Inherently Disordered Regions of Axonemal Dynein Assembly Factors

Abstract

The dynein-driven beating of cilia is required to move individual cells and to generate fluid flow across surfaces and within cavities. These motor enzymes are highly complex and can contain upwards of twenty different protein components with a total mass approaching 2 MDa. The dynein heavy chains are enormous proteins consisting of ~4,500 residues and ribosomes take approximately fifteen minutes to synthesize one. Studies in a broad array of organisms ranging from the green alga Chlamydomonas to humans has identified nineteen cytosolic factors (DNAAFs) that are needed to specifically build axonemal dyneins; defects in many of these proteins lead to primary ciliary dyskinesia in mammals which can result in infertility, severe bronchial problems, and situs inversus. How all these factors cooperate in a spatially and temporally regulated manner to promote dynein assembly in cytoplasm remains very uncertain. These DNAAFs contain a variety of well-folded domains many of which provide protein interaction surfaces. However, many also exhibit large regions that are predicted to be inherently disordered. Here I discuss the nature of these unstructured segments, their predicted propensity for driving protein phase separation, and their potential for adopting more defined conformations during the dynein assembly process.

Introduction

Cilia are membrane-bound organelles built on a core of microtubule doublets that protrude from the cell surface. These structures are present in most eukaryotes with some notable exceptions such as angiosperm plants and red algae (Carvalho-Santos et al., 2011; Wickstead, 2018) and in many systems have become highly modified to accomplish dedicated tasks (Falk et al., 2015). Cilia can act as sensory structures detecting discrete signaling molecules in the environment (Singla and Reiter, 2006) and also as a platform for the regulated processing and secretion of bioactive products (Luxmi et al., 2019; Luxmi et al., 2022; Wang et al., 2014; Wood et al., 2013). In addition, many cilia contain a highly conserved and complex motile machine - the axoneme (Figures 1a and b)- capable of rhythmic beating and which can propel individual cells such as sperm and a myriad of protozoans or generate fluid flow across surfaces, e.g., moving mucus in the airways. This motile behavior is driven by the action of the inner and outer rows of dynein arms that are arranged in a complex repeating pattern along the axonemal outer doublet microtubules (Lin and Nicastro, 2018; Ma et al., 2019; Oda et al., 2014; Wakabayashi et al., 2001).

Figure 1. Organization of Motile Cilia and the Outer Dynein Arm.

Longitudinal (a) and transverse (b) electron micrographs of cilia from the chlorophyte alga Chlamydomonas illustrating the overall axonemal organization that has been highly conserved throughout phylogeny. Bars represent 250 and 100 nm in panels a and b, respectively. Abbreviations: BB, basal body; CP, central pair microtubule complex; dMT, doublet microtubule; IDA/N-DRC, inner dynein arms and nexin-dynein regulatory complex; ODA, outer dynein arms; TZ, transition zone. c) Ribbon diagram colored by protein component of a cryo-electron microscopy reconstruction of the outer dynein arm from the alveolate Tetrahymena (PDB 7K5B; (Rao et al., 2021). The three HCs are colored magenta, light blue, and light green; each contains a ring of six different AAA+ domains. The microtubule binding domains (MTBDs) are located at the end of antiparallel coiled coils that derive from the AAA4 domains. The IC/LC complex associates with the HC N-terminal domains and is involved in cargo attachment to the A-tubule of the outer doublets via a multimeric docking complex (not shown).

Dyneins generate work by coupling ATP binding, hydrolysis and product release to inter-doublet microtubule sliding which is subsequently converted to a bending motion by other axoneme-associated structures. In order to generate discrete waveforms, regions of dynein activity must propagate along the ciliary length and switch between different subsets of microtubule doublets to yield principal and reverse bends (see (King and Sale, 2018) for a brief historical perspective). Both these activities must occur at a rate consistent with the ciliary beat frequency that in some organisms can exceed 100 Hz. Several axonemal substructures including the central pair microtubule complex and radial spokes, are thought to play a role in modulating dynein-driven waveforms and mutants lacking these structures are normally immotile. However, it is clear from the rescue of motility in these otherwise paralyzed mutants by suppressor mutations in several integral dynein components (Porter et al., 1994; Rupp et al., 1996) and by the induction of ciliary beating in these same mutant cilia under high hydrostatic pressure (Yagi and Nishiyama, 2020) that they are not absolutely fundamental to the axonemal beating mechanism.

Axonemal dyneins are organized in two general patterns (King, 2018; King et al., 2023). The outer dynein arms and inner arm I1/f consist of two (or for outer arms in some non-metazoan species three) heavy chain (HC) motor units (Figure 1c). These each contain an N-terminal region involved in motor assembly followed by a linker domain that moves across the face of a hexameric ring of AAA+ domains. The microtubule binding domain (MTBD) is located at the end of an antiparallel coiled that derives from AAA4 and is supported by a second coiled coil from AAA5. Motor function is driven by ATP binding and hydrolysis at the AAA1 domain, and its actions and interconnectivity with the MTBD modified by nucleotide binding to AAA2-AAA4. These motor units associate with a series of WD-repeat containing intermediate chains (ICs) and multiple light chain (LC) components. Depending on the source, a fully formed outer arm dynein can have as many as twenty different protein subunits and a mass approaching 2 MDa. The second dynein organizational pattern is found in a series of inner arms that consist of single HC motors associated with a molecule of actin and either a dynein-specific LC or the Ca²⁺ binding protein centrin (King et al., 2023). Outer arms occur with a spacing of 24 nm along the entire axonemal length, while the inner arms are arranged in a pattern that repeats every 96 nm and consists of one copy of inner arm I1/f followed by six different monomeric dyneins arranged in pairs to either side of the three radial spokes which protrude towards the central microtubule complex (Lin and Nicastro, 2018). There are also several minor monomeric inner arm dyneins that occur in very restricted regions of the axoneme (Yagi et al., 2009).

Both inner and outer dynein arms are synthesized and assembled in the cytoplasm, and the fully formed motors then trafficked in an inactive state (Mali et al., 2021) into the growing cilium where they are docked at precise locations (Owa et al., 2014). This cytoplasmic assembly process is highly complex and time consuming (King, 2021). Although these HCs have a modular organization, they cannot form a fully folded structure until the sixth AAA domain has been synthesized thereby completing the hexameric AAA ring. Furthermore, it is likely that partially synthesized HCs can assemble with their binding partners once the interacting regions have been built. The complexity of this process is illustrated by the observation that to date nineteen specific cytosolic factors have been identified as necessary for axonemal dynein assembly, in addition to all the ribosomal, chaperone and prefoldin proteins that are generally involved in protein synthesis and folding (see (Braschi et al., 2022; Mitchell, 2018) for recent reviews).

In mammals, mutations in the axonemal dynein subunits themselves or in the cytosolic assembly factors lead to the syndrome primary ciliary dyskinesia (PCD) (Fliegauf et al., 2007; Mirra et al., 2017; Reiter and Leroux, 2017). This manifests as a variable collection of phenotypes (Shoemark and Harman, 2021) including male and female infertility due to poor sperm motility and reduced or absent fluid flow in the Fallopian tubes. Furthermore, PCD patients often have severe bronchial problems as mucus in the lungs cannot be cleared – neonates exhibit respiratory distress. Other phenotypes include nasal congestion, recurrent ear infections, and approximately half of all PCD patients exhibit situs inversus with rearranged internal organs due to disrupted fluid flow at the embryonic node and the consequent failure to set up the correct left-right body axis during development. Defective ciliary motility has also been linked to hydrocephalus (Wallmeier et al., 2019), various forms of epilepsy (Faubel et al., 2022; King, 2006) and to congenital heart abnormalities (Li et al., 2015).

Sequence analysis and structure predictions suggest that many of the axonemal dynein assembly factors contain protein interaction domains associated with, in some cases, large regions that are inherently disordered. Here I discuss the assembly of axonemal dyneins in cytoplasm and examine the properties of these inherently disordered regions (IDRs) which reveals clear differences in overall charge, composition, and propensity for phase separation suggesting they exhibit defined and distinct roles in the complex process of axonemal dynein formation.

The Scale and Biosynthetic Burden of Axonemal Dynein Assembly

Building each 10-μm long cilium requires about 10,000 dynein arms containing ~15,000 HCs that have an approximate mass of 8 GDa (King, 2021). Each HC consists of ~4,500 residues and thus a ribosome will take nearly fifteen minutes at a rate of ~5 residues/sec to assemble one. During this time, the nascent HC must be kept soluble and associations with other components of the holoenzyme directed. As a HC mRNA is ~15 kb or ~5-μm in length, it can likely accommodate perhaps fifty or more ribosomes each with a growing HC of varying lengths. Furthermore, HC-HC and HC-IC/LC complex interaction sites are formed early in HC synthesis potentially allowing for multivalent crosslinking of different HC polysomes to yield enormous cellular structures aided and maintained by the plethora of assembly factors. Several studies have identified DNAAFs that are needed at particular times in the dynein assembly process e.g., (Mali et al., 2018; Yamamoto et al., 2020). Importantly though, with the exception of DNAAF9 (shulin) (Mali et al., 2021) there is currently little experimental data on the required stoichiometry of these factors per HC or per dynein particle and on how or whether this might change as HC synthesis proceeds. In mammals, multiciliated cells can contain upwards of three hundred cilia (Meunier and Azimzadeh, 2017). Thus, the biosynthetic burden placed on these cells is enormous requiring about one and a quarter million hours of ribosomal synthetic activity to build just the ~2.4 TDa of HCs needed to generate motility (King, 2021).

Multiple Cytosolic Factors Needed for Axonemal Dynein Assembly Contain Disordered Regions

Studies in a broad array of organisms ranging from the green alga Chlamydomonas to humans has so far identified nineteen cytoplasmic components necessary for the cytosolic preassembly of axonemal dyneins (Table 1) (Braschi et al., 2022; Mitchell, 2018). Our recent analysis of the N-terminal processing and modification of dynein proteins indicates that a methionine aminopeptidase and at least two distinct multi-component N-terminal acetyl transferase complexes are also necessary to generate dynein holoenzymes (Sakato-Antoku, Balsbaugh and King, unpublished). Furthermore, one might anticipate that additional, as yet unidentified, factors perhaps involved in dynein-specific protein quality control may also occur. Somewhat surprisingly, none of these cytosolic factors appear needed to build either canonical cytoplasmic dynein or the closely related motor that powers retrograde intraflagellar transport (IFT), even though the time scales, complexity, and folding requirements for generating these motors are essentially identical to those of axonemal dyneins. Indeed, several organisms that assemble canonical cytoplasmic dynein completely lack the genes encoding these assembly factors (e.g., budding yeast and nematodes), and to date no axonemal dynein assembly factor mutation has been reported to cause a phenotype that might be ascribed to cytoplasmic or IFT dynein dysfunction.

Table 1.

Human Axonemal Dynein Assembly Factors (DNAAFs)^†

DNAAF #	Aliases†	UniProt ID	Residues (n)	Mass (Da)	pI	Properties*
DNAAF1	LRRC50 ODA7/DAU1	Q8NEP3	725	80,026	4.57	LRR barrel + IDRs.
DNAAF2	Kintoun PF13/DAP1	Q9NVR5	837	91,114	5.09	PIH domain + IDRs.
DNAAF3	C19orf51 PF22/DAB1	Q8N9W5	541	59,409	5.66	α/β domain$ + IDR.
DNAAF4	DYX1C1	Q8WXU2	420	48,526	8.88	N-terminal antiparallel β sandwich followed by an extended α helix, and IDR and three C-terminal TPRs.
DNAAF5	HEATR2	Q86Y56	855	93,521	5.98	Nineteen HEAT repeats.
DNAAF6	PIH1D3 Twister	Q9NQM4	214	24,068	3.99	IDR + PIH domain.
DNAAF7	ZMYND10	O75800	440	50,343	5.81	Mainly α helical with a C-terminal Zn2+ finger.
DNAAF8	C16orf71	Q8IYS4	520	55,681	4.83	Several α helices interconnected by extended IDRs.
DNAAF9	C20orf194, Shulin	Q5TEA3	1,177	132,286	6.11	α helical with four antiparallel β sheets.
DNAAF10	WDR92 Monad	Q96MX6	357	39,740	8.32	β propeller consisting of seven WD repeats.
DNAAF11	LRRC6 Mot47 Seahorse	Q86X45	466	54,254	6.08	LRR barrel, three α helical regions, antiparallel β sandwich and extended C-terminal IDR.
DNAAF12	LRRC56 ODA8/DLU2	Q8IYG6	542	58,733	8.05	LRR barrel with N- and C-terminal IDRs.
DNAAF13	SPAG1	Q07617	926	103,638	6.46	Multiple α helical domains interconnected by IDRs.
DNAAF14	PIH1D1 MOT48/DAP2	Q9NWS0	290	32,362	5.05	PIH domain followed by IDR and α/β domain and a C-terminal α helix.
DNAAF15	PIH1D2	Q8WWB5	315	35,956	5.96	PIH domain followed by IDR and α/β domain and a C-terminal α helix.
DNAAF16	CFAP298 C21orf59 FBB18/DAB2	P57076	290	33,224	6.99	N-terminal α/β structure with extended C-terminal helix.
DNAAF17	CFAP300 C11orf70	Q9BRQ4	267	30,859	6.39	α/β structure.
DNAAF18	WDR69 ODA16/DAW1	Q8N136	415	45,776	6.12	β propeller consisting of eight WD repeats.
DNAAF19	CCDC103	Q8IW40	242	27,162	5.73	N-terminal α helix followed by IDR and C-terminal RPAP3_C domain.

^†See (Braschi et al., 2022) for a detailed discussion of dynein assembly factors and their human genome organization (HUGO)-approved nomenclature and for additional aliases.

^†Various aliases commonly used for Chlamydomonas, zebrafish and/or Tetrahymena orthologs and human placeholder names are indicated.

^*Abbreviations: HEAT, α solenoid repeat found in huntingtin, elongation factor 3, protein phosphatase 2A and TOR1; IDR, inherently disordered region; LRR, leucine-rich repeat; PIH, protein that interacts with HSP90; RPAP3_C, C-terminal domain found in RNA polymerase II-associated protein 3; TPR, tetratricopeptide repeat; WD, approximately forty residue repeat usually ending in tryptophan (W)-aspartate (D).

^$M. Sakato-Antoku, R.S. Patel-King, J.L. Balsbaugh, and S.M. King – in preparation.

These cytosolic dynein assembly factors fall into several distinct groups based on their structural/functional properties (Braschi et al., 2022). Several (DNAAF2, DNAAF6, DNAAF14 and DNAAF15) contain PIH domains that recruit heat-shock protein HSP90, and multiple studies have revealed that different axonemal dyneins require distinct PIH proteins and that this can vary as assembly proceeds (Omran et al., 2008; Yamaguchi et al., 2018; Yamamoto et al., 2020). Several other factors (e.g., DNAAF4, DNAAF5 and DNAAF13) (Knowles et al., 2013; Tarkar et al., 2013; Yamamoto et al., 2017) contain protein interaction modules and appear to act as scaffolds providing surfaces to bind various factors and build variant forms of the R2TP complex with the RuvBL1/2 AAA ATPases and different PIH proteins (Kakihara and Houry, 2012; Patel-King et al., 2019; Yamamoto et al., 2020; zur Lage et al., 2018). Others contain leucine-rich repeat (LRR) domain barrels (DNAAF1, DNAAF11 and DNAAF12) (Desai et al., 2015; Duquesnoy et al., 2009; Zhao et al., 2013) or WD-repeat β-propellers (DNAAF10 and DNAAF18) (Ahmed et al., 2008; Liu et al., 2019; Patel-King and King, 2016; Patel-King et al., 2019; zur Lage et al., 2018) that again provide protein interaction surfaces. The DNAAF9 protein (also known as Shulin) binds to fully formed axonemal dyneins to keep them in an inactive state prior to their transport into and assembly within the ciliary axoneme (Mali et al., 2021). Finally, one assembly factor (DNAAF8) (Lee et al., 2020) contains several short α helical segments but is predicted to be mainly disordered.

Sequence analysis and examination of AlphaFold2 structure predictions (Jumper et al., 2021) indicates that eleven of these assembly factors - DNAAF1, DNAAF2, DNAAF3, DNAAF6, DNAAF8, DNAAF11, DNAAF12, DNAAF13, DNAAF14, and DNAAF15 - contain one or more IDRs usually associated with at least one well folded domain. In contrast, DNAAF5, DNAAF7, DNAAF9, DNAAF10, DNAAF16, DNAAF17, DNAAF18 and DNAAF19 lack significant regions that are inherently disordered. Examples of AlphaFold2 structures for four DNAAFs containing varying numbers of IDRs are illustrated in Figure 2.

Figure 2. Examples of Human DNAAF Proteins Containing Inherently Disordered Regions.

Alphafold2 predictions for four human DNAAF proteins (DNAAF1, DNAAF2, DNAAF4 and DNAAF8) are shown as ribbon diagrams displayed using the PyMOL molecular graphics system. Folded domains containing β sheets are indicated in light blue, α helical segments in light pink, antiparallel β hairpins in pale green and inherently disordered/unstructured loop regions in wheat. DNAAF1 consists of a leucine-rich repeat barrel and several helical regions interconnected by IDRs. DNAAF2 has an N-terminal α-crystallin domain followed by a PIH domain and IDRs while DNAAF4 has an N-terminal α-crystallin domain followed by an extended α helix and disordered region and terminates in three tetratricopeptide repeats. DNAAF8 has several predicted helical segments that do not associate to form a compact structured domain connected by disordered regions. As each amino acid adds a length of 0.38 nm to an extended protein chain (Miller and Goebel, 1968), DNAAF8 (with 520 residues) may exist as a highly extended structure 150–200 nm in length that could span across multiple nascent dynein particles.

Properties and Classification of Inherently Disordered Regions in Dynein Assembly Factors

In general, IDRs are of relatively low amino acid complexity and have insufficient bulky hydrophobic residues to drive cooperative folding to a compact structure implying the presence of considerable levels of charged and polar residues (Romero et al., 2001; Uversky et al., 2000). The pattern of charge (e.g., alternating clusters of positive and negative like charges) along these regions can also influence whether they are highly extended or more collapsed, and this feature can be altered by post-translational modifications of short linear motifs embedded within the IDRs. These segments are generally dynamic and sample large volumes of conformational space (Figure 3). IDRs can have a variety of properties that provide key functions. For example, they can contain short binding motifs such that a single IDR might interact with multiple partners, be altered by post-translational mechanisms, and/or take on different overall conformations when binding distinct partner proteins (Babu, 2016).

Figure 3. Disordered Regions of DNAAF1 Sample a Large Conformational Space.

Two independent AlphaFold2 predictions for human DNAAF1 (light blue and light green) were overlaid in PyMOL using the commands “align” and “super”. The N-terminal leucine-rich repeat barrel and the extended α helix that follows it are both very closely superimposed. Different locations for the second helix and the unstructured regions are indicated by yellow lines joining equivalent residues and illustrate the potentially enormous conformational space these segments might sample on a nsec to μsec time-scale.

Binding of IDRs to a partner protein is driven by the thermodynamics of the system and can result in several distinct outcomes (Morris et al., 2021). In situations where the enthalpic contribution from binding is enough to overcome the high entropic penalty a well-folded structure may emerge (i.e., a disorder to order transition). However, if the enthalpic changes are insufficient, the IDR may retain a conformational state exhibiting both static (disorder to disorder transitions) and dynamic disordered segments that mediate transient weak interactions – these are so-called “fuzzy” complexes in which the entropic penalty is minimized (Miskei et al., 2020). In general, for any given IDR, interactions with other complexes are governed by embedded short linear functional motifs as well as the overall electrostatics and charge density properties.

The disordered regions identified in dynein assembly factors from AlphaFold2 structure predictions range from 21 to 229 residues in length with all hydrophobic residues constituting approximately 40–60% of the total. However, the bulky hydrophobic residues (such as Ile, Leu, Val, Phe, Trp and Tyr) that usually drive globular protein folding to form the hydrophobic core are present in much lower amounts (≤27% of total residues and in several cases <10%). Furthermore, these regions present a broad range of properties that imply distinct functions in the dynein assembly process (see Table 2). Many of these segments contain potential phosphorylation sites corresponding to various kinase consensus sequences, although there is currently no experimental evidence to indicate that any are actually modified in this manner.

Table 2.

Composition and Properties of Inherently Disordered Regions in Human DNAAF Proteins

DNAAF #	IDR #	Residues (n)	Mol. Wt. (Da)	pI	# Residues (%)
					Basic$	Acidic$	Polar#	Hydrophobic*
								Total	Bulky
DNAAF1	1	1–88 (88)	9,217	4.71	8 (9.1)	18 (20.5)	22 (25.0)	37 (42.0)	7 (7.9)
	2	391–619 (229)	24,294	3.91	17 (7.4)	58 (25.3)	40 (17.5)	114 (49.8)	47 (20.5)
	3	640–725 (86)	8,918	4.14	5 (5.8)	11 (12.8)	26 (30.2)	44 (51.2)	15 (17.4)
DNAAF2	1	348–524 (177)	17,107	4.63	15 (8.4)	26 (14.7)	40 (22.6)	96 (54.2)	17 (9.6)
	2	670–804 (135)	15,098	4.42	11 (8.1)	27 (20.0)	54 (40.0)	43 (31.8)	30 (22.2)
DNAAF3	1	462–541 (80)	8,020	3.88	1 (1.2)	9 (11.3)	23 (28.7)	45 (56.2)	16 (20.0)
DNAAF4	1	199–254 (56)	6,226	10.90	11 (17.8)	5 (8.9)	16 (28.6)	25 (44.6)	14 (25.0)
DNAAF6	1	1–109 (109)	12,160	3.55	4 (3.7)	30 (28.5)	29 (26.6)	46 (42.2)	23 (21.1)
DNAAF8	1	1–21 (21)	2,123	5.59	1 (4.8)	1 (4.8)	6 (28.6)	13 (61.9)	2 (9.5)
	2	34–187 (154)	16,534	4.08	11 (7.1)	32 (20.8)	34 (22.1)	71 (46.1)	31 (20.1)
	3	214–260 (41)	4,127	5.59	4 (9.7)	5 (12.2)	7 (17.1)	23 (56.1)	6 (14.6)
	4	282–306 (25)	2,863	9.52	3 (12.0)	2 (8.0)	7 (28.0)	13 (52.0)	5 (20.0)
	5	329–430 (102)	10,743	4.45	9 (8.8)	19 (18.6)	32 (31.4)	42 (41.1)	12 (11.7)
	6	447–520 (74)	7,690	10.39	12 (16.2)	7 (9.4)	13 (17.5)	42 (56.7)	8 (10.8)
DNAAF11	1	359–466 (108)	12,217	9.17	19 (17.6)	16 (14.8)	33 (30.5)	40 (37.0)	19 (17.6)
DNAAF12	1	1–55 (55)	6,295	9.29	9 (16.3)	7 (12.7)	16 (29.1)	23 (41.8)	12 (21.8)
	2	259–333 (75)	7,897	4.84	6 (8.0)	9 (12.0)	18 (24.0)	42 (56.0)	18 (24.0)
	3	393–542 (150)	15,850	10.98	19 (12.6)	12 (8.0)	37 (24.7)	82 (54.7)	26 (17.3)
DNAAF13	1	119–165 (47)	5,497	9.47	11 (23.4)	7 (14.9)	10 (21.2)	19 (40.4)	9 (19.1)
	2	335–445 (111)	11,118	9.77	20 (18.0)	14 (12.6)	17 (15.3)	58 (52.2)	6 (5.4)
	3	742–805 (64)	7,019	6.52	13 (20.3)	13 (20.3)	12 (18.7)	26 (40.6)	8 (12.5)
DNAAF14	1	180–209 (30)	3,372	6.77	5 (16.6)	5 (16.6)	6 (20.0)	14 (46.7)	4 (13.3)
DNAAF15	1	184–234 (51)	5,612	5.01	4 (7.8)	6 (11.7)	16 (31.4)	25 (49.0)	14 (27.4)

^$Charged residues include Arg and Lys (basic) and Asp and Glu (acidic).

^#Polar residues include Asn, Cys, Gln, His, Ser and Thr.

^*Total hydrophobic residues include Ala, Gly, Ile, Leu, Met, Phe, Pro, Trp, Tyr, and Val. Bulky hydrophobic residues comprise Ile, Leu, Phe, Trp, Tyr, and Val.

There is a very distinct division of these IDRs into three sets defined by their pI. Of the twenty-three IDRs, ten have an acidic pI<5.0 (and in several cases <4.0) and eight have a basic pI>8.0 (two segments have a pI of almost 11); the remaining five have an intermediate pI near neutrality. For example, the 102-residue DNAAF8 IDR#5 contains 19 acidic, 9 basic and 32 polar residues with an overall pI = 4.45. Hydrophobicity plots indicate that many of the IDRs (e.g., DNAAF2 IDR#1, DNAAF11 IDR#1 and DNAAF12 IDR#3) exhibit mainly overall hydrophilic characteristics with only a few short hydrophobic segments (Figure 4). Indeed, several IDRs, including DNAAF12 IDR#1, DNAAF13 IDR#1 and DNAAF14 IDR#1, have essentially no overall hydrophobic moment at all, although these tend to be only about 50 residues or less in length. As such, these regions are unlikely ever to take on a stable well-folded conformation.

Figure 4. Generalized Hydrophilicity of Disordered Regions in Dynein Assembly Factors.

Hydropathy plots (Kyte-Doolittle) generated using ProtScale for six IDRs from various DNAAF proteins are shown. Those on the top row exhibit a few short regions of significant hydrophobic tendency whereas those on the bottom row are almost completely hydrophilic. A positive hydropathy score represents hydrophobicity and a negative score hydrophilicity.

One clear exception to this general rule is the second IDR within human DNAAF1 which exhibits a very distinct repeating pattern of hydrophilicity interspersed with strongly hydrophobic regions (Figure 5). Indeed, sequence alignments reveal four consecutive repeating units of twenty-six residues with sixteen residues being completely conserved in all four repeats. As this region is predicted to be fully extended, it potentially might bind four copies of a target protein or interact with multiple repetitive elements within a single target. Similarly, detailed analysis of the highly acidic IDR at the C-terminus of DNAAF3 also revealed a repetitive motif (Figure 6). This extended region is amphiphilic, exhibiting an array of acidic patches on one face, while the opposite face is much more hydrophobic. To further assess these properties, the binding mode landscape for the disordered region of DNAAF3 was analyzed with FuzPred, which compares the probability of a disorder-to-disorder (or disorder-to-order) transition upon association with a partner and the potential for different modes of binding to distinct partners (either a single or multiplicity of binding modes) (Miskei et al., 2020). This revealed that the DNAAF3 IDR likely engages in multimodal binding to multiple targets and does not exhibit a strong probability of attaining order upon binding.

Figure 5. Repetitive Hydropathy Variations in a Disordered Region of Human DNAAF1.

a) Dot plot comparison of the second 229-residue IDR from DNAAF1 revealing the repetitive nature of the N-terminal ~125 residues of this segment. b) The Kyte-Doolittle hydropathy plot illustrates the repeating pattern of hydrophobicity (positive score) and hydrophilicity (negative score). c) Sequence analysis reveals a highly conserved element repeated four times within the N-terminal section of this IDR (DNAAF1 residues 394–497). Color code: charged residues, purple; polar residues, yellow; hydrophobic residues, cyan. Intriguingly, other mammalian DNAAF1 orthologs have a single copy of this repeat. d) Structural model of the twenty-residue conserved region of the second repeat. These unstructured regions exhibit an amphipathic nature with one face being mainly hydrophobic, whereas the opposite face has a prominent centralized acidic patch. The model was generated using the builder interface within PyMOL. Hydrophobicity is indicated on the stick model using the command “color_h” (red, hydrophobic; white, hydrophilic). The Poisson-Boltzmann electrostatic potential from +3.0 (blue) to −3.0 (red) K_bT/e_c was calculated using the APBS plugin within PyMOL and painted on the molecular surface with transparency set to 40%.

Figure 6. Repeat Features within the C-terminal Disordered Region of Human DNAAF3.

a) Dot plot comparison of the C-terminal IDR from human DNAAF3. b) Hydropathy plot reveals several hydrophobic segments (positive score) surrounding more hydrophilic regions. c) Three short, repeated elements of an acid residue surrounded by hydrophobic ones accounts for the repeats seen in the dot plot. d) For the DNAAF3 IDR the probability of each residue undergoing a disorder-to-disorder transition upon binding a target and whether it might adopt a single or multiple binding mode is plotted. The plot can be considered in four quadrants defined by the likelihood of unimodal versus multimodal binding and whether this results from a disordered or structured bound state. The coordinates were generated using FuzPred (Miskei et al., 2020). For this region of DNAAF3, the residues strongly favor multiple binding modes and a bound structure at the border of a disorder-to-order transition. e) The C-terminal IDR AlphaFold model is shown with the residues colored by hydrophobicity and the superimposed molecular surface painted with the electrostatic potential (as described in Figure 5). The two structures are related by a rotation around the y-axis of ~180°. One face appears mostly uncharged, whereas the opposite surface exhibits distinct acidic patches along its length.

Do Dynein Assembly Factor Disordered Regions Drive Phase Separation?

Studies in mammalian ciliated cells expressing several different fluorescently-tagged dynein assembly factors and integral dynein IC and LC components demonstrated that axonemal dynein assembly occurs within discrete particles or regions of cytoplasm (Huizar et al., 2018). Photobleaching experiments indicated that the tagged DNAAFs were mobile, and that fluorescence recovery could occur rapidly suggesting constant interchange between the dynein assembly region and bulk cytoplasm. In contrast, the fluorescent signal from integral dynein proteins did not recover over the time course of the experiments and indeed different dynein signals were found to occur in non-overlapping regions of the assembly particles (Lee et al., 2020). These data lead to the suggestion that dynein assembly occurs within liquid-liquid phase separated compartments (Huizar et al., 2018; Lee et al., 2020), but what might promote condensate formation remains unclear.

Phase separation is involved in formation of various cellular membrane-less protein accumulations such as nucleoli, pyrenoids, paraspeckles and numerous other membrane-less cellular accumulations or condensates (Hyman et al., 2014; McSwiggen et al., 2019). This process is often driven by disordered protein regions that respond to specific cytoplasmic conditions to form viscous liquid droplets (Brangwynne et al., 2015). Potentially, the driving force for the establishment of these dynein assembly particles or factories might derive, at least in part, from the sheer scale of the process itself (King, 2021). The 5-μm long HC mRNAs with numerous ribosomes and HCs in various stages of assembly would allow for nascent dynein HCs to interact with each other, IC/LC complexes and/or individual LCs to generate multivalent conglomerations of growing dyneins, numerous mRNAs, ribosomes, chaperones, and assembly factors. Within these, assembly factors might be loosely associated and thus in flux with a cytoplasmic pool, while integral dynein components are tethered to the polysome-associated growing HCs and only released following the completion of the synthetic process which might take almost 15 minutes.

To assess whether DNAAF unstructured regions might contribute to or even drive the predicted dynein assembly particle phase separation, the IDRs identified in AlphaFold2 structures were analyzed using the ParSe v2 tool which uses hydrophobicity, β-turn propensity, and a polymer scaling exponent (v_model; this provides information on the distribution of self-interactions versus those with solvent) to predict whether a given IDR has a strong tendency to undergo phase separation (Ibrahim et al., 2023). Examination of all the assembly factors identified eight regions within five DNAAF proteins which may have the propensity to undergo phase separation and potentially then drive dynein assembly particle formation (Table 3); with one exception, all these regions occur within protein segments with no predicted tertiary structure. For example, within DNAAF2 the C-terminal seventy-six residues of a large-disordered segment encompassing residues 347–524 have phase separation potential. The predicted outlier is a 28-residue (residues 100–128) segment of DNAAF2 which consists of two anti-parallel β strands and a loop forming a β hairpin. Five of these eight sequences have a basic pI>8.0, and these segments all have relatively low sequence complexity and are enriched in several amino acid types – most notably Gly, Pro and Ser – and indeed, only one of these regions, from DNAAF13, contains less than 10% Ser. FuzPred analysis of DNAAF2 identified many of the residues suggested to drive phase separation within a region of the binding mode landscape dominated by high likelihood of a disorder-to-disorder transition and of a disordered unimodal bound structural state (Figure 7 upper panel); both parameters are characteristic of sequences that undergo phase separation (Hatos et al., 2023). In contrast, although two short regions of DNAAF8 (of 40 residues total) are predicted to phase separate, only a subset actually occur within the upper left quadrant of the FuzPred plot (Figure 7, lower panel) which is usually occupied by residues strongly associated with this property.

Table 3.

Inherently Disordered Regions with Predicted Propensity for Phase Separation^*

	Properties
Protein	Residues	Sequence	pI	Enriched Residues#
DNAAF1	52–83	GSSDTSYHSQQKQSGDNGSGGHFAHPREDRED	5.28	Asp, Gly, Ser
DNAAF2	100–128	APSSRPGSGGDRGAAPGSHWSLPYSLAP	8.80	Ala, Gly, Pro, Ser
	448–523	PGSVEEPSPGGENSPGGGGSPCLSSRSLAWGSSAGRESARGDSSVETREESEGTGGQRSACAMGGPGTKSGEPLCP	4.61	Glu, Gly, Pro, Ser
DNAAF8	214–233	KVTRDACGPTSSDKGGVKEA	8.18	Ala, Asp, Gly, Lys, Ser, Thr, Val
	337–357	PQDTKEADSGSRCASRKQGSQ	8.60	Gln, Ser
DNAAF12	18–44	VRVRELSWQGLHNPCPQSKGPGSQRDR	10.69	Arg, Gln, Gly, Pro, Ser
	434–467	PSLASEPSGTSSQHLVPSPPKHPRPRDSGSSSPR	10.74	Pro, Ser
DNAAF13	428–451	AGGGATGHPGGGQGAENPAGLKSQ	6.79	Ala, Gly

^*Determined using ParSe v2 (Ibrahim et al., 2023).

^#Amino acid enrichment is defined as a single residue type accounting for ≥10% of the total composition.

Figure 7. Binding Mode Landscapes for Human DNAAF2 and DNAAF8.

For each residue of human DNAAF2 (upper panel) and DNAAF8 (lower panel) the probability of undergoing a disorder-to-disorder transition (P_DD) and for exhibiting multiple binding modes for different partners is plotted. Residues with high propensity for phase separation usually occupy the upper left quadrant of the plots. Residues with this property predicted by ParSe v2 occupy the 448–523 region of DNAAF2 and are indicated in blue; the other residues within the same IDR (which spans residues 348–524) and mainly occupy the same region are colored purple; all other residues are in light brown. For DNAAF8, residues 214–233 (blue) and 337–357 (purple) may promote phase separation as predicted by ParSe v2; all other residues are in light brown. However, the first DNAAF8 segment actually occupies a region of the binding mode landscape not usually associated with this property. The coordinates for these plots were generated using FuzPred (Miskei et al., 2020).

It is noteworthy that only a few short segments of assembly factor IDRs are predicted to potentially participate in phase separation suggesting that the majority of these regions play other roles in the dynein assembly process. As the concentration of individual factors within regions of dynein assembly is unknown, it is uncertain whether these short segments from a few DNAAFs are present in sufficient amount to actually drive the protein phase separation process or whether the occurrence of cytoplasmic regions dedicated to axonemal dynein assembly is caused and/or influenced by other processes (Huizar et al., 2018; King, 2021).

Phylogenetic Conservation of Assembly Factor Disordered Regions

To assess the general level of conservation in the IDR sequences within DNAAF proteins, four mammalian sequences (human, mouse, rat, and bovine) for DNAAF2 which contains two IDRs were aligned. There are gaps of varying length in several of the sequences when compared to human. The first IDR (177 residues) which in humans contains a C-terminal region of seventy-six residues predicted to exhibit phase separation propensity was very poorly conserved with only nineteen identical residues present in all four proteins. Even so, these regions of mouse, rat and bovine DNAAF2 also exhibit phase separation propensity as does the equivalent IDR in Chlamydomonas. Thus, although the sequence is not well conserved even among mammals, the overall properties of this protein region are maintained in a broad array of organisms with motile cilia. The second IDR (135 residues) exhibits a considerably higher degree of conservation with forty-three residues identical in all four sequences and a further thirty-one with high similarity. Similarly, the first IDR of DNAAF1 is highly conserved between these four mammalian species with 39 out of 88 residues identical. In contrast, the human sequence of the second DNAAF1 IDR contains an ~80 residue (depending on species comparisons) insertion with only three short segments showing much relationship. Also, the final IDR in DNAAF3 contains a single conserved six residue motif with the remaining eighty residues exhibiting wide sequence divergence. These low levels of sequence identity between equivalent IDRs in a variety of mammalian orthologous sequences are in stark contrast to those found in, for example, the well-folded LRR region of DNAAF1 (residues 90–271) that exhibits 84.6% identity.

The dramatic differences in conservation between IDRs within orthologous mammalian proteins suggest these regions may play fundamentally different roles whereby in one class, sequence specificity is paramount as might occur if, for example, it binds a particular nascent dynein protein segment, whereas in the other class it is the overall biophysical properties of the IDR that appear conserved with the actual primary sequence being of less importance. This prediction could be readily tested by substituting individual IDRs with sequences from different organisms to test if they are functionally equivalent.

What Role(s) Might Disordered Regions Play in Dynein Assembly

Dynein assembly factors present two very distinct protein contexts – well-folded scaffolds or modules that provide clearly defined interaction surfaces and IDRs whose disordered nature might support various potential functional roles. At least some of the DNAAF IDRs likely transiently bind and stabilize nascent dynein HCs during their assembly perhaps by masking exposed hydrophobic segments. Depending on their sequence, these IDRs might attain a well-folded structure on binding their target mediating a high affinity association or undergo a disorder-to-disorder transition with an unstructured segment(s) interacting with low affinity. Given the extensive nature of these IDRs, they might also allow for multimeric associations either with several regions of the same protein or with different proteins. This may be especially important in directing the assembly of complex multimers such as axonemal dyneins. Furthermore, IDRs may adopt different conformations when binding to different proteins perhaps allowing DNAAFs to interact with multiple different HCs rather than a dedicated DNAAF being required for each different HC. As detailed above, several of these segments have a propensity for phase separation and potentially they might drive the formation of dynein assembly particles provided they are present in sufficient quantity. However, the stoichiometry of these factors with respect to nascent dynein proteins is unknown. It is also feasible that one or more IDRs might mediate interactions with dynein mRNAs helping to maintain assembly factories as defined units, and there is even a recent report that DNAAF11 (LRRC6) is involved in the transport of FOXJ1 – the key regulator of mammalian ciliary gene transcription - into the nucleus (Kim et al., 2023).

Conclusions and Prospects

Axonemal dyneins are highly complex motors and the assembly of multiple different dynein types in the same cytoplasmic milieu presents an extraordinary challenge in macromolecular assembly. That nineteen cytosolic factors are known to be dedicated to this particular process, in addition to the ribosomes, chaperones, prefoldins and other proteins normally needed for all protein syntheses, illustrates the complexity of the process. We still lack a comprehensive accounting for all interactions that occur between DNAAFs and for how they associate with nascent dyneins. Furthermore, there is little information on the needed stoichiometry of these factors per dynein or per polysome. One key problem in understanding dynein assembly is the likely transient nature of many interactions which makes isolating functional intermediates in the dynein folding/assembly pathway and defining the correct biological context for IDR function very difficult. Thus, there remains an enormous amount to learn about axonemal dynein assembly that will require application of a combined arsenal of genetic, biochemical, and biophysical methods and tools. One approach that might yield significant insight would be to apply focused ion beam milling combined with cryo-electron microscopy and tomography to the cytoplasm of multiciliated cells to examine axonemal dynein assembly particles at high resolution as has been so successful with deciphering nucleolar (Erdmann et al., 2021) and pyrenoid (He et al., 2020) organization, and the in vivo assembly of intraflagellar transport trains at the ciliary base (van den Hoek et al., 2022).

Protein Analytics - Algorithms and Display Methods

Sequences and AlphaFold2 structure predictions (Jumper et al., 2021) for all nineteen human axonemal dynein assembly factors were downloaded from UniProt (https://www.uniprot.org). To assess the conformational flexibility of IDRs, additional structural models of DNAAF1 were built using AlphaFold2 running on ColabFold (Mirdita et al., 2022) and overlaid in the PyMOL molecular graphics system (Schrödinger LLC) using the commands “align” and “super”. Basic sequence properties were obtained using ProtParam (https://web.expasy.org/protparam/). Kyte-Doolittle hydropathy plots were calculated using the ProtScale hydrophobicity tool (https://web.expasy.org/cgi-bin/protscale/protscale.pl?1). IDRs were delineated from the AlphaFold2 predictions, and the structures displayed using PyMOL. Structural models were generated using the builder interface within PyMOL. Stick diagrams were colored by hydrophobicity using the command “color_h” and Poisson-Boltzmann electrostatic potentials calculated using the APBS plugin with the range adjusted to ±3.0 K_bT/e_c; this gradation was painted onto the molecular surface with transparency set to 40% Disorder in select DNAAFs was further analyzed with IUPRED2 (https://iupred2a.elte.hu) and FuzPred (https://fuzpred.bio.unipd.it/predictor); the latter calculates probability functions for disorder-to-disorder and disorder-to-order transitions and the potential binding modes during associations with protein targets. The propensity for phase separation was predicted using ParSe v2 (https://stevewhitten.github.io/Parse_v2_web/). Multiple sequence alignments were performed using CLUSTALW (https://www.genome.jp/tools-bin/clustalw). Graphical displays were generated using GraphPad Prism v.7.05. Figures were assembled and labeled using Adobe Photoshop and Illustrator.

Abbreviations

AAA

ATPase associated with various cellular activities

DNAAF

dynein axonemal assembly factor

HC

heavy chain

IC

intermediate chain

IDR

inherently disordered region

LC

light chain

LRR

leucine-rich repeat

MTBD

microtubule-binding domain

PCD

primary ciliary dyskinesia

PIH

protein that interacts with HSP90

Data Availability

All sequence data and structure predictions used in this review are freely available at UniProt (see Table 1 for accession numbers).