Scientific Papers

Reanalysis of a μ opioid receptor crystal structure reveals a covalent adduct with BU72 | BMC Biology

The missing atom is not magnesium

Placing an Mg2+ ion in the unexplained density followed by refinement confirmed the earlier reports of a good fit, with no excess or unexplained density above 2.5 σ (Fig. 3a; data in Additional files 1 and 2). However, contrary to these prior reports, the N–Mg bonds were unrealistically short (1.9 and 1.7 Å). Compare the N–Mg bond lengths in structures of subatomic resolution: 2.19 ± 0.06 Å, mean ± s.d. (standard deviation) [11]. These proposed bonds are thus extreme outliers, with Z scores of −5 and −9, respectively. The high resolution of the structure (2.1 Å) allows strong conclusions about bond lengths, with a diffraction precision index (DPI) of 0.22 Å for the Mg2+ ion [12]. Note also that despite these short distances, the ion was not centered in the density, suggesting that the actual bonds must be even shorter (Fig. 3a). This resulted in a poor real-space R value (RSR) of 0.32 for the Mg2+ ion, despite good values for His54 (0.11) and BU72 (0.08).

Fig. 3
figure 3

Proposed magnesium complex. a Bond lengths and B-factors (red). b Proposed third bond from Mg2+ to Tyr1483×33. c Comparison with the salt bridge to Asp1473×32

A later report from the same group added a third bond to the model [10], from Mg2+ to Tyr1483×33 (Fig. 3b), expressed in generic GPCRdb numbering [13]. However, this proposal requires an O−Mg bond length of 3.1 Å; compared with high-resolution structures (2.10 ± 0.04 Å), this is untenable (Z = 25) [11]. It is instead suggestive of a hydrogen bond to another element. Note also the large gap in the electron density along this proposed bond, unlike the strong and uninterrupted density for the other bonds (Fig. 3b). Additionally, note the highly asymmetrical geometry required, with a bond angle of 105°, compared to 90° for the N atoms: magnesium complexes are symmetrical [11].

Other evidence against Mg2+ was revealed by CheckMyMetal, a metal binding site validation server [14]. The values of five of the eight parameters evaluated were classed as dubious, including three that strongly suggest a misidentified element:

  • A much higher temperature factor (B-factor) for the ion than the bonding partners (Fig. 3a); since bonds transmit thermal motion, this is implausible [15].

  • Bonding to a protonated amine (NH+); Mg2+ favors neutral or negatively charged bonding partners [16].

  • An incomplete coordination sphere. The expected number of bonds is six, or in rare cases four or five; a value of two is extremely rare in high-resolution structures [17].

While unresolved water molecules might complete the coordination sphere, this is implausible since the rest of the complex is well resolved with full occupancy, as are many structured water molecules elsewhere in the binding pocket [3].

Finally, no source of magnesium is mentioned in the experimental method [3]. Collectively, the above evidence firmly excludes Mg2+ as a candidate.

The missing atom forms covalent bonds to both BU72 and His54

The fit of the Mg2+ ion to the density establishes that a non-hydrogen atom is present in this approximate position. As noted above, this missing atom is likely nearer to both His54 and BU72 than the modelled position of Mg2+; that is, < 1.9 Å from each (Fig. 3a). This is much too close for non-covalent interactions (≥ 2.4 Å) [18], which would also not result in strong, uninterrupted electron density connecting the three atoms. For instance, the protonated tertiary amine of BU72 forms a charge-assisted hydrogen bond to Asp1473×32 (Fig. 3c); these are among the shortest of all noncovalent interactions [18]. Nonetheless, the NO distance is 2.6 Å, and the regions of high electron density are well separated, in striking contrast to the continuous density surrounding the proposed Mg2+ complex. Therefore, the unidentified atom is covalently bonded to both BU72 and μOR; that is, they form an adduct.

While this evidence does not definitively establish the identity of the missing atom, it is inconsistent with the published model of BU72 and the receptor as discrete entities. One way to resolve this would be to model the adduct, but leave the bridging atom unidentified. Many Protein Data Bank (PDB) models include unidentified atoms (ligand identifier UNX). Nonetheless, the evidence is sufficient to exclude some elements, as discussed below.

The missing atom is very unlikely to be a metal, but may be oxygen

The CheckMyMetal validation report for the magnesium complex suggested alternative metals as better candidates: copper, iron, cobalt, nickel, manganese, and zinc. However, each of these also gave multiple outliers when validated. Also, of these metals, only nickel was present during preparation of the crystals; it was used for affinity purification [3]. The bond lengths are more plausible than for magnesium, since N−Ni bonds are short (1.88 ± 0.03 Å) [11]. However, as noted above, nickel did not fit the electron density, leaving a substantial excess [9]. Further evidence against nickel and other heavy metals is the lack of anomalous scattering noted in the original report [3].

The only metal in the buffer solution, sodium, also gave five dubious values in CheckMyMetal, including even more extreme outliers from typical N−Na bond lengths (2.46 ± 0.02 Å, Z =  −29 and −40) [11], and a much worse fit to the density than magnesium [9]. Indeed, no metal forms coordination bonds to N shorter than 1.76 Å [11]. It is thus extremely implausible that the missing atom is a metal.

Given the above, it appears that the missing atom is a non-metal approximately isoelectronic with magnesium, but forming shorter bonds. The element must also be at least divalent, and can probably form hydrogen bonds given its distance to Tyr1483×33 (~ 3.1 Å). One candidate meeting these criteria is oxygen; water molecules in crystal structures are frequently misidentified as magnesium [19].

A known source of reactive oxygen species contacts the unexplained density

Formation of an oxygen-bridged adduct between the secondary amine of BU72 and the imidazole ring of His54 would require harsh conditions. Reactive oxygen species (ROS), for instance, can oxidize secondary amines [20] and histidine [21]. But how might these ROS arise? Surprisingly, several potential sources were present. The BU72-μOR complex was purified and crystallized in HEPES buffer, which generates hydrogen peroxide on exposure to light [22]. HEPES has also been reported to enhance metal-catalyzed generation of other ROS from hydrogen peroxide [23]. A further potential source is the N-terminus, which contains a sequence motif known to generate ROS. The N-terminus used was truncated, leaving glycine as the first residue and histidine as the third [3]. This sequence motif (H-Gly-Xaa-His-) forms redox-active nickel coordination complexes [24]. Moreover, a nickel affinity column was used for purification [3], and the H-Gly-Xaa-His- motif can capture Ni2+ ions from these columns [25,26,27]. The resulting square planar nickel complexes catalyze the decomposition of hydrogen peroxide to other ROS, including the hydroxyl radical [24, 28], which has been described as “the most reactive biological oxidant” [29]. Thus, the conditions used were sufficient to generate ROS near His54, potentially oxidizing both the residue itself and BU72.

A search of PDBeMotif [30] revealed eight protein structures in which square planar Ni2+-Gly-Xaa-His- complexes were resolved: PDB entries 1JVN [31], 1XMK [32], 2RJ2 [33], 3RDH [34], 3UM9 [35], 3ZUC [36], 4I71 [37], and 4OMO [38]. In three of these cases, the nickel was not added during crystallization, but unexpectedly captured during affinity chromatography: 1JVN [25], 3UM9 [26], and 3ZUC [27]. Intriguingly, in 1JVN the electron density was not consistent with the expected ligand structure; no density supported several of the atoms, suggesting partial decomposition [25]. The buffer used, PIPES, is an analog of HEPES that also generates hydrogen peroxide [39] and other ROS [23]. This provides a plausible explanation for the decomposition of the ligand.

Proposed formation and structure of an oxygen-bridged adduct

Two previous reports of adduct formation between aminoxyl radicals and imidazole rings are shown in Fig. 4a [21, 40]. These suggested potential structure 6 for an adduct between BU72 and His54 (Fig. 4b). The stereochemistry of the histidine derivative was dictated by the observed density. A possible intermediate aminoxyl radical is also shown; these can form via oxidation of secondary amines by ROS [20].

Fig. 4
figure 4

Adduct structures. a Previously reported adducts 4 ([21], Fig. 7c) and 5 ([40], Scheme 2). b Proposed adduct 6, with the nickel complex and a possible aminoxyl intermediate

This proposal finds support in a puzzling result from the original report. Despite the very strong interactions apparent between BU72 and His54, removal of the side-chain of His54 by receptor mutagenesis had no detectable effect on the affinity or potency of BU72 [3]. This seeming paradox, however, is consistent with the mechanism proposed here. Affinity and potency were measured using cells and cell membranes rather than purified proteins, so no nickel was added. Moreover, the cells expressed the full-length receptor, which lacks the N-terminal motif that forms nickel complexes [24]. Thus, the reactions proposed above could not occur, and the assays would be unaffected by the presence or absence of His54.

The oxygen-bridged adduct fits the unexplained density

Substituting adduct 6 for His54 and BU72 gave an excellent fit, with no excess or unexplained density even at 2 σ (Fig. 5; data in Additional files 3, 4, 5 and 6) [41]. Both bonds to oxygen were of typical length (1.5 Å) and were resolved up to 4.2 σ—that is, higher density than most of the ligand itself and surrounding side-chains. Unlike Mg2+, the oxygen atom was well centered in the density. Oxygen also gave a superior B-factor to Mg2+, both lower and more consistent with its bonding partners, making this a much more plausible candidate element (Fig. 5) [15]. The lower B-factor for oxygen results in a more precise fit (DPI 0.14 vs 0.22 Å). Indeed, it is among the most precisely-resolved atoms in the entire structure. The bridging oxygen and modified histidine moiety make favorable polar contacts with Tyr1483×33, which are close to the length of weak hydrogen bonds.

Fig. 5
figure 5

Fit of adduct 6 to density, with B-factors (red) and polar contacts to Tyr1483×33

The adduct is highly strained

The bound geometry of adduct 6 gave acceptable ligand validation metrics, which were superior to the original model of BU72, 1a (Table 1; data in Additional file 7).

Table 1 Ligand validation: geometry relative to Grade restraints, and electron density fit from PDB validation reports

The only severe outlier was the bond angle at the bridging oxygen (131° vs the ideal 109°: Z = 7.2). There are several indications that this is real strain rather than a fitting artifact, however. The angle is clearly resolved at high density and is consistent with tension from the tethered N-terminus. The phenyl group is bent 11° out of plane, consistent with being pulled against the adjacent residue Ile1443×29 by the same tension (Fig. 6b). This bend is also clearly resolved and is comparable to those seen in severely strained aromatic residues at subatomic resolution [42]. It also yields a more complementary fit to Ile1443×29 than the original model, as well as eliminating another small pocket of unexplained density (Fig. 6).

Fig. 6
figure 6

Fit of phenyl group to adjacent residue Ile1443×29, shown with solvent-accessible surfaces. a Original model (5C1M v.1.5). b Adduct

Strain is also evident in the N-terminus itself: in both this model and the original (5C1M v.1.5), Thr60 adopts a rare and high-energy cis-peptide bond, and there are many energetically unfavorable clashes along the peptide backbone (Fig. 7).

Fig. 7
figure 7

Polar contacts (< 3.6 Å) and clashes of the N-terminus in the adduct model. Note the high-energy cis-peptide bond at Thr60

Alternate modelling can eliminate the cis-peptide bond, as in the revised version of the original model (5C1M v.2). However, this results in a worse fit to the density, which is extremely weak in this region: several side-chains and even parts of the backbone are unresolved at 1 σ, yielding eight RSR outliers in the N-terminus, five of which are severe (Fig. 8). Atomic displacements in the N-terminus are also extremely high: the occupancy-weighted average B-factor (OWAB) of the last seven residues (58–64) are higher than 95% of residues in the structure. Indeed, Gln59 has the highest value in the entire structure, 159 Å2, compared to a median of 46. The above features (poor density coverage, high B-factors, clashes and a probable cis-peptide bond) establish that the N-terminus is constrained in an extremely unfavorable high-energy state by the tethered ligand.

Fig. 8
figure 8

The N-terminus in the revised original model (5C1M v.2), colored by B-factor. Note poor electron density coverage for some residues; severe RSRZ outliers (> 5) are given in brackets

Source link