ImmunoPET Imaging with ⁸⁹Zr-Labeled Atezolizumab Enables In Vivo Evaluation of PD-L1 in Tumorgraft Models of Renal Cell Carcinoma

This article is featured in Highlights of This Issue, p. 4833

Renal cell carcinoma (RCC) accounts for approximately 4% of all cancer cases in the United States, with estimated 79,000 new cases and 13,920 deaths in 2022 (1). The treatment landscape for RCC has changed dramatically over the past 7 years with the introduction of immune checkpoint inhibitor (ICI) therapy (2–7). Among immune checkpoint proteins leveraged for immunotherapy, the programmed cell death 1 receptor (PD-1) and its ligand (PD-L1) play a central role (8, 9). PD-1/PD-L1 dampen the host immune system to protect against autoimmunity (8, 10). While PD-1 expression is restricted to activated hematopoietic cells including lymphocytes and antigen-presenting cells, PD-L1 is found in both hematopoietic and non-hematopoietic tissues (11). In many malignancies, the PD-1/PD-L1 pathway is hijacked by cancer cells to suppress an antitumor immune response (12). To date, the FDA has approved six mAbs targeting PD-1 or PD-L1 across 19 cancer types and two tissue-agnostic conditions (13).

Clinical responses to ICI therapies are heterogeneous. While long-lasting responses occur in a subset of patients with RCC, the disease ultimately progresses in over 50% (2, 7, 9, 14–16). Predictive biomarkers are lacking and this represents an unmet medical need. Tissue-based PD-L1 analyses have failed to show consistent results. In CheckMate 025, a trial leading to FDA approval of nivolumab for patients with previously treated metastatic RCC, a correlation between overall survival and PD-L1 expression was not observed (7). The CheckMate 214 study, which compared the combination of an anti-PD-1 mAb (nivolumab) with an anti-CTLA4 mAb (ipilimumab) to standard-of-care sunitinib for patients with naïve metastatic RCC, showed improved survival in patients with baseline PD-L1 tumor expression (≥ 1% PD-L1), but this failed to reach statistical significance (2). These results likely reflect confounding factors. Tissue-based PD-L1 analyses are limited by the samples utilized. Biopsies provide minimal amounts of tissue from a single site, and given PD-L1 heterogeneity, the results may not be representative. An alternative is afforded by surgical resection specimens, which enable broader studies, but the specimens are often dated and may not be representative of the remaining sites of disease in the patient. In addition, IHC assays are limited by the particular anti-PD-L1 antibody utilized, and substantial heterogeneity has been observed (17, 18). Finally, there are limitations with scoring pertaining to the type of cell, signal intensity and localization, as well as the percentage of positive cells used for positivity or negativity.

Recent technological advances in immuno–positron emission tomography (immunoPET) have made it possible to noninvasively and quantitatively assess immune system pathways in real time. Indeed, immunoPET imaging of PD-L1 with a ⁸⁹Zr-labeled deferoxamine conjugate of atezolizumab, a fully humanized therapeutic anti-PD-L1 mAb with a mutant Fc fragment ([⁸⁹Zr]Zr-DFO-ATZ), has been evaluated in a clinical trial in patients with bladder, breast, and non–small cell lung cancer (19). Despite the small size (n = 22), tumoral PET outperformed IHC as a predictive marker of response to ATZ in this first-in-human study of immunoPET imaging of PD-L1.

We explored immunoPET with [⁸⁹Zr]Zr-DFO-ATZ in RCC and previously reported a case study showing increased uptake of [⁸⁹Zr]Zr-DFO-ATZ in a patient-derived tumorgraft (TG) mouse model corresponding to a patient with metastatic RCC, who experienced a remarkable and lasting response to nivolumab (20). In this work, we present systematic analyses of RCC TG lines (n = 8) expressing a wide range of PD-L1 (0%–85%), including the case study. Through a double-blind analysis, we evaluated the correlation of tumoral PET signal with corresponding PD-L1 by IHC to validate the use of immunoPET for noninvasive PD-L1 assessment.

The preparation of DFO-ATZ was carried out by our previously reported procedure with minor modifications (20). Specifically, we introduced a liquid chromatography/mass spectrometry (LC/MS) method to quantify the molar ratio of DFO (Macrocyclics, #B-705) versus ATZ (Genentech/Roche) in the conjugate. For each batch of DFO-ATZ production, our quality control procedures included assessment of chemical purity by high-performance liquid chromatography (HPLC), radiolabeling efficiency with ⁸⁹Zr radionuclide, and binding specificity to the target PD-L1 protein by cell assays.

The ⁸⁹Zr labeling of DFO-ATZ was performed in the Cyclotron and Radiochemistry Program at University of Texas Southwestern Medical Center (UT Southwestern, Dallas, TX) using a GE PETtrace 880 cyclotron (GE Healthcare) for in-house ⁸⁹Zr production following current good manufacturing practices for radiotracer production for clinical use.

The averaged molar ratio of DFO/ATZ in the prepared DFO-ATZ was determined by LC/MS. The sample was loaded to an Agilent 6540 Q-TOF LC/MS system equipped with an Agilent 300 SB-C3 RRHD 1.8 μm LC column, 2.1 × 100 mm. The HPLC gradient was set to: 90% to 80% A [0.1% formic acid (Sigma-Aldrich, #399388) in water] from 0 to 4 minutes, 80% to 10% A from 4 to 5 minutes, and then 90% to 100% B [0.1% formic acid in acetonitrile (Sigma-Aldrich, #AX0156)] from 5 to 10 minutes. Dual Agilent Jet Stream Electrospray Ionization was used as the time-of-flight mass spectrometry ion source, and the sample was run at the positive polarity mode. Profile MS data were acquired for protein deconvolution data analysis using the Agilent MassHunter Qualitative Analysis with BioConfirmProteinDigest version B06.00. From the deconvolution results, peaks corresponding to unmodified ATZ and DFO-ATZ were manually integrated. The area of each peak was obtained for the calculation of DFO number per ATZ: N = R₀ × 0 + R₁ × 1 + R₂ × 2 + … + R_n × n. (N is the chelator number per ATZ; R₀, R₁, R₂ or R_n is the percentage of the peak area corresponding to DFO-ATZ with 0, 1, 2, or n of DFO chelator, respectively).

Binding affinity experiments utilized the PD-L1–expressing A20 lymphoma cell line (ATCC, #TIB-208, RRID: CVCL_1940), which was maintained at 37°C in 5% CO₂ in RPMI1640 medium (Sigma-Aldrich, #R8758), supplemented with 10% FBS (Sigma-Aldrich, #F2442) and 1% penicillin-streptomycin antibiotics (Sigma-Aldrich, #P0781). Cells were incubated with the DFO-ATZ conjugate at different dilutions (from 10 μg/mL to 5 pg/mL, 5-fold dilutions) for 20 minutes at room temperature. After incubation, the cells were washed twice with 1× phosphate buffered saline (PBS; Sigma-Aldrich, #D8537), and stained by PE-anti-human IgG secondary antibody (BioLegend, #366903, RRID: AB_2876689). Cells were then washed again with PBS and analyzed using a CytoFLEX (Beckman Coulter) flow cytometer. The binding affinity was determined by estimating the EC₅₀ values from mean fluorescent intensity using GraphPad Prism 8.0.

Eight RCC patient-derived TG lines with varying levels of PD-L1 expression were expanded using NOD/SCID (NOD.CB17-Prkdc^scid) mice by surgical implantation of the tumors in the shoulder flank or thigh regions (21). They were as follows: XP783 [cohort (c) 17: 60%–85% PD-L1, n = 3], XP945 (c1: 40%–60%, n = 3; c6: 55%–70% PD-L1, n = 3), XP925 (c1: 10%–15% PD-L1, n = 3), XP813 (c8: 1%–5% PD-L1, n = 3), XP955 (c1: 60%–70% PD-L1, n = 3), XP818 (c24: 0% PD-L1, n = 4), XP258 (c29: 1%–10% PD-L1, n = 2), and XP490 (c25: 40%–75% PD-L1, n = 3). The percentage of PD-L1 expression reported was obtained through ex vivo IHC analysis of the explanted TG tumors after immunoPET. Throughout the experiment, animals were housed in laminar flow cages maintained at approximately 22°C with approximately 55% relative humidity in a 12-hour light/dark cycle. Mice were permitted free access to autoclaved water and commercial food throughout the experiment. When tumor sizes reached ≥ 50 mm³, immunoPET scans were performed on the tumor-bearing mice, from whom the tumors would be later excised for IHC analysis at the end of imaging. Imaging reconstruction and quantitative analyses were conducted by the immunoPET team, while histology studies were completed by the histopathology core of the Kidney Cancer Program. Experiments were based on an Institutional Animal Care and Use Committee protocol in compliance with the United States Public Health Service Standards and NIH guidelines.

ImmunoPET scans were performed on a Siemens Inveon PET/CT Multi-modality System with an effective spatial resolution of 1.4 mm at the center of field of view. Each mouse received approximately 3.7 MBq of [⁸⁹Zr]Zr-DFO-ATZ (equivalent to ∼29 to 52 μg of DFO-ATZ) intravenously via the tail vein (day 0). Static PET scans were performed from day 1 up to day 7 postinjection (p.i.) for 15 to 45 minutes on mice anesthetized with isoflurane. CT scanning was performed to obtain the required anatomic information for data analysis. CT projections (360 steps/rotation) were acquired with a power of 80 kVp, current of 500 μA, exposure time of 142 ms, binning of 4, and effective pixel size of 102 μm. The acquired static PET data were reconstructed using the three-dimensional Ordered Subsets Expectation Maximization (OSEM3D/MAP) algorithm. The CT reconstruction protocol with a downsample factor of 2 was set to interpolate bilinearly and used a Shepp-Logan filter. The obtained PET and CT images were coregistered using the manufacturer's software package for image and data analysis. As part of the double-blinded study, PET data quantification was blinded to the grouping information of the TG-bearing mice. Guided by CT, the regions of interest (ROI) were drawn manually, encompassing the coronal, axial, and sagittal planes of the tissue. The uptake of [⁸⁹Zr]Zr-DFO-ATZ in each ROI was calculated as percentage injected dose per gram (% ID/g).

A voxel-based analysis of PET ROI data (22) was performed to evaluate the intratumoral uptake distribution of [⁸⁹Zr]Zr-DFO-ATZ using two representative TG lines (XP945, n = 6 and XP818, n = 4). The resultant voxel-based data were exported in excel, in which histograms of data were generated (0.5% ID/g bin width, starting from 0.25% ID/g) using GraphPad Prism 8.0. Because the absolute frequency of voxels falling into a bin partly depends on the number of voxels analyzed, the probability distribution was then calculated by normalizing the histograms to the total number of voxels analyzed, which varied with tumor size. A mixed model statistical analysis was performed to compare the radiotracer uptake curves using a compound symmetric covariance structure for PET voxels measured within the same mouse, where the uptake in the muscle was used as the background signal (nonspecific uptake) in each mouse.

After PET/CT scans, the animals were sacrificed. TGs were collected immediately, fixed in 10% neutral-buffered formalin for 48 hours, and then stored in 70% ethanol at −20°C. The samples were processed and embedded in paraffin blocks for histology and PD-L1 IHC after the radioactivity decayed (∼33 days). Prior to fixation, the tumors were measured for volume estimations (length, width, and height). PD-L1 (1:300, BioCare, #ACI 3171A) IHC was performed using a Dako Autostainer Link 48 after its standardization and validation using 10 clear-cell RCC (ccRCC) tumors with a wide range of PD-L1 expression as controls (Dako, #22C3 antibody) from our Clinical Laboratory Improvement Amendments–certified laboratory. Positive and negative controls with known PD-L1 expression were used with each run of immunostaining. Combined positive scores (CPS) were recorded as the number of PD-L1 staining cells (both tumor and other stromal cells) divided by the total number of tumor cells evaluated, multiplied by 100 (23). The analysis was performed by a trained pathologist (P. Kapur) blinded to other results using recommended guidelines (https://pathology-education.agilent.com/en-us/home.html).

Under the purview of an Institutional Review Board–approved protocol (STU 012011-190), clinical and pathologic information of the 8 patients contributing tissues for TG generation was retrospectively compiled. Histologic data were confirmed by a genitourinary pathologist (P. Kapur). For patients who received immunotherapy (n = 4), International Metastatic RCC Database Consortium (IMDC) risk scores were calculated at the time of immunotherapy initiation. Best objective response to ICI was determined using principles from the RECIST version 1.1. Tumor burden was reported as the sum of the longest diameters of target lesions. For clinical data analyses, baseline clinicopathologic characteristics were tabulated and evaluated using descriptive statistics.

GraphPad Prism 8.0 was used for data analyses, except where mixed models were used, which were conducted in SAS 9.4 (SAS). All data analyses were completed at an alpha value = 0.05 and two-tailed P values without correction for multiple comparisons. Mixed-effects correlation analyses were performed using the method described by Lam and colleagues (24), which removes the variation due to individual TGs. A mixed model statistical analysis was performed to evaluate the association of tumor uptake values and tumor-to-muscle (T/M) uptake ratios, individually, with the corresponding PD-L1 IHC measurements, using the pooled data (n = 25) from all eight RCC TG lines. These analyses were performed with a compound symmetric covariance structure for mice within the same TG model. The distribution of tumor uptake voxels (mean 4.62% ID/g for XP945 vs. 3.5% ID/g for XP818) in two representative TG lines were also compared using a mixed model analysis with a compound symmetric covariance structure for voxels from the same mouse, where the uptake in the muscle was used as the background signal (nonspecific uptake) in each mouse. In addition, we centered the voxel distributions by XP line on the same median and used the Kolmogorov–Smirnov two-sample test to determine whether there was a difference in the shapes of the two distributions.

The data used and analyzed in this article are available from the corresponding authors (J. Brugarolas and X. Sun) upon request.

We conjugated DFO to ATZ. As shown in Fig. 1A, the DFO-ATZ conjugation resulted in four species varying the DFO chelator number per ATZ molecule from 0 to 3. The apparent number of DFO per ATZ was calculated to be 0.98 according to the molecular distribution determined by LC/MS analysis. An in vitro cell assay using the PD-L1–expressing A20 lymphoma cell line (Fig. 1B) showed that the PD-L1 binding affinity of DFO-ATZ was not significantly undermined relative to the unconjugated antibody (DFO-ATZ: EC₅₀ = 21.14 ± 11.9 pmol/L versus intact ATZ: EC₅₀ = 34.9 ± 13.5 pmol/L). The immunoreactivity of [⁸⁹Zr]Zr-DFO-ATZ was determined by a Lindmo assay to be 86 ± 4% (Fig. 1C). Furthermore, the PD-L1 binding specificity of [⁸⁹Zr]Zr-DFO-ATZ was confirmed by a cell-based blocking assay, in which coincubation with approximately 100 times excess of ATZ (∼ 6 ng) completely blocked the binding of [⁸⁹Zr]Zr-DFO-ATZ (Fig. 1D).

To further evaluate [⁸⁹Zr]Zr-DFO-ATZ, we leveraged our TG resource involving >100 TG models generated from surgical specimens of patients with RCC implanted orthotopically in immunocompromised mice without additives or disaggregation (21). We selected TG lines with PD-L1 expression ranging from 0% to 85% (n = 8). TGs were derived from a clinically heterogeneous group of patients (Table 1). Two (25%) cases were derived from women and three (37.5%) from ethnic minorities. Four (50%) TGs were generated from metastatic sites [bone, spinal cord (n = 2), and retroperitoneum] and the remainder from primary tumors. Histologically, 7 (87.5%) were ccRCC and one was a papillary RCC (pRCC). Generally, TG PD-L1 expression by IHC was similar to that of the corresponding patient tumor.

TG-bearing mice were injected with approximately 3.7 MBq (equivalent to ∼ 29–52 μg of DFO-ATZ) [⁸⁹Zr]Zr-DFO-ATZ (n ≥ 3 per line, except XP258 with n = 2). This dose was selected on the basis of a previously reported dose-escalation study, where a dose-dependent biodistribution was observed upon coinjection of 30 to 90 μg of unlabeled ATZ with approximately 8 μg ¹¹¹In-labeled ATZ (25). ImmunoPET/CT imaging was performed from day 1 up to day 7. PET data were reconstructed using an OSEM3D/MAP algorithm and the PET and CT images were coregistered. Guided by the CT, ROIs were drawn encompassing coronal, axial, and sagittal planes. The uptake of [⁸⁹Zr]Zr-DFO-ATZ in each ROI was calculated as % ID/g. Different organs were evaluated and kinetic analyses were performed. Highest levels of [⁸⁹Zr]Zr-DFO-ATZ were observed in the tumors as well as heart, liver, kidneys, and spleen (Supplementary Fig. S1 and S2). PET signal demonstrated a gradual and steady decline in the heart, liver, kidneys, as well as tumors over several days, likely reflective of clearance of the antibody from the circulation over time. In contrast, bone uptake increased by 1.5- to 2-fold over the study period, which can be attributed to the accumulation of ⁸⁹Zr after lysosomal degradation of [⁸⁹Zr]Zr-DFO-ATZ, and is consistent with other reports (26, 27). As a reservoir of PD-L1–expressing immune cells, the spleen showed high and fairly consistent uptake throughout the imaging window. Muscle uptake was low (∼ 1.5% ID/g) and remained steady throughout the imaging period (Supplementary Fig. S1). Thus, we chose to focus our analyses on T/M uptake ratios (Supplementary Table S1). Given maximal systemic clearance of [⁸⁹Zr]Zr-DFO-ATZ at day 6 ± 1 p.i. (Fig. 2; Supplementary Fig. S2), this timepoint was used for correlation analyses.

Like patient RCCs, TGs grow at variable rates within their mouse hosts (21, 28), and we asked whether the PET signal may be affected by tumor size. As shown in Supplementary Fig. S3, [⁸⁹Zr]Zr-DFO-ATZ tumor uptake was fairly consistent across a broad range of tumor sizes (53–3,960 mm³) for the different TG lines.

Representative immunoPET images from seven TG lines on day 6 p.i. of [⁸⁹Zr]Zr-DFO-ATZ are shown in Fig. 2 (an image of XP258 is not shown as only one mouse survived to the end of the experiment). Following image acquisition, tumors were harvested for histologic and IHC analyses (Fig. 2). Tumor samples from the corresponding patient tumors were incorporated as a reference. To evaluate the performance of [⁸⁹Zr]Zr-DFO-ATZ immunoPET imaging, we conducted a repeated measures linear regression analysis. We compared immunoPET tumor uptake values and T/M signal ratios (Supplementary Table S1) to the corresponding IHC-derived tumor PD-L1 expression levels (n = 25 samples representing eight TG lines). The regression model accounted for the IHC variation among individual mice within a TG line. These studies showed a statistically significant association between IHC PD-L1 level and [⁸⁹Zr]Zr-DFO-ATZ immunoPET using either T/M (P = 0.0014) or tumor uptake (P = 0.030). The correlation (ρ_XY) was 0.78 for T/M with a slope of 0.029 ± 0.008 [95% confidence interval (CI) = 0.013–0.045; Fig. 3A], and 0.57 for tumor uptake with a slope of 0.014 ± 0.006 (95% CI = 0.001–0.027; Supplementary Fig. S4).

We observed intratumoral variation in both immunoPET signal and PD-L1 expression by IHC (Fig. 2). To further assess the capability of immunoPET to detect heterogeneous intratumoral PD-L1 expression, we carried out a voxel-to-voxel analysis of tumor uptake in XP945 (PD-L1 IHC 40%–70%) using the corresponding data in XP818 (PD-L1 IHC 0%) as control. A mixed model analysis was performed to compare the uptake distribution difference of [⁸⁹Zr]Zr-DFO-ATZ in these two TG lines, in which the uptake in the muscle was used as nonspecific signal in each mouse to level out the background differences among mice. The analysis revealed not only a statistically significant difference (P = 0.0024) in the intratumoral voxel distributions of [⁸⁹Zr]Zr-DFO-ATZ, but also a difference in the shapes of the distributions centering on the same median (Kolmogorov–Smirnov P < 0.0001; Fig. 3B). While the radiotracer's distribution in XP818 showed a relatively narrow peak with low PET signal intensity, the probability distribution curve in XP945 was wide and right shifted, suggestive of heterogeneous PD-L1 expression within the tumor (Fig. 3B; Supplementary Table S2).

Among the TGs included in this study, there were four derived from metastatic sites and 3 of these patients received ICI therapy. Among the rest, XP945 was generated from a T4 tumor invading into the adrenal gland that was resected and also received ICI therapy for metastatic disease. In total, 4 (50%) TG lines in this study were derived from patients who received ICI therapy. Clinical characteristics of these patients at the time of ICI initiation are summarized in Table 2. One patient received ICI as first-line therapy, while the remainder received ICI as third or later line. All patients received nivolumab monotherapy and 1 had nivolumab/ipilimumab combination after progression on nivolumab monotherapy. Best responses to ICI included a partial response in 1 patient, a mixed response in 1 patient, and progressive disease in 2 patients (Fig. 4A and B). The TG derived from patient XP945, who had a partial response to ICI (20), had the highest levels of immunoPET signal among all four TGs examined (Fig. 4B; corresponding to high PD-L1 by IHC on both the TG and patient tissue). Interestingly, patient XP818 experienced a mixed response to immunotherapy with shrinkage of pulmonary lesions but progression in the central nervous system (CNS). The XP818 TG line was derived from a new lesion that developed on therapy (after ICI) and had very low PD-L1 enhancement by immunoPET which corresponded with 0% PD-L1 expression by IHC.

ImmunoPET, which evolved from conventional radioimmunodiagnostics, is a promising imaging technique for in vivo tumor characterization with potential applications in ICI treatment selection and monitoring treatment response (13, 29, 30). IHC-measured PD-L1 expression has been shown to predict ICI response in several tumor types, but there is conflicting evidence for RCC (14, 16, 31). We hypothesized that this may be due to limitations of tissue-based PD-L1 analyses which oftentimes rely on minute tumor fragments or archival tissue that may not be representative of the current disease in the patient. ImmunoPET, which delivers noninvasive information for all lesions in real-time, offers a potential alternate strategy. In one clinical study evaluating ATZ monotherapy in a variety of tumor types, immunoPET with [⁸⁹Zr]Zr-DFO-ATZ was found to correlate more strongly with treatment response than PD-L1 expression by IHC (19). However, the utility of immunoPET in RCC remains to be determined. Herein, we report preclinical data evaluating immunoPET with [⁸⁹Zr]Zr-DFO-ATZ for noninvasive PD-L1 imaging in RCC patient-derived TG models expressing a wide range of PD-L1.

While the preparation of [⁸⁹Zr]Zr-DFO-ATZ has been previously reported by our group (20) and others (19, 27, 32), we observed batch-to-batch variations in the immunoreactivity and radiolabeling efficiency of the antibody conjugates. We found that the main determinant is the molar ratio of DFO:ATZ, which is preferably 1:1. At this ratio, the binding affinity of DFO-ATZ to PD-L1 was indistinguishable to that of intact ATZ as measured by a cell-based assay. The analytic method we developed herein using LC/MS provides a rapid measurement to assess the “apparent number of DFO per ATZ,” which assists in the quality control of DFO-ATZ synthesis and reduces the batch-to-batch variation. This method has been implemented in our ongoing clinical trial (NCT04006522).

TGs, which preserve histologic and molecular features from the corresponding patient tumors, are an invaluable tool for the development of diagnostic and therapeutic agents (21). Using such TGs, we were able to evaluate immunoPET with [⁸⁹Zr]Zr-DFO-ATZ in RCC. Kinetic analyses determined that optimal timing for immunoPET imaging was approximately 7 days following tracer administration. Inasmuch as ATZ can bind to mouse PD-L1, this platform also allowed the measurement of [⁸⁹Zr]Zr-DFO-ATZ uptake across normal tissues. Consistent with this notion, high levels were observed in the spleen. Of note, this is likely driven by the fact that the Fab portion of the mAb is mutated. In contrast, [⁸⁹Zr]Zr-DFO-ATZ uptake in the muscle was relatively low and remained constant over time. Thus, we normalized tumor uptake to muscle uptake. The T/M ratios strongly correlated with PD-L1 IHC measurements on explanted tumors. Absolute tumor uptake of [⁸⁹Zr]Zr-DFO-ATZ also showed a significant, albeit weaker, correlation.

We retrospectively assessed the clinical responses of 4 patients who received ICI therapy. PD-L1 expression in the TGs, both by IHC and immunoPET, tended to correlate with clinical outcomes. One noteworthy exception is the patient from whom XP818 was derived. This patient had a mixed response to therapy with shrinkage in pulmonary metastases but progression in the CNS. The corresponding TG line was derived from a spinal cord metastasis that had developed on therapy and had low PD-L1 levels (by both IHC and radiotracer uptake).

Heterogeneity may exist not only across sites of disease, but also within tumors (33, 34). To address whether immunoPET can identify this heterogeneous PD-L1 target expression, we performed voxel-based quantitative analyses. Even in the absence of PD-L1 expression, the intratumoral distribution of the radiotracer was wide. However, there was not only a rightward shift, but also a broadening of the radiotracer uptake distribution in PD-L1–expressing tumors, and the difference between the two distribution curves was statistically significant. This suggests that immunoPET may also be helpful in assessing intratumoral heterogeneity, which was indeed observed in a clinical trial participant recently reported by our group (21).

Tumoral PD-L1 expression can be labile and dependent on several variables (35–37). The PD-1/PD-L1 pathway has been reported to have an inverse relationship with the VEGF pathway in ccRCC (38). Inflammatory cytokines, such as IFNγ and TNFα, upregulate PD-L1 (39). Moreover, PD-L1 expression can be altered by antineoplastic agents (40, 41) as well as radiotherapy (42). Inasmuch as immunoPET is noninvasive, it can be used to probe in vivo determinants of PD-L1 expression over time. In addition, immunoPET may also shed light into mechanisms of resistance by identifying instances of PD-L1 downregulation, as well as resistant lesions that retain PD-L1 expression (suggesting an alternative mechanism of resistance; refs. 15, 43, 44).

This study has several limitations. First, although both immunoPET and IHC were intended for the measurement of PD-L1, the imaging and ex vivo procedures were neither performed at the same time, nor in the same biological context. As PD-L1 expression is dynamic and heterogeneous in tumors, discrepancies are expected between the two measurements. This was particularly true in tumors with low PD-L1 expression wherein the immunoPET readouts were higher than the IHC measurements. Possible explanations include radiotracer retention in the tumor blood pool and potential residualization of ⁸⁹Zr-containing catabolites in tumor cells after lysosomal degradation of [⁸⁹Zr]Zr-DFO-ATZ (45). Whether these issues will be confounding factors in patients remains to be seen, but they could limit the ability to distinguish small differences in PD-L1 levels. Nevertheless, our double-blinded study revealed a statistically significant (P = 0.0014) association between immunoPET quantification of PD-L1 expression and its corresponding IHC. In addition, the clinically relevant thresholds of immunoPET uptake remain to be determined. PD-L1 is expressed in a variety of cell types, and the contribution of the tumor versus tumor microenvironment compartments to the overall uptake cannot be dissected using [⁸⁹Zr]Zr-DFO-ATZ immunoPET. Considering this, we used the CPS IHC scoring method, which reports PD-L1 expression across all cell types, and not the tumor proportion score (23, 46, 47). Finally, though patient-derived TG mouse models have been shown to recapitulate the host (patient) molecular features, they are grown in immunocompromised mouse hosts, which limit their application in evaluating ICIs (21). Despite these limitations, immunoPET appears to be capable of monitoring PD-L1 expression in RCC and holds promising potential as a predictive biomarker. Furthermore, the studies performed in this work supported an FDA IND application and a clinical trial that is currently accruing patients (NCT04006522).

P. Kapur reports personal fees from ClearNano Inc outside the submitted work. A. Christie reports grants from NCI/NIH during the conduct of the study, as well as grants from CPRIT outside the submitted work. J. Brugarolas reports personal fees from Eisai, Johnson & Johnson, Exelixis, Arrowhead, and Calithera outside the submitted work. No disclosures were reported by the other authors.

A. Mulgaonkar: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. R. Elias: Data curation, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. L. Woolford: Validation, investigation, methodology, writing–review and editing. B. Guan: Investigation, methodology, writing–review and editing. K. Nham: Data curation, software, methodology, writing–review and editing. P. Kapur: Resources, data curation, formal analysis, supervision, investigation, methodology, writing–review and editing. A. Christie: Resources, data curation, software, formal analysis, supervision, investigation, methodology, writing–review and editing. V.T. Tcheuyap: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. N. Singla: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. I.A. Bowman: Formal analysis, investigation, methodology, writing–review and editing. C. Stevens: Investigation, methodology, writing–review and editing. G. Hao: Software, formal analysis, investigation, methodology, writing–review and editing. J. Brugarolas: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. X. Sun: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing.

We acknowledge the patients that donated tissues enabling these studies. This research was supported by grants from the V Foundation (T2018-011, I.A. Bowman, X. Sun, J. Brugarolas), Cancer Prevention and Research Institute of Texas (RP110771 and RP170638, X. Sun), the Kidney Cancer Research Program of the Department of Defense (W81XWH-19-1-0711, X. Sun), and National Institutes of Health (P50CA196516, A. Christie, P. Kapur, J. Brugarolas). We acknowledge the services provided by the Tissue Shared Resource at UT Southwestern, which is supported in part by the Harold C. Simmons Cancer Center through an NCI Cancer Center Support Grant, P30CA142543.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.