Epidermal Growth Factor Receptor Activation in Glioblastoma through Novel Missense Mutations in the Extracellular Domain

DNA Samples

Genomic DNA was extracted from eight glioblastoma cell lines (A172, SF268, SF295, SF539, T98G, U87, U118, and U251) and 143 fresh frozen glioma samples. The clinical glioma samples comprised glioblastomas (n = 132), World Health Organization grade III anaplastic astrocytomas (n = 3), grade III mixed gliomas (n = 4), and grade III oligodendrogliomas (n = 4). Germline genomic DNA was extracted from peripheral blood samples. To confirm the match between germline and tumor DNA for each patient, we performed mass spectrometric genotyping of 24 single-nucleotide polymorphism (SNP) loci. These loci included 23 SNP loci represented on both 50K Xba and Hind arrays (Affymetrix, http://www.affymetrix.com) and one AmelXY locus for sex determination (Table S1). Collection and analysis of all clinical samples was approved by the University of California Los Angeles Institutional Review Board.

Reagents

Erlotinib was purchased from WuXi Pharmatech (http://www.pharmatechs.com). The following antibodies were used in this study: anti-EGFR, anti-phospho-Y1068-EGFR anti-phospho-Y845-EGFR, and anti-phosphoinositide 3-kinase (PI3K) p85 (all from Cell Signaling Technology, http://www.cellsignal.com); anti-phosphotyrosine 4G10 (Upstate Biotechnologies, now Millipore, http://www.upstate.com); and anti-actin, anti-ERK1/2, and anti-P-ERK1/2 (all from Santa Cruz Biotechnology, http://www.scbt.com).

Sequencing and Mass Spectrometric Genotyping

PCR reactions for each exon and flanking intronic sequences contained 5 ng of genomic DNA, 1× HotStar Buffer, 0.8 mM dNTPs, 1 mM MgCl₂, 0.2 U HotStar Enzyme (Qiagen, http://www.qiagen.com), and 0.2 μM forward and reverse primers in a 6 or 10 μl reaction volume. PCR cycling parameters were: one cycle of 95 °C for 15 min; 35 cycles of 95 °C for 20 s, 60 °C for 30 s, and 72 °C for 1 min; followed by one cycle of 72 °C for 3 min.

The resulting PCR products were sequenced using bidirectional dye-terminator fluorescent sequencing with universal M13 primers. Sequencing fragments were detected via capillary electrophoresis using ABI Prism 3730 DNA Analyzer (Applied Biosystems, http://www.appliedbiosystems.com). PCR and sequencing were performed at Agencourt Bioscience Corporation (http://www.agencourt.com) or at the Broad Institute of Harvard and MIT (http://www.broad.mit.edu). Forward (F) and reverse (R) chromatograms were analyzed in batch with Mutation Surveyor 2.51 (SoftGenetics, http://www.softgenetics.com), followed by manual review.

A minimum of 21 of 28 (75%) EGFR exon sequence coverage was accomplished for 151 samples. An exon for each individual sample was considered covered if 90% of the sequence trace within the exon had a phred quality score of 30 or greater, a signal-to-background noise ration of 15% or less, and signal intensity greater than 25% of the signal intensity of the sequencing plate. High-quality sequence variations found in one or both directions were scored as candidate mutations. Exons harboring candidate mutations were reamplified from the original DNA sample and resequenced.

For mass spectrometric genotyping, PCR and extension primers (Table S2) were designed using SpectroDESIGNER software (Sequenom, http://www.sequenom.com). Unincorporated nucleotides from PCR reactions were dephosphorylated with shrimp alkaline phosphatase (Amersham, http://www.amersham.com) followed by primer extension with ThermoSequence polymerase (Amersham). Primer extension reactions were loaded onto SpectroCHIPs (Sequenom) and analyzed using a MALDI-TOF (matrix-assisted laser desorption/ionization time-of-flight) mass spectrometer (SpectroREADER, Sequenom) [10]. Mass spectra were processed with SpectroTYPER (Sequenom) to determine genotypes based on peaks intensities corresponding to the expected extension products.

Affymetrix 100K SNP Arrays

Genomic DNA was processed and hybridized following the guidelines of the manufacturer (Affymetrix) and arrays were scanned with a GeneChip Scanner 3000. Genotyping calls and signal quantification were obtained using GeneChip Operating System 1.1.1 and Affymetrix Genotyping Tools 2.0 software. Data were normalized at the probe level to a baseline array with median signal intensity using invariant set normalization. After normalization, the signal values for each SNP in each array were obtained with a model-based (perfect-match/mismatch) method [11]. Signal intensities at each probe locus were compared with a set of normal reference samples representing 36 ethnically matched individuals to generate log₂ ratios. Log₂ ratios were smoothed using the breakpoint analysis method in the R package GLAD (Gain and Loss Analysis of DNA) [12]. Regions were considered amplified if their smoothed log₂ ratio exceeded 0.3 (half the variation seen with a single-copy gain).

Fluorescence In Situ Hybridization

Dual-probe fluorescence in situ hybridization (FISH) was performed on paraffin-embedded sections with locus-specific probes for EGFR and the centromere of Chromosome 7 as previously described [9].

Determination of EGFRvIII Expression

RNA was extracted from fresh frozen tumor samples and EGFRvIII expression determined by two independent RT-PCR assays for each sample. Primer pairs included: #1F 5′-CTTCGGGGAGCAGCGATGCGAC-3′, #1R 5′-ACCAATACCTATTCCGTTACAC-3′, #2F 5′-GAGCTCTTCGGGGAGCAG-3′, and #2R 5′-GTGATCTGTCACCACATAATTACCTTTCTT-3′. EGFRvIII expression was also examined by immunohistochemistry and/or immunoblotting depending on the availability of tissue samples.

Quantification of Mutant EGFR Alleles

The abundance of missense and wild-type EGFR alleles in tumor DNA samples was determined by PCR-cloning and sequencing of respective EGFR exons. PCR products were ligated into PCR2.1-Topo vectors (Invitrogen) and transformed into E. coli. After transformation, bacteria were plated onto selection plates and grown overnight. For each DNA sample, 65–94 colonies were isolated using a colony picking robot (QPix2, Genetix Limited, http://www.genetix.com), grown overnight, and bidirectionally sequenced at the Broad Institute. Sequence traces were analyzed using Mutation Surveyor software (SoftGenetics).

EGFR Expression Constructs

Retroviral EGFR expression constructs containing puromycin (pBabe-puro-EGFR) [13] or neomycin-resistance genes (pLXSN-neo-EGFR) [14] were used for site-directed mutagenesis using the Quick-Change Mutagenesis XL kit (Stratagene, http://www.stratagene.com). pLXSN-neo-EGFR retroviral constructs for EGFR and EGFRvIII were generously provided by David Riese 2nd (Purdue University, West Lafayette, Illinois, United States) and Webster Cavenee (Ludwig Institute for Cancer Research, La Jolla, California, United States). pBabe-Puro-based viral stocks were generated by transfecting the Phoenix 293T packaging cell line (Orbigen, http://www.orbigen.com) with the pBabe-Puro retroviral constructs using Lipofectamine 2000 (Invitrogen, http://www.invitrogen.com). pLXSN-Neo-based viral stocks were generated by transfecting the human amphotrophic 293-T cell line with pLXSN-Neo retroviral constructs using Lipofectamine 2000 (Invitrogen). Supernatants were collected 24–48 h post-transfection, filtered (0.45 μM), and used to infect NIH-3T3 cells, Ba/F3 cells, and human astrocytes.

Expression of EGFR Alleles in NIH-3T3 Cells

Cells cultured in DMEM supplemented with 10% calf serum were infected with pBabe-Puro-based viral stock in the presence of polybrene. Beginning 2 d after infection, cells were selected in puromycin (2 μg/ml) for 3 d. Pooled NIH-3T3 cells stably expressing respective EGFR alleles at comparable EGFR protein levels were examined for their ability to induce colony formation in soft agar and tumor growth in nude mice. For soft agar assays, 1 × 10⁵ NIH-3T3 cells were suspended in a top layer of DMEM supplemented with 10% calf serum and 0.4% Select Agar (Gibco/Invitrogen) and plated on a bottom layer of DMEM supplemented with 10% calf serum and 0.5% Select Agar. EGF (10 ng/ml) was added to the top agar where indicated. Pictures of colonies were taken 2–3 wk after plating. Colonies were counted from ten random images (40× magnification) taken from each well. Colonies were counted from three replicate wells with the average number represented. In vivo tumorigenicity assays were performed in three mice (two injections/mouse) for each cell line. For each injection, 2 × 10⁶ cells were injected subcutaneously into each nude mouse (Taconic, http://www.taconic.com) and three-dimensional tumor volumes calculated 3–4 wk following injection.

Expression of EGFR Alleles in Ba/F3 Cells

Murine Ba/F3 pro-B lymphocytes [15] were cultured in RPMI 1640 (Cellgro, Mediatech, http://www.cellgro.com) supplemented with 10% FCS, 100 units/ml penicillin and 100 μg/ml streptomycin, 1% L-glutamine, and 10% WEHI-3B-conditioned media. To derive Ba/F3 subclones stably expressing various EGFR alleles, Ba/F3 cells were “spinfected” with pBabe-puro-EGFR-based viral supernatants and spinfection repeated after 48 h. Cells were selected for neomycin or puromycin resistance and maintained in the presence of interleukin-3 (IL-3). IL-3-independent subclones were derived through prolonged passage in IL-3-depleted media. To determine sensitivity to erlotinib, 1 × 10³ cells were seeded in 96-well flat-bottomed plates with the indicated concentrations of erlotinib. Cell proliferation was assessed 48 h postplating using the WST-1 assay (Roche, http://www.roche.com). Each data point represents the median of six replicate wells for each Ba/F3 subclone and erlotinib concentration.

Expression of EGFR Alleles in Human Astrocytes

Viral supernatants (pLXSN-neo-EGFR) were used to infect immortalized human astrocytes expressing the catalytic subunit of the telomerase holoenzyme and human papillomavirus 16 E6/E7 [16]. Astrocytes were then selected in G418 (Invitrogen) for approximately 10 d.

Missense Mutations in Glioblastoma Cluster in the Extracellular Domain of EGFR

Encouraged by the recent success in identifying oncogenic kinase mutations through resequencing of kinase-encoding genes [17–19], we sequenced the entire coding sequence of EGFR in 143 human glioma samples and eight glioblastoma cell lines. Analysis of the initial Sanger sequencing results in these 151 samples revealed several novel sequence variations in the coding region of the EGFR. To validate these candidate mutations via a complementary method, all DNA samples were reexamined using allele-specific genotyping by MALDI-TOF mass spectrometry.

In all, we identified EGFR missense mutations in 14.4% (19/132) of glioblastomas, 12.5% (1/8) of glioblastoma cell lines, and none (0/11) in lower-grade gliomas. Only one tumor sample harbored a missense mutation in the EGFR kinase domain (L861Q), the location of EGFR mutations in lung cancer, supporting the recent conclusion from other groups that EGFR kinase domain mutations appear to be a rare event in this disease [20–22]. The remainder of the EGFR mutations (18/132 glioblastomas) were located in the extracellular ligand-binding (I, III) or cysteine-rich (II, IV) domains of the receptor (Figure 1A). Two evolutionarily highly conserved amino acid residues (Figure S1) were affected by mutations in five samples each (R108 and A289). Examination of peripheral blood DNA, matched to the tumor DNA by genotyping of 24 SNP loci, showed that eight of the 12 distinct missense mutations were unambiguously somatic, and one mutation (E330K) was germline. Three additional missense mutations (A289D, A289T, and R324L) were found in tumors for which no normal tissue was available (Table 1). None of the missense mutations were detected in germline DNA from 270 normal control individuals.

To define which fraction of the EGFR pool represented the mutant allele in gliomas with EGFR missense mutations, we employed a PCR-cloning strategy previously used by our laboratories for mutation detection in clinical samples [23]. The mutant EGFR allele represented 30%–98% of the receptor pool in two-thirds (10/16) of all examined cases and over 50% in at least one tumor representing the most common amino acid changes: R108K, T263P, A289V, and G598V (Table 1). Lower abundance of the mutant EGFR allele in other samples might be due to contaminating stromal tissue, because genomic DNA was extracted from frozen tumor aliquots without prior microdissection.

We also genotyped genomic DNA from 119 primary lung tumors to detect EGFR ectodomain mutations. While 13.4% (16/119) of these lung tumor samples harbored mutations in the EGFR kinase domain, we did not detect any of the glioma-related EGFR ectodomain mutations in this sample set.

EGFR Ectodomain Mutations Are Associated with Increased EGFR Gene Dose

Since EGFR is amplified in about 40% of human glioblastomas [4], we determined the relationship between EGFR missense mutation and EGFR gene dose in our tumor samples. Of 17 tumors with EGFR missense mutations, 58.8% (10) showed evidence for EGFR amplification by FISH and/or Affymetrix 100K SNP genotyping arrays (Figure 1B; Table 1). This distribution suggests that EGFR missense mutations are associated with EGFR amplification and raises the question of whether EGFR missense mutations in glioblastoma co-occur with or are mutually exclusive of the EGFRvIII mutation, which is found almost exclusively in glioblastomas with increased gene dosage [24]. Using at least two independent assays for the determination of EGFRvIII status, we identified the EGFRvIII allele in 28.3% (13/46) of gliomas without EGFR missense mutation and 6.3% (1/16) tumors with EGFR missense mutation (Figure 1C; Table 1); note that this tumor showed vastly lower levels of EGFRvIII (Figure 1C, lane 12). These findings suggest that EGFR ectodomain mutations occur independently of EGFRvIII in glioblastoma and provide an alternative mechanism for EGFR activation in this disease.

EGFR Ectodomain Mutants Are Oncogenic

To test the oncogenicity of the glioma-related EGFR missense mutations, we transduced NIH-3T3 fibroblasts with retroviruses encoding either wild-type EGFR or selected EGFR missense mutants (encoding R108K, T263P, A289V, G598V, and L861Q). Ectopic expression of all EGFR mutants examined in NIH-3T3 cells conferred anchorage-independent colony formation in soft agar (Figure 2A). In contrast, expression of wild-type EGFR induced a transformed phenotype only in the presence of exogenous EGF, as previously reported [25,26].

To further analyze the oncogenic potential of the EGFR mutants, NIH-3T3 subclones stably expressing the same missense mutant receptors (encoding R108K, T263P, A289V, G598V, and L861Q) were inoculated subcutaneously into nude mice. NIH-3T3 cells infected with empty vector or wild-type EGFR-expressing virus did not yield any measurable tumors within the four-week observation period. In contrast, NIH-3T3 cells expressing each of the tested EGFR missense mutants produced large tumors at the inoculation site in all mice within three to four weeks (Figure 2B).

EGFR Ectodomain Mutants Are Basally Phosphorylated and Are Responsive to Ligand

Signal transduction through EGFR is determined by its basal catalytic activity, receptor activation by ligand, and signal termination through intracellular compartmentalization of the receptor-ligand complex, receptor dephosphorylation, and degradation [27]. To explore the biochemical basis for the gain of function observed with EGFR ectodomain mutants, we first examined the basal catalytic activity of A289V-EGFR in transiently transfected 293T cells using EGFR autophosphorylation as a readout for receptor activation. EGFR autophosphorylation was determined by measuring the total phosphotyrosine content of the immunoprecipitated receptor (Figure 3A, left blot) and by immunoblotting of whole cell lysates with several phosphosite-specific anti-EGFR antibodies (Figure 3A, right blot). Compared to wild-type EGFR, the ectodomain mutant A289V-EGFR showed a marked increase in receptor autophosphorylation in the absence of ligand or serum. We subsequently examined a more extensive panel of EGFR missense mutants (T263P, A289V, G598V, L861Q) in immortalized human astrocytes stably transduced with these receptors. Compared to astrocytes overexpressing wild-type EGFR, sublines expressing EGFR missense mutants showed an increased phosphotyrosine content of EGFR and several other unidentified proteins under serum-free conditions (Figure 3B).

We also expressed selected EGFR mutants (R108K, T263P, A289V, G598V, L861Q) in murine hematopoietic cells (Ba/F3 cells) which do not express any EGFR family members [14] but otherwise retain functional properties of the EGF-signaling pathway [28–30]. Consistent with our findings in 293T cells and astrocytes, all examined EGFR ectodomain mutants showed increased tyrosine phosphorylation under serum-starved conditions and were responsive to exogenous EGF (Figure 3C). We also noted that EGF stimulation led to a more pronounced drop of EGFR levels in Ba/F3 cells expressing wild-type EGFR than in subclones expressing EGFR ectodomain mutants (Figure 3C), reminiscent of the impaired ligand-induced receptor downregulation reported for selected EGFR kinase domain mutants [31].

Sensitivity of EGFR Ectodomain Mutants to EGFR Kinase Inhibitors

The presence of identical missense mutations in multiple patient samples and their oncogenicity in standard transformation assays suggest that these mutants play a role in gliomagenesis. It also raises the question whether these mutations might sensitize transformed cells to EGFR kinase inhibitors. Ba/F3 cells provide a unique model system to examine kinase inhibitor sensitivity [32–35] because stable expression of oncogenic kinases in these cells can relieve them from their intrinsic dependence on IL-3 for survival [15,36]. As expected from our results in NIH-3T3 cells, expression of the tested EGFR missense mutants but not wild-type EGFR was able to relieve Ba/F3 cells from IL-3 dependence. Addition of the EGFR kinase inhibitor erlotinib to the media had little or no effect on the viability of parental Ba/F3 cells growing in the presence of IL-3 or on Ba/F3 cells expressing the drug-resistant EGFR double-mutant L858R/T790M-EGFR. However, erlotinib did induce dose-dependent cell death in Ba/F3 subclones expressing the EGFR ectodomain mutants (missense and vIII truncation) or EGFR kinase domain mutants (L858R and L861Q) (Figure 4A). Of note, erlotinib-induced cell death of Ba/F3 cells expressing EGFR ectodomain mutants occurred at IC-50 values of 50–150 nM, drug concentrations that are well below the concentrations achieved in human plasma [37]. These data suggest that EGFR missense mutants sensitize transformed cells to EGFR kinase inhibitors similar to EGFRvIII or lung cancer-related kinase domain mutants, both of which have been associated with clinical responses to EGFR kinase inhibitor therapy [9,19,38,39].

We recently reported the results of a glioblastoma clinical trial with EGFR kinase inhibitors which associated clinical responses to the coexpression of EGFRvIII and PTEN [9]. To investigate whether clinical responses might also be linked to the presence of EGFR ectodomain mutations, we reexamined all available tumor DNA samples from this clinical trial. We identified the ectodomain mutant R108K-EGFR in 14% (1/7) gliomas that responded to erlotinib. This tumor, however, also expressed EGFRvIII, raising the possibility of independent clones arising from a common progenitor with EGFR amplification. We also identified the R108K EGFR mutation in 7% (1/5) gliomas that failed EGFR kinase inhibitor therapy, but loss of PTEN in this tumor provides a potential explanation for treatment failure (Table S3). Larger clinical trials are required to ascertain the contribution of EGFR missense mutants to EGFR kinase inhibitor response in glioblastoma.

Accession Numbers

The GenBank (http://www.ncbi.nlm.nih.gov) accession numbers of the proteins discussed in this paper are human EGFR protein (NM_005228), ERBB2 26 (NM_004448), ERBB3 (NM_001982), ERBB4 (NM_005235); and the EGFR proteins of Mus musculus (NM_207655), Rattus norvegicus (NM_031507), Sus scrofa (NM_214007), Danio rerio (NM_194424), Drosophila melanogaster (NM_057410), and Xiphophorus (X56319)