Structure-based statistical analysis of transmembrane helices

Recent advances in determination of the high-resolution structure of membrane proteins now enable analysis of the main features of amino acids in transmembrane (TM) segments in comparison with amino acids in water-soluble helices. In this work, we conducted a large-scale analysis of the prevalent locations of amino acids by using a data set of 170 structures of integral membrane proteins obtained from the MPtopo database and 930 structures of water-soluble helical proteins obtained from the protein data bank. Large hydrophobic amino acids (Leu, Val, Ile, and Phe) plus Gly were clearly prevalent in TM helices whereas polar amino acids (Glu, Lys, Asp, Arg, and Gln) were less frequent in this type of helix. The distribution of amino acids along TM helices was also examined. As expected, hydrophobic and slightly polar amino acids are commonly found in the hydrophobic core of the membrane whereas aromatic (Trp and Tyr), Pro, and the hydrophilic amino acids (Asn, His, and Gln) occur more frequently in the interface regions. Charged amino acids are also statistically prevalent outside the hydrophobic core of the membrane, and whereas acidic amino acids are frequently found at both cytoplasmic and extra-cytoplasmic interfaces, basic amino acids cluster at the cytoplasmic interface. These results strongly support the experimentally demonstrated biased distribution of positively charged amino acids (that is, the so-called the positive-inside rule) with structural data.     

Structural features of transmembrane helices

A total of 160 transmembrane helices of 15 non-homologous high-resolution X-ray protein structures have been analyzed in respect of their structural features. The dihedral angles and hydrogen bonds of the helical sections that span the hydrophobic interior of the lipid bilayer have been investigated. The Ramachandran plot of protein channels and solute transporters exhibit a significant shift Delta (phi- and psi-angles) of Delta mean (+4.5 degrees and -5.4 degrees ), compared to a reference group of 151 alpha-helices of the same average length derived from water-soluble globular proteins. At the C-termini of transmembrane helices structural motifs equivalent to the Gly-caps of helices in globular proteins have been found, with two third of the transmembrane Gly-caps taking up a primary structure that is typically not found at helix termini exposed to a polar solvent. The structural particularities reported here are relevant for the three-dimensional modelling of membrane protein structures.     

Proline-induced distortions of transmembrane helices

Proline residues in the transmembrane (TM) alpha-helices of integral membrane proteins have long been suspected to play a key role for helix packing and signal transduction by inducing regions of helix distortion and/or dynamic flexibility (hinges). In this study we try to characterise the effect of proline on the geometric properties of TM alpha-helices. We have examined 199 transmembrane alpha-helices from polytopic membrane proteins of known structure. After examining the location of proline residues within the amino acid sequences of TM helices, we estimated the helix axes either side of a hinge and hence identified a hinge residue. This enabled us to calculate helix kink and swivel angles. The results of this analysis show that proline residues occur with a significant concentration in the centre of sequences of TM alpha-helices. In this location, they may induce formation of molecular hinges, located on average about four residues N-terminal to the proline residue. A superposition of proline-containing TM helices structures shows that the distortion induced is anisotropic and favours certain relative orientations (defined by helix kink and swivel angles) of the two helix segments.     

A turn propensity scale for transmembrane helices.

Using a model protein with a 40 residue hydrophobic transmembrane segment, we have measured the ability of all the 20 naturally occurring amino acids to form a tight turn when placed in the middle of the hydrophobic segment. Turn propensities in a transmembrane helix are found to be markedly different from those of globular proteins, and in most cases correlate closely with the hydrophobicity of the residue. The turn propensity scale may be used to improve current methods for membrane protein topology prediction.     

Tryptophan supports interaction of transmembrane helices

Interactions of transmembrane helices play an important role in folding and oligomerization of integral membrane proteins. The interfacial residues of these helices frequently correspond to heptad repeat motifs. In order to uncover novel mechanisms underlying these interactions, we randomised a heptad repeat pattern with a complete set of amino acids. Those sequences that were capable of high-affinity self-interaction upon integration into bacterial inner membranes were selected by means of the POSSYCCAT system. A comparison between selected and non-selected sequences reveals that high-affinity sequences were strongly enriched in tryptophan residues that accumulated at specific positions of the heptad motif. Mutation of Trp in selected clones significantly reduced self-interaction of the transmembrane segments without affecting their efficiency of membrane integration. Conversely, grafting Trp onto artificial transmembrane segments strongly enhanced their interaction. We conclude that tryptophan supports interaction of transmembrane segments.     

The stability of transmembrane helices: A molecular dynamics study on the isolated helices of bacteriorhodopsin

Bacteriorhodopsin (bR) continues to be a proven testing ground for the study of integral membrane proteins (IMPs). It is important to study the stability of the individual helices of bR, as they are postulated to exist as independently stable transmembrane helices (TMHs) and also for their utility as templates for modeling other IMPs with the postulated seven-helix bundle topology. Toward this purpose, the seven helices of bR have been studied by molecular dynamics simulation in this study. The suitability of using the backbone-dependent rotamer library of side-chain conformations arrived at from the data base of globular protein structures in the case TMHs has been tested by another set of 7 helix simulations with the side-chain orientations taken from this library. The influence of the residue's net charge on the helix stability was examined by simulating the helices III, IV, and VI (from both of the above sets of helices) with zero net charge on the side chains. The results of these 20 simulations demonstrate in general the stability of the isolated helices of bR in conformity with the two-stage hypothesis of IMP folding. However, the helices I, II, V, and VII are more stable than the other three helices. The helical nature of certain regions of III, IV, and VI are influenced by factors such as the net charge and orientation of several residues. It is seen that the residues Arg, Lys, Asp, and Glu (charged residues), and Ser, Thr, Gly, and Pro, play a crucial role in the stability of the helices of bR. The backbone-dependent rotamer library for the side chains is found to be suitable for the study of TMHs in IMP. © 1996 John Wiley & Sons, Inc.     

TMSEG: Novel prediction of transmembrane helices

Transmembrane proteins (TMPs) are important drug targets because they are essential for signaling, regulation, and transport. Despite important breakthroughs, experimental structure determination remains challenging for TMPs. Various methods have bridged the gap by predicting transmembrane helices (TMHs), but room for improvement remains. Here, we present TMSEG, a novel method identifying TMPs and accurately predicting their TMHs and their topology. The method combines machine learning with empirical filters. Testing it on a non‐redundant dataset of 41 TMPs and 285 soluble proteins, and applying strict performance measures, TMSEG outperformed the state‐of‐the‐art in our hands. TMSEG correctly distinguished helical TMPs from other proteins with a sensitivity of 98 ± 2% and a false positive rate as low as 3 ± 1%. Individual TMHs were predicted with a precision of 87 ± 3% and recall of 84 ± 3%. Furthermore, in 63 ± 6% of helical TMPs the placement of all TMHs and their inside/outside topology was correctly predicted. There are two main features that distinguish TMSEG from other methods. First, the errors in finding all helical TMPs in an organism are significantly reduced. For example, in human this leads to 200 and 1600 fewer misclassifications compared to the second and third best method available, and 4400 fewer mistakes than by a simple hydrophobicity‐based method. Second, TMSEG provides an add‐on improvement for any existing method to benefit from. Proteins 2016; 84:1706–1716. © 2016 Wiley Periodicals, Inc.     

Evidence for hetero-association of transmembrane helices of integrins

Integrins are important transmembrane cell-surface receptors, which mediate interactions of the cell with other cells or the extracellular matrix. Integrins are heterodimers composed of an alpha- and a beta-subunit. They can switch between different activation states depending on intra- or extracellular signals. Inside/out and outside/in signaling is mediated via integrins across the membrane. A biologically important and yet still unanswered question is the role of the transmembrane domains in the signaling event. Here it is shown by simulated annealing/molecular dynamics calculations that recently published structural data of the cytoplasmic domains of integrin alphaIIbbeta3 are supporting a structure with interacting transmembrane helices. This corroborates a model of transmembrane domains that are actively involved in the transmembrane signaling event.     

Identifying interactions between transmembrane helices from the adenosine A2A receptor

The human adenosine A(2A) receptor (A(2A)R) is an integral membrane protein and a member of the G-protein-coupled receptor (GPCR) superfamily, characterized by seven transmembrane (TM) helices. Although helix-helix association in the lipid bilayer is known to be an essential step in the folding of GPCRs, the determinants of their structures, folding, and assembly in the cell membrane are poorly understood. Previous studies in our group showed that while peptides corresponding to all seven TM domains of A(2A)R form stable helical structures in detergent micelles and lipid vesicles, they display significant variability in their helical propensity. This finding suggested to us that some TM domains might need to interact with other domains to properly insert and fold in hydrophobic environments. In this study, we assessed the ability of TM peptides to interact in pairwise combinations. We analyzed peptide interactions in hydrophobic milieus using circular dichroism spectroscopy and F?rster resonance energy transfer. We find that specific interactions between TM helices occur, leading to additional helical content, especially in weakly helical TM domains, suggesting that some TM domains need a partner for proper folding in the membrane. The approach developed in this study will enable complete analysis of the TM domain interactions and the modeling of a folding pathway for A(2A)R.     

Chaperoning transmembrane helices in the lipid bilayer

Elimination of membrane proteins often requires recognition of their transmembrane domains (TMDs) in the lipid bilayer. In this issue, Arines et al. (2020. J. Cell Biol. https://doi.org/10.1083/jcb.202001116) show that in Saccharomyces cerevisiae, the vacuole-associated Rsp5 ubiquitin ligase uses a TMD in substrate adaptor Ssh4 to recognize membrane helices in Ypq1, which targets this lysine transporter for lysosomal degradation during lysine starvation.

In eukaryotic cells, protein quality control (PQC) mediates the degradation of not only aberrant but also unwanted polypeptides, safeguarding both the quality and quantity of the cellular proteome (1). A central goal in PQC research is to delineate the mechanism of substrate selection, which, if inappropriately executed, could lead to undesired destruction of functional proteins and thus the collapse of the proteostasis network. For soluble proteins that succumb to PQC, it is usually the surface exposure of hydrophobic elements that alerts cellular chaperones to potential folding catastrophe (2). Chaperones often serve a dual triaging role: while giving their clients additional time to fold, they can also interface with degradation machineries such as the ubiquitin proteasome system or lysosomes, causing the elimination of terminally misfolded or unwanted polypeptides.Unlike PQC of soluble proteins, substrate recognition for membrane proteins bearing abnormal transmembrane domain (TMD) is largely unknown, even for the best characterized PQC process, ER-associated degradation (ERAD; 3). Early studies on PQC of unassembled T cell receptor α chain (TCRα) showed that the single TMD of TCRα contains two charged residues, which are thermodynamically unfavored in the lipid environment and thus must be shielded when TCRα assembles with CD3σ. Accordingly, unassembled TCRα is eliminated by ERAD via a mechanism dependent on these charged residues (4), but TMD-specific chaperones responsible for recognizing charged residues in the lipid bilayer have not been identified. Likewise, recent investigations into the function of the Hrd1 ubiquitin ligase suggested a role for the TMDs of Hrd1 in recognition of specific aberrant membrane proteins in ERAD (5). Cryo-EM studies further showed two juxtaposed central cavities with a lateral gate poised to receive TMDs in the yeast Hrd1 complex (6), but how aberrant TMDs in ERAD substrates are recognized and retrotranslocated by Hrd1 remains an open question.The issue of substrate recognition becomes even more complex for feedback-regulated degradation of unwanted membrane proteins. In this case, substrates are initially stable and functionally essential, but a change in environmental cues renders them dispensable and results in a short-lived fate. One such example is the sterol-regulated degradation of a sterol-synthesizing enzyme called HMG-CoA reductase (HMGR). HMGR is a stable ER protein when the sterol level is low, but an increase in membrane sterol abundance alters the conformation of a sterol-sensing domain in HMGR, exposing an element functionally equivalent to a degron in short-lived proteasomal substrates (7). Despite extensive studies, the molecular signature of the degron in HMGR is still undefined, let alone the molecular basis of its recognition. In this issue, Arines and colleagues investigate how Ypq1, a multi-spanning lysine transporter of the yeast vacuole, is regulated by lysine availability, a regulated membrane protein turnover event analogous to HMGR degradation. Their study identifies critical residues in Ypq1 TMDs for its turnover and establishes the Rsp5 ubiquitin ligase adaptor Ssh4 as a TMD-specific chaperone that recognizes these elements (8).Ypq1 is a seven-transmembrane, PQ loop–containing lysine transporter localized to the yeast vacuole membrane. Under lysine-replete conditions, Ypq1 is stable as it uses a PQ loop–dependent conformational cycle to import excess lysine into the vacuole. When lysine is depleted, Ypq1 is sorted into the multivesicular body (MVB) for degradation (Fig. 1). This process is initiated once Ypq1 is ubiquitinated by the ubiquitin ligase complex Rsp5–Ssh4, but how Ypq1 is targeted by Rsp5–Ssh4 has been unclear (9).Open in a separate windowFigure 1.Regulated recognition of Ypq1 by Ssh4. When lysine in the cytosol is abundant, Ypq1 undergoes a rapid conformational cycle to transport lysine from the cytosol into the vacuole lumen. Under lysine-depleted conditions, the transporter is trapped in a conformation recognizable by Ssh4, which recruits Rsp5 to catalyze Ypq1 ubiquitination and internalization into the MVB. Ub, ubiquitin.To understand the mechanism of Ypq1 recognition, Arines et al. first engineered a Ypq1 mutant that uncouples ligase-mediated degradation from lysine availability. This Ypq1 mutant is constitutively degraded in an Ssh4-dependent manner even under lysine-replete conditions. With this tool in hand, they performed a random mutagenesis-based suppressor screen, which identified many suppressor mutants. Mapping these mutations revealed several elements in Ypq1 that are critical for ligase recognition, which include two TMDs (TM5 and TM7) and a cytosolic loop. Importantly, when these mutations were introduced back into wild-type Ypq1, they also block Ssh4-dependent, lysine-regulated Ypq1 degradation. As expected, coimmunoprecipitation showed that Ypq1 suppressor mutants have reduced affinity to Ssh4. Since the cytosolic loop contains a previously known Rsp5 recognition motif, they further characterized the role of Ypq1 TMDs in ligase recruitment.Structural modeling suggests that TM5 and TM7 are juxtaposed to each other. Systematic mutagenesis targeting each residue of these two TMDs further consolidated the residues essential for Ssh4-mediated degradation. A similar mutagenesis study on Ssh4 revealed an important role for the Ssh4 TMD in Ypq1 degradation. Interestingly, for both Ssh4 and Ypq1, many identified residues are clustered on one side of the membrane helices. Arines et al. propose that Ssh4 uses its TMD to recognize TM5 and TM7 in Ypq1 based on a charge complementation experiment: a charged residue introduced into TM5 of Ypq1 abolished Ssh4-mediated degradation, but introducing an opposite charge into the TMD of Ssh4 restored lysine-regulated Ypq1 degradation.The recognition of Ypq1 by Ssh4 appears to occur when Ypq1 adopts a specific conformation during lysine transport because charge complementarity–based degradation of Ypq1 depends on the PQ loop, which is required for lysine transport. Additionally, structural modeling of Ypq1 suggested that in the inward-open and occluded conformations, TM5 is packed against TM7, but the two TMDs become distant from each other in the outward-open conformation, which exposes residues critical for Ssh4 recognition. These findings suggest that the rapid conformational cycling during lysine transport may prevent Ssh4 recognition, but lysine depletion stalls Ypq1 in a conformation recognizable by Ssh4 (Fig. 1).The study, together with the recent discovery of the ER membrane protein complex (EMC) in the biogenesis of multi-spanning membrane proteins at the ER, suggests a new class of chaperones that recognize specific features in TMDs. While emerging evidence suggests that the EMC recognizes exposed charged or polar residues in TMDs (10), the molecular basis of Ssh4 substrate interaction remains unclear. Like cytosolic chaperones, the EMC at the ER appears to play a dual role: while initially shielding charged/polar residues to facilitate TMD assembly, it may eventually target misassembled membrane proteins for degradation. By contrast, TMD-specific chaperones in other organelles like Ssh4 may have a more dedicated function in PQC. Clearly, more TMD-specific chaperones await to be discovered. Additionally, future studies will surely reveal not only the range of substrates and functions for each TMD-specific chaperone but also the structural basis of TMD recognition.     

A specific interface between integrin transmembrane helices and affinity for ligand

Conformational communication across the plasma membrane between the extracellular and intracellular domains of integrins is beginning to be defined by structural work on both domains. However, the role of the α and β subunit transmembrane domains and the nature of signal transmission through these domains have been elusive. Disulfide bond scanning of the exofacial portions of the integrin α_IIβ and β₃ transmembrane domains reveals a specific heterodimerization interface in the resting receptor. This interface is lost rather than rearranged upon activation of the receptor by cytoplasmic mutations of the α subunit that mimic physiologic inside-out activation, demonstrating a link between activation of the extracellular domain and lateral separation of transmembrane helices. Introduction of disulfide bridges to prevent or reverse separation abolishes the activating effect of cytoplasmic mutations, confirming transmembrane domain separation but not hinging or piston-like motions as the mechanism of transmembrane signaling by integrins.     

Calculating the free energy of association of transmembrane helices

A large number of experimental studies have been devoted to quantifying the interaction between transmembrane (TM) helices in detergent micelles and, more recently, in bilayers. Theoretical calculation of association free energy of TM helices would be useful for predicting the propensity of given sequences to oligomerize and for understanding the difference between association in micelles and in bilayers. In this article, the theoretical foundation for calculating the standard association free energy of TM helices is laid out and is applied to glycophorin A in both micelles and bilayers. The standard association free energy is decomposed into the effective energy, translational, rotational, and conformational entropy terms. The effective energy of association is obtained by molecular dynamics simulations in an implicit membrane model. The translational and rotational entropy of association is calculated from the probability distribution of the translational and rotational degrees of freedom obtained from the molecular dynamics simulations. The side-chain conformational entropy of association is estimated from the probability distribution obtained by rigid rotation of all side-chain dihedral angles. The calculated standard association free energy of glycophorin A in N-dodecylphosphocholine micelles is in good agreement with the experimental value. The translational entropy cost is larger, whereas the rotational entropy cost is smaller in bilayers than in micelles. The standard association free energy in 1,2-dimyristoyl-sn-glycero-3-phosphocholine bilayers is calculated to be approximately 1.3 kcal/mol more favorable than in N-dodecylphosphocholine micelles, consistent with available experimental data.     

A synergistic effect between cholesterol and tryptophan-flanked transmembrane helices modulates membrane curvature

The aim of this study was to gain insight into the structural consequences of hydrophobic mismatch for membrane proteins in lipid bilayers that contain cholesterol. For this purpose, tryptophan-flanked peptides, designed to mimic transmembrane segments of membrane proteins, were incorporated in model membranes of unsaturated phosphatidylcholine bilayers of varying thickness and containing varying amounts of cholesterol. Analysis of the lipid organization by (31)P NMR and cryo-TEM demonstrated the formation of an isotropic phase, most likely representing a cubic phase, which occurred exclusively in mixtures containing lipids with relatively long acyl chains. Formation of this phase was inhibited by incorporation of lysophosphatidylcholine. These results indicate that the isotropic phase is formed as a consequence of negative hydrophobic mismatch and that its formation is related to a negative membrane curvature. When either peptide or cholesterol was omitted from the mixture, isotropic-phase formation did not occur, not even when the concentrations of these compounds were significantly increased. This suggests that formation of the isotropic phase is the result of a synergistic effect between the peptides and cholesterol. Interestingly, isotropic-phase formation was not observed when the tryptophans in the peptide were replaced by either lysines or histidines. We propose a model for the mechanism of this synergistic effect, in which its dependence on the flanking residues is explained by preferential interactions between cholesterol and tryptophan residues.     

Optimal bundling of transmembrane helices using sparse distance constraints

We present a two-step approach to modeling the transmembrane spanning helical bundles of integral membrane proteins using only sparse distance constraints, such as those derived from chemical cross-linking, dipolar EPR and FRET experiments. In Step 1, using an algorithm, we developed, the conformational space of membrane protein folds matching a set of distance constraints is explored to provide initial structures for local conformational searches. In Step 2, these structures refined against a custom penalty function that incorporates both measures derived from statistical analysis of solved membrane protein structures and distance constraints obtained from experiments. We begin by describing the statistical analysis of the solved membrane protein structures from which the theoretical portion of the penalty function was derived. We then describe the penalty function, and, using a set of six test cases, demonstrate that it is capable of distinguishing helical bundles that are close to the native bundle from those that are far from the native bundle. Finally, using a set of only 27 distance constraints extracted from the literature, we show that our method successfully recovers the structure of dark-adapted rhodopsin to within 3.2 A of the crystal structure.     

The determinants of hydrophobic mismatch response for transmembrane helices

Hydrophobic mismatch arises from a difference in the hydrophobic thickness of a lipid membrane and a transmembrane protein segment, and is thought to play an important role in the folding, stability and function of membrane proteins. We have investigated the possible adaptations that lipid bilayers and transmembrane α-helices undergo in response to mismatch, using fully-atomistic molecular dynamics simulations totaling 1.4 μs. We have created 25 different tryptophan-alanine-leucine transmembrane α-helical peptide systems, each composed of a hydrophobic alanine–leucine stretch, flanked by 1–4 tryptophan side chains, as well as the β-helical peptide dimer, gramicidin A. Membrane responses to mismatch include changes in local bilayer thickness and lipid order, varying systematically with peptide length. Adding more flanking tryptophan side chains led to an increase in bilayer thinning for negatively mismatched peptides, though it was also associated with a spreading of the bilayer interface. Peptide tilting, bending and stretching were systematic, with tilting dominating the responses, with values of up to ~ 45° for the most positively mismatched peptides. Peptide responses were modulated by the number of tryptophan side chains due to their anchoring roles and distributions around the helices. Potential of mean force calculations for local membrane thickness changes, helix tilting, bending and stretching revealed that membrane deformation is the least energetically costly of all mismatch responses, except for positively mismatched peptides where helix tilting also contributes substantially. This comparison of energetic driving forces of mismatch responses allows for deeper insight into protein stability and conformational changes in lipid membranes.     

Computational differentiation of N-terminal signal peptides and transmembrane helices

Signal peptides and transmembrane helices both contain a stretch of hydrophobic amino acids. This common feature makes it difficult for signal peptide and transmembrane helix predictors to correctly assign identity to stretches of hydrophobic residues near the N-terminal methionine of a protein sequence. The inability to reliably distinguish between N-terminal transmembrane helix and signal peptide is an error with serious consequences for the prediction of protein secretory status or transmembrane topology. In this study, we report a new method for differentiating protein N-terminal signal peptides and transmembrane helices. Based on the sequence features extracted from hydrophobic regions (amino acid frequency, hydrophobicity, and the start position), we set up discriminant functions and examined them on non-redundant datasets with jackknife tests. This method can incorporate other signal peptide prediction methods and achieve higher prediction accuracy. For Gram-negative bacterial proteins, 95.7% of N-terminal signal peptides and transmembrane helices can be correctly predicted (coefficient 0.90). Given a sensitivity of 90%, transmembrane helices can be identified from signal peptides with a precision of 99% (coefficient 0.92). For eukaryotic proteins, 94.2% of N-terminal signal peptides and transmembrane helices can be correctly predicted with coefficient 0.83. Given a sensitivity of 90%, transmembrane helices can be identified from signal peptides with a precision of 87% (coefficient 0.85). The method can be used to complement current transmembrane protein prediction and signal peptide prediction methods to improve their prediction accuracies.     

Asymmetric stability among the transmembrane helices of lactose permease

Combining structure determinations from nuclear magnetic resonance (NMR) data and molecular dynamics simulations (MD) under the same environmental conditions revealed a startling asymmetry in the intrinsic conformational stability of secondary structure in the transmembrane domain of lactose permease (LacY). Eleven fragments, corresponding to transmembrane segments (TMs) of LacY, were synthesized, and their secondary structure in solution was determined by NMR. Eight of the TMs contained significant regions of helical structure. MD simulations, both in DMSO and in a DMPC bilayer, showed sites of local stability of helical structure in these TMs, punctuated by regions of conformational instability, in substantial agreement with the NMR data. Mapping the stable regions onto the crystal structure of LacY reveals a marked asymmetry, contrasting with the pseudosymmetry in the static structure: the secondary structure in the C-terminal half is more stable than in the N-terminal half. The relative stability of secondary structure is likely exploited in the transport mechanism of LacY. Residues supporting proton conduction are in more stable regions of secondary structure, while residues key to substrate binding are found in considerably unstable regions of secondary structure.     

MemBrain: improving the accuracy of predicting transmembrane helices

Prediction of transmembrane helices (TMH) in alpha helical membrane proteins provides valuable information about the protein topology when the high resolution structures are not available. Many predictors have been developed based on either amino acid hydrophobicity scale or pure statistical approaches. While these predictors perform reasonably well in identifying the number of TMHs in a protein, they are generally inaccurate in predicting the ends of TMHs, or TMHs of unusual length. To improve the accuracy of TMH detection, we developed a machine-learning based predictor, MemBrain, which integrates a number of modern bioinformatics approaches including sequence representation by multiple sequence alignment matrix, the optimized evidence-theoretic K-nearest neighbor prediction algorithm, fusion of multiple prediction window sizes, and classification by dynamic threshold. MemBrain demonstrates an overall improvement of about 20% in prediction accuracy, particularly, in predicting the ends of TMHs and TMHs that are shorter than 15 residues. It also has the capability to detect N-terminal signal peptides. The MemBrain predictor is a useful sequence-based analysis tool for functional and structural characterization of helical membrane proteins; it is freely available at http://chou.med.harvard.edu/bioinf/MemBrain/.