Gene Expression Ratio Test Distinguishes Normal Lung from Lung Tumors in Solid Tissue and FNA Biopsies

Abstract

Lung cancer is the leading cause of cancer-related deaths worldwide. Prognosis and survival are dependent on cell type, early detection, and surgical treatment. Hence, optimal screening strategies and new therapies are urgently required. Although surveillance with low-dose computed tomography can reduce lung cancer mortality by 20%, the number of false-positive detections is significant. Tissue diagnosis aids in the identification of benign nodules, reducing the number of false positive detections. To determine whether molecular testing of fine-needle aspirations (FNAs) can reduce false-positive detections, we developed a gene expression–based test that distinguishes normal from cancerous lung tissues. The test first was applied to published microarray data, showing overall sensitivity and specificity values of 95% (95% CI, 90%–98%) and 100% (95% CI, 40%–100%), respectively. Subsequently, it was validated on 30 solid and ex vivo FNA lung cancer tumor samples and matched normal lung specimens using real-time PCR. The validation test was 93% (95% CI, 78%–99%) sensitive and 100% (95% CI, 88%–100%) specific for the detection of tumor versus normal lung on solid samples, whereas FNA specimens yielded a sensitivity of 91% (95% CI, 72%–99%) and a specificity of 94% (95% CI, 70%–100%). This study supports the hypothesis that the gene-ratio approach reliably distinguishes normal lung from cancerous tissues in FNA samples and can be optimized to diagnose benign nodules.

Lung cancer is the leading cause of cancer-related deaths worldwide for both men and women.¹ Approximately 85% of lung cancers have histologic features characterized as non-small cell lung cancer (NSCLC).² The histologic subtypes of NSCLC include squamous cell carcinoma, adenocarcinoma, adenosquamous carcinoma, and large cell carcinoma.³ Management of NSCLC usually is determined by stage of disease at diagnosis. Early stage cancers (I to IIIA) are treated surgically (with or without neoadjuvant or adjuvant therapy), and more advanced stage cancers (IIIB to IV) are treated with chemotherapy with or without radiation therapy.^4,5 The high mortality rate of lung cancer is attributed to the following: a diagnosis occurring in advanced stages as a result of the late onset of symptoms, the relative lack of screening strategies, and ineffective treatment approaches for advanced cancer. Currently, a definitive diagnosis of cancer in patients with documented pulmonary nodules can be established by surgical resection, percutaneous image-guided needle biopsy, or bronchoscopy with bronchioalveolar lavage.^6–8

Percutaneous image-guided, fine-needle aspiration (FNA) biopsy is a minimally invasive technique that allows sampling of tumors for diagnosis. Cells are aspirated directly from the solid lesion and analyzed by a cytopathologist.⁹ The accuracy of this approach, however, diminishes with the size of the suspected nodule(s) as a consequence of technical challenges associated with accessing small lesions and the difficulty of obtaining sufficient amounts of cytologically diagnostic material.¹⁰ Until recently, lung cancer subtyping in FNA specimens was based entirely on morphology.¹¹ The difficult interpretation of poorly differentiated tumors and the presence of inadequately preserved samples strongly affected the cytopathologist's ability to render an accurate diagnosis.¹¹ In addition, the relative lack of tissue architecture and cellular dispersion in cytology samples further limited correct tumor subtyping. In more recent years, these barriers to accurate diagnosis have been removed by increasing reliance on improved immunocytochemistry and novel adjunctive molecular biology techniques for more precise tumor diagnosis.^12,13

We previously developed a bioinformatic algorithm using gene expression ratio–based tests to translate comprehensive expression profiling data into simple clinical tests based on the expression levels of a relatively small number of genes.^14–17 In addition, we have used this algorithm to generate a gene ratio–based test for the differential diagnosis of solitary lung nodules using 23 genes, and we have tested it in 29 ex vivo FNA biopsies from normal lung and suspected tumor nodules.¹⁸ This algorithm was 87% and 100% accurate in detecting cancer in FNAs from normal tissue and tumor, respectively.

In this study, we used published microarray data¹⁹ to develop an expression-based test capable of distinguishing normal lung from lung cancer as a pilot study to determine the feasibility of using molecular gene-based ratio tests for the diagnosis of lung cancer. We compared the diagnostic accuracy of applying the test to FNA biopsy specimens and to corresponding solid tumor samples from the resected specimens. Our aim was to determine whether a molecular test that was capable of distinguishing between normal lung and cancerous tissues using FNA biopsy could be added to cytologic tests to improve the detection and diagnosis of lung cancer.

Materials and Methods

Tumor and FNA Samples

Studies using human tissues were approved by and conducted in accordance with the policies of the institutional review board at Brigham and Women's Hospital and the Dana Farber Cancer Institute (Boston, MA). Solid tumors and ex vivo FNAs were collected prospectively from 282 consecutive lung cancer resection specimens at the Brigham and Women's Hospital using standard cytopathology techniques as previously described.²⁰ Briefly, immediately after the tumors were removed, the specimens were transferred to the frozen section room where FNAs were obtained by a technician who had been trained by a cytopathologist to reproduce the in vivo technique using a 22-gauge needle. Solid areas were visualized, palpated, and FNAs were obtained. Chunks of matching tumor and normal lung samples from the same patients were acquired and fresh frozen for comparative analysis. A subset of 30 NSCLC samples and 10 tumor samples of different histologies included in the differential diagnosis of solitary pulmonary nodules was selected randomly. In situ adenocarcinomas were not included in this analysis. The histologic distribution of the samples and the number of solid samples and FNA biopsies available for each case are displayed in Table 1. The frozen tissues were not microdissected to enrich their tumor content because it previously has been shown that the repeatability of the gene ratio is not affected by or related to low levels of tumor content in tissue specimens.¹⁴ The frozen samples and the ex vivo FNA biopsies were placed in RNA extraction buffer (TRIzol reagent; Life Technologies Corporation, Grand Island, NY) and RNA was extracted according to the manufacturer's instructions. The RNA was quantified using an ND-1000 spectrophotometer (NanoDrop, Fisher Thermo, Wilmington, DE). RNA samples that showed a 260:280 ratio less than 1.4 were considered inadequate for the purposes of the study.

Table 1.

Samples Included in the Study

Total number of analyzed cases	Histology	Number of analyzed solid tumor samples	Number of analyzed solid normal samples	Number of analyzed tumor FNAs	Number of analyzed normal FNAs
16	ADCA	16	16	14	15
14	SQCA	14	14	11	14
2	Carcinoid	2	2	2	2
1	Metastatic breast cancer	1	1	1	1
1	Metastatic colon cancer	1	1	1	1
1	Metastatic renal cell carcinoma	1	1	1	1
1	Metastatic stomach carcinoma	1	1	1	1
1	Metastatic transitional cell carcinoma	1	1	1	1
1	NSC, mixed histology	1	1	1	1
1	NSC, poorly differentiated	1	1	1	1
40	Total	40	40	35	39

ADCA, adenocarcinoma; NSC, non-small cell; SQCA, squamous cell carcinoma.

Data Analysis and Gene Expression Ratio Test

To generate the gene expression ratio test for normal versus tumor specimens, we analyzed a published microarray data set¹⁹ to identify genes that were expressed differentially between tumor and normal samples. The histologic distribution and the number of samples in this data set are shown in Table 2.¹⁹ The arrays were searched to identify the probes with a highly significant difference in expression (P ≤ 0.01) and at least twofold absolute expression difference between the two-sample sets. Eleven genes were selected for further analysis and several gene ratio tests were calculated. The diagnostic accuracy of each gene expression ratio test was determined as previously described.¹⁵ Finally, the relative expression levels of four genes (AGER, GPR56, SSP1, and DDR1) were used to calculate the geometric mean of three individual gene-pair expression ratios (AGER/GPR56, AGER/SSP1, and AGER/DDR1). Samples were assigned a diagnosis of normal lung when the combined score was greater than 1 and a diagnosis of lung cancer when the combined score was less than 1.

Table 2.

Samples Included in the Published Microarray¹⁹

Histology	Number of analyzed solid tumor samples
ADCA	139
SQCA	21
Carcinoid	20
SCLC	6
Normal lung	17

ADCA, adenocarcinoma; SCLC, small cell lung cancer; SQCA, squamous cell carcinoma.

Real-Time Quantitative RT-PCR

One microgram or 100 ng of total RNA was reverse-transcribed using a TaqMan reverse-transcription reagent (Life Technologies Corporation). RT-PCR was performed using a SYBR Green fluorometry-based detection system (Life Technologies Corporation), as previously described.¹⁶ Primer sequences were synthesized by Invitrogen (Life Technologies Corporation), and used for RT-PCR as previously published.^15,21 The sequences of the primers are listed in Table 3. The relative expression levels of the four selected genes were determined by RT-PCR, and the geometric mean of the three individual gene-pair expression ratios was calculated. All samples were analyzed at least in duplicate. Equivocal tests (score = 1.0) were repeated starting from the reverse-transcriptase reaction. Samples receiving an equivocal score (1.0) in two different experiments were excluded from the analysis because it was impossible to classify the sample as either tumor or normal lung using the ratio test.

Table 3.

List of the Primers for RT-PCR

Primer name	Sequence
AGERq33F	5′-CAGGACCAGGGAACCTACAG-3′
AGERq33R	5′-CTGCTCATTGGGGTCATCTT-3′
GPR56q56F	5′-GCGTGACTTCTTGCTGAGTG-3′
GPR56q56F	5′-CTCAAAAGGGACCTCCAGCT-3′
SPP1q34F	5′-CAGTTGCAGCCTTCTCAGC-3′
SPP1q34R	5′-GGCTAAACCCTGACCCATCT-3′
DDR1q25F	5′-TTGCAGGTGGATCTACAACG-3′
DDR1q25R	5′-GAGTGGTGCTGAAGGACCTT-3′

Data and Statistical Analysis

The final diagnosis for a patient using the gene ratio expression test was determined by majority rule pooling across all replicates from an analyzed sample of solid tissue or FNA.¹⁴ In cases in which an equal number of samples with different assignments were found, a final diagnosis could not be determined, the test result was considered equivocal, and the case was excluded from the analysis of evaluable samples. The 95% CIs for rates of sensitivity and specificity were based on the exact binomial distribution.

Results

From a published microarray data set (Table 2¹⁹), a training set of 26 tumors (13 adenocarcinomas and 13 squamous cell carcinomas) and 13 normal lung samples was selected arbitrarily to develop a four-gene, three-ratio, lung cancer diagnostic test (AGER/GPR56, AGER/SSP1C, and AGER/DDR1B). The test was examined further in the remaining samples of the published microarray data¹⁹ to determine the accuracy of the proposed ratio combination in a test cohort. The overall sensitivity was 95% (95% CI, 90%–98%) and the specificity was 100% (95% CI, 40%–100%). The ability of the test to classify SCLC and metastatic tumors then was investigated in the remaining 26 microarray samples. Five SCLC and 19 (92%) pulmonary carcinoid tumor samples were classified correctly as lung tumor.

For validation by RT-PCR, a test set of 30 matched tumor and normal snap-frozen samples was assembled by randomly selecting the samples from our tumor collection. The overall sensitivity of the test using RNA extracted from solid samples was 93% (95% CI, 78%–99%) and the specificity was 100% (95% CI, 88%–100%). The same test was applied to RNAs collected from the matched FNA biopsies, showing a sensitivity of 91% (95% CI, 72%–99%) and a specificity of 94% (95% CI, 70%–100%).

To determine whether the test also was accurate in detecting samples included in the differential diagnosis of a solitary pulmonary nodule, 10 matched normal and tumor metastatic solid tissues and FNA biopsies from the same specimens also were analyzed. The test performed in metastatic solid tissues by RT-PCR was 60% (95% CI, 26%–88%) sensitive and 90% (95% CI, 55%–100%) specific. When the test was performed using the RNA from the FNA biopsies, the sensitivity was 80% (95% CI, 44%–97%) and the specificity was 78% (95% CI, 40%–97%). A comprehensive representation of the results is provided in Table 4.

Table 4.

Schematic Representation of the Results by Microarray and RT-PCR

Samples	Tumor	Normal	Sensitivity	Specificity
Microarray data set
Analyzed test set	133	4
Correctly classified	127	4	95% (95% CI, 90%–98%)	100% (95% CI, 40%–100%)
RT-PCR data set
Lung cancer solid tissues
Analyzed samples	30	30
Correctly classified samples	28	30	93% (95% CI, 78%–99%)	100% (95% CI, 88%–100%)
Lung cancer FNA biopsies
Analyzed samples	23	16
Correctly classified	21	15	91% (95% CI, 72%–99%)	94% (95% CI, 70%–100%)
Metastatic solid tissues
Analyzed samples	10	10
Correctly classified samples	6	9	60% (95% CI, 26%–88%)	90% (95% CI, 55%–100%)
Metastatic FNA biopsies
Analyzed samples	10	9
Correctly classified samples	8	7	80% (95% CI, 44%–97%)	78% (95% CI, 40%–97%)

Discussion

We have previously shown that the gene-ratio algorithm can be applied to diagnostic and prognostic tests using samples collected by FNA biopsy.^18,20 This algorithm identifies genes that are expressed differentially in two clinically distinct conditions and calculates ratios for gene pairs that can predict the condition associated with a test sample. We have reported that combinations of a small number of carefully chosen and validated gene expression ratios can be used to develop diagnostic and prognostic tests for many types of cancer.^{14,15,17,22–24} The benefit of this approach is that it can be used as a stand-alone test with minimal amounts of nucleic acid material. In addition, we previously have shown that the FNA biopsy technique captures sufficient RNA for RT-PCR analysis in the majority of cases, but that there is considerable variability in the quality of RNA extracted per sample, which more likely is linked to the technical performance of the procedure rather than to the specific tumor evaluated.²⁰ In this study, we successfully extracted a sufficient quality and quantity of RNA from ex vivo FNA biopsies from both normal lung and tumor to show the feasibility of their use in molecular diagnostic testing for lung cancer. Although the study highlighted some technical limitations of this approach, most FNAs yielded accurate and sufficient diagnoses.

The gene expression ratio–based test developed for this study reliably distinguishes normal lung from pulmonary adenocarcinoma and squamous cell carcinoma. The test is not only very sensitive and specific in solid tissues, but also in FNA-acquired specimens, which showed 91% (95% CI, 72%–99%) sensitivity and 94% (95% CI, 70%–100%) specificity. In addition, the same test can be used to distinguish other primary pulmonary tumors and metastatic cancers in both microarray and RT-PCR data, albeit with lower accuracy, indicating that additional refinements are required to improve the specificity of the test in metastatic tumors.

One of the limitations of gene ratio-based testing of FNA biopsy material is the variable RNA quality inherent in these samples. This limitation ultimately may be overcome by improving the collection method. In addition, some of the biopsies yielded contradictory results when RT-PCR was repeated, which most likely was the result of discrepancies in the sampling area. In such cases, more than one biopsy may be required to obtain an accurate diagnosis, but this does not usually entail significant risk or discomfort for the patient. Finally, we believe that these molecular tests should be performed in combination with cytologic analysis to take advantage of a combination of orthogonal technologies to optimize diagnostic accuracy.

To reduce the mortality of lung cancer, the leading cause of cancer-related deaths worldwide, more effective screening strategies and therapies are needed. The challenge is to find an effective, economical screening strategy for early detection because more than 94 million current and former smokers are at increased risk. The National Lung Screening Trial was conducted to determine whether screening with low-dose computed tomography could reduce mortality from lung cancer.^25,26 Although a 20% decrease in mortality from lung cancer was observed in the low-dose computed tomography group compared with the radiography group, the rate of false-positive results was higher with low-dose computed tomography screening because of the presence of benign intrapulmonary lymph nodes or noncalcified granulomas, which was confirmed noninvasively by the stability of the findings on follow-up computed tomography scans. It is estimated that this screening program, if instituted, will detect approximately 1,000,000 nodules annually, the majority of which will be benign. Therefore, more accurate tests are required to identify malignant nodules. One possible approach is to use percutaneous FNA in combination with molecular techniques to characterize newly discovered pulmonary nodules. We recommend a combination of molecular and cytologic testing to enhance the diagnostic accuracy of FNAs.

Microarray analysis allows the simultaneous measurement of expression levels of thousands of genes in a single experiment to determine molecular variations among different tissues.²⁷ However, there are still many limitations in its direct clinical application. The gene-ratio test is a bioinformatic method that overcomes these limitations and analyzes tissues for differences in gene expression. Therefore, it is a valid method for discriminating malignant from benign nodules.²⁸

Some of the genes identified in this study also have been validated in other disease processes by other studies, including our own.^18,29 AGER (also known as the receptor for advanced glycation end products) encodes a member of the immunoglobulin superfamily of cell surface molecules that plays an important role in the development and progression of vascular disease.³⁰ Recently, disruption of AGER levels has been implicated in the pathogenesis of a variety of pulmonary disorders including cancer.^31,32 GPR56 encodes a member of the G-protein–coupled receptor family; initially, it was reported to suppress tumor growth and metastasis in xenograft models using human melanoma cell lines.³³ Lately, however, its role in other cancer types appears to be more complex.³⁴ DDR1 is a receptor tyrosine kinase thought to play a key role in the communication of cells with their microenvironment. A recent study that aimed to investigate the prognostic impact of DDR1 expression in NSCLC showed that DDR1 up-regulation tends to be observed more frequently in invasive adenocarcinoma compared with DDR1 expression in adenocarcinoma in situ and that overexpression of DDR1 in lung cancer cells results in a significant increase of cell motility and invasiveness (P < 0.001), and the interaction of DDR1 with collagen facilitates the invasiveness of NSCLC cells.³⁵ The selection of the diagnostic genes for the normal lung versus lung cancer test was based solely on bioinformatic rules (eg, stated criteria of selection), not on biology.

The proposed gene ratio diagnostic approach is suitable for analyzing FNA biopsies from normal lung and lung tumors and, ultimately, will be fine-tuned to define benign nodules and the transitional zone around tumors. The diagnostic test for normal lung versus adenocarcinoma and squamous cell carcinoma shows 91% to 93% sensitivity (95% CI, 72%–99%) and 94% to 100% specificity (95% CI, 70%–100%) in solid samples as well as FNA, thereby proving that FNA specimens are comparable with solid tumors and are appropriate for molecular diagnosis. In addition, most diagnostic tests with high sensitivity rarely are sufficiently specific to be useful for broad clinical applications, whereas the gene expression ratio test is able to distinguish normal lung from lung cancer and shows both high sensitivity and high specificity in solid tissues and FNAs. Therefore, this technique may complement and extend current cytopathologic techniques for the evaluation of indeterminate and suspicious lung nodules.