Cytoplasmic factories for axonemal dynein assembly

Dyneins are highly complex molecular motors that power microtubule-based transport activities in many eukaryotes (Wickstead, 2018). For example, cytoplasmic dyneins traffic vesicular and other cargoes within cells, enable the breakdown of the nuclear envelope, are involved in mitosis and mediate retrograde axonal transport in vertebrates (Hinchcliffe and Vaughan, 2018; Paschal and Vallee, 1987; Pfister, 2000; Salina et al., 2002). An evolutionarily closely related dynein powers retrograde movement of intraflagellar transport (IFT) particles that are required for ciliary assembly, maintenance and signaling (Pazour et al., 1999; Porter et al., 1999; Witman and Hou, 2018), and also mediates trafficking of some membrane components to cilia (Cao et al., 2015). In contrast, axonemal dyneins are incorporated into the microtubular superstructure (the axoneme) of motile cilia (Bell et al., 1979; Pfister et al., 1982) and generate rhythmic beating of the organelle, leading to movement of individual cells and/or the generation of fluid flow in various compartments, cavities and organs (Fig. 1) (Bustamente-Marin and Ostrowski, 2017; Faubel et al., 2016; Meunier and Azimzadeh, 2017). In mammals, defects in ciliary dyneins underlie many ciliopathies – complex syndromes that can affect the development and/or function of multiple organs and cell types (Fliegauf et al., 2007; Reiter and Leroux, 2017).

There are two general ways in which dyneins have been categorized: (1) structurally, based on the number of heavy chain (HC) motor units (Fig. 2A) in the holoenzyme, whereby those with two or three motor units and associated components are distinguished from a subset of monomeric HC motor axonemal dyneins with distinct accessory subunits, and (2) phylogenetically, where cytoplasmic and IFT dynein HCs are considered distinct from those which power motility in the ciliary axoneme (King, 2018; Wickstead, 2018).

Axonemal dyneins are assembled in the cytoplasm and then transported into cilia where they are incorporated into the axonemal superstructure at very precise sites (Ma et al., 2019; Oda et al., 2014; Wakabayashi et al., 2001). The complexity of the axonemal dynein pre-assembly process is clearly illustrated by the observation that ∼20 different cellular factors have been described as essential for formation of these motors in the cytoplasm (Mitchell, 2018). Remarkably, most of these factors appear to be specific to axonemal dynein assembly and have not been reported to markedly affect formation of either canonical cytoplasmic dynein or the closely related dynein that powers retrograde IFT (discussed in Patel-King et al., 2019); this implies that these latter HC motors may have distinct assembly factor(s), even though their general folding requirements are essentially identical to those of axonemal dynein HCs. Intriguingly, although both cytoplasmic and IFT dynein HCs have been expressed and purified from various heterologous systems, such as human embryonic kidney and/or insect cells (for example, Ichikawa et al., 2011; Toropova et al., 2019; Trokter et al., 2012), these cell types normally synthesize endogenous cytoplasmic dynein, and thus presumably already contain any needed assembly factors.

Liquid–liquid phase separation (LLPS; see Box 1 and figure within box) has become a major unifying theme in modern biology aimed at explaining the formation and behavior of self-assembling protein conglomerates or condensates as membrane-less functional units or organelles (Hyman et al., 2014). LLPS occurs through solution de-mixing and often involves proteins containing inherently disordered or low complexity regions. This behavior is exquisitely sensitive to the local concentration of components and also shows dependence on other parameters, such as temperature, pH and ionic strength (Alberti et al., 2019; Brangwynne et al., 2015). Cellular systems suggested to form by this process include nucleoli, nuclear paraspeckles whose formation is directed by long non-coding RNAs, pyrenoids that occur within chloroplasts and contain the CO₂-fixing enzyme Rubisco, cytoplasmic P bodies, which function in mRNA decay and silencing, and many others (see McSwiggen et al., 2019b for a detailed tabulation). Recently, axonemal dynein assembly has also been proposed to occur in membrane-less phase-separated compartments termed dynein axonemal particles (DynAPs), which exhibit liquid-like properties within the cytoplasm of vertebrate multi-ciliated cells (Huizar et al., 2018). However, based on an extensive analysis of the measured parameters that are typically reported to support LLPS as a mechanism for condensate formation in a broad array of cellular systems, it has been recently suggested that the general approaches and procedures often used to define LLPS in vivo may not readily distinguish between LLPS per se and alternative biochemical mechanisms (McSwiggen et al., 2019b).

Box 1. General principals of liquid–liquid phase separation

LLPS refers to the partitioning or de-mixing of solubilized component(s) into two distinct realms resulting in a condensate containing high levels of the component and a bulk phase that is depleted of the component (see Brangwynne et al., 2015; Hyman et al., 2014; McSwiggen et al., 2019b for detailed discussion). This phenomenon is highly concentration dependent and can often be readily altered by varying other conditions of the solution, such as temperature, ionic strength and pH (see figure on the features of LLPS). At low concentrations, the component will be present only in the bulk phase. Condensates occur once a critical concentration is reached leading to two phases of differing component concentration. However, once condensates form, the component concentration in them does not increase further as the total component amount in the system continues to climb. Rather, any subsequent increase of component quantity leads to formation of more and/or larger condensates until, eventually, the entire volume becomes occupied by a single condensate phase (see figure). Thus, a phase diagram of an LLPS system will show two distinct regions in which a component exists in a single-phase or in two-phases depending on the precise solution conditions. LLPS condensates exhibit a variety of liquid-like properties including approximately spherical geometry, fusion and/or fission, and rapid exchange of components both between the condensate and bulk phase and also between different regions within the condensate itself. LLPS is readily observed using various purified components in vitro, but is more difficult to assess in the highly complex and crowded environment of a living cell (McSwiggen et al., 2019b); this is especially the case when tagged proteins are overexpressed, which can potentially affect the critical concentration dependence. In the figure, increasing protein concentration is illustrated by darker green coloration.

Here, I examine the spatial and temporal parameters for axonemal dynein formation in the cytoplasm. I consider the length of the HC mRNAs, the rate at which these very large motor proteins can be synthesized by ribosomes, the size of the polysomes with partially assembled HCs bound to various transiently interacting assembly factors needed for stability and folding, and the potential for inter-polysome associations via partially synthesized dynein subunits during co-translational folding and assembly. These parameters support the concept that dynein synthesis might involve self-organization into large ‘conglomerates’ or ‘factories’ in a manner that is independent of a change in phase. Indeed, the formation of dynein factory compartments may be a direct and immediate consequence of the sheer scale of the assembly process itself, and, as argued here, thus perhaps provides an alternate model to the LLPS hypothesis suggested previously to explain DynAP formation and behavior.

The ciliary axoneme consists of nine outer doublet microtubules surrounding a central pair of singlet microtubules; individual doublets are interconnected by the multi-subunit nexin–dynein regulatory complex. (Fig. 1A, right insert). Axonemal dyneins are arranged in two rows along the doublet microtubule length, forming the inner and outer dynein arm systems (Fig. 1A, left insert) that generate interdoublet microtubule sliding, which underlies ciliary bend formation and propagation. Dyneins are multi-component enzymes built around the HC motor units, which have molecular masses in the 460–650 kDa range; although most dynein HCs are ∼530 kDa and contain ∼4500 residues, this wider range reflects the considerable variation within a subset of monomeric inner-arm HC motors (King et al., 2021).

Dynein HCs consist of an N-terminal region involved in HC–HC interactions, association with other dynein components and binding to cargoes, a linker that spans the plane of a six-membered ring of AAA+ domains (AAA1 to AAA6) and a microtubule-binding domain located at the tip of an anti-parallel coiled coil that emanates from AAA4 and is supported by a second coiled-coil buttress or strut derived from AAA5 (Burgess et al., 2003; Carter et al., 2011; Kon et al., 2012) (Fig. 2A). Many dyneins also have an additional C-terminal domain that forms a cap partially covering one face of the AAA+ ring and which appears to regulate processivity and force generation (Nicholas et al., 2015b). Changes driven by ATP binding, hydrolysis and product release at AAA1 lead to alterations in linker conformation, providing a power stroke, combined with a change in register of the antiparallel coiled coil that results in transitions between high-affinity microtubule-binding and release (Carter et al., 2008; Kon et al., 2005; Redwine et al., 2012; Roberts et al., 2009). Nucleotide binding at several other AAA+ domains appears to impart regulatory control over conformational changes propagating through the AAA+ ring (Dewitt et al., 2015; Kon et al., 2004; Nicholas et al., 2015a; Schmidt and Carter, 2018). Dyneins with two or more motor units also contain a core subcomplex formed from two WD-repeat intermediate chains (ICs) and light chain (LC) dimers of three distinct classes (King, 2018); outer arms also associate with a docking complex needed for precise axonemal assembly. Furthermore, individual dyneins often have distinct additional subunits based on their function and mode of regulation. For example, Chlamydomonas axonemal outer arm dyneins have several different LCs involved in responses to Ca²⁺ and redox changes, as well as microtubule–motor interactions (Benashski et al., 1999; Ichikawa et al., 2015; King and Patel-King, 1995; Patel-King et al., 1996; Toda et al., 2020; Wakabayashi and King, 2006).

Taking the outer dynein arm from Chlamydomonas as a well-studied example and incorporating key studies from other (often vertebrate) systems allows the general scale of both the structural components of the holoenzyme and the factors required for their assembly to be appreciated.

This fully assembled Chlamydomonas dynein (see Fig. 2) contains three ∼530 kDa HCs (α, β and γ); the α- and β-HCs have thioredoxin-like LCs (LC5 and LC3, respectively) associated with their N-terminal regions, while the γ-HC has an N-terminal-associated calmodulin homolog (LC4) and a leucine-rich repeat protein (LC1) bound to the microtubule-binding domain (MTBD) (King, 2018). The β- and γ-HCs interact with each other via their N-terminal domains. They also both associate with an IC–LC complex consisting of two WD-repeat ICs (IC1 and IC2), dimers of three distinct LC types – the LC8 class also known as DYNLL1 and DYNLL2 in mammals (formed from the related isoforms LC6, LC8 and LC10), the Tctex1 class (an LC2–LC9 heterodimer) and the LC7/Roadblock class (an apparent LC7a–LC7b heterodimer) (King, 2018; Walton et al., 2021). This multimeric assembly forms a dynein core that is homologous to the two-motor unit outer dynein arms of vertebrates. The α-HC and its associated LC5 interact with the β-HC of this core dynein complex via an unusual N-terminal β-propeller domain consisting of kelch repeats and two immunoglobulin-like domains (King, 2018; Mali et al., 2021; Rao et al., 2021 preprint; Walton et al., 2021). Furthermore, another integral sub-assembly – the docking complex – which is essential for the incorporation of this dynein into the axonemal superstructure in vivo and stable association of the αβ- and γ-HC subunits in vitro, is formed from two coiled-coil proteins (DC1 and DC2) and a redox-sensitive calmodulin homologue (DC3) (Casey et al., 2003; Owa et al., 2014; Takada and Kamiya, 1994). In total, the complete outer arm has a mass of ∼2 MDa (Fig. 2B).

An important aspect of axonemal dynein formation is the coordinated expression of individual components such that they are made in the appropriate stoichiometry. This feature likely has its roots in co-transcriptional regulation of dynein gene expression. Indeed, in Chlamydomonas, ciliary component expression is upregulated following deciliation in a synchronized manner (Lefebvre and Rosenbaum, 1986). Furthermore, Chlamydomonas cells (and presumably others) apparently monitor dynein protein levels to achieve an appropriate component balance. For example, when a tagged gene for the Chlamydomonas LC1 protein (DNAL1 in mammals) was inserted into a wild-type background, the resulting cells (with both endogenous and tagged LC1 gene copies under control of native promoters) coordinated LC1 expression such that the normal total protein amount was generated, consisting of approximately equal quantities of both tagged and untagged proteins (Patel-King and King, 2009). Given this temporal coordination, it seems reasonable that protein expression also may be synchronized spatially such that individual components can readily associate to form the final motor product.

Depending on the organism, ∼20 different cytoplasmic factors have been identified as key to axonemal dynein assembly (Mitchell, 2018; Patel-King et al., 2019), including the AAA+ ATPase proteins RUVBL1 and RUVBL2 (pontin and reptin), which form hetero-hexamers that can dimerize to dodecamers (Gorynia et al., 2011). These complexes bind various scaffold [RNA polymerase II associated protein 3 (RPAP3) or Sperm antigen 1 (SPAG1)] and PIH (protein interacting with heat shock protein HSP90; PIH1D1, PIH1D2 or DNAAF2) proteins, forming the R2TP complex (consisting of RUVBL1 and RUVBL2, PIH1D1 and RPAP3) (Kakihara and Houry, 2012), and its variants, R2SP (RUVBL1 and RUVBL2 plus PIH1D2 and SPAG1) and R2SD (RUVBL1 and RUVBL2 plus DNAAF2 and SPAG1) that recruit HSP90 co-chaperones and are needed for formation of distinct groups of dynein HCs (Maurizy et al., 2018; Yamaguchi et al., 2018). In addition, these co-chaperone recruitment complexes associate with WDR92 (also known as DNAAF10), which is thought to target a prefoldin-like complex to dynein assembly sites (Liu et al., 2019; Patel-King et al., 2019; zur Lage et al., 2018). It remains somewhat unclear why certain dynein HCs need different scaffolds and chaperone recruitment proteins for their formation, but potentially these factors act at distinct steps as folding proceeds (Yamamoto et al., 2020); indeed the assembly factor ZMYND10 (DNAAF7) has been shown to act in a sequential chaperone relay (Mali et al., 2018). Additional proteins that participate and play key roles in this process contain a variety of protein–protein interaction domains including WD-repeat β-propellers, tetratricopeptide repeats, leucine-rich repeats, HEAT repeats, coiled coils and MYND zinc finger domains (see Mitchell, 2018 for a detailed review). Several additional factors are involved in transport and functional control of fully formed dyneins, such as the outer-arm IFT adaptor ODA16 (encoded by DAW1) (Ahmed et al., 2008) and shulin (DNAAF9), which binds to assembled outer arms to inhibit their motor activity prior to axonemal insertion (Mali et al., 2021). In many cases, the stoichiometry of these factors per HC, and whether this changes as HC formation proceeds, remains uncertain. Even so, the mass they add to each growing HC during synthesis is likely quite considerable – for example, a single RUVBL1/RUVBL2 dodecamer has a mass of ∼0.5 MDa.

For simplicity, I consider here only the HCs as they account for much of the total mass of a dynein, ranging from ∼75% for outer dynein arms to almost 90% for the monomeric motor inner dynein arms. The inner and outer dynein arms are arranged along the length of the ciliary doublet microtubules in a characteristic 96 nm repeating pattern, which consists of four outer dynein arms spaced 24 nm apart and an inner arm system containing one inner dynein arm I1/f followed by six different monomeric HC dyneins (Bui et al., 2012; Heuser et al., 2012; Nicastro et al., 2006). Each 96 nm-length portion of the cilium requires 168 HCs; for two 10-µm axonemes, this amounts to 35,000 HCs (for calculations see Box 2). In addition, Chlamydomonas cells maintain a pool of axonemal components sufficient to rebuild two approximately half-length cilia in the absence of further protein synthesis (Rosenbaum et al., 1969). Thus, the total number of HCs synthesized per Chlamydomonas cell approaches 52,500, which is equivalent to ∼28 GDa of protein. Vertebrate multiciliated cells assemble as many as 300 cilia (Meunier and Azimzadeh, 2017) each containing ∼15,000 HCs, indicating that every cell needs to synthesize ∼4.5 million HCs with a total mass over 2 TDa; this represents a very considerable biosynthetic burden (upwards of one million ribosome-hours of synthetic activity; see Box 2).

Given the massive size of dynein motors and the large number needed in a motile cilium, it is helpful to consider the spatial and temporal scales of the assembly process itself. In Chlamydomonas, dynein HCs are each encoded by a >20 kb gene yielding a spliced mRNA of ∼15 kb (mRNAs for several of the minor monomeric dynein HCs will be several kb longer or shorter). As each nucleotide adds ∼0.33 nm (Phillips et al., 2009), this leads to a total unfolded HC mRNA length of ∼5 µm (Box 2). Clearly, these will not exist as single linear extended structures, but rather will undergo folding to form complex local secondary (e.g. stem-loops and pseudoknots) and perhaps tertiary organizations that likely change as ribosomes transit along their length. Indeed, modeling the folding of small regions of the outer arm β-HC mRNA (the entire mRNA is considerably beyond the allowable limits set by the RNAstructure prediction server; https://rna.urmc.rochester.edu/RNAstructureWeb/Servers/Predict1/Predict1.html) revealed extensive, high-probability local folding.

How many ribosomes simultaneously bind along a 5-µm long mRNA will depend on multiple factors, including some that are intrinsic to the particular message, and may also be limited by local ribosome availability and steric clashes between polysome-associated partially assembled HCs, other tightly associated dynein components and transiently interacting cytoplasmic assembly factors. Based on average ribosome density on mRNAs (Hendrickson et al., 2009), the inverse relationship between mRNA length and ribosome density (Arava et al., 2003) and modeling that indicates this latter feature becomes less impactful as ribosome availability increases (Fernandes et al., 2017), a reasonably conservative estimate might be one ribosome per 100–200 nm of mRNA, leading to perhaps 25–50 ribosomes per HC mRNA (see Box 2).

Eukaryotic ribosomes incorporate new amino acids into a growing polypeptide chain at the rate of ∼5 residues/s (Lodish et al., 1995). Thus, it may take ∼15 min for a ribosome to fully assemble a single 530 kDa HC consisting of ∼4500 residues. If there are 50 ribosomes arrayed equally spaced along an mRNA, the total mass of the mRNA–ribosome–HC complex (ignoring specific assembly factors) will be ∼237 MDa, which would occupy a volume of ∼2.78×10⁻⁴ µm³ assuming that all components are packed tightly together, and clearly represents an absolute lower estimate (see Box 2 for calculations).

Following deciliation, a Chlamydomonas cell will regrow two new cilia and replenish cytoplasmic pools of axonemal components in ∼90 min (Lefebvre et al., 1978); to achieve this, the cell rapidly upregulates transcription of ciliary genes by 10-fold or more (Lefebvre and Rosenbaum, 1986; Stolc et al., 2005). Although the absolute expression levels of dynein HC genes in Chlamydomonas under these conditions is uncertain, this process likely occurs rapidly as eukaryotic RNA polymerase II can synthesize mRNA at rates in the 2–4 kb/min range depending on the organism and the particular mRNA (Ardehali and Lis, 2009). Based on the predicted ribosome density and the rates of polypeptide formation, in the 90 mins taken for complete ciliary assembly, an individual mRNA might template ∼300 HCs, suggesting that ∼160 HC mRNAs (∼10 for each inner arm HC and ∼33 for each outer arm HC) are needed by a single cell (see Box 2 for calculations). The requirement in vertebrate multiciliated cells will be much greater – perhaps by a factor of ∼50- to 100-fold or more – further emphasizing the large amount of cellular synthetic machinery that needs be dedicated to axonemal dynein formation.

The hypothesis that dynein assembly in the cytoplasm of vertebrate multiciliated cells occurs in compartments (DynAPs) generated by LLPS (Huizar et al., 2018) is predicated on several key observations in experiments tracking the behavior of GFP-tagged dynein assembly factors and integral dynein components. Importantly, the dynein subunits localized within DynAPs were either outer arm or inner arm I1/f ICs, or a LC present in a subset of monomeric inner arms (Huizar et al., 2018; Lee et al., 2020); to date, there is no information on the location of HCs within these structures. Direct observation has revealed that DynAPs are dynamic and undergo fission and fusion on a timescale of a few minutes. In addition, fluorescence recovery after photobleaching (FRAP) methods have demonstrated that tagged dynein assembly factors exchanged between the condensate and bulk cytoplasm on a timescale of a few seconds, when fluorescence in the entire condensate was bleached. Similarly, rapid intra-DynAP movement and fluorescence recovery was also observed when only part of the condensate was bleached (Huizar et al., 2018). A subsequent study demonstrated that although cytoplasmic assembly factors are present throughout DynAPs, integral components specific to individual dyneins (such as outer and inner dynein arm ICs – see Fig. 2B) show a more restricted, and in some cases non-overlapping, distribution within individual compartments (Lee et al., 2020). As component concentration throughout a liquid phase is usually constant, this observation suggests that integral dynein components are spatially confined and thus may not be in a ‘liquid-like’ state. This study also proposed the alternative possibility that instead of acting as assembly compartments, these dynein-rich structures might represent stores of fully formed enzymes that are ‘stockpiled’ for axonemal incorporation later on (Lee et al., 2020). Although a clear possibility, this concept is seemingly at odds with the presence of cytoplasmic assembly factors, whose function is to recruit and/or scaffold chaperones and other components needed for protein folding, throughout DynAPs (Huizar et al., 2018; Lee et al., 2020). Furthermore, in the past, assembled axonemal dyneins have been purified from numerous sources (including a broad array of vertebrates and invertebrates, as well as ciliates and green algae; see Inaba, 2018 for a review) under various solution conditions and in a wide range of concentrations, but have never been reported to exhibit phase separation in vitro, suggesting other factors are likely needed for the formation of these cytosolic entities.

Although LLPS is readily demonstrated and confirmed in vitro, recently, concern has been raised that the general methods used to define LLPS systems in vivo (including the ‘gold standard’ FRAP analysis) do not readily distinguish between LLPS and other possible mechanisms (McSwiggen et al., 2019b); this is not to imply that LLPS-generated condensates do not exist in vivo, but rather that in many cases the evidence for LLPS as a driving formative mechanism is not definitive. Much of the concern surrounds the generally descriptive, rather than quantitative, nature of the evidence provided for LLPS (e.g. roundness and fission/fusion properties of the individual compartments), overexpression of tagged components that might disrupt the key concentration dependence, as well as interpretation of FRAP data, which, as has been argued, is not a true test of liquid-like properties (McSwiggen et al., 2019b). For example, structures that had once been considered to form by this physicochemical process include nucleoli (Brangwynne et al., 2011) and the nuclear replication compartments that occur following Herpes virus infection of mammalian cells (Taylor et al., 2003). However, although exhibiting certain features predicted for LLPS, other studies support the involvement of different mechanisms in the formation of these structures, such as non-specific, transient protein–nucleic acid interactions (Mao et al., 2011; McSwiggen et al., 2019a; Shevtsov and Dundr, 2011). Thus, for axonemal dynein assembly the question becomes whether this process indeed occurs in cytoplasmic condensates that are formed by LLPS, or if there might be alternative mechanisms or considerations that modify or replace the general LLPS concept for dynein assembly.

As dynein HCs have a motor unit that consists of a six-membered ring of AAA+ domains, they cannot form a stabilized structure until the terminal AAA6 domain has been synthesized to complete the ring. Even so, partially synthesized HCs would still have stable N-terminal regions that could mediate HC–HC interactions and associations with the IC–LC and/or docking complexes and regulatory LCs. In addition, the MTBD-associated LC could also bind once that structure is formed mid-way through HC synthesis (Fig. 3A). For example, partially synthesized N-terminal β- and γ-HC segments and/or the α-HC β-propeller and the β-HC N-terminal region might associate, thereby bringing into close apposition their individual large polysome structures, and thus essentially crosslinking them together for periods of up to 10 min or more as the C-terminal HC motor domains are synthesized. Indeed, evidence for the colocalization of four different HC mRNAs within ribonucleoprotein granules has been obtained in Drosophila premeiotic spermatocytes by fluorescent in situ hybridization (Fingerhut and Yamashita, 2020). Similarly, once associated, these oligomeric regions would provide high-affinity binding sites for the ∼200-kDa IC–LC complex, individual LCs and potentially the docking complex as well. It is quite conceivable, given the long HC mRNA length, that multiple, for example, γ-HC polysome complexes might associate with one or more β-HC polysome complexes and so on, thereby forming an intricately interconnected multivalent network, cross-linked by partially synthesized components; this would dramatically increase the size of the dynein factory with each incorporated polysome adding almost 0.25 TDa (Fig. 3B). These general considerations of size and multivalent interactions might also explain the fascinating observation that subunits of different dyneins can occupy distinct locations within DynAPs (Lee et al., 2020). This might occur due to the association of different HC polysomes via shared assembly scaffolds or other factors, or possibly through simple physical entanglement of these enormous linear arrays.

Can this concept explain the observations and time scales concerning the dynamic behavior of DynAPs discussed above in the cytoplasm of multiciliated cells without invoking LLPS? A key underpinning of LLPS is the fission and fusion of condensates, and DynAPs clearly undergo both transitions on a time frame of several minutes (Huizar et al., 2018). However, if the underlying reason for their formation is crosslinking between HC polysomes mediated by partially synthesized HCs as outlined above, ‘fusion’ might occur if, for example, a γ-HC polysome contacted and thus cross-linked to a β HC polysome; potentially a single high-affinity HC-HC interaction might suffice to mediate this. In contrast, ‘fission’ might be facilitated by the completion of HC synthesis and thus the release of fully formed dyneins, with the resulting loss of key crosslinks keeping the polysomes in contact. Thus, both events might occur through, and be driven by, the rate of HC synthesis and the high affinity of HC–HC associations. Alternatively, as polysomes for different dyneins may occupy distinct regions of the same DynAP potentially kept together by assembly factors used in common or physical entanglement, fission might result from dynamic changes in these parameters as synthesis proceeds. As observed for LLPS condensates in vitro, another important feature of an LLPS system is the ‘roundness’ or aspect ratio of the condensate image due to liquid surface tension effects (McSwiggen et al., 2019b); membrane-less compartments with a round aspect ratio, such as the chloroplast-located pyrenoid (Fig. 1A) (He et al., 2020), are also observed in vivo. Although no measures of this parameter have yet been described for DynAPs, published DynAP images (e.g. Huizar et al., 2018; Lee et al., 2020) reveal a variety of condensate structures with various protrusions that do not readily provide support for an approximately spherical organization rather than one that is more topologically complex.

A major parameter used in the literature to support in vivo LLPS is FRAP, whereby fully or partially photo-bleached structures are followed as they recover fluorescence due to exchange of bleached components with unbleached ones, either from bulk cytoplasm or from elsewhere in the condensate. However, it is uncertain why this behavior should necessarily invoke or derive exclusively from LLPS, as any factor loosely associated with a large scaffold might be expected to behave in a similar manner. In the case of DynAPs, GFP-tagged cytoplasmic assembly factors, such as DNAAF2, LRRC6, DNAAF3, DNAAF4 and RUVBL2 do undergo FRAP (Huizar et al., 2018). However, it is likely that these cytoplasmic factors only weakly bind to partially folded proteins as their association must dynamically respond as dynein synthesis and assembly proceeds. These factors do not readily co-purify with dyneins during cytoplasmic fractionations, providing support for the generally low affinity and/or transient nature of the interactions (see, for example, Omran et al., 2008; Patel-King et al., 2019; Yamamoto et al., 2017); one well-characterized exception is shulin, which binds fully assembled motors (Mali et al., 2021). In contrast, tagged structural components of dynein, such as the outer arm ICs (DNAI1 and DNAI2) or the monomeric inner arm dynein-associated protein DNALI1 show little evidence for rapid FRAP over a 30-s timescale (Huizar et al., 2018). This might be expected as these are integral dynein subunits that will be tightly and stably bound to the N-terminal regions of semi-synthesized HCs; indeed, in several systems (e.g. Chlamydomonas and sea urchin outer arms), dissociating the IC–LC complex from the HCs in vitro requires extensive biochemical treatments aimed at disrupting strong hydrophobic interactions (Pfister and Witman, 1984; Tang et al., 1982). As such, any bleached ICs would only leave the dynein factory once HC synthesis was complete, which might take another ten or more minutes. Therefore, their fluorescence recovery would require additional ribosome loading and partial HC synthesis to provide new binding sites and/or new synthesis of GFP-tagged dynein ICs if those polysomes were also incorporated within the factory.

Loss of a dynein assembly factor (Heatr2) also leads to changes in DynAP behavior and altered dynamics of GFP-tagged DNAAF2 (Huizar et al., 2018). Specifically, the number of DynAPs per cell was reduced, as was the intensity and mobility of DNAAF2, and this has been suggested to provide further evidence for liquid-like properties (Huizar et al., 2018). However, this observation might equally result from HC aggregation due to the failure of protein folding in the absence of this assembly factor. This in turn might lead to ribosome stalling and altered association dynamics of DNAAF2 with misfolded HCs within the assembly factory.

In summary, axonemal dynein assembly appears to occur in cytoplasmic factories (DynAPs) that exhibit some features previously ascribed to LLPS (Huizar et al., 2018; Lee et al., 2020). However, it is now thought that some key experimental standards and approaches used to support the existence of LLPS in vivo may not actually directly test for this physicochemical process (McSwiggen et al., 2019b). Consideration of the features of cytoplasmic factories for axonemal dyneins and the enormous scale of the process suggests that the observed segregation of dynein assembly into discrete membrane-less cytoplasmic structures might be a direct consequence of the number and size of dynein HC mRNAs and the time needed to synthesize different dynein subunits. When combined with high-affinity interactions linking partially assembled HCs and/or associations mediated by cytoplasmic assembly factors used in common for folding and/or stability of different nascent HCs, this could yield enormous, self-organized complexes containing mRNA, ribosomes, assembly factors and partially synthesized dynein subunits, which, however, are distinct from systems generated by LLPS.

Further understanding the axonemal dynein assembly process will require determining whether translationally active HC mRNAs drive DynAP formation, how partially assembled HC motors behave within these cytosolic compartments and whether their translation is coordinated both spatially and temporally to allow intra-dynein associations to occur during the long synthetic process. One way to examine this might be through expression of multiple HCs tagged with different fluorescent proteins within the same cell driven by the native promoters to ensure normal expression levels. This is potentially quite challenging given the size of mammalian dynein HC genes (e.g. the human DHAH11 gene is 359 kb). However, addressing assembly and co-translation really only requires the native promoter followed by the sequence for a fluorescent marker protein, and the N-terminal HC domain involved in HC–HC and HC–IC interactions. Expression of such tagged HC constructs would allow determination of whether they colocalize with other components of the same dynein within cytoplasmic compartments. Furthermore, if tagged with appropriate fluorescent donor–acceptor pairs, in vivo fluorescence resonance energy transfer approaches (Hirata et al., 2012) potentially could assess when newly synthesized HCs come into close contact during the dynein assembly process.

The micrographs shown in Fig. 1B were obtained by Drs Dhivya Kumar and Panteleimon Rompolas. I also thank Dr Win Sale (Emory University School of Medicine) for comments on a preliminary draft of this manuscript, and an anonymous reviewer for insightful suggestions.