Structural determination of VSG11

Two constructs of VSG11 NTD were used in the structural studies in this manuscript (all VSGs, unless otherwise noted, are from the strain Lister 427). The first was the wild type sequence (VSG11_WT) that (1) crystallized as a monomer in the asymmetric unit (with a crystallographic trimer) in sodium-potassium tartrate with two structures determined: VSG11_WT-Iodine, a crystal soaked in Na/K iodine diffracting to 1.27Å, and VSG11_WT-Oil, a crystal diffracting to 1.23 Å resolution cryo-cooled in oil, and (2) crystallized as two monomers in the asymmetric unit (not a dimer and with a crystallographic trimer–see Results below) in ammonium sulfate (diffracting to 1.75 Å, denoted VSG11_WT-AS). The second construct of VSG11 was a chimeric form consisting of the VSG11_WT NTD connected to the VSG2 CTD, denoted VSG11_N2C. This construct crystallized with 18 monomers of VSG11_N2C in an asymmetric unit comprised of six trimeric assemblies (denoted VSG11_N2C-18mer). VSG11_WT-18mer crystals were also obtained that diffracted to nearly 3Å resolution, but the best diffracting crystals were produced by the VSG11_N2C construct (2.6 Å resolution). Therefore, the presence of the VSG2 CTD was not required for the formation of the 18mer form, and since in both cases the CTDs were not present in the crystallized form, it is unlikely that the 18mer form is tied in any manner to the chimeric form.

Plasmids used to generate VSG11_WT and VSG11_N2C are described in [17]. Plasmids were first linearized by EcoRV (New England Biolabs), and then transfected into VSG2-expressing cells (2T1): 10ug of each plasmid were mixed with 100ul of cells (at a concentration of between 2.5x10⁷ and 3x10⁷) in Tb-BSF buffer (90mM Na₂HPO₄, pH 7.3, 5mM KCl, 0.15mM CaCl₂, 50mM HEPES, pH7.3), using an AMAXA nucleofector (Lonza) program X-001, as previously described [29]. Blasticidin at a concentration of 100ug/ml was added after 6h and single-cell clones were obtained by serial dilutions in 24-well plates and collected after 5 days.

Clones initially screened with flow cytometry for VSG2 loss of expression using a monoclonal VSG2_WT antibody [30] and VSG11 gain of expression with anti-VSG11 antisera. To determine the binding of antisera to live trypanosomes, 1 x 10⁶ parasites were collected and incubated with VSG2_WT antisera (1:4000) or VSG11_WT (1:1000) together with Fc block (1:200, BD Pharmingen) in cold HMI-9 without FBS for 10 min at 4°C. Cells were washed once with cold HMI-9 and resuspended in 200μl cold HMI-9 with rat anti-mouse IgM-FITC (1:500, Biolegend). After one wash with cold HMI-9, cells were resuspended in 150μl HMI-9 and immediately analyzed with FACSCalibur (BD Bioscences) and FlowJo software (v10). For a second step, clones were sequenced by isolating RNA using the RNeasy Mini Kit (Qiagen), followed by DNAse treatment with the TURBO DNA-free kit (Invitrogen) and cDNA synthesis with ProtoScript II First Strand cDNA Synthesis (New England Biolabs). The sequences were then amplified, using Phusion High-Fidelity DNA Polymerase (New England Biolabs), a forward primer binding to the spliced leader sequence and a reverse binding to the VSG 3´untranslated region. The final products were purified by gel extraction from a 1% gel with the NucleoSpin Gel and PCR clean-up kit (Macherey-Nagel) and sent for Sanger sequencing.

All VSG11 constructs were expressed in T. b. brucei cultured at 37°C and 5% CO₂ in HMI-9 media (PAN Biotech) supplemented with 10% fetal calf serum (Gibco), L-cysteine and ß-mercaptoethanol. The cells from 3.6 liter culture were pelleted and the VSG11 proteins purified through modifications of previously published protocols [31]. The cells were lysed with 40 ml 0.4 mM ZnCl₂, after centrifugation (10.000 g, 10 min) the pellet was resuspended in 30 ml 20 mM Hepes/NaOH pH = 8.0, 150 mM NaCl (42°C) and centrifuged (10.000 g, 10 min). The supernatant was passed through a 25 ml Q-sepharose Fast-flow column (GE Healthcare) equilibrated with 20 mM Hepes/NaOH pH = 8.0, 150 mM NaCl. The flow through was collected and concentrated. The protein was further purified on a HiLoad 16/600 Superdex 200 pg column (GE Healthcare) equilibrated in 10 mM HEPES/NaOH pH = 8.0, 150 mM NaCl. The fractions containing the VSG protein were pooled and concentrated.

VSG11_WT crystals containing only the N-terminal domain appeared after several days at 22°C using the hanging drop method with 6 mg/ml VSG11 (full-length) protein in a 1:1 volume ratio against 100 mM Tris/HCl pH = 7.5 1.6–1.75M sodium-potassium tartrate. CryoOil (MiTeGen) was used as cryoprotectant and the VSG11_WT-Oil crystals flash-cooled in liquid nitrogen. The loss of the CTD is presumed to have occurred during the crystallization stage. VSG11_WT-Iodine crystals soaked in 200 mM KI, 100 mM Tris/HCl pH 8.0 and 1.7 M NaKTartrate were flash-cooled directly in liquid nitrogen.

VSG11_WT-AS crystals were grown at 22°C by vapor diffusion using hanging drops with a 1:1 volume ratio of 6mg/ml protein to an equilibration buffer consisting of 0.1M sodium acetate pH 4.5 and 2M ammonium sulfate. For cryoprotection the crystals were transferred to the same buffer as that used for equilibration but supplemented with 25% v/v glycerol and were flash-cooled in liquid nitrogen.

VSG11_N2C-18mer crystals were grown at 22°C by vapor diffusion using hanging drops with a 1:1 volume ratio of 6 mg/ml protein to equilibration buffer containing 19% (w/v) PEG 2000MME, 0.2 M NaCL 0.1 M MES pH 6.0. For cryoprotection the crystals were transferred to the same buffer as used for the equilibration with 25% (v/v) glycerol and were flash-cooled in liguid nitrogen.

Native VSG11_WT datasets were collected at a wavelength of 1.0 Å at the Paul Scherrer Institut Villingen. For phasing, an iodine soaked crystal was collected at 1.54 Å on a Rigaku X-ray generator and DECTRIS PILATUS3 R detector. The structure was solved by single wavelength anomalous diffraction (SAD) using SHELX [32] and HKL3000 suite [33]. The initial model was built using Arp/wARP [34] with PHENIX [35], COOT [36] and PDB_REDO [37] for model optimization and refinement. That model was placed in a high-resolution VSG11_WT-Iodine dataset (collected at 1Å at the SLS synchrotron–see S1 Table) by molecular replacement with PHASER [38] and the model optimized and refined with PHENIX-REFINE [35] and with cycles of manual model building using COOT [36]. The structures of VSG11_WT-Oil, VSG11_WT-AS, and VSG11_N2C-18mer were solved by molecular replacement using the refined VSG11_WT-Iodine model with the PHASER package [38] of PHENIX, and the models optimized and refined with PHENIX-REFINE [35] and with cycles of manual model building using COOT [36]. The structure of VSG11_N2C-18mer is characterized by a high Wilson B factor (66.92Ų) and disorder in many regions of the model. Final statistics are shown in S1 Table.

Structural determination of VSG615

T. brucei brucei expressing VSG615 was obtained as a kind gift of Dr. Hee-Sook Kim [39] and were cultured at 37°C and 5% CO₂ in HMI-9 media (PAN Biotech) supplemented with 10% fetal calf serum (Gibco), L-cysteine and ß-mercaptoethanol. The cells from 4L culture were pelleted and the VSG615 protein purified through modifications of previously published protocols [31]. The cells were lysed with 40 ml 0.2 mM ZnCl₂, after centrifugation (10.000 g, 10 min) the pellet was resuspended in 30 ml 20 mM HEPES/NaOH pH = 7.5, 150 mM NaCl (42°C) and centrifuged (10.000 g, 10 min). The supernatant was passed through a 25 ml Q-sepharose Fast-flow column (GE Healthcare) equilibrated with 20 mM HEPES/NaOH pH = 7.5, 150 mM NaCl. The flow-through was collected and concentrated to final concentration 1 mg/ml.

To generate the NTD of VSG615, the concentrated protein was subjected to limited proteolytic digestion using trypsin. The VSG at the concentration of 1 mg/ml was mixed with trypsin (5 mg/ml) (Sigma Aldrich) at 1:50 trypsin:VSG ratio and incubated for 3 hours on ice. The reaction was terminated by adding PMSF to 1 mM final concentration. The protein was further purified (separating the NTD from the CTD) by size exclusion chromatography on a HiLoad 16/600 Superdex 200 pg column (GE Healthcare) equilibrated in 20 mM HEPES/NaOH pH = 7.5, 150 mM NaCl. The fractions containing the NTD of VSG615 from size exclusion chromatography were concentrated to 10 mg/ml in 500 μl of final volume. The lysine residues on the protein were subsequently methylated by reductive alkylation [40]. 10 μl of 1M borane dimethylamine complex (DMAB) and 20 μl of 1M formamide into the protein solution and mixed gently. The mixture was incubated for 2 hours in the dark at 4°C with rotation and the entire process repeated. Prior to overnight incubation, 5 μl of 1M DMAB was added. To stop the reaction, 1M Tris pH 7.5 was added to bring the reaction to a final volume of 1 ml. The buffer was exchanged to 20 mM HEPES pH 7.5, 150 mM NaCl by size exclusion chromatography with Superdex 200 10/300 GL column. The fractions containing the methylated protein was further concentrated to 10 mg/ml for crystallization.

The methylated VSG615 protein was crystallized at 22°C by vapour diffusion using hanging drops formed from mixing a 1:1 volume ratio of the protein with an equilibration buffer consisting of 23% (w/v) PEG 4000, 100 mM sodium cacodylate pH = 6.0, 10 mM ZnCl₂. For data collection, crystals were soaked in the same buffer augmented to 20% glycerol, flash-cooled in liquid nitrogen. A native dataset was collected at the Paul Scherrer Institut, Villingen. The structure was solved with molecular replacement using a model of a VSG615 trimer predicted by AlphaFold using the PHENIX package (loop regions with a pLDDT below 50 were removed for the search model). The structure is characterized by a very high Wilson B factor (85.19Ų), high atomic temperature factors, and correspondingly high disorder in many regions of the model, although the electron density for the O-linked sugars (a principle reason for pursuing the VSG615 structure) was clear. Final model statistics are shown in S1 Table.

Structural determination of VSG558

VSG558 was amplified from genomic DNA of T. brucei brucei strain Lister 427 VSG2 expressing cells. The plasmid used to generate VSG558-expressing trypanosomes was a modification of previous plasmids designed for integration into the trypanosome genome [30]. The plasmid was first linearized by EcoRV (New England Biolabs), and then transfected into VSG2-expressing cells (2T1): 10ug of plasmid was mixed with 100ul of 4x10⁷ cells in Tb-BSF buffer (90mM Na₂HPO₄, pH 7.3, 5mM KCl, 0.15mM CaCl₂, 50mM HEPES, pH7.3), using an AMAXA nucleofector (Lonza) program X-001, as previously described [29]. After 6h hygromycin was added to a concentration of 25 ug/ml and single-cell clones were obtained by serial dilutions in 24-well plates and collected after 5 days.

Clones were initially screened with flow cytometry for VSG2 loss of expression using a monoclonal VSG2_WT antibody [30]. To determine the binding of antisera to live trypanosomes, 2 x 10⁶ parasites were collected and incubated with 200ul FITC conjugated VSG2_WT antisera (FITC conjugation kit, Abcam ab102884) (1:200) in cold HMI-9 without FBS for 10 min on ice. Cells were washed twice with cold HMI-9, resuspended in 200μl cold HMI-9 and immediately analyzed with a Guava EasyCyte 4HT Flow Cytometer (Luminex). For a second step, negative clones were sequenced by isolating RNA using the RNeasy Mini Kit with on-column DNAse digestion (Qiagen) and cDNA synthesis with Superscript IV first strand synthesis system (Thermo Fisher). The sequences were then amplified, using Phusion High-Fidelity DNA Polymerase (New England Biolabs), a forward primer binding to the spliced leader sequence and a reverse binding to the VSG 3´ untranslated region. The final products were purified by gel extraction from a 1% gel with the NucleoSpin Gel and PCR clean-up kit (Macherey-Nagel) and verified by Sanger sequencing.

VSG558 was expressed and purified from trypanosomes in the same manner as VSG11. Purified full-length VSG558 was mixed with trypsin at 1:50 trypsin:VSG ratio and incubated 1 hour on ice and the NTD domain was isolated by size exclusion chromatography on a Superdex 200 10/300 GL column equilibrated in 10 mM HEPES/NaOH pH = 8.0, 150 mM NaCl. After purification the VSG558 NTD was concentrated to 5 mg/ml and crystallized at 22°C by vapour diffusion using hanging drops formed from mixing a 1:1 volume ratio of the protein with an equilibration buffer consisting of 100 mM Citric acid/NAOH pH = 5.5 and 17.5% PEG 3350. The crystals appeared after one week and the same condition supplemented with 25% PEG 400 or 25% MPD was used as a cryoprotectant for crystals flash-cooled in liquid nitrogen. The crystals diffracted to 1.74 Å and were collected at a wavelength of 1.0 Å at the Paul Scherrer Institut Villingen. The structure was solved by molecular replacement using a model of a dimer predicted with Alphafold [27], followed by model optimization and refinement using PHENIX and COOT. Final model statistics are shown in S1 Table.

Structural determination of VSG21

A VSG21 stably-expressing T.b. brucei strain was obtained as a kind gift of Dr. Hee-Sook Kim [39]. VSG21 was expressed and purified from trypanosomes according to the same protocol as VSG11. Full-length VSG21 at the concentration of 12.3 mg/ml was crystallized at 22°C by vapour diffusion using sitting drops formed from mixing a 1:1 volume ratio of the protein with an equilibration buffer consisting of 100 mM Tris/HCl pH = 7.0, 200 mM CaCl2 and 20% (W/V) PEG 3335. For data collection, crystals were soaked in 100 mM Tris/HCl pH = 7.0, 25% (V/V) ethylene glycol and 20% (W/V) PEG 3335 and flash-cooled in liquid nitrogen. A native dataset was collected at the Paul Scherrer Institut, Villingen. The structure was solved with molecular replacement using a model of a VSG21 dimer predicted by AlphaFold using the PHENIX package, followed by model optimization and refinement using PHENIX and COOT. Final model statistics are shown in S1 Table. VSG21 was found to be truncated, with the far C-terminal sequence corresponding to the lower lobe removed.

Structural determination of VSG545

VSG545 cloning and verification were performed in the same manner as VSG558 (except that for selection a hygromycin concentration of 5μg/ml was used). For purification of VSG545, cells from 2.4 L culture were pelleted and lysed with 20 ml 0.4 mM ZnCl2, containing a protease inhibitor cocktail (Roche cOmplete), after centrifugation (10.000 g, 10 min) the pellet was resuspended in 20 ml 10 mM Sodium Phosphate buffer, pH8 containing protease inhibitors (42°C) and centrifuged (10.000 g, 10 min). The supernatant was passed through a 20 ml Q-sepharose Fast-flow column (Cytiva) equilibrated with 10 mM Sodium Phosphate buffer, pH8. The flow through was collected and concentrated. To remove protease inhibitors, the protein was purified on a Superdex 200 Increase 10/300 GL column (GE Healthcare) equilibrated in 10 mM HEPES/NaOH pH = 8.0, 150 mM NaCl. The fractions containing the VSG protein were pooled and concentrated.

To generate N-terminal domain, the concentrated protein was subjected to limited proteolytic digestion. VSG545 at the concentration of 2 mg/ml was mixed with endoproteinase LysC (New England Biolabs) at 1:1200 LysC:VSG ratio and incubated for 1 hour at 37°C. The reaction was terminated by adding TLCK to 50μg/ml final concentration. The protein was further purified (separating the NTD from the CTD) by size exclusion chromatography.

In order to obtain well-diffracting crystals, N-linked oligosaccharides were removed by treatment of the purified NTD with PNGase F. VSG545 NTD was mixed with PNGase F (New England Biolabs, P0704S) to final concentrations of 0.4 mg/ml and 15,000 U/ml respectively, incubated for 3 hours at 37°C and purified again by size exclusion chromatography.

The protein was concentrated to 2 mg/ml and was crystallized at 22°C by vapour diffusion using sitting drops formed from mixing a 1:1 volume ratio of the protein with an equilibration buffer consisting of 100 mM MES pH = 6.5, 22% PEG 8000 and 300mM Li₂SO₄. Crystals appeared after about 8 weeks and the same condition supplemented with 25% PEG 400 was used as a cryoprotectant for crystals flash-cooled in liquid nitrogen.

The VSG545 data set was highly anisotropic and the Staraniso server [41] was used to determine the anisotropic diffraction cutoff and the output, modified data file was used to solve the structure. A dataset was collected at a wavelength of 1.0 Å at the Paul Scherrer Institut Villingen. The structure was solved by molecular replacement using a model of a dimer predicted with Alphafold [27]. After model building and refinement using PHENIX and COOT, the final model contains a truncated dimer in the asymmetric unit likely produced by further proteolysis in the crystallization drop. Final model statistics are shown in S1 Table. Like with VSG21, VSG545 was found to be truncated, with the far C-terminal sequence corresponding to the lower lobe removed.

Protein structural prediction with AlphaFold (ColabFold)

The Trypanosoma Brucei Brucei Lister 427 VSG sequences were obtained from a publically accessible database at the Rockefeller University (https://tryps.rockefeller.edu/). Structural prediction was performed only for the N-terminal domain sequences (spanning approximately 350 amino-acids, depending on the VSG). As the signal peptide is not present in the mature VSG protein, the signal sequence predicted by SignalP 6.0 (https://services.healthtech.dtu.dk/services/SignalP-6.0/) was removed. The end of the helix that connects the N-terminal bottom lobe with C-terminal domain terminates the N-terminal domain.

For the structure predictions, LocalColabFold [42] version 1.3.0 was installed on a local computer (using https://github.com/YoshitakaMo/localcolabfold). The LocalColabFold command line interface was used with arguments to specify an input FASTA file, an output directory, using the default options for structure predictions. For monomer prediction “colabfold_batch—amber—templates—use-gpu-relax—num-recycle 3 input.fasta output_directory” was used. For the oligomer prediction the input FASTA file contained the VSG NTD sequence 2 x for class A and 3 x for class B separated by a colon, and adding—model-type AlphaFold2-multimer-v2 to the colabfold_batch command line.

The multiple sequence alignment (MSA) was generated for each sequence by the MMseq2 web server implemented in Colabfold, using the Uniref, PDB70 and Environmental sequence database [42]. The server aligns the input sequence against the database and prepares the input files for the structure prediction that ran locally on our Nvidia P-4000 GPU. Improvement of the structural models was achieved through recycling three times (the software default). In the final step in Colabfold a relaxation/energy minimization is performed using AMBER (Assisted Model Building with Energy Refinement) [42]. At the end, Colabfold created 5 predictions for each sequence, ranked by pLDDT for a monomer and TM-score for the oligomers.

Model quality was assessed quantitively using calculations of root-mean-square-deviation (RMSD) and global distance test total score (GDT_TS) [43] as calculated by the AS2TS system (http://as2ts.proteinmodel.org/) [44]. Parameters for the structural alignments and scoring were taken from default suggestions to match those used for the scoring in CASP (https://proteopedia.org/wiki/index.php/Calculating_GDT_TS), where for the comparison only one chain of an oligomer was used.

Additional experimental structures to map VSG diversity

Three novel VSG structures published between 2018 and 2021—VSG3, VSG13, and VSGsur [17,18]—have demonstrated that there is far more diversity in the VSG coat proteins in fold, oligomerization, topology, and post-translational modification than was appreciated. To further map the diversity space of the VSGnome, we have undertaken X-ray crystallographic studies of numerous VSGs. Guided by recent bioinformatic analyses [25,26], we sought to examine VSGs from different clustering classes. In a separate paper [16], we determined structures of three metacyclic VSGs: VSG397, VSG531, and VSG1954, representing three clustering classes. In this paper we present five new experimental structures from T. brucei brucei strain Lister 427: VSG11, VSG21, VSG545, VSG558, and VSG615.

VSG11 and VSG615 are related to VSG3 (the topologically-similar class of monomer/trimers with O-linked glycan PTMs), denoted as class B [25] or N4 [26] in the literature. These were examined to confirm the fold and oligomerization of this class as well as the presence of O-linked glycosylation. Previously determined structures of VSG3 [17] and VSG1954 [16] showed that both possess highly similar trimers in the crystals. Further confirmation of the trimeric oligomerization state of members of this class came from a biochemical analysis of several VSGs that showed evidence of concentration dependent monomeric and trimeric forms of VSG9 in solution [24]. However, unlike VSG3, the structure of VSG1954 showed no evidence of O-linked glycans, despite possessing a conserved glycosylation sequence [16,17]. Whether this absence is due to unknown regulatory processes yet to be identified, to the sugar being labile and removed during purification/crystallization as has been observed for VSG3 [45], or to a difference between metacyclic and bloodstream VSGs, was unclear. Therefore, pursuing additional studies of related VSGs could aid in determining how many VSGs are so modified.

NTD structures of VSG11 and VSG615 were solved to 1.23Å and 3.2Å resolution, respectively (Methods, Figs 1A and 1B and S1 and S1 Table). Both reveal a fold similar to VSG3 (Fig 1C) as well as the presence of O-linked glycans on the top surface of the protein–one observed on VSG11 and two on VSG615 (Fig 1A and 1B). This crystal form of VSG11 (VSG11_WT-Oil, see Methods and S1 Table) crystallizes as a monomer in the asymmetric unit of the crystal, but forms a trimer along a crystallographic three-fold axis of symmetry that is nearly identical to the crystallographic trimers seen with VSG3 and VSG1954 (Fig 2A). Interestingly, VSG615 crystallized with two trimers in the asymmetric unit, and these trimers align well with those of VSG3, VSG11, and VSG1954 (Fig 2A), providing additional support that class B trimer formation is not a crystallization artifact, but could represent the relevant, biological assembly.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Crystal Structures of Five VSGs from Trypanosome brucei.

(A) and (B) Structures of the VSG11 and VSG615 NTD monomers shown as a ribbon diagram colored light blue. The N-linked glycans attached to the bottom lobe are shown in a space-filling representation colored red. Disulfide bonds (labeled S-S bonds) are shown in a space-filling representation, colored cyan. The O-linked glycans on the top of the molecules are shown in a space-filling representation colored purple and green (for the second sugar in VSG615). (C) Alignment of four B-class VSG monomers (VSG3, VSG11, VSG615, and VSG1954 [16]) shown as thin “worm” drawings with each separate protein colored as indicated in the figure labels. The N- and O-linked sugars at the bottom and top of the molecules (respectively) are differentially colored as noted in the figure. (D) Structure of the VSG558 dimer illustrated as in panel (A), which each chain of the dimer colored differently. (E and F) Structures of the VSG21 and VSG545 dimers illustrated as in panel (A), the disordered regions at the C-terminal portion (e.g., missing bottom lobe) denoted by dotted lines. Illustrations were produced with CCP4mg [57].

https://doi.org/10.1371/journal.pntd.0011621.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Class-B VSG Trimers.

(A) Both crystallographic (VSG3 and VSG1954) and non-crystallographic (VSG11 and VSG615) trimeric arrangements aligned and illustrated as ribbon (gray, VSG3) or worm tracing of the mainchain (other VSGs). Two orientations are shown rotated 90 degrees. The three-fold axis of symmetry is shown in red. “C” indicated a crystallography three-fold symmetry for the trimer and “NC” indicates a non-crystallographic three-fold symmetric trimer in the asymmetric unit. (B) The full asymmetric unit of the VSG_WT-18mer is shown on the left with the proteins drawn as ribbon diagrams and each set of trimers colored differently. On the right is a single trimer isolated from the other six. Illustrations produced using CCP4mg.

https://doi.org/10.1371/journal.pntd.0011621.g002

An additional crystal form of VSG11 shows non-crystallographic trimerization in the asymmetric unit, while another form also raises the possibility of an inherent flexibility that could be present in the VSG proteins. In a second crystal form (VSG11_N2C-18mer, determined to 2.6 Å resolution, Methods and S1 Table), the asymmetric unit contains six trimers of the same arrangement as observed previously for all class B VSGs, for a total of 18 independent molecules (hence “18mer”, Fig 2B). In addition to helping establish that this specific trimeric arrangement is the preferred oligomerization of this class (at least at higher concentrations), the independent monomers in this asymmetric unit display a range of alternative conformations, particularly in the bottom lobe of the protein and even in some limited regions of the three-helix bundle (Fig 3A). An overall structural alignment of the monomers to themselves gives an average RMSD of 1.33 Å (RaptorX), a somewhat high value for aligning a protein to itself. Examination of the set of alignments from the monomers of the 18mer show that it appears that the divergence primarily occurs in the lower lobe (Fig 3A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Lower Lobe Conformational Variability Observed in VSG11 Structures.

(A) Alignment of the monomers in the asymmetric unit of VSG_WT-18mer, each monomer given a separate color, shown in two orientations related by a 90-degree rotation about the axis of the helical bundle. (B) Alignment of the two monomers in the asymmetric unit of the VSG11_WT-AS crystals. Two views are shown rotated by 90 degrees and the rotation of the bottom lobes relative to each other is indicated. Illustrations produced using CCP4mg.

https://doi.org/10.1371/journal.pntd.0011621.g003

The third crystal form (VSG11_WT-AS grown in ammonium sulfate and solved to 1.75Å resolution, Methods, S1 Table) contains two molecules in the asymmetric unit (S2A Fig). However, these two monomers are not arranged in a standard class A type dimeric arrangement in which the three-helix bundles, top, and bottom lobes align alongside each other, but are instead “tail-to-tail” with the lower lobe of one monomer interacting with the lower lobe of the other (S2A Fig). In contrast, the crystal packing is characterized by a three-fold trimer identical to those seen in all the class B protein structures solved to date, including the alternative crystal forms of VSG11 (S2B Fig). Therefore, the assembly of VSG11_WT-AS is most accurately described as trimeric and not dimeric (and certainly not dimers in the sense of those seen in class A VSGs), providing an interesting example in which the contents of the asymmetric unit of the crystal do not present the likely biological assembly, whereas the crystallographic arrangement does. Intriguingly, these two independent monomers are characterized by a significant conformational change in bottom lobe of the protein. This change consists of a 12-degree rotation of the bottom lobe from one monomer to the other (Fig 3B).

In addition, alignments of the VSG11_WT-Iodine crystal form with the VSG11_WT-Oil form (Methods and S1 Table) show that the second helix of the three-helix bundle displays a significant conformational change in the region near the middle of the bundle, the helix becoming disordered and forming a random coil in one monomer while keeping helical structure in the other (S2C Fig). Finally, when the different monomers of the 18mer are compared with the other crystal forms described, a wide variety of conformational changes is seen, although these changes are limited to the bottom lobe and lower regions of the bundle (S3 Fig).

These alternative conformations, seen in several crystal forms where the protein is not highly constrained by the energetics of crystal packing (non-crystallographic symmetry), suggests that the VSGs on the membrane could potentially be characterized by a collection of conformers, each with molecular surfaces that differ in these regions. It is possible that such differing conformers could present distinct antigenic surfaces and impact the immunogenicity of the VSGs, combining with other forms of variability such as surface protein features and post-translational modifications to augment the diversity space of the coat.

The structure of the VSG558 NTD was determined to 1.74 Å resolution (Fig 1D, Methods, S1A Fig, and S1 Table). It has a highly similar structure to VSG13 [18] with a 2.54 Å RMSD over 324 residues. This includes a broad and flat beta-sheet top-lobe that forms an intermolecular beta-sandwich in the dimeric assembly, a distribution of disulfides throughout the length of the rod-like structure, an N-terminus located in the “middle lobe” of the protein (instead of at the top, most membrane distal face of the VSG), and the presence of N-linked glycans directly below the top lobe. A notable difference with all the previously solved VSG NTD structures published is the presence of two N-linked glycans at the middle lobe of the VSG558 protein. To date, all VSG structures determined have been characterized by a single N-linked glycan (or none, as seen in the case of ILTat1.24). However, it has been established biochemically that some VSGs do possess more than one such sugar in the NTD, such as VSG5 [46]. The structure of VSG558 confirms the presence of multiple N-linked glycans for another VSG. This shows that not only are the locations of the N-linked glycans variable (found in both the bottom and middle lobes), but that the number of possible such carbohydrates is also variable.

The structures of VSG21 and VSG545 were also determined, both harboring closely related folds and biochemical properties that distinguish them from other VSGs (Fig 1E and 1F). Of note is the fact that both structures were obtained with proteolytic removal of the bottom lobe (contained in the C-terminal region of the NTD sequence). This cleavage occurred spontaneously during crystallization experiments performed with full-length VSG21 and the NTD of VSG545 that had been produced by limited proteolysis with Endoproteinase Lys-C (LysC) (see Methods). Presumably these cleavage events in the crystallization experiments occurred from residual contaminating endogenous protease(s) from T. brucei (and/or possibly residual LysC for VSG545). While such proteolysis has been seen before in the purification and crystallization of the VSG proteins [16,17,19,23], the cleavage typically occurs between the NTD and CTD of the protein in the linker sequence. In contrast, for VSG21 and VSG545, the cleavage in the crystallization drops produced a truncated NTD missing the bottom lobe sequence but maintaining the 3-helix bundle and top lobe of the protein. Superpositions with predicted models from AlphaFold (which were accurate enough to solve the experimental structures by molecular replacement) show possible conformations of the full NTD for each protein (S4 Fig). Finally, it should be noted that for VSG545, crystals were only obtained with PNGase treated material, removing the N-linked glycan (Methods). No sugar is seen in the VSG21 electron density, and no predicted site for N-linked glycosylation (Asn-Xaa-Ser/Thr) was found in the sequence of the NTD.

Both structures overall resemble the structures of VSG13, VSG397, VSG558, and VSGsur, possessing an N-terminus of the protein located approximately a third of the way down the long-axis of the VSG, a beta-sandwich fold for the top lobe, and disulfides both in the top lobe and spread elsewhere in the VSG. VSG21 and VSG545 are more similar to each other than these other VSGs, however, distinguished from them by the tightly twisted nature of the top lobe and, directly beneath it, the presence of a second three-stranded beta-sheet perpendicular to the axis of the 3-helix bundle (Fig 1E and 1F). VSG21 and VSG545 align well to each other in the top lobe but not to the other VSGs with beta-sheet top lobes (S5 Fig).

Overall classification of the VSGs based on structures

Altogether, as of the writing of this manuscript, there are fourteen published, experimentally determined protein structures of the VSG NTD from Trypanosoma brucei in hand: VSG1, VSG2, VSG3, VSG11, VSG13, VSG21, VSG397, VSG531, VSG545, VSG558, VSG615, VSG1954, VSGsur, and IlTat1.24. These fourteen structures not only map well to the bioinformatic clustering schemes published previously, but they better discriminate between them and put the classification schemes on an architectural foundation. We have sought to take these structures, with an eye to the bioinformatics work, and create what we conclude is a more explanatory organizational scheme for the VSG proteins (Fig 4). To provide continuity with the previous classification schemes of Hutchinson [15] and Cross [25], we have preserved the A/B designation for classes. However, this has required a change in the meaning of subdivisions within class A to reflect insight from the protein structures (discussed below). Finally, this schema was then tested by modeling hundreds of VSG proteins with the deep learning system AlphaFold and comparing the resultant folds to the classes we created.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Structure-based VSG Classification Scheme.

The two broad subclasses of VSG protein discussed in the main text are denoted at the top of the figure (class A and class B). Subclasses are shown in different colors with a collection of structures illustrating each class contained in a color-coordinated box matching the subclass name. Structures are drawn as in previous figures with the exception that O-linked glycans are colored red like N-linked glycans. Beneath each subclass name in gray are the clustering designations of the classes in previous papers discussed in the main text. Cyan backgrounds indicate structures solved prior to 2017. “N” and “C” refer to the N- and C-termini of the respective NTDs (colored by chain and placed nearby the terminal residue itself). As discussed in the text, VSG21 and VSG545 crystal structures are missing the bottom lobe sequence in the models due to truncation of the protein during crystallization.

https://doi.org/10.1371/journal.pntd.0011621.g004

To begin, there are two broad “super-families” of VSG structures based on the topological arrangement of the bottom lobe in the primary sequence (classes A and B, Figs 4 and S6). All VSG NTD structures determined have a bottom lobe subdomain structure (with the exceptions of VSG21 and VSG545 in this report that had this domain proteolytically removed). In class A the bottom lobe residues are present at the C-terminal portion of the NTD sequence, directly following the final helix of the bundle. In contrast, in class B the bottom lobe residues are present at the N-terminal portion of the NTD sequence between the amino acids that form the first and second helices of the bundle (S6 Fig). Further cementing this broad division into two main super-families are two observations. The first is that all class A VSGs studied to-date are found as dimers in solution and in protein crystals (the latter either as non-crystallographic dimers in the asymmetric unit or as monomers in the asymmetric unit where a crystallographic two-fold symmetry produces the dimer, [18,21,45]), whereas all studied class B VSGs are characterized by the same trimeric arrangement in the crystals (either through crystallographic or non-crystallographic symmetry, with monomers or trimers in the asymmetric unit, respectively), with biochemical evidence of monomer to trimer transitions based on protein concentration [24]. The second fact is that many of the class B VSGs are post-translationally modified by O-linked carbohydrates, whereas none of the class A VSGs have been found so modified. When comparing with previous efforts to classify the VSGs, our class A would correspond to class “A” in Cross [25], classes N1-N3/N5 in Weirather [26], and the older class A of Carrington [47]. Class B would correspond in these sources to classes B, N4, and B, respectively.

Within our class A superfamily are two large structural subclasses, A1 and A2 (Fig 4). In class A2 are found all the VSG structures that were solved prior to 2018 (Iltat1.24, VSG1, and VSG2), illustrating that the set of conclusions about VSG structure and function accepted for over a quarter century were based on a very limited subset of VSGs containing highly related protein folds. This subclass is characterized by a top lobe that contains all the cysteine disulfides in the NTD, a top lobe fold that is a hodge-podge of alpha helices and beta-strands, and N-linked glycan chains located in the bottom lobe of the NTD. In sharp contrast, subclass A1 VSGs are significantly longer due to the presence of a large beta-sheet subdomain for the top lobe (forming a beta-sandwich in the dimer). Further distinguishing A1 from A2 is the distributed nature of the disulfide bonds (present throughout the length of the VSG in A1), the presence of a “middle lobe” of secondary structure straddling the beginning of the three-helix bundle, and the location of the N-linked sugar(s) just below the beta-sandwich top lobe. This dramatically different arrangement in structural elements is reflected by the differing positions of the N-terminus of the NTD, namely toward the middle of the VSG fold in A1 but located at the very top of the fold in A2.

Additionally, the folds of the A1 VSGs from experimental and predicted structures (see below) suggest that this subclass can be further subdivided into three groups based on the size, conformation, and twist of the top lobe beta-sheet and the width of the space between the three-helix bundles in the dimer. These subclasses of A1 were in previous sequence clustering classification systems denoted as the separate classes A1 and A3[25] and N1, N3, and N5 [26]. However, all these subclasses are structurally similar to each other in the manners described above while differing markedly from the A2/N2 classes. Therefore, we considered it better to combine A1/A3 and N1/N3/N5 into a single class, A1 (a family contrasted to A2), and then subdivide them within A1: A1a, A1b, A1c (Fig 4). In this subdivision, we split the A3 class from Cross into A1b and A1c (which correspond to classes N5 and N3 in Weirather, respectively) based on differences in the top lobe architecture (e.g., the smaller top lobe beta-sandwich in A1c compared to A1b and the presence of a second beta-sheet over the middle lobe in A1c that is not present in A1b).

Finally, Cross et al. [25] divide the class B VSGs into two subgroups, whereas Weirather et al. [26] do not subdivide their equivalent class, N4. In contrast to the marked structural divergences between classes A1 and A2, and even the differences within the distinct subclasses of A1, we find no broad structural differences within the class B VSGs (examining features such as the protein fold, disulfides, N- or O-linked glycans, or oligomerization). However, three more subtle differences between members of the B class can be used to divide them into two subgroups. For example, two helical regions differ between the B1 and B2 subgroups. One helix that exists in the top lobe of the B2 class is not generally present in the B1 class (S7A Fig). Secondly, one of the bundle helices in class B1 is disordered in places relative to the same helix in many class B2 members (S7B Fig). Thirdly, in class B2 several VSGs possess an NTD with more amino acids. In the AlphaFold modelling discussed below, these additional residues are predicted to form both an extended loop in the disordered region of the helix and also longer loops in the top lobe (S7A Fig). Buttressing this subdivision of class B, when multiple B1 and B2 VSGs (from experimental and AlphaFold predicted models) are analyzed by structure (using the Dali Server [48]), these are divided into two classes consistent with the B1 and B2 groupings produced by sequence analysis (see the structural dendogram in S7C Fig). We have therefore denoted a split in Fig 4 between the B1 and B2 classes.

Mapping the VSGnome with AlphaFold

We sought to utilize the structural prediction system of AlphaFold [27,28,42] to quickly examine the sequences of hundreds of VSG proteins and compare the structural predictions both to experimentally determined VSGs as well as assess the fit of our classification scheme to the VSGnome as a whole. In the work that follows, all our predictive modeling was performed with AlphaFold2 [27,28,42] (see Methods), although we will simply refer to it as “AlphaFold”. Models were assessed by two quantitative metrics: (1) the root mean square deviation in the Cα residues of the protein main chain between experimental models and AlphaFold predictions (measured in Å with a lower value indicating a more accurate model) and (2) the global distance test total score (GDT_TS, scored as a % from 0–100, with higher score indicating a more accurate model) that is also computed over the Cα positions between model and experimental structure but which is less sensitive to small, localized discrepancies that occur in flexible/disordered loops [43]. The GDT_TS score is a primary metric for evaluating protein structural accuracy in the Critical Assessment of Structure Prediction (CASP), a global competition between protein structure prediction algorithms [49]. In the CASP14 competition, AlphaFold2 performed better than other submitted software systems in fold prediction and is now widely utilized and available through online servers and as open source software. Because RMSD values are measured in Å, values below 2Å are approaching the length of a carbon-carbon bond (1.54Å). Models with a GDT_TS value of 80% or higher are considered to have modeled most of the protein accurately [50], and above 90% are considered to be as accurate as experimental structures [51]. Therefore models with a score of ~80% constitute an accuracy sufficient for our efforts in classifying a given VSG with respect to our scheme discussed above (Fig 4).

We began by assessing the accuracy of AlphaFold against known VSG structures that existed in the PDB database at the time of writing this manuscript (VSG1, VSG2, VSG3, IlTat1.24, VSG13, and VSGsur), and which would therefore be accessible to AlphaFold as structural templates for training the deep learning neural networks. In addition, as the VSGs are multimers (dimers and trimers) and AlphaFold can predict both isolated folds and folds as oligomers [42], we generated models for monomer VSG sequences as well as for the appropriate oligomer for a given VSG sequence. Using ColabFold (Methods) and an input sequence of these VSGs, we examined five ranked models output by the software from which the rank 1 model (highest pLDDT score, see Methods) was compared with structures in the PDB database. All VSG structures predicted by AlphaFold were of high accuracy (Fig 5A), with oligomer predictions showing better values compared to monomer predictions (structures were predicted for the monomer and the oligomer using dimers for class A and trimers for class B VSG sequences, respectively). The predicted local distance test (pLDDT) of model confidence as well as the comparisons of model to experimental structures (as scored by the GDT_TS and RMSD) all showed statistically significant improvements in the predictions for oligomeric assemblies compared to monomers (S8A–S8C Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Comparison of Experimental VSG Structures with AlphaFold Models.

(A) Alignment of experimentally determined structures (blue) and AlphaFold models (gold) for structures whose coordinates were available at the time of the writing of this manuscript for use by AlphaFold as a template. (B) Alignment of experimentally determined structure (blue) and AlphaFold model (gold) for structures whose coordinates were not yet available at the time of the writing of this manuscript for use by AlphaFold as a template. RMSD, GDT-TS, and pLDDT are defined in the main text and Methods. Illustrations of alignments produced by CCP4mg.

https://doi.org/10.1371/journal.pntd.0011621.g005

Next, we repeated this analysis against structures not yet available in the protein database at the time of the analysis, namely VSG397, VSG531, and VSG1954 [16] and VSG11 and VSG615 from this manuscript. These recently determined structures were well-predicted by AlphaFold, although with slightly higher RMSD and lower GDT_TS values as compared to the predictions for VSG structures already in the database (Fig 5B). Frequently in these models, the three-helix bundle region showed the most accurate alignment, while discrepancies appeared mostly in loop regions connecting secondary structure. The predicted location of disulfide bonds matches very closely those observed in the experimental structures (e.g., S8D Fig). Finally, as noted in the Methods, the structures of VSG21, VSG545, VSG558, and VSG615 were solved by molecular replacement using an AlphaFold model of these VSGs as search models, indicating a very close fit of the model to the experimental data (a general idea for X-ray phasing implemented by several groups due to the accuracy of AlphaFold models [52–55]). These robust results from AlphaFold provide some confidence that it can be used as a tool to rapidly and robustly map the structural diversity of the VSGnome.

Focusing on the Lister427 strain database (Methods), we examined 215 NTD sequences with AlphaFold and compared these predictions to the existing structures known. All of the predicted structures conform to our classification scheme and no structure was predicted that conflicted with our schema (Fig 6). We also examined 83 sequences from a second Trypanosoma brucei strain, T. brucei brucei 927. All VSGs examined from Tb 927 were predicted to adopt structures fitting our classification scheme (S9 Fig). Extending our analysis to more distally related species and VSGs like T. vivax showed less confidence in the modeling (S10 Fig), and therefore it is likely that template experimental structures (which currently do not exist for this species) may be required on which the AlphaFold system can train to produce more reliable models.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. AlphaFold Models of Sequences from VSG Structural Classes.

Structures are drawn in different colors with a thin worm style. Class A1a with models for the NTD sequences of VSGsur (experimental) and AlphaFold predictions of VSG478, VSG483, VSG485, VSG507, VSG520, VSG523, VSG529, VSG535, VSG537 Class A1b with models for the NTD sequences of VSG13 (experimental) and AlphaFold predictions of VSG510, VSG650, VSG660, VSG661, VSG662, VSG663, VSG671, VSG717, VSG719 Class A1c with models for the NTD experimental structures of VSG21 and VSG545 together with the AlphaFold predictions of sequences of VSG391, VSG479, VSG493, VSG526, VSG545, VSG584, VSG612, VSG659 Class A2 with models for the NTD sequences of VSG1(experimental) and AlphaFold predictions of, VSG472, VSG476, VSG494, VSG495, VSG504, VSG514, VSG517, VSG525, VSG527, VSG536 Class B with models for the NTD sequences of VSG3 (experimental) and AlphaFold predictions of VSG473, VSG475, VSG480, VSG486, VSG491, VSG506, VSG513, VSG516, VSG528, VSG534. Alignments generated and illustrated with CCP4mg, each structure aligned in a different color [57].

https://doi.org/10.1371/journal.pntd.0011621.g006

From this we hypothesize that the space of protein structure for the T. brucei VSGs is well-described by the classes we have presented in this manuscript, although it is quite possible that some variants will be discovered outside this scheme. As an overarching observation, both the experimental structures and the predicted AlphaFold models all display unique molecular surfaces in terms of the distribution of chemical properties exposed to the immune system (charge, hydrophobicity, topography, etc., S11 Fig). This is of course consistent with the notion that the VSGs are each antigenically distinct, presenting unique faces to the immune system.