Research Article

895

The structural basis of lipid interactions in lipovitellin, a soluble lipoprotein

TA Anderson1, DG Levitt2 and LJ Banaszak1*
Background: The conformation and assembly of lipoproteins, proteins containing large amounts of noncovalently bound lipid, is poorly understood. Lipoproteins present an unusual challenge as they often contain varying loads of lipid and are not readily crystallized. Lipovitellin is a large crystallizable oocyte protein of approximately 1300 residues that contains about 16% w/w lipid. Lipovitellin contains two large domains that appear to be conserved in both microsomal triglyceride transfer protein and apolipoprotein B-100. To gain insight into the conformation of a lipoprotein and the potential modes of binding of both neutral and phospholipid, the crystal structure of lamprey lipovitellin has been determined. Results: We report here the refined crystal structure of lipovitellin at 2.8 resolution. The structure contains 1129 amino acid residues located on five peptide chains, one 40-atom phosphatidylcholine, and one 13-atom hydrocarbon chain. The protein contains a funnel-shaped cavity formed primarily by two sheets and lined predominantly by hydrophobic residues. Conclusions: Using the crystal structure as a template, a model for the bound lipid is proposed. The lipid-binding cavity is formed primarily by a singlethickness -sheet structure which is stabilized by bound lipid. This cavity appears to be flexible, allowing lipid to be loaded or unloaded.
Addresses: 1Department of Biochemistry, University of Minnesota, 4-225 Millard Hall, 435 Delaware St SE, Minneapolis, MN 55455, USA and 2Department of Physiology, University of Minnesota, 4-225 Millard Hall, 435 Delaware St SE, Minneapolis, MN 55455, USA. *Corresponding author. E-mail: len_b@dccc.med.umn.edu Key words: apolipoprotein B, lipoprotein, lipovitellin, microsomal triglyceride transfer protein, X-ray crystallography Received: 14 April 1998 Revisions requested: 7 May 1998 Revisions received: 26 May 1998 Accepted: 28 May 1998 Structure 15 July 1998, 6:895909 http://biomednet.com/elecref/0969212600600895 Current Biology Ltd ISSN 0969-2126

Introduction
Soluble lipoproteins have a vital role in lipid transport and metabolism. Because they usually contain varying amounts of heterogeneous lipid, they are difficult to crystallize and have not been amenable to X-ray diffraction studies. Lipovitellin was the first lipoprotein to be crystallized with a substantial portion of associated heterogeneous lipid [1]. As the amino acid sequence of this protein was not known at the time of the preliminary X-ray studies, the electron density was interpreted in terms of an unrefined polyalanine model [2,3]. The subsequent determination of the cDNA sequence [4] has permitted the assignment of the amino acid sequence and refinement of the X-ray data. Lipovitellin is the major lipidprotein complex found in the yolk of egg-laying animals [5]. The protein can be loaded with varying amounts of noncovalently bound lipid, up to about 16% w/w lipid, two thirds of which is phospholipid [6,7]. Lipovitellin is formed from the precursor, vitellogenin, which is the product of a single gene coding for 1659 to 1852 amino acids. Vitellogenin is synthesized in the liver where it is loaded with lipid, phosphorylated and glycosylated, and where it interacts with at least two Ca2+ ions and one Zn2+ ion [8]. After secretion of the vitellogenin dimer into the blood stream and receptor-mediated endocytosis by oocytes, vitellogenin undergoes specific and limited cleavage into

several polypeptide chains; the lipid-binding product is now referred to as lipovitellin. The vertebrate lipovitellin is comprised of a heavy chain (LV1) of approximately 120 kDa and a light chain (LV2) of about 30 kDa [5]. In addition, the proteolytic processing of vertebrate vitellogenin releases a smaller fragment, called phosvitin, which ranges in molecular weight from 10 to 35 kDa, and which remains loosely associated with lipovitellin. Phosvitin has a high serine content, approaching 50% in some vertebrates; most of these serine residues are phosphorylated [5]. A detailed study of the proteolytic processing has not been carried out and it is possible that in some species, vitellogenin fragments are lost during this maturation process. As the amino acid sequence of the lamprey form was obtained from the vitellogenin cDNA, this post-translational proteolytic processing complicated the assignment of the derived amino acid sequence to the X-ray diffraction data, as will be described below. Raag et al. [2] obtained X-ray diffraction data to 2.8 from crystals of lipovitellin purified from lamprey eggs and derived a multiple isomorphic replacement electron-density map. In the absence of an amino acid sequence, it was only possible to build a preliminary polyalanine model of 984 residues into this initial map. The phases and electron density were slightly improved by Poliks [3] who added an additional 26 alanines. Because of weak or missing electron density, the chain tracing was interrupted at many

896

Structure 1998, Vol 6 No 7

Figure 1 Schematic representation of homology regions among vitellogenin (VTG), microsomal triglyceride transfer protein (MTP), apolipoprotein B (apoB) and pro-von Willebrand factor (pro-vWF). The homologous regions found between vitellogenin, lipovitellin, MTP and apoB are indicated in green. A small region of vitellogenin near its C terminus is homologous with a repeating domain in provWF and is shown in pink. The pro-vWF has four full copies of the C-terminal region of vitellogenin and one partial copy. The amino acid numbering above the schematics includes the leader sequence, shown in black, of all of the proteins. The numbering represents the first and last residues of the homology regions. The chain length of each protein is given at the C terminus.

17

688

1566

1823

VTG
22 676 894

MTP
28 703 4563

apoB
34 386 745784 865 1241 1947 2295 2813

pro-vWF

Structure

points, mostly on the surface, and the initial model consisted of about 30 separate segments. In addition, with the exception of the helical domain, the polarity or direction of the chain could not be determined. The lipid contained in the complex is bound in a noncovalent manner and is extractable with organic solvents. However, the lipovitellin used in this study, after precipitation and lyophilization, remains in the native state with bound lipid. Although chemical analysis indicated that the soluble protein contains about 16% w/w lipid [1], no electron density could be assigned to lipid in the earlier work of map interpretation and model building [2,3]. The identification of a large cavity bordered by two sheets as the major site of lipid binding was confirmed by low-resolution neutron diffraction with contrast variation [9]. In addition to its overall molecular organization, the lipovitellin structure has broad physiological significance because of its sequence homology with both apolipoprotein B-100 (apoB) and the lipid-carrying protein required for very low density lipoprotein (VLDL) formation, microsomal triglyceride transfer protein (MTP) [10,11]. There is a sequence of approximately 670 amino acids in the N-terminal region of the lipovitellin segment LV1 which is homologous to a similar-sized region in the N termini of MTP [11,12] and apoB [10]. This homology is summarized in Figure 1 and is marked in green. The evolutionary relationship between lipovitellin, MTP and apoB is further confirmed by the location of intron exon boundaries [12,13] and site-directed mutagenesis studies (C Shoulders et al., unpublished data). In addition, there is a cysteine-rich sequence of approximately 260 amino acids in the C-terminal region of lipovitellin (pink in Figure 1) that has four complete copies and one partially complete copy in pro-von Willebrand factor (pro-vWF; Figure 1) [14].

In the results described below, the amino acid sequence of lipovitellin derived from the cDNA has been placed in the electron-density map and the resulting coordinates have been refined. The resulting model still has disordered regions where chain continuity is lacking. Nonetheless, it is now possible to describe details of the conformation that were not apparent with the initial polyalanine model. On the basis of the earlier neutron diffraction study and the shape of the cavity found in the crystal structure, the nature of the amino acid sidechains that form the lipid cavity surface can be described. An attempt to analyze potential modes for the positioning of the lipid components has also been carried out. The present model structure contains 1129 amino acids refined to an R factor of 0.19 at 2.8 resolution.

Results and discussion

In order to use the cDNA sequence for interpreting the electron-density map of lipovitellin [4], the cleavage sites in the vitellogenin precursor had to be determined. Some were obtained by N-terminal analyses of purified peptides; others were estimated by counting residues based on molecular weights obtained with sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDSPAGE) results. The primary sequence for the vitellogenin precursor and how it relates to the X-ray study is given in Figure 2. The complete sequence of lamprey vitellogenin can also be found in GenBank with the accession code M88749. Most lipovitellins contain a large and a small polypeptide chain, the former being derived from the N-terminal end of the precursor. In the lamprey system the larger molecular weight chain, LV1, is cleaved into two segments, hereafter referred to as LV1n and LV1c. Thus, the crystal structure is formed from three separate polypeptide chains: LV1n, LV1c and LV2. Figure 2 tabulates the regions where amino acid structural assignments have been made in the crystallographic model and is meant to aid in understanding

Research Article Structure of crystalline lipovitellin Anderson, Levitt and Banaszak

897

Figure 2 Amino acid sequence of lamprey vitellogenin and the crystal structure components. The amino acid sequence of the lipovitellin precursor, vitellogenin, is shown. Residues not present in the X-ray model of lipovitellin are colored a medium gray unless otherwise noted. Residues are indicated as being present in one of the three lipovitellin chains by colored lines below the residue. In addition, a colored bar is located to the left of the N-terminal residue of each chain. As we lack experimental detail definitively defining the C-terminal residues, the presumed termini are not bound by lines. Chain LV1n is indicated with green lines beginning with Gln17, a pyroglutamate, and ending with Tyr688. Chain LV1c is indicated with red lines, beginning with Arg708 and with the assumed ending at Pro1074. Chain LV2 is indicated with blue lines beginning with Lys1306 and ending at Lys1624, as described in the text. In contrast to the underlining, the color of the one letter amino acid code indicates its location in the domain structure of LV as obtained from the crystallographic analysis. Colors correspond to the illustration in Figure 3b. Green characters indicate the residues that form the N-sheet domain, roughly Gln17 to Val296. Cyancolored characters mark those residues forming the large -helical domain, Thr297 to Ser614. Red is used to denote the amino acids forming the C-sheet domain, Lys615 to Tyr688 and Trp729 to Lys758. Dark blue letters mark the residues forming the large A-sheet domain, Gln778 to Lys948, Ser991 to Pro1074 and Pro1358 to Phe1529. The three potential N-glycosylation sites (Asn1097, Asn1298 and Asn1675) are indicated with yellow boxes. Between Ser1131 and Ser1284 there are five polyserine regions (18, 12, 14, 8 and 23 amino acids long) drawn in magenta; this region is within the putative phosvitin region of lamprey lipovitellin. (The figure was prepared using the program ALSCRIPT [52].)

1 51 101 151 201 251 301 351 401 451 501 551 601 651 701 751 801 851 901 951 1001 1051 1101 1151 1201 1251 1301 1351 1401 1451 1501 1551 1601 1651 1701 1751 1801

M WK L L L V A L A F A L A D A Q F Q P G K V Y R Y S Y D A F S I S G L P E P G V N R A G L S G E M K I E I H G H T H N Q A T L K I T Q V N L K Y F L G P WP S D S F Y P L T A G Y D H F I Q Q L E V P V R F D Y S A GR I GD I Y A P P QV T D T A V N I V R GI L N L F QL S L K K N QQT F E L QE T GV E GI C QT T Y V V QE GY R T N E MA V V K T K D L N N C D H K V Y K T MGT A Y A E R C P T C QK MN K N L R S T A V Y N Y A I F D E P S GY I I K S A H S E E I QQL S V F D I K E GN V V I E S R QK L I L E GI QS A P A A S QA A S L QN R GGL MY K F P S S A I T K MS S L F V T K GK N L E S E I H T V L K H L V E N N Q L S V H E D A P A K F L R L T A F L R N V D A G V L Q S I WH K L H Q Q K D Y R R WI L D A V P A M A T S E A L L F L K R T L A S E Q L T S A E A T Q I V Y S T L S N QQA T R E S L S Y A R E L L H T S F I R N R P I L R K T A V L GY GS L V F R Y C A N T V S C P D E L L QP L H D L L S QS S D R A D E E E I V L A L K A L GN A GQP N S I K K I QR F L P GQG K S L D E Y S T R V QA E A I MA L R N I A K R D P R K V QE I V L P I F L N V A I K S E L R I R S C I V F F E S K P S V A L V S MV A V R L R R E P N L QV A S F V Y S QMR S L S R S S N P E F R D V A A A C S V A I K ML GS K L D R L GC R Y S K A V H V D T F N A R T MA GV S A D Y F R I N S P S GP L P R A V A A K I R GQGMGY A S D I V E F GL R A E GL QE L L Y R GS QE QD A Y GT A L D R Q T L L R S G Q A R S H V S S I H D T L R K L S D WK S V P E E R P L A S G Y V K V H G Q E V V F A E L D K K M M Q R I S Q L WH S A R S H H A A A Q E Q I R A V V S K L E Q G M D V L L T K G Y V V S E V R Y MQP V C I GI P MD L N L L V S GV T T N R A N L H A S F S QS L P A D MK L A D L L A T N I E L R V A A T T S MS QH A V A I MGL T T D L A K A GMQT H Y K T S A GL GV N GK I E MN A R E S N F K A S L K P F QQK T V V V L S T ME S I V F V R D P S GS R I L P V L P P K MT L D K G L I S Q Q Q Q Q P H H Q Q Q P H Q H G Q D Q A R A A Y Q R P WA S H E F S P A E Q K Q I H D I MT A R P V MR R K QH C S K S A A L S S K V C F S A R L R N A A F I R N A L L Y K I T GD Y V S K V Y V QP T S S K A QI QK V E L E L QA GP QA A E K V I R MV E L V A K A S K K S K K N S T I T E E GV GE T I I S QL K K I L S S D K D K D A K K P P GS S S S S S S S S S S S S S S S S S D K S GK K T P R QGS T V N L A A K R A S K K QR GK D S S S S S S S S S S S S D S S K S P H K H GG A K R QH A GH GA P H L GP QS H S S S S S S S S S S S S S S A S K S F S T V K P P MT R K P R P A R S S S S S S S S D S S S S S S S S S S S S S S S S S S S S S S S E S K S L E WL A V K D V N Q S A F Y N F K Y V P QR K P QT S R R H T P A S S S S S S S S S S S S S S S S S S S D S D MT V S A E S F E K H S K P K V V I V L R A V R A D GK QQGL QT T L Y Y GL T S N GL P K A K I V A V E L S D L S V WK L C A K F R L S A H M K A K A A I G WG K N C Q Q Y R A M L E A S T G N L Q S H P A A R V D I K WG R L P S S L Q R A K N A L L E N K A P V I A S K L E M E I M P K K N Q K H Q V S V I L A A M T P R R M N I I V K L P K V T Y F Q Q G I L L P F T F P S P R F WD R P E G S Q S D S L P A Q I A S A F S GI V QD P V A S A C E L N E QS L T T F N GA F F N Y D MP E S C Y H V L A QE C S S R P P F I V L I K L D S E R R I S L E L QL D D K K V K I V S R N D I R V D GE K V P L R R L S QK N Q Y G F L V L D A G V H L L L K Y K D L R V S F N S S S V Q V WV P S S L K G Q T C G L C G R N D D E L V T E M R M P N L E V A K D F T S F A H S WI A P D E T C G G A C A L S R Q T V H K E S T S V I S GS R E N C Y S T E P I MR C P A T C S A S R S V P V S V A MH C L P A E S E A I S L A MS E GR P F S L S GK S E D L V T E ME A H V S C V A

Structure

the structural and chemical results. For example, although the LV1n chain begins at Gln17 and ends at Tyr688 in Figure 2, this segment actually encompasses more than two structural domains. The largest of the lipovitellin protein components, LV1n, has a Mr = 66,800 Da. LV1n resisted repeated attempts at N-terminal sequencing using the Edman reaction, and the N-terminal amino group was assumed to be blocked [4]. The X-ray structure suggested that the N terminus was a pyroglutamic acid occurring at residue 17 in the vitellogenin sequence (see Figure 2). A neural networks program developed by Nielsen et al. [15] determined that the leader peptide sequence of vitellogenin included only the first 16 residues with Gln17 being the N terminus of mature vitellogenin. The green underline in Figure 2 ends

at Tyr688 because this is the last residue to be fit into the X-ray model from the LV1n chain. Using Edman degradation, the N-terminal residue of LV1c (Mr = 40,750 Da) was determined to be Arg708, the beginning of the red underline in Figure 2. This means that either 20 amino acids are disordered on the C-terminal end of LV1n or are lost during the post-translational proteolytic processing of vitellogenin. Furthermore, the X-ray coordinates start again at Trp729, as shown in red letters in Figure 2, so that 21 amino acids belonging to the N-terminal region of LV1c are also disordered. Two more disordered segments were found in LV1c. No electron density could be found for the 19 residues in the region of Met759 to Ala777, which would run across the large lipid cavity opening. A second break begins at Met949 and

898

Structure 1998, Vol 6 No 7

continues for 42 residues to Phe990; this is one of the gray lettered segments underlined in red (Figure 2). The third chemically identified polypeptide chain, LV2, stains as a phosphoprotein in SDSPAGE experiments [1], and has an N-terminal amino acid sequence beginning with Lys1306 (Figure 2): KYVPQRKPQTSRRHTPASSSSSSSSSSSSSSSSSSSDSDMTVSAESFEKHSK1357. This segment is not visible in the electron-density map, and most of its serine and threonine residues are thought to be phosphorylated. Although not detected within the lamprey lipovitellin complex, the 231 residues from the end of LV1c to the beginning of LV2 represent the phosvitin region of the lamprey vitellogenin. This putative phosvitin contains 94 serine and seven threonine residues; approximately half of these residues appear to be phosphorylated as shown by 31P NMR experiments [1]. In summary, of the 1823 amino acids present in lamprey vitellogenin, the lipovitellin model contains 672 residues in LV1n, 285 in LV1c and 172 in LV2. The crystal structure includes several breaks so that there are a total of five peptide chains, one 40-atom phosphatidylcholine, one 13carbon hydrophobic chain, and 59 water molecules. Clearly, there are residues missing in the electron-density for LV1c and LV2. Interestingly, the 672 residues found by X-ray crystallography for LV1n gives a molecular weight of about 74,000, which is considerably greater than the 66,800 Da determined by SDSPAGE.
Domain organization of lipovitellin

adjacent solvent. LV1n is the longest piece of the primary structure and contains two domains, the N sheet and helical domain. In addition, five of the seven strands of the C sheet are from LV1n. Contacts between either the N sheet or the helical domain and the lipid cavity are limited. As will be described later, the lipid cavity is formed primarily from segments derived from LV1c and LV2, which comprise all of the A sheet and about a third of the C sheet (see Figure 3b). The overall model properties of the domains, as observed in the crystal structure, are described in Table 1. The A-sheet and C-sheet domains form the bulk of the lipidbinding cavity and generally are characterized by the weakest electron-density. To some degree, this can be seen by the B factors observed in the A and C sheets which are generally higher than those of the N sheet and the helical domain. Although the A and C sheets form most of the lining of the lipid-binding cavity and must have extensive contacts with the bound lipids, they have little contact with each other. This is apparent in Figure 3b. The contact areas were calculated: N sheet to helical domain, 2400 2; N sheet to A sheet, 6400 2; N sheet to C sheet, 2300 2; helical domain to A sheet, 5000 2; helical domain to C sheet, 4600 2; and A sheet to C sheet, 900 2. The major stabilizing domain of the entire complex is the helical domain. It has a large buried surface area relative to the A and C sheets. The smaller, seven-stranded sheet (the C sheet) is formed from LV1n (residues 196197 and 615688) and LV1c (residues 729758). It has been so labeled because it makes many contacts with the C terminus of the helical domain. A proteolytic cleavage site must occur between residues 688 and 708, and more likely between 700 and 708. The 700708 segment is suspected because it is part of a peptide insertion which is not found in the other vertebrate vitellogenins; LV1 in most species is not cleaved into two fragments. The larger sheet is composed of 23 strands formed primarily from LV1c (residues 778948 and 9911074) and LV2 (residues 13581529) with one strand coming from LV1n (residues 188190). It has been referred to as the A sheet because it has interactions with the amino terminus of the helical domain. As noted above, this domain forms most of the molecular surface for the lipid cavity. When compared to other lipovitellin family members, this domain contains the smallest number of conserved residues (16%), probably because it has the fewest proteinprotein contacts and the most sidechain contacts with lipid. The extent of conservation in the other domains, as determined by multiple sequence alignment, ranges from 25 to 31%. The N sheet and helical domains that have the most

The crystal structure can be divided into the three polypeptide components mentioned above or into a series of domains which conform more closely to the three-dimensional organization. The latter is clearly more useful in describing the crystallographic model. Both the polypeptide components and the general shape and domain structure of a single subunit are shown in Figures 3a and 3b. The domain structure and the large cavity are shown in Figure 3b. In this representation of the crystal structure, there are three predominantly antiparallel -sheet domains, referred to as the N sheet, A sheet and C sheet, and a large helical domain. The 12 strands of the N sheet domain are formed from LV1n and include residues 17296. Exceptions are the two strands located within a long loop (residues 186207). The main sheet structure is only one strand short of closing off to form a barrel-like domain. The connection between the domain structure and the proteolytic processing of the precursor is obscure. It is possible, however, to speculate about the location of the missing segments. Both the long phosvitin region between LV1c and LV2 and the shorter connection between LV1n and LV1c may be parts of the polypeptide chain which loop out from the core lipovitellin dimer into the

Research Article Structure of crystalline lipovitellin Anderson, Levitt and Banaszak

899

Figure 3 Stereoview ribbon diagrams of the lamprey lipovitellin monomer. (a) The view looking down into the funnel-shaped cavity with the widened anterior. The cartoon is color-coded to show the amino acid sequence organization in the crystal structure. Residues belonging to LV1n (Gln17 to Tyr688) are shown in green, residues belonging to LV1c (Trp729 to Lys758, Gln778 to Lys948 and Ser991 to Pro1074) are in red, and LV2 residues (Pro1358 to Phe1529) are in dark blue. The lipid ligands, L1 and L2, are drawn in orange, with the phosphate atom of L1 shown as a sphere in purple. The N and C termini of each chain are labeled with a residue number, as are all cysteine residues forming disulfide bonds. In addition, approximately every twentieth residue is labeled. Bonds have been drawn in yellow along the C-S-S-C atom positions for the five disulfide bonds (Cys156Cys182, Cys198Cys201, Cys443Cys449, Cys1014Cys1025 and Cys1408Cys1429). It was impossible to have them all appear readily visible and the reader may find it useful to also observe them in Figure 7. (b) The colors on the protein ribbon have been changed to illustrate the domain organization of the monomer. The structural domains are color-coded: the N-sheet domain is shown in green, the helical domain in cyan, the C-sheet domain in red, and the A-sheet domain in dark blue. (The figures were prepared using the program SETOR [53].)

proteinprotein contacts are the most highly conserved 31% and 30%, respectively. There are four regions of the vitellogenin sequence which are chemically known to be present in the complex, and which we were unable to build into the electron-density: Arg708Asp728, Met759Ala777, Met949Phe990 and Lys1306 Lys1357. Eight of the residues in Arg708Asp728 are due to the insertion responsible for the cleavage of LV1 into two chains. Met759Ala777 chemically links the C sheet to the A sheet; this 19 amino acid chain stretches across the large opening of the lipid cavity and presumably is surrounded on all sides by phospholipid headgroups. Met949Phe990 is very polar, including 12 glutamines and five histidines. The location of this gap (see Figure 3a) suggest that this very polar segment projects into the solvent region, forming a partially disordered minidomain. Part of this missing segment may interact with the posterior surface of the

lipid cavity as shown in Figure 3a. As mentioned previously, Lys1306Lys1357 contains a large number of serines believed to be phosphorylated, and this rather large segment would extend from the middle of the A sheet in an unknown conformation. On the basis of lipid and 31P NMR analyses, the number of phosphorylated serine and threonine residues per monomer was estimated at 55 [1,16]. No electron density was found to be consistent with a phosphoserine or phosphothreonine residue. In particular, we were unable to find any density for the polyserine segment including residues 13231341. This segment must be disordered and located in the surrounding milieu. It is also possible that a small phosvitin chain is missing in the X-ray model. This would consist of a subfragment between Ser1131 and Phe1305. This segment is 231 residues long and contains 94 serine residues believed to be phosphorylated. A

900

Structure 1998, Vol 6 No 7

Table 1 Temperature (B) factors for the domains in crystalline lipovitellin. No. of mainchain atoms N sheet Helical domain C sheet A sheet PLC C13 Waters 1120 1272 418 1711 No. of sidechain atoms 1084 1238 393 1582 40 13 59

Figure 4

Average B factor 53 50 67 77

Average B factor 55 55 72 81 43 45 38 Cartoon representation of the helical domain of lipovitellin. This enlarged view shows the layered nature of the helical domain. The position of the Cys443Cys449 disulfide bond is marked in yellow. The four partially or fully buried salt bridges in the helical domain are labeled: () Arg358Glu390, () Lys478Glu513, () Arg519Glu556 and () Arg547Glu574. Their respective sidechains are colored with anionic residues in red and cationic residues in blue. (The figure was prepared using the program SETOR [53].)

polypeptide chain of this composition probably would not stain with Coomassie blue on SDSPAGE gels.
Helical domain

The helical domain in lipovitellin is an unusual form of supersecondary structure not commonly found in a globular protein. It consists of a coil of helices which forms a clamp linking together the two sheets that form the walls of the lipid-binding cavity [2,9]. Viewed from the edge, the helical domain has a crescent shape and consists of two layers, one toward the center of the molecule, the other on the surface. Only about 30 sidechains of this domain interact directly with the cavity. The helical domain is found in LV1n, residues 297614, as shown in Figure 3. The long loop from 498 to 505 connecting helices 12 and 13 is in a region of poor electron-density, resulting in a bad torsional angle at Gln499. Arrangement of all the helices (with the exception of helix 8) in a right-handed supercoil results in a roughly parallel grouping of helices within each layer and an antiparallel grouping between layers. Much like helical domains in other proteins, the interhelical axes are canted slightly away from 0, corresponding to parallel, and 180, indicating antiparallel. The angular relationship between helical axes was determined using Promotif [17]. For helices in the same plane, the interhelical angles range from 15 to 66; for interactions between planes, the angles range from 118 to 164. In addition to the 17 helices of the supercoil, one short helix, helix 8, is present in a loop connecting the longer helical segments (helices 7 and 9 in Figure 4). The two layers of the helical domain lead to three sidechain surfaces: an outer layer, an inner layer and a central portion containing the interface between the two layers. The outer layer faces the solvent with the exception of

helices 14, 16 and 18 which make contact with the other monomer. Clearly less hydrophobic than the inner face, the outer layer contains nearly all polar and most of the ionizable sidechains in the helical domain. One of the most interesting features of the helical domain is exemplified by the amino acid types in the interface layer, which are largely hydrophobic. The nonpolar character of the middle surface suggests that this domain has a core resembling most other globular proteins. One would predict that the helical domain would form a stable globular polypeptide in the absence of the rest of the lipovitellin structure. Of the 59 water molecules found in the crystal structure, 25 are hydrogen bonded to sidechains in this domain. There is one disulfide bridge connecting the end of helix 9 to a loop right before the start of helix 10 (see Figure 4). There are also four partially buried salt bridges between neighboring helices each involving two hydrogen bonds between the electrostatically paired sidechains (Arg358Glu390, Lys478Glu513, Arg519Glu556 and Arg547Glu574). These bridges are labeled as , , and , respectively, in Figure 4. The first three bridges restrain helices on the inner surface of the helical domain: helices 4 to 6, helices 11 to 13 and helices 13 to 15. The fourth salt bridge occurs between helix 15 of the inner surface and helix 16 on the solvent-accessible surface. Nearly a third of the helical domain residues are strictly conserved in six of the seven vertebrate vitellogenins that were

Research Article Structure of crystalline lipovitellin Anderson, Levitt and Banaszak

901

compared. Among the conserved residues is a strikingly high number of conserved leucine (16) and valine residues (5). The arrangement of leucine, valine and isoleucine residues is somewhat similar to that found in the leucine zipper motif. Instead of two helices, however, the hydrophobic sidechains intercalate from four neighboring helices. A search for an internal consensus sequence for this supersecondary structure met with little success. The helical domain has 12 proline residues; nine are conserved within the lipovitellin family. Of the prolines found in the helical domain of lamprey lipovitellin, six are at the beginning of helices (Pro326, Pro425, Pro450, Pro486, Pro526 and Pro596), three are in turns (Pro497, Pro575 and Pro559), and three are in the middle of a helix (Pro366, Pro456 and Pro535). Although it is possible that these proline residues impose conformational restrictions favoring the parallel structure, it is likely that the packing of the helices is influenced more by the locations of the hydrophobic residues mentioned above. The helical domain of lipovitellin was the first-discovered superhelical right-handed, coil-coiled structure with a twohelix repeat unit [2]. A similar supersecondary structure was subsequently found in two domains of soluble lytic transglycosylase (SLT) from Escherichia coli [18]. In SLT, the N-terminal domain contains 22 helices in 360 amino acids packed in a U-shaped conformation. In combination with a second smaller domain containing four helices, a torus is formed with a large central hole 2535 in diameter. The helical domain in lipovitellin and SLT leads to a series of parallel helixhelix interactions that are in marked contrast to the all antiparallel helical contacts found in typical helical-bundle proteins, such as apolipoprotein E receptorbinding domain or apolipophorin III [19,20].
N-sheet domain

have comparatively low temperature factors (Table 1) and good geometry with the exception of the type IV turn from 165168 (in Figures 3a and 3b, this turn is behind residue 220). This segment forms crystal contacts between the lipovitellin dimers and is within a few ngstrms of a crystallographic twofold rotation axis. In addition, the program Errat [21] indicated that the -strand region from 4348 was found in less then 2% of its library of structures even though the electron-density for this region is very good. Finally, 20 of the 59 water molecules found in the crystal structure form hydrogen-bonding contacts with residues from this domain. A 14-residue helix is in the center of the shell formed by the N sheet. It consists of mainly uncharged residues and has the amino acid sequence T120DTAVNIVRGILLFQ. The single ionizable sidechain (Arg128) extends to the aqueous surface while most of the other helical sidechains are buried. The domain also contains two disulfide bridges found in the LV1n chain linking Cys156 to Cys182 and Cys198 to Cys201. The 156182 disulfide bridge appears to be conserved in both MTP and apoB. An extended connection from 185208 forms two short strands, one interacting with the C sheet and the other with the A sheet through typical strand strand hydrogen bonds. The combination of the 185208 segment and a depression in the surface of the N sheet leads to an unusual crevice or pit that accommodates the ends of several of the long strands that form the walls of the lipid-binding cavity. The strands belong to the A sheet and are curved to conform to the direction of the strands in the N sheet. Strandstrand hydrogen bonding occurs and a short segment of sheet forms from segments of polypeptide chain that are long distances apart in the primary structure. These interactions have the appearance of a balland-socket junction which appears to have the potential to be flexible. This junction could have an inherent function in the loading of lipid to the folded or partially folded polypeptide chain.
The A and C sheets and the lipid-binding cavity

Consisting of residues 17296 of LV1n, this domain contains the putative N-terminal pyroglutamate (pyrrolidone carboxylic acid) identified by the placement of the amino acid sequence in the electron-density map. The presence of the modified N terminus suggests that the first 16 amino acids comprise the endoplasmic reticulum (ER) signal sequence and are removed during translocation of vitellogenin into the ER. A glutamine sidechain at the N terminus will spontaneously cyclize, eliminating ammonia to form pyroglutamate. As can be seen in Figures 3b and 5, the N-sheet domain resembles the compact structure of a typical globular protein. It contains 4 helices and 12 strands, 11 of which form a barrel-like conformation. The barrel has a gap between two strands which prevents formation of a continuous surface by the structure. Of the four -helical segments, two are only four residues long. The electron-density for this portion of crystalline lipovitellin was generally good. Atoms in the N-sheet domain

Lamprey lipovitellin, as noted previously, must contain anywhere from 2540 molecules of lipid (mostly phospholipid) per subunit. Using 31P NMR, it has been suggested that the phospholipid is present in the form of a condensed state such as a monolayer or bilayer [16]. The precise number of bound lipid molecules is not easily determined and indeed may vary somewhat from molecule to molecule. Experimental evidence for lipid domain variations comes from the observation that additional spin-labeled fatty acids can be taken up by lipovitellin when it is in the solution state [22]. Either the protein is very rigid, and water and other molecules are exchanged to accommodate these spin probes, or the cavity can expand somewhat to accommodate additional lipid molecules.

902

Structure 1998, Vol 6 No 7

Figure 5 Stereoview ball-and-stick representation of the lipovitellin monomer showing the polarity of its sidechains. The backbone atoms (N, C, C and O) are all colored light gray. Hydrophobic sidechains (alanine, valine, leucine, isoleucine, methionine, phenylalanine and proline) are colored green, anionic sidechains (glutamate and aspartate) are colored red and cationic sidechains (lysine, arginine and histidine) are colored blue; all other residues are colored yellow. The preponderance of hydrophobic residues along the inner surface of the lipid cavity is apparent. In addition, there are rings of basic residues occurring at positions that could interact with the phosphate moiety of the bound phospholipids. (The figure was prepared using the program SETOR [53].)

Earlier crystallographic models of lipovitellin indicated a large cavity in each of the subunits of the dimer. Because of weak electron-density around the binding cavity, particularly at the solvent-accessible turns connecting the antiparallel strands, it was the most difficult region to interpret. In fact, as shown in Figure 2, several segments of LV1c and LV2 are still missing in the map and X-ray model. Neutron-scattering studies with contrast variation provided strong evidence that the cavities contained the bound lipid [9]. Once it was demonstrated that the cavity contained the lipid, the X-ray crystallographic results showed that there must be two noncontiguous lipidbinding domains present in the dimer.

The lipid-binding cavity, as visible in Figures 3a and 3b, is formed primarily by the A and C sheets and a small section of the helical domain. It is roughly funnel- or cupshaped and is accessible to solvent at both ends. Viewing the monomer (Figures 3 and 5), a third breach in the protein surface appears, which is closed off by dimerization (Figure 6). The two -sheet domains consist of mainly antiparallel strands. Residues 778948 represent the first contiguous segment of the A sheet. This first segment has two bends in most of the strands leading to a Z appearance when the sheet is viewed edge-on. The topmost part of the Z fits into the pit in the N domain and forms the ball-and-socket arrangement referred to earlier.

Figure 6 Stereoview ribbon diagram of the lipovitellin homodimer. The crystalline lipovitellin dimer is depicted in cartoon representation. Each monomer is color-coded as in Figure 3b: N sheet green, helical domain cyan, C sheet red, A sheet dark blue, disulfide bonds yellow, lipids (L1 and L2) orange, and the phosphate of L1 purple. Two black lines above and below the homodimer mark the position of the dimer dyad. Most of the A sheet is unsupported by monomermonomer contacts. (The figure was prepared using the program SETOR [53].)

Research Article Structure of crystalline lipovitellin Anderson, Levitt and Banaszak

903

Two other unusual bends lead to the appearance of a bulge on the otherwise relatively smooth sheet. The strand appears to be distorted beginning at residues 915 and 1056 with the amino acid sequences P915FQQKT and P1056TSSKA, respectively. The distortion appears to accommodate a change in orientation of the sheet belonging to LV2 (dark blue segment in Figure 3a). As noted in the section describing the amino acid sequence, residues from 13061358 are missing. Half of the residues in this segment are serines that are believed to be phosphorylated. The polarity of the sidechains that line the cavity is visible in Figure 5. The sidechains pointing inward towards the cavity are predominantly hydrophobic with the additional presence of a few neutral polar groups from threonine, serine, asparagine, glutamine, tryptophan, cysteine or tyrosine residues. The hydrophobic residues are relatively uniformly distributed over the cavity surface. Around the larger opening to the cavity is a ring of basic and polar sidechains including Lys757, Lys758, Lys787, Lys1359, Tyr1382, Thr392, Ser388, Gln1071, Asn1387 and Thr1440. In the hypothetical initial positioning of phospholipid molecules described below, we considered these sidechains to be of significance. In this modeling, the hydrophobic lipid fills the cavity; the charged hydrophilic phospholipid headgroups form the aqueous interface at the mouth of the funnel and interact with the ring of basic and polar sidechains.
Ordered lipid molecules

At the L1 site, three sidechains, Tyr187, Thr189 and Thr192, belonging to the N sheet, form hydrogen bonds with the phosphate of the headgroup. None of these residues is conserved in the other vertebrate vitellogenins. Van der Waals contacts are made between the hydrocarbon chain and residues of the A sheet, Val805, Tyr807 and Ala871, and the C sheet, Tyr669. Only Val805 and Ala871 are strongly conserved. The second lipid-binding site, L2, is difficult to describe in any detail. It could be anything from a segment of a phospholipid to a triglyceride. The only crystallographically ordered fragment was modeled as a C13 hydrocarbon chain. It appears in a region that is relatively distant from the lipid cavity and has no direct connection to the lipid funnel. L2 is located in a nonpolar region of the protein and has hydrophobic interactions with residues from the N sheet: Ala30, Leu71, Phe93, Thr122 and Ile126. The other part of the binding site derives from the A sheet and includes sidechains from Ile813, Ile815, Thr877 and Leu879. The two isoleucines form an antiparallel turn having the sequence Ile-Gly-Ile. In the other vertebrate vitellogenins, three of these residues, Thr122, Ile126 and Thr877, are strongly conserved and the fourth position, Leu879, is always hydrophobic.
Zn2+ site?

There were two regions of electron-density in the final map which could not be interpreted by the polypeptide chain. One region was accurately modeled by a phosphatidylcholine with carbon chains of length 16 and 6, labeled L1 in Figures 3a and 3b. The other region resembled an extended 13-atom hydrocarbon chain, and is labeled L2 in Figures 3a and 3b. Both of these compounds must be bound with sufficient affinity to be ordered but still may have disordered segments. We believe they are lipid molecules and are in a different state than most of the lipid bound by lipovitellin. The electron-density for both L1 and L2 was strong (35) and both structures refined with relatively low temperature factors as given in Table 1. Both L1 and L2 are located near the juncture of the N and the A sheets. The interaction between these two domains forms the ball-and-socket relationship described earlier. The phosphatidylcholine, L1, binds in a large loop formed by residues 185206 at the narrow end of the lipid funnel. This loop region of the N sheet is stabilized on both ends by disulfide bridges linking Cys156 to Cys182 and Cys198 to Cys201. Loop 185206 is the nexus of the structure, interacting with residues in the helical domain, the A and C sheets, and L1.

Both calcium and zinc ions have been found bound to lipovitellins and probably have an important role in differentiation and development of the oocyte. Recent EXAFS studies indicated the presence of at least one Zn2+ per monomer in most lipovitellins. This data suggested that the Zn2+ ion was liganded to two histidine residues [23]. Montorzi et al. [8] speculated that Zn2+ would not be found in our lipovitellin crystal structure because of the presence of 1 mM EDTA in the buffer used for crystallization. This appears to be the case as we were unable to find any evidence of Zn2+ density. In spite of the absence of any sign of metal ions in the electron-density map, the crystallographic model was examined for possible metal-binding sites. Of the 23 histidines present in the model, only two regions have histidine pairs that could possibly form the Zn2+ binding site. They are His312 and His322 in the helical domain, and His868 and His887 in the A sheet. The two sites are only 12 apart. The His312/ His322 site is more highly conserved and is found in the chicken, toad, trout, lamprey, and one of the killifish sequences. The immediate environment around this site contains several additional residues which could interact with Zn2+: Asn316, Asp324 and Lys328. The His868/His887 site is less conserved. Although His868 is found in all but one of the killifish sequences, His887 is conserved only in lipovitellin from lamprey, toad and sturgeon. The prediction that a Zn2+ site is present and involves His312/His322 may be important to embryogenesis.

904

Structure 1998, Vol 6 No 7

Interdomain and subunitsubunit interactions

Table 2 Table of polar interactions between monomers in lamprey lipovitellin. Monomer 1 E38 OE1 E38 OE2 E233 OE1 R635 NH1 K1043 NZ T176 OG1 N215 OD1 E233 OE1 K244 NZ K244 NZ Q997 OE1 I1001 O Monomer 2 R1506 NH1 R1506 NH2 R572 NH1 E1482 OE1 E1482 OE2 A562 O R570 NE R572 NH1 S591 O S593 O K1489 NZ N1508 ND2 Distance () 3.13 2.80 3.34 2.94 3.10 2.89 2.75 3.34 3.24 3.23 3.26 2.96

Because of the unusual framework of the lipoprotein, added significance must be given to interdomain interactions. Although ten of the 14 cysteine residues present in a lipovitellin subunit form disulfide bonds, none of these is an interdomain connection. Instead, there is a large number of conserved salt bridges involved in important interdomain interactions. They are characterized by dual hydrogen-bond separations between polar atoms of basic and acidic sidechains. For example, four of the 12 conserved and partially buried salt bridges form in the interface between the helical domain and the C sheet. There is also a conserved ionic bridge between the helical domain and the A sheet (Arg337 to Glu804). In contrast, the contacts between the N-sheet domain and A sheet are more nonpolar in character. The hydrophobic residues of the -turn connecting strands 2 and 3 inserts into a cleft present in the N sheet, just to the right of L2 in Figures 3a and 3b. In addition, the loops connecting strands 4 and 5, 6 and 7, and 8 and 9 of the A sheet all have sharp, nearly 90 bends, which present residues involved in hydrophobic interactions with the N sheet. As both lipovitellin and vitellogenin are dimers in solution, the subunitsubunit interface is important to the overall structure. The early X-ray studies demonstrated that the molecule has C2 symmetry and in the crystals the dyad is congruent with a crystallographic twofold rotation axis. Figure 6 shows one view of the dimer which has the appearance of an X. A single subunit runs along each of the diagonals. It has been color coded to highlight the arrangement of the main cavity-forming domains the N sheet and the helical domain. The latter two domains form the principal contact surfaces between subunits. Close to the molecular dyad, the A sheet also makes subunitsubunit contacts. There are strong monomermonomer contacts that stabilize the noncavity regions and which support the regions of the cavity sheets where they attach to the compact protein region (Table 2). Of the area of the dimer contact, 41% is between the two A sheets. The remainder is between the N sheet of one subunit and the helical domain of the other. This suggests that both the N sheet and helical domains have an important role in organizing and supporting the cavity forming A and C sheets, and that dimerization plays an integral role in establishing the functional lipid-binding structure.
Hypothetical phospholipid binding in the cavity

or bilayer [16]. The disorder in the crystal structure suggests that multiple conformations are likely but this would still be in agreement with the NMR results. By taking into account the presence of nonpolar residues lining the cavity and the ring of basic and polar residues identified and mentioned above, possible orientations for a phospholipid microdomain were tested. Only a monolayer was considered in detail as the funnel shape of the cavity would otherwise necessitate modeling layers with different numbers of phospholipid on the two sides and would not provide for an adequate volume of neutral lipid. The monolayer shown in Figure 7 was constructed by manually positioning the dipalmitoyl phosphatidyl choline (DPPC) molecules along the edges of the mouth of the cavity, being careful to place phosphate groups close to the positively charged sidechains present at the outer ring of the cavity. A total of 38 molecules of DPPC were added to the crystallographic coordinates of the protein. Several rounds of conjugate gradient energy minimization using X-PLOR followed by manual rebuilding were carried out. During these calculations, the protein coordinates were fixed and the electrostatic terms turned off. The structure of the DPPC monolayer as shown in Figure 7 satisfies simple steric considerations. Moreover, the energy dropped from 5.5 109 to 760 (kcal mol1, relative scale). Because of the initial placement of DPPC molecules interacting with the ring of basic residues, the hydrocarbon chains of one DPPC molecule ended up nearly perpendicular to another across the opening. As an increasing number of DPPC molecules were added, a

The total absence of electron-density for any lipid molecules in the binding cavity leaves a major unanswered question regarding the orientation and protein interactions of the bound lipid. On the basis of the 31P NMR results, the phospholipid should be present as either a monolayer

Research Article Structure of crystalline lipovitellin Anderson, Levitt and Banaszak

905

Figure 7 Stereoview of the lipovitellin monomer perpendicular to a modeled DPPC monolayer. The crystal structure shown includes bound phospholipid modeled as described in the text. Each monomer is colored as in Figure 3b: N sheet green, helical domain cyan, C sheet red, A sheet dark blue, disulfide bonds yellow, lipids (L1 and L2) orange, and the phosphate of L1 purple. The 38 energyminimized DPPC molecules are drawn in pink except for their phosphate atoms which were drawn as purple spheres. A representation of biliverdin (pink) was placed in the inner lipid region of the cavity. The biliverdin marks the portion of the cavity assigned to neutral lipids. (The figure was prepared using the program SETOR [53].)

bulge was necessary in the monolayer near the center. The net effect was the visible dome-shaped curvature on the surface of the monolayer. With the phospholipid microdomain present in the crystal structure of the protein, the lipovitellin molecule had a smoother overall surface. When examined closely, the hydrocarbon chains of the microdomain have neighbors with similar orientations. This puts each phospholipid molecule in an anisotropic environment. Therefore, in localized segments, the DPPC orientations resemble that found in a monolayer. Over the entire microdomain and because of the curvature, however, the domain also has a crude micellar appearance. The current model is thus a hybrid between a microdomain of a single-layered liposome and a micelle. The inner surface of the phospholipid microdomain consists of the hydrocarbon tails of the phospholipids and therefore is completely nonpolar. To complete the description of the lipid cavity shown in Figure 7, several additional points are noteworthy. A nearby segment consisting of residues 759777 is disordered, and could be responsible for reducing part of the headgroup surface area in the present model. In addition, a relatively large cavity still remains near the inner concave surface formed by the hydrocarbon tails of the phospholipid. This could function to accommodate the neutral lipid (4.5% w/w) known to be present. No attempt has been made to model the neutral lipid, but the site is marked in Figure 7 by the bile pigment biliverdin (pink porphyrin ring) which is known to be present at a stoichiometry approaching one per dimer [24]. This same internal cavity could be the location of other neutral lipids such as triglycerides.

Implications of lipovitellin structure for MTP and apoB

As discussed above, the 672 N-terminal residues of lipovitellin are homologous to the N-terminal regions of the lipoproteins MTP and apoB. MTP is an intracellular protein involved in the transfer of lipid to apoB in the first stage of VLDL formation. ApoB is the major protein constituent of VLDL and low density lipoprotein (LDL). As all three of these proteins are involved in lipid transport, the amino acid sequence homology can be used to make some predictions about structures. In terms of evolutionary analysis, the sequence relationships suggest that vitellogenins evolved first, and that MTP and apoB arose by divergent evolution [25]. Consistent with this analysis, a BLAST [26] search of the Caenorhabditis elegans genome yielded several copies of vitellogenin, but no MTP or apoB [27]. In terms of the crystal structure of lipovitellin, the amino acid sequence similarities predict that all of the N sheet, the helical domain and most of the C-sheet region are present in the homologous family. Only the A sheet which forms the bulk of the extended lipid-binding cavity (see Figures 3b and 6), contains no significant similarity to the amino acid sequences of the other family members. As the A sheet in lipovitellin consists of 450 residues and only about 200 residues are available to form a similar domain in MTP, this cavity-forming region must either be significantly truncated, or it must involve an altogether different conformation. The former hypothesis is consistent with the much smaller lipid:protein monomer ratio of about 3 in MTP [28,29] versus 38 or more in lipovitellin [1]. The 676 residues which are homologous to lipovitellin represent only about 15% of the polypeptide chain of apoB. Predictions about the structure of the LDLs based

906

Structure 1998, Vol 6 No 7

only on the N-terminal homology of lipovitellin with apoB would not be reliable. Nonetheless, because the homologous region in the three proteins is a relatively compact globular unit, this segment could be the portion that serves as the framework for domains which are more directly bound to aggregated lipid. Expression of only the N-terminal 583 amino acids of apoB (the N sheet and helical domain of lipovitellin), yields a soluble protein which cannot bind enough lipid to form a lipoprotein [30]. One other major functional difference in these three proteins is related to the nature of the lipid components. Phospholipid is the principal component in lipovitellin, while triglycerides and other neutral lipids make up the majority of the lipid in LDL and VLDL (apoB). In the crystal structure of lipovitellin, headgroups of the phospholipids provide the hypothetical interface between the cavity and the aqueous environment. The greater predominance of triglycerides in MTP and apoB suggests that the lipid cavity in these two proteins must have significantly smaller aqueous interfaces relative to total cavity volume than is found in crystalline lipovitellin.
Structure and function of lipovitellin

The lipid domain of lamprey lipovitellin is known to contain both phospholipid and a smaller amount of neutral lipid. The crystal structure of the mature protein suggests one sterically reasonable form of phospholipid organization. In the model tested, the outermost segment would contain the phospholipid organized in a monolayer form. As noted above, the presence of a phospholipid monolayer added to the crystal structure leaves an unoccupied inner cavity to serve as the site for the neutral lipid. In view of the protein framework of lipovitellin, a reasonable assembly model is that lipid loading occurs cotranslationally and before the presence of the fully folded polypeptide chain as visualized in the crystal structure reported here. Simultaneous addition of neutral and phospholipid during protein folding eliminates the need for conformational changes which would be required to give access to the inner or neutral lipid segment of the cavity. If the current positioning of the phospholipid is accepted, its loading could involve a single-step removal from the inner leaflet of any nearby bilayer membrane.
Summary

The lack of direct information on the positioning of bound lipid in the crystal structure of lipovitellin makes it difficult to speculate on the most obvious question of function. By what mechanism are the lipid components loaded into the precursor structure? This question is further complicated by the fact that the loading process and subsequent secretion occurs in the precursor, vitellogenin. The mature lipovitellin form, for which the crystal structure is known, is missing significant portions (678 amino acids) of the polypeptide chain that must be present during the lipid assembly. One can more easily speculate about the removal of the lipid portion. In this case, it is known that the presence of the bound lipid is essential for the stability of the proteolytically processed lipovitellin. Once delipidated, some of the polypeptide chains are insoluble and it is unlikely that the removal of lipid can be reversed [31]. Recall that the lipid-binding cavity is formed principally by the A sheet. Formed primarily from two separate peptide chains, the A sheets overall structure is dictated by the presence of lipid in the cavity. In the precursor vitellogenin dimer, the apoprotein is a single polypeptide chain. It is likely that the precursor also binds the two molecules of lipid, L1 and L2, mentioned above. This process may mark the first step in the assembly process. Although it is unknown whether these two sites affect the conformation of the protein, the binding sites are near the ball-and-socket site. This location is where conformational changes probably occur as an increasing amount of lipid is added to the apoprotein.

The placement of the amino acid sequence into the electron-density map agrees with all aspects of the available X-ray and sequence data. The interpretation of the electron-density map was very difficult for several reasons. One of them was the fact that no chemical sequence was easily obtainable from the C-terminal end of each chain. Derived endings based on SDSPAGE molecular weights turned out to be unsatisfactory. In addition, many polypeptide segments were disordered and weak electron-density, particularly at surface turns, made tracing the chains difficult. In spite of these difficulties, most of the secondary structure was plainly visible and the X-ray coordinates refined to a reasonable agreement with the diffraction data. The X-ray crystallographic studies could be carried one step further. The crystals, which are identical to those found in situ [32], can be studied by very low temperature data collection methods. This might reduce some of the observed disorder, permit visualization of additional lipid molecules, and extend the resolution of the X-ray study.

Biological implications
The secretion and uptake of soluble lipoprotein molecules represents the major pathway for the relocation of large quantities of lipid in higher organisms. Mechanisms for the assembly of these complex molecules are difficult to envision because of a lack of information on their molecular structures. Vitellogenins proteolytically processed subform, lipovitellin, is a relatively homogeneous stable lipidprotein complex that has been crystallized and analyzed by X-ray crystallography. Because of its size and abundance during reproductive cycles, vitellogenin mRNA was one of the first mRNAs

Research Article Structure of crystalline lipovitellin Anderson, Levitt and Banaszak

907

to be identified. Like apolipoprotein B-100, vitellogenin is synthesized in the liver, secreted into the serum, and imported into target tissues (e.g., oocytes) through receptor-mediated endocytosis with receptors that are homologous to low density lipoprotein (LDL) receptors. Subsequent to the receptor-mediated uptake, the lipidprotein complex vitellogenin is proteolytically processed into several polypeptide chains which form lipovitellin and phosvitin. In its final deposited form, lipovitellin contains about 16% w/w lipid mostly phospholipid with some neutral lipid. The proteolytic processing does not appear to alter lipid binding, and the X-ray model of lipovitellin is probably closely related to the molecular structure of vitellogenin. The molecular structure derived from X-ray crystallographic analysis provides insights into the lipid stabilization and guidelines for potential assembly mechanisms. The large cavity or indentation that appears in the refined X-ray model contains the tightly bound lipid. The location of two other unique lipid sites suggest that they may have a role in the nucleation of lipid uptake during assembly of the complex. Because the cavity has a surface which is exposed to the aqueous environment, the alignment of phospholipid headgroups in the hypothetical monolayer could provide a required interface to the milieu. A segment of the polypeptide chain containing a large number of phosphorylated serines was missing in the X-ray studies, but is known to exist from NMR studies. The lack of electron-density for this segment indicates it is in a disordered state relative to the rest of the lipoprotein. On the basis of N-terminal analyses and molecular weight studies, several other small segments of lipovitellin are also missing in the X-ray crystallographic model. As they remain with the complex throughout the purification and crystallization procedures, they also must be an integral part of the lipidprotein complex. The lack of electron-density and the assumed fluidity of these segments suggests that they may be interacting directly with the lipid-containing domain. Comparison of the amino acid sequence and exonintron boundaries of lipovitellin with those of two other lipoproteins microsomal triglyceride transfer protein (MTP) and apolipoprotein B-100 (apoB) reveals regions of homology. The crystal structure described here therefore offers a template for the structure of MTP and a segment of apoB. These three proteins exhibit dramatically different levels of lipid binding and explanation for this may lie outside of the purported homology regions. One hypothesis would be that the fourth domain of lipovitellin, the A sheet, is responsible for the size of the lipid-binding cavity. The larger the potential A sheet of the molecule the larger the total amount of lipid which can associate

with the protein. MTP has only about 200 amino acids to form a putative A sheet, the lipovitellin A sheet comprises over 400 amino acids, and apoB has several thousand amino acids to form a putative A sheet. This increase of residues to the C terminus of the homology region correlates with the increased number of lipid molecules bound by each protein.

Materials and methods

X-ray data
Crystalline lamprey lipovitellin belongs to the monoclinic class, C2 with cell dimensions of a = 192.9, b = 87.1, c = 91.0 and = 100.92. The X-ray data was collected on the Xuong-Hamlin multiwire detector and is described in an earlier report by Raag et al. It contains reflections to 2.8 with an Rsym of 6.1%. Heavy-atom phasing statistics from four derivatives lead to an overall figure of merit of 0.49. Although lipovitellin is a dimer, the molecular symmetry is coincident with a crystallographic twofold rotation axis.

Sequence assignment
The N-terminal sequence and molecular weight assignment for two of the polypeptide chains were taken from earlier publications [1,4]. According to the model derived from the X-ray data, LV1n has a molecular weight of about 75 kDa, a difference of nearly 8 kDa from the molecular weight inferred from SDSPAGE studies. An independent measure of molecular weight was determined from the amino acid composition. Protein chains were separated from each other using preparative SDSPAGE and gel filtration under denaturing conditions. Approximately 10 g of each isolated chain was submitted for amino acid analysis on a Beckman 6300. A comparison of a histogram of the amino acid composition and the predicted cDNA sequence suggests a molecular weight of 71.5 kDa for LV1n. Similar comparisons for LV1c and LV2 were within 1 kDa of the SDSPAGE results. The identification of pyroglutamate at Gln17 as the blocking group at the N-terminal end of LV1n was based on the following arguments. First, two other vertebrate LV1 chains which are not blocked, toad and a killifish, have an N-terminal glutamate or asparagine that is homologous to Gln17. Second, there is no detectable electron density before Gln17. Third, the electron density on an omit map at Gln17 is closely fit by pyroglutamate. Fourth, pyroglutamate is a commonly modified form of an N-terminal glutamine [33]. Fifth, Gln17 in vitellogenin marks the most probable beginning of the LV1n chain of lipovitellin as determined by a neural networks program developed by Nielsen et al. [15].

X-ray refinement
The initial map used for building the model was based on the previous multiple isomorphic replacement model [2], and a new round of solvent flattening [34]. A computer program was developed to aid in assigning residues to the model using an eight-point scale to characterize sidechain density. The underlying method compared a region of sidechain density of defined character to an input amino acid sequence and selected the sequences with the best match. It also permitted the linking together of several small segments with variable gaps. Using this methodology, most of the amino acid sequence of the helical domain (Thr297Met611) and a large portion of the A sheet (Gly791 to Leu945) were placed in the electron-density map. For this sequence assignment, the model building program MAID [35] was used for visualization, and its real-space molecular dynamics options were used to improve portions of the fit of the backbone to the map. Where the sequence assignment could be made, sidechains were built into the density using the rotamer database available in O [36], while the rest of the model was refit with polyserine. An initial model consisting of 470 residues with assigned sidechains and 418 polyserines was then refined (Rfree = 0.45 and R = 0.31) in X-PLOR [37,38]. The standard

908

Structure 1998, Vol 6 No 7

simulated annealing protocol and the geometry library of Engh and Huber [39] were used with X-ray data from 2.88.0 [40]. The Rfree dataset was approximately 3% of the overall X-ray data [38]. Using the phases obtained after density modification [3], the heavyatom parameters were re-refined for data between 2.8 and 20 using MLPHARE [41] in the CCP4 package [42,43]. This map was then subjected to density modification using the program DM (CCP4) [43,44] including solvent flattening with 51% of the total unit-cell volume estimated as solvent, histogram matching and iterative skeletonization to obtain a map from 2.846 . Model phases were calculated using SFALL (CCP4) and combined with DM phases using SIGMAA (CCP4) for phases from 2.85.0 [43,45]. Electron-density maps were calculated using DM/model phases in the resolution range 2.85.0 and DM phases in the range of 5.046 . Starting with this map, iterative cycles of model building, restrained positional refinement, and phase recombination were performed. After several such cycles, a model was obtained with 1070 residues on 10 chains. All further cycles with X-PLOR were done with a bulk solvent correction applied to the model structure factors using the values fsolvent = 0.33e/3 and Bsolvent = 552. Rounds of restrained positional energy refinement and individual B-factor refinement for all reflections in the resolution range of 462.8 with no sigma cutoff were followed by model rebuilding into Sim-weighted 2|F o| |Fc| and |Fo| |Fc| maps. This led to the final structure with sequence assigned to 1129 residues on five chain fragments. Disulfide bonds were restrained after at least one cycle of refinement as free SH groups. Lipid molecules were included in the structure if they appeared in Sim-weighted |Fo| |Fc| map at 3 and 2|Fo| |Fc| map at 1. A diundecanoyl phosphatidylcholine was used as a template for L1, a palmitoyl-hexanoyl-phosphatidylcholine. The tridecane (C 13) was built from a subset of stearic acid (PDB; 1lif). Both lipid molecules were positioned into |Fo| |Fc| density at a 1 cutoff using torsion angle manipulations. Parameter and topology files for use in X-PLOR were generated with XPLO2D [46]. The 59 water molecules were built into 4 |Fo| |Fc| peaks if there was at least one nearby hydrogenbond donor or acceptor. The final Rfree = 0.255 for all reflections 2.850.0 with no sigma cutoff. A final cycle of positional and temperature factor refinement was done on the model using all data including the Rfree reflections, and the final R = 0.194. The mean B factors for LV and its ligands are listed in Table 1. The root mean square deviations (rmsd) from ideality in bond lengths, angles and improper angles are 0.005 , 1.38 and 1.02, respectively. The model has good stereochemistry, with only seven outliers in a Ramachandran plot (PROCHECK) [47]. Besides possessing good through-space contacts between the various fragments, correct assignment of protein sidechains was verified with the ERRAT [21] and Profile [48] programs. The three strongest (CH3)HgCl heavy-atom sites were all very close to three of the four free sulfhydryls present in the model.

We refer frequently to homologous amino acid sequences in the vitellogenin family. An alignment of the amino acid sequences of seven different species has been made, including the amino acid sequences of vitellogenin from toad (Y00354, EMBL), chicken (M18060, GenBank), sturgeon (U00455, GenBank), trout (X92804, EMBL), lamprey (M88749, GenBank) and two sequences from killifish (U70826 and U07055, GenBank). The alignment covers approximately 1900 residues and is available from the authors upon request.

Accession numbers
Coordinates for the refined model and structure factors have been deposited in the Protein Data Bank, Brookhaven, with accession code 1LLV.

Acknowledgements
We thank the National Institutes of Health for its continued support of this work through grant GM13925. We are indebted to the Minnesota Supercomputer Institute for supercomputing grants without which both the refinement and lipid model building would have been difficult. We acknowledge frequent discussions with all members of the Banaszak laboratory and for numerous helpful discussions with J Scott and C Shoulders at Hammersmith Hospital in London, UK regarding the homology with MTP. We recognize the contribution of E Hoeffner through his diligent maintenance of the computational facilities. The amino acid analyses were conducted by the Microchemical Facility at the University of Minnesota directed by E Eccleston.

References
1. Meininger, T., Raag, R., Roderick, S. & Banaszak, L.J. (1984). Preparation of single crystals of a yolk lipoprotein. J. Mol. Biol. 179, 759-764. 2. Raag, R. Appelt, K., Xuong, N.-H. & Banaszak, L. (1988). Structure of the lamprey yolk lipidprotein complex lipovitellinphosvitin at 2.8 resolution. J. Mol. Biol. 200, 553-569. 3. Banaszak, L., Timmins, P. & Poliks, B.J. (1990). The crystal structure of a lipoprotein: lipovitellin. In Molecular Biology of Atherosclerosis: Proceedings of the 20th Steenbock Symposium, June 35, 1990, University of Wisconsin-Madison. (Attie, A.D. ed.), pp. 135-140, Elsevier, New York. 4. Sharrock, W.J., et al., & Banaszak, L. (1992). Sequence of lamprey vitellogenin: implications for the lipovitellin crystal structure. J. Mol. Biol. 226, 903-907. 5. Byrne, B.M., Gruber, M. & Ab, G. (1989). The evolution of egg yolk proteins. Prog. Biophys. Mol. Biol. 53, 33-69. 6. Ohlendorf, D.H., Barbarash, G.R., Trout, A., Kent, C. & Banaszak, L.J. (1977). Lipid and polypeptide components of the crystalline yolk system from Xenopus laevis. J. Biol. Chem. 252, 7992-8001. 7. Norberg, B. & Haux, C. (1985). Induction, isolation and a characterization of the lipid content of plasma vitellogenin from two salmon species: rainbow trout (Salmo gairdneri) and sea trout (Salmo trutta). Comp. Biochem. Physiol. B 81, 869-876. 8. Montorzi, M., Falchuk, K.H. & Vallee, B.L. (1995). Vitellogenin and lipovitellin: zinc proteins of Xenopus laevis oocytes. Biochemistry 34, 10851-10858. 9. Timmins, P.A., Poliks, B. & Banaszak, L. (1992). The location of bound lipid in lipovitellin complex. Science 257, 652-655. 10. Baker, M.E. (1988). Is vitellogenin an ancestor of apolipoprotein B100 of human LDL and human lipoprotein lipase? Biochem. J. 255, 1057-1060. 11. Shoulders, C.C., et al., & Scott, J. (1993). Abetalipoproteinemia is caused by defects of the gene encoding the 97 kDa subunit of a microsomal triglyceride transfer protein. Hum. Mol. Gen. 2, 2109-2116. 12. Shoulders, C.C., et al., & Banaszak, L.J. (1994). The abetalipoproteinemia gene is a member of the vitellogenin family and encodes an -helical domain. Nat. Struct. Biol. 1, 285-286. 13. Knott, T.J., et al., & Scott, J. (1986). Complete cDNA and derived protein sequence of human apolipoprotein B-100. Nucleic Acids Res. 14, 7501-7503. 14. Baker, M.E. (1988). Invertebrate vitellogenin is homologous to human von Willebrand factor. Biochem. J. 256, 1059-1063. 15. Nielsen, H.N., Engelbrecht, J., Brunak, S. & von Heijne, G. (1997). Identification of prokaryotic and eukaryotic signal peptides and prediction of their cleavage sites. Protein Eng. 10, 1-6. 16. Banaszak, L.J. & Seelig, J. (1982). Lipid domains in the crystalline lipovitellin/phosvitin complex: a phosphorus-31 and deuterium nuclear magnetic resonance study. Biochemistry 21, 2436-2443.

Building the phospholipid bilayer in the lipid cavity

The DPPC coordinates were generated by using BIOSYM [49] to add five carbon atoms to each chain of a diundecanoyl phosphatidylcholine (PDB, 1lpa). The monolayer shown in Figure 7 was constructed by manually positioning the phosphatidylcholine molecules in the mouth of the cavity, being careful to place phosphate groups close to the positively charged sidechains that are present at the outer ring of the cavity. A round of conjugate gradient energy minimization was then carried out using X-PLOR with fixed model protein coordinates and with the electrostatic terms turned off. This was followed by minor lipid relocation and further refinement.

Other methods and computer programs

Secondary structure assignments were done using the protocol of Kabsch and Sander [50] as implemented in Promotif [17]. The interhelical angles listed were calculated by Promotif. Solvent-accessible calculations were done using GRASP with a 1.4 solvent probe radius [51].

Research Article Structure of crystalline lipovitellin Anderson, Levitt and Banaszak

909

17. Hutchinson, E.G. & Thornton, J.M. (1996). PROMOTIF a program to identify and analyze structural motifs in proteins. Protein Sci. 5, 212-220. 18. Thunnissen, A.-M.W.H., et al., & Dijkstra, B.W. (1994). Doughnutshaped structure of a bacterial muramidase revealed by X-ray crystallography. Nature 367, 750-753. 19. Wilson, C., Wardell, M.R., Weisgraber, K.H., Mahley, R.W. & Agard, D.A. (1991). Three-dimensional structure of the LDL receptor-binding domain of human apolipoprotein E. Science 252, 1817-1822. 20. Breiter, D.R., et al., & Holden, H.M. (1991). Molecular structure of an apolipoprotein determined at 2.5 resolution. Biochemistry 30, 603-608. 21. Colovos, C. & Yeates, T.O. (1993). Verification of protein structures: patterns of nonbonded atomic interactions. Protein Sci. 2, 1511-1519. 22. Birrell, G.B., Anderson, P.B., Jost, P.C., Griffith, O.H., Banaszak, L.J. & Seelig, J. (1982). Lipid environments in the yolk lipoprotein system. A spin-labeling study of the lipovitellin/phosvitin complex from Xenopus laevis. Biochemistry 21, 2444-2452. 23. Auld, D.S., Falchuk, K.H., Zhang, K., Montorzi, M. & Vallee, B.L. (1996). X-ray absorption fine structure as a monitor of zinc coordination sites during oogenesis of Xenopus laevis. Proc. Natl Acad. Sci. USA 93, 3227-3231. 24. Redshaw, M.R., Follett, B.K. & Lawes, G.J. (1971). Biliverdin: a component of yolk proteins in Xenopus laevis. Int. J. Biochem. 2, 80-84. 25. Davis, R.A. (1997). Evolution of processes and regulators of lipoprotein synthesis: from birds to mammals. J. Nutr. 127, 795S-800S. 26. Altschul, S.F., Gish, W., Miller, W., Myers, E.W. & Lipman, D.J. (1990). Basic local alignment search tool. J. Mol. Biol. 215, 403-410. 27. Coulson, A. (1996). The Caenorhabditis elegans genome project. C. elegans genome consortium. Biochem. Soc. Trans. 24, 289-291. 28. Atzel, A. & Wetterau, J.R. (1993). Mechanism of microsomal trigylceride transfer protein catalyzed lipid transport. Biochemistry 32, 10444-10500. 29. Atzel, A. & Wetterau, J.R. (1994). Identification of two classes of lipid molecules binding sites on the microsomal triglyceride transfer protein. Biochemistry 33, 15382-15388. 30. Graham, D.L., Knott, T.J., Jones, T.C., Pease, R.J., Pullinger, C.R. & Scott, J. (1991). Carboxyl-terminal truncation of apolipoprotein B results in gradual loss of the ability to form buoyant lipoproteins in cultured human and rat liver cell lines. Biochemistry 30, 5616-5621. 31. Franzen, J.S., Bobik, C.M., Stuchell, R.N. & Lee, L.D. (1968). Polypeptide subunits of lipovitellin. Arch. Biochem. Biophys. 123, 127-132. 32. Lange, R.H. (1984). Lattice parameters as revealed by electron microscopy and a comparison between lipoprotein crystals from cyclostome eggs formed in vivo and in vitro. J. Mol. Biol. 179, 765-768. 33. Abraham, G.N. & Podell, D.N. (1981). Pyroglutamic acid: nonmetabolic formation, function in proteins and peptides, and characteristics of the enzymes effecting its removal. Mol. Cell Biochem. 38, 181-190. 34. Wang, B.-C. (1985). Resolution of phase ambiguity in macromolecular crystallography. Methods Enzymol. 115, 90-112. 35. Levitt, D.G. & Banaszak, L.J. (1993). A new routine for thinning, editing, and fitting MIR maps using real-space molecular dynamics. J. Appl. Cryst. 26, 736-745. 36. Jones, T.A., Zou, J.-Y., Cowan, S.W. & Kjeldgaard, M. (1991). Improved methods for building protein models in electron density maps and the location of errors in these models. Acta. Cryst. A 47, 110-119. 37. Brnger, A.T. (1992). X-PLOR v. 3.1 Manual. Yale University, New Haven, CT, USA. 38. Brnger, A.T. (1992). The free R value: a novel statistical quantity for assessing the accuracy of crystal structures. Nature 355, 472-474. 39. Engh, R.A. & Huber, R. (1991). Accurate bond and angle parameters for X-ray protein structure refinement. Acta Cryst. A 47, 392-400. 40. Brnger, A.T., Krukowski, A. & Erickson, J. (1990). Slow-cooling protocols for crystallographic refinement by simulated annealing. Acta Cryst. A 46, 585-593. 41. Otwinoski, Z. (1991). Maximum likelihood refinement of heavy atom parameters. In Isomorphous Replacement and Anomalous Scattering (Wolf, W., Evans, P.R. & Leslie, A.G.W., eds), pp. 80-86, Science & Engineering Research Council Daresbury Laboratory, Warrington, UK. 42. Cura, V., Krishnaswamy, S. & Podjarny, A.D. (1992). Heavy-atom refinement against solvent-flattened phases. Acta Cryst. A 48, 756-764. 43. CCP4 (1994). The CCP4 suite: programs for protein crystallography. Acta Cryst. D 50, 760-763. 44. Cowtan, K. (1994). DM: an automated procedure for phase improvement by density modification. In Joint CCP4 and ESFEACBM Newsletter on Protein Crystallography. 31, 34-38.

45. Read, R.J. (1986). Improved fourier coefficients for maps using phases from partial structures with errors. Acta Cryst. A 42, 140-149. 46. Kleywegt, G.J. (1995). Dictionaries for heteros. ESF/CCP4 Newslett. 31, 45-50. 47. Laskowski, R.A., MacArthur, M.W., Moss, D.S. & Thornton, J.M. (1993). PROCHECK: a program to check the stereochemical quality of a protein structure. J. Appl. Cryst. 26, 283-291. 48. Luthy, R., Bowie, J.U. & Eisenberg, D. (1992). Assessment of protein models with three-dimensional profiles. Nature 356, 83-85. 49. Insight II User Guide, (1995). Biosym/MSI, San Diego, CA. 50. Kabsch, W. & Sander, C. (1983). Dictionary of protein secondary structures: pattern recognition of hydrogen-bonded and geometrical features. Biopolymers 22, 2577-2637. 51. Nicholls, A., Bharadwaj, R. & Honig, B. (1993). GRASP: a graphical representation and analysis of surface properties. Biophys. J. 64, A166. 52. Barton, G.J. (1993). ALSCRIPT: a tool to format multiple sequence alignments. Protein Eng. 6, 37-40. 53. Evans, S.V. (1993). SETOR: hardware-lighted three-dimensional solid model representations of macromolecules. J. Mol. Graph. 11, 134-138.