High epiregulin expression in human U87 glioma cells relies on IRE1α and promotes autocrine growth through EGF receptor

Culture media were from Invitrogen (Cergy-Pontoise, France). Antibodies against ErbB1 were purchased from BD Biosciences (San Diego, USA). Anti-ErbB2 and anti-phospho-JNK (Thr183/Tyr185) were from Cell Signaling (Saint-Quentin-en-Yvelines, France). Anti-phospho-Tyr1173-ErbB1 was from Millipore (Molsheim, France). Anti-β-actin and anti-JNK antibodies were from Santa Cruz Biotechnology (Santa Cruz, USA). Recombinant EREG, monoclonal and polyclonal antibodies against EREG and control mouse monoclonal (isotype IgG₁) antibodies were from R&D Systems (Minneapolis, USA). Secondary goat-anti-mouse antibodies coupled to biotin or to peroxidase were from DAKO (Trappes, France). Humanized anti-ErbB1 (Erbitux®, cetuximab) and anti-ErbB2 (Herceptin®, trastuzumab) antibodies were kindly provided by Merck Serono (Darmstadt, Germany) and by Roche (Mannheim, Germany), respectively. Primers are indicated in Additional file 1.

The dominant-negative IRE1 RNase mutant (IRE1Δ899; GenBank accession number JQ425696) was obtained by truncation of the carboxy-terminal 78 amino acids of IRE1α. The mutant was obtained by inserting a gatc motif at position 2812 of the BglII restriction site ²⁷⁹⁹tctgtcagagatc “gatc” tcctccgagccatgagaaataa²⁸³³. The frameshift insertion generates a stop codon 19 bases later. The wild type IRE1α amino acids sequence at positions 896–907 is –SVRDLLRAMRNK- and the C-terminal sequence of the mutant is –SVRDRSPPSHEK-COO^–. The final sequence was controlled by DNA sequencing and was cloned in a pcDNA3 plasmid before transfection in U87wt cells and selection at 800 μg/ml G418.

U87-MG (U87wt) cells were from ATCC (HTB-14). SF126 and SF188 cells were kindly provided by Dr. M. Czabanka (Charité Universitätsmedizin, Berlin). Cells were grown at 37°C, 10% CO₂ in DMEM, 4.5 g/l glucose supplemented with 10% FBS, L-glutamine and antibiotics. Empty plasmid U87 (U87Ctrl) cells, U87 IRE1dn (U87dn) cells [22] and U87 IRE1Δ899 (U87Δ899) cells were grown in the presence of 500 μg/ml G418 and were used at passages 8–13 after transfection. The immortalized human astrocyte NHA/TS cell line and its tumorigenic NHA/TSR counterpart were kindly provided by Drs K. Sasai and S. Tanaka and were grown as reported [24].

Proliferation assay was performed in 96-well plates with DMEM containing 1% FCS and 30 ng/ml EREG. Serial propagation of cells in the absence of serum was developed as previously reported [25]. Briefly, cells were plated at 10 000 cells/cm² in fibronectin-precoated 24-well plates. The serum-free complete medium consisted of a 1 to 1 mixture of DME/F12 medium, 1 mg/ml fatty-acid free BSA, 50 μg/ml high-density lipoproteins, 5 μg/ml transferrin, 5 μg/ml insulin with or without 10 ng/ml EREG. The medium was renewed every 3 days and cells were passaged after 9 days of culture. Cells were counted by using a cell counter (Coultronics, Margency, France). The transwell migration assays was performed as described previously [22]. Results were analyzed after counting of at least 15 fields of 150 μm² each per condition and by three independent investigators.

Subconfluent cells were lysed at 4°C with 100 mM Tris–HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 1 mM Na₃VO₄, 5 mM NaF, protease inhibitors (P8340; Sigma), SDS 1%. The cytosolic fraction was obtained by centrifugation for 2 min at 7000 rpm. After migration on SDS-PAGE, proteins were transferred to a nitrocellulose membrane and probed using antibodies against phospho- and total ErbB proteins, phospho- and total JNK proteins, β-actin or α-tubulin. Primary antibodies were revealed with a secondary HRP-antibody and detected by ELS Western bloting detection reagents (Amersham), or with a secondary antibody coupled to IRDye 800CW using the Odyssey infrared imaging system (Li-Cor Biosciences, Nebraska, US).

Conditioned media were obtained after a 16 h-incubation of cells in serum-free medium containing 1 mg/ml BSA. Proteins were precipitated in the presence of 80% ammonium sulfate, solubilized and dialyzed against PBS. A sandwich-type ELISA was developed for detection of human EREG using 3 μg/ml goat polyclonal antibodies for coating on 96-well plates and a mouse monoclonal anti-EREG (1 μg/ml) as the second antibody. Presence of EREG was indirectly measured using goat anti-mouse antibodies coupled to biotin and revelation was carried out using streptavidin peroxidase and the TMB substrate. Standard curves were obtained using recombinant hEREG and assays were performed in duplicate or triplicate. Measures were obtained with a SPECTRAmax spectrophotometer and calculations were developed from linear curves (r > 0.98).

Total RNAs extraction, real-time quantitative PCR (qPCR) and PCR analyses were carried out as previously described [22] using HPRT1, S16, α-tubulin and β-actin as reference genes. Experiments were performed in triplicate or tetraplicate from two or three independent cell cultures or from chicken and mouse tissues as indicated below. XBP1 splicing was monitored as reported before [22].

U87 cells were plated at a density of 10⁵ cells per well in six-well plates. Small interfering RNA (siRNA) against human IRE1α (5′-GCGUCUUUUACUACGUAAUCU-3′) was from Eurofins MWG Operon (Ebersberg, Germany). ON-TARGETplus siRNA against XBP-1 (GCUCUUUCCCUCAUGUAUAC) and non-targeting siRNA (#D-001810-01-20) were from Dharmacon (Lafayette, CO). Transfection was performed for 48 h using lipofectamine RNAiMAX (Invitrogen) in accordance with the manufacturer’s protocol, with siRNA at a final concentration of 100 nM.

The Chorio-allantoic membrane (CAM) assay was developed as previously described [22]. At day 4 after implantation, tumors were excised from the CAM and pooled (n = 5 for each condition) before RNA extraction using Trizol reagent. Intracranial implantation was performed as follows: U87, SF126, SF188, NHA/TS and NHATSR cells were orthotopically implanted in 8–9 weeks of age RAG2/γ_c immunodeficient mice [22]. Cells (2.5x10⁵ cells, 3 μl) were implanted in the striatum of the left cerebral hemisphere, 0.1 mm posterior to bregma, 2.2 mm lateral and 3 mm in depth. For Kaplan–Meier survival analyses, 18 mice were implanted with U87Ctrl cells and half of them were treated by subcutaneous injection of 400 μg Erbitux® three times a week from day 4 to day 32 post-implantation. In vivo experiments were performed at the animal facility Université Bordeaux 1 (agreement n° B33-522-2) according to ethical criteria approved by the Ministère de l′Enseignement Supérieur et de la Recherche (MESR).

Tumors were xenografted in mice as described above. Brains were recovered at different times and frozen at −80°C. Tissue sections (30 μm) were obtained at −20°C using a CM3050 S microtome (Leica) and were mounted on PEN-membrane 1 mm glass slides (P.A.L.M. Microlaser Technologies AG, Bernried, Germany) that had been pretreated to inactivate RNase. Frozen sections were fixed by incubation for 1 min in pre-cooled (−20°C) 80% ethanol and stained with H&E for 30 s. Sections were then rinsed with RNase-free water for 30 s, dehydrated in a series of pre-cooled ethanol baths (30 s in 50%, 70% and 1 min in 100%) and air-dried. Immediately after dehydratation, LCM was performed using a PALM MicroBeam microdissection system version 4.0-1206 equipped with a P.A.L.M. RoboSoftware (P.A.L.M. Microlaser Technologies AG, Bernried, Germany). Microdissection was performed at 5X or 20X magnification. Total volumes of tumor tissues captured on one single cap were in the 0.8- to 8.7 x 10⁶ μm³ range and random areas were chosen within tumors. RNA samples with a RNA-Integrity Number (RIN) above 8 were kept for qPCR analyses after NanoDrop and Agilent validation. Three tumors were analyzed for each condition and qPCR were carried out in triplicates. Primers specifically recognized cognate human sequences and did not significantly cross-react with any mouse sequences as determined both in total mouse brain tissues and mouse brain sections obtained by LCM. Control qPCR were also performed from tumor tissues after omitting the reverse transcriptase step, giving no detectable signals after 40 complete run cycles.

Expression of EREG and HB-EGF, two members of the EGF family, was analyzed in U87 cells in culture conditions. Using transcriptome analysis, we observed that the two transcripts were abundant both in wild type U87 (U87wt) cells and in cells transfected with the empty vector (U87Ctrl cells), whereas ~100-fold (EREG) and 8-fold (HB-EGF) decreases were monitored in cells expressing an IRE1α dominant-negative protein (U87dn cells) (Figure 1a). Similar results were obtained by qPCR in independent cell cultures as well as in U87wt cells transfected with small interfering RNAs targeting IRE1α (si.IRE1α) (Figure 1a). Thus, both dominant-negative and siRNA knockdown approaches led to a significant decrease in EREG mRNAs in cells under-expressing IRE1α. As positive controls, SPARC and THBS1 genes were upregulated to different extents. Consistent values were obtained at the protein level by using an ELISA against EREG (Figure 1b). U87Ctrl cells released ~270 pg of diffusible EREG per million cells daily, whereas EREG immunoreactivity was undetectable in U87dn cell-conditioned media (< 20 pg per million cells per day).

Presence of EREG and HB-EGF mRNAs in U87 cells was also monitored in human tumor xenografts using the chicken chorio-allantoic membrane (CAM) and the mouse brain models. U87Ctrl and U87dn cells were implanted onto the CAM and tumors were grown for 4 days. Under these conditions, U87dn-tumors were small and merely avascular, compared to massive and angiogenic U87Ctrl-tumors (Figure 1c, upper panel) [22]. Tumors were then excised and total mRNA was extracted for qPCR analysis. EREG and HB-EGF mRNAs were present in smaller amounts (~5- and ~2.5-fold decreases, respectively) in U87dn-derived tumors as compared to U87Ctrl tumors (Figure 1c, lower panel). These transcripts were also quantified in the orthotopic glioma implantation model in mice using LCM coupled to qPCR analysis (Figure 1d). In these conditions, EREG and HB-EGF mRNAs were readily detected in U87Ctrl-derived tumors but not in U87dn-derived tumors (Figure 1d, panel vii). Thus, mRNA production of these growth factors occurred in an IRE1α-dependent manner in U87 glioma cells.

The effect of EREG on U87 cells was examined in cell cultures at low serum concentration. U87dn cells incubated for three days in the presence of EREG underwent notable scattering, which was not observed with U87Ctrl cells (Figure 2a). Such an effect has already been described using HeLa epithelial cells [15]. In addition to its morphological effect, EREG induced proliferation and migration of the two cell variants, these effects being more important in U87dn cells (Figure 2b). These results suggest the presence of functional ErbB proteins on the membrane of U87 cells.

Transcript and protein expression levels of ErbB1-4 were analyzed comparatively and quantitatively in the two cell types. EREG was reported to bind preferentially to ErbB1 and ErbB4, whereas ErbB2 does not bind any known ligand but contributes as a co-receptor to signal transduction [13, 14]. Transcriptomic and qPCR analyses indicated that the respective amounts of ErbB1, ErbB3 and ErbB4 mRNAs are similar in the two U87 cell variants (Figure 2c), the level of ErbB3 transcript being almost undetectable. Besides, the amount of ErbB2 mRNA increased by ~1.5- to 4-fold in U87dn cells vs. U87Ctrl cells. Only ErbB1 and ErbB2 proteins were detected by immunoblotting (Figure 2d; data not shown), which is consistent with results reported by others in this cell model [6, 26]. Finally, treatment of U87Ctrl and U87dn cells with EREG stimulated phosphorylation of the EGFR (ErbB1) protein at Tyr-1173 residue (~10% and ~40% increases in the two cell variants, respectively).

Next, we investigated the respective contribution of ErbB1 and ErbB2 to cell proliferation promoted by EREG. Cells were incubated in the presence of EREG under low-serum conditions, with or without inhibitory antibodies directed against either ErbB1 (Erbitux®) or ErbB2 (Herceptin®). As shown in Table 1, Erbitux® almost completely abrogated EREG-induced cell proliferation of U87Ctrl and U87dn cells, whereas Herceptin® had no significant effect. Thus, the effect of EREG on U87 cell proliferation was mediated mainly through ErbB1.

In order to validate the existence of an EREG autocrine loop, a serial propagation of U87 cells was performed for four passages in a serum-free medium in the absence of growth factors. The culture medium was designed to allow better detection of endogenous growth promoting activities, including those of the EGF family [25]. Again, stimulation with EREG in these conditions resulted in a significantly higher growth rate of both U87Ctrl and U87dn cells (Table 2). This effect was reverted by adding either Erbitux® or anti-EREG antibodies. Interestingly, EREG blocking antibodies also consistently increased by 14% the U87Ctrl cell division time in the absence of exogenous EREG and this effect was not observed in U87dn cells under-expressing EREG. Thus, U87Ctrl cells, but not U87dn cells, actively stimulated themselves by producing both EREG and ErbB1.

The autocrine effect of EREG was then examined in a xenograft tumor model. After implantation of U87wt cells in mice brain, animals were treated for four weeks with or without Erbitux® and tumor aggressiveness was determined. As shown in Additional file 2, no significant effect of Erbitux® was evidenced in this experimental setting (see also ref. [27]), which may result of a limited antibody delivery to tumor tissues. Besides, the autocrine contribution of EREG is likely to be reduced in the U87 glioma model, as these fast-growing tumors secrete other growth-promoting and angiogenic polypeptides and may exploit alternative signaling pathways for expansion [22, 28].

EREG mRNA and protein levels were monitored in several human glioma cell lines. As shown in Figure 3a, U87wt, SF126 and SF188 cells were highly tumorigenic in the orthotopic implantation model in mice and released highly variable amounts of EREG protein (up to 200-fold differences). Moreover, non-tumorigenic NHA/TS human astrocytes produced about five-times more EREG than their highly oncogenic Hras-transformed (NHA/TSR) counterparts. These results are consistent with those obtained at the mRNA levels (Figure 3b) and indicated that the release of EREG by these glioma cell lines did not strictly correlate with tumor malignancy.

We then evaluated the clinical significance of EREG expression in human gliomas, of which a significant percentage accumulates high levels of ErbB proteins. We documented EREG mRNA production by transcriptome mining using the Gene Expression Omnibus (GEO) and Oncomine databases (Additional file 3). Microarray analyses of gliomas at different grades of malignancy indicated that EREG transcripts were detected in highly variable amounts in tumor tissues, although no clear relationship was established between EREG mRNA levels and the glioma grade or brain tumor type. Individual cases presenting EREG upregulation were also observed by using PCR approaches in both anaplastic astrocytoma and glioblastoma, as compared to normal brain tissues [21].

The relationship identified between IRE1α invalidation and the decrease in EREG mRNA level was further monitored in U87 glioma cells incubated with tunicamycin, an antibiotic that inhibits N-linked protein glycosylation and triggers ER-stress. In order to assess the respective effects of the protein kinase and RNase cytoplasmic domains of IRE1α on EREG expression, we designed an IRE1α mutant (IRE1Δ899) truncated by 78 amino acids at the C-terminal and invalidated for RNase activity (Figure 4a). Three cell clones (R2, R3 and R7) were selected for their expression of the artificial IRE1α isoform and inhibition of ≥ 90% of XBP1 pre-messenger splicing under tunicamycin treatment (Figure 4b). Low levels of MIST1 transcripts were consistently detected in U87Δ899 cells (Figure 4c), in keeping with the fact that MIST1 is a target gene of the mature XBP1 transcription factor [29]. Conversely, IRE1α autophosphorylation (phospho-Ser724-IRE1) was still effective in U87Δ899 clones and was upregulated with tunicamycin (Figure 4d). Thus, the IRE1Δ899 construct acts as a selective dominant-negative mutant of IRE1 RNase and does not notably affect IRE1 kinase activity.

Kinetic expression of EREG was analyzed in U87 cell mutants. EREG mRNA levels were similar in U87Ctrl and in U87Δ899 cells in basal conditions and were transiently and modestly (~2.5-fold) increased in the two cell variants in response to either tunicamycin (Figure 5a) or thapsigargin (not shown) treatments. Again, U87dn mutant cells defective in both IRE1 kinase and IRE1 RNase activities produced much lower amounts of EREG under basal condition, a partial recovery of EREG transcript accumulation being observed after 4 to 8 h of incubation with tunicamycin (Figure 5a). Thus, invalidation of IRE1 RNase activity did not compromise EREG expression whereas the absence of both kinase and RNase functions strongly affected its production. siXBP1 knockdown, which achieved significant silencing of the XBP1 gene, confirmed that EREG expression was independent of the IRE1 RNase/XBP1 axis (Figure 5b).

Since JNK activation can be controlled by IRE1α kinase activity [30], we further investigated EREG production in the presence of the specific pan-JNK inhibitor SP600125. Notably, inhibition of JNK compromises tunicamycin-mediated induction of EREG in both U87Ctrl and U87Δ899 cells after 6h of incubation (Figure 6a). Thus, involvement of the JNK pathway for IRE1-dependent regulation of EREG was irrespective of the IRE1 RNase status. Moreover, tunicamycin partially restored the ability of U87dn cells to accumulate EREG transcripts and this inducible effect was also strongly hindered by treatment with SP600125. Thus, both IRE1-dependent and IRE1-independent pathways may converge in U87 cells toward JNK signaling and EREG expression under tunicamycin treatment. This is also consistent with the fact that JNK phosphorylation was increased by tunicamycin in all cell variants, including U87dn cells (Figure 6b).