A radial axis defined by semaphorin-to-neuropilin signaling controls pancreatic islet morphogenesis

Pancreatic islets are named for their most characteristic feature: islets are endocrine cell clusters dispersed throughout an abundant sea of exocrine tissue. The molecular and signaling origins of this conserved hallmark morphology, which are established during fetal development, remain undefined. Islets are endocrine micro-organs that are crucial for regulation of metabolism, including glucose control. Within the islets, β cells secrete insulin to stimulate glucose uptake by peripheral tissues and α cells secrete glucagon to mobilize glucose from target organs such as the liver. Islet structure may influence function: specialized vascular and neural crest-derived structures ramify within islets to regulate hormone secretion and blood flow, and interactions among islet cells can influence hormone secretion (Brissova et al., 2006; Cleaver and Dor, 2012; Lammert et al., 2003; Muñoz-Bravo et al., 2013; Reinert et al., 2014; Rodriguez-Diaz et al., 2011; van der Meulen et al., 2015). Islet morphogenesis comprises several distinct processes. Initially, islet formation begins with delamination of endocrine precursor cells from the primitive core of the branching ductal epithelium. After exiting this epithelial layer, newborn islet cells migrate away from the epithelium into the surrounding mesenchyme, cluster and begin to form recognizable islets (Benitez et al., 2012; Gouzi et al., 2011; Rukstalis and Habener, 2007). Vascularization and innervation of islets begins in the embryo and continues postnatally (Reinert et al., 2014), and islets appear to continue to disperse from large central ductal structures after birth, although this latter process has not been measured. Thus, islet morphogenesis could be parsed into delamination, migration, clustering and remodeling, with the possibility that each of these processes could be controlled by distinct signals. Despite the importance of islet structure to function, little is known regarding long- or short-range signals that control islet morphogenesis.

Prior studies have identified secreted signals controlling cell differentiation and proliferation in pancreas development (reviewed by Benitez et al., 2012; Gittes, 2009; Kim and Hebrok, 2001; Puri and Hebrok, 2010; Serup, 2012). For example, classical fetal organ culture studies showed that pancreatic mesenchyme provides cues for epithelial expansion, morphogenesis and differentiation (Gittes et al., 1996; Golosow and Grobstein, 1962; Landsman et al., 2011; Wessells and Cohen, 1967). Studies by Bhushan et al. (2001) revealed that Fgf10, which is expressed throughout the pancreatic mesenchyme at the inception of pancreas morphogenesis, is required for pancreatic progenitor proliferation. Other studies demonstrated that vascular endothelium induces foregut expression of Ptf1a and Pdx1, crucial transcriptional regulators of pancreatic growth and development (Lammert et al., 2001; Yoshitomi and Zaret, 2004). At later stages, Notch signaling – possibly through short-range lateral inhibition – controls endocrine cell specification and acinar cell differentiation from progenitor cells (Afelik et al., 2012; Apelqvist et al., 1999; Cras-Méneur et al., 2009; Esni et al., 2004; Murtaugh et al., 2003; Shih et al., 2012). Short-range signals associated with neural development, including netrins and Eph-ephrin signaling, have been associated with pancreatic cell migration, growth and islet function (Konstantinova et al., 2007; Yang et al., 2011; Yebra et al., 2003). In islet morphogenesis, extracellular signals, including EGF, HGF and TGFβ, and intracellular signal transducers, including Cdc42, Rac1, Tm4sf4 and Grg3 have been suggested to influence fetal islet morphogenesis (Anderson et al., 2011; Blum et al., 2014; Greiner et al., 2009; Guo et al., 2013; Kesavan et al., 2014; Metzger et al., 2012; Miettinen et al., 2000; Miralles et al., 1998; Pagliuca et al., 2014; Rezania et al., 2014; Sanvito et al., 1994; Tulachan et al., 2007). Thus, although many signals are known to regulate pancreas and islet development, spatially localized cues that might control islet morphogenesis have not yet been discovered.

To define signals controlling islet cell migration and movement, we designed a screen to identify secreted factors sufficient to alter β cell localization in development. We characterized one factor identified in the screen, semaphorin 3a, as a regulator of fetal islet cell migration during islet morphogenesis. The Sema3 family is composed of secreted proteins initially identified as inducers of axon growth cone collapse, and later characterized as chemoattractant or chemorepulsive factors that regulate multiple specific stages of nervous system development, including axon targeting, neuron migration and neuron polarization (Chen et al., 2008; Kolodkin et al., 1993; Polleux et al., 1998; Shelly et al., 2011; Tran et al., 2009). In addition, semaphorins have been demonstrated to function in other organ systems, including bone homeostasis and cardiovascular development (Degenhardt et al., 2013; Epstein et al., 2015; Fukuda et al., 2013; Ieda et al., 2007). Secreted semaphorins usually signal through heterodimeric receptor complexes composed of neuropilins and plexins (Chen et al., 1997; Giger et al., 2000; Kolodkin et al., 1997; Takahashi et al., 1999; Tamagnone et al., 1999; Winberg et al., 1998). Neuropilin 2 (Nrp2) expression has been reported in adult islets and Sema3a mRNA was enriched in E11.5 mouse pancreatic mesenchyme (Cohen et al., 2002; Guo et al., 2013), but a functional role for semaphorin signaling has not yet been reported in pancreatic development or physiology.

Here, we provide evidence that semaphorin signaling through Nrp2 receptors during pancreas development provides guidance cues along a previously unrecognized proximodistal axis that is essential for regulating islet morphogenesis and dispersion. This developmental signaling axis in the pancreas has striking homology to radial patterning cues required for cortical lamination during neural development, unexpectedly revealing shared use of a signaling module to establish radial pattern in the brain and pancreas.

To define signals controlling islet cell migration, we identified 21 candidate secreted factors based on existing genome-wide expression datasets from fetal pancreatic mesenchyme (Guo et al., 2013) and developing islet cells (Benitez et al., 2014). To assay for effects on islet development, we implanted factor-soaked beads in cultured E13.5 Ins1-GFP transgenic mouse fetal pancreas (Fig. 1A, Fig. S1) (Hara et al., 2003). By measuring changes in the distribution of β cell-derived GFP signal around beads over 5 days of development, we identified seven growth factors that increased β cell proximity to beads (Fig. 1A). These included growth factors with established roles in islet biology, including PDGF, VEGF and HGF (Cai et al., 2012; Chen et al., 2011; Reinert et al., 2013, 2014; Roccisana et al., 2005), and ligands with unknown function in islet development, including Cxcl13 and Sema3a. Based on this β cell clustering and the pronounced α cell phenotypes described below, we chose semaphorin signaling for further study.

Pancreatic β cell distribution was strongly shifted toward beads soaked in Sema3a (P=0.01 relative to PBS control beads, Fig. 1B,C,E,F). We also observed clustering of glucagon⁺ α cells adjacent to Sema3a-soaked beads by immunofluorescence analysis (Fig. 1D,G). Quantification of α cell fluorescence relative to beads demonstrated that glucagon⁺ cells were significantly increased near Sema3a-soaked beads compared with PBS-soaked control beads (P=0.04, two tailed t-test: Sema3a=2348±254 versus PBS=1506±61). Many α cells lacked E-cadherin expression, consistent with acquisition of a migratory phenotype (Acloque et al., 2009). Moreover, unlike α cells in PBS controls, α cells adjacent to Sema3a-soaked beads were deep in the mesenchyme several cell diameters from nearby ductal epithelia, suggesting that Sema3a might affect islet cell migration over a relatively long range (Fig. S1). Expression of Ki67 in glucagon⁺ cells remained unchanged by Sema3a-soaked beads at multiple time points tested, suggesting that increased α cell proliferation did not underlie the observed phenotypes (Fig. S1). Likewise, expression of the islet progenitor marker Neurog3 was indistinguishable in organs implanted with PBS or Sema3a-soaked beads, indicating that Sema3a did not induce endocrine differentiation (Fig. S1). Other semaphorin family members, including Sema3b and Sema3f, were expressed in the pancreas at E15.5, and in bead experiments had similar effects on fetal α cell localization (Fig. S1). These data suggested that semaphorin signaling might provide guidance cues to migrating fetal islet cells.

Based on established roles of semaphorins as chemoattractant or chemorepulsant guidance cues in other contexts, we next assessed whether semaphorin signaling might provide directional cues in islet morphogenesis. At E15.5, in situ hybridization revealed a striking concentration of Sema3a transcripts at the pancreatic mesenchymal periphery. By contrast, we observed uniform distribution of Polr2a transcripts encoding RNA polymerase II (Fig. 2A-C). Developing islet cells, including glucagon⁺ α cells, were localized to the core of the organ, adjacent to the central epithelium (Fig. 2A-C). Cells expressing Sema3a co-expressed the fibroblast marker vimentin and were enriched in FACS-purified mesenchymal cells in the fetal pancreas, supporting the view that peripheral fibroblasts expressed Sema3a (Fig. S2). We observed a similar peripheral mesenchymal localization of Sema3d using Sema3d^GFP/Cre knock-in mice (Katz et al., 2012) and by measuring gene expression in FACS-purified cell populations (Fig. S2). Compared with Sema3a expression, Sema3d^gfp expression appeared to extend several cell layers deeper, suggesting that a semaphorin gradient composed of multiple types of semaphorins could instruct islet morphogenesis. Alternatively, this difference in observed expression pattern could reflect differences in detecting Sema3a by in situ hybridization and Sema3d by GFP expression.

In contrast to Sema3 ligand production in peripheral mesenchyme, we detected the Sema3 receptor Nrp2 in both α and β cells at E13.5 (Fig. 2D,E). From E15.5 to adult stages, Nrp2 expression was primarily detected in α cells (Fig. S3). Nrp2 was not expressed in endocrine progenitor cells marked by Neurog3⁺ nuclei, suggesting expression is acquired only after differentiation into α or β cells (Fig. S3). Nrp1 was not detectable in fetal islet cells, but was present in other pancreatic cell types (Fig. S3). Human fetal α cells expressed NRP2, whereas somatostatin-expressing δ cells expressed NRP1, indicating some signaling features are conserved in human pancreas development (Fig. S3). Upon implanting Sema3a-soaked beads in cultured Nrp2 knockout mouse pancreas, we did not detect α cell aggregation around beads (Fig. 2F-I). Thus, Nrp2 is required for islet cell responses to Sema3a. These findings also indicate that other receptors like Nrp1 did not compensate for Nrp2 loss, as observed in other systems (Takashima et al., 2002).

Neuropilins act as co-receptors with plexin proteins (Takahashi et al., 1999; Tamagnone et al., 1999). Multiple mRNAs encoding plexins were detected in E15.5 mouse fetal pancreas by RT-PCR (Fig. S3). Assessment of mRNA expression of selected plexin co-receptors in FACS-purified cell populations from the E15.5 pancreas detected enrichment of Plxna3 in fetal endocrine cells relative to whole pancreas, pancreatic epithelial cell (EpCAM⁺), or endothelial cell subsets (CD31⁺; Fig. 2J). Plexins B1 and B2 were expressed in the pancreas, but were not similarly enriched in islet cells (Fig. S3). These data indicate that the expected neuropilin co-receptors are present in the appropriate cell types in the pancreas to facilitate responses to semaphorin cues. Together these findings suggest that semaphorin signals from distal mesenchyme to central Nrp2⁺ islet cells could define an endogenous long-range developmental axis.

Semaphorin guidance signals can be repulsive or attractive, depending on the cellular context (Tran et al., 2007). Based on islet cell attraction to Sema3a-soaked beads and the radially asymmetric distribution of Sema3a and Sema3d transcripts, we hypothesized that semaphorins function as a chemoattractant cue for developing islet cells. If so, loss of Nrp2 should impair islet cell migration outward from their origin in the central ductal epithelium. To test this hypothesis, we assessed islet development in mice with homozygous inactivation of Nrp2 (Giger et al., 2000). Because of high-frequency perinatal mortality in homozygous Nrp2 mutants (Giger et al., 2000), we focused our analysis of pancreas development at E15.5 and postnatal day 1 (P1). At E15.5, islet cells in control littermates were distributed throughout the pancreas as discrete clusters or single islet cells, corresponding to nascent islets and cells migrating to join islets (Fig. 3A). By contrast, in E15.5 Nrp2 mutants, islet cells formed long streams of hormone⁺ cells along ducts (Fig. 3B). In controls at P1, islets formed as rounded structures distributed throughout the pancreas, with β cells in the islet interior surrounded by α cells (Fig. 3C). In Nrp2 mutants, we observed abnormal islet cell aggregates enveloping ductal structures in central regions of the pancreas (Fig. 3D,F). Within these aggregates, typical islet architecture appeared preserved, but individual islets were abnormally clustered near other islets and ducts. The ductal and acinar tissue in Nrp2 mutants appeared similar to controls.

To quantify these phenotypes, we measured specific features of islet development and morphogenesis. Islet α cell and β cell quantities were not detectably altered in Nrp2 mutants at P1 (P=0.32, α cells; P=0.28, β cells; Fig. 3G), suggesting islet morphogenetic defects did not reflect altered proliferation or differentiation. Measurement of the distance from islets to the nearest ductal structure marked by DBA lectin indicated a 40% reduction in duct-to-islet distance in Nrp2 mutants at P1 (P=0.03, Fig. 3H). Subsequent analysis of rare surviving adult Nrp2-knockout mice indicated that this islet phenotype persisted in adults, with a halving of duct-islet distance detected (P=0.04, Fig. 3H and Fig. S4). We also quantified the extent of contact between ducts and islets and noted a significant increase in islet contact with ductal surfaces in Nrp2 mutants (97°±2 in controls, 147°±6 in mutants: P=2×10⁻⁵, Fig. 3I). Total encirclement of ducts by islets occurred rarely in controls (1.8%) compared with islets in Nrp2 mutants (8.5%, P=0.0009, Fisher's exact test). Thus, islet separation from ducts appeared to be impaired in Nrp2 mutants. We did not detect obvious alterations in islet vascularization or innervation in Nrp2 mutants (Fig. S4). To assess the requirement for Nrp2 specifically in the pancreatic islets, we attempted conditional inactivation of Nrp2 using a Cre-Lox approach. We did not observe a reduction of Nrp2 expression by immunofluorescence using either Neurog3-Cre or Pdx1-Cre (Fig. S4). Collectively, the phenotypes observed support the view that radial migration of islet cells away from their ductal origin was impaired in Nrp2 mutants.

Prior studies of islet morphogenesis have been limited to reconstruction of three-dimensional islet phenotypes from two-dimensional imaging. To characterize islet developmental defects in the intact pancreas of Nrp2 mutants at P1, we performed whole-organ confocal imaging using CLARITY (Chung et al., 2013; Tomer et al., 2014). This method permits optical clearing of large tissues while preserving protein and nucleic acid localization, enabling phenotyping of intact organs with cellular resolution. After clearing the pancreas at P1, we assessed islet hormones and ductal markers by immunofluorescence in Nrp2 wild-type and knockout mice. In controls, islets formed as round clusters distributed both near to and remote from the central ductal network (Fig. 4A,C,E,F; Movie 1). In contrast, islet cells in Nrp2 knockout mice grew in streams along large central ductal structures (Fig. 4B,D,G,H; Movie 2). Thus, whole-organ imaging of intact pancreata corroborated the morphometric quantification of islet distribution we obtained with sectioned tissue, and supported a model in which Nrp2 signaling is essential for separation and dispersion of islets from their ductal origin during development.

Our analysis of fixed tissues suggested that semaphorin-Nrp2 signaling is essential for migration of developing pancreatic endocrine cells. To test this hypothesis directly, we used live cell imaging to quantify movement and cell shape changes in fetal islet cells (Pauerstein et al., 2015). We used mice harboring a Neurog3-tdTomato transgene, which labels all Neurog3⁺ endocrine progenitors and their hormone-expressing progeny (Sugiyama et al., 2013). Time lapse confocal imaging of cultured E13.5 Nrp2^+/+; Neurog3-tdTomato control organs for 24 h revealed extensive migration and deformation of developing endocrine cells, including extension of filopodia (Fig. 5A,B,E-H; Movie 3). In contrast, movement of Neurog3-tdTomato⁺ cells in Nrp2 mutants was limited, and these cells failed to undergo deformation (Fig. 5C,D,I-L; Movie 4). Mutant islet cells often remained rounded and near their origin throughout imaging, and many cells did not extend detectable filopodia (Fig. 5E-L; Movie 4). Quantification of cell displacement confirmed a reduction in total distance traveled by single islet cells in Nrp2 mutants (P=2×10⁻¹¹; Fig. 5M, Fig. S5). Measurement of total cell shape change during imaging confirmed a significant reduction of cell deformation (P=0.003, Fig. 5N). These experiments indicate that Nrp2 is required for fetal islet cell migration and deformation, including cell biological processes such as the filopodia extension that is integral to migration, and provide a cellular mechanism linking defective Nrp2 signaling to defects in islet morphogenesis.

We used classical organ culture and genetics to identify mesenchyme-derived cues influencing islet development, and found that semaphorins are unrecognized chemoattractants for α and β cells. Purified Sema3a was sufficient to redirect fetal α and β cell migration toward beads, indicating that semaphorin signaling provides instructive, not merely permissive, cues to these islet cells. Spatially localized production of Sema3a or other semaphorins by peripheral mesenchymal cells could provide instructive cues to guide radial migration of Nrp2⁺ islet cells (Fig. 6). Consistent with this model, loss of Nrp2, a receptor for secreted semaphorins, resulted in impaired cell migration and defects in islet morphogenesis. Thus, our findings suggest that semaphorins are potent long-range chemoattractants for fetal α and β cells, defining a radial axis that controls islet morphogenesis at both unicellular and multicellular levels. To our knowledge, long-range or spatially localized guidance cues coordinating islet morphogenesis have not previously been identified. Future studies, perhaps using conditional genetics in specific cell subsets, could assess the principal semaphorins responsible for regulating islet morphogenesis. Our data raise the likelihood that multiple semaphorins, including Sema3a and Sema3d, could serve this function.

While Sema3a stimulated migration of both fetal α and β cells, this effect appeared more pronounced with α cells. We also found differences in Nrp2 expression between α and β cells: fetal α cell Nrp2 expression was durable and persisted in adult α cells, unlike in β cells, which showed only transient fetal-stage expression. To the extent that Nrp2 appears to be essential for Sema-mediated signaling in fetal islets, our findings raise the possibility that Sema-Nrp2 signaling primarily guides α cell migration and initial β cell migration. In the later stages of islet morphogenesis, it is possible that α cells then influence β cell migration through cues yet to be identified. Our findings also suggest possible functions for Nrp2 signaling in adult islet α cells. These models could be tested by conditional inactivation of Nrp2 in fetal or adult α cells, though we found that the extant conditional ‘floxed’ Nrp2^f allele (Walz et al., 2002) was resistant to inactivation by Cre recombinase transgenes, including Pdx1-Cre or Ngn3-Cre. These have been used previously to inactivate other floxed alleles in the pancreas or fetal islets (Goodyer et al., 2012; Gu et al., 2002). Thus, alternative approaches, perhaps including the generation of additional Nrp2^f alleles, may be required to pursue conditional genetics in the pancreas.

Semaphorin signaling through Nrp-plexin receptors can be transduced through multiple intracellular pathways that regulate cytoskeletal dynamics leading to oriented cell movement, including Rho GTPases, the Ras pathway or via Cdk5 signaling (Ahmed and Eickholt, 2007; Eickholt et al., 2002; Takahashi et al., 1999). Our preliminary data support a role for Nrp2 signaling through Cdk5 and Rho intermediaries in semaphorin-induced migration of nascent islet cells (P.T.P., K.T. and S.K.K., unpublished). In addition to Sema3a signaling through plexin receptors to cytoskeleton effector proteins, it is also possible that Sema3a signaling leads to an interaction between Nrp2 and integrins to affect cell migration. Nrp2 interactions with integrins at focal adhesions within the same cell (cis) permit integrin binding to laminin in the extracellular matrix, whereas Nrp2 binding to integrins on endothelial cells (trans) allows cancer cell intravasation and extravasation; both of these processes lead to an increase in cancer cell migration or metastasis and poor prognosis (Goel et al., 2012; Cao et al., 2013). Considering that many cancer mechanisms have been adapted from normal developmental events, it is conceivable that such a mechanism might also occur in normal islet morphogenesis.

Distribution of the islets of Langerhans throughout the exocrine tissue is a cardinal pancreatic feature in most vertebrates – including fish, birds, snakes, rodents and primates – suggesting selective forces have maintained this anatomy (Conlon et al., 1988; Moscona, 1990; Steiner et al., 2010). Thus, islet morphogenesis and dispersion is likely controlled by multiple conserved mechanisms. Initially, islet progenitor cells within the fetal ductal epithelium delaminate and, while differentiating, migrate into nascent islet clusters. Multicellular clusters enlarge, reflecting continued cell aggregation and proliferation, and at later fetal and postnatal stages islet cells rearrange to form morphologically mature islets dispersed throughout the exocrine pancreas (Benitez et al., 2012). Our analysis of Nrp2 mutant mice suggests that the initial differentiation, delamination and clustering of islet cells do not require Sema-Nrp signaling. However, in Nrp2 mutants, islet cells failed to establish normal dispersed islet morphology, leading to abnormally large islet cell aggregates surrounding and connected to major ducts. We postulate that normal islet dispersion could reflect the combination of at least two distinct processes: (1) active directional islet cell migration mediated by a combination of long- and short-range cues, both attractive and repulsive; and (2) growth of intervening non-islet tissues, most abundantly acinar and ductal cells. Other signals are likely involved in islet dispersion: we identified multiple candidate signaling pathways with effects on islet cell development in our screen. However, we focused on semaphorin signaling because of the observed phenotypes in our assays and the availability of reagents to perturb the semaphorin pathway in the pancreas. As pancreatic growth was not detectably impaired in Nrp2 mutants at birth, tissue growth is likely not sufficient to disperse islets during embryogenesis. In an evolutionary context, the arrangement of islet cells contiguous with their originating epithelium observed in Nrp2 mutants is reminiscent of findings from jawless fish, where cells expressing pancreatic hormones such as insulin, pancreatic polypeptide and glucagon are found adjacent to or within homologous foregut intestinal epithelium (Conlon et al., 1988; Falkmer, 1979).

The sharing of genetic or signaling modules used to build morphologically or phylogenetically disparate features has been called ‘deep homology’ (Shubin et al., 1997, 2009). Unexpectedly, we found that key elements of a radially oriented signaling pathway regulating cortical lamination in central nervous system development are also deployed for dispersion of islets during pancreas development. During mammalian brain development, a stereotyped laminar organization of neurons in the cortex is essential for cortical function. In mice, semaphorins at the distal pial surface signal to postmitotic neurons derived from proximal ventricular zone neural stem cells, directing radial migration toward the pial surface along radial glial cells as cortical lamination proceeds (Chen et al., 2008). In that context, plexin receptors signal through Rho GTPases to direct cell migration towards outer cortical layers (Azzarelli et al., 2014; Pacary et al., 2013). Signaling via Cdk5 also controls multiple aspects of neuronal radial migration and cortical lamination (Chae et al., 1997; Su and Tsai, 2011). Ectopic Sema3a or loss of Nrp1 is sufficient to disrupt development of this layering. Thus, cortical lamination and pancreatic islet morphogenesis appear to share strikingly similar signaling molecules, cellular processes and developmental axis arrangement (Fig. 6). Sema3a signaling in brain cortical lamination may extend up to several hundred micrometers, comparable with or longer than the distances involved in mouse islet development. Our findings also support prior work that revealed striking similarities between mammalian islet cells and neurons (Ohta et al., 2011; Rulifson et al., 2002; Van Noorden and Falkmer, 1980). As in nervous system development, additional signals conveying chemoattractive, chemorepulsive or ‘stop’ cues might influence islet development. We speculate that an undetected radial or laminar patterning of islets may be linked to specialized functions, responsiveness or control. Consistent with this notion is the observation that human islet inflammation, destruction or preservation in type 1 diabetes mellitus may occur in a heterogeneous pattern (Gaglia et al., 2015; In't Veld, 2014; Poudel et al., 2015), and that autonomic nerves link groups of islets (Reinert et al., 2014; P.T.P., B.H., K.D. and S.K.K., unpublished).

Discovery of a long-range islet chemoattractant could also influence regenerative approaches to islet reconstitution. In hematopoietic stem cell transplantation, chemokine-guided migration of stem cells to developmental niches illustrates the therapeutic applications of cellular homing (Copelan, 2006). We speculate that studies of islet guidance cues might help establish similar homing and engraftment paradigms for islet cells or their progenitors.

All animal experiments were conducted in accordance with Stanford University IACUC guidelines. Mouse Insulin1 promoter-GFP transgenic mice were purchased from Jackson Laboratories (Hara et al., 2003) and maintained on a CD1 genetic background. Nrp1 floxed (Gu et al., 2003) and Nrp2 knockout mice (Giger et al., 2000) were obtained with permission from Dr David Ginty (Harvard Medical School, Boston, MA, USA) and maintained on a Black6 background. Neurog3-tdTomato transgenic mice have been previously reported and were maintained on a CD1 background (Sugiyama et al., 2013), as were Sema3d^GFP-Cre knock-in mice (Katz et al., 2012). Neurog3^gfp/+ mice were obtained from Klaus Kaestner (University of Pennsylvania, USA) and maintained on a CD1 background (Lee et al., 2002). Wnt1-Cre (Danielian et al., 1998; Lewis et al., 2013) and Rosa-mTmG (Muzumdar et al., 2007) mice were obtained from the Jackson Laboratories and maintained on a mixed Black6 and CD1 background. Transgenic Ngn3-Cre and Pdx1-Cre mice have been previously described (Gu et al., 2002). Floxed Nrp2 mice (Walz et al., 2002) were obtained from the Kolodkin lab (Johns Hopkins School of Medicine and HHMI, MD, USA) and maintained by the Epstein and Kim labs. Experiments and morphometry analysis were performed on mice at embryonic day (E) 13.5, E15.5, postnatal day 1 and 6 months; both male and female mice were used in all experiments.

For organ culture experiments, fetal pancreatic rudiments were dissected from E13.5 mouse embryos, embedded in type 1 collagen gel (EMD Millipore) and cultured in DMEM/F12 supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin for experiments not involving beads (Gittes et al., 1996; Puri and Hebrok, 2007). The FBS concentration was reduced to 0.5% for bead assays in the screen. Medium was changed every other day for experiments not involving beads. For bead assays, one half of the media volume was refreshed daily for the duration of experiments. Organs were fixed in 4% PFA and then processed for whole-mount immunofluorescence or for cryosectioning.

For live-imaging experiments, E13.5 organs were dissected and visually assessed for transgene presence. For Nrp2 knockout experiments, rapid genotyping was performed using the Phire Tissue Direct kit (Thermo Fisher). Samples were mounted in type 1 collagen gel (EMD Millipore) in glass-bottomed dishes (MatTek) and cultured in organ culture media without Phenol Red (described above) supplemented with 1% insulin-transferrin-selenium. Images were acquired using a Leica SP8 inverted confocal system equipped with a white light laser and an environmental control chamber using a 20× water immersion objective. For detection of tdTomato, samples were excited at 561 nm. Images were acquired at 20 min intervals for all time-lapse experiments. Data were saved in LIF format and analyzed using Volocity image analysis software (Perkin Elmer) or ImageJ. Quantification of cell shape changes have been described elsewhere (Pauerstein et al., 2015). Cell displacement and velocity measurements were calculated using the MTrackJ plugin in ImageJ (Meijering et al., 2012).

Tissue was fixed in 4% PFA overnight and processed for cryosectioning. Samples were blocked in 1% BSA and 0.2% nonfat evaporated milk supplemented with 0.5% Triton X-100 and 1% DMSO, and antibody incubations were performed in blocking buffer. AlexaFluor- and Dylight-conjugated secondary antibodies (Life Technologies and Jackson Immunoresearch) were used for detection. Antibodies used are indicated in Table S3.

For all morphometric analyses, images representing entire pancreatic sections were acquired using a Zeiss AxioM1 microscope. At least 12 sections of 10 μm thickness and separated by at least 120 μm were analyzed for each organ. For quantification of α and β cell areas, tissue was immunostained for E-Cadherin and appropriate hormone markers. E-Cadherin-positive areas were manually traced in ImageJ, and images were thresholded to calculate hormone-positive areas using the ‘Analyze Particles’ function. Total hormone⁺ areas were divided by total E-Cadherin⁺ areas for each biological sample. For duct-islet distance measurements, tissue was labeled using DBA lectin and antibodies to ChgA. Distances from ChgA⁺ islet edges to nearest DBA lectin⁺ structure were measured using the ‘Line’ function in ImageJ. All islet clusters larger than 4 cells were measured in all sections analyzed for each pancreas. For assessment of duct contact with islets, ducts cut in cross-section and adjacent to islets were visually identified. The ImageJ ‘Angle’ tool was used to measure the angle defined by the center of the duct and the two edges of islet contact with the basal surface of the duct. Data were exported to Microsoft Excel and Graphpad Prism for analysis. Distance or angle measurements were averaged for each independent sample, and biological groups were compared using statistical methods described below. Measurements were performed by individuals blinded to the genotypes of samples.

Whole organ analysis of pancreatic islet morphology was performed using CLARITY (Chung et al., 2013; Hsueh et al., 2017) with several modifications. Tissue was dissected in ice-cold PBS and immediately incubated in hydrogel monomer solution for 3-5 days at 4°C. The hydrogel monomer solution is 4% PFA, 4% acrylamide, 0.25% VA-044 initiator and 0.05% saponin in PBS. Bis-acrylamide, included in the original formulation (Chung et al., 2013), was omitted to prevent polymerization of hydrogel outside the tissue. Tissue was cleared passively for 5-14 days by incubating in clearing buffer (4% SDS, 200 mM boric acid at pH 8.5) at 55°C (Tomer et al., 2014). After clearing, tissue was washed in PBS and immunostained using standard protocols, with antibody incubations extended to 3-5 days to permit penetration of thick samples. Samples were mounted in FocusClear (CelExplorer Labs) and imaged using an Olympus confocal microscope equipped with 5× air and 10× water immersion objectives. Tiled z-stack datasets were stitched into a single z-stack using Fiji (Schindelin et al., 2012) and were visualized using Volocity image analysis software. 3D reconstruction videos were created using Arivis Vision4D software.

Pancreatic rudiments were dissected from E15.5 embryos and processed as previously described (Sugiyama et al., 2007), with the exception that organs were dissociated using Accutase (eBioscience). Single-cell suspensions were stained using antibodies described in Table S1 and were sorted using a BD FACS Aria II. Red blood cells were depleted from the cell fractions using RBC lysis buffer (BioLegend). Cells were collected in PBS supplemented with 2% BSA and 10 mM EGTA. RNA was purified using the PicoPure RNA extraction system (Life Technologies), cDNA was synthesized using Superscript III reverse transcriptase (Life Technologies), and gene expression was measured using qPCR with Taqman probes and an ABI7500 qPCR system (Applied Biosystems).

For analysis of plexin gene expression in FACS-purified cell populations, E15.5 Neurog3^gfp/+ pancreas was obtained and processed for FACS as above, with fractions defined as follows: GFP⁺CD133^lowEpCAM^low/negCD31⁻ cells were considered to be ‘endocrine’, EpCAM⁺GFP⁻ cells were ‘EpCAM’, CD31⁺GFP⁻EpCAM⁻ cells were ‘CD31’ and EpCAM⁻GFP⁻CD31⁻ cells were considered ‘Negative’ and represent a mixed population of cells not encompassed by the other cell fractions (Benitez et al., 2014; Sugiyama et al., 2007). CD45⁺ white blood cells were excluded from all fractions, and red blood cells were depleted as described above.

Gene expression analysis using quantitative RT-PCR was performed using the following Taqman probes (Life Technologies): Sema3a, Mm00436469; Sema3d, Mm01224783; Pecam, Mm01242584; S100a4, Mm00803372; vimentin, Mm01333430; Acta2, Mm00725412; Syp, Mm00436850; Sox9, Mm00448840; Ins2, Mm00731595; Plxna3, Mm00501170; Plxnb1, Mm00555359; and Plxnb2, Mm00507118.

To assess plexin family gene expression by RT-PCR, RNA was isolated from microdissected whole E15.5 wild-type mouse pancreas (CD1 background) using the PicoPure RNA Isolation Kit (Life Technologies). cDNA was synthesized using Superscript III reverse transcriptase, and gene expression was assessed by PCR using BioMix Red (Bioline) and the primer sets indicated in Table S4.

RNA in situ hybridization was performed using the RNAscope system (Advanced Cell Diagnostics, ACD). Probes for mouse Sema3a (ACD 412961) and Polr2a (ACD 312471) were purchased from ACD, as were RNAscope 2.0 (Brown reagents). For fluorescence detection, tyramide signal amplification reagents (Life Technologies) were used in place of DAB reagents. In situ hybridizations were performed on freshly sectioned cryoembedded tissue samples at stages indicated in the text.

Experiments involving human tissue samples were approved by the Stanford University School of Medicine Institutional Review Board. Human pancreatic tissue between gestational days (GD) 75 and 110 was obtained through the Birth Defects Research Laboratory, University of Washington (Seattle, WA, USA). All donors provided informed consent.

Data are presented as mean±s.e.m. unless otherwise indicated. Student's t-test was used for statistical comparisons unless otherwise noted. P values less than 0.05 were considered significant. Statistical analysis was performed using Graphpad Prism and Microsoft Excel software.

We thank members of the Kim Lab for helpful discussion and advice, Dr Jon Mulholland and Kitty Lee of the Stanford Cell Sciences Imaging Facility for assistance with confocal microscopy, and Drs P. Lwigale (Rice University), M. Scott (Stanford University) and R. Nusse (Stanford University) for sharing reagents.