A Combination of Molecular Markers Accurately Detects Lymph Node Metastasis in Non–Small Cell Lung Cancer Patients

Materials and Methods

Identification of potential markers

An extensive literature and public database survey was conducted to identify any potential markers relevant to lung cancer. Resources for this survey included PubMed, OMIM, UniGene (http://www.ncbi.nlm.nih.gov/), GeneCards (http://bioinfo.weizmann.ac.il/cards), and Cancer Genome Anatomy Project (http://cgap.nci.nih.gov). Our survey criteria were somewhat flexible but the goal was to identify genes with moderate to high expression in lung cancer tissue and low expression in normal lymph nodes. In addition, genes reported to be up-regulated in lung cancer and genes with restricted tissue distribution were considered potentially useful. Finally, genes reported to be cancer specific, such as the cancer testis antigens and human telomerase reverse transcriptase, were evaluated.

Tissues and pathologic evaluation

Tissue specimens were obtained from tissue banks at the University of Pittsburgh Medical Center through institutional review board–approved protocols. One half of each lymph node, and a small piece of each primary tumor, were snap frozen in liquid nitrogen and later embedded in OCT for frozen sectioning and RNA isolation. The remaining tissue was used for routine diagnostic pathology evaluation. Twenty 5-μm sections were cut from each frozen tissue piece for RNA isolation. In addition, sections were cut and placed on slides for H&E and immunohistochemical analysis at the beginning, middle (between the 10th and 11th sections for RNA), and end of the sections for RNA isolation. All three H&E slides from each specimen underwent pathologic review to confirm presence of tumor, percentage of tumor, and to identify the presence of any contaminating tissues (such as lung tissue on a lymph node). All of the unstained slides were stored at −20°C. Immunohistochemistry evaluation was done using the AE1/AE3 antibody cocktail (DAKO, Carpinteria, CA), and Vector Elite ABC kit and Vector AEC Chromagen (Vector Laboratories, Burlingame, CA). Immunohistochemistry was used as needed to confirm the H&E histology. Only tissues on which diagnostic pathology and frozen section pathology were concordant were included in the screening tissue sets.

Marker screening

The screening was conducted in two phases. All potential markers entered the primary screening phase and expression was analyzed in six primary NSCLC and 10 benign lymph nodes obtained from patients without cancer (five RNA pools with two lymph node RNAs per pool). Markers that showed good characteristics for lymph node metastasis detection (i.e., high expression in primary tumors and low expression in benign lymph nodes) passed into the secondary screening phase. The secondary screen consisted of expression analysis on 21 primary tumors, 21 histologically positive lymph nodes, and 21 benign lymph nodes from 21 patients without cancer (called the screening tissue set). All tumors and positive lymph node samples were from independent patients.

Marker validation

To externally validate the classification accuracy of markers tested in the secondary screen, an independent, validation set of lymph nodes was analyzed. The validation set consisted of 32 benign lymph nodes from independent patients without cancer and 38 pathology-positive lymph nodes obtained from 25 patients undergoing surgery for NSCLC. Diagnostic pathology and frozen section pathology were concordant on all nodes in the validation set.

Analysis of lymph nodes from pN₀ NSCLC patients

Three markers were selected from the screening and used to identify occult metastasis in 462 nodes from 68 patients with NSCLC (mean 6.8 nodes per patient, median six nodes per patients). Clinical information for these 68 patients is shown in Table 1 . Expression of markers was determined as described below and was compared with expression in 30 benign nodes previously analyzed in the screening and validation phases. Cancer patient lymph nodes were considered positive for occult metastasis if expression of any marker was higher than the maximum expression observed in the benign nodes.

RNA isolation, cDNA synthesis, and quantitative PCR

RNA was isolated using the RNeasy minikit (Qiagen, Valencia, CA) essentially as described by the manufacturer. All RNAs were DNase treated using the DNA-free kit (Ambion, Austin, TX) and reverse transcription was done in 100 μL reaction volumes with random hexamer priming and Superscript II (Invitrogen, Carlsbad, CA) reverse transcriptase (10). All quantitative PCR for the marker screening was done on the ABI Prism 7700 Sequence Detection Instrument. Quantitative PCR on lymph nodes from pN₀ lung cancer patients (and 30 benign lymph nodes) was done on the Stratagene Mx3000P instrument. Relative expression of the marker genes was calculated using the δC_T methods previously described (11) and with β-glucuronidase as the endogenous control gene.

All assays were designed for use with 5′ nuclease hybridization probes although the primary screening was done using SYBR green quantification to save cost. Assays were designed using the ABI Primer Express Version 2.0 software and, where possible, amplicons spanned exon junctions to provide cDNA specificity. In addition, all primer pairs were tested empirically for amplification from 100 ng genomic DNA and primers were redesigned when necessary. With the exception of CK19, cDNA-specific primer sets were successfully identified for all genes in this study. Finally, optimal annealing temperature was determined by testing primers for generation of a single band on gels using cDNA templates and annealing temperatures of 60°C, 62°C, and 64°C. Further details describing our methods for primer design and testing have recently been published (12). PCR efficiency was estimated using SYBR green quantification before use in the primary screen. Further optimization and more precise estimates of efficiency were done with 5′ nuclease probes for all assays used in the secondary screen. Briefly, reactions were done with a probe concentration of 200 nmol/L and a 60-second anneal/extend phase at 60°C for β-GUS, CK19, and SFTPB; 62°C for CK7, PVA, and squamous cell carcinoma antigen (SCCA) 1/2; and 64°C for carcinoembryonic antigen (CEA), lung-specific X protein (LUNX), and TACSTD1. The sequences of primers and probes (purchased from IDT, Coralville, IA) for genes evaluated in the secondary screen are listed in Supplementary Table S1. The primer sequences for markers used in the primary screen will be provided upon request. The details of RNA preparation, reverse transcription, and qRT-PCR have been described in our previous studies (13, 14).

Data analysis

In the primary screen, data from the melt curve was analyzed using the ABI Prism 7700 Dissociation Curve Analysis 1.0 software (Applied Biosystems, Foster City, CA). The first derivative of the melting curve was used to determine the product T_m as well as to establish the presence of the specific product in each sample. In general, samples were analyzed in duplicate PCR reactions and the average C_t value was used in the expression analysis. However, in the secondary screen, triplicate reactions were done for each individual benign node and the lowest C_t value was used to calculate relative expression and thus obtain the highest (most conservative) estimate of background expression for the sample.

Statistical analysis

Generation of prediction rules. Six markers that passed the secondary screen were evaluated individually and in combination with other markers. The characteristics used to evaluate markers were sensitivity, specificity, classification accuracy, and the area under the receiver operating characteristic curve. For individual markers, a cutoff value was determined that maximized the classification accuracy (proportion of lymph nodes correctly classified). In cases where classification accuracy was 100%, the cutoff was set at the midpoint between the highest expressing negative node and the lowest expressing positive node.

Validation of prediction rules. Internal validation of prediction rules was conducted by leave-one-out cross-classification in which each case is classified by a decision rule formulated from the remaining cases. The average sensitivity, specificity, and classification accuracy over all cases is reported. External validation was conducted by applying the most accurate cutoff from the secondary screen (n = 42) to a new set of benign and histologically positive lymph nodes (n = 70).

Results

Primary screen. Our literature and database surveys identified a total of 42 genes for evaluation in the primary screen. All of these genes were analyzed for expression in 6 primary NSCLC and 10 benign lymph nodes. Primary screening was done using SYBR green instead of hybridization probes. Resulting data for the 20 genes with the highest median expression in tumors is shown in Supplementary Fig. S1 and similar data for all genes in the primary screen are available at http://www.mssm.edu/labs/godfrt01/research/charts.htm. Median relative expression (relative to the endogenous control gene) in the primary tumors and in benign nodes was calculated for each gene in the primary screen and is reported in Supplementary Table S2. In addition, we also calculated the ratio of relative expression between the lowest expressing tumor and the highest expressing benign node and between the median expression in tumors and the highest expressing benign node. Some genes, such as LUNX, MAGEA4, and MAGEA10, had no detectable expression in benign nodes and, therefore, ratios could not be calculated. With the exception of LUNX, however, all genes with undetectable expression in benign nodes also have very low median expression in the primary tumors and, as a result, these genes are unlikely to be sensitive markers for detection of occult disease.

When using median expression in the tumors as the numerator, five genes (SFTPB, TACSTD1, CK7, CK19, and CEA) clearly stand out as having tumor/highest benign node ratios >200. When the minimum difference between tumors and benign nodes is calculated as a ratio (lowest tumor/highest benign node), all five of these genes still distinguish tumors from benign nodes very well. Thus, these five genes were selected for analysis in the secondary screen. In addition, whereas PVA and SCCA1/2 were only expressed highly in two (both squamous cell) and three (two squamous, one adenocarcinoma) of six primary tumors, respectively, these genes are primarily squamous cell markers and we would therefore only expect them to show high expression in squamous cell lung tumors. Because ∼30% of NSCLCs are squamous cell tumors, we therefore decided to evaluate these two markers in the secondary screen also. Finally, LUNX is frequently mentioned as a lung cancer marker and although it did not show high expression in all tumors, no background expression was detected at all in benign nodes. For this reason, LUNX was also analyzed in the secondary screen.

Secondary screen. In this stage of the screening procedure, histologic evaluation of the 21 primary tumor specimens revealed 2 large cell tumors, 7 squamous cell carcinomas, and 12 adenocarcinomas with a median tumor percentage of 70% (range: 5-95%). Similarly, in the 21 histologically positive nodes, the median tumor percentage was 70% (range: 2-100%; six tumors had ≤10% tumors). The histologies of the primary tumor for these specimens were as follows: 1 large and small cell mixed cancer, 3 large cell cancers, 6 squamous cell carcinomas, 10 adenocarcinomas, and 1 adenosquamous cell carcinoma (Supplementary Table S3).

The relative expression profiles of eight markers in the primary lung tumors, histologically positive nodes, and benign lymph nodes are shown in Fig. 1 . With the exception of LUNX and SFTPB, the expression levels observed in positive lymph nodes were very similar to those in primary tumors. LUNX expression was consistently lower in positive nodes that in tumors, whereas SFTPB expression was lower in ∼50% of positive nodes. Using an absolute cutoff value equal to that of the highest expression in benign lymph nodes (100% specificity), the sensitivity to detect tumor in the histologically positive nodes achieved with each marker was 100% for CK19 and TACSTD1, 95% for CEA, 61.9% for LUNX, 57.1% for PVA, 61.9% for SFTPB, 76.2% for CK7, and 38.1% for SCCA1/2. Although both CK19 and TACSTD1 provided 100% sensitivity in this data set, the minimum difference in expression between positive nodes and benign nodes was only 1.4-fold for CK19, whereas it was 7.9-fold for TACSTD1. Thus, in a single-gene assay, TACSTD1 is likely to be a more robust discriminator than CK19.

Because NSCLC consists of three major histologic subtypes (adenocarcinoma, squamous cell carcinoma, and large cell carcinoma), we also determined the sensitivity (with 100% specificity) of each marker based on histologic subtype (Table 2 ). The most important finding from this analysis is that whereas SFTPB, CK7, PVA, and SCCA1/2 were not strong markers overall, SFTPB and CK7 may be very useful in adenocarcinomas, whereas PVA and SCCA1/2 are promising in squamous cell carcinomas. To explore this further, the secondary screen data was replotted to show fold change determined as the expression value of each marker in each positive node, divided by the highest expression value of this marker observed among the 21 benign nodes (Fig. 2A ). This analysis showed that in the adenocarcinoma samples, 6 of 11 positive nodes had the greatest fold change with SFTPB, three with TACSTD1, and one each with CEA and CK7. In the squamous cell carcinomas, PVA was highest in five of the six nodes with CEA being highest in the remaining node. In the four large cell specimens, TACSTD1 was highest in two, CEA was highest in one, and PVA was highest in one. Thus, a combination of TACSTD1, PVA, and SFTPB would provide the best discrimination of positive from benign nodes in all but three cases. In these three cases, CEA was the best marker and could be considered in a multiple-marker analysis. However, even in these three cases, the TACSTD1, PVA, and SFTPB marker combination easily distinguished these nodes as positive and the addition of CEA is not really necessary. Therefore, we concluded that the best marker combination marker for NSCLC is probably TACSTD1, PVA, and SFTPB, as shown in Fig. 2B. Using this combination, the median difference between positive nodes and the highest expressing benign lymph node was 606-fold with a minimum of 7.8-fold in this set of 21 positive and 21 benign lymph nodes.

External marker validation. To perform a more robust evaluation of the marker combination identified in the secondary screen, PVA, SFPTB, and TACSTD1 were further evaluated in an independent, validation set of positive and benign lymph nodes. Pathology analysis of 38 positive nodes in this set determined that the median tumor percentage was 55% with a range of 1% to 100% (10 had ≤10% tumor). The histology of the primary tumors in this set was as follows: 24 adenocarcinomas, 10 squamous cell carcinomas, 3 large cell carcinomas, and 1 large cell and small-cell mixed cancer (Supplementary Table S3). Similar to the screening data, 11 of 24 positive nodes from adenocarcinomas had the highest fold change with SFTPB and the remaining 14 were highest for TACSTD1 (Fig. 2C). All 10 nodes from squamous cell carcinomas were highest with PVA and all four large-cell carcinoma nodes were highest for TACSTD1. The median difference between positive nodes and highest benign lymph node was 489.8-fold, with a minimum of 25.7-fold among this set of 38 positive nodes and 32 benign lymph nodes.

Association with histologic subtype. The joint expression levels of TACSTD1, SFTPB, and PVA were plotted in three dimensions (Fig. 3 ). The three-dimensional view clearly shows complete separation for (a) benign nodes versus positive nodes and (b) positive nodes in patients with squamous cell carcinomas and adenocarcinomas. Nodes associated with large-cell lung carcinomas, however, cannot be distinguished by this view. The ability of these three markers to discriminate by histologic origin is primarily due to PVA. Although PVA by itself did poorly in discerning positive lymph nodes (validated classification accuracy of 57%), it was capable of distinguishing positive lymph nodes from squamous versus adenocarcinoma primaries with 98% accuracy. Conversely, TACSTD1 perfectly discriminated between benign and all histologic types of positive nodes but had no ability to discern histologic subtype.

Analysis of lymph nodes from pN₀ NSCLC patients. Analysis of 462 lymph nodes from 68 pN₀ NSCL patients using TACSTD1, PVA, and SFTPB identified 32 (7%) lymph nodes from 20 (29%) patients as potentially positive for occult metastasis (Fig. 4 ). Of these positive nodes, 3 were positive for TACSTD1 (3 patients), 7 were positive for PVA (5 patients), and 26 were positive for SFTPB (16 patients). In four instances, a lymph node was positive for two markers and in four cases a patient had one or more lymph nodes positive for two markers. As in the screening data, nodes positive for PVA were predominantly from patients with squamous cell tumors (5 of 7), whereas those positive for SFTBP were predominantly from patients with adenocarcinoma (21 of 26). TACSTD1 showed no obvious tendency toward any histologic subtype.

Discussion

The presence of occult metastatic tumor spread to lymph nodes is recognized as a significant staging problem in many tumor types. Two characteristics of routine histologic analysis of lymph nodes are responsible for this understaging: limited sampling of each node and insensitivity for detection of isolated tumor cells or small tumor foci. These limitations can be overcome by analyzing sections of each lymph node at multiple levels, to increase sampling, and by adding immunohistochemistry to improve sensitivity of tumor detection (5, 15). Although many studies have reported the prognostic significance of occult lymph node metastases using these methods (5, 15, 16), the additional labor and cost involved in such a detailed analysis of each node has proved to be a major barrier to the routine implementation of this approach. Two exceptions to this are breast cancer and melanoma. In these tumors, the use of sentinel lymph node biopsy identifies typically only one to three nodes, thus allowing for detailed examination.

An alternative approach to improve lymph node staging has been to use modern molecular techniques, such as RT-PCR and mutation or epigenetic alteration detection. There are now hundreds (if not thousands) of such reports in the literature, the vast majority of which have used RT-PCR to detect tumor cells in histologically negative lymph nodes. Almost invariably, these reports indicate that RT-PCR finds positive results in histologically negative nodes and in cases where clinical follow-up is available, many studies show prognostic significance of RT-PCR positivity (17–19). In most studies, however, the frequency of RT-PCR positivity in histologically node-negative patients is much higher than disease recurrence rates and thus this literature is commonly criticized based on poor specificity of the assay. It is now becoming clear that background or ectopic expression of the genes used in these studies is present in normal lymph nodes and that this is responsible for at least some of the false-positive results. Furthermore, despite the large volume of RT-PCR reports, the number of genes used as metastasis markers remains very small with CEA and various cytokeratins (CK19 and CK20 in particular) being by far the most popular in most tumor types except melanoma and prostate. This is probably due to the historical use of these genes as targets for immunohistochemical staining in pathology and is not a result of systematic searches to identify targets specifically for RT-PCR assays. Thus, it is likely that alternative molecular markers could be found and that these markers may provide better specificity than previously obtained. Indeed, we have already reported successful searches for new markers in esophageal adenocarcinoma (13), breast cancer (20), and head and neck cancer (14) and Mitas et al. (21) have done a similar study, also in breast cancer. In lung cancer, however, there is only one study [also by Mitas et al. (23)] looking at six genes in 27 mediastinal lymph nodes of NSCLC patients. For this reason, we designed the current project to perform a large-scale marker survey and identify the best marker or combination of markers for lymph node metastasis detection by RT-PCR in NSCLC.

Of the 42 markers analyzed in the primary screen, eight (CEA, CK7, CK19, LUNX, PVA, SCCA1/2, SFTPB, and TACSTD1) were chosen for more detailed analysis in a secondary screen. Only two of these markers (CK19 and TACSTD1) provided 100% classification accuracy, regardless of primary tumor histology. CK19 is an acidic keratin that is expressed at high levels in many epithelial tissues and their derived tumors. CK19 is one of the keratins targeted by the AE1/AE3 antibody cocktail and, as such, is frequently used by pathologists for immunohistochemical detection or confirmation of lymph node metastases in many cancer types. CK19 has also been used in many RT-PCR-based studies to detect occult lymph node disease in breast cancer (21, 24, 25). Unfortunately, this gene has at least two pseudogenes that make it difficult to develop a cDNA-specific assay. As a result, the utility of this marker may be limited in clinical settings given the potential for false positive results. TACSTD1 is also known as EPCAM, EGP-2, KS1/4, GA-733-2, and MIC-18 and encodes a cell surface antigen that is defined by the monoclonal antibody AUAI. TACSTD1 is also the antigen for the antibody Ber-Ep4 and as such has been used in several studies to detect lymph node micrometastases in a variety of tumor types, including NSCLC (8, 25, 26). The recent studies by Wallace et al. (22) and Mitas et al. (23) also found TACSTD1 (KS1/4 in their article) to be a good RT-PCR-based marker for identification of lymph node metastasis in mediastinal lymph nodes from NSCLC patients. From our study and the Mitas et al. study, TACSTD1 seems to be superior to CK19 for two reasons. First, the observed variation in expression within either positive or benign nodes is much less for TACSTD1 than for CK19. Second, the expression difference between positive and benign nodes is consistently greater for TACSTD1 compared with CK19. In our data, there is at least a 7.9-fold difference between the positive and benign nodes with TACSTD1, with a median fold difference of 90. However, for CK19, the minimum difference is only 1.4-fold with a median of 36. Therefore, although CK19 distinguished positive from negative nodes in our relatively small sample set, CK19 is less likely to stand up as an independent marker in larger data sets. Therefore, of all the markers that underwent the screening process, TACSTD1 showed the most promise as a stand-alone marker for nodal involvement in NSCLC patients.

Overall, the next best marker in our study was CEA, with a sensitivity of 95%. CEA (CEACAM5) is a cell adhesion molecule that is frequently found in adenocarcinomas of endodermally derived digestive system epithelia. Historically, CEA has been the marker of choice for identifying adenocarcinoma of the gastrointestinal tract and was among the first markers to be used for RT-PCR analysis of lymph nodes (27). In lung cancer, D'Cunha et al. (28) analyzed expression of CEA in 53 primary tumors and 232 lymph nodes from 53 patients with stage I NSCLC. In this study, 90.5% of primary tumor samples were positive for CEA mRNA expression as were 25.4% of histologically negative lymph nodes. Fifty-seven percent of the patients had at least one CEA-positive lymph node and would theoretically be upstaged to stage II. Unfortunately, no follow-up was reported in this article so the clinical relevance of upstaging is currently unknown. CEA was also evaluated in the Mitas et al. study but high expression in some control lymph nodes led to relatively poor sensitivity and specificity compared with TACSTD1.

None of the remaining markers in the secondary screen gave very good sensitivity overall (<77%) but on further analysis by histologic subtype of the primary tumor, CK7 and SFTPB seem to be good markers for adenocarcinoma, whereas PVA and SCCA1/2 are good markers for squamous cell carcinoma. CK7 (KRT7) is a type II keratin of simple, nonkeratinizing epithelia and is typically found in tissues, such as the epidermis, bronchus, and mesothelium. In our study, CK7 gave 100% sensitivity (11 of 11) for detecting lymph node metastases from adenocarcinoma tumors; however, based on fold expression difference, TACSTD1 was superior to CK7 in 9 of the 10 cases. SFTPB is one of a family of surfactant molecules secreted by type II alveolar cells. Betz et al. (29) evaluated this gene as a marker for metastatic lung adenocarcinoma using standard gel-based RT-PCR, but discarded it citing expression in nonneoplastic lymphatic tissue. However, in our study, the ability to quantify the expression of mRNA allowed us to distinguish this background expression from expression associated with metastatic cancer cells in 9 of 11 adenocarcinoma cases. Furthermore, in seven of these cases, the fold expression difference was much higher than for either CK7 or TACSTD1. Thus, SFTPB, but not CK7, may be a useful marker in combination with TACSTD1 to improve the detection of adenocarcinoma foci in lymph nodes.

SCCA is a member of the ovalbumin family of serine proteinase inhibitors. The SCCA protein is expressed in neutral and acidic forms, designated as SCCA1 and SCCA2, and is detected in the superficial and intermediate layers of normal squamous epithelium. The expression of SCCA2 in cancer has been associated with an aggressive phenotype and this gene has been used in several studies for detection of squamous cell carcinoma metastases to lymph nodes (30). In our study, SCCA was able to identify metastatic tumor in four of six lymph nodes from patients with squamous cell carcinoma of the lung. PVA (also known as desmoglein 3) is a 130-kDa surface glycoprotein that is the serologic target in the autoimmune skin disease pemphigus vulgaris. Apart from our previous study in head and neck cancer (14), PVA has not been described as a marker for lymph node metastasis detection by RT-PCR. In this lung cancer study, we found that PVA is also an excellent marker for squamous cell lung cancer, providing 100% sensitivity in this histologic subtype. Furthermore, as with SFTPB in adenocarcinoma, PVA is superior to TACSTD1 in five of six cases and could, therefore, be useful in a combination marker assay. The last gene included in our secondary screen, LUNX, was first isolated and reported as a novel lung-specific gene of unknown function by Iwao et al. (31). Using RT-PCR, LUNX expression was detected in all 35 NSCLC tumor samples and in 80% (16 of 20) of histologically positive lymph nodes. In our hands, however, the LUNX gene did poorly as a marker for nodal disease, with a sensitivity of only 61.9% overall and with no obvious improvement in either histologic subtype.

In summary, we have shown that TACSTD1 may be used alone as a molecular marker to detect metastatic lymph node involvement in NSCLC patients and that combining this marker with PVA and SFTPB may provide a more robust assay. Ongoing studies are using these three markers for detection of occult lymph node metastases in pN₀ patients and in the mediastinal nodes of pN₁ patients. Preliminary data presented here indicates that ∼7% of histologically negative lymph nodes from NSCLC patients are positive for one or more of these markers and this identifies ∼30% of patients as potentially having occult lymph node metastasis. These numbers are very similar to the frequencies found in studies using immunohistochemistry but a larger patient series and more follow-up is needed to evaluate the prognostic significance of qRT-PCR positivity.