Antigen-Dependent Inducible T-Cell Reporter System for PET Imaging of Breast Cancer and Glioblastoma

Visual Abstract

T cells are engineered to respond to cancer cells expressing a specific protein target, inducing rapid cell division and clonal expansion within the tumor microenvironment and activating immune response to target cells via local secretion of cytokines, interleukins, and growth factors (Supplemental Fig. 1A; supplemental materials are available at http:// jnm.snmjournals.org) (4,5). Hundreds of clinical trials have been initiated globally to study CAR T cells, with 2 of the most popular targets being CD19 (6) and B-cell maturation antigen (seen in multiple myeloma) (7,8). Importantly, there were engineered T-cell clinical trials resulting in patient deaths due to off-target effects, which may have been mediated by recognition of normal lung (9) or cardiac (10,11) tissues. We currently do not have tools to detect the engineered T cells engaging antigens in off-target tissues in vivo before advanced tissue damage, which can be identified only via biopsy or autopsy. Noninvasive methods are therefore critical in evaluating the safety of preclinical CAR T therapies and CAR T clinical trials by providing surrogate real-time maps for patient-specific T-cell-target antigen interactions.
Current limitations in predicting CAR T safety in vivo are potentially addressed by the PET-compatible synthetic intramembrane proteolysis receptor (SNIPR) T cells described in this article. The SNIPR system has produced a powerful new class of chimeric receptors that bind to target surface antigens and induce transcription of exogenous reporter genes via release of a transcription factor domain by regulated intramembrane proteolysis (Supplemental Fig. 1B). Importantly, SNIPRs can be designed to have an identical single-chain variable fragment (scFv) domain as CARs, thereby providing a surrogate map for CAR-antigen interaction. The SNIPR contains the regulatory transmembrane domain of human Notch receptor but bears an extracellular antigen recognition domain (e.g., scFv) and an intracellular transcriptional activator domain (Gal4-VP64). When the SNIPR engages its target antigen on an opposing cell, intramembrane cleavage is induced, releasing the intracellular transcriptional domain Gal4 and allowing it to enter the nucleus to activate transcription of target genes (12,13).
Despite the great versatility of synNotch and SNIPR, their diagnostic potential has not yet been explored. Current PET approaches producing antigen-dependent signals are dominated by immuno-PET, whereby a monoclonal antibody is labeled with a radioisotope such as [ 89 Zr] (14)(15)(16). In the described SNIPR approach, PET signals also depend on the interaction between an antigen and its corresponding scFv but occur via T-cell-based overexpression of the HSV-TK reporter. In this study, we combined CAR and SNIPR technologies to develop a new T-cell-based molecular sensor that can image T cells engaged with their target antigens. On binding an antigen target, CAR produces rapid T-cell division within the tumor microenvironment, and SNIPR activates the overexpression of PET imaging reporter genes. We developed human epidermal growth factor receptor 2 (HER2) and epidermal growth factor receptor variant III (EGFRvIII)-specific SNIPR T cells that were successfully imaged in vivo on interaction with their corresponding antigenexpressing tumors. We also compared HER2-specific SNIPR T cells with the naked [ 89 Zr]-modified anti-HER2 monoclonal antibody ([ 89 Zr]trastuzumab) used in immuno-PET to validate the specificity of the cell-based method. This proof-of-concept study provides the foundation for applying the SNIPR PET reporter to high-sensitivity cell-based antigen detection as well as mapping of engineered T-cell-antigen interactions in vivo.

MATERIALS AND METHODS
The supplemental materials provide detailed information on molecular biology, radiosynthesis, and several imaging studies not reported in the main text.
Reporter Design. The reporter constructs used are fully described in the supplemental materials. All reporter constructs were cloned into either a modified pHR'SIN:CSW vector containing a Gal4-upstream activation sequence response element with CMV (Gal4UAS-RE-CMV) promoter followed by multiple-cloning-site pGK promoter and mCherry or a modified pHR'SIN:CSW vector containing a Gal4UAS-RE-CMV promoter followed by multiple-cloning-site 3-phosphoglycerate kinase (pGK)-promoter and mCitrine. The HSV-derived thymidine kinase with SR39 mutation and GFP fusion (HSV-TKSR39-GFP) construct was cloned from HSV-thymidine kinase-GFP fusion gene with eukaryotic translation elongation factor 1a promoter (cEF.TK-GFP) (plasmid 33308; Addgene), which was deposited by Pomper et al. (19) using site-directed mutagenesis as described in Supplemental Figure 2A. The HSV-TKSR39-T2A-interleukin 2 superkine (sIL2) construct was cloned from HSV-TKSR39-GFP using In-Fusion cloning (Takara Bio) after adding 6 C-terminal amino acids (EMGEAN) that were deleted in the original HSV-TK in cEF.TK-GFP, as shown in Supplemental Figure 2B.

Preparation of SNIPR T Cells
Primary Human T-Cell Isolation and Culture. A full description of T-cell isolation and culture is presented in the supplemental materials. Primary CD41 and CD81 T cells were isolated by negative selection (catalog nos. 15062 and 15063; STEMCELL Technologies). CD41 T cells and CD81 T cells were separated using a Biolegend MojoSort human CD4 T-cell isolation kit (catalog no. 480130; Biolegend) following the manufacturer's protocol. For experiments involving the induction of interleukin 2 (IL2) superkine, primary T cells were maintained in human T-cell medium supplemented with IL2 until experimentation, whereupon medium was replaced with medium without added IL2. Lentiviral Transduction of Human T Cells. Human T cells were transduced as described in the supplemental materials. Pantropic VSV-G pseudotyped lentivirus was produced via transfection of Lenti-X 293T cells (catalog no. 11131D; Clontech) with a pHR'SIN:CSW transgene expression vector and the viral packaging plasmids pCMVdR8.91 and pMD2.G using TransIT-Lenti (catalog no. MIR6606; Mirus). Primary T cells were thawed the same day, and after 24 h in culture, they were stimulated with human T-activator CD3/CD28 Dynabeads (catalog no. 11131D; Life Technologies) at a 1:3 cell-to-bead ratio. At 48 h, viral supernatant was harvested, and the primary T cells were exposed to the virus for 24 h. At day 5 after T-cell stimulation, the CD3/CD28 Dynabeads were removed, and the T cells were sorted for assays with a Beckton Dickinson (BD) FACSAria II.
In Vitro Fluorophore and Luciferase Reporter Assay. For in vitro SNIPR T-cell stimulations, 2 3 10 5 T cells were cocultured with 1 3 10 5 cancer cells and analyzed at 48-72 h for reporter expression. Production of fluorophores (GFP and cyan fluorescent protein) were assayed using flow cytometry with a BD LSR II, and the data were analyzed with FlowJo software (TreeStar). Production of firefly luciferase was assessed with the ONE-Glo luciferase assay system (catalog no. E6110; Promega), and production of nanoLuc luciferase was assessed with the Nano-Glo luciferase assay system (catalog no. N1110; Promega). Bioluminescence was measured with a FlexStation 3 (Molecular Devices).
In Vitro Radiotracer Uptake Assay. Radiosyntheses of 9-(4-18 Ffluoro-3-[hydroxymethyl]butyl)guanine ([ 18 F]FHBG) was performed using established techniques, summarized in Supplemental Figure 3 and the supplemental methods for radiosynthesis (20). For in vitro SNIPR T-cell stimulations, 1 3 10 6 T cells were cocultured with 5 3 10 5 cancer cells and analyzed at 48-72 h for radiotracer uptake. On the day of the radiotracer uptake experiment, T cells and cancer cells were resuspended and 74 kBq (2 mCi) of [ 18 F]FHBG were added to each well and incubated for 3 h at 37 C, 5% CO 2 . After washing, retained radiotracer activity was measured using a Hidex g-counter.
Reporter Assays with Varying Receptor Affinity and Abundance (Heat Map). Heat maps for reporter expression were generated as described in the supplemental materials, with SNIPR receptors of varying HER2 binding affinities (4D5-3, 4D5-5, 4D5-7, and 4D5-8 scFv, in order of increasing binding affinity) and cancer cells with varying amounts of surface HER2 expression (293T, MCF7, and SKBR3, in order of increasing HER2 expression level).

In Vivo Studies
Murine Models/Tumor Cohorts Studied. Both luciferase-based and PET reporter data were acquired, and 2 dual-xenograft models were studied. After determining the optimal time point using optical imaging, (8-10 d), PET imaging was performed at several time points, with killing of the animals to verify tissue tracer accumulation (g-counting) and to perform histology and antigen staining (21-23).
Optical Imaging. As shown in Supplemental Figure 2, luciferasebased studies were performed initially (21 d (20,24). After radiotracer administration, mice were anesthetized under isoflurane, transferred to an Inveon small-animal PET/CT system (Siemens), and imaged using a single static 25-min PET acquisition followed by a 10-min small-animal CT scan for attenuation correction and anatomic coregistration.
[ 18  On completion of imaging, the mice were killed, and biodistribution was analyzed. Harvested tissues were g-counted using a Hidex automatic g-counter.

Data Analysis and Statistical Methods
Fluorescence-activated cell sorting analysis data were processed using FlowJo (BD Biosciences). All data graphs are depicted with error bars corresponding to the SEM. All statistical analyses of in vitro data were performed using Microsoft Excel, programming language R (https://www.R-project.org/), and Prism software (version 7.0; Graph-Pad). Data were analyzed using 1-way ANOVA or unpaired 2-tailed Student t tests. Small-animal PET/ CT data were analyzed using the open-source software AMIDE (25), and percentage injected dose (%ID) per volume was used for quantitative comparison. A 95% CI was used to distinguish significant differences in all cases.

Synthesis and In Vitro Validation of PET-Compatible Anti-HER2-SNIPR T Cells
To investigate the SNIPR PET imaging approach, we started with one of the most extensively studied tumor antigens, HER2 (26). Following the published protocol by Roybal et al. (26), we transduced human CD41 T cells with 2 plasmids encoding a SNIPR containing an anti-HER2 scFv (4D5-8), and an inducible reporter. For anti-HER2 SNIPR, we used the SNIPR with an anti-HER2 scFv binding head, an optimized truncated CD8a hinge region, the human Notch1 transmembrane domain, the intracellular Notch2 juxtamembrane domain, and a transcriptional element composed of Gal4-VP64 ( Fig. 1A) (27). For the reporter, we used SR39 mutant herpes simplex virus-thymidine kinase (HSV1-sr39TK)-GFP fusion protein and enhanced firefly luciferase (fLuc) (19,28,29). In this study, we used the hyperactive mutant HSV1-sr39TK to maximize the detection sensitivity (Supplemental Fig. 2A) (28,30). When thymidine kinase (TK) reporter expression on SNIPR activation was being evaluated, HSV1-sr39TK-GFP fusion protein was applied instead for ready assessment using flow cytometry (Supplemental Fig. 1B). In all the subsequent radiotracer uptake experiments, however, HSV1-sr39TK was cloned with self-cleavage sequence T2A followed instead by IL2 superkine for a higher level of T-cell activation on SNIPR activation (Supplemental Fig. 1B).

Activated Anti-HER2-SNIPR T Cells Show Antigen-Dependent [ 18 F]FHBG Accumulation In Vitro
Next, we designed SNIPR T cells to secrete IL2 superkine on binding to target antigens (33). We created an inducible vector with HSV1-sr39TK and IL2 superkine, linked by T2A self-cleaving peptides (Supplemental Figs. 2B and 5A) (34). We induced the anti-HER2 SNIPR T cells by coculturing them with SKBR3 or MD468 cells for 48 h. [ 18 F]FHBG was added to the medium, and SNIPR T cells were incubated for 3 more hours (Supplemental Fig. 5B). After removing the supernatant and washing, residual intracellular [ 18 F]FHBG was measured using a g-counter. Anti-HER2 SNIPR T cells cocultured with SKBR3 cells accumulated over 20-fold higher radiotracer levels than did anti-HER2 SNIPR T cells cocultured with MD468 cells (n 5 3, P 5 0.013) (Supplemental Fig. 5C). Once again, we tested [ 18 F]FHBG radiotracer uptake in suboptimal conditions by using scFvs with lower binding affinities and cancer cells with a lower antigen abundance (n 5 3 per combination). As expected, [ 18 F]FHBG incorporation correlated with both SNIPR binding affinity and target antigen abundance (Supplemental Fig. 5D).

Imaging Using Anti-HER2 SNIPR PET Shows High Antigen Specificity In Vivo
Luciferase-Based Optical Imaging. Since IL2 superkine was not strong enough to induce T-cell survival and proliferation in vivo, we introduced anti-HER2 CAR to the SNIPR T cells to more strongly induce T-cell proliferation and survival as well as reporter (HSV1-sr39TK or fLuc) expression on antigen binding (Supplemental Fig.  6A). The fluorescence-activated cell sorting yield of T-cell transduction with 3 plasmids was in the acceptable range, about 10% (Supplemental Fig. 6B). In this system, anti-HER2 CAR is constitutively expressed but only activates T cells and induces proliferation on binding to its target antigen HER2. As a pilot experiment, we again generated similar-sized SKBR3 (HER21) and MB468 (HER2-) xenografts by subcutaneous injection, followed by anti-HER2 SNIPR T-cell injection and bioluminescence imaging for the next 21 d (Supplemental Fig. 6C). The luciferase-SNIPR T-cell signal was stronger within the HER21 tumor, which is maximized at day 9, likely reflecting active proliferation within the tumor microenvironment (145-fold, no P value) (Supplemental Figs. 6D and 6E). The signal, however, decreased over time afterward, likely secondary to minimal target-killing activity of CD4 and clearing of target cells (Supplemental Fig. 6F).
PET Imaging. On the basis of the luciferase data, we generated SNIPR T cells with constitutively expressed anti-HER2 SNIPR and anti-HER2 CAR and with conditionally expressed HSV-sr38TK-T2A-sIL2 (Fig. 1A). We first tested the efficacy of in vitro [ 18 F]FHBG uptake on SNIPR activation in the SNIPR-CAR system, following the same experimental scheme as for the SNIPR-only system in Supplemental Figure 5B. As expected, when induced by coculturing with SKBR3 (HER21) cells, the SNIPR T cells accumulated a 27-times higher amount of [ 18 F]FHBG than the SNIPR T cells cocultured with MD468 (HER2-) cells (n 5 8, P , 0.001) (Fig. 1B). We also chose day 8 to image the SNIPR T cells with anti-HER2 CAR and inducible HSV1-sr39TK-T2A-sIL2 in the HER21/HER22 xenograft model (Fig. 1C). Again, we generated similar-sized HER21 (SKBR3) and HER2-(MB468) xenografts by subcutaneous injection into NCG mice, 4 wk before the injection of SNIPR-CAR T cells with HSV-sr38TK-T2A-sIL2 reporter (n 5 7). [ 18 F]FHBG imaging was performed at 3, 6, 8, and 10 d by microPET/CT (mPET/CT) (35,36). Greater [ 18 F]FHBG uptake on the HER21 side was observed than on the HER22 side, indicating that the T cells localized around the HER21 xenograft (n 5 7) (Figs. 1D and 1E). There was an approximately 10-fold higher [ 18 F]FHBG uptake in the HER21 xenograft than in the HER22 xenograft (P , 0.001), and there was approximately 13-fold higher [ 18 F]FHBG uptake in the HER21 xenograft than in the shoulder muscle (P , 0.001), based on region-of-interest (ROI) analysis. In contrast, there was no significant difference in [ 18 F]FHBG signal between HER2-xenograft and background muscle (P . 0.05). As seen in the SNIPR-CAR system with luciferase reporter in Supplemental Figure 5E, [ 18 F]FHBG signal decreased 10 d after Tcell intravenous injection, at which time the tumors were harvested for ex vivo biodistribution analysis (Fig. 1F). The difference in PET signal between HER21 and HER2-xenografts was 5-fold, with a P value of 0.003 at day 10, and the difference in ex vivo radioactivity between HER21 and HER2-xenografts was 4.2-fold, with a P value of 0.036 (Figs. 1F and 1G). Images from the PET study showed marked radiotracer in the stomach, intestine, and gallbladder, consistent with the known biodistribution of [ 18 F]FHBG in wild-type animals due to hepatobiliary excretion, as shown in both humans and rodents (Fig. 2D) (37)(38)(39). This result was corroborated using tissue extraction and ex vivo g-counting at day 10 (Fig. 1G).

Comparing Anti-HER2 SNIPR PET with Anti-HER2 [ 89 Zr]Trastuzumab and [ 18 F]FDG
Trastuzumab has the same anti-HER2 scFv binding moiety as our SNIPR, thereby reflecting the affinity-based interaction of the same antigen-antibody pair (40). We used [ 89 Zr]trastuzumab (anti-HER2) PET imaging and [ 18 F]FDG PET in the same animal model as for the SNIPR-CAR system (n 5 4). Overall, different biodistributions of the 2 tracers were observed, consistent with distinct metabolism and excretion pathways (Fig. 3A). Both immuno-PET with [ 89 Zr]trastuzumab and SNIPR PET with [ 18 F]FHBG demonstrated statistically significant increased radiotracer enrichment in HER21 tumor compared with HER2-tumor (9.9-fold, with P , 0.001, and 9.3-fold, with P 5 0.002, respectively) (Fig. 3B). The relative radiotracer enrichment within HER21 tumor compared with HER22 tumor was not statistically significant between immuno-PET and SNIPR PET (P . 0.05) (Fig. 3C). Likewise, the relative radiotracer enrichment within HER21 tumor compared with background was also not statistically significant between immuno-PET and SNIPR PET (P . 0.05) (Fig. 2D). Imaging results using [ 89 Zr]trastuzumab were corroborated via ex vivo analysis of harvested tissues (Supplemental Fig. 7). Although not statistically significant, the trend of higher [ 18 F]FDG accumulation in HER2-tumor than in HER21 tumor on a %ID/cm 3 basis correlated with the higher growth rate of MD468 (HER2-) than of SKBR3 (HER21) that we observed both in vitro and in vivo (Figs. 3B and 3D).

SNIPR PET Can Be Extended to the Glioblastoma Antigen EGFRvIII
To demonstrate the feasibility of SNIPR PET in other cancers, we chose the glioblastoma-specific antigen EGFRvIII. We designed T cells to constitutively express SNIPR that targets EGFRvIII and to conditionally express CAR that targets the distinct antigen IL13 receptor a-2, analogous to our previous approach to targeting glioblastoma with cytotoxic CD8 T cells (12). At baseline, those T cells express only anti-EGFRvIII SNIPR, without CAR or reporter. When they recognize EGFRvIII, they express IL13-mutein (IL13m)-CAR, which strongly binds to the more widely expressed but less specific target IL13 receptor a-2, as well as reporter genes ( Fig. 2A). To demonstrate in vitro reporter activation on anti-EGFR-vIII SNIPR activation, we generated 3 different T cells with blue fluorescent protein (BFP), nano-luciferase (nLuc), and TKSR39 reporters. Those reporters were coexpressed with IL13m-CAR and then cleaved at the intervening T2A sequence to generate separate CAR and reporter proteins. As expected, T cells bearing anti-EGFR-vIII SNIPR receptor induced a significantly increased level of BFP and nLuc activity 48 h after coculturing with EGFRvIII1 U87 cells, compared with coculturing with EGFRvIII-U87 cells (BFP: n 5 4, 25-fold, P , 0.001; nLuc: n 5 4, P 5 0.002) (Fig. 2B). T cells bearing anti-EGFRvIII SNIPR receptor and inducible IL13m-CAR-T2A-TKSR39 demonstrated a significantly increased level of [ 18 F]FHBG uptake when cocultured with EGFRvIII1 U87 cells compared with the same T cells cocultured with EGFRvIII-U87 cells (n 5 8, 32.5-fold, P , 0.001). To demonstrate the potential for in vivo imaging, we again generated a mouse model with EGFRvIII1 U87 and EGFRvIII-U87 xenografts by subcutaneous injection, followed by SNIPR T-cell injection and PET imaging for the next 10 d (n 5 4) (Fig. 2C). ROI analysis demonstrated a significantly higher level of radiotracer within the EGFRvIII1 xenograft than within the EGFRvIII-xenograft, and this increase was maximized at day 8, based on per-volume radioactivity (%ID/ cm 3 ) (14-fold, P , 0.001) (Figs. 2D and 2E). As seen with the HER2 SNIPR-CAR system, the EGFRvIII SNIPR CAR system demonstrated a decrease in PET signal at day 10, likely secondary to low target killing activity of activated CD41 T cells (Fig.  2D). Again, EGFRvIII1 and EGFRvIII2 xenograft mice were killed at day 10 for ex vivo analysis, which confirmed significantly higher enrichment of radiotracer within the EGFRvIII1 xenograft than within the EGFRvIII-xenograft based on per-weight radioactivity (%ID/g) (5-fold excess, P , 0.001) (Fig. 2F).

DISCUSSION
We have developed an antigen-specific, PET-compatible SNIPR T-cell reporter system in response to the rapidly increasing interest in T-cell-mediated treatment of human tumors. This customizable synthetic receptor platform can provide an essential companion to CAR T-mediated treatment, in addition to the recent application of syn-Notch to oncologic challenges (12). Applying several rounds of synthetic biology, we engineered SNIPR T cells that are antigeninducible, as detected by GFP fluorescence, luciferase luminescence, and the HSV-TK PET reporter [ 18 F]FHBG, and can successfully be imaged in dual-xenograft animal models. We have demonstrated application of this imaging modality to 2 different tumor antigens: HER2 and EGFRvIII. This technology might be used as a companion biomarker for CAR T therapies, such as to characterize off-target effects or verify tumor engagement. As the technology matures, SNIPR PET might be used to detect and characterize molecular profiles of early cancers without biopsy. CONCLUSION We have developed an antigen-inducible T cell PET reporter system using the versatile SNIPR. This reporter system may be used in a variety of T cell treatments for imaging antigen engagement.  62/905,251, and 62/905,263). Kole Roybal is a cofounder of Arsenal Biosciences, as well as a consultant, science advisory board member, and stockholder; is an inventor on patents for synthetic Notch receptors (WO2016138034A1 and PRV/2016/62/333,106) and receives licensing fees and royalties; was a founding scientist/consultant for and stockholder in Cell Design Labs, now a Gilead Company; holds stock in Gilead; and is on the science advisory board of Ziopharm Oncology and an Advisor to Venrock. No other potential conflict of interest relevant to this article was reported.

KEY POINTS
QUESTION: Can an inducible, cell-based PET system image tumor antigens in vivo?
PERTINENT FINDINGS: An inducible PET reporter system can be coexpressed with CAR in therapeutic T cells to image HER2 and EGFRvIII.
IMPLICATIONS FOR PATIENT CARE: As cell-based therapies mature, SNIPR PET might be added to any therapeutic T cell to image antigen engagement.