The CRIPTO/FRL-1/CRYPTIC (CFC) domain of human Cripto Functional and structural insights through disulfide structure analysis Susan F. Foley, Herman W. T. van Vlijmen, Raymond E. Boynton, Heather B. Adkins, Anne E. Cheung, Juswinder Singh, Michele Sanicola, Carmen N. Young and Dingyi Wen Biogen, Inc., Cambridge Center, Cambridge, MA, USA The disulfide structure of the CRIPTO/FRL-1/CRYPTIC (CFC) domain of human Cripto protein was determined by a combination of enzymatic and chemical fragmentation, followed by chromatographic separation of the fragments, and characterization by mass spectrometry and N-terminal sequencing. These studies showed that Cys115 forms a disulfide bond with Cys133, Cys128 with Cys149, and Cys131 with Cys140. Protein database searching and molecular modeling revealed that the pattern of disulfide linkages in the CFC domain of Cripto is the same as that in PARS intercerebralis major Peptide C (PMP-C), a serine protease inhibitor, and that the EGF-CFC domains of Cripto are predicted to be structurally homologous to the EGF-VWFC domains of the C-terminal extracellular portions of Jagged 1 and Jagged 2. Biochemical studies of the interactions of ALK4 with the CFC domain of Cripto by fluorescence-activated cell sorter analysis indicate that the CFC domain binds to ALK4 independent of the EGF domain. A molecular model of the CFC domain of Cripto was constructed based on the nuclear magnetic resonance structure of PMP-C. This model reveals a hydrophobic patch in the domain opposite to the presumed ALK4 binding site. This hydrophobic patch may be functionally important for the formation of intra or intermolecular complexes. Keywords: Cripto; disufide structure; CFC domain; model. Cripto is a member of a family of proteins that includes human Cripto and Criptic, murine Cripto and Criptic, frog FRL-1, zebrafish one-eyed pinhead protein (oep) and chick Cripto [1–3]. The involvement of these proteins in early embryonic development is well established [1,2,4–11], and other recent investigations indicate that Cripto is over- expressed in a number of human cancers [2]. These proteins are characterized by two cysteine-rich structural motifs: an epidermal growth factor (EGF)-like domain and a CRI- PTO/FRL-1/Cryptic (CFC) domain, the latter of which is considered unique to this family. Previous characterization of human recombinant Cripto (residues 1–169) showed that mature protein begins at Leu31, that Asn79 is N-glycosyl- ated with >90% occupancy, Ser40 and Ser161 are partially O-glycosylated [2], and Thr88 is modified with a single O-linked fucose [12]. Ser161 is the predicted x-site for propeptide cleavage and glycosylphosphatidylinositol (GPI) attachment, and the segment comprising residues 170–188 is the predicted signal peptide for GPI-anchorage (Fig. 1) [1]. Evidence of an EGF-like domain structure in Cripto-related proteins is based on amino acid sequence homology [1,2,4,13,14], molecular modeling [1,14], and gene structure [1,2,4,14]. The CFC region has no predictive model for disulfide linkage of its six cysteines. The role of the EGF-CFC family of proteins in embryogenesis is still being elucidated, but information to date suggests that Cripto is required for Nodal binding to the ActRIIB/ALK4 receptor complex and for Nodal activation of similar to mothers against decapentaplegic peptide-2 (Smad-2) [15–18]. Moreover, point mutation experiments with Cripto have shown that the EGF domain is necessary for binding to Nodal and the CFC domain is responsible for binding to ALK4 [16,17,19]. A naturally occurring Pro125 fi Leu mutation in the CFC domain of Cripto has been correlated with developmental anomalies in the midline and forebrain in human fetuses, and an engineered construct with the same Pro125 fi Leu muta- tion was inactive in a rescue model of the oep phenotype in zebrafish [19]. These findings highlight the biological importance of Cripto and underline the functional signi- ficance of the CFC domain. In the present work, we have solved the disulfide structure of the CFC domain and have conducted biochemical studies that detail the interactions of ALK4 with the CFC domain. From molecular modeling studies, we have shown that the CFC domain of Cripto is structurally homologous to the von Willebrand Factor type C-like domain and that Cripto protein is structurally similar to the C-terminal extracellular portions of Jagged 1 and Jagged 2. Correspondence to Dingyi Wen, Biogen, Inc., 14 Cambridge Center, Cambridge, MA 02142, USA. Fax: + 1 617 679 2616, Tel.: +1 617 679 2362. E-mail: Dingyi_Wen@biogen.com Abbreviations: CFC, CRIPTO/FRL-1/Cryptic; PMP-C, PARS inter- cerebralis major Peptide C; oep, zebrafish one-eyed pinhead protein; EGF, epidermal growth factor; LTbR, Lymphotoxinb Receptor; FACS, fluorescence-activated cell sorter; PTH, phenylthiohydantoin; Cripto delC-Fc, Cripto (amino acids 1–169) fused to the hinge and Fc region of human IgG1; NEM, N-ethylmaleimide; IAM, iodo- acetamide; NES, 2-(N-ethylsuccinimidyl); ESI, electrospray ionization; VWFC, von Willebrand Factor C domain; GPI, glycosylphosphatidylinositol. (Received 7 May 2003, revised 3 July 2003, accepted 11 July 2003) Eur. J. Biochem. 270, 3610–3618 (2003) Ó FEBS 2003 doi:10.1046/j.1432-1033.2003.03749.x Experimental procedures Protein expression and purification Recombinant human Cripto-1 was expressed in Chinese hamster ovary cells as a C-terminally truncated form, comprising amino acid residues 1–169, and was purified by immunoaffinity chromatography on an anti-Cripto mAb column [12]. Fluorescence-activated cell sorter (FACS) analysis Analysis of Cripto–ALK4 interactions by flow cytometry was performed essentially as described earlier [19]. Briefly, human 293 cells were transfected with plasmids expressing human ALK4 (provided by M. Whitman, Harvard Medical School), full length wild-type Cripto, Cripto (N85G/T88A) or Cripto (H120G/W123G) using Fugene (Roche) according to the manufacturer’s instructions. After 48 h, cells were processed for flow cytometry. Approximately 5 · 10 5 cells were incubated with 10 lgÆmL )1 of either human Cripto delC-Fc, CFC-Fc, EGF-Fc, LTbR (Lymphotoxinb Receptor)-Fc or human ALK4-Fc (R&D Systems) followed by a PE (phycoery- thrin)-conjugated anti-human Fc secondary Ig (Jackson Immunoresearch). Estimation of free thiol groups Approximately 50 lg of human Cripto (residues 74–169, which covers the combined EGF-CFC domains) was incubated in 100 m M N-ethylmeleimide (NEM), 6 M guanidine HCl, 72 m M Mes, pH 6.0, at 37 °Cfor1h. The control sample was incubated in the same way, but without NEM. Samples were desalted using ethanol precipitation [20]. This treatment was followed by complete reduction of cystines with 30 m M dithiothreitol in 8 M guanidine HCl, 150 m M Tris HCl, pH 8.5, at 45 °C for 45 min, followed by alkylation with 75 m M iodoacetamide (IAM) at room temperature for 1 h. After desalting using ethanol precipitation, the sample was deglycosylated with PNGase F (Glyko) in 2 M urea, 50 m M sodium phosphate, pH 7.6, 20 m M methylamine HCl, 5 m M EDTA, at 37 °C overnight. Intact mass was measured on-line using a ZMD (electrospray ionization) mass spectrometer (Waters). The molecular masses were generated by deconvolution with the MAXENT 1 program. Generation of disulfide-linked CFC fragments of Cripto using CNBR and endoproteinase lys-c Nonreduced Cripto was treated with 1 M CNBr in 70% formic acid at room temperature for 36 h. To remove residual CNBr, the treated sample was dried under vacuum in a SpeedVacÒ concentrator, suspended in HPLC-grade water and dried again. The water wash and drying steps were repeated, after which the final pellet was dissolved in 200 m M Mes, pH 6.0. Approximately 60 lg of protein in 160 m M Mes, pH 6.0, 20% 2-propanol, was digested with 6 lg of endoproteinase Lys-C (WAKO) at room tempera- ture for 24 h. An additional 6 lg of the enzyme was added and the sample again incubated for 24 h. Solid guanidine HCl was added to a final concentration of 6 M to quench the reaction. Separation of disulfide-linked CFC fragments of Cripto Fragments of Cripto were separated by reverse-phase HPLC (rp-HPLC) on a 1-mm · 250-mm Vydac C 4 column using a Waters Separation Module, Model 2690. Solvent A was 0.1% (v/v) trifluoroacetic acid in water and solvent B was 0.08% (v/v) trifluoroacetic acid in 75% acetonitrile. A linear gradient from 0 to 60% acetonitrile over 160 min was applied at flow rate of 0.07 mLÆmin )1 .Thecolumn temperature was 30 °C. Fractions were collected at 1-min intervals. Fractions that were determined by mass analysis to be enriched in the CFC-containing fragment were pooled and concentrated to dryness under vacuum. The pellet was dissolved in 160 m M Mes, pH 6.0, 20% (v/v) 2-propanol, 1m M CaCl 2 . Digestion was carried out at room tempera- ture using the following regimen: 1.8 lg of thermolysin was added at time 0 and after 24 h, and then 1.8 lgof endoproteinase Lys-C was added after 48 h. The progress of the digestion reaction was monitored by MALDI- TOF MS. For this purpose, aliquots were removed after 24, 48 and 72 h and desalted using Millipore C 18 ZIP TIPs TM with or without reduction with 25 m M dithiothreitol in 4 M guanidine HCl, 80 m M Tris HCl, pH 8.5, at room tem- perature for 1–2 h prior to mass analysis. The enzymatic digest was stopped by the addition of solid urea to 6 M .The thermolysin and endoproteinase Lys-C digests were separ- ated on an rp-HPLC Vydac C 18 column using a 100-min linear gradient of 0–75% acetonitrile. Solvent A was 0.03% (v/v) trifluoroacetic acid in water, and solvent B was 0.024% (v/v) trifluoroacetic acid in 75% (v/v) acetonitrile. One- minute fractions were collected and concentrated to dryness under vacuum, and the residue was resuspended in 5 lLof 0.1% (v/v) trifluoroacetic acid, 30% (v/v) acetonitrile. Peptide analysis using MALDI-TOF MS MALDI-TOF MS was carried out on a Voyager-DE TM STR mass spectrometer (Applied Biosystems) in either linear or reflector mode using a-cyano-4-hydroxy cinnamic acid as matrix. Results generated using the linear mode are expressed as average, protonated masses, those collected in the reflector mode, as protonated, monoisotopic masses. An aliquot of 0.5 lLor1lL of each test sample was applied to the target plate. After partial evaporation of the sample droplet at room temperature, 0.5 lL of matrix (10 mgÆmL )1 Fig. 1. The predicted sequence of human Cripto-1 encoded from DNA is shown: the signal peptide is italicized and corresponds to residues 1–30. Cripto del-C covers residues 31–169, with the CFC domain in bold. The arrow indicates the predicted x-site for propeptide cleavage and GPI-attachment; the signal region for GPI anchorage is underlined. Ó FEBS 2003 Structure of the human Cripto CFC domain (Eur. J. Biochem. 270) 3611 in 50% acetonitrile/0.1% trifluoroacetic acid, v/v) was applied. Data acquisition and analysis were controlled by GRAMS /32 software (version 4.11, Level 2). N-terminal sequencing Sequencing was carried out on an Applied Biosystems Procise 494 cLC sequencer that was run in the pulsed liquid mode. The resulting PTH (phenylthiohydantoin) amino acids were separated using an ABI 140D Solvent Delivery System with a 0.8-mm · 250-mm, C 18 PTH column and were monitored on-line using an ABI 785A programmable absorbance detector. Data were analyzed using the ABI 610 A data analysis software. Homology search A BLAST search [21] of the SWISSPROT database was carried out using the primary sequence of the CFC region (residues 113–154) of human Cripto as the query motif. Disulfide pattern search The experimentally defined disulfide bond pattern of the CFC region of CRIPTO was used to query an in-house disulfide database built from annotations in SWISSPROT. The search method reports all proteins with the same disulfide topology, e.g. C1-C4/C2-C6/C3-C5 (C1 is the first cysteine in the domain, C2 is the second, C3 is the third, etc.), and ranks them according to sequence spacing between the cysteines. Comparative modeling The 3-D structures of the EGF-like domain and the CFC domain were modeled separately using the MODELER module [22] of the INSIGHT II software package (Accelrys, Inc., San Diego, CA, USA). The NMR structure of mouse EGF [Protein Data Bank code (pdb): 1epi] was used for modeling of the EGF-like domain. The CFC domain was built using the NMR structure of the proteinase inhibitor, PMP-C (pdb: 1pmc). Motif search Motif elements, identified independently by homology or disulfide pattern, were used to query the nonredundant database TREMBL using the DART program (a domain motif search algorithm). Results The Cripto CFC domain is a functional unit The predicted amino acid sequence (Fig. 1) of mature human Cripto contains 12 cysteine residues, six in the EGF domain and six in the CFC domain. To test whether the CFC domain could retain its function independent of the EGF domain, we generated a soluble form of the CFC domain, comprised of the signal peptide and amino acids 112–169, fused to the hinge and Fc region of human IgG1 (CFC-Fc), and tested its ability to bind to ALK4. Previously, we showed in a FACS assay that full length Cripto (amino acids 1–169) fused to human Fc (Cripto delC-Fc) bound to human 293 cells expressing ALK4, but not to control 293 cells lacking ALK4 [19]. We have now evaluated the binding of soluble CFC-Fc to ALK4-293 expressing cells by FACS assay, using Cripto delC-Fc as the positive control and LTbR-Fc as a negative control. Figure 2A shows the results of this comparison. A signifi- cant shift in mean fluorescence for ALK4-293 cells was seen in the presence of either Cripto delC or CFC-Fc, but not with LTbR-Fc (Fig. 2A2). A small shift was also seen with EGF-Fc, but this shift was not dependent on ALK4 expression. These experiments also show that the shift in mean fluorescence for CFC-Fc (Figs 2 and 3) binding to ALK4-293 cells is of similar magnitude as the shift of the positive control, Cripto delC-Fc (Fig. 2A1), and therefore that the CFC domain is sufficient for the interaction of Cripto with ALK4. To verify the role of the CFC domain in ALK4 binding, we analyzed the effects of point mutations in the EGF and CFC domains by FACS analysis, using mutations in the EGF and CFC domains known to disrupt function, specifically either downstream signaling or ALK4 binding [12,16]. We have also compared the ability of both types of mutants, i.e. the CFC domain mutant, H120G/W123G, and the EGF domain mutant, N85G/T88A, to bind to ALK4 Fig. 2. FACS analysis of the interactions between Cripto and ALK4. (A) Incubation of soluble Cripto delC-Fc (A1), EGF-Fc (A2), and CFC-Fc (A3) with 293 cells expressing ALK4. The cells expressing ALK4 (bold, solid curve) are compared to the control cells that do not express ALK4 (solid curve). Incubation of LTbR-Fc with 293 cells expressing ALK4 was used as a control for the Fc portion of the proteins (dashed curve). (B) Incubation of ALK4-Fc with 293 cells expressing full length wild-type Cripto (B1), Cripto N85G/T88A (B2), or Cripto H120G/W123G (B3). Cells expressing Cripto or mutants (bold, solid curve) are compared to the control cells that do not express any Cripto proteins (solid curve). 3612 S. F. Foley et al. (Eur. J. Biochem. 270) Ó FEBS 2003 by FACS (Fig. 2B). The results showed that ALK4-Fc binds well to cells expressing either wild type Cripto (Fig. 2B1) or the EGF domain mutant, N85G/T88A (Fig. 2B2), but does not bind to cells expressing the CFC mutant, H120G/W123G (Fig. 2B3). This and the previous experiments demonstrate that the CFC domain is involved in ALK4 binding. Determination of disulfide linkages in the CFC domain Determination of whether there are free thiol groups in the protein was done by alkylation of the protein with NEM under nonreducing conditions followed by alkylation with IAM under reducing conditions. Alkylation of a cysteine with NEM will add 125.1 Da to the mass of the protein or peptide, whereas alkylation with IAM will add a mass of 56.9 Da. The results from ESI mass spectrometric analysis showed a range of molecular masses corresponding to residues 74–169 completely alkylated with IAM, with heterogeneity in glycosylation. Masses corresponding to protein containing 2-(N-ethylsuccinimidyl)-cysteine (NES- Cys) residues were not detected. Therefore, we conclude that all of the cysteine residues in the protein are disulfide-linked. To study the disulfide structures of the CFC domain, a double cleavage strategy was developed using CNBr treatment followed by endoproteinase Lys-C cleavage. This strategy took advantage of a Lys residue (Lys112) between the EGF-like domain and the CFC domain and a Met residue (Met154) between the last Cys in the CFC domain and the O-linked glycosylation site at residue Ser161. The dual digest was then separated by rpHPLC and the fractions containing the CFC domain were identified by MALDI- TOF MS and were pooled for further analysis (see below). In the CFC region, there are three Lys residues that might be cleaved by endoproteinase Lys-C and two Trp residues that could be oxidized during CNBr treatment [23]. Additional cleavage can take place on the C-terminal side of oxidized Trp [24]. The observed protonated mass (MH + ) of the major component in the pooled fractions containing the CFC domain was 4702.4 Da (Fig. 3), which is consis- tent with fragments having either two oxidized Trp residues and one cleavage at a Lys residue or one oxidized Trp and cleavages at two of the Lys residues. In-source fragment ions, MH + ¼ 1599.5 Da and MH + ¼ 3105.9 Da (Fig. 3) indicated that the 4702-Da component was derived mainly from the peptides 113–126 (calculated m/z ¼ 1599.8) and 127–154 (calculated m/z ¼ 3106.8), linked by a disulfide bond. A minor component generated by an additional cleavage after oxidized Trp123 (calculated MH + ¼ 4365.12 Da, based on disulfide linked peptides 113–123 and 127–154) was also identified (Fig. 3). The pooled fractions were analyzed by MALDI-TOF MS after reduc- tion also. The results support the identification of the CFC peptides predicted from in-source fragmentation. N-terminal sequencing results also supported this inter- pretation (data not shown). The CFC domain-containing fractions were further digested with thermolysin, followed by endoproteinase Lys-C. Twenty percent propanol was added to the digest to promote preferential cleavages by thermolysin at the N-terminus of leucine, isoleucine, and phenylalanine [25]. The extent of proteolytic cleavage between cysteine residues was monitored by MALDI-TOF MS after reduction (data not shown). Figure 4 shows the mass spectrum of the nonreduced digest after all enzyme treatment. For the sake of simplicity, we use C1 for the first cysteine residue in the CFC domain, C2 for the second, C3 for the third, etc. We will use this nomenclature in the following discussion. Interpretation of the data for the nonreduced sample is supported by identification of the peptides necessary to form the predicted disulfide bonds. For example, mass signal detected at m/z ¼ 2243.1 was interpreted as a disulfide-linked component composed of peptides 113–126 [C1 (Cys115)] and 133–137 [C4 (Cys133)] (Fig. 4). Corres- ponding peptide 113–126 (m/z cal ¼ 1598.7) and peptide 133–137 (m/z cal ¼ 646.2) were detected both under reducing Fig. 4. MALDI-TOF mass spectrum of the nonreduced CFC domain after all enzymatic treatments. The spectrum was derived in the reflector mode and all masses correspond to protonated monoisotopic mass. Enzyme fragment peaks are identified with asterisks and in-source fragments are underlined. Fig. 3. MALDI-TOF mass spectrum of the CFC domain-containing fractions under nonreducing conditions. Peptide a, ENCGSVPHD TW OX LPK; peptide b, ENCGSVPHDTW OX and peptide c, KCSLC KCW OX HGQLRCFPQAFPQAFLPGCDGLVM. The spectrum was obtained in the linear mode and all masses correspond to protonated average masses. Oxidized Trp residues are represented as W OX and the Met residue converted to homoserine lactone is in italics. Masses corresponding to intact CFC domain were not present. In-source fragments are indicated with asterisks. Ó FEBS 2003 Structure of the human Cripto CFC domain (Eur. J. Biochem. 270) 3613 conditions and as in-source fragment ions derived from the disulfide-linked peptide (Fig. 4). The mass spectrometric data also clearly demonstrate that C3 (Cys131) forms a disulfide bond with C5 (Cys140) as evidenced by disulfide- linked peptides at masses 957.5, 1066.6, 1123.7, and 1194.7. In addition, certain in-source fragments expected from these disulfide-linked peptides are present, i.e. at 763.4 and 834.5 Da (Fig. 4). As C1 is disulfide-bonded to C4 and C3 is disulfide-bonded to C5, it can be deduced that C2 must be linked to C6, although the corresponding mass was not detected, presumably due to ion suppression. To confirm this deduction, the thermolysin digest was separated by rpHPLC and the peaks were analyzed by MALDI- TOF MS. In one of the major peaks, masses of 1042.5 and 914.5 corresponding to the disulfide-linked peptides FLPGC(6)DG with KC(2)S and FLPGC(6)DG with C(2)S, respectively, were detected. Other disulfide-linked peptides, such as C1-C4 and C3-C5, were also identified by MALDI-TOF MS in different fractions. The fractions containing disulfide-linked peptides were evaluated by N-terminal sequencing. The mass spectrometric and N-terminal sequencing results confirmed that C1 is linked toC4,C2toC6,andC3toC5. Primary structure search The amino acid sequence information for the CFC region of human Cripto was used to carry out a BLAST search of the combined SWISSPROT/TREMBL database. An initial search showed matches to the VWFC (von Willebrand Factor C)-like domain in human and chicken a-1 collagen, mouse and human NELL 2, and chicken NEL, with low homology (e-value > 0.1), in addition to other Cripto and Cripto-like proteins. The VWFC domain is defined by a pattern of 10 cysteine residues of undetermined connectivity, but, the similarity of the Cripto CFC domain to the above-listed proteins is confined to the portion of the VWF-C motif containing the first six cysteine residues. Motif search Subsequent searches of the protein database with DART , using combined EGF-like/VWFC sequences as queries, provided additional matches, some of which are listed in Table 1. Many of the identified proteins contain both EGF- like and VWFC domains, but, only in human Jagged 1 and Jagged 2, Drosophila Serrate, and NELL 1 were both domains adjacent to and in the same order as the putative EGF-like and CFC regions of human Cripto. Furthermore, only in Jagged 2 are the regions adjacent to the membrane interface (transmembrane) region. To examine the strength of this relationship, we aligned the amino acid sequences of human Jagged 2 and Cripto (Fig. 5). The conservation of residues such as cysteine, proline, glycine, and tryptophan, which are important for the protein folding, is highlighted in the alignment [26]. Disulfide pattern search and comparative modeling The disulfide pattern that was determined experimentally for human Cripto (i.e. C1-C4, C2-C6, C3-C5) was used to query a disulfide database compiled from SWISSPROT annotations (van Vlijmen, H. W. T., Gupta, A. & Singh, J., Biogen Corp., unpublished observations). This is an ortho- gonal method for exploring relationships, and revealed two small, structurally related serine protease inhibitors, PMP- D2 and PMP-C [27], that were not uncovered using BLAST on the SWISSPROT/TREMBL database. Based on the NMR structure of PMP-C (Protein Data Bank code, 1pmc) and the sequence alignment shown in Fig. 6, a 3-D model was built for the CFC domain of Cripto (Fig. 7). In the Table 1. Summary of some of the proteins identified as containing VWFC-like domains. Definition of motifs as VWFC-like are based on SWISS- PROT annotations and NCBI DART predictions. Protein name No. of EGF-like domains No. of VWFC-like domains No. of Cys in VWFC-like domains Cripto (human or mouse) 1 1 6 VWF (human and porcine) 0 3 10 NEL (chicken) 6 5 10 in domains 1–4, 8 in domain 5 NELL1 (rat) 6 5 10 in domains 1–4, 8 in domain 5 NELL2 (human and rat) 6 5 10 in domains 1–4, 8 in domain 5 Protein kinase C (BP) 6 5 10 in domains 1–4, 8 in domain 5 Jagged 1 and Jagged 2 (human) 15 1 10 Serrate and Drosophila 14 1 10 Fig. 5. Alignment of the sequences of human Cripto and human Jagged 2. Conserved residues are framed with solid lines and homologous residues are framed with dashed lines. The sequence identity over the alignment length is 26%. 3614 S. F. Foley et al. (Eur. J. Biochem. 270) Ó FEBS 2003 computed model of the Cripto CFC domain, one side of the molecule has a high concentration of hydrophobic residues, including Trp134, Leu138, Phe141, Pro142, Phe145, and Leu146. These hydrophobic residues are on the side of the protein opposite to residues His120 and Trp123 that have been implicated in binding of Cripto to ALK4. The hydrophobic residues may play a role in the folding of Cripto by interacting with the EGF-like domain, or they may constitute the interaction site with other signaling components. A 3-D model of the EGF-like domain of Cripto was also built, based on the NMR structure of murine EGF (Protein Data Bank code, 1epi), by aligning the cysteine residues as described previously [28]. Two theoretical models of the full-length Cripto protein were constructed by connecting the EGF and CFC modules. In the first model (Fig. 7A) the domains are arranged in an extended conformation, analogous to the conformation found for the solution structure of a covalently linked pair of EGF domains from human fibrillin-1 [29]. The second model has a more globular structure in which the EGF and Fig. 7. Model for EGF-CFC domains of Cripto. Hypothetical structures for EGF-CFC domains are shown in an extended conformation (A) or closed conformation (B). In the CFC domain, disulfide bonds Cys115-Cys133, Cys128-Cys149, and Cys131-Cys140 are indicated by DS1, DS2 and DS3, respectively. Residues H120 and W123 have been implicated in Alk4 binding and are shown in purple. Residues N79 and T88 (shown in red) are modified through N-linked glycosylation and O-linked fucosylation, respectively. Residues N79, N85, R104, and E107 (the latter three shown in blue) have been shown to be important in Nodal induction of Smad2 phosphorylation [16]. Nter designates the location of the expressed amino- terminus; Cter designates the location of the expressed carboxyl-terminus. Fig. 6. Alignment of the sequences of the CFC domain of human Cripto and PMP-C. Conserved residues are framed with solid lines and homologous residues are framed with dashed lines. Ó FEBS 2003 Structure of the human Cripto CFC domain (Eur. J. Biochem. 270) 3615 CFC modules have a large number of noncovalent contacts (Fig. 7B), analogous to the crystal packing of the EGF-like domains from human factor IX [30]. We have also modeled the EGF-like (15th EGF domain) and adjacent VWF-C domains of human Jagged 2, using the same approach described for the Cripto EGF-like and CFC domains and found that there are no structural incompatibilities. Discussion We have used chemical and enzymatic fragmentation, mass spectrometry, and N-terminal sequence analysis to charac- terize the disulfide linkages of the cysteine residues in the CFC region of human Cripto. From these studies, we show that the six cysteines are linked in three disulfide bonds, C1-C4, C3-C5, C2-C6. We performed these experiments on a truncated, recombinant version of human Cripto, con- taining residues 31–169 of wild type human Cripto. We consider the results a valid representation of the wild type structure because a seventh Cys residue, Cys181, is located in the predicted GPI Ôsignal sequenceÕ that would normally be cleaved off during processing of the wild type protein [31,32]. Furthermore, it has been demonstrated that this soluble, C-terminally truncated recombinant human Cripto protein is biologically active [2]. Using both the primary sequence of the CFC region and the experimentally defined disulfide pattern to query protein sequence and disulfide databases, we obtained matches to a group of proteins containing a VWFC-like motif. The VWFC-like motif is believed to play an important role in the formation of certain protein complexes, examples including thrombo- spondin 1 (TSP1), which binds to CD36 on endothelial cells [29], and procollagen IIA and chordin, which bind to bone morphogenic protein [33]. The binding properties of these proteins have led to the hypothesis that proteins containing VWFC-like domains (Cys-rich) act as ÔTGFbeta sinksÕ in modulating development [33]. Although most of the docu- mented VWFC-like motifs contain 10 cysteine residues, there are several instances where such regions have fewer than 10, e.g. the C-terminal VWFC domain of NEL (chicken), NELL 1 (rat) and NELL 2 (human and rat), and the last VWFC region of murine tectorin – all of which contain only eight cysteine residues. In all of the examples of proteins containing shortened VWFC-like domains, the motif is abbreviated by loss of the C-terminal region, covering residues Cys9 and Cys10. These observations suggest that the CFC region in Cripto can be considered as a truncated form of the VWFC-like domain. Assuming that the CFC region of Cripto is VWFC-like, we infer that the EGF-CFC family of proteins is a variation of an already described theme for which there are many examples in modular proteins. Among them are several that have a juxtaposition of the EGF and CFC domains seen in Cripto, forexample,NELL1,NELL2,JAGGED1and JAGGED 2, in which at least one of the EGF-like domains is N-terminal to a VWFC-like domain (Table 1). For JAGGED 1 and JAGGED 2, the similarity extends to the position of the membrane attachment sequence, specifically, a trans-membrane domain that is C-terminal to the VWFC-like domain, and we found that there was a striking degree of structural similarity between Cripto EGF-CFC and human JAGGED 2 (Fig. 5). As with Cripto, human JAGGED 2 is involved in signal transduction as a ligand for the NOTCH receptor, another EGF homolog [34]. Moreover, similar to Cripto, a major function of JAGGED 2 is in patterning and morphogenesis in early embryonic development [35,36]. Although JAGGED 2 is not fucosylated as Cripto is [2,12], the function of NOTCH ligand is reportedly regulated by fucosylation of the Notch receptor [35]. The specific role of the individual domains of human JAGGED has not been delineated, but Serrate, the Drosophila version of JAGGED, has been investigated. Hukriede et al. [37] have shown that a truncated form of Serrate, lacking the VWFC region [38], binds to NOTCH but does not activate NOTCH signaling. The functions of the domains in Cripto are still being investigated, but initial information published previously by Yeo et al. [16] and described here indicate that the EGF and CFC domains have different functions. Yeo et al. showed that ALK4 was coimmunoprecipitated with the CFC domain of murine Cripto, but not with the CFC mutant (H120G/W123G) [16]. Here we have confirmed and expanded upon these findings using ALK4 and human Cripto, and have demon- strated that the CFC domain alone is sufficient for ALK4 binding. These experiments highlight the important role of the CFC domain, like other VWFC-domains [29,33], in complex formation. Recently, Minchiotti et al. [7] postulated a structural model of human Cripto based on the beta-trefoil fold of basic FGF. In this model, the EGF-like and CFC domains form the second and third lobes of the trefoil structure, respectively. We now believe this model to be incorrect because it cannot accommodate the actual disulfide connectivities in the CFC domain of Cripto described here. Using our experimentally determined disulfide pat- tern in the CFC domain to search a disulfide database compiled from SWISSPROT, we identified a structurally known homologue, chymotrypsin inhibitor PMP-C. Because of amino acid sequence similarities and disulfide linkage identity between the Cripto CFC domain and PMP-C, we built a model of the Cripto CFC domain using the NMR solution structure of PMP-C as a template (Fig. 7). Our model is consistent with data from previous functional studies [7,16,19], as well as from the current study, in particular, the observation that mutations in the CFC domain at His120 and Trp123 abolish ALK4 interactions (Fig. 2B). In our model (Fig. 7), the side- chains of His120 and Trp123 are solvent-exposed, allowing for possible protein–protein interaction. Interestingly, in our CFC model, we have identified a hydrophobic patch consisting of Trp134, Leu138, Phe141 and Pro142. Leu138 and homologues of Trp134 and Phe141 are conserved throughout the Cripto family [1] and are clustered on the side of the CFC domain opposite the presumed ALK4 binding site (which includes His120 and Trp123). This hydrophobic patch may be important for protein–protein interactions. Two possible structural models for full-length Cripto protein – a linear (open) configuration (Fig. 7A) and a closed configuration (Fig. 7B) – have been constructed by connecting an EGF-like module [28] and our CFC module (Fig. 7). However, at this point we do not have enough data to favor one model over the other. Both models fulfill the 3616 S. F. Foley et al. (Eur. J. Biochem. 270) Ó FEBS 2003 predictions for the structure of the EGF-like domain, namely, solvent exposure of the fucosylation site at Thr88 and the N-linked glycosylation site at Asn79, and both allow for potential protein–protein interactions via the above-described hydrophobic patch. Structure determin- ation of human Cripto by NMR is in progress to address these questions. In summary, the disulfide bond pattern for the six cysteine residues in the CFC domain of human Cripto has been experimentally defined as C1-C4, C2-C6, C3-C5, and biochemical studies have shown that the CFC domain binds to ALK4 independent of the EGF domain. Database searches based on the primary sequence have uncovered similarities between Cripto EGF-CFC domains and the EGF-VWFC domains of the C-terminal extracellular portions of Jagged 1 and Jagged 2. A 3-D structural model of the CFC domain was constructed based on the NMR structure of PMP-C, a serine protease inhibitor having the same disulfide connectivity. This model revealed a hydro- phobic patch that is probably important for protein binding. Two possible models for intact Cripto have also been proposed. By exploring the structural features of Cripto, as defined by our models, we hope to increase the understand- ing of the role of Cripto in the Nodal signal transduction pathway. Acknowledgements We would like to thank Dr R. Blake Pepinsky for his review and editing of this manuscript. We would also like to thank Dr Joseph Rosa, Drs Kevin Williams, Alphonse Galdes, and Alex Buko for their valuable insights. References 1. Colas, J.F. & Schoenwolf, G.C. (2000) Subtractive hybridization identifies chick-cripto, a novel EGF-CFC ortholog expressed during gastrulation, neurulation and early cardiogenesis. Gene 255, 205–217. 2. Saloman, D.S., Bianco, C., Ebert, A.D., Khan, N.I., De Santis, M., Normanno, N., Wechselberger, C., Seno, M., Williams, K., Sanicola, M., Foley, S., Gullick, W.J. & Persico, G. (2000) The EGF-CFC family: novel epidermal growth factor-related proteins in development and cancer. Endocr. Relat. Cancer 7, 199–226. 3. Bamford,R.N.,Roessler,E.,Burdine,R.D.,Saplakoglu,U.,dela Cruz, J., Splitt, M., Towbin, J., Bowers, P., Ferrero, G.B., Marino, B.,Schier,A.F.,Shen,M.M.,Muenke,M.&Casey,B.(2000) Loss-of-function mutations in the EGF-CFC gene CFC1 are associated with human left-right laterality defects. Nat. Genet. 26, 365–369. 4. Dono, R., Scalera, L., Pacifico, F., Acampora, D., Persico, M.G. & Simeone, A. (1993) The murine cripto gene: expression during mesoderm induction and early heart morphogenesis. Development 118, 1157–1168. 5. Ding, J., Yang, L., Yan, Y.T., Chen, A., Desai, N., Wynshaw- Boris, A. & Shen, N.N. (1998) Cripto is required for correct orientation of the anterior-posterior axis in the mouse embryo. Nature 395, 702–707. 6. Gritsman, K., Zhang, J., Cheng, S., Heckscher, E., Talbot, W.S. & Schier, A.F. (1999) The EGF-CFC protein one-eyed pinhead is essential for nodal signaling. Cell 97, 121–132. 7. Minchiotti, G., Manco, G., Parisi, S., Lago, C.T., Rosa, F. & Persico, M.G. (2001) Structure-function analysis of the EGF-CFC family member Cripto identifies residues essential for nodal sig- nalling. Development 128, 4501–4510. 8. Saloman, D.S., Bianco, C. & De Santis, M. (1999) Cripto: a novel epidermal growth factor (EGF) -related peptide in mammary gland development and neoplasia. Bioessays 21, 61–70. 9. Schier, A.F., Neuhauss, S.C., Helde, K.A., Talbot, W.S. & Driever, W. (1997) The one-eyed pinhead gene functions in mesoderm and endoderm formation in zebrafish and interacts with no tail. Development 124, 327–342. 10. Xu, C., Liguori, G., Persico, M.G. & Adamson, E.D. (1999) Abrogation of the Cripto gene in mouse leads to failure of post- gastrulation morphogenesis and lack of differentiation of cardio- myocytes. Development 126, 483–494. 11. Zhang, J., Talbot, W.S. & Schier, F. (1998) Positional cloning identifies zebrafish one-eyed pinhead as a permissive EGF-related ligand required during gastrulation. Cell 92, 241–251. 12. Schiffer, S.G., Foley, S.F., Kaffashan, A., Hronowski, X., Zichittella, A.E., Yeo, C.Y., Miatkowski, K., Adkins, H.B., Domon. B., Whitman, M., Salomon, D., Sanicola, M. & Williams, K.P. (2001) Fucosylation of Cripto is required for its ability to facilitate nodal signaling. J. Biol. Chem. 276, 37767–37777. 13. Ciccodicola, A., Dono, R., Obici, S., Simeone, A., Zollo, M. & Persico, M.G. (1989) Molecular characterization of a gene of the ÔEGF familyÕ expressed in undifferentiated human NTERA2 teratocarcinoma cells. EMBO J. 8, 1987–1991. 14. Shen, M.M., Wang, H. & Leder, P. (1997) A differential display strategy identifies Cryptic, a novel EGF-related gene expressed in the axial and lateral mesoderm during mouse gastrulation. Development 124, 429–442. 15. Reissmann, E., Jornvall, H., Blokzijl, A., Andersson, O., Chang, C., Minchiotti, G., Persico, M.G., Ibanez, C.F. & Brivanlou, A.H. (2001) The orphan receptor ALK7 and the Activin receptor ALK4 mediate signaling by Nodal proteins during vertebrate develop- ment. Genes Dev. 15, 2010–2022. 16. Yeo, C.Y. & Whitman, M. (2001) Nodal signals to Smads through Cripto-dependent and Cripto-independent mechanisms. Mol. Cell 7, 949–957. 17. Bianco, C., Adkins, H.B., Wechselberger, C., Seno, M., Normanno,N.,DeLuca,A.,Sun,Y.,Khan,N.,Kenny,N., Ebert, A., Williams, K.P., Sanicola, M. & Salomon, D. (2002) Cripto-1 activates nodal- and ALK4-dependent and – independ- ent signaling pathways in mammary epithelial Cells. Mol. Cell Biol. 22, 2586–2597. 18. Yan, Y., Liu, J., Luo, Y.E.C., Haltiwanger, R.S., Abate-Shen, C. & Shen, M.M. (2002) Dual roles of Cripto as a ligand and coreceptor in the nodal signaling pathway. Mol. Cell. Biol. 22, 4439–4449. 19. De la Cruz, J.M., Bamford, R.N., Burdine, R.D., Roessler, E., Barkovich, A.J., Donnai, D., Schier, A.F. & Muenke, M. (2002) A loss-of-function mutation in the CFC domain of TDGF1 is associated with human forebrain defects. Hum. Genet. 110, 422–428. 20. Pepinsky, R.B. (1991) Selective precipitation of proteins from guanidine hydrochloride-containing solutions with ethanol. Anal. Biochem. 195, 177–181. 21. 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. 22. Sali, A. & Blundell, T.L. (1993) Comparative protein modelling by satisfaction of spatial restraints. J. Mol. Biol. 34, 779–815. 23. Morrison, J.R., Fidge, N.H. & Gergo, B. (1990) Studies on the formation, separation, and characterization of cyanogen bromide fragments of human AI apolipoprotein. Anal. Biochem. 186, 145–152. Ó FEBS 2003 Structure of the human Cripto CFC domain (Eur. J. Biochem. 270) 3617 24. Boulware, D.W., Goldsworthy, P.D., Nardella, F.A. & Mannik, M. (1985) Cyanogen bromide cleaves Fc fragments of pooled human IgG at both methionine and tryptophan residues. Mol. Immunol. 22, 1317–1322. 25. Welinder, K.G. (1988) Generation of peptides suitable for sequence analysis by proteolytic cleavage in reversed-phase high- performance liquid chromatography solvents. Anal. Biochem. 174, 54–64. 26. Naismith, J.H. & Sprang, S.R. (1998) Modularity in the TNF- receptor family. Trends Biochem. Sci. 23, 74–79. 27. Mer, G., Hietter, H., Kellenberger, C., Renatus, M., Luu, B. & Lefevre, J.F. (1996) Solution structure of PMP-C: a new fold in the group of small serine proteinase inhibitors. J. Mol. Biol. 258, 158–171. 28. Lohmeyer, M., Harrison, P.M., Kannan, S., DeSantis, M., O’Reilly, N.J., Sternberg, M.J., Salomon, D.S. & Gullik, W.J. (1997) Chemical synthesis, structural modeling, and biological activity of the epidermal growth factor-like domain of human cripto. Biochemistry 36, 3837–3845. 29. Dawson,D.W.,Pearce,S.F.,Zhong,R.,Silverstein,R.L.,Frazier, W.A. & Bouck, N.P. (1997) CD36 mediates the In vitro inhibitory effects of thrombospondin-1 on endothelial cells. J. Cell Biol. 138, 707–717. 30. Rao,Z.,Handford,P.,Mayhew,M.,Knott,V.,Brownlee,G.G.& Stuart, D. (1995) The structure of a Ca (2+)-binding epidermal growth factor-like domain: its role in protein–protein interactions. Cell 82, 131–141. 31. Ferguson, M.A. & Williams, A.F. (1988) Cell-surface anchoring of proteins via glycosyl-phosphatidylinositol structures. Annu. Rev. Biochem. 57, 285–320. 32. Englund, P.T. (1993) The structure and biosynthesis of glycosyl phosphatidylinositol protein anchors. Annu. Rev. Biochem. 62, 121–138. 33. Larrain, J., Bachiller, D., Lu, B., Agius, E., Piccolo, S. & De Robertis, E.M. (2000) BMP-binding modules in chordin: a model for signalling regulation in the extracellular space. Development 127, 821–830. 34. Muskavitch, M.A. (1994) Delta-notch signaling and Drosophila cell fate choice. Dev. Biol. 166, 415–430. 35. Hicks, C., Johnston, S.H., diSibio, G., Collazo, A., Vogt, T.F. & Weinmaster, G. (2000) Fringe differentially modulates Jagged1 and Delta1 signalling through Notch1 and Notch2. Nat. Cell Biol. 2, 515–520. 36. Lanford, P.J., Lan, Y., Jiang, R., Lindsell, C., Weinmaster, G., Gridley, T. & Kelley, M.W. (1999) Notch signalling pathway mediates hair cell development in mammalian cochlea. Nat. Genet. 21, 289–292. 37. Hukriede, N.A. & Fleming, R.J. (1997) Beaded of Goldschmidt, an antimorphic allele of Serrate, encodes a protein lacking trans- membrane and intracellular domains. Genetics 145, 359–374. 38. Hukriede, N.A., Gu, Y. & Fleming, R.J. (1997) A dominant- negative form of Serrate acts as a general antagonist of Notch activation. Development 124, 3427–3437. 3618 S. F. Foley et al. (Eur. J. Biochem. 270) Ó FEBS 2003 . The CRIPTO/ FRL-1/CRYPTIC (CFC) domain of human Cripto Functional and structural insights through disulfide structure analysis Susan F biological importance of Cripto and underline the functional signi- ficance of the CFC domain. In the present work, we have solved the disulfide structure of the CFC domain and