Proteomic analysis of serum yields six candidate proteins that are differentially regulated in a subset of women with endometriosis

Abstract

Objective: To identify potential novel biomarkers that differ between subjects with and without endometriosis and that might aid in developing a noninvasive, serum-based diagnostic test.

Design: Case-control evaluation of a diagnostic test.

Setting: University medical center.

Patient(s): Consenting women of reproductive age undergoing laparoscopy for indications of pain, infertility, elective tubal ligation, tubal reanastomosis, or other benign indication.

Intervention(s): Diagnostic laparoscopy and peripheral venipuncture.

Main outcome measure(s): Concentrations of low-molecular-weight proteins in serum; surgical staging of endometriosis.

Result(s): Six proteins were found that were differentially expressed between those with and without disease and that had good diagnostic properties. Taken together in a two-step diagnostic algorithm, we were able to diagnose 55% of subjects, with 99% accuracy as to the status of disease. Further combining this algorithm with that derived by our previous study of serum putative markers (monocyte chemoattractant protein-1, migration inhibitory factor, leptin, and CA-125) improved our diagnostic capability to 73% of subjects, with 94% overall accuracy.

Conclusion(s): This study is the critical first step in the identification of potential novel biomarkers of endometriosis. Future identification of the proteins and further validation in a second population is needed before applying these findings in clinical practice.

Figure 1

Sample CART output. This non-parametric statistical procedure creates two child nodes from each parent node by splitting on the independent variable that best classifies subjects into binary groups. In this hypothetical example, all 100 subjects are classified using 2 independent variables, with 30 subjects deemed disease positive.

Figure 2

Protein Marker Classification Tree Optimizing Sensitivity: Sensitivity=100% (95% CI: 95–100%) and specificity=45% (95% CI 34–57%). The Class assignment of subjects in each node is shown under the node number. Class CONT is the control group and Class ENDO is the Endometriosis group. CART initially categorizes all study subjects in Node 1 as ENDO. Colored bars give a graphical representation of the proportion of subjects from each group assigned to that child node, red=control, blue=endometriosis. Splitting variables are shown in the parent node (by protein size in Da), with the cut-off for the split shown above the child node in grey. N=number of subjects. Protein sizes are indicated in each node as the mass preceeded by “M” (e.g. M3774 is a protein with a molecular mass of 3774 Da).

Figure 3

Protein Classification Tree Optimizing Specificity: Specificity=99% (95% CI 93–100%) and sensitivity=66% (95% CI 52–77%).