The L6 domain tetraspanin Tm4sf4 regulates endocrine pancreas differentiation and directed cell migration

The pancreas develops through a series of intricate steps involving highly coordinated signaling, transcriptional and morphological events that are tightly regulated to produce mature ductal, exocrine and endocrine tissue compartments (for reviews, see Gittes, 2009; Jorgensen et al., 2007; Murtaugh, 2008; Oliver-Krasinski and Stoffers, 2008; Puri and Hebrok, 2010). Despite the identification of many different transcriptional components that are important for pancreas development, the downstream mechanisms that ultimately lead to specification or differentiation of pancreatic cell types are still not well understood.

During mouse pancreatic development, the critical stage of islet cell differentiation occurs between embryonic day (e) 12.5 and e15.5 (Oliver-Krasinski and Stoffers, 2008; Rutter et al., 1968), at which time endocrine progenitor cells, which are demarcated by their expression of the basic helix-loop-helix transcription factor Ngn3 (Neurog3 – Mouse Genome Informatics), localize and multiply in the branched epithelial pancreatic ducts. As the Ngn3⁺ population differentiates, Ngn3 becomes downregulated once differentiating endocrine cells lose polarity, delaminate and migrate from the ducts into the surrounding mesenchyme and developing acinar environment (Gouzi et al., 2011; Slack, 1995). Identifying the extracellular factors that are important for maintenance of the progenitor population by regulating adhesion, migration and/or differentiation is crucial for understanding pancreas development, as well as for potential therapeutic design.

In zebrafish, many aspects of islet development are similar to those in the mouse. The posterior (dorsal) bud gives rise to the principal endocrine islet and appears at 24 hours post-fertilization (hpf) whereas the anterior (ventral) anlage is not evident until 40 hpf and develops into exocrine, ductal and some endocrine tissues (Field et al., 2003; Tiso et al., 2009). It has been proposed that, similar to mouse, Notch-responsive cells initiate a secondary transition in the developing zebrafish pancreas (Parsons et al., 2009). Although zebrafish embryos lack ngn3 (neurog3 – Zebrafish Information Network) expression during early pancreas development (Zecchin et al., 2007), endocrine precursors express the transcription factor sox4b in a domain ventral to the developing principal islet at 24 hpf (Mavropoulos et al., 2005; Soyer et al., 2010). sox4b is expressed transiently in endocrine precursors and responds to Notch signaling in the same manner as Ngn3 in mice (Mavropoulos et al., 2005).

Although the endocrine progenitor program may be initiated by different transcription factor(s) in mouse and zebrafish, many of the upstream and downstream transcription factors or signaling molecules have conserved roles in the two organisms (Biemar et al., 2001; Kinkel and Prince, 2009). The homeodomain transcription factor Nkx2.2 is crucial for both mouse and zebrafish islet cell differentiation. Nkx2.2^–/– mice completely lack β cells, have an 80% decrease in α cells and a modest reduction in PP cells. Conversely, ghrelin⁺ (ε) cells are significantly increased and fill the islets in Nkx2.2^–/– mice (Prado et al., 2004; Sussel et al., 1998). These islet cell differentiation defects are conserved in zebrafish pancreas organogenesis when nkx2.2a translation is blocked (Pauls et al., 2007).

Analysis of the genome-wide expression changes that occur in the mouse Nkx2.2^–/– pancreas at onset of the secondary transition identified a tetraspanin-like protein, Tm4sf4, as a direct target of Nkx2.2 that is highly upregulated in Nkx2.2^–/– pancreata (Anderson et al., 2009a; Hill et al., 2011). Tetraspanins are defined by their similar protein topology and predicted tertiary structure. They can interact with other tetraspanins, integrins, growth factor receptors and signaling molecules that occur both extracellularly and intracellularly to form large tetraspanin-enriched microdomains (TEMs). TEMs have been shown to function as signaling centers crucial for adhesion, migration and proliferation (Hemler, 2005; Hemler, 2008; Yunta and Lazo, 2003).

Tm4sf4 is classified as a member of the more divergent tetraspanin L6 domain family owing to a lack of characteristic cysteine residue motifs in the EC2 extracellular domain (Wright et al., 2000). Other members of the L6 family include Tm4sf1 (L6-Ag), Tm4sf5 and Tm4sf18 (L6D) (Wright et al., 2000). Tm4sf1 and Tm4sf5 have been shown to be upregulated in multiple tumors, have roles in epithelial-to-mesenchymal transition (EMT) and affect migratory mechanisms crucial to cancer invasion and metastasis (Kao et al., 2003; Lee et al., 2008; Marken et al., 1992; Muller-Pillasch et al., 1998; Muschel and Gal, 2008; Storim et al., 2001). Unlike its homologs, little mechanistic information is known about Tm4sf4. Tm4sf4 levels appear to increase when non-dividing epithelial cells differentiate and migrate out of intestinal crypts (Wice and Gordon, 1995). In the liver, Tm4sf4 is expressed in non-dividing hepatocytes that retain high proliferative potential when given the correct stimulus, and is upregulated during liver injury (Liu et al., 2001; Qiu et al., 2007). In the context of Nkx2.2 loss of function, when endocrine differentiation is severely disrupted, Tm4sf4 is highly upregulated. In this study, we describe novel roles for Tm4sf4 in regulating islet cell fates and inhibiting cellular migration that might provide a principal mechanism for Nkx2.2 function in the pancreas.

Nkx2.2^+/– heterozygous mice (Sussel et al., 1998) and Ngn3^+/EGFP (Lee et al., 2002) heterozygous mice were maintained on Swiss Black (Taconic) backgrounds. Genotyping of mice and embryos was performed as previously described (Sussel et al., 1998; Lee et al., 2002). Mice were housed and treated according to Columbia University and UCD-AMC IACUC approval protocols.

Zebrafish were maintained and embryos staged according to standard protocols (Kimmel et al., 1995). For initial experiments screening morpholino phenotypes and function, wild-type AB, TAB and EKK strains were used. All subsequent experiments were done with single mating wild-type EKK embryos.

Mouse RNA ISH was performed as previously described (Prado et al., 2004). A full-length mouse Tm4sf4 cDNA clone (Open Biosystems/ThermoScientific, AL, USA) was used to generate the antisense probe. Pictures were acquired on a Leica CTR 5000 with 20× magnification.

Zebrafish whole-mount ISH was performed as described by Thisse and Thisse (Thisse and Thisse, 2008), digoxigenin (DIG)-labeled antisense RNA probes: tm4sf4 (PCR cloned: Forward 5′-ATCATGTGCTCTGGAAATTTCGCC-3′, Reverse 5′-TTACTTTATTCCTTGCAGCAGCCG-3′); insulin, glucagon, somatostatin and ghrelin [synthesized from cDNAs provided by Chris Wright (Vanderbilt University, TN, USA)] and sox4b [Marianne Voz (GIGA, Sart-Tilman, Belgium)]. Images were acquired on an Olympus BX51 with 20× magnification.

Mouse immunofluorescence was performed on frozen 8 μm sections. Images were acquired on a Leica DM5500. For cell staining, cells were grown and fixed on coverslips or Boyden Chamber filters and images acquired on a Zeiss META LSM510. Rhodamine phalloidin was incubated for the last 30 minutes of the secondary antibody incubation.

Whole-mount 48 hpf zebrafish embryos were fixed in 4% paraformaldehyde (PFA) for 1 hour at 25°C. Embryos were deyolked, followed by 2×5 minute water washes and 1 hour in blocking solution [2% goat serum, 2% BSA, 1% DMSO, 0.2% Triton X-100, in 1× PBS]. Embryos were imaged on a Zeiss META LSM 510. Confocal z-stacks were taken every 2 μm and compiled by 3D rendering using Imaris software (Bitplane Scientific Software, MN). Glucagon- and Insulin-positive cells were counted per islet, n=6 for both uninjected and tm4sf4 morphant conditions.

Antibodies and markers used were: goat anti-CpA (1:800; R&D Systems, MN, USA), mouse anti-FLAG (1:500; Sigma, MO, USA); rabbit anti-ghrelin (1:800; Phoenix Pharmaceuticals, CA, USA); guinea pig anti-glucagon (1:1000; Millipore, MA, USA); mouse anti-Glucagon (1:300; Sigma); guinea pig anti-insulin (1:1000 or 1:200; Millipore); rabbit anti-Sox9 (1:1000; Chemicon); DBA-lectin (1:100; Vector Laboratories, CA, USA); Rhodamine-phalloidin (1:100; Invitrogen, CA, USA); donkey anti-guinea pig-Cy3 (1:200; Jackson ImmunoResearch, PA, USA); donkey anti-guinea pig-Cy5 (1:300; Jackson ImmunoResearch); donkey anti-rabbit-Cy2 (1:200; Jackson ImmunoResearch); donkey anti-mouse-Cy2 (1:200; Jackson ImmunoResearch) and DAPI (1:1000; Invitrogen).

Injection bolus size was measured on a 1 mm ruled:0.01 mm division micrometer (12-561-SM1, Sigma). Ten ng of nkx2.2atgMO (MOnk-5UTR, 5′-TGGAGCATTTGATGCAGTCAAGTTG-3′) (Pauls et al., 2007), 4-8 ng of Translation (tm4atgMO, 5′-AATTTCCAGAGCACATGATTGAGTC-3′) or splice blocking (tm4splMO, 5′-GTTATTGTTTTTCTCACCGCAAATC-3′) morpholinos (Gene Tools, OR, USA) were injected into embryos at the 1- to 4-cell stage. Injected embryos were screened by rhodamine or cascade blue dextran prior to experimental analysis.

RNA from fluorescence-activated cell (FAC) sorted populations (White et al., 2008) or tissue culture cells were isolated with an RNeasy Mini Kit (Qiagen, CA, USA). cDNA was transcribed using a Superscript III Kit (Invitrogen). Pre-designed mouse Tm4sf4 (Mm00523755_m1, Life Technologies, CA, USA) and custom Ngn3 primer/probe Taqman assays were used to measure mRNA expression levels and were normalized to the internal control gene, cyclophilin B, with a custom Taqman primer/probe set. Relative quantification used a standard curve from wild-type cDNA. For zebrafish samples, ten wild type and ten tm4sf4 morphants for each time point were collected and pooled into one sample for each condition. Samples were processed as described above. Each gene of interest was normalized to β-actin expression and the ΔΔCt method (Life Technologies) was used for relative quantification. Mouse and zebrafish sequences are listed in Table S1 in the supplementary material.

mPacL20, NIH 3T3 and 293T cells were cultured in DMEM with 10% FBS. For Tm4sf4-FLAG overexpression, the Tm4sf4 cDNA (MMM1013-65619, Open Biosystems) was cloned into pCMV Sport6 with a reverse primer containing FLAG nucleotides. For overexpression, cells were transfected with 0.8 μg Tm4sf4-FLAG and lipofectamine 2000 (NIH 3T3-5:1 ratio, 293T-4:1 ratio) (Invitrogen). For siRNA knockdown, mPacL20 cells were transfected with 10 pmol predesigned siControl (scrambled siRNA not known to degrade any RNA) or siTm4sf4 (sc-154303, Santa Cruz Biotechnology, CA, USA) and 1 μl lipofectamine 2000. At 40 hours post-transfection, cells were serum starved for 30 minutes, trypsinized and added to medium lacking FBS. Half the cells were added to the Boyden chamber insert (ECM508, Millipore), which was then added to a 24-well plate containing medium with FBS. Cells were allowed to migrate for 24 hours before fixing and staining nuclei. Three 20× magnification fields were imaged per well for each condition in triplicate (total n=9), and the number of migrated cells was counted manually. In knockdown cells, morphology differences were identified by broadly spread cells with intense and disorganized actin cytoskeleton, n=6.

For Rho family activator or inhibitor treatments, mPacL20 cells were allowed to adhere to the insert membrane for 2 hours prior to drug additions with the following final concentrations: 0.25 units/ml Rac/Cdc42 activator (CN02, Cytoskeleton, Denver, CO, USA), 1 μg/ml Rho Inhibitor/cell permeable C3 transferase (CT04, Cytoskeleton), 100 μM ROCK inhibitor Y-27632 (Y0503, Sigma-Aldrich), 0.2 or 0.4 μg/ml pan-small GTPase inhibitor Toxin B (616377, EMD Chemicals, Gibbstown, NJ, USA). Cell migration occurred for 18 hours prior to cell fixation and quantification.

In wild-type pancreata, Tm4sf4 mRNA is expressed at low levels throughout the central trunk domain (Anderson et al., 2009a). Tm4sf4 mRNA is expressed in the pancreatic epithelium as early as e10.5 (Fig. 1A). In the absence of Nkx2.2, Tm4sf4 became highly upregulated at e10.5, e12.5 and e15.5 (Fig. 1A,C,F,I,K,O). Within the endocrine compartment, Tm4sf4 was only co-expressed with a rare number of insulin⁺ cells (Fig. 1S), and high Tm4sf4 mRNA-expressing (Tm4sf4^HI) cells did not appear to overlap with glucagon or ghrelin protein expression in wild-type or Nkx2.2^–/– pancreata (Fig. 1A-D,F,G,I,J). The lack of correlative localization with ghrelin⁺ cells in the Nkx2.2^–/– pancreata suggests that upregulation of Tm4sf4 is not due to elevated numbers of ε cells (Anderson et al., 2009a). Analysis of Tm4sf4 mRNA expression with carboxypeptidase A (CpA) immunohistochemical staining also suggested that CpA^– ductal trunk cells are enriched with Tm4sf4, with lower expression visible in the CpA⁺ distal tip domains and exocrine tissue (Fig. 1S,T). Immunofluorescence staining for the ductal markers DBA-lectin and Sox9 on adjacent sections to Tm4sf4 mRNA expression confirmed enrichment of the Tm4sf4 expression domain in the developing ductal network (Fig. 1E,F,H,I,K,L,O,P).

Based on the predominant localization of high Tm4sf4 mRNA levels to the pancreatic ductal regions, we hypothesized that Tm4sf4^HI is localized to the endocrine progenitor cell population. Using fluorescence-activated cell sorting (FACS) technology on Ngn3-EGFP mouse pancreata (Lee et al., 2002), GFP⁺ (Ngn3⁺, 5.9%) and GFP^– (Ngn3^–, 91.6%) cell populations were collected from e14.5-16.5 pancreata at the peak of Ngn3 expression and cell specification (Fig. 2A). In agreement with the expression analysis, we found that the Ngn3⁺ endocrine progenitor population was highly enriched with Tm4sf4 mRNA during the secondary transition (Fig. 2B). Ngn3^–/– mouse pancreata (Lee et al., 2002), which lack all endocrine cell populations, expressed similar levels of Tm4sf4 (Fig. 2C-E), to further support the localization of Tm4sf4 expression in the central ductal (non-endocrine) domains of the pancreas.

Previous studies have shown that Tm4sf4 is expressed in both liver and intestine (Liu et al., 2001; Qiu et al., 2007; Wice and Gordon, 1995). Although Nkx2.2 is expressed in a subset of intestinal enteroendocrine cells and has a functional role in cell fate choices (Desai et al., 2008), Tm4sf4 intestinal mRNA expression appeared to be unaltered in Nkx2.2^–/– mice (Fig. 1N,R; see Fig. S1 in the supplementary material). Consistent with the absence of Nkx2.2 expression in the liver, Tm4sf4 mRNA levels were unchanged in the liver of Nkx2.2^–/– mice (Fig. 1M,Q). We also determined that the other closely related L6 family members were not expressed at high levels in the pancreas (see Fig. S1 in the supplementary material) and did not display elevated expression in Nkx2.2^–/– pancreata (see Fig. S1 in the supplementary material) (Hill et al., 2011). These findings suggest that regulation of Tm4sf4 by Nkx2.2 is specific to the pancreatic domain and is unique among its L6 counterparts.

To determine the function of tm4sf4 in the developing pancreas, we used zebrafish to knockdown tm4sf4 using antisense morpholino (MO) technology. We identified a single zebrafish Tm4sf4 ortholog (zgc:92479) with 79% similarity to mouse Tm4sf4 (Fig. 3A). Performing temporal qRT-PCR analysis, we determined that tm4sf4 expression initiates at 10 hpf and increases until 168 hpf, with spikes of expression occurring at 20.5, 48 and 144 hpf (Fig. 3B,C). The first spike in expression corresponds to a time period (15-24 hpf) at which endocrine progenitor markers are being induced and β cells are specified (Biemar et al., 2001). Furthermore, 18-21 hpf represents a stage when the endoderm first begins to organize radially near pancreatic and liver domains, forming what will later become the polarized gut tube (Wallace and Pack, 2003). Similar to murine Tm4sf4, zebrafish tm4sf4 was expressed in liver, intestine and pancreas of 48 hpf embryos (Fig. 3D).

A conserved role for Nkx2.2 in pancreas gene regulation, islet specification and duct morphogenesis has previously been shown in zebrafish (Anderson et al., 2009a; Pauls et al., 2007). To determine whether regulation of Tm4sf4 by Nkx2.2 is also conserved in zebrafish, we assessed tm4sf4 expression in nkx2.2MO-injected embryos. We confirmed that the nkx2.2a translation blocking morpholino was functional by measuring insulin and ghrelin mRNA levels at three stages during pancreas development. As previously described (Pauls et al., 2007), insulin was reduced approximately twofold at 52 hpf; however, we also showed this reduction to be apparent as early as 19.5 hpf (Fig. 3H). We determined that ghrelin is also increased by approximately twofold, as shown by Pauls et al. (Pauls et al., 2007), and that this increase in ghrelin occurs by 28 hpf (Fig. 3I). Similar to mice, morpholino knockdown of nkx2.2a resulted in increased tm4sf4 expression specific to pancreas as seen by ISH at 36 hpf (Fig. 3E,F). When mRNA levels were measured by qRT-PCR, tm4sf4 was increased in nkx2.2a morphant embryos by approximately threefold at 19 hpf, ∼1.7-fold at 28 hpf, but appeared to be unchanged by 52 hpf (Fig. 3G). This apparent loss of a tm4sf4 phenotype probably results from analyzing mRNA extracted from whole zebrafish embryos; the high and unaltered expression of Tm4sf4 in the developing liver and intestine could mask any pancreas-specific phenotypes. For a summary of morpholino penetrance see Table 1.

To determine whether tm4sf4 plays a role during pancreas development, we used a translation-blocking morpholino (Tm4atgMO) and a splice-blocking morpholino (Tm4splMO, designed against the exon2-intron2 boundary) to knockdown tm4sf4 expression globally throughout early zebrafish embryogenesis. Using the splice-blocking morpholino, we assessed efficiency of knockdown by detecting the appearance of aberrantly spliced products that would result in severely truncated or altered amino acid sequence (see Fig. S2 in the supplementary material). Injections of a nonsense control morpholino had no effect on hormone, transcription factor or tm4sf4 expression analyzed by qRT-PCR or ISH (data not shown).

We hypothesized that loss of tm4sf4 would lead to an increase in insulin-producing β cells because Nkx2.2^–/– mouse pancreata contain no β cells, tm4sf4 is highly upregulated and these phenotypes are conserved in the nkx2.2a morphant zebrafish embryos. Indeed, knockdown of tm4sf4 by both Tm4atgMO and Tm4splMO resulted in a significant increase in insulin mRNA expression and β cell numbers within the pancreas at 30 hpf by ISH in a dose-dependent manner (see Fig. S3 in the supplementary material). The splice-blocking morpholino (Tm4splMO) was used for all subsequent experiments to allow quantification of knockdown via qRT-PCR. We confirmed that wild-type tm4sf4 transcript was reduced by 50-75% when using 5 ng of the Tm4splMO morpholino (Fig. 4Q). Expression of the liver marker hhex and intestine marker foxa3 appeared to be unaffected (data not shown). To ensure that the phenotype was specific to knockdown of Tm4sf4, we showed that tm4sf4 mRNA co-injected with Tm4splMO could rescue the β cell phenotype and significantly reduce the number of β cells compared with uninjected embryos (see Fig. S3 in the supplementary material).

The phenotype associated with loss of tm4sf4 function was analyzed by measuring endocrine hormone RNA levels at three consecutive stages during pancreas development (16.5-21 hpf, 24-30 hpf and 48-52 hpf). Using ISH analysis, we observed a persistent and significant twofold increase in insulin mRNA expression across the pancreatic developmental time points, beginning at 16.5 hpf, soon after β cells are initially specified (Fig. 4A). ISH for insulin on 20.5 hpf embryos confirmed an increase in the number of insulin⁺ cells (Fig. 4B,C). Between 15 hpf and 24 hpf, bilaterally specified β cells migrate to form a tight cluster adjacent to the midline (Biemar et al., 2001; Kim et al., 2005; Kinkel et al., 2008). Interestingly, the larger β cell population in tm4sf4 morphant embryos appeared less clustered and extended more anteriorly than that of their uninjected counterparts (Fig. 4B,C). Quantification of insulin⁺ β cells by immunofluorescence revealed a significant 1.4-fold increase in β cells per islet in tm4sf4 morphants at 52 hpf (Fig. 5A-C). Similar results were observed in Tg(insulin::dsRed/glucagon:GFP) zebrafish (provided by R. Anderson and D. Stainier, University of California, San Francisco, CA, USA) at 30 hpf (data not shown). It remains possible that the increase in β cells is due to an increase in proliferation of differentiated β cells rather than to an increase in specification; however, temporal analysis suggests that the increase in insulin mRNA occurs early and does not continue to expand over time (Fig. 4A). In addition, we performed 5-ethynyl-2′-deoxyuridine (EdU) incorporation at 30 hpf to label cells actively in S phase of the cell cycle. Morphant embryos displayed no change in EdU⁺ cells within the pancreatic insulin⁺ population or the entire pancreatic domain when compared with uninjected embryos (Fig. 5F,G). Similarly, co-staining of Insulin and Phospho-histone H3 (PH3) on 20.5 hpf embryos did not identify proliferating PH3⁺/Insulin⁺ cells (data not shown). These findings suggest that knockdown of tm4sf4 increases β cell number via increased specification or differentiation.

We determined that glucagon transcript levels were significantly elevated fourfold in tm4sf4 morphant embryos by 16.5 hpf (Fig. 4D) and that by 20.5 hpf, pancreatic α cell number appeared increased, similar to that of β cells (Fig. 4E,F). However, the large early increase in glucagon mRNA disappeared by 24 hpf and remained unchanged compared with uninjected embryos up to 52 hpf (Fig. 4D). At 52 hpf, we confirmed that α cell numbers were unaltered in tm4sf4 morphant embryos (Fig. 5A,B,D). Upon closer analysis of the trunk domain at 20.5 hpf, the tm4sf4 morphant embryos contained a large midline stream of glucagon⁺ cells along the developing intestinal tract, posterior to the pancreas (Fig. 4H,J). These cells underwent apoptosis by 28 hpf (Fig. 5H-K), suggesting that the intestinal glucagon⁺ population in morphant embryos is transient and might explain the lack of a glucagon phenotype at later stages (Fig. 4D). Glucagon⁺ cells within the intestine of uninjected clutch mates at 20.5 hpf were not detected (Fig. 4G,I). The emergence of glucagon⁺ cells at this stage could be due to mis-specification of pancreatic α cells in the intestine or unregulated differentiation of enteroendocrine cells. However, we have only detected intestinal tm4sf4 expression within a short section of the developing gut immediately posterior to the liver and pancreas (Fig. 3B). It remains possible that α cells, glucagon⁺ intestinal cells, or their precursors, are born more anteriorly and migrate posteriorly within the developing gut tube of tm4sf4 morphant embryos.

We determined whether Tm4sf4 also functions downstream of Nkx2.2 to regulate ghrelin-expressing ε cells. Consistent with the increase of α and β cells, ghrelin mRNA expression was decreased in tm4sf4 morphant embryos when compared with uninjected embryos at 48 hpf (Fig. 4K). Although it has recently been suggested that ghrelin-expressing cells are present as early as 22 hpf (Soyer et al., 2010), ghrelin mRNA expression was not significantly changed until 48 hpf in tm4sf4 morphants (Fig. 4K). However, as wild-type ghrelin levels are extremely low prior to 48 hpf, it remains possible that detection of a decrease at these early stages might be challenging. Even by 48 hpf, the zebrafish pancreas contains a relatively low number of cells expressing ghrelin (Fig. 4L) (Pauls et al., 2007). At 48 hpf, tm4sf4 morphant zebrafish had significantly fewer ghrelin-expressing cells (Fig. 4L,M, Fig. 5E).

The early pancreatic domain in zebrafish expresses pdx1 and comprises an endocrine progenitor pool marked by the transcription factor sox4b (Biemar et al., 2001; Mavropoulos et al., 2005; Soyer et al., 2010; Tiso et al., 2009). The Sox4b⁺ population appears to be distinct from hormone⁺ cells at 24 hpf (Soyer et al., 2010). To identify whether the endocrine progenitor/precursor pool could be affected with the loss of Tm4sf4 function, we analyzed the progenitor markers pdx1 and sox4b mRNA levels at 24 hpf. Interestingly, both pdx1 and sox4b levels were significantly increased 1.4- and 1.5-fold, respectively (Fig. 6A-D), in embryos with reduced tm4sf4 (Fig. 4Q). An expanded endocrine precursor pool could explain the increase of α and β cells, but a similar increase in δ and ε cells would also be expected. Instead, we observed a decrease in ghrelin⁺ (ε) cells (Fig. 4K-M, Fig. 5E) and, although we observe subtle increases and decreases in somatostatin mRNA levels by qRT-PCR, we were unable to detect any noticeable variation in somatostatin mRNA or cell numbers by ISH (Fig. 4N-P) or immunofluorescence (data not shown). It remains possible that zebrafish tm4sf4 can coordinate regulation of pdx1 and sox4b expression levels within endocrine precursors, which could affect specific islet cell fate competency (Fukuda et al., 2008; Johansson et al., 2007; Wang et al., 2010). Alternatively, as Pdx1 is also expressed in β cells and expression of sox4b in a small subset of β cells at 20 hpf has been described (Mavropoulos et al., 2005), the increase in pdx1 and sox4b expression might be secondary to the increase in β cell numbers at 24 hpf. However, we were unable to detect any sox4b⁺insulin⁺ double-positive cells in either uninjected or Tm4MO embryos at 20 hpf (data not shown), suggesting that the increase in sox4b is independent of increased β cell numbers.

We have determined that Ngn3⁺ endocrine progenitor cells in mice are enriched with Tm4sf4 (Fig. 2B). Ngn3 can drive EMT in endocrine progenitors, resulting in Snail2 (Snai2 – Mouse Genome Informatics) expression and a change in polarity preceding delamination, migration and differentiation into the five islet cell types (Gouzi et al., 2011; Kim and MacDonald, 2002; Pictet and Rutter, 1972; Rukstalis and Habener, 2007; Rutter et al., 1964). A gap remains in our understanding of the mechanisms linking transcriptional control to cell behaviors crucial for islet development. Interestingly, tetraspanin and L6 domain family members have been implicated in migration and adhesion in many tissues (Hemler, 2005; Wright et al., 2000). To analyze the functional role of Tm4sf4 in different cellular processes, we took advantage of in vitro cell culture systems to overexpress and knockdown Tm4sf4. To confirm the cellular context of Tm4sf4 expression in vitro we first compared mRNA levels in several differentiated mouse pancreatic cell lines; mPacL20 (pancreatic duct cell, Nkx2.2^–), αTC1 (α cell, Nkx2.2⁺) and βTC3 (β cell, Nkx2.2⁺) cells (Gasa et al., 2004; Yoshida and Hanahan, 1994). Consistent with the high levels of ductal staining in the mouse (Fig. 1E,F,H,I), we found that Tm4sf4 is expressed in the ductal mPacL20 cells that lack Nkx2.2, but is undetectable in αTC1 cells and expressed at very low levels in βTC3 cells (see Fig. S4 in the supplementary material).

To determine the intracellular localization of Tm4sf4, we overexpressed a Tm4sf4-FLAG fusion protein in NIH 3T3 and 293T fibroblast cells. As would be expected for an L6 domain tetraspanin molecule, Tm4sf4-FLAG protein was localized to the membrane and was apparent in vesicles (cytoplasmic punctae) (Fig. 7B). In addition, we observed that cells expressing Tm4sf4-FLAG displayed a striking cell morphological phenotype that was characterized by a multitude of filipodial cell protrusions that were not present in the mock-transfected controls (Fig. 7B). Not only did Tm4sf4-FLAG⁺ cells have more protrusions, but a large number of very thin filipodia appeared to be consistently concentrated on one edge of the cell. Frequently, multiple filipodial remnants also appeared to be detached or `ripped' from the cell body (Fig. 7B,C). Membrane ripping has been shown to occur during slower fibroblast and keratinocyte migration where Integrin⁺ cell remnants are thought to provide a track for subsequent cell attachments (Bard and Hay, 1975; Chen, 1981; Kirfel et al., 2003; Palecek et al., 1998). Based on the cell morphological changes resulting from Tm4sf4-FLAG overexpression, we hypothesized that Tm4sf4 might act to stabilize cell protrusions and matrix contacts, resulting in migratory defects.

To test whether Tm4sf4 has a role in migration, we performed directed migration assays in Boyden chambers using both overexpression of Tm4sf4-FLAG in 293T cells, and knockdown of Tm4sf4 in mPacL20 cells using siRNA technology. Using this assay, we determined Tm4sf4-FLAG overexpression in 293T cells (data not shown) or mPacL20 cells (Fig. 7F) caused a significant reduction in directed cellular migration (Fig. 7G). Conversely, we discovered that ∼70% knockdown of endogenous Tm4sf4 in mPacL20 cells (Fig. 7D) led to a significant increase in migrating cells (Fig. 7E). Based on these results, we believe that Tm4sf4 probably functions to inhibit cell detachment and/or cell migration, a crucial mechanism during islet morphogenesis and differentiation.

Directed cell migration requires alterations in the actin cytoskeleton organization and, therefore, we analyzed whether Tm4sf4 affects F-actin organization. NIH3T3 cells that overexpress Tm4sf4-FLAG (Fig. 8A, arrow) had more disorganized F-actin stress fibers emanating from an intense nucleation site. NIH3T3 and 293T cells that overexpress Tm4sf4-FLAG (Fig. 8A-C, arrows) exhibited a loss of defined cortical F-actin at the cell boundaries compared with that of characteristic lamellipodia in Tm4sf4^– cells (Fig. 8A-C, arrowheads). Instead, F-actin appeared to be more concentrated in the many filipodial extensions and cell protrusions of the less migratory cells expressing Tm4sf4-FLAG (Fig. 8C). mPacL20 cells endogenously expressing Tm4sf4 displayed a variety of morphologies (Fig. 8D-F). However, upon treatment with siTm4sf4, threefold more cells within the population spread broadly in all directions with pronounced cortical F-actin and lamellipodial cell boundaries (Fig. 8G-I,K). In addition, cells with presumably lower levels of Tm4sf4 harbored more robust F-actin stress fibers and multiple cytoskeleton nucleation sites (Fig. 8G-I) than was seen in siControl cells.

The Rho family of GTPases (RhoA, Rac1, Cdc42) is known to regulate actin cytoskeleton organization and direct cellular migration and polarity. To identify whether Tm4sf4 dependent inhibition of migration requires Rho family GTPase activity, we first treated siControl or siTm4sf4 mPacL20 cells with the pan-Rho-GTPase inhibitor toxinB at 0.2 or 0.4 μg/ml, which repressed the migration induced by loss of Tm4sf4 (Fig. 8L). To determine which Rho pathway Tm4sf4 might be involved with, we also treated siControl or siTm4sf4 cells with more specific pharmacological effectors: a Rac/Cdc42 activator, RhoA/B/C inhibitor or a Rho kinase (ROCK) inhibitor. None of the drug treatments significantly affected normal siControl mPacL20 cell migration (Fig. 8M) regardless of visual perturbations to the actin cytoskeleton (see Fig. S5 in the supplementary material). Interestingly, the only pharmacological modulator to inhibit the increased migration resulting from a loss in Tm4sf4 expression was the cell-permeable Rho inhibitor, C3 Transferase (Fig. 8M). Importantly, inhibition of the downstream effector of Rho, ROCK, did not suppress the increased migration caused by loss of Tm4sf4 (Fig. 8M). Rho can also modulate the mammalian Diaphenous-related (mDia) formin pathway, which has been shown to localize in the lamella of migrating cells and is important for the stabilization of cortical actin, focal adhesion turnover and, thus, migration velocity (Gupton et al., 2007). Our data suggests that Tm4sf4 inhibits Rho activity independent of ROCK and could affect migration through the mDia/formin pathway.

Nkx2.2^–/– mice completely lack β cells and lack the majority of their α cells, which are replaced by an increased ghrelin⁺ cell population. The mechanism by which Nkx2.2 provides proper cell specification remains unknown. In previous studies, we identified Tm4sf4 as a novel direct target of Nkx2.2 that is markedly increased in Nkx2.2^–/– pancreata (Anderson et al., 2009b; Hill et al., 2011). In this study, we have confirmed that Tm4sf4 is regulated by Nkx2.2 as early as e10.5 during pancreas organogenesis and that Ngn3⁺ endocrine progenitor cells are enriched with Tm4sf4. In the zebrafish model, tm4sf4 inhibits α and β cell specification, but appears to be necessary for ε cell fates. Therefore, the tm4sf4 morphant phenotype and Nkx2.2^–/– or nkx2.2a morphant phenotypes are reversed, suggesting that Tm4sf4 might be a crucial factor downstream of Nkx2.2 controlling cell fate decisions. In support of this idea, co-injection of Nkx2.2atgMO and Tm4splMO into 1- to 4-cell stage zebrafish embryos causes a phenotype intermediate to wild-type and tm4sf4 morphant embryos and suppresses the nkx2.2a morphant phenotype when analyzing insulin and ghrelin levels (Fig. 9A,B). Our data provides evidence to support the idea that Tm4sf4 lies downstream of Nkx2.2 and that its aberrant upregulation contributes to the Nkx2.2^–/– phenotype.

Not surprisingly, the increase in β cells and decrease of ε cells resulting from loss of Tm4sf4 function were restricted to the pancreas. However, we observed that, in addition to increased α cells in the pancreas, transient glucagon⁺ cells were present more posteriorly along the ventral midline in response to the reduction in tm4sf4 in 20.5 hpf embryos. The ectopic glucagon⁺ cells are located in a thin layer of endoderm that will later become the primitive gut (Ng et al., 2005; Wallace and Pack, 2003). Interestingly, the ectopic glucagon⁺ cells are localized in a pattern reminiscent of Tg[nkx2.2a:mEGFP]⁺ cells that appear later during development of the posterior gut at 52 hpf (Ng et al., 2005). Nkx2.2 is expressed in the mouse intestine between e14.5 and e15.5 at the onset of epithelial differentiation. Nkx2.2^–/– mice harbor defects in intestinal enteroendocrine cell fate decisions in addition to the more prominent pancreatic phenotype (Desai et al., 2008). A complete temporal analysis of enteroendocrine marker expression during zebrafish development is lacking, but PP⁺ enteroendocrine cells appear to be restricted to the anterior intestine (bulb) and do not arise until 96 hpf (Wallace et al., 2005), whereas glucagon⁺ somatostatin⁺ double-positive enteroendocrine cells are restricted to the posterior intestine as early as 74 hpf (Ng et al., 2005). In support of previous observations, ectopic glucagon⁺ cells are absent from the intestinal bulb domain adjacent to the pancreas; however, the appearance of glucagon in the primitive mid and posterior gut domains at 20.5 hpf in tm4sf4 morphants occurs before intestinal tube morphogenesis is complete at 34 hpf. It has been reported that the earliest intestinal epithelium differentiation in zebrafish does not occur until 74 hpf (Wallace et al., 2005). We have only observed tm4sf4 expression in liver, pancreas and intestinal bulb domains, so unless posterior intestinal tm4sf4 is below detection limits, increased pancreatic α cells or aberrant intestinal bulb glucagon⁺ cells are most likely to have differentiated outside the posterior intestine and migrated to the domain before undergoing cell death. Future experiments in both mouse and zebrafish models will allow us to investigate how Tm4sf4 might function in the pancreas and primitive gut.

In this study, we have shown that Tm4sf4 inhibits migration of both pancreatic ductal cells and non-pancreatic fibroblasts. Several L6 family members, Tm4sf1 and Tm4sf5, are implicated in cell adhesion and migration during tumor progression and metastasis (Lee et al., 2008; Lee et al., 2006; Lekishvili et al., 2008; Muschel and Gal, 2008). Tm4sf4 is involved in intestinal progenitor cell adhesion and migration during differentiation of intestinal crypt cells (Diosdado et al., 2004; Wice and Gordon, 1995). Recently in Xenopus, Tm4sf3 was described to regulate dorsal and ventral pancreatic bud fusion and ventral bud cell migration, and independently appeared to inhibit ventral β cell differentiation (Jarikji et al., 2009). We believe Tm4sf4 plays a role during endocrine pancreas specification, promoting progenitor cell adhesion to the ductal epithelium and inhibiting migration of the differentiating Ngn3⁺ population. As further evidence of an adhesion/migration role for Tm4sf4, we have observed that in Nkx2.2^–/– mice (high Tm4sf4) the increased ghrelin⁺ cell population form tight clusters that are strictly associated with the ductal epithelium (see Fig. S6 in the supplementary material). Although some wild-type ghrelin⁺ ε cells are near pancreatic ducts at e16.5, ε cells are normally found distributed further from the local ductal environment within the migrating islets (see Fig. S6 in the supplementary material). It is possible that the loss of Nkx2.2 and subsequent upregulation of Tm4sf4 impairs delamination and migration of the differentiating endocrine progenitors away from the ductal epithelium. This would suggest that α and β cell differentiation programs might be inhibited by prolonged exposure to the ductal environment, whereas this environment favors ε cell differentiation.

In support of this hypothesis, when nkx2.2a expression is knocked down in zebrafish, we not only see an increase in tm4sf4 and decreased β cells (Pauls et al., 2007) (Fig. 3G,H; see Fig. S7 in the supplementary material), but also multiple small clusters of insulin-expressing cells that do not aggregate to form a primary islet (see Fig. S7 in the supplementary material). These data imply that nkx2.2a morphant β cells have defects in cell migration, in addition to β cell fate initiation, which might correlate with the increase in tm4sf4 expression. A previous study in zebrafish showed that Wnt5a/Fz-2 (Fzd2 – Zebrafish Information Network) signaling was critical for β cell migration and islet formation (Kim et al., 2005). Recently, the tetraspanin Tspan12 has been shown to interact with Fzd4 and increase β-catenin signaling, crucial for proper retinal vasculature development (Junge et al., 2009). It is possible that Tm4sf4 might have a role in coordinating Wnt signaling and migration during endocrine pancreas specification.

Downstream of the non-canonical Wnt pathway lie the Rho-GTPase family of proteins (Rho, Rac, Cdc42) that modulate actin cytoskeleton, cell polarity, and migration. The cell migration regulator, Rac1, was recently shown to regulate E-cadherin associated cell adhesion and subsequent migration of pancreatic β cells, presumably through actin remodeling (Kesavan et al., 2009). Our results have shown that Tm4sf4, which is highly expressed in the endocrine progenitors, appears to inhibit migration in a Rho GTPase (RhoA, RhoB, RhoC) dependent manner, allowing sustained contact with the ductal environment and affecting cell fate decisions. RhoA activity is mediated through downstream effectors ROCK and mDia. Multiple mDia genes have been implicated in driving migration during development and disease by promoting actin-rich cell protrusions and membrane blebbing (DeWard et al., 2010). Our data suggests that Tm4sf4 might inhibit cell migration via inhibition of an mDia-dependent Rho signaling pathway and not through ROCK. Interestingly, another member of the Rho-GTPase family, Cdc42, has been shown to control pancreatic tubulogenesis of the ducts and to maintain cell polarity. Pancreas-specific knockout of Cdc42 led to a disrupted ductal architecture/microenvironment that results in fewer differentiated α and β cells (Kesavan et al., 2009). We believe that Tm4sf4 is involved in coordinating extracellular cues, cell migration and differentiation specifically within the endocrine pancreas domain.

In vivo, loss of Tm4sf4 function results in an increase in both α and β cell specification and inhibition of ε cell fates. In vitro studies demonstrate that Tm4sf4 promotes increased filipodia and inhibits cell migration mediated through ROCK-independent Rho-GTPase activity. Based on these complementary experimental approaches, we have developed a model whereby endocrine progenitor cells initially express high levels of Tm4sf4, which is downregulated by elevated expression of Nkx2.2. Once Tm4sf4 function is diminished, the endocrine progenitors detach from the surrounding epithelial ducts, actively rearrange the actin cytoskeleton and begin to delaminate, migrate and differentiate (Fig. 9C). We believe that Tm4sf4 functions to inhibit β and α cell specification within the Ngn3⁺ progenitors prior to islet precursor stages. It remains possible that Tm4sf4 could play a later role as a component of signaling pathways to drive cell fate switching independent of cell migration. We will need to utilize mouse model systems to take advantage of conditional knockout and overexpression technology to answer these types of questions confidently. Tm4sf4 promises to become a potential target for small molecule or function-blocking therapies that might provide an avenue to promote and activate quiescent pancreas progenitors in an in vivo or ex vivo environment.

We thank Lee Niswander and David Clouthier (UCAMC) for the use of their microscopes and tissue culture equipment; Christopher Wright (Vanderbilt University) for ISH probes; Ryan Anderson and Didier Stainier (UCSF) for transgenic zebrafish; Morgan Singleton for zebrafish line maintenance; Christina Chao for ISH troubleshooting; and Randall Wong/UCAMC BDC DERC Molecular Core for the use of the ABI 7000. We thank Shouhong Xuan for providing the Flag-tagged Tm4sf4 construct. We thank Teresa Mastracci, Kristi Lamonica and Davalyn Powell for critical reading of the manuscript. This project was supported by the NIDDK Beta Cell Biology Consortium (BCBC) grant U01 DK0272504 (L.S. and K.B.A.) and a BCBC Collaborative Bridging Project grant U19 DK0724473 (L.S., K.R.A. and K.B.A.). Deposited in PMC for release after 12 months.