Integrins establish dendrite-substrate relationships that promote dendritic self-avoidance and patterning in Drosophila sensory neurons

Summary

Dendrites achieve characteristic spacing patterns during development to ensure appropriate coverage of territories. Mechanisms of dendrite positioning via repulsive dendrite-dendrite interactions are beginning to be elucidated, but the control, and importance, of dendrite positioning relative to their substrate is poorly understood. We found that dendritic branches of Drosophila dendritic arborization sensory neurons can be positioned either at the basal surface of epidermal cells, or enclosed within epidermal invaginations. We show that integrins control dendrite positioning on or within the epidermis in a cell autonomous manner by promoting dendritic retention on the basal surface. Loss of integrin function in neurons resulted in excessive self-crossing and dendrite maintenance defects, the former indicating a novel role for substrate interactions in self-avoidance. In contrast to a contact-mediated mechanism, we find that integrins prevent crossings that are non-contacting between dendrites in different three-dimensional positions, revealing a requirement for combined dendrite-dendrite and dendrite-substrate interactions in self-avoidance.

Introduction

For many types of neurons, dendrites represent the most expansive membrane compartment, with large surface areas in extensive contact with the surfaces of other neurons as well as the substrates upon which they grow. Thus, the molecular interplay between growing dendrites, the extracellular environment, and substrate must be highly regulated. The molecular basis of dendrite-substrate interactions in vivo and the implications for dendrite morphogenesis remain incompletely understood. As dendrites elaborate, one important step in their patterning is the proper spacing of branches from the same cell, or sister dendrites, via repulsive dendrite-dendrite interactions (Grueber and Sagasti, 2010; Jan and Jan, 2010). Self-avoidance, which ensures complete and non-redundant coverage of sensory or synaptic inputs, is most clearly observed in neurons that grow in a planar pattern, such as retinal ganglion cells, leech sensory neurons, and Drosophila dendritic arborization (da) neurons (Grueber and Sagasti, 2010; Jan and Jan, 2010; Kramer and Stent, 1985). Although self-avoidance is probably not limited to two-dimensional arbors (Zhu et al., 2006), the robustness of self-avoidance in such processes implies that molecules and substrates that restrict growth to a plane may influence repulsive interactions. The extent of this influence, and the impact on current molecular models of self-avoidance, is not known.

Drosophila dendritic arborization (da) neurons have proven useful for studies of dendritic morphogenesis and self-avoidance. da neurons can be segregated into four classes (classes I–IV) distinguished both by dendritic morphology and central axon projections (Grueber et al., 2002; Grueber et al., 2007). Numerous molecules have been implicated in control of dendrite-dendrite repulsion. For example, the Down syndrome cell adhesion molecule 1 (Dscam1) family of homophilic adhesion molecules permits selective recognition between the surfaces of sister dendrites and initiation of repulsive responses between them (Corty et al., 2009; Hattori et al., 2008; Hughes et al., 2007; Matthews et al., 2007; Soba et al., 2007). Dscam1 endows different neurons with unique surface identities via extensive alternative splicing to permit self vs. non-self discrimination (Corty et al., 2009; Jan and Jan, 2010; Millard and Zipursky, 2008). Several genes have been found to promote repulsion between branches of class IV neurons, including tricornered (trc), which encodes a serine threonine kinase, furry (fry), and turtle (tutl), encoding an immunoglobulin superfamily member, however these appear to function independently of Dscam1 (Emoto et al., 2004; Long et al., 2009; Soba et al., 2007). Consequently, how Dscam1 and other factors combine to support self-avoidance is not currently known. One notable distinction is that Dscam1 is required for self-avoidance in all classes of neurons (Hughes et al., 2007; Matthews et al., 2007; Soba et al., 2007), whereas action of other molecules appears to be limited to the highly complex class IV neurons. It is not clear how self-repulsion mechanisms might differ between different classes of neurons, but understanding this distinction should begin to extend current models.

The identification of molecules that prevent crossing and promote dendrite spacing has been aided by treating da neuron dendrites as a largely two-dimensional array on the basal surface of the epidermis. Such an organization simplifies mechanistic models since dendrites will, in principle, have equivalent capacities to interact with other nearby dendrites. Consequently, when crossing of dendrites was observed, the underlying cause has been attributed to defects in the machinery underlying branch recognition or repulsion. Conversely, in such a system, the potential for non-contacting crossings, or crossing in three dimensions, should be negligible. However, the relationship between da neuron dendrites, the extracellular matrix (ECM), and epidermal cells has not been examined at high enough resolution to validate this view, so more complex interactions between dendrites and their substrate that impact avoidance between dendrites and arbor patterning remain an interesting possibility.

Here we investigate dendrite-substrate relationships in da sensory neurons and their impact on dendritic morphogenesis. We show using electron microscopy that dendrites are positioned at the basal surface of the epidermis in contact with the ECM, or deeper within the epidermis where they become enclosed by epidermal cell membrane. We provide evidence that integrins, transmembrane receptors that provide a physical and signaling link between the ECM and the cytoskeleton (Bokel and Brown, 2002; Hynes, 2002), promote positioning on the basal epidermal surface. Integrins likewise prevent self-crossing between class IV da neuron dendrites and support dendritic maintenance. Our analysis suggests that integrins limit self-crossing not by controlling recognition or repulsion directly, but by impacting dendritic enclosure and, consequently, the ability of dendrites to participate in contact-mediated repulsion mediated by Dscam1. We propose that dendrite-substrate relationships established by integrins, and dendrite-dendrite repulsion regulated by Dscam1, control the positioning and spacing of sensory arbors in three dimensions during development for appropriate coverage of sensory territories.

Results

Integrins function in sensory dendrite morphogenesis and self-avoidance

We examined how molecular interactions between dendrites and the ECM influence da neuron morphogenesis by focusing on integrin receptors, which provide a major link between cell surfaces and the ECM. Functional integrin receptors are heterodimers of α and β integrin subunits. The Drosophila genome encodes two β subunits, βPS and βν (MacKrell et al., 1988; Yee and Hynes, 1993) and five α subunits. βPS-integrin, encoded by the myospheroid (mys) gene, predominates in all tissues except the midgut (Yee and Hynes, 1993). To determine whether integrins function cell autonomously in neurons during dendrite development, we generated mys mutant MARCM clones (Lee and Luo, 1999). We tested two different mys alleles, mys¹ and mys^XG43, and both caused reductions in total dendritic length and branch number in class I neurons (Figures 1A–1B and 1E–1F). To determine which α subunit(s) might be important, we examined clones lacking multiple edematous wings (mew), which encodes αPS1 and inflated (if), which encodes αPS2. mew, if double mutant clones showed similar reductions in length and branching as mys clones, indicating that one, or both, of these genes is important for dendrite morphogenesis (Figures 1C and 1E–1F). We examined roles for individual α subunits by transgenic RNAi-based knock down (Dietzl et al., 2007) and found that depletion of mew, but not if, transcripts in class I neurons using 221-Gal4 led to a defect in dendritic arborization similar to that caused by RNAi of mys (Figures 1I–1J). Consistent with these results, we did not observe a dendrite branching or length phenotype in if MARCM clones (p>0.05; data not shown). Thus, PS1 (αPS1βPS) probably plays a primary role in dendritic morphogenesis, although these data do not exclude a possible neuronal role for PS2 (αPS2βPS). Finally, consistent with a role for integrin-mediated adhesion in dendritic arborization, mutations in Talin, a protein encoded by the rhea gene that promotes high affinity interactions between integrins and the ECM (Moser et al., 2009), caused defects that were similar to those caused by mys mutations in class I neurons (Figures 1D, 1G–1H). Together, these results reveal a cell-autonomous requirement for integrins in da neuron dendritic elaboration and/or dendritic branch maintenance, likely reflecting a requirement for adhesive interactions between dendrites and the ECM.

Figure 1. Integrins are required cell autonomously for sensory dendrite morphogenesis.

(A) Wild-type (FRT19A) MARCM clone of the class I neuron ddaE.

(B) mys^XG43 ddaE MARCM clone shows a simplified dendritic arbor.

(C) Similar dendrite phenotype is seen in neurons mutant for two α integrin subunits, mew (αPS1) and if (αPS2).

(D) rhea13-8 ddaE mutant clone has dendrite defects that are similar to mys ddaE clones.

(E) Quantification of branch points for wild-type (+/+) FRT19A, mys¹, mys^XG43, and mew if ddaE MARCM clones (all generated with FRT19A). n values are indicated in bars.

(F) Quantification of total dendrite length for FRT19A control, mys¹, mys^XG43, and mew if ddaE MARCM clones. n values for each genotype are the same as in (E).

(G) Quantification of branch points for wild-type (FRT2A) and FRT2A rhea ddaE MARCM clones. n values are indicated in bars.

(H) Quantification of total dendrite length for FRT2A and FRT2A rhea ddaE MARCM clones. n values are the same as indicated in (G).

(I) Quantification of branch points for ddaE neurons expressing UAS-dicer2, UAS-mys-RNAi, UAS-mew RNAi, or UAS-if RNAi (all with UAS-dicer2) under the control of 221-Gal4, UAS-mCD8::GFP. n values are indicated below boxplots.

(J) Quantification of total dendrite length for ddaE neurons expressing UAS-dicer2, UAS-mys-RNAi, UAS-mew RNAi, or UAS-if RNAi.

(K) Wild-type FRT19A control class IV neuron clone. Arrowheads indicate dendritic crossing points.

(L) mys^XG43 class IV MARCM clone. Arrowheads indicate dendritic crossing points.

(M) Quantification of the number of branch points for class IV ddaC neurons in wild-type (+/+) and mys mutant clones. n values are indicated in bars.

(N) Quantification of sister dendrite overlaps in wild-type and mys clones standardized to dendrite length. n values are indicated in bars.

Scale bars = 50 µm.

Barplots show mean + standard deviation (S.D.). p values are indicated as: *=p < 0.05, ** = p < 0.01, and *** = p < 0.001 as assessed by pairwise t-tests with Bonferroni correction.

Boxplots show median (thick line), quartiles Q1–Q3 (25%–75% quantiles; grey box), and data in the 1.5× quartile range (dashed bars). *=p < 0.05, ** = p < 0.01, and n.s. = not significant, as assessed by Wilcoxon rank-sum test (I) or pairwise t-tests with Bonferroni correction (J).

Anterior is to the left and dorsal top for these and subsequent confocal images.

See also Figure S1.

We next used MARCM to examine the requirements for integrins in dendritogenesis of the complex class IV neuron, ddaC. Like class I neurons, ddaC mys clones showed a decrease in dendritic branch points (Figure 1K–1M). Class IV dendrites also normally show robust self-repulsion between branches with only occasional crossing errors (Figure 1K). We found that mys ddaC clones showed increased self-crossings and thus appeared to be defective in this repulsive response (Figures 1L, 1N, S1A). By contrast, sister dendrite crossing as a proportion of total branch number or total length was not significantly affected in class I mys clones (both p > 0.05, Wilcoxon rank-sum test). Excessive dendrite self-crossing observed in class IV neurons suggested that integrin-mediated dendrite-ECM interactions promote dendritic self-avoidance.

We next examined expression patterns of integrins in the peripheral body wall at third instar larval stages. Immunolabeling with anti-βPS, αPS1, and αPS2 integrin revealed localization in puncta on the basal surface of the epidermis, and enrichment alongside dendrites (Figure S1B–S1F). Expression across the epidermis prevented unambiguous assessment of expression in da neuron dendrites, however examination of arbors growing over mys epidermal clones that were devoid of βPS integrin provided support for dendritic localization (Figures S1C–S1D'). In these cases, labeling was most consistently observed at class I dendrites, while localization at class IV dendritic branches was barely detectable or below the limit of detection at least within these regions examined (Figure S1D'). Together, analysis of mutant phenotypes and expression supports a cell autonomous requirement for integrins in sensory dendrite morphogenesis.

Drosophila da sensory neuron dendrites lie on the basal surface of the epidermis and can become enclosed by epidermal membrane

The above results suggested that interactions between dendrites and the ECM were important for dendrite development. To examine the relationship between dendrite surfaces and their substrate in larval da neurons we performed transmission electron microscopy (TEM). Larval dendrites appear to project largely in two-dimensions across the basal surface of the epidermis when viewed with light microscopic resolution, but dendritic positioning relative to the epidermis has not been resolved at high resolution. In thin sections of abdominal segments cut en face to the body wall, processes containing arrays of multiple parallel microtubules were identified near the basal surface of the epidermis (Figure 2A). To determine the relationship between dendritic branches and epidermal cells, we made transverse sections to visualize processes in profile (Figure 2B). A notable feature of dendrites in cross section was their variable depth in relation to the basal surface of the epidermis. One population of arbors sat in shallow depressions of epidermal membrane in contact with ECM (Figures 2C–2D). One or more electron-dense putative junctions were often seen adjacent to these dendrites (Figures 2C–2D, asterisks). In contrast to this population of surface dendrites, other dendrites were located within invaginations of epidermal cell membrane that could be long and sinuous (Figures 2E–2F). Dendrite depth below the basal surface of the epidermis ranged between approximately 80–890 nm in our sampling (n=11 branch profiles). Measurements of dendrite diameters ranged between 140–1250 nm, with the finest dendrites that were identified (less than approximately 360 nm across) residing on the basal surface and other dendrites residing either on the surface or within invaginations (n=31 branch profiles). These EM studies therefore show positioning of larval sensory neuron dendrites along the basal surface of the epidermis in contact with the ECM and also reveal enclosure within epidermal cell invaginations (Figures 2G–2H). We speculated that the arrangement of dendrites on the basal surface or within invaginations may have important implications for arbor development, and investigated mechanisms of its control.

Figure 2. Electron microscopy of the dendrite-epidermal interface.

(A) Electron microscopy of a branched process containing multiple parallel microtubules in third instar larva cut parallel to the epidermal surface. Key: d = dendrite.

(B) Schematic of transverse sectioning for transmission electron microscopy (TEM).

(C–D) Representative electron micrographs of dendrites of varying size located at the basal surface of the epidermis. Key: d = dendrite, ECM = extracellular matrix, * = junction. Apical-basal orientation of epidermis is indicated in (C).

(E–F) Electron micrographs of dendrites that are enclosed by epidermal membrane. Key is the same as in (C). Arrowheads indicate membrane segments that extend between the dendrite and the basal surface of the epidermis. Note apparent junctions at the basal end of these membrane segments (asterisk).

(G) Schematic of surface dendrite.

(H) Schematic of enclosed dendrite.

Scale bars = 200 nm.

Identification of markers of dendrite-substrate relationships

The body wall is covered by dendrites of several distinct classes of da neurons that differ in branching morphology. To determine how dendritic enclosure relates to da neuron class and characterize the distribution of enclosures across dendritic arbors, we sought markers of enclosed and surface branches. We examined a collection of GFP trap lines for expression associated with da neurons (Buszczak et al., 2007; Morin et al., 2001; Quinones-Coello et al., 2007) and observed that several septate junction resident proteins, including Discs large, Scribble, and ATPalpha showed intermittent enrichments along class IV dendritic arbors (Figures S2A–S2A' and data not shown). Antibodies against the FERM protein Coracle, which also localizes to septate junctions (Fehon et al., 1994), showed similar enrichment (Figures S2B–S2C'). We observed that anti-Coracle enrichments were associated primarily with class IV dendrites, with less extensive labeling along the trajectories of class III, II, and I neurons (Figure S2D).

To test for association between anti-Coracle labeling and enclosed dendrites, we sought an additional independent marker of these regions. We reasoned that dendritic branches that are enclosed by epidermal membrane should be at least partially protected from surface labeling by HRP antibodies, which recognize cell surface antigens contributed by numerous neuronal proteins (Jan and Jan, 1982; Paschinger et al., 2009). We labeled animals carrying the class IV marker ppk-Gal4, UAS-mCD8GFP sequentially with anti-HRP in the absence of detergent (Triton X-100), followed by Triton treatment and anti-GFP to mark sensory dendrites and anti-Coracle to mark the epidermis. As a control, Triton was included during all antibody incubations. In the presence of Triton, anti-HRP labeling was fairly uniform along the dendrites of all neuronal classes (Figures 3A–3B). By contrast, when anti-HRP labeling was performed without Triton, we observed alternating strong and weak HRP-like immunoreactivity along dendrites (Figures 3C–3D). The ends of terminal branches, but not necessarily the entire terminal branch, usually remained strongly labeled (Figure 3D'). Class III neurons also showed diminished labeling along some major dendrites (Figure 3C' and data not shown). Labeling of membrane-bound GFP in class IV dendritic branches, performed in the presence of Triton, did not co-vary with anti-HRP signal (Figures 3C'' and 3D''). Thus, it appeared that diminished anti-HRP labeling arose from lowered accessibility of dendrites when labeling was restricted to membrane surfaces. Combining analysis of anti-HRP and anti-Coracle labeling, we observed a negative correlation between the intensity of anti- HRP and anti-Coracle along class IV dendrites when anti-HRP labeling was performed without Triton (Figures 3H–3J; Spearman's rank correlation rho = −0.709, p < 0.001), but not when all labeling was performed in the presence of Triton (Figures 3E–3G; Spearman's rank correlation rho = 0.278, p > 0.05). These data suggest that anti-Coracle labeling is intermittently enriched where dendritic branches show lower membrane accessibility.

Figure 3. Identifying markers of enclosed dendrites.

(A–B) Labeling of third instar ppk-Gal4, UAS-mCD8-GFP larva with anti-HRP in Triton. Anti-GFP and anti-Coracle channels are omitted for clarity.

(C–C") Labeling of third instar ppk-Gal4, UAS-mCD8-GFP larva with anti-GFP and anti-HRP. Anti-Coracle labeling is omitted for clarity. Anti-HRP labeling was performed prior to Triton X-100 exposure. HRP channel shows different labeling intensities along major branches of class I, III, and IV neurons (arrows). GFP labeling, performed after Triton exposure, is fairly uniform along class IV dendrites.

(D–D") Labeling of third instar ppk-Gal4, UAS-mCD8-GFP larva with anti-GFP and anti-HRP. Anti-Coracle labeling is omitted for clarity. Anti-HRP labeling was performed prior to Triton X-100 exposure. (D') HRP channel shows differential labeling intensities along higher order branches of class IV neurons (arrowheads). (D") GFP labeling, performed after Triton exposure, is fairly uniform along class IV dendrites.

(E–E"') Labeling of third instar ppk-Gal4, UAS-mCD8-GFP larva with anti-GFP, HRP, and Coracle, all after detergent treatment, reveals uniform labeling with GFP and HRP, and intermittent labeling of Coracle along a class IV dendrite. A dashed arrow lies next to the dendritic region quantified in the linescan (G).

(F) Schematic of labeling patterns shown in (E).

(G) Linescan of dendrite marked with a dashed arrow in (E) showing labeling intensity along length of the dendrite for GFP (green), HRP (red), and Coracle (blue). Note that while the Coracle intensity varies, the trajectories of GFP and HRP linescans are fairly uniform and equivalent. A high-Coracle region is indicated by a horizontal blue bar.

(H–H"') Dendrites were labeled with anti-HRP in non-detergent conditions, treated with detergent (Triton), and then labeled with anti-GFP and anti-Coracle. GFP immunoreactivity is fairly uniform, but anti-HRP immunoreactivity is variable. Low HRP labeling is observed along a high-Coracle dendritic region and stronger HRP labeling is observed along low-Coracle dendritic region. A dashed arrow lies next to the dendritic region quantified in the linescan (J).

(I) Schematic of labeling patterns shown in (H)

(J) Linescan of the dendrite marked by a dashed arrow in (H) shows that HRP intensity is dampened relative to GFP intensity where Coracle labeling is high (blue horizontal bars), and that anti-HRP fluorescence intensity increases where Coracle levels drop.

Scale bars = 25µm

See also Figure S2.

To further test for an association between anti-Coracle labeling and enclosure, we correlated light microscopic observations of anti-Coracle localization with electron micrographs of dendrites in cross section. Whole mounts of body walls viewed en face using light microscopy revealed enrichments of anti-Coracle labeling along the epidermis (Figures S2E and S2F). Regions of interest were re-examined in cross section at the light microscopic level to identify the labeled landmarks and select these regions for thin sectioning (Figures S2E' and S2F'). We examined thin sections using TEM and found enclosed dendrites whose position correlated well with the locations of Coracle enrichments seen at the light microscopic level (Figures S2E'' and S2F'').

This analysis therefore supported an association between anti-Coracle labeling and dendritic enclosure. Based on these results, we concluded that anti-Coracle and anti-HRP labeling could provide useful markers to study the effects of integrins on dendrite morphogenesis.

Integrins counteract enclosure of dendrites

In principle, enclosure of dendrites might involve part of a neuron pushing into a host epidermal cell. We noted previous studies showing that a cell-in-cell phenomenon may arise from reduced integrin-based adhesion to the ECM leading to invasion of one cell into another (Overholtzer et al., 2007). We examined the effect of integrins on enclosure of dendrites by generating mys mutant clones and labeling with anti-Coracle and anti-HRP antibodies. We focused our analysis on class I neurons because their dendrites normally showed minimal signatures of enclosure (Figure S2D). Class I mys^+/− neurons likewise showed very rare apparent enclosure, primarily along a main proximal branch (Figures 4A–4B, and 4E). By contrast, we found that homozygous mutant (mys^−/−) class I MARCM clones acquired significant enrichments of anti-Coracle labeling (Figures 4C–4E; n=5). Similar to wild-type class IV neurons, strong anti-Coracle labeling along mys clones was interrupted by low-Coracle regions (Figures 4C–4D) and the strength was negatively correlated with anti-HRP labeling performed without Triton (Figures 4F–4H; Spearman’s rank correlation rho = −0.644; p < 0.001; n=30 dendritic regions from 5 clones). These results suggested that integrins counteract enclosure of sensory dendrites.

Figure 4. Loss of integrins leads to dendritic enclosure.

(A–A") mys+/− class I ddaE neuron, acquired from the same animal as the MARCM clone in (C), labeled for anti-GFP (ddaE is not labeled), anti-HRP, and anti-Coracle. Isolated GFP and Coracle channels are shown to the right of the merged image. ddaE arbor is indicated by small yellow arrowheads. Class IV ddaC dendrites are indicated by red arrows. Immunolabeling was performed in the presence of Triton throughout. Coracle enrichments show little co-localization with class I ddaE dendrites (small yellow arrowheads) but more extensive co-localization with class IV ddaC dendrites (red arrows). Large yellow arrowheads indicate location of ddaE cell body.

(B) Schematic drawing of labeling pattern shown in (A) with class I ddaE neuron drawn with thick lines and class IV dendrites drawn with thin lines.

(C–C") mys¹ mutant MARCM clone labeled with anti-GFP, anti-HRP, and anti-Coracle. Isolated GFP and Coracle channels are shown to the right of the merged image. Immunolabeling was performed in the presence of Triton throughout. Coracle is enriched along the trajectories of mys mutant class I dendrites (yellow arrowheads). Coracle labeling along class IV neuron dendrites is indicated by red arrows.

(D) Schematic drawing of labeling patterns shown in (C).

(E) Quantification of mean length of ddaE arbors showing Coracle enrichments in matched heterozygous (mys+/−; n=5) and mys¹ neurons (n=5).

(F–F") mys^XG43 mutant MARCM clone showing that enrichment of Coracle correlates with decreased HRP immunofluorescence when anti-HRP labeling is performed prior to detergent treatment. Isolated HRP and Coracle channels are shown to the right of the merged image. A dashed arrow in (F) lies adjacent to the region of dendrite quantified in the linescan shown in (H).

(G) Schematic drawing of labeling patterns shown in (F).

(H) Linescan of class I mys MARCM clone dendrite labeled with dashed arrow in (F) showing HRP (red line) and Coracle (blue line) fluorescence intensities (in arbitrary units, A.U.). GFP labeling is omitted from linescan for clarity. Anti-HRP intensities are lowest in regions of high Coracle, and increase where Coracle levels decline.

Scale bars = 50 µm.

Barplot shows mean + S.D. p values are indicated as ** = p < 0.01 as assessed by Student’s t-test.

We next asked whether, conversely, overexpression of integrins could reduce the normal enclosure of class IV dendrites. Consistent with this notion, co-expression of UAS-if (αPS2) and UAS-mys (βPS) along with UAS-mCD8::GFP under the control of the class IV neuron driver ppk-Gal4 could reduce Coracle enrichments along dendrites compared to larvae expressing only UAS-mCD8::GFP (Figures 5A–5D). Together, these results support a role for integrins in the positioning of dendrites on the basal surface of the epidermis in contact with the ECM. Given our loss-of-function and overexpression results with α subunits, it may be that αPS2 and αPS1 have at least partially interchangeable ability to promote basal positioning of sensory dendrites, consistent with evidence for their functional interchangeability in some other contexts (Martin-Bermudo et al., 1997; Roote and Zusman, 1996).

Figure 5. Integrin overexpression can suppress markers of dendritic enclosure.

(A–A") Labeling of ppk-Gal4, UAS-mCD8-GFP/+; ppk-eGFP/+ third instar larva with anti-mCD8 (green), anti-HRP, and anti-Coracle reveals Coracle labeling along class IV neuron dendrites (yellow arrowheads). Labeling is also seen at junctions between epidermal cells. Anti-HRP labeling was performed in PBS without detergent. A red asterisk indicates a dorsal external sensory neuron.

(B) Schematic of class IV dendrites (green) and Coracle labeling (blue).

(C–C") Labeling of ppk-Gal4, UAS-mCD8-GFP/+; ppk-eGFP/UAS-αPS2, UAS-βPS third instar larva with anti-mCD8 (green), anti-HRP, and anti-Coracle reveals diminished anti- Coracle labeling of class IV neuron dendrites. Labeling is retained at epidermal cell-cell junctions. Anti-HRP labeling was performed in PBS without detergent. Yellow arrowheads indicate areas of dendrite-associated Coracle labeling. A red asterisk indicates a dorsal external sensory neuron.

(D) Schematic of class IV dendrites (green) and Coracle labeling (blue).

Scale bars = 50 µm.

Role of integrins in the maintenance of sensory dendrites

Branching and dendritic length reductions in mys mutant MARCM clones could conceivably arise from decreased dendrite outgrowth, disrupted dendrite maintenance, or both. To test for possible effects of integrins on dendritic maintenance, we imaged wild-type and mys mutant MARCM clones starting at second instar larval stages, processed for immunohistochemistry approximately two days later as third instars, and assessed the status of terminal and internode branches (Figures 6A and 6B). Consistent with prior studies (Parrish et al., 2009; Sugimura et al., 2003), branches of wild-type class I neurons nearly all lengthened during this interval, with the only exception being some short branches (less than approximately 20 µm) that were more dynamic and could lengthen, shorten, or fully retract (Figures 6A and 6C). In mys mutant class I clones, shorter branches could likewise be dynamic, however, unlike wild-type clones, several longer terminal dendritic segments had shortened (Figures 6B and 6D; mean initial length of regressed branches = 39.4 µm, mean length of dendrite regression = 10.9 µm; n=23 branches from 4 neurons). Notably, examination of third instar mys MARCM clones revealed “tails” of anti-Coracle labeling that extended beyond dendritic endings but showed no obvious tracking of other dendrites in the vicinity (Figure 6E). The majority of tails were associated with dendrites that showed net decreases in length between second and third instar stages (Figure 6D). Moreover, the paths of tails closely matched the positions and orientations of lost branch segments (compare Figures 6B and 6E). These observations together support a role for integrins in the maintenance of terminal dendritic branches of class I neurons. We speculate that tails may represent markings left in the epidermis upon regression of enclosed endings.

Figure 6. Defective dendritic maintenance in class I integrin mutant clones.

(A) Wild-type MARCM clone of class I neuron ddaE imaged in second and third instar larval stages. Very little addition or retraction of branches is observed, however branches elongate and maintain territory coverage.

(B) mys¹ MARCM clone of class I neuron ddaE imaged in second and third instar larval stages. Arrowheads indicate areas of dendritic arbor shown enlarged in (E).

(C) Plot of branch lengths in wild-type ddaE MARCM clones in second and third instar stages (n=4 clones). Blue circles represent non-terminal branch segments and yellow circles represent terminal branches. The dashed line represents the isolength line.

(D) Plot of branch lengths in mys ddaE MARCM clones in second and third instar stages (n=4 clones). Blue circles represent non-terminal branch segments, yellow circles represent terminal branches, and red circles represent terminal branches that were identified as having Coracle tails. The dashed line represents the isolength line.

(E) Dendrites of mys MARCM clone in third instar larva that have regressed since second instar. Coracle tails are indicated by red arrows. Branches correspond to branches with arrowheads in (B).

Scale bars = 50 µm (A, B), 25 µm (E).

Identification of non-contacting crossings in class IV neurons that associate with sites of dendrite enclosure

Class IV da neurons have provided insights into mechanisms that prevent dendritic crossing and promote non-redundant territory coverage. The dual effect of integrins on dendrite enclosure and dendrite crossing led us to examine the consequences of three-dimensionality for dendritic self-avoidance in class IV neurons. We first asked whether sister dendrites that occasionally cross each other in wild-type class IV neurons show evidence for differences in dendrite depth. We used markers of enclosure (anti- HRP without detergent and anti-Coracle) to examine self-crossings in class IV neurons labeled with ppk-Gal4, UAS-mCD8::GFP. We found occasional self-crossings and, in all but a few crossovers (26/28 or 93% of crossings, n=10 cells), at least one of the crossing branches extended along a region of Coracle enrichment (either along a putative enclosure or at a junction between two epidermal cells; Figures 7A–7B, 7E). Anti-HRP labeling was also diminished in branches that showed high Coracle labeling (Figures 7A''). Enclosure was less often observed at the more numerous non-sister (heterotypic) dendritic crossings between class IV neurons and other classes of neurons (Figures 7C–7E). This observation suggests that enclosures are unlikely to arise solely as a consequence of two dendrites crossing. Thus, dendrite self-crossings in wild-type class IV neurons were almost exclusively a non-contacting type of dendrite crossing. We noted similar immunohistochemical signatures of high Coracle and low HRP at crossings between branches from different class IV neurons, suggesting that non-contacting crossings can also lead to apparent violations in class IV neuron tiling (Figure S3A).

Figure 7. Dendritic enclosure and basis for self-crossing in integrin and Dscam1 deficient neurons.

(A–A"') Isoneuronal overlaps can occur where one branch is labeled strongly by anti-Coracle and another branch is unlabeled. Yellow arrow indicates point of overlap. Note low levels of anti-HRP in one of the crossing branches (A").

(B) Schematic drawing of labeling patterns shown in (A).

(C–C"') Heteroneuronal overlap between class III and class IV neuron. Yellow arrowhead indicates point of overlap. Note lack of Coracle enrichment along either branch.

(D) Schematic drawing of labeling patterns shown in (C).

(E) Summary of Coracle immunohistochemical signatures of crossing dendrites in ppk-Gal4, UAS-mCD8::GFP third instar larvae. Contacting crossings are scored by absence of Coracle along both dendrites (left column). Non-contacting crossings are scored by presence of Coracle along at least one dendrite at crossing (right column). The predicted dendritic arrangements are schematized as two crossing dendrites on the basal surface of the epidermis, or one surface dendrite and one enclosed dendrite. The proportions of isoneuronal crossings (n=28 crossings; 10 cells) and heteroneuronal/heterotypic crossings (n=154 crossings from 5 cells) in each category are shown.

(F) Stacked barchart showing percentages of dendrite crossings in mys and Dscam1 clones scored as contacting crossing (red bar) or non-contacting crossing (blue bar) as assessed by Coracle labeling.

(G) Fluorescence intensity plots of Coracle labeling (measured in arbitrary units, A.U.) in 17 dendrite-dendrite crossings in mys clones in which relative depth of crossing dendrites could be discriminated. Coracle labeling intensity along the more apical branch is indicated by a blue box and paired with the associated basal branch by an “X”. In each crossing the apical-most branch shows higher levels of Coracle.

(H) Quantification of fluorescence intensity along crossing dendrites that could be resolved as “basal” and “apical” in 1µm confocal Z-steps (n=17 pairs). Shown are median of median values of Coracle immunofluorescence in linescans.

(I–I') Confocal image of Dscam1 class IV MARCM clone showing two isolated non-contacting crossings (arrows) and an area of dendrite crossings that includes putative contacting crossings (arrowhead). Anti-GFP is in green and anti-Coracle is in blue. Coracle channel is shown in isolation (I') with arrows and arrowheads indicating regions of non-contacting and contacting crossing, respectively.

(J) Schematic drawing of labeling patterns shown in (I).

Datasets are presented in boxplots as median (thick line), quartiles Q1–Q3 (25%–75% quantiles; blue box), and data in 1.5× quartile range (dashed bars). *** = p < 0.001 by Wilcoxon rank-sum test.

Scale bars = 12.5 µm.

See also Figure S3

Different basis for self-crossing defects in myospheroid and Dscam1 mutant neurons

Given the strong tendency for non-contacting self-crossing in class IV neurons, we next examined types of crossings in class IV MARCM clones mutant for either mys or Dscam1, a gene that is required for self-avoidance in all classes of da neurons (Hughes et al., 2007; Matthews et al., 2007; Soba et al., 2007). We identified instances of dendrite crossing in clones and assessed the evidence for enclosure along the trajectory of crossing dendrite(s) using anti-Coracle labeling. In mys MARCM clones, anti-Coracle was associated with all but a small fraction of crossing dendrites (96% or 182/190; n=9 neurons; Figure 7F). Crossovers occurred both at junctions between two epidermal cells (that label strongly with anti-Coracle), and at non-border anti-Coracle enrichments. We examined whether the dendrite associated with Coracle enrichment indeed resided deeper in the epidermal layer, and consistent with this, found that in each crossing that could be separated in successive confocal sections, Coracle labeling correlated with the path of the deeper, more apically positioned dendrite (correlation between Coracle labeling and apical dendrite positioning: p < 0.001, n=17; Figures 7G, 7H, S3B–S3E). These data therefore suggest that loss of integrins impacts dendrite crossing by affecting the three-dimensional positioning of dendrites and inflating the number of non-contacting crossings.

In contrast to mys clones, Dscam1 MARCM clones showed a smaller proportion of crossings that could be associated with Coracle enrichments (56/89 or 63% putative non-contacting crossings and 37% putative contacting crossings; n=4 clones examined; Figure 7F). These results suggest that many, but not all, self-crossings seen in Dscam1 mutant class IV neurons result from defects in contact-mediated repulsion rather than being almost solely non-contacting crossings. Crossings in Dscam1 mutant neurons often occurred in clusters of crossing and bundling along, or at the ends of, major dendrites (Figures 7I–7J). The majority of crossings that were scored as contacting (97%) occurred in these regions. Non-contacting crossings were scored at approximately equal frequency as contacting crossings within crossing clusters (32 contacting, 34 non-contacting), but appeared to dominate in unbundled crossings that occurred in isolation between two terminal branches or between a lower-order branch and terminal branch (22 non-contacting, 1 contacting; Figures 7I–7J). One scenario that might explain these observations is that Dscam1 mutations lead to defects in contact-mediated repulsion in class IV neurons that are manifest primarily as bundles and clusters of crossings. Non-contacting crossings might arise with certain probability elsewhere due to enclosure of class IV dendrites. Tight clustering of dendrites in Dscam clones could also conceivably have a secondary effect of enhancing apparent non-contacting crossings within crossing clusters if bundling precedes ingression. Together, these results suggest that integrins support self-avoidance by promoting the positioning of dendrites on the basal surface of the epidermis, where sister dendrites can reliably recognize and repel each other through the action of Dscam1.

Discussion

The molecular interplay between growing dendrites, ECM, and surrounding cells is likely intricate and relevant for diverse morphological and functional properties of dendrites. We provide evidence that da sensory neurons in Drosophila develop a complex spatial relationship with the ECM and epidermis that is under the control of integrin receptors. Our results further suggest that the relationships that are established between dendrites and their substrate have important implications for dendritic morphology, self-avoidance, and maintenance.

Integrins control dendrite positioning along the basal surface of the epidermis

EM analysis indicated that larval da neuron dendrites can reside either on the basal surface of epidermal cells in contact with ECM or intermittently enclosed within epidermal invaginations. Enclosed dendrites appear separated from the extracellular space by sheets of closely apposed epidermal cell membrane that originate at the basal surface. Based on marker expression, these arbors can become enclosed along major proximal regions, and also intermittently along higher-order branches. By contrast, terminal endings may remain on the basal surface of the epidermis. Within the enclosure, epidermal and dendritic membranes appeared to be closely juxtaposed. A prior TEM study in blowfly described ensheathment of dendrites by glia containing fluid-filled spaces, and termination of dendrites within epidermal invaginations (Osborne, 1964). These differences might point to additional diversity in dendrite-epidermal relationships, perhaps arising due to differences in the identities (or regions) of the sensory neurons examined, or perhaps species differences. An important goal will be to identify additional markers to extend this characterization, as well as examine other molecules at the dendrite-epidermal interface for possible roles in the establishment or maintenance of specific interactions.

Our results provide molecular insight into how dendrites at the basal surface are segregated from enclosed dendrites. Loss of integrin function in neurons increases dendrite enclosure according to marker expression, whereas neuronal overexpression of integrins has the opposite effect, diminishing markers of enclosure along class IV dendrites. Reduced integrin-based adhesion, and perhaps a weakening of dendrite-ECM attachment by mechanical forces on the dendrites, might establish conditions that favor enclosure. Interactions with epidermal cells could likewise be important. For example, in tumor cells, integrin-based adhesion to the extracellular matrix is thought to counterbalance adherens junction-based compaction forces between cells to prevent cell invasion (Overholtzer et al., 2007). In cells that are detached from the matrix, adhesive contacts are predicted to shift to predominantly cell-cell adhesion with imbalanced compaction forces pushing one cell into another (Overholtzer et al., 2007). Although the precise mechanism for how enclosure of da neuron dendrites arises is presently unknown, it will be interesting to examine whether, on a local scale of dendrite segments, balanced adhesion may play a role.

The physiological consequences of placement of dendrites in proximity to the ECM or in enclosures are unknown. The ECM might influence the transduction of mechanical forces to the neuronal cytoskeleton and impact mechanosensation (Du et al., 1996; Emtage et al., 2004), and studies in C. elegans suggest roles for integrin signaling in touch sensitivity (Calixto et al., 2010). In the da neuron system, class IV neurons are thought to sense noxious mechanical, thermal, or photic stimuli, whereas class I neurons appear to function as proprioceptors (Hughes and Thomas, 2007; Hwang et al., 2007; Song et al., 2007; Xiang et al., 2010). Mechanosensation could be affected by the specific relationship between sensory arbors and surrounding tissues. For example, mechanical stimuli or compression impinging on the body wall could distort surface versus enclosed dendrites in different ways (Osborne, 1964). Intermittent enclosure could also result in spaced tetherings of dendrites, which could conceivably establish local foci for mechanosensation across an arbor (Hall and Treinin, 2011). Finally, it is worth noting that among the different sensory neuron types, enclosure was observed predominantly along neurons with more highly arborized dendrites. One speculative possibility is that this arrangement could isolate dendritic membrane and conceivably impact signal transduction along more expansive arbors. Behavioral analyses should begin to address these and other possible functional consequences of the relationship between da neuron sensory dendrites and their substrate.

A role for integrins in sensory dendrite maintenance

Integrin-deficient class I neurons showed reduced dendritic length and branching complexity and also acquired markers of dendritic enclosure, including Coracle immunoreactivity and intermittent protection from surface anti-HRP labeling. How are these phenotypes related? Fly sensory neurons show ongoing growth of dendrites during larval development so that territory coverage scales with overall expansion of the body wall (Parrish et al., 2009; Sugimura et al., 2003). Dendrite morphology defects observed in integrin mutant neurons could therefore arise for any of several reasons, including defective branch initiation, stabilization, scaling growth, and/or maintenance. While these underlying causes are not mutually exclusive, our results suggest that the phenotype is contributed at least in part by a failure in dendritic maintenance and susceptibility of arbors to regression in the absence of integrin-based ECM interaction. Branch maintenance defects are consistent with prior studies of the vertebrate retina, which showed that β1-integrins are required for the maintenance of mature dendrites (Marrs et al., 2006). Integrins may also be involved in Abelson (Abl) and Abl-related gene (Arg)-dependent maintenance of cortical dendrites (Moresco et al., 2005). One notable feature of regressed dendritic endings in da sensory neurons is that they appeared to leave markings of enclosure in their wake. These results imply that positioning of dendritic terminal endings of at least some classes of da neurons on the basal surface of the epidermis in contact with ECM is important for their maintenance. It will be interesting in the future to examine whether other pathways that are important for dendritic maintenance (Parrish et al., 2007) might act by modulating interactions between dendrites and the ECM.

Dendrite-substrate interactions influence self-avoidance between sensory neuron dendrites

Dendritic self-avoidance depends on recognition between sister dendrites that leads to repulsion and separation. Whereas sister branches self-avoid, branches from different cells can overlap. Such self-repulsion is widespread in nervous systems and ensures non-redundant coverage of territories (Grueber and Sagasti, 2010). The homophilic transmembrane receptor Dscam1 is required for self-avoidance in Drosophila in both central and peripheral neurons, including all classes of da neurons (Hattori et al., 2008; Hughes et al., 2007; Matthews et al., 2007; Soba et al., 2007). In addition to Dscam1, self-crossing, specifically of class IV dendrites, is prevented by the action of several additional molecules, including Furry and the serine/threonine kinase Tricornered (Emoto et al., 2004), target of rapamycin, Sin1, and Rictor (Koike-Kumagai et al., 2009), and Turtle (Long et al., 2009). One interpretation of the specificity towards class IV neurons is that robust self-avoidance between dendrites could require several independent pathways (Long et al., 2009). For example, dendrites with high branch complexity or surface area may require multiple signals for self-recognition or repulsion across all parts of the arbor.

As shown here, integrin receptors likewise prevent excessive self-crossing of class IV dendrites, and our data support the conclusion that crossing in integrin deficient neurons arises because of dendritic enclosure within membrane of epidermal cells, resulting in almost exclusively non-contacting crossing between dendrites. The conclusions with integrins were different from the extensive crossing and bundling observed in Dscam1 mutant neurons, which our data suggest is contributed both by defects in contact-mediated repulsion and by non-contacting crossings. These data therefore provide cell biological support for previous studies of Dscam1 in the control of contact-mediated recognition and repulsion, and reveal an important role for substrate interactions in promoting self-avoidance. da neurons, and class IV neurons in particular, have become a model for studies of dendritic self-avoidance and tiling mechanisms. Separating two causes of crossing in these cells should enable the identification of key molecules that regulate repulsive signaling between dendrites, as well as mechanisms that establish relationships between dendrites and other surrounding cell types that impact dendrite development and, perhaps also, function.

Materials and Methods

Fly stocks

Alleles used were mys¹ (Bloomington Stock Center), and mys^XG43 (linked to markers y, w, f), mew^M6 if^k27e (linked to markers y, f), and if^k27e (linked to marker f) on FRT19A, and rhea^tendrils13-8 on FRT2A (all provided by Dr. M. Krasnow, Stanford University) (Levi et al., 2005), and Dscam1²³ on FRT42D (Matthews et al., 2007). GFP protein trap lines were provided by Drs. L. Cooley (Princeton University) and B. Ohlstein (Columbia University). RNAi lines were obtained from the Vienna RNAi Collection (Dietzl et al., 2007). UAS-αPS2 (if), UAS-βPS (mys) flies were provided by Dr. K. Broadie (Vanderbilt University). 221-Gal4, ppk-Gal4, and clh201-Gal4 lines have been described previously (Grueber et al., 2003; Grueber et al., 2007; Hughes and Thomas, 2007).

Mosaic analysis

MARCM experiments were performed as described (Grueber et al., 2002; Lee and Luo, 1999) by crossing FRT lines to either hsFLP, C155-Gal4, UAS-mCD8::GFP; FRT2A tubPGal80 or hsFLP, tubPGal80, FRT19A; 109(2)80-Gal4, UAS-mCD8::GFP. For time-lapse analysis of MARCM clones, we examined mid-stage second instar larvae for the presence of dorsal cluster clones. Selected animals were imaged live under halocarbon oil (Sigma, St. Louis) and a coverslip, recovered to yeasted grape plates, raised to late third instar at 25°C, then dissected and labeled with anti-HRP, anti-GFP, and anti-Coracle.

Immunohistochemistry

Larvae were processed for immunohistochemistry largely as described (Grueber et al., 2002). Antibodies and dilutions used were CF.6G11 (anti-βPS, 1:10; developed by D. Brower), DK.1A4 (anti-αPS1, 1:10; developed by D. Brower), CF.2C7 (anti-αPS2, 1:10; developed by D. Brower), c556.9 and c615.16 (anti-Coracle, 1:20; developed by R. Fehon), 4F3 (anti-discs large, 1:10; developed by C. Goodman). These antibodies were obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the NICHD and maintained by the University of Iowa, Department of Biology. Other primary antibodies were chicken anti-GFP (Abcam; 1:1,000) and goat anti-HRP (Sigma; 1:200). Species-specific fluorophore-conjugated secondary antibodies (Jackson Immunoresearch) were used at 1:200 in PBS with 0.3% Triton X-100 (PBS-TX). For labeling of αPS1 and αPS2, dissected animals were incubated with primary antibody (1:10 in PBS) for 30 minutes at room temperature, rinsed twice for 2 min each, then incubated for 30 minutes with fluorophore-conjugated donkey anti-mouse. Animals were rinsed again in PBS (2 × 2 minutes) and then fixed for 20 minutes in 4% paraformaldehyde. Animals were then rinsed in PBS-TX and labeled successively with goat anti-HRP overnight and donkey anti-goat for 1 hour at room temperature. For anti-HRP labeling without detergent, goat anti-HRP at 1:200 in PBS was added for 1 hour at room temperature, the animals were rinsed for 30 minutes in PBS, 30 minutes in PBS-TX, then incubated with other primary antibodies and secondary antibodies in PBS-TX each overnight at 4°C.

Quantitative analysis

Arbors were traced in Neurolucida (Microbrightfield, Natick, MA) and analyzed in Neurolucida Explorer. For time-lapse analysis, neurons were quantified using the Simple Neurite Tracer plugin for FIJI. Arbors were traced as stacks (class I) or confocal projections (class IV). Linescan analysis was performed using Metamorph software (Molecular Devices, Downingtown, PA). For determination of HRP immunoreactivity in relation to Coracle labeling in class IV neurons, regions of arbors were categorized as either “high Coracle” (n=24 regions) or “low Coracle” (n=15 regions) in confocal projections. Cumulative average fluorescence intensities were 137 arbitrary units (A.U.) for high Coracle regions and 66 for low Coracle regions. Regions for linescan analysis were selected as informative if no Coracle-rich epidermal cell membranes intersected the linescan and no other dendrites crossed the linescan. For quantification of HRP and Coracle labeling intensity in mys mutant class I clones, cumulative average fluorescence intensities were 163 for high Coracle regions and 82 for low Coracle regions. For quantification of Coracle immunofluorescence intensities in crossing dendrites in Figure 7G, linescans were performed up to the point at which dendrites crossed. Statistical analysis was performed using R (R Development Core Team). Normality of datasets was assessed using a Shapiro-Wilk test. All p values are indicated as: *=p < 0.05, ** = p < 0.01, and *** = p < 0.001.

Electron microscopy

Mature third instar larvae were dissected in PBS and fixed immediately with 3% glutaraldehyde in 0.1M phosphate buffer (PB). Specimens were fixed for a total of 20 minutes, with 60 seconds of the fixation time in a Pelco 3451 Microwave System. Fixed tissue was washed 3 × 20 min in 0.1M PB, post-fixed with 1% osmium tetroxide in 0.1M PB in a microwave for 2 × 40 seconds (each 40 second exposure in fresh osmium), then washed 3 × 10 minutes in 0.1M PB. Tissue was dehydrated in the microwave in ethanol grades of 50%, 70%, 95% (all 1 × 40 seconds), and 100% (2 × 40 seconds). Dehydrated tissue was infiltrated in epon (Fullam Epox 812) and ethanol (1:1) for 15 minutes in the microwave, then in 100% epon resin (2 × 15 minutes each with fresh epon) in the microwave. Specimens were then mounted between 2 plastic slides with epon and polymerized overnight at 60°C. Areas of interest were identified in the epon wafer, placed either flat or perpendicular to the bottom of the tip of a Dykstra flat embedding mold, and polymerized in epon for 18–24 hrs at 60°C. The block was trimmed to include the area of interest and 10 µm serial sections were cut using a diamond Histo-knife with an ultramicrotome. Relevant regions were selected for thin sectioning and remounted on blank epon blocks using a small amount of fresh epon and allowed to polymerize overnight. Thin sections were collected on formvar-coated slot grids and stained with uranyl acetate and lead citrate. Grids were viewed using a JEOL 1200EX electron microscope and photographed using a digital camera. For Coracle labeling prior to electron microscopy, animals were fixed in 2.5% paraformaldehyde/0.5% glutaraldehyde in phosphate buffer, and primary antibody labeling was performed with 1:10 anti-Coracle in 0.1% PBS-TX. We used peroxidase conjugated goat anti-mouse at 1:200 in 0.1% PBS-TX, followed by detection using 1:20 diaminobenzidine in 0.1% PBS-TX with NiCl2 and 3 µl of a 3% hydrogen peroxide solution. The reaction was terminated by several rinses in PBS. Preparations were then mounted as above and photographed on a Zeiss A1 microscope fitted with a Zeiss digital camera and software prior to sectioning for TEM.