Bacteroidota polysaccharide utilization system for branched dextran exopolysaccharides from lactic acid bacteria

Dextran is an α-(1→6)-glucan that is synthesized by some lactic acid bacteria, and branched dextran with α-(1→2)-, α-(1→3)-, and α-(1→4)-linkages are often produced. Although many dextranases are known to act on the α-(1→6)-linkage of dextran, few studies have functionally analyzed the proteins involved in degrading branched dextran. The mechanism by which bacteria utilize branched dextran is unknown. Earlier, we identified dextranase (FjDex31A) and kojibiose hydrolase (FjGH65A) in the dextran utilization locus (FjDexUL) of a soil Bacteroidota Flavobacterium johnsoniae and hypothesized that FjDexUL is involved in the degradation of α-(1→2)-branched dextran. In this study, we demonstrate that FjDexUL proteins recognize and degrade α-(1→2)- and α-(1→3)-branched dextrans produced by Leuconostoc citreum S-32 (S-32 α-glucan). The FjDexUL genes were significantly upregulated when S-32 α-glucan was the carbon source compared with α-glucooligosaccharides and α-glucans, such as linear dextran and branched α-glucan from L. citreum S-64. FjDexUL glycoside hydrolases synergistically degraded S-32 α-glucan. The crystal structure of FjGH66 shows that some sugar-binding subsites can accommodate α-(1→2)- and α-(1→3)-branches. The structure of FjGH65A in complex with isomaltose supports that FjGH65A acts on α-(1→2)-glucosyl isomaltooligosaccharides. Furthermore, two cell surface sugar-binding proteins (FjDusD and FjDusE) were characterized, and FjDusD showed an affinity for isomaltooligosaccharides and FjDusE for dextran, including linear and branched dextrans. Collectively, FjDexUL proteins are suggested to be involved in the degradation of α-(1→2)- and α-(1→3)-branched dextrans. Our results will be helpful in understanding the bacterial nutrient requirements and symbiotic relationships between bacteria at the molecular level.

Bacteria produce various enzymes, including glycoside hydrolases (GHs) and polysaccharide lyases, to degrade and utilize carbohydrates depending on their habitat. Bacteria belonging to the Bacteroidota phylum possess polysaccharide utilization loci (PUL) that are involved in the systematic recognition and degradation of glycans and transport of the degradation products. PULs encode proteins such as cell surface sugar-binding proteins, sugar transporters, carbohydrateactive enzymes, such as GH and polysaccharide lyases, and transcriptional regulators (1). The study of PULs began when Salyers discovered the starch utilization system (Sus) of Bacteroides thetaiotaomicron, a human intestinal bacterium (2)(3)(4). Sus is composed of SusR, a transcriptional regulator, cell surface sugar-binding proteins (SusD, SusE, and SusF), TonBdependent transporter (SusC), and GHs (SusA, SusB, and SusG) (5)(6)(7). These proteins cooperate to capture starch and break it down to oligosaccharides on the cell surface and then to D-glucose in the periplasm (8)(9)(10)(11). The sensor/regulator protein SusR recognizes maltose in the periplasm, leading to the rapid upregulation of Sus genes (6). Since then various PULs have been reported and predicted from many Bacteroidota species, including B. thetaiotaomicron. As of January 2023, 55,351 PULs from 1760 Bacteroidota species have been registered in the PUL database (http://www.cazy.org/PULDB/) (12). In recent years, several PULs that target fructan, chitin, hemicellulose, pectin, glycosaminoglycan, N-glycan, and mucin-type O-glycan have been reported, and many GHs with novel activities and amino acid sequences have been found (13)(14)(15)(16)(17)(18)(19)(20). Although the number of reports on PULs for plant cell wall polysaccharides and mammalian glycoconjugates increases, there are few reports of PULs targeting microbial exopolysaccharides (EPSs), except for one targeting mycobacterial D-arabinan (21).

Biochemical and structural characterization of FjGH66
FjGH66 is a member of the GH66 family, which contains dextranase and cycloisomaltooligosaccharide glucanotransferases (CITase) (50). FjGH66 shares 47.6% and 39.7% amino acid sequence identity with dextranases from B. thetaiotaomicron and Thermoanaerobacter pseudethanolicus (TpDex), respectively. In addition, FjGH66 was predicted to be a dextranase because there is no CITase-specific CBM35 domain insertion, which facilitates intramolecular transglycosylation to produce cycloisomaltooligosaccharides, into the catalytic module of FjGH66 (Fig. S2) (50). Recombinant FjGH66 showed the highest hydrolytic activity at pH 5.5 and 45 C when dextran 40,000 was used as substrate (Fig. S3, A and B). FjGH66 exhibited superior hydrolytic activity toward linear dextran, and the resultant products comprise glucose, IG2, IG3, IG4, and IG5 (Table 1 and Fig. 3A). FjGH66 degraded IG3, IG4, and IG5 while displaying little hydrolytic activity against IG2, indicating that FjGH66 is a typical endo dextranase (Fig. 3A). FjGH66 showed slight hydrolytic activity toward S-32 α-glucan and S-64 α-glucan and did not hydrolyze soluble starch and pullulan ( Table 1). The kinetic parameters for dextran 40,000 and dextran 200,000 have the same order of magnitude but are slightly different; therefore, its substrate-binding and hydrolytic activity may be affected by the molecular weight of the substrate ( Table 2).
Substrate specificity of FjDex31A on isomaltooligosaccharides and α-glucans FjDex31A not only hydrolyzes dextran but also catalyzes transglycosylation when IG2 is used as a substrate (45,47). In this study, we investigated how FjDex31A acts on IG3, IG4, and branched dextran in addition to IG2 and linear dextran. FjDex31A showed hydrolytic activity against S-64 α-glucan and S-32 α-glucan as well as FjGH66 (Table 1). By contrast, FjDex31A showed glucose-producing activity against IG3 and IG4 (Table 1 and Fig. 3B). FjDex31A produced glucose from the initial stage of the reaction when isomaltooligosaccharides were used as substrates and catalyzed transglycosylation against IG3 and IG4, promoting the formation of longer isomaltooligosaccharides (Fig. 3B). The activity of FjDex31A against IG3 and IG4 was almost the same, and these specific activities were almost 47-and 3-fold higher than those against IG2 and dextran 40,000, respectively ( Table 1). The k cat /K m values for dextran 40,000 were two times higher than that for dextran 200,000 (Table 2).

Degradation of branched dextran by Bacteroidota bacterium
PspAG97A in specificity in terms of substrate length and types of linkage other than the α-(1→6)-linkage.

Degradation of branched dextran by Bacteroidota bacterium
the anomeric C1 atom of glucose at subsite −1 of the glucose complex at a distance of 2.4 Å (Fig. 5B). The reducing end of IG2 is directed toward the solvent, whereas 6-OH of Glc +1 at the nonreducing end is blocked by Arg74 (Fig. 5B).

Hydrolysis of α-(1→2)and α-(1→3)-branched dextran is orchestrated by FjDexUL GHs
Based on the substrate specificity of each FjDexUL GH examined, we investigated whether the GHs work together to degrade α-(1→2)and α-(1→3)-branched dextran. Dextran 200,000, S-64 α-glucan, and S-32 α-glucan were treated with various combinations of FjDexUL GHs (Fig. 6A). Glucose production from S-64 and S-32 α-glucans by a single enzyme was low (<0.21 mM glucose produced). When S-32 α-glucan was treated with FjDex31A, FjGH97A, and FjGH65A, 1.20 mM glucose was produced. Glucose production increased 1.8-fold (2.17 mM) when FjGH66 was included in the reaction (Fig. 6A). Because FjGH66 generates oligosaccharides from S-32 α-glucan, which is a substrate for exoacting enzymes (FjGH97A and FjGH65A), glucose production increased in the presence of FjGH66. Thus, FjGH66 can release α-(1→2)and α-(1→3)-glucosylated isomaltooligosaccharides from S-32 α-glucan. The concentration of glucose released from S-32 α-glucan was 21.1% lower when FjGH65A was not included in the reaction. The most significant difference in glucose production was observed in reaction mixtures with or without FjGH97A. The addition of FjGH97A increased the amount of glucose produced from S-32 α-glucan by 1.50 mM (Fig. 6A). Moreover, the amount of glucose produced from S-64 α-glucan was approximately onehalf that of S-32 α-glucan in the presence of all FjDexUL GHs (Fig. 6A). This implies that FjDexUL is more proficient in breaking down S-32 α-glucan-type structures than other αglucans. The degradation products of S-32 α-glucan were analyzed by normal-phase high-performance liquid chromatography (HPLC). The HPLC system can separate glucose, IG2, IG3, IG4, and IG5. The retention times of glucose, IG2, IG3, IG4, and IG5 were 5, 6, 7.5, 9, and 11 min, respectively (Fig. 6B). When S-32 α-glucan was treated with FjGH66, peaks corresponding to glucose, IG2, and IG3 were detected; however, with FjDex31A, peaks of glucose and IG2, as well as a distinct peak at retention time 8.5 min was detected. This peak is likely a minor product of transglycosylation and did not disappear after prolonged incubation or additional enzyme treatment (data not shown). When S-32 α-glucan was treated with all four enzymes, the IG3 peak (produced by FjGH66) disappeared, and the glucose peak was significantly larger. This result suggests that exo-type enzymes play an important role in glucose production from S-32 α-glucan (Fig. 6B). Moreover, endodextranase, α-glucoside hydrolase, and kojibiose hydrolase synergistically degrade S-32 α-glucan into glucose.

Binding profiles of FjDusD and FjDusE
FjDusD and FjDusE belong to the SusD and SusF_SusE superfamilies and share 24% and 21% sequence identity with SusD and SusF, respectively. The AlphaFold2 models of FjDusD and FjDusE are shown in Fig. S6 (56). The FjDusD model contains a tetratricopeptide repeat, similar to SusD (PDB 3CKC), and most likely has one ligand binding site (57). A comparison of the FjDusD model and crystal structures of SusD and the SusD homolog BT1762, derived from the levan utilization locus of B. thetaiotaomicron (PDB 6ZAZ), showed that the overall structures are similar (Fig. S7A). However, the conformation of the ligand-binding cleft appeared to vary based on the bound ligand (Fig. S7, B-D). The amino acid residues forming the predicted ligand-binding cleft of FjDusD were mostly conserved in closely-related homologs of FjDusD from other bacteria (Fig. S8). FjDusE has an immunoglobulin superfamily (Ig) domain and three β-sandwich carbohydratebinding modules (FjDusEa, FjDusEb, and FjDusEc) similar to SusF (PDB 4FE9) and may have three ligand-binding sites (58). The amino acid residues responsible for ligand recognition in FjDusE were found to be partially conserved in SusF (Fig. S9).  First, the affinity of these proteins toward carbohydrates was evaluated using native polyacrylamide gel electrophoresis with gels containing 0.5% (w/v) linear dextran, S-32 α-glucan, S-64 α-glucan, soluble starch, and inulin (fructan). The protein mobility profiles indicated that FjDusE binds tightly to linear dextran, S-32 α-glucan, and S-64 α-glucan and weakly to soluble starch (Fig. 7). By contrast, FjDusD displayed no significant interaction with these glucans. Next, the binding ability of FjDusD for oligosaccharides was analyzed using isothermal titration calorimetry (ITC). FjDusD showed affinity toward IG3, IG4, IG5, and IG6 but not IG2, kojibiose, kojitriose, nigerose, maltose, and maltotriose (Table 4 and Fig. S10). The K d and ΔG values of FjDusD were IG5 < IG4 < IG6 < IG3. Notably, IG5 exhibited the highest binding affinity with FjDusD, based on the lowest K d and ΔG values observed.
Three distinct regulatory systems that are specific to Bacteroidota have been proposed to be responsible for the transcriptional regulation of PUL genes in Bacteroidota bacteria: SusR-like regulators, hybrid two-component systems (HTCSs), and extracytoplasmic function sigma/anti-sigma factors. These regulators perceive signals by directly interacting with oligosaccharide products through their periplasmic domains. After maltose binds to SusR, the expression of all Sus components is upregulated (6,62). The periplasmic region of the HTCS regulator of PUL, which targets plant cell wall polysaccharides found in B. thetaiotaomicron and  (63). Therefore, the PUL regulator recognizes specific linkage types in oligosaccharide inducers. S-32 α-glucan remarkably upregulated the expression of FjDexUL genes. However, linear dextran, kojibiose, and IG2 did not upregulate gene expression to the same extent as S-32 α-glucan. Because the upregulation of gene expression may be influenced by factors such as ease of degradation by FjGH66 and ease of uptake by the FjDusC-FjDusD complex, it is difficult to determine the structure of the FjDusR ligand from our results. However, considering that gene expression is also induced by short oligosaccharides, IG2, and kojibiose, oligosaccharides containing α-(1→6) and α-(1→2) linkages are likely involved. Compared with S-32 αglucan, S-64 α-glucan contains other types of linkages, such as α-(1→4) linkages and different proportions of linkages. Therefore, it is less susceptible to FjGH66, resulting in a smaller amount of oligosaccharides, which are ligands for FjDusR. Consistently, FjGH66 activity against S-64 α-glucan is slightly lower than that against S-32 α-glucan, and the synergistic effect of FjDexUL GHs on S-64 α-glucan is lower. This suggests that FjDexUL proteins strictly recognize differences in structure and branch frequency between α-(1→2)and α-(1→3)-branched dextran.
Several studies on the functions of SusD and SusE homologs have been reported. Deletion of susD or mutation in the sugarbinding site of SusD inhibited the growth of B. thetaiotaomicron on starch (2,57). SusD helped import degradation products from starch into the periplasm and showed affinity only for maltooligosaccharides longer than maltoheptaose (57). In this study, FjDusD had an affinity toward oligosaccharides like SusD, but no detectable affinity for dextran unlike SusD, which had an affinity for amylose (Table 3 and Fig. 7) (3). A few studies have reported SusD homologs that do not bind polysaccharides (64)(65)(66). Reverse genetic analysis and mutagenesis studies have shown that the binding of polysaccharides by SusD homologs is not essential for microbial growth in the presence of SusE homologs (7,(65)(66)(67)(68)(69). The affinity of FjDusE for linear and branched dextrans suggests that FjDusE contributes to the binding of branched dextran on the cell surface. In this study, FjDusD strictly recognized α-(1→6)-linkages. Further analysis is  needed to evaluate whether FjDusD can bind isomaltooligosaccharides with α-(1→2)or α-(1→3)-branches. The SusD homolog (BT1762) of the levan utilization locus of B. thetaiotaomicron binds a fructooligosaccharide with a β-(2→1)-linked branch (Fig. S7) (70). To elucidate the glucanbinding modes and physiological functions of FjDusD and FjDusE, crystallographic studies with ligands and fjdusD and fjdusE deletion strains are warranted. Several studies on dextran with α-(1→2)and α-(1→3)branches have been reported; the polysaccharide is synthesized as an EPS by Leuconostoc spp. found in wheat sourdough and fermented products and by Lactobacillus sp. from the digestive tract of poultry (22,71,72). EPSs have been demonstrated to play a crucial role in bacterial colonization, adherence, stress resistance, host-bacterium interactions, and immunomodulation (72). This study provided insight into the mechanism of α-(1→2)and α-(1→3)-branched dextran degradation by Gram-negative bacteria, which is different from that of α-(1→2)-branched dextran degradation by the Gram-positive bacterium M. dextranolyticum (34). M. dextranolyticum removes α-(1→2) branches extracellularly, degrades dextran, and uptakes isomaltooligosaccharides into the cytosol, whereas F. johnsoniae extracellularly degrades branched dextran to oligosaccharides (with branches) and hydrolyzes all linkages, including branches, in the periplasm. In addition, α-(1→2)and α-(1→3)-branched dextran exhibited resistance to hydrolysis by digestive enzymes of both human and animal origin (73). Such resistance is attributed to the presence of branches, and the degree of resistance to hydrolysis by degrading enzymes, such as endodextranases, tends to increase with the percentage of branches. Intestinal microorganisms, such as Bifidobacterium and Bacteroides spp., can ferment α-(1→2)-branched dextran and α-glucooligosaccharides, containing α-(1→2)and α-(1→6)-glucosidic linkages, to produce short-chain fatty acids (73)(74)(75). Given its prebiotic potential, branched dextran is a promising candidate for enhancing gut health. The findings of this study revealed that the presence or absence of GH97 and GH65 in the dextran utilization locus of Bacteroides is a potential indicator for intestinal bacteria that can effectively metabolize α-(1→2)and α-(1→3)-branched dextran. Notably, several species of Bacteroides harbor GH97 and GH65 in their putative dextran utilization locus (Fig. 1).
In conclusion, we discovered the PUL targeting EPS from lactic acid bacteria and comprehensively investigated the gene expression and function of proteins present in FjDexUL, demonstrating that FjDexUL recognizes and degrades α-(1→2)and α-(1→3)-branched dextran. Because bacterial species possessing both GH65 and GH97 genes in dextran utilization loci are limited, it is likely that F. johnsoniae acquired FjDexUL genes to utilize α-(1→2)and α-(1→3)branched dextran. Our results will be helpful for understanding complex polysaccharide-mediated interactions between the microbiota in nature. Furthermore, FjDexUL enzymes are expected to be a useful tool for the structural determination of branched dextran.
qRT-PCR F. johnsoniae cells were precultured on 0.2% yeast extract and 0.1% MgSO 4 and inoculated in 5 ml of 0.2% yeast extract and 0.1% MgSO 4 containing 0.5% (w/v) carbohydrates (glucose, αglucobioses, and α-glucans). The cells were cultured for 48 h at 30 C and harvested by centrifugation (20,640g, 4 C, 5 min). Total RNA was extracted using NucleoSpin RNA plus (Takara Bio, Shiga, Japan). DNA was removed by DNase treatment using NucleoSpin RNA Clean-up (Takara Bio), and RNA quality was assessed by calculating absorbance ratios at 260/280 and 260/ 230 nm using the NanoDrop 2000c (Thermo Fisher Scientific). Then 1 μg RNA was used immediately for reverse transcription with PrimeScript RT Reagent Kit (Takara Bio). The expression of fjdusR, fjdusC, fjdusD, fjdusE, fjdex31a, fjgh66, fjgh97a, and fjgh65a was analyzed using quantitative PCR using THUN-DERBIRD SYBR qPCR Mix (Toyobo). The gene-specific primers are listed in Table S1. The program for thermal cycling was performed using the Mx3000P system (Agilent Technologies). The cycling conditions were 15 s denaturation at 95 C and 60 s annealing/extension at 60 C. Data were normalized to 16S rRNA transcript levels, and change in expression level was calculated as fold-change compared with media containing glucose cultures. The experiment was performed using three independent biological replicates. Quantification of relative transcript abundance was achieved using the ΔΔCt method. The primers for quantitative PCR were designed using the Primer-BLAST server (https://www.ncbi.nlm.nih. gov/tools/primer-blast/) (76).

Recombinant protein production and purification
Recombinant FjDex31A and FjGH65A were expressed and purified as described (46,47). The signal peptides of FjGH66, FjGH97A, FjDusD, and FjDusE were evaluated by the SignalP server (48). The genes for FjGH66 (GenBank ID, ABQ07435.1), FjGH97A (ABQ07433.1), FjDusD (ABQ07437.1), and FjDusE (ABQ07436.1) were amplified from F. johnsoniae NBRC 14942 (= ATCC 17061 = UW101) by colony-direct PCR using KOD ONE DNA polymerase (Toyobo) and each pair of primers is listed in Table S1. The amplified gene product was ligated into a pET28a (+) vector (Merck Millipore) using the NheI and XhoI restriction sites. Constructed plasmids were transformed into chemically competent E. coli BL21 (DE3) for overexpression. All proteins were produced in 1 L Luria-Bertani medium containing 50 μg/ml kanamycin. Cells were grown at 37 C with shaking until culture OD 600 reached 0.6 to 0.8, at which point isopropyl β-D-thiogalactopyranoside was added at a final concentration of 0.1 mM. Protein expression was induced by overnight shaking at 20 C. The cells were harvested, resuspended in 50 mM Tris-HCl buffer (pH 7.5) containing 300 mM NaCl and 20 mM imidazole and disrupted by sonication. The cell lysate was centrifuged to remove debris. The supernatant of the lysate was loaded onto a Ni-Sepharose excel (GE Healthcare) column, and unbound proteins were washed with the same buffer. Proteins were eluted with 50 mM Tris-HCl buffer (pH 7.5) containing 300 mM NaCl and 250 mM imidazole and then concentrated by ultrafiltration using an Amicon Ultra 30,000 molecular cutoff filter (Merck). FjGH97A was further purified by cation exchange chromatography using Mono S 5/50 GL (GE Healthcare). Equilibration was performed in 20 mM HEPES-NaOH buffer (pH 7.5) and eluted with a linear gradient of 0 to 1 M NaCl. FjGH66 was further purified with anion exchange chromatography using Mono Q 5/50 GL (GE Healthcare). Equilibration was performed in 20 mM Tris-HCl (pH 7.5) and eluted with a linear gradient of 0 to 1 M NaCl.

Synergistic activities of FjDexUL GHs on branched dextran
To determine if FjDexUL GHs function synergistically, a combination of 100 μg/ml FjDexUL GHs was used to hydrolyze dextran 200,000, S-64 α-glucan, and S-32 α-glucan at 0.5% (w/v) final concentration, pH 5.5, and 30 C for 10 min or 24 h. The released glucose was determined by the glucose oxidaseperoxidase method and degradation products were detected using HPLC. The reaction mixtures were then applied to a TSK-GEL amide-80 column (4.6 mm × 250 mm; Tosoh) immediately after incubation and eluted with 60% (v/v) acetonitrile at a flow rate of 1.0 ml/min at 30 C. The reaction products were detected using a refractive index detector (RID-10A, Shimadzu, Kyoto, Japan).

Crystallization and structure determination
FjGH65A (30 mg/ml in 20 mM sodium citrate buffer, pH 6.0, and 150 mM NaCl) was crystallized as described (46) and then soaked with reservoir solutions containing 10 mM IG2 for 1 min. FjGH66 (20 mg/ml in 10 mM HEPES-NaOH buffer; pH 7.0) was crystallized at 20 C using the hanging drop vapor diffusion method, where 1 μl of protein was mixed with the same volume of reservoir solution, consisting of 0.1 M Tris-HCl buffer (pH 8.5 and 9.0), 0.2 M lithium sulfate, and 20% (v/v) PEG 4000. For the glucose-and IG2-complexes, crystals of FjGH66 were soaked with reservoir solutions containing 10 mM glucose or IG2 for 1 min. For the IG3-complex, FjGH66 was cocrystallized using a reservoir solution containing 10 mM IG3. All crystals were cryoprotected with 20% (v/v) glycerol in reservoir solution and flash-frozen in liquid nitrogen. Diffraction data were collected at the BL5A beamline (Photon Factory). The data were processed with XDS and then scaled using AIMLESS, as implemented in the CCP4i2 package (81,82). The initial phase was determined with the molecular replacement method using MOLREP (83) with PDB 7FE3 and the Alpha-Fold2 model as a search model for the FjGH65A-IG2 complex and FjGH66, respectively (56). Manual model building was performed using COOT (84), and refinement was performed using REFMAC5 (85). Molecular images were made using PyMOL (Schrödinger LLC). Structural similarity searches were performed using the Dali server (86). Table 4 summarizes data collection and refinement statistics. All AlphaFold2 models were downloaded from the AlphaFold Protein Structure Database (56,87).

Isothermal titration calorimetry
All ITC experiments were performed using MicroCal iTC200 (Malvern Panalytical Ltd) calibrated to 25 C. All titrations were performed in 10 mM HEPES-NaOH buffer (pH 7.0). Purified FjDusD and FjDusE (0.1 mM each) were loaded into the sample cell, and the syringe was loaded with 5 mM IG5 and IG6 and 10 mM kojibisoe, kojitriose, nigerose, maltose, maltotriose, IG2, IG3, and IG4. An initial injection of 0.2 μl was followed by 19 injections of 2 μl spaced 150 s apart, with an injection duration of 4 s. Titration results were analyzed using MicroCal Origin ITC (Malvern Panalytical Ltd,). Data were fitted to a standard one-site binding model (n = 1), with ligand concentration as a variable. The average and standard error were calculated using data from at least three tests.

Data availability
The atomic coordinates and structure factors were deposited in the Worldwide Protein Data Bank (http://wwpdb.org/) under the accession codes 8IU8, 8IU9, 8IUA, 8IUB, and 8IUC. All other data are contained within the article.

Supporting information-This article contains supporting information.
Acknowledgments-We thank the staff of the Photon Factory for Xray data collection. This research was approved by the Photon Factory Program Advisory Committee (proposals 2019G097 and 2021G013). We also thank Enago (www.enago.jp) for the English language review. Funding and additional information-This work was supported in part by the Japan Society for the Promotion of Science KAKENHI (grant Nos. 19K15748 and 23K05039).
Conflicts of interest-The authors declare that they have no conflicts of interest with the contents of this article.