FULL PAPER Molecular Modeling of the Three-Dimensional Structure of Human Sphingomyelin Synthase Zhang,Ya°(张亚)Lin,F(林赋)Deng, Xiaodong°(邓晓东) 王任小)Ye, Deyong*(叶德泳) School of pharmacy, Fudan University, Shanghai 201203, China b State Key Lab of Bioorganic Chemistry, Shanghai Institute of Organic Chemistry, Chinese Academy of Sciences, Shanghai 200032, Ch Sphingomyelin synthase (SMS) produces sphingomyelin and diacylglycerol from ceramide and phosphatidyl- soline. It plays an important role in cell survival and apoptosis, inflammation, and lipid homeostasis, and therefore has been noticed in recent years as a novel potential drug target. In this study, we combined homology modeling, molecular docking, molecular dynamics simulation, and normal mode analysis to derive a three-dimensional struc- ture of human sphingomyelin synthase(hSMs )in complex with sphingomyelin. Our model provides a reasonable explanation on the catalytic mechanism of hSMSl. It can also explain the high selectivity of hSMsI towards phos- hocholine and sphingomyelin as well as some other known experimental results about hSMSl. Moreover, we also derived a complex model of D609, the only known small-molecule inhibitor of hSMSI so far. Our hSMSI model may serve as a reasonable structural basis for the discovery of more effective small-molecule inhibitors of hSMSI Keywords sphingomyelin synthase, molecular modeling, molecular dynamics Introduction round of catalysis Due to its pharmaceutical implications, SMS Sphingomyelin synthase(SMS) is the enzyme that been noticed as a potential drug target in recent ye functions at the last step in the synthesis of shingo- myelin. It recognizes ceramide and phosphatidylcholine Huitema et al. reported the sequence of SMs using (PC)as substrates to produce sphingomyelin(SM)and functional cloning strategy in yeast.But the diacylglycerol (DAG)(Figure 1). Its activity influences three-dimensional structure of SMS remains unresolved the levels of SM, PC, ceramide, and DAG directly in so far. Without such structural information, it remains as living body, and is closely related with cell survival and a challenge to understand the catalytic mechanism of apoptosis, inflammation, and atherosclerosis SMS and discover potent inhibitors of SMS accordingly SMS has two known subtypes, SMSI and SMS2, In fact, very few small-molecule compounds that can which are classified by their cellular localizations. regulate the biological function of SMS have been re- SMSI is found merely in the trans-Golgi apparatus, and ported in literature. To the best of our knowledge, the SMS2 is primarily found in the plasma membranes. only known SMS inhibitor so far is D609(Figure 2) As most lipid phosphate phosphatase family, SMs which was reported to have a weak inhibitory activity catalyzes the choline phosphotransferase reaction possi-(ICso=500 umol-L-') against SMS in vitro, but no ac- bly through a similar mechanism. First, a double-chain tivity in vivo due to its unstable chemical structure.13-15 choline phospholipid(PC or SM)enters and binds to a In this study, we combined homology modeling, mo- single site of the enzyme. Then, a nucleophilic attack on lecular docking, and molecular dynamics simulation to the lipid-phosphate ester bond is executed by His328 in derive a three-dimensional structural model of human the assistance by Asp332. After the formation of a cho- line phosphohistidine intermediate and the release of sphingomyelin synthase 1(hSMS1). Our model can DAG or ceramide, His285 acts as a nucleophile to at reasonably explain some known experimental results tack on the carbon attached to the primary hydroxy regarding hSMSl. It can be applied to future structure- group on ceramide or DAG. Finally, the product (SM or based discovery of novel small-molecule inhibitors of PC) is released from the active site to allow the nex hSMS I dyye@shmu.edu.cn,wangrx/@mailsiocaccn;Tel 021-51980117,0086-021-54925128,Fax:0086-021-51980125 d January 5, 2011; revised February 18, 2011; accepte China(Nos. 30973641, 20902013), a special research fund for the Doctoral Program of Education from the Chinese Ministry of Education 0090071ll and an open grant from the State Key Laboratory of Bio-organic and Natural Products Chemistry, Chinese Academy of Sciences WILEY Chin. . Chem. 2011, 29, 1567--1575 C2011 SIOC, CAS, Shanghai, WILEY-VCH Verlag gmbH Co KGaA, Weinheim ONLINE LIBRARY
FULL PAPER * E-mail: dyye@shmu.edu.cn.; wangrx@mail.sioc.ac.cn.; Tel.: 0086-021-51980117, 0086-021-54925128; Fax: 0086-021-51980125 Received January 5, 2011; revised February 18, 2011; accepted Apil 28, 2011. Project supported by the National Natural Science Foundation of China (Nos.30973641, 20902013), a special research fund for the Doctoral Program of Higher Education from the Chinese Ministry of Education (No. 20090071110054), and an open grant from the State Key Laboratory of Bio-organic and Natural Products Chemistry, Chinese Academy of Sciences. Chin. J. Chem. 2011, 29, 1567—1575 © 2011 SIOC, CAS, Shanghai, & WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 1567 Molecular Modeling of the Three-Dimensional Structure of Human Sphingomyelin Synthase Zhang, Yaa (张亚) Lin, Fub (林赋) Deng, Xiaodonga (邓晓东) Wang, Renxiao*,b (王任小) Ye, Deyong*,a (叶德泳) a School of Pharmacy, Fudan University, Shanghai 201203, China b State Key Lab of Bioorganic Chemistry, Shanghai Institute of Organic Chemistry, Chinese Academy of Sciences, Shanghai 200032, China Sphingomyelin synthase (SMS) produces sphingomyelin and diacylglycerol from ceramide and phosphatidylcholine. It plays an important role in cell survival and apoptosis, inflammation, and lipid homeostasis, and therefore has been noticed in recent years as a novel potential drug target. In this study, we combined homology modeling, molecular docking, molecular dynamics simulation, and normal mode analysis to derive a three-dimensional structure of human sphingomyelin synthase (hSMS1) in complex with sphingomyelin. Our model provides a reasonable explanation on the catalytic mechanism of hSMS1. It can also explain the high selectivity of hSMS1 towards phosphocholine and sphingomyelin as well as some other known experimental results about hSMS1. Moreover, we also derived a complex model of D609, the only known small-molecule inhibitor of hSMS1 so far. Our hSMS1 model may serve as a reasonable structural basis for the discovery of more effective small-molecule inhibitors of hSMS1. Keywords sphingomyelin synthase, molecular modeling, molecular dynamics Introduction Sphingomyelin synthase (SMS) is the enzyme that functions at the last step in the synthesis of sphingomyelin. It recognizes ceramide and phosphatidylcholine (PC) as substrates to produce sphingomyelin (SM) and diacylglycerol (DAG) (Figure 1).1 Its activity influences the levels of SM, PC, ceramide, and DAG directly in living body, and is closely related with cell survival and apoptosis, inflammation, and atherosclerosis.2-9 SMS has two known subtypes, SMS1 and SMS2, which are classified by their cellular localizations. SMS1 is found merely in the trans-Golgi apparatus, and SMS2 is primarily found in the plasma membranes.10,11 As most lipid phosphate phosphatase family, SMS catalyzes the choline phosphotransferase reaction possibly through a similar mechanism. First, a double-chain choline phospholipid (PC or SM) enters and binds to a single site of the enzyme. Then, a nucleophilic attack on the lipid-phosphate ester bond is executed by His328 in the assistance by Asp332. After the formation of a choline phosphohistidine intermediate and the release of DAG or ceramide, His285 acts as a nucleophile to attack on the carbon attached to the primary hydroxyl group on ceramide or DAG. Finally, the product (SM or PC) is released from the active site to allow the next round of catalysis.12 Due to its pharmaceutical implications, SMS has been noticed as a potential drug target in recent years. Huitema et al. reported the sequence of SMS using functional cloning strategy in yeast.10 But the three-dimensional structure of SMS remains unresolved so far. Without such structural information, it remains as a challenge to understand the catalytic mechanism of SMS and discover potent inhibitors of SMS accordingly. In fact, very few small-molecule compounds that can regulate the biological function of SMS have been reported in literature. To the best of our knowledge, the only known SMS inhibitor so far is D609 (Figure 2), which was reported to have a weak inhibitory activity (IC50＝500 µmol•L－1 ) against SMS in vitro, but no activity in vivo due to its unstable chemical structure.13-15 In this study, we combined homology modeling, molecular docking, and molecular dynamics simulation to derive a three-dimensional structural model of human sphingomyelin synthase 1 (hSMS1). Our model can reasonably explain some known experimental results regarding hSMS1. It can be applied to future structurebased discovery of novel small-molecule inhibitors of hSMS1
FULL PAPER Ceramide HaNON Figure 1(A)hSMSI-catalyzed synthesis of sphingomyelin( SM) from phosphatidylcholine(PC).(B) Some other phosphatides related to ceramide, including phosphatidylethanolamine(PE), phosphatidic acid(Pa), phosphatidylserine(PS)and phosphatidylglycerol(PG) The transmembrane domain of hSMSI is composed of six transmembrane helices(TMs). We employed 15 SK different computational methods to predict the locations of these TMs on the hSMSI sequence. Most of them produced consistent predictions (Table 1). In order to select an appropriate template for modeling the TMs of Figure 2 Chemical structures of D609(ICs0=500 umolL-I hSMSl. we retrieved a total of 61 entries with six TMs from the membrane structural proteins database against SMS) and the corresponding dimer d609-dixanthogen (no inhibition activity against SMs). PdbTm(htTp: //pdbtm. enzim. hu). After careful evaluations, which will be explained in the later Results Computational methods and Discussion section, we selected the crystal structure of Escherichia coli GlpG(PDB entry: 2IC8)as the tem Homology modeling of the hSMSl/lipid complex plate(Table 2)35 The Modeler function(as imple- structure mented in the Discovery Studio software suite), was The amino acid sequence of hSMSI used in our employed to generate a total of 30 structural models study, which has 413 residues in total, was retrieved based on this template from PubMed (access ID=NP 671512). The hSMSI As for the extracellular or intracellular loop of has an N-terminal domain, a transmembrane domain, hSMSl, loop 3(residues 235--274)is relatively long and a C-terminal domain. 6 Jiang et al. demonstrated (Figure 3). Therefore, it should be modeled based on an recently that truncation of N-terminal and C-terminal of appropriate template. PDB entry IBWO(residues 144 core structure of hSMSI(M130-Q353)was modeled in 463%quence identity =34. 1%; sequence similarity hSMSI would not abort its activity. Thus, only the 181)(se as selected as the template for this purpose our study Locations of the extracellular or intracellular This structure was selected throughout the entire PDB loop and transmembrane domain on the sequence of database according to the sequence similarity computed hSMSI were predicted by using the PSI-PRED Server by the FASTa algorithm. , No qualified template wa he predicted extracellular or intracellular loops and found for other loops of hSMSI though. These loops transmembrane segments are given in Figure 3. This were constructed from scratch by using the modeler prediction is basically consistent with the results re module in the Discovery Studio software suite. A total ported by Huitema et al. in a previous study of 30 structural models also generated for each 1568www.cjc.wiley-vch.deO2011SioC,Cas,Shanghai&WileY-VchVerlagGmbh&Co.Kgaa,WeinheimChin.j.chem.2011,29,1567-1575
FULL PAPER Zhang et al. 1568 www.cjc.wiley-vch.de © 2011 SIOC, CAS, Shanghai, & WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chin. J. Chem. 2011, 29, 1567—1575 HO O O H O O HO H HN O HO PC SM Ceramide O O P O O H O O O O N O P O H HN O HO O O N Diacylglycerol + + SMS + - + - A O O P O O H O O O H3N O O P HO O H O O O O O P O O H O O O O H3N O O O O P O O H O O O OH HO H PE PA PS PG + + - O - - - - O O B Figure 1 (A) hSMS1-catalyzed synthesis of sphingomyelin (SM) from phosphatidylcholine (PC). (B) Some other phosphatides related to ceramide, including phosphatidylethanolamine (PE), phosphatidic acid (PA), phosphatidylserine (PS) and phosphatidylglycerol (PG). O S S K O S SS O S D609 D609 dixanthogen + - Figure 2 Chemical structures of D609 (IC50＝500 µmol•L－1 against SMS) and the corresponding dimer D609-dixanthogen (no inhibition activity against SMS). Computational methods Homology modeling of the hSMS1/lipid complex structure The amino acid sequence of hSMS1 used in our study, which has 413 residues in total, was retrieved from PubMed (access ID＝NP_671512). The hSMS1 has an N-terminal domain, a transmembrane domain, and a C-terminal domain.16 Jiang et al. demonstrated recently that truncation of N-terminal and C-terminal of hSMS1 would not abort its activity.17 Thus, only the core structure of hSMS1 (M130-Q353) was modeled in our study. Locations of the extracellular or intracellular loop and transmembrane domain on the sequence of hSMS1 were predicted by using the PSI-PRED Server.18 The predicted extracellular or intracellular loops and transmembrane segments are given in Figure 3. This prediction is basically consistent with the results reported by Huitema et al. in a previous study.10 The transmembrane domain of hSMS1 is composed of six transmembrane helices (TMs). We employed 15 different computational methods to predict the locations of these TMs on the hSMS1 sequence.19-33 Most of them produced consistent predictions (Table 1). In order to select an appropriate template for modeling the TMs of hSMS1, we retrieved a total of 61 entries with six TMs from the membrane structural proteins database PDBTM (http://pdbtm.enzim.hu).34 After careful evaluations, which will be explained in the later Results and Discussion section, we selected the crystal structure of Escherichia coli GlpG (PDB entry: 2IC8) as the template (Table 2).35 The Modeler function (as implemented in the Discovery Studio software suite)36,37 was employed to generate a total of 30 structural models based on this template. As for the extracellular or intracellular loop of hSMS1, loop 3 (residues 235—274) is relatively long (Figure 3). Therefore, it should be modeled based on an appropriate template. PDB entry 1BW0 (residues 144— 181) (sequence identity＝34.1%; sequence similarity＝ 46.3%) was selected as the template for this purpose. This structure was selected throughout the entire PDB database according to the sequence similarity computed by the FASTA algorithm.38,39 No qualified template was found for other loops of hSMS1 though. These loops were constructed from scratch by using the Modeler module in the Discovery Studio software suite. A total of 30 structural models were also generated for each
Molecular Modeling of the Three-Dimensional Structure of Human Sphingomyelin Synthase CHEMISTRY ++++十十十十十+十十++++++++++十十十十十十十十十+++十十++++十十十十十十十十++++++++++十十十十十十十十十+++++++++十十十+十十十十十+++++++++++++十 MKEVVYWSPKKVADWLLENAMPEYCEPLEHFTGQDLINLTQEDEKKPPLCRVSSDNGORLLDMIETLKMEHHLEAHKNGHANGHLNIGVDIPTPDGSESI 110 130 140 150 170 180 200 +++++++十+++十+++++++++++++十++十工工工工工工 KIKPNGMPNGYRKEMIKIPMPELERSQYPMEWGKTFLAFLYALSCEVLTTVMISVVHERVPPKEVQPPLPDTFFDHFNRVQWAFSICEINGMILVGLWL 210 220 260 +++++++工工工工工X 00000XXXX工工工工工++++ OLLLKYKSIISRRFECIVGILYLYRCITMYVTILPVPGMHENCSPKLEGDWEAQLRRIMKLIAGGGLSITGSHNMCGDYLYSGHTVMLTLTYLEIKEYS 310 340 370 390 +++++++工工工工工工XXXo0000o--00000X 工工工++++++++++++十十++++++++++++十+++++++++++++++++++ PRRLWWYHWICWLLSVVGIECILLAHDHYTVDVVVAYYITTRLEWWYHTMANQQVLKEASQMNLLARVWWYRPFQYEEKNVQGIVPRSYHWPFPWPVVHL Figure 3 Transmembrane topology of hSMSl predicted by MEmsAt3. Here, the characters with broad lines, with thin lines and with- out line stand for transmembrane domains, loop regions, and terminal domains respectively (+ intracellular domains; the extracellular loop; O: Outside helix cap; X: Central transmembrane helix segment; 1: Inside helix cap) Table 1 Predicted locations of the trans-membrane helices on hSMS I Method N TMI TM3 TM4 TM5 TM6 AEMSTAT36M130-S154(25)Nl78W202(25)1210-1234(25)N275F295(21)1308-D332(25)V335-Q354(20) PSIPRED06E131H157(27)W821204(23)S209-1229(21)H285-E298(14)L304A325(22)V31-A351(21) Ssp26W32-H157(26)W82-K206(25)S2091228(20)T286F295(10)w305-A325(21)Y329353(25) Beta Pred26W132-R159(28)F184-k206(23)111-1234(24)T286-Y299(14)W306-A325(20)Y329-Q353(25) ConPred I235F136-V156(21)S185L205(21)F2151235(21) D279-￥299(21)Y307-D327(21) PROF- 6G133V156(24)R179K206(28)K2081233(26)T286K297(12)W306-L324(19)T330-M350(21) 6M130H157(28)Q181-L204(24)S209-1229(21)Y280-Y299(20)W305-A325(21)V331Q353(23) DPM 6M130-V160(31)Q181-k206(26)1210-1234(25)T286-F298(13)R302-H326(25)T330-Q353(24 DSC27 6M30-H157(28)W182-L205(24)1210-1233(24)V287-K297(11)R302-1323(22)D332Q353(22) GOR128 6M30-E158(29)R179L205(27)S209-P238(30)M276-k297021)I310-Y329(20)V331-Q353(23) GOR329 6M130V156(27)86-Y207(22)1210-1233(24)T286R302(17)L304A325(22)V331-Q353(23) 6W132-H157(26)1186-L205(20)S209-234(26)Y280-K297(18)R303-A325(23)V331-Q353(23) PHD 6E131-R159(29)Nl78L205(28)S209L235(27)C277-K297(21R303-L324(22)T330-Q353(24) Predator26Tl3V156(22)F184L205(2)S209T234(26)T286K297(12)W305-A325(21)T33353(24) 6T35-R159(25)W182-L205(24)S209-1234(26)M276k297(22)L304-324(21) ted number of Tms. Numbers in brackets are the lengths of tm Table 2 Comparison of the length of transmembrane helices on which the three key residues(H285, H328, and D332) hSMSI and GlpG(PDB entry: 21C8) could not form a reasonable arrangement in the catalytic Protein TM number TMI TM2 TM3 TM4 TM5 TM6 ock Iso excluded. Among the remaining hSMS I 6 252525212520 models, the one with the lowest probability density function(PDF)energy computed by Modeler was cho- GlpC 271720182118 sen for further refinement. In order to relax the steric repulsions between side chains, side chain refinement loop. All of the resulting models were then visually with restraints on backbone was performed. Each loop checked to exclude those having serious internal col was refined by high-level optimizations by using Mod- sions. Moreover, as for the third extracellular loop(loop eler. Side-chain refinement was performed 5, residues 327-339), two highly conserved residues cause the conformation of backbone could have bee H328 and D332 on this loop formed hydrogen bond, changed during the optimization performed at the pre- which could be used as an additional criterion to select vious step appropriate model for loop 5 Then. all of the whole structural models of hSMSI Molecular dynamics simulation of the hSMSI/SM were visually inspected to exclude those containing complex structure crossing loops or serious internal steric collisions Based on the final representing model of hSMSI Models on which the catalytic pocket were too small to SM was manually docked into the catalytic pocket in accommodate PC or SM were excluded. Models on favor of the SN2 nucleophilic substitution reaction. The Chin J. Chem. 2011, 29, 1567--1575 C2011 SIOC, CAS, Shanghai, WILEY-VCH Verlag gmbH Co KGaA, Weinheim wwcjc. wiley-vch. de 1569
Molecular Modeling of the Three-Dimensional Structure of Human Sphingomyelin Synthase Chin. J. Chem. 2011, 29, 1567—1575 © 2011 SIOC, CAS, Shanghai, & WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cjc.wiley-vch.de 1569 Figure 3 Transmembrane topology of hSMS1 predicted by MEMSAT3. Here, the characters with broad lines, with thin lines and without line stand for transmembrane domains, loop regions, and terminal domains respectively (＋: intracellular domains; -: the extracellular loop; O : Outside helix cap; X : Central transmembrane helix segment; I: Inside helix cap). Table 1 Predicted locations of the trans-membrane helices on hSMS1 Method Na TM1 TM2 TM3 TM4 TM5 TM6 MEMSTAT319 6 M130-S154 (25)b N178-W202 (25) I210-T234 (25) N275-F295 (21) I308-D332 (25) V335-Q354 (20) PSIPRED20 6 E131-H157 (27) W182-L204 (23) S209-T229 (21) H285-E298 (14) L304-A325 (22) V331-A351 (21) APSSP21 6 W132-H157 (26) W182-K206 (25) S209-I228 (20) T286-F295 (10) W305-A325 (21) Y329-Q353 (25) BetaTPred222 6 W132-R159 (28) F184-K206 (23) I211-T234 (24) T286-Y299 (14) W306-A325 (20) Y329-Q353 (25) ConPred II23 5 F136-V156 (21) S185-L205 (21) F215-L235 (21) D279-Y299 (21) Y307-D327 (21) PROF24 6 G133-V156 (24) R179-K206 (28) K208-T233 (26) T286-K297 (12) W306-L324 (19) T330-M350 (21) SSpro25 6 M130-H157 (28) Q181-L204 (24) S209-T229 (21) Y280-Y299 (20) W305-A325 (21) V331-Q353 (23) DPM26 6 M130-V160 (31) Q181-K206 (26) I210-T234 (25) T286-E298 (13) R302-H326 (25) T330-Q353 (24) DSC27 6 M130-H157 (28) W182-L205 (24) I210-T233 (24) V287-K297 (11) R302-L323 (22) D332-Q353 (22) GOR128 6 M130-E158 (29) R179-L205 (27) S209-P238 (30) M276-K297 (21) I310-Y329 (20) V331-Q353 (23) GOR329 6 M130-V156 (27) I186-Y207 (22) I210-T233 (24) T286-R302 (17) L304-A325 (22) V331-Q353 (23) MLRC30 6 W132-H157 (26) I186-L205 (20) S209-T234 (26) Y280-K297 (18) R303-A325 (23) V331-Q353 (23) PHD31 6 E131-R159 (29) N178-L205 (28) S209-L235 (27) C277-K297 (21 R303-L324 (22) T330-Q353 (24) Predator32 6 T135-V156 (22) F184-L205 (22) S209-T234 (26) T286-K297 (12) W305-A325 (21) T330-Q353 (24) SOPM33 6 T135-R159 (25) W182-L205 (24) S209-T234 (26) M276-K297 (22) L304-324 (21) T330-Q353 (24) a Predicted number of TMs. b Numbers in brackets are the lengths of TMs. Table 2 Comparison of the length of transmembrane helices on hSMS1 and GlpG (PDB entry: 2IC8) Protein TM number TM1 TM2 TM3 TM4 TM5 TM6 hSMS1 6 25 25 25 21 25 20 GlpG 6 27 17 20 18 21 18 loop. All of the resulting models were then visually checked to exclude those having serious internal collisions. Moreover, as for the third extracellular loop (loop 5, residues 327—339), two highly conserved residues H328 and D332 on this loop formed hydrogen bond, which could be used as an additional criterion to select appropriate model for loop 5. Then, all of the whole structural models of hSMS1 were visually inspected to exclude those containing crossing loops or serious internal steric collisions. Models on which the catalytic pocket were too small to accommodate PC or SM were excluded. Models on which the three key residues (H285, H328, and D332) could not form a reasonable arrangement in the catalytic pocket were also excluded. Among the remaining models, the one with the lowest probability density function (PDF) energy computed by Modeler was chosen for further refinement. In order to relax the steric repulsions between side chains, side chain refinement with restraints on backbone was performed. Each loop was refined by high-level optimizations by using Modeler. Side-chain refinement was performed again because the conformation of backbone could have been changed during the optimization performed at the previous step. Molecular dynamics simulation of the hSMS1/SM complex structure Based on the final representing model of hSMS1, SM was manually docked into the catalytic pocket in favor of the SN2 nucleophilic substitution reaction. The
FULL PAPER Zhang et al. binding mode of SM was then optimized within the 1.0 MPa. After these preparative steps, a production of structural restraints of hSMSl(Figure 4) Minimization 10 ns long was conducted under the same temperature was conducted using the adopted basis New. and pressure. During this process, the temperature was ton-Raphson algorithm with the CHARMm force field controlled with the Hoover method whereas the pres- implemented in the discovery studio software sure was controlled by coupling to a pressure bath using extended system algorithm. The mass of the pressure piston was 1000 amu. Langevin piston collision fre et to 25.0 ps The dielectric constants of 332 protein and water molecules were set to 1.0 and 80.0 respectively. Distance cutoff in generating the list of pairs was 14. 0 A Switching function was used between 10. A and 12.0 A to treat non-bonding interactions RMSD values relative to the initial structure were moni- tored as an indication of equilibrium along the MD trajectory. The last snapshot on the resulting md tra- jectory was retrieved, minimized without restraint in couple with the implicit membrane model. This final 8 efined model of the hSMSI/SM complex is shown in Figure 5. In order to evaluate this structural model, the PROCHECK program(version 3. 5.4 was applied to check its stereochemical quality. The Ramachandran plot of this hSMS I structure produced by PROCHECK Figure 4 The binding mode of sphingomyelin to hSMS1 Three lipid complex structure nesis study on the hSMSl/ Computational muta residues, i.e. H285, H328 and D332, in the catalytic site are criti- cal for the catalytic mechanism of hSMSI We then performed computational mutagenesis on the hSMSl complex structure to evaluate the impor In order to refine the structural model of the tance of both the functional groups on the lipid substrate hSMSI/SM complex, it was further subjected to a and several amino acid residues on hSMSISuch results long-time molecular dynamics (MD)simulation will provide additional proofs on the catalytic mecha- which an implicit GB/SA model of bilayer membrane nism of hSMS1. Two types of mutations were pe was applied. Major parameters for setting the implicit formed accordingly: on the substrate side, SM was mu membrane GB/SA model include: Grid spacing for tated in turn into PC, PE, PA, PS, and PG(Figure 1) lookup table (DGP)=1.5 A: Half membrane switching whereas on the hSMsl side, several important amino length(MSW)=2.0 A; Half smoothing length(SW) 3 A; Non-polar surface tension coefficient turn into alanine. Both types of mutations were done (SGAMMA)=0. 418 kJ/(mol-A2). Number of angular using the Discovery Studio software by deleting or ntegration points 50 was used for volume integration adding some atoms while keeping original atoms as GB/SA calculation. To keep the catalytic site stable for much as possible. After mutation, two rounds of mini- SN2 nucleophilic substitution reaction, three pairs of mization were applied to the hSMSIAipid complex harmonic distance restraint were applied with harmonic structure with the same force field parameters described force constant 418(kJ/molA): atom pairs 1-2, 4-5. above. Finally, binding energies between the substrate and 6-7(Figure 4). Firstly, hSMS 1/SM was positioned molecule and hSMS I were calculated using in the center ot the implicit membrahe model. The pla- hSMSi/ipid complex (Table 3 and Table 4) Iting tered at ==0 with a membrane thickness of 30 A. To- Normal mode analysis and protein motion analy pology and coordinate files were generated with the of hSMS Discovery Studio software. Minimization was per formed using the program CHARMm(release c33b2) We conducted protein domain motion analysis for Atotalof5000stepsofsteepestdescentminimizationtheDyndomon-lineserver(http://fizz.cmp.uea.ac.uk/ and 5000 steps of conjugate gradient minimization were dyndomn 4 The initial and the last configuration of the performed subsequently As the first step of MD simulation, the entire system hSMS I/SM complex in MD simulation were submitted was heated froma to in100 ps. Then, the as inputs. The Dyn Dom server returned the information whole system was equilibrated for 400 ps under a con- of the " hinge axes"(Figure 7). In order to confirm the stant temperature of 310 K and a constant pressure of mes of protein domain motion ana 1570www.cjc.wiley-vch.deO2011SioC,Cas,Shanghai&WileY-VchVerlagGmbh&Co.Kgaa,WeinheimChin.j.chem.2011,29,1567-1575
FULL PAPER Zhang et al. 1570 www.cjc.wiley-vch.de © 2011 SIOC, CAS, Shanghai, & WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chin. J. Chem. 2011, 29, 1567—1575 binding mode of SM was then optimized within the structural restraints of hSMS1 (Figure 4). Minimization was conducted using the adopted basis Newton-Raphson algorithm40 with the CHARMm force field implemented in the Discovery Studio software. P O O O O N N H H N N H N O O H285 H328 D332 1 2 3 4 5 6 7 8 Figure 4 The binding mode of sphingomyelin to hSMS1. Three residues, i.e. H285, H328 and D332, in the catalytic site are critical for the catalytic mechanism of hSMS1. In order to refine the structural model of the hSMS1/SM complex, it was further subjected to a long-time molecular dynamics (MD) simulation in which an implicit GB/SA model of bilayer membrane41 was applied. Major parameters for setting the implicit membrane GB/SA model include: Grid spacing for lookup table (DGP)＝1.5 Å; Half membrane switching length (MSW)＝2.0 Å; Half smoothing length (SW)＝ 0.3 Å; Non-polar surface tension coefficient (SGAMMA)＝0.418 kJ/(mol•Å2 ). Number of angular integration points 50 was used for volume integration in GB/SA calculation. To keep the catalytic site stable for SN2 nucleophilic substitution reaction, three pairs of harmonic distance restraint were applied with harmonic force constant 418 (kJ/mol•Å2 ): atom pairs 1—2, 4—5, and 6—7 (Figure 4). Firstly, hSMS1/SM was positioned in the center of the implicit membrane model. The planar membrane is perpendicular to the z axis and centered at z＝0 with a membrane thickness of 30 Å. Topology and coordinate files were generated with the Discovery Studio software. Minimization was performed using the program CHARMm (release c33b2).42 A total of 5000 steps of steepest descent minimization and 5000 steps of conjugate gradient minimization were performed subsequently. As the first step of MD simulation, the entire system was heated from 0 K to 310 K in 100 ps. Then, the whole system was equilibrated for 400 ps under a constant temperature of 310 K and a constant pressure of 1.0 MPa. After these preparative steps, a production of 10 ns long was conducted under the same temperature and pressure. During this process, the temperature was controlled with the Hoover method43 whereas the pressure was controlled by coupling to a pressure bath using extended system algorithm.44 The mass of the pressure piston was 1000 amu. Langevin piston collision frequency was set to 25.0 ps－1 . The dielectric constants of protein and water molecules were set to 1.0 and 80.0, respectively. Distance cutoff in generating the list of pairs was 14.0 Å. Switching function was used between 10.0 Å and 12.0 Å to treat non-bonding interactions. RMSD values relative to the initial structure were monitored as an indication of equilibrium along the MD trajectory. The last snapshot on the resulting MD trajectory was retrieved, minimized without restraint in couple with the implicit membrane model. This final refined model of the hSMS1/SM complex is shown in Figure 5. In order to evaluate this structural model, the PROCHECK program (version 3.5.4)45 was applied to check its stereochemical quality. The Ramachandran plot of this hSMS1 structure produced by PROCHECK is shown in Figure 6. Computational mutagenesis study on the hSMS1/ lipid complex structure We then performed computational mutagenesis on the hSMS1 complex structure to evaluate the importance of both the functional groups on the lipid substrate and several amino acid residues on hSMS1. Such results will provide additional proofs on the catalytic mechanism of hSMS1. Two types of mutations were performed accordingly: on the substrate side, SM was mutated in turn into PC, PE, PA, PS, and PG (Figure 1); whereas on the hSMS1 side, several important amino acid residues near the binding pocket were mutated in turn into alanine. Both types of mutations were done using the Discovery Studio software by deleting or adding some atoms while keeping original atoms as much as possible. After mutation, two rounds of minimization were applied to the hSMS1/lipid complex structure with the same force field parameters described above. Finally, binding energies between the substrate molecule and hSMS1 were calculated using an implicit GB/SA membrane model for each resulting hSMS1/lipid complex (Table 3 and Table 4). Normal mode analysis and protein motion analysis of hSMS1 We conducted protein domain motion analysis for hSMS1 to study its major conformational motions with the DynDom on-line server (http://fizz.cmp.uea.ac.uk/ dyndom/).46 The initial and the last configuration of the hSMS1/SM complex in MD simulation were submitted as inputs. The DynDom server returned the information of the "hinge axes" (Figure 7). In order to confirm the outcomes of protein domain motion analysis, normal
Molecular Modeling of the Three-Dimensional Structure of Human Sphingomyelin Synthase CHEMISTRY D3325 A28 R3 Figure 5 The hSMs 1/SM complex structure as the last snapshot of 10 ns molecular dynamics simulation. (A) hSMSI is rendered in ribbons. Several key residues around SM are shown in stick models. Dashed lines stand for hydrogen bonds. (B)hSMSI is rendered in the solvent accessible surface. SM is rendered in stick model. The arrow indicates where the choline moiety on SM is buried inside Table 3 Computed binding energies between hSMSl and some lipid substrates Complex Binding energy/(k.mol) ASMSI/PE 62.80 hSMS1/PA hSMSI/PS 63.76 hSMSI/PG 32.80 The binding energy between hSMSI and SM is taken as the energy rererence. Table 4 Computed binding energies between hSMSl mutations and SM 180-135-90-5 hhSMS I mutation Binding energy/ (kJmol) Phi° Figure 6 Ramachandran plot of the structural model of hSMSI D332A after molecular dynamics refinement. A: core alpha; a: allowed H285A 26.27 alpha; -a: general alpha; B: core beta; b: allowed beta, -b: gen-H328A 73.47 eral beta; L: core left-handed alpha; I: allowed left-handed alpha, R342A -l: general left-handed alpha, p: allowed epsilon,-p: genera Y338A epsilon. Glycines are shown as triangles. Here, 87.5% of residues 21.08 are in the most favored regions(A, B, L), 10.5% of residues are in the additionally allowed regions(a, b, 1, p), 1.0% of residues H274A 18.95 are in the generally allowed regions (a, b, -1, -p), and only L281A 1.0% of residues are in the disallowed regions F173A 17.61 mode analysis(NMA) was conducted to analyze the F3A+1A+74A+1281A5414 ntrinsic motions of the hSMSl structure. The eINemo The binding energy between SMSI and SM is set as the refer on-lineserver(http://igs-server.cnrs-mrs.fr/elnemo/ence Chin J. Chem. 2011, 29, 1567--1575 C2011 SIOC, CAS, Shanghai, WILEY-VCH Verlag gmbH Co KGaA, Weinheim wweje. wiley-vch.de 15
Molecular Modeling of the Three-Dimensional Structure of Human Sphingomyelin Synthase Chin. J. Chem. 2011, 29, 1567—1575 © 2011 SIOC, CAS, Shanghai, & WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cjc.wiley-vch.de 1571 Figure 5 The hSMS1/SM complex structure as the last snapshot of 10 ns molecular dynamics simulation. (A) hSMS1 is rendered in ribbons. Several key residues around SM are shown in stick models. Dashed lines stand for hydrogen bonds. (B) hSMS1 is rendered in the solvent accessible surface. SM is rendered in stick model. The arrow indicates where the choline moiety on SM is buried inside. Figure 6 Ramachandran plot of the structural model of hSMS1 after molecular dynamics refinement. A: core alpha; a: allowed alpha; ~a: general alpha; B: core beta; b: allowed beta; ~b: general beta; L: core left-handed alpha; l: allowed left-handed alpha; ~l: general left-handed alpha; p: allowed epsilon; ~p: general epsilon. Glycines are shown as triangles. Here, 87.5% of residues are in the most favored regions (A, B, L), 10.5% of residues are in the additionally allowed regions (a, b, l, p), 1.0% of residues are in the generally allowed regions (~a, ~b, ~l, ~p), and only 1.0% of residues are in the disallowed regions. mode analysis (NMA) was conducted to analyze the intrinsic motions of the hSMS1 structure. The elNémo on-line server (http://igs-server.cnrs-mrs.fr/elnemo/ Table 3 Computed binding energies between hSMS1 and some lipid substrates Complex Binding energy/(kJ•mol－1 ) hSMS1/SM 0.00a hSMS1/PC －27.90 hSMS1/PE 62.80 hSMS1/PA 123.80 hSMS1/PS 63.76 hSMS1/PG 32.80 a The binding energy between hSMS1 and SM is taken as the energy reference. Table 4 Computed binding energies between hSMS1 mutations and SM hhSMS1 mutation Binding energy/(kJ•mol－1 ) Wild type 0.00a D332A －0.25 H285A 26.27 H328A 73.47 R342A 81.79 Y338A 21.08 F177A 7.57 H274A 18.95 L281A 14.60 F173A 17.61 F173A＋F177A＋H274A＋L281A 54.14 a The binding energy between SMS1 and SM is set as the reference
FULL PAPER Zhang et al. index. html) was employed in our study for this pur- Table 5 Results of the normal mode analysis of the hSMSI This server makes use of the elastic Network tructure which provides a fast simple tool to compute ze, and analyze low-frequency normal modes of requena biological macromolecules. The structural model of mode 7 0.6182 hSMS I/SM was submitted for the NMA analysis. Key mode 8 107 0.6494 parameters used in computation included: DQMIN mode 9 1.23 0.6552 100, DQMAX=100, DQSTEP=20 and NRBL auto" The default cutoff of eight residues was used to identify elastic interaction ranges. A total of 100 normal mode 11 1.87 0.6801 modes with the lowest frequencies were computed Essential features of the top ten low-frequency normal 0.6152 modes, including its frequency, are summarized in Ta- 0.6192 ble 5. The two lowest-frequency normal modes ( mode 15 2.70 0.5726 modes 7 and 8 in Table 5)are illustrated in Figure 8 0.3389 the 10 normal modes with lowest frequencies are listed here, Collectivity indicates the percentage of residues that are involved in a certain normal mode Figure 8 The two primary normal modes of hsMSI with the lowest frequencies (Left: bending of the extracellular loops to- open state and a closed state, in which the extracellular loops tracellular loops on the top of the taa ight: rotation of the ex- Figure 7 The two major states of the hSMS I structure, i.e. an wards the transmembrane domain adopt a different orientation from the transmembrane domain he initial structure of hSMSI before MD refinement is rendered top-ranked binding poses were output after docking was in pale ribbons, the refined structure after 10 ns molecular dy- finished These binding poses predicted by GOld were namics simulation is rendered in dark ribbons. The arrow on the quite similar to each other. Therefore, we chose the top stands for the hinge axe around which the extracellular loops binding pose with the best binding score, and the corre rotate, which is given by the DynDom program. sponding hSMS 1/D609 complex structure was mini- mized with the same force field parameters mentioned Molecular docking of d609 to hSMSI above in couple with the implicit membrane GB/SA To understand how D609 affects the catalytic activ- model. The final model of the hSMSl/D609 complex is ity of hSMSl, the GOLd program(version 4.1, released illustrated in Figure 9 by CCDC Inc. was employed to perform automated lar docking to explore the possible binding poses Results and discussion between hSMSI and D609. The active site was defined as the residues within 10 A from the reference ligand. Homology modeling of the hSMS1/SM complex i.e. SM. Other key parameters used in docking included structure population size=100, number of GA operations In our study, it was of particular difficulty to select 500000, mutation rate=95%, crossover rate=95% and an appropriate template for homology modeling of the scoring function=Chemscore During docking process, hSMSI structure In fact, no other membrane protein the hSMSI structure was kept fixed. A total of 20 with known three-dimensional structure has a sequence 1572www.cjc.wiley-vch.de02011SioC,Cas,Shanghai&WIley-vcHVerlaggmbh&Co.Kgaa,WeinheimChinj.chem.2011,29,1567-1575
FULL PAPER Zhang et al. 1572 www.cjc.wiley-vch.de © 2011 SIOC, CAS, Shanghai, & WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chin. J. Chem. 2011, 29, 1567—1575 index.html)47 was employed in our study for this purpose. This server makes use of the Elastic Network Model, which provides a fast simple tool to compute, visualize, and analyze low-frequency normal modes of biological macromolecules. The structural model of hSMS1/SM was submitted for the NMA analysis. Key parameters used in computation included: DQMIN＝ －100, DQMAX＝100, DQSTEP＝20 and NRBL＝ “auto”. The default cutoff of eight residues was used to identify elastic interaction ranges. A total of 100 normal modes with the lowest frequencies were computed. Essential features of the top ten low-frequency normal modes, including its frequency, are summarized in Table 5. The two lowest-frequency normal modes (i.e. modes 7 and 8 in Table 5) are illustrated in Figure 8. Figure 7 The two major states of the hSMS1 structure, i.e. an open state and a closed state, in which the extracellular loops adopt a different orientation from the transmembrane domain. The initial structure of hSMS1 before MD refinement is rendered in pale ribbons; the refined structure after 10 ns molecular dynamics simulation is rendered in dark ribbons. The arrow on the top stands for the hinge axe around which the extracellular loops rotate, which is given by the DynDom program. Molecular docking of D609 to hSMS1 To understand how D609 affects the catalytic activity of hSMS1, the GOLD program (version 4.1, released by CCDC Inc.) was employed to perform automated molecular docking to explore the possible binding poses between hSMS1 and D609. The active site was defined as the residues within 10 Å from the reference ligand, i.e. SM. Other key parameters used in docking included: population size＝100, number of GA operations＝ 500000, mutation rate＝95%, crossover rate＝95% and scoring function＝Chemscore. During docking process, the hSMS1 structure was kept fixed. A total of 20 Table 5 Results of the normal mode analysis of the hSMS1 structure Modea Frequency Collectivityb mode 7 1.00 0.6182 mode 8 1.07 0.6494 mode 9 1.23 0.6552 mode 10 1.85 0.6581 mode 11 1.87 0.6801 mode 12 2.07 0.5377 mode 13 2.16 0.6152 mode 14 2.65 0.6192 mode 15 2.70 0.5726 mode 16 2.75 0.3389 a Only the 10 normal modes with lowest frequencies are listed here; b Collectivity indicates the percentage of residues that are involved in a certain normal mode. Figure 8 The two primary normal modes of hSMS1 with the lowest frequencies (Left: bending of the extracellular loops towards the transmembrane domain. Right: rotation of the extracellular loops on the top of the transmembrane domain). top-ranked binding poses were output after docking was finished. These binding poses predicted by GOLD were quite similar to each other. Therefore, we chose the binding pose with the best binding score, and the corresponding hSMS1/D609 complex structure was minimized with the same force field parameters mentioned above in couple with the implicit membrane GB/SA model. The final model of the hSMS1/D609 complex is illustrated in Figure 9. Results and discussion Homology modeling of the hSMS1/SM complex structure In our study, it was of particular difficulty to select an appropriate template for homology modeling of the hSMS1 structure. In fact, no other membrane protein with known three-dimensional structure has a sequence
Molecular Modeling of the Three-Dimensional Structure of Human Sphingomyelin Synthase CHEMISTRY p),1.0% of residues are in the generally allowed regions (a, -b, -l, -P), and only 1.0% of residues are in the disallowed regions. Thus, the quality of the hSMSI structural model derived in our study is acceptable in this aspect According to our model ( Figure 5A), residues H285 H328, and D332 are fully involved in the catalytic mechanism. D332 can form hydrogen bond with H328 which may act as hydrogen donor during catalysis. The choline group on the Sm molecule stretches well into a hydrophobic patch formed by F173, F177, H274, and L281. Y338 and R342 can form hydrogen bond with SM. With these two kinds of interactions, the catalytic center on SM, i.e. the phosphorus atom, is well posi tioned inside the catalytic pocket of hSMSl(Figure 5B) Our model(Figure 4)is generally consistent with the possible catalytic mechanism of hSMSl, which is given in the Introduction section of this article Figure 9 The binding mode of D609 with hSMSI derived Mutagenesis study on hSMSl/ipid complexes through molecular docking. hSMSI is rendered in ribbons whereas D609 is rendered in stick model. The shaded region in- According to our computation results(Table 4),mu- dicates the lipid bilayer tation of SM into other possible lipid substrates, in- cluding PG, PE, PA, and PS(Figure 1), leads to a de similarity high enough with hSMSI, especially for the crease in binding energy of hSMSI/PG,hSMSI/PE transmembrane domain. To tackle this problem, we re SMSI/PA, and hSMS 1/PS complexes as compared to trieved all known membrane proteins with six TMs, 61 that of the hSMSl/SM complex. However, mutation of PDBTM34 Then, we examined them visually one by above. The binding energy between hSMSl and PC template. Firstly, those largely different from hSMSI in hSMS1/SM. Based on the computed binding energies, terms of biological function, such as aquaporin-like the choline group on SM is very important to binding proteins, were excluded because they usually form a between hSMsI and SM. These results are consistent channel-like shape encircled by transmembrane helices. with Huitema's experimental proofs: none of the In addition, qualified candidates must be an enzyme non-choline phospholipids Secondly, for obvious reason, the TMs on the qualified than PC can trigger the production of sM. On the other candidates should be similar to those of hSMSI in terms hand, PC was efficiently recognized as a substrate of of length. After careful evaluation of all 61 entries, we SMS and SM itself had a donor of the phosphorylcho- chose the crystal structure of Escherichia coli GlpG line group, which can also be transformed back to PC (PDB entry: 2IC8)as the template. Indeed, the length of Therefore our model can, at least in a qualitatively TMs in GlpG is comparable to the counterparts in manner, explain the high selectivity of hSMSI towards hSMSI (Table 2). In addition, there is an internal cavity PC and sm on GlpG that harbors the Ser-His catalytic site, which is To better estimate the contribution of key residues to similar to hSMSl that catalyzes PC into SM. An ex pinding affinity between hSMSI and SM, computa- bi tracellular loop on GlpGare is also very similar to the tional mutagenesis, i.e. alanine scanning, was conducted counterpart on hSMSl with two conserved amino acid in our study on some residues near the catalytic site on esidues(H328 and D332) hSMSI. For example, H285, H328, and D332 are Our initial model of the hSMSI/SM complex struc- known to be three highly conserved amino acids in SMS ture was constructed through homology modeling ar families, even in lipid phosphatases/phosphotransferases, molecular docking. It was then subjected to a long-time and are key components participating in the catalytic MD simulation for refinement. An implicit membrane reaction. Mutation of these three key residues on model, which is basically a GB/SA model, was em- hSMSI will lose catalytic activity completely. In Ta ployed in simulation to mimic the real environment ble 4 and Figure 5A, one can see that mutation of round hSMSl. The final model of the hSMSI/SM com- D332A has little influence on the binding affinity be- plex structure is shown in Figure 5. The Ramachandran tween hSMSI and SM. However, the carboxyl group on lot of this hSMSl structure produced by Procheck d332 functions as a hydrogen acceptor to stabilize the (Figure 6)indicates that 87.5% of residues on our model transition state during the catalysis process. As for H285 are in the most favored regions(A, B, L), 10.5% of and H328, they are very important in both the binding residues are in the additionally allowed regions(a, b, 1, process and the subsequent catalytic process. Mutatio Chin J. Chem. 2011, 29, 1567--1575 C2011 SIOC, CAS, Shanghai, WILEY-VCH Verlag gmbH Co KGaA, Weinheim wwcjc. wiley-vch. de 1573
Molecular Modeling of the Three-Dimensional Structure of Human Sphingomyelin Synthase Chin. J. Chem. 2011, 29, 1567—1575 © 2011 SIOC, CAS, Shanghai, & WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cjc.wiley-vch.de 1573 Figure 9 The binding mode of D609 with hSMS1 derived through molecular docking. hSMS1 is rendered in ribbons whereas D609 is rendered in stick model. The shaded region indicates the lipid bilayer. similarity high enough with hSMS1, especially for the transmembrane domain. To tackle this problem, we retrieved all known membrane proteins with six TMs, 61 in total, from the membrane structural proteins database PDBTM.34 Then, we examined them visually one by one with two criteria to select the most suitable one as template. Firstly, those largely different from hSMS1 in terms of biological function, such as aquaporin-like proteins, were excluded because they usually form a channel-like shape encircled by transmembrane helices. In addition, qualified candidates must be an enzyme. Secondly, for obvious reason, the TMs on the qualified candidates should be similar to those of hSMS1 in terms of length. After careful evaluation of all 61 entries, we chose the crystal structure of Escherichia coli GlpG (PDB entry: 2IC8) as the template. Indeed, the length of TMs in GlpG is comparable to the counterparts in hSMS1 (Table 2). In addition, there is an internal cavity on GlpG that harbors the Ser-His catalytic site, which is similar to hSMS1 that catalyzes PC into SM. An extracellular loop on GlpGare is also very similar to the counterpart on hSMS1 with two conserved amino acid residues (H328 and D332). Our initial model of the hSMS1/SM complex structure was constructed through homology modeling and molecular docking. It was then subjected to a long-time MD simulation for refinement. An implicit membrane model, which is basically a GB/SA model, was employed in simulation to mimic the real environment round hSMS1. The final model of the hSMS1/SM complex structure is shown in Figure 5. The Ramachandran plot of this hSMS1 structure produced by PROCHECK (Figure 6) indicates that 87.5% of residues on our model are in the most favored regions (A, B, L), 10.5% of residues are in the additionally allowed regions (a, b, l, p), 1.0% of residues are in the generally allowed regions (~a, ~b, ~l, ~p), and only 1.0% of residues are in the disallowed regions. Thus, the quality of the hSMS1 structural model derived in our study is acceptable in this aspect. According to our model (Figure 5A), residues H285, H328, and D332 are fully involved in the catalytic mechanism. D332 can form hydrogen bond with H328, which may act as hydrogen donor during catalysis. The choline group on the SM molecule stretches well into a hydrophobic patch formed by F173, F177, H274, and L281. Y338 and R342 can form hydrogen bond with SM. With these two kinds of interactions, the catalytic center on SM, i.e. the phosphorus atom, is well positioned inside the catalytic pocket of hSMS1 (Figure 5B). Our model (Figure 4) is generally consistent with the possible catalytic mechanism of hSMS1, which is given in the Introduction section of this article. Mutagenesis study on hSMS1/lipid complexes According to our computation results (Table 4), mutation of SM into other possible lipid substrates, including PG, PE, PA, and PS (Figure 1), leads to a decrease in binding energy of hSMS1/PG, hSMS1/PE, hSMS1/PA, and hSMS1/PS complexes as compared to that of the hSMS1/SM complex. However, mutation of SM into PC is quite different from lipid without choline above. The binding energy between hSMS1 and PC is a bit lower, i.e. more favorable, if compared to hSMS1/SM. Based on the computed binding energies, the choline group on SM is very important to binding between hSMS1 and SM. These results are consistent with Huitema’s experimental proofs:10 none of the non-choline phospholipids (PG, PE, PA, and PS) other than PC can trigger the production of SM. On the other hand, PC was efficiently recognized as a substrate of SMS and SM itself had a donor of the phosphorylcholine group, which can also be transformed back to PC. Therefore our model can, at least in a qualitatively manner, explain the high selectivity of hSMS1 towards PC and SM. To better estimate the contribution of key residues to binding affinity between hSMS1 and SM, computational mutagenesis, i.e. alanine scanning, was conducted in our study on some residues near the catalytic site on hSMS1. For example, H285, H328, and D332 are known to be three highly conserved amino acids in SMS families, even in lipid phosphatases/phosphotransferases, and are key components participating in the catalytic reaction. Mutation of these three key residues on hSMS1 will lose catalytic activity completely.17 In Table 4 and Figure 5A, one can see that mutation of D332A has little influence on the binding affinity between hSMS1 and SM. However, the carboxyl group on D332 functions as a hydrogen acceptor to stabilize the transition state during the catalysis process. As for H285 and H328, they are very important in both the binding process and the subsequent catalytic process. Mutation
FULL PAPER Zhang et al. of H285A and H328A will lead to a decrease in binding The binding mode of a small-molecule inhibitor of with value of 26.27, and 73. 47 kJ/mol respec- hSMsI These results demonstrate the rationale of our D609 is essentially the only known small-molecule inhibitor of hSMSI so far. Interestingly, dimerization of Notably, our model also suggests some key residues D609, i.e. D609 dixanthogen(Figure 2), was observed which have not been well studied before. Mutation of to have no inhibition activity at all. Based on the R342A caused great loss of binding energy with value structural model of the hSMS I/SM complex derived in of 81.79 kJ/mol. This datum indicates that the hydrogen our study, a binding mode of D609 was derived through bonding interaction between SM and R342 is important molecular docking. Our model of the ASMS1/D609 for locking the conformation of SM to facilitate the complex(Figure 9)indicates that D609 occupies the catalyzed reaction. Mutation of Y338A also causes a catalytic pocket, which blocks the entry of substrate decrease in binding energy of ca. 21.08 kJ/mol. One can molecules, such as PC and SM. The key residues in- see in Figure 5A that Y338 can form favorable hydro- volved in the binding of D609 include R342 and H285 phobic interactions with one of the carbon chain on SM. In particular, the negative charged sulfur atom on D609 In addition, Y338 can form a hydrogen bond with R342 forms a strong salt-bridge interaction with R342, which In contrast, F173A, F177A, H274A, and L281A muta- mimics the two critical hydrogen bonds between tions do not have a major impact on the binding energy SMSI and SM. d609 dixanthogen does not have the of SM(Table 4). However, the quadruple mutant F173A negatively charged sulfur atom, leading to the loss of +F177A+H274A+ L281A leads to a large decrease of the salt-bridge with R342. In addition, the steric hin- 54. 14 kJ/mol in binding energy. It indicates that the hy- drance near R342 does not allow the second D609 drophobic interactions involving these residues are co- moiety on D609 dixanthogen to go into the binding operative even though the contribution of each individ- pocket deeply. Thus, our model provides a reasonable lal residue is trivial. These residues form a hydrophobic explanation on the different hSMSl inhibition activities patch surrounding the choline group of SM, illustrating between D609 and D609 dixanthogen the importance of choline group in the PC and SM. 0 Activation mechanism of hsMsi based on nma and Conclusions We have obtained a three-dimensional structural Protein domain motion analysis indicated that the model of hSMSI through homology modeling, molecu- extracellular loops of hSMSI rotate around the hinge lar docking, and extensive molecular dynamics simula axis shown in Figure 7. In order to further detect such tions. Our model reasonably explains how hSMs conformational motions, we performed normal mode transforms a phosphocholine moiety on ceramide to analysis(NMA)on the final snapshot on the MD tra oduce sphingomyelin. It also explains the high selec jectory of the hSMSI/SM complex. The two lowest- tivity of hSMsI towards phosphocholine and sphingo- frequency normal modes(i.e. modes 7 and 8 in Table 3) myelin. Normal mode analysis on the low-frequency are illustrated in Figure 8: one mode features bending of motions of hSMSI suggests a gating mechanism of the loop towards the transmembrane domain whereas immobilizing phosphatidylcholine at the catalytic site the other features rotating of the loop on top of the and releasing sphingomyelin afterwards. Computational transmembrane domain. This motion prompts that there mutagenesis results explore the key residues for the can be two major states existing in the physiological binding of SM to hSMSl. Currently, very few state: an open state and a closed state. These two states small-molecule inhibitors of hSMs I have been reporte are connected by low-frequency conformational mo- publicly in literature. The potential value of our struc- tions. This gating mechanism of the extracellular loops tural model is that it can be applied to the discovery of on hSMSl illustrates how the reactant PC is locked after effective hSMSI inhibitors, for example, through virtual it enters the catalytic pocket and how the product SM is eleased afterwards. It also illustrates how an inhibitor may block the function of hSMSl. In particular, the References choline moiety on PC can form favorable hydrophobic I Merrill, A H Jr, Jones, DD. Biochim. Biophys. Acta 1990, interactions with F173. F177. H274. and L281 in the 1044.1 closed state, which immobilizes the choline moiety and 2 Nagao, K, Takahashi, K, Hanada, K, Kioka, N; Matsuo, locks the shape of the catalytic center(Figure 5A). The M. Ueda, K.J. Bio. Chem. 2007, 282, 14868 lowest frequency motions that induce the movement of Sano, O, Kobayashi, A, Nagao, K, Kumagai, K, Kioka F173, F177, and H274 on the loops away from the N, Hanada, K; Ueda, K, Matsuo, M. J. Lipid Res 2007, binding site will disassemble the hydrophobic patch 48,2377 originally accommodating the choline moiety in the 4 Schutze, S: Potthoff, K. Machleidt, T, Berkovic, D closed state. Consequently, SM will be released from wiegmann, K, Kronke, M. Cell 1992, 71, 765 the catalytic site in the open state 5 Luberto. C. Yoo. D. S: Suidan. H. S: Bartoli. G. M 1574www.cjc.wiley-vch.deO2011SioC,Cas,Shanghai&WileY-VchVerlagGmbh&Co.Kgaa,WeinheimChin.j.chem.2011,29,1567-1575
FULL PAPER Zhang et al. 1574 www.cjc.wiley-vch.de © 2011 SIOC, CAS, Shanghai, & WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chin. J. Chem. 2011, 29, 1567—1575 of H285A and H328A will lead to a decrease in binding energy with value of 26.27, and 73.47 kJ/mol respectively. These results demonstrate the rationale of our model further. Notably, our model also suggests some key residues which have not been well studied before. Mutation of R342A caused great loss of binding energy with value of 81.79 kJ/mol. This datum indicates that the hydrogen bonding interaction between SM and R342 is important for locking the conformation of SM to facilitate the catalyzed reaction. Mutation of Y338A also causes a decrease in binding energy of ca. 21.08 kJ/mol. One can see in Figure 5A that Y338 can form favorable hydrophobic interactions with one of the carbon chain on SM. In addition, Y338 can form a hydrogen bond with R342. In contrast, F173A, F177A, H274A, and L281A mutations do not have a major impact on the binding energy of SM (Table 4). However, the quadruple mutant F173A ＋F177A＋H274A＋L281A leads to a large decrease of 54.14 kJ/mol in binding energy. It indicates that the hydrophobic interactions involving these residues are cooperative even though the contribution of each individual residue is trivial. These residues form a hydrophobic patch surrounding the choline group of SM, illustrating the importance of choline group in the PC and SM.10 Activation mechanism of hSMS1 based on NMA and DynDom analysis Protein domain motion analysis indicated that the extracellular loops of hSMS1 rotate around the hinge axis shown in Figure 7. In order to further detect such conformational motions, we performed normal mode analysis (NMA) on the final snapshot on the MD trajectory of the hSMS1/SM complex. The two lowestfrequency normal modes (i.e. modes 7 and 8 in Table 3) are illustrated in Figure 8: one mode features bending of the loop towards the transmembrane domain whereas the other features rotating of the loop on top of the transmembrane domain. This motion prompts that there can be two major states existing in the physiological state: an open state and a closed state. These two states are connected by low-frequency conformational motions. This gating mechanism of the extracellular loops on hSMS1 illustrates how the reactant PC is locked after it enters the catalytic pocket and how the product SM is released afterwards. It also illustrates how an inhibitor may block the function of hSMS1. In particular, the choline moiety on PC can form favorable hydrophobic interactions with F173, F177, H274, and L281 in the closed state, which immobilizes the choline moiety and locks the shape of the catalytic center (Figure 5A). The lowest frequency motions that induce the movement of F173, F177, and H274 on the loops away from the binding site will disassemble the hydrophobic patch originally accommodating the choline moiety in the closed state. Consequently, SM will be released from the catalytic site in the open state. The binding mode of a small-molecule inhibitor of hSMS1 D609 is essentially the only known small-molecule inhibitor of hSMS1 so far. Interestingly, dimerization of D609, i.e. D609 dixanthogen (Figure 2), was observed to have no inhibition activity at all.13 Based on the structural model of the hSMS1/SM complex derived in our study, a binding mode of D609 was derived through molecular docking. Our model of the hSMS1/D609 complex (Figure 9) indicates that D609 occupies the catalytic pocket, which blocks the entry of substrate molecules, such as PC and SM. The key residues involved in the binding of D609 include R342 and H285. In particular, the negative charged sulfur atom on D609 forms a strong salt-bridge interaction with R342, which mimics the two critical hydrogen bonds between hSMS1 and SM. D609 dixanthogen does not have the negatively charged sulfur atom, leading to the loss of the salt-bridge with R342. In addition, the steric hindrance near R342 does not allow the second D609 moiety on D609 dixanthogen to go into the binding pocket deeply. Thus, our model provides a reasonable explanation on the different hSMS1 inhibition activities between D609 and D609 dixanthogen. Conclusions We have obtained a three-dimensional structural model of hSMS1 through homology modeling, molecular docking, and extensive molecular dynamics simulations. Our model reasonably explains how hSMS1 transforms a phosphocholine moiety on ceramide to produce sphingomyelin. It also explains the high selectivity of hSMS1 towards phosphocholine and sphingomyelin. Normal mode analysis on the low-frequency motions of hSMS1 suggests a gating mechanism of immobilizing phosphatidylcholine at the catalytic site and releasing sphingomyelin afterwards. Computational mutagenesis results explore the key residues for the binding of SM to hSMS1. Currently, very few small-molecule inhibitors of hSMS1 have been reported publicly in literature. The potential value of our structural model is that it can be applied to the discovery of effective hSMS1 inhibitors, for example, through virtual screening. References 1 Merrill, A. H. Jr.; Jones, D. D. Biochim. Biophys. Acta 1990, 1044, 1. 2 Nagao, K.; Takahashi, K.; Hanada, K.; Kioka, N.; Matsuo, M.; Ueda, K. J. Biol. Chem. 2007, 282, 14868. 3 Sano, O.; Kobayashi, A.; Nagao, K.; Kumagai, K.; Kioka, N.; Hanada, K.; Ueda, K.; Matsuo, M. J. Lipid Res. 2007, 48, 2377. 4 Schutze, S.; Potthoff, K.; Machleidt, T.; Berkovic, D.; Wiegmann, K.; Kronke, M. Cell 1992, 71, 765. 5 Luberto, C.; Yoo, D. S.; Suidan, H. S.; Bartoli, G. M.;
Molecular Modeling of the Three-Dimensional Structure of Human Sphingomyelin Synthase CHEMISTRY Hannun, Y. A.J. Biol Chem. 2000, 275, 14760 5 Cheng, J. Randall, A. Z, Sweredoski, M. J, Baldi, P. 6 Hailemariam, T.K., Huan, C, Liu, J. Li, Z, Roman, C Nucleic Acids Res 2005.33 W72 G. Arterioscler. H H Peake. D.A.: Kuo, M S; Cao, 26 Deleage,G, Roux, B Protein Eng.1987, 1,289 Kalbfeisch. M.. Bui omb.asc.Biol.2008,28,1519 27 King, R. D; Sternberg, M.J. Protein Sci. 1996, 5, 2298 7 Miyaji, M., Jin, Z.X., Yamaoka, S, Amakawa, R; Fuku- 28 Garnier, J; Osguthorpe, D.J.; Robson, B.J.Mol. Biol. 1978 hara, S; Sato, S. B. Kobayashi, T ; Domae, N; Mimori, T. 120.97. Bloom, E. T. Okazaki, T, Umehara, H.J.Exp. Med. 2005, 29 Gibrat, J. F: Garnier, J: Robson. B.J. Mol Biol. 1987.198 202.249 8 Van der Luit, A. H-, Budde, M. Zerp, S. Caan,W, 30 Guermeur, Y; Geourjon, C. Gallinari, P; Deleage, G Klarenbeek Verheij, M, Van Bli Bioinformatics 1999, 15, 413 Biochem.. 2007. 40/ 541 31 Rost, B: Sander, C.J. Mol. Bio/. 1993, 232, 584 9 Ding, T ; Li, Z, Hailemariam, T; Mukherjee, S: Maxfield, F.R., Wu, M. P. Jiang, X. C.J. Lipid Res. 2008, 49, 3 32 Frishman, D ; Argos, P Protein Eng. 1996, 9, 133 33 Geourjon, C; Deleage, G Protein Eng. 1994, 7, 157 10 Huitema, K, van den Dikkenberg, J; Brouwers, J.F., 34 Tusnady, G. E; Dosztanyi, Z; Simon, I. Bioinformatics thuis. J. C. Embo. 2004. 23. 33 004.20.2964 11 Yamaoka, S; Miyaji, M. Kitano, T; Umehara, H; Okazaki T.J.Biol.Chem.2004,279,18688 35 Wang, Y Zhang, Y Ha, Y Nature 2006, 444, 179 12 Tafesse. F. G. Ternes. P. Holthuis J. C. J. Bio/. Chem 36 Marti-Renom, M. A, Stuart, A. C, Fiser, A Sanchez, R: 2006,281,2942 Melo, F. Sali, A. Annu. Rev. Biophys. Biomol. Struct. 2000, 29.291 13 Meng, A, Luberto, C ; Meier, P, Bai, A; Yang, X, Han- nun,Y A, Zhou, D Exp Cell 37 Discovery Studio Sofware, Version 1. 7, Accelyrs Inc, San 4 Bai, A. Meier, G. P; Wang, Y, Luberto, C; Hannun,YA Diego, California, U.S. A. 2007 Zhou, D.J. Pharmacol. Exp Ther. 2004, 309, 1051 38 Bhat, T N. Bourne, P Feng, Z Nucleic Acid Res. 2001, 29 15 Dong, J.B.; Liu, J; Lou, B. Li, Z. Q; Wu, M P; Jiang, X. 39 Pearson, W.R. Methods Enymol 1990,183,63 C.J. Lipid Res.2006,47,1307 16 Sigal. Y. J. McDermott. M.I.: Morris. A. J. Biochem. 40 Momany, F. A, Rone, R.J. Comp 1992,I3,888 2005.387,281 41 Im, w, Feig, M; Brooks, C L. Biophys J. 2003, 85, 2900 17 Yeang, C: Varshney, S; Wang, R; Zhang, Y, Ye, D. Y: 42 Brooks, B. R,; Bruccoleri,R. E; Olafson, B. D; States, D Jiang, x. C. Biochim. Biophys. Acta 2008, 1781, 610 J, Swaminathan, S, Karplus, M.J. Comput. Chem. 1983, 4 Sodhi. J S. Jones D. T. Nucleic Acids Res. 2005. 33. W36 43 Hoover, W.G. Phys. Rev. 1985, 31, 1695 19 Jones, D. T Bioinformatics 2007, 23, 53 44 Andersen, H. C.J. Chem. Phys. 1980. 72, 2384 20 Jones. D. T J Mol Biol. 1999.. 19 45 Laskowski. R. A: Mac Arthur. M. W: Moss. D. S 21 Raghava, G. P S CASP4 2000, 75 Thornton, J. M.J. Appl. Cryst. 1993, 26, 283 22 Kaur, H; Raghava, G. P. Protein Sci. 2003, 12, 627 46 Lee, R. A; Razaz, M. Hayward, S Bioinformatics 2003, 19, 23 Arai, M: Mitsuke, H ; Ikeda, M 1290. Satake, M.: Shimizu, T. Nucleic Acids Res. 2004, 32, W30 Sanejouand, Y.H. Nucleic Acid Res. 2004, 32, Ouali, M. King, R. D. Protein Sci. 2000, 9, 1162 W610 ng, F; Fan, Y) Chin J. Chem. 2011, 29, 1567--1575 C2011 SIOC, CAS, Shanghai, WILEY-VCH Verlag gmbH Co KGaA, Weinheim wweje. wiley-vch.de 1575
Molecular Modeling of the Three-Dimensional Structure of Human Sphingomyelin Synthase Chin. J. Chem. 2011, 29, 1567—1575 © 2011 SIOC, CAS, Shanghai, & WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cjc.wiley-vch.de 1575 Hannun, Y. A. J. Biol. Chem. 2000, 275, 14760. 6 Hailemariam, T. K.; Huan, C.; Liu, J.; Li, Z.; Roman, C.; Kalbfeisch, M.; Bui, H. H.; Peake, D. A.; Kuo, M. S.; Cao, G. Arterioscler. Thromb. Vasc. Biol. 2008, 28, 1519. 7 Miyaji, M.; Jin, Z. X.; Yamaoka, S.; Amakawa, R.; Fukuhara, S.; Sato, S. B.; Kobayashi, T.; Domae, N.; Mimori, T.; Bloom, E. T.; Okazaki, T.; Umehara, H. J. Exp. Med. 2005, 202, 249. 8 Van der Luit, A. H.; Budde, M.; Zerp, S.; Caan, W.; Klarenbeek, J. B.; Verheij, M.; Van Blitterswijk, W. J. Biochem. J. 2007, 401, 541. 9 Ding, T.; Li, Z.; Hailemariam, T.; Mukherjee, S.; Maxfield, F. R.; Wu, M. P.; Jiang, X. C. J. Lipid Res. 2008, 49, 376. 10 Huitema, K.; van den Dikkenberg, J.; Brouwers, J. F.; Holthuis, J. C. Embo. J. 2004, 23, 33. 11 Yamaoka, S.; Miyaji, M.; Kitano, T.; Umehara, H.; Okazaki, T. J. Biol. Chem. 2004, 279, 18688. 12 Tafesse, F. G.; Ternes, P.; Holthuis, J. C. J. Biol. Chem. 2006, 281, 29421. 13 Meng, A.; Luberto, C.; Meier, P.; Bai, A.; Yang, X.; Hannun, Y. A.; Zhou, D. Exp. Cell. Res. 2004, 292, 385. 14 Bai, A.; Meier, G. P.; Wang, Y.; Luberto, C.; Hannun, Y. A.; Zhou, D. J. Pharmacol. Exp. Ther. 2004, 309, 1051. 15 Dong, J. B.; Liu, J.; Lou, B.; Li, Z. Q.; Wu, M. P.; Jiang, X. C. J. Lipid Res. 2006, 47, 1307. 16 Sigal, Y. J.; McDermott, M. I.; Morris, A. J. Biochem. J. 2005, 387, 281. 17 Yeang, C.; Varshney, S.; Wang, R.; Zhang, Y.; Ye, D. Y.; Jiang, X. C. Biochim. Biophys. Acta 2008, 1781, 610. 18 Bryson, K.; McGuffin, L. J.; Marsden, R. L.; Ward, J. J.; Sodhi, J. S.; Jones, D. T. Nucleic Acids Res. 2005, 33, W36. 19 Jones, D. T. Bioinformatics 2007, 23, 538. 20 Jones, D. T. J. Mol. Biol. 1999, 292, 195. 21 Raghava, G. P. S. CASP4 2000, 75. 22 Kaur, H.; Raghava, G. P. Protein Sci. 2003, 12, 627. 23 Arai, M.; Mitsuke, H.; Ikeda, M.; Xia, J. X.; Kikuchi, T.; Satake, M.; Shimizu, T. Nucleic Acids Res. 2004, 32, W390. 24 Ouali, M.; King, R. D. Protein Sci. 2000, 9, 1162. 25 Cheng, J.; Randall, A. Z.; Sweredoski, M. J.; Baldi, P. Nucleic Acids Res. 2005, 33, W72. 26 Deleage, G.; Roux, B. Protein Eng. 1987, 1, 289. 27 King, R. D.; Sternberg, M. J. Protein Sci. 1996, 5, 2298. 28 Garnier, J.; Osguthorpe, D. J.; Robson, B. J. Mol. Biol. 1978, 120, 97. 29 Gibrat, J. F.; Garnier, J.; Robson, B. J. Mol. Biol. 1987, 198, 425. 30 Guermeur, Y.; Geourjon, C.; Gallinari, P.; Deleage, G. Bioinformatics 1999, 15, 413. 31 Rost, B.; Sander, C. J. Mol. Biol. 1993, 232, 584. 32 Frishman, D.; Argos, P. Protein Eng. 1996, 9, 133. 33 Geourjon, C.; Deleage, G. Protein Eng. 1994, 7, 157. 34 Tusnady, G. E.; Dosztanyi, Z.; Simon, I. Bioinformatics 2004, 20, 2964. 35 Wang, Y.; Zhang, Y.; Ha, Y. Nature 2006, 444, 179. 36 Marti-Renom, M. A.; Stuart, A. C.; Fiser, A.; Sanchez, R.; Melo, F.; Sali, A. Annu. Rev. Biophys. Biomol. Struct. 2000, 29, 291. 37 Discovery Studio Software, Version 1. 7, Accelyrs Inc., San Diego, California, U. S. A. 2007. 38 Bhat, T. N.; Bourne, P.; Feng, Z. Nucleic Acid Res. 2001, 29, 214. 39 Pearson, W. R. Methods Enzymol. 1990, 183, 63. 40 Momany, F. A.; Rone, R. J. Comp. Chem. 1992, 13, 888. 41 Im, W.; Feig, M.; Brooks, C. L. Biophys. J. 2003, 85, 2900. 42 Brooks, B. R.; Bruccoleri, R. E.; Olafson, B. D.; States, D. J.; Swaminathan, S.; Karplus, M. J. Comput. Chem. 1983, 4, 187. 43 Hoover, W. G. Phys. Rev. 1985, 31, 1695. 44 Andersen, H. C. J. Chem. Phys. 1980, 72, 2384 45 Laskowski, R. A.; MacArthur, M. W.; Moss, D. S.; Thornton, J. M. J. Appl. Cryst. 1993, 26, 283. 46 Lee, R. A.; Razaz, M.; Hayward, S. Bioinformatics 2003, 19, 1290. 47 Suhre, K.; Sanejouand, Y. H. Nucleic Acid Res. 2004, 32, W610. (E1101051 Cheng, F.; Fan, Y.)