Introduction

In mice, pancreas development begins when pancreatic and duodenal homeobox 1 (PDX1), pancreas transcription factor 1 subunit α (PTF1A) and sex-determining region Y (SRY)-related high mobility group (HMG) box (SOX)9-expressing epithelial cells bud from the posterior foregut endoderm at embryonic day (E) 9 [1–3]. Signals from adjacent tissues support budding and as development proceeds, two waves of differentiation occur [4–8]. The primary transition, from E8.5–10.5, is marked by low numbers of hormone-immunoreactive (+) cells; the secondary transition, from E12.5–15.5, is denoted by a large wave of exocrine and endocrine cell differentiation [9]. The restriction towards a terminally differentiated endocrine cell occurs progressively [10].

Endocrine differentiation requires inhibition of NOTCH signalling and transient activation of neurogenin 3 (NEUROG3) in epithelial trunk progenitor cells, and is accompanied by exit from the cell cycle and migration into the mesenchyme [11–18]. Downstream, regulators of beta cell specification include: paired box 4 (PAX4), neurogenic differentiation 1 (NEUROD1), regulatory factor X, 6 (RFX6), NK2 homeobox 2 (NKX2.2), NK6 homeobox 1 (NKX6.1) and v-maf avian musculoaponeurotic fibrosarcoma oncogene homologue A (MAFA), among others [19–25]. While a number of these factors are essential for normal beta cell development, it is unclear if any transcription factor couples the endocrine progenitor fate choice with downstream terminal differentiation. Here, we demonstrate that SOX4 may play this bridging role during endocrine cell genesis.

SOX proteins are an evolutionarily ancient family of complex-forming transcriptional activators and repressors that are identified by a conserved HMG DNA-binding domain [26, 27]. A plethora of SOX genes has been demonstrated to be expressed within the developing pancreas, including most of the C, D, E, F, G and H subfamily members [28]. Early during pancreas formation, expression of an E group family member, SOX9, is critical for the maintenance of pancreatic progenitor identity [1].

Previous work has demonstrated SOX4 expression in the embryonic human, mouse and zebrafish pancreas [28–30]. Null mutation of Sox4 results in cardiac abnormalities, defects in blood flow and embryonic lethality by E14.5 prior to completion of the bulk of endocrine cell differentiation. However, significant reductions of insulin (INS) and/or glucagon (GCG) expression were observed by us and others [29, 31].

This study expands on these observations to define the spatiotemporal importance of SOX4 during endocrine formation in vivo. SOX4 is demonstrated to be required in a cell-autonomous capacity for efficient endocrine cell differentiation both prior to and downstream of NEUROG3. As such, the function of SOX4 during pancreatic endocrine cell formation mirrors roles of SOX family genes in other systems: to bridge key developmental transitions and couple fate decisions.

Methods

Chemicals and reagents

Chemicals were purchased from Sigma Aldrich (Oakville, ON, Canada) or Fisher Scientific (Ottawa, ON, Canada). The antibodies used are listed in electronic supplementary material (ESM) Table 1. Oligonucleotides were purchased from IDT (Coralville, IA, USA; ESM Table 2). Tissue culture reagents were from Hyclone (Logan, UT, USA) and cultureware from BD-Falcon (Fisher Scientific).

Tissues

The University of British Columbia (UBC) Animal Care Committee approved all mouse breeding and experiments, with E0.5 set as noon after discovery of a vaginal plug. Mouse strains included membrane-targeted tdTomato/membrane-targeted enhanced (e) green fluorescent protein (GFP; mT/mG) [32], Neurog3-Cre recombinase (Cre) [33] (Jackson Laboratory, Bar Harbor, ME, USA), Pdx1-Cre [17] (Mutant Mouse Regional Resource Centers [MMRRC], Davis, CA, USA) and Sox4 flox [34]. Pdx1-Cre Sox4 flox/flox (PS4KO) and Neurog3-Cre Sox4 flox/flox (ES4KO) controls were Cre negative and Sox4 flox/WT or Sox4 flox/flox littermates (where WT is wild type). Neurog3-Cre mT/mG Sox4 flox/flox (EMS4KO) controls were: Neurog3-Cre mT/mGflox/WT or mT/mGflox/flox and Sox flox/WT double heterozygotes. The University of Alberta Research and Transplant Islet Isolation Cores obtained human islets under informed consent.

Tissue harvest and preparation

Tissue was harvested and fixed in 4% (wt/vol.) paraformaldehyde in PBS (4°C). Tissue samples were then dehydrated and paraffin embedded or frozen as described [35]. Sections, 5 μm paraffin or 10 μm frozen, were rehydrated as necessary prior to antigen retrieval.

Immunostaining and imaging

Please see ESM for antibodies, in situ hybridisation, TUNEL and 5-ethynyl-2′-deoxyuridine (EdU) staining methods. Antigen retrieval was performed in citrate buffer (pH 6 at 95°C for 20 min). Slides were blocked for 1 h each in 5% (vol./vol.) horse serum followed by donkey anti-mouse IgG. Sections were incubated overnight at 4°C with primary antibody and for 1 h at 20°C with secondary antibodies plus nuclear stains. Tyramide signal amplification (TSA) was carried out for detection of SOX4 (Perkin Elmer, Woodbridge, ON, Canada). Imaging was carried out with sequential scanning on a Leica SP8 confocal imaging system (Concord, ON, Canada).

Morphometric analyses and quantification

Serial sections were chosen at 100 or 50 μm intervals for E18.5 or E15.5 embryos, respectively. The whole section was imaged and tiled and immunopositive cells were counted using CellProfiler (Cambridge, MA, USA) [36]. Counts were normalised to total pancreatic nuclei using CellProfiler or the sum of pancreatic area calculated using Image-Pro Analyzer (Media Cybernetics, Rockville, MD, USA).

Adenoviral generation, infection and cell culture

Sox4 adenovirus was generated using the pAdTrack system (Addgene, Cambridge, MA, USA) [37]. Viruses were purified with the Adeno-X Maxi adenovirus purification system (Clontech, Mountain View, CA, USA). Please see ESM Methods for gene targeting, human embryonic stem cells (hESCs) and FACS methods. Mouse pancreatic ductal adenocarcinoma (mPACs) cells were infected at a multiplicity of infection (MOI) 100:1 for 2 h. Cells were washed briefly in PBS, media was replaced and they were cultured for 48 h.

Real-time PCR

RNA was isolated with TRIzol and DNAse treated and cDNA was synthesised using Superscript II reverse transcriptase (Life Technologies, Burlington, ON, Canada). A 40 ng sample of cDNA was used as template for quantitative (q)PCR carried out in a ViiA7 qPCR machine (Life Technologies). Expression was determined using the ∆∆Ct method with glucouronidase B (Gusb) as a reference gene [35].

Luciferase assay

mPACL20 cells were seeded (2 × 105 cells) in 24-well plates and transfected using Lipofectamine 2000 (Life Technologies) with 1 μg reporter, 50 ng cytomegalovirus promoter (pCMV)-renilla luciferase and 100 ng expression vectors, balanced with empty expression vector. Cells were incubated for 48 h prior to determination of luciferase activity using the Dual-Luciferase Reporter (DLR) Assay (Promega, Madison, WI, USA) on a SpectraMaxL luminometer (Molecular Devices, Sunnyvale, CA, USA).

Chromatin immunoprecipitation

Chromatin immunoprecipitation (ChIP) was performed as previously described [35]. The only change being that ChIP was carried out in the presence of 160 mmol/l NaCl, 0.25% (wt/vol.) deoxycholate and 0.025% (vol./vol.) Triton X-100.

Statistical analysis

All statistical analyses were performed with Prism5 (GraphPad Software, La Jolla, CA, USA). Two-tailed unpaired Student’s t tests or one-way ANOVA with Tukey post-hoc tests were performed, with p ≤ 0.05 considered significant. Error bars represent SEM.

Results

SOX4 is dynamically expressed during mouse pancreas formation

As seen in Fig. 1a, Sox4 is expressed throughout pancreas development, with the highest levels at E11.5. A gradual decline in Sox4 message was observed; however, significant expression was observed during the secondary transition when the majority of NEUROG3+ endocrine progenitors are derived (E12.5–14; Fig. 1a). During human embryonic stem cell differentiation SOX4 is expressed at high levels in stage 3 and 4 pancreatic progenitors (Fig. 1b).

Fig. 1
figure 1

SOX4 is expressed during pancreas development and becomes restricted to endocrine cells. qPCR analyses of Sox4 expression in E11.5 to E18.5 mouse pancreas (a) and following directed differentiation of hESCs (b). Coimmunostaining of SOX4 (green) (c–f) with: (c) NKX6-1 (red), E11.5; (d) PDX1 (red), E12.5; and (e, f) NEUROG3 (red) (e) E14.5 and (f) E15.5. n ≥ 3; scale bars, 50 μm. TBP, TATA-binding protein

Full size image

Because suitable antibodies were not initially available, in situ hybridisation was carried out to localise Sox4 expression (ESM Fig. 1). The epithelium at this stage was cord-like with GCG+ cells found in proximity to the trunk (ESM Fig. 1a). Low expression of Sox4 mRNA was observed throughout the epithelium (E14.5; ESM Fig. 1b). Occasional cells expressing higher levels of Sox4 were noted scattered through the epithelium (arrowheads ESM Fig. 1b): an expression pattern reminiscent of the bipotent trunk cells that activate NEUROG3 [13, 38]. Next, immunostaining was used to determine the spatiotemporal distribution of SOX4. Early during pancreatic organogenesis (E11.5 and E12.5), SOX4 co-localised with PDX1 and NKX6.1 in the multipotent progenitors and trunk cells respectively (Fig. 1c, d; ESM Fig. 2b) [39]. Following the secondary transition, SOX4 was progressively excluded from pancreatic epithelial cells; however, it remained highly expressed in all NEUROG3-expressing endocrine progenitors (Fig. 1e, f) and in nascent endocrine cells (ESM Fig. 2b). By E17.5, SOX4 expression was confined largely to endocrine cells (ESM Fig. 2d).

To verify that SOX4 was expressed in NEUROG3+ endocrine progenitor cells, the mouse mPACL20 pancreatic ductal cell line, which has previously been used to model endocrine cell differentiation, was used [40, 41]. Using transcription activator-like effector nuclease (TALEN)-driven homologous recombination in the mPACL20 cells, a haemagglutinin (HA) tag followed by 2A-GFP was knocked onto Sox4 (mPAC-Sox4-HA; Fig. 2a) and 11 clonal lines were derived with 23% targeting efficiency.

Fig. 2
figure 2

Endogenous SOX4 is induced by NEUROG3 in the mPAC model of endocrine development. Schematic of Sox4-HA-2A-eGFP knock-in add-on allele; 3′ red box is stop codon (a; see ESM Methods for details). Immunofluorescence analyses show induction of HA-tagged SOX4 and GFP following treatment with Ad-Neurog3 in mPAC-Sox4-HA cells (b) but not in control cells (c). Western blotting of mPAC-Sox4-HA cell lysates showed an increase in HA and GFP immunoreactivity at predicted electrophoretic mobilities (∼47 kDa and ∼70 kDa+) following Ad-Neurog3 transduction (d). Sox4 expression is increased by Ad-Neurog3 in unmodified mPACL20 cells (e). n ≥ 3; *p ≤ 0.05. βgal is also known as bgaL. GAPDH, glyceraldehyde-3-phosphate dehydrogenase

Full size image

As predicted from the immunostaining results (Fig. 1c–f), when this clonal mPAC-Sox4-HA cell line was transduced with a Neurog3 adenovirus (Ad-Neurog3) [41], marked increases in endogenous promoter activity led to appreciable HA and GFP that were detectable by both immunostaining (Fig. 2b, c) and western blotting (Fig. 2d). In support, unmodified mPACL20 cells transduced with Ad-Neurog3 showed a 240-fold increase in Sox4 message expression (Fig. 2e).

These observations show that SOX4 is widely produced in the multipotent pancreatic progenitors prior to endocrine differentiation and that its expression is maintained within the endocrine progenitors. In order to assess whether there is an important cell-autonomous role for SOX4 during pancreatic or endocrine cell formation, loss-of-function approaches were used.

SOX4 is necessary for endocrine progenitor cell formation

To determine if SOX4 is important prior to endocrine progenitor formation, mice harbouring a floxed Sox4 allele (Sox4 flox) were crossed with Pdx1-Cre transgenic mice to generate pancreatic Sox4-knockout embryos (PS4KO). Lineage analyses indicate that recombination is extremely efficient early during pancreas organogenesis (>95% of epithelial cells at E12.5; ESM Fig. 2a). In agreement, SOX4 immunostaining at E15.5 verified that SOX4 protein level was significantly reduced compared with controls (ESM Fig. 2b, c).

Quantification of endocrine cell types at E18.5 revealed that pancreatic deletion of Sox4 significantly reduced INS+ GCG+ (Fig. 3a, b, i, j), somatostatin+ ([SST+] Fig. 3c, d, l), and pancreatic polypeptide+ ([PPY+] Fig. 3e, f, m) cells, without changes in ghrelin+ (GHRL+) cells (Fig. 3c, d, k). To address the possibility of a defect in early endocrine cell differentiation, bipotent SOX9+ cells and NEUROG3+ endocrine progenitor cells were quantified. A significant reduction in NEUROG3+ cells, but not SOX9+ cells, was observed (Fig. 3g, h, n, o); suggesting that SOX4 is dispensable until endocrine differentiation is initiated within the SOX9+ progenitor [38, 42]. As expected, neither apoptosis (ESM Fig. 3e–h) nor proliferation (ESM Fig. 3c, d) accounted for the differences in NEUROG3+ cells at E15.5.

Fig. 3
figure 3

PS4KO mice have defects in endocrine cell genesis. Immunohistochemical analyses of hormone immunoreactive (+) cells in control (a, c, e, g) and PS4KO (b, d, f, h) pancreas: (a, b) INS+ (green) and GCG+ (red), E18.5; (c, d) GHRL+ (green) and SST+ (red) E18.5; (e, f) PPY+ (green) and INS+ (red), E18.5; (g, h) NEUROG3+ (green) and SOX9+ (red), E15.5. PS4KO (black bars) had significantly reduced numbers of INS+ (i), GCG+ (j), SST+ (l), PPY+ (m) and NEUROG3+ (n) cells but not in GHRL+ (k) or SOX9+ (o) cells compared with controls (white bars). Sox4 expression (p) was significantly reduced at E15.5 in PS4KO pancreas without changes in Sox11 (q) or Sox12 (r). n ≥ 3; *p ≤ 0.05; scale bars, 50 μm; nuclei are stained blue

Full size image

Despite the marked phenotype and loss of SOX4 protein (ESM Fig. 2c), Sox4 message was significantly reduced but detectable in knockout pancreas by qPCR (Fig. 3p). This may result from a long half-life of Sox4 message or expression from neurons, blood vessels and mesenchyme that are closely associated with the epithelium. Expression of related SOXC class family members, Sox11 and Sox12 (Fig. 3q, r), was not significantly different, suggesting their upregulation did not compensate for loss of Sox4. In sum, these findings demonstrate SOX4 is essential for efficient endocrine progenitor formation.

SOX4 activates the Neurog3 promoter and induces NEUROG3 expression

To investigate whether the cooperative actions of NEUROG3 and SOX4 are important for endocrine cell genesis, unmodified mPACL20 pancreatic ductal cells were transduced with Ad-Sox4 and Ad-Neurog3 (Fig. 4a) [41, 42]. Co-transduction of mPACL20 cells resulted in a significant increase in endogenous Neurog3 expression (Fig. 4a).

Fig. 4
figure 4

SOX4 directly regulates Neurog3 expression in the mPAC model of endocrine development. Ad-Neurog3 and Ad-Sox4 cooperatively increase expression from the Neurog3 locus in mPACL20 cells (a). HA-ChIP (b) following Ad-Neurog3 showed HA-tagged SOX4 protein enriched at the Neurog3 promoter. Luciferase assays demonstrate that SOX4 activates a truncated 2 kb fragment of Neurog3 in mPACL20 cells (c). n ≥ 3; *p ≤ 0.05. RLU, relative light units

Full size image

In order to determine if SOX4 binds and directly regulates NEUROG3 in the context of development, the modified mPAC-Sox4-HA cells were transduced with Ad-Neurog3 to induce endogenous Sox4 (Fig. 2e) and ChIP was carried out using anti-HA antibodies (Fig. 4b). Significant enrichment of SOX4-HA at a conserved region of the Neurog3 promoter containing two SOX4 consensus binding sites (ESM Fig. 4) was observed. Luciferase assays next demonstrated that SOX4 can transactivate a 2 kb fragment of the Neurog3 promoter (Fig. 4c) [43]. Collectively, these results suggest that SOX4 expression within a subset of bipotent trunk progenitors enhances NEUROG3 expression and drives subsequent endocrine progenitor cell formation.

SOX4 regulates endocrine cell differentiation downstream of NEUROG3

SOX4 expression is maintained at high levels in both mouse and human islets (ESM Figs 2 and 5); however, NEUROG3 expression is not [44]. In order to test if the maintenance of Sox4 expression downstream of Neurog3 remains important for islet cell formation, mice harbouring Sox4 flox were crossed with the Neurog3-Cre transgenic mice (ES4KO). Because reduction of SOX4 expression would likely occur with some delay following Neurog3 induction in the ES4KO mice, we hypothesised that no difference in the number of NEUROG3+ cells would be observed. To test this, we analysed the number of NEUROG3+ cells in the ES4KO pancreas at E15.5 and found no difference between genotypes (Fig. 5a–c). Despite this, quantification of endocrine cells within E18.5 ES4KO pancreas demonstrated significant reductions of INS+, GCG+ (Fig. 5d–g), SST+ (Fig. 5h, i, k) and SOX4+ (ESM Fig. 2e) cells but not GHRL+ (Fig. 5h–j) or PPY+ (Fig. 5l–n) cells.

Fig. 5
figure 5

ES4KO embryos have defective endocrine cell formation. Pancreatic immunostaining for (a, b) NEUROG3 (green) and SOX9 (red) and subsequent morphometric analyses showed no difference in NEUROG3+ (c) or SOX9+ cells (not shown) in the ES4KO (compared b vs a). Immunostaining in control (d, h, l) and ES4KO (e, i, m) E18.5 pancreases for: (d, e) INS+ (green) and GCG+ (red), (h, i) GHRL+ (green) and SST+ (red) and (l, m) PPY+ (green) and INS+ (red). Morphometric analyses showed that ES4KO (black bars) have significantly reduced INS+ (f), GCG+ (g) and SST+ (k) cell populations but normal GHRL+ (j) and PPY+ (n) cell populations compared with controls (white bars). n ≥ 3; *p ≤ 0.05; scale bars, 50 μm; nuclei are stained blue

Full size image

Loss of SOX4 does not change the number of Neurog3 lineage cells

Because a significant reduction in endocrine mass was observed in the ES4KO mouse and SOX4 has previously been shown to be important for cell survival and cell proliferation [45], the effects of loss of SOX4 on proliferation (ESM Fig. 6a–c) and apoptosis (ESM Fig. 6d–g) were tested: neither was affected.

Previous studies have suggested ‘dedifferentiated’ endocrine lineage cells maintain expression of the neuroendocrine secretory granule protein chromogranin A (CHGA) [46]. Therefore, in order to address the possibility that islet endocrine cells do not fully differentiate in ES4KO mice, CHGA immunostaining was carried out. Remarkably, despite the reduction in INS+ (Figs 5d, e and 6b) and other endocrine cells (Fig. 5) no difference in the number of CHGA+ cells (Fig. 6a) was observed. Furthermore, co-staining for INS and GCG with CHGA uncovered a significantly increased number of cluster-forming (dotted area) CHGA-expressing cells that did not co-stain for either of these hormones (Fig. 6c, d). These data suggest that loss of SOX4 in the ES4KO impacts normal islet endocrine cell differentiation or maturation downstream of NEUROG3.

Fig. 6
figure 6

Sox4-null endocrine progenitor cells remain in the endocrine lineage but do not differentiate into islet endocrine cells. CHGA+ (a), but not INS+ (b), cells are present in normal numbers in the E18.5 ES4KO (black bars). Significant numbers of CHGA+ INS− GCG− cells can be appreciated in the ES4KO (d, dotted area) but not in control animals (c, arrowheads). There is no change in the total number of labelled endocrine lineage GFP+ cells in the E18.5 ES4KO pancreas (e). However, significantly increased numbers of GFP+ cells that did not co-stain for either INS or GCG were observed (f, h, dotted area) compared with littermate controls (white bars, g, arrowheads). n ≥ 3; *p ≤ 0.05; scale bars, 50 μm; nuclei are stained blue

Full size image

To show that these CHGA-expressing non-INS+/GCG+ cells were derived from the Sox4-null Neurog3 lineage, mice harbouring a ROSA26 mT/mG reporter allele [32] were crossed with ES4KO mice (EMS4KO). In this model, Cre expression in endocrine progenitors drives a switch from red fluorescence (RF) to green fluorescence (GF). Neurog3-lineage GFP+ cells were quantified and no reduction in the total number of GFP+ cells in the EMS4KO mice was found (Fig. 6e), supporting the observation that neither cell proliferation nor death is impacted. Despite a similar number of pancreatic GFP+ cells, a significant increase in non-INS+/GCG+-staining GFP+ cells was observed in the EMS4KO mice (Fig. 6f–h). Notably, no differences in the numbers of GFP+ cells that expressed the acinar cell marker amylase (ESM Fig. 7a, b, e) or the ductal cell marker Dolichos biflorus agglutinin (DBA)-lectin (ESM Fig. 7c, d, f) were observed. These experiments reveal that Sox4-knockout cells remain present within the pancreas, incapable of differentiating into islet hormone-expressing endocrine cells and that they persist in a state not normally present in the pancreas, with CHGA+ and synaptophysin+ (not shown) but no islet hormone expression.

SOX4 regulates factors downstream of Neurog3

As lineage tracing revealed that Sox4-null Neurog3 lineage cells remained fated to a neuroendocrine lineage, the possibility that SOX4-regulated genes important for beta cell differentiation downstream of Neurog3 was explored. To that end, qPCR expression analyses were carried out at E17.5 (Fig. 7a–j). The expression profiles of the endocrine hormones (Fig. 7a–e) mirrored tissue morphometry results (Fig. 5). Importantly, gene expression of at least two transcription factors known to be downstream of endocrine cell specification, Neurod1 and Pax4, was significantly reduced (Fig. 7f, g).

Fig. 7
figure 7

SOX4 potentiates the induction of pro-beta cell genes during endocrine pancreas development. Ins2 (a), Gcg (b), Sst (d), Ppy (e), Pax4 (f), Neurod1 (g), and Sox4 (h) message levels were significantly reduced, while Ghrl (c), Sox11 (i), and Sox12 (j) were not in E17.5 ES4KO (black bars). Pax4 (k) and Neurod1 (l) expression was induced following combined Ad-Neurog3 and Ad-Sox4 in mPACL20 cells. SOX4 significantly induced Neurod1 promoter-driven luciferase activity in mPACL20 cells (m). ChIP demonstrated that HA-tagged SOX4 was enriched at the Pax4 and Neurod1 promoters in the Ad-Neurog3-transduced mPAC-Sox4-HA cells (n; n = 3). qPCR analyses of Sox4 (o); Neurog3 (p); Neurod1 (q); and Pax4 (r) expression in newborn endocrine cells from E17.5 control and EMS4KO mice. A, Tomato+; B, Tomato+ GFP+ (newborn); and C, GFP+ (Neurog3 lineage). n ≥ 3; * p ≤ 0.05. RLU, relative light units

Full size image

In order to test whether SOX4 binding to the Pax4 and Neurod1 promoters was important for beta cell formation, mPACL20 cells were transduced with Ad-Neurog3 and/or Ad-Sox4, and both Pax4 (Fig. 7k) and Neurod1 (Fig. 7l) expression was assessed using qPCR. Remarkably, in this system the combination of both Ad-Neurog3 and Ad-Sox4 induced Pax4 expression ∼200-fold and Neurod1 expression ∼tenfold compared with Neurog3 alone. Neurod1 promoter-driven luciferase transcription [47] was significantly induced with co-transfection of Sox4 (Fig. 7m). HA-ChIP on Ad-Neurog3-transduced mPAC-Sox4-HA cells demonstrated interaction of SOX4-HA at potential binding sites within both the Pax4 and Neurod1 (Fig. 7n) promoters, providing further evidence for the direct regulation of these beta cell transcription factors by SOX4.

In order to understand if Pax4 and Neurod1 are SOX4 targets in vivo, FACS purification of ‘newborn’ Neurog3-lineage cells was carried out [32, 48] (Fig. 7o–r; ESM Fig. 8) using the Neurog3-Cre mT/mG mouse. We reasoned that newly born cells would possess both RF and GF and FACS purification of these cells demonstrated significant enrichment of both Neurog3 and Sox4, indicating that SOX4 is indeed enriched in the newborn endocrine progenitor pool in vivo (Fig. 7o, p). Assessment by FACS suggests that Neurog3-Cre mediated SOX4 inactivation is robust in the EMS4KO line, with over 80% loss of Sox4 message in the endocrine lineage (Fig. 7o). More robust Sox4 message detected by whole pancreas qPCR (Fig. 7h) may be due to low-level expression in the exocrine tissues, which form the bulk of the pancreas. In agreement with in vitro studies, significant reductions in expression of Pax4, Neurod1 and Ins2 (but not aristaless related homeobox [Arx]) were observed in the Neurog3-lineage cells from the EMS4KO (Fig. 7q, r; ESM Fig. 8c, d). The previous demonstrations that Pax4 and Neurod1 are directly downstream of Neurog3 and the observation that expression of these factors was reduced in the newborn (RF + GF) endocrine cells that expressed normal Neurog3 levels (Fig. 7p), lends credence to them being direct, biologically important, SOX4 targets in vivo.

Discussion

Many of the developmental transitions that underlie the formation of beta cells are marked by expression of unique SOX proteins. In embryonic stem cells, SOX2 helps maintain the network of factors that are important for the pluripotent state [49]. SOX factors remain important downstream of SOX2 during definitive endoderm formation, when octamer-binding transcription factor 4 (OCT4) complexes with SOX17 to regulate endoderm induction [50]. SOX17 then likely interacts with undefined factors until expression wanes in the pancreatic buds at the time when SOX9 is induced [51]. SOX9 establishes the gene expression programme necessary for pancreatic budding and growth and remains important in the formation of the endocrine progenitor cells [52]; however, it is silenced in the endocrine lineage.

This study provides the first evidence of a cell-autonomous role for SOX4 during normal endocrine cell formation. Based on our data, we propose a model (Fig. 8) where SOX4 within a subset of multipotent epithelial progenitor cells stabilises Neurog3 transcription and endocrine progenitor fate. SOX4 then couples endocrine progenitor cell differentiation with subsequent development through cooperation with NEUROG3.

Fig. 8
figure 8

Model highlighting a role for SOX4 during pancreas development. SOX4-expressing bipotent progenitor cells (red) activate high levels of NEUROG3 expression (yellow). SOX4 and NEUROG3 cooperate following Neurog3 activation to increase NEUROG3 levels and also to increase expression of downstream targets Pax4 and Neurod1, genes necessary for mature endocrine cell formation (light and dark blue)

Full size image

SOX4 has been implicated in development of a wide array of mammalian tissues. As such, the previous pancreatic studies did not definitively show that cell-autonomous actions of SOX4 in the pancreatic epithelium were necessary [28, 31, 53]. Furthermore, the early studies did not uncover when SOX4 becomes important for islet cell formation or how it functions to regulate islet cell genesis. Finally, these previous studies did not attempt to assess whether SOX4 expression in the exocrine progenitors is important for their differentiation. Herein, we address all of these previous questions and for the first time show that SOX4 expression within the trunk cells is necessary for endocrine cell development and that loss of SOX4 impacts both formation of progenitors and terminal differentiation of all islet cells but not the exocrine tissue.

Of note, Danio rerio Sox4b, an orthologue of Sox4, was transiently expressed in endocrine precursor cells and knockdown of SOX4B uncovered a requirement for alpha cell differentiation. Other endocrine cell types, however, were unchanged [29]. The differences in observed phenotypes between our study and this former one may be attributed in part to unique endocrine differentiation pathways present in fish and mice or to the differences in knockout/knockdown approaches used.

In the multipotent pancreatic progenitor cells, Neurog3 expression is repressed through active NOTCH signalling [54, 55]. Endocrine cell differentiation requires loss of NOTCH signalling and binding of a number of transcription factors to the Neurog3 promoter including: one cut homeobox 1 (ONECUT1), SOX9, hepatocyte nuclear factor 1β (HNF1B), forkhead box protein A2 (FOXA2) and NEUROG3 itself [15, 42, 54, 56–58]. This study adds to the cadre of factors that regulate NEUROG3 by demonstrating that SOX4 participates in this NEUROG3 regulatory network.

SOX4 remains important for endocrine differentiation downstream of its initial role in activating NEUROG3 protein expression. Pax4 and Neurod1, which lie directly downstream of Neurog3 and are essential during beta cell formation [24, 59], were significantly reduced in the ES4KO pancreas. Null mutation of Neurod1 leads to reductions in alpha, beta and delta cells [20] while pancreatic Pax4-knockouts have reductions of both beta and delta cells and increases in the other endocrine cells including GHRL+ epsilon cells [24, 60, 61]. Notably, a dramatic increase in epsilon cells was not observed following loss of SOX4, suggesting that residual Pax4 expression is sufficient to suppress epsilon cell formation. Interestingly, the highly abundant epsilon cells in the Pax4-null pancreas also express CHGA [60]; thus, the possibility remains that loss of SOX4 drives endocrine differentiation to an unappreciated terminally differentiated non-islet endocrine cell type.

A number of studies have characterised transcription factors and their regulatory networks in pancreatic development. This knowledge has catalysed advancements of directed differentiation protocols and has brought cell-based strategies for the generation of functional INS-producing cells to clinical trials. In conclusion, this work expands the network of transcription factors that are directly important for beta cell formation, and future studies that aim to understand how SOX4 is induced within endocrine progenitors may refine protocols for differentiation of beta cells from hESCs.