Roles of the respective loops at complementarity determining region on the antigen-antibody recognition

Accepted Manuscript Title: Roles of the respective loops at complementarity determining region on the antigen-antibody recognition Author: Tomonori Osajima Tyuji Hoshino PII: DOI: Reference:

S1476-9271(16)30177-3 http://dx.doi.org/doi:10.1016/j.compbiolchem.2016.08.004 CBAC 6574

To appear in:

Computational Biology and Chemistry

Received date: Revised date: Accepted date:

13-4-2016 16-8-2016 18-8-2016

Please cite this article as: Osajima, Tomonori, Hoshino, Tyuji, Roles of the respective loops at complementarity determining region on the antigen-antibody recognition.Computational Biology and Chemistry http://dx.doi.org/10.1016/j.compbiolchem.2016.08.004 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Page 1 of 48

Roles of the respective loops at complementarity determining region on the antigen-antibody recognition Tomonori Osajima, Tyuji Hoshino Graduate School of Pharmaceutical Sciences, Chiba University, Inohana 1-8-1, Chuo-ku, Chiba 260-8675, Japan *

To whom correspondence should be addressed. Tyuji Hoshino Graduate School of Pharmaceutical Sciences, Chiba University Phone: +81-43-226-2936, Fax: +81-43-226-2936 E-mail: [email protected]

Page 2 of 48

Graphical Abstract

Page 3 of 48

Highlights ● The direct hydrogen bond with the shortest distance is principally observed in the heavy chain. ● Tyr in the heavy chain (especially at H2 and H3) primarily contributes to the binding free energy. ● The appearances of Asn and Gln are large in the light chain. ● Fab is more favorable than scFv for antigen-antibody interaction from the energetic viewpoint. ● The length of H3 CDR loop has no correlation to the binding affinity.

Page 4 of 48

Abstract For the rational design of antibody, it is important to clarify the characteristics of the interaction between antigen and antibody. In this study, we evaluated a contribution of the respective complementarity determining region (CDR) loops on the antibody recognition of antigen by performing molecular dynamics simulations for 20 kinds of antigen-antibody complexes. Ser and Tyr showed high appearance rates at CDR loops and the sum of averaged appearance rates of Ser and Tyr was about 20 - 30 % at all the loops. For example, Ser and Tyr occupied 23.9% at the light chain first loop (L1) and 23.6% at the heavy chain third loop (H3). The direct hydrogen bonds between antigen and antibody were not equally distributed over heavy and light chains. That is, about 70% of the hydrogen bonds were observed at CDRs of the heavy chain and also the direct hydrogen bond with the shortest distance mainly existed at the loops of the heavy chain for all the complexes. It was revealed from the comparison in contribution to the binding free energy among CDR loops that the heavy chain (especially at H2 and H3) had significant influence on the binding between antigen and antibody because three CDR loops of the heavy chain showed the lowest binding free energy (ΔGbind) in 19 complexes out of 20. Tyr in heavy chain (especially in H2 and H3) largely contributed to ΔGbind whereas Ser hardly contributed to ΔGbind even if the number of the direct hydrogen bond with Ser was the fourth largest and also the appearance rate at CDR was the highest among 20 kinds of amino acid residues. The contributions of Trp and Phe, which bear aromatic ring in the side chain, were often observed in the heavy chain although the energetic contribution of these residues was not so high as Tyr. The present computational analysis suggests that Tyr plays an outstanding role for the antigen-antibody interaction and the CDR loops of the heavy chain is critically important for antibody recognition of antigen.

Keywords: antigen-antibody complex, complementarity determining region, binding free energy, molecular dynamics simulation, energetic contribution, amino acid residues

Page 5 of 48

1. Introduction Antibodies occupy an important position in biochemical research, diagnosis, and therapy, in recent years (Rodrigues et al., 2010; Huamg et al., 2011). Regarding therapeutics, the contribution of antibody is increasing and nowadays antibody drugs account for one third of all newly approved medicines for cancers, arthritis, asthma, psoriasis, virus infection, transplant rejection, and so on (Banerjee, 2010; Reichert, 2013; Pai et al., 2009). Antibodies are a family of glycoproteins which are specifically bound to the respective antigens. A remarkable feature of the antigen-antibody interaction is its high specificity and high affinity (Kumagai and Tsumoto, 2001; Altshuler et al., 2010; Li et al., 2007; Desmyter et al., 2001). A binding strength between an antigenic determinant in an antigen (epitope) and an antigen-binding site in an antibody (paratope) is related to their affinity. It is important to understand the characteristics of molecular binding between antigen and antibody for the rational design of an antibody or the enhancement of its affinity. We clarified the following features in our previous study (Osajima et al., 2014). (i) Ser shows the highest appearance among 20 kinds of amino acid residues at CDR loops of antibody, and Tyr has the second highest appearance. (ii) Tyr has the largest number of the direct hydrogen bonds at CDRs, and Asn, Asp, and Ser show the second, third, and forth largest numbers, respectively. As for number of the direct hydrogen bonds, non-charged polar residues (about 60%) and charged polar residues (about 30%) occupy approximately 90% of the total. On the other hand, non-polar residues are only involved in about 10% of the total. Moreover, water molecules contribute to the stabilization of complex structures not only by occupying the space at the antigen-antibody interface but also by intermediating the indirect hydrogen bond between antigen and antibody. The number of the direct hydrogen bonds at the antigen-antibody interface is much larger than that in the case of the complex of low molecular-weight compounds and their target proteins. (iii) The electrostatic energy significantly contributes to the formation of the complex contrary to the case of low molecular-weight compound in

Page 6 of 48

which van der Waals energy is predominant. (iv) The shape complementarity of the antigen–antibody interface is good, but not so high as low molecular-weight compounds bound to the target proteins. In this study, we will report the roles of the respective CDR loops from the statistical and energetic viewpoints. The analysis is focused on the appearance rate of specific amino acid residues at each CDR loop (L1, L2, L3, H1, H2, and H3), the distribution of the direct hydrogen bonds at each loop, and the contributions of the respective loops to the binding free energy. These examinations will deepen the understanding of the characteristics of molecular interaction between antigen and antibody and the recognition of antibody to antigen. Molecular dynamics (MD) simulations were performed for 20 kinds of the antigen-antibody complexes which were used in our previous study (Osajima et al., 2014). Twenty antigen-antibody complexes were selected based on the following condition; (1) antigen is not a linear short peptide, (2) the X-ray diffraction has fine resolution for the crystal structure, and (3) the secondary structures at the contact area of antigen shows diversity such as helix, sheet, loop etc. MD simulations are one of the major computational approaches for analyzing the molecular interaction. MD simulation can provide us the atomic level understanding on the binding mechanism between a target protein and its ligand. For example, the stability of an antigen-antibody binding is examined by the energy change through the simulation trajectory and the strength of a hydrogen bond is evaluated from the time for which the hydrogen bond is established during simulation. Hence, statistical analysis with MD simulation was applied to the antigen- antibody interaction, which will suggest the roles of the respective CDR loops of antibody.

Page 7 of 48

2. Methods 2.1. Construction of computational model Three-dimensional structures of the antigen–antibody complexes used in this study were extracted from the X-ray crystallographic data in the protein data bank (PDB). The PDB codes of the complexes are 1A14 (Malby et al., 1998; Kortt et al., 1999; Lee and Air, 2002), 1AFV (Momany et al., 1996; Malbet et al., 2000), 1BGX (Murali et al., 1998), 1DQJ (Li et al., 2000; Li et al., 2003), 1DVF (Braden et al., 1996; Goldbaum et al., 1997), 1FJ1 (Ding et al., 2000; Nassal et al., 2005), 1H0D (Chavali et al., 2003) , 1IGC (Derrick and Wigley, 1994; Khoury and Lowe, 2013) ,1IQD (Spiegel et al., 2001; Dimitrov et al., 2010), 1JPS (Faelber et al., 2001; Eigenbrot et al., 2003), 1KB5 (Housset et al., 1997; Mazza et al., 1999), 1NDG (Li et al., 2003; Nakanishi et al., 2008; Mohan et al., 2009), 1NDM (Li et al., 2003; Mohan et al., 2003), 1NL0 (Huang et al., 2004; Shi et al., 2005), 1UA6 (Nakanishi et al., 2008; Kumagai et al., 2003; Acchionea et al., 2009; Yokota et al., 2010), 2JEL (Prasad et al., 1998; Smallshaw et al., 1999), 2VIR (Fleury et al., 1998; Edwards et al., 2001; Lee et al., 2012), 2VIS (Fleury et al., 1998; Edwards et al., 2001; Lee et al., 2012), 2VIT (Fleury et al., 1998; Edwards et al., 2001; Lee et al., 2012) and 3HFL (Cohen et al., 2005; Długosz et al., 2009). The initial structure of each calculation model was constructed from the atom coordinates of the crystal structure with modifications by adding the missing residues and generating hydrogen atoms. The residues extracted from the crystal structure for the modeling are listed in Table S1 in Supplementary Materials by its chain ID and residue number. The missing residues were manually added with setting the adequate xyz-coordinates for main-chain atoms and then the coordinates of the side-chain atoms were automatically generated by leap module of AMBER11 program package (Case et al., 2005). However, no residue was added for the missing residues at the N- and/or C-terminus in the crystal structure. The N- and C-terminal sides of protein molecules were capped by ACE and NME groups, respectively. Each complex model was placed in a rectangular periodic-boundary box and solvated with TIP3P water molecules using leap module. The solvated box was made neutral by

Page 8 of 48 generating the counter ions. The box size, the number of ions, and the number of water molecules in the respective computational models are also shown in Table S1.

2.2. Calculation condition for MD simulation MD simulation was performed with sander module of AMBER11 (Case et al., 2005). AMBER ff03 force field was applied. MD simulation was executed in three steps of minimization, heating, and equilibration. Molecular geometry with the least total energy was obtained in the minimization step. The minimization was performed with the steepest descent method for the earlier 3,000 cycles and with the conjugated gradient method for the later 10,000 cycles, with only water molecules permitted to move freely. Subsequently, the minimization was performed again in a similar manner without any positional constraint on the atoms (Brooks et al., 1983).

In the heating step, the temperature was gradually increased to 300 K under

the NVT ensemble condition. Then, the equilibration calculation was performed under the NPT ensemble condition with a temperature of 300 K and a pressure of 1 atm. A periodic boundary condition was applied to all the xyz-directions, and the pressure and the temperature were kept constant. The cutoff distance for van der Waals and Coulomb forces in a real space was set to 12.0 Å. The particle mesh Ewald method was applied to estimate the influence of long-distance electrostatic force. For all of the twenty complexes, 10 ns MD simulation was carried out and the trajectory for the last 2 ns was collected for analysis. Snapshot structures were extracted from the trajectory every 10 ps in order to analyze the average structure, hydrogen bonding, and the interaction energy of each CDR loop and amino acid residue.

2.3. Analysis of simulation data The average structure of each complex was calculated using the last 2 ns of trajectory. Protein structures, in particular, the positions of the CDR loops were visualized with PyMOL (DeLano, 2012). The calculations

Page 9 of 48

should be sufficiently equilibrated for the sake of reliable analysis. The root-mean-square deviation (RMSD) during MD simulation was monitored with respect to the main chain Cα, N, and C atoms to confirm the equilibration. Further, RMSD changes separately obtained for the antibody and antigen molecules, in which each molecule was fitted to the reference structure independently. The formation of hydrogen bond was defined in terms of distance and angle. The combination of donor D, hydrogen H, and acceptor A atoms with a D-H∙∙∙A configuration was regarded as being in the hydrogen bond formation when the distance between donor D and acceptor A was shorter than 3.5 Å and the angle H-D-A was smaller than 60.0°. The hydrogen bonds were examined using the trajectory of the last 2 ns simulation and the distances were averaged over the trajectories. The binding free energy (ΔGbind) of the complex and the contribution of each CDR loop to ΔGbind were calculated by MM-GB/SA method (Kollman et al., 2000; Gohlke and Case, 2004) in a similar manner to our previous study (Osajima et al., 2014). Additionally, the contributions of the respective amino residue to ΔGbind was calculated by decompose method (Gohlke et al., 2003; Ode et al., 2007; Chen et al., 2012). ΔGbind is obtained from the equation : ΔGbind = ΔEele + ΔEvdW + ΔGsol – TΔS, where ΔEele, ΔEvdW, ΔGsol, and ΔS are the changes in electrostatic energy, van der Waals energy, the solvation free energy, and conformational entropy, respectively (Chong et al., 1999). In calculation of ΔGbind and its energy decomposition, we neglected - TΔS from the following two reasons. First, the aim of the energetic analysis is to compare the interactions of the respective CDR loops and amino residues to ΔGbind. Second, the predictive accuracy of - TΔS is relatively low (Hou et al., 2002; Lee et al., 2010).

2.4. Analysis of appearance rates of amino acid residues at the respective CDR loops. The residues at CDR loops for all the complexes were determined by following Kabat numbering

Page 10 of 48

scheme. The averaged appearance rates of amino acid residues at 6 CDR loops were classified into three groups from the physicochemical property, namely, non-polar residues (Ala, Gly, Ile, Leu, Met, Phe, Pro, Trp, and Val), uncharged polar residues (Asn, Cys, Gln, Ser, Thr, and Tyr), and charged polar ones (Arg, Asp, Glu, His, and Lys). The averaged appearance rate of an amino acid residue was calculated by the percentage ratio of the sum of the number of the residue at one CDR loop in all the antibodies to the sum of all the residues at the same CDR loop in all the antibodies. The direct hydrogen bonds were separately examined with respect to 6 CDR loops and amino acid residues.

Page 11 of 48

3. Results 3.1. Averaged structures of the complexes and positions of the respective CDR loops For all the complexes, RMSD for the main chain Cα, N, and C atoms was calculated to confirm the equilibration of the simulations and to examine the dynamic stability of the complexes (Figs.S1-1 and S1-2 in supplementary material). The system was judged to be equilibrated because the fluctuation of the RMSD was almost within 2.0±0.5 Å during the last 2 ns MD simulation (Table S2). Then, the trajectories of the last 2 ns were used to analyze the antigen-antibody interaction. The changes of RMSD values were also monitored for the antibody and antigen molecules separately (Figure S2-1 and S2-2). In most of the complexes, the fluctuation in RMSD of the antigen was smaller than that of the antibody. In other words, a rise and fall was sometimes observed in RMSD value for the antibody. The rise and fall was reflected in the averaged RMSD in Table S2, in which the antibody showed a high RMSD compared to the antigen in many complexes. Further the averaged RMSD value for the complex is usually larger than those for the antibody and antigen part only. This suggest that the antibody-antigen binding structure is likely to change compared to the other part of the complex molecules. It is also interesting to note that RMSD fluctuations for scFv complexes; 1A14, 1AFV, 1DVF, and 1UA6, are smaller than those for Fab. The averaged structures and the locations of CDR loops of the antibodies without water molecules are shown in Figs. 1 and 2, respectively. All the averaged structures in this study were almost the same as the previous study (Osajima et al., 2014). In all the antibodies except for 1IGC, both of L3 and H3 were located at the center of the binding domain between antigen and antibody. This L3 - H3 location is involved in the characteristic interaction between antigen and antibody.

3.2. Averaged appearance rates of amino acid residues at the respective CDR loops The domain of each CDR loop in antibody for all complexes was determined by following Kabat

Page 12 of 48

numbering scheme. The averaged appearance rates (%) of non-polar amino acid residues at the respective CDR loops in antibodies were calculated for all the complexes and shown in Fig. 3A. Phe and Trp, each of which possess an aromatic ring, showed the highest averaged appearance rates at H3 (6.7%) and L3 (8.3%), respectively. The averaged appearance rate of Ala was the highest at L2 (13.6%) and that of Ala at L1 (9.8%) was the second highest. Gly had the highest averaged appearance rate at H2 (13.3%) and the second highest at H3 (9.1%). Leu was frequently observed in the light chain rather than in the heavy chain and the averaged appearance rate was 4.1% at L1, 5.7% at L2, and 7.2% at L3. The averaged appearance rate of Met was noticeably high at H1 (7.6%). Pro showed the highest averaged appearance at L3 (7.7%) and the second highest at H2 (4.8%), followed by L2 (3.6%). The averaged appearance rates of Ile and Val was the highest at H2 (6.9%) and at L1 (4.9%), respectively. The averaged appearance rates (%) of uncharged polar amino acid residues are illustrated in Fig. 3B. It was clarified that Ser and Tyr always existed in either loop in all the complexes. The averaged appearance rate of Ser was the highest at L1 (23.9%) and the lowest at H3 (3.4%). On the other hand, Tyr showed the highest averaged appearance rate at H3 (23.6%) and the lowest at L2 (6.4%). The sum of averaged appearance rates of Ser and Tyr was approximately 20 - 30%. These two amino residues showed the high occupancy for all the loops. The averaged appearance rate of Asn was the highest at L1 loop (13.0%). Interestingly, the averaged appearance rate of Gln at L3 (14.9%) was significantly higher than the other loop whereas that of Gln was 0% at H1, and the appearance of Gln tended to be higher in the light chain than the heavy chain. Thr showed almost the same averaged appearance rate (10.0－11.6%) among L2, L3, and H1, and the averaged appearance rate of Thr was the lowest at H3 (2.4%). No Cys was observed at any of 6 CDR loops in all the antibodies. The averaged appearance rates (%) of charged polar residues are shown in Fig. 3C. Interestingly, Asp at H3 (17.8%) had the highest averaged appearance rate among all the loops. Additionally, the averaged

Page 13 of 48

appearance rate of Asp was higher in the heavy chain than the light chain. Arg and His had the highest averaged appearance rates at H3 (8.7%) and H1 (8.5%). The averaged appearance rate of Lys at H2 (6.6%) was the highest among all the loops. The averaged appearance rate of Glu was low as well as Lys through all the loops.

3.3. Distribution of the direct hydrogen bonds between antigen and antibody The numbers of the direct hydrogen bonds involved in the antigen-antibody interaction are shown in Fig. 4A. The direct hydrogen bonds were observed for all the 20 complexes and the numbers ranged from 4 to 38 (14.6 ± 7.7). Additionally, the direct hydrogen bonds more than 50% of the total for 1BGX existed at non-CDR. Furthermore, all the direct hydrogen bonds for 1IGC existed at non-CDR because the antigen is bond to the outside domain of CDR. Fig. 4B represents the number of the respective amino residues related to the direct hydrogen bonds at CDRs. The number of amino residues contributed to the hydrogen bonding decreased in order of Tyr, Asn, Asp, and Ser from the highest. The number of the hydrogen bonds at each loop for all the complexes is shown in Fig. 4C. The maximum number of the direct hydrogen bonds among CDR loops was 5 at L1 for 1NDM, 4 at L2 for 1DQJ, 6 at L3 for 1JPS, 8 at H1 for 1JPS, 12 at H2 for 1BGX, and 11 at H3 for 1IQD. The hydrogen bond in the heavy chain is always observed among all the complexes except for 1IGC whereas no hydrogen bond was observed in the light chain for 1DVF, 1NL0, 2VIR, 2VIS, and 2VIT. Fig. 4D illustrates the distribution of the direct hydrogen bonds at the respective loops for all the complexes. The number of the direct hydrogen bonds at each loop decreased in order of H2, H3, H1, L1, L3, and L2. Interestingly, the number of the direct hydrogen bonds in the heavy chain occupied approximately 70% of the total.

Page 14 of 48

The direct hydrogen bonds with the shortest distance between antigen and antibody are summarized in Table 1. The steady hydrogen bond, whose bond distance is 2.5－2.6 Å, was observed at H2 for 1DQJ, at H2 for 1NDM, and at H1 for 3HFL. Surprisingly, the strong hydrogen bond, whose bond distance is < 2.5 Å, was seen at H2 for 1UA6. The direct hydrogen bond with the shortest distance in each complex was dominantly distributed to the heavy chain among all the antibodies (17/21: about 80%). This result is well correlated with the high existence rate (approximately 70%) of all the direct hydrogen bonds at CDR loops in the heavy chain as described above. The details of the direct hydrogen bond between antigen and antibody for the respective complexes are listed in Table S3‐S22.

3.4. Binding free energy at the respective CDR loops ΔGbind and its energy components for all the complexes are shown in Table 2. ΔGbind values for all the complexes were negative and in the range of -167.5 kcal/mol (1BGX) to -25.8 kcal/mol (1AFV). ΔEele was lower than ΔEvdW in 18 out of 20 complexes. Furthermore, ΔEvdW was negative for all the complexes but ΔEele was positive for 1KB5 (40.5 kcal/mol). Table 3 presents ΔGbind for each CDR loop of antibodies for all the complexes. The lowest ΔGbind among 6 CDR loops was observed in the heavy chain for all the complexes except for 1IGC. The sum of ΔGbind of three loops in the heavy chain was lower than that in the light chain for all the complexes except for 1IGC. Many complexes had the lowest ΔEele (Fig. 5A) and/or ΔEvdW (Fig. 5B) in the heavy chain (85% and/or 80% of the total, respectively). The number of the complexes that had the lowest ΔGbind at any loop in the heavy chain was large and the number was 2 at H1, 10 at H2, and 7 at H3 (Fig. 5C). The results of Table 3 are graphically illustrated in Figs. S3-1 and S3-2. In addition, the details of ΔGbind and its energy components for the respective loops of all the complexes are shown in Table S23‐S42.

Page 15 of 48

3.5. Binding free energy of the respective amino acid residues ΔGbind and its energy components for each amino acid residue of antibody were calculated for all the complexes by MM-GB/SA decomposition method. Calculated ΔGbind for each residue is shown in Fig. 6-1 to 6-2. In all the complexes except for 1IGC, the key amino residues that significantly contribute to ΔGbind were mainly observed at CDR loops. Table 4 summarizes the lowest ΔGbind among amino residues at CDR loops of all the complexes except for 1IGC. The residue with the lowest ΔGbind was Tyr in 15 out of 19 complexes (15/19: about 80%). Additionally, this high contribution of Tyr was dominantly observed in the heavy chain (14/15). The details of ΔGbind and its energy components for the amino residues at CDR loops are shown in Fig. S4‐S23. ΔEele rather than ΔEvdW significantly contributed to the antigen-antibody interaction except for 1IGC. Except for 1IGC, Asp showed the lowest ΔEele among 20 kinds of amino acid residues at all the loops and Asp occupied approximately 60% (12/19) of the total residues with the lowest ΔEele. Lys and Glu showed the second (4/19) and the third (3/19) lowest ΔEele. It should be noted that these residues are classified into the charged polar residue.

Page 16 of 48

4. Discussion 4.1. Averaged appearance rates of amino acid residues at the respective CDR loops It is important to know what kind of amino acid residues or how many residues exist at CDR for deeply understanding the recognition mechanism of antibody to antigen. The appearance rates of amino acid residues at the whole domain or CDR have been frequently reported so far. However, the appearance rates of amino residues at the respective loops of antibodies have not been almost reported. In this study, we estimated the appearance rates of amino acid residues at the respective CDR loops for 20 kinds of antigen-antibody complexes. The averaged appearance rates of amino residues at 6 CDR loops were classified into, non-polar, uncharged polar, and charged polar residues from the physicochemical property. As for non-polar amino acid residues, the residues with an aromatic ring play an important role in the antigen-antibody binding based on the π-π or CH-π interaction (Chen et al., 2012; Yuki et al., 2007; Sivasakthi et al., 2013). According to our analysis, Phe has the highest averaged appearance rate in the heavy chain (especially at H2 and H3) whereas Trp has the highest averaged appearance rate in the light chain (especially at L3). In addition, the sum of the averaged appearance rates of the residues with an aromatic ring is higher at the loops in the heavy chain than those in the light chain. The sum is significantly high at H3 loop and is about 34%, which is attributed to the high averaged appearance rate of Tyr at H3. The above results suggest that amino acid residues bearing an aromatic ring and existing at CDR loops in the heavy chain greatly contribute to the antigen-antibody interaction. Gly has the highest averaged appearance rate at H2 and exists in the heavy chain rather in the light chain. It has been reported that Gly was highly abundant at CDRs in all the heavy chains in antibodies (Kunik and Ofran, 2013) and Gly markedly occupied at H2 in the clustering analysis (North et al., 2011). Our results are consistent with their reports. Additionally, the highest averaged appearance rates were observed at L2 (13.6%) for Ala, at H2 (6.9%) for Ile, at L3 (7.2%) for Leu, at H1 (7.6%) for Met, at L3 (7.7%) for Pro, and at L1 (4.9%) for Val.

Page 17 of 48

Regarding uncharged polar amino acid residues, Ser has the highest averaged appearance rate at L1 and the lowest at H3 whereas Tyr has the highest averaged appearance rate at H3 and the lowest at L2 (Fig. 3B). Furthermore, the distribution of Ser is higher in the light chain than the heavy chain. The sum of the averaged appearance rates of Tyr and Ser is approximately 20 - 30% through all CDR loops and both residues highly occupy every loop of antibody. This implies that many Ser and Tyr residues at CDR make a large contact with antigen and these two residues are intrinsically responsible for molecular recognition (Fellouse et al., 2004; Fellouse et al., 2005; Fellouse et al., 2006). Asn has the highest averaged appearance rate at L1（13.0%）, followed by L2 (9.3%). It has been revealed that Asn was abundant in paratope of antibody and also existed in the light chain (Kringelum et al., 2013). Our findings are consistent with their results. The averaged appearance rate of Gln is the highest at L3 (14.9%) and Gln tends to abundantly exist in the light chain rather than in the heavy chain. Additionally, no Gln is observed at H1 in all the antibodies. It has been clarified that the appearance of Gln was higher in the light chain than in the heavy chain and the presence of Gln at H1 was low (Ofran et al., 2008). They also found that the presence of Thr was low at H3. Furthermore, the significant absence of Cys at all the CDR loops has been shown (Ofran et al., 2008). Therefore, these facts demonstrate the validity of our analytical results for Gln and Thr. As for charged polar amino acid residues, the averaged appearance rate of Asp is the highest at H3 (17.8%) and also higher in the heavy chain than the light chain. It was reported that Asp was abundant in paratope of antibody and also abundantly existed in the heavy chain, especially at H3 (Kringelum et al., 2013; Ofran et al., 2008). On the other hand, the averaged appearance rate of Glu is low although Glu exists at every CDR loop (≤ 3.0%). It has been elucidated that the appearance of Gln was low at all the CDR loops (Kringelum et al., 2013; Ofran et al., 2008). In addition, our results for Arg, His, and Lys are almost the same as the previous report (Ofran et al., 2008). We calculated the averaged appearance rates for amino acid residues at 6 CDR loops with 20 kinds of antigen-antibody complexes. The averaged appearance rate of

Page 18 of 48

residues obtained in our study is almost consistent with the expression frequency (poor and/or rich) of residues analyzed with many samples as described above.

4.2. Comparison of the direct hydrogen bonds among six CDR loops The hydrogen bond is a main factor to stabilize the structure of antigen-antibody complex and the electrostatic energy significantly contributes to the binding between antigen and antibody. Hence, the comparison of the hydrogen bonds among CDR loops will be informative to understand which loop of antibody works significantly and how degree its loop contributes to the binding to the antigen. Then, we focused on and evaluated the direct hydrogen bonds among 6 CDR loops because the direct hydrogen bond is a basic component of electrostatic energy and greatly contributes to the stability of the molecular binding. According to our analysis, the number of the direct hydrogen bonds between antigen and antibody is the most at H2 among all the loops and the second most at H3, followed by H1, L1, L3, and L2. The number of the direct hydrogen bonds in the heavy chain is more than that in the light chain (approximately 70% of the total). Additionally, the direct hydrogen bond with the shortest distance in each complex is principally observed in the heavy chain. Moreover, strong hydrogen bonds with short bonding distance are observed only in the heavy chain. These results clearly suggest that the heavy chain significantly contributes to the interaction between antigen and antibody. On the other hand, the L2 loop is the smallest contribution to the antigen-antibody interaction. The sum of the number of the direct hydrogen bonds at the respective CDR loops of antibodies for all the complexes except for 1IGC is illustrated in Fig. 7. The number of the direct hydrogen bonds by Tyr, Asn, Asp, and Ser are the most at H2 (17), at L1(20), at H3(18), and at H2(11), respectively. Interestingly, the correlation is observed between the number of the direct hydrogen bonds and the averaged appearance rates

Page 19 of 48

of Asn and Asp. In contrast, the correlation is hardly observed for Tyr and Ser. In case of Tyr, the averaged appearance rates at H2 and H3 are 11.7% and 23.6%, respectively. Therefore, the averaged appearance rate of Tyr at H2 is a half of that at H3. The averaged appearance of Ser at H2 (12.0%) is a half of that at L1 (23.9%) as well. Furthermore, the number of the direct hydrogen bond of Ser at loop L2 is zero nevertheless the averaged appearance rate of Ser is 22.9% and it is the second highest value among all the loops.

4.3. Comparison of binding free energy among CDR loops It is informative to examine the interaction between antigen and CDR loop of antibody from the energetic viewpoint. The energetic contribution of the respective loops of antibody with MD simulation has been hardly reported. Accordingly, we calculated ΔGbind between antigen and each loop of the antibodies for 20 kinds of complexes by MM-GB/SA method. According to our analysis, the CDR loops having the lowest ΔGbind are always seen in the heavy chain (10 complexes at H2, 7 complexes at H3, and 2 complexes at H1) except for 1IGC. Further the sum of ΔGbind of the three CDR loops in the heavy chain is lower than that of ΔGbind of those in the light chain for all the complexes except for 1IGC. This result suggests that the heavy chain rather than the light chain significantly contributes to the antigen-antibody binding in terms of energy. Robin et al. reported that the contribution ratios of the respective CDR loops to the binding energy for 206 kinds of the antigen-antibody complexes were 9% at L1, 0% at L2, 12% at L3, 10% at H1, 22% at H2, and 22% at H3 (Robin et al., 2014). Our findings are compatible with their report. In addition, their suggestion that the contribution ratio of L2 was 0% strongly supports our results because the number of the direct hydrogen bonds at L2 was the least among all the loops.

4.4. Contribution of amino acid residues at the respective CDR loops to binding free energy

Page 20 of 48

It is meaningful for the rational design of the antibody to clarify how degree its residue contributes to ΔGbind between antigen and antibody. Hence, we calculated ΔGbind of the amino residues at each CDR loop for all the complexes by the decompose method. According to our analysis, the amino residue with the lowest ΔGbind is Tyr in 15 out of 19 complexes except for 1IGC and the Tyr having the lowest ΔGbind mainly appears in the heavy chain in 14 out of 15 complexes. Additionally, it has been reported that the criterion for the significant contribution to make an intermolecular interaction and a stable antigen-antibody complex is ΔGbind ≤ -2.0 kcal/mol per one amino residue (Lee et al., 2010; Ahmad et al., 2009; Huang et al., 2011). Therefore, we examined kind, number, and distribution of amino residues with ΔGbind ≤ -2.0 kcal/mol at the respective CDR loops for all the complexes except for 1IGC (Fig. 8). According to the count of the respective residues in Fig. 8, the number of Tyr is the most among all the loops (Fig. 9A) and Tyr mainly exists in the heavy chain (especially at H2 and H3) (Fig. 9B). As for Tyr and Asn, the correlation between the number of the direct hydrogen bonds and the number of the residues with ΔGbind ≤ -2.0 kcal/mol is observed at all the loops. The correlation between the averaged appearance rate and the number of the amino residues with ΔGbind ≤ -2.0 kcal/mol is, however, scarcely observed. In contrast to Tyr, the contribution of Ser to ΔGbind is very small nevertheless Ser shows the highest averaged appearance rate among 20 kinds of residues at CDR loops (Fig. 9A). Robin et al. reported that Ser hardly contributes to ΔGbind (Robin et al., 2014). Trp and Phe which bears an aromatic ring are frequently observed in the heavy chain (especially at H2 and H3) although the contribution to ΔGbind of these residues is not large. These amino residues will contribute to ΔGbind through π-π and CH-π interaction in addition to the hydrogen bonding. An interesting property was obtained for molecular recognition of Ser and Tyr. Ser has the highest averaged appearance rate among 20 kinds of amino acid residues at CDR. However, the contribution of Ser to ΔGbind is much small in spite that the number of the direct hydrogen bonds between antigen and antibody for Ser are the fourth most residue at CDR loops. On the other hand, Tyr has the largest contribution to the

Page 21 of 48 direct hydrogen bonds and to ΔGbind among all the residues. The importance of these residues has been often suggested from the viewpoint of molecular recognition (Fellouse et al., 2004; Fellouse et al., 2005; Fellouse et al., 2006). The reason why CDR loops in the heavy chain largely contribute to the antigen-antibody interaction is not clear. This point attracts the further interests for understanding antigen-antibody interaction.

4.5. Comparison of scFv and Fab in the binding affinity to antigen In the 20 kind of complexes calculated in this work, 4 complexes are comprised of an antigen and its antibody with the single chain Fv (scFv) form. The antibodies in the other 16 complexes have the Fab form. Fab form is usually more stable than scFv because the contact area between the heavy chain and the light one becomes large. This difference in stability may be reflected in the antigen-antibody binding affinity. Hence, the binding free energy (ΔGbind) and the number of the direct hydrogen bonds were compared between scFv and Fab. The complex 1BGX was excluded in this comparison since the antigen-antibody interaction was outside of CDR. The average number of hydrogen bonds for scFv was 10.3 while that for Fab was 13.8. The average total binding free energy for scFv was -48.1 kcal/mol. In contrast, that for Fab was -70.2 kcal/mol. These results suggest a tendency that Fab has a strong affinity to antigen compared to scFv. The averages and the standard deviations for the number of hydrogen bonds and the binding free energy for all the 19 complexes were 13.1 ± 5.2 and -65.5 ± 31.7 kcal/mol, respectively. Therefore, the differences in values between scFv and Fab are within the standard deviations. Further the number of the calculated complexes was not so large. Accordingly, there is room for further study for the robust conclusion on the stronger affinity of Fab compared to scFv. The number of the residues at H3 loop can largely vary among antibodies. Hence, the relationship between the number of the residues at H3 and the contribution of H3 to the binding free energy was examined as shown in Fig. S24. There was no clear correlation between the loop length and the contribution to the binding energy. In several complexes in which the energetic contribution of H3 was small, the H2

Page 22 of 48

contribution was large. That is, the CDR loops compensated for each other in terms of the contribution to the binding free energy. Therefore, no clear relationship between the length of H3 loop and the binding affinity will be observed for the antigen-antibody complexes other than those in our present study.

5. Conclusions The roles of the respective CDR loops on antibody recognition of antigen were investigated with MD simulations for 20 kinds of antigen-antibody complexes. The following findings were obtained. (i) Ser and Tyr have the highest averaged appearance rates at L1 and at H3 among all CDR loops, respectively. The sum of the averaged appearance rates of Tyr and Ser is approximately 20－30% at each loop. (ii) The number of the direct hydrogen bonds between antigen and antibody in the heavy chain is larger than that in the light chain. Approximately 70% of the total is relevant to the heavy chain. The direct hydrogen bond with the shortest distance in each complex is principally observed in the heavy chain. (iii) The amino residues in the heavy chain (especially at H2 and H3) rather than the light chain energetically contribute to the stable binding of antibody to antigen. (iv) Tyr plays an outstanding role for the antigen-antibody interaction. Tyr in the heavy chain (especially at H2 and H3) primarily contributes to the binding free energy (ΔGbind) compared with other 19 kinds of amino acid residues. Ser scarcely contributes to ΔGbind even if the number of the direct hydrogen bonds with Ser is the fourth highest and also has the highest averaged appearance rate at CDR among 20 kinds of amino acid residues. (v) The energetic contribution of Trp and Phe, each of which bears an aromatic ring, are frequently observed in the heavy chain (especially at H2 and H3) although the contribution of these residues to the binding free energy at CDR loops is not large. Accordingly, the heavy chain of antibody plays an important role in interaction between antigen and antibody. Our results will provide helpful information

Page 23 of 48

for the rational design of a novel antibody with high affinity.

Associated contents Acknowledgments Calculations were performed at Research Center for Computational Science, Okazaki, Japan, and Information Technology Center of the University of Tokyo, and also by the high-performance computer system at Institute for Media Information Technology of Chiba University. A part of this work was supported by Grant-in-Aid for Scientific Research (C) (#15K08458) from Japan Society for the Promotion of Science (JSPS).

Supplementary data Figures S1 – S2 are RMSD changes in the MD simulations. Figures S3 – S23 are binding free energy obtained by MM-GB/SA method and binding free energy and its components with respect to the respective amino acid residues obtained by the decompose method. Figure S24 is the relationship between the length of H3 loop and its contribution to the binding free energy. Table S1 is information on the composition of computational models. Table S2 is averaged RMSD value and the standard deviation. Tables S3 – S22 are details of the direct hydrogen bonds between antigen and antibody for all the complexes. Tables S23 – S42 are details of binding free energy and its components obtained by MM-GB/SA method with respect to the respective loops.

Page 24 of 48

References Acchionea, M., Lipschultza, C., DeSantisa, M., Shanmuganathana, A., Li, M., Wlodawer, A., Tarasov, S., Smith-Gill, S.J., 2009. Light chain somatic mutations change thermodynamics of binding and water coordination in the HyHEL-10 family of antibodies. Mol. Immunol. 47, 457–464. Ahmad, R., Brandsdal, B.O., Michaud-Soret, I., Willassen, N.P., 2009. Ferric uptake regulator protein: Binding free energy calculations and per-residue free energy decomposition. Proteins 75, 373－386. Altshuler, E.P., Serebryanaya, V., Katrukha, A.G., 2010. Generation of recombinant anti-bodies and means for increasing their affinity. Biochemistry (Moscow) 75, 1584–1605. Banerjee, J., 2010. Antibodies are challenged. Indian J. Med. Sci. 64, 144－147. Braden, B.C., Fields, B.A., Ysern, X., Dall’Acqua, W., Goldbaum, F.A., Poljak, R.J., Mariuzza, R.A., 1996. Crystal structure of an Fv–Fv idiotope–anti-idiotope complex at 1.9Å resolution. J. Mol. Biol. 264, 137–151. Brooks, B.R., Bruccoleri, R.E., Olafson, B.D., States, D.J., Swaminathan, S., Karplus, M., 1983. CHARMM: a program for macromolecular energy, minimization, and dynamics calculations. J. Comput. Chem. 4, 187–217. Case, D.A., Cheatham III, T.E., Darden, T., Gohlke, H., Luo, R., Merz, K.M., Jr., Onufriev, A., Simmerling, C., Wang, B., Woods, R., 2005. The Amber biomolecular simulation programs. J. Comput. Chem. 26, 1668–1688. Chavali, G.B., Papageorgiou, A.C., Olson, K.A., Fett, J.W., Hu, G., Shapiro, R., Acharya, K.R., 2003. The crystal structure of human angiogenin in complex with an anti-tumor neutralizing antibody. Structure 11, 875–885. Chen, J., Zhang, D., Zhang, Y., Li, G., 2012. Computational studies of difference in binding modes of peptide and non-peptide inhibitors to MDM2/MDMX based on molecular dynamics simulations. Int. J. Mol. Sci. 13, 2176－2195. Chong, L.T., Duan, Y., Wang, L., Massova, I., Kollman, P.A., 1999. Molecular dynamics and free-energy

Page 25 of 48 calculations applied to affinity maturation in antibody 48G7. PNAS 96, 14330－14335. Cohen, G.H., Silverton, E.W., Padlan, E.A., Dyda, F., Wibbenmeyer, J.A., Willson, R.C., Davies, D.R., 2005. Water molecules in the antibody–antigen interface of the structure of the Fab HyHEL-5–lysozyme complex at 1.7Å resolution: comparison with results from isothermal titration calorimetry. Acta Crystallogr. D: Biol.Crystallogr. 61, 628–633. DeLano, W.L., 2012. The PyMOL Molecular Graphics System. Schrödinger, LLC. (www.pymol.org). Derrick, J.P., Wigley, D.B., 1994. The third IgG-binding domain from streptococcal protein G: an analysis by X-ray crystallography of the structure alone and in a complex with Fab. J. Mol. Biol. 243, 906–918. Desmyter, A., Decanniere, K., Muyldermans, S., Wyns, L., 2001. Antigen specificity and high affinity binding provided by one single loop of a camel single-domain antibody. J. Biol. Chem. 276, 26285–26290. Dimitrov, J.D., Roumenina, L.T., Plantier, J.L., Andre, S., Saboulard, D., Meslier, Y., Planchais, C., Jacquemin, M., Saint-Remy, J.M., Antanasov, B.P., Kaveri, S.V., Lacroix-Desmazes, S., 2010. A human FVIII inhibitor modulates FVIII surface electro-statics at a VWF-binding site distant from its epitope. J. Thromb. Haemost. 8, 1524–1531. Ding, W., Huang, X., Yang, X., Dunn, J.J., Luft, B.J., Koide, S., Awson, C.L., 2000. Structural identification of a key protective B-cell epitope in Lyme disease antigen OspA. J. Mol. Biol. 302, 1153–1164. Długosz, M., Antosiewicz, J.M., Trylska, J., 2009. pH-dependent association of proteins: the test case of monoclonal antibody HyHEL-5 and its antigen hen egg white lysozyme. J. Phys. Chem. B 113, 15662–15669. Edwards, M.J., Dimmock, N.J., 2001. A haemagglutinin (HA1)-specific FAb neutralizes influenza A virus by inhibiting fusion activity. J. Gen. Virol. 82, 1387–1395. Eigenbrot, C., Meng, Y.G., Krishnamurthy, R., Lipari, M.T., Presta, L., Devaux, B., Wong, T., Moran, P., Bullens, S., Kirchhofer, D., 2003. Structural insight into how an anti-idiotypic antibody against D3H44 (anti-tissue factor antibody) restores normal coagulation. J. Mol. Biol. 331, 433–446.

Page 26 of 48 Faelber, K., Kirchhofer, D., Presta, L., Kelley, R.F., Muller, Y.A., 2001. The 1.85Å resolution crystal structures of tissue factor in complex with humanized Fab D3h44 and of free humanized Fab D3h44: revisiting the solvation of antigen combining sites. J. Mol. Biol. 313, 83–97. Fellouse, F.A., Barthelemy, P.A., Kelley, R.F., Sidhu, S.S., 2006. Tyrosine plays a dominant functional role in the paratope of a synthetic antibody derived from a four amino acid code. J. Mol. Biol. 357, 100－114. Fellouse, F.A., Li, B., Compaan, D.M., Peden, A.A., Hymowitz, S.G., Sidhu, S.S., 2005. Molecular recognition by a binary code. J. Mol. Biol. 348, 1153－1162. Fellouse, F.A., Wiesmann, C., Sidhu, S.S., 2004. Synthetic antibodies from a four-amino-acid code: a dominant role for tyrosine in antigen recognition. Proc. Natl. Acad. Sci. U.S.A. 101, 12467－12472. Fleury, D., Wharton, S.A., Skehel, J.J., Knossow, M., Bizebard, T., 1998. Antigen distortion allows influenza virus to escape neutralization. Nat. Struct. Biol. 5, 119–123. Gohlke, H., Case, D.A., 2004. Converging free energy estimates: MM-PB(GB)SA studies on the protein– protein complex Ras–Raf. J. Comput. Chem. 25, 238–250. Gohlke, H., Kiel, C., Case, D.A., 2003. Insights into protein-protein binding by binding free energy calculation and free energy decomposition for the Ras-Raf and Ras-RalGDS complexes. J. Mol. Biol. 330, 891－913. Goldbaum, F.A., Velikovsky, A., Dall’Acqua, W., Fossati, C.A., Fields, B.A., Braden, B.C., Poljak, R.J., Mariuzza, R.A., 1997. Characterization of anti-anti-idiotypic antibodies that bind antigen and an anti-idiotype. Proc. Natl. Acad. Sci. U. S. A. 94, 8697–8701. Hou, T., Guo, S., Xu, X., 2002. Predictions of binding of a diverse set of ligands togelatinase-A by a combination of molecular dynamics and continuum solvent models. J. Phys. Chem. B 106, 5527－5535. Housset, D., Mazza1, G., Grégorire, C., Piras, C., Malissen, B., Fontecilla-Camps, J.C., 1997. The three-dimensional structure of a T-cell antigen receptor VαVβ heterodimer reveals a novel arrangement of the Vβ domain. EMBO J. 16, 4205–4216.

Page 27 of 48 Huang, B., Liu, F.F., Dong, X.Y., Sun, Y., 2011. Molecular mechanism of the affinity interactions between protein A and human immunoglobulin G1 revealed by molecular simulations. J. Phys. Chem. B 115, 4168–4176. Huang, J., Liang, J., Tang, Q., Wang, Z., Chen, L., Zhu, J., Feng, Z., 2011. An active murine–human chimeric Fab antibody derived from Escherichia coli, potential therapy against over-expressing VEGFR2 solid tumors. Appl. Microbiol. Biotechnol. 91, 1341－1351. Huang, M., Furie, B.C., Furie, B., 2004. Crystal structure of the calcium-stabilized human factor IX Gla domain bound to a conformation-specific anti-factor IX antibody. J. Biol. Chem. 279, 14338–14346. Khoury, G.E., Lowe, C.R., 2013. A biomimetic protein G affinity adsorbent: an Ugi ligand for immunoglobulins and Fab fragments based on the third IgG-binding domain of protein G. J. Mol. Recognit. 26, 190–200. Kollman, P.A., Massova, I., Reyes, C., Kuhn, B., Huo, S., Chong, L., Lee, M., Lee, T., Duan, Y., Wang, E., Donini, O., Cleplak, P., Srinivasan, J., Case, D.C., Cheatham III, T.E., 2000. Calculating structures and free energies of complex molecules : combining molecular mechanics and continuum models. Acc. Chem. Res. 33, 889–897. Kortt, A.A., Nice, E., Gruen, L.C., 1999. Analysis of the binding of the Fab fragment of monoclonal antibody NC10 to influenza virus N9 neuraminidase from tern and whale using the BIAcore biosensor: effect of immobilization level and flow rate on kinetic analysis. Anal. Biochem. 273, 133–141. Kringelum, J.V., Nielsen, M., Padkjær, S.B., Lund, O., 2013. Structural analysis of B-cell epitopes in antibody :protein complexes. Mol. Immunol. 53, 24－34. Kumagai, I., Nishimiya, Y., Kondo, H., Tsumoto, K., 2003. Structural consequences of target epitope-directed functional alteration of an antibody: the case of anti-hen lysozyme antibody, HyHEL-10. J. Biol. Chem. 278. 24929–24936. Kumagai, I., Tsumoto, K., 2001. Antigen–Antibody Binding. Encyclopedia of Life Sciences, Nature

Page 28 of 48 Publishing Group, New Jersey (NJ), US, pp. 1–7. (www.els.net). Kunik, V., Ofran, Y., 2013. The indistinguishability of epitopes from protein surface is explained by the distinct binding preferences of each of the six antigen-binding loops. Protein Eng. Des. Sel. 26, 599–609. Lee, J.T., Air, G.M., 2002. Contacts between influenza virus N9 neuraminidase and monoclonal antibody NC10. Virology 300, 255–268. Lee, P., Yoshida, R., Ekiert, D.C., Sakai, N., Suzuki, Y., Takada, A., Wilson, I.A., 2012. Heterosubtypic antibody recognition of the influenza virus hemagglutinin receptor binding site enhanced by avidity. Proc. Natl. Acad. Sci. U. S. A. 109, 17040–17045. Lee, V.S., Tue-ngeun, P., Nangola, S., Kitidee, K., Jitonnom, J., Nimmanpipug, P., Jiranusornkul, S., Tayapiwatana, C., 2010. Pairwise decomposition of residue interaction energies of single chain Fv with HIV-1 p17 epitope variants. Mol. Immunol. 47, 982－990. Li, Y., Li, H., Smith-Gill, S.J., Mariuzza, R.A., 2000. Three-dimensional structures of the free and antigen-bound Fab from monoclonal antilysozyme antibody HyHEL-63. Biochemistry 39, 6296–6309. Li, Y., Li, H., Yang, F., Smith-Gill, S.J., Mariuzza, R.A., 2003. X-ray snapshots of the maturation of an antibody response to a protein antigen. Nat. Struct. Biol. 10, 482–488. Li, Y., Parry, G., Chen, L., Callahan, J.A., Shaw, D.E., Meehan, E.J., Mazar, A.P., Hung, M., 2007. An anti-urokinase plasminogen activator receptor (uPAR) antibody: crystal structure and binding epitope. J. Mol. Biol. 365, 1117–1129. Li, Y., Urrutia, M., Smith-Gill, S.J., Mariuzza, R.A., 2003. Dissection of binding interactions in the complex between the anti-lysozyme antibody HyHEL-63 and its antigen. Biochemistry 42, 11–22. Malbet, S.M., Colominas, C.B., Novelli, A., Battaï, N., Piga, N., Cheynet, V., Mallet, F., Cusack, S., 2000. Mutual conformational adaptations in antigen and antibody upon complex formation between an Fab and HIV-1 capsid protein p24. Structure 8, 1069–1077. Malby, R.L., McCoyl, A.J., Kortt, A.A., Hudson, P.J., Colman, P.M., 1998. Three-dimensional structures of

Page 29 of 48 single-chain Fv-neuraminidase complexes. J. Mol. Biol. 279, 901–910. Mazza, G., Housset, D., Piras, C., Grégorire, C., Fontecilla-Camps, J.C., Malissen, B., 1999. Structural features of the interaction between an anti-clonotypic antibody and its cognate T-cell antigen receptor. J. Mol. Biol. 287, 773–780. Mohan, S., Kourentzi, K., Schick, K.A., Uehara, C., Lipschultz, C.A., Acchione, M., DeSantis, M.E., Smith-Gill, S.J., Willson, R.C., 2009. Association energetics of cross-reactive and specific antibodies. Biochemistry 48, 1390–1398. Mohan, S., Sinha, N., Smith-Gill, S.J., 2003. Modeling the binding sites of anti-hen egg white lysozyme antibodies HyHEL-8 and HyHEL-26: an insight into the molecular basis of antibody cross-reactivity and specificity. Biophys. J. 85, 3221–3236. Momany, C., Kovari, L.C., Prongay, A.J., Keller, W., Gitti, R.K., Lee, B.M., Gorbalenya, A.E., Tong, L., McClure, J., Ehrlich, L.S., Summers, M.F., Carter, C., Rossmann, M.G., 1996. Crystal structure of dimeric HIV-1 capsid protein. Nat. Struct. Biol. 3, 763–770. Murali, R., Sharkey, D.J., Daiss, L., Murthy, H.M., 1998. Crystal structure of Taq DNApolymerase in complex with an inhibitory Fab: the Fab is directed against an intermediate in the helix coil dynamics of the enzyme. Proc. Natl. Acad. Sci. U.S. A. 95, 12562–12567. Nakanishi, T., Tsumoto, K., Yokota, A., Kondo, H., Kumagai, I., 2008. Critical contributionof VH–VL interaction to reshaping of an antibody: the case of humanization of anti-lysozyme antibody, HyHEL-10. Protein Sci. 17, 261–270. Nassal, M., Skamel, C., Kratz, P.A., Wallich, R., Stehle, T., Simon. M.M., 2005. A fusion product of the complete Borrelia burgdorferi outer surface protein A (OspA) and the hepatitis B virus capsid protein is highly immunogenic and induces protective immunity similar to that seen with an effective lipidated OspA vaccine formula. Eur. J. Immunol. 35, 655–665. North, B., Lehmann, A., Jr., Dunbrack, R.L., 2011. A new clustering of antibody CDR loop conformations. J.

Page 30 of 48 Mol. Biol. 406, 228–256. Ode, H., Matsuyama, S., Hata, M., Hoshino, T., Kakizawa, J., Sugiura, W., 2007. Mechanism of drug resistance due to N88S in CRF01_AE HIV-1 protease, analyzed by molecular dynamics simulations. J. Med. Chem. 50, 1768－1777. Ofran, Y., Schlessinger, A., Rost, B., 2008. Automated identification of complementarity determining regions (CDRs) reveals peculiar characteristics of CDRs and B cell epitopes. J. Immunol. 181, 6230－6235. Osajima, T., Suzuki, M., Neya, S., Hoshino, T., 2014. Computational and statistical study on the molecular interaction between antigen and antibody. J. Mol. Graphics Modell. 53, 128－139. Pai, J.C., Sutherland, J.N., Maynard, J.A., 2009. Progress towards recombinant anti-infective antibodies. Recent Pat Antiinfect Drug Discov. 4, 1－17. Prasad, L., Waygood, E.B., Lee, J.S., Delbaere, L.T.J., 1998. The 2.5Å resolution structure of the Jel42 Fab fragment/HPr complex. J. Mol. Biol. 280, 829–845. Reichert, J.M., 2013. Antibodies to watch in 2013. MAbs. 5, 1－5. Robin, G., Sato, Y., Desplancq, D., Rochel, N., Weiss, E., Martineau, P., 2014. Restricted diversity of antigen binding residues of antibodies revealed by computational alanine scanning of 227 antibody-antigen complexes. J. Mol. Biol. 426, 3729－3743. Rodrigues, M.E., Costa, A.R., Henriques, M.J., Azeredo, Oliveira, R., 2010. Technological progresses in monoclonal antibody production systems. Biotechnol. Prog. 26, 332－351. Shi, X., Li, Y., Bian, C., Zhao, G., Huang, M., 2005. Crystallization of an anti-factor IX antibody and its complex. Acta Crystallogr. D: Biol. Crystallogr. 61, 701–703. Sivasakthi, V., Anbarasu, A., Ramaiah, S., 2013. π-π Interactions in structural stability: role in RNA binding proteins. Cell Biochem. Biophys. 67, 853－863. Smallshaw, J.E., Georges, F., Lee, J.S., Waygood, E.B., 1999. Synthesis, cloning and expression of the single-chain Fv gene of the HPr-specific monoclonal antibody, Jel42: determination of binding constants

Page 31 of 48 with wild-type and mutant HPrs. Protein Eng. 12, 623–630. Spiegel, P.C., Jacquemin, M., Saint-Remy, J.R., Stoddard, B.L., Pratt, K.P., 2001. Structure of a factor VIII C2 domain–immunoglobulin G4k Fab complex: identification of an inhibitory antibody epitope on the surface of factor VIII. Blood 98, 13–19. Yokota, A., Tsumoto, K., Shiroishi, M., Nakanishi, T., Kondo, H., Kumagai, I., 2010. Contribution of asparagine residues to the stabilization of a proteinaceous antigen–antibody complex, HyHEL-10-hen egg white lysozyme. J. Biol. Chem. 285, 7686–7696. Yuki, H., Hata, Y., Ishikawa, M., Neya, S., Hoshino, T., 2007. Implementation of π-π interactions in molecular dynamics simulation. J. Comput. Chem. 28, 1091－1099.

Page 32 of 48

Figure Captions Figure 1: Computational average structures for 20 kinds of antigen-antibody complexes. The average structure was obtained from the last 2 ns trajectory of 10 ns MD simulation. The antigen is colored yellow. The heavy and light chains of the antibody are shown in blue and green, respectively.

Figure 2: Position of the respective CDR loops of antibodies for 20 complexes.

Figure 3: Averaged appearance rates of amino acid residues at the respective CDR loops of antibodies for all the complexes. (A) Non-polar residues. (B) Uncharged polar residues. (C) Charged polar residues.

Figure 4: (A) Number of the direct hydrogen bonds between antigen and antibody for 20 complexes. (B) Number of amino acid residues at CDR related to the direct hydrogen bonds between antigen and antibody for all the complexes. (C) Number of the direct hydrogen bonds at the respective CDR loops of antibodies for 20 complexes, (D) Distribution of the direct hydrogen bonds at 6 CDR loops of antibodies for all the complexes.

Figure 5: Distribution of the lowest energy in 6 CDR loops for all the complexes. (A) Electrostatic energy. (B) van der Waals energy. (C) Binding free energy. The term “other” in pie graph means 1IGC complex in which the antigen is bound to the outside domain of CDR.

Figure 6: Binding free energy of each amino acid residue of antibodies obtained by the MM-GB/SA decompose method.

Figure 7: Sum of the numbers of the direct hydrogen bonds at the respective CDR loops of antibodies, summarized in terms of kinds of amino acid residues.

Page 33 of 48 Figure 8: Sum of the numbers of the residues with ΔGbind below -2.0 kcal/mol at the respective CDR loops with regard to 20 kinds of amino acid residues for all the complexes except for 1IGC.

Figure 9: (A) Number of the amino acid residues with ΔGbind below -2.0 kcal/mol at CDR loops of antibodies for all the complexes except for 1IGC. (B) Distribution of the residues with ΔGbind below -2.0 kcal/mol at CDR loops for all the complexes except for 1IGC.

Page 34 of 48

Figure 1

Page 35 of 48

Figure 2-1

Page 36 of 48

Figure 2-2

Page 37 of 48

Figure 3

Page 38 of 48

Figure 4

Page 39 of 48

Figure 5

Page 40 of 48

Figure 6-1

Page 41 of 48

Figure 6-2

Page 42 of 48

Figure 7

Page 43 of 48

Figure 8

Page 44 of 48

Figure 9

Page 45 of 48 Tables Table 1 Direct hydrogen bonds with the shortest distance between antigen and antibody for 20 complexes PDB code

CDR

Distance (Å)

Angle (degree)

1A14

H2

Asp447

OD2

Ser290

HG

2.66

6.0

1AFV

H3

Ser519

HG

Glu75

O

2.62

4.7

1BGX

H2

Lys1104

HZ2

Glu388

OE1

2.68

28.3

1DQJ

H2

Ser267

HG

Asp527

OD2

2.58

14.4

1DVF

H1

Asp364

OD1

Arg207

2HH2

2.63

18.2

1DVF

Non

Tyr274

HH

Asp162

OD1

2.63

6.9

1FJ1

Non

Hie49

HE2

Ser613

O

2.70

23.2

1H0D

L1

Asn31

2HD2

Ser478

O

2.61

15.7

1IGC

Non

Pro340

O

Tyr473

HH

2.66

15.6

1IQD

H3

Asp312

OD2

Ser506

HG

2.63

6.5

1JPS

H1

Lys244

O

Lys631

HZ3

2.61

32.2

1KB5

H1

Tyr481

OH

Asp26

OD2

2.62

2.7

1NDG

H2

Ser269

HG

Asp527

OD1

2.61

10.2

1NDM

H2

Ser267

HG

Asp527

OD2

2.60

20.2

1NL0

H2

Tyr267

H

Phe449

O

2.90

14.9

1UA6

H2

Ser160

HG

Asp324

OD1

2.43

5.3

2JEL

H2

Ser273

HG

Glu513

OE1

2.76

19.0

2VIR

H2

Asn267

OD1

Ser549

HG

2.61

13.0

2VIS

H2

Gly265

H

Ser549

OG

2.77

24.2

2VIT

H1

Ser242

HG

Gln524

O

2.70

6.1

Antibody

Antigen

Page 46 of 48 3HFL

H1

Asp244

OD2

Ser510

HG

2.56

OD2 and its components obtained by the MM-GB/SA method Table 2 Binding free energy PDB code

ΔHele Ser510 HG (kcal/mol)

ΔHvdW (kcal/mol)

ΔGsol (kcal/mol)

ΔGbind (kcal/mol)

1A14

2.56 8.1 -64.6 ± 13.7

-69.8 ± 5.0

97.1 ± 12.4

-37.4 ± 4.4

1AFV

-139.9 ± 22.0

-49.5 ± 3.8

163.6 ± 19.1

-25.8 ± 4.8

1BGX

-603.2 ± 29.4

-290.3 ± 7.6

726.0 ± 24.3

-167.5 ± 9.0

1DQJ

-269.4 ± 23.1

-102.3 ± 4.9

298.6 ± 21.3

-73.1 ± 5.5

1DVF

-146.9 ± 15.3

-101.3 ± 4.5

174.0 ± 12.6

-74.2 ± 5.4

1FJ1

-109.1 ± 21.0

-86.4 ± 4.1

135.5 ± 17.9

-60.0 ± 5.0

1H0D

-138.7 ± 14.5

-80.7 ± 3.8

159.8 ± 13.6

-59.6 ± 3.2

1IGC

-187.23 ± 14.0

-69.3 ± 4.5

197.66 ± 13.0

-58.9 ± 3.6

1IQD

-674.0 ± 31.3

-110.2 ± 5.6

679.2 ± 29.8

-105.0 ± 4.9

1JPS

-444.3 ± 30.5

-115.6 ± 4.8

486.7 ± 27.7

-73.2 ± 6.3

1KB5

40.5 ± 37.5

-118.8 ± 6.8

6.5 ± 33.4

-71.8 ± 9.8

1NDG

-274.5 ± 18.4

-96.0 ± 4.2

299.4 ± 15.5

-71.0 ± 5.0

1NDM

-270.5 ± 11.4

-100.2 ± 4.8

308.8 ± 11.8

-61.9 ± 4.2

1NL0

-1845.9 ± 60.2

-63.5 ± 3.8

1857.9 ± 57.8

-51.5 ± 3.6

1UA6

-287.8 ± 18.6

-85.3 ± 4.4

318.3 ± 16.7

-54.8 ± 4.9

2JEL

-349.4 ± 21.5

-81.9 ± 4.3

384.7 ± 20.4

-46.6 ± 4.1

2VIR

-158.2 ± 26.7

-67.7 ± 4.2

188.1 ± 24.3

-37.8 ± 5.4

2VIS

-119.3 ± 16.4

-72.4 ± 3.7

149.9 ± 16.2

-41.8 ± 3.4

2VIT

-131.9 ± 13.3

-68.0 ± 3.9

159.2 ± 11.1

-40.7 ± 3.7

3HFL

-346.7 ± 21.8

-100.0 ± 6.8

355.7 ± 18.9

-90.9 ± 6.5

Mean ± standard deviation. ΔHele; electrostatic energy, ΔHvdW; van der Waals energy, ΔGsol;

8.1

Page 47 of 48 solvation energy, ΔGbind; binding free energy.

Page 48 of 48 Table 3 Binding free energy obtained by the MM-GB/SA method for the respective CDR loops of antibodies ΔGbind(kcal/mol)

PDB code L1

L2

L3

H1

H2

H3

1A14

-3.6 ± 1.8

-1.1 ± 0.2

-11.9 ± 1.4

2.9 ± 0.3

-12.9 ± 2.7

-6.6 ± 1.5

1AFV

-1.1 ± 1.4

0.3 ± 0.1

-4.9 ± 1.1

0.3 ± 0.6

-5.5 ± 1.6

-5.7 ± 2.8

1BGX

-7.5 ± 2.1

-2.7 ± 1.7

-10.6 ± 2.0

6.9 ± 0.7

-37.3 ± 3.8

-18.7 ± 2.5

1DQJ

-6.3 ± 1.7

-1.0 ± 1.7

-10.3 ± 1.5

-14.9 ± 2.1

-19.7 ± 2.8

-5.5 ± 1.0

1DVF

1.5 ± 0.6

2.9 ± 0.4

-4.9 ± 1.4

-1.3 ± 2.0

-6.0 ± 1.9

-49.5 ± 3.4

1FJ1

-1.4 ± 0.9

-8.7 ± 1.8

-0.5 ± 1.0

-12.2 ± 2.5

-4.9 ± 2.7

-10.0 ± 1.6

1H0D

-12.6 ± 1.2

1.1 ± 0.1

-10.0 ± 0.3

0.4 ± 0.6

-12.4 ± 2.3

-21.0 ± 1.7

1IGC

0.0 ± 0.0

-0.1 ± 0.1

0.0 ± 0.0

0.0 ± 0.0

0.0 ± 0.0

0.0 ± 0.0

1IQD

-10.2 ± 2.0

2.2 ± 1.3

-9.0 ± 1.4

-10.0 ± 1.8

-17.0 ± 1.6

-37.0 ± 3.5

1JPS

-0.5 ± 2.0

1.4 ± 0.7

-16.3 ± 1.9

-18.3 ± 2.1

-11.3 ± 3.4

-10.2 ± 1.8

1KB5

-9.4 ± 1.7

3.6 ± 0.7

-6.6 ± 3.0

-0.5 ± 2.5

-20.0 ± 2.6

-20.5 ± 3.1

1NDG

-5.5 ± 1.9

1.0 ± 1.7

-8.7 ± 1.6

-18.2 ± 2.1

-18.7 ± 2.2

-7.2 ± 2.0

1NDM

-4.6 ± 1.8

-0.6 ± 1.8

-7.0 ± 1.6

-12.7 ± 2.0

-20.2 ± 3.1

-5.1 ± 1.9

1NL0

1.8 ± 0.2

-2.1 ± 0.2

5.0 ± 1.3

-5.6 ± 1.1

-17.9 ± 4.2

-9.4 ± 1.1

1UA6

-6.2 ± 2.0

-1.0 ± 1.9

-6.5 ± 1.6

-10.2 ± 1.5

-15.6 ± 2.3

-5.8 ± 1.4

2JEL

-6.8 ± 2.7

-2.1 ± 0.7

-0.2 ± 1.4

-0.1 ± 1.2

-17.5 ± 2.0

-6.4 ± 1.3

2VIR

1.1 ± 0.1

1.6 ± 0.4

0.8 ± 0.3

-3.3 ± 1.8

-14.9 ± 1.8

-15.9 ± 5.0

2VIS

1.2 ± 0.1

1.8 ± 0.3

0.9 ± 0.3

-5.4 ± 1.6

-14.6 ± 2.0

-19.1 ± 2.2

2VIT

1.1 ± 0.1

1.6 ± 0.1

0.0 ± 1.4

-5.1 ± 1.4

-16.8 ± 1.9

-14.2 ± 2.2

3HFL

-8.0 ± 2.4

2.9 ± 0.8

-17.5 ± 2.8

-13.4 ± 2.2

-22.5 ± 4.4

-6.3 ± 1.9

Mean ± standard deviation. ΔGbind; binding free energy.

Page 49 of 48 Table 4 Lowest binding free energy obtained by the MM-GB/SA decompose method among the residues at CDR loops of antibodies except for 1IGC PDB code

Loop

Residue

ΔGbind(kcal/mol)

1A14

L3

Phe603

-3.7

1A14

H2

Asn445

-3.7

1A14

H3

Tyr495

-3.7

1AFV

L1

Tyr336

-3.5

1BGX

H3

Trp1143

-6.8

1DQJ

H1

Tyr248

-5.4

1DVF

H3

Tyr435

-11.8

1FJ1

H1

Tyr246

-5.3

1H0D

H3

Tyr323

-5.5

1IQD

H3

Pro313

-4.3

1JPS

H1

Tyr247

-5.1

1KB5

H2

Tyr504

-5.6

1NDG

H1

Tyr248

-4.7

1NDG

H3

Trp313

-4.7

1NDM

H1

Tyr248

-5.3

1NL0

H2

Lys271

-7.0

1UA6

H1

Tyr141

-4.5

2JEL

H2

Tyr272

-4.7

2VIR

H3

Tyr313

-5.7

2VIS

H3

Tyr313

-7.0

2VIT

H3

Tyr313

-5.2

3HFL

L3

Trp90

-6.3

ΔGbind; binding free energy.