Search and Characterization of ITM2A as a New Potential Target for Brain Delivery

Céline Cegarra (  celine.cegarra@sano .com ) Sano -Aventis Recherche Developpement Chilly-Mazarin Longjumeau https://orcid.org/0000-00021710-9796 Catarina Chaves Sano -Aventis Recherche Développement Chilly-Mazarin: Sano -Aventis Recherche Developpement Chilly-Mazarin Longjumeau Catherine Déon Sano -Aventis Recherche Développement Chilly-Mazarin: Sano -Aventis Recherche Developpement Chilly-Mazarin Longjumeau Tuan Minh Do Sano -Aventis Recherche Développement Chilly-Mazarin: Sano -Aventis Recherche Developpement Chilly-Mazarin Longjumeau Bruno Dumas Sano -Aventis Recherche et Développement Vitry-sur-Seine: Sano -Aventis Recherche et Developpement Vitry-sur-Seine Alfortville André Frenzel Universität Braunschweig: Technische Universitat Braunschweig Philip Kuhn Universität Braunschweig: Technische Universitat Braunschweig Valérie Roudieres Sano -Aventis Recherche Développement Chilly-Mazarin: Sano -Aventis Recherche Developpement Chilly-Mazarin Longjumeau Jean-Claude Guillemot Sano -Aventis Recherche Développement Chilly-Mazarin: Sano -Aventis Recherche Developpement Chilly-Mazarin Longjumeau Dominique Lesuisse Sano -Aventis Recherche Développement Chilly-Mazarin: Sano -Aventis Recherche Developpement Chilly-Mazarin Longjumeau


Abstract Background
Integral membrane protein 2A (ITM2A) is a transmembrane protein whose function is not well described. This target was identi ed as highly enriched in human brain vs peripheral endothelial cells by transcriptomic and proteomic studies during the European Collaboration on the Optimization of Macromolecular Pharmaceutical Innovative Medicines Initiative (COMPACT IMI) consortium. The object of the present paper is to report the work we have undertaken to characterize this protein as a potential brain delivery target.

Methods
A series of ITM2A constructs, cell lines and speci c anti-human and mouse ITM2A antibodies were generated. Binding and internalization in Human Embryonic Kidney 293 (HEK293) cells overexpressing ITM2A and in brain microvascular endothelial cells from mouse and non-human primate (NHP) were performed with these tools. The best antibody was evaluated in an in vitro human blood brain barrier (BBB) model and in a mouse in vivo pharmacokinetic study to validate blood brain barrier crossing.

Results
Antibodies speci cally recognizing extracellular parts of ITM2A or tags inserted in its extracellular domain showed selective binding and uptake in ITM2A-expressing cells. However, despite high RNA expression in mouse and human microvessels, this protein, is rapidly downregulated upon endothelial cells culture, probably explaining why transcytosis could not be evidenced in vitro. An attempt to directly demonstrate in vivo transcytosis in mice turned out unconvincing, using a cross reactive anti ITM2A antibody on one hand and in vivo phage panning of an anti ITM2A phage library on the other hand.

Conclusions
The present article describes the work we have undertaken to explore the potential of ITM2A as target to mediate transcytosis through BBB. This work highlights the multiple challenges linked to the identi cation of new brain delivery targets.

Background
Brain is a highly protected tissue. The endothelial cells lining the blood vessels that are at the interface of blood and brain are different than the peripheral endothelial cells, and as opposed to them are extremely tight, non-fenestrated and equipped with many e ux systems. This blood brain barrier is only permeable to very small lipophilic compounds but is actively preventing most molecules to enter and in particular large or polar molecules such as biotherapeutics and antibodies (1,2). This explains why some enormous medical needs remain to be addressed in particular for targets for which biologics are the main modality in therapeutic area such as neurosciences, oncology (e.g. central nervous system (CNS) lymphoma or glioblastoma) or rare diseases. For instance, the use of therapeutic antibodies for CNS disorders such as Alzheimer's, Parkinson's, Huntington's diseases or brain cancers has been very limited so far (3) owing to the presence of the BBB. This is also explaining why so few biologics are in development in CNS. The biologics that are on the market in neurology are most certainly acting peripherally (or else given intrathecally) (4). Therefore, strategies to increase brain exposure of biotherapeutics are key to the success of biotherapeutics in this area.
So far, the most successful strategy to carry biotherapeutics to the brain using systemic route has been making use of a ligand or antibody against a receptor-mediated transcytosis (sometimes referred to as the 'Trojan horse' approach). Several receptors such as insulin (5), transferrin(6), lipoprotein-related proteins (7,8), low density lipoprotein (9) or Insulin-like Growth Factor 1 (10) receptors have been used. Several challenges however remain in the eld. One of them is linked to the fact that all these receptors are ubiquitously expressed. So far, no brain-speci c receptor capable of mediating such transcytosis has been discovered.
During the COMPACT IMI consortium (https://www.imi.europa.eu/projects-results/projectfactsheets/compact ) an approach integrating different omics levels of proteomic, MicroArray and RNA sequencing approaches was engaged from human primary endothelial cells from brain, liver and lung aimed at identifying candidate membrane proteins highly enriched in brain vs liver or lung (11). The resulting over 60000 RNA's were then processed through several lters, downsizing to mRNA's not detected at all in liver and/or lung or with high differential level in brain, mRNA's with mostly transmembrane expression, association to the BBB vasculature, expression and selectivity comparable between rodents and humans and high degree of conservation between orthologs. This led to a few mRNA's such as basigin (12) and Low-density lipoprotein receptor-related protein 8 (LRP8)(13) already reported as brain transport mechanisms validating therefore the approach. ITM2A was identi ed among these mRNA's. ITM2A (alias E25A or BRICD2A) is a 263-amino acid protein with a single transmembrane domain (14). Its ubiquitous expression is higher in thymus where it was shown to be an activation marker in thymocyte development (15). ITM2A is mainly believed to be associated with cell differentiation during myogenesis(16, 17), chondrogenesis(18-22) and odontogenesis (23). The overall homology between mouse and human is more than 95% in the extracellular domain (24). The protein has a motopsin-binding Brichos domain within the COOH-terminal extracellular domain(25) whose signi cance is poorly understood (25). Brichos domains appear to bear a chaperone function in different biological situations and particularly bind amyloid brils (26). Recently ITM2A expression was shown to negatively regulate autophagic ux by inhibiting lysosomal function through physical interaction with vacuolar Adenosyl Tri Phosphatase (27).
Expression in mouse and human brain has been reported to be homogenous in all brain (Protein Atlas ITM2A) and quite speci c of endothelial cells vs neurons, microglia, oligodendrocytes or astrocytes as shown in brain RNA seq database (Fig. 1). In fact, it is referred to as an endothelial cell-speci c gene(28).
An analysis of ve human and murine cell type-speci c transcriptome-wide RNA expression data sets that were generated within the past several years also identi ed ITM2A as one of the top expressed genes in endothelial cells (29). A recent single cell RNA seq analysis of 20 organs in mice also established the ITM2A gene speci c for brain endothelial cells (30).
Two precedents point to speci c association of ITM2A with the blood brain barrier. ITM2A was identi ed as microvasculature speci c through the screening of subtractive cDNA libraries from rat brain capillaries versus kidney/liver on one hand (31,32) and from porcine brain and aortic endothelial cells on the other hand (33). In addition, the expression of ITM2A found in freshly isolated porcine Brain Microvascular Endothelial Cells (BMEC) was shown to be lost upon culture similarly to some known BBB markers (33).
ITM2A as an endothelial brain speci c transmembrane protein has not been associated in the literature with brain transcytosis or transport. The present report describes the work we have done to characterize this target as a potential brain delivery candidate.

Animals
Male and Female Cynomolgus monkeys (Macaca fascicularis) aged from 4.8 to 5.9 years were purchased from Le Tamarinier and Noveprim Ltd. (Mahebourg, Mauritius). Six animals were grouphoused in aviaries or interconnected mobile cages and two animals were individually housed in interconnected mobile cages. Animals were housed under controlled conditions (20-2 4°C, 40-70% humidity, 10-15 renewals per hour of ltered, non-recycled air, 12-h light cycle) with free access to ltered tap water and daily distribution of expanded diet (sodium dodecyl sulfate SDS, France) and fruits or vegetables. Animals used for the isolation of brain microvessels were at disposal following pre-clinical studies. Prior to the isolation of brain cortical microvessels, animals were submitted to a washout period of a minimum of 1 month.
Male mice C57BL6/J aged from 6 to 8 weeks were purchased from Charles River Laboratories (France) Pregnant mice C57BL/6JRj were purchased from Janvier Lab's (France) between E10 and E12.
After arrival, mice were housed individually (for pregnant mice) or grouped (for male, 6 animals per cage) in an enriched environment in a pathogen-free facility at a constant temperature of 22 ± 2°C and humidity (50 ± 10%) on a 12-h light/dark cycle with ad libitum access to food and water.
Male rats Wistar: crl WI were purchased from Charles River Laboratories (Germany) After arrival, rats were housed individually in an enriched environment in a pathogen-free facility at a constant temperature of 22 ± 2°C and humidity (50 ± 10%) on a 12-h light/dark cycle with ad libitum access to food and water.
Isolation of Brain Microvessels from Cynomolgus Monkey and rodents Cortex Brains from Cynomolgus monkeys or rodents were collected shortly after the euthanasia of the animal into ice-cold Hibernate A medium (ThermoFisher). All subsequent steps were performed at 4 °C and under a biological safety cabinet. Brain cortex was isolated and placed in petri dishes containing ice-cold Hanks' Balanced Salt solution HBSS. The meninges and the cortical white matter were removed. The collected tissues were transferred into a new sterile container with HBSS, nely minced with a scalpel, and then pelleted by centrifugation (5 min at 600 g, 4°C). The pellet was resuspended in a collagenase/dispase solution (Roche, Meylan, France, Collagenase 0.1 U/mL; Dispase 0.8 U/mL prepared in Ca2+/Mg2+ free HBSS) containing type I DNAse at 20 U/mL and Tosyl-L-lysyl-chloromethane hydrochloride (TLCK) at 0.147 g/mL, and incubated at 37°C for 60 min, under vigorous agitation. The digested tissue was carefully homogenized, and centrifuged for 5 min at 600 g, 4°C. The resultant pellet was resuspended in HBSS containing 20% Bovine serum albumin (BSA) and centrifuged at for 30 min at 2000 g, 4°C. The myelin ring-containing supernatants were discarded, and the vessel-containing pellet was resuspended and re-incubated in the collagenase/dispase solution in presence of DNAse and TLCK for 30 min at 37°C (except microvessels from mice). This suspension was re-pelleted by centrifugation 5 min at 600 g, 4°C, and the nal pellet (named passage 0 Day in vitro 0 (P0D0) fraction from this point onwards) is resuspended in endothelial cell medium endothelial basal medium (EBM-2) supplemented with Kit endothelial cell growth medium micro vascular (EGM-2 MV) Single Quots, Lonza, Basel, Switzerland) containing 3 g/mL puromycin, before set up in pre-coated (collagen IV 100 g/mL, bronectin 10 g/mL, Sigma, Saint Quentin Fallavier, France) cell culture asks, and incubated at 37°C, 5% CO2 for 7 days. Every 2 days the cell medium was changed, the supplemented puromycin concentration lowered to 2 g/mL, and subsequently removed. Following 7 days of expansion at P0D7, Brain Endothelial Cell (BEC)s from cortex, were further singularized and re-plated de novo for further 7-day cell expansion (P1D7).

Plasmid constructs
Various human in uenza hemagglutinin (HA) human ITM2A cDNA with coding sequence: NM_004867 and its variants were synthetically made by GeneART and introduced in a PiggyBac® transposon mammalian expression vector pBH 6450 with a Cytomegalo virus (CMV) promoter. HA tag is inserted at different position in ITM2A sequence: at NH2 or COOH side or in the Brichos domain: at NH2 or COOH side of the Brichos domain or in the middle in order to disrupt the eventual function of this domain. Plasmids were named: pBH-hITM2A HA NH2, pBH-hITM2A HA COOH, pBH-hITM2A wild type (wt), pBH-hITM2A Brichos HA COOH, pBH-hITM2A Brichos HA NH2, pBH-hITM2A4 Brichos HA mid.
Cells and cloning HEK293/CVCL_0045 cells were obtained from Deutsche Sammlung von Mikroorganismen und Zellkulturen (DSMZ). For thawing: cells are thawed rapidly in water bath at 37°C, centrifuge at 900 g for 4mn, pellet is re-suspended in culture medium. Cells are cultured in the following medium: Dulbecco's Modi ed Eagle Medium Gibco 21969 (glutamine-free: selection pressure for HA tag or Blasticidin 15µg/mL: selection pressure for GFP tag); 10% Foetal Calf Serum Eurobio CVFSVF06 heat inactivated Australian; Penicillin / Streptomycin Gibco 15140 100 U / mL nal. At con uence, the cells are rinsed with Phosphate-buffered saline (PBS) (-Ca2+, -Mg2+), detached by enzymatic treatment (Accutase Sigma A6964 or trypsin) at 37°C for 3mn, centrifuge at 900g for 4mn. The pellet is re-suspended in culture medium and diluted to 1/10 in fresh medium for seeding in new asks.
Human Cerebral Microvascular Endothelial Cell (hCMEC)/D3 cells were obtained from Cedarlane. Cells are cultured in the medium cell biologics add with supplements for endothelial cells H1168 as described above.
HEK293/CVCL_0045 cells were transfected with pcDNA6.2™-EmGFP or cotransfected with PiggyBac® transposon expression vectors bearing wt ITM2A and variants cDNAs and transposase plasmids 6209 (10:1) by using Lipofectamine 2000®, following manufacturer instructions. Transfections were performed at 50% con uence in 24 wells plate with 500 ng of plasmids. For stable transfections, clones are obtained by limit dilution in 96 well plates with blasticidin selection for pcDNA 6.2 or in glutamine free media corresponding to the glutamine synthase selection marker of pBH 6450. Each individual cell is ampli ed in a 6 wells plate, then ITM2A expression is checked rst with uorescent microscope for GFP tag, second by Western Blot for all ITM2A constructions.

Antigen preparation
The human and murine ITM2A extra-cellular domain (ECD) gene sequence was synthesized and fused to a human Fc in a mammalian expression vector. After preparation of transfection-grade DNA, a transient transfection of HEK293 cells was performed. After 7 days of cultivation, the ITM2A-Fc containing culture supernatant was isolated and puri ed by Protein A a nity chromatography according to standard protocols. After buffer exchange to PBS, the puri ed antigen was analyzed by UV/VIS spectrometry and SDS-PAGE.

Antibody-phage selection
The target protein (ITM2A-human Fc) and negative antigen (Protein N Standard, Siemens, QQIM13) were immobilized onto the wells of an Enzyme-Linked Immunosorbent Assay (ELISA) plate (Corning, 9018) for 1 h at room temperature (RT) (1 µg each). After removal of non-bound antigen, ELISA wells were blocked with a 2% BSA solution for 16 h at 4 °C. After washing of the plates with PBS-T (PBS containing 0.05% Tween 20), the antibody-phage library was added to the immobilized negative antigen and incubated for 1 h at RT to remove Fc speci c or polyreactive antibody-phage. Additionally, 5 µg Protein N Standard was added as soluble competitor. Non-bound antibody-phage were recovered and incubated on the immobilized target antigen for 2 h at RT. Non-bound or weakly bound antibody-phage were removed by washing with PBS-T (10x) before antigen-speci c phage were recovered by Trypsin (10 µg/ml) elution for 30 min at 37 °C. The antibody-phage were rescued by infection of TG1 cells (OD600=0.5) for 30 min at 37°C . After propagation of the cells for additional 30 min at 37 °C and 500 rpm, ampicillin (100 µg/ml) and glucose (100 mM) were added to the 2YT culture medium. Bacterial propagation was continued for 1 h at 37 °C and 500 rpm. Then, bacteria were co-infected with M13K07 helper phage, incubated at 30 min at 37°C followed by another incubation for 30 min, 37 °C and 500 rpm. Double-infected bacteria were centrifuged (4000 g, 10 min) and the cell pellet was resuspended in fresh 2YT, containing Ampicillin (100 µg/ml) and Kanamycin (50 µg/ml). For the ampli cation of antibody-phage particles, the incubation was continued for 16 h at 30 °C. Then, the culture was centrifuged (4000 g, 10 min) and antibody-phage containing supernatant was recovered and used for the next panning cycle. Three panning cycles were performed in total. In each cycle, the number of washing steps was increased (cycle 2: 20x, Cycle 3: 30x) to increase the stringency of the selection.

Antibody screening and sequence analysis
After the third panning cycle, the eluted phage were used to infect XL1 (OD600=0.5) for 30 min at 37 °C and streaked out on 2YT agar plates, containing ampicillin (100 µg/ml), glucose (100 mM) and tetracycline (20 µg/ml). Incubation was continued at 37 °C until single colonies were observed. Single clones were isolated and transferred into 96-well plates, containing 2YT, ampicillin (100 µg/ml), glucose (100 mM) and tetracycline (20 µg/ml). The bacteria were cultivated for 16 h at 37 °C and 300 rpm. Then, 15 µl of the overnight cultures were used to inoculate new 96-well plates, containing 2YT medium, ampicillin (100 µg/ml), tetracycline (20 µg/ml) and IPTG (50 µM). The bacteria were cultivated for 16 h at 30 °C and 300 rpm. The cultures were centrifuged (4000 g, 10 min) and Single-Chain Variable Fragment (scFv) containing supernatants recovered. The scFv containing supernatants were used for antibody screening. In brief, supernatants were diluted with a 2% BSA solution (in PBS, containing 0.05% Tween20), added to the immobilized antigens in a 384 well ELISA plate (20 ng/well) and incubated for 1 h at RT. After washing, binding of the scFv antibodies to human ITM2A-human Fc, murine ITM2A-human Fc, Protein N Standard or BSA was detected via a Myc-tag using a secondary horseradish Peroxidase (HRP) conjugated antibody. The binding was quanti ed by TMB reaction and absorbance reading at 450 nm. Target speci c antibody clones were isolated, and the DNA sequence of the respective scFv antibody analyzed by sanger sequencing.

Immunoglobulin G (IgG) expression
The VH and VL sequence of selected antibody clones was ampli ed by PCR and cloned into mammalian IgG expression vectors. After preparation of transfection-grade DNA, a transient transfection of HEK293 cells was performed. After 7 days of cultivation, the IgG containing culture supernatant was isolated and puri ed by Protein A a nity chromatography according to standard protocols. After buffer exchange to PBS, puri ed antibodies were analyzed by UV/VIS spectrometry, SDS-PAGE and ow cytometry analysis for cell binding.

Transcriptomic
Transcriptome Sequencing (RNAseq) and mRNA Expression Analysis COMPACT IMI consortium As described previously by Li et al (11).

RNA Samples for Library Preparation
As described previously by Chaves et al (34) Frozen cell pellets from the cellular P0D0, P0D7 and P1D7 fractions were lysed using QIAzol Lysis Reagent (Qiagen, #79306, Courtaboeuf, France). Total RNA was then isolated from the lysates on a QIAcube instrument (Qiagen) using the Rneasy Mini QIAcube kit (Qiagen, #74116) and following the manufacturer's instructions. The RNA concentration was determined using the Qubit RNA HS Assay Kit (Invitrogen, #Q32852, Illkirch-Graenstaden, France) and the quality and integrity was assessed on a Bioanalyzer 2100 (Agilent Technology, Les Ulis, France) with an Agilent RNA 6000 nano kit (Agilent Technology, #5067-1511).

RNA Sequencing (RNAseq)
As described previously by Chaves et al (34), the RNAseq libraries were prepared with 30 ng of input total RNA using the NEBNext Ultra II Directional RNA Library Prep Kit for Illumina (New England Biolabs, #E-7765S, Évry-Courcouronnes, France) with the NEBNext rRNA Depletion Kit (New England Biolabs, #E6310L) and following the manufacturer's instructions. The libraries were then paired end sequenced (75 cyclesx2) on the NovaSeq 6000 instrument (Illumina, Paris, France) using the NovaSeq 6000 SP Reagent Kit (300 cycles; #20027465, Illumina).

Reverse transcription quantitative Polymerase Chain Reaction (RTqPCR)
Cells or tissus were lysed with RLT buffer from Qiagen added with 1% β-mercapto-ethanol according to manufacturer instructions. mRNA was extracted with Qiagen MiniKit followed by DNase step using the

Colocalization confocal imaging
Confocal microscope images were acquired with SP8 LEICA microscope, 40X objectives. For the acquisition ITM2A was directly visualized by EmGFP 488. Organelles were stained with following primary antibodies anti Lysosomal-associated membrane protein 1 (LAMP1) ab24170 for lysosome detection, anti-giantin ab37266 for Golgi. Antibody anti-organelle were detected by staining with appropriate secondary antibody Alexa 633 goat anti mouse A21050 1/1000 or goat anti rabbit A21070 1/1000 Western Blot NHP-derived BEC were lysed using ice-cold radio-immunoprecipitation assay (RIPA) or cell lysis buffer containing protease inhibitor cocktail (ThermoFisher) centrifuged at 15,000 g for 15 min, and supernatant fractions were collected. Samples were added with SDS and loading buffer then denaturized by heat 95°C for 5 min. These were loaded into 4-12% Tris-Glycine SDS-page gels (Invitrogen), and let to migrate for 1h at 180V. Samples were then transferred onto polyvinylidene uoride (PVDF) or nitrocellulose membranes using an iBlot 2 Dry Blotting System (Invitrogen) on the P0 program (20 V for 1 min, 23 V for 4 min, 25 V for 2 min). PVDF membranes were then rinsed with Tris-buffered saline with 0.1% Tween 20 (TBST) and blocked for 1 h in 5% non-fat dry milk in TBST (blocking buffer). Membranes were rst probed overnight at 4°C with primary antibodies in blocking buffer (anti-ITM2A polyclonal AF4876 and 18306-1-AP, EmGFP A11122, anti-HA 901509, anti α-tubulin T9026) and then probed with secondary antibodies diluted in TBST for 1 h at RT (1:10,000 diluted HRP-coupled goat anti-mouse IgG or goat antirabbit IgG, GEHealthcare). Following secondary antibody incubation, membranes were rinsed thoroughly with TBST, imaged using a LICOR Odyssey Imager and bands quanti ed using Multi-Gauge v3.0.

Mass spectrometry
Post-mortem human brain samples (occipital cortex, devoid of pathological ndings) from three 69 to 79year-old, male, non-demented, control donors were obtained from external biological resource centers in full accordance with legislation and ethical standards. Microvessels were isolated as described above for monkeys. Tissus or cells were lysed in Preomics 2X buffer then crushed gently with gentle MACS Dissociator Miltenyi Biotec, program Protein for 1min. Samples were centrifugated 10min at 4000g, 4°C, supernatants were collected and dilute to 1X Preomics buffer. Benzonase 1/100 was added and incubated 10 min à 95°C, 1000 rpm. Samples were centrifugated 20mn at 13000g and supernatants were collected. Proteins are quanti ed by spectrophotometry using bicinchoninic acid method and spectraMax i3x. 25 µg of protein were digested on Preomics lter in 50µL following manufacturer instructions during 3h at 37°C, after evaporation, samples were diluted at 0,5µg/µL in 50µL LC-Load buffer, vortexed and sonicated. Heavy peptides were added to the sample solution to inject 5 fmol of heavy peptides and 2µg of proteins. 7 heavy stable isotope labelling (SIL) peptides spiked in nal digest prior to Parallel Reaction Monitoring-Mass Spectrometry analysis on Q-Exactive HF /NanoRSLC 3500. From ITM2A_MOUSE Integral membrane protein 2A Q61500, spiked SIL peptides for detection were: IAFNTPTAVQK, NLVELFGK, EDLVAVEEIR, DLLLGFNK. Spiked SIL peptides used for hTFRC detection were: DSAQNSVIIVDK, LTVSNVLK, SGVGTALLLK, AAAEVAGQFVIK, LTTDFGNAEK.

Transport assays
Internalization assay HEK293 ITM2A cells are cultured on 96 wells plate coated with poly-L-lysine at 50000 cells/well. After 24h cells were incubated 1h at 4°C in cell medium with 5µg/mL of studied antibody (Anti ITM2A RandDsystem AF4876, Yumab antibodies anti ITM2A YU147-A01, E02 and H07, YU93-G04, GFP antibodies A31852 or A11122, anti-HA 901509). Cells were washed with ice cold PBS then incubated 1h at 37°C or at 4°C for control. Cells were washed 2 times with ice cold PBS, then 2mn with acid buffer (glacial acid acetic 1/167 in PBS, pH = 2,5) and nally 2 times with ice cold PBS. Cells were xed and stained as described above. Images were acquired with Operetta High Content Screening (PerkinElmer) objective 40X water confocal mode.
In Vitro Transcytosis and Permeability Measurements Brainplotting models All human samples were provided by Brainplotting (35)(iPEPS, Institut du Cerveau et de la Moelle épinière, Hôpital Universitaire de la Pitié-Salpêtrière, Paris, France) in partnership with Sainte-Anne Hospital, Paris (neurosurgeon Dr. Johan Pallud) and harvested during tumor scheduled resection surgery with written informed consent from the patients (authorization number CODECOH DC-2014-2229). Human brain microvessels were obtained from surgical resections of one patient: a 35-years-old female suffering from of a diffuse oligoastrocytic grade III glioma. Microvessels were isolated from peritumoral or healthy brain tissue using an enzymatic procedure(34) adapting methods previously published for rats(36, 37). Brie y, tissue samples were carefully cleaned from meninges and excess of blood; then, an enzymatic mix was used to dissociate the tissues and microvessels were isolated by retention on a 10 mM mesh. Cells were cultured in EBM-2 medium (Lonza, Basel, Switzerland) supplemented with 20% serum and growth factors (Sigma)(36, 37). After seeding brain capillaries in petri dishes, brain primary microvascular endothelial cells were shortly ampli ed and seeded (P1D0) on Transwell (Corning) with microporous membranes (pore size: 0.4 mm) in monoculture.
Test (1 µg/mL of internal anti-human/cynomolgus Transferrin Receptor C (TFRC) antibody or anti-ITM2A YU93-G04) and control antibodies (1µg/mL mouse IgG, clone MG1-45, BioLegend) were added onto the upper chamber on day adapted to the transport assay de ned by Brainplotting. Fresh endothelial cell medium with none of these compounds was added onto the bottom chamber. Final aliquots from both chambers were taken 240 min following incubation at 37°C, 5% CO2. Compound levels in mother solutions (T = 0  In vivo experimentations Mouse brain collection Each mouse was anaesthetized in iso urane gas chamber then transcardiacally perfused with Liheparinate solution at nal concentration 20 U/mL in sterile PBS. Perfusion was realized with 48mL delivered at the speed of 8 mL/mn. Brain samples were collected. Each mouse was decapitated immediately after perfusion. Perfused brain was removed, cerebellum and brainstem were separated and eliminated. Brain cortex were washed in ice-cold PBS, collected in preweighted Precellys tube and stored immediately at -80°C freezer (or dry ice) until use. Then, the preweighted hemispheres were thawed and homogenized in 5 vol. (v/w) of brain lysis buffer (1% NP-40 in PBS containing complete mini ethylene diamine tetra-acetic acid -free protease inhibitor cocktail tablets, Pierce) using bead homogenizer. Homogenized brain samples were then rotated at 4°C for 1 hour before centrifuge at 20000 g for 20 min.
Pharmacokinetic study in vivo in mouse 5-10 mg/kg (35-70 nmol/kg) in a single dose of anti-ITM2A (clone YU93-G04) or control (anti-trinitrophenyl (TNP), batch VA2-17-419-1, internal production) antibodies were administrated by caudal intravenous injection into mice, C57Bl/6, male, 20-25g (n = 3/condition). 5h post-injection, plasma and saline-perfused brains were collected, then concentration was determined by ELISA immunoassay as described above In vivo panning Yumab in mice Yumab provided ampli ed phage library to Sano . First, a naïve human antibody phage library was enriched for antigen speci c antibodies against ITM2A proteins. The antibody phage output was ampli ed and puri ed by poly-ethylene glycol (PEG)/ NaCl puri cation. Puri ed antibody phages were directly used for the in vivo panning: 1011 antibody phage particles (~10 µl) were mixed from each panning output. The antibody phage mix was injected into the mouse. Brains were isolated after 1 h and 24 h (2x each). Brains were collected as described above.
Antibody phages in the homogenate were used for infection of E. coli. Bacteria were selected and used for production of monoclonal scFv antibodies. About 800 clones were picked and screened by scFv in ELISA assay.

Statistical Analysis
The statistical signi cance of differences between groups was analyzed using GraphPad Prism v9.

Results
Our strategy is outlined in Fig. 2. The rst objective was to produce tools to study the mechanisms: cells expressing ITM2A and the extracellular domain of the protein to produce antibodies preferably mousehuman cross reactive. With this in hand, uptake, internalization, and tra cking in either the above overexpressing cells or in BEC's would provide the rst lter. Transcytosis and in vivo brain pharmacokinetic would lead to nal validation (Fig. 2).

Building up cell lines expressing ITM2A
In the absence of available monoclonal antibodies, we devised a strategy that could bring clues on the potential of this protein to carry biologics into the brain using speci c well characterized cargos. To this end, we engineered several constructs of human or murine ITM2A. One plasmid, named wt, was designed with no tag, the others bearing either an HA-tag (HA-tag are amino acids 98-106 of human in uenza hemagglutinin glycoprotein) or a GFP-tag at different positions, namely on the C-Terminal (C-Ter) extracellular domain or on the N-Terminal (N-Ter) intracellular domain of ITM2A. In addition, the HA tags were positioned at three distinct locations of the Brichos domain within the extracellular portion to bring additional insight on the potential position involved in internalization. This information could be exploited later for the generation of antibodies speci cally recognizing a part of the antigen to optimize their transcytosis. Designed plasmids are summarized in Fig. 3.
We selected adherent HEK293 cells well suited for immuno uorescence analyses and produced 10 clonal cells lines by recombination with the various tagged or untagged mouse or human plasmids displayed in Fig. 3 (Table 1). Clones were obtained after limit dilution and selection by either uorescence and Western Blot for GFP tag or by Western Blot for HA tag and wt ITM2A (Additional le 1).  (Fig. 4). GFP uorescence was detected in all N-Ter and C-Ter constructs in HEK293 expressing both human and mouse ITM2A.

Cellular localization of ITM2A
Aside from membrane localization as could be showed by the above binding experiments, ITM2A could be localized in Golgi of the overexpressing HEK293 cells after evidencing co-localization with Giantin ( Fig. 5A). No ITM2A could be seen in the lysosomes as shown by LAMP1 co-labeling (Fig. 5B).

Cell uptake of antibodies
With these cell lines in hand, we went on to evaluate their capacity to uptake antibodies after binding with the protein. We could perform these experiments using several handles.
First, we performed immuno uorescence studies on HEK293 cells expressing human ITM2A with two commercially available polyclonal antibodies against extracellular epitopes of human ITM2A: AF4876 and 14407-1-AP ( Fig. 6A and B). The rst one displayed membrane labeling at 4°C while the second one could not be detected at 4°C maybe linked to a lower a nity or different behavior after acid washing (performed before analysis). Both antibodies readily internalized at 37°C leading to punctate labelling and demonstrating active uptake. The weaker signal observed with AF4876 on the C-Ter GFP cell line could result from steric hindrance of the GFP tag for antibody binding.
The same results were obtained using anti-GFP antibodies either directly uorescent (A31852, Fig. 6C), or after detection of an unlabeled anti-EmGFP antibody (A11122) with a secondary labeled Alexa 647 anti-Fc rabbit antibody (Fig. 6D). In both cases, very clear membrane labeling could be observed at 4°C followed by internalization at 37°C with the HEK293 cells expressing human and murine C-Ter GFP ITM2A, but not N-Ter GFP ITM2A con rming the need for the GFP to be extracellular for recognition.
We generated analogous results with anti HA antibodies on HEK293 cells (Fig. 6E) where clear internalization could be observed when the label was in C-Ter leading to punctate labelling con rming that ITM2A was functional and able to internalize cargos such as antibodies into cells.
Similar results were obtained with anti GFP antibodies and mouse ITM2A GFP constructs in HEK293 cells expressing mouse ITM2A with C-Ter and N-Ter GFP demonstrating functional internalization of mouse ITM2A too (Fig. 7): Production of monoclonal ITM2A antibodies A campaign was launched in collaboration with Yumab and the COMPACT IMI consortium to generate mouse/human cross-reactive monoclonal antibodies. The genes of the murine and human ITM2A ECD were synthesized and fused to the human IgG1 Fc part on a mammalian antigen expression vector. After transient transfection of HEK293 cells, the ITM2A-Fc fusion proteins were expressed, secreted by the cells into the culture medium and puri ed. A total of four different antibody selections were performed using two antibody libraries consisting of human kappa or lambda antibodies, comprising together a diversity of more than 10 10 different antibody sequences.
From 2304 clones tested for binding activity on the murine and human ITM2A, 220 cross-reactive and 15 mouse speci c hits were identi ed. 19 cross reactive antibodies with unique sequences were selected, produced and puri ed as fully human IgG in HEK293 cells. Testing these antibodies on human ITM2A overexpressing HEK293 cells revealed low correlation between binding to the recombinant protein and cell binding. 4 IgG antibodies did exhibit potent cell binding. These antibodies could demonstrate binding and internalization analogously as shown previously. For instance, when HEK293 cells expressing wt, C-Ter or N-Ter GFP-ITM2A were exposed to G04, internalization occurred readily at 37°C (Fig. 8A).
Similarly Yumab A01, E02 et H07 and the control AF4876 displayed binding at 4°C with the cells. Upon washing and warming to 37°C they were internalized as evidenced by a punctiform labeling inside cells (Fig. 8B).

ITM2A expression in BEC's
With antibodies demonstrating binding and uptake in hand, our next step was to evaluate their capacity to perform transcytosis. Before we went on to evaluate our ITM2A antibodies in models of transcytosis we checked whether we could nd ITM2A expression in our BEC's.
First, we checked ITM2A mRNA level in brain endothelial primary cells and cell lines. RNA levels were rapidly lost upon culture of non-human primate brain endothelial cells (Fig. 9A, cortex shown). Similar downregulation after culture was also observed in mouse primary BEC's (Fig. 9B). In comparison, cluster of differentiation 31's (CD31) expression, a speci c marker of endothelial cells retained high expression throughout the culture. This downregulation most certainly also explains why no expression of itm2a could be found in bEnd.3 cell lines (Fig. 9B).
Likewise, we could not detect endogenous ITM2A protein in primary endothelial cells from brain cortex of mouse or rats by immunolabelling compared to mouse ITM2A overexpressing HEK293 cells which displayed nice immunolabeling with anti-mouse ITM2A antibodies (AF4876, 18306-1-AP and Yumab G04). (Fig. 10) Similarly, we could not detect endogenous ITM2A protein in hCMEC/D3 by immunolabelling compared to TFRC probably due to the downregulation of ITM2A in culture ( Fig. 11) As immunolabelling showed no signal for ITM2A, we checked the level of ITM2A by WB using commercially available polyclonal antibody AF4876 and as a control we evaluated the mouse ITM2A-GFP fusion in HEK293 cells using an anti GFP antibody. The protein was well detected at the expected molecular weight (MW) (30 kDa for ITM2A+ 27 kDa for GFP) in the HEK293 overexpressing cells but could not be detected in mouse, rat or monkey primary endothelial cells or astrocytes nor could it be detected in other cell lines (bEnd.3, Madin-Darby Canine Kidney (MDCK1), hCMEC/D3) (Fig. 12).
We checked mRNA expression at several mice ages but could not see any marked difference. Decent expression (itm2a Cycle Threshold (Ct) = 25 vs gapdh Ct = 21) could be counted throughout post birth (P1) to six weeks of age in the cortex of mice (Fig. 13).
However, when these brain fractions were analyzed by WB, no band at 30 kDa could be identi ed (Fig.   14).
Proteomic studies to quantify ITM2A To clarify whether ITM2A protein was or not present in the mouse, proteomic studies were engaged in mouse cortex and muscle of newborn (P1 and P2 stage) and adult mice, along with freshly prepared astrocytes and brain microvessels in comparison with HEK293 cells overexpressing ITM2A. By Liquid Chromatography/ Mass Spectrometry (LC/MS), six endogenous peptides belonging to C-Ter or N-Ter ITM2A were used for detection and quanti cation. The results are shown in Fig. 15. While ITM2A was well detected in newborn muscle samples and in brain microvessels, less ITM2A was detected in both newborn and adult cortex samples and in adult muscle sample and no ITM2A was detected in astrocytes.
Protein intensities were calculated by averaging peptides intensity values (Fig. 15). After quanti cation of ITM2A in each sample, it could be calculated that the newborn cortex and muscle, brain microvessels and HEK293 cells overexpressing mITM2A averaged 0,1; 1,7; 0,6 and 3,6 fmol of ITM2A per µg of total protein content respectively.
From these experiments, it was concluded that ITM2A could be quanti ed in mouse brain and muscle albeit to a much lower extent in adult muscle and in cortex from newborn and adult. The protein was higher in brain microvessels freshly isolated from adult wt mice than in cortex homogenate pointing to endothelial cell enrichment. Levels of TFRC in mouse brain microvessels were quanti ed and found around the same range (0.8 fmol/µg).
These results suggest that WB conditions might not have been sensitive enough to detect the levels of ITM2A present. This was veri ed by diluting HEK293 cell lysate overexpressing mouse ITM2A. From the above-determined levels of 3.6 fmol ITM2A/µg of protein we determined that our limit of detection using Western blotting (additional le 4) was 7.2 fmol of ITM2A, over 10-fold the levels found in brain microvessels.
Finally, ITM2A was quanti ed in human microvessels and found to be four times lower than TFRC (ITM2A 0.11 fmol/µg of total protein vs TFRC 0.4 fmol/µg of total protein which could suggest lower transcytosis e cacy (Fig. 16).

In vitro transcytosis in human primary BBB model
As the ITM2A protein could be quanti ed, we decided to study one of our anti-ITM2A antibodies in a transcytosis model. Because of the strong ITM2A downregulation upon culture, cell line derived models were excluded. Even primary models which require a minimal culture to access enough endothelial cells were ruled out as we showed that in our NHP primary model ITM2A was already strongly downregulated. Preparation of human primary endothelial cells suitable for transcytosis was di cult owing to the hurdle of obtaining very fresh post-mortem human brains. We decided to turn to a human primary BEC model prepared from freshly resected brain tissue from glioblastoma or epilepsy surgeries provided by Brainplotting.
Anti ITM2A YU93-G04 antibody was evaluated in the model. As seen on Fig. 17A, the antibody did not perform better than a control antibody. In this model, an anti-TFRC receptor antibody performed systematically better than the control antibody (Fig. 17B).
In vivo mouse brain exposure of ITM2A antibodies To nally conclude on the potential of this protein to enhance brain exposure of antibodies, two in vivo mouse experiments were conducted.
A single administration study in mice of the anti ITM2A monoclonal antibody (mAb) G04 vs a control anti-TNP mAb was performed and levels of the mAbs in brain and plasma after 5h were determined. The mice were perfused with PBS before brain collection. The level of the antibodies was quanti ed using an ELISA assay. There was a weak (>= 2-fold) increase in brain exposure of the anti-ITM2A mAb compared to the anti-TNP (control) antibody (Fig. 18).
In a second experiment, a naïve human antibody-phage library was enriched for antigen speci c antibodies against ITM2A. The antibody-phage output was ampli ed and puri ed by PEG/N antibody puri cation. The puri ed phage display library of >6 million phages (panning campaign PC084, Strategy S1-1-10, panning rounds 3, diversity 6.3 10E6, titer ~1.10E13 cfu/ml) anti ITM2A ScFv's was used for in vivo panning and injected in mice. Brain were isolated after 1h and 24h and the brain homogenates were used to perform infection of E.coli. Picking of ~800 clones, screening of scFv supernatant in ELISA and on cells did not allow to identify any ITM2A phages in the brain.

Discussion
Endothelial cells form biological barriers that regulate exchanges and maintain a low and selective permeability to uid and solutes under normal physiological conditions. Understanding their speci c or enriched membrane proteins has been critical to facilitate drug delivery to speci c organs. Within these organs, brain is indeed the most highly protected tissue. The blood brain barrier with its network of tight junctions, e ux pumps and speci c metabolic systems represents a huge challenge for xenobiotics, drugs and especially large and polar biomolecules (1,2). Several strategies are actively pursued to enhance brain exposure of biotherapeutics, one the most successful is certainly making use of an endogenous transcytotic receptor located on the BBB such as transferrin or insulin receptor(4). However, the mechanisms so far identi ed for brain enhancement have been mostly ubiquitously expressed leading to exposure in other tissues than brain which could potentially lead to pleiotropic and adverse effects. Identi cation of brain-speci c mechanisms remains the ultimate unreached goal and is the active focus of current research in this area.
Two main work ows have been reported for the search of new mechanisms of brain delivery: On one hand transcriptomic and proteomic approaches from either brain microvessels or endothelials cells of human(38), cynomolgus monkey(39), bovine (40), rat (31,41) or mouse (12,(42)(43)(44)(45), including human(32, 46-48) diseased brains. On the other hand, phenotypic in vitro or in vivo screening of antibody's and peptide libraries displayed in various formats including phage and yeast (49)(50)(51). Only a few of them have delivered new brain delivery targets. Proteomics of rodent BEC's have led to CD98 heavy chain (a solute carrier) and Basigin (a matrix metalloprotease) along with known Lrp1 and InsR. Phenotypic panning of naïve lama single-domain antibody phage display for binding and internalizing in primary human BEC versus primary human lung endothelial cells led to FC5 and FC44 (50). It was later shown that FC5 binds to Cdc50A (energy-dependent clathrin endocytosis) (52). Our approach is combining both strategies.
Using proteomic, MicroArray and RNA sequencing approaches, the COMPACT IMI consortium identi ed candidate genes with high enrichment for brain, liver or lung of human primary endothelial cells (53).
These identi ed proteins could have potential in understanding biological differences among these barriers and developing drugs to target speci c organ. Analysis of the Next Generation Sequencing (NGS) data from human brain, liver and lung endothelial cells, selection of the genes with the highest expression in brain, differential expression versus peripheral tissues, annotation of human tissue, cell type and membrane localization using several public databases, led to few genes, some of them previously reported as brain delivery receptors such as LRP8 (54) or Basigin(12) (54). Among them, ITM2A stood out with the highest differential expression. This protein has been reported as brain endothelial speci c and identi ed from other omics efforts on rat(31) (32) and pig(33) BEC's. However, the function of this protein remains largely unknown, and it was never reported as a brain transporter.
To validate a putative transport mechanism for ITM2A along with potential for delivering drugs to the brain, we rst developed cells overexpressing the protein with the aim to validate the protein membrane localization and look at binding and uptake of anti-ITM2A antibodies. At the start of this effort, we did not have access to monoclonal anti-ITM2A antibodies. Several anti-ITM2A antibodies were reported (15,27) or commercially available but mostly polyclonal. We reasoned that engineering the protein with GFP and HA tags C-Ter (extracellular domain) position of the protein, could serve the double purpose of visualizing its cellular location using GFP uorescence and allow the study of binding and uptake with antibodies against these tags. This strategy of tagging a protein to circumvent the absence of monoclonal antibodies has actually been reported for ITM2A for deciphering its role in hedgehog signalling pathway (55). Tags at the N-Ter were also engineered as controls along with several positions within the extracellular Brichos domain which could later bring information on the precise site for endocytosis.
Exposure of the cells to anti-GFP antibodies rst allowed to con rm ITM2A membrane localization. Cellular localization of ITM2A has been looked at in a few cell systems and shown on the plasmic membrane along with large cytoplasmic vesicles, possibly endosomes and the Golgi apparatus (15). In particular, cytosolic localization has been reported in HEK293 cells overexpressing ITM2A and that the protein could be found colocalized with LAMP1 in lysosomes (27). We have con rmed the presence of ITM2A in Golgi but not in lysosomes by colocalization experiments using GFP tags. In addition to binding, HEK293 cells demonstrated nice uptake of anti GFP or HA antibodies respectively. These binding and uptake were speci c of cells overexpressing ITM2A bearing extracellular tags. The cells overexpressing ITM2A bearing intracellular tags did not lead to binding or uptake with these antibodies con rming that this uptake was speci cally linked to extracellular binding to ITM2A. Internalization of receptors genetically engineered with extracellular tags such as HA, cMyc, EGFP, have been documented in the literature with some G-coupled receptors (56), Transforming Growth Factor β(57) or erythropoietin(58) receptors. However, this uptake is far from systematic and many antibodies do not internalize upon binding their antigen receptors as was shown for instance by Jacobsen et al with an anti-Myc antibody and myc-engineered GPR6 and β2-adrenergic receptors (59). The speci c uptake of these antibodies upon binding to the extracellular part of the protein was interpreted as a positive signal and gave us the second go for our validation owchart. Monoclonal anti ITM2A antibodies were later designed and generated using the extracellular domain of ITM2A as antigen and the resulting antibodies con rmed the uptake seen previously.
Our next objective was to demonstrate that this uptake could lead to transcytosis in endothelial polarized cells. For this we needed a transcytosis model expressing the protein. It could be rodent or human as our anti ITM2A antibodies were cross reactive. From the COMPACT IMI consortium studies, it had been shown that the protein could no longer be found after an additional cell passage. We showed that ITM2A expression is strongly downregulated upon culture of either non-human primate (34) or mouse primary BEC's ( Fig. 9) as is the case for several BBB genes after cell line establishment or culture (60, 61) and has already been shown speci cally with ITM2A (33).
Most expression studies reported in the literature about ITM2A are transcript-based. In human, a good amount of the ITM2A transcript can be found in the brain as detailed in databases such as Gtex, Stanford, GenCard or open target platform (Fig. 1). In addition, according to Zhang et al, in humans, the amount of ITM2A mRNA in endothelial cells was evaluated at 150 Fragments Per Kilobase Million (FPKM) versus 23 FPKM for TFRC a well-known transcytotic receptor while in total brain the two proteins accounted for 2 FPKM. In mice, the amount of ITM2A mRNA was evaluated at 2000 FPKM in endothelial cells against 800 FPKM for TFRC and respectively 80 and 60 FPKM in the total brain for the two proteins (28).
Our own RNA seq study in non-human primate brain microvessels had shown intermediate expression of ITM2A in cortex, hippocampus and septum (FPKM 9-16) while very low expression in liver (34). By comparison in the same study, TFRC had shown a high expression (FPKM 75-125) in brain structures (34). Mitsui (25) reported that ITM2A protein was strongly detected in the lysates of mouse cerebral cortex between P0 and P10, and gradually decreased towards adulthood. Our own experiments on mice from P0 to 6 weeks of age showed that their ITM2A mRNA content remains constant throughout age. However, ITM2A protein could not be detected by WB in any of the samples, nor could it be detected in mouse, rat or monkey primary endothelial cells or astrocytes or in other cell lines (bEnd.3, MDCK1, hCMEC/D3) as opposed to the HEK293 cells overexpressing ITM2A where a strong band could be seen (Fig. 12). The theoretical MW of ITM2A is of ~30 kDa (Gencards or Proteintech) and it was reported that post translational modi cations lead to an apparent MW of 43 and 45 kDa probably resulting from Nglycosylation at amino acid position 166 (15). Nevertheless, no protein could be detected around this MW either. Fluorescence gave the same results with strong signal for the HEK293 cells and no signal for endothelial rat or mouse cells or hCMEC/D3 cells con rming results from Masuda et al(62).
Using more sensitive proteomics, we were able to quantitate levels of ITM2A in mouse brain microvessels. These levels have been con rmed to be under the limit of detection of our WB conditions.
Nevertheless, as these levels were in the same range as the ones found for TFRC in the same study, we considered that it was worth engaging into a transcytosis experiment.
We decided to evaluate one of our monoclonal anti ITM2A antibody in a model based on primary cultures of human BMEC from Brainplotting (35). These cells are prepared from fresh brains derived from surgical resections after very short time culture which gives them a better chance to keep their phenotype. In this model, a TFRC antibody reproducibly shows enhanced apparent permeability vs a control antibody. However, when the ITM2A antibody was evaluated in the same conditions, no difference in permeability vs the control could be evidenced (Papp ITM2A 0,97.10-6 vs control 1,02.10 -6 cm.min-1). As we had no information regarding the status of ITM2A levels in this model, we could not clearly conclude on this experiment.
The area of predictivity of in vitro blood brain barrier models is still the matter of intense research and debate. Even if some antibodies brain exposures have been linked to their apparent permeabilities in in vitro transcytosis models(63, 64), this was shown in cases where in vitro and in vivo experiments were performed in the same species as described by Stanimirovic et al in rat(63). In vivo brain exposure, distribution and pharmacokinetics are dependent on a series of dynamic processes, linked also to target engagement, localization and cellular tra cking. All these would be di cult to recapitulate in an in vitro model even more so for species-to-species predictions where additional parameters such as anatomy, capillary bed density, molecular composition, as well as the density of speci c BBB transporters(63) need to be taken into account.
To conclude about the potential of ITM2A to transport antibodies to the brain, we engaged one of our speci c anti ITM2A antibodies directly in vivo in mice to quantify its brain exposure. When injected at 5 mg/Kg YU93-G04 could be quantitated in brain parenchyma 5h after injection with a 2-fold higher level than a control antibody. The brain/plasma ratio were not very different though for ITM2A and control antibodies (around 0.3%). This ratio was in the range of what is described in the literature regarding brain /plasma ratio for antibodies with no modi cation to enter the brain : 0.1% in the rat(65, 66) and 0.01% in primates(65-67) .To further evaluate if this modest brain exposure increase observed was mechanismrelated and monitor an early time after injection (1h), we performed in vivo panning of a library of anti ITM2A antibodies. A naïve human antibody-phage library was enriched for ITM2A speci c antibodies against recombinant protein ITM2A and the antibody-phage output was ampli ed and puri ed by PEG/NaCl puri cations before injection to a single mouse. The brain was harvested at 1 and 24h after injection and the homogenates were used for infection of E. coli. From this no hit was identi ed suggesting that none of them was able to speci cally reach the brain parenchyma. Phages are huge entities, and their barrier crossing might be more di cult than isolated antibodies. In addition, our anti ITM2A antibody could be trapped in the vessels or recycling(68). Alternatively, the epitope recognized by the antibody we selected for in vivo study might not be the one leading to transcytosis. To nally conclude about the fate of ITM2A antibodies after in vivo injection and their potential to enhance brain exposure, several antibodies recognizing distinct epitopes should be compared, and the antibodies quantitated in both parenchyma and vessels. At this stage, we considered that the enhancement obtained was not at the level that could be of interest for potential application to a therapeutic project.

Conclusions
Our work combines both transcriptomic pro ling leading to selection of ITM2A as a potential brain speci c target, and in vivo phage panning of an anti ITM2A phage library. Our approach illustrates the complexity of such endeavor. Not even mentioning the technical challenge of getting access to pure human primary endothelial cells, highly expressed targets are often down regulated upon culture making it di cult to study them in functional cellular models, cross-reactive monoclonal antibodies are necessary for validation in rodent models. In addition, targets might have different function in rodent and human although we have no indication that this could be the case for ITM2A. ITM2A might remain a valid target for human brain enhancement but its validation might prove quite complex. Availability of data and materials Human or mouse GFP tagged ITM2A expression in HEK293 cells. HEK293 cells are stably recombined with human or mouse ITM2A with GFP tag at different positions. GFP tag ITM2A is visualized in green, and nuclei are stained in blue by Hoechst.    Immuno uorescence staining of hCMEC/D3 with TFRC and ITM2A antibodies. Labelling is visualized with secondary antibodies anti human Alexa 647 in red. Nuclei are labelled with Hoechst in blue.