Graf regulates hematopoiesis through GEEC endocytosis of EGFR

Epidermal growth factor receptor (EGFR) family members play pivotal roles in regulating cell proliferation and differentiation during animal development (Avraham and Yarden, 2011; Shilo, 2003). Their aberrant expression or activation is strongly associated with the etiology of several human epithelial cancers (Patel and Leung, 2012). Ligand binding to EGFR induces receptor dimerization and activation, which subsequently activates several downstream signaling cascades, including the Ras/mitogen-activated protein kinase (MAPK) and phosphatidylinositol 3-kinase (PI3K)/Akt pathways (Avraham and Yarden, 2011). In Drosophila, four ligands [i.e. Spitz (Spi), Karen, Gurken and Vein] activate a single EGFR to regulate many aspects of development, primarily via the canonical Ras-MAPK pathway (Shilo, 2003).

Ligand-induced activation of EGFR also triggers its rapid internalization to early endosomes, followed by either recycling to the plasma membrane or lysosomal degradation (Sorkin and Goh, 2009). Interestingly, two distinct modes of endocytosis differentially influence the fate of internalized EGFRs (Sigismund et al., 2008). Clathrin-dependent (CD) endocytosis, which functions at all EGF concentrations, leads to EGFR recycling and prolonged signaling. In contrast, clathrin-independent (CI) endocytosis, which is additionally activated under conditions of high EGF, is primarily linked to receptor degradation and signal attenuation. Although CI endocytosis of EGFR is known to be associated with Cbl-mediated receptor ubiquitylation at the plasma membrane (Sigismund et al., 2013), its precise molecular features remain poorly defined.

One prevalent type of CI endocytosis, which mediates the uptake of various GPI-anchored proteins (GPI-APs) as well as bulk fluid, involves the formation of pleiomorphic tubular intermediates known as clathrin-independent carriers (CLICs) (Kirkham et al., 2005; Sabharanjak et al., 2002). Once formed, CLICs undergo homotypic fusion to form glycosylphosphatidylinositol (GPI)-enriched endocytic compartments (GEECs), which then fuse with early/sorting endosomes derived from the CD endocytic system (Gupta et al., 2009; Kalia et al., 2006). This type of CI endocytosis has been shown to depend critically on the activity of the small GTPase Cdc42 and on GTPase regulator associated focal adhesion kinase 1 (GRAF1) (Lundmark et al., 2008; Sabharanjak et al., 2002).

GRAF1 contains N-terminal BIN/amphiphysin/Rvs (BAR) and pleckstrin homology (PH) domains that work together to generate and/or stabilize CLIC/GEEC endocytic membranes (Lundmark et al., 2008). In addition, this protein harbors a central RhoGAP domain that displays GAP activity toward Cdc42, raising the possibility that GRAF1 coordinates membrane remodeling and Cdc42-dependent actin polymerization to facilitate GEEC endocytosis. Apart from this role in endocytosis, GRAF1 has been implicated in regulation of myeloid cell development. Mutations and deletions in the human GRAF1 gene are associated with acute myeloid leukemia (AML) or myelodysplastic syndrome (MDS) (Borkhardt et al., 2000). Furthermore, the GRAF1 promoter is abnormally methylated in bone marrow samples from individuals with AML or MDS (Bojesen et al., 2006; Qian et al., 2011), reducing its expression (Qian et al., 2010). Despite these studies, it is unclear whether the two seemingly distinct roles of GRAF1 in hematopoiesis and GEEC endocytosis are related.

The Drosophila hematopoietic system provides a simple genetic model for the investigation of molecular aspects of myeloid cell development. Drosophila blood consists of only a few types of blood cells that functionally resemble the mammalian myeloid lineage. In addition, several key mechanisms controlling hematopoiesis are evolutionarily conserved (Crozatier and Vincent, 2011; Evans et al., 2003). Hematopoiesis in Drosophila occurs in two phases during development (Evans et al., 2003). The first phase takes place within the embryonic head mesoderm (Lebestky et al., 2000; Tepass et al., 1994) and produces two major classes of blood cells or hemocytes: macrophage-like plasmatocytes involved in phagocytosis and megakaryocyte-like crystal cells involved in melanization. The second phase of hematopoiesis occurs in larval lymph glands (LGs), which do not release differentiated plasmatocytes and crystal cells into circulation until the onset of pupariation (Holz et al., 2003). Lamellocytes, the final hemocyte type, are rarely found in normal healthy larvae but differentiate in response to parasitic infections (Lanot et al., 2001).

To understand the role of the GRAF protein family in blood cell development, we have studied the Drosophila GRAF1 ortholog Graf. Graf is expressed at high levels in plasmatocytes, where it plays an essential role in GEEC endocytosis. We provide evidence that Graf functions cell-autonomously to restrain plasmatocyte proliferation by inhibiting EGFR signaling. Consistent with this, EGFR signaling has been shown to stimulate plasmatocyte proliferation (Asha et al., 2003; Zettervall et al., 2004). Notably, GEEC endocytosis acts as a major pathway for EGFR internalization at high Spi and is preferentially associated with receptor degradation and signal attenuation. Finally, we find that high Spi treatment induces EGFR ubiquitylation mediated by Drosophila Cbl (D-Cbl), and that this modification subsequently promotes receptor association with Graf. This is the first demonstration of a role for Graf-dependent GEEC endocytosis in downregulating a specific signaling pathway and reveals a molecular mechanism underlying the recruitment of ubiquitylated cargoes to this CI endocytic pathway.

A BLAST search using the human GRAF1 sequence identified a single Drosophila homolog: Graf (CG8948). Conversely, a search using Drosophila Graf showed the closest matches to three members of the human GRAF family of RhoGAPs (GRAF1-GRAF3). The Drosophila Graf protein has an identical domain organization (Fig. 1A) and overall 43-46% identity to its human homologs. The Graf gene spans ∼9.0 kb of genomic DNA and contains another gene (i.e. CG8260) within its second intron (Fig. 1B).

To generate Graf-null mutants, we imprecisely excised a P-element (G275) inserted in the Graf transcription unit (Fig. 1B). Sequence analysis identified a deletion (Graf¹) spanning the entire first exon and part of the first intron. Homozygous Graf¹ flies and heterozygotes over the deficiency Df(1)BSC756 (hereafter referred to as Df) are null for Graf expression but express normal levels of the adjacent genes mRpL3 and CG8260 (Fig. 1C). Homozygous Graf¹ flies are viable and fertile.

To characterize the Graf protein, we raised an antibody against a Graf peptide (see Materials and Methods). In western blots of extracts derived from hemocytes of wild-type third instar larvae, this antibody detected a single major band of 110 kDa, the expected molecular weight of Graf, which was absent in extracts of Graf¹/Df mutants (Fig. 2A). To prove this band corresponds to Graf, we transiently overexpressed the full-length protein in Drosophila S2R+ cells. Levels of the 110 kDa product were higher in extracts of transfected cells than in non-transfected cells (Fig. 2A).

To explore Graf expression in the embryonic hematopoietic system, we first analyzed wild-type and Graf¹/Df embryos expressing a membrane-tethered GFP (UAS-mCD8-GFP) under the control of the hemocyte driver srp-GAL4. Double staining of wild-type embryos with anti-Graf and anti-GFP antibodies revealed Graf expression in the developing gut from stage 12 (data not shown), and beginning with stage 13, in migrating Srp-GFP⁺ plasmatocytes (Fig. 2B,C). Interestingly, Graf had a punctate distribution in embryonic plasmatocytes (Fig. 2B,C; enlargements). Although we observed some non-specific staining in the trachea and pharynx, we were unable to detect anti-Graf signal in the gut or Srp-GFP⁺ plasmatocytes of Graf¹/Df embryos (Fig. 2D), confirming the specificity of anti-Graf staining. To examine Graf expression in embryonic crystal cells, we also performed a similar experiment with wild-type and Graf¹/Df embryos expressing UAS-mCD8-GFP under the control of the crystal cell-specific lozenge (lz)-GAL4 driver (Lebestky et al., 2000). We observed punctate anti-Graf signal in Lz-GFP⁺ crystal cells clustered around the proventriculus of wild-type but not Graf¹/Df embryos (Fig. 2E and data not shown). Finally, to further assess the subcellular localization of Graf puncta, we analyzed GFP-positive hemocytes isolated from embryos carrying UAS-mCD8-GFP and srp-GAL4 or lz-GAL4. Additional nuclear staining with DAPI clearly revealed that Graf puncta had a cytoplasmic distribution in Srp-GFP⁺ and Lz-GFP⁺ hemocytes (Fig. 2F,G).

Next, we explored Graf expression in the larval hematopoietic system. To do this, we doubly stained hemocytes from wild-type and Graf¹/Df third instar larvae carrying both lz-GAL4 and UAS-mCD8-GFP with anti-Graf and an antibody against plasmatocyte-specific Nimrod C1 (NimC1). We found that Graf was expressed in a punctate pattern in most NimC1⁺ plasmatocytes and Lz-GFP⁺ crystal cells from wild-type larvae (Fig. 2H). Anti-Graf signal was dramatically decreased in NimC1⁺ or Lz-GFP⁺ hemocytes from Graf¹/Df larvae (Fig. 2H). We also observed specific Graf expression throughout the medullary and cortical zones of the primary LG lobe but not in the posterior signaling center (PSC) (Fig. S1).

To assess the role of Graf in the regulation of hematopoiesis, we compared hemocyte levels in wild-type and Graf¹/Df third instar larvae. We found approximately threefold higher levels of circulating and total (i.e. circulating plus sessile) hemocytes in Graf¹/Df larvae than in wild-type controls (Fig. 3A,B; Fig. S2A). Using the hemocyte reporter HmlΔ-DsRed (Makhijani et al., 2011), we confirmed that the sessile compartment of Graf¹/Df larvae also contains increased levels of hemocytes (Fig. 3C). Finally, Graf¹/Df larvae had 30% more hemocytes in the primary LG lobe (Fig. S2B). Thus, hemocyte numbers are increased in all hematopoietic compartments of Graf mutant larvae.

We were able to ascribe the increase in the larval hemocytes to a lineage-specific increase in plasmatocytes rather than other hemocyte types. First, Graf¹/Df larvae displayed a threefold increase in the concentration of circulating hemocytes labeled by the plasmatocyte lineage marker crq-GAL4, UAS-mCD8-GFP (Fig. 3D). However, loss of Graf did not significantly alter the concentration of circulating hemocytes labeled by the crystal cell lineage marker lz-GAL4, UAS-mCD8-GFP (Fig. 3D). Second, Graf¹/Df larvae had similar numbers of sessile crystal cells as wild-type larvae (Fig. S2C). Third, the percentage of lamellocytes, marked by the L1 antigen, was not significantly different between wild-type and Graf¹/Df mutant larvae (wild type, 0.1±0.01%; Graf¹/Df, 0.15±0.03%; P=0.11). Finally, plasmatocyte-specific expression of UAS-Graf-HA using crq-GAL4 fully rescued the hemocyte overproduction phenotype of Graf¹/Df larvae (Fig. 3B; Fig. S2A).

To confirm the cell type-specific effect of Graf on hematopoiesis, we stained primary LG lobes for NimC1, Lz, L1 and the PSC marker Antennapedia (Antp). Compared with wild-type controls, Graf¹/Df mutants show a significant increase in the proportion of NimC1⁺ plasmatocytes in the primary LG (Fig. S2D). However, the populations of Lz⁺ crystal cells and Antp⁺ PSC cells were not affected in the Graf mutants (Fig. S2E,F). Moreover, as in wild type, the number of L1⁺ lamellocytes per lobe remained low in Graf¹/Df mutant larvae (wild-type, 0.8±0.27; Graf¹/Df, 0.9±0.22; P=0.77). Thus, hemocyte overproduction in the Graf mutant LG is also due to a selective increase in the plasmatocyte population.

We next asked whether this Graf phenotype is due to an increase in hemocyte proliferation. Staining total hemocytes using an anti-phospho-histone H3 (pH3) antibody revealed a significantly higher proportion of mitotically active (pH3-positive) hemocytes in Graf¹/Df larvae than in wild-type controls (1.9- and 4.2-fold in second and third instar larvae, respectively; Fig. 3E,F). We also quantified hemocyte proliferation using S/G2/M-Green fucci, a fluorescent ubiquitylation-based cell cycle indicator (Makhijani et al., 2011; Sakaue-Sawano et al., 2008). Indeed, Graf mutations increased the proportion of fucci-positive hemocytes to a similar extent (1.7- and 2.6-fold in second and third instar larvae, respectively; Fig. S2G,H). Thus, the overabundance of plasmatocytes in Graf mutants can be attributed primarily to increased cellular proliferation.

Elevated EGFR/Ras/MAPK signaling has been shown to cause plasmatocyte overproliferation without affecting other hemocyte lineages (Zettervall et al., 2004). We therefore hypothesized that the plasmatocyte overproliferation phenotype of Graf-null mutants was associated with abnormal EGFR signaling. To test this, we first examined levels of EGFR and its downstream signaling components in primary Graf¹/Df hemocytes. Staining with an anti-EGFR antibody revealed upregulation of EGFR expression in Graf¹/Df hemocytes compared with wild-type controls (Fig. 4A,B). We also observed parallel elevation of MAPK activation by staining for diphospho-ERK (dp-ERK), a downstream target of EGFR signaling (Fig. 4A,B). In sharp contrast, levels of phospho-AKT (p-AKT), another downstream target, remained normal (Fig. 4B). We confirmed this selective increase in dp-ERK in Graf mutants by western blot analysis of larval extracts (Fig. 4C,D). Thus, Graf mutant hemocytes display aberrant upregulation of EGFR/MAPK signaling.

We next looked for genetic interactions between Graf and components of the EGFR signaling pathway. Loss of one copy of Egfr, Ras85D/Ras1 or the MAPK gene rolled (rl) did not affect the level of circulating hemocytes in the wild-type background, but partially or completely suppressed the hemocyte overproduction observed in Graf mutants (Fig. 4E). Moreover, hemocyte overproduction in Graf was further suppressed by removing both copies of rl (Fig. 4E). These data indicate that hemocyte overproduction in Graf requires EGFR signaling and further that Graf restrains plasmatocyte proliferation by inhibiting EGFR signaling.

Finally, we examined genetic interactions between Graf and spi, the latter which encodes the principal ligand for EGFR in most tissues (Rutledge et al., 1992). In a control experiment, we first examined the role of Spi in hemocyte development. Overexpression of UAS-sSpi-GFP (i.e. a secreted Spi::GFP fusion protein) using spi-GAL4 or the pan-neuronal driver C155-GAL4 potently increased hemocyte production (Fig. 4F), phenocopying EGFR gain of function (Zettervall et al., 2004). In contrast, expression of UAS-spi^RNAi with the same GAL4 drivers had the opposite effect, mimicking the rl mutation (Fig. 4F). Importantly, loss of one copy of spi, which had no effect on hemocyte abundance alone, fully suppressed the hemocyte overproduction observed in Graf mutants (Fig. 4G). These results indicate that hemocyte overproduction in Graf requires EGFR activation.

Mammalian GRAF1 is an essential component of the GEEC endocytic pathway (Lundmark et al., 2008). We confirmed that this role is conserved in Drosophila Graf. Not only was Graf prominently localized to CLIC/GEEC endocytic membranes in Drosophila primary hemocytes and S2R+ cells (Fig. S3A-G), but Graf was also required for efficient fluid-phase uptake (Fig. S3H,I).

How does the endocytic protein Graf attenuate EGFR activation? In a variety of mammalian cells, low EGF (≤2 ng/ml) treatment causes clathrin-dependent (CD) endocytosis of EGFR and subsequently its recycling to the plasma membrane, whereas high EGF also induces clathrin-independent (CI) endocytosis and degradation of the receptor (Sigismund et al., 2008, 2005). We therefore hypothesized that Graf-dependent GEEC endocytosis plays an important role in endocytic downregulation of EGFR under conditions of high EGF. To test this, we used several approaches.

We first examined the role of GEEC endocytosis in internalizing EGFR at low and high Spi. To do so, we stimulated primary larval hemocytes with 4 or 30 ng/ml sSpi-GFP (i.e. ∼1.3 or 10 ng/ml sSpi) and quantified the amount of ligand internalized for up to 2 min by anti-GFP immunostaining. The amount of internalized ligand in wild-type hemocytes stimulated with high sSpi-GFP was significantly greater than in wild-type hemocytes stimulated with low sSpi-GFP (Fig. 5A,B). Importantly, loss of Graf or knockdown of Arf1 and Arf-GEF (guanine nucleotide exchange factor) Garz, two well-established regulators of GEEC endocytosis (Gupta et al., 2009; Kumari and Mayor, 2008), reduced ligand internalization by ∼36-50% under high sSpi-GFP, but had no effect under low sSpi-GFP (Fig. 5B). In sharp contrast, knockdown of clathrin heavy chain (Chc) strongly impaired ligand internalization only under low sSpi-GFP (Fig. 5B). Finally, consistent with a previous report demonstrating that GEEC endocytosis in hemocytes is dynamin independent (Guha et al., 2003), treatment of hemocytes with dynasore did not affect sSpi-GFP internalization under high sSpi-GFP. In control experiments, dynasore treatment strongly impaired ligand internalization under low sSpi-GFP (Fig. 5B). Collectively, these results demonstrate that ligand-activated EGFR can be internalized via both CD and GEEC endocytosis and that high Spi stimulation skews the partitioning of EGFR endocytosis toward the latter route.

To corroborate the above conclusions, we treated primary wild-type hemocytes with low or high sSpi-GFP and visualized the early steps of ligand internalization compared with endogenous Graf and Chc. Immediately after low sSpi-GFP stimulation (2 min), the internalized ligand appeared intracellularly in a punctate pattern and these puncta colocalized more extensively with Chc than with Graf (27% versus 9%, respectively; Fig. 5D). However, under high sSpi-GFP conditions, the internalized ligand showed the opposite behavior (13% with Chc versus 30% with Graf; Fig. 5C,D), supporting that high Spi stimulation skews the partitioning of EGFR endocytosis toward the GEEC pathway.

We then examined the effect of high Spi treatment on GEEC endocytosis-mediated fluid phase uptake. We pretreated primary hemocytes from wild-type larvae with both FITC-labeled dextran (FITC-Dex; a GEEC endocytosis marker) and low or high sSpi-GFP at 4°C for 30 min and subsequently allowed uptake of FITC-Dex at 25°C for 2 min. Both the total amount of FITC-Dex internalized and the number of intracellular puncta double positive for Graf and FITC-Dex in high Spi-treated hemocytes were not significantly different from those observed in low Spi-treated hemocytes (Fig. 5E,F), indicating that high Spi induces GEEC endocytosis of EGFR without affecting the activity of the GEEC pathway.

Next, we investigated whether GEEC endocytosis is associated with efficient sorting of internalized EGFR to lysosomes. We treated primary larval hemocytes with low or high Spi and visualized ligand internalized for up to 30 min in comparison with the lysosome marker Lysotracker. The extent of sSpi-GFP-Lysotracker colocalization in wild-type hemocytes treated with high sSpi-GFP was significantly greater than in wild-type hemocytes treated with low sSpi-GFP (51% versus 17%; Fig. 5G,H). This high Spi-induced effect was strongly abolished by loss of Graf, Arf1 or Garz but not by loss of Chc. In control experiments, malelylated bovine serum albumin (a CD endocytosis marker) was normally internalized and targeted to lysosomes in Graf mutant hemocytes under high Spi (data not shown), confirming that the effect of interfering GEEC endocytosis is not due to global defects in endocytic protein trafficking to lysosomes. Under low sSpi-GFP, loss of Chc resulted in a small but significant reduction in sSpi-GFP-Lysotracker colocalization, whereas loss of Graf, Arf1 or Garz had no effect (Fig. 5H). Thus, GEEC endocytosis, but not CD endocytosis, of EGFR is associated with its efficient sorting to lysosomes.

Finally, we directly investigated the impact of GEEC endocytosis on EGFR degradation at high Spi. When S2R+ cells overexpressing Flag-EGFR were stimulated with 30 ng/ml sSpi-GFP, we observed robust Flag-EGFR degradation and enhanced MAPK activation beginning within 5 min that was almost completely abolished by 60 min (Fig. 5I,J). However, loss of Graf or Arf1, but not Chc, strongly impaired ligand-induced Flag-EGFR degradation, leading to more sustained activation of ERK (Fig. 5I,J). Thus, GEEC endocytosis, but not CD endocytosis, of EGFR is indispensable for its efficient degradation under the condition of high Spi.

In mammalian cells, CI endocytosis of EGFR at high EGF requires its ubiquitylation (Sigismund et al., 2013, 2005), which is mediated by the E3 ligase Cbl (Levkowitz et al., 1999, 1998). To investigate the mechanistic link between EGFR ubiquitylation and GEEC endocytosis, we initially assessed the extent of receptor ubiquitylation in S2R+ cells stimulated with either low or high Spi. EGFR ubiquitylation was detected only under high Spi, as demonstrated by immunoblotting using the P4D1 antibody that recognizes ubiquitin (Ub) monomers and polymers (Fig. S4A). Given the effect of high Spi on GEEC endocytosis, this result establishes a positive correlation between EGFR ubiquitylation and GEEC endocytosis of the receptor.

We then analyzed the relevance of Drosophila Cbl (D-Cbl) to GEEC endocytosis of EGFR. For this, we prepared primary hemocytes from larvae carrying a hypomorphic D-cbl mutation (D-cbl^KG03080) and stimulated them for 2 min with high Spi (i.e. 30 ng/ml sSpi-GFP). The amount of internalized sSpi-GFP in D-cbl^KG03080 hemocytes was decreased by 37% compared with control crq-GAL4/+ hemocytes (Fig. 6A,B). Importantly, this defect was concurrent with a substantial decrease in sSpi-GFP-Graf colocalization (18% in D-cbl^KG03080 versus 31% in crq-GAL4/+; Fig. 6C,D), demonstrating a reduction in sSpi-GFP internalization via GEEC endocytosis. Expression of the longer isoform of D-Cbl (D-CblL) in D-cbl^KG03080 hemocytes simultaneously increased sSpi-GFP internalization and sSpi-GFP-Graf colocalization to control levels, while similar expression of an E3 ligase-defective D-CblLΔ70Z mutant did not (Fig. 6B,D). These results support the importance of the E3 ligase activity of D-Cbl for GEEC endocytosis of EGFR.

Next, we directly assessed the role of D-Cbl-mediated receptor ubiquitylation in GEEC endocytosis of EGFR. To do so, we first demonstrated that the Y1271F point mutation in EGFR significantly impaired its ability to recruit D-Cbl and undergo ubiquitylation in response to high Spi (Fig. S4B-E). We then examined the impact of this mutation on the GEEC endocytosis of EGFR. S2R+ cells expressing Flag-EGFR or Flag-EGFR-Y1271F in combination with exogenous GFP-GPI (a GEEC pathway marker) were analyzed for early trafficking steps of transfected EGFR protein. After prelabeling surface Flag-EGFR and GFP-GPI with anti-Flag and anti-GFP antibodies at 4°C, we stimulated cells with high sSpi-HA (10 ng/ml) at 25°C for 2 min to allow Flag-EGFR endocytosis. A major fraction (∼66%) of the internalized Flag-EGFR pool colocalized with GFP-GPI-containing endosomes, whereas only 24% of internalized Flag-EGFR-Y1271F receptors did so (Fig. S4F,G). We also examined the impact of the Y1271F mutation on EGFR degradation. Under high sSpi-HA, Flag-EGFR-Y1271F showed less degradation than wild-type Flag-EGFR (Fig. S4H,I), confirming the notion that GEEC endocytosis of EGFR is indispensable for efficient receptor degradation. Thus, D-Cbl-mediated ubiquitylation of EGFR is required for its GEEC endocytosis and subsequent degradation.

To investigate the in vivo role of D-Cbl in downregulating EGFR, we first examined hemocyte development in D-cbl^KG03080 mutants. We found significantly higher levels of circulating hemocytes in homozygous D-cbl^KG03080 larvae than in wild-type larvae (Fig. 6E,F). In addition, D-cbl^KG03080 larvae showed a 1.8-fold increase in the fraction of pH3-positive hemocytes (Fig. 6G). Plasmatocyte-specific expression of D-CblL, but not D-CblLΔ70Z, rescued the hemocyte overproduction observed in D-cbl^KG03080 mutants (Fig. 6F), underscoring the importance of the E3 ligase activity of D-Cbl in regulating plasmatocyte proliferation. As an additional approach, we also examined genetic interactions between D-cbl and Egfr (Fig. 6H). Plasmatocyte-specific overexpression of EGFR induced hemocyte overproliferation. Importantly, this phenotype was strongly suppressed by overexpressing D-Cbl, which had no effect on plasmatocyte proliferation in the wild-type background. In contrast, overexpression of D-CblLΔ70Z failed to suppress the EGFR overexpression phenotype. These results clearly establish a role for the E3 ligase activity of D-Cbl in EGFR downregulation during plasmatocyte development.

To further understand the mechanism underlying GEEC-mediated EGFR internalization, we first tested whether Graf interacts physically with EGFR. In GST pull-down experiments, GST-Graf, but not GST alone, precipitated Flag-EGFR from S2R+ cell lysates (Fig. 7A). We were also able to co-immunoprecipitate Flag-EGFR and endogenous Graf from S2R+ cell lysates using an anti-Flag antibody (Fig. 7B,C). Importantly, the EGFR-Graf interaction was enhanced by stimulating S2R+ cells with high sSpi-HA (Fig. 7A-C). This interaction was direct, as demonstrated by GST pull-down experiments using a GST fusion to the EGFR intracellular domain (GST-EGFR-IC) and purified His₆-Graf (Fig. 7D,E).

Next, we addressed the issue of whether D-Cbl-mediated EGFR ubiquitylation is required for the Spi-induced interaction of EGFR with Graf. We found that loss of D-Cbl, which had no effect on the basal EGFR-Graf interaction by itself, completely abolished the effect of high Spi treatment on the EGFR-Graf interaction (Fig. 7A-C). In addition, the Y1271F point mutation in EGFR also blocked the high Spi-induced EGFR-Graf interaction (Fig. 7A-C). Finally, the addition of a single ubiquitin molecule to GST-EGFR-IC enhanced its interaction with His₆-Graf (Fig. 7D,E). In a control experiment, GST-Ub, but not GST alone, also bound to His₆-Graf, revealing an intrinsic ability of Graf to bind Ub. These results support a model in which D-Cbl-mediated receptor ubiquitylation facilitates GEEC endocytosis of EGFR by promoting its physical interaction with Graf.

The above model predicts that Graf acts at the plasma membrane to recruit EGFR to GEEC endocytosis. However, in line with previous work demonstrating that the assembly of mammalian GRAF1 at the cell surface is very transient (Francis et al., 2015), Graf was barely detected at the cell surface in hemocytes and S2R+ cells (Fig. S3A-E). We therefore sought to stabilize the localization of Graf to nascent endocytic structures during EGFR internalization. To this end, we transfected S2R+ cells with Graf-GFP and Flag-EGFR in combination with a constitutively active Cdc42 mutant (Cdc42-Q61L), which has been shown to result in accumulation of assembled GRAF1 at the plasma membrane (Francis et al., 2015). Live cells were prelabeled for surface Flag-EGFR at 4°C and stimulated with high sSpi-HA at 25°C. After 2 min, the localization of Graf-GFP was visualized in comparison with that of internalized Flag-EGFR and sSpi-HA by immunocytochemistry. Under these conditions, Graf-GFP localized together with internalizing Flag-EGFR and sSpi-HA to punctate or tubular structures connected to the cell surface (Fig. 7F), supporting our conclusion that Graf recruits EGFR to GEEC endocytosis at the plasma membrane.

Deletions, truncations and mutations in the human GRAF1 gene are associated with myeloid malignancies such as AML and MDS (Borkhardt et al., 2000; Panagopoulos et al., 2004; Wilda et al., 2005), but little is known about the role of GRAF family members in blood cell development. Here, we provide evidence that Graf, the only fly ortholog of mammalian GRAF proteins, acts cell-autonomously to regulate the proliferation of macrophage-like plasmatocytes. First, Graf is expressed in plasmatocytes. Second, Graf mutant larvae have more plasmatocytes than wild-type controls owing to increased cellular proliferation. Third, this Graf mutant phenotype is fully rescued by plasmatocyte-specific expression of Graf. Our data further imply that Graf restrains plasmatocyte proliferation by inhibiting the EGFR-Ras-MAPK pathway. Not only do Graf mutant hemocytes show elevated levels of EGFR and activated ERK, but hemocyte overproliferation in Graf mutants requires both the EGFR ligand Spi and the EGFR-Ras-MAPK cascade.

Unlike plasmatocytes, the populations of crystal cells and lamellocytes in both circulation and the LG are not significantly affected by loss of Graf. This finding is consistent with a previous report demonstrating that overexpression of wild-type EGFR in circulating hemocytes induces selective overproliferation of plasmatocytes without significantly affecting other hemocyte lineages (Zettervall et al., 2004). However, in another study, overexpression of a constitutively active Egfr mutant in both circulating hemocytes and the LG cortical zone was found to potently induce the generation of lamellocytes (Sinenko et al., 2011). The discrepancy with regard to lamellocyte differentiation could arise from differences in the timing and strength of signaling mediated by wild-type and constitutively active Egfr alleles. Another, not mutually exclusive, possibility is that discrete hemocyte populations respond to different thresholds of EGFR signaling activity.

Previous work has proposed that, upon parasitic infection, the PSC cells of the LG secrete Spi to induce lamellocyte differentiation in a non-cell-autonomous manner (Sinenko et al., 2011). However, it remains unknown whether the PSC cells also secrete Spi to induce plasmatocyte proliferation under normal developmental conditions. Our data raise the possibility that the nervous system is a physiological source of secreted Spi in the hemolymph. We find that pan-neuronal expression of Spi RNAi using C155-GAL4 is sufficient to reduce hemocyte numbers in circulation. Conversely, expression of secreted Spi with the same driver has the opposite effect on hemocyte numbers in circulation. Furthermore, spi-GAL4 is found to be highly active in a small subset of neurosecretory cells that innervate the ring gland, the major endocrine organ in Drosophila (S.K. and S.L., unpublished). It will be interesting in the future to determine whether the Spi section activity of the nervous system is controlled by developmental cues or infection.

Like human GRAF1, Drosophila Graf plays an important role in GEEC endocytosis. This raises the obvious issue of whether the two roles that Graf plays in regulating GEEC endocytosis and EGFR signaling are related. In mammalian cells, CI endocytosis of EGFR results in receptor degradation and signal attenuation in the presence of high ligand concentrations (Sigismund et al., 2008). Until now, the precise nature of the CI endocytic pathway mediating EGFR downregulation remained unclear. Here, we provide evidence that Graf-dependent GEEC endocytosis is a major CI endocytic route of EGFR internalization. Upon exposure to high concentrations of Spi, a substantial fraction of internalized Spi and EGFR colocalizes with Graf- or GFP-GPI-labeled CLICs/GEECs. Interfering with GEEC endocytosis by knocking down Graf or Arf1 strongly impairs this high-Spi-induced internalization of EGFR. We also provide evidence that GEEC endocytosis is essential for efficient EGFR degradation at high Spi. Loss of Graf or Arf1 significantly impairs high-Spi-induced EGFR degradation and signal attenuation. The EGFR-Y1271F mutant, which cannot be efficiently partitioned toward the GEEC pathway (see below), is much more resistant to high-Spi-induced degradation than wild-type EGFR. Our findings thus imply that Graf attenuates EGFR signaling by facilitating its GEEC endocytosis under conditions of high Spi (see Fig. 8).

In mammalian cells, ubiquitylation of EGFR by Cbl proteins seems to divert the receptor from CD to CI endocytosis at high EGF concentrations (Sigismund et al., 2013). Consistently, GEEC endocytosis of Drosophila EGFR also depends on Cbl-dependent receptor ubiquitylation. We find that high Spi treatment, which promotes the GEEC endocytosis of EGFR, potently induces EGFR ubiquitylation by the fly Cbl ortholog D-Cbl. In addition, the E3 ligase activity of D-Cbl is required for EGFR partitioning into the GEEC endocytic pathway. Furthermore, the ubiquitylation-defective EGFR-Y1271F mutant fails to undergo GEEC endocytosis at high Spi.

How are ubiquitylated EGFRs sorted preferentially to the GEEC endocytic pathway? We provide biochemical evidence that Graf itself participates in this sorting process via physical interactions with the ubiquitin (Ub) moiety on EGFR (Fig. 8). We show that Spi enhances the physical interactions of EGFR with Graf and that this enhancement requires D-Cbl recruitment to EGFR. We also show that the addition of a single Ub moiety to the EGFR intracellular domain enhances its ability to interact with Graf in vitro. Given this Graf/Ub-dependent sorting mechanism, it will be interesting in the future to determine whether other receptor tyrosine kinases (RTKs) subjected to D-Cbl-mediated ubiquitylation are internalized through Graf-dependent GEEC endocytosis, resulting in their degradation.

At present, the mechanism by which loss of GRAF1 causes AML or MDS remains unknown. Our Drosophila studies reveal that EGFR overexpression is the principal force driving the selective overproliferation of a myeloid-like lineage (plasmatocytes) in Graf-null mutant larvae, suggesting a potential link between EGFR dysregulation and myeloid malignancies. Consistent with this possibility, one recent clinical study reported elevated expression of EGFR protein in ∼33% of primary AML samples and associated this elevated expression with poor clinical outcomes (Sun et al., 2012). In addition, another clinical study reported EGFR mutations in primary malignant cells from individuals with MDS (Bejar et al., 2011). Furthermore, administration of the EGFR inhibitor erlotinib can cause complete remission of AML in individuals with concomitant non-small-cell lung cancer, although its off-target effects on Jak2 were invoked to explain this anti-neoplastic activity (Chan and Pilichowska, 2007; Pitini et al., 2008). These clinical data, combined with our current findings, suggest that EGFR dysregulation may contribute to the development and progression of a subset of myeloid malignancies. Future experiments will further address this important issue.

Flies were maintained at 25°C. The fly strains used and the generation of transgenic lines are detailed in the supplementary Materials and Methods.

A full-length cDNA for Graf was amplified by reverse transcription (RT)-PCR of total RNA extracted from Drosophila S2R+ cells and then cloned into the pGEM-T Easy vector (Promega). For transgenic rescue experiments, a full-length cDNA for Graf was subcloned into pcDNA3.1-HA, a derivative of the pcDNA3.1(+) vector (Invitrogen), and then subcloned again with the C-terminal HA-tag sequence into the pUAST vector (Brand and Perrimon, 1993) to produce UAS-Graf-HA. For expression of Graf-GFP in S2R+ cells, the Graf cDNA insert was subcloned into the pAc-EGFP vector (Invitrogen). For expression of GST-Graf and His₆-Graf in E. coli, the full-length Graf cDNA was subcloned into the pGEX6P1 vector (GE Healthcare) or the pRSET A vector (Invitrogen), respectively.

For expression of additional genes in S2R+ cells or E. coli, their cDNAs were subcloned into the pUAST, pAc5.1, or pGEX6P1 vector. See the supplementary Materials and Methods for the construction of UAS-Myc-arf1, UAS-Myc-Rab5, UAS-Myc-Cdc42-Q61L, UAS-Flag-Egfr, UAS-Flag-Egfr-Y1271F, pAc-sSpi-GFP, pAc-sSpi-HA, pAc-HA-Ub and pGEX6P1-Egfr-IC-Ub. The UAS-GFP-GPI construct was obtained from the Drosophila Genomics Resource Center.

To characterize the Graf excision lines, genomic DNA was isolated from homozygous animals and used as a template for PCR using the following primers: 5′-TACAGGATCATACCTGTGACTAC-3′ and 5′-GCTTGACCATCGATTATTTCTGG-3′. PCR products were subcloned and subjected to DNA sequencing to identify the precise breakpoints of each Graf deletion. RT-PCR was performed to assess the effect of the Graf¹ mutation on the expression of Graf, as detailed in the supplementary Materials and Methods.

For RNAi experiments in S2R+ cells, Graf, chc, arf1 and D-cbl dsRNAs were generated by in vitro transcription of their cognate DNA templates containing T7 promoters at both ends, as previously described (Lee et al., 2007). See the supplementary Materials and Methods for the generation of DNA templates for in vitro transcription.

Drosophila S2R+ cells were maintained at 25°C in Schneider's medium (Invitrogen) supplemented with 10% heat-inactivated (30 min, 55°C) fetal bovine serum (FBS) and antibiotics. S2R+ cells were transfected using Cellfectin (Invitrogen), according to the manufacturer's instructions. Typically, 10⁶ cells were transfected in serum-free medium with 2 µg of plasmid DNA or 5 µg of double-stranded RNA (dsRNA). Expression of all UAS plasmids was driven by co-transfection of RK241, an actin 5C-GAL4 plasmid.

For production of Spi-conditioned medium, transfection of S2R+ cells with pAc-sSpi-HA or pAc-sSpi-GFP was performed in serum-free medium for 4 h. Transfected cells were incubated for 48 h in serum-containing medium and then for 24 h in fresh serum-free medium. Levels of secreted sSpi-GFP and sSpi-HA in the conditioned medium were measured by ELISA (Cell Biolabs) and western blot analysis using a multiple-epitope tag (GenScript). To treat primary larval hemocytes and S2R+ cells, the conditioned medium was diluted with serum-free Schneider's medium to 4 ng/ml or 30 ng/ml sSpi-GFP (to obtain a final concentration of about 1.3 ng/ml or 10 ng/ml sSpi) or 10 ng/ml sSpi-HA.

Antibodies used for immunostaining and western blotting are described in the supplementary Materials and Methods.

Western blotting, co-immunoprecipitation and GST pull-down assays were performed as described previously (Nahm et al., 2010), with some modifications. For details, see the supplementary Materials and Methods.

Fixed and permeabilized embryos, primary hemocytes, and S2R+ cells were stained with the indicated antibodies and imaged by confocal microscopy. For further details, see the supplementary Materials and Methods.

For sSpi-GFP internalization assays, primary hemocytes from early third instar larvae were incubated in conditioned medium containing 4 or 30 ng/ml of sSpi-GFP for 30 min at 4°C to allow the ligand to bind to surface EGFRs. In some experiments, 20 µM dynasore (Sigma) was also pretreated to inhibit dynamin. After washing with ice-cold PBS, hemocytes were incubated in serum-free Schneider's medium for 2 min at 25°C to allow internalization of sSpi-GFP. Hemocytes were immediately fixed on ice for 10 min with ice-cold 4% formaldehyde in PBS and permeabilized with PBST-0.2 for 10 min and then stained with an anti-GFP antibody and an appropriate fluorescent-conjugated secondary antibody. For each cell, a z stack of images was taken with a LSM 700 laser-scanning confocal microscope (Carl Zeiss) using a Plan Apo 63×1.4 NA oil objective. To quantify the amount of internalized sSpi-GFP, total fluorescence per cell was determined by integrating intracellular fluorescence on all planes of the z stack after correcting for background fluorescence.

For sSpi-GFP trafficking studies, primary hemocytes pretreated with 4 or 30 ng/ml of sSpi-GFP for 30 min at 4°C were incubated at 25°C for 2 min or 25 min. Hemocytes were immediately processed for staining with anti-GFP and anti-Graf or anti-Chc antibodies (2 min internalization) or further incubated with 1 µM Lysotracker (Molecular Probes) for 5 min (25 min internalization), prior to staining with an anti-GFP antibody. Fluorescence imaging was done with a LSM 700 laser-scanning confocal microscope (Carl Zeiss) using a Plan Apo 63×1.4 NA oil objective. For quantitative analysis of anti-GFP and Graf/Chc/Lysotracker colocalization, a z stack of two-dimensional images (0.35 µm thick) of cells was obtained through FITC (sSpi-GFP) and Cy3 (Graf, Chc, or Lysotracker) filter channels and deconvoluted using the AutoQuant software (Media Cybernetics). Quantitative analysis was performed on deconvoluted z stack images using the JACoP plug-in for ImageJ to determine Manders' coefficient M₁, which represents the ratio of the summed intensities of pixels from the green image for which the intensity on the red channel is above 0 to the total intensity in the green channel. Details on the analysis of endocytic probe internalization with primary hemocytes and EGFR trafficking studies in S2R+ cells can be found in the supplementary Materials and Methods.

S2R+ cells overexpressing Flag-EGFR were stimulated with 30 ng/ml sSpi-GFP for the indicated time and subjected to western blot analysis. For further details, see the supplementary Materials and Methods.

For hemocyte counting, late-wandering third instar larvae were staged using food containing 0.5% green household food dye as described previously (Zettervall et al., 2004). For further details, see the supplementary Materials and Methods.

The collected values were analyzed for normality with the D'Agostino-Pearson omnibus test and used for significance testing. To determine statistical significance, we performed the Student's t-test or one-way ANOVA followed by post hoc pairwise comparisons of means using Tukey-Kramer test.

We are grateful to the Bloomington, VRDC and Kyoto Drosophila stock centers, to the DSHB hybridoma bank, and to the following individuals for stocks and reagents: I. Ando, K. Bruckner, J. Jiang, M. Haenlin, E. Hafen, L.-M. Pai, and B.-Z. Shilo.