β-Cell pre-mir-21 induces dysfunction and loss of cellular identity by targeting transforming growth factor beta 2 (Tgfb2) and Smad family member 2 (Smad2) mRNAs

Objective β-cell microRNA-21 (miR-21) is increased by islet inflammatory stress but it decreases glucose-stimulated insulin secretion (GSIS). Thus, we sought to define the effects of miR-21 on β-cell function using in vitro and in vivo systems. Methods We developed a tetracycline-on system of pre-miR-21 induction in clonal β-cells and human islets, along with transgenic zebrafish and mouse models of β-cell-specific pre-miR-21 overexpression. Results β-cell miR-21 induction markedly reduced GSIS and led to reductions in transcription factors associated with β-cell identity and increased markers of dedifferentiation, which led us to hypothesize that miR-21 induces β-cell dysfunction by loss of cell identity. In silico analysis identified transforming growth factor-beta 2 (Tgfb2) and Smad family member 2 (Smad2) mRNAs as predicted miR-21 targets associated with the maintenance of β-cell identity. Tgfb2 and Smad2 were confirmed as direct miR-21 targets through RT-PCR, immunoblot, pulldown, and luciferase assays. In vivo zebrafish and mouse models exhibited glucose intolerance, decreased peak GSIS, decreased expression of β-cell identity markers, increased insulin and glucagon co-staining cells, and reduced Tgfb2 and Smad2 expression. Conclusions These findings implicate miR-21-mediated reduction of mRNAs specifying β-cell identity as a contributor to β-cell dysfunction by the loss of cellular differentiation.


INTRODUCTION
With a prevalence of 30.2 million people in the US alone, diabetes poses a tremendous domestic and international health burden [1]. A commonality between both type 1 diabetes (T1D) and type 2 diabetes (T2D) is reduced functional b-cell mass; either in association with autoimmune b-cell destruction (T1D) or with prolonged exposure to insulin resistance, systemic elevations in proinflammatory cytokines, and saturated free fatty acids (T2D) [2]. In both T1D and T2D, b-cells may exhibit maladaptive signaling responses to inflammatory stress, potentially exacerbating b-cell dysfunction and death or accelerating b-cell autoimmune destruction [3]. An improved understanding of these molecular signaling pathways may pave the way for novel therapies targeting b-cell dysfunction before or after diabetes development.
MicroRNAs (miRNAs) are small RNA molecules that classically repress translation through either direct inhibition or mRNA destabilization [4]. Islet miRNA expression profiling and analyses have identified multiple b-cell miRNAs as critical regulators of b-cell differentiation, development, death, function, and as mediators of the complex b-cell response to inflammatory stress [4,5]. This work has identified that b-that miR-21-5p mimic transfection increased b-cell death by inhibition of the pro-survival mRNA B cell lymphoma 2 (Bcl2), despite decreased PDCD4 [6]. Using RNA duplexes or mimic transfection, several groups have also shown a negative effect of miR-21-5p overexpression on GSIS [6,7] and in vitro inhibition, using a miR-21 inhibitor improved insulin release from cytokine-treated MIN6 cells [7]. However, more comprehensive studies identifying mechanisms of pre-miR-21's effects on b-cell function and the in vivo roles of b-cell miR-21 are required. To bridge this knowledge gap, we developed an in vitro lentiviral model to define the effects of b-cell pre-miR-21 (hereafter referred to as miR-21) induction at levels comparable to those observed in models of islet inflammatory stress [6]. This model demonstrated that miR-21 induction reduced insulin secretion in concert with the expression of key transcription factors associated with b-cell identity. Based on a target prediction analysis, we hypothesized that miR-21 induces b-cell dysfunction by the inhibition of mRNAs critical for b-cell function and identity: transforming growth factor-beta 2 (Tgfb2) and Smad family member 2 (Smad2) mRNAs in the Tgfb2 pathway. Tgfb2 is a member of the TGF-b superfamily of proteins that is involved in diverse roles across different cell types by signaling through a group of transcription factors called Smads [13]. Specifically, Tgfb2 has been shown to play an important role in b-cell identity and function [14]. To further test the roles of b-cell miR-21 in vivo, we developed zebrafish and mouse models of inducible b-cell-specific miR-21 overexpression. Our results implicate miR-21 as a regulator of b-cell identity in part, by direct targeting of Tgfb2 and Smad2 mRNAs.

Islet MiR-21 predicted target analysis
In silico analysis was performed to identify predicted miR-21-5p or -3p targets overlapping with human islet mRNAs downregulated under conditions of inflammatory stress and diabetes (workflow described in detail in Supplementary Figure 1 and identified targets listed in Supplementary File 1) [16e25].

RNA sequencing
Isolated RNA was used to prepare dual-indexed non stranded cDNA libraries using SMART-Seq v4 Ultra Low Input RNA Kit (Clontech) [26]. mRNA sequencing was performed with greater than 20 million reads per sample. Libraries were sequenced with a HiSeq 4000 system (Illumina).
2.6. MiR-21 induction in zebrafish All animal experiments were carried out in accordance with the National Institutes of Health guide for the care and use of laboratory Original Article animals. To generate Tg(hs:CS-bmiR-21) zebrafish, a zebrafish pre-miR-21 amplicon was put in place of the H2B-GFP coding sequence contained in the transgenesis vector used to make the Tg(hs:CSH) transgenic line. This was generated using high-fidelity PCR, followed by subcloning to a site downstream of the lox-mCherry-STOP-lox cassette. Tg(hs:CS-bmiR-21) fish were intercrossed with Tg(ins:-Cre) s924 fish [30] to generate fish exhibiting heat-shock inducible miR-21 overexpression, specifically within b-cells. Embryos were heatshocked for 10 min at 39 C. RNA from 15 embryos/clutch and 20 islets/clutch was used for PCR analysis. Glucose colorimetric assays (Bio Vision #K686) were performed using 20 embryos/clutch. Zebrafish embryos were fixed with 3% formaldehyde in PEM buffer at 4 C overnight and deyolked for immunostaining. Glucose colorimetric assays (BioVision #K686) were performed using 20 embryos/clutch.

Immunofluorescence
Zebrafish embryos and mouse pancreata were fixed and immunostained as described [30]. Primary antibodies (Supplementary Table 2) were detected with 1:500 dilutions of Alexa-conjugated secondary antibodies (Jackson Immu-noResearch). Confocal imaging was performed using a Zeiss LSM700 microscope and quantified by measuring pixel density per insulinpositive cell (Fiji software). To measure nuclear vs. cytoplasmic intensity of markers in cells, a DAPI signal was used as a mask to quantify only pixel density within the nucleus. To quantify Insulin þ glucagon þ polyhormonal cells, all visible individual islet cells that exhibited both insulin staining and glucagon staining were counted.

Human islet transduction
Human islets were obtained from the IIDP [6]. Dispersed cells from 300 islets were transduced with 50 ml of concentrated lentivirus as above, and then treated with 10 mg/ml of doxycycline for 48 h followed by a 24 h recovery period.

Statistical analysis
Statistical analyses were performed using GraphPad Prism Version 7.1 (GraphPad software). Data are presented as means AE standard error of the mean (SEM). Student's t-tests or KolmogoroveSmirnov tests were used for comparison between the experimental and control groups as indicated. One-way ANOVA with Tukey's post-test for multiple comparisons was used when comparing >2 groups. A p-value of 0.05 was considered significant 2.11. Data and resource availability The datasets generated and/or analyzed during the present study are included in the published article (and its online supplementary files).

RESULTS
3.1. Induction of miR-21 in INS1 b-cells leads to b-cell dysfunction and loss of identity Because miRNA mimic transfection leads to supraphysiologic increases in miRNAs [34], we generated a tetracycline-on, doxycycline dependent system of lentiviral pre-miR-21 (hereafter referred to as miR-21) induction to define the effect of smaller fold increases in b-cell miR-21 ( Figure 1A). This system allowed for pre-miR-21 induction at a relative expression more comparable to that of b-cells or islets treated with proinflammatory cytokines (Figure 1BeC) [6]. Compared to INS1scramble cells, INS1-miR-21 cells exhibited a pronounced reduction in insulin secretion at baseline and a response to high glucose ( Figure 1D). Quantification of cytoplasmic Rab37 as a marker of secretory granules demonstrated reduced staining in INS1-miR-21 cells compared to scramble controls ( Figure 1E). Staining and quantification of proinsulin to insulin express expression showed an increase in the ratio of immature proinsulin to mature insulin in INS1-miR-21 cells as compared to INS1-scramble control cells ( Figure 1F).
Recent data have identified b-cell dedifferentiation, loss of identity, or reversion to a progenitor-like state, as a compensatory response to islet inflammatory stress, with evidence of b-cell dedifferentiation in models of T1D and T2D [35]. To test whether our observed phenotype could be associated with this phenomenon, RT-PCR was performed to validate changes in gene expression associated with loss of b-cell identity in vitro. We first assessed transcription factors classically associated with b-cell identity and function ( Figure 2A). Here, miR-21 induction decreased mRNA expression of MAF BZIP transcription factor A (Mafa), NK6 homeobox 1 (Nkx6.1), both insulin genes (Ins1 and Ins2), neuronal differentiation 1 (Neurod1), and solute carrier family 2 member 2 (Glut2). We also performed western blots to test changes in protein expression with miR-21 induction. Here, we measured a trend toward decreased protein expression level of Pdx1 and a significant decrease in MafA ( Figure 2CeE). No significant decrease in protein levels of Glut2 was measured, as detected by western blot analysis (Supplementary Figure 4A). Immunostaining was also performed to measure decreases in protein levels of transcription factors associated with b-cell identity and function, which demonstrated a trend toward decreased nuclear (relative to cytoplasmic) levels of Nkx6.1 ( Figure 2F). No decrease in urocortin staining was measured (Supplementary Figure 4B). We next assessed markers associated with dedifferentiation and b-cell progenitor markers. Consistent with a shift towards a more progenitorlike state, we observed increased neurogenin 3 (Ngn3), Nanog homeobox (Nanog), L-myc 1 proto-oncogene (L-Myc), and aldehyde dehydrogenase 1a3 (Aldh1a3) expression after miR-21 induction  To further probe molecular pathways and identify potential direct mRNA targets impacted by increased islet miR-21 during diabetes development, we performed an analysis of predicted mRNA targets of miR-21-5p and 3p using target prediction software, and overlapped these results with mRNAs reduced in publicly available sequencing datasets from human islets treated with cytokines or with T2D (Supplementary Figure 1). Notably, several members of the transforming growth factor-beta 2 (Tgfb2) pathway, including Tgfb2 and Smad2, were identified as potential direct targets (Supplementary Figure 1B). Because this pathway has been implicated in the Original Article 4 regulation of b-cell identity and commitment [14] and also because miR-21 is predicted to directly target several genes within the pathway, we chose to focus on Tgfb2 and Smad2 as potential direct mRNA targets that could contribute to observed effects of miR-21 induction on b-cell identity in vitro. Consistent with a negative effect of miR-21 on these mRNAs, RT-PCR analysis demonstrated a significant decrease in both Tgfb2 and Smad2 transcripts after miR-21 induction ( Figure 3A). Immunoblot analysis also demonstrated decreased protein expression of both Tgfb2 and Smad2 after miR-21 induction (Figures 3B,3C). To test whether Tgfb2 and Smad2 mRNAs directly bind to miR-21, we performed a streptavidin bead-based pulldown after transfection with biotinylated miR-21-5p, miR-21-3p, or cel-miR-67 duplex control construct. Compared to control pull-downs, Tgfb2 and Smad2 mRNAs were significantly enriched within the biotinylated miR-21 pulldown, suggesting direct binding to miR-21-5p and -3p, respectively ( Figure 3D). To determine whether miR-21 leads to functional inhibition of Tgfb2 and Smad2 translation, we also performed luciferase reporter assays (Figure 3EeF). Here, consistent with functional inhibition, miR-21 overexpression reduced luciferase activity in constructs containing wild type 3 0 UTRs for both Tgfb2 and Smad2. By contrast, miR-21 had no effect on Tgfb2 and Smad2 3' UTRs with mutated predicted binding sites.
3.3. Inhibition of miR-21 or overexpression of Tgfb2 can partially abrogate reductions in mRNAs linked to b-cell identity Islet inflammatory stress increases islet miR-21 expression and is associated with the altered b-cell identity [38]. To test whether miR-21 inhibition can block the effect of inflammatory cytokines on altered b-  Figure 5C). Furthermore, reduced numbers of insulin þ b-cells were observed in Tg(HS:bmiR-21) islets ( Figure 5D). Consistent with the loss of differentiation in association with miR-21 overexpression, Tg(HS:bmiR-21) islets also exhibited increased numbers of insulin þ glucagon þ co-staining cells, marked by white arrows ( Figure 5D).
Although immunostaining for genes associated with mature b-cell identity was limited by available antibodies exhibiting specific cross-reactivity with zebrafish antigens, we identified depletion of Nkx6.1, a key transcription factor in the maintenance of b-cell function and maturation, in insulin þ cells from Tg(HS:bmiR-21) islets ( Figure 5E) [39]. Consistent with our data in INS1-miR-21 cells, Tgfb2 immunostaining was decreased in Tg(HS:bmiR-21) islets ( Figure 5F). RT-PCR also showed reductions in MafA and Pdx1 mRNAs in Tg(HS:bmiR-21) islets ( Figure 5G). Although Smad2 antibodies were not available for immunofluorescence in zebrafish, both Tgfb2 and Smad2 mRNA levels were decreased in Tg(HS:bmiR-21) islets ( Figure 5H).

b-cell MiR-21 induction in a mouse model leads to glucose intolerance and a phenotype consistent with loss of b-cell identity, in association with reduced Tgfb2 and Smad2 expression
To define the effects of b-cell miR-21 induction on glucose homeostasis in a mammalian system, we generated tamoxifen-inducible bcell-specific transgenic (Tg(bmiR-21)) mice (Figure 6AeB). Compared to tamoxifen-treated littermate controls, Tg(bmiR-21) mice exhibited mild glucose intolerance on IPGTTs, without significant differences in insulin tolerance (Figure 6CeD). Similar results were verified in Cre þ versus Cre À controls to rule out the effect of Cre on glucose tolerance (Supplementary Figure 6A). No changes in miR-21 expression were observed in other tissues tested (Supplementary Figure 6B). Ex vivo peak insulin secretion was decreased in Tg(bmiR-21) islets ( Figure 6E). Insulin positive b-cell mass was also decreased in Tg(bmiR-21) mice ( Figure 6F). Next, we assessed endpoints that could point to changes in b-cell identity. Similar to our zebrafish model, altered islet architecture with increased insulin þ glucagon þ costaining cells was also observed in the Tg(bmiR-21) mice (white arrows) ( Figure 6G). Additionally, Tg(bmiR-21) mice demonstrated an increase in glucagon þ cell area and a decrease in insulinþ cell area ( Figure 6G). Consistent with phenotypes observed in Tg(HS:bmiR-21) zebrafish, RT-PCR analysis of Tg(bmiR-21) mouse islets showed decreased expression of Tgfb2, Smad2, MafA, and Pdx1( Figure 7A). Furthermore, immunostaining of Tg(bmiR-21) mouse islets concurrently showed However, even with the inclusion of this donor's data, consistent with our model systems in vitro and in vivo, miR-21 induction resulted in significantly increased Aldh1a3 mRNA expression ( Figure 8B) and significant reductions in Pdx1, along with Tgfb2 and Smad2 mRNA expression ( Figure 8C).

DISCUSSION
Prior studies have linked loss of b-cell identity and dedifferentiation to b-cell dysfunction in models of insulin resistance and reduced islet mass [38,40]. Recent data have also identified this reversion to a progenitor-like state, as a compensatory response to islet inflammatory stress, with evidence of b-cell dedifferentiation in models of T1D [35]. These studies have collectively characterized this phenomenon . Additionally, reduced prohormone processing enzyme expression has also been described in models of islet dedifferentiation, a phenotype observed in our models, with an increase in proinsulin relative to insulin staining, and decreased expression of processing enzymes [44]. A recent study identified gene signature profile changes in murine embryonic b-cells and adult b-cells following STZ treatment to identify b-cell dedifferentiation and dysfunction markers using single-cell RNA-seq [45]. Interestingly, an overlap analysis between the RNA-seq dataset generated by our miR-21 inducible INS1 cell line and the Sachs et al. dataset displayed an overlap of several statistically significant (FDR<0.5) genes (58 upregulated and 165 downregulated) (included in Supplementary File 4). Differences in this phenotype between studies likely reflect differences in terminology, differences in models used, and importantly, the likely existence of this phenomenon on a spectrumdwith heterogeneous effects between disease states, individuals, or even between beta cells within islets.
As features defining loss of b-cell identity continue to be elucidated, the determination of underlying molecular mechanisms contributing to these effects is needed. Here, we identify a novel relationship linking cytokine-induced increases in b-cell miR-21 to reduced expression of mRNAs specifying b-cell identity and b-cell function. The sequence of miR-21 is highly conserved across multiple species including rats, mice, zebrafish, and humans (Supplementary Figure 9). This allowed for the usage of multiple model systems in vitro and in vivo to validate a conserved role of increased b-cell miR-21 in loss of b-cell identity, suggesting that this pathway could be an important physiologic response to islet inflammation.
Our results showed that miR-21 exerts effects on b-cell identity in part through direct targeting of mRNAs in the Tgfb2 pathway. This signaling pathway has also been implicated in the development of the endocrine pancreas [13], b-cell development, and postnatal b-cell identity and function [14]. Tgfb2 overexpression in rat islets in vitro increases insulin secretion [46]. Our data also suggest that increasing Tgfb2 and Smad2 partially abrogated miR-21's effects on mRNAs critical for b- cell function and identity. Prior work has shown that inhibition of Tgfb2 and Smad2 was associated with islet dedifferentiation [47]. By contrast, combined pharmacological inhibition of human b-cell DYRK1A and the TGF-b superfamily did not lead to a dedifferentiated phenotype [48]. These differences could potentially result from combined treatment, off-target effects of pharmacologic inhibitors, or the impact of miR-21 on multiple mRNAs. Both zebrafish and mouse models of islet miR-21 induction displayed reduced expression of transcription factors specifying b-cell identity and in insulin þ cells, with increases in double positive insulin þ and glucagon þ islet cells, and hyperglycemia or glucose intolerance. However, there was a more drastic increase in double positive insulin þ and glucagon þ islet cells in the zebrafish compared to our mouse model. Differences in islet findings could be related to several differences in the model systems. The degree of b-cell miR-21 induction in zebrafish was higher than that observed in mice. Additionally, although we designed both systems to achieve post conception inducible miR-21 expression, because of the nature of our zebrafish model, miR-21 induction occurred 3-days post fertilization vs. 8-weeks after birth in the mouse model, which could impact effects on b-cell fate.
Notwithstanding these differences, the overall similarities between our findings across model systems support the idea of miR-21 as a conserved modulator of b-cell identity.
The use of miRNA mimics can saturate RISC complexes and displace other endogenous miRNAs, causing disproportionately increased binding with lower affinity targets that may not be as dramatically  were seen in insulinþ cells in Tg(bmiR-21) islets. (I) Immunofluorescence quantification is displayed as mean AE SEM for littermate controls and Tg(bmiR-21) mice, respectively. Immunofluorescence intensity was quantified for 3 islets per mouse for n ¼ 3e4 mice per group; *p < 0.05.
Original Article 10 impacted by lower level increases in the miRNA of interest [34]. Furthermore, mimic transfection yields overexpression of the predicted sense strand (5p strand) of the miRNA, while in vivo induction of pre-miRNA transcripts could lead to differential effects owing to activities of the antisense strand of the miRNA duplex [34]. In this study, to address this issue, we designed a lentiviral system of pre-miR-21 induction to model increases in pre-miR-21 on the scale of those observed in models of islet inflammation and diabetes [18]. A limitation of our study is the variability in observed impacts of miR-21 and TGFB/Smad signaling within and across our different systems, especially in human islet studies, where islets from one donor did not show an effect of miR-21 on b-cell identity, consistent with heterogeneity in human disease. Although inter-species differences in effects may exist, we consider the inclusion of multiple model systems a strength, and the fact that significant changes in features associated with b-cell identity are present across each of these systems is suggestive of a conserved physiologic response of the b-cell to increases in miR-21. Our GSIS in INS1 cells was not normalized for differences in cell death, which we previously observed on a larger scale in experiments performed using miR-21 mimics. However, lentiviral induction of more physiologic increases in miR-21 appeared to have a differential effect on b-cell function; consistent with this, mimic experiments associated with larger increases in beta cell apoptosis demonstrated higher insulin secretion at baseline for miR-21-5p mimic-transfected cells, with no increase in insulin secretion following high glucose treatment [6]. In contrast, in the present study, baseline and stimulated GSIS were both reduced in concert.
In conclusion, these studies have defined a new mechanism that links increases in b-cell miR-21 to b-cell dysfunction during diabetes development. Furthermore, our study has identified a novel upstream molecular modulator of b-cell identity, and a mechanistic pathway initiating b-cell dedifferentiation in the context of islet inflammatory stress. The use of several model systems and human islets ensure that these results are robust and relevant to human diabetes. Future studies should test the potential for therapeutic targeting of islet miR-21 and its molecular signaling pathways to preserve functional b-cell mass in diabetes.

ACKNOWLEDGMENTS
Emily Sims is the guarantor of this study and had full access to all the data and takes responsibility for the integrity of the data and the accuracy of the data analysis.