PtdIns(3,5)P2 is one of the seven regulatory PPIn (polyphosphoinositides) that are ubiquitous in eukaryotes. It controls membrane trafficking at multiple points in the endosomal/lysosomal system and consequently regulates the size, shape and acidity of at least one endo-lysosomal compartment. PtdIns(3,5)P2 appears to exert this control via multiple effector proteins, with each effector specific for a subset of the various PtdIns(3,5)P2-dependent processes. Some putative PtdIns(3,5)P2 effectors have been identified, including Atg18p-related PROPPIN [β-propeller(s) that bind PPIn] proteins and the epsin-like proteins Ent3p and Ent5p, whereas others remain to be defined. One of the principal functions of PtdIns(3,5)P2 is to regulate the fission/fragmentation of endo-lysosomal sub-compartments. PtdIns(3,5)P2 is required for vesicle formation during protein trafficking between endo-lysosomes and also for fragmentation of endo-lysosomes into smaller compartments. In yeast, hyperosmotic stress accelerates the latter process. In the present review we highlight and discuss recent studies that reveal the role of the HOPS–CORVET complex and the vacuolar H+-ATPase in the process of endo-lysosome fission, and speculate on connections between these machineries and the Fab1p pathway. We also discuss new evidence linking PtdIns(3,5)P2 and PtdIns5P to the regulation of exocytosis.
IDENTIFICATION AND BIOSYNTHESIS
The existence of PtdIns(3,5)P2 was suggested in 1989 , but it was not formally identified until almost a decade later [7,8]. This lipid is present in all eukaryotic cells that have been examined to date and appears to control several membrane trafficking events, as well as one arm of the cellular response to hyperosmotic stress . Like the other eukaryotic regulatory glycerolipids, known as PPIn (PolyPhosphoInositides), PtdIns(3,5)P2 is synthesized from PtdIns [7,8] (Figure 1).
Structure of PtdIns and PtdIns(3,5)
PtdIns(3,5)P2 biosynthesis from PtdIns is achieved by two sequential PPIn kinases, with each enzyme specific for a particular hydroxy group of the inositol ring (Figure 2) . The first step in PtdIns(3,5)P2 biosynthesis is catalysed by the PtdIns 3-kinase Vps34p (mVps34 in animals; where Vps is vacuolar protein sorting), an endosome-resident enzyme that phosphorylates the 3-hydroxy group of the inositol ring of PtdIns, but not other PPIn . Vps34p-related Type III PI3Ks (phosphoinositide 3-kinases) are found in almost all eukaryotes , and all members of this subfamily appear to display the same substrate specificity .
PPIn metabolism in eukaryotes and PtdIns(3,5)
P2 synthesis and degradation in S.cerevisiae
As indicated in Figure 2, PtdIns3P has its own functions and effectors. It is involved in at least three distinct membrane trafficking processes: in post-Golgi sorting of lysosomal enzymes, in autophagy and also in the retromer pathway  (and see Figure 3). This latter process recycles membrane-bound receptors for soluble cargos, for example the M6R (mannose-6-phosphate receptor), and sorts them from endosomes to the TGN (trans-Golgi network) . PtdIns3P is made at multiple sites in the cell, including endosomes, pre-autophagic structures and the TGN [2,11]. In contrast, PtdIns(3,5)P2 is probably generated on the vacuole (endosome/lysosome in higher eukaryotes; see Appendix 1), although the lack of a suitable probe for this lipid has prevented an exact definition of its whereabouts in cells . Indeed PtdIns(3,5)P2 synthesis in other compartments is strongly implied by some studies [13,14]. Like PtdIns3P, PtdIns(3,5)P2 also appears to have multiple functions in cells (see Figure 4 for one hypothesis capable of explaining this).
Pathways of membrane trafficking in yeast
PPIn as foci for formation of membrane-bound complexes
PIPkIIIs (Type III PtdInsP kinases) are the enzymes that catalyse 5-phosphorylation of PtdIns3P to PtdIns(3,5)P2 and are found in almost all eukaryotic cells . These large enzymes, ranging from ∼1500 kDa to 2850 kDa, are all similarly organized, comprising a PtdIns3P binding FYVE domain located near the N-terminus, followed by a regulatory CCT-like chaperone domain which is connected to the lipid kinase domain via a unique sequence rich in cysteine and histidine residues . The FYVE domain is dispensable for Fab1p (where Fab is formation of aploid and binucleate cells) activity since expression of a FYVE-deleted mutant Fab1p in cells recovers all of the phenotypes associated with loss of Fab1p . However, the enzyme is no longer tightly localized to the vacuole. The CCT domain is known to be regulatory because of the isolation of several constitutively hyperactive mutants of Fab1p that result in amino-acid changes in this region [16,17]. In addition, at least one regulator of Fab1p also binds in this area (see below).
PIPkIIIs are exemplified by Fab1p in Saccharomyces cerevisiae and PIKfyve in mammals [1,18,19]. Both are active PtdIns3P 5-kinases In vitro  and in vivo [20,21], although PIKfyve has also been reported to phosphorylate PtdIns to PtdIns5P . We find this to be a very minor activity in yeast and it remains an open question as to whether PtdIns5P represents a bona fide biological product or a non-biological artefact of In vitro assays . PIKfyve also acts as a protein kinase: it is reported both to be autophosphorylated and to phosphorylate at least one other protein, the Rab9 effector p40 (see below) .
In yeasts, Fab1p-like PtdIns3P 5-kinases provide the sole route of PtdIns(3,5)P2 synthesis: deletion of the FAB1 gene (in S. cerevisiae) or the STE12 gene (Schizosaccharomyces pombe) results in a complete absence of PtdIns(3,5)P2 [20,21,25].
However, it is possible that PIKfyve might not provide the only route to PtdIns(3,5)P2 in mammals. In a recent study of NIH 3T3 cells, a PIKfyve inhibitor (YM201636) diminished the 32P-radiolabelled peak representing PtdIns(3,5)P2 by only ∼85% . This raises the possibility of a second route to PtdIns(3,5)P2 in animals, possibly via PtdIns3P phosphorylation by Type I PtdIns4P 5-kinases . This route does not appear to function in S. cerevisiae , despite the presence therein of a Type I PtdIns4P 5-kinase (Mss4p). It remains to be determined whether this ‘YM201636-insensitive pool of PtdIns(3,5)P2’ is genuine or represents some other co-chromatographing 32P-radiolabelled compound. Whatever the answer, it is certain that PIKfyve makes most of the PtdIns(3,5)P2 in mammals. Moreover, PIKfyve can functionally replace S. cerevisiae Fab1p in fab1Δ yeast; indeed it can even recover hyperosmotic-stress-stimulated PtdIns(3,5)P2 synthesis, contrary to what we originally reported .
PtdIns(3,5)P2 is of low abundance in both yeast and animal cells. In unstressed mammalian cells, it typically constitutes 0.04–0.08% of total inositol phospholipid [7,8]: steady-state levels of PtdIns(3,5)P2 are approx. one-tenth to one-third of the concentrations of PtdIns3P and it is many-fold less abundant than either PtdIns4P or PtdIns(4,5)P2. In response to hyperosmotic stress, the concentration of PtdIns(3,5)P2 in yeast becomes elevated transiently: up to 20-fold in both S. cerevisiae and S. pombe . More modest increases have been observed in plant cells and also in at least one differentiated animal cell line, namely 3T3 L1 adipocytes [28,29].
P2 SYNTHESIS ALSO REQUIRES Vac7p, Vac14p AND Fig4p
Three other proteins, Vac7p, Vac14p and Fig4p, are required if Fab1p is to sustain both basal and hyperosmotically stimulated PtdIns(3,5)P2 synthesis in S. cerevisiae [16,17,30,31]. Inactivation of any of these three proteins causes deficient PtdIns(3,5)P2 synthesis, with the cells displaying most or all of the defects characteristic of fab1Δ cells (see below). In addition, one of the effectors of PtdIns(3,5)P2, Atg18p, also modulates Fab1p activity, but this time in a negative fashion (see below and Figure 5).
Atg18p mediates feedback inhibition of Fab1p
BLAST searches indicate that Vac7p is present in ascomycete fungi and maybe their close relatives, but not in other eukaryotes . Vac7p is an integral vacuolar membrane protein that is predicted to possess a single transmembrane domain; detergent is needed to extract it from vacuole membranes . Its C-terminal segment, which extends into the interior of the yeast vacuole, is highly conserved. In contrast, much of the rest of the protein, including its large N-terminal cytosolic domain, shows low sequence complexity and considerable sequence variation, even among close relatives of S. cerevisiae, suggesting that it is not evolutionarily constrained. Vac7p is also required for Atg18p, a PtdIns(3,5)P2 effector, to associate with the vacuole membrane (see below) . Atg18p homologues are found in almost all eukaryotes [1,32], so it might be expected that non-fungal eukaryotes will possess some as yet unrecognized protein that functionally substitutes for Vac7p.
Vac14p and its murine homologue mVac14 (also known as ArPIKfyve) are large peripheral membrane proteins that appear to be composed largely of HEAT-like domains [17,31,33]. Vac14p and mVac14 deletion cause, respectively, an ∼90% decrease in PtdIns(3,5)P2 synthesis in S. cerevisiae [17,30,31] and at least a 50% reduction in knockout mice . Some Vac14p is mislocalized in fab1Δ yeast, suggesting that Fab1p is needed for efficient association of Vac14p with the vacuole membrane . Indeed recent work has shown that this interaction between Fab1p and Vac14p is mediated by a sequence, T1017ILLR1021, that is present in the CCT-like domain of Fab1p . PIKfyve and mVac14, can also be co-immunoprecipitated from animal cells , and this association is probably mediated by the same interaction sites. Vac14p also physically interacts with the PtdIns(3,5)P2-specific 5-phosphatase, Fig4p: Vac14p and Fig4p co-immunoprecipitate, suggesting that they co-exist in a membrane-associated complex, and they have also been functionally linked by yeast two-hybrid analyses [17,31,35]. Vac14p is partially mislocalized in a fig4Δ mutant, much as is it is a fab1Δ mutant, suggesting that Fab1p and Fig4p are both required to keep Vac14p on the vacuole membrane . Indeed recent data has shown that Fab1p, Vac14p and Fig4p are present in a complex that probably contains at least two copies of Vac14p . It seems paradoxical that Vac14p should associate with both a PtdIns3P 5-kinase and a PtdIns(3,5)P2-directed 5-phosphatase, so it is possible that there is some spatial or temporal discrimination between these Vac14p functions (but see below) .
The mVac14 protein also keeps other interesting company, for unexplained reasons: it binds to the PDZ domain of nNOS [neural isoform of NOS (nitric oxide synthase)] in vertebrates , as well as associating with the Tax protein produced by the retrovirus HTLV-1 (human T-cell lymphotrophic virus-1) .
Fig4p is a Sac1p-like PPln phosphatase that removes the 5-phosphate of PtdIns(3,5)P2 to produce PtdIns3P (see Figure 2) . In the one published study of recombinant Fig4p activity, PtdIns(3,5)P2 was the only PPln attacked by Fig4p . This observation is surprising given the similarity that the catalytic site of Fig4p bears to the relatively non-specific Sac1p PPIn phosphatase . Fig4p displays a classical CX5R catalytic motif and an extensive C-terminal coiled-coil domain that mediates interaction with Vac14p: deletion of Vac14p or removal of this C-terminal segment renders Fig4p completely cytosolic, as does deletion of FAB1 . Surprisingly, inactivation of this PtdIns(3,5)P2 5-phosphatase, in yeast or in animal cells, causes a profound defect in PtdIns(3,5)P2 synthesis [17,41–43]. Recent work has unravelled this apparent conundrum by demonstrating that Fig4p is required as a scaffold protein which is necessary for Fab1p and Vac14p to interact . Hence Fab1p-like lipid kinases appear, like Vps34p, to associate with their cognate lipid phosphatases.
MYOTUBULARINS DEGRADE PtdIns(3,5)
The MTMs (myotubularins) are a family of evolutionarily related ‘dual-specificity protein phosphatase-like’ PPIn phosphatases, comprising MTM1 and the MTMR (MTM-related) proteins (MTMR1–MTMR13 and MTMR14/JUMPY) [44–46]. Their catalytic sites include CX5R motifs akin to those in Fig4p and the PtdIns(3,4,5)P3 3-phosphatase PTEN (phosphatase and tensin homologue deleted on chromosome 10) . They were initially considered to be PtdIns3P phosphatases, but were then shown to 3-dephosphorylate PtdIns(3,5)P2 to the novel PPln PtdIns5P, a molecule which may itself have signalling functions [47–50]. One recent hypothesis suggests that active MTMs naturally target PtdIns3P, but that the inactive versions can heterodimerize with the active phosphatases and direct them to PtdIns(3,5)P2 .
Mutations in members of the MTM family are causal in the genetically transmitted progressive human diseases X-linked myotubular myopathy (MTM1 or Jumpy respectively, in X-linked and autosomal forms of the disease) and CMT (Charcot-Marie-Tooth) disease (MTMR2 or MTMR13, in two CMT subtypes) . X-linked myotubular myopathy is characterized by aberrant muscle cells with central nuclei like those present in foetal myotubes. CMT disease involves faulty nerve myelination, resulting in poor conduction and neurodegeneration . Moreover, fig4−/− mice display a condition similar to human CMT disease , and FIG4 is inactivated in some humans with a form of this disease known as CMT4J [42,52].
The molecular mechanism causing these defects is still not clear, but appears to result from altered myelination . In fibroblasts cultured from mouse models of these diseases, inactivation of MTMs leads to an increase in PtdIns3P and PtdIns(3,5)P2, but to a reduction in PtdIns5P , whereas PtdIns-(3,5)P2 levels are much reduced in fig4−/− mice . Exactly how the dysregulation of PtdIns(3,5)P2 and/or PtdIns3P and/or PtdIns5P metabolism and of functions controlled by these PPIn is causal for X-linked myotubular myopathy and the CMTs remains to be determined from detailed functional studies of the affected tissues . One recent hypothesis proposes that PtdIns5P controls exocytosis, and hence plasma membrane area . Since the neurological problems in the CMT-related diseases seem to result from aberrant out-foldings of the plasma membrane, this hypothesis clearly has merit. It might also explain the effect of PIKfyve on the trafficking to the cell surface of certain permeases and ion channels (see below).
PUTATIVE EFFECTORS OF PtdIns(3,5)
P2 INCLUDE PROPPINS [β-PROPELLER(S) THAT BIND PPIn] AND EPSINS
The search for effector proteins that bind to PtdIns(3,5)P2 and mediate its effects on various cellular processes has identified several candidates .
Perhaps the most convincing are the seven bladed β-propeller PROPPIN proteins exemplified by Atg18p (also known as Svp1p, Aut10p, Nmr1p or Cvt18p) [32,53,54]. These proteins were first associated with PtdIns(3,5)P2 because deletion of ATG18 results in an enlargement of the yeast vacuole that largely phenocopies the defect seen in fab1Δ cells . PROPPINs all include the PPIn-binding sequence motif Glu/Gln-ψ-Arg-Arg-Gly (where ψ represents a hydrophobic residue), and are conserved throughout all eukaryotes, with the exception of some pathogenic micro-organisms . Most organisms have multiple PROPPINs: three in S. cerevisiae (Atg18p, Atg21p, also known as Hsv1p, and Hsv2p), and four (WIPI-1 to WIPI-4) in mammals [1,32,55–57]. Atg18p has a role in autophagy that appears to be separate from its role in the Fab1p/PtdIns(3,5)P2 pathway; autophagy does not appear to require PtdIns(3,5)P2, at least in yeast (see Appendix 2) [54,58] (see also Box 1 of the Supplementary Information to ). Atg21p is also somehow involved in the autophagy-related Cvt (cytoplasm-to-vacuole trafficking) pathway that delivers the protease aminopeptidase I from the cytosol to the vacuole lumen [59–62] (see Figure 3), and Hsv2p was previously identified as a factor required for micronucleophagy [32,63].
The PPIn-binding properties of PROPPINs suggest that they are genuine PtdIns(3,5)P2 effectors . They bind PtdIns(3,5)P2 with high affinity (Kd ∼500 nM) and selectivity, and most of the PROPPINs tested display no appreciable binding to other PPln when assayed using quantitative approaches such as PPIn competition assays and SPR (surface plasmon resonance; Biacore) analysis of PROPPIN binding to PPIn-doped phospholipid monolayers [32,64,65]. The affinities of Atg18p and Hsv2p for PtdIns(3,5)P2 are similar to the affinity of the PH (pleckstrin homology) domain of PLCδ (phospholipase Cδ) for PtdIns(4,5)P2, which allows PLCδ to interact robustly with the plasma membrane . However, PtdIns(3,5)P2 is many-fold less abundant in unstressed cells than PtdIns(4,5)P2, so it seems likely that additional protein–protein interactions might be needed to reinforce the recruitment of PROPPINs to membranes, in accordance with the principles alluded to in Figure 4. Vac7p or a Vac7p-associated protein seems to fulfil this function in yeast .
By contrast, studies using ‘lipid blot’ assays reported an apparent preference of yeast Atg21p and human WIPI-1 for binding PtdIns3P [61,62]. The physiological meaning of this is not clear, since quantitative methods such as SPR and lipid-competition assays have suggested that the PtdIns(3,5)P2 affinities of almost all PROPPINs [except for Dm3, a Drosophila melanogaster PROPPIN with equal affinities for PtdIns3P and PtdIns(3,5)P2] far outweighs their modest affinity for PtdIns3P .
Atg21p and Hsv2p appear to be mislocalized from membranes in vps34Δ, but not fab1Δ cells, suggesting that PtdIns3P may indeed contribute to their localization in vivo [61,62]. In fact a small pool of Atg18p, Atg21p and Hsv2p all localize to endosomes as well as the vacuole, and this endosomal localization is dependent upon PtdIns3P generated by the portion of Vps34p associated with Vps38p: the so-called ‘complex II’ form of this enzyme . Since the affinities of Atg18p, Atg21p and Hsv2p are much too low for PtdIns3P to be the key determinant of their association with membranes, why should a lack of PtdIns3P mislocalize these proteins? It now seems likely that some modification of the lipid-binding specificity of PROPPINs can occur in vivo, although no mechanism is currently defined. Alternatively, they may associate with protein partners that themselves depend upon PtdIns3P for their localization.
The functions of Atg18p-like proteins were illuminated by a study that examined a version of Atg18p that was membrane-localized by virtue of fusion to a transmembrane domain derived from the vacuolar ALP (alkaline phosphatase), Pho8p . This Atg18–Pho8p fusion rescued the atg18Δ strain in terms of the defects associated with the Fab1p pathway, but failed to rescue the autophagy defect of atg18Δ cells [66,67]. This confirms what we originally suggested; namely that the PtdIns(3,5)P2 binding and autophagic functions of Atg18p are distinct and separable . It seems logical to suppose that the same could be true of Atg21p and Hsv2p (and indeed all other PROPPINs); that they participate in PtdIns(3,5)P2-dependent and -independent processes. This raises the question of why PROPPINs should continue to support two such apparently unrelated processes throughout evolution? One hypothesis we favour is that PROPPINs have evolved to carry-out membrane and protein recycling and that they participate in this process in a number of different pathways. In the case of Atg18p, one of these recycling steps is PtdIns(3,5)P2-dependent but the other, supporting autophagy, is not and may instead rely upon PtdIns3P. Hence PPIn may act just to localize the recycling machinery to particular sites in the cell, with different PPIn specific for each compartment.
The two epsin-like yeast proteins Ent3p and Ent5p represent a second class of PtdIns(3,5)P2 effectors . Epsins, first identified as proteins required for endocytosis, include PPIn-binding ENTH or ANTH domains [66,67]. The first described epsins bound PtdIns(4,5)P2, but epsinR (the mammalian homologue of Ent3p/Ent5p) appears to bind PtdIns4P [69,70]. By contrast, Ent3p and Ent5p have both been reported to bind PtdIns(3,5)P2 selectively In vitro and to require Fab1p for vacuolar localization in vivo , but this is put in some doubt by the fact that others have failed to detect this PtdIns(3,5)P2 binding either In vitro or in vivo . As discussed below, however, both Ent3p and Ent5p appear to be implicated in AP1 (adaptor protein 1)-dependent trafficking of the chitin synthase Chs3p , a process now known to be dependent on Fab1p and PtdIns(3,5)P2 [14,18,20,21,25,43,71–74]. It is therefore possible that, despite displaying a lack of selectivity for PtdIns(3,5)P2 as a sole ligand, Ent3p/Ent5p may be involved in some of the functions of PtdIns(3,5)P2 .
LOSS OF PtdIns(3,5)
P2 SYNTHESIS LEADS TO DEFECTS IN DISPARATE PROCESSES
The consequences of ablating PtdIns(3,5)P2 synthesis have now been extensively studied [genetically, by siRNA (small interfering RNA), or by expression of ‘dominant-negative’ Fab1p or PIKfyve variants] in several model organisms, including S. cerevisiae, S. pombe, Caenorhabditis elegans, D. melanogaster and various mammalian cells in culture . Notably, yeast cells in which the wild-type FAB1 gene has been replaced with a kinase-dead version of Fab1p (fab1D2134R) replicate most of the defects seen in S. cerevisiae fab1Δ mutants [18,20,21,75], so it seems that the key event in these mutants is a failure of PtdIns(3,5)P2 synthesis rather than absence of Fab1p.
These studies implicate PtdIns(3,5)P2 in several cellular processes, centred around the homoeostasis of acidic organelles in the endosomal/lysosomal system [17,30]. These, to be discussed individually, include: (i) control of the size and shape of endosomes and/or lysosomes [32,76]; (ii) acidification and inheritance of these organelles [14,71]; (iii) retrieval of membrane and proteins from the vacuole/lysosome to the endosome [7,29,77–79]; (iv) ubiquitin-dependent sorting of some cargo proteins into MVBs (multi-vesicular bodies) ; and (v) an ill-defined role in stress signalling, most likely involving a still undefined stress signalling pathway [80–82], stress-provoked changes in one or more of the above processes, or some combination of these [83,84].
In addition to these obviously endosomal/lysosomal defects, loss of PtdIns(3,5)P2 is reported to restrict exocytosis and the trafficking of some membrane transporters to the cell surface in animal cells [34,85,86] and of the chitin synthase Chs3p in S. cerevisiae , and there is evidence that M6R recycling may be perturbed in cells in which the PtdIns(3,5)P2 supply is compromised [18,42]. The ability of a single lipid signal to regulate all of these processes is probably possible due to mechanisms like those outlined in Figure 4.
CONTROL OF ENDOSOME/LYSOSOME SIZE AND TRAFFICKING
Deletion of FAB1, VAC14 or FIG4, or their fungal or metazoan homologues, produces cells in which the cytoplasm becomes extensively vacuolated. One or more membrane-delineated compartments, often of ill-defined origin or identity, become greatly enlarged . In some cases, enlargement of the aberrant compartment(s) can perturb the normal inheritance of other organelles [71,72]. In the S. cerevisiae fab1Δ mutant the enlarged organelle is so large that it can preclude partitioning of the two nuclei between mother and daughter cells during mitosis [75,87,88].
The engorged yeast compartment is undoubtedly the vacuole, since vacuole markers decorate its surface and it remains distinct from the adjacent and much smaller endosomal structures, markers for which (e.g. Snf7p) are not mislocalized to the vacuole membrane (S. K. Dove, unpublished work). However, its identity in animal cells remains a matter of debate, partly because many of the analyses involve examination of terminal phenotypes that are the results of long-standing gene deletion, so it is difficult to identify the processes that gave rise to the enlarged vacuolar structures. Most studies have concluded that late endosomes or lysosomes, or both, contribute to the vacuolated structures , but some have failed to detect markers of any defined compartment(s) on these aberrant organelles . By contrast, a recent study that followed the emergence of the vacuolated condition during siRNA suppression of PIKfyve expression in HeLa cells suggests that early endosomes are the first organelles to be affected . Work has just started using YM201636, a novel inhibitor of PIKfyve, with which cells could be switched between the normal and vacuolated states . The enlarged vacuolar structures in inhibitor-treated cells carried markers for both late endosomes and lysosomes . Such examinations of the progressive inhibition of PtdIns(3,5)P2 synthesis at early times should permit better definition of the primary PtdIns(3,5)P2-dependent events. One caveat is that the YM201636-treated cells retain a little ‘PtdIns(3,5)P2’, as discussed earlier, so some PtdIns(3,5)P2-dependent events might persist in the presence of the inhibitor.
MOLECULAR BASIS OF PtdIns(3,5)
P2-DEPENDENT MEMBRANE TRAFFICKING: A VACUOLE-TO-ENDOSOME RECYCLING PATHWAY
It seems likely that at least two distinct molecular defects contribute to the ‘swollen vacuole’ phenotype of fab1Δ yeast and PIKfyve-inhibited animal cells. The first defect was uncovered by following trafficking of RS-ALP (retention sequence-ALP), a chimaeric membrane protein that traffics to the vacuole from the Golgi directly by the AP-3 pathway, by-passing all endosomes . RS-ALP then recycles out of the vacuole compartment  by a retrograde trafficking (retrieval) pathway that takes it through the PVE (pre-vacuolar endosome; see the green arrow on Figure 3), and mutants that prevent exit from the PVE can trap RS-ALP there . The SNARE (soluble N-ethylmaleimide-sensitive fusion protein-attachment protein receptor) Pep12p is required for fusion of vacuole-derived vesicles with the PVE, demonstrating that this pathway proceeds via transport intermediates. Vti1p, another SNARE, undergoes recycling by the same route .
Recycling of RS-ALP from the vacuole requires the Fab1p activator Vac7p , and fab1Δ cells are also defective for RS-ALP recycling . This implies that the retrieval of membrane proteins from the vacuole is PtdIns(3,5)P2-dependent and that, in PtdIns(3,5)P2-deficient cells, a failure of membrane recycling out of the vacuole to the PVE might provide a mechanism by which the vacuole could accumulate membrane and become enlarged .
The PROPPIN Atg18p/Svp1p is a PtdIns(3,5)P2 effector that mediates the effects of PtdIns(3,5)P2 on this retrieval pathway ; RS-ALP recycling out of the vacuole to the PVE is blocked in atg18Δ cells. However, the PtdIns(3,5)P2 concentration in unstressed atg18Δ cells is more than 10-fold higher than in wild-type cells, indicating that feedback inhibition by PtdIns(3,5)P2-ligated Atg18p would normally limit PtdIns(3,5)P2 synthesis at the vacuole (as schematically depicted in Figure 5) [15,32]. Other PtdIns(3,5)P2-dependent events remain normal in atg18Δ cells, confirming that PtdIns(3,5)P2 must mediate its diverse effects through multiple effectors [15,32].
This inhibitory control of Fab1p by Atg18p that had been localized to the vacuole by PtdIns(3,5)P2 was unexpected, but clearly it helps to maintain normal PtdIns(3,5)P2 levels . Whether other PPIn effectors might also influence local concentrations of their PPIn ligands is not known, but it does seem an elegant possible means of control of steady-state PPIn levels in cells.
We originally suggested that PtdIns(3,5)P2 might allosterically regulate the effect of Atg18p on PtdIns(3,5)P2 synthesis, but it now seems likely that the key role of PtdIns(3,5)P2 is simply to bring Atg18p to the vacuole membrane where Fab1p resides . This conclusion emerges from the properties of an Atg18p–Pho8p construct that lacks the FRRG motif and does not bind PtdIns(3,5)P2, but is vacuole-localized by the transmembrane domain of Pho8p . This Atg18–Pho8p chimaera is at the vacuole even in the absence of PtdIns(3,5)P2. PtdIns(3,5)P2 levels are normal in atg18Δ cells expressing this chimaera , suggesting that Atg18p and Fab1p only need to be on the same membrane for Atg18p to restrict the activity of the lipid kinase.
This Atg18–Pho8p chimaera also corrects vacuole enlargement in atg18Δ cells and in vac14Δ cells [despite their PtdIns(3,5)P2 deficit] , but not in fab1Δ or vac7Δ cells. This suggests that correcting one membrane trafficking defect can overcome a lack of Atg18p or of Vac14p, but that a second, and still undefined, functional defect contributes to vacuole enlargement in fab1Δ or vac7Δ cells. This idea is supported by the fact that the vacuole defects in fab1Δ cells are more severe than atg18Δ cells . For example, raising the temperature to 37 °C exaggerates the fab1Δ vacuole defect, but not that in atg18Δ cells (S. K. Dove, unpublished work).
Moreover, fab1Δ cells often display an ‘open figure eight morphology’, in which the vacuole straddles the junction between the mother and daughter cell, but in some strain backgrounds atg18Δ cells do not (S. K. Dove, unpublished work). We originally attributed this difference to the fact that the Cvt pathway (see Figure 3), one of three main routes that deliver proteases to the lumen of the S. cerevisiae vacuole, is blocked in atg18Δ, but not fab1Δ, cells. We hypothesized that the slowing of the Cvt pathway would mean that atg18Δ vacuoles would be less enlarged than fab1Δ vacuoles. However, we constructed fab1Δ cells in which the Cvt pathway was simultaneously inactivated and they still had larger vacuoles than atg18Δ cells and more frequently displayed the ‘open figure eight’ morphology (S.K. Dove, unpublished work).
These results all suggest that a second PtdIns(3,5)P2-dependent process, requiring Fab1p and Vac7p but not Atg18p, achieves some of the membrane exit from the vacuole, and that this process contributes to the vacuole enlargement in fab1Δ and vac7Δ cells. Obvious participants might have been the two other Atg18p-related yeast PROPPINs, Hsv2p and Atg21p in this process. However, single and double mutants of the genes encoding these showed no vacuolar defects, even though they bind PtdIns(3,5)P2 avidly . Indeed, they have no known PtdIns(3,5)P2-dependent functions as yet, although deletion of either partially reduces vacuole fragmentation in response to hyperosmotic stress (S. K. Dove, unpublished work).
CONTROL OF VACUOLE MORPHOLOGY BY THE RELATED HOPS AND CORVET COMPLEXES
In exponentially growing cells, the S. cerevisiae vacuole normally exists as 3–6 distinct, but tethered, sub-compartments . This morphology changes during growth so that at high-glucose concentrations, the vacuole is more subdivided, but at higher cell densities it becomes progressively larger and less fragmented (see Figure 6). Vacuole morphology can also be perturbed both by hyperosmotic stress, which further sub-compartmentalizes the vacuole into many smaller organelles [92,93], and by hypo-osmotic stress, which triggers the sub-compartments to fuse into one large vacuole (see Figure 6) . The machinery for homotypic vacuole fusion is relatively well characterized , whereas that for fission was undefined until very recently . However, a new recognition that the two processes use common components begins to provide an understanding of the mechanisms underlying vacuole fusion/fission.
CORVET–HOPS, the V-ATPase and the Fab1p pathway co-operate to regulate endo-lysosomal fusion/fission status
Proteins encoded by the class C VPS genes, which are conserved in most eukaryotes, make up the HOPS tethering complex, which controls homotypic vacuole–vacuole fusion . In the original VPS mutant screens, class C genes were those whose deletion or mutation led to failure to produce a distinct vacuole: the vesicles from the four or more pathways feeding into the vacuole failed to fuse resulting in the formation of a number of tiny aberrant organelles . Several class C Vps proteins (Vps11p, Vps16p, Vps18p and Vps33p) form the core of the HOPS complex . Fusion of vacuolar elements mediated by the HOPS complex is regulated by the Rab7-like Ypt7p. Vam6p/Vps39p, which binds preferentially to GTP-ligated Ypt7p, appears to be its effector, and the activity of Ypt7p is controlled by the GEF (guanine-nucleotide-exchange factor) Vam6p .
It now emerges that HOPS can interconvert between a tethering complex for fusion between vacuole sub-compartments and a slightly different CORVET complex that catalyses either vacuole fission into sub-compartments or vacuole–endosome fusion, or both . It does so by exchanging Ypt7p and its partners for Vps21p (yeast Rab5) plus its effector (Vps8p) and GEF (Vps3p) . The physiological process that influences this shift from fusion to fission is unknown, but overexpression of Vps3p can drive a striking fission of the vacuole  that is not unlike the fragmentation that occurs during hyperosmotic shock.
It will be important for future studies to examine whether the CORVET complex might link PtdIns(3,5)P2/Fab1p and vacuole sub-compartmentation. The CORVET regulator Vps21p controls entry of proteins into the Pep12p-positive PVE that is adjacent to the vacuole (see Figure 3) , so it is clearly possible that PtdIns(3,5)P2 influences vacuolar fission and maybe also retrograde vesicle trafficking from the vacuole to the PVE trafficking in some fashion, promoting HOPS–CORVET subunit exchange. A link between the Fab1p and CORVET machineries is certainly suggested by the finding that deletion of some CORVET and/or HOPS subunits can cause mislocalization of Atg18p, Atg21p and Hsv2p (Figure 7) .
P2-dependent events in control of vacuole size
P2-DEPENDENT VACUOLE ACIDIFICATION IS REQUIRED FOR VACUOLE FISSION
The V (vacuolar)-ATPase was recently revealed as another component of the vacuolar fission machinery . V-ATPase is a large H+-pumping protein complex that drives the acidification of endosomes and lysosomes/vacuoles . Related to the H+-gradient-driven ATP synthase of bacteria and bioenergetic organelles, it is composed of a membrane-embedded V0 subcomplex and an associated peripheral V1 subcomplex . The V-ATPase has been implicated in vacuole–vacuole fusion for some time, although in a manner that does not require a transmembrane H+ gradient. Previous work has shown that vacuole fission in yeast cells requires both the V-ATPase and vacuole acidification [20,21,30,31]. Inhibition of V-ATPase with bafilomycin prevents vacuole fission, as does inactivation of any V-ATPase subunit [21,30]. These treatments also prevent the rapid fragmentation of the vacuole in response to hyperosmotic stress. This may also explain why, at high cell densities or in the presence of low glucose, the vacuole tends to fuse to form one large compartment; a low concentration of glucose causes the V1 sector to dissociate from the V0 component and this prevents fission while maintaining fusion, because only the former process requires the H+ gradient (see Figure 6).
It has long been known that PtdIns(3,5)P2 is required for vacuolar acidification because fab1Δ, vac7Δ and vac14Δ cells all have neutral vacuoles, a defect that appears to occur independently of vacuole enlargement [71,72]. Several studies have seen no difference between the abundances of V-ATPase on the vacuoles of wild-type and fab1Δ yeast, so the acidification defect in cells that cannot make PtdIns(3,5)P2 is not simply due to mislocalization of the V-ATPase [34,43]. This defective acidification phenotype also occurs in C. elegans and in D. melanogaster cells where the Fab1p pathway is defective: Lysotracker failed to stain the endo-lysosomes or lysosomes in worms or flies encoding a partially functional PtdIns3P 5-kinase . The PtdIns(3,5)P2 dependence of endo-lysosome acidification therefore appears to be widespread, but it has not been reported in knockout mice, probably because the PtdIns(3,5)P2 complements in the vac14−/− and fig4−/− mice were only modestly compromised (∼50% of wild-type) [17,31]. Since less PtdIns(3,5)P2 is required to support vacuole acidification than to preserve normal vacuole morphology and to localize Atg18p to the vacuole , it is not surprising that vacuolar acidification defects have not been observed in these mutant organisms. Thus some mutants, such as yeast vac14Δ cells expressing a single copy of FAB1, can have substantially enlarged vacuoles that are still acidic .
Since vacuolar acidification is essential, like Atg18p localization, for vacuolar fission, it is perhaps not surprising that PtdIns(3,5)P2 regulates both processes . How PtdIns(3,5)P2 regulates vacuole acidification is presently unknown, but is likely to involve undefined effector proteins that somehow activate the V-ATPase either directly or indirectly.
ROLE OF PtdIns(3,5)
P2 IN STRESS SIGNALLING/ADAPTATION
Another process that probably involves CORVET, the V-ATPase and PtdIns(3,5)P2 is stress signalling to the vacuole, especially in response to hyperosmotic stress. Stress signalling occurs in response to sudden changes in the extracellular environment and allows a cell to respond with concomitant adjustments in metabolism and physiology [7,77]. Many of the responses to stress are well characterized and include the induction of heat shock and other chaperone-like proteins as well as up-regulation of genes, such as catalase, by the general stress-response pathway. The sensors of the various stresses and the signalling pathways they initiate are much less well understood .
The levels of PtdIns(3,5)P2 in yeast, plant and some animal cells rise in response to heat, hyperosmotic and alkaline stresses in a manner that suggests that this lipid may be a regulator of stress-driven cell responses [16,35]. In the case of hyperosmotic stress, the rise is transient, with the duration and magnitude of the accumulation of PtdIns(3,5)P2 proportional to the degree of hypertonicity. Modest stresses (e.g. 0.4 M NaCl) provoke small PtdIns(3,5)P2 accumulations that reverse within ∼10 min, but more intense stresses (such as 0.9 M NaCl) provoke much larger accumulations of PtdIns(3,5)P2 that take up to 40 min to revert to pre-stress values ( and S. K. Dove, unpublished work).
How stress-induced increases in PtdIns(3,5)P2 occur is not known: they could be caused by increased Fab1p activity or decreased PtdIns(3,5)P2 5-phosphatase activity, for example of Fig4p . Vac14p, Vac7p and Fig4p are all required for the full stress-induced accumulation of PtdIns(3,5)P2 , but this is not informative because these proteins all seem to be required for full Fab1p activity .
P2-DEPENDENT VACUOLE FISSION ACCOMPANIES STRESSES
A consequence of transient PtdIns(3,5)P2 accumulation in response to stress is a rapid Fab1p/PtdIns(3,5)P2-dependent fragmentation/fission of the vacuole (see Figure 6) . This also requires Atg18p, although a small amount of fragmentation is still observed in its absence or in the absence of the CORVET subunit Vps3p (F. Williams and S. K. Dove, unpublished work), or of the V-ATPase .
The vacuole remains ‘locked’ in the hyperfragmented state for at least 1 h even after the burst of PtdIns(3,5)P2 has passed , for incompletely understood reasons. This process requires the palmitoylated vacuolar casein kinase Yck3p, which traffics to the vacuole via the AP-3 pathway: in yck3Δ cells the fragmented vacuoles reassemble after only a few minutes . The HOPS complex Ypt7p effector Vps41p is phosphorylated by Yck3p in response to hyperosmotic stress , and this might act to prevent HOPS-mediated vacuole reassembly, so long as Yck3p is present and active. Alternatively other Yck3p substrates might be involved.
It has been widely assumed that vacuolar fragmentation in response to hyperosmotic stress, which greatly decreases the volume contained in the multiple vacuole sub-compartments, is somehow adaptive and helps restore the tonicity of the cytosol . This is clearly possible, as considerable water remains in the vacuole after even acute hyperosmotic stress . Indeed, the increase in vacuolar surface area is accompanied by a considerable decrease in vacuolar volume and we can safely assume that the excess vacuolar water is indeed extruded into the cytosol. Sedimentation of yeast at high g forces can also activate PtdIns(3,5)P2 accumulation, with concomitant fragmentation of the vacuole (S. K. Dove, unpublished work). In a possibly related phenomenon, hyperbaric stress to 40–60 MPa, provokes a transient increase in vacuolar acidity [80,81]. It is tempting to conclude that hyperosmotic and hyperbaric stresses trigger a PtdIns(3,5)P2-dependent increase in V-ATPase activity that somehow drives vacuolar fission.
P2 AND CONTROL OF EXOCYTOSIS
We now turn to other cell activities that appear to be influenced by PtdIns(3,5)P2, but for which there are as yet no known mechanisms. First, there seems to be a link between PtdIns(3,5)P2 and the regulation of exocytosis in several contexts: (i) addition of IRAP (insulin-regulated aminopeptidase) and GLUT4 (glucose transporter 4) to the plasma membrane of adipocytes ; (ii) delivery of multiple transporters to the plasma membrane in response to activation of the protein kinase SGK1 (serum- and glucocorticoid-induced protein kinase 1) ; and (iii) neurosecretion . However, the relationship between PtdIns(3,5)P2 formation and these process remains unclear, although the hypothesis that PtdIns5P might regulate exocytosis seems a plausible explanation . The first observation was that PIKfyve could be phosphorylated by PKB (protein kinase B; also known as Akt) in response to insulin, apparently at a phosphorylation site (Ser318) midway between the FYVE domain and the DEP domain that is only present in animal PIPkIIIs , but this is now in some doubt (see below). PKB-mediated phosphorylation apparently enhanced PIKfyve kinase activity, albeit quite modestly, but overexpression of a mutant PIKfyve with an S318A mutation, unexpectedly led to enhanced delivery of IRAP to the plasma membrane . It has also been reported that greatly overexpressing PIKfyve modestly inhibits exocytosis from neurosecretory cells [80,81]. Since PtdIns3P is essential for priming in neurosecretory cells, it is perhaps not surprising that PIKfyve overexpression should turn out to be inhibitory as it will effectively reduce PtdIns3P levels .
Reports of the effects of PIKfyve phosphorylation by SGK1 are more convincing . SGK1 is a serine/threonine-directed protein kinase that is activated in response to many different types of stress (including osmotic stress), leading to increased cell-surface residency of several ion channels and transporters, presumably because of increased exocytosis from intracellular pools . In vitro, Ser318 of PIKfyve is phosphorylated by SGK1 rather than PKB , and it seems that phosphorylation of this site is required for the stimulatory effect of SGK1 on the exocytosis of at least three membrane channels: the creatine transporter SLC6A8 , the potassium channel KCNQ1 , and the Na+/glucose co-transporter SLGT-1 . The SGK1 effect on these transporters is likely to involve PtdIns(3,5)P2 and/or PtdIns5P production, but the identity of the effector mediating this response remains to be determined. Exactly how the Ser318 phosphorylation influences PIKfyve activity is also unknown.
P2, AP-1 AND Ent3p/Ent5p MAY CONTROL Chs3p TRAFFICKING TO THE S. CEREVISIAE CELL SURFACE
Chs3p, a chitin synthase that makes yeast cell wall chitin, undergoes regulated secretion to the mother-bud neck in S. cerevisiae, where it strengthens this growing area of the cell wall. This transport may be from an intracellular compartment known as the ‘chitosome’ , possibly a specialized organelle unique to yeasts, to the cell surface , but other studies have suggested that Chs3p shuttles between the TGN and the Snc1p-positive post-Golgi endosome (see Figure 3) [69,70]. Chs3p sequestration and secretion requires the hetero-tetrameric AP-1 adaptin complex which complexes protein cargoes to the surface of clathrin-coated vesicles at the TGN . Several studies now link PtdIns(3,5)P2 and the putative PtdIns(3,5)P2 effectors Ent3p and Ent5p to the regulation of Chs3p trafficking to and from the cell surface . From these studies, it emerges that at least some AP-1 functions in yeast require PtdIns(3,5)P2, but with PtdIns(3,5)P2 participating downstream of AP1: deletion of any AP-1 subunit decreases PtdIns(3,5)P2 synthesis . It seems that AP-1 is another regulator of Fab1p activity, and that this control is needed for AP1 to work properly . In accordance with this idea, fab1Δ cells have a defect in Chs3p recycling and intracellular retention similar to that in AP-1 mutants . Ent3p and Ent5p have also been implicated in intracellular retention of Chs3p [85,89]. These two epsin-like cargo adaptors have previously been shown to bind clathrin, AP-1 and Gga2p  and, as discussed above, are relatively non-specific PPIn binders.
P2 AND RETROMER
PtdIns(3,5)P2 may also influence the recycling of M6R from the early/late endosome to the TGN . M6R binds precursors of soluble lysosomal proteases, such as cathepsin D, and carries them to late endosomes, whence they are transported to the lysosome. In a Rab9-dependent process catalysed by the retromer complex, the M6R is then rapidly recycled back to the TGN to pick up more cargo . Retromer comprises mVps35p (which binds cargo), the SNXs (sorting nexins) 1/2 and 5/6 (which appear to form the vesicle coat and to drive membrane deformation), and mVps29p and mVps26p (which mediate the attachment of mVps35p to the coat) .
The first connection between this process and Fab1p/PIKfyve was the observation that PIKfyve can associate with and phosphorylate the Rab9 effector, p40 . The effect of this phosphorylation was to promote p40 attachment to membranes and in some way enhance endosome-to-TGN recycling of cargoes . This result was borne out by a study which identified PIKfyve and Rab9 as two of several proteins required for endosome-to-TGN trafficking of a viral Gag protein . Moreover, mammalian HeLa cells treated with siRNAs to knockdown PIKfyve levels display a defect in the recycling of M6R and sortillin from endosomes to the TGN , as do vac14−/− mice .
Unexpectedly, however, treatment of NIH 3T3 cells with the PIKfyve inhibitor YM201636 did not reproduce this phenotype: their M6R trafficking was normal . The authors of the inhibitor study therefore ascribed the PIKfyve-dependent endosome-to-TGN defect to an indirect effect of preventing other aspects of membrane trafficking that require PtdIns(3,5)P2. Given the variety of trafficking routes that have been described, and their variations between cell types, clear conclusions will have to await further work, as will the identification of PtdIns(3,5)P2 effectors that are involved. Two candidates have been suggested: WIPI-1, an Atg18p-like PROPPIN , and SNX1 . The PPIn-binding specificities of both of these proteins are contested, with evidence that they may also bind the more abundant PtdIns3P. The case of WIPI-1 is open: all other PROPPINs bind PtdIns(3,5)P2 preferentially and we have no reason to expect WIPI-1 to behave differently . SNX1 contains a classical PX (Phox homology) domain and binds PtdIns3P with a considerably higher affinity than PtdIns(3,5)P2 .
One suggestion that has appeared in several places is that defective endosome-to-TGN trafficking is likely to be responsible for the vacuole enlargement in cells devoid of PtdIns(3,5)P2 . This is unlikely, at least in yeast, for the following reason: trafficking of Vps10p, the classical sortillin homologue, is well mapped because it is the protein used originally to isolate the retromer complex  and all the VPS proteins from yeast ; Vps10p recycles between the Pep12p-positive PVC endosome and the TGN . When we examined the localization of Vps10p in fab1Δ cells, we saw no difference from its localization in wild-type cells (F. Williams and S. K. Dove, unpublished work). In addition, retromer mutants have different phenotypes to those defective in PtdIns(3,5)P2 biosynthesis . Thus, although PIKfyve may assist retromer in higher organisms, it is still unclear if this depends upon lipid kinase activity and so upon PtdIns(3,5)P2.
Our view of the world of PtdIns(3,5)P2, a rare but lively phosphoinositide, continues to expand, and key patterns are emerging. First, it is likely that PtdIns(3,5)P2 regulates more or less the same core processes across the eukaryote super-kingdom, although we badly need evidence from groups of organisms that branched deeper in the evolutionary tree than fungi and metazoans . Secondly, the PROPPINs represent an evolutionarily conserved group of PtdIns(3,5)P2 effectors that may mediate both PtdIns(3,5)P2-dependent and -independent retrograde trafficking at many sites in the cell, one of which is essential for autophagy. Thirdly, outlines of mechanisms of PtdIns(3,5)P2-dependent retrograde trafficking from vacuole/lysosome to endosome and of vacuole fission have emerged, and detail is now needed. Fourthly, links between PtdIns(3,5)P2 and other cell machinery, including endo-lysosomal pH control by V-ATPase and CORVET-regulated processes are becoming apparent. Although there are still many more hints and challenges than firm facts, we hope readers of the present review will be as excited by the future prospects as we are.
We would like to thank Mark Lemmon, Peter Parker, Frank Cooke, Paul Whitley, Harald Stenmark, Sylvie Urbe and Mike Clague for helpful and stimulating discussions over the past decade. S. K. D. and R. H. M. would also like to thank past and present members of the Phosphoinositide Laboratory for their contributions. We would also like to apologize to those authors whose work has been omitted for reasons of space.
formation of aploid and binucleate cells
Type III PtdInsP kinase
protein kinase B
β-propeller(s) that bind PPIn
serum- and glucocorticoid-induced protein kinase 1
small interfering RNA
soluble N-ethylmaleimide-sensitive fusion protein-attachment protein receptor
surface plasmon resonance
vacuolar protein sorting
Appendix 1: Will the real yeast lysosome please step forward?
A significant impediment to understanding the role of Fab1p and PtdIns(3,5)P2 in higher organisms is the fuzzy issue of compartment identity in yeast; it is not always clear which yeast compartment corresponds to the mammalian lysosome or to the early, late or recycling endosomes . Part of the problem arises because these organelles in yeast are often defined on the basis of genetic rather than morphological evidence.
The Snc1p (t-SNARE)-positive compartment that underlies the plasma membrane in S. cerevisiae seems well placed to be either an early or recycling endosome. This appears to be confirmed by the observation that FM4-64 and other dyes that are endocytosed seem to enter this compartment at early time points . Yet there appears to be no direct trafficking pathway from this compartment to the cell surface; all cargo requiring transport to the plasma membrane must first pass through the yeast Golgi (see Figure 3). This has resulted in some individuals dubbing this compartment the ‘post-Golgi endosome’ or PGE to avoid confusion with mammalian early endosomes, and we shall do the same .
In contrast, the Pep12p (t-SNARE)-positive compartments adjacent to the vacuole do appear to connect bidirectionally with the plasma membrane, at least via genetic evidence; the uracil permease Fur4p can recycle back to the plasma membrane from the Pep12p-positive compartment, seemingly without passing through either the Snc1p-positive compartment or the Golgi (see Figure 3) . Yet the Pep12p-positive compartment is the site of formation of MVBs, a function normally associated with late endosomes in higher eukaryotes. Further obscuring this issue is the fact that the Rab GTPase, Vps21p, controlling entry into the Pep12p-positive compartment most closely resembles Rab5, the Rab controlling the early endosome in animals . An additional layer of complexity is added by observations suggesting that the Pep12p compartment and the MVB compartment are not the same organelle, but that the former matures into the latter . On balance, the evidence appears to imply that this compartment is, in fact an early endosome that can mature or acquire late-endosome characteristics. Again, the Pep12p-positive compartment has been named the PVC (pre-vacuolar compartment) or PVE (pre-vacuolar endosome) in an attempt to sidestep this issue and avoid confusion and in the present review we adopt this practise (see Figure 3).
The most enigmatic of all yeast organelles is the vacuole. This compartment is very large, acidic (pH 6.0) and is a major nutrient store. The vacuole has commonly been assumed to be the yeast lysosome and functionally it may perform this role as it is the site of protein and organelle degradation and contains many proteolytic enzyme activities . However, the vacuole is vastly greater in volume than the endosome in yeast and the pH is only as acidic as a mammalian endosome, as lysosomes generally have pHs that are below 5.5. Most intriguing of all is the fact that the vacuolar Rab is Ypt7p, a protein that is homologous with mammalian Rab7. In eukaryotes, Rab7 controls entry to, and is localized on, late endosomes. It seems very likely that the yeast vacuole is in fact a modified late endosome or endosome–lysosome hybrid and has no direct equivalent in animal cells. Might this mean that the MVB machinery could catalyse the inward invagination of the vacuole membrane as it does at the PVE? We have in fact observed this phenomenon under certain conditions (S. K. Dove, unpublished work).
Appendix 2: PtdIns(3,5)
P2 and autophagy
PtdIns(3,5)P2, Fab1p and PROPPINs seem, in some fashion, to be associated with autophagy, despite the fact that in S. cerevisiae, fab1Δ cells have no gross defect in this process; fab1Δ cells do display a kinetic defect in the rate of maturation of Pho8Δ60p, a marker of autophagy, but it is not as profound as the defect in atg18Δ cells (S. K. Dove, unpublished work). Fab1p in D. melanogaster is indeed required for correct autophagy, highlighting the fact that the morphological description of autophagy in yeast and in higher animals appears to be different. In yeast the double membrane-bound autophagosome appears to fuse directly with the vacuole whereas, in animal cells, an amphisome is first formed that then matures into an autolysosome via fusion with the late endosome. Fab1p appears to be required for this latter maturation effect because amphisomes accumulate in the cytoplasm of fab1-deleted D. melanogaster cells. Since the D. melanogaster study was performed on purely morphological criteria it is difficult to know whether protein degradation is inhibited in fab1 fly cells. Whether these differences between autophagy in yeast and animals are real or are artefacts of the disparate way the two processes have traditionally been studied (by electron microscopy in animals and by yeast genetics in S. cerevisiae) remains to be determined; yet if the vacuole is indeed a late endosome, not a lysosome (see Appendix 1), perhaps yeast genuinely do have a ‘shorter’ more truncated pathway that has become PtdIns(3,5)P2 independent?
If this is the case, why do PROPPINs appear to have PtdIns(3,5)P2-dependent and -independent functions in many organisms? Might this family of proteins mediate the same process (retrograde trafficking) at a number of distinct sites inside the cell? In this model, only one of these sites requires PtdIns(3,5)P2 to localize the PROPPIN, whereas others are independent of this lipid and require other (protein or lipid) factors to recruit the PROPPIN-associated recycling machinery. One of these PtdIns(3,5)P2-independent retrograde events may therefore be required to sustain autophagy and the Cvt pathway.
This work was supported by the Royal Society; the Biotechnology and Biological Sciences Research Council [grant number BB/D522197/1]; the Wellcome Trust [grant number WT076480MA]; the Adrian Brown postgraduate scholarship fund; and the Dorothy Hodgkin award for postgraduate students. S. K. D. is a Royal Society University Research Fellow.