2.1. Protein expression, purification and crystallization

The expression, purification and crystallization of MhsT in complex with its different substrates have been described previously (Focht et al., 2021) based on earlier studies (Malinauskaite et al., 2014; Quick & Javitch, 2007).

2.2. Crystal morphology and data collection

Crystals of MhsT in complex with aromatic substrates (Phe, 4-F-Phe or Tyr), Val or Leu were small, three-dimensional rod-like crystals (Fig. 1a) with lengths ranging from 30 to 50 µm, similar to MhsT–Trp crystals (Malinauskaite et al., 2014). Data collection was performed on beamlines I24 and I04 at Diamond Light Source (DLS).

Figure 1 Crystals of MhsT. (a) P21 crystal form of MhsT–Val exhibiting P2221 pseudosymmetry and negligible twinning and (b) a similar P21 crystal form of MhsT–Leu with twin fraction 0.43. (c) MhsT–Ile crystallizes in a different P21 form with translational noncrystallographic symmetry

In contrast, the crystals of MhsT–Ile were flat plates with dimensions of up to 70 µm and thicknesses of about 5–10 µm (Fig. 1b). The data sets were collected on beamline PXI at the Swiss Light Source (SLS). The MhsT–Ile crystals were very sensitive to radiation damage and complete data sets could not be obtained from single crystals.

3.1. Data processing of MhsT–Val and MhsT–Leu data sets

The MhsT–Val and MhsT–Leu data sets were initially processed using the XDS package (Kabsch, 2010) as well as POINTLESS and AIMLESS from the CCP4 package (Winn et al., 2011) in the orthorhombic space group P222₁ with systematic absences along c. The processing resulted in an overall R_meas of 0.233 in the case of MhsT–Val and 0.133 in the case of MhsT–Leu, indicating seemingly acceptable merging statistics and space-group assignment. Initial phases were obtained by the use of molecular replacement in Phenix Phaser-MR with Mhst–Trp (PDB entry 4us3; Malinauskaite et al., 2014) without TM5 and ligands as a search model, which identified one molecule in the asymmetric unit (Matthews coefficient of 2.57 ų Da⁻¹ and solvent content of 52.1% for MhsT–Val and Matthews coefficient of 2.55 ų Da⁻¹ and solvent content of 51.9% for MhsT–Leu; Matthews, 1968). Molecular replacement gave clear solutions in both cases (PAK = 0, LLG = 822 and TFZ = 18.2 for MhsT–Val and PAK = 0, LLG = 1262 and TFZ = 19.1 for MhsT–Leu), However, real- and reciprocal-space refinement stalled in both cases at an R_work and R_free of about 0.38 and 0.43, respectively.

These R-factor statistics indicated that the structure did not explain the diffraction data well, and that the assignment of P222₁ space-group symmetry was potentially incorrect. The presence of pseudosymmetry and/or twinning in the data sets was suspected to make the diffraction pattern resemble an orthorhombic space group due to a fortuitous value of β ≃ 90°. Therefore, the three monoclinic Translationengleiche sub­groups of P222₁ were explored to investigate which twofold or screw operator present in the orthorhombic space group remained valid (if any) as a crystallographic operator in a monoclinic space group.

In the case of MhsT–Val, the merging R factors for the two different settings of space group P2 were markedly increased (Table 2), suggesting that they also were not valid space groups for this data set, and indeed they also resulted in high model R factors in refinement. However, processing in space group P2₁ with the long axis (the c axis in P222₁) now along b merged with proper statistics and an overall R_meas of 0.138. Molecular replacement located two molecules in the asymmetric unit (Matthews coefficient of 2.57 ų Da⁻¹, solvent content of 52.1%), yielding a single solution with PAK = 0, LLG = 5863.578 and TFZ = 53.9. Model refinement proceeded smoothly and converged with an R_work and R_free of 0.207 and 0.237, respectively, with high-quality electron-density maps for unmodeled parts.

Similarly, parallel runs of molecular replacement and initial refinement were performed for MhsT–Leu, yielding initial R_work and R_free values of 0.26 and 0.31, respectively, for P2₁, 0.35 and 0.42, respectively, for P2_b=44.26 Å and 0.31 and 0.37, respectively, for P2_b=50.17 Å. These results clearly indicated that space group P2₁ was again the correct assignment, similar to MhsT–Val. The complete processing and refinement statistics for the two data sets processed in P2₁ can be seen in Table 3.

However, for MhsT–Leu the data-scaling statistics looked comparable in all three monoclinic assignments, i.e. with the two P2 space groups having only a slightly increased R_measure compared with P2₁. This seemingly different behaviour of the MhsT–Val and MhsT–Leu data was investigated further and is explained below.

3.2. Pseudosymmetry of the MhsT–Val and MhsT–Leu crystals

The presence and orientation of crystallographic and non­crystallographic rotational and screw axes in the two data sets were further analysed by use of the Patterson self-rotation function (Rossmann & Blow, 1962). Fig. 2 presents a stereographic projection of the self-rotation function at κ = 180°, indicating 222 symmetry or 222 pseudosymmetry with the NCS axes lying close to the crystallographic axes, which are therefore also candidates as a possible twin axis.

Figure 2 Self-rotation function of (a) MhsT–Val and (b) MhsT–Leu in the κ = 180° section. The low-resolution limit is 7 Å, the high-resolution limit is 3 Å and the radius of integration is 32 Å. The crystallographic twofold screw symmetry axis is present along b, the noncrystallographic twofold symmetry axis and twin axis are present along a and the third twofold axis is present along c.

The P222₁ pseudosymmetry prompted us to investigate the NCS operators with `Find NCS operators' in Phenix. The following operators were found for the two data sets. For MhsT–Leu,

and for MhsT–Val,

Hence, the NCS operator approximates a crystallographic twofold axis along a, which together with the crystallographic twofold axis along b generates a third twofold operator parallel to c and a P222₁ pseudosymmetry. This operator is assumed to be crystallographic when processing is performed in the orthorhombic space group, but because the NCS operator diverges from the crystallographic operator the refinement in the orthorhombic space group stalls at high model R factors, as indicated above.

3.3. Pseudomerohedral twinning of MhsT–Leu

Intensity analyses were performed on the monoclinic data sets in phenix.xtriage (Zwart et al., 2005) as β approaches 90° for both MhsT–Val and MhsT–Leu, making pseudomero­hedral twinning possible. The Wilson ratio and secondary intensity moments (Stanley, 1972; Rees, 1980) of centric and acentric reflections hinted at twinning in both data sets (data not shown) with the h, −k, −l twin law. Additionally, a more robust local intensity difference analysis of the data sets was performed by use of The Merohedral Twinning Server (https://services.mbi.ucla.edu/Twinning/) applying the Padilla–Yeates algorithm (Padilla & Yeates, 2003). The presence of twinning is indicated in both data sets (Fig. 3), but with a markedly higher degree of twinning in the case of MhsT–Leu.

Figure 3 Graphs presenting statistics in the (a) MhsT–Leu and (b) MhsT–Val data sets.

The presence of a twofold NCS operator along a can bias twinning detection to indicate a twin operator along a, because the analysis of the scaling statistics assumes that the related reflections are otherwise independent (Yeates & Rees, 1987; Lebedev et al., 2006), although it is already clear from the NCS operator that they are not. At the same time, cases in which the NCS operator coincides with a crystal axis make twinning highly probable (Lebedev et al., 2006). Generally, twin fractions can be determined in several ways (Britton, 1972; Rees, 1980; Fisher & Sweet, 1980; Murray-Rust, 1973; Yeates, 1997); however, many of these tests do not give an accurate estimate of the twin fraction in cases such as this. Therefore, a maximum-likelihood test that takes the NCS axis into consideration was used in phenix.xtriage (Liebschner et al., 2019). The obtained twin fraction for MhsT–Val was 0.065, while it was 0.443 for MhsT–Leu. In comparison, the Britton test (Britton, 1972) gave twin fractions of 0.173 for MhsT–Val and 0.447 for MhsT–Leu.

A significant decrease (for example 3–10%) in model R factors is to be expected in cases with high degrees of twinning when comparing refinement with and without a twin law. Indeed, for the MhsT–Leu data set structure refinement with the h, −k, −l twin law resulted in a decrease in the refinement R factors from an R_work and R_free of 0.266 and 0.298, respectively, to 0.185 and 0.222. However, in the MhsT–Val complex with a low degree of twinning, R_work and R_free were 0.207 and 0.237, respectively, without the use of the twin law and 0.194 and 0.223, respectively, when the h, −k, −l twin law was used (a refined twin fraction of 0.06); i.e. there was an almost negligible difference. As refinement with the twin law also did not improve the electron-density maps for MhsT–Val, we refined the MhsT–Val structure without the use of the twin law.

The significant difference in the twin fractions observed between the two data sets explains the low merging factors of the MhsT–Leu data set in all three monoclinic subgroups of P222₁. In the case of a data set belonging to P222₁, proper scaling and low merging factors would be expected in its type I maximal non-isomorphous subgroups. Therefore, in the case of two monoclinic data sets containing twin fractions of ∼0 and ∼0.443 it could be expected that the data set with almost perfect twinning would scale better in all three subgroups, as the diffraction pattern approximates an orthorhombic setting more closely than the data set with the lower twin fraction. However, in the case where pseudosymmetry is present at the same time, as here, the merging statistics will also appear valid in the case of P222₁, even though the twin fraction is small or non-existent (Parsons, 2003).

3.4. Crystal packing explains the pseudosymmetry of MhsT–Val and MhsT–Leu

In order to visualize the pseudosymmetry in the crystal structure, a closer analysis of the crystal packing was made. The two molecules in the P2₁ asymmetric unit, molecules A and B (Figs. 4a and 4b), are related by twofold rotational NCS, with a C^α r.m.s.d. of 0.232 Å. Smaller conformational deviations between the molecules are present in the loop regions 248–251 and 419–424. However, a main difference is that the C-terminal end of molecule A can only be traced to Phe448, whereas for molecule B the entire C-terminus ending at Asn453 is visible in the maps (Fig. 4d), with this region being stabilized by local interactions with the neighbouring molecule A (Fig. 4c). In the case of molecule A, however, the distance between the C-terminus and molecule B is larger and no interaction is observed (Fig. 4f). In other words, molecules A and B are not identical, the symmetry operations superimposing them are imperfect and orthorhombic symmetry is not present.

Figure 4 (a) Two molecules in the asymmetric unit of MhsT–Val, also representing MhsT–Leu. (b) Two unit cells are shown. Distances between molecules in the asymmetric unit are shown in blue and distances between molecules in different asymmetric units are shown in red. (c) Interactions between the C-­terminus in molecule B and two residues in molecule A. (d) 2Fo − Fc electron-density map of the additional residues in the C-terminus of molecule B contoured at 1 r.m.s.d. (e) Distances between molecules in the asymmetric unit. (f) Distances between molecules in different asymmetric units.

The variations in crystal packing may be due to subtle ligand-induced conformational changes of the MhsT structure, and we also cannot exclude that the presence of amino-acid ligands at ∼0.5 mM concentration may affect the lipid–detergent phase diagram and therefore the crystallization conditions. As mentioned earlier, the aromatic substrate complexes can crystallize in both P2 and P2₁ forms, and we therefore performed a systematic crystallization approach with controlled protein:lipid ratios for the MhsT–Trp complex. At protein:lipid ratios of 3:0.8(w:w) and 3:1.0(w:w) MhsT–Trp mainly crystallized in the P2₁ form, whereas at a ratio of 3:2.25 it crystallized mainly in the P2 form. These ratios are dependent on both the protein and lipid batch, and cryoprotection procedures also seem to have an effect, but it seems clear that, for example, protein:lipid ratios affect crystal-packing preferences and highlight the importance of exploring and controlling these ratios in crystal screening and optimization.

4.1. Data processing and pseudo-translation of MhsT–Ile

The data sets obtained from the MhsT–Ile crystals were processed with the XDS package (Kabsch, 2010) and the CCP4 suite of programs (Winn et al., 2011) as described for the other complexes, although autoindexing failed in some cases. For most crystals a data set could not be collected due to low resolution or poor quality of the diffraction, with streaky or split reflections. The crystal form was again monoclinic, but it was unclear whether systematic absences were present along b because this direction had been poorly sampled by the data-collection runs before the onset of radiation damage and also due to the anisotropic diffraction properties of the thin plate crystals in the loop. Furthermore, the crystals were generally not isomorphous, but two fairly isomorphous data sets were identified and merged, and despite a low completeness of ∼80% the data were of sufficient quality that we could distinguish space-group assignments and perform structure determination and limited refinement. Important structural features of the complex, especially inside the binding pocket, could be obtained and compared with the other substrate complexes (Focht et al., 2021).

Importantly, analysis of the data set in phenix.xtriage revealed a non-origin peak with a size of 59.6% of the origin peak in the Patterson map at fractional coordinates (0.337, 0.5, −0.312), indicating translational NCS (tNCS).

As we were unable to distinguish P2 and P2₁ in scaling without data along b, molecular replacement was performed in both space groups. Molecular replacement was performed in Phenix Phaser-MR using MhsT–Trp (PDB entry 4us3 without TM5 and ligands) as a search model, proposing a model in space group P2₁ with two molecules in the asymmetric unit (Matthews coefficient 2.52 ų Da⁻¹, solvent content 51.2%). However, the model exhibited negative LLG values and R factors close to 50–60%. Additionally, the model could not explain the presence of the large non-origin peak in the Patterson map. When molecular replacement was extended to P1, searching for four molecules in the unit cell, different relationships between the molecules were revealed. Here, as expected, two pairs of molecules could be distinguished; however, surprisingly, they were related by a twofold NCS axis almost parallel to b. Discovering the possible relationships in the asymmetric unit and guided by the non-origin peak in the Patterson map, models of molecules were created by the use of `Apply NCS operators' in Phenix in space group P2₁, with two molecules related by a NCS twofold axis parallel to b with the translational matrix derived from the non-origin peak in the Patterson map, as only this relationship would explain the coordinates of the peak. A small deviation (only 0.23°) in the orientation of the molecules related by tNCS (Fig. 5b) to­gether with the streaky reflections explained the problems with indexing and data processing of the data sets (Read et al., 2013).

Figure 5 Visualization of the translational noncrystallographic symmetry of MhsT–Ile. (a) Two molecules, A and B, related by NCS. The NCS axis is parallel to b. (b) Molecule A1 has almost the same orientation as molecule B2 and these two molecules are only related by translation. TM1 is coloured in dark grey to more easily visualize the orientation of the molecules.

Fig. 6(a) presents the Harker section at v = 0.5, whereas Fig. 6(b) shows the stereographic projection of the self-rotation function at κ = 180° with the crystallographic twofold screw axis and the noncrystallographic twofold axis both along b.

Figure 6 (a) Harker section at v = 0.5 of the Patterson map for MhsT–Ile visualizing the non-origin peak with a size of 59.6% of the origin peak. The map is drawn with a minimum contour level at 1.0σ with 1.5σ increments. (b) Self-rotation function of MhsT–Ile. The crystallographic twofold screw axis as well as the NCS axis are both present along b.

This solution not only explained the non-origin peak in the Patterson map, but also caused an immediate decrease in the R factors to an R_work and R_free of 0.313 and 0.347, respectively, in the initial round of refinement. The model was further refined in phenix.refine, ending at a final R_work and R_free of 0.277 and 0.305, respectively, which was deemed to be acceptable considering the presence of tNCS, the low completeness of the data and the overall lower resolution and quality of the data set. However, combined with an accurate and overall identical model from other high-resolution structures, a meaningful analysis could be obtained (Focht et al., 2021). Processing and refinement statistics are summarized in Table 3.