Interaction of clinical-stage antibodies with heme predicts their physiochemical and binding qualities

Immunoglobulin repertoires contain a fraction of antibodies that recognize low molecular weight compounds, including some enzymes’ cofactors, such as heme. Here, by using a set of 113 samples with variable region sequences matching clinical-stage antibodies, we demonstrated that a considerable number of these antibodies interact with heme. Antibodies that interact with heme possess specific sequence traits of their antigen-binding regions. Moreover they manifest particular physicochemical and functional qualities i.e. increased hydrophobicity, higher propensity of self-binding, higher intrinsic polyreactivity and reduced expression yields. Thus, interaction with heme is a strong predictor of different molecular and functional qualities of antibodies. Notably, these qualities are of high importance for therapeutic antibodies, as their presence was associated with failure of drug candidates to reach clinic. Our study reveled an important facet of information about relationship sequence-function in antibodies. It also offers a convenient tool for detection of liabilities of therapeutic antibodies.


Introduction
Antibodies (Abs) represent essential element of the adaptive immune defense. Typically they recognize unique molecular motifs (epitopes) displayed on biological macromolecules such as proteins, and carbohydrates. Nevertheless, immunoglobulin (Ig) repertoires contain also Abs that bind to various low molecular weight molecules. Thus, pioneering works with human and mouse myeloma-derived monoclonal Abs revealed unexpected high frequencies of molecules that interact with nitoarene compounds, especially 2,4-dinitrophenyl 1,2,3 . Further studies showed that the human Abs can also recognize other aromatic and heterocyclic molecules, including cofactors or prosthetic groups exploited by enzymes, namely -riboflavin, FAD, ATP, cobalamin, protoporphyrin IX and heme 4,5,6,7,8,9,10,11,12,13,14 . Human Abs also recognize a xenogenic disaccharide molecule i.e. galactosyl-(1−3)-galactose (α-Gal) 15,16,17 . These studies also suggested that the binding of cofactors or other low molecular weight molecules occurs in variable region of Abs. Of note, the interaction of Abs with some cofactors, results in dramatic effect on their functions. Thus, upon contact with heme some Abs acquire reactivity towards large panels of previously unrecognized antigens i.e. they become polyreactive 10,14,18,19,20 .
Detection of Abs interacting with low molecular weight compounds in human immune repertoires is not anecdotic but these Abs represent a considerable fraction of circulating Igs.
The particular physicochemical properties of heterocyclic molecules can explained the high reactivity with Abs. As for example, heme is a complex of tetrapyrole macrocylic compound protoporphyrin IX with iron ion (Fig 1A). The heme molecule displays different types of chemical groups offering possibility for establishing non-covalent interactions of various nature -hydrophobic, π-electrostatic, ionic bonding, metal coordination, and hydrogen bonding. Not surprisingly heme is very promiscuous molecule; apart from proteins that use heme constantly as prosthetic group (hemoproteins), large number of proteins bind heme in a less stable or transient manner 24,25 . Moreover, heme was reported to interact with lipids, nucleic acids, peptides and other aromatic compounds 26,27,28 . Large-scale investigations of the architecture of the hemebinding sites of proteins revealed some clear-cut tendencies. Thus, the binding sites for heme are predominantly enriched in certain amino acids such as histidine, cysteine, methionine, tyrosine and amino acids with basic side chains 29,30,31 . All these amino acids can establish direct interactions with different parts of heme molecule.
The fact that a significant portion of human Ab molecules interact with heme implies that these Abs may have a common molecular features of their binding sites that are suitable for accommodation of the cofactor molecules. We hypothesize that that interaction with heme can inform for qualities of antigen-binding sites that are expressed further as particular physicochemical and functional behavior of the Abs. To test this hypothesis, we used a repertoire of clinical-stage therapeutic monoclonal Abs. These Abs were previously extensively and meticulously characterized by battery of 12 methods that assess different physiochemical and functional properties 32 . Our results obtained with 113 samples with V region sequences corresponding to the clinical-stage therapeutic monoclonal Abs expressed on human IgG1 framework, revealed that interaction with heme correlated with specific sequence characteristics of antigen-binding site. Notably, the interaction with heme could be used as a potent predictive surrogate for number of molecular and functional qualities of Abs i.e. hydrophobicity, stability, natural polyreactivity, as well as propensities for self-binding and cross-reactivity with other Abs. Thus, in addition to providing evidence for the molecular organization of the paratopes of heme-binding Abs, this study also points to the utility of heme for prediction of the physicochemical features of V region and functional behavior of Abs. These data might extend our basic comprehension about the Ab molecule and pave the way for development of convenient analytical assays for early detection of some unwanted properties of candidate therapeutic Abs.

Interaction of therapeutic Abs with heme
Previous studies have demonstrated that significant fraction of human IgG interact with heme (Fe-protoporphyrin IX, Fig. 1A) 10 . To examine whether interaction with heme reflects certain physicochemical and functional qualities of Abs we used a repertoire of 113 monoclonal therapeutic Abs all expressed as human IgG1. At present, 44 of those Abs are approved for clinical use, and rest of the molecules reached at least Phase II or III clinical trials. These Abs were characterized thoroughly by diverse assays measuring physiochemical and binding properties 32 .
First, to assess the binding of Abs to heme, we used a method based on immune sorbent assay, where heme is covalently attached to pre-coated carrier protein. The obtained results clearly demonstrate that a significant portion of Abs bind to immobilized heme (14 % of Abs binding with 10-fold higher intensity to immobilized heme as compared to carrier alone) (Fig. 1B). Next, the interaction of the therapeutic Abs set with another aromatic cofactor molecule, folic acid (folate), was evaluated. Number of Abs binding to surface-immobilized folate was considerably lower as compared to heme-binding Abs (Fig. 1B). Nevertheless, Spearman's correlation analyses demonstrated a significant correlation between the reactivities of Abs against both heterocyclic cofactors (ρ = 0.69, P < 0.0001). The Abs that recognized heme with the highest intensity were subjected to additional analysis. Figure 1C depicts the Ab-concentration dependent binding to immobilized heme. Data indicates that those Abs bind with high avidity to heme (50 % of the maximal binding for most of the Abs was achieved by < 100 nM of IgG1).
Two Abs (figitumumab and fulranumab) recognized surface-immobilized heme with ca. 10-fold higher avidity as compared to other Abs. To determine the binding affinity of immobilized Abs to soluble heme, we used surface plasmon resonance-based technology. The obtained real-time interaction data of binding of heme to selected heme-binding monoclonal Abs characterized an interaction with K D values of 3.13 and 5.09 × 10 -7 M for fulranumab and figitumumab, respectively (Fig. 1D).
In addition to the approach based on surface immobilization of heme (or Abs), we analyzed the interaction of heme with Abs in solution by high-throughput UV-vis absorbance spectroscopy, and by examining the changes in the catalytic activity of heme (Fig. 2). These techniques confirmed that a fraction from the panel of therapeutic monoclonal Abs is capable of binding to heme. Indeed, the increase of absorption intensity in Soret region of heme, expressed as ΔA max (Fig. 2B), usually occur as result of change of molecular environment of oxidized form of heme from more polar, typical for the aqueous solution to more hydrophobic, typical for binding site on protein. The shift of the wavelength of the absorbance maximum in Soret region, expressed as λ max shift (Fig. 2C), indicates interaction of amino acid residues with heme's iron, thus also proving interaction of heme with Ab. Indication for binding of heme to Abs is also changes in the intrinsic peroxidase activity of the former (Fig. 2D).

Acquisition of antigen-binding polyreactivity after interaction of therapeutic Abs with heme
To analyse whether the exposure to heme affects the antigen-binding properties of Abs, we next investigated the interaction of the panel of therapeutic Abs with a panel of unrelated proteinshuman Factor VIII, human C3 and LysM domains of AtlA from Enterococcus faecalis. The binding of all Abs to immobilized proteins was assessed before and after the exposure to heme. Figure 3A illustrates that in their native state some of the therapeutic Abs have capacity to recognize the studied proteins. These Abs displayed natural antigen-binding polyreactivity since none of the proteins is cognate antigen of the studied therapeutic Abs. The presence of polyreactive Abs in the panel of therapeutic monoclonal Abs was also detected by three different assays in a previous study 32 . The obtained results also demonstrated that a fraction of therapeutic Abs acquired capacity to recognize the proteins following contact with heme (Fig. 3A). The heme-induced Ab reactivity to a given protein strongly correlated with binding to the other two proteins and to average binding reactivity against the three proteins (Fig. 3B). This result implies that the heme-sensitive Abs acquired antigen-binding polyreactivity upon exposure to heme. The Abs that demonstrated the highest gain in reactivity, as assessed by ELISA, were further subjected to immunoblot analyses. The observed recognition of multiple proteins from bacterial lysate confirmed the ability of heme to transform the binding behaviour of certain Abs (Fig. 3C).
Previous works suggested that heme endows Abs with polyreactivity by binding to the antigenbinding site and serving as interfacial promiscuous cofactor facilitating contact with unrelated proteins 10,14,33 . We applied computational Ab modeling (Rosetta algorithm) to generate the most probable structural configuration of the antigen-binding site of a heme-binding Ab. The selected Ab (sample corresponding to the sequences of tigatuzumab) displayed the highest sensitivity to heme in terms of both binding and acquisition of polyreactivity (Fig. 3A). By using ligand-docking algorithm, we predicted the most probable position of macrocyclic system of heme on the variable region. As can be seen on Figure 3D, the putative binding site for heme is overlapping with the central part of antigen-binding region. This model suggests that heme molecule could establish interaction with CDR loops of both light and heavy Ig chains. The docking analyses of the Ab displaying highest sensitivity to heme are in full accordance with previous findings that heme binding to antigen-binding site can be used as an interfacial cofactor for extension of antigen-recognition activity.
Heme influenced selectively only Fab-dependent functions of Abs. Thus, SPR-based analyses revealed that the binding of native and heme-exposed Abs to human neonatal Fc receptor (FcRn) was characterized with no significant differences in the binding affinity (Fig. 4).

Sequence characteristics of antigen-binding site of Abs interacting with heme
To provide understanding about the molecular features of Abs responsible for interaction with heme as well as acquisition of binding polyreactivity, we elucidated sequence characteristics of antigen-binding site of 113 Abs that overlap with molecules previously described in 32 . We focused our analyses on the most diverse regions of the antigen-binding site i.e. CDR3 of heavy and light chains, as well as CDR2 of heavy chain, since the latter is the second most diverse region in antigen-binding site after CDRH3 and provides the second largest molecular surface for contact with protein antigens 34 . By applying Spearman correlation analyses, it was observed that the length of CDRH3, CDRH2 and CDRL3 did not correlate with binding to heme or folate or with heme-induced polyreactivity (Fig. 5). No significant correlation was observed with the number of somatic mutations in the V regions either.
The analysis of frequencies of amino acid residues in CDRs, however, revealed a number of statistically significant correlations. Thus, the ability of Abs to recognize heme positively correlated with the presence of a higher number of basic amino acid residues in CDRH3, particularly with the higher number of Lys (Fig. 5). Positive correlation of binding to heme was also observed for the numbers of Arg residues in CDRH2. This analysis also showed that both heme-and folate-binding Abs had a significantly decreased frequency of acidic residues (Asp and Glu) in CDRH2. The heme-binding Abs have significantly higher number of Tyr residues in CDRH3. Frequency of another aromatic residue -Phe in CDRH3 and CDRH2 negatively correlated with higher binding to immobilized heme and folate. Interestingly, no statistically significant correlation was observed between frequency of any amino acid residue type in CDRL3 and the binding to heme or folate.
Next, we investigated the correlation of the heme-mediated acquisition of polyreactivity and sequence features of the antigen-binding site of Abs. The heme-induced polyreactivity of the therapeutic Abs negatively correlated with the presence of acidic residues in CDRH3. Similarly, as in the case of binding to heme, the Abs that acquired polyreactivity had significantly higher number of Tyr residues in CDRH3. Heme-sensitive Abs also presented an elevated frequency of Tyr in CDRH2. Abs acquiring polyreactivity following exposure to heme were characterized with lower frequencies of Ala and Leu in CDRH2 and CDRH3.
Structural models of the selected Abs with the most pronounced heme-binding capacity and/or heme-induced polyreactivity revealed that positive charges and Tyr residues are well displayed on the molecular surface of the antigen-binding site (Fig. 6). The surface accessibility of positively charged amino acid residues and Tyr on the molecular surface of antigen-binding sites and their increased frequency suggest that these residues might play a direct role in contacts with heme's carboxyl groups and aromatic system, respectively. Tyr residues are also known to establish metal-coordination interaction with heme's iron ion.

Biophysical properties of the therapeutic Abs that interact with heme
In a previous study, Jain and colleagues quantified different physicochemical properties of a repertoire of 137 monoclonal therapeutic Abs 32 . The measured features of Abs are includeexpression yield in HEK cells (HEKt), thermodynamic stability (assessed as melting temperature, Tm), accelerated stability (AS, assessed as long-term storage stability at elevated temperature i.e. 40 °C), tendency for self-association (ACSIN assay, which is based on gold nanopartilces, and CSI assay -based on biolayer interferometry, for measuring propensity for homotropic binding of Abs), binding to polyclonal human IgG (CIC, assay where the binding of the Abs to sepharose-immobilized pooled IgG was estimated), antigen-binding polyreactivity (PSR assay, where the reactivity of Abs to soluble membrane proteins from CHO cells was assessed in solution; ELISA assay where the reactivity of Abs to surface immobilized antigens -KLH, LPS, ssDNA, dsDNA and insulin was measured, and BVP assay where binding of Abs to baculovirus particles was evaluated by ELISA) and hydrophobicity (SGAC100 assay where hydrophobicity of Abs was studied by salting-out effect of ammonium sulfate, HIC and SMAC assays, where tendency of Abs for binding to two types of hydrophobic matrixes was assessed by chromatography). All these features have been shown to be of immense importance for successful entry of Ab in the clinical practice 32,35 .
The panel of therapeutic Abs used in the present study overlaps with the repertoire of Abs that was used in the work of Jain (113 of the studied monoclonal Abs correspond to identical samples). This allowed us to comprehensively analyse relationships between the physicochemical properties of the Abs assessed by different analytical techniques with their capacity to bind to heme or acquire polyreactivity upon exposure to heme. Statistical analyses revealed that the potential of therapeutic Abs to recognize immobilized heme strongly correlated with several biophysical properties of Abs (Fig. 7). Thus, Abs that have propensity to bind to heme were characterized with significantly higher natural polyreactivity, as assessed by three independent polyreactivity assays (PSR, ELISA and BVP) (Fig. 7). Moreover, the increased binding to immobilized heme positively correlated with the tendency of the therapeutic Abs for self-binding and binding to other IgG molecules as estimated by three different assays (ACSINS, CSI and CIC) (Fig. 7). These analyses revealed that the heme-binding Abs are generally aggregating at lower concentration of ammonium sulphate, i.e. they are characterized by elevated propensity for aggregation mediated by hydrophobic interactions (SGAC100 assay, Fig. 7).
Interestingly, stronger association of heme had also significant negative correlation with expression titer of the therapeutic Abs (Fig. 7).
Further, we investigated the relationship of biophysical properties of the Abs with heme-induced polyreactivity. The heme-sensitive Abs had significantly higher overall hydrophobicity, as determined by three independent assays (SGAC100, HIC and SMAC). The Abs acquiring polyreactivity following contact with heme were also characterized by their intrinsic propensity for self-binding and cross-reactivity to human polyclonal IgG (Fig. 7).
Statistical analyses of results from different assays involving heme additionally suggested that heme-induced polyreactivity positively correlates with heme binding in solution and increase in ABTS oxidation (catalytic assay) (Fig. 7). Moreover, these analyses demonstrate that the study of Ab binding to immobilized heme and heme-induced Ab polyreactivity has considerably higher power in predicting developability issues of Abs as compared with the heme-binding assays in solution (absorbance spectroscopy and ABTS oxidation).

Clinical status of the heme-binding Abs
The studied panel of samples sourced V region sequences from therapeutic Abs molecules that have been already approved for clinical use or have undergone evaluation in phase II and phase III clinical trials. In previous work, it was shown that the Abs that entered clinical practice tend to be characterized by significantly lower number of negative for therapeutic developability physicochemical and binding properties as compared with Abs that had merely reached clinical trials 32 . The threshold value of measures defining negative developability traits was set as the worst 10 % of value from different assays. By selecting Abs that bind heme with 10 % highest intensity (11 Abs in total), we found that among these Abs only two molecules have been approved for the clinical use (i.e. 18 %). Among the therapeutic Abs that bind less or do not bind at all to heme, 41 % have been approved for use in clinic. Most of the Abs (9 out of 11) showing the highest heme-binding intensity crossed at least one of the negative thresholds set for other developability assays in the study by Jain et al 32 . This result suggests that the interaction with heme is an indicator for the presence of liabilities of monoclonal Abs that might affect the final approval for clinical use.

Discussion
In the present study, we demonstrated that a considerable fraction of monoclonal therapeutic Abs recognizes heme. Following the exposure to heme some of the monoclonal Abs acquired antigen-binding promiscuity. We identified sequence patterns of the antigen-binding site that are associated with interaction with heme. Moreover, the interaction with heme significantly correlated (P < 0.005) with three different features of Ab molecule i.e. hydrophobicity, propensity for self-association and intrinsic antigen-bidding polyreactivity. Although with lower significance (P < 0.05), heme binding also negatively correlated with expression titer of Abs in eukaryotic cells, a parameter strongly dependent on protein stability 32 .
Previous studies have demonstrated that heme-binding sites on distinct proteins are enriched of specific amino acid residues 29,31,36 . Generally heme has tendency to associate with more hydrophobic regions of proteins 29 . Elucidation of sequence characteristics of the variable region of Abs that bind to heme revealed a positive correlation of the recognition of heme with the number of Tyr residues in the most diverse region of the Ig molecule i.e., CDRH3. The number of Tyr was also significantly elevated in the CDRH3 and CDRH2 of Abs that acquire polyreactivity after interaction with heme. Tyr is a well-known amino acid that can interact with heme. It is one of the amino acid residues particularly enriched in the heme binding sites on proteins 29,30,37 . Tyr can interact with heme both by aromatic (π-stacking) interactions or by coordination of central iron ion (through its hydroxyl group). Aromatic amino acid residues such as tyrosine, phenylalanine and tryptophan are usually localized in the interior of proteins. But Igs are unusual in that they carry large number of aromatic amino acids exposed to the protein surface, and more specifically in CDRs 38,39 . Thus, presence of Tyr in the antigen-binding site can provide appropriate environment for binding of aromatic compounds such as heme (and other cofactors). Besides Tyr residues, the binding of Abs correlated with elevated numbers of positively charged residues such as Lys and Arg in the CDRs. These residues can interact through ion bonds with carboxyl groups of heme (Fig. 1A), thus further facilitating the attachment of heme molecule to IgG. It is noteworthy that conventional heme-binding sites on diverse proteins are also enriched in basic amino acid residues 29 . The sequence analyses also showed that Abs that bind heme and acquire polyreactivity were depleted by amino acid residues with acidic side chains. This can be explained by potential repulsion of these side chains with the carboxyl groups of heme. The lower number of acidic residues in CDR regions can also explain elevated hydrophobicity of the Abs.
Our data revealed that heme could specifically detect elevated number of surface exposed positive amino acid residues and aromatic residues in Abs. The capacity of heme to detect both an increased presence of aromatic and positively charged amino acid residues in the V region well explains the strong correlation of certain physicochemical and functional parameters of V regions with the interaction with heme. Thus, the higher frequency of Tyr in the V regions was associated with an augmented hydrophobicity of the Abs and an increased tendency for homophilic binding (self-association) 40,41 . On the other hand, the higher prevalence of basic amino acid residues or surface patches with positive charge in the antigen-binding site has been demonstrated to correlate both with an antigen-binding polyreactivity and tendency for selfassociation 41,42,43,44,45 . These observations imply that heme preferentially interacts with V regions of Abs with particular molecular characteristics. Therefore, these results suggest that the molecular features of heme allow this compound to predict some important physicochemical and functional qualities of Abs. The utility of heme as a probe most probably stems from its distinctive chemical structure (Fig. 1A), allowing establishment of various types of non-covalent molecular interactions with proteins.
The present study also reveled that the set of clinical-stage Abs rcontains a fraction of molecules that specifically bind another cofactor -folate. Similarly as heme, the binding to folate correlated with some features of Abs such as propensity for self-binding and polyreactivity. Notably, the reactivity to heme strongly correlated with reactivity to folate (ρ = 0.69, P < 0.0001). This suggests that the recognition of different heterocyclic compounds might be an intrinsic property of a specific fraction of Abs in Ig repertoires. Nevertheless the Abs recognizing folate were significantly lower number and this molecule was not able to predict the hydrophobicity of V region or expression titer of Abs, suggesting that heme has superior capacity to characterize Abs. In conclusion, the present study demonstrated that significant fraction of clinical-stage Abs interact with heme. Specific sequence traits identify the set of cofactor-binding Abs. Moreover, those Abs that were able to interact with heme presented with particular physicochemical and functional qualities. This study contributes for better understanding of the fraction of Abs that recognizes cofactors. It may also have repercussions for applied science as heme molecule has the capacity to predict several liabilities of monoclonal Ab that can compromise their introduction into the clinical practice.

Antibodies
In the present study, we used 113 samples with variable region sequences corresponding to therapeutic antibodies that are approved for use in clinic or preceded as far as Phase II and Phase III clinical trials, with vesencumab being the lone exception. All antibodies were expressed as human IgG1 in HEK293 cells. A detailed description of production of this set of recombinant antibodies was provided in Jain et al 32 . All reagents used in the study were with the highest purity available.

Antibody reactivity to surface immobilized heme
For covalent in situ immobilization of heme or folic acid through their carboxyl groups, we applied amino-coupling procedure previously described in 63

Real time analyses of interaction of heme with antibodies
Surface plasmon resonance (SPR)-based assay (Biacore 2000, Biacore GE Healthcare, Uppsala, Sweden) was applied to elucidate kinetics of interaction of heme with therapeutic antibodies.
Antibodies demonstrating substantial binding to immobilized heme -figitumumab, fulranumab and tigatuzumab were covalently coupled to surface of CM5 sensor chips (Biacore) using amino-

Immunoblot
Lysate of Bacillus anthracis was loaded on a 4-12 % gradient NuPAGE Novex SDS-PAGE gel (Invitrogen, Thermo Fisher). After migration, proteins were transferred on nitrocellulose membranes (iBlot gel transfer stacks, Invitrogen, Thermo Fisher) by using iBlot electrotransfer system (Invitrogen, Thermo Fisher). Membranes were blocked overnight at 4 °C in PBS buffer containing Tween 0.1% (PBS-T). Next, the membranes were fixed on Miniblot system (Immunetics, Cambridge, MA) and incubated for 2 hours at RT with 10 µg/ml of selected native or heme-exposed therapeutic antibodies. The antibodies were pre-treated at 50 µg/ml in PBS with 5 µM hemin for 10 min and diluted in PBS-T before loading. Membranes were washed (6 × 10 min) with excess of PBS-T before being incubated for one hour at RT with an alkaline phosphatase-conjugated goat anti-human IgG (Southern Biotech, Birmingham, AL), diluted 3000 × in PBS-T. Membranes were then thoroughly washed (6 × 10 min) with PBS-T before revealed with SigmaFast NBT/BCIP substrate solution (Sigma-Aldrich).

Real time analyses of interaction of antibodies with human FcRn
The binding of selected heme-sensitive antibodies to human FcRn before and after exposure to heme was studied by SPR-based biosensor technology (Biacore). Pro, microplate reader. The spectral resolution was 2 nm. All measurements were done at RT. To assess the changes in spectral properties of heme upon interaction with antibodies, the shift in maximal absorbance wavelength was defined as follows: λ-shift = λ max of hemin aloneλ max of hemin in the presence of antibody. The differential spectra (absorbance spectrum of hemin in the presence of antibody -absorbance spectrum of hemin alone) allowed quantification of the maximal increase in the absorbance intensity (ΔA).

Peroxidase assay
To evaluate how interaction with antibodies influences the intrinsic peroxidase activity of heme, we performed colorimetric catalytic assay. The antibodies from the studied repertoire were first diluted in PBS to 100 µg/ml (670 nM) and exposed to 5 µM final concentration of hemin. The absorbance at 414 nM was recorded at 10 and 20 min after mixing by using Tecan Infinite 200 Pro, microplate reader. The peroxidase activity of BSA at 150 µg/ml exposed to 5 µM hemin was used as a control.

Structural modeling of the variable regions of antibodies and molecular docking
For modeling of three dimensional structures of the variable domains of heavy and light chains of selected heme-binding and heme-sensitive antibodies, we used sequence-based modeling algorithm -RosettaAntibody3 program 64,65,66,67 . The amino acid sequences of the variable regions were loaded to ROSIE online server (http://antibody.graylab.jhu.edu/). Relaxed 3D models with lowest free energy were visualized by using Chimera UCSF Chimera package.
Chimera is developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco (supported by NIGMS P41-GM103311) 68 . Chimera software was also used for visualization of Coulombic electrostatic potential of antigen-binding sites as well as to highlight the surface exposed Tyr residues in the antigen-binding site.
The structural model of antibody variable region of tigatuzumab was used as an input for docking of protoporphyrin IX. To this end, we used SwissDock web service, which uses CHARM force field and it is dedicated to the prediction of protein-small molecule interaction 69,70 . The three-dimensional structures of the most probable complex were visualized by using Chimera software.

Statistical analyses
Abs that meet the following criteria were included in statistical analyses: availability of sequence