Detailed Mechanistic Insights into HIV-1 Sensitivity to Three Generations of Fusion Inhibitors*

Peptides based on the second heptad repeat (HR2) of viral class I fusion proteins are effective inhibitors of virus entry. One such fusion inhibitor has been approved for treatment of human immunodeficiency virus-1 (T20, enfuvirtide). Resistance to T20 usually maps to the peptide binding site in HR1. To better understand fusion inhibitor potency and resistance, we combined virological, computational, and biophysical experiments with comprehensive mutational analyses and tested resistance to T20 and second and third generation inhibitors (T1249 and T2635). We found that most amino acid substitutions caused resistance to the first generation peptide T20. Only charged amino acids caused resistance to T1249, and none caused resistance to T2635. Depending on the drug, we can distinguish four mechanisms of drug resistance: reduced contact, steric obstruction, electrostatic repulsion, and electrostatic attraction. Implications for the design of novel antiviral peptide inhibitors are discussed.

The functional trimeric Env spike on HIV-1 virions consists of three gp120 and three gp41 molecules that are the products of cleavage of the precursor gp160 by cellular proteases such as furin (4,5). The gp120 surface subunits are responsible for binding to the cellular receptors, whereas the gp41 subunits anchor the complex in the viral membrane and mediate the fusion of viral and cellular membranes. Env undergoes several conformational changes that culminate in membrane fusion. The gp120 subunit binds the CD4 receptor, resulting in creation and/or exposure of the binding site for a coreceptor, usually CCR5 or CXCR4 (6,7). Two ␣-helical leucine zipper-like motifs, heptad repeat 1 (HR1) and heptad repeat 2 (HR2), located in the extracellular part of gp41, play a major role in the following conformational changes. Binding of the receptors to gp120 induces formation of the pre-hairpin intermediate of gp41 in which HR1 is exposed and the N-terminal fusion peptide is inserted into the target cell membrane (1, 8 -12). Subsequently, three HR1 and three HR2 domains assemble into a highly stable six-helix bundle structure that juxtaposes the viral and cellular membranes for the membrane merger. Other viruses with class I viral fusion proteins use similar HR1-HR2mediated membrane fusion for target cell entry.
Peptides based on the HR domains of class I viral fusion proteins have proven to be efficient inhibitors of virus entry for a broad range of viruses (13)(14)(15)(16)(17). The HIV-1 fusion inhibitor T20 (enfuvirtide (Fuzeon)) has been approved for clinical use. T20 mimics HR2 and can bind to HR1, thereby preventing the formation of the six-helix bundle ( Fig. 1) (18 -21). T1249 is a second-generation fusion inhibitor with improved antiviral potency compared with the first-generation peptide T20 (22)(23)(24)(25). Recently, a series of more potent third-generation fusion inhibitors were designed (26,27). These include T2635, which has an improved helical structure that increases stability and activity against both wild type (WT) HIV-1 and fusion inhibitor resistant variants.
Both the in vitro and in vivo selection of resistance has been described for T20 (28 -33) and T1249 (23, 34 -36). Resistance is often caused by mutations in the HR1 binding site of the fusion inhibitor. In particular, substitutions at positions 36 (G36D/M/ S), 38 (V38A/W/M/E), and 43 (N43D/K) of gp41 can cause resistance. Strikingly, substitutions at position 38 can cause resistance to both T20 and T1249, but distinct amino acid substitutions are required. At position 38 only charged amino acids (V38E/R/K) cause resistance to T1249 (35). Surprisingly, none of the known T20 and T1249 resistance mutations at position 38 affect the susceptibility to the third generation inhibitor T2635.
We hypothesized that the use of HIV-1 as a model system could provide a more detailed understanding of resistance to fusion inhibitors. We, therefore, analyzed the effect of all 20 amino acids at resistance hotspot 38 on Env function, viral fitness, biochemical properties of gp41, and resistance to the fusion inhibitors. From the results we can propose four resistance mechanisms that differ in the way the drug-target interaction is affected at the molecular level. Furthermore, we can deduce general principles on the mechanisms of resistance against fusion inhibitors and the requirements for effective antiviral drugs.

EXPERIMENTAL PROCEDURES
Peptide Synthesis-Peptides T20, T1249, and T2635 were synthesized as described previously (35). A more detailed description of the procedure is given in the supplemental material.
Construction of HIV-1 LAI Molecular Clones-The full-length molecular clone of HIV-1 LAI (pLAI) (37) was used to produce WT and mutant viruses. The plasmid pRS1 was used to introduce mutations at position 38 as described previously (35,38), and the entire env gene was verified by DNA sequencing. Mutant env genes in pRS1 were cloned back into pLAI as SalI-BamHI fragments. Each virus variant was transiently transfected in C33A cells by calcium phosphate precipitation as previously described (39). The virus-containing supernatant was harvested 3 days post-transfection, filtered, and stored at Ϫ80°C, and the virus concentration was quantitated by capsid CA-p24 enzyme-linked immunosorbent assay as described previously (40).
IC 50 and Infectivity Determination-The TZM-bl reporter cell line (32,41) stably expresses high levels of CD4 and HIV-1 co-receptors CCR5 and CXCR4 and contains the luciferase and ␤-galactosidase genes under the control of the HIV-1 long-terminal-repeat promoter. The TZM-bl cell line was obtained through the NIH AIDS Research and Reference Reagent Pro-gram, Division of AIDS, NIAID, National Institutes of Health (John C. Kappes, Xiaoyun Wu, and Tranzyme Inc. (Durham, NC)). Single-cycle infection experiments and inhibition experiments using TZM-bl cells were performed as described (35). A more detailed description of the procedure is given in the supplemental material.
We note that some differences in relative infectivity were found compared with our previous studies (35). Similar small inconsistencies compared with our previous studies were observed with the biophysical experiments. We can attribute these differences to the different Env background in which the substitutions were introduced. We used the HIV-1 LAI sequence here, whereas a N125S mutation was present in the earlier studies to mimic the HR2 sequence from the patient isolate we studied (33,35). The exact consequences of this substitution for virus infectivity and gp41 function in general are currently under investigation.
Homology Modeling-The structures of the HIV-1, strain LAI, HR1 trimer in complex with an inhibitory peptide was based on homology modeling using the x-ray structure of SIV gp41 six-helix bundle (PDB 2EZO and 1QCE), which is the most complete example of six-helix bundles in the data base, and it contains the largest part of the interaction site of the HR1 trimer with the HR2-based inhibitory peptides. Part of the HR2 helix was used as a model for T20, T1249, and T2635. In case of T2635, the helix corresponding to HR2 was extended with seven residues. Model building was performed using the automated protein-modeling software on the SWISSMODEL protein-modeling server (42,43). Model quality was verified using WhatCheck (44).
Binding Energy Calculations-To get an estimate of the binding energy of the various mutants, the models of the various inhibitory peptides and their mutants were subjected to a gentle refinement in explicit solvent (8 Å water shell) using HAD-DOCK2.0 (45,46). For this, randomization of starting orientations and rigid-body energy minimization were turned off, and only the final water refinement stage was performed consisting of a heating phase (100 molecular dynamics (MD) steps at 100, 200, and 300 K), a sampling phase (1250 MD steps at 300 K), and a final cooling phase (500 MD steps at 300, 200, and 100 K). Non-bonded interactions were calculated with the optimized potentials for liquid stimulations force field (47) using a cutoff of 8.5 Å. The electrostatic potential (E elec ) was calculated by using a shift function, whereas a switching function (between 6.5 and 8.5 Å) was used to define the van der Waals potential (E vdw ). The binding energy was approximated by the average HADDOCK score of the best four models after water refinement. The HADDOCK score consists of a weighted sum of intermolecular non-bonded energies and an empirical desolvation term (48): HADDOCK score ϭ 1.0 ϫ E vdw ϩ 0.2 ϫ E elec ϩ 1.0 ϫ E desolv . We should remark that the binding energies we are referring to are estimates based only on the intermolecular electrostatic and van der Waals energies with an additional empirical desolvation term. They clearly should not be interpreted in terms of free energies and do not account for any entropic effects (except for the desolvation term that implicitly accounts for the solvent entropy gain upon binding). From all the calculated terms, the HADDOCK score was the one giving the best correlation with the experimental data (data not shown).
Peptide Expression and Purification-The recombinant N36(L6)C34 model peptide and its variants were expressed in the Escherichia coli strain BL21(DE3)/pLysS by a modified pET3a vector (Novagen, San Diego, CA) and purified as described (35,49,50). The sequence of N36(L6)C34 is SGIV-QQQNNLLRAIEAQQHLLQLTVWGIKQLQARILSGGRG-GWMEWDREINNYTSLIHSLIEESQNQQEKNEQELL (with the six-residue linker underlined). Substitutions were introduced into the pN36/34 plasmid by the method of Kunkel (51) and were verified by DNA sequencing. Peptide identities were confirmed by electrospray mass spectrometry (Voyager Elite; PerSeptive Biosystems, Framingham, MA). Protein concentrations were determined by the method of Edelhoch (52).
Biophysical Analysis-Circular dichroism (CD) experiments were performed on an 62A/DS (Aviv Associates, Lakewood, NJ) spectropolarimeter equipped with a thermoelectric temperature control at 10 M peptide concentration in phosphate-buffered saline as described (35,49,50). An ellipticity value at 222 nm ([] 222 ) value of Ϫ33,000 degrees cm 2 dmol Ϫ1 was taken to correspond to 100% helix (53). Thermal stability was determined by monitoring [] 222 as a function of temperature as described. All melts were reversible. Temperatures of midpoint unfolding transitions (T m ) were estimated by evaluating the maximum of the first derivative of [] 222 in relation to the temperature data (54). Equilibrium ultracentrifugation measurements were carried out on an XL-A analytical ultracentrifuge equipped with an An-60 Ti rotor (Beckman Coulter, Fullerton, CA) at 20°C as described previously (55) (35,49,50). Solvent density and protein partial specific volume were calculated according to solvent and protein composition, respectively (56). The apparent molecular masses of all N36(L6)C34 variants were within 10% of that calculated for an ideal trimer, with no systematic deviation of the residuals.

RESULTS
Virus Infectivity-We generated a set of 20 viruses based on the CXCR4-using HIV-1 LAI strain, each with a different amino acid at position 38 in gp41. To compare the infectivity of these viruses, we performed single cycle infection assays. Luciferase activity in TZM-bl reporter cells was measured 2 days postinfection (Fig. 2). Mutant viruses V38H/T/W revealed infectiv-ity close to WT, but all other viruses showed decreased infectivity compared with WT, and some substitutions severely reduced infectivity (V38K/P/R). The lower infectivity of these viruses may be explained by a reduced efficiency of six-helix bundle formation because of an altered HR1-HR2 interaction (see below). In addition, substitutions at position 38 may have effects on protein folding or on subsequent conformational changes. The relatively dramatic impact of changes at position 38 on infectivity may explain the high level of conservation of this residue (Los Alamos data base; www.hiv.lanl.gov) (57).
Sensitivity to Fusion Inhibitors-Next, we investigated the susceptibility of our panel of viruses to the three generations of peptide inhibitors, T20 (first generation), T1249 (second generation), and T2635 (third generation), and determined their respective 50% inhibitory concentrations (IC 50 ) ( Table 1, supplemental Fig. 1). These three inhibitors are based on the HR2 sequence or are derivatives thereof, and position 38 is part of the predicted binding site of all three inhibitors (see Fig. 1 and Fig. 5, A-C). Consistent with this, almost every substitution at position 38 caused high resistance to T20, except for the V38I and V38L substitutions. All other substitutions caused resistance to T20 ranging from 7.4-fold (V38M) to Ͼ200-fold (V38D/E/P). Resistance to T1249 was only observed with five amino acids (V38D/E/K/P/R; Ͼ4-fold resistance compared with WT), whereas other substitutions did not alter sensitivity to T1249 significantly. Interestingly, all charged amino acids caused resistance to T1249, the negatively charged amino acids (V38D/E) causing the highest resistance (28.8-and 34.1-fold). Despite the location within the predicted binding site for T2635, none of the substitutions at position 38 conferred significant resistance to the third generation peptide. It may be surprising that most variants are resistant to T20, because only a limited number of these (V38A/M/W) are frequently observed in different in vitro and in vivo settings. However, all previous studies are based on virus evolution, either in cell culture or in patients, which means that the likelihood of the underlying mutation favors certain escape routes. In an evolutionary setting, the virus will try to find a balance between resistance, viral fitness, and the ease of emergence of the codon change required for escape.
Biophysical Properties-To obtain mechanistic insight into the role of HR1 position 38 in fusion and resistance to fusion inhibitors, we studied the biophysical properties of the six-helix bundle. Analyzing the effects of substitutions at position 38 may provide explanations for differences in infectivity as sixhelix bundle formation is an essential step during fusion. Furthermore, it can teach us about the resistance to fusion inhibitors as the interaction of these drugs with HR1 is comparable with the interaction of HR2 with HR1.
We introduced all 20 amino acids at position 38 in the recombinant HIV-1 LAI N34(L6)C28 peptide, which represents the six-helix bundle (49,58), and compared their biophysical properties. Sedimentation equilibrium analysis showed that the molecular weights of the 20 six-helix bundle variants were all close to those calculated for the ideal trimer, indicating that all variants formed trimers ( Table 2). CD was used to determine the extent of ␣-helical structure of the six-helix bundle variants. The WT peptide is ϳ100% ␣-helical, as indicated by the typical wavelength dependence pattern shown in Fig. 3A. Most variants showed a similar pattern, indicating an ␣-helical content of Ͼ90% ( Fig. 3A; Table 2). A number of amino acid substitutions decreased the ␣-helical content to Ͻ90% (V38E/K/N/P/R), whereas the V38D substitution reduced the ␣-helicity most dramatically (to 64%). The reduced helix content of the V38P six-helix bundle can be easily explained by the helix-breaking properties of this amino acid, and it is likely that the primary effect of the substitution is on the helix content of HR1. This effect may explain why V38P causes resistance to T20 and T1249. The reduced ␣-helicity of other substitutions is less easily explained. For example, Glu is extremely well accommodated in ␣-helices (59 -61), but the V38E six-helix bundle nev-ertheless showed reduced ␣-helicity. A plausible explanation is that the V38E substitution does not directly affect the ␣-helix content of HR1 but indirectly affects that of HR2, possibly interfering with proper packing of HR2 with HR1. Similar effects may occur for V38K/N/P/R. These findings are also relevant for the drug-HR1 interactions and have implications for the observed resistance (see below).
The thermal stability of the variant six-helix bundles was determined by monitoring the CD signal at 222 nm over a temperature range. The midpoints of thermal denaturation (T m ) are given in Table 2, and selected denaturation curves are shown in Fig. 3B. Under these conditions, the T m of the WT peptide was 80°C, whereas the T m of the mutants ranged from 58 to 80°C. Almost all substitutions destabilized the six-helix bundle to some extent, with the V38A/D/E/G/P/S substitutions causing the largest effects (Ͼ10°C drop in T m ). Not surprisingly, substitutions that reduced ␣-helicity often caused a decrease in thermal stability (e.g. for V38D/E/N/P, less evident for V38K/R). However, the small amino acids Ala/Gly/Ser caused a profound reduction of thermal stability (T m 65-69) but without a noticeable effect on the helix content (95-100%). Taken together these results indicate that substitutions at position 38 of gp41 can lead to a substantial reduction of ␣-helical content and destabilization of the six-helix bundle.
Energetics of Drug-HR1 Interactions-Resistance to HIV-1 inhibitors is often caused by mutations that lower the affinity of the target for the drug. Many common resistance mutations to HIV-1 reverse transcriptase and protease inhibitors exemplify this type of resistance (62)(63)(64), and a similar mechanism of resistance has been proposed for T20 resistance mutations at position 38 (65). An important contributor to drug affinity is the energy that is released upon binding of the drug (binding energy). The binding energy is composed of energy loss, caused among other things by the decrease in entropy, and energy gain, caused by the acquisition of drug-target contacts. We studied the binding energies of the 20 HR1 variants with the three gen-  erations of peptide inhibitors in silico. Models of the T20-HR1, T1249-HR1, and T2635-HR1 complexes were generated based on the core structure of gp41 and the binding energies of the 60 complexes were calculated using HADDOCK2.0. We found a correlation between the calculated binding energy and the thermal stability of the six-helix bundle observed in our biophysical experiments (p ϭ 0.0024; data not shown). A higher calculated binding energy corresponds with a higher thermal stability. This provided confidence in the in silico analyses on the energetics of the drug-HR1 interactions. HR1 in complex with T20, T1249, and T2635 revealed binding energies of 160, 208, and 254 kcal mol Ϫ1 , respectively, consistent with the escalating inhibitory capacities of the three generations of inhibitors (Fig. 4). We next investigated the binding energies of T20 with the 20 HR1 variants. Comparable binding energies were observed for V38C/F/H/I/K/L/M/P/T/W/Y (150 -167 kcal mol Ϫ1 , Fig. 4, supplemental Table S1), whereas V38A/D/E/G/N/Q/R/S caused a substantial decrease (binding energies Ͻ150 kcal mol Ϫ1 ). Interestingly, all aromatic residues showed higher binding energy than WT. Correlation and regression analysis showed a trend between the binding energy and sensitivity to T20 (p ϭ 0.035; data not shown). An increased resistance to T20 correlated with a lower binding energy of T20 to HR1. However, the aromatic residues did not fit this trend.
We were intrigued by the anomalous binding energies of the aromatic amino acids. One would expect that the bulky aromatic rings would disturb the interactions with T20, causing a decrease in binding energy that would explain the observed resistance. Surprisingly, we found an increased binding energy for the aromatic residues Phe and Trp and only a small decrease for Tyr. A potential explanation could lie in the methodology. The calculations of the binding energies were based on T20-HR1 complexes in which the 38 side chains are accommodated optimally within the complex as a result of the optimization process during the modeling. The aromatic rings can be accommodated very well in the T20-HR2 complex after refinement, but it is highly unlikely that when T20 docks onto HR1, the 38 side chain is in this minimized "ideal" conformation. On the contrary, the aromatic side chains may stick out and obstruct proper T20 docking and/or binding, which would explain the observed resistance. We, therefore, constructed an additional model of T20 in complex with the V38F mutant and forced the aromatic side chain in a conformation that protrudes from the HR1 surface (F2; Fig. 4 and supplemental Table S2). The binding energy based on this complex was indeed lower (152.7 versus 166.0 kcal mol Ϫ1 for T20 binding).
The binding energies of the set of HR1 mutants with T1249 and T2635 ranged from 187 to 223 and 232 to 260 kcal mol Ϫ1 , respectively, and this variation was similar to the T20-HR1 complexes (ϳ30 kcal mol Ϫ1 ; Fig. 4 and supplemental Table S1). Furthermore, the trends observed were similar. Thus, the amino acids that caused decreased binding energies for T20 mostly had similar effects on the binding energies for T1249 and T2635. However, despite these similarities, we could not observe a correlation between the T1249-HR1 and T2635-HR1 binding energies and resistance, as we had seen with T20. A likely explanation is the overall higher binding energy of the former two inhibitors and a relatively decreased contribution of residue 38 to the drug-HR1 binding energy. Thus, T1249 may be less dependent and T2635 not dependent on the contribution of residue 38 to the overall binding energy.
Effect of Charge Interactions-Strikingly, charged residues at position 38 caused the highest resistance to T20, and charged residues could confer resistance to T1249 but not to T2635. Further inspection of the peptide-HR1 interface revealed that residue 38 of gp41 interacts with residues Gln-142, Asn-145, Glu-146, and Leu-149 in HR2 or the inhibitor (Fig. 5, A-D). The presence of Glu-146 in HR2 may explain why charged residues at position 38 cause an effect on six-helix bundle helix content and resistance. The electrostatic repulsion by Glu-146 in T20 and T1249 readily explains why V38D/E causes resistance. The resistance of V38K/R is counterintuitive, but it appears that formation of a salt bridge between V38K/R and Glu-146 is incompatible with packing of HR2/T20/T1249 onto HR1 as exemplified by the decreased helical content and lower six-helix bundle stability (Fig. 3, Table 2). Glu-146 is also present in T2635. However, in T2635 Glu-146 is involved in an intramo- lecular salt bridge with Arg-150 as part of the design to stabilize the helical structure of the peptide (27). This resulting neutral net charge may explain why even the charged residues at position 38 do not cause resistance to T2635 (Figs. 5C and 6).
To verify this possibility, we introduced this salt bridge in T1249 and removed it from T2635. Thus, we introduced Q150R in T1249 to pair with Glu-146 (T1249 mut , Figs. 1 and 5D), and we removed the specific salt bridge from T2635 by the reverse R150Q change, resulting in exposure of the charged Glu-146 (T2635 mut ). We then performed single cycle infection experiments in the absence and presence of these variant peptides and determined the IC 50 for WT and mutants containing a charged residue at position 38 (V38D/E/K/R) (Fig. 5, E and F, Table 3). The IC 50 for T1249 mut for WT virus was slightly decreased compared with the parental T1249 (2-fold), consistent with the helix-stabilizing effect of the salt bridge (27). Conversely, the removal of the salt bridge from T2635 caused a minor IC 50 increase. Importantly, the Arg substitution in T1249 mut almost completely abrogated the resistance caused by the charged residues at position 38 (from up to 40-fold to only 4-fold), confirming the hypothesis that electrostatic interactions of Glu-146 with charged residues at position 38 (negative or positive) are responsible for the resistance. We did not observe significant resistance differences between T2635 and T2635 mut . Apparently, the total binding energy of the T2635-HR1 interaction is high enough to accommodate even an electrostatic clash between Glu-146 and V38D/E. There are two alternative explanations why T2635 could be less sensitive to mutations at position 38. First, the binding site of T2635 on the HR1 trimer is different. The T2635 binding site overlaps for a large part with those of T20 and T1249 but is shifted three helical turns. Therefore, position 38 binds close to the C terminus of T2635, and this may not be the energetic core of the binding site. Second, T2635 is a stabilized helix that may not depend on the same type of docking as T20 and T1249; the latter two have to fold onto the hydrophobic groove of the HR1 trimer to acquire their helical structure. T2635 already has a preformed, large contact surface that may have a different docking mechanism that is less easy to escape from. This will be further addressed in the discussion.

DISCUSSION
Mutation of position 38 in HIV-1 gp41 can induce resistance to T20 and T1249, both in vivo and in vitro. Here, we have determined the impact of all 20 amino acids at this gp41 position for the sensitivity against three generations of peptide fusion inhibitors: T20, T1249, and T2635. In addition, we have collected biophysical and computational data on these gp41 variants. Combined, these data provide detailed insight into the underlying mechanisms of fusion inhibitor resistance.

Estimated Binding Energy Versus Docking
We found a weak correlation between increased T20 resistance and decreased binding energy of the T20-HR1 complex. Thus, the binding energy can only in part explain the resistance profiles of the 38 variants. For T1249 and T2635 the resistance cannot readily be explained by the variations in binding energy. It is, therefore, likely that residue 38 does not only contribute to the binding energy but is important for the initial docking of the peptides onto HR1. Indeed, it has been described that the LLS-GIV stretch (residues 32-38 of gp41) is a critical docking site for T20 (66), and this may also explain the critical role of position 38 in resistance development.
If residue 38 is essential for binding/docking of T20, one would expect it to be crucial for the HR2 interaction as well as T20 and HR2 are similar in sequence. Indeed, we found a cor- relation between T20-HR1 binding energy and six-helix bundle stability (p ϭ 0.0042), and we observed dramatic effects on sixhelix bundle formation and infectivity with some substitutions.
However, it should be noted that HR2 may be less dependent on such an initial docking event as it is already tethered to HR1 via the gp41 loop domain. Therefore, resistance mutations may have a more dramatic impact on the T20-HR1 interaction than on the HR2-HR1 interaction, and this is exactly a scenario that viruses would prefer for the selection of drug resistance mutations.
The increase in total binding energy of T2635-HR1 and to a lesser extent T1249-HR1 may explain why these drugs are less dependent on residue 38. The relative contribution of this single residue to the total binding energy is decreased. Another issue that may be taken into account is the level of structure of the drugs in solution. T20 exists predominantly as a random coil (ϳ12% helix content) and needs the docking onto the HR1 coiled coil to acquire an ␣-helical structure, whereas T1249 is ϳ50% helical in solution. T2635 was specifically designed to be more stable (ϳ75% helix content and a T m of 86°C in complex with HR1 compared with 59°C for T1249 and Ͻ5°C for T20) and, therefore, may be less dependent on the initial docking event to obtain its structure (27,67).
Because the binding site of T1249 and T2635 is shifted compared with that of T20, their binding and/or docking may be more dependent on a hydrophobic pocket which is located near the C-terminal end of the HR1 coiled coil (68). The pocket binding domain, which is absent in T20, may contribute to the enhanced binding energy for T1249 and T2635 observed in our in silico analysis. We note that the binding of a lipid binding domain at the C terminus of T20 is absent in our analysis (68 -70). This may constitute an underestimation of the binding energy for T20.

Mechanisms of Resistance
From our results, we can deduce four well defined mechanisms of resistance to fusion inhibitors (excluding the resistance of V38P that probably causes a kink in HR1) (Fig. 6).
Reduced Contact-Small amino acids such as Ala/Gly/Ser at position 38 are smaller than the WT Val, creating a "gap" in the binding site for T20. This reduces the number of contacts between HR1 and T20, resulting in a lower binding energy and resistance (Fig. 6). Obviously, the same should be the case for the HR2-HR1 interaction. Indeed we observed dramatically reduced six-helix bundle stability despite the fact that A/G did not alter six-helix bundle formation itself (as indicated for example by the high ␣-helix content).
Steric Obstruction-The opposite effect is induced by large residues Phe/His/Met/Trp/Tyr that create a "bulge," which sterically hinders the attachment of T20 to HR1. Although these larger residues can be accommodated quite well in the six-helix bundle complex (as indicated by high helix content, considerable six-helix bundle stability, relatively high infectivity, and large binding energies for T20), they probably affect the initial docking of T20 onto HR1. FIGURE 6. Four mechanisms of resistance to fusion inhibitors. Molecular models (A) and schematic representation (B) of four different mechanisms of fusion inhibitor resistance. Small amino acids (Ala/Gly/Ser) create a hole in the contact site, whereas large residues (Phe/His/Met/Trp/Tyr) form a bulk that causes steric repulsion. Negatively charged residues (Asp/Glu) cause electrostatic repulsion, whereas positively charged amino acids (Lys/Arg) can form a salt bridge with the inhibitory peptide, possibly causing non-optimal helix packing and/or docking.  Fig. 5B and supplemental Fig. S2.

HIV-1 Resistance to Fusion Inhibitors
These two types of resistance mechanisms suggest that there is a correlation between the size of an amino acid at position 38 and resistance to T20. We indeed observed such a trend (data not shown), although it was not significant. Both small and large amino acid caused resistance to T20, whereas intermediate sized amino acids displayed T20 sensitivity (Val/Ile/Leu). Several amino acids did not fit this trend. In particular, the charged amino acids behaved differently, which calls for alternative resistance mechanisms.
Electrostatic Repulsion-A third type of resistance proposed for negatively charged residues is electrostatic repulsion (Asp/ Glu). This is facilitated by the juxtaposition of a negatively charged residue (Glu-146) in HR2, T20, and T1249. The presence of a negative charge at position 38 caused a decrease in ␣-helical content, six-helix bundle stability, and T20-HR1 binding energy. Despite the fact that amino acids like Asp and especially Glu, are well accommodated in ␣-helices (60, 61), a large decrease in helical content is observed. This suggests that these amino acids do not disturb the formation of the HR1 helix but affect the HR2 helix. This explains the high level of resistance to T20 and the resistance to T1249. The structural design of the T2635 peptide reveals why this peptide is not repulsed by the negative HR1 charge. In the design of T2635, this Glu-146 residue is specifically placed in a helix stabilizing salt bridge, resulting in a net neutral charge in that area of the peptide (Fig. 5C). To directly test this scenario, we paired the negative charge of T1249 in an intramolecular salt bridge (T1249 mut ), which reduced the resistance profile of the V38D/E mutants dramatically.
Electrostatic Attraction-Interestingly, positive charges also resulted in resistance to both T20 and T1249. We do not have an entirely satisfying explanation for this, but the potential formation of a salt bridge between Arg/Lys and Glu-146 may not be accommodated in a configuration that is compatible with the proper T20-HR1 or HR2-HR1 packing (as confirmed by a reduced helicity and stability of the six-helix bundle). The introduction of a positive charge affects the electrostatic potential and might result in improper steering of the components when docking. Although a salt bridge might form, this is probably not compatible with an optimal HR1-peptide interface.
Other common T20 and T1249 resistance mutations identified in vitro and in vivo are G36D/E and N43D/K, representing changes from neutral to negative and positive charges (29,30,35,(71)(72)(73) that may influence the HR1-peptide interaction in a similar fashion as V38D/E/K/R. Residue 43 in HR1 is in close contact with residues Glu-137 in HR2 or peptide inhibitors (74). We studied the binding energies of G36D, V38D, and N43D HR1 variants with the three generations of peptide inhibitors in silico (supplemental Fig. S3 and Table S1). G36D, V38D, and N43D caused a decrease in binding energy of T20 corresponding with the observed resistance (29,30,32). N43D, but not G36D and V38D, caused a decrease in binding energy of T1249, confirming that binding energy is not the only determinant for sensitivity to T1249. All three HR1 variants displayed decreased binding energy with T2635, but the differences are relatively small when the total binding energy is taken into account. Thus, the resistance mechanisms we describe are likely to apply to other positions than 38 as well.

Implications for the Design of Novel Antiviral Fusion Inhibitors
We can make two important recommendations for the design of novel peptide fusion inhibitors. First, we give indications for the relative minimal binding energy that is required for potent inhibition and that may prevent easy viral escape by a single mutation in the binding site. To prevent a single substitution in the binding site from causing resistance, the binding energy for the peptide-target interaction, which can be computed from the six-helix bundle structure, should be high such that a single mutation only causes a relatively small change in the overall binding energy. We cannot give absolute values as our calculated binding energies are only an approximation and different methods might lead to different estimates. An additional advantage of such a mutation-tolerant mechanism of inhibition is that more natural virus variants will display sensitivity to the drug. The accumulation of multiple resistance mutations may provide resistance but will require more time (75) and will likely have a more dramatic effect on viral fitness.
Second, our data suggest that the presence of exposed charges on the peptide at the drug-target interface, unless involved in an intramolecular salt bridge, are not desirable as it provides the virus with an easy possibility to generate the most powerful mechanism of resistance; that is, electrostatic repulsion. These findings may guide the design of novel fusion inhibitors targeting viruses with class I fusion proteins.