Cell lines were obtained from the American Type Culture Collection (Manassas, VA) and submitted to no more than 20 passages, which were routinely tested for mycoplasmas (MycoAlert Mycoplasma Detection Kit, Lonza) and cultured in a humidified environment at 37°C and 5% CO₂. Ovarian cancer cell lines OVCAR 3 (ATCC^® HTB-161™) were cultured in RPMI 1640 supplemented with 20% foetal bovine serum (FBS) respectively and 0.1% bovine insulin. A1847 (Cellosaurus, CVCL_9724) cells and AsPc1 (ATCC^® CRL-1682™) were cultured in RPMI 1640 supplemented with 10% FBS. HEK 293T HEK 293T/17 (ATCC^® CRL-11268™), MDA-MB-231 (ATCC^® HTB-26™), and HeLa (ATCC^® CRM-CCL-2™) were maintained in Dulbecco modified Eagle medium supplemented with 10% FBS.

Adherent HEK293-T cells (70-80% confluence) were transfected with GFP-MSLN (Human Mesothelin/MSLN Gene ORF cDNA clone expression plasmid C-GFPSpark tag, SinoBiologicals) using Lipofectamine™ 3000 Transfection Reagent (Invitrogen) diluted in Opti-MEM (Gibco) according to the manufacturer’s instructions. MSLN expression was evaluated 12-24h hours post transfection by flow cytometry using 10 nM mAb K1 (Genetex) and Alexa647-conjugated goat anti-mouse (1/300, Miltenyi).

A1847-derived tumour spheroids were generated as previously described (13). Briefly, cells were harvested and seeded (10⁴ cells/well, > 95% viability) into Corning^® Costar^® 96 well ultra-low attachment round bottom plates. Plates were centrifuged at 400xg for 2 min and allowed to incubate at 37°C under standard conditions. After 3 days, ATTO-647N-conjugated nanobodies (50 nM) were added to the wells. Spheroids were carefully washed with PBS 1x at the indicated time and then fixed with 4% paraformaldehyde for 30 min at room temperature. Spheroids were first washed with PBS 1x/2% BSA and then PBS 1x before clearing with CUBIC I solution (25% urea, 25% N,N,N’,N’-tetrakis (2-hydroxypropyl)ethylenediamine, and 15%Triton X-100) (14). Spheroids were embedded in warm 1% Low Melting Point agarose into thin glass capillaries and imaging was performed using a Zeiss LightSheet Z.1 Microscope and a 5x/0.16 objective. Data analysis was performed with the Imaris viewer software.

A nanobody library was constructed in E. coli TG1 strain after immunisation of a llama (Lama gluma) with the recombinant human MSLN (rhMSLN-His, R&D Biotechnology) as previously described (15, 16). The first round of selection was performed on rhMSLN (10 µg/ml) immobilised on Maxisorp 96-well plates. Before the second round of selection performed at 4°C on OVCAR-3 cells (22x10⁶ cells), non-relevant epitopes were masked with anti-HEK nanobodies as previously described (17). Individual TG1 colonies (186 clones) from the selection outputs (round 2) were randomly picked and grown overnight at 37°C in 2YTAG (2YT complemented with 100µg/ml ampicilline and 2% glucose) in 96-microwell plates for the screening step. Overnight cultures were used to inoculate fresh 2YTA medium. After growing for 2h at 37°C, the production of nanobodies was induced by the addition of 0.1 mM IPTG and overnight growth at 30°C. Supernatants were harvested and used for screening on GFP-hMSLN-transfected HEK 393 T cells using non-transfected HEK 293 T cells as a negative control. The binding of nanobodies was detected using an anti-HIS antibody (Novagen, 1/500) and Alexa 647-conjugated goat anti-mouse IgG (Alexa 647-GAM, 1/300 Miltenyi). Selected clones were sequenced to identify distinct nanobodies (Genecust).

After the transformation of E. coli BL21 DE3 strain by positive phagemids, the nanobody production was performed as described in Behar et al. (16) after induction by 0,1 mM IPTG. Bacteria were pelleted and lysed in Bugbuster lysis buffer (Merck Milliopore) supplemented with benzonase (25 U/ml, and lysozyme (20 µg/ml). The his-tagged nanobodies were purified by affinity chromatography on TALON superflow™ cobalt resin (GE Healthcare, 28-9575-02) followed by a desalted step on Sephadex G-25 resine (Cytiva, 17085101). Nanobodies were stored in PBS. The protein concentration was determined spectrophotometrically (Direct Detect^®). Protein purity was evaluated by SDS-PAGE on a 4-20% Mini-PROTEAN^® TGX Stain-Free™ Protein stain free gel (BioRad) under reducing conditions. Western blotting was performed on nitrocellulose membrane using a trans-Blot Turbo Transfer System (BioRad). Precision Plus Protein™ unstained and Prestained Standards (BioRad) were used for SDS-PAGE and Western blot respectively.

To generate the sortag-nanobody, the coding sequence of the sortase A recognition sequence LPETG was introduced upstream of the C-terminal His-tag in the nanobody coding sequences. The resulting sequences were cloned in frame behind the pelB leader sequence in the pJF55 vector. Plasmids were transformed in E. coli BL21DE3 for standard protein expression and nanobody-sortag were purified by size exclusion chromatography (Superdex™ Increase 75 10/300GL (GE Healthcare)). The integrity and binding capacity of nanobody-sortag were verified by SDS-PAGE 4-20%, flow cytometry on A1847 cells, and biolayer interferometry.

Pentamutant sortase A plasmid (Addgene, plasmide #75144) was modified to replace the 6HIS tag by the Twin-Strep-tag (SA-WSHPQFEK-(GGGS)2-GGSA-WSHPQFEK) and the resulting enzyme was produced in E. coli BL21DE3 and purified by affinity chromatography on Strep-Tactin XTSuperflow^™ resin (IBA lifescience^®) according to manufacturer’s instructions. The peptides for sortase A-mediated ligation, GGGWWSSK(NODAGA)-OH, and H-GGGYK-biotin were purchased from Pepscan. NODAGA: 2,2′-(7-(1-carboxy-4-((2,5-dioxopyrrolidin-1-yl)oxy)- 4-oxobutyl)-1,4,7-triazanonane-1,4-diyl)diacetic acid). The sortase reaction was performed at 25°C for 2h in 50 mM Tris-HCl, 150 mM NaCl, and 10 mM CaCl₂ buffer pH 7,5 using a molar ratio of sortase/Nb-sortag/peptide-NODAGA of 1/10/100. The sortase was depleted on Strep-Tactin XTSuperflow™ resin and unbound peptide-NODAGA was removed by size exclusion chromatography on Superdex™ Increase 75 10/300GL (GE Healthcare) with PBS 1x pH 7.5 as running buffer.

The integrity and binding capacity of nanobody-sortag-Biot/NODAGA were verified by SDS-PAGE 4-20% and flow cytometry on A1847 cells, respectively. Matrix-Assisted Laser Desorption Ionization Time-of-Flight Mass Spectrometry (MALDI-TOF MS) was carried out for assessing the presence of biotin or NODAGA groups.

All flow cytometry experiments were performed on a MACSQuant cytometer (Miltenyi Biotec) using V-bottom 96-well microtiter plates. Cells were gated on single-cell populations and 10⁴ events were collected for each sample. Data were analyzed with the MACSQuant software and the results were expressed as the median of fluorescence intensity.

MSLN-positive cells (2x10⁵ cells/well) were first saturated using PBS/BSA 2% for 1 h at 4°C to avoid non-specific binding and incubated with serial dilutions of anti-MSLN nanobody for 1 h at 4°C in PBS/BSA 1%. Bound nanobodies were detected by staining 1 h at 4°C with a mouse anti-HIS mAb (1/500, Novagen) and Alexa 647-GAM. Three washes in PBS/BSA 2% were performed between each incubation step. The binding of monoclonal antibody K1 (Genetex) was detected with an Alexa-647- GAM. An irrelevant nanobody and/or Alexa 647 GAM were used as negative controls.

rhMSLN-HA-His (5 µg/ml) was coated on Nunc MaxiSorp™ ELISA 96 flat bottom microplates overnight at 4°C in PBS. After a saturation step with PBS/5% milk for 1 hour at RT, serial dilutions of nanobodies were added for 1 h at 4°C under shaking. Bound nanobodies were detected using an anti-cmyc antibody followed by an HRP-conjugated goat anti-mouse IgG (1/1000). The detection of peroxidase activity was performed using TMB (3,3’,5,5’-Tétraméthylbenzidine - KPL) substrate, and OD_450nm was measured on a Tecan Infinite^® M1000 plate reader after the addition of HCL 1N stop solution.

In both cases, the curves were fit with a sigmoidal dose–response equation, and EC50 values were calculated using the Prism 5 software.

Bio-layer interferometry (BLI) on the Octet R2 system (Pall ForteBio) was used to measure binding kinetics between nanobody and biotinylated rhMSLN-Fc. Streptavidin biosensor was rehydrated in binding buffer (PBS supplemented with 1% BSA and 0.05% Tween 20) for 10 min at 25°C. Biotinylated rhMSLN (10µg/ml) in binding buffer was bound to streptavidin sensor for 120 s. After an equilibration step in binding buffer for 30 s at 25°C, the MSLN-bound sensor was exposed to various concentrations of nanobody (50, 12.5, and 3.13 nM) for 300 s (association step) and then to a nanobody-free binding buffer for 300 s for the dissociation step. Kinetic constants were determined by fitting data with a 1:1 stoichiometry using the Octet analysis studio software.

MSLN and MUC16 binding capacity of tumour cell lines was quantified by DAKO QIFIKIT (DAKO Cytomation), according to the manufacturer’s protocol. Briefly, tumour cells were first labelled with anti-MSLN mAb K1 (150 nM, Genetex) or anti-CA125 mAb X75 (100 nM, ThermoFisher Scientific) on ice for 60 min. After several washes in PBS-BSA 2%, FITC-conjugated goat anti-mouse F(ab’)2 antibody diluted 1:50 (QIFIKIT, Agilent Dako) was used for labelling of both calibration and set-up beads (QIFIKIT, Agilent Dako) as well as tumour cells. Set-up beads were used to establish the window of analysis and the calibration beads were used to construct a calibration curve. The MSLN or MUC16 densities were determined by extrapolation on the calibration curve and expressed as specific antibody-binding capacity units after subtracting the background from the isotype control.

Epitope mapping was carried out by ELISA using different recombinant MSLN home-made constructs corresponding to MSLN domain 1 (aa296-390, DIH-Fc), truncated domain 1 (aa 296-354, DIL-Fc), domain II/III (aa391-598, HA-His-tagged DII/DIII) based on the putative MSLN structure (18). All the constructs were produced in the eukaryotic system using the Gibco™ EXPI 293™ Expression System Kit (Fisher Scientific) following the procedure provided by the manufacturer and purified by affinity chromatography on a GE Talon^® Superflow™ cobalt resin column. Correct conformation of the DII/DIII protein was assessed using the SD1-Fc fusion protein described by Tang et al. (19). ELISA procedure was as described above and the concentration of nanobodies was fixed at 100 nM. Affinity measurements were performed on D1H-Fc and D1L-Fc by ELISA and data were analyzed with GraphPad Prism software.

MSLN nanobody were site-specifically biotinylated using sortase A-mediated conjugation (eSrtA, Addgene) and a GGGYK-biotin peptide (Pepscan) at a molar ratio of nanobody-sortag/sortase/peptide-biotin of 1/0.1/20. For competition experiments A1847 cells (2x105 cells/well) in PBS/BSA 1% were incubated for 1 h at 4°C with serial dilutions of MSLN nanobody and their biotinylated counterpart at their EC50. After 2 washes in PBS/BSA 1%, cells were incubated with streptavidin-Alexa FluorX® (1/300, BioLegend) and analysed by flow cytometry. Data were analysed with GraphPad Prism software. Epitope binning was also analyzed by biolayer interferometry using an Octet R2 system (Sartorius). Biotinylated human MSLN (10 µg/ml) was immobilised on streptavidin sensors. In the first step, antibody 1 (Nbs A1 or S1 or amatuximab) in PBS (100 nM) was bound for 500 seconds to MSLN-bound biosensors. In the second step, antibody 1 (100 nM) was mixed with antibody 2 (100 nM, Nbs A1, S1, or amatuximab) to avoid a potential displacement of the already bound antibody 1. No wavelength shift should be observed if both antibodies share the same epitope.

To test the blocking property of MSLN nanobodies, hrMSLN-Fc was mixed with a 10-fold molar excess of anti-MSLN- or irrelevant nanobodies in PBS/BSA 1%. After a 30 min incubation at room temperature, the mixture was added to Nunc maxisorp 96-well plates pre-coated with rhCA125 (5µg/ml, R&D Systems^®) for 1 h at RT. After 3 washes in PBS/Tween 0.1% followed by 3 washes in PBS, MSLN-Fc binding was detected by the addition of HRP-conjugated goat anti-human Ig (1/1000, Life Technologies) for 30 min at RT. The detection of peroxidase activity was performed as described above. The percentage of binding inhibition was determined using the following formula: % blocking=100*(1-(A450nm assay/A450nm”no Ab condition”). (n=3)

OVCAR 3 cells (4 × 10⁴) were seeded in triplicate in black Corning^® 96 well flat clear bottom black microplates (3603). Two days later, GFP-MSLN transfected HEK 293 T cells (3 × 10⁵) were incubated in the presence or absence of anti-MSLN/ANef nanobodies (1 µM) at 4°C, 30 min in RPMI 10% FCS then added to the OVCAR-3 monolayer for 1 hr at 37°C. GFP signals were recorded at 508 nm before and after 7 washes in PBS using a fluorescent plate reader (Tecan Infinite^® M1000 - Life Technologies). The percentage of adhesion was calculated using the formula: (FAW/FBWsample)/(FAW/FBWmedium) *100 as described by Bergan et al. (20) with FAW= fluorescence after washes and FBW=fluorescence before washes. Incubation without antibody corresponds to the reference condition (n=3).

ATTO 647N labelling was performed using the Zedira TGase Protein Q-Labelling kit (L107) according to the manufacturer’s protocol. A size exclusion chromatography on GPC column (L107 kit) was carried out to remove the excess of ATTO 647N and the degree of labelling (DOL, dye-to-protein ratio) was calculated as follows: DOL= (A646nm*Eprot)/((A280nm-A646nm*CF280)*Emax) with A646nm: Absorbance at 646 nm, A280nm: Absorbance at 280nm, Eprot: Extinction coefficient of the protein in M-1cm-1, CF 280: Attenuation coefficient of ATTO647 at 280 nm (= 0.03) and Emax: Extinction coefficient of the fluorophore. The integrity and functionality of ATTO 647N conjugated nanobodies were assessed by 4-20% SDS-PAGE and flow cytometry.

A1847 cells (1.5x10⁴ cells/well) were grown on a glass coverslip immerged in 24 well plates for 2 days at 37°C. After washing the cells with PBS, a saturation step was carried on in PBS/BSA 3% for 1 h at 4°C. Then the cells were incubated for 1h at 4°C or 37°C with HA-His-tagged nanobodies (500 nM). The coverslips were washed with PBS-BSA1%, fixed with 4% p-formaldehyde for 30 min at RT, and permeabilised in PBS/0.5% Triton-X100 for 10 min before a 1 h-incubation with AlexaFluor 488-conjugated anti-HA antibody (1/200, LifeTechnologies) at RT. After several washes, the nuclei were stained with DAPI (1/2000 ThermoFischer) for 5 min. Fluorescence was evaluated using an Apotome fluorescent microscope (Zeiss), magnification: x63.

In accordance with the European Directive 2010/63/EU on the use of animals for scientific purposes, all procedures using animals were approved by the Institution’s Animal Care and Use Committee of Aix-Marseille University. The corresponding Project Authorizations (agreements APAFIS#28902 (TrGET platform) and #32157 (CERIMED)) were delivered by the French Ministry of Research and Higher Education. The animals were housed in enriched cages placed in a temperature-and hygrometry-controlled room with daily monitoring. Food and water were provided ad libitum.

A1847 MSLN^pos (5x10⁶ cells) and MDA-MB-231 MSLN^low (4.4x10⁶ cells) cells in a 1/2 (v/v) Matrigel (Corning Life Sciences, Bedford, MA, USA) suspension were implanted subcutaneously in 8-week-old female NOD-SCID IL-2Rgamma(null) (NSG) mice (n=24/cell line, n=8/group) and grown until all the tumours reached an average volume between 250 and 300 mm³. ATTO647N™ conjugated nanobodies (27 µg with an average DOL of 0.57) were injected via the tail vein and in vivo whole body fluorescence images were acquired using a Photon Imager (BioSpace Lab), at the following time points: 1, 6, and 24 h. Background fluorescence was determined on a xenografted mouse without antibody. Fluorescence signals within the regions of interest are expressed as photons per square centimetre per second per steradian (ph/cm²/s/sr) and determined using the following formula: Signal from ROI tumour - signal from ROI negative. After the final timepoint, animals were euthanised by cervical dislocation, and fluorescence imaging of individual organs was performed. Results were expressed as ph/cm2/s/sr (photon per square centimetre per second per steradian) or as a percentage of total signal (100*(organ signal - signal of non-injected mouse)/total fluorescence). The optical fluorescence measurements were performed with the PhotoAcquisition software (Biospace Lab), on 2 or 3 mice belonging to the same experimental group at the same time. We used identical acquisition parameters for all animals and for all time points (1, 6, and 24h). The analyses were carried out using the M3Vision software (Biospace Lab). The analysis parameters used for the 1 h time point were different from those used for the 6 and 24h time points. However, for each time point, the analytical parameters are the same for all animals. The fluorescence scales are indicated in the figures.

A1847 MSLN ^pos (10x10⁵ cells) and MDA-MB-231 MSLN^low (5x10⁵ cells) cells in a 1/1 (v/v) Matrigel (Corning Life Sciences, Bedford, MA, USA) suspension were implanted subcutaneously in 6-week-old female NMRI-Foxn1<nu>mice (A1847, n=20; MDA-MB-231, n=20) and grown until the tumours reached an average volume of 100 – 300 mm³ (average sizes of MDA-MB-231 and A1847 tumours were 197 +/-147 and 266 +/-186 mm³ respectively). For radiolabeling, nanobody S1 (50 µg in PBS) was first diluted in fresh 4M NH4OAc pH 5,0 (50 µl) and then mixed with 500 µl of Gallium-68 chloride ([⁶⁸Ga]GaCl). The mixture was stirred for 10 min at room temperature. The radiochemical purity was assessed by radio-thin-layer chromatography (solid phase: ITLC-SG) in two different mobile phases (sodium citrate 1 M, pH 5,0 and a solution of methanol/1M NH4OAc (1:1, v:v)) using a miniGITA radio-TLC scanner detector (Raytest, Straubenhardt, Germany). Radiolabeling stability was evaluated by iTLC after incubation of ⁶⁸Ga-nanobody in human plasma or NaCl 0,9% at 37°C for 30 and 120 min after radiosynthesis.

For dynamic microPET/CT imaging, mice were maintained under 1.5% isoflurane anaesthesia and imaged for 2 hr, immediately after intravenous injection of 68Ga-nanobody S1 (5-7 MBq/mouse) in the tail vein. A catheter 26G was placed into the tail vein of the mice to facilitate a rapid radiotracer injection. For the static experiment, microPET images were acquired for 20 min, 2h after intravenous injection of the radiotracer (5 MBq/mice). Image acquisition was performed on a NanoScan PET/CT camera (Mediso, Budapest, Hungary). After each PET recording, micro-CT scans were acquired for anatomical coregistration. Region-of-interest (ROI) analysis of the PET signal was performed on attenuation- and decay-corrected PET images using VivoQuant v.4.0 software (InVicron, Boston, USA) and tissue uptake values were expressed as a mean percentage of the injected dose per gram of tissue (%ID/g) ± SD. At the end of the static experiments, mice were sacrificed by cervical dislocation, and the radioactivity of individual organs was measured using a gamma counter (Wizard™ from Perkin Elmer). For the blocking experiment, mice (n=6) were pretreated with a 150-fold molar excess of unlabelled S1 (i.v.). Results were expressed as the percentage of the injected dose per gram of tissue (%ID/g) +/- SD.

Statistical analyses were performed with GraphPad Prism software (V5.01). Unpaired two-tailed t-tests or Mann-Whitney tests were used to do a pair-wise comparison. Two-way ANOVA followed by Turkey post hoc test were used for in vivo competing experiments. All statistical tests and resulting P values are indicated in the figure panels or the figure legends. P-values below 0.05 were considered statistically significant.

MSLN nanobodies were isolated from a phage-nanobody library generated after immunisation of llama with the mature recombinant human MSLN protein. Two successive rounds of selection were performed, first on recombinant MSLN protein and then on high-grade serous ovarian adenocarcinoma cell line OVCAR3. After screening for MSLN binding on HEK 293T cells transfected with human mature MSLN, three clones displaying different sequences (A1, C6, S1) were isolated, two of which (A1 et C6) have been described previously (15). The clone S1 was therefore selected for further characterisation. As Nb A1, Nb S1 displayed the hallmark residues of the VHH genes in the framework 2 regions (21). Nb S1 was produced at a large scale in E. coli as previously described (16) and purified by affinity chromatography on TaLon and size exclusion.

The capacity of nanobody S1 to target MSLN+ cells was evaluated by flow cytometry on a panel of cancer cell lines from different cancers, expressing various levels of MSLN and of its ligand MUC16 ( Figure 1A ; Supplementary Figures 1A, B ). The binding of Nb S1 was efficient on all cell lines as in all cases more than 90% of cells were labelled. The titration curves demonstrated that Nb S1 binds mature MSLN in a dose-dependent manner with an apparent affinity in the nanomolar range regardless of the presence of MUC16 ( Table 1 ). The kinetic parameters of Nb S1 ( Table 2 ) were determined by bio-layer interferometry. As shown in Figure 1B , a rapid distribution of the of ATTO 647N-labeled Nb S1 inside A1847-derived spheroids was observed compared to irrelevant aNef Nb, leading to a homogenous labelling throughout the entire spheroid in 2 hrs. The specificity of binding was confirmed by the absence of binding in the presence of an excess of soluble MSLN ectodomain. As well, ELISA on murine and human recombinant MSLN demonstrated that, unlike Nb A1, no binding of Nb S1 was observed on the murine protein.

The epitope targeted by Nb S1 was first investigated by bio-layer interferometry using streptavidin sensors pre-coated with biotinylated human mesothelin. The sensorgrams ( Figure 1C ) showed that Nb S1 binds to an epitope distinct from that of Nb A1 or amatuximab and in the same way whatever the order of its addition in the reaction. As expected from previous data on Nb A1 (13, 15), Nb A1 and amatuximab recognise the same epitope which overlaps the MUC16 binding site. Competition experiments on A1847 cells using biotinylated and non-biotinylated A1 and S1 confirmed these results as no competition was observed between Nb S1 and biotinylated Nb A1 or vice versa.

The immunoblotting experiment revealed that Nb S1 was able to detect human recombinant MSLN on western blotting in reducing conditions suggesting that Nb S1 recognises a linear epitope, as Nb A1 and mAb K1.

Next, truncated mutants of mesothelin were constructed based on the hypothesis that mature MSLN is organised in 3 distinct domains ( Supplementary Figure 1C ): a membrane distal domain I (residues 296–390), domain II (residues 391–486) and a proximal membrane domain III (residues 487-581) (22). Domain 1 (residues 296–390, D1) and truncated domain 1 (residues 296–354, D1L) were generated as Fc-fusions while the D2/D3 fusion protein was generated as a monomeric HA-HIS-tagged protein. The binding of Nb S1 was assessed by ELISA on immobilised mature and truncated MSLN ( Figure 1D ). Nb S1 binds both mature rhMSLN and D1 but not D2-D3 indicating that Nb A1 and Nb S1 bind the membrane distal domain. The apparent affinity of Nb S1 was assayed by ELISA on mature MSLN, D1-Fc, and D1L-Fc. While the apparent K_D of Nb A1 on the 3 targets were similar ( Table 3 ), Nb S1 displayed a significant decrease of apparent affinity for D1L-Fc, highlighting the importance of amino acids 359-390 for Nb S1 binding to MSLN.

Mesothelin is used both as a tissue marker and as a serum marker in association with CA-125 in several cancers. Many antibodies targeting MSLN recognise an epitope located in the MUC16/MSLN binding site (22), suggesting that the detection of MSLN can be affected by the presence of MUC16. To determine whether MUC16/MSLN interaction is hindered by Nb S1 binding, we evaluated the adhesion of GFP-MSLN transfected HEK 293 T on OVCAR-3 (MSLN+, MUC16+) monolayer in the presence or absence of Nb S1. As shown in Figure 2A , the presence of S1 did not impede the adhesion between the 2 cell lines in contrast to A1 which blocked more than 90% of MSLN-transfected cells adhesion. Similar results were obtained by ELISA on plate-bound recombinant human CA125.

To test whether Nb S1 is internalised upon MSLN binding, HA-His-tagged Nb S1 was incubated with OVCAR-3 cells at 37°C, a permissive temperature for internalisation, and at 4°C as control. As shown in Figure 2B , at 4°C, fluorescent staining was localised at the plasma membrane with both Nb A1 and Nb S1. However, A1 staining is fainter than with S1, which could be related to the expression of MUC16 on these cells. At 37°C, fluorescence signals were mainly localised in the cytoplasm, in favour of the internalisation of Nb S1 and Nb A1 upon MSLN binding.

The targeting capacities of Nb S1 were evaluated in vivo by fluorescence optical imaging using ATTO 647N-conjugated MSLN nanobodies and NSG mice xenografted with either A1847 cells (MSLN^high) or MDA-MB-231 (MSLN^low) cells. The absence of impact of site-directed conjugation of MSLN Nb with ATTO 647N was checked by ELISA ( Supplementary Figure 2A ). Once tumours reached around 200-300 mm³, mice were injected intravenously with 27 µg of ATTO 647N-S1, ATTO 647N-A1, or irrelevant aNef-ATTO 647N nanobodies. Whole-body fluorescence images were obtained at different time points (1, 6, and 24 h) ( Figure 3A ). Accumulation of ATTO 647N-S1 and ATTO 647N-A1 in the A1847 tumours were visible as early as 1 h post-injection and up to 24 h ( Supplementary Figure 2B ; Figure 3A ) in contrast to irrelevant-ATTO 647N nanobody. Quantification of fluorescence intensity at the tumour site revealed that the tumour uptakes of ATTO 647N-S1 and ATTO 647N-A1 nanobodies were significantly higher than that of aNef-ATTO 647N nanobody at all time points. In MDA MB 321 tumour-bearing mice, the fluorescence quantification showed that the 3 nanobodies generated similar signals that decreased in the same way over time ( Supplementary Figure 2B ).

To determine the biodistribution profiles of ATTO 647N-labeled S1 and A1, mice were sacrificed 24h post-injection, and ex vivo analysis of the fluorescent signal in resected tumours and organs was performed. Compared to other organs, kidney uptake was high, reaching up to 75-80% of the total fluorescence ( Figures 3B, C ), a well-described phenomenon for nanobodies due to their rapid blood clearance and retention by the kidney (23). As shown in Figure 3C , more than 40% of the total fluorescence signal was found in the A1847 tumours 24h post-injection of ATTO 647N-A1 and ATTO 647N-S1 compared to a mean of 17% with ATTO 647N-ANef. The fluorescence signal was significantly higher with ATTO 647N-S1 and ATTO 647N-A1 than with the irrelevant Nb in A1847 tumour-bearing mice ( Figure 3D ). In MDA-MB-231 tumour-bearing mice, the signal detected in the tumour in the presence of anti-MSLN Nb was not significantly different from that detected with the irrelevant Nb, highlighting the specific tumour capture in relation to the expression level of mesothelin.

After engineering anti-MSLN nanobodies for inserting a sortase A-recognition motif (LPETG), C-terminal specific conjugation with NODAGA chelator was performed using a pentamutant sortase A-twin-strepTag. The reaction efficiency was estimated by western blot through the decrease of nanobody-sortag-His tag. The reaction mixture was purified by a two-step process on Strep-Tactin XTSuperflow™ to remove the sortase followed by a size exclusion chromatography. Matrix-assisted laser desorption ionisation mass spectrometry (MALFI-TOFTOF) confirmed the presence of a major species corresponding to the NODAGA-conjugated S1 (14 910 Da) ( Supplementary Figure 3A ). IMAC purification on the Talon metal affinity column to remove the non-conjugated nanobody was not possible because the NODAGA cage chelates Co2+ ions with a fairly good affinity hindering the subsequent radiolabeling. Taking into account all steps, the overall conversion yield of unconjugated to conjugated nanobody was ranging from 30 to 60%.

The binding properties of NODAGA-conjugated Nb S1 on immobilised recombinant MSLN were also assessed by biolayer interferometry ( Supplementary Figure 3B ; Table 2 ). Also, competitive binding assays by flow cytometry confirmed that the NODAGA-conjugated S1 retained its binding properties on MSLN⁺ cells ( Supplementary Figure 3C ).

After several optimisation steps, NODAGA-Nb S1 was successfully radiolabeled with gallium-68 as evidenced by the radiochemical purity (RCP) of 97% in 1M sodium citrate pH 5.0 and 90% in 1M NH₄OAc/MetOH evaluated by thin layer radio-chromatography ( Supplementary Figure 3D ) and an apparent specific activity of 2-3 GBq/mg. In human plasma the ⁶⁸Ga-labelled nanobody was stable overtime 2 h at 37°C, (RCP>95% in 1M sodium citrate pH 5.0) ( Supplementary Figure 3E ).

Mice bearing A1847 or MDA-MB-231 tumours were injected with [⁶⁸Ga]Ga-NODAGA-S1 to determine its in vivo kinetics and distribution. As shown in Figure 4A , the nanobody was able to target MSLN+ tumours with a signal detectable 10 min post-injection, clearly visible 60 min post-injection and retained through 120 min scan. The time activity curves within the tumours presented in Figure 4B showed a rapid uptake in A1847 tumours that remained up to 2 h post-injection while MDA-MB-231 tumour uptake decreased over time. Kidneys and bladder showed high radioactivity accumulation in agreement with the well-described kidney retention and rapid blood clearance of nanobody.

PET static scans were also performed on mice bearing either A1847 or MDA-MB-231- tumours 2 h post-injection. Ex vivo biodistribution analyses demonstrated a significantly higher uptake of [⁶⁸Ga]Ga-NODAGA-S1 in MSLN^high tumours ( Figure 4C ) compared to low expressing/negative MDA-MB-231 tumours. The specific tumour uptake was validated by a competition experiment with an excess of unlabeled Nb ( Figure 4D ). No significant difference was noted in the presence of a 150-fold molar excess of cold S1 in the MSLN^low tumours while a 50% decrease of the signal was observed in the MSLN^high tumours. The biodistribution of ⁶⁸Ga-labeled nanobody was similar in all organs for the 2 groups of mice, except for the tumours ( Figure 4E ). As expected, high uptake of the radiotracer was observed in the kidneys in both MDA-MB-231 and A1847 tumour-bearing mice. Surprisingly, a higher uptake was observed in the liver and lung of 2-3 mice. One hypothesis to explain these results could be the presence of emerging tumour foci as in healthy mice no radiotracer uptake is observed in these organs.