Associate Professor of Medicine Divisions of Endocrinology, Oncology, and Human Cancer Genetics The Ohio State University and Arthur G. James Cancer Center Columbus, Ohio 43210
Address all correspondence and requests for reprints to: Matthew D. Ringel, M.D., Ohio State University, Department of Endocrinology, 455D McCampbell Hall, 1581 Dodd Drive, Columbus, Ohio 43210. E-mail: ringel-1{at}medctr.osu.edu.
Measurements of circulating serum thyroglobulin (Tg) and whole body radioiodine scanning, both of which rely on thyroid-specific gene transcription or function, are highly specific tests for detecting recurrent or residual thyroid cancer and are commonly used in clinical practice. With the development of more accurate Tg immunoassays, there is now a greater reliance on Tg monitoring in clinical paradigms (1). However, there are several clinically important limitations of serum Tg immunoassays: 1) circulating anti-Tg antibodies interfere with Tg measurement in approximately 20% of patients, and 2) TSH stimulation is required for adequate clinical sensitivity in patients at greatest risk of tumor recurrence (1). There has therefore been an interest in developing new assays for thyroid cancer detection that are even more sensitive and are not altered by anti-Tg antibodies. These seemingly straightforward goals have proven remarkably difficult to achieve using immunoassay detection methods. However, the development of molecular RT-PCR-based methods to diagnose viral diseases that identify specific mRNAs, rather than proteins, suggested that a highly sensitive method for detecting thyroid cancer that would be unencumbered by antibody interference could be developed. Now, 7 yr after the initial promising report, it is an appropriate time to make a critical appraisal of the status of these assays for thyroid cancer and to determine whether we are able to detect a reliable "signal" over a more "noisy" background than was initially recognized.
The first Tg mRNA detection assay was reported by Ditkoff et al. (2) who isolated total RNA from 100 individuals, including 87 with thyroid cancer, six with benign thyroid disease following total thyroidectomy, and five normal subjects. Tg mRNA was detected in nine of nine patients with metastatic thyroid cancer, seven of 78 patients thought to be free of disease, and none of 11 patients having surgery for benign disease or normal control subjects. Although detailed clinical information and TSH levels were not reported, the investigators had clearly demonstrated that Tg mRNA could be amplified from peripheral blood and that its presence appeared to correlate with the stage of disease. Tallini et al. (3) subsequently reported data using RT-PCR assays for detecting Tg, thyroid peroxidase (TPO), and RET/PTC1 mRNA from peripheral blood isolated preoperatively, postoperatively, or at both time points, from 44 patients. Of those patients, 24 had thyroid cancer (16 with metastases and 8 free of disease) and 20 had benign nodules. Of the 24 thyroid cancer patients, 56% of the patients with either local or distant metastases had positive Tg mRNA assays, compared with 63% of those thought to be free of disease. Of those thought to be free of disease who had positive Tg mRNA assays, 80% had cervical adenopathy at diagnosis and were felt to be at high risk of tumor recurrence. Of the patients with benign disease, two of 20 patients had a positive Tg mRNA assay and both reverted to negative after surgery.
We reported our initial work in this area in 1998 (4) using a different method for detecting Tg mRNA in peripheral blood. For these assays, total RNA was isolated from whole blood placed directly into an RNA-stabilization solution, a method that resulted in a more sensitive assay. Of the 87 individuals with thyroid cancer evaluated, Tg mRNA was detected in all 14 with cervical or distant metastases during L-T4 therapy, in 65% of patients with thyroid bed uptake, and in 20% of patients with no uptake on scan. These data suggested both a higher sensitivity and lower specificity of the assay than the prior studies. Of concern was that similar to Tallini et al. (3) and distinct from Ditkoff et al. (2), circulating Tg mRNA was detectable in normal subjects and in 20% of athyreotic patients. These results raised two possibilities: 1) a low assay specificity due to the presence of ectopically transcribed Tg or splice-variants of Tg in nonthyroid cells; or 2) a high assay sensitivity, i.e. detection of very early minimal residual or recurrent disease, a hypothesis supported by the lower percentage of positive results in athyreotic patients and the identification of rare cells expressing TSH receptor and Tg in peripheral blood of normal subjects (at that time believed to be evidence of thyroid cells). Tg mRNA and immunoassay results did not correlate, a factor common to all studies that could be related to lower specificity, higher sensitivity, or differences in the parameters being measured. Based on these data, we concluded that Tg mRNA was more sensitive than Tg immunoassay, particularly when TSH levels were low, and that the specificity may be acceptable for clinical practice, but long-term clinical follow-up was needed.
Additional data have been published from many groups using similar approaches to amplify Tg mRNA and other "thyroid" mRNA transcripts from peripheral blood. The results have been remarkably variable, with some groups demonstrating excellent correlation between tumor stage and results (5, 6, 7, 8), whereas others demonstrate no correlation with tumor stage (9, 10, 11). Some have demonstrated that the assay is more useful for papillary than follicular cancer (5), whereas others report optimal screening for tumor recurrence by combining Tg mRNA assays with Tg immunoassays (7). Taken together, nearly all groups have confirmed the finding that circulating Tg mRNA is detectable in peripheral blood of normal subjects and in a subset of athyreotic patients, suggesting that ectopic transcription of Tg or splice variants of Tg can be detected if the sensitivity is high enough, but that there also may be a correlation with the amount of residual or recurrent thyroid tissue present.
What reasons could account for such variable results from excellent research laboratories throughout the world? Several recent articles have begun to address this question. Bojunga et al. (9) reported data using low and high sensitivity qualitative Tg mRNA assays in patients with thyroid cancer by altering the number of PCR amplifications performed and using different primers. Using the lower sensitivity assay, circulating Tg mRNA was detected in 69% of patients with metastatic disease, 46% of patients with thyroid cancer thought to be free of disease, 25% of patients with benign thyroid disease, and 18% of control patients. The higher sensitivity assay was positive in nearly all evaluated subjects, regardless of group. Thus demonstrating that they could enhance sensitivity, but at the expense of specificity, suggesting that qualitative Tg mRNA detection may not be robust when applied to clinical samples. Gupta et al. (12) addressed the issue of Tg splice variants by creating PCR primers designed to carefully avoid amplification of all known Tg and TSH receptor mRNA splice variants. These authors reported detection of thyroid transcripts in 83% of thyroid cancer patients with positive radioiodine scans compared with 5% of patients with negative scans. All normal volunteers were negative. They found excellent correlation between Tg and TSH receptor mRNA results for individual patients, suggesting a high specificity for their assay system. Similarly, Savagner et al. (13) designed Tg primers that amplified known splice variants and others that did not. They determined that the splice variants account for approximately one third of the total amplified Tg mRNA, and that results obtained using only the full-length Tg primers correlated with the volume of thyroid tissue and TSH concentration. Thus, it appears that the avoidance of splice variants and the number of PCR cycles can improve the "signal-to-noise" ratio of thyroid mRNA assays.
Another potentially effective method to circumvent the background noise of ectopic transcription would be to develop a quantitative detection system that would allow for determination of clinical relevant levels of circulating thyroid mRNA. With the advent of real-time quantitative RT-PCR methods, this approach seemed feasible and uncomplicated. However, similar to qualitative RT-PCR, the methodological considerations have turned out to be much greater than initially predicted. A major issue regarding whole blood application of real-time RT-PCR for any clinical or laboratory application is appropriate normalization to control mRNA transcripts. Traditionally, quantitation of target mRNA levels in Northern blot analysis is achieved by normalizing to mRNA levels of "housekeeping genes" that are constitutively expressed and are therefore not regulated by external factors. Glyceraldehyde-3-phosphate-dehydrogenase (GAPDH) and ß-actin are most frequently used; however, tremendous variability in expression of these control transcripts has been reported in human samples when evaluated by RT-PCR (14, 15), including peripheral blood, and a surprisingly large number of pseudogenes exist for GAPDH and ß-actin that can be inadvertently amplified by PCR if tiny amounts of DNA contaminate RNA preparations. For these, and other reasons, nearly all authorities consider normalization to GAPDH or ß-actin alone to be inaccurate for human tissue samples (16, 17). Alternative approaches include normalization to total RNA, to rRNA (18S), or to the original blood volume (16, 17), although these methods may not be appropriate for thyroid cancer because their abundance may result in inability to detect small changes in RNA quality that are important for quantitation of rare transcripts. The use of a "geometric" panel of mRNA markers has recently been suggested (17), but this is untested in thyroid cancer. Thus, it is apparent that normalizing to different control transcripts will alter the assay results (M. Saji, and M. D. Ringel, unpublished observations) and, to date, no standard method has been applied by all laboratories, making it difficult to compare results from one assay to another.
Wingo et al. (18) reported the first quantitative Tg mRNA assay. The interassay variability of this assay was 1722% due primarily to RNA stability, RNA handling, and the reverse transcriptase reaction. The PCR portion of the assay displayed reproducible results over a three log concentration range when quantified to thyroid RNA from a normal thyroid sample. This assay was then used to analyze peripheral blood RNA from 107 patients with thyroid cancer; including 84 during L-T4 therapy, 14 after L-T4 withdrawal, and nine before and after T4 withdrawal (19). Twenty-three patients had circulating anti-Tg antibodies. Because most subjects had detectable levels using this more sensitive method, an arbitrary cut-point that best correlated with clinical status was established (36 cycles), and patients were classified as either positive or negative for detection. Using this cut-point, the Tg mRNA assay was more sensitive than Tg immunoassay but was less specific. Overall, 38% of patients with no evidence of thyroid tissue had detectable Tg mRNA levels compared with 75% of patients with thyroid bed uptake, 89% with local/regional metastasis, and 94% with distant metastasis. However, although there was a statistical correlation between a positive result and the presence of thyroid tissue on scan, there was considerable overlap between the groups, and the mean Tg mRNA levels did not rise with the stage of disease at the time of evaluation. Thus, for individual patients, it was concluded that the absolute value of Tg mRNA did not appear to be diagnostically useful, but the "positive vs. negative" results might be helpful, particularly in patients with anti-Tg antibodies in whom Tg immunoassay is unreliable.
Savagner et al. (13) also developed a quantitative assay for measurement of Tg mRNA in peripheral blood. In this study, the cut-point of a positive or negative assay was determined to be the amount of circulating prostate-specific antigen mRNA as a control transcript, and results were reported per total RNA amount. The results in this study were similar to the prior study (19), in that using a mean value, there was a statistical correlation with the absence or presence of residual or recurrent thyroid tissue, but there was significant overlap between all groups for individual data.
Similar to the experience with qualitative Tg mRNA assays, variable results have also been reported with the quantitative approach by many different groups. Takano et al. (20) performed a study evaluating Tg mRNA from peripheral blood, and similar to prior results, identified this transcript in all patients. They were not able to correlate mean levels with stage of disease; however, the results were normalized in a different manner (GAPDH vs. total RNA), different PCR primers were used, and DNase I treatment was not performed. Takano et al. (20) also report similar disappointing results amplifying TPO as a tumor marker that did not agree with those of Roddiger et al. (10), who reported a better correlation using TPO mRNA amplification than Tg mRNA in patients with thyroid cancer. Eszlinger et al. (21) also did not demonstrate correlation between quantitative Tg mRNA levels and the presence or absence of thyroid tissue. They evaluated several different methods of blood collection and also described important differences in results depending on the types of tubes used for phlebotomy and the time between the sample collection and RNA isolation, further complicating the potential sources of differences in results between groups. These authors used a new set of primers and normalized to ß-actin, factors that distinguish their method from others. To further clarify the importance of recognition of assay differences between groups, Span et al. (11) used the same Tg PCR primers as earlier initial reports (19), but were not able to confirm a relationship between stage of disease and level of Tg mRNA. However, the authors used a different method of RNA isolation and normalized their results to ß-actin, both important differences in assay methods that can alter results (see above).
The issues in clinical applicability of RT-PCR assays for detection of early cancer metastases are not specific to thyroid cancer. Remarkably similar results and controversies can be found in the prostate cancer, breast cancer, colon cancer, and melanoma literature (22). In prostate cancer, for example, sensitivity and specificities for RT-PCR detection of circulating cancer cells have ranged from 2590%, depending on the study, the method used, and the patient population (23). In addition, increased levels of circulating prostate-specific antigen mRNA have been detected after biopsy, suggesting, similar to thyroid cancer, that real "signal" can be detected over background "noise" if the signal is strong enough. To address these issues, an international consortium has been formed to coordinate a comparative study of different molecular prostate cancer detection assays (24). Because of the similarities in the results for different diseases, it is likely that differences between the published results for all of these detection systems may be related to several common factors, including: 1) methodological differences such as the use of different PCR primers that might detect splice variants and the inconsistent use of DNase I treatment to avoid amplification of pseudogenes; 2) differences in assay normalization; 3) the use of a cut-point of clinical detection vs. comparison of results quantitatively without a cut-point; 4) variability inherent in the assay method (i.e. instability of RNA); and 5) interpretive differences.
This brings us to the current study of Elisei et al. (25) published in this issue of JCEM. The authors carefully performed Tg mRNA RT-PCR after DNase I treatment using a new set of PCR primers designed to avoid splice variant-Tg and normalized their results to normal thyroid RNA. The method itself displayed similar calibration characteristics to that of Wingo et al. (18), and an arbitrary cut-point was used to classify results as positive or negative. Similar to prior studies (11, 13, 19, 20, 21), all subjects had quantifiable Tg mRNA levels provided enough PCR cycles were run, further confirming the likelihood that ectopic expression of Tg can be detected in nonthyroid cells. They evaluated this assay in 100 subjects, 80 with thyroid cancer and 20 with no known thyroid disease, and classified patients using stimulated Tg levels and radiographically as being free of disease, having thyroid bed uptake with detectable Tg or without detectable Tg, or with regional or distant metastases. Using these groups, the sensitivity of the assay overall was 82.3%, whereas the specificity was 24.2%. The mean levels tended to be higher in patients with metastasis. Disappointingly, using this method, the patients with an isolated positive Tg mRNA assay did not appear to have a higher incidence of recurrent cancer during follow-up over 4 yr (n = 9). Thus, it appears that over 4 yr, this Tg mRNA assay did not predict later tumor recurrence, suggesting that they were clinically false-positive results, although the number of patients is small and the follow-up period is short.
So where does that leave us regarding the use of Tg mRNA assays in clinical practice now, 7 yr after the initial report? Can we distinguish the signal from the noise well enough for clinical application in patients with anti-Tg antibodies or have we determined that accurate detection is not possible with currently available technologies? Several facts seem clear, based on available data: 1) "thyroid-specific" mRNAs are not nearly as specific as was thought 7 yr ago; 2) splice variants (and probably pseudogenes) exist for Tg that must be accounted for by careful primer design and removal of contaminating DNA; 3) sample handling accounts for significant variability in results; 4) highly sensitive Tg mRNA assays detect transcripts in patients with no known thyroid tissue, which may be false-positives based on follow-up data in a small number of patients; 5) quantitation of the assay is remarkably difficult due to issues with normalization and the rarity of the transcript in peripheral blood; 6) in some studies, Tg mRNA results correlate with results for other "thyroid-specific" transcripts; and 7) patients with larger amounts of thyroid tissue, or perhaps high TSH levels, are more likely to have positive results than those with smaller amounts in many studies. Taken together, these facts suggest that there tends to be a greater signal in patients with obvious residual or recurrent thyroid cancer, but that the background noise is high and variable and may be difficult to separate from the bona fide signal using available methods. Clearly, as has been suggested (14), standards for methods for molecular detection assays need to be established to allow for quality control and interpretation of new data, as has been recently initiated by the prostate cancer research community.
New advances in the ability to store RNA samples, to develop "geometric" control groups of transcripts for normalizing results, the use of additional thyroid and nonthyroid-specific transcripts (26), and elimination of amplification of Tg splice variants could all enhance the method over time. In addition, based on data from molecular detection of other malignancies, enrichment of blood samples for tumor cells using separation techniques (22) and more global evaluation of cDNA (27) and protein (28) expression patterns in peripheral blood may result in earlier detection of metastasis or enhance preoperative diagnosis for many cancers, thyroid included.
Thus, based on the published data, it appears that Tg mRNA assays, at this time, are not clinically better than new Tg immunoassays, and use in the general thyroid cancer population as an adjunctive test is not yet warranted. However, the apparent correlation with extent of tissue in many studies, emerging data using more specific primers for bona fide Tg mRNA, and the development of new methods to enrich and analyze peripheral blood samples are tantalizing enough to continue work in this area for patients with circulating anti-Tg antibodies in whom Tg immunoassay is inaccurate and in the area of preoperative diagnosis, with the caveat that it will probably take more time than predicted to determine whether the "signal to noise ratio" will be acceptable for clinical practice.
Footnotes
Abbreviations: GAPDH, Glyceraldehyde-3-phosphate-dehydrogenase; Tg, thyroglobulin; TPO, thyroid peroxidase.
Received November 11, 2003.
Accepted November 11, 2003.
References