In order to clarify some of the conflicting data from clinical studies, we performed an in vitro study of deregulation of the EGFR signaling pathway in a large panel of lung cancer cell lines and correlated the results with TKI sensitivity. In contrast to SCLC, mutations or CNGs of the studied genes were frequent in NSCLC – 84.4% of NSCLC cell lines had a mutation, one or more CNGs or both. Thus the EGFR pathway is deregulated at high frequency in NSCLC.
Somatic mutations or CNGs of the studied genes were rare in SCLC and further studies were limited to NSCLC. About half of the NSCLC lines had a mutation in one of the genes of the EGFR
pathway. The relative distribution of the mutations of EGFR
pathway genes in established NSCLC cell lines were very similar to those reported from large clinical studies 
(13%) and KRAS
(24.7%) mutations were more frequently observed than other gene mutations. We found a single mutation of HER2
4 and no mutations of EGFR
3. Mutations of BRAF
were intermediate in frequency. In general mutations were more common in adenocarcinoma subtype and relatively rare in squamous cell carcinomas except for PIK3CA
which was more common in squamous cell carcinomas, as also occurs in tumors 
. Mutations were mutually exclusive except for a single cell line that had both KRAS
mutations. We have previously reported that mutations of EGFR
, BRAF and HER2
were mutually exclusive in 691 resected NSCLC tumors indicating that a single activating mutation in the EGFR
-RAS-RAF signaling pathway may be sufficient for the pathogenesis of many lung cancers 
. However, the mutational status of PIK3CA
was not mutually exclusive, which is similar to another recent report 
. We have postulated that the PIK3CA
signaling pathways closely interact but that PI3K signaling represents a partially independent pathway 
Copy number gains are believed to be an important mechanism for activation of the EGFR
pathway genes and the downstream signaling network. CNGs have been difficult to interpret due to the differences stemming from the use of different methods. Our comparison of the three methods yielded excellent concordance for binary values after we used kappa statistics to determine the optimal cut off values for increased copy numbers. However, the three methods we compared have their individual advantages and disadvantages. CNGs result from gene over-representation due to chromosomal polysomy or focal amplification. FISH, a widely used technique to assess gene copy number takes can identify gains due to both mechanisms. On the other hand, not only is it expensive and time consuming but it analyzes only a small subset of cells (~100 cells) to assess the copy number, and tandem segmental duplications may not always be detected. Thus intratumoral heterogeneity is difficult to assess. Both aCGH and qPCR analyze DNA from many thousands of cells, but because they compare values for the test gene against an internal standard (a reference locus), they can detect focal amplifications more readily than polysomy. In the case of qPCR, if the amplicon encompasses the reference locus, then amplification may be underestimated. In the case of aCGH, the copy number deduced from the signal ratio can appear lower than those obtained by the other two techniques, for aCGH measures relative and not absolute copy number. It does not take into account changes in ploidy and therefore will dampen ratio shifts for copy number gains in samples with a high DNA index 
. Thus, the absolute copy number for a given chromosome region is best determined by using a combination of complementary methods 
. Despite the shortcomings of each method, we found a high concordance among the three techniques when using binary values (normal/increased), similar to findings in a recent comparison 
. We had most extensive data for gene copy number generated via qPCR and used it for all our further analysis.
Sixty seven percent of the NSCLC lines had CNGs for one or more EGFR pathway genes. CNGs for EGFR (40.3%), HER2 (18.2%) and KRAS (14.3%) were more frequent than the other genes. BRAF CNGs (13%) were more common in adenocarcinoma subtype and PIK3CA CNGs (36.4%) were more frequent in squamous cell carcinoma as compared to the other subtypes. Other genes did not show subtype bias. Unlike mutations, CNGs were not mutually exclusive either with other CNGs or with mutations.
However, CNGs were not completely random. We identified that EGFR
CNGs were significantly more frequent in EGFR
mutant cell lines respectively. Using subcloning, we determined that in almost all mutant lines with CNGs, the mutant allele was in great excess in comparison to the wild type allele. In some lines with diploid amounts of the mutant gene, the wild type allele was absent or in minute amounts. This finding for oncogenes (as compared to tumor suppressor genes) has been described, particularly in hematologic malignancies 
, and represents a form of acquired uniparental disomy. Taken together, we term this phenomenon as mutant allele specific imbalance (MASI). Thus MASI may result from specific amplification of the mutant allele, or by loss of the wild type gene, or by a combination of these events. Of interest, wild type KRAS
may function as an inhibitor of tumorigenesis and thus represent a form of tumor suppressor gene 
. These results suggest that the combination of two methods of activating the oncogenes EGFR
(i.e. mutation and MASI) may confer a greater growth or survival advantage to the malignant cell than a single method. These findings also indicate that mutations must precede CNGs in cells harboring both changes. Other evidence for the combination of mutations and CNGs in tumor cells exists 
, and for the concept that mutations are the initial event 
EGFR mutations were exclusively found in adenocarcinoma histology and in never smokers or individuals with low smoke exposure (<15 PYR). In addition EGFR mutations were more common in a comparatively younger age group (<55 years). EGFR mutations did not demonstrate gender bias. Most mutant lines were derived from Caucasians, which is not surprising considering that 90% were established in the USA. In rank order of sensitivity to gefitinib, nine of the EGFR mutant lines were represented in the 12 most sensitive lines, in keeping with the previously discussed importance of mutations in sensitivity to TKIs.
We found an excellent concordance between the IC50 values between gefitinib and erlotinib. Depending on the previously determined in vitro threshold for gefitinib and the bimodal curve we classified our NSCLC cell lines into 3 categories: sensitive, intermediate and resistant. We had nine sensitive cell lines which included seven of the 10 EGFR
mutant cell lines, one cell line with EGFR
CNG and one with HER2
CNG. Of the remaining 3 EGFR
mutant cell lines, two had the resistance associated secondary T790M mutation. The third resistant EGFR
mutant line had a deletion of the PTEN
gene, a finding associated with TKI resistance in other systems 
Pao et al demonstrated that mutations in KRAS
are associated with a lack of sensitivity to TKIs 
. We tested the hypothesis that KRAS
mutations confer intrinsic resistance to TKIs and our univariate analyses demonstrated that indeed KRAS
mutations conferred in vitro resistance to gefitinib. Thus KRAS
mutations are associated with both clinical and in vitro resistance to gefitinib.
None of the other gene mutations or copy number gains showed any significant correlation with gefitinib sensitivity. In the clinical setting, approximately 10–20% of NSCLC patients who do not harbor identifiable EGFR mutations respond partially to gefitinib. Evidently EGFR mutations are one of the most important but not the sole determinant of TKI response.
We evaluated the effect of EGFR mutation, EGFR CNGs and HER2 CNGs on TKI sensitivity using a multivariate regression analysis. To summarize the multivariate analysis, all 3 parameters, i.e. EGFR mutations, EGFR CNG and HER2 CNG, showed a correlation with the TKI sensitivity (used as a continuous variable) in decreasing order of importance. Thus three previously identified factors related to patient response to TKIs are major factors in the in vitro response. However, the most important independent factor is the presence of activating EGFR mutations. Our findings may assist in the prediction of response and the selection of patients for targeted therapies. Another important finding was the frequent presence of selective imbalance of the mutant form (MASI) of oncogenes in tumor cells that harbor mutations of EGFR and KRAS. While MASI may confer a selective advantage to the tumor cell, its effect on clinical course or response to therapy remains to be determined.