UBC Faculty Research and Publications

Pattern similarity study of functional sites in protein sequences: lysozymes and cystatins Nakai, Shuryo; Li-Chan, Eunice; Dou, Jinglie May 18, 2005

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-12858_2004_Article_98.pdf [ 1.63MB ]
JSON: 52383-1.0223875.json
JSON-LD: 52383-1.0223875-ld.json
RDF/XML (Pretty): 52383-1.0223875-rdf.xml
RDF/JSON: 52383-1.0223875-rdf.json
Turtle: 52383-1.0223875-turtle.txt
N-Triples: 52383-1.0223875-rdf-ntriples.txt
Original Record: 52383-1.0223875-source.json
Full Text

Full Text

ralssBioMed CentBMC BiochemistryOpen AcceResearch articlePattern similarity study of functional sites in protein sequences: lysozymes and cystatinsShuryo Nakai*, Eunice CY Li-Chan and Jinglie DouAddress: Food, Nutrition and Health, The University of British Columbia, 6650 Marine Drive, Vancouver, B.C., CanadaEmail: Shuryo Nakai* - shuryo.nakai@ubc.ca; Eunice CY Li-Chan - eunice.li-chan@ubc.ca; Jinglie Dou - jinglie.dou@ubc.ca* Corresponding author    AbstractBackground: Although it is generally agreed that topography is more conserved than sequences,proteins sharing the same fold can have different functions, while there are protein families withlow sequence similarity. An alternative method for profile analysis of characteristic conservedpositions of the motifs within the 3D structures may be needed for functional annotation of proteinsequences. Using the approach of quantitative structure-activity relationships (QSAR), we haveproposed a new algorithm for postulating functional mechanisms on the basis of pattern similarityand average of property values of side-chains in segments within sequences. This approach wasused to search for functional sites of proteins belonging to the lysozyme and cystatin families.Results: Hydrophobicity and β-turn propensity of reference segments with 3–7 residues wereused for the homology similarity search (HSS) for active sites. Hydrogen bonding was used as theside-chain property for searching the binding sites of lysozymes. The profiles of similarity constantsand average values of these parameters as functions of their positions in the sequences couldidentify both active and substrate binding sites of the lysozyme of Streptomyces coelicolor, which hasbeen reported as a new fold enzyme (Cellosyl). The same approach was successfully applied tocystatins, especially for postulating the mechanisms of amyloidosis of human cystatin C as well ashuman lysozyme.Conclusion: Pattern similarity and average index values of structure-related properties of sidechains in short segments of three residues or longer were, for the first time, successfully appliedfor predicting functional sites in sequences. This new approach may be applicable to studyingfunctional sites in un-annotated proteins, for which complete 3D structures are not yet available.BackgroundIn their recent review of protein sequence analysis in silico,Michalovich et al. [1] described the methodology fortransferring functional annotation of known proteins to anovel protein. Computer-assisted technology is used tosearch for and assign the similarity from databases of well-BLAST and PSI-BLAST, respectively. Meanwhile, the Hid-den Markov model is more efficient in searching for a dis-tant family. Furthermore, structure-based annotationconducted by using a combination of PSI-BLAST and Gen-Threader (matching of substitution energy in evolution)may facilitate rapid functional annotation from structurePublished: 18 May 2005BMC Biochemistry 2005, 6:9 doi:10.1186/1471-2091-6-9Received: 08 July 2004Accepted: 18 May 2005This article is available from: http://www.biomedcentral.com/1471-2091/6/9© 2005 Nakai et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.Page 1 of 17(page number not for citation purposes)maintained and previously annotated sources. Sequence-based and profile-based searches are conducted using[1]. However, proteins sharing the same fold can have dif-ferent functions, and structure determination and analysisBMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9will not always mean that function can be derived [2].There are examples of protein families, such as the four-helical cytokine and cytochrome super families, whosesequence similarities are either very low or not detectable[3]. Instead, their topography is more conserved thantheir sequences. This is rational, since protein functionsare classified based on function per se, regardless ofwhether their sequences or 3D structures are similar ordifferent. An example is the classification of a protein aspossessing the function of lysozyme activity, as long as theprotein possesses the ability of hydrolyzingpeptidoglycans.Another direct approach for peptide QSAR has beensimultaneously investigated in peptide sequence analysis[4]. A critical difference between those two approaches,namely bioinformatics and QSAR, is the prerequisite of3D structure information on the basis of evolutionaryconservation in the case of former; on the other hand, the3D information is helpful but not always indispensable inthe case of latter, by substituting with simpler stericparameters to account for the functional mechanism [5].For example, Hellberg et al. [4] used altogether 29 proper-ties of side chains of bioactive peptides. After dimensionreduction using principal components analysis (PCA) theresultant three main PC scores, i.e., z1, z2 and z3, represent-ing hydrophobicity, molecular size and electronic param-eter, respectively, were used as independent variables inregression analysis on the dependent variable of function-ality [4].Meanwhile, by using the homology similarity analysis(HSA), we have found the importance of functional seg-ments within 15-residue sequences of lactoferricin deriva-tives to correlate with the minimum inhibitoryconcentration (MIC) [6]. Pattern similarity constant (acorrelation coefficient) of the pattern of segments withina test derivative, in comparison to the reference pattern ofthe corresponding segment and the average of propertyvalues of the amino acid side-chains in the most potentderivative, was computed and correlated with MIC of thederivatives. In order to obtain the best (lowest) MIC, thepattern similarity should be close to 1.0 and the averageproperty value should be close to that of the referencepotent peptide (template). In the case of the above lactof-erricin derivatives, higher correlation coefficients wereobtained for log MIC predicted by HSA vs. measured logMIC computed as the output variables of regression ANN(artificial neural networks) than by sequence analysisbased on the Hellberg approach [4]. More recently, a dif-ferent approach, namely "additive QSAR" obtained bysubstituting with other amino acid residues at differentpositions in the same sequences, was reported to correlateLejon et al. [8] reported that in PCA analysis of peptidesequences, information of the positions of side chains inthe sequence should be included for improving R2X valuecompared to the results obtained by computing withoutside chain position data (0.99 vs. 0.60, respectively). OurHSA approach, by segregating segments with and withoutα-helix propensity, was in good agreement with theirs(R2X of 0.90–0.94 compared to corresponding value of0.60 but with a much larger number of derivatives). Wehave further extended this approach to infer the mecha-nism of emulsifying capacity of peptides with 10–32 resi-dues as a function of hydrophobic periodicity [9]. For thestudy of emulsification function, a new homology simi-larity search (HSS) was introduced to plot similarity con-stants and average property values of segments (3–7residues) by shifting the segment stepwise from N-termi-nus towards C-terminus of the sequences; the referencesegment used was ELE, i.e., alternate cycle of charged (E),hydrophobic (L) and charged (E) residues. However,emulsification ability is a rather general function of pep-tides that is not dependent on specific active sites withinthe sequences; overall, the emulsification ability of pep-tides was highly correlated with hydrophobic periodicityof their entire sequences.There are cases of "peptides" which do not have definitivefunctional sites but requiring specific segments, or "func-tions" which do require neither specific sites nor seg-ments. The lactoferricin derivatives described in the abovestudy [4] are an example of the former since all of themutants were prepared as derivatives of the correspondingwild-type lactoferricin 15-residue sequence, which hasdistinct helical and cationic segments. In contrast, thepeptide emulsions [9] are an example of the latter. Typicalexamples of proteins with definitive functional sites areenzymes, for which the positions of active sites are criticalto elucidate the functional mechanisms. Defective proteinfolding leading to amyloid fibril formation has been asso-ciated with various human diseases, such as Alzheimer'sand Creutzfelds-Jacob diseases. In 1993, hereditary non-neuropathic systemic amyloidosis was reported to becaused by naturally occurring variants of human lysozymethat aggregated in the liver [10]. Similarly, cystatin Cmutation in an elderly man was reported to be the causeof amyloid angiopathy and intracerebral hemorrhage[11].The recent discovery of a new-fold enzyme named Cello-syl [12] led us to select the lysozyme family in this studyas an important one to use for validating the HSSapproach to search for functional sites [13]. Meanwhile,loss of papain inhibitory activity in recombinant humancystatin C was reported to be due to insolubilization [14].Page 2 of 17(page number not for citation purposes)well with peptide functions [7]. Assuming that this loss was induced by an amyloidosis,changes of helix-to-strand in the inhibitory sites as well asBMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9the binding sites with papain could also be used for arational example of application of the HSS approach inthis study.The objective of this paper was to extend application ofthis new HSS approach to search for functional sites, suchas active and substrate binding sites in lysozyme and amy-loidosis of cystatin families, to verify the reliability of ournew method. The intention was to validate the hypothesisthat the evaluation of pattern similarity of short segmentswith 3–7 residues or even slightly longer in proteinsequences is useful in predicting functionality, assumingthat they are within allowable topographical units.Accordingly, it is not our intention to replace the 3Dapproach by the new peptide QSAR proposed in thisstudy; rather, it is anticipated to be supplemental.ResultsLysozyme familiesPCS classificationFigure 1 shows a scattergram derived from principal com-ponents similarity (PCS) analysis of 25 lysozymesequences using the hydrophobicity index of side chainsand hen lysozyme as a reference. A scattergram similar tothis figure was also obtained when a charge index wasand Streptomyces globisporus. The v-type lysozymes were T4and PA2 phage lysozymes, while g-type lysozymes werefrom goose, black swan, cassowary, ostrich and chicken G.Sixteen lysozymes (chicken C, human, horse, dog, rat,mouse, red deer, rainbow trout, pigeon, turkey, duck, Cal-ifornia quail, Japanese quail, common bobwhite, fruit flyDrosophila, and tobacco hornworm) belong to the c-typefamily. The large deviations in slope (about 2) of CH-typelysozymes as seen in Figure 1 are suggestive of explicit dif-ference in their molecular structures from those of otherlysozyme types.Segment pattern similarity search for active sitesThe HSS computer program was applied to the sequencealignment patterns shown in Figure 2 using the active sitesof hen lysozyme at positions 54–57(F34ESN) and 80–83(T51DYG) as the references. The position numbers inparenthesis are the positions in un-gapped sequences ofindividual lysozymes, while the immediately precedingnumbers outside of parenthesis are the position numbersof the multiple sequence alignment (gapped) in Figure 2.Since the peptidoglycan-lysing activity is associated withglutamic 55(E35) and aspartic 81(D52) side-chains in thehen lysozyme [13], the segments flanking these residuesin the sequences were the focus of pattern similaritysearch using the HSS.The search patterns illustrated in Figure 3A (charge) and3B (turn propensity) are a trial run to validate the HSSapproach, applied to human lysozyme vs. hen lysozyme,employing the five residues flanking E35 (K33FESN of hen)as a search unit. The rules herein for selecting active seg-ments are that the greater the similarity approaching to1.0 and the nearer the average value to that of the refer-ence, the more likely to be active sites in the test sequence.In Figure 3A, in addition to E35 and D53, three residues ofE7, S80 and D120 show about the same similarity constants(upper pattern) as well as similar average charge values(shown by arrows on the lower pattern). However, noneof the above latter three positions are acceptable as theactive site, as N44, I59, C65, Q86, and Q126 in Figure 3B forβ-turn search do not have matching peaks in Figure 2A forcharge search. Therefore, only E35 and D53 are qualified asthe active positions of human lysozyme. The same resultwas obtained when five residues flanking D52 (S50TDYG)of hen lysozyme were used as an alternative reference seg-ment. The two regions around E35 and D53 determined forhuman lysozyme are in good agreement with those beingreported for the active site in the literature [13]. Despitethe fact that the charge is of prime importance in definingactive sites of lysozymes, the turn propensity values of thesegment rather than its pattern similarity appear to playan important role in the enzyme, as exemplified in a low53 35PCS scattergram of lysozyme families (C-, G-, V-, and CH-types) wh n hydr phobicity w s us d as property i exFigure 1PCS scattergram of lysozyme families (C-, G-, V-, and CH-types) when hydrophobicity was used as property index. Hen lysozyme was used as the reference with [coefficient of determination] = 1.0 and [slope] = 1.0.Page 3 of 17(page number not for citation purposes)used (instead of the hydrophobicity index) for classifica-tion. CH-type lysozymes used herein were from a fungussimilarity value of 0.30 for D compared to 0.98 for E ,whereas similar average turn values of 1.2 and 1.1,BMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9respectively, were computed (Fig. 3B). All these results The same search was conducted for goose and T4 lys-Multiple sequence alignment of five lysozymes belonging to four familiesFigure 2Multiple sequence alignment of five lysozymes belonging to four families.Page 4 of 17(page number not for citation purposes)may imply that the exposure of active sites to react withthe substrate binding sites is essential.ozymes as well as the new fold CH-lysozyme [12], withthe assumption that it is an un-annotated sequence, toBMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9confirm validation of the HSS approach (Table 1). Twopositions of 79(E73) and 103(D97) were determined as theactive sites of the goose lysozyme sequence. D97 may notbe essential for the catalytic activity of the goose lysozyme[15]. In comparison, 11(E11) and 20(D20) were noted forT4 lysozyme, which are in good agreement with thosereported in the literature [13]. As shown in Figure 4, in thecase of the sequence of Cellosyl (CH-type), 14(D9),104(D98) and 106(E100) were identified as candidates tobe the active positions of catalysis, which are in goodagreement with Rau et al. [12]. These results would sup-port the reliability of the HSS approach.HSS search for substrate binding sitesCompared to the active sites that are explicitly negative incharge at single positions, the substrate binding sites oflysozymes are rather loosely defined, mainly due to theis generally agreed that the substrates locate inside thecleft formed between the helix lobe and the strand lobe ofmost lysozyme molecules, except for Cellosyl, the bacte-rial muramidase from Streptomyces coelicolor, which can beattributed to structural difference in the catalytic crevice[12]. The six-residue segment at alignment positions 84–89 (I55LQINS) of hen lysozyme was employed as the ref-erence segment [16] using the hydrogen bonding scale(Table 2). Based on the high pattern similarity constantsand the average hydrogen bonding index values, eightpotential sites were identified for hen and human lys-ozymes. The same rule as that for selecting active site wasused herein for selecting of binding sites. Position77(D48) and 188(W111) shown in "Hen 1" of Table 2 withlower similarity constant and hydrogen bonding value,respectively, may not be the potent substrate binding sitesin the hen lysozyme. In the 3D structure [17], those twoHSS search patterns for active sites of human lysozymeFigur  3HSS search patterns for active sites of human lysozyme. Segment 53–57 of hen lysozyme was used as the reference. A: HSS search pattern based on charge. B: HSS search pattern based on turn propensity.Page 5 of 17(page number not for citation purposes)greater complexity of determining the 3D structure ofenzyme-substrate complexes than that of enzyme alone. Itpositions are far away from the catalytic cleft where thesubstrates snugly fit in. In the human lysozyme,BMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9188(W112) is more likely to be the binding site than102(A73), with higher pattern similarity and strength ofhydrogen bonding of 0.89/0.45 than 0.80/0.32, respec-tively. Not only strong hydrogen bonding, but also highpattern similarity of a segment may be required to bequalified for substrate binding sites.For the goose lysozyme, six potential binding positionswere detected (Table 2); I113 and G163appear more likelyto be the binding sites than other positions consideringthe location of the cleft in the molecule. Similarly, ninesites were found to be potential sites of T4 lysozyme, espe-cially three positions, i.e. M6, L66 and S136 (Table 2). In thecase of Cellosyl, eight positions were identified as thepotential binding sites (Table 2). Probably due to the con-siderable 3D-structure difference of this lysozyme fromthose of other lysozyme families [12], the alignment posi-tions 40–45 (T34EGTNY) instead of hen's 84–89(I55LQINS) were used as a reference segment for obtainingmore rational search results. Instead of hydrogen-bondingmotivated interactions, less polar van der Waals interac-tion with the aromatic side chains in CH-lysozyme maybe regarded as the second important stereochemical forcesin the substrate binding [16].In the literature, the most frequently cited substrate-bind-ing sites in c-type lysozymes family have been W62 andD101 of hen lysozyme [13]. Since Figure 2 includes the dis-tant family of CH-type, the segment similarity search wasrepeated within the c-type lysozymes alone to restrict thesearch within similar fold. The results are shown as "hen2" and "human 2" in Table 2. Those results almost per-fectly match to the substrate binding mechanism based on101 103 104 107these side-chains are very close or adjacent to the seg-ments listed in "Hen 1" and "Hen 2" of Table 2.Substrate binding sites reported by site-directed mutagenesisAmong three mutants obtained by replacing W62 with Y, For H, the W62H mutant, and especially the doublemutant W62H/D101G, reduced substrate binding drasti-cally [15]. This change can be explained by a decrease inthe hydrogen bond average value from 0.58 to 0.54 andfrom 0.46 to 0.29 in V62H and D101G, respectively,when the index values employed in this study were usedin computation. The double mutant changed substrate-binding mode while maintaining the overall proteinstructure almost identical to that of the wild type [18]. Anextensive cluster of hydrophobic structure is involved indistinct regions of the sequence, but is all disrupted by asingle point mutation of W62G located at the interface ofthe two major structural domains in the native lysozyme[19]. Similar effects were observed in mutants Y63L andD102E of human lysozyme [20]. The double mutantsR41N/R101S and V74R/Q126R of human lysozyme werebetter catalysts for lysis of Micrococcus lysodeikticus [18].The average hydrogen bond value of both R41N andR102S was shown to increase in our HSS search, but sim-ilar effects could not be observed for V74R/Q126R. Aninteresting finding is that these two mutations have bothresulted in the side chains being identical to those of henlysozyme. R41 and V74 are near A42 and A73, respectively.Importance of R115 in substrate binding of human lys-ozyme was reported [21], which is in good agreement ofW112 within the same subsite F (Table 2).Table 1: Determination of active sites in sequences of lysozymes in different familiesFamily Active sites Position Charge 1 Charge 2 Turn 1 Turn 2Hen 1 55(35)E 1.0/5.0 .98/5.0 1.0/1.122 81(52)D .98/5.2 1.0/5.2 1.0/1.21Hman 1 55(35)E 1.0/5.9 .98/5.0 .98/1.172 81(53)D .98/5.2 1.0/5.2 1.01/1.22Goose 1 79(73)E .67/6.9 .78/6.9 .06/1.032 103(97)D .83/7.3 .83/7.3 .67/,92T4 1 11(11)E .61/4.5 .55/4.5 .76/1.162 20(20)D .55/5.4 .55/5.4 .53/1.07CH 1 14(9)D .98/5.2 1.0/5.2 .95/.952 104(98)D .74/4.5 .65/4.5 .81/.923 106(100)E .95/5.6 .86/5.6 .87/.97Position: the first number is the multiple sequence alignment position shown in Figure 2. Number in bracket is that in the sequence of each lysozyme. Charge 1 and 2 and Turn 1 and 2: Sites 54–57 and 80–83 (alignment position numbers in Fig. 2) were used as references. These position numbers correspond to 34–37 and 50–54, respectively in the sequence of hen lysozyme. The first and second digits n each cell of Charge and Turn are similarity constant and average of property value.Page 6 of 17(page number not for citation purposes)X-ray crystallographic analysis, e.g. D , N , N , A ,V109, E35, N46, V110, E52, N59, and W63 [15]. Almost all ofBMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9CystatinsPCS classificationThe PCS scattergram of the 17 cystatins using humanof the index to α-helix and β-strand propensities did notappreciably change the grouping results. The three groupsinclude human cystatins C, D, S, SA, SN and hen cystatinHSS search patterns for active sites in Strep. coelicolor lysozymeFigur  4HSS search patterns for active sites in Strep. coelicolor lysozyme. Segment 79–83 (STDYG) of hen lysozyme was used as the reference based on charge.Page 7 of 17(page number not for citation purposes)cystatin C (HCC) as a reference and hydrophobicity asside-chain property index is shown in Figure 5; alteration(Group I), human cystatins E, F and M (Group II), andhuman cystatins A and B (Group III) which are the stefinBMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9group cystatins that are smaller in molecular size and haveslightly lower papain inhibitory activity than HCC [22].Active sitesThe HSS patterns using hydrophobicity to search foractive sites of the four cystatins (HCC, EWC, HCA andHCB), when L9VGG of HCC has been employed as the ref-erence, are shown in Figure 6. The active sites shown witharrows are at about the same location in all cystatinssequences used in this study. Similar results were obtainedwhen bulkiness was used as side-chain index. In thesecases, the similarity peaks are the major clue for identify-ing active sites. However, when similarity peaks appear inthe neighbourhood, the average index values wouldbecome more reliable for identifying active sites as shownin Figure 6B; hen cystatin has two probable active posi-tions side by side with the same similarity constants. Theactive sites, therefore, should be near the N-terminus withProsite-type patterns of [L,I,M]-x(4)-G- [G,A]; the activesites are L9VGG and L7LGA in human cystatin C and hencystatin, respectively. The similarity constants and theaverage hydrophobicity of the active sites are shown inTable 3. Stefin B (HCB) shows much lower similarity con-stant and average hydrophobicity than those of othercystatins. This result is in good agreement with ki differ-Substrate binding sitesA HSA study similar to our previous paper [6] was con-ducted at the active and two binding sites of cystatins,yielding results (Table 3) which are in good agreementwith Turk et al. [22]. Substrate-binding site 1 had the pat-tern Q-x(3)-V- [S,A]-G, while substrate-binding site 2 hadthe pattern [L,I,V]-P-x(3)-x(3)- [N,G]. Similarity constantsof binding loop 2 of egg white cystatin (EWC) and HCAare lower than that of HCC, whereas not only similaritybut also average hydrophobicity are lower in HCB.Similarity constants at the active site (against 1.0 for HCC)using hydrophobicity index were >0.8 for cystatins A, D, Fand hen, ~0.5 for E and M, and 0.1–0.2 for cystatins B, S,SA and AN. Similarity constants at binding loop 2, whenstrand propensity was used for PCS computation, werelower for stefins A and B with values of 0.8 and 0.6,respectively, than >0.9 for other cystatins. On the otherhand, similarity constants for strand at binding site 1 werenot much different among different cystatins, with values>0.9 (not included in Table 3).It is interesting to note that stefins A and B do not have thePW pair which is in the binding site 2 of HCC and EWC;instead they have PG and PH pairs, respectively (Table 3).The W → G replacement increased strand propensity,Table 2: Determination of substrate binding sites in amino acid sequences of lysozyme familiesPotential sitesHen 1 55(35)E 70(42)A 77(48)D 84(55)I 90(61)R 102(72)S 117(83)L 118(111)W.81/.71 .79/.53 .34/.65 1.0/.41 .56/.58 .78/.56 .61/.42 .66/.35Hen 2 36S 45R 49G 56L 62W 73R 90A 100S 112R.82/.71 .86/.44 .79/.47 1.0/.39 .59/.58 .91/.47 .98/.41 .55/.46 .76/.29Human 1 54(34)W 70(42)A 77(49)D 84(56)I 90(62)R 102(73)A 116(83)A 118(112)W.48/.52 .42/.51 .34/.65 .91/.47 .50/.61 .80/.32 .61/.46 .89/.45Human 2 27N 43T 50R 57F 63Y 74V 84L 101R 117Q.56/.39 .56/.51 .61/.47 1.0/.45 .55/.61 .72/.33 .48/.47 .58/.29 .79/.45Goose 43(73)I 99(93)L 119(113)I 143(133)S 181(143)G 201(163)G.85/.31 .76/.34 .67/.46 .81/.55 .82/.36 .81/.58T4 6(6)M 16(16)K 53(49)A 66(57)V 75(66)L 94(85)K 123(112)A 186(136)S 207(153)F.74/.34 .99/.38 .78/.30 1.0/.42 .52/.54 .67/.38 .73/.45 .58/.51 .78/.36CH 40(35)T 47(42)D 55(50)T 80(74)A 92(86)W 124(118)T 153(147)C 186(180)T1.0/.58 .63/.52 .80/.51 .51/.44 .73/.49 .83/.59 .51/.56 .56/.54The first/second digits are similarity constants and average hydrogen bond index values of binding site with segments of sox residues beginning with the positions shown. Bold digits show more likely binding sites than non-bold digits.Hen 1 and Human 1: from the alignment shown in Figure 2; Hen 2 and Human 2; from the alignment of the lysozyme C family exclusively.Page 8 of 17(page number not for citation purposes)ence reported by Abrahamson [23]. while W → H replacement did so moderately. The valuesshown in Table 3 were almost inversely proportional toBMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9PCS scattergram of cystatins when hydrophobicity was used as property indexFigure 5PCS scattergram of cystatins when hydrophobicity was used as property index. Human cystatin C (HuC) was used as the reference. Human cystatins M, E, S, SA, SN, D, F, A and B are labelled as 1, 2, 7, 8, 9, 10, 15, 16 and 17. Labels 3, 4, 6, and 11–14 are for cystatins from mouse C, rat C, bovine, hen, rainbow trout, chum salmon and carp, respectively.Table 3: Active and substrate binding sites of cystatinsHCC EWC HCA HCBActive siteSegment L9VGG L7LGA I2PGG M2SGASimConst/Av.Hydroph* 1.00/1.35 0.87/1.80 0.97/1.00 0.55/-0.01Binding site 1Segment Q55IVAG Q53LVSG Q46VVAG Q46VVAGSimConst/Av.Hydroph 1.00/0.90 0.70/0.92 0.90/0.95 0.90/0.95Binding site 2Segment V104PWQG I102PWLN L73PGQN L73PHENSimConst/Av.Hydroph 1.00/1.01 0.35/1.95 0.30/0.98 -0.16/0.42SimConst/Av.Turn** 1.00/1.05 0.90/0.96 0.56/1.22 0.99/1.06* Pattern similarity constant / average hydrophobicity** Pattern similarity constant / average turn propensitySimilar to the corresponding tables for lysozymes (Tables 1 and 2), Sim/Const is 1.0 for reference HCC, and the greater the Av.Property the greater the property strength.Page 9 of 17(page number not for citation purposes)BMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9the equilibrium inhibition constant ki except for humancystatin S that was weak in the inhibitory activity, whichmight have been due to the difference in phosphorylationof serine at N-terminal region [23]. Although stefins A andB are classified differently from other groups on PCS scat-tergram (Fig. 5), the weak binding at the binding site 2may not have considerable effects on the ki values.Turk et al. [22] have stated that the differences in the bind-ing constants between cystatins and various cysteine pro-teases arise primarily from differences in the structure ofenzyme active site clefts. The inhibition of endopepti-dases, i.e. papain and cathepsins S and L, by cystatins isextremely tight and rapid, whereas the inhibition offree to accommodate inhibitors, while in the case ofexopeptidases, the active site cleft contains extra residuesin it. In the N-terminal region of cystatins, it was observedthat the affinity for target proteases decreased with bothsize and charge of substituting residues [22]. Theseobservations are in good agreement with the results whenbulkiness of side chains was used for the HSS computa-tion for the binding site 1, "SimConst /Av.bulk" valueswere 1.00/12.3, 0.92/14.2, 0.98/11.4 and 0.91/11.34 forHCC, EWC, HCA and HCB, respectively. As expected, ste-fins A and B were less bulky. Furthermore, HCC and EWCinclude longer chains at the N-terminal sides with bulkierresidues than those of stefins. These findings are in goodagreement with the effect of bulkiness of G4 in the stefin AHSS search patterns for active sites of cystatins against human cystatin C based on hydrophobicityFigur  6HSS search patterns for active sites of cystatins against human cystatin C based on hydrophobicity. A: HCC (reference), B: EWC (egg white cystatin), C: HCA and D: HCB.Page 10 of 17(page number not for citation purposes)exopeptidases, i.e. cathepsins B and H, is considerablyweaker. The active site cleft of known endopeptidases issequence, implying that the bulkier the residue at position4, the weaker the papain inhibitory activity [24].BMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9AmyloidosisLysozymesSimilarity constants and average propensities of α-helixand β-strand were computed for G54IL and G54ILQIN ofhen lysozyme (G55IF and G55IFQIN in the case of humanlysozyme) as shown in Table 4. The amyloidogenicmutant I55T showed increased strand propensity from0.82 to 0.85, without a substantial change in the similarityconstants. With regard to the helix structure, the similarityconstant decreased from 1.00 to 0.93 while the averagevalue increased from 1.08 to 1.14, thereby implicating adecrease in helix, since the helix index used herein isinversely related to the content of helix structure. Thesechanges are favourable for amyloidosis. The fact that posi-tion 56 in human lysozyme is near its active position atD53 may explain its dramatic effect on the enzymaticactivity, more effective than other positions in thesequence. The six-residue computation did not show asclear a difference as was found in the three-residue com-putation. Nine double-site mutations in addition to I55Tat other positions selected at random in the sequences ofthe amyloidogenic mutant I55T by using the RCG pro-gram did not restore the activity of wild-type lysozyme(unpublished). These results infer that the amyloid, onceformed by detrimental mutation at the active site, cannotbe restored by mutation at other locations in thesequence.CystatinsHeat treatment of HCC induced its dimer formation at anearly stage of separation, resulting in a complete loss of itsactivity [25]. Based on a dramatic decrease in the mono-mer form as shown by its CD spectrum, polymerization(positions 1–35) of the helix domain of HCC, 17 residueswere mutated in the 22 single-site mutants using the RCGprogram [14 (Table 1)]. When 33 mutants obtained byadding one extra residue each in both side of the original17 residues after eliminating duplication were used forPCS computation, the resultant PCS demonstrated thathelix propensity and bulkiness were playing importantroles in thermostability (data not shown). Employmentof three residues flanking the mutated residue was impor-tant in pattern similarity computation. This conclusion isin good agreement with Hall et al. [26] who did anexhaustive study showing that mutations at positions 8–10 enhanced thermostability of cystatin. With regard tothe papain inhibitory activity, the importance ofhydrophobicity and bulkiness was demonstrated (the PCSscattergrams, similar to Fig. 7, are not shown here).Of 86 residues in the strand domain (positions 36–121)of HCC, 21 residues (positions 36–120) were mutated inthe 23 double-site mutants using the RCG program [14].Thirty-seven residues were used for the PCS computationof single-site mutations as described above. Hydrophobic-ity appeared to be playing an important role inthermostability, while strand propensity was importantfor inhibitory activity (data not shown). Strand and helixpropensities in the strand domain were influential to thepapain inhibitory activity of HCC (Fig. 7). The figuresshow 12 data points only by eliminating data from singlesite mutation in the helix domain, which did not showdistinct trends with broader scatter in these figures. Thesecond mutations in addition to the above single muta-tions were conducted at the strand domain of the enzyme[14]. Coefficient of determination of 1.0 and slope of 1indicate perfect match with the reference sample (5*) thatis mutant G12W/H86V with the lowest strand propensityalong with highest helix propensity in the strand domainamong 23 double mutants. It is worth noting that the PCSis a classification program comparing pattern similaritywithout demonstrating quantitative relationships withfunctions but providing the information of the extent ofinvolvement of side chain properties in the functions ofinterest.For mutant G12W/H86V that gained the greatest activityincrease of 4.98 ± 0.09 times (mean ± SD at n = 3) that ofrecombinant wild-type [14], the strand propensitydecreased from 0.78 to 0.69 (H86V) with a slight increasein the helix propensity (corresponding to decrease in theindex values). The same was true for mutant D15P/H86Iwith 2.65 ± 0.30 times activity increase. A similar resultwas observed in mutant G4L/D40I with 2.11 ± 0.29 timesactivity increase, due to strand decrease along with almostno change in helix (D40I). However, mutant V10S/R93GTable 4: HSA computation for I55T lysozymeHen Human I55T (Hen)HelixG54IL(F)QINSSimConst* 1.000 0.917 0.952Average 0.960 0.993 0.985G54IL(F)SimConst 1.000 1.000 0.927Average 1.082 1.082 1.141StrandG54IL(F)QINSSimConst 1.000 0.978 0.997Average 0.788 0.762 0.797G54IL(F)SimConst 1.000 1.000 0.998Average 0.822 0.822 0.845L56 for hen and F57 for human. * Similarity constant.Page 11 of 17(page number not for citation purposes)such as amyloidosis could be a cause of the loss of papaininhibitory activity of mutated HCC [14]. Of 35 residueswith 4.50 ± 0.07 times activity increase behaved differ-ently with increased strand and simultaneous decrease inBMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9helix. It is worth noting that the single mutation of V10Salone increased the activity 2.96 ± 0.06 times, therefore,changes in the helix domain may have a more predomi-nant effect on the inhibitory activity than mutations in thestrand domain. The activity change due to mutation helix→ strand in the strand domain in the sequence may beslight in this case.It is well known that a single-point mutation of humanlysozyme, namely I56T, has been identified as the originof hereditary systemic amyloidosis [27]. The amyloidog-enic nature of the lysozyme variants arises from a decreasein the stability of the native fold relative to partially foldedintermediates. Accordingly, in a low population ofsoluble, partially folded species, the protein can aggregatein a slow and controlled manner to form amyloid fibrils.Similarly, sporadic amyloid angiopathy and intracerebralhemorrhage was reported in an elderly man due tocystatin C mutation [11]. In the case of human cystatin C,the decrease in strand along with an increase in helixmight have prevented amyloidosis, despite the fact thathelix change was not always as evident as in the case ofmutation optimization [14] may be due to lack of the dataof single-site mutation in the strand domain of the cysta-tin sequence. Unfortunately, the objective of that study[14] was for mutation optimization and not for investiga-tion of the mechanism of amyloidosis. It has beenreported that stefin B (HCB) readily formed amyloid [28],which may imply declined importance of the role beingplayed by the binding site 2 in amyloidosis of HCC.DiscussionIn a review on the quest to deduce protein function fromsequences [29], the author stated that the searching of pat-tern databases would be more sensitive and selective thansearching of sequence database. It was predicted that thesequence pattern databases, especially by comparing thepattern similarity, would play an increasingly importantrole, as the post-genome quest to assign functional infor-mation to raw sequence data gains pace [29]. Pattern sim-ilarity computation requires at least three residues insegments to represent a nonlinear curve, which is unlikelyto be due to the effect of a single point mutation per se.Effects of mutating the strand domain of human cystatin C (mutated 21 residues)Figure 7Effects of mutating the strand domain of human cystatin C (mutated 21 residues). A: Strand index, B: Helix index. The numbers show multiples of activity increases from wild-type. 1 shows no increase of inhibitory activity. 5* is the reference mutant 12W86V with the highest inhibitory activity.Page 12 of 17(page number not for citation purposes)lysozyme. Some inconsistency in the amyloidosis as acause of inhibitory activity of human cystatin C in ourWith regard to an apparent effect of the single residuemutations of hen lysozyme on substrate binding, theBMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9structural analysis by NMR of the position-62 mutant ofhen lysozyme [18,30] found major changes in the chemi-cal shift of back bone protons, especially in a loop region(positions 61–78), which contains W62 influencing thelocal folding. Similarly, Muraki et al. [20] reported thatcompared to the wild-type human lysozyme, the N-acetylglucosamine residue at subsite B of the L63 mutantmarkedly moved away from the 63rd residue, with sub-stantial loss of hydrogen-bonding interaction. In Figure 5of Ref 17, involvement of not only Y63 but also W64 isevident. These results are supportive of the importance ofpattern similarity of ≥ 3 residues, which are affected bysingle-residue mutation.The predictability of the active and binding sites solely onthe basis of protein sequences [31] may be useful forinvestigating the underlying mechanisms of unknownfunctions of human genes after translation to proteinsequences. Usually, two or three essential residues aredirectly involved in the bond making and breaking stepsleading to formation of enzyme catalysis; however, theremoval of an essential group often does not abolish activ-ity, but can significantly alter the catalytic mechanism[32]. T4 lysozyme was cited by Peracchi [32] as an exam-ple of the alteration of catalytic mechanisms; the lyticactivity of lysozyme changed to that of a transglucosidase.An approach utilizing the property of side chains in asequence for identifying functional motifs has alreadybeen utilized in the computer-assisted selection of anti-genic peptide sequences [33]. The authors stated that anantibody produced in response to a simple linear peptidewith 7–9 residues in a protein would most likely recog-nize a linear epitope. Furthermore, this epitope must besolvent-exposed to be accessible to the antibody. In a largescale data mining study, Binkowski et al. [34] describedthe importance of local sequence and spatial surface pat-terns in inferring functional relationships of proteins. Thegeneral feature of protein structure that would correspondto these criteria could be turns or loop structures, whichare generally found on the molecular surface connectingto other elements of secondary structure, and the area ofhigh hydrophobicity, especially for those containingcharged residues.Successful identification of active sites of new-fold of CH-lysozyme using the HSS approach in this study suggeststhat this approach could be applied to query proteinstranslated from unknown RNA segments of the humangenes against templates with known functions, when their3D structure information is still unavailable. It has beenshown that the inhibition of the papain family by cystatinis due to a tripartite wedge-shaped structure with a goodrelationship as seen in Figure 8, lysozyme functions as anO-glycosyl hydrolase, while α-lactalbumin lacks thisactivity and instead regulates the substrate specificity ofgalactosyltransferase. The active site of peptidoglycan lysisis disrupted in α-lactalbumin. The ESS computationshowed that although a pattern equivalent to hen's D53exists in the form of TEYG/YDYG, there is no residueequivalent to hen's E35 around corresponding positions asin the form of HTSG/WESG. Two side-chain carboxyl rad-icals are required for the lysozyme activity within the crev-ice between the helix and strand domains of the moleculebelonging to C-type family [16]. According to Alvarez-Fernandez et al. [35], the three parts of the cystatinpolypeptide chain included in the enzyme-bindingdomain are the N-terminal segment, a central loop-form-ing segment with motif QXVXG and second C-terminalloop typically containing a PW pair [31,37,38].For multiple sequence alignment of an uncharacterizedprotein or peptide, many Web alignment servers are avail-able for use [1], such as Blast and NPSA, as was done inthis study. For classification of uncharacterized sequences,the PCS scatterplots are also useful as shown in Figures 1,5 and 7. The PCA demonstrated the classifying capacitysuperior to that of distance-based cluster analysis [39].The PCS is more flexible than cluster analysis as differentpattern similarity patterns can be drawn by rotating thereference segment for searching. It implies that similarityis not always [1 – dissimilarity]. This difference resulted inthe possibility of selecting outliers, which is critical inderiving true classes or ranking [40]. Most of the currentlyavailable peptide QSAR, such as the method of Hellberget al. [4], intends to be based on whole sequence data. Thenew HSS approach reported in this study could be just thebeginning of more detailed, reliable peptide QSAR to bedeveloped in the future. Analysis of a variety of bioactiveproteins contributing to human health is a potentialfuture application of the HSS software package as well asmultifunctional PCS. Considering the multifunctionalnature of human diseases, the functionality of food pro-teins also can be manipulated based on combinations ofbioactive segments in different or even single natural pro-tein sequences. Therefore, for an uncharacterized proteinor peptide, a new plan is proposed: (1) A referencesequence is chosen from multiple sequence alignment(MSA) as discussed above; PCS scattergrams would assistthis selection in addition to BLAST search. (2) Based onsegments with high similarity in MSA, segments to beused for search are selected within the reference sequence.Then, (3) HSS search is conducted to identify functionalsegments in the uncharacterized sequence. (4) From theabove PCS computation, important PC scores arescreened (PCA is a subroutine subprogram of PCS). (5)Page 13 of 17(page number not for citation purposes)supplement to the active site clefts of the enzyme [35].Todd [36] stated that despite highly homologousRegression neural networks are conducted using selectedPC scores as input variables as exemplified in our lactofer-BMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9An example of enzyme vs. non-enzymeFigure 8An example of enzyme vs. non-enzyme. Adopted from Figure 10.3b [36]. Fig. 8A: 1IWT human α-lactalbumin, Fig. 8B: 1B9O human lysozyme C using Swiss-PdbViewer (spdbv). Blast 2 sequences [1] showed that the identities and the positives Page 14 of 17(page number not for citation purposes)between the two proteins were 35% and 55% respectively.BMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9ricin derivative study [6]. (6) RCG would be useful forconfirming the HSS data and also to find the best segmentor sequence as exemplified in our HCC mutation [14].One of the original purposes of our new approach inunsupervised data mining was to verify the hypothesisthat there might be adaptability of different human cystat-ins to better inhibit different human cathepsins [41]. Thishypothesis has not been fully pursued in the past, proba-bly because of costly separation of pure cystatins andcathepsins. An advantage of our approach is to derivepotential hypothesis for enzyme/substrate interactionsexclusively from their sequence data. Although the verifi-cation of those hypotheses may need to await future 3D-structure study, it is important that most of the usefulQSAR data could become available, which would pro-mote the functional mechanism study based on 3D struc-ture. However, we admit that more examples ofapplication should be performed in the future to morethoroughly verify and establish this method for predictingfunctions based on sequences. This work is underway inour laboratory.ConclusionAlthough the importance of pattern similarities of motifswith 20–30 residues as a whole has been reported for pep-tide QSAR in the past, the importance of a search for seg-ments with three or more residues as functional sites ofprotein sequences has not been investigated. Lysozymesand cystatins were used as examples of proteins to demon-strate the capacity of segment pattern similarity analysis topredict functions, such as active and binding sites,amyloidosis and thermostability as a tool for quantitativefunctional sequence analysis.MethodsAmino acid sequences of proteinsMultiple sequence alignments of lysozymes were con-ducted using the Network Protein Sequence Analysis ofPôle Bio-Informatique Lyonnais [42] based on Clustal W.Similarly, multiple sequence alignments were obtainedfor human cystatins A (HCA), B (HCB) and C (HCC) andhen egg white cystatin (EWC) as well as for papain as hostproteases of cysteine protease inhibitors, i.e. cystatins. ForPCS analysis, a total of 17 cystatins were used: human A,B, C, D, E, F, M, S, SA, SN, hen (EWC), bovine, ratC,mouseC, Chum salmon, Rainbow trout and carp.Principal components similarity analysis of protein sequencesThe method described in the previous papers [9,39] wasfollowed. Principal components analysis (PCA) was mod-ified to principal components similarity (PCS) byplot. The PCS was then modified to apply to peptidesequences.Homology similarity searchHomology similarity search (HSS) was conducted asreported previously [9]. The similarity constant used inthis study is eventually a correlation coefficient [43]. Apreliminary study was carried out by changing the size ofsegment (normally 3–7) flanking the potential functionalposition to determine the most appropriate size of seg-ment in differentiating the functional site from othersegments within the sequence of lysozymes and cystains.The property indices used for amino acid side chains werehydrophobicity, charge, propensities of α-helix, β-strandand β-turn, hydrogen bonding, and bulkiness as reportedpreviously [14,44]. Segments with pattern similarity closeto 1.0 and average values similar to that of the referencesegment were sought within each gapped sequence.All software used in this study along with the instructionson how to use the computer programs are available in theform of ftp files on the Web [45] to download to PCcomputers.List of abbreviationsEWC Egg white cystatin or hen cystatin.HCC Human cystatin C.HCA Human cystatin A or stefin A.HCB Human cystatin B or stefin B.HSA Homology similarity analysis: the PCS software wasmodified to compute pattern similarity constants andaverage side-chain property index values of segments insequences [6].HSS Homology similarity search: A step-wise search pro-gram initiated from N-terminus of query sequences byshifting the search unit (reference segment) towards C-ter-minus based on similar segments in terms of patternsimilarity constant and average property values comparedto those of template sequences [9].MIC Minimum inhibitory concentration.PCA Principal components analysis.PCS Principal components similarity: PCA modified formulti-functional variables using linear regression of devi-ation of PC scores on the reference PC scores. Scatter plotis drawn as slope vs. coefficient of determination (r2) [44].Page 15 of 17(page number not for citation purposes)incorporating linear regression of PC scores to be able toaccount for more than three PC scores on a 2D scatterBMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/9RCG Random-centroid optimization of site directedmutagenesis.QSAR Quantitative structure-activity relationships.Authors' contributionsEL participated in laboratory investigation to verify thehypothesis set in this study, while JD was taking care ofthe computer programming. SN was responsible mainlyfor the creation and application of software used in thisstudy. All authors read and approved the final manuscript.AcknowledgementsThis work was financially supported by a Multidisciplinary Network Grant entitled "Structure-function of food biopolymers" (Dr. Rickey Y. Yada of University of Guelph as the principal investigator) from the Natural Sci-ences and Engineering Research Council of Canada. The authors acknowl-edge the collaboration of all co-authors listed in our past publications as shown in the following references. Especially, the drawing of 3D structures to compare lysozyme and α-lactalbumin by Dr. Yasumi Horimoto is highly appreciated.References1. Michalovich D, Overington J, Fagam R: Protein sequence analysisin silico : application of structure-based bioinformatics togenomic initiatives.  Cur Opinion Pharmacol 2002, 2:574-580 [Blast(www.ncbi.nlm.nih.gov/blast), NPSA (http://npsa-pbil.ibcp.fr/)].2. Norin M, Sundsröm M: Structural proteomics: developments instructure-to-function predictions.  Trends Biotechnol 2002,20:79-84.3. Hill EE, Morea VM, Chothia C: Sequence conservation in familieswhose members have little or no sequence similarity: thefour-helical cytokines and cytochromes.  J Mol Biol 2002,322:205-233.4. Hellberg S, Sjöström M, Skagerberg B, Wold S: Peptide quantita-tive structure-activity relationships, a multivariateapproach.  J Med Chem 1987, 30:1126-1135.5. Giliani A, Benigni R, Zbilut JP, Webber CL Jr, Sirabella P, Colosimo A:Nonlinear signal analysis methods in the elucidation of pro-tein sequence-structure relationships.  Chem Rev 2002,102:1471-1491.6. Nakai S, Chan JCK, Li-Chan EC, Dou J, Ogawa M: Homology simi-larity analysis of sequences of lactoferricin and itsderivatives.  J Agric Food Chem 2003, 51:1215-1223.7. Doytchinova IA, Walshe VA, Jones NA, Closter SE, Borrow P, FlowerDR: Coupling in silico and in vitro analysis of peptide-MHCbinding: A bioinformatics approach enabling prediction ofsuperbinding peptides and anchorless epitopes.  J Immunol2004, 172:7495-7502.8. Lejon T, Ström MB, Svendsen JS: Is information about peptidesequence necessary in multivariate analysis?  Chemom Intell LabSyst  2001, 57:93-95.9. Nakai S, Alizadeh-Pasdar N, Dou J, Buttimor R, Rousseau D, PaulsonA: Pattern similarity analysis of amino acid sequences forpeptide emulsification.  J Agric Food Chem 2004, 52:927-934.10. Pepys MB, Hawkins PN, Booth DR, Virgushin DM, Tennent GA,Soutar AK, Totty N, Nuguyen O, Blake CCF, Feest TG, Zalin AM,Hsuan JJ: Human lysozyme gene mutations cause hereditarysystemic amyloidosis.  Nature 1993, 362:553-557.11. Graffagnino C, Herbstreith MH, Schmechel DE, Levy E, Roses AD,Alberts MJ: Cystatin C mutation in an elderly man with spo-radic amyloid angiopathy and intracerebral hemorrhage.Stroke 1995, 26:2190-2193.12. Rau A, Hogg R, Marquardt R, Hulgenfeld R: A new lysozyme fold:crystal structure of the muramidase from Streptomyces coe-licolor at 1.65 Å.  J Biol Chem 2001, 276:31994-31999.14. Ogawa M, Nakamura S, Scaman CH, Jing H, Kitts DD, Dou J, Nakai S:Enhancement of proteinase inhibitory activity of recom-binant human cystatin C using random-centroidoptimization.  Biochim Biophys Acta 2002, 1599:115-124.15. Strynadka NCJ, James MNG: Lysozyme: a model enzyme in pro-tein crystallography.  In Lysozyme: Model Enzymes in Biochemistryand Biology Edited by: Jollès P. Berlin: Birkhäuzer Velag; 1996:185-221. 16. Imoto T: Engineering of lysozyme.  In Lysozymes: Model Enzymesin Biochemistry and Biology Edited by: Jollés P. Berlin: Birkhäuser Verlag;1996:163-181. 17. Song HW, Inaka K, Maenaka K, Matsushima M: Structure-changeof active site cleft and different saccharide binding modes inhuman lysozyme co- crystallized with hexa-N-acetyl-chito-hexaose at pH 4.0.  J Mol Biol 1994, 244:522-540.18. Maenaka K, Matsushima M, Kawai G, Kidera A, Watanabe K, KurokiR, Kumagai I: Structural and functional effect of Trp-62 → Glyand Asp-101 → Gly substitutions on substrate-bindingmodes of mutant hen egg-white lysozyme.  Biochem J 1998,333:71-76.19. Klein-Seetharaman J, Oikawa M, Grimshaw SB, Wirmer J, DuchardtE, Ueda T, Imoto T, Smith LJ, Dobson CM, Schwalbe H: Long-rangeinteractions within a nonnative protein.  Science 2002,295:1719-1722.20. Muraki M, Harada K, Sugita N, Sato K: Protein-carbohydrateinteraction in human lysozyme probed by combining site-directed mutagenesis and affinity labelling.  Biochemistry 2000,39:292-299.21. Muraki M, Morikawa M, Jigami Y, Tanaka H: Engineering of humanlysozyme as a polyelectolyte by the alteration of molecularsurface charge.  Protein Eng 1988, 2:49-54.22. Turk B, Turk V, Turk D: Structural and functional aspects ofpapain-like cysteine proteinases and their protein inhibitors.J Biol Chem 1997, 378:141-150.23. Abrahamson M: Cystatins.  Methods Enzym 1994, 244:685-700.24. Shibuya K, Kaji H, Ohyama Y, Tate S, Kainosho M, Inagaki R, SamejimaT: Significance of the highly conserved Gly-4 residue inhuman cystatin A.  J Biochem 1995, 118:635-642.25. Ekiel I, Abramhamson M: Folding-related dimerization ofhuman cystatin C.  J Biol Chem 1996, 271:1314-1321.26. Hall A, Håkannsson K, Mason RW, Crubb A, Abrahamson M: Struc-tural basis for the biological specificity of cystatin C.  J BiolChem 1995, 270:5115-5121.27. Canet D, Sunde M, Last AM, Miranker A, Spencer A, Robinson CV,Dobson M: Mechanistic studies of the folding of human lys-ozyme and the origin of amyloidogenic behavior in its dis-ease-related variants.  Biochemistry 1999, 38:6419-6427.28. Žeronik E, Pompe-Novak M, Škarabot M, Ravnikar M, Muševič I, TurkV: Human stefin B readily forms amyloid fibrils in vitro.  Bio-chim Biophys Acta 2002, 1594:1-5.29. Attwood TK: The quest to deduce protein function fromsequence: the role of pattern databases.  Int J Biochem Cell Biol2000, 32:139-155.30. Kumagai I, Maenaka K, Sunada F, Takeda S, Miura K: Effect of sub-site alterations on substrate-binding mode in the active siteof hen egg-white lysozyme.  Eur J Biochem 1993, 212:151-156.31. Karplus K, Barrett C, Cline M, Diekhans M, Grate L, Hughey R: Pre-dicting protein structure using only sequence information.Proteins 1999:121-125.32. Peracchi A: Enzyme catalysis: removing chemically 'essential'residues by site-directed mutagenesis.  Trends Biochem Sci 2001,26:497-503.33. Grant GA: Synthetic peptides for production of antibodiesthat recognize intact proteins.  In Current Protocols in MolecularBiology Volume 2. Edited by: Ausubel EM. New York: WileyInterscience; 2002:11.16.1-11.16.9. 34. Binkowski TA, Adamian L, Liang J: Inferring functional relation-ships of proteins from local sequence and special surfacepatterns.  J Mol Biol 2003, 332:505-526.35. Alvarez-Fernandez M, Barrett AJ, Gerhartz B, Dando PM, Ni J, Abra-hamson M: Inhibition of mammalian legumain by some cystat-ins is due to a novel second reactive site.  J Biol Chem 1999,274:19195-19203.36. Todd AE: From protein structure to function.  In Bioinformatics:Genes, Proteins and Computers Edited by: Oreng C, Jones D, ThorntonPage 16 of 17(page number not for citation purposes)13. Frastrez J, Höltje J-V: Phage lysozyme, Bacterial lysozymes.  InLysozymes: Model Enzymes in Biochemistry and Biology Edited by: JollésP. Berlin: Birkhäuser Verlag; 1996:35-74. J. Oxford: BIOS Scientific Publisher; 2003:151-174. Publish with BioMed Central   and  every scientist can read your work free of charge"BioMed Central will be the most significant development for disseminating the results of biomedical research in our lifetime."Sir Paul Nurse, Cancer Research UKYour research papers will be:available free of charge to the entire biomedical communitypeer reviewed and published immediately upon acceptancecited in PubMed and archived on PubMed Central BMC Biochemistry 2005, 6:9 http://www.biomedcentral.com/1471-2091/6/937. Stubbs MT, Laber B, Bode W, Huber R, Jerala R, Lenarcid B, Turk V:The refined 2.4Å X-ray crystal structure of recombinanthuman stefin B in complex with the cysteine proteasepapain: a novel type of proteinase inhibitor interaction.EMBO J 1990, 9:1939-1947.38. Abrahamson M, Titonja A, Brown MA, Crubb A, Machleidt W, Bar-rett AJ: Identification of the probable inhibitory reactive siteof the cysteine proteinase inhibitors human cystatin C andchicken cystatin.  J Biol Chem 1987, 262:9688-9694.39. Nakai S, Amantea G, Nakai H, Ogawa M, Kanagawa S: Definition ofoutliers using unsupervised principal component similarityanalysis for sensory evaluation of foods.  Int J Food Prop 2002,5:289-306.40. Nakai S, Dou J, Richards JF: New multivariate strategy for panelevaluation using principal component similarity.  Int J Food Prop2000, 3:149-164.41. Bromme D, Kaleta J: Thiol-dependent cathepsins: pathophysio-logical implications and recent advances in inhibitor design.Current Pharm Design 2002, 8:1639-1658.42. Network Protein Sequence Analysis of Pôle Bio-Informatique Lyonnais[http://npsa-pbil.ibcp.fr].43. Krzanowski WJ: Principles of Multivariate Analysis Oxford: Oxford Sci-ence Publications; 1988:26. 44. Nakai S, Ogawa M, Nakamura S, Dou J, Funane K: A computer-aided strategy for structure-function study of food proteinsusing unsupervised data mining.  Int J Food Prop 2003, 6:25-47.45. Nakai S: Computer software used in this study.  2003 [ftp://ftp.agsci.ubc.ca/foodsci/].yours — you keep the copyrightSubmit your manuscript here:http://www.biomedcentral.com/info/publishing_adv.aspBioMedcentralPage 17 of 17(page number not for citation purposes)


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items