Deciphering the importance of culture pH on CD22 CAR T-cells characteristics

The products included in this study were derived from a cohort of 20 children and young adults who were diagnosed with recurrent or refractory CD22-expressing B-cell malignancies and were enrolled on a Phase 1/2 clinical trial of anti-CD22 CAR T-cells (NCT 02315612). All patients were treated in the Pediatric Oncology Branch of the National Cancer Institute, National Institutes of Health at a dose level of 0.3 × 10⁶ cells.

All anti-CD22 CAR T-cell products were manufactured at the Center for Cellular Engineering (CCE) between September 2019 and August 2022. The CCE is compliant with good manufacturing practice (GMP) and good clinical laboratory practice (GCLP) standards and has manufactured numerous CAR T-cell products for phase 1 and 2 clinical trials at the National Institutes of Health. All the products were manufactured using autologous mononuclear cells collected by apheresis. T-cells from fresh (n = 5) or cryopreserved (n = 15) autologous mononuclear cells were selected on CliniMACS (Miltenyi Biotec) using anti-CD4 and anti-CD8 magnetic beads and stimulated with anti-CD3/CD28 Dynabeads (ThermoFisher) with a ratio of 1 cell:3 beads. Cells were seeded into culture bags at a density of 1.5–2.0 × 10⁶ viable CD3⁺ cells. Cell culture was maintained in AIM-V medium (Gibco) containing heat inactivated human AB serum (Valley Biomedical), Glutamax (Gibco) and IL2 (Aldesleukin) in a 37 °C, 5% CO₂ incubator for 9 days. On day 2 of the culturing process, cells were transduced with GMP grade CD22 lentiviral vector (Lentigen) using bag spinoculation (1000 g for a period of 2 h). On day 3 (D3), transduction was stopped by media exchange. Debeading and cell density adjustment to 0.4 × 10⁶ viable cells/mL was performed on day 4 (D4) of the culturing process. Cell density was further adjusted on Day 7 (D7) to 0.6–1 × 10⁶ viable cells/mL. All cultures were harvested on Day 9 (D9) and were either freshly infused (n = 7) or cryopreserved (n = 9). A simplified 9-day manufacturing process is illustrated in Additional file 1: Supplementary Figure 1a.

Culture media supernatants were collected at four different time points during the CD22 CAR T-cell culture (on Day 2, 3, 4 and 9). 10 mL of culture supernatant was collected before transduction on D2, when transduction was stopped on D3, and before debeading on D4. Final supernatants were collected during the cell harvest on D9. All supernatants were aliquoted (1 mL per vial), immediately frozen and stored at − 80 °C until further analysis.

A Vi-CELL MetaFLEX bioanalyzer (Beckman Coulter) was used to measure pH, glucose, lactate, pCO₂ and pO₂ levels during the manufacturing process. At the time of the analysis, supernatants collected at various time-points during the culturing process were thawed and warmed to 24 °C. 65 uL of temperature adjusted supernatants were loaded into the MetaFLEX device for testing.

Culture media cytokine, chemokine, and growth factor levels were measured at 2 different time points (D2 and D9) during the culture process using the Bio-Plex 200 system (Bio-Rad Laboratories). Bio-Plex assays were performed according to the manufacturers’ instructions using commercially available Bio-Plex Pro Human Cytokine Screening 48-plex Panel Assay. All assay values are reported in pg/mL.

We have previously described the bulk RNA sequencing procedures in detail [18]. In brief, total mRNA from CAR T-cell products was isolated using the miRNeasy Mini Kit. The concentration and quality were assessed using Nanodrop 8000 and 2100 Bioanalyzer, respectively. DNA libraries were prepared using the TruSeq Stranded Total RNA kit and sequenced on the Illumina Nextseq550 platform.

Raw fastq files underwent quality control using FastQC and were further processed with Trimmomatic to remove adapter sequences and low-quality reads. After filtering, the reads were aligned to the human reference genome (GENCODE hg38) using the STAR aligner. Gene expression levels were quantified using subread (featureCounts), and TPM values were utilized for downstream analysis. For differential expression analysis, we employed the limma package in RStudio with custom scripts. Threshold for definition of differentially expressed genes (DEGs) is p value < 0.01 & |FoldChange|≥ 2. We then performed gene set enrichment analysis (GSEA), a bioinformatic method to determine whether a predefined set of genes showed statistically significant, concordant differences between two biological states. We divided the gene expression from our RNA-seq data into two sets: up-regulated and down-regulated set. Subsequently, we examined the presence of genes associated with each pathway in either the up-regulated or down-regulated set. The degree of enrichment was quantified by a normalized enrichment score (NES). A significant positive NES value suggests that members of the selected pathway gene set tend to appear in the up-regulated set, whereas a significant negative NES value indicates the opposite scenario. The analysis was conducted by the built-in function of the cluster Profiler package [19] in RStudio.

Transduction efficiency, viability and percentage of cells expressing CD3, CD4, CD8, CD14, CD15, and CD56 were measured using flow cytometry following staining with fluorescently labelled antibodies (BD Biosciences). The CD22-CAR vector transduction efficiency was detected using protein L (ThermoFisher). Untransduced cells were used to determine the positive gate for transduction efficiency. The gating strategy of protein L expression on CD22 CAR T-cells is displayed in Additional file 1: Supplementary Figure 1c. Frequencies of the above-mentioned markers were acquired on the BD FACS CantoII (BD Biosciences). Thirty thousand events in total were collected for each sample. Data were analyzed using BD FACSDiva software.

The clinical characteristics of patients whose CD22 CAR T-cell products were included in the study are detailed in Table 1. In total, twenty CD22 CAR T products were analyzed. All 20 products met the 0.3 × 10⁶/kg target dose. Four products were not infused, as the patients experienced rapidly progressive disease during the manufacturing process. Seven patients received infusions of fresh CAR T-cells, and nine patients received cryopreserved CAR T-cells.

During the manufacturing process, viable total nucleated cells (TNC) were counted on days 0, 2, 4, 7 and 9. The mean TNC at the start of the manufacturing process was 470 × 10⁶ cells (range 288–565 × 10⁶). On Day 2, a mean of 343 × 10⁶ total nucleated cells underwent transduction with lentiviral vector. On Day 4, there was a drop in total nucleated cells with a mean of 52 × 10⁶ cells (range 26–76 × 10⁶). The total number of cells harvested on Day 9 was 1.1 × 10⁹ (range 0.3–1.7 × 10⁹, Fig. 1a). Fold expansion was determined between the following days of the cell culture: 0–2, 2–4, 4–7 and 7–9. The maximum fold expansion was achieved on Days 4–7 of the culturing process (mean 8.5, range 4.6–11.9) with a decrease in expansion between Days 7 and 9 (mean 2.9, range 2.0–3.9, Fig. 1b). As shown in Fig. 1c, the percentage of CD3⁺ cells increased from 47.9 in the apheresis bag to 91.2 post-CD4/CD8 selection to 99.7 at the time of the harvest of CAR T-cells. The apheresis bag and selected CD4/CD8 T-cells were analyzed for additional markers such as CD56, CD14, or CD15. Additional file 1: Supplementary Figure 1b shows the analysis of these markers. The transduction efficiency (percent of protein L⁺ cells) was assessed on Days 7 and Day 9 of the manufacturing process. On Day 7, after the fastest period of expansion, 31% transduction was observed (data not shown). By Day 9, a significantly larger proportion of cells expressed protein L, (42.8%, p = 0.0073, Fig. 1d), suggesting preferential expansion of transduced cells. The final CD4/CD8 ratio was 2.8 (Fig. 1e).

Culture media supernatants were collected at 4 different time points throughout CD22 CAR T-cell manufacturing: Days 2, 3, 4, and 9. Our primary focus was on the findings on Day 2 and Day 9. Prior to transduction on Day 2, the pH of cell culture media ranged from 7.1–7.5 (Fig. 2a). pH level showed a statistically significant negative correlation with the D2 TNC count and fold expansion. On retrospective evaluation, we separated the products into 2 groups: 12 samples had relatively low pH on Day 2 (ranging from 7.093–7.289; mean 7.2), and 8 samples had high pH (ranging from 7.323–7.541; mean 7.4). The differences in pH levels between these 2 groups were statistically significant (p < 0.0001, Fig. 2a). The cell counts, fold expansion, and metabolic activity of T-cells were compared between the low pH and high pH groups. Cell products in the low pH group had significantly elevated total nucleated cell counts (414 vs 236 × 10⁶, p < 0.0061, Fig. 2b), greater fold expansion (0.85 vs 0.53, p < 0.0015, Fig. 2c), higher pCO₂ (23.2 vs 17.6 mmHg, p < 0.001, Fig. 2d) and higher lactate levels (0.73 vs 0.38 g/L, p < 0.0001, Fig. 2f). Those in the low pH group also had reduced glucose levels (1.9 vs 2.3 g/L, p < 0.0003, Fig. 2e) compared with the high pH group. These results demonstrate the relationship between lower extracellular pH and T-cell growth, expansion, and glycolysis as pH levels correlated with pCO₂, glucose, and lactate levels. On the other hand, Day 2 pH levels did not correlate with CD3% (Additional file 1: Supplementary Figure 2a) or pO₂ levels (Additional file 1: Supplementary Figure 2b).

Cytokine and growth factor levels were also compared between low and high pH groups. On Day 2 of the manufacturing process, six analytes demonstrated significantly greater levels among samples in the low pH group compared with the high pH group (Fig. 3). Increased levels of stem cell factor (SCF, p < 0.04), macrophage colony-stimulation factor (M-CSF, p < 0.004), leukemia inhibitory factor (LIF, p < 0.0052), interleukin 15 (IL-15, p < 0.0066), interleukin receptor 2 alpha (IL2Rα, p < 0.0002), and tumor necrosis factor beta (TNF-β, p < 0.0001) were observed in products with more acidic pH. In the context of increased proliferation associated with lower pH as shown above, these findings were consistent with our expectation that more active T-cells will secrete more cytokines and growth factors into the cell culture media.

Next, we investigated whether the pH level measured at the end of manufacturing was associated with cell or metabolic parameters. Overall, we observed a slight decline in pH values during the manufacturing process with higher values on Day 2 compared to Day 9. The Day 9 pH ranged from 7.0 to 7.4 and based on its values, the final CAR T-cell products were also separated into 2 groups representing low (9 patients, pH ranged from 7.008–7.235 with the mean pH of 7.1) and high pH (11 patients, pH ranged from 7.252 to 7.432 with mean pH of 7.3). The difference in final pH levels between these two groups was statistically significant (p < 0.0001, Fig. 4a). However, unlike the Day 2 pH groups, these final pH groups did not differ in terms of cell counts (Fig. 4b), fold expansion (Fig. 4c), CD3% (Fig. 4d), pCO₂ (Fig. 4e), or pO₂ (Fig. 4f). Higher pH at harvest correlated with higher glucose (1.91 vs 1.55 g/L; p < 0.0003, Fig. 4g) and lower lactate production (0.64 vs 0.89 g/L; p < 0.0002, Fig. 4h), suggesting reduced glycolysis and lower metabolic activity.

The Day 9 levels of 7 cytokines differed between the high and low pH groups. All 7 were elevated in the low pH group compared to the high pH group [interleukin 6 (IL6, p < 0.008), interleukin 12 subunit p70 (IL-12p70, p < 0.003), tumor necrosis factor alpha (TNF-α, p < 0.008), leukemia inhibitory factor (LIF, p < 0.007), macrophage migration inhibitory factor (MIF, p < 0.002), macrophage colony-stimulation factor (M-CSF, p < 0.003) and granulocyte macrophage colony-stimulation factor (GM-CSF, p < 0.02)] (Fig. 5). The levels of LIF and M-CSF were also greater in the low pH group on day 2 (Fig. 3), although others did not differ significantly.

CD22 CAR T-cell final product characteristics were analyzed by RNA sequencing (Fig. 6). Sequencing analysis of 17 final products assigned 78 differentially expressed genes between pH groups at harvest: 15 downregulated and 63 upregulated under low pH. GSEA analysis revealed that genes involved in T-cell activation, lymphocyte activation, differentiation, and mediated immunity, tend to be downregulated in low pH conditions. This highlights the value of pH as an indicator during CD22 manufacturing as it can help identify more activated CAR T-cell products (Table 2 shows the list of genes involved in each pathway).

To determine if product pH on Day 2 correlated with clinical outcomes, we compared disease response rates between the low and high pH groups (Fig. 7). Of the 12 patients whose samples were in the low pH group on D2, 10 were infused and 2 were not due to progressive disease. Among the patients receiving the cells, 8 (80%) achieved a complete response (CR), which is defined as complete bone marrow response at 28 days. The high pH group contained 8 patients. Six patients received CAR T-cell infusions and 2 did not due to rapidly progressive disease. In contrast to the low pH group, among those with high pH, only 2 patients (33%) achieved a clinical response, while 4 (67%) did not. The relationship between low starting pH and CR was most consistent in the subsets of patients with high baseline disease burden (M3 marrow, > 25% blasts) (Additional file 1: Supplemetary Figure 3a) and prior hematopoietic stem cell transplant (HSCT) (Additional file 1: Supplemetary Figure 3b).

Moreover, we observed a difference in toxicity patterns between the low and high pH groups. In the low pH group, 8 patients (all those with CR) developed cytokine release syndrome (CRS), 5 (62.5%) of them with Grade 1 and 3 with (37.5%) Grade 2. Non-responders in this group did not have any toxicities. Only 25% of the patients with clinical response in the low pH group experienced neurotoxicity and 50% had IEC-HS (immune-effector cell HLH-like syndrome) (Table 3). In the high pH group, both patients with clinical response developed Grade 2 CRS, neurotoxicity, and IEC-HS. Furthermore, in the high pH group, all non-responders experienced toxicities as well: all had CRS, 2 of them had neurotoxicity, and one had IEC-HS.

Analysis of pH changes over the course of manufacturing demonstrated that patients who achieved CR were more likely to have cells with stable or slightly increased pH values during the manufacturing process, whereas those who did not have CR had a drop in pH (Fig. 8a). Similar to the pattern observed with starting pH, the association between a drop in pH and lack of clinical response was most pronounced among patients with high baseline disease burden or history of prior HSCT (Fig. 8b and c).