Vascular endothelial growth factor coordinates islet innervation via vascular scaffolding

Coordinated vascularization and innervation is often a crucial requirement for organogenesis and normal physiological function, but the mechanisms driving these processes during embryonic and postnatal development are far from understood. In many organs, vascular and neural structures are closely aligned (Bearden and Segal, 2005; Correa and Segal, 2012; Mukouyama et al., 2005; Mukouyama et al., 2002; Stubbs et al., 2009), reflecting their interdependent existence (Quaegebeur et al., 2011) and synchronized function (Storkebaum and Carmeliet, 2011). Defining how these vascular and neural elements are patterned is important for a full understanding of organ formation, as well as pathological conditions.

Multiple mechanisms have been proposed to direct neurovascular development. First, vessels and nerves can independently respond to the same signaling factors, including angiogenic factors (VEGF and ephrin families), neurotrophic factors [nerve growth factor (NGF) and glial cell line-derived neurotrophic factor (GDNF) families], and axon guidance molecules (netrins, semaphorins and slits) (reviewed by Carmeliet and Tessier-Lavigne, 2005). For example, large blood vessels and nerves in the developing quail forelimb follow the same semaphorin 3A gradient, without any interdependence between the two structures (Bates et al., 2003). Alternatively, the cells that ultimately form blood vessels (Damon et al., 2007; Honma et al., 2002) and nerve fibers (Mukouyama et al., 2002; Mukouyama et al., 2005) may also release reciprocal guidance cues in a process termed mutual guidance.

The pancreatic islets, miniature endocrine organs that coordinate whole-body glucose homeostasis, are both highly vascularized (Bonner-Weir and Orci, 1982; Brissova et al., 2006; Lammert et al., 2003a) and richly innervated (Ahrén, 2000; Chiu et al., 2012; Rodriguez-Diaz et al., 2011a). Intra-islet capillaries are thicker, denser and more tortuous than capillaries of the exocrine pancreas, and are highly fenestrated (Brissova et al., 2006; Lammert et al., 2003b), allowing for the rapid exchange of nutrients and hormones between the islet and bloodstream. Input from the nervous system fine-tunes islet hormone secretion (Ahrén, 2000), regulates islet blood flow (Atef et al., 1992; Jansson and Hellerström, 1986), and influences β cell mass (Imai et al., 2008; Lausier et al., 2010; Nekrep et al., 2008; Plank et al., 2011). To acquire their specialized vasculature, islet endocrine cells produce angiogenic factors, such as vascular endothelial growth factor A (VEGF), which is crucial for islet vascularization, revascularization and function (Brissova et al., 2006; Cai et al., 2012; Lammert et al., 2003a; Magenheim et al., 2011). Although several recent studies have examined the developmental timing of pancreatic innervation (Burris and Hebrok, 2007; Nekrep et al., 2008; Plank et al., 2011), very little is known about the molecular mechanisms that direct this process, and whether the mechanisms controlling islet vascularization and innervation are interdependent or distinct.

Here, we used genetic mouse models to define the mechanisms that direct neurovascular development within pancreatic islets, with the initial hypothesis that VEGF derived from islet endocrine cells is also a neurotrophic factor. Although formation of the intra-islet vasculature occurs concomitantly with endocrine cell clustering during embryogenesis, the expansion and integration of nerve processes into the islets was significantly delayed until birth, and was completed by the time of weaning. Inactivation of VEGF early in pancreas organogenesis rapidly and dramatically reduced islet vascularization and impaired postnatal nerve fiber ingrowth. By contrast, β cell-specific overexpression of VEGF caused rapid hypervascularization and hyperinnervation of the islet, associated with increased production of extracellular matrix (ECM) components and axon guidance molecules by the intra-islet endothelium. Surprisingly, intra-islet nerves did not express VEGF receptors, but instead were dependent on the VEGF-mediated patterning of the intra-islet vasculature. Collectively, these data demonstrate a new role for VEGF in directing islet innervation.

Although it is well established that adult pancreatic islets are densely vascularized (Brissova et al., 2006; Lammert et al., 2003a) and richly innervated (Ahrén, 2000), little is known about the spatial and temporal relationship of intra-islet capillaries and nerves and the signals guiding their development. To resolve the neurovascular architecture in adult islets, we fluorescently labeled pancreatic sections using markers for nerve fibers (neuronal class III β-tubulin, TUJ1; TUBB3 - Mouse Genome Informatics) and endothelial cells (platelet endothelial cell adhesion molecule 1, PECAM1) (Fig. 1). Confocal microscopy analysis indicated that the majority of TUJ1⁺ fibers was either in a complete or partial alignment with PECAM1⁺ capillaries (Fig. 1A′, closed arrowheads), but some TUJ1⁺ fibers were not adjacent to endothelial cells (Fig. 1A′, open arrowheads). A similar relationship of nerve fibers and intra-islet endothelial cells was confirmed by an additional pan-neuronal marker, synapsin (supplementary material Fig. S1A-A′).

To determine how changes in islet VEGF production affect islet innervation and the relationship of intra-islet capillaries and nerve fibers, we used Pdx1-Cre;Vegfa^fl/fl (abbreviated VEGF^Down) mice, in which VEGF is genetically inactivated throughout the pancreas during embryogenesis, resulting in a nearly 90% decrease in islet vascularization (Lammert et al., 2003b; Reinert et al., 2013). To increase islet vascularization, we used a Tet-on inducible system, in which treatment with doxycycline (Dox) induces expression of VEGF in insulin⁺ β cells (Cai et al., 2012). We treated adult RIP-rtTA;TetO-hVegfa (abbreviated VEGF^Up) mice with Dox for one week, which led to an increase in VEGF secretion, a dramatic expansion of intra-islet endothelial cells, recruitment of macrophages, and a reduction in β cell number (Brissova et al., 2014). Compared with littermate controls (Fig. 1A-A′), islets in adult VEGF^Down mice showed reduced innervation (Fig. 1B-B′), as measured by a 52% reduction in the number of TUJ1⁺ nerve fibers present within the insulin⁺ area of the islet (Fig. 1D), as well as a 50% reduction in the length of those fibers (Fig. 1E). By contrast, hypervascularized islets in VEGF^Up mice were more highly innervated, with nerve fibers more closely associated with endothelial cells than with β cells (Fig. 1C-C′). VEGF-overexpressing islets showed a 23% increase in the number of TUJ1⁺ nerve fibers (Fig. 1D) and a 29% increase in fiber length (Fig. 1E). The changes in islet innervation in VEGF^Down and VEGF^Up mice were further confirmed using synapsin labeling (supplementary material Fig. S1). Taken together, these data indicate that the abundance of islet innervation is closely related to the degree of islet vascularization. This suggests that islet innervation may be regulated directly by islet endocrine cell-derived VEGF or signal(s) from intra-islet endothelial cells.

Mouse pancreatic islets are primarily innervated by autonomic nerves (Ahrén, 2000; Rodriguez-Diaz et al., 2011a). To determine whether the changes in islet innervation following altered VEGF expression selectively affected sympathetic or parasympathetic nerve fibers, we labeled pancreata from both VEGF^Down and VEGF^Up mice (and their respective controls) for tyrosine hydroxylase (TH) and vesicular acetylcholine transporter (VAChT).

Control islets contained many TH⁺ sympathetic nerve fibers and also a few TH-expressing β cells (Fig. 2A). Surprisingly, VEGF^Down islets contained few TH⁺ fibers (Fig. 2B), but the number of TH-expressing β cells dramatically increased. In VEGF-overexpressing islets, TH⁺ β cells were rare, but these islets had an increased abundance of TH⁺ fibers (Fig. 2C) compared with controls. VAChT labeling showed that changes in the density of parasympathetic nerve fibers in VEGF^Down and VEGF^Up islets also coincided with islet VEGF production and islet vascular density (Fig. 2D-F). Regardless of islet VEGF expression, vascularization or innervation status, we did not find evidence of VAChT labeling in islet endocrine cells, in contrast to previous observations in human pancreatic islets (Rodriguez-Diaz et al., 2011b; Rodriguez-Diaz et al., 2011a). These data suggest that islet VEGF expression determines the extent of both islet sympathetic and parasympathetic innervation.

To follow the establishment of islet innervation and determine when VEGF is important for this process, we assessed several stages throughout pancreas and islet development. Prior lineage-tracing analysis in the Wnt1-Cre;R26R^YFP/+ system demonstrated that neural crest cells enter the pancreatic primordium at approximately embryonic day (E) 10.0, and neuronal processes arrive at developing islets as endocrine cell clusters begin delaminating from the pancreatic epithelium around E13.5 (Plank et al., 2011). However, it is not known when nerve fibers expand into islet clusters and whether this process is coordinated with development of islet vasculature.

Confocal microscopy analysis of whole-mount pancreas at E14.5 revealed that developing islets (visualized by labeling for glucagon⁺ and insulin⁺ endocrine cells) were not only in contact with neural crest derivatives but were also interconnected by the neuronal network traveling between endocrine cell clusters (Fig. 3; supplementary material Movies 1A,B; Fig. S2). This observation brings a new insight into how autonomic nerves may coordinate and synchronize hormone secretion from hundreds of islets dispersed throughout the gland, allowing for oscillations of islet hormone secretion, and optimizing islet hormone secretory output during metabolic stress (Ahrén, 2000). However, nerve fibers remained at the periphery of PDX1^high-expressing islet clusters at E16.5 (Fig. 4A; supplementary material Movie 3) and E18.5 (supplementary material Fig. S2C).

Global pancreatic inactivation of VEGF in VEGF^Down embryos differentially affected the vascular and neuronal tissue compartments. Whereas VEGF loss led to a dramatic reduction in pancreatic vascularization as early as E14.5, the density of TUJ1⁺ neuronal processes and their association with developing endocrine clusters (marked by PDX1^high expression) were very similar in mutants and controls (Fig. 4B; supplementary material Movie 4; Fig. S2A-D), indicating that VEGF is not required for the establishment of pancreatic innervation or the migration of nerve processes to developing islets. By contrast, VEGF overexpression by β cells in VEGF^Up embryos resulted in greatly increased VEGF production in the vicinity of epithelial tubes (Cai et al., 2012), leading to overall pancreatic hypervascularization (Fig. 4C; supplementary material Movie 5). Three-dimensional reconstructed images of E16.5 pancreata from VEGF^Up embryos demonstrated that hypervascularized pancreatic regions were also highly innervated (Fig. 4C; supplementary material Movie 5), suggesting that although VEGF and/or endothelial cells are not required for the development of pancreatic innervation, they greatly enhance and facilitate this process.

As pancreatic innervation in VEGF^Down embryos was not affected by reduced pancreatic vascularization, we asked if nerves in the developing pancreas conversely influence vascular development. To answer this question, we used the Wnt1-Cre;Foxd3^fl/- mouse model to ablate neural crest cells in the pancreas (Plank et al., 2011; Teng et al., 2008). Because these mice die perinatally, we evaluated pancreatic vascularization in embryos at E16.5 (supplementary material Fig. S3A,B). As expected, the pancreatic epithelium in Wnt1-Cre;Foxd3^fl/- embryos was depleted of TUJ1⁺ nerve fibers (supplementary material Fig. S3B). By contrast, PECAM1⁺ endothelial cells still formed a dense capillary network in Wnt1-Cre;Foxd3^fl/- embryos, and this was unchanged from controls (supplementary material Fig. S3C). This indicates that neural crest derivatives do not influence pancreatic vascularization. Moreover, these data suggest that the development of pancreatic nerves and blood vessels is guided by different signals.

Given that nerve processes in the embryonic pancreas were still localized at the periphery of developing endocrine clusters at E18.5, we sought to determine the time course of their expansion and integration into islets during postnatal development. At postnatal day (P) 1 and P7, the islet capillary network in controls (Fig. 5A; supplementary material Fig. S4A) resembled that of adult islets, but VEGF^Down islets were severely depleted of endothelial cells (Fig. 5B; supplementary material Fig. S4B). Surprisingly, TUJ1⁺ nerve processes continued to be localized at the islet periphery with occasional short projections between peripheral endocrine cells in islets of both genotypes (Fig. 5A,B; supplementary material Fig. S4A,B), indicating that formation of the intra-islet capillary network precedes islet innervation. In control islets, projections of nerve fibers extended into the insulin⁺ core by the time of weaning (P21) (Fig. 5D), but this process was arrested in VEGF^Down islets (Fig. 5E). In sharp contrast to control and VEGF^Down islets, VEGF^Up islets showed a dramatic increase in the number of PECAM1⁺ endothelial cells (Fig. 5C,F; supplementary material Fig. S4C) that disrupted typical β cell clustering owing to greatly increased endothelial cell mass (Cai et al., 2012). However, at each time point, these hypervascularized islets were highly innervated by a dense network of TUJ1⁺ nerve fibers (Figs 5C,F; supplementary material Fig. S4C). These data demonstrate that expansion of neuronal processes into islets occurs postnatally and that normal levels of islet VEGF are crucial for this process.

To further establish when the increase in TH⁺ β cells in hypoinnervated islets is first evident, we examined TH expression at multiple stages of development in VEGF^Down mice. At E14.5, E17.5 and P1, few insulin⁺ β cells were TH⁺ in control or VEGF^Down mice (supplementary material Fig. S5A-C,F-H). However, TH⁺ β cells became more abundant in VEGF^Down islets at P7 (supplementary material Fig. S5D,I) and at P21 (supplementary material Fig. S5E,J), which coincided with the timing when projections of nerve fibers failed to extend into their core (Fig. 5E,E′). By contrast, hypervascularized islets from VEGF^Up mice at P7 showed many TH⁺ fibers, but very few β cells expressed TH (supplementary material Fig. S5K). These results show that the number of postnatal β cells expressing TH is inversely correlated with islet innervation.

Because expansion of nerve processes into islets was dynamically controlled by islet VEGF production, we wanted to determine if VEGF directly regulates islet innervation. In cells of the nervous system, VEGF signaling is mediated through VEGF receptor 2 (VEGFR2; KDR - Mouse Genome Informatics) and neuropilin 1 (NRP1). To co-label these receptors with islet neuronal components (nerve fibers and Schwann cells), we used Wnt1-Cre;R26-EYFP mice, in which neural crest-derived cells are indelibly labeled with YFP (Plank et al., 2011; Teng et al., 2008). Although both VEGFR2 and NRP1 displayed strong colocalization with PECAM1⁺ capillaries in the islet (Brissova et al., 2006) (data not shown), YFP⁺ neural crest cells did not co-express either VEGF receptor in adult islets or postnatally, when nerve processes begin to grow into the islets (Fig. 6; supplementary material Fig. S6A-C,E). A few YFP⁺ neural crest cells transiently expressed NRP1, but not VEGFR2, at E16.5 (supplementary material Fig. S6D,F, arrowheads). Therefore, these data indicate that VEGF regulates islet innervation indirectly, through its effects on intra-islet endothelial cells.

Because formation of the intra-islet vasculature was crucial for the development of normal islet innervation patterns, and because islet hypervascularization enhanced islet innervation, we reasoned that intra-islet endothelial cells were important mediators of islet neurodevelopment. We hypothesized that intra-islet endothelial cells could act on nerve fibers in two possible ways: by releasing neurotrophic or axon guidance factors and/or by forming a vascular scaffold for migrating nerve fibers. For example, NGF, a signal known to promote sympathetic innervation of the islet (Edwards et al., 1989), may be produced by endothelial cells (Cabrera-Vásquez et al., 2009; Gibran et al., 2003; Tanaka et al., 2004). Alternatively, the ability of a neuron to extend a process over a biological substrate requires a direct interaction with the ECM or basal lamina (Myers et al., 2011).

To identify transcripts of neurotrophic and axon guidance factors, and ECM molecules expressed by intra-islet endothelial cells, we performed transcriptome analyses of control islets, endothelial cell-enriched VEGF^Up islets and endothelial cells isolated from VEGF^Up islets. We found that members of several classic axon guidance families, such as slits, netrins, semaphorins and ephrins had significantly increased expression in VEGF^Up islets (Fig. 7A; supplementary material Table S5), and were highly abundant in the islet endothelial cell fraction (Slit3, Sema3f, Sema6d and Efnb2) (Fig. 7B; supplementary material Table S5). Prior studies have shown that, depending on biological context, these molecules may direct either axonal attraction or repulsion (Kolodkin and Tessier-Lavigne, 2011). A few axon guidance molecules that may serve as axon repellents (Efna5, Slitrk6) were downregulated in VEGF^Up islets, together with β cell-specific genes (Ins2, Nkx6-1, Pdx1, Pax6) (Fig. 7A), suggesting that these molecules may be β cell-derived.

Surprisingly, intra-islet endothelial cells did not express any known neurotrophic factors (Fig. 7A; supplementary material Table S5). However, we found that Bdnf (brain-derived neurotrophic factor) expression was appreciably increased in VEGF^Up islets and we detected a modest increase in Ngf (Fig. 7A; supplementary material Table S5). Because of prior work demonstrating NGF expression in developing islet endocrine and endothelial cells (Cabrera-Vásquez et al., 2009), we also performed quantitative RT-PCR analysis on VEGF^Up and VEGF^Down islets. Ngf expression was increased approximately threefold in VEGF^Up islets compared with control (supplementary material Fig. S7A) but was unchanged in endothelial cell-depleted VEGF^Down islets (supplementary material Fig. S7B), indicating that NGF is primarily an endocrine cell-derived factor. Furthermore, these data suggest that NGF produced by islet endocrine cells is not sufficient for ingrowth of nerve fibers into islets, and that this process requires intra-islet endothelial cells.

In addition to axon guidance molecules, VEGF^Up islets had upregulated expression of several components of the islet basement membrane (Nikolova et al., 2006), including integrins (Itga1, Itga6 and Itgb1), collagens (Col4a1 and Col4a2) and laminins (Lama4, Lamb1 and Lamc1) all of which were highly expressed in intra-islet endothelial cells (Fig. 7A; supplementary material Fig. S7A; Table S5). Moreover, immunohistochemical analysis showed that intra-islet nerve fibers were closely aligned with endothelial cells expressing collagen IV α1 and laminin during the postnatal period, when nerve fibers begin to penetrate islets, and in adult islets, when innervation is completed (Fig. 7C-D; supplementary material Figs S8-S11). Collectively, these data suggest that intra-islet endothelial cells contribute to islet innervation through synthesis of axon guidance molecules and the vascular basement membrane, which functions as a scaffold for nerve ingrowth.

Vascularization and innervation are crucial for normal organ physiology. Although neurovascular development is often synchronized, diverse mechanisms can achieve this physiologically important ‘goal state’. We elucidated the mechanisms directing neurovascular development in pancreatic islets, the normal function of which depends on dynamic intercommunication between the vascular and nervous systems. By increasing or decreasing local VEGF production during islet development and in mature islets, we showed that: (1) establishment of the intra-islet vasculature by VEGF is essential for the postnatal maturation of islet innervation; (2) intra-islet endothelial cells provide crucial signals for nerves through production of axon guidance molecules and ECM components; and (3) β cells may compensate for a lack of neuronal input via neuro-islet cell plasticity. Therefore, we propose a novel role for VEGF in coordinating neurovascular development within the pancreatic islet. In this organ, VEGF serves as the initial signal in an intra-islet signaling cascade in which endocrine cells establish the vasculature that supports islet endocrine cell development and provides a scaffold for the later ingrowth of nerve fibers to form essential autonomic nervous system connections.

Both VEGF inactivation and VEGF overexpression within pancreatic islets affected islet innervation. In the VEGF-deficient pancreas, the loss of intra-islet endothelial cells resulted in hypovascularized and hypoinnervated islets. Neural crest-derived cells did associate with these early-stage endocrine cell clusters, but differentiated nerve fibers failed to penetrate the islet core in the postnatal period because of the lack of an established, VEGF-directed intra-islet vascular plexus. Therefore, VEGF-directed formation of the intra-islet vasculature is required for the most essential aspects of final islet innervation. However, the role of VEGF in islet innervation is indirect, because the neural crest-derived cells did not express the VEGF receptors VEGFR2 or NRP1 at any stage analyzed during the relevant postnatal maturation period. By contrast, islet-specific VEGF overexpression led to dramatic hypervascularization and consequent hyperinnervation.

Our studies show that the recruitment of neural crest-derived cells to the pancreatic bud and the subsequent penetration of nerve fibers into the islet involve distinct stages that are differentially dependent on VEGF and vascularization. We propose a model in which VEGF and its patterning of islet vascularization direct the final maturation of islet innervation (Fig. 7E). During early organogenesis, endothelial cells are required for induction of the pancreatic buds starting at E8.5 (Lammert et al., 2001). Neural crest-derived cells migrate into the buds between E10.0 and E12.5 (Burris and Hebrok, 2007; Nekrep et al., 2008; Plank et al., 2011; Shimada et al., 2012). VEGF is essential for the initial recruitment of endothelial cells to the developing pancreas, and nascent endocrine cell clusters start to become vascularized and exposed to blood flow as early as E13.5 (Brissova et al., 2006; Shah et al., 2011). Neural crest-derived cells, traveling to the pancreatic anlagen from the foregut (Kirchgessner et al., 1992), associate with early islet clusters independently of VEGF signaling, in part recruited by exocrine pancreas-derived netrin signaling (Jiang et al., 2003). Although differentiated nerves associate with developing islet clusters between E13.5 and E15.5 (Burris and Hebrok, 2007; Nekrep et al., 2008; Plank et al., 2011; Shimada et al., 2012), potentially serving as a broad interconnecting system between the emerging islets, they remain localized at the periphery of vascularized islets at birth. Islet innervation reaches its mature state around the time of weaning, weeks after birth (Burris and Hebrok, 2007), when nerve fibers are finely interspersed within the vascularized islets.

The temporal asynchrony of islet innervation and vascularization fits the proposal that these processes are driven by sequential and independent mechanisms. Additionally, genetic disruption of pancreatic innervation did not affect patterning of the pancreatic vasculature, and did not result in any obvious deterioration of the preformed vascular network. Thus, pancreatic innervation differs from the neurovascular alignment process in limb skin, in which ingrowing peripheral nerves direct arterial differentiation and co-alignment of existing blood vessels through VEGF expression (Mukouyama et al., 2002; Mukouyama et al., 2005). In our model of coordinated islet neurovascular development, endocrine cell- and endothelial cell-derived factors work postnatally to signal nerve fibers to follow capillaries and the endothelial cell-derived vascular basement membrane into the islet core.

The requirement for vessels in islet innervation is similar to the interdependence seen in the avian gut, wherein the disruption of endothelial cells prevented migration of undifferentiated neural crest cells, ultimately impairing formation of the enteric nervous system (Nagy et al., 2009). In that study, signaling through β1 integrin, a receptor for ECM components, was central to the neural crest-derived cell-endothelial cell interaction. We postulate an analogous interaction in the islet: endothelial cells deposit basement membrane (Nikolova et al., 2006) as a scaffold for traveling nerve fibers.

Islet hypervascularization led to increased production of ECM components and hyperinnervation. These results are similar to a model of inducible angiogenesis in the rat mesentery, in which nerve growth slowly followed the pattern of new vessels, resulting in accurate neurovascular alignment down to the capillary level, though the mechanisms guiding this arrangement were unexplored (Stapor and Murfee, 2012). Transcriptome analysis of hyperinnervated VEGF^Up islets demonstrated an increase in the expression of several axon guidance molecules by intra-islet endothelial cells (Slit3, Sema3f, Sema6d and Efnb2) which could coordinate resulting neurovascular alignment (Kolodkin and Tessier-Lavigne, 2011). At the same time, decreased expression of β cell-derived axon repellent factors in VEGF^Up islets (Efna5) due to a concurrent decline in β cell population (marked by decreased expression of β cell genes) could contribute to their increased innervation (Frisén et al., 1998; Kolodkin and Tessier-Lavigne, 2011). Investigation of the role of specific ECM-axon interactions during axon migration into islets will require using mutants with disrupted expression of ECM or integrin molecules. However, these mutant models are not viable at the postnatal stage when the islet innervation takes place (Pöschl et al., 2004; Watt and Hodivala, 1994).

Our data also demonstrate the interesting finding that nerve fibers are closely associated with, yet do not penetrate, pancreatic islets during development and the early postnatal period, despite the presence of an established intra-islet vascular plexus. The reason for this remains unclear. The incomplete alignment of capillaries and nerve endings in adult islets suggests that although intra-islet capillaries form the structural scaffold for nerve fiber migration, endocrine cells are a likely source of neuroattractants that guide the very final stages of innervation, yielding the requisite meshwork of autonomic nervous system connections. Supporting this concept, purified β cells were reported to be more efficient than other islet endocrine cells in promoting reinnervation of islet cell grafts (Myrsén et al., 1996). Alternatively, it is possible that α cells are responsible for producing factors that establish their own neural connections, as the islet periphery is particularly innervated by sympathetic fibers (Chiu et al., 2012; Rodriguez-Diaz et al., 2011a). Islet endocrine and endothelial cells were previously shown to produce NGF throughout development (Cabrera-Vásquez et al., 2009), and experimental overexpression of NGF by β cells led to sympathetic hyperinnervation (Edwards et al., 1989). However, our transcriptome analysis demonstrated that Ngf is not produced by intra-islet endothelial cells. Moreover, Ngf expression was unchanged in VEGF^Down islets, which were depleted of endothelial cells and nerve fibers, indicating that NGF is primarily an endocrine cell-derived factor and is not sufficient for ingrowth of nerve fibers into islets in the absence of intra-islet endothelial cells.

VEGF-deficient islets also showed an increased number of β cells expressing TH, normally a marker of catecholaminergic nerves. In control islets, consistent with prior reports (Rodriguez-Diaz et al., 2011a; Teitelman et al., 1988), a small subset of β cells expressed TH. TH expression in β cells has several proposed, yet unconfirmed, roles: (1) a marker of endocrine precursor cells during early pancreas development (Alpert et al., 1988; Teitelman and Lee, 1987; Teitelman et al., 1993); (2) a marker of post-proliferative β cells on the path to senescence (Teitelman et al., 1988); or (3) an indicator of synthesis of endogenous islet catecholamines (Borelli and Gagliardino, 2001; Borelli et al., 2003). The number of TH⁺ β cells in VEGF-deficient islets begins to increase postnatally, when islet innervation should be reaching its mature state. Thus, we postulate that these hypoinnervated, hypovascularized islets detect a lack of neuronal input, such that some β cells begin to assume the role of catecholamine biosynthesis in compensation. Supporting this hypothesis is the recent observation that genetic ablation of sympathetic nerves during embryonic development leads to a dramatically increased number of TH⁺ β cells in postnatal islets (Borden et al., 2013). Normal β cells express several other neuronal markers (reviewed in Arntfield and van der Kooy, 2011) and can synthesize neurotransmitters such as dopamine and serotonin (Kim et al., 2010; Ustione and Piston, 2012). Similarly, human α cells express VAChT and produce acetylcholine as a paracrine signal to enhance islet hormone secretion (Rodriguez-Diaz et al., 2011b). Therefore, the increased number of TH⁺ endocrine cells in VEGF-deficient islets may be an indicator of neuro-islet cell plasticity (Ahrén et al., 2006), though further work is necessary to test this hypothesis.

The morphology of human islet innervation has only recently been explored (Rodriguez-Diaz et al., 2011b). Although human islets display fewer nerve fibers than mouse islets, those nerve fibers appear to be more intimately associated with intra-islet capillaries than endocrine cells. It was proposed that this arrangement allows islet neurotransmitters to signal first to arterial smooth muscle cells, thus allowing changes in blood flow to regulate islet function. Alternatively, islet neurotransmitters may be released into the bloodstream before they reach endocrine cells (Rodriguez-Diaz et al., 2011a). Because the close physical relationship between the intra-islet vascular and nervous systems is maintained in human islets, we predict that the vascular-mediated nerve patterning observed in mouse islets is a mechanism that is conserved in humans.

Collectively, our findings emphasize the importance of intercellular signaling networks in organ formation. A detailed understanding of these networks is essential for efforts to develop successful tissue regeneration and transplantation strategies.

Animal studies were approved by the Institutional Animal Care and Use Committee at Vanderbilt University Medical Center, and were performed in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. For all developmental studies, noon of the date of the observed vaginal plug was considered to be E0.5. Mice were considered adults at 10 weeks of age. Tail biopsies were obtained from all mice for genotyping by PCR using the primers listed (supplementary material Table S1).

For a model of reduced pancreatic VEGF expression (VEGF^Down), we crossed Pdx1-Cre mice (Gu et al., 2002) with mice in which exon 3 of the Vegfa gene is flanked by loxP sites (Vegfa^fl/fl) (Gerber et al., 1999). Unless otherwise stated, age-matched Vegfa^fl/fl mice are shown as controls. For a model of inducible overexpression of VEGF in β cells (VEGF^Up), we bred RIP-rtTA mice (Milo-Landesman et al., 2001) with TetO-hVegfa¹⁶⁴ mice (Ohno-Matsui et al., 2002). To induce VEGF expression, pregnant dams or adult mice were given light-protected drinking water containing 2 mg/ml doxycycline (Dox; Sigma-Aldrich) and 1% Splenda (zero-calorie sucralose sweetener) (Cai et al., 2012). For developmental studies, Dox was administered from E5.5. Adult mice were treated with Dox for 1 week. The Dox solution was freshly prepared every other day.

Neural crest-derived cells were genetically labeled in the pancreas using Wnt1-Cre mice (Danielian et al., 1998) crossed with R26-EYFP reporter mice (The Jackson Laboratory) (Srinivas et al., 2001). For a model of neural crest cell ablation, Wnt1-Cre;Foxd3^fl/- embryos were used (Plank et al., 2011; Teng et al., 2008).

Postnatal mice and pregnant dams were anesthetized with an intraperitoneal injection of a 90 mg/kg ketamine and 10 mg/kg xylazine solution (Henry Schein) before surgical removal of the pancreas. Pancreata were further dissected and washed in ice-cold 10 mM PBS, fixed in 4% paraformaldehyde in 1× PBS, cryoprotected in 30% w/w sucrose, and frozen in Tissue-Tek Optimal Cutting Temperature (OCT) compound (VWR Scientific Products) before cryosectioning. Cryosections (10-30 μm) were labeled using immunohistochemistry as described previously (Brissova et al., 2004), with the primary and secondary antibodies listed in supplementary material Tables S2 and S3. Slides were mounted with SlowFade Gold antifade reagent (Invitrogen) and sealed with fingernail polish before imaging.

For whole-mount imaging, embryos were dissected in 10 mM PBS to isolate the pancreas, stomach, duodenum and spleen. For each of the following steps, embryonic tissues were placed in cryogenic vials and rocked at 4°C. Tissues were fixed in 4:1 methanol:DMSO overnight, washed twice with 100% methanol, and stored in methanol at -20°C until further use. Before immunolabeling, tissues were rehydrated (washed once in 50% methanol for 30 minutes and twice in 10 mM PBS for 30 minutes each), then blocked and permeabilized in PBSBT (10 mM PBS with 2% bovine serum albumin and 0.5% Triton X-100) in two washes over two hours. On consecutive days, tissues were incubated overnight with primary or secondary antibodies diluted in PBSBT. After each antibody step, tissues were rinsed and washed three to four times with PBSBT for one hour each. Before imaging, immunolabeled tissues were dehydrated (washed once each in 50% and 80% methanol for 30 minutes, and washed twice in 100% methanol over one hour), optically cleared in a 1:2 solution of benzyl alcohol:benzyl benzoate (BABB; Sigma-Aldrich), and placed on a glass coverslip.

Tissue images were obtained using an Olympus BX41 fluorescence microscope, a Leica DMI 6000 B, a Zeiss LSM 510 META laser scanning confocal microscope and a ScanScope FL slide scanner (Aperio Technologies).

Pancreatic islets were isolated from adult mice as described previously (Brissova et al., 2004), and hand-picked to purity. RNA from 150-400 purified islets (from one to five mice) was extracted as described previously (Dai et al., 2012; Dai et al., 2013). Following extraction, RNA quality and purity was assessed by the Vanderbilt Genome Sciences Resource using a Nanodrop ND-1000 spectrophotometer and an Agilent 2100 Bioanalyzer. Only RNA with a RIN score of >7 was used further.

Quantitative RT-PCR was performed using a TaqMan primer-probe approach, using the primer-probes shown in supplementary material Table S4. Data were analyzed using the ΔΔC_t method (Livak and Schmittgen, 2001), as calculated by the iQ5 system software. Expression of each gene was normalized to Tbp (encoding TATA box binding protein) as a reference gene (Dai et al., 2012). The relative expression of Tbp-normalized genes in each sample from Pdx1-Cre;Vegfa^fl/fl mice was compared with each sample of age-matched Vegfa^fl/fl controls (three to four islet preparations per group), then averaged. Similarly, samples from one-week-Dox-treated RIP-rtTA;TetO-hVegfa mice were compared with those from untreated RIP-rtTA;TetO-hVegfa controls (four islet preparations per group). Quantitative RT-PCR experiments were performed according to the Minimum Information for Publication of Quantitative Real-Time PCR Experiments (MIQE) guidelines (Bustin et al., 2009).

Islets were isolated as described previously (Brissova et al., 2004), handpicked in Clonetics EGM MV Microvascular Endothelial Cell Growth Medium (Lonza, Basel, Switzerland), washed three times with 2 mM EDTA/1× PBS, and then dispersed by incubating with Accutase (Innovative Cell Technologies) at 37°C for 10 minutes with constant pipetting. After quenching Accutase with EGM MV media, cells were incubated at 4°C with rat anti-PECAM1-PE antibody (1:500; BD Pharmingen) in FACS buffer (2 mM EDTA/2% fetal bovine serum/1× PBS). Anti-rat Ig, κ CompBead Plus Compensation Particles (BD Biosciences) were used for single color compensation. DAPI (0.25 μg/1,000,000 cells; Invitrogen Molecular Probes) was added to samples for non-viable cell exclusion and cells were sorted using a FACSAria III cell sorter (BD Biosciences). RNA was extracted from whole islets and sorted endothelial cells as described above.

RNA sequencing (RNA-Seq) was performed as described (Brissova et al., 2014). Briefly, RNA from control and VEGF^Up islets, and endothelial cells derived from VEGF^Up islets (RNAqueous, Ambion) was amplified using a NUGEN Technologies Ovation RNA amplification kit, and sequenced using standard Illumina methods as described (Malone and Oliver, 2011; Mortazavi et al., 2008). Raw sequencing reads were mapped to the mm9 mouse reference genome using TopHat v2.0 (Trapnell et al., 2009). The aligned reads were then imported onto the Avadis NGS data analysis platform (Strand Scientific Intelligence, Bengalor) where they were filtered by their quality metrics. Duplicate reads were then removed and quantification of normalized gene expression was performed using the TMM (trimmed mean of M-values) algorithm (Dillies et al., 2012; Robinson and Oshlack, 2010). Principal component analysis (PCA) and hierarchal clustering analysis were used to compare the transcriptional profiles from each sample group (control islets, VEGF^Up islets, and sorted endothelial cells isolated from VEGF^Up islets). Differential expression was calculated based on fold change ≥2.0 and the P-value was estimated using z-scores (≤0.05) determined by the Benjamini Hochberg false discovery rate (FDR) method (Benjamini and Hochberg, 1995). The RNA-Seq data set from the same biological samples was also used for other studies (Brissova et al., 2014).

Quantification of immunohistochemistry was performed on original, unadjusted images with MetaMorph software (Universal Imaging). To represent innervation within islet centers accurately, only islets with a cross-sectional diameter >100 μm were included in the analysis. Morphometric analysis was performed on at least 30 islets per mouse, with n≥4 mice per group. To assess vascularization in whole mount-labeled Wnt1-Cre;Foxd3^fl/- embryonic pancreata, vessel density was measured in every other optical section from individual confocal z-stacks, totaling >30 slices analyzed per sample (Reinert et al., 2013). Statistics were performed with GraphPad Prism software, using Student’s t-test or one-way ANOVA to compare groups.

We thank G. Gu, N. Ferrara, S. Efrat and P. Campochiaro for kindly providing mice. We thank A. L. Kolodkin and R. Brekken for their generous antibody gifts. We thank Greg Poffenberger and Alena Shostak for technical assistance. We are grateful to Maureen Gannon, Danny Winder, Wenbiao Chen, Masakazu Shiota, Elizabeth Blackburn and E. Danielle Dean for their helpful discussions.