Cytonemes are required for the establishment of a normal Hedgehog morphogen gradient in Drosophila epithelia.

Hedgehog (Hh) signalling is important in development, stem cell biology and disease. In a variety of tissues, Hh acts as a morphogen to regulate growth and cell fate specification. Several hypotheses have been proposed to explain morphogen movement, one of which is transport along filopodia-like protrusions called cytonemes. Here, we analyse the mechanism underlying Hh movement in the wing disc and the abdominal epidermis of Drosophila melanogaster. We show that, in both epithelia, cells generate cytonemes in regions of Hh signalling. These protrusions are actin-based and span several cell diameters. Various Hh signalling components localize to cytonemes, as well as to punctate structures that move along cytonemes and are probably exovesicles. Using in vivo imaging, we show that cytonemes are dynamic structures and that Hh gradient establishment correlates with cytoneme formation in space and time. Indeed, mutant conditions that affect cytoneme formation reduce both cytoneme length and Hh gradient length. Our results suggest that cytoneme-mediated Hh transport is the mechanistic basis for Hh gradient formation.

INTRODUCTION

Morphogen signalling is crucial for developmental patterning. Morphogens are secreted molecules that act at a distance and in a concentration-dependent manner, controlling the differential activation of target genes according to the distance from their source[1]. To generate this activity gradient, morphogen production, transport and reception must be tightly controlled.

The mechanism of morphogen movement from producing to receiving cells has been the subject of intense study[2]. Several models have been proposed, including restricted diffusion[3], planar transcytosis[4], movement by exosomes[5] or argosomes[6] and lipoprotein particles[7], and transport by cellular extensions called cytonemes that orient towards the morphogen source. Cytonemes were first identified in the Drosophila wing disc, where they have been postulated to connect sending and receiving cells, enabling transfer of surface-associated cargoes[8,9]. However, the involvement of cytonemes in morphogen transport is still mysterious and the mechanisms underlying cytoneme function are still elusive.

To study the role of cytonemes in morphogen gradient formation, we investigated Hh signalling in two experimental paradigms in Drosophila, the wing disc and the abdominal epidermis. Hh acts as a morphogen, specifying cell identities in initially homogenous fields of cells[2,10]. Hh is post-translationally modified by the addition of cholesterol[11] and palmitic acid[12], which confer a highly hydrophobic character on Hh. These modifications suggest that association with cell membranes or lipoprotein particles is involved in Hh transport[11].

In the wing disc, Hh patterns the central region of the wing[13,14]. The wing is divided into anterior (A) and posterior (P) compartment[15]. P compartment cells produce Hh, which moves across the compartment border, decreasing in concentration as it spreads into the A compartment[16,17]. Dependent on this signal, the Hh receptor Patched (Ptc) is upregulated in a graded manner in the A compartment[18,19]. Ptc can thus serve as readout of Hh signalling.

In the abdominal epidermis, Hh also acts as a morphogen and is expressed in the P compartment[20,21]. Since the abdomen comprises multiple segments, Hh influences A compartment cells both anteriorly and posteriorly of its source[20]. In the larval epithelial cells (LECs), Ptc is expressed in A compartment cells anterior and posterior to each P compartment[20]. In the adult histoblasts, Ptc is expressed in a graded manner anterior to and in a narrow stripe posterior to each P compartment[20-22]. This pattern of Ptc expression is established during adult tissue formation. Ptc expression starts in the anterior dorsal histoblast (ADH) nest[23] before it fuses with the posterior dorsal histoblast (PDH) nest at the beginning of the process[21] (initially, both nests are separated by a row of LECs[23]). Histoblasts then spread and replace the LECs[23]. Only after histoblasts have met at the segment (P/A) border, Ptc expression begins in the histoblast stripe posterior to the P compartment[22].

To understand how Hh is released and transported, analysis of the Hh co-receptor Interference hedgehog (Ihog)[24] has proven useful. Drosophila Ihog interacts directly with the glypican Dally-like (Dlp), Hh and Ptc in the A compartment[25-27]. It is necessary for membrane localisation of Ptc[24] and required to maintain normal extracellular Hh levels[26,28]. We have recently described that fluorescently tagged forms of Ihog localise to the plasma membrane and label cellular processes in the baso-lateral part of wing discs[28]. We refer to these processes as ‘cytonemes’, as they resemble the previously described processes implicated in morphogen transport[8].

Here, we characterise these cytonemes in detail and study their role in Hh gradient formation. We show that, in the wing disc and in the abdominal epidermis, cytonemes are present in the Hh morphogenetic field. Hh and several Hh signalling components localise to these actin-based protrusions. Interestingly, we observe punctate structures, probably exovesicles, some of which are associated with and move along cytonemes. Using in vivo imaging, we show that cytonemes extend and retract dynamically and that Hh gradient establishment correlates with cytoneme formation in space and time. Furthermore, we found a strong correlation between maximum cytoneme length and Hh gradient length. These results strongly suggest that cytonemes play a role in Hh gradient establishment. Overall, our data support a model of Hh transport in which cytonemes act as routes for Hh movement in the extracellular matrix (ECM).

RESULTS

Wing disc cells generate basal actin-based cytonemes

To analyse wing disc cytonemes[28] in detail, we expressed fluorescently tagged Ihog in either A or P compartment cells. We found that both cell types displayed these processes basally (Fig. 1a-e). P compartment cells extended a dense array of shorter cytonemes (~27 μm long) and individual longer cytonemes (up to 70 μm long) perpendicular to the A/P border into the neighbouring compartment. Cytonemes of A compartment cells were about 40-50% shorter than P compartment cytonemes. The processes were mostly linear (0.1-0.2 μm diameter), but some were branched, especially towards their distal end. The cytonemes covered most of the Hh gradient in the A compartment (7-12 cell diameters). We could also detect lateral processes, but these were considerably shorter (1-2 cell diameters).

Figure 1

Wing disc cells generate actin-based basal cytonemes.

Fluorescently labelled Ihog marks cytonemes in both the A and the P compartment. Bars, 20 μm. White dotted lines indicate A/P boundaries. (a,b) hh.Gal4>UAS.Ihog-RFP discs, co-labelled with α-Hh antibody. (a) Overview of a basal z-section showing P compartment cytonemes that point anteriorly into the A compartment. (b) Detail of the region highlighted in (a). (c,d) ptc.Gal4>UAS.Ihog-RFP discs, co-labelled with α-Hh antibody. (c) Overview of a basal z-section showing A compartment cytonemes that point posteriorly into the P compartment. Note that protrusions also point anteriorly. (d) Detail of the region highlighted in (c). Like P compartment cells (b), A compartment cells show a dense array of shorter protrusions reaching into the neighbouring compartment as well as individual longer protrusions. (e) Cytonemes are basal structures. Lateral view of the P compartment cells shown in (b). Red and green bars indicate apical and basal sections, respectively. (f) Ihog-CFP labels punctate structures that not only decorate basal cytonemes, but are also found basolaterally, without being associated to cytonemes. (g,h) Immunoelectron microscopy of the Ihog protein in hh.Gal4/UAS.Ihog-CFP wing disc cryosections. Immunogold labelling was performed with an anti-GFP antibody followed by protein A conjugated to 15-nm gold particles (n=9 cryosections). (g) Ihog labelling at the basolateral membrane of a thin cytoplasmic protrusion (depicted in brown). (h) Ihog staining at basolateral membranes and exovesicles. Nuclei (green), cytoplasm (brown), extracellular space (blue) and exovesicles (red) are depicted by colours. (h’) Detail of the area delimited in (h). Arrowheads indicate gold particles on membranes and exovesicles; N, nucleus; LD, lipid droplet; Bars, 200 nm (g,h’), 1 μm (h). In all figures, anterior is to the left.

Fluorescently tagged Ihog also labelled punctate structures, some of which decorated cytonemes (Fig. 1f). Other puncta were not associated with cytonemes and were located laterally in a region apical to the basal cytonemes (Fig. 1f). To explore whether these puncta could be exovesicles, we performed immunoelectron microscopy of wing discs expressing Ihog-CFP in P compartment cells. Immunogold labelling with α-GFP antibodies detected Ihog at baso-lateral plasma membranes of P compartment cells (Fig. 1g,h), including cytoplasmic protrusions (Fig. 1g), and in clusters of exovesicles (50 to 200 nm) present in discrete baso-lateral extracellular spaces (Fig. 1h).

Besides fluorescently labelled Ihog, cytoskeletal markers such as Moe-Cherry (Fig. 2a) and GMA (an actin-binding fragment of Moesin that labels actin filaments[29]; Fig. 2b) labelled cytonemes. This corroborates the notion that cytonemes are actin-based structures as previously reported[8]. Moreover, various membrane markers labelled cytonemes. Fluorescently labelled Flotillin-2 (Flo2/Reggie-1), a major component of membrane microdomains[30], as well as CD4-Tomato[31] and glycosylphosphatidyl-inositol-YFP (GPI-YFP)[6] localised to cytonemes (Fig. 2c,d,e). Similar to Ihog, these markers also labelled puncta along cytonemes (Fig. 2d,e).

Figure 2

Cytoskeletal and membrane markers label cytonemes in the wing disc.

Cytonemes are labelled by actin cytoskeletal markers Moe-Cherry (a) and GMA (b) as well as by membrane markers Flo2-YFP (c) CD4-Tomato (d) and GPI-YFP (co-expressed with Ihog-CFP) (e), all driven by hh.Gal4. Note that GPI-YFP and Ihog-CFP colocalise. (f,g) Expression of Ihog stabilises cytonemes. Compared to the expression of cytoskeletal markers alone (a,b), co-expression of IHog-CFP and Moe-Cherry (f) as well as of Ihog-RFP and Act5c-GFP (g) increases visibility and number of cytonemes. Note that Moe-Cherry and Actin5c-GFP only label cytonemes, but not the dotted structures seen with Ihog. Bars, 20 μm.

Interestingly, cytonemes were more easily visible when Ihog was co-expressed (Fig. 2e-g) than when cytoskeletal and membrane markers were expressed alone (Fig. 2a-d). It is hence possible that Ihog stabilises cytonemes, maybe by recruiting some actin-stabilising factor. That cytonemes could be observed without Ihog overexpression using a variety of markers showed that cytonemes were not an Ihog overexpression artefact.

Hh pathway components label cytonemes

If cytonemes were involved in morphogen signalling, they would be expected to carry signalling pathway components[9]. Indeed, we have previously observed that Ihog overexpression increases Hh levels along cytonemes[28,32]. We thus examined whether we could detect Hh pathway components other than Ihog along cytonemes. Using α-Hh antibody or Hh-GFP, we found Hh in puncta that appeared to be organised in threads running perpendicular to the A/P border in the basal part of the wing disc epithelium (Fig. 3a,b). These structures became more evident after we ‘froze endocytosis’ by expressing a dominant negative form of Dynamin (ShiK44A) or by using a temperature-sensitive dynamin mutant (shi) before staining for Hh or Ptc (Fig. 3c-e). Freezing endocytosis leads to an accumulation of Hh or Ptc prior to internalisation, which improves their visualisation[33].

Figure 3

Components of the Hh pathway are associated to cytonemes in wing discs.

Endogenous Hh (a) and ectopic Hh-GFP driven by hh.Gal4 (b) can be observed in puncta that appear to be organised in threads and reach into the A compartment. (c) ‘Freezing endocytosis’ in shi mutants increases the visibility of Hh-GFP in cytonemes. shi wing disc shown after one hour at restrictive temperature. (d,e) ‘Freezing endocytosis’ in the receiving cells by expressing ptc.Gal4>UAS.shi increases visibility of endogenous Ptc (d) and Hh (e) in puncta. These puncta appear to be organised in threads and reach into the P compartment. Wing disc shown after 14 hours at restrictive temperature. (f,g) Both Dlp-GFP (f) and Disp-YFP (g), expressed in clones of cells, label cytonemes (arrows). Dotted lines indicate A/P boundaries. Bars, 20 μm.

In addition, Dispatched (Disp), which is required for Hh release[34], and Dlp, which is required for both Hh release[28] and reception[35], localised to basal protrusions (Fig. 3f,g). These findings confirm that Hh pathway components are associated with cytonemes.

Cytonemes are present at sites of Hh signalling in the abdominal epidermis

To study cytonemes in vivo, we turned to the abdominal epidermis for live imaging. Using fluorescently labelled Ihog and a GFP-trap in the ptc-promotor (ptc-promotor-trap::GFP) allowed us to label cytonemes and simultaneously monitor Ptc expression (Fig. 4a; Supplementary Fig. S1a). We found that both histoblasts and LECs of P compartment cells generated basal cytonemes, resembling the protrusions in wing discs (Fig. 4b-f). Importantly, the presence and length of these protrusions spatially correlated with Ptc expression, both at the compartment (A/P) and segment (P/A) borders in larval and adult epithelia (Supplementary Fig. S1b):

Figure 4

Hh signalling and cytonemes in the abdominal epidermis.

(a) Hh signalling as reported by Ptc-promotor-trap::GFP. Early stage (left): ADH and PDH nests still separated by a row of LECs (red arrow). In the ADH nest, the reporter is switched on faintly in a few cells (white arrows). ‘Intermediate’ stage (middle): The histoblast mass (hatched line) moves in dorsal direction (arrow) and replaces LECs. A row of A compartment LECs posterior to the P compartment expresses Ptc-promotor-trap::GFP (white asterisks). Furthermore, A compartment LECs anterior to the P compartment express the reporter (cyan asterisks). In histoblasts, Ptc-promotor-trap::GFP shows graded expression anterior to the A/P border. Late stage (right): adult epithelium almost formed. Graded reporter expression visible anterior to the A/P border in histoblasts. Furthermore, reporter expression starts in histoblast regions where neighbouring segments touch (grey arrows). LECs still present at segment boundaries also express GFP (asterisks). Cyan and yellow lines indicate A and P compartments, respectively. Red line indicates A/P border. See Supplementary Fig. S1 for scheme. (b) P compartment cells display cytonemes in anterior (white arrows) but not posterior (yellow arrows) direction into the A compartments. hh.Gal4>UAS.Ihog-CFP labels cells at end of abdominal closure. (c) P compartment cytonemes are basal and reach far into the A compartment, covering the Ptc-promotor-trap::GFP gradient. Note dense array of shorter protrusions and individual longer protrusions (white arrows). hh.Gal4>UAS.Ihog-RFP labels P compartment cells, and Ptc-promotor-trap::GFP labels A compartment cells responding to Hh. Small box shows lateral view. (d) Cytonemes of A compartment cells (white arrows) are rare and shorter than those of P compartment cells. ptc.Gal4>UAS.Ihog-RFP labels cells. (e) P compartment LECs display cytonemes mainly in anterior but also posterior direction (white arrows) during their posterior migration phase. hh.Gal4>UAS.Ihog-CFP labels cells. See Supplementary Video S4. (f) When morphogenetic process begins (around 15 h APF), P compartment LECs generate cytonemes (arrow) that ‘surround’ LECs of first row of neighbouring segment (asterisks) – these A compartment LECs also begin to express ptc-promotor-trap::GFP (Supplementary Video S2). At the tip of the protrusions, vesicles emanate (arrowhead). hh.Gal4>UAS.Ihog-CFP labels LECs. See Supplementary Video S5. White hatched line indicates A/P border in (c,d). Bars, 10 μm.

Histoblasts

(1) At the compartment border: P compartment histoblasts generated protrusions that reached anteriorly into the A compartment, covering most of the Ptc-promotor-trap::GFP gradient (Fig. 4b,c; Supplementary Videos S1,S2). We observed both a dense array of shorter protrusions (about 17 μm, 4 cell diameters) and individual longer protrusions (about 40 μm, ~9 cell diameters). These protrusions could also be observed by labelling the actin cytoskeleton with Utrophin-GFP[36] (Supplementary Fig. S1c,d). In contrast to P compartment histoblasts, A compartment histoblasts rarely sent shorter cytonemes posteriorly into the P compartment (Fig. 4d; Supplementary Video S3). This is a prominent difference to wing discs, where both A and P compartment cells produced cytonemes extensively (Fig. 1b,d). (2) At the segment border: P compartment histoblasts very rarely sent short protrusions posteriorly into the neighbouring segment (Fig. 4b). This correlated with the short-range signalling event occurring at this boundary – here, only a narrow stripe of histoblasts switched on ptc expression[22].

LECs

(1) At the compartment border: P compartment LECs sent protrusions anteriorly into the A compartment (Fig. 4e; Supplementary Video S4), where also Ptc expression could be observed (Fig. 4a). (2) At the segment border: P compartment LECs sent thick bundles of cytonemes posteriorly, which ‘surrounded’ the first row of A compartment LECs of the neighbouring segment (Fig. 4f; Supplementary Video S5). This row of cells also expressed Ptc-promotor-trap::GFP (Fig. 4a).

Our observation of threads of Hh puncta running perpendicular to the compartment border (Fig. 3a,b) suggested that cytonemes might be oriented along the anterior-posterior (a-p) axis. To test this hypothesis, we generated FLP-out clones in abdomen, which expressed Ihog-RFP. These clones confirmed that P compartment cytonemes point anteriorly (Supplementary Fig. S1e,f).

Cytoneme formation and Hh signalling gradient establishment correlate over time

Using live imaging also allowed us to explore how Hh signalling and cytonemes behave over time. Before histoblast nest fusion, Ptc-promotor-trap::GFP expression could be observed in the ADH nest. Only after ADH and PDH nests had fused, histoblast cytonemes started to form (Fig. 5a-c; Supplementary Videos S2, S6). Simultaneously, Ptc expression in the ADH nest began to increase (Supplementary Video S6). Early Ptc-promotor-trap::GFP expression therefore seems to be independent of cytonemes. Eventually, the cytonemes grew and the Ptc gradient formed (Supplementary Videos S2,S6). Intriguingly, maximum cytoneme length correlated with Hh activity gradient length throughout development (Fig. 5d-f; Supplementary Video S6). Furthermore, the shape of the Hh activity gradient and the density of cytonemes along the a-p axis correlated: The short array coincided with the brightest section of the gradient; then both curves declined similarly (Fig. 5d; Supplementary Video S7). This correlation was also observed in wing discs (Supplementary Fig. S2).

Figure 5

Cytonemes behave dynamically, and their development correlates with Hh signalling gradient establishment.

(a-c) Cytonemes develop over time. hh.Gal4>UAS.Ihog-RFP labels P compartment cells. (a) No cytonemes observed before ADH and PDH nests fuse. Yellow hatched line outlines PDH nest; asterisk indicates LECs separating ADH and PDH nests. (b) After nest fusion, basal cytonemes start to form (cyan arrow). (c) Four hours later, cytonemes fully developed - note dense array of shorter protrusions and prominent longer protrusions (white arrow). See Supplementary Video S2. (d-f) Cytoneme development correlates with gradient formation. hh.Gal4>UAS.Ihog-RFP labels P compartment cytonemes; Ptc-promotor-trap::GFP indicates Hh activity gradient. (d) Four frames from Supplementary Video S7. Top panels – Hh activity gradient shown by expression of the Ptc-promotor-trap::GFP. Bottom panels – cytonemes in a basal section of Ihog-RFP channel. Middle panels – profile plots of vertically averaged pixel intensities for panels above (green curve) and below (red curve). Blue line marks A/P border. 0 min: ADH and PDH nest fusion, no cytonemes present. 738 min: End of video, cytonemes present, and gradient established. Maximum cytoneme length (red line) correlates with gradient width (black line). The Hh activity gradient (green curve) and cytoneme density along the a-p axis (red curve) also correlate: the short array (cyan dotted line) coincides with the brightest gradient section, both curves decline similarly (pink arrow). This relation is already visible during gradient formation (two centre frames). (e) Gradient length correlates with maximum cytoneme length over time. (f) Kymograph illustrating establishment of the Ptc-promotor-trap::GFP gradient over time. Pixel intensities per time point for Supplementary Video S6 (top panel, 2.5 min intervals) were vertically averaged. The resulting horizontal lines were then plotted along the y-axis, illustrating change in fluorescence over time. Black arrow indicates time when ADH and PDH nests fuse. (g,h) P compartment cytonemes labelled with the actin cytoskeleton marker GMA driven by hh.Gal4. (g,g’) Apical (red) and basal (green) z-sections. Some cytonemes indicated with white arrows. (h) Close-up of individual cytonemes labelled with GMA. See Supplementary Video S9. White hatched lines indicate A/P border. Bars, 10 μm.

Also the formation of the thick cytoneme bundles of P compartment LECs (Fig. 4f) correlated with the onset of Ptc expression in the receiving cells (Supplementary Video S2). Overall, our findings show that cytonemes and Hh signalling are closely associated in space and time.

Cytonemes are dynamic structures

Next we analysed the behaviour of individual cytonemes. As in wing discs, Ihog-RFP/-CFP expression stabilised cytonemes also in abdomen – we could observe individual protrusions for up to 9.5 hours. To study cytoneme dynamics, we therefore used GMA instead of Ihog-RFP/CFP to visualise cytonemes. Without Ihog, histoblast cytonemes appeared thinner and shorter (up to 26 μm) (Fig. 5g,h; Supplementary Video S8) and had a lifetime of about 11 ± 2.6 min (n=6) (Supplementary Video S9). With a growth rate of approx. 5 μm/min, they grew faster than filopodia observed in other in vivo systems (3.3 μm/min[37]) and in vitro assays (2.5 μm/min[38]). Cytonemes were very dynamic; they grew to their final length and immediately shrank (Supplementary Fig. S3; Supplementary Video S10). We could also image cytonemes using the membrane markers CD4-Tomato[31] and Gap43-Venus[39] (Supplementary Videos S11,S12).

Interfering with cytoneme formation affects the Hh gradient

Our observations suggest that cytonemes might be important in Hh transport. Thus, a manipulation of cytoneme formation should affect the Hh gradient. To test this hypothesis, we manipulated cytonemes in wing discs using RNAi against various actin-binding proteins and analysed the effects on gradient formation. We chose the following proteins: (1) Capping protein (CP): CP restricts the accessibility of actin filaments, thus inhibiting addition or loss of actin monomers[40]. In Drosophila, mutations in either subunit (cpa or cpb) lead to accumulation of F-actin[41]. In accordance with this, we found an accumulation of actin and myosin in wing disc cells after Cpa-RNAi (Supplementary Fig. S4). Cell culture experiments with a CP inhibitor have furthermore shown that at high inhibitor concentrations, actin polymerisation is inhibited[42]. (2) SCAR: SCAR/WAVE is involved in the generation of lamellipodia[43] and filopodia[44,45]. Loss of SCAR function results in the loss of F-actin and protrusions within the baso-lateral domain of the Drosophila pupal notum without affecting cell morphology, size, polarity or endocytosis[46]. (3) Pico: Pico/Lamellipodin interacts with Ena/Vasp actin regulators and is involved in lamellipodia formation[47]. We found that RNAi knock-down of all these proteins in the P compartment affected maximum cytoneme length as well as Hh gradient length, both of which were significantly shorter compared to wild-type (Figs. 6a-d,g-j; 7a,b).

Figure 6

Interfering with cytoneme length affects the Hh activity gradient.

(a-f) Fluorescence images of wing discs, in which the cytonemes are labelled with hh.Gal4>UAS.Ihog-RFP (left panel; detail shown in the adjacent panel) and the Hh activity gradient is visualised using Ptc-promotor-trap::GFP (right panel). The merge of both channels is shown in the centre. In addition, RNAi- or UAS-constructs were co-expressed in the P compartment to interfere with cytoneme formation. (a) Control comprising Ihog-RFP only. (b) Cpa-RNAi and Ihog-RFP. (c) Scar-RNAi and Ihog-RFP. (d) Pico-RNAi and Ihog-RFP. (e) UAS.Flo2 and Ihog-CFP. (f) Flo2-RNAi and Ihog-YFP. In the RNAi wing discs (b-d,f), both the cytonemes and the gradient are shorter than in the control (a). In UAS.Flo2 wing discs (e), cytonemes and gradient are longer than in the control (a). (g-l) Quantification of gradient length in the wing discs shown in (a-f) using profile plots. The graphs depict the vertically averaged pixel intensities along the horizontal (a-p) axis of the detail of the Ptc-promotor-trap::GFP image shown in the top left corner. The red arrows indicate the length of the gradient, up to the point where the average pixel intensity drops below 5% of the maximum intensity. The green arrows show the average length of the two longest cytonemes in the analysed disc. These parameters have been chosen to determine cytoneme reach and gradient length in our fixed samples as accurately as possible (see Methods for further information). However, this quantification has its limitations, since the parameters have been chosen to some extent arbitrarily. Bars, 20 μm.

Figure 7

Gradient length correlates with cytoneme reach, both with (a-d) and without (e-g) Ihog-RFP overexpression.

(a-c) Experiments with Ihog-RFP overexpression. (a) Boxplot comparing cytoneme length between control discs and treatments affecting cytoneme formation. Boxes show median and interquartile range. Whiskers show minima/maxima, or 1.5 times interquartile range if outliers are present. Four RNAi treatments reduced cytoneme length significantly compared to UAS.Ihog-RFP (Kruskal-Wallis-Test, P<0.001 and pairwise Wilcoxon Rank Sum Test, P<0.01). Overexpression of Flo2 increased cytoneme length in some individuals, but with no overall significant difference from UAS.Ihog-RFP. (b) Boxplot comparing gradient length between control discs and treatments. Similar to cytoneme length, there were significant differences between UAS.Ihog-RFP and four RNAi treatments (Kruskal-Wallis-Test, P<0.001 and pairwise Wilcoxon Rank Sum Test, P<0.01). Longest gradients were obtained in Flo2 overexpression. (c) Scatter plot of gradient length against cytoneme length, showing strong correlation (Spearman’s rho= 0.89, n=44 wing discs; P<0.001). (d) Boxplot showing gradient length in control discs, RNAi treatments and UAS.Flo2 treatment when co-expressed with UAS.Ihog-RFP or expressed alone. Expression of UAS.Ihog-RFP shortens the gradient (Wilcoxon Rank Sum Test; P<0.001). Similar to Ihog overexpression experiments (b), there are significant differences between wild-type and treatments without Ihog overexpression (Kruskal-Wallis-Test; P<0.001 and Wilcoxon Rank Sum Test; P<0.05). Fig. 6 and Supplementary Fig. S5a-f show original data. (e-g) Cytoneme length and Hh activity gradient length also correlate without Ihog overexpression. hh.Gal4>UAS.CD4-Tomato labels cytonemes and Ptc-promotor-trap::GFP visualises Hh activity gradient. Cytonemes in wing disc (e) and abdomen (e’) shown. (f) Quantification of gradient length and cytoneme reach in wing discs using profile plots. Graph depicts vertically averaged pixel intensities of Ptc-promotor-trap::GFP image shown in the top left corner. Red arrow indicates gradient length, up to the point where average pixel intensity drops below 5% of maximum intensity. Green arrow shows average length of the two longest cytonemes in the disc analysed. (f’) Scatter plot showing correlation between gradient length and cytoneme length (Spearman’s rho= 0.78, n=14 wing discs; p<0.01). (g) As for Ihog-RFP overexpression (Fig. 5d; Supplementary Fig. S2), the shape of the Hh activity gradient (green profile plot) and density of cytonemes along the a-p axis (red profile plot) correlate: the short array (cyan line) coincides with the brightest gradient section, both curves decline similarly. Methods show details of statistical analysis (a-d,f). Bars, 20 μm.

In contrast, ectopic expression of Flo2/Reggie-1 in the P compartment extended the Hh gradient. Also cytonemes were longer compared to control experiments (Figs. 6e,k; 7a,b). This agrees with the observation that overexpression of Flo2/Reggie-1 induces filopodia-like protrusions in various cell lines[48]. It furthermore suggests that the mechanism by which Flo2/Reggie-1 stimulates Hh secretion and promotes its diffusion[49] might be cytoneme-related. Accordingly, knock-down of Flo2/Reggie-1 in P compartment cells produced the opposite phenotype; both cytonemes and the Hh gradient were significantly shorter than in wild-type (Figs. 6f,l; 7a,b).

Importantly, although the strength of the phenotypes varied from individual to individual, there was a strong positive correlation between cytoneme reach and gradient length for all individuals across all experiments (Fig. 7c). This strongly suggests that cytonemes play a role in Hh gradient establishment.

All above experiments were done with and without co-expression of fluorescently labelled Ihog. A comparison of gradients with and without Ihog-RFP/CFP showed that Ihog led to a shortening of the Ptc-promotor-trap::GFP gradient in both wing discs (Fig. 7d; Supplementary Fig. S5a-f) and abdomen (Supplementary Fig. S5g). This could be due to the above-mentioned cytoneme-stabilising effect of Ihog, which might impair cytoneme function. Alternatively, Ihog’s ability to sequester Hh[26,28] might reduce the amount of Hh being released from cytonemes overexpressing Ihog and therefore impair signalling[32]. Our findings, however, do not represent an artefact of Ihog-RFP overexpression, because also without Ihog, the differences in gradient length between the above experiments remained (Fig. 7d). This suggests that overexpression of Ihog only affects the range of Hh signalling, but not Hh signalling per se.

To further rule out that our findings represent an artefact of Ihog overexpression, we replaced Ihog-RFP with the membrane marker CD4-Tomato to label cytonemes (Fig. 7e). As with Ihog-RFP, the reach of CD4-Tomato-labelled cytonemes correlated with the length of the Hh activity gradient (Fig. 7f; Supplementary Video S13). The shape of the Hh activity gradient and the density of cytonemes along the a-p axis also correlated (Fig. 7g).

In all experiments, Hh levels were not altered by the RNAi knock-down (Supplementary Fig. S6), excluding the possibility that the effects on gradient length were due to reduced morphogen levels in the signal-producing cells.

In addition, RNAi against genes involved in exovesicle production/release had a significant effect on the length of the Hh activity gradient. A range of RNAi treatments were used, including the ESCRT complex members TSG101, Alix, HRS and VPS4[50,51] as well as proteins found in exosomes, such as AnxB11 and Rab11[50]. In all experiments, a shortening of the Hh gradient was observed (Supplementary Fig. S7).

Cytoneme-mediated Hh transport

To further explore the role of cytonemes in Hh transport, we analysed the relationship of cytonemes and ECM components, such as heparan sulfate proteoglycans (HSPGs), which are known to mediate Hh transport. Toutvelu (Ttv) and Brother of ttv (Botv) are essential for the biosynthesis of HSPGs[52,53]. Hh cannot cross ttv−/− or ttv−/−, botv−/− mutant clones abutting the A/P compartment border and can therefore not signal anterior to the clone[52,53]. However, if cytonemes were involved in Hh transport, they would be able to deliver the signal across the clone. Thus, the ttv mutant phenotype and the cytoneme model were only compatible, if cytonemes could not cross mutant territory. We tested this hypothesis by generating ttv−/−, botv−/− mutant clones abutting the compartment border and labelling cytonemes with Ihog-YFP (Fig. 8a). We found that, in most cases, cytonemes did not cross mutant clones (Fig. 8b). As shown before[52,53], no Hh response could be detected anterior to these clones. However, if the clones were very narrow, cytonemes crossed the mutant territory (Fig. 8c) and, interestingly, we could detect a Hh response in the wild-type territory, into which the cytonemes reached.

Figure 8

Cytonemes rarely cross large HSPG mutant clones, however, they do cross narrow clones.

(a-c) ttv−/−, botv−/− double-mutant clones located at the A/P compartment border in a Hh.Gal4>UAS.Ihog-YFP wing disc are shown. Clones are outlined with a hatched line. (a) Merge of Ihog-YFP, α-Ptc and α-βGal, the absence of which labels mutant cells. Note that the clone has a very different shape in the apical and basal part of the disc. Ptc expression can be best observed in the sub-apical part and cytonemes in the basal part of the disc. A scheme shows the genotypes of the clonal and the surrounding cells as well as the areas of Hh and Ptc expression. (b) Two wing discs, in which cytonemes do not cross the clones. Note that the Hh signal, as assayed by Ptc expression, is not received anterior to the clones. Only a single row of Ptc expression within the clones can be detected, as described before[52] (cyan arrows). In the heterozygous territory, however, cytonemes and Ptc expression can be observed. Note that basolaterally, Ihog puncta can be observed only in regions, in which also cytonemes are present (red arrows). (c) Wing disc, in which cytonemes cross a narrow area of the clones (green arrow). Note that the Hh signal, as assayed by Ptc expression, is received anterior to the clones (yellow arrows). As in (b), basolateral Ihog puncta can only be observed in regions, in which also cytonemes are present (red arrows). Bars, 20 μm.

In addition, we observed that Ihog-labelled baso-lateral puncta were only present in regions where cytonemes were present as well, but not in or anterior to mutant clones (Fig. 8b,c). This suggests that vesicles might reach their target territory via cytonemes.

These results suggest that the inability of cytonemes to cross ttv−/−, botv−/− mutant clones could explain why Hh fails to signal anterior to these clones. The inability of cytonemes to cross wider clones could be due to a destabilisation of the ECM due to the lack of HSPGs, which might deprive cytonemes of structural support. Narrow clones can be crossed by cytonemes, suggesting that clones do not repel but rather destabilise them. These results not only give a possible mechanistic explanation for the behaviour of ttv−/−, botv−/− mutant clones, but also implicate cytonemes in Hh transport.

DISCUSSION

The localisation of several Hh signalling components at cytonemes[28,32] (this study) suggested a role of these structures in Hh signalling. Here, we characterise cytonemes in two Drosophila paradigms, the wing disc and the abdominal epidermis, and investigate their role in Hh gradient formation. We present evidence that cytonemes play an active role in gradient formation: (1) The establishment of the Hh signalling gradient correlates dynamically in space and time with cytoneme formation in vivo. (2) Experimental shortening and lengthening of cytonemes affects the gradient accordingly. (3) The analysis of ttv−/−, botv−/− mutant clones implicates cytonemes in Hh transport. Overall, our results support a model in which cytonemes of signal-producing cells are involved in long-range Hh transport (Supplementary Fig. S8).

In wing discs, however, both sending and receiving cells generate cytonemes raising the question which role the cytonemes of receiving cells play. Expression of Ihog in A compartment cells leads to a depletion of Hh from the P compartment cells close to the A/P border[32], which suggests that A compartment cytonemes might actively engage in Hh reception. Hence, cytonemes of both sending and receiving cells might contribute to Hh transport. Interestingly, A compartment cytonemes are rare in histoblasts, suggesting that cytonemes of receiving cells play a minor role in the abdomen.

We observed Ihog-RFP puncta that associated with and moved along cytonemes (Supplementary Video S14). Frequently, such puncta were released from cytonemes (Supplementary Videos S5,S15). We also observed puncta when labelling cytonemes with CD-4-Tomato (Supplementary Video S11). This suggests that cytonemes might transport exovesicles that act as a vehicle for Hh or are the structure where exovesicles are being released. Accordingly, the knock-down of genes involved in exosovesicle production/release has a significant effect on Hh gradient length, and Ihog can be detected in baso-lateral exovesicles at the ultrastructural level. However, the characterisation of these exovesicles as well as their implication in Hh gradient formation requires further analysis. A role of exosomes in morphogen gradient formation has recently been suggested. Active Wnt proteins are secreted in exosomes in cultured cells and in the wing disc[5]. In addition, vesicular release of SonicHh has been implicated in the determination of left-right asymmetry in vertebrates[54]. Very recently, particles containing SonicHh and CDO (the vertebrate homologue of Ihog) that travel along filopodia-like extensions have been described in the chicken limb bud[55].

The mechanisms by which cytonemes could transport morphogens to their targets must ensure specificity and accuracy. One possibility is that cytonemes established contact between sending and receiving cells[9,56]. Alternatively, cytonemes could act as a structure of morphogen release and uptake without cell-cell contacts involved. Our in vivo imaging showed that cytonemes are dynamic structures. Cytonemes might grow towards a receiving cell and then retract after a signalling event has taken place, or their dynamics could be determined intrinsically by the stability of their cytoskeleton. Moreover, not just cytoneme length but also their number could shape the gradient, as its brightest section coincides with the dense array of shorter cytonemes. This cytoneme-based model challenges the previous diffusion-based models[57-59] (Supplementary Fig. S8).

Cytonemes have been described in a variety of signalling pathways. The Dpp receptor Thickveins is present in punctate structures moving along cytonemes[60]. Air sac precursors extend cytonemes towards FGF-expressing cells[61]. Tracheal cells were reported to have at least two types of cytonemes; one type that carries an FGF receptor, and another type that carries the Dpp receptor[9]. This suggests that cytonemes are ligand specific[9]. In the context of Notch signalling, filopodia mediate lateral inhibition between non-neighbouring cells of the pupal notum[56,62]. Interestingly, the dynamic behaviour of these processes is crucial for signalling[56]. Spitz/EGF is delivered through polarised actin protrusions to spatially bias the specification of a particular cell of the Drosophila leg[63]. In another example, short cytonemes mediate the delivery of a juxtacrine Hh signal to maintain germ line stem cells in the Drosophila ovary[64]. Here, we show that cytonemes also play a pivotal role in long-range Hh signalling in wing disc cells, histoblasts and LECs. Therefore, we believe that cytonemes are a general feature of signalling events of all epithelial cells.

METHODS

Fly mutants

A description of the mutations, insertions and transgenes is available at http://flybase.org. tub.Gal80, FLP122 (BDSC, Indiana, USA), shits1[65] ,ttv, botv[53].

Overexpression experiments

The following Gal4 drivers were used for ectopic expression using the Gal4/UAS system[66]: hh.Gal4, ap.Gal4, ptc.Gal4.

The pUAS-transgenes were: UAS.HhGFP[33], UAS.Dlp-GFP[67], UAS.Disp-YFP[28], UAS.Shi [68], UAS.Ihog[24], UAS.Ihog-YFP and UAS.Ihog-RFP[28]. UAS.Moe-Cherry[69], UAS.GPI-YFP[6], UAS.GMA[29], UAS.RedStinger, UAS.CD4-tdTom[31], UAS.actin5c-GFP[70], UAS.gap43-Venus[39], sqh.utrophin-GFP[36], Ptc-promotor-trap::GFP (CB02030; http://flytrap.med.yale.edu), UAS.SCAR-RNAi (BDSC 36121), UAS.Cpa-RNAi (VDRC v16731), UAS.Pico-RNAi (VDRC v16371), UAS.Flo2-RNAi (VDRC v31524, v31525), UAS.TSG101-RNAi (BDSC 38306), UAS.Alix-RNAi (BDSC 33417), UAS.VPS4-RNAi (VDRC v105977), UAS.HRS-RNAi (BDSC 33900), UAS.AnxB11-RNAi (VDRC v101313) and UAS.Rab11-RNAi (BDSC 27730). For generating UAS.Ihog-CFP the vector pENTR/D-TOPO–Ihog was introduced by recombination into the destination vectors pTWC (pUAS-CFP).

Experimental genotypes and clonal analysis

Transient expression of the UAS-constructs using Gal4 drivers and tub.Gal80 was done by keeping the flies at 18°C and inactivate the Gal80ts for 24 to 30 hours at the restrictive temperature (29°C).

Mutant clones were generated by FLP-mediated mitotic recombination. Larvae were incubated at 37°C for 1 hour at 24-48 hours after egg laying, or for 45 minutes at 48-72 hours AEL. Flp-out clones in the abdomen were induced for 1 h at 35.5°C. The genotypes of the flies were:

Fig. 3f,g: The transgene actin>CD2>Gal4[71] was used to generate ectopic FLP-out clones that expressed UAS.Disp-YFP[28] or UAS.Dlp-GFP[67].

Fig. 8: hs.FLP122; FRT42D, ttv

Supplementary Fig. S1e,f: y w hs.FLP122; UAS.RedStinger / If or CyO; UAS.Ihog-RFP / tub

Microscopy of imaginal discs

Imaginal discs were fixed in 4% paraformaldehyde (PFA) for 25 min at room temperature (RT) and either imaged directly or used for immunostaining. Exceptions were the wing discs used for the visualisation of cytonemes with the membrane marker UAS.CD4-tomato (Fig. 7e-g). These discs were fixed in 4% PFA on ice for 10 min. Cytonemes that did not overexpress Ihog-RFP/CFP were sensitive to fixation. In around 60% of fixed discs that express CD4-Tomato, we observed cytonemes. This might be due to the stabilising effect of Ihog overexpression on cytonemes. A laser scanning confocal microscope (Zeiss LSM510) was used for fluorescence imaging of the discs. ImageJ software (NIH, Bethesda) was used for image processing.

Immunostaining of imaginal discs

Immunostainings were performed according to standard protocols[18]. Antibodies were used at the following dilutions: rabbit polyclonal α-Hh[53] 1/800 and mouse monoclonal α-Ptc (Apa 1.3; Developmental Studies Hybridoma Bank) 1/30. All performed immunostainings were highly reproducible. n-numbers of analysed wing discs in each experiment were: Fig. 1a-f, n>200; Fig. 2a, n>50; Fig. 2b, n>25; Fig. 2c, n>200; Fig. 2e, n>25; Fig. 2f, n>50; Fig. 2g, n>25; Fig. 3a, n>200; Fig. 3b,c, n>50; Fig. 3d,e, n>200; Fig. 3f,g, n>20; Fig. 8, n>200.

Immunoelectron microscopy

Larvae were inverted in PBS and fixed in 2% (w/vol) PFA and 0.2% (w/vol) glutaraldehyde in 0.2 M phosphate buffer (PB, pH 7.4) for 2 h at RT and kept in 1% (w/vol) PFA in PB at 4°C. Subsequently, wing discs were dissected, embedded in 10% (w/vol) gelatine, and processed for cryosectioning. Discs were then cut orthogonal to the apical/basal axis on an EM FCS cryo-ultramicrotome (Leica Microsystems) at −120°C. For immunogold labelling, thawed 75-nm thick cryosections were incubated with rabbit α-GFP (1:500, A-6455; Invitrogen) followed by protein A conjugated to 15-nm gold particles (EM Laboratory, Utrecht University, The Netherlands). Sections were stained with a mix of 1.8% methylcellulose and 0,4% uranyl acetate.

In vivo imaging of the abdomen

Pupae were filmed through a window in the pupal case as described[72]. The analysis focused on the dorsal side (tergite) of abdominal segment A2. All imaged flies developed into pharate adults and many hatched. Z-stacks of around 40 μm with a step size of 2.5 or 3.0 μm were recorded every 150 or 180 s using a Leica SP5 confocal microscope at 23±2°C. All images and videos shown are projections of z-stacks unless stated otherwise. Figures and videos were made using Adobe Illustrator, Adobe Image Ready, Adobe Photoshop, ImageJ (NIH, Bethesda), Volocity (Improvision), and Quicktime Pro (Apple Inc.).

n-numbers of recorded pupae: Video S1, n=10; Video S2, n=10; Video S3, n=11; Video S4, n=3; Video S5, n=7; Videos S6,S7, n=4; Video S8, n=8; Video S9, n=4; Video S10, n=5; Videos S11,S13, n=8; Video S12, n=3; Video S14, n=8; Video S15, n=6.

n-numbers of videos of which frames were used for the Figures: Fig. 4a, n=7; Fig. 4b, n=6; Fig. 4c, n=10; Fig. 4d, n=11; Fig. 4e, n=3; Fig. 4f, n=7; Fig. 5a-c, n=10; Fig. 5d-f, n=4; Fig. 5g, n=8; Fig. 5h, n=4; Fig. 7e’g, n=8; Fig. S1c,d, n=3; Fig. S1e,f, n=4; Fig. S5g, n=3 (Ptc-trap::GFP) and n=5 (Ptc-trap::GFP; Ihog-RFP).

Quantification of cytoneme and gradient length

(1) Gradient length: An area of five cell diameters in height and 120 μm in length, covering the gradient and the A/P border in Ptc-promotor-trap::GFP images, was cropped out to calculate the ‘plot profile’ using ImageJ. This procedure plots the vertically averaged pixel intensities along the horizontal axis of the image. Using the ‘plot profile’ data, the length of the gradient was determined in Microsoft Excel. Gradient length has been defined as the distance from the A/P border up to the point where the gradient ‘dipped’ below 5% of the maximum fluorescence. We chose this 5% threshold, since at this pixel intensity the signal cannot be clearly distinguished from background noise. We used this 5%-criterion rather than the decay length, because we found that the gradients did not follow an exponential distribution, so that the decay length would not describe the gradient correctly.

(2) Cytoneme reach: To quantify the maximum length of cytonemes, which is the crucial parameter to correlate cytoneme length and gradient length, we measured the length of the two longest protrusions in the wing pouch, from the A/P border to the tip, using the ‘line tool’ in ImageJ. Then, the average length of the two longest cytonemes was used for the analysis. We used the two longest cytonemes to ensure that we did not use a single outlier for the analysis.

(3) The Kymograph in Fig. 5f was generated using the ‘reslice’ tool of ImageJ.

Animal models and statistical analysis

All used Drosophila melanogaster strains are described above. Third instar larvae and staged pupae were used for the experiments. The sex of the experimental animals was not determined.

All statistical analysis was carried out in R[73] based on a threshold significance level of 0.05. No statistical method was used to predetermine sample sizes. Instead, sample sizes were determined by the available material that we could process during each experiment. No samples have been excluded. The experiments were not randomised. The investigators were only blinded to allocation during the measurement of cytoneme length (Figs. 6,7). To determine the appropriate statistical tests, the data was tested for normal distribution and for equivalence of variance. The following tests, which are justified as appropriate, were conducted:

Tests for treatment effects on cytoneme length (Fig. 7a)

A Kruskal-Wallis test of cytoneme length against the six treatments (UAS.Ihog-RFP, UAS.Pico-RNAi + UAS.Ihog-RFP, UAS.Cpa-RNAi + UAS.Ihog-RFP, UAS.Scar-RNAi + UAS.Ihog-RFP, UAS.Flo2-RNAi + UAS.Ihog-RFP and UAS.Flo2 + UAS.Ihog-RFP) showed a significant effect of treatment (Chi2=28.84, df=5, P<0.001). Pairwise comparisons using Wilcoxon rank sum tests corrected for multiple comparisons showed significant differences of UAS.Ihog-RFP and UAS.Flo with each of the 4 RNAi treatments (P<0.01 in all cases).

Tests for treatment effects on gradient length (Fig. 7b)

A Kruskal-Wallis test of gradient length against the six treatments (UAS.Ihog-RFP, UAS.Pico-RNAi + UAS.Ihog-RFP, UAS.Cpa-RNAi +UAS.Ihog-RFP, UAS.ScarRNAi + UAS.Ihog-RFP, UAS.Flo2-RNAi +UAS.Ihog-RFP and UAS.Flo2 + UAS.Ihog-RFP) showed a significant effect of treatment (Chi2=30.03, df=5, P<0.001). Pairwise comparisons using Wilcoxon rank sum tests corrected for multiple comparisons showed significant differences of UAS.Ihog-RFP and UAS.Flo with each of the 4 RNAi treatments (P<0.01 in all cases).

Correlation of gradient length with cytoneme length (Fig. 7c)

Gradient length was significantly positively correlated with cytoneme length for the five treatments and the control (Spearman’s rho = 0.89, n = 44 wing discs, P<0.001).

Tests for gradient length with and without Ihog (Fig. 7d)

A Wilcoxon rank sum test of gradient length for the treatments with Ihog (n = 44 wing discs) versus without Ihog (n = 31 wing discs) showed that gradient length was significantly shorter under the Ihog treatments (W = 1074, P<0.001). For the six treatments without Ihog overexpression (wild-type, UAS.Pico-RNAi, UAS.Cpa-RNAi, UAS.ScarRNAi, UAS.Flo2-RNAi and UAS.Flo2), a Kruskal-Wallis test showed a significant effect of treatment on gradient length (Chi2=25.42, df=5, P<0.001). Wilcoxon rank sum tests corrected for multiple comparisons showed significant differences of wild-type and UAS.Flo2 with each of the 4 RNAi treatments (P<0.05 in all cases).

Correlation of gradient length with cytoneme length for CD4-Tomato (Fig. 7f’)

Using the CD4-Tomato expression to visualise cytonemes, there was a significant positive correlation between gradient length and cytoneme length (Spearman’s rho = 0.78, n=14 wing discs, P<0.01).

Tests for treatment effects on gradient length related to exosome production (Supplementary Fig. S7e)

Analysis of variance of ln (gradient length) against treatment for the seven treatments related to exosome production (wild-type, UAS.TSG101-RNAi, UAS.VPS4-RNAi, UAS.HRS-RNAi, UAS.AnxB11-RNAi, UAS.Alix-RNAi, UAS.Rab11-RNAi) showed a significant effect of treatment (F6,67=4.83, P<0.001). Tukey’s Honest Significant Difference tests showed that confidence intervals did not overlap for wild-type with the RNAi treatments (P<0.05 in all cases).