The Inactivation of Arx in Pancreatic α-Cells Triggers Their Neogenesis and Conversion into Functional β-Like Cells

Recently, it was demonstrated that pancreatic new-born glucagon-producing cells can regenerate and convert into insulin-producing β-like cells through the ectopic expression of a single gene, Pax4. Here, combining conditional loss-of-function and lineage tracing approaches, we show that the selective inhibition of the Arx gene in α-cells is sufficient to promote the conversion of adult α-cells into β-like cells at any age. Interestingly, this conversion induces the continuous mobilization of duct-lining precursor cells to adopt an endocrine cell fate, the glucagon+ cells thereby generated being subsequently converted into β-like cells upon Arx inhibition. Of interest, through the generation and analysis of Arx and Pax4 conditional double-mutants, we provide evidence that Pax4 is dispensable for these regeneration processes, indicating that Arx represents the main trigger of α-cell-mediated β-like cell neogenesis. Importantly, the loss of Arx in α-cells is sufficient to regenerate a functional β-cell mass and thereby reverse diabetes following toxin-induced β-cell depletion. Our data therefore suggest that strategies aiming at inhibiting the expression of Arx, or its molecular targets/co-factors, may pave new avenues for the treatment of diabetes.


Introduction
The mature pancreas is organized into two compartments with distinct functions: the exocrine pancreas is involved in the production of digestive enzymes and their transport into the duodenum via a branched ductal tree, whereas the endocrine compartment consists of highly vascularized functional units called islets of Langerhans. These contain five distinct cell subtypes, a-, b-, d-, eand PP-cells, responsible for the production of the hormones glucagon, insulin, somatostatin, ghrelin and pancreatic polypeptide (PP), respectively [1][2][3][4][5]. Type 1 Diabetes Mellitus (T1DM) is caused by an autoimmune reaction leading to the loss of insulin-producing b-cells resulting in chronic hyperglycemia. Current therapies allow a measure of control of blood glucose levels in T1DM patients. However, due to the difficulties encountered in maintaining appropriate glycemic levels, such patients exhibit an increased risk of vascular complications [6]. Thus, in an effort to design improved protocols, current research efforts aim at elucidating the genetic programs underlying pancreatic endocrine cell lineage determination and/or reprogramming, an approach that could be of interest for the development of alternative therapies for T1DM. Pancreatic development involves an elaborate process of morphological events accompanied by a complex pattern of cellular differentiation and lineage selection. These events are mediated in great part by tissue interactions, signaling pathways and directed cascades of gene expression that determine cell fate [7,8]. Thus, Pdx1 plays a critical role in early pancreas development [9][10][11][12][13], whereas the bHLH-containing transcription factor Neurogenin 3 (Ngn3) specifies the endocrine cell lineage [14][15][16][17][18]. Subsequent endocrine cell fate allocation is largely dependent on the interplay between additional transcription factors, including Arx and Pax4.
These were shown to display opposing activities in the embryonic processes underlying the specification toward the endocrine subtype destinies: while Pax4 is instrumental for the allocation to the band d-cell lineages, Arx is involved in specifying the a-cell fate [19][20][21]. Subsequent studies demonstrated that the misexpression of Arx in adult b-cells induces their conversion into cells exhibiting aor PP-cell characteristics [22]. Most interestingly, the ectopic expression of Pax4 in glucagon-expressing cells was found to induce their regeneration and subsequent conversion into cells displaying most of the typical features of b-cells, these being able to counteract the effects of chemically-induced hyperglycemia [23,24].
Due to (1) the opposing functions of Pax4 and Arx during embryonic development and (2) the persistent expression of Arx in glucagon-producing cells, we sought to determine whether the inhibition of Arx expression in a-cells could induce their conversion into b-like cells. To this end, we generated different transgenic models allowing the constitutive or conditional inactivation of Arx in glucagon-expressing cells. Using lineage tracing, we provide evidence that glucagon-producing cells, even at relatively advanced ages, may be converted into functional b-like cells solely upon Arx inactivation. In addition, following the loss of glucagon-producing cells, a cycle of endocrine cell regeneration is initiated whereby glucagon + cells are subsequently acquiring a b-cell phenotype, ultimately leading to islet hypertrophy due to a b-like cell hyperplasia. Interestingly, the further inactivation of Pax4 in these animals does not impact these processes, suggesting that Arx is the main player in a-cell reprogramming. Most importantly, upon the chemical induction of diabetes, Arx mutants display a clear regeneration of b-cells, reversion of diabetes and an extended lifespan compared to controls.

The inactivation of Arx in a-cells of different ages results in insulin-producing cell hyperplasia
To determine the consequences of the loss of Arx in a-cells, we first generated animals allowing the constitutive deletion of Arx in all glucagon-producing cells ( Figure S1 Left) by crossing the ArxcKO mouse line (in which the second exon of the Arx gene is flanked by LoxP sites [25]) with Glu-Cre transgenic animals (generated using a transgene composed of the glucagon promoter driving the expression of the phage P1 Cre recombinase [26]). The resulting double transgenics (referred to as Glu-ArxKO) were further crossed with ROSA26-LoxP-Neomycin Resistance-STOP-LoxP-b-gal animals (containing a transgene encompassing the ubiquitous ROSA26 promoter in front of the neomycin resistance gene with STOP codons flanked by LoxP sites and followed by the b-galactosidase cDNA [27] -henceforth referred to as Rosa) for lineage tracing purposes.
A number of tests were performed to determine the efficiency of this strategy. First, by combining several immunohistochemical approaches, we analyzed 2 week-old homozygous Glu-Cre::Rosa animals to further verify the efficiency of glucagon-mediated Cre activity ( Figure S2). A quantitative analysis demonstrated approximately 7267% of glucagon + b-gal + cells in the pancreas of these animals ( Figure S2A-C), a result in line with previously published data [24,26,28]. Next, age-matched Glu-ArxKO pancreata were assayed for Arx expression by immunohistochemistry: our data indicated a loss of Arx expression in approximately 6766% of glucagon-producing cells ( Figure S2G-I) as compared to ArxcKO or wild-type controls ( Figure S2D-F), a proportion matching the ratio of Cre + glucagon-expressing cells. Of note, in a number of these Arx 2 Glucagon + cells, Pax4 was clearly detected, suggesting that the loss of Arx results in the ectopic expression of Pax4 in a-cells ( Figure S2G-I arrowheads). Lastly, qPCR was used to quantify the Arx transcript content ( Figure S2J): a comparison of controls versus Glu-ArxKO pancreata outlined a 74% reduction in Arx transcripts upon glucagon-mediated Arx deficiency, a result in line with cell quantification. Taken together, our results suggest that, in this experimental model, Cre can efficiently lead to the inactivation of Arx in glucagon-producing cells.
In a second set of experiments, to develop an animal model permitting the inducible deletion of Arx in adult a-cells ( Figure S1 Right), the ArxcKO mouse line was crossed with the Glu-rtTA transgenic line (containing a transgene composed of the rat glucagon promoter [29] upstream of the reverse tetracycline-dependent transactivator) and further mated with TetO-Cre animals (whose transgene includes the Tet operator upstream of Cre Recombinase cDNA [30]). Of note, Glu-rtTA::TetO-Cre were previously characterized [23]. The resulting triple-transgenic mice will henceforth be referred to as IndGlu-ArxKO (inducible Glu-ArxKO). From hereon, mice treated with Doxycycline for x months will be referred to as xmDox+. In all cases, untreated animals were found phenotypically similar to their wild-type (WT) counterparts; they will be referred to as ''Dox2'' or ''controls''. The combined analysis of Arx-, glucagon-, and Pax4-producing cells in 1mDox+ IndGlu-ArxKO outlined a loss of Arx in approximately 8769% of glucagon-labeled cells, few of which initiated Pax4 expression ( Figure S3A-F), indicating that, in this inducible model, Arx can be efficiently inactivated in glucagon-producing cells.
Both Glu-ArxKO and Dox+ IndGlu-ArxKO were found viable and fertile, their life expectancy and basal glycemia remaining within normal range (Table S1-S2). Interestingly, a substantial increase in islet size was noted in both animal models (Table S1-S2). Further analyses indicated that a large insulin + cell hyperplasia was the reason underlying the observed islet hypertrophy ( Figure 1D-I, as compared to controls in Figure 1A-C). In Glu-ArxKO animals, a correlation between age and islet overgrowth was apparent, albeit a plateau phase was observed following 4 months of age (Table S1). Similarly, in Dox+ IndGlu-ArxKO, the degree of islet hyperplasia was found to Author Summary Type 1 diabetes is a condition that results from the loss of insulin-producing b-cells. Despite current therapies, diabetic patients are prone to vascular complications. Using the mouse as a model, we previously found that pancreatic glucagon-expressing cells can be regenerated and converted into b-like cells by the forced expression of a single gene, Pax4. Here, we generated transgenic mice allowing both the permanent labeling of a-cells and the inactivation of Arx solely in this cell subtype. Our results indicate that, upon Arx inactivation, a-cells can be continuously regenerated from duct-lining precursors and converted into b-like cells. Importantly, the additional loss of Pax4 does not impact these processes, suggesting that Arx is the main trigger of a-cell-mediated b-like cell neogenesis. Most interestingly, upon chemical induction of diabetes/b-cell loss, while control animals die or remain severely hyperglycemic, a normalization of the glycemia, a clear regeneration of the b-like cell mass, and an extended lifespan are noted in animals with the conditional inactivation of Arx. Our data therefore suggest that strategies aiming at inhibiting the expression of Arx, or its molecular targets/co-factors, may pave new avenues for the treatment of diabetes.
a-to-b-Like Cell Conversion upon Arx Inactivation depend on the duration of Dox treatment rather than on the age of Dox induction (Table S2). As important was the finding that, in both cases, a dramatic increase in islet number was apparent, suggestive of islet neogenesis (Table S1-S2). The increased number of islets and the insulin + cell hyperplasia were further demonstrated by means of optical projection tomography allowing the examination of insulin + cells in the entire pancreas ( Figure 1J-K, Movie S1). Indeed, a global pancreatic increase of 17169% in the number of insulin + cells was thereby outlined in 5 month-old Glu-ArxKO pancreata as compared to age-matched controls.
Taken together, our data suggest that the inactivation of Arx in adult a-cells (even of relatively advanced ages) results in a clear islet hypertrophy caused by an insulin + cell hyperplasia and a substantial increase in islet number.

Insulin + cells in Arx mutants display a b-cell phenotype
To ascertain the identity of the insulin + cells in Glu-ArxKO and Dox+ IndGlu-ArxKO pancreata, we assayed the expression of several endocrine cell marker genes. Our analyses demonstrated that all insulin + cells observed in Glu-ArxKO and Dox+ IndGlu-ArxKO  Figure S4S-X). These were found to lack the a-cell determinants, such as Brn-4 and glucagon, as well as somatostatin or PP (see hereafter and data not shown).
Since these results suggested that the insulin + cells observed in Glu-ArxKO and Dox+ IndGlu-ArxKO pancreata displayed a bcell phenotype, we sought to ascertain these observations using electron microscopy combined with insulin detection by immunogold labeling. Following the examination of more than 200 sections of pancreas per animal, it appeared that, in both genotypes, all insulin-producing cells displayed a b-cell ultrastructure ( Figure 3C Figure S5B). Altogether, our data indicate that the deletion of Arx in a-cells at any age leads to a substantial increase in the insulin + cell content, such cells displaying most features of true b-cells. In addition, non-b-cells display an abnormal localization within the islet, close to ducts.

a-cells are converted into b-like cells upon Arx inactivation
To further characterize the effects of the inactivation of Arx in acells, we sought to trace their lineage through the analysis of Glu-ArxKO::Rosa pancreata combining X-Gal staining and immunohistochemical approaches ( Figure 5A-F). In Glu-Cre::Rosa control pancreata, b-gal-labeled cells were solely found in the mantle zone of the islets where glucagon + cells typically reside ( Figure 5A). Importantly, in Glu-ArxKO::Rosa pancreata, such labeling was Interestingly, a similar analysis performed in IndGlu-ArxKO::Rosa animals treated with Dox for one month also outlined numerous insulin + b-gal + cells ( Figure 5G-I). A quantitative analysis revealed that approximately 30 to 40% of neo-generated b-like cells passed through a glucagon-expressing transition phase in both models (Table S3). While impressive, this contribution is lower compared to the initial labeling of a-cells (approximately 70 to 80% - Figure S2, S3), suggesting the involvement of additional processes. Of note, the co-detection of b-gal with glucagon or somatostatin expectedly revealed glucagon + b-gal + cells while all somatostatin + cells were found negative for b-gal ( Figure 5J-L), suggesting that the supplementary somatostatin + cells do not derive from cells having expressed the glucagon hormone. Together, our results provide conclusive evidence that a-cells of different ages can be converted into insulin + cells upon the sole inactivation of the Arx gene.
While the conversion of a-cells into b-like cells is of interest, it cannot account for the dramatic b-like cell hyperplasia noted in our animal models, nor for the continuous detection of glucagon + cells despite their conversion. Hence, to gain further insight into the mechanisms underlying these processes, we assayed proliferating cells combining BrdU labeling and immunohistochemical detection of the KI67 marker. Importantly, in 2mDox+ IndGlu-ArxKO pancreata of animals subjected to a long-term 10-day BrdU pulsechase, we observed a thickened ductal epithelium (insets in Figure 6B-C, E-F, H-I), most of the BrdU + cells being detected in some epithelial cells and in the adjacent ductal lining while a few appeared located at a pole of the islet where non-b-cells are detected ( Figure 6A-I). Similar observations were made using KI67 labeling in Glu-ArxKO animals, even in mice as old as 13 months of age ( Figure 6J-L). We therefore investigated whether Ngn3 could be re-expressed in animals lacking Arx in a-cells. Indeed, Ngn3 was previously found to be re-expressed in duct-lining cells in animals that underwent pancreatic duct ligation [31,32] or with the forced misexpression of Pax4 in a-cells [23,24]. Additional work suggested that Ngn3 + cells could be continuously generated and converted into endocrine cells [23,24,32]. Hence, we assayed Glu-ArxKO pancreata for Ngn3 using immunohistochemistry. Ngn3 has been previously detected in endocrine cells being expressed at very low levels [33]: we confirmed these observations ( Figure 7A, D). However, a much stronger expression of Ngn3 was noted mainly in cells located in the pancreatic ductal lining of animals lacking Arx in a-cells ( Figure 7B-C, E-F). A time course analysis of Ngn3 + cells in pancreata of IndGlu-ArxKO animals subjected to increasing durations of Dox treatment outlined Ngn3 + cells within the ductal epithelium and lining, as well as in islet cells located close to ducts, as early as 10 days post-Dox treatment initiation, a similar trend being detected 11 days later ( Figure 7G-I). Importantly, the downstream target of Ngn3, Rfx6 [34], was also found ectopically expressed in the ductal lining ( Figure 7J-L) of Glu-ArxKO animals. These results suggesting a reactivation of, at least, a part of the endocrine differentiation program, we queried the mechanisms triggering such processes. One possibility could be the loss of a-cells, and the resulting shortage in glucagon, provoked by their conversion into b-like cells. To verify this hypothesis, we initiated long-term glucagon supplementation experiments in IndGlu-ArxKO animals. Interestingly, while Dox+ IndGlu-ArxKO pancreata showed a significant augmentation of the insulin + or somatostatin + cells mass following 3 weeks of treatment with saline and Dox (as compared to controls - Figure 8A), animals treated with glucagon and Dox exhibited a diminished increase in insulin + , glucagon + , and somatostatin + cell counts ( Figure 8A). Taken together, these results provide evidence that, upon a-cell-mediated Arx deficiency, a-cells are converted into b-like cells. This, in turn, induces glucagon shortage-dependent regeneration processes characterized by the continuous proliferation of duct-lining cells, the reexpression of the developmental factors Ngn3 and Rfx6, and the compensatory neogenesis of endocrine cells, neo-formed glucagon + cells acquiring a b-like cell identity upon glucagon expression and subsequent Arx inactivation. The sole inactivation of Arx can induce the regeneration of a functional b-like cell mass following chemically-induced diabetes In a continuous effort to ascertain the identity of the supplementary insulin + cells in animals with an a-cell-specific inactivation of Arx, numerous tests were performed. It is important to reiterate that both Glu-ArxKO and Dox+ IndGlu-ArxKO animals exhibited normal basal glycemia levels (Tables S1-S2). However, upon intraperitoneal glucose tolerance test, these were found to display an improved response with a lower rise in a-to-b-Like Cell Conversion upon Arx Inactivation glycemia as compared to control animals ( Figure 8B-C). This suggests an improved capacity to release insulin upon glucose stimulation. Indeed, when we combined such glucose challenges with insulin measurements, while the basal insulin levels were found within normal range, a dramatic increase in circulating insulin content was noted when compared to controls ( Figure 8B-C Tables). Our results therefore support the notion of an increased b-like cell mass able to respond to a glucose bolus. Importantly, upon insulin challenge and despite the augmented numbers of b-like cells, the resulting glycemic level variations were found similar in Glu-ArxKO, Dox+ IndGlu-ArxKO, and control animals ( Figure 8D-E). This indicates that, despite a b-like cell hyperplasia, such animals do not exhibit any insulin resistance.
To determine whether neo-formed insulin + cells could functionally replace endogenous b-cells, we injected IndGlu-ArxKO animals with a high dose of streptozotocin to obliterate the endogenous b-cell mass. Two groups of IndGlu-ArxKO animals, and matching controls, were used: the first (''long pre-treatment'') a-to-b-Like Cell Conversion upon Arx Inactivation corresponding to animals for which Dox treatment was initiated 1.5 months prior to, and the second (''short pre-treatment'') 0 to 10 days prior to, streptozotocin administration. Dox treatment was continued for both groups until examination ( Figure 8F-G, respectively). All control animals developed extreme hyperglycemia leading to the death of the majority of these ( Figure 8F-G). Interestingly, all Dox+ IndGlu-ArxKO animals also developed a strong hyperglycemia but none died ( Figure 8F-G). As important was the observation that, following this hyperglycemic phase, a progressive return to normal levels was outlined ( Figure 8F-G). It is interesting to note that a return to euglycemia was also evident when Dox-mediated Arx inactivation in a-cells was triggered as late as on the day of streptozotocin injection ( Figure 8G), suggestive of efficient/rapid mechanisms allowing the restoration of an appropriate glycemic balance.
By immunohistochemical analyses, we observed a classical loss of b-cells shortly after streptozotocin treatment ( Figure 9A-B and data not shown). However, a clear neogenesis was outlined in these Dox+ IndGlu-ArxKO animals, resulting in the replenishment of the b-cell mass ( Figure 9B-E). Importantly, when the same  Figure 9C-E), suggesting glucagon + cell neogenesis prior to the acquisition of a b-like cell identity upon Arx inactivation. Taken together, these results further demonstrate that the loss of Arx in glucagon-producing cells is sufficient to induce their conversion into b-like cells. Our data also suggest that upon such a conversion, compensatory mechanisms are activated and lead to the neogenesis of endocrine cells among which glucagon + cells are again converted into b-like cells upon Arx inactivation. Importantly, this cycle of b-like cell regeneration is able to counter toxin-induced diabetes, such cells being able to maintain euglycemia.
The additional inactivation of Pax4 in Arx mutants does not impact the a-cell-mediated b-like cell neogenesis It was previously shown that the forced expression of Pax4 in glucagon + cells was sufficient to induce their neogenesis and conversion into b-like cells [23,24]. Here, we show that, in fact, the inactivation of Arx initiated in embryonic, but also in adult, a-cells is sufficient to induce a similar outcome. One may therefore conclude that the misexpression of Pax4 in a-cells could induce the  To determine the impact of the glucagon shortage provoked by the conversion of a-cells into b-like cells, 2.5 month-old IndGlu-ArxKO animals were supplemented (or not) with exogenous glucagon for 3 weeks as well as with Dox. While saline-treated animals developed significant insulin + and somatostatin + cell hyperplasia, as compared to controls, their glucagon-supplemented counterparts exhibited a diminished increase in insulin + and somatostatin + cell counts accompanied by a decrease in glucagon + cell number. n = 3 ** p,0.01, * p,0.07 using ANOVA comparing saline-and glucagon-treated animals. (B-G) 2.5 month-old Glu-ArxKO (and age/sex-matched controls) were subjected to an intraperitoneal glucose tolerance test (B). Mutant animals displayed an increased capacity to counteract the glucose bolus with a lower peak in glycemia, suggestive of an increased b-cell mass, further indicated by the augmented levels of down-regulation of Arx and thereby trigger a-cell mediated b-like cell neogenesis. However, the opposite could also be true, that is, that the deletion of Arx could promote such processes by upregulating Pax4. To discriminate between these two possibilities, we generated double-mutant animals allowing the conditional deletion of Arx and Pax4 specifically in a-cells. To achieve this purpose, we crossed ArxcKO animals with Pax4cKO animals (generated by knock-in of two LoxP sites within the Pax4 locus [35]). The resulting double transgenic animals were subsequently crossed with Glu-Cre mice to generate Glu-Cre::ArxcKO:: Pax4cKO animals (referred to as Glu-ArxKO/Pax4KO). Using immunohistochemistry on 6 month-old triple transgenic pancreata, most glucagon-producing cells were found to be negative for both Arx and Pax4 ( Figure 10A). Importantly, a number of insulin-producing cells were also found to lack Pax4, such cells most likely corresponding to a-cells converted into Arx 2/Y Pax4 2/ 2 b-like cells ( Figure 10B-C). Further examination of these triple transgenic pancreata by immunohistochemistry outlined, yet circulating insulin in Glu-ArxKO animals compared to their WT counterparts (Table inserted in B). Similar results were also evident in 1.2mDox+ IndGlu-ArxKO animals (C and Table inserted), where a faster reduction in glucose levels and return to euglycemia were observed compared to both WT and Dox-controls (C). The challenge of both transgenic lines with insulin resulted in no significant difference compared to their control counterparts (D-E), indicating that insulin remains fully active despite the b-like cell hyperplasia. IndGlu-ArxKO animals were subsequently subjected to streptozotocin treatment after 1.5 months (F) or, only 0-10 days (G) following Dox treatment initiation. In both cases, by the monitoring of glycemic levels, following an initial peak in glycemia, a steady recovery was noted in the induced animals, while controls either maintained their hyperglycemic state or succumbed to excessive glycemic levels. n.6 for all experiments *** p,0.001, * p,0.05 using ANOVA. doi:10.1371/journal.pgen.1003934.g008 Figure 9. The sole inactivation of Arx in a-cells can induce a regeneration of the b-like cell mass following streptozotocin treatment. 1.5mDox+ IndGlu-ArxKO::Rosa animals were treated with streptozotocin or sodium citrate (A) and their pancreatic composition was examined by immunohistochemistry. Next to the classical loss of b-cells upon high-dose streptozotocin treatment (data not shown), few insulin-producing cells could be detected as early as 10 days post-streptozotocin treatment (B). Interestingly, after 100 days of Dox treatment, the islets in streptozotocintreated animals appeared reformed and displayed b-gal + cells located in the core of the islet (C). Further examination demonstrated that such b-gal + cells were insulin + (D-E), suggesting that Arx inactivation in glucagon-expressing cells allows their regeneration and conversion to functional insulin + cells capable of counteracting the effects of streptozotocin. (Stz: Streptozotocin, each photograph is representative of at least three independent animals). doi:10.1371/journal.pgen.1003934.g009 a-to-b-Like Cell Conversion upon Arx Inactivation again, a substantial increase in the islet number and a clear islet hypertrophy caused by an insulin + cell hyperplasia that was found to be similar to the one observed in animals with Arx deletion ( Figure 10D compared to 1E-H). Quantitative analyses confirmed this augmentation in insulin + cell numbers ( Figure 11A-E), but also in the content in somatostatin + cells, non-b-cells being again found preferentially located close to ducts within the islets ( Figure 10E-I). Of note was the observation that, despite the lack of Pax4 in a number of b-cells, no alteration in basal glycemia of 4 month-old double-mutants could be detected as compared to controls (12767 mg/dl and 12164 mg/dl, respectively). Interestingly, upon glucose challenge, Glu-ArxKO/ Pax4KO animals displayed a significantly improved response as previously seen in Glu-ArxKO mice ( Figure 11F), suggestive of an increased functional b-like cell mass. Altogether, our analyses indicate that the combined loss of Arx and Pax4 in glucagonproducing cells results in a phenotype similar to that of Arx mutants, sustaining the notion that Arx represents the main player involved in a-cell-mediated b-like cell neogenesis processes.

Discussion
In this study, we report that the inactivation of the Arx gene in pancreatic a-cells at different ages results in hypertrophic islets mainly composed of cells displaying a b-cell phenotype. Our data support the notion of a conversion of a-cells into b-like cells upon the sole inactivation of Arx. These processes trigger compensatory mechanisms, associated with the re-expression of Ngn3 and Rfx6 in duct-lining cells, resulting in the neogenesis of endocrine cells among which glucagon + cells are subsequently converted into blike cells upon Arx inactivation. Importantly, the newly-formed blike cells are functional and can reverse chemically-induced diabetes. As interesting was the finding that the additional loss of Pax4 does not impact these mechanisms, suggesting that Arx represents the main inducer of a-cell-mediated b-like cell neogenesis.

Arx inactivation induces a-cells to acquire a b-like cell identity
We initially analyzed two transgenic mouse models allowing either (1) the constitutive inactivation of Arx in all glucagonproducing cells as soon as they initiate hormone expression (that is, during embryogenesis), or (2) the inducible loss of Arx in a-cells at different ages. In both instances, the inactivation of Arx was found to be fairly efficient with 70-80% of glucagon + cells appearing Arxdeficient. Interestingly, a number of Arx 2/Y glucagon + cells were found to ectopically express the b-cell-specific gene Pax4. As Pax4 and Arx were previously found to mutually inhibit each other's transcription during the course of pancreas morphogenesis and endocrine cell fate allocation [21], it is conceivable that the loss of Arx may result in Pax4 reactivation in adult a-cells. This would suggest that Arx acts in a-cells to maintain their phenotype through the repression of Pax4 expression. Supporting this notion were the phenotypic alterations observed in both models reminiscent of those found in animals with the ectopic expression of Pax4 in glucagon-producing cells [23,24]. Indeed, a progressive insulin + cell hyperplasia/islet hypertrophy and a substantial increase in islet numbers were noted. However, unlike in animals constitutively misexpressing Pax4 in glucagon + cells [24], both Glu-ArxKO and Dox+ IndGlu-ArxKO mice displayed a normal life expectancy, a normal basal glycemia, and a restricted increase in overall islet/insulin + cell populations. These discrepancies can most likely be attributed to differences in the efficiency of the transgene expression, but also, in the case of Dox+ IndGlu-ArxKO, to the fact that Arx deficiency is triggered in acells at older ages. A recent report by Wilcox et al. [36] using a similar approach, and mostly focusing on embryogenesis, is in agreement with the notion of a-to-b-like cell conversion upon Arx inactivation, although milder phenotypic alterations were observed, most likely due to low efficiency in Cre-mediated recombination processes. However, an induction of Arx inactivation in the adult context did not result in any cell conversion, most likely due to the very short induction times and the use of a non-specific Cre expression system. Herein, using lineage tracing, we demonstrate that the loss of Arx in a-cells induces their conversion into cells displaying most features of true b-cells, as outlined through marker gene analyses, electron microscopy examination, and functional assays. Such plasticity of a-cells had been previously suggested using alternative approaches, but never using gene inactivation [24,37,38]. It is interesting to note the similarity in phenotypic alterations in Glu-ArxKO mice and IndGlu-ArxKO animals treated with Dox at different ages. Indeed, this indicates that aging is not a limiting factor in this conversion process, or in the resulting insulin + cell neogenesis. In other words, our data demonstrate that adult a-cells, having been subjected to aging and environmental signals, retain their potential for conversion into b-like cells by the selective inhibition of a single gene, Arx.

Origin of the neo-generated b-like cells
While we demonstrated a conversion of glucagon-producing cells into b-like cells, we also consistently observed the continued presence of glucagon + cells in both animal models. In addition, quantification analyses did not show significant variations in their content as compared to controls. Of note, such glucagon + cells were found to be PC1/3 2 GLP1R 2 , indicating that they display a mature a-like cell phenotype. However, the lack of additional a-cell markers calls for caution concerning the degree of maturity of such glucagon + cells. The continued detection of glucagon + cells suggests that neogenesis processes are activated in order to compensate for the loss of the converted a-cells. Accordingly, the glucagon shortage provoked by this conversion was found to contribute to this neogenesis, as seen in animals misexpressing Pax4 in glucagon + cells [24]. Interestingly, numerous proliferating cells were observed in Arx mutant pancreata. These were mostly detected in the ductal lining instead of within the islet. In addition, cells re-activating the expression of the developmental and pro-endocrine gene Ngn3 were also detected in the same location as seen in animals that underwent pancreatic duct ligation [31,32] or with forced misexpression of Pax4 in a-cells [23,24]. Along the same line, an Ngn3 target, Rfx6, was also found ectopically expressed in the same location, suggesting a recapitulation of, at least, a part of the endocrine differentiation program. Whether known additional developmental targets of Ngn3, such as IA1 or Myt1, are also reactivated is a challenging question. Indeed, Ngn3 + cells appear to quickly migrate from the ductal compartment to the islet in which most endocrine cells are positive for IA1 or Myt1. Thus, while our data support the notion of the existence of a putative pancreatic precursor cell niche, the exact nature and potential of the Ngn3 reexpressing cells remains to be determined. Such cells do not appear to be stem-like cells as, using qPCR, we could not detect any variation in the transcripts of the classical stem-cell markers genes (such as Oct4 or Nanog -data not shown). However, our results suggest that such Ngn3 + cells have the potential to adopt any of the four endocrine cell lineages. Altogether (see model in Figure 11G), our analysis of a-cell-specific Arx mutants supports the notion that, upon the conversion of a-cells into b-like cells, compensatory mechanisms are activated and result in the proliferation of duct-lining cells, the re-expression of Ngn3, as well as the ectopic expression of Rfx6, and the acquisition of an endocrine cell identity. Among these neo-generated islet cells, glucagon + cells are further converted into b-like cells upon Arx inactivation, such a cycle resulting into b-like cell hyperplasia. The validation of this concept would require the inducible tracing of the lineage of Ngn3-, Rfx6-, HNF1b, and/or Ptf1a-expressing cells. However, having employed the Cre/Lox system to inactivate Arx in these animal models, this strategy could not be used to follow the progeny of these different cell subtypes.
While lineage tracing experiments allowed us to conclude that continuously neo-generated glucagon + cells can be converted into b-like cells upon Arx inactivation, the initial 70-80% labeling of acells did not fully match the subsequent 30-40% marking of b-like cells. One could therefore ponder the origin of neo-generated bgal 2 supplementary b-like cells. As Ngn3 is re-expressed in cells of the ductal lining and since some of the endocrine cell developmental pathways appear to be reactivated, one could expect Ngn3 + cells to mainly adopt a b-like cell fate as seen during pancreas morphogenesis, such cells would thus be b-gal 2 . Another  Figures S2-S3). Interestingly, a number of insulin + cells appeared Pax4 2 , such cells most likely deriving from Arx 2/Y Pax4 2/2 glucagon + cells (A-C). As noted in Glu-ArxKO and Dox+ IndGlu-ArxKO pancreata, an increase in islet size compared to controls (Figure 1), caused by an insulin + cell hyperplasia, was observed in Glu-ArxKO/Pax4KO pancreata (A-I). Non-b-cell endocrine hormone-expressing cells displayed a preferential localization at poles of the islets, adjacent to neighboring ducts (E-I), reminiscent of the phenotype of the sole inactivation of Arx in glucagon + cells. (Each photograph is representative of at least three independent animals). doi:10.1371/journal.pgen.1003934.g010 a-to-b-Like Cell Conversion upon Arx Inactivation Figure 11. Quantitative analyses upon the dual inactivation of Arx and Pax4 in glucagon-producing cells. (A-E) Quantitative comparison of the number of insulin-(A), glucagon-(B) and somatostatin-(C) expressing cells between 4 month-old Glu-ArxKO/Pax4KO animals and age-/sex-matched WT controls. A significant increase in the numbers of insulin-and somatostatin-expressing cells was observed in Glu-ArxKO/ Pax4KO animals, whilst no significant variation in glucagon + cells was noted. Interestingly, both islet count (D) and size (E) were significantly increased in these animals compared to their WT counterparts, suggesting a process of islet neogenesis in addition to an increased insulin + cell mass. (F) 4 month-old Glu-ArxKO/Pax4KO animals (and age/sex-matched WT controls) were challenged with glucose. Double-mutant animals displayed an increased capacity to counteract the glucose bolus with a lower peak in glycemia, suggestive of a functional increased b-cell mass. n$3 in all experiments, ** p,0.01, * p,0.05 using ANOVA. (G) Schematic detailing the consequences of Arx (and Pax4) inactivation triggered in a-cells. Following the inactivation of Arx (and Pax4), a-cells are converted into b-like cells (1-2). The resulting shortage in glucagon (and/or putative additional signals -3) promotes the proliferation of duct-lining cells, some of which re-express the developmental factor Ngn3. Our results indicate that such Ngn3 + cells adopt an endocrine cell identity, the glucagon + cell fate being clearly favored (4). Whether neo-generated somatostatin + cells contribute to the supplementary b-like cell mass remains to be determined (''?''). Similarly, one could assume that Ngn3 + cells could directly give rise possibility concerns somatostatin + cells. Indeed, we observed a large increase in their number in both animal models, such an increase not progressing over time; thus, one could hypothesize a continuous neogenesis of these cells prior to a putative conversion into b-like cells, as seen with glucagon + cells. Lastly, one could also hypothesize a contribution of acinar cells as elegantly demonstrated by Pan et al. using Ptf1a lineage tracing in animals that underwent pancreatic duct ligation [31]. As mentioned previously, the determination of the exact mechanisms involved would be complicated as the tracing of Ngn3 + /HNF1b + /Ptf1a + cells would, yet again, be required. Similarly, tracing somatostatin + cells is, to the best of our knowledge, currently impossible as a working Somatostatin-Cre mouse line still remains to be generated. It is worth noting that, upon streptozotocin treatment, a large proportion of the regenerated b-like cells appeared to have transitioned through a glucagon-expressing cell phase, suggesting that additional mechanisms/facultative precursor cells could be mobilized upon such severe b-cell depletion. Yet again, the determination of the exact processes involved will have to await the generation of the proper mouse lines.

The inactivation of Arx in a-cells can reverse the consequences of toxin-induced diabetes
In a continuous effort to ensure the identity of the supplementary b-like cells observed in Glu-ArxKO and Dox+ IndGlu-ArxKO mice, we monitored their physiological parameters and performed functional assays, including glucose and/or insulin level measurements upon glucose or insulin challenges. In all cases, our results support the notion of an increased b-like cell mass capable of responding to different challenges. Of note, despite a b-like cell hyperplasia, these animals were found to exhibit a normal basal glycemia and did not display any insulin resistance, suggestive of a feedback loop allowing optimal regulation. Importantly, our analyses showed that such b-like cells could functionally replace their endogenous counterparts upon streptozotocin-induced b-cell ablation and thereby prevent animal death or chronic hyperglycemia. Interestingly, our data also suggest that such regenerative processes can lead to a replenishment of the b-cell mass, and thereby allow animal survival, even when triggered concomitantly with toxin-mediated hyperglycemia. This indicates a relatively efficient and rapid b-like cell neogenesis. Thus, our findings provide evidence that the sole loss of the Arx gene in a-cells is sufficient to initiate a continuous cycle of glucagon + cell neoformation and their conversion into b-like cells, and thereby lead to the regeneration a functional b-like cell mass, a concept of importance in the context of diabetes research.
Arx as a main player in b-like cell neogenesis As mentioned previously, in a number of Glu-ArxKO and Dox+ IndGlu-ArxKO glucagon-producing cells, we noted an ectopic expression of Pax4. While the apparently rapid a-to-b-like cell conversion did not allow us to observe this misexpression in all Arx 2/Y glucagon + cells, one could hypothesize that Pax4 reexpression in a-cells could, at least in part, contribute to their conversion into b-like cells as previously reported [24]. Thus, to conclusively determine whether the forced expression of Pax4 or the inactivation of Arx is at the origin of the conversion of glucagon + cells into b-like cells, we generated an animal model allowing the inactivation of both genes in glucagon + cells.
Interestingly, we observed phenotypic alterations similar to that of Arx-deficient or Pax4-misexpressing animals [24], including an insulin + cell hyperplasia, an augmentation in the islet number, and a preferential location of non-b-cells close to ducts. It is worth noting that a number of Pax4 2 insulin + cells were observed in these double-mutant pancreata, providing information on their origin and thereby indicating that these were derived from glucagon-producing cells. Thus, these results demonstrate that Pax4 is dispensable for these regeneration processes and allow us to conclude that Arx represents the main factor allowing the conversion of a-cells into b-like cells and their subsequent regeneration. We therefore propose that the development of therapies aiming at inactivating/inhibiting Arx in a-cells could potentially open new avenues for diabetes research and/or aid the design of efficient b-cell differentiation protocols in the context of stem cell research.

Ethics statement
All mouse work was conducted according to French ethical regulations. This project received the approval from our local ethics comity (NCE/2011-22).

Mouse manipulations
To characterize the effects of the invalidation of the Arx gene in pancreatic glucagon-producing cells, different transgenic lines were used. Firstly, using the Cre-LoxP system, Glu-Cre::ArxcKO animals were generated by the crossing of the previously described ArxcKO mouse line [25], generated by homologous recombination by inserting two LoxP sites around the second exon of the Arx gene, and the classical glucagon-Cre mouse line, described by Herrera et al. [26]. Lineage tracing experiments were achieved by the crossing of Glu-Cre::ArxcKO animals with the ROSA26-loxstop-lox-b-Gal mouse line [27] to generate Glu-Cre::ArxcKO::R-OSA26-b-Gal triple-transgenic mice. Secondly, taking advantage of the TetOn system (Clontech), double transgenic mice obtained by the crossing of Glu-rtTA and TetO-Cre mice [23], were further mated with the ArxcKO::ROSA26-b-Gal mouse line. In the resulting transgenic mouse line, upon Doxycycline treatment, Arx is specifically inactivated in glucagon-expressing cells whose lineage can be followed. Note that the analyzed animals were mostly males as Arx is located on the X chromosome. For best results, we mostly used animals homozygous for the other transgenes. Doxycycline (Sigma) was administered via the drinking water prepared freshly once a week at a concentration of 2 g/L. To assess cellular proliferation, animals were treated with BrdU (in drinking water) for 10 days prior to examination (1 mg/ml solution). To determine the effects of glucagon supplementation on islet size, mice were injected intraperitoneally twice daily with 5 mg of glucagon (Sigma) for 3 weeks. To analyze the inactivation of Pax4 in Arx mutants, Pax4 conditional knockout mice, previously described by Kordowich et al. [35], were crossed with Glu-Cre::ArxcKO animals (Glu-Cre::ArxcKO::Pax4cKO).

b-galactosidase-based lineage-tracing experiments
Pancreatic tissues were isolated and fixed for 30 min at 4uC in a solution containing 1% formaldehyde, 0.2% glutaraldehyde, 0.02% NP40. The tissues were dehydrated in 25% sucrose overnight at 4uC. Prior to sectioning, tissues were embedded in freezing medium. For b-galactosidase activity assessment, the tissues were washed in PBS and then incubated overnight at 30uC in staining solution (500 mM K 3 Fe(CN) 6 , 250 mM K 4 Fe(CN) 6 , 0,5 M MgCl 2 , 40 mg/ml X-gal in DMF).

Induction of streptozotocin-mediated diabetes
To induce hyperglycemia, STZ (Sigma) was dissolved in 0.1 M sodium citrate buffer (pH 4.5), and a single dose was administered intraperitoneally (100-200 mg/kg) within 10 min of dissolution. Diabetes progression was assessed by monitoring the blood glucose levels and/or survival rates of mice.

Challenges and blood glucose levels measurement
For challenge purposes, animals were fasted for 16-18 h and injected intraperitoneally with glucose (2 g/kg of bodyweight) or insulin (0.75 U/kg). Blood glucose levels were measured at the indicated time points post-injection with a ONETOUCH Vita glucometer (Life Scan, Inc., CA).

RNA analysis
RNA was isolated (RNAeasy, Qiagen) and cDNA synthesis (SuperScript choice system, Invitrogen) was performed according to the manufacturer's instructions. Quantitative RT-PCR was carried out using validated primers (Qiagen) and the QuantiTect SYBR Green RT-PCR Kit (Qiagen) following manufacturer's instructions. PCR reactions and detection were performed on a Mastercycler ep realplex cycler using GAPDH and HPRT1 as internal controls for normalization purposes.

Electron microscopy
For ultrastructural analyses, anesthetized mice were perfused transcardially with physiological serum, then with 2% glutaraldehyde in 0.1 M cacodylate buffer. Pancreas were dissected, immerged in fixative for hours, post-fixed for 2 h in 1% osmium tetroxide in 0.1 M cacodylate buffer and embedded in epoxy resin. Contrasted ultrathin sections (70 nm) were analyzed under a JEOL 1400 transmission electron microscope mounted with a Morada Olympus CCD camera.
For immuno-gold staining, 200 islets, isolated by collagenase (1 mg/ml) digestion, were fixed with 4% paraformaldehyde, 0.2% glutaraldehyde in 0.1 M phosphate buffer (PB) (pH 7.4) overnight at 4uC and were processed for ultracryomicrotomy according to a slightly modified Tokuyasu method [39]. Immunostainings were processed with an automated immuno-gold labeling system Leica EM IGL using guinea-pig anti-insulin primary antibody (1/500) and 15 nm colloidal gold conjugated protein AG (CMC, University Medical Center, Utrecht, The Netherlands). Sections were examined under a JEOL 1400 transmission electron microscope.
Optical projection tomography, volumetric quantification and image analysis OPT analysis was performed as previously described [40,41]. Each specimen was scanned using the Bioptonics 3001 OPT scanner with a resolution of 102461024 pixels and reconstructed with the NRecon version 1.6.1.0 (Skyscan) software. Quantification of the insulin-producing cell mass was undertaken using Imaris software (Biplane). Volumes were calculated by applying a ''find objects by intensity'' task to select voxels above a specified intensity. The intensity threshold value was manually determined for each image stack. All pancreata was scanned and analyzed blinded. Figure S1 Generation of Glu-ArxKO and IndGlu-ArxKO animals. (Left) ArxcKO animals (in which the second exon of the Arx gene is flanked by LoxP sites) were crossed with the Glu-Cre mouse line (generated using a transgene composed of the glucagon promoter driving the expression of the phage P1 Cre recombinase). The resulting double transgenics (referred to as Glu-ArxKO) were further crossed with ROSA26-LoxP-Neomycin Resistance-STOP-LoxP-b-gal animals (''Rosa'', containing a transgene encompassing the ubiquitous ROSA26 promoter in front of the neomycin resistance gene with a STOP codon flanked by LoxP sites and followed by the b-galactosidase cDNA) for lineage tracing purposes. (Right) In a second mating scheme, ArxcKO animals were mated to Glu-rtTA mice (containing a transgene composed of the rat glucagon promoter upstream of the reverse tetracycline-dependent transactivator) and further mated with TetO-Cre animals (whose transgene includes the Tet operator upstream of Cre Recombinase cDNA). The resulting triple-transgenic mice, referred to as IndGlu-ArxKO, were subsequently mated with Rosa animals for lineage tracing purposes (IndGlu-ArxKO::Rosa). (TIF) Figure S2 Validation of the Glu-ArxKO mouse line. (A-C) Aiming to further demonstrate the efficiency of the glucagonmediated expression of the Cre recombinase, Glu-Cre animals were initially crossed to Rosa mice. Using immunohistochemistry with antibodies raised against glucagon or b-galactosidase, we showed a clear co-expression of the b-galactosidase with the glucagon hormone. In fact, quantitative analyses showed that 72% of glucagon-producing cells were b-galactosidase + , further demonstrating the relatively high efficiency of the Cre recombinase expression in this cell subtype. (D-I) Pancreata of 2 month-old controls (D-F) and of Glu-ArxKO (G-I) animals were subjected to quantitative analyses using a co-detection of Arx, Glucagon and Pax4. While in control pancreata, 9763% of glucagon + cells were found labeled with Arx (D, F), only 3066% of glucagon + cells appeared Arx + in the Glu-ArxKO pancreata (G, I), suggesting an efficient deletion of Arx in approximately 70% of glucagonproducing cells. Interestingly, while Pax4 was not detected in acells (E-F) in controls, a small number of Arx 2 glucagon + cells were found to be Pax4 + (1269% of glucagon cells) in Glu-ArxKO pancreata (H-I), indicating an ectopic expression of Pax4 in glucagon + cells upon Arx deficiency. (J) Using qPCR, a significant 74% reduction in the Arx transcript content was noted in Glu-ArxKO as compared to controls. (TIF) Figure S3 Validation of the IndGlu-ArxKO mouse line. Pancreata of 3 month-old controls (A-C) and 1mDox+ IndGlu-ArxKO (D-F) were subjected to quantitative analyses using the codetection of Arx, Pax4 and glucagon. In the control pancreata, 9862% of glucagon + cells were found to be labeled with Arx (A-C), whilst only 1167% of glucagon + cells appeared to be Arx + in Dox+ IndGlu-ArxKO pancreata (D-F), suggesting an efficient deletion of Arx in approximately 90% of glucagon-expressing cells upon Dox treatment. Interestingly, though no expression of Pax4 was observed in glucagon + cells in control pancreata (B-C), a small number of Arx 2 glucagon + cells were found to be Pax4 + (1664% of glucagon + cells) in Dox+ IndGlu-ArxKO animals (E-F), suggesting an ectopic expression of Pax4 in such glucagon + cells upon Arx deficiency. Movie S1 OPT examinations. Pancreata from 5 month-old Glu-ArxKO animals (Bottom) and of age-/sex-matched WT controls [23] (Top) were stained with an anti-insulin antibody and visualized using Optical Projection Tomography to highlight the islet hypertrophy and increase in islet number upon Arx inactivation in glucagon-expressing cells.

(MOV)
Table S1 Assessment of the life expectancy, glycemic levels, islet size, and islet number in Glu-ArxKO animals. Glu-ArxKO mice were examined at the indicated ages. Life expectancy and basal glycemia (monitored weekly) were found within normal ranges, as compared to controls. Glu-ArxKO animals displayed a clear increase in islet size dependent on age, however this increase appeared to plateau at approximately 4 months of age. An increase in the islet number was also observed, suggestive of islet neogenesis, which peaked at an average of x1.98 compared to controls. (DOCX)

Table S2
Assessment of the life expectancy, glycemic levels, islet size, and islet number in IndGlu-ArxKO animals. The IndGlu-ArxKO mice were treated with Dox at different ages and were examined after the indicated lengths of Dox treatment (values sorted based on Dox treatment duration). Life expectancy and basal glycemia (monitored weekly) were found within normal ranges, as compared to controls, in all conditions analyzed. Interestingly, a near doubling (1.86) in islet size was observed after just 14 days of Dox with a steady increase in islet size dependent on length of Dox treatment until a plateau appeared to be reached after approximately 4 months of Dox treatment. No large variations were observed in islet number between the different durations of Dox treatment however, on average, a 1.9-fold increase was observed in Dox+ IndGlu-Arx animals compared to age-matched controls. (DOCX)

Table S3
Quantification of glucagon + cell-derived b-like cells. Using immunohistochemical analysis on Glu-ArxKO and Dox+ IndGlu-ArxKO pancreata, insulin 2 and insulin + b-gal-producing cells were manually counted and compared to controls. To determine the proportion of supplementary b-like cells expressing the b-gal tracer, the increase in b-like cell number was factored in using the following formula: (% of labeled b-cell6b-like cell increase)/(b-like cell increase -1). Our data demonstrate a similar contribution (between 30 to 41%) of a-cells to the supplementary b-like cell count across models and indicated conditions. (DOCX)