Affibody Molecules Intended for Receptor-Mediated Transcytosis via the Transferrin Receptor

The development of biologics for diseases affecting the central nervous system has been less successful compared to other disease areas, in part due to the challenge of delivering drugs to the brain. The most well-investigated and successful strategy for increasing brain uptake of biological drugs is using receptor-mediated transcytosis over the blood–brain barrier and, in particular, targeting the transferrin receptor-1 (TfR). Here, affibody molecules are selected for TfR using phage display technology. The two most interesting candidates demonstrated binding to human TfR, cross-reactivity to the murine orthologue, non-competitive binding with human transferrin, and binding to TfR-expressing brain endothelial cell lines. Single amino acid mutagenesis of the affibody molecules revealed the binding contribution of individual residues and was used to develop second-generation variants with improved properties. The second-generation variants were further analyzed and showed an ability for transcytosis in an in vitro transwell assay. The new TfR-specific affibody molecules have the potential for the development of small brain shuttles for increasing the uptake of various compounds to the central nervous system and thus warrant further investigations.


Introduction
Antibodies and other biological drugs have had a profound impact on the treatment of many different diseases, such as different forms of cancers and auto-immune disorders [1,2]. However, for diseases of the central nervous system (CNS), the restricted uptake over the blood-brain barrier (BBB) [3,4] makes development of therapeutics and diagnostic tools more challenging.
The BBB protects the sensitive neuronal environment by substantially decreasing the paracellular diffusion of substances from blood. Various receptors and transporters such as the transferrin receptor-1, the insulin-like growth factor 1 receptor, the L-type amino acid transporter, and the glucose transporter isoform 1 control the transcytosis of molecules in and out between the brain and blood compartments [3,5,6]. Transfer can either be by adsorption [7][8][9][10], passive or active transportation [5], or receptor-mediated transcytosis (RMT) [5]. It has been shown that about 0.1-0.2% of systemically administered antibodies pass into the CSF [11], which is potentially too low to achieve efficient disease-modifying effects [12]. Therefore, efforts have been made to find a transportation system across the BBB for biological drugs [5,[13][14][15][16][17]. One of the more well-studied approaches is RMT via the transferrin receptor-1 (TfR), which naturally transports transferrin (Tf) across the BBB to the neuronal environment [7,18,19]. An example is the anti-murine TfR-specific 8D3 shuttle antibody [20], which has shown up to 80-fold increased uptake to the brain in rodent models [21]. For antibodies targeting TfR, several parameters have to be considered, such as affinity [22], valency [21,23], and antibody effector functions to minimize potential adverse effects [24]. (B) Z library sequence with randomized positions marked with "X" aligned to sequences for ZTfR#14 and ZTfR#18.

Production of His6-Tagged and ABD-Fused Affibody Molecules
Nineteen different affibody molecules from the selections ( Figure S2) were produced in both a His6-ZTfR-ABD and a ZTfR-His6 format. Affibody molecules can be fused to an albumin-binding domain (ABD) to prolong their half-life in circulation by interaction with human serum albumin (HSA) [36,41,42]. The production indicated a higher yield from the smaller format compared to the ABD-fusions (Tables S1 and S2). Molecular weight was verified by MS and correlated well with expected values (Tables S1 and S2). The purity was verified by SDS-PAGE ( Figures S3 and S4), and all constructs showed the expected alpha-helical secondary structure content as measured using circular dichroism (CD) spectroscopy ( Figure S5).

Flow Cytometry Analysis of ZTfR Candidates Aimed to Bind to Murine and Human Cells
The affibody candidates in ZTfR-ABD format were analyzed using flow cytometry for binding to human SK-OV-3 and murine brain endothelial cells (bEnd.3) expressing human and mouse TfR ( Figure 2). Two of the affibody molecules (ZTfR#14 and ZTfR#18) demonstrated binding to both human SK-OV-3 and murine bEnd.3 ( Figure 2, Table S3). The two binders originated from the selection track with alternating mTfR and hTfR as targets ( Figure S1). ZTfR#14 and ZTfR#18 showed low sequence homology, which is typically an indication of non-overlapping epitopes ( Figure 1B). (B) Z library sequence with randomized positions marked with "X" aligned to sequences for Z TfR#14 and Z TfR#18 .

Production of His 6 -Tagged and ABD-Fused Affibody Molecules
Nineteen different affibody molecules from the selections ( Figure S2) were produced in both a His 6 -Z TfR -ABD and a Z TfR -His 6 format. Affibody molecules can be fused to an albumin-binding domain (ABD) to prolong their half-life in circulation by interaction with human serum albumin (HSA) [36,41,42]. The production indicated a higher yield from the smaller format compared to the ABD-fusions (Tables S1 and S2). Molecular weight was verified by MS and correlated well with expected values (Tables S1 and S2). The purity was verified by SDS-PAGE ( Figures S3 and S4), and all constructs showed the expected alphahelical secondary structure content as measured using circular dichroism (CD) spectroscopy ( Figure S5).

Flow Cytometry Analysis of Z TfR Candidates Aimed to Bind to Murine and Human Cells
The affibody candidates in Z TfR -ABD format were analyzed using flow cytometry for binding to human SK-OV-3 and murine brain endothelial cells (bEnd.3) expressing human and mouse TfR ( Figure 2). Two of the affibody molecules (Z TfR#14 and Z TfR#18 ) demonstrated binding to both human SK-OV-3 and murine bEnd.3 ( Figure 2, Table S3). The two binders originated from the selection track with alternating mTfR and hTfR as targets ( Figure S1). Z TfR#14 and Z TfR#18 showed low sequence homology, which is typically an indication of non-overlapping epitopes ( Figure 1B).  The first bar in both groups is negative control with cells incubated only with a secondary reagent (*H). The dashed line represents the signal from the negative control affibody molecule (Ztaq-ABD) (*C). A construct scFv8D3-ZSYM73-ABD (*P) with an scFv fragment of the murine TfR-specific antibody 8D3 is included as a positive control. Values are given as mean ± s.d. based on n = 3 samples.
Next, TfR-positive human SK-OV-3 and brain endothelial hCMEC/D3 cells were incubated with a dilution series of ZTfR#14 and ZTfR#18, and binding was analyzed by flow cytometry, showing concentration-dependent signal ( Figures S6 and 3). The bar chart shows the MFI (mean fluorescent intensity) of TfR-binding, and signals are normalized to blank cells for respective cell lines. The first bar in both groups is negative control with cells incubated only with a secondary reagent (*H). The dashed line represents the signal from the negative control affibody molecule (Ztaq-ABD) (*C). A construct scFv8D3-Z SYM73 -ABD (*P) with an scFv fragment of the murine TfR-specific antibody 8D3 is included as a positive control. Values are given as mean ± s.d. based on n = 3 samples.
Next, TfR-positive human SK-OV-3 and brain endothelial hCMEC/D3 cells were incubated with a dilution series of Z TfR#14 and Z TfR#18, and binding was analyzed by flow cytometry, showing concentration-dependent signal ( Figure S6 and Figure 3). Cell-specific binding was investigated by co-incubating the His6-ZTfR-ABD construct with ZTfR-His6 at different molar excess ratios, showing a substantial decrease in signal in comparison to no blocking (Table S4). Moreover, flow cytometry was also used for epitope binning by co-incubating the His6-ZTfR#14-ABD construct with ZTfR#18-His6 and vice versa, showing a negligible decrease in binding and thus confirming non-overlapping epitopes (Table S4).
The high concentration of Tf in blood and results from previous studies demonstrating negative effects on reticulocyte count for transferrin-blocking agents in animal models [29] suggest that competitive binding with transferrin (Tf) is probably not optimal. Another epitope-binning experiment was thus performed, where the ZTfR candidates #2, #4, #10, #14, and #18 in ZTfR-ABD format were pre-incubated with fluorescently labeled transferrin (Tf-488) prior to analysis of cell-binding. For clones #2, #4, and #10, the Tf-488 signal decreased after co-incubation, whereas no shift in signal was observed for co-incubation with ZTfR#14 and ZTfR#18 ( Figure S7). Furthermore, the signal from ZTfR#14 and ZTfR#18 was maintained at a 5-fold molar excess of Tf ( Figure 4). A non-TfR binding control affibody (Ztaq) [43] was included in the experiment, showing no effect on TfR binding (Table S5) and supports the indication that the TfR epitope is not shared between the ZTfR#14, ZTfR#18, and Tf. Cell-specific binding was investigated by co-incubating the His 6 -Z TfR -ABD construct with Z TfR -His 6 at different molar excess ratios, showing a substantial decrease in signal in comparison to no blocking (Table S4). Moreover, flow cytometry was also used for epitope binning by co-incubating the His 6 -Z TfR#14 -ABD construct with Z TfR#18 -His 6 and vice versa, showing a negligible decrease in binding and thus confirming non-overlapping epitopes ( Table S4).
The high concentration of Tf in blood and results from previous studies demonstrating negative effects on reticulocyte count for transferrin-blocking agents in animal models [29] suggest that competitive binding with transferrin (Tf) is probably not optimal. Another epitope-binning experiment was thus performed, where the Z TfR candidates #2, #4, #10, #14, and #18 in Z TfR -ABD format were pre-incubated with fluorescently labeled transferrin (Tf-488) prior to analysis of cell-binding. For clones #2, #4, and #10, the Tf-488 signal decreased after co-incubation, whereas no shift in signal was observed for co-incubation with Z TfR#14 and Z TfR#18 ( Figure S7). Furthermore, the signal from Z TfR#14 and Z TfR#18 was maintained at a 5-fold molar excess of Tf ( Figure 4). A non-TfR binding control affibody (Ztaq) [43] was included in the experiment, showing no effect on TfR binding (Table S5) and supports the indication that the TfR epitope is not shared between the Z TfR#14 , Z TfR#18 , and Tf.

pH-Dependent Binding to TfR Expressing Cells
During transcytosis across the brain endothelial cells in the BBB, the pH of the endosomes containing TfR decreases to around 5.5 [18,19]. Thus, the dissociation of the affibodies from cells at different pH was investigated by employing a flow-cytometric assay similar to previously described studies [44] ( Figure 5A). The analysis was performed at 4 °C to decrease internalization, and fluorescently labeled Tf (Tf-488) was included for comparison ( Figure 5B). After incubation of binders with SK-OV-3 cells, cells were washed with buffers of different pH, followed by an analysis of cell-binding using flow cytometry. The pH-dependent binding of Tf-TfR was verified, shown as a decrease in signal for pH below 6.0 ( Figure 5B). Interestingly, ZTfR#14 showed a pH-independent profile, and ZTfR#18 had opposite dependence compared to Tf, where the dissociation appeared slower at lower pH ( Figure 5B). CD spectroscopy was used to verify secondary structure content, thermal stability, and refolding in the pH range (5.5-7.4) ( Figure S8). The spectra measured at different pH were overlapping, indicating that the structure was similar and independent of pH in the given range ( Figure 5C,D). The thermal melting point (Tm) was estimated to be 56 °C for ZTfR#14 and 58 °C for ZTfR#18, and spectra before and after the variable temperature measurements were overlapping, indicating full refolding ( Figure S9).

pH-Dependent Binding to TfR Expressing Cells
During transcytosis across the brain endothelial cells in the BBB, the pH of the endosomes containing TfR decreases to around 5.5 [18,19]. Thus, the dissociation of the affibodies from cells at different pH was investigated by employing a flow-cytometric assay similar to previously described studies [44] ( Figure 5A). The analysis was performed at 4 • C to decrease internalization, and fluorescently labeled Tf (Tf-488) was included for comparison ( Figure 5B). After incubation of binders with SK-OV-3 cells, cells were washed with buffers of different pH, followed by an analysis of cell-binding using flow cytometry. The pH-dependent binding of Tf-TfR was verified, shown as a decrease in signal for pH below 6.0 ( Figure 5B). Interestingly, Z TfR#14 showed a pH-independent profile, and Z TfR#18 had opposite dependence compared to Tf, where the dissociation appeared slower at lower pH ( Figure 5B). CD spectroscopy was used to verify secondary structure content, thermal stability, and refolding in the pH range (5.5-7.4) ( Figure S8). The spectra measured at different pH were overlapping, indicating that the structure was similar and independent of pH in the given range ( Figure 5C,D). The thermal melting point (T m ) was estimated to be 56 • C for Z TfR#14 and 58 • C for Z TfR#18, and spectra before and after the variable temperature measurements were overlapping, indicating full refolding ( Figure S9).

Surface Plasmon Resonance Analysis of Binding between ZTfR and TfR
The affinity of the ZTfR#14-ABD and ZTfR#18-ABD to recombinant TfR was estimated by surface plasmon resonance (SPR) using an immobilized extracellular domain of human and murine his-tagged TfR. Both affibody molecules showed a higher affinity for murine TfR with a KD of 90 nM for ZTfR#14 and 170 nM for ZTfR#18. For human TfR, the KD was 150 nM for ZTfR#14 and 710 nM for ZTfR#18 ( Figure S9 and Table S6).

Single Amino Acid Mutagenesis of ZTfR#14 and ZTfR#18
Single amino acid mutagenesis was performed on ZTfR#14 and ZTfR#18 to investigate the contribution of individual residues in binding to TfR. The previously 14 randomized

Surface Plasmon Resonance Analysis of Binding between Z TfR and TfR
The affinity of the Z TfR#14 -ABD and Z TfR#18 -ABD to recombinant TfR was estimated by surface plasmon resonance (SPR) using an immobilized extracellular domain of human and murine his-tagged TfR. Both affibody molecules showed a higher affinity for murine TfR with a K D of 90 nM for Z TfR#14 and 170 nM for Z TfR#18 . For human TfR, the K D was 150 nM for Z TfR#14 and 710 nM for Z TfR#18 ( Figure S9 and Table S6).

Single Amino Acid Mutagenesis of Z TfR#14 and Z TfR#18
Single amino acid mutagenesis was performed on Z TfR#14 and Z TfR#18 to investigate the contribution of individual residues in binding to TfR. The previously 14 randomized positions of respective affibodies were mutated to either histidine or the wildtype amino acid from the original IgG-binding Z domain [45]. The genes for the mutated affibody molecules were subcloned in fusion to a gene encoding an engineered albumin-binding domain (ABD) [41] into an expression vector for Adhesin Involved in Diffuse Adherence (AIDA) 1-mediated display on the surface of E. coli [34]. E. coli expressing the different mutants on the surface were analyzed by flow cytometry for assessment of TfR-binding, and the original binders (Z TfR#14 and Z TfR#18 ) were included for comparison. In addition to the analysis of TfR-binding, cells were also incubated with saturating concentrations of fluorescently labeled HSA to assess surface expression levels.
For both affibody molecules, most mutations had a negative impact on the binding to TfR ( Figure 6). However, two mutations with improved binding to TfR, for respective binder and with the introduction of histidine, were found, corresponding to A9H and L27H for Z TfR#14, and M14H and I11N for Z TfR#18 , although M14H seemed to have a negative effect on the cell surface expression (Figures 6 and S10). positions of respective affibodies were mutated to either histidine or the wildtype amino acid from the original IgG-binding Z domain [45]. The genes for the mutated affibody molecules were subcloned in fusion to a gene encoding an engineered albumin-binding domain (ABD) [41] into an expression vector for Adhesin Involved in Diffuse Adherence (AIDA) 1-mediated display on the surface of E. coli [34]. E. coli expressing the different mutants on the surface were analyzed by flow cytometry for assessment of TfR-binding, and the original binders (ZTfR#14 and ZTfR#18) were included for comparison. In addition to the analysis of TfR-binding, cells were also incubated with saturating concentrations of fluorescently labeled HSA to assess surface expression levels.
For both affibody molecules, most mutations had a negative impact on the binding to TfR ( Figure 6). However, two mutations with improved binding to TfR, for respective binder and with the introduction of histidine, were found, corresponding to A9H and L27H for ZTfR#14, and M14H and I11N for ZTfR#18, although M14H seemed to have a negative effect on the cell surface expression (Figures 6 and S10).

Characterization of Second-Generation Affibody Molecules Binding TfR
The four mutants (ZTfR#14_A9H, ZTfR#14_L27H, ZTfR#18_I11N, and ZTfR#18_M14H) that showed improved binding in the mutagenesis study were subcloned for expression of soluble affibody molecules in a (HE)3-ZTfR-cys format. In addition, two double mutants were generated and included in the study, corresponding to ZTfR#14_A9H_L27H and ZTfR#18_I11N_M14H, as well as a control affibody with an affinity for the HER2 receptor [46] (Table S7, Figure S11). After purification, the C-terminal cysteine in the proteins was conjugated with biotin and FITC followed by verification of secondary structure content, molecular mass, thermal stability, and binding to TfR-positive cells ( Figure S12, Tables S7 and S8).
The secondary structure content of the six mutants was similar to ZTfR#14 and ZTfR#18, and thermal stability was either improved or similar (Table 1). However, ZTfR#14_L27H and ZTfR#18_I11N showed non-complete refolding after heat-induced denaturation. All six mutants demonstrated binding to both SK-OV-3 and brain endothelial bEnd.3 cells and the

Characterization of Second-Generation Affibody Molecules Binding TfR
The four mutants (Z TfR#14_A9H , Z TfR#14_L27H , Z TfR#18_I11N , and Z TfR#18_M14H ) that showed improved binding in the mutagenesis study were subcloned for expression of soluble affibody molecules in a (HE) 3 -Z TfR -cys format. In addition, two double mutants were generated and included in the study, corresponding to Z TfR#14_A9H_L27H and Z TfR#18_I11N_M14H , as well as a control affibody with an affinity for the HER2 receptor [46] (Table S7, Figure S11). After purification, the C-terminal cysteine in the proteins was conjugated with biotin and FITC followed by verification of secondary structure content, molecular mass, thermal stability, and binding to TfR-positive cells ( Figure S12, Tables S7 and S8). The secondary structure content of the six mutants was similar to Z TfR#14 and Z TfR#18, and thermal stability was either improved or similar (Table 1). However, Z TfR#14_L27H and Z TfR#18_I11N showed non-complete refolding after heat-induced denaturation. All six mutants demonstrated binding to both SK-OV-3 and brain endothelial bEnd.3 cells and the highest signals were observed for Z TfR#14_L27H and Z TfR#18_M14H ( Figure S12). The pHdependency of cell binding was similar for the mutants compared to Z TfR#14 and Z TfR#18 ( Figure S13).

Flow Cytometric Endocytosis Assay
To study the potential endocytosis of the affibody constructs, a previously described flow cytometry-based assay was used [47]. In the assay, the pH sensitivity of FITC is exploited. One emission maximum (FL1 525 nm) of FITC is pH-dependent, whereas the other maximum (FL2 575 nm) is not. Dextran-FITC was added to bEnd.3 cells and incubated at different pH, followed by flow-cytometric analysis of the fluorescence in FL1 and FL2. The pH was plotted against the FL1/FL2 ratio for a standard curve ( Figure S14). The FITC-labeled affibody molecules were thereafter incubated with bEnd.3 cells, followed by washing with acidic buffer (pH 5.0) and neutral (pH 7.0) to release remaining membranebound affibodies. After neutralization with buffer (pH 7.4), FL1 and FL2 fluorescence were analyzed using flow cytometry.
The TfR-specific BBB shuttle antibody 8D3 has previously been analyzed using the assay, revealing a pH of around 6, which corresponds to recycling TfR endosomes [18,19,47]. The obtained FL1/FL2 ratios for the affibodies corresponded to a pH of around 5-6 for Z TfR#18 -derived variants and to a pH of around 6-7 for the Z TfR#14 -derived variants ( Table 2).

Transcytosis across a Murine In Vitro BBB Model
To study potential transcytosis across the BBB, a previously described in vitro assay was employed [47]. The assay is based on transwells with recombinant spider silk nanomembranes supporting confluent monolayers of brain endothelial cells. The FITClabeled affibody molecules were added to the apical side of bEnd.3 cell membranes together with a fluorescently labeled antibody (IgG-AF647) that is used as an internal control for the barrier integrity of the membranes. The HER2-specific affibody (Z HER2 ) was included as a control. After 90 min, the media on both sides were collected, and the AF647 and FITC fluorescence were measured using a fluorescence spectrophotometer. The fluorescence was finally used to calculate the apparent permeability (p app ) [47].
All six TfR-specific affibody molecules except for Z TfR#14 _ A9H showed higher apparent permeability than the internal control antibody (Table S9, Figure 7). Three of the variants also showed higher permeability compared to Z HER2 , indicating transcytosis over TfRpositive bEnd.3 cells. fluorescence were measured using a fluorescence spectrophotometer. The fluorescence was finally used to calculate the apparent permeability (papp) [47]. All six TfR-specific affibody molecules except for ZTfR#14_A9H showed higher apparent permeability than the internal control antibody (Table S9, Figure 7). Three of the variants also showed higher permeability compared to ZHER2, indicating transcytosis over TfR-positive bEnd.3 cells.

Discussion
The limited efficacy of many CNS-targeted biological drugs could potentially be improved by increased transportation across the BBB. Utilizing small alternative scaffolds (e.g., affibodies) as brain shuttles would potentially have an advantage over antibodies due to the faster biodistribution in the brain parenchyma after transcytosis. The smaller mass of the complex might additionally decrease potential systemic toxic effects from long serum half-life in the periphery [31,48]. RMT via the transferrin receptor-1 (TfR), which shuttles transferrin (Tf) and iron over the BBB to the brain [7,18,19], is one of the more well-investigated strategies. However, targeting the TfR receptor is still problematic as the receptor is abundantly present in other organs and cells in the body, resulting in a sink effect and potential toxicity issues, which have been observed in previous studies [40].
Herein, we selected TfR-specific affibody molecules using phage display technology. Nineteen candidates were selected from the output, and the two most promising binders were further characterized in vitro. Affinity, binding kinetics, epitope, valency, and pHdependent binding are examples of properties that have been suggested in the literature to be important for RMT via TfR [23,29,49,50]. Moreover, cross-reactivity to TfR orthologues from model animals would facilitate future preclinical evaluation and clinical translation. When analyzing the two top candidates in flow cytometry on TfR-positive SKOV-3 cells, it was noted that the cell-binding signal decreased when labeling and analysis were performed at room temperature (data not shown), perhaps indicating internalization of the binders prior to the addition of secondary detection reagents.
The two candidates, ZTfR#14 and ZTfR#18, were cross reactive for murine and human TfR and bound to an epitope that was non-overlapping with the site for Tf. Avoiding the binding site of Tf is probably important as the high concentration of Tf in blood would otherwise greatly limit available receptors for RMT. Furthermore, blocking the interaction between Tf and TfR has previously been connected to toxicity in mice [29]. Using multi-color

Discussion
The limited efficacy of many CNS-targeted biological drugs could potentially be improved by increased transportation across the BBB. Utilizing small alternative scaffolds (e.g., affibodies) as brain shuttles would potentially have an advantage over antibodies due to the faster biodistribution in the brain parenchyma after transcytosis. The smaller mass of the complex might additionally decrease potential systemic toxic effects from long serum half-life in the periphery [31,48]. RMT via the transferrin receptor-1 (TfR), which shuttles transferrin (Tf) and iron over the BBB to the brain [7,18,19], is one of the more well-investigated strategies. However, targeting the TfR receptor is still problematic as the receptor is abundantly present in other organs and cells in the body, resulting in a sink effect and potential toxicity issues, which have been observed in previous studies [40].
Herein, we selected TfR-specific affibody molecules using phage display technology. Nineteen candidates were selected from the output, and the two most promising binders were further characterized in vitro. Affinity, binding kinetics, epitope, valency, and pHdependent binding are examples of properties that have been suggested in the literature to be important for RMT via TfR [23,29,49,50]. Moreover, cross-reactivity to TfR orthologues from model animals would facilitate future preclinical evaluation and clinical translation. When analyzing the two top candidates in flow cytometry on TfR-positive SKOV-3 cells, it was noted that the cell-binding signal decreased when labeling and analysis were performed at room temperature (data not shown), perhaps indicating internalization of the binders prior to the addition of secondary detection reagents.
The two candidates, Z TfR#14 and Z TfR#18 , were cross reactive for murine and human TfR and bound to an epitope that was non-overlapping with the site for Tf. Avoiding the binding site of Tf is probably important as the high concentration of Tf in blood would otherwise greatly limit available receptors for RMT. Furthermore, blocking the interaction between Tf and TfR has previously been connected to toxicity in mice [29]. Using multi-color flow cytometry, we could show the simultaneous binding of Tf and respective affibody to cells at molar excess of either Tf or affibody. Thus, these affibodies will likely not be toxic due to decreased uptake of Tf.
It has been hypothesized and demonstrated for some TfR-specific antibodies in in vitro models that pH-dependent binding is important for transcytosis [49,51]. The idea is, in principle, to mimic the natural pH dependency in the range between pH 5.0 to 7.4, which is part of the iron transport mechanism via Tf and TfR. However, it should be noted that several TfR-specific brain shuttles without pH-dependent binding have been reported, and it is speculated that the importance of pH dependency is linked to TfR epitope and affinity, where both fast on rate and off rate are considered important [15,28,50,52]. The pHdependent binding can, however, be of importance since certain epitopes, in combination with bivalent binders, are believed to cross-link the receptor during endocytosis, directing it to lysosomal degradation. Still, release by pH or by fast off rates after endocytosis probably has the potential to improve transcytosis [49,51]. The two candidates, Z TfR#14 and Z TfR#18 , displayed different pH-dependent behavior, where Z TfR#14 was not affected by pH in the given range, and Z TfR#18 had a slower off rate from the receptor with lower pH.
Single amino acid mutagenesis was performed on the two candidates, where histidine or the Z wt [45] amino acid was incorporated in the 14 previously randomized positions.
The results from the mutagenesis study indicated the contribution of individual residues to the interaction with TfR. Moreover, a few mutants showed increased binding to TfR, and two single amino acid mutants for each affibody were analyzed further. The two mutations were also combined in double mutants to evaluate potential additive effects.
The new variants (Z TfR#14_A9H , Z TfR#14_L27H , Z TfR#14_A9H_L27H, Z TfR#18_I11N , Z TfR#18_M14H , and Z TfR#18I_11N_M14H ) were site-specifically conjugated to FITC and showed binding to both SK-OV-3 and brain endothelial bEnd.3 cells in flow cytometry. Internalization was analyzed using a fluorescence assay, exploiting the pH-dependency of FITC and indicating that the binders were located in a cell compartment with a pH of around 5-6 for the Z TfR#18derived variants and around 6-7 for the Z TfR#18 -derived variants. The results suggest that the Z TfR#18 -derived variants are more efficiently internalized, although a pH of around five might indicate that it is partly directed to lysosomes, which is undesired.
Finally, the six affibodies were studied with a method where nanofibrillar membranes of recombinant silk seeded with brain endothelial cells are used as a simplified BBB model to assess active transcytosis. The results showed higher apparent permeability for three of the TfR-binders compared to negative control. As epitope, affinity, and internalization behavior only give indications about the capability of transcytosis over BBB, this assay is important as a predictor for the in vivo functionality.
The results are promising, and further optimization of the binders is warranted before studies on brain uptake in animal models.

Protein Labeling
Biotinylation of recombinant human and mouse transferrin receptor-1 (TfR: Sino Biological Inc., Bejing, China) was performed using a Biotin-XX Microscale Protein Labeling Kit (Invitrogen, Waltham, MA, USA) according to supplier's recommendations. The concentration of the proteins was determined using absorbance at 280 nm.

Phage Display Selections
A combinatorial phage library of the Z domain with randomization in 14 positions was essentially prepared as described previously [53] (Figure 1). Selection and amplification were performed in phosphate-buffered saline with tween (PBST: 0.1% Tween-20) with bovine serum albumin (BSA: Saveen and Werner, Limhamn, Sweden) (PBSTB: 3 w/v% BSA) at room temperature, as described previously [35]. Prior to the first cycle of bio-panning, a negative selection of the library was conducted by incubating the library with streptavidincoated magnetic beads. The remaining phage particles were panned against recombinant TfR for 1 h in five cycles with an increasing number of washes over the cycles. Different strategies for panning of binders were implemented based on recombinant human TfR, recombinant murine TfR, or cross-selection strategies. In the tracks where the target was kept constant, the concentration of the receptor was decreased from 100 nM in the first panning cycle to 12.5 nM in cycle five. In the cross-selection strategies, human and mouse receptor targets were altered in the five panning cycles, with target concentration lowered from 100 nM in the first two cycles to 40 nM in the last cycle ( Figure S1). Selections were followed by DNA sequencing (Microsynth AG, Balgach, Switzerland) of 174 randomly picked colonies.

Production of Recombinant Affibody Molecules in E. coli
The affibody molecules selected for further characterization were produced with either a C-terminal hexahistidine (His 6 ) tag or an N-terminal hexahistidine (His 6 ) in combination with a C-terminal albumin binding domain (ABD) [41] for purification and characterization strategies. For His 6 -tagged proteins, the genetic sequences were subcloned into the pET26b(+) vector, introducing a C-terminal His 6 -tag and yielding the protein constructs [Z TfR ]-His 6 . For ABD-fused proteins, the affibody genes were cloned into a pT7 vector introducing a C-terminal ABD 035 molecule [41], yielding the final proteins To analyze pH-dependent binding, SK-OV-3 cells were treated and analyzed as above, except for the following differences. His 6 -[Z TfR ]-ABD and HSA-647 were preincubated for 20 min and then added to SK-OV-3 cells. After incubation, cells were washed in PBSB (1 w/v%) with different pHs in the range 5.0-7.4 for 30 min at 4 • C. All samples were analyzed in triplicate. Experimental set-up with inspiration from Neiveyans et al. [44].
To verify TfR-specific binding, a blocking experiment was performed by co-incubating the Z TfR -ABD construct with 5-10× molar excess of Z TfR -His 6 . The SK-OV-3 cells were treated and analyzed as above. To

Circular Dichroism Analysis for Secondary Structure
Circular dichroism (CD) spectroscopy was used to verify the secondary structure content of the affibody molecules in (HE) 3 -[Z TfR ]-bio or His 6 -[Z TfR ]-ABD format. CD spectra between 195-260 nm at 20 • C were recorded with a Chirascan system (Applied Photophysics, Leatherhead, UK) using a 1 mm High precision cell (110-1P-40 cuvettes, Hellma Analytics, Munich, Germany). Five scans were recorded for each protein sample at a concentration of 0.2 mg × mL −1 in PBS.
The thermal melting point for the affibody molecules was determined using a variable temperature measurement (VTM) at 221 nm with a temperature gradient of 5 • C or 1 • C per minute (depending on samples as indicated in results) with five readings at each temperature point. After cooling down to 20 • C, refolding was accessed by recording spectra and compared to the spectra before denaturation.

Biosensor Analysis of the Z TfR and TfR Interaction of Both Murine and Human TfR
The affinity and kinetics of the interaction between the His 6 -[Z TfR ]-ABD constructs and TfR of human and murine origin were analyzed using surface plasmon resonance (SPR) on a Biacore T200 system (Cytiva, Marlborough, MA, USA) at 25 • C with PBST (0.05% Tween-20) as running buffer. Approximately 1100 response units (RU) of hTfR-His 6 and mTfR-His 6 (Sino Biological Inc., Bejing, China) were immobilized on Series S CM5 chips (Cytiva, Marlborough, MA, USA) via amine coupling as per the manufacturer's recommendations. The Z TfR constructs were injected for 300 s at 30 µL × min −1 in a dilution series between 582-19.4 nM, in duplicate. Dissociation was monitored for 1000 s before regeneration with 10 mM HCl for 30 s, followed by a stabilization period of 60 s. Sensorgrams were double-referenced with the blank surface and a buffer injection of PBST and analyzed with Biacore evaluation software using 1:1 binding model.

Single Amino Acid Mutagenesis and E. coli Surface Display Binding Analysis
The Z TfR#14 and Z TfR#18 affibodies were mutated at 14 positions by introducing codons for histidine and the wildtype amino acid of the Z domain [45] at each position. Oligos encoding the mutated affibody genes (HT genparts, GenScript, Piscataway, NJ, USA) were subcloned to the E. coli display vector pBad2.2 [34] using Gibson assembly according to the supplier's recommendations. Four fragments were assembled simultaneously in the Gibson assembly method (New England Biolabs, Ipswich, MA, USA) with a two times molar excess of the insert to pBad2.2 vector and transformed to BL21STAR (DE3) E. coli (Thermo Scientific, Waltham, MA, USA) by standard heat shock protocol. Colonies were sequence-verified by the Sanger sequencing (Eurofins Genomics, Ebersberg, Germany) and further transformed into electrocompetent E. coli JK321 [54].
Next, cells were resuspended in ice-cold buffer pH 7.4 (25 mM HEPES, 25 mM NaCl, 125 mM KCl) buffer without nigericin and with 250 nM of respective FITC-labeled affibody molecule and incubated for 15 min, 15 rpm, at room temperature before spun down and washed once in a buffer of pH 5.0 and once at pH 7.0 (no nigericin) for 5 min at 15 rpm. The sample was resuspended in a pH 7.4 buffer and analyzed in the flow cytometer with the same settings as for the standard curve.

Transcytosis In Vitro Murine BBB Model
A previously described method based on recombinant spider silk nanomembranes with confluent monolayers of murine brain endothelial cells (bEnd.3) was used for the assessment of TfR-mediated transcytosis [47]. 500 nM of respective FITC-labeled affibody molecule together with an AF647-labeled internal-control IgG2a antibody was added in complete pre-warmed cell media to the silk membrane apical side with (n = 3) or without confluent bEnd.3 cell layer (n = 3) and placed into a well. Cell media was added to the basal side and incubated for 90 min (37 • C, 5% CO 2 ). The apical and basal samples were aspired, and together with the starting sample, the fluorescent intensity was measured at 483-14/530-30 nm and 575-20/621-10 nm with 1200/3000 gain at 25 • C in a CLARIOStar Plus (BMG Labtech, Ortenberg, Germany). Values were averaged by technical and biological replicates, and signals from complete cell media were subtracted. Statistical analysis was performed in Microsoft Excel using a two-sided Student's t-test with equal variance for the apparent permeability (p app , cm × s −1 ) calculated as by Equation ()1, where A is the surface area (cm 2 ), dQ/dt the steady-state flux (mmol × s −1 ), V R is the volume of the receiver chamber (cm 3 ) and C 0 the initial concentration of the affibody (mM).

Conclusions
To conclude, we describe the development of affibodies targeting TfR by phage display technology. Two top candidates were evaluated in vitro and further improved by single amino acid substitutions, yielding a total of six new variants. The six affibodies were studied in terms of cell binding and transcytosis, demonstrating higher apparent permeability for three of the TfR-binders compared to a negative control. Although the affibody constructs evaluated herein show promising results, further optimization of the binders is likely warranted prior to studies on brain uptake in animal models.