MLL3 and MLL4 Methyltransferases Bind to the MAFA and MAFB Transcription Factors to Regulate Islet β-Cell Function

Diabetes mellitus is a disease that affects the body’s ability to maintain euglycemia, with type 1 characterized by a loss of insulin-producing islet β-cells and type 2 (T2DM) by peripheral insulin resistance and β-cell dysfunction. One proposed treatment for type 1 diabetes is to replace diseased β-cells with those generated from human embryonic stem cells (hESCs) or induced pluripotent stem cells (1). The principal limitation in producing functional β-cells has been directing the final postnatal maturation steps (2), which involves expression of proteins required for glucose sensitivity and insulin secretion (3).

Islet-enriched transcription factors are essential for embryonic formation and postnatal function of β-cells (4–6). For example, very early exocrine and endocrine pancreatic development is driven by Pdx-1 starting at embryonic day 8.5 (e8.5) in mice, with mice and humans both lacking a functional copy suffering from pancreatic agenesis (4,5). In contrast, Ngn3 expressed from e9.5 is only required in the formation of endocrine cell types (i.e., β, α [hormone glucagon producing], δ [somatostatin], ε [ghrelin], and pancreatic polypeptide) (6). MafA is expressed even later during development and only in β-cells (i.e., e13.5), contributing in postnatal maturation steps (7). Interestingly, the induction of glucose-sensitive insulin secretion in vivo from transplanted hESC-derived endocrine progenitors correlates with MAFA expression (8). Moreover, the production of only Pdx-1, Ngn3, and MafA is sufficient to reprogram mouse exocrine, intestinal, and liver cells into insulin⁺ β-like cells in vivo (9–11). Although these examples clearly illustrate the fundamental importance of islet-enriched activators to β-cells, the transcriptional mechanisms involved are not well defined.

Transcription factors primarily regulate gene activation by recruitment of coregulators, which often influence expression by directly binding to the basal transcriptional machinery and/or through epigenetic remodeling of the chromatin structure. These coregulators can have positive (coactivator) and negative (corepressor) actions on target gene transcription (12), thus conferring a second level of specificity to the transcriptional response. Coregulator recruitment is, in turn, controlled by the spatial and temporal expression patterns and posttranslational modifications of the transcription factor and/or coregulator. Unfortunately, little is known about the coregulators recruited by islet-enriched transcription factors. Although there are hundreds of known coregulators (http://www.nursa.org/), such knowledge is limited to candidate studies linking, for example, Pdx-1 to p300 (13), Set7/9 (14), HDAC1/2 (15), PCIF1 (16), and Bridge-1 (17). In contrast, MafA has only been linked to p/CAF (18).

In this study, we used an “in cell” reversible cross-link immunoprecipitation (Re-CLIP) and mass spectrometry (MS) approach to isolate coregulators of MafA from mouse β-cells. Notably, all nine subunits of the Mll3 and Mll4 histone 3 lysine 4 (H3K4) methyltransferase complexes were identified in the MafA immunoprecipitates, but none of the unique subunits of the other mammalian Mll complexes were detected (e.g., Menin, Mll1, Mll2 of Mll1/2 complexes; Set1A, Set1B, Wdr82 in Set1A/B complexes). (Mll3 and Mll4 will be referred to as Mll3/4 for simplicity.) These methyltransferases were also found to bind MAFB, a closely related transcription factor essential to mouse β-cell development and coexpressed with MAFA in adult human islets (19,20). Notably, mouse islet β-cell function is compromised in a heterozygous null mutant of NCoA6, a key subunit of Mll3/4 (21). Decreased NCOA6 levels reduced expression of a subset of MAFA- and MAFB-regulated genes in human and rodent β-cell lines, with evidence provided that this results from limiting gene transcriptional start site (TSS) H3K4 trimethylation. Interestingly, there is essentially complete overlap of MafA and Mll3/4 transcriptional control in islets isolated after embryonic β-cell–specific removal of NCoA6 or MafA in vivo. These results suggest that MLL3/4 coactivator recruitment by MAFA and MAFB is important to the formation and adult function of islet β-cells.

Mouse βTC-3 cells were maintained in DMEM containing 25 mmol/L glucose, 10% FBS, and 100 units/mL penicillin/streptomycin (22). Human EndoC-βH1 (23) cells were grown in DMEM containing 5.6 mmol/L glucose, 2% BSA, 50 μmol/L 2-mercaptoethanol, 10 mmol/L nicotinamide, 5.5 μg/mL transferrin, 6.7 ng/mL selenite, 100 units/mL penicillin, and 100 units/mL streptomycin. βTC-3 cells were infected with an adenovirus overexpressing GFP alone or a Flag-tagged MafA (provided by Dr. Qiao Zhou, Harvard University) at a multiplicity of infection of 200 under conditions described previously (24). Nuclear extract was prepared 48 h postinfection as described below. Primary antibodies for immunoblotting (25) are listed in Supplementary Table 1. Horseradish peroxidase–conjugated anti-rabbit, anti-mouse, or anti-goat secondary antibody was used at 1:2000 (Promega). Immunoblots were quantitated with ImageJ software (National Institutes of Health).

The Re-CLIP protocol was adapted from Smith et al. (26). Briefly, ∼10⁹ βTC-3 cells were cross-linked using 0.1 mmol/L dithiobis (succinimidyl propionate) (DSP) in PBS for 30 min at 37°C. Cells were then harvested and nuclear extract was prepared by high salt (400 mmol/L NaCl) extraction (25). Magnetic protein G beads (Life Technologies) were covalently cross-linked with MafA antibody (20 μg) and were preincubated with an MafA-blocking peptide (³²KKEPPEAERFC⁴², 100-fold excess antibody) or with PBS alone. Extract and beads were incubated for 3 h at 4°C, washed with radioimmunoprecipitation (RIPA) buffer (10 mmol/L Tris, pH = 8.0, 140 mmol/L NaCl, 0.5% Triton X-100, 1% sodium deoxycholate, 0.1% SDS), and eluted in RIPA supplemented with 200 mmol/L dithiothreitol (DTT). The eluted proteins were visualized by PAGE/silver staining, and protein identification was determined by liquid chromatography–tandem MS (LC/MS/MS) or Multidimensional Protein Identification Technology (MudPIT) analysis in the Vanderbilt University Proteomics Core. Immunoprecipitation experiments were performed with the antibodies listed in Supplementary Table 1 at 10 μg each, as described above for Re-CLIP, omitting the DSP step and substituting PBS for RIPA.

Sucrose gradients were performed as described earlier (25). Briefly, βTC-3 nuclear extract (600–1000 μg) was collected by high salt extraction (25) and separated over a 5–35% sucrose gradient (4.5 mL total volume). Fractions (300 μL each, excluding the first 500 μL) were analyzed by immunoblotting, rat insulin II C1 element gel shift (25), and immunoprecipitation analysis. HeLa cells were transfected with expression plasmids encoding human MAFA and/or MAFB (25). Antibodies used for gel shift are listed in Supplementary Table 1.

Knockdown in βTC-3 and EndoC-βH1 cells was achieved using ON-TARGETplus small interfering (si)RNAs of mouse MafA (#J-041353–09), mouse NCoA6 (#J-041129–6), human MAFA (#L-027343–01), human MAFB (#L-009018–00), and human NCOA6 (#L-019107–00). The targeting siRNA (0.5 nmol) or a nontargeting control (#D001810; GE Healthcare Dharmacon) was introduced into βTC-3 cells (4 × 10⁶) using Buffer V (Lonza #VVCA-1003) with an Amaxa Nucleofector 2 (Program G-016; Lonza, Walkersville, MD). EndoC-βH1 cells (2 × 10⁶) were transfected with siRNAs (50 pmol/L) using the Dharmafect #1 reagent (GE Healthcare Dharmacon #T-2001), following the manufacturer’s protocol. Nuclear extract, RNA, and chromatin were collected 72 h after transfection.

RNA was collected from βTC-3 cells using the RNeasy kit (Qiagen). The Trizol reagent (Life Technologies) and the DNA-Free RNA Kit (Zymo Research) were used for EndoC-βH1 cells and mouse islets. The iScript cDNA synthesis kit (Bio-Rad Laboratories, Inc.) was used for cDNA synthesis. The quantitative (q)PCR reactions were performed with the gene primers listed in Supplementary Table 2 on a LightCycler 480 II (Roche), and analyzed by the ΔΔCT method (27). Significance was calculated by comparing the ΔCT values in all but Fig. 7A and B, wherein each mutant was normalized to a control littermate. The Qiagen RT² Profiler PCR array for human diabetes genes (Cat. No. 330231 PAHS-023ZA) was screened with 1 ng cDNA from EndoC-βH1 cells, following the manufacturer’s guidelines.

Chromatin was prepared from βTC-3 cells and EndoC-βH1 cells by cross-linking for 10 min with 1% formaldehyde, followed by lysis in buffer containing 1% NP-40. The centrifuged pelleted nuclei were lysed in SDS buffer (50 mmol/L Tris, 10 mmol/L EDTA, 1% SDS), and chromatin was sonicated using a Diagenode Bioruptor. DNA (10 μg) was incubated with 5 μg of antibody or species-matched IgG, and chromatin immunoprecipitation (ChIP) was performed as described previously (28). Antibodies are listed in Supplementary Table 1, and the qPCR primers used for the analysis are listed in Supplementary Table 2.

Floxed (fl) MafA (19) or NCoA6 (29) mice were crossed with rat insulin II promoter-driven Cre (RIP-Cre) (30) transgenic mice to delete MafA (termed MafA^Δβ) or NCoA6 (NCoA6^Δβ) specifically in β-cells. All mice were maintained on a mixed background (C57BL/6J, I129, Balb/c, 129S6/SvEvTac, FVB/N), although MafA^Δβ mice were mostly C57BL/6J. MafA^Δβ, NCoA6^Δβ, MafA^fl/fl and/or NCoA6^fl/fl mice were fasted for 6 h in the morning before the intraperitoneal glucose tolerance test. There was no change in total body weight in MafA^Δβ or NCoA6^Δβ mice (Supplementary Fig. 2A). Blood glucose levels were measured before (time 0) and then 15, 30, 60, and 120 min after an intraperitoneal injection of glucose (2 mg/g body weight) prepared in sterile PBS (20% w/v). Islets from 8-week-old mice were isolated using collagenase P (Roche) digestion and hand picking (31). The Vanderbilt University Institutional Care and Use Committee approved all of these studies.

Pancreata were dissected and fixed in 4% paraformaldehyde in PBS for 4 h, followed by paraffin embedding (32). The entire e15.5 pancreas was serially sectioned (6 µm per section) and every 10th section analyzed by immunofluorescence. Three sections at least 120 μm apart were examined at 8 weeks for islet architecture and for Slc30a8 and Slc2a2 expression. The primary antibody list is provided in Supplementary Table 1. Species-matched secondary antibodies were used for immune detection at 1:1000 (Jackson ImmunoResearch Laboratories). Slides were mounted with Dapi Fluoromount-G (SouthernBiotech #0100-20) and images acquired on a Zeiss Axio Imager M2 widefield microscope with Apotome.

EndoC-βH1 cells (2 × 10⁶) were placed in medium containing low glucose (1.1 mmol/L) for 12–14 h. Low or high glucose (15.5 mmol/L) medium was then added for 1 h. Medium and cell lysates (lysis buffer: 20 mmol/L Tris, 1% Triton X-100, 10% glycerol, 137 mmol/L NaCl, 2 mmol/L EGTA, 1X protease inhibitor tablet [Roche Diagnostics]) were collected and analyzed for insulin content by the Vanderbilt Hormone Assay Core. Glucose-stimulated insulin secretion was performed with 8-week-old islets in the Vanderbilt Islet Procurement and Analysis Core. Briefly, islets were size matched, incubated overnight in 5.6 mmol/L glucose and then treated with 5.6 mmol/L or 16.7 mmol/L glucose for 1 h. The insulin content in the medium and cell lysate was analyzed. Insulin secretion capacity was calculated as media insulin over cell lysate insulin content. Glucose-stimulated levels are presented as the 15.5 mmol/L and 1.1 mmol/L values for EndoC-βH1 cells and 16.7 mmol/L glucose and 5.6 mmol/L glucose for isolated islets.

Mean differences were tested for significance using a Student two-tailed t test. The minimal level of statistical significance is listed in the figure legend.

MafA interacting proteins were isolated from mouse βTC-3 cells using a combined Re-CLIP/MS strategy (Fig. 1A) (26). The Re-CLIP procedure was selected over other isolation methods (e.g., tandem affinity purification [33]), because the dimerization, DNA binding, and activation capacity of MafA is regulated by many phosphorylation events (25). Amine-reactive DSP was used to covalently cross-link MafA to neighboring proteins, and MafA antibody was used to precipitate bound proteins. The signal-to-noise of the procedure was improved by using the MafA antibody blocked by MafA peptide in the control immunoprecipitation (as opposed to IgG) and by washing precipitated proteins with RIPA buffer. Bound proteins were eluted from MafA by the addition of DTT to cleave the disulfide bond in the cross-linker. As expected, Pdx-1 (22) and p/CAF (18), known MafA-interacting proteins, were bound to MafA after Re-CLIP (Fig. 1B). Moreover, many proteins were selectively bound to MafA in the immunoprecipitation conducted in the absence of blocking peptide (Fig. 1C).

Numerous MafA bound proteins were detected upon MS analysis of the Re-CLIP products, with their activities consistent with a role in mediating MafA activity (Supplementary Table 3). It is notable that no other islet-enriched transcription factors or the p/CAF coactivator was identified by MS, presumably reflecting their relatively low abundance (e.g., Pdx-1, p/CAF) or inability to interact. This indicates that the results in Supplementary Table 3 represent only a portion of MafA-interacting proteins.

Our analysis focused on determining the significance of Mll3/4 coactivator binding to MafA because the evolutionary conserved COMPASS family of methyltransferases is strongly associated with gene activation due to their ability to mono-, di-, and trimethylate H3K4 within the enhancer and/or promoter region (34). MS identified all nine proteins of the ∼1.5 MDa Mll4 complex (Fig. 1D), and Mll3 was additionally detected by the more sensitive MudPIT analysis (Supplementary Table 3). However, none of the unique COMPASS subunits of mammalian Mll1/2 (i.e., Mll1, Mll2, menin, Hcfc1/2, Psip1/2) or Set1A/B (i.e., Set1A, Set1B, Cfp1, Wdr82, Bod1, Bod1l) were detected. The ability of MafA to bind to Mll3/4 was also independently demonstrated in βTC-3 cells producing an adenovirus-driven Flag-tagged MafA (Supplementary Fig. 1).

βTC-3 nuclear proteins were separated by size in a 5–35% sucrose gradient to determine if the mobility of the ∼90kDa MafA dimer was affected by interaction with Mll3/4. MafA was found in two distinct gradient size fractions. The faster mobility MafA, containing fractions 6–8, were surrounded by islet-enriched transcription factors of relatively low molecular weight (Fig. 2A) (Pdx-1, 31 kDa, fractions 3–5; Nkx6.1, 38 kDa, fractions 3–5; Hnf1α, 67 kDa, fractions 6–8; Nkx2.2, 30 kDa, fractions 3–5; Pax6, 50 kDa, fractions 6–8), whereas the high–molecular weight fraction 14 comigrated with the Rbbp5 (59 kDa), NCoA6 (220 kDa), Wdr5 (36 kDa), and Utx (154 kDa) subunits of the Mll3/4 complex. In contrast, few other islet-enriched transcription factors were found in this high–molecular weight fraction. Antibody supershift and competitor analysis illustrated that the binding properties of MafA in fractions 7 and 14 were identical (Fig. 2B). In addition, immunoprecipitation experiments conducted with antibodies to NCoA6 or MafA demonstrated that MafA interacted with several Mll3/4 components in fraction 14 (Fig. 2C). These results strongly indicate that a significant fraction of MafA is associated with the Mll3/4 complex in β-cells.

We next analyzed how Mll3/4 affected MafA activity by depleting MafA or NCoA6 from βTC-3 cells by siRNA-directed knockdown. The NCoA6 subunit lacks a DNA binding domain and is involved in recruiting Mll3/4 to a variety of transcription factors and nuclear receptors (35–38). Furthermore, Mll3 and Mll4 have redundant roles in the liver, and NCoA6 is required for activity of both methyltransferase complexes (36). Notably, mouse NCoA6^+/− mutants have impaired glucose clearance (21), whereas overexpression of NCoA6 in rat INS-1 β-cells increases cell function (39). In addition, polymorphisms in NCOA6 are associated with decreased human β-cell function (39). Our objective was to determine how closely Mll3/4 is linked to MafA-mediated transcriptional control.

MafA and NCoA6 protein levels in βTC-3 cells were both effectively (>80%) and specifically depleted by siRNA treatment (Fig. 3A). The effect of these conditions on expression of β-cell genes bound by MafA and downregulated in pancreas-specific deletion MafA^Δpanc mice was determined (40, 41) (Supplementary Table 4). As expected, the expression of most of these genes was decreased upon depletion of MafA (Fig. 3B; Slc2a2, Ins2, Lifr, Prss53, Atp2a2), but these conditions had no effect on NCoA6 levels. In contrast, only a small subset of these MafA-activated genes was affected in the NCoA6 knockdown (i.e., Slc2a2, Prss53). This outcome was considered likely because a number of other potential MafA coregulators were identified here and previously (e.g., p/CAF (18) (Supplementary Table 1), although it is possible that some fraction of Mll3/4 activity is independent of NCoA6. In addition, ChIP analysis showed that Mll3 and MafA were bound within the same control regions of the Slc2a2 and Prss53 genes (Fig. 3C). Moreover, H3K4 trimethylation and elongating carboxy terminal tail phosphorylated RNA polymerase II levels were significantly reduced within the TSS region of Prss53 upon knockdown of MafA or NCoA6, the regulatory pattern expected for MafA-recruited Mll3/4 activity (Fig. 3D and E).

Our next objective was to determine if human MAFA and closely related MAFB bind MLL3/4. MafB is produced in insulin⁺ cells only during mouse embryogenesis and plays an important role in β-cell formation (19). In contrast, human MAFB is expressed in developing and postnatal β-cells, being coproduced with MAFA in adult islet β-cells (20). These basic leucine-zipper proteins bind to DNA with identical specificity as homo- or heterodimers (32,42), as also observed in MAFA- and MAFB-transfected HeLa cell extracts (Fig. 4A). Gel shift analysis indicates that both MAFA/MAFB and MAFB₂ complexes are also formed in nuclear extracts prepared from the human EndoC-βH1 β-cell line (Fig. 4B), although they were difficult to distinguish due to their similar size (MAFA at ∼45 vs. MAFB at ∼42 kDa). Supershift analysis with MAFA and MAFB antibodies provided further support for these protein associations (Fig. 4A: lanes 3–5; Fig. 4B: lane 3). Significantly, immunoprecipitation experiments demonstrated that MAFA and MAFB were both able to associate with multiple proteins of the MLL3/4 complex in EndoC-βH1 cells (NCOA6, UTX, RBBP5) (Fig. 4C), although whether this results from binding MAFB₂ and/or MAFA/MAFB is unclear. In addition, MAFA and MAFB were able to interact with the islet-enriched PAX6 and ISL1 transcription factors as well as P/CAF; however, these likely represent relatively weak or indirect binding due to their absence from the Re-CLIP/MS analysis performed in mouse βTC-3 cells (Supplementary Table 1).

MAFA, MAFB, and NCOA6 were siRNA depleted from EndoC-βH1 cells to analyze their significance to human β-cell gene expression and function. Protein levels were reduced by ∼50–70% under these conditions (Fig. 5A). The effect on transcriptional control was determined on genes regulated by mouse MafA and a candidate set available on a commercial array composed of 84 genes associated with T2DM onset, development, and progression.

INSULIN, G6PC2, PRSS53, and SLC2A1 were significantly affected by MAFA or MAFB depletion in EndoC-βH1 cells (Fig. 5B). However, their expression was unchanged by NCOA6 depletion, presumably due to differences in human MAFA/MAFB:MLL3/4 control in relation to rodents and/or the β-cell lines used in our analysis. Notably, the levels of several genes linked to T2DM and important to glucose-stimulated insulin secretion were reduced upon NCOA6 depletion (Fig. 5C), specifically, AMP-kinase subunit PRKAA1 (43), glucose-6-phosphate dehydrogenase (44), and angiotensin (AGT) (45). In addition, the expression of each was compromised in MAFB siRNA-treated EndoC-βH1 cells, whereas MAFA depletion only affected AGT. Moreover, MAFA, MAFB, and MLL3 were all corecruited to the same endogenous 5′-flanking control region of the AGT gene (Fig. 5D). Furthermore, glucose-stimulated insulin secretion in EndoC-βH1 cells was compromised upon knockdown of MAFA, MAFB, or NCOA6 (Fig. 5E). Collectively, these data support an important role for MLL3/4 in MAFA and MAFB activity in human β-cells.

To evaluate the influence of MafA:Mll3/4 on islet β-cells in mice, we compared the phenotype of β-cell–specific loss of NCoA6 with that of MafA. Rat insulin II–driven Cre transgenic mice (30) were bred with MafA^fl/fl (19) and NCoA6^fl/fl (46) mice to remove these proteins specifically from insulin⁺ cells in MafA^Δβ and NCoA6^Δβ mice during development (Fig. 6A). Approximately 70% of mutant insulin⁺ cells lacked the MafA or NCoA6 proteins by e15.5 (Fig. 6B), a time at which MafB is still expressed in developing β-cells (47).

Like the pancreas-specific Pdx1-Cre removal of MafA (MafA^Δpanc) (41), MafA^Δβ mice were euglycemic and had impaired islet architecture, where glucagon⁺ α-cells were no longer restricted to the outer mantle of the islet (Fig. 6C). Adult NCoA6^Δβ mice were also euglycemic, but their islet architecture was unchanged. In addition, only MafA^Δβ mice displayed the impaired whole-body glucose clearance observed in MafA^Δpanc mice (41) (Supplementary Fig. 2B–E).

Strikingly, most of the genes regulated by MafA in 8-week-old MafA^Δβ mouse islets were now also significantly compromised in 8-week-old NCoA6^Δβ islets (Fig. 7A). In addition, glucose-stimulated insulin secretion was impaired in islets isolated from both MafA^Δβ and NCoA6^Δβ mice (Fig. 7C), as found upon MAFA, MAFB, or NCOA6 depletion in human EndoC-βH1 cells (Fig. 5D). The basal insulin secretion level of NCoA6^Δβ islets was unusually high in relation to wild-type controls and MafA^Δβ (Fig. 7C), a sign of β-cell immaturity (48). The more penetrant Mll3/4 control properties observed in NCoA6^Δβ mice likely represent the combined actions of Mll3/4 on MafB activity during development and MafA postnatally.

Gene induction involves the recruitment of coregulators by enhancer-bound transcription factors such as MAFA and MAFB. These protein-bound effectors ultimately influence recruitment of the RNA polymerase II transcriptional machinery. We used an unbiased “in cell” chemical cross-linking, immunoprecipitation, and MS strategy to identify coregulators of the MafA transcription factor, an essential regulator of postnatal rodent islet β-cell proliferation and activity. Many distinct coregulator candidates were identified in our biochemical screen, and we focused on the Mll3/4 methyltransferase complex because of the established role of this coactivator in gene transcription. Our studies are consistent with a model where MLL3/4 recruitment by MAFA and MAFB is essential to islet β-cells.

MLL3 and MLL4 are members of the evolutionary-conserved COMPASS family of methyltransferases that are strongly associated with gene activation due to their ability to methylate H3K4 within enhancer and/or promoter regions. Strikingly, all of the proteins in the Mll3/4 complexes were identified in the MafA antibody precipitates by MS, and not the other mammalian Mll complexes (e.g., Mll1/2 complex or Set1A/B complex). MLL3/4 binding to MAFA was also confirmed by immunoprecipitation analysis in mouse and human β-cell lines (Figs. 2 and 4). Moreover, a significant fraction of MafA was bound to the ∼1.5 MDa Mll3/4 complex by sucrose gradient centrifugation (Fig. 2). Interestingly, the islet-enriched Nkx2.2 and Nkx6.1 transcription factors were also found in the same high–molecular weight fraction. Although the identity of these islet-enriched complexes are unknown, a very high–molecular weight complex would be predicted from the association in β-cells of Nkx2.2 with the histone deacetylase-1 corepressor, the Groucho-3 transcription factor repressor, and the DNA (cytosine-5)-methyltransferase 3A corepressor to prevent expression of the α-cell specification factor Arx (49). Likewise, Pax6 comigration with MafA:Mll3/4 could reflect binding to high–molecular weight proteins like p300 (50). However, Nkx2.2, Nkx6.1, and Pax6 were not detected by MafA antibody in Re-CLIP experiments, and no other islet-enriched transcription factors were either (Supplementary Table 3), strongly suggesting that these proteins are not directly, and certainly not abundantly, associated with MafA:Mll3/4. At least some interaction is likely with PAX6 and ISL1 in human β-cells based on the MAFA and MAFB antibody precipitation results in human EndoC-βH1 cells (Fig. 4C). Collectively, these data strongly implied that MafA recruitment of Mll3/4 was highly specific and important to β-cell activity.

Immunoprecipitation experiments conducted with the EndoC-βH1 cell line showed that human MAFA and MAFB are both capable of binding and recruiting the MLL3/4 complexes. However, reducing the protein levels of these transcription factors or the core NCOA6 subunit of the MLL3/4 complexes in human (Fig. 5) and mouse (Fig. 3) β-cell lines disclosed that the coactivator only influenced some MAFA/MAFB activated genes. For example, only the Slc2a2 and Prss53 genes were bound and coregulated by MafA and Mll3/4 among a variety of tested MafA-regulated genes in βTC-3 cells (Fig. 3). Importantly, and as expected, recruitment of Mll3/4 by MafA regulated H3K4 trimethylation levels of the Prss53 gene as well as RNA polymerase II engagement.

In contrast to what was found upon reduction of NCoA6 levels in β-cell lines, nearly all of the islet MafA target genes were dependent on NCoA6 in 8-week-old NCoA6^Δβ islets (Fig. 7A). Because of the early action of Cre-recombinase in developing MafA⁺ and MafB⁺ NCoA6^Δβ β-cells, this more penetrant phenotype is likely due to the activity of Mll3/4 on MafB-bound enhancers during development. Notably, this supposition is consistent with analysis of the MafB^−/− (19) and MafA^Δpanc (41) mutants, which established that many genes first regulated by MafB developmentally were subsequently influenced by MafA after birth (41). In NCoA6^Δβ mice, the lack of Mll3/4 activity on MafA- and MafB-bound enhancers during development may explain why glucose-stimulated insulin secretion was differentially affected in NCoA6^Δβ and MafA^Δβ islets (Fig. 7C). The results also indicate that Mll3/4 is involved in some, but not all, MafA- and MafB-regulated gene expression. Thus, Mll3/4 was necessary for glucose-stimulated insulin secretion in both human and mouse β-cells but did not have the same influential effect on islet cell architecture as MafA (Fig. 7A).

Our understanding of the genes regulating glucose-stimulated insulin secretion in human β-cells is very limited. The results in EndoC-βH1 cells suggest that regulation by MAFA/B:MLL3/4 of the AMP-kinase subunit PRKAA1 (43), glucose-6-phosphate dehydrogenase (44), and angiotensin (45) expression may be important (Fig. 5C). For example, angiotensin appears to have a direct role in regulating β-cell activity (51). However, this candidate list certainly represents only a fraction of the gene products involved in β-cell function, and future efforts will be aimed at identifying more using ChIP-Seq and RNA-Seq. This analysis should also be conducted with human islets, because the transformed and proliferative nature of the EndoC-βH1 cell line could possibly influence how MAFA and MAFB regulate gene expression. It is also important to note how similar the phenotype of the MAFA, MAFB, and NCOA6 regulatory response was in our β-cell line and mouse studies, and yet different from the other islet-enriched transcription factors present in high–molecular weight complexes (Fig. 2). Thus, islet β-cell–specific deletion of Pax6, Nkx2.2, or Nkx6.1 produces a much more severe effect on cell identity and causes lethality soon after birth due to hyperglycemia (52–54).

In conclusion, our combined proteomic, cell line, and mouse strategies have led to the identification and characterization of MLL3/4 as a novel coactivator of the MAFA and MAFB transcription factors in islet β-cells. Future efforts will be aimed at determining how other isolated candidate coregulators affect MAFA and MAFB control, such as CoREST (LSD1, RCOR1, RCOR3, HDAC2) and histone deacetylase 6, both of which are known to play a major regulatory role in other cell contexts (55,56). It will also be of interest to determine if coregulator recruitment is influenced by homodimeric (i.e., MAFA₂, MAFB₂) or heteromeric (MAFA/MAFB) transcription factor composition and the physiological or pathophysiological state of the β-cell.

Acknowledgments. Adenoviral Flag-tagged MafA was kindly provided by Dr. Qiao Zhou, Harvard University. The authors thank Dr. Raphael Scharfmann, INSERM, Paris, France, and Dr. Philippe Ravassard, French National Centre for Scientific Research, Paris, France, for generously providing EndoC-βH1 cells.

Funding. This work was supported by grants from the National Institutes of Health (DK-090570 to R.S., F32-DK-102283-01 and T32-DK-007061 to H.A.C., and CA-112403 and DK-058242 to J.X.) and the Vanderbilt Diabetes Research and Training Center (DK-20593). Imaging was performed with National Institutes of Health support from the Vanderbilt University Medical Center Cell Imaging Shared Resource (CA-68485, DK-20593, DK-58404, HD-15052, DK-59637, and EY-08126) and Vanderbilt University Medical Center Islet Procurement and Analysis Core (DK-20593).

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

Author Contributions. D.W.S. and H.A.C. researched data and wrote the manuscript. L.L. and J.X. provided animals. A.R. provided technical support. S.G. researched data. R.S. wrote the manuscript. R.S. 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.