The human melanoma cell line A375M was purchased from the American Type Culture Collection. Cell lines were confirmed to be negative for microbial contamination and were authenticated on August 6, 2018, by BaseClear using short tandem repeat profiling. A375M cells were routinely cultured in Roswell Park Memorial Institute 1640 medium (Invitrogen) containing 10% fetal calf serum (Bodinco BV), under humidified conditions at 37°C with 5% CO₂. Cells were passaged 1:10, twice a week. For in vivo experiments, cells in the exponential growth phase were used.

First, the buffer of pembrolizumab (25 mg/mL, Merck) was exchanged for NaCl 0.9% (Braun) using a Vivaspin-2 concentrator (30 kDa) with a polyethersulfon filter (Sartorius). Next, pembrolizumab was conjugated with the tetrafluorphenol-N-succinyldesferal-Fe(III) ester (TFP-N-sucDf; ABX) as described earlier, in a 1:2 TFP-N-sucDf:mAb ratio.30 Conjugated product was purified from unbound chelator using Vivaspin-2 concentrators and stored at −80 °C. On the day of tracer injection, N-sucDf-pembrolizumab was radiolabeled with ⁸⁹Zr, delivered as ⁸⁹Zr-oxalate dissolved in oxalic acid (PerkinElmer), as described previously.30 For in vivo studies, pembrolizumab was radiolabeled at a specific activity of 250 MBq/mg. IgG₄ control molecule (Sigma-Aldrich) was conjugated with TFP-N-sucDf at a 1:3 molar ratio, followed by radiolabeling with ⁸⁹Zr at similar specific activity of 250 MBq/mg.

Size exclusion high-performance liquid chromatography (SE-HPLC) was used to determine the final number of TFP-N-sucDf ligands per antibody (chelation ratio). SE-HPLC analysis was also performed to assess potential aggregation and fragmentation for both N-sucDf-pembrolizumab and ⁸⁹Zr-pembrolizumab. An HPLC system (Waters) equipped with an isocratic pump (Waters), a dual wavelength absorbance detector (Waters), in-line radioactivity detector (Berthold) and a TSK-GEL G3000SWXL column (Tosoh Biosciences) was used with phosphate buffered saline (PBS, sodium chloride 140.0 mmol/L, sodium hydrogen phosphate 0.9 mmol/L, sodium dihydrogen phosphate 1.3 mmol/L; pH 7.4) as mobile phase (flow 0.7 mL/min). Radiochemical purity of ⁸⁹Zr-pembrolizumab was measured by trichloroacetic acid precipitation assay.31 Immunoreactivity of ⁸⁹Zr-pembrolizumab was analyzed by a competition binding assay with unlabeled pembrolizumab. Nunc-immuno break apart 96-wells plates (Thermo Scientific) were coated overnight at 4°C with 100 µL of 1 µg/mL PD-1 extracellular domain (R&D Systems) in PBS, set to pH 9.6 with Na₂CO₃ 2M. Plates were washed with 0.1% Tween 80 in PBS and blocked for 1 hour at room temperature (RT) with 150 µL 1% human serum albumin (Albuman, Sanquin) in PBS. Multiple 1:1 mixtures of ⁸⁹Zr-pembrolizumab with unlabeled pembrolizumab were prepared, using a fixed concentration of ⁸⁹Zr-pembrolizumab (7000 ng/mL) and varying concentrations of unlabeled pembrolizumab (from 3.75 ng/mL to 12.5×10⁶ ng/mL). Of each mixture, 100 µL was added to the 96-wells plate and incubated for 2 hours at RT. After washing twice with washing buffer, radioactivity in each well was counted using a gamma counter (Wizard² 2480–0019, SW 2.1, PerkinElmer). Counts were plotted against the concentration of competing unlabeled pembrolizumab. The half maximal inhibitory concentration (IC₅₀) was calculated using GraphPad Prism 7 (GraphPad software). Immunoreactivity was expressed as the IC₅₀ value divided by the ⁸⁹Zr-pembrolizumab concentration to calculate the immune reactive fraction (IRF).

All animal studies were approved by the Institutional Animal Care and Use Committee of the University of Groningen. Studies were performed in humanized NOG mice (NOD.Cg-Prkdc^scid Il2rg^tm1Sug/JicTac, Taconic) and non-humanized NOG mice (Taconic) were used for control experiments. HuNOG mice are sublethally irradiated 3 weeks after birth and subsequently reconstituted with human CD34⁺ hematopoietic stem cells derived from fetal cord blood to express a functional human immune system including B-cells, T-cells, NK-cells, dendritic cells and monocytes. HuNOG and NOG mice were subcutaneously xenografted with 5×10⁶ A375M human melanoma cells in 300 µL of a 1:1 mixture of PBS and Matrigel (BD Biosciences) on the right flank. Tumor growth was assessed by caliper measurements. When tumor volumes reached 100 to 200 mm³ (after 2 weeks), 2.5 MBq ⁸⁹Zr-pembrolizumab (10 µg) was administered via retro-orbital injection. Mice were anesthetized using isoflurane/medical air inhalation (5% induction, 2.5% maintenance).

The first group of huNOG mice received 10 µg ⁸⁹Zr-pembrolizumab (n=5). In addition, a second group of huNOG mice xenografted with the same tumor model received a co-injection of 10 µg ⁸⁹Zr-pembrolizumab and 90 µg unlabeled pembrolizumab (n=4). To a third group of huNOG mice, 2.5 MBq ⁸⁹Zr-IgG₄ control (10 µg) was administered (n=4). Control NOG mice received 10 µg ⁸⁹Zr-pembrolizumab (n=4).

On day 7 post tracer injection (pi), PET scanning was performed. We selected this day based on optimal tumor-to-blood ratio and technical aspects, including feasible tracer specific activity and animal welfare. Mice were placed in a Focus 220 rodent scanner (CTI Siemens) on heating matrasses. Acquisition time was 60 min. A transmission scan of 515 s was performed using a ⁵⁷Co point source to correct for tissue attenuation. After scanning, mice were sacrificed for ex vivo biodistribution. Bone marrow was collected from the femur bone by centrifugal-based separation. All other organs were dissected and counted in a gamma-counter (Wizard² 2480–0019, SW 2.1, PerkinElmer). Tracer uptake in each organ was expressed as percentage of the injected dose per gram tissue weight, calculated by the following formula: %ID/g = (activity in tissue (MBq)/total injected activity (MBq))/tissue weight (g)×100. To compare ex vivo and in vivo uptake, ex vivo uptake was also calculated as mean radioactivity per gram tissue, adjusted for total body weight (SUV_{mean ex vivo} ), calculated with the following formula: SUV_{mean ex vivo} = (activity in tissue (MBq)/total injected activity (MBq))×mouse weight (g). Calculations are corrected for decay and background.

PET data was reconstructed and in vivo quantification was performed using PMOD software (V.4.0, PMOD technologies LCC). Three-dimensional regions of interest were drawn around the tumor. For other organs and tissues, a size-fixed sphere was drawn in representative tissue parts. PET data was presented as mean standardized uptake value (SUV_{mean in vivo} ), calculated by the following formula: SUV_mean (g/mL) = (activity concentration (Bq/mL)/applied dose (Bq))×weight (kg)×1000.

Tumor and spleen from ex vivo biodistribution studies were formalin-fixed and paraffin embedded (FFPE). FFPE tissue blocks where cut into slices of 4 µM. These slices were exposed to a phosphor imaging screen (PerkinElmer) for 72 hours and then scanned using a Cyclone phosphor imager (PerkinElmer).

Subsequent slices of the same tumor, spleen and mesenteric lymph node tissue were stained for H&E, CD3, CD8 and PD-1. FFPE tumor, spleen and lymph node tissue were cut into 4 µm slices using a microtome (Microm Hm 355 s, Thermo Scientific) and mounted on glass slides. Tissue sections were deparaffinized and rehydrated using xylene and ethanol. Heat-induced antigen retrieval was performed in citrate buffer (pH=6) at 100°C for 15 min. Endogenous peroxidase was blocked by 30 min incubation with 0.3% H₂O₂ in PBS. For CD3 staining, slides were incubated with rabbit anti-human CD3-antibody (Spring bioscience; clone SP162) in a 1:100 dilution in PBS/1% bovine serum albumin (BSA) at RT for 15 min. For CD8 staining, slides were incubated with rabbit anti-human CD8-antibody (Abcam; clone SP16) in a 1:50 dilution in PBS/1% BSA at 4°C overnight. For PD-1 staining, slides were incubated with rabbit anti-human PD-1-antibody (Abcam, clone EPR4877(2)) in a 1:500 dilution in PBS/1% BSA at RT for 30 min. Human tonsil or lymph nodes tissues sections served ad positive control and were incubated with either CD3, CD8 or PD-1 antibody. As a negative control human tonsil or lymph nodes sections were incubated with rabbit IgG monoclonal antibody (Abcam, clone EPR25A) or PBS/1% BSA.

For CD3, CD8 and PD-1 staining, incubation with secondary antibody (anti-rabbit EnVision⁺, Dako) was performed for 30 min, followed by application of diaminobenzidine chromogen for 10 min. Hematoxylin counterstaining was applied and tissue sections were dehydrated using ethanol and imbedded using mounting medium (Eukitt). H&E staining served to analyze tissue viability and morphology. Digital scans were acquired by a Nanozoomer 2.0-HT multi slide scanner (Hamamatsu).

To enable clinical application, GMP-compliant ⁸⁹Zr-pembrolizumab was developed. First, N-sucDf-pembrolizumab intermediate product was produced on a larger scale (60 mg batch, divided in 2.5 mg aliquots) and subsequently radiolabeled with ⁸⁹Zr, followed by purification, dilution and sterile filtration (online supplemental figure S1). Release specifications were defined, as shown in online supplemental table S1. All analytical methods for quality control (QC) were validated. According to protocol validation of both N-sucDf-pembrolizumab and ⁸⁹Zr-pembrolizumab, manufacturing consisted of three independent validation runs, including complete release QC. Stability of N-sucDf-pembrolizumab stored at −80 °C was studied up to 6 months and stability of ⁸⁹Zr-pembrolizumab was determined up to 168 hours at 2°C to 8°C stored in a sterile, type 1 glass injection vial. In addition, in use stability was demonstrated at RT in a polypropylene syringe for up to 4 hours (online supplemental table S2).

Data are presented as median±IQR. A Mann-Whitney U test, followed by a Bonferroni correction was performed to compare groups (GraphPad, Prism 7). P values ≤0.05 were considered significant. If not indicated otherwise, results were not statistically significant.

We optimized the conjugation processes of pembrolizumab with the TFP-N-sucDf chelator and its subsequent radiolabeling with ⁸⁹Zr. For in vivo studies, N-sucDf-pembrolizumab was produced with >60% yield and average 1.7 chelators per antibody (online supplemental figure S2, table S1). N-sucDf-pembrolizumab was subsequently radiolabeled with ⁸⁹Zr at a specific activity of 250 MBq/mg, with radiochemical purity of >95% after purification. Both N-sucDf-pembrolizumab and ⁸⁹Zr-pembrolizumab were stable, as shown in online supplemental table S1, S2 and figure S2. Immunoreactivity was not impaired by conjugation or radiolabeling.

PET imaging revealed ⁸⁹Zr-pembrolizumab uptake in tumor, but also in healthy tissues, including liver, spleen and lymph nodes, of A375M tumor-bearing huNOG mice (figure 1A,B). Consistent with these results, ex vivo biodistribution at day 7 pi showed highest ⁸⁹Zr-pembrolizumab uptake in spleen (SUV_mean 30.5, IQR 15.8 to 67.7), mesenteric lymph nodes (SUV_mean 20.4, IQR 8.0 to 25.2), bone marrow (SUV_mean 14.5, IQR 6.1 to 32.8), thymus (SUV_mean 1.3, IQR 1.1 to 2.1), liver (SUV_mean, IQR 6.0, IQR 3.4 to 9.9) and tumor (SUV_mean 5.1, IQR 3.3 to 8.9) (figure 1C, online supplemental table S3).

Tumor uptake of ⁸⁹Zr-pembrolizumab was variable and slightly higher than tumor uptake observed for ⁸⁹Zr-IgG₄ control, however not significant due to small groups of mice (SUV_mean 5.1, IQR 3.3 to 8.9 vs SUV_mean 3.5, IQR 2.7 to 4.4) (figure 1C). This may be explained by low PD-1 expression found in all tumors by immunohistochemical (IHC) analysis (figure 2). ⁸⁹Zr-pembrolizumab tumor-to-blood ratio also did not differ from ⁸⁹Zr-IgG₄ control (figure 1D).

⁸⁹Zr-pembrolizumab in huNOG mice showed higher uptake in lymphoid tissues compared with ⁸⁹Zr-IgG₄ control: spleen (SUV_mean 13.9, IQR 7.1 to 21.4, NS, p=0.254), mesenteric lymph nodes (SUV_mean 2.3, IQR 1.4 to 4.4, NS, p=0.114), salivary gland (SUV_mean 2.1, IQR 1.2 to 2.9, NS, p=0.635), bone marrow (SUV_mean 8.8, IQR 7.6 to 10.0, NS, p=1.714) and thymus (SUV_mean 0.5, IQR 0.4 to 1.1, p=0.1714), indicating that ⁸⁹Zr-pembrolizumab uptake in these tissues is, at least partly, PD-1-mediated. ⁸⁹Zr-pembrolizumab tissue-to-blood (T:B) and tissue-to-muscle (T:M) ratios in lymphoid organs confirmed high uptake in these tissues (figure 1D,E). Additionally, relatively high ⁸⁹Zr-IgG₄ uptake was found in spleen, bone marrow and liver compared with other organs, suggesting ⁸⁹Zr-pembrolizumab uptake in these tissues is also due to Fcγ receptor (FcγR)-binding of the antibody’s Fc-tail. High ⁸⁹Zr-IgG₄ uptake was less evident in lymph nodes and thymus.

⁸⁹Zr-pembrolizumab spleen uptake in huNOG mice was blocked by the addition of a 10-fold excess unlabeled pembrolizumab (SUV_mean 30.5, IQR 15.8 to 67.7 versus SUV_mean 5.1, IQR 4.3 to 7.0, p=0.032) (figure 1B,C). Uptake in other lymphoid organs and liver was also reduced by addition of unlabeled mAb dose, whereas uptake in non-lymphoid tissues was unaffected (online supplemental table S3). Tracer activity in blood pool was increased by addition of unlabeled mAb (SUV_mean 0.1, IQR 0.0 to 1.8 to SUV_mean 2.2, IQR 1.4 to 7.4), but uptake in tumor did not change.

Autoradiography confirmed PET imaging results on a macroscopic level, showing high uptake in spleens of huNOG mice compared with spleens of mice that received an additional unlabeled pembrolizumab dose (figure 3). Furthermore, comparable tumor uptake was found for different dose groups. IHC analysis on spleen and lymph node tissue of huNOG mice revealed that PD-1, CD3 and CD8 positive cells were present. CD3 and CD8 cells were also present in tumor tissue of huNOG mice (figure 2), however, PD-1 staining of these tumors was negative.

⁸⁹Zr-pembrolizumab biodistribution in NOG control mice clearly showed a different pattern than in huNOG mice, with high uptake in liver (SUV_mean 16.9, IQR 5.1 to 26.2) and spleen (SUV_mean 49.6, IQR 16.6 to 135.6), whereas ⁸⁹Zr-pembrolizumab tumor uptake in NOG mice was similar to huNOG mice (SUV_mean 9.3, IQR 4.5 to 15.7 vs SUV_mean 5.1, IQR 3.3 to 8.9) (online supplemental figure S3). High ⁸⁹Zr-pembrolizumab spleen uptake in this model may be unexpected, since limited T-cells are present in NOG mice (online supplemental figure S3). However, high spleen uptake in severely immunocompromised mice has been described previously and is potentially Fcγ receptor-mediated.23 24 32 Moreover, spleen weights in NOG mice were lower than in huNOG mice (NOG: 0.017 g±0.015 g; huNOG: 0.037 g±0.016 g, p=0.036), which resulted in higher tracer uptake expressed as %ID per gram spleen tissue for NOG mice. A low spleen weight may result from high radiosensitivity of NOG splenocytes, which can lead to toxicity.33

The production processes for N-sucDf-pembrolizumab intermediate product and ⁸⁹Zr-pembrolizumab for in vivo studies were modified to comply with GMP requirements. In the conjugation reaction, pH is increased from 4.5 to 8.5, performed in small titration steps, as described earlier by Verel et al.30 During this pH transition, precipitation occurred at 6.5 to 7.0, which was re-dissolved at pH >7.5. No precipitation was observed when pH was changed abruptly, for example, by buffer exchange, to pH 8.5 during conjugation and to pH 4.5 for removal of Fe(III). This indicates potential instability of pembrolizumab at pH 6.5 to 7.0. Formation of aggregates may be explained by the fact that pembrolizumab is an IgG₄ type mAb, which forms non-classical disulfide bonds. In contrast, IgG₁ type antibodies can only form classical disulfide bonds. There are many other determinants of antibody stability besides disulfide bond formation, however, this phenomenon was not seen previously with the radiolabeling of IgG₁ type antibodies.31 33 34

Immunoreactivity was not affected when pembrolizumab showed precipitation during pH transition, demonstrated by comparable IRF for precipitated N-sucDf-pembrolizumab and for non-precipitated N-sucDf-pembrolizumab (online supplemental figure S4). However, it is unknown whether the pembrolizumab structure is modified by the formation of precipitates. Therefore, the method for pH transition by buffer exchange was incorporated in the conjugation protocol for pembrolizumab. Production of clinical grade ⁸⁹Zr-pembrolizumab was performed as previously described by Verel et al. 30

Three consecutive batches of conjugated and radiolabeled pembrolizumab were produced at clinical scale and complied with all release specifications (online supplemental tables S1 and S2), indicating that our process for manufacturing clinical grade ⁸⁹Zr-pembrolizumab is consistent and robust. ⁸⁹Zr-pembrolizumab was obtained with a specific activity of 37 MBq/mg and mean IRF of 1.35±0.6 (n=3). Stability studies revealed that N-sucDf-pembrolizumab remained compliant to release specifications up to 6 months storage at −80°C, therefore N-sucDf-pembrolizumab shelf-life was set at 6 months. Stability studies are ongoing and shelf-life may be extended if future time points remain within specifications. Data obtained during process development and validation were used to compile the investigational medicinal product dossier (IMPD), which includes all information regarding quality control, production and validation of ⁸⁹Zr-pembrolizumab. Based on this IMPD, ⁸⁹Zr-pembrolizumab has been approved by competent authorities for use in clinical studies.