Disrupting the DREAM complex enables proliferation of adult human pancreatic β cells

Resistance to regeneration of insulin-producing pancreatic β cells is a fundamental challenge for type 1 and type 2 diabetes. Recently, small molecule inhibitors of the kinase DYRK1A have proven effective in inducing adult human β cells to proliferate, but their detailed mechanism of action is incompletely understood. We interrogated our human insulinoma and β cell transcriptomic databases seeking to understand why β cells in insulinomas proliferate, while normal β cells do not. This search reveals the DREAM complex as a central regulator of quiescence in human β cells. The DREAM complex consists of a module of transcriptionally repressive proteins that assemble in response to DYRK1A kinase activity, thereby inducing and maintaining cellular quiescence. In the absence of DYRK1A, DREAM subunits reassemble into the pro-proliferative MMB complex. Here, we demonstrate that small molecule DYRK1A inhibitors induce human β cells to replicate by converting the repressive DREAM complex to its pro-proliferative MMB conformation.

Proximity Ligation Assay. Dispersed human islet cells were seeded in eight-well chamber slides with harmine, DMSO or virus transduction as described above. The cells were treated with harmine or adenovirus for 72-96 hours before being fixed in 4% paraformaldehyde for 10 min, washed with PBS, and permeabilized for 10 min using 0.5% Triton-X100. The proximity of p130 to LIN52, or HA (HA-p130) to V5 (V5-WTor Mut-LIN52) was assessed using the Duolink PLA In Situ Orange Starter Kit (Mouse/Rabbit) (Sigma-Aldrich, St. Louis, MO, USA) according to the manufacturer's protocol. Blocking solution provided with the kit was added and slides were incubated in a humidified chamber at 37°C for 60 min. Slides were incubated with polyclonal rabbit LIN52 antibody (1:100) and monoclonal mouse p130 antibody (1:100) or monoclonal rabbit anti-HA (Santa Cruz Sc-805) (1:500) and monoclonal mouse V5 (Life Technologies) (1:500) overnight at 4 °C. Slides were then washed twice for 5 min in 1x DuoLink Wash Buffer A at room temperature, followed by incubation with Duolink PLA PLUS and MINUS probes diluted 1:5 in Duolink Antibody Diluent at 37 °C for 1 h. Slides were then washed with 5x Duolink Ligation Buffer diluted 1-to-5 in high-purity water, followed by incubation in Ligation Buffer for 60 min at 37 °C. All subsequent steps were performed in the dark. The 5x amplification buffer was diluted 1:5 in high-purity water, and slides were washed as above. DuoLink Polymerase was added to diluted Amplification Buffer (1:80) and slides were incubated at 37 °C for 100 min. Slides were washed twice for 10 min with 1x DuoLink Wash Buffer B. Finally, the cells were immunolabeled for C-peptide or alexa fluor-conjugated anti-glucagon (R&D Systems IC 1249C) as described above. Cover slips were mounted on slides using Duolink In Situ Mounting Medium with DAPI and sealed. Fluorescent images were captured using a Leica SP5 DMI microscope. Figures are representative of three independent experiments. Fluorescence intensity was measured using ImageJ. A minimum for 300 cells was measured for each treatment.
qPCR Methods. RNA was isolated and quantitative RT-PCR was performed as described previously (8). Gene expression in dispersed islets was analyzed by real-time PCR performed on an QuantStudio System. Primers are listed in Suppl. Table 5.
qPCR-ChIP Methods. p130 ChIP was performed using the EZ-ChIP Kit (Millipore) according to manufacturer's protocol. Experimental repeats (N) are indicated for each primer set in Fig. 5. Human cadaveric islets (500-700 IEQs, or 500,000-700,000 cells) were used per experiment for each p130 immunoprecipitation {#SC-317 (20) Santa Cruz}. Briefly, chromatin preparations were prepared using according to the EZ-ChIP Kit protocol. Primer sets were designed in the putative promoter regions of all genes tested. GAPDHS and MYBL2 primer sets have previously been published (39). Immunoprecipitated DNA was quantified using QuantStudio 5 real-time quantitative PCR detection system (Applied Biosystems). Data are presented as fold-enrichment of ChIP signal over the IgG signal. GAPDHS was used as a negative control. Primers used were: Bioinformatics and Transcription Factor Enrichment Analysis. The Bisque4 insulinoma module genes and genes co-correlated with the module eigenvector of Bisque4 (i.e. genes reflecting module membership) at P<0.01 (total 253 genes, Suppl. Table 1 and Wang et al (46) were queried in the iRegulon app (version 1.3)(50) within Cytoscape (version 3.7.2)(63). The iRegulon options used included: Motif collection (10K 9713 PWMs); track collection (750K ChIP-seq tracks-ENCODE uniform signals) and putative regulatory region of 500bp centered around transcription start site. The enrichment score threshold was 3.0 and the maximum FDR on motif similarity was 0.001. Predicted direct target genes governed by TP53, DREAM, MMP-FOXM1 and RB-E2F were curated from Fischer et al (39). Fisher's Exact Test was used to assess significance of the geneset enrichment with a Benjamini-Hochberg multiple test correction.
Proteomics and Mass Spectrometry. Mass spectrometry samples were prepared for relative quantification as described previously (64). TMTpro reagents were used for quantification (65,66). We fractionated the pooled, labeled peptide sample using BPRP HPLC (67) and an Agilent 1200 pump equipped with a degasser and a UV detector (set at 220 and 280 nm wavelength) into 96 fractions which were concatenated into 24 super-fractions.
Database searching included all entries from the human UniProt Database (downloaded: August 2019) with a reversed database. Searches were performed using a 50-ppm precursor ion tolerance for total protein level profiling. The product ion tolerance was set to 0.9 Da. TMTpro labels on lysine residues and peptide N-termini +304.207 Da), as well as carbamidomethylation of cysteine residues (+57.021 Da) were set as static modifications, while oxidation of methionine residues (+15.995 Da) was set as a variable modification. Peptidespectrum matches (PSMs) were adjusted to a 1% false discovery rate (FDR) (68,69). PSM filtering was performed using a linear discriminant analysis, as described previously (70) and then assembled further to a final proteinlevel FDR of 1% (68). Proteins were quantified by summing reporter ion counts across all matching PSMs, also as described previously (71). Reporter ion intensities were adjusted to correct for the isotopic impurities of the different TMTpro reagents according to manufacturer specifications. The signal-to-noise (S/N) measurements of peptides assigned to each protein were summed and these values were normalized so that the sum of the signal for all proteins in each channel was equivalent to account for equal protein loading. Finally, each protein abundance measurement was scaled, such that the summed signal-to-noise for that protein across all channels equals 100, thereby generating a relative abundance (RA) measurement.   (45), were genes involved in cell cycle progression (e.g., E2F1, CDK1) (46). C. Since the Hub and Bisque4 modules were enriched for genes involved in later stages of the cell cycle, we expanded our search for "earlier" genes, assembling a Bisque4 module membership (Bisque4 expanded ) group of predicted upstream genes, predicted to be important in driving insulinoma cell cycle progression (46). This group was further explored for upstream driver transcription factors using iRegulon, as described in the text and Fig. 2. Note that Panels A and B were previously published in reference 46, and are provided as background. Fig. 2C showing a network representing the predicted direct target genes governed by TFDP1, E2F4, FOXM1, and MYBL2 as determined by the iREGULON analysis. Nodes are colored according to the predicted transcriptional regulator to which they are connected in the network (yellow=FOXM1; cyan=MYBL2; green=TFDP1; orange=E2F4). Genes (nodes) which have edges/connections coming from multiple transcription factors are colored grey. See text and Fig. 2 Legend for details.  Figure 5. Confirmation of canonical DREAM members in human islets in by immunoblot. A. Immunoblotting was performed on three different sets of human islet and probed with antibodies directed against E2F4, RBBP4, p130, DP1 and GAPDH as a loading control. In each case, an adenovirus expressing a corresponding shRNA was used to confirm specificity of the antibody. These results confirm and extend immunohistochemical and RNAseq findings in Fig. 3, and Tables 1 and 2. B. MYBL2, a target of the repressive DREAM complex, is undetectable at the immunohistochemical level, as predicted by the DREAM model (Fig. 2B) and RNAseq in Tables 1 and 2. To confirm that the antibody used is able to detect MYBL2, we overexpressed MYBL2 by adenovirus in human islets and assessed these islets for MYBL2 expression in beta cells as shown in the far right panel. Clear MYBL2 immunolabeling is apparent. Each experiment shown is representative of experiments in three different human islet preparations. Figure 6. Silencing E2F4 + E2F5 or all three Rb family augments beta cell proliferation, and DREAM target gene expression but not the DREAM members or the CDK4/6-Cyclin D family. A. Simultaneous silencing of E2F4 and E2F5 or pRb, p107 and p130 induces human beta cell Ki67 immunolabeling. B-D. qPCR analysis of human islets in which E2F4 and E2F5 were simultaneously silenced. Silencing these two DREAM members had no effect on other DREAM members or CDK4, CDK6 or D-cyclins, but led to almost uniform upregulation of canonical DREAM target genes. E-G. qPCR analysis of human islets in which pRb, p107 and p130 were simultaneously silenced. Again, silencing of Rb family members had no effect on DREAM members or CDK4, CDK6 or D-cyclins, but led to almost uniform upregulation of canonical DREAM target genes. Compare to G.