GPI-anchored FGF directs cytoneme-mediated bidirectional signaling to self-regulate tissue-specific dispersion

How signaling proteins generate a multitude of information to organize tissue patterns is critical to understanding morphogenesis. In Drosophila, FGF produced in wing-disc cells regulates the development of the disc-associated air-sac-primordium/ASP. Here, we show that FGF is Glycosylphosphatidylinositol-anchored to the producing cell surface and that this modification both inhibits free FGF secretion and activates target-specific bidirectional FGF-FGFR signaling through cytonemes. FGF-source and recipient ASP cells extend cytonemes that present FGF and FGFR on their surfaces and reciprocally recognize each other over distance by contacting through CAM-like FGF-FGFR binding. Contact-mediated FGF-FGFR binding induces bidirectional signaling, which, in turn, promotes ASP and source cells to polarize cytonemes toward each other and reinforce signaling contacts. Subsequent un-anchoring of FGFR-bound-FGF from the source cell membrane dissociates cytoneme contacts and delivers FGF target-specifically to ASP cytonemes for paracrine functions. Thus, GPI-anchored FGF organizes both source and recipient cells and self-regulates its cytoneme-mediated tissue-specific dispersion and signaling.

). We also examined the inter-cytoneme interactions during the development of 174 the ASP from the early-to-late L3 larval stages. Despite dynamic morphological changes 175 in the growing ASP and disc, the relative positions of the ASP, bnl-source, and the site 176 of inter-cytoneme interactions were maintained throughout the development 177 significantly reduced the long, polarized ASP tip cytonemes . In contrast, short, 204 randomly oriented ASP cytonemes were unaffected. Thus,  are required for the formation of the polarized Bnl-receiving ASP cytonemes. 206 207 We next removed ASP cytonemes by expressing dia-i under btl-Gal4 and recorded non-208 autonomous effects on mCherryCAAX-marked source cytonemes. The dia-i expression 209 had to be controlled with Gal80 ts to avoid lethality (see Methods). Tracheal dia-i 210 expression not only reduced ASP cytonemes but also non-autonomously reduced 211 source cytonemes (Fig.2l-n'). Similarly, tracheal expression of a dominant-negative form 212 of Btl (Btl-DN) was known to suppress ASP growth and cytoneme formation without 213 affecting the wing disc development 19 . When both source and Btl-DN-expressing 214 tracheal cells were marked, the complete loss of ASP and ASP cytonemes was found to 215 produce a corresponding loss of bnl-source cytonemes (Fig.2o,p). Thus, Btl-presenting 216 ASP cytonemes are required to produce source cytonemes that polarize toward the 217 ASP. Collectively, these results suggested that the source and recipient cytonemes 218 reciprocally guide each other to form signaling contacts. 219 220

CAM-like Btl-Bnl binding drives bidirectional contact matchmaking of cytonemes. 221
The above results also suggested that the inter-cytoneme interactions might recruit and 222 activate a bidirectional signaling mechanism, responses of which could induce ASP 223 cells to extend Btl-containing cytonemes toward the source and activate source cells to 224 extend Bnl-containing cytonemes toward the ASP. We hypothesized that such selective 225 matchmaking between source and ASP cytonemes could be mediated by the binding of 226 surface-displayed Btl and Bnl. In this model, Btl and Bnl are analogous to cell-227 recognition or cell-adhesion molecules (CAMs), physical interaction of which can 228 produce selective cell-cell adhesion and contact-mediated bidirectional signaling 23,32,33 . 229 The initiation of CAM-like interactions might not require Btl to activate canonical 230 transcriptional outputs 34 . An alternative possibility is that the Bnl-Btl binding activates 231 MAPK signaling and transcription of target genes in the ASP, and these gene products, 232 in turn, non-autonomously act on the wing disc bnl-source to induce a response.  Figure 2b) and can heterodimerize with WT Btl, but cannot activate 236 nuclear MAPK signaling due to the lack of its intracellular kinase domain 19,21 . As 237 observed before 13,19 , while most wing discs with Btl-DN-expressing trachea (TC) 238 completely suppressed ASP/ASP cytonemes, occasionally a few produced partially 239 suppressed nascent ASP with variable numbers of ASP cytonemes orienting toward the 240 source. Strikingly, in each of these discs, the appearance of polarized ASP cytonemes 241 correlated with the concomitant appearance of similar numbers of polarized source 242 cytonemes, forming direct cytoneme:cytoneme contacts (Fig.3a-c). These results were 243 consistent with the idea that source and ASP cells reciprocally promote each other's 244 cytoneme-forming polarity by contacting each other via cytonemes and that the Btl-245 DN::Bnl binding is likely to induce the reciprocal cytoneme-mediated matchmaking. 246 247 Indeed, non-permeabilized Bnl ex staining showed that Bnl ex was selectively enriched at 248 these inter-cytoneme contact sites (Fig.3d in S2 cells, GFP and mCherry tags were separated, and, importantly, while the 328 Bnl:GFP 3 portion was localized on the cell surface (detected with GFP ex 329 immunostaining), the C-terminal mCherry remained intracellular ( Fig.4b-b"'). The C-330 terminal mCherry tag did not alter the predicted topology and physico-chemical 331 properties of Bnl (see Supplementary information). In fact, when Bnl:GFP 3 Cherry c was 332 expressed in the wing disc source under bnl-Gal4, the mCherry-tag was retained in 333 source cells and the Bnl:GFP 3 portion was efficiently delivered to the ASP (Fig.4c).  Figure 5b). This was consistent with previous reports of 420 ER retention of the uncleaved pro-GPI-APs, in contrast to the normal trafficking of the 421 same protein with deleted SS 45,46 . These results indicated that Bnl is cleaved at its C 422 terminus and added with a GPI moiety, which both facilitated Bnl externalization and 423 inhibited its free secretion.

GPI-anchored Bnl acts as a CAM to produce target-specific cytoneme contacts. 426
To test if GPI anchoring is required for Bnl's CAM-like activity, we employed the cell  Table 1). 480 The increased stability of inter-cytoneme adhesion might account for the higher intensity and recipient cytonemes, we produced randomly-localized mCherryCAAX-marked wing 486 disc clones that expressed Bnl:GFPC-TM (Fig.7f-j). Clones in the wing disc pouch that 487 occurred far away from the ASP were unable to establish contact with the ASP. These 488 clones had only short, randomly oriented signal-containing cytonemes (Fig.7f,g,h). In 489 contrast, ASP-proximal clones extended long polarized cytonemes and establish 490 contacts with the ASP (Fig.7f,i,j). These results were consistent with the contact-491 dependent activation of a retrograde signaling response in the Bnl-source that 492 reinforced the ASP-specific source cytoneme polarity. ASP cells also can extend 493 cytonemes toward ectopic Bnl-expressing clones 19 and the source-specific polarity can 494 be reinforced by Bnl signaling feedbacks 9 . Moreover, when we produced randomly-495 localized Btl:GFP-expressing wing disc clones, these clones and the mCherryCAAX-496 marked Bnl-source were found to extend polarized cytonemes toward each other 497 ( Fig.7k-m'). These results showed that the cytoneme-mediated Bnl:Btl signaling is 498 bidirectional and that this bidirectional signaling is required for the reciprocal guidance 499 of source and recipient cytonemes.  To investigate if GPI-anchoring of Bnl facilitates its target-specific release, we compared 516 the spatial distribution of GPI-modified (Bnl:GFP) and non-GPI-modified Bnl:GFP 517 constructs expressed from the mCherryCAAX-marked wing disc Bnl source. As 518 observed before 9,36 , despite overexpression in the disc bnl-source, Bnl:GFP puncta 519 were exclusively transferred from the disc source to the ASP (Fig.8a; Supplementary  520 Movie 12). In contrast, two C variants (Bnl:GFPC and Bnl:GFPC 168 ) showed 521 dispersion in the non-specific disc areas surrounding the source (Fig.8b, and only their truncated Bnl:GFP 3 portions were transferred to the ASP (Fig.8f,g,j,m). 536 However, when Furin inhibitors were added to the culture media, uncleaved signals 537 (HA-probed) were received by the ASP 36 . Therefore, the fraction of the HA-probed 538 uncleaved signal relative to the total levels of the GFP-probed signal (pre-existing 539 Bnl:GFP 3 + HA 1 Bnl:GFP 3 ) accumulated in the ASP during a Furin-inhibited period 540 provided a semi-quantitative estimate of the rate of signal uptake in the ASP (Fig.8f'). 541

542
As observed before 36 , the levels of HA 1 Bnl:GFP 3 (control) uptake in the ASP gradually 543 increased with increasing duration of the culture. In comparison, the levels of 544 HA 1 Bnl:GFP 3 C-TM and HA 1 Bnl:GFP 3 C in the ASP did not change dramatically, 545 indicating a slow rate of ASP-specific transfer of these variants (Fig.8g-p). Notably, even 546 after a 5 h of culture, HA 1 Bnl:GFP 3 dispersed exclusively to the ASP (Fig.8q). In 547 contrast, within a 5 h of incubation period, HA 1 Bnl:GFP 3 C was randomly localized in 548 the source-surrounding disc areas, but was barely received by the ASP from the same 549 disc (Fig.8j-l,q-s). These results suggested that GPI anchoring is required to both inhibit 550 free Bnl secretion/dispersion and activate its directed contact-dependent release. Bnl:GFP-receiving cells also induced dpERK (Fig.9a,a',d; Supplementary Figure 7g). In 562 comparison, Bnl:GFPC was received by many cells surrounding its clonal source, but 563 only a few randomly located Bnl:GFPC-receiving cells induced dpERK (Fig.9b, Bnl was also known to chemoattract tracheal migration toward its source 47,48 . To further 577 assess the morphogenetic potency of Bnl variants, we examined their ability to 578 chemoattract tracheal branches to an ectopic expressing source, such as larval salivary 579 gland, a non-essential, trachea-free organ, which normally does not express bnl 36,48 . 580 bnl-Gal4 was reported to be non-specifically expressed in the salivary glands, and bnl-581 Gal4-driven Bnl expression in the salivary glands induced tracheal invasion into this 582 trachea-free organ 36 . Therefore, we expressed comparable levels of Bnl:GFP, contacts that the Btl-Bnl binding helps to form also bring Btl and Bnl molecules together 656 to interact. Consequently, not only is the signal exchange cytoneme/contact-dependent, 657 but the cytoneme contacts are also formed signal-or tissue-specifically. 658 659 Free secretion and dispersion of paracrine signals are presumed to be required for long-660 range morphogen-like signaling. In this general paradigm, source surface retention is 661 inhibitory to long-range dispersion and activity. In contrast, we found that surface 662 localized display of Bnl is critical for its morphogenetic potency. This is consistent with 663 our previous finding showing that the Bnl availability on the source surface controls the 664 levels and range of Bnl signaling in the ASP 9,36 . How signal retention might facilitate 665 recipient-specific long-range patterning can be explained by our previous findings of 666 self-generation of cytoneme-mediated signaling by Bnl-signaling feedbacks 9,36 . 667 Although Bnl release is critical for its long-range morphogen like functions, we do not 668 know how GPI-anchored Bnl is released from the source membrane and how the 669 release mechanism is specifically activated at the cytoneme contact sites. We speculate 670 that an enzymatic shedding 51 of Bnl might be activated at the cytoneme contact sites. other hand, here we found that the CAM activity of Bnl is a prerequisite for its 692 subsequent target-specific release and morphogen-like roles. Consequently, a readily 693 secreted/dispersed non-GPI Bnl, although can activate receptors, fails to induce both 694 CAM-like local and morphogen-like large-scale organization (Fig.10b). Whereas, TM-695 tethered Bnl efficiently acts as a CAM, but due to its poor release from the source, 696 causes only a scaled-down ASP patterning within a narrow range (Fig.10c). The standard immunostaining and the extracellular immunostaining under live-cell non-773 permeabilized condition (GFP ex for GFP or Bnl ex for Bnl) was carried out following 774 standard protocols 9,36 . Supplementary Table 4 lists all antibodies and dilutions used. 775

DNA constructs 777
All constructs generated and used here are described in Supplementary Table 4. For PIPLC assay in wing discs, third instar larvae were prepared following ex-vivo organ 854 culture method 36 and transferred to 1.5 ml Eppendorf tubes containing 1 ml of either 855 WM1 media (control) or WM1 media with PIPLC (1U/mL). Tissues were incubated for 856 30 min at 20-25 o C with gentle rotation. Then the PIPLC reaction was stopped by 857 removing the solution and washing the tissues 3 times with WM1 media. Tissues were 858 then prepared for extracellular staining as described before.

Quantitative analyses of fluorescence intensities in tissues 882
For intracellular and extracellular surface Bnl levels, all fluorescent intensity 883 measurements were background corrected. The density of fluorescence intensity (e.g., 884 spatial range and density of signals) was measured from maximum-intensity projections 885 encompassing the wing disc, ASP, or salivary gland sections from a selected region of 886 interest (ROI) using Fiji. For each genotype, at least 3 samples were used to obtain the 887 average plot profile. Quantitative estimates of levels of Bnl:GFP variants and signaling 888 outcomes are normalized with internal controls to avoid variations among samples. For 889 example, to compare between Bnl variants, we compared the ratio of surface levels of 890 each protein (red, anti-GFP non-permeabilized immunofluorescence) to total expression 891 (total GFP fluorescence) in the same ROI of wing disc source (Fig.5) and the salivary 892 glands (Fig. 9i-l). Similarly, to assess MAPK signaling patterns of different Bnl variants 893 Excel. P values were determined using the unpaired two-tailed t-test for pair-wise 912 comparisons, or the one-way ANOVA followed by Tukey's honestly significant different (HSD) test for comparison of multiple groups. p < 0.05 is considered significant. All 914 experimental results were analyzed from at least three independent experiments. The 915 sample size (n) for each data analysis is indicated in the figures/ figure  916 legends/methods/source data. Graphs in Figure 4f  Bnl:GFP- m (n=16) and Bnl:GFPC (n=9). Rose plots were generated by R software as 926 described in 9 . 927

RNA isolation and RT-PCR 928
Total RNA was extracted from 20 wing discs of the w 1118 L3 larvae using TRI reagent 929 (Sigma-Aldrich) followed by Direct-zol RNA purification kits (Zymo Research). 930 Expression analyses of bnl PA and PC isoforms are described in Supplementary  931 Information. 932

DATA AVAILABILITY 934
All data generated and analyzed are included in the manuscript and supporting files. 935 Source data are provided with this paper. 936

CODE AVAILABILITY 938
The code for R plots is provided in the Supplementary Information.