Transmembrane phospholipid translocation mediated by Atg9 is involved in autophagosome formation

Macroautophagy (hereafter referred to as autophagy) is a conserved catabolic pathway in eukaryotes. In the budding yeast Saccharomyces cerevisiae, autophagy initiates at the preautophagosomal structure by formation of the isolation membrane (phagophore), followed by its maturation to the double-membrane autophagosome, which eventually fuses with the vacuole (yeast lysosome), leading to degradation of engulfed cargoes (Bento et al., 2016; Mizushima et al., 2011; Nakatogawa, 2020). The proteins that are necessary for the autophagic process have been analyzed extensively, and the molecular mechanisms underlying the various steps are beginning to emerge. Nevertheless, there are still many unanswered questions, one of which is how the phospholipids for isolation membrane/autophagosome formation are supplied.

A recent study has shown that de novo phospholipid synthesis driven by acyl-CoA synthetase, Faa1, is critical for the expansion of isolation membranes (Schütter et al., 2020). Phospholipids synthesized through this process are thought to be supplied from the ER exit site (ERES) to isolation membranes via nonvesicular transport mediated by Atg2 (Graef et al., 2013; Kishi-Itakura et al., 2014; Suzuki et al., 2013; Velikkakath et al., 2012). Atg2 and its mammalian homologue, ATG2, have a hydrophobic cavity that can accommodate many phospholipid molecules, tether two liposomes, and mediate phospholipid transport between liposomes (Chowdhury et al., 2018; Kotani et al., 2018; Maeda et al., 2019; Osawa et al., 2019; Valverde et al., 2019). Thus, it has been thought that Atg2 binds to the ERES on one end and the isolation membrane on the other end, thereby supplying phospholipids for the isolation membrane formation (Osawa et al., 2019; Valverde et al., 2019).

The Atg2-dependent process, however, can only transport phospholipids to the cytoplasmic leaflet of isolation membranes. For isolation membranes to expand, phospholipids need to be provided to the noncytoplasmic leaflet as well. In this context, we recently showed that yeast Atg9, the only transmembrane protein in the core Atg proteins, translocates phospholipids across liposome membranes, suggesting that Atg9 may function as a phospholipid scramblase in the isolation membrane (Matoba et al., 2020). Distribution of phosphatidylinositol 3-phosphate (PtdIns(3)P) in both leaflets of yeast autophagosomal membranes (Cheng et al., 2014) is consistent with this putative Atg9 functionality, but PtdIns(3)P is a minor phospholipid and the transbilayer movement of this lipid alone would not be sufficient for isolation membrane expansion.

In the present study, we explored whether phospholipids other than PtdIns(3)P are also distributed in both leaflets of the isolation membrane/autophagosome. Specifically, we examined two major phospholipids, phosphatidylcholine (PtdCho) and phosphatidylserine (PtdSer), as well as phosphatidylinositol 4-phosphate (PtdIns(4)P), by quick-freeze and freeze-fracture replica labeling EM (QF-FRL; Fig. S1 A), which revealed lipid distribution in individual membrane leaflets (Tsuji et al., 2019). We also examined how de novo–synthesized PtdCho is incorporated into autophagic membranes by combining QF-FRL and metabolic labeling with a clickable choline analogue (Iyoshi et al., 2014). We found that autophagosomes and autophagic bodies are the only membranes in which all these phospholipids show symmetrical distribution across the lipid bilayer. Moreover, de novo–synthesized PtdCho was found to be distributed in both leaflets of autophagosomal membranes within 30 min after synthesis without dispersing to the ER and other cellular membranes. Incorporation of de novo–synthesized PtdCho into the noncytoplasmic leaflet of autophagosomal membranes was compromised in yeast expressing an Atg9 mutant that was decreased in the phospholipid translocation activity in liposomes (Matoba et al., 2020). Our results indicate that highly efficient transbilayer phospholipid transport that is mediated by Atg9 occurs in isolated membranes/autophagosomes.

PtdCho is a major phospholipid in the yeast autophagosomal membrane (Schütter et al., 2020). We previously clarified the distribution patterns of PtdCho in the budding yeast S. cerevisiae by metabolic labeling with the clickable choline analogue propargylcholine and click reaction in freeze-fracture replicas (Fig. S1 B; Iyoshi et al., 2014). We found that PtdCho is distributed symmetrically across the phospholipid bilayer of all organelles, but not in the plasma membrane (Iyoshi et al., 2014). In the present study, using the same methodology, we analyzed PtdCho in autophagic membranes. Yeast was cultured overnight in synthetic complete (SC) medium containing 1 mM propargylcholine to label PtdCho in all cellular membranes, and then transferred to nitrogen- and carbon-deficient synthetic medium (S(–NC)) supplemented with 1 mM propargylcholine for 4 h. (In this and subsequent experiments, vacuolar proteolysis was blocked by adding 1 mM phenylmethylsulphonyl fluoride or by using pep4Δ cells to observe autophagic bodies.) This technique revealed intense PtdCho labeling in comparable densities in the cytoplasmic and noncytoplasmic leaflets, in both the outer and inner autophagosomal membranes (Fig. 1, A and B). Autophagic bodies that were derived from the inner autophagosomal membrane were also labeled similarly in the two leaflets (Fig. 1 C). The specificity of the labeling was confirmed by omitting propargylcholine from the culture medium or by using the click reaction mixture without biotin-azide (Iyoshi et al., 2014).

For PtdSer, endogenous lipids were labeled by using the evectin-2 pleckstrin-homology (PH) domain (Tsuji et al., 2019). In yeast in a logarithmic growth phase, PtdSer is distributed in both leaflets of most organelles, whereas it is confined to the cytoplasmic leaflet in the vacuole and the plasma membrane (Tsuji et al., 2019). When autophagy was induced by culturing in S(–NC) for 4 h, PtdSer was intensely labeled in both the cytoplasmic and noncytoplasmic leaflets of autophagosomes (Fig. 2 A). Autophagic bodies were also labeled in both leaflets (Fig. 2 B). Furthermore, after autophagic induction, the lumenal leaflet of vacuoles showed a significant increase in PtdSer (Fig. 2 C), which is thought to be caused by the fusion of outer autophagosomal membranes to vacuoles. The labeling was negligible when a PH domain mutant lacking the affinity to PtdSer was used, confirming its specificity (Tsuji et al., 2019). The overall labeling density of PtdSer is comparable to that of PtdCho, but this does not indicate that the actual concentrations of PtdSer and PtdCho are similar, because the labeling efficiency is different for each phospholipid.

The results showed that PtdCho and PtdSer are present in both leaflets of autophagosomes. This distribution could be generated if isolation membranes/autophagosomes were to expand by the fusion of vesicles derived from other organelles with symmetrical PtdCho/PtdSer distributions. Atg9 vesicles, which contain both PtdCho and PtdSer (Sawa-Makarska et al., 2020), are involved in isolation membrane/autophagosome formation, but the amount of phospholipid that is imported by these vesicles is thought to be much smaller than that transported by the Atg2-dependent nonvesicular mechanism (Chowdhury et al., 2018; Kotani et al., 2018; Maeda et al., 2019; Osawa et al., 2019; Valverde et al., 2019; Yamamoto et al., 2012). Thus, the symmetrical distribution of PtdCho and PtdSer is more likely to be caused by the transbilayer movement of phospholipids in isolation membranes/autophagosomes.

PtdIns(4)P is known to play important roles in autophagy (Palamiuc et al., 2020). To label PtdIns(4)P in QF-FRL, we adopted the PtdIns(4)P-binding domain of SidM/DrrA (P4M) of Legionella pneumophila, which has a high affinity for this lipid (Brombacher et al., 2009; Schoebel et al., 2010) and has been used successfully as a fluorescence biosensor in live cells (Hammond et al., 2014). We first confirmed that GST-tagged P4M (GST-P4M) specifically labels PtdIns(4)P in liposome freeze-fracture replicas (Fig. S2 A). With this probe, intense labeling for PtdIns(4)P was found in the cytoplasmic leaflet of the Golgi, the vacuole, and the plasma membrane in yeast in growth condition (Fig. S2, B and D). GST-P4M^K568A, a P4M mutant that lacks the PtdIns(4)P binding affinity due to one amino acid substitution (Del Campo et al., 2014), did not bind to any membrane significantly, verifying the specificity of this labeling method (Fig. S2, C and D).

In yeast cultured in S(–NC) for 4 h, PtdIns(4)P was labeled in autophagosomes and autophagic bodies in both the cytoplasmic and the noncytoplasmic leaflets (Fig. 3, A and B). Vacuoles in autophagic cells showed a significant increase in PtdIns(4)P labeling in the lumenal leaflet compared with the amount observed in growth condition (Fig. 3 C), as observed for PtdSer. The result in autophagosomes is similar to that in a recent report that used an anti-PtdIns(4)P antibody as a probe, but the PtdIns(4)P antibody method yielded only a low level of labeling in the autophagic bodies and in the cytoplasmic leaflet of vacuoles (Kurokawa et al., 2019; Tomioku et al., 2018). The reason for these discrepancies is not clear, but the PtdIns(4)P distribution detected by GST-P4M is consistent with the fact that autophagic bodies originate from inner autophagosomal membranes and with the presence of PtdIns 4-kinase Lsb6 on the vacuolar membrane (Han et al., 2002).

The symmetrical distribution of PtdIns(4)P in autophagosomes and autophagic bodies is conspicuous, because this lipid is generated by kinases and phosphatases in the cytosol (Strahl and Thorner, 2007), and no other membrane shows significant labeling in the noncytoplasmic leaflet (Fig. S2 D). This distribution is similar to that of PtdIns(3)P, which is localized in the noncytoplasmic leaflet of autophagic membranes, despite its synthesis in the cytoplasmic leaflet (Cheng et al., 2014). If only one lipid were to show such distribution, it might appear that some special mechanism was operating only on that lipid. The presence of both PtdIns(3)P and PtdIns(4)P in the noncytoplasmic leaflet, however, suggests a general mechanism that enables transbilayer phospholipid movement in isolation membranes/autophagosomes.

To further test the idea of phospholipid translocation in isolation membranes/autophagosomes, we examined whether PtdCho attains symmetrical distribution immediately after synthesis. For this purpose, PtdCho was labeled metabolically for only 30 min. First, to examine how de novo–synthesized PtdCho is incorporated into yeast in growth condition, propargylcholine was added to the SC medium only for the last 30 min before freezing. The PtdCho label was distributed most densely in the outer nuclear membrane, the cortical ER, and the outer mitochondrial membrane, and much less densely in the inner mitochondrial membrane, the vacuole, and the plasma membrane (Figs. 4 Aand S3 A). This result was consistent with the putative distribution of Cpt1 and Ept1, enzymes of the Kennedy pathway synthesizing PtdCho, in the ER and mitochondria (Henry et al., 2012).

Distribution of de novo–synthesized PtdCho in autophagy was similarly examined. After autophagy was induced for 3.5 h in S(–NC), propargylcholine was added; 30 min later, yeast was frozen. By this means, autophagosomal membranes were significantly labeled, whereas autophagic bodies and other organelles, including the nuclear membrane, the cortical ER, and the outer mitochondrial membrane, were labeled only scarcely (Figs. 4 Band S3 B). The difference between autophagosomes and other organelles was evident even when the labeling density was compared within the same cell, refuting the possibility of cell-to-cell variation (Fig. 4 C). The contrasting distribution of de novo–synthesized PtdCho in growth condition and after autophagy induction was confirmed by fluorescence microscopy. In yeast in SC medium, intense click labeling was seen in the ER and mitochondria; after incubation in S(–NC), in contrast, the labeling was observed only as a weak dot overlapping with GFP-Atg8 (Fig. S3 C).

Remarkably, de novo–synthesized PtdCho was distributed in both leaflets of autophagosomes at comparable densities (Fig. 4 D). Phospholipid scrambling is thought to occur constitutively in the ER (Pomorski and Menon, 2016), but the negligible labeling in the ER makes it unlikely that the symmetrical PtdCho distribution in autophagosomes is generated by the putative ER mechanism. Rather, our result more likely indicates phospholipid translocation in isolation membranes/autophagosomes, which is probably mediated by Atg9 (Matoba et al., 2020). Consistently, in atg9Δ cells that are reconstituted with an Atg9 mutant, Atg9^Basic, incorporation of de novo–synthesized PtdCho into the noncytoplasmic leaflet of the autophagosomal membranes was decreased (Fig. 4 E). The decrease was relatively small, but this is reasonable considering that the difference of the in vitro phospholipid translocation activity between WT Atg9 and Atg9^Basic is ∼18% (Matoba et al., 2020).

The above result also indicates that PtdCho synthesized by the Kennedy pathway during autophagy is preferentially incorporated into isolation membranes. This is in line with the recent report showing that phospholipid synthesis for isolation membrane expansion is driven by Faa1 that exists in the vicinity (Schütter et al., 2020). Phosphatidylinositol synthase was also shown to be enriched in an ER subdomain near the autophagosome formation site (Nishimura et al., 2017). These results corroborate the idea that a dedicated molecular machinery for phospholipid synthesis and transport is involved in isolation membrane formation.

Based on the available data, including those on the molecular functions of Atg2 and Atg9 in vitro (Kotani et al., 2018; Matoba et al., 2020; Osawa et al., 2019), we propose that autophagy in yeast involves the following sequence of events: Phospholipids are synthesized in an ER domain near the ERES, transported to nascent isolation membranes via Atg2 at the membrane contact site, and distributed into both leaflets of the membrane by Atg9, thereby inducing proportionate membrane growth (Fig. 4 F). Considering that mammalian ATG2A and ATG9A share functionality with their respective yeast counterparts (Chowdhury et al., 2018; Maeda et al., 2019; Maeda et al., 2020; Matoba et al., 2020; Valverde et al., 2019), autophagy in mammalian cells is likely to involve similar processes. In this respect, it is peculiar that PtdIns(3)P in autophagosomal membranes is significantly labeled only in the cytoplasmic leaflet in mammalian cells, whereas it is labeled in both leaflets in yeast (Cheng et al., 2014). We speculate that this apparent difference may have several causes, which are discussed below. Phospholipid scrambling by Atg9 (ATG9A) is likely to occur near Atg2 (ATG2A) at the ER isolation membrane contact site (Gómez-Sanchez et al., 2018; Tang et al., 2019; Guardia et al., 2020), suggesting that phospholipids that are transported from the ER are preferentially translocated to the noncytoplasmic leaflet. Here, mammalian autophagosomes are larger than yeast autophagosomes, and they require more phospholipid influx. In this setting, PtdIns(3)P, which is present in a relatively sparse manner in mammalian autophagosomes (Cheng et al., 2014), may not undergo efficient scrambling by ATG9A. Additionally, the distribution density of mammalian ATG9A in autophagosomal membranes is likely to be smaller than that of yeast Atg9 (Koyama-Honda et al., 2013; Orsi et al., 2012; Yamamoto et al., 2012), and PtdIns(3)P-binding proteins that are likely to restrict PtdIns(3)P scrambling might exist more abundantly in mammalian cells than in yeast (Dall’Armi et al., 2013; Lystad and Simonsen, 2016). A combination of these and other quantitative differences probably gives rise to the distinct PtdIns(3)P distribution patterns in yeast and mammalian autophagosomal membranes.

Despite the emergence of a plausible model for the isolation membrane formation, the in vivo functionalities of Atg2 and Atg9 are yet to be defined in detail, as are the means by which they are regulated. Moreover, it is not clear how phospholipids synthesized in a limited ER domain can be funneled to isolation membrane formation rather than diffusing into the general ER membrane. These questions need to be addressed in further studies before we can fully understand the mechanisms of isolation membrane formation.

For PtdIns(4)P labeling, the P4M domain of SidM/DrrA (Brombacher et al., 2009) was cloned from L. pneumophila cDNA (provided by Dr. Koji Kimura, Nagoya University Graduate School of Medicine, Nagoya, Japan) and inserted to a pGEX-6P vector (GE Healthcare). Recombinant GST-P4M and its mutant GST- P4M^K568A were produced in E. coli of the BL21 (DE3) strain and affinity purified by using glutathione-agarose (Sigma-Aldrich). For PtdSer labeling, vectors carrying the PH domain of human evectin-2 in tandem (Uchida et al., 2011) were kindly provided by Dr. Tomohiko Taguchi (Tohoku University, Sendai, Japan), and the recombinant GST-evectin-2-2xPH domain was produced after subcloning to pGEX-6P (Tsuji et al., 2019). Rabbit anti-GST (A190-122A; Bethyl) and mouse anti-biotin (200–002-211; Jackson ImmunoResearch) antibodies were obtained from their respective suppliers. Colloidal gold (10 nm)–conjugated Protein A (PAG10; The University Medical Center Utrecht) and goat anti-mouse IgG antibody (GAM12; Jackson ImmunoResearch) were also purchased.

1-Palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (MC-6081; Nichiyu), L-α-phosphatidylinositol (bovine liver; 840042P; Avanti), PtdIns(3)P (910; CellSignals), PtdIns(4)P (912; CellSignals), phosphatidylinositol 5-phosphate (914; CellSignals), phosphatidylinositol 3,4-bisphosphate (904; CellSignals), phosphatidylinositol 3,5-bisphosphate (906; CellSignals), phosphatidylinositol 4,5-bisphosphate (902; CellSignals), and phosphatidylinositol 3,4,5-trisphosphate (908; CellSignals) were dissolved in chloroform/methanol (2:1) at defined molar ratios, dried with dry nitrogen gas in a hot bath, and desiccated in a vacuum overnight. Dried samples were vortexed with 20 mM Hepes buffer containing 0.1% rhodamine-dioleoyl phosphatidylethanolamine (810150; Avanti) at 37°C, and the solution was pushed through a mini-extruder (Avanti) with a polycarbonate membrane filter (pore size, 0.2 µm) 10 times to create unilamellar vesicles. These were centrifuged for 30 min at 50,000 g in a Beckman Optima TLX ultracentrifuge, and the pellet was quick frozen.

The S. cerevisiae strains used in this study are derived from SEY6210 and W303-1A and are listed in Table S1. Deletion mutants were generated according to a PCR-based method using pFA6a plasmids as templates (Kaiser et al., 1994). For comparison of WT Atg9 and Atg9^Basic, atg9Δ cells expressing pRS316-ScAtg9^WT-EGFP (pRS316-ScAtg9^WT) and pRS316-ScAtg9^Basic-EGFP (p9KM129) were used (Matoba et al., 2020).

Yeast cells were grown to the exponential phase in SC medium (0.17% yeast nitrogen base without amino acids and ammonium sulfate [233520; Becton Dickinson], 0.5% ammonium sulfate, 2% glucose, and dropout mix) at 30°C. For metabolic labeling of PtdCho, the medium was supplemented with 1 mM propargylcholine, which was synthesized as described (Jao et al., 2009). To induce autophagy, cells were washed with distilled water and incubated for 4 h at 30°C in S(–NC). Vacuolar proteolysis was blocked by adding 1 mM phenylmethylsulphonyl fluoride (10837091001; Sigma-Aldrich) or by using pep4Δ cells to observe autophagic bodies.

A copper EM grid (100 mesh) immersed with yeast or liposome pellets was sandwiched between a 20-µm-thick copper foil and a flat aluminum disc (242; Engineering Office M. Wohlwend) and frozen using an HPM 010 high-pressure freezing machine (Leica). For freeze fracture, the sandwiched sample was transferred to the stage of a Balzers BAF 400 and fractured at −115°C to −105°C under a vacuum of ∼10⁻⁶ mbar. Replicas were made by electron-beam evaporation in three steps: carbon (C; 5 nm in thickness) at an angle of 90° to the fractured surface, platinum-C (2 nm) at an angle of 45°, and C (10 nm) at an angle of 90°. The thickness of evaporation was adjusted by referring to a quartz crystal thickness monitor.

Replicas were treated sequentially with 2.5% SDS in 0.1 M Tris-HCl (pH 8.0) overnight at 60°C, with 0.1 mg/ml Zymolyase 100T (07665–55; Nacalai) in PBS containing 0.1% Triton X-100, 1% BSA (01281–26; Nacalai), and a protease inhibitor cocktail (25955–11; Nacalai) for 2 h at 37°C, and again with 2.5% SDS in PBS overnight at 60°C. In some experiments, the Zymolyase step was replaced with digestion with 0.5% Westase (9005; Takara Bio) in McIlvain citrate phosphate buffer (pH 6.0) containing 10 mM EDTA and 30% FCS (Tsuji et al., 2019). The replicas were stored in buffered 50% glycerol in PBS at −20°C until use.

PtdCho in all cellular membranes was labeled metabolically by culturing yeast overnight in growth medium supplemented with 1 mM propargylcholine. For labeling de novo–synthesized PtdCho, yeast cultured in medium without propargylcholine was treated with 1 mM propargylcholine only for 30 min before quick freezing. Click labeling of PtdCho was performed as described (Iyoshi et al., 2014) with a slight modification. For EM, freeze-fracture replicas were treated with 2% cold fish gelatin (G7041; Sigma-Aldrich) and 0.001% Triton X-100 in 0.1 M Tris-HCl (pH 7.5) for 10 min and incubated in 0.1 M Tris-HCl (pH 7.5) containing 1 mM copper sulfate, 2.5 mM ascorbic acid, 100 µM Tris[(1-benzyl-1H-1,2,3-triazol-4-yl)methyl]amine (AS-63360; AnaSpec), 2% cold fish gelatin, and 5 µM biotin-azide (BCFA-003-1; baseclick) for 30 min at RT. After rinsing, the replicas were blocked with 2% cold fish gelatin in PBS for 30 min and incubated with mouse anti-biotin antibody (8.7 µg/ml) in PBS containing 1% cold fish gelatin overnight at 4°C, followed by GAM12 for 30 min at 37°C in PBS containing 1% BSA. For fluorescence microscopy, yeast cells were fixed in 4% formaldehyde in 0.1 M sodium phosphate buffer for 15 min, permeabilized with 0.1% Triton X-100 in PBS for 10 min, and subjected to the click reaction for 20 min at RT. The reaction solution contained 0.1 mM copper sulfate, 0.1 M ascorbic acid, and 500 nM AFDye546 picolyl-azide (1283-1; Click Chemistry Tools) in 0.1 M Tris-HCl (pH 7.5). After rinsing, cells were observed under an Axiovert 200 M microscope (Zeiss) equipped with Apotome2 using an Apochromat 63× objective lens. The color, brightness, and contrast of the presented images were adjusted using Adobe Photoshop CC.

PtdSer was labeled as described (Tsuji et al., 2019). Briefly, freeze-fracture replicas were blocked with a mixture of 3% BSA and 2% cold fish gelatin in 0.1 M sodium phosphate buffer (pH 6.0) and incubated at 4°C overnight with GST-evectin-2 2xPH in 0.1 M sodium phosphate buffer containing 1% BSA and 1% cold fish gelatin. Samples were then successively incubated in PBS containing 1% BSA at 37°C for 40 min with a rabbit anti-GST antibody, then for another 40 min with colloidal gold (10 nm)–conjugated protein A.

For PtdIns(4)P labeling, replicas were washed with PBS containing 0.1% Triton X-100, blocked with 3% BSA and 2% cold fish gelatin in PBS, and incubated overnight at 4°C with GST-P4M or GST-P4M^K568A (80 ng/ml) in PBS containing 1% BSA and 1% cold fish gelatin. Samples were subsequently treated with rabbit anti-GST antibody (10 µg/ml) followed by colloidal gold (10 nm)–conjugated protein A (1:50 dilution of the supplied solution), both for 30 min at 37°C in 1% BSA in PBS.

The labeled replicas were picked up on Formvar-coated EM grids after a brief rinse in distilled water and observed under a JEOL JEM-1011 electron microscope. Digital images were taken using a charge-coupled device camera (Gatan). The number of colloidal gold particles was counted manually, and areas were measured using ImageJ (National Institutes of Health). The labeling density in the selected structure was calculated by dividing the number of colloidal gold particles by the area. For each structure, more than 15 different micrographs that were taken randomly were used. Each experiment was repeated more than three times to confirm reproducibility. Statistical differences between samples were tested by two-tailed Mann–Whitney U test using Prism 8 (GraphPad Software). The labeling intensity in the autophagosome of each cell was compared with that in the outer nuclear membrane of the same cell by the Wilcoxon signed-rank test using the statistical language R (Fig. 4 C).

Fig. S1 shows the outline of freeze-fracture replica labeling. Fig. S2 shows the labeling of PtdIns(4)P. Fig. S3 shows the labeling of de novo–synthesized PtdCho. Table S1 lists the yeast strains used in this study.

We thank Dr. Koji Kimura (Nagoya University) for L. pneumophila cDNA; Dr. Tomohiko Taguchi (Tohoku University) for plasmids carrying the evectin-2 PH domain in tandem; Drs. Hitoshi Nakatogawa and Tetsuya Kotani (Tokyo Institute of Technology, Tokyo, Japan) for W303-1A strains; Ms. Tsuyako Tatematsu, Dr. Jinglei Cheng (Nagoya University), and Ms. Hiroko Osakada (Juntendo University) for excellent technical assistance; and Drs. Hitoshi Nakatogawa and Nobuo N. Noda (Institute of Microbial Chemistry) for critical reading of the manuscript.

This study was supported by Grants-in-Aid for Scientific Research from the Japan Society of the Promotion of Science (18K15004 to M. Orii, 19K07265 and 20H05339 to T. Tsuji, and 15H05902 and 18H04023 to T. Fujimoto).

The authors declare no competing financial interests.

Author contributions: Conceptualization: T. Fujimoto; investigation: M. Orii, T. Tsuji, and Y. Ogasawara; writing, original draft: M. Orii and T. Fujimoto; writing, review and editing: T. Fujimoto; funding acquisition: M. Orii, T. Tsuji, and T. Fujimoto.