Polar Residues and Their Positional Context Dictate the Transmembrane Domain Interactions of Influenza A Neuraminidases

Background: Transmembrane domain (TMD) interactions in bitopic proteins are less understood than in multispanning proteins.

Results: The interaction of the neuraminidase TMDs from influenza A viruses increases with decreasing hydrophobicity.

Conclusion: Neuraminidase TMD interactions are dependent on the helix localization and positioning of their polar residues in the membrane bilayer.

Significance: Polar-mediated TMD interactions are related to their membrane-integration properties.

Abstract

Interactions that facilitate transmembrane domain (TMD) dimerization have been identified mainly using synthetic TMDs. Here, we investigated how inherent properties within natural TMDs modulate their interaction strength by exploiting the sequence variation in the nine neuraminidase subtypes (N1–N9) and the prior knowledge that a N1 TMD oligomerizes. Initially, consensus TMDs were created from the influenza A virus database, and their interaction strengths were measured in a biological membrane system. The TMD interactions increased with respect to decreasing hydrophobicity across the subtypes (N1–N9) and within the human N1 subtype where the N1 TMDs from the pandemic H1N1 strain of swine origin were found to be significantly less hydrophobic. The hydrophobicity correlation was attributed to the conserved amphipathicity within the TMDs as the interactions were abolished by mutating residues on the polar faces that are unfavorably positioned in the membrane. Similarly, local changes enhanced the interactions only when a larger polar residue existed on the appropriate face in an unfavorable membrane position. Together, the analysis of this unique natural TMD data set demonstrates how polar-mediated TMD interactions from bitopic proteins depend on which polar residues are involved and their positioning with respect to the helix and the membrane bilayer.

Introduction

Approximately 25% of human genes encode membrane proteins, and these are equally divided between bitopic proteins with a single transmembrane domain (TMD)³ and proteins with multiple TMDs (1, 2). In both cases, the TMDs can be involved in a range of crucial functions from protein topology and trafficking, to assembly, transport across the membrane, signal transduction, and energy production. To acquire these functions, newly synthesized TMDs must first partition to the membrane where they can fold, insert, and make lateral interactions (3–6). Currently, how TMDs from bitopic proteins fold and interact laterally is less characterized than in multispanning transmembrane proteins.

The properties that contribute to TMD interactions differ from those for soluble proteins as they take place in the hydrophobic environment of the membrane. Initially, TMD interactions were described as “knobs into holes” where the hydrophobic residues or “knobs” of one TMD helix pack into the “holes” of an adjacent TMD helix thereby limiting the disorder to the linear lipid alkyl side chains (5, 7, 8). More recently, TMDs have been shown to interact through salt bridges and hydrogen bonds between polar and charged side chains, which lower the energetic cost of positioning these groups in the hydrophobic membrane (9–11). In addition, the membrane composition, membrane protein concentration, and the juxtamembrane sequences can also influence TMD interactions (12, 13).

These fundamental concepts have provided a basis for understanding TMD interactions and led to the identification of several TMD interaction motifs (8, 14). However, these motifs are dependent on their surrounding sequence and local environment in unpredictable manners, indicating that our understanding of how TMDs associate with specificity in limited sequence space remains rudimentary (8, 12, 15–17).

The positional analysis of amino acids in an α-helix led to the identification of amphipathic properties within these segments (18). By segregating the hydrophobic and polar residues on opposing faces, amphipathic α-helices can bind lipids, insert into membranes, contribute to membrane curvature, create channels in the membrane, and regulate protein function and signaling (19, 20). Based on their functional diversity, amphipathic helices are found in a variety of proteins either as a TMD or soluble helix. This includes influenza A viruses (IAVs), where M2 uses an amphipathic TMD to assemble into a tetrameric ion channel and a neighboring soluble amphipathic helix to facilitate cholesterol-dependent budding and membrane scission (21, 22).

In addition to M2, IAVs also encode for HA, a type I membrane glycoprotein, and neuraminidase, a type II membrane glycoprotein (23). HA mediates host cell binding and fusion of the viral envelope, whereas neuraminidase promotes viral release and prevents interparticle aggregation by removing the cell surface receptor (sialic acid) to which HA binds (24–26). Neuraminidase is attached to the viral envelope by an N-terminal TMD that is located within its signal anchor sequence (27). Thus, the TMD targets neuraminidase to the endoplasmic reticulum and dictates its topology, but it also contributes to its assembly and directs apical plasma membrane sorting (28–30). All of these functions are crucial for viral entry, replication, and propagation.

Numerous studies have shown that many different residues can facilitate TMD interactions with synthetic and individual protein TMDs, which can loosely be grouped as polar and nonpolar (for review, see Refs. 8, 14, 16). Recently, it has been quantitatively demonstrated how each of the 20 amino acids vary in their contribution to membrane insertion based on their positioning in the membrane bilayer (31, 32). Here, we combined these two principles to investigate whether a correlation exists between the strength of TMD oligomerization and their membrane integration properties using the large number of natural neuraminidase TMDs from the IAV database. To do this, the interaction strength was measured for consensus TMDs created for the nine subtypes of neuraminidase (N1–N9) found in avian IAVs and the two subtypes, N1 and N2, from human IAVs.

Based on our results, it was determined that the neuraminidase TMD interactions are dictated by their polar properties and this correlated with a conserved amphipathic pattern of two adjacent polar faces within the TMDs (except for those from N2). Using this information, the polar residues responsible for the interactions were identified based on their location within the TMD helix and how favorable their positioning is in the membrane with respect to integration. The sequence context of the interactions was demonstrated using the hydrophobic variation in the human N1 TMDs. These results provide the first example of a predictable TMD interaction strength for a protein and put polar mediated TMD interactions in the context of their polar residue positioning within the helix and the membrane bilayer.

EXPERIMENTAL PROCEDURES

Plasmids and Constructs

All mammalian expression vectors were created by PCR overlap cloning (33) using pcDNA3.1A-myc-His (Invitrogen) containing full-length neuraminidase from influenza A/WSN/33 (H1N1) attached to the C-terminal myc-His tag. For N1_1–74, N1_1–49, N1_1–40, and N1_1–35 the indicated residues were fused to the C-terminal myc epitope. The GALLEX constructs were derived from the pBLM-GpA plasmid kindly provided by Dirk Schneider (Johannes Gutenberg University Mainz). All neuraminidase TMD-encoding segments, with the indicated juxtamembrane sequences listed in Table 1, were inserted between lexA and MalE by overlap PCR cloning. Face mutations were created by site-directed mutagenesis, and all constructs were verified by sequencing (Eurofins MWG Operon).

TABLE 1.

Properties of the IAV neuraminidase TMDs analyzed in this study

Listed below are the consensus sequences for amino acids 2–35 for each avian (a) and human (h) neuraminidase subtype (N1–N9) created from the IAV database. The TMDs in each consensus sequence were determined by their ΔG_app for membrane insertion and are displayed in upper case. The juxtamembrane sequence is in lower case, the amino acid substitutions in the mutants are shown in bold, and the TMD face of each residue is in italics at the top. Listed to the right of each sequence is their hydrophobicity (ΔG_app in kcal/mol) and the prevalence of the consensus sequence in the unique neuraminidase protein sequences obtained from the IAV database for the indicated avian and human subtypes.

Transfections and Immunostaining

Vero cells (ATCC) were cultured in DMEM with 10% FBS, 100 units/ml penicillin and streptomycin and maintained at 37 °C in a 5% CO₂ humidified incubator. The Vero cells were transfected with the indicated constructs using the transfection reagent LT-1 (Mirius) and seeded onto coverslips in 3.5-cm dishes (28). 24 h after transfection, the coverslips were washed with PBS and fixed with ice-cold methanol (10 min at 4 °C). Cells were rehydrated in PBST-Az (PBS, pH 7.4, 0.1% Tween 20, 0.2 mg/liter sodium azide) for 1 h at 4 °C and stained with a monoclonal myc antibody (Cell Signaling) at 1:50 in 1% BSA PBST-Az for 1 h at 37 °C. After three PBST-Az washes, coverslips were incubated with Alexa Fluor 488 secondary (Invitrogen) for 1 h at 37 °C in the dark and mounted with Vectashield. Images were acquired with a Zeiss LSM700 inverted microscope and processed with Zen software.

GALLEX Assay

SU101 Escherichia coli cells with a constitutively expressed lacZ gene regulated by the LexA operator were provided by Dirk Schneider with an established protocol (34). Fresh transformations of SU101 cells with each pBLM expression plasmid were grown overnight in 100 μg/ml ampicillin. Cultures were back diluted to A_{600 nm} = 0.1 in LB with 100 μg/ml ampicillin and induced with the indicated IPTG concentration at 37 °C for 2.5 h until A_{600 nm} ≈ 0.6. β-Galactosidase activity was determined as described previously (35). β-Galactosidase activity (X) from each sample (X_S) was then normalized to that of cells containing an empty vector (X_V) where X_Norm = X_S/X_V and the relative interaction (R.I.) strength was calculated by the formula R.I. = 100%(1 − X_Norm).

Immunoblotting

At the indicated time after induction with IPTG, the SU101 E. coli (2 ml of A_{600 nm} ≈ 0.6) was sedimented at 15,000 × g for 5 min, resuspended in 200 μl of reducing (0.1 m DTT) Laemmli sample buffer, and subjected to a short sonication on ice. The samples were heated to 37 °C for 5 min, separated by SDS-PAGE, and transferred to a PVDF membrane. The membranes were processed using standard immunoblotting protocols with a monoclonal maltose-binding protein (MBP) antibody (NEB). Immunoblots were developed with ECL-prime (GE Healthcare), and images were acquired with a CCD camera (Fuji).

Consensus Neuraminidase TMD Sequences

The available neuraminidase protein sequences from IAVs were retrieved in August 2012 from the NCBI Influenza Virus Resource Database and grouped according to host (avian (a), human (h)) and subtype (N1–N9). Within these groups, the TMDs were all predicted to exist within the first 35 amino acids by the ΔG predictor (31), so each sequence was truncated to the first 35 amino acids, redundant sequences were removed, and consensus sequences were created. The initial hN1 TMD represents the consensus TMD for the human H1N1 viruses from 1918 to 1957, which is equivalent to the TMD found in the influenza A/WSN/33 strain.

Separation of the hN1 and hN2 Sequences and Calculation of Their ΔG_app-face Values

Based on when the strains were sequenced from the human population and their strain lineage, the hN1 sequences were separated into three data sets (hN1_H1N1 1918–1957, hN1_H1N1 1977–2011, and pandemic hN1_pH1N1 2009–present). The hN2 sequences were separated into two data sets (hN2_H2N2 1957–1968 and hN2_H3N2 1968–present). Consensus TMDs were then generated from the unique TMD sequences in each set. To calculate the hydrophobicity for each face of the neuraminidase TMDs (ΔG_app-face), each TMD was treated as an α-helix with seven faces (designated a to g), and the positional ΔG^aa(i)_app values (32) were summed for each face.

Statistics

Error bars represent S.D. from three or more experiments, or from all unique TMDs of each set. The R² and p value of the linear correlation were computed using the square of Pearson's correlation coefficient. Two-tailed p values were calculated from unequal variance t tests (*, p ≤ 0.05; **, p ≤ 0.01; ***, p ≤ 0.001; ****, p ≤ 0.0001). All statistical calculations were made with GraphPad Prism 6.0b.

RESULTS

TMD Oligomerization Strength Varies between the Neuraminidase Subtypes in IAVs

The TMD from influenza neuraminidase exists within its N-terminal signal anchor sequence, and for N1 from the common laboratory strain influenza A/WSN/33, it is predicted to span residues 7–34 (Fig. 1A) (28). To determine the minimal length necessary for the N1 TMD to target to the endoplasmic reticulum and assemble, the cell surface localization of a series of C-terminal truncations was analyzed after expression in mammalian cells by confocal microscopy (Fig. 1B). For visualization, a small myc epitope was attached to the C termini of the truncations, which encoded residues 1–74, 1–49, 1–40, and 1–35 of N1 from WSN33. By 24 h after transfection, all of the constructs including N1_1–35 were observed to accumulate on the cell surface. This indicated that the 35 N-terminal residues of N1, which is mainly comprised of the TMD, can target to the endoplasmic reticulum and correctly traffic to the plasma membrane.

FIGURE 1.

TMD oligomerization strength varies between the neuraminidase subtypes in IAVs. A, schematic representation of the neuraminidase TMD with respect to the structural data for the enzymatic head domain (Protein Data Bank ID code 3BEQ; A/WSN/33 numbering). B, confocal microscopy sections of fixed Vero cells that were transfected with the indicated hN1 (A/WSN/33) C-terminal truncations and immunostained to demonstrate the plasma membrane localization. C, neuraminidase consensus TMDs from all of the human (hN1 and hN2) and avian (aN1–aN9) subtypes analyzed in the GALLEX system to measure the strength of their interaction. The interaction strength is determined by the TMD-mediated dimerization of LexA that results in the repression of LacZ transcription, which determines the β-galactosidase activity. D, relative interaction (R.I.) strengths of the consensus neuraminidase TMDs from IAVs (hN1 and hN2 and aN1–aN9 (Table 1)) after induction for 2.5 h with 0.05 mm IPTG in the GALLEX system. β-Galactosidase activity (X) from each sample (X_S) was normalized to that of cells containing an empty vector (X_V) where X_Norm = X_S/X_V and the relative interaction strength was calculated by the formula R.I. = 100%(1 − X_Norm). The immunoblot shows the similarity in the expression levels for all the neuraminidase TMDs, the endogenous MBP, and the dashed line is where lanes from the same blot were spliced together. The inset shows the R.I. with respect to an IPTG titration (0.0001 to 1 mm in logarithmic scale), demonstrating the interaction span where the assay is sensitive ∼20–80%. Error bars represent S.D. from three or more experiments.

Based on the ability of N1_1–35 to reach the plasma membrane and that all influenza neuraminidase TMDs are predicted to be within the first 35 amino acids, consensus sequences for amino acids 2–35 (excluding the initiation Met) were created for each avian (a) and human (h) neuraminidase subtype (N1–N9) from the IAV sequence database (Table 1). For each subtype consensus sequence, the predicted TMD is shown as well as the juxtamembrane sequence, the ΔG_app for membrane insertion, and its prevalence in the unique neuraminidase protein sequences for the indicated subtype.

To measure the TMD interactions quantitatively in a biologically relevant context, we chose to use the GALLEX system developed by Engelman and colleagues, which has been used to characterize other eukaryotic plasma membrane TMDs (34, 36–39). In this system, the TMD interactions are measured within the inner membrane of E. coli by the suppression of β-galactosidase and hence its activity (Fig. 1C). The consensus TMDs were introduced into the GALLEX system with their 5 identical N-terminal juxtamembrane residues (amino acids 2–6) to include as much contextual information as possible (Fig. 1C and Table 1).

Accurate measurements in the GALLEX system require that they are made within its linear range, which is ∼20–80% relative interaction strength (Fig. 1D, inset). This is dependent on the inherent interaction strength of the TMD and the expression level of the TMD that can be adjusted by the concentration of IPTG used for induction. Based on these factors, an IPTG concentration (0.05 mm) that enabled the majority of the TMDs to fall into the linear range was chosen. Upon analysis, the aN1–aN9, hN1, and hN2 TMD constructs were all synthesized to similar levels, and significant subtype-dependent variation was observed in the TMD interactions (Fig. 1D). Regarding the human subtypes, this included very weak interactions for the hN2 TMD compared with reasonably strong interactions for the hN1 TMD.

The TMD Interactions for Neuraminidase from IAVs Are Linked to Their Hydrophobicity

TMD interactions can be driven by either polar or nonpolar residues. To try to discriminate between these two possibilities, the amino acid frequency in all of the unique TMDs for each subtype were evaluated with respect to the nonredundant TMDs from the MPtopo database (Fig. 2A and Ref. 40). Comparably, the neuraminidase TMDs contain approximately a 2-fold higher frequency of Cys, the hydrophobic amino acid Ile, and the polar residues Asn, Gln, Ser, and Thr, but were absent of Pro, Tyr, and charged amino acids.

FIGURE 2.

The interactions of the neuraminidase TMDs from IAVs are inversely proportional to their hydrophobicity. A, amino acid frequencies in the unique neuraminidase TMD sequences from the IAV database. Only residues enriched in the neuraminidase TMDs compared with TMDs from the MPtopo database are shown. B, relative interaction strength of each neuraminidase TMD plotted with respect to its hydrophobicity determined by the ΔG predictor. The gray areas highlight where the values are outside the linear range of the assay. The relative TMD interaction strength of aN3, aN6, and hN1_pH1N1, which are outside the linear range with 0.05 mm IPTG, are shown with a lower IPTG concentration (0.02 mm) as an inset. Error bars represent S.D. from three or more experiments, or from all unique TMDs of each set.

Due to the abundance of both Ile and polar residues, a second analysis was performed where the relative interaction strength of each TMD was plotted with respect to its hydrophobicity (ΔG_app for membrane insertion) (Fig. 2B). Surprisingly, the TMD interactions increased in a linear manner with decreasing hydrophobicity giving a reasonable straight line with an R² value of 0.86. The three potential outlier TMDs that had relative interaction strengths above the linear threshold (∼80%) were analyzed separately at lower IPTG levels (0.02 mm) to show that the correlation was still valid (Fig. 2B, inset). These results indicate that the TMD interaction strength across the different neuraminidase subtypes in IAVs is driven by their polar characteristics, making them inversely proportional to their hydrophobicity.

TMD Interactions within the Human N1 Subtype Also Correlate with Their Polar Characteristics

In contrast to the avian IAV sequences, the number of available human N1 and N2 IAV sequences is quite more expansive. To exploit this large data set, we examined the hydrophobicity variation in all of the unique hN1 and hN2 TMDs and discovered substantial variation in the hN1 TMD data set, but not the hN2 TMD data set. Upon the temporal analysis of the unique hN1 TMDs an obvious decrease in hydrophobicity was observed with respect to when the IAV strain was sequenced from humans (Fig. 3A). Thus, we reasoned to divide the hN1 and hN2 TMDs based on the different IAV lineages that have been sequenced from the human population and their associated hemagglutinin (H) subtype (Fig. 3B). The hN1 TMDs were grouped as either H1N1 (1918–1957), H1N1 (1977–2011), and the pandemic pH1N1 of swine origin (2009–present), whereas N2 was grouped as H2N2 (1957–1968) and H3N2 (1968-present) (23, 41).

FIGURE 3.

The TMD interactions are linked to polar characteristics within the hN1 subtype. A, hydrophobicities of all of the subtype 1 neuraminidase TMDs from human H1N1 IAVs are plotted with respect to their date of isolation. B, time line showing the periods of the indicated IAV lineages in the human population. For each strain, the subtype for the surface antigens hemagglutinin (H) and neuraminidase (N) are listed. C, scatter plots displaying the hydrophobicity for each unique neuraminidase (hN1 and hN2) TMDs and their respective M2 and HA TMDs, from the indicated IAV strains. The black bar represents the average for each set, and the p value for each stepwise decrease in the neuraminidase (hN1) TMD hydrophobicities with respect to the subsequent time period is indicated by stars. D, relative interaction strengths for the indicated hN1 neuraminidase TMDs shown with their statistical significance compared with the hN1_≤1957 TMD. (These points were plotted with respect to their hydrophobicity in Fig. 2B). The immunoblot shows the similar expression levels for the different hN1 TMD constructs, the endogenous MBP provides a loading control, and the dashed line is where lanes from the same blot were spliced together. Error bars represent S.D. from three or more experiments, or from all unique TMDs of each set. *, p ≤ 0.05; **, p ≤ 0.01; ****, p ≤ 0.0001.

After this division, a clear difference was observed between the hydrophobicity of the TMDs from hN1 and hN2, but more striking was the trend toward decreasing hydrophobicity in the hN1 TMDs of the different IAV lineages (Fig. 3C). This was specific to hN1, as the hydrophobicites of the predicted 26 amino acid HA TMDs and the 23 amino acid M2 TMDs from these IAVs have remained hydrophobic. The predicted lengths of all the TMDs agree with those for eukaryotic secretory proteins (42), and changes in the predicted TMD lengths were not observed with respect to their hydrophobicities.

Based on these findings, consensus sequences for the TMDs from each hN1 lineage were created. The analysis of the three different hN1 consensus TMDs revealed that the strength of their interaction also increased with decreasing hydrophobicity. When these results were superimposed on the original analysis of the relative TMD interactions for the different subtypes with respect to hydrophobicity, they were found to follow the same linear trend (Fig. 2B). Together, these results demonstrate that the relationship between the polar characteristics and the interaction strength of the neuraminidase TMDs from IAVs exists across and within subtypes.

Amphipathicity Exists within the TMDs of Neuraminidase from IAVs

The observation that the TMD interaction increased with decreasing hydrophobicity indicated that the polar residues were likely responsible for the interaction. To investigate whether the polar residues localized to potential interfaces, each TMD was treated as an α-helical segment with seven faces (a, b, c, d, e, f, and g), and the positional hydrophobicities for each amino acid in the respective face were summed (Fig. 4A). This analysis revealed a distinct pattern of two adjacent hydrophilic faces (a and e) in every neuraminidase TMD from IAVs with the exception of the N2 TMDs, which possess a single polar face (e) and interact very weakly. These results suggested that residues within the two polar faces of these amphipathic TMDs are responsible for their interactions.

FIGURE 4.

Polar residues in unfavorable membrane positions drive the amphipathic TMD interactions. A, the unique neuraminidase TMDs from IAVs were treated as an α-helix with seven faces (a–g), and the hydrophobicity for each face was calculated by summing their positional ΔG_app^aa(i) values (31, 32). The wheel illustrates how the faces a to g are positioned relative to each other. B, removal of the polar residues in the a- and e-face that are the least favorable with respect to membrane integration can synergistically, or additively, decrease the neuraminidase TMD interactions. Each of the indicated face mutants within the hN1_pH1N1, aN3, and hN2 TMDs (see Table 1) were directly compared with their consensus TMD sequence. Error bars represent S.D. from three or more experiments, or from all the unique TMDs of each set. The immunoblots display similar expression levels for each TMD, as well as the endogenous MBP. **, p ≤ 0.01; ***, p ≤ 0.001; ****, p ≤ 0.0001.

Polar Residues in Unfavorable Membrane Positions Drive the TMD Interactions

The least favorable residues in the neuraminidase TMDs from the different IAVs are the polar residues Asn, Gln, Ser, and Thr, which all affect membrane integration in a position-dependent manner with respect to the membrane (31, 32). Of these residues, Asn, Gln, and Ser are preferentially localized to the polar faces (a and e) of the amphipathic TMDs. This suggested that these residues could make interactions that drive the TMD assembly and by doing so decrease their unfavorable localization in the membrane. Thus, different combinations of Ala mutations were made in the Asn, Gln, and Ser residues positioned in the a and e faces of the two strongly interacting TMDs, hN1_pH1N1 and aN3. From the analyzed constructs, the hN1_pH1N1 TMD interaction was almost entirely abolished in a Δae mutant, where the two Asn residues on the a-face and the single Gln on the e-face were changed to Ala (Fig. 4B and Table 1). Independently, removal of both Asn residues in the Δa mutant moderately decreased the TMD interaction, whereas removing the Gln residue in the Δe mutant had no effect.

In a similar Δae mutant of the aN3 TMD, its interaction was also significantly lost (Fig. 4B and Table 1). However, individual mutations of each Asn residue (Δa₁ and Δa₂) and the Ser residue (Δe) all decreased the aN3 TMD interaction. The Δa mutant, where both Asn residues were removed, interacted as weakly as the combined Δae mutant. Substituting the Gln residue in the sole hydrophilic face of the hN2 TMD (Δe) did not alter its weak TMD interaction. These results indicated that the polar interactions can function either synergistically (hN1_pH1N1) or additively (aN3) and that the key polar residues can be identified by where they are located on the TMD and how favorable their position is within the membrane.

Enhancing the hN1 TMD Interaction Required Small Neighboring Side Chains around the Larger Polar Residue

The correlation between the polar residue positioning and TMD interactions of these natural TMDs implied that the proper placement of polar residues alone could increase the TMD interaction strength. To investigate this possibility, we identified the residues in hN1_pH1N1 TMD which were responsible for its stronger interaction compared with the closely related hN1_H1N1≤1957 TMD (Fig. 5A). Of the six changes found in the hN1_pH1N1 TMD, the three mutations (Met-Ala-Asn) in the TMD core enhanced the hN1_H1N1≤1957 TMD interaction to levels equivalent to the hN1_pH1N1 TMD (Fig. 5B residues 13–15 MAN). Alone, the Asn-15 substitution showed no effect even though this causes a large hydrophobicity decrease in the TMD and is positioned on the a-face (Fig. 5B and Table 1). However, when the Asn-15 mutation included the neighboring Met-13 or Ala-14 residues, a moderate increase in the TMD interaction was observed. This suggests that the proper packing of the large polar Asn side chain benefits from slightly polar neighboring side chains that are smaller and less branched. Together, these results demonstrate that the interaction of the neuraminidase TMDs from IAVs is dictated by their polar residue positioning with respect to the TMD and the membrane bilayer and imply that the observed correlation occurs because the local changes that optimize each amphipathic TMD assembly exist within the IAVs in nature.

FIGURE 5.

Contributions of polar residues to the hN1 TMD interactions require proper packing. A, comparison of the consensus hN1_≥1957 and hN1_pH1N1 TMD sequences that made moderate and strong interactions. The amino acids that differ in the hN1_pH1N1 TMD are underlined, and residues 13–15 that were analyzed in this study are shown in red. B, the relative interaction strength for the hN1_≥1957 TMD (IIS) and the constructs with the indicated mutations within residues 13–15 of this TMD (shown in red) compared with the hN1_pH1N1 TMD. The p values (*, p ≤ 0.05; **, p ≤ 0.01) are shown with respect to the value for the hN1_≥1957 TMD (IIS). The immunoblot shows a similar expression profile for all of the constructs and the endogenous MBP. C, schematic depiction of how the TMDs from N1 and N3–N9 could assemble from a dimer to a tetramer based on the localization of the polar a- and e-faces and their expected behavior within the membrane. N2 lacks one polar face and interacts weakly. D, model demonstrating how the neuraminidase TMD interactions are dependent on which polar residues are present, their localization within the TMD α-helix, and their positioning with respect to the membrane as their H-bond contribution appears to be inversely related to their hydrophobicity or contribution to membrane integration.

DISCUSSION

The large number of membrane proteins places enormous pressure on their TMDs to achieve lateral interaction specificity in limited sequence space. Many studies on how this can occur have identified several properties, residues, and motifs predominantly using synthetic TMDs and structures of multispanning TMD proteins. We investigated whether these known features can be applied to identify and predict the TMD interactions of a natural bitopic protein by exploiting the large number of neuraminidase sequences from the IAV database. The interaction strength of each consensus TMD was analyzed with respect to its amino acid composition and residue positioning within the helix and in the membrane bilayer. By combining these three principles, the TMD interactions were determined to be polar-mediated, and the key interacting residues could be identified, which presumably make hydrogen bonds in the membrane. Our results demonstrate that polar TMD interactions depend on the particular residue involved, their location on the helix, and the favorability of their position in the membrane bilayer. This likely explains why the strength of the neuraminidase TMD interactions correlated with their ΔG_app for membrane insertion (Fig. 5D).

TMD assembly can be broken down into membrane partitioning, α-helix formation, and integration followed by the interhelical associations that contribute to the final structure and function of the protein (3–5). Utilizing this as a premise, we proposed a simple model for the neuraminidase TMD association (Fig. 5C). The model is based on the conserved amphipathicity within the TMDs, with the exception of N2, the positioning of interacting residues on adjacent polar faces, and the previous observations that the hN1 TMD tetramerizes and likely contributes to its rapid co-translational oligomerization (28, 30, 43). Upon integration, the hydrophobic lipid environment drives the polar faces of the amphipathic TMD together resulting in their rapid dimerization because the unfavorable polar residues prefer helix-helix H-bonds over any potential helix-lipid interactions. This would explain the inverse correlation between the polar residue contributions to the TMD interactions with their positional effects on membrane integration (Fig. 5D). One can then envision that when two dimers come into contact, the TMDs make a subtle shift that opens up the two hydrophilic faces and creates a polar core resulting in tetramerization. The final structure would be similar to that of synthetic TMDs with two polar Ser faces (44), but additional biochemical and structural data of the neuraminidase TMDs are needed to confirm this idea.

The presence of motifs and characteristics known to be involved in TMD interactions does not necessarily mean that they actually facilitate an interaction due to the unpredictable influence of the surrounding sequence context (15–17). In terms of the neuraminidase TMDs from IAVs, the initial TMD sequence analysis gave no indication of a conserved interaction motif as GXXXG, QXXS, periodic Ile/Leu/Val motifs, and the polar residues Asn, Gln, and Ser were all found. Thus, the consensus TMDs and the different mutants were experimentally analyzed using the GALLEX system. This system was chosen based on its success with other plasma membrane TMDs and biological relevance in terms of membrane protein concentration and lipid composition, with the exception that eukaryote-specific lipids are absent (34, 36–39).

The consensus TMDs that were examined are identical to those found in more than one-third of the unique neuraminidase protein sequences in the IAV database. To retain as much sequence context as possible, they also included the 5 identical N-terminal juxtamembrane residues, but excluded the variable C-terminal residues. Initially, the TMDs were analyzed independently using single-residue Trp and Ala mutations with little success at disrupting the interactions. However, this indicated that synergistic and/or multiple interactions were likely involved. Thus, the different TMD interactions were compared based on the premise that they likely interact and assemble in a similar manner that could be extracted from the sequence characteristics even though the TMDs range in identity from ∼20 to 90%. The comparison revealed a clear trend between the interaction strength and decreasing hydrophobicity and ultimately provided the sequence context required to identify the key interactions as the polar residues partitioned to two adjacent faces in all but the N2 TMDs.

All of the polar residues in the neuraminidase TMDs from IAVs (Asn, Ser, Gln, and Thr) have been shown to be capable of mediating interactions of other TMDs, but with great variation (44–48). As expected, the ability of these residues to drive interactions is related to their localization within the TMD. To identify the key interacting residues, we combined these observations with the idea that the H-bonding contribution would depend on how favorable the polar residue is positioned with respect to the membrane bilayer. For instance, Asn-14, Asn-28, and Ser-18, which are unfavorably positioned toward the center of the aN3 TMD, contributed to the interaction; whereas Asn-32, favorably positioned toward the phospholipid head groups, did not (Fig. 4B and data not shown). This is similar to synthetic TMDs where an Asn residue placed in the center of the TMD made a stronger contribution than one placed toward the edge of the bilayer (49). These findings suggest that the contribution of a particular polar residue to a TMD interaction should correlate with its negative contribution to membrane integration, making their identification easier.

Currently, the ability to predict TMD interactions is quite limited and based on a small set of sequence motifs. So why can the interaction strength between the neuraminidase TMDs be estimated by their hydrophobicity? One likely factor is that the TMD of influenza neuraminidase is located within its signal anchor sequence, and it performs a variety of functions from targeting to the endoplasmic reticulum to assembly, apical trafficking, and viral incorporation (27–29). These functional requirements probably constrain the neuraminidase TMD sequences in IAVs resulting in conserved structures, which make it easier to identify what dictates their lateral interactions.

In terms of specificity and packing, polar residues alone did not enhance the N1 TMD interaction without additional local changes. Presumably, this is because the more polar residues have larger side chains requiring the neighboring side chains to be smaller (Ala), unbranched (Met), and less hydrophobic (Ala and Met) to accommodate their packing. Based on this, it can be speculated that the neuraminidase sequences found in nature have undergone selection for their ability to interact. In support of this concept, the interaction strength could not be predicted for the mutants in this study. For instance, the Gln to Ala (e-face) mutation in the hN1_pH1N1 TMD remarkably increased its hydrophobicity (∼1.2 kcal/mol) without affecting the interaction strength. Similarly, the Ser to Asn (a-face) mutation in the hN1_H1N1>1957 TMD significantly decreased the hydrophobicity (∼0.7 kcal/mol) also without affecting the interaction strength. This packing requirement ultimately generates specificity within the TMD interactions in addition to the use of different key residues and varying their positions.

One surprising discovery during this investigation was the difference in hydrophobicity and interaction strength between the hN1 and hN2 TMDs. To date, the N1 TMD in the pandemic (p) H1N1 virus of swine origin is by far the least hydrophobic of all IAVs that have been sequenced and show the strongest interactions. More interestingly, this TMD should not integrate into the membrane very well, if at all, based on predictions and that neuraminidase only has one TMD (31). Potentially, sequestering the polar faces functions to overcome their negative impact on membrane insertion, suggesting that TMD localization bias also contributes to insertion.

Polar residue substitutions in membrane proteins are the most common disease-causing mutations (50, 51). Although it is debatable about how much these residues actually contribute to TMD interactions, this study demonstrates in natural TMDs that the contribution is significant and highly dependent on their position in the membrane in addition to where they are localized on the helix. This feature would be expected to add another degree of specificity to how proper lateral TMD interactions are achieved. Together, analyzing the TMD amino acid properties with respect to their helix localization and favorability toward membrane integration should increase our ability to identify polar-mediated interactions in the large number of uncharacterized TMDs from bitopic proteins.