Nix interacts with WIPI2 to induce mitophagy

Abstract

Nix is a membrane‐anchored outer mitochondrial protein that induces mitophagy. While Nix has an LC3‐interacting (LIR) motif that binds to ATG8 proteins, it also contains a minimal essential region (MER) that induces mitophagy through an unknown mechanism. We used chemically induced dimerization (CID) to probe the mechanism of Nix‐mediated mitophagy and found that both the LIR and MER are required for robust mitophagy. We find that the Nix MER interacts with the autophagy effector WIPI2 and recruits WIPI2 to mitochondria. The Nix LIR motif is also required for robust mitophagy and converts a homogeneous WIPI2 distribution on the surface of the mitochondria into puncta, even in the absence of ATG8s. Together, this work reveals unanticipated mechanisms in Nix‐induced mitophagy and the elusive role of the MER, while also describing an interesting example of autophagy induction that acts downstream of the canonical initiation complexes.

The minimal essential region of Nix recruits WIPI2 to mitochondria for initiation of autophagy that does not need ATG8 proteins.

Introduction

Autophagy is the engulfment of intracellular material into double‐membraned structures termed autophagosomes and subsequent fusion with lysosomes for degradation. A canonical order of events in autophagy has been determined, with the initiation complex consisting of FIP200, Ulk1 or Ulk2, Atg13, and Atg101 recruiting the PI3K nucleation complex which generates PI3P (Itakura & Mizushima, 2010; Kishi‐Itakura et al, 2014), which recruits WIPI proteins, which also bind ATG16L1 (Dooley et al, 2014), followed by the Atg12 conjugation machinery and ATG8 lipidation machinery (Mizushima et al, 1998; Ichimura et al, 2000; reviewed by Dikic & Elazar, 2018). There are six ATG8 proteins in human cells (hATG8s) consisting of LC3A, LC3B, LC3C, GABARAP, GABARAPL1, and GABARAPL2, which are attached to the growing autophagosome while keeping exposed a region that can bind to the LC3‐interacting region (LIR) of target cargo (Noda et al, 2008; Wirth et al, 2019). Initiation of this process can be controlled at the cellular level by AMPK, which has an activating phosphorylation site on ULK1/2, and mTOR, which inhibits ULK1/2 via phosphorylation of different residues (Kim et al, 2011).

Mitophagy is a crucial process for removing mitochondria by selective autophagy. PINK1 and PRKN gene mutations are linked to Parkinson's disease and code for proteins well‐studied in this process. The Pink1‐Parkin pathway is a damage‐sensing method of coating impaired mitochondria with ubiquitin, tagging defective mitochondria for degradation, and is therefore considered a means of mitochondrial quality control in cells. The ubiquitin chains on these mitochondria are then bound to autophagy receptors Optineurin (OPTN) and NDP52 (Wong & Holzbaur, 2014; Heo et al, 2015; Lazarou et al, 2015), which can bind multiple autophagy proteins including hATG8s. In the case of Pink1‐Parkin mitophagy, ubiquitin chains recruit OPTN and NDP52, which either recruit the Ulk1 complex via FIP200 (Ravenhill et al, 2019; Vargas et al, 2019; Zhou et al, 2021) or ATG9 (Yamano et al, 2020) and the VPS34 complex (Nguyen et al, 2023). Local concentration of the Ulk1 complex on cargo is sufficient for trans‐autophosphorylation of Ulk1 and autophagy induction without input from AMPK (Vargas et al, 2019). While OPTN and NDP52 both have LC3‐interacting regions (LIRs), hATG8 proteins are not required for Pink1‐Parkin–induced autophagy initiation (Vargas et al, 2019) or engulfment into autophagosomes (Padman et al, 2017) (but GABARAPs are required for autophagosome fusion with lysosomes; Nguyen et al, 2016).

Other autophagy receptors that are not part of the SLR (sequestosome‐like receptors) family include integral membrane‐spanning proteins localized to the ER or mitochondria. On the outer mitochondrial membrane, homologs BNIP3 and BNIP3L/Nix (herein referred to as Nix) are required for certain developmental mitophagy processes (Schweers et al, 2007; Sandoval et al, 2008; Brennan et al, 2018; Ordureau et al, 2021). The role of Nix was first found in reticulocytes as necessary for clearance of mitochondria during erythroid differentiation in mice (Schweers et al, 2007). Nix was later found to have an LIR domain capable of binding hATG8s (Schwarten et al, 2009; Novak et al, 2010), although with substantially lower binding affinity than the p62 LIR (Novak et al, 2010). Generating phosphomimetic mutations in the Nix homolog BNIP3 around the LIR was found to enhance binding of LC3 to BNIP3 (Zhu et al, 2013), implying a role for post‐translational modifications modulating BNIP3 activity. Both BNIP3 and Nix can homo‐ and heterodimerize through their transmembrane regions (Ohi et al, 1999; Sulistijo & MacKenzie, 2006) and this dimerization may also be affected by phosphorylation (Liu & Frazier, 2015). The requirement of the Nix LIR has been debated—in reticulocyte maturation, a minimal essential region (MER) is necessary for mitochondrial clearance and the LIR can be completely removed with only partial effects on mitophagy (Zhang et al, 2012). BNIP3 was shown to induce LC3 lipidation even when overexpressed with an LIR mutation (Hanna et al, 2012), indicating that there may be another form of autophagy induction beyond binding to ATG8 proteins. Proposed mechanisms for Nix‐induced autophagy induction include competitive binding with Beclin‐1 for Bcl‐2 (Zhang et al, 2008; Weindel et al, 2022) and binding to the mTOR activator Rheb (Li et al, 2007), although Nix did not have the same effect on mTOR signaling as BNIP3 (Li et al, 2007). Thus, there remain key open questions on the mechanism of Nix‐mediated mitophagy.

Results

Engineering a minimal system to directly study Nix‐mediated mitophagy

Nix is constitutively anchored in the outer mitochondrial membrane via a hydrophobic domain near the C‐terminus, as is the homologous protein BNIP3. We designed an overexpression system to study Nix without co‐inducing BNIP3 expression and, to explore the molecular mechanism of Nix with greater temporal specificity, we inducibly targeted it to mitochondria using chemically induced dimerization (CID; Vargas et al, 2019). The CID system was designed by replacing the C‐terminus including the transmembrane domain with a FKBP dimer and monitored by conjugating mEGFP to the N‐terminus (EGFP‐Nix_1–188‐FKBP). This construct was stably expressed in HeLa cells (Fig EV1A) also expressing FRB fused to the Fis1 C‐terminus (Fis1_93–152) that localizes to the mitochondrial outer membrane (FRB‐Fis1T; Fig 1A) and mKeima targeted to the mitochondria matrix (herein referred to as mtKeima) as a pH‐dependent reporter of mitophagy. Nix conjugated to FKBP can be targeted to the FRB construct on mitochondria when cells are treated with the A/C heterodimerizer agent, an analog of rapamycin incapable of inhibiting mTOR (Bayle et al, 2006; herein referred to as Rapalog). Stably expressed EGFP‐Nix_1–188‐FKBP was observed in the cytoplasm of untreated cells and upon Rapalog treatment targeted to mitochondria, (Fig 1B and C). EGFP‐Nix_1–188‐FKBP in the absence of the transmembrane domain induces a steadily increasing mitophagy signal (Fig 1B and D) corroborated by FACS analysis of mtKeima (Fig EV1B and C). We tested whether the orientation of Nix relative to the mitochondrial membrane is important for activity by placing the FKBP on the N‐terminus of Nix as well as the related protein BNIP3 and found no significant difference in mitophagy between C‐terminal‐ and N‐terminal‐targeted Nix or N‐terminal‐targeted BNIP3 (Fig EV1B and C). Notably, there was a substantial lag from the time of Nix recruitment peaking at 3 h to the time of maximal mitophagy seen after 20 h (Fig 1C and D), implicating slower mitophagy dynamics than with PINK1/Parkin.

Figure EV1. CID of Nix localizes to mitochondria and induces mitophagy.

1. Western blot comparing endogenous Nix in HeLa cells, untreated or with 24 h phenanthroline (50 μM), to overexpressed EGFP‐Nix_1–188‐FKBP.
2. FACS plots of untreated (top) and 24‐h Rapalog‐treated (bottom) mtKeima cells expressing EGFP‐Nix_1–188‐FKBP (left), FKBP‐EGFP‐Nix_1–188 (middle), or FKBPEGFP‐BNIP31‐163 (right).
3. FACS analysis of mtKeima of Nix with FKBP attached via the C‐terminus or N‐terminus and compared to BNIP3 with FKBP attached via the N‐terminus with either no treatment for 24 h or Rapalog, indicating percentage of cells above threshold line.
4. mtKeima analysis from microscopy images of CID‐Nix‐expressing cells after either no treatment or 24 h of Rapalog in different knockout backgrounds. 5KO cells lack NDP52, OPTN, TAX1BP1, p62, and NBR1. 6KO cells lack all six hATG8 (LC3 and GABARAP) proteins.
5. Recruitment of EGFP‐Nix_1–188‐FKBP to FRB‐Fis1T by Pearson's correlation coefficient between KeimaEx488 and EGFP of the cells quantified in Panel D.
6. Confirmation of Ulk1 ko in Ulk1/2 dko cells, Ulk2 confirmed by sequencing but not by western blot.
7. Time course of mtKeima reported mitophagy in WT or Ulk1/2 dko cells either with or without a previously validated TBK1 inhibitor GSK8612.

Data Information: Area shading indicates standard deviation between wells. Over 1,000 cells are quantified per condition. Statistics: two‐way ANOVA; ** indicates a P value < 0.01, **** indicates a P‐value < 0.0001. ns indicates not significant. Experiments are repeated at least three times.

Figure 1. CID of Nix localizes to mitochondria and induces mitophagy.

1. Schematic of full‐length Nix (left) compared to CID Nix (right) and their mitochondrial localization.
2. Representative images of EGFP‐Nix‐FKBP (green) Rapalog‐induced recruitment to mtKeima‐labeled mitochondria (Keima excitation 488, cyan) and subsequent mitophagy (Keima excitation 561, magenta). Scale bar: 10 μm.
3. Recruitment dynamics of CID Nix after different treatments measured by Pearson's correlation between EGFP and Keima Exc. 488.
4. Mitophagy as indicated by percent mtKeima pixels in EGFP‐positive cells over a threshold set from no treatment.
5. Confirmation of decreased mitochondrial proteins in WT but not in FIP200 ko cells expressing CID Nix, either untreated or treated for 24 h with Rapalog by immunoblotting.
6. Quantification of MTCO2 relative to actin over three independent experiments after treatment with Rapalog over 24 h by immunoblotting.
   Cells in this figure express mtKeima‐P2A‐FRB‐Fis1tT and EGFP‐Nix_1–188‐FKBP.

Data Information: Area shading indicates standard deviation between wells. Over 1,000 cells are quantified per condition. Statistics performed by two‐way ANOVA; ** indicates a P‐value < 0.01; *** indicates a P‐value < 0.001. Experiments are repeated at least three times.

Source data are available online for this figure.

To confirm degradation of mitochondria, we performed CID of Nix over 24 h and assessed levels of mitochondrial proteins by Western blot and found a decrease in MTCO2 in WT, but not in FIP200 KO cells (Fig 1E and F). To further confirm that canonical autophagy machinery is important for CID‐Nix‐mediated mitophagy, we expressed EGFP‐Nix_1–188‐FKBP in FIP200 KO and 6KO (lacking all hATG8 proteins) cells (Nguyen et al, 2016) and, because of their essential role in canonical autophagy, saw no induction of mitophagy following Rapalog treatment (Fig EV1D) despite equal recruitment of Nix in all cell lines (Fig EV1E). We also assayed cells lacking five other autophagy receptors—Tax1BP1, OPTN, NDP52, SQSTM1 (p62), and NBR1 (5KO; Lazarou et al, 2015)—and found that there was no decrease in mitophagy (Fig EV1D), consistent with the idea that Nix functions as an organelle‐anchored autophagy receptor that acts independently of five other major cytosolic autophagy receptors. While we saw a complete block in mitophagy with an Ulk1 inhibitor (Fig 1D) to rule out off‐target effects, we used Ulk1/2 DKO cells (Fig EV1F) and found a nearly complete block of mitophagy by CID‐Nix (Fig EV1G). Because TBK1 can be redundant with Ulk1/2 in some cases (Nguyen et al, 2023), we used a TBK1 inhibitor in the presence and absence of Ulk1/2 and found no effect on CID‐Nix‐mediated mitophagy (Fig EV1G). This indicates that Ulk1/2, but not TBK1, is important for Nix‐mediated mitophagy.

The transmembrane domains of Nix and BNIP3 not only target Nix and BNIP3 to mitochondria but they also mediate homodimerization or heterodimerization with BNIP3 or Bcl‐2. Such homodimerization of BNIP3 or Nix has been reported to affect mitophagy and hATG8 affinity (Hanna et al, 2012; Marinković et al, 2020). CID‐induced targeting of Nix lacking the C‐terminal membrane‐spanning domain indicates that transmembrane‐mediated dimerization is not necessary for mitophagy induction in this overexpression system. While dimerization may play a role in increasing the rate or magnitude of mitophagy, we continued with monomeric CID Nix to identify which factors are necessary for activity beyond known modulators of Nix activity.

Nix induces mitophagy through the minimal essential region (MER)

Nix is thought to promote mitophagy via binding of its LIR motif to lipidated hATG8 moieties on phagophores (Novak et al, 2010). Recent work on other autophagy receptors such as NDP52 and p62, once thought to mediate mitophagy via LIR LC3 interaction, reveals that they bind upstream autophagy machinery in addition to hATG8s to initiate selective autophagy (Ravenhill et al, 2019; Turco et al, 2019; Vargas et al, 2019). To assess the role of the LIR in Nix‐mediated mitophagy using our CID approach, we performed mitophagy assays with Nix containing key mutations in its LIR domain (Fig EV2A, LIR disruption, right box) and Nix with an N‐terminal truncation that completely removes the LIR (Nix_71–188, Fig 2A). Surprisingly, we found about half the mtKeima measured mitophagy activity remains intact when the key LIR residues are mutated. However, a truncation of the N‐terminus including the LIR motif up to the MER domain (Nix_71–188) removes nearly all detectable mitophagy using the CID system (Fig EV2B).

Figure EV2. Nix domains are required for mitophagy despite apparent lack of structure.

1. Schematic of different residues mutated to disrupt a possible FIR or the LIR of Nix.
2. Percent mtKeima cells by FACS over ratiometric threshold after 24 h of Rapalog treatment indicating relative mitophagy between WT and LIR mutants.
3. Percent mtKeima cells by FACS over ratiometric threshold after 24 h of Rapalog treatment indicating relative mitophagy between each FIR mutation with or without Rapalog.
4. Coomassie gel showing purity of Nix_1–188 for circular dichroism and NMR.
5. The circular dichroism spectrum of Nix‐C1 displays a single minimum of around 200 nm. This is consistent with the random coil nature of Nix_1–188.
6. The 1HN‐15N HSQC NMR spectrum of Nix shows very little dispersion with amide proton chemical shifts ranging from 6.6 to 8.8 ppm. In addition, the side‐chain indole signals for the three tryptophan are all overlapping around 10 ppm. These all indicate a random coil conformation of Nix_1–188.
7. All truncation mutants in Fig 2B are recruited with the same dynamics and to the same magnitude as indicated by Pearson correlation coefficient between GFP and mtKeima Exc. 488.
8. Recruitment dynamics of Nix_1–70 upon different treatments by Pearson's correlation coefficient between KeimaExc488 and EGFP of the cells quantified in Fig 2C.
9. Recruitment dynamics of Nix_1–87 upon different treatments by Pearson's correlation coefficient between KeimaExc488 and EGFP of the cells quantified in Fig 2D.

Data Information: Over 1,000 cells quantified per condition. Statistics: two‐way ANOVA; ** indicates a P‐value < 0.01, *** indicates a P‐value < 0.001. Experiments are repeated at least three times.

Figure 2. Truncation mutants reveal the Nix MER and LIR are both required for mitophagy.

1. Schematic of truncation mutants compared to full length (top). C‐terminal truncations (left) and N‐terminal truncations (right).
2. Mitophagy time course by mtKeima of different Nix truncation mutants recruited with CID, measured by percentage of Keima‐positive pixels above threshold.
3. Mitophagy by mtKeima of Nix_1–70 when treated with Torin, Rapalog, or both compared to control.
4. Mitophagy by mtKeima of Nix_1–87 when treated with Torin, Rapalog, or both compared to control.

Data Information: Area shading indicates standard deviation between wells. Over 1,000 cells are quantified per condition. Experiments are repeated at least three times.

Source data are available online for this figure.

To understand how mutating the essential LIR residues only showed a partial effect compared to an LIR truncation, we asked if the Nix LIR may also be a FIP200‐interacting region (FIR), a motif very similar to the LIR motif but with different key residues (Zhou et al, 2021). Indeed, Nix has two serine residues N‐terminal to the LIR domain that, if phosphorylated, would match canonical FIRs (Zhou et al, 2021). To remove potential FIR activity without disrupting the putative LIR, we mutated Serines 34 and 35 to glycine (Fig EV2A, FIR modulation, left box)—a common residue in that position for an LIR (Johansen & Lamark, 2020) but not an FIR (Zhou et al, 2021). To enhance potential FIP200 binding, we also mutated the Serine residues to a phosphomimetic glutamate (Zhou et al, 2021). While S34G and S35G Nix had a somewhat lower magnitude, all constructs retained similar degrees of increased mitophagy with Rapalog (Fig EV2C). These results indicate that, although the region containing the LIR is important for mitophagy, the mechanism by which mitophagy is induced is not through hATG8 binding the LIR motif or FIP200 binding an FIR motif, although these interactions may modulate the magnitude of mitophagy.

To identify which regions of Nix are required for mitophagy, we aimed to perform truncations from either terminus. However, removal of large regions of a protein can prevent proper folding, making it difficult to identify important domains and motifs. Interestingly, the AlphaFold model for Nix (AF‐O60238‐F1; Jumper et al, 2021; Varadi et al, 2022) shows that most of the protein is predicted to be unstructured, apart from the transmembrane domain, the BH3 domain, and the MER domain, which are predicted to be alpha helical. To test this structure model, we performed circular dichroism on purified Nix_1–188, lacking its transmembrane domain (Fig EV2D), and found a single minimum around 200 nm that would indicate the absence of an alpha helix or beta sheet (Fig EV2E). To confirm a lack of structure, we performed NMR spectroscopy and found very little dispersion in its NMR resonances, typical of random coil with no obvious structure at all in the absence of binding partners or post‐translational modifications (Fig EV2F). This apparent lack of secondary or tertiary structure emboldened us to perform large‐scale truncations for identification of motifs that may facilitate mitophagy.

We found above that Nix induces mitophagy to the same extent regardless of its orientation to the mitochondrial membrane (Fig EV1B and C), allowing the freedom to link both EGFP and FKBP to either the C‐ or N‐terminal side of Nix. We used this finding to simplify the construction of a series of truncation mutants with FKBP‐EGFP tagged to the N‐terminus (Fig 2A). We used these truncations, from both the N‐ and C‐termini, to explore which regions of Nix induce mitophagy in the tightly controlled CID system. We found that both the MER and LIR domains of Nix are required for mitophagy when localized to mitochondria with Rapalog, as Nix_1–70 had no increase and Nix_60–188 had a marginal increase in mitophagy, whereas most of the protein can be deleted with minimal effect as with Nix_1–87 (Fig 2B) despite nearly identical recruitment of the Nix variants to mitochondria (Fig EV2G). While the LIR of Nix has long been considered capable of inducing mitophagy by binding to hATG8 proteins, the MER domain has an unknown role in Nix mechanism, despite having been found essential for its function in vivo (Zhang et al, 2012).

BNIP3 has been reported to bind Rheb, an activator of mTOR, causing inhibition of mTOR (Li et al, 2007) and activation of general autophagy at a cellular scale. We considered that the two regions of Nix could serve two roles: one to induce bulk autophagy via the MER and the other acting as an attachment point between hATG8 proteins and the mitochondrial surface by the LIR. To test this hypothesis, we targeted a truncated form of Nix containing the LIR but lacking the MER, Nix_1–70, to mitochondria with Rapalog and used Torin to inhibit mTOR and induce global autophagy. We found high recruitment of Nix to mitochondria when treated with Rapalog or Rapalog and Torin but no recruitment with Torin alone (Fig EV2H). While there is some increase in mitochondrial autophagy with the construct containing the LIR but lacking the MER (Fig 2C), it is very small compared to Nix containing an MER domain treated with Rapalog alone (Fig 2D). The two constructs showed similar recruitment dynamics over the first 6 h but then Nix_1–87 showed decreased colocalization, presumably due to quenching of EGFP upon mitophagy (Fig EV2H and I). Furthermore, we found a large increase in Torin‐treated cells when the LIR and MER are targeted to mitochondria (Fig 2D), implying that activation of global autophagy potentiates Nix‐mediated mitophagy instead of vice versa. Together, these data indicate that the LIR and MER have a large effect on mitophagy that cannot be reconstituted with only an LIR plus mTOR inhibition and that mTOR inhibition allows for a substantial increase in Nix‐mediated mitophagy.

Nix stimulates mitophagy by binding WIPI2

To identify effectors through which the MER stimulates mitophagy, we utilized an Immunoprecipitation coupled with mass spectrometry (IP‐MS) approach. We compared interacting proteins of mitophagy‐deficient Nix_1–70 (only containing the LIR domain) and mitophagy‐competent Nix_1–87, containing LIR and MER motifs. Intriguingly, WIPI2 was the only member of the core autophagy machinery enriched in NIX_1–87 IPs. In addition, WIPI2 ranked among the interactors with the highest significance difference between Nix_1–87 and Nix_1–70 (Fig 3A and B). Validation by IP Western blot also shows a strong pulldown of WIPI2 by Nix_1–188 and by Nix_1–87, but not Nix_1–70 lacking the MER domain (Fig 3C). This suggests the MER may function to recruit WIPI2 to mitochondria to mediate mitophagy.

Figure 3. Nix binds to WIPI2 via the MER.

1. Volcano plot generated from one‐tailed t‐test of proteins that interact with Nix_1–87 more so than Nix_1–70 by IP‐mass spectrometry. Red dots indicate proteins over twofold enrichment and P‐value < 0.05
2. Table of top significant interactors from Fig 4A.
3. IP western blot of FKBP‐EGFP‐Nix_1–188, Nix_1–87, or Nix_1–70, comparing 1% input to GFP‐pulldown and blotted against GFP or WIPI2 after 4 h of Rapalog.
4. HeLa cells expressing FKBP‐EGFP‐ p62_327–348 or FKBP‐EGFP‐WIPI2b were treated for 16 h with phenanthroline to induce endogenous Nix and BNIP3 expression, then underwent a GFP pulldown and compared to 1% input by immunoblot.
5. IP western blot of FKBP‐EGFP‐Nix_1–188 compared to Nix_60–188, comparing 1% input to GFP pulldown and blotted against GFP or WIPI2 after 4 h of Rapalog.

Data Information: Statistics: one‐tailed unpaired Student's t‐test. Fold change calculated from spectral counts (sc) by (sc_1–87 + 1)/(sc_1–70 + 1) so that zero values could be included. IP experiments were repeated at least three times.

Source data are available online for this figure.

To confirm this interaction with endogenous Nix, we performed a GFP pulldown in FKBP‐GFP‐WIPI2 cells treated for 16 h with phenanthroline to induce endogenous Nix and BNIP3 expression (Fig 3D). We found that WIPI2 pulls down both Nix and BNIP3, but a similar FKBP‐EGFP construct containing p62_327–348 does not. This indicates that the WIPI2‐Nix interaction is not due to the CID system and is conserved between Nix and its homolog BNIP3.

Because recruitment of WIPI2 to mitochondria is sufficient to induce robust mitophagy (Le Guerroué et al, 2023), whereas Nix lacking the N‐terminus and LIR does not show substantial mitophagy (Fig 2B: Nix_60–188), we questioned if the LIR may also be important for WIPI2 binding. GFP pulldown of Nix_60–188, which lacks the LIR but retains the MER (Fig 2A), shows a decrease in the amount of WIPI2 pulled down, in comparison to Nix_1–188 (Fig 3E). Together these data show that the MER domain in Nix is essential for efficient interaction with WIPI2 and that Nix amino acids 1–59 augment binding.

Nix recruits WIPI2 to mitochondria independent of common autophagy proteins

Nix has long been thought to initiate mitophagy by binding to hATG8 proteins (Novak et al, 2010), so to confirm that WIPI2 recruitment is upstream of this event, we performed immunofluorescence experiments in the absence of all hATG8 proteins (6KO cells). Using CID of Nix_1–188 in WT or 6KO cells, we monitored endogenous WIPI2 puncta forming on mitochondria and found a substantial increase regardless of the presence of hATG8 proteins (Fig 4A and B). Because Nix was found to bind more tightly to GABARAP than to LC3B (Novak et al, 2010), we also measured endogenous GABARAP puncta formation on mitochondria in the presence or absence of WIPI2 and found a significant increase in WT, but not in WIPI2 KO cells (Fig 4C and D). To further explore GABARAP recruitment, we performed CID of Nix_1–87 or Nix _1–70 to determine if the Nix LIR is sufficient to recruit GABARAP and found an increase in GABARAP puncta on mitochondria when the MER is present (Nix_1–87), but not with the LIR alone (Nix_1–70; Fig 4E). Because FIP200 is recruited by some autophagy receptors to initiate selective autophagy (Ravenhill et al, 2019; Turco et al, 2019; Vargas et al, 2019; Zhou et al, 2021), we also aimed to determine if Nix may bind WIPI2 indirectly through interacting with FIP200. While we were unable to determine FIP200 binding to Nix via an IP because of the appearance of non‐specific bands, we did find that WIPI2 interacts with Nix in both WT and FIP200 KO cells (Fig 4F). This supports the hypothesis that Nix directly interacts with WIPI2 for its recruitment to mitochondria.

Figure 4. Nix binds WIPI2 independently of hATG8 or FIP200 proteins.

1. Representative images of WT or 6ko HeLa cells expressing EGFP‐Nix_1–188‐FKBP (green), mtKeima, and FRB‐Fis1T with or without 4 h of Rapalog, immunostained for endogenous Tom20 (red) and WIPI2 (white). Scale bar: 50 μm.
2. Automated puncta counting of WIPI2 in WT or 6ko cells normalized to cell number in control conditions or after 4 h with Rapalog.
3. Representative images of WT or WIPI2 ko HeLa cells expressing EGFP‐Nix_1–188‐FKBP (green), mtKeima, and FRB‐Fis1T with or without 4 h of Rapalog immunostained for endogenous Tom20 (red) and GABARAP (white). Scale bar: 50 μm.
4. Automated puncta counting of GABARAP in WT or WIPI2 ko cells normalized to cell number in control conditions or after 4 h with Rapalog.
5. Automated puncta counting of GABARAP in WT cells expressing FKBP‐EGFP‐Nix_1–87 or Nix_1–70 normalized to cell number in control conditions or after 4 h with Rapalog.
6. GFP pulldown of EGFP‐Nix_1–188‐FKBP in WT, WIPI2 ko, or FIP200 ko cells after 4 h of Rapalog treatment. FIP200 cropped due to the appearance of non‐specific bands after IP.

Data Information: Error bars indicate standard deviation. Statistics performed by two‐way ANOVA; ** indicates a P‐value < 0.01, *** indicates a P‐value < 0.001. ns indicates not significant. Over 1,000 cells are quantified per condition. Experiments are repeated at least three times.

Source data are available online for this figure.

To confirm that WIPI2 is required for Nix‐mediated mitophagy, we monitored mtKeima in WT or WIPI2 KO cells treated with the iron‐chelating agent phenanthroline (Fig EV3A), and found a complete block in pH4 signal in the absence of WIPI2. Because recent data also implicate Nix in mediating pexophagy (Wilhelm et al, 2022; Barone et al, 2023), we attached Keima to an SKL sequence for peroxisome targeting and stably expressed it in WT and WIPI2 KO cells and treated with phenanthroline, finding a measurable amount of pexophagy, and finding a complete block of pexophagy in WIPI2 KO cells (Fig EV3B).

Figure EV3. WIPI2 is essential for Nix‐mediated autophagy. CID of Nix and WIPI2 have varied dynamics from the p62 FIR.

1. Time course of mitophagy reported by mtKeima in WT or WIPI2 ko HeLa cells untreated or treated with phenanthroline (50 μM).
2. Time course of pexophagy reported by Keima‐SKL in WT or WIPI2 ko HeLa cells untreated or treated with phenanthroline (50 μM).
3. Mitophagy dynamics of EGFP‐Nix‐FKBP treated with Rapalog or Rapalog and Torin compared to control.
4. Mitophagy dynamics of FKBP‐EGFP‐WIPI2b treated with Rapalog or Rapalog and Torin compared to control.
5. Mitophagy dynamics of FKBP‐EGFP‐p62FBD treated with Rapalog or Rapalog and Torin compared to control.
6. Slope of mitophagy from Panels C‐E between 6 and 12 h.

Data Information: Area shading indicates standard deviation between wells. Over 1,000 cells are quantified per condition. Statistics: two‐way ANOVA; * indicates a P‐value < 0.05, ** indicates a P‐value < 0.01. Experiments are repeated at least three times.

We have previously seen that WIPI2 recruitment by CID is sufficient for mitophagy induction (Le Guerroué et al, 2023), and that inhibition of mTOR by co‐treatment of Torin with Rapalog dramatically increased the rate and magnitude of mitophagy by Nix_1–87 (Fig 2D). We assessed if there is a difference in mitophagy dynamics between using CID to directly recruit WIPI2b to mitochondria, using CID of Nix to indirectly recruit WIPI2 to mitochondria, and CID of the p62 FIR domain, p62_327–348, which recruits FIP200 to mitochondria (Turco et al, 2019). We compared the CID systems of the three using Rapalog or Rapalog co‐treated with Torin. CID of WIPI2b showed very similar dynamics and magnitude to that of Nix_1–188 (Fig EV3C, D, and F). In contrast, CID of p62_327–348 had both a higher rate and slightly larger effect on mitophagy than Nix (Fig EV3E and F). We found that despite p62_327–348 having the highest rate of mitophagy when treated with Rapalog alone, Torin treatment had a more modest improvement in mitophagy compared to CID of either WIPI2b or Nix, which showed a more substantial increase to an even greater rate of mitophagy than p62_327–348 (Fig EV3F). These data indicate that CID of Nix has nearly identical dynamics to CID of WIPI2, and both are dissimilar to FIP200 recruitment by p62.

To further characterize the recruitment of endogenous WIPI2 to mitochondria, we assessed Halo‐WIPI2 knock‐in U2OS cells generously shared by the Schmidt Lab (Broadbent et al, 2023). We compared different truncation mutants in their ability to induce WIPI2 puncta formation and recruit WIPI2 to mitochondria (Fig 5A). Nix_1–70 appears to be incapable of recruiting WIPI2 to mitochondria or mediating WIPI2 puncta formation (Fig 5A–C). CID of Nix_1–87 induced a large increase in WIPI2 puncta on mitochondria (Fig 5A and B) and co‐localizes with mitochondria via Pearson's correlation (Fig 5A and C). Interestingly, we found that CID of Nix_60–188 did not increase WIPI2 puncta formation (Fig 5A and B) despite showing a large degree of recruitment of WIPI2 to mitochondria upon Rapalog treatment (Fig 5A and C). Thus, as seen also by co‐IP and mitophagy experiments, the amino acids 1–59 of Nix are not required for WIPI2 binding, but function otherwise to mediate WIPI2 puncta formation and mitophagy independent of LC3 family protein binding.

Figure 5. Nix recruits WIPI2 to mitochondria via MER.

1. Representative images of endogenous Halo‐WIPI2 in U2OS cells stained with JF646 Halo dye (magenta)‐expressing mtKeima (cyan), FRB‐Fis1T, FKBP‐EGFP‐Nix_1–87, and FKBP‐EGFP‐Nix_60–188 or FKBP‐EGFP‐Nix_1–70 (green) after 16 h of Rapalog treatment.
2. Automated puncta counting of endogenous Halo‐WIPI2 in U2OS cells expressing FKBP‐EGFP‐Nix_1–87 (LIR and MER), Nix_60–188 (MER), or Nix_1–70 (LIR) after 16 h of Rapalog treatment then fixed.
3. Pearson's correlation coefficient between Halo‐WIPI2 and mtKeima pH7 signal in cells is described in Panel B.

Data Information: Error bars indicate standard deviation between wells. Over 1,000 cells are quantified per condition. Statistics performed by two‐way ANOVA; * indicates a P‐value < 0.05, ** indicates a P‐value < 0.01, **** indicates a P‐value < 0.0001. ns indicates not significant. Experiments are repeated at least three times.

Source data are available online for this figure.

AlphaFold predicted binding interface between Nix and WIPI2

With evidence supporting the direct interaction of WIPI2 and Nix, we used AlphaFold2 (Jumper et al, 2021) to generate a model of this interaction with the recently developed multimer settings (preprint: Evans et al, 2022). We found that Nix is predicted to have very little secondary or tertiary structure (Fig EV4A), much like the AlphaFold model of Nix alone (AF‐O60238‐F1; Jumper et al, 2021; Varadi et al, 2022) and our CD and NMR data (Fig EV2E and F), whereas WIPI2d modeled similarly to its crystal structure (Strong et al, 2021). In the multimer model of Nix and WIPI2d, the only region of Nix predicted to interact with WIPI2d is the MER domain and neighboring residues, which interact at the same location on WIPI2d as ATG16L1 (Figs 6A and EV4B; Dooley et al, 2014; Strong et al, 2021). The majority of Nix modeled with fairly low confidence (Fig EV4C) and with high predicted aligned error (PAE; Fig EV4D), consistent with it being disordered. The MER and adjacent residues, however, showed low PAE in reference to WIPI2d residues (Fig EV4D) as they consistently aligned to the same region of WIPI2d across each generated model (Fig EV4E). These models predict that L75 of Nix inserts into the same groove of WIPI2d as leucine residues of ATG16L1 (Strong et al, 2021) between blades 2 and 3 (Figs 6B and EV4B). Furthermore, WIPI2d residues K88, R108, R125, and K128 form electrostatic interactions with Nix residues D77, E81, Q79, and E72 (Fig 6B).

Figure EV4. AlphaFold Multimer Model of Nix with WIPI2d finds interaction between MER and WIPI2d.

1. Full proteins modeled by AlphaFold2 shown as aquamarine cartoon for Nix except for residues 60–87 shown in red. WIPI2d shown in green cartoon format with transparent gray surface.
2. Closer view of the protein interface showing Nix L75 inserting into a groove of WIPI2d. WIPI2d hydrophobic residues L68, I92, and I124 colored are colored in yellow; WIPI2d‐positive residues K88, R108, and R125 are colored in blue; the rest of WIPI2d presented in gray.
3. Heatmap with overlapping plot showing pLDDT scored for Nix and WIPI2d in the model presented in Panels A and B and Fig 6.
4. Predicted aligned errors (PAE) for Nix and WIPI2d residues compared to other residues in multimer model.
5. Overlapping aligned structures for each of the five models generated by AlphaFold2, each showing the Nix MER in the same groove of WIPI2d.

Figure 6. A ribbon representation of the AlphaFold model of Nix interaction with WIPI2d.

1. Left panel is a top view of the complex between Nix and WIPI2d. The WIPI2d regions consisting of the beta strands forming the blades are color coded based on their blade numbers. The helices between residues W273‐Y287 and S290‐N297 are colored in gray, as well as the flexible loops connecting the WIPI2d blades. The region of Nix predicted to bind WIPI2d as a helix is shown in cyan. For clarity, only residues I66 to S87 of Nix are displayed. Right panel is rotated 90° relative to the left panel.
2. A zoom‐in region of the interaction site between Nix and WIPI2d. Right panel is 90° rotated relative to the left panel. Only WIPI2d blades 2 and 3 are visible. Residues crucial for the interaction between the two proteins are shown as sticks. Residues in WIPI2d blade 2 are colored orange, whereas those in blade 3 are green. Nix residues are in cyan. Those belonging to WIPI2d are labeled in red, while Nix residues are in black. There is clear charge complementarity between D77 of Nix and K88 of WIPI2d, E81, and R108, as well as E72 and K128. In addition, Nix residues L75 and A78 form close hydrophobic contacts with WIPI2d residues L69, V83, I92, and I124. Finally, there is a potential hydrogen bond formation between the side chain of Q79 of Nix and backbone of I124 of WIPI2d.
3. A heatmap indicating the calculated distance between each of all Nix residues (top) and WIPI2 protein in the AlphaFold predicted model, where red indicates amino acids at the interface between both proteins. The bottom heatmap shows an annotated zoom of Nix residues 50–92 with a plot representing Ångstrom distance between each residue and WIPI2 overlaid on the heatmap. L75 is highlighted as a residue of interest.
4. A heatmap indicated the calculated distance between each of all WIPI2 residues (top) and Nix protein as in Panel C. The bottom heatmap shows an annotated zoom of residues 60–138 with a distance plot overlaid. Residues R108 and R125 are highlighted as residues of interest.

Source data are available online for this figure.

Measuring the distance between each Nix residue from the WIPI2d chain reveals the interface between Nix and WIPI2d as Nix residues 58–82, and the only other region with relatively low distance is the transmembrane domain (Fig 6C), which AlphaFold predicted to be near the PI3P‐binding region of WIPI2d (Fig EV4A). While the AlphaFold model shows a high‐confidence interaction between the MER and WIPI2d, it found no proximity with the LIR or any residues upstream of Nix E58 (Fig 6C). This is inconsistent with biochemical data showing that Nix_60–188 bound far less tightly with WIPI2d than Nix_1–188 by IP (Fig 3E) and with the recruitment of Halo‐WIPI2 to mitochondria by Nix_60–188 assessed by imaging (Fig 5C), although it is unclear if this is a shortcoming of the multimer model or owing to a more complex interaction of the complex with another protein. We also measured the distance between each residue of WIPI2d from the Nix chain (Fig 6D), showing the full interface from the WIPI2d side of the interaction. From the measured distances in Fig 6C and D, and the interactions shown in Fig 6B, we hypothesized that mutating Nix L75 would prevent WIPI2 binding, and R108/R125 mutations that block WIPI2d interaction with ATG16L1 (Dooley et al, 2014; Strong et al, 2021) would block a Nix interaction as well.

To test our predicted model, we first compared full‐length EGFP‐Nix overexpressed as WT, with a double LIR point mutation (W36A/L39A), or with L75A mutation in the MER and measured mitophagy by mtKeima. We found that LIR point mutations significantly blocked mitophagy induced by Nix as previously reported (Novak et al, 2010; Ordureau et al, 2021), but not as substantially as the L75A mutation (Fig 7A and B). This experiment was performed in Nix/BNIP3 DKO cells to prevent any dimerization through the transmembrane tail to endogenous Nix and BNIP3 (Fig EV5A).

Figure 7. Investigation of residues predicted by AlphaFold to modulate Nix‐WIPI2 interaction.

1. FACS plots measuring mtKeima in Nix/BNIP3 dko HeLa cells expressing EGFP or EGFP conjugated to full‐length Nix as WT, or with W36A/L39A, or L75A mutations 1 week after viral transduction.
2. Percent cells above threshold line shown in Panel A as indication of mitophagy measured by mtKeima after 1 week of stably expressing EGFP, or EGFP‐Nix as WT or with W36A/L39A or L75A mutations.
3. Automated puncta counting of endogenous Halo‐WIPI2 in U2OS cells expressing FKBP‐EGFP‐Nix_1–188 as WT or with W36A/L39 (LIR) or L75A (MER) mutations after 16 h of Rapalog treatment then fixed.
4. Pearson's Correlation Coefficient between Halo‐WIPI2 and mtKeima pH7 signal in cells is described in panel D.
5. GFP pulldown of EGFP alone, EGFP‐Nix_1–188‐FKBP as WT, W36A/L39A, or L75A mutants after 4 h of Rapalog treatment.
6. GFP pulldown of FKBP‐EGFP‐WIPI2 as WT or with R108E/R125E mutations after 24 h of phenanthroline treatment.

Data Information: Error bars indicate standard deviation between wells. Over 1,000 cells are quantified per condition. Statistics: Panel B—two‐tailed unpaired student's t‐test, Panels C and D—two‐way ANOVA; ** indicates a P‐value < 0.01, **** indicates a P‐value < 0.0001. ns indicates not significant. Experiments are repeated at least three times.

Source data are available online for this figure.

Figure EV5. Investigation of residues predicted by AlphaFold to modulate Nix‐WIPI2 interaction.

1. Confirmation of Nix/BNIP3 dko HeLa cells by immunoblot.
2. Representative images of endogenous Halo‐WIPI2 (magenta) in U2OS cells expressing mtKeima (cyan), FRB‐Fis1T, and EGFP‐Nix_1–188‐FKBP as WT or with W36A/L39A or L75A mutations (green). Scale bar: 50 μm. Experiments are repeated at least three times.

We then compared WT Nix to W36A/L39A and L75A mutants in their ability to recruit WIPI2 to mitochondria using the CID system in U2OS cells with endogenously tagged Halo‐WIPI2 (Fig EV5B). WT Nix_1–188 induced WIPI2 puncta formation on mitochondria in contrast to both W36A/L39A and L75A Nix mutants that displayed no significant increase when treated with Rapalog (Figs 7C and EV5B), consistent with the truncation mutant experiments in Fig 5C. Mutation of L75 in the MER prevented recruitment of WIPI2 to mitochondria, whereas WT Nix and W36A/L39A mutations in the LIR retained WIPI2 recruitment to mitochondria upon Rapalog treatment (Figs 7D and EV5B).

We confirmed by co‐IP that the L75A mutation in full‐length Nix blocked WIPI2 interaction in Nix/BNIP3 DKO cells, and found that W36A/L39A Nix retained its ability to interact with WIPI2 (Fig 7E). To further assess the role of MER domain interaction with WIPI2, we mutated two amino acids in WIPI2b at the interface with the Nix MER domain (R108E/R125E highlighted in Fig 6D). We added phenanthroline to induce Nix and BNIP3 expression and found that mutations R108E/R125E in WIPI2b prevented both Nix and BNIP3 interactions by IP, as they do with ATG16L1 (Fig 7F). These data support the accuracy of the AlphaFold‐generated model and that the WIPI2‐Nix interaction is direct. These data also indicate that the Nix LIR, although important for both mitophagy and WIPI2 puncta formation, is not required for WIPI2 interaction.

Discussion

Selective autophagy receptors are considered proteins with an LIR motif allowing for attachment to lipidated hATG8 proteins and engulfment into a growing autophagosome. This model explains how soluble protein cargo turnover such as p62 and p62 clients are selectively turned over during steady‐state or starvation‐induced autophagy. However, this model does not indicate how selective autophagy would be initiated by the receptor binding to hATG8 proteins through the LIR. Furthermore, there is a great deal of competition for hATG8 binding as, based on the iLIR Autophagy Database (Jacomin et al, 2016), there are more than 6,000 proteins in the human genome that contains an LIR. Several autophagy receptors bind autophagy machinery proteins upstream of hATG8s in the canonical autophagy pathway, notably to FIP200 through an FIR domain (Ravenhill et al, 2019; Turco et al, 2019; Vargas et al, 2019; Zhou et al, 2021; Le Guerroué et al, 2023), which can facilitate the initiation of autophagy at the point of cargo engulfment, while also not requiring mTor inhibition or AMPK activation for autophagy initiation (Vargas et al, 2019). Because both BNIP3 and Nix can stimulate mitophagy, we explored a similar upstream autophagy machinery protein‐binding event. We found that the Nix MER functions by interacting with the autophagy protein WIPI2, solving a long‐lasting question in the Nix literature (Zhang et al, 2012; Ney, 2015). Furthermore, our AlphaFold model predicted that Nix L75 would support this interaction by inserting into a groove of WIPI2, which we validated by IP and imaging. This is the same residue in the MER (L74 in mouse Nix) that Paul Ney's group discovered to be essential for mitochondrial clearance during erythrogenesis (Zhang et al, 2012).

The Nix LIR may act as an attachment point for hATG8 proteins at the terminus of autophagy, but induction of mitochondrial autophagy appears to begin with WIPI2 binding the Nix MER domain. The Nix LIR also retains another function in inducing autophagy as, surprisingly, WIPI2 recruitment to mitochondria and WIPI2 puncta formation and subsequent mitophagy do not always align in our data. We find that truncating residues 1–59 from Nix or including point mutations in the LIR prevent both WIPI2 puncta formation (Figs 5B and 7C) and mitophagy (Figs 2B and 7A). In contrast, the 1–59 Nix deletion (Fig 3E) does not completely block WIPI2 binding and the LIR‐mutant Nix does not appear to inhibit WIPI2 binding (Fig 7E). These functions of the LIR, and other residues in Nix_1–59, in WIPI2 recruitment and mitophagy do not appear to be through hATG8 proteins as the 6KO (LC3 GABARAP null) cell line still displayed increased WIPI2 puncta upon CID of Nix_1–188 (Fig 4A and B). We consider two possible reasons that deleting Nix residues 1–59 decreases WIPI2 binding—either AlphaFold was unable to model this disordered region to a binding interface of WIPI2d or Nix_1–59 binds to another protein that helps stabilize interaction with WIPI2. The Nix LIR, however, is not required for WIPI2 interaction, but is required for robust mitophagy and WIPI2 puncta formation—we speculate that the LIR motif may be required for binding an autophagy‐promoting protein beyond hATG8 members.

While Nix binds WIPI2 upstream of ATG8 proteins in the mitophagy pathway, WIPI2 is canonically about halfway through the starvation‐induced autophagy pathway—upstream of ATG16L1 and the hATG8 conjugation machinery and downstream of the Ulk1 initiation complex and Vps34 complex (Fig 8). Similarly, OPTN has been found to initiate mitophagy in an atypical manner as it can bind to ATG9 (Yamano et al, 2020) and the PI3K complex via TBK1, which then recruits the FIP200 complex (Nguyen et al, 2023), revealing that recruitment of some components out of conventional order is also capable of initiating mitophagy. When the Fip200 initiation complex is recruited by NDP52 or p62, Ulk1/2 clustering could allow for trans‐autophosphorylation of Ulk1/2 (Bach et al, 2011; Vargas et al, 2019) and subsequent activation of selective autophagy (Fig 8, top). TBK1 can also autophosphorylate (Ma et al, 2012; Helgason et al, 2013) and play redundant roles with Ulk1/2 (Nguyen et al, 2023), so a similar mechanism through OPTN binding to TBK1 may exist in the case of OPTN‐stimulated autophagy. How autophagy is initiated when WIPI2 is recruited by Nix remains unclear, but it is possible that the Fip200 initiation complex and the downstream VPS34 complex are recruited by an interaction between ATG16L1 and FIP200 (Gammoh et al, 2013), although ATG16L1 and Nix bind WIPI2 at the same site, which might lead to competition between these steps.

Figure 8. Comparison of soluble autophagy receptor NDP52 recruitment of FIP200 (top) with Nix and BNIP3 recruitment of WIPI2 (bottom) to induce mitophagy.

Graphical diagram depicting typical selective autophagy initiation by NDP52 to FIP200 and subsequent trans‐autophosphorylation of Ulk1/2, followed by recruitment of VPS34 complex, PI3P, WIPI2, ATG16L1, and then the ATG12/ATG8 conjugation machinery (top). The bottom of the diagram shows BNIP3 and Nix recruiting WIPI2 for atypical autophagy initiation, and the known binders of WIPI2: Rab11, PI3P, and ATG16L1. Created with BioRender.com

To understand the mechanism of Nix‐mediated mitophagy, we engineered a minimal system to study Nix‐mediated mitophagy. By using an overexpression CID system, we either suppressed or completely abolished Nix regulatory steps such as transcriptional control, C‐tail recruitment to mitochondria or peroxisomes, dimerization, and potentially phosphorylation and degradation. Dimerization can increase Nix‐mediated mitophagy and now that we have uncovered monomeric Nix binding to WIPI1, it will be interesting to understand how dimerization may affect WIPI2 recruitment. Nix and BNIP3 are also heavily phosphorylated, which has been implicated both in their interaction with LC3 and in dimerization. What kinase or kinases phosphorylate Nix and how the Nix dimer may interact differently with autophagy components are important open questions to understanding Nix biology.

Limitations of the study

There are several limitations to this study. Notably, much of the work done on Nix, and all of the work using Nix mutations, are performed by overexpressing proteins—a necessary method when induction of endogenous protein by iron chelation or hypoxia would lead to increased BNIP3 expression possibly other relevant proteins as well. Much of this work is also performed with Nix which cannot form a dimer through its transmembrane domain, which may affect protein–protein interactions or activity in some way. We also note that mitophagy is measured by matrix‐targeted Keima in most cases. While we have substantial experience with this probe and performed the best controls available to us, we note that the mitochondrial matrix is a location with pH fluctuations through proton pumping. These limitations and the use of cell lines show that further work building upon this mechanism with endogenous protein in tissues or animals would be beneficial to the field.

Materials and Methods

Cell culture

Cells were grown in DMEM supplemented with 10% FBS, 1× sodium pyruvate, and 1x GlutaMax at 37°C, 5% CO₂, and 75% humidity. Cells were passaged with 0.05% Trypsin–EDTA and aliquots frozen in BamBanker at −80°C. Cells were tested monthly for mycoplasma. Rapalog treatments were done with 500 nM A/C heterodimerizing agents. For Torin‐1 treatments, cells were incubated with 250 nM. For Ulk1 inhibition, cells were treated with 1 μM Ulk1 inhibitor MRT68921. For TBK1 inhibition, cells were treated with 10 μM GSK8612 as in Le Guerroué et al (2023).

Cell lines

Nix/BNIP3 DKO cells were generated with CRISPR/Cas9 by gRNAs targeting exon 2 of each gene cloned into pSpCas9(BB)‐2A‐Puro (PX459) V2.0 (Addgene plasmid no. 62988) for Nix and BNIP3, respectively. HeLa cells were transiently transfected with these plasmids and selected with 1.5 μg/ml puromycin for 2 days. Surviving cells were then diluted to approximately one cell per well and grown in a 96‐well plate to isolate single colonies. Single colonies were screened by PCR of the targeted region and Sanger sequencing. Sequencing results were then analyzed by Interference of CRISPR Edits (ICE) through Synthego's online tool (https://ice.synthego.com/) to identify frame‐shifting deletions and insertions (InDels). The DKO clones were further validated by Western blot. Ulk1/2 DKO cells were generated similarly to Nix/BNIP3 DKO cells. gRNA sequences for each gene are listed in Appendix Table S1. Ulk1 deletion was validated by both Western blot and Sanger sequencing, and Ulk2 deletion was only validated by Sanger sequencing due to the lack of working commercial antibody. Halo‐WIPI2 knock‐in cells were graciously provided by Jens Schmidt and previously reported (Broadbent et al, 2023).

Cloning and viral transduction

All cloning was performed by PCR and Gibson assembly (New England Labs) in pHAGE vectors and verified by Sanger sequencing. pHAGE vectors were transfected into HEK293t cells with Tat1b, Rev1, MGMp2, and VSVg helper plasmids using polyethylenimine (PEI). After 2 days, virus was filtered (0.45 μm) and added to HeLa cells of the appropriate genetic backgrounds for infection. A list of plasmids used in this manuscript can be found in Appendix Table S1 and a detailed protocol was previously described (Wang, 2020).

FACS analysis and sorting

To establish full coverage of stable protein expression, we sorted cells for fluorescence of mtKeima and EGFP. All Nix constructs were sorted using the same gates for GFP and Keima. Some lines were re‐sorted to establish the same protein level after passaging. After sorting, cells were grown in penicillin–streptomycin–neomycin (PSN) antibiotic mixture for 5 days before switching back to antibiotic‐free media. No experiments were performed in antibiotic‐containing media.

For acquisition and analysis of mtKeima‐reported mitophagy, cells were measured with a CellStream flow cytometer (Luminex Corporation, CS‐100196). To determine percentage of cells undergoing mitophagy, we used a custom MATLAB script to automatically fit a line to the untreated cells, then raise the y‐intercept of that line until 1% of the population was above it. The line was then held constant for the treated sample, and percent cells above the threshold line were calculated. Separate experiments are displayed as the average across 3 days plus or minus standard deviation.

Live cell imaging

Cells were imaged on a Nikon Ti‐2 CSU‐W1 spinning disk confocal system operated by Nikon elements AR microscope imaging software either using a 20× air objective (NA 0.75) or a 40× water objective (NA 1.15) with the automated TI2‐N‐WID water immersion dispenser. The microscope is equipped with a live cell chamber that was set to 37°C and 50% CO₂ with constant humidity. Cells were imaged in MGB096‐1‐2‐LG‐L 96‐well imaging plates (DOT Scientific Inc). mKeima was excited with 488 and 561 lasers for neutral and acidic pH compartments, respectively, and each was assessed with a 698/70 nm emission filter.

Immunofluorescence

Cells seeded in 96‐well imaging plates described above were fixed with 4% warm paraformaldehyde (PFA) in PBS for 10 min, washed three times with PBS, permeabilized with 0.5% Triton‐X100 in PBS, blocked with 3% goat serum, 1% BSA, and 0.1% Triton‐X100 in PBS for 1 h. Cells were then incubated overnight in 1:500 dilution of primary antibodies in blocking solution on a rocker at 4°C. Cells were washed three times, incubated with a 1:1,000 dilution of secondary antibodies for 1 h at room temperature in blocking solution on a rocker, and then washed three times before imaging.

Image analysis

All image analysis was performed using custom MATLAB scripts. We found the median value of multiple sites within a well, then the average and standard deviation taken across three wells. Entire experiment was repeated at least three times.

For mitophagy analysis, images were automatically background corrected based on the mode pixel value for the lower half of the image histogram, then cells were segmented by a manual intensity threshold followed by size exclusion, and pixels within cells were assessed as above or below a manual ratiometric threshold. To generate the ratiometric threshold, pixels were plotted for untreated samples, and a line (with both slope and y‐intercept values) was generated to manually determine what values yielded approximately 1% mitophagy and maintained a slope parallel to the pixel population. The method of quantifying Keima pixels as fraction over a threshold to determine percent mitophagy is based on the method paper by Sun et al (2017).

For recruitment by colocalization, we performed background subtraction and cell segmentation as above. We then found that performing a singular Pearson correlation over all cells of an image was comparing bright cells against dim cells for each channel, so we used a water‐shedding algorithm to segment cells from each other and ran individual correlations for each segmented object. We then took the median of the individual correlation coefficients in the image for further analysis against other sites and wells. For live cell imaging of Halo dye JF646 in combination with mtKeima, we found crosstalk from Keima_Ex.561 into the 640 channel and thus performed a subtractive compensation determined from cells expressing mtKeima but not treated with dye. Fixed cells showed no crosstalk from Keima to the 640 channel.

Immunoprecipitation

Immunoprecipitation experiments for Western blots were performed with samples prepared from nearly confluent 10 cm dishes, treated for 4 h with Rapalog. We then followed the GFP‐Trap protocol but took 5 μl (1%) for input and eluted in 40 μl of 2× SDS buffer, which was split between two gels, as was the input.

For mass spectrometry, four 15 cm dishes were prepared for each replicate and two replicates per condition, also treated for 4 h with Rapalog. Each condition was incubated with 100 μl of beads instead of 25 μl to accommodate for sample size. Samples were then acid eluted as per the GFP‐Trap protocol then TCA precipitated overnight. Samples were then washed with acetone, dried with a speed vac, and resuspended in solubilizing buffer (5% SDS, 8 M urea, and 50 mM triethylammonium bicarbonate (TEAB), pH7.6) by shaking at 50°C. We then reduced the solution with 5 mM DTT at 50°C for 30 min then added methyl methanethiosulfonate (MMTS) to 25 mM and incubated at room temperature for 45 min before acidifying with 1% phosphoric acid. Acidified protein was then added with trypsin to a pre‐equilibrated S‐Trap column, washed, and then incubated with MS‐grade Trypsin overnight. Peptides were eluted with 50 mM TEAB, again with 0.2% formic acid, then a final time with 50% acetonitrile with 0.2% formic acid.

Immunoblotting

HeLa cells were washed once with PBS and then aspirated completely before freezing the plate at −80°C. Cells were then scraped in RIPA buffer containing cOmplete protease inhibitor cocktail and incubated on ice for 1 h. Lysates were then centrifuged at 21,130 g for 15 min and the supernatant was collected. Protein sample concentration was measured using a Pierce BCA protein assay kit and the lysates were mixed with 4× LDS sample buffer containing 4% β‐mercaptoethanol and boiled at 98°C for 10 min. A 20 or 30 μg of protein was loaded per well in a 4–12% Bis‐Tris SDS‐PAGE gel. After running, gels were transferred to either nitrocellulose or PVDF membranes and blocked with 3% milk in TBS‐T for 1 h before incubating with primary antibodies in 3% BSA in TBS‐T overnight, rocking at 4°C. After washing, membranes were incubated with HRP‐conjugated secondary antibodies for 1 h, rocking at room temperature. After washing, membranes were imaged with either Amersham ECL or SuperSignal West Femto ECL using a BioRad ChemiDoc Imaging System.

Mass spectrometry

Tryptic digests were analyzed using an Orbitrap Fusion Lumos Tribrid mass spectrometer interfaced to an UltiMate3000 RSLC nano HPLC system (Thermo Scientific) using data‐dependent acquisition. Initial protein identification was carried out using Proteome Discoverer (V2.4) software (Thermo Scientific). Search results from Proteome Discoverer were incorporated into Scaffold4 for relative quantification using spectral counting. All spectral counts within a sample were normalized to spectral counts of GFP. To obtain fold change and P‐value, each normalized count was added to by +1 then each subjected to averaging and division of values from Nix_1–87 divided by values from Nix_1–70, or to a student's t‐test.

Protein expression and purification

An hNIX‐C1 expression vector was engineered using the In‐Fusion^® (Takara Bio USA) recombinational cloning approach. In a nutshell, the hNIX‐C1 (corresponding to amino acid sequence 1–185 of the hNIX‐pDONR223^Ref) was seamlessly inserted (without non‐sequence‐related additional amino acids) in‐frame upstream of the intein tag of a PTYB1 destination vector. The sets of primers (5′ GAAGGAGATATACATATGTCGTCCCACCTAGTCGAG 3′, 5′ ACCCTTGGCAAAGCACTTCAGAAATTCTGCGGAGAAAATACCC 3′ and 5′ TGCTTTGCCAAGGGTACCAAT 3′, 5′ ATGTATATCTCCTTCTTAAAGTTAAACAAAATTATTTCTAGAGGGG 3′) used to synthesize the PCR products corresponding to the insert and the destination vector, respectively, were designed using the SnapGene® In‐Fusion cloning tool (GSL Biotech) and synthesized by Eurofins (Eurofins Genomics). The ligation‐independent cloning was carried out following Takara's recommendation. The entire coding region of the fusion protein was sequenced (Psomagen) with synthetic primers designed specifically for the construct, and the integrity of the full‐length plasmid was further verified by nanopore sequencing (Plasmidsaurus).

BL21‐Gold (DE3)‐competent cells (Agilent), transformed with the above plasmid, were grown overnight at 37°C in 1 l of M9 minimal media supplemented with ¹⁵NH₄Cl, in the presence of 100 μg/ml ampicillin (Sigma). After dilution with 1 l of corresponding media containing the antibiotic, the cells were grown at 37°C for 1/2 h before induction with 1 mM IPTG (EMD Millipore) for 4 h at 37°C.

The cell pellets, harvested at 3,381 g for 20 min at 4°C, were resuspended in 10 mM Tris–HCl pH 8.0, 100 mM NaCl, and 0.1 mM EDTA buffer containing one cOmplete™ Protease Inhibitor Cocktail tablet (Roche Diagnostics) and disrupted by three passages through an M‐110P homogenizer (Microfluidics™). Cell debris removal was carried out at 125,500 g for 45 min at 4°C. The supernatant was loaded onto a column containing 15 ml chitin resin (New England Biolabs, Ipswich, MA, USA) that was equilibrated with the above lysis buffer. The column was washed with 20‐column volume of the lysis buffer. The protein was cleaved from the resin by incubating it in the lysis buffer containing 50 mM DTT for 3 days at 4°C. After cleavage, the Nix‐C1 protein was collected, and buffer was exchanged into 1X PBS and 1 mM DTT. The protein sample was then loaded onto a HiLoad 26/60 Superdex 75 prep‐grade column (GE Healthcare) that was equilibrated with 1X PBS and 1 mM DTT. The eluted protein containing fractions confirmed by SDS–PAGE were pooled and concentrated (Amicon, 3000 MWCO; Millipore). The protein's final concentration was estimated using their measured A₂₈₀ and their respective calculated molar extinction coefficient (http://web.expasy.org/protparam/). The protein solution was stored at −20°C.

SDS–PAGE was used to check for protein sample. A 10‐ to 20‐fold dilution of our protein sample was mixed at 1:1 ratio with SDS–PAGE denaturing sample buffer containing 300 mg/ml Urea, 100 mg/ml SDS, 240 mg/ml DTT, and bromophenol blue in Tris pH 8 buffer. Following 3 min of sample heating, the protein was loaded onto a 10–20% Tris‐Glycine gel (Invitrogen) and run using Tris‐Glycine‐SDS buffer (BioRad). The gel was stained using AcquaStain (Bulldog Bio) overnight followed by multiple destaining steps with deionized water.

NMR spectroscopy

The ¹⁵N Nix‐C1 NMR sample was in a buffer containing 1X PBS, 1 mM DTT, and 10% D₂O. The final protein concentration was 750 μM in a 250 μl volume in a Shigemi tube (Shigemi Co. Ltd). The HSQC spectrum was acquired at 298 K on Bruker 800 MHz spectrometer equipped with cryoprobe using 16 scans and 1 s recycle delay. Spectrum was processed using NMRPipe (Delaglio et al, 1995) and plotted using PIPP (Garrett et al, 1991).

Circular dichroism spectroscopy

The Nix‐C1 sample for CD measurement was prepared by diluting the above NMR sample with 20 mM potassium phosphate buffer at pH 7 to a final concentration of 9 μM. The CD spectrum was acquired on an Applied Photophysics ChirascanTm Q100 spectropolarimeter. The CD spectrum was recorded from 260 to 190 nm using a 1 mm pathlength cell (Hellma) at 20°C. The signal was acquired in 1 nm increments with 1 s acquisition per increment. The final spectrum was an average of three accumulated scans. The data acquisition and baseline subtraction were performed using the Chirascan software.

AlphaFold2 prediction

Alphafold 2.3.1 was run using the NIH High‐Performance Computing Biowulf cluster using multimer settings and a max template date of 2023‐04‐27. Structures were displayed with Open‐Source PyMOL and distances between each residue and the other protein were calculated using a Python “for” loop within the PyMOL GUI. pLDTT and PAE values were found using the AF2Align Python script (https://github.com/CYP152N1). Values were displayed in heatmap format with custom MATLAB scripts.

Statistics

All P‐values were generated with a two‐way ANOVA using Tukey's multiple comparisons test with Prism9 software except for the mass spectrometry results that were compared with a one‐tailed student's t‐test in MATLAB_2017b. All error bars and filled shading represent standard deviation from the mean.

Author contributions

Eric Newman Bunker: Conceptualization; data curation; software; formal analysis; supervision; validation; investigation; visualization; methodology; writing – original draft; project administration; writing – review and editing. François Le Guerroué: Validation; investigation; methodology; writing – review and editing. Chunxin Wang: Conceptualization; resources; supervision; methodology; writing – review and editing. Marie‐Paule Strub: Investigation. Achim Werner: Resources; data curation; investigation; writing – review and editing. Nico Tjandra: Resources; investigation; visualization; methodology; writing – original draft; writing – review and editing. Richard J Youle: Conceptualization; supervision; funding acquisition; writing – original draft; project administration; writing – review and editing.

Disclosure and competing interests statement

The authors declare that they have no conflicts of interest.

Supporting information

Appendix

Expanded View Figures PDF

PDF+

Source Data for Figure 1

Source Data for Figure 2

Source Data for Figure 3

Source Data for Figure 4

Source Data for Figure 5

Source Data for Figure 6

Source Data for Figure 7

The EMBO Journal (2023) 42: e113491

Data availability

Uncropped microscopy images can be found at: https://www.ebi.ac.uk/biostudies/bioimages/studies/S‐BIAD769 (S‐BIAD769). Mass spectrometry data can be found at: https://www.ebi.ac.uk/pride/archive/projects/PXD043441 (PXD043441).