Integration of JAK/STAT receptor–ligand trafficking, signalling and gene expression in Drosophila melanogaster cells

The Janus kinase/signal transducer and activator of transcription (JAK/STAT) signalling pathway regulates a variety of cellular events, including proliferation and apoptosis, throughout development and in adult life (Villarino et al., 2017). According to the canonical model, JAK/STAT signalling involves the activation of homo- or hetero-dimerised cell-surface transmembrane receptors by ligands, including cytokines, growth factors and hormones, which causes a conformational change in the cytoplasmic tail of the receptor. This stimulates activation of the Janus kinases (JAKs) that are constitutively associated with the receptor. JAK activation leads to specific Tyr phosphorylation of both the kinase and the receptor, subsequently allowing recruitment of signal transducer and activator of transcription (STAT) transcription factors through Src-homology 2 (SH2) domains. This association in turn allows JAK to phosphorylate STATs at a highly conserved C-terminal Tyr residue, leading to STAT dimerisation and translocation to the nucleus. In the nucleus, STATs bind to palindromic DNA sequences to alter expression of target genes, resulting in developmental, haematological and immune-related responses (O'Shea et al., 2015; Stark and Darnell, 2012). Dysregulation of the JAK/STAT pathway is involved in the pathogenesis of diseases such as gigantism, asthma, myocardial hypertrophy, myeloproliferative neoplasia and severe combined immunodeficiency (O'Shea et al., 2015).

The JAK/STAT pathway has been highly conserved throughout evolution, with invertebrates such as Drosophila melanogaster having a full complement of pathway components. However, whereas mammals have multiple copies of receptors, JAKs and STATs, in Drosophila the signalling pathway is composed of a single positively acting receptor, Domeless (Dome) (Brown et al., 2001); a negatively acting receptor, Latran (also known as eye transformer, et; Makki et al., 2010); one JAK, Hopscotch (Hop); and one STAT, STAT92E (Hou et al., 1996; Yan et al., 1996; Zeidler and Bausek, 2013). Therefore, Drosophila provides an excellent model in which to investigate JAK/STAT pathway regulation, without the difficulties of compensation and signalling crosstalk inherent in mammalian systems. In fact, investigating JAK/STAT signalling in Drosophila has led to key breakthroughs in understanding the impact of its dysregulation in human disease (Ekas et al., 2010).

The repeated use of the JAK/STAT pathway in a variety of contexts begs the question as to how transcriptional outputs are differentially regulated in a cell- and tissue-specific manner. One potential mechanism to explain this diversity of outputs is regulation by endocytosis (Sigismund and Scita, 2018; Villaseñor et al., 2016; Weinberg and Puthenveedu, 2019). Activated receptors can be internalised into cells by multiple endocytic pathways, of which clathrin-mediated endocytosis (CME) is the best characterised. Receptor complexes internalised by CME are clustered into clathrin-coated pits. The assembled clathrin lattice is linked to the cytoplasmic domains of transmembrane receptors via adaptor proteins, including the AP2 adaptor complex (Mettlen et al., 2018; Owen et al., 2004). In addition to CME, several clathrin-independent (CIE) pathways exist, which are important for the uptake of particular cargoes (Mayor et al., 2014). Following internalisation, activated receptors are delivered to the early endosome where they may be recycled or targeted to late endosomes and lysosomes for degradation. The endosomal sorting complexes required for transport (ESCRT) protein complexes are key for sorting receptors into late endosomes and lysosomes. Hrs is a component of ESCRT-0, acting as an adaptor to select ubiquitylated cargo for targeting to lysosomes. TSG101 is a component of ESCRT-I complexes, which recruit other ESCRT complexes – a process key to allowing the inward invaginations of the late endosome to form intraluminal vesicles (Henne et al., 2013). Results from in vivo and in vitro experiments indicate that endocytosis can regulate receptor signalling quantitatively through removal of activated receptors from the cell surface and targeting them to lysosomes for degradation. Endocytosis can also qualitatively regulate signalling by establishing ‘signalosomes’, which are membrane microdomains within endosomal compartments that allow the recruitment of specific scaffolds, adaptors, kinases and phosphatases, thus resulting in different downstream signalling outputs (Carroll and Dunlop, 2017; Lawrence et al., 2019; Moore et al., 2018; Sigismund and Scita, 2018; Villaseñor et al., 2016). The route of entry of activated receptors (CME versus CIE) can also influence signalling output, as demonstrated for Notch signalling in Drosophila (Shimizu et al., 2014) and TGF-β signalling in mammalian cells (Di Guglielmo et al., 2003). CME is a major entry portal that has been shown to regulate JAK/STAT signalling following activation of several different cytokine receptors in mammalian cells (Cendrowski et al., 2016; Chmiest et al., 2016; German et al., 2011; Kermorgant and Parker, 2008; Marchetti et al., 2006).

In vivo studies in Drosophila suggest that Dome-dependent border cell migration requires ligand-dependent CME and delivery to multivesicular bodies (Devergne et al., 2007). Mutation of endocytic components, including clathrin heavy chain (CHC), prevents Dome internalisation and decreases STAT92E expression and nuclear translocation in follicle cells. In contrast, endocytosis appears to negatively regulate JAK/STAT signalling in Drosophila Kc₁₆₇ cells (Müller et al., 2008; Vidal et al., 2010). These varying results likely reflect differences due to cell context, as has been observed for endocytic regulation of receptor tyrosine kinases such as epidermal growth factor receptor (EGFR) (Sousa et al., 2012; Vieira et al., 1996; Villaseñor et al., 2015). The underlying regulatory mechanisms of context-dependent signalling are, however, largely unknown.

Canonical signalling by STAT requires phosphorylation at a conserved tyrosine residue (Tyr704 in Drosophila STAT92E, isoform C used in this study), which allows for parallel dimerisation of STATs via their SH2 domains and translocation into the nucleus. There is also evidence that other posttranslational modifications, in addition to phosphorylation of the conserved Tyr, regulate STAT activity (Chung et al., 1997; Costa-Pereira et al., 2011; Grönholm et al., 2010; Karsten et al., 2006; Wang et al., 2005).

Here, we show that in Drosophila S2R+ cells, endocytosis is essential for the expression of some, but not all, JAK/STAT pathway target genes. We demonstrate that STAT92E is the target for endocytic regulation and, importantly, that endocytosis qualitatively regulates STAT92E activity. In addition, we have identified a novel phosphorylation site, Thr702, which is crucial for phosphorylation of STAT92E on Tyr704.

To understand mechanisms of Dome internalisation, we first asked how Dome and its ligand Upd2 are taken into cells. Similar to mammalian cells, Drosophila cells can internalise material by a variety of CME and CIE mechanisms (Shimizu et al., 2014). It has been shown that Dome is internalised into Drosophila Kc₁₆₇ cells by CME (Müller et al., 2008; Vidal et al., 2010). To investigate whether this is the case in S2R+ cells, we measured internalisation of Upd2–GFP, as a proxy for receptor internalisation, using an anti-GFP ELISA assay (Wright et al., 2011). We first treated cells with dsRNA targeting Dome and found that there is a significant reduction in the rate (−38%) and extent (−50%) of uptake of Upd2–GFP at both high (20 nM; Fig. 1A) and low (3 nM; Fig. S1A) concentrations of Upd2–GFP. Under these conditions, levels of Dome mRNA are reduced by ∼90% (Fig. S1B). The residual uptake of Upd2–GFP in the absence of Dome is likely due to non-specific fluid-phase uptake of ligand. When cells were incubated with 20 nM Upd2–GFP, knockdown of CHC and AP2 reduced the uptake of Upd2–GFP by ∼60% compared to knockdown of Dome alone (Fig. 1A). Because levels of CHC and AP2 mRNA were reduced by ∼80% following dsRNA knockdown (Fig. S1B and data not shown), this suggests that the Upd2–GFP complex can be internalised by CIE as well as CME, as has been shown for several receptors in mammalian cells (Sigismund et al., 2005; Vander Ark et al., 2018) and for Notch and Delta in Drosophila (Shimizu et al., 2014). By contrast, when S2R+ cells were incubated with low concentrations of Upd2–GFP (3 nM), knockdown of CHC reduced the uptake of Upd2–GFP to the level observed following Dome knockdown (Fig. S1A). Taken together this suggests that at low concentrations of Upd2–GFP, Dome is primarily internalised by CME, but that increasing concentrations of ligand results in Dome also being internalised via CIE.

Sorting of cargo into clathrin-coated pits requires internalisation motifs in the cytoplasmic tails of receptors that include both Tyr- and di-leucine (di-Leu)-based motifs (Traub, 2003). Dome is most similar in sequence and structure to gp130 (also known as IL6ST) of vertebrates, which is a co-receptor shared by receptors for IL-6 (Fig. 1B). Internalisation of gp130 requires a di-Leu motif (⁷⁸⁶LL⁷⁸⁷) in its cytoplasmic domain (Dittrich et al., 1996) while an upstream serine within the sequence ⁷⁸⁰SESTQPLL⁷⁸⁷ has also been shown to be important for rapid internalisation (Dittrich et al., 1996). Strikingly, the cytoplasmic tail of Dome also contains a di-Leu motif, ⁹⁸⁵LL⁹⁸⁶, in a similar context to that of the di-Leu motif in gp130 (Fig. 1C). In order to test the potential significance of this motif, we generated a series of FLAG-tagged Dome mutant constructs where individual elements of the di-Leu cassette were mutated either alone or in combination (Fig. 1C), and transfected these constructs into S2R+ cells. To quantitatively measure ligand-dependent uptake of the engineered Dome constructs, proteins on the surface of transfected S2R+ cells were biotinylated prior to addition of Upd2–GFP. This showed that, although expression of the mutants was somewhat more efficient than that following transfection of wild-type Dome (Fig. S1C), plasma membrane expression of all the constructs was comparable (Fig. S1D). Following ligand internalisation, cell-surface biotin was removed by treatment with the reducing agent, 2-mercaptoethanesulfonic acid sodium salt (MESNa), while internalised cell-surface proteins were protected and remained biotinylated. This allowed the amount of internalised wild-type and mutant Dome to be quantitated. As has been demonstrated previously for Dome (Ren et al., 2015), we observed ligand-independent internalisation of Dome (Fig. S1E). We found that mutation of the entire di-Leu cassette to AAASKAA (defined from now on as Dome^allA) inhibited internalisation of Dome. Mutation of the di-Leu motif alone (Dome^LL985AA–FLAG) did not significantly reduce internalisation. Using site-directed mutagenesis in which we progressively replaced elements of the putative cassette, we established that Glu980 and ⁹⁸⁵LL⁹⁸⁶ together represent essential residues required for Dome internalisation (Fig. 1D,E). Mutation of Glu980 alone did not significantly affect Dome internalisation (Fig. S1F,G). Although uptake of Dome ^{E980G/LL985AA}–FLAG was significantly inhibited (∼66%), the effect on internalisation was less than that observed for the Dome^allA–FLAG mutant, suggesting that other determinants may also be present within the sequence that are important for Dome internalisation (Fig. 1D,E). Taken together, these results identify a di-Leu-containing cassette as being essential for Dome internalisation.

Dome signalling is known to be regulated by endocytosis in Kc₁₆₇ cells (Müller et al., 2008; Vidal et al., 2010) and in vivo (Devergne et al., 2007). To test whether it is similarly regulated in S2R+ cells, we measured the expression of the exogenous reporter 10×STAT–Luciferase, which expresses the firefly luciferase enzyme under the control of a minimal promoter downstream of ten STAT92E binding sites (Baeg et al., 2005). As expected, this reporter is activated in S2R+ cells by exogenous Upd2–GFP, in a dose-dependent manner (Fig. 2A), indicating that these cells express the JAK/STAT pathway components required for activation. We next measured Upd2–GFP-dependent 10×STAT–Luciferase reporter activity in control cells and those expressing FLAG-tagged wild-type Dome (Dome ^wt–FLAG) or Dome^allA–FLAG (Fig. 2B). Although expression of Dome^WT–FLAG did not significantly affect signalling, expression of Dome^allA–FLAG had a strong dominant-negative effect on Upd2–GFP-mediated pathway stimulation. This effect was comparable to the level observed in cells expressing Dome^Y966A/Q969A–FLAG and Dome^P925I–FLAG, mutants which have been previously reported to have reduced signalling because of their inability to bind STAT92E (Stahl and Yancopoulos, 1994) and Hop (Fisher et al., 2016), respectively. Levels of expression of the transfected proteins are shown in Fig. S2A. Taken together, these data demonstrate that Dome mutants that cannot be internalised also alter JAK/STAT signalling and are consistent with a model where activation of 10×STAT–Luciferase by Upd2–GFP is dependent on Dome internalisation.

To further explore a role for endocytosis in regulating signalling downstream of Dome, we asked whether knocking down components of the endocytic machinery might differentially affect expression of Dome target genes. We therefore examined the expression of the 10×STAT–Luciferase reporter and the endogenous target genes socs36E and lama (Flaherty et al., 2009; Karsten et al., 2002) in cells treated with dsRNA to knock down endocytic components. We targeted AP2, an adaptor whose knockdown is predicted to result in accumulation of receptors at the cell surface (Robinson, 2004); Hrs, an adaptor whose knockdown is likely to result in accumulation of ubiquitylated receptors in early endosomes; and TSG101, which is required for the sorting of receptors into intraluminal vesicles and whose knockdown is likely to lead to an accumulation of receptors on the limiting membrane of late endosomes (Henne et al., 2013). Treating cells with dsRNA to knockdown Dome (levels of Dome mRNA were reduced by ∼90%; Fig. S1B) resulted in almost complete abolition of 10×STAT–Luciferase expression, demonstrating that both background, and Upd2–GFP-stimulated, reporter activation are receptor dependent (Fig. 2C). In the absence of exogenous ligand, activation of 10×STAT–Luciferase in cells treated with dsRNA targeting AP2, Hrs or TSG101 was, however, unchanged compared to cells treated with control dsRNA (Fig. 2C). We speculate that this ligand-independent activation is due to expression of ligands and growth factors that may crosstalk with the JAK/STAT pathway in S2R+ cells (Cherbas et al., 2011). By contrast, knockdown of AP2 significantly reduced ligand-dependent 10×STAT–Luciferase activation, whereas knockdown of Hrs or TSG101 had no effect. This indicates that activation of this reporter requires delivery of activated Dome either to, or beyond, an AP2-positive endocytic compartment but prior to an Hrs-positive endosomal compartment. We also examined an endogenous target of Dome, socs36E (Stec et al., 2013) and found that, in contrast to 10×STAT–Luciferase expression, knockdown of both AP2 and Hrs inhibited socs36E mRNA expression, whereas knockdown of TSG101 had no effect (Fig. 2D). This indicates that activated Dome must be trafficked to an Hrs-positive compartment, or beyond, to allow downstream pathway activation to trigger socs36E transcription. Taken together, these results indicate that the location of the activated Upd2–Dome complexes within the endocytic pathway can lead to qualitatively different signalling outputs. It is important to note that not all Dome target genes are regulated by endocytosis. For example, expression of lama, a well-characterised target of STAT92E (Flaherty et al., 2009), was unaffected when endocytosis was perturbed, suggesting that expression of this target gene mRNA can be driven by activated Upd2–Dome complexes that are located on the plasma membrane (Fig. S2B).

Upon ligand activation of Dome, STAT92E is phosphorylated by Hop at a conserved Tyr residue (Y704) (Yan et al., 1996). This residue is conserved across all vertebrate STATs, and its phosphorylation is essential for canonical STAT activity and target expression. We therefore asked whether Tyr704 phosphorylation of STAT92E was sensitive to endocytic regulation. One approach to assaying STAT92E phosphorylation utilises its change in electrophoretic mobility on SDS–PAGE gels (Shi et al., 2008), caused by changes in charge and conformation that occur following phosphorylation (Mao et al., 2005; Wenta et al., 2008). Using this experimental approach, we observed an Upd2-dose-dependent change in the electrophoretic mobility of STAT92E following ligand stimulation (Fig. 3A,B), which was reversed by phosphatase treatment (Fig. 3C,D). Strikingly, perturbation of the endocytic pathway by knockdown of AP2 (Fig. 3E,F), or Hrs or TSG101 (Fig. S3), did not affect the temporal dynamics of STAT92E phosphorylation, a finding that was also confirmed by mass spectrometry (Fig. 3G; data available via ProteomeXchange PXD020719). These data demonstrate that phosphorylation on Tyr704 of STAT92E is not regulated by endocytosis and that other mechanisms must be responsible for the pathway's sensitivity to endocytic regulation.

Canonical JAK/STAT pathway signalling requires nuclear import of the STAT92E transcription factor to activate gene expression. We therefore investigated whether knockdown of AP2 impaired translocation of STAT92E into the nucleus. Nuclear accumulation could be visualised in S2R+ cells transfected with STAT92E–GFP. In the absence of ligand there appeared to be low levels of STAT92E–GFP in the nucleus. This is consistent with reports that STATs shuttle between the nucleus and cytoplasm in a phosphorylation-independent manner and that unphosphorylated nuclear STATs can perform non-canonical functions (Brown and Zeidler, 2008). The levels of nuclear STAT92E–GFP we observed in the absence of Upd2 is also in keeping with reports of GFP-tagged proteins entering the nucleus independently of a nuclear localisation signal (Seibel et al., 2007). When cells were treated with Upd2–GFP (Fig. 4A,B), a maximum accumulation was reached after 30 min stimulation. This is comparable to the nuclear accumulation of mammalian STATs (McBride et al., 2000) and the time-point at which STAT92E phosphorylation is maximal (data not shown). Consistent with previous studies (Begitt et al., 2000; Schindler et al., 1992), mutation of STAT92E Tyr704 (Y704F), to prevent phosphorylation, abolished nuclear accumulation (Fig. 4C). Although knockdown of Dome almost completely abolished nuclear accumulation of STAT92E–GFP, knockdown of either AP2 or Hrs had no significant effect, indicating that endocytic trafficking of Upd2–Dome does not regulate nuclear accumulation of STAT92E (Fig. 4D). This demonstrates that the loss of target gene expression following AP2 and Hrs knockdown is not likely to be the result of a defect in the translocation of STAT92E into the nucleus.

Given that Y704 phosphorylation is necessary but not sufficient for STAT92E-driven pathway gene expression, we wanted to investigate whether other posttranslational modifications of STAT92E might be associated with pathway activation. We expressed STAT92E–GFP in S2R+ cells, stimulated them with Upd2–GFP and subjected samples isolated using GFP–TRAP beads to mass spectrometry analysis. In addition to Tyr704, this analysis identified Thr47, Ser227 (Fig. 5A) and Thr702 (with lower confidence) on STAT92E as being phosphorylated (data available via ProteomeXchange with identifier PXD020719). We therefore decided to test the potential physiological relevance of these newly identified phosphorylation sites using an S2R+ cell line lacking endogenous STAT92E. We used CRISPR-Cas9 to engineer STAT92E-negative S2R+ cells, demonstrating that the cell line no longer had detectable STAT92E by western blotting (Fig. S4A) and T7 endonuclease assay (Fig. S4B) and was no longer able to activate 10×STAT–Luciferase in response to Upd2–GFP treatment (Fig. 5B). As expected, expression of wild-type STAT92E was able to rescue both Upd2–GFP-dependent and -independent 10×STAT–Luciferase activity (Fig. S4C) in these STAT92E-negative cells, whereas ligand-dependent 10×STAT–Luciferase activity was further enhanced by expression of STAT92E^K187R, a mutant form of STAT92E that cannot be SUMOylated and that has previously been shown to increase luciferase activity (Grönholm et al., 2010). Taken together, these results demonstrate the utility of the STAT92E-negative S2R+ cells for rescue experiments (Fig. S4C).

We next generated mutant forms of STAT92E lacking both known and candidate phosphorylation sites (T47V, S227A, T702V and Y704F), expressed them in STAT92E-negative S2R+ cells and measured their ability to activate 10×STAT–Luciferase. Following ligand stimulation with 0.75 nM Upd2–GFP, STAT92E^T47V, STAT92E^S227A and STAT92E^WT resulted in comparable levels of 10×STAT–Luciferase activation, whereas cells expressing STAT92E^T702V and STAT92E^Y704F showed no activation (Fig. 5C). This indicates that phosphorylation of Thr702 as well as Tyr704, but not Thr47 or Ser227, is required for JAK/STAT signalling.

To further explore the role of Thr702 phosphorylation in STAT92E-mediated gene activation, we generated phosphomimetic mutants of Thr702 (STAT92E^T702D, STAT92E^T702E) and tested their effects on activation of the 10×STAT–Luciferase reporter. Using the STAT92E-negative S2R+ cell assays, we first showed that expression of ‘loss-of-phosphorylation’ mutants STAT92E^T702V and STAT92E^Y704F did not stimulate reporter activity above background levels (Fig. 5D). By contrast, expression of either phosphomimetic mutant, STAT92E^T702D or STAT92E^T702E, was sufficient to increase both ligand-dependent and ligand-independent 10×STAT–Luciferase expression, with STAT92E^T702D more effective in both cases. Taken together, we have thus identified a novel posttranslational modification of STAT92E that is essential for triggering transcriptional activity in this assay.

We next asked whether Thr702 phosphorylation is required for nuclear translocation of STAT92E and found that Upd2–GFP does not stimulate STAT92E^T702V translocation into the nucleus (Fig. 6A,B). Using mass spectrometry, we found that STAT92E^T702V showed a substantial reduction in Tyr704 phosphorylation (Fig. 6C; data available via ProteomeXchange with identifier PXD020719). This indicates that phosphorylation of Thr702 is essential for efficient phosphorylation of Tyr704, which, in turn, is essential for the bulk of canonical JAK/STAT gene expression.

In this work we have explored regulatory mechanisms of JAK/STAT signalling following Upd2-dependent Dome activation in Drosophila S2R+ cells. We have identified an evolutionarily conserved internalisation motif in the cytoplasmic tail of Dome. We have demonstrated that internalisation and endocytic trafficking of activated Dome allows for compartmentalised signalling to regulate subsets of Drosophila JAK/STAT transcriptional targets, through a mechanism that is independent of Tyr704 phosphorylation of STAT92E. We have also demonstrated that phosphorylation of Thr702 is essential for Tyr704 phosphorylation of STAT92E, its translocation to the nucleus and its activity as a transcription factor.

It has been shown that Dome enters cells by CME in vivo in Drosophila (Devergne et al., 2007) and in vitro in Kc₁₆₇ cells (Müller et al., 2008; Vidal et al., 2010). Our results also support a role for CME in Dome uptake in S2R+ cells, because dsRNA-mediated knockdown of CHC and AP2 reduced Upd2–GFP internalisation. There are a number of defined motifs that allow the inclusion of transmembrane receptors into clathrin-coated pits, through interactions with adaptor molecules such as AP2. A di-Leu motif is one such motif, which is well documented to bind to the α–σ2 hemicomplex of AP2 (Doray et al., 2007; Kelly et al., 2008). In this work, we have demonstrated that such a motif is part of a cassette that is essential for efficient internalisation of Dome. Interestingly, a di-Leu-containing cassette is also required for the internalisation of gp130, the closest vertebrate homologue of Dome and the co-receptor for IL-6R, which is necessary for IL-6R internalisation (Dittrich et al., 1996). Similar to gp130, mutation of the di-Leu motif alone in Dome was insufficient to completely abolish internalisation. In the case of gp130, a Ser residue upstream of the di-Leu motif was also shown to be involved in rapid internalisation. We found that mutation of the equivalent Ser, in combination with mutation of the di-Leu motif, further reduced Dome internalisation, although still not to the same extent as in the Dome^allA mutant. An acidic residue (Glu or Asp) at the −4 position is commonly found adjacent to di-Leu motifs, and its mutation has previously been shown to drastically decrease binding to the α–σ2 hemicomplex of AP2 (Doray et al., 2007). Mutation of this charged residue alone had no effect on receptor internalisation, whereas mutation of both the Glu and di-Leu residues reduced internalisation by ∼66% compared to internalisation of Dome^WT. This suggests that, although the Glu and di-Leu are important, other residues may also influence Dome internalisation. It also points to an important evolutionary conservation in mechanisms of Dome internalisation that is in line with the conservation of JAK/STAT pathway components across species.

Our results support a role for CIE, in addition to CME, in uptake of activated Dome in S2R+ cells. While dsRNA-mediated knockdown of CHC and AP2 inhibits internalisation of Upd2–GFP bound to Dome, the extent of inhibition depends on the concentration of the Upd2–GFP ligand. At low concentrations (3 nM) of Upd2–GFP, there is an absolute requirement for CHC and AP2, whereas at higher concentrations (20 nM), uptake of Upd2–GFP bound to Dome in cells treated with dsRNA targeting CHC and AP2 is inhibited by ∼50% compared to cells treated with dsRNA targeting Dome. This is consistent with studies in Drosophila where uptake of Notch and Delta through different endocytic pathways (CME and CIE) leads to delivery to different endosomal compartments and differential signalling, with the balance of flux between these pathways allowing cells to respond to different environmental conditions (Shimizu et al., 2014). Similarly, in mammalian cells, activated receptor tyrosine kinases, such as TGF-β receptors and EGFR, can be taken up by CME and CIE, with CME being favoured at lower ligand concentrations (Di Guglielmo et al., 2003; Sigismund et al., 2005). As with Notch signalling, the route of entry of the receptors can determine signalling outcome and receptor fate (Sigismund et al., 2013; Vander Ark et al., 2018). The concept of endocytosis modulating Dome target gene expression in different cells and tissues is supported by previous in vitro and in vivo studies (Devergne et al., 2007; Silver et al., 2005; Vidal et al., 2010). Our experiments, which have focussed on CME of activated Dome, indicate that endocytosis also regulates a subset of Dome signalling in S2R+ cells. Mutation of the internalisation motif not only prevented Dome uptake but also prevented Dome activation of 10×STAT–Luciferase, consistent with a role for endocytosis in activation of target genes. It is noteworthy that we observed constitutive internalisation and recycling of Dome in the absence of ligand, similar to that observed for other cytokine receptors in mammalian cells (Thiel et al., 1998). Regulation of constitutive recycling provides cells with a mechanism to control cell-surface levels of receptor, which in turn will impact on the magnitude of signalling (Moore et al., 2018).

Strikingly, we have demonstrated that endocytosis of Dome allows an additional level of regulatory control, in that delivery to distinct endosomal populations can further affect signalling outcome. Endocytosis is not required for expression of all genes, for example, lama is still expressed even when components of the endocytic machinery are ablated with dsRNA. By contrast, expression of 10×STAT–Luciferase requires delivery to, or beyond, an AP2-positive compartment, and expression of socs36E only occurs when activated Dome has trafficked through an Hrs-positive compartment, but before it has reached a TSG101-positive compartment (Fig. 6D). Our data thus demonstrate that qualitatively different signalling outputs can occur depending on the location of the activated receptor within the endocytic pathway. This strongly supports the concept that the rate at which receptors, in this case Dome, move through the pathway (endocytic flux) is key for signalling outputs and will have profound effects on downstream cell behaviours. This is consistent with studies of EGFR signalling, which imply that receptor signalling can modulate the endocytic machinery to determine the rate of receptor flux (Villaseñor et al., 2015). Although mechanistic details for endocytic regulation of signalling are better understood for receptor tyrosine kinases (RTKs) and G-protein-coupled receptors (GPCRs), there is a considerable body of emerging evidence to support a role for endocytic regulation of cytokine receptors in mammalian cells (Cendrowski et al., 2016). Our data are thus consistent with a variety of studies in mammalian cells demonstrating an instructive role for endocytosis in JAK/STAT signalling (Cendrowski et al., 2016; Chmiest et al., 2016; German et al., 2011; Kermorgant and Parker, 2008; Marchetti et al., 2006).

In Drosophila, STAT92E is the single transcription factor utilised by the JAK/STAT pathway to control expression of many different target genes, which are expressed in a tissue-specific and developmentally-regulated manner. The essential role of Tyr704 phosphorylation in JAK/STAT signalling is well established (Yan et al., 1996). We eliminated the possibility that endocytosis is required for STAT92E phosphorylation by demonstrating that STAT92E is phosphorylated to the same extent when components of the endocytic machinery, such as AP2, are knocked down by dsRNA. More importantly, our data demonstrate that STAT92E Tyr704 phosphorylation, although necessary, is not sufficient for the expression of all Dome target genes. Our data are consistent with previous studies showing that a mutant form of STAT92E that cannot be methylated is hyper-phosphorylated, but has a dominant-negative effect on target gene expression (Karsten et al., 2006).

When the endocytic pathway is disrupted, phosphorylated STAT92E can still translocate into the nucleus, but it is no longer fully signalling competent. This implies that Dome needs to reach a particular endosomal subcompartment or microdomain in order to allow STAT92E to become transcriptionally competent. Of particular interest is the post-Hrs and pre-TSG101 compartment required for socs36E expression (Fig. 6D). Hrs is a component of the ESCRT-0 complex, which recognises ubiquitylated signalling cargo destined to be packaged into inward invaginations of the endosomal membrane to form intraluminal vesicles and, ultimately, multivesicular bodies. TSG101 is required for later stages of intraluminal vesicle formation (Vietri et al., 2019). As such, both these components are found within the same limiting membrane. It has been proposed that membrane microdomains of defined composition, containing signalling molecules, must be able to form within endosomal membranes to generate local signalling-competent (signalosome) domains (Shimizu et al., 2014; Teis et al., 2002). Within these specialised signalosomes, STAT92E is likely either to undergo additional posttranslational modifications or to acquire a chaperone protein that facilitates its ability as a transcription factor for a subset of target genes. Support for a Hrs (also known as HGS in mammals) signalosome comes from studies that demonstrate that the Hrs-interacting protein STAM is required for downstream signalling following IL2R activation (Takeshita et al., 1997; Tognon et al., 2014). In mammals, STAMs are phosphorylated in response to a range of cytokines and growth factors (Pandey et al., 2000). The Hrs–STAM complex remains an interesting link between signalling and endocytosis, because it has been shown to have both positive and negative roles in the regulation of RTK signalling in Drosophila that are dependent on specific tissue and developmental stages (Chanut-Delalande et al., 2010).

Previous studies in mammalian cells have shown that endosomal location is required for STAT3 activation by activated c-Met (also known as MET), which is classed as a weak activator, and it was proposed that by localising STAT3 activation in endosomes, nuclear import is facilitated (Kermorgant and Parker, 2008). Here, we show the importance of localisation at different points along the endocytic pathway to nuance Dome signalling to allow different signalling outputs, with STAT92E being a target for endocytic regulation.

Mass spectrometry analysis revealed Thr702 as a novel phosphorylation site on STAT92E that is functionally important. Mutation to valine, which is structurally similar but cannot be phosphorylated, prevented STAT92E Tyr704 phosphorylation and nuclear translocation, whereas phosphomimetic forms of Thr702 rescued this phenotype. Alignment (Waterhouse et al., 2018) of STAT92E with the published crystal structure of STAT1 (Chen et al., 1998) suggests that Thr702 and Tyr704 are located in a flexible loop region (Fig. 6E). Phosphorylation is likely to have significant effects on the conformation of this region. Intriguingly, the Thr residue at this position is conserved in STAT1 and is a phosphomimetic residue in STAT5, suggesting that it may play a role in ensuring effective Tyr phosphorylation of STATs across species.

In summary, we have shown that endocytosis regulates JAK/STAT signalling in Drosophila S2R+ cells, resulting in qualitatively different signalling outputs. We therefore suggest that the endocytic flux of activated Dome provides a mechanism by which JAK/STAT can regulate different cellular behaviours depending on cell context. In the course of our studies we have shown that although phosphorylation of Tyr704 on STAT92E is necessary, it is not sufficient for expression of some JAK/STAT target genes. Moreover, for some targets, delivery to an endosomal subcompartment is required in order to make STAT92E transcriptionally competent.

S2R+ cells were cultured at 25°C in Schneider's Insect Tissue Culture medium (Gibco, UK), supplemented with 10% heat inactivated FBS (Sigma-Aldrich, UK.), penicillin (1000 units/ml) and streptomycin (0.1 mg/ml) (Sigma-Aldrich, UK) and 2 mM L-glutamate (Gibco, UK). Cells were grown to confluency in T75 flasks and routinely passaged at a 1:3 dilution every 3–4 days.

For expression of STAT92E–GFP (Addgene plasmid 126662, deposited by Ron Vale) or Dome–FLAG, cells were seeded a day prior to transfection. They were transfected at a ratio of 2 µg DNA per 1×10⁶ cells in a 6-well plate, using Effectene Reagent (Qiagen Ltd, UK) and used 2 days later for experiments. Cells were routinely tested to ensure that they were free of mycoplasma.

Upd2-GFP conditioned medium was produced essentially as described previously (Wright et al., 2011), with the following modifications: S2R+ cells were seeded at 1×10⁶ cells per well of a 6-well plate 1 day prior to transfection. pAct-Upd2–GFP (2 μg per well) was transfected using Effectene Transfection Reagent (Qiagen Ltd, UK) following the manufacturer's instructions. After 2 days, 3 wells of transfected cells were transferred to a T75 flask and incubated for a further 4 days. Cells were centrifuged at 1000 g for 3 min, and the medium was filtered, aliquoted and snap frozen in liquid N₂, then stored at −80°C. The concentration of Upd2–GFP was determined using an ELISA for GFP (see below). Mock conditioned medium (referred to as mock treatment) was produced by transfecting cells with 2 μg pAc5.1 (Müller et al., 2005) and processed as above.

dsRNAs were obtained from the Sheffield RNAi Screening Facility, whose dsRNA database is based on the Heidelberg 2 library (Boutros laboratory, DKFZ, Heidelberg, Germany), generated with Next-RNAi (Horn et al., 2010). It is the redesigned, non-off-target-effect library, HD2.0, generated using the software next-RNAi (Horn et al., 2010). Low-complexity regions and sequence motifs that induce off-target effects have been excluded. dsRNA probe sizes vary from 81 to 800 bp covering ∼14,000 protein encoding genes and ∼1000 non-coding genes (∼98.8% coverage). The dsRNA design covers every isoform of each gene and has been optimised for specificity and avoidance of low complexity regions. The following dsRNA amplicons were used: α-adaptin (BKN20148); CHC (BKN20463); Dome (BKN25660); Hrs (BKN27923); and TSG101 (BKN28961). Negative control dsRNA was a mixture of three amplicons targeting Caenorhabditis elegans mRNA (BKN70003, BKN70004 and BKN70005). Amplification of dsRNA was carried out using a MEGAscript RNAi Kit (Life Technologies, AM1626), and purified via ethanol precipitation with sodium acetate, followed by resuspension in sterile water.

Cells were seeded 1 day prior to knockdown, and resuspended in serum-free medium on the day of knockdown. The desired number of cells was added to wells already containing dsRNA and incubated for 1 h at 25°C (15 µg of dsRNA plus 1×10⁶ cells per well in a 6-well plate). After incubation, an equal volume of fresh medium containing 20% FBS was added. Cells were incubated at 25°C for a total of 5 days before subsequent experiments. Transfection with STAT92E–GFP was performed on day 3 of dsRNA treatment.

sgRNA were designed to target the N-terminal coding region of STAT92E and showed <1% chance of off-target activity (crispr.mit.edu). Sequences were also verified using NCBI BLAST to eliminate potential off-target hits. The NGG sequence was then removed, and a G was added to the 5′ end of the sgRNA sequence to allow transcription from the U6 promoter in pAc-sgRNA-Cas9 vector (Addgene plasmid 49330, depositied by Ji-Long Liu). sgRNA oligos (Table S1) were cloned into the pAc-sgRNA-Cas9 expression vector following a previously published protocol (Bassett et al., 2014). S2R+ cells were plated at 5×10⁵ cells per well in a 12-well plate and transfected with 1 μg pAc-sgRNA-Cas9 construct using Effectene (Qiagen Ltd, UK). After 3 days, puromycin (5 µg ml⁻¹) selection was performed for 7 days before subsequent analysis (Bassett et al., 2014).

To detect Cas9-induced mutations within the genomic DNA of S2R+ CRISPR cell lines, a T7 endonuclease assay was carried out to identify mismatched, heteroduplex DNA. PCR products were first produced by amplifying an ∼1 kb region around the Cas9 cut site in a 50 μl PCR reaction, following a previously published method (Guschin et al., 2010). Following verification of size on agarose gels, PCR products were denatured and annealed to form heteroduplexes in the following reaction: 5–10 μl PCR products, 2 μl NEBuffer 2 (NEB) made up to 19 μl with nuclease-free water. The reaction was heated at in a 95°C heat block for 10 min and allowed to cool to room temperature. 1 μl of T7 endonuclease was then added to reactions and incubated at 37°C for 15 min. The reaction was stopped by addition of 1.5 μl 0.25 M EDTA before running on an agarose gel.

The anti-GFP ELISA was performed essentially as described previously (Wright et al., 2011). Briefly, a 96-well high-binding EIA plate (Costar) was coated with 0.0625 μg ml⁻¹ goat anti-GFP antibody (Abnova; PAB10341) in 100 mM sodium bicarbonate overnight at 4°C. The plate was washed three times with wash buffer [0.2% (w/v) BSA, 0.5% Triton-X 100 in phosphate-buffered saline (PBS)] and then blocked in the same buffer for 1 h at room temperature. A serial dilution of recombinant GFP (Cellbiolabs; STA-201), starting at 5 ng ml⁻¹, was plated for reference. Samples were incubated for 3 h at 37°C. After washing, the plate was incubated with rabbit anti-GFP (Abcam; Ab290) at 1:20,000 for 2 h at room temperature. After further washes, the plate was incubated with a secondary HRP-linked anti-rabbit antibody (Santa Cruz Biotechnology; sc-2004) at 1:5000 for 1 h at room temperature. Following washing, 200 μl per well of freshly prepared HRP developing solution [0.012% H₂O₂, 0.4 mg ml⁻¹ o-phenylenediamine in HRP assay buffer: 51 mM Na₂HPO₄, 27 mM citric acid, pH 5.0, (filtered)] was added to the plate and colour change was observed. To stop the reaction, 50 μl of 2 M H₂SO₄ was added per well, and the absorbance read at 492 nm on a BMG Labtech plate reader.

Cells were seeded in a 24-well plate (2×10⁵ cells per well) a day prior to the experiment. The medium was replaced with conditioned medium containing established concentrations of Upd2–GFP and incubated at 25°C for various times. Endocytosis was stopped by placing cells on ice and washing twice with ice-cold PBS. Cell-surface ligand was removed by acid washing twice with 0.2 M glycine, 0.15 M NaCl, pH 2.5 for 2 min. Cells were then washed again in PBS before lysis in ELISA lysis buffer [PBS containing 1 mM MgCl₂, 0.1% (w/v) BSA and 0.5% Triton-X 100 supplemented with cOmplete™ Mini, EDTA-free protease inhibitor cocktail (Roche; 11836170001)].

All reactions were carried out on ice unless specified. Growth medium was aspirated from cells, which were then washed twice with ice-cold PBS. Cells were incubated for 1 h on ice with freshly prepared EZ-link™ Sulfo-NHS-SS-Biotin (Thermo Fisher Scientific; 0.25 mg ml⁻¹) before biotin was quenched by washing twice with PBS containing 100 mM glycine. Internalisation was allowed to proceed for various times by adding pre-warmed Upd2–GFP and incubating at 25°C. Cells were returned to ice and washed twice with PBS. Cell-surface biotin was cleaved by washing cells three times for 20 min with MESNa [100 mM 2-mercaptoethanesulfonate, added fresh for each incubation to 50 mM Tris-HCl pH8.6, 100 mM NaCl, 1 mM EDTA and 0.2% (w/v) BSA]. Cells were then washed three times in PBS. Reduced disulphide bonds were alkylated for 10 min with 500 mM iodoacetamide in PBS, before two final PBS washes. Cells were then lysed for 30 min and lysates were centrifuged at 18,000 g for 10 min. Streptavidin–agarose (15 µl) was washed three times with lysis buffer and incubated with cell lysate (10–30 μg) overnight at 4°C with rotation. Beads were then washed three times with lysis buffer and boiled for 5 min at 95°C in 20 μl Laemmli SDS–PAGE buffer, before SDS–PAGE and western blotting. Lysis buffer: 20 mM Tris-HCl pH7.5, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton X-100, 1 mM β-glycerophosphate, 25 mM Na-Pyrophosphate, 1 mM Na₃VO₄, 1 μg ml⁻¹ microcystin and 25 mM N-ethylmaleimide supplemented with cOmplete™ Mini, EDTA-free protease inhibitor cocktail (Roche; 11836170001).

Cells were seeded in a 12-well plate at 5×10⁵ cells per well 1 day prior to transfection. Cells were transfected with 0.5 μg 10×STAT–Luciferase and 0.5 μg pAct-Renilla (pAc-Ren, internal control for transfection; Müller et al., 2005) for 1 day and then transferred to a 96-well plate at 5×10⁴ cells per well. Cells were treated with conditioned medium containing Upd2–GFP for 18 h. Luciferase activity was measured using the Dual-Glo Luciferase Assay System (Promega), following the manufacturer's instructions, using a 1:5 dilution of DualGlo-luciferase in distilled water. The Dual-Glo Stop and Glo Luciferase Assay reagent (1:5 dilution) was added to the plate at an equal volume to the culture medium in the wells and incubated for at least 10 min. The firefly luciferase signal was measured using a Thermo Scientific Varioskan Flash Luminometer. An equal volume of Dual-Glo Stop and Glo reagent was then added and incubated for at least a further 10 min to allow measurement of the Renilla luciferase (RL) signal. Luciferase activity was calculated as firefly luciferase value normalised to the internal transfection control (RL).

STAT92E was immunoprecipitated from cells lysed in lysis buffer [PBS containing 1 mM MgCl₂, 0.1% (w/v) BSA, 0.5% Triton-X 100 supplemented with cOmplete™ Mini, EDTA-free protease inhibitor cocktail (Roche)]. Calf intestinal phosphatase (CIP; M0290S, NEB) was added at 1 unit per 1 μg protein then incubated for 1 h at 37°C. The reaction was stopped by addition of Laemmli SDS–PAGE buffer and boiling at 95°C for 5 min.

RNA extraction was carried out using TRI reagent (Sigma-Aldrich; T9424) and reverse transcribed using the High Capacity RNA-to-cDNA™ kit (Applied Biosystems; 4387406). cDNA was diluted 1:10 and relative mRNA levels of socs36E, Dome, lama, AP2, Hrs and TSG101 were quantified using qPCR. This was performed using SYBR Green JumpStart™ Taq ReadyMix™ (Sigma-Aldrich; S4438) and the primers listed in Table S2, on a Bio-Rad CFX96 real-time system, C100 Touch thermal cycler or Applied Biosystems QuantStudio 12K Flex. A standard curve of diluted template was used to interpolate the quantity of target gene in the test samples. Results for each target were normalised to levels of the reference gene, ribosomal protein L32 (Rpl32) mRNA, within each well.

Site-directed mutagenesis was carried out using a QuikChange Site-Directed Mutagenesis kit (Agilent Technologies), according to the manufacturer's instructions. Sequencing of plasmid DNA was carried out at the University of Sheffield's Core Genomic Facility, and results were analysed using ApE (ApE, A plasmid Editor v2.0.50; https://jorgensen.biology.utah.edu/wayned/ape/).

STAT92E–GFP samples, prepared from GFP-trap beads or gel bands, were processed for mass spectrometry (MS) according to standard procedures. Samples were reduced with TCEP, followed by alkylation with iodoacetamide and digestion with trypsin or GluC. For some samples additional processing steps were required; for example, the use of HiPPR Detergent Removal spin columns (Thermo Fisher Scientific) for initial Triton removal or microscale solid phase extraction TiO₂ tips (Thermo Fisher Scientific) to enrich phosphorylated peptides. STAT92E–GFP peptides were analysed in an LTQ Orbitrap Elite (Thermo Fisher Scientific) hybrid ion trap-orbitrap mass spectrometer equipped with an EASY-Spray Ion Source hyphenated to a Dionex Ultimate 3000 uHPLC (Thermo Fisher Scientific). Peptides were trapped with an Acclaim™ PepMap™ 100 C18 trap column (3 µm particle size, 75 µm × 150 mm) and separated on an EASY-Spray™ C18 column (2 µm particle size, 50 µm × 150 mm) using an 80 min gradient with 0.1% formic acid in water (mobile phase A) and 0.1% formic acid in 80% acetonitrile (mobile phase B). Positive MS survey scans were acquired in the FT-orbitrap analyzer using an m/z window from 375 to 1600, and a resolution of 60,000. The 20 most intense precursor ions were selected for the acquisition of MS/MS spectra in the ion trap (Normal Scan Rate) using collision-induced dissociation (CID) with normalized collision energy of 35% and isolation width of 2 Th. Raw MS/MS data files were used for protein identification using MaxQuant software (Cox and Mann, 2008) with default parameters, except for the following: protein database was the Drosophila melanogaster Uniprot proteome (downloaded on 25 August 2016) and a FASTA file containing the theoretical sequence of STAT92E–GFP and STAT92E^T702V–GFP; LFQ quantification and iBAQ values were selected for calculation; variable modifications, oxidation (M) and phosphorylation (STY); fixed modifications, carbamydomethylation (C). Relevant spectra were manually inspected after annotation with pLabel software (Li et al., 2005; Wang et al., 2007). A detailed description of mass spectrometry methods (sample preparation, mass spectrometry analysis and data processing), together with mass spectrometry data and annotated relevant spectra (phosphorylated Y704, T47, S227 and T702), has been deposited at the ProteomeXchange Consortium via the PRIDE partner repository (Deutsch et al., 2020). The identifier number of the dataset is PXD020719.

Following dsRNA treatment or transfection with STAT92E constructs, cells were plated on sterile coverslips, pretreated overnight with 0.1 mg/ml poly-L-lysine. Following treatment with Upd2–GFP, cells were fixed in 4% paraformaldehyde for 20 minutes, then quenched with two 5-minute washes with 50 mM ammonium chloride in PBS. Following permeabilisation with 0.2% Triton X-100 in PBS for 5 minutes, coverslips were washed three times with PBS and incubated with 0.2% fish skin gelatin (FSG) for 1 h at room temperature. Following three washes with PBS, coverslips were incubated with anti-GFP antibody (Abnova, PAB10341; diluted 1:2000 in PBS containing 0.2% FSG) and then incubated with secondary antibody for 45 minutes in the dark. To stain cell nuclei, a final concentration of 1 μg/ml DAPI was added for 5 minutes. Coverslips were then washed three times in 0.2% FSG in PBS, before mounting onto slides with ProLong™ Gold Antifade mountant (Thermo Fisher Scientific, P10144). A DeltaVision/GE Healthcare OMX optical microscope (version 4) with oil-immersion objective (60× NA 1.42, PlanApochromat Olympus) was used for wide-field and SIM immunofluorescence image acquisition. Deconvolution and image registration (for alignment of SIM images) was carried out using the DeltaVision OMX softWoRx 6.0 software. Analysis of microscopy images was carried out using ImageJ (NIH, Bethesda, MD, USA). Four regions of interest (ROI) of equal size were drawn within each transfected cell: two within the nucleus and two within the cytoplasm. Intensity measures were averaged for the nucleus and divided by the average intensity for the cytoplasm.

We acknowledge Addgene for the STAT92E–GFP and pAc-sgRNA-Cas9 constructs. Images were collected in the Wolfson Light Microscopy Facility using a Nikon wide-field camera, funded by MRC Shima award (MR/K015753/1). Mass spectrometry was performed in the University of Sheffield, Faculty of Science biOMICs Facility.

The peer review history is available online at https://jcs.biologists.org/lookup/doi/10.1242/jcs.246199.reviewer-comments.pdf