Quarterly Reviews of Biophysics 25, 4 (1992), pp. 433-457 Printed in Great Britain
433
Structure and dynamics of polypeptides and proteins in lipid membranes
HORST VOGEL Swiss Federal Institute of Technology, Lausanne
1.
INTRODUCTION
433
2.
STRUCTURAL MODELS OF MEMBRANE PROTEINS DERIVED FROM RAMAN SPECTROSCOPY AND PREDICTION METHODS
3. STRUCTURAL DYNAMICS OF MODEL PEPTIDES
I.
434 441
4.
THE ROLE OF PROLINE RESIDUES IN TRANSMEMBRANE HELICES
5.
ACKNOWLEDGEMENT
6.
REFERENCES
445
453
453
INTRODUCTION
The elucidation of the molecular mechanisms whereby ions and polar molecules are translocated across the hydrophobic barrier of a lipid bilayer in biological membranes is one of the most challenging problems in biological research. Specific membrane proteins, such as pumps, carriers and channels, play the central role in the various translocation pathways. Recent progress in expression cloning has provided the sequence of a number of biologically important membrane proteins and in principle the door is open to investigate every protein which might be of importance in the central signal transduction and transport processes. Unfortunately, to date there are only a few examples where the threedimensional structure of membrane proteins are known at atomic resolution. The photosynthetic reaction centres from purple bacteria (Deisenhofer et al. 1985), bacteriorhodopsin (Henderson et al. 1990) and the large porin channel of Rhodobacter capsulata (Weiss et al. 1991). According to these structural data membrane proteins seem to fold in general in membrane-spanning a-helices and /?-strands in order to saturate hydrogen bonds. Only these two motifs seem to form stable structures which can be in contact with the hydrophobic lipid interior of a membrane. The hydrophobic or amphipathic segments may be arranged such that the hydrophobic surfaces are in contact with the hydrophobic part of the lipid Postal address: Ecole Polytechnique Federate de Lausanne, Institut de Chimie Physique, CH-1015 Lausanne, Switzerland.
434
H. Vogel
1500
1000 •
1630
1650 1670 Wavenumber [crrr'l
1690
Fig. i. 15-point representation of the amide I Raman spectra of a typical a-helical membrane protein, lactose permease (A), and two typical /^-structure membrane proteins, porin (O) and maltoporin ( • ) . From Vogel & Jahnig (1986a).
bilayer or other hydrophobic protein segments, whereas the hydrophilic protein surfaces are facing other hydrophilic protein segments in the interior of the membrane embedded protein. Such hydrophilic parts might create a central hydrophilic pathway for the membrane translocation of ions and polar molecules, yet the real existence of such structures has to be proven experimentally in each particular case. In spite of the lack of experimental evidence, detailed models have been published of how the polypeptide chains of nearly every sequenced membrane proteins are thought to be folded. These models are mainly based on the distribution of hydrophobic and hydrophilic regions within the protein sequence. Here, we summarize our work on the structure and dynamics of proteins and polypeptides in lipid membranes, based on optical spectroscopic measurements, secondary structure prediction and molecular dynamics calculations. 2. STRUCTURAL MODELS OF MEMBRANE PROTEINS DERIVED FROM RAMAN SPECTROSCOPY AND PREDICTION METHODS
In spite of intensive research, most of the investigated membrane proteins have resisted attempts to grow three-dimensional crystals of sufficient quality. In this situation, one may turn to other techniques which provide structural information, in general less than high resolution X-ray diffraction, but do not require threedimensional crystals. Spectroscopic techniques are ideally suited for this purpose. High-resolution, multidimensional NMR spectroscopy is not applicable to membrane proteins due to excessive protein immobilisation in lipid bilayers. In the case of small, membrane active proteins, this problem might be circumvented
Structure and dynamics of polypeptides and proteins
435
by incorporating the polypeptides in detergent micelles. Solid-state NMR of suitably labelled proteins might be an alternative to obtain the orientational distribution of the specially-labelled groups of the protein in a lipid bilayer, as has been shown recently in the case of filamenteous phage proteins in membranes (Opella et al. 1987; Shon et al. 1991). On the other hand, optical spectroscopy such as circular dichroism (CD), infrared absorption (IR), Raman scattering and fluorescence spectroscopy deliver information on protein structure and mobility. CD spectroscopy can provide both qualitative and quantitative assessment of overall protein secondary structure and has been shown to yield quite reliable results for the average a-helix content of proteins (Provencher & Glockner, 1981; Hennesey & Johnson, 1981). However, /?-turns and /?-strand structures are difficult to distinguish and the technique might give rise to artefacts in the case of membrane proteins (Vogel & Gartner, 1987). Vibrational spectroscopy such as Raman and IR spectroscopy has been shown in recent years to be a powerful tool for the determination of membrane protein structures (Vogel et al. 1985 ; Vogel & Jahnig, 1986a, b; Vogel & Gartner, 1987; Lee et al. 1991; Braiman & Rothschild, 1988; Surewicz & Mantsch, 1988). It has benefited from both the progress made in the calculation of protein spectra (Krimm & Bandekar, 1986) and the development of quantitative methods for the evaluation of structural information (Williams, 1983). In an IR or Raman spectrum of a protein, the amide I band in the region of 1600—1700 cm"1 is the most sensitive vibrational band for investigating the average protein secondary structure. The amide I band originates primarily from the C—O stretching vibrations of peptide amide groups coupled to the in plane N-H bending and C~N stretching modes (Krimm & Bandekar, 1986). The frequency of this vibration depends on the nature of the hydrogen bonding involving the C = O and N~H moieties, and therefore in turn the frequency of the amide I mode reflects the particular secondary structure adopted by the polypeptide backbone. Although, in principle, it would be possible to determine the average protein conformation and to distinguish the different structural elements from their particular amide I vibrational frequencies, this strategy is not straightforward in practice because the amide I band of proteins normally appears as a relatively structureless, broad band in the IR and Raman spectra. This is demonstrated in Fig. 1 showing the Raman amide I spectra of typical a-helix and /?-strand structure membrane proteins. Different resolution enhancement algorithms have been applied to solve the problem but this method is not free of artefacts, and furthermore, certain resolved amide I vibrational components cannot always be assigned with confidence to a particular protein structure (Surewicz & Mantsch, 1988). An alternative method has been successfully used for quantitative estimation of the protein secondary structure. The spectrum to be analysed is fitted by a linear combination of a set of reference proteins of known three-dimensional structure. The set of reference proteins is transformed to a set of orthogonal 'quasi' or eigenspectra which may be combined to generate the original data within experimental error. Since the structure of the reference proteins is known, the structure of the protein under investigation can be
436
H. Vogel
determined usually in percentages of different structure classes, such as ordered helix (Ho), when the amino acid residue participates in hydrogen bonds to two sides along the helix, disordered helix (Hais) when less then two hydrogen bonds are formed, e.g. at helix ends or within a Pro induced a-helix bend, antiparallel /?strand (Sa) and parallel /?-strand (Sp), turn (T) and other undefined (U) structures (Williams, 1983). This method was used to determine the average secondary structure of various membrane proteins and membrane-incorporated peptides by Raman and IR spectroscopy (Vogel et al. 1985; Vogel & Jahnig, 1986a, b; Vogel & Gartner, 1987; Vogel, 1987; Williams, 1986; Braiman & Rothschild, 1988; Surewicz & Mantsch, 1988; Lee et al. 1991), with the same method of analysis being applied to CD spectra as well. In particular, Raman spectroscopy has been shown to give the most reliable results in the determination of the secondary structure of membrane proteins, as demonstrated in the case of bacteriorhodopsin and porin, where excellent agreement was obtained with the corresponding high-resolution protein structure derived from diffraction methods (Vogel & Jahnig, 1986a; Vogel, 1987). In the following, we will show for 3 examples how spectroscopic methods combined with structure prediction can give useful models of the folding of membrane proteins. Lactose permease of Escherichia coli is a hydrophobic, polytopic membrane protein that catalyses active transport of galactosides across the cytoplasmic membrane in symport with a proton (for reviews see Kaback et al. 1990). The primary structure is known and consists of 417 amino acid residues. The permease has been solubilized from the membrane, purified and reconstituted in lipid membranes. Due to genetic modifications, biochemical and biophysical studies lactose permease is one of the best characterized membrane transport protein. Yet, neither the three-dimensional protein structure nor the molecular mechanism of the sugar transport is clear at this time. The secondary structure of the lactose permease reconstituted in lipid membranes was determined by Raman spectroscopy (Fig. 1) to consist of 70% a-helix content, less than 10% ^-strands and 15 % /?-turns (Vogel et al. 1985). Changes induced in the Raman spectrum by exchanging H2O for D2O showed that about § of the residues in a-helices and most other residues are exposed to water. Also, the effect of substrate binding was investigated using /?-D-galactosyl-i-thio-/?-D-galactoside, which induced a slight increase of the a-helix content at the expense of /?-strand structure, which decreased slightly. Further effects of substrate binding were observed on several vibrational bands originating from vibrations of Glu and/or Asp, Tyr and Trp residues. In addition, the secondary structure and the folding of lactose permease was predicted. In general, the methods developed for the prediction of the structure of water soluble proteins are not reliable for the application to membrane proteins. In the latter case, however, the following simple considerations are useful to predict lipid bilayer-traversing protein segments (Vogel et al. 1985; Vogel & Jahnig, 1986 a): An intrinsic membrane protein is composed of membranespanning a-helices or /^-strands, which might be either totally hydrophobic or amphipathic i.e. consisting of a hydrophobic and a hydrophilic surface. Because
Structure and dynamics of polypeptides and proteins
437
intramolecular hydrogen bonds are formed at the polypeptide backbone either as intrachain bonds along a helix or as interchain bonds between neighbouring /?strands, these two structures can directly contact the apolar lipid bilayer with their hydrophobic surfaces. In order to traverse a lipid bilayer, a minimal number of 20 residues is required for an a-helix and about 10 residues for a /?-strand. Totally hydrophobic membrane-spanning segments were predicted according to Kyte & Doolittle (1982) by searching for hydrophobic stretches of about 10 to 20 residues along the protein sequence according to equation (1). H(i) = [h(i±3) + h(i±2) + h(i±i) + h(i)]/7.
(1)
A hydrophobic index h is attributed to each amino acid residue and the mean hydrophobicity H(i) along a protein sequence calculated as an average over seven neighbouring residues on the sequence position i. A membrane-spanning a-helix is predicted if H ^ i-6 over a region of about 20 residues. The values of i-6 is typical for membrane-spanning helices such as in bacteriorhodopsin as well as in the photosynthetic reaction centres. Amphipathic membrane-spanning a-helices or /^-structures were predicted by simply searching stretches of about 20 or 10 residues whose hydrophobicity varies with a period of 3'6 or two residues, respectively. Actually, the search is performed on a region of 20 or 10 residues by a summation of the hydrophobicities over one side of an a-helix or a /?-strand, respectively, according to equations (2) and (3). An amphipathic helix is predicted if HJi) > i-6 and Ha(i+2)
433
Structure and dynamics of polypeptides and proteins in lipid membranes
HORST VOGEL Swiss Federal Institute of Technology, Lausanne
1.
INTRODUCTION
433
2.
STRUCTURAL MODELS OF MEMBRANE PROTEINS DERIVED FROM RAMAN SPECTROSCOPY AND PREDICTION METHODS
3. STRUCTURAL DYNAMICS OF MODEL PEPTIDES
I.
434 441
4.
THE ROLE OF PROLINE RESIDUES IN TRANSMEMBRANE HELICES
5.
ACKNOWLEDGEMENT
6.
REFERENCES
445
453
453
INTRODUCTION
The elucidation of the molecular mechanisms whereby ions and polar molecules are translocated across the hydrophobic barrier of a lipid bilayer in biological membranes is one of the most challenging problems in biological research. Specific membrane proteins, such as pumps, carriers and channels, play the central role in the various translocation pathways. Recent progress in expression cloning has provided the sequence of a number of biologically important membrane proteins and in principle the door is open to investigate every protein which might be of importance in the central signal transduction and transport processes. Unfortunately, to date there are only a few examples where the threedimensional structure of membrane proteins are known at atomic resolution. The photosynthetic reaction centres from purple bacteria (Deisenhofer et al. 1985), bacteriorhodopsin (Henderson et al. 1990) and the large porin channel of Rhodobacter capsulata (Weiss et al. 1991). According to these structural data membrane proteins seem to fold in general in membrane-spanning a-helices and /?-strands in order to saturate hydrogen bonds. Only these two motifs seem to form stable structures which can be in contact with the hydrophobic lipid interior of a membrane. The hydrophobic or amphipathic segments may be arranged such that the hydrophobic surfaces are in contact with the hydrophobic part of the lipid Postal address: Ecole Polytechnique Federate de Lausanne, Institut de Chimie Physique, CH-1015 Lausanne, Switzerland.
434
H. Vogel
1500
1000 •
1630
1650 1670 Wavenumber [crrr'l
1690
Fig. i. 15-point representation of the amide I Raman spectra of a typical a-helical membrane protein, lactose permease (A), and two typical /^-structure membrane proteins, porin (O) and maltoporin ( • ) . From Vogel & Jahnig (1986a).
bilayer or other hydrophobic protein segments, whereas the hydrophilic protein surfaces are facing other hydrophilic protein segments in the interior of the membrane embedded protein. Such hydrophilic parts might create a central hydrophilic pathway for the membrane translocation of ions and polar molecules, yet the real existence of such structures has to be proven experimentally in each particular case. In spite of the lack of experimental evidence, detailed models have been published of how the polypeptide chains of nearly every sequenced membrane proteins are thought to be folded. These models are mainly based on the distribution of hydrophobic and hydrophilic regions within the protein sequence. Here, we summarize our work on the structure and dynamics of proteins and polypeptides in lipid membranes, based on optical spectroscopic measurements, secondary structure prediction and molecular dynamics calculations. 2. STRUCTURAL MODELS OF MEMBRANE PROTEINS DERIVED FROM RAMAN SPECTROSCOPY AND PREDICTION METHODS
In spite of intensive research, most of the investigated membrane proteins have resisted attempts to grow three-dimensional crystals of sufficient quality. In this situation, one may turn to other techniques which provide structural information, in general less than high resolution X-ray diffraction, but do not require threedimensional crystals. Spectroscopic techniques are ideally suited for this purpose. High-resolution, multidimensional NMR spectroscopy is not applicable to membrane proteins due to excessive protein immobilisation in lipid bilayers. In the case of small, membrane active proteins, this problem might be circumvented
Structure and dynamics of polypeptides and proteins
435
by incorporating the polypeptides in detergent micelles. Solid-state NMR of suitably labelled proteins might be an alternative to obtain the orientational distribution of the specially-labelled groups of the protein in a lipid bilayer, as has been shown recently in the case of filamenteous phage proteins in membranes (Opella et al. 1987; Shon et al. 1991). On the other hand, optical spectroscopy such as circular dichroism (CD), infrared absorption (IR), Raman scattering and fluorescence spectroscopy deliver information on protein structure and mobility. CD spectroscopy can provide both qualitative and quantitative assessment of overall protein secondary structure and has been shown to yield quite reliable results for the average a-helix content of proteins (Provencher & Glockner, 1981; Hennesey & Johnson, 1981). However, /?-turns and /?-strand structures are difficult to distinguish and the technique might give rise to artefacts in the case of membrane proteins (Vogel & Gartner, 1987). Vibrational spectroscopy such as Raman and IR spectroscopy has been shown in recent years to be a powerful tool for the determination of membrane protein structures (Vogel et al. 1985 ; Vogel & Jahnig, 1986a, b; Vogel & Gartner, 1987; Lee et al. 1991; Braiman & Rothschild, 1988; Surewicz & Mantsch, 1988). It has benefited from both the progress made in the calculation of protein spectra (Krimm & Bandekar, 1986) and the development of quantitative methods for the evaluation of structural information (Williams, 1983). In an IR or Raman spectrum of a protein, the amide I band in the region of 1600—1700 cm"1 is the most sensitive vibrational band for investigating the average protein secondary structure. The amide I band originates primarily from the C—O stretching vibrations of peptide amide groups coupled to the in plane N-H bending and C~N stretching modes (Krimm & Bandekar, 1986). The frequency of this vibration depends on the nature of the hydrogen bonding involving the C = O and N~H moieties, and therefore in turn the frequency of the amide I mode reflects the particular secondary structure adopted by the polypeptide backbone. Although, in principle, it would be possible to determine the average protein conformation and to distinguish the different structural elements from their particular amide I vibrational frequencies, this strategy is not straightforward in practice because the amide I band of proteins normally appears as a relatively structureless, broad band in the IR and Raman spectra. This is demonstrated in Fig. 1 showing the Raman amide I spectra of typical a-helix and /?-strand structure membrane proteins. Different resolution enhancement algorithms have been applied to solve the problem but this method is not free of artefacts, and furthermore, certain resolved amide I vibrational components cannot always be assigned with confidence to a particular protein structure (Surewicz & Mantsch, 1988). An alternative method has been successfully used for quantitative estimation of the protein secondary structure. The spectrum to be analysed is fitted by a linear combination of a set of reference proteins of known three-dimensional structure. The set of reference proteins is transformed to a set of orthogonal 'quasi' or eigenspectra which may be combined to generate the original data within experimental error. Since the structure of the reference proteins is known, the structure of the protein under investigation can be
436
H. Vogel
determined usually in percentages of different structure classes, such as ordered helix (Ho), when the amino acid residue participates in hydrogen bonds to two sides along the helix, disordered helix (Hais) when less then two hydrogen bonds are formed, e.g. at helix ends or within a Pro induced a-helix bend, antiparallel /?strand (Sa) and parallel /?-strand (Sp), turn (T) and other undefined (U) structures (Williams, 1983). This method was used to determine the average secondary structure of various membrane proteins and membrane-incorporated peptides by Raman and IR spectroscopy (Vogel et al. 1985; Vogel & Jahnig, 1986a, b; Vogel & Gartner, 1987; Vogel, 1987; Williams, 1986; Braiman & Rothschild, 1988; Surewicz & Mantsch, 1988; Lee et al. 1991), with the same method of analysis being applied to CD spectra as well. In particular, Raman spectroscopy has been shown to give the most reliable results in the determination of the secondary structure of membrane proteins, as demonstrated in the case of bacteriorhodopsin and porin, where excellent agreement was obtained with the corresponding high-resolution protein structure derived from diffraction methods (Vogel & Jahnig, 1986a; Vogel, 1987). In the following, we will show for 3 examples how spectroscopic methods combined with structure prediction can give useful models of the folding of membrane proteins. Lactose permease of Escherichia coli is a hydrophobic, polytopic membrane protein that catalyses active transport of galactosides across the cytoplasmic membrane in symport with a proton (for reviews see Kaback et al. 1990). The primary structure is known and consists of 417 amino acid residues. The permease has been solubilized from the membrane, purified and reconstituted in lipid membranes. Due to genetic modifications, biochemical and biophysical studies lactose permease is one of the best characterized membrane transport protein. Yet, neither the three-dimensional protein structure nor the molecular mechanism of the sugar transport is clear at this time. The secondary structure of the lactose permease reconstituted in lipid membranes was determined by Raman spectroscopy (Fig. 1) to consist of 70% a-helix content, less than 10% ^-strands and 15 % /?-turns (Vogel et al. 1985). Changes induced in the Raman spectrum by exchanging H2O for D2O showed that about § of the residues in a-helices and most other residues are exposed to water. Also, the effect of substrate binding was investigated using /?-D-galactosyl-i-thio-/?-D-galactoside, which induced a slight increase of the a-helix content at the expense of /?-strand structure, which decreased slightly. Further effects of substrate binding were observed on several vibrational bands originating from vibrations of Glu and/or Asp, Tyr and Trp residues. In addition, the secondary structure and the folding of lactose permease was predicted. In general, the methods developed for the prediction of the structure of water soluble proteins are not reliable for the application to membrane proteins. In the latter case, however, the following simple considerations are useful to predict lipid bilayer-traversing protein segments (Vogel et al. 1985; Vogel & Jahnig, 1986 a): An intrinsic membrane protein is composed of membranespanning a-helices or /^-strands, which might be either totally hydrophobic or amphipathic i.e. consisting of a hydrophobic and a hydrophilic surface. Because
Structure and dynamics of polypeptides and proteins
437
intramolecular hydrogen bonds are formed at the polypeptide backbone either as intrachain bonds along a helix or as interchain bonds between neighbouring /?strands, these two structures can directly contact the apolar lipid bilayer with their hydrophobic surfaces. In order to traverse a lipid bilayer, a minimal number of 20 residues is required for an a-helix and about 10 residues for a /?-strand. Totally hydrophobic membrane-spanning segments were predicted according to Kyte & Doolittle (1982) by searching for hydrophobic stretches of about 10 to 20 residues along the protein sequence according to equation (1). H(i) = [h(i±3) + h(i±2) + h(i±i) + h(i)]/7.
(1)
A hydrophobic index h is attributed to each amino acid residue and the mean hydrophobicity H(i) along a protein sequence calculated as an average over seven neighbouring residues on the sequence position i. A membrane-spanning a-helix is predicted if H ^ i-6 over a region of about 20 residues. The values of i-6 is typical for membrane-spanning helices such as in bacteriorhodopsin as well as in the photosynthetic reaction centres. Amphipathic membrane-spanning a-helices or /^-structures were predicted by simply searching stretches of about 20 or 10 residues whose hydrophobicity varies with a period of 3'6 or two residues, respectively. Actually, the search is performed on a region of 20 or 10 residues by a summation of the hydrophobicities over one side of an a-helix or a /?-strand, respectively, according to equations (2) and (3). An amphipathic helix is predicted if HJi) > i-6 and Ha(i+2)
