal-to-noise ratio).

The baseline measurement was performed on the day of sample collection. Subsequently, the specimen was stored at ambient temperature without addition of fixatives or stabilizers and analyzed daily for up to five consecutive days with the CAR T cell detection protocol described above. Sample stability at a particular time point of measurement was assessed on the basis of the percentage of viable leukocytes and the relative percent difference (RPD) of the CAR T cell value from baseline.

To determine the sensitivity of our assay, cryopreserved cells (in RPMI medium with 20% FCS and 10% DMSO) from rinsed tisagenlecleucel infusion tubing were thawed, washed twice, and spiked into CAR T cell negative whole blood. The spike-in sample was serially diluted in whole blood and measured in triplicates if a sufficient blood volume was available. Un-spiked negative controls were run to assess the mean of blank samples, i.e., the mean event number in the CAR T cell gate. The Limit of Blank (LOB) was calculated as the mean of blank samples + 1.645 standard deviations (SD). Based on this, the Limit of Detection (LOD) was estimated as the LOB + 1.645 SD. Accurate CAR T cell quantification is highly dependent on the number of T cell events acquired in the run. Therefore, LOB and LOD were calculated for CAR T cell event numbers rather than concentrations. The Lower Limit of Quantification (LLOQ) was defined as the lowest concentration above LOD at which the coefficient of variation (CV) of the triplicate measurement achieved < 30% CV (27). Subsequently, the CAR T cell detection protocol was run on leftover CAR T cell negative patient specimens to verify the specificity.

Intra-assay precision was assessed with technical duplicates in a single measurement run on a NAVIOS flow cytometer. Afterwards, the flow cytometry tubes were transferred to a DxFLEX flow cytometer (Beckman Coulter, Krefeld, Germany), and the measurement was repeated on the alternative instrument.

Finally, we established the inter-assay precision on the initial instrument. For this purpose, sample processing and acquisition were performed independently for two more measurement runs. Prior to acquisition, the flow cytometer was powered-down, and daily quality control was repeated. Acceptance criteria were defined as ≤ 25% CV or 30-35% CV at LLOQ, as suggested by Sarikonda et al. (27).

For the detection of CAR T cells, we established a real-time PCR system. The amplicon is located in the region spanning from CD8 to 4-1BB. This targeted combination is unique for the chimeric construct and minimizes any background signals, which might impact detection of less frequent CAR T-cell levels. After sequencing the CAR from FMC63 IGHV to TCRζ, following primers were designed: forward (CTTCTCCTGTCACTGGTTATCAC), reverse (CTCGTTATAGAGCTGGTTCTGG), and probe (CAAACGGGGCAGAAAGAAACTCCT). Real-time PCR was tested with leftovers from one patient’s infusion bag showing 30% positive cells as confirmed by flow cytometric analysis. Isolated leftover DNA was diluted stepwise in mixed buffy DNA, originating from six CAR T cell negative samples, to simulate lymphocyte background. The determined quantitative range of the assay was 1E-04 (1 positive cell in 10,000 negative cells).

Immediately after CAR T cell detection with our flow cytometric assay, the sample was reincubated for 15 min with CD62L-FITC, CD45RO-ECD, CD8-APC-A700, CD4-APC-A750, and CD45RA-PacB (all purchased from Beckman Coulter Immunotech, Marseille, France) and acquired on the NAVIOS instrument again. Subpopulations were defined as CD62L⁺ CD45RO⁺ central memory T cells, CD62L^- CD45RO⁺ effector memory T cells, CD62L^- CD45RA⁺ T EMRA cells, and CD62L⁺ CD45RA⁺ naïve T cells. Stem cell memory T cells are comprised in the naïve population as no discriminating markers were assessed.

Dot plots were generated using NAVIOS Software version 1.3 (Beckman Coulter, Krefeld, Germany) or FlowJo Software v10.6.2 (FlowJo, LLC, Ashland, OR). Statistical analyses and graphs were done using Prism 9.0.1. (GraphPad Software, San Diego, CA). Statistically significant differences between subpopulations of T cells and CAR T cells were assessed by two-tailed paired t-tests. As normality testing suggested a departure from normality for the naïve subpopulations, the nonparametric two-tailed paired Wilcoxon signed rank test was chosen for this subset. The Bonferroni-Dunn correction was used to account for multiple comparisons. Adjusted p values were considered statistically significant below 0.05. Spearman’s rank correlation analysis was carried out between flow cytometry and real-time PCR test results.

CD19 CAR Detection Reagent and Anti-Biotin PE volumes were titrated to determine the reagent volume that results in optimal separation of CAR positive and CAR negative populations. 1 µl, 2.5 µl, and 5µl were considered within the optimal range (Figure 1), and 1 µl was subsequently chosen to be continued with in our CAR T cell monitoring.

In clinical routine, patient samples frequently arrive at the laboratory with delay, which makes their timely analysis on the same day difficult. More importantly though, many centers administering CAR T cells have not yet established CAR T cell detection methods, and patient specimens have to be shipped to cooperating centers for analysis. To assess the impact of such measurement delays on cell viability and the reported percentage of CAR T cells, we repeated testing of seven patient specimens at daily intervals. All samples for stability testing were collected within the first month after CAR T cell infusion. While initial leukocyte viability was very high and expectedly declined slowly over time (Figure 2A), the RPD of the median CAR T cell value plummeted strikingly early and exceeded our acceptability criterion of 20% difference even on the first day after sample collection (Figure 2B). As demonstrated in Figure 2C, measurement delays noticeably skewed the result towards lower CAR T cell concentrations. However, neither the general tendency of the course nor the order of magnitude or the detectability were affected.

To verify the validity of our reported results, we determined a number of assay performance characteristics. First, the LLOQ was established by means of a dilution series in CAR T cell negative whole blood (Figure 3A). Results from three independent experiments suggest precise (maximally 27.2% CV at LLOQ) and linear quantification of the CAR T cell fraction down to a LLOQ of 0.05% of T cells, which corresponded to 22 events.

We used negative controls and near-blank dilutions to calculate LOB and LOD at 8 and 13 CAR T cell events, respectively. The mean event number of viable T cells in the parent gate was 45,061. Below the LOD, a departure from linearity was observed, and CAR T cells could not be distinguished from background staining.

Following this, our assay was run on 21 CAR T cell negative patient samples, which were thought to be more comparable with CAR T cell patient specimens than healthy donor blood. Except for one outlier, which showed substantial background staining, all negative samples fell below the detection limit.

We next evaluated the precision of our assay at different levels across the detection range. A high positive (1277 CAR positive events), an intermediate positive (579 CAR positive events), and a low positive (58 CAR positive events) sample were selected, and three different precision parameters were assessed for each of them. The imprecision ranged between 0% and 16.8% CV, with the low positive sample displaying the highest imprecision among them. A fourth sample at the very detection limit (12 CAR positive events) clearly showed more intra-assay (40% CV), inter-assay (32.7% CV), and inter-instrument (20.3% CV) imprecision and thereby exceeded recommended limits for precision.

46 CAR T cell monitoring samples were independently assessed with both methods, flow cytometry and real-time PCR. Although the methods assay different features of CAR T cells and report in different units, we observed a clear correlation (Figure 3B), as shown by Spearman’s correlation coefficient (r = 0.723; p < 0.0001). Of note, the level of agreement between the methods seems to be relatively constant throughout the detection range, supporting acceptable accuracy down to very low frequencies.

The cellular kinetics of CAR T cells follow uniform patterns after infusion (Figure 4). Nevertheless, there is substantial variability between individual patients. Among pediatric ALL patients who received their first CAR T cell infusion, we observed the median peak expansion on day 12 (n = 13; range, 6 to 20) at 37.33% (n = 13; range, 3.53% to 69.36%) CAR T cells. At the last follow-up, 71% (12/17) of patients with more than six months follow-up were in ongoing BCA, which demonstrates functional CAR T cell persistence in the majority of patients (Figure 5A).

While we confirmed the maintenance of high CAR T cell levels for 1645 and 406 days in two patients (#1, #11) (Figure 5B upper left), most patients with BCA for more than 6 months (10/12) exhibited CAR T cell levels around the detection limit (Figure 5B upper right). In several of these patients (#4, #5, #6, #10, #13, #14), the CAR T cell percentage fell below and rose above the detectability threshold at different time points. And yet, the ongoing BCA and CAR T cell detectability by real-time PCR in these patients provide evidence of continuous low-level persistence. These inconsistently positive flow cytometric results might in part be explained by variations of the CAR T cell frequency in the patients. Another reason are differences in the T cell count per microliter between samples. If, for example, a sensitivity level of 0.05% is aimed for, a minimum of 26,000 T cell events would be required in the flow cytometric measurement to reach our experimentally established LOD of 13 CAR T cell events. An analysis of the relationship between T cell count and T cell event number showed that most samples with T cells <1000/µl did not meet this sensitivity target, whereas those with T cells ≥1000/µl did (Supplementary Figure 2).

On the contrary, 29% (5/17) of patients experienced recovery of B cells. To our surprise, we were intermittently able to detect low-level CAR T cell persistence during B cell recovery up to 18.6% in three patients (#7, #9, #12). In patient #12, the coexistence of CAR T cells and B cells was repeatedly confirmed by real-time PCR testing.

Interestingly, while CAR T cell concentrations were almost always ≤ 0.05% in samples showing B cell recovery, we found that a transient reoccurrence of BCA in two patients (#8, #12) after prolonged periods of B cell recovery was accompanied by an increase of CAR T cell levels to ca. 0.2% (Figure 5B lower left).

Among five patients who received more than one CAR T cell infusion, one patient (#6) achieved durable CAR T cell persistence above the detection limit with BCA lasting for one year before the patient was given other anti-B cell therapies for CD19 negative relapse. A second and a third infusion in patient #7 and a second infusion in patients #9, #12, and #15 though, showed limited re-expansion with a CAR T cell peak at 12.3%, 0.12%, 1.97%, 0.10%, and 0.85%, respectively. Additionally, CAR T cell levels declined rapidly below the detection limit. We observed B cell recovery 328 and 29 days after the second and third infusion in patient #7 and 118, 27, and 23 days after the second infusion in patients #9, #12, and #15 (Figure 5B lower right).

The phenotypic composition of 11 CAR T cell monitoring samples from seven pediatric ALL patients and four adult NHL patients in their initial expansion phase was evaluated. The mean proportion of CD4 or CD8 CAR T cells and CD4 or CD8 T cells did not differ significantly (Figure 6A). We observed an inversed CD4/CD8 ratio, which was 0.44 (27.53%/62.92%) for all T cells and 0.79 (41.72%/52.4%) for CAR T cells. Yet, there was large variation between individual patients and the CAR T cells of two of them were almost completely composed of CD4 CAR T cells with CD4/CD8 ratios above 20.

Comparison of CAR T cell subsets with their T cell counterparts showed a significantly higher frequency of highly differentiated effector memory T cells (p = 0.0346) and a significantly lower frequency of less differentiated naïve T cells (p = 0.0039) among CAR T cells (Figure 6B). In addition, the terminally differentiated T EMRA subset was significantly decreased (p = 0.0063).

Among CAR T cells, effector memory (66.39%) and central memory (25.69%) subsets accounted for the vast majority of cells. Naïve CAR T cells (2.20%) and CAR T EMRA cells (2.62%) were only detectable in low frequencies.

To resolve rare subsets, the phenotypic analysis of circulating CAR T cells requires high CAR T cell numbers. Consequently, most patient samples in the persistence phase of CAR T cells are not evaluable as they tend to be at the limit of detection. One patient (#1), however, presented 4 years after CAR T cell infusion with a frequency of 0.74% CAR T cells, allowing us to perform a subpopulation analysis. In contrast to the CAR T cell composition in tisagenlecleucel infusion bags or during the initial expansion period, we found 33.33% naïve CAR T cells in this patient (Figure 6C).

The notion of increased proportions of naïve cells during CAR T cell persistence was corroborated by the longitudinal follow up of another patient who maintained high CAR T cell numbers 27, 55, 63, 70, 205 and 379 days after CAR T cell infusion (Figure 6D). While naïve CAR T cells ranged between 0% and 1.29% within the first 90 days, they represented 4.88% and 6.52% of CAR T cells on days 205 and 379 post infusion, respectively. The rise of the naïve CAR T cell fraction was accompanied by an increase of naïve T cells among all T cells as well as a continuous decline of the overall CAR T cell frequency among peripheral blood T cells.