The Effectiveness of Checkpoint InhibitorCombinations and Administration Timing CanBe Measured by Granzyme B PET ImagingBenjamin M. Larimer1, Emily Bloch1, Sarah Nesti1, Emily E. Austin1, Eric Wehrenberg-Klee1,Genevieve Boland2, and Umar Mahmood1
Abstract
Purpose: The lack of a timely and reliable measure ofresponse to cancer immunotherapy has confounded under-standing of mechanisms of resistance and subsequent thera-peutic advancement. We hypothesized that PET imaging ofgranzymeBusing a targeted peptide,GZP, could beutilized forearly response assessment across many checkpoint inhibitorcombinations, and that GZP uptake could be comparedbetween therapeutic regimens and dosing schedules as anearly biomarker of relative efficacy.
Experimental Design: Two models, MC38 and CT26, weretreated with a series of checkpoint inhibitors. GZP PET imag-ing was performed to assess tumoral GZP uptake, and tumorvolume changes were subsequently monitored to determineresponse. The average GZP PET uptake and response of eachtreatment group were correlated to evaluate the utility of GZPPET for comparing therapeutic efficacy.
Results: In both tumor models, GZP PET imaging washighly accurate for predicting response, with 93% sensi-tivity and 94% negative predictive value. Mean tumoralGZP signal intensity of treatment groups linearly corre-lated with percent response across all therapies andschedules. Moreover, GZP PET correctly predicted thatsequential dose scheduling of PD-1 and CTLA-4 targetedtherapies demonstrates comparative efficacy to concurrentadministration.
Conclusions: Granzyme B quantification is a highlysensitive and specific early measure of therapeutic effi-cacy for checkpoint inhibitor regimens. This work pro-vides evidence that GZP PET imaging may be useful forrapid assessment of therapeutic efficacy in the context ofclinical trials for both novel drugs as well as dosingregimens.
IntroductionThe success of approved immune checkpoint inhibitors target-
ing PD-1, PD-L1, and CTLA-4 has generated significant interest inthe application of these and other therapies across malignancies(1–3). Although many challenges exist in the rapidly progressingfield of immuno-oncology, 2 major concerns include improvingthe ability to monitor response to both approved and clinicallytrialed drugs and reducing potentially severe side effects whilemaintaining maximum antitumor efficacy (4, 5). Currently,response is monitored by anatomic measurements of tumorvolume and overall survival, the latter of which may take yearsto resolve (4). Adding to the complication of long delays betweentherapy initiation and outcome is the frequency of severeimmune-related adverse events, especially in therapy combina-tions such as anti-CTLA-4 and anti-PD-1 (5, 6). Several preclinicaland clinical studies have demonstrated that concurrent adminis-
tration of combination therapies may not be necessary, and thatthe timing of when each checkpoint inhibitor is delivered mayreduce side effects while maintaining efficacy (7).
Although a substantial amount of evidence surrounds theefficacy of PD-1 and CTLA-4 blockade, not all combinations ofhypothesized checkpoints under investigation are strongly sup-ported by biological rationale. One such checkpoint of interest,TIM-3, is upregulated in murine and human cancers followingtreatment with PD-1 checkpoint inhibitors, providing a rationaltarget for combination studies (8). However, the role of TIM-3 inoncologic immunotherapy response has yet to be completelyelucidated, as evidence supports its role in both immunogenicand tolerogenic pathways (9, 10). The ability to noninvasivelycompare the effects of therapies targeting both well-understoodcheckpoint inhibitors, like PD-1 and CTLA-4, and those withoutconsensus, such as TIM-3, could accelerate the clinical develop-ment of novel therapies.
Prior to therapy, certain characteristics, such as T-cell infiltra-tion, PD-L1 expression, microsatellite instability, and mutationalburden have been demonstrated to enrich response (11–13).Although these are important population-based characteristics,they are not effective for predicting individual tumoricidalimmune activity. Recently, we developed a PET imaging agenttargeting a marker of effector T-cell activation, granzyme B, thatpermits direct quantification of the antitumor response beforetumor volume changes (14). Extracellular granzyme B has abiological half-life of 14 days, representing a stable target ofimmune activation. Using a highly specific peptide PET imagingagent for granzyme B (GZP), we found that GZP PET signal was
1Center for Precision Imaging, Department of Radiology, Massachusetts GeneralHospital, Boston, Massachusetts. 2Department of Surgery, Massachusetts Gen-eral Hospital, Boston, Massachusetts.
Note: Supplementary data for this article are available at Clinical CancerResearch Online (http://clincancerres.aacrjournals.org/).
Corresponding Author: Umar Mahmood, Massachusetts General Hospital, 14913th St, Rm 5.407, Charlestown, MA 02129. Phone: 617-726-6477; Fax: 617-726-7422; E-mail: [email protected]
predictive of eventual response to immunotherapy, with high-signal tumors responding to therapy, and low-signal tumorsprogressing. By designing the GZP imaging agent to only bindto the active, secreted form of granzyme B, we have developed asystem inwhich PET imaging canquantitatively integrate pro- andantitumor immune signaling, as well as discriminate betweenactive and exhausted T cells to provide insight beyond T-cellinfiltration (15, 16). Given the initial success of this agent, amore comprehensive analysis of the potential clinical applica-tions of GZP, including as a broad predictive agent and clinicaltrial assessment tool was warranted. We hypothesized that GZPPET imaging would both be predictive of response to additionalimmunotherapy regimens across multiple tumor types for anindividual tumor and provide a quantitative biomarker thatwould allow for early comparison of therapeutic regimens effi-cacy. We hypothesized that it could also be used as an efficacymonitoring tool in an adaptive therapy approach of dose sched-uling, permitting a precision approach to immunotherapy.
Materials and MethodsMaterials and cell lines
Unless otherwise stated, all chemicals were purchased fromSigma-Aldrich. CT26murine colon carcinoma cells were obtainedfrom ATCC and cultured in Roswell Park Memorial Institutemedium with 10% FBS. MC38 murine colon carcinoma cellswere obtained from Kerafast and cultured in DMEM with 10%FBS. Monthly mycoplasma testing was performed by PCR screen-ing and cells were discarded after 15 passages. All cell-basedexperiments were done with cells acquired within 6 months inorder to ensure fidelity of the cell line identity.
Murine modelsMice were housed and maintained by the Center for Compar-
ative Medicine following animal protocols approved by theMassachusetts General Hospital Institutional Animal Care andUse Committee. Approximately 1 � 106 cells per xenograft ofCT26 (n ¼ 5–12) and MC38 (n ¼ 5–8) were diluted 1:1 (v/v)in Matrigel (Corning) and injected into the upper right flankof balb/c or C57BL/6 mice (Charles River Laboratories), respec-tively. CT26 and MC38 tumors were chosen based on their
demonstrated responsiveness to checkpoint inhibitor therapy andwhole-exome sequencing confirmed differences in mutation bur-den (14, 17–19). Anti-mouse PD-1 (clone RPM1-14), anti-mouseCTLA-4 (clone 9H10), and anti-mouse TIM-3 (clone RMT3-23)therapies were obtained from Bio X Cell. Mice were treated ondays 3, 6, and 9 following tumor inoculation by intraperitonealinjection with saline (vehicle), 200-mg anti-PD-l (monotherapy),200-mg anti-PD-1 and 100-mg anti-CTLA-4 (PþC combinationtherapy), or 200-mg anti-PD-1 and 250-mg anti-TIM-3 (PþT com-bination therapy). These doses were chosen based on previousstudies that demonstrated efficacy at ranges that approximate the3 to 10mg/kg dosing used for human clinical studies (18, 20, 21).In subsequent studies, scheduling of immunotherapy doses weremodified to 100-mg anti-CTLA-4 on day 3 then 200-mg anti-PD-1on days 6 and 9 (CþP schedule therapy) or 200-mg anti-PD-1 and100-mg anti-CTLA-4 on day 3 and PD-1 maintenance on days 6and 9 (PCþP schedule therapy). A schematic of all treatments canbe found in Supplementary Fig. S1. Mice were either imaged orsacrificed for ex vivo analysis onday 12posttumor inoculation. Forsurvival imaging experiments, tumor dimensions were mea-sured with calipers every 2 to 3 days beginning on day 5 forMC38 and day 10 for CT26 and ending on day 20. Mice weresacrificed once tumors exceeded a volume of 500 mm3 ordeveloped ulceration.
Biochemical analysis of tumorsTumors were excised from sacrificedmice for ex vivo analysis on
day 12 posttumor inoculation. Tumors designated for Westernblot analysis were lysed in 1% SDS solution supplemented withprotease inhibitors prior to analysis by Western blotting. Anti-granzyme B (4275S; Cell Signaling Technologies), anti-CTLA-4(ab134090; Abcam), anti-PD-1 (12A7D7; ThermoFisher), anti-TIM-3 (ab185703; Abcam), anti-CD4 (sc19643; Santa CruzBiotechnology) anti-CD3 (sc-20047; Santa Cruz), anti-CD8(53-6.7; ThermoFisher), and anti-b-actin (4970S; Cell SignalingTechnologies) antibodieswere used and subsequently detected byhorseradish peroxidase-conjugated goat anti-rabbit polyclonalantibody (ab6721; Abcam) as per manufacturers' recommenda-tions. Bands were detected using SignalFire ECL Reagent (CellSignaling Technologies) and imaged on an iBright FL1000 Imag-ing System (ThermoFisher). Images were quantitatively analyzedby ratio of net relative optical intensity of the target of interest tothat of b-actin for each sample.
For IHC staining of patient samples, all specimens wereacquired from patients following informed consent and accord-ing to a Massachusetts General Hospital institutional reviewboard-approved clinical protocol. Samples were obtained frompatients undergoing pembrolizumab treatment between 21 and42 days in order to analyze samples for potential markers ofresponse. Immunofluorescent staining was performed on forma-lin-fixed paraffin embedded sections following antigen retrievalin EDTA buffer using standard heat-based antigen retrieval tech-niques. Samples were blocked with goat serum before detectinggranzyme B expression with an anti-granzyme B antibody(ab5049; Abcam). Bound antibody was detected by AlexaFluor-488 goat anti-rabbit (Life Technologies) for immunofluo-rescent visualization. Fluorescentmicroscopywas performed on aBioTek Cytation5 inverted microscope using Gen5 software(BioTek).
Flow cytometric analysis was performed on PþC treatedand vehicle-treated mice by excising murine tumors on day 12
Translational Relevance
As checkpoint inhibitor therapy treatments continue to gainFDA approval across a number of cancers, previous challengesmonitoring therapeutic efficacy become amplified. Previously,our group developed a clinically translatable PET imagingagent for granzyme B. Here, we demonstrate the predictivecapabilities of granzyme B PET imaging across a range oftumors and treatments, revealing greater than 90% sensitivityand negative predictive value. Granzyme B PET imaging couldalso be used to compare therapies, with a correlation betweenmeanPET signal andoverall response. Finally, wedemonstratethat the effectiveness of the order in which CTLA-4 and PD-1blockade is administered can be ascertained by PET imaging.The data suggest granzyme B PET imaging could provide anearly readout of immunotherapy efficacy with benefits to bothindividual patients and novel therapy evaluation.
Granzyme B PET Immunotherapy Response Quantification
www.aacrjournals.org Clin Cancer Res; 25(4) February 15, 2019 1197
following tumor inoculation and generating a single-cell sus-pension by incubating in RPMI (ThermoFisher) supplementedwith 1 mg/mL collagenase type IV for 35 minutes at 37�C.Cells were filtered into a single-cell suspension through a 70-mm cell filter (Greiner Bio-One) and stained for live cells byZombie Violet Fixable Viability Kit (Biolegend), and murineanti-CD45 (clone 104; Biolegend), anti-CD8 (clone 53-6.7;Biolegend), anti-PD-1 (clone RMP1-30; Biolegend), anti-EOMES (ab2574227; ThermoFisher), and anti-granzyme B(Grb17; ThermoFisher) according to the manufacturer's speci-fications. Flow cytometry was performed on a cytometry cellanalyzer BD LSRFortessa X20 (BD Biosciences) and gatingestablished using FlowJo software (FlowJo LLC).
Peptide synthesis and radiolabelingNOTA–b-Ala–Gly–Gly–Gly–Ile–Glu–Phe–Asp–CHO (NOTA-
GZP)was synthesized using standard FMOC chemistry and puritywas measured by mass spectroscopy and HPLC (20). 68Ga waseluted from a 68Ge/68Ga generator (RadioMedix) with 0.1MHCl.The eluent was raised to pH 3.5 to 4.0 with 2M HEPES beforeaddition of 100 mg of NOTA-GZP. After allowing the labelingreaction to proceed for 10 minutes at room temperature, thepeptide was purified on a reverse-phase C18 Sep-Pak mini car-tridge (Waters) and eluted in 200 mL 70% ethanol. Saline wasadded to create a final formulation of <10% ethanol concentra-tion and approximately 9.5 to 28 MBq per mouse.
Murine PET imagingOn day 12 postinoculation, mice were intravenously injected
via tail vein with 68Ga-NOTA-GZP and subsequently imaged after1 hour. All imaging was performed on a rodent Triumph PET/CT(GE Healthcare). PET images were acquired for 15 minutes andfollowed by CT acquisition. Reconstruction was achieved using3D-MLEM (4 iterations, 20 subsets) and corrected for scatter andrandoms. VivoQuant software (InviCRO) was used for imageprocessing and analysis. Individual tumors were identified man-ually by drawing a 3D region of interest using CT-anatomiccorrelation. Background blood pool radioactivity was measuredby identifying the left ventricle of the heart as a region of interest.Specific 68Ga NOTA-GZP uptake was quantified by dividing totaltumor uptake by background blood pool to derive a tumor-to-blood ratio (TBR).
Statistical analysisGraphPad Prism Version 7 software (GraphPad Software, Inc.)
wasused for statistical analysis and graphing.Unpaired t testswithWelch's correction were used for all comparisons between immu-notherapy-treated and vehicle-treated tumors. A first-order regres-sionweightedby group sizewas used to correlate percent responsewith average TBR for each treatment arm. Statistical inferencebased on Pearson's correlation coefficient was used to define 95%confidence bands of the best-fit line. Comparison of therapyschedules was performed using 1-way ANOVA with a Kruskal–Wallis test for multiple comparisons. Survival benefit was calcu-lated by log-rank test for trend.
ResultsWestern blot analysis of relevant checkpoint inhibitor targets
MC38 and CT26 syngeneic tumors treated with concurrentcombination anti-PD-1 and CTLA-4 therapy were excised and
evaluated by immunoblot on day 12 after 3 combination doses todetermine posttreatment markers that may differentiate responserate prior to therapy. Expression after 1 cycle of treatment of thecheckpoint inhibitor target molecules, PD-1, CTLA-4, and TIM-3,immune cell markers CD3, CD8, CD4, and granzyme B wereanalyzed (Fig. 1A; Supplementary Fig. S2). Significant differenceswere found to exist between moderate and highly responsivetumors for CTLA-4 (MC38CTLA-4:b-actin¼ 2.113 vs. CT26 0.43,P ¼ 0.02) as well as CD3 (MC38 CD3:b-actin ¼ 2.153 vs. CT260.50, P¼ 0.006). Granzyme Bwas also evaluated by immunobloton day 12 and expression was found to be significantly higher intreatedMC38 tumors (granzymeB:b-actin¼ 6.79) than in treatedCT26 tumors (0.52 P ¼ 0.002). This 13-fold higher expression inhighly responsive MC38 tumors posttreatment was the largestmagnitude of difference for any of the markers analyzed.
Despite the significant differences obtained by Western blotanalysis, as a destructive technique, it was not possible to deter-mine whether the detected granzyme B was from active, inactive,or exhausted T cells. PD-1 and eomesodermin (EOMES) werechosen as markers of chronic activation and exhaustion, andgranzyme B positive T cells found in the tumors were assessedby flow cytometry to determine relative expression levels of these2 proteins (Fig. 1B). Analysis in anti-PD-1 plus anti-CTLA-4combination treated and vehicle-treated CT26 tumors on day12 demonstrated that although there were roughly similar levelsof granzyme B negative T cells in both treatment groups, thepercentage of granzyme Bþ/PD-1 negative T cells was muchhigher in vehicle-treated tumors, whereas a significantly highernumber of granzyme Bþ/PD-1þ cells were found in treated cells.A similar trend in granzyme Bþ/EOMESþ CD8þ T cells supportsmore T-cell exhaustion, as previous studies have linked thisphenotype to T-cell exhaustion, although the active level ofgranzyme B release cannot be distinguished (22).
Immunofluorescent microscopy of human melanomaThe inability to distinguish active granzyme B release using
ex vivo techniques such as flow cytometry andWestern blot analysisrepresents a significant problem for using granzyme B as a proteinbiomarker, as only actively released granzyme B is capable ofinitiating apoptosis and generating an antitumor response. Todemonstrate this concept in immunotherapy patients, immuno-flourescent microscopy of humanmelanoma specimens posttreat-ment of anti-PD-1 illustrates that although similar levels of gran-zyme Bmay be present in 2 separate tumors, the ability to quantifyreleased granzyme B is critical to distinguishing patient nonrespon-ders (Fig. 1C) from responders (Fig. 1D). The biological relevanceand magnitude of dynamic range between response and nonre-sponseofgranzymeB, combinedwith thenecessity toquantifyonlyactivated, released granzyme B makes PET imaging with the GZPpeptide a logical choice for response monitoring.
Granzyme B PET imagingPreviously, we demonstrated response prediction in a moder-
ately responsive CT26model of tumor immunotherapy, howeverbaseline levels of granzyme B were low, which leads to minimaldifferences between stored granzyme B not contributing to theantitumor response and secreted granzyme B that is activelyinducing tumoral killing. To examine whether the predictivecapability of PET imaging could be expanded into a tumormodel with higher levels of granzyme B at baseline, both CT26and MC38 tumor-bearing mice were examined. Prior to the
Larimer et al.
Clin Cancer Res; 25(4) February 15, 2019 Clinical Cancer Research1198
comparison of these models, biodistribution was performed inMC38 mice, which showed that the GZP peptide had similarnontarget organ distribution and clearance routes, confirmingthat comparison across 2 different mouse strains was appropriate(Supplementary Fig. S3). To confirm that accumulation of thepeptide was due to granzyme B-targeted binding and not perfu-sional differences between responding and nonrespondingtumors, imaging was performed in combination anti-PD-1 plusanti-CTLA-4 treated mice with both GZP and 1 of 2 nontargetedpeptides on the same day. Significantly higher peptide accumu-lation was observed in responding tumors for GZP as comparedwith either control peptide, confirming that perfusional differ-ences were not the major driving factor for peptide accumulation(Fig. 2A and B). In addition, the peptide was tested at 10� timesthe imaging dose in a group of combination treated mice andcompared with mice injected with a control peptide. No differ-ences in either tumor growth rate or complete responses were seenbetween the groups, indicating that the fraction of granzyme Bbound by the GZP peptide was not significant enough to alterantitumor response (Fig. 2C).
Following these observations, both MC38 and CT26 micewere treated with monotherapy, anti-PD-1þanti-CTLA-4 (PþC)combination therapy, anti-PD-1þanti-TIM-3 (PþT) combina-tion therapy, or saline and imaged using 68Ga-NOTA-GZP onday 12 to noninvasively assess granzyme B levels, and subse-quently compared to previous imaging of CT26 PþC (14). PETimaging revealed common organ uptake patterns in all miceanalyzed, including kidney and bladder accumulation consis-tent with renal clearance characteristic of small peptides.Degree of tumor uptake, however, varied based on tumormodel and therapy regimen (Fig. 3). In the CT26 model,vehicle-treated mice had similar tracer accumulation in thetumor and in the left ventricle of the heart, which was usedas a measure of background radioactivity, resulting in a base-line average tumor-to-background ratio (TBR) of 0.97 � 0.07.In comparison, CT26 tumor-bearing mice receiving PD-1monotherapy and PþC combination therapy had significantlyhigher tumor uptake with mean TBRs of 1.32 � 0.15 (P < 0.05)and 1.48 � 0.19 (P < 0.005), respectively (Fig. 4A). CT26xenograft mice treated with PþT combination therapy
Figure 1.
A, Western blot quantification of potential predictive proteins associated with antitumor response in CT26 (black) and MC38 (gray) tumors treated withanti-PD-1 and anti-CTLA-4 combination therapy. Bars represent the mean of 4 tumors, error bars denote standard error measurement (SEM). B, Flowcytometry quantification of CT26 tumors treated with anti-PD-1 and anti-CTLA-4 combination therapy for granzyme B (GZB), PD-1 and EOMES in CD45þ T cellsfrom vehicle (black) and combination therapy. Bars represent the mean of 4 tumor replicates, with error denoted as SEM. C, Tumor specimens frommelanoma patients treated with anti-PD-1 therapy stained for nuclei with DAPI (blue) and anti-granzyme B (green). A more diffuse pattern is seen in (D)consistent with activated T cells � , P < 0.05; �� , P < 0.01.
Granzyme B PET Immunotherapy Response Quantification
www.aacrjournals.org Clin Cancer Res; 25(4) February 15, 2019 1199
demonstrated no elevated tumor-specific tracer accumulationin comparison to the vehicle-treated mice (TBR ¼ 0.99 � 0.14).
To experimentally define a TBR threshold to predict responseacross the 3 therapy regimens, the highest TBR for vehicle-treated
CT26 tumors, 1.27, was defined as the value above which tumorswould be classified as high-uptake, and below which they wouldbe classified as low-uptake.Using this cutoff for theCT26model, 5of 12monotherapy, 7 of 12 PþC combination therapy, and 1 of 6
Figure 2.
TBR ratios of subsequently-responding mice bearing CT26 tumors treated with anti-PD-1 plus anti-CTLA-4 combination therapy and imaged using PET/CTwith (A) GZP followed 4 hours later by a scrambled peptide or (B) GZP followed 4 hours later with a second peptide targeting a human protein. BackgroundTBR ¼ 1. � , P < 0.05. C, Schematic of treatment with anti-PD-1 and anti-CTLA-4 therapy followed by 10� mass dose injection of unlabeled GZP. Averagetumor volume of 10� GZP-treated mice are shown by orange lines and circles, with uninjected tumors shown by blue lines and squares.
Figure 3.
Representative PET/CT coronal and axial images takenfrom CT26 (top row) or MC38 (bottom row) tumor-bearing mice. Mice were treated with either vehicle,anti-PD-1, anti-PD-1 plus anti-CTLA-4 or anti-PD-1 plusanti-CTLA-4 therapy as denoted in the heading of eachrow. Tumors are outlined by a white circle and labeledwith a white "T" and kidneys, the major route of imagingagent clearance are marked by a black "K." All PETintensities are displayed in the same intensity, with thescale displayed on the right in Bq/mL.
Larimer et al.
Clin Cancer Res; 25(4) February 15, 2019 Clinical Cancer Research1200
PþT combination therapy-treated mice were classified as high-uptake.
MC38 tumor-bearing mice also demonstrated differentialtumor uptake based on therapy regimen. Saline-treated MC38-bearing mice had an average TBR of 1.37 � 0.016, significantlyhigher than that of the saline-treated CT26 tumor-bearing mice(P < 0.005; Fig. 4B) and indicative of a comparatively higherbaseline granzyme B level in MC38 tumors consistent withWestern blot results. As in the CT26model, MC38 tumor-bearingmice treated with both monotherapy and PþC combinationtherapy had significantly higher TBRs of 1.99 � 0.22 (P < 0.05)and 2.17 � 0.21 (P < 0.005), respectively, compared with therelevant control mice. However, unlike in the CT26model, MC38tumors treated with PþT combination therapy also resulted inhigher uptake than control tumors (TBR¼ 1.74� 0.12, P < 0.05),although this group still had the lowest mean TBR among the 3treatments. The threshold for high- and low-uptake for MC38tumors was defined as 1.45 using the same methodology as forCT26 tumors, with an exception made for a single vehicle-treated tumor with high uptake that later demonstrated delayedgrowth in comparison to all other vehicle-treated tumors. Thenumber of high-uptake MC38 tumors included 4 of 5 mono-therapy, 8 of 8 PþC combination therapy, and 5 of 7 PþTcombination therapy-treated mice.
Growth curve analysisTumor volumes were used as a surrogate response measure of
overall survival. CT26 tumor growth was dichotomous, as tumors
regressed completely by day 20, or continued to grow untildeveloping ulcerations, reaching maximum volume, or the endof the study. Based on these growth patterns, mice were catego-rized as responders if the tumor volume decreased to 0 mm3
or nonresponders if tumor volume continued to increase. Figure4C–E illustrates the changes in CT26 tumor volume for eachtherapy regimen, with an average of vehicle-treated tumors repre-sented in black. None of the vehicle-treated CT26 tumorsregressed in size, but 3 of 12 monotherapy (Fig. 4C), 7 of 12PþC combination therapy (Fig. 4D), and 0 of 6 PþT combinationtherapy treated mice (Fig. 4E) responded to therapy.
In the MC38 tumor model, 3 groups were identified based onresponse patterns; complete responders, nonresponders, and athird group classified as partial responders. Responding tumorsregressed in size by day 20, nonresponding tumors grew to500 mm3 in size on or before day 14, and partially respondingtumors reached 500 mm3 at a delayed time point later than day14. Figure 4F–H highlights these temporal patterns of tumorvolume changes inMC38by therapy regimenwith vehicle-treatedtumor growth curves in black. Of note, 1 of the 7 vehicle-treatedMC38 tumors showed significantly delayed tumor growth com-pared with the other vehicle-treated tumors and was therebyclassified as a partial responder (Fig. 4F). This tumor also had aTBR greater than 1.45 and as such is shown in blue. A higherpercentage of MC38 tumors responded to therapy compared toCT26 tumors, with 4 of 5 monotherapy treated and 7 of 8 PþCcombination therapy treated tumors completely regressing in size.Unlike CT26, MC38 tumors responded to PþT combination
Figure 4.
Individual TBR ratios for mice treated with vehicle (Veh, n ¼ 8 CT26, n ¼ 7 MC38) anti-PD-1 (PD-1 n ¼ 12 CT26, n ¼ 5 MC38), anti-PD-1 plus anti-CTLA-4(PþC, n¼ 12 CT26, n¼ 7 MC38), or anti-PD-1 plus anti-TIM-3 (PþT n¼ 7 CT26, n¼ 7 MC38) therapies for (A) CT26 and (B) MC38 tumor bearing mice. Green circlesdenote mice with responses and pink boxes represent nonresponsive tumors. Individual tumor growth curves for CT26 (C–E) and MC38 (F–H) tumors treatedwith the same therapy combinations are shown, with green lines corresponding to high GZP PET TBRs and red lines to low GZP PET TBRs. The black lines are arepresentative mean of 8 vehicle mice and blue lines are representative of a single mouse with high GZP PET TBR that exhibited a partial response.
Granzyme B PET Immunotherapy Response Quantification
www.aacrjournals.org Clin Cancer Res; 25(4) February 15, 2019 1201
therapy, with 1 of 7 tumors responding completely, and anadditional 3 of 7 demonstrating delayed growth in response totherapy, characteristic of partial response.
Analysis of response prediction using GZP PETTo compare the predictive value of GZP PET across therapeutic
regimens, we defined the threshold for discriminating high fromlow GZP uptake among treated tumors as the highest TBR seenamong vehicle-treated tumors, as described above, resulting inTBR thresholds of 1.27 for CT26 tumors and 1.45 for MC38tumors. Using this classification system, 26 of 28 fully or partiallyresponding mice were categorized as high-uptake by GZP PETimaging, whereas 33 of the 38 non responding mice were clas-sified as low-uptake (Table 1). This reflects an overall sensitivityand specificity of 93% and 87% respectively for the ability of68Ga-GZP PET to predict immunotherapy response. A morecomprehensive examination as to the reason for the discrepancybetween PET and response for the discordant tumors was soughtthrough analysis of the positive and negative predictive values ofGZP PET. Most of the non-predictive scans were found in tumorsthat had high granzyme B PET signal but ultimately did notrespond resulting in a positive predictive value of 84% (26/31)but anegativepredictive valueof 94%(33/35), indicating that lowgranzyme B PET signal was highly correlated with non-response.
Next, the accuracy of granzyme B PET imaging was tested todetermine if group-based prediction of efficacy was possible bycompiling the mean GZP TBR within each treatment group andcorrelating with overall percent response. Treatments with thebest response survival outcomes also had the highest mean GZPTBR based on a Kaplan–Meier plot of survival for each therapy(Fig. 5A and B). To determine whether the relationship betweenpercent response and mean tumor GZP TBR was correlated,percent response was plotted against TBR (Fig. 5C). The curverevealed a linear relationship with R2 ¼ 0.84 and a significantlynonzero slope (P < 0.0005), demonstrating a direct relationshipbetween GZP TBR for a treatment group and percent survival for agiven treatment group.
Combination therapy dose schedule assessmentBecause GZP PET could successfully predict the efficacy of
various combinations of therapy, we next sought to determineif it could also quantify the effectiveness of the order in which asingle combination was delivered. The combination of anti-PD-1and anti-CTLA-4 was given either repeatedly as was done with theprevious experiments (PCþPC), as a single concurrent combina-tion followed by single agent anti-PD-1 administration (PCþP),or as a single agent anti-CTLA-4 therapy given once followed byanti-PD-1 therapy (CþP). PET imaging of each of these groups
revealed statistically similar TBRs (Fig. 5D; PCþPC¼ 1.83� 0.18,PCþP ¼ 1.98 � 0.29, CþP ¼ 1.65 � 0.21). As predicted by GZPPET imaging, neither mean tumor volume (Fig. 5E) nor overallsurvival (Fig. 5F) were different between any of the treatmentgroups. The predictive nature of GZP PET imaging in assessing thetiming of anti-PD-1 and anti-CTLA-4 therapy suggests it can alsobe used to determine when combinations should be given, inaddition to which therapies are most effective.
DiscussionIn this article, we demonstrate GZP PET imaging to be a
promising method for measuring active antitumor immuneresponse and for comparing novel checkpoint inhibitors andcombination regimens. By classifying tumors as responders basedon GZP PET signal intensity, we establish that granzyme B PETimaging is highly sensitive and selective for response across avariety of tumor models and treatments. Even comparison ofvehicle-only groups revealed that CT26 tumors had low baselinelevels of granzyme B and moderate responsiveness to immuno-therapy, whereas MC38 had high baseline levels of granzyme Band high responsiveness to immunotherapy. In fact, 1 MC38vehicle-treated tumor that was identified as having levels ofgranzyme B consistent with response spontaneously regressed,even without therapy. It has been demonstrated that MC38tumors have aDNAmismatch repair deficiency phenotype, whichleads to increased tumor mutational burden and thereforeincreased probability of immune recognition and high baselinegranzyme B as seen in this study (19). Our data support that amore inflamed tumor microenvironment induces high granzymeB secretion even at baseline, and GZB PET could serve as apotential pre-treatment biomarker of response for mismatchrepair deficient tumors.
Across both the MC38 and CT26 treatment groups, GZP PEThad 94% negative predictive value. The high negative predictivevalue indicates that the absence of granzymeB is a strong predictorof failure to activate an antitumor response. Conversely, highGZPPET may not always signify response, as immune cell activationmaynot be enough to clear a tumor in cases of rapid tumor growthor a strong tumor evasion. However, an 84% positive predictivevalue suggests that this outcome occurs in a minority of cases.
In addition to individual response, we also illustrate thatsubsequent survival for each therapy group correlates with theaverage PET uptake of that cohort, suggesting that GZP PETimaging could provide a method for early comparison of novelimmunotherapy regimens. The most effective therapy of thisstudy, anti-PD-1 and anti-CTLA-4 in combination, yielded highermean GZP PET uptake in comparison to the other 2 regimens,
Table 1. Summary of all tumor and treatment types analyzed by GZP PET
NOTE: Data are presented for sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of GZP PET for each individual tumor andtreatment combination, with a global summary in the bottom row.
Larimer et al.
Clin Cancer Res; 25(4) February 15, 2019 Clinical Cancer Research1202
and subsequently resulted in higher rates of response. Anunexpected diminishment in therapeutic efficacy of PD-1 andTIM-3 combination therapy especially highlights the potentialof granzyme B PET imaging to detect and predict combinationsthat may not be effective. In addition, despite significantdifferences in the immunogenicity of each of the cell linesmonitored, a combined analysis of both tumor types revealed anonzero slope and a linear correlation between granzyme B
levels and percent response, which corroborates the hypothesisthat granzyme B PET imaging can be used to determine ther-apeutic effectiveness on a broad scale.
GZP PET also permitted a rapid evaluation of precise dosetiming of anti-PD-1 and anti-CTLA-4 therapy, with GZP PETaccurately predicting that a single administration of anti-CTLA-4followed by anti-PD-1 therapy was as effective as repeated con-current administration of anti-PD-1 and anti-CTLA-4 therapy. This
Figure 5.
A and B, Kaplan–Meier curves for each tumor and therapy combination with average TBR included in parentheses next to each label. C, Linear plot of overallresponse versus TBR for each tumor/treatment combination,whereR2¼0.84 (P<0.0001) anddotted lines represent the 95%confidence interval.D,GZPPETTBRofconcurrent anti-PD-1 and anti-CTLA-4 therapy (PCþPC, blue, n ¼ 14), single concurrent anti-PD-1 and anti-CTLA-4 followed by single agent anti-PD-1 therapy(PCþP, green, n ¼ 8), single-dose anti-CTLA-4 therapy followed by anti-PD-1 monotherapy (CþP, purple, n ¼ 14) and vehicle treatment (Veh, black, n ¼ 8) aredenoted by bars representing the mean of 8 to 14 mice for each group. Error bars represent SEM (E) average tumor volumes of CPþCP (blue circle), CPþP(green square), CþP (purple triangle), and Vehicle (black triangle) treated tumors, with error bars representing SEM. F, Kaplan–Meier curve of the 3treatment schedules and vehicle-treated tumors. �� , P < 0.01.
Granzyme B PET Immunotherapy Response Quantification
www.aacrjournals.org Clin Cancer Res; 25(4) February 15, 2019 1203
is concordant with other studies that demonstrate the timing ofcheckpoint inhibitor therapy is crucial (7, 23). GZP PET imagingcan be used to gather early evidence regarding proper timing ofcombination therapy delivery, providing a personalized approachto an adaptive immunotherapy regimen
Because of the favorable pharmacokinetics and high degree ofsimilarity between the murine and human versions of GZP, ourimaging agent provides a readily available route of clinical trans-lation. Previous analysis by our group suggests that differences inthe granzyme B expression between responding and nonrespond-ing patients can be seen as early as 21 days posttherapy initiation(14). Taken together, these data justify planned clinical trialsutilizing the human version of the peptide. This is an importantnext step, as GZP PET imaging has, to date, only been tested insubcutaneousmurinemodels, whichmay have differentmechan-isms of response to immunotherapy. Additionally, the datashould be expanded to ensure additional tumor models, immu-notherapy regimens, image timing, and dosing schedules in largersample sizes and over longer time frames continue to follow thesame linear relationship between GZP uptake and tumor respon-siveness. GZP PET holds promise as a significant addition to theevolving immuno-oncology landscape, both for its ability topredict individual response to therapy and for its ability toevaluate novel therapies and combination regimens as theyemerge.
Disclosure of Potential Conflicts of InterestB.M. Larimer and E. Wehrenberg-Klee have ownership interests (including
patents) at and are consultant/advisory boardmembers forCytosite Biopharma.
G. Boland reports receiving commercial research support from Takeda Oncology.U. Mahmood is an employee of, has ownership interests (including patents) in,and is a consultant/advisoryboardmembers forCytositeBiopharma.Nopotentialconflicts of interest were disclosed by the other authors.
Authors' ContributionsConception and design: B.M. Larimer, E. Bloch, E. Wehrenberg-Klee,U. MahmoodDevelopmentofmethodology:B.M.Larimer, S.Nesti,E.E.Austin, E.Wehrenberg-Klee, U. MahmoodAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): B.M. Larimer, E. Bloch, S. Nesti, E.E. Austin, G. BolandAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): B.M. Larimer, E. Bloch, S. Nesti, E.E. AustinWriting, review, and/or revision of the manuscript: B.M. Larimer, E. Bloch,S. Nesti, E.E. Austin, E. Wehrenberg-Klee, U. MahmoodAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): E. Bloch, G. Boland, U. MahmoodStudy supervision: U. Mahmood
AcknowledgmentsThe authors would like to thank Sengchan Khamhoung and Catharina
Dekker for technical assistance in preparing the manuscript. This publicationis based on research supported by the Melanoma Research Alliance, and NIHgrants R01CA214744, K99CA215604, P50CA090381.
The costs of publication of this articlewere defrayed inpart by the payment ofpage charges. This article must therefore be hereby marked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Received July 26, 2018; revised September 5, 2018; accepted October 12,2018; published first October 16, 2018.
References1. Herbst RS, Baas P, Kim DW, Felip E, Perez-Gracia JL, Han JY, et al.
Pembrolizumab versus docetaxel for previously treated, PD-L1-positive,advanced non-small-cell lung cancer (KEYNOTE-010): a randomised con-trolled trial. Lancet 2016;387:1540–50.
2. Hodi FS,O'Day SJ,McDermott DF,Weber RW, Sosman JA,Haanen JB, et al.Improved survivalwith ipilimumab in patients withmetastaticmelanoma.N Engl J Med 2010;363:711–23.
3. Massard C, GordonMS, Sharma S, Rafii S, Wainberg ZA, Luke J, et al. Safetyand efficacy of durvalumab (MEDI4736), an anti–programmed cell deathligand-1 immune checkpoint inhibitor, in patients with advanced urothe-lial bladder cancer. J Clin Oncol 2016;34:3119–25.
4. Wolchok JD, Hoos A, O'Day S, Weber JS, Hamid O, Lebb�e C, et al.Guidelines for the evaluation of immune therapy activity in solidtumors: immune-related response criteria. Clin Cancer Res 2009;15:7412–20.
5. Marrone KA, Ying W, Naidoo J. Immune-related adverse events fromimmune checkpoint inhibitors. Clinical Pharmacol Ther 2016;100:242–51.
6. PostowMA, Chesney J, Pavlick AC, Robert C, Grossmann K,McDermott D,et al. Nivolumab and ipilimumab versus ipilimumab in untreated mela-noma. N Engl J Med 2015;372:2006–17.
7. Messenheimer DJ, Jensen SM, Afentoulis ME, Wegmann KW, Feng Z,Friedman DJ, et al. Timing of PD-1 blockade is critical to effective com-bination immunotherapy with anti-OX40. Clin Cancer Res 2017;23:6165–77.
8. Koyama S, Akbay EA, Li YY, Herter-Sprie GS, Buczkowski KA, RichardsWG,et al. Adaptive resistance to therapeutic PD-1 blockade is associated withupregulation of alternative immune checkpoints. Nat Commun 2016;7:10501.
9. Huang Y-H, Zhu C, Kondo Y, Anderson AC, Gandhi A, Russell A, et al.CEACAM1 regulates TIM–3–mediated tolerance and exhaustion. Nature2015;517:386–90.
10. Gorman JV, Starbeck-Miller G, Pham N-LL, Traver GL, Rothman PB, HartyJT, et al. Tim-3 directly enhances CD8 T cell responses to acute Listeriamonocytogenes infection. J Immunol 2014;192:3133–42.
11. Herbst RS, Soria J-C, Kowanetz M, Fine GD, Hamid O, Gordon MS, et al.Predictive correlates of response to the anti-PD-L1 antibody MPDL3280Ain cancer patients. Nature 2014;515:563–67.
12. Dudley JC, Lin M-T, Le DT, Eshleman JR. Microsatellite instability as abiomarker for PD-1 blockade. Clin Cancer Res 2016;22:813–20.
13. Rizvi NA, Hellmann MD, Snyder A, Kvistborg P, Makarov V, Havel JJ, et al.Mutational landscape determines sensitivity to PD-1 blockade in non–small cell lung cancer. Science 2015;348:124–8.
14. Larimer BM,Wehrenberg-Klee E, Dubois F, Mehta A, Kalomeris T, FlahertyK, et al. Granzyme B PET imaging as a predictive biomarker of immuno-therapy response. Cancer Res 2017;77:2318–27.
15. Larimer BM, Wehrenberg-Klee E, Caraballo A, Mahmood U. QuantitativeCD3 PET imaging predicts tumor growth response to anti-CTLA-4 therapy.J Nucl Med 2016;57:1607–11.
16. Tavar�e R, Escuin-OrdinasH,Mok S,McCrackenMN, Zettlitz KA, Salazar FB,et al. An effective immuno-PET imaging method to monitor CD8-dependent responses to immunotherapy. Cancer Res 2016;76:73–82.
17. Selby MJ, Engelhardt JJ, Johnston RJ, Lu L-S, Han M, Thudium K,et al. Preclinical development of ipilimumab and nivolumab com-bination immunotherapy: mouse tumor models, in vitro functionalstudies, and cynomolgus macaque toxicology. PLoS One 2016;11:e0161779.
18. Lau J, Cheung J, Navarro A, Lianoglou S, Haley B, Totpal K, et al. Tumourand host cell PD-L1 is required to mediate suppression of anti-tumourimmunity in mice. Nat Commun 2017;8:14572.
19. Efremova M, Rieder D, Klepsch V, Charoentong P, Finotello F, HacklH, et al. Targeting immune checkpoints potentiates immunoeditingand changes the dynamics of tumor evolution. Nat Commun 2018;9:32.
Clin Cancer Res; 25(4) February 15, 2019 Clinical Cancer Research1204
20. Kang C-W, Dutta A, Chang L-Y, Mahalingam J, Lin Y-C, Chiang J-M, et al.Apoptosis of tumor infiltrating effector TIM-3þCD8þ T cells in coloncancer. Sci Rep 2015;5:15659.
21. Holmgaard RB, Zamarin D, Lesokhin A, Merghoub T, Wolchok JD. Target-ing myeloid-derived suppressor cells with colony stimulating factor-1receptor blockade can reverse immune resistance to immunotherapy inindoleamine 2,3-dioxygenase-expressing tumors. EBioMedicine 2016;6:50–8.
22. Buggert M, Tauriainen J, Yamamoto T, Frederiksen J, Ivarsson MA,Micha€elsson J, et al. T-bet and EOMES are differentially linked to theexhausted phenotype of CD8þ T cells in HIV infection. PLoS Pathog2014;10:e1004251.
23. Weber JS, Gibney G, Sullivan RJ, Sosman JA, Slingluff CL, Jr, Lawrence DP,et al. Sequential administration of nivolumab and ipilimumab with aplanned switch in patients with advanced melanoma (CheckMate 064): anopen-label, randomised,phase2 trial. TheLancetOncology2016;17:943–55.
www.aacrjournals.org Clin Cancer Res; 25(4) February 15, 2019 1205
Granzyme B PET Immunotherapy Response Quantification
2019;25:1196-1205. Published OnlineFirst October 16, 2018.Clin Cancer Res Benjamin M. Larimer, Emily Bloch, Sarah Nesti, et al. ImagingAdministration Timing Can Be Measured by Granzyme B PET The Effectiveness of Checkpoint Inhibitor Combinations and
Updated version
10.1158/1078-0432.CCR-18-2407doi:
Access the most recent version of this article at:
