**3. Results**

## *3.1. Peptide Configurations*

To investigate HAfp configuration space and its alteration by two known functionaffecting mutations, we conducted a set of tREMD simulations for surface and deeply inserted peptides. Taking into account structures based on experimental data, the simulations of surface bound peptides started from all hairpin or all boomerang structures. In turn, all deeply inserted peptides were initially modeled in hairpin conformation, based on our previous results obtained using the self assembly technique [20].

## 3.1.1. Surface Placement

Regardless of initial configurations, surface bound peptides started to interconvert between tight helical hairpin and open boomerang forms, as is apparent based on the gradual formation of bimodal RMSD distributions (Figure 1a). Representative structures, obtained in each case as the centroids of the most populated conformation clusters, correspond well to experimentally determined boomerang and hairpin geometries, with minor differences between the considered peptide versions (Figure 1b and Figure S1). The insertion depth of surface configurations turns out to be the same for each variant (Figure 1e and Figure S1) and displays only slight conformation-dependent alterations. In general, the N-helices do not change their position relative to membrane surface, irrespective of hairpin opening or closing, while the C-helices of boomerang configurations are slightly more deeply inserted.

**Figure 1.** (**a**) Time evolution for distributions of backbone atoms RMSD with respect to tight helical hairpin (NMR structure, PDB id 2kxa, model 1) at 310 K from tREMD simulations. (**b**) Representative conformations for two most populated states. (**c**) Time evolution of hairpin fraction in tREMD simulations based on fitted kinetic equations. (**d**) Equilibrium hairpin fractions at 310 K. (**e**) Insertion depths of N-terminal amino group nitrogen and C*α* atoms of wt HAfp ± one standard deviation (dotted lines). Dashed lines are the membrane surface (the maximum of phosphate atoms density).

Given the adopted tREMD protocol and relatively small size of the peptides under study, we aimed to determine equilibrium populations of surface conformations in each case. Unfortunately, despite the long accumulated sampling time (85, 73, and 28 μs for wt, W14A, and E11A, respectively), the RMSD distributions were still far from equilibria, indicating rather slow conformational dynamics on membrane surface. To estimate the final proportion of closed and open configurations, we assumed that the transition between them is a two-state process and considered a set of temperature-dependent kinetic equations operating on stochastic tREMD trajectories (Figure 1c) [41]. The extrapolation of fitted solutions to infinite time for tREMD simulations starting from all hairpin, and, likewise, all boomerang structures should, in principle, lead to the same equilibrium hairpin population at *T* = 310 K. Accordingly, we estimated this population as an average from both kinds of tREMD runs and obtained 0.13 ± 0.02, 0.05 ± 0.05, and 0.005 ± 0.005 hairpin fractions for wt HAfp, W14A, and E11A mutants, respectively, at *T* = 310 K (Figure 1d). In qualitative agreemen<sup>t</sup> with experimental data, this result indicates greater preference towards hairpin structure for wt HAfp compared to W14A mutant [11,14]. We did not find direct experimental evidence for the nature of the dominant E11A conformation, but, based on its reported thermodynamic binding signature with moderately unfavorable entropy component similar to W14A and rather different from highly entropy penalized wt HAfp [14], the dominance of less constrained boomerang structure is likely also in its case.

Our predicted population of 0.13 closed structures for wt HAfp23 is considerably lower compared to estimates based on NMR in DPC micelles that sugges<sup>t</sup> exclusively hairpin conformation at neutral pH, exemplified by 2kxa PDB structure [11]. In the same system, but at pH = 4, to which our simulation conditions are adjusted (protonated E11), closed conformations were found to constitute 0.8 of the entire population and were estimated to interchange with the open ones on ∼25 μs time scale [49]. Furthermore, as compared to detergent micelles, the closed form is apparently less stable in membranes, where its population of 0.7 was reported already at neutral pH [50]. Taken this into account, as well as the fact that sole W14A substitution is enough to permanently open up the hairpin, it is plausible that the stabilization of the closed structure is not very strong, and that its interconversion with non-negligible boomerang fraction indeed occurs in POPC membranes for wt HAfp.

## 3.1.2. Intramembrane Placement

Although shorter than typical transmembrane helical fragments, HAfp in hairpin conformation may be capable of adopting intramembrane position [20]. It can do so by inducing local membrane thinning that provides hydration of both hydrophilic hairpin poles necessary for the stabilization of the membrane-spanning orientation (Figure 2a). In the case of wt HAfp, the deeply inserted hairpin is located centrally within the membrane, with residues 6 and 18 positioned at the bilayer midplane (Figure 1e). Notably, the nuclear Overhauser effect between peptide amide groups and terminal methyl groups of lipid acyl chains was observed exactly and exclusively for these two residues in the study of wt HAfp23 in DPC micelles [11]. Given relatively shallow insertion depths observed in simulations of open structures (Figure 1e), this would be not possible at all without assuming intramembrane configurations.

**Figure 2.** (**a**) Representative membrane-spanning configuration of wt HAfp with average water density isosurface at half of bulk density. (**b**) Free energy profiles for unrestrained peptides center of mass in membrane-spanning configurations.

Hairpin conformation appears to be well preserved for deeply inserted peptides (Figure 1a), although we observed solitary unfolding events that lead to surfacing of the Cterminal arm at the opposite membrane side to the N-terminus—we note, however, that this would be impossible, if the peptide was attached via a linker to the remaining part of the HA2 subunit. Our simulations indicate that the intramembrane configuration is generally metastable, and can be abandoned by peptide escaping to the surface in the direction of hairpin opening. We observed no successful, complete egress in the opposite direction (i.e., towards the hairpin apex) in any of our simulations, likely due to firm anchoring of the positively charged N-terminal amino group within the aqueous compartment.

The stability of intramembrane localization appears to be affected by amino acid composition of the peptide. Wt HAfp remains most deeply inserted among the considered variants, and is separated from surface configuration by ∼1 kcal/mol free energy barrier (Figure 2b). Intriguingly, the deprotonation of E11, which may be plausible owing to the exposure of hairpin apex to the neutral pH of cellular environment on the outer side of the endosomal membrane, further deepens and stabilizes peptide insertion. The W14A mutant is least stable, with practically no free energy barrier for surfacing, and roughly 3 Å shallower insertion depth compared to wt HAfp. We attribute this behavior to the decrease of hairpin apex hydrophilicity upon the removal of W14 side chain. In turn, E11A substitution leads to intermediate insertion depth, albeit relatively well stabilized within the membrane core. In this case, in contrast to other variants, a partial remodeling of the C-terminal hairpin arm is also observed (Figure S1).

To check whether the stability of intramembrane configurations is not merely a spurious effect of the selected Amber force field family, we carried out an independent set of tREMD simulations for wt using unrelated Charmm force field (peptide and lipids). We observed a stable, membrane-spanning hairpin, with even greater free energy barrier for surfacing of ∼2.5 kcal/mol than in the Amber force field (Figure S5).

#### *3.2. Potentials of Mean Force*

To investigate peptides partitioning between membrane surface and interior, we evaluated free energy differences between the respective configurations (Figure 3). To this end, we took into account the free energy cost of peptide restraining to hairpin conformation at membrane surface assuming hairpin to boomerang ratios summarized in Figure 1d, the potential of mean force (PMF) for restrained hairpin translocation into the membrane core, and free energy gain upon the release of the restraining potential thereof. To further complement the calculations, we assessed the cost of hairpin unbinding from surface and unfolding in bulk solvent (for details, see the Supplementary Materials).

The calculated free energies of peptides binding to membrane surface (Table 1, Δ *GB*→*<sup>S</sup>*) are in the order of 11 kcal/mol. They do not include, however, the effect of peptide liberation from a fixed position in the bulk, which should generally decrease the calculated affinities by 1–2 kcal/mol, depending on the assumed free peptide concentration. Our estimate of the wt HAfp binding strength is ∼1 kcal/mol above that of the W14A mutant, which is in fair agreemen<sup>t</sup> with our own measurements (Table 1, Δ *Gexp*, Figure S9) and other published data [14]. Free energy profiles for further peptide transition from membrane surface towards the core rise steeply in all three considered cases, and barriers leading to relatively shallow intramembrane minima are achieved when peptides centers of mass are around 6 Å from bilayer midplane (Figure 3). The estimated free energy cost of peptide transitioning from membrane surface to the core, Δ *<sup>G</sup>S*→*D*, is 10 kcal/mol for the wt HAfp and 11 kcal/mol for both mutants. The overall transition cost from surface to deep minimum is a combination of three effects. The first is the cost of hairpin restraining on membrane surface, − Δ *G<sup>S</sup> h*→0. Here, naturally closed peptides have an advantage over those assuming predominantly open conformation, as is reflected by almost 3 kcal/mol difference between wt HAfp and E11A.

The second, is the cost of restrained hairpin translocation to intramembrane position. Somewhat surprisingly, the results for wt HAfp and W14A mutant are quite similar in this respect (see Figure S3 for full, detailed PMF). It indicates that, instead of changing intrinsic peptide propensity to move from membrane surface to the core, W14A mutation rather affects peptide behavior within both free energy wells by promoting the open surface form and weakening the stability of the membrane-spanning configuration. In turn, membrane penetration of the restrained E11A mutant is least energetically costly, most likely owing to increased hydrophobicity of the hairpin apex, that is the peptide part that actually penetrates through the core. The third free energy contribution, Δ *G<sup>D</sup> h*→0, comes from the removal of the restraining potential for membrane-spanning configurations. We do not see large differences here between the considered HAfp versions. In all cases, free energy gain is smaller than the corresponding effect on the surface, indicating greater hairpin rigidity within the membrane core.

**Table 1.** Experimental, <sup>Δ</sup>*Gexp*, and calculated free energy differences in kcal/mol (error estimates in subscript): Δ*GB*→*<sup>S</sup>* binding from bulk solvent to membrane surface, Δ*GS*→*<sup>D</sup>* transition from surface to deep configuration, <sup>Δ</sup>*GB*0→*<sup>h</sup>* the cost of peptide restraining to hairpin in bulk solvent, Δ*GB*→*<sup>S</sup> h* transition of restrained peptide from bulk solvent to membrane surface, <sup>Δ</sup>*<sup>G</sup>Sh*→<sup>0</sup> restraints removal for surface configuration, Δ*GS*→*<sup>D</sup> h* transition of restrained peptide from surface to deep configuration, <sup>Δ</sup>*<sup>G</sup>Dh*→<sup>0</sup> restraints removal for deep configuragion. ∗ experimental energies are for peptides with solubility tags, and † value for E11A based on experimental difference wrt. wt.


**Figure 3.** PMF for unrestrained peptides in deep and surface free energy minima. *G* = 0 corresponds to free peptides in the aqueous phase. Relative depths of free energy wells for surface and deep configurations were obtained based on: continuous PMF for restrained peptides, (un)restraining free energy calculations, and free peptides simulations (Table 1 and Figure S4). Dashed line for E11A reflects the fact that its surface minimum was estimated based on binding free energy for wt.

Taken together, the overall cost of peptide translocation into the membrane core appears to be rather high and, based on the considered simulation setup, indicates that the population of deeply bound peptides is very small. Nevertheless, the order wt > E11A > W14A observed both for the availability and stability of membrane-spanning configurations correlates well with peptides fusogenic activities (Figure S8), suggesting a hypothesis that deep insertion may be relevant to their mechanism of action.

## *3.3. N-Helix Orientation*

Experimental evidence concerning peptide orientation within the membrane often comes from ATR-FTIR spectroscopy [51]. Assuming the dominance of helical structure, at least within the N-helix, the obtained order parameter, *Sh*, allows determination of the average angle between helix axis and membrane normal, *z*. In this respect, *Sh* for an at least 23 residue wt HAfp is reported to be in the range from 0.32 to 0.59, depending on pH and membrane composition [17,18,52], which is interpreted as an oblique helix insertion, with 30–50 degrees average tilt with respect to the *z* axis. A similar oblique insertion angle of 50 degrees was suggested based on EPR measurements [10]. Our simulations do not support this view. In all surface configurations, the N-helix remains parallel to membrane plane (Figures 1e and 4), and its capping NH+3 group does not reveal any tendency for deep membrane penetration. On the contrary, if the peptides are forced to move into the membrane during umbrella sampling runs or supervised insertions (see below), they penetrate the hydrophobic core exclusively with the hairpin apex. In doing so, however, they do not assume any intermediate, oblique orientations, but rather shift sharply from parallel to almost perpendicular orientation upon passing the free energy barrier between surface and deep configurations (Figure 4).

**Figure 4.** N-helix angle with respect to membrane normal in degrees as a function of insertion depth. Lines represent data from umbrella sampling simulations for restrained hairpin conformations, while points are data from unrestrained simulations. Error bars correspond to one standard deviation. Inset: The estimated N-helix order parameter as a function of deep configurations fraction. Squares are the experimental data for *Sh*: (A) W14A [14]; (B) wt HAfp20 [53]; and (C) wt HAfp23 [17,18,52].

This simulation based picture suggests a two states contribution to the order parameter. If we calculate *Sh* with the assumption of bimodal tilt angle distribution taken from simulation data, the experimental values for wt HAfp23 are recovered for 0.5–0.6 fraction of deep configurations (Figure 4, inset). This fraction would be lower for HAfp20 (reported *Sh* = 0.18 [53], deep fraction 0.4) and nonexistent for the W14A mutant (reported *Sh* = −0.48 [14]), in agreemen<sup>t</sup> with surface only configurations in this case.

## *3.4. TRP Fluorescence*

To further confront our simulations with experimental findings, we analyzed depthdependent tryptophan fluorescence quenching by brominated lipids [54]. Here, the decrease in TRP fluorescence intensity with respect to a reference value, *F*/*F*0, caused by Br atoms located at known positions along lipid acyl chains reveals an approximate insertion depth of TRP side chains. Assuming distance-dependent model of fluorescence quenching (see Section 2.1.5 for details) [54] and applying it to peptide-lipids configurations from our tREMD simulations, we assessed what combination of major macrostate populations: surface hairpin, surface boomerang, and deeply inserted hairpin, best reproduces experimental data.

For all three considered peptides, the measured quenching effect is strongest for Br probes that are closest to bilayer surface (Br4,5) and generally decreases towards membrane interior (Figure 5a). Importantly, however, the readouts differ between peptide mutants, indicating higher fluorescence quenching by deeply buried probes in the case of the E11A mutant and wt HAfp, compared to the W14A mutant. In all cases root mean square errors (RMSE) between experimental *F*/*F*0 ratios and predictions based on MD simulations clearly disfavor standalone surface hairpin configurations (Figure 5b). Since for the wt HAfp the hairpin is indicated as a dominant structure by NMR [11], it suggests the need to include deep configurations in order to explain fluorescence data. Indeed, 20% of deep configurations in addition to surface hairpin leads to the overall lowest RMSE in this case. For the W14A mutant, the hairpin structure is unlikely [14], which leaves surface boomerang in best agreemen<sup>t</sup> with fluorescence data. Notably, in this case, the inclusion of deep configurations does not improve RMSE. On the contrary, favorable agreemen<sup>t</sup> for the

E11A mutant clearly requires the assumption of 30% of intramembrane locations, but the results do not discriminate between boomerang and hairpin configurations.

**Figure 5.** (**a**) Experimentally determined fluorescence quenching ratio, *F*/*F*0, for peptides in brominated lipids; and (**b**) RMSE values for best fit between experimental and calculated *F*/*F*0. Circle plots indicate the fraction of transmembrane configuration providing the lowest RMSE.

The results unanimously sugges<sup>t</sup> a greater tendency to penetrate membrane interior for the wt HAfp and E11A mutant compared to the W14A mutant. While there is certainly no quantitive match between population fractions assessed from the PMF calculations and the estimates based on TRP fluorescence quenching or ATR-FTIR spectroscopy discussed above, we note that the simulation model most likely overestimates the free energy difference between surface and deep minima due to limited system size that enhances the compactness of the considered membrane slab, but it still captures qualitative differences between the considered peptide variants.

#### *3.5. Membrane Perturbation*

#### 3.5.1. Lipid Tail Protrusions

The extent of peptide-promoted lipid tail protrusions [30] has often been invoked as a predictor of their ability to induce lipid mixing between membranes, as a proxy for fusogenic activity [23]. According to our simulations, the presence of surface bound peptides indeed increases the likelihood of lipid tail protrusions, but only roughly sixfold compared to pure POPC membrane (Figure 6a). Aside from being relatively weak and showing little dependence on peptide conformations, the effect observed for surface configurations is equally strong for fusion active wt HAfp and inactive W14A mutant, and, hence, does not explain the loss of function in this latter case.

**Figure 6.** Ratios of lipid tail protrusions within 7 Å distance from peptides to those in pure POPC membrane for: (**a**) surface; and (**b**) deep configurations.

The situation changes, if membrane-spanning configurations are considered. Firstly, they appear to be capable of increasing the intensity of lipid tail protrusions roughly 250 fold compared to pure membrane (Figure 6b). It is most likely a consequence of deeper insertion of the positively charged N-terminal amino group (Figure 1e and Figure S2), which allows more effective dragging of phosphate groups below membrane surface, thus promoting lipid tilting and splaying. Secondly, the assumption of only surface bound configurations for the W14A mutant, inferred from previous results, would explain its inactivity. Thirdly, comparably lower protrusion promoting capability of the E11A mutant compared to the wt HAfp corresponds well to their relative fusogenic activity. The difference is possibly related to shallower insertion depth the E11A mutant, and consequently lower burial of the Nterminal peptide charge (Figure S2). Finally, of note is the highest rate of lipid protrusions generated by the wt− HAfb with E11 deprotonated, whose occurrence in real life scenarios cannot be excluded given the exposure of the hairpin apex to cellular environment.

#### 3.5.2. Membrane Water Permeability

To further assess membrane perturbation induced by the peptides, we calculated relative membrane water permeability and compared it with experimental leakage assays of LUV encapsulated calcein.

According to simulations, peptides located at the membrane surface do not lead to any increase in its water permeability compared to pure POPC membrane (Figure 7a). However, a sharp onset of permeability is observed upon transition to deep minimum. This effect is somewhat less pronounced for the E11A mutant, possibly due to its greater hydrophobicity and smaller insertion depth. The highest permeation was observed for the most deeply inserted wt− variant. The simulated results are in fair agreemen<sup>t</sup> with the experimental leakage assays (Figure 7b). Apparently lower leakage caused by W14A mutant as compared to other peptides may be explained by its less stable deep configuration (Figure 3). Notably, however, relative magnitudes of leakage for the wt HAfp and E11A mutant observed in experiments are opposite to those suggested by simulations. In agreemen<sup>t</sup> with the interpretation of fluorescence quenching experiment (Figure 5b), this may indicate that the fraction of deeply inserted E11A configurations may be actually somewhat greater than in the case of wt. Taken together with greater stability but shallower depth of E11A transmembrane free energy minimum (Figure 2b), this would imply that comparably greater fusogenic activity of the wt is predominantly the effect of its ability to localize deeper within the membrane core. To check whether the difference between W14A and wt/E11A peptides was not driven by a diminished surface concentration resulting from an order of magnitude lower partition coefficient for W14A ( ∼1.6 kcal/mol difference in Gibbs binding free energy, Table 1), we also performed the experiment at lower (1:500) peptide/lipid ratio. All three leakage curves looked more similar to each other; however, wt and E11A peptide still led to more efficient permeation (Figure S10).

**Figure 7.** (**a**) POPC membrane water permeation for 1:162 peptide to lipid ratio, relative to pure membrane slab composed of 162 lipids. Dots correspond to values obtained from umbrella sampling simulations for restrained hairpin conformations, squares to values obtained from unrestrained simulations. (**b**) Peptide-induced calcein leakage from POPC LUVs.

## *3.6. Supervised Insertion*

As indicated by our PMF calculations, the peptides need to traverse a substantial free energy barrier of roughly 10 kcal/mol in order to reach membrane interior from surface configurations. To check whether such putative transitions are feasible for unconstrained peptides and to assess their time scale, we devised a supervised insertion scheme. It is based on consecutive simulation rounds, each including relatively short (5 ns) multiple (50) unconstrained MD runs. After each round, a structure that is most advanced along the assumed pathway is selected among all runs and is used as a seed for a subsequent round of simulations. We focussed on wt HAfp and considered the position of W14 C*α* atom along the *z* axis as a measure of insertion progression.

Of four independent supervised insertion procedures initiated from diverse starting configurations two turned out to be successful in peptide reaching the deep free energy well within 10 simulation rounds (Figure 8a, Top). In each such case, prior to achieving membrane-spanning configuration, the peptide induced the formation of a thin water wire between W14 side chain and aqueous compartment on the opposite membrane side (Figure 8b). The apparent role in this process was played by W14 indole nitrogen atom. It was effective, however, only in one of three possible side chain conformers that provided cross-membrane facing nitrogen orientation (Figure 8c), and without assuming this conformer peptide progression was stalled. Intriguingly, this conformer appears more frequently among closed rather than open surface configurations (Figure 8d), thus contributing further to hairpin insertion readiness, in addition to its already more favorable shape. To test an alternative hypothesis of N-terminus driven insertion, we also carried out simulations based on supervised terminal amino group nitrogen atom position, but none of the four trials for hairpin structure was successful (Figure S6).

**Figure 8.** (**a**) (Top) Supervised insertion trajectories for two successful (green) and two unsuccessful trials (red); and (Bottom) the probability of insertion advancing along *z* axis. (**b**) Water wire established by interaction with W14 at percolation onset (trajectory snapshot). The water molecules within 10 Å of protein atoms and *z* coordinates restricted to protein-distal membrane leaflet are shown. The orange spheres represent membrane phosphate atoms. (**c**) Three main W14 conformations. (**d**) Probability distribution for finding W14A conformers in surface configurations.

The time scale of complete peptide insertion was roughly 20 ns. By simply evaluating the probability of completing the entire route as a product of ratios of productive to unproductive runs in each of 10 rounds (see methods for details), we obtain the probability of success in a single attempt, *p* ∼ 10−<sup>12</sup> (Figure 8a, bottom). Following a coarse reasoning, for a GUV with 10 μm diameter and 160 lipids to peptide ratio, this gives an estimate of

>150 insertions per second, per vesicle, thus approaching the time scales of experimental fusion observation.

#### *3.7. Possible Insertion Modes*

The dominant view of wt HAfp23 configurations based on experimental insights to date corresponds to a tightly closed hairpin [11] which remains partially inserted into membrane, with buried N-terminus and solvent-exposed kink, such that the overall tilt angle of the N-helix with respect to membrane normal is ∼40 degrees (Figure 9c) [10,14]. We do not find support for this view in our simulations. We also note that such configuration would entail at least partial burial of the C-terminal hairpin fragment (together with a few residues from subsequent linker region, if complete HA structure was considered) whereas its strictly conserved sequence, 21WYG23, suggests rather a preference to the membrane–water interface region. Instead, we propose a more complex model, in which the peptides fluctuate between closed and open surface conformations (Figure 9a,b) and occasionally dive into the membrane core, adopting more or less stable transmembrane configuration (Figure 9d,e). In this view, the fusogenic HAfp activity would be related either to this latter configuration alone or to membrane-perturbing transitions of buried hairpins back to membrane surface.

**Figure 9.** Possible insertion modes of HAfp into lipid membrane: (**<sup>a</sup>**,**b**) surface hairpin and boomerang, respectively; (**c**) obliquely inserted hairpin; (**d**) deeply inserted transmembrane hairpin (wt); and (**e**) shallowly inserted transmembrane hairpin (E11A).
