Structural model of the SARS coronavirus E channel in LMPG micelles

Abstract

Coronaviruses (CoV) cause common colds in humans, but are also responsible for the recent Severe Acute, and Middle East, respiratory syndromes (SARS and MERS, respectively). A promising approach for prevention are live attenuated vaccines (LAVs), some of which target the envelope (E) protein, which is a small membrane protein that forms ion channels. Unfortunately, detailed structural information is still limited for SARS-CoV E, and non-existent for other CoV E proteins. Herein, we report a structural model of a SARS-CoV E construct in LMPG micelles with, for the first time, unequivocal intermolecular NOEs. The model corresponding to the detergent-embedded region is consistent with previously obtained orientational restraints obtained in lipid bilayers and in vivo escape mutants. The C-terminal domain is mostly α-helical, and extramembrane intermolecular NOEs suggest interactions that may affect the TM channel conformation.

Graphical abstract

Highlights

• •SARS-CoV E protein is a pentameric ion channel.
• •The SARS-CoV E protein (8–65) is almost completely α-helical in LMPG micelles.
• •Ten inter-monomeric NOEs have been identified.
• •The SARS-CoV E protein (8–65) pentameric model has been obtained.
• •Orientation of key residues, e.g. Val25 is consistent with previous in vivo results.

1. Introduction

Coronaviruses (CoV) typically affect the respiratory tract and gut of mammals and birds. Approximately 30% of common colds are caused by two human coronaviruses - OC43 and 229E. Of particular interest are the viruses responsible for the severe acute respiratory syndrome (SARS), which produced a near pandemic in 2003 [1], and the recent Middle East respiratory syndrome coronavirus (MERS-CoV) [2].

No effective licensed treatments exist against coronavirus infections [[3], [4], [5]], but live attenuated vaccines (LAVs) [[6], [7], [8], [9], [10]] and fusion inhibitors [11] are promising strategies. One CoV component critical for pathogenesis is the envelope (E) protein, as reported in several coronaviruses, e.g., MERS and SARS-CoVs [[12], [13], [14]]. The CoV envelope (E) proteins are short polypeptides (76–109 amino acids) with a single α-helical transmembrane (TM) domain [[15], [16], [17], [18], [19], [20], [21]] that form homopentameric ion channels (IC) with poor ion selectivity [22,23]. CoV E proteins are mostly found in the endoplasmic reticulum-Golgi intermediate compartment (ERGIC) [[24], [25], [26], [27], [28], [29]]. In animal models, deletion of SARS-CoV E protein reduced pathogenicity and mortality [30], whereas cellular models displayed up- and down-regulation of stress response and inflammation host genes, respectively [31]. The importance of E protein in pathogenesis has led to the development of LAVs based on deletion of E protein in SARS- and MERS-CoVs, although this led to compensatory mechanisms that recover virulence [32,33].

Specific critical features in the SARS-CoV E protein sequence have been identified that determine virulence, e.g., at the C-terminal tail [34] or in the TM domain [30], and precise structural characterization of these regions could help in the design of E protein-based CoV LAVs. However, detailed structural knowledge is still very limited in the case of SARS-CoV E, and non-existent for other CoV E proteins.

A pentameric model for SARS-CoV E was initially proposed by the authors after an in silico conformational search [15] of TM domain oligomers. In that report, two pentameric models (termed ‘A’ and ‘B’) that were separated by a ~50° rotation of their α-helices were selected. In model A, V25 adopts a more lumenal position, whereas in model B, the position of this residue is clearly interhelical (Fig. 1 ). The pentameric organization of SARS-CoV E has been confirmed experimentally in various detergents: PFO, DPC or C-14 betaine [17,18], not only for synthetic TM (E_TM), but also for an 8–65 (E_TR) construct and for full length E protein (E_FL).

Fig. 1.

Comparison of orientation of residue V25 in SARS-CoV E_TM pentameric models. Orientation of computational models A (orange) and B (cyan) [15], where the side chain of V25 (F26 is only used to guide the eye) is indicated. The ‘A-like’ model obtained by NMR [20] is shown in red. In model B, the position of V25 is clearly interhelical.

To confirm experimentally the orientation of the α-helices in the pentameric model, site specific infrared dichroism (SSID) measurements [35] were obtained in hydrated lipid bilayers, with ¹³C = ¹⁸O isotopically labeled synthetic E_TM. However, the orientation of the α-helices turned out to be strongly dependent on the presence of 2 flanking lysine residues at each end of the peptides [16]: with flanking lysine residues, the orientation was a hybrid between models A and B (residues 17–24 were oriented consistent with model B, but from residue 24 onwards, orientation was as expected for model A), consistent with a ‘bend of the α-helices around residues 25–27’ [16]. Without terminal lysines, however, the orientation of the central five labeled consecutive residues, L21 to V25, was entirely consistent with model A [17].

These initial results suggested that the conformation of the E_TM pentamer may be very sensitive of the presence of extra residues and probably also, extramembrane domains. An NMR study was performed on a synthetic E_TM (residues 8–38) in DPC detergent micelles, where E_TM was selectively labeled [20]. E_TM was ¹⁵N-labeled at A22, V24, V25, and ¹³C, ¹⁵N-labeled at L18, L19 and L21. Intermonomeric NOEs were assigned indirectly, i.e., when cross-peaks could not be explained by intramonomer interactions. Of these, derived from difference 2D homonuclear ¹H^N, ¹H^aromatic band-selected NOESY, only four NOEs were labeled ‘strong’, and involved the ¹H^δε phenyl ring of Phe23, to ¹H₃ ^δ1/¹H₃ ^δ2 of either Leu18 (two NOEs) or Leu21 (two NOEs). These intermolecular NOEs were insufficient to distinguish between models A and B, and the monomer structure was fit to a model A template.

More recently, recombinant SARS-CoV escape mutants were recovered after introducing a V25F channel-inactivating mutation in the E protein, [36], that led to attenuation in a mouse model [30]. Revertant mutants regained fitness and pathogenicity whereas mutated E protein regained channel activity [30]. Surprisingly, escape mutations in E protein clustered along the helix interface opposite to residue V25, consistent with an interhelical orientation of this residue, as found in model B (Fig. 1, cyan).

In the present paper, we report a more accurate model of the SARS-CoV E protein pentamer, in LMPG micelles. The construct we have used prolongs the TM domain with another 27 residues in the C-terminal domain (residues 8–65). Following established protocols [37], two types of monomers were mixed, bearing different isotopical labels, that allowed unambiguous identification of ten intermonomeric NOEs. In a nutshell, the results are consistent with a TM model that appears to be a hybrid between models A and B: while overall being closer to model A, residue V25 has a clear ‘model B-like’ interhelical orientation, consistent with the revertant mutants that appeared in vivo.

2. Materials and methods

2.1. Protein expression and purification

The expression and purification methods for the truncated SARS-CoV E construct corresponding to residues 8–65 (E_TR) have been described previously [19]. This construct does not have cysteines, as these are not required for oligomerization [18,19,28,38]. In the present work, M9 media was supplemented with an appropriate combination of ¹⁵NH₄Cl, ¹³C-glucose, ²H-glucose, and ²H₂O (Cambridge Isotope Laboratories) to produce¹⁵N-, ¹³C-, ¹⁵N/¹³C- and ¹⁵N/²H-labeled E_TR samples. For preparation of fully deuterated ¹⁵N/²H-labeled samples, freshly transformed E. coli cells were doubly-selected in LB agar plates and media prepared with 30% and 60% ²H₂O, successively, and later grown in M9 media prepared with 99.9% ²H₂O [39,40].

2.2. Gel electrophoresis

Blue-native PAGE (BN-PAGE) was performed as described previously [41]. Lyophilized E_TR protein was solubilized (0.1 mM) in sample buffer containing LMPG (lyso-myristoyl phosphatidylglycerol, Anatrace) at the indicated concentrations.

2.3. Residue rotational pitch calculations

For α-helical bundle models, the rotational pitch angle of a residue, ω, defined arbitrarily as 0° or 180° when transition dipole moment, helix director, and the z-axis all reside in a single plane, was calculated as described elsewhere [42]. The final result is the average of the ω values calculated in each monomer. For a canonical α-helix, it is expected that Δω between two consecutive residues is ~100°.

2.4. NMR sample preparation

Lyophilized E_TR protein (0.67 mM) was solubilized in 20 mM sodium phosphate pH 5.5, 50 mM NaCl, and 200 mM LMPG, i.e., a protein:detergent (P/D) molar ratio of 1:300. The same protein concentration and P/D ratio was used for the mixture of ¹⁵N-D and ¹³C-labeled samples. The solution was vortexed and sonicated several times until a clear solution was obtained, indicating protein reconstitution into detergent micelles.

2.5. NMR spectroscopy

NMR experiments were performed at 308 K using an Avance-II 700 NMR spectrometer with cryogenic probes. Sodium 2,2-dimethyl-2-silapentane-5-sulfonate (DSS) was used as the internal reference for ¹H nuclei. The chemical shifts of ¹³C and ¹⁵N nuclei were calculated from the ¹H chemical shifts. The NMR data were processed using TopSpin 3.1 (www.bruker-biospin.com) and analyzed using CARA (www.nmr.ch). Sequence-specific assignment of backbone ¹H^N, ¹⁵N, ¹³C′ and ¹³C^α was achieved by using 2D [¹H-¹⁵N]-TROSY-HSQC, 3D HNCA and HN(CO)CA experiments on a ¹⁵N/¹³C-labeled E_TR protein. Side-chain resonances were assigned using 3D ¹⁵N-resolved NOESY-HSQC (120 ms mixing time), (H)CCH-TOCSY and ¹³C-resolved NOESY-HSQC (120 ms mixing time). To identify membrane-embedded residues, the NMR sample was lyophilized overnight and reconstituted in 99% D₂O. Immediately after reconstitution, 2D [¹H-¹⁵N]-TROSY was collected. The titration experiments with 5-(N,N-hexamethylene) amiloride (HMA) were performed with ¹⁵N-labeled E_TR sample. Chemical shift perturbation (CSP) values and chemical shift differences were calculated using the formula CSP $=\sqrt{△{\mathrm{\delta H}}^2+{\left(0.23\ast △\mathrm{\delta N}\right)}^2}$.

2.6. Structure calculation

Intra-monomeric NOE distance restraints were obtained from ¹⁵N-NOESY-HSQC and ¹³C-NOESY-HSQC spectra (both with a mixing time of 120 ms). Backbone dihedral angle restraints (φ and ψ) were derived from ¹³C′, ¹³C^α, ¹³C^β, ¹H^α and ¹H^β chemical shift values using TALOS+ [43]. Short-range and medium range NOE connectivities were used to establish sequence-specific ¹H NMR assignments and to identify elements of the regular secondary structure. Hydrogen bonds were derived from the NOE connectivity, and supported by the H/D exchange data. Monomer structure calculations were performed using CYANA 3.0 [44,45] and visualized using PyMOL (Delano Scientific). All of the restraints used in the calculations to obtain a total of 10 monomer structures, and all the structure statistics, are summarized in Supplementary Tables S1 and S2.

Inter-monomeric NOE restraints were obtained from 3D ¹⁵N-resolved NOESY-HSQC (250 ms mixing time) of two sets of asymmetrically deuterated samples: (1) ¹⁵N/²H-labeled E_TR sample (ND), and (2) an equimolar mixture of ¹⁵N/²H-labeled and a non-deuterated ¹³C-labeled E_TR sample (ND + C). NOE cross-peaks appearing in sample ND + C but not in sample ND were assigned to inter-monomeric contacts. Conversely, resonances also appearing in the ND sample were attributed to incomplete deuteration, and were assigned to intra-monomeric NOEs.

The pentamer structure was calculated using HADDOCK 2.2 [46] according to standard protocols. Ten inter-monomeric NOE restraints (defined as above) were described as ambiguous and unambiguous 5.0 Å distance restraints. Two segments were described as fully flexible: residues 37–47 and 40–54. A C5 symmetry restraint between all 5 subunits and pairwise non-crystallographic symmetry restraints between neighbouring subunits were applied. Initial rigid-body docking yielded 1000 structures, out of which 200 top-scoring structures (i.e., based on HADDOCK target function score) were selected for refinement by semi-flexible simulated annealing. These were then clustered based on RMSD, and the top-scoring cluster was selected (all 16 structures within the said cluster were grouped to form an ensemble).

3. Results and discussion

3.1. Helical structure and TM domain of SARS-CoV E monomer (E_TR) in LMPG micelles

Despite phospholipid isotropic bicelles may have been more membrane-like than detergent micelles, in our hands, phospholipid bicelles did not produce suitable spectra of E_TR (not shown). Examples of significant differences observed in bicelles vs micelles have been reported, e.g., in the study of the integrin TM heterodimers [[47], [48], [49], [50], [51], [52]] or in viral channels [53].

Nevertheless, we have shown previously that E_TR is pentameric in various detergents [17,18], although none of them was suitable for NMR studies of E_TR or E_FL (not shown). E_TR only produced reasonably good NMR spectra in DPC when SDS was also present [19], but since SDS disrupts E_TR oligomerization, we searched for other micellar environments. Lipid-like LMPG was found to produce good NMR spectra for E_TR, although not for E_FL. Therefore, E_TR in LMPG was used in subsequent experiments. The use of the E_TR construct instead of the full-length E protein (E_FL) is justified since the ¹³Cα chemical shifts of E_TR and E_FL protein in SDS or SDS/DPC were almost identical for residues 8–65 [19]. In addition, the secondary structure, obtained by CD/FTIR [18], of E_TR and E_FL is similar and predominantly α-helical, whether in DPC, SDS, mixed (1:2 M ratio) SDS/DPC micelles or DMPC synthetic membranes [18,19].

Comparison of the HSQC spectrum of E_TR/LMPG before and after exposure to D₂O (Fig. 2A) shows that only 20 residues are protected from hydrogen/deuterium (H/D) exchange. The protected residues correspond to the stretch L18-L37, unequivocally indicating the presence of a single TM domain in SARS-CoV E. This result is consistent with the stretch L18-L39 found to be protected in SDS micelles [19]. The chemical shift index (CSI)-based secondary structure of E_TR (calculated by using TALOS+) obtained in LMPG (Fig. 2B), has significantly higher helicity in C-terminal residues 52–55, when compared with the data obtained SDS or with a mixture SDS/DPC [19].

Fig. 2.

Hydrogen-deuterium exchange protected region and secondary structure of E_TR monomer in LMPG. (A) [¹H-¹⁵N]-TROSY-HSQC spectra in H₂O (left) and 99% D₂O (right), with cross-peaks labeled by one-letter code and residue number; (B) Secondary structure prediction obtained using TALOS+ [43], comparing E_TR in LMPG, SDS, and SDS/DPC [19]. (Layout note: 1 column).

The structure of E_TR was calculated from 10 E_TR monomer structures (Fig. 3A) and the structure statistics are summarized in Supplementary Table S1. The E_TR monomer in LMPG consists of three helical segments: the one encompassing the TM domain (H1, residues 12–37), a juxtamembrane middle helical segment (H2, residues 39–47), and a C-terminal helix (H3, residues 52–65) (Fig. 3B). In contrast, E_TR in DPC/SDS [19] was formed by only two helical segments separated by a long flexible link (Fig. 3C). Compared to the results in SDS or SDS/DPC [19], in LMPG helix H3 is extended by 3 residues on its N-terminal side, whereas a new helical segment, H2, is formed.

Fig. 3.

E_TR consists of three α-helical segments in LMPG. (A) Ensemble of 10 calculated E_TR monomer structures in LMPG showing the backbone as line representation; (B) for clarity, the helical segments shown in (A) are superimposed locally and the side chains are shown as line representation; (C) graphical comparison of α-helical stretches and H/D protection (showing the TM domain) in LMPG obtained herein and in SDS/DPC environments [19]. Structure statistics in LMPG are summarized in Supplementary Table S1. (Layout note: 1.5 columns).

3.2. Oligomeric state of SARS-CoV E in LMPG

To assess the oligomerization of E_TR in LMPG micelles, its migration in a BN-PAGE gel was analyzed at various protein-to-detergent (P/D) ratios (Fig. 4 ). At the lowest P/D molar ratio (1:1000), E_TR migrates as a ladder of increasingly larger oligomers where the fastest migrating band is assumed to correspond to monomers (lower star), ~8 kDa, whereas at a high P/D ratio (1:125), E_TR migrates with an apparent molecular weight of ~150 kDa. These results are almost identical to those obtained previously for MERS-CoV E, for which a pentameric oligomer was determined using analytical ultracentrifugation in C-14 betaine. In that case, migration in BN-PAGE gels was also observed as a single ~150 kDa band in detergents DPC, DHPC and LMPG [21], and the ladder observed at higher detergent concentration conveniently provided an internal reference that served as a oligomeric size marker. Similar to E_TR, by comparison with that ladder, we confidently assigned the single band observed for MERS-CoV E to pentameric oligomers. It should be noted that in BN-PAGE gels of membrane proteins, molecular weights can appear up to 80% higher due to a contribution of the dye [54]. We have shown this for tetrameric AQPZ, which migrated at ~170 kDa instead of the expected ~100 kDa, and with a viroporin, the SH protein pentamer [41], which migrated as ~66 kDa instead of ~40 kDa. In the case of envelope E proteins, the effect is even more pronounced. In both SARS-CoV E_TR and MERS-CoV E, the pentameric form appears at ~150 kDa, therefore the monomer should appear at >30 kDa. This is consistent with its migration above the AQPZ monomer (~25 kDa). The ladder ends with a pentamer, which is the predominant band at high P/D ratios. The proportion of large oligomers naturally decrease at low P/D ratios, but a significant amount of pentamer species is still present even at the 1:1000 P/D ratio. The NMR data was collected at a P/D molar ratio of 1:300, which should mostly be formed by pentamers.

Fig. 4.

Oligomeric state of SARS-CoV E in LMPG. BN-PAGE of E_TR in lipid-like LMPG detergent (peptide-to-detergent ratio is indicated). A ladder of oligomeric sizes is indicated by stars (*). The membrane protein aquaporin Z from E. coli (AqpZ) is used as reference, in monomeric and tetrameric forms (AqpZ:1 and AQPZ:4, respectively). (Layout note: 1 column).

3.3. HMA binding to oligomeric E_TR

In a previous paper, we showed that monomeric E_TR in SDS micelles was not affected by addition of the drug HMA [19]. However, after addition of DPC to SDS, to a SDS/DPC 1:4 M ratio, HMA induced clear chemical shift perturbations (CSPs), concomitant with E_TR oligomerization. The oligomerization in DPC/SDS was not homogeneous, which precluded a more detailed study, whereas in LMPG a predominant oligomeric size is observed at a high protein-detergent ratio (Fig. 4). Therefore, in LMPG the changes observed after HMA addition should more reliably represent the binding of HMA to E_TR. HMA-induced CSPs were detected herein after addition of 7.75 mM HMA to 0.25 mM E_TR in 200 mM LMPG micelles (P/D molar ratio 1:800) (Fig. 5 ).

Fig. 5.

E_TR oligomer chemical shifts perturbation (CSP) by HMA. (A) Superposition of TROSY-HSQC spectra of uniformly ¹⁵N-labeled E_TR protein (0.25 mM monomer concentration) in the absence (red) and presence (blue) of 7.75 mM HMA. Peaks that undergo significant shifts upon HMA addition are highlighted; (B) selected regions in the TROSY-HSQC spectrum at varying HMA concentration: 0 (red), 0.25 (pink), 0.75 (purple), 1.75 (yellow), 3.75 (light green), 7.75 (green), 9.75 (light blue), 11.75 mM HMA (blue); (C) chemical shift perturbation (CSP) of the backbone amide resonances of 0.25 mM ¹⁵N-labeled E_TR protein upon titration with 7.75 mM HMA. Mean CSP value across all residues and the standard deviation are shown by dashed and dotted line, respectively. Missing/overlapping residues are omitted. (Layout note: 1 column).

The average CSP value was 0.019 ppm, and several residues showed CSP >1 S.D. from the average value, notably Thr-9, Leu-12, Ile-13, Ala-36 and Val-47. These results suggest the presence of two binding sites located at both ends of the TM domain. Given the long distance between Ala-36 and Val-47, the two HMA-interacting residues may be located in different monomers.

3.4. Pentameric model of E_TR

A pentameric model was obtained by docking the monomeric form of E_TR using HADDOCK 2.2 [46], which incorporated 10 inter-monomeric NOE restraints (Fig. 6A). We note that 2 inter-monomeric NOEs are located at the extramembrane C-terminal tail: L39 HN - Y57 HB and V47 HN - N64 HN. The same figure shows a representative example of NOE E_TR inter-monomer connectivity (Fig. 6B). The remaining plots of inter-monomeric NOEs are shown in Fig. S1. Structure statistics are summarized in Supplementary Table S2.

Fig. 6.

Inter-monomeric NOEs in E_TR pentamer. (A) List of inter-monomeric NOE contacts, with those located in the extramembrane C-terminal region in bold; (B) a representative example of NOE E_TR inter-monomer connectivity (green lines). Selected strips correspond to a ¹⁵N–NOESY-HSQC spectrum and NH protons of V14 for samples ¹⁵N/²H-labeled (ND), ¹⁵N/²H-labeled + ¹³C-labeled (ND + C), and ¹⁵N/¹³C-labeled (NC). The NOE strips from the NC sample are shown as reference, as they contain both intra and inter-monomer contacts. Strips corresponding to the remaining NOE connectivity are shown in Fig. S1. (Layout note: 1 column).

The E_TR pentamer is a right handed α-helical bundle where the C-terminal tails coil around each other (Fig. 7A) likely owing to the 2 inter-monomeric restraints between the two C-terminal helices. Each pentamer subunit (Fig. 7B) has better defined structure compared to the monomer alone (Fig. 3A). This is mainly due to decreased flexibility at the inter-helical segments, which were kept flexible during the docking, as the two C-terminal helices now adopt a relatively fixed conformation. This is also apparent from the RMSD values; the pentamer subunit RMSD values are significantly reduced as compared to the monomer (Fig. 7C).

Fig. 7.

Structure of the E_TR pentamer. (A) Top view of the E_TR pentamer showing an ensemble of 16 structures obtained using HADDOCK and 10 inter-monomeric NOE restraints (see Materials and Methods); (B) Side view of one subunit of the pentamer showing the backbone as line representation; (C) RMSD values (per-residue) of the monomer ensemble (see Fig. 3A, black), structured helical segments of the monomer (see Fig. 3B, blue), and the pentamer ensemble (Fig. 7A, red). The average RMSD value of the monomer (dashed line) and ±1 S.D. values (grey band) are indicated. (Layout note: 1.5 column).

Notably, in this pentameric model, the location of V25 is interhelical (Fig. 8B–C), whereas in the previously proposed model it was closer to a lumenal orientation [16]. The rotational pitch of the residues in the TM domain of this pentameric model were measured individually [35] and compared to those from the computational models A and B [15] (Fig. 8D). While values for residues 25–27 are closer to model B, the rest of the sequence is similar to model A, except at residue 28 which deviates from both models. For comparison, the rotational pitch close to model A for residues in E_TM obtained previously by NMR in DPC micelles [20] is also shown. The present model has been constructed independently from A and B model templates, and the result appears to be a hybrid between the two [15]. This is not surprising since the in silico study assumed a certain rigidity in the TM α-helices [15]. Most of the residues in the model we report have an orientation consistent with model A. This is not surprising, since model A had the lowest energy value for each individual E protein homologs [15]. However, the model gets closer to model B in the turn that contains V25 (Fig. 8D). This enables V25 to adopt a more interhelical orientation consistent with the revertant mutants that appeared in vivo [30]. Additionally, the helix kink region suggested by infrared dichroism data in lipid bilayers [16] is also observed, which supports the validity of the membrane-mimic environment used herein.

Fig. 8.

Orientation of the E_TR pentamer. (A) Top view and (B) side view of average structure of the E_TR pentamer bundle in cartoon representation. The N- and C-terminus of one monomeric unit is indicated; (C) top view of a monomer-monomer TM interaction, showing the distances between the side chain of V25 and those of residues appearing in SARS-CoV E V25F revertant mutants [30]; (D) differences in TM residue rotational orientation, ω, between the experimental model proposed here (LMPG) versus computational models A and B [15] and that of E_TM obtained by NMR in DPC micelles [20]. The region with larger differences between the present model (LMPG) and model A (residues 25–28) is highlighted. (Layout note: 1.5 columns).

Finally, in LMPG micelles, the C-terminal tail of SARS-CoV E protein is α-helical, more so than observed in mixed DPC/SDS micelles [19], and the presence of extramembrane NOEs suggest interactions between the C-terminal domains that may affect the pentameric conformation.

Accession numbers

The atomic coordinates have been deposited in the Protein Data Bank (PDB ID: 5X29). Assigned chemical shifts have been deposited at the Biological Magnetic Resonance Bank (BMRB ID: 36049).

Transparency document

Transparency document.

Appendix A. Supplementary data

Supplementary material