The current paradigm in intracellular trafficking is that there are four AP complexes, each mediating a different post-Golgi pathway. Additional homology has been reported between the AP complexes and the COPI F subcomplex, and there are also several proteins with adaptin homology domains, including C14orf108 
. However, until now, the relationship between C14orf108 and the μ-adaptins has been unclear.
Here we show that C14orf108 has more in common with the μ-adaptins than with other MHD-containing proteins such as the stonins and FCHO1/2. C14orf108's homology to the μ-adaptins extends upstream beyond the MHD, and secondary structure predictions indicate that the μ-adaptins and C14orf108 adopt very similar folds. In addition, C14orf108 binds to a previously uncharacterised protein, DKFZp761E198, which is homologous to the β-adaptins and has an almost identical predicted secondary structure. Further evidence for the interaction between C14orf108 and DKFZp761E198 comes from the destabilisation of C14orf108 when DKFZp761E198 is knocked down, from their similar knockdown phenotypes, and from the taxonomic distribution of the two genes. In eukaryotes from five different supergroups, the genes are always found either together or not at all, suggesting coordinate loss. For all of these reasons, we have proposed that C14orf108 and DKFZp761E198 are the μ5 and β5 subunits of a novel complex, AP-5. While this work was in progress, Słabicki et al. 
demonstrated that μ5 and β5 can be coimmunoprecipitated with two other novel proteins, KIAA0415/SPG48 and C20orf29. These two proteins have a number of properties that suggest that they are the other large subunit (ζ) and the small subunit (σ5) of the AP-5 complex, including similar knockdown phenotypes to μ5 and β5.
How can we reconcile the results of the present study with the results of Słabicki et al., who identified KIAA0415 in a screen for DNA repair genes? The DNA repair phenotype appears to be robust and on-target, because the authors were able to rescue it with RNAi-resistant KIAA0415. However, the phenotype could be indirect. For instance, Słabicki et al. also pulled out a Rho GEF, ARHGEF1, in their screen, and they proposed that it might have a signalling role which, when disrupted, could affect DNA repair, e.g. by interfering with protein phosphorylation. A similar scenario could apply to KIAA0415, since there are many connections between cell signalling and membrane traffic, especially in the endocytic pathway 
. The authors also proposed that KIAA0415 might be a helicase, based on a bioinformatics analysis. However, our own data argue against such a role. We have localised tagged versions of two of the subunits to endosomes; knockdowns affect endosomal trafficking; and HHpred searches with KIAA0415 pull out AP large subunits as the top hits (). Further down the list of hits are other proteins with α-helical solenoids, such as importins, but there are no helicases on the list. Furthermore, helicases have a number of motifs that are essential for enzymatic activity 
, and these motifs are missing in KIAA0415, indicating that KIAA0415 is unlikely to be a helicase.
In addition to the other candidate AP-5 subunits, Słabicki et al. coimmunoprecipitated SPG15 and SPG11 with KIAA0415. SPG15 and SPG11 are two proteins that are mutated in hereditary spastic paraplegia (HSP), and both proteins have features that are consistent with a role in the AP-5 pathway. First, the taxonomic distributions of these two proteins closely match those of the four proposed AP-5 subunits (). Second, SPG15 has a FYVE domain, a zinc finger domain that binds to the endosomal phosphoinositide phosphatidylinositol 3-phosphate (PI3P), and like other FYVE domain-containing proteins, SPG15 has been localised to endosomes 
. Similarly, many of the proteins associated with endocytic CCVs are recruited onto the membrane by binding to the plasma membrane phosphoinositide, phosphatidylinositol 4,5-bisphosphate (PIP2), so SPG15 may be another component of the AP-5-containing coat, such as an “alternative adaptor” 
. Third, when SPG11 was analysed by HHpred, the top hit was clathrin heavy chain, followed by β′-COP, a component of the COPI coat which, like clathrin, is thought to drive vesicle formation by assembling into a cage-like structure () 
. Thus, SPG11 may be a component of the outer part of the AP-5-containing coat, possibly acting as a membrane-deforming scaffold.
What is the function of AP-5? Western blotting indicates that it cycles on and off membranes but is not associated with CCVs. The lack of any identifiable clathrin binding sites in β5 and the lack of colocalisation between tagged AP-5 and clathrin also indicate that AP-5 is clathrin-independent. In addition, the lack of key residues in μ5 for binding YXXΦ motifs indicates that, if AP-5 is an adaptor, it must be recognising some other type of sorting signal, in the same way that the MHD protein stonin recognises a non-canonical sorting signal on the membrane protein synaptotagmin 
. The partial colocalisation of tagged μ5 with LAMP1, the altered appearance of the CIMPR and Vps26 in AP-5-depleted cells, and the accumulation of swollen MVBs in such cells all point to a role in the endocytic pathway. However, the compartment that is morphologically altered in AP-5-depleted cells is positive for the CIMPR and Vps26, and AP-5 does not show much colocalisation with either of these proteins. Together, these observations suggest that the site of action of AP-5 is the late endosome and/or lysosome, as defined by the presence of LAMP1 (), but that when it is depleted, there are indirect effects on earlier, retromer-positive endosomes, possibly because a bottleneck is created. Although trafficking out of late endosomes has never been formally established, there are a number of late endosomal membrane proteins that need to be recycled, such as receptors for lysosomal hydrolases and different types of SNAREs 
, so AP-5 may be part of a coat that facilitates vesicle budding from this compartment. So far, relatively little machinery has been identified for the later stages of the endocytic pathway. A protein complex associated with the yeast vacuole was recently discovered, called the SEA complex, which also has some features of a coat, although structurally it is more closely related to tethering complexes 
. Thus, at present the AP-5 complex appears to be the best candidate for a late endosomal coat. The connection with hereditary spastic paraplegia, a group of genetic disorders that already have a number of links with membrane traffic 
, provides a promising lead for future investigations into AP-5 function.
Updated diagram of trafficking machinery and pathways.
Based on the taxonomic distribution of AP-5, we deduce that it is an ancestral complex, but that it has been lost frequently throughout eukaryotic history. In most instances we observed coordinated absence of all six proteins examined. However, in a few instances we failed to identify one or more subunits in specific taxa. These missing subunits could indicate that the complex is in the process of being lost in those lineages. Investigation of publicly available databases, however, showed that in Arabidopsis
, and Toxoplasma
, nearly all of the identified AP-5 subunits were expressed (unpublished data), arguing instead for bioinformatic false negatives in the homology searches of the missing components. Indeed, sensitive homology searching was needed in many cases to identify the AP-5 homologues, which are clearly divergent sequences in phylogenetic analysis (, S4
–S8, and S11). In the case of SPG11 and SPG15, alternate accessory proteins might be in use in diverse eukaryotic lineages, as recently described for the endosomal trafficking complex retromer 
Nonetheless, AP-5 represents a most extreme case of the more general evolutionary expendability of adaptor complexes, as AP-2, AP-3, and AP-4 have each been lost from various organisms 
. This may reflect the evolutionary plasticity of the endocytic machinery, which needs to be adapted to diverse life strategies, and it could potentially reflect the functional overlap seen between the adaptor complexes (e.g., see 
). The toxic nature of μ5 when overexpressed in mammalian cells 
could provide some hints for the basis of the repeated loss of the complex.
We have previously suggested that the duplication giving rise to COPI and the ancestral adaptor complex was coincident with the origin of the Golgi proper and the TGN 
. Given the involvement of the various adaptor complexes with the endocytic system, particularly the basally emerging AP-3 and AP-5 complexes, we now speculate that this duplication may well have correlated with the specialisation of a Golgi compartment (COPI-associated) and a primordial endosome/TGN compartment, representing a first integration of the phagosomal endocytic pathway and the secretory pathway (). Hypothetically, this primordial endosome/TGN with both secretory and endocytic features would later further expand to become the various endosomes and the TGN ().
Why wasn't the AP-5 complex discovered earlier? One reason is that AP-5 does not appear to be present in several of the major model organisms used to study membrane traffic, such as Saccharomyces cerevisiae
. In addition, AP-5's resemblance to the rest of the family is relatively weak. Even when comparing the same AP-5 subunit in closely related species like humans and mice, the degree of conservation is surprisingly low. For instance, human and mouse β2 are 99.9% identical: there is only a single conservative amino acid substitution in the 951-residue protein. In contrast, human and mouse β5 are only 85% identical: when one aligns the two sequences, there are 136 amino acid substitutions and four gaps. More generally, as seen from the phylogenetic analyses (, S4
–S8, and S11), the AP-5 components always represent divergent sequences, pointing to a lack of selective pressure on AP-5, which has made the subunits very difficult to identify even with sophisticated bioinformatic techniques. But although over 10 years have elapsed since the last AP complex, AP-4, was discovered 
, it now appears that what we all thought was the “final recount” of adaptins 
was not so final after all and that there may be additional surprises in store.