Noninvasive Imaging of Tumor PD-L1 Expression Using Radiolabeled Anti–PD-L1 Antibodies

T cells play an essential role in the anticancer immune response. T-cell activation depends on the initial antigen-specific signal, presented via the antigen-loaded MHC to the T-cell receptor, and on activation of the costimulatory molecule CD28 by binding of CD80/86. T cells also express coinhibitory molecules that are capable of downregulating the immune response (1). One major coinhibitory receptor is programmed death 1 (PD-1). PD-1 has two ligands, programmed death ligand-1 (PD-L1) and PD-L2, of which PD-L1 is most widely expressed. Binding of PD-L1 to PD-1 transduces an inhibitory signal to the T cell, resulting in inhibition of T-cell proliferation, reduced secretion of effector cytokines, and potentially exhaustion. By upregulating PD-L1 expression levels, tumor cells are capable of escaping immune recognition and attack (2–4).

PD-L1 is expressed on a wide variety of tumors, including breast cancer, gastric cancer, renal cell cancer, ovarian cancer, non–small lung cancer, melanoma, and hematologic cancers (5). In general, PD-1 and PD-L1 have been demonstrated to be poor prognostic factors as high-expression levels are associated with poor outcome of cancer patients (5). Preclinical studies with anti–PD-1 and anti–PD-L1 antibodies have shown promising antitumor effects and have led to the initiation of several clinical investigations. Early clinical trials demonstrated objective and durable (≥1 year) responses in patients with treatment-refractory, advanced melanoma, renal cell carcinoma, non–small cell lung cancer, and ovarian cancer (6–11). Because of these impressive results, phase II/III studies are currently further exploring the therapeutic efficacy of these agents. Because of the impressive efficacy in melanoma patients, the FDA has recently granted accelerated approval of pembrolizumab (anti–PD-1 antibody) for the treatment of patients with advanced or unresectable melanoma following progression on prior therapies (11).

Although studies with PD-1/PD-L1–targeted therapies show encouraging results, not all patients respond to this type of treatment and there is an urgent need for a predictive biomarker. In this regard, several studies have postulated that tumor PD-L1 expression might be required for response to anti–PD-1– and anti–PD-L1–targeted therapy (7, 12–14). However, immunohistochemical analysis of PD-L1 expression as determined on archival tissue samples should be interpreted with caution, because PD-L1 expression may undergo changes due to alterations in tumor microenvironment or previous treatment. For example, IFNγ and TNFα can upregulate PD-L1 expression whereas vascular endothelial growth factor (VEGF) can downregulate tumor PD-L1 expression (15–17). Also, certain chemotherapeutics such as doxorubicin can decrease the expression of PD-L1 on the tumor cell surface, whereas radiotherapy and paclitaxel are capable of upregulating PD-L1 expression (18–20). There are no strict criteria to define PD-L1 positivity by IHC (e.g., cytoplasmic vs. membranous staining, number of positive tumor cells), and misinterpretation may occur due to heterogeneous expression within or between tumor lesions (e.g., primary vs. metastatic lesions; refs. 7, 12–14).

Molecular in vivo imaging with radiolabeled anti–PD-L1 antibodies may overcome some of the limitations of immunohistochemical analysis of PD-L1 expression in tumor biopsies. It allows measurement of PD-L1 expression at time of start of therapy, of whole tumor lesions and their metastases, thereby avoiding sampling errors and, thus, misinterpretation due to intratumoral and interlesional heterogeneity. This may result in more accurate detection of PD-L1 expression and accessibility, compared with immunohistochemical analysis. It can potentially be used as biomarker to select patients for PD-1/PD-L1–targeted therapy, and it allows longitudinal monitoring of PD-L1 expression during disease progression and treatment.

Therefore, the aim of this study was to develop a noninvasive in vivo imaging technique to measure tumor PD-L1 expression and accessibility. For this purpose, the anti–PD-L1 antibody PD-L1.3.1 was radiolabeled with Indium-111 (¹¹¹In) and tumor targeting was analyzed by SPECT/CT imaging in mice bearing subcutaneous cancer xenografts with different PD-L1 expression levels.

The breast cancer cell lines MDA-MB-231, SK-Br-3, and MCF-7 were cultured in RPMI-1640 (GIBCO, BRL Life Sciences Technologies), supplemented with 2 mmol/L glutamine (GIBCO) and 10% FCS (Sigma-Aldrich Chemie BV) at 37°C in a humidified atmosphere with 5% CO₂. SUM149 was cultured in Ham's F12 medium (GIBCO) supplemented with 5% FCS, 10 mmol/L 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES, GIBCO), hydrocortisone (1 μg/mL, Sigma-Aldrich Chemie BV), and insulin (5 μg/mL, Sigma-Aldrich Chemie BV). BT474 was cultured in RPMI-1640, 2 mmol/L glutamine, 10% FCS, and 10 μg/mL insulin.

PD-L1 expression of MDA-MB-231, SK-Br-3, SUM149, BT474, and MCF-7 cells was determined by FACS analysis. Cells were incubated with PD-L1-PE (557924; BD Biosciences) or mouse IgG1-PE (400114; Biolegend) for 30 minutes at 4°C. Cells were washed and, subsequently, analyzed using the Gallios flow cytometer (Beckman Coulter).

The murine monoclonal IgG1 antibody PD-L1.3.1 is specifically directed against human PD-L1 and does not cross react with murine PD-L1 (21). It was conjugated with isothiocyanatobenzyl-diethylenetriaminepentaacetic acid (ITC-DTPA, Macrocyclis) in 0.1 mol/L NaHCO₃, pH 9.5, at a 21-fold molar excess of ITC-DTPA, for 1 hour at room temperature (RT). Unbound ITC-DTPA was removed from the reaction mixture by dialysis against 0.1 mol/L 2-(N-morpholino)ethanesulfonic acid (MES; Sigma-Aldrich Chemie BV) buffer, followed by purification on a disposable G25M Sephadex column (PD10, GE Healthcare Life Sciences, Eindhoven), eluted with 0.25 mol/L ammoniumacetate buffer, pH 5.4 (Sigma-Aldrich Chemie BV).

DTPA-conjugated PD-L1.3.1 was incubated with ¹¹¹In (Mallinckrodt BV) in 0.5 mol/L MES buffer, pH 5.4, 20 minutes at RT, under strict metal-free conditions (22). After incubation, 50 mmol/L ethylenediaminetetraacetic acid (EDTA) was added to a final concentration of 5 mmol/L to chelate unincorporated ¹¹¹In. Labeling efficiency was determined using instant thin-layer chromatography on silica gel chromatography strips (Agilent Technologies), using 0.1 mol/L citrate buffer (Sigma-Aldrich Chemie BV), pH 6.0, as the mobile phase. In case labeling efficiency was below 95%, the reaction mixture was purified on a PD-10 column, eluted with PBS, containing 0.5% BSA (Sigma-Aldrich Chemie BV). Radiochemical purity of ¹¹¹In-DTPA-PD-L1.3.1 (¹¹¹In-PD-L1.3.1) exceeded 95% in all experiments.

The breast cancer cell lines MDA-MB-231, SK-Br-3, SUM149, BT474, and MCF-7 were cultured to confluency in 6-well plates and incubated with 32 pmol/L ¹¹¹In-PD-L1.3.1 (1 kBq) for 4 hours at 37°C in a humidified atmosphere with 5% CO₂, or for 4 hours on ice, in RPMI-1640 containing 0.5% BSA. Separate wells were coincubated with a 1,000-fold excess of unlabeled PD-L1.3.1 to determine nonspecific binding. After incubation, cells were washed with PBS and the cell-associated activity was measured in a shielded well-type gamma counter (Perkin-Elmer). Specific binding was calculated by subtracting the nonspecific binding from the total binding.

The immunoreactive fraction of ¹¹¹In-PD-L1.3.1 was determined essentially as described by Lindmo and colleagues (26). A serial dilution of MDA-MB-231 cells (3.3 × 10⁵–8.4 × 10⁷ cells/mL) in RPMI-1640 containing 0.5% BSA was incubated with 8 pmol/L ¹¹¹In-PD-L1.3.1 (0.2 kBq). Nonspecific binding was determined by adding an excess of unlabeled PD-L1.3.1 (67 nmol/L) to a duplicate of the lowest cell concentration. After 1-hour incubation at 37°C, cells were centrifuged and the activity in the cell pellet was measured in a shielded 3-inch well-type gamma counter (PerkinElmer). The inverse of the specific cell bound activity was plotted against the inverse of the cell concentration, and the immunoreactive fraction was calculated from the y-axis intercept using GraphPad Prism (version 5.03 for Windows).

MDA-MB-231 cells were cultured to confluency in 6-well plates. The 50% IC₅₀ of PD-L1.3.1 blocking of ¹¹¹In-PD-L1.3.1 binding was determined by incubating the cells for 4 hours on ice in 1 mL RPMI-1640 0.5% BSA, containing 20 pmol/L ¹¹¹In-PD-L1.3.1 (1 kBq) and increasing concentrations of unlabeled PD-L1.3.1 (1–1,000 pmol/L). After incubation, cells were washed with PBS and the cell-associated activity was measured in a gamma counter. The IC₅₀ was defined as the antibody concentration that was required to inhibit binding of the radiolabeled antibody by 50%. IC₅₀ values were calculated using GraphPad Prism.

Scatchard analysis was performed to determine the dissociation constant (K_d) of ¹¹¹In-PD-L1.3.1 and to quantitatively measure PD-L1 expression on MDA-MB-231, SK-Br-3, and SUM149 cells. Cells were cultured to confluency in 6-well plates and were incubated for 4 hours on ice with increasing concentrations ¹¹¹In-PD-L1.3.1 (3–3,000 pmol/L) in 1 mL RPMI-1640 containing 0.5% BSA. Nonspecific binding was determined by coincubation with 100 nmol/L PD-L1.3.1. After incubation, cells were washed with PBS and the cell-associated activity was measured in a shielded well-type gamma counter. The specific binding (total binding—nonspecific binding) was plotted against the bound/free ratio. Data were analyzed by linear regression to determine PD-L1 receptor density per cell and to determine the K_d of ¹¹¹In-PD-L1.3.1.

MDA-MB-231 cells were cultured in 6-well plates and were incubated for 2, 4, or 24 hours with 75 pmol/L ¹¹¹In-PD-L1.3.1 (1 kBq) in RPMI-1640 containing 0.5% BSA at 37°C in a humidified atmosphere with 5% CO₂. Nonspecific binding and internalization was determined by coincubation with 17 nmol/L unlabeled PD-L1.3.1. After incubation, acid wash buffer (0.1 mol/L HAc, 0.15 mol/L NaCl, pH 2.6) was added for 10 minutes to remove the membrane-bound fraction of the cell-associated ¹¹¹In-PD-L1.3.1. Subsequently, cells were harvested from the 6-well plates and the amount of membrane bound and internalized activity was measured in a gamma counter. Specific binding and internalization were calculated by subtracting the nonspecific binding and internalization from the total binding and internalization.

Animal experiments were performed on female BALB/c nude mice (Janvier) and were conducted in accordance with the principles laid out by the revised Dutch Act on Animal Experimentation (1997) and approved by the institutional Animal Welfare Committee of the Radboud University Nijmegen. At 6 to 8 weeks of age, mice were inoculated s.c. with 5 × 10⁶ MDA-MB-231, SK-Br-3, SUM149, MCF-7, or BT474 cells (mixed 2:1 with Matrigel, BD Biosciences; Pharmingen). Mice receiving MCF-7 or BT474 cells were, before tumor cell inoculation, implanted s.c. with a slow release estradiol pellet (0.18 mg, 60 days, Innovative Research of America) under general anesthesia (isoflurane/O₂). Experiments started when tumors reached a size of approximately 0.1 cm³.

Seven groups (n = 6) of mice with subcutaneous MDA-MB-231 xenografts received an intravenous injection of 0.2 MBq ¹¹¹In-PD-L1.3.1 (specific activity 0.4 MBq/μg) in the tail vein. To study the effect of the antibody protein dose on the biodistribution of ¹¹¹In-PD-L1.3.1, groups received increasing protein doses of PD-L1.3.1 (0.3–300 μg/mouse). Three days after injection, mice were euthanized using CO₂/O₂-asphyxiation. The biodistribution of the radiolabel was determined ex vivo. Tumor, blood, muscle, lung, heart, spleen, pancreas, intestine, kidney, liver, bone, and bone marrow were dissected and weighed. Activity was measured in a gamma counter. To determine the uptake of radiolabeled antibodies in each sample as a fraction of the injected dose, aliquots of the injected dose were counted simultaneously. The results were expressed as percentage injected dose per gram tissue (%ID/g).

Three groups (n = 6) of mice with subcutaneous MDA-MB-231 xenografts and three groups with MCF-7 xenografts received an i.v. injection of 0.2 MBq ¹¹¹In-PD-L1.3.1. Separate groups of mice were coinjected with an excess of 300 μg unlabeled PD-L1.3.1 to block PD-L1 in vivo. At 1, 3, and 7 days after injection of radiolabeled PD-L1.3.1, mice were euthanized, and the ex vivo biodistribution of radiolabeled PD-L1.3.1 was determined as described previously.

Three mice with subcutaneous MDA-MB-231 and 3 mice with subcutaneous MCF-7 xenografts received an i.v. injection of 15.5 MBq ¹¹¹In-PD-L1.3.1 (protein dose 1.5 μg). Immediately after injection and 1, 3, and 7 days after injection, images were acquired with the U-SPECT-II/CT (MILabs; ref. 23). Mice were scanned under general anesthesia (isoflurane/O₂) for 30 to 90 minutes using the 1.0-mm diameter pinhole mouse high sensitivity collimator tube, followed by a CT scan (spatial resolution 160 μm, 65 kV, 615 μA) for anatomical reference. Scans were reconstructed with MILabs reconstruction software, using an ordered-subset expectation maximization algorithm, with a voxel size of 0.2 mm. SPECT/CT scans were analyzed and maximum intensity projections (MIP) were created using the Inveon Research Workplace software (IRW, version 4.1). A 3D volume of interest was drawn around the tumor and uptake was quantified as the percentage injected dose per gram (%ID/g), assuming a tissue density of 1 g/cm³.

Tumor targeting of ¹¹¹In-PD-L1.3.1 to MDA-MB-231, SK-Br-3, SUM149, MCF-7, and BT474 xenografts was determined in 10 mice bearing subcutaneous tumors on both flanks (n = 2 mice/4 tumors/xenograft model). Mice received intravenous tail vein injections of 1 μg (10 MBq) ¹¹¹In-PD-L1.3.1 and 3 days later, mice were euthanized by CO₂/O₂ asphyxation and SPECT/CT images were acquired for 90 minutes, as described previously. Subsequently, the ex vivo biodistribution of radiolabeled PD-L1.3.1 was determined. Tumors were fixed in 4% formalin or frozen at −80°C for autoradiography and IHC.

Frozen and formalin-fixed tumor sections (5 μm) from mice injected with 10 MBq of ¹¹¹In-PD-L1.3.1 (1 μg) were exposed to a Fujifilm BAS cassette 2025 overnight (Fuji Photo Film). Phospholuminescence plates were scanned using a Fuji BAS-1800 II bioimaging analyzer at a pixel size of 50 × 50 μm. Images were analyzed with Aida Image Analyzer software (Raytest).

Frozen tumor sections of MDA-MB-231, SK-Br-3, SUM149, BT474, and MCF-7 were fixed for 10 minutes in ice-cold acetone (−20°C). Endogenous mouse Ig staining was blocked using a mouse-on-mouse blocking kit (BMK-2202, Vector). Subsequently, sections were incubated with 10 μg/mL PD-L.3.1, followed by incubation with a peroxidase-conjugated rabbit-anti-mouse antibody (P0260, DAKO). Finally, 3-3′-Diaminobenzidine (DAB) was used to visualize peroxidase activity in the sections.

Statistical analyses were performed using PASW Statistics version 18.0 and GraphPad Prism version 5.03 for Windows. Differences in uptake of radiolabeled PD-L1.3.1 were tested for significance using the nonparametric Kruskal–Wallis and Mann–Whitney U test. The correlation between tumor uptake measured by SPECT and ex vivo biodistribution was calculated with the Spearman correlation coefficient. A P value below 0.05 was considered significant.

Flow-cytometry analysis showed that the percentage of tumor cells positive for PD-L1 was 89.4%, 2.9%, 8.9%, 0.2%, and 0.1% for MDA-MB-231, SK-Br-3, SUM149, BT474, and MCF-7, respectively. PD-L1.3.1 was labeled with ¹¹¹In, obtaining specific activities up to 10 MBq/μg antibody. ¹¹¹In-PD-L1.3.1 showed the highest binding to MDA-MB-231 cells while binding to SUM149 and SK-Br-3 was significantly lower (P = 0.002). PD-L1–negative BT474 and MCF-7 cells did not show any specific binding of ¹¹¹In-PD-L1.3.1 (Fig. 1A). The number of binding sites for PD-L1.3.1 was determined quantitatively with scatchard analysis and was 47,700 ± 2,900 for MDA-MB-231, 2,000 ± 100 for SK-Br-3, and 3,600 ± 400 for SUM149. On the basis of these results, MDA-MB-231 cells were used in subsequent binding assays.

The immunoreactive fraction of ¹¹¹In-PD-L1.3.1 was 82% and the IC₅₀ of unlabeled PD-L1.3.1 was 0.15 nmol/L (Fig. 1B). Scatchard analysis showed that the affinity of ¹¹¹In-labeled PD-L1.3.1 was 0.97 ± 0.15 nmol/L (Fig. 1C). PD-L1.3.1 was slowly internalized by MDA-MB-231 cells. After 24 hours of incubation, 25% of the cell-associated activity was internalized and 75% was still membrane-bound (Fig. 1D). These data demonstrate that ¹¹¹In-PD-L1.3.1 antibody specifically binds to PD-L1–expressing tumor cells.

The antibody protein dose-escalation study showed high and specific tumor accumulation of ¹¹¹In-PD-L1.3 (Fig. 2A). Tumor uptake in PD-L1–positive MDA-MB-231 was the highest in mice injected with 0.3 or 1 μg of antibody (37.5 ± 12.5 and 35.7 ± 5.8%ID/g, respectively, Fig. 2A). At antibody doses of ≥3 μg, tumor uptake significantly decreased (3 μg: 16.9 ± 3.8%ID/g, P = 0.002). Tumor uptake was the lowest in mice injected with 300 μg PD-L1.3.1 (7.6 ± 1.6%ID/g). Other organs did not show specific targeting of ¹¹¹In-PD-L1.3.1.

Uptake of ¹¹¹In-PD-L1.3.1 by PD-L1–positive MDA-MB-231 xenografts was observed as early as 1 day after injection and further increased at days 3 and 7 (Table 1, Fig. 2B). PD-L1–negative MCF-7 xenografts did not show specific uptake at any time point. Tumor-to-blood ratios increased over time for MDA-MB-231 and were the highest 7 days post injection (3.9 ± 1.0). Tumor-to-blood ratios for MCF-7 did not exceed 0.9 and were not significantly increased compared with the tumor–blood ratios in mice that received an excess of unlabeled PD-L1.3.1. These data demonstrate that ¹¹¹In-PD-L1.3.1 can discriminate between PD-L1–positive and PD-L1–negative xenografts.

MDA-MB-231 xenografts were clearly visualized with ¹¹¹In-PD-L1.3.1 SPECT/CT, with increasing contrast between the tumor and normal tissue with time. Typical examples of SPECT/CT scans, and the quantification of tumor and liver uptake are presented in Fig. 3. Tumor uptake in MDA-MB-231 xenografts increased over time, whereas the uptake in MCF-7 xenografts did not exceed uptake in normal organs, such as the liver. Intratumoral distribution of ¹¹¹In-PD-L1.3.1 was heterogeneous, as is visualized in a zoomed high-resolution image of the tumor in Fig. 3B. There was a strong correlation between the tumor uptake as measured by SPECT and by counting of dissected tissues at 7 days postinjection (Spearman r = 0.94).

SPECT/CT images demonstrated high uptake of ¹¹¹In-PD-L1.3.1 in MDA-MB-231 and SK-Br-3 xenografts and low uptake in SUM149, BT474, and MCF-7 xenografts (Fig. 4), which was confirmed in the ex vivo biodistribution study. Tumor uptake at 3 days after injection was 25.2 ± 2.9%ID/g, 22.0 ± 5.1%ID/g, 8.4 ± 0.2%ID/g, 10.0 ± 0.7%ID/g, and 8.1 ± 1.4%ID/g, for MDA-MB-231, SK-Br-3, SUM149, BT474, and MCF-7 xenografts, respectively.

Autoradiographic analysis of the tumor sections showed that ¹¹¹In-PD-L1.3.1 antibody was distributed heterogeneously within the tumor. In general, highest tumor uptake was observed in the periphery of the tumor, whereas the uptake in the tumor center was lower. This heterogeneous distribution was also observed on cross-sections of the SPECT/CT scan (Fig. 5). After autoradiographic analysis, the same slides were used for hematoxylin and eosin (H&E) staining, which showed that the tumors contained areas with vital tissue (mostly in the periphery of the tumor) and areas with necrosis (mostly in the center of the tumor). In MDA-MB-231 and SK-Br-3 xenografts, highest uptake of ¹¹¹In-PD-L1.3.1 was found in the vital part of the tumor.

Immunohistochemical analysis of tumor sections for PD-L1 expression showed that MDA-MB-231 tumors expressed the highest levels of PD-L1. SK-Br-3 xenografts also clearly expressed PD-L1, although the expression levels varied largely between the tumors. Immunostaining of SUM149, BT474, and MCF-7 showed low PD-L1 expression (Fig. 5).

In summary, ¹¹¹In-PD-L1.3.1 SPECT/CT can discriminate between xenografts with high and low PD-L1 expression levels.

Patients with PD-L1–expressing tumors have a poor prognosis. By upregulating PD-L1, tumor cells can escape immune recognition and attack. Anti–PD-1/PD-L1 antibodies have shown impressive antitumor effects. However, currently, there is no useful biomarker to predict response to anti–PD-1/PD-L1 targeted therapy, although there are first indications that tumor PD-L1 expression seems to be required for response to treatment. We have shown for the first time that it is feasible to noninvasively detect tumor PD-L1 expression. SPECT/CT imaging with the monoclonal antibody PD-L1.3.1 showed specific and efficient accumulation in PD-L1–expressing xenografts, whereas no specific uptake was measured in PD-L1–negative tumors.

Molecular imaging with radiolabeled anti–PD-L1 antibodies has important advantages over immunohistochemical analysis of PD-L1 expression. First of all, imaging allows measurement of PD-L1 expression of whole tumor lesions and their metastases, thereby avoiding sampling errors and thus misinterpretation due to intratumoral and interlesional heterogeneity. Second, it allows monitoring of expression during the course of disease, without the need of repetitive biopsies. This is of clinical importance because PD-L1 expression can change during the course of disease, due to disease progression and/or the effects of treatment (15, 18, 19). Third, in vivo imaging techniques also take into account the in vivo target accessibility for agents that are administered systemically. Several factors may determine whether antibodies will reach the tumor cells, such as blood vessel density, vascular permeability, and interstitial fluid pressure (24, 25). Therefore, in vivo imaging with radiolabeled anti–PD-L1 antibodies may provide more relevant information about PD-L1 expression and accessibility than IHC, it may allow patient selection for PD-1/PD-L1–targeted therapy, and can be used to monitor PD-L1 expression during the course of disease.

Our in vitro studies showed that ¹¹¹In-PD-L1.3.1 has very good characteristics for in vivo imaging. Importantly, the immunoreactivity was retained after radiolabeling, and its affinity for PD-L1 was high (K_d = 1.0 nmol/L). Moreover, upon binding of PD-L1.3.1 to the tumor cell, PD-L1.3.1 was slowly internalized. This is a favorable characteristic for imaging, because ¹¹¹In-DTPA is trapped in the tumor cell after internalization and degradation of the antibody (26). This will result in enhanced tumor-to-background contrast in SPECT/CT imaging. In vivo studies showed that optimal contrast between tumor and normal tissue was obtained at antibody protein doses of ≤1 μg PD-L1.3.1. At higher antibody doses, the PD-L1–binding sites on the tumor cells became saturated, resulting in decreased contrast in SPECT/CT imaging. Because ¹¹¹In-PD-L1.3.1 is an intact IgG molecule, the clearance from the circulation was relatively slow. Therefore, the optimal time point for imaging was 3 to 7 days after injection.

To investigate the specificity of ¹¹¹In-PD-L1.3.1 imaging, we compared the uptake of ¹¹¹In-PD-L1.3.1 in five breast cancer xenografts with different PD-L1 expression levels. In vitro analysis showed that MDA-MB-231 expressed high levels of PD-L1, SK-Br-3, and SUM149 showed moderate to low PD-L1 expression levels and BT474 and MCF-7 did not express PD-L1. In vivo, MDA-MB-231 xenografts showed high uptake of ¹¹¹In-PD-L1.3.1, whereas uptake in BT474 and MCF-7 was significantly lower. The low uptake of ¹¹¹In-PD-L1.3.1 observed in PD-L1–negative tumors was not PD-L1 mediated, because blocking of PD-L1 with an excess of unlabeled PD-L1.3.1 did not further reduce ¹¹¹In-PD-L1-3.1 uptake. The nonspecific uptake is most likely due to the enhanced permeability and retention effect. Macromolecules larger than 40 kDa selectively leak out from tumor vessels and accumulate in tumor tissue and not in normal tissue. This is caused by the abnormal architecture of tumor vasculature, lack of effective lymphatic drainage, and the production of vascular mediators such as VEGF, which enhance the extravasation and accumulation of macromolecules into tumor tissue (27). The levels of nonspecific uptake observed with PD-L1.3.1 in MCF-7 and MDA-MB-231 xenografs are in the same range as those observed with other antibodies and in other antigen-negative xenograft models (28–30).

The uptake of ¹¹¹In-PD-L1.3.1 in the two models with moderate to low PD-L1 expression in vitro (SK-Br-3 and SUM149) differed significantly from each other in vivo. SK-Br-3 xenografts showed high uptake of ¹¹¹In-PD-L1.3.1, whereas the uptake in SUM149 was comparable with that in the PD-L1–negative models. The low uptake in SUM149 xenografts may be due to the fact that this is an inflammatory breast cancer xenograft. Inflammatory breast cancers are highly vascular tumors that produce large amounts of VEGF (31–33). It has been shown that VEGF expression negatively correlates with PD-L1 expression, which may explain the low PD-L1 expression level of SUM149 in vivo (16).

The present preclinical studies show that ¹¹¹In-PD-L1.3.1 is a promising tracer to noninvasively determine PD-L1 expression and accessibility in cancer lesions. PD-L1.3.1 is specifically directed against human PD-L1, and does not cross react with murine PD-L1. Therefore, in animal models, ¹¹¹In-PD-L1.3.1 targets PD-L1 expressed on human tumor cells, while it has no affinity for murine PD-L1 expressed on murine cells, such as dendritic cells, macrophages, and natural killer cells (5). In addition, PD-L1 mRNA was found in non-lymphoid tissue such as the heart, skeletal muscles, placenta, and lungs. However, in these tissues, PD-L1 protein expression was not detectable with IHC, indicating that these tissues have very low-expression levels (15, 34, 35). Future in vivo studies will have to demonstrate whether tumor targeting of ¹¹¹In-PD-L1.3.1 is affected by the presence of PD-L1–expressing human cells. Also, to translate noninvasive PD-L1 imaging into the clinic, anti–PD-L1 antibodies should be preferably radiolabeled with the positron emitter zirconium-89 (⁸⁹Zr) for PET/CT imaging. For preclinical purposes, microSPECT/CT has a similar or even better resolution compared with microPET/CT. However, in the clinical setting, the resolution, sensitivity, and quantification of PET/CT are superior to that of SPECT/CT.

Immunohistochemical analysis of PD-L1 expression seemed to correlate with tumor uptake of ¹¹¹In-PD-L1.3.1, although there were some discrepancies. MDA-MB-231 and SK-Br xenografts both showed clear PD-L1 expression as determined immunohistochemically and high tumor uptake of ¹¹¹In-PD-L1.3.1. However, in the PD-L1–negative models (BT474 and MCF-7) that did not show specific uptake of ¹¹¹In-PD-L1.3.1, we also observed some immunostaining for PD-L1. This illustrates the difficulty to adequately score PD-L1 expression in immunohistochemically stained sections. At the moment, there are no strict criteria to define PD-L1 positivity (e.g., cytoplasmic vs. membranous staining, intensity of the staining, percentage of positive cells) and different immunohistochemical staining procedures have been reported in literature (7, 12–14).

Next to PD-L1 expression, other biomarkers have been studied for their potential to predict response to PD-1/PD-L1–targeted therapy. Several immunohistochemical features, including PD-1, PD-L1, and PD-L2 expression, patterns of immune cell filtration, and lymphocytes subpopulations, have been analyzed for their association with clinical outcome after treatment with anti–PD-1/PD-L1 antibodies. These studies showed that tumor cell PD-L1 expression significantly correlated with objective response, whereas PD-L2 expression did not (12–14, 36). Also, PD-1 expression and the presence of tumor-infiltrating lymphocytes did not correlate to response to anti–PD-1 antibodies (12). Thus, so far, PD-L1 expression seems to be the most powerful predictor of response to both PD-1– and PD-L1–targeted therapy.

This study shows the feasibility to noninvasively determine PD-L1 expression and accessibility of tumor lesions using radiolabeled anti–PD-L1 antibodies and SPECT/CT imaging. These encouraging results warrant future research to explore the feasibility of this approach in patients. Ultimately, this imaging technique may serve as biomarker to predict clinical response to PD-1/PD-L1–targeted therapy.

No potential conflicts of interest were disclosed.

Conception and design: S. Heskamp, W. Hobo, H. Dolstra, O.C. Boerman

Development of methodology: S. Heskamp, J.D.M. Molkenboer-Kuenen, D. Olive, O.C. Boerman

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Heskamp, W. Hobo, J.D.M. Molkenboer-Kuenen, H. Dolstra

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Heskamp, J.D.M. Molkenboer-Kuenen, W.J.G. Oyen

Writing, review, and/or revision of the manuscript: S. Heskamp, W. Hobo, J.D.M. Molkenboer-Kuenen, W.J.G. Oyen, H. Dolstra, O.C. Boerman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Heskamp, J.D.M. Molkenboer-Kuenen

Study supervision: S. Heskamp, D. Olive, O.C. Boerman

The authors thank Bianca Lemmers-van de Weem, Henk Arnts, Iris Lamers-Elemans, and Kitty Lemmens-Hermans for technical assistance with the animal experiments.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.