Human Proislet Peptide Promotes Pancreatic Progenitor Cells to Ameliorate Diabetes Through FOXO1/Menin-Mediated Epigenetic Regulation

A hallmark of either type 1 or type 2 diabetes is the inadequate number of functional pancreatic β-cells at the late stage (1). Regeneration of β-cells, therefore, is an ideal approach to ameliorating diabetes (2). Various approaches have been explored to increase the number of β-cells. One is the transplantation of islets from donors to patients with diabetes (3). However, an insufficient number of donors (from cadavers) as well as transplant rejection through immune response are hurdles to overcome (4,5). Differentiation of inducible progenitor stem cells into β-cells also has been reported (6), but the functionality of the induced cells often is low. Human fetus–derived pancreatic progenitor cells (HFPPCs) can be induced into glucose-sensitive insulin-secreting cells capable of protecting nude mice from streptozotocin-induced diabetes (7). Of note, these cells exhibit characteristics of β-cells and produce circulating human insulin and C-peptide in transplanted mice (7). Thus, HFPPCs have been an effective source of functional β-cells to control diabetes in preclinical studies (8). However, the underlying mechanism is poorly understood.

To promote these embryonic progenitor cells to commit toward β-cell differentiation, an attractive approach is to administer trophic agents such as islet neogenesis-associating protein (INGAP) (9), which is encoded by the REG gene in hamster. INGAP induces islet neogenesis and can normalize hyperglycemia in diabetic mice and humans (10,11). The REG genes are expressed primarily during fetal development (12), whereas in adulthood, they are upregulated in response to acute pancreatic injury or conditions of increased insulin demands (13). Human proislet peptide (HIP), a 14–amino acid bioactive fragment encoded by the human REG3a gene, is suggested as a potential therapeutic peptide for the treatment of diabetes because it increases β-cell mass and improves glycemic control (14).

The forkhead transcription factor (TF) FOXO1 is a prominent mediator in controlling pancreatic β-cell mass (15). FOXO1 represses β-cell differentiation in the human fetal pancreas partly by suppressing the TFs, such as NGN3 and NKX6.1 (16), but how FOXO1 does so remains unknown. Of note, menin has been reported to be upregulated by FOXO1, which directly binds to the promoter of the MEN1 gene (17). Deletion of the menin gene results in increase of β-cell mass and prevents the development of streptozotocin-induced diabetes in mice (18). On the other hand, conditional deletion of menin leads to the reversal of preexisting hyperglycemia in several mouse models of diabetes (19). Thus, menin plays a key role in regulating β-cell mass in both physiological and diabetic conditions. However, whether and how menin regulates the crucial proislet TFs, such as PDX-1, MAFA, and NKX6.1, during differentiation of HFPPCs are poorly understood.

Differentiation of fetal progenitor cells toward a specific cell lineage requires the highly regulated and tightly controlled expression of several key TFs. The expression of TFs, such as PDX-1, MAFA, and NKX6.1, is induced during differentiation of progenitor cells into β-cell lineage (20). These TFs are closely linked to the differentiation of fetal progenitor cells into mature pancreatic cells (21). PDX-1 is essential for the early stage of β-cell maturation and maintenance of mature β-cell function (22). MAFA, a basic leucine zipper TF, is expressed solely in β-cells, controlling expression of the insulin gene through the cis-regulatory element RIPE3b1 (23) as a potent transactivator for the insulin gene (24). NKX6.1 is also a key TF for β-cell differentiation (25). Therefore, understanding whether and how these TFs are regulated by the menin pathway is important, given the crucial role of menin in regulating β-cell differentiation (26). In this study, we found that HIP promotes HFPPC differentiation into insulin-expressing cells through increasing the phosphorylation of FOXO1, represses menin expression, and reduces menin-mediated recruitment of SUV39H1 H3 lysine 9 (H3K9) trimethylation (H3K9me3) at the promoters of proislet TFs, enhancing HFPPC differentiation and thus ameliorating hyperglycemia in diabetic mice.

This study was approved by the Health Science Research Ethics Board at Shenzhen University (Shenzhen, China). Human fetal pancreata at 10–12 weeks fetal age were collected in accordance with protocols approved by the Clinical Research Ethics Board of China-Japan Friendship Hospital (Beijing, China) and conducted according to the principles of the Declaration of Helsinki. HFPPCs were isolated and cultured as previously described (7). Briefly, pancreatic tissues were dissected and digested by collagenase XI (Sigma, Shanghai, China). The islet-like structures were suspended and cultured in medium described in Fig. 1. Adherent cells that grew from the islet-like structures were trypsinized for passage with 0.1% trypsin/0.1% EDTA solution at confluence.

HFPPCs were induced in M199 medium containing 15% FBS, 10 mmol/L nicotinamide, 30 ng/mL all-trans retinoic acid, and 42 ng/mL glucagon-like peptide-1 (Abcam, Cambridge, U.K.) and treated with either 10 μg/mL scrambled peptide (SP) or 10 μg/mL HIP (Kaijie Peptide Company, Chengdu, China) for 7 days for RNA or protein extraction or for 14 days for transplantation.

After 2 weeks of differentiation, the cells were resuspended in M199 medium containing 20% FBS, 2 mmol/L glutamine, 5 μg/mL type IV collagen, 2 μg/mL laminin, 3 μg/mL fibronectin, 1.5 mmol/L Ca²⁺, and 1 mmol/L ATP. The mixture was then transferred into 50-mL centrifuge tubes and cultured in a 37°C, 5% CO₂ incubator. After 24 h incubation, small cell aggregations were harvested by centrifugation at 1,200 rpm for 8 min. The morphology of islet-like structures was examined with a stereomicroscope.

Male BALB/c nude mice (8–10 weeks of age) were purchased from the Experimental Animal Center of Guangdong Academy of Medical Science (Guangzhou, China) and bred in the specific-pathogen-free animal house at Shenzhen University. Mice were injected intraperitoneally with streptozotocin (Sigma, St. Louis, MO) at 180 mg/kg in citrate buffer and randomized into the study when blood glucose levels were ≥20 mmol/L for 3 consecutive days. Blood glucose levels were measured daily with an Accu-Chek Active glucometer (Roche Diagnostics, Mannheim, Germany).

Mouse islets were isolated and cultured as described previously (27). The mice were killed by cervical dislocation, and pancreatic islets were isolated by collagenase P digestion (Roche Life Science, Indianapolis, IN). The islets were handpicked and cultured overnight in DMEM with 15% FBS, 100 μmol/L penicillin G, and 100 mg/mL streptomycin.

Recipient nude mice for transplantation were randomly divided into four groups: sham-operated controls, transplantation with 200 mouse islets, or HFPPC-derived islet-like cell clusters treated with SP or HIP. After anesthetization, the diabetic nude mice underwent transplantation with the differentiated islet-like structures from 6 million HFPPCs or the equivalent number of islets under the left-side kidney capsule. The animal procedures were performed according to the Guide for the Care and Use of Laboratory Animals and approved by the Shenzhen University Animal Care Committee.

Total RNA from induced HFPPCs was extracted by using RNAiso Plus (Takara Bio, Dalian, China). Single-stranded cDNAs were generated with Bestar qPCR RT Kit (DBI Bioscience, Shanghai, China). Quantitative real-time PCR (qRT-PCR) was conducted by using Bestar SybrGreen qPCR mastermix (DBI Bioscience) with ABI Prism 7500 Sequence Detection System. Relative gene expression was measured by qRT-PCR and the 2^−ΔΔCT method.

Chromatin immunoprecipitation (ChIP) assay was performed with a ChIP Assay Kit (EMD Millipore, Temecula, CA). Briefly, 10⁶–10⁷ formaldehyde cross-linked cells were lysed and sonicated to obtain sheared DNA fragments. Lysates were incubated with either control IgG or antigen-specific antibodies at 4°C overnight followed by the addition of protein A agarose beads to be incubated for 1 h. Samples were eluted from the beads and reverse cross-linked at 65°C overnight for quantification by PCR.

pMX-puro-menin (10 μg), pMX-puro-FOXO1-AAA, or short hairpin RNAs (shRNAs) or vector for retroviral packaging were cotransfected with 10 μg psi-2 helper plasmid into 293T cells by using the calcium chloride precipitation method. The supernatant containing generated recombinant virus was collected every 12 h. Collected virus was filtered through a 0.45-μm filter and transfected into HFPPCs twice in medium containing 8 μg/mL polybrene followed by selection in 2 μg/mL puromycin (Sigma) for 4 days. Transfection of FOXO1-small interfering RNA (siRNA) and control siRNA was performed according to the RNAiMAX transfection procedure.

Total protein from HFPPCs was extracted with radioimmunoprecipitation assay lysis buffer (Beyotime Biotechnology, Nantong, China) and immunoblotted as previously described (19). The antibodies used are listed in Supplementary Table 1.

Islet-like cell clusters for insulin content assay were collected and lysed for protein extraction. Cell lysates (25 μL) were used for insulin measurement with human insulin ELISA kits (ALPCO, Salem, NH).

Retroviral plasmid pMX-puro-FOXO1-AAA was generated from pMX-puro-FOXO1 by using the QuikChange Site-Directed Mutagenesis kit (Agilent, Santa Clara, CA). pMX-puro-menin was constructed by inserting PCR-amplified menin cDNA into the BamHI/NotI site of the retroviral vector pMX-puro. Construction of shRNA for sh-MEN1 was performed, as previously described (28).

Data are presented as mean ± SEM for the indicated number of experiments. Statistical significance was evaluated using Student t test. Data were considered significant when P < 0.05.

HFPPCs can differentiate into insulin-producing cells in vitro and normalize hyperglycemia upon transplantation into diabetic mice (7). We examined the profile of endocrine markers and stem-cell markers in these HFPPCs and found that they expressed pancreatic endocrine markers PDX-1 and NGN3 as well as stem-cell markers OCT4 and NANOG consistent with the previous observations (7). We next determined the effect of the induction medium on expression of these markers. The HFPPCs were treated with the induction medium containing nicotinamide, all-trans retinoic acid, and glucagon-like peptide-1 for 7 days (Fig. 1A) followed by analysis of mRNA levels. We found that the 7-day induction substantially increased expression of PDX-1, MAFA, and NKX6.1 (Fig. 1B and C). Of note, the increase of these TFs was accompanied by increased transcription of the genes important for β-cell maturation and function, including insulin, GLUT2, SUR1, KIR6.2, and voltage-dependent Ca²⁺ channels (VDCC) (Fig. 1D and E). By contrast, the mRNA levels of TFs defining the stem cells, such as OCT4, NANOG, and CK19, were decreased (data not shown), as observed previously (7).

We next explored whether HIP influences HFPPC differentiation. HFPPCs were treated with the induction medium containing either 10 μg/mL HIP or 10 μg/mL SP for 7 days. mRNA analysis revealed that HIP increased expression of PDX-1, MAFA, and NKX6.1 by ∼1.5-fold (P < 0.05), ∼1.6-fold (P < 0.05), and ∼1.6-fold (P < 0.01), respectively, compared with the SP controls (Fig. 1B and C). Consistently, expression of insulin, GLUT2, SUR1, KIR6.2, and VDCC was also increased (Fig. 1D and E). These findings are consistent with increased insulin immunoreactivity (Fig. 1F and Supplementary Fig. 1) and insulin content (Fig. 1G) in HFPPCs.

To evaluate the function of these HIP-treated HFPPCs in vivo, we transplanted the HIP-treated islet-like structures from the differentiated HFPPCs under the left-side kidney capsule of diabetic nude mice. Induction and formation of islet-like structures were performed according to the experimental flowchart shown in Fig. 1H. The diabetic recipient mice were transplanted with either 200 isolated normal mouse islets or SP- or HIP-treated HFPPC clusters. As shown in Fig. 1I, transplantation with HIP-treated HFPPC islet-like clusters led to a significant reduction of blood glucose of the diabetic mice. The blood glucose level remained at ∼15 mmol/L (P < 0.01) as opposed to that (>20 mmol/L) of diabetic controls throughout 6 days posttransplantation. By contrast, transplantation with SP-treated islet-like structures had a much weaker effect (Fig. 1I). Of note, human insulin-positive cells were found under the capsule of the left-side kidney (Supplementary Fig. 2), confirming the function and survival of the graft.

Because FOXO1 negatively regulates β-cell differentiation (29), we examined whether HIP influences expression of FOXO1 in HFPPCs by treating HFPPCs with HIP or SP in the induction medium. Western blot results revealed that HIP potently decreased FOXO1 protein levels (Fig. 2A). Because FOXO1 is inhibited by its phosphorylation and the phosphorylated FOXO1 is sequestered in the cytoplasm (30), we determined whether HIP affects FOXO1 phosphorylation by using an antibody specific for FOXO1 phosphorylated at serine 256. Treatment with HIP dose dependently increased FOXO1 phosphorylation (Fig. 2B). Long-term treatment of HFPPCs with HIP, but not SP, increased the phosphorylation of FOXO1 (Fig. 2C), whereas the total amount of FOXO1 was reduced (Fig. 2C). The same observations were made for primary mouse islets (Supplementary Fig. 3).

Given that FOXO1 is regulated by phosphorylation, mutation of the phosphorylation sites serine to alanine, such as the AAA mutant (31), yields constitutively active FOXO1. We reasoned that constitutively active FOXO1 mutant FOXO1-AAA would suppress basal and HIP-induced expression of NKX6.1, PDX-1, and MAFA. To this end, HFPPCs were transduced with vector retroviral recombinant viruses or the viruses expressing FOXO1-AAA. The resulting cells showed the obvious FOXO1 expression (Fig. 2D). Ectopic expression of FOXO1-AAA reduced the basal levels of the TFs as determined by qRT-PCR (Fig. 2E), and the constitutively active FOXO1 potently suppressed HIP-induced expression of PDX-1, NKX6.1, and MAFA (Fig. 2F). These results indicate that HIP suppresses FOXO1 and contributes to the induction of proislet TFs, including PDX-1, NKX6.1, and MAFA, thus enhancing HFPPC differentiation.

Given that menin plays a crucial role in maintaining β-cell homeostasis (32) and stabilizing lineage identity (28) and that HIP suppresses FOXO1 by increasing its phosphorylation (Fig. 2A–C), we explored whether FOXO1 affects expression of menin in HFPPCs. The experiments were conducted by treating HFPPCs in expansion medium containing siRNA targeting FOXO1 for 2 days. Western blot results showed that siRNA knockdown of FOXO1 leads to reduction of menin protein expression (Fig. 2G). Consistently, treatment of the HFPPCs with FOXO1 inhibitor AS1842856 also reduced menin expression (Fig. 2H).

We next determined whether HIP affects expression of menin. Treatment of HFPPCs in either control or induction medium with either SP or HIP for 7 days showed that HIP reduced the menin mRNA level (Fig. 3A). Consistently, HIP also reduced the menin protein level, whereas SP had little effect (Fig. 3B and Supplementary Fig. 3). These findings indicate that HIP reduced menin expression. We also determined the effect of menin on expression of TFs in HFPPCs by knocking down menin with an MEN1 shRNA and found that menin knockdown led to increased expression of PDX-1, MAFA, and NKX6.1 (Fig. 3C and D).

These findings prompted us to investigate whether HIP stimulates these TFs through repressing menin. To this end, we treated HFPPCs with HIP or SP in the induction medium and performed ChIP assay to determine whether HIP affects menin binding and, if any, the histone modifications at the promoters of these TFs. Treatment of HFPPCs with HIP, but not SP, significantly reduces the binding of menin to the PDX-1 promoter (Fig. 3E). Of note, HIP also reduced the amount of histone H3K9me3, a marker for repressing gene expression, whereas the total amount of histone 3 at the promoter was not affected (Fig. 3E). Furthermore, HIP reduced recruitment of menin to the promoter of MAFA and NKX6.1 as well as H3K9me3 histone modification in HFPPCs (Fig. 3F and G). These results indicate that HIP suppresses expression of menin and subsequently reduced menin recruitment to the promoter of PDX-1, MAFA, and NKX6.1, thereby reducing their expression in HFPPCs.

To confirm the role of menin in regulating expression of these TFs in HFPPCs, we performed ChIP assay by using the control and menin knockdown HFPPCs. Menin knockdown substantially diminished its binding to the promoter of PDX-1 (Fig. 4A), MAFA (Fig. 4B), and NKX6.1 (Fig. 4C). Consistently, menin knockdown also reduced H3K9me3 marker levels but had no effect on total histone H3 at the promoters. To investigate the role of menin in mediating HIP-regulated expression of these TFs, we treated the control cells and menin-expressing cells with SP or HIP for 7 days in the induction medium and analyzed expression of these TFs. As expected, HIP induced expression of PDX-1, MAFA, and NKX6.1 in control vector cells (Fig. 4E). However, ectopic menin expression inhibited HIP-induced induction of these TFs, indicating that HIP promotes the expression of these proislet TFs by reducing menin expression.

Because menin knockdown reduced H3K9me3 at the menin and FOXO1 target genes (Fig. 4A–C) and SUV39H1, the gene expressing H3K9 methyltransferase, is one of the key histone modifiers generating H3K9me3 (33), we determined whether menin interacts with SUV39H1 in HFPPCs. Coimmunoprecipitation with anti-FOXO1 or antimenin antibodies showed that menin, but not FOXO1, pulled down SUV39H1 (Fig. 5A, lane 2). We also examined whether SUV39H1 binds to the promoters of the TFs. The ChIP assay indicated that menin knockdown markedly decreased SUV39H1 binding to the promoter of PDX-1, MAFA, and NKX6.1 (Fig. 5B–D). We next explored whether SUV39H1 is involved in mediating menin-dependent suppression of the proislet TFs. We treated the cells with either chaetocin, an SUV39H1 inhibitor (34), or control DMSO, and found that chaetocin increased expression of the proislet TFs (Fig. 5E). Furthermore, ChIP assay showed that the SUV39H1 inhibitor reduced the detection of H3K9me3 at the promoter (Fig. 5F–H). Thus, these data suggest that menin recruits SUV39H1 to increase H3K9me3 markers at the promoters of the TFs, thereby repressing their expression.

Because HIP increased FOXO1 phosphorylation (Fig. 2B and C) and reduced expression of FOXO1 (Fig. 2A and C) and menin (Fig. 3A and B), we determined whether HIP acts through activation of AKT, which is a serine/threonine protein kinase that directly phosphorylates and suppresses FOXO1 (35). Treatment of HFPPCs with HIP led to increased AKT phosphorylation as well as FOXO1 phosphorylation (Fig. 6A, lane 2 vs. 1). Consistently, cotreatment of the cells with HIP and LY294002, an inhibitor of PI3K, abolished HIP-induced phosphorylation of FOXO1 (Fig. 6A, lane 4). A similar observation showed the presence of another PI3K inhibitor, wortmannin (Fig. 6B). To further confirm the effect of AKT on HIP-induced phosphorylation of FOXO1, we cotreated the cells with HIP and/or MK2206, an allosteric inhibitor of AKT, and determined phosphorylation of FOXO1. The results showed that HIP treatment increased FOXO1 phosphorylation (Fig. 6C, lane 2), but inhibition of AKT with MK2206 abrogated the HIP-induced AKT phosphorylation and phosphorylation of FOXO1 (Fig. 6C, lane 4). Thus, these results demonstrate that HIP induces AKT phosphorylation and activation and FOXO1 phosphorylation and inactivation, leading to a reduction of menin expression.

The current results are consistent with a model that accounts for these findings (Fig. 6D). HIP promotes expression of the β-cell–specific transcription factors PDX-1, MAFA, and NKX.6.1 through reducing the ability of SUV39H1 to bind to the promoters of these TFs by suppressing the FOXO1/menin pathway. We found that HIP suppresses FOXO1 by increasing PI3K/AKT-mediated phosphorylation of FOXO1, reducing the FOXO1-induced expression of menin and menin-mediated recruitment of SUV39H1 to the promoter of these TFs and thereby promoting expression of the TFs and β-cell differentiation (Fig. 6D).

We demonstrate that HIP promotes differentiation of HFPPCs into insulin-secreting cells, which significantly ameliorate hyperglycemia in streptozotocin-induced diabetic mice. For the first time in our knowledge, we show that HIP stimulates β-cell differentiation through suppressing the FOXO1/menin-mediated increase of H3K9me3 at the promoter of the TFs. Of note, we found that HIP-treated HFPPCs display a significant increase in insulin content and the ability to reverse hyperglycemia in diabetic mice. Thus, these studies substantially implicate that HIP may be used ex vivo to promote differentiation of HFPPCs into functional β-cells for transplantation into human recipients to ameliorate diabetes. The induction of islet neogenesis of human islets ex vivo for transplantation can resolve the issue of the current scarcity in supply of human islets for transplantation to treat diabetes clinically.

Human Reg-encoded peptide HIP stimulates islet neogenesis (14). However, the precise underlying molecular mechanism remains unclear. We found that HIP stimulated phosphorylation of AKT in HFPPCs and that HIP-activated AKT phosphorylates downstream transcription factor FOXO1, thereby increasing phosphorylation of FOXO1 and suppressing the FOXO1 function. Of note, HIP-induced suppression of FOXO1 is accompanied by a reduction in menin expression in both HFPPCs (Fig. 3A and B) and primary β-cells (Supplementary Fig. 3), consistent with the observation that FOXO1 is capable of binding to the promoter of the MEN1 gene and inducing menin expression in pancreatic β-cells (17). Therefore, knockdown of FOXO1 (Fig. 2G) or treatment with FOXO1 inhibitor (Fig. 2H) results in reduced menin expression. Our earlier X-ray crystallographic studies showed that menin interacts with multiple partners and acts as a scaffold to regulate multiple pathways (36). Menin also is involved in regulating β-cell differentiation (37). However, it was previously unclear how menin regulates the process. We uncovered that menin binds to the promoters of proislet TFs, including PDX-1, MAFA, and NKX6.1, and that suppression of menin promotes β-cell differentiation (Fig. 6D). Unlike the situation where the deletion of the MEN1 gene promotes β-cell proliferation in adult mice (17,18), MEN1 siRNA transfection had little effect on the proliferation index in HFPPCs (Supplementary Fig. 4). Although the detailed mechanism remains to be elucidated, menin-controlled signaling circuitry is possibly different between pancreatic progenitor cells and adult β-cells. As such, the Men1 knockdown may enhance proliferation of adult β-cells but not pancreatic progenitor cells. Nevertheless, the current findings suggest a more prominent role for menin in cell differentiation in human fetal pancreas.

Although menin has been reported to interact with SUV39H1 in regulating neuroendocrine cells (38), whether HIP can regulate menin and SUV39H1 is unclear. The current findings indicate that HIP-mediated repression of FOXO1/menin leads to a reduction in recruitment of SUV39H1. Moreover, SUV39H1-mediated H3K9me3 can potently repress gene transcription, and suppression of SUV39H1 by its inhibitor reduces both expression of these proislet TFs and the level of the repressive H3K9me3 marker at the pro-β-cell TFs. These findings suggest that HIP signaling promotes HFPPC differentiation through the FOXO1/menin/H3K9me3 axis. Of note, treatment with chaetocin increased TF expression more than menin knockdown, suggesting the existence of a menin-independent mechanism that also may be partly mediated by SUV39H1. Nevertheless, the current results unravel the previously unappreciated link between the HIP signaling and menin-mediated suppression of the proislet TFs through SUV39H1 and histone methylation (Fig. 6D).

Taken together, the findings suggest that the HIP/FOXO/menin/SUV39H1 axis plays a crucial role in the differentiation of HFPPCs and the maintenance of their function ex vivo and that the HIP-treated HFPPCs are better equipped to produce more insulin and reduce blood glucose levels in diabetic mice. These findings are important because previously, little was known about whether and how HIP regulates differentiation of HFPPCs and whether HIP can be used to enhance the function of HFPPCs in restoring normoglycemia in diabetic mice. These studies unravel the role of the HIP/FOXO/menin/SUV39H1 axis in regulating multiple key factors critical for the differentiation and function of human islets. Given that HFPPCs are able to ameliorate hyperglycemia in diabetic mice after transplantation (7,39,40), the current findings demonstrate that HIP treatment can further enhance HFPPC differentiation, thereby paving the way to enhance the function of HFPPCs for transplantation and improved treatment of the diabetes.

Funding. This work was supported by the National Natural Science Foundation of China (no. 81370914, no. 81670708, no. 81570733, no. 81600597), National Basic Research Program of China (no. 2012CB966402), Strategic Funds for Scientific and Innovative Development of Shenzhen Municipality (no. CXZZ20130329101949981, no. JSGG20130918150446437), and Shenzhen Peacock Plan (no. KQTD20140630100746562, no. KQC201108300039A).

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

Author Contributions. Z.J. analyzed data and prepared figures. Z.J., D.S., Y.T., J.T., W.Z., B.X., and J.W. performed the experiments. S.L., J.L., and J.-Å.G. contributed to the discussion and reviewed the manuscript. X.H. and X.M. designed the experiments and wrote the manuscript. X.M. initiated the project. X.M. 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.