The quantitation of biomarkers on tissue microarrays (TMAs) provides numerous technical challenges for the development of automated systems capable of reading microarrays and translating image information into useable data. Unlike DNA/RNA expression arrays, each spot (or histospot) on a TMA represents a miniature histologic section of tissue that contains complex spatial information that can dramatically affect the quantitative analysis of biomarkers [1–3]. The location and expression level of biomarkers on TMAs is generally determined using target-specific antibodies tagged with fluorescent dyes or enzymes that deposit chromogenic and/or fluorescent substrates. In some cases a given biomarker may only be represented in a subset of cells within a TMA spot (e.g. tumor vs. stromal cells) or may be found differentially expressed in different sub-cellular compartments within cells (e.g. membrane vs. cytoplasm vs. nucleus). In some cases, the cell of interest may only represent a small percentage of the analyzed area, such that an analysis of biomarker intensity within the entire area is meaningless. Accurate quantification, therefore, requires accurate measurement of both biomarker expression and location. Indeed, the importance of preserving the spatial information has proven to be an essential component of automated TMA analysis and is incorporated, in some degree, into most systems currently available for automated TMA analysis.
Quantitative visual analysis of tissue microarrays
The simplest method of biomarker quantitation relies on manual (visual) inspection, usually by a pathologist or someone trained in basic histology, to determine accurate expression levels in histologically complex tissues. In general, expression of biomarkers is scored on an ordinal (0, 1+, 2+, 3+) scale, which in some cases is combined with a scored interpretation of the markers' overall distribution [4,5]. Manual scoring has been successfully used in hundreds of manuscripts on TMAs with a variety of tissue types. Of course, there are limitations to visual inspection of microarrays. At best, manual inspection is only semi-quantitative, reducing biomarker expression – which generally occurs in nature as a continuous, normal distribution – to an ordinal scale. Visual inspection can also be confounded by the inherently subjective nature of human observation, affected by context (e.g. the amount of tumor present, background staining, stromal staining, and even the order in which histospots are observed) [6]. These issues can lead to low inter- and intra-observer variability [7,8]. In some cases, they can preclude the discovery of subtle sub-populations that cannot be identified using manual analysis [9,10].
Current methods of automated tissue microarray analysis
The potential benefits of automated analysis were realized early on as a way to promote TMAs as a method for high-throughput biomarker discovery [11]. Indeed, perhaps the greatest argument against manual scoring, aside from its inherent subjectivity, is the time consuming and tedious burden of repetitively scoring hundreds to thousands of histospots. More recently, systems for the automated reading and image archiving have eliminated some of the most tedious aspects of TMA analysis and permit simultaneous, web based, visual inspection of individual histospots at multiple institutions [12–15]. Several researchers have developed TMA analysis systems out of existing technologies and software [16–19]. These systems dovetail into the recent development of commercially available, dedicated TMA readers and software packages (Table 1), which provide automated quantitation of biomarkers in addition to image acquisition [1]. Unfortunately, the cost of these systems (usually greater than $200 000) is prohibitive for many researchers. Most of these systems utilize chromogenic (i.e. ‘brown-stain’) substrates to quantify biomarkers. Localization of biomarkers is achieved by counterstaining the TMA with traditional histochemical dyes (e.g. hematoxylin, eosin). Automated systems then use morphometric analyses based on size, shape, and color to distinguish tumor from stroma, nuclei from cytoplasm, etc. This technology has achieved higher accuracy than manual scoring for a number of biomarkers including Her-2/neu, ER, and p21 [20–24]; although its use in tissue microarrays per se is in its infancy [25,26]. However, since chromogenic substrates rely on opacity (i.e. optical density) to assess intensity, there is a finite limit to detection (namely 100% opacity) as governed by Beer's law [27]. As optical density is a logarithmic rather than linear representation of the amount of light absorbed by a chromophore, large changes in protein concentration can affect only small changes in optical density. This is particularly true in the range of optical densities used in immunohistochemistry (OD: 0.5–2.0) [28], thus complicating quantitation using chromophores.
Table 1. Commercially available tissue microarray analyzers.
An alternative approach has been the development of systems for immunofluorescence-based antigen detection [10,29]. In theory, such systems should provide quantitation that is more sensitive, easier to localize, and with a broader dynamic range. Unlike chromogenic dyes, immunofluorescence is epifluorescent rather than transmitted, so there is no theoretical limit to detection. Fluorophores come in a wide array of colors, permitting the use of multiple fluorescently tagged antibodies on a single microarray. Rather than using morphometric analysis to localize biomarker signals, fluorescent analysis co-localizes these signals using specific tags to cells and/or sub-cellular compartments of interest. For example, epithelial tumor cells might be distinguished from surrounding stroma using a fluorescently labeled anti-cytokeratin antibody. Nuclei can be distinguished from cytoplasm using diamidino-2-phenylindole (DAPI). Multiple publications have demonstrated the utility of fluorescence-based automated TMA analysis in a variety of tumor types including breast, colon, melanoma, oropharyngeal, prostate, and carcinoid [9,10,30–35].
Of course, there are several disadvantages for fluorescent-based TMA analysis. First, epifluorescent microscope equipment is generally higher cost and more complex than traditional light microscopy. Second, although some biomarkers can be visualized with specific antibodies directly conjugated to fluorophores, most require some form of enzymatic amplification. Similar to chromogenic methods, fluorescent systems generally use peroxidase-conjugated antibodies to catalyze the deposition of tyramide-containing fluorescent substrates [36–38]. This enzymatic process, though capable of dramatic amplification of low level biomarkers (up to 100-fold [39,40]) is, like all enzymatic reactions (e.g. polymerase chain reaction (PCR)), inherently non-linear particularly at high expression levels. Third, formalin fixed tissues exhibit strong background autofluorescence in the wavelengths of the most common fluorophores (e.g. fluresein isothiocyanate (FITC), rhodamine, phycoerythrin). Fortunately, tissue autofluorescence diminishes markedly in the far red to infrared spectrum, and the development of dyes (e.g. Cy5, Cy7, Alexa-647, and Alexa-750) and digital cameras capable of visualizing signals into the near-infrared have minimized this issue. Fluorescence-based systems have the added potential to multiplex TMA reading (i.e. the simultaneous staining and analysis of multiple biomarkers each tagged with a different fluorophore, on a single TMA). The flexibility of fluorescence-based systems to co-localize targets tagged with different fluorophores, permits the quantification and localization of biomarkers in a wide array of fluorescently tagged sub-cellular compartments (e.g. DAPI-stained nuclei, antibody tagged membranes, mitochondria, vessicles, or golgi). This method has proven beneficial in the study of several biomarkers (e.g. beta-catenin in colon carcinoma, AP-2 in melanoma) that exhibit sub-cellular compartmentalization [10]. Furthermore, it provides the potential to develop ‘virtual’ compartments. Thus, a researcher might define a compartment as a particular growth factor receptor (e.g. using a receptor-specific antibody) and then analyze the amount of phosphorylated receptor within that compartment (using a phospho-specific antibody).
The future of automated tissue microarray analysis
The ultimate (and highly demanding) goal of automated TMA analysis is to provide a standard, reproducible, efficient, sensitive, and specific method of biomarker quantitation. Such a system would ultimately report biomarker expression as molecules per-cell or unit-area. However the attainment of this goal will necessarily include advancements in every aspect of tissue microarray production, staining, reading and analysis. New and better methods for tissue preservation (in contrast to formalin fixation) may do a better job of preserving antigenicity and eliminating the need for antigen retrieval [41]. Newer fluorophores and non-enzymatic amplification procedures may provide linear signals across an even broader dynamic range, potentially permitting the detection of single molecules. In particular, quantum dots have shown some promise in this regard [42]; although their successful use on TMAs is, as yet, unproven. The development of fluorophores and cameras capable of imaging even farther into the infrared spectrum (outside of tissue autofluorescence) may facilitate the study of multiple biomarkers on a single array. Also important is the development of quality control standards consisting of specific analyte molecules at set concentrations. These can ultimately be incorporated into microarrays as a series of cell lines or spots containing recombinant protein to provide a dilution curve from which specific biomarker concentrations can be determined [1]. Perhaps the most vital component of automated TMA analysis is the development of databases and biostatistical tools to rigorously analyze the millions of biomarker assays likely to come from high-throughput analysis. Fortunately, dedicated methods for cutpoint analysis [43] and hierarchical clustering of TMAs [44–51], akin to those created for RNA/DNA expression arrays, are currently being developed.
Acknowledgements
Supported by grants from the Breast Cancer Alliance and the NIH (K08 ES11571).
