Determination of N- and C-terminal borders of the transmembrane domain of integrin subunits.

Previous studies on the membrane-cytoplasm interphase of human integrin subunits have shown that a conserved lysine in subunits alpha(2), alpha(5), beta(1), and beta(2) is embedded in the plasma membrane in the absence of interacting proteins (Armulik, A., Nilsson, I., von Heijne, G., and Johansson, S. (1999) in J. Biol. Chem. 274, 37030-37034). Using a glycosylation mapping technique, we here show that alpha(10) and beta(8), two subunits that deviate significantly from the integrin consensus sequences in the membrane-proximal region, were found to have the conserved lysine at a similar position in the lipid bilayer. Thus, this organization at the C-terminal end of the transmembrane (TM) domain seems likely to be general for all 24 integrin subunits. Furthermore, we have determined the N-terminal border of the TM domains of the alpha(2), alpha(5), alpha(10), beta(1), and beta(8) subunits. The TM domain of subunit beta(8) is found to be 22 amino acids long, with a second basic residue (Arg(684)) positioned just inside the membrane at the exoplasmic side, whereas the lipidembedded domains of the other subunits are longer, varying from 25 (alpha(2)) to 29 amino acids (alpha(10)). These numbers implicate that the TM region of the analyzed integrins (except beta(8)) would be tilted or bent in the membrane. Integrin signaling by transmembrane conformational change may involve alteration of the position of the segment adjacent to the conserved lysine. To test the proposed "piston" model for signaling, we forced this region at the C-terminal end of the alpha(5) and beta(1) TM domains out of the membrane into the cytosol by replacing Lys-Leu with Lys-Lys. The mutation was found to not alter the position of the N-terminal end of the TM domain in the membrane, indicating that the TM domain is not moving as a piston. Instead the shift results in a shorter and therefore less tilted or bent TM alpha-helix.

Integrins are heterodimeric receptors composed of an ␣ subunit noncovalently associated with a ␤ subunit. Each subunit has an N-terminal extracellular domain, a transmembrane (TM) 1 region and a cytoplasmic domain. The human ␣and ␤ subunits constitute two unrelated protein families of 18 and 8 members, respectively (1,2).
Integrins mediate cell adhesion to the pericellular matrix and to neighboring cells (1). In addition to the anchoring function, ligand binding to integrins generates intracellular signals required for several cellular processes, including cell migration and proliferation. The ability to bind ligands is regulated by mechanisms acting on the cytoplasmic part of the protein, an unusual receptor feature. Integrin activation by cytoplasmic signals has been shown to involve transmembrane conformational changes (3,4). Subsequent ligand binding induces further structural rearrangements, as monitored by exposure of new epitopes, in the extracellular as well as in the intracellular domains (5,6).
Recently, significant progress has been made in the elucidation of the mechanisms controlling integrin activation ("inside-out signaling") and ligand-induced signaling ("outside-in signaling"). The cytoplasmic protein talin was found to bind to the membrane-proximal region of the ␤ 1 , ␤ 2 , and ␤ 3 subunits and thereby activates the integrins (7)(8)(9)(10). Integrin activation has been shown to require separation of the ␣ and ␤ subunit cytoplasmic domains from each other (9,11), and this is presumably the way by which talin activates integrins. In addition, recent reports have suggested that the TM domains of the subunits mediate integrin clustering after ligand binding (12). TM domains therefore appear to contribute to signaling in both directions across the membrane rather than serving merely to connect the intra-and extracellular domains. Evidence for the important functions of integrin TM domains is further provided by their high degree of conservation within the integrin ␣and ␤-protein families and also between species for individual subunits.
Several models have been proposed to explain the transmembrane signaling of integrins. These are based on different types of movements of the TM domains, such as rotation, tilting, and piston movement (13)(14)(15)(16)(17). As a step toward the identification of the mechanisms used for outside-in and inside-out signaling, we have in the present study defined the borders of the TM domains from five selected integrin subunits. This information has allowed us to test the piston model for integrin ␣ 5 ␤ 1 .

MATERIALS AND METHODS
Enzymes and Chemicals-Unless stated otherwise, the enzymes were purchased from Promega, MBI Fermenta AB and New England Biolab. For PCR puReTaq TM Ready-To-Go TM PCR beads from Amersham Biosciences were used. PCR primers were from DNA Technology and TAG Copenhagen. DNA manipulations were made using the TOPO kit from Invitrogen, the Rapid Ligation kit from Roche Applied Sciences, and the QuikChange TM site-directed mutagenesis kit from Stratagene. Ribonucleotides, the cap analogue m7G(5Ј)ppp(5Ј)G, and [ 35 S] Met were from Amersham Biosciences. Dithiothreitol, bovine serum albumin, RNasin ribonuclease inhibitor, plasmid pGEM1, rabbit reticulocyte lysate, and amino acid mixture without methionine were from Promega. Spermidine was from Sigma.
Expression in Vitro-The Lep vector constructs were transcribed by SP6 RNA polymerase and translated in reticulocyte lysate in the presence and absence of dog pancreas microsomes as described (19). The proteins were analyzed by SDS-polyacrylamide gel electrophoresis, and the bands were quantitated by phosphorimaging on a Fluorescent Image Reader FLA-3000 using the Image Reader 1.1 software. The extent of glycosylation of a given construct was calculated as the quotient between the intensity of the glycosylated band divided by the summed intensities of the glycosylated and nonglycosylated bands. In general, the glycosylation efficiency varied by no more than Ϯ5% between different experiments. The "minimal glycosylation distance" (MGD), i.e. the number of residues between the acceptor site and the lipid bilayer required to reach half-maximal glycosylation efficiency, previously determined for poly-Leu TM segments (18) was used to identify the water-lipid interface.

RESULTS
Glycosylation Mapping-The glycosylation mapping technique has previously been described in detail (18). Briefly, the assay is based on the ability of the lumenally disposed endoplasmic reticulum enzyme oligosaccharyl transferase (20) to add a glycan to the Asn residue in Asn-Xaa-(Ser/Thr) glycosylation acceptor sites in target proteins. The minimal number of residues between the end of the model transmembrane segment (25LV) composed of 25 consecutive leucines and one valine and flanked by polar residues and the acceptor Asn required for half-maximal glycosylation is ϳ10 residues when the acceptor site is C-terminal to the TM segment and ϳ14 residues when it is N-terminal to the TM segment (18). Using a glycosylation site scanning approach to identify the corresponding MGD for a TM segment of interest, one can thus estimate the position of this TM segment in the endoplasmic reticulum membrane by comparison with the MGD for the model TM segment (15).
To locate both the N-and C-terminal ends of integrin TM segments relative to the endoplasmic reticulum membrane, TM regions of chosen integrin subunits were cloned into two series of vectors (boxes A and B in Fig. 1) based on the well characterized integral membrane protein Lep. The A series was used for determining the position of the C-terminal end, and the B series was used for determining the position of the N-terminal end of the TM segments. The vectors in each series differ only in the position of the glycosylation acceptor site relative to the TM segment. The constructs were transcribed and translated in vitro in the presence and absence of dog pancreas rough microsomes. The measured MGD values were used to estimate the positions of the chosen integrin TM segments in the endoplasmic reticulum membrane by comparison with the 25LV model TM described above.
Determination of N-terminal Borders of the Transmembrane Domain of Integrin Subunits-Five integrin subunits were selected for determination of the N-terminal border of their TM domains. ␣ 2 , ␣ 5 , and ␤ 1 were chosen as representatives of ␣ and ␤ subunits with TM domains typical for each of the two protein families, whereas ␣ 10 and ␤ 8 are examples of interesting deviations from the consensus sequences. For the analysis, segments of integrin subunits ␣ 2 (aa 1126 -1162), ␣ 5 (aa 992-1029), ␣ 10 (aa 1115-1154), ␤ 1 (aa 722-760), and ␤ 8 (aa 673-708) were inserted into the B series vectors (Fig. 1).
The results of in vitro transcriptions/translations of the constructs are summarized in Fig. 2. As expected, a rapid drop in glycosylation efficiency is seen when the acceptor site is moved closer to the TM segment (Fig. 2, B and C). For example, the glycosylation efficiency is reduced from 68% for the ␣ 2 TM domain in vector 14 (where the acceptor Asn is 14 residues upstream of Val 1134 ) to 11% for same TM segment in vector 13 (where the acceptor Asn is 13 residues upstream of Val 1134 ; Fig.  2C). By comparison with the MGD value of 14 residues determined for the model 25LV TM described above (18), the residue in the ␣ 2 TM domain located in the equivalent position in the membrane as the N-terminal Leu in the model segment is thus Val 1134 . We conclude that the N-terminal membrane border is approximately at Val 1134 for ␣ 2 , at Leu 999 for ␣ 5 , at Leu 1123 for ␣ 10 , at Pro 731 for ␤ 1 , and at Tyr 682 for ␤ 8 (Fig. 3).
Determination of C-terminal Borders of the Transmembrane Domain of Integrin Subunits-The C-terminal borders of the TM domains of integrin subunits ␣ 2 , ␣ 5 , and ␤ 1 have been determined previously (15). In this study we have determined the C-terminal border for the ␣ 10 and ␤ 8 subunits. Segments of ␣ 10 (aa 1115-1154) and ␤ 8 (aa 673-708) were cloned into the A series vectors. The glycosylation efficiency of in vitro expressed proteins was tested as described above. The results are shown in Fig. 4. By comparison with the MGD value of 10 residues determined for the model 25LV TM described above (18), the C-terminal membrane border is at Ala 1151 for ␣ 10 and at Ile 704 for ␤ 8 (Fig. 3).
Testing the Piston Model-Several models for integrin-de- The MGD is the number of residues between the end of the integrin TM segment and the Asn in the engineered Asn-Ser-Thr glycosylation acceptor site (*) needed for half-maximal glycosylation of the protein.
OST, oligosaccharyl transferase pendent signal transduction across the membrane have been proposed (13)(14)(15)(16)(17), with special attention given to the highly conserved regions flanking the membrane-cytoplasm interphase. The C-terminal part of the TM domain has been suggested to move out of the membrane either by sliding of the TM helices in a piston-like motion (13), by changes in tilt (15), or by a an uncoiling process (14) (see Fig. 6).
We have previously shown that the C-terminal end of the TM segment in ␤ 1 shifts relative to the membrane if a second lysine is introduced next to the conserved membrane-embedded lysine, i.e. by mutating Leu 753 to Lys in ␤ 1 (15). In the present study we analyzed whether the N-terminal end of the TM domain would move relative to the membrane when the Cterminal end is forced out of the membrane in this way. The MGD determination was repeated for ␣ 5 and ␤ 1 constructs carrying the L-to-K mutation at the C-terminal end (␣ 5 L1023K and ␤ 1 L753K). The results show that no change in the position of the N-terminal end of ␣ 5 and ␤ 1 TM domains occurs (Fig. 5). Thus, the luminal N-terminal ends of the ␣ 5 and ␤ 1 TM segments do not move relative to the membrane when their Cterminal ends are forced out of the membrane on the cytoplasmic side. DISCUSSION The role of the cytoplasmic domains of ␣ and ␤ subunits in integrin activation and signaling is well established (9,(21)(22)(23). Accumulating data indicate active roles also for the TM domains (12,24), as well as for the membrane-proximal parts flanking the TM domains (8,25). However, it is not yet known what kind of molecular movements of the TM domains are linked to these events. To better understand the role of integrin TM domains, we previously determined the membrane-cyto- plasm interface for ␣ 2 , ␣ 5 , ␤ 1 , and ␤ 2 using an in vitro glycosylation mapping assay (15). Unexpectedly, the transmembrane domain was found to include an additional 5-6 amino acids at the C-terminal end compared with earlier predictions; this result was subsequently confirmed by NMR studies of the ␤ 3 TM and cytoplasmic domains in dodecylphosphocholine micelles (26). A basic amino acid, which is conserved in all human integrin subunits residue, Arg in ␣ V and ␤ 7 and Lys in all other subunits, is thus located in the plasma membrane in the absence of interacting proteins. A basic residue at this position is likely to influence interactions with membrane proteins and/or the orientation of the TM domain in the lipid bilayer.
In the present study, the characterization of integrin TM domains has been extended with (i) determination of the Cterminal border of ␤1-associated subunit ␣ 10 ; (ii) determination of the N-terminal borders of ␣ 2 , ␣ 5 , ␣ 10 , and ␤ 1 ; (iii) determination of both ends of the strongly divergent ␤ 8 TM domain; and (iv) a test of the validity of the piston model (13) as a possible mechanism for propagating conformational changes across the plasma membrane. The amino acid motif (K/R)XGFFKR is present at the membrane-cytoplasm interface in all 18 integrin ␣ subunits except ␣ 8 (KCGFFDR), ␣ 9 (KLGFFRR), ␣ 10 (KLGFFAH), and ␣ 11 (KLGFFRS). In view of the high degree of conservation of the motif, minor deviations such as those in ␣ 8 and ␣ 10 may be functionally significant. Analysis by the in vitro glycosylation assay showed that the ␣ 10 TM domain extends 1-2 amino acid residues further at the C terminus compared with the TM domain of other ␣ subunits. This result is not unexpected considering the absence in ␣ 10 of the strongly charged dipeptide KR. Thus, the membrane-embedded lysine in ␣ 10 resides even deeper inside the membrane than in other ␣ subunits.
It is not obvious from the primary sequences where the N-terminal borders of integrin TM domains are located. The border has usually been predicted to be located 23 amino acids or more upstream of the conserved membrane-embedded lysine (e.g. Lys 752 in ␤ 1 ) (27-31). However, not all integrin subunits may necessarily have TM domains of identical length, and the ␣ subunits in particular have variable numbers of nonpolar amino acids upstream of the predicted 23 residues that may influence the length of the TM segment.
Applying the in vitro glycosylation method, the N-terminal borders for ␣ 5 and ␣ 10 were found to be located at the same distance upstream of the membrane-embedded lysine (Fig. 3). Thus, both these subunits have a tryptophan in position to interact with the carbonyl group of phospholipids, a commonly found arrangement in membrane proteins (32). In ␣ 5 , Pro 998 immediately outside of the TM domain will promote disruption of the ␣-helix. Subunit ␣ 10 has a weakly polar serine residue at the corresponding position and a proline located 4 residues further upstream, suggesting that the ␣-helix may continue approximately one turn into the extracellular domain. For the ␣ 2 subunit the water-lipid interface was found to reside 2 residues further toward the C terminus than in ␣ 5 and ␣ 10 , if the conserved KLGFF motif is used as the reference point. Thus, in ␣ 2 , Gly 1133 and Pro 1131 are located approximately 1 and 3 residues outside the membrane, respectively, and may serve as helix-breakers. According to the results of the glycosylation mapping assay, the approximate length of the membrane spanning segment is 25 residues in ␣ 2 , 27 residues in ␣ 5 , and 29 residues in ␣ 10 . One implication of these results is that the TM ␣-helix of the three selected representatives of integrin ␣ subunits are not running perfectly perpendicular to the membrane but rather have to be bent, tilted, and/or coiled in slightly different ways to fit into the membrane.
The TM segment of ␤ 1 was found to be ϳ26 aa in length, with Pro 731 at the N-terminal border instead of the commonly predicted Asp 728 . The TM segment of ␤ 8 was originally suggested to consist either of the 15 hydrophobic amino acids that are flanked by Arg 684 and Lys 700 at the N and C termini, respectively, or by a 30-amino acid segment (33). Our experimental data indicate that both Arg 684 and Lys 700 are located inside the membrane, resulting in a TM segment of 22 residues in length.
Still, the TM segment in ␤ 8 is significantly shorter than that in ␤ 1 . The ␤ 8 TM domain also exhibits several other unique features. The sequence around the membrane-cytoplasm interface, WKLLXX(I/F)HDR(R/K)E, is conserved in ␤ 1 , ␤ 3 , ␤ 5 , and ␤ 6 , whereas significant deviations from the motif are present in ␤ 2 , ␤ 7 , and ␤ 8 ; ␤ 4 shows only weak similarity in this part of the protein, as well as in the cytoplasmic domain. Our measurements show that the ␤ 8 TM domain continues four residues beyond the membrane-embedded lysine, compared with approximately 6 residues in ␤ 1 . Other notable differences between ␤ 8 and ␤ 1 are the absence of Trp in front of the conserved C-terminal Lys, and the replacement of HDRRE with another polar sequence. Furthermore, the membrane-embedded arginine (Arg 684 ) is only found in ␤ 8 , whereas the ␤ 8 TM domain lacks both the glycine and alanine residues that are present at specific positions in most other ␤ subunits.
Whether these structural features confer any particular function to ␤ 8 is presently not known. However, ␤ 8 , as well as ␤ 1 , ␤ 3 , ␤ 5 , and ␤ 6 , associate with the ␣ V subunit, and therefore the unusual structure of the ␤ 8 TM domain most likely does not influence the selection of the ␣ subunit partner. Because ␤ 8 lacks key talin-binding residues in the membrane proximal and cytoplasmic domains (10), i.e. Ile-His 758 , Trp 775 , and Asn-Pro- FIG. 6. Hypothetical models of transmembrane signaling. In the piston model (A) the TM regions (red) are assumed to slide as a rigid piston through the membrane, moving the conserved lysine (K) of one or both subunits in and out of the membrane. In the coiled model (B) the two TM regions (red) are coiled around each other as a coiled coil. When uncoiled, the TM ␣-helices are too long to run perpendicular to the plasma membrane. Instead the C-terminal end of the TM region would be moved into the cytoplasm. In the tilting model (C) the TM regions adapt to the bilayer by tilting. Changes in the tilt angle will push the C-terminal end of the TM into the cytosol. The separation of the TM regions after talin binding is not included in these models because it is not clear in which conformation this occurs. If model B is correct, the coiled coil structure would correspond to a conformation before activation by talin.
FIG. 5. Comparison of N-terminal borders of the ␣ 5 and ␤ 1 TM segment with the borders of the ␣5L1023K and ␤1L753K TM segments. Glycosylation efficiencies of ␣ 5 and ␣5L1023K was analyzed in three different B vectors, and the same was done for ␤ 1 and ␤1L753K. As seen in the graph, the MGD values were not altered by the mutations.
Ile-Tyr 783 in human ␤ 1 , ␣ V ␤ 8 may have a different mechanism of activation than other integrins. Possibly, this is reflected in the structure of the TM domain. Relatively little is yet known about the signaling properties of ␣ V ␤ 8 , and further studies may clarify whether the ␤ 8 TM domain has any specific role in this context.
Under the conditions of the glycosylation assay, the ␣-carbon of Lys 752 in the isolated TM domain of ␤ 1 and the corresponding lysine in other integrin ␣and ␤-TM domains is clearly located in the lipid bilayer. A similar position for the lysine was found when a ␤ 3 fragment consisting of the TM and cytoplasmic domains was analyzed by NMR spectroscopy (26). However, the presence of a tryptophan or tyrosine at the position immediately preceding the conserved Lys/Arg in all integrin subunits except ␤ 8 suggests that the (W/Y)(K/R) motif may be found at the membrane-cytoplasm interphase in certain integrin conformation(s). The membrane proteins commonly have a tryptophan or a tyrosine at the ends of the TM segments where they can serve as anchors by interacting both with the fatty acid chains and the carbonyl group of the phospholipids via hydrophobic and hydrogen bonds, respectively (32). The basic residue (e.g. Lys 752 in ␤ 1 ) may serve as a flexible anchor that can interact via its long side chain with the negatively charged phosphate groups of phospholipids even if the ␣-carbon moves a short distance in or out of the membrane.
It has been suggested that movement of the conserved Cterminal end of the TM domain in or out of the membrane could occur if the whole TM helix slides as a rigid piston through the membrane (13) (Fig. 6A). Because the extracellular region immediately outside of the TM domains analyzed in this study contains a short stretch of nonpolar or weakly polar residues, such a model appeared to be possible. However, we find that the position of the N-terminal end of the TM helix remains unaltered when the position of the C-terminal end is forced to shift from Phe-Lys 1027 to Tyr-Lys 1022 for ␣ 5 and from Ile-His 758 to Trp-Lys 752 for ␤ 1 by replacing a leucine with lysine at positions 1023 and 753, respectively (15). Therefore, the piston model seems unlikely for the ␣ 5 and ␤ 1 subunits. If the C-terminal end of the TM domains is induced to move into the cytoplasm by physiological stimuli, e.g. by a protein-protein interaction, altered tilting and/or uncoiling seem more likely mechanisms to account for the shortening of the membrane-spanning segment. Two schematic models for such shortening of the TM helix is pictured in Fig. 6 (B and C). Further experiments will be needed to test whether alterations in the orientation of the membrane-proximal region of one or both integrin subunits are linked to the active, inactive, or ligandstimulated conformations.