Biomarkers in the age of omics: time for a systems biology approach

Abstract

Limitations to biomarker discovery are not only technical or bioinformatic but conceptual as well. In our attempt to offer a solution, we are highlighting three issues that we think are limiting progress in biomarkers discovery. First, the confusion stemming from the imposition of a pathology-type immunohistochemical marker (IHCM) concept on omics data without fully understanding the characteristics and limitations of IHCMs as applied in clinical pathology. Second, the lack of serious consideration for the scope of disease heterogeneity. Third, the refusal of the biomedical community to borrow from other biological disciplines their well established methods for dealing with heterogeneity. Therefore, real progress in biomarker discovery will be attained when we recognize that an omics biomarker cannot be assigned and validated without a priori data modeling and subtyping of the disease itself to reveal the extent of its heterogeneity, and its omics' clonal aberrations (drivers) underlying its subtypes and pathways' diversity. To further support our viewpoints, we are contributing a novel a systems biology method such as parsimony phylogenetic approach for disease modeling prior to biomarker circumscription. As an analytical approach that has been successfully used for a half of a century in other biological disciplines, parsimony phylogenetics simultaneously achieves several objectives: it provides disease modeling in a hierarchical phylogenetic classification, identifies biomarkers as the shared derived expressions or mutations--synapomorphies, constructs the omics profiles of specimens based on the most parsimonious arrangement of their heterogeneous data, and permits network profiling of affected signaling pathways as the biosignature of disease classes.

FIG. 1.

Pathogenesis as a heterogeneous process in cancer, degenerative disease, and temporary illness disturbs the differentiation of the affected cells. The process is reversible in temporary illness, but irreversible in cancer and degenerative diseases. The latter two are associated with mutations and permanently deregulated expressions. Multiple arrows signify that several pathways may produce the same disease phenotype, a condition that complicates biomarkers discovery and calls for a priori modeling of the disease to reveal its classes.

FIG. 2.

A most parsimonious cladogram produced by PHYLIP's MIX using Camin-Sokal parsimony algorithm. Dataset GDS1439 (http://www.ncbi.nlm.nih.gov/sites/entrez?db=gds) is comprised of six benign specimens, as well as seven primary and six metastatic prostate tumors. The cladogram shows a major bifurcation that delineates two clades; the first composed of all primary tumors and four benign specimens (node 6, supported by 24 synapomorphies), and the second composed of all metastatic tumors and two benign specimens (node 13, supported by 717 synapomorphies). A clade of primary tumors is delimited by 1,018 synapomorphies (node 8), whereas a clade of the metastatic tumors is delimited by 4,494 synapomorphies (node 15). Synapomorphies at nodes 6, 8, 13, and 15 are considered clonal (driver) expression aberrations. Pooled primary tumor specimens PX1 and PX2 grouped into a clade (node 12), whereas pooled metastatic specimens, WX1 and WX2 formed a clade (node 18).

FIG. 3.

Schematic summary of network analyses produced by Genomatix's BiblioSphere. (A) A summary of affected nodal pathways in primary prostate tumors at node 8 of Figure 2, and (B) in metastatic prostate tumors at node 13 of Figure 2. More details are provided in Supplementary Figures 1 and 2.