Clinical relevance of biomarkers
Personalized medicine requires the discovery, validation and application of unambiguous prognostic, predictive and pharmacodynamic biomarkers to guide therapeutic decisions [37, 38]. High-throughput screening methods, using genomic and transcriptomic profiling, have greatly increased our knowledge of the molecular basis of tumorigenesis, cancer progression and therapeutic response [39, 40]. The ambition of individualized treatment regimens can thus be viewed as an achievable goal. Microscopic evaluation of tissue sections taken from a tumour remains the gold standard for determining a cancer diagnosis, including the establishment of the tumour type in most cases. However, there is a great need for better stratification of tumours to optimize patient handling and therapeutic intervention. The heterogeneous and complex nature of cancer can, in part, be untangled by gene sequencing and other emerging molecular biological technologies; however, adding a layer of information regarding protein expression on top of morphology appears to be beneficial for tumour stratification in a clinical setting. IHC prevails as an invaluable method and provides such a tool for the visualization of protein expression patterns in cells from a section of tumour tissues. The role of antibodies is most likely to involve diagnostic, prognostic and predictive biomarker development, and despite the relative success of antibody-based assays for both the oestrogen receptor and human epidermal growth factor receptor 2 (Her2) in breast cancer, the lack of development of clinically implemented assays has not even nearly kept pace with the rate of biomarker discovery .
To date, the majority of the approximately 200–300 antibodies used in clinical pathology are mainly used for diagnostic purposes and to a lesser extent for grading of malignancy and stratification of different tumour types, with the exception of haematological malignancies for which antibody-based classification is required for the subtyping of lymphoma and leukaemia. Only rarely are prognostic markers used in clinical routine and only exceptionally are biomarkers routinely used for the purpose of personalized medicine, by predicting or determining which therapeutic regimen(s) a patient would benefit from [16, 41]. The most prominent exceptions are overexpression of Her2 in breast cancer for determining susceptibility to treatment with trastuzumab [42, 43] and the detection of KIT (CD117) expression for diagnosing gastrointestinal stromal tumours susceptible to imatinib treatment [44, 45]. With only a small repertoire of biomarkers for use in personalized medicine, there is an unmet need to identify sets of markers to subclassify both disease states and patients for accurate assessment of prognosis and for selecting the most favourable regimens for treatment [14, 15, 37].
TMA-based biomarker discovery
Tissue-based clinical research requires specimens from a large set of patients, in the form of frozen or paraffin-embedded tissue blocks. To enable efficient data collection in combination with high-throughput IHC, the TMA technique has proven powerful for paraffin-embedded material, as it allows for the detection of molecular targets in a multitude of samples simultaneously [7, 25, 46]. This technique also conserves the original tissue samples as only a small part of the specimen is needed to generate representative tissue cores for TMA construction. The TMA technology also requires small amounts of reagents and decreases errors because of experimental variability, as all samples are analysed using a single protocol for the entire TMA section. Depending on the design and number of cores sampled from each block of donor tissue, analyses of core specimens can provide a representation of the entire original tissue specimen with >95% accuracy .
Within the Human Protein Atlas project, the database is actively mined for potential biomarkers with the aim of identifying protein expression patterns that indicate whether a particular protein could be used as a biomarker. The focus is to identify and validate cancer biomarker candidates that can fulfil currently unmet clinical needs in pathology and oncology. To address such clinical needs, clinical questions are defined and appropriate patient cohorts determined to enable the collection and assembly of tumour material and clinical data for creating cancer-research TMAs with corresponding comprehensive clinical databases. These specifically designed cancer-research TMAs are used for extended analysis of protein expression patterns to test and validate candidate proteins as diagnostic, prognostic and/or predictive cancer biomarkers (Fig. 4, upper row).
Figure 4. Strategy for tissue-based biomarker discovery: identification of clinical questions and unmet needs for novel biomarkers. Based on the clinical question, patient cohorts are defined and the corresponding tissue microarrays (TMAs) and clinical databases are constructed. The database is mined for potential biomarkers and putative candidates are selected based on, e.g., heterogeneous expression in cancers. A selected candidate antibody is scrutinized with respect to specificity to detect its target protein based on the basic validation performed in the Human Protein Atlas pipeline, extended validation platforms and available external data. The reproducibility and quality of immunohistochemistry staining is tested on screening TMAs before selected antibodies are used with clinical material. The first clinical cohort will fail or qualify the protein as a potential biomarker, and promising results are verified in independent clinical cohorts. Monoclonal antibodies are raised against verified biomarker proteins by using, e.g., information derived from epitope mapping of the msAbs used to detect the protein.
Download figure to PowerPoint
Biomarker candidate selection and antibody verification
The protein expression pattern across 12 individual tumour samples for each cancer type is analysed in the basic screening performed on all approved antibodies in the Human Protein Atlas . This set-up enables the identification of differentially expressed proteins, a prerequisite for finding markers of biological relevance and for determining correlations between protein expression and clinical parameters such as disease-free survival. Proteins exclusively expressed in a particular form of cancer or cell type are equally interesting as such markers could be potential diagnostic markers. Proteins showing differential expression in cancer cells as compared to corresponding normal cells suggest an alteration associated with the malignant phenotype and could provide a marker for distinguishing benign from malignant lesions. All novel proteins with the above-mentioned expression patterns provide a lead for further investigation and require in-depth studies to explore functional aspects. Extended studies in larger well-defined patient cohorts are also needed before a candidate can be established as a useful biomarker.
Identified potential biomarker proteins are scrutinized prior to being selected for further analyses. Foremost, the available technical validation of the antibody used to detect the protein must provide convincing evidence of target-specific binding. Experimental data including the known or presumed function and interaction partners of the target protein are also considered. Identified candidate biomarkers are tested in a series of additional technical verification steps to determine the specificity and reliability of the antibody, including the use of siRNA to knockdown the corresponding transcript and additional immunoassays to ascertain the functionality and reproducibility of the antibody (Fig. 4, lower row).
Qualification of biomarker value
Verified candidate antibodies are subsequently used to analyse the expression pattern in specific cancer-research TMAs, typically containing tumour tissues from 200 to 300 patients. The outcome of immunostaining is analysed and annotated by a certified pathologist to obtain a score based on the intensity and fraction of positive tumour cells for each patient sample. The annotation data are imported to the corresponding clinical database and used for biostatistical analysis to detect possible associations between levels of protein expression and clinical parameters. Using this approach, proteins identified as potential biomarkers are further investigated in a series of independent cancer-research TMAs to verify the original results (Fig. 4, upper row). If the results are reproducible in several different and independent patient cohorts, the biomarker is deemed to be of clinical significance.
For a biomarker to be truly useful in IHC and other applications, the antibodies used to detect the target protein must selectively bind to the intended protein with the highest level of certainty. The in-house-generated, affinity-purified polyclonal antibodies have the advantage that they are normally directed to several binding sites (epitopes) of the target protein, but they have the disadvantage that binding to multiple epitopes increases the risk of cross-specificity towards other proteins. The preferred option for launching a new biomarker is therefore to generate renewable, highly characterized, single epitope-specific monoclonal antibodies that selectively bind to the target protein. This can be accomplished by epitope mapping of the original PrEST antibody to define the epitopes with highest specificity , and the corresponding peptides can be synthesized and subsequently used as antigens for producing monoclonal antibodies (Fig. 4, lower row).
The exploitation of the Human Protein Atlas database in combination with the analysis of protein expression patterns in specific cancer TMAs has proven a successful strategy for biomarker discovery efforts. Cell type–specific proteins are rare, and the detection of such novel proteins in cells related to cancer provides a starting point for exploring the potential of a new diagnostic biomarker. The homeobox transcription factor SATB2 was identified with a selective pattern of expression and, within cells of epithelial lineages, SATB2 expression is restricted to glandular cells lining the lower gastrointestinal tract. The expression of SATB2 protein is largely preserved in cancer cells of colorectal origin. The results from a recent study of over 2000 tumours showed that SATB2 is a sensitive and highly specific marker for colorectal cancer with distinct positivity in 85% of all colorectal cancers. Moreover, SATB2 and cytokeratin 20 combined identified 97% of all tumours of colorectal origin .
Recently, the RNA-binding protein RBM3 and the rate-limiting enzyme in the mevalonate pathway HMG-CoAR were identified as potential biomarkers of human cancer. RBM3 and HMG-CoAR were initially identified as potential biomarkers through searches in the database, and the extent of protein expression was subsequently found to correlate with clinically relevant parameters in several types of cancer. Both RBM3 and HMG-CoAR were found to be independent predictors of recurrence-free survival in patients with breast and epithelial ovarian cancer [50–54].
Patients with tumour cells expressing high levels of RBM3 have a better overall prognosis compared with patients with tumours showing lower expression levels. Moreover, in vitro experiments have demonstrated a role of RBM3 with respect to cisplatin sensitivity. The cisplatin-resistant ovarian cancer cell line A2780/Cp70 has relatively lower levels of RBM3 compared with the cisplatin-sensitive parental cell line A2780, and siRNA knockdown of RBM3 in the latter resulted in an increased resistance to cisplatin [55, 56]. This indicates that patients whose tumours express high levels of RBM3 could benefit from cisplatin treatment, whereas alternative drugs may be considered in patients with a lack of or low RBM3 expression.
Similarly, decreased levels of HMG-CoAR have been coupled to decreased in vitro sensitivity to tamoxifen and, intriguingly, HMG-CoAR has also been identified as a predictor of response to adjuvant tamoxifen treatment in breast cancer, regardless of oestrogen receptor status . Moreover, the activity of HMG-CoAR can be inhibited by statins that are known to have antineoplastic effects [57–59], and studies on breast and ovarian tumours suggest that HMG-CoAR may prove useful as a surrogate marker of response to statin treatment in these cancers [51, 53]. Taken together, these recent findings offer an emerging panel of biomarkers for personalized medicine with respect to cisplatin, tamoxifen and statin treatments in breast and ovarian cancers.
Several additional promising cancer biomarkers have been explored, and protein expression patterns of use for clinical cancer research have been suggested for a number of tumour types. First, for colorectal cancer, a high expression of tumour-associated trypsin inhibitor was shown to correlate with liver metastasis and poor prognosis . There was also a correlation between low expression of tryptophanyl-tRNA synthetase in tumour tissue and increased risk of recurrence and shorter survival . Furthermore, growth differentiation factor 15 was a negative prognostic marker with high expression in tumour tissue and high plasma levels correlating with an increased risk of recurrence and reduced overall survival . Second, in malignant melanoma, decreased expression of the protein syntaxin-7 in the cells of melanocytic lineage appeared to be associated with more aggressive tumours . Additionally, SOX-10 was identified as a transcription factor selectively expressed in melanocytic cells, with highest levels in benign lesions and lowest levels of expression in melanoma metastases . Also, the amplification of topoisomerase I was associated with more advanced tumours and poor prognosis . Third, in prostate cancer, GAD-1 showed specific expression in both benign and malignant prostatic tissues, suggesting a role as a prostate-specific tissue biomarker . Moreover, the three novel markers somatic cytochrome c, intestinal cell kinase and inhibitor of nuclear factor-κB kinase subunit beta showed higher expression in prostate tumours as compared to benign prostatic tissue .
Antibody-based proteomics and the Human Protein Atlas resource can also be used to complement more basic tumour biology studies, by providing expression data from human tissues and clinical cancer samples. In a recent tumour biology study using a mouse model, large numbers of granulin-expressing bone marrow–derived haematopoietic cells were found in the tumour stroma of breast cancers responding to instigating signals. These cells were shown to induce a local inflammatory response and remodel the extracellular milieu through paracrine interactions with resident fibroblasts.This study also showed that the expression of granulin in human breast cancer was strongly correlated with the triple negative/basal-like breast tumour subtypes and that breast cancer patients with tumours positive for granulin staining had a significantly worse outcome in terms of overall survival .
Studies of diseases other than cancer can also benefit from the data and reagents generated within the Human Protein Atlas project. In a recent effort to develop new tools for measuring beta cell mass to use as an end-point in diabetes-related studies, a screen for new cell surface markers with specific expression on beta cells as compared to surrounding cell types was performed. In this first study, several new candidates were characterized and tetraspanin-7 was identified as a promising candidate for future development of new PET tracers for beta-cell imaging . In a study of epilepsy, the disc large homologue-5 gene encoding for the synapse-associated protein 102 was found to have a strictly neuronal pattern of expression and the results suggested a role of this protein in cortical hyperexcitability and epileptogenicity of malformations of cortical development .