Lysine 117 Residue Is Essential for the Function of the Hepatocyte Nuclear Factor 1α Free

Hepatocyte nuclear factor 1α (HNF1α) plays essential roles in controlling development and metabolism by regulating a large group of downstream genes (1,2). HNF1α was originally identified as a key transcription factor for liver-specific gene transcription (3,4). Subsequently, HNF1α was detected in the pancreas, kidney, and digestive tract (5–7). Complete deficiency of HNF1α in mice results in type 2 diabetes, hypercholesterolemia, hepatic dysfunction, dwarfism, renal Fanconi syndrome, and progressive wasting syndrome but not intestinal dysfunction (8). Liver dysfunction in HNF1α-null mice largely occurs because of decreased enzyme production, while kidney phenotypes of glycosuria/phosphaturia result from inadequate transport proteins in the renal proximal tubule (8). Moreover, HNF1α-null mice have a severe transcriptional defect in the phenylalanine hydroxylase (Pah) gene, leading to elevated phenylalanine (PHE) levels in plasma (8,9). HNF1α-null mice also develop diabetes owing to the alteration of the genes that regulate β-cell responses to secretagogues (8–10). Heterozygous mutations in the HNF1α gene are clearly linked to the occurrence of maturity-onset diabetes of the young (MODY3) in humans (11–17). Heterozygous mutation of HNF1α causes a progressive insulin deficiency in humans, which is characterized by mild hyperglycemia in childhood and diabetes in early adulthood (18,19). However, HNF1α heterozygous-deficient mice are perfectly normal (8,9).

HNF1α is a lysine-rich protein containing 22 lysine residues within different functional domains. Several lysine residues have been shown to be important for HNF1α functions, and their mutations may disrupt DNA interactions of HNF1α (K158N, K205Q), or POU-specific (POUs)–POU_H interactions (K205Q) and are associated with MODY3 (20,21). Lysine 117 residue (K117) is highly conserved among different species and between members of the POUs domain transcription factor family (22). Importantly, the K117 to glutamic acid (K117E) mutation in the HNF1α protein was clinically associated with MODY3 (22), implicating a crucial role for lysine 117 in HNF1α functions, but no data are available on this variant. Lysine 117 residue is located in the POUs domain of HNF1α protein, which is more sensitive to single amino acid changes than the transactivation domain. So far, over 20 MODY3 mutations have been identified in this domain (21,23). The functional characterization of nearby variant substitutions, including P112L, L107R, and Y122C, revealed an association with reduced DNA binding and transcriptional activity of HNF1α (24–26), indicating that this region affects HNF1α functions in many ways. Crystal structure analysis did not provide the functional clues of 117 lysine residue in HNF1α (21). In addition, acetylation of HNF1α lysine residues have been shown to be required for transcriptional activation of HNF1α (27,28). Whether lysine 117 can be modified by acetylation is a very interesting question. For the above reasons, we were interested in knowing whether and how lysine 117 affects the properties of HNF1α.

In this article, we addressed the role of lysine 117 in HNF1α functions using a knock-in animal model and site-directed mutagenesis. Our results clearly demonstrate that lysine 117 residue is essential for the function of the HNF1α, and we support the fact that K117 mutation in HNF1α is loss-of-function mutation associated with MODY3.

HNF1α K117E homozygous (K117E Hom) mice, harboring the human pathogenic mutation (replacement of K with E) in the mouse ortholog, were generated using the CRISPR/Cas9 system on a C57BL/6J background. Seven heterozygotes (Het) were obtained and then backcrossed with wild-type (WT) C57BL/6J mice. DNA isolated from the mouse tail was used for sequencing and PCR genotyping. For sequencing, the following primers were used: HNF1α primer F (5′-CCAGGTTCAGCAGAGAGAAGGAGG-3′) and HNF1α primer R (5′-TGCAAGTACGACTTGACCATTTCG-3′). All animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC-DWZX-2021-760) of Beijing Institute of Radiation Medicine.

Both glucose tolerance test and insulin tolerance test were carried out as previously described (29).

The RNA-sequencing (RNA-seq) analysis was performed as previously described (30). Briefly, RNA libraries were prepared using the RNA sample prep kit (Invitrogen) and sequenced on the MGISEQ-2000. The GEO accession numbers for the RNA-seq data reported in this article are GSE207758 and GSE232242.

The Deseq2 package was used to identify down-expressed genes. Genes with log two-fold change ≥1 or ≤−1 and P value ≤ 0.01 (without adjustment) were identified as differentially expressed. Only protein-coding genes were used for comparisons with microarray data of HNF1α-null mouse. Microarray data of HNF1α-null liver were downloaded from ArrayExpress (ID: E-MEXP-1709). To identify the target genes of HNF1α, the bed file of HNF1α HepG2 chip-seq was downloaded from the ENCODE portal (https://www.encodeproject.org/) with file identifier ENCFF540TRC (31). Homer software (32) was used to identify HNF1α binding sites.

In vitro and in vivo luciferase assays were performed as previously described (33,34). Luciferase activities were expressed as -fold induction relative to values obtained from controls. At least three independent experiments were conducted in duplicate.

Electrophoretic mobility shift assay (EMSA) was carried out as previously described using a LightShift chemiluminescent EMSA kit (20148; Pierce) (33,34).

Bacterial-expressed GST-tagged truncated WT HNF1α and K117 mutants HNF1α (1–287 aa) and His-tagged full-length WT HNF1α were purified and immobilized as previously described (33,34).

For the pull-down of HNF1α response elements (HREs), double-stranded DNA oligos containing 12 different HREs (35) were synthesized as affinity reagents to enrich HNF1α proteins from nuclear extracts of mouse liver or HepG2 cells transfected with WT or mutant HNF1α as described (36).

All data are presented as the mean ± SEM and analyzed using GraphPad Prism (version 8; GraphPad Software, San Diego, CA). Student t tests were used for comparisons between two groups. Log-rank test was used to compare the survival distributions between groups. Two-way ANOVA was used for multiple group comparisons. P < 0.05 was considered statistically significant.

Further applied methods are described in the Supplementary Materials.

All data and the resource information are contained within this article.

To determine the impact of K117E mutation on HNF1α functions, we generated a C57BL/6J knock-in mouse model wherein the lysine 117 (K117) was replaced with glutamic acid (E117) in the endogenous murine HNF1α locus by CRISPR/Cas9 and further verified by sequencing analysis (Supplementary Fig. 1). Quantitative PCR and immunoblot analysis revealed that both HNF1α mRNA and protein levels in the liver and kidney of K117E homozygous (K117E Hom) mice were comparable with those in WT mice (Fig. 1A and B). In addition, the protein levels of HNF1β, HNF3β, HNF4α, and HNF6 were unaffected in the liver of Hom mice (Fig. 1C). The Hom mice were born much less frequently than expected and displayed growth retardation phenotypes (Fig. 1D). At the age of 9 weeks, the body weight of Hom mice was only 64.5% that of their heterozygous (Het) and WT littermates (Fig. 1E and F). Life span analysis showed that only 36.7% Hom mice survived to 12 weeks, and less than 15% survived to 16 weeks (Fig. 1G). These data indicate that Hom mice suffer from progressive wasting syndrome as reported for HNF1α-null mice (8), albeit at a slower rate.

At 3 and 9 weeks after birth, K117E Het mice showed normal blood glucose levels compared with WT mice, but Hom mice displayed higher blood glucose concentrations than those of Het and WT littermates (Fig. 2A). The circulating insulin levels and staining of pancreas sections with an insulin antibody indicate that Hom mice have relative insulin deficiency (Fig. 2B and C). RNA-seq analysis revealed 51 up-expressed genes and 39 down-expressed genes (P value ≤0.01, |log two-fold change| ≥1) between Hom and WT pancreas. Gene set enrichment analysis using the KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway database showed that several pathways, including pancreatic secretion, insulin resistance, carbohydrate digestion, and absorption and glycolysis, were enriched in differentially expressed genes (DEGs) of Hom pancreas (Fig. 2D). Quantitative PCR validated several DEGs including Cela2a, Ctrb1, Pdx1, Hnf4α, Shp1, Hkdc1, Glut2, and Nr5a2 (Fig. 2E). Insulin tolerance tests showed that 9-week-old Hom mice were not resistant to insulin (Fig. 2F). Interestingly, Hom mice had 3–10 times higher diuresis than control mice and an approximately 20,000 times higher urine glucose level than that of Het and WT littermates (Fig. 2G). These data indicate that both insulin secretion defects and renal glucose wasting contribute to abnormal glucose metabolism in mutant animals (8,9).

Next, we examined glucose metabolism in Het mice and found that Het mice at different ages exhibited fasting blood glucose and serum insulin levels comparable to their WT littermates (Fig. 3A and B). Nevertheless, Het mice develop progressive glucose intolerance with increasing age, while insulin sensitivity remains normal (Fig. 3C and D).

Liver/body weight ratio was greater in Hom mice than in WT mice, but no change occurred in kidney/body weight ratio (Fig. 4A and B). The levels of serum alanine aminotransferase and aspartate aminotransferase were markedly increased in Hom mice, indicative of liver damage (Fig. 4C). Moreover, elevated levels of serum triglycerides, total cholesterol, and bile acids were observed in Hom mice as compared with WT mice, while serum albumin levels were slightly decreased (Supplementary Fig. 2A). Compared with WT mice, the crystalline osmotic pressure showed no significant difference (Supplementary Fig. 2A). Urine chemistry analysis revealed that Hom mice developed severe renal dysfunction including glucosuria, phosphaturia, and elevated creatinine levels (Figs. 2G and 4D and Supplementary Fig. 2B). The mRNA and protein levels of sodium-glucose cotransporter-2 (SGLT2) in the kidney of Hom mice were significantly lower than their WT littermates, implying inadequate transport proteins in the renal proximal tubule of Hom mice (Fig. 4E). Furthermore, Hom mice exhibited hyperammonemia including hydroxyproline, citrulline, sarcosine, and L-glutamine (Fig. 4C). Unexpectedly, the serum PHE contents of Hom mice were normal during the 9 weeks of postnatal development, which is inconsistent with previous reports on HNF1α-null mice (8,9) (Fig. 4F). Accordingly, both the mRNA and protein expression of PHE hydroxylase in the liver of Hom mice were comparable to those of WT mice (Fig. 4G). Collectively, our data suggest that Hom mice develop hepatic and renal dysfunction similar to the phenotypes observed in HNF1α-null mice, except for hyperphenylalaninemia.

To support the notion that K117E mutation impaired HNF1α target gene expression, we conducted RNA-seq analysis using liver RNAs from Hom mice; 412 up-expressed genes and 487 down-expressed genes (P value ≤ 0.01, |log two-fold change| ≥1) between Hom and WT liver were identified. Considering the same cutoff (log two-fold change ≤−1, P value ≤ 0.01), K117E Hom mouse liver and HNF1α-null mouse liver (37) shared 181 down-regulated genes. The down-regulated genes in K117E liver were significantly enriched in the down-regulated genes set in HNF1α-null liver (P value = 5.339e-201, hypergeometric test). Among the top 100 down-regulated genes, 39 genes displayed a reduced expression pattern in HNF1α-null liver (Supplementary Table 1). We further analyzed the biological pathways enrichment using the KEGG pathway database, and the results showed that the DEGs in K117E liver were significantly enriched in several pathways, including steroid hormone biosynthesis, chemical carcinogenesis, drug metabolism, and bile acid secretion (Supplementary Table 2). Among the 20 top-ranked pathways down-regulated in K117E liver, steroid hormone biosynthesis pathways, metabolism of xenobiotics by cytochrome P450, drug metabolism–cytochrome P450, and arachidonic acid metabolism were listed in the 20 top-ranked pathways down-regulated in HNF1α-null liver (37). These data clearly demonstrate that K117E Hom mice liver and HNF1α-null mice liver share highly similar gene expression profiles. Among 181 genes down-regulated in both K117E liver and HNF1α-null liver, 41 genes contain HNF1α binding sites (Supplementary Table 3). Twenty-seven genes containing HNF1α binding sites were only down-regulated in HNF1α-null liver (Supplementary Table 4). In addition, another 31 target genes were only observed to be down-expressed in K117E liver, indicating that HNF1α K117E mutant may have distinct transcriptional regulatory properties (Supplementary Table 5).

We measured the overall transcriptional activities of WT and K117E mutants by performing luciferase reporter assays using the promoter region of HNF1α target gene α fetoprotein (AFP) in either HepG2 cells, which express endogenous HNF1α, or HeLa cells, which do not express endogenous HNF1α. As shown in Fig. 5A, overexpression of WT HNF1α significantly increased HNF1α-dependent luciferase activity, but the K117E mutant produced a greater reduction in transcriptional activity in both HepG2 and HeLa cells (Fig. 5A and B). Consistent results were obtained with GLUT2-luc and β28-Luc promoters (38), two other targets of HNF1α, in HepG2 cells (Fig. 5C and D). Increasing amounts of the K117E mutant did not interfere with WT HNF1α activity in HeLa cells, indicating that K117E mutant HNF1α have no dominant-negative effect on WT HNF1α (Fig. 5E). Furthermore, the reduced transcriptional activity of K117E mutant HNF1α was confirmed by in vivo luciferase assays as previously described (30) (Fig. 5F). Our results further showed that the expression of several genes previously identified to be induced upon HNF1α transduction (22) increased in cells overexpressing WT HNF1α; however, K117E mutants significantly abrogated this induction (Fig. 5G). Similarly, HNF1α K117E mutant in RINm5F cells (rat insulinoma cell line) failed to stimulate Pck1, Ins2, and cyclophilin transcription, compared with enforced expression of WT HNF1α in the same cells (Fig. 5H). These data suggested that K117E mutation caused a significant reduction in overall transactivation of HNF1α.

To investigate whether the DNA binding potential of HNF1α was affected by K117E mutation, we compared WT and K117E mutants for their ability to interact with a DNA oligonucleotide probe containing Alb, Fgb, and Sglt2 promoter elements (35). As expected, EMSA using purified HNF1α-GST protein showed that WT HNF1α recognized DNA target baits (Fig. 6A). In contrast, K117E mutant showed markedly decreased binding. We further examined the effect of K117E mutation on DNA binding in RINm5F cells using chromatin immunoprecipitation assays (Fig. 6B). The results revealed that the occupancy of WT HNF1α in Tmem27, Kif12, Tmed6, Gc, Fbp1, Glut2, Ins, and HNF4α regions of the DNA promoter increased in WT HNF1α-transfected RINm5F cells, but little or no PCR product was generated in cells transfected with K117E mutant or normal IgG as negative control. To further confirm this, we synthesized 12 different HREs as affinity reagents to enrich HNF1α proteins from liver nuclear extracts of both WT mice and K117E Hom mice (Supplementary Fig. 3) as previously described (36). Immunoblot analysis showed that WT HNF1α was efficiently enriched by HREs constructs, whereas HNF1α K117E was significantly decreased. The specificity of binding was identified by mutant HREs constructs (Fig. 6C and D). The findings indicate that K117E mutation impairs the DNA binding activity of HNF1α.

To further determine the molecular mechanism of lysine 117 affecting HNF1α functions, we generated two other HNF1α mutants, K117R and K117Q, in which the lysine 117 residue was replaced with arginine (R) and l-glutamine (Q) residues, respectively. The K117R mutation retained a positive charge, avoiding modifications such as acetylation and ubiquitin, while the K117Q mutation neutralized the positive charges of lysine-117 residue (39,40). Both mutants had obvious defects in activation of AFP-, Glut2-, and β28-luc promoter, compared with WT HNF1α (Fig. 7A–C). Similar to the K117E mutant, the reduced transcriptional activities of K117R and K117Q mutants were not due to their decreased protein stability or cellular localization (Supplementary Fig. 4 and Fig. 7D and E). Moreover, EMSA revealed that the DNA binding ability to target probes significantly decreased in all mutant proteins compared with that of WT HNF1α (Fig. 7F). Further, HRE pull-down identified that HNF1αK117 mutation attenuated its interaction with response elements of target genes (Supplementary Fig. 5). These data demonstrate that lysine 117 is required for transactivation activity and binding to DNA of HNF1α.

Both HNF1α K117R and K117Q mutants exhibited functional defects, suggesting K117 may not affect HNF1α function through acetylation modification. To further confirm, we demonstrated the acetylation of HNF1α and its mutants in HepG2 cells after treatment with nicotinamide and Trichostatin A using a pan-antiacetyl lysine antibody. The results showed that the acetylation of HNF1α was more pronounced under nicotinamide/Trichostatin A treatment, but this process was not affected by K117 mutants, indicating that lysine117 may not be a major acetylation site in HNF1α (Supplementary Fig. 6). Together with the results that both K117Q and K117R mutants were dysfunctional, we conclude that acetylation of residue lysine 117 may not be involved in regulation of HNF1α functions.

Homodimerization of HNF1α is essential for its DNA binding and transcription activity (41). According to crystal structure, K117E mutant is not predicted to directly contact the DNA or disturb POU_S and POU_H domains interaction and nuclear localization, which suggests that lysine 117 residue may affect HNF1α function through other mechanisms. The monomer and dimer structures of K117E mutant predicted by AlphaFold Protein Structure Database were similar to those of WT HNF1α (Supplementary Fig. 7). Then we examined the ability of the mutant protein to interact with itself by coimmunoprecipitation assay. As expected, when flag-HNF1α and myc-HNF1α were coexpressed, both were found to coimmunoprecipitate with anti-flag antibody, confirming that WT HNF1α undergoes homodimerization (Fig. 8A). In contrast, K117 mutations significantly attenuated the self-association. While K117E almost completely failed to self-associate, K117R and K117Q mutants showed a significant decreased self-association (Fig. 8A).

To further evaluate the effect of K117 mutation on HNF1α homointeraction in cells, we performed FRET (Fluorescence Resonance Energy Transfer). The coexpression of WT HNF1α-YFP with WT HNF1α-CFP generated a strong FRET signal, indicating that WT HNF1α self-associated efficiently (Fig. 8B). In contrast, transfection of K117 mutant HNF1α-CFP with K117 mutant HNF1α-YFP resulted in a significantly lower FRET signal. Collectively, our data clearly demonstrate that K117 mutation impairs the HNF1α activity through disrupting the self-association process.

The dimerization of HNF1α is mediated by dimerizing cofactor for HNF1α (DCoH), which directly associates with the dimerization domain of HNF1α to stabilize the dimeric (42), DNA-binding form of HNF1α. Coimmunoprecipitation assay showed that K117E mutant did not affect DCoH binding with HNF1α in the liver of K117E Hom mice (Fig. 8C).

In this study, we demonstrated that HNF1α K117E Hom mice developed dwarfism, hepatic dysfunction, renal Fanconi syndrome, diabetes, and progressive wasting syndrome. These phenotypes were very similar to those of mice with complete HNF1α deficiency, although the deficient mice present a more rapid disease progression (8). Moreover, K117E Hom mice shared similar liver expression profiles with HNF1α-deficient mouse (37). These findings presented solid genetic evidence that K117E mutation in HNF1α is a loss-of-function mutation. HNF1α is important for Pah transcription, as that deficiency in mice results in hyperphenylalaninemia (8,9). However, the plasma content of PHE and hepatic expression of PHE hydroxylase in K117E Hom mice were normal compared with those of heterozygous and WT mice. The exact mechanism of this discrepancy is not yet clear. One possible explanation is that the failure to transcribe the Pah gene in the liver of HNF1α-null mice is correlated with the absence of an open chromatin configuration and was accompanied by hypermethylation in the Pah enhancer regions (43), indicating that the molecular mechanism of HNF1α regulating Pah expression is different from many other known HNF1α target genes in liver. The K117E mutant probably only slightly affects the chromosomal remodeling function of HNF1α, or the residual function of HNF1α K117E is sufficient to accomplish the remodeling of chromatin structure in Pah enhancer. In addition, strain differences between the K117E Hom mice (C57BL/6J background) generated here and the previously reported knockout mice (129 Svj 3 C57BL/6J or 129 Svj 3 C57BL/6J 3 FVB/N background) may also be part of the reason (8,9). Further investigation of the chromatin structure and methylation status of the enhancer region of the Pah promoter in K117E Hom mice will help reveal the different mechanisms by which HNF1α regulates chromosome remodeling or promoter transcription. Since the same wasting phenotype was observed in hph1 and hph5 knockout mice with hyperphenylalaninemia, the wasting syndrome presented by HNF1α-null mice is thought to be associated with hyperphenylalaninemia (8). Our current study found that Hom mice develop wasting syndrome without hyperphenylalaninemia, suggesting that hyperphenylalaninemia may not be the main cause of wasting syndrome in HNF1α-null mice.

K117E Hom mice exhibit diabetes by 3 weeks of age. The relative deficiency of serum insulin levels and the normal response to insulin treatment in diabetic K117E Hom mice closely resemble the characteristics of MODY3 disorder in humans (44). Transcriptome analysis showed that DEGs in pancreas of Hom mice were significantly enriched in pathways involved in pancreatic secretion, carbohydrate digestion and absorption, glycolysis, and others, such as Cele2a, Ctrb1, Pdx1, Hnf4α, Shp1, Glut2, and Nr5a2. Abnormal expression of these genes has been reported to affect insulin levels (10,45–47). Therefore, the K117E Hom mice reported here provide an opportunity to further examine the cause, at the molecular level, of diabetes due to MODY3 in humans. Previous reports suggest that nondiabetic carriers of the MODY3 mutation display an intermediary phenotype, with subnormal insulin secretion but decreased glucose tolerance (48). Interestingly, we found a progressive severity of glucose intolerance but unchanged insulin sensitivity in aged K117E heterozygous mice, the mechanism of which has not been elucidated in present studies, and further studies may help to elucidate the role of HNF1α heterozygous mutations in pathogenesis of MODY3 in human.

Previous studies show that the consequences of variant P112L and L107I substitutions reduced DNA binding and transcriptional activity of HNF1α but showed normal nuclear targeting (24,25). Another variant Y122C substitution decreased DNA binding together with reduced protein stability (26). In addition, A116V mutation was predicted to disrupt DNA recognition indirectly and decreas protein stability (21). These data indicate that at least codons 107–122 are important for functions of HNF1α. We here examined the possible mechanisms of how K117 mutants affect the functions of HNF1α; the results showed that K117 mutants are not likely to have decreased protein stability and prevention nuclear localization. Moreover, we provide evidence showing that lysine 117 is not a major acetylation site of HNF1α and does not affect HNF1α function through acetylation. In contrast, we found that K117 mutants were completely or partially incapable of binding to WT HNF1α or themselves but did not affect the association between HNF1α and DCoH, which is required for dimerization of HNF1α (49). Our present study thus reveals a previously unappreciated role of POUs domain in HNF1α homodimerization, which is essential for DNA binding and transcription activity of HNF1α (41). A deletion mutant containing only the N-terminal 72 amino acids has been shown to form cross-linked dimers, which suggests that sufficient structural information for homodimerization is contained within the first 72 amino acids at the N terminus of HNF1α (50). We therefore speculate that the dimerization defect caused by K117 mutation may be associated with the disruption of the tertiary structure of HNF1α protein, including the N terminus. Further studies on the role of K117 in the tertiary structure of HNF1α may help to reveal the molecular mechanism of dimerization defect caused by K117 mutation.

Collectively, our findings clearly demonstrate that the lysine 117 residue is essential for HNF1α function and we reveal a previously unappreciated role of POU domain in HNF1α homodimerization. Our results provide important clues for identifying the molecular basis of HNF1α-related diseases such as MODY3.

Acknowledgments. The authors thank Lifeng Wang from Beijing Institute of Radiation Medicine, PengJun Wang from Beijing Institute of Lifeomics, ZongHeng Fu from School of Chemical Engineering and Technology of Tianjin University, Xian Wang from Beijing Institute of Lifeomics, and YiQun Zhan from Beijing Institute of Radiation Medicine for technical support.

Funding. This work was supported by the National Natural Science Foundation of China (31871192), the China Postdoctoral Science Foundation (2018M643903), the Original Exploration Program of the National Natural Science Foundation of China (82150110), and Beijing Hospitals Authority’s Ascent Plan (DFL20221502).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. X.Y. and M.Y. conceptualized the study. Y.C., L.Z., X.L., and C.Z. provided methodology. Y.C., L.Z., X.L., S.C., and M.Y. provided formal analysis. S.X., J.L., X.W., Y.W., D.D., S.Y., C.L., R.Y., and G.R. provided investigation. H.C. provided resources and data curation. Y.C. and X.Y. wrote the original draft. X.Y. and M.Y. wrote, reviewed, and edited the article. M.Y. provided project administration. Y.C., J.L., and M.Y. acquired funding. All authors have read and agreed to the published version of the manuscript. M.Y. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.