Preclinical Development and Evaluation of Allogeneic CAR T Cells Targeting CD70 for the Treatment of Renal Cell Carcinoma

CD70 is highly expressed in renal cell carcinoma (RCC), with limited expression in normal tissue, making it an attractive CAR T target for an immunogenic solid tumor indication. Here we generated and characterized a panel of anti-CD70 scFv-based CAR T cells. Despite the expression of CD70 on T cells, production of CAR T from a subset of scFvs with potent in vitro activity was achieved. Expression of CD70 CARs masked CD70 detection in cis and provide protection from CD70 CAR T-mediated fratricide. Two distinct classes of CAR T cells were identified with differing memory phenotype, activation status, and cytotoxic activity. Epitope mapping revealed that the two classes of CARs bind unique regions of CD70. CD70 CAR T cells displayed robust antitumor activity against RCC cell lines and patient-derived xenograft mouse models. Tissue cross-reactivity studies identified membrane staining in lymphocytes, thus matching the known expression pattern of CD70. In a cynomolgus monkey CD3-CD70 bispecific toxicity study, expected findings related to T cell activation and elimination of CD70-expressing cells were observed, including cytokine release and loss of cellularity in lymphoid tissues. Lastly, highly functional CD70 allogeneic CAR T cells were produced at large scale through elimination of the T cell receptor by TALEN-based gene editing. Taken together, these efficacy and safety data support the evaluation of CD70 CAR T cells for the treatment of RCC and has led to the advancement of an allogeneic CD70 CAR T candidate into phase I clinical trials.


INTRODUCTION
Renal cell carcinoma (RCC) is an area of high unmet need and represents a substantial patient population, with 74,000 new cases and 15,000 deaths estimated in the US each year (1). New therapies for RCC, including PD-1-targeting agents and combinations, show promising initial response, but low CR (complete remission) rates of 6-16%, highlighting the need for additional treatment options (1-6).
RCC is a highly T cell infiltrated tumor type and thus may be amenable to a T cell-based therapy (2-4,7). Adoptive transfer of T cells expressing chimeric antigen receptors (CARs) is a promising therapy showing substantial benefit in hematologic malignancies, including the recent approval of CD19-and BCMA-targeting CAR T therapies (8)(9)(10). Approved CAR T therapies are based on use of a patient's own T cells (autologous). While highly effective in the clinic, this approach has challenges, including significant time for production and variability in performance of patient-derived cells (11,12). Allogeneic CAR T cell therapy, or "off-the-shelf" therapy is a next-generation CAR T modality that utilizes cells from healthy donors and may overcome many of these challenges by increasing both the number of patients treated and product activity and consistency (13)(14)(15)(16)(17)(18)(19)(20).
Allogeneic CAR T cells can be created by TALEN® gene-editing at the TRAC loci to avoid graft versus host disease (GvHD) and the CD52 loci to confer resistance to ALLO-647, an anti-CD52 antibody used as part of the conditioning regimen to extending the window of T cell engraftment (21).
To translate this approach for RCC treatment, public expression data were mined to identify targets with a high prevalence in RCC and low/absent expression in normal tissues.
CD70 was identified as a gene expressed in a high proportion of patients with RCC and has limited normal tissue expression in a small subset of activated lymphocytes, stromal cells of the 5 thymic medulla, and antigen presenting cell (APCs) (22-24). CD70 is the ligand for the T cell costimulatory receptor CD27, activation of which generally leads to increased T cell persistence and memory formation (25,26). CD70 is also aberrantly expressed at high levels in a variety of hematological malignancies and solid tumors (27,28). The role of CD70 in cancer is only partially elucidated. CD70 is thought to play an immunosuppressive role in solid tumors by possibly overstimulating CD27 in the absence of productive co-stimulation, leading to T cell apoptosis and immune escape (29,30).
Previous studies have reported the use of truncated CD27-based CARs to target CD70 (30-32). This study is the first to demonstrate in vitro and in vivo analysis of multiple single chain fragment variable (scFv)-based anti-CD70 CAR T cell clones and identify a potential clinical product. Despite expression of CD70 on activated T cells, CAR T generation was successful, and cells did not succumb to mass fratricide. Avoidance of fratricide may be due to cis masking of CD70 by CAR expression. Several selected candidates were tested for appropriate specificity in tissue-cross-reactivity assays and one candidate was shown to have an acceptable toxicity profile in a cynomolgus monkey study when formatted as a CD70/CD3 bispecific.
Allogeneic CD70 CAR T cells were generated from healthy donor T cells and manufactured successfully in a large-scale process allowing for the treatment of many patients with a single run.
The impressive preclinical efficacy and safety data of the lead candidate CD70 CAR T cells presented in this manuscript and the promise of next-generation allogeneic CAR T cell therapy support the clinical investigation of allogeneic CD70 CAR T cell for the treatment RCC.

Staining and quantification of CD70 by flow cytometry
Cells were stained with 1 or 10 μg/mL PE-conjugated clone 41D12, generated at Allogene based on published sequences (33). Antibody binding capacity was quantified using Quantibrite PE beads (BD Biosciences) following manufacturer's protocol. Samples were acquired on CytoFLEX flow cytometer (Beckman Coulter).

CAR characterization in Jurkat and primary T cells
Jurkat cells or primary T cells were transduced with CAR-containing lentiviral vector (LVV) as described in Supplemental Methods. Activation (CD69) was determined four days posttransfection by flow cytometry. TRAC, CD52, or CD70 were edited on Day 6 by electroporation with TALEN® mRNA (Cellectis & TriLink Biotechnologies). Cells were stained on Day 9 with anti-CD70, anti-CD25, and anti-4-1BB, and on Day 14 with anti-CD62L and anti-CD45RO.
Recombinant human IL-2 was added throughout T cell culture and TCRαβ-positive cells were depleted using TCRαβ isolation kit (Miltenyi Biotec) at the end of process. T cells were cryopreserved in 90% FBS/10% DMSO. 7

In vitro cytotoxicity assays
Luciferase-expressing 786-O, ACHN, and REH target cells were co-cultured with CAR + T cells at defined E:T ratios for 72 hours. Target viability was determined by ONE-Glo reagent (Promega). For long-term serial killing assays, target cells and CAR T cells were co-cultured as described above. Every three-or four-days cells were mixed and half the T cells were transferred to a new plate of target cells. Target cell viability in the spent plate was read out by ONE-Glo reagent. Additional details are provided in Supplemental Methods.

CD70 masking and protection assays
Luciferase-expressing ACHN cells were transduced with CAR LVVs and stained with CAR Fabs at 2.5 μg/mL followed by anti-His antibody at 1:50 dilution. CAR-expressing ACHN cells were subjected cytotoxicity assays as described above.

RCC xenograft models
All animal studies were performed under approval by the Allogene Therapeutics Institutional Animal Care and Use Committee (IACUC). 786-O cells were implanted subcutaneously with Matrigel at 5 × 10 6 cells/mouse. Tumors were measured twice weekly using calipers. ACHN-nucLucGFP cells were injected IV at 1 × 10 6 cells/mouse. Bioluminescence measurements were performed twice weekly using the IVIS Spectrum. NSG mice bearing patient derived RCC xenograft tumors were purchased from Jackson Laboratory. Tumor volume was measured twice weekly by calipers. CAR T cells generated as previously described were thawed and dosed via IV injection. Animals were euthanized when they exhibited disease model-specific endpoints.
Additional details are provided in Supplemental Methods.

Normal kidney staining and cytotoxicity
Fresh kidney tissues dissected into cortex and medulla (Biobank Online) were dissociated using multi-tissue dissociation kit 1 (Miltenyi). Cells were stained with CAR 23 or isotype scFv-Fc protein followed by anti-Fc-PE antibody. Dissociated kidney and CAR 23 cells were co-cultured at a 1:1 ratio for 24 hours in RPMI1640 plus 10% FBS. Cells were stained with anti-CD69 antibody and viability dye before acquisition.

Cynomolgus monkey study
CAR 23 and CD3 antibodies were reformatted into a bispecific antibody. Two cynomolgus monkeys (one male and one female), ~3 years old at initiation of dosing, were administered the bispecific at 30 μg/kg by IV injection on Day 1. On Day 8 the same animals were administered a dose of 100 μg/kg. Vehicle (PBS) was administered to the control group. Clinical observations, body weight measurements, qualitative food consumption, and body temperature measurements were collected prior to initiation of dosing (PID) and at additional specific timepoints. Samples were collected for hematology, coagulation, clinical chemistry, immunophenotyping, and cytokine analysis PID and/or at additional specific timepoints. Cytokine analysis was performed using the Milliplex MAP Non-Human Primate Cytokine Magnetic Bead Panel (Millipore).
Necropsy was performed at the end of the study for microscopic and macroscopic examination of tissues.

Statistical analysis
All statistics were performed in GraphPad Prism 8. Animal tumor volume or flux (log transformed) was matched into groups with equal means and variance. Each group was then randomly assigned a treatment. Average mean tumor volume or bioluminescence was analyzed 9 for all mice and statistics were performed on replicate measurements from the day of CAR T dosing until the day the animals in the control group were euthanized. For statistical analysis, tumor growth in the treatment groups was compared against the control group, using repeatedmeasures (RM) one-way analysis of variance (ANOVA) with either Dunnett's or Tukey's multiple comparison test. Gaussian distribution and sphericity were assumed.

Data availability statement
All data associated with this study are presented in the paper or Supplemental Material. Reagents can be provided by and at Allogene's sole discretion pending scientific review and completion of a material transfer agreement with Allogene.

CD70 expression on renal cell carcinoma cell lines and tumors
To identify suitable RCC tumor associated antigens (TAAs), RNA expression data from the TCGA and GTEX databases were analyzed. CD70 was found to be highly expressed in RCC and had minimal expression in normal tissues (Fig. 1A). CD70 RNA in RCC tumors displayed median expression slightly lower than that observed in ACHN cells and a maximum equivalent to 786-O cells (Fig. 1B). CD70 protein staining by IHC was localized to the cytoplasm and cell membrane, with strong/moderate staining in 69% of clear cell RCC (ccRCC; Fig. 1C and D). cell lines and primary patient samples (Fig.1E) with a median receptor density in patient samples of 7k/cell ( Figure 1F). Expression was also evaluated on T cells and was observed on activated but not resting cells (Fig 1G).

Unique classes of CD70 CAR T cells with distinct behaviors and cytolytic activity
Phage display and hybridoma anti-CD70 antibody generation was performed (34,35). Forty-six antibodies were selected based on binding and specificity by ELISA and flow cytometry. These antibodies were formatted into second generation 4-1BB CARs and screened to eliminate clones exhibiting autoactivation (target-independent CAR aggregation) (36). CD70 wildtype and CD70 knockout (KO) Jurkat cells transduced with CARs were used and activation in the absence of target was monitored. Eight strongly auto-activating CARs displayed high CD69 in the absence of target (CD70 KO Jurkats) and were removed from further evaluation ( Fig. 2A).
CARs were next transduced into primary T cells using lentiviral vectors. The majority of clones (92%) displayed transduction efficiency greater than 25% (  Table S1).
To evaluate CAR functionality a 3-day cytotoxicity assay was performed against cell lines with a range of CD70 expression. In general, class 1 CARs were highly effective and lysed target cells across the CD70 expression range, whereas only three class 2 CARs displayed target lysis >50% (Fig 2F, Supplementary Fig S1, Supplementary Table S1). Overall, CARs with lower levels of CD70 detection at the end of culture (<5%) had the greatest short-term cytotoxic potential/activity (Fig. 2G).
A long-term serial-killing assay was developed to evaluate CAR susceptibility to exhaustion following repeated target exposure. All CARs were able to kill target cells initially, but many lost activity after repeated re-challenge. Class 1 CARs performed well against all target cells. Class 2 CAR were highly active against CD70-high targets but performed poorly when CD70 expression was low (Fig. 2H).

CD70 CAR expression protects CAR T cells from fratricide
The relationship between CD70 detection and cytotoxic activity (Fig. 2G) prompted experiments to better understand the differences between class 1 and 2 CARs and to determine if lack of CD70 detection on class 1 CAR T cells is due to the killing of CD70 + cells in culture (fratricide) or prevention of detection through cis masking of CD70 by the CAR, obscuring the detection epitope (Fig. 3A). Jurkat cells were used because they are a CD4-derived cell line incapable of cytotoxicity and fratricide. CD70 detection on CAR Jurkat cells mirrored what was seen for primary CAR T cells, with the same subset of CARs being CD70 negative, thus the lack of CD70 detection must be due to masking ( Fig. 3B and C).
Next, affinities of clones were examined. The binding domains of CARs were reformatted as Fabs (since not all CARs could be expressed as soluble scFvs). Affinities for recombinant human CD70 protein ranged between 0.15nM-19nM for class 1 CARs and 1.4nM-47nM for class 2 CARs as determined by biosensor assays (Supplementary Fig. S2). No clear trend in affinity and CD70 detection or CAR T activity was observed.
CAR Fabs were also subjected to binning and epitope mapping. The two classes of CARs fit largely into two distinct bins. An outlier was CAR 24, which did not co-bind with any of the other Fabs and was designated subclass 2b ( Supplementary Fig. S2). To perform epitope mapping, point mutations of CD70 residues were generated and key residues to binding of each CAR Fab identified. All class 1 CARs targeted epitopes ranging from the apex to the side of CD70, while class 2 CARs targeted epitopes located at the bottom of CD70 (Fig. 3D, Supplementary Fig. S3).
Experiments were carried out to demonstrate the ability of cis masking to obscure CD70 detection. ACHN or T cells transduced with or without CARs were stained for CD70 with either class 1 or class 2 Fabs (Fig. 3E, Supplementary Fig. S4A-B). Class 1 Fabs were unable to stain cells expressing class 1 CARs but were able to stain cells expressing class 2 CARs and vice versa. The exception to this observation was the inability of any of the Fabs to stain cells expressing CAR 23, suggesting this CAR may have unique properties and fall into a subclass (designated 1b). CD70 was not detected on any cells by the CAR 24 Fab, potentially because this Fab has a fast dissociation rate ( Supplementary Fig. S2).
To examine biological consequences of masking, ACHN cells expressing the various CARs were subjected to cytotoxicity assays with primary CD70 CAR T cells. Expression of class 1 CARs by target cells was able to protect against lysis by class 1 CAR T cells, but not class 2 CAR T cells (Fig. 3F). The inverse was true for class 2 CARs (Fig. 3G). Despite the ability of CAR 23 to prevent detection by both class 1 and class 2 Fabs (Fig. 3E), it was only susceptible to killing by class 2 CAR T cells. CAR 24 continued to stand apart from other class 2 CARs in that it was only able to protect against killing by itself.

Effects of CD70 knockout (KO) on CAR T cell activity in vitro and in vivo
Despite the potential protective effects of cis masking in CD70 CAR T cells, interaction of CAR and untransduced CD70 + T cells in culture may still lead to overactivation and exhaustion. For were eliminated from further evaluation.

CD70 CAR T with rituximab-based off-switches
A CAR off-switch may be desired to eliminate CAR T cells in the case of unexpected adverse activity (37,38). A rituximab-based off-switch system was selected, as rituximab is a widely used anti-CD20 antibody and such a system can effectively modulate CAR activity (38-40). Three rituximab formats (R-formats) were evaluated with varying mimotope number and location ( were converted into R-formats and transduced in primary T cells. All R-format CAR T cells with transduction above 40% ( Fig. 5B; Supplementary Fig. S6A) were tested in cytotoxicity assays and were able to eliminate target cells effectively, although some differences in activity between formats were observed (Fig. 5C&D). Optimal R-format CARs were generally equal or better than their "naked" CAR counterparts and this was particularly true for CAR 23, which was significantly more effective both in vitro and in vivo in the QR3 format as compared to the "naked" CAR ( Additionally, rituximab dosed at the time of CAR T administration completely abrogated the CAR T anti-tumor activity in an ACHN xenograft model (Fig. 5H). Based on the above data, the three most active candidates, CARs 3, 17, and 23, were advanced into additional safety assessment.

Tissue Cross-Reactivity study identified minimal off-target binding for CARs 3 and 23
Initial screening of antibodies against a panel of CD70-negative cell lines by flow cytometry identified no off-target binding for CARs 3, 17, and 23 (Sup. Fig. S1B). A tissue cross reactivity (TCR) assay was also performed in which soluble binding domains from CARs 3, 17, and 23 were screened for binding against a panel of 36 normal tissues by immunohistochemistry.
Appropriate staining was observed on positive and negative control cell lines (Fig. 6A). All three CARs had membrane staining of leukocytes in lymph nodes and cytoplasmic staining of epithelial cells in the thymus. No additional staining was observed for CAR 3. Staining unique to CAR 17 included cytoplasmic staining in all tissues examined and it was de-prioritized given this broad non-specific staining. CAR 23 displayed cytoplasmic staining of kidney epithelial cells and this was not observed with either of the other CARs, suggestive that the staining is not ontarget binding to CD70. A follow-up GLP TCR study was performed with CARs 3 and 23 and findings were similar to those observed previously, with the exception of additional membrane and cytoplasmic staining of resident, migrating, infiltrating, and/or intravascular mononuclear cells (APCs and lymphocytes) in multiple tissues, including lymph node, gut and bronchialassociated lymphoid tissue, and cervix. Given that staining was not widespread and intracellular protein is not accessible to CAR T cells, staining of CAR 3 and CAR 23 was deemed acceptable.
To confirm the CAR 23 staining in normal kidney was limited to the intracellular compartment, normal kidney cells from three donors were evaluated by flow cytometry and subjected to cytotoxicity assays. Normal kidneys cells were negative for CD70 staining and did not induce CAR T activation or cytotoxicity (Fig. 6B-F). These data suggest that the CAR 23 staining is of low toxicologic risk. Following the second dose, animals showed signs of more severe CRS, including mild tremors, decreased activity and fever, and were euthanized. An extensive panel of normal tissues were collected and subjected to macroscopic and microscopic histopathologic analysis. There were no test article-related macroscopic findings. No microscopic findings were observed in the large majority of tissues examined ( Figure 7E). Key findings included decreased cellularity in lymphoid tissues, likely due to elimination of on-target CD70-positive lymphocytes. Additional findings included minimal to mild decreased zymogen granules in salivary glands and the pancreas, an increased myeloid to erythroid ratio that correlated to increased white blood cells, and decreased hemoglobin and platelets in one animal, which may be secondary to CRS and inflammation. Findings were expected based on the expression of CD70 on lymphocytes and the mechanism of action of a CD3 bispecific.

Allogeneic CD70 CAR T cells generated by TALEN® gene-editing at clinical-scale
Success of an allogeneic CAR T cell therapy relies on the ability to produce cells at large scale and treat many patients from a single production run. To verify that CD70 CAR T cells can be generated en masse, a large-scale manufacture process was developed and performed utilizing CAR 23. CAR T cells gene-edited to disrupt the TRAC and CD52 loci were generated from healthy donor PBMCs using an 18-day culture process (Fig. 8A). Cell viability remained high (greater than 95%) throughout the production and CAR T cells expanded more that 30-fold from day 8 to 18, resulting in greater than 4x10 10 cells at the end of the process (Fig. 8B). The percentage of CD52 and TCRαβ negative cells was high at the end of production (>60%; Fig.   8C). Remaining TCR-positive cells could contribute to GvHD and were thus purified out successfully using bead-based elimination, resulting in 98.4 % TCRαβ negativity after purification. The CD70 CAR-positive percentage was ~60% after purification (Fig. 8C). The activity of CD70 CAR 23 T cells produced in the large-scale process was similar to that of CAR T cells generated at small-scale when compared in an in vitro cytotoxicity assay (Fig. 8D-E).
These data demonstrate the ability to generate highly functional ALLO TM CAR T targeting CD70 at scale and support the use of allogeneic CAR T cells clinically.

DISCUSSION
In this study we found that CD70 has homogenous expression in a high percentage of RCC suggesting it could be an appealing CAR T target. In agreement with previous reports (22) we also found expression on activated T lymphocytes. Such expression might be expected to lead to CAR T cell fratricide, as has been reported for other tumor targets expressed on T cells such as CD38 and CD7 (45,46). Despite the potential for fratricide, a large number of scFv-based CD70 CARs were successfully transduced and CAR T cells generated.
Extensive in vitro analysis of the panel of potential candidate CD70 CAR T cells revealed two general classes based on detection of CD70, one in which CAR T cells displayed increased activation markers, a differentiated memory phenotype, and were highly potent in short-term cytotoxicity assays and the other in which CAR T cells were less activated, less differentiated, and generally displayed inferior cytotoxic activity. Both classes of CARs were capable of masking/protection from fratricide: overexpression of CD70 CARs on ACHN tumor cells was able to protect the cells from lysis by CARs recognizing the same epitope. CD70 was still detected on the surface of the ACHN-CD70 CAR tumor cells with antibodies from another epitope class, suggesting that CD70 is not downregulated on cells upon binding CARs in cis, but rather the CAR is "masking" CD70 and preventing binding. A similar phenomenon has been reported for CD19 and CD5 CARs (47,48). While not followed extensively in our studies, it is possible that CAR T candidates susceptible to fratricide were screened out early in our selection process by low transduction or expansion. Given that both classes of CARs were capable of masking, this phenomenon does not explain the difference between cytotoxic activity of the two classes of CARs.
The binding affinity of Fabs generated from the CARs was very similar between both classes and is likely not the reason for activity difference. A sandwiching assay revealed that the two classes of CARs generally fall into two epitope bins, and this was confirmed by epitope mapping, in which class 1 CARs bound to the membrane distal region of CD70. This region contains the predicted binding epitope for the detection reagent, thus explaining why CD70 is not detectable on class 1 CARs (33). Binding membrane distal epitopes may result in enhanced CAR spacing and sensitivity/potency, thus allowing class 1 CARs to recognize low levels of CD70 expression present on T cells during the CAR T production and leading to activation and differentiation. These CARs were indeed more potent in cytotoxicity assays. Despite generally falling into two classes, unique CARs within each class emerged in certain assays. CAR 23 prevented detection of CD70 by all Fabs when expressed on ACHN cells and this could be due to the specific binding epitope of CAR23 or not only masking, but downregulation of CD70 on the cell surface. CAR 23 ACHN cells were still able to be lysed by CARs of the opposite class, so available CD70 on these cells may be below the detection limit by flow cytometry but still high enough to induce cytotoxicity. The data presented here identify the existence of multiple subsets of CD70 CAR T cells, differing in respect to functionality and highlight the need for evaluating a large number of CAR clones to obtain an optimal candidate.
Despite the potential for CD70 expression on non-transduced cells to drive CAR T exhaustion, CD70 KO did not enhance CAR T expansion, in vitro cytotoxicity, or anti-tumor activity of CD70 CAR T cells against a model regarded as having the most physiologically relevant expression. Others have reported enhancements with CD70 KO (49) and we did find enhanced activity of one candidate in a high expression model, suggesting that the benefits of CD70 KO are not universal, but rather scFv and context-dependent. Rituximab off-switch formats were evaluated to mediate CAR T control and in some cases had the added benefit of enhancing CAR T activity, as in the case of CAR 23 in the QR3 format. Given previous published works on the importance of spacers to manipulate CAR T activity (50-52) it could be speculated that for different scFvs, one format of off-switch over another could create the ideal spacing necessary and this may vary by epitope.
To be effective for patients, CAR T cells must possess robust anti-tumor activity, but also demonstrate an appropriate safety profile though extensive toxicological evaluation. A tissue cross-reactivity (TCR) assay was utilized to evaluate binding of the scFvs for three of the most       Cells were incubated at a 1:1 E:T ratio for 24 hours. CAR 23 T cells did not kill dissociated kidney medulla E, or cortex F, cells in a 24-hour cytotoxicity assay.           days A c t iv a t io n T r a n s d u c t io n T A L E N ® g e n e -e d it in g T C R α c e ll d e p le t io n a n d c r y o p r e s e r v a t io n E x p a n s io n Allogeneic CAR T manufacturing process