Immuno-PET Molecular Imaging of RANKL in Cancer

Simple Summary Receptor activator of the nuclear factor kappa B ligand (RANKL) has been recently identified as a target of interest in the tumor microenvironment (TME), specifically in facilitating an immunosuppressive environment and subsequent resistance to immunotherapy. However, assessment of RANKL in the TME remains difficult due to its heterogeneous nature and suboptimal sampling methods. In our study we developed an anti-RANKL immuno-PET tracer to non-invasively monitor RANKL expression in the TME and help to understand the RANK/RANKL pathway. Abstract Purpose: The involvement of RANK/RANKL signaling in the tumor microenvironment (TME) in driving response or resistance to immunotherapy has only very recently been recognized. Current quantification methods of RANKL expression suffer from issues such as sensitivity, variability, and uncertainty on the spatial heterogeneity within the TME, resulting in conflicting reports on its reliability and limited use in clinical practice. Non-invasive molecular imaging using immuno-PET is a promising approach combining superior targeting specificity of monoclonal antibodies (mAb) and spatial, temporal and functional information of PET. Here, we evaluated radiolabeled anti-RANKL mAbs as a non-invasive biomarker of RANKL expression in the TME. Experimental design: Anti-human RANKL mAbs (AMG161 and AMG162) were radiolabeled with 89Zr using the bifunctional chelator DFO in high yield, purity and with intact binding affinity. After assessing the biodistribution in healthy CD-1 nude mice, [89Zr]Zr-DFO-AMG162 was selected for further evaluation in ME-180 (RANKL-transduced), UM-SCC-22B (RANKL-positive) and HCT-116 (RANKL-negative) human cancer xenografts to assess the feasibility of in vivo immuno-PET imaging of RANKL. Results: [89Zr]Zr-DFO-AMG162 was selected as the most promising tracer for further validation based on biodistribution experiments. We demonstrated specific accumulation of [89Zr]Zr-DFO-AMG162 in RANKL transduced ME-180 xenografts. In UM-SCC-22B xenograft models expressing physiological RANKL levels, [89Zr]Zr-DFO-AMG162 imaging detected significantly higher signal compared to control [89Zr]Zr-DFO-IgG2 and to RANKL negative HCT-116 xenografts. There was good visual agreement with tumor autoradiography and immunohistochemistry on adjacent slides, confirming these findings. Conclusions: [89Zr]Zr-DFO-AMG162 can detect heterogeneous RANKL expression in the TME of human cancer xenografts, supporting further translation of RANKL immuno-PET to evaluate tumor RANKL distribution in patients.


Introduction
The receptor activator of nuclear factor kappa B ligand (RANKL) and its receptor RANK are members of the tumor necrosis factor (TNF) and TNF receptor (TNFR) superfamily. The binding of RANKL to RANK results in the trimerization of the receptor and recruitment of TNF receptor-associated factors (TRAF), adaptor proteins, and activation of downstream signaling pathways [1]. The RANKL/RANK system was initially discovered during research into TNFR homologs expressed on T cells and dendritic cells [2]. Subsequently, the RANK/RANKL interaction's pivotal role in osteoclastogenesis and bone homeostasis was elucidated [3]. RANKL binds osteoprotegerin (OPG) and the leucinerich repeat-containing G-protein coupled receptor 4 (LGR4), resulting in an inhibition of downstream signaling and acting as a negative feedback mechanism preventing excess activation. These findings have led to the development of RANKL-targeting therapies, of which the fully human monoclonal antibody denosumab has been in clinical use for almost a decade. Indeed, denosumab has proven benefits for patients with osteoporosis, cancer-related bone disease, and other skeletal conditions [4][5][6].
More recently, RANK/RANKL signaling was also found to be an essential component in carcinogenesis, specifically in the maintenance of self-renewal of cancer stem cells and up-regulation of anti-apoptotic pathways, making the RANK/RANKL axis an attractive therapeutic target [7]. Moreover, inhibiting RANKL has been shown to improve the effectiveness of immune-checkpoint inhibitors (ICI) targeting CTLA-4 or PD(L)-1 in preclinical models of cancer [8,9]. While immunotherapy has revolutionized contemporary oncology, it is typically only beneficial in a small subset of patients. This limitation of ICIs has fueled interest in an improved understanding of the intricate interplay of cancer cells, immune cells and cytokines in the tumor microenvironment (TME) [10]. To this end, current methods to assess RANKL expression typically include serum assays or ex vivo immunohistochemistry (IHC) of tissue biopsies. However, these techniques are hampered by issues with sensitivity, variability, and spatial heterogeneity within the TME and negatively impact patient comfort when tissue sampling is required, if tissue sampling is even remotely possible [11,12]. Immuno-positron emission tomography (PET) can provide solutions to the problems above by combining the superior sensitivity of PET with the benefits of high targeting specificity of monoclonal antibodies, which can provide information on whole-body biomarker distribution or tumor target expression and act as a companion diagnostic in vivo in a non-invasive and longitudinal manner.
We aimed to develop a non-invasive imaging biomarker using PET to study the expression and spatial heterogeneity of tumor expressed RANKL in the TME. In particular, 89 Zr-anti-RANKL radioimmunoconjugates were synthesized in a reproducible way, characterized for antigen affinity and optimal biodistribution, and validated as markers of RANKL expression and heterogeneity in the TME in human cancer xenograft models.
The loading efficiency of the DFO chelates in the final bioconjugates was determined by ESI mass spectrometer (Centre for Proteomics, University of Antwerp) using a Q-TOF2 instrument (Waters, Milford, MA, USA), as previously described [13].
For 89 Zr labeling, 185 MBq of 89 Zr-oxalate (BV cyclotron VU, Perkin Elmer, Waltham, MA, USA) was diluted to 0.2 mL with 1 M oxalic acid (Sigma Aldrich, St. Louis, MO, USA), neutralized with 2 M Na 2 CO 3 (Sigma Aldrich, St. Louis, MO, USA) and added to chelex-treated HEPES buffer (0.5 M, pH 7.2, Sigma Aldrich, St. Louis, MO, USA). This was followed by the addition of 1 mg of DFO-AMG162 or DFO-IgG2, or 0.75 mg of DFO-AMG161, and volume adjustment to 2 mL with HEPES buffer. Radiolabeling was performed for 1 h at 37 • C, after which the radiolabeled antibodies were purified using a PD-10 desalting column and concentrated using Amicon centrifugal filter units (cutoff 50 kDa). The radiochemical yield and purity were evaluated by size exclusion chromatographyhigh performance liquid chromatography (SEC-HPLC, Cytiva, Marlborough, MA, USA, Superdex 200 increase 5/150, phosphate buffer 0.05 M pH 6.7, λ = 280 nm) and instant thinlayer chromatography (iTLC, Elysia, Angleur, Belgium glass microfiber chromatography paper impregnated with a silica gel (SG), 20 mM citric acid/acetonitrile (9:1, (v/v)). iTLC strips were cut in half, and the bottom and top parts were counted for radioactivity in an automatic gamma-counter (Wizard 2480, PerkinElmer, Waltham, MA, USA). Radiolabeled antibody remained at the origin (Rf [ 89 Zr]Zr-DFO-Antibody = 0), while free 89 Zr and 89 Zr-DFO moved with the solvent front (Rf = 1). The concentration of the radiolabeled antibody in the final PBS formulation was calculated using a spectrophotometer (Genesys 10S UV-VIS, λ = 280 nm), which was used to calculate the specific activity.
The radiotracers' stability was evaluated over seven days in vitro by incubation in the final formulation (PBS) at room temperature and in human plasma and mouse plasma at 37 • C. Samples were spotted on TLC strips, and subsequent iTLC analysis was performed as described above.
The immunoreactivity of DFO-AMG161 and DFO-AMG162 was determined by using a non-cell based binding assay, as previously described [14].

Mice
Immunodeficient CD-1 nude female mice (Charles River Laboratories, Wilmington, MA, USA, RRID:IMSR_CRL:086), age 5-7 weeks, weight 20-25 g, were kept under environmentally controlled conditions (12 h light/dark cycle, 20-24 • C and 40-70% relative humidity) in individually ventilated cages with food and water ad libitum. Experimental procedures and protocols were performed following European Directive 86/609/EEC Welfare and Treatment of Animals and were approved by the local ethical committee (2018-48, University of Antwerp, Belgium). Sample size was estimated by power analysis (α = 0.05, power 0.8) and animals were assigned to experimental groups using simple randomization.
The transduction efficiency was estimated by assessing RANKL expression using flow cytometry on a BD FACSAria II (BD Biosciences, San Jose, CA, USA) cytometer with an APC-conjugated antibody (#347508, Biolegend, San Diego, CA, USA). Positivity was determined using the overton algorithm.
Animals were eligible for the experimental design in the imaging study when tumor volume reached 100 mm 3 . Tumor bearing animal in which tumor growth was halted or did not exceed 100 mm 3 were excluded from the study.
For quantitative analysis, volumes of interest were manually drawn on the PET images using PMOD (PMOD, v 3.6; PMOD Technologies, RRID:SCR_016547) to delineate the tumors. For an absolute measure of tracer uptake, normalized images were scaled according to the percent injected dose (% ID/mL = tissue uptake [kBq/mL]/injected dose [kBq] × 100%). After the last PET/CT imaging acquisition, mice were sacrificed via cervical dislocation, and ex vivo biodistribution was performed as described earlier.
Moreover, adjacent frozen tumor sections (10 µm) were taken at regular intervals across the entire tumor volume and used for histologic analysis of RANKL expression. Quantification of RANKL staining was performed by calculating the percentage of DABstained (brown) area across two non-sequential whole-tumor sections using the immunohistochemistry (IHC) profiler plugin for ImageJ v1.53 (RRID:SCR_003070), as previously described [15]. Three to five tumors were evaluated per tumor type. The mean percentage of positive stained area per tumor was used to calculate differences between groups. RANKL levels were correlated to the corresponding ex vivo radiotracer uptake in the tumor.

Statistical Analysis
The data are presented as mean % ± one standard deviation (SD). Graphs, twotailed unpaired t-tests, Pearson correlation and linear regression analysis were performed with GraphPad Prism version 6.01 (RRID:SCR_002798). p value < 0.05 was considered as statistically significant.

Biodistribution of Both Radiotracers in Healthy CD-1 Nude Mice
Differences in the biodistribution between the IgG1 and IgG2 isotype of the radiotracer were evaluated in vivo during 7 days in CD-1 nude mice. [ 89 Zr]Zr-DFO-AMG162 showed an expected and slow antibody clearance from the blood, while undesirable non-specific radiotracer uptake in various organs was absent ( Figure 1A, Table S1). In contrast, [ 89 Zr]Zr-DFO-AMG161 demonstrated immediate high and variable sequestration in the liver and spleen in the majority of animals, clearing the antibody from circulation and undesirably reducing the exposure time of the radiotracer to a potential target ( Figure 1B, Table S2). In light of these suboptimal findings, [ 89 Zr]Zr-DFO-AMG161 was not further explored in subsequent xenograft experiments. On the other hand [ 89 Zr]Zr-DFO-AMG162 showed favorable biodistribution and good characteristics as potential radiotracer.

Biodistribution of Both Radiotracers in Healthy CD-1 Nude Mice
Differences in the biodistribution between the IgG1 and IgG2 isotype of the radiotracer were evaluated in vivo during 7 days in CD-1 nude mice. [ 89 Zr]Zr-DFO-AMG162 showed an expected and slow antibody clearance from the blood, while undesirable nonspecific radiotracer uptake in various organs was absent ( Figure 1A, Table S1). In contrast, [ 89 Zr]Zr-DFO-AMG161 demonstrated immediate high and variable sequestration in the liver and spleen in the majority of animals, clearing the antibody from circulation and undesirably reducing the exposure time of the radiotracer to a potential target ( Figure 1B, Table S2). In light of these suboptimal findings, [ 89 Zr]Zr-DFO-AMG161 was not further explored in subsequent xenograft experiments. On the other hand [ 89 Zr]Zr-DFO-AMG162 showed favorable biodistribution and good characteristics as potential radiotracer.
Tumor-to-organ ratios for different xenografts are shown in Table 2, showing the best ratios for [ 89 Zr]Zr-DFO-AMG162 in RANKL positive UM-SCC-22B xenografts (Tables S5-S7).  Figure 3D). Taken together these data show specific binding of [ 89 Zr]Zr-DFO-AMG162 to RANKL in human head and neck squamous cancer xenografts UM-SCC-22B, compared to isotype control and a RANKL-negative xenograft.

Validation of the Radiotracer Uptake in Tumor Xenografts
Autoradiography (ARG) and IHC were performed in each group of animals, and the patterns of radiotracer distribution seen on ARG were compared with IHC staining on adjacent tumor sections. ARG of ME-180-RANKL tumor sections with [ 89 Zr]Zr-DFO-AMG162 showed hot spots that could be matched entirely with IHC RANKL stainings (mean 24.0 ± 6.9% positively stained tumor area) in adjacent slides ( Figure 4A). In the blocking experiments, fewer hot areas could be observed on ARG of ME-180-RANKL tumors, as expected ( Figure 4B). However, these regions still demonstrated overlap with RANKL IHC stain (mean 23.8 ± 4.4% positively stained tumor area) ( Figure S4).

Discussion
The introduction of a RANKL-targeting treatment using the monoclonal antibody denosumab (AMG162) has improved clinical outcomes in patients with various skeletal conditions, including osteoporosis, metastatic bone disease, multiple myeloma, and giant cell tumor of bone [16]. Moreover, data from the Cancer Genome Atlas (TCGA) suggests that RANKL gene expression is associated with patient outcome in multiple cancer types in an explorative analysis of survival ( Figure 5) [17,18]. This illustrates the potential of anti-RANKL therapies and novel methods for RANKL quantification, including immuno-PET. However, both in clinical practice and in the setting of clinical trials, questions remain regarding patient selection, optimal duration, long-term safety, maintenance dose, and sequencing of therapies that include a RANKL inhibitor [22,23]. These issues remain largely unaddressed, in part because of a lack of reliable non-invasive biomarkers for RANKL. Moreover, the observation that the benefit of the combination of RANKL and ICI may be sequence-dependent supports the concept of a biomarker that allows sequential assessment without the need for invasive procedures [24]. We initiated the search for an imaging biomarker of RANKL in the TME by radiolabeling AMG161 (IgG1) and AMG162 (IgG2). The difference in isotype showed no impact on in vitro characterization, with both yielding good radiolabeling, stability and unchanged affinity. However, a significant dif- More recently, the potential for repurposing denosumab as a modulator of the immune response in improving the efficacy of ICI in cancer treatment is actively being explored [19,20]. A phase II study in breast cancer patients supports this approach by showing an increase in lymphocytes and CD8+ T-cells in tumors exposed to single-agent denosumab (D-BEYOND; ClinicalTrials.gov Identifier: NCT01864798) [21]. Of note are two ongoing randomized trials, one being the CHARLI trial (ClinicalTrials.gov Identifier: NCT03161756) that is a phase Ib/II study including patients with unresectable stage III/IV melanoma treated with nivolumab in combination with four doses of denosumab, with or without ipilimumab (primary end-point: progression-free survival). The POPCORN trial (ACTRN12618001121257) is a phase Ib/II translational study including patients with stage IA to IIIA non-small-cell lung cancer receiving neoadjuvant treatment with two doses of nivolumab with or without denosumab (following nivolumab), followed by surgery and as primary end-point translational research into the tumor-immune correlates of combination therapy. Similar translational efforts are ongoing in phase I/II studies in cervical cancer (DICER; ISS20177041) and melanoma (ClinicalTrials.gov Identifier: NCT03620019) [1].
However, both in clinical practice and in the setting of clinical trials, questions remain regarding patient selection, optimal duration, long-term safety, maintenance dose, and sequencing of therapies that include a RANKL inhibitor [22,23]. These issues remain largely unaddressed, in part because of a lack of reliable non-invasive biomarkers for RANKL. Moreover, the observation that the benefit of the combination of RANKL and ICI may be sequence-dependent supports the concept of a biomarker that allows sequential assessment without the need for invasive procedures [24]. We initiated the search for an imaging biomarker of RANKL in the TME by radiolabeling AMG161 (IgG1) and AMG162 (IgG2). The difference in isotype showed no impact on in vitro characterization, with both yielding good radiolabeling, stability and unchanged affinity. However, a significant difference could be observed in vivo between both isotypes: [ 89 Zr]Zr-DFO-AMG161 expressed different levels of radiotracer sequestration in spleen and liver, a phenomenon which could be related to mouse Fc receptor binding. Generally, IgG1 antibodies are more immunoreactive than their IgG2 counterparts and more prone to bind Fc receptors, an effect that is even more pronounced in immunodeficient mice models [25]. Even though the translational significance of this sequestration of the radiotracer for the biodistribution in humans is uncertain, it is an undesirable characteristic that can result in lower target uptake and more rapid metabolization of the tracer. Radiotracers revealed clear (5-10% ID/g) bone uptake, a phenomenon related to instability of the DFO complex, remarkably in patients the unwanted bone uptake is hardly an issue [26]. New chelators with improved characteristics have been developed (DFO*/HOPO) but were not yet commercially available at the start of this study [27]. Currently, AMG162 is an FDA-approved biopharmaceutical used in clinical care, whereas AMG161 is not, and this may limit its potential for successful clinical translation. For these reasons, [ 89 Zr]Zr-DFO-AMG162 was selected for further experiments.
The selective uptake of [ 89 Zr]Zr-DFO-AMG162 was demonstrated in RANKL-transduced ME-180 xenografts, with clear visualization on PET and high radiotracer uptake in the xenografts. Subsequent ARG and IHC showed corresponding regions with radiotracer uptake and staining, respectively. Moreover, a control blocking experiment reduced the radiotracer uptake demonstrating the specific uptake of the radiotracer.
Subsequent experiments were performed on patient-derived cell lines, UM-SCC-22B (RANKL positive) and HCT-116 (negative control). The RANKL positive cell line UM-SCC-22B was of interest since both membranes bound and soluble RANKL expression was reported. However, prior to the start of the study RNA RANKL expression was not assessed in different cell lines to select the highest RANKL expression cell line and is a limitation of the study.
[ 89 Zr]Zr-DFO-AMG162 showed specific uptake of on immuno-PET in UM-SCC-22B xenografts and not of the radiolabeled isotype control [ 89 Zr]Zr-DFO-IgG2. While we could visualize physiologically relevant amounts of tumor derived RANKL in the TME with [ 89 Zr]Zr-DFO-AMG162, the RANKL expression of UM-SCC-22B was substantially lower compared to the transduced ME-180-RANKL cell line. In contrast, [ 89 Zr]Zr-DFO-AMG162 uptake was significantly lower in HCT-116 xenografts (a non-RANKL expressing model) compared to UM-SCC-22B xenografts (a RANKL expressing cell line), supporting the specificity of tracer uptake. While the overall differences in mean SUV of [ 89 Zr]Zr-DFO-AMG162 uptake between UM-SCC-22B and HCT-116 xenografts or over isotype control were low, remarkable differences in regional uptake were evident, confirming the considerable heterogeneity of RANKL in the TME, which may be missed by other sampling techniques (IHC). Autoradiography and immunohistochemistry were used to explore the contribution of RANKL heterogeneity on overall uptake. This showed convincing spatial congruency of [ 89 Zr]Zr-DFO-AMG162 and RANKL expression in UM-SCC-22B xenografts. In contrast, autoradiography only demonstrated limited overlap in areas with high vasculature or stroma when using [ 89 Zr]Zr-DFO-IgG2 (in UM-SCC-22B xenografts) or [ 89 Zr]Zr-DFO-AMG162 (in HCT-116 xenografts), suggesting non-specific uptake. Indeed, the large size of antibodies (~150 kDa) represents an intrinsic limitation for efficient tumor diffusion, and due to their high avidity, they remain close to the periphery of the vasculature [28]. In addition, the long-blood half-lives of several days to weeks allows antibody-based radiotracers to achieve high uptake-values in targeted tissues, but simultaneously leads to elevated overall non-specific accumulation [29].Taken together, these data illustrate the potential benefits of immuno-PET for in vivo RANKL assessment compared to other sampling techniques, such as biopsies that are prone to error due to sampling bias or serum assays that are only able to provide a global biomarker quantification. In addition, immuno-PET has the unique capability of non-invasively visualizing actual drug-delivery to the TME.
Importantly, it is noted that while murine RANKL shares 83% sequence homology with human RANKL, the anti-RANKL antibodies used in our experiments have no affinity for murine RANKL [30]. For the purpose of our study, this difference is not of importance, but evidence does implicate the host-derived RANKL in the TME [31]. Therefore, further preclinical work using [ 89 Zr]Zr-DFO-AMG162 to elucidate the tumor and host interactions in the TME will require more translationally appropriate clinical models. For example, the use of transgenic models and humanized mice models may better reflect the human TME and recapitulate the complex interactions of RANKL between the tumor and surrounding cells [32,33]. Finally, a more human-like TME may impact the levels of tumor-derived RANKL expression, resulting in improved imaging characteristics.

Conclusions
In conclusion, we describe for the first time in vivo assessment of RANKL expression in the TME using immuno-PET imaging. Our results suggest that RANKL imaging offers advantages over more traditional approaches for longitudinal tumor characterization and merits further investigation. [ 89 Zr]Zr-DFO-AMG162 showed favorable stability, high binding affinity, and specific tumor uptake with very good visual agreement to the spatial distribution of RANKL in the TME as assessed with histology, supporting further translation to evaluate tumor RANKL distribution in patients.