Biochimie (1996) 78, 771-779

SociEt6 franqaise de biochimie et biologie moi6culaire / Elsevier, Paris

Computer modeling of 3D structures of cytochrorne P450s

YT Chang, OB Stiffelman, GH Loew
Molecular Resealz'h Institute, 845 Page Mill Road, Palo Alto. CA 94304. USA

(Received 7 June !996; accepteo 14 August 1996)

Summary - - The understanding of structure-function relationship of enzymes requires detailed information of their three-dimensional structure. Protein structure determination by X-ray and NMR methods, the two most frequently used experimental procedures, are often difficult and time-consuming. Thus computer modeling of protein structures has become an increasingly active and attractive option for obtaining predictive models of three-dimensional protein structures. Specifically, for the ubiquitous metabolizing heme proteins, the cytochrome P450s, the X-ray structures of four isozymes of bacterial origin, P450cam, P450terp, P450BM-3 and P450eryF have now been determined. However, attempts to obtain the structure of mammalian forms by experimental means have thus far not been successful. Thus, there have been numerous attempts to construct models of mammalian P450s using homology modeling methods in which the known structures have been used to various extents and in various strategies to build models of P450 isozymes. In this paper, we review these efforts and then describe a strategy for structure building and assessment of 3D models of P450s recently developed in our laboratory that corrects many of the weaknesses in the previous procedures. The results are 3D models that for the first time are stable to unconstrained molecular dynamics simulations. The use of this method is demonstrated by the construction and validation of a 3D model for rabbit liver microsomal P450 isozyme 2B4, responsible for the oxidative metabolism of diverse xenobiotics including widely used inhalation anesthetics. Using this 2B4 model, the substrate access channel, substrate binding site and plausible surface regions for binding with P450 redox partners were identified. homology modeling / P450choP / P4502B4 / benzphetamine / androste'~edione Introduction
Cytochrome P450s are a ubiquitous superfamily of metabolizing heme proteins. Using recombinant DNA technology, mole than 250 isozymes belonging to --40 different families have been cloned and sequenced [ll. These cytochrome P450 enzymes are present in a broad range of species including plants, fungi, bacteria, insects and mammals, where they participate in the phase I metabolism of many endogenous and exogenous compounds. The er, zymatic cycle of these enzymes is believed to include substrate binding, electron transfer, oxygen bindin~ a~d activation, and substrate oxidation. All P450s contain a common prosthetic group, an iron-protoporphyrin IX complex as the donor of the reactive oxygen atom during substrate oxidation. In addition, all P450s contain a cysteine residue as an axial ligand of the iron of this prosthetic group. Substrates bind in the opposite or distal side of the heme unit. In contrast to the common heme unit and conserved proximal cysteine ligand, there is a large variation in this binding site region in both amino acid composition and architecture among these isozymes leading to their unique substrate specificity and regio- and/or stereo-product specificity. Furthermore, there are also other important differences among the P450s such as the source of the two electrons, the various redox partners required, and the involvement of the membrane, for nonbacterial systems, in the enzymatic cycle. In order to elucidate the structure-function relationship of these P450s and the differences among them, sequence and structural information are required. Unlike the determination of a P450 sequence, which is now rather straightforward due to ihe d e v e l o p m e n t of r e c o m b i n a n t DNA techniques, experimental determination of a P450 structure, especially for the membrane-bound isozymes, has proven very difficult. Advances in our understanding of the structure and function relationship of P450s have benefited from the first resolved P450 crystal structure, the cytochrome P450cam, of a bacterial source [2]. Recently, three more bacterial P450 structures including P450terp [3, 4], P450BM-3 [5, 6] and P450eryF [7] have been solved by X-ray crystallography. However, although the crystallization of a mammalian P450scc (CYPI 1o~) has been reported [8], no structure of this isozyme or any other mammalian P450s has yet been achieved. Lack of structural information for these mammalian P450s is now one of the main deterrents to further understanding of their function. Because of the importance of this problem, a number of experimental efforts have been devised to obtain partial information at least of the binding site architecture. One strategy, developed by Paul Ortiz de Montellano and coworkers [9-13] has been to monitor the site-specific formation of covalent adducts of hydrazine substrates such as phenyldiazene, phenylhydrazine, or arylhydrazines with the four possible pyrrole nitrogen atoms of the heme porphyrin ring of various bacterial and mammalian P450 isozymes.

772 This method thus provides a measure of the relative accessible space above each pyrrole rings in the substrate binding site. The effect of site specific mutations on substrate and/or product specificity is another method for proriding information about important residues in the active site. A l t h o u ~ very useful, the overall amino acid composition and the spatial arrangement of the binding site residues cannot be deduced from these methods alone and they are most effectively used together with structure based guidelines. With the advances of computer graphics technology and various protein modeling methods, computer modeling of protein structures has become an increasingly used alternative for obtaining a three-dimensional model st ucture. The resolved X-ray structures of the four bacterial is~zymes provide a particularly propitious opportunity f,~r building model structures of other cytochrome P450s. Tt ~ r~ suits of these studies can provide structure based guidel nes for selection of mutations that could affect enzyme ft~action and substrate and product specificity for further experimental study. R e results of these experimental studies can, in turn, used to further assess and refine the model structure. Such iterative interaction between computational structural biology and biochemical and molecular biology efforts currently provides the most promising pathway to obtaining reliable and useful 3D models of P450 isozymes. In this paper, a brief review of all of the reported computer modeling of cytochrome P450 structures is presented. A strategy is then described that has been developed recently in our laboratory for constructing and evaluating P450 model structures using the bacterial P450choP enzyme as the test system. Finally, we report the results of using these methods for the construction and assessment of a 3D model of the rabbit P450 2B4 isozyme, and use of this model to identify a substrate access channel, binding site and surface ~sidues important for reduction of the enzyme by cytochrome bS. Specific mutants have been suggested for further validation of this model. For consistency, the cytochrome P450s are identified by their corresponding gene symbol (for example, CYPI A 1) as recommended by Nelson et al [1]. Listed in table I are the reported computer modeling studies of various P450s using structural information from this isozyme. They can be divided into two groups: those that built a substrate binding site only, and those that attempted to build a complete model of the protein. Among the modeling studies using P450cam as a template for constructing the substrate binding region are the modeling of CYPI, 2B, 3, and 4 by Lewis and Moereels [ 14], CYP 2C2 by Ramarao and Kemper [15], 3A4 (P450 NF) by Ferenczy and Morris [16], CYP 17 and 21 by Linet al [17], and CYPI07A1 (P450eryF) by Andersen et al [18]. These models were generated simply by extracting the P450cam binding site residues from the crystal structure and substituting the coordinates of these residues with those of the modeled P450s based on sequence alignment results. The modeling of CYP 2D6 by Koymans et al[ 19] included a binding site region, helices B, B', C, D, F, G, I, J, K, L, and 1~3/134regions. In additi6n, Iwasaki et al [20, 21 ] have used the P450cam crystal structure directly as a model for CYP 2A4/2A5 and proposed possible steroidbinding orientations and key residues (209 and 481) in 2A4/2A5 that regulate the regioselectivity of their hydroxylase activity. The earlier modeling work performed by Lewis et al of CYP 1,2B, and 2E [22, 23], by Ishida et al of CYP 51 (P450 14DM) [24] and by GrahamLorence et al [25] of CYPI9 (aromatase) included only small part (the I helix region) of the substrate binding site. Full protein models of CYPIAI (by Zvelebil et al [26l), 2Bl (by Szklarz et al [27]), 5 (TXA synthase, by Ruan et al [28]), 1IA 1 (scc, by Vijayakumar and Salerno [29]), 17ct (by Laughton [30]), 19 (aromatase, by Laughton etal [31]), 51 (14DM from Saccharomyces cerevisiae, by Morris and

Table I. Computer modeling of cytochrome P450s using the X-ray structure of P450cam as a template.
Cytochrome P450 Model Reference

CYP! CYP I A I CYP2A4/2A5 CYP2B CYP2B 1 CYP2C2 CYP2D6 CYP2E CYP3 CYP3A4 (nf) CYP4 CYP5 (TXA synthase) CYPI IAI (scc) CYPI 7 (17~) CYP19 (arom) CYP21 CYP51 (14DM) CYP51 (CA) CYPI05AI/IO5B ! (SUI/SU2) CYPI07A 1 (eryF)

Materials and methods

Review of previous computer models of P450s

Most of the previously reported attempt to build models of P450s use only the X-ray crystal structure of P450cam, the first known structure of P450s. The X-ray structure of this isozyme re~ealed all overall topology composed of a helix-rich domain and a helix-poor domain containing mainly anti-parallel It-sheets. The heine prosthetic group is embedded between the I and L helices, The substrate binding site on the distal side of the prosthetic group, opposite the cysteine ligand, was found to be surrounded mainly by hydrophobic residues in the B' helix, F-G helices, I helix, a 13sheet region after the K helix, and a 13hairpin region near the C-terminus.

Active site, I helix only Active site Full protein No model Active site, I helix only Active site Full protein Active site Partial protein with active site Active site, I helix only Active site Active site Active site Full protein Full protein Full protein Active site Full protein Active site, I helix only Active site Active site, I helix only Full protein Full protein Full protein Active site

22 14 26 20, 21 22 14 27 15 19 23 14 !6 14 28 29 30 17 31 25 17 24 32 33 34 18

773 Richards 1132]), 51 (Candida albicans, by Boscott and Grant [331), 105A 1/105BI (by Braatz et al [34]) have also been reported. The models based on the structure of P450cam are not reliable [35] for a number of fundamental reasons: i) all of the modeled P450s have very low sequence identity (-15 to 20% for the mammalian P450s, and < 25% for the bacterial ones) with the P450cam sequence. Thus an incorrect sequence alignment between P450cam and the target was obtained with the automated sequence alignment techniques used in all of these studies, especially in the N terminus half of the P450s. Without a reliable sequence alignment, it is not possible to derive structural information for a target protein, from that of a protein with known structure used as a template; and ii) the assumption made in these studies that the binding site architecture of P450cam is conserved and can be used to represent other P450s is not valid. In fact, the binding site of each P450 is in the most variable region of the protein. Direct experimental evidence for this variability has been obtained from the crystal structures of three additional bacterial P450s; P450terp, P450BM-3 and P450eryE Alignment of these four isozymes based on structural information clearly indicate both sequence and spatial variability for the residues in the substrate binding region. These comparisons therefore provide direct evidence that "~model structure bas,zd solely on P450cam is not reliable. The use of P450BM-3 as a template should be a better alternative for modeling the mammalian P450s since this isozyme, like the mammalian ones, is a class II P450 isozyme which utilizes FAD and FMN containing reductases as redox partners. In contrast, the bacterial P450cam, terp, and eryF isozymes are class I enzymes whose redox partners are FAD containing reductase and iron-sulfur proteins. In addition, compared to the other three bacterial P450s, P450BM-3 is known to have higher sequence homology with many mammalian P450 isozymes. Recently, as listed in table II, Ruan et al [28] used ~f-'450cam and P450BM-3 separately to model CYP5, and concluded that the model based on P450BM-3 is more reliable. Lewis [361 used the BM-3 structure as a template to build full protein models of 'CYP 1A I, 1A2, 2A6, 2B6, 2C9, 2D6, 2E I, 3A4, 4A4, 4A i 1, and 19. in that work, an alignment between BM-3 and 2A4 was used as the starting point to obtain a multiple sequence alignment of BM-3 with the target sequences of interest by taking into accouni several additional factors such as secondary" structure information, results from site-directed mutagenesis studies, known substrate specificity, and specific antibody binding sites. A significant improvement m the methodology for construction of 3D models of P450 isozymes was made possible by the observation of highly conserved structural regions among the three known structures, P450cam, terp, and BM-3 by Hasemann et al [37] and between P450cam and P450eryF by Cupp-Vickery and Poulos [7]. These comparisons have shown that the overall fold and secondary structure content among the four isozymes are similar despite their low sequence homology. There are now three reports of construction of 3D models of P450s using three templates: P450cam, P450terp, and P450BM-3. These include the work: 1) by Graham-Lorence et al [38] of a model structure of CYP 19; 2) by Szklarz et ai [39] of an improved model ofCYP 2B 1; and 3) by Modi et al [40] of CYP 2D6. All three investigations had as a common step the generation of a multiple sequence alignment of these three templates with their target isozyme relying on structurally conserved regions of the templates. They differed in strategies used to remodify this alignment. In the work by Graham-Lorence et ai and by Szklarz et al, an alignment of the target sequence with those of cam, terp, and BM-3 was optimized using data from site-directed mutagenesis. The work by Modi et al [40] used NMR studies of substrate (codeine) binding. In that work, NMR measurement of the distances between the heine iron and protons of a bound substrate (codeine) provided additional information for improving the accuracy of the binding site region. Many of the reported homology modeling efforts of the P450s assumed that by developing an energy minimized structure, one had a model of sufficient accuracy to proceed with substrate/inhibitor docking studies. Since, both substrate and product specificity are dynamic properties, it is important that the model structures are hydrated with bound waters and are stable to unconstrained full protein molecular dynamics simulations. However, of all of the P450 models reported, only P450SU I/SU2 by Braatz etal [34] and 2B 1 by Szklarz et ai [27, 39] have water molecules added, and none were subjected to completely unconstrained MD simulations. In addition to these limitations in the construction of the models,, very little effort has been devoted to the examination of the overall quality of the models before using them for the problem of interest. Among all model P450s reported, only CYP105A I/105B I 134] was examined with QPACK [411 for its packing quality, 2B ! [27,391 and CYPI9 [381 were examined with Profiles-3D [421 for the residue local environment and compatibility between sequence and 3D profile, and CYPI9 [38] was examined with Procheck [431 for its residue backbone conformations. Recently, we have developed and carefully evaluated improved strategies for constructing reliable 3D models of P450s [44] using all four known structures including P450 cam, terp, BM-3 and eryF as templates and P450choP as the test system. The evaluation included systematic assessment of the model by comparisons with known structures in the protein data bank (PDB), and the requirement that it be stable to completely unconstrained MD simulations. We have used these protocols, described in the next section, to construct and assess a model for the rabbit microsomai P450 2B4 isozyme.
Protocols developed for the construction and evaluation of 3D structures of mammalian P450 isozymes

Table III. Computer modeling of full protein structures of cytochrome P450s using P4501?,M-3 and/or the other bacterial P450s with known X-ray structures as template(s).
Cytochrome P450 Template(s) Reference

CYP1AI/IA2 CYP2A6 CYP2B ! CYP2B4 CYP2B6 CYP2C9 CYP2D6 CYP2E1 C~/P3A4 CYP4A4 CYP4A 11 CYP5 (TXA synthase) CYPI 9 (arom) CYPI05CI (choP)

BM-3 BM-3 cam/terp/BM-3 caln/terp/BM-3/eryF BM-3 BM-3 BM-3 cam/terp/BM-3 BM-3 BM-3 BM-3 BM-3 cam or BM-3 cam/terp/BM-3 BM-3 cam/terp/BM-3/eryF

36 36 39 (this work) 36 36 36 40 36 36 36 36 28 38 36 41

The first and most important step in building structures for the P450s is to obtain an accurate sequence alignment of the target with tile template proteins. In particular, in developing our current

774

strategies [44], we have clearly demonstrated that sequence alignments using automated alignment procedures for the P450s are not reliable. This assessment was based on the observation of large errors in the sequence alignment of the tbur isozymes with known structures (ie cam, terp, BM-3 and eryF), obtained from automated procedures compared to the alignment obtained using structural info~ation. Therefore, the procedure chosen to generate a sequence alignment among the template proteins (cam, terp, BM-3 and eryF) is based solely on the 3D structurally conserved regions (indicated with boxes in table III). This approach has also been used in other recent homology modeling work by Graham-Lorence et al

Table IlL Multiple sequence alignment between P4502B4 and the four templates. Structurally conserved regions are indicated with boxes.
Call tm erjP Im-] 2~4 cam tern eryr 2N NLAPLI)~V11~v~YHPS NZU~U4TZ PlU4X/d~'V Zl, : RI'TVPD : TgRI~PQPIrI~GEQ~II, P : I~1~* - - S- - - ~-~-~--JUPLAGLLLLLIq~3N P8JOIGRLPiq;PS PLIW1.GNLL : 32 17 6 18 44

: m.s~w~m,~.mv-

- p[~'m~

b1-1

c~ - .

bl-2

!1 h e l i x

~e
6~ ~5 65

I ~ t ~ D g V Z Y P M ' 8 z r , JU~Q I Zq.MOm | Zi ~ t n P ~ I~Z ~ I VaU~VZ~Z~K01P : AZZJU~A I ZW'~Z'W I rZ,~Q-Z~ Im, V~ I m r m ~ , J ~ . s z ) I z, zaar,~w0~u.mzz~m~L--O ! z z r m ~ I A ~ - V ~ I~VZ.S I S ~ U . Z m C ~ I S I~t-5 il' helix IU~qlg .............. plr | PlUUtGIL4y. . . . . . . . . . . I~ZPTS~4 : G/,JPDI~. . . . . . . . . . . . H 1 LYI~NNL4.wNRSI SOGCIPI4',/IDslr.TSM : -UDiqKN[YFGVI~VI~P&Z~FEDVNffr ........... A111tIGTS * ~ ............... 0,M,KP'VN)PA . . . . . . . . . . GDGI,7'I'SW z MUUP~GR . . . . . . . . . . . . . . . Gg'XAVVDPZ . . . . . . . . . . I~GYGVIF helix C* h e l i x DPPI~-I~P ~ D helix -- I/a~m~;~u;i~ I SLI~P9 . . . . ~

te~ ew~ De-~ 2~

103 105 95 90 115 146

te~l~ ery? 3B4

. ~

.... [L

z61

terp ery~ DN-3 384

b$-1 R h e l i x B h e l i x I~ h e l i x " ~ t ~ 11[~hJk][IP~ Ol,Pl[l[ . . . . . . DZPNIA(Y~TD~I4TRPD.I~lP~ JN'rI~ALY ] Y F i n A l . i ~PlLt) . . . . . . DgPIJ4LKLTODFJPGVHgP z VVDI I V D U N I P [ L P Z I N I e g L I . I 0 . A ~ K . . . . . . 8~UGn~SSI~Zl.V~D.t H| I~V [ ;PIQ~-59R i I,I'IATF l O l e o Jiq4YIUGISpyp,~pHPF',r T S I e V S ~ D ~ z ~ZAUd I~TL~PH ~ V ~ G K R I z'. . . . DYKDPVIPIJU,A~/,APF~SFS

lee 193 Z~e 196 ~O?

O helix z ........................... O~~YDYZ.ZPZ Zg98 z I~O~WdkP . . . . . . . . ~D.... WH~rZ~,t~ptIGm~ "" ""-- .................. P ~ 8 0 0 ~ Z LDI.~lZRR , ,' gI,QllU~HPD-. . . . . . . DP&Y . . . . . ~ R Q P Q r ~ I ~ I , Q1DXIIAI[~ =114 LISSP$1K~IPI[I, . . . . PP~I,lq)flr PGTHR0%YIU41,.QEIIf?IeIGQSV~H ~e~p ery~

211 a3o 203 223 252

te=p ery~ 2H

t~x--~r--m,-n-~ - -

I ~ s x v A . Ir.Ov--.~.P---xT ~
I SgL,-m~'--

II h e l i x

hS.l

b5-2

2~helix

2sl
=?o

' Uc"" PXD'-- I ~

-ZD I~YZ~U~YYVAZATJ~;HD ~ I~ IT[~

I~Z8 IVO~mx;n. --za I~u.Tsz~.v~u~o~z Ima.qmm.q IC~npm~zp-- -up I ~ z a v o z ZTrLZA~,Z

.
8p . . . . . . . . . . . . . . . . . .

244
267

~ T z a p . m p s ~ ~ s z ) p s s ~ . helix : I ' w ~ ~ S p l g 4 ~

3ol 283 2?6

ter~ ery~ liN-~ ~B4 ce~ ~erp e=yF 1~-3

: JUIVSLZGIQI~LLL? I Hiq~I.kLVRR ] DP. . . . . . . . . . . . . . . . . .

~ ~ z

zm.nl,pz~z.va~sl Dp. . . . . . . . . . . . . . . . . .

~lzPn

SA I I.PN

3o2

I Y z m v ' r z m ~ I f 1 F&VZGSHSPPALDDRA~PY ~ 8 helix bl~4 Io3-1 b2-2 b1-3 K' h e l i x

3 $1

~ ~
~

~ ~

[ e ' ~ IZ'nW~lmPmzol~-M ZP~VS'rvLvlk'~,~DP. I ~ - ~ irss,v ~ I surv~i varn'l ~.xa~.s~vJ,-z ~

~I q ~ L
mender

324
36~
400

GY'V ~

IYt,~eXLqDF B
L helix

elm te~ e~yr a~ c~ te=p eryF 284

b-bulge

l WlS-mlmqOml~ ........ m - - . ~ l m c , ~ l ~ z ~

3so
364
449

: ]fir ! li"?l~ll ~ J

I ~ ~ l q l p

Ll~l,(ll~ZCb I GI~ I k.q?ZI, lCLlr

L helix bZ.Z b4-1 h4-2 b3-2 : ~ D~8 ] A " i~AQIQIIR" SG~ ] P l ~ DPA- -~'l'l~V nm.LPm~ isv~a---oPPnz.v~'rmP',~t~ IvPz~Vl~w uu~am~el~Sz~z~ux)vv, m~u.u~az~ I~v~u.l~ M~MI~H@D I ie~:~I'--14NgLI)ZS-gTLTL~iq~G [ ~ ] KSKSZ PLGG I~HFS ! X~St~IPi~DID~.TF~p~IVI.~

414 ~a 4o~ 4 $? 491

[38], Szklarz et ai [391 and Mode et al [40]. In err previously reported methods, developed using P450choP, alignment of the target protein with the template proteins was obtained tl,sing its predicted secondary structure together with a pairwi~;e sequence alignment with one of the templates, 1:'450 eryE to guide the alignment. For the modeling of P450 2B4, because of the low sequence similarity between 2B4 and the template sequences (~ 18% identity with all four templates), and because of the existence of other sequences highly homologous to 2B4, a modified strategy for aligning the target sequence with the templates was developed. In this improved strategy, we have profited from the fact that P450 2B4 has a high sequence identity of >75% with other 2B sequences, and > 45% identity with sequences in different 1)450 2 subfamilies. Therefore, the alignment of 2B4 with its closely related proteins in the 1:'450 2 family can be unambiguously determined by automated multiple sequence alignment techniques. Some of these related proteins exhibit a higher sequence similarity with the template proteins than does P450 2B4. Therefore, a more confident alignment could be performed between those related P450s and the template proteins than between 2B4 and the templates. Once a related protein was aligned to the template proteins, P450 2B4 could also be aligned to the templates because its alignment to the related protein had already been determined. Upon obtaining a multiple sequence alignment, the remainder of the procedures used in producing an initial model structure were as described in detail in our previous paper [44]. Briefly, in the next step, the backbone coordinates in the structurally conserved regions (SCRs) of the target protein were generated by copying those from the templates. Then, structurally uncon~erved regions of the same length as the template were generated by copying the structures of a template and for those of variable length by a loop search procedure. Generation of initial side chain conformations, another very challenging aspect of model building, was performed using a simulated annealing/Monte Carlo method called Torso [45]. The reliability of this procedure to predict correct side chain conformations was assessed in our previous study using the four template structures and found to predict more than 80% of Z I angles correctly to within 40 degrees [44]. Energy minimization using Amber 4, I of the structurally unconservod regions and the side chains were then performed by fixing the backbone of the SCRs producing an initial full protein model. We have profited from the recent development of various protein structure analysis programs to evaluate the quality of an initial model built using any alignment procedure. We have chosen to use the Prosa program developed by Sippl [46] for this evaluation. Specifically, using the Prosa program, the interaction energy of each residue with the rest of the protein was calculated, based on a mean-field potential derived from known protein structures in the protein data bank. A large repulsive energy peak indicates a possible region of incorrect sequence alignment or bad steric contact. In our initial 2B4 model, four regions including the B' helix, F-G helices, K helix, and the C terminus regions, all of them located near substrate binding regions were found to have repulsive energies, Therefore, an adjustment of the sequence alignment and/or the local structures in these regions was then performed in order to reduce the repulsive energy peaks; leading to an improved alignment and an improved model. After this improvement, the repulsive energy in these four regions either disappeared completely or was reduced significantly. In our previous work of modeling P450choP, we have found that un~constrained full protein minimization showed a small degree of protein contraction. These results pointed to the importance of addin~g structural waters to the model. The presence of bound waters in the binding site region is especially important in a substrate-free

775 model for maintaining the integrity of the binding site structure. As mentioned earlier, addition of bound waters to model P450 structures has been generally overlooked. Of all of the P450 models reported, only P450SUI/SU2 by Braatz et al [34] and 2BI by Szklarz et al [27, 39] have water molecules added. In these studies, however, water molecules were added by either copying the structural waters from P450cam to the target protein or by simply soaking the protein with waters. In our work, structural waters were added systematically using an in house program called HYDRAS. This program examines the local H-bonding network and steric interaction to assign the positions of the structural waters. UnconstrL-ained full protein energy minimization including the ferryl heme unit and structural waters using Amber 4.1 was then performed. The force field parameters including the point charges for this ferryl heme was developed previously in this laboratory [47]. This optimized, hydrated full protein structure was then subjected to a series of further tests for reliability and stability. As mentioned earlier, in past P450 homology modeling efforts, only a few model structures have been examined with the goal of assessing the overall quality of the model structure. In our work, a set of five rigorous tests were used to evaluate the reliability and stability of the resulting model structure of P450choP and P450 2B4. These included: i) comparisons of their backbone conformations with the allowed regions in known protein structures in the protein data bank using Quanta/Protein Health [48]; ii) calculation of the residue interactions with the remaining protein using Prosa [46]: iii) calculation of the residue local environment and comparing it with the environment in proteins of known structure in the protein data bank with Profiles-3D [42]; iv) examination of the residue packing with Whatif Quality Control [49]; and (v) determination of the overall protein stability with more than 100 ps of molecular dynamics simulations in which neither the heme unit, the structural waters nor the residues is constrained. A detailed description of the principle of these methods and their recommended criteria for structural evaluation can be found in the original references and in our previous work 1441. After these evaluations, the model 2B4 structure was then used to identify and characterize the putalive substrate binding site, substrate access channel, and surface regions |br interaction with redox partnets. The identification of the substrate binding site was made by determining the location of water clusters above the heme unit. The substrate access channel was identified by the location of a continuous aggregation of waters from this distal side of the heine to the surface. The method used to find surface regions tbr binding redox partners was based on a geometric fit algorithm for protein-protein docking [50, 51]. No physico-chemical properties were considered in this protein-protein docking process.

Fig 1. Ribbon representation of crystal structure of P450BM-3 and model structure of P450 2B4 in a familiar view of P450s. Structurally conserved o~ helices, [3 sheets, 13 bulge and meander regions are colored magenta, yellow, red and green, respectively. Structurally unconserved regions ate colored white. Heine unit is in red with vdw surface shown.

Results
The result of the sequence alignment between P450 2B4 and the four templates is shown in table III. The nomenclature of the secondary structures used in this table is adopted from the work by Hasemann et a / [ 3 7 ] . A ribbon representation of the unconstrained energy minimized full protein model of 2B4 is shown side by side with one of the templates, BM-3 in figure 1. We see from this figure that there is a large conservation of 3D structural motifs. The results of the evaluation of the unconstrained energy minimized (full

protein + heme + water) model of 2B4 are as foilows. Using Quanta/Protein Health, the backboi-~e ~/~ angles o f - 9 3 % of the residues in the 2B4 model were found to be in allowed region, compared to 98 to 99% for the four templates. Using Prosa, the normalized z score over sequence length was found to be 0.65 for this model, compared to 0.97 to I. 12 for the four templates. The recommended criterion using Prosa of a good model is its normalized z score >0.7. Tile current model is then not significantly outside the recommended score. Using Profiles-3D, the normalized s-score was found to be 0.79 compared to 1.00 to 1.12 for the four templates, and it is well above the 0.45 value recommended for a reasonable model. The Whatif quality control value of this 2B4 model was found to be -1.7, compared to -0.5 to -0.75 for the four templates. It was recommended that if the Whatif score is below -3, the ~nodel is definitely of low quality. If the Whatif value is above -1, it is recommended as a good structure. The quality of a model with a Whatif value between -1 and - 2 is more difficult to assess. In order to aid in this assessment, we have found three crystal structures (pdb155c, pdblabp, and pdblprh) in the PDB databank with resolution of 2.4, 2.5 and 3.5/~ whose Whatif quality value are - 2 . 2 5 , - 1 . 8 6 and -1.36, respectively. Taken together, the current model of P450 2B4, though not as good as high resolution crystal structures, is of comparable quality to these structures. Moreover, assessment of the stability of the model by examining the total energy, the potential energy and radius of gyration during 140 ps of

776 fully unconstrained molecular dynamics simulations, indicates that the model equilibrates and remains stable after about 80 ps of dynamics simulation and has a constant radius of gyration during the entire simulation. Taken together, these assessments provide a reliable and stable model suitable for characterization of explicit interactions with substrates and redox partners. The finding of a cluster of waters above the heme unit allowed the identification of the residues that enclose the substrate binding site. These include F95, R98, V 103, I 107, Y111, G112 and V113 in the B' helix region, F202 and F206 in the F helix, $294, F297, A298, T302 in the I helix, I363, F365, G366, V367, P368 between the K helix and 13!-4, and V477 in the 134 hairpin region (the nomenclature of the 13 sheets by Hasemann et al [37] was adopted here). Consistent with the finding among the known proteins themselves, these residues are largely in the variable region of the protein. Examination of the binding site topology revealed that the space above rings C and D and the two propionate groups of the heine unit is more open than that above rings A and B. The substrate binding site in the model structure is not completely buried. Examination of the continuously aggregation of water molecules from the binding site to the surface suggests a possible substrate access channel. The upper region (ie, region that is closer to the surface) of this channel is narrower than the middle and lower regions since the water population in this upper region is smaller. The residues that form this channel are in the A helix, F helix, 131 and 134 regions. Among all of the known substrates of CYP2B4, benzphetamine and androstenedione show the most pronounced product specificity. The preferred N-demethylation observed of 2B4 sugge~t~ ~hat the N-CH3 group of benzphetamine has to be closest to the ferryl oxygen. For androstenedione, P450 2B4 catalyzes 161]hydroxylation, with very little 16ot hydroxylation, suggesting that the 16J] H atom should be closest to the ferryl oxygen, Therefore, as a further assessment of the binding site identified, we determined whether benzphetamine and androstenedione could be docked in the substrate binding site in orientations that would lead to these observed products. Two possible orientations of the lowest energy conformer of benzphetamine in the 2B4 binding site were found that could lead to N-demethylation. These two 2B4-benzphetamine complexes were energy minimized with AMBER4. I. As shown in figure 2a, b), the N-methyl group in both orientations is closer to the ferryl oxygen than any other paint of benzphetamine, implying Ndemethylation. The distance between the ferryl oxygen and the N atom of benzphetamine r (O-N) in the first orientation is 4.4 A and that between ferryl oxygen and the N-methyl carbon atom r (O-C)is abou~ 2.9/~,. The r (O-N) and r (O-C) distances in the s.~cond orientation are 5.0 and 4.0 A, respectively. Two possible orientations of androstenedione were also found that could lead to 1613hydroxylation. The two 2B4-androstenedione complexes were then energy optimized with AMBER 4.1. Figure 2c, d shows the resulting optimized orientations of androstenedione. In both orientations, the C 16 [3 hydrogen is closer to the ferryl oxygen than any other atoms in androstenedione. These two androstenedione orientations differ by an overall rotation such that the two methyl groups in androstenedione point in different directions in these two orientations. In the first orientation, the 161] hydrogen to ferryl oxygen distance (r (HI]-O)) is about 2.9 A, shorter than the 4.0 A distance between 16ot and the ferrvl oxygen (r ,(Hot-O)). In the second orientation, these distances are 2.1 A (r (H1]-O)) and 3.2/~ (r (Hot-O)). The results imply that 161] hydroxylation is favored over 16ot hydroxylation, in agreement with experimental data. The ability of these optimized complexes to account for the observed product selectivity of benzphetamine and androstenedione provides additional support for the reliability of the binding site and the full protein model that led to its identification. The cytochromes P450 require the input of two electrons to complete the enzymatic cycle and to transform substrates into products. Among the three classes of P450 redox partners, only cytochrome b5 has an available crystal structure from bovine liver. Therefore, this cytochrome b5 structure (pdb l cyo.ent from PDB databank) was used here, together with our model P450 2B4 structure, to find possible regions of interactions between these redox partners. Using a geometric fit method [50, 51 ], the 100 lowest energy positions of the smaller cytochrome b5 protein relative to the model 2B4 structure gener'~ted from this program were analyzed and three distinct cius ers of b5 positions were found. These three clusters were then used to identify three surface regions on the model 2B~ structure that could possibly interact with cytochrome b5. The first surface region is located on the distal face of the model 2B4 and it includes residues in segments of the A helix, E* helix, F helix, I helix and in the ~4 region. The second surface region is located t,n a face perpendicular to the heme plane of 2B4, and it includes residues in the B' helix region of 2B4. The third surface region is located on the proximal face of 2B4. The third face found in our study was also suggested to be the b5 binding site by Stayton et al [52] from manual docking experiments between the crystal structures of P450cam and cytochrome b5. From our study, the surface residues of 2B4 in this face that may contact cytochrome b5 include residues in the CC* helices, I]5, K helix, meander, 1]-bulge, and L helix. Mutation of some of the surface residues to alanine in these three regions are currently in progress in the laboratory of our collaborator, Dr Lucy Waskell.
Conclusion

This work reviewed all of the previously reported computer modeling studies of cytochrome P450s and described a strategy used to construct and evaluate P450 model structure rei:ently developed in our laboratory to address the

777

(a)
R9 ~5

V113

(b) R ~ 9 ~~/""'3
1107

V113

"107 V367 ;294 S294 F365

F3

-(

(c)

F95

v1

Id)

r95

v113

II07 V367)~ $294 F365 V367

II07

$294

Fig 2. Binding orientations of benzphetamine and androstenedione in the model 2B4 binding site viewi,~g I','onl top of the heine unit. The structure of the substrate is represented with thick lines. The ferryi oxyge, of the heine is labeled, a, b. Two different orientations of benzphetamine, e-d. Two different orientations of androstenedione. for the reliability of the model. Using a geometric fit method to dock the model 2B4 structure with the known cytochrome b5 crystal structure, three regions on the surface of the model 2B4 structure were suggested to be possible sites for interaction with cytochrome b5. Residues that may interact with cytochrome b5 have been identified, and mutagenesis studies changing some of these residues to alanine are currently in progress.

major weaknesses of these previous studies. An application of this strategy to the modeling of the rabbit P450 2B4 isozyme was also presented. The model structure of 2B4 obtained was evaluated for its overall quality, using available protein analysis programs and found to be satisfactory. It was also found to be stable to unconstrained full protein molecular dynamics simulations. The 3D model of 2B4 was then used to identify and characterize three important regions related to its function; a substrate access channel, substrate binding site and candidate surface contact regions with a redox partner, cytochrome b5. When two different substrates, benzphetamine and androstenedione, that are metabolized by P450 2B4 with pronounced product specificity, were docked into the putative binding site, two orientations were found for each substrate that could lead to the observed preferred products. These results provide evidence

Acknowledgments We wish to thank Dr llya Vakser of the Rockefeller University for using his program to identify the candidate P450 2B4 surface regions that interact with cytochome b5. We also gratefully acknowledge financial support from NIH grants GM35533 and GM27943.

778 References
I Nelson DR, Kamataki T, Waxman DJ, Guep.gerich FE Estabrook RW, Feyereisen R, Gonzalez FJ, Coon MJ, Gunsalus IC, Gotoh O, Okuda K, Nebert DW (19931 The P450 superfamily:update on new sequences, gene mapping, accession numbers, early trivial names of enzymes, and nomenclature. DNA Cell Biol 12, 1-51 2 Poulos TL, Finzel BC, Howard AJ (1986) Crystal structure of substratefree Pseudomonas putida cytochrome P450.J Am Chem Soc 25, 53145322 3 Boddupalli SS, Hasemann CA, Raviehandran KG, Lu J-Y, Goldsmith EJ, Deisenhofer J, Peterson JA (19921 Crystallization and preliminary X-ray diffraction analysis of P450terp and the hemoprotein domain of P450BM-3, enzymes belonging to two distinct classes of the cytochrome P450 supeffamily. Proc Nati Acad Sci USA 89, 5567-5571 4 Hasemann CA, Ravichandran KG, Peterson JA, Deisenhofer J ( 19941 Crystal structure and refinement of cytochrome P450terp at 2.3 A, resolution. J Moi Bio1236, 1169-1185 5 Ravichandran KG, Boddupalli SS, Hasemann CA, Peterson JA, Deisenhofer J (19931 Crystal structure of hemoprotei,! domain of P450BM-3, a prototype for microsomal P450's. Sciem'e 261, 731-736 6 Li H, Poulos TL (1995) Modeling protein-substrate interactions in the heme domain of cytochrome P450BM-3.Acta Crystallogr D 51, 2 !-32 7 Cupp-Vickery JR, Poulos TL (1995) Structure of cytochrome P450eryF involved in erythromycin biosynthesis. Sn'uct Biol 2, 144-153 8 lwa.,nato ~: Tsubaki M, Hiwatashi A, Ichikawa Y (1988) Crystallization of cytochrome P450scc from bovine adrenocortical mitochondria.FEBS Lett 2 I0, 31-36 9 Raag R, Swanson BA, Poulos TL, Ortiz de Montellano P (1990) Formation, crystal structure, and rearrangement of a cytochrome P450cam Iron-phenyi complex. Biochemistry 29, 8119-8126 10 Swanson BA, Dutton DR, Lunetta JM, Ortiz de Montellano PR (1990) The active sites of cytochromes P450 ! A I, liB 1, lIB2, and lie I: topological analysis by in situ rearrangement of phenyl-iron complexes. Bi,Jchemistry 266, i 9258-19264 II Swanson BA, Bornheim LM, Halpert JR, Ortiz de Montellano PR (19911 Topological analysis of the active sites of cytochromes P4501IB4 (rabbit), P45011B ! 0 (mouse), and P45011B I i (dog) by in situ rearrangement of phenyl-ir(m complexes. Arch l?iochem Biophys 292, 42.-46 12 'luck SF, Peterson JA, Ortiz de Monteilano PR (19921 Active site topologies of bacterial cytochromes P450101 (p450cam), P450108 (P450tcrp), and P450102 (P450BM-31. J Biol Chem 267, 5614-5620 13 Tuck SE Ortiz de MonteUano (1992) '|bpological mapping of the active sites of cytochromes P4502B ! and P4502B2 by in situ rearrangement of aryl-iron complexes. Biochemistry 31, 691 i-69 i 6 14 Lewis DFV, Moereels H (1992) The sequence homologies of cytochrome P-450 and active site geometries. J Comput-Aided Mol Design 6, 235-252 15 Ramarao M, Kemper B (1995) Substitution at residue 473 confers progesterone 21-hydroxylase activitiy to cytochrome P450 2C2, Mol Pharmaco1484, 4 ! 7--424 16 Ferenczy GG, Morris GM (19891 The active site of cytochrome P-450 nifcdipine oxidase: a model-building study. J Mol Graphics 7,206-2 ! 1 17 Lin D, Zhang LH, Chiao E, Miller WL (19941 Modeling and mutagenesis of the active site of human P450c 17. Mol Endocrinoi 8, 392-402 18 Andersen JF, Tatsuta K, Gunji H, Ishiyama 1, Hutchinson CR (19931 Substrate specificity of 6-deoxyerythronolide B hydroxylase, a bacterial cytochrome P450 of erythromycin A biosynthesis. Biochemistry 32, 1905-1913 19 Koymans LMH, Vermeulen NPE, Baarslag A, den Kelder GMD (19931 A preliminary 3D model for cytochrome P450 2D6 constructed by homology model building. J Comput-Aided Mol Design 7, 281-289 20 lwasaki M, Darden TA, Pedersen LG, Davis DG, Juvonen RO, Sueyoshi T, Negishi M (19931 Engineering mouse P450coh to a novel corticosterone 150t-hydroxylase and modeling steroid-binding orientation in the substrate pocket. J Biol Cl~em 268, 759-762 21 Iwasaki M, Darden TA, Pedersen LG, Davis DG, Negishi M (19951 Altering the regiospecificity of androstenedione hydroxylase activity in P450s 2a-4/5 by a mutation of the residue at position 48 I. Biochem. istry 34, 5054-5059 22 Lewis DFV, loannides C, Parke DV ( 19871 Structural requirements for substrates of cytochromes P-450 and P-448. Chem Biol Interactions 64, 39--60 23 Lewis DFV (19871 Quantitative structure-activity relation.;hips in a series of alcohols exhibiting inhibition of cytochrome P450-mediated aniline hydroxylation. Chem Bio! Interactions 62, 271-280 24 Ishida N, Aoyama Y, Hatanaka R, Oyama Y, lmajo S, lshiguro M, Oshima I", Nakazato H, Noguchi T, Maitra US, Mohan VP, Sprinson DB, Yoshida Y (19881 A single amino acid substitution converts cytochrome P45014DM to an inactive form, cytochrome P450SGI: complete primary structures deduced from cloned DNAs. Biochem Biophys Res Commun 155, 317-323 25 Graham-Lorence S, Khalil MW, Lorence MC, Mendelson CR, Simpson ER (19911 Structure-function relationships of human aromatase cytochrome P450 using molecular modeling and site-directed mutagenesis. J Biol Chem 266, ! 1939-! 1946 26 Zvelebil MJJM, Wolf CR, Sternberg MJE (19911 A predicted threedimensional structure of human cytochrome P450: implications for substrate specificity. Protein Eng 4, 271-282 27 Szklarz GD, Ornstein RL, Halpert JR (19941 Application of 3-dimensional homology modeling of cytochrome P450 2B 1 for interpretation of site-directed mutagenesis results. J Biomol Struct Dyn 12, 61-78 28 Ruan KH, Milfeld K, Kulmacz RJ, Wu KK (1994) Comparison of the construction of a 3-D model for human thromboxane synthase using P450cam and BM-3 as templates: implications for the substrate binding pocket. Protein Eng 7, ! 345-1351 29 Vijayakumar S, Salerno JC (1992) Molecular modeling of the 3D structure of cytochrome P-450scc. Biochim Biophys Acta 1160, 281-286 30 Laughton CA, Neidle S, Zvelebil MJJM, Sternberg MJE (19901 A molecular model for the enzyme cytochrome P450 17a, a major target for the chemotherapy of prostatic cancer. Biochem Biophys Res Commun 171, ! 160-1174 31 Laughton CA, Zvelebil MJJM and Neidle S ( 19931 A detailed molecular model for human aromatase. J Steroid Biochem Mol Bio144, 399-407 32 Morris GM, Richards WG (19911 Molecular modeling of the sterol C- 14 demethylase of Sacchatomyces cerevisiae. Biochem Soc Trans 19, 793-795 33 Boscott PE, Grant GH ( 19941 Modeling cytochrome P450 14a demethylase (Candida albicans) from P450cam. J Moi Graphic 12, 185-192 34 Braatz JA, Bass MB, Ornstein RL (19941 An evaluation of molecular models of the cytochro.ne P450 Streptomvces griseolus enzymes P450SU ! and P450SU2. J Comput.Aided Mol Des 8, 607-622 35 Poulos TL (1991 ) Modeling of mammalian P450s on basis of P450cam X-ray structure. Methods Enzymo1206, i 1-30 36 Lewis DFV (19951 Three-dimensional models of human and other mammalian microsomal P450s constructed from an alignment with P450102 (P450BM3). Xenobiotica 25,333-366 37 Hasemann CA, Kurumbail RG, Boddupalli SS, Peterson JA, Deisenhofer J (19951 Structure and function of cytochrome P450: a comparative analysis of three crystal structures. Structure 3, 41--62 38 Graham-Lorence S, Amarneh B, White RE, Peterson JA, Simpson ER (19951 A three-dimensional model of aromatase cytochrome P450. Protein Sci 4, 1065-1080 39 Szklarz GD, He YA, Halpert JR (1995) Site-directed mutagenesis as a tool for molecular modeling of cytochrome P450 2B I. Biochemistry 34, 14312-14322 40 Modi S, Plaine MJ, Sutcliffe MJ, Lian LY, Primrose WU, Wolf CR, Roberts GCK ( 19961 A model lor human cytochrome P450 2D6 based on homology modeling and NMR studies of substrate binding. Biochemistry 35, 454,0--4550 4 ! Gregoret LM, Cohen FE ( 19901A novel method for the rapid evaluation of packing in protein structures. J Mol Biol 211,959-974 42 Luthy R, Bowie JU, Eisenberg D (19921 Assessment of protein modelswith three-dimer~sional profiles. Nature 356, 83-85 43 Laskowski RA, MacArthur MW, Moss DS, THronton JM (19931 Procheck: a program to check the sterochemical quality of protein structures. J Appi Crystallogr 26, 283-291

779
44 Chang YT, Loew GH {! 996) Construction and evalualion of a three-dhnensional structure of P450choP {CYPI05C! ) enzyme. Prowin Eng, in press 45 Holm L, Sander C (1991) Database algorithm for generating protein backbone and side-chain coordinates from a Ca trace. J Mol Biol 218, 183-194 46 Sippl MJ (1993) Boltzmann's principle, knowledge-based mean fields and protein folding. J Comput-Aided Mol Design 7, 473-501 47 Harris D, Loew GH (1995) Prediction of regiospecific hydroxylation of camphor analogs by cytochrome P450cam. J Am Chem Sot" i 17, 2738-2746 48 Quanta, version 4.0 (1994) Molecular Simulations Inc, Burlington, MA 49 Vriend G, Sander C {i993) Quality control of protein models: dtrectional atomic contact analysis. J Appl CpTst 26, 47-60 50 Katchalski-Katzir E, Shariv 1. Eisenstein M, Friesem AA, Allal(~ C. Vakser IA (1992) Molecular surface recogriition:determination of geometric fit between proteins and their ligands by correlation techniques. Proc Natl Acad Sci USA 89, 2 ! 95-2199 51 Vakser IA (1995) Protein docking for low-resolution structures. Protein Eng 8, 371-377 52 Stayton PS, Poulos TL, Sligar SG (1989) Association: a proposed molecular model for a cytochrome P450cam electron-transfer complex. Biochemistry 28, 8201-8205