Direct Visualization of ATP-Binding Cassette Protein A1 Mediated Nascent High-Density Lipoprotein Biogenesis by High-Speed Atomic Force Microscopy

Abstract

The ATP-binding cassette protein A1 (ABCA1) is crucial for high-density lipoprotein (HDL) formation, maintaining cholesterol homeostasis. However, the mechanism of ABCA1-mediated HDL formation remains unclear. Here, we reconstituted this process and directly visualized it using high-speed atomic force microscopy by employing a novel linker-based immobilization strategy for side-view imaging of ABCA1 nanodiscs. Our data revealed an approximately 1.6-fold volumetric expansion of the ABCA1 extracellular domain (ECD), coupled with a contraction of the ABCA1-embedded nanodisc, suggesting ATP-dependent transfer of over 300 lipids to the ECD for transient storage. Subsequent addition of apolipoprotein A-I, a major HDL protein, allowed ABCA1 to facilitate nascent HDL particle formation, as it does at the cell membrane. Notably, this study provides the first direct observation of a single ABC protein executing its unique functional cycle with submolecular spatial and subsecond temporal resolution, offering critical insights into “HDL biogenesis”, with potential implications for ABCA1-targeted therapeutics for dyslipidemia.

High-density lipoprotein (HDL) particles help transport excess cholesterol from peripheral tissues back to the liver, a process that is known to prevent coronary artery disease. This transport was previously believed to occur through passive diffusion, driven by the concentration gradient of cholesterol between the plasma membrane and HDL particles (reviewed in ref ). However, in 1999, the genetic analysis of Tangier disease, a condition characterized by quite low levels of HDL in the blood, revealed that the ATP-binding cassette protein A1 (ABCA1) is essential for this process, suggesting that active transport is involved. −

Human ABCA1 is 2,261 amino acids long and consists of 12 transmembrane helices. It also has two cytoplasmic nucleotide binding domains (NBDs) and a large extracellular domain (ECD) , (Figure S1). ATP hydrolysis at the NBDs provides energy for the function of ABCA1, while the ECD is believed to be responsible for binding to apolipoprotein A-I (apoA-I). − Laboratory studies have found that ABCA1 is responsible for exporting cholesterol and phosphatidylcholine (PC) to lipid-free apoA-I on the cell surface. − This activity leads to the formation of nascent HDL particles, which take the shape of nanodiscs with a diameter of 6 to 12 nm. , Each nascent HDL particle consists of 200–700 lipids wrapped by two to four molecules of apoA-I. ,,

The mechanism by which ABCA1 loads several hundred lipid molecules onto lipid-free apoA-I on the cell surface has been a long-standing mystery. , We have studied ABCA1-mediated nascent HDL production using cell-based assays ,− and proposed a mechanism in which ABCA1 temporarily stores more than 500 lipid molecules transported to its large ECD before loading them onto lipid-free apoA-I. Contrary to our hypothesis, a cryo-EM structure analysis showed that the ECD of human ABCA1 forms a hydrophobic tunnel that can only accommodate less than ten lipid molecules. ,, To solve this contradiction, here we used high-speed atomic force microscopy (HS-AFM) , to directly observe the process of ABCA1-mediated nascent HDL generation. The ability of HS-AFM to directly visualize protein movements with high spatiotemporal resolution led us to expect that it would reveal the dynamic behavior of ABCA1 during HDL biogenesis, providing insights beyond those obtainable from cryo-EM and X-ray crystallography.

To observe the process of ABCA1-mediated nascent HDL generation, purified human ABCA1 was reconstituted into nanodiscs (Figure a). Since ABCA1 produces nascent HDL particles containing 200–700 lipids, it is essential to embed purified ABCA1 in nanodiscs containing a sufficient number of lipid molecules for transport. Nanodiscs with a 10 nm diameter, formed by membrane scaffold protein MSP1D1 (211 amino acids long) as used in the previous study, provide ABCA1 with only a limited number of lipids as transport substrates (Figure S2a). In contrast, more than 600 lipids can be available as substrates for ABCA1 when reconstituted in 17 nm diameter nanodiscs formed by MSP2N1 (401 amino acids long) (Figure S2b). ABCA1 was reconstituted into MSP2N1 nanodiscs containing PC, phosphatidylserine (PS), and cholesterol in a 2:2:1 ratio. PC and cholesterol serve as substrates for ABCA1, and PS is necessary for ABCA1 function.

1.

Oriented immobilization of side-view ABCA1 nanodiscs on mica via engineered linkers. (a) Gel filtration chromatogram of human ABCA1 reconstituted into nanodiscs, obtained using a Superose 6 Increase 10/300 column. Coomassie-stained SDS–PAGE analysis of the peak fraction corresponding to ABCA1 nanodiscs is shown, with molecular weight markers included for reference. Predicted models of the ABCA1 nanodisc and empty nanodisc, generated using AlphaFold , and the CHARMM-GUI Nanodisc Builder, , are displayed above the corresponding peaks. Inset: SDS-PAGE analysis of silver-stained, FLAG-affinity-purified ABCA1 prior to nanodisc reconstitution. (b) Initial AFM imaging of ABCA1 nanodiscs on untreated mica. The upper panel shows molecules predominantly observed from either the intracellular or extracellular side (cyan circles) or displaying unstable fluctuations (red circle). The lower panel shows schematic representations of ABCA1 nanodiscs from a lateral perspective. (c) Synthesis of a bifunctional Ni²⁺-Glutaraldehyde-Ni²⁺ linker. Glutaraldehyde is conjugated at both termini with Ni²⁺–NTA, forming a dumbbell-shaped molecule. (d) Oriented immobilization of nanodiscs on mica. The negatively charged mica surface electrostatically attracts one Ni²⁺ terminus of the linker. The N-terminal hexahistidine (His₆) tag of the membrane scaffold protein MSP2N1 coordinates with the exposed Ni²⁺–NTA moiety, enabling side-view immobilization of ABCA1 nanodiscs. The illustration is not to scale and is intended to highlight the immobilization mechanism. (e) AFM imaging of ABCA1 nanodiscs on mica functionalized with engineered linkers. The upper panel shows efficient side-view immobilization of ABCA1 nanodiscs (white ellipses). Quantitative analysis was restricted to molecules with sufficient height; particles with low height (dashed ellipses) were excluded. The lower panel shows a schematic representation of an ABCA1 nanodisc from a lateral perspective.

Simultaneous observation of ABCA1’s substantial cytosolic and extracellular domains is essential to elucidate its functional mechanism. However, initial HS-AFM on unmodified mica, using conventional methods, predominantly yielded top (extracellular) or bottom (intracellular) views, or revealed unstable fluctuations of ABCA1 nanodiscs, hindering side-view analysis (Figure b). To address this, we systematically explored mica surface modifications (Table S1) and developed an oriented immobilization strategy for side-view ABCA1 analysis. This strategy employs an engineered linker comprising glutaraldehyde, N-(5-amino-1-carboxypentyl)­iminodiacetic acid (AB-NTA), and nickel­(II) chloride at a 1:2:2 molar ratio (Figure c). This modification significantly increased the occurrence of side-view orientations to over 50% (Table S1). Glutaraldehyde cross-links AB-NTA, while nickel­(II) coordinates AB–NTA to both the mica surface and the hexahistidine (His₆)-tagged N-terminus of the nanodisc’s membrane scaffold protein (MSP2N1) (Figure c,d). Given the multiple MSP2N1 molecules per nanodisc, both single and double binding events to the modified mica are possible (Figure d). We focused our HS-AFM analysis on ABCA1 molecules captured in side-view orientations (Figure e). In these images, the hydrophilic domains (ECD and NBDs) of ABCA1 protruded vertically from the nanodisc in opposing directions (Figure a). The nanodisc diameter ranged from 15 to 21 nm, and the total molecular length was 22–29 nm. After correcting for the tip–sample convolution effect (2 nm), the corrected values were 13–19 nm and 20–27 nm, respectively. These values were consistent with cryo-EM data ,, and predictive models of ABCA1 within a large nanodisc (Figure a,d, Figure a, Figure S2b). This confirms the successful capture of the entire ABCA1 molecule by HS-AFM.

2.

ATP- and time-dependent enlargement of ABCA1 ECD. (a) Representative HS-AFM images of ABCA1 nanodiscs. ABCA1 nanodiscs immobilized on the mica surface were incubated with (molecules 1, 2, and 3) or without Mg-ATP (molecule 4). The incubation times before and after adding Mg-ATP are indicated. Green and orange arrowheads indicate the nanodisc and ECD, respectively. Dotted lines delineate nanodisc diameters. Some images are rotated to position the ECD on the right side. The images were cropped from the following scan areas: 100 × 100 nm² with 100 × 100 pixels (molecule 1), 60 × 54 nm² with 60 × 54 pixels (molecule 2), 80 × 80 nm² with 80 × 80 pixels (molecule 3), and 100 × 100 nm² with 80 × 80 pixels (molecule 4); frame time, 700 ms (molecule 1), 194 ms (molecule 2), 711 ms (molecule 3), and 877 ms (molecule 4). Predicted models of the ABCA1 nanodiscs, generated using AlphaFold , and the CHARMM GUI Nanodisc Builder , are shown on the left, aligned to best fit the experimental data. (b) The anti-ECD antibody reacted with the enlarged domain. Scan area, 42 × 44 nm² with 42 × 44 pixels; frame time, 1084 ms. (c, e) Time-dependent increases in the length (c) and apparent volume (e) of ABCA1 ECD (molecule 1 in (a)). Prior to ATP addition, the plot is outlined in black. (d, f) Relative changes of the length (d) and apparent volume (f) of ABCA1 ECD in 10 min (n = 6). For each molecule (n = 6), mean values were calculated over the 10 s preceding t = 0 and t = 600. Ratios (t = 600 s/t = 0 s) were log-transformed, and statistical significance was determined using a one-sample t test. In the box plots, the center line represents the median, the square (□) denotes the mean, the box edges indicate the first and third quartiles, and the whiskers represent the minimum and maximum values. The dashed gray line indicates the reference value (μ₀ = 1.0). (g) ATP- and time-dependent changes of the volume of the nanodisc. Green points, molecule 1; blue points, molecule 4. The average value during the 10 s prior to ATP addition was taken as 100%. Prior to ATP addition, the plot is outlined in black. (h) ATP-dependent changes of the volume of the nanodisc in 10 min. Blue box plots, without ATP (n = 3); green box plots, with ATP (n = 6). The statistical analysis and box plot representation are consistent with the descriptions provided in panels d and f. (i) ATP-dependent changes of the diameter of the nanodisc in 10 min. Blue bars, without ATP (n = 3); green bars, with ATP (n = 6). (c, e–i) All data points represent measured values. (d) The ratio was calculated based on the corrected values.

ABCA1 mediates nascent HDL biogenesis predominantly at the cell surface, except when cholesterol is heavily accumulated intracellularly. This ABCA1-mediated process, occurring at the cell surface, is physiologically crucial for all somatic cells to effectively maintain cholesterol concentration in their plasma membranes. In this study, we aimed to reveal the mechanism of HDL generation at the cell surface. The experimental conditions in this study (pH 7.4 and 50 mM NaCl) were also suited for HS-AFM observation.

ABCA1 nanodisc molecule exhibited no substantial morphological changes before ATP addition (see molecule 4, Figure a and Supporting video 4). Following ATP addition to the reaction chamber, a progressive enlargement and dynamic movement of one of the two hydrophilic domains became evident (see molecules 1–3, Figure a, and Supporting videos 1–3). To determine the orientation of the molecules, a monoclonal antibody against the ECD of ABCA1 was added to the chamber. It reacted with the enlarged domain of ABCA1, indicating that the domain was indeed the ECD (Figure b). We quantified changes in the ECD by measuring its length (Figure c,d) and apparent volume (Figure e,f). Both the length and the volume of the ECD of molecule 1 gradually increased until approximately 200–250 s after ATP addition (Figure c,e). At 600 s after ATP addition, the length and apparent volume of the ECD increased on average 1.9 ± 0.4-fold (±SD, n = 6) and 1.60 ± 0.15-fold (±SD, n = 6), respectively (Figure d,f). Because the volume of molecule 4 did not change in the absence of ATP, we concluded that the increase was accompanied by ATP hydrolysis (Figure a,d,f). Conversely, the apparent total volume of the nanodisc and ABCA1 NBDs (which HS-AFM could not resolve individually), gradually decreased to 57 ± 11% (±SD, n = 6) over 10 min (Figure g,h). Given that the relatively constant volume of the NBDs during the ATP hydrolysis cycle, and considering the clear contraction observed in the central region of individual ABCA1 nanodisc molecule (Supporting video 1), the observed volume decrease likely originated within the nanodisc. Indeed, the average diameter of the nanodisc gradually decreased from 16.8 ± 3.0 nm to 13.2 ± 1.8 nm (±SD, n = 6) after correcting for the tip–sample convolution effect (2 nm) (Figure i). In the absence of ATP, no significant structural changes or decrease in the volume or diameter of the nanodisc were observed (Figure g–i).

Since the diameter of the nanodisc is determined by the number of lipids, the decrease in diameter suggests that lipids are being moved from the nanodisc, a movement mediated by the function of ABCA1. After correcting for the tip–sample convolution effect (2 nm), the diameter of the nanodisc containing ABCA1 was found to be 16.8 ± 3.0 nm. This is consistent with the reported diameter (17 ± 1 nm) of nanodiscs formed with MSP2N1. , The diameter then decreased to 13.2 ± 1.8 nm after ATP addition. A 17 nm nanodisc containing ABCA1 is estimated to comprise approximately 650 lipid molecules (Figure S2b), whereas a 13 nm nanodisc is estimated to contain around 300 lipid molecules (Figure S2c). Based on this information, we estimated that over 300 lipid molecules were transferred from the nanodisc and are likely stored in the enlarged ECD of ABCA1. This conclusion is consistent with a previous biochemical study, which showed that the trypsin-sensitive sites of ABCA1 on the cell surface, most likely the ECD, can store up to 800 lipid molecules. In summary, these results suggest that ABCA1 transfers lipids from the nanodisc, where it is embedded, to its ECD through ATP hydrolysis and stores the lipids there.

The experiments shown in Figure were conducted without apoA-I, which is necessary for ABCA1-mediated HDL production. After a 10 min reaction with ATP, we added apoA-I to the chamber and incubated the reaction for another 10 min. As a result, we observed many small particles with diameters ranging from 6 to 11 nm on the mica surface after correcting for the tip–sample convolution effect (2 nm) (Figure a,d), but only a few (less than 14%) of such particles were observed in the control experiment with neither Mg-ATP nor apoA-I (Figure b,e), indicating that ABCA1 produced these particles in the presence of apoA-I. The size distribution of the particles (Figure g) showed an average diameter and thickness of 8.3 ± 1.8 nm and 3.5 ± 1.1 nm (±SD), respectively, which fits well with previous studies describing disc-shaped reconstituted HDL, , suggesting that the produced particles are nascent HDL. To our surprise, when we added EDTA after the 10 min reaction with ATP, we did not observe particles even after the addition of apoA-I (Figure c,f). These results suggest that ATP hydrolysis is necessary not only for transferring lipids to the ECD but also for loading stored lipids onto apoA-I to produce nascent HDL.

3.

ATP and apoA-I dependent nascent HDL generation by ABCA1. (a) An HS-AFM image after incubation with Mg-ATP for 10 min, followed by incubation with apoA-I for 10 min. The arrow represents an ABCA1 nanodisc, and the green arrowhead represents nascent HDL. (b) An HS-AFM image of an ABCA1 nanodisc after incubation for 10 min with neither Mg-ATP nor apoA-I. (c) An HS-AFM image after incubation with Mg-ATP for 10 min, followed by incubation with apoA-I in the presence of EDTA for 10 min. The images are displayed with a higher contrast than in Figure . Scan area, 300 × 300 nm² with 150 × 150 pixels (a), 300 × 300 nm² with 130 × 130 pixels (b, c); frame time, 2015 ms (b), 1607 ms (a), 2903 ms (c). (d, e, f) The distributions of the diameters and quantities of generated nascent HDL under the conditions (a), (b), and (c), respectively. (g) The thicknesses and diameters of the generated nascent HDL under condition (a). (d–g) All data points represent measured values.

Subsequently, we analyzed how ABCA1 loaded the stored lipids onto apoA-I using HS-AFM. When apoA-I was added to the chamber, it diffused as small structures (14 ± 7.9 nm (±SD) in length, 1.2 ± 0.4 nm (±SD) in height, n = 40) in the buffer (Supporting video 5). Because of the time resolution (0.5–1 image per second), it was hard to observe the direct interaction of apoA-I with the enlarged ECD and nascent HDL generation. However, we succeeded in visualizing the dynamic restructure of the enlarged ECD of ABCA1 just before nascent HDL generation (Figure , Supporting video 6). A new structure (bright spot, shown by arrows) was created in the ECD and grew larger. Finally, a small particle (green arrowhead in the bottom right panel in Figure ) of diameter 7.5 nm and thickness 4.5 nm was released. As a result, the apparent volume of the enlarged ECD was reduced by approximately 30% (Figure S3). These results suggest that a significant amount of lipids stored in the ECD were transferred to apoA-I interacting with the ECD within 15 s.

4.

Loading of lipids stored in the ECD onto apoA-I. Visualization of the nascent HDL generation. The incubation times (seconds) before nascent HDL generation are indicated in the images. Arrows represent new structures appearing in the ECD. The green arrowhead in the bottom right panel represents a released nascent HDL. The images were cropped from the following scan area: 100 × 100 nm² with 100 × 100 pixels; frame time, 812 ms.

The generation of HDL is a key process for cholesterol homeostasis in the human body. However, it remained a mystery how HDL particles are generated by ABCA1 on the cell surface. While cryo-EM structure analyses have been performed, ,, they have not explained how ABCA1 produces a nascent HDL particle consisting of 200–700 lipids in the absence of strong interaction with apoA-I. In this study, we directly observed the process of ABCA1-mediated nascent HDL generation using HS-AFM and uncovered a previously unseen mechanism.

A transporter moves one or two substrates (excluding ions) in a single transport cycle. ABCA1 is an ATP-dependent transporter and hydrolyses 1 to 2 ATPs per second. , As a result, it would take 1.5 to 6 min for ABCA1 to transport 200–700 lipid molecules, which make up one nascent HDL particle. Therefore, if this loading process is coupled with ATP-dependent transport, apoA-I would need to remain bound to ABCA1 for 1.5 to 6 min; despite the interaction between ABCA1 and apoA-I being neither strong nor specific. ,, To reconcile this contradiction, a plausible model suggests that ABCA1 transiently stores hundreds of lipids within its ECD, thus serving as a platform to load the entire lipid pool onto a weakly bound apoA-I at once.

HS-AFM single-molecule observations suggest a multistep process for ABCA1-mediated nascent HDL generation, as illustrated in Figure . The first step involves ABCA1 transporting lipids from the membrane outward and relies on ATP hydrolysis. In the second step, the transported lipids are temporarily stored in the ECD of ABCA1. This storage causes the enlargement of ECD, as shown in Figure a. Moreover, we observed the dynamic movement of enlarged ECD (Supporting video 1). Because ATP hydrolysis is required for lipid loading onto apoA-I, this movement may be required for apoA-I binding to the enlarged ECD and lipid loading in the subsequent step. In the final step, lipids are loaded onto apoA-I by the dynamic restructuring of the enlarged ECD. It is conceivable that the discoidal nanodiscs with diameters of 6 to 11 nm shown in Figure a,d,g are created through this process in the presence of apoA-I, rather than by direct coupling with ATP-dependent lipid transport. This restructuring may be driven by the ATP-induced rotation of the ECD.

5.

Proposed model for ABCA1-mediated nascent HDL biogenesis. Schematic depicting the proposed dynamic mechanism of ABCA1-mediated nascent HDL biogenesis. The multistep mechanism is as follows: Step I, ATP-dependent outward transport of phospholipids and cholesterol from the membrane. Step II, transient lipid storage within the ECD. Step III, dynamic movement of the enlarged ECD. Step IV, apoA-I interaction with the enlarged ECD, triggering nascent HDL generation through the massive restructuring of the lipid-stored ECD. Predicted amphipathic helices within the ECD are indicated in magenta.

While previous cryo-EM studies of ABCA1 ,, revealed a hydrophobic tunnel within its ECD suggestive of lipid binding, the mechanism by which it accommodates hundreds of lipid molecules remained entirely unclear. Our findings unequivocally demonstrate that ABCA1 accumulates lipids in its ECD in an ATP-dependent manner. We propose the following mechanism: The ABCA1 ECD is predicted to contain numerous dispersed amphipathic helices (Figure , Figure S1). Cryo-EM structures ,, indicate that the upper half of the ECD is flexible. We hypothesize that this region undergoes ATP-dependent unfolding during lipid transport, allowing the embedded amphipathic helices to surround and sequester hundreds of lipid molecules. This ATP-dependent nascent HDL biogenesis is quite distinct from very-low-density lipoprotein (VLDL) formation.

As described above, nascent HDL generation is a complicated pathway that includes ABCA1-mediated lipid transport, lipid storage, and lipid loading onto apoA-I. If all three processes occur simultaneously, analysis would be difficult. Therefore, we separated the lipid loading step from the lipid transport and storage steps in this study. In this way, we successfully showed that massive conformational changes occur in the ABCA1 ECD by storing several hundred lipids, thereby exposing the apoA-I binding site(s) on the surface of the ECD, as previously shown by biochemical experiments. , We also showed that several hundred lipid molecules are loaded onto apoA-I within a considerably short period to generate nascent HDL.

The enlargement of the ECD and the contraction of the nanodisc appear to proceed via discrete steps rather than a continuous, gradual process (Figure c,e,g). In the absence of ATP, the ECD is predicted to adopt a stable conformation, stabilized by interactions among the amphipathic helices present within the ECD, as depicted in Figure . Given the constrained capacity of the ECD for accommodating a large number of lipid molecules, extensive structural reconfigurations, rather than continuous and gradual conformational alterations, are anticipated to occur to incorporate several hundred lipids. The observed stepwise contraction of the nanodisc may thus reflect these substantial ECD reconfigurations. Furthermore, it is likely that the nanodisc itself undergoes stepwise reorganization in response to the reduction of lipid content. The cooperativity between the ECD enlargement and nanodisc contraction, and the specific thresholds governing these stepwise conformational changes, remain to be elucidated.

In previous HS-AFM studies of membrane proteins, observations have mainly been limited to a top-view orientation, , although some side-view imaging of membrane protein-embedded nanodiscs has been reported. , In this study, we established a methodology for efficient side-view imaging of human ABCA1 in near-native functional conformations. This was accomplished by employing engineered linkers for the oriented immobilization of side-view ABCA1 nanodiscs on mica (Figure c,d,e). The approach holds strong potential for broad application across a wide range of membrane protein systems. Using this technique, we successfully visualized the complex process of HDL generation by ABCA1, revealing a unique mechanism distinct from those observed in other proteins. This advancement significantly expands the utility of HS-AFM for mechanistic study of membrane protein dynamics.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.nanolett.5c03116.

• Materials and Methods; Figure S1, The 3D structure of human ABCA1 generated by Alphafold2; Figure S2, Human ABCA1 reconstitution within nanodiscs of different sizes; Figure S3, Time-resolved changes in ECD volume and nascent HDL appearance; Table S1, Modification of mica surface to obtain side-view ABCA1 nanodiscs; Detailed descriptions of supporting videos (PDF)

• Video 1: The conformational change of molecule 1 in Figure a captured by HS-AFM (MOV)

• Video 2: The conformational change of molecule 2 in Figure a captured by HS-AFM (MOV)

• Video 3: The conformational change of molecule 3 in Figure a captured by HS-AFM (MOV)

• Video 4: The conformational change of molecule 4 in Figure a captured by HS-AFM (MOV)

• Video 5: HS-AFM images of apoA-I added to the chamber at a concentration of 1 nM (MP4)

• Video 6: The conformational change of the molecule in Figure captured by HS-AFM MP4)

∇.

AK and RA contributed equally. The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. KUe, AK, and NK designed the project. AK purified the proteins and reconstituted them in nanodiscs. RA and AK performed the HS-AFM and collected the data. AK and RA analyzed the data. NK supervised the HS-AFM experiments. KUm assisted with the HS-AFM experiments. FO and YK helped with the data analysis and contributed discussions on the results. AK and RA prepared the figures and videos. AK, NK, and KUe wrote the manuscript with input from all authors.

This work was supported by JSPS KAKENHI Grant Numbers 18H05269 (KUe, NK, YK), 23H00326 (KUe), 20H00327 (NK), 20H02912 (AK), 22KF0153 (RA), 22H02258 (YK), and 20K15449 (FO) and CREST-JST Grant number JPMJCR1762 (NK).

The authors declare no competing financial interest.

§.

JSPS International Research Fellow.