|Home | About | Journals | Submit | Contact Us | Français|
Cholesterol is a critical component of cell membranes, and cellular cholesterol levels and distribution are tightly regulated in mammals. Recent evidence has revealed a critical role for pancreatic β cell–specific cholesterol homeostasis in insulin secretion as well as in β cell dysfunction in diabetes and the metabolic response to thiazolidinediones (TZDs), which are antidiabetic drugs. The ATP-binding cassette transporter G1 (ABCG1) has been shown to play a role in cholesterol efflux, but its role in β cells is currently unknown. In other cell types, ABCG1 expression is downregulated in diabetes and upregulated by TZDs. Here we have demonstrated an intracellular role for ABCG1 in β cells. Loss of ABCG1 expression impaired insulin secretion both in vivo and in vitro, but it had no effect on cellular cholesterol content or efflux. Subcellular localization studies showed the bulk of ABCG1 protein to be present in insulin granules. Loss of ABCG1 led to altered granule morphology and reduced granule cholesterol levels. Administration of exogenous cholesterol restored granule morphology and cholesterol content and rescued insulin secretion in ABCG1-deficient islets. These findings suggest that ABCG1 acts primarily to regulate subcellular cholesterol distribution in mouse β cells. Furthermore, islet ABCG1 expression was reduced in diabetic mice and restored by TZDs, implicating a role for regulation of islet ABCG1 expression in diabetes pathogenesis and treatment.
Cholesterol is an essential component of cell membranes, and cellular cholesterol homeostasis is a tightly regulated process (1). Membrane cholesterol content and distribution must be maintained at finely tuned levels, and conditions of both cholesterol overload and cholesterol deficiency can result in cellular dysfunction and disease. One hallmark of type 2 diabetes is impaired insulin secretion with progressive pancreatic β cell dysfunction in the face of peripheral insulin resistance (2). Many potential pathways for β cell dysfunction in diabetes have been proposed (3), but the precise mechanisms remain elusive. Associations between dyslipidemia and diabetes have long been recognized (4), but the reasons for this relationship have not been clear. Recent work has demonstrated an important role for β cell–specific cholesterol homeostasis in β cell function (5). Inactivation of the ABC transporter A1 (ABCA1) in β cells led to islet cholesterol accumulation and markedly impaired insulin secretion (6). Importantly, loss of β cell ABCA1 abrogated the whole-animal metabolic response to the antidiabetic thiazolidinedione (TZD) rosiglitazone, implicating an important role for β cell cholesterol homeostasis in the response to clinical therapy. Studies in mouse models of diabetes and dyslipidemia have also shown increased total islet cholesterol content and demonstrated a role for this increased cholesterol in the impairment of glucose sensing and insulin secretion (7). In addition to studies of islet cholesterol accumulation, studies of cholesterol depletion, using either the cholesterol scavenger methyl-β-cyclodextrin (MβCD) (8) or an inhibitor of endogenous cholesterol synthesis (9), have demonstrated a requirement for cholesterol in regulated insulin secretion.
The ABC transporter G1 (ABCG1) has been demonstrated to promote cholesterol efflux to HDL (10), but the mechanism by which ABCG1 mediates cholesterol efflux is not well understood. In contrast to ABCA1, which specifically couples cholesterol efflux to the acceptor ApoA1 (11, 12), the efflux activity of ABCG1 is relatively nonspecific, as it can promote efflux not only to HDL but also to LDL and to cyclodextrin (10). Additionally, though ABCG1 can traffic to the plasma membrane, several studies in different cell types have shown the bulk of it to be intracellular (13, 14). Whether ABCG1 is mainly mobilized to the cell surface to support cholesterol efflux (14) or regulates intracellular cholesterol distribution remains unclear.
Studies from our laboratory have shown decreased expression of ABCG1 in macrophages from diabetic mice (15) and humans (16), and this effect could be recapitulated by chronic culture in high glucose in vitro (15). Furthermore, ABCG1 is transcriptionally upregulated by the nuclear receptor PPARγ (17, 18), which is the pharmacologic target of TZDs. Despite these findings and an emerging role for cellular cholesterol homeostasis in β cell function, no study to date, to our knowledge, has investigated ABCG1’s role in this context. Here, we establish that ABCG1 supports insulin secretion, but more importantly, we highlight that this involves what we believe to be a novel intracellular action. ABCG1 largely localizes to insulin granules and appears to promote organelle cholesterol retention that is required for the function of the regulated secretory pathway. Furthermore, we show that islet ABCG1 expression is downregulated in disease and upregulated by TZD treatment, implicating a role for the maintenance of intracellular cholesterol distribution by ABCG1 in diabetes pathogenesis and treatment.
To investigate the role of ABCG1 in pancreatic β cell function, we first examined the expression pattern of ABCG1 mRNA and protein in pancreatic tissues. Conventional RT-PCR on isolated C57BL/6 mouse islets and 2 mouse pancreatic β cell lines, βTC-3 and MIN6, showed high levels of ABCG1 mRNA in these tissues (Supplemental Figure 1A; supplemental material available online with this article; doi: 10.1172/JCI41280DS1). This expression was enriched compared with whole pancreas and was comparable to expression levels in whole brain, where ABCG1 is highly expressed (13). Western blotting confirmed protein expression and validated both loss of the ABCG1 band by siRNA-mediated knockdown or gene ablation and induction by liver x receptor (LXR) agonist (Supplemental Figure 1B).
Given the enrichment of ABCG1 expression in pancreatic β cells, we next examined the role of ABCG1 in whole-animal glucose metabolism. Fasting blood glucose values of 10- to 12-week-old male Abcg1–/– mice (19) on a C57BL/6 background (backcrossed at least 10 generations) were slightly but significantly elevated as compared with WT (Figure (Figure1A).1A). Intraperitoneal glucose and insulin tolerance testing revealed impaired glucose clearance beginning at 20 minutes (Figure (Figure1B)1B) with normal insulin sensitivity (Supplemental Figure 2). This was accompanied by a 50% reduction in plasma insulin at 15 minutes after i.p. glucose injection (Figure (Figure1C),1C), consistent with a primary β cell defect. Experiments with female mice yielded similar results (Figure (Figure1,1, D–F), suggesting that these metabolic effects are not gender specific.
Our in vivo findings in the Abcg1–/– mice led us to hypothesize that ABCG1 may play an important role in pancreatic β cell function. Immunohistochemical analysis of pancreata from WT and Abcg1–/– mice showed no change in total insulin staining (Figure (Figure2,2, A and B), suggesting that loss of ABCG1 does not alter β cell mass. Instead, isolated islets from Abcg1–/– mice showed reduced glucose-stimulated insulin secretion (GSIS) as compared with WT (Figure (Figure2C).2C). The decrease observed in Abcg1–/– islets was not accounted for by increased basal secretion, in agreement with the unaffected fasting plasma insulin levels (Figure (Figure1,1, C and F). Together, these results imply that islet-specific ABCG1 plays a role in regulated insulin secretion.
Insulin secretion is initiated by glucose uptake through the β cell’s constitutive GLUT2 transporter, and the ensuing glucose phosphorylation, oxidative metabolism, and ATP generation lead to KATP channel closure. The resulting membrane depolarization and calcium (Ca2+) influx through voltage-gated Ca2+ channels trigger soluble NSF attachment protein receptor (SNARE) complex–mediated exocytosis of insulin granules (20). To begin to dissect the mechanism of impaired insulin secretion in Abcg1–/– islets, we assessed glucose sensing as judged by changes in intracellular calcium concentration ([Ca2+]i) measured by the ratiometric calcium dye fura-2. Raising the extracellular glucose concentration from 3 mM to 28 mM produced similar net increases in [Ca2+]i in WT and Abcg1–/– islets (Figure (Figure2,2, D and E), and the initial response in both cases was followed by characteristic [Ca2+]i oscillations (21, 22). These data suggest that loss of ABCG1 does not impair glucose sensing. Furthermore, stimulation by direct depolarization with 30 mM KCl revealed an insulin secretory defect similar to that observed in response to glucose (Figure (Figure2C),2C), indicative of a defect distal to KATP channel closure. Interestingly, Abcg1–/– islets exhibited increased insulin content (Figure (Figure2F),2F), consistent with a slight but nonsignificant increase in pancreatic insulin content (WT = 169 ± 13 μg/g tissue; Abcg1–/– = 181 ± 9 μg/g tissue, n = 6 each) as well as normal mRNA levels of the processed mature-insulin transcript, the unprocessed preinsulin transcript, and the regulatory enzyme glucokinase (Figure (Figure2G).2G). These findings indicate that Abcg1–/– β cells sense glucose and synthesize insulin normally and suggest that the defect in secretion is at the level of insulin release.
Insulin secretion typically occurs in 2 kinetic phases; the first is mediated by morphologically docked and primed granules, and the second by mobilization and release from a reserve pool (20). To better understand the defect in granule exocytosis in ABCG1-deficient islets, we approximated first- and second-phase insulin secretion using static incubations, similar to previous studies (23). Both phases were impaired in Abcg1–/– islets (Figure (Figure2,2, H and I). In aggregate, our in vitro studies of insulin secretion indicate that in the absence of ABCG1, there is a change in insulin granule releasability that impairs stimulus-secretion coupling.
Given the established role for ABCG1 in cholesterol efflux to HDL, we hypothesized that the secretory defect in Abcg1–/– islets may be due to β cell cholesterol accumulation secondary to impaired cholesterol efflux. Therefore, we assessed the role of ABCG1 in cellular cholesterol homeostasis using both ABCG1-directed siRNA in the MIN6 β cell line and isolated islets from Abcg1–/– mice. MIN6 cells showed robust efflux of cholesterol to HDL, similar to levels observed in other cell types (Figure (Figure3A)3A) (10). However, siRNA knockdown of ABCG1 protein levels by 90% (Figure (Figure3B)3B) had no effect on HDL-mediated cholesterol efflux at any time point or HDL concentration tested (Figure (Figure3,3, C and D) and no effect on ApoAI-mediated cholesterol efflux (Figure (Figure3E).3E). Additionally, WT and Abcg1–/– islets showed no difference in total-, free-, or esterified-cholesterol content (Figure (Figure3F).3F). Furthermore, transcriptional profiling of islets from WT and Abcg1–/– mice showed no significant change in oxysterol-regulated (ABCA1 and SREBP-1c) or cholesterol-regulated (SREBP-2, LDL receptor, HMG-CoA reductase, and HMG-CoA synthase) mRNA expression (Figure (Figure3G).3G). Finally, there is some evidence that in adipocytes, ABCG1 may play a role in triglyceride storage (24). Given the established role of free fatty acids and triglycerides in β cell function, we also measured triglyceride content in isolated islets from WT and Abcg1–/– mice. Results from these studies showed no difference between groups (Figure (Figure3H).3H). Importantly, Abcg1–/– mice had normal plasma lipoprotein profiles (Supplemental Figure 3), so the Abcg1–/– β cells were not exposed to different cholesterol-loading conditions in vivo prior to islet isolation. Together, these findings suggest that, at least under basal conditions, ABCG1 does not detectably regulate total β cell cholesterol content and that alterations in whole-cell cholesterol content do not mediate impairment of insulin secretion in ABCG1 deficiency.
In an effort to identify the site of secretory impairment in ABCG1 deficiency, we investigated the subcellular localization of ABCG1 in pancreatic β cells. Recent studies in other cell types have suggested that ABCG1 may be mostly concentrated intracellularly (13, 14). Because deficiency of ABCG1 did not impair cellular cholesterol efflux (Figure (Figure3),3), we hypothesized that β cell ABCG1 might also be primarily intracellular. The distribution of endogenous ABCG1 was assessed using multiple approaches including subcellular fractionation, surface biotinylation, and confocal immunofluorescence microscopy. To our surprise, differential centrifugation of MIN6 cells indicated that a significant proportion of ABCG1 was recovered in the granule-enriched fraction and that it cofractionated with the granule marker secretogranin (SG) and not the plasma membrane marker Na/K-ATPase (Figure (Figure4A).4A). To distinguish whether ABCG1 was actually associated with granules and not with a variety of other organelles that are potential contaminants, we subfractionated the enriched granule fraction on 2 separate density gradients and compared the distribution of ABCG1 to a variety of organelle markers. Western blotting of fractions from both density gradients indicated that ABCG1 comigrated quite closely with the mature granule markers secretogranin and insulin. In contrast, syntaxin 6 (25) and vesicle-associated membrane protein 4 (VAMP4) (markers of the immature granules and trans-Golgi network [TGN]; refs. 25, 26), secretory carrier membrane protein SCAMP3 (a TGN and endosomal marker; refs. 27, 28), and lysosomal-associated membrane protein 1 (LAMP1, a late endosomal/lysosomal marker) were all restricted to lower densities (Figure (Figure4,4, B–C, including densitometry profiles). These results suggest that ABCG1 resides to a substantial extent in the regulated secretory pathway as a membrane protein. Similar results were obtained upon fractionation of the rat adrenal chromaffin-derived PC12 cell line (Supplemental Figure 4), suggesting that ABCG1 may mark the regulated secretory pathway in multiple neuroendocrine cell types.
We also fractionated isolated C57BL/6 mouse islets. Results showed similar corecovery of ABCG1 and insulin in the granule-enriched fraction (Figure (Figure4D).4D). Subsequent centrifugation of this fraction over 1.5 M sucrose (29) revealed partial resolution of ABCG1 and insulin in that ABCG1 distributed 75%/25% between the interface (<1.5 M) and pellet (>1.5 M), whereas insulin distributed 50%/50% (Figure (Figure4D). 4D).
As a second methodology, and to directly address any potential localization of ABCG1 to the plasma membrane, we performed surface biotinylation experiments in MIN6 cells. As positive and negative controls, we compared biotinylation of the plasma membrane marker Na/K-ATPase and the cytoplasmic protein γ-adaptin. Results from these studies showed that only approximately 10%–12% of ABCG1 was surface labeled, whereas essentially all of Na/K-ATPase and none of γ-adaptin were detected (Figure (Figure4E).4E). These studies support the results from the cell fractionation experiments and demonstrate that the bulk of ABCG1 in β cells is intracellular.
Finally, we directly addressed the colocalization of ABCG1 with insulin in MIN6 cells using dual immunofluorescence staining (Figure (Figure4,4, F–K). As detection of endogenous ABCG1 protein by this method has proved difficult in other cell types, we used confocal microscopy to allow better spatial resolution and minimize background, and we validated the specificity of staining by modulating ABCG1 levels using siRNA and LXR agonism. Staining of control-, ABCG1 siRNA–, and LXR agonist–treated samples showed that ABCG1 immunofluorescence was reduced by ABCG1 siRNA and induced by LXR agonism in parallel with ABCG1 protein levels as assessed by Western blot in identically treated samples (Supplemental Figure 5A), demonstrating specificity of staining for ABCG1. Examination of dual stained cells at high resolution revealed significant colocalization of ABCG1 and insulin in cytoplasmic puncta (Figure (Figure4,4, F–K). Importantly, confocal z-stack series imaging showed that the colocalization extended throughout the depth of the sample, demonstrating that these are localized to the same (not adjacent but overlapping) structures (Supplemental Figure 5C). Additionally, costaining for the ER chaperone BiP and ABCG1 showed no colocalization (Supplemental Figure 5B), further demonstrating the specific intracellular colocalization of ABCG1 with insulin. In aggregate, localization data gathered from multiple methodologies demonstrates consistently that the bulk of ABCG1 protein localizes to insulin granules.
In light of the realization that ABCG1 largely resides in the regulated secretory pathway, we sought insight into its function in this compartment. Using transmission electron microscopy (TEM) on WT and Abcg1–/– islets, we noticed an apparent increase in size of β cell granules where the characteristic insulin crystal cores were surrounded by enlarged electron-lucent haloes (Figure (Figure5,5, A–D). Measurements made on images documented this as a significant increase in mean granule membrane delimited cross-sectional area (Figure (Figure5E)5E) and as a rightward shift when plotted as a frequency distribution (Figure (Figure5F).5F). TEM analysis of adrenal chromaffin cells from Abcg1–/– mice showed a similar shift in size distribution as compared with WT (Supplemental Figure 6), further supporting a general role for ABCG1 in multiple neuroendocrine cell types.
To complement these observations in a separate cell system, we used siRNA-mediated knockdown of ABCG1 in MIN6 cells and subjected cell homogenates to velocity sucrose gradient ultracentrifugation. Western blotting of fractions for insulin revealed greater migration into the gradient of granules from the ABCG1-deficient cells than from control cells, indicative of increased granule size (Figure (Figure5G,5G, including densitometry).
Interestingly, the TGN and secretory granule membranes are known to have a relatively high cholesterol content (30, 31), which has been shown to be important for secretory granule function. Some granule proteins associate with cholesterol-rich sites of budding in the TGN (31, 32), and cholesterol ultimately is required for membrane fusion during exocytosis (33). Furthermore, in mouse models of Smith-Lemli-Opitz syndrome (SLOS) and lathosterolosis, in which cholesterol biosynthesis is impaired, secretory granule morphology is remarkably similar to what we have observed in ABCG1 deficiency (34). Since ABCG1 activity is thought to alter membrane cholesterol distribution, we hypothesized that it may regulate granule cholesterol content. To assess this possibility, we performed siRNA knockdown of ABCG1 in MIN6 cells, labeled them with [3H]cholesterol overnight, isolated the granule-enriched fractions, subfractionated these on continuous density gradients, and assayed the recovery of [3H]cholesterol. We normalized these results to phospholipid as a measure of total membrane. The cholesterol content of the granule fraction from ABCG1-deficient cells was significantly reduced by an average of approximately 10% (Figure (Figure5H).5H). This outcome was consistent across 4 independent experiments. Importantly, whole-cell [3H]cholesterol labeling was equivalent between control and ABCG1 knockdown cells (Supplemental Figure 7A), so these differences were not due merely to changes in [3H]cholesterol uptake.
Notably, profiles of cholesterol recovery across granule density gradients indicate that this reduction in granule cholesterol in ABCG1 deficiency may be restricted to lower density granules and that cholesterol progressively declined at higher densities in both groups (Figure (Figure5I).5I). Incubation with [3H]cholesterol for an extended 48-hour period produced essentially identical results, indicating that the decline at higher densities is unlikely to reflect inaccessibility of these more dense granules to radiolabel. Furthermore, though there may be some contamination by other organelles, the recoveries of insulin and phospholipid mirror one another across the gradients, suggesting that insulin granules are the major source of membrane lipid across the gradients (Supplemental Figure 7B). The underlying basis for the variations in cholesterol content at different densities remains the subject of ongoing investigation.
One important consideration is that if granule cholesterol is reduced but total cellular cholesterol remains unchanged (Figure (Figure3F),3F), then the lost granule cholesterol must redistribute to another organelle. Therefore, in order to corroborate our findings and further evaluate the role of ABCG1 in β cell subcellular cholesterol distribution, we considered the cholesterol content of 2 main candidate organelles: the ER and the plasma membrane. Transcriptional profiling of cholesterol-regulated genes (Figure (Figure3G),3G), an exquisitely sensitive measure of ER cholesterol content, showed no difference between genotypes, so it does not appear that granule cholesterol redistributes to the ER pool. To assess relative plasma membrane cholesterol, we treated MIN6 cells with control or ABCG1 siRNA, labeled them overnight with [3H]cholesterol, then chilled the cells to 10°C to inhibit membrane recycling, and measured the rate of plasma membrane cholesterol extraction by methyl-β-cyclodextrin (MβCD), similar to previous protocols with minor modifications (35–37). Results from these studies showed a significant increase in the rate of plasma membrane cholesterol extraction by MβCD from ABCG1-deficient cells (Figure (Figure5J),5J), indicative of an increase in accessible cholesterol relative to control. Importantly, this increase cannot be due to impaired removal of plasma membrane cholesterol because efflux to physiological acceptors was not impaired (Figure (Figure3,3, C–E). Together, these data confirm the findings from measures of granule cholesterol, and they suggest that the plasma membrane is at least one site of granule cholesterol redistribution.
To further interrogate the link between reduced granule cholesterol, altered granule morphology, and impaired insulin secretion in ABCG1 deficiency, we undertook a series of experiments utilizing addition of exogenous cholesterol with cholesterol-loaded methyl-β-cyclodextrin (chol-CD). Treatment of ABCG1-deficient MIN6 cells with chol-CD overnight restored granule cholesterol content to the level observed in control cells (Figure (Figure6A).6A). Strikingly, exogenous cholesterol also restored insulin granule size, as indicated by the leftward shift of insulin recovered in velocity gradient fractions (Figure (Figure6B;6B; compare with Figure Figure5G).5G). Similarly, cholesterol addition to isolated Abcg1–/– islets restored a near-normal granule size distribution as measured from TEM images (Figure (Figure6,6, C–E; mean granule areas from untreated and chol-CD–treated islets from both genotypes are shown together in Figure Figure6F).6F). Finally, addition of exogenous cholesterol also markedly improved insulin secretion in Abcg1–/– islets (Figure (Figure6G),6G), and paralleled the changes in granule area (Figure (Figure6F).6F). Together, these results indicate that ABCG1 is a novel regulator of cholesterol content within the regulated secretory pathway and that reduced granule cholesterol is one underlying cause of impaired insulin secretion in ABCG1 deficiency.
Prior studies from our group have shown that ABCG1 expression is reduced in macrophages isolated from diabetic mice (15) and humans (16) and that this reduction could be recapitulated with chronic high glucose in vitro (15). To determine whether this phenomenon occurs similarly in β cells, we isolated islets from diabetic db/db mice and control db/+ mice and measured ABCG1 levels. Hyperglycemic db/db mice showed a marked reduction in ABCG1 mRNA (Figure (Figure7A)7A) and protein (Figure (Figure7B),7B), while normoglycemic db/db mice did not differ from db/+ controls. Plotting islet ABCG1 mRNA as a function of blood glucose revealed a significant negative correlation as assessed by linear regression analysis (Figure (Figure7C).7C). Additionally, chronic culture of WT islets in high D-glucose in vitro led to significant reductions in ABCG1 mRNA at 3 and 5 days of treatment, as compared with the nonmetabolizable L-glucose control (Figure (Figure7,7, D and E). Together, these results demonstrate that islet ABCG1 expression is reduced in a murine model of diabetes, and they suggest that this downregulation may be mediated by chronic high glucose. Furthermore, these data suggest that reduced ABCG1 expression may be one mechanism contributing to progressive β cell dysfunction in chronic hyperglycemia.
TZDs such as rosiglitazone and pioglitazone are commonly used drugs for the clinical treatment of diabetes and have numerous beneficial effects on β cell function (21, 38), including significantly elevated cellular granule content (39). A recent elegant study by Brunham et al. showed that ABCA1 in β cells was required for the whole-animal metabolic response to rosiglitazone (6), which suggests that β cell cholesterol homeostasis may be a critical factor in this process. Importantly, ABCG1 is a known transcriptional target of PPARγ agonism in other tissues (17, 18). To determine whether PPARγ could upregulate ABCG1 expression in β cells, we treated MIN6 cells for 24 hours with pioglitazone and measured ABCG1 mRNA. This treatment significantly upregulated ABCG1 expression (Figure (Figure7F),7F), indicating that pioglitazone can activate β cell ABCG1 expression in a cell-autonomous manner. We observed a similar increase in human islets treated with pioglitazone in vitro (Figure (Figure7G),7G), indicating that this effect extends to primary human cells. We then asked whether β cell ABCG1 could be regulated by PPARγ agonism in the context of β cell dysfunction. We measured ABCG1 expression in islets from db/db mice treated for 6 weeks with pioglitazone (Pio-db/db). Pio-db/db mice had significantly reduced random blood glucose, improved glucose tolerance, and increased serum insulin levels as compared with untreated db/db mice, indicative of markedly improved β cell function (21). Interestingly, islet ABCG1 expression in Pio-db/db mice was increased 4-fold over untreated db/db mice and was indistinguishable from nondiabetic control levels (Figure (Figure7H).7H). This demonstrates that the transcriptional machinery required to upregulate ABCG1 expression is intact in the failing β cell, and it suggests that increased β cell ABCG1 may contribute to improved β cell function in response to TZDs.
Cellular cholesterol homeostasis has emerged as an important modulator of insulin secretion, and altered β cell cholesterol homeostasis has been implicated in islet dysfunction in disease and the response to pharmacologic therapy. ABCG1 has been shown to regulate cholesterol content in other cell types, but its role in β cell function has not been explored. The studies reported here extend the role of ABCG1 in cellular cholesterol homeostasis to β cells and link this to modulation of insulin secretion. More importantly, they demonstrate the primary localization of ABCG1 to an intracellular compartment and subsequently uncover a role for ABCG1-mediated cholesterol transport in the function of that compartment, the regulated secretory pathway. These findings have important implications, not only for how we understand β cell function and regulated secretion, but also for how we assess the general role of ABCG1 in cellular cholesterol homeostasis.
The first data implicating ABCG1 in β cell function came from our studies of glucose metabolism in WT and Abcg1–/– mice, wherein Abcg1–/– mice had impaired glucose tolerance and insulin secretion, consistent with a primary β cell defect (Figure (Figure1).1). In parallel with these in vivo observations, Abcg1–/– islets had decreased insulin secretion in vitro. This defect appeared to be distal to glucose sensing and to extend to both first- and second-phase insulin secretion (Figure (Figure2),2), suggesting that loss of ABCG1 leads to impairment of later steps in insulin granule release, perhaps at the level of granule priming or membrane fusion.
These secretory defects are similar to those shown in LXRβ-deficient islets (40), which have an 80% reduction in islet ABCG1 mRNA, and in ABCA1-deficient islets (6). However, both LXRβ- and ABCA1-deficient islets had some measure of either cholesterol or neutral lipid accumulation. Surprisingly, we found no change in β cell cholesterol efflux or content in ABCG1 deficiency (Figure (Figure3).3). In order to better understand how loss of ABCG1 could impair insulin secretion without an increase in cellular sterol content, we utilized multiple methodologies to assess the subcellular localization of endogenous ABCG1 protein in both MIN6 and primary mouse β cells (Figure (Figure4).4). Surprisingly, only about 10% of ABCG1 was present on the cell surface, whereas most of the remaining 90% localized to insulin granules. This localization may be limited to a subset of insulin granules, as both immunofluorescence and density gradient fractionation of MIN6 cells and islets showed a proportion of insulin staining without significant ABCG1. To the extent that increased granule density correlates with increased granule maturity (41), these ABCG1-deficient granules may be older or more mature. This raises the interesting possibility that ABCG1 may not be retained throughout the granule lifetime, though its retention substantially exceeds that of VAMP4 and syntaxin 6 — membrane components that are restricted to immature granules (25, 26). In any event, ABCG1 is a long-term resident of the regulated secretory pathway; how it enters and possibly exits the pathway remain subjects for further investigation.
In light of this subcellular localization of ABCG1, we assessed granule morphology and cholesterol content in ABCG1 deficiency. Interestingly, the granule morphology in Abcg1–/– islets resembled that seen in mice with impairments in cholesterol biogenesis (34), and measurements of granule lipid content showed decreased granule cholesterol in ABCG1-deficient cells (Figure (Figure5).5). Profiling of granule cholesterol content across density gradients suggested that this decrease may be limited to less dense granules, as the difference in granule cholesterol between control and ABCG1 siRNA was blunted at higher densities. This is consistent with a possible restriction of ABCG1 to this lower density granule subset. Importantly, addition of exogenous cholesterol restored secretory compartment cholesterol and granule size and rescued insulin secretion in Abcg1–/– islets (Figure (Figure6),6), indicating that this reduced granule cholesterol is at least one mechanism of impaired insulin secretion in ABCG1 deficiency. Granule cholesterol deficiency could negatively impact the regulated secretory pathway at various levels. Interestingly, cholesterol itself supports lipid bilayer rearrangement during membrane fusion, and studies in a cell-free system showed that a reduction of granule membrane cholesterol by only 10% reduced exocytotic events to 60%–70% of control (33): both similar values to what we observe in ABCG1 deficiency. Cholesterol-enriched membranes also support SNARE-mediated exocytosis (42), and reduced granule cholesterol content could alter the assembly or function of SNARE complexes. Alternatively, the impaired insulin secretion could reflect retention of a negative regulator of exocytosis during granule maturation (26), particularly since cholesterol contributes to assembly of membrane microdomains during vesicle budding and the granule morphology we observe is similar to that observed in other examples of impaired membrane sorting (43).
The precise cellular mechanisms for ABCG1 action in cholesterol homeostasis are unknown. The best evidence on ABCG1-mediated cholesterol efflux underscores 2 points: first, ABCG1 probably has to be at the plasma membrane to promote cholesterol efflux (14, 44); and second, ABCG1 redistributes membrane cholesterol to efflux accessible pools (44). Most of the plasma membrane cholesterol resides in the cytoplasmic leaflet (45), and one mechanism whereby ABCG1 could promote cholesterol efflux would be to facilitate its redistribution to the outer leaflet. In the secretory granule, this would correspond topologically to the inner (lumenal) leaflet. Interestingly, other intracellular ABC transporters act to drive their substrates into their respective intracellular compartments in a similar manner (46, 47), and in fact the Drosophila homolog of ABCG1, white, localizes to pigment granule membranes in the fly eye, where it facilitates uptake of pigment precursors into the granule (48). Thus, we propose that ABCG1 maintains secretory granule cholesterol by facilitating its concentration in the granule membrane inner leaflet, thereby limiting cholesterol dispersion to other cellular sites (Figure (Figure8).8). This concept is particularly interesting in the β cell because of the relative proportion of cellular lipid that is partitioned to the granule membrane. Stereological analysis of islets showed that each β cell contains approximately 9,000 granules and that the total surface area of these granules amounts to 4.5 times that of the plasma membrane (49). Furthermore, the cholesterol content of granule membranes is similar to that in the plasma membrane (50), so though the plasma membrane may contain the bulk of cellular-free cholesterol in other cell types (51), in the β cell, the granule accounts for a substantial portion of the cholesterol pool. It is perhaps not quite as surprising then that a cholesterol transporter localizes to this compartment where it can play a pivotal role in cholesterol handling. This realization also has important implications for how we interpret measures of whole-cell cholesterol content in β cells and underscores the importance of assessing subcellular cholesterol distribution.
Given our data on the role of ABCG1 in insulin secretion, we sought to extend previous findings on the regulation of ABCG1 expression in other cell types to β cells (Figure (Figure7).7). As in macrophages, ABCG1 expression was significantly reduced in islets isolated from hyperglycemic db/db mice, and chronic culture of isolated islets in high glucose produced a similar effect. These results suggest that islet ABCG1 expression is probably reduced later in disease when hyperglycemia has already ensued and that this effect might propagate metabolic dysfunction in a positive feedback cycle, reminiscent of β cell failure in advanced disease. The mechanism for downregulation of ABCG1 expression by high glucose is unclear. Importantly, mRNA levels of LXRβ, the primary regulator of ABCG1 expression in islets, were unchanged, whereas mRNAs of other transcriptional targets of LXR (ABCA1 and SREBP-1c) were downregulated (Supplemental Figure 8), consistent with altered LXR activity. This effect could be mediated by posttranscriptional or posttranslational modifications of LXR (52, 53); similar mechanisms have been shown in islets for the transcription factors Pdx1 and MafA (54, 55). Alternatively, reduced LXR activity may reflect reduced ligand availability, as shown during T lymphocyte expansion (56). Interestingly, pioglitazone increased islet ABCG1 expression in vitro and restored islet ABCG1 expression to normal levels in diabetic animals. Though TZDs are widely used to treat metabolic dysfunction in diabetes, clinical data have generated controversy as to the efficacy of TZDs in prevention of cardiovascular disease and mortality (57). Based on our data, at least at the level of the β cell, specific targeting of ABCG1 expression may be a viable alternative approach. Future studies aimed at testing this hypothesis will require β cell–specific deletion or overexpression of ABCG1 in vivo in models of diabetes.
In summary, we have shown here that ABCG1 has a critical intracellular role in pancreatic β cell function. Our paradigm indicates that ABCG1 does not modulate insulin secretion through regulation of total cellular cholesterol content, but instead through regulation of subcellular cholesterol distribution and maintenance of membrane cholesterol content in a cholesterol-rich organelle, the secretory granule. Furthermore, we have shown that modulation of islet ABCG1 expression may play an important role in diabetes pathogenesis and treatment. Many studies of ABCG1 in a variety of other cell types and biological processes, such as atherosclerosis, have produced unexpected and discrepant results that may not be explained simply by total cellular cholesterol accumulation (58, 59). Future studies addressing the subcellular localization and trafficking of ABCG1 in these other cell types will likely shed light on the mechanisms of ABCG1 action in diverse cellular processes and may thereby elucidate the roles of ABCG1 in human disease.
C57BL/6 mice and db/db C57BLKS mice were purchased from Jackson Laboratory. Abcg1–/– mice (19) were purchased originally from Deltagen. All mice were fed standard rodent chow, housed in microisolator cages in a pathogen-free facility, and maintained under a 14-hour light/10-hour dark cycle, with lights on at 6:30 am, lights off at 8:30 pm. For pioglitazone studies, pioglitazone was initially dissolved in ethanol at a mixture of 10% w/v, then diluted with PBS to a final ethanol concentration of 1.8%. Mice were given 20 mg/kg body weight pioglitazone in 400 μl ethanol/PBS (or ethanol/PBS vehicle alone) daily by oral gavage as described (21). All procedures were reviewed and approved by University of Virginia and Indiana University Laboratory Animal Care and Use Committee review boards.
For glucose tolerance testing, mice were fasted overnight and challenged with D-glucose (1–3 g/kg body weight) via i.p. injection. For insulin tolerance testing, mice were fasted 4 hours and, between the hours of 2:00 and 3:00 pm, challenged with recombinant human insulin (0.75 U/kg body weight; Novolin R, Novo Nordisk). Blood glucose levels were measured via tail clip with a OneTouch Ultrasmart glucometer. Plasma insulin was measured using Ultrasensitive Mouse Insulin ELISA (ALPCO).
Primary mouse islets were isolated via intraductal collagenase (Roche) digestion and density centrifugation essentially as described previously (60). Following isolation, islets were hand picked and maintain in DMEM without phenol red, 5.5 mM glucose plus 10% (v/v) FBS (Sigma-Aldrich), and penicillin and streptomycin. Human islets were obtained from Beta-Pro LLC and upon arrival were placed in DMEM without phenol red, 5.5 mM glucose plus 10% (v/v) fetal bovine serum, and penicillin and streptomycin, and allowed to recover overnight before initiation of experiments. MIN6 cells were a gift of Chien Li (University of Virginia) and were maintained in DMEM with 15% FBS and penicillin and streptomycin. PC12 cells were maintained at 10% CO2 in DMEM with 10% horse serum and 5% FBS (61). For in vitro studies with pioglitazone, cells were cultured in DMEM, 5.5 mM glucose plus 1% (v/v) lipoprotein-deficient fetal bovine serum (Biomedical Technologies) with either vehicle (DMSO) or 5–10 μM pioglitazone for 24 hours. For treatment with soluble cholesterol, cells were incubated overnight with 20 μg/ml cholesterol-loaded methyl-β-cyclodextrin (Chol-CD) (Sigma-Aldrich) in the presence of 10% (islets) or 15% (MIN6) FBS. Chol-CD was not included in the medium during GSIS in order to prevent acute cholesterol loading of the plasma membrane. Knockdown of ABCG1 in MIN6 cells was done with heteroduplex siRNA constructs (Dharmacon, L-040312) and using Lipofectamine RNAiMAX (Invitrogen) according to manufacturer instructions. 1.2 × 106 MIN6 cells were plated with 25–50 pmol of siRNA/well of a 6-well dish for protein harvest or fractionation, or 3 × 105 cells with 6.25–12.5 pmol siRNA/well of a 24-well dish for efflux assays. Experiments were performed 60 hours following initiation of siRNA unless otherwise described. For LXR agonism, T0901317 was from Caymen, and GW3965 was from Sigma-Aldrich.
Measurements of β cell mass were performed as described (62). Mice were euthanized and perfused through the heart with PBS and 4% paraformaldehyde. Fixed pancreata were embedded in paraffin, and 5 μM sections were stained with rabbit polyclonal anti-insulin antibody (1:500; Santa Cruz Biotechnology Inc.) and secondary antibody peroxidase conjugate to enable quantification of positive-stained areas. Digital images of each section at ×10 magnification were acquired on an Axio-Observer Z1 microscope (Zeiss) fitted with an AxioCam high resolution color camera. Relative β cell area was calculated using Axio-Vision Software.
Glucose- and potassium-induced insulin secretion were measured as described (23, 62). Islets (50 per sample) were treated according to specified experimental conditions, preincubated in 0 mM glucose HEPES-buffered Krebs-Ringer solution (KRB) for 1 hour, then incubated for 1 hour in 3 mM glucose KRB, followed by 1 hour in KRB containing either 30 mM KCl or 28 mM glucose. Following incubations, total islet insulin was extracted with acid ethanol. Insulin concentration in the media and extracts was measured using the Ultrasensitive Mouse Insulin ELISA from ALPCO. Insulin secretion was normalized to total insulin content. For measurements of first- and second-phase insulin secretion, islets were incubated in 3 mM glucose KRB plus 30 mM KCl for 5 minutes, followed by incubation in 28 mM glucose KRB for an additional 10 minutes. Intracellular calcium measurements were made using the ratiometric calcium dye fura-2 (63). Islets were loaded with 3 μM fura-2-acetoxymethylester (fura-2 AM) for 30 minutes, transferred to a low-volume chamber (Warner Instruments), mounted on the stage of an Olympus BX51WI fluorescence microscope, and subsequently perifused with 3 mM followed by 28 mM glucose KRB delivered by a peristaltic pump (Gilson) maintained at 37°C with an in-line heater (Warner Instruments). During perifusion, fura-2 AM was excited by xenon bulb light source, with sequential excitation at 340 and 380 nm and emission recorded at 510 nm. Relative Ca2+ concentration was determined by the ratio of emission from the 2 excitation wavelengths.
Cholesterol efflux assays were performed as described with minor modifications (15). MIN6 cells were plated in 24-well culture dishes and treated according to experimental conditions. On the day of the experiment, cells were washed 3 times with DMEM, preincubated in DMEM plus 0.02% fatty acid–free bovine serum albumin (FAFBSA) for 2 hours, followed by incubation with media containing either FAFBSA alone or FAFBSA plus the lipoprotein acceptor, which, depending upon the experiment, was either 50 μg/ml HDL (Intracel) for 30 minutes to 8 hours, 5–100 μg/ml HDL for 4 hours, or 15 μg/ml human ApoAI (isolated as previously described in ref. 64) for 4 hours. Following incubation in acceptor media, the media were removed, cells were dried, and lipid was extracted overnight in isopropanol. [3H]cholesterol in media and extract was quantified by liquid scintillation counting. HDL- or ApoAI-mediated cholesterol efflux was calculated by subtracting efflux to FAFBSA containing media from efflux to media with FAFBSA plus acceptor, and was normalized to cellular [3H]cholesterol content. For plasma membrane cholesterol extraction assays, cells were treated as described above, but following the 2-hour preincubation period, cells were chilled to 10°C, extraction media containing 0.2% FAFBSA plus 10 mM methyl-β-cyclodextrin was added, and aliquots were taken and replaced with fresh media at 1, 2, 5, 10, and 15 minutes. Cells and samples were then processed as described above. For assessment of subcellular [3H]cholesterol distribution, cells were labeled with 2 μCi/ml [3H]cholesterol overnight, subjected to fractionation as described, and radioactivity was measured via scintillation counting. In experiments in which cells were treated with cholesterol-loaded methyl-β-cyclodextrin, both FBS and chol-CD were prelabeled for 1 hour at room temperature with equivalent amounts of [3H]cholesterol to yield a final concentration of 2 μCi/ml [3H]cholesterol. Cholesterol content was normalized to phospholipid content following extraction (65) and assay of lipid-derived inorganic phosphate (66). Total islet cellular cholesterol content was quantified by gas-liquid chromatography (67). MIN6 total cholesterol content was determined by enzymatic assay (WAKO). Triglyceride content was quantified using the Infinity Triglycerides Reagent enzymatic assay (Thermo Scientific).
MIN6 (29, 68) and PC12 (69) cells were fractionated using previously described procedures with minor modifications. Briefly, MIN6 cells were grown to 70%–80% confluence and homogenized by 10–12 passes through a 27-gauge needle in homogenization buffer (20 mM Tris-HCl, pH 7.4, 0.5 mM EDTA, 0.5 mM EGTA, 250 mM sucrose, 1 mM DTT, and 1:100 Sigma-Aldrich protease inhibitor cocktail). Homogenization efficiency was monitored by phase-contrast microscopy. Whole-cell homogenate was centrifuged at 900 g for 10 minutes to pellet nuclei and plasma membrane (PM). The post-nuclear supernatant (PNS) was spun at 5,400 g for 15 minutes, and the post-mitochondrial supernatant (PMS) was subsequently spun at 25,000 g for 20 minutes to pellet the granule-enriched fraction. The remaining supernatant was then spun at 100,000 g for 1 hour to pellet microsomes from cytosol. For PM collection, the nuclear/PM pellet was resuspended in 1 ml 0.25 M sucrose plus 2 ml 2 M sucrose, overlaid with 0.25 M sucrose, and subsequently spun at 113,000 g for 1 hour. The interface was collected, diluted, and pelleted at 8,000 g to generate the PM fraction. In preparation for equilibrium density gradient centrifugation (to separate on the basis of organelle density), the PMS was adjusted to 10 mM EDTA, and the granule-enriched fraction was sedimented onto either a high-density sucrose or iodixanol (OptiPrep; Nyegard) cushion. The collected fraction was loaded on either a continuous sucrose (0.5–2.0 M) or iodixanol (7.5%–30%) gradient and spun at 150,000 gmax for 5 hours. For velocity sucrose gradient centrifugation (to separate on the basis of organelle size), the PNS was adjusted to 10 mM EDTA, loaded on a continuous sucrose gradient (0.3–1.2 M), and spun at 110,000 g for 1 hour. 1 ml fractions were collected, and the pellet (fraction 10), which included some whole cells and possibly organelle aggregates, was excluded from distribution analysis. Primary mouse islets were fractionated as previously described (29).
Immunofluorescence staining and microscopy were performed as previously described (28). MIN6 cells were plated on tissue culture–treated glass chamber slides (Lab-Tek) and where applicable were treated with siRNA or T0901317 as described. Cells were fixed with 4% paraformaldehyde, quenched in PBS containing 50 mM glycine, and permeabilized in PBS plus 0.1% Triton X-100. Samples were blocked with 2% goat serum, labeled with primary antibodies to rabbit anti-ABCG1 (2–5 μg/ml), guinea pig anti-insulin (1:1500), and mouse anti-BiP (1:200), and subsequently labeled with Alexa Fluor–conjugated secondary antibodies (1:500; Alexa Fluor 488 for ABCG1, Alexa Fluor 647 for insulin and BiP). Images were acquired on a NikonC1 confocal microscope. Quantitative analysis of colocalization was done using the JACoP plug-in in ImageJ software (70).
Surface biotinylation of MIN6 cells was performed similarly to what was previously described (71). Cells were grown in 10 cm dishes, plated 3 days before the experiment, labeled on ice for 30 minutes with 1 mg/ml EZ-Link Sulfo-NHS-SS-Biotin (Thermo Scientific) in Ca2+- and Mg2+-containing PBS, pH 8; quenched with 1% BSA, and lysed and scraped in 0.1% NP-40 in PBS, 2 mM EDTA, 2.5 mM iodoacetamide, protease inhibitor cocktail (Roche), 1 ml per plate. A portion of the lysate was then removed, and biotinylated proteins were precipitated with Neutravidin beads for 2 hours at room temperature. Following binding, beads were washed and bound proteins were eluted with sample buffer plus DTT. SDS-PAGE and Western blotting were performed from gels loaded with 3 different volumes each of lysate and bead eluate to confirm proportionate antigen recovery, and blots were developed and quantified using a Fuji LAS-3000 imager and ImageQuant software.
Primary mouse islets were isolated, fixed in phosphate-buffered 4% PFA, 2.5% glutaraldehyde, post-fixed with 1% osmium tetroxide, dehydrated in acetone, and embedded in EPON. Ultrathin sections were stained with aqueous uranyl acetate and lead citrate. Dissected mouse adrenal glands were similarly processed except that initial fixation was in 0.1 M cacodylate-buffered 2% formaldehyde, 3% glutaraldehyde, and the osmium used for post-fixation contained 0.1% potassium ferrocyanide. Specimens were stained in-block with 0.5% uranyl acetate, pH 6, prior to dehydration and embedding. Digital images were acquired using a JEOL 1230 electron microscope in the University of Virginia Advanced Microscopy Facility and analyzed using Image-Pro software (Media Cybernetics Inc). For islet granule analysis, insulin granules were identified by the characteristic appearance of their zinc-insulin condensed core with surrounding halo. For chromaffin granule analysis, care was taken to focus on a single cell type in which granules had dense cores surrounded by membrane-limited haloes.
Western blotting and quantitative RT-PCR (qRT-PCR) analysis were performed essentially as previously described (15, 72). For Western blotting, RIPA buffer (50 mM TrisHCl, pH 8.0, 150 mM NaCl, 1% Igepal, 10 mM NaF, 2 mM Na3VO4, and Sigma-Aldrich proteases inhibitor cocktail) was added to cells to generate whole-cell lysates. Total protein was quantified using a protein assay kit (Bio-Rad). Proteins were separated by SDS-PAGE and transferred to nitrocellulose (materials and buffers from Invitrogen); membranes were washed with Tris-buffered saline with 0.1% (v/v) Tween-20 (TBST) and blocked in TBST with 2.5% milk (w/v). Proteins were detected using specific primary antibody and horseradish peroxidase–conjugated secondary antibody, and visualized by chemiluminescence. Antibodies used were obtained as follows: ABCG1 (Novus), β-actin (Santa Cruz Biotechnology Inc.), BiP (Stressgen), insulin (rabbit, Santa Cruz Biotechnology Inc.; guinea pig, Linco), LAMP1 (Developmental Studies Hybridoma Bank, University of Iowa), Na/K-ATPase (Upstate), SCAMP3 (generated as described; ref. 73), secretogranin (isoform II; Biodesign), syntaxin 6 (Stressgen), and VAMP4 (Affinity Bioreagents). For qRT-PCR, RNA was isolated using the RNeasy Mini Kit (QIAGEN) as per manufacturer’s instructions. RNA was quantified, and cDNA from equivalent starting material for each sample was synthesized using the iScript cDNA Synthesis Kit (Bio-Rad). Total cDNA was then diluted 1:10 in water, and a serial dilution standard and negative control (no reverse transcriptase) were used. A Bio-Rad MyIQ Single Color Real-Time PCR Detection System and the Bio-Rad iQ SYBR Green Supermix were used to run the reactions and quantify.
Data were analyzed by ANOVA, 2-tailed Student’s t tests, or linear regression, where appropriate, using GraphPad Prism4 statistical software. All results are presented as mean ± SEM unless otherwise specified.
The authors would like to acknowledge M.D. Skaflen and A.K. Gebre for their expert technical assistance, J.A. Redick and S. Guillot for their electron microscopy services, and C.S. Nunemaker for helpful discussion. J.M. Sturek was supported by an NIGMS-NHLBI predoctoral fellowship. This work was supported in part by grants from the NIH (P01 HL55798 and R01 HL085790 to C.C. Hedrick; R01 DK073380 to J.D. Castle; K08 DK 80225 to C. Evans-Molina; R01 HL094525 and P01 HL049373 to J.S. Parks; and R01 DK60581 to R.G. Mirmira) and an investigator-initiated grant from Takeda Pharmaceuticals to R.G. Mirmira.
Conflict of interest: The authors have declared that no conflict of interest exists.
Citation for this article: J Clin Invest. 2010;120(7):2575–2589. doi:10.1172/JCI41280.