Electron Microscopy and Particle Classification
Negatively stained samples of yeast γ-TuSC adopted a preferred orientation on carbon grids, resulting in characteristic Y-shaped projections (A, filled arrowheads). The particles are roughly 120 Å wide at the tips of the Y, tapering to 40 Å at the narrow end, and 160 Å long (B). A smaller number of side views of the complex were also seen, which are the same length as the predominant Y-shaped view, but only half the width (A, empty arrowheads). Random conical tilt (RCT) reconstruction is the method of choice for three-dimensional reconstructions of particles with a limited number of orientations on the grid (Radermacher et al., 1986
). RCT makes use of tilted and untilted views of each particle, allowing very accurate determination of its orientation. An additional advantage of RCT is that the untilted particle images can be classified to ensure homogeneous data sets for reconstruction. We acquired 163 micrograph pairs by using the automated conical tilt software of the UCSF Tomo package (Zheng et al., 2007a
), and 6464 individual particles were picked from the micrograph pairs.
Figure 1. Negative stain images of γ-TuSC. (A) Section of an untilted micrograph. Y-shaped views (filled arrowheads) predominate, but most images also included i-shaped side views (empty arrowheads). Only Y-shaped views were selected for reconstruction. (more ...)
An initial reference-free alignment of all of the untilted particles produced a clear average with the expected Y shape. However, the variance map of the aligned particles had strong peaks in the regions around the ends of the arms, indicating heterogeneity in the data set. The globally aligned data set was separated into more homogeneous classes by using PCA and hierarchical clustering. The first round of clustering yielded two distinct classes; the primary difference between the two was the branching angle between the two arms. The alignments were improved using the two class averages as references for several rounds of iterative supervised classification of the entire data set. Further PCA and hierarchical clustering within each of the two reference-based classes resulted in five unique classes, which were used as references for several more rounds of iterative supervised classification. The homogeneity of the classes was evaluated throughout the procedure by inspection of variance maps of the class averages. Further subdivision within each of the five final classes did not yield averages with any appreciable differences, another indication of homogeneity. The five class averages seem to be closely related, and differ primarily in the angle between the two arms of the Y, and in the partial bifurcation of the body in classes 3 and 4 (C).
Random Conical Tilt Reconstructions
Three-dimensional density maps were generated for each of the five particle classes. Preliminary orientations for each particle were assigned by combining the data collection tilt geometry with the in-plane rotation determined by alignment of the untilted images. An initial volume was generated by back-projection of the particles, and translational shifts were iteratively refined by cross-correlation of each particle with its corresponding projection. The reconstructions have resolutions of 25–27 Å as measured by the Fourier shell correlation by using the 0.5 cutoff criterion ().
FSC resolution measurements of the RCT reconstructions in . The resolution of the reconstructions is taken as the spatial frequency at which the FSC drops below 0.5.
Each of the structures is roughly Y-shaped (). The body of the Y is a rough hexahedron 70 Å long × 40 Å wide × 30 Å deep. Each of the arms is ~100 Å long, and consists of a narrow (roughly 20 Å) constriction near the branch point, and two terminal bulbous lobes roughly 40 Å wide and 60 Å long. The two arms are asymmetric, with one lying flat against the carbon support film and the other arching slightly near the branch point. The branching angle between the arms varies from ~35° in classes 3, 4, and 5, to 45° in class 2 and 50° in class 1. The branching angle variation leads to differences in the center-to-center distance between the two terminal lobes: 85 Å in class 1, 75 Å in class 2, and 70 Å in classes 3, 4, and 5.
Figure 3. Three-dimensional reconstructions of γ-TuSC. Volumes were generated with tilted particles corresponding to classes 1, 2, and 3 in C. Four orthogonal views of each particle are shown, and the rotations between the views are indicated at (more ...)
Classes 3, 4, and 5, which account for two thirds of the particles, have essentially identical structures related by rotations about the long axis of the molecule, presumably due to small differences in how the particles lay down on the grid. A rotation of 10° brings volume 4 into alignment with volume 3, whereas a rotation of −10° brings volume 5 into alignment with volume 3 (Supplemental Figure S1). It is likely that the particles actually adopt a continuum of orientations, and the classification into these three groups represents the finest division possible at this resolution.
Volumes 1 and 2 are distinguished by rotation of the arched arm away from the other (). Rotation around the long axis of these structures—by 10° for volume 2 and 15° for volume 1—brings the body and the fixed arm into alignment with volume 3. When the densities are superimposed, it is clear that rotations about the base of the mobile arm relate the three structures (, right-hand column). Relative to volume 3, the mobile arm is rotated 8° in volume 2 and 15° in volume 1. Again, it is likely that the three calculated structures represent points along a continuum, and subdivision into a larger number of classes is limited by the resolution of the images.
To calculate an average structure, each of the five volumes was segmented into two parts: the mobile arm, and the body and fixed arm (). The corresponding segments from each structure were aligned to the segments of volume 3. An average was then calculated for each segment, weighted for the size of the corresponding classes. The approximate mass of each part of the structure is 115 kDa for each arm and 70 kDa for the body. The position of the mobile arm can be altered to generate the states observed in the different reconstructions.
Figure 4. Average γ-TuSC map. Each of the reconstructed volumes was segmented into the constant region and the mobile arm, as indicated in . An average was calculated for each segment, weighted for class size. In this view, the two average maps (more ...)
Effect of Nucleotide State
γ-Tubulin binds GTP and GDP with affinities similar to those for the exchangeable site of β-tubulin (Aldaz et al., 2005
). We investigated the possibility that the nucleotide state of Tub4p may alter the conformation of γ-TuSC. Images of 2321 particles of γ-TuSC incubated with 1 mM GTPγS, a nonhydrolyzable GTP analog, were acquired with the conical tilt data setup. There were no obvious differences between the two-dimensional averages of GDP- and GTPγS–γ-TuSC, and the class distribution was similar with either nucleotide. The three-dimensional structures are also very similar, as measured by FSC. We conclude that nucleotide state does not induce any large-scale conformational changes in γ-TuSC. This result is not surprising, in light of crystal structures of human γ-tubulin with nearly identical conformations with either nucleotide bound (Aldaz et al., 2005
; Rice, Montabana, and Agard, personal communication).
Position of Tub4p Determined by Gold Labeling
To determine the location of Tub4p within γ-TuSC, complexes with N-terminally His-tagged Tub4p were labeled with Ni2+-NTA-Nanogold, and a data set of 612 particles was collected. The presence of gold label in a fraction of the data set was immediately obvious from a variance map of the aligned images. PCA and clustering of the aligned images yielded a class of 68 labeled particles (A). The class average of labeled γ-TuSC clearly shows the 18-Å nanogold particle bound between the two bulbous lobes at the end of each arm (B). The location of the gold label was confirmed by a difference map calculated between labeled and unlabeled particles, with a 9.3σ peak at the location of the gold particle (D). The presence of a single gold label (rather than the 2 that might be expected for labeling of each Tub4p subunit) is explained by both of the His-tags in the complex binding the same multivalent Ni2+-NTA-Nanogold. In keeping with this explanation, it proved impossible to subclassify the labeled particles into the different classes found in the unlabeled data set (, classes 1, 2, and 3), likely due to the label constraining the distance between the two heads.
Figure 5. Labeling of the Tub4p N-terminus. (A) Individual particles of γ-TuSC with an N-terminal 6xHis-tag on Tub4p incubated with Ni2+-Nanogold. (B) Average of 68 labeled particles. (C) Average of 68 unlabeled particles. (D) Difference map generated by (more ...)
Based on the gold localization, we manually docked the crystal structure of γ-tubulin, the human Tub4p orthologue, into the average EM density (). The size and shape of the arm density allows an excellent fit of the crystal structure, which accounts for roughly two third of the density of the terminal lobes.
Figure 6. Docking γ-tubulin in the average γ-TuSC structure. (A) The crystal structure of γ-tubulin (gold), the human Tub4p homologue, was manually fit into the lobes at the ends of the arms. (B) Side views, rotated 90° relative (more ...)
Relative Orientations of Spc97p and Spc98p N and C Termini Determined by FRET
The relative positions of the γ-TuSC components were examined by FRET, with various combinations of CFP- and YFP-tagged proteins used as donors and acceptors (). FRET was measured in vivo, and in all cases cell viability depended on the tagged proteins being functional. The FRET interactions are summarized in A. In the absence of energy transfer, the predicted value of FRETR
is 1, and statistical analysis of the γ-TuSC data (see Materials and Methods
; Muller et al., 2005
) shows that values greater than 1.2 are significant. Although the linkers and large size of the fluorescent protein tags make it difficult to convert the FRETR
signals into absolute distances, for our purposes it is sufficient to correlate stronger FRETR
with shorter distances.
Figure 7. FRET relationships and the overall architecture of γ-TuSC. (A) The average FRETR values for the CFP- and YFP-tagged proteins presented in . The lines connect the protein pairs that showed significant FRET, and they are not meant to correspond (more ...)
The FRET measurements indicate that the C termini of Spc97p and Spc98p are in proximity to each other, and that the C terminus of Spc97p is near both termini of Tub4p. Despite the homology of Spc97p and Spc98p, the C terminus of Spc98p could not tolerate a fluorescent protein tag when Tub4p was also tagged at either terminus. Whether this synthetic lethality is due to a significant structural difference between Spc97p and Spc98p is unclear; however, it is consistent with the asymmetry seen in the EM density. Both the observed FRET and the synthetic lethality imply that Tub4p interacts with the C-terminal regions of Spc97p and Spc98p.
Significant FRET values are also observed between the N-terminally tagged Spc97p and Spc98p, suggesting that the N termini of these two proteins are near each other. The lack of FRET between N-terminally tagged Spc97p and Spc98p and any of the other termini in γ-TuSC suggests that the N-terminal regions of Spc97p and Spc98p are distal to the Tub4p heads. Although these results may be complicated by intermolecular FRET, the simplest interpretation would pair Spc97p and Spc98p with their N termini toward the base of the Y, and their C termini interacting with Tub4p in the arm regions (B).
We were also interested in determining which regions of γ-TuSC are involved in binding Spc110p, which anchors γ-TuSC to the SPB via interactions with its N-terminal domain (Knop and Schiebel, 1997
; Sundberg and Davis, 1997
). N-terminally tagged Spc110p showed significant FRET with the C-termini of all three γ-TuSC components. Although N-terminally tagged Spc97p and Spc98p were significantly different from baseline by Tukey–Kramer statistical analysis, we label these interactions as tentative on the basis of their standard deviations (). The N-terminal Spc110p FRET values suggest that it binds γ-TuSC near the branch point connecting the two arms to the body.