Tubulin cofactors and Arl2 are cage-like chaperones that regulate the soluble αβ-tubulin pool for microtubule dynamics

Several parts of the work presented – e.g. the biochemical characterisation of the chaperone components and the yeast TBCC C-terminal domain crystal structure – shed new light on aspects of the chaperone mechanism. However, there are some significant concerns, particularly about the 3D reconstruction using negative stain EM data, which would need to be addressed. The data on microtubule dynamics appear difficult to interpret, and the authors should very seriously consider saving them for another manuscript, where the effect of the complex on microtubule growth is investigated more thoroughly.

We thank the editor and the reviewers for their enthusiasm for our manuscript, and for the thoughtful advice in revising the manuscript. We have fully revised the manuscript in order to address all comments. Major areas of revision include:

1) Supporting negative-stain electron microscopy analyses. We present revised multivariant statistical analysis, and comparisons of the model projections to image class averages for each of the structures. We also provide a parallel molecular docking approach independent of segmentation to support our docking of various molecular models into the refined EM density.

2) As suggested by the reviewers, we have removed our TIRF-based experiments on the effects of the tubulin cofactors on microtubules dynamics. We are further revising these studies and we will be publishing those studies in separate manuscript elsewhere.

3) We have significantly revised our results and discussion of GTP hydrolysis studies, to make both the results and our interpretations clearer.

4) We have revised and improved the manuscript throughout to clarify the language and improve cohesion of the manuscript.

Essential revisions:

1) Table 1 indicates that TBCC peptide coverage is 0.9%. Is this a typo? If not, what is the protein that has been studied?

The mass-spectrometry analysis of the tubulin cofactor complex was carried out from a bacterial co-expression of TBCA, TBCB, TBCC, TBCD, TBCE and Arl2 with an individual his-tag on TBCD. The sample analyzed included purified materials following nickel affinity chromatography. The low coverage of 9% for TBCC is due to the low affinity of TBCC to the TBC-DEG complex, which is successively lost in the later purification steps. These have been further explained in the manuscript.

2) In the first paragraph of the subsection “Tubulin cofactors D, E and Arl2 GTPase form a stable heterotrimeric TBC-DEG chaperone” the author states that ‘Monomeric TBCD, TBCE and Arl2 subunits were not observed in vitro at any concentration and the TBC-DEG complex behaves as a single biochemical entity’. Could it be that the subunits were not recovered on the column due to their tendency to precipitate?

We agree with the reviewers that it is likely that any individual TBC-DEG subunits precipitate if not integrated into the stable complex. This is supported by our finding that the individually expressed TBCD, TBCE and Arl2 subunits are insoluble.

3) 3D reconstruction using negative stain EM data is a significant proportion of the assembled evidence. Although intrinsically limited in achievable resolution, use of negative stain is legitimate with such small and apparently aggregation-prone complexes. However, as outlined below, the analysis of these EM data does not appear to be sufficiently robust and needs to be improved, in order to ensure that all potential errors and/or ambiguities are eliminated.

We appreciate the reviewers’ suggestions on revising the electron microscopy analyses. These suggestions have improved our manuscript and the revised manuscript presents these changes as described below.

The authors have taken a systematic approach to reconstructing a series of multi-subunit complexes applying standard methodologies. However, for these standard methods to work, several conditions must be met:

A) Individual particles and class averages must be centred: based on the views in Figure 3–figure supplement 1B, this does not appear to be the case – e.g. top row, second from right, fourth row, first on left;

We have revised our multi-variant statistical analysis (MSA) particle image classification for the structures described. These data now show centered class averages in Figure 4–figure supplement 1B, Figure 5–figure supplement 1B and Figure 7–figure supplement 1B.

B) Classes must be homogeneous – however multiple class averages show a lack of crispness that is suggestive of member heterogeneity – e.g. second row 9 along; third row 7 along. This problem is particularly acute in the larger complexes shown in Figure 6–figure supplement 1B but also in Figure 4–figure supplement 1B; note also there are no stars in these figures, as suggested by the legend.

We have revised our multivariant statistical analysis (MSA) particle image classification to calculate fewer class averages to encompass more images per class average. Our revised analysis show crisp class averages, as expected, and show more of the features of each view of the complex. These are now shown in Figure 4–figure supplement 1B, Figure 5–figure supplement 1B and Figure 7–figure supplement 1B.

When these conditions are not met, structures cannot be considered to be reliable. Unreliability is further supported by the lack of match between class averages and re-projections e.g. in Figure 3–figure supplement 1D, Figure 4–figure supplement 1E and Figure 6–figure supplement 1E.

We present a revised projection-matching using the approach described by Lyumkis et al. 2013 (Science, 342 (6165) 1484-1490), where global angular search was carried out using the MSA class averages to identify and refine the matching model projections to each class average. Our revised analyses show the MSA class averages closely match projections from models presented for each structure. These are presented in Figure 4–figure supplement 1C, Figure 5–figure supplement 1C and Figure 7–figure supplement 1C.

The segmentation imposed on such lower resolution reconstructions is essential for interpretation, but appears a bit arbitrary. Putting aside concerns about the structures themselves, the evidence presented for subdomain assignment is not currently justified. First, the high cross-correlation (CC) values from the docking calculations are not by themselves evidence of a unique assignment – for example, what was the CC when the HEAT repeats were fit in the pink density rather than the blue density?

We completely agree with the reviewers that the segmentation of low-resolution negative stain electron microscopy maps involves a degree of arbitrariness. However, our maps provide enough shape information that molecular models can be fit to unique positions within these maps even without using segmentation. There is a moderate to high sequence homology between TBCD and TBCE-LLR (40% identical) to the paralog structures being used for the fitting. We used an independent approach for fitting, presented in Figure 4–figure supplement 3. Low-resolution models for TBCD, TBCE paralogs and Arl2, generated from high-resolution structures or homology models, were cumulatively fit into the TBC-DEG map. The paralog structure for TBCD, the largest subunit, has a distinctive “ring with rod” shape, which can be fit fairly unambiguously. Once TBCD is positioned and that region of the map is excluded, the TBCE-LLR can only be fit reasonably into the bow density with two globular densities, attached both ends of the bow, likely to be TBCE Cap-Gly and C-terminal ubiquitin domains. Finally, the more globular Arl2 matches the size of the remaining pillar density.

The lack of connectivity between the TBCE LRR and Ubiquitin is particularly unsatisfying – and the CAPGly and LRR are separated by ∼100A in the final +tub +TBCC reconstruction, which seems unlikely.

The lack of connectivity between the TBCE-LLR and Ubiquitin domain is due to our choice of contour-threshold chosen to present the TBC-DEG map or potential density defects caused by negative stain. We describe these details in the revised manuscript in the subsections “The TBC-DEG complex is a cage-like chaperone with a hollow central core” and “Electron microscopy and Single particle image analysis”.

We have reevaluated the positioning of the Cap-Gly domain in the TBC-DEG Q73L:αβ-tubulin:TBCC map based on the reviewer’s comment. We agree that the Cap-Gly domain placement was likely incorrect in the previous model. We have generated a revised model for the TBC-DEG-αβ-tubulin: TBCC complex in comparison to its previous positions in the TBC-DEG and TBC-DEG:αβ-tubulin maps. We believe the revised model takes into account the positioning of TBCE-terminal domains more accurately. We have further described these details in the revised manuscript in the subsections “TBCC β-helix wedge interfaces with Arl2 and αβ-tubulin dimer in the TBC-DEG chaperone” and “Electron microscopy and Single particle image analysis”.

However, higher resolution structures will be required to further understand the conformations of individual domains in each of the different states.

Second, the data localising the GFP density in two different tagged versions of the TBC-DEG complex (Figure 3–figure supplement 2) are unconvincing because: a) the density for the GFP in the class averages is not always visible and b) it is not obvious that the same projections are being compared {+/-}GFP.

We have revised the MSA particle image classification of GFP-fusion TBC-DEG complexes, as described above for the native complex. We have used projection matching to compare the TBC-DEG and TBC-DE(GFP)G projections and have identified the correct projections more accurately. The GFP density appears clearly in most revised projections.

4) The experiments with dynamic microtubules represent the weakest part of the story. The authors show that in the presence of 6μM tubulin, nanomolar concentrations of TBC-DEG moderately reduce microtubule growth rate and promote rescues in a concentration-dependent manner. At concentrations above 20 nM, the effect on rescues is reduced and a bi-modal distribution of catastrophe frequencies is observed. When the concentration is increased further, microtubule pausing is observed. When the TBC-DEG complex version containing the GTP-locked Arl2-Q73L is used, very slow growth or even microtubule pausing is observed. The authors interpret the experiments in terms of the effects on the tubulin dimer (dimer re-activation, sequestration or decay, dependent on concentration, although there is very little evidence to make a justified choice between these different possibilities). However, it is well known that when tubulin concentration is increased, microtubules normally should grow faster. Slow growth, pausing and rescues – very prominent phenotypes observed by the authors – are all typically associated with proteins acting on microtubule ends or on the microtubule lattice. The authors suggest that their complexes do not bind to microtubule ends, but the data shown (Figure 7K) are clearly not of sufficient quality to make a strong conclusion about this: TBC-DEG-Q73L strongly binds to the GMPCPP seeds and the overall background is high. Further, it is not shown whether the ReAsH labelled complex has the same activity as the unlabelled one. The only way to exclude the direct effects of the cofactors on the microtubules would be to pre-incubate the tubulin with the cofactor for different times, remove the cofactor and perform microtubule dynamics assays. Based on these considerations, the authors are strongly encouraged to remove this section and save the results for a subsequent study, where the effects of the complex on microtubule dynamics are researched properly.

We agree with the reviewers that the regulatory effects of tubulin cofactor chaperone on the soluble tubulin pools and their dynamic polymerization into microtubules are complex. The defects in GTP hydrolysis observed in the Arl2 Q73L mutant, however, lead to pausing of microtubule dynamics, that we observed both in vivo and in vitro.

As the reviewers requested, we have removed these data from the manuscript.

5) It remains difficult to understand the effect and the importance of GTP hydrolysis by Arl2 for the TBC-DEG cycle. While no additional experiments seem to be necessary, the authors should make a much better effort to describe their interpretation of the results discussed in the section “Sequential binding of tubulin and TBCC activates maximal GTP hydrolysis in TBC-DEG chaperones” and in the Discussion (second paragraph). What do the authors mean when they say “Arl2 GTPase state likely controls the TBC-DEG chaperone state” – to which part of the chaperone cycle does this refer? For many small GTPases, locking them a GTP state makes them constitutively active, and indeed Q73L-Arl2 promotes formation of the complex of TBC-DEG with tubulin and TBCC. Is tubulin recycling blocked in this case? If so, the complex containing Q73L-Arl2 will act as a stable tubulin sequestering agent.

We have revised the text in the Results and Discussion sections to further clarify the role GTP-hydrolysis to address all the points described above regarding the mechanism of the tubulin cofactors/Arl2 chaperone system.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

1) Figure 4–figure supplement 1 seems to contain colored asterisks and some other material hidden behind panel B. This probably reflects inaccurate figure preparation and must be removed.

We thank the reviewer for noticing this error. We have removed these asterisks in the revised Figure 4–figure supplement 1.

2) In Figure 4–figure supplement 1 (panels B and D), some of the particles are still not centered within the box. This is most apparent with donut-shaped averages where dark center of particle should be in middle of box e.g. top row, 3rd from left: original critique of these data still applies.

The reviewer is correct that the class average images in Figure 4–figure supplement 1 are slightly off-center. Indeed, in these images all of the class averages are 1-2 pixels to the left of center. This is an image masking/display artifact, and has no effect on the quality of the resulting reconstruction or on the quality of our multivariant statistical analyses classification. Each individual image used for the electron microscopy reconstructions and validation of model projections are independently centered prior to their inclusion in projection matching or angular refinement algorithms.

3) Please compare your work to the paper by Serna et al. in J Cell Science (PMID: 25908846).

We agree that the study above is important, and have added the reference in the Results section of the manuscript where appropriate. Our conclusions, however, potentially differ from the conclusions of Serna et al., and we do not observe a TBCB-TBCE complex in our work. Our approach to co-express the TBC proteins with different tags did not lead to TBCB interacting with TBCE under any condition. In addition, we never observed isolated TBCE and thus our data indicate that recombinant TBCE is likely insoluble without TBCD and Arl2.

It is possible that the TBCE protein described in the Serna et al. manuscript is metastable. Alternatively, TBCD and Arl2 may have co-purified from insect cells where the purification of TBCE was carried out. The lack of mass-spectrometry data in this manuscript prevents us from comparing the TBCE preparation to that we present.

Despite these outstanding issues, we think it is significant that the overall shape and organization of the low resolution TBCE reconstruction presented by Serna et al. suggests a similar organization to the region we assign as TBCE in our map. The crescent shape that we assign to the TBCE LLR domain with its two globular densities supports our conclusions regarding TBCE organization, with terminal Cap-Gly and ubiquitin domains.