Development and evaluation of Hsp90-targeting nanobodies for visualisation of extracellular Hsp90 in tumours using PET imaging

Abstract

Background

The extracellular localisation of the Heat shock protein 90 (Hsp90) is associated with the diseased state and wound healing and presents a promising opportunity for cancer targeting using Positron Emission Tomography (PET) imaging and molecularly targeted radiotherapy. The aim of this work is to develop a radiotracer with low nanomolar binding affinity to target the extracellular and particularly membrane pool of Hsp90, evaluate it in vitro, and conduct preliminary PET studies in vivo in mouse tumour models. Variable Heavy domain of Heavy chain antibodies, often referred to as Nanobodies, are suitable targeting vectors for the extracellular targets due to their favourable pharmacokinetic properties and low nanomolar target affinities. The main objective of the study is to target tumours expressing extracellular and membrane Hsp90 phenotype with minimal tracer accumulation in the non-target organs, which limited the translation of previously studied small molecule cytosolic Hsp90 tracers suffering from high non-Hsp90 specific background in the abdominal area.

Results

Six nanobodies were obtained after llama immunization with recombinant Hsp90α and ELISA biopanning, produced in E. coli and screened for stability and affinity. We selected one nanobody, 4DAM26, with good thermal stability, no aggregation at elevated temperatures, and low nanomolar affinity towards Hsp90α and Hsp90β isoforms for translation as a PET radiotracer. The nanobody was bioconjugated to p-NCS-NODAGA and radiolabeled with gallium-68 with 75 ± 11% radiochemical yield and > 99% radiochemical purity and remained stable up to 3 h in phosphate buffered saline and mouse serum. Pilot in vivo evaluation using µPET/CT and ex vivo biodistribution demonstrated a favourable pharmacokinetic profile, but the tumour uptake was non-distinguishable from the background tissue.

Conclusion

Compared to the small molecule Hsp90 tracers, the studied Nb-based tracer has improved pharmacokinetics properties including renal clearance and almost no accumulation in the non-target organs. Tumour uptake, on the other hand, was minimal and could not be differentiated from the background in µPET/CT. Our experiments indicate that in the studied models, membrane and extracellular expression of Hsp90 is majorly an artifact of cellular death, as only dead/dying cells had accessible pools of Hsp90 by flow cytometry, a consequence of a leaky membrane. More fundamental research is required to reassess the role of extracellular Hsp90 in cancer, and our future efforts will be focused on improving our inventory of cytosolic Hsp90 tracers with proven Hsp90-specific tumour accumulation.

Supplementary Information

The online version contains supplementary material available at 10.1186/s41181-025-00331-0.

Background

Hsp90 is an abundant molecular chaperone that is essential for maintaining proteostasis by assisting in protein folding, disaggregation, and degradation of misfolded proteins (Chen and Jeffery 2019). In humans, there are four Hsp90 paralogues expressed in three different cellular compartments: two in the cytosol, of which Hsp90α is inducible upon stress, and Hsp90β is constitutively expressed but can be upregulated via internal ribosome entry site (IRES) in certain cases (Bhattacharya et al. 2022); glucose-regulated protein 94 (GRP94) in the endoplasmic reticulum; and tumour necrosis factor receptor associated protein 1 (TRAP1) in the mitochondria. A small proportion of the cytosolic Hsp90 also localizes to the nucleus (Berbers et al. 1988). Cytosolic Hsp90 constitutes 1–2% of total cellular protein and has a growing list of unique clients, with multiple hundreds of documented proteins identified to date (Echeverría et al. 2011; Finka and Goloubinoff 2013).

The dependence of a vast number of proteins on Hsp90 implicates it in a variety of diseases, where it may either exert cytoprotective functions, such as in inflammation and wound healing, or exacerbate the disease severity, such as in viral diseases, neurodegeneration, and cancer. Elevated levels of Hsp90 have been documented across multiple cancer types, such as in melanoma and dermal carcinoma cell lines and tissues which express higher Hsp90 levels relatively to their non-cancerous counterparts (Becker et al. 2004a; Sahu et al. 2012), in breast cancer (Pick et al. 2007), leukemia (Flandrin et al. 2008), and colon cancer (Kubota et al. 2010). In cancer cells, Hsp90 stabilizes several proteins associated with tumourigenic processes, including cyclin-dependent kinase 4 (CDK4), human epidermal growth factor receptor 2 (HER2), mutant tumour suppressor p53, and matrix metalloproteinase-2 (MMP2), all of which contribute to the hallmarks of cancer (Hanahan and Weinberg 2011; Whitesell and Lindquist 2005). Together, the pro-tumourigenic roles of Hsp90 sparked interest in exploring Hsp90 as a pan-cancer target for therapy and diagnostics. Most attempts to target Hsp90 have focused on selectively targeting cytosolic paralogues of Hsp90. However, cell permeable molecules explored so far inhibited Hsp90 function in both healthy and cancer cells, leading to hepatic, neurological, and ocular toxicity, significantly impeding the clinical translation of Hsp90 inhibitors (Li and Luo 2022). Despite decades of research and numerous clinical trials, only one Hsp90 inhibitor has been approved for clinical use so far, namely Pimitespib (Jeselhy®, Taiho Pharmaceutical) for the treatment of gastrointestinal stromal tumours in Japan (Kurokawa et al. 2022).

Over the past decades, multiple studies have emerged showing that Hsp90 can be found in the extracellular space of cells exposed to the environmental stress, where it may associate with the membrane (Becker et al. 2004b; Hance et al. 2012), be found in the exosomes (Eguchi et al. 2018; Ono et al. 2018), or enter the circulation, collectively referred to as extracellular Hsp90 (eHsp90). The presence of eHsp90 has been reported in various cancer types in vitro and in vivo, including prostate cancer (Hance et al. 2012), breast cancer (Kaneko et al. 2020), and melanoma (Becker et al. 2004b), where it is implicated in metastasis and invasion through the activation of MMP2 and interaction with the low density lipoprotein receptor-related protein 1 (LRP1) (Gopal et al. 2011). Elevated levels of eHsp90 were also found in the serum of cancer patients which correlated with poor prognosis (Yuan et al. 2022). Notably, Hsp90α isoform, and not Hsp90β, is responsible for the tumourigenic functions in the extracellular space, even though both isoforms can be detected outside of the cell (Zou et al. 2017). Targeting the extracellular pool of Hsp90 can mitigate the cytotoxic effects of pan-Hsp90 inhibitors and provide a selective route to cancer cells without affecting surrounding healthy cells, where intracellular Hsp90 plays an essential role, as opposed to eHsp90, which mainly comes into play during wound healing and tumourigenesis. Strategies to target eHsp90 have been investigated preclinically for therapy using cell membrane-impermeable small molecule inhibitor DMAG-N-oxide, monoclonal antibody 4C5, and a series of fluorescent and near-infrared probes such as HS-27 and HS-201 (Barrott et al. 2013a; Kaneko et al. 2020; Stellas et al. 2007; Tsutsumi et al. 2008). These compounds selectively target cancer cells and mitigate cancer cell growth, migration, and invasion without exhibiting cytotoxic effects due to the selectivity for eHsp90. Interestingly, two compounds, namely HS-131 and HSP90 inhibitor − SN38 conjugates were shown to selectively internalize in tumour cells via eHsp90-dependent endocytosis, although the mechanistic details of eHsp90 internalization are not fully elucidated (Crowe et al. 2017; Zhu et al. 2020). The ability to visualize tumours in vivo for applications such as fluorescence-guided surgery and photodynamic therapy was also demonstrated (Barrott et al. 2013b; Osada et al. 2022). Combined, this provides evidence that eHsp90 is a not only a suitable therapeutic target in cancer that could help overcome challenges associated with targeting intracellular Hsp90, but also an interesting imaging biomarker.

Attempts to visualize Hsp90 in cancer using PET and Single-photon emission computed tomography (SPECT) mostly focused on the cytosolic Hsp90 pool. We and others have previously reported that Hsp90-overexpressing tumours can be targeted by small-molecule PET radiotracers. Such tracers include [¹¹C]NMS-E973, [¹¹C]HSP990, and ¹⁸F-labeled Dimer-Sansalvamide A cyclodecapeptide, all of which accumulate at the tumour site in the multiple xenograft mouse models with moderate to good tumour uptake (Cools et al. 2023a; Vermeulen et al. 2019; Wang et al. 2022). However, further optimization was needed due to the lack of contrast in the abdominal area caused by the high uptake in the liver, bile, and gastrointestinal tract as a result of hepatobiliary clearance. This limits their diagnostic and potential therapeutic applications since these off-target areas could be areas of interest in metastatic disease. Current study explores targeting of the extracellular and specifically membrane-bound pool of Hsp90 in cancer using non-invasive PET imaging. The aim of this approach is to limit uptake by healthy tissues, as well as to reduce non-tumour-associated background due to the hepatobiliary clearance. We believe that this strategy would result in improved image contrast compared to the imaging of cytosolic Hsp90. To achieve this goal, we chose nanobodies as targeting vectors for eHsp90, given their favourable pharmacokinetic profile. Relatively to the antibodies, their small molecular weight of 15–20 kDa allows them to pass glomerular filtration and be excreted to urine via the kidneys. Renal clearance has been observed for the majority of the nanobody-based PET tracers both in the in vivo mouse models, as well as in patients (Keyaerts et al. 2016; Vaidyanathan et al. 2016; Xavier et al. 2016). One of the potential limitations of using nanobodies targeting approach is that their clearance can be too rapid to allow for sufficient tumour accumulation and high kidney uptake, which can be overcome via conjugation with the albumin binders to prolong blood circulation (Song et al. 2023). Such a tracer could become a useful tool in diagnosing multiple cancer types and aid in patient selection for Hsp90-targeting therapies. Additionally, if sufficiently selective tumour uptake is achieved, translation to therapeutic applications of the probes using labeling with alpha- and beta-emitters may be envisaged.

Methods

Immunization & initial selection

Two llamas were subcutaneously injected with (N-terminal) His₆-tagged recombinant human Hsp90α (in house, active) and separate VHH libraries were constructed from each animal’s lymphocytes to screen for the presence of antigen-specific Nbs. The VHH libraries were expressed in E. coli TG1 cells and panned on the solid-phase coated Hsp90α plates. The enzyme-linked immunosorbent assay (ELISA) screening of the colonies selected during panning produced 69 clones that scored positive for Hsp90α, out of which 6 unique full-length Nbs with distinct complementarity-determining region 3 (CDR3) groups were selected for further studies. The immunizations, generation of VHH libraries, panning and ELISA screenings were performed by the VIB Nanobody Core. Immunizations and handling of the llamas were performed according to directive 2010/63/EU of the European parliament for the protection of animals used for scientific purposes and approved by the Ethical Committee for Animal Experiments of Lamasté (permit No. NSF 2021–1).

Recombinant protein expression

Plasmid encoding (N-terminal) His₆-tagged Hsp90α was a kind gift from Dr. Blagg’s group. Nbs were encoded with a (N-terminal) PelB sequence and a (C-terminal) FLAG₃HIS₆-tag for affinity purification, which were synthesized and cloned into pET-24a( +) plasmid by GenScript. All proteins were expressed in chemically competent BL21 E. coli cells.

Hsp90α production: 1 L of lysogeny broth (LB) supplemented with Kanamycin was inoculated with 10 mL of the overnight culture transformed with plasmid encoding Hsp90α and grown at 37 °C until the optical density at 600 nm (OD600) reached ~ 0.6. The culture was cooled on ice for 2 min before induction with 1 mM Isopropyl β-D-1-thiogalactopyranoside (IPTG) and grown overnight at 16 °C. The pellet was collected by centrifugation at 5,000 revolutions per minute (RPM) or 6,238 × g (JLA 8.100 rotor) for 15 min at 4 °C and lysed by sonication in the ice-cold lysis buffer (50 mM Tris–HCl, 500 mM NaCl, 20 mM imidazole, protease inhibitors, pH 8), followed by incubation with DNAse for 20 min at 4 °C. The lysate was cleared by centrifugation at 18,000 RPM or 39,191 × g (JA-25.5 rotor) for 30 min at 4 °C, protein was filtered through 0.22 µM polyvinylidene fluoride (PVDF) syringe filter, purified using affinity chromatography (HisTrap FF 5 mL, Cytiva) and eluted with 20–500 mM imidazole, followed by size exclusion chromatography (Superdex S200 16.600, Cytiva) in phosphate-buffered saline or PBS (500 mM NaCl, pH 7.5) using an AKTA purification system (Cytiva). Purity was assessed by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) and Coomassie blue staining.

Nbs production: cultures were grown at 37 °C until the OD600 reached ~ 1, cooled to 28 °C for 1 h, and induced with 1 mM IPTG overnight at 16 °C. The pellet was collected as described above. Nbs were extracted from the periplasm by stirring in 24 mL of ice-cold lysis buffer (20 mM Tris–HCl, 500 mM sucrose, protease inhibitors, pH 8.3) for 30 min. 40 mL of ice-cold water was added and stirred for an additional 30 min, after which the lysate was cleared as above. Nbs were purified by affinity chromatography (HisTrap FF 5 mL, Cytiva) and eluted with 20–250 mM imidazole, endotoxins were removed by an additional wash with 0.1% Empigen® in PBS. Purity was assessed by SDS-PAGE and Coomassie blue staining and molecular weight determined experimentally by LC–MS (LCMS-2020, Shimadzu) equipped with a Zorbax Extend-C18 column (2.1 × 100 mm, 3.5 µm, 300 A, Agilent) and a narrow-bone Zorbax 300 Extend-C18 guard column (2.1 × 12.5 mm, 5 µm, Agilent) using a standard method (buffer A—Milli-Q® water with 0.1% formic acid, buffer B—acetonitrile with 0.1% formic acid; running method—5–100% Buffer B, 6.5 min at 0.35 mL/min, oven at 80 °C). All proteins were dialysed into PBS overnight at 4 °C and stored in small aliquots at -80 °C in 5% glycerol.

Biophysical characterization and affinity determination

Melting and aggregation temperatures of the Nbs were determined on Uncle instrument (Unchained Labs) at 1 mg/mL Nbs concentration in PBS; all samples were filtered through PVDF membranes directly before the measurement. To assess thermal stability parameters, protein unfolding was monitored during thermal denaturation via intrinsic fluorescence, while aggregation was simultaneously detected using right-angle light scattering (RALS). This approach provided melting (T_m) and aggregation onset (T_agg) temperatures using the barycentric mean of the whole spectrum fluorescence and the side light scatter at 266 nm over 15–95 °C gradient, respectively. All biophysical assays were done in technical and biological triplicates.

Binding affinities were determined by Bio-Layer Interferometry (BLI) using Octet-RED96 instrument (ForteBio). Nbs were biotinylated with EZ-Link™ NHS-PEG₄-Biotin reagent (Thermo Scientific™), loaded on Octet® SA Biosensors (Sartorius) and tested against the concentration gradient of recombinant Hsp90α (in house) and Hsp90β (ab80033, Abcam) in the concentration range of 5–500 nM in 20 mM Tris buffer (0.1% bovine serum albumin or BSA, 0.02% Tween, 0.5 mM ATP). 4DAM26 and Hsp90α binding curves were measured in six replicates. For the remaining measurements, experimental conditions such as Hsp90 concentration and dissociation times were optimized to obtain the single best fit which was used to calculate the binding kinetics parameters. Fitting was performed in Octet Analysis software with a 1:1 model (global fit) and quality of the fits was assessed by examining R-squared (R²) and chi-squared (χ²) values, where R² > 0.95 and χ² < 3 indicate a good fit (Apiyo 2022).

Epitope binning assay was performed using a premix method in both directions, where each Nb was immobilized and treated with a mixture of Hsp90α and 100X molar excess of every other Nb that was pre-incubated for 1 h at 25 °C before the measurement. Cross reactivity of the Nbs was assessed by screening the immobilized Nbs with the same amount of blocking Nbs as in the epitope binning in the absence of antigen. Self-blocking study in the absence of antigen was performed to ensure complete blocking occurs at the concentrations chosen for the assay (Salim et al. 2024).

Cell culture methods

MDA-MB-231 and U87 cells were obtained from ATCC and cultured in Dulbecco’s Modified Eagle’s Medium, high glucose (Gibco™) supplemented with 10% fetal bovine serum (Gibco™), 1X MEM Non-Essential Amino Acids Solution (Gibco™) and 1 mM Sodium Pyruvate (Gibco™) in the humidified chamber at 37 °C and 5% CO₂. Cells were subcultured using TrypLE™ Select Enzyme (1X), no phenol red (Gibco™) and the passage number was kept under 20 for all experiments performed in this study. Cells were regularly tested negative for mycoplasma contamination.

Hsp90 knockdown in cells

For dsiRNA Hsp90 knockdown (KD) experiments, 100,000 cells/well were seeded in 6-well plates. Hsp90 KD experiment was performed as follows: 10 nmol dsiRNA targeting Hsp90α (dsiRNA 13.1, IDT) and Hsp90β (dsiRNA 13.2, IDT), or negative control dsiRNA (IDT) were formulated in 0.5 mL Opti-MEM™ I Reduced Serum Medium, no phenol red (Gibco™) and added to the wells, followed by addition of 5 µL Lipofectamine™ RNAiMAX Transfection Reagent (Invitrogen™) directly to the wells. The reagents were equilibrated for 20–30 min followed by the addition of 2 mL of regular cell media containing the corresponding number of cells. Cells were incubated for 48 h in the humidified chamber at 37 °C and 5% CO₂. KD was verified by assessing protein levels of Hsp90 isoforms using Western Blot (WB) and expressed as percent binding calculated by normalizing signal to GAPDH and negative control lysate: %binding = (Hsp90 KD signal / GAPDH KD signal) / (Hsp90 NC signal / GAPDH NC signal) × 100%.

General Western Blot protocol

Cells were harvested by first rinsing with ice-cold PBS followed by lysing with RIPA Lysis and Extraction Buffer (Thermo Scientific™). Cells were sonicated (0.1 s on, 0.1 s off, 50%, 30 s) on ice, and lysates were cleared by centrifugation at 14,000 × g for 20 min at 4 °C. Protein concentration in all cell lysates and conditioned media (CM) was determined with BCA Rapid Gold assay (Thermo Scientific™) and equal amounts of protein were loaded on the gel (any KD™ Mini-PROTEAN® TGX™ Precast Protein Gels, BioRad) and transferred to the nitrocellulose membrane (Trans-Blot Turbo Mini, BioRad). The membrane was blocked in 5% BSA in Tris Buffered Saline with Tween 20 (TBS-T) for 1 h at room temperature (RT) and incubated with primary antibodies (1:1000 anti-Hsp90α, 68/Hsp90 Abcam; 1:1000 anti-Hsp90β, E296 Abcam; 1:1000 anti-Hsp90 total, C45G5 Cell Signaling; 1:1000 anti-LDH, H-10 Santa Cruz sc-133123) or with nanobodies at 1 µM overnight at 4 °C. Membranes were washed with TBS-T buffer 3 × 10 min and incubated with 1:10,000 Rhodamine Anti-GAPDH and 1:2500 StarBright700 secondary antibodies (BioRad) in blocking buffer for 1 h, washed and imaged using ChemiDoc™ MP Imaging System (BioRad). Image analysis was performed in Image Lab software (BioRad).

Immunofluorescence in MDA-MB-231 and U87 cells

Extracellular staining protocol was adapted from our previous work (Vermeulen et al. 2019). 10,000 cells/well were seeded in poly-L-lysine coated 96-well plates (Phenoplate, Perkin Elmer). For Hsp90 isoforms KD experiments, both dsiRNAs (10 nmol) and Lipofectamine (0.2 µL/well) were pre-formulated in transfection media, 10 µL of each reagent was added to the wells and allowed to equilibrate for 20–30 min. Appropriate number of cells was added in the 100 µL of regular cell media on top of the dsiRNA transfection reagents and incubated for 48 h in the humidified chamber at 37 °C and 5% CO₂. After 48 h, cells were fixed with 4% paraformaldehyde in PBS for 15 min, blocked in 1% BSA for 1 h and treated with 0.5 µM Nbs for 2 h at RT, followed by 1 h in anti-FLAG mAb (1:350, VIB protein core) and 1 h in secondary AlexaFluor488 antibody (1:1000). Primary antibody staining was performed overnight at 4 °C (1:50 anti-Hsp90α, 68/Hsp90 Abcam; 1:100 anti-Hsp90β, E296 Abcam; 1:100 anti-S6, 5G10 Cell Signaling). Plates were imaged with Operetta CLS High-Content Analysis System (Perkin Elmer) and the corresponding Harmony software (Perkin Elmer). Background signal was subtracted and data normalized to the negative control and expressed as percent bound signal: %binding = (Hsp90 KD signal—background signal) / (Hsp90 NC signal—background signal) × 100%.

p-NCS-benzyl-NODAGA bioconjugation

4DAM26 was conjugated with a p-NCS-benzyl-NODAGA (Chematech) chelator at 10X molar excess overnight in 0.1 M carbonate buffer pH 9.5, purified by dialysis and analysed by SDS-PAGE and Coomassie blue staining, molecular weight was determined using standard LC–MS method described earlier. To test that NODAGA-conjugated Nb retains its affinity to Hsp90α, a BLI blocking study was performed by pre-incubating Hsp90α with excess of non-conjugated and NODAGA-conjugated 4DAM26 for 1 h before binding to the biosensors with immobilized 4DAM26. The difference in blocking ability of native and NODAGA-conjugated 4DAM26 would be indicative of the difference in the binding affinities.

⁶⁸Ga-radiolabelling and tracer stability studies

Gallium-68 was obtained from the germanium-68/gallium-68 generator (2.7 GBq, Eckert and Ziegler). Elution and purification were done on Scintomics automated synthesis unit using cassette and consumables kit for ⁶⁸Ga-labelled peptides (ABX). Gallium-68 was eluted from the generator with 1 mL 0.1 M HCl, purified with a Chromafix PS-H⁺ cation exchange cartridge (Fisher Scientific), and eluted from the cartridge with 1 mL 5 M NaCl. Pre-purification of the eluate was done to reduce the metal impurities that can influence radiolabelling. The eluate was then buffered with ~ 1 mL 2.5 M sodium acetate buffer to reach a pH of 4. An aliquot of 500 µL (± 50 MBq) of the buffered eluate was added to 5 nmol of NODAGA-4ADAM26 and incubated at 37 °C for 15 min while shaking. Purification of the labelling crude was performed with a PD-10 desalting column (Cytiva) to remove free gallium-68 and final tracer was eluted in PBS. Purified tracer was analysed by radioTLC (0.1 M Sodium Citrate buffer, pH 5) and radioHPLC (PBS, 0.75 mL/min, 40 min, Superdex 75 Increase 10/300 GL (Cytiva)). For stability studies, tracer was diluted 1:3 in either PBS or mouse serum and incubated for 3–4 h at RT followed by radioHPLC analysis.

[⁶⁸Ga]Ga-NODAGA-4DAM26 µPET/CT study in U87 and MDA-MB-231 tumour xenograft mice

All animal studies were approved by the ethical committee (P190-2019) and handled in accordance with Directive 2010/63/EU and the Belgian code of practice for the care and use of animals. Female SCID/Bg mice (CB17.Cg-Prkdc < scid > Lyst < bg-J > /Crl; Charles River laboratories, Sulzfeld, Germany) were inoculated with U87 (N = 6) and MDA-MB-231 (N = 4) cells at 10 weeks of age. Inoculation was performed as previously described with 1 × 10⁶ cells mixed 1:1 with Cultrex (R&D systems), and a volume of 100 µL was implanted subcutaneously into the right shoulder of each animal (Cools et al. 2023b). Once tumours reached visible size, the imaging study was performed.

Animals were anesthetized using 5% isoflurane in O₂ at a flow rate of 1 L/min, before being transferred to a heated pad and cannulated. They were then transferred to the imaging cell (Molecubes, Ghent, Belgium) and placed on the scanner (Molecubes β-CUBE, Molecubes, Ghent, Belgium). 2–4 MBq of [⁶⁸Ga]Ga-NODAGA-4DAM26 was injected intravenously at the beginning of a 90 min dynamic PET scan, followed by a CT scan performed for anatomical co-registration using an X-cube CT scanner (Molecubes, Ghent, Belgium) using the ‘General Purpose’ protocol with the following parameters: 50 kVp, 480 exposures, 85 ms/projection, 100 μA tube current, rotation time 60 s. To assess Hsp90-specific binding, a blocking study was performed by co-injection with a 100-fold molar excess of 4DAM26 alongside the PET tracer in the same animals on separate days. PET data were histogrammed into 24 frames (1 × 7.5 s, 3 × 15 s, 1 × 37.5 s, 3 × 1 min, 1 × 2 min, 4 × 3 min, 1 × 4 min, 7 × 5 min, 1 × 7.5 min, 2 × 10 min) and reconstructed into a 192 × 192 image matrix with 0.4 mm voxels using 30 iterations the native Maximum-Likelihood Expectation–Maximization (MLEM) algorithm with corrections for randoms, scatter, attenuation and decay. CT data were reconstructed using a regularized statistical (iterative) image reconstruction algorithm with non-negative least squares, yielding an isotropic 200 µm voxel size. CT data were scaled to Hounsfield Units (HUs) after calibration against a standard air/water phantom. Using PFUS v4.0 (PMOD Technologies GmbH, Zurich, Switzerland), fused PET-CT images were displayed, and volumes of interest were manually drawn over the tumor and kidneys, with 2 mm spheres placed over the liver. Data were normalized based on the injected dose and body weight to derive tissue time-activity curves (TACs), expressed as Standard Uptake Value (SUV). TACs were represented as mean SUV ± SD, visualized with GraphPad Prism v. 10 (GraphPad Software).

[⁶⁸Ga]Ga-NODAGA-4DAM26 ex vivo biodistribution studies in U87 and MDA-MB-231 tumour xenograft mice and ex vivo autoradiography

Tumour-bearing mice were injected with ~ 1 MBq [⁶⁸Ga]Ga-NODAGA-4DAM26 30 min prior to sacrifice by an overdose of isoflurane (5%) followed by decapitation. Post-sacrifice, different organs and tissues were dissected, and activity was counted in a gamma counter (Perkin Elmer). Counts were corrected for background, detector dead time and decay. Total radioactivity in blood, muscle and bone was calculated assuming the masses were 7%, 40% and 12% of the total body mass (Cools et al. 2023b). The activity was expressed as SUVs calculated as (radioactivity (in counts per minute, cpm) in organ / weight of organ in g) / (total cpm recovered / total body weight in g). Following the biodistribution study, tumour, muscle and kidney were frozen and sectioned using cryotome and exposed to a phosphor storage screen (super-resolution screen; Perkin Elmer) overnight. The autoradiograms were obtained by reading the screens using a Cyclone Plus system (Perkin Elmer). Autoradiography images were analysed using Optiquant software (Perkin Elmer).

Extracellular Hsp90 expression validation using Western Blot

1 × 10⁶ of U87 glioblastoma and MDA-MB-231 breast cancer cells were plated in 75 cm² flasks and grown for 48 h. Cells were rinsed with 3 × PBS (calcium, magnesium), and incubated in serum-free media for 20 h. The media was collected, centrifuged for 5 min at 300 × g to remove cell debris, and concentrated 30-fold using Amicon® Ultra Centrifugal Filters (10 kDa MWCO, Millipore). Cell pellets were collected, lysed, and 10 µg of protein was loaded on the gel. General Western Blot procedure was followed. Protein content in CM vs lysates is represented as % total and calculated as (signal CM) / (signal CM + signal lysates) × 100%.

Extracellular Hsp90 expression validation using flow cytometry

U87 and MDA-MB-231 cells were harvested and resuspended in 1% FBS in PBS at the concentration of 1 × 10⁶ cells/mL. Cells were incubated with PE-Hsp90 mAb at 1:100 dilution (AC88, Enzo) or 1:100 PE-IgG isotype control (Enzo) for 30 min at 4 °C. Cells were washed and resuspended in Annexin V binding buffer (BioLegend) prior to incubation with Annexin V for 15 min at 4 °C. DAPI was added to the cells directly before the measurement. To assess Nbs binding to the intracellular Hsp90, cells were collected and stained with fixable viability dye eBioscience 780 (1:1000) for 15 min at 4 °C, fixed in 4% PFA in PBS for 20 min followed by permeabilization with Triton-X for 10 min. Cells were incubated with Nbs at 1 µM concentration for 30 min at 4 °C, followed by PE anti-FLAG mAb (L5 BioLegend) incubation at 1:500 dilution for 30 min, as well as PE-Hsp90 and PE-IgG mAbs to control for target expression. Cells were analyzed by LSRFortessa™ X-20 with HTS (BD Bioscience) and imaged with FACSDiscover S8 with BD CellView™ Image Technology (BD Bioscience). Data were analyzed using FlowJo software and FACSChorus™ Software (BD Bioscience), respectively. The gating strategy is shown in the supplementary materials.

Results

Immunization & initial selection

A total of six nanobodies with distinct CDR3 domains were identified during phage display and ELISA panning (Fig. 1A–B). The relatively small panel size can be attributed to the low immunogenicity of the antigen, as human Hsp90α shares over 99% amino acid sequence identity with llama Hsp90α. Only one of the two immunized llamas produced an immune response, from which a library was constructed. All selected nanobodies demonstrated specific binding to the recombinant Hsp90α in ELISA, with a signal to blank ratio ranging from 4.0 to 15.4 (Table 1).

Fig. 1.

A Schematic process of nanobodies generation. B Sequence alignment of the CDR3 regions of the nanobodies panel. Created in BioRender. Kleynhans, J. (2025) https://BioRender.com/n43k428

Table 1.

ELISA biopanning results of selected nanobodies, protein yield after purification and percent purity as determined by SDS-PAGE and Coomassie staining

Nanobody	ELISA HSP90α	ELISA Blank	ELISA Ratio	Yield (mg/L culture)	%Purity	MW (theor.)	MW (LC–MS)	pI
2DAM5	1.2	0.1	9.0	38.9	96.6	16,724	16,710	6.1
3DAM59	0.9	0.1	6.2	20.4	97.0	16,665	16,651	5.3
3DAM106	6.0	0.4	15.4	31.2	95.3	16,880	16,866	5.8
3DAM144	2.3	0.2	11.5	38.3	92.1	17,571	17,557	5.8
4DAM26	0.5	0.1	4.0	61.2 ± 5.9	96.4	16,530	16,516	5.8
4DAM84	1.1	0.2	6.4	7.9 ± 10.0	96.5	17,023	17,008	5.7

Yields for 4DAM26 and 4DAM84 are reported based on two different protein productions, N = 1 for the rest of the Nbs. Theoretical molecular weight (MW) and isoelectric point (pI) were calculated with ProtParam (Expasy) (Gasteiger et al. 2005), and MW experimentally determined with LC–MS

Affinity determination

BLI was used to determine the affinity of the nanobodies to Hsp90α. Due to the significant size difference between the nanobodies (< 18 kDa) and Hsp90α (> 86 kDa monomer, or ~ 172 kDa in its native homodimer state), immobilizing the antigen in the BLI assay was impractical, as the signal amplitude upon nanobody binding was too weak (0–0.1 nm at > 500 nM Nbs concentration). Therefore, all nanobodies were biotinylated and immobilized on SA biosensors instead of Hsp90. All nanobodies exhibited binding kinetics to Hsp90α, with equilibrium dissociation constants (K_D) ranging from 2 to 122 nM. Notably, only 2DAM5 showed rapid target dissociation, while the rest of the panel displayed very slow dissociation kinetics, with k_dis values in the range of 10^–4 to 10^–5 1/s (Fig. 2, Table 2). Additionally, we tested whether these nanobodies bind to the same epitope on Hsp90α by pre-incubating the target protein with a 100-fold molar excess of each nanobody. Only 3DAM59 and 4DAM26 were found to bind unique epitopes, as evidenced by the lack of blocking effect by other Nbs, while 3DAM106, 3DAM144, and 4DAM84 blocked binding to Hsp90 (Supplementary Fig. 1). Interestingly, 3DAM59 also blocked binding of Hsp90α to 2DAM5 but not the other way round, likely due to inferior binding properties of 2DAM5. In addition, 2DAM5, 3DAM106, 3DAM144 and 4DAM84 showed some degree of cross-reactivity in the absence of antigen and incomplete self-blocking due to self-binding of these nanobodies to themselves. Binding to Hsp90β was also assessed, with only three Nbs, namely 3DAM59, 4DAM26 and 4DAM84 demonstrating binding to this isoform with similar binding kinetics to that of Hsp90α. This result is not unexpected given that the two isoforms share 86% of their amino acid sequence identity and have a very similar tertiary structure. Overall, we isolated three Hsp90α-specific Nbs that share the same epitope, and two Nbs with unique epitopes that bind to both Hsp90α and Hsp90β isoforms with similar binding kinetics.

Fig. 2.

BLI Binding curves of Nbs to A Hsp90α and B Hsp90β isoforms. Raw data are shown as black curves, fitting results used for the calculation of binding kinetics parameters are shown in red. Hsp90 protein concentration and association and dissociation time windows were optimized for each nanobody. Measured response varied from < 0.1 nm for lowest antigen concentrations to > 1 nm for highest antigen concentrations with 3DAM59 and 4DAM26

Table 2.

Summary of BLI binding kinetics (best fit or mean ± SD N = 6 for 4DAM26) and thermal stability (N = 3)

	Binding kinetics: Hsp90α				Binding kinetics: Hsp90β				Thermal stability
Nb	KD (nM)	kon (1/Ms)	kdis(1/s)	Χ2	KD (nM)	Kon (1/Ms)	kdis(1/s)	Χ2	Tm (°C)	Tagg (°C)
2DAM5	122.0 ± 1.0	2.2⋅104	1.7⋅10–3	0.4	No binding kinetics				74.1 ± 6.4	65.0 ± 0.4
3DAM59	2.1 ± 0.0	1.2⋅104	2.5⋅10–5	0.3	1.1 ± 0.0	4.0⋅104	4.2⋅10–5	1.7	73.0 ± 0.0	NA
3DAM106	3.8 ± 0.0	2.5⋅104	9.4⋅10–5	1.1	No binding kinetics				66.0 ± 0.0	55.8 ± 0.0
3DAM144	2.4 ± 0.0	9.3⋅103	2.2⋅10–5	0.8	No binding kinetics				71.3 ± 0.5	62.6 ± 0.0
4DAM26	2.9 ± 0.8	6.2⋅104	1.9⋅10–4	1.1	0.8 ± 0.0	2.7⋅105	2.1⋅10–4	3.7	66.1 ± 0.1	NA
4DAM84	13.1 ± 11.4	1.9⋅104	2.3⋅10–4	0.4	13.6 ± 15.6	5.6⋅104	4.4⋅10–4	1.4	48.3 ± 2.4	50.0 ± 0.1

R² > 0.99 for all binding curve fits except for 4DAM84 with Hsp90β (R² > 0.97). Chi-squared value Χ² < 3 indicates a good fit, NA = no aggregation

Thermal stability and aggregation

In addition to affinity, another key biophysical property of nanobodies that determines their applicability and developability is thermodynamic stability, where higher stability is better, and aggregation propensity, where aggregation is undesired. In the current panel, only 3DAM59 and 4DAM26 showed no aggregation at temperatures below 95 °C. The other nanobodies exhibited signs of aggregation starting at as low as 50 °C for 4DAM84, indicated by the presence of a side light scatter peak at 266 nm, typical of small aggregates (Fig. 3, Table 2). The onset of aggregation either precedes or coincides with the melting temperature of these proteins, suggesting that protein denaturation exposes thermodynamically unstable regions of the nanobodies, leading to aggregation. These results are consistent with BLI data where all but two Nbs showed some degree of self-binding which can be interpreted as aggregation.

Fig. 3.

Melting and aggregation temperatures of the Nbs panel. Purple line depicts a melting curve expressed as barycentric mean/nm over a temperature gradient of 15–95 °C, whereas blue line shows aggregation profile measured by side light scatter (SLS) at 266 nm (small aggregates). Aggregation is characterized by the presence of the SLS peak at 266 nm and precedes or coincides with the melting temperatures of 2DAM5, 3DAM106, 3DAM144 and 4DAM84 nanobodies

Detection of Hsp90 by Nbs in western blot

Next, we examined whether cellular Hsp90 could be detected by running cell lysates derived from MDA-MB-231 and U87 cells (Fig. 4A). Isoform specificity was assessed by including lysates of dsiRNA KD cells (Fig. 4B–C and Supplementary Fig. 2). All nanobodies except 4DAM26 detected a band at around 90 kDa, corresponding to Hsp90’s molecular weight of ~ 86 kDa, with the signal significantly reduced in Hsp90α KD lysates for all nanobodies. 4DAM26 did not show any signal in WB, suggesting that it recognizes a conformational epitope and requires a folded protein for binding, while the others likely bind to a linear epitope that is accessible when the protein is denatured (Forsström et al. 2015). In addition, 3DAM59 was the only Nb that demonstrated loss of signal upon Hsp90β KD, consistent with the fact that 2DAM5, 3DAM106 and 3DAM144 do not exhibit any binding to Hsp90β in BLI, and binding of 4DAM84 to Hsp90β observed in BLI might be not entirely specific, which is supported by its aggregation profile and incomplete self-blocking.

Fig. 4.

Evaluation of Hsp90 detection by Nbs in WB. A Detection of Hsp90 isoforms in MDA-MB-231 and U87 cell lysates upon dsiRNA Hsp90 KD of either Hsp90α, Hsp90β, or both Hsp90 isoforms. B Quantification of fluorescent signal in MDA-MB-231 and C U87 cell lysates upon dsiRNA Hsp90 KD of either Hsp90α, Hsp90β, or both Hsp90 isoforms. The statistical analysis was performed using ordinary one-way ANOVA and Dunnett’s multiple comparisons test in GraphPad Prism v. 10 (GraphPad Software). P ≤ 0.05 (*), P ≤ 0.01 (**), P ≤ 0.001 (***), P < 0.0001 (****)

In vitro evaluation of the Nbs panel

We tested the Nbs panel in two cell lines, U87 and MDA-MB-231 and found that all Nbs exhibit cell binding in fixed cells in immunofluorescence experiment under non-permeabilizing conditions, suggesting the presence of eHsp90 (Fig. 5A). The diffuse membrane staining pattern was in agreement with the previously observed results in other cancer cell lines (Vermeulen et al. 2019). Upon Hsp90 dsiRNA KD, only 3DAM59 and 4DAM26 showed significant reduction in fluorescence, with the highest reduction observed upon KD of both Hsp90 isoforms and comparable to the signal loss observed with Hsp90 mAbs (Fig. 5B–C and Supplementary Fig. 3). No significant reduction was observed for any other Nbs, suggesting that while they are isoform-specific, they might have some off-target binding in cells. This explanation is also supported by WB results, where some of the isoform-specific Nbs stained the protein ladder and exhibited weak binding to a band at the molecular weight corresponding to GAPDH.

Fig. 5.

A Evaluation of the Nbs panel in vitro using confocal fluorescent microscopy in fixed MDA-MB-231 and U87 tumour cells. B Nbs binding evaluation in MDA-MB-231 and C U87 cells upon Hsp90α, Hsp90β and Hsp90αβ dsiRNA KD expressed as %binding. Only 3DAM59 and 4DAM26 signal significantly decreased in all KD conditions in MDA-MB-231 and U87 cells, and no statistically significant change is observed for the other Nbs. The statistical analysis was performed using ordinary one-way ANOVA and Dunnett’s multiple comparisons test in GraphPad Prism v. 10 (GraphPad Software). P ≤ 0.05 (*), P ≤ 0.01 (**), P ≤ 0.001 (***), P < 0.0001 (****)

Radiolabeling & stability

4DAM26 was chosen as a lead Nb for further studies due to its high affinity, thermal stability and Hsp90-specific binding in cells. The effective p-NCS-NODAGA conjugation ratio was determined by LC–MS and is estimated 1:1 for 70% of all NODAGA-labeled 4DAM26, with the remaining 30% having the ratio of 2:1 chelators per nanobody (LC–MS data: unlabeled MW 16515 (100%), 1:1 labeled MW 17039 (81%), 2:1 labeled MW 17559 (35%)). A slight shift to higher MW is also observed in SDS-PAGE followed by Coomassie staining with > 96% purity. No effect of NODAGA bioconjugation was observed on Hsp90α binding affinity in BLI blocking studies, where both 4DAM26 and NODAGA-4DAM26 blocked Hsp90α binding to the same extent in the concentration-dependent manner, whereas no blocking was observed with the irrelevant control nanobody (Fig. 6A–B). For optimization of the radiolabelling conditions, Nb amounts of 1.25 nmol, 2.5 nmol, or 5 nmol per reaction were tested after 5 min, 10 min or 15 min incubation with ~ 50 MBq of gallium-68 at room temperature. Doubling the amount of nanobody used per reaction roughly doubled the radiochemical yield (RCY), whereas increasing the incubation time did not have a pronounced effect on the yield. The use of 10 nmol of 4DAM26 at 37 °C for 15 min did not further improve the yield compared to 5 nmol Nb. For all further experiments, 5 nmol Nb, 15 min incubation at 37 °C for was chosen as the optimal radiolabelling condition. The 75 ± 12% RCY and > 99% radiochemical purity (RCP) was consistently attained for all further studies (Fig. 6C). [⁶⁸Ga]Ga-NODAGA-4DAM26 is stable in mouse serum after 3 h at RT with 99% RCP (Fig. 6D). Tracer binding was evaluated in U87 and MDA-MB-231 cells as described previously, however > 99.9% of the activity was found in the washes. Given that Hsp90 is not a transmembrane protein, and that the mechanisms of its membrane association are currently under investigation, it is possible that Hsp90 is only weakly associated with the cells and did not withstand the washes during the experimental conditions. Therefore, we wanted to see whether in vivo we can observe tumour binding, in addition to investigation of the tracer’s pharmacokinetic profile.

Fig. 6.

[⁶⁸Ga]Ga-NODAGA-4DAM26 radiotracer development. A Effect of p-NCS-NODAGA-bioconjugation on Hsp90α affinity evaluated in a BLI blocking assay, with B inhibition of binding occurring in a concentration-dependent manner producing a sigmoidal-shaped curve typical for specific binding. C Radiochemical yield and radiochemical purity of the [⁶⁸Ga]Ga-NODAGA-4DAM26 tracer. D RadioHPLC analysis of [⁶⁸Ga]Ga-NODAGA-4DAM26: top chromatograph shows the UV absorbance peak at 280 nm expressed in milli-absorbance units (mAUs) corresponding to the non-radiolabeled precursor, followed by radioactive peaks measured in counts per second (CPS) of [⁶⁸Ga]Ga-NODAGA-4DAM26 after PD10 purification; after 3–4 h stability in PBS; and stability in mouse serum. The statistical analysis was performed using ordinary one-way ANOVA and Dunnett’s multiple comparisons test in GraphPad Prism v. 10 (GraphPad Software). P ≤ 0.05 (*), P ≤ 0.01 (**), P ≤ 0.001 (***), P < 0.0001 (****)

In vivo evaluation of [⁶⁸Ga]Ga-NODAGA-4DAM26 in U87 and MDA-MB-231 mouse tumour xenografts

[⁶⁸Ga]Ga-NODAGA-4DAM26 tracer showed favourable pharmacokinetic profile in vivo in MDA-MB-231 (Fig. 7A) and U87 tumour xenograft mice (Supplementary Fig. 4A), with little to no accumulation in the major organs except for kidneys. Renal clearance is an expected major elimination route for the low molecular weight compounds (< 70 kDa) such as Nbs that pass glomerular filtration. No tumour uptake is observed in µPET/CT, however blocking with excess of unlabelled tracer resulted in lower kidney retention and higher bladder uptake (Fig. 7B, Supplementary Fig. 4B, Supplementary Fig. 5). Ex vivo biodistribution data showed that 83.5% of the injected dose is found in the kidneys, with only 0.4% of the dose located in tumour. When expressed as SUV, the tumour uptake is 0.1–0.2, which is the same or only moderately higher than muscle SUV (Fig. 7C, Supplementary Fig. 4C). Ex vivo autoradiography showed tracer binding in kidneys, but not in muscle or U87 xenograft (Supplementary Fig. 4D), consistent with the in vivo results.

Fig. 7.

[⁶⁸Ga]-Ga-NODAGA-4DAM26 tracer evaluation in vivo and ex vivo in MDA-MB-231 tumour xenograft mice. A Prone maximum intensity projection (60 min p.i.) of MDA-MB-231 xenograft with axial slices of selected regions. B Time activity curves of tracer accumulation in tumours under baseline (n = 7) and blocking (n = 3) conditions expressed as averaged SUV (g/mL). C Ex vivo biodistribution (n = 4) in selected organs expressed as mean SUV ± SD

One reason for the very low tumour uptake could be fast plasma clearance of the tracer which prevents sufficient tumour accumulation, in which case different strategies could be explored to prolong the plasma half-life, e.g. by conjugation to an albumin-binding moiety in order to increase its blood circulation time. However, given no observable binding of our tracer in neither in vitro nor in vivo models, we first wanted to investigate additional methods to validate the presence of extracellular/membrane bound Hsp90.

Target validation

Both U87 and MDA-MB-231 cell lines have reported pools of extracellular Hsp90α, which we also detected by WB of the CM after 20 h of serum starvation (Fig. 8A). Interestingly, observable levels of both isoforms of Hsp90 were detected in the CM, despite some reports claiming that Hsp90α is the sole isoform present in the extracellular space. GAPDH and Actin bands in the CM are either absent or present at levels just above the detection limit, suggesting Hsp90 detected in CM is not associated with the intracellular pool, which could be a result of cellular contamination of the CM. Relatively low levels of lactate dehydrogenase (LDH), a commonly used marker of cell death in the extracellular space, were detected in the CM. When quantified fluorescent signal for the CM fraction is expressed as %total protein, relative levels of Hsp90α, Hsp90β and Hsp90total are consistently higher than those of LDH (Fig. 8B). In MDA-MB-231 cells, the relative levels of Hsp90α and Hsp90β are significantly higher than those of LDH when unpaired t test is used for analysis (P values 0.0375 and 0.0465, respectively). No significance was found in U87 cells (P values 0.2248 and 0.1112). Given the small sample size, no significance was found using Welch’s test in either of the two cell lines. Together, this suggests that at least a fraction of Hsp90 is actively secreted from these cell lines rather than passively released due to cell death, in which case similar ratios would be observed for Hsp90 isoforms and LDH.

Fig. 8.

Target validation. A Hsp90 levels in the whole cell lysates and in CM after 20 h of serum starvation in U87 and MDA-MB-231 cell lines, GAPDH and Actin are used as both loading controls in lysates and as controls for cell contamination in the extracellular fraction, LDH is used as a control for cell death in CM. B Ratio of WB fluorescent signal in CM / total signal in lysates and CM expressed in %. C Flow cytometry analysis of U87 and D MDA-MB-231 cells. A small Hsp90-positive population indicated with a black arrow is observed in both cell lines if data are visualized in a histogram and if dead cells are not excluded from the analysis. Hsp90-positive cells are double positive for DAPI and Annexin V and number of Hsp90-positive cells is proportional to the level of cell death. Samples with 50% dead cells were used for better visual example, however the same is true in samples with low levels of cellular death. E Morphological analysis of different MDA-MB-231 cell populations acquired with imaging modality of BD FACSDiscover S8 and live/dead and Hsp90-positive populations determined using flow cytometry modality. Differences in membrane integrity are clearly visible across populations, with dead and Hsp90 positive cells lacking defined membrane compared to live cells. Fluorescent signal in yellow is located intracellularly in cells lacking defined membrane, with no membrane staining pattern observed in any cells. The statistical analysis was performed using unpaired t test and Welch’s test in GraphPad Prism v. 10 (GraphPad Software)

Lastly, we examined both MDA-MB-231 and U87 cell lines by flow cytometry, which is one of the most commonly used techniques to evaluate membrane Hsp90 presence together with immunofluorescence. We saw a slight increase in fluorescence in cells stained with anti-Hsp90 mAb vs IgG control, and a small population with markedly higher expression of Hsp90. Unexpectedly, all Hsp90-positive cells were positive for DAPI, suggesting the membrane of these cells is damaged and can signify cell death. To further investigate whether Hsp90-positive cells are in fact dead or dying cells, we included Annexin V in addition to DAPI. We see that the vast majority of the Hsp90 mAb-bound cells are double positive for Annexin V and DAPI, meaning that these cells do not have a distinct pool of membrane Hsp90 and that compromised membrane integrity provided access to the intracellular pool of Hsp90 instead (Fig. 8C–D). The population of Hsp90-positive cells is higher when live cells are mixed 1:1 with dead cells, suggesting that Hsp90 only becomes available upon cell death due to the membrane rupture, rather than due to the membrane localization of the protein in live cells in the cell lines examined. We also looked at the cell morphology of these cells by using the imaging modality of the S8 Discover flow cytometer (Fig. 8E). Morphological analysis further supports the proposed Hsp90 mAb uptake mechanism, namely membrane rupture of the dead/dying cells providing access to the abundant cytosolic Hsp90. It can be clearly seen that Hsp90-positive cells lack a defined cell membrane which is observed in the cells double negative for DAPI and Annexin V. Even a small subset of live Hsp90-positive cells identified by gating out Annexin V/DAPI positive cells exhibits the signs of membrane damage.

Next, we decided to use flow cytometry to validate our immunofluorescent staining protocol for extracellular Hsp90 (Supplementary Fig. 6 and Supplementary Fig. 7). When we test for Hsp90 expression in fixed and permeabilized cells by flow cytometry, we see that all cells are positive for Hsp90, as expected given its essential function in all human cells. All Nbs also bound to fixed and permeabilized cells with a signal being significantly higher than with irrelevant Nb (p < 0.001). In line with the results observed for membrane Hsp90 target validation, none of the Nbs were bound to live cells, underlining the importance of using additional cell death markers in flow cytometry analysis and not solely relying on the cell morphology (i.e. forward and side scatter). In fixed but non-permeabilized cells, two populations appear with roughly half of all cells positive for both Hsp90 and Nbs by flow cytometry. This finding indicates that even very mild fixation conditions without detergent lead to membrane permeabilization of at least half the cells in flow cytometry and the majority in immunofluorescent imaging, resulting in the unwanted cytosolic staining which was mistaken for eHsp90.

Altogether, we were unable to detect membrane Hsp90 using previously reported techniques, and the lack of tumour uptake in PET images serves as powerful evidence of the lack of target expression in the chosen cell models.

Discussion

In this study, we confirm that the radiolabeled nanobodies possess the expected favourable pharmacokinetic profile compared to the previously reported small molecule based Hsp90 tracers, underscoring their potential as powerful probes for cancer cell imaging. However, a better understanding of the mechanisms underlying Hsp90’s extracellular and membrane localization is essential for further pursuing extracellular Hsp90 as a potential theranostic target in cancer using radiopharmaceuticals.

Our findings suggest that Hsp90 is actively secreted into the extracellular space, given that the relative ratios of secreted to cytosolic Hsp90αβ isoforms are higher than that of LDH. The leakage due to cell death, however, cannot be completely ruled out given some LDH was clearly still present in the CM. Unfortunately, we could not validate the presence of Hsp90 on the outer cell membrane surface. What we observed by flow cytometry is that Hsp90 becomes available primarily when there is cell membrane damage, such as in the case of cell death. Previous studies reporting membrane Hsp90 presence in MDA-MB-231 cells by flow cytometry did not explicitly mention the exclusion of dead cells from the analysis using cell death markers such as Annexin and PI (Devarakonda et al. 2015; Osada et al. 2017). This oversight raises the possibility that observed membrane staining may reflect the presence of dead cells rather than membrane-localized Hsp90 in viable cells. In addition, one study showed that Hsp90 becomes surface-accessible in flow cytometry experiment upon treatment with cytotoxic compounds doxorubicin and CM-EE (Quan et al. 2020). Similarly, the exclusion of dead cells from the analyses was not explicitly stated. This leaves it open to interpretation whether cytotoxic compounds that induce cell death cause Hsp90 to localize to the cell membrane, or whether a loss of membrane integrity resulting from cell death allows the antibodies access to the cytosolic pool of Hsp90 that is otherwise inaccessible in live cells. Morphological analysis showing the lack of defined membrane of cells positive for Hsp90 by flow cytometry further confirms this hypothesis.

In addition, the MDA-MB-231 cell line that is most often used for studying membrane Hsp90 is also frequently used as a model for cell blebbing—a phenomenon associated with both cell death and cancer cell migration. Given the exact mechanism of Hsp90 release is still under investigation, and that Hsp90 is often found associated with the membrane-enclosed extracellular vesicles in cancer, an effort to discern between exosomes, oncosomes, apoptotic bodies or blebs, all of which can be released by tumour cells, could shed some light on whether the release of Hsp90 is regulated or occurs during cell death. Concerns regarding whether eHsp90 is an artifact of cell death or is actively secreted has been previously raised by other researchers, however no definite consensus has been reached (Li et al. 2012).

Interestingly, the group of Philip J. Hogg has been actively working on exploring Hsp90 as a biomarker of cancer cell death. Their compound GSAO covalently binds to the cytosolic Hsp90 which becomes accessible upon cell death, and [¹¹¹In]-In-DTPA-GSAO and [⁶⁷Ga]Ga-DOTA-GSAO radiotracers have been successfully used in vivo to visualize cell death in mouse tumour models using SPECT imaging (Ho Shon et al. 2020; Park et al. 2011; Shon et al. 2022). In addition, it has been reported that more aggressive tumours have higher levels of apoptosis (Morana et al. 2022). Similarly, elevated plasma levels of Hsp90α are also associated with tumour aggressiveness and poor prognosis across multiple types of cancer (Yuan et al. 2022). Therefore, it is plausible that there might be a direct link between elevated plasma Hsp90α levels and high degree of apoptosis in aggressive cancers with poor clinical prognosis. This makes us rethink the idea of targeting eHsp90 in cancer from membrane target to the cell death marker.

Imaging of cell death is a viable option for cancer treatment monitoring in patients undergoing other types of therapy. Therapeutic translation of tracers targeting cell death has not been widely explored, although ¹³¹I-labeled hypericin targeting necrosis has shown effectiveness in reducing tumour volume following the vascular disrupting treatment (Shao et al. 2015). Even though secreted Hsp90 can be further pursued as a cell death target to monitor therapy efficacy by quantifying cell death, we believe that Nbs might not be the best candidates as targeting vectors. For this purpose, a tracer that binds covalently to Hsp90 such as previously explored GSAO-based tracers might offer a better retention in the tumour environment relatively to the nanobodies with a fast washout. However, monitoring plasma levels of cell death markers might offer a less invasive and a lot less costly alternative.

Given the inconclusive validation of the membrane Hsp90 as a target in our chosen cell lines, we intend on screening more cell lines with reported membrane Hsp90 to see whether we could identify a more suitable model to evaluate our tracer. In parallel, our future research will focus on further development of the cytosolic Hsp90 tracers. One option is to investigate strategies such as conjugating nanobodies with the cell-penetrating peptides to improve their cellular uptake and access cytosolic Hsp90 (Herce et al. 2017). Additionally, we could explore optimizing the pharmacokinetic profile of small molecule tracers targeting intracellular Hsp90. Structural modifications that maintain Hsp90 affinity while enhancing stability and solubility can be explored. Ultimately, our findings highlight the need for a more nuanced understanding of Hsp90 in the extracellular space before we can confidently proceed with using it as a cancer cell target for PET diagnostics and radiotherapy.

Conclusions

Given our previous experience with small molecule based Hsp90 tracers, Nbs offer a great advantage when it comes to favourable pharmacokinetics and biodistribution and have rightly earned their place as a targeting vector for radiopharmaceuticals. However, we believe that more understanding is needed about the eHsp90 as a target for imaging and therapy. Namely, the interplay between extracellular Hsp90, cancer and cell death, as well as the nature of membrane Hsp90, i.e. whether it is an artifact of cell death or an actively expressed form by the living cells, must be more thoroughly investigated before we can consider eHsp90 as a viable biomarker for PET imaging and for now it remains controversial. Our future research will be focused on the development of the tracers for targeting intracellular Hsp90.

Abbreviations

(e)Hsp90

(Extracellular) heat shock protein 90

BLI

Bio-layer interferometry

BSA

Bovine serum albumin

CDR3

Complementarity-determining region 3

CM

Conditioned media

cpm

Counts per minute

CPS

Counts per second

CT

Computed tomography

ELISA

Enzyme-linked immunosorbent assay

IPTG

Isopropyl β-d-1-thiogalactopyranoside

K_D

Equilibrium dissociation constant

KD

Knockdown

k_off

Dissociation rate

k_on

Association rate

LB

Lysogeny broth

LDH

Lactate dehydrogenase

MMP2

Matrix metalloproteinase-2

MW

Molecular weight

Nb

Nanobody

OD600

Optical density at 600 nm

PBS

Phosphate buffered saline

PET

Positron emission tomography

PVDF

Polyvinylidene fluoride

R²

R-squared value

RCP

Radiochemical purity

RCY

Radiochemical yield

RPM

Revolutions per minute

RT

Room temperature

SDS-PAGE

Sodium dodecyl sulfate–polyacrylamide gel electrophoresis

SLS

Side light scatter

SPECT

Single-photon emission computed tomography

SUV

Standard uptake value

TACs

Time-activity curves

T_agg

Aggregation temperature

TBS-T

Tris buffered saline with tween 20

T_m

Melting temperature

χ²

Chi-squared value

Author contributions

VN conceptualization, data acquisition, data analysis, writing—original draft. CC data analysis (µPET/CT image analysis), writing—review & editing. JK data acquisition (⁶⁸ Ga-radiolabeling optimization), visualization (µPET/CT images), writing—review & editing. JS conceptualization, supervision, writing—review, funding acquisition. FR conceptualization, supervision, writing—review, funding acquisition. GB conceptualization, supervision, writing—review & editing, funding acquisition. All authors have read and approved the final manuscript.

Funding

The research conducted in this paper was funded by the VIB (C0401) and a grant from the KU Leuven Special Research Fund (C24E/19/070). JK was supported financially by the Fonds Wetenschappelijk Onderzoek—Vlaanderen through a senior postdoctoral grant [1226524 N-7029]. The PET-CT equipment was funded via an FWO medium infrastructure grant (AKUL15-30/G0H1216N). CC was supported through FWO I000321N.

Availability of data and materials

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.