Physicochemical properties of GPCR amino acid sequences for understanding GPCR-G-protein coupling

Transcription

1 Chem-Bio Informatics Journal, Vol. 8, No. 2, (2008) Physicochemical roerties of GPCR amino acid sequences for understanding GPCR-G-rotein couling Ganga D. Ghimire 1, 2*, Hideki Tanizawa 2, Masashi Sonoyama 2, and Shigeki Mitaku 2 1 Venture Business Laboratory, Nagoya University Furocho, Chikusa-ku, Nagoya , JAPAN 2 Nagoya University, School of Engineering, Deartment of Alied Physics, Furocho, Chikusa-ku, Nagoya , JAPAN * ghimire@b.nua.nagoya-u.ac. (Received January 31, 2008; acceted August 5, 2008; ublished online August 26, 2008) Abstract G-rotein couled recetors (GPCRs) bind with G-roteins uon activation by ligands. Understanding the mechanisms of secific binding between GPCRs and G-roteins is one of the most imortant issues in bioinformatics research. In this study, the hysical roerties of various regions were analyzed in order to classify GPCRs by G-rotein family and to better understand binding secificity. We focused on cytolasmic loos (IL1, IL2 and N/C-terminus of IL3), extracellular loos (NTL, EL1 and N/C-terminus of EL2) and cytolasmic termini of transmembrane helices, excet for helices that connect to C-terminus loos. The distribution of hydrohobicity, charge density, lysine and arginine densities, and loo length enabled discrimination of GPCRs with more than 90% accuracy. Key Words: GPCR, G-rotein, signal transduction, roteomics, bioinformatics Area of Interest: Bioinformatics and Bio comuting Coyright 2008 Chem-Bio Informatics Society htt:// 49

2 1. Introduction G-rotein couled recetors (GPCRs) form one of the largest rotein families and consist of seven transmembrane (TM) helices. GPCRs lay a key role in cellular signal transduction and reresent one of the most imortant roteins in the harmaceutical industry [1], as they are targets of more than half (52%) of currently known drugs [2]. Uon activation by an extracellular ligand, GPCRs undergo couling with G-roteins. G-roteins are classified into three maor families, G s, G i/o and G q/11, based on their subunit [3]; G s and G i/o stimulate and inhibit, resectively, adenylate cyclase [4] [5], while G q/11 stimulates hosholiase C. Each family has a secific influence on the cell. Therefore, the mechanisms of secific binding between GPCRs and G-roteins are imortant in understanding signal transduction. The urose of this work is to study (1) which hysical roerties of GPCRs are resonsible for secific binding to G-roteins, and (2) the develoment of an accurate system for classifying GPCRs based on hysicochemical arameters. In this article, we refer to the recetors couled with G s, G i/o and G q/11 as R s, R i/o and R q/11, resectively. The classification of GPCRs by homology analysis is difficult because the function-structure relationshi is unclear. For examle, some homologous GPCR airs with the same ligands bind to different tyes of G-rotein; some airs bind to the same tye of G-rotein bind to different ligands; and some GPCR airs bind to both the same ligand and the same G-rotein, desite showing sequence similarities of less than 25% [6]. Therefore, discriminating each GPCR family based on hysicochemical roerties is necessary in order to clarify the reasons for differences in GPCR families. With regard to discrimination method, the use of suort vector machines is one of the most oular statistical methods for this urose due to higher accuracy[7][8], and has attracted much attention after its state-of-art erformances and the GRIFFIN [9], which imlemented this method quite successfully. On the other hand, rincial comonent analysis (PCA) is a simler method and has been widely used in statistical data analysis. Furthermore, the results of PCA can be easily exlained based on the contributions of various arameters to the rincial comonent of discrimination. In this study, we alied PCA to the roblem of the classifying GPCR families, and attemted to exlain the mechanisms of secific couling between GPCRs and G roteins The same aroach was reviously alied to the roblem of discriminating G-roteins, and we found that the hydroathy and charge density of the amino- and carboxyl-terminal segments are essential for the binding secificity of G-roteins [10]. Because the comlementary nature of the two roteins results in comlex formation, we considered that the classification of GPCRs is ossible by analyzing the same roerties at the binding regions. In this work, we alied the hysical fingerrint method to the roblem of classifying GPCRs by G-rotein family, based on hydrohobicity, charge density, lysine and arginine densities, and loo length. We obtained good redictors that were able to classify GPCRs into three families; R s, R i/o and R q/11. The results showed that the hysical fingerrint method is alicable to the roblem of classifying G-rotein-couled recetors. 50

3 Chem-Bio Informatics Journal, Vol. 8, No. 2, (2008) 2. Method 2.1 Amino acid sequence data for G-rotein-couled recetors Table 1. Dataset for analysis Recetor couling to G rotein families No. of data Recetor couling to G s family (R s ) 68 Recetor couling to G i/o family (R i/o ) 116 Recetor couling to G q/11 family (R q/11 ) 94 Total no. of data 278 PRED-COUPLE dataset [11] of amino acid sequences of GPCRs were used for the training and testing. Table 1 shows the number of recetors that coule with different families of G-rotein. 2.2 Physical fingerrint method for classification of rotein In the hysical fingerrint method, the average values for amino acid indices, which have well-defined hysicochemical meanings, were calculated for intracellular and extracellular loos, and TM domains, and the average values were used for discrimination analysis. To calculate these arameters, the boundaries of the transmembrane helix and loo regions of GPCR sequences were determined by the SOSUI membrane rotein rediction system.[12] To select the regions for analysis, we first observed the structure of rhodosin (PDB ID: 1HZX) as shown in Figures 1A and 1B. The seven TM helices are connected by alternating extracellular and cytolasmic loos. The loos that are thought to directly contact the G-rotein are indicated in green, blue and cyan in Figures 1A and 1B. The NTL, EL1 and N/C of EL2 in the extracellular domain and IL1, IL2 and N/C of IL3 in the cytolasmic domain, as well as the cytolasmic termini of helices, excet for the helix connected to the C-terminus loo, were used for classification of GPCRs by G-rotein tye. The focus regions for analysis are marked in red, as shown in Figure 1C. The hysicochemical roerties used for the analysis were the hydrohobicity index, charge density, lysine and arginine densities, and loo length. Loos, as well as hysical roerties, were selected after extensive discrimination analyses and we finally omitted the C-terminal segment (CTL) from analysis. The CTLs of GPCRs showed marked variety in length and hysical roerties, even within the same GPCR family. Therefore, we hyothesized that these segments are not suitable for classification of GPCR families. The arameters and equations for analysis were as described reviously [10][13]. The average hydrohobicity index and average charge density were calculated using the following equations. i 3 X () i X ( k) (1) 7 k i 3 where, X 1 X 2, X 3 and X 4 indicate the hydrohobicity index, electric charge, and lysine and arginine densities, resectively. The Kyte and Doolittle index was used for the hydrohobicity index of amino acids [14]. Average values X () i were further averaged for the selected regions. 51

4 Figure 1. Structure of rhodosin (PDB ID: 1HZX) and schematic diagram (A) Viewed along lane arallel to the membrane. (B) View of rhodosin from cytolasm. (C) Schematic diagram of GPCR. The region used for the analysis is shown in red. Parts thought to be most imortant for secificity of GPCR/G-rotein couling are the N-terminus (NTL), first loo (EL1) and N/C terminus of the second loo (EL2) of the extracellular domain, and the first loo (IL1), second loo (IL2), N/C terminus of third loo (IL3) and cytolasmic termini of helices, excet for the helix connected to the C-terminus loo. 52

5 Chem-Bio Informatics Journal, Vol. 8, No. 2, (2008) X X () i / l( ) (2) i loo( ) The double average value loo of length of l(). Here, we defined the third arameter X reresents the average of the -th arameter for the -th X 5 using the following equation. X 5 l( ) (3) When two families were discriminated, the sequences in one family are assumed to be the ositive data and the sequences in the other were assumed to be negative data. We calculated the weighted deviation DX, in which the difference is weighted by the average difference between the ositive and negative data as follows: DX X N P N (4) where P and N are the values of X for all ositive and negative data, resectively. Finally, the discrimination score between the ositive and negative data were calculated by discrimination analysis, leading to the following discrimination function, Score a1dp1 a2dp2 a3dp3 a4dp4 a5dp5 (5) 3. Result For analysis of GPCR families by the hysical fingerrint aroach, we first calculated the moving average of hydrohobicity < H >, charge density < C > lysine density < K >, and arginine density < R > using equation (1) and (2). The double averages of each selected region were then calculated for the R s, R i/o and R q/11 classes. Comarison of these arameters and loo length between each class is shown in Figure 2. The difference in the rofiles among families is subtle, but certainly observable. For examle, several roerties in the first and second internal loos, as well as both ends of third internal loos, showed significant differences. The roerties of external loos were generally similar among the three families, and are consistent with the fact that binding with the G-rotein occurs on the internal side. However, the contribution of external loos to the discrimination was not negligible, and the roerties of these regions were also used. Analysis of differences in the rofiles of the hysicochemical arameters enabled calculation of scores for discrimination between families: R s vs. R i/o ; R s vs. R q/11 ; and R i/o vs. R q/11. The three-ste rocess of discrimination of the three GPCR families (R s, R i/o and R q/11 ) is shown in Figure 3. First, we discriminated between R s and R i/o. As shown in Figure 3A, these GPCRs could be discriminated very clearly. Proteins belonging to R q/11 were not discriminated well in this ste; therefore, the data for R q/11 were divided into two arts; low- and high-score data from the threshold are reresented by R q/11a and R q/11b, resectively. These GPCRs were discriminated in the next two stes: R s vs. R q/11 (Figure 3B) and R i/o vs. R q/11 (Figure 3C). All data were discriminated with high accuracy by a combination of the three discrimination scores. Table 2 shows the results of discrimination of GPCRs using the training dataset and the results of the cross-validation test. The accuracy for R s, R i/o and R q/11 in analysis of the training dataset was 100%, 99.1% and 100%, resectively. The average accuracy for R s, R i/o, and R q/11 in 1000 rounds of five-fold cross-validation was 92.1%, 93.8% and 85.2%, resectively. 53

6 Figure 2. Profiles of hysicochemical roerties; Average values for hydroathy, charge density, lysine density, arginine density and loo length were comared. 54

7 Chem-Bio Informatics Journal, Vol. 8, No. 2, (2008) Figure 3. Three sets of histograms in which three tyes of GPCR are discriminated by rimary comonent analysis. The first ste (A) reresents the discrimination between R s and R i/o. All data for R s are discriminated against R i/o. However, R q/11 scattered in both regions. Therefore, in the second ste (B) R s and R q/11 a were discriminated. The final ste (C) reresents the discrimination between R i/o and R q/11 b. R i/o could be discriminated against R q/11 b. Combining three stes, all three families of GPCR could be classified. Table 2. Results of classification of three GPCR families using training data. Classification of GPCRs by G-rotein tye based on hysical fingerrint analysis. The accuracy of training set was ercentage that correctly redicted. The accuracy of cross validation was average accuracy among 1000 round of 5-fold cross validations. Predicted class G s G i/o G q/11 Accuracy of training set Accuracy of cross validation Actual class G s % 92.1% G i/o % 93.8% G q/ % 85.2% total 99.7% 90.3% 55

8 4. Discussion Uon activation by extracellular ligands, GPCR couling with heterotrimeric G-rotein leads to a hysiological resonse. Therefore, classifying GPCRs by G-rotein tye using amino acid sequence data is one of the most imortant roblems in bioinformatics. In this work, we analyzed the IL1, IL2 and N/C terminus of IL3 of the cytolasmic domain, which directly contacts G-roteins. Several mutational analyses have suggested the imortance of these regions [3] for determining couling secificity. We also used analysis of the cytolasmic terminus of TM near the IL1, IL2 and IL3 domains, as well as the NTL, EL1 and N/C terminus of the extracellular domain for accurate discrimination of the three tyes of GPCR. However, these regions do not directly contact the G-rotein, and there is no exerimental evidence regarding their influence on the binding secificity of GPCRs and G-roteins; nonetheless, these regions may have some indirect role. Loo length was also used for classification of GPCRs. We found that there is a difference in loo length among GPCR families. It is difficult to classify GPCRs by G-rotein tye, as the binding secificity is not closely linked to sequence similarity. Therefore, we alied the hysical fingerrint method, which is alicable to classifying G-rotein families. IL1, IL2, CIL3, NIL3, CHIL1, CHIL2, CHIL3 and NHIL3 in the cytolasmic domain, and NTL, EL1, CEL2 and NEL2 in the extracellular domain were analyzed. Loo length, IL1, IL1, IL3, CTL NTL, EL1, EL2 and EL3 were used in the analysis. The differences in hysical roerties of the binding regions among the three GPCR classes are subtle but significant, thus allowing accurate classification of GPCRs by G-rotein binding secificity. The accuracy of the cross-validation test was better than 90%. We reviously classified G-roteins, the counterarts of GPCRs, by hydroathy and charge density of amino acid sequences alone [10]. In the current work, we used these arameters, as well as lysine density and arginine density, and were able to classify GPCRs leading to the conclusion that the hysical roerties are comlimentary at the binding sites of GPCRs and G-roteins. For examle, there is more charge in the G s family than in the other G-rotein families. In the current rofiles, the ositive charge in the R s family was less than in the other families of GPCR (R i/o and R q/11 ). We also found comlementary rofiles for hydroathy when comared with revious rofiles [10]. In the future, we lan to classify the sub-families of GPCRs by G-rotein tye. 56

9 Chem-Bio Informatics Journal, Vol. 8, No. 2, (2008) References [1] J. Drews, "Drug discovery: a historical ersective.," Science, vol. 287, , [2] J. Drews, "Genomic sciences and the medicine of tomorrow.," Nat Biotechnol, vol. 14, , [3] J. Wess, "Molecular basis of recetor/g-rotein-couling selectivity.," Pharmacol Ther, vol. 80, , [4] D. R. Benamin, D. W. Markby, H. R. Bourne, and I. D. Kuntz, "Solution structure of the GTPase activating domain of alha s.," J Mol Biol, vol. 254, , [5] C. A. Johnston and V. J. Watts, "Sensitization of adenylate cyclase: a general mechanism of neuroadatation to ersistent activation of Galha(i/o)-couled recetors?," Life Sci, vol. 73, , [6] A. Gaulton and T. K. Attwood, "Bioinformatics aroaches for the classification of G-rotein-couled recetors.," Curr Oin Pharmacol, vol. 3, , [7] T. Muramatsu and M. Suwa, "Statistical analysis and rediction of functional residues effective for GPCR-G-rotein couling selectivity.," Protein Eng Des Sel, vol. 19, , [8] K. R. Sreekumar, Y. Huang, M. H. Pausch, and K. Gulukota, "Predicting GPCR-G-rotein couling using hidden Markov models.," Bioinformatics, vol. 20, , [9] Y. Yabuki, T. Muramatsu, T. Hirokawa, H. Mukai, and M. Suwa, "GRIFFIN: a system for redicting GPCR-G-rotein couling selectivity using a suort vector machine and a hidden Markov model.," Nucleic Acids Res, vol. 33,. W148--W153, [10] G. D. Ghimire, et al., "Physicochemical roerties of amino acid sequences of G-roteins for understanding GPCR-G-rotein couling," Chem-Bio Informatics Journal, vol. 6,. 1-16, [11] N. G. Sgourakis, P. G. Bagos, P. K. Paasaikas, and S. J. Hamodrakas, "A method for the rediction of GPCRs couling secificity to G-roteins using refined rofile Hidden Markov Models.," BMC Bioinformatics, vol. 6,. 104, [12] T. Hirokawa, S. Boon-Chieng, and S. Mitaku, "SOSUI: classification and secondary structure rediction system for membrane roteins.," Bioinformatics, vol. 14, , [13] K. Imai and S. Mitaku, "Mechanisms of secondary structure breakers in soluble roteins," BIOPHYSICS, vol. 1, , [14] J. Kyte and R. F. Doolittle, "A simle method for dislaying the hydroathic character of a rotein.," J Mol Biol, vol. 157, ,