Lung carcinoma is the first cause of death by cancer in the world, mostly because patients have an advanced stage disease at diagnosis 1 . Adenocarcinoma (ADC), the most frequent histological type is morphologically and biologically heterogeneous. Different architectural patterns have been described and different molecular pathways are involved in the carcinogenesis process 2 . Attention has largely been focused on proliferation pathways with the identification of mutations in oncogenes such as KRAS, EGFR, ALK, HER2, PI3KCA and BRAF that are potential or validated drug targets 3 . The first identified target in NSCLC was the EGF receptor. In 2004, EGFR activating mutations were identified in lung ADC and rapidly associated with response to EGFR-TKI 4 5 . Clinico-pathological features that correlate with these mutations include east-Asian ethnicity, adenocarcinoma histology, female sex, and never smoking history. In lung cancer the prevalence of EGFR mutations varies from 10% in Caucasians to more than 40% in Asian populations 6 . They are mainly located in the tyrosine kinase domain and 90% consist of either small deletions in exon 19 (DEL19) or a missense mutation in codon 21 that changes the leucine 858 in an arginine (p.L858R). Concerning rare alterations, about 3% of the mutations occur at codon 719, resulting in the substitution of glycine by a cysteine, alanine or serine (p.G719X) or at codon 861 (p.L861Q) 7 8 . In addition, there are rare 1 to 2% in-frame insertion mutations in exon 20 9 . The predictive value of frequent alterations (DEL19 and p.L858R) is more or less equivalent but some studies have reported a higher sensitivity and longer PFS for patients with DEL19 mutated NSCLC 10 . Concerning rare alterations, sensitivity to EFGR-TKI and PFS are globally lower 11 . In 2010, results from large phase III trial led to the restriction of EGFR tyrosine kinase inhibitors to EGFR mutated tumors in first line treatment 12 . EGFR mutational status has therefore became mandatory to determine which therapy will be the most appropriate to patients with stage IV diseases. In this context, genetic heterogeneity is an obstacle to correct determination of EGFR status on small biopsies specimens. Previous studies showed that the EGFR mutation status was discordant in different parts of the tumor or between primary or secondary metastatic sites 13 14 15 16 . At the opposite, Yatabe et al showed in a series of Asiatic patients that discordant cases where extremely rare 17 and it was suggested that discrepancies regarding EGFR mutations distribution could be due to methodological procedures 16 17 18 . If EGFR genotyping results depend on sample types it will negatively impact treatment decisions. In Caucasians, EGFR molecular status at various tumor sites remains to be examined in standard testing conditions to validate EGFR molecular testing as a diagnostic tool. Finally, EGFR mutational heterogeneity could also explain the occurrence of secondary EGFR-TKI resistances. Patients undergoing EFGR-TKI treatments will ultimately relapse. Recurrences are related to various mechanisms among which the emergence of EGFR p.T790M clones seems to be the most frequent. This alteration, present as a minor sub-clone before treatment seems selected by EGFR-TKI treatments 19 . Most methods used for molecular diagnostic are not sensitive enough to detect minor p.T790M subclones (<1%) and this alteration is rarely identified in untreated patients. Indeed, the reported frequency of baseline EGFR p.T790M mutations varies widely in the literature, ranging from 1% of all EGFR-mutant lung cancers 11 to 35% of all EGFR-mutant lung cancers 20 and depends of the sensitivity of the assays used and their ability to identify minor clones within a tumor. The prognostic significance of baseline EGFR p.T790M has not been reported. In the acquired resistance setting, it has been demonstrated that the presence of p.T790M predicts a favorable prognosis and indolent progression, compared to the absence of p.T790M after TKI failure 19 .

Because no large Caucasian series was tested for EGFR genetic heterogeneity, we addressed this question in clinical testing conditions thanks to a French nationwide EGFR mutation characterization program in advanced lung cancer (National Cancer Institute, INCa).

The aim of this study was to answer several questions of clinical relevance: -i- is EGFR mutation distribution, including p.T790M, heterogeneous within the primary tumor?; -ii- is EGFR mutation distribution heterogeneous between the primary tumor and thoracic metastases ?; -iii- Are microbiopsy or cytology samples suitable for EGFR screening? and iiii- is EGFR copy number heterogeneous within the tumor?

1- EGFR mutation status on multiple spatially separated samples:

From January 2010 to March 2011, 357 patients managed at the Hotel-Dieu hospital for lung cancer had EGFR testing for clinical purpose. Among the 57 EGFR mutated tumors, 40 patients had enough available tissue for multiple EGFR testing. Nineteen patients had an exon 19 deletion (DEL19), 12 a p.L858R point mutation, 6 an exon 20 insertion (INS20), 2 a p.G719A mutation and 1 a p.L861Q/p.T790M double alteration. For those 40 patients, 153 DNA samples were extracted from various tumor localizations and reanalyzed for the initial alteration. Four samples could not be amplified (2 DEL19 and 2 INS20 tumors). The initial EGFR mutation was identified in 144 of the 149 informative samples. Five samples with an expected DEL19 (n = 3) or INS20 (n = 2) were found wild type. For those 5 cases, 2 cytologic specimens (pleural effusion and bronchial aspiration), 1 small endobronchial biopsy and 2 surgical specimens, the proportion of tumor cells was low, ranging from 2% to less than 10%. One specimen (P3) was rescued by DEL19-CAST probes while both samples from P16 remained wild type (Table  2). CAST probes were not available to test INS20. There was no discordance for samples with expected p.L858R, p.L861Q/p.T790M or p.G719A mutations (Table  2 and Additional file 1: Table S1).

All specimens were tested for the p.T790M mutation. This alteration was identified in one tumor (P40). Patient was naïve of EGFR-TKI, the tumor was p.L861Q/p.T790M mutated, was divided into 19 parts and both mutations were homogeneously distributed with similar allele intensity (Additional file 1: Table S1). Samples that were initially diagnosed EGFR wild-type, were found wild-type on all sub-specimens (Additional file 2: Table S2).

2- Concordant EGFR mutation status was found within the tumor, and between primary and thoracic metastasis.

In our series, 22/40 patients had surgical resection for adenocarcinoma allowing classification according the IASLC/ATS/ERS recommendations 2 . All architectural patterns were represented except the mucinous pattern. Most tumors had a lepidic counterpart (9/22) but it was not necessary the predominant pattern (Additional file 1: Table S1). For these 22 adenocarcinomas, we analyzed different tumor area, displaying similar or different architectural patterns and no EGFR heterogeneity was seen within these specimens.

For 10 patients we analyzed both primary tumor and lymph node metastases (n = 8) or pleural metastases (n = 2). A single discordant result was found between the primary tumor (INS20 mutation) and one metastatic lymph node containing 15% of tumor cells. For this patient, the mutation was present in the other lymph node metastasis (P27, Table  2, Additional file 1: Table S1).

Two patients showed a pre-invasive lesion (atypical adenomatous hyperplasia and in situ adenocarcinoma) located at distance of the invasive adenocarcinoma. These pre-invasion lesions had the same EGFR mutation as the invasive counterpart (Additional file 1: Table S1).

3- Microbiopsy and cytology samples allow EGFR mutation analysis.

Twenty non-surgical specimens (8 pleural effusions, 2 bronchial aspirations, 5 tomography-guided needle lung biopsies, 5 bronchial biopsies) were analyzed including 8 with less than 15% tumor cells. Nineteen samples were contributive, 16 were EGFR mutated and 3 were wild type: 1 pleural effusion, 1 bronchial aspirate and 1 bronchial biopsy with 5, 5 and 10% tumor cell content respectively (Table  2).

4- Determination of EGFR copy number.

EGFR copy number (CN) was available in 132/153 samples. In this series, 26/132 samples had an EGFR CN > 2.5 and 2 an EGFR CN >5. Half of the patients (18/40) had at least one sample showing an increased EGFR CN but only one patient had homogeneous copy number increase on all fragment analyzed (CN >4 on 3 pleural biopsy fragments, P11) (Table  3).

Shows the type and number of samples tested for EGFR copy number variation (CNV) in 40 patients. Non-contributive samples are non-amplified specimens or samples with Ct >30.

Lung biopsy: * and bronchial biopsy: £.

Concerning specimens with very low tumor cell content (≤10%, n = 25), 5 were found wild type and 2 were non informative, none of these 7 specimens had an increased EGFR CN. At the opposite, among the mutated specimens with low tumor cell content 4 /18 had a CN ≥ 2.5 suggesting that EGFR CN impacts on mutation detection for low tumor cell content specimens.

We next examined whether increased EGFR CN was associated with specific morphologic features. In our experience, there was no clear association with high-grade lesions (solid or micropapillary predominant patterns) but in one patient, high EGFR CN was found in the solid pattern and not in the lepidic counterpart (Figure  1). At the opposite, 1 out of the EGFR mutated precursor lesions (in situ adenocarcinoma and atypical adenomatous hyperplasia), had an increased EGFR CN (CN: 3.4). No EGFR copy number increase was found in EGFR wild type tumor samples.

Tyrosine kinase inhibitors (TKI) have changed advanced EGFR mutated lung carcinoma clinical practice with global improved short-term survival and fewer side effects 27 . In routine clinical practice EGFR mutation screening is mandatory to decide which first line treatment would be the most appropriate 28 29 . Heterogeneous repartition of EGFR mutations within tissue or between different metastatic sites is an obstacle to accurate molecular screening. It was reported in different studies and remains an important question for clinicians 13 14 15 30 31 32 . Table  4 summarizes previous works and compared samples and technologies. Main differences are, the inclusion of various cancer types including squamous-cell cancers that are not targets for EGFR-TKI treatments or pre-treated samples, different proportion of smokers, different types of metastatic site and finally different technologies and different panels of mutations tested. It seems that higher proportion of smokers, heterogeneous tumors specimens and the use of low sensitivity methods yielded to higher rates of inconsistencies in EGFR mutation results 33 34 . Moreover lower rates of concordance are also found for rare EGFR variants 34 35 . It seems therefore important to analyze in a series of patients prospectively tested for EGFR mutation in clinical settings, the impact of heterogeneity on diagnosis. Indeed, non-surgical specimens from either the primary or metastatic site often constitute the only tissue available for molecular diagnosis in patients eligible for TKI with advanced stage cancers 13 30 32 36 37 . Are these small specimens reliable for molecular testing? We addressed this question in Caucasian patients with EGFR mutated lung tumors. Multiple samples were obtained either from different localizations or different loci within the primary tumor. As expected in Caucasians, the frequency of EGFR mutated tumor was 15% 38 . The p.T790M resistance mutation was studied in all patients (153 specimens) that were EGFR-TKI naïve. In untreated patient, the p.T790M mutation is usually described as a subclonal alteration that is difficult to identify even with highly sensitive methods 39 . Here the alteration was found once concomitantly to a p.L861Q mutation, both alleles were equally represented and the distribution was homogeneous suggesting that in this case, the p.T790M alteration is a driver mutation altogether with the p.L861Q. Routine EGFR testing gives a qualitative analysis of EGFR mutations, tumor is either mutated or non-mutated and the proportion of mutated allele is not taken into consideration. Using this definition, we found that 149/153 samples were contributive; among these 149 samples, only 5 showed a discordant genotype (WT/M). Discordant results were from low tumor cell content specimens expected to be either DEL19 or INS20. Fragment analysis, the method used to detect these alterations has a higher detection threshold as compared to allelic discrimination (10% versus 1%) 24 , suggesting that these results might be false negative. Among the samples that could be tested using the DEL19 TaqMan® Mutation Detection assay one was found positive while both specimens from patient P16 remained negative. This patient had 2 primary cancers, concomitant adenocarcinoma and squamous cell cancer. Initial EGFR mutation was found on a bronchial biopsy sample showing an ADC, a second biopsy and the bronchial aspiration were found EGFR wild type even though high sensitivity CAST PCR was used. For this patient, the existence of 2 cancers might explain the presence of EGFR mutated and non-mutated samples. Indeed diagnosis on small specimens may be equivocal. To validate the accuracy of EGFR wild type status, 10 patients with a diagnosis of EGFR wild type tumor were re-analyzed at different loci. No EGFR alteration was found (mutation or increased CN). It suggests that negative results can be trusted as long as the method detection cut-off matches the tumor cell content. Although our work cannot rule out the existence of minor wild type subclones, we believe that any sample allows accurate EGFR mutations detection at initial diagnostic. This fits the results of Yatabe et al 17 . However, discrepancies may be increased by the use of low sensitivity methods such as sequencing. In this work, 38 samples were analyzed by direct sequencing. We found 23 concordances, 3 discordances and 12 informative sequences because of high background noise. We stopped the comparison as this method is not used in diagnosis in our lab and was shown to be inappropriate for biopsy or cytology FFPE samples. Indeed results depend on the estimation of the tumor cell content, on the method’s detection cut-off and on EGFR copy number. What threshold of tumor nuclei should we use? It is assumed that more than 50% of tumor cells allows accurate molecular testing and that samples under 10% may lead to false negative results 40 . Between 10% and 50% the reliability depends on the laboratory experience. Interestingly, we had detectable EGFR mutations in 18/23 contributive samples with ≤ 10% of tumor cells. It highlights the fact that if a positive result can be found in a sample with low tumor content, negative results have to be validated according to the specimen’s tumor content. Pleural metastatic evolution is frequent in lung cancer and pleural effusion and/or tissue are convenient materials for molecular biology.

Schematic review of previously published series comparing primary tumor and metastasis or different loci within primary tumor. Tumor type, smoking status, detection methods, mutation tested and metastatic sites are given. PT : primary Tumor, M: metastasis, ADC: adenocarcinoma, SCC: squamous cell carcinoma, ADSQ : adenosquamous carcinoma, LCC: large cell carcinoma mt: mutated, wt : wild type, ARMS: amplification refractory mutation system.

We secondly tested whether EGFR CN increase could explain why EGFR mutation assessment, remain possible for cases with very few tumor cells. We found that EGFR copy number was heterogeneous between different fragments from the same tumor. Thus, the allelic ratio of each specimen depends not only on the tumor cell content but also on the EGFR copy number and on the number of mutated allele in tumor cells. In clinical settings the proportion of EGFR mutated cells or the presence of an associated increased copy number, which is frequent in EGFR mutated tumors, is not taken into account. In samples with various tumor cell contents and various EGFR gene copy number the quantification of the mutated allele/wild type allele ratio is difficult. It is clear to us that this ratio varies between tumor sites and that low tumor cell content specimens are rescued by the presence of a high mutant/wild type allele ratio. At the opposite, high tumor cell content specimens may show low mutant/wild type allele ratio estimated by the intensity of the mutant probe signal. This variability may explain discrepancies between series using different detection strategies or using microdissection of very few tumor cells as PCR amplifications in those conditions may lead to amplification errors 44 . However, it would be important to set up assays to analyze the allelic ratio’s impact on response to treatment. One other question was the link between histology and EGFR mutation or CN. In our experience, no difference in EGFR mutation status was found according to adenocarcinoma architectural patterns. This is in accordance with a report on a small series of samples (n = 11) from a Swedish group 18 . As already reported EGFR copy number was variable within the same tumor and may vary according to the architectural pattern 17 45 . Here, for one tumor, EGFR increased copy number was restricted to the solid counterpart as compared to lepidic pattern. But this could not be generalized as other solid or micropapillary predominant pattern tumors had no EGFR CN increase and at the opposite a preinvasive lesion showed a high EGFR copy number. Finally, our results suggested that the role of EGFR amplification in cancer progression might not be directly linked to the histological tumor grade.

Patients undergoing EGFR TKI treatment will develop resistance after several months. Heterogeneity has been described as a phenomenon that could drive secondary resistance. Issues are either the development of a p.T790M or a wild type subclone 15 30 39 . In our series, there was no p.T790M heterogeneity. Considering that our detection method (CAST-PCR) can detect approximately 1 to 2% of mutated alleles in a background of wild type this might not have been sufficient for p.T790M subclone detection 24 . It was also suggested that secondary resistances might be due to EGFR mutation loss. Our study was not designed to detect the presence of EGFR wild type subclones however, if this wild type tumor cell population exists, it is not an obstacle to initial EGFR molecular diagnosis defined by the qualitative presence of the EGFR mutation.

This retrospective series reports multiple EGFR testing in lung cancer in routine diagnostic conditions and validates that molecular testing from single tumor-biopsy sample before first line EGFR-TKI may be conducted on any specimens. This is an important result for clinical practice, it indicates that EGFR testing is relevant on biopsies, cytological samples, lymph nodes and metastatic sites using standardized and validated procedures with a defined detection cut-off. This existence of multiple primary tumors with possible distinct genotypes needs to be considered and the development of sensitive methods is recommended because the wild type/mutated allele ratio is unpredictable. Finally, the clinical impact of the associated EGFR copy number increase in a subset of samples remains to be evaluated.

Following authors have declared that they do not have any conflict of interests: Mansuet-Lupo A, Zouiti F, Alifano M, Charpentier MC, Ducruit V, Devez F, Lemaitre F, Damotte D and Blons H Pierre Laurent-Puig declared being a consultant or advisory role for Amgen, Roche, Merck-Serono, Genomic Health, Astellas, Sanofi, Integragen and Raindance.

AL carried out the molecular genetic studies, carried out tumor classification according to the IASLC/ATS/ERS recommendation, analyzed the data, and drafted the manuscript. FZ, FL carried out the molecular genetic studies, analyzed molecular data. MA recruited the patients, drafted the manuscript. CMC carried out tumor classification according to the IASLC/ATS/ERS recommendation. DV, DF carried out the pathology laboratory work. DM, PLP participated in the design of the study, performed analysis and helped to draft the manuscript. HB conceived of the study participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript.