Abstract
Recently, chimeric antigen receptor (CAR) T cell technology has revolutionized cancer immunotherapy. This strategy uses synthetic CARs to redirect T cells to specific antigens expressed on the surface of tumor cells. Despite impressive progress in the treatment of hematological malignancies with CAR T cells, scientific challenges still remain for use of CAR T cell therapy to treat solid tumors. This is mainly due to the hostile tumor microenvironment and CAR-related toxicities. As the glycans decorating the T cell surface are implicated in T cell activation, differentiation, proliferation, and in the interaction of human T cells with tumor cells, we studied the role of human T cell glycosylation in more depth by manipulating their glycome. In this context, there is in vitro evidence that β-galactoside binding lectins (Galectins) can have a strong impact on the functionality of tumor-infiltrating T cells. The high-affinity poly-LacNAc N-linked galectin ligands are mainly synthesized onto the β1,6-GlcNAc branch introduced by N-acetylglucosaminyltransferase V (GnTV, encoded by Mgat5). We showed that knocking out Mgat5 in CD70 targeting CAR T cells leads to lower densities of poly-LacNAc modifications on the CAR T cell surface. Most interestingly, our results indicate that MGAT5 KO CD70 CAR T cells show enhanced potency to control primary tumors and relapses.
Introduction
Immunotherapy with T cells that are genetically modified to express chimeric antigen receptors (CARs), which target tumor-associated molecules, has shown impressive efficacy in several malignancies1. The advent of second-generation CAR T cells, in which activating and costimulatory signaling domains are combined, has led to encouraging results in patients with chemo-refractory B cell malignancies2,3. However, the translation of this clinical success to the treatment of solid tumors requires overcoming multiple obstacles4. In general, it is required to generate robust and stable populations of T cells that are able to infiltrate the tumor and escape the immunosuppressive effect of the tumor microenvironment (TME). Further issues in CAR T cell therapy include antigen escape, CAR T cell therapy-related toxicities and the relatively high occurrence of tumor relapse.
Cell surface glycosylation plays an important role in the interaction of human T cells with tumor cells and often contributes to escape mechanisms adopted by the tumor to evade T cell anti-tumor immunity5. For example, the expression of immune checkpoint inhibitors such as PD-1 and CTLA-4 is tuned by glycosylation6–8. Further, there is in vitro evidence that β-galactoside binding lectins (Galectins) can have a strong impact on the functionality of tumor-infiltrating T cells9. Galectin-1 controls T cell effector function homeostasis by regulating activation, differentiation, survival and cytokine production10. Galectin-9 is one of the ligands of Tim-3 and negatively regulates T cell immunity11. Binding of Galectin-3 to glycoproteins has both pro- and anti-apoptotic effects on T cells, depending on its localization. Intracellular Galectin-3 blocks apoptosis by stabilizing the mitochondrial membrane and preventing cytochrome c release,12 while extracellular Galectin-3 binds to glycoproteins such as CTLA-4 and Lag3 on the T cell surface, leading to inhibition and cell death of activated T cells13,14. Endogenous Galectin-3, produced by activated T cells, is recruited to the immunological synapse. There it negatively regulates T cell activation by destabilizing the immunological synapse through direct interactions with glycoproteins associated with the T cell receptor, and by promoting downregulation of the TCR15,16. Another interesting finding is that binding of Galectin-3 to antigen-specific activated CD8+ T cells inhibits their effector function within the tumor microenvironment13. It was shown that Galectin-3 prevents the formation of a functional secretory synapse by trapping LFA-1 in glycan-Galectin lattices, leading to reduced cytokine secretion17. Ex vivo treatment of T cells with an anti-Galectin-3 antibody or a Galectin competitive binder such as N-acetyllactosamine (LacNAc) resulted in the detachment of surface Galectin-3 leading to increased cytotoxicity and ability to secrete cytokines such as IFN-γ9,18.
The high-affinity poly-LacNAc N-linked galectin ligands are mainly synthesized onto the β1,6-GlcNAc branch introduced by N-acetylglucosaminyltransferase V (MGAT5) (Figure 1.A). Knocking out Mgat5 should thus also strongly reduce the density of poly-LacNAc modifications on the cell surface. MGAT5 deficiency was shown to markedly increase TCR clustering and signaling at the immune synapse, resulting in a lower T cell activation threshold and increased incidence of autoimmune disease in vivo and in human19.
To evaluate the impact of altered cell surface glycosylation on cytotoxic T cell functionality, specifically in a cancer immunotherapy setting, we used CD70 as the CAR target. Nanobodies targeting CD70 have been thoroughly evaluated as antigen-binding module in a CAR format (CD70 nanoCAR) in the lab of Prof. Dr. Bart Vandekerckhove (Department of Clinical Chemistry, Microbiology and Immunology, Ghent University)20. We specifically aimed to evaluate the impact of glyco-engineering via Mgat5 KO on the CAR T cell glycome and on their in vitro and in vivo activation, proliferation, differentiation and anti-tumor functionality.
We could demonstrate that MGAT5 KO CD70 nanoCAR T cells are functional and even perform better than CD70 nanoCAR T cells, both in vitro and in vivo. Both the average tumor volume and the tumor growth rate of primary and secondary tumors are significantly lower in the MGAT5 KO CD70 nanoCAR T cell treated group, as compared to the CD70 nanoCAR T cell treated mice. These results show that disrupting N-glycosylation modifications on CAR T cells enhances their capability to control primary tumors and subsequent relapses. Interestingly, MGAT5 KO CD70 nanoCAR T cells are present in higher numbers than CD70 nanoCAR T cells both in vitro and in peripheral blood and spleen upon specific antigen recognition.
Results
Engineering of MGAT5 KO CD70 nanoCAR T cells
We optimized a workflow for the combined CRISPR-Cas9 mediated glyco-gene editing and retroviral CAR delivery to purified, activated human CD3+ T cells. The presence of both CD4+ and CD8+ T cell subsets in the final CAR T cell product is indispensable for efficient anti-tumor immunity.
To efficiently combine CRISPR-Cas9-based glyco-gene editing and retroviral CAR delivery, various experimental steps were optimized. Optimal editing and transduction efficiencies were obtained when CD3+ T cells were stimulated with Immunocult for three days, after which activated T cells were first subjected to Cas9 RNP nucleofection, followed by a 1-hit retroviral transduction on the same day. Engineering efficiencies were assessed on day 10. The experimental timeline is depicted in Figure 1.B.
CRISPR editing efficiencies were determined by Sanger sequencing of the region of interest followed by ICE analysis, and the mean editing efficiency as percentage insertions and deletions (% indel) for the Mgat5 locus over multiple experiments was consistently high (exceeding 80% indel) as is depicted in Figure 1.C. Flow cytometry was used to measure both CD70 nanoCAR expression and GFP expression as read outs of the retroviral transduction efficiency. High CD70 nanoCAR transduction efficiencies were consistently obtained over multiple experiments, irrespective of the simultaneous glyco-gene engineering as shown in Figure 1.D.
MGAT5 KO CD70 nanoCAR T cells show an altered glycocalyx
In order to be able to assess the extent of the intended glycosylation changes upon glyco-gene engineering, we developed a lectin-based flow cytometry assay. For the detection of poly-LacNAc structures, we used the lectin from Datura stramonium (DSL) (Figure 1.E). This lectin is reported to bind well to LacNAc and oligomers containing repeating LacNAc sequences next to its preferred N-acetylglucosamine oligomer ligand.
When comparing the DSL lectin stain intensity of mock engineered CD70 nanoCAR T cells with that of MGAT5 KO CD70 nanoCAR T cells, we observe a clear reduction in signal, indicating that we successfully eliminated N-glycan β1,6-branching and subsequent elongation of this branch with poly-LacNAc modifications.
As a complementary method to profile the CAR T cell surface glycosylation, we adapted the DSA-FACE method developed in our research group to enable the analysis of cell surface N-glycosylation. We aimed to directly release the N-glycans from the cell surface by applying the PNGaseF digest on living cells in suspension. We established an optimized protocol in which we incubate 1 × 106 cells per sample in the presence of 0.125 IU PNGaseF in PBS for 2 hours at 37°C. Subsequently, the cells are removed by centrifugation and the crude digest is labeled with APTS for 1 hour at 70°C. After two rounds of clean-up over Sephadex resin to remove excess label and salts, labeled N-glycans are resuspended in water and analyzed by DSA-FACE. The complete protocol is schematically depicted in Figure 1.F.
When CAR T cells are engineered for MGAT5, the N-glycan profile is clearly different from that of mock-engineered CAR T cells (Figure 1.F). The peaks in P6 disappear while the peaks in P4 show a higher intensity relative to P2 and P3. This shift in electrophoretic mobility is consistent with the removal of one LacNAc unit (two monosaccharide units) or a shift from a tetra-antennary to a tri-antennary N-glycan. These DSA-FACE results are also in agreement with the lectin-staining experiments, where we observed a reduction in DSL staining intensity upon MGAT5 engineering (Figure 1.E). When comparing to the annotated N-glycan profile of human plasma (data not shown), this observation indeed confirms that P6 corresponds to a tetra-antennary N-glycan, while peaks in P4 and P5 correspond to tri-antennary N-glycan structures.
Characterization of the CD70 expressing tumor cell lines
In order to study the anti-tumor functionality of the MGAT5 KO CD70 nanoCAR T cells, two tumor cell lines were used in our studies. THP-1 cells are a M4 subtype acute myeloid leukemia (AML) cell line and SKOV-3 cells are a serous adenocarcinoma cell line. We confirmed the cell surface CD70 expression on these cells by flow cytometry (Supplementary Figure 1.A). Jurkat cells (immortalized line of human T cells) were included as a negative control. Further, we performed the anti-CD70 cell surface staining on non-transduced (NTC) and CD70 nanoCAR transduced CD3+ T cells and did not identify auto-antigen expression.
Galectins exert a broad range of effects during different aspects of T cell-mediated immunity by the formation of lattices on the T cell surface5. In anti-tumor immunity, it has been shown that Galectin-1 and Galectin-3 in the TME lead to tolerogenic signaling and immune suppression. LacNAc is the ligand recognized by Galectins and the affinity of the interaction is proportional to the LacNAc content of the glycan structure. We hypothesized that by eliminating MGAT5 expression in order to reduce the poly-LacNAc content on cytotoxic T cells, the inhibitory effect of Galectins on T cell immunity can be reduced. To this end, we first verified that the tumor cell lines used in our study indeed express Galectin-1 and Galectin-3.
Secretion and subsequent cell surface binding of Galectin-1 and -3 was detected by performing a flow cytometry experiment with anti-Galectin-1 and -3 antibodies. The results are shown in Supplementary Figure 1.B. As positive control, cells were incubated with recombinant Galectin-1 or -3 before performing the cell surface staining. As a negative control, cell surface Galectin binding was abolished by the addition of the competitive inhibitor lactose. Jurkat cells were included as negative control cells. Galectin-1 expression is detected for both the THP-1 and SKOV-3 cell lines. Further, galectin-3 expression is clearly observed for the SKOV-3 cell line but only slightly for the THP-1 cell line. No secretion and cell surface binding of galectins is seen on primary CD3+ T cells and Jurkat cells. Binding of recombinant Galectin-3 to T cells leads to an increase in signal, while recombinant Galectin-1 does not seem to bind to the primary T cells.
Additionally, we confirmed the expression of Galectin-1 and -3 in tumor sections from tumor-bearing NSG mice (Supplementary Figure 1.C). The latter were obtained by ectopically inoculating human SKOV-3 cells. The SKOV-3 tumor model is used in the experiments described below.
MGAT5 KO CD70 nanoCAR T cells are functional in vitro
In a first set of experiments, we evaluated the viability and functionality of MGAT5 KO CD70 nanoCAR T cells in vitro. Viability is maintained for each condition as is depicted in Figure 2.A. After engineering and culturing, most of the cells in the total CD3+ T cell pool are CD4+ T cells. Even with the 4-1BB signal in the CAR construct, which is believed to support a moderate rise in the CD8+ T cell fraction, the CAR T cell groups show a decrease in the CD8+ population. Furthermore, this decrease is even more pronounced when CD70 nanoCAR T cells were CRISPR-Cas9 engineered (both mock Cas9 and MGAT5 KO), which suggests that the viral transduction and nucleofection procedures might affect the growth of CD8+ T cells more than that of CD4+ T cells (Figure 2.B)
The antitumor effects of CAR T cells depend on their capacity to secrete cytokines upon exposure to antigens. Therefore, we evaluated the cytokine production of the glyco-engineered CD70 nanoCAR T cells after challenging them with the THP-1 and SKOV-3 target cell lines (Figure 2.C-E). Target cells were co-incubated for 16 hours with MGAT5 KO CD70 nanoCAR T cells. Unstimulated cells were included as negative control and Immunocult stimulation was included as positive control. Subsequently, T cells were labelled for intracellular TNF-α, IFN-γ and IL-2. The MGAT5 KO CD70 nanoCAR T cells are able to produce cytokines upon antigen stimulation and the proportion of cytokine-producing cells is similar to, or even higher than what is observed for mock nucleofected CD70 nanoCAR T cells. This cytokine expression is dependent on CD70 nanoCAR expression, given that non-transduced T cells (NTC) fail to express cytokines or express only very low levels in the presence of CD70 positive cells (but do show expression of cytokines after polyclonal Immunocult stimulation).
In order to evaluate the combined proliferative and cytotoxic potential of MGAT5 KO CD70 nanoCAR T cells, T cells were co-cultured with THP-1 target cells at different ratios for a period of 14 days. The number of THP-1 cells left in culture was determined by flow cytometry every three to four days. At day 7, a second challenge was performed by adding target THP-1 cells to the co-cultures. Results obtained with three independent T cell donors are depicted in Figure 2.F-I.
Figure 2.F and G shows the results corresponding to an effector/target (E/T) ratio of 0.15, that is 20 000 THP-1 target cells co-cultured with 3000 CD70 nanoCAR effector cells. At this ratio, all target cells get killed by day 4, in the wild-type, mock engineered and MGAT5 KO CD70 nanoCAR T cell conditions (Figure 2.F). Even at a very low E/T ratio of 0.015 (20 000 target cells co-cultured with only 300 CD70 nanoCAR T cells), all target cells are eliminated by day 4, irrespective of the engineering condition (Figure 2.H).
Most interestingly, from day 11 onwards, at both E/T ratios the number of CD70 nanoCAR T cells is higher for those that are knockout in Mgat5, indicating a higher expansion of these glyco-engineered CAR T cells over time Figure 2.G and I. We found that, on day 11, the number of MGAT5 KO nanoCAR T cells is about 1.74 times higher (95% CI: 1.36 to 2.21) than the number of nanoCAR T cells starting from the same conditions. On day 14, the number of MGAT5 KO nanoCAR T similarly is about 1.70 times higher (95% CI: 1.33 to 2.18). All these estimates are averaged over the two E/T ratios and the three independent donors.
Treatment of tumor-bearing mice with MGAT5 KO CAR T cells leads to a better outcome in terms of tumor control
After validating the in vitro activity of MGAT5 KO CD70 nanoCAR T cells, we aimed to evaluate whether MGAT5 KO CD70 nanoCAR T cells are also capable of clearing a tumor upon adoptive transfer in vivo.
The NOD.SCID IL2rγnull (NSG) mouse strain has been widely used in the pre-clinical evaluation of CAR T cell efficacy. Immune-deficient NSG mice lack functional mouse T cells, B cells, NK cells and are deficient in cytokine signaling through the common γC receptor21. Human tumor xenograft models were established in NSG mice by subcutaneous injection of luciferase-expressing SKOV-3 cells in the flank. Ten days after tumor cell inoculation, the presence of a subcutaneous tumor was evaluated by measurement with a slide caliper and through bioluminescent imaging (BLI) performed using an in vivo imaging system (IVIS).
After establishment of a solid, palpable tumor, mice were treated with either mock Cas9-engineered or MGAT5 KO CD70 nanoCAR respectively. As control groups, mice were treated with PBS to evaluate tumor development, or with non-transduced T cells (NTC) to evaluate graft versus host disease (GvHD) and non-specific anti-tumor effects. Throughout the experiment, tumor burden was measured every two days with a caliper and every 4 days through IVIS. A schematic representation of the experimental timeline is depicted in Figure 3.A.
At day 34, after the first phase of the experiment, the presence and phenotype of CAR T cells was evaluated in the blood and spleen. Furthermore, mice were followed-up in time and challenged between day 87 and day 90 with a second tumor to evaluate long-term anti-tumor efficacy. Again, tumor burden was evaluated over time and the mice were sacrificed between day 118 and day 123 for end-point analyses.
In the following sections, we will describe the results obtained for two independent experiments (Experiment A and Experiment B), which were performed with T cells from different donors. For the analysis, the treatment groups were divided in three treatment groups of interest: The ‘No CAR’ group contains the data from all the mice that did not receive any CD70 nanoCAR T cells, and thus includes untreated mice and mice treated with PBS or NTC. The ‘CAR’ group contains the data from all the mice that received a CD70 nanoCAR T cell treatment, with or without mock Cas9 engineering. The ‘CD70 nanoCAR - MGAT5 KO’ group contains data from the mice that received MGAT5 knockout CD70 nanoCAR T cells.
The outcome of the treatment was defined by 4 subtypes for the primary tumor challenge. (1) Full control meaning the tumor becomes undetectable and no relapse follows. (2) Full control but occurrence of a relapse later on. (3) Partial control meaning a halt in tumor growth but the tumor remains detectable and all mice also experience a relapse after long-term follow-up. (4) No control of tumor growth throughout the duration of the experiment.
As is clear from Figure 3.B and the table in Figure 3.C, the primary tumor is not controlled by the mice that did not receive CAR T cells, meaning that they were all sacrificed at the humane end-point. When we compare CD70 nanoCAR treated groups with MGAT5 KO CD70 nanoCAR treated groups, we see that more mice control tumor growth when they were treated with MGAT5 KO CD70 nanoCAR T cells, and that all of these mice show full control, with or without relapse, of primary tumor growth. Contrary to this, a considerable number of mice in the CAR treated group show only partial or even no control at all of the primary tumor.
As opposed to experiment A, in which we did not observe any relapse of the primary tumor over time, the tumor did regrow in some of the treated mice in experiment B. A survival analysis was performed to evaluate whether a difference could be observed in either the number of relapses and the time of their onset between CD70 nanoCAR and MGAT5 KO CD70 nanoCAR T cell treated mice in experiment B (Supplementary Figure 3). When we look at the Kaplan-Meier curves, we indeed observe a difference. The CD70 nanoCAR group seems to have more relapses with an earlier onset in time, leading to a shorter median tumor free survival time of 55 days as compared to the MGAT5 KO CD70 nanoCAR treated group in which the median tumor free survival time is 72 days.
For the secondary tumor, we defined three types of tumor control as no relapse of tumor growth was observed in any of the mice that cleared the secondary tumor. (1) Full control meaning the tumor never develops or becomes undetectable after an initial growth phase. (2) Partial control meaning the tumor stops growing but remains detectable. (3) No control of tumor growth throughout the duration of the experiment.
As is clear from Figure 3.B and the table in Figure 3.D, MGAT5 knockout CD70 nanoCAR T cell treatment also lead to better tumor control after a secondary challenge. While the majority of the mice show no or only partial control of the secondary tumor in the CAR treated groups (52.9% in total), this image is shifted in the MGAT5 KO CAR treated groups (35% in total). In the latter, the majority of the mice completely clear the tumor before the end of the experiment (64% of the mice in total). In the mice that did not clear the secondary tumor completely, the majority of mice treated with the MGAT5 knockout CD70 nanoCAR T cells experienced partial control (21%) while the majority of mice treated with wild type CD70 nanoCAR T cells showed no control at all (47%).
Treatment of tumor-bearing mice with MGAT5 KO CAR T cells leads to a better control of tumor growth rate
To evaluate differences in tumor growth or resolution between the treated mice, a piecewise linear mixed model (with interactions) was fitted (see Supplementary Figure 4 to Supplementary Figure 8) that allows to model the mean tumor volumes in each group. For these analyses, we made a distinction between Experiment A and Experiment B. The main reason for this is that the model would become unnecessarily complex because the timescales (design) of both experiments differ slightly as do the times at which the mice start to respond to the CAR T cell therapy. The latter is possibly due to inherent differences between the CAR T-cell batches (i.e. a donor effect).
In experiment A, we did not observe a difference in the speed of primary tumor resolution in mice treated with CD70 nanoCAR T cells and those treated with MGAT5 KO CD70 nanoCAR T cells (Supplementary Figure 4). As even non-glyco-engineered CAR T cells already cleared the primary tumor in a very short time span, there was not much scope for improvement. However, when we look at the response to treatment in the secondary tumor (Supplementary Figure 5), differences were observed. While the secondary tumor in the untreated mice grows at 12% (95% CI +9% to +16%) per day (which is consistent with the primary tumor growth), the average growth rate in the CD70 nanoCAR T cell treated group is slower, only 3% per day (95% CI: −15% to +26%). When we look at the MGAT5 KO CD70 nanoCAR T cell treated group, the secondary tumor actually decreases with 10% (95% CI: −24% to +7%) each day, indicating that MGAT5 KO CD70 nanoCAR T cells control tumor growth more efficiently in the secondary phase. However, due to the highly variable responses of individual mice in these groups, the difference in tumor growth rate between the MGAT5 KO CD70 nanoCAR T cell and nanoCAR T cell treated groups is therefore not statistically significant (ratio of growth rates: 95% CI = 0.67 to 1.13, adj. p-value = 0.505).
In experiment B, we do observe a difference in primary tumor clearance when we compare CD70 nanoCAR T cell treated mice with those that received MGAT5 KO CD70 nanoCAR T cells (Supplementary Figure 6). The primary tumor loses about half of its volume (49% with 95% CI: −53% to −45%) each day between day 22 and 33 compared to 29% (95% CI: −41% to −24%) each day for the CD70 nanoCAR T cell group. This difference is statistically significant (tumor shrinkage rate in CD70 nanoCAR T cell treated group is only 72% of shrinkage rate in MGAT5 KO CD70 nanoCAR T cell treated group with a 95% CI of 0.59 to 0.86, adj. p-value <0.001). Moreover, while the primary tumor completely disappears in all mice in the MGAT5 KO CD70 nanoCAR T cell treated group in the subsequent part of the experiment (day 33 to about day 84), this is not the case for the CD70 nanoCAR cell treated group, where the tumor volume remaining at day 60 is 5.07 mm3 (95% CI: 0.32 to 80.07) on average. The confidence interval is quite wide, probably due to the large spread in the CD70 nanoCAR T cell treated group, were some mice clear the tumor completely, some partially and some not at all. Unlike what was observed in experiment A, some of the primary tumors did regrow in the course of the experiment B. From the analysis of these relapsed primary tumors shown in Supplementary Figure 7, it is clear that, although the tumor growth rate is the same, the tumor volume is significantly lower in the MGAT5 KO CD70 nanoCAR T cell treated group, as compared to the CD70 nanoCAR T cell treated mice.
The response of CD70 nanoCAR T cell therapy on a secondary tumor challenge in experiment B is summarized in Supplementary Figure 8. From day 101 onwards, we see that the tumor size in the MGAT5 KO CD70 nanoCAR T cell treated group decreases with 10% each day (95% CI: −34% to +23%), while the tumor size in the untreated and CD70 nanoCAR T cell treated groups increases with 9% (95% CI: +1% to +19%) and 6% (95% CI: −10% to +25%) daily respectively, again indicating that MGAT5 KO CD70 nanoCAR T cells lead to a better tumor control after a secondary challenge. However, the difference in growth rate between CD70 nanoCAR T cell treated mice and MGAT5 KO CD70 nanoCAR T cell treated mice is not statistically significant (growth speed in MGAT5 KO CD70 nanoCAR T treated mice is 0.85 times the growth rate in CD70 nanoCAR T cell treated mice, 95% CI 0.59 to 1.22). This is most probably due to the large variability and relatively few available mice within the treatment groups.
MGAT5 KO CD70 nanoCAR T cells are present in higher numbers than CD70 nanoCAR T in peripheral blood and spleen following tumor control
End-point analysis on day 34 was performed on peripheral blood (both experiments) and spleen (experiment A) by flow-cytometry (Figure 4-A,B,F). Human CD3+ T cells were detected in blood and spleen of mice treated with mock Cas9 or MGAT5 KO CD70 nanoCAR T cells and around 75% of these cells were found to be CD70 nanoCAR T cells (data not shown), based on GFP expression. The number of MGAT5 KO CD70 nanoCAR T cells in the spleen (Figure 4-B) and blood (Figure 4-F) is markedly increased as compared to mock Cas9 CD70 nanoCAR T cells. We did not analyze splenocytes on day 34 in experiment B, since we kept all mice for rechallenge, enabling statistics on larger groups. CAR T cells were still present in the blood at day 80 (Figure 4-C,G). We see a trend of higher numbers of MGAT5 KO CD70 nanoCAR T cells compared to CD70 nanoCAR T cells, mostly pronounced in experiment B, however, the difference is not statistically significant. End-point analysis between day 118 and day 123 was performed on peripheral blood and spleen (Figure 4-E,I). Similarly, in both experiments we measure higher numbers of MGAT5 KO CD70 nanoCAR T cells compared to CD70 nanoCAR T cells, however, the difference is not statistically significant.
Discussion
In this paper, we described the impact of cell surface glycosylation alterations on T cell fate and functions through MGAT5 KO induced in CD70 nanoCAR T cells. As the alteration of cellular glycosylation has an impact on multiple cell surface receptors and their signal transduction, we measured the integrated results of all of these alterations on cellular behavior, both in vitro and in vivo.
β1,6-N-acetylglucosaminyltransferase-V (MGAT5) is the enzyme responsible for the initiation of GlcNAc-β- (1,6)-branching on N-glycans and is involved in multiple aspects of T cell activation. β-(1,6)-N-glycan branching leads to an increase in LacNAc modifications, the ligand of Galectins. It has been demonstrated that absence of Mgat5 and thus a decrease in LacNAc, lowers T cell activation thresholds in vitro by enhancing TCR clustering due to the absence of Galectin-glycoprotein lattice formation22,23. This Galectin-mediated lattice is responsible for holding CD45 and the TCR signaling complex in close proximity via their O- and N-linked glycans (respectively) to prevent low-avidity T cell activation24. Greco et al. recently demonstrated, by knockout of Mgat5 in pancreatic adenocarcinoma, that N-glycans protect tumors from CAR T cell killing by interfering with proper immunological synapse formation and reducing transcriptional activation, cytokine production and cytotoxicity25. It is known MGAT5 is a primary target of the Golgi-resident intramembrane protease Signal peptide peptidase-like 3 (SPPL3)26. Along the same line as described by Greco et al., Heard and colleagues identified expression of SPPL3 in malignant B cells as a potent regulator of resistance to CAR therapy 27.
An increased incidence of autoimmune disease is seen in the absence of Mgat5 in vivo19. Furthermore, negative regulation of TCR signaling by β1,6-GlcNAc-containing N-glycans promotes development of Th2 over Th1 responses, enhances Th2 polarization, and suggests a mechanism for the increased autoimmune disease susceptibility observed in Mgat5−/− mice28. On the other hand, Mgat5 expression can be induced by the anti-inflammatory cytokine IL-10, decreasing antigen sensitivity of CD8+ T cells during chronic infection29.
Our results indicate that MGAT5 KO CD70 nanoCAR T and CD70 nanoCAR T largely behave in the same way as control cells in vitro; MGAT5 elimination had no clear impact on T cell activation or viability. Furthermore, anti-tumor cell responses by MGAT5 KO CD70 nanoCAR T cells were maintained in vitro and very interestingly, our results indicate that MGAT5 KO nanoCAR T cells show enhanced anti-tumor potential and control upon a primary and secondary tumor challenge, as compared to mock engineered CAR T cells. In the case of very potent CAR T cells (cfr in experiment A), the improved protective effect of MGAT5 KO CD70 nanoCAR T cells over CD70 nanoCAR T cells seems to be more pronounced upon rechallenge, however, when CAR T cells are less potent (cfr in experiment B), enhanced capability seems to be more explicit in clearance of the primary tumor. Notably, increased numbers of MGAT5 KO CD70 nanoCAR T cells were observed upon specific antigen recognition, both in vitro and in vivo.
It was previously shown that the inhibition of binding to LacNAc glycans via competitive inhibition with carbohydrate analogs increased the number of infiltrating tumor-specific T cells30. In a recent study by Ye et al.31, MGAT5 was discovered as one of the top hits in a CRISPR screen in murine CD8+ T cells in a syngeneic model of glioblastoma in immunocompetent mice. MGAT5 knockout enhanced the efficacy of adoptive T cell transfer against glioblastoma in mice with both immunocompetent and antigen-specific transgenic TCR models in terms of increased tumor infiltration and overall survival of tumor bearing mice.
A possible explanation for the higher numbers of MGAT5 KO CD70 nanoCAR T cells compared to control CD70 nanoCAR T cells, could be that MGAT5 KO CAR T cells are less susceptible to Galectin-3-mediated apoptosis. We already confirmed Galectin-3 overexpression by the tumor cell lines used in our models and we are currently evaluating whether Galectin-3 binding to MGAT5 KO (CAR) T cells is indeed reduced. To capture the transcriptional programs that are differentially regulated between glyco-engineered and wild type CAR T cells, we will perform transcriptome profiling by bulk mRNA sequencing on cells that are cultured in the absence and presence of antigen expressing cells. Gene set enrichment and pathway analyses can then reveal a signature of gene upregulation or downregulation specific to knockout cell populations32. Results of this experiment are expected soon.
Taken together, it is clear from our data that disruption of N-glycosylation modifications on CAR T cells can have a major impact on their antitumor efficacy, and thus might have important implications for future design of cell-based immunotherapies.
Materials and methods
Ethical approval
All experiments were approved and performed according to the guidelines of the ethical committee Medical Ethics of Ghent University, Belgium.
The breeding of NSG mice is covered by file E-726 and animal experiments are covered by file EC2020-009.
Cell lines
THP-1 cells were obtained from ATCC and cultured in RPMI medium (Gibco) supplemented with 10% fetal calf serum (FCS), 0.03% L-Gln, 0.4 mM sodium pyruvate and 50 µM β-mercaptoethanol. SKOV-3 cells expressing luciferase were kindly provided by Prof. De Wever (Ghent University, Faculty of Medicine and Health Sciences) and were cultured in DMEM medium (Gibco) supplemented with 10% FCS and 1% penicillin/streptomycin. Jurkat cells were obtained from ATCC and were cultured in RPMI medium (Gibco) supplemented with 10% FCS, 2mM L-Gln and 0.4 mM sodium pyruvate. All cell lines were maintained in a 37°C, 5% CO2, fully humidified incubator and passaged twice weekly.
Human CD3+ T cell isolation and culturing
Leukocyte-enriched buffy coat samples were obtained from healthy donors attending the Red Cross center after informed consent and ethical committee approval (EC2019-1083). Peripheral blood lymphocytes were prepared by a Ficoll-Paque density centrifugation as described in the instruction manual for Leucosep™ (Greiner bio-one). CD3+ T cells were isolated by negative selection with antibodies against CD14, CD15, CD16, CD19, CD36, CD56, CD123 and CD235 (MojoSort™ Human CD3 T cell selection kit, Biolegend) according to the manufacturer’s protocol. Cells were cultured in IMDM + Glutamax medium (Gibco-BRL) supplemented with 10% heat-inactivated FCS and stimulated with Immunocult™ Human CD3/CD28 T cell Activator (Stemcell Technologies) (25 µL/ 106 cells) for 3 days at 37°C in the presence of 10 ng/ mL IL-12 (Biolegend).
Prior to cell seeding, cells were washed twice with PBS before putting them in culture with rhIL-7 at 10ng/mL (Miltenyi) and rhIL-15 at 10 ng/mL (Miltenyi). Cytokines and medium were replaced every 2-3 days. Cell densities were maintained between 1 × 106 and 3 × 106 cells/ mL.
Guide RNA
We designed gRNAs using the Synthego design tool (https://www.synthego.com/products/bioinformatics/crispr-design-tool). Guides were ordered as chemically modified synthetic sgRNAs (with 2’O-Methyl at 3 first and 3 last bases and 3’ phosphorothioate bonds between first 3 and last 2 bases) and reconstituted at 100 µM in TE buffer. An overview of the guides used in this study can be consulted in Supplementary Figure 2.A.
RNP electroporation
Recombinant Cas9-GFP protein was purchased from the VIB protein core (https://vib.be/labs/vib-protein-core). Cas9 RNP was made by incubating Cas9 protein with sgRNA at a molar ratio of 1:2 at 37°C for 15 min immediately prior to electroporation in T cells. Electroporation was performed using the Lonza Amaxa 4D Nucleofector X unit (Program EH-115) and the P3 primary cell kit with the following conditions: 1 × 106 cells/20 µL P3 buffer per cuvette (16-well strips) with 20 µM Cas9-RNP. Following nucleofection, 80 µL pre-warmed medium was added per well and cells were allowed to rest for 30 mins at 37°C, 5% CO2.
Analysis of genome editing efficiency
0.1 × 106 cells were collected and lysed in QuickExtract™ (Lucigen Epicentre) according to the supplier’s instructions. The target site was amplified by PCR and Sanger Sequenced. Sequencing data was analyzed with the ICE tool (Inference of CRISPR Edits, Synthego) to infer the percentage of insertions and deletions (INDEL score) and the percentage of insertions and deletions that are out of frame (knock out (KO) score)33.
Production of retroviral vectors
Retroviral constructs encoding the nanoCAR sequences were previously cloned in the LZRS-IRES-eGFP vector and were obtained from Prof. Dr. Bart Vandekerckhove (Department of Diagnostic Sciences, Ghent University, 9000 Ghent, Belgium). Viral particles were produced using standard Ca3(PO4)2 transfection of the Phoenix ampho packaging cell line. Retroviral supernatant was collected at day 14 after transfection and puromycin selection and kept at −80°C until use.
Generation of CD70 nanoCAR Expressing Human T cells
Immunocult-stimulated human CD3+ T cells were retrovirally transduced on Retronectin-coated plates (TaKaRa). 500 µL of cells per well at 0.5 × 106 cells/mL were supplemented with 0.5 mL retroviral supernatant and centrifuged for 90 minutes at 900 g at 32°C. Transduced cells were detected by eGFP expression or by an anti-VHH antibody (Genscript) directed against the nanobody constituting the extracellular domain of the CAR and analyzed by flow cytometry.
Lectin-based flow cytometry
For the evaluation of the poly-LacNAc content on the cell surface, we used the lectin from Datura stramonium at a staining concentration of 10 µg/ mL (Biotinylated DSL, Vector laboratories, B-1185-2). 2 × 105 cells per condition were collected and rinsed three times with PBS. Cells were incubated with fixable viability dye eFl780 (eBioscience) and biotinylated lectin in lectin binding buffer (PBS with 0.1 mM CaCl2) for 30 minutes at 4°C. After rinsing with lectin binding buffer, cells were incubated with PE-coupled neutravidin (Invitrogen, 5 µg/mL) for 30 minutes at 4°C. After rinsing the cells with PBS, samples were resuspended and analyzed by flow cytometry. A minimum of 50 000 events was recorded.
PNGaseF digest
In order to prepare cell surface N-glycans for DSA-FACE analysis, 1 × 106 cells were collected per condition and washed three times with PBS to reduce the presence of medium-derived glycans. Cell culture medium was collected for N-glycan labeling. PNGaseF digest (0.125 IU/ 1×106 cells, in-house production) was performed in 25 µL final volume in PBS for 2 hours at 37°C. Cells were removed by centrifugation (5 min at 300g) and the supernatant was subjected to another centrifugation step (15 min at 15 000 rpm) to remove cell debris. The remaining liquid portion of the sample was stored at −20°C until APTS labeling and DSA-FACE analysis.
N-glycan analysis using DSA-FACE
The remaining N-glycan samples were labelled by adding an equal volume (20 µL) of labeling mix consisting of a 1/1 v/v mix of 1M morpholine borane in 20% DMSO, 20% SDS and 4M Urea mixed with 350 mM APTS in 2.4M citric acid and 14% SDS immediately prior to labeling. The labeling reaction was incubated at 70°C for 1 hour and allowed to cool down at 4°C before purification. Size exclusion chromatography (Sephadex G-10 resin with an exclusion limit of 700 Da prepared in a 96-well setup in Multiscreen-Durapore plates) was performed twice to desalt the samples and to remove free unreacted APTS34. The labeled glycans were then dried in a speedvac.
Purified labelled and dried N-glycans were resuspended in 10 µL ultrapure water and analyzed with capillary electrophoresis on an eight-capillary DNA sequencer (Applied Biosystems 3500 Genetic analyzer). A proprietary internal standard (GlyXera) was added to the samples to be able to align profiles from different samples. Samples were injected on a 50 cm capillary at 15 kV for 10 seconds, using POP7 polymer and 100 mM TAPS, pH 8,0, containing 1 mM EDTA as the running buffer. N-glycan profiles were analyzed through the Genemapper 6 software.
Flow cytometry analysis
Flow cytometry analysis was performed on 0.2 × 106 cells per sample collected in a 96-well V bottom plate. Cells were rinsed with FACS buffer (PBS containing 0.5% BSA and 2mM EDTA) for 3 min at 300g and incubated with Fc Receptor Blocking solution (Human TruStain FcX™, Biolegend) for 10 minutes prior to cell surface staining with fluorescently labeled antibodies in Brilliant Stain buffer (BD Biosciences) for 30 minutes at 4°C.
For human CD3+ T cell phenotyping, cells were labeled with fluorescent antibodies against human CD8, CD62L and CD45RA (BD Biosciences) and CD3, CD4, CD25, CD69, and CD279 (PD-1) (Biolegend). A Fixable dye eFluor™ 780 (eBioscience) was used to evaluate live/dead cells.
Flow cytometer calibration was performed using CS&T beads (BD Biosciences). The gating strategy was set based on fluorescence minus one (FMO) controls and retained for all samples. Jurkat, THP-1 and SKOV-3 cell lines and primary human T cells were labeled with fluorescent antibody against human CD70 or isotype control (Biolegend) to verify antigen expression as described before35.
Galectin expression by tumor cell lines was evaluated by cell surface staining with a fluorescent antibody against Galectin-3 and an antibody against Galectin-1. The latter was detected by a fluorescent anti-goat antibody. As a positive control, cells were incubated with 200 µg/mL recombinant Galectin-1 and Galectin-3 (Biolegend). Galectin binding was competitively inhibited by adding 50 mM lactose during the staining procedure.
In all analyses, following doublet exclusion, live cells were identified using a fixable viability dye (Molecular Probes, Life Technologies). Data were acquired on a BD Symphony A5 equipped with five lasers (355, 405, 488, 561, 640nm) (BD Biosciences) and analyzed using FlowJo software (Tree Star, Ashland, OR).
In vitro analysis of cytokine production
Glyco-engineered CD70 nanoCAR T cells were stimulated in vitro by co-incubation with THP-1 or SKOV-3 tumor cell lines expressing CD70 in a 96-well plate in duplicate. After 1 hour of co-incubation, BD GolgiPlug (BD Biosciences) was added and after an additional 15 hours of stimulation, the cells were harvested, labelled with fluorescent antibodies against CD3, CD4 and CD8, fixed and permeabilized (eBioscience) and labelled for intracellular expression cytokines with fluorescent antibodies against TNF-α (BD Biosciences), IFN-γ and IL-2 (Biolegend). Samples were analyzed by flow cytometry as described above.
In vitro analysis of tumor cell killing
Glyco-engineered CD70 nanoCAR T cells were incubated with 2 × 104 THP-1 cells at different effector/target ratios (0; 0.0015; 0.015 and 0.15) in IMDM medium with Glutamax (Gibco) containing 10% FCS and 1% penicillin/streptomycin. Cells were labelled with fluorescent antibodies against CD3, CD4 and CD8 for the analysis of T cells and CD33 for the analysis of THP-1 cells at the start of the co-culture (day 0) and at day 3, 7, 10 and 14. At day 7 of co-culture, 2 × 104 THP-1 cells were added to the remaining wells. Cell numbers were determined by flow cytometry.
In vivo analysis of glyco-engineered CD70 nanoCAR T cell efficacy
NSG mice (breeder pairs obtained from The Jackson Laboratory, breeding in house) between 8-12 weeks of age were subcutaneously (in the flank) injected with 2 × 106 SKOV-3 cells in 50 µL PBS. The cells were allowed to form a solid mass tumor and CD70 nanoCAR T cells were intravenously injected on day 13 (in 200 µL total volume in PBS). Body weight and tumor progression was followed up by caliper and bioluminescence imaging (BLI). A dose of 150 mg/kg D-luciferin potassium salt (Perkin Elmer) was injected intraperitoneally 10 minutes before BLI. Imaging data were analyzed using Living Image Software and reported as photons/second. In experiment A, we started with 6 mice in the PBS and NTC group and 12 mice in the CD70 nanoCAR mock Cas9/MGAT5KO groups. At day 34, all control mice and 6 mice from the CD70 nanoCAR mock Cas9/MGAT5KO groups were sacrificed for analysis. The other 6 mice from the CD70 nanoCAR mock Cas9/MGAT5KO groups were kept for rechallenge. In experiment B, we started with 6 mice in the PBS and NTC group and 9 mice in the CD70 nanoCAR mock Cas9/MGAT5KO groups. All mice were kept for rechallenge, unless humane endpoint was reached (control mice). At the start of each experiment, we also kept a group of 8 (experiment A) or 6 (experiment B) mice to be used as a control (=PBS) group in the rechallenge phase of the experiment, to ensure age-matched controls.
End-point analysis on spleen and blood
At indicated time points, mice were euthanized. Peripheral blood was collected following severing of the right atrium of the heart and transferred to EDTA coated Microvettes (Sarstedt). The volume of blood was determined and red blood cells were removed using ammonium-chloride-potassium (ACK) lysis buffer (Lonza) prior to antibody staining for flow cytometry analysis.
The spleen was collected and processed to a cell suspension through a 70 µM cell strainer. Erythrocytes were removed using ACK lysis buffer followed by washes. Cells were counted prior to antibody staining for flow cytometry analysis.
Tumors were isolated from non-treated controls and fixed in 4% PFA. Subsequently, tumor tissue was embedded in paraffin for downstream immunohistochemistry analysis.
Immunohistochemistry and microscopic analysis of galectin expression in tumor tissue
Immunofluorescent staining was performed on 4 μm thick formalin-fixed, paraffin embedded (FFPE) sections of tumor samples from untreated mice. After antigen retrieval using citrate buffer pH 6 (Vector, H-3300), sections were incubated with 1% goat serum in PBS + 0.5% BSA + 0.1% Tween20 for 30 minutes to block non-specific binding. Subsequently, monoclonal rabbit anti-galectin-1 (1:200, Cell Signaling, 13888S) or monoclonal rabbit anti-galectin-3 (1:200, Cell Signaling, 87985S) diluted in 1% w/v goat serum in PBS + 0.5% BSA + 0.1% Tween20 were incubated at overnight at 4°C. Alexa Fluor 568 labelled goat anti-rabbit (1:500, Thermofisher, A11036), was incubated at room temperature for two hours. Counterstaining was performed using DAPI (1:1000). Slides were mounted using 1% n-propyl-gallate in glycerol (pH7). Images of the galectin staining were acquired with a LSM880 confocal microscope (Zeiss) and analyzed through ZEN Microscopy Software (Zeiss).
Statistical analysis in vitro experiments assessing tumor cell killing
To analyze the data of the coculture experiment, we started from flow cytometry-based count data. Since the counts had been normalized using counting beads, they were not necessarily integers so we rounded all to the closest integer. We considered each setup with the same donor, E/T Ratio and type of CAR T cells (i.e. CD70 nanoCAR or CD70 nanoCAR MGAT5 KO) as a cluster. Since we had two measurements (repeats) at each day and the measurements were performed at day 0, 4, 7, 11 and 14, this means we had 10 measurements in each cluster. Furthermore, we observed a slight rise in the counts of the NTC cells over time in the control setups. We corrected the CAR T cell counts for this background (per cluster and at each timepoint) by subtracting the mean background count from the measurements.
We analyzed the background-corrected counts with a generalized linear mixed model (GLMM) to allow for modeling the within-cluster correlation over time. Since the data showed considerable overdispersion, we used a negative binomial model. The GLMMadaptive package36 allows to fit such a model in R37 using adaptive Gaussian quadrature (AGQ). We did not have enough data to fit a random slope model, so we settled for a random intercept model of the following form: In this model, all fixed effects are coded as a factor. We chose to also model the time variable as a factor since the log-transformed counts are not linear with time. The model fit was evaluated using the DHARMa package38 and contrasts were estimated using the multcomp package39.
Statistical analyses in vivo experiments
Multinomial logistic regression
We analyzed the distributions of outcomes for the primary and secondary tumors in the different groups of mice. To do so, we first had to define several possible outcomes. For the primary tumor, there are four possible outcomes: - Full control of the tumor, meaning that the tumor becomes undetectable both by caliper measurement and on BLI, and also no relapse. - Full control of the tumor but with a relapse after a period of the tumor being undetectable. - Partial control, meaning that the tumor stops growing but remains detectable, all these mice also had a relapse. - No control of tumor meaning that the tumor continually keeps growing. For the secondary tumor we only have full control, partial control or no control. The follow up time was not long enough to also consider relapses. To analyze these data in R37, we used multinomial logistic regression (with a proportional odds assumption) as implemented in the polr function of the MASS package40. We analyzed the outcomes of experiments A and B together making experiment an additional predictor apart from group. Using likelihood ratio testing, we tested for an interaction effect between experiment and group and found that this was not significant in the primary nor secondary tumor. We used the multcomp package39 to calculate contrast estimates with 95% confidence intervals. We also used the ggpredict function from the ggeffects package41 to calculate experiment-wise predictions with 95% confidence intervals for the predicted outcomes.
Survival analysis (time to relapse)
To analyze the time to relapse, we first defined the start of follow up as the moment the primary tumor was controlled or partially controlled. We define control as the first day the tumor became completely undetectable on BLI and by caliper measurement. We define partial control as the first day a tumor (that never fully disappears) stopped increasing in size according to caliper measurements. Next, we define a relapse event as the moment a tumor starts growing again. We take the last day before the tumor has increased in size again or became detectable again as the onset of relapse. The time to event is then the time between start of follow up and a relapse event and the follow up time is the time between start of follow up and either an event or the end of follow up in case of no relapse. We used R37 with the survival42,43 and survminer44 packages to generate Kaplan-Meier plots with estimates of the median survival times and a corresponding risk and events table. Since relapses were only observed in experiment B, we ran a straightforward analysis with group as the only predictor (groups: CD70 nanoCAR or CD70 nanoCAR - MGAT5 KO). We tested for the difference in survival probability in these groups with a logrank test as implemented in the survival package.
Longitudinal analyses
Tumor volumes were measured by measuring the length and width of a tumor and using the length* width*width/2 (this is a half cube or cuboid) approximation of the volume of a sphere. The smallest tumor length/width that can be reliably measured with a caliper is about 0.5 cm. The minimal tumor volume that can be calculated in this way is 0.5*0.5*0.5/2 = 0.0625 cm3 (which can be regarded as the limit of quantification). We also cross-checked with BLI data for the small tumors, since this gives a better indication on whether there actually is still a tumor present or not. Whenever a small tumor was measured or a zero volume was registered, BLI was used to verify whether a tumor was actually present or not and the caliper measurements were adapted accordingly: when no tumor was found on BLI, we set small caliper measurements to zero and when a tumor was found on BLI but not measured by caliper, we set the tumor volume to 0.5. Uncontrolled tumor growth is exponential so we log-transformed (with a base 2 log) all tumor volume data to simplify the mean structures of the fitted models and to correct for the mean-variance relationship we observed during data exploration. To avoid problems when the tumor volume is zero, we first added 0.0625 (the detection limit) to all volumes before log-transforming. We then analyzed the transformed data of each experiment (A and B) and each phase (primary tumor before and after rechallenge and secondary tumor) separately by fitting a linear mixed model to each using the lme4 package45 and the nlminb fitting algorithm from the optimx package46 in R37. Where needed, we used piece-wise linear models with up to two knots for the time variable to allow for changes in growth rate over time. Random effects included a per-mouse random intercept and one or more random slopes for the time variable to model within-mouse correlation over time. For each model, we started with mean and covariance structures that were as saturated as possible based on the available data. Pruning the models was done via likelihood ratio testing first using Residual maximum likelihood (REML) to test for the random effects and then maximum likelihood (ML) to test for fixed effects. The final models were fitted using REML. In all models, we observed residual heteroscedasticity, even with the log-transformed data, so we used robust covariance estimators from the clubSandwich package47 (vcovCR, type ‘CR0’) in conjunction with the multcomp package39 to calculate adjusted p-values and/or adjusted 95% confidence intervals for parameters and contrasts.
Declaration of interest statement
EDB, NF and NC are co-inventors on a PCT International Patent application (PCT/EP2022/086474) by the VIB and Ghent University, which incorporate discoveries and inventions described here. All the other authors declare no conflict of interest.
Tables
Supplementary information
Acknowledgements
NF and LM were staff scientists of VIB. EDB was a predoctoral fellow at FWO during the project and has currently a doctor-assistant mandate at UGhent. EP was a research associate of VIB, AVH and EW are research associates of UGhent. This work was supported by grants G050420N and G028220N of FWO Vlaanderen and by a Young Investigator Proof of Concept (YIPOC) grant of the Cancer Research Institute Ghent (CRIG). We are grateful to M. Goossens and L. De Pryck for help with the caliper/IVIS measurements, splenocyte preparations and collecting SKOV3 cells. We thank Prof. Dr. Y. Chen (Parker Institute for Cancer Immunotherapy Center at UCLA, Los Angeles, CA, USA) for intensive experimental training in the CAR T field. We thank the VIB Bioimaging core Ghent (https://vib.be/labs/vib-bioimaging-core-ghent) and VIB Flow Core (https://vib.be/labs/vib-flow-core-ghent) facilities for their services.