Why are enteric ganglia so small? Role of differential adhesion of enteric neurons and enteric neural crest cells.

The avian enteric nervous system (ENS) consists of a vast number of unusually small ganglia compared to other peripheral ganglia. Each ENS ganglion at mid-gestation has a core of neurons and a shell of mesenchymal precursor/glia-like enteric neural crest (ENC) cells. To study ENS cell ganglionation we isolated midgut ENS cells by HNK-1 fluorescence-activated cell sorting (FACS) from E5 and E8 quail embryos, and from E9 chick embryos. We performed cell-cell aggregation assays which revealed a developmentally regulated functional increase in ENS cell adhesive function, requiring both Ca 2+ -dependent and independent adhesion. This was consistent with N-cadherin and NCAM labelling. Neurons sorted to the core of aggregates, surrounded by outer ENC cells, showing that neurons had higher adhesion than ENC cells. The outer surface of aggregates became relatively non-adhesive, correlating with low levels of NCAM and N-cadherin on this surface of the outer non-neuronal ENC cells. Aggregation assays showed that ENS cells FACS selected for NCAM-high and enriched for enteric neurons formed larger and more coherent aggregates than unsorted ENS cells. In contrast, ENS cells of the NCAM-low FACS fraction formed small, disorganised aggregates. This suggests a novel mechanism for control of ENS ganglion morphogenesis where i) differential adhesion of ENS neurons and ENC cells controls the core/shell ganglionic structure and ii) the ratio of neurons to ENC cells dictates the equilibrium ganglion size by generation of an outer non-adhesive surface.


Introduction
The enteric nervous system (ENS) is derived from the neural crest (NC), a population of migratory mesenchyme cells originating in the dorsal neural tube. Most of the ENS arises from the caudal hindbrain (vagal) NC (Yntema & Hammond, 1954), chiefly from the level of somites s3 to s5 (Epstein et al., 1994). These cells migrate to the nearby foregut, changing en route (Simkin et al., 2013) to become enteric NC (ENC) cells which are capable of exploiting the gut mesoderm. ENC cells migrate in the mesoderm along the midgut and hindgut to colonise the entire gastro-intestinal tract. This migration takes the form of intersecting narrow chains of motile ENC cells (Druckenbrod & Epstein, 2005;Epstein et al., 1991;Young et al., 2004;Young et al., 2014). Later the ENS comprises a network of numerous small, closely-spaced ganglia with many types of neurons and glia, with each ganglion connected via neurites and glial cells to other ganglia and to the smooth muscle and the mucosa (Conner et al., 2003;Epstein et al., 1991;Fairman et al., 1995). This distributed ENS network controls peristalsis as well as other gut activities (Furness, 2012).
Developmental disorders of the structure, size and organization of the ENS ganglia have consequences for ENS function. Hypoganglionosis (fewer, smaller ganglia) is associated with persistent constipation, and mice heterozygous for the neurotrophic factor GDNF are hypoganglionic (Flynn et al., 2007). Defects involving an overabundance of ENS cells and disturbance of their distribution are also known; hyperganglionosis in the form of enteric ganglioneuromas occurs in multiple endocrine neoplasia type 2 (MEN2B) syndrome as a result of constitutive activation of RET (the receptor for GDNF on ENS cells) and is accompanied by dysfunction of the ENS (Takahashi et al., 1999). Mouse Zic2 mutants show an increased number of ENS neurons (Zhang & Niswander, 2013). Other ENS disorganizations have also been described. In mice where HAND2 is knocked out in ENS cells, the migration of ENC cells is not impaired but segregation into ganglia is abnormal (D'Autreaux et al., 2007;Lei & Howard, 2011). Mice with NC-specific knockout of β1-integrin show ENS ganglia which are morphologically different from normal (Breau et al., 2006).
How ENC cell chain migration evolves into a ganglionated network, and how this is disturbed in some pathologies, is not well understood. NC derivatives elsewhere form relatively large ganglia, such as the dorsal root ganglia (DRG) and sympathetic ganglia. In the forming DRG and sympathetic ganglia, early differentiating neurons occupy the centre of cell aggregates with NC cells or glioblasts surrounding this core. This segregation is maintained in part by Notch signalling which suppresses neuronal differentiation in the peripheral cells (Tsarovina et al., 2008;Wakamatsu et al., 2000). DRG and sympathetic ganglia are in stereotyped positions which are clearly related to and dependent on segmentally spaced cues from their mesodermal microenvironment (Teillet et al., 1987). Formation of each the NC-derived sympathetic ganglia, for example, relies on the scattered NC cells self-aggregating, driven innately by increased N-cadherin homophilic adhesion. This is combined with growth factor and cytokine attraction and repulsion from a patterned microenvironment to provide the positioning of each of the ganglia (Kasemeier-Kulesa et al., 2006;Kasemeier-Kulesa et al., 2010). The size of the ganglia is also regulated.
In the DRG, experimental NC overload and NC ablation suggest that the initial size of the ganglia reflects both the number of NC cells that give rise to them, together with early proliferation. The final size of the ganglia is adjusted in normal and NC cell overload conditions by population reduction by programmed cell death, due at least in part to logistic competition for survival factors. Conversely in NC cell under-supply conditions, compensatory mitosis occurs to increase the DRG cell population (Barde, 1994;Kalcheim et al., 1987;Zarzosa et al., 2014).
The final ENS cell population is enormous but ENS ganglia are each very small and are located mainly in two narrow layers associated with the intestinal smooth muscle layers, the myenteric ganglia between the longitudinal and circular muscle layers and the submucosal ganglia internal to the circular muscle. Typically migratory ENC cells occupy the myenteric layer early, prior to visible structural or molecular correlations including smooth muscle differentiation (Newgreen & Hartley, 1995). A spatial association of the early ENS with the pre-existing intestinal vascular layer has been suggested but this is controversial (Delalande et al., 2014;Hackett-Jones et al., 2011;Hatch & Mukouyama, 2015;Nagy et al., 2009;Schrenk et al., 2015;Young et al., 2004). The submucosal ENS cells originate later from this outer layer by centripetal migration (except in the avian hindgut), a process involving the response of DCC-expressing ENC cells to netrin produced by the endoderm (Jiang et al., 2003).
Cues governing the size, shape and location of each ENS ganglion are not well understood, but are clearly reliant on molecules from their local mesoderm microenvironment. With different-sized starting populations of ENC progenitors, resultant ENS ganglia achieve a similar, small size (compared to DRG) and a similar density of distribution (Allan & Newgreen, 1980;Zhang et al., 2010), suggesting some regulatory ability. Cell death is of great importance in most of the nervous system but is slight in the ENS (Chalazonitis et al., 2012). However, it does occur normally in the early ENSfated population, at and before gut colonisation, well before the ganglia form. Inhibition of this early cell death leads later to an increase in the ENS cell population in those regions of the gut that are first colonised, and this increase takes the form of more densely distributed ganglia, but significantly the ganglia are of normal size and shape (Wallace et al., 2009).
The early ENS is strongly active mitotically (Simpson et al., 2007) in response to growth factors from the local environment. In particular the gut mesoderm cells produce GDNF (glial cell line-derived neurotrophic factor). GDNF is a survival factor and also initially a mitogen (Hearn et al., 1998) and chemotactic factor (Young et al., 2001) via its receptor RET on ENC cells. Later however, its role changes to an inducer of neuronal differentiation (which reduces proliferation) and axon growth. Down-regulation of GDNF during migration not only reduces ENC proliferation but also triggers premature neuronal differentiation (Mwizerwa et al., 2011). Endothelin-3 (ET-3) is also important at early stages, via its receptor EDNR-B on ENC cells. ET-3 is thought to dampen the differentiation response to RET activation, thereby prolonging the proliferative phase of ENC cells (Hearn et al., 1998;Wu et al., 1999) and aiding distal intestinal colonisation (Nagy & Goldstein, 2006). The mesodermal factor BMP2/4 is also important in ENS morphogenesis and differentiation. However, the ENS cell proliferation and neuronal differentiation response to BMP2/4 signalling by ENS lineage cells is complex, and may be influenced by factor concentration and by the time of exposure to alter proliferation and differentiation (Chalazonitis et al., 2004). Early BMP inhibition impairs aggregation of ENC cells into ganglia and leads to hypoganglionosis (Goldstein et al., 2005). BMP also has an effect on enteric gliogenesis, probably by priming ENC cells to respond to factors including Glial Growth Factor-2 (Chalazonitis et al., 2011). Furthermore, BMP increases the number of smooth muscle cells which are a major source of GDNF. In addition, BMP promotes polysialylation of NCAM on ENS cells which likely down-modulates cell-cell adhesivity (Faure et al., 2007); this may favour morphogenetic movement of ENS cells to allow structure changes such as establishment of the sub-mucosal plexuses.
The morphogenesis of the ENS is affected not only by mesodermderived soluble factors but also by structural elements such as extracellular matrix (ECM) (Newgreen & Hartley, 1995). The shape, size and pattern of ganglia can be altered by manipulation of ECM adhesion properties. NC-restricted loss of β1-integrin receptor for ECM adhesion leads to larger, rounder, sparser and abnormally patterned ENS ganglia (Breau et al., 2006). It is assumed that a major β1-integrin ligand is the ECM adhesive molecule fibronectin. Interestingly, the morphogenetic disturbance of ENS gangliogenesis caused by genetic ablation of β1-integrin can be partially corrected by simultaneous deletion of the cell-cell adhesion molecule N-cadherin (Broders-Bondon et al., 2012). ENS neurons, glia and ENC cells show differential labelling for various cell-cell adhesion molecules, and also differ from their surrounding mesodermal cells (Hackett-Jones et al., 2011;Nagy et al., 2012). Here we describe cell-cell adhesion molecules in the ENS, and the roles of cell-cell adhesion in aggregation tests in vitro. In particular we wished to explore why the ENS ganglia are similar and small in size and why the initial cell disposition in ganglia has the pattern of central neurons surrounded by ENC progenitor cells.

Materials and methods
Embryo and tissue origin Fertilised quail (Coturnix japonica) and White Leghorn/Black Australorp cross chicken (Gallus gallus domesticus) eggs were obtained respectively from Lago Game Supplies and Research Poultry Farm, Vic., Australia. Eggs were incubated at 38°C in a 60% humidity incubator. Embryos were staged according to the number of embryonic days (E) and Hamburger and Hamilton stages (HH) (Hamburger & Hamilton, 1951). Animal ethics permission was obtained for the Royal Children's Hospital Animal Ethics Committee, AEC677.

Wholemount midgut preparation
The midgut (defined as the intestine caudal to the stomach to rostral to about half way along the caecum) was removed from quail (Q) embryos at half-day intervals from QE4.5 (HH25) to QE8 (HH34) and then at intervals to QE14 (HH43). These were fixed from times varying from 1 h to overnight in 4% PFA. Antigen retrieval of fixed specimens employed 10 mM citrate buffer pH 6 for 20 minutes at 95°C. Specimens were washed in phosphate-buffered saline (PBS) for 10 minutes, then blocked and permeabilised overnight with 1% horse serum (CSL, Melb., Aust.) and Triton X-100 (Sigma-Aldrich, USA) at 0.1% in PBS. These were then incubated for 1-2 days at 4°C sequentially in primary and secondary antibodies (see Supplemental Table T1) prepared in blocking/permeabilising solution. Between treatments, the specimens were washed extensively in PBS. Gut tissue and aggregate specimens were mounted in Vectashield antifade reagent (Vector Laboratories, Inc., CA, USA) between two coverslips with coverslip spacers.

Counts of myenteric ENS cells in wholemounts
Four to eight areas of 100×100 μm were selected along the midguts from QE4.5 to QE8.5 and cell counts of Hu +ve and SoxE +ve cells were made from optical sections of the myenteric plexus. Cells outside this layer, such as submucosal ENS cells, were not counted.

Cell dissociation
The midgut was removed from QE5 (HH27) and from QE8 embryos and 9-day chick embryos (ChE9), about HH34-35 (Supplemental Figure S1). The intestinal tissue pooled from 15-60 embryos was digested for 35 minutes at 37 o C in Ham's F12 media (Gibco Cell Culture, Invitrogen, USA) with 0.5% w/v Dispase II (Roche, USA) and 0.05% w/v CLSAFA Collaganase (Worthington, USA). To disrupt cadherin-based cell interactions ethylenediaminetetraacetic acid (EDTA; Sigma-Aldrich, USA)) was added to a concentration of 1 mM for a further 10 minutes. The tissue was mechanically triturated and the cell suspension was washed in F12 media with 5% BSA. Cells from mid-trunk dorsal root ganglia (DRG) from the QE8 embryos were dissociated in the same way.

Cell identification and FACS validation
After selection by FACS, HNK-1 +ve and HNK-1 -ve cells (as well as unselected cells) were examined by q-PCR for the NC marker Sox-10 and the neuronal marker Hu-D, and by cell culture.
Total RNA was isolated from cells using the RNeasy mini kit (Qiagen, USA) and contaminant genomic DNA removed with DNAfree reagents (Ambion, USA). Primer sequences were designed using Primer3 (http://frodo.wi.mit.edu/primer3/) and are listed below. For quantitative reverse transcription and polymerase chain reaction qRT-PCR, oligo-dT primed cDNA was synthesised from 200 ng total RNA using Murine Moloney Leukaemia Virus reverse transcriptase (Promega, USA). qRT-PCR was performed on an ABI Prism ® 7500 Real Time PCR System using SYBR green master mix (Applied Biosystems, USA) according to the manufacturer's protocols. Relative gene expression values were obtained by normalization to the reference gene GAPDH using the −2 ΔΔCt method, where −2ΔΔCt = ΔCt sample−ΔCt calibrator as described (Peterson et al., 2010). All fold changes were calibrated to the negative sort population. Results are shown in Supplemental Figure S3.
Both q-PCR and culturing of FACS sorted cells indicated that the HNK-1-based sorting accurately selected for virtually all cells in the gut that expressed NC and neuronal markers.

Short term cell aggregation assays
After cell dissociation and FACS analysis, the HNK-1 +ve cells remained for 1 h at 37°C in cell culture medium of F12 with 2% heat inactivated FCS, 0.5% BSA and pen/strep, to recover cell-cell adhesive potential (Steinberg et al., 1973;Takeichi, 1977). A low degree of aggregation occurred in the recovery period, indicated by a 6-15% reduction in particle number from the count recorded at the time of FACS. In three dissociation runs Calcein AM (1/4000; Invitrogen/Molecular Probes) was added for 20 min to reveal live cells, with the cells then centrifuged into fresh medium. This indicated that about 90% of cells were alive at this stage. The cell suspension was then aliquoted into Eppendorf tubes (100 μl cell suspension/tube) with particle density adjusted to 0.3-0.5×10 6 cellular particles/ml. A minimum of three replicate tubes were prepared for each assay, and each assay was repeated at least twice. In parallel tubes ethylene glycol tetraacetic acid (EGTA) was added to 1 mM; this chelates Ca 2+ and therefore prevents cadherin-dependent cellcell adhesions. The tubes were then incubated on a rotating platform (120 rpm, radius 1 cm) at 37°C. At t=0, 15, 30, 60 and 120 min., 10 μl samples were withdrawn from each tube and the particles were counted (see below). A particle was defined as a single cell or group of contacting cells of any size. Cell aggregation was indicated by a decreasing particle number.
Longer term cell aggregation assays Cells were allowed to aggregate as above, but with aggregation time of 2h, 4h, 6h, 18h and 48h. Rotation rates of 150, 120, 100, 75 and 0 rpm were tested. Starting cell densities were varied from 0.167×10 6 to 1.0×10 6 particles/ml. Aggregation assays were also performed with the inclusion of BrdU (Amersham-GE Healthcare, USA) for 4 h prior to fixation. At the end of the incubation period, aggregates were collected for imaging and image measurement by allowing them to settle in the Eppendorf tube for 5 min then removing the bottom 20 μl of medium plus aggregates. This was placed as a standing drop on a non-TC Petri dish which was oscillated at 80 rpm for 5 minutes to centralise the aggregates, which were then imaged. For fixation, 200 μl of 4% PFA in PBS was added to each Eppendorf tube. After overnight fixation, the fixative was washed out with PBS prior to immunolabelling as for midgut wholemounts.
Imaging, cell and aggregate counting and evaluation Samples were screened using an Olympus IX70 microscope (Olympus Optical Co., Tokyo, Japan), under selective Texas Red, FITC and AMCA filters, and by phase contrast. Images were recorded using a Spot Monochrome camera model 2.1.1 with Image-Pro Plus 4.5 (MediaCybernetics, Silver Spring, MD, USA). Confocal imaging was prepared on a Leica TCS SP2 with image processing via Leicalite and Image-Pro-Analyser 6.1 (MediaCybernetics). For short-term aggregation assays, particles (cells and cell groups) at each time point were counted in a haemocytometer chamber with 10-20 microscope field images each of 1.14 mm 2 recorded using an Olympus IX70 microscope (Olympus Optical Co., Tokyo, Japan) with 10× objective. Particle counts were made from these images by operators blinded to the assay conditions. For the long term aggregation assays, aggregate diameters were measured from phase contrast images (×20 objective) of at least 50 aggregates per treatment and time. Aggregates were chosen for measurement on the basis of roundness and defined edges, and very small and or loose cell clusters and single cells were ignored.

Statistical analysis
Unless specified, data were expressed as mean± standard error of mean (SEM). All statistical tests were performed using GraphPad Prism version 6. A difference between two groups was determined using a two-tailed Student's t-test and for nonparametric data Mann-Whitney test was used. For differences among multiple groups, statistical comparisons were performed using one-way analyses of variance (one-way ANOVA) followed with Fisher's LSD post-test. A p-value of <0.05 was considered significant.

Results and discussion
The developing ENS in vivo has a constant ENC cell/ neuron ratio The sparse ENS cell population in the nascent myenteric plexus of the midgut was dominated by SoxE +ve ENC cells. The total ENS cell density (cells/unit plexus area) continued to increase over the period QE4.5 to QE8 ( Figure 1) and the plexus area increased by exponential gut growth (Binder et al., 2008). From about QE6, correlating with the assembly of ENS cells into coherent groups, the proportion of Hu +ve neurons increased to reach a ratio of 1.2:1 Hu +ve : Sox E+ve cells. This ratio was maintained until at least QE8 ( Figure 1). The constancy of the ratio of Sox +ve cells to Hu +ve cells while the population number and density increased suggests an effective co-ordinate control between neurons and ENC cells.
Inhibition of Notch activity in mice by NC cell-specific knockout of the Pofut1 gene, and Notch inhibition by DAPT in mouse and human enteric neurospheres in vitro (Okamura & Saga, 2008;Theocharatos et al., 2013) led to loss of Sox10 +ve ENC cells and a bias towards enteric neuron differentiation. This strongly suggests that after cell aggregation has been achieved by morphogenetic cell re-arrangements, the Notch system forms part of the intercellular signaling agency maintaining the ENC cell/neuron balance in the developing ENS. MG1 neural cell count (Rollo et al., 2015a).

ENS ganglia undergo progressive morphogenesis in vivo
In the quail embryonic midgut at HH27 (QE4.5 to 5), shortly after arrival of vagal ENC cells at HH25/6 (QE4.25), the relatively sparse ENS cell population was mainly SoxE +ve ENC cells (Hu-ve) distributed in chains (Figure 2A). The number of Hu +ve neurons (SoxE -ve ) increased in the nascent myenteric plexus and they commenced forming clusters by QE6 ( Figure 2B). Later (QE8), the Hu +ve cells and the SoxE +ve cells formed co-aggregates with almost all the SoxE +ve cells segregated to the periphery of each neuronal Hu/SoxE ratio Embryonic age  cluster ( Figure 2C). The ENS cell aggregates increased in size and developed increasingly smooth borders and by QE14 the SoxE +ve cells were found not only surrounding the neuron groups but also between the individual neurons, as well as along axon tracts ( Figure 2D). In other NC-derived ganglia like the DRG a similar sequence of events, but chronologically earlier, has been observed, with the late-appearing intraganglionic cells being differentiated ganglionic glia (Henion et al., 2000).
We sought an explanation in differential cell-cell adhesion (Steinberg, 2007) for the progressive aggregation of ENS cells into ganglia, with internal neurons and a shell of ENC cells. Time-lapse microscopy in mouse intestine has revealed that ENS cells (both neurons and ENC cells) are motile for a considerable period after the initial colonization phase (Hao et al., 2009;Young et al., 2014). It can be imagined that such motile ENS cells might, by differential adhesion, finally collect together to form few very large ganglia; we therefore also sought reasons for the ENS forming only small ganglia.

HNK-1 FACS effectively selects for NC-derived cells
Unsorted dissociated midgut cells showed modest NC (Sox10) and neuronal markers (Hu) by q-PCR. The HNK-1 +ve sorted moiety showed high levels of these neural sequences whereas the HNK-1 -ve moiety had very low expression (Supplemental Figure S3). HNK-1 +ve sorted cells plated on fibronectin or laminin surfaces showed neural immunoreactivity, including SoxE (recognizes Sox9 and Sox10), HNK-1, HuC/D, Tuj1 and E-C8. A few cells (<2.5% at 18 h in vitro) were negative for neural markers but fibroblast-like in appearance and smooth muscle actin (SMA) +ve . We regard these as contaminating gut mesoderm cells. In contrast HNK-1 -ve cells when plated were virtually entirely fibroblast-like in appearance and SMA +ve (Supplemental Figure S4). This cell sorting procedure therefore provides highly enriched ENS cells for performance of cell aggregation assays.
ENS cell clusters in vitro are due to cell aggregation, not proliferation Dissociated HNK-1 +ve midgut ENS cells in low serum aggregation assays rapidly formed clusters which were relatively small and uniform. We examined ENS cell aggregates (N=7) at 22 h with confocal microscopy after 4 h bromodeoxyuridine (BrdU) exposure, and examined 18 h aggregates (N=6) with phosphohistone-H3 antibody and detected no cells labeled by these markers of proliferation. We conclude that there was little or no cell proliferation in vitro, and therefore the cell clusters under these conditions are due to cell aggregation.
ENS cell aggregation indicates several adhesive mechanisms are operative and increase with age Avian ENS cells in previous studies showed immunoreactivity for N-cadherin and NCAM and also for Ng-CAM (L1CAM) at the stages equivalent to the early stage shown above (Hackett-Jones et al., 2011;Nagy et al., 2012). Ng-CAM soon became almost undetectable while N-cadherin and NCAM labelling became more intense on both SoxE +ve and Hu +ve cells. By QE8 in the midgut NCAM was clearly more strongly labelled on the Hu +ve neurons compared to the SoxE +ve cells (Hackett-Jones et al., 2011). This suggests that there may be a general increase in cell-cell adhesion in the ENS and a further increase in adhesion between neurons.
The short-term rotating cell aggregation assays indicated that dissociated QE5 and QE8 midgut ENS cells adhered progressively ( Figure 3A, Figure 4A-D), but aggregation was faster and more complete in cells from the older embryos. This confirms a developmental increase in ENS cell cohesion. QE5 and QE8 ENS cell aggregation occurred over the first 30 minutes at the same rate with Ca 2+ chelation (i.e. 1 mM EGTA) as with normal medium but later the level of cell aggregation was impaired ( Figure 3B, C, Figure 4E). The sensitivity to EGTA showed involvement of Ca 2+ -dependent (i.e. cadherin) mechanisms, but the residual aggregation indicated Ca 2+ -independent mechanisms (such as NCAM) were operative as well, and suggests that the initial phase of adhesion in these conditions was largely due to Ca 2+ -independent adhesion. This is in accord with the immunoreactivity for both N-cadherin and NCAM in situ noted above and in these cells in vitro ( Figure 6A, B). Aggregates at QE5 and QE8, control and with EGTA. Aggregation was indicated by decrease in particle count and is expressed as % of time t=0 min (Rollo et al., 2015b).

ENS cell aggregation is progressive
The earliest stage of aggregation for HNK-1 +ve E8 quail (and E9 chick) midgut ENS cells was as small clumps and strings, at about 2 h in rotating culture. They formed spheres by 4 h, and initially cells bulged from the surface of spheres (resembling a "bunch of grapes") but the spheres became more smooth-surfaced, and maintained this from 18 h to 48 h ( Figure 4A-D). The diameter of the aggregates increased between 4 h and 48 h ( Figure 4D, Figure 5A) and the range of diameters recorded became wider (Supplemental Figure S5) but aggregation into a few huge cellular masses did not occur. ENS cells from quail and chick behaved identically in these assays ( Figure 5B).
The spherical form of the aggregates suggests isotropic adhesion forces while the evolution of the aggregates from a "bunch of grapes" appearance to smooth-surfaced spheres indicates an increase in cell-cell adhesion strength in vitro with time after initial cellcell adhesion. A time-dependent increase in adhesive bond energy has been observed in direct measurement of cadherin-mediated adhesion maturation in biophysical tests (Chu et al., 2004).
SoxE +ve /Hu -ve (ENC cells) and SoxE -ve /Hu +ve cells (neurons) occurred first in tiny clumps, with SoxE +ve /Hu -ve cells predominating in the strings; these ENC cells displayed both N-cadherin and NCAM immunoreactivity ( Figure 6A', A"). The transient presence of cell-strings has also been described in cells with only cadherin adhesive mechanisms operative (Takeichi, 1977). Cadherins move on the plane of the membrane and cluster in cis, via intercadherin bonds extracellularly and via binding to the cytoskeleton intracellularly (Hong et al., 2013). The generation of cell strings by SoxE +ve ENS cells suggests that the ENC cells have a limited number of adhesive molecules on their surface, this cis-clustering may restrict cadherin to a few patches capable of mediating adhesion in trans. .001 for QE8), except for QE5 at 60 min vs 120 min, where the particle counts were not significantly different. In addition, aggregation was greater for QE8 ENS cells compared to QE5 ENS cells (*** p<0.001 QE8 vs QE5 at each time point), consistent with a developmentally increasing adhesive capacity. B. With EGTA, early aggregation (0 min, 15 min and 30 min) of QE8 ENS cells proceeded rapidly and was not significantly different from particle counts in control medium at the same time points. At later time points (60 min and 120 min) particle counts with EGTA were significantly greater than from the same time points in control medium (θθθ p<0.001 EGTA vs Ctrl). This strongly indicates that early aggregation events in these assays are largely calcium-independent but after about 30 min aggregation is dependent on cadherin function. C. QE5 ENS cells showed a similar early rate of decline in particle number with EGTA medium at matched time points of 0 min, 15 min and 30 min. Later, particle number decline decreased in EGTA (φφφ, p<0.001 EGTA vs Ctrl at 60 min; φ p=0.0225, EGTA vs Ctrl at 120 min;). This indicates that for QE5 ENS cells early aggregation is largely calcium-independent but further aggregation requires cadherin function. Error bar=SEM. Aggregate size from 2h to 48h (Rollo et al., 2015c). Aggregate diameters from QE8, ChE9 and mixed populations (Q+Ch) at 18h in vitro (Rollo et al., 2015d).
ENS cell aggregation is insensitive to fluid shear but sensitive to initial cell density Varying the speed of rotation (0 and 75 rpm) had little effect on aggregate form or size at 18 h, while aggregates at 150 rpm were somewhat larger in diameter ( Figure 4F-H, Figure 5C). Since increasing rotation rates did not prevent aggregation, the intercellular cell-cell adhesions of these ENS cells must display a rapid "catch" to initiate adhesion, with initial adhesion strength sufficiently high to resist external distractive forces in the highest shear used here. On the other hand, the similar size and shape of the aggregates even down to zero rpm ( Figure 4F, Figure 5C) suggests an overriding intrinsic adhesive mechanism that regulates not only accumulation of cells into aggregates but also governs the preferred size of the aggregates under these conditions. Aggregate diameter at 18 h in vitro was starting cell density-related. Starting cell density at 1×10 6 cells/ml produced aggregates of least diameter. With the gradually reduced starting cell density, the aggregate diameter increased (relative to 1×10 6 cells/ml: *** p<0.001; relative to 0.67×10 6 cells/ml: # p=0.0186, ### p<0.001). However, this increase plateaued, with no significant difference in aggregate diameter at 0.167, 0.33 and 0.5×10 6 cells/ml. Error bar=SEM. Varying the initial density of ENS cells in suspension over a 6-fold range (0.167×10 6 to 1.0×10 6 cells/ml) led, counter-intuitively, to smaller aggregates in much larger numbers at higher starting cell densities ( Figure 4I-L, Figure 5D).

The spatial patterning within aggregates indicates ENS neuron cohesion is greater than ENC cell cohesion
Confocal examination of 4-6 h spherical aggregates showed mixed Sox10 +ve and Hu +ve cells ( Figure 6B), but by 18 h cells in the aggregates showed most of the Hu +ve cells located in the centre, with the SoxE +ve cells forming the outer part of the spheres ( Figure 6C). Labelling for NCAM showed that this adhesion molecule was most strongly expressed on the internal Hu +ve cells with lower immunoreactivity on the SoxE +ve cells, and especially low on the outer surface ( Figure 6D). Such differential labelling was also present with labelling for N-cadherin ( Figure 6E).
Confocal examination of four aggregates (diameter range: 63.8-98.7 μm; cell number range 276-861) revealed a remarkably uniform average cell density of 0.24 ± 0.02 cells/(10 μm) 3 . Likewise the neuron/ENC cell ratio of 1.19 ± 0.06 (Hu +ve /SoxE +ve cells) was identical to that in the ENS in vivo (Figure 1). The spatial order of central neurons and peripheral ENC cells in the aggregates in vitro strikingly resembled that in the ENS ganglia in vivo ( Figure 2C) (Hackett-Jones et al., 2011).
In assays of this kind, cells move within aggregates ("sort out") with their final equilibrium positions determined by the most favoured adhesive balance, that is, with the least surface free energy of adhesion (Foty & Steinberg, 2005). To satisfy this, the external position of the ENC cells relative to neurons indicates that the ENC cells must have lower overall adhesive capacity than the neurons, and the lower levels of NCAM immunoreactivity in SoxE +ve

Figure 6. Formation of aggregates by ENS cells in cell-cell adhesion assays correlates with adhesion molecule expression pattern. A' A''. At 2 h QE8 N-cadherin and NCAM +ve ENS cells formed small aggregates (mixed Hu +ve neurons and SoxE +ve ENC cells) and chains (mostly ENC cells). B.
At 6 h spherical aggregates were formed but neurons (HuC/D +ve ) and ENC cells (SoxE +ve ) were only partially segregated. C. At 18 h most HuC/D +ve neurons formed the centre of each spherical aggregate surrounded by SoxE +ve ENC cells. D. NCAM immunoreactivity was low on the external face of SoxE +ve ENC cells (indicated by dotted line), but high around HuC/D +ve neurons, whether these were located centrally or on the periphery (arrows). E. N-cadherin immunoreactivity was associated with all cells in the aggregate, but was less distinct on the external surface (indicated by dotted line).
cells is in accord with this. Cell variants with a step difference in adhesiveness segregate especially rapidly in co-aggregates (Zhang et al., 2011), and this is likely to be represented by the two ENS cell types -neurons and ENC cells-used here.

ENS cells are mobile within aggregates, allowing sorting out
In vivo the neurons form recognisable groups first ( Figure 2B), at a time when the ENC cells still appear randomly placed, and the ENC cells co-assemble around neurons later (Fairman et al., 1995;Hackett-Jones et al., 2011). In contrast, the final internal/external distribution observed here would be attained from any starting distribution of the cell types (Steinberg, 2007). Indeed, when we followed aggregation in vitro, the neurons and ENC cells were initially mixed ( Figure 6B) and only later did the relative positions of neurons and ENC cells develop ( Figure 6C).
To test whether cells were able to sort out within aggregates we combined pre-formed (E9 chick) ENS cell aggregates with freshly dissociated QE8 ENS cells. A few quail cells adhered to the surface of pre-formed ENS cell aggregates when combined for 4 h in rotating culture ( Figure 7A). By 18 h some of these quail ENS cells had penetrated deep into the chick ENS cell aggregates ( Figure 7B). This confirms that cells can move within the aggregates. This is consistent with the SoxE +ve /Hu -ve ENC cells and SoxE -ve /Hu +ve neurons physically sorting out, although it does not preclude an additional spatial differentiation whereby internally placed ENC cells differentiate mostly into neurons.
The external surface of ENS cell aggregates is poorly adhesive for ENS cells When freshly dissociated QE8 HNK-1 +ve cells were added to preformed (18 h) chick E9 ENS cell aggregates, QCPN-labelling showed that only a few quail cells adhered to the outer surface of pre-formed chick cell aggregates ( Figure 7A), and most quail ENS cells formed separate aggregates entirely of quail cells ( Figure 7C). This separation was not caused by species-specific adhesive differences, because when freshly dissociated E9 chick and QE8 ENS cells were mixed at the time of dissociation, all cell aggregates were a mixture of both chick and quail cells ( Figures 5B, Figure 7D).
This shows that the outer surface of pre-formed aggregates is relatively non-adhesive, and is consistent with the observation that the outer surface of the peripheral cells of aggregates showed low immunoreactivity for CAMs ( Figures 6D, E). We therefore propose that the outer SoxE +ve ENC cells segregate a limited number  of CAMs mainly to the internal face to bind to similar but more numerous molecules on the neurons, leaving the external surface relatively non-adhesive. This process would automatically restrict the size of ENS cell aggregates by preventing new cells from binding to the surface.
Altering the ENC cell/neuron ratio alters the aggregate size If the peripherally located Sox10 +ve ENC cells in the aggregates form an insulating coating preventing ever-larger aggregates from forming, then reducing the number of these cells should allow larger aggregates to form. By combining NCAM with HNK-1 FACS, we produced HNK-1 +ve populations of higher and lower NCAM levels ( Figure 6D, Supplemental Figure S2C). Previous immunolabelling in vivo (Hackett-Jones et al., 2011) and here in aggregates in vitro shows that the NCAM-high sub-population will be enriched for neurons, and the NCAM-low sub-population depleted in neurons. Culturing these cells in Terasaki wells confirmed this was the case, both populations had SoxE +ve cells but HuC/D +ve cells with E-C8 +ve , Tuj1 +ve neurites occurred only in the NCAM high fraction. Aggregation of NCAM-high and NCAM-low sub-populations produced respectively larger and smaller irregularly shaped aggregates than was usual for unsorted HNK-1 +ve cells ( Figure 8A-C).

Different NC cell ganglia show aggregation differences
Quail E8 trunk DRG were dissociated as for the ENS cells and placed in aggregation assays. Like the ENS cells, DRG cells rapidly formed aggregates with similar average size although the range of size of DRG aggregates was larger ( Figure 9A). Interestingly, the internal structure of the aggregates was strikingly different: SoxE + /Hu -ve cells and SoxE -/Hu +ve neurons were mixed not segregated ( Figure 9B inset). In addition, although all Hu+ve neurons were strongly NCAM +ve and all outer SoxE +ve cell membranes were deficient in NCAM labelling (as in equivalent ENS cells), the internal SoxE +ve cells were strongly NCAM-labelled ( Figure 9B, C).
In vivo, NC-derived cells coalesce to form DRG as early as about E4 (in chick; equivalent to E3.5 in quail), and at these early stages the DRG display NC cell/neuron segregation as seen for ENS cells, with all neurons placed centrally (Wakamatsu et al., 2000). As early as E5 in chick, however, early DRG glial cells identified by transitin expression (which are also SoxE +ve ) spread centrally between the DRG neurons (Henion et al., 2000). We propose that the different sorting behavior of QE8 ENS cells and QE8 DRG cells represents a difference in developmental stage of the two ganglion types of the same chronological age, with DRG being much further advanced in ganglion cell differentiation, marked by the appearance of highly NCAM +ve glial cells. The morphogenetic consequence of this drives the internal relocalisation of glial cells among the neurons they support by E8 in trunk DRG. We propose that a similar process evolves later, by E14, in midgut ENS ( Figure 2D). Box plots show the first quartile to inter quartile range and whiskers show minimum and maximum range of data, while the median is represented by a vertical line. There was no difference between DRG and ENS in aggregate size when analysed using Mann-Whitney test for nonparametric data. Starting cell density was 0.5 ×10 6 cells/ml. B. Unlike QE8 ENS cell aggregates, Hu +ve neurons and SoxE +ve cells are not segregated in DRG aggregates (inset), but like ENS aggregates, peripheral SoxE +ve cells show little outer NCAM labelling (arrows). C. In contrast, internal SoxE +ve cells (stars) as well as surface and internal Hu +ve cells show strong NCAM labelling in DRG cell aggregates. Aggregate size for QE8 DRG and ENS HNK-1 +ve cells (Rollo et al., 2015h). EN cell: neuron ratio and aggregate size (Rollo et al., 2015j).
Conclusions: intrinsic regulation of cell distribution and size of ENS ganglia A mechanism of separation by sorting of more and less adhesive cells would produce the observed core/shell spatial distribution of SoxE +ve and Hu +ve cells ( Figure 10A). Translocation of a numerically limited number of CAMs in the plane of the cell membrane to the internal-facing side of the less adhesive cells (i.e. the Sox10 +ve cells) to engage the more numerous homophilic CAMs on the more adhesive cells (i.e. Hu +ve neurons) ( Figure 10B) would automatically limit the size of the aggregates by denuding the external surface of CAMs. This mechanism would predict that the size of the aggregate depends at least in part on the ratio of more adhesive neurons to less adhesive ENC cells. We suggest that this operates also in vivo and is of importance for ensuring that the ENS consists of relatively small and uniform ganglia. However, alteration of adhesion of ENS cells to ECM leads to ENS ganglia of abnormal size, shape and pattern of distribution in vivo (Breau et al., 2009;Broders-Bondon et al., 2012). This shows that the balance of cell-ECM adhesion as well as cell-cell adhesion, needs to be taken into account to achieve correct ENS ganglionic morphogenesis.  which are free to move in the cell surface, but the neurons have more adhesive molecules. Cell-cell adhesion requires juxtaposition in trans of two adhesive molecules (red-red, blue-blue or red-blue).

Data availability
To minimise adhesive free energy, adhesive molecules on ENC cells move (pink arrows) in order to form an adhesion, thereby denuding the outer face of adhesion molecules. This insulates the group from further adhesion.

Competing interests
No competing interests were disclosed.

Grant information
This work was supported by National Health and Medical Research Council grants 436971 and 607379. MCRI facilities are supported by the Victorian Government's Operational Infrastructure Support Program.
I confirm that the funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Supplemental Figure S1. QE5 and QE8 midgut ENS donor tissue. Intestine (midgut) caudal to the gizzard (giz.) to about half way along the cacum (cec.) was obtained. This was removed from E5 (about HH27) and E8 quail embryos (about HH34-35). HH34-35 midgut was also obtained from 9-day chick embryos. The intestine was cut at the dorsal border of the mesentery to exclude cells of the Nerve of Remak. The tissue used did not include the cecal root because here the Nerve of Remak is applied to the gut so closely that contamination is unavoidable. C-r = colo-rectum, prov. = proventriculus To assess the enrichment for ENC markers following FACS for HNK1, QE8 midgut-derived cells were assessed for expression of SOX10 (a NC marker) and Hu-D (a neuronal marker). Cell populations analysed by qPCR were cells which were not sorted (pre-sort), sorted cells which were not HNK1 +ve (-ve sort) and HNK1 +ve sorted cells (HNK1). HNK1 +ve cells showed a significant increase in expression of the markers tested when compared with pre-sorted cells indicating successful enrichment of ENC cells by FACS. Full qPCR analysis of HNK-1 sorted cells (Rollo et al., 2015i).  Figure S5. HNK-1 +ve ENS cells aggregate progressively. Aggregate diameter increased on average and the range of diameters increased over the period 2 h to 18 h in rotation aggregation assays. Starting cell density was 0.3×10 6 cells/ml.

Increase in aggregate diameter and range
Mat.+Meth.: "Counts of myenteric ENS cells in wholemounts". One does not know what "Hu" stands for and why "SoxE" is mentioned here but Sox-10 further down on the page (last line). Explanation of abbreviations here or before the paper starts?
"Cell dissociation". While a non-expert might imagine how a "midgut" is removed, the skill that is needed to harvest DRG may be beyond his/her imagination. Add perhaps 1-2 sentences for explanation?
"The intestinal tissue pooled from 15-60 embryos": why 15-60? Is this one of many similar preparations using between 15 and 60 guts? Otherwise you should indicate a fixed number.
One gets no information about the proportion of ENS cells obtained from the myenteric and submucosal plexuses. Are the ganglia in both places equal in size and structure? At certain developmental stages there could be many more Hu-positive or SoxE-positive ganglia in the myenteric than in the submucosal plexuses because the latter are settled earlier.
Section: "Fluorescent labelling and fluorescence-activated cell sorting (FACS)": It is hardly imaginable that from such a broth obtained by filtering disaggregated tissue or single cells through a 30 µm strainer cells were obtained that could form such beautiful aggregates as shown in Fig. 6. May be you can add a sentence about the consistency of the filtered cells? This section is ENS cell clusters are due to cell aggregation, not proliferation". in vitro nearly too short to merit section status -could it be subsumed somewhere else? I only wonder whether this ENS cells are mobile within aggregates, allowing sorting out". initial mobility of ENS and ENC cells and their later sorting out and finally the outer placement of the ENC cells with restricting ENS cells is not all a little contradictory?  Fig. 6C with DRG size in Fig. 9. May be you could discuss that at the very end of the paper? I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.

Results
No competing interests were disclosed. This paper represents a vanguard analytical investigation on the role of adhesion molecules in regulating the size of enteric ganglia. This paper demonstrates that both and , Hu enteric neural in vitro in vivo crest cells showing a neuronal phenotype tend to "sort out" within the centre of ganglia and adhere together while Hu enteric neural crest cells show less adhesion and are located more in the periphery. I have a few comments that I would like to be addressed in order to increase the clarity and impact of this manuscript for the reader.
In the Introduction the authors state ": Hypoganglionosis (fewer, smaller ganglia) is associated with persistent constipation…" Could they please provide a citation for this statement?
On page 2 the authors wrote. "Formation of each NC-derived sympathetic ganglia, for the example, relies on the scattered NC cells self-aggregating…" Please delete "the" from this statement.
On page 4 "The tubes were then incubated on a rotating platform (120 rpm, radius 1 cm) at 37 C. At t=0, 15, 30, 60 and 120 min.," Please clarify the position that the tubes were incubated, vertical or horizontal? On page 6. "HNK-1+ve sorted cells plated on fibronectin or laminin surfaces showed neural immunoreactivity, including SoxE (recognizes Sox9 and Sox10). In the description of the antibodies it states that this antibody also recognizes Sox8. Therefore, could Sox8 please be added within the parentheses.
In Fig. 1 . The y axis label reads cells per 100mm2, I think that this might be error and should be cells per 100 microns squared. Could the authors please amend this?
On Page 6. The authors wrote, "The sensitivity to EGTA showed involvement of Ca -dependent (i.e. cadherin) mechanisms, but the residual aggregation indicated Ca -independent mechanisms (such as NCAM) were operative as well, suggests that the initial phase of adhesion in these conditions was largely due to Ca -independent adhesion." This section could be made more clear with some revision. As the words residual aggregation suggest that it is a later time point and so maybe the removal of the word residual would make this sentence easier to read and understand what the authors are trying to say. Figure 6 needs to be modified so that each of the fluorophores are shown separately, especially 6D. It would be useful for the reader to be informed where the optical section was taken from. Are we looking at a section from the surface or deeper within the aggregate as this changes how we can interpret what we are looking at. 10.
between some Hu cells. Some quantification of the intensity of the NCAM staining would greatly improve the validity of the authors statements here. Fig 7B. This image doesn't support the assertion from the authors that the Hu cells are found in the center of the aggregates as many Hu cells are found on outside in this image. Do the authors have an alternative image that could be used to demonstrate their conclusions? Alternatively could they make sure that they state the proportion of aggregates that had Hu cells on their surface.
On page 10 the authors wrote, "This shows that the outer surface of pre-formed aggregates is relatively non-adhesive, and is consistent with the observation that the outer surface of the peripheral cells of aggregates showed low immunoreactivity for CAMs ( Figures 6D, E)." The outer surface of the aggregates must be sticky at initial stages or the aggregates would not have increased in size during these stages since the authors have already shown us that aggregate formation at the early stages is not being regulated by cell proliferation. Could the authors please amend this statement to ensure that it is a true reflection of what is occurring within their experiments?
On page 11. "Culturing these cells in Terasaki wells confirmed this was the case, both populations had SoxE+ve cells but HuC/D+ve cells with E-C8+ve, Tuj1+ve neurites occurred only in the NCAM high fraction." It would be really good if the authors could show this data. If this is not possible then could they please state, data not shown here. Fig 9A. "Box plots show the first quartile to inter quartile range and whiskers show minimum and maximum range of data, while the median is represented by a vertical line." The authors are very imaginative but some readers don't shave. Please change to vertical lines show minimum and maximum range of data respectively, while the median is represented by a horizontal line.
On page 11 the authors state, "Interestingly, the internal structure of the aggregates was strikingly different: SoxE+/Hu-ve cells and SoxE-/Hu+ve neurons were mixed not segregated ( Figure 9B inset)". Please add "in the DRG" at the end of this sentence.
We have read this submission. We believe that we have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.
No competing interests were disclosed. Competing Interests: +ve +ve