Integrative modeling reveals the molecular architecture of the intraflagellar transport A (IFT-A) complex

  1. Caitlyn L McCafferty
  2. Ophelia Papoulas
  3. Mareike A Jordan
  4. Gabriel Hoogerbrugge
  5. Candice Nichols
  6. Gaia Pigino
  7. David W Taylor
  8. John B Wallingford
  9. Edward M Marcotte

  • Curated by eLife

    eLife logo

    eLife assessment

    This paper will be of interest to scientists working on cilia, intraflagellar transport, and structural modeling. Using an integrative modeling approach, the paper provides a fundamental structural model for a part of the molecular machinery that is responsible for cilium assembly. However, additional approaches would improve confidence in the as yet incomplete structure model.

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Intraflagellar transport (IFT) is a conserved process of cargo transport in cilia that is essential for development and homeostasis in organisms ranging from algae to vertebrates. In humans, variants in genes encoding subunits of the cargo-adapting IFT-A and IFT-B protein complexes are a common cause of genetic diseases known as ciliopathies. While recent progress has been made in determining the atomic structure of IFT-B, little is known of the structural biology of IFT-A. Here, we combined chemical cross-linking mass spectrometry and cryo-electron tomography with AlphaFold2-based prediction of both protein structures and interaction interfaces to model the overall architecture of the monomeric six-subunit IFT-A complex, as well as its polymeric assembly within cilia. We define monomer-monomer contacts and membrane-associated regions available for association with transported cargo, and we also use this model to provide insights into the pleiotropic nature of human ciliopathy-associated genetic variants in genes encoding IFT-A subunits. Our work demonstrates the power of integration of experimental and computational strategies both for multi-protein structure determination and for understanding the etiology of human genetic disease.

Article activity feed

  1. Version published to 10.7554/elife.81977 on eLife
  2. eLife

    Author Response

    Reviewer #1 (Public Review):

    This manuscript by McCafferty et al. presents the integrative computational structural modelling of the IFT-A complex, which is important to proper cilium organelle formation in eukaryotic cells. Recent advances in protein structure prediction (AlphaFold) allowed the authors to model the structures of the 6 individual subunits of the IFT-A complex. Interactions between IFT-A proteins were experimentally investigated by purifying Tetrahymena cilia, isolating IFT complexes, and utilizing chemical crosslinking and mass spectrometry (MS). In addition, the authors present a somewhat improved 23Å cryo-electron tomography (cryo-ET) map of the IFT-A complex (previously determined cryo-ET structures of IFT trains have resolutions of 24 - 40 Å). Integrative modelling using the predicted structures of the 6 IFT-A proteins and the experimental data as restraints allows the authors to present a structural model for the entire IFT-A complex. This model is analysed in the context of the polymeric IFT train structure, interactions with the IFT-B complex, and the structural position of ciliopathy disease variants.

    This is in principle a timely and interesting study that attempts to push the limits of structural modelling of large protein complexes using structure prediction in combination with experimental data. Unfortunately, the study has several shortcomings and the data providing restraints for the integrative modelling are not optimal.

    1. Chemical crosslinking and MS were used to obtain both intra-molecular crosslinks used to validate the structural models of the individual IFT-A proteins as well as inter-molecular crosslinks used as restraints in the structural modelling of the hexameric IFT-A complex. It is mentioned on p. 4, line 9, that IFT-A complexes were enriched from the flagellar lysate M+M fractions using SEC and that fractions from SEC containing IFT-A complexes were crosslinked for MS analysis. However, the authors do not show the data for this sample, neither SEC profiles, SDS-PAGE nor data of the cross-linked samples. On p. 7 the authors write that their SEC profile corresponds to monomeric IFT-A, but this is not shown anywhere in the manuscript. The reason this is so important is that the IFT-A complex assembles into linear polymeric structures together with the IFT-B complex as so-called IFT trains in cilia. Data obtained from isolated IFT trains would thus have additional crosslinks between subunits in neighbouring IFT-A complexes that, if used to restrain the position of subunits within a hexameric IFT-A complex, would likely result in a wrong architecture. The fact that the authors also observe crosslinks between IFT-A and IFT-B proteins strongly suggests that they indeed carried out the crosslinking experiment on polymeric rather than monomeric IFT complexes.

    These are excellent points, and we apologize for previously omitting these data. In the new Figure 1—figure supplement 2, we now include size exclusion chromatography elution profiles for IFT-A along with molecular weight calibrants, plotting the mass spectrometrically-determined abundances of IFT-A subunits. Based on these data, we experimentally determined the molecular weight of the IFT-A particles that we analyzed to lie between 720 kDa and 1.1 MDa, consistent with the expected monomeric molecular weight of 772 kDa.

    These samples were isolated directly from Tetrahymena cilia and were composed of ~3% each of IFT-A and IFT-B. However, as we now note on p. 11, the samples were subsequently concentrated before crosslinking. We speculate that concentrating the particles could have induced some degree of oligomerization and interactions with IFT-B, which may in turn explain the small number of crosslinks consistent with IFT-A/IFT-A and IFT-A/IFT-B interactions. However, we have now removed all discussion of specific IFTA/B contacts in the paper and present only the general orientation of the two complexes as determined by cryo-ET.

    1. Given that the crosslink/MS data are unlikely to provide sufficient restraints for IFT-A structure assembly (and may even be misleading), the cryo-ET data become increasingly important. Unfortunately, the 23Å cryo-ET map does not provide sufficient detail to unambiguously fit domains of the IFT-A subunits as several of these have similar architectures consisting of WD-repeats followed by TPRs.

    We now address this comment using a different approach, which we describe in full on p. 5, 14-15, and Figure 2—figure supplements 1-4 of the paper.

    In particular, we used AlphaFold-Multimer (AF-Multimer) to identify confidently-modeled rigid-body domains and domain-domain interactions for directly contacting protein pairs (see Figure 2—figure supplements 1-2), which we used as starting models for integrative modeling (see Figure 2—figure supplements 3-4). We incorporated our cross-links as distance restraints for the modeling. This approach allowed us to model the entire IFT-A complex in a manner compatible both with our experimental structural data and the computationally derived restraints. We suspect this will be a very useful strategy for others to adopt, as the approach should be generalizable to many other large molecular assemblies that are too big to predict using AF-Multimer alone. Importantly, we see high concordance between the AlphaFold intermolecular constraints and our crosslinks (as plotted in the new Figure 2—figure supplement 4), and the models produced by this strategy agree well with the two structures presented in the newly posted preprints, which were arrived at using very distinct methodologies.

    This approach allowed us to withhold the cryo-ET tomogram from the modeling altogether in order to generate a fully independent model. We could then compare the final model to the subtomogram average and, by docking the model into the cryo-ET tomogram, to build a model of polymeric IFT-A, as described on p. 6 and presented in the new Figure 4, Figure 4—figure supplement 1, and Figure 4—animation 1.

    1. Two preprints of the IFT-A structure appeared over the last few weeks. Hesketh et al., (https://www.biorxiv.org/content/10.1101/2022.08.09.503213v1) have obtained a single particle cryo-EM structure of the human IFT-A complex at 3.5Å resolution for the IFT121/122/139 part of the complex providing amino acids side-chain information. In addition, Lacey et al. (https://www.biorxiv.org/content/10.1101/2022.08.01.502329v1) provide a 10-18Å resolution cryo-ET structure of the Chlamydomonas IFT trains containing both IFT-A and IFT-B. It is noteworthy that the model outlined in the current manuscript is very different from the IFT-A models of Hesketh et al., and Lacey et al. (the Lacey et al. manuscript by the way shares an author with the McCafferty et al., manuscript). In both Hesketh et al., and Lacey et al. the IFT121 and IFT122 subunits interact via the N-terminal WD-repeats and the C-terminal TPRs with the beta-propellers (WD-repeats) positioned parallel and in close contact. In the model proposed by McCafferty, the beta-propellers of IFT121 and IFT122 are positioned far away from each other (>50Å) and are perpendicular to each other. Several other large discrepancies are found in the relative positions of IFT-A subunits. This suggests serious problems with the structural model of IFT-A proposed by McCafferty and needs to be addressed with great care.

    This is an important point that we have indeed considered with great care. Our new model now positions the WD-40 domains of IFT121 and IFT122 proximal to each other and broadly matches the 2 preprints in general placement and orientation of all subunits, including the placement of IFT43, for which only we and Hesketh provide models.

    We now include an extensive comparison to the structures reported in the other two preprints. Note that a direct 3D alignment of the structures was not possible, as we were the only group to deposit our atomic coordinates. However, we now include a new Figure 4—figure supplement 2 orienting our structure to match figures appearing in those preprints, and use this as the basis for comparison, which can be found on p. 10. While it is not possible to calculate a quantitative measure of agreement (e.g. RMSD), our IFT43/120/121/139 structure visually agrees with the structure of Hesketh et al., even to the placement of IFT43, which is highly disordered for the most part, and which is omitted from Lacey et al. Our structure also generally agrees with that of Lacey et al. in this region, with the exception of what appears to be a re-orientation of the N-terminus of IFT139 in the Lacey structure relative to that of ours and Hesketh, which appear to be concordant with each other (again, with the caveat that we are limited in the comparisons we can make without having access to atomic coordinates.) Most importantly, all three structures agree with respect to the nature of the IFT-A monomer-monomer interactions in the polymeric train, with IFT140 acting to bridge adjacent monomers. Differences in the resolutions of the cryo-ET subtomogram averages (which range from 18 to 30 Å) are relatively small across the 3 studies and, at least as best we can tell by this necessarily crude comparison at this stage, do not obviously lead to any major changes in the structures of the polymeric assemblies.

    1. The authors observe crosslinks between the IFT-A proteins (IFT122 and IFT140) and IFT-B proteins (IFT70, IFT88, and IFT172) as discussed on pg. 6 and shown in figure 5A. To accommodate these crosslinks into the structural model of the IFT train shown in Figure 5A, the authors place the IFT-B subunits IFT70 and IFT88 far apart in the IFT-B complex. However, these subunits are known to interact directly (Taschner et al. JCB 2014) and indeed sit in proximity to the IFT train structure as observed by Lacey et al. While the crosslinking data may well be correct, the incorrect structural model of IFT-A likely forces an incorrect positioning of IFT-B proteins to fulfill the crosslinking data.

    It is now clearly evident that the earlier segmentation of the monomeric unit from within the polymeric IFT-A chain, which we based on the published segmentation of Jordan et al. Nat Cell Biol. 2018, did not properly capture the true boundaries of an IFT-A monomer, especially with regard to IFT140, which extends outward to connect adjacent monomers. The use of this artificially truncated monomer as a molecular envelope in our initial modeling effectively forced the IFT-A subunits to pack in a reversed orientation in order to fit the truncated density.

    In order to address this issue, we omitted the cryo-ET data from the modeling altogether and instead incorporated evidence capturing domain-domain structures of interacting protein pairs from AlphaFold-Multimer. This substantially reduced the number of degrees of freedom to be explored by the integrative modeling process in order to satisfy the available structural restraints, leading in turn to significantly better convergence of independent modeling runs and high concordance with the input data (Figure 2—figure supplement 3 and Figure 2—figure supplement 4), and a significantly improved structural model of the IFT-A monomer. Docking this refined monomer structure into the (now fully independent) cryoET tomogram produced a model of the polymer that fit well into the cryo-ET density (Figure 4 and Figure 4—figure supplement 1) and agreed in large part with those derived by Hesketh and Lacey, as described above and visualized in Figure 4—figure supplement 2.

  3. eLife

    eLife assessment

    This paper will be of interest to scientists working on cilia, intraflagellar transport, and structural modeling. Using an integrative modeling approach, the paper provides a fundamental structural model for a part of the molecular machinery that is responsible for cilium assembly. However, additional approaches would improve confidence in the as yet incomplete structure model.

  4. eLife

    Reviewer #1 (Public Review):

    This manuscript by McCafferty et al. presents the integrative computational structural modelling of the IFT-A complex, which is important to proper cilium organelle formation in eukaryotic cells. Recent advances in protein structure prediction (AlphaFold) allowed the authors to model the structures of the 6 individual subunits of the IFT-A complex. Interactions between IFT-A proteins were experimentally investigated by purifying Tetrahymena cilia, isolating IFT complexes, and utilizing chemical crosslinking and mass spectrometry (MS). In addition, the authors present a somewhat improved 23Å cryo-electron tomography (cryo-ET) map of the IFT-A complex (previously determined cryo-ET structures of IFT trains have resolutions of 24-40Å). Integrative modelling using the predicted structures of the 6 IFT-A proteins and the experimental data as restraints allows the authors to present a structural model for the entire IFT-A complex. This model is analysed in the context of the polymeric IFT train structure, interactions with the IFT-B complex, and the structural position of ciliopathy disease variants.

    This is in principle a timely and interesting study that attempts to push the limits of structural modelling of large protein complexes using structure prediction in combination with experimental data. Unfortunately, the study has several shortcomings and the data providing restraints for the integrative modelling are not optimal.

    1. Chemical crosslinking and MS were used to obtain both intra-molecular crosslinks used to validate the structural models of the individual IFT-A proteins as well as inter-molecular crosslinks used as restraints in the structural modelling of the hexameric IFT-A complex. It is mentioned on p. 4, line 9, that IFT-A complexes were enriched from the flagellar lysate M+M fractions using SEC and that fractions from SEC containing IFT-A complexes were crosslinked for MS analysis. However, the authors do not show the data for this sample, neither SEC profiles, SDS-PAGE nor data of the cross-linked samples. On p. 7 the authors write that their SEC profile corresponds to monomeric IFT-A, but this is not shown anywhere in the manuscript. The reason this is so important is that the IFT-A complex assembles into linear polymeric structures together with the IFT-B complex as so-called IFT trains in cilia. Data obtained from isolated IFT trains would thus have additional crosslinks between subunits in neighbouring IFT-A complexes that, if used to restrain the position of subunits within a hexameric IFT-A complex, would likely result in a wrong architecture. The fact that the authors also observe crosslinks between IFT-A and IFT-B proteins strongly suggests that they indeed carried out the crosslinking experiment on polymeric rather than monomeric IFT complexes.

    2. Given that the crosslink/MS data are unlikely to provide sufficient restraints for IFT-A structure assembly (and may even be misleading), the cryo-ET data become increasingly important. Unfortunately, the 23Å cryo-ET map does not provide sufficient detail to unambiguously fit domains of the IFT-A subunits as several of these have similar architectures consisting of WD-repeats followed by TPRs.

    3. Two preprints of the IFT-A structure appeared over the last few weeks. Hesketh et al., (https://www.biorxiv.org/content/10.1101/2022.08.09.503213v1) have obtained a single particle cryo-EM structure of the human IFT-A complex at 3.5Å resolution for the IFT121/122/139 part of the complex providing amino acids side-chain information. In addition, Lacey et al. (https://www.biorxiv.org/content/10.1101/2022.08.01.502329v1) provide a 10-18Å resolution cryo-ET structure of the Chlamydomonas IFT trains containing both IFT-A and IFT-B. It is noteworthy that the model outlined in the current manuscript is very different from the IFT-A models of Hesketh et al., and Lacey et al. (the Lacey et al. manuscript by the way shares an author with the McCafferty et al., manuscript). In both Hesketh et al., and Lacey et al. the IFT121 and IFT122 subunits interact via the N-terminal WD-repeats and the C-terminal TPRs with the beta-propellers (WD-repeats) positioned parallel and in close contact. In the model proposed by McCafferty, the beta-propellers of IFT121 and IFT122 are positioned far away from each other (>50Å) and are perpendicular to each other. Several other large discrepancies are found in the relative positions of IFT-A subunits. This suggests serious problems with the structural model of IFT-A proposed by McCafferty and needs to be addressed with great care.

    4. The authors observe crosslinks between the IFT-A proteins (IFT122 and IFT140) and IFT-B proteins (IFT70, IFT88, and IFT172) as discussed on pg. 6 and shown in figure 5A. To accommodate these crosslinks into the structural model of the IFT train shown in Figure 5A, the authors place the IFT-B subunits IFT70 and IFT88 far apart in the IFT-B complex. However, these subunits are known to interact directly (Taschner et al. JCB 2014) and indeed sit in proximity to the IFT train structure as observed by Lacey et al. While the crosslinking data may well be correct, the incorrect structural model of IFT-A likely forces an incorrect positioning of IFT-B proteins to fulfill the crosslinking data.

  5. eLife

    Reviewer #2 (Public Review):

    The authors use XL-MS and AlphaFold to predict the structure and interactions of the six individual IFT-A proteins of Tetrahymena. As this data set still allows for numerous possible 3D structures of the hexameric complex, the authors fitted their models to the low-resolution 3D structure of the IFT-A densities of Chlamydomonas IFT trains in situ obtain by cryo-EM and image averaging. While not optimal, this cross-species approach is possible as IFT proteins are highly conserved and the identified crosslinks fit the Tetrahymena and Chlamydomonas AlphaFold structures almost equally well. The result is a best-fitting model, which was further "validated" by accounting for previously established interactions between IFT-A proteins (and IFT-A to -B interactions). The manuscript also provides a scholarly comparison of the IFT-A particle and protein structure with other cellular protein of similar domain structure and observe that many such proteins participate in intracellular transport.

    The structure of the IFT-A complex presented here is modeled rather than based on direct imaging. In as much, this is probably an intermediate step. However, because the fine structure of the IFT-A particle remains unknown, this indirect approach appears useful and appropriate. The model presented here fits the available data and likely can be tested further in future experiments. Probably, the approach could be also used to predict the structure of other multiprotein complexes. The work elegantly demonstrates how the structures of single proteins provided by AlphaFold can drive structure predictions of protein complexes.

  6. Version published to 10.1101/2022.07.05.498886v2 on bioRxiv
  7. Version published to 10.1101/2022.07.05.498886v1 on bioRxiv