ABSTRACT
In creating vaccines against infectious agents, there is often a desire to direct an immune response toward a particular conformational epitope on an antigen. We present a method, called Protect, Modify, Deprotect (PMD), to generate immunogenic proteins aimed to direct a vaccine-induced antibody response toward an epitope defined by a specific monoclonal antibody (mAb). The mAb is used to protect the target epitope on the protein. Then, the remaining exposed surfaces of the protein are modified to render them non-immunogenic. Finally, the epitope is deprotected by removal of the mAb. The resultant protein is modified at surfaces other than the target epitope. We validate PMD using the well-characterized antigen, hen egg white lysozyme (HEWL). Then, we demonstrate the utility of PMD using influenza virus hemagglutinin (HA). Specifically, we use a mAb to protect a highly conserved epitope on the stem domain of HA. Exposed surface amines are then modified by introducing short polyethylene glycol (PEG) chains. The resultant antigen shows markedly reduced binding to mAbs that target the variable head region of HA, while maintaining binding to mAbs at the epitope of interest in the stem region. This antigenic preference is also observed with yeast cells displaying antibody fragments. Antisera from guinea pigs immunized with the PMD-modified HA show increased cross-reactivity with HAs from other influenza strains, as compared to antisera obtained with unmodified HA trimers. PMD has the potential to direct an antibody response at high-resolution and could be used in combination with other such strategies. There are many attractive targets for the application of PMD.
INTRODUCTION
Vaccines are among the most profound accomplishments of biomedical science and provide cost-effective protection against infectious disease. Many vaccines work by eliciting a neutralizing antibody response that prevents infection (1, 2). However, for some infectious agents it has not been possible to create an efficacious vaccine and, for others, the protection provided by vaccines is strain-specific.
In the case of influenza, the majority of antibodies elicited by vaccination target the trimeric viral surface glycoprotein, hemagglutinin (HA) (3–5). The three-dimensional structure of HA consists of two regions, the head and the stem (6). Most of the HA-directed antibody response focuses on the head region, which is therefore considered immunodominant (3–5, 7). Amino acid residues on the surface of this immunodominant head region vary substantially among different strains and change continuously in a phenomenon referred to as antigenic drift (8, 9). This variability, which leads to new circulating virus strains, coupled with the immunodominance of the head region, necessitates the production of new seasonal vaccines against influenza (10, 11).
Strikingly, there is an epitope within the stem region of HA that is highly conserved among different influenza strains and not subject to seasonal variation (12), likely because residues that form this epitope are critical for the viral fusion mechanism mediated by HA (13, 14). There is not a significant immune response towards the stem region during infection. Nonetheless, Okuno and coworkers (15) isolated a monoclonal antibody (mAb) that targets this conserved epitope and demonstrated that it had broad neutralizing activity. Since the discovery of this broadly neutralizing antibody (bnAb) 25 years ago (15), many other HA stem-binding bnAbs have been characterized (3, 16–22). In addition, expression of such bnAbs protects mice from lethal challenges with a broad range of influenza subtypes (23). Taken together, these results suggest that if antibodies targeting the conserved stem epitope could be elicited it might be possible to create a universal flu vaccine (9, 24–29). Such a vaccine might provide cross-strain protection against all circulating strains of influenza, as well as against future pandemic influenza strains (i.e., new strains transmitted to humans from another animal, such as those that led to the 1918, 1957, 1968 and 2009 pandemics) (30).
Towards this goal, there has been substantial interest in directing a vaccine-induced antibody response toward the conserved stem region of influenza HA (31, 32). This would require avoiding the normal, immunodominant antibody response against the head (33). Strategies that aim to direct the immune system towards a particular region of a protein are referred to as “immunofocusing” (34).
Previous immunofocusing work, either against influenza or other infectious agents, has utilized a variety of approaches. The five most prominent examples are (i) epitope masking (31—43), (ii) epitope scaffolding (48–53), (iii) protein dissection (54–57), (iv) antigen resurfacing (58–60), and (v) cross-strain boosting (22, 61–64). Epitope masking is a method in which an immunodominant region of a protein is shielded, often using unnatural glycosylation sites, to discourage antibody formation. Epitope scaffolding aims to transplant a conformational epitope of interest onto a unique protein scaffold. Protein dissection removes undesirable or immunodominant epitopes from the native antigen. Antigen resurfacing utilizes site-directed mutagenesis to install less immunogenic residues at regions outside the epitope of interest. Finally, cross-strain boosting employs sequential immunizations with other strains or chimeric proteins that vary at off-target epitopes.
Significant progress has been made with these immunofocusing strategies. These methods, however, have inherent limitations. They are not easily generalizable, making it challenging to apply them to new antigens. With the exception of epitope scaffolding (which requires extensive protein engineering) these immunofocusing methods are also generally ‘low-resolution’ (i.e., directed toward a region of the protein that is significantly larger than a typical antibody epitope). Moreover, it can be challenging with some of these methods to maintain the precise three-dimensional structure of the epitope.
Here, we introduce a method that has the potential to provide high-resolution immmunofocusing, in a generalizable manner, with minimal protein engineering. The method utilizes a bnAb as a molecular stencil to generate an antigen aimed at focusing the immune response toward the bnAb epitope. Although bnAbs have been used previously to inform and guide immunogen design, we are not aware of their use as reagents in the creation of vaccine candidates.
We refer to the method as ‘Protect, Modify, Deprotect’ (PMD). The steps for PMD are: (1) protection of an epitope on a particular antigen by binding of a bnAb, (2) chemical modification of exposed sites to render them non-immunogenic, and (3) deprotection of the epitope of interest by dissociation of the antibody-antigen complex. This produces an immunogen where the only unmodified region is the epitope mapped by the bnAb (Figure 1).
To establish the PMD method, we use hen egg white lysozyme (HEWL) because it is a stable, monomeric protein with well-characterized epitopes (65, 66). We protect an epitope on HEWL by binding it to a mAb-conjugated resin (67). Then, we modify surface amines to add short polyethylene glycol (PEG) chains, which are known to decrease immunogenicity locally (64, 68–71). The modified HEWL derivatives, isolated following dissociation from the mAb resin, have antigenic properties consistent with those expected based on the location of surface amines in antibody co-crystal structures.
We then use PMD to generate an influenza HA antigen designed to skew the immune response toward a conserved epitope on the stem. We confirm that the PMD-generated HA is properly folded based on biophysical studies of the protein and binding to conformation-specific mAbs. The PMD-generated HA displays markedly reduced binding to mAbs that target the HA head, while maintaining binding to mAbs that target the stem. We also use the PMD-generated HA as bait in fluorescence activated cell sorting (FACS) experiments with a polyclonal yeast mini-library displaying scFvs and obtain significant enrichment for stem-directed clones. Finally, antisera from guinea pigs immunized with this PMD-generated HA show a skewed immune response toward the stem as demonstrated by a more cross-reactive antibody response compared to antisera obtained with animals immunized with unmodified HA.
RESULTS
Establishing the Protect, Modify, Deprotect method with hen egg white lysozyme (HEWL)
The initial validation of the PMD method was done using hen egg white lysozyme (HEWL), a well-characterized protein with known antigenic epitopes (Figure 2A). We chose to use amine-reactive N-hydroxysuccinimide-esters (NHS-esters) as our non-specific modifying reagent, because NHS-esters rapidly react with lysine residues and the N-terminal amino group at neutral pH (Figure 2B). There are three major non-overlapping, conformation-dependent epitopes on HEWL mapped by monoclonal antibodies, four of which are (i) HyHEL10 (72) and F9.13.7 (73), (ii) D11.15 (74, 75), and (iii) HyHEL5 (76) (Figure 2A, supporting information (SI) Figure 1A). Crystal structures are available for each of these mAbs bound to HEWL. The epitope mapped by HyHEL10 contains two lysine residues, K96 and K97 (SI Figure 1A). This epitope is partially shared by F9.13.7, which also binds over K96 and K97 (SI Figure 1A). D11.15 binds over a different lysine residue, K116. Finally, HyHEL5 does not contain any reactive amines (lysine residues or the N-terminus) in its epitope (SI Figure 1A).
We selected HyHEL10 as the protecting mAb for our proof of concept PMD study because (i) it bound over 2 lysine residues, (ii) it shares a significant portion of its epitope with F9.13.7, allowing for a separate test of epitope protection, and (iii) it does not contain the lysine residue present in the D11.15 epitope.
During the deprotection step in PMD, there is a need to separate the modified antigen from the protecting mAb. To facilitate this separation, we conjugated HyHEL10 to resin. We determined that HEWL bound to this HyHEL10 resin can be eluted at low pH (100 mM glycine, pH 1.5).
For the modification step we investigated different length PEG chains using NHS-polyethylene glycoln-methyl, where n denotes the number of ethylene glycol units (referred to as NHS-PEGn-me; n = 2, 4, 8, 12, or 24 (Figure 2B)). HEWL antigens that were PEGylated on an HyHEL10 resin and then dissociated (following the PMD protocol) are referred to as HEWL-pron. We simultaneously produced HEWL antigens that were PEGylated in solution, without antibody protection, and refer to them as HEWL-soln.
PMD-HEWL decreases antigenicity at off-target sites while maintaining on-target antigenicity
We used biolayer interferometry (BLI) to compare the binding of the four HEWL mAbs described above to wild type (WT) HEWL, the five HEWL derivatives PEGylated on HyHEL10 resin and the five HEWL derivatives PEGylated in solution. BLI measures the kinetics of protein-protein interactions and allowed us to determine dissociation constants (KD) for these 44 interactions (-log(KD) values with an overlaid heat map are shown in Figure 2C).
The top two rows of the heat map show that both HyHEL10 and F9.13.7 do not bind to HEWL-soln antigens, presumably because modification of the two lysine residues (K96 and K97) in their epitopes interferes with binding. Interestingly, HEWL-sol2 did not fully ablate HyHEL10 binding suggesting that PEG2 is too short to fully disrupt antibody binding, while PEG4 is sufficient. Conversely, the same two antibodies, HyHEL10 and F9.13.7, retain their binding to antigens produced using PMD (Figure 2C). This demonstrates that immobilization of HEWL on a HyHEL10 resin during PEGylation sufficiently protects the conformation-dependent HyHEL10 epitope from modification.
HyHEL5 bound to all HEWL derivatives (Figure 2C). There are no amines within the epitope for this mAb. These results indicate that PEGylation at other amines in the protein, even with long PEG chains, does not interfere with binding of HyHEL5. We refer to such epitopes that retain their antigenicity, even after the PMD protocol, as ‘holes’.
Finally, D11.15 bound to the WT protein but did not bind to any of the PEGylated proteins. D11.15 binds over a lysine residue outside of the HyHEL10 epitope. Thus, PMD can effectively modify antigenic sites outside of the epitope of interest. Enzyme linked immunosorbent assays (ELISAs) measuring binding of the four mAbs to ELISA plates coated with the modified HEWL derivatives yield results that are fully consistent with these BLI results (SI Figure 1C).
We further analyzed the proteins PEGylated on and off of the HyHEL10 resin using SDS-PAGE followed by Ponceau S staining and western blotting (SI Figure 1B). The results are generally consistent with those obtained by BLI (Figure 2C). The Ponceau S and western-blot analyses, however, reveal a ‘laddering’ phenomenon that is particularly prominent when longer PEGylation reagents are used (SI Figure 1B). Specifically, multiple, discrete forms of PEGylated HEWL derivatives are observed, with molecular weight differences consistent with those expected for integral differences in the number of PEGn units. This suggests that PEGylation is not complete in some cases. Likely candidate sites on HEWL that are incompletely PEGylated are K1 (the N-terminal residue), K96 and K97. Modification of the ε-amino or α-amino group of residue K1 may interfere with modification of the other, and modification at either K96 or K97 may act to hinder modification of the adjacent residue. Therefore, longer PEG chains could be detrimental in efforts to fully PEGylate amines within unprotected epitopes.
Taken together these results demonstrate that (i) protection with a monoclonal antibody is required to retain the epitope of interest, since HyHEL10 did not bind to HEWL PEGylated in solution, (ii) PMD can selectively ablate binding of off-target antibodies (in this case D11.15), (iii) use of longer PEGylation reagents can lead to incomplete modification and (iv) the antigenicity of modified HEWLs can be predicted reasonably well with co-crystal structures, suggesting that holes can be predicted from three-dimensional structural information.
PMD with influenza hemagglutinin using a conserved stem-binding mAb (MEDI8852)
Given the ability to conserve binding to a distinct epitope on HEWL after PMD, we sought to design an immunogen that would elicit an antibody response to the conserved stem of influenza HA by reducing the immunogenicity of the head. Such an immunogen should focus the immune system on the conserved HA stem, producing a more cross-reactive antibody response in immunized animals. Thus, we selected the stem-directed bnAb, MEDI8852 as our protecting antibody (21).
To prepare a PMD-HA antigen, we started with HAΔSA, which is based on A/New Caledonia/20/1999(H1N1), as previously described (77). We introduced a point mutation at the HA1/HA2 cleavage site to maintain the construct as HA0 (78) (SI Figure 2A) and added a foldon trimerization domain and purification tags at the C-terminus (see Methods). We refer to this construct as H1 WT. We utilized the crystal structure of a similar H1 HA (PDB ID: 4EDB (79)) to identify potential holes that are predicted to remain after PMD (i.e., regions lacking surface lysine residues). We utilized deep mutational scanning data (80, 81) to identify residues within these predicted holes that can be replaced with lysine. In this way, nine lysine substitutions were made in the head region of H1 WT. We refer to this protein as H1+9 (SI Figure 2B).
To enable elution of H1+9 off of MEDI8852 resin following PMD while avoiding the irreversible conformational change that occurs with HA at low pH (13, 14), we used the cocrystal structure of MEDI8852 with HA (PDB ID: 5JW4 (21)) to install two point mutations in the MEDI8852 heavy chain (R52A and Y54A). We refer to this mutated antibody as MEDI8852*. These mutations lower the affinity of binding and facilitated elution (82) of H1+9 off of a MEDI8852* resin in 2M KSCN at pH 7.4.
PMD was carried out as follows (SI Figure 2C). H1+9 was bound to MEDI8852* resin. The complex was PEGylated with NHS-PEG4-me. The PEGylated H1+9 was eluted off the resin at neutral pH. The final protein is referred to as H1+9+PEG.
H1+9+PEG is a properly folded antigen
We sought to confirm that the structure of the protein was not perturbed by the PEG modifications. Thus, we compared the H1+9+PEG antigen to both H1 WT and H1+9 using gel electrophoresis, circular dichroism (CD) spectroscopy, gel filtration chromatography, and calorimetry. H1+9+PEG has a higher MW than H1 and H1+9, as determined by SDS-PAGE. The MW difference is consistent with that expected for PEGylation of ∼20 amines on the surface of H1+9 (SI Figure 2D). Indistinguishable CD spectra for H1 WT, H1+9, and H1+9+PEG suggest that these proteins have the same folded structure (Figure 3A). The gel filtration results for all three proteins are consistent with those expected for a trimer (Figure 3B), with H1+9+PEG exhibiting a slightly earlier elution, consistent with an increased molecular weight due to PEGylation. Finally, calorimetry indicates that the proteins have a similar melting temperature (SI Figure 2E). Taken together, these results provide strong evidence that the HA antigen generated using PMD, H1+9+PEG, retains a native conformation.
In order to investigate whether it was necessary to protect the epitope during modification, we produced an antigen, denoted H1+9+sol, by PEGylating H1+9 in solution in the absence of Medi8852* (schematic in SI Figure 3A). H1+9+sol has a slightly higher molecular weight than H1+9+PEG as determined by SDS-PAGE analysis (SI Figure 3B), suggesting that additional PEGylation occurs in the absence of the mAb. The CD spectra of H1+9+sol and H1+9+PEG are different (SI Figure 3C). H1+9+sol melts at a lower temperature than H1+9+PEG, with an apparent pre-transition, as determined by calorimetry (SI Figure 3D). Thus, H1+9+sol appears to exhibit a notable conformational change compared to H1+9+PEG.
H1+9+PEG decreases head antigenicity
To determine if the PMD protocol could decrease antigenicity of the head region of HA, we used BLI to compare antibody binding to a set of six human monoclonal antibodies: three targeting the head and three targeting the stem (Figure 3C). All six antibodies bound to H1 WT (Figure 3C, left). Head directed antibody binding decreased after lysine substitutions (H1+9) and was further reduced after PEGylation (H1+9+PEG) (red, Figure 3C). Notably, H1+9+PEG showed reduced but not ablated binding to the head antibody H2897, indicating the presence of a hole in the head of H1+9+PEG.
In contrast to head mAb binding, stem directed antibodies retain their binding after lysine substitution (H1+9) and after PEGylation (H1+9+PEG) (blue, Figure 3C right graphs). This demonstrates that the conformation of the HA stem is retained in the case of H1+9+PEG.
Importantly, stem-directed antibodies show decreased binding to H1+9+sol compared to H1+9+PEG (SI Figure 3E), indicating that the PMD protocol is required to retain on-target antigenicity. It is likely that PEGylation of a single lysine residue on the periphery of the MEDI8852 epitope and/or the conformational change that occurs when H1+9 is PEGylated in solution (see above) is responsible for this difference in binding.
Yeast expressing antibody-fragments show preferential stem binding towards H1+9+PEG
Given that H1+9+PEG shows reduced binding of head antibodies, while retaining the binding of stem antibodies, we sought to investigate antigenicity in a high-avidity situation (e.g., as would occur with a B cell population in vivo). A set of mAbs were expressed on the surface of yeast cells in the form of single chain variable fragments (scFvs). It has been estimated that ∼50,000 copies of scFv are expressed per cell using this protocol (83). Tetramers of either H1 WT or H1+9+PEG, prepared by incubating biotinylated antigens with streptavidin, were used as bait in fluorescence activated cell sorting (FACS) experiments with four head-directed and six stem-directed yeast clones (representative FACS sorts are shown in SI Figure 4). Consistent with results obtained with isolated mAbs, the results with this high avidity system indicate that all of the stem antibodies that bind to H1 WT also bind to H1+9+PEG, while binding of the head antibodies is significantly reduced (Figure 4A). In addition, the previously identified H2897 hole is apparent (∼3% of clones were antigen positive).
Yeast display of scFvs also offers the possibility of generating libraries that can be used to detect holes in PMD antigens using FACS. As an initial experiment, we produced a minilibrary of yeast expressing 22 different scFvs that bind to HAs of various subtypes. We pooled the 22 clones at an approximate equimolar ratio (Figure 4B top), performed FACS with either H1 WT or H1+9+PEG tetramers, and sequenced the selected antigen-positive yeast. When the yeast library was sorted with H1 WT, there was no significant enrichment for either head or stem directed clones (Figure 4B bottom, left). However, when the library was sorted with H1+9+PEG, there was a profound enrichment for stem-directed clones (Figure 4B bottom, right). These results show that H1+9+PEG is capable of enriching a polyclonal library for stem directed clones and suggest that much larger libraries of scFv-displayed clones could be used to efficiently detect holes in PMD antigens in a high-throughput manner.
H1+9+PEG elicits more cross-reactive serum compared to H1 WT
To evaluate the in vivo immunofocusing ability of PMD, we immunized guinea pigs with either H1 WT or H1+9+PEG in Imject Alum adjuvant (ThermoFisher). Animals were boosted with the same composition at day 20. This immunization experiment was done twice. The first immunization contained three animals in each group and the second immunization contained four animals in each group. A single animal (GP5) in the H1+9+PEG group from the first immunization produced a significantly weaker immune response (SI Figure 5A) and was therefore omitted from further data processing.
On average, day 30 antisera from animals immunized with H1+9+PEG show slightly less binding to H1 WT as determined by ELISA than those immunized with H1 WT (Figure 5A left). This trend is also apparent at the individual animal level, as illustrated by their serum EC25 titers, but the difference is not statistically significant (Figure 5B left). PEGylation is known to decrease the overall immunogenicity of proteins (68, 69, 84).
In contrast, ELISA results with the same H1+9+PEG antisera show more cross-strain binding to H5 HA (A/Viet Nam/1203/2004 (H5N1)) (85) as compared to the H1 WT antisera (Figure 5A). The difference is even more pronounced with binding to H2 HA (A/Japan/305/1957 (H2N2)) (86) (Figure 5A). Comparing EC25 titers at the individual animal level indicates that these differences are significant (Figure 5B).
As a second method to evaluate antisera cross-reactivity, we used BLI. Day 30 antisera from each group were pooled in equal amounts from each animal. These BLI experiments confirm that antisera from H1 WT immunized animals bound better to H1 WT than H1+9+PEG immunized animals but bound worse to H5 or H2 HA antigens (SI Figure 5B). Taken together, these ELISA and BLI results suggest that immunization with H1+9+PEG skews the antibody response towards the conserved stem epitope.
DISCUSSION
Our results demonstrate that PMD can be used as a generalizable immunofocusing method requiring minimal protein engineering. With HEWL, we show that the PMD protocol keeps the epitope of interest intact, while decreasing antigenicity elsewhere on the protein. With HA, we show that a PMD-generated antigen shows greatly reduced mAb binding at the head region, while retaining robust binding to the stem region. This selective antigenicity was maintained in a high-avidity comparison with yeast-displayed scFvs. Using PMD-HA as bait in FACS sorting experiments, yeast clones expressing scFvs that bind to the stem region of HA were selectively enriched from a mini-library. Finally, when this PMD-HA antigen was used to immunize guinea pigs, the resultant antisera was more cross-reactive to HAs from other influenza strains, compared to animals immunized with unmodified HA. Although the in vivo derived effects are modest, taken together, our experiments demonstrate the viability of PMD for use in immunogen design.
Possible immediate steps to improve the efficacy of H1+9+PEG as an immunogen include: (i) introducing additional lysine substitution(s) to eliminate the hole on the HA head identified by the mAb H2897, (ii) altering the PEG length or modifying reagent and/or (iii) utilizing other chemistries outside of NHS-esters (70). It will also be important to discover new holes that need to be eliminated with additional mAbs (e.g., with yeast-display scFv libraries). We imagine such improvements to be iterative, where new PMD candidates can be sequentially screened in vitro as outlined above before use in immunization experiments in vivo. We also note that PMD vaccine candidates can be prioritized based on human B cell binding experiments (e.g. (53, 87)).
Importantly, the PMD strategy is generalizable. It requires an antigen of interest and a mAb with an epitope against which one would like to direct a vaccine-induced antibody response. Three-dimensional structural information is helpful but not absolutely required. Generating PMD antigens with a binding partner that is not a mAb is also conceivable (e.g., using cell-surface receptors such as CD4 for HIV-1 (88), or SR-B1 for HCV (89)). Indeed, there are many attractive targets for the application of PMD.
We anticipate that another advantage of PMD is its potential to produce high-resolution epitope-focused vaccines. This is because individual residues on an antigen either are, or are not, protected from chemical modification by a binding partner. Consequently, in theory, PMD could be used to create immunofocusing antigens at the resolution of specific residues. For example, it is conceivable that PMD could lead to vaccine candidates that avoid eliciting non-neutralizing antibodies that bind to epitopes overlapping with those of neutralizing antibodies (see e.g., (90, 91)).
Of many possible applications of PMD, HIV-1 is particularly interesting to consider. The initial, immunodominant antibody responses to HIV-1 are strain specific (92–95). While rare, many bnAbs have been isolated from infected subjects and these bnAbs can be mapped to a few epitopes on HIV-1 Env (96, 97). The sequences of these bnAbs indicate that an extensive degree of somatic hypermutation generally occurs during years of viral and host co-evolution (98, 99). PMD offers the possibility of creating immunogens to determine the possibility of eliciting a bnAb-like response in the absence of extensive somatic hypermutation, if other strain-specific antibody responses against HIV-1 are avoided.
Today, most vaccines are produced using methods developed many decades ago. Although in some cases these have had tremendous success, most notably the eradication of smallpox, they have failed to address some of largest medical needs in the field of vaccinology, like HIV-1 and influenza. Modern immunofocusing methods and the discovery of bnAbs have reignited the field to target such historically intractable diseases. Since PMD utilizes these bnAbs and can be used in combination with other immunofocusing strategies, we hope that it will aid in creating new vaccines.
ACKNOWLEDGMENTS
We thank A. E. Powell and members of the Kim lab for helpful comments on early drafts of this manuscript; B. N. Bell and A. E. Powell for helpful discussion; J. R. Cochran for access to the CD spectrometer and Accuri flow cytometer; and J. E. Pak for access to the calorimeter. This work was supported by the Virginia and D. K. Ludwig Fund for Cancer Research and the Chan Zuckerberg Biohub.
REFERENCES
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.
- 18.
- 19.
- 20.
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.
- 26.
- 27.
- 28.
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.
- 36.
- 37.
- 38.
- 39.
- 40.
- 41.
- 42.
- 43.↵
- 44.
- 45.
- 46.
- 47.
- 48.↵
- 49.
- 50.
- 51.
- 52.
- 53.↵
- 54.↵
- 55.
- 56.
- 57.↵
- 58.↵
- 59.
- 60.↵
- 61.↵
- 62.
- 63.
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵
- 89.↵
- 90.↵
- 91.↵
- 92.↵
- 93.
- 94.
- 95.↵
- 96.↵
- 97.↵
- 98.↵
- 99.↵
- 100.↵
- 101.↵