The structural basis of intraflagellar transport at a glance

Cilia are highly ordered and conserved microtubule-based organelles of eukaryotic cells that play fundamental roles in cell motility, sensing and signaling. Their assembly and maintenance depend on the intraflagellar transport (IFT) system. Large macromolecular machines, called IFT trains, powered by molecular motors, deliver ciliary building blocks to the assembly site at the distal tip (Kozminski et al., 1993; Lechtreck, 2015). Despite moving in a confined space between the ciliary membrane and the microtubule tracks, oppositely directed IFT trains move fast, up to ∼3.5 µm/s (Kozminski et al., 1993), and rarely collide (Stepanek and Pigino, 2016). Additionally, trains that move toward the tip carry retrograde motors required to return to the base of the cilium. Nevertheless, these trains move processively and unidirectionally (Kozminski et al., 1993), indicating that oppositely directed motors do not engage in a tug-of-war. Recent advances in structural biology have provided fascinating insights into how these processes are regulated.

IFT trains are repetitive structures and assemble from two large protein complexes, IFT-A and IFT-B, which serve as molecular adaptors between the motors and various cargo proteins (Taschner and Lorentzen, 2016). Anterograde trains are driven by the plus-end-directed motor kinesin-2 complex (Cole et al., 1998), whereas retrograde trains are powered by the minus-end-directed IFT-dynein complex, called dynein-2 in humans and dynein-1b in Chlamydomonas (Porter et al., 1999; Pazour et al., 1998). Although a detailed picture of the transport process has been obtained from mutant analysis and fluorescence microscopy (Taschner and Lorentzen, 2016), detailed structural information is required to understand how interactions between individual components control this unique machinery.

Since the first observation of IFT in the green alga Chlamydomonas reinhardtii (Kozminski et al., 1993), it has been shown that the IFT system is conserved across eukaryotic species, and structural studies have begun to provide insights into the role of its various components. Our understanding of how these components work together to produce a functional IFT machinery, however, is still some way from being complete. Recent studies have begun to address this gap, for instance by providing structural insights into the mechanism and assembly of anterograde IFT trains, the regulation of the dynein motor and the binding of specific cargoes. These and other advances in the field have helped to place observations from earlier studies in the context of complete IFT trains. In this Cell Science at a Glance article and the accompanying poster, we discuss these insights and explore the work still required to obtain a complete structural and mechanistic understanding of IFT.

IFT-A and IFT-B protein complexes build the scaffold of the trains while also serving as versatile adaptors between motor proteins and cargoes (see poster). Structural information is available for some individual IFT subunits, but the mechanisms that build the complete complexes and provide their functionality are unknown. IFT-A and IFT-B were first identified on the basis of their separation once extracted from cilia in low-salt concentrations (Piperno and Mead, 1997; Cole et al., 1998), and currently six subunits are known to belong to IFT-A and 16 subunits to IFT-B (see Table 1). Crystallographic analysis and structural predictions of IFT proteins revealed typical protein–protein interaction domains as major motifs, such as tetratricopeptide repeats (TPR), WD40 repeats and coiled-coils (Cole, 2003; Taschner et al., 2012), which, besides promoting IFT–IFT interactions, also support their adaptor function (Taschner et al., 2012). Crystal structures of several individual or pairs of IFT-B components (Bhogaraju et al., 2011, 2013; Taschner et al., 2014, 2016, 2018; Wachter et al., 2019) have provided important insight into their function, but we cannot yet explain how complexes are formed or trains assembled. Cryo-electron tomography (ET) and subtomogram averaging of whole trains in cilia from living cells have provided intermediate-resolution structures for both complexes, and show how they arrange into IFT trains (Jordan et al., 2018). However, because of the gap in resolution between cryo-ET and crystallography data, fitting of the available crystal structures remains challenging. To overcome this, the complexes will need to be investigated at higher resolution, presumably by single-particle electron microscopy (EM). Together with biochemical interaction maps (Taschner and Lorentzen, 2016; Nakayama and Katoh, 2020), these data will help us understand how IFT components organize into active functional machines.

Cryo-ET of anterograde trains shows that elongated IFT-B complexes form the core of the trains (see poster). This central role of IFT-B results in the inability of most IFT-B mutants to form trains or even assemble cilia (Richey and Qin, 2012; Hou et al., 2007). At high salt concentrations, IFT-B splits into two subcomplexes of ten and six proteins, known as IFT-B1 and IFT-B2, respectively (Lucker et al., 2005; Taschner et al., 2014, 2016). So far, interaction maps of IFT-B subunits have been obtained mainly by co-expression and pulldown assays or yeast two-hybrid analysis (Lucker et al., 2005; Taschner et al., 2016; Katoh et al., 2016). In the future, comparative structural analyses may provide insights into how IFT-B subcomplexes arrange within the overall structure of IFT-B.

Less detailed information is available for IFT-A. Three of its six proteins [IFT144 (also known as WDR19), IFT140 and IFT122] form a stable core (Mukhopadhyay et al., 2010). Four IFT-A proteins are predicted to contain N-terminal WD40 repeats that fold into β-propellers and C-terminal α-helical TPR repeats (Cole, 2003). These WD40 repeats could provide transient protein–protein interactions between IFT and cargo proteins (Cole, 2003). In anterograde trains, IFT-A is close to the membrane (Jordan et al., 2018) and is involved in the import of membrane-bound cargo proteins (Liem et al., 2012; Mukhopadhyay et al., 2010; Picariello et al., 2019). Several other cargoes of IFT have been identified by mutant analysis (Hou et al., 2007; Ishikawa et al., 2014; Taschner and Lorentzen, 2016), raising the question of how IFT proteins organize to associate with diverse cargo proteins.

Many ciliary precursors, such as tubulin (Craft et al., 2015), radial spokes (Lechtreck et al., 2018), outer dynein arms (ODAs) (Dai et al., 2018) and DRC4 (also known as GAS8) (Wren et al., 2013), have been shown to co-migrate with IFT trains through total internal reflection fluorescence (TIRF) microscopy in Chlamydomonas. Although direct demonstration of interactions between cargo and IFT proteins remains sparse, the IFT-B complex is responsible for the import of several axonemal factors (see poster). Some cargoes require additional specific adaptors for their transport, such as ODAs, which are transported by IFT46 via the assembly factor ODA16 (also known as DAW1) (Hou et al., 2007; Ahmed et al., 2008; Dai et al., 2018) and possibly ODA8 (LRRC56 in mammals) (Desai et al., 2015). Similarly, the Chlamydomonas inner dynein arm (IDA) I1 requires IDA3 (Hunter et al., 2018) with IFT56 (TTC26), which is otherwise dispensable for ciliogenesis (Ishikawa et al., 2014). The retrograde motor IFT-dynein is carried on anterograde trains by IFT-B (Pedersen et al., 2006; Jordan et al., 2018), via IFT54 (TRAF3IP1) (Zhu et al., 2020) and IFT172 (Williamson et al., 2012). IFT70 (TTC30A and TTC30B, also known as Fleer in zebrafish and DYF-1 in C. elegans) was proposed to be an adaptor for tubulin polyglutamylation enzymes (Pathak et al., 2007; Yang et al., 2021). Tubulin, the most abundant ciliary protein, can enter cilia by both diffusion and IFT, but requires IFT to support rapid ciliary growth (Craft et al., 2015; Craft Van De Weghe et al., 2020). Two tubulin-binding sites on IFT have been identified that could independently mediate this process – the calponin homology (CH) domain of IFT81 and the IFT74 N-terminal domain form a tubulin-binding module (Bhogaraju et al., 2013), while the IFT54 CH domain binds tubulin directly, but is dispensable for ciliogenesis (Taschner et al., 2016; Kubo et al., 2016; Zhu et al., 2017). In contrast, IFT-A is more important for the transport of membrane proteins, such as components of the sonic hedgehog pathway (Liem et al., 2012) and GPCRs via TULP3 (Mukhopadhyay et al., 2010). The BBSome complex, (discussed in Box 1) is another important cargo and adaptor of the IFT system. However, with the exception of the IFT-dynein on anterograde trains (Jordan et al., 2018), it is not known where all the other cargoes are positioned in the train, and how trains accommodate a large variety of different cargoes. As the cargo load is increased in growing or regenerating cilia (Wren et al., 2013; Craft et al., 2015), the use of such samples together with computational classification of IFT particles from cryo-ET data could help to identify cargoes in the trains.

Anterograde trains are linear assemblies of IFT-A and IFT-B. At the transition fibers of the ciliary base, nascent IFT trains assemble and queue for their release into the cilium (Deane et al., 2001; Wingfield et al., 2017). IFT-B proteins occupy a broad area from the transition fibers to the basal body proximal end, whereas IFT-A proteins colocalize only with the more distal part of the IFT-B region (Brown et al., 2015) (see poster). IFT-B, as the core of the trains, appears to be the first to polymerize, partially from proteins that remained within the basal body pool after their return with retrograde trains (Wingfield et al., 2017). In contrast, IFT-A proteins, which have been seen to move away from the basal body pool, are recruited from the cell body together with the motors to assemble new anterograde trains (Wingfield et al., 2017). Dynein and tubulin cargo associate with new trains only shortly before anterograde IFT begins (Wingfield et al., 2017).

The anterograde train core is made from elongated IFT-B complexes that repeat tightly with a spacing distance of 6 nm (Jordan et al., 2018). This IFT-B core appears to be necessary and sufficient to form anterograde trains together with kinesin-2 motors, whereas IFT-A is mainly expendable for anterograde train formation. IFT-A associates between the membrane and IFT-B (Jordan et al., 2018), consistent with its adaptor role for membrane cargo transport (Liem et al., 2012; Mukhopadhyay et al., 2010). IFT-A has a more globular shape than IFT-B and polymerizes with a repeating distance of 11 nm (Jordan et al., 2018). This raises a mismatch with the 6-nm periodicity of IFT-B, which suggests that they assemble independently before associating to form complete trains at the ciliary base. The mismatched periodicities may also provide a stable yet flexible connection between the two polymers while the train travels through an actively beating cilium. The majority of mature anterograde trains are ∼310 nm long, with some reaching up to 700 nm (Jordan et al., 2018). Thus, most trains contain at least 50 IFT-B complexes. Therefore, an intriguing question for future investigation is what factors limit the length of the trains. In addition, the extremely tight repeat within the trains leaves little space for cargo binding. Known cargo proteins, such as ODAs or radial spokes, are themselves large and comprise quite diverse assemblies that would be expected to interfere with loading at the same IFT-B or -A repeats. This suggests that there is a competitive loading of trains and makes it unclear how cargo diversity can be guaranteed. These questions await structural investigations of IFT–cargo interactions in vitro or eventually of whole trains in situ.

Anterograde IFT trains are powered by the heterotrimeric kinesin-2 (Cole et al., 1998) (see poster). The two motor subunits 2α and 2β dimerize via coiled-coil rods, thereby forming a flexible stalk (Yamazaki et al., 1995; Vukajlovic et al., 2011) that links the motor domains to the train through the kinesin-2-associated protein (KAP; KAP3 or KIFAP3 in mammals) subunit (Doodhi et al., 2009; Mueller et al., 2005). In Chlamydomonas and mouse, kinesin-2 is thought to bind to IFT-B (Liang et al., 2014; Baker et al., 2003; Jordan et al., 2018). An immunoprecipitation study in human cells showed binding of kinesin-2 to the IFT-B tetramer IFT38–IFT57–IFT52–IFT88 (IFT38 is also known as CLUAP1 in mammals) (Funabashi et al., 2018), and recent evidence indicated a role for IFT54 in kinesin binding (Zhu et al., 2020). Additionally, in humans, IFT46 and IFT56 are needed for ciliary entry of the homodimeric kinesin-2 motor (KIF17); however, it appears dispensable for IFT trafficking (Funabashi et al., 2017). Given that kinesin has a step size of 8 nm and includes large head domains, it is unlikely to be associated with every 6-nm repeat of the IFT-B complex in trains. While an IFT train contains ∼50 copies of IFT-B (Jordan et al., 2018), a much lower number of kinesins was indeed estimated (Engel et al., 2009; Laib et al., 2009). EM images from flat-embedded samples suggest two connections between the anterograde train and the microtubules (Stepanek and Pigino, 2016), but the finite position of kinesin has yet to be determined, and approaches such as cryo-ET will be needed to understand how kinesins power the trains.

Anterograde IFT appears more complex in C. elegans than in humans or Chlamydomonas, where a fast homodimeric kinesin-2 (OSM-3) supports the slower heterotrimeric kinesin-2, consisting of KLP-20, KLP-11 and KAP-1 (Snow et al., 2004). The heterotrimeric kinesin-2 is responsible for import and transport at the transition zone. In a ‘handover zone’, IFT trains are then transferred to the more processive OSM-3, which serves as long-range transporter to the ciliary tip along the microtubule singlets of the distal segment (Prevo et al., 2015). In contrast to Chlamydomonas and mouse, C. elegans heterotrimeric kinesin-2 binds to IFT-A, whereas OSM-3 binds to IFT-B (Ou et al., 2005), likely via recruitment by DYF-1 (the homolog of IFT70) (Mohamed et al., 2018). Until recently, it has been impossible to perform in situ cryo-ET studies in C. elegans, therefore precluding the cryo-EM investigation of its sensory cilia that are embedded within a complex tissue structure. Consequently, very little is known about the structural organization of IFT trains and their motors in this system. However, the use of combinations of new technologies, such as cryo-focused ion beam (FIB) lift-out and cryo-ET to visualize molecular assemblies in large frozen samples (Schaffer et al., 2019), may now make such studies possible, opening the door to comparative studies in the coming years.

After reaching the tip, anterograde trains transition into retrograde trains that return to the ciliary base (see poster). In motile cilia, the available space for IFT trains is limited by the number of microtubule doublets and the ODAs that create additional spatial constraints. In Chlamydomonas, although all nine doublets are used by both types of train without bias for either direction, TIRF microscopy shows no collision between the oppositely directed trains (Stepanek and Pigino, 2016). Correlation of these live, time-resolved TIRF microscopy data with 3D-EM images has revealed that anterograde trains move along the B-tubule, whereas retrograde trains use the A-tubule of each microtubule doublet (Stepanek and Pigino, 2016). Thus, in Chlamydomonas, anterograde and retrograde motions are spatially separated on different tracks (see poster). In Trypanosoma, IFT uses only a small subset of the microtubule doublets (Bertiaux et al., 2018), thus confirming that regulation of IFT includes microtubule selectivity. However, the logistics of anterograde and retrograde trains remains unknown in other organisms than Chlamydomonas, such as in mammalian primary cilia and C. elegans sensory cilia, where very long microtubule singlets are present (Kuhns and Blacque, 2016).

It has been proposed that IFT-kinesin motors migrate towards the B-tubule by an intrinsic left-handed helical movement around the microtubules that is sterically blocked by the ODAs and thus results in a straight movement along the B-tubule (Stepp et al., 2017). Reduction of the motor sidestepping behavior in C. elegans did not result in traffic jams (Xie et al., 2020), suggesting that sidestepping is not involved in the positioning of kinesins on the B-tubule in Chlamydomonas or that a different IFT logistics co-evolved with the highly specialized axonemal structure of C. elegans cilia.

Post-translational modifications may also play a role in outlining motor-specific microtubule tracks in axonemes. A- and B-tubules differ in detyrosination (Johnson, 1998) and polyglutamylation levels (Kubo et al., 2010; Orbach and Howard, 2019). In mammalian cilia, rapid deglutamylation of tubulin leads to altered anterograde IFT (Hong et al., 2018), whereas, in Chlamydomonas, IFT activity is unaffected in polyglutamylation mutants (Kubo et al., 2015).

To explain the unperturbed movement of anterograde trains in cilia, another mechanism is needed. Retrograde trains are powered by IFT-dynein, a multiprotein complex that contains two heavy chains, consisting of a tail and a motor domain. The motor domain is formed by an AAA+ ring and a stalk, the tail binds to a variety of different light and intermediate chains (Roberts, 2018; see poster and Table 1). Before dynein can power retrograde transport, it has to be delivered to the tip by anterograde trains (Reck et al., 2016). Bidirectional transport processes that involve the simultaneous presence of oppositely directed motors can result in a tug-of-war when the motors compete for the direction of movement (Hancock, 2014). However, anterograde trains move processively towards the tip, and pausing or reversals of direction are only observed rarely (Dentler, 2005; Chien et al., 2017). To explain this, cryo-ET of native anterograde trains has revealed that IFT-dynein is transported in an auto-inhibited conformation (Jordan et al., 2018) in which its motor domains are stacked and the stalks crossed (Toropova et al., 2017; Jordan et al., 2018). In this conformation the dynein ‘legs’ are tied together, thus preventing the motor from pulling the train in the opposite direction (see poster). Likewise, this conformation has been shown to inhibit walking activity along microtubules in in vitro experiments using artificially dimerized motor domains (Toropova et al., 2017). A similar mechanism might be involved in preventing binding of ODAs to microtubules during anterograde IFT; ODAs isolated from the cytoplasm of Tetrahymena cells are locked in a similarly inhibited crossed-conformation by the protein Shulin (Mali et al., 2021). Since the crossed-conformation of IFT-dynein can still bind to microtubules (Toropova et al., 2017), it could act as a break during anterograde transport. Strikingly, cryo-ET of anterograde trains has revealed that dyneins are positioned on the train such that their microtubule-binding domains are kept away from the microtubule track (Jordan et al., 2018). In this way, the combination of cryo-EM subtomogram averaging and single-particle analyses has made it possible to reveal the elegant inhibition mechanism of IFT dynein.

Retrograde trains have a repeat of 45 nm, which is larger than any repeat seen in anterograde trains (see poster). They appear as broad zig-zag assemblies, and their appearance indicates that a major reassembly of the train structure must take place during turnaround at the tip (Stepanek and Pigino, 2016; Jordan et al., 2018). These structural changes correlate with the switch in function of the train and support the transition to a different motor and the transport of different cargoes. For example, whereas IFT-A does not mediate anterograde transport of IFT-dynein (Pedersen et al., 2006; Jordan et al., 2018), it is thought to change its position to dock active dyneins during retrograde transport (Williamson et al., 2012). Furthermore, in Chlamydomonas, kinesin-2 does not associate with retrograde trains. Instead, after its inactivation by phosphorylation at the tip (Liang et al., 2014), it returns via an IFT-independent mechanism to the cell body, probably by diffusion (Pedersen et al., 2006; Engel et al., 2012; Chien et al., 2017). This is in contrast to what occurs in mammalian cells and C. elegans, where kinesin is actively carried to the base by IFT (Williams et al., 2014; Prevo et al., 2015; Mijalkovic et al., 2017).

In Chlamydomonas, the conversion from anterograde to retrograde trains takes ∼3 s; in this period, the trains ‘rest’ at the tip, remodel and depart as retrograde trains (Chien et al., 2017). In C. elegans, only IFT-B rests at the tip, while IFT-A and dynein change direction without pause (Mijalkovic et al., 2018). Thus, the remodeling process is thought to include disassembly of individual trains and the combination of material from multiple trains to build a new retrograde train (Chien et al., 2017; Mijalkovic et al., 2018). However, no structure of retrograde trains is currently available that could help understand the conformational changes and the new positions that IFT-A, IFT-B and dynein adopt. Retrograde trains are structurally more heterogeneous, which makes subtomogram averaging more difficult than for anterograde trains (Jordan et al., 2018). Additionally, identifying the three major components IFT-A, IFT-B and dynein by mutant analysis, as was done for anterograde trains, is challenging, because lack of IFT-A or dynein compromises retrograde IFT (Picariello et al., 2019; Pigino et al., 2009; Lin et al., 2013), and without IFT-B, anterograde IFT does not occur and cilia are often not even assembled (Richey and Qin, 2012; Hou et al., 2007). Therefore, the localization of the main IFT components in retrograde trains is likely to require a comparison with available protein structures.

During the course of IFT, IFT-dynein changes its functional state by adopting different conformations. Whereas it is auto-inhibited during anterograde transport, it becomes activated at the tip in a controlled way to power retrograde transport. In addition to the crossing of the stalks, the auto-inhibited dynein adopts a kinked conformation of the tail domain that allows association with the tight repeat of IFT-B on the anterograde train (Jordan et al., 2018; Toropova et al., 2019). This conformation is different from the straight-tail conformation of auto-inhibited dynein-1, which occurs in the cell cytoplasm (Zhang et al., 2017). High-resolution single-particle analysis of recombinantly expressed IFT-dynein has shown that the kink is established by a zig-zag conformation of one heavy chain, while the other chain remains straight (Toropova et al., 2019). Since the kinked conformation is also adopted in vitro (Toropova et al., 2019), the contact with IFT-B is not required for its formation and it may rather be a prerequisite for the loading of dyneins on the train.

Once at the ciliary tip, dyneins must be released from the train. Here again, transporting dynein in its auto-inhibited conformation appears to be crucial to avoid released dyneins starting to ‘walk back’ to the cell before they are incorporated into retrograde trains (Jordan et al., 2018). An intermediate ‘open’ conformation of dynein with separated motor domains (Chowdhury et al., 2015; Zhang et al., 2017) has been proposed to play a role in the controlled activation of dynein at the tip (Jordan et al., 2018). However, the factors that are involved in this process are not yet known.

To power retrograde trains, dynein must transition to an active conformation with its heavy chains in a parallel orientation and the stalks able to step along the microtubule, directed towards the minus end, as shown for dynein-1 in the cell cytoplasm (Chowdhury et al., 2015). However, the common dynein-1 regulators LIS1 (also known as PAFAH1B1) and BICD2 do not appear to be associated with IFT-dynein (Asante et al., 2014). Since mutations in IFT-A subunits often result in phenotypes similar to those arising from mutations in the retrograde motor (Piperno et al., 1998; Iomini et al., 2009), IFT-A is thought to dock with IFT-dynein in retrograde trains (Williamson et al., 2012), and it seems plausible that IFT-A binds to the dynein tail and acts in a similar way to known dynein activators (Cianfrocco et al., 2015). However, understanding how dynein is kept active during retrograde transport awaits determination of the retrograde train structure.

Overall, structural analyses have provided valuable insights into various aspects of the IFT system, with most information being available on the anterograde trains. However, many questions that require high-resolution structural information remain open. For instance, how are IFT-A and IFT-B complexes assembled from their subunits? How do anterograde trains interact with their kinesin motor? How do the complexes and their subunits interact with cargoes and other adaptors, such as the BBSome, to load the trains? Which factors tell the train that the tip has been reached? What triggers dynein activation? Which rearrangements are performed during the conversion from anterograde into retrograde trains? How is dynein held in an active conformation? A structure from cryo-ET of native retrograde trains is required to understand how they are formed from anterograde trains and how they are powered by dynein.

Further questions focus on train assembly and entry at the ciliary base. For example, how is the train length controlled? How do IFT trains pass the barrier of the transition zone? It has been shown that RabL2, IFT74 and IFT81 play a role in regulating IFT train entry (Kanie et al., 2017); however, mechanistic and structural insight into train entry is sparse. Analysis of cytoplasmic membrane vesicles has indicated that IFT proteins are also involved in the transport of axonemal and ciliary components from the Golgi to the ciliary base, likely by forming a coatomer, similar to COPI coats (Wood and Rosenbaum, 2014; Quidwai et al., 2020 preprint). Structural insight into the ciliary base and surrounding regions will require cryo-ET of FIB-milled lamellae, which would allow the visualization of structures of the thick ciliary base. Together with high-resolution structures of IFT-A and IFT-B complexes, these data promise to reveal the mechanisms that explain how the large and highly effective IFT machinery assembles the cilium.

We thank Iain K. Patten and Dennis R. Diener for fruitful discussions, comments and corrections to the manuscript.