A Heptad Motif of Leucine Residues Found in Membrane Proteins Can Drive Self-assembly of Artificial Transmembrane Segments*

Specific interactions between α-helical transmembrane segments are important for folding and/or oligomerization of membrane proteins. Previously, we have shown that most transmembrane helix-helix interfaces of a set of crystallized membrane proteins are structurally equivalent to soluble leucine zipper interaction domains. To establish a simplified model of these membrane-spanning leucine zippers, we studied the homophilic interactions of artificial transmembrane segments using different experimental approaches. Importantly, an oligoleucine, but not an oligoalanine, se- quence efficiently self-assembled in membranes as well as in detergent solution. Self-assembly was maintained when a leucine zipper type of heptad motif consisting of leucine residues was grafted onto an alanine host sequence. Analysis of point mutants or of a random sequence confirmed that the heptad motif of leucines mediates self-recognition of our artificial transmembrane segments. Further, a data base search identified degenerate versions of this leucine motif within transmembrane segments of a variety of functionally different proteins. For several of these natural transmembrane segments, self-interaction was experimentally verified. These results support various lines of previously reported evidence where these transmembrane segments were implicated in the oligomeric assembly of the corresponding proteins.

Specific interactions between ␣-helical transmembrane segments are important for folding and/or oligomerization of membrane proteins. Previously, we have shown that most transmembrane helix-helix interfaces of a set of crystallized membrane proteins are structurally equivalent to soluble leucine zipper interaction domains. To establish a simplified model of these membrane-spanning leucine zippers, we studied the homophilic interactions of artificial transmembrane segments using different experimental approaches. Importantly, an oligoleucine, but not an oligoalanine, sequence efficiently self-assembled in membranes as well as in detergent solution. Self-assembly was maintained when a leucine zipper type of heptad motif consisting of leucine residues was grafted onto an alanine host sequence. Analysis of point mutants or of a random sequence confirmed that the heptad motif of leucines mediates self-recognition of our artificial transmembrane segments. Further, a data base search identified degenerate versions of this leucine motif within transmembrane segments of a variety of functionally different proteins. For several of these natural transmembrane segments, self-interaction was experimentally verified. These results support various lines of previously reported evidence where these transmembrane segments were implicated in the oligomeric assembly of the corresponding proteins.
In any type of cell, a multitude of integral membrane proteins is simultaneously synthesized and integrated into various membranes followed by association to homo-or heterooligomeric complexes. To ensure specific assembly, their subunits must present complementary recognition domains to each other. These domains may be located on the ectodomains and/or the transmembrane segments (TMSs). 1 Interactions between TMSs are currently intensely studied, since they usually form autonomous ␣-helices and have been found to direct subunit assembly or support correct folding of many membrane proteins (1,2). Biochemical and functional analyses, molecular modeling, and structural studies indicated that the self-assembly of transmembrane helices is driven by a close packing of their characteristically shaped surfaces. These packing interactions may result in pairs of ␣-helices with a right-handed twist as exemplified by glycophorin A (3,4) and probably by synaptobrevin II (5). Other TMS interactions involve a leucine zipper type of side-chain packing as known from certain soluble proteins. Within soluble leucine zippers, the interacting residues form repeated heptad (abcdefg) motifs. Residues at a-and d-positions constitute the hydrophobic core of the interfaces; side-chains at the e-and g-positions are frequently charged, form salt bridges to each other, and make hydrophobic contacts to the core (6). Heptad motifs were also suggested to form the TMS interfaces of phospholamban (7,8) and the M2 proton channel (9). Based on a quantitative evaluation of high resolution structures, we recently confirmed previous observations (10,11) in demonstrating that TMSs primarily interact via a leucine zipper type of packing within bacteriorhodopsin, the photosynthetic reaction center, and cytochrome c oxidase. There, the heptads are repeated on average 2-3 times, and the motif gaxxdexgaxxdexga covers the central parts of the membrane-spanning interfaces. Salt bridges are absent due to the hydrophobic nature of most membrane-embedded residues (12).
To establish a simplified model of membrane-spanning leucine zipper domains, we designed artificial TMSs on the basis of leucine and alanine residues. We show that an oligoleucine sequence or a gaxxdexgaxxdexga motif of leucine residues elicits specific self-assembly in membranes and in detergent solution. Interestingly, variants of this motif are found within the TMSs of a diverse set of natural membrane proteins, where they appear to be important for oligomeric assembly.

EXPERIMENTAL PROCEDURES
Plasmid Constructs-Construction of plasmids pToxR⌬TM and pSNiR⌬TM was described previously (5,13). All other pToxR constructs were made by ligating synthetic oligonucleotide cassettes encoding the desired sequences into the plasmid pHKToxR(TM Il4 )MalE (14) previously cut with NheI and BamHI. For the nuclease A fusions, the oligonucleotide cassettes were ligated into plasmids pSNiR (5) or pSNiR2 previously cut with NheI and BamHI. Details on the pSNiR and pSNiR2 plasmids will be described elsewhere. All constructs were verified by dideoxy sequencing.
ToxR Activity Assays-Transcription activation was determined upon expression of the pToxR constructs in the indicator strain FHK12 as described (15). 0.4 mM isopropyl 1-thio-␤-D-galactopyranoside was added to the cultures to enhance the dynamic range of the produced ␤-galactosidase signals (in Miller units (MU), means Ϯ S.D.) elicited by the different constructs in several independent experiments. This effect is thought to result from isopropyl 1-thio-␤-D-galactopyranoside-induced expression of an FЈ-plasmid-encoded truncated ␤-galactosidase, which competes with full-length enzyme in the formation of functional tetramers. The previously (15) described construct pToxR/GPA13 elicited 1240 Ϯ 298 MU under these conditions.
Gel Filtration Chromatography-pSNiR and pSNiR2 fusion proteins were expressed in BL21(DE3)pLysS cells (Novagen), solubilized in 25 mM HEPES, pH 7.9, 0.5 M NaCl, 2% CHAPS, 1 mM EDTA and quantitated as described (5). Volumes of 300 l at concentrations of 4 or 20 M fusion protein were separated on a Superdex 200HR 10/30 column (Amersham Pharmacia Biotech FPLC system) using a flow rate of 0.5 ml/min and 25 mM HEPES, pH 7.9, 0.5 M NaCl, 1% CHAPS, 1 mM EDTA as running buffer. Fractions of 0.5 ml were collected and analyzed for fusion protein with a dot blot procedure (16) using the 9E10 monoclonal antibody directed against the c-myc marker epitope for detection. The elution profiles were constructed from the antigen content, and the apparent molecular weights were calculated with reference to standards given in the legend to Fig. 3.
Data Base Searching-The Swiss-Prot data base (release 35.0) was searched with the LLXXLLXLLXXLLXLL motif using the Findpatterns option of the HUSAR sequence analysis package made available by the German Cancer Research Center (Heidelberg). Up to three mismatches were allowed. To selectively retrieve TMSs, any amino acid except the charged residues lysine, arginine, glutamate, aspartate, or the helixbreaker proline was allowed for those positions not occupied by leucine.
Miscellaneous Methods-Western blotting was done as described with an antiserum recognizing the maltose-binding protein (MalE) moiety of the constructs, and the bands were quantitated densitometrically (13,15). The ability of our constructs to complement the MalE deficiency of PD28 cells was tested by measuring the cell densities of transformed bacteria in minimal medium containing maltose at 640 nm after different growth periods (13). NaOH extraction was done as described (17) by vortexing whole bacteria with cold 0.1 M NaOH followed by centrifugation to separate soluble from membrane-bound proteins.

A Model of Membrane-spanning Leucine Zipper Domains-
Leucine is the most prevalent amino acid within the interface of leucine zippers (18), which is probably related to its ability to adopt multiple conformations (19). We therefore reasoned that the flexible leucine side chain may be particularly well suited to form a well packed membrane-spanning leucine zipper. The methyl side chain of alanine, in contrast, is expected to be too small for efficient interaction with other alanine residues. This prediction was tested by comparing the self-association of oligoleucine and oligoalanine sequences, which are known to form stable ␣-helices (20,21).
One of the experimental approaches we used is based on an engineered version of the ToxR transcription activator. This protein is anchored by a single TMS of choice within the inner membrane of expressing Escherichia coli cells, where it is thought to exist in a monomer/dimer equilibrium. The dimeric form binds to the cholera toxin promoter, thus activating expression of a downstream lacZ gene in a reporter strain ( Fig. 1; Ref. 14). ␤-Galactosidase expression is therefore diagnostic of ToxR self-assembly in the membrane. We previously established this system as a sensitive tool to study TMS interactions using the structurally well characterized glycophorin A TMS dimer for reference (13,15).
Here, we found that a sequence of 16 leucine residues (designated L16) elicited strong transcription activation (924 Ϯ 209 MU; mean Ϯ S.D.). In contrast, a sequence of 16 alanine residues (designated A16), elicited only a weak signal (210 Ϯ 53 MU) (Fig. 2, A and B). This suggests that the oligoleucine sequence self-assembles in the membrane, whereas the oligoalanine sequence stays largely monomeric. Thus, the latter can be used as host for a leucine zipper motif. Based on the gaxxdexgaxxdexga motif representing the central parts of most transmembrane helix-helix interfaces within crystallized membrane proteins (12), a simplified version of a membrane-spanning leucine zipper interaction domain was designed. In this model, the a, d, e, and g positions are occupied by leucine and all others by alanine. The construct with this hybrid sequence (AZ2) self-interacted to a similar degree (929 Ϯ 186 MU) as the parental L16 protein (Fig. 2, A and B). To demonstrate that the leucine residues contained within AZ2 constitute the helixhelix interface, we mutated some of them to alanine and assessed the consequences for self-interaction. None of the single mutations made (L2A, L5A, L9A) significantly reduced the signal (data not shown). However, when either four a and d (L2A/L5A/L9A/L12A) or four g and e (L6A/L8A/L13A/L15A) positions were simultaneously mutated, the signal dropped by about 50% (516 Ϯ 106 or 596 Ϯ 102 MU). Thus, the leucine residues are critical for the interaction and, hence, most likely make up the interface. Further, ad-and eg-positions seem to be of similar importance for helix-helix packing. Introducing a glycine-proline pair into the center of the AZ2 sequence (L9G/ A10P) similarly affected the interaction (584 Ϯ 100 MU), consistent with the known destabilization of ␣-helices by glycine (22) and their kinking by proline (23) residues. We also replaced the leucines of AZ2 by three different random sequences consisting of the most abundant residues found within TMSs (leucine, isoleucine, valine, phenylalanine, alanine) (24) while maintaining total hydrophobicity and side-chain surface (25). Compared with AZ2, these random sequences also self-assembled much less efficiently, thus emphasizing the superior suitability of the leucine side chain for helix-helix packing (e.g. "random," 446 Ϯ 72 MU; Fig. 2, A and B, and data not shown). The reductions in signal strength of the mutants compared with AZ2 are statistically highly significant (two-tailed Student's t test, p Ͻ 0.001).
Comparing the concentrations of our ToxR constructs by Western blot analysis indicated that most of them were expressed at similar levels, whereas consistent overexpression was noted for the A16 construct (Fig. 2C). When we extracted the cells with NaOH to separate membrane proteins (pellet) from soluble proteins (alkali supernatant) (17), all constructs cosedimented quantitatively with the membranes as expected except A16, which could be partially alkali-extracted (Fig. 2C). Thus, a fraction of the A16 protein seems to remain in a soluble compartment, which is probably due to the comparably low hydrophobicity of the oligoalanine sequence. This fraction is thought not to interfere with the assay. To assess correct integration of the proteins into the inner membrane, we tested their ability to functionally complement the MalE deficiency of PD28 cells. Due to a MalE deletion, this E. coli strain is unable to grow in minimal medium with maltose as the only carbon source (26). In cells expressing correctly inserted ToxR membrane proteins with the ToxR moiety facing the cytoplasm and the MalE domain exposed to the periplasmic space (see Fig. 1), however, the MalE domain allows maltose uptake and thus cell growth (13,14). Here, expression of all constructs including A16 complemented the MalE deficiency of PD28 cells to comparable degrees (Fig. 2D). In contrast, a control construct where the TMS is deleted (ToxR⌬TM) proved unable to support cell growth as expected from its presumed cytoplasmic localization. In sum, equivalent amounts of all ToxR proteins analyzed here for self-assembly appear to be correctly integrated into the inner bacterial membrane, and the obtained ␤-galactosidase activities can thus be directly compared.
To examine self-assembly of our artificial TMSs by an independent approach, their oligomeric states were directly compared in detergent solution (Fig. 3). The L16, A16, AZ2, ⌬TM, L9G/A10P, and "random" sequence segments were genetically fused to the C terminus of a fusion moiety based on Staphylococcus aureus nuclease A, a monomeric soluble protein. The fusion proteins were overexpressed in E. coli, solubilized with CHAPS, and subjected to gel filtration chromatography at concentrations of 4 or 20 M. When injected at 20 M, both L16 and AZ2 fusion proteins eluted as broad peaks with mean apparent molecular masses of ϳ300 kDa plus minor peaks at 47 kDa. At 4 M, the 300-kDa peaks were decreased in favor of the 47-kDa peaks, indicating equilibrium between both forms of the proteins (data not shown). Whereas the 300-kDa peaks clearly indicate assembly to multimers whose stoichiometry is currently not clear, the 47-kDa peaks most likely reflect monomers that may migrate at increased apparent molecular weights due to bound detergent (calculated molecular masses: L16, 21.2 kDa; AZ2, 20.9 kDa). In contrast to that, the ⌬TM, A16, L9G/ A10P, and "random" constructs gave rise to major peaks at 17,

FIG. 2. Transcription activation, expression, and membrane incorporation of ToxR constructs with artificial TMSs.
A, TMS sequences aligned to the underlying heptad pattern. Leucine residues of the zipper variants are shaded for clarity. B, different levels of transcription activation elicited by the different constructs in FHK12 cells indicate sequence-specific TMS assembly in the membrane. The bars represent mean specific ␤-galactosidase activities calculated from numbers of data points given for each construct; error bars denote S.D. C, expression level and membrane association in FHK12 cells. tot, the total cell content of most ToxR proteins was similar as revealed by the staining intensities of the 65-kDa proteins upon Western blotting (densitometric quantitation of seven independent blots established that the average levels of the mutant TMSs ranged from 98 to 111% of the parental AZ2 protein), whereas ToxRA16 was overexpressed; P, the alkali-extracted membrane pellet quantitatively retained all constructs except ToxRA16; SN, the alkali supernatant contained part of ToxRA16 but none of the other proteins. The order of samples corresponds to that in B. D, functional complementation of MalE deficiency to assess correct membrane incorporation. All constructs except the control construct ToxR⌬TM allowed for similar rates of PD28 cell growth, thus confirming their correct N in -C out integration. The individual data points represent means from five independent experiments. 22, 31, and 41 kDa at both concentrations. These peaks are consistent with monomers (calculated masses: 19.5, 20.5, 20.9, and 20.9 kDa, respectively) whose migration may be influenced by different amounts of bound detergent depending on the presence and the hydrophobicity of the hydrophobic segments.
Taken together, two independent experimental approaches indicate that both the oligoleucine sequence and the model leucine zipper motif AZ2 self-assemble in a sequence-specific way in membranes as well as in detergent solution.
Self-assembly of Leucine-rich Natural Transmembrane Segments-Given the self-assembly of the AZ2 model, we assessed whether TMSs with similar leucine patterns exist in naturally occurring proteins. The Swiss-Prot data base was searched for hydrophobic sequence segments with the motif LLXXLLX-LLXXLLXLL allowing for up to three mismatches. This search yielded 38 predicted N-terminal signal sequences, 30 TMSs predicted within polytopic membrane proteins, and 15 predicted TMSs from bitopic membrane proteins when homologous proteins from different species were counted only once. Whereas the signal sequences and TMSs of polytopic proteins were not further investigated here, the TMS sequences corresponding to the bitopic proteins are shown in Table I. Selfinteraction of a subset was examined with the ToxR system. The TMS eliciting the strongest signal was derived from the erythropoietin receptor followed by the TMSs of the Friend spleen focus-forming virus envelope protein, E-cadherin, and hemagglutinin of canine distemper virus. Other TMSs corresponding to papillomavirus E5 protein, mouse poliovirus receptor homolog, and chick asialoglycoprotein receptor gave rise to intermediate values suggesting lower levels of self-assembly ( Fig. 4A and Table I). A Western blot run for control revealed roughly similar expression levels (Fig. 4B).
The data predict that these TMSs are important for oligomerization of the corresponding proteins. A survey of previously reported experimental evidence and our own experiments indicated this indeed to be the case for several of these proteins or related homologs as discussed below. DISCUSSION We demonstrate that an artificial TMS of leucine residues efficiently self-assembles in membranes and in detergent solution. A heptad motif of leucine residues suffices to elicit selfassembly, which therefore is thought to be driven by the type of side-chain packing known from leucine zipper interaction domains. The main implications of our results are 2-fold. (i) They establish a simplified model system of short membrane-spanning leucine zippers. (ii) They suggest that similar interaction domains may play a role in subunit-subunit recognition of certain natural membrane proteins.

Structural Aspects of Membrane-spanning Leucine Zippers
We assume that the L16 and AZ2 TMSs form ␣-helical bundles upon self-assembly. Self-assembly is thought to involve self-complementary helix surfaces that associate which each other via a "knobs-into-holes" type of side-chain packing characteristic of leucine zippers (6). The highly flexible leucine side chain (19) may be particularly well suited for this type of packing interaction. Consistent with this concept, leucine-rich heptad motifs have previously been applied in the design of helix bundles forming transmembrane ion channels (27) or of a folded polytopic membrane protein (28). On the other hand, leucine helices have frequently been used as experimental models to study TMS interactions with lipid bilayers. For some of these studies (20, 29 -31), both termini of the leucine helices were capped with lysine residues whose repulsive interaction may keep them in a monomeric state (31). In other cases (17, 32, 33), their self-assembly as implied by our data should be considered in interpreting the results.
A leucine zipper type of side-chain packing also accounts for TMS interactions within phospholamban (7,8), the M2 proton channel (9), and different polytopic membrane proteins (12). In contrast to our leucine-based model, these heptad motifs are made up of different hydrophobic amino acids, which may generate the characteristically shaped helix surfaces ensuring specific, stoichiometric, and/or heterophilic assembly of these natural proteins.

Leucine Zipper Motifs in Natural Membrane Proteins
Data base searching identified leucine-rich heptad motifs within different naturally occurring TMSs, and an analyzed subset of these indeed exhibited various levels of self-interaction. This predicts a role of TMS interactions in the assembly of the corresponding membrane proteins. This is also implied by studies on the corresponding full-length proteins as will be briefly discussed below.
Cadherins-Cadherins are calcium-dependent homophilic b Sequences representing those parts of the TMSs that cover the query pattern. The sequence positions of the N-terminal residues are stated, and leucine residues within the heptad pattern given above the sequences are in boldface type.
c ␤-Galactosidase activity as determined with the ToxR system (MU, mean Ϯ S.D.). cell-cell adhesion molecules. Their function depends on lateral clustering within the plasma membrane (34), which is believed to involve interactions between extracellular (35) and juxtamembrane domains (36). On the other hand, leucine-rich heptad motifs are evolutionarily conserved in the TMSs of different cadherin families, and our data demonstrating selfinteraction of the E-cadherin TMS suggest a role of TMS interactions in clustering. Strong support for this hypothesis is provided by our recent experimental evidence indicating that mutations reducing the TMS interaction likewise affect the adhesive properties of full-length E-cadherin expressed in eukaryotic cells. 2 Erythropoietin Receptor-The erythropoietin receptor (EpoR) is required for erythrocyte maturation. In analogy to other growth factor receptors, erythropoietin binding is thought to trigger homo-dimerization followed by receptor activation (37). Apart from the case of the Neu oncogene product, where a point mutation within the TMS triggers ligand-independent receptor activation (38), the role of the TMS in growth factor receptor activation is currently not clear. Since ligand binding is translated into activation of cytoplasmic domains, it has been postulated that the subunit-subunit interface of growth factor receptors extends across the membrane and that TMS interactions contribute to ligand-induced subunit assembly in a nonspecific way (1). Our finding that the EpoR TMS is capable of self-interaction indeed suggests its contribution to ligand-induced receptor assembly. Alternatively, the EpoR may exist as a preformed dimer activated by ligand binding. Precedence for the latter model is given by the insulin receptor or the aspartate chemoreceptor; in both cases, ligand-binding activates preformed receptor oligomers (39). Ligand-independent dimerization has also been proposed for the epidermal growth factor receptor (40).
Viral Envelope Proteins-Enveloped viruses enter the cytoplasm of host cells upon fusion of viral and cellular membranes mediated by fusogenic viral envelope proteins. These proteins exist as oligomeric complexes (41), and both their fusogenicity and oligomerization appear to depend on their TMSs. For example, the influenza hemagglutinin TMS is required for full membrane fusion (42) and stabilizes the trimeric complex (43). Also, mutations of conserved leucine residues within the TMS of the hemagglutinin-neuraminidase of Newcastle disease virus affected tetramerization and fusion promotion (44). Extending these findings, our data suggest a role of the TMS in oligomerization of hemagglutinin-neuraminidase from canine distemper virus and of the Friend leukemia virus envelope protein. Apart from homooligomerization, a heterophilic and functionally important interaction has been reported between the EpoR and the gp55 protein of Friend spleen focus-forming virus, which is derived from its envelope protein (45). At the surface of infected erythroid cells, the EpoR and gp55 form a noncovalent complex, which results in erythropoietin-independent cell differentiation (46). Complex formation is therefore thought to cause persistent EpoR activation (47). Notably, both the gp55 TMS and the EpoR TMS have been shown to be crucial for this heterophilic interaction (48,49). Since both TMS sequences have been identified by our data base search and shown to self-interact, we propose that formation of the heteromeric complex proceeds from preformed gp55 and EpoR homomers.
E5 Protein-The papillomavirus E5-protein is a transforming membrane protein that exists as a disulfide-bonded dimer (50). Its transforming activity presumably rests on interaction with, and ligand-independent activation of, the receptors for epidermal growth factor, colony-stimulating factor (51), or platelet-derived growth factor (52). In the case of the plateletderived growth factor receptor, binding to the E5 protein has been directly demonstrated to involve the TMSs plus extracellular flanking regions of both receptor and E5 protein (53). Although the E5 protein extracellular region and the glutamine residue within the TMS are important for activity (54), we speculate that the leucine-rich surface of its TMS aids in homodimer formation and/or binding to the various growth factor receptor TMSs.
Asialoglycoprotein Receptor-The hepatic asialoglycoprotein receptors remove abnormally glycosylated proteins from blood circulation (55). The chick homolog exists as a homotrimer whose formation and stability depends on the TMS and flanking sequences (56,57). This is consistent with self-interaction of its TMS shown here.
These examples demonstrate that assembly of several different natural membrane proteins depends on their TMSs as predicted by the presence of leucine-rich heptad repeats. Future studies will show whether these TMS interactions are based on the leucine zipper type of packing as inferred for our self-assembling model TMSs L16 and AZ2. TMS interactions may be modulated by the lipid composition of the respective host membrane. Further, they may not be the exclusive cause of subunit-subunit recognition but may be complemented by interactions between extramembraneous domains in particular cases. FIG. 4. Transcription activation and expression of ToxR constructs with natural TMSs. A, transcription activation in FHK12 cells reflects various levels of self-assembly of the TMSs whose amino acid sequences are given in Table I. The Swiss-Prot identifiers are explained in the legend of Table I. The bars represent mean specific ␤-galactosidase activities averaged from 24 -32 data points; error bars denote S.D. Arrowheads indicate the signals elicited by the L16 and A16 sequences for comparison (see Fig. 2). B, Western blotting revealed roughly similar expression levels for the different proteins. The order of samples corresponds to that in A.