While it is frequently reported in the literature that a TMA has been previously validated, details pertaining to this validation are often absent or are not well described. In many instances, the observed association of TMA biomarker expression with a single clinicopathologic feature of the cancer of interest is provided as evidence that the TMA has been validated and TMA technology can therefore be used to study other biomarkers in the same cancer or the same biomarker in other cancers. As such, prior to preparing a TMA, the number of TMA cores necessary to accurately represent whole tissue section biomarker expression should be established on a biomarker-specific and cancer-specific basis. We established the use of a simulated TMA as a cost-effective tool for this purpose.
The first objective of this study was to evaluate agreement between whole tissue section quantification and quantification using a simulated TMA to determine the number of TMA cores necessary to adequately represent whole tissue section expression for 6 biomarkers that have been shown to be associated with ccRCC patient outcome [16
]. The biomarkers studied displayed different frequencies and patterns of expression; consequently, the number of TMA cores necessary to adequately represent whole tissue sections was not the same for all biomarkers. In particular, using a simulated TMA, we determined that 2-to-3 cores appeared adequate for Ki-67, IMP3, B7-H3, and CAIX, but even as many as 10 cores resulted in poor agreement for B7-H1 and survivin with their whole tissue section counterparts. This demonstrates that the number of TMA cores necessary to represent whole tissue specimens is biomarker-specific and thus a single TMA is not always appropriate for a set of markers.
The second objective of this study was to evaluate the ability of a TMA to discover associations between biomarker expression and patient outcome. For most of the biomarkers studied, 2-to-4 cores appeared adequate to identify the associations between expression and outcome observed with whole tissue sections. However, while whole tissue section B7-H1 was significantly associated with RCC-specific death, no significant associations with patient outcome were detected using as many as 10 cores for this focal and rarely expressed biomarker, demonstrating that a TMA may not be an appropriate tool for some biomarkers. The suitability of TMA analysis of B7-H1 for other studies and tumor types need to be further evaluated. It is important to identify biomarkers that are not well suited for TMA analysis in order to eliminate false-negative conclusions obtained from TMA analyses. Likewise, the sample size necessary to detect associations is not equivalent for all biomarkers as a result of the varying expression patterns across biomarkers.
TMAs are commonly used to test various biomarkers for their ability to predict disease outcome or response to therapy since TMAs allow investigators to study many tissue specimens using a uniform experimental process, while simultaneously preserving limited tissue resources [24
]. As the results of the current study demonstrate, the general guideline that 3-to-4 cores are sufficient to adequately represent whole tissue specimens does not hold true for all biomarkers, and presumably for all malignancies. Our results corroborate the conclusions of Fons et al. [5
] who suggested that the concordance between TMA and whole tissue section quantification depends on the expression pattern of the biomarker. Biomarkers with focal and rare expression may not be well suited for TMA analysis, as Linderoth et al. [25
] demonstrated for BCL6 expression in diffuse large B-cell lymphoma, whereas biomarkers with diffuse expression may be adequately represented by a limited number of TMA cores. Although it may be concluded that TMAs are being successfully used since they are uncovering significant associations between biomarker expression and outcome, we highlight the likelihood of obtaining false-negative findings when using a single TMA to quantify multiple biomarkers with heterogeneous patterns of expression.
One limitation of our study is that we did not fully evaluate potential sources of variability in biomarker expression due to non-informative cores and core location. However, as shown in Table , we did observe that the outer ring of cores on the simulated TMA were more likely to contain primarily normal, stromal, artifact, necrotic, degenerative tissue or no tissue at all in comparison to the inner ring of cores. The non-informative cores were retained throughout the analyses, reasoning that they represented missing cores on a real TMA (i.e., cores that are lost during processing or that contained minimal tumor). For example, Fons et al. [5
] reported that 10%, 8%, and 9% of TMA cores were not assessable for their study of oestrogen and progesterone receptor, p53, and epithelial membrane antigen, respectively. Similarly, Gillett et al. [26
] reported that 12% and 13% of TMA cores had floated off during the immunohistochemical technique and 10% and 12% of the stained sections did not contain invasive tumor when evaluating estrogen and progesterone receptors, respectively. Thus, the number of non-informative cores observed with our simulated TMA is similar to what others have observed using real TMAs. Additionally, using the simulated TMA, we were unable to evaluate variability in biomarker expression associated with core size or different slices of the TMA block. With respect to core size, Lesnikova et al. [24
] evaluated cervical neoplasia specimens and concluded that 1 mm tissue cores were more appropriate than 0.6 or 2 mm cores and provide a suitable compromise of being large enough to be representative yet small enough to be high throughput. With respect to evaluating different slices of the TMA block, Hager et al. [4
] evaluated RCC specimens and concluded that the percentage of lost or non-informative cores tripled from the first slice to the last slice of the TMA block primarily due to the presence of limited (> 25%) tumor tissue, core folding, and necrotic tissue. Lastly, the use of a TMA template mask placed on previously stained whole tissue sections did not allow us to address additional variability that can be introduced by day-to-day fluctuations in staining procedures or TMA core edge effects.
Number of non-informative cores observed (out of 100 patients) on the simulated TMA
While non-informative cores, core location, core size, repeated sectioning of the TMA block, and staining fluctuations certainly contribute to variability in biomarker expression, variability may also arise by the use of a single TMA to study multiple biomarkers without an understanding of their varied expression patterns in whole tissue sections. This often underestimated variability may contribute to the inability of promising biomarkers to achieve clinical utility. As McShane et al. [27
] remarked, while numerous biomarkers have been studied, the number of biomarkers that have attained clinical utility is "pitifully small". Initial reports of a biomarker's promising ability to predict a clinical outcome are rarely substantiated by subsequent studies of the same biomarker or related biomarkers. McShane and colleagues cite a number of reasons for these inconsistent findings including insufficient sample size, inappropriate statistical methods, and use of biomarker assays that are not standardized and reproducible. Herein, we demonstrate that studies that employ TMA technology without first evaluating agreement in biomarker expression with whole tissue sections may further contribute to washout of biomarker utility in the clinical setting.
In summary, TMAs are useful for studying a given biomarker provided that the number of TMA cores necessary to accurately represent whole tissue section biomarker expression is established on a biomarker-specific and tumor-specific basis as there is not a one-size-fits-all TMA. We recommend that guidelines for evaluating the appropriateness of a TMA for a given biomarker and tumor type should be developed, similar to the REMARK guidelines established for prognostic studies [27
]. We further recommend that any claims suggesting that a TMA can be used to represent biomarker expression in whole tissue sections be supported by published data that disclose rigorous evidence of these validation steps.