DFCP1 binds and hydrolyses ATP
The main structural features of DFCP1 are two C-terminal FYVE domains, which, in conjunction with an ER-targeting region, bind to PtdIns3P on ER domains and by this marks the sites of autophagosome formation 4 (Fig. 1a). In contrast, the N-terminus of DFCP1 does not carry any characterized domains, but we found it interesting that it contains a P-Loop motif 4. A search in the Interpro databases revealed that the DFCP1 domain structure, including the P-loop, the ER-binding domain and the two FYVE domains, is an ancient architecture which evolved more than 500 million years ago, with homologues in most metazoan phyla (Extended Data Fig. 1a,b). P-loops interact with phosphorylated nucleotides, but no function of the N-terminal domain has so far been described for DFCP1.
Structure prediction using AlphaFold 11, as well as structural homology modelling using the Phyre2 server 12 revealed that the DFCP1 P-Loop domain shows structural homology to large nucleotide binding proteins, especially to the GTPases Atlastin and GBP1 (Fig. 1b,c, Extended Data Fig. 1c). Based on this finding, we tested if DFCP1 is able to bind nucleotides. We purified the N-terminus of DFCP1, including the predicted nucleotide-binding domain (Extended Data Fig. 2a). To measure nucleotide binding, we used N-methylanthraniloyl (mant)-modified nucleotides, mantATP and mantGTP, which are not fluorescent in solution but emit fluorescence upon binding to a protein. We observed no binding to GTP within the time course of our measurements, whereas Cdc42 – a GTP-binding protein – showed robust binding to GTP (Extended Data Fig. 2b). In contrast, DFCP1 efficiently and specifically bound to ATP (Fig. 1d). DFCP1 also bound to ADP (Extended Data Fig. 2c), suggesting that it could act as a molecular switch, depending on the loaded nucleotide. Binding of ATP/ADP to DFCP1 was rapid, with the reaction reaching saturation within minutes (Fig. 1d). To address whether DFCP1 has ATPase activity, we measured DFCP1-dependent phosphate release from ATP. Indeed, adding ATP to the purified DFCP1 ATPase domain resulted in the release of free phosphate (Fig. 1e). Thus, DFCP1 is a functional ATPase.
To understand the functional importance of the ATP binding and hydrolysis by DFCP1, we aimed to generate mutations that affect these biochemical properties. We used a Phyre2-generated homology model based on Atlastin, which showed the highest degree of sequence similarity (Fig. 1b,c). Using this model, we performed structural alignments against other nucleotide-binding proteins (Atlastin, GBP1, N-RAS and K-RAS) to identify key residues necessary for nucleotide binding and hydrolysis.
NTPases have a characteristic GKS motif, which is critical for nucleotide binding by coordinating a magnesium ion 13. We identified this motif at residues 192–194 of DFCP1. Based on homologies with the small GTPases K-RAS and N-RAS, we further identified the residues T189 and G190 as potentially critical residues. T189 is at the same position as the G12 residue of K-RAS, whereas G190 corresponds to G13 in K-RAS. Both amino acids are critical for K-RAS GTPase activity (Fig. 1b,c).
Based on these predictions, we generated and purified DFCP1 nucleotide-binding defective mutants (Extended Data Fig. 2a). Similar to other nucleotide binding proteins, mutation of the GKS motif (K193A, S194N) resulted in a loss of ATP binding (Fig. 1d). The same was the case with mutations of the K-Ras G12/G13 analogous amino acids (T189V and G190V) (Fig. 1d).
We also aimed to identify an ATPase-defective DFCP1 mutant and found that the mutation T189V, which corresponds to the hydrolysis-defective K-Ras G12V allele, showed slightly reduced hydrolysis activity, but showed also reduced nucleotide binding (Fig. 1d,e). Importantly, however, another mutation of T189, T189A, showed robust nucleotide binding but only minimal ATP hydrolysis activity (Fig. 1f,g), thus constituting an ATP-locked DFCP1 allele.
DFCP1 dimerizes upon ATP binding
Many NTPases, such as Dynamin and Atlastin, oligomerize as a consequence of nucleotide binding 14. To elucidate if DFCP1 shows nucleotide-dependent oligomerization, we loaded the purified DFCP1 N-terminus with ADP or the non-hydrolysable ATP analogue, ATPγS, and performed size exclusion chromatography. Interestingly, whereas ADP-bound DFCP1 migrated as a monomer, ATPγS-loaded DFCP1 migrated as dimer (Fig. 1h, Extended Data Fig. 2d). As a control, we performed the same assay using DFCP1 K193A, the nucleotide-binding defective mutant we identified. This mutant migrated as a monomer and did not dimerize in the presence of ATPγS (Fig. 1h, Extended Data Fig. 2d), in line with our findings that dimerization of DFCP1 is ATP-dependent. Surprisingly, DFCP1 T189A, which binds ATP but is hydrolysis-deficient, also migrated as a monomer in the presence of ATPγS (Fig. 1h, Extended Data Fig. 2d). This suggests that dimerization of DFCP1 could be required for ATP hydrolysis.
DFCP1 ATP binding and hydrolysis are necessary for efficient omegasome constriction
To address whether cells expressing the DFCP1 mutants show defects in omegasome formation, we generated a knockout (KO)-based complementation system (Extended Data Fig. 3a-d). DFCP1 KO cells were transduced with lentiviral vectors expressing low levels of either mNeonGreen (mNG)-tagged DFCP1 wild-type (WT) or the two ATP binding or hydrolysis mutants K193A or T189A, in addition to mCherry- (mCh)-p62 as an autophagosomal marker. Structured illumination microscopy (SIM) revealed that mNG-DFCP1 WT localized to ring shaped omegasomes containing LC3B and p62 (Fig. 2a). To assess omegasome dynamics, the cells were starved with EBSS for 15 min and images were acquired for 10 minutes with one frame taken every 2 s in starvation conditions (Fig. 2b, movie 1).
Autophagosome formation at omegasomes is a reproducible course of events 4,15. Based on published data and the tracking of DFCP1 WT omegasomes by live microscopy, we divided the omegasome formation process into three phases – initiation, maturation, and constriction (Fig. 2b). The initiation phase is characterized by a DFCP1 spot which is formed de novo, and to which p62 and LC3B are recruited a few seconds later 15,16 Fig. 2b, Extended Data Fig. 4d). The spot grows and forms a ring-like structure, as previously observed 4 15. We have defined the appearance of a ring with a visible lumen as the start for the maturation phase. LC3B and p62 are recruited to the initial DFCP1 spot and during the maturation, the DFCP1 ring grows, and LC3B and p62 span the lumen forming a disk-like structure (Fig. 2a; Extended Data Fig. 4b,e 15–17). At the end of maturation, a p62 and LC3B-labelled phagophore extrudes out of the ring and forms a pocket (Fig. 2a; Extended Data Fig. 4b,e). This separation of the DFCP1-positive omegasome from the p62/LC3B positive phagophore initiates the third phase – the constriction of the omegasome. The phagophore bends, buds out and is separated from the DFCP1-positive omegasome, thereby forming the nascent autophagosome. During this process, the shrinking omegasome remains connected to the growing autophagosome, which is positive for both p62 and LC3B, but also weakly for DFCP1 (Fig. 2a; Extended Data Fig. 4b,e). Finally, the weak DFCP1 signal leaves the autophagosome, and DFCP1 at the collapsed omegasome disappears, leaving a p62/LC3B-positive-finished autophagosome (Fig. 2a).
We next analysed the dynamic formation of omegasomes in DFCP1 mutant cells. Whereas the omegasomes appeared morphologically similar to the WT omegasomes (Extended Data Fig. 4b,e), detailed analysis and tracking revealed a significant delay in the omegasome formation process (Fig. 2c, Extended Data Fig. 4a, movie2). While DFCP1 WT used approximately 330 sec to form omegasomes as previously described 16 4,15, the ATP-binding mutant DFCP1 K193A used 500 sec, whereas the ATPase-defective mutant DFCP1 T189A needed 450 sec to complete the process.
To understand which step in omegasome biogenesis was affected in the mutants, we measured the duration of the three individual phases. Omegasomes in cells expressing DFCP1 WT used approximately 100 sec for each of the two first phases, whereas the last phase took 150 sec. Surprisingly, cells expressing either of the two DFCP1 mutants showed a nearly unchanged duration for the first two phases. In contrast, the constriction phase was markedly delayed compared to WT (Fig. 2d, Extended Data Fig. 4c). We confirmed these findings also using DFCP1 knockdown cell lines, which were rescued with stable expression of low levels of siRNA resistant WT or mutant mNG-DFCP1 in combination with SNAP-LC3B (Extended Data Fig. 3e-g, Extended Data Fig. 4d-i, movie 3, movie 4). Based on these data, we conclude that the ATP binding and hydrolysis activities of DFCP1 are necessary for the efficient constriction of omegasomes.
DFCP1 ATPase mutants cause increased numbers of omegasomes
Based on our finding that DFCP1 ATPase is required for omegasome constriction, we asked which consequences it has for autophagy. Upon starvation with EBSS, cells expressing the ATP-binding or ATPase-defective mutants showed an increased number of DFCP1 puncta compared to DFCP1 WT, and both mutants had a higher number of larger omegasomes (Fig. 3a, b).
The increased number of omegasomes in the DFCP1 mutants can be either a consequence of increased omegasome formation or prolonged persistence. To investigate whether omegasome formation was increased, we made use of another omegasome marker, WIPI2 9,10. WIPI2 is recruited by PtdIns3P to phagophores originating at omegasomes and appears nearly at the same time as DFCP1 but dissociates earlier 9. Thus, WIPI2 represents a good marker for omegasome formation independently of DFCP1. We analysed endogenous WIPI2 puncta 2 hrs after EBSS treatment in DFCP1 WT and mutants. As expected, a portion of the WIPI2 dots co-localized with DFCP1 positive omegasomes, likely representing early stages of autophagosome formation (Extended Data Fig. 5a). Interestingly, although the number of DFCP1 puncta was higher in cells expressing the mutants, neither the number of WIPI2 puncta nor the portion of DFCP1 positive WIPI2 puncta was changed (Extended Data Fig. 5b), suggesting that omegasome formation was not increased. On the other hand, the portion of DFCP1 puncta positive for WIP2 was reduced in cells expressing the mutants, indicating that the DFCP1 positive omegasomes that accumulated in the mutant cells likely represent a later, WIPI2 negative, stage of omegasomes. We conclude that whereas the number of omegasomes increases in the DFCP1 mutant cells, the onset of omegasome formation is not affected.
Autophagic flux of p62 is compromised in DFCP1 ATPase mutants
It has remained a paradox that depletion of DFCP1 by siRNA does not have any effect on the flux of LC3B (Axe 2008). However, how LC3B flux is affected upon specific modulate of DFCP1 ATPase activity was not clear. First, we measured the sum intensity of endogenous LC3B puncta in DFCP1 WT and DFCP1 mutant rescue cells by automated analysis of confocal micrographs. As expected, numerous LC3B puncta were detected upon EBSS starvation in DFCP1 WT cells (Extended Data Fig. 5c). Interestingly, DFCP1 ATP-binding deficient K193A mutant cells showed a small, but significant increase in the sum intensity of LC3B dots, whereas cells expressing the ATP-locked T189A mutant only showed a tendency to accumulate LC3B as compared to DFCP1 WT (Fig. 3c, Extended Data Fig. 5c).
It has been reported that p62 localizes to phagophores which are formed at omegasomes, and that this recruitment occurs independently of LC3B 18. In line with this, we noticed that while LC3B was present on many other structures in addition to omegasomes, the majority of p62 puncta localized with DFCP1 positive omegasomes (Extended Data Fig. 5c). Importantly, when we measured the sum intensity of endogenous p62 puncta, both DFCP1 ATPase mutant cell lines showed a more than two-fold increase in p62 dot intensity as compared to DFCP1 WT cells upon EBSS treatment (Fig. 3d). Moreover, in cells expressing high amounts of DFCP1, p62 hyperaccumulated inside DFCP1 ATPase deficient omegasomes (Extended Data Fig. 5d).
To address whether the increased level of p62 dots observed in the mutants could be explained by an impaired autophagic flux, we measured the sum intensity of p62 dots per cell in fed or EBSS starved cells in the presence or absence of Bafilomycin A1, which inhibits lysosomal degradation activity. This analysis confirmed our previous result, with a close to three-fold increase in p62 dot sum intensity in the EBSS treated mutant cells (Fig. 3e). In addition, we observed a similar effect in fed cells, indicating that also basal autophagy was impaired in DFCP1 depleted or ATPase deficient cells (Fig. 3e, Extended Data Fig. 6a-e). Addition of Bafilomycin A1 indeed prevented the lysosomal degradation of p62, as the sum intensity of p62 dots clearly increased (Fig. 3e). Importantly, when lysosomal degradation was blocked with Bafilomycin A1, there was no difference in the p62 levels between DFCP1 WT or ATPase defective cells (Fig. 3e). This indicates that p62 accumulates in the DFCP1 ATP-binding and ATPase defective mutants due to a delayed autophagic flux, rather than increased onset of autophagy, consistent with our finding that the number of WIPI2 puncta were unaffected. Taken together, our data show that DFCP1 is required for efficient autophagic flux of p62, and that this depends on its ability to bind and hydrolyse ATP.
DFCP1 ATPase is required for selective autophagy
As expected from our and published 4 results on LC3B lipidation upon DFCP1 depletion, we found that DFCP1 is not involved in bulk autophagy, using cytoplasmic mKeima reporter assays 19,20 in U2OS and RPE-1 cells (Extended Data Fig. 6f-i). However, since we observed that p62 accumulates in DFCP1 ATPase mutants in both fed and starved conditions, we asked if the DFCP1 ATPase mutants are impaired in selective autophagy.
p62 has been implicated in several types of selective autophagy, such as aggrephagy and mitophagy 21–23. To induce aggrephagy, we treated stably expressing mCh-DFCP1 cells with Puromycin. Puromycin treatment leads to the formation of large p62-containing aggregates, which are heavily ubiquitinated and cleared by autophagy 24. Notably, DFCP1 localized to a subset of these aggregates, forming characteristic omegasomes positive for LC3B (Fig. 4a, Extended Data Fig. 7a,b). Strikingly, DFCP1 KO cells had more than a two-fold increase in p62 aggregates upon Puromycin treatment compared to the WT (Fig. 4b). This could be rescued by WT DFCP1, but not the ATPase mutants (Fig. 4c, Extended Data Fig. 7c). Thus, DFCP1 and its ATPase activity are necessary for efficient aggrephagy.
To investigate if DFCP1 is necessary for autophagic degradation of mitochondria, we used the iron chelator Deferiprone (DFP) to induce mitophagy 25 in cells stably expressing mNG-DFCP1 and a mitochondrial marker (mitoSNAP) and performed live imaging. We observed that DFCP1 formed omegasomes at the surface of mitochondria, and that mitochondrial fragments were channelled through the omegasome ring (Fig. 4.d, Extended Data Fig. 7d, movie 5, movie 6). Importantly, the mitochondrial fragments were surrounded by the weakly DFCP1-positive phagophore, indicating that it had been engulfed by the autophagosome (insets Fig. 4d, Extended Data Fig. 7d). Following engulfment and constriction of the omegasome, autophagosomes were released from the mitochondria (Extended Data Fig. 7d).
The importance of DFCP1 for mitophagy was quantified using RPE-1 cells stably expressing a mitochondrial mKeima probe, mito-mKeima 19,26. When mito-mKeima is transported to acidic lysosomes for degradation, the excitation maximum changes and mitophagy can be determined by flow cytometry as an increased ratio of lysosomal mito-mKeima. While control cells displayed a nearly two-fold increase in lysosomal mito-mKeima when mitophagy was induced, DFCP1 depleted cells had a strongly reduced capacity to deliver mito-mKeima to lysosomes (Fig. 4e, Extended Data Fig. 7e).
To address whether DFCP1 ATPase activity is necessary for mitophagy, we stably expressed DFCP1 WT or mutants in the mito-mKeima RPE-1 cells depleted of endogenous DFCP1 and performed mitophagy assays (Fig. 4f, Extended Data Fig. 7f,g). Both ATP-binding and ATPase defective mutants of DFCP1 were impaired in mitophagy (Fig. 4f, Extended Data Fig. 7f,g), indicating that the ATPase activity of DFCP1 is necessary for efficient mitophagy.
When examining the phenotype of U2OS DFCP1 KO cells, we noticed that they contained an increased number of micronuclei. Image quantifications showed that DFCP1 KO and ATPase deficient cells had a more than two-fold increase in the number of micronuclei per cell as compared to parental or DFCP1 WT cells (Fig. 4h, Extended Data Fig. 7h). This phenotype was also confirmed in two independent clones of A431 DFCP1 KO cells (Extended Data Fig. 7k,l). Similarly, U2OS cells acutely depleted for DFCP1 showed an increase in the number of micronuclei per cell (Extended Data Fig. 7i,j). Importantly, we noticed that DFCP1 localized to a subset of micronuclei, colocalizing with p62. Super-resolution microscopy revealed that endogenously GFP-tagged DFCP1 formed characteristic omegasome rings at the surface of micronuclei together with p62 (Fig. 4g). These results suggest that DFCP1 promotes autophagic degradation of micronuclei, in addition to mitochondria and protein aggregates.
DFCP1 associates with ubiquitinated proteins
How could it be explained mechanistically that DFCP1 mediates selective autophagy whereas it is dispensable for bulk autophagy? We noticed that omegasomes stain positive with antibodies against conjugated ubiquitin (Extended Data Fig. 8a, c). As in the case of aggrephagy (Fig. 4a), also omegasomes decorating mitochondria or micronuclei were positive for ubiquitin (Extended Data Fig. 8a, c), raising the possibility that DFCP1 could interact with ubiquitinated cargoes. To investigate this notion, we co-transfected HeLa cells with Myc-epitope-tagged ubiquitin and GFP-DFCP1 and investigated whether affinity isolated GFP-DFCP1 would co-precipitate ubiquitinated proteins. Interestingly, whereas no ubiquitinated proteins co-precipitated with GFP as revealed by immunoblotting with anti-Myc, a smear of Myc-Ubiquitin positive bands was pulled down with GFP-DFCP1 (Fig. 4i, Extended Data Fig. 8b). This indicates that DFCP1 associates with ubiquitinated cargoes and provides a plausible model for its function.