Glioblastoma is one of the most lethal forms of adult cancer, with a median survival of ∼15 mo. Targeting glioblastoma stem-like cells (GSCs) at the origin of tumor formation and relapse may prove beneficial. In situ, GSCs are nested within the vascular bed in tight interaction with brain endothelial cells, which positively control their expansion. Because GSCs are notably addicted to apelin (APLN), sourced from the surrounding endothelial stroma, the APLN/APLNR nexus has emerged as a druggable network. However, how this signaling axis operates in gliomagenesis remains underestimated. Here, we find that the glycoprotein GP130 interacts with APLNR at the plasma membrane of GSCs and arbitrates its availability at the surface via ELMOD1, which may further impact on ARF-mediated endovesicular trafficking. From a functional standpoint, interfering with GP130 thwarts APLNR-mediated self-renewal of GSCs ex vivo. Thus, GP130 emerges as an unexpected cicerone to the G protein–coupled APLN receptor, opening new therapeutic perspectives toward the targeting of cancer stem cells.
Glioblastoma multiforme (GBM), an aggressive adult primary brain tumor, remains one of the most pernicious malignancies. In defiance of intrusive treatments, including neurosurgery and combined radiation and chemotherapy with temozolomide, median survival has leveled off at ∼15 mo since the introduction of the so-called Stupp protocol more than a decade ago (Stupp et al., 2005). However, most patients inexorably relapse and shortly succumb to their disease. The intratumor and intertumor heterogeneity of GBM surfaced upon systematic profiling of patient samples (Verhaak et al., 2010; Lathia et al., 2015; Jin et al., 2017; Gimple et al., 2019). Within these multifarious masses exists a population of rare cells with stem-like properties termed glioblastoma stem-like cells (GSCs; Lathia et al., 2015; Gimple et al., 2019). Rather than possessing a molecular identity based on gene signature (i.e., expression and mutations), GSCs are phenotypically defined by their ability to self-renew and expand in vitro as nonadherent spheroid culture in mitogen-defined, serum-free medium. GSCs can also differentiate, escape from chemotherapy- and radiation-provoked damages, and initiate and propagate tumors upon serial implantations in immunocompromised mice (Singh et al., 2004; Lathia et al., 2015; Gimple et al., 2019). The expression of classical stem markers, such as NESTIN, sex-determining Y-box2, octamer binding transcription factor 4, and nanog homeobox, is nonetheless frequently associated with the GSC population. Because GSCs are believed to be at the origin of gliomagenesis, tumor expansion, and treatment failure, they represent an interesting opportunity for targeted therapies.
Comparable to neural stem cells, GSCs reside, at least in part, in a confined and protective microenvironment, which provides essential cues to maintain their self-renewal and dictates their fate (Shen et al., 2004; Calabrese et al., 2007; Gilbertson and Rich, 2007; Prager et al., 2019; Day et al., 2019). The neural stem cell niche notably integrates the blood vasculature, as neo-neurons arise along vessels in adulthood, and employs intermingled vacular endothelial growth factor (VEGF) signaling (Bao et al., 2006; Oka et al., 2007; Batchelor et al., 2007; Treps et al., 2016). Likewise, GSCs interact with endothelial cells and pericytes from the tumor vasculature (Calabrese et al., 2007; Cheng et al., 2013; Harford-Wright et al., 2017). Seminal work highlighted that NESTIN-positive brain tumor cells physically interface with endothelial cells (Calabrese et al., 2007). In the same vein, the endothelial secretome recapitulated in vitro the supportive action of endothelial cells on GSC subsistence (Galan-Moya et al., 2011; Harford-Wright et al., 2017; Jacobs et al., 2017). In this scenario, the localization of GSCs, in close proximity to endothelial cells, facilitates communication between these cells, delivering a paracrine supply of angiocrine factors essential to nurture GSCs. Few factors, such as basic fibroblast growth factor, have been definitively implicated in the endothelial-ignited communication toward GSCs (Takahashi et al., 1991; Batchelor et al., 2007; Loilome et al., 2009). Our recent mass spectrometry analysis of brain endothelial cell conditioned media identified the vasoactive peptide apelin (APLN) as critical for GSC maintenance and growth (Harford-Wright et al., 2017). Indeed, APLN could maintain GSC identity in the absence of all other mitogens, while the pharmacological blockade of the apelin receptor (APLNR; also known as APJ) drastically reduced tumor growth in experimental animal models, and sensitized GSCs to temozolomide treatment in vitro (Harford-Wright et al., 2017). This endogenous peptide, which may emanate from endothelial cells throughout diverse organs and tissues, engages the G protein–coupled receptor APLNR, and conducts a large repertoire of physiological, protective and regenerative actions, as exemplified in heart and skeletal muscles (Yang et al., 2015). In the context of GBM, APLN functions may deviate in order to guard the pool of GSCs. The potent role of APLN in tumorigenesis has been further elucidated in diverse experimental tumor models (Uribesalgo et al., 2019). Moreover, endothelial cell–produced APLN contributes to hematopoietic stem cell repopulation, which in turn supports vessel integrity, further emphasizing the intertwined connection between stem cells and vessels (Chen et al., 2019). A pillar of this reciprocal communication may thus be APLN.
Based on the essential role we recently ascribed to endothelial APLN in glioma progression via its cognate receptor, APLNR, we explored in depth the molecular mechanisms involved in APLN-based signaling in GBM. A recent study demonstrated that APLNR is linked to the Janus kinase (JAK) pathway in cancer cells and therefore suggests that APLN might prime atypical signaling, beyond G-protein coupling (Patel et al., 2017). Here, we report that the glycoprotein GP130, the transducing receptor for JAK-associated cytokine receptors, interacts with APLNR and orchestrates its availability at the plasma membrane for further ligand-dependent activation in GSCs. Our results help to better understand the molecular basis for tumor perversion of the physiological APLN/APLNR signaling axis in the course of glioblastoma progression.
APLNR interacts with GP130 at membrane microdomains at the surface of GSCs
We and others recently showed that GSCs are addicted to APLN secreted by surrounding endothelial cells, thereby placing the APLN/APLNR nexus as a druggable network in glioblastoma (Harford-Wright et al., 2017; Uribesalgo et al., 2019; Mastrella et al., 2019). Of note, the nonreceptor tyrosine kinase JAK1 that bridges the transducer cytokine receptor GP130 to the activation of the STAT (signal transducer and activator of transcription) transcription factors was found bound to a mutated version of APLNR in tumor cells (Patel et al., 2017). Although none of these recently identified APLNR mutations were found in the genomic DNA of three different glioblastoma patient–derived cells (see Materials and methods), the implication of GP130 in maintaining GSCs (Jacobs et al., 2017; Shi et al., 2017) prompted us to examine the connection between APLNR and GP130. First, we found that blocking JAK activity with ruxolitinib largely impaired self-renewal and sphere formation of GSCs in APLN-only mitogen-free (MF) medium (Fig. S1, A and B). This was however not the case in mitogen-supplemented medium (NS34; Fig. S1, A and B), suggesting a specific involvement of JAK in APLN-ignited signaling rather than a selective yet toxic effect of this drug. Similar results were obtained with filgotinib, an additional JAK inhibitor (Fig. S1 B). In line with this functional connection between the JAK pathway and the APLN cascade, confocal analysis revealed that APLNR and GP130 coalesce at the plasma membrane of GSCs (Fig. 1 A). Moreover, coimmunoprecipitation experiments demonstrated that GP130 binds to APLNR (Fig. 1 B). This was further confirmed with a proximity ligation assay, which highlighted that the interaction between GP130 and APLNR occurs within a maximum 30–40 nm distance (Fig. 1 C). Of note, this GP130/APLNR association was recapitulated in ectopic expression system, using transfected HEK293T cells (Fig. 1 D).
To gain insights into the distribution of the two receptors at the plasma membrane, confocal analysis was next performed upon manipulation of the plasma membrane topology. Strikingly, we observed that APLNR and GP130 staining were restricted to a crescent-shape portion of the plasma membrane in the absence of nonionic detergents (i.e., triton, saponin, and digitonin; Fig. 1, E and F; and Fig. S2). Likewise, cholesterol depletion using methyl-β-cyclodextrin (MBCD) resulted in a more homogeneous distribution of APLNR around the plasma membrane (Fig. 1 E). These membrane microdomains segregated in zones also enriched with glycophosphatidylinositol (GPI) anchors, as detected with fluorescein-labeled proaerolysin (FLAER), further reinforcing the idea of membrane clustering (Fig. 1 F). Finally, structured-illuminated microscopy detailed the intrabiomembrane organization of APLNR/GP130 nanoclusters (Fig. 1 G). The superresolution images showed that APLNR and GP130 indeed coalesce at a confined region at the surface of GSCs.
GP130 is required for APLN-mediated GSC expansion
We next assessed the functional impact of the GP130/APLNR association on the stem properties of GSCs. As expected (Harford-Wright et al., 2017), APLN alone promotes self-renewal of patient-derived GSCs, as monitored via limiting dilution assays (LDAs) and secondary spheroid assays (tumorspheres; Fig. 2, A and B). However, this was reduced upon neutralization of GP130 signaling and expression, with blocking antibodies and RNA interference, respectively (Fig. 2, B–D). To delve into the role of GP130 in APLN supporting action on patient-derived GSCs, GP130 was knocked out via CRISPR/Cas9-based gene editing (Fig. 2 E). Single clones were selected by fluorescence-activated cell sorting and genomic DNA was sequenced to identify bi-allelic mutants in the GP130 gene (Fig. 2 E). GP130 knockout (KO) was confirmed at the protein level on two independent clones (Fig. 2, F and G). In agreement with the obliteration of GP130-mediated signaling, the two KO clones exhibited a sharp reduction in STAT3 Y705 phosphorylation (Fig. 2 G; Guanizo et al., 2018), which reflects canonical JAK-mediated activation of this transcription factor downstream of GP130, in growing GSCs. It is noteworthy that STAT3 S727 phosphorylation was not affected, as this posttranslational modification might be caused by an alternative signaling cascade, such as mTOR (mechanistic target of rapamycin) and MAPK (Guanizo et al., 2018; Fig. 2 G). To next determine the importance of GP130 expression in APLN signaling, self-renewal capabilities were estimated with LDAs and spheroid assays (Fig. 2, H and I). We found that depletion of GP130 expression strongly impaired APLN-dependent tumorsphere formation. Previous findings had highlighted the contribution of the glycogen synthase kinase 3 β (GSK3β) activity in gliomagenesis and more precisely that GSK3β inactivation by S9 phosphorylation was associated with APLN inhibition (Zhou et al., 2016; Harford-Wright et al., 2017). In line with this, the level of inactive S9 phosphorylation was increased in GP130 KO cells, as compared with parental cells grown in APLN-only MF medium (Fig. 2 J). S6 phosphorylation engaged upon mTOR activation was, however, intact (Figs. 2 J and S3 A). Thus, GP130 expression licenses APLN to operate in GSCs.
GP130 contributes to APLNR availability at the plasma membrane
Because GP130 interacts with APLNR and is coopted in APLN signaling, we wondered whether GP130 could directly modulate APLN binding to its receptor in GSCs. However, while the competitive antagonist of APLNR, namely MM54, precluded APLN binding, anti-GP130 antibodies failed to do so (Fig. S3 B; Harford-Wright et al., 2017). We therefore inferred that presentation of APLNR at the cell surface could be the mechanisms by which its downstream signaling is modulated. Supporting this hypothesis, silencing of GP130 expression decreased the membrane expression of APLNR, as monitored by flow cytometry, in three different patient-derived GSCs (Fig. 3 A). Conversely, interfering with APLNR expression did not mitigate the level of GP130 detected at the plasma membrane, therefore suggesting an unidirectional relationship within the APLNR–GP130 complex (Fig. 3 B). Furthermore, GSC#1, with CRISPR/Cas9-mediated KO of GP130 gene expression, phenocopied transient knockdown cells in terms of diminished APLNR membrane expression (Fig. 3 C). Of note, GP130 deletion did not impinge on the expression of JAMC (junctional adhesion molecule C), an alternate membrane receptor expressed in GSCs, ruling out a broad action of GP130 expression on protein presentation at the plasma membrane (Fig. 3 C). This effect was most likely posttranslational, as neither APLNR silencing nor GP130 knockdown and KO altered the global level of expression of the partner receptor (Fig. S3, C and D). One nonexclusive alternative mechanism might be that GP130 cotraffics with APLNR from the ER along the secretory pathway. To challenge this hypothesis, APLNR was ectopically expressed in HEK293T together or not with GP130 and further treated with Endoglycosidase H (endo H) and peptide/glycosidase (PGNase), which removes asparagine-linked high mannose glycans from immature and mature glycoproteins, respectively. The proportion of EndoH-sensitive and EndoH-resistant APLNR was not overly modified with GP130 expression (Fig. S3 E). Importantly, the reintroduction of GP130, with GP130 episomal expression in KO cells, was sufficient to restore APLNR membrane availability (Fig. 3 D). Echoing flow cytometry analysis, the measurement of mean intensity of APLNR-conjugated fluorescent staining by high-content microscopy revealed a drastic reduction in APLNR signal when GP130 expression was abolished (Fig. 3 E). This most likely results from the dispersion and/or processing of the APLNR immunosignal into intracellular compartments. Of note, FLAER staining of GPI-enriched membrane microdomains was not modified, and remaining surface-exposed APLNR localized within this zone (Fig. 3 F). We next challenged APLNR trafficking from and to the plasma membrane (i.e., internalization and recycling, respectively) in WT and KO cells (Fig. 3 G). Tracking the internalized fraction of APLNR demonstrated an enlarged proportion of APLNR-positive intracellular puncta within RAB5-stained (i.e., early endosome), RAB7-negative (i.e., late endosome) vesicles in GP130 KO GSCs (Fig. 3 H). Flow measurements identified an increase in anti-APLNR antibody uptake in GP130 KO clones, as compared with WT (Fig. 3 I). Conversely, APLNR recycled back at the plasma membrane solely in WT cells 1 h following antibody uptake (Fig. 3 J). The receptor failed to do so in GP130 KO cells (Fig. 3 J). This was further quantified with flow cytometry analysis over time (Fig. 3 K). Altogether, our data support the notion that GP130 expression dictates the level, but not the clustering, of APLNR surface presentation.
The expression of the GTPase activating protein ELMOD1 is modulated by GP130
To gain further mechanistic insights into GP130-orchestrated APLN-based signaling, whole-transcriptome shotgun sequencing was performed in WT and GP130 KO GSC#1 (Fig. 4 A). Principal-component analysis and unsupervised clustering indicated a minimal culture drift between independent samples (Fig. 4 A), with a sparse set of genes whose expression was modified upon GP130 deletion, rather than massive differences between WT and GP130 KO clones. RNA sequencing (RNA-seq) data were next confirmed by qPCR on a panel of 11 differentially expressed genes (5 down and 6 up) shared between the two KO clones (Fig. 4 B). Among those was ELMOD1 (engulfment and cell motility proteins [ELMO] domain containing 1), a protein with GTPase-activating protein (GAP) for the ARF (ADP-ribosylation factor) family of small G proteins (Zhang et al., 1998; Donaldson and Jackson, 2011; Ivanova et al., 2014; Sztul et al., 2019). Consistent with the observed decrease in mRNA level, ELMOD1 protein abundance was also reduced when analyzed by immunostaining in GP130 KO cells (Fig. 4 C). Of note, this was not the case for ELMOD2- and ELMOD3-related members (Fig. 4 D). Likewise, transient silencing of GP130, but not APLNR, reiterated the reduction in ELMOD1 mRNA (Fig. 4 E). GP130 signaling culminates in STAT3 Y705 phosphorylation, a signaling event, which is abolished in GP130 KO clones (Fig. 2 G). Supporting the idea that ELMOD1 expression is controlled by a GP130/STAT3 axis, our analysis of publicly available datasets unveiled putative STAT3 binding sites around ELMOD1 gene, similar to the ones identified in the ELMOD2 and ELMOD3 regions (Figs. 4 F and S4). In keeping with this idea, inhibiting STAT3 in GSC#1 led to a reduction in ELMOD1 transcription (Fig. 4 G), along with the expected increase and decrease in the level of TP53 and BCL2L1, respectively, used as internal controls (Louault et al., 2019). Conversely, neither GP130 nor APLNR mRNA levels were modified upon STAT3 pharmacological blockade (Fig. 4 G). The Cancer Genome Atlas (TCGA) was then interrogated for the correlated genes associated with ELMOD1 expression in primary GBM patients. This reveals 635 up-regulated and 257 down-regulated genes, positively and negatively correlated with ELMOD1 expression, respectively (Fig. 4 H). Interestingly, gene ontology function analysis partitions the ELMOD1-associated signature to “transport,” “cell cycle,” “synapse,” “ion homeostasis,” and “NTP,” which agrees with reported functions of ELMODs (Ivanova et al., 2014; Sztul et al., 2019). Collectively, these results suggest that GP130 expression governs the expression of ELMOD1 in GSCs.
ELMOD1 is involved in the GP130-dependent trafficking of APLNR
We then wondered whether ELMOD1 drives the phenotype of GP130 KO cells. First, ELMOD1 silencing in parental GSC#1 attenuated APLNR level at the cell surface (Fig. 5, A and B). Likewise, a portion of GP130 was withdrawn from the plasma membrane in ELMOD1-knockdown GSCs, while APLNR internalization was enhanced (Fig. 5, B and C). Favoring a specific involvement of GP130 in APLNR trafficking via ELMOD1, ELMOD2 and ELMOD3 silencing did not affect APLNR surface presentation (Fig. 5 D). At the functional level, APLN-mediated secondary sphere formation was reduced when ELMOD1 mRNA was silenced (Fig. 5 E). Finally, this correlation between ELMOD1 and GP130 was discernible at the level of their expression in clinical GBM samples (Fig. 5 F). Data mining of the TCGA further unveiled that the high levels of ELMOD1 and, to a lesser extent, GP130 mRNA were associated with a worsen prognosis in primary GBM (Fig. 5 G), supporting the notion that impairing plasma membrane signaling could be of therapeutic interest.
ARF GTPases contribute to APLNR internalization
Because of the prominent role of ARFs in receptor trafficking and intracellular dynamic processes (Donaldson and Jackson, 2011; Sztul et al., 2019), we next examined whether GP130 extinction modifies the expression of ARF family members. However, the RNA expression for the seven members tested here, namely ARF1, ARF3, ARF4, ARF5, ARF6, ARL1, and ARL2, remained even in both WT and GP130 KO cells (Fig. S5, A and B). Because ARF enzyme activity could be controlled by the GAP activity of ELMOD1 (Schweitzer et al., 2011; Ivanova et al., 2014), we then challenged whether interfering with the expression of two ARF G-protein exchange factors (GEFs), namely BRAG2 (IQSEC1) and ARNO (CYTH2; Donaldson and Jackson, 2011; Sztul et al., 2019; Peurois et al., 2017), modulates APLNR membrane expression (Fig. S5, C and D). We observed that ARNO, but not BRAG2, silencing indeed increased APLNR plasma membrane staining in GSC#1, suggesting an ARF-dependent control of APLNR surface expression (Fig. S5, C and D). This augmentation was not, however, detected in GP130 KO cells (Fig. S5 D). Although we cannot exclude complex, parallel mechanisms, it is tempting to speculate that ARFs cannot be further inactivated in the GP130 KO background as a consequence of the loss of ELMOD1 expression. It is therefore conceivable that the functions of ARFs in vesicular trafficking are altered in GP130 KO cells. To test this hypothesis, seven ARF members were individually knocked down in GSC#1 (Fig. 6 A; Sztul et al., 2019). Interestingly, ARF3 silencing boosted APLNR expression at the plasma membrane and was the only tested ARF, whose silencing inversely mirrored the phenotype observed in GP130- and ELMOD1-silenced cells (Fig. 6 A). Indeed, ARF3 silencing increases APLNR membrane presentation in GSC#1 (Fig. 6, B and C), which might be explained rather via an increased recycling activity than APLNR endocytosis in WT cells (Fig. 6 D). Importantly, ARF3 knockdown was sufficient to moderately rescue the GP130 KO phenotype on APLNR membrane trafficking (Fig. 6, B and C). In these cells, where APLNR was less directed toward the plasma membrane (Fig. 3), ARF3 silencing partially opposed to the effects induced by GP130 deletion (Fig. 6 D). Internalization was slightly but significantly reduced, while recycling was increased, suggesting that ARF3 silencing may in part normalize APLNR flux in GP130 KO cells. However, this partial effect suggested the coexistence of combined and paralleled mechanisms. For instance, it is noteworthy that ARF6 silencing negatively affects the level of membrane-exposed APLNR (Figs. 6 A and S6), suggesting counteracting action between ARF3 and ARF6 in regulating APLNR membrane. From a functional standpoint, ARF3 knockdown enhanced APLN-mediated secondary sphere formation, again inversely mirroring ELMOD1 and GP130 silencing, as well as ARF6 silencing phenotype (Figs. 6 E and S6). This complex phenotype might result from a combined, stepwise involvement of multiple ARFs cooperating in parallel and in a timely manner (Figs. 6 A and S6).
Thus, GP130 not only interacts with APLNR but also contributes to their copresentation at the plasma membrane by transcriptionally enabling ELMOD1 expression, which in turn involves a fine-tuned balance between endocytosis/recycling ARF-based activities (Donaldson and Jackson, 2011; Sztul et al., 2019; Schweitzer et al., 2011). This could be attributed to the modulation of receptor membrane trafficking at the plasma membrane and/or coreceptor clustering, as previously reported for VEGFR2 and EGFR (Zhu et al., 2017; Kulasekaran et al., 2021). Our data therefore suggest that this endocytic–recycling pathway may constitute an interesting process to target in cancer cells dependent on APLN supply.
Here, we found that the glycoprotein GP130 commissions APLN-mediated action on GSC self-renewal. From a molecular standpoint, interfering with GP130 lowers APLNR availability at the plasma membrane, thereby dampening APLN signaling. In fact, GP130 dictates APLNR surface presentation by enabling the expression of ELMOD1. This might ultimately detune ARFs, which are instrumental molecular switches of vesicular trafficking (Donaldson and Jackson, 2011; Sztul et al., 2019). In line with previous reports pointing to the role of endovesicular trafficking in GBM cell vitality (Shingu et al., 2017; Jacobs et al., 2020; Kulasekaran et al., 2021), our study reports on how modulating receptor dynamics might limit nononcogene signaling involved in tumor progression.
The role of the APLN peptide has initially been investigated in the cardiovascular system, highlighting its protective actions against ischemia–reperfusion injury, elevated blood pressure, and aging (Yang et al., 2015). Further involvement of the apelinergic system in metabolic disorders, such as diabetes and obesity, has been unmasked (Chaves-Almagro et al., 2015). This hormonal peptide operates through the G protein–coupled receptor APLNR, which is widely distributed throughout the body, with a preponderant expression in smooth muscle cells and endothelial cells. The canonical Gi/G0-coupling toward cAMP inhibition conducts the main functions of APLN (Chaves-Almagro et al., 2015). Another layer of complexity is added by the fact that APLNR coalesces with other G protein–coupled receptors, including angiotensin, bradykinin, opioid, and neurotensin receptors (Chaves-Almagro et al., 2015). Recent findings had further highlighted the role of the vasoactive peptide APLN in the paracrine communication between endothelial cells and their neighboring cells in the tumor ecosystem (Harford-Wright et al., 2017; Uribesalgo et al., 2019; Chen et al., 2019; Mastrella et al., 2019). Indeed, genetic and pharmacological targeting of APLNR abrogates APLN- and endothelial-mediated expansion of tumor cells in vitro and suppresses tumor growth in vivo (Harford-Wright et al., 2017; Uribesalgo et al., 2019; Mastrella et al., 2019). Functionally, selective competitive antagonists of APLNR are safe and effective in reducing tumor growth and lengthening the survival of intracranially xenografted mice (Harford-Wright et al., 2017). Therefore, the APLN–APLNR signaling nexus may operate as a paracrine signal that sustains tumor cell expansion and progression. Our study shows that GP130 controls the magnitude of APLNR expression at the cell surface of GSCs. It does so by interfering with ligand-independent APLNR endocytosis, although alternate mechanisms such as cotrafficking from the ER secretory pathway might also contribute. Future work will delineate the contribution of GP130 in the other facets of APLNR functions besides GSCs.
APLN appears to modulate proliferation, survival and migration in varied cell types (Chaves-Almagro et al., 2015). In line with the discovery of these multiple cellular actions sparked by APLN, the downstream mechanisms have been extended to the activation of intracellular kinase pathways, such as MAPK and phosphoinositide 3-kinase/AKT (Chaves-Almagro et al., 2015). Our data suggest that APLN increases the phosphorylation of the ribosomal protein pS6, a typical target of mTOR (mechanistic target of rapamycin), constitutively active in most glioma and GSCs (Galan-Moya et al., 2011). However, GP130 deletion did not impair either the constitutive activation of the mTOR pathway or in response to APLN, as pS6 phosphorylation was left intact in both conditions. Recent works also demonstrated the importance of the GSK3β activation in GSCs, notably in the response to APLN (Zhou et al., 2016; Harford-Wright et al., 2017). Likewise, the inhibitory phosphorylation of GSK3β was elevated in GP130 KO cells upon APLN exposure. Thus, in the absence of GP130, APLN stimulation can no longer maintain GSK3β activity and its effect on GSC self-renewal. Moreover, GP130 KO interfered with the constitutive STAT3 activation in proliferative GSCs. Pharmacological blockade of JAK–STAT signaling mimics the impact of GP130 deletion on APLN-driven self-renewal of GSCs (Harford-Wright et al., 2017; Jacobs et al., 2017). However, our ongoing observations suggest that the level of STAT3 phosphorylation is not altered in APLN-stimulated GP130 KO cells, indicating that STAT3 activation is not a direct consequence of APLN stimulation. Rather, it is more likely due to an autocrine stimulation of the GP130 receptor via IL-6 (Wang et al., 2009; Shi et al., 2017). Thus, GP130-dependent activation of JAK/STAT in GSCs might impact the capacity of APLNR to conduct downstream molecular signaling via a bystander effect. Hence, GP130 deletion selectively opposes APLN downstream signaling (i.e., GSK3β inhibition), but not mTOR activation. Although the decisional mechanism involved is still unclear, it might be linked to the observed defined localization of APLNR at the cell membrane, enriched in FLAER-positive zones. Because the GPI-enriched nanodomains integrate signaling platforms and sort signaling hubs (Helms and Zurzolo, 2004), these detergent-resistant membrane domains most likely contribute to compartmentalize APLNR signaling and trafficking. The restricted, polarized localization of APLNR occurs in a ligand-independent manner, at steady-state in GSCs. However, whether this localization guides APLN signaling is still unclear. Interestingly, our data showed that APLNR was addressed, at a slow pace, to such lipid domains of the plasma membrane, even in the absence of GP130 expression. Therefore, GP130 might modulate APLNR surface presentation rather than its guided delivery toward these domains. How GP130 further controls APLNR membrane availability might be a multi-modal mechanism, such as internalization and/or recycling, cotrafficking through the ER secretory pathway, and, membrane stability. While we cannot entirely rule out any of these latter dynamic processes, our data point toward the regulation of APLNR ligand-independent internalization via the modulation of the endocytosis/recycling machinery, as illustrated by anti-APLNR antibody uptake experiments.
In keeping with this idea of GP130 modulating APLNR membrane presentation, our RNA-seq analysis in GP130 KO GSCs unveils changes in ELMOD1 expression, a putative negative regulator of ARF GTPases (Donaldson and Jackson, 2011; Ivanova et al., 2014; Sztul et al., 2019). The ELMO family consists of six paralogs in mammals, with ELMOD1 exhibiting in vitro GAP activity toward ARFs (Donaldson and Jackson, 2011; Ivanova et al., 2014; Sztul et al., 2019). Our experimental results corroborate the analysis of TCGA clinical data showing a positive correlation between GP130 and ELMOD1 gene expression. In addition, the ELMOD1 loci harbor putative STAT3-binding sites, which may in turn account for the decreased level of ELMOD1 mRNA in GP130 KO cells. Likewise, pharmacological blockade of STAT3 reduces ELMOD1 expression. ELMOD1 is highly expressed within the brain, marking stereotyped neuronal structures within the hippocampus (Johnson et al., 2012). ELMOD1 has been reported to contribute to the stabilization of apical structures in vestibular cells during development, while mutations in the human gene have been linked with mental retardation and autism disorder syndromes (Johnson et al., 2012; Krey et al., 2018; Miryounesi et al., 2019). Although ELMOD1 mutations have not been documented in cancer patients, TCGA analysis predicts that a low level of its transcripts suggests a better prognosis. It is therefore tempting to speculate here that ELMOD1 conveys permissive nononcogene signaling, at least via its action on APLNR in GSCs. Whether ELMOD1 acts on the magnitude of other membrane receptors will require further investigation. Additionally, it will be important to decipher whether ELMOD1 operates directly on ARFs in GSCs, alone or in conjunction with the GEF ARNO (CYTH2) but possibly other GEFs besides BRAG2 (IQSec1; Peurois et al., 2017; Sztul et al., 2019). In this context, our siRNA screening of seven ARF members unveils that at least two of them, namely ARF3 and ARF6, exert opposing effects on APLNR trafficking from and to the plasma membrane. Alternatively, such ARFs might operate in sequence. Yet, how exactly ELMOD1 controls APLNR surface expression requires further investigation, as ELMOD1 might target ARF GTPases either by its GAP activity and/or binding, reminiscent of SAT (suppressors of Arf ts) identified in yeast (Zhang et al., 1998). Overall, this suggests a complex modulation in endovesicular dynamics, which recently emerged as a potent checkpoint for tumor cell expansion in GBM cells (Shingu et al., 2017; Jacobs et al., 2020; Kulasekaran et al., 2021).
In conclusion, we report that GP130 interacts with APLNR and guides its surface presentation and availability for ligand binding. This supports the notion that impairing membrane signaling might be achieved in multiple ways, including blocking ligand binding and camouflaging the receptor from agonists, and therefore offers new opportunities for anticancer therapeutic options.
Materials and methods
Informed consent was obtained from all patients before sample collection for diagnostic purposes. This study was reviewed and approved by the institutional review boards of Sainte Anne Hospital (Paris, France) and Laennec Hospital (Nantes, France) and performed in accordance with Helsinki protocol.
Cell culture, siRNA, and DNA transfection
Stealth nonsilencing control duplexes (low-GC 12935111; Life Technologies), and siRNA duplexes (Stealth RNAi; Life Technologies) were transfected using RNAiMAX Lipofectamine (Life Technologies). The following duplexes targeting the respective human genes were used: GP130, 5′-CUCACUUGCAACAUUCUUA-3′; APLNR, 5′-CCAUCAUGCUGACCUGUUACUUCUU-3′; ELMOD1 HSS124561 (si.1), HSS124562 (si.2), and HSS124563 (si.3), ELMOD2 5′-AUAUAAAUUCUGUUGAAUG-3′ (si.1) and 5′-UCAAUUUAAUACAAAUACC-3′ (si.2), ELMOD3 HSS130688 (si.1) and HSS130687 (si.2), ARF1 HSS141267, ARF3 5′-CCAUUGGGUUCAAUGUGGAGACAGU-3′ (si.1), 5′-UGUGGAGACAGUGGAGUAU-3′ (si.2) and 5′-ACAGGAUCUGCCUAAUGCU-3′ (si.3), ARF4 5′-GAUAGCAACGAUCGUGAA-3′, ARF5 5′-UGAGCGAGCUGACUGACAA-3′, ARF6 HSS100646 (si.1), HSS180192 (si.2) and HSS180193 (si.3), ARL1 5′-AAGAAGAGCUGAGAAAAGCCA-3′, ARL2 5′-CAACCAUCCUGAAGAAGUU-3′, ARNO/CYTH2 HSS145078 (si.1) and HSS145079 (si.2), BRAG2/IQSEC1 HSS115015 (si.1) and HSS115016 (si.2).
pORF-GP130 (OriGene) and HA-tagged APLNR (Chun et al., 2008) plasmid DNA were introduced in GSCs and HEK-293T using the Neon electroporation system and Lipofectamine, respectively, according to the manufacturer’s instructions (Life Technologies).
Single gRNAs (sgRNAs) targeting GP130 were chosen in the sgRNA library and cloned into a lentiviral lentiCRISPRv2 (GeCKO; ZhangLab) backbone (Sanjana et al., 2014; Shalem et al., 2014). The sgRNA used was 5′-GTGGATGCTGTGTCTTCAGG-3′. For infections, lentiviral particles were produced in HEK293T cells by cotransfection of the various constructs together with pVSV-G and psPAX2 plasmids. Supernatants were collected after 2 d and applied on GSC#1 during a 1,000 g centrifugation for 90 min in the presence of 8 µg/ml polybrene (Santa Cruz Biotechnology). Cells were cultured with 1 µg/ml puromycin to select infected cells and sorted by fluorescence-activated cell sorting (anti-GP130; Abcam; and ab34324, BD FACS Aria III; BD Biosciences). Viable single clones were tested by flow cytometry to control for GP130-negative staining, and GP130 KO was confirmed by genomic DNA sequencing (forward, 5′-CAAGGCTGTGCCTCAAAAA-3′; reverse, 5′-CCCACTTGCTTCTTCACTCC-3′). Two bi-allelic KO clones were further selected.
Antibodies and reagents
The following primary antibodies were used (dilution 1:1,000 for Western blot, 1:200 for immunofluorescence and flow cytometry): GAPDH (mouse SC-25778; Santa Cruz Biotechnology), GP130 (rabbit SC-656 for Western blot, mouse SC-376280 for immunofluorescence; Santa Cruz Biotechnology; and mouse ab34324 for flow cytometry; Abcam), APLNR (mouse FAB8561R and MAB856; R&D Systems), JAMC (mouse FAB11891A; R&D Systems), STAT3 (rabbit 9132; Cell Signaling), pS727-STAT3 (rabbit 9134; Cell Signaling), pY705-STAT3 (rabbit 9145; Cell Signaling), pS235/S236-S6 (rabbit 2211 for Western blot and rabbit 5316 for flow cytometry; Cell Signaling), pS9-GSK3β (rabbit 9336; Cell Signaling), ELMOD1 (rabbit ab127541; Abcam), RAB5 (rabbit ab218624; Abcam) and RAB7 (rabbit 9367; Cell Signaling). HRP-conjugated secondary antibodies (anti-rabbit, mouse Ig, mouse IgG1, mouse IgG2a, and mouse IgG2b) were purchased from Southern Biotech. Alexa-conjugated secondary antibodies were from Life Technologies. FLAER (Alexa Fluor 488–conjugated proaerolysin) was purchased from Cedarlane, MBCD was from Sigma-Aldrich, and Stattic, ruxolitinib, and filgotinib were all from Selleckchem. [Pyr1]-Apelin-13 was resuspended and aliquoted according to the supplier’s recommendation (Phoenix).
Exploration of public datasets
The STAT3 consensus binding sequence was obtained from the JASPAR database, together with the list of matching sites across the genome. A closer look was then applied to the 500-kb region surrounding the ELMOD1, ELMOD2, and ELMOD3 genes. They were further compared with STAT3 transcription factor chromatin immunoprecipitation sequencing (ChIP-seq) database from ENCODE. TCGA was explored in Agilent databases via the Gliovis platform (http://gliovis.bioinfo.cnio.es/; Bowman et al., 2017). Optimal cutoffs were set. All subtypes were included, and histology was the only selective criteria.
To analyze tumorsphere formation, GSCs (100 cells/µl) were seeded in triplicate in NS34, MF media and MF media supplemented with Pyr-APLN (1 µM; Phoenix) as previously described (Harford-Wright et al., 2017). Cells were dissociated manually each day to reduce aggregation influence and maintained at 37°C 5% CO2 until day 5 (day 4 for siRNA). Tumorsphere formation was assessed in response to anti-GP130 (2 µg/ml, rabbit ab234105; Abcam) or control antibodies (swap70, 2 µg/ml, rabbit ab228846; Abcam). Tumorspheres per field of view (FOV) were calculated by a manual, single-blinded count of the total number of tumorspheres in five random FOVs for each well.
To evaluate the self-renewal of GSCs, LDAs were performed as previously described (Tropepe et al., 1999). GSCs were plated in a 96-well plate using serial dilution ranging from 2,000 to 1 cell per well with eight replicates per dilution and treated as indicated. After 14 d, each well was binary evaluated for tumorsphere formation. Stemness frequency was then calculated using ELDA software (Hu and Smyth, 2009). The mean stemness frequency for each treatment was calculated by averaging across two independent experiments.
Flow cytometry analyses were performed on FACSCalibur (BD Biosciences and Cytocell) and processed using FlowJo software. Staining was performed on living cells (incubation, 1 h at 4°C). Intracellular flow cytometry experiments were performed using BD Cytofix/Cytoperm Kit (BD Biosciences) following the manufacturer’s protocol.
Cells were fixed in PBS-paraformaldehyde 4% and processed for immunostaining with or without permeabilization with Triton X-100 (0.05%, 5 min). Immunofluorescence was performed using Alexa Fluor 488–, Alexa Fluor 568–, and Alexa Fluor 640–conjugated secondary antibodies (dilution 1:500; Life Technologies). Samples were mounted in DAPI-containing mounting medium (Prolong gold anti-fade reagent; Life Technologies). APLNR–GP130 interaction was visualized through the Duolink in situ kit following the manufacturer’s instructions (Sigma-Aldrich). All images were acquired on confocal Nikon A1 RSi using a 60× oil-immersion lens with NIS-Element Software at the Nikon Excellence Center (Micropicell, SFR François Bonamy, Nantes, France). Signal intensity quantification was performed on Imaris software (Oxford Instruments). Structured illumination microscopy images were acquired with a Nikon N-SIM microscope. Z-stacks of 0.12 µm were performed using a 100× oil-immersion lens with a 1.49 aperture and reconstructed in 3D using the NIS-Element Software. Reconstructed images validated with the SIMcheck plugin 1.2. All images were analyzed using the Image J software.
Antibody uptake and recycling experiment
To investigate endocytosis, cells were first placed at 4°C for 30 min to stop endocytosis processes and stained with anti-APLNR AF647-coupled antibody (mouse IgG1, 1 h, 4°C, FAB8561R, R&D Systems). After three ice-cold PBS washes, cells were transferred into prewarmed NS34 at 37°C and harvested into ice-cold PBS at different time points (0, 2.5, 5, 10, 15, and 30 min). An acid wash (50 mM glycine and 100 mM NaCl, pH 3, for 10 min at 4°C) stripped membrane-bound antibody and left intact intracellular staining to be processed for flow cytometry or confocal microscopy. To track recycling processes, a similar procedure was deployed using uncoupled anti-APLNR antibody (mouse IgG1, 1 h, 4°C, MAB8561; R&D Systems). After 15 min of endocytosis in prewarmed NS34 at 37°C (pulse), membrane-bound antibody was stripped using the acid wash solution described above. Then, cells were placed back at 37°C and harvested into ice-cold PBS at different time points (0, 5, 10, 15, 30, and 60 min, chase). APLNR membrane staining was revealed with Alexa-conjugated secondary anti-mouse IgG1 antibodies (1 h, 4°C; Life Technologies) followed by three ice-cold PBS washes. Samples were analyzed by flow cytometry and confocal microscopy.
Total protein lysates from GP130- and/or APLNR-transfected HEK293T cells were denaturated (55°C, 10 min) and treated overnight with EndoH and for 1 h with PGNase (New England Biolabs) according to the manufacturer’s instructions.
Immunoblotting and immunoprecipitation
Briefly, cells were lysed at 4°C in TNT lysis buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 1% Triton X-100, 1% Igepal, and 2 mM EDTA, supplemented with protease inhibitor) for 30 min and cleared by centrifugation at 8,000 g. For immunoprecipitation, samples were precleared by a 30-min incubation with Protein G agarose, and then incubated for 2 h at 4°C with Protein G agarose and 5 µg of indicated antibodies. Protein concentrations were determined by bicinchoninic acid. Equal amount of 5–10 µg proteins were resolved by SDS-PAGE and transferred to nitrocellulose membranes. Membranes were revealed using a chemiluminescent HRP substrate and visualized using the Fusion imaging system (Vilber-Lourmat).
Parental and GP130 KO GSC#1 were harvested in three independent experiments and snap-frozen on dry ice. RNA extraction (all RNA integrity number >9.0), library preparation, RNA-seq and bioinformatics analysis was performed at Active Motif (Carlsbad, CA). Briefly, 2 µg of total RNA were isolated using the Qiagen RNeasy Mini Kit and further processed in Illumina’s TruSeq Stranded mRNA Library kit. Libraries are sequenced on Illumina NextSeq 500 as paired-end 42-nt reads. Sequence reads are analyzed with the STAR alignment – DESeq2 software pipeline described in the Data Explanation document. The list of differentially expressed genes from DESeq2 output was selected based on 10% adjusted P value level and false discovery rate of 0.1. Gene ontology and KEGG pathway enrichment analysis was done using DAVID bioinformatics resources portal.
RNA extraction was done using Qiagen RNeasy kit on three biological replicates (3.106 GSC#1). Equal amounts of RNA were reverse transcribed using the Maxima Reverse transcription kit, and 30 ng of the resulting cDNA was amplified by qPCR using PerfeCTa SYBR Green SuperMix Low ROX. Data were analyzed using the 2-ΔΔCt method and normalized by the housekeeping genes ACTB and HPRT1. The following primers were used: ACTB forward, 5′-GGACTTCGAGCAAGAGATGG-3′; ACTB reverse, 5′-AGCACTGTGTTGGCGTACAG-3′; HPRT1 forward, 5′-TGACACTGGCAAAACAATGCA-3′; HPRT1 reverse, 5′-GGTCCTTTTCACCAGCAAGCT-3′; GP130 forward, 5′-TGTAGATGGCGGTGATGGTA-3′; GP130 reverse, 5′-CCCTCAGTACCTGGACCAAA-3′; APLNR forward, 5′-GAGTGCTGGGAAGGACTCTG-3′; APLNR reverse, 5′-ACTGGTTGTCTGCCCCATAG-3′; FAM20C forward, 5′-CCTCAGCTCCCAGAGGATCA-3′; FAM20C reverse, 5′-TCCCTCGTTTGTTTCATGGGT-3′; GRIK1 forward, 5′-CTCCCTGCAGTGCCATAGAC-3′; GRIK1 reverse 5′-TCCCAATCTTTTCAGTTCCTTCC-3′; ELMOD1 forward, 5′-GGAAGGCGGAGATGAAGACC-3′; ELMOD1 reverse, 5′-GGATCAACATTCTCAGGAAGTGC-3′; ELMOD2 forward, 5′-GGAAGGCGGAGATGAAGACC-3′; ELMOD2 reverse, 5′-TGTCGCCTTCTGTAAAACCTTA-3′; ELMOD3 forward, 5′-GCTCAGTGTGGCCTGGATAG-3′; ELMOD3 reverse, 5′-TCGGCAAGGTCTTTGAGTCC-3′; IGSF1 forward, 5′-GCCAGCAGGCTATCTCACC-3′; IGSF1 reverse, 5′-GCAGAGAAGGGCCGTGACTA-3′; MRC2 forward, 5′-CGGACACCAAACTCCGGTAT-3′; MRC2 reverse, 5′-CGGGTTCTGATTCACCGAAGA-3′; CDH10 forward, 5′-ATGGCAAGGGACCCAGATTC-3′; CDH10 reverse, 5′-GGGATTGTTGATTTCAGCAGCA-3′; ROBO2 forward, 5′-TGCAGACTTGCCAAGAGGAA-3′; ROBO2 reverse, 5′-AAACTGTGGGGGAGCAACAG-3′; DRAXIN forward, 5′-CTGCCACAGAAGAGTCCCTG-3′; DRAXIN reverse, 5′-ACCGCGGTGTTTCTCTTTCT-3′; LAMP5 forward, 5′-TCTTTGTAAAGGAAAGCCACAACAT-3′; LAMP5 reverse, 5′-CAGAGAGGTGGTGCGAGTTG-3′; GRIP2 forward, 5′-TAGCGAACATGCTGTGTGGG-3′; GRIP2 reverse, 5′-CCCTCGGAACTCCTCTGGAA-3′; TNFSF10 forward, 5′-GCAGCTCACATAACTGGGACC-3′; TNFSF10 reverse, 5′-TTGCTCAGGAATGAATGCCC-3′; BCL2L1 forward, 5′-TTCAGTGACCTGACATCCCA-3′; BCL2L1 reverse, 5′-TCCACAAAAGTATCCCAGCC-3′; TP53 forward, 5′-GGAGCACTAAGCGAGCACTG-3′; TP53 reverse, 5′-TCTCGGAACATCTCGAAGCG-3′; ARF1 forward, 5′-GGGCCGCAGAGACGTT-3′; ARF1 reverse, 5′-TCCACGTTGAAGCCTATGG-3′; ARF3 forward, AGGAAAACCGTCCGAGGAG-3′; ARF3 reverse, 5′-TGGTCCCTGGTATGGAAGACT-3′; ARF4 forward, 5′-AGCTGCAGAAAATGCTTCTGGT-3′; ARF4 reverse, 5′-TGACAGCCAGTCAAGTCCTTC-3′; ARF5 forward, 5′-CGTGTCCGCGCTCTTTTC-3′; ARF5 reverse, 5′-CGCATCCAAGCCAACCATGA-3′; ARF6 forward, 5′-GTGACCACCATTCCCACTGT-3′; ARF6 reverse, 5′-GCCCACATCCCATACGTTGA-3′; ARL1 forward, 5′-CAGTCTGTTTGGAACTCGGGA-3′; ARL1 reverse, 5′-ACATCTCCAGTATGGCCTGA-3′; ARL2 forward, 5′-CTGGAGCACCGAGGATTCAAG-3′; ARL2 reverse, 5′-AAAGATGAGGAGGGTTGCTCC-3′; ARNO/CYTH2 forward, 5′-AGGACGGCGTCTATGAACC-3′; ARNO/CYTH2 reverse, 5′-CGATCTCCTCGGGTGTGTTC-3′; BRAG2/IQSEC1 forward, 5′-CATAGAGGCGTTCAGCCAGC-3′; BRAG2/IQSEC1 reverse, 5′-CTCACGGGGAATGTCCTCAC-3′.
Data are representative of at least three independent experiments, unless otherwise stated. Statistical analysis was performed with GraphPad Prism8 using one-way and two-way ANOVA. For each statistical test, a P value of <0.05 was considered significant. Data distribution was assumed to be normal, but this was not formally tested.
Online supplemental material
Fig. S1 shows that JAK inhibitors impair APLN-based GSC maintenance. Fig. S2 reports the effects of nonionic detergents on APLNR membrane localization. Fig. S3 shows that GP130 does not affect all APLN/APLNR signaling activities in terms of downstream S6 phosphorylation, APLN binding to APLNR, and APLNR expression and glycosylation. Fig. S4 documents STAT3 predictive consensus binding sites on the three loci surrounding ELMOD1, ELMOD2, and ELMOD3 genes. Fig. S5 shows the modulation of APLNR membrane expression by the ARF-related GEFs ARNO and BRAG2. Fig. S6 highlights that the ARF GTPase ARF6 contributes to modulate APLNR availability at the plasma membrane.
We thank Signaling in Oncogenesis, Angiogenesis, and Permeability team members, especially Tiphaine Douanne, Yanis Mace, Thomas Maurice, Margot Machu, and Margo Fernandez (Université de Nantes, Institut National de la Santé et de la Recherche Médicale [INSERM], Centre National de la Recherche Scientifique, Nantes, France). We are grateful to Thomas Bonneaud (Université de Nantes, INSERM, Centre National de la Recherche Scientifique, Nantes, France) for providing the STAT3 inhibitor and technical advice, as well as Eric Chevet (INSERM, Rennes, France) for helpful discussion. We acknowledge Micropicell and Cytocell core facilities (Structure Fédérative de Recherche Santé François Bonamy, Nantes, France).
This research was funded by Fondation pour la Recherche Médicale (Equipe Labellisée DEQ20180339184); Fondation ARC contre le Cancer (J. Gavard; PJA20171206146 and PJA20191209477); Ligue Nationale contre le Cancer Comités de Loire-Atlantique, Maine et Loire, Vendée, Ille-et-Vilaine (J. Gavard and N. Bidère); and Région Pays de la Loire et Nantes Métropole under Connect Talent Grant (J. Gavard). K.A. Jacobs received Nantes Métropole and Fondation ARC contre le Cancer PhD fellowships. The team is part of the SIRIC ILIAD (INCA-DGOS-INSERM_12558).
The authors declare no competing financial interests.
Author contributions: K. Trillet and K.A. Jacobs: conceptualization, formal analysis, investigation, methodology, visualization, and writing (review and editing). G. André-Grégoire, A. Thys, C. Maghe, J. Cruard, and S. Gonzalez Diest: formal analysis, investigation, and visualization. J. Cruard and S. Minvielle: resources, visualization. G. Montagnac: formal analysis, resources, and methodology. N. Bidère: conceptualization, formal analysis, funding acquisition, and writing (review and editing). J. Gavard: conceptualization, formal analysis, resources, funding acquisition, and writing (original draft). All authors approved the manuscript.