Forty crystal structures of Fab fragments were retrieved from the PDB and organized into couples, so that free antibody could be compared to the hapten antigen-bound form. We included all structures with acceptable resolution ( ≲ 3 °Å) in the analysis to allow a confident classification of the surface topography and structural changes (McDonald and Thornton, 1994; Almagro, 2004; Raghunathan et al., 2012)^. Antibody structures were considered similar if their chains (heavy and light) shared 100% sequence identity, and had similar sequence to the antigen bound form. The effect of crystal contacts in the PDB structures used in this analysis were considered negligible, because the structural changes were analyzed in terms of backbone movements rather than the side-chains that are most likely affected by crystal formation. Therefore, the assumption is made that the crystal contacts in the PDB structures do not strongly affect the backbone structure; this assumption can be explored further in future work.

Amino acid positions were sequentially numbered to avoid conflicts at residue positions, and to ensure accurate structural measurements. A few sequences were omitted from the study at different stages (such as 1JNH, 1JN6, and 1JGL) due to the fact that these polypeptide chains were not continuous. In total, 40 crystal structures (32 mouse, four fully human, and four chimeric with human C_H1 domain) were identified and selected; all the potential sequences in the PDB were considered, so that the analysis is comprehensive within the stated selection criteria. A few of the selected antibodies were crystalized against more than one antigen, and full details of these crystal structures are summarized in Supplementary Tables S1.1.

The Fab sequences were retrieved from the PDB, and analyzed using BioEdit Sequence Alignment Editor, version 7.2.5 (Hall, 1999). ClustalW Multiple alignment was used to align sequences of the same type (heavy or light chains) and origin (human or mouse). All the gathered and analyzed sequences are listed in Supplementary Tables S3. The three loops in the variable domains were denoted CDRs 1–3, and the three loops in the constant domains (CH and CL) were named C_Loop 1–3, according to their location throughout the sequence from the N-terminal to the C-terminal of the entire chain.

The structural variability of similar proteins or different conformations of the same protein can be quantified using RMSD (Carugo and Pongor, 2001).

Here the RMSD is defined as R M S D = ∑ i = 1 N | r → i , A − r → i , F | 2 N (1) where N is the number of (backbone) atoms in the protein structure and r → i , A , r → i , F is the position of the ith atom in the antigen-bound (A) and free (F) Fab structure. In order to calculate the RMSD, the two protein structures to be examined are firstly considered as rigid bodies, then overlapped (aligned) using only translations and rotations. The RMSDs of the backbone atoms were only calculated for particular regions, such as variable (residues 1–105), linker (106–113) and constant (114–218) domains, as well as each couple for the entire antibody chain (residues 1–218). The side chains of residues were excluded because insignificant differences in their orientations could introduce additional noise.

RMSF is another useful tool to track structural changes in proteins, where the RMSD is calculated and reported for each protein residue in the compared structure. It is regularly denoted as “fluctuations” as it calculates a time-average amplitude of residue movement (fluctuation) from each compared position in the aligned structures. Despite the wide adaptation of RMSF in dynamic structures, it can be used in static structures as it provides RMSDs calculated per particular residue. With this form of RMSF, it is easy to track the residues that are most responsible for the total conformational alterations, the creation of pockets and, for potential allosteric signaling. Therefore, following Eq. 1 we define RMSF in Eq. 2, and where symbols have the same meaning as before. R M S F i = | r → i , A − r → i , F | (2)

The orientation of the V_H/V_L domains vs. C_H1/C_L domains was examined to complement the RMSD/RMSF measurements. One conserved cysteine residue (Cys) was chosen in each of the V_L, C_L, V_H, and C_H1 domains (See Domain Orientation). Moreover, one conserved amino acid in each of the V_L/C_L and V_H/C_H1 linkers was also selected. These amino acids were serine (Ser), arginine (Arg), and glutamine (Gln) in the heavy, light (K), and light (λ) chains, respectively (Figure 1). The six amino acids selected in each Fab are highlighted in turquoise in the full sequences file (Supplementary S3). The angles of the heavy (Cys-Ser-Cys) and light (Cys-Gln-Cys or Cys-Arg-Cys) chains were measure by PyMOL (The PyMOL Molecular Graphics System, Version 1.7.4 Schrödinger, LLC.). The distances between the Cys residues in the heavy and light chains were also measured by PyMOL.

Analysis of anti-hapten Fabs can provide information about the changes in the binding surfaces upon antigen recognition. Anti-hapten antibodies have shown preference to recognize their targeted antigens through the formation of pockets in their binding sites. Comparison of the free vs. antigen-bound crystal structures Fabs (23 couples) have shown three binding surface (S1–3) trends:

• S1: the presence of a pocket binding site on the antibody-antigen complex but not on the antibody-free counterparts, as can be observed in three mouse Fab couples (1Q0Y_1Q0X, 3CFB_3CFC, and 1I9J_1I9I) and two human Fab couples (1D6V_1D5B and 1AJ7_2RCS). Illustration of these figures can be observed in Supplementary Figures S2.1, S2.10, S2.18, S2.22, and S2.23.

An example from each of the three groups is demonstrated in Figure 2, and detailed observations are shown in Supplementary Figure S2. The antibody (3CFB; Figure 2A) can bind the antigen [4-(4-styryl-phenylcarbamoyl)-butyric acid] through side chain of amino acid Q96, and CDR H3 of the bound form has slightly moved inwards to support the pocket formation, as illustrated in Supplementary Figure S.2.1F. In contrast, the antigen (cocaine) can bind the antibody (1Q72; Figure 2B) through residue Y32 of the light chain, and CDR H3 of the bound form has moved away from the binding site center to cause an opening of the binding pocket (Supplementary Figure S.2.4F). Whilst, the third example was demonstrated by the antibody (4COX; Figure 2C) that can bind the antigen (methotrexate) through two amino acids at the light chain (E39 and Y41) and one amino acid at the heavy chain (N36) (Supplementary Figure S.2.17F). The three residues are located at the bottom of the pocket, and no major changes were noticed by comparing the antigen-bound verses the free form of the antibody.

Antibodies can generally bind antigens through multiple conformational states rather than a rigid lock and key method. Although structural changes between free and antigen-bound antibodies were usually correlated with an induced-fit mechanism, structural evidence for a pre-existing equilibrium was also identified (James et al., 2003). This was evident in analyzing anti-IL1β and anti-IL6 Fabs and scFvs that have shown the core of the antigen binding sites, in particular the CDR3 loops, to exist in multiple conformational states (Addis et al., 2014). This behavior seems to be a conserved feature of many high affinity protein-protein interaction sites, suggesting a key role in the formation of tight protein complexes (Waters et al., 2007; Veverka et al., 2009; Holdsworth et al., 2012). Consequently, slight structural changes between the free and bound structures would be expected if the free conformation that was crystallized is the same one bound to the antigen. These slight changes were observed in group S3 (15 couples), supporting this picture.

On the other hand, significant changes are expected if a different pre-existing antigen-unbound conformation was crystallized. These structural changes could reflect the anti-hapten Fabs of groups S1 (5 couples) and S2 (3 couples). A distinction between these two groups can be achieved through analysis of their RMSD/RMSF differences, and heavy and light chain packing (see below).

The conformational changes were determined by calculating the total RMSD difference between the antibody-antigen complex and the antibody-free counterparts. Out of the analyzed 23 couples, seven have shown more deviations in the light chains when compared to heavy chains, and the opposite was observed in the remaining 16 couples (Table 1). In addition, the constant domains (C_H1 and C_L) of these Fabs have demonstrated more movements than the variable domains (V_H and V_L), as reported in Table 1; and the quantitative information is summarized in Supplementary Table S1.2.

The high movements of the constant domains relative to the variable domains and to each other can increase the likelihood of a suitable V_H-V_L relative orientation and signal propagation from the variable to constant domains (Padlan, 1994; Wilson and Stanfield, 1994; Braden and Poljak, 1995). For instance, these movements were reflected in the complement activation of two antibodies that differed only in their V_H domain (Horgan et al., 1992). In other work, two human monoclonal antibodies, having similar variable domains but different constant domains, were shown to bind their target with significantly different affinities (Pritsch et al., 1996). Furthermore, a comparison of the binding affinity of Fabs to smaller variable fragments (scFv) by Adachi et al. also revealed a large difference in affinity (Adachi et al., 2003). We have noticed these movements through analysis of an entire antibody in molecular dynamics simulations (Al Qaraghuli et al., 2018), and in a comprehensive analysis of anti-protein antibodies (Al Qaraghuli et al., 2020). So our findings suggest that the high movement of the constant domains could be a reflection of essential structural movements that can help trigger the immune response. These movements are expected to be greater in the heavy chain when compared to light chains, since the Fab-to-Fc movements occur in these heavy chains and through the main Fab-Fc linker.

The RMSF of each amino acid position in the 40 crystal structures, which represent 23 free antibody and antigen-complexed couples, were also analyzed. In general, four different classifications (B1–4) for structural change upon binding were identified, following the classification scheme identified with anti-protein Fabs (Al Qaraghuli et al., 2020):

• B1: Random movements throughout the entire Fab illustrated by five anti-hapten couples (1NGP_1NGQ, 2CGR_1CGS, 1D6V_1D5B, 1AJ7_2RCS, and 1FL6_1FL5).

Examples from each trend were selected from the anti-hapten Fabs, as illustrated in Figure 3, and Supplementary Figure S2. The RMSF was calculated for the entire heavy and light chains, and also separately for their specific domains (VH, CH, VL, and CL) to separate out contributions from the domain orientation changes reported below.

These results can reflect changes in the constant domains as well as the variable domains. These changes, however, were either random, with big (B1) or small (B4) movements, or at specific positions (CDRs and/or C_Loop1 in classes B3/B2). We note that this range of structural change agrees with our previous analysis of protein-binding Fabs, with the addition of B4 which is unique to certain anti-hapten Fabs.

Movements at the CDRs are expected to reflect the antigen binding process. Furthermore, Sela-Culang et al. (2012) have shown that the C_Loop₁ was the highest moving loop, which in B2 has moved even more than CDRH3 in B3. Various studies have correlated the C_Loop1, which is located at C_H1 domain, to detected changes in binding, because it was recognized as the only region with sequence diversity among the examined antibodies (Pritsch et al., 1996; Pritsch et al., 2000; Torres et al., 2007). Furthermore, the C_Loop₁ was suggested to be intrinsically disordered, associated in the interaction between the heavy and light chains, and linked to complement binding (Sela-Culang et al., 2012). We note that this loop is close to the main hinge region, and a conformational change in this loop may affect the relative orientation of the C_H1 domain vs. the entire Fc, thus influencing the overall effector function of the Fc region (Patil et al., 2010; Sela-Culang et al., 2012). Therefore, structural changes in the C_Loop₁ (in the Fab region) could stimulate changes in the Fc region to facilitate triggering of the immune response. The random movements of B1 and B4 can be better understood through an analysis of the various domain orientations, as will be discussed in the following section.

The heavy and light chain angles (Figure 1) of the anti-hapten Fabs have shown large changes in class B1, as observed in 1NGP/1NGQ, 2CGR/1CGS, 1FL6/1FL5, 1D6V/1D5B and 1AJ7/2RCS (Table 2 and Supplementary Table S4). These large changes originate from a reduction in the angles between the heavy chains and an increase between the light chains, and are in line with the distance changes between the selected Cys residues on each domain. The linker-linker distance of this class has also increased in the bound form when compared to their free counterparts. The small angle changes of class B4 were also reflected in the relatively low changes in the Cys-Cys and linker-linker distances. Likewise, both class B3 (with large CDR movement) and class B2 (large C_Loop1 movement) have shown relatively low angle changes, and this can be observed in the remaining crystal structure pairs (Supplementary Table 4).

Binding Surface class S2 (with a pocket binding site on the antibody-free structure, but not on the antibody-antigen complex) is associated with B3 (with CDR changes only). In addition, none of class B3 have shown binding surface class S1 (pockets on the antibody-antigen complex, but no pocket on the free antibody form). Therefore, these results could suggest that large CDR movements are associated with the closure of the pockets in the antibody-antigen complexes, and not involved in the opening of these pockets if the antibody-free structure displays a non-pocket binding site.

The presence of pockets in the antibody-antigen complexes, but not in the free form (S1) or the presence of pockets on both the free and bound forms (S3) were associated with either classes B1 or B4 (random movements) or class B2 (C_Loop1 specific movements). Consequently, the presence of a pocket in the binding sites of the antibody-bound form could be correlated to movements in C_Loop1 and/or with high movements in the entire antibody. However, the high movement of C_Loop1, and the generation of these pockets in the antibody-bound form could also be a result of an induced-fit mechanism or by binding to one of the pre-existing free conformations.

Taken together, these data illustrate the complexity of the antibody-antigen interaction process that could be restricted to the binding sites, or extended through the constant domains to propagate structural changes beyond these sites. In order to further illustrate these concepts, we now examine and discuss one specific example in detail.

We analyzed the antibody 7A1 and compared it to its set of antigen-bound forms. This antibody hydrolyses cocaine to produce the non-psychoactive metabolites ecgonine methyl ester and benzoic acid. Crystal structures of 7A1 Fab (2AJU) and six complexes with substrate cocaine (2AJV), the transition state analogue (2AJX), the products ecgonine methyl ester and benzoic acid together (2AJY) and individually (2AJZ and 2AK1), as well as with heptaethylene glycol (2AJS), have all been resolved by Zhu et al. (2006) and are all available from the PDB. These authors identified significant structural rearrangements that occur along the reaction pathway, which they proposed to be limited to the binding site (Zhu et al., 2006). This suggestion was based on identifying modest CDR loop movements (up to 2.3 Å), along with substantial side chain rearrangements (up to 9 Å) that have altered the shape and size (∼320–500 Å³) of the antibody active site from open (for the substrate) to closed (transition state) to open (product states) (Zhu et al., 2006).

Antibody 7A1 is IgG2a-Kappa, and the binding sites transition from open (2AJV) to closed (2AJX) to open (2AJZ, 2AJY, and 2AK1) can be observed in Figure 4. This closure can be attributed to the twisting of the benzene ring of the transitional state antigen of 2AJX when compared to cocaine in 2AJV. The cocaine bound crystal structure (2AJU) also shows closure of a very small binding pocket in the center of the binding sites of the remaining free and bound crystal structures (Figure 4). Zhu et al. have crystalized antibody 7A1 in different antigen-bound forms, and comprehensively reported these atomistic interactions (Zhu et al., 2006).

RMSD comparison of the bound vs. free forms has confirmed suggestions by Zhu et al. that the VH and VL domains have moved more than the C_H1 and C_L counterparts. In addition, the light chains of 2AJY, 2AK1, and 2AJV have moved more than the heavy chains (Supplementary Table S1.2). Therefore, this movement could be attributed to the binding of ecgonine methyl ester to the light chains, and the benzene ring to the heavy chain, which cause further light chains movements.

Despite the high movement of the variable domains (V_H and V_L), the RMSF analyses have also shown additional movement of the constant domains (C_H1 and C_L) (Figure 5). 2AJV has shown specific movement of CDRL1, CDRL3 and an area next to CDRH1 to accommodate cocaine through an open binding state (Figure 5). Whilst in the transition state of 2AJX, the active site demonstrates a closed form, in which the CDRH2 progress toward the active site (Figure 5 and Supplementary Figure S.2.1.13). The remaining five antigen-bound states have demonstrated general fluctuation with greater focus on the loops (Figure 5). An important region for structural analysis is C_Loop1 of the heavy chain. The movement of this loop was between 1 and 2 Å in all bound states except the main anti-cocaine antibody 2AJV (∼0.55 Å). This reduced movement of C_Loop1 in 2AJV could be correlated with the disappearance of the small pocket at the center of the groove binding site (Figure 4). We note that the scale of the C_Loop1 changes is much lower than found in classes B1 and B2 where clear evidence of signal propagation is found. Indeed, all these crystal structures have shown class B3 structural changes, and S3 surfaces apart from 2AJV (class S2), which is the crystal structure of the cocaine bound antibody. These structural changes match the function of this antibody in hydrolyzing cocaine rather than propagation of an immune response.