See editorial on pages 4370–1 and related article on pages 4486–94, this issue.
The discovery of distinct subsets of nonsmall cell lung cancer (NSCLC) characterized by activation of driver oncogenes has greatly affected personalized therapy. It is hypothesized that the dominant oncogene in NSCLC would be associated with distinct patterns of metastatic spread in NSCLC at the time of diagnosis.
A total of 209 consecutive patients with stage IV nonsquamous NSCLC with an EGFR (epidermal growth factor receptor) mutation (N = 39), KRAS (v-Ki-ras2 Kirsten rat sarcoma viral oncogene homolog) mutation (N = 49), ALK (anaplastic lymphoma receptor tyrosine kinase) gene rearrangement (N = 41), or wild-type for all 3 (triple negative, N = 80) were included. The percentage of patients with metastatic disease at a given site was compared between each molecular cohort (EGFR, KRAS, or ALK) and the triple negative cohort.
ALK gene rearrangement was significantly associated with pericardial disease (odds ratio [OR] = 4.61; 95% confidence interval [CI] = 1.30, 16.37; P = .02) and pleural disease (OR = 4.80; 95% CI = 2.10, 10.97; P < .001). Patients with ALK gene rearrangements (OR = 5.50; 95% CI = 1.76, 17.18; P = .003) and patients with EGFR mutations (OR = 5.17; 95% CI = 1.63, 16.43; P = .006) were predisposed to liver metastasis compared to the triple negative cohort. No molecular cohort had a predisposition to pulmonary nodules, or adrenal, bone, or brain metastasis compared to the triple negative cohort. The mean number of metastatic disease sites in patients within the ALK rearranged cohort was significantly greater than that of the triple negative cohort (mean = 3.6 sites vs 2.5 sites, P < .0001).
For a long time, nonsmall cell lung cancer (NSCLC) was treated as a single entity without regard to histology or molecular status. Over the last decade, it was recognized that histology can predict both efficacy and safety of drugs used for the treatment of NSCLC.1, 2 Molecular analysis has provided an even more detailed classification of NSCLC. Activating mutations in the epidermal growth factor receptor (EGFR) gene are both prognostic and predictive in that they are associated with improved survival, irrespective of therapy, and are associated with a significant response to EGFR tyrosine kinase inhibitors such as gefitinib or erlotinib.3, 4 Patients with EGFR mutations also show a significant improvement in progression-free survival (PFS) compared to standard chemotherapy.5 More recently, fusions involving the anaplastic lymphoma receptor tyrosine kinase (ALK) gene were discovered in NSCLC.6 Patients with ALK gene rearrangements detected by fluorescence in situ hybridization (FISH) demonstrate significant objective response rates and PFS times to the oral ALK inhibitor crizotinib.7 The prognostic significance of ALK is somewhat unclear, because studies in untreated, unselected populations are not yet available, although it was recently reported that ALK did not portend a favorable prognosis in NSCLC.8 Despite being the earliest recognized and the most frequently activated oncogene in lung cancer, KRAS (v-Ki-ras2 Kirsten rat sarcoma viral oncogene homolog) mutations do not currently predict for benefit from any targeted or chemotherapeutic drugs and are associated with a worse survival.9-11
Subclassification of patients with NSCLC through use of molecular diagnostics has permitted us to reexamine the characteristics and outcomes of patients with NSCLC. Indeed, evaluation of PFS for patients treated with pemetrexed showed a significant benefit for ALK-positive patients compared to patients without ALK gene rearrangement, EGFR mutation, or KRAS mutation (known as the triple negative cohort).12 We initially made a clinical observation that a number of ALK-positive patients had metastatic disease to the pericardium. We hypothesized that the biology of the tumor would, at least in part, regulate patterns of metastatic spread. We therefore formally analyzed patterns of metastatic spread by comparing different molecular cohorts based on current dominant oncogenes recognized in NSCLC. Here, we describe the first study to examine variations in the patterns of metastatic spread in NSCLC by oncogenic driver status.
MATERIALS AND METHODS
The University of Colorado Thoracic Oncology Program began in 2008 to screen tumor biopsies from patients with NSCLC for a series of different molecular drivers. In addition to routine EGFR and KRAS mutational testing, ALK gene rearrangements were also assessed to identify patients for entry into defined molecular cohorts treated within the phase 1 study of crizotinib (PF-02341066).5, 8, 11 Our initial ALK screening strategies intentionally enriched for those likely to be ALK-positive, including not testing those who were previously proven as EGFR or KRAS mutant, leading to a higher than expected prevalence of ALK-positive patients at the University of Colorado.13 However, in a desire to capture all ALK-positive cases, we later adopted a policy of screening all NSCLC cases with available tissue.14
A protocol approved by the institutional review board permits clinical correlates to be made on all patients in whom molecular analyses have been conducted within the Colorado Molecular Correlates Laboratory (CMOCO). All patients with NSCLC who had stage IV cancer (classified by TNM 7th edition15) tested within CMOCO from June 2008 to May 2011 were eligible for assessment if there was either full triple testing results available (EGFR and KRAS mutation status and ALK gene rearrangement as determined by FISH) or at least 1 positive result (ALK-positive as determined by FISH, KRAS mutation, or EGFR mutation) was found. Patients with more than 1 positive result in ALK, EGFR, and KRAS were excluded. Data were collected by retrospective chart review and review of imaging for each patient, capturing NSCLC histology, age at metastatic diagnosis, sex, and smoking status and presence of metastatic disease in the predefined sites of brain, bone, adrenal, liver, pleura, pericardium, and intra- and extrathoracic lymph nodes at the time of diagnosis or metastatic recurrence. Other sites of metastatic disease were noted but not formally analyzed due to the predicted low prevalence. Patients with squamous cell histology were excluded out of concern that histology may be significantly associated with sites of metastatic disease and the fact that the prevalence of the predominant gene abnormalities currently tested for would be significantly lower in this population, thereby skewing the triple negative cohort. Patients were classified as either never smokers if they smoked fewer than 100 cigarettes, light smokers if they smoked ≤10 pack-years, former smokers if they smoked >10 pack-years and quit more than 1 year prior to diagnosis, and current smokers if they smoked >10 pack-years and were still smoking within 1 year of diagnosis. Patients were excluded if clinical information was incomplete.
Mutational analyses were conducted, as previously described, at CMOCO.12 Briefly, EGFR exons 18-21 and KRAS exon 2 were amplified and sequenced using an ABI model 3730 capillary gel sequencer. Mutations were identified by visual inspection of the resulting chromatograms that were aligned with Mutation Surveyor, version 3.24 (or higher). In some cases, mutation analysis in EGFR and KRAS was confirmed by SNaPshot analysis, as previously described.16 The occurrence of an ALK gene rearrangement was assessed by FISH, as previously described, in the cytogenetics laboratory within CMOCO.13 Patients were deemed ALK FISH–positive if >15% of tumor cells showed split red and green signals and/or single red (residual 3′) signals. Otherwise, the specimen was classified as ALK FISH–negative.
Descriptive analyses were performed on 4 molecularly defined groups: ALK FISH–positive, EGFR mutant, KRAS mutant, and triple negative and associated clinical data. Within each site of metastasis, a Fisher exact test was used to compare the proportion of patients falling in 1 molecular cohort to that for the triple negative cohort; thus, 27 comparisons were analyzed. One-way analysis of variance was used to analyze the differences on number of metastatic sites among the 4 different molecular cohorts. Statistical analyses were performed by the University of Colorado Biostatistics and Bioinformatics Core using SAS/BASE and SAS/STAT software, version 9.2 of the SAS System for Windows (SAS Institute Inc., Cary, NC).
A total of 209 consecutive patients with stage IV nonsquamous NSCLC were evaluated in this study. One patient with an EGFR mutation and 1 patient with a KRAS mutation were excluded due to lack of records documenting sites of disease at the time of diagnosis of metastatic disease. One patient was excluded because of the presence of both an EGFR (HV773_774LM) and a KRAS mutation (G12V). One patient was excluded because of the presence of an ALK gene rearrangement and an EGFR mutation (S768I).
An ALK gene rearrangement was identified in 20% of patients, an EGFR mutation in 19% of patients, a KRAS mutation in 23% of patients, and 38% of patients had no abnormality in the 3 genes (Table 1). The vast majority of patients in this study displayed adenocarcinoma histology, with only a few patients demonstrating large cell lung cancer with or without neuroendocrine features or NSCLC not otherwise specified. The majority of patients analyzed here were categorized as stage IV at the time of diagnosis, but approximately 20% of patients had recurrent cancer.
Baseline characteristics of evaluable patients are shown in Table 1. The proportion of heavy smokers significantly differed across the molecular cohorts, with the triple negative and KRAS mutation groups showing the highest proportions and the EGFR- and ALK-positive groups showing the lowest proportions (P < .0001). Patients with EGFR and KRAS gene mutations were more often female than those of the triple negative cohort, whereas patients with ALK gene rearrangements exhibited a similar sex distribution; overall, the distribution of molecular cohorts differed significantly between males and females (P = .03). Patients positive for the EGFR and KRAS mutations were diagnosed with metastatic disease at a similar age as those of the triple negative cohort, and ALK-positive patients (P < .0001) tended to be younger, which is consistent with previous reports.17
The majority of patients underwent testing for all 3 molecular markers (80%). A smaller number underwent testing for only 2 of the 3 biomarkers (18%) (Table 2). Only 3 patients, all with an ALK gene rearrangement, had only 1 test performed. Biopsy material from the primary tumor was most often used for molecular testing, followed by metastatic sites, then lymph nodes (Table 3). Only 11 patients had more than 1 site biopsied. Documentation of imaging modalities across the molecular cohorts did not differ significantly (Table 4). Positron emission tomography/computed tomography and brain magnetic resonance imaging, for example, were performed and documented in the majority of cases across all molecular cohorts.
Only 18 of 209 patients (9%) exhibited evidence of pericardial spread at the time of diagnosis, but 8 of 41 (20%) patients in the ALK-positive cohort had pericardial spread (Fig. 1A). Patients with an ALK gene rearrangement were significantly more likely to have metastatic spread to the pericardium than patients from the triple negative cohort (odds ratio [OR] = 4.61; 95% confidence interval [CI] = 1.30, 16.37; P = .02). EGFR (OR = 1.03; 95% CI = 0.18, 5.87; P = 1.0) and KRAS (OR = 1.69; 95% CI = 0.40, 7.09; P = .48) mutation positive patients were similar to the triple negative patients with respect to metastatic spread to the pericardium.
A large proportion of patients, 90 of 209 (43%), exhibited pleural disease (Fig. 1C). ALK-positive patients were also more likely than patients in the triple negative cohort to have spread to the pleura (OR = 4.80; 95% CI = 2.10, 10.97; P < .001). There was no difference in the EGFR (OR = 1.22; 95% CI = 0.56, 2.68; P = .69) and KRAS cohorts (OR = 0.78; 95% CI = 0.36, 1.66; P = .57) with respect to pleural spread compared to patients from the triple negative cohort.
As expected, the majority of patients, 144 of 209 (69%), displayed spread to intrathoracic lymph nodes, whereas only 37 of 209 (18%) had involvement of extrathoracic lymph nodes (Fig. 2). Patients with ALK gene rearrangements demonstrated a numerically higher incidence of spread to intrathoracic lymph nodes (OR = 2.34; 95% CI = 0.92, 5.98; P = .09) and extrathoracic lymph nodes (OR = 2.25; 95% CI = 0.96, 5.29; P = .07) compared to patients from the triple negative cohort, but this did not reach statistical significance.
Patients with ALK gene rearrangements (OR = 5.50; 95% CI = 1.76, 17.18; P = .003) and patients with EGFR mutations (OR = 5.17; 95% CI = 1.63, 16.43; P = .006) were predisposed to liver metastasis compared to those of the triple negative cohort (Fig. 3). Of the 29 patients in this study with liver metastasis, 21 (72%) patients were positive for ALK or EGFR. Given the high proportion of patients with “actionable” genetic changes (ie, targeted by a highly active drug) found to have liver metastases, we determined the likelihood ratios for an EGFR mutation or ALK gene rearrangement in the presence of liver, pleural, or pericardial metastases and the predictive values of these sites of disease (Fig. 4).
No molecular cohort demonstrated a significant predilection to pulmonary nodules or adrenal, bone, or brain metastases compared with patients from the triple negative cohort (Fig. 5). Brain imaging was not documented in 27% of patients, although the lack of imaging in each cohort was similar (Table 4). To determine whether this factor influenced our findings, we reanalyzed the data using only patients who had brain imaging and found similar results (data not shown). KRAS mutation was not associated with increased risk of any metastatic site compared with patients from the triple negative cohort.
The majority of patients in this study had metastatic disease at diagnosis. However, 21% of patients included in this analysis had metastatic disease at recurrence (ie, patients initially with a stage I to III NSCLC that later recurred with sites of metastatic disease) (Table 1). To determine whether the inclusion of “recurrent” metastatic patients could have influenced the results presented here, we reanalyzed the data with only patients who had metastatic disease at first diagnosis. A similar trend for increased prevalence of pleural and pericardial effusion in ALK-positive patients and for liver metastases in patients with ALK-positive or EGFR mutant NSCLC was observed (Fig. 6). For all other sites of disease, the relative prevalence among the molecular cohorts at each organ site was similar (data not shown).
We also evaluated the number of metastatic disease sites for each cohort, limiting our analysis to the disease sites described above (Fig. 7). The EGFR (mean = 2.5 sites, P = .79) and KRAS mutation cohorts (mean = 2.4 sites, P = .72) exhibited a similar number of metastatic sites to that of the triple negative cohort (mean = 2.5 sites), whereas the ALK-positive cohort exhibited more sites of metastatic disease than that of the triple negative cohort (mean = 3.6 sites, P < .0001).
Less common sites of metastatic spread were also recorded for each patient in this study. No formal analyses were performed, given the small number of patients displaying each of these rare sites. Interestingly, 3 patients were found to have retinal metastases, 2 in the ALK-positive cohort and 1 in the EGFR mutation cohort, whereas no patients in any other molecular cohort had this finding. Retinal metastases have been reported previously in lung cancer, but before molecular testing was routinely performed.18
Here, we report analysis of the association between molecular oncogene status and patterns of metastatic spread in treatment-naive patients with NSCLC. We observed a higher incidence of pericardial, pleural, and liver metastasis in ALK-positive patients compared with patients who had no EGFR, KRAS, or ALK oncogene abnormality. Patients with an EGFR mutation also had a higher rate of liver metastases compared with those of the triple negative cohort.
A much higher than expected number of ALK-positive patients were present in this study, partly because of our role as a referral site for the phase 1 study of crizotinib and our initial screening strategy, which enriched the detection of these patients. Undoubtedly, these elevated numbers have facilitated the identification of oncogene-specific patterns of spread for ALK that might otherwise have been missed, given its relative rarity as a molecular subtype of NSCLC. An expected percentage of EGFR and KRAS mutant patients were identified in our study.3, 19
We recognize that the triple negative cohort is a heterogeneous cohort, and a number of patients evaluated in this study underwent evaluation for mutations in other molecular oncogenes such as BRAF, MET, and HER2. Currently, KRAS, EGFR, and ALK are the most established and commonly tested oncogenes in NSCLC, and our testing patterns dictated the categorization used in this study. As we collect more data on other oncogenes in NSCLC, we expect to refine the model described here. The vast majority of patients underwent triple testing for all 3 molecular markers analyzed in this study; however, 20% of patients had only 1 or 2 tests performed. As necessitated by the entry criteria for this study, all of the patients with incomplete testing for all 3 biomarkers demonstrated a positive result for 1 of the biomarkers. We believe that this criterion is justified given the low likelihood of a patient harboring more than 1 positive biomarker result within this subset of analytes.13, 14 Indeed, data from this study show that of the 211 patients with nonsquamous NSCLC who underwent double or triple testing, only 2 patients (∼1%) were found to have more than 1 positive biomarker result, and both were associated with less common forms of mutation in EGFR.
We chose to collect data only at the time of diagnosis, because a longitudinal study might be skewed by survival time. We included patients with recurrent metastatic disease in this study; however, analysis of only patients with metastatic disease at first diagnosis yielded remarkably similar results, which is consistent with our hypothesis that biology is a critical factor in determining patterns of metastatic spread. Furthermore, treatment with either chemotherapy or targeted therapies may induce patterns of spread that do not reflect the natural history of the disease. As an example, patients treated over a significant time period may develop more brain metastases because of poor central nervous system penetration by many of the drugs used to treat NSCLC.20, 21
We used clinical staging (TNM 7th edition15) to classify sites of metastatic disease based on imaging studies, because it would not be feasible to perform pathologic confirmation for each of the metastatic sites documented. The rates of site-specific metastases are difficult to compare with previously published data, given the recent shift of pleural and pericardial disease from a T4 classification in the 6th edition TNM to an M1a classification based on the 7th edition TNM.15, 22 A higher incidence of liver metastases in ALK-positive patients was recently reported.23, 24 Although this is the first study to examine patterns of spread by different molecular oncogenes, a pattern of miliary spread of pulmonary metastases has previously been reported in association with EGFR in-frame deletions of exon 19, consistent with oncogene status driving the specific clinical presentation of the disease.25
In addition to the organ-specific patterns observed here, patients with ALK gene rearrangements were found to have more involved sites of metastasis at the time of diagnosis compared with patients in the triple negative cohort. On the surface, this finding might be considered an indicator of poor prognosis. Currently, the prognosis of ALK-positive patients compared with other molecular cohorts has not been determined definitively.8, 14, 26 Analysis of prognosis in this group is hindered by the fact that most patients who have been identified as ALK-positive have been treated with crizotinib.8 Alternatively, this finding could be explained by a lag in time to diagnosis in this population of patients who are mostly younger and tend to be never or light smokers. Finally, it is possible that a slower proliferating cancer with similar metastatic potential might take longer to present with symptoms and explain this result.
The hypothesis presented here is that biology drives metastasis, but one limitation of this study is that patients were categorized based on a limited set of oncogene abnormalities. In addition, there is clearly significant genetic heterogeneity within all of the molecular cohorts, such as specific type of EGFR mutations (eg, L858R vs exon 19 deletions) or KRAS mutations (eg, G12C vs G12D). However, the number of patients in the study did not allow characterization based on the specific type of EGFR or KRAS mutations. The fact that KRAS behaved similarly to the triple negative cohort with respect to sites of metastatic spread could reflect greater biological heterogeneity in the KRAS cohort compared to the EGFR and ALK cohorts. Alternatively, these results may reflect the role of KRAS in tumor initiation rather than always being a dominant oncogene to which tumors are “addicted.”27EGFR- and ALK-positive tumors may be more uniform based on their stricter dependence on these dominant signaling pathways.6, 28
The clinical results reported here are supported by preclinical models showing that specific genetic pathways mediate the sites of cancer metastases.29-31 Gene expression signatures derived from cell lines that repeatedly metastasize to a given organ site suggest that this process is programmed rather than just a stochastic result.32 Animal studies using tail vein injections of NSCLC cell lines with different oncogene drivers could be used to further study the genetic programs underlying site-specific organ metastatic patterns in lung cancer. The number of patients in each molecular cohort is relatively small, and none of the significant associations identified are absolute, thus this data should not be used as the only criteria for deciding which patients should undergo molecular testing. Although we would advocate testing all patients with NSCLC for relevant molecular markers, the clear association of distinct patterns of metastatic spread with certain highly targetable oncogenes could provide additional clinical prompts for molecular testing among those oncologists who do not conduct widespread screening in their lung cancer patients at the present time.
We thank Dr. Peter Sachs in the Department of Radiology for assisting with radiology images.
This work was supported by the University of Colorado Lung SPORE grant with a Career Development Award to R.C. Doebele, support to M. Varella-Garcia, and support to the Lung SPORE Biostatistics, Informatics and Bioinformatics Core (P50CA058187). The University of Colorado Cytogenetics Core also provided technical support for this work (P30CA046934).