Protein Engineering vol.14 no.6 pp.397–402, 2001
`
`A critical view on conservative mutations
`
`Per Harald Jonson and Steffen B.Petersen1
`
`Biostructure and Protein Engineering Group, Department of Life Sciences,
`Aalborg University, Sohngaardsholmsvej 49, DK-9000 Aalborg, Denmark
`1To whom correspondence should be addressed. E-mail: sp@bio.auc.dk
`By analysing the surface composition of a set of protein
`3D structures, complemented with predicted surface com-
`positional information for homologous proteins, we have
`found significant evidence for a layer composition of protein
`structures. In the innermost and outermost parts of proteins
`there is a net negative charge, while the middle has a net
`positive charge. In addition, our findings indicate that the
`concept of conservative mutation needs substantial revision,
`e.g. very different spatial preferences were found for
`glutamic acid and aspartic acid. The alanine screening
`often used in protein engineering projects involves the
`substitution of residues to alanine, based on the assumption
`that alanine is a ‘neutral’ residue. However, alanine has a
`high negative correlation with all but the non-polar res-
`idues. We therefore propose the use of, for example, serine
`as a substitute for the residues that are negatively correlated
`with alanine.
`Keywords: amino acid properties/protein engineering/solvent
`accessibility/spatial contacts/structural preference
`
`Introduction
`Upon folding of a peptide chain into a 3D protein structure,
`some residues are transferred from a polar environment to a
`more non-polar environment
`in the interior of the folded
`protein. This transfer is driven by the thermodynamic properties
`of the amino acids and the solvent. Throughout molecular
`evolution nature has selected for suitable function and stability
`of the resulting protein. For small to medium sized proteins—
`in the folded structure—only a few residues are totally buried
`(Chothia, 1976; Miller et al., 1987; Petersen et al., 1998),
`whereas most residues are only partially buried. The variation
`in solvent accessibility is dependent on the properties of the
`residue in question and is reflected in the amino acid composi-
`tion throughout the protein structure. These differences in the
`solvent accessibility profile have found wide applications in
`various structure prediction methods (Holbrook et al., 1990;
`Rost and Sander, 1994; Thompson and Goldstein, 1996). Also,
`the use of environment specific substitution matrices (Donnelly
`et al., 1994; Wako and Blundell, 1994) have proven valuable.
`The sequential neighbourhood of amino acids has been investi-
`gated previously (Vonderviszt et al., 1986) and its use has
`been found in, for example, loop prediction (Wojcik et al.,
`1999) and secondary structure prediction (Chou and Fasman,
`1978; Chandonia and Karplus, 1999; Jones, 1999). No signific-
`ant correlation between residues sequential neighbour prefer-
`ence was discovered.
`The spatial neighbourhood around individual residues has
`
`© Oxford University Press
`
`also been previously investigated (Burley and Petsko, 1985;
`Bryant and Amzel, 1987; Miyazawa and Jernigan, 1993;
`Petersen et al., 1999). Further, spatial contacts have been
`studied to derive contact potentials for the different amino
`acid interactions (Brocchieri and Karlin, 1995; Miyazawa and
`Jernigan, 1996, 1999). The common strategy is to study the
`number of contacts within a given distance cut-off. However,
`the literature seems devoid of investigations of distance-
`dependent contacts and also of reports utilizing the embedded
`information of the solvent accessibility of the residues involved.
`A two-state prediction of solvent accessibility correlation
`between hydrophobicity, buried contact propensity and the
`location in the prediction window has been reported
`(Mucchielli-Giorgi et al., 1999). However, it does not describe
`any correlation between individual residue distributions.
`It is important to be able to discriminate between correctly
`folded and misfolded model structures. It has been pointed
`out that potential energy-based methods do not discriminate
`well between folded and misfolded structures. However, struc-
`tural features such as buried polar surface (Overington et al.,
`1992) and number of polar contacts (Bryant and Amzel, 1987;
`Golovanov et al., 1999) have proven valuable.
`In protein engineering the concept of conservative mutations
`is frequently used. The general idea is that a substitution of
`an amino acid with another amino acid with similar physico-
`chemical properties will not influence the stability and function
`of the protein. The present paper shows that
`the spatial
`preferences for similar residues can be dramatically different
`in protein structures under similar circumstances (in this
`context solvent accessibility).
`The results of the neighbour analysis will be valuable in
`model validation, as a tool for structure prediction and especi-
`ally as a guide in the search for stability enhancing mutations.
`
`Methods
`The sequences used are a subset of the 25% sequence identity
`set of non-homologous structures (Hobohm et al., 1992;
`Hobohm and Sander, 1994) derived from the protein structure
`databank PDB (Bernstein et al., 1977). Only single-chain
`protein sequences were used. The resulting dataset consisted
`of 336 single-chain sequences with a maximum pairwise
`sequence identity of 25%. The subset was expanded through
`the use of the corresponding HSSP-files (Dodge et al., 1998).
`The total data set contained 8379 aligned sequences and
`1 415 986 residues. This corresponds to 6.7% of all residues
`in version 34 of SWISS-PROT (Bairoch and Apweiler, 1997).
`The length of the sequences was between 64 and 1017 residues.
`The resolution of the X-ray structures used varied between
`1.0 and 3.0 Å, with an average of 2.0 Å. Further,
`the
`subset contained 31 structures solved by NMR. However, all
`hydrogen-atom co-ordinates were discarded. To check for a
`possible bias introduced by the use of
`the homologous
`sequences the complete analysis was done with and without the
`aligned sequences. No significant differences were observed,
`
`397
`
`Bausch Health Ireland Exhibit 2035, Page 1 of 6
`Mylan v. Bausch Health Ireland - IPR2022-00722
`
`

`

`P.H.Jonson and S.B.Petersen
`
`although the reduced size of the smaller of the two datasets,
`as expected, gave rise to more noise.
`The spatial neighbours of each residue were determined
`based on solvent accessibility and spatial distance. The solvent
`accessibility was taken from the respective HSSP-files (Dodge
`et al., 1998). For each surface residue the neighbouring surface
`residues were grouped according to their distance to the residue
`in question. The distance between two residues was computed
`as the shortest distance among the set of all possible pairs of
`atoms in the two residues. We assume that the alignment in
`the HSSP-file implies that neighbours in the main sequence
`are also neighbours in the aligned sequences and that the
`solvent accessibility is conserved (Andrade et al., 1998;
`Goldman et al., 1998). The expected number of neighbour
`interactions between residues of type i and j are calculated by
`⫽ xj|d,ACC · xj|d,ACC · N0|d,ACC
`N expected
`ij|d,ACC
`where xi and xj are the fraction of amino acid i and j in the
`dataset for the distance range d and at a solvent accessibility
`larger than the cut-off ACC and N0 is the total number of
`observed neighbour contacts. The score, Sij|d,ACC, is calcu-
`lated by
`
`(1)
`
`(3)
`
`(2)
`
`ij|d,ACC / expectedij|d,ACC )
`Sij|d,ACC ⫽ ln (N observed
`
`This gives a negative score for disfavoured neighbour-pairs
`and a positive score for favoured interactions. The score value
`Sij|d,ACC can be transformed into an apparent thermodynamic
`parameter by multiplication with RT.
`The net charge in each layer of the protein was calculated.
`Aspartic acid and glutamic acid are considered negatively charged
`and arginineand lysine are consideredpositively charged. Histidine
`is either considered as uncharged or positively charged. The relat-
`ive net charge, ∆qrel, we define as
`∆qrel ⫽ (NPositive – NNegative) /N Total
`where NPositive is the number of positive residues, NNegative the
`number of negative residues and NTotal the total number of
`residues in that particular layer.
`The PDB identification codes for the structures used are
`1ptx, 2bbi, 1hcp, 1iml, 1cdq, 1vcc, 1nkl, 1tiv, 2abd, 2hts, 1tpg,
`1fbr, 1pco, 1who, 1beo, 2ncm, 1fim, 1tlk, 1xer, 1onc, 1rga,
`1erw, 1fd2, 1put, 1fkj, 1jpc, 1thx, 1jer, 1ccr, 1wad, 2tgi, 1pls,
`1neu, 4rhn, 1rmd, 1hce, 1hfh, 1tam, 2pf1, 1bip, 1whi, 1yua,
`1bp2, 1zia, 4fgf, 7rsa, 1bw4, 2vil, 1eal, 1rie, 1doi, 3chy, 1cpq,
`1msc, 1mut, 1rcb, 1lzr, 1htp, 1lid, 1lis, 1lit, 1kuh, 1nfn, 1irl,
`1poc, 2tbd, 1cof, 1pms, 1rsy, 1snc, 1eca, 1jvr, 2end, 1anu,
`5nul, 1fil, 1jon, 1lcl, 1itg, 1tfe, 1maz, 1pkp, 1lba, 1vsd, 2fal,
`1ash, 1def, 2hbg, 1div, 1gds, 1grj, 1i1b, 1ilk, 1rcy, 1sra, 1ulp,
`1mbd, 1aep, 1jcv, 2gdm, 1phr, 1rbu, 1esl, 1hlb, 1mup, 1vhh,
`1gpr, 1btv, 1cyw, 1klo, 1l68, 3dfr, 2cpl, 1sfe, 1huw, 5p21,
`1ha1, 1wba, 1lki, 2fha, 1prr, 2fcr, 1amm, 1cid, 1hbq, 1cdy,
`2stv, 153l, 1rec, 1xnb, 2sas, 1gky, 1knb, 1ryt, 1zxq, 1har, 1cex,
`1chd, 2tct, 2ull, 1gen, 1iae, 1nox, 1rnl, 2gsq, 1cfb, 1dyr, 1nsj,
`2hft, 1fua, 2eng, 1thv, 1hxn, 2abk, 9pap, 1lbu, 3cla, 1vid,
`2ayh, 2dtr, 1gpc, 1dts, 1jud, 1emk, 1ois, 1akz, 1sgt, 1ad2,
`1nfp, 1din, 1lrv, 1dhr, 1bec, 1lbd, 1dpb, 1jul, 1mrj, 1fib, 1hcz,
`1mml, 1vin, 1dja, 2cba, 3dni, 1lxa, 1arb, 1rgs, 1tys, 3tgl, 1ako,
`1eny, 1ndh, 2dri, 1xjo, 1drw, 1kxu, 2prk, 1cnv, 1tfr, 1ytw,
`1iol, 2ebn, 1tml, 1han, 1xsm, 1pbn, 1amp, 1ryc, 1bia, 1vpt,
`1csn, 2ora, 1ctt, 1bco, 1fnc, 1gym, 1pda, 1cpo, 1esc, 2reb,
`1mla, 1sig, 8abp, 1ghr, 1iow, 2ctc, 1gca, 1sbp, 1ede, 1pgs,
`2cmd, 1anv, 1gsa, 1tag, 1dsn, 2acq, 1cvl, 1tca, 2abh, 2pia,
`
`398
`
`Fig. 1. The net relative charge in layers of protein structures with different
`solvent accessibility (ACC). The net relative charge is defined as the net
`charge per residue found in a particular layer (number of positive charges –
`number of negative charges/number of residues). Aspartic acid and glutamic
`acid are considered to be negative, arginine, lysine and protonated histidine
`positive (dotted line). The solid line include all of the above-mentioned
`residues except histidine.
`
`1pot, 1vdc, 1axn, 1msk, 1hmy, 2bgu, 1ldm, 1dxy, 1ceo, 1nif,
`1arv, 1xel, 1uxy, 1rpa, 2lbp, 3pte, 1uby, 1fkx, 1pax, 3bcl, 1air,
`1mpp, 2mnr, 1eur, 1cem, 1fnf, 1pea, 1omp, 2chr, 1pud, 1kaz,
`1mxa, 1edg, 2sil, 1ivd, 1pbe, 1svb, 1ars, 1oyc, 1inp, 1oxa,
`1eft, 1phg, 1cpt, 1iso, 1qpg, 2amg, 1uae, 1gnd, 2dkb, 1gpl,
`1csh, 4enl, 1pmi, 1lgr, 1nhp, 1gcb, 1bp1, 1geo, 2bnh, 3grs,
`1gln, 1gai, 2pgd, 2cae, 2aaa, 1byb, 1smd, 2myr, 3cox, 1dpe,
`1pkm, 1ayl, 1crl, 1ctn, 1clc, 1tyv, 2cas, 1ecl, 1oxy, 1vnc, 1gal,
`1dlc, 1sly, 1dar, 1gof, 1bgw, 1aa6, 1vom, 8acn, 1kit, 1taq,
`1gpb, 1qba, 1alo and 1kcw.
`
`Results and discussion
`The distribution of charged residues in different layers of the
`protein 3D structure and the total net charge are shown in
`Figure 1. In the innermost and outermost parts of proteins
`there is a net negative charge, while the middle has a
`net positive charge. This apparent three layer structure with
`alternating charge exposing the negatively charged outermost
`layer to the solvent is interesting. Such organisation will secure
`some level of radial charge neutralisation, and may possibly
`contribute to tight packing of the protein. Likewise this charge
`organisation of the surface layer could provide important
`electrostatic guidance during the folding event. Conversely,
`changing pH to acidic or alkaline conditions at which subsets
`of the titratable residues becomes uncharged will destabilise
`the packing of residues at the surface of the protein. Buried,
`acidic amino acids can be found in several different protein
`structures and these residues play important functional roles
`in, for example, trypsin (McGrath et al., 1992), ribonuclease
`T1 (Giletto and Pace, 1999) and thioredoxin (Dyson et al.,
`1997; Bhavnani et al., 2000). The reported three layer structure
`is observed both with and without the aligned sequence and
`is therefore not caused by a bias introduced by the conservation
`of the buried, charged groups within a protein family.
`The spatial neighbours around each type of residue were
`calculated without any discrimination for solvent accessibility.
`With the notable exceptions of tryptophan and cysteine, amino
`acids were not frequently observed as spatial neighbours to
`identical residue types. This trend was not dependent upon the
`choice of distance cut-off (results not shown). The differences
`in distribution were remarkably small between the different
`
`Bausch Health Ireland Exhibit 2035, Page 2 of 6
`Mylan v. Bausch Health Ireland - IPR2022-00722
`
`

`

`Critical view on conservative mutations
`
`Fig. 2. The significantly over- or under-represented neighbour pairs. All residues have a solvent accessibility higher than 20%. The distance in Å between
`the residues is given along the vertical axis. Red and green represent areas where the number of pairs are less and higher than expected, respectively.
`(A) Tryptophan; (B) glycine; (C) proline; (D) histidine; (E) lysine; (F) aspartic acid.
`
`amino acids for an 8 Å distance cut-off suggesting that 8 Å
`is a large enough distance for the distribution to become
`independent of the nature of the central residue. This observa-
`tion led to the use of 8 Å as the largest distance between
`neighbours investigated in detail.
`Figure 2 shows the score values for all amino acid neighbour
`pairs involving tryptophan, glycine, alanine, proline, serine,
`histidine, lysine and aspartic acid for neighbour pairs with at
`least 20% solvent accessibility. The results for the other amino
`acids are available on our homepage (http://www.bio.auc.dk/).
`Score values have been calculated similarly for other solvent
`accessibility cut-offs. The aromatic residue tryptophan is one
`of only two residues showing a clear preference for contacts
`
`with the same residue type (the other is cysteine). Also
`interactions with the other aromatic residues are preferred.
`Interestingly the interactions between tryptophan and the two
`acidic residues (aspartic acid and glutamic acid) seem different.
`While tryptophan and glutamic acid are observed less fre-
`quently than expected, the opposite is observed for tryptophan
`and aspartic acid. Glycine shows the typical negative score
`for interactions with the same residue type. Also, glycine does
`not seem to have neighbours in the close spatial neighbourhood
`(艋3.5 Å). This under-representation of neighbours close by is
`even clearer for proline. We interpret this under-representation
`as a sign of the preference for loop that proline residues have.
`The lack of interactions with all other amino acids in its
`
`399
`
`Bausch Health Ireland Exhibit 2035, Page 3 of 6
`Mylan v. Bausch Health Ireland - IPR2022-00722
`
`

`

`P.H.Jonson and S.B.Petersen
`
`Fig. 3. Over- and under-represented neighbour pairs as a function of solvent accessibility (ACC) and distance (Å). The distance between the residues is given
`along the vertical axis and the solvent accessibility along the horizontal axis. (A) Lysine–aspartic acid; (B) glutamic acid–aspartic acid; (C) tryptophan–
`glutamic acid; (D) tryptophan–aspartic acid; (E) histidine–aspartic acid; (F) serine–histidine.
`
`vicinity point to most contacts being with solvent molecules.
`However, proline has an abundance of contacts at a larger
`distance (4–5 Å). Histidine is interesting in that it shows signs
`of its aromatic properties, through preference for contacts with
`aromatic residues (~3.5 Å), and its polarisable nature, through
`preferred contacts with the negatively charged residues (~3 Å).
`The basic amino acid lysine has as expected a clear negative
`score for contacts with other lysines. The favourable electro-
`static interactions with the acidic amino acids is evident.
`Some of the most interesting pair interactions are shown in
`Figure 3. Figure 3A depicts the salt bridge pair lysine–aspartic
`acid. The strong over-representation seen at 3 Å separation is
`consistent with the classical salt bridge concept. The over-
`
`representation of lysine–aspartic acid pairs in the most solvent
`exposed layers observed at 5.5 to 6 Å is unexpected. We
`propose that charge networks on the protein surface could
`cause this observation. In Figure 3B the result for the glutamic
`acid–aspartic acid pair is shown. The most obvious feature is
`the expected under-representation of this pair. However, close
`to the protein surface the same restriction does not appear to
`be present. Again we propose that surface located charge
`networks are contributing to this observation. In Figures 3C
`and D the amino acid pairs tryptophan–glutamic acid and
`tryptophan–aspartic acid are shown. The common belief that
`a glutamic acid to aspartic acid mutation is conservative is
`contrary to the observations shown. The tryptophan–glutamic
`
`400
`
`Bausch Health Ireland Exhibit 2035, Page 4 of 6
`Mylan v. Bausch Health Ireland - IPR2022-00722
`
`

`

`acid pair is highly under-represented in the highly solvent
`exposed layers of the proteins. Surprisingly, the same cannot
`be said for the tryptophan–aspartic acid pair, where an over-
`representation is observed for the 3.5 to 6 Å distance interval.
`Similar, but less pronounced, observation was made for the
`tyrosine–glutamic acid and tyrosine–aspartic acid pairs. No
`significant differences were observed between the phenylalan-
`ine–glutamic acid and phenylalanine–aspartic acid pairs. The
`only difference between the glutamic acid and the aspartic
`acid is the length of the side chain. Common to both tryptophan
`and tyrosine is their polarisability, in contrast to phenylalanine.
`We believe that surface located tryptophans involved in defining
`protein functionality are polarised by their local electrostatic
`environment. Although we cannot provide a quantitative
`explanation, it is plausible that the differences between the
`different chain length of glutamic acid and aspartic acid may
`put preference on the proximity to tryptophan. It has been
`shown that aspartic acid has a tendency to have favourable
`interactions between the side chain carbonyl group and the
`backbone carbonyl group (Deane et al., 1999), resulting in a
`ring-like structure. Similar conformations have not been
`observed for glutamic acid. In Figures 3E and F the histidine–
`aspartic acid and serine–histidine pairs are shown. Since these
`three residues constitute the active site residues of a wide
`range of hydrolases they have particular interest. There is an
`over-representation of histidine–aspartic acid pairs in the highly
`solvent accessible areas. The distance is larger than the typical
`distance observed in active site crevasses. However, the small,
`but significant, over-representation in the 3 Å range conforms
`with the
`classical histidine–aspartic
`acid distances
`in
`hydrolases. Figure 3E shows the clear preference for contacts
`between buried histidines and aspartic acids. We believe that
`this feature is an important part of the molecular evolution of
`de novo catalytic sites. Storing possible catalytic ‘triads’ in
`non-functional environments makes the number of amino acid
`substitutions necessary to activate the site smaller.
`The most distinct feature in Figure 3F is the clear under-
`representation of serine–histidine pairs in highly solvent
`exposed environments. A weak over-representation of the
`serine–histidine pair is seen at 3 Å in the less solvent accessible
`areas. Thus the presence of the catalytic triad apparently is
`determined mostly by the preference of the histidine–aspartic
`acid pair although the serine–histidine pair reveal similar, but
`much weaker, trends.
`The amino acid composition of each solvent accessibility
`layer was determined. As expected the buried parts of the
`proteins are composed of a higher amount of non-polar residues
`than the more solvent exposed layers. The correlation between
`the amino acid composition was calculated from the data of
`the composition of the individual structural layers. Amino
`acids that have similar preferences for solvent contact and
`local environment are expected to show a high positive
`correlation because of similar trends in their distribution.
`Hence, amino acids showing negative correlation will have
`different preferences for local environment and are therefore
`not believed to be compatible, i.e. a single site mutation of
`this type at this location is not recommended. As the non-
`polar residues are abundant in the core and show a gradual
`decrease as the solvent accessibility increases in general the
`correlation between the non-polar
`residues
`is positive
`(Figure 4). In contrast, the polar residues are more abundant
`in the highly exposed parts and hence are negatively correlated
`with the non-polar residues. Histidine and threonine behave
`
`Critical view on conservative mutations
`
`Fig. 4. Correlation between distribution of amino acids in proteins. The
`correlation is calculated based on the amino acid composition of the
`different layers of solvent accessibility layer of the protein structure. The
`green areas represent positive correlation, whereas the red areas represent
`negative correlation. Areas with low degree of correlation are in white.
`
`markedly differently. They show positive correlation to each
`other, but little correlation with any of the other columns, with
`the exception of arginine and glycine. This is caused by the
`low occurrence of histidine and threonine in both the buried
`and highly exposed areas and their relatively high occurrence
`in the medium exposed layers. Histidine has positive correlation
`with two aromatic residues, tryptophan and tyrosine, and with
`the weakly polar threonine and the polar arginine. Again we
`interpret this as a sign of both the aromatic properties and the
`charge properties of histidine. The weakly polar residues do
`not have the same clear similarity in distribution as the polar
`and non-polar residues. Proline and serine seem to be more
`closely related to the polar residues. The weakly polar residue
`alanine has positive correlation only with the non-polar res-
`idues. We propose that mutations between residues with high
`positive correlation have a high chance of maintaining the
`thermodynamic stability of the 3D structure. This is particularly
`so for charged residues. In contrast, the residues with a high
`degree of negative correlation are typically residues with
`different physical-chemical properties, which cannot be inter-
`changed without changing the physical chemistry of the protein.
`The non-correlated residues involve residues with a special
`role in the structure, e.g. some residues often involved in
`catalysis. We believe that the observation that proline in our
`study behaves similarly to polar residues is related with the
`structural role of proline residues and its preference for
`loops and turns. The alanine screening often used in protein
`engineering projects involves the substitution of residues to
`alanine, based on the assumption that alanine is a ‘neutral’
`residue. However, our data shows that alanine has a high
`negative correlation with all but the non-polar residues. We
`therefore propose the use of, for example, serine as a substitute
`for the residues that are negatively correlated with alanine.
`In the authors’ opinion the present paper provides important
`new information about protein structural organisation. The
`protein surface should be viewed as a multi-layered structural
`feature of the protein, where each layer has its specific
`composition and resulting characteristics. This simple key
`
`401
`
`Bausch Health Ireland Exhibit 2035, Page 5 of 6
`Mylan v. Bausch Health Ireland - IPR2022-00722
`
`

`

`P.H.Jonson and S.B.Petersen
`
`observation we believe will be of significant importance for
`many protein engineering strategies that target modification of
`solvent exposed residues.
`
`Acknowledgements
`P.H.J. thanks the Research Council of Norway for financial support (NFR-
`116316/410). S.B.P. expresses his gratitude for financial support from Obelsk
`Familiefond as well as Mål-2.
`
`and
`
`References
`Andrade,M.A., O’Donoghue,S.I. and Rost,B. (1998) J. Mol. Biol., 276,
`517–525.
`Bairoch,A. and Apweiler,R. (1997) Nucleic Acids Res., 25, 31–36.
`Bernstein,F.C., Koetzle,T.F., Williams,G.J., Meyer,E.E.,Jr, Brice,M.D.,
`Rodgers,J.R., Kennard,O., Shimanouchi,T. and Tasumi,M. (1977) J. Mol.
`Biol., 112, 535–542.
`Bhavnani,M., Lloyd,D., Bhattacharyya,A., Marples,J., Elton,P.
`Worwood,M. (2000) Gut, 46, 707–710.
`Brocchieri,L. and Karlin,S. (1995) Proc. Natl Acad. Sci. USA, 92, 12136–
`12140.
`Bryant,S.H. and Amzel,L.M. (1987) Int. J. Pept. Protein Res., 29, 46–52.
`Burley,S.K. and Petsko,G.A. (1985) Science, 229, 23–28.
`Chandonia,J.M. and Karplus,M. (1999) Proteins, 35, 293–306.
`Chothia,C. (1976) J. Mol. Biol., 105, 1–14.
`Chou,P.Y. and Fasman,G.D. (1978) Annu. Rev. Biochem., 47, 251–276.
`Deane,C.M., Allen,F.H., Taylor,R. and Blundell,T.L. (1999) Protein Eng., 12,
`1025–1028.
`Dodge,C., Schneider,R. and Sander,C. (1998) Nucleic Acids Res., 26, 313–315.
`Donnelly,D., Overington,J.P. and Blundell,T.L. (1994) Protein Eng., 7, 645–
`653.
`Dyson,H.J., Jeng,M.F., Tennant,L.L., Slaby,I., Lindell,M., Cui,D.S., Kuprin,S.
`and Holmgren,A. (1997) Biochemistry, 36, 2622–2636.
`Giletto,A. and Pace,C.N. (1999) Biochemistry, 38, 13379–13384.
`Goldman,N., Thorne,J.L. and Jones,D.T. (1998) Genetics, 149, 445–458.
`Golovanov,A.P., Volynsky,P.E., Ermakova,S.B. and Arseniev,A.S.
`(1999)
`Protein Eng., 12, 31–40.
`Hobohm,U. and Sander,C. (1994) Protein Sci., 3, 522–524.
`Hobohm,U., Scharf,M., Schneider,R. and Sander,C. (1992) Protein Sci., 1,
`409–417.
`Holbrook,S.R., Muskal,S M. and Kim,S.H. (1990) Protein Eng., 3, 659–665.
`Jones,D.T. (1999) J. Mol. Biol., 292, 195–202.
`McGrath,M.E., Vasquez,J.R., Craik,C.S., Yang,A.S., Honig,B.
`Fletterick,R.J. (1992) Biochemistry, 31, 3059–3064.
`Miller,S., Janin,J., Lesk,A.M. and Chothia,C. (1987) J. Mol. Biol., 196,
`641–656.
`Miyazawa,S. and Jernigan,R.L. (1993) Protein Eng., 6, 267–278.
`Miyazawa,S. and Jernigan,R.L. (1996) J. Mol. Biol., 256, 623–644.
`Miyazawa,S. and Jernigan,R.L. (1999) Proteins, 36, 347–356.
`Mucchielli-Giorgi,M.H., Tuffery,P. and Hazout,S. (1999) Theor. Chim. Acta.,
`101, 186–193.
`Overington,J., Donnelly,D., Johnson,M.S., S˘ ali,A. and Blundell,T.L. (1992)
`Protein Sci., 1, 216–226.
`Petersen,M.T.N., Jonson,P.H. and Petersen,S.B. (1999) Protein Eng., 12,
`535–548.
`Petersen,S.B., Jonson,P.H., Fojan,P., Petersen,E.I., Petersen,M.T.N., Hansen,S.,
`Ishak,R.J. and Hough,E. (1998) J. Biotechnol., 66, 11–26.
`Rost,B. and Sander,C. (1994) Proteins, 20, 216–226.
`Thompson,M.J. and Goldstein,R.A. (1996) Proteins, 25, 38–47.
`Vonderviszt,F., Ma´trai,G. and Simon,I. (1986) Int. J. Pept. Protein Res., 27,
`483–492.
`Wako,H. and Blundell,T.L. (1994) J. Mol. Biol., 238, 693–708.
`Wojcik,J., Mornon,J.P. and Chomilier,J. (1999) J. Mol. Biol., 289, 1469–1490.
`
`and
`
`Received October 17, 2000; revised January 21, 2001; accepted February
`15, 2001
`
`402
`
`Bausch Health Ireland Exhibit 2035, Page 6 of 6
`Mylan v. Bausch Health Ireland - IPR2022-00722
`
`