Macroautophagy (herein referred to as autophagy) is a regulated process that sequesters and delivers portion of the cytoplasm to the lysosomes for degradation. Currently the autophagic process in mammals is poorly understood. Identification and characterization of mammalian autophagy proteins are crucial to elucidate details of mammalian autophagy. Beclin 1 (encoded by Becn1
, the orthologue of yeast ATG6/Vps30
), is among the earliest characterized mammalian autophagy proteins1
. Initially identified as a Bcl-2 binding protein2
, Beclin 1 has been shown both in vitro
and in vivo
to participate in autophagy regulation and to play important roles in development3
, tumorigenesis1, 3–5
. Like yeast Atg6, Beclin1 also forms a complex with Vps34/class III phosphatidylinositol 3-kinase9–11
. Yeast has at least two Atg6/Vps34 protein complexes: one containing Atg14 and participating in autophagy, and the other containing Vps38 and functioning in non-autophagic pathway10
. However, no concrete evidence has indicated multiple Beclin 1-Vps34 complexes or multiple functions associated with the Beclin 1-Vps34 in mammals11
To reveal the mechanism whereby the Beclin 1-Vps34 interaction regulates autophagy, we combined mouse genetics and biochemistry to identify Beclin 1-associated protein complexes in vivo. We genetically engineered mice in which endogenous Beclin 1 is functionally replaced by an enhanced green fluorescent protein-tagged Beclin 1 protein (Beclin 1-EGFP) ( & S1a–b). In these mice (Becn1−/−; Becn1-EGFP/+), only the Beclin 1-EGFP fusion protein, but not endogenous Beclin 1 was detected by anti-Beclin 1 antibody (). These mice were born at the expected Mendelian ratio (Fig. S1c), survived postnatally and were phenotypically normal at adult stage, suggesting a full “rescue” of the embryonic lethality of Becn1−/− mice by functional Becn1-EGFP transgene.
Identification of novel Beclin 1-interaction proteins from Becn1−/−; Becn1-EGFP/+ mice
Using these “rescued” mice, we isolated the Beclin 1-EGFP protein complexes by affinity purification from liver, brain () and thymus (data not shown), and identified their components using mass-spectrometry (Fig. S2). The “rescued” mice, not the control mice, are associated with at least six readily detectable protein bands common to both liver and brain (). These bands include Beclin 1-EGFP (#5) (~90 kD), three previously reported Beclin 1-binding proteins p150/Vps15 (#1)12
, Vps34 (#3)9–11
and UVRAG (#4)13
, and two novel proteins (#2 and #6, asterisks). The first novel protein (#6, ~60 kD, gi|27369860) contains 492 amino acids and has a conserved SMC (S
aintenance of C
hromosomes) motif or two coiled-coil domains (aa 75–95 and aa 148–178) near the N-terminus (). Interestingly, the sequence of this protein shows modest similarity to yeast Atg14 (overall 15% identity) (Fig. S3a). Thus, we named this protein Atg14L for yeast Atg14-Like
. The second novel protein (#2, ~124 kD, gi|45708948) contains 941 amino acids and has a conserved RUN domain (aa 49–190) near the N-terminus, a cysteine-rich domain (aa 837–890) near the C-terminus, and a coiled-coil domain (aa 488–508) in the central region (). Thus, we named this protein Rubicon for RUN domain and cysteine-rich domain containing, Beclin 1-interacting protein
. No sequence homology was observed between Rubicon and Vps38 or Atg14 (data not shown). Noticeably, the protein levels of affinity purified Beclin 1-EGFP, p150/Vps15, Vps34 and UVRAG and are comparable and reproducibly higher than those of Atg14L and Rubicon (), suggesting a stable “core” Beclin 1-Vps34 complex consisting of Beclin 1, Vps34, p150 and UVRAG. Additionally, we did not detect the previously identified Beclin 1-associated proteins, such as nPIST6
, thus raising a possibility that their interactions with Beclin 1 may be relatively unstable, transient or occur only under specific conditions.
We next examined the specific binding of Atg14L or Rubicon to Beclin 1 in transfected mammalian cells. We showed that FLAG- or EGFP-tagged Atg14L or Rubicon co-immunoprecipitated with endogenous Beclin 1 (Fig. S3b–c) as well as Vps34 (Fig. S3d-e). We also constructed a series of deletion mutants to analyze the sequence domains required for Beclin 1-Atg14L/Rubicon associations (Fig. S3f). We found that while the coiled-coil domain (CCD) of Beclin 1 is sufficient for the binding of Atg14L, the CCD and evolutionarily conserved domain (ECD) of Beclin 1 are necessary for the binding of Rubicon (Fig. S3g-h). Furthermore, both CCD domains of Atg14L are required for the efficient binding of Atg14L to Beclin 1 and Vps34 (Fig. S3i), whereas the central region of Rubicon containing the CCD domain is important for the binding of Rubicon to Beclin 1 and Vps34 (Fig. S3j). Interestingly, the RUN or cysteine-rich domain of Rubicon appears to be inhibitory to the binding of Rubicon to Beclin 1 and Vps34 (Fig. S3j).
We then characterized the composition of the Beclin 1 complexes. Using anti-Atg14L and anti-Rubicon antibodies, we found that Atg14L is co-immunoprecipitated with Vps34 (Fig. S4a) and Beclin 1 (data not shown), but not with Rubicon (Fig. S4a); Rubicon is co-immunoprecipitated with Vps34 and UVRAG, but not with Atg14L (Fig. S4a). Therefore, Atg14L and Rubicon appear to exist in separate Beclin 1 complexes.
We performed gel filtration experiments with the tissue extract prepared from either wild-type () or “rescued” (Fig. S4b) mouse liver. For each sample, eighty fractions of the eluent were collected and analyzed by immunoblotting. We found that the endogenous Vps34, Beclin 1 (or Beclin 1-EGFP), Atg14L and Rubicon proteins were primarily co-eluted in the fractions 38–45 ( & S4b), suggesting that these fractions contain a major Beclin 1-Vps34 complex (size > 700 kD) that includes both Atg14L and Rubicon. We also performed gel filtration experiments with the cell lysate prepared from stable cell lines expressing either Atg14L-EGFP (Fig. S4c) or Rubicon-EGFP (Fig. S4d). Again, endogenous Beclin 1 was co-eluted with Atg14L-EGFP (Fig. S4c) and Rubicon-EGFP (Fig. S4d). Interestingly, starvation treatments of these stable cells did not affect the elution profiles of Atg14L-EGFP, Rubicon-EGFP and Beclin 1 (Fig. S4c-d).
To test the possibility that Atg14L and Rubicon are in separate protein complexes while co-eluted, we added anti-Rubicon antibody to the tissue extract before the gel filtration run and immunoblotted the resulting fractions with anti-Atg14L antibody. Our data show that Atg14L was co-eluted with the anti-Rubicon antibody (Fig. S4e, bands labeled by *, ** and ***), suggesting that the binding of anti-Rubicon antibody to, and therefore the presence of Rubicon in, the Atg14L-containing complex.
Moreover, we observed mutual co-immunoprecipitation (co-IP) of Atg14L and Rubicon from transfected cells, further supporting that Atg14L and Rubicon can be present in the same protein complex (); UVRAG was also co-immunoprecipitated with Atg14L or Rubicon, and the interaction between UVRAG and Rubicon is significantly enhanced in the presence of Beclin 1 (Fig. S4f-h).
Taken together, we conclude that Atg14L, Rubicon, UVRAG, Beclin 1, p150/Vps15 and Vps34 can form a major Beclin 1-Vps34 complex in vivo. However, Atg14L was also eluted in the later fractions (51–56) containing Beclin 1 or Beclin 1-EGFP (but not Rubicon) ( & S4b), suggesting that Atg14L is also associated with a smaller Beclin 1 complex without Rubicon.
We performed several autophagy assays to dissect the role of Atg14L in autophagy. First, we knocked down Atg14L expression in cultured cells by RNA interference (RNAi) and analyzed the levels of LC3II, a lipid-conjugated form of LC3 that is normally localized on autophagosomes, by immunoblotting 16–18
. Like Beclin 1 siRNA, Atg14L siRNA caused increased levels of LC3II, as compared to control siRNA (, left). Second, we examined levels of p62/SQSTM1, a known autophagy substrate that is normally accumulated upon autophagy impairment19–21
. Again, like Beclin 1 siRNA, Atg14L siRNA resulted in increased p62/SQSTM1 levels (, left), and the increase in p62/SQSTM1 and LC3II levels upon the Beclin 1 or Atg14L siRNA treatment was also significant after starvation (, right). Therefore, knocking-down of Atg14L or Beclin 1 impaired the autophagy-mediated clearance of p62/SQSTM1 and LC3II.
Atg14L positively regulates autophagy; Beclin 1 and Atg14L synergistically promote double membrane formation
Third, we knocked down Atg14L expression in MLE12 cells that stably expressed GFP-LC3. In control siRNA-treated cells, many small GFP-LC3 puncta were observed, reflecting the presence of basal levels of autophagosomes (Fig. S5a, left). In contrast, Atg14L siRNA transfection resulted in accumulation of large-size GFP-LC3 puncta (Fig. S5a, right). These large GFP-LC3 puncta were co-colocalized with p62/SQSTM1 (Fig. S5b) and ubiquitin (Fig. S5c), indicating that these are ubiquitinated protein inclusions as previously shown in Atg5
-deficient mouse-tissues22, 23
. Ultrastructural analysis showed that Atg14L siRNA transfection results in reduced number of autophagosomes (data not shown). These above analyses suggest that reduced Atg14L expression abolishes autophagosome formation and increases ubiquitinated protein levels.
Fourth, as compared to control siRNA treatment, Atg14L siRNA treatment caused a slight decrease in the rate of long-lived protein degradation (~10%, p=0.007) under nutrient-rich conditions and a strong reduction in the rate of long-lived protein degradation (~37%, p=5E-6) after nutrient withdrawal; and this effect of Atg14L siRNA treatment was diminished in the presence of 3-methyladenine (3MA), an inhibitor of autophagy ().
Fifth, we investigated whether Atg14L modulates Vps34 kinase activity using a novel kinase assay which included Vps15/p15024
. Our result showed that co-expression of FLAG-Atg14L with myc-Vps34-Vps15 plasmids resulted in Vps34 activity that was 2.5-fold as high as resulted from co-expression of control FLAG with myc-Vps34-Vps15 plasmids (). Interestingly, the Atg14L-mediated stimulation of Vps34 activity occurred only when co-expressing Beclin 1. This result suggests that over-expression of Atg14L enhances Vps34 activity in a Beclin 1-dependent manner.
We observed that Atg14L-EGFP or Beclin 1-EGFP stably expressed in cells was primarily diffuse in cytoplasm (Fig. S5d), which is consistent with the previous report in Beclin 1-EGFP transgenic tissues25
. However, co-expression of Atg14L-EGFP and Beclin 1-AsRed resulted in their colocalization on punctate structures (). Electron microscopy (EM) analysis of these Atg14L-EGFP and Beclin 1-AsRed co-transfected cells revealed many large size “organelles” (~3–5 μm) (). Some of these structures display concentric “rings” with double membranes (, asterisks); many are large vacuole-like structures filled with materials of high electron density (, asterisks) and enwrapped with double-membranes ( inset), which are readily distinguishable from typical aggresomes or protein aggregates (usually not associated with limiting membranes). These structures are positive for Atg14L-EGFP, as shown by immuno-EM (). We also observed increased number of autophagosomes in these transfected cells (, blue arrows).
We next studied the nature of these Beclin 1-Atg14L-resided structures. We found that these structures were negative for Golgi (Fig. S5e) or ER (Fig. S5f) markers. In contrast, they were co-localized with GFP-LC3 (Fig. S5g), suggesting that these Beclin 1-Atg14L structures likely recruit LC3. Moreover, they were co-localized with co-expressed EGFP-Atg12 () or EGFP-Atg5 (), suggesting that these Beclin 1-Atg14L structures may be involved in the early steps of autophagosome biosynthesis by recruiting Atg12 and Atg5.
To investigate the role of Rubicon in autophagy, we knocked down endogenous Rubicon protein levels. In contrast to Atg14L or Beclin 1 siRNA, Rubicon siRNA caused reduced steady-state levels of LC3II and p62/SQSTM under normal or nutrient starvation conditions (), suggesting that knock-down of Rubicon promotes autophagic activity. Conversely, in cells stably or transiently transfected with Rubicon-EGFP, the p62/SQSTM1 protein levels were markedly enhanced as compared to those in the cells transfected with EGFP (), suggesting that over-expression of Rubicon inhibits autophagy.
Rubicon is a negative regulator of autophagy
To examine whether Rubicon also modulates the Vps34 lipid kinase activity, we performed the above described lipid kinase assay. Our result showed that co-expression of FLAG-Rubicon with myc-Vps34-Vps15 resulted in remarkably reduced Vps34 activity, but only in the absence of Beclin 1-EGFP over-expression (). This result suggests that over-expression of Rubicon inhibits the Vps34 kinase activity, and this effect does not require Beclin 1..
Previously mCherry-GFP-LC3 has been used to examine autophagosome maturation, e.g., autophagosome acidification following fusion with late endosomes/lysosomes26
. We found that cells co-expressing mCherry-GFP-LC3 and FLAG-Rubicon contained primarily yellow fluorescent mCherry-GFP-LC3 puncta (immature autophagosomes) (, lower panel, white arrows), whereas cells expressing only mCherry-GFP-LC3 (, lower panel, yellow arrows) or co-expressing mCherry-GFP-LC3 and control vector FLAG (, upper panel) contained considerable numbers of red fluorescent mCherry-GFP-LC3 puncta (mature autophagosomes) (). This result suggests that over-expression of Rubicon may block autophagy through inhibiting autophagosome maturation.
Interestingly, Rubicon-EGFP (or FLAG-Rubicon, data not shown) expression exhibited punctate subcellular localization (). While the Rubicon-EGFP puncta (some were “ring”-shaped), were occasionally labeled with the early endosomal marker EEA1 (Fig. S5h), they were primarily co-localized with the late endosomal/lysosomal marker Lamp1 (). Moreover, some of the Rubicon-EGFP puncta were positively stained with an antibody against lysobisphosphatidic acid (LBPA) (), an unusual eukaryotic lipid found only in multi-vesicular body (MVB)27
, suggesting that some of the Rubicon-EGFP structures may be related to MVB28
. EM analysis of Rubicon-EGFP-transfected cells showed many abnormal large vacuole-like structures (1–5 μm in diameter) (). Some of these structures contained high electron density (Fig. 4c1–2, orange arrows), characteristic of late endosomes/lysosomes; some had relatively less content with overall low electron density, which may represent enlarged early stage endosomes (, black arrows); notably, some appeared to enclose numerous small vesicles of multiple-layers (, purple arrows), while others resembled MVB28
(, blue arrows). Through immuno-EM using anti-GFP gold particles, we observed that these vacuole-like structures in the Rubicon-EGFP-transfected cells were positive for Rubicon-EGFP; moreover, Rubicon-EGFP was associated with the limiting membranes of these particular structures (). Therefore, these structures corresponded to the fluorescent Rubicon-EGFP puncta (). In addition, our immuno-EM result also confirmed the co-localization of Rubicon-EGFP and Lamp1 at ultrastructural level ().
Over-expressing Rubicon causes aberrant expansion of late endosomes/lysosomes
Bioinformatic analysis revealed that the cysteine-rich domain of Rubicon shares sequence homology to the FYVE domain (), a well-characterized motif specific for PtdIns(3)P binding29
. When examined experimentally, unlike the control PtdIns(3)P-binding protein 2×FYVE-EGFP, Rubicon-EGFP was not pulled down by PtdIns(3)P-conjugated sepharose beads (data not shown). However, co-expressed Rubicon-AsRed and p40 (phox)-PX-EGFP, another reporter for PtdIns(3)P binding, showed extensive co-localization (, upper), suggesting that the Rubicon-associated structures are enriched in PtdIns(3)P. Moreover, wortmanin, an inhibitor of Vps34 kinase, effectively dispersed the p40 (phox)-PX-EGFP puncta but not the Rubicon-AsRed structures (, lower), suggesting that the maintenance of these Rubicon-associated structures does not depend on PtdIns(3)P. Furthermore, through immunofluorescent imaging of Rubicon truncation mutants, we found that the cysteine-rich domain of Rubicon is required for the formation of the Rubicon-positive structures that are enriched in PtdIns(3)P and associated with the aberrant endosomes/lysosomes (). Finally, we found that the distribution of Beclin 1-AsRed or FLAG-Beclin 1-CE was exclusive from Rubicon-EGFP puncta (); and Beclin 1 siRNA treatment did not affect the formation of the Rubicon-EGFP puncta (). Therefore, the formation of these Rubicon-associated late endosomal-lysosomal structures are Beclin 1-independent.
Over-expressed Rubicon is localized on PtdIns(3)P enriched-structures in a Beclin 1-independent manner
In summary, our study uncovers Atg14L and Rubicon, two novel components in the Beclin 1-Vps34 protein complexes, and reveals their distinct roles in regulating autophagy and Vps34 kinase activity. We show that Atg14L and Rubicon may regulate autophagy via modulating Vps34 activity. Our study also suggests the existence of multiple Beclin 1 protein complexes that are engaged in distinct functions in autophagy regulation (). The significance of these distinct Beclin 1 complexes remains to be fully elucidated. The dynamic change in protein composition between different functional Beclin 1-Vps34 complexes may play a central role in mediating the Beclin 1-Vps34 activity that governs multiple cellular events including autophagy.