Pii: s0378-1119(01)00640-0

Positively selected amino acid sites in the entire coding region of Center for Information Biology and DNA Data Bank of Japan, National Institute of Genetics, 1111 Yata, Mishima-shi, Shizuoka-ken 411-8540, Japan Received 6 December 2000; received in revised form 10 April 2001; accepted 26 July 2001 To predict the amino acid sites important for the clearance of hepatitis C virus (HCV) subtype 1b in vivo, positively selected amino acid sites were detected by analyzing the sequence data collected from the international DNA databank. The rate of nonsynonymous substitutionsper nonsynonymous site was compared with that of synonymous substitutions per synonymous site for each codon site in the entire codingregion. As a result, 13 out of 3010 amino acid sites were found to be positively selected. Among the 13 positively selected amino acid sites,eight were located in the structural proteins and five were in the nonstructural proteins. Moreover, eight were located in B-cell epitopes andtwo were in T-cell epitopes. These observations suggest that both the antibody and the cytotoxic T lymphocyte are involved in the clearanceof HCV subtype 1b in vivo. These positively selected amino acid sites represent candidate vaccination targets for HCV subtype 1b. q 2001Elsevier Science B.V. All rights reserved.
Keywords: Hepatitis C virus; Synonymous substitution; Nonsynonymous substitution; Positive selection; Epitope; Vaccine et al., 1998). In the polyprotein of HCV, a region includingthe N-terminal 27–31 amino acid sites in E2 is known to be Hepatitis C virus (HCV), the sole member of the genus the most variable and is called hypervariable region 1 Hepacivirus in the family Flaviviridae (van Regenmortel et (HVR1; Hijikata et al., 1991; Weiner et al., 1991). HVR1 al., 2000), is an enveloped, non-segmented, single-stranded, has been proposed as a major target of the immune response and positive-sense RNA virus (Choo et al., 1989). The (Weiner et al., 1992; Farci et al., 1996; Zibert et al., 1997a).
genome of HCV is approximately 9.5 kilobases long, encod- More than 50% of humans infected with HCV establish ing a polyprotein of approximately 3000 amino acids (Kato chronic hepatitis, which may progress to cirrhosis and hepa- et al., 1990). The polyprotein is co- and post-translationally tocellular carcinoma (Alter et al., 1992). Interferon alpha cleaved into core protein (C), envelope glycoprotein 1 (E1), and ribavirin are used for treatment of HCV infection.
E2, p7, nonstructural protein 2 (NS2), NS3, NS4A, NS4B, However, they are not highly effective, especially for NS5A, and NS5B in order from its N-terminus by the cellu- HCV subtype 1b (Davis et al., 1998; McHutchison et al., lar signalase and the viral proteinases (Grakoui et al., 1993).
1998). Therefore, development of effective therapies and The genomic sequences from different HCV isolates are vaccines against HCV is an urgent subject worldwide.
highly divergent (Kato et al., 1989). According to the phylo- For developing effective vaccines against HCV, it is genetic analysis, HCV has been classified into six clades, in important to identify epitopes involved in the clearance of which various numbers of subtypes are included (Robertson HCV in vivo, because they may be the candidate vaccinationtargets. Although it may be difficult to identify thoseepitopes experimentally because of the lack of an efficient Abbreviations: C, core protein; DDBJ, DNA databank of Japan; E, envel- in vitro cell culture system and an in vivo animal model ope glycoprotein; HCV, hepatitis C virus; HVR, hypervariable region; NS,nonstructural protein; P, probability; c system for proliferation of HCV (Blight et al., 2000), they tions per codon; cN, number of nonsynonymous substitutions per codon; dS, may be identified as the positively selected amino acid sites.
number of synonymous substitutions per synonymous site; dN, number of This is because the amino acid mutations at those epitopes nonsynonymous substitutions per nonsynonymous site; sS, number of may provide selective advantage to the mutants by allowing synonymous sites per codon; sN, number of nonsynonymous sites per codon them to escape from the immune response (Endo et al., * Corresponding author. Tel.: 181-559-81-6847; fax: 181-559-81-6848.
E-mail address: [email protected] (T. Gojobori).
1996; Fitch et al., 1997; Nielsen and Yang, 1998; Suzuki 0378-1119/01/$ - see front matter q 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 8 - 1 1 1 9 ( 0 1 ) 0 0 6 4 0 - 0 Y. Suzuki, T. Gojobori / Gene 276 (2001) 83–87 and Gojobori, 1999; Yamaguchi-Kabata and Gojobori, CLUSTAL W (Thompson et al., 1994). We then extracted sequences which did not contain any gaps within each Positive selection operating at the amino acid sequence coding region in the pairwise alignment. The subtype 1b level can be detected by comparing the rate of nonsynon- sequences were identified by reconstructing a phylogenetic ymous substitution per nonsynonymous site (rN) with that of tree for each coding region. The numbers of nucleotide synonymous substitution per synonymous site (rS). A higher sequences collected for C, E1, E2, NS2, NS3, NS4, value of rN than rS is an indicator of positive selection, NS5A, and NS5B were 129, 135, 59, 74, 68, 70, 69, and whereas the inverse is an indicator of negative selection 61, respectively. These sequences originated largely from (Hughes and Nei, 1988, 1989). By comparing rN and rS at unrelated patients and only a few were from the same single codon sites, it is possible to detect positive selection patients. In this paper, the amino acid positions are at single amino acid sites (Fitch et al., 1997; Nielsen and Yang, 1998; Suzuki and Gojobori, 1999; Yamaguchi-Kabata and Gojobori, 2000).
In this study, we identified positively selected amino acid sites in the entire coding region of HCV subtype 1b by using A multiple alignment was made for each coding region by the method of Suzuki and Gojobori (1999) to predict the using CLUSTAL W. The positively selected amino acid amino acid sites important for the clearance of HCV subtype sites were identified by using the method of Suzuki and Gojobori (1999). Briefly, a phylogenetic tree was recon-structed by the neighbor-joining method (Saitou and Nei,1987) using the number of synonymous substitutions (Nei and Gojobori, 1986). The ancestral sequence was inferred ateach node in the phylogenetic tree by using the maximum parsimony method (Hartigan, 1973). Then, the averagenumbers of synonymous (sS) and nonsynonymous (sN) In this study, we focused on HCV subtype 1b because of sites and the total numbers of synonymous (cS) and nonsy- its clinical importance and a large number of sequence data nonymous (cN) substitutions throughout the phylogenetic deposited in the international DNA databank. Note that a tree were estimated for each codon site. A probability (P) relatively large number of sequences are required for detect- of obtaining the observed or more biased numbers of synon- ing positively selected amino acid sites by using the method ymous and nonsynonymous substitutions was computed for of Suzuki and Gojobori (1999). Actually, the numbers of each codon site, assuming a binomial distribution. In the sequences for another subtype were too small to detect posi- computation, sS/(sS 1 sN) and sN/(sS 1 sN) were used as the tively selected amino acid sites in the entire coding region.
probabilities of the occurrences of synonymous and nonsy- For detecting positively selected amino acid sites in the nonymous substitutions, respectively. The significance level entire coding region of HCV subtype 1b, the simplest way is was set at 5%. A significantly larger value of cN than cS was to analyze only sequence data which contain the entire considered as an indicator of positive selection, whereas the coding region. However, the number of such sequences inverse was considered as an indicator of negative selection.
was relatively small. Thus, we divided the entire coding The number of synonymous substitutions per synonymous region of HCV subtype 1b into eight regions encoding C, site (dS) and that of nonsynonymous substitutions per nonsy- E1, E2, NS2, NS3, NS4, NS5A, and NS5B and analyzed nonymous site (dN) were estimated by cS/sS and cN/sN, each region separately. p7 and E2 were combined as E2, and NS4A and NS4B were combined as NS4, because p7 (63amino acid sites) and NS4A (54 amino acid sites) were tooshort to analyze positively selected amino acid sites. The numbers of amino acid sites for C, E1, E2, NS2, NS3, NS4,NS5A, and NS5B were 191, 192, 426, 217, 631, 315, 447, The results for identifying positively selected amino acid sites in the entire coding region of HCV subtype 1b are A total of 7262 entries which included the successive summarized in Fig. 1. dS exceeded dN at most of the terms ‘hepatitis C virus’ in their organism names were amino acid sites (2426/3010, 80.60%), and negative selec- collected from the international DNA databank (DDBJ tion was detected at more than half (1560/3010, 51.83%) of release 40). These entries included all clades and subtypes all amino acid sites. The HCV polyprotein contains many B- of HCV. To collect nucleotide sequences for each coding cell and T-cell epitopes. For example, nearly the entire region of HCV subtype 1b, we defined HCV-JS (Accession coding region of E1 and E2, and NS3 have been reported number: D85516; Tanaka et al., 1995) as a reference as B-cell and T-cell epitopes, respectively (Zibert et al., sequence for HCV subtype 1b and made 7261 pairwise 1997a, 1999; Tabatabai et al., 1999). However, the amino alignments, each of which consisted of HCV-JS and one acid sites involved in the clearance of HCV subtype 1b in of the other sequences, by using the computer program vivo may be limited, because most of the sites in the entire Y. Suzuki, T. Gojobori / Gene 276 (2001) 83–87 Fig. 1. Distribution of the value of (1 2 P) in the entire coding region of HCV subtype 1b. The entire coding region is divided into eight coding regions (C, E1,E2, NS2, NS3, NS4, NS5A, and NS5B). The abscissa indicates the amino acid positions and the ordinate indicates the value of (1 2 P) for each amino acid site.
When dN is larger than dS, the value is indicated above the abscissa, whereas in the opposite situation, the value is indicated below the abscissa. Dotted linesindicate the 5% significance level. A filled rectangle in E2 indicates the position of HVR1.
coding region, including E1, E2, and NS3, are negatively Table 1Functions of positively selected amino acid sites in the entire coding region Indeed, dN exceeded dS at only 265 amino acid sites (8.80%), and positive selection was detected at only 13 amino acid sites (0.43%) (Fig. 1). Among the 13 positively selected amino acid sites, eight were located in the structural proteins (C, E1, and E2) and five were in the nonstructural proteins (NS2, NS3, NS4, NS5A, and NS5B). Since the structural proteins occupy only 26.9% (809/3010) of the HCV polyprotein, positively selected amino acid sites seemed to be located more densely in the structural proteins than in the nonstructural proteins (P ¼ 0:009).
Table 1 summarizes the functions of positively selected amino acid sites for HCV subtype 1b. Most of the sites are located in B-cell (Zibert et al., 1997a, 1999; Jolivet- Reynaud et al., 1998; Pereboeva et al., 1998, 2000; Nakano et al., 1999) and T-cell (Tabatabai et al., 1999; Wang and Eckels, 1999) epitopes, suggesting that both antibodies and cytotoxic T lymphocytes (CTLs) are involved in the clear- ance of HCV subtype 1b in vivo. However, a larger numberof positively selected amino acid sites were located in B-cell a The amino acid position is numbered according to HCV-JS.
epitopes than in T-cell epitopes. This is probably because Y. Suzuki, T. Gojobori / Gene 276 (2001) 83–87 the recognition of T-cell epitopes is restricted by the haplo- type of human leukocyte antigen, whereas that of B-cellepitopes is not. The sequence data used in this study were Allain, J.-P., Dong, Y., Vandamme, A.-M., Moulton, V., Salemi, M., 2000.
collected from the international DNA databank, which Evolutionary rate and genetic drift of hepatitis C virus are not correlatedwith the host immune response: studies of infected donor-recipient included HCV subtype 1b sequences largely from unrelated clusters. J. Virol. 74, 2541–2549.
patients. Therefore, the positive selection may be more effi- Alter, M.J., Margolis, H.S., Krawczynski, K., Judson, F.N., Mares, A., ciently detected in B-cell epitopes than in T-cell epitopes in Alexander, W.J., Hu, P.Y., Miller, J.K., Gerber, M.A., Sampliner, this study. This is consistent with the above observation that R.E., Meeks, E.L., Beach, M.J., 1992. The natural history of commu- positively selected amino acid sites are more densely nity-acquired hepatitis C in the United States. N. Engl. J. Med. 327, located in the structural proteins than in the nonstructural Blight, K.J., Kolykhalov, A.A., Rice, C.M., 2000. Efficient initiation of proteins, because B-cell epitopes are mainly located in HCV RNA replication in cell culture. Science 290, 1972–1974.
structural proteins. Since most positively selected amino Bush, R.M., Bender, C.A., Subbarao, K., Cox, N.J., 1999. Predicting the acid sites were located in B-cell and T-cell epitopes, evolution of human influenza A. Science 286, 1921–1925.
amino acid positions 345, 827, 2719, and 2968 may also Choo, Q.-L., Kuo, G., Ralston, R., Weiner, A.J., Overby, L.R., Bradley, be parts of B-cell and T-cell epitopes.
D.W., Houghton, M., 1989. Isolation of a cDNA clone derived from ablood-borne non-A, non-B hepatitis genome. Science 244, 359–362.
The positively selected amino acid sites in B-cell and T- Davis, G.L., Esteban-Mur, R., Rustgi, V., Hoefs, J., Gordon, S.C., Trepo, cell epitopes may be the vaccination targets against HCV C., Shiffman, M.L., Zeuzem, S., Craxi, A., Ling, M.-H., Albrecht, J., subtype 1b, because these sites should be highly immuno- 1998. Interferon alfa-2b alone or in combination with ribavirin for the genic and involved in the clearance of HCV subtype 1b in treatment of relapse of chronic hepatitis C. N. Engl. J. Med. 339, 1493– vivo. However, it should be noted that these sites are often highly variable, so that HCV may escape from the immune Endo, T., Ikeo, K., Gojobori, T., 1996. Large-scale search for genes on which positive selection may operate. Mol. Biol. Evol. 13, 685–690.
response by producing antigenic mutants (Nowak et al., Farci, P., Shimoda, A., Wong, D., Cabezon, T., De Gioannis, D., Strazzera, 1991; Weiner et al., 1992; Farci et al., 2000). This may be A., Shimizu, Y., Shapiro, M., Alter, H.J., Purcell, R.H., 1996. Preven- facilitated by a high evolutionary rate of HCV (Ina et al., tion of hepatitis C virus infection in chimpanzees by hyperimmune 1994; Smith et al., 1997; Allain et al., 2000; Suzuki et al., serum against the hypervariable region 1 of the envelope 2 protein.
2000) and shifting immunodominance of the immune Proc. Natl. Acad. Sci. USA 93, 15394–15399.
Farci, P., Shimoda, A., Coiana, A., Diaz, G., Peddis, G., Melpolder, J.C., response (Nowak et al., 1995). Nevertheless, it has been Strazzera, A., Chien, D.Y., Munoz, S.J., Balestrieri, A., Purcell, R.H., reported that if the immune response is sufficiently strong Alter, H.J., 2000. The outcome of acute hepatitis C predicted by the in an acute infection, HCV is unable to escape from evolution of the viral quasispecies. Science 288, 339–344.
the immune response even if it is directed against highly Fitch, W.M., Bush, R.M., Bender, C.A., Cox, N.J., 1997. Long term trends variable epitopes (Missale et al., 1996; Zibert et al., in the evolution of H(3) HA1 human influenza type A. Proc. Natl. Acad.
1997a,b; Farci et al., 2000). Moreover, the composite Grakoui, A., Wychowski, C., Lin, C., Feinstone, S.M., Rice, C.M., 1993.
vaccines containing different amino acid residues, particu- Expression and identification of hepatitis C virus polyprotein cleavage larly predicted future amino acid residues, at positively products. J. Virol. 67, 1385–1395.
selected amino acid sites may be useful for preventing Hartigan, J.A., 1973. Minimum mutation fits to a given tree. Biometrics 29, proliferation of escape mutants (Bush et al., 1999). Since both antibodies and CTLs seem to be involved in the Hijikata, M., Kato, N., Ootsuyama, Y., Nakagawa, M., Ohkoshi, S., Shimo- tohno, K., 1991. Hypervariable regions in the putative glycoprotein of clearance of HCV subtype 1b, it may be more effective to hepatitis C virus. Biochem. Biophys. Res. Commun. 175, 220–228.
use both B-cell and T-cell epitopes as the vaccination Hughes, A.L., Nei, M., 1988. Pattern of nucleotide substitution at major histocompatibility complex class I loci reveals overdominant selection.
In conclusion, we identified 13 positively selected amino acid sites in the entire coding region of HCV subtype 1b.
Hughes, A.L., Nei, M., 1989. Nucleotide substitution at major histocompat- ibility complex class II loci: evidence for overdominant selection. Proc.
These sites may be candidate vaccination targets against Ina, Y., Mizokami, M., Ohba, K., Gojobori, T., 1994. Reduction of synon- ymous substitutions in the core protein gene of hepatitis C virus. J. Mol.
Evol. 38, 50–56.
Jolivet-Reynaud, C., Dalbon, P., Viola, F., Yvon, S., Paranhos-Baccala, G., Piga, N., Bridon, L., Trabaud, M.A., Battail, N., Sibai, G., Jolivet, M.,1998. HCV core immunodominant region analysis using mouse mono- The authors thank Dr Allison Wyndham at the National clonal antibodies and human sera: characterization of major epitopes Institute of Genetics, Japan, for providing valuable useful for antigen detection. J. Med. Virol. 56, 300–309.
comments on an earlier version of the manuscript. We are Kato, N., Ohkoshi, S., Shimotohno, K., 1989. Japanese isolates of the non- grateful to two anonymous reviewers for suggestions to A, non-B hepatitis viral genome show sequence variations from the improve the manuscript. This work was supported, in part, original isolate in the U.S.A. Proc. Jpn. Acad. 65, 219–223.
Kato, N., Hijikata, M., Ootsuyama, Y., Nakagawa, M., Ohkoshi, S., Sugi- by grants from the Ministry of Education, Culture, Sports, mura, T., Shimotohno, K., 1990. Molecular cloning of the human hepa- Science, and Technology, Japan. Y.S. is supported by the titis C virus genome from Japanese patients with non-A, non-B JSPS Research Fellowships for Young Scientists.
hepatitis. Proc. Natl. Acad. Sci. USA 87, 9524–9528.
Y. Suzuki, T. Gojobori / Gene 276 (2001) 83–87 McHutchison, J.G., Gordon, S.C., Schiff, E.R., Shiffman, M.L., Lee, W.M., Tabatabai, N.M., Bian, T.-H., Rice, C.M., Yoshizawa, K., Gill, J., Eckels, Rustgi, V.K., Goodman, Z.D., Ling, M.-H., Cort, S., Albrecht, J.K., D.D., 1999. Functionally distinct T-cell epitopes within the hepatitis C 1998. Interferon alfa-2b alone or in combination with ribavirin as initial virus non-structural 3 protein. Hum. Immunol. 60, 105–115.
treatment for chronic hepatitis C. N. Engl. J. Med. 339, 1485–1492.
Tanaka, T., Kato, N., Cho, M.-J., Shimotohno, K., 1995. A novel sequence Missale, G., Bertoni, R., Lamonaca, V., Valli, A., Massari, M., Mori, C., found at the 30 terminus of hepatitis C virus genome. Biochem.
Rumi, M.G., Houghton, M., Fiaccadori, F., Ferrari, C., 1996. Different Biophys. Res. Commun. 215, 744–749.
clinical behaviors of acute hepatitis C virus infection are associated Thompson, J.D., Higgins, D.G., Gibson, T.J., 1994. CLUSTAL W: improv- with different vigor of the anti-viral cell-mediated immune response.
ing the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties and weight matrix Nakano, I., Fukuda, Y., Katano, Y., Hayakawa, T., 1999. Conformational choice. Nucleic Acids Res. 22, 4673–4680.
epitopes detected by cross-reactive antibodies to envelope 2 glycopro- van Regenmortel, M.H.V., Fauquet, C.M., Bishop, D.H.L., Carstens, E.B., tein of the hepatitis C virus. J. Infect. Dis. 180, 1328–1333.
Estes, M.K., Lemon, S.M., Maniloff, J., Mayo, M.A., McGeoch, D.J., Nei, M., Gojobori, T., 1986. Simple methods for estimating the numbers of Pringle, C.R., Wickner, R.B., 2000. Virus Taxonomy. Academic Press, synonymous and nonsynonymous substitutions. Mol. Biol. Evol. 3, Wang, H., Eckels, D.D., 1999. Mutations in immunodominant T cell Nielsen, R., Yang, Z., 1998. Likelihood models for detecting positively epitopes derived from the nonstructural 3 protein of hepatitis C virus selected amino acid sites and applications to the HIV-1 envelope have the potential for generating escape variants that may have impor- tant consequences for T cell recognition. J. Immunol. 162, 4177–4183.
Nowak, M.A., Anderson, R.M., McLean, A.R., Wolfs, T.F., Goudsmit, J., Weiner, A.J., Brauer, M.J., Rosenblatt, J., Richman, K.H., Tung, J., Craw- May, R.M., 1991. Antigenic diversity thresholds and the development ford, K., Bonino, F., Saracco, G., Choo, Q.-L., Houghton, M., Han, J.H., 1991. Variable and hypervariable domains are found in the regions of Nowak, M.A., May, R.M., Phillips, R.E., Rowland-Jones, S., Lalloo, D.G., HCV corresponding to the Flavivirus envelope and NS1 proteins and McAdam, S., Klenerman, P., Koppe, B., Sigmund, K., Bangham, the Pestivirus envelope glycoproteins. Virology 180, 842–848.
C.R.M., McMichael, A.J., 1995. Antigenic oscillations and shifting Weiner, A.J., Geysen, H.M., Christopherson, C., Hall, J.E., Mason, T.J., immunodominance in HIV-1 infections. Nature 375, 606–611.
Saracco, G., Bonino, F., Crawford, K., Marion, C.D., Crawford, K.A., Pereboeva, L.A., Pereboev, A.V., Morris, G.E., 1998. Identification of anti- Brunetto, M., Barr, P.J., Miyamura, T., McHutchinson, J., Houghton, genic sites on three hepatitis C virus proteins using phage-displayed M., 1992. Evidence for immune selection of hepatitis C virus (HCV) peptide libraries. J. Med. Virol. 56, 105–111.
putative envelope glycoprotein variants: potential role in chronic HCV Pereboeva, L.A., Pereboev, A.V., Wang, L.F., Morris, G.E., 2000. Hepatitis infections. Proc. Natl. Acad. Sci. USA 89, 3468–3472.
C epitopes from phage-displayed cDNA libraries and improved diag- Yamaguchi-Kabata, Y., Gojobori, T., 2000. Reevaluation of amino acid nosis with a chimeric antigen. J. Med. Virol. 60, 144–151.
variability of the human immunodeficiency virus type 1 gp120 envelope Robertson, B., Myers, G., Howard, C., Brettin, T., Bukh, J., Gaschen, B., glycoprotein and prediction of new discontinuous epitopes. J. Virol. 74, Gojobori, T., Maertens, G., Mizokami, M., Nainan, O., Netesov, S., Nishioka, K., Shin-i, T., Simmonds, P., Smith, D., Stuyver, L., Weiner, Zibert, A., Kraas, W., Meisel, H., Jung, G., Roggendorf, M., 1997a. Epitope A., 1998. Classification, nomenclature, and database development forhepatitis C virus (HCV) and related viruses: proposals for standardiza- mapping of antibodies directed against hypervariable region 1 in acute tion. Arch. Virol. 143, 2493–2503.
self-limiting and chronic infections due to hepatitis C virus. J. Virol. 71, Saitou, N., Nei, M., 1987. The neighbor-joining method: a new method for reconstructing phylogenetic trees. Mol. Biol. Evol. 4, 406–425.
Zibert, A., Meisel, H., Kraas, W., Schulz, A., Jung, G., Roggendorf, M., Smith, D.B., Pathirana, S., Davidson, F., Lawlor, E., Power, J., Yap, P.L., 1997b. Early antibody response against hypervariable region 1 is asso- Simmonds, P., 1997. The origin of hepatitis C virus genotypes. J. Gen.
ciated with acute self-limiting infections of hepatitis C virus. Hepatol- Suzuki, Y., Gojobori, T., 1999. A method for detecting positive selection at Zibert, A., Kraas, W., Ross, R.S., Meisel, H., Lechner, S., Jung, G., single amino acid sites. Mol. Biol. Evol. 16, 1315–1328.
Roggendorf, M., 1999. Immunodominant B-cell domains of hepatitis Suzuki, Y., Yamaguchi-Kabata, Y., Gojobori, T., 2000. Nucleotide substi- C virus envelope proteins E1 and E2 identified during early and late tution rates of HIV-1. AIDS Rev. 2, 39–47.
time points of infection. J. Hepatol. 30, 177–184.

Source: http://www.nsc.nagoya-cu.ac.jp/~yossuzuk/MyPapers.dir/83.pdf