**2. Results**

#### *2.1. Structural Properties of Modeled Membrane Proteins*

Firstly, we aligned the primary sequences of two proteins, murine UCP2 protein and bovine ANT protein (Figure 1). Although the homology of both proteins is 24%, the crucial amino acid positions and motifs characteristic for even and odd-numbered transmembrane helices, namely πGπxπG (helices 1, 3 and 5) and πxxxπ (helices 2, 4 and 6), are conserved. Furthermore, the positions of amino acids corresponding to matrix or cytosolic salt bridge network, as well as proline kinks at odd-numbered helices, also remain preserved (Figure 1).


**Figure 1.** Alignment of the primary sequences of UCP2 protein (PDB code: 2LCK) and bovine ANT protein (PDB code: 1OKC). Secondary structure alpha helices are denoted in the yellow shade, and missing residues from the crystallographic structure are presented with red boxes. Residues that constitute the salt bridge network at the cytoplasmic side are shown in dashed black boxes, while potential residues that could constitute the salt bridge network at the matrix side are enclosed in solid black boxes. πGπxπG (helices 1, 3 and 5) and πxxxπ (helices 2, 4 and 6) motifs are depicted in violet and green boxes, respectively. Residues responsible for proline kinks are enclosed in yellow boxes.

> Taking into account that strategic pillars, including cytosolic and matrix salt bridges as well as shape-forming proline kinks in both structures were conserved, we felt that it was safe to take the ANT structure as a starting point for subsequent MD simulations. In our previous MD simulations of two different crystallographic structures belonging to differently open ANT states towards the cytosolic [38] or matrix [42] side of the inner mitochondrial membrane, we have shown that it is possible to capture important conformational changes of the protein embedded in the membrane within microsecond MD simulations [44]. Moreover, we have also revealed that an order shorter timescales of ca. 200 ns, used in previous MD simulations of UCP2 NMR structures in DOPC bilayers [20], were not sufficiently long for a more relevant description of critical regions in the protein, such as reversible salt bridge breaking and forming [44]. In addition to MD simulations of the UCP2h, we also repeated simulations of UCP2NMR based on the NMR structure

of UCP2 [20] and compared them to referent ANT simulations at relevant microsecond timescales. A schematic representation of the UCP2 structure is depicted in Figure S1.

Figure 2 shows the time evolution of root mean square deviation (RMSD), which indicates the three studied protein structures' stability in time. RMSD deviation was the largest for UCP2NMR, which was not surprising, given the fact that this structure obtained in alkylphosphocholine detergent was determined in a non-optimal environment [20,22,33]. It was visible that the extension of simulations by Zoonens et al. [20] showed even larger deformations of the structure (especially after 1 μs), which will be analyzed in more detail in later sections. MD simulations of UCP2h were more stable, although a small increase in the RMSD occurred at the end of simulation time. However, these instabilities were not as severe as in the UCP2NMR simulations and were of a similar order of magnitude as the RMSD oscillations observed for the referent ANT structure. However, although very useful for the general description of protein stability, the RMSD analysis was a simple "one-number" analysis and did not contain information on the conformational changes of specific residues [45]. For this reason, we turned to root mean square fluctuation (RMSF) analysis (Figure 3), which provided important (but not time-resolved) data on the flexibility of particular residues. Importantly, we saw that the UCP2NMR structure was more flexible (and less stable) around residues found in the water phase, oriented to the matrix side (especially around residues 250–270 and C-terminus) compared to UCP2h and referent ANT structures. Figure S2 shows the time evolution of the UCP2NMR and UCP2h secondary structures which remained preserved for both structures in the simulation time.

Finally, another very useful analysis of general structural parameters was obtained by the principal component analysis (PCA) of backbone carbon atoms of the protein. PCA is a procedure that reduces a multidimensional complex set of all possible conformational degrees of freedom to lower dimensions along which the main conformational changes of protein are identified. The PCA analysis of UCP2NMR and UCP2h is shown in the Figure 4. We can see that the area spanned by the first two principal components (PC1 and PC2) was much larger in the case of UCP2NMR structure in comparison to UCP2h structure. This further supports the above analysis, showing that UCP2NMR structure was more flexible and less stable in DOPC phospholipid bilayer. The analysis indicates that the protein structure tried to find its optimal position and a proper fold in the membrane, which was not attainable at a microsecond time scale and probably orders of magnitude longer simulation times were needed. In contrast, the area spanned by PC1/PC2 in the UCP2h structure was much smaller and relatively compact, demonstrating that it was stable in the bilayer within our simulation time. It was in line with RMSD and RMSF analyses and with reported microsecond MD simulations of ANT protein [44].

**Figure 2.** Time propagation of the RMSD values for UCP2NMR, UCP2h and the referent structure ANT. The RMSD is calculated for backbone carbon atoms (Cα) of protein.

**Figure 3.** RMSF analysis of (**a**) UCP2NMR and UCP2h structures and (**b**) referent ANT structure. The RMSF is calculated for backbone carbon atoms (Cα) of protein. Light red color corresponds to protein residues inside the phospholipid bilayer, light blue represents protein residues immersed in water at the cytosolic side, whereas dark pink corresponds to the protein residues immersed in water at the matrix side.

**Figure 4.** PCA analysis-2D projection of UCP2NMR and UCP2h protein conformations onto common first and second principal components (PC1 and PC2) are presented in blue and red color, respectively.

#### *2.2. Stability of Salt Bridges Exposed to the Cytosolic and Matrix Side of the Inner Mitochondrial Membrane*

As a next step, we now focused on the stability of the salt bridge networks formed at the cytosolic and matrix sides of the modeled UCP2 structure (Figure S1). Opening and closing the cytosolic and matrix side of the ANT protein via salt bridges, which are connected to the transport of ADP and ATP nucleotides across inner mitochondrial membranes, involves at least 10 kcal mol−<sup>1</sup> for breaking the salt bridge network [40–42]. However, it is essential that water does not leak through the protein interior since it would abolish strictly controlled proton transport due to water-mediated ion exchange, as had been shown by functional leakage assays [20]. Therefore, as an initial prerequisite for controlled proton transfer, the UCP2 protein should be impermeable to water in order not to allow short-circuiting of the system, which is possible only if the salt bridge network is closed and constricts the protein at the matrix [38] or cytosolic side [42] as found in the corresponding crystallographic structures of ANT and subsequent MD simulations [44]. These experimental results further indicated that the proton transport mechanism, either via UCPs or ANT [6], was not controlled by direct transport of proton through the protein interior, but involved the transport of FA anion (and in turn proton) alongside the protein/lipid interface [1,8,9,46].

The salt bridge networks analysis showed that in the case of the UCP2NMR structure (Figure 5a,b), only one residue pair (Asp35-Lys141), located at the matrix side of the protein, permanently formed a salt bridge within our simulation time. In contrast, two other salt bridges located at the matrix side (Asp236-Lys38 and Asp138-Lys239) were not making a salt bridge, as well as three other salt bridge pairs at the cytosolic side (Asp198-Lys104, Asp101-Lys295, and Glu292-Lys201). On the other hand, salt bridges at the matrix side formed in the case of UCP2h structure (Asp35-Lys141, Asp236-Lys38, and Asp138-Lys239) were stable and persistent (Figure 5c) just as in the case of the analogous salt bridges in the referent ANT structure (Glu29-Arg137, Asp231-Lys32, Asp134-Arg234) presented in Figure 5e. Cytosolic salt bridges were partially closed in UCP2h (Figure 5d). In contrast, they were fully opened in the case of ANT (Figure 5f). These results imply that water leakage should be largely suppressed in the case of the UCP2h structure due to the closed matrix side of the protein, which is the pivotal condition for the protein structure to have a relevant functional role in the proton transfer mechanism. We reached similar conclusions by analysis of the referent ANT structure. However, in the case of UCP2NMR structure, we showed that due to the opened matrix side of the protein, water leakage was possible across the protein interior (more details are found in the next section). It was similar to the observations by Zoonens et al. from their shorter analogous MD simulations [20]. We should also mention that in the case of UCP2h and ANT structures, the distances between pairs of negatively charged residues at the matrix side (i.e., EGmotif), which were highly conserved across mitochondrial ADP/ATP carriers [47], kept three-fold pseudosymmetry in contrast to the UCP2NMR structure where this motif was not conserved, and distances between the negatively charged residues were larger (Figure S3). In this way, we further showed that the UCP2NMR structure was unstable and functionally irrelevant when embedded in phospholipid bilayers, which were structurally significantly different compared to the alkyl phosphocholine environment serving as an extracting agen<sup>t</sup> [20,22,33].

**Figure 5.** Analysis of the salt bridge network for UCP2 model based on (**<sup>a</sup>**,**b**) the UCP2 NMR structure (UCP2NMR), (**<sup>c</sup>**,**d**) the UCP2 model based on the crystallographic structure of ANT (UCP2h), and (**<sup>e</sup>**,**f**) the referent ANT structure. Distances between residues that can form a salt bridge network at the matrix side are shown in panels (**<sup>a</sup>**,**c**,**<sup>e</sup>**). Distances between residues that can form a salt bridge network at the cytoplasmic side are shown in panels (**b**,**d**,**f**). Distances are calculated between centers of mass of the corresponding residues. A top-down view on the matrix exposed side of selected protein snapshots of UCP2NMR and UCP2h structures is shown on the right.

#### *2.3. Water Leakage across the Protein and Permeability Calculations*

The analysis of the salt bridge networks in the previous section suggests that UCP2NMR structure should be more water permeable due to the simultaneously open matrix and cytosolic sides of the protein in contrast to the partially closed UCP2h structure (Figure 5). To quantitatively analyze this assumption, we performed a detailed analysis of the water density inside the protein for both structures and calculated corresponding water osmotic permeability coefficients *Pf* using the method described in Zoonens et al. [20]. The analysis of averaged water density inside the protein along the *z*-axis for UCP2NMR, UCP2h and the referent ANT structure is shown in Figure 6a, with the time evolution shown in Figure S4. Interestingly, although the averaged number density of water was averaged across *z*-coordinate and did not include the differences in the *x*- and *y*- directions, the minimal value of the number density was similar for all structures, being less than a half of the water molecule per nm3. Thus, it was not very informative of the possible formation of a continuous water channel, which would possibly enable water-mediated direct proton transfer leading to the inactive UCP protein [44]. However, we should mention here that the presence of a continuous water channel is not a key prerequisite for efficient proton transfer across the membrane protein and that electrostatic effects resulting in a high energy barrier for proton transfer predominate, such as in a case of aquaporins [48,49].

**Figure 6.** (**a**) *z*-averaged water number density based on 2 μs simulations for UCP2h, UCP2NMR and referent ANT structures, (**b**) snapshots presenting volume map of water in transparent blue with surface isovalue set to 0.2 for UCP2NMR structure (left side) and homology modeled UCP2h structure (right side). Cytosolic and matrix sides of UCP2 protein are indicated.

A better look at Figure 6a revealed that the area, and in turn the total volume of water, was largest in the case of the UCP2NMR structure (blue curve), in contrast to UCP2h and ANT number density profiles, which were considerably wider (red and black curve, respectively). This is better visualized in Figure 6b, where we saw that the average volume map of water was continuous along the protein interior (left panel) in the case of UCP2NMR structure. In contrast, two disjointed volume maps existed in the case of the UCP2h structure, indicating that the water channel was not formed (right panel), similar to ANT protein.

However, these analyses are still not fully quantitative, and therefore we turned to water osmotic permeability calculations *Pf*, to compare the data to other systems. The results of the calculations are shown in Table 1.

**Table 1.** Water osmotic permeability coefficients calculated for four distinct membrane protein structures in different simulation times.


Osmotic permeability coefficients were calculated for four different membrane structures. First, we calculated water osmotic permeability for the equilibrated UCP2NMR structure (i.e., only after short initial equilibration), which closely corresponded to the

experimental NMR structure. The calculated *Pf* is (5.7 ± 0.4) × 10−<sup>13</sup> cm<sup>3</sup> s<sup>−</sup>1, which is comparable to the value of 5.3 × 10−<sup>13</sup> cm<sup>3</sup> s<sup>−</sup><sup>1</sup> obtained by Zoonens et al. for an analogous system [20]. Similarly to their observations, the *Pf* decreased after 200 ns to (3.2 ± 0.2) × 10−<sup>13</sup> cm<sup>3</sup> s<sup>−</sup>1) and finally after 2 μs it assumed the value of (1.3 ± 0.1) × 10−<sup>13</sup> cm<sup>3</sup> s<sup>−</sup><sup>1</sup> which showed a certain collapse of the water pore in the protein. However, this number was still comparable to the water osmotic permeability of the α-hemolysine, where this value was calculated to be 1.9 × 10−<sup>12</sup> cm<sup>3</sup> s<sup>−</sup><sup>1</sup> [50]. Since these values were comparable, it was clear that the UCP2NMR structure was behaving quite similarly to the water channel, which was physiologically irrelevant for UCP2 function [20]. On the other hand, the water permeability coefficient of UCP2h structure, calculated after 2 μs, was by three orders of magnitude lower being *Pf* = (2.0 ± 0.5) × 10−<sup>16</sup> cm<sup>3</sup> s<sup>−</sup>1. This was additionally confirmed by simple counting of water molecules that crossed across the protein, where we saw that UCP2h and ANT protein were virtually impermeable in contrast to the UCP2NMR structure (Table S1). These results are in accordance with the general structure analysis described in the previous sections, and promote UCP2h structure as a potentially relevant structure for further MD simulation and mechanistic studies. We should mention here that the MD simulations of UCP2 based on ANT homology structure presented in Reference [37] agreed with the presented MD simulations. In particular, the constriction at the matrix side and opening of the cytosolic side of UCP2 protein had been observed as well, together with the low number density of water inside the protein cavity similar to the UCP2h structure (Figures 5 and 6).

#### *2.4. Binding of ATP in the UCP2 Cavity*

To further verify key functional elements of the UCP2h structure, we performed a series of MD simulations (see Simulation Details) to evaluate the binding properties of ATP nucleotide in the UCP2 cavity which are known to inhibit proton transport in UCPs. It has been suggested in the literature that three positively charged arginine residues in the UCP1 cavity bind a negatively charged phosphate group, which leads to the conformational change of the protein and inhibition of proton transport [2,51–53]. This mechanism can be extended to other UCP proteins as well since the arginine residues responsible for nucleotide binding are conserved in other homologs as illustrated in Figure S5. MD simulations of ATP binding in the UCP2h cavity show that the ATP phosphate group binds tightly in the protein cavity, having all three phosphate groups bound to arginines R88, R185, and R279 (Figure 7). This can be inspected by analyzing average distances between phosphorous atoms present in ATP and the center of mass of arginine residues in the UCP2 structure (Figure S6). In particular, we observed that in the case of UCP2h structure, the binding of phosphate to arginine residues was tight, with all three phosphate groups bound to arginine residues R88, R185 and R279. This was in striking contrast to UCP2NMR structures after 20 ns and also after 2 μs where simultaneous binding of ATP phosphate groups to arginine residues did not occur and average distances between the groups were significantly larger (Figure S6), implying in turn weaker binding of ATP. This is in line with previously suggested binding motifs of ATP in UCP1 and UCP3 [54] and very tight binding of GDP in the UCP1 cavity determined by titration calorimetry experiments [52]. Moreover, it has been found that in the case of AAC3 protein, binding of carboxyatractyloside (CATR) inhibitor at the analogous location in the protein cavity as ATP was by several orders of magnitude weaker if the AAC protein structure was obtained by extraction with DPC detergent in contrast to the native crystallographic structure [24]. Therefore, we believe that the molecular description of ATP binding in the UCP2h cavity further promotes the relevance of the homology modelled structure for further MD simulation studies.

**Figure 7.** Simultaneous ATP nucleotide binding to three arginines (R88, R185, R279) in the case of UCP2h protein, with R279 (depicted in gray) being found to primarily bind to P<sup>α</sup>. R88 (shown in blue) binds to <sup>P</sup>β (occasionally to Pα), while R185 (depicted in red) binds predominantly to <sup>P</sup>γ. Water molecules are omitted for the sake of clarity.

#### *2.5. Binding of Fatty Acid to UCP2*

In previous sections, we focused on the general structural parameters of the UCP2 protein obtained by MD simulations. Finally, we turned to additional experimental verification of MD results by complementary experiments using model membranes to prove whether the suggested UCP2h structure might be physiologically relevant.

NMR titration experiments, as well as proton flux assay measurements, sugges<sup>t</sup> that a patch of positively charged residues around R60 residue in UCP2 (consisted of K271, R267, R40, and R71 residues) is relevant for the proton transport mechanism since it serves as a binding site for FA anion [25]. We should stress here that the proton flux assay measurements presented in Reference [25] were performed in liposomes with correctly folded protein. It suggests that the active site in UCP2 is preserved regardless of the protein extraction medium. We performed experiments with the recombinant mutant UCP2 to check whether the mutation from R60 to S60 affected the proton conductance (see details in Materials and Methods). We also performed 500 ns of complementary MD simulations with the AA anion (AA−) to visualize the binding process at the molecular level. Figure 8c shows the comparison of the experimental total membrane conductance (*Gm*) in two systems: the wild type UCP2-WT protein and the UCP2-R60S mutant protein. Both proteins were measured in the presence and absence of AA. We observed two effects. Firstly, the addition of AA− was essential for the increase of *Gm* as shown in our previous works [9,10], as the conductance increased by order of magnitude compared to the neat WT protein. Secondly, the mutation from R60 to S60 had a significant effect on the *Gm*, thus further supporting the hypothesis that R60 is a possible binding site for AA<sup>−</sup>. It agreed with NMR titration experiments and proton flux assays [25]. Moreover, the addition of ATP, which is an efficient inhibitor of the UCP2 proton-transporting function [3,10,43,55], also showed that it was effective only in the UCP2-WT and less effective in the UCP2-R60S (Figure S7), confirming that R60 had an important role in the proton transport mechanism. However, based on the present MD simulation results, it is still questionable whether other positively charged residues along the outer protein ring at the matrix side could also serve as potential binding sites of AA<sup>−</sup>.

**Figure 8.** Snapshots of 500 ns simulations of UCP2 homology modeled structures (UCP2h) with added arachidonic acid anion (AA−). A starting structure of protein used for binding calculations is taken from UCP2h MD simulations after 2 μs. UCP2h structure with AA− (**a**) bound to R60 and (**b**) not bound to the mutated binding site S60. A volume map of phosphorus atoms of DOPC lipid is presented in transparent gray color (surface isovalue set to 0.0064) and volume map of AA− is presented in transparent orange color (surface isovalue set to 0.02). (**c**) Specific membrane conductance of lipid bilayers in the absence of protein (white), and presence of UCP2-WT (orange) or UCP2-R60S (brown). Membranes were made of 45:45:10 mol.% DOPC:DOPE:CL reconstituted with 15 mol.% AA where indicated. Lipid and protein concentrations were 1.5 mg/mL and 4 μg per mg of lipid, respectively. The buffer solution contained 50 mM Na2SO4, 10 mM TRIS, 10 mM MES and 0.6 mM EGTA at pH = 7.34 and *T* = 306 K. Data are represented as the mean and standard deviation from three independent experiments. (**d**) Zoomed region of AA− binding to UCP2h structure. (**e**) Zoomed region of AA− binding to UCP2-R60S mutant structure. (**f**) Total number of contacts within 0.35 nm between all AA− atoms and UCP2-WT and (red color) and UCP2-R60S mutant structures (blue color), respectively. Total number of contacts within 0.35 nm between AA− carboxyl atom towards UCP2h(orange color) and UCP2-R60S structures (green color).

A detailed analysis of MD simulations was in full agreemen<sup>t</sup> with conductance measurements using model membranes. In Figure 8a,d (which is a zoomed region of the binding site) we can see that AA− fitted nicely to the binding site, at the same time having two distinct modes of interaction, i.e., salt bridge formation between R60 and AA− and stabilizing hydrophobic interactions between AA− and protein α-helices. This was also clearly visible in the analysis of total contacts between AA− and protein (Figure 8f, red and orange curves). Conversely, interaction between AA− and S60 was severely diminished (Figure 8b,e), resulting in no formation of a salt bridge between S60 and AA<sup>−</sup>, as well as a greatly reduced number of total contacts between AA− and protein α-helices within 0.35 nm (blue and green curve, Figure 8c). Interestingly, the number of contacts within 0.35 nm between carboxylic carbon atom C1 of AA− was similar in both cases. The present analysis implies that the unsaturated 20:4 AA anion, with its four cis double bonds, which are conformationally quite restricted, fits much better to the UCP2h-WT structure than the UCP2-R60S mutant.

Interestingly, the activation of UCP2-WT with saturated 20:0 arachidic acid (ArA) showed lower *Gm* than with its unsaturated 20:4 counterpart [10], thus further pointing out the importance of hydrophobic contacts for proper binding of FA anion to R60, which were reduced and entropically unfavored in a far more flexible fully saturated ArA.

Finally, we also analyzed how AA− binds to the UCP2NMR structure in two cases, after equilibration (to mimic the NMR experimental structure) and after 2 μs of MD simulations (to see the effect of relaxation in the bilayer). In the case of the equilibrated NMR structure, AA− bound very poorly to R60, showing no permanent salt bridge formation (evidenced by a smaller number of contacts between C1 atom and protein in comparison to the UCP2h structure shown in Figure 8) and very few total contacts with the protein (Figure 9a,c,d). In addition to the increased permeability of the protein to water, which is a fundamental structural problem, we also saw that binding of AA− to R60 was not established, which even further disqualified UCP2NMR structure in regard to new mechanistic studies. After 2 μs of MD simulations, the total number of contacts remained low and the situation was actually even worse as AA− could not achieve proper hydrophobic interactions with one of the α-helices which was displaced from the rest of the UCP2 protein structure (Figure 9b,e,f). This implies a possible denaturation or even disintegration of the experimental UCP2NMR structure when transferred from the alkyl phosphonate detergent environment to the phospholipid bilayer milieu.

**Figure 9.** Snapshots of 500 ns simulations of UCP2NMR structures with added arachidonic acid anion (AA−). The starting structures of protein used for binding calculations are taken from UCP2NMR MD simulations (**a**) after equilibration and (**b**) after 2 μs. A volume map of phosphorus atoms of DOPC lipid is presented in transparent gray color (surface isovalue set to 0.0064) and volume map of AA− is presented in transparent orange color (surface isovalue set to 0.02). (**c**) Total number of contacts within 0.35 nm between all AA− atoms and UCP2NMR after equilibration. (**d**) Zoomed region of AA− binding to UCP2NMR structure after equilibration. (**e**) Zoomed region of AA− binding to UCP2NMR structure after 2 μs. (**f**) Total number of contacts within 0.35 nm between all AA− atoms and UCP2NMR after 2 μs.
