Multiparameter Longitudinal Imaging of Immune Cell Activity in Chimeric Antigen Receptor T Cell and Checkpoint Blockade Therapies.

Longitudinal multimodal imaging presents unique opportunities for noninvasive surveillance and prediction of treatment response to cancer immunotherapy. In this work we first designed a novel granzyme B activated self-assembly small molecule, G-SNAT, for the assessment of cytotoxic T lymphocyte mediated cancer cell killing. G-SNAT was found to specifically detect the activity of granzyme B within the cytotoxic granules of activated T cells and engaged cancer cells in vitro. In lymphoma tumor-bearing mice, the retention of cyanine 5 labeled G-SNAT-Cy5 correlated to CAR T cell mediated granzyme B exocytosis and tumor eradication. In colorectal tumor-bearing transgenic mice with hematopoietic cells expressing firefly luciferase, longitudinal bioluminescence and fluorescence imaging revealed that after combination treatment of anti-PD-1 and anti-CTLA-4, the dynamics of immune cell trafficking, tumor infiltration, and cytotoxic activity predicted the therapeutic outcome before tumor shrinkage was evident. These results support further development of G-SNAT for imaging early immune response to checkpoint blockade and CAR T-cell therapy in patients and highlight the utility of multimodality imaging for improved mechanistic insights into cancer immunotherapy.

Introduction

Cancer immunotherapies, mainly immune checkpoint blockade and adoptive cell transfer (e.g., chimeric antigen receptor-CAR T-cell therapy), have significantly improved survival rates in many cancers by providing robust and sustained therapeutic effects.[1,2] Yet, many hurdles such as complex immune evasion mechanisms, dysfunction of T lymphocytes, and the immunosuppressive tumor microenvironment[3−11] limit treatment efficacy.[12,13] Additionally, life-threatening immune-related side effects are not uncommon.[14] Tremendous efforts are now being devoted to understanding and targeting these mechanisms for the development of more effective therapies with improved safety profiles. Noninvasive molecular imaging is a promising strategy for monitoring whole-body immune responses both during disease progression and upon treatment induction.[15]

Radiolabeled antibodies,[16−19] peptides,[20] or small molecules[21,22] that target surface markers or show uptake by specific immune cell populations have been extensively studied for imaging the immune response. The presence of specific tumor-infiltrating immune cells, for example, CD8+ cytotoxic T lymphocytes (CTLs), has been shown to correlate with a favorable response to checkpoint blockade therapy.[23−25] Nevertheless, several known immunotolerant mechanisms associated with the suppressive tumor microenvironment such as T cell anergy, exhaustion, or senescence may compromise the accuracy for predicting the therapeutic response solely based on the presence of cell subsets.[26,27] Alternatively, granule-mediated cytotoxicity by natural killer (NK) cells and CTLs has been explored for monitoring the immune activity responsible for tumor eradication.[28,29] Granule-mediated cytotoxicity, one of the dominant cytotoxic mechanisms used by immune effectors, involves the release of granzyme B (gzmB)-loaded granules to induce the lysis of target cells.[30] Secreted gzmB is intrinsically stable, representing an ideal biological target of cytotoxicity.[31] A number of imaging probes have been developed to study gzmB function in vitro(32,33) and image gzmB activity in vivo,[34−37] supporting gzmB as a useful biomarker of cytotoxic cell activation in anticancer responses.[28,38,39] However, these studies did not monitor the gzmB activity together with the immune cell trafficking and tumor infiltration. Given the highly complex and regulated nature of the immune system, sequential imaging of multiple events of immune response in the same animals can help reveal the complicated spatiotemporal dynamics of immune activity and uncover immune evasion mechanisms.

In this study, we first developed a sensitive fluorescent imaging probe for noninvasive surveillance of gzmB function. This probe, termed a gzmB sensitive nanoaggregation tracer (G-SNAT-Cy5), was designed based on the TESLA (target-enabled in situ ligand aggregation), a versatile probe platform built on the bioorthogonal condensation reaction between an aromatic nitrile and cysteine.[40−42] The TESLA platform has been demonstrated for the development of small-molecule imaging probes for multimodality imaging of proteolytic enzyme activity.[43−49] A TELSA probe is made of a scaffold containing an aromatic nitrile such as cyanoquinoline or pyrimidinecarbonitrile and a cysteine whose amino group is conjugated to a caging moiety—a peptide substrate of gzmB in the case of G-SNAT. Upon uncaging by the target, for example, gzmB, the cysteine group will react with the aromatic nitrile intermolecularly or intramolecularly to form macrocyclic products. Promoted by hydrophobic and π–π interactions, these products will form molecular aggregates at the nano scale, resulting in preferential retention at the target site. A cyanine 5 fluorophore (Cy5) labeled G-SNAT-Cy5 probe was validated for imaging the activity of gzmB both in vitro and in clinically relevant mouse models of CAR T-cell and checkpoint blockade therapies. We discovered that gzmB could maintain partial hydrolytic activity under acidic conditions within cytotoxic granules. G-SNAT-Cy5 not only detected gzmB activity within the CTLs but also reported the CAR T cell mediated cytotoxicity in vivo, a major advantage over conventional T cell-tracking techniques. Employing longitudinal bioluminescence imaging of transgenic mice with hematopoietic cells engineered to stably express firefly luciferase, we simultaneously determined the immune cell trafficking and tumor infiltration in response to immunotherapies. This combined imaging of immunological events revealed distinct whole-body patterns of immune cell migration and associated cytotoxicity among checkpoint inhibitors treated responder, nonresponder, and untreated groups and allowed earlier and reliable prediction of immunotherapeutic outcomes. These results highlight the value of multimodal longitudinal imaging for monitoring complex immune dynamics and their orchestration in response to immunotherapies.

Results

Design of gzmB-Sensitive Nanoaggregation Probes

In effective CAR T-cell or checkpoint blockade therapies (Figure A), CTLs extravasate, infiltrate, and engage cancer cells to establish immunological synapses which enable rapid, polarized release of cytotoxic granules containing effector proteins perforin and gzmB at the synaptic cleft (Figure B). Perforin, as named, directly perforates the target-cell plasma membrane and oligomerizes in a calcium-dependent manner into a conduit which allows the passive diffusion of gzmB to trigger programed cell death (Figure C). GzmB is a serine protease that cleaves an IEFD (Ile-Glu-Phe-Asp) peptide motif (preferred substrate in mice) to activate downstream caspase signaling and trigger DNA fragmentation and apoptosis (Figure C).[50−53] This unique mechanism in the cytotoxic immune response provides efficient delivery of active gzmB into target cells, in addition to an endocytosis-dependent mechanism that has also been proposed.[54,55]

Figure 1

Mechanism of in vivo imaging of gzmB activity in tumors by G-SNAT. (A) Treatment of a patient with checkpoint inhibitors or CAR T cells. (B) Effector T cells (blue) can extravasate and infiltrate tumors to kill cancer cells (yellow). (C) Production of gzmB and perforin by CTLs and their delivery across an immunological synapse into cancer cells. (D) Proposed gzmB and reduction-controlled conversion of G-SNAT into G-SNAT-cyclized through the bioorthogonal intramolecular cyclization, followed by self-assembly into nanoaggregates in situ. Blue square, the 2-cyanopyrimidine; brown and orange, amino and thiol groups of d-cysteine, respectively; yellow, thioethyl masking group; green, the capping peptide residues; pink, fluorophore Cy5. Part of this figure was created with BioRender.com.

Based on our highly modular TESLA, G-SNAT was designed (Supporting Information Figures S1 and S2) to uniquely enable the projection of specific catalytic activity of gzmB during immune response assessment to retained molecular aggregates. As illustrated in Figure D, G-SNAT contains (1) a 2-cyanopyrimidine group (blue), (2) a cysteine residue coupled to the IEFD substrate (green) at the amino group (brown) with a disulfide bond at the mercapto group (orange), and (3) a propargylglycine residue between pyrimidine and the cysteine for labeling with a fluorophore, radioisotope, or other contrast agents (pink). After hydrolysis and cleavage of IEFD by gzmB as well as the reduction of the disulfide bond by intracellular glutathione (GSH), the product cyclizes intramolecularly into macrocyclics which are rigid, hydrophobic, and susceptible to intermolecular interactions that trigger nanoaggregation in situ. For studies in vitro, both a Cy5 preconjugated G-SNAT-Cy5 probe (Supporting Information Figure S2) and post cell culture click reaction at the propargylglycine alkyne handle with an azido Cy5 were applied for pinpointing the aggregated G-SNAT in cells.

Macrocyclization and Nanoaggregation of G-SNAT In Vitro

Incubation of G-SNAT with recombinant mouse gzmB enzyme (1 μg/mL) and tris(2-carboxyethyl)phosphine (TCEP, 2 mM) to mimic the intracellular reducing environment at 37 °C overnight enabled the macrocyclization of G-SNAT (10 μM; retention time, TR = 14.2 min) to give G-SNAT-cyclized (TR = 8.8 min), as shown in high-performance liquid chromatography (HPLC) and confirmed by mass spectrometry (Figure A). Since IEFD has been reported to be a preferred substrate of mouse gzmB and its specificity has been validated,[35,56] we conducted a kinetic study with caspase 3 as a control, which was believed to share a pool of substrates with gzmB like the nuclear mitotic apparatus protein (NuMA, Val-Leu-Gly-Asp) and DNA-dependent protein kinase catalytic subunit (DNA-PKcs, Asp-Glu-Val-Asp).[57] It was shown that gzmB, not caspase 3, activated G-SNAT (10 μM) and converted it to cyclized product (Figure B and Supporting Information Figure S3). The assembled nanoaggregation was imaged with transmission electron microscopy (TEM), and the size distribution was acquired by dynamic light scattering (DLS) (Figure C and D). The average diameter of the aggregated nanostructures was 396 nm, ranging from 190 to 955 nm. No DLS signal could be detected when G-SNAT remained uncleaved by gzmB. The nanoaggregates were found nearly neutral by zeta potential analysis (Figure E). Together, these results demonstrate that gzmB specifically hydrolyzes IEFD and induces macrocyclization of the products leading to intermolecular interaction and nanoaggregation. The stability of G-SNAT-Cy5 in mouse serum was evaluated by HPLC and indicated a half-life of approximately 8 h (Figure F and Supporting Information Figure S4), which is sufficiently stable to capture serial immunological synapse activity that usually lasts less than 15 min each.

Figure 2

In vitro characterization of the G-SNAT probes. (A) HPLC traces of G-SNAT in gzmB assay buffer (black, TR = 14.2 min) and the incubation of G-SNAT (10 μM) with TCEP (blue, TR = 12.8 min) and recombinant mouse gzmB (1 μg/mL) overnight at 37 °C (red, TR = 8.9 min). (B) The enzymatic reaction kinetics and specificity studies by longitudinal monitoring of the percentage conversion of G-SNAT (10 μM) into G-SNAT-cyclized after incubation with equal amounts (100 U) of recombinant mouse gzmB (0.05 μg/mL) and human caspase-3. (C) TEM image of nanoaggregates after the incubation of G-SNAT (100 μM) with recombinant mouse gzmB (1 μg/mL) overnight at 37 °C in assay buffer. (D) DLS analysis of diluted (2×) G-SNAT-cyclized (50 μM) showing the distribution of particle sizes. (E) Measurement of the zeta potential of G-SNAT-cyclized. (F) Analysis of the stability of G-SNAT in mouse serum by HPLC. The percentage of probe remaining after the incubation of 100 μM of G-SNAT-Cy5 in mouse serum at 37 °C for indicated times was obtained by calculating the percentage of the peak area (mAU*min) of the probe on the corresponding HPLC trace; error bars (S.D.) were calculated from two separate experiments.

G-SNAT-Cy5 Imaging Report gzmB Activity in Cytotoxic T Cells and CAR T Cell Engaged Cancer Cells

To examine whether G-SNAT-Cy5 (Figure A) could report gzmB activity in cytotoxic T cells, we isolated CD8+ T cells from BALB/c mice, activated them with anti-CD3/CD28-coated beads, and divided these cells into two groups: one underwent retroviral transduction to generate CD19-28ζ CAR-T cells, as previously described,[58] while the other group was left untransduced. Nonactivated naive T cells (CD44lowCD62Lhigh) were prepared as the control (Supporting Information Figure S5). After 2.5 h of incubation with G-SNAT-Cy5 (5 μM), confocal microscopic imaging revealed fluorescence in both activated untransduced CD8+ and CAR T cells but not in naive T cells (Figure B and Supporting Information Movies S1, S2, and S3). After its synthesis, gzmB was packaged and activated in lytic granules to prevent self-killing and to facilitate trafficking to the immunological synapse.[59] The images were consistent with such granular sequestration of intracellular active gzmB and suggested the retention of in situ aggregated G-SNAT-Cy5 (Supporting Information Figure S6). Intriguingly, gzmB-packed granules displayed a denser fluorescent cluster in CAR T versus untransduced activated CD8+ T cells (Figure B). This might concur with faster lytic granule recruitment to nonclassical CAR T cell immune synapses described by Davenport et al.,[60] which was characterized by a lack of Lck clustering at small immunological synapses, in contrast to wild type T cells with the Lck-rich, large immunological synapses. Incubation of activated untransduced CD8+ T cells at 4 °C significantly attenuated G-SNAT-Cy5 uptake (Supporting Information Figure S7), suggesting an energy-dependent endocytic process rather than passive membrane diffusion. To validate that gzmB retained enzymatic activity in granules, we tested recombinant mouse gzmB under acidic (pH 5.5) conditions and found that the enzyme maintained about 26.6% and 24.3% of its activity at pH 5.5 after 2 and 4 h incubation with G-SNAT-Cy5 at 37 °C, respectively (Figure C and Supporting Information Figure 8). Bioluminescence assay with BALB/c B cell lymphoma line derived A20 cells expressing firefly luciferase (A20Luc+) confirmed that the cytotoxic function of these CAR T cells was not affected by overnight incubation with G-SNAT or G-SNAT-Cy5 (Figure D and Supporting Information Figure S9). Further viability study showed that both G-SNAT and G-SNAT-Cy5 were well tolerated by A20Luc+, CD8+ T, and CAR T cells (Supporting Information Figure S10). Collectively, these results suggest activation and retention of G-SNAT-Cy5 by gzmB in the granules of cytotoxic T cells in a bioorthogonal manner.

Figure 3

GzmB activity in naive CD8+ T, activated untransduced CD8+ T, CD19-28ζ CAR T, and A20 cancer cells. (A) Structure of G-SNAT-Cy5. (B) Microscopic imaging of T cells incubated with G-SNAT-Cy5 (5 μM) for 2.5 h and then stained with Hoechst (blue, Ex390/Em440) and FITC conjugated CD3 antibody (green, Ex488/Em520). Magenta (Ex650/Em670) represents retained G-SNAT-Cy5. (C) HPLC traces of G-SNAT-Cy5 (black) and incubation with gzmB (0.05 μg/mL) in MES buffer (pH 5.5, blue) or assay buffer (pH 7.5, red) at 37 °C for 2 h; 8.38% and 31.51% indicate the percent conversion relative to the G-SNAT-Cy5 peak as calculated from the peak areas (mAU*min). (D) G-SNAT or G-SNAT-Cy5 treatment (0, 2.5, and 5 μM) showed no impact on the cytotoxic function of CD19-28ζ CAR T cells against A20Luc+ cells. Error bars (S.D.) are calculated from triplicated experiments. (E) Flow cytometry analysis of PBS or G-SNAT-Cy5 treated naive CD8+ T, activated untransduced CD8+ T and CAR T cells with or without A20 cells, and A20 cells incubated with or without activated untransduced CD8+ T and CAR T cells. 10 000 cells were analyzed. T cells were incubated with A20 cancer cells at a 1:1 ratio for 2.5 h. The gates indicate high G-SNAT-Cy5 retention in cells. (F) Quantitative RT-PCR analysis of gzmB mRNA. ND: nondetectable. (G) Western blot analysis of gzmB in cell lysate (25 μg).

Next, we investigated whether G-SNAT-Cy5 could differentiate antigen specific cytotoxic killing by CAR T cells from activated untransduced CD8+ T cells in two established CAR T-cell therapy models in vitro: CD19–28ζ CAR T cells cocultured with A20 cells naturally expressing CD19 and GD2-4-1BBζ CAR T cells (transduced from total T cells) cocultured with SB28 murine glioblastoma cells engineered to stably express red fluorescent protein (RFP) and mouse GD2 (SB28-RFP/GD2). Both CAR T cells were freshly prepared and maintained according to previously reported protocols.[58,61] Naive, activated untransduced CD8+ and CD19-28ζ CAR T cells were first incubated with A20 cells in suspension at a 1:1 ratio (effector:target) in the presence of G-SNAT-Cy5 (5 μM) for 2.5 h. The live cells were analyzed by flow cytometry. Cancer and T cells were gated (Supporting Information Figure S11), and their mean fluorescence intensity of Cy5 (MFI-Cy5) was individually quantified. As shown in both 2D plots (side scatter-y-axis; MFI-Cy5-x-axis) (Figure E) and statistical analysis (Supporting Information Figure S12), exposure to cognate antigens greatly enhanced the G-SNAT-Cy5 activation in CAR T cells. A20 cells under CAR T cell-mediated cytotoxic killing had also elevated G-SNAT-Cy5 retention, although a small number of nonspecific killings were observed with untransduced CD8+ T cells (Supporting Information Figure S12A). Without an antigen, CAR T cells had also higher G-SNAT-Cy5 retention than activated untransduced CD8+ T cells (Supporting Information Figure S12B). We analyzed the expression of gzmB in these cancer and T cells at both transcriptional (Figure F) and protein levels (Figure G). GzmB was robustly expressed in activated CD8+ and CAR T cells, negligible in naive T cells, and completely nondetectable in A20 cells. CD19-28ζ CAR transduction allowed higher gzmB expression in CAR T cells, which correlated to the higher activation of G-SNAT-Cy5. Inclusion of a competitive gzmB inhibitor (Ac-IETD-CHO)[62] significantly compromised the retention of G-SNAT-Cy5 in CAR T cells (Supporting Information Figure S12C). To rule out that Cy5 fluorophore dominated the uptake and nanoaggregation of G-SNAT-Cy5, we repeated the study in GD2-4-1BBζ CAR T-cell treated SB28-RFP/GD2 cells with a postclick imaging strategy as previously described (Supporting Information Figure S13).[47] A caspase-3 sensitive nanoaggregation probe C-SNAT4 developed previously[42] was included to show apoptosis triggered by cytotoxic killing. Consistent with the prelabeled probe, G-SNAT was highly activated and retained in CAR T and engaged cancer cells after 2.5 h incubation at a 2:1 ratio. Additional caspase-3 imaging pinpointed the apoptotic cells without nonspecific detection of gzmB (Supporting Information Figure S14). Together, these results demonstrate that G-SNAT probes can image the activity of gzmB from CTLs during immunotherapy.

Bioluminescence and G-SNAT-Cy5 Fluorescence Imaging Reveal gzmB-Mediated Cytotoxic Activity in CAR T Cells Treated Lymphoma Models

To assess the ability of G-SNAT-Cy5 to report gzmB activity during CAR T cell mediated cytotoxicity in vivo, we tested a gzmB-releasing model by subcutaneously (s.c.) implanting A20 cells into the upper right flank of BALB/c Rag2–/–γc–/– mice and treated them with CD19-28ζ CAR TLuc+ cells (Figure A). These CAR-TLuc+ cells were generated by retroviral transduction of activated CD8+ TLuc+ cells from transgenic Fluc+ BALB/c mice with hematopoietic cells engineered to stably express firefly luciferase.[63] When the tumor volume reached ∼300 mm3, 6 × 106 activated untransduced CD8+ or equivalent CAR TLuc+ cells (∼70% transduction efficacy) were intratumorally injected. The abundance, viability, and distribution of these T cells were monitored by bioluminescence imaging with standard intraperitoneal (i.p.) administration of D-luciferin (3 mg) at 24 h (day 12 post A20 implantation) and 48 h (day 13) post T cell injection. Associated gzmB activity was longitudinally imaged at 1, 2, 4, and 20 h post tail vein administration of G-SNAT-Cy5 (5 nmol) at day 12. At day 13, the tumors were removed, imaged, and analyzed by western blot and immunofluorescence staining after another round of G-SNAT-Cy5 imaging (Figure A). As shown in Figure B, both activated untransduced CD8+ and CAR TLuc+ cells maintained viability 24 h post intratumoral injection, although a faster decay of CAR TLuc+ cells within the semisolid tumor microenvironment was noticed (Supporting Information Figure S15). The regions of interest (ROIs) on tumors were defined to quantify both bioluminescent and fluorescent intensity (Supporting Information Figure S16). The highest fluorescent intensity of G-SNAT-Cy5 was observed at 1 h postinjection in CAR TLuc+ cells treated tumors which was on average 3.1-fold of the PBS and 1.3-fold of the activated untransduced CD8+ TLuc+ cells treated tumors (Figure C). When the bioluminescence from TLuc+ cells was utilized to normalize the total fluorescence in tumors, a significantly higher G-SNAT-Cy5 activation and retention were observed in CAR TLuc+ cells treated than activated untransduced CD8+ TLuc+ cells treated tumors (Figure D). Further ex vivo analysis showed that the detection of G-SNAT-Cy5 highly correlated to the expression of gzmB in semisolid tumors (Figure E and F), and the colocalization of CD8α, gzmB, and G-SNAT-Cy5 was confirmed by immunofluorescence staining (Figure G). Hence, the significant and stable detection of our G-SNAT in an acidic semi-solid tumor microenvironment after extended time periods of adoptive cell transfer provides promising features that warrant further investigation in diverse tumor murine models.

Figure 4

Multimodal optical imaging with G-SNAT-Cy5 and D-luciferin predicts lympoid tumor response to CAR T-cell therapy. (A) Illustration of the workflow to generate the subcutaneous A20 lymphoid tumor/CD19-28ζ CAR TLuc+ cell therapy model and imaging study with D-luciferin and G-SNAT-Cy5. (B) Longitudinal bright-field, bioluminescence, and fluorescence imaging with D-luciferin and G-SNAT-Cy5 (5 nmol, Ex640/Em690) of A20 implanted (bottom), activated untransduced CD8+ TLuc+ (middle), or CAR TLuc+ cells (top) treated tumor-bearing mice. T cells were injected at the tumor site. Representative mice are shown here while full panels are shown in Supporting Information Figures S15 and S16. (C) A comparison of the relative fluorescence intensity acquired by defining ROIs on PBS, untransduced activated CD8+ TLuc+, or CAR TLuc+ cells treated A20 tumors at 1, 2, 4, and 20 h postinjection. **p < 0.0021-A20 vs CD8+ TLuc+ cells treated; ****p < 0.0001-A20 vs CAR TLuc+ cells treated. (D) A comparison of the normalized intensity (FI, fluorescence imaging/BLI, bioluminescence imaging) of untransduced activated CD8+ TLuc+ and CAR TLuc+ cells treated A20 tumors at 1 h postinjection. *p < 0.0332. (E) Bright-field and fluorescence imaging of PBS, CD8+ TLuc+, or CAR TLuc+ cells treated tumors at day 13. Major ruler unit is cm. (F) Western blot analysis of gzmB in tumor lysate (50 μg) from (E). (G) Immunofluorescence staining analysis of the tumors in (E). (H) Illustration of the workflow to generate the systemic lymphoma/CD19-28ζ CAR TLuc+ cell therapy model and imaging study with D-luciferin and G-SNAT-Cy5. (I) Longitudinal bright-field, bioluminescence, and fluorescence imaging with D-luciferin and G-SNAT-Cy5 (5 nmol, Ex650/Em670) of A20Luc+ implanted (bottom), activated untransduced CD8+ T (middle), or CAR T cells (top) treated tumor-bearing mice. Representative mice are shown here while full panels are shown in Supporting Information Figure S17. (J) Cartoon illustrating the defined ROI on liver and spleen (middle circle) and bone marrow (lower two circles) regions to estimate the fluorescence intensity. (K) A comparison of the relative fluorescence intensity acquired by defining the ROI on liver and spleen regions of A20Luc+, activated untransduced CD8+, or CAR T cells treated tumor-bearing mice. Error bars represent S.D. *p < 0.0332; **p < 0.0021, ns: not significant. (L) A comparison of the relative fluorescence intensity acquired by defining the ROI on bone marrow regions of A20Luc+, activated untransduced CD8+, or CAR T cells treated tumor-bearing mice. Error bars represent S.D. ****p < 0.0001.

To better mimic the clinical scenario, we employed a systemically delivered lymphoma model in which A20Luc+ cells were intravenously injected (i.v., tail vein) into sublethally (4.4 Gy) irradiated BALB/c mice (Figure H). In agreement with what has been reported previously,[64] by the time of CD19-28ζ CAR T cell administration (2.4 × 106, i.v., retro orbital), 7 days after tumor inoculation, A20Luc+ cells were infiltrating the liver and lymphoid organs, including bone marrow (Figure I and J and Supporting Information Figure S17). Longitudinal bioluminescence imaging with D-luciferin revealed the therapeutic effect of CAR T cells. Fluorescence imaging performed 1 h post tail vein administration (i.v.) of 5 nmol G-SNAT-Cy5 at day 11 showed a significantly higher retention of the probe in the liver, spleen, and bone marrow regions of CAR T cells treated mice (Figure K, L). Collectively, these results suggest that G-SNAT-Cy5 detects both intracellular and exocytosed gzmB during CAR T cells mediated cancer cell killing and tumor eradication which might be useful in reporting tumor response to CAR T-cell therapy as well as trafficking of cytotoxic T cells.

Bioluminescence and G-SNAT-Cy5 Fluorescence Imaging Monitor Tumor Response to Checkpoint Blockade Therapy

To visualize the immune activation to checkpoint blockade therapy, we performed longitudinal bioluminescence and fluorescence imaging of a syngeneic mouse model treated with a combined regimen of anti-PD1 and anti-CTLA4 antibodies. As outlined in Figure A, transgenic Fluc+ BALB/c mice were implanted subcutaneously with 1 × 106 CT26 murine colorectal cancer cells and treated with 200 μg of anti-PD1 and 100 μg of anti-CTLA4 intraperitoneally at day 9, 12, and 15 postinoculation. As indicated by tumor growth curves, this combination therapy triggered a heterogeneous response (Figure B). We defined those treated without complete tumor regression at the end point (day 35) as nonresponders (gray) and found that the tumor sizes in responders and nonresponders diverged at day 15. All treated responder, nonresponder, and nontreated cohorts were imaged longitudinally with D-luciferin, G-SNAT-Cy5, and a previously developed gzmB specific bioluminogenic substrate GBLI2 (Supporting Information Figure S18) at day 9, 12, 15, 18, and 20. With D-luciferin, the expansion and whole-body trafficking of hematopoietic cells in response to the checkpoint blockade could be monitored. Additional GBLI2 imaging provided a reference of gzmB activity for the validation of G-SNAT-Cy5.

Figure 5

Imaging with G-SNAT-Cy5, GBLI2, and D-luciferin to predict colorectal tumor reponse to checkpoint blockade therapy. (A) Illustration of the workflow to generate the subcutaneous CT26 colorectal tumor in Fluc+ BALB/c mice treated with anti-PD-1 and anti-CTLA-4 and imaging studies with G-SNAT-Cy5, GBLI2, and D-luciferin. (B) Growth of the tumors in treated responder (T.R.), treated nonresponder (T.N.R.), and nontreated cohorts. Three of the mice were sacrificed at day 16 for ex vivo analysis. Day 12 ns: not significant, day 14 ****p < 0.0001. (C) Bright-field, bioluminescence, and fluorescence imaging of healthy Fluc+ BALB/c mice showing the baseline uptake and distribution of D-luciferin (3 mg, i.p), GBLI2 (200 μg, i.v. retro orbital), G-SNAT-Cy5 (5 nmol, i.v. retro orbital, Ex640/Em690), and Cy5 (5 nmol, i.v. retro orbital, Ex640/Em690). (D) Longitudinal bioluminescence imaging with D-luciferin (3 mg, i.p.), GBLI2 (200 μg, i.v. retro orbital), and fluorescence imaging with G-SNAT-Cy5 (5 nmol, i.v. retro orbital) of nontreated (bottom), treated nonresponder (middle), and responder (top) groups. White circles indicate tumors. Representative mice from each group are shown. (E), (F), and (G) Relative bioluminescent and fluorescent intensity of tumors at day 9, 12, 15, 18, and 20 were quantified by defining ROIs on tumors. The percent differences of bioluminescence and fluorescence intensity among treated responder, treated nonresponder, and nontreated over the course of imaging are plotted using the treated nonresponder as the baseline; n = 5 for Treated R., n = 4 for Treated N.R., and n = 5 for Nontreated. *p < 0.0332, ***p < 0.0002. Error bars represent S.D. (H) Bright-field and G-SNAT-Cy5 fluorescence imaging of the mice at day 16 before euthanasia and tumor collection. White circles indicate tumors. (I) Bright-field and fluorescence imaging of tumors collected from (H). Major ruler unit is cm. (J) Western blot analysis of gzmB from (I). 50 μg of tumor lysate was loaded for analysis. (K) Immunofluorescent staining analysis of treated responded tumor from (I).

The baseline uptake and distribution of these probes were first evaluated with healthy Fluc+ BALB/c mice (Figure C). Via i.p. injection (3 mg), the bioluminescence from D-luciferin was mainly restricted to the abdomen, which is ideally suitable for imaging immune cell trafficking to the tumor site. GBLI2 imaging (200 μg, i.v. retro-orbital) gave approximately two magnitudes of weaker bioluminescence but which was highly concentrated in the spleen (Figure C) and to a lesser extent in the bone marrow, which suggested the major homing organs of CTLs. G-SNAT-Cy5, 1 h post i.v. injection (5 nmol, retro orbital), showed some uptake in the upper abdomen, in contrast to fast clearance of Cy5 fluorophore alone (5 nmol, i.v. retro orbital) (Figure C). Further dissection and organ imaging revealed that the uptake of G-SNAT-Cy5 was mainly associated with the gastrointestinal tract (Supporting Information Figure S19). Next, we imaged tumor-implanted mice receiving checkpoint blockade therapy. Consistent with growth curves, bright-field imaging showed divergent tumor responses to the checkpoint blockade (Figure D and Supporting Information Figure S20). Responders showed an increase in tumor volume initially at day 12, 3 days post the first treatment, and then a sustained shrinkage from day 15 to 20. Nontreated mice and treated nonresponders had continuous tumor growth. When imaged with D-luciferin, an activation of immune cells was seen in treated cohorts at day 12, indicated by enhanced bioluminescence surrounding tumors. GBLI2 imaging showed the activation of CTLs in all three cohorts regardless of treatment, but only the treated responders had obvious infiltration and gzmB activity in the tumors. Through longitudinal imaging, a distinct, ringlike pattern which might represent gradual immune cell exclusion from tumors was observed in nonresponder and nontreated cohorts. Such a ring pattern, however, was not observed in responders through the course of treatment (day 12–20). Compared to D-luciferin and GBLI2, G-SNAT-Cy5 imaging revealed an overall similar distribution but afforded a greater confined fluorescence signal and improved contrast within the tumor (Figure D).

To quantify the dynamic change of and compare the bioluminescent and fluorescent signals in different cohorts, we defined ROIs on tumors and calculated the relative percent difference using the treated nonresponders as the reference (Figure E–G and Supporting Information Figure S21). By tracking total immune cells with D-luciferin, an elevated tumor infiltration was observed since day 12, and the infiltrates were maintained through the whole process of tumor eradication in responders. GBLI2 and G-SNAT-Cy5 correlated well to immune cell infiltration in responders since day 12 which predicted the therapeutic outcome prior to tumor volume divergence. Follow-up mouse and tumor imaging at day 16 confirmed a higher G-SNAT-Cy5 in responders (Figure H and I) which correlated to the level of gzmB expression (Figure J) and localization (Figure K and Supporting Information Figure S22). Taken together, by visualizing the dynamics of immune cell trafficking and gzmB-mediated cytotoxicity, we observed the essential role of early and sustained CTLs infiltration in battling solid tumors. G-SNAT-Cy5 gzmB imaging could reliably differentiate and predict tumor response to checkpoint blockade therapy earlier than conventional imaging of tumor burden.

Discussion

Conventional immune profiling offers single-cell and molecular analysis of biopsy samples but has many known limitations such as invasive nature and lack of longitudinal information.[15] Molecular imaging has yielded different strategies to noninvasively monitor tumor response to immunotherapies such as the immunoPET strategy that combines the specificity of monoclonal antibodies and inherent sensitivity of PET to detect immune cell population, expression, and distribution of target antigens. Nonetheless, several gaps remain with immunoPET: (1) Membrane associated immune markers are often expressed by more than one cell population. For instance, CD8 has been mainly identified on CTLs but could also be found on regulatory T cells, NK cells, and dendritic cells,[65,66] all of which could be present in the tumor microenvironment at an early stage.[67] (2) The mere presence of immune cells may not represent their functions. (3) Expression of immune targets may not fully correlate to therapeutic response such as PD-L1.[68] (4) Antibody-based imaging probes usually have poor tissue/tumor penetration and a long circulation half-life (days to weeks), which necessitates long half-life radioisotopes associated with high radiation exposure.[69] New strategies, especially functional imaging at the cell level, are thus needed to fill the gaps for noninvasive imaging of immune activity.

One of the alternative approaches is to image the activity of gzmB, a key enzyme involved in the granule-mediated cytotoxicity by NK cells and CTLs. Based on the versatile TESLA strategy, we first developed G-SNAT for imaging the activity of gzmB in mouse models of CAR T-cell and checkpoint blockade therapies. In comparison to other gzmB-imaging probes, G-SNAT was initially administered in a format of small molecule and was converted by active gzmB in situ to nanoparticles. In this work, we proved the fluorescence imaging by labeling G-SNAT with a near-infrared dye Cy5. As demonstrated in multimodality imaging of caspase-3 and methionine aminopeptidase II activity,[45−49] the modular design of the TESLA strategy should allow us to label G-SNAT with a positron-emitting isotope like 18F for translational PET imaging that will overcome the limited tissue penetration with fluorescence imaging.

GzmB is produced in the cytosol as a zymogen with an N-terminal Gly-Glu dipeptide that inhibits the assembly of a functional catalytic triad.[59] In the Golgi, pro-gzmB is tagged with a mannose 6-phosphate for targeting the granules[70] where dipeptidyl peptidase I (DPPI, cathepsin C) cleaves the Gly-Glu dipeptide to generate active gzmB.[71] The active enzyme is then deposited on a scaffold of serglycin[72,73] under acidic conditions (∼pH 5.5).[74,75] It is believed that these mechanisms tightly regulate gzmB function within effector lymphocytes prior to being released from granules.[32,33] However, there are also reports that gzmB remained partially active in granules and turned on fluorogenic small[76] or cell permeable macromolecules.[35] Our results showed that gzmB could process G-SNAT-Cy5 in a buffer system at ∼ pH 5.5 and maintain ∼25% of its activity at the optimal pH 7.4 (Figure C and Supporting Information Figure S8), consistent with an early study.[77] Taking into account that granule-released, serglycin-bound gzmB could trigger apoptosis of target cells as efficient as free enzymes,[78] we believe the activation and retention of G-SNAT-Cy5 within CTLs are dependent on the gzmB activity. The different observations of gzmB activity in CTLs may be explained by the varied permeability of probes to cell or granule membranes. The unique property of G-SNAT-Cy5 allowed it to detect the remaining gzmB activity in CTLs. We observed no impact of G-SNAT-Cy5 labeling on the granule releasing and cytotoxic killing (Figure D). Our previous work has shown that the in vivo assembled nanoaggregates were cleared from the target site with a half-life of 244 min.[48] It is thus attempting to evaluate if G-SNAT can be applied to label and track CTLs in vivo. This new feature represents another advantage of our TESLA-based granzyme B probes.

To date, five CAR T-cell therapies have been approved by the FDA to treat several types of lymphomas and leukemias, as well as multiple myeloma. Despite great success in hematological malignancies, harnessing CAR T cells for treating solid tumors has seen little progress. Current research focuses on improving tumor infiltration, persistence, and potency in heterogeneous and immunosuppressive tumor microenvironments.[79] As demonstrated in our preclinical lymphoma tumor models, G-SNAT gzmB imaging in combination with bioluminescence imaging may facilitate these efforts by reporting the distribution of CAR T cells and the onset of cytotoxic killing. In the subcutaneously implanted, semisolid lymphoid tumors, bioluminescence imaging indicated more activated, untransduced CD8+ T cells than CAR T cells post intratumoral injection; on the other hand, G-SNAT gzmB imaging showed more activity from CAR T cells. This example demonstrated the value of sequential imaging of multiple targets in the same tumors in reporting the complex immune activity. Bioluminescence imaging also revealed a decrease of CAR T cell viability post intratumoral injection (Figure B and Supporting Information Figure S15). This decay might reflect the gradual exhaustion or senescence of CAR T cells overwhelmed by CD19 antigens within the tumor microenvironment. A20 lymphoma cells were known to express high levels of PD-L1, and a combined treatment of anti-PD-L1 antibody and ibrutinib, an approved Bruton’s tyrosine kinase inhibitor, could cure established A20 tumors.[80] It is worthwhile to investigate in the future how these CAR T cells lost persistence within the semisolid A20 tumors and whether a checkpoint inhibitor could reverse that. It will also be interesting to examine if G-SNAT gzmB imaging could be utilized to differentiate malfunctioning CAR T cells in vivo since cumulative data have suggested that senescent T cells have a reduced expression of gzmB[81] and exhausted CTLs have impaired granzyme packaging and degranulation[82,83] and thus a defect in gzmB release and cytotoxicity. The modular design of the TESLA strategy allows G-SNAT to be readily engineered by replacing the mouse gzmB peptide substrate with the sequence preferred by human gzmB. It should enable the detection of human CAR T cell mediated cytotoxicity in treating human cancer xenografts in mouse models and potentially in clinical trials.

While checkpoint inhibitors have been adopted to treat a variety of solid tumor types, only a small percentage of patients exhibit complete response, and early detection of antitumor immunity remains challenging. By imaging CT26 colorectal tumors whose responses to anti-PD-1/anti-CTLA4 were dichotomous, we found that checkpoint blockade therapy cured cancer by stimulating and maintaining early immune cell infiltration including activated and expanded CTLs, which led to gradually intensified cytotoxicity at the tumor site that was inversely proportional to the tumor volume (Figure ). These findings are consistent with a previous immune profiling by flow cytometry and gene expression analysis showing that responded CT26 tumors contained expanded T and NK cell populations with CD8+ T cells that nearly doubled 3 days after the third round of anti-PD-1/anti-CTLA-4 treatment.[84] At this point, G-SNAT or GBLI2 imaging was not able to differentiate gzmB activity from NK and certain gzmB expressing regulatory T cells or between newly infiltrated and tumor-surrounding effector lymphocytes, but unlike conventional immune profiling, G-SNAT gzmB imaging would advance spatially and temporally correlated surveillance of cytotoxicity and predict tumor early response in vivo. Our longitudinal imaging revealed a distinct ring pattern surrounding the tumors or so-called “immune dessert” in both advanced untreated and treated nonresponding tumors. With current imaging, we could not rule out the possibility that the rings were formed due to lack of tumor-infiltrated immune cells or CTLs gradually losing their viability. But a similar pattern displayed by longitudinal immunoPET with an 89Zr-labeled PEGylated variable region segment of camelid heavy chain-only antibody targeting CD8 from Rashidian et al.[18] supported our conclusion that CTLs were exiled from advanced tumors. Future combination of G-SNAT functional imaging and cell tracking immunoPET should sketch a more comprehensive picture of how the immune system interacts with tumor cells upon checkpoint blockade therapy and may generate new insights leading to the development of more potent therapies.

In conclusion, we developed a novel gzmB-activated self-assembly imaging probe and proved conceptually that G-SNAT-Cy5 fluorescence imaging could be utilized to monitor cytotoxic activity and predict tumor response to CAR T-cell and checkpoint blockade therapies. Together with longitudinally whole-body bioluminescence imaging, the trafficking and tumor infiltration of total splenocytes or CTLs in responding to immunotherapies were visualized, which revealed an essential role of early immune infiltration and persistent cytotoxic activity in curing cancer. These results support further development of G-SNAT for PET imaging of the immune response in patients and also highlight the value of multiparameter imaging for studying the cytotoxic function in the context of immune cell interplay, tumor microenvironments, and new cancer immunotherapy.