Emerging data indicate that molecular analysis of patterns of somatic changes within tumor DNA has the potential to improve diagnostic classifications of second tumors as either metastases or independent second primaries. Thus far, studies have focused primarily on cancer sites where second primaries are common, such as those involving the bladder, head and neck, and breast. The typical strategy has been to genotype markers at candidate loci where somatic mutations are known to occur commonly in the tumor under investigation. More recently, the use of aCGH has been applied to this problem, in recognition of the attractiveness of using a comprehensive genome-wide strategy [20
], especially in breast tumors [9
]. Studies of lung cancer using this technology have been rare and not directly targeted at the classification of multiple lung tumors [27
]. Our goal in this study was to apply aCGH and mutational profiling to the problem of multiple primary NSCLCs to determine their usefulness in assessing clonality of tumors.
Our results provide strong evidence that genomic profiling can help distinguish clonal tumors from independent primaries. However, the application of aCGH in normal clinical practice would be challenging. First, the need for fresh frozen tissue would make the technique more immediately applicable for evaluating synchronous tumors. For metachronous tumors, frozen tissues would need to be stored and accessible in an easy manner. Patients who have tumors resected at different institutions would require additional coordination for genomic profiling. Second, standardized software for routine automated statistical comparisons of the array data would need to be widely available, although software for the application of the methods in this article is available from the authors. Third, aCGH is expensive and can be a time-consuming and labor-intensive procedure. Fourth, aCGH requires relatively large amounts of DNA and array quality can be variable, depending upon differences in tumor content, tissue handling, specimen storage, and specimen processing. For example, despite using fresh-frozen tumor samples in an experimental (non-clinical) setting, we found arrays from 2 of the 42 tumors to be of insufficient quality to reach a diagnosis with confidence. The requirement for abundant DNA also may preclude such analysis on DNA derived from needle biopsies prior to surgery, where such results may actually be more useful in deciding whether patients should undergo surgery in the first place.
A more practical solution may be to develop strategies based upon mutational profiling that can be performed using DNA extracted from formalin fixed, paraffin embedded tissue rather than frozen samples. Although mutational profiling is less comprehensive than aCGH, it may yet have sufficient sensitivity for improved diagnoses, as evidenced by the strong concordance of the genomic and mutational profiling results in this study. The strength of the evidence from individual mutational matches depends on the rarity of the mutation. For example, three of our matching cases involved matching KRAS G12V mutations. This specific KRAS alteration occurs at a frequency of 4.4% in lung adenocarcinomas [29
]. Thus, the probability of a match in independent tumors, given that the first tumor has the mutation, is 0.044. Viewed in this context, the simultaneous independent occurrence of this mutation in all 3 tumors for case 20 is an unlikely event with a probability of 0.002 (0.0442
), providing further evidence that these three tumors are of clonal origin. The corresponding population frequencies of the other matches in our study have been estimated to be 5.2% for KRAS G12D and 1.2% for BRAF V600E [30
Conversely, mutational profiling is limited in practice by the infrequency of the occurrences of specific mutations. Indeed, for the panel of markers in our study, none of the mutations were observed for 20 of the 42 tumors, and none were observed in either tumor for 7 of the 20 patients. Mutational profiling would be more powerful if assays could detect all known loci that are commonly mutated at the somatic level in NSCLC. A panel of about 20 commonly occurring markers would have good statistical power for detecting clonal tumors, provided that the preponderance of the somatic events observed occur in the originating clonal cell [31
]. Additionally, mutational profiling would allow the use of partially-fragmented DNA, such as can be obtained from paraffin-embedded tissue, or of whole-genome-amplified DNA, when starting tumor DNA quantities are limited (i.e. derived from bronchoscopic or needle biopsies). Importantly, one of us (ML) has demonstrated that similar results are obtained from mutation testing of EGFR and KRAS using matching formalin fixed and frozen specimens (M Ladanyi, unpublished data).
Our results suggest surprisingly that diagnostic misclassifications may occur frequently when using the Martini-Melamed guidelines with histologic classification according to traditional WHO criteria. Out of 22 evaluable comparisons of tumor pairs, the genomic classification contradicted the clinical diagnoses in 4 (18%), identifying clonal patterns of metastasis (as in the case highlighted in ) in 3 cases diagnosed as independent primaries. Based upon our statistical algorithm, it is highly unlikely that such identical or near-identical allelic changes would have occurred in the tumors independently by chance. In another 3 cases deemed multiple primaries by clinical criteria but diagnosed as equivocal by aCGH, the identification of matching somatic point mutations by mutational profiling suggested that the tumors were of clonal origin. Thus, potentially up to 7 of the 22 cases (32%) were misclassified by the traditional clinical/histologic criteria alone.
We recognize that we cannot state with absolute certainty that the diagnoses based on genomic profiling are “correct” and that the clinical diagnoses were “wrong”. However, the kinds of precise matching of regions of allelic losses and gains that we observed seem compelling as a definitive basis for diagnosis from a logical standpoint. Prospective studies with much larger sample sizes will be necessary to validate the usefulness of genomic analyses in relation to patient survival. We further note that this study was based upon a highly selected, small sample size. To be included for study, the metachronous patients all had to survive long enough to develop multiple primaries and undergo resection.
Distinguishing multiple primary tumors from metastases may actually not change the initial management of these patients, as surgery has been shown to increase survival in these instances [4
]. However, in the 3 cases initially considered as independent primaries, reclassification as metastases using genomic profiling could have led to consideration of adjuvant chemotherapy and/or to a closer surveillance monitoring plan. Conversely, patients found to have multiple primaries rather than metastases could be spared from adjuvant chemotherapy.
In summary, our results suggest that genome-wide examination of copy number changes using aCGH and mutational profiling of multiple loci can help distinguish multiple independent primaries from metastases. Molecular diagnostics such as these should be honed for practical application in the clinic.
STATEMENT OF TRANSLATIONAL RELEVANCE
The incidence of multiple primary lung cancers is rising. When patients are found to have multiple primary lung lesions, clinicians must decide whether they have independent tumors or metastases and tailor treatment accordingly. Decisions are currently made using clinical and pathological criteria. Here, we performed a comprehensive molecular and clinico-pathological analysis of 20 patients originally presumed to have multiple primary lung cancers. Tumors were analyzed by array comparative genomic hybridization (aCGH) and profiled for the presence of multiple somatic mutations. Surprisingly, classification based upon genomic/mutational profiling contradicted the clinico-pathologic diagnosis in 4 (18%) of the comparisons. Although the study involved limited numbers of samples, the results suggest that clinico-pathologic criteria used to distinguish multiple primaries from metastases could be further refined by the routine use of molecular profiling.