HEATR5B associates with dynein-dynactin and selectively promotes motility of AP1-bound endosomal membranes

The dynein motor complex mediates polarised trafficking of a wide variety of organelles, intracellular vesicles and macromolecules. These functions are dependent on the dynactin complex, which helps recruit cargoes to dynein’s tail region and activates motor movement. How dynein and dynactin orchestrate trafficking of diverse cargoes is unclear. Here, we identify HEATR5B, an interactor of the AP1 clathrin adaptor complex, as a novel player in dynein-dynactin function. HEATR5B is one of several proteins recovered in a biochemical screen for proteins whose association with the human dynein tail complex is augmented by dynactin. We show that HEATR5B binds directly to the dynein tail and dynactin and stimulates motility of AP1-associated endosomal membranes in human cells. We also demonstrate that the HEATR5B homologue in Drosophila is an essential gene that promotes dynein-based transport of AP1-bound membranes to the Golgi apparatus. As HEATR5B lacks the coiled-coil architecture typical of dynein adaptors, our data point to a non-canonical process orchestrating motor function on a specific cargo. We additionally show that HEATR5B promotes association of AP1 with endosomal membranes in a dynein-independent manner. Thus, HEATR5B co-ordinates multiple events in AP1-based trafficking.


INTRODUCTION
Microtubule motors play a central role in the trafficking of cellular constituents through the cytoplasm. Whilst multiple kinesin family members are tasked with transporting cargoes towards microtubule plus ends, a single motor -cytoplasmic dynein-1 (dynein) -is responsible for almost all minus end-directed movement (Reck-Peterson et al., 2018).

25
Dynein's diverse cellular cargoes include mRNAs, protein complexes, nuclei, mitochondria, lysosomes, the Golgi apparatus and multiple classes of vesicle. It is unclear how one motor orchestrates trafficking of so many cargoes.
Dynein is a highly conserved, 1.3-MDa complex of six subunits, which are each present in two 30 copies (Reck-Peterson et al., 2018;Schmidt & Carter, 2016). The heavy chain subunit contains a motor domain and a tail domain (Figure 1A). The motor domain has forcegenerating ATPase activity and a microtubule-binding site, which work in concert to drive movement along the track. The tail domain mediates homodimerisation and recruits the other dynein subunits, which make important contributions to complex stability and cargo binding 35 (Lee et al., 2018;Reck-Peterson et al., 2018;Schroeder et al., 2014).
In vitro studies have shown that mammalian dynein needs additional factors for processive movement on microtubules (McKenney et al., 2014;Miura et al., 2010;Schlager et al., 2014;Trokter et al., 2012). The paradigmatic activation mechanism uses a combination of the 1.1-

40
MDa dynactin complex and one of a number of coiled coil-containing proteins -so called 'activating adaptors' (Olenick & Holzbaur, 2019;Reck-Peterson et al., 2018) -that interact with cargo-associated proteins. The activating adaptor stimulates the interaction of dynein's tail with dynactin, which re-positions the motor domains for processive movement (McKenney et al., 2014;Schlager et al., 2014;Splinter et al., 2012;Zhang et al., 2017). Dynactin is also 45 important for cargo recruitment to dynein as it stabilises the association of the motor with activating adaptors (McKenney et al., 2014;Schlager et al., 2014;Splinter et al., 2012).
The activating adaptors and cargo-associated proteins that connect different cargoes to dynein's tail and dynactin have been defined in only a small number of cases (Hoogenraad & 50 Akhmanova, 2016;Olenick & Holzbaur, 2019;Reck-Peterson et al., 2018). Thus, for many cargoes the proteins that provide the bridge to the motor complex and activate motor movement are not known. Identifying such factors is a prerequisite for understanding principles of cargo recognition and motor activation, as well as for dissecting the cellular functions of specific dynein-based transport processes.

55
We set out to address this issue by identifying novel biochemical interactors of the dynein tail, including those whose binding is enhanced by dynactin. Functional analysis of one dynactinstimulated interactor, the non-coiled-coil protein HEATR5B, shows that it binds directly to the dynein tail and dynactin and has an evolutionarily conserved function in promoting motility of 60 AP1-associated endosomal membranes. This work reveals a critical contribution of a protein that lacks coiled-coil architecture to dynein-based transport and provides novel insights into dynein functions during membrane trafficking. We also show that HEATR5B promotes association of AP1 with endosomal membranes independently of dynein. Thus, we have identified a factor that co-ordinates the recruitment of AP1 to endosomal membranes with 65 microtubule-based transport of these structures.

A biochemical screen for dynein tail interactors
It was recently found that autoinhibitory interactions involving the dynein motor domain reduce 70 association of the dynein tail with dynactin and cargo adaptors (Htet et al., 2020;Zhang et al., 2017). We therefore sought to increase the likelihood of identifying novel tail interactors by performing affinity purifications with a recombinant human dynein complex that lacks the motor domains ( Figure 1B and Figure S1). This 'tail complex' (comprising residues 1-1079 of the DYNC1H1 heavy chain, the DYNC1I2 intermediate chain, the DYNC1LI2 light intermediate 75 chain and the DYNLL1, DYNRB1 and DYNLT1 light chains) was produced in insect cells and coupled to beads via epitope tags. The beads were then incubated with extracts from mouse brain, which provides a concentrated source of potential binding partners. Pull-downs were performed with both N-terminally and C-terminally tethered tail complexes to prevent loss of any interactions obstructed by coupling to beads in a specific orientation. In a subset of pull-80 downs, brain extracts were spiked with purified dynactin (Figure S1), which we reasoned would facilitate capture of tail interactors involved in cargo transport processes. Following mild washing of beads, retained proteins from three technical replicates per condition were analysed with label-free quantitative proteomics (Supplementary Datasets S1-S4).

85
In the absence of exogenous dynactin, 57 proteins met our criteria for enrichment in the dynein tail pull-downs compared to controls in which only the epitope tags were coupled to the beads (log2(fold change) >3.322 (i.e. (fold change) >10) and Welch's t-test q-value <0.05; Figure 1C, D). 22 of these proteins were core components of dynein or dynactin (Figure 1C,. These factors included isoforms of dynein subunits that were not present in the recombinant 90 human tail complex (Dync1i1, Dync1li1, Dynll2 and Dynlrb2), which were presumably recovered either because they exchange with their counterparts in the recombinant tail complex or are part of a second dynein complex that can be bridged by dynactin (Grotjahn et al., 2018;Urnavicius et al., 2018). The other 35 proteins specifically captured by the tail ( Figure   1C, D -bold; Table S1) included two known coiled-coil containing cargo adaptors for dynein-95 dynactin -BicD2 and Mapk8ip3 (also known as Jip3) -and the dynein regulator Ndel1 (Reck-Peterson et al., 2018). The other tail-enriched factors had not previously been shown to interact with dynein or dynactin. These proteins included ribosomal subunits, RNA binding proteins and other proteins with diverse biochemical functions (Table S1).

100
We next determined which proteins were captured on the dynein tail versus the epitope tag controls when exogenous dynactin was spiked into the extracts. In addition to dynein and dynactin components (Figure 1E,, 28 proteins were enriched on the tail with dynactin spiking (Figure 1E,. Seventeen of these proteins were not captured by the tail in the absence of exogenous dynactin (Figure 1E, F -bold and blue; Table S2). These 105 'dynactin-stimulated' interactors included Strip1 (Striatin-interacting protein 1), a component of the Stripak (Striatin-interacting phosphatases and kinases) complex, and the Stripakassociated factor Cttnbp2 (Cortactin binding protein 2) (Kuck et al., 2019). Drosophila Stripak components associate with dynein and dynactin and regulate transport of endosomes, autophagosomes and dense-core vesicles (Neisch et al., 2017;Sakuma et al., 2014). Our 110 data strengthen evidence that interactions of STRIPAK proteins with dynein-dynactin are conserved in mammals (Goudreault et al., 2009). The other dynactin-stimulated proteins had not been linked with dynein or dynactin in previous studies.
Collectively, our pull-down experiments identified ~50 novel interacting proteins of the dynein 115 tail, several of which had their association enhanced by dynactin.

HEATR5B associates with the dynein tail and dynactin
From our list of candidate tail interactors, we were particularly drawn to Heatr5B (Heat repeat containing protein 5B; also known as p200a (Hirst et al., 2005)) because this protein was the 120 only factor whose recruitment to both the N-terminally and C-terminally tethered dynein tail complexes was stimulated by exogenous dynactin (boxed labels in Figure 1E, F). Moreover, a previous proteomic study found that Heatr5B is present on dynactin-associated membranes isolated from mouse brain (Hinckelmann et al., 2016).

125
Heatr5B (HEATR5B in humans) is a 225 kDa protein that lacks a coiled-coil domain and is predicted to mostly comprise HEAT repeats (Yoshimura & Hirano, 2016); https://www.uniprot.org/uniprot/Q9P2D3). Humans also have a related protein, HEATR5A (p200b), which was not recovered in our screen for tail interactors. A complex of HEATR5B, Aftiphilin (AFTPH/AFTIN) and γ-synergin (SYNRG/AP1GBP1) interacts with the AP1 (Adaptor 130 protein-1 (AP1) complex and the Golgi-localized, gamma adaptin ear-containing, ARF-binding (GGA) protein (Hirst et al., 2005;Lui et al., 2003), adaptors that orchestrate formation and cargo loading of a subset of clathrin-coated vesicles from intracellular membranes (Nakatsu et al., 2014;Nakayama & Wakatsuki, 2003;Sanger et al., 2019). Neither AFTPH or SYNRG were found in our dynein tail pulldowns, suggesting they do not associate with the motor 135 complex or interact less strongly with it than HEATR5B does. RNAi-mediated knockdowns have implicated HEATR5B in AP1-mediated cargo sorting in HeLa cells (Hirst et al., 2005), Drosophila imaginal discs (Le Bras et al., 2012) and C. elegans epidermal cells (Gillard et al., 2015). The importance of HEATR5B proteins is underscored by the recent finding that hypomorphic mutations in the human gene are associated with the neurodevelopmental 140 syndrome pontocerebellar hypoplasia (Ghosh et al., 2021). However, the molecular function of HEATR5B is not clear in any of these systems.
We first asked if HEATR5B is part of a complex with dynein and dynactin in human cells by performing immunoprecipitations from HEK293 cell lines that stably express either GFP-145 tagged HEATR5B or GFP alone. As expected, association of AFTPH, SYNRG and the AP1g subunit was detected with GFP-HEATR5B but not the GFP control (Figure 2A). Dynein and dynactin components were also specifically precipitated with GFP-HEATR5B (Figure 2A), corroborating the results of our tail pull-down experiments. To determine if HEATR5B can interact directly with dynein or dynactin, we performed in vitro pull-downs with purified proteins.

150
Whilst truncated versions of HEATR5B were unstable and could not be purified, we could produce a full-length GFP-tagged version of the protein in insect cells (Figure S1). We detected binding of the recombinant dynein tail complex to recombinant GFP-HEATR5B but not GFP alone (Figure 2B). Purified dynactin also interacted specifically with recombinant GFP-HEATR5B ( Figure 2B). Our previous finding that association of HEATR5B with the 155 dynein tail in brain extracts is stimulated by dynactin ( Figure 1E, F) suggests that HEATR5B can interact simultaneously with both complexes. Compatible with this notion, we did not observe competition between the purified dynein tail and dynactin for HEATR5B binding in our in vitro binding assay when both complexes were added simultaneously to the beads ( Figure   2B). However, we cannot rule out the possibility that a competitive interaction was masked by 160 binding sites on one of the components not being saturated. Nonetheless, we can conclude from this set of experiments that HEATR5B complexes with endogenous dynactin and dynein in cell extracts and can interact with both complexes directly.

165
We next investigated the possibility that HEATR5B is part of a link between AP1-positive membranes and dynein-dynactin. We first asked if HEATR5B co-localises with AP1. Although antibodies to HEATR5B work in immunoblots, they are not suitable for immunostaining of cells. We therefore stained HeLa cells that have a stable integration of the GFP-HEATR5B construct with an antibody to the g subunit of the adaptor complex (AP1g). GFP-HEATR5B 170 was enriched in small puncta in the cytoplasm, ~ 30% of which co-localised with AP1g puncta (Figure 3A,B). In contrast, the population of AP1g that associated with the perinuclear trans-Golgi network (TGN, marked with TGN46 antibodies) rarely overlapped with GFP-HEATR5B ( Figure 3A and Figure S2A, B). GFP-HEATR5B puncta also seldomly associated with EEA1positive early endosomes or LAMP1-positive lysosomal membranes (Figure 3B  We then tested if the cytoplasmic structures containing HEATR5B and AP1 are motile by live imaging of GFP-HEATR5B HeLa cells transfected with a plasmid coding for an RFP-tagged 180 σ1 subunit of AP1. As expected from our immunostaining results, HEATR5B and AP1σ1 were frequently enriched together on punctate structures in the cytoplasm ( Figure 3C, Figure S3A and Movie S1). During several minutes of filming, the vast majority of the dual-labelled punctate structures exhibited short, oscillatory movements (Movie S1). This behaviour is reminiscent of the behaviour of early endosomal cargoes for microtubule motors in HeLa cells, 185 which can take tens of minutes to traverse the cytoplasm (Flores-Rodriguez et al., 2011;Tirumala et al., 2022;Zajac et al., 2013). However, a small fraction of AP1σ1 puncta exhibited unidirectional movements of between 1.5 and 10 μm both towards and away from the perinuclear region with a mean instantaneous velocity of 260 ± 4 nm/s (± SEM; 22 particles) (Figure 3D and Movie S2). Thus, AP1-associated membranes can be subjected to long-range 190 transport.
AP1 is implicated in the anterograde movement of cargoes from the TGN to recycling endosomes, as well as a reverse process that retrieves unbound receptors and SNAREs back to the TGN in order to sustain anterograde trafficking (Cancino et al., 2007;Hirst et al., 2012; 195 Robinson et al., 2010). To confirm that motile HEATR5B-positive structures in the cytoplasm are associated with recycling endosomal membranes, we transfected GFP-HEATR5B HeLa cells with a DsRed-tagged version of the recycling endosome marker RAB11A. HEATR5B and RAB11A signals frequently overlapped in the cytoplasm (Figure 3E and Figure S3B). As was the case with AP1 and GFP-HEATR5B, most of the structures positive for RAB11A and GFP-
The above observations indicate that HEATR5B can associate with AP1-bound endosomal 205 membranes that are capable of directed movement. To determine if dynein contributes to trafficking of these structures, we examined the distribution of AP1g in cells treated with an siRNA pool that depletes DYNC1H1 ( Figure S3C). As microtubule minus ends are enriched at the perinuclear microtubule-organising centre (Brinkley, 1985), a role of dynein in AP1 transport should be reflected in more peripheral AP1g localisation when the motor complex is 210 inhibited. This is indeed what we saw, with AP1g-associated structures more dispersed in the DYNC1H1 siRNA conditions than in controls treated with a non-targeting control siRNA pool (Figure 3G,H). The dispersed AP1g-associated structures included TGN material (as judged by strong TGN46 staining), which was previously shown to depend on dynein for perinuclear clustering (Burkhardt et al., 1997). However, AP1g puncta that were positive for RAB11A but 215 lacked robust TGN46 signals, and thus corresponded to the free recycling endosome compartment (Fujii et al., 2020a), were also localised more peripherally when DYNC1H1 was depleted ( Figure 3G). Based on these data, we conclude that dynein promotes retrograde trafficking of AP1-associated endosomal membranes.

HEATR5B promotes membrane localisation and motility of AP1
We next sought to determine if HEATR5B contributes to trafficking of AP1-associated membranes. To this end, we used CRISPR/Cas9-mediated mutagenesis to generate clonal human U2OS cell lines with frameshift mutations in the HEATR5B gene that disrupt protein expression ( Figure S4A, B, and Table S3).

225
We first used immunostaining to examine the effect of disrupting HEATR5B on the distribution of AP1 in fixed cells. Compared to parental, wild-type cells, HEATR5B mutant cells had a striking reduction in the number of AP1g puncta in the cytoplasm, as well as lower intensity of the puncta that were present . This phenotype was seen in three independent 230 HEATR5B mutant U2OS clones and was fully rescued by transfection of a GFP-HEATR5B construct (Figure S5 and S6A), confirming the causal nature of the HEATR5B mutation. To better understand the nature of the phenotype, we visualised RAB11A together with AP1g in control and mutant cells ( Figure 4A). As expected, in the control cells punctate signals of both proteins frequently overlapped with each other within the cytoplasm. RAB11A and AP1g 235 signals did not, however, co-localise precisely, in keeping with a report that AP1 is present on tubular endosomes that have different constituents spatially segregated (Klumperman & Raposo, 2014). In HEATR5B mutant cells, RAB11A-positive structures were still abundant.
However, there was a strong reduction in the amount of AP1g signal associated with them.
These data indicate that disrupting HEATR5B reduces the association of AP1g with 240 endosomal membranes. We also observed modestly reduced association of AP1g with the TGN in the HEATR5B mutant cells, which was rescued by transfection of the GFP-HEATR5B construct (Figure S6A,B). Consistent with reduced AP1 interaction with the recycling compartment, disruption of HEATR5B caused excessive tubulation of Transferrin receptorpositive membranes that were associated with the TGN (Figure S7A, B). The decreased 245 interaction of AP1g with endosomal membranes and the TGN in mutant U2OS cells is consistent with the reduction in punctate AP1 signal observed when the HEATR5B orthologue was knocked down with RNAi in C. elegans (Gillard et al., 2015) and Drosophila wing discs (Le Bras et al., 2012). However, these earlier studies did not determine if targeting HEATR5B affects expression, stability or membrane recruitment of AP1. Immunoblotting of extracts 250 showed that the overall level of AP1g protein was not altered in the HEATR5B mutant U2OS cells ( Figure 4D). We therefore conclude that HEATR5B promotes recruitment of AP1 to endosomal membranes and, to a lesser extent, the TGN.
Our previous observation that bright AP1g puncta are abundant in cells treated with DYNC1H1 255 siRNA ( Figure 3G) revealed that HEATR5B does not co-operate with dynein to promote AP1 membrane localisation. However, a small but significant dispersion of total AP1g signal towards the periphery of HEATR5B mutant U2OS cells (Figure 4E) was consistent with HEATR5B having an additional function in dynein-based motility of AP1-bound membranes.
To directly assess the contribution of HEATR5B to motility of AP1, we performed high-speed 260 imaging of AP1σ1-RFP in live wild-type and mutant U2OS cells (Movie S5). As expected from our fixed cell analysis, AP1σ1 puncta were dimmer in the HEATR5B deficient cells than in controls ( Figure S7C). In both genotypes, the low intensity of AP1σ1-RFP puncta meant that many of these structures could only be followed for a few seconds before the signals bleached.
Nonetheless, mean square displacement (MSD) analysis over this timescale revealed that the 265 motility of AP1σ1 puncta in the cytoplasm of mutant cells was much less persistent than in controls ( Figure 4F). The impaired transport of AP1-positive structures when HEATR5B was disrupted was not an indirect effect of reduced AP1 association with membranes, as the motility defect was still evident for AP1σ1 puncta that had equivalent intensities in control and HEATR5B mutant cells ( Figure S7C). Staining of HEATR5B deficient cells with an antibody to 270 a-Tubulin indicated that altered AP1σ1 motility was also not due to impaired integrity of the microtubule network ( Figure S7D). Collectively, these data indicate that HEATR5B directly promotes motility of AP1-positive structures in U2OS cells.
We next asked if HEATR5B is sufficient to redistribute AP1-associated membranes by strongly 275 expressing GFP-HEATR5B in U2OS cells via transfection. Compared to control cells in which only GFP was expressed, GFP-HEATR5B expressing cells had increased perinuclear clustering of RAB11A-associated membranes that were also positive for AP1g and the DCTN1 subunit of the dynactin complex ( Figure 4G -I). These observations suggest that HEATR5B can stimulate retrograde trafficking of AP1-associated endosomal membranes by dynein-280 dynactin.

The Drosophila HEATR5B homologue is an essential gene
Our experiments in human tissue culture cells revealed that HEATR5B promotes recruitment of AP1 to endosomal membranes, as well as the motility of these structures. To assess the 285 importance of HEATR5B function at the organismal level, as well as in polarised cell types, we generated a strain of the fruit fly Drosophila melanogaster with an early frameshift mutation in the single HEATR5B homologue (CG2747, hereafter called Heatr5) ( Figure S8A). This was achieved by combining a Cas9 transgene that is active in the female germline (nos-cas9) with a transgene expressing two gRNAs that target Heatr5 (gRNA-Hr5 1+2 ) ( Figure S8A). Zygotic

290
Heatr5 homozygous mutants (Hr5 1 /Hr5 1 ) failed to reach adulthood ( Figure 5A), with most animals dying during the 2nd larval instar stage ( Figure S8B). The lethal phenotype was not complemented by a pre-existing deletion of a genomic region that includes the Heatr5 gene ( Figure S8B) but was fully rescued by a wild-type Heatr5 transgene ( Figure 5A). Thus, the lethality observed was due to the Heatr5 mutation and not an off-target effect of the gRNAs.

295
To examine the maternal requirement for Heatr5, we followed the development of the embryos laid by nos-cas9 gRNA-Hr5 1+2 mothers. The vast majority of embryos did not hatch into larvae (Figure 5B), instead arresting during late embryogenesis. These embryos, which presumably had biallelic disruption of Heatr5 in the female germline, typically had denticle hairs that were 300 either absent or short and thin ( Figure 5C and Figure S8C). This phenotype is reminiscent of that of mutants for Syntaxin-1A, which promotes apical protein secretion (Moussian et al., 2007;Schulze & Bellen, 1996). Taken together, these results demonstrate that Heatr5 has essential zygotic and maternal functions in Drosophila.

305
Heatr5 strongly promotes dynein-based transport of AP1-positive structures in the fly embryo We next set out to understand the effect of disrupting Heatr5 on AP1-based trafficking in Drosophila. For these experiments, we used the syncytial blastoderm embryo as a model.
This system is attractive because the microtubule cytoskeleton is highly polarised with minus 310 ends nucleated apically above the nuclei and plus ends extended basally (Karr & Alberts, 1986;Warn & Warn, 1986). This means that the activity of dynein and kinesin motors can be distinguished by the direction of cargo movement (Shubeita et al., 2008). Moreover, membranes can be readily visualised by microinjection of antibodies coupled to bright fluorophores into the shared cytoplasm of the syncytium (Papoulas et al., 2005;Sisson et al., 315 2000).
We first analysed the effect of Heatr5 depletion on AP1 distribution in blastoderm embryos by immunostaining embryos laid by control and nos-cas9 gRNA-Hr5 1+2 mothers with antibodies to AP1g. Bright AP1g puncta were abundant in the cytoplasm of control embryos, particularly 320 in the region basal to the nuclei ( Figure 5C). In contrast, embryos of nos-cas9 gRNA-Hr5 1+2 females had AP1g puncta that were much fewer in number and much dimmer ( Figure 5C, D), reminiscent of the situation in HEATR5B deficient human cells. This phenotype was confirmed with an independent pair of Heatr5 gRNAs ( Figure S9A, B), as was the failure of mutant embryos to develop to larval stages ( Figure S9C). As in HEATR5B deficient human cells, the 325 change in AP1g distribution in nos-cas9 gRNA-Hr5 embryos was not due to altered total amounts of AP1g protein ( Figure S9D). These observations reveal a conserved role of HEATR5B proteins in localising AP1 to membranes.
We next examined motility of AP1-positive structures in live embryos. This was achieved by labelling the AP1g antibodies with Alexa555-coupled secondary antibodies, injecting the We next injected the fluorescent AP1g antibody conjugates into nos-cas9 gRNA-Hr5 1+2 embryos. As expected, the AP1g puncta in these embryos were considerably dimmer than 340 those observed in the control. Nonetheless, the antibody labelling method meant they were bright enough to be followed throughout the period of filming. In the mutant embryos, the rate of net apical movement of AP1g puncta was strongly impaired. The puncta mostly exhibited short-range saltatory movements or pausing behaviour ( Figure S10B and Movie S6) and consequently rarely reaching the region beneath the blastoderm nuclei during the period of 345 image acquisition ( Figure 6A). A defect in the rate of net apical transport in mutant embryos was confirmed by automated tracking of particle movement, which additionally revealed significant reductions in the apical velocity and run length of AP1g signal compared to the control ( Figure 6B), as well as more modest decreases in the basal velocity and run length (Figure S11A). These observations show that AP1 undergoes net apical transport in the 350 Drosophila embryo and that this process is strongly promoted by Heatr5.
As actin structures are concentrated above the nuclei at blastoderm stages (Karr & Alberts, 1986), it is likely that microtubules are the tracks for long-distance transport of AP1 in the basal cytoplasm. Supporting this notion, AP1g transport in wild-type embryos was arrested by microinjection of the microtubule targeting agent colcemid and rapidly reinitiated when the 355 drug was inactivated with a pulse of UV light (Movie S7) (Czaban & Forer, 1985). The localisation of microtubule minus ends above the blastoderm nuclei strongly suggests that apical AP1 transport is driven by dynein and we confirmed this is the case by injecting wildtype embryos with a function-blocking antibody to Dynein intermediate chain (Dic) (Bullock et al., 2006) prior to injecting the AP1g antibody conjugate ( Figure 6C and Movie S8). Consistent 360 with the interdependence of dynein and kinesin motors in several bidirectional transport systems (Jolly & Gelfand, 2011), the Dic antibody also impaired some features of plus-enddirected AP1g motility ( Figure S11B). Thus, the modest impairment of plus-end-directed motion of AP1g in Heatr5 mutant embryos could be an indirect effect of dynein inhibition.
Together, these experiments demonstrate that Heatr5 promotes dynein-dependent transport 365 of AP1 along microtubules.

membranes
To next sought to shed light on the trafficking routes of the transported AP1-associated structures in the embryo. To this end, we used CRISPR-mediated homology-directed repair to generate a fly strain in which the trans-Golgi marker Golgin-245 is endogenously tagged with GFP ( Figure S12) and confirmed that the fusion protein is correctly localised to the 380 dispersed 'mini-stacks' that constitute the Golgi apparatus in Drosophila cells (Kondylis & Rabouille, 2009) (Figure S13A). Injecting the fluorescent AP1g antibody conjugate into the GFP-Golgin-245 embryos revealed that AP1 puncta were often transported to, and engaged with, the periphery of Golgin-245-positive structures or were trafficked together with them ( Figure 7A, Figure S13B and Movie S10 -12). Consistent with these observations, AP1g 385 puncta were frequently located adjacent to Golgin-245 puncta in fixed, uninjected embryos ( Figure S13C). These findings suggest that microtubule-based transport of AP1-associated membranes facilitates their interaction with Golgi membranes.
Staining of fixed YFP-Rab11 knock-in embryos (Dunst et al., 2015) demonstrated that many 390 of the AP1g-positive structures in the basal cytoplasm associated with Rab11 ( Figure S13D).
However, AP1g and Rab11 signals rarely overlapped precisely. This finding suggests that, as in U2OS cells (Figure 4A), these proteins are present on closely opposed or conjoined endosomal membrane structures. In contrast, AP1g was not enriched in the vicinity of the pool of Rab11 that is located at the apically positioned microtubule-organising centre (Pelissier et

400
Finally, we investigated if the AP1 membrane recruitment and trafficking defects in Heatr5deficient embryos are accompanied by defects in Golgi morphology. We compared the distribution in control and Heatr5 mutant embryos of the Golgin-245, GM130 and Golgin-84 golgin proteins, which are enriched, respectively, in the trans-Golgi, cis-Golgi and rims of the Golgi stack (Kondylis & Rabouille, 2009;Munro, 2011). In contrast to the small Golgi stacks 405 distributed throughout the basal cytoplasm in the wild-type embryo, the Heatr5 mutant embryos had large conglomerations of all three proteins in the region of the basal cytoplasm just beneath the nuclei ( Figure 7B; quantification in Figure S14). Of the three Golgi proteins analysed, Golgin-245 showed a particularly strong conglomeration phenotype in the mutant.
Conglomeration of golgin signals in Heatr5 mutant embryos may therefore reflect impaired release of material from the Golgi due to inefficient targeting of AP1 complexes to this organelle. Based on these observations, we propose that Heatr5-mediated transport of AP1 415 complexes to the Golgi stimulates post-Golgi trafficking in the embryo. We additionally discovered that a fraction of Golgin-245 was ectopically localised in the yolk in Heatr5 mutant embryos (Figure 7B; quantification in Figure S14). This result raises the possibility that Heatr5 also contributes to transport of this protein from an internal pool to the cytoplasm.

Novel interactors of the dynein tail
In addition to known co-factors, our recombinant tail-based screening strategy revealed ~50 novel candidate interactors of the dynein complex in brain extracts. Several of these proteins had their association with the tail enhanced by dynactin. The novel dynein tail interactors may not have been found in previous dynein 'interactomes ' (Gershoni-Emek et al., 2016;Redwine 425 et al., 2017) because these studies used full-length DYNC1H1, in which the presence of the motor domain can impair cargo association with the tail (Htet et al., 2020;Zhang et al., 2017), as well as different cell types or subcellular compartments as a source of potential interactors.
Sequence analysis revealed that only ~20% of the total set of dynein-interacting proteins in our experiments contain a predicted coiled-coil domain (Tables S1 and S2), suggesting that 430 many of them interact indirectly with the motor complex or bind directly through a mode distinct from that of canonical activating adaptors. Whilst we prioritised HEATR5B -one of the dynactin-stimulated interactors -for mechanistic analysis in this study, we anticipate that investigating other hits from our screen will shed further light on dynein's cargo linkage and regulation. Of particular note, we observed a large number of ribosomal proteins and other 435 RNA-associated proteins in our dynein tail interactomes. These factors are candidates to participate in dynein-mediated trafficking of messenger ribonucleoprotein particles (mRNPs), a process that can dictate the site of protein function in cells (Mofatteh & Bullock, 2017). Other intriguing hits include the dynactin-stimulated tail interactor Wdr91, a Rab7 effector implicated in endosomal recycling and lysosomal function (Liu et al., 2022;Xing et al., 2021), as well as 440 reovirus infection (Snyder et al., 2022). In addition to analysing the current set of dynein interactors, it may be possible in the future to adapt our pulldown approach to identify more transient interactors of the motor complex, for example by incorporating a biotin ligase on the dynein tail for proximity-dependent biotinylation (Samavarchi-Tehrani et al., 2020).

HEATR5B promotes AP1 membrane localisation and motility
HEATR5B was first identified in human cells through its association with AFTPH, which contains a motif that binds the AP1g 'ear' domain (Hirst et al., 2005;Lui et al., 2003). It was shown that AFTPH and HEATR5B form a stable complex with SYNRG and that knocking down the function of this assembly causes AP1 cargoes to be partially re-routed from the TGN 450 to a more peripheral compartment (Hirst et al., 2005). RNAi-based knockdowns subsequently implicated HEATR5B orthologues in AP1-based trafficking of components of the Notch signalling pathway in Drosophila imaginal discs (Le Bras et al., 2012) and E-cadherin in C. elegans epidermal cells (Gillard et al., 2015). How HEATR5B mechanistically influences AP1 function in these systems was not, however, investigated.

455
It was previously shown that the budding yeast homologue of Heatr5B -Laa1 -is required for membrane recruitment of AP1 (Fernandez & Payne, 2006;Zysnarski et al., 2019). We have shown that HEATR5B proteins also promote association of AP1 with membranes in human and Drosophila cells. Thus, HEATR5B has widespread roles in membrane targeting 460 of AP1. In budding yeast, the ability of Laa1 to promote AP1 membrane association depends on another protein, Laa2 (Zysnarski et al., 2019). It is unclear how HEATR5B promotes AP1 membrane association in higher eukaryotes, as an overt orthologue of Laa2 is missing.
However, our data indicate that this process is independent of HEATR5B's ability to bind dynein. We provide evidence that HEATR5B's interaction with the motor complex instead 465 promotes microtubule-based trafficking of AP1-associated endosomal membranes. Thus, HEATR5B has a central role in controlling the function of AP1, co-ordinating its association with membranes and microtubule-based transport of these structures.
Work by several groups has demonstrated the importance of coiled-coil-containing activating adaptors in linking dynein to cargoes and dynactin, and thus initiating long-distance cargo 470 transport (Olenick & Holzbaur, 2019;Reck-Peterson et al., 2018). Whilst our data demonstrate a role for HEATR5B proteins in promoting AP1 transport, including long-range dynein-based movements in the fly embryo, several observations suggest they are unlikely to act analogously to an activating adaptor as a primary link between AP1-associated membranes and dynein. Firstly, HEATR5B proteins lack the coiled-coil domains that are typical of 475 activating adaptors. Secondly, our observation of residual dynein-based motility of AP1positive membranes in nos-cas9 gRNA-Hr5 Drosophila embryos shows the cargo can still be linked to the motor complex when HEATR5B is disrupted. And thirdly, in ongoing work, we have failed to detect in vitro activation of dynein-dynactin motility by purified HEATR5B. We therefore favour a scenario in which HEATR5B enhances the function of a dynein-dynactin-480 activating adaptor complex on AP1-associated membranes. The involvement of an as-of-yet unidentified activating adaptor may also explain why we observed stimulation of dynein's association with HEATR5B by dynactin in cellular extracts but not with purified proteins.
HEATR5B consists almost entirely of repeats of HEAT domains, ~40 amino acid motifs of antiparallel a-helices separated by a short linker (Andrade & Bork, 1995;Groves et al., 1999). The 485 ability of HEAT repeat proteins to act as flexible scaffolds for protein-protein interactions (Grinthal et al., 2010;Yoshimura & Hirano, 2016) raises the possibility that HEATR5B promotes transport by stabilising a dynein-dynactin-activating adaptor assembly and/or its association with the membrane. Intriguingly, another non-coiled-coil protein involved in cargo transport by dynein-dynactin, Ankyrin-B (Lorenzo et al., 2014), also contains a large number 490 of α-helical ankyrin repeats. It is tempting to speculate that the repeated units in Ankyrin-B and HEATR5B play an analagous role in scaffolding dynein-dynactin-activating adaptor-cargo complexes.
Future efforts will be directed at identifying additional proteins that link AP1 to dynein and dynactin and determining how HEATR5B affects the motility of the entire machinery when it 495 is reconstituted in vitro. It will also be important to determine if and how HEATR5B puncta contributes to transport of other cargoes by dynein. Whilst our microinjection of mRNAs in mutant fly embryos show that HEATR5B is not a general regulator of dynein activity, the finding that a substantial fraction of HEATR5B puncta in human cells do not overlap with AP1 raises the possibility of involvement in transport of additional cargoes.

The role of HEATR5B and dynein in AP1-based membrane trafficking
Dynein plays a role in multiple trafficking events in the endocytic pathway, including transport of peripheral early endosomes, late endosomes and lysosomes, as well as sorting of internalised receptors through these compartments (Burkhardt et al., 1997;Driskell et al., 505 2007;Guo et al., 2016;Hong et al., 2009;Horgan et al., 2010;Jongsma et al., 2023;Jordens et al., 2001;Lalli et al., 2003;Loubery et al., 2008;Traer et al., 2007). The range of dynein functions makes it challenging to study specific trafficking processes by targeting the motor complex. Our analysis of HEATR5B highlights a novel dynein-based process for retrograde trafficking of AP1-associated endosomal material to the Golgi apparatus (Hirst et al., 2012; 510 Robinson et al., 2010). This process appears to be distinct from a previously identified dyneinand RAB11FIP3-dependent process for moving material between RAB11A-positive recycling endosomes and the the TGN (Horgan et al., 2010;McKenney et al., 2014) because RAB11A and AP1 are not enriched on the same domain of tubular endosomal structures. We envisage the two dynein-based pathways acting in parallel to ensure efficient delivery of endosomal 515 material to the TGN or translocating distinct sets of proteins that are sorted into endosomal membrane domains enriched with AP1 or RAB11A. Our observation in human cells that HEATR5B is concentrated on peripheral AP1-positive endosomal membranes but not the perinuclear AP1-positive TGN additionally suggests a mechanism for locally modulating dynein-based transport. Limiting dynein activity on AP1-bound membranes at the TGN 520 presumably facilitates kinesin-driven trafficking of AP1 cargoes in the anterograde direction (Nakagawa et al., 2000;Schmidt et al., 2009), thus ensuring bidirectional trafficking.
In addition to trafficking cargoes, the HEATR5B-mediated transport process may promote delivery of the AP1 complex from endosomal membranes to the TGN, where it is needed for clathrin-mediated budding of post-Golgi membranes. In support of this notion, we observed 525 long-range transport of AP1-associated membranes from the basal cytoplasm to Golgi stacks in the wild-type Drosophila embryo. Moreover, impairment of this process in  The column was washed with eight column volumes of 10% (v/v) elution buffer (lysis buffer containing 500 mM imidazole) and the protein complex eluted using a step gradient of elution 585 buffer (from 10% to 40%). The relevant fractions were collected, pooled and filtered through a 0.22-μm filter before ion exchange purification on a MonoQ column (10 mL, 5/50 GL; Sigma) that was pre-equilibrated in lysis buffer. After washing with ten column volumes of 10% buffer B (lysis buffer including 1M imidazole), the protein complex was eluted using a step gradient of buffer B (from 10% to 50%). Fractions were collected and analysed by SDS-PAGE, followed 590 by pooling of fractions of interest, dispensing into aliquots, and flash freezing in N2 for storage at -80˚C.

GFP-HEATR5B
Sf9 cells (2-L cell suspension at 2 x 10 6 cells/mL) were infected with baculovirus incorporating the GFP-HEATR5B expression cassette and cultured for a further 72 h before cell pelleting by

610
Dynactin was purified from pig brain as described previously (Schlager et al., 2014;Urnavicius et al., 2015) and analysed by SDS-PAGE. Aliquots of the complex were flash frozen in N2 for storage at -80˚C.

615
Following dissection, brains from 16 C57BL/6j-OlaHsd male mice strain were transferred to bijou tubes containing solubilisation buffer (10 mM Hepes pH 7.3, 150 mM NaCl and 2 mM EDTA) on ice. Two brains were transferred to a pre-chilled 30-mL glass tube Wheaton homogenizer containing 5 mL of cold 'solubilisation-plus' buffer (solubilisation buffer with 1X COMPLETE protease inhibitor and 1X PhosSTOP phosphatase inhibitor (Roche)). The brains 620 were homogenised in a coldroom using an electric homogeniser with a Teflon arm (1100-1300 rpm and 10 strokes). An extra 1 mL of solubilisation-plus buffer was added to the tube, followed by addition of two more brains and further homogenisation (15 strokes) with the same speed. The lysate was then transferred to a pre-chilled 50-mL falcon tube and kept on ice.
The same procedure was repeated three times, followed by pooling of lysates from the 16 625 brains, addition of Triton X-100 (0.1% (v/v) final concentration) and incubating on ice for 15 min. The lysate was then split into eight pre-chilled, thick-walled polycarbonate tubes (3.2 mL, Beckman Coulter) and centrifuged in an ultracentrifuge (Beckman Optima TLX) using a TLA110 pre-chilled rotor at 70,000 rpm for 20 min at 4˚C. The supernatants were pooled in a 50-mL falcon tube on ice and then flash frozen in N2 in 500 μL aliquots. Protein concentration 630 of the extracts was ~ 10 mg/mL.

Pull-downs using glutathione beads
Glutathione magnetic beads (Pierce) were washed twice with GST binding buffer (125 mM μL IgG binding buffer with rotation for 1 h at 4˚C. The beads were washed three times in IgG binding buffer and incubated with 500 μL mouse brain extract (5 mg total protein) or 500 μL mouse brain extract spiked with 50 µg purified dynactin (41.67 pmol (48.45 nM)) in 660 solubilisation-plus buffer for 2.5 h at 4˚C with rotation. The samples were washed and prepared for immunoblot analysis and LC-MS/MS following the same procedure described above for glutathione beads.

675
Data-dependent analysis was carried out with a resolution of 60,000 for the full MS spectrum (collected over a 200-1800 m/z range), followed by ten MS/MS spectra in the linear ion trap (collected using 35-threshold energy for collision-induced dissociation).

680
Raw mass-spectrometry data from pull-down samples were processed with MaxQuant software (versions 1.5.6.2; (Tyanova et al., 2016a)) using the built-in Andromeda engine to search against the UniprotKB mouse proteome (Mus musculus; release 2012_02) containing forward and reverse sequences. The iBAQ algorithm and "Match Between Runs" option were additionally used. Carbamidomethylation was set as a fixed modification, and methionine 685 oxidation and N-acetylation were set as variable modifications (using an initial mass tolerance of 6 ppm for the precursor ion and 0.5 Da for the fragment ions). For peptide and protein identifications, search results were filtered with a false discovery rate (FDR) of 0.01. Datasets were further processed with Perseus software (version 1.6.13.0; (Tyanova et al., 2016b)).
Protein tables were filtered to eliminate identifications from the reverse database, as well as 690 common contaminants. Only proteins identified on the basis of at least two peptides and a minimum of three quantification events in at least one experimental group were taken forward.
iBAQ intensity values were normalised against the median intensity of each sample (using only those peptides that had intensity values recorded across all samples and biological replicates), followed by log2-transformation and filling of missing values by imputation with 695 random numbers drawn from a normal distribution calculated for each sample, as previously described (Neufeldt et al., 2019;Plaszczyca et al., 2019). Proteins that were statistically significantly enriched between pairs of datasets were identified with Welch's t-tests with permutation-based false discovery rate statistics. We performed 250 permutations and the FDR threshold was set at 0.05. The parameter S0 was set at 0.1 to separate background 700 from specifically enriched interactors. Volcano plots of results were generated in Perseus.
UniprotKB accession codes of all protein groups and proteins identified by mass spectrometry are provided in Supplementary datasets 1 -4.

705
HEK293 or HeLa Flp-In cells in wells of 6-well plates (8.5 x 10 5 cells/well) were transfected with 1 μg plasmid DNA (pcDNA5 FRT/TO GFP-linker or pcDNA5 FRT/TO GFP-HR5B with pOG44 in a 1:1 or 2:1 ratio, respectively). The DNA was mixed with 200 μL OptiMEM (Gibco) and 5 μl 1mg/mL Polyethyleneimine (PEI) "MAX" MW 40,000 (Polysciences) in sterile phosphate-buffered saline (PBS). After vortexing, the mixture was incubated for 15 min at 710 20˚C before adding to the cells in a drop-wise manner and gentle swirling of the plate. Cells were then incubated for 24 h at 37˚C before removal of the media and splitting into a T25 flask containing 150 μg/mL hygromycin B and 5 μg/mL blasticidin (both from Gibco). The next day, and once every following week, the selective media was changed until single colonies appeared, which were detached with trypsin and pooled together for subsequent 715 experimentation.

GFP-Trap immunoprecipitation from extracts
To induce expression of GFP or GFP-HR5B, HEK293 or HeLa Flp-In T-REx GFP-linker or GFP-HR5B cell lines cells were cultured in the presence of 1 μg/mL tetracycline for at least 720 48 h. Cells from three 15-cm dishes were harvested for each immunoprecipitation sample and lysed in 1.2 mL of ice-cold lysis buffer (10mM Tris-HCl pH 7.4, 150 mM NaCl, 0.5mM EDTA, 5% NP-40, 1 mM PMSF, 1X COMPLETE protease inhibitors and 1X PhosSTOP phosphatase inhibitors) for 30 min by pipetting extensively every 10 min. Lysates were transferred to prechilled thick polycarbonate tubes and centrifuged at 30,000 rpm for 30 min in a TL110 rotor 725 using a Beckman Optima TLX ultracentrifuge. The supernatant was transferred to a precooled 15-mL falcon tube, followed by addition of 1.5 volumes of ice-cold dilution buffer (10 mM Tris/HCl pH 7.4, 150 mM NaCl, 0.5mM EDTA, 1X COMPLETE protease inhibitor and 1X PhosSTOP phosphatase inhibitor). The diluted supernatant was added to 60 μL of equilibrated GFP-Trap®_MA bead slurry (Chromotek) and the samples tumbled end-over-end for 4 h at 730 4˚C. Beads were washed in 1 mL Citomix buffer (with mixing by pipetting up and down four times), transferred to fresh tubes and denatured with LDS/DTT as described above for pulldowns from mouse brain extracts.

735
For each sample, 25 μL of GFP-Trap®_MA bead slurry (Chromotek) was equilibriated with 1 mL of cold dilution buffer (DB: 10 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.5 mM EDTA, 1X COMPLETE protease inhibitors and 1X PhosSTOP phosphatase inhibitors), followed by blocking of non-specific binding sites on the beads with 4% bovine serum albumen (BSA) in DB for 50 min at 4˚C with end-over-end tumbling. Beads were washed twice in DB and 740 transferred to fresh 1.5-mL tubes. Following removal of the supernatant, 80 pmol of purified GFP or GFP-HR5B in 0.5 mL of DB containing 2% NP-40 and 0.4 mM PMSF was mixed with the beads by end-over-end tumbling at 4˚C for 2 h. Beads were washed three times in RIPA buffer (10 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.5 mM EDTA, 0.1% SDS, 1% Triton X-100, 1 X COMPLETE protease inhibitors and 1X PhosSTOP phosphatase inhibitors) and two times 745 in protein binding buffer (PBB: 10mM HEPES pH 7.6, 150 mM NaCl, 0.5 mM EDTA, 0.1% Triton X-100, 1% deoxycholate and 1X COMPLETE protease inhibitors), followed by transfer to fresh tubes and removal of the supernatant. Dynein tail (20 pmol), dynactin (10 pmol) or dynein tail with dynactin (20 pmol and 10 pmol, respectively) were incubated with the beads in 0.5 mL PBB for 2 h at 4˚C with end-over-end tumbling. Subsequently, beads were washed 750 twice in 1 mL RIPA buffer (with mixing by pipetting up and down four times), transferred to a fresh tube and processed for immunoblotting as described above. One third of the sample was loaded per gel lane, alongside 0.1 μg of dynein tail or dynactin alone for molecular weight comparison.

755
Immunoblotting Proteins were separated on NuPAGE Bis-Tris gels (4% -12%) (ThermoFisher Scientific), with ECL Rainbow Full Range Marker (Cytiva) used for molecular weight standards. Following transfer to methanol-activated PDVF membrane (Immobilon P, Millipore) using the XCell II blot system (ThermoFisher Scientific), membranes were blocked with 5% (w/v) milk powder 760 (Marvel) in PBS, and incubated with primary antibodies in 1% milk powder or 1% BSA in PBS overnight at 4˚C. After washing with PBS/0.5% Tween-20 or PBS/1% Tween-20, membranes were incubated with secondary antibodies in 1% milk powder or 1% BSA in PBS for 1 h at room temperature. Details of primary and antibodies are provided below. Signals were developed using the ECL Prime system (Cytiva) and Super RX-N medical X-ray film

Antibodies for immunoblotting and immunofluorescence of human cells
The following antibodies were used for immunoblotting (IB) and immunofluorescence (   The camera was configured in 12-bit dynamic range mode (gain 1) to maximise sensitivity, and a 600 x 600 pixel region of interest selected to maximise acquisition speed. Sequential imaging of a single z-plane was performed with a total exposure time for both frames of 0.206 875 s. Absolute timestamps were recorded to ensure accurate tracking. Acquisition was controlled by Metamorph software (7.10.1.161).

Immunofluorescence of human cells
High-speed imaging of AP1s1-RFP motility in human cells was performed using the Nikon Tibased custom spinning disk confocal microscopy system described above, with the exception 880 that the camera was operated in streaming mode with an effective frame-rate of 17.5 Hz (50 ms exposure plus ~7 ms read-out time). Particle tracking and analysis of these time series was performed in Fiji (Schindelin et al., 2012) and Matlab 2021b (Mathworks). AP1s1-RFP particles were automatically detected in a threshold-free manner by 2D gaussian fitting using the TwoTone TIRF-FRET matlab package (Holden et al., 2010). Trajectories were then 885 automatically tracked using a MATLAB implementation by D. Blair and E. Dufresne of the IDL particle tracking code initially developed by D. Grier, J. Crocker and E. Weeks (http://site.physics.georgetown.edu/matlab/index.html). Since short tracks make MSD analysis inaccurate (Zahid et al., 2018), tracks that lasted for fewer than 10 time points were discarded. We also manually drew Regions of Interest (ROI) around cells and excluded all 890 tracks corresponding to tracked objects that were not located in cells. For each track, the MATLAB class MSD Analyzer (Tarantino et al., 2014)  Drosophila melanogaster strains were cultured using 'Iberian' food (5.5% (w/v) glucose, 3.5% (w/v) organic wheat flour, 5% (w/v) baker's yeast, 0.75% (w/v) agar, 16.4 mM methyl-4hydroxybenzoate (Nipagin), 0.004% (v/v) propionic acid). Stocks and crosses were maintained in an environmentally controlled room set to 25 ± 1ºC and 50 ± 5% relative humidity with a repeating 12h-light/12h-dark regime. The following previously generated strains were 905 used in the study: nos-Cas9 ZH-2A (Bloomington Drosophila Stock Center stock number: BL54591; (Port et al., 2014)), Df(3R)BSC222 (BL9699; containing a genomic deficiency that includes the Heatr5 locus); and YFP-Rab11 (Dunst et al., 2015). Wild-type flies were of the w 1118 strain.

Generation of Drosophila Heatr5 mutant, GFP-Heatr5B and GFP-Golgin-245 strains
Flies expressing a pair of gRNAs that target the 5' region of the Heatr5 gene (gRNAs 1+2; Figure S8A) from the U6:3 promoter were generated using the pCFD4 plasmid, as described (Port et al., 2014) (see Table S4 for sequences of oligos used for gRNA cloning). Males expressing this transgene were crossed with nos-Cas9 ZH-2A females (which express Cas9 915 specifically in the germline), with mutations identified in the offspring of the progeny by Sanger sequencing of PCR products derived from the targeted genomic region (Port & Bullock, 2016).
The GFP-Heatr5 expression construct was generated by restriction enzyme-mediated cloning of PCR-amplified eGFP and Heatr5 coding sequences into a pCASPER-based plasmid that expresses proteins under the control of the ubiquitously active a-tubulin84B promoter 920 (Dienstbier et al., 2009). The Heatr5 sequence was derived from the full-length cDNA clone GH08786 (Drosophila Genome Resource Center). Flies were transformed with this plasmid by P-element-mediated integration using standard procedures. The GFP-Golgin-245 knock-in strain was generated by CRISPR/Cas9-mediated homology-directed repair (HDR), as described (Port & Bullock, 2016;Port et al., 2014). Briefly, embryos of nos-Cas9 ZH-2A mothers

925
were injected with a mixture of three plasmids: two pCFD3-gRNA plasmids targeting the Golgin245 genomic sequences shown in Figure S12 (see Table S4 for oligo sequences used for cloning) and a pBlueScript-based donor plasmid (pBS-GFPG245) that has eGFP flanked by ~1-kb Golgin245 homology arms. The final concentration of each plasmid in the injection mix was 100 ng/μL for each of pCDF3-gRNA plasmids and 150 ng/μL for pBS-GFPG245.

930
Successful HDR was confirmed by PCR-based analysis of the progeny of the offspring of surviving embryos, as described (Port & Bullock, 2016;Port et al., 2014;Port et al., 2015).

CRISPR-based disruption of Heatr5 in the female Drosophila germline
Mothers doubly heterozygous for nos-Cas9 ZH-2A and one of two pCDF4 transgenes that 935 express independent pairs of gRNAs targeting Heatr5 (gRNAs 1+2 or gRNAs 3+4; Figure S8A and S9A) were crossed with wild-type males and introduced into egg-laying cages mounted on plates of apple juice agar (1.66% (w/v) sucrose, 33.33% (v/v) apple juice, 3.33% (w/v) agar, and 10.8 mM methyl-4-hydroxybenzoate). Embryos from this cross were collected in timed egg lays and processed for immunofluorescence, microinjection or immunoblotting as 940 described below.
For cuticle preparations, 0 -4 h egg collections were incubated for a further 28 h at 25˚C.

955
Unhatched embryos were then dechorionated, washed and mounted in lacto-hoyers solution before clearing by heating in an oven at 65˚C for 16 h.

Analysis of immunofluorescence signals in human cells and fly embryos
Quantification of spread of fluorescent signals away from the nucleus in U2OS cells was 960 performed using a custom ImageJ script (available at https://github.com/jboulanger/imagejmacro/tree/main/Cell_Organization). The nuclear DAPI signal was segmented either using StartDist (Schmidt et al., 2018) or a median filter, followed by a rolling ball, a threshold controlled by a probability of false alarm, a watershed and the extraction of connected components. The signed euclidean distance transform was then computed for each region of 965 interest, which allowed association of each pixel with the closest nucleus. Background pixels were discarded by segmenting the maximum projection intensity of all channels using a smoothing step and a threshold based on a quantile of the intensity distribution. Regions touching the border of the image were discarded to avoid analysis of incomplete cells. The signal of interest was segmented using a sharpening, a median filter, a rolling ball and finally 970 a threshold controlled by a user defined false alarm rate. For each cell, the spread was computed as the sum of the squared deviation to the centroid weighted by the signal intensity in each segmented mask. This procedure was implemented as a macro to process batches of images. The number and length of tubules was determined manually in ImageJ using the freehand selection tool. ImageJ was also used for quantification of particle number and total 975 particle intensity in mammalian cells and fly embryos. The non-adjusted channel of interest was thresholded and segmented, followed by removal of particles smaller than five pixels. In mammalian cells, particle intensity and number of particles were measured per individual cell.
In fly embryos, rectangular regions of 300 µm 2 were analysed in the apical cytoplasm, upper basal cytoplasm, lower basal cytoplasm and yolk; the upper basal and lower basal cytoplasmic 980 regions were defined by the midpoint of the distance between the edge of the yolk and the basal side of the nuclei.

Analysis of AP1 and mRNA motility in Drosophila embryos
Motility of AP1-positive structures was assessed with an affinity purified α-Drosophila AP1g 985 rabbit antibody that has been shown to be highly specific (Hirst et al., 2009). Fluorescent AP1g antibody conjugates were generated by incubating 1 µL of the α-AP1g antibody with 0.5 µL of Alexa555-conjugated donkey α-rabbit IgG (H+L) highly cross-adsorbed secondary antibody (ThermoFisher Scientific) and 3.5 µL PBS at room temperature for 30 min. The antibody solution was then dialysed by dropping on a 0.025 μm-pore filter MCE membrane (13 mm-990 diameter; Millipore) that was floating on 25 mL of PBS, and incubating for 1 h. The drop was recovered and centrifuged briefly to remove any aggregates before loading into a laser-pulled borosilicate Microcap needle (Drummond). Primary antibodies conjugated to the secondary antibody as negative controls were: rabbit α-HA (Sigma H6908) and rabbit α-dinitrophenol (Invitrogen A6430). Alexa488-UTP-labelled bcd mRNA was generated by in vitro transcription 995 as described (Bullock et al., 2006) and loaded into the injection needle at a concentration of 750 ng/µL in dH2O.
Antibody and mRNA injections into embryos were performed at room temperature as described (Bullock et al., 2006) using a micromanipulator fitted to an Ultraview ERS spinning 1000 disk imaging system (PerkinElmer). Time-lapse images were acquired at 1 frame/s (AP1g antibody conjugates) or 2 frame/s (bcd mRNA) using a 60×/1.2 NA UPlanApo water objective and an OrcaER camera (Hamamatsu). Centroid-based automatic tracking of fluorescent antibody and mRNA particles in the cytoplasm and quantification of their movements was performed using a custom script written in Mathematica (Wolfram) by A. Nicol 1005 and D. Zicha (Bullock et al., 2006). To assess the contribution of microtubules to motility of AP1-positive puncta, wild-type embryos were injected with a 20 ng/µL solution of colcemid 2 min prior to injection of the fluorescent AP1g antibody conjugate. After imaging the antibody signal for ~10 min, colcemid was inactivated by a 10 s exposure to UV light through the mercury lamp attached to the microscope, with filming resumed 60 s afterwards. To determine 1010 the role of dynein in transport of AP1 positive structures, embryos were preinjected 2 min before AP1g antibody conjugate injection with a 1 µg/µL solution of mouse α-Dic antibody (clone 74.1; Millipore) or a 1 µg/µL solution of mouse α-GFP antibody (mixture of clones 7.1 and 13.1; Sigma) as a control. These antibodies were dialysed before use, as described above. Multiple injection sessions were performed for each experiment to confirm the 1015 consistency of results. When assessing the effects of mutating Heatr5, the control and mutant embryo injections were interleaved within each imaging session.
To visualise both Golgin-245 and AP1 in the live blastoderm, embryos from GFP-Golgin-245 homozygous mothers were injected with Alexa555-labelled AP1g antibody conjugates as 1020 described above. To maximise the intensity of the relatively dim GFP-Golgin-245 signal, the focal plane was set closer to the surface of the embryo than for AP1g or mRNA tracking. This meant that it was typically not possible to follow transported AP1g puncta all the way from the yolk to the upper basal cytoplasm. In these experiments, images were subjected to 2 x 2 binning, with sequential imaging of the two channels (exposure times of 500 ms for the 1025 AP1g signal and 1 s for the GFP-Golgin-245 signal).

Immunoblotting of Drosophila embryo extracts
Drosophila embryo extracts were generated for immunoblotting as described (McClintock et al., 2018), with a Coomassie (Bradford) Protein Assay (ThermoFisher Scientific) used to 1030 ensure that the amount of total protein loaded per gel lane was equivalent between genotypes.

Statistical analysis and data plotting
Statistical evaluations and plots were made with Prism (version 9; Graphpad) or Perseus.
Details of sample sizes and types of tests are included in the figure legends. Individual data points are shown in plots, except when N > 40 (when box and whisker plots are used).

DATA AVAILABILITY
The mass spectrometry-based proteomics data will be deposited at the ProteomeXchange      . '% apical' is the percentage of the particle trajectory time that is classed as apical transport. Circles are mean values for individual embryos; columns and error bars represent means ± S.D. of these mean values; number of embryos analysed is shown above columns (at least 24 particles analysed per embryo); statistical significance was evaluated with a t-test: ****, p < 0.0001; ***, p <0.001; *, p <0.05.  . Purified protein samples. Images of Coomassie-stained gel lanes after electrophoresis of the human dynein tail complex (DT), pig brain dynactin (DCTN) and GFP tagged human HEATR5B (HR5B). Note that DCTN1 exists as two different isoforms (p150 and p135). M, protein markers; MW, molecular weight of protein markers.

Figure S2. Localisation of AP1g and GFP-HEATR5B with respect to other structures in fixed human cells. (A, B)
Representative confocal images of wild-type (A) and stable GFP-HEATR5B (B) HeLa cells stained with antibodies to the indicated proteins (note that GFP signal in panel B was amplified with GFP antibodies). Dashed box shows area magnified in right-hand images. Panel A shows that AP1g is clustered at the periphery of the TGN. Panel B shows that association of GFP-HEATR5B (HR5B) with TGN46, EEA1 and LAMP1 is rarely observed. Scale bars: main panels, 10 µm; insets, 2.5 µm.   Table S3. (B) Immunoblots showing loss of HEATR5B (HR5B) protein in U2OS KO1-2 clone. Equivalent data for KO1-17 clone is shown in Figure 4D.    In the sequence alignment, the position of the PAM and target sequence is transposed to the opposite strand for simplicity. (B) Lethal phase analysis of Hr5 1 zygotic mutants. As none of the genotypes exhibited significant lethality during embryogenesis, only the rate of survival of L1 larvae of the indicated genotypes to L2 or L3 stages was recorded (Df: chromosomal deficiency Df(3R)BSC222, which uncovers the Hr5 locus). Data are expressed as a percentage of initial number of L1 larvae for each genotype, with the number of L1 larvae followed for each genotype shown above columns. Hr5 1 /+ and Hr5 1 homozygous larvae were siblings from the same cross, as were 'Hr5 1 /+ or Df/+' and Hr5 1 /Df larvae. (C) Quantification of cuticle defects of unhatched embryos from nos-cas9 gRNA-Hr5 1+2 females. Embryos in the 'No denticles' category had reached late stages of embryogenesis as judged by the development of obvious internal structures, such as abdominal segments and/or mouth parts; 'Early arrest' embryos had no obvious internal structures; 'Other' represents rare instances of axial patterning defects. Data are pooled from egg lays of females generated in three independent crosses, across which the results were very consistent. N is number of embryos analysed.    cas9) and nos-cas9, gRNA-Hr5 1+2 mothers (A, C) or wild-type embryos pre-injected with function-blocking Dic antibodies or control GFP antibodies (B). '% stationary' and '% basal' are the percentages of particle trajectory time that are classed as immobile or undergoing basal transport, respectively. Circles are mean values for individual embryos; columns and error bars represent means ± S.D. of these mean values; numbers of embryos injected is shown above columns. At least 24 particles were analysed per embryo. Statistical significance was evaluated with a t-test: ****, p < 0.0001; **, p <0.01; *, p <0.05.

Supplementary dataset 1. Mass spectrometry-derived iBAQ values for dynein tail-GST and GST only pulldowns when exogenous dynactin was absent from both conditions.
In these and other datasets, the first tab shows the complete dataset and the second tab shows statistically significantly enriched proteins in the dynein-tail vs control pulldown (q value <0.05). For calling of hits (highlighted in pink), an additional threshold of iBAQ value fold change for dynein tail vs control of >10 was applied. In tab 2, proteins that are not core components of the dynein or dynactin complex are labelled in bold.

Supplementary dataset 2. Mass spectrometry-derived iBAQ values for dynein tail-GST and GST only pulldowns when exogenous dynactin was present in both conditions.
Supplementary dataset 3. Mass spectrometry-derived iBAQ values for ZZ-dynein tail and protein A only pulldowns when exogenous dynactin was absent from both conditions.

Supplementary dataset 4. Mass spectrometry-derived iBAQ values for ZZ-dynein tail and protein A only pulldowns when exogenous dynactin was present in both
conditions.