Diagnosis of early stage ovarian cancer by 1H NMR metabonomics of serum explored by use of a microflow NMR probe

Abstract

We show that (1)H NMR based metabonomicsof serum allows the diagnosis of early stage I/II epithelial ovarian cancer (EOC) required for successful treatment. Because patient specimens are highly precious, we conducted an exploratory study using a microflow probe requiring only 20 μL of serum. By use of logistic regression on principal components (PCs) of the NMR profiles, we built a 4-variable model for early stage EOC prediction (training set: 69 EOC specimens, 84 healthy controls; test set: 40 EOC, 44 controls) with operating characteristics estimated for the test set at 80% specificity [95% confidence interval (CI): 65-90%], 63% sensitivity (95% CI: 46-77%), and an area under the Receiver Operator Characteristic Curve (AUC) of 0.796. Independent validation (50 EOC, 50 controls) of the model yielded 95% specificity (95% CI: 86-99.5%), 68% sensitivity (95% CI: 53-80%) and an AUC of 0.949. A test on cancer type specificity showed that women diseased with renal cell carcinoma were not incorrectly diagnosed with EOC, indicating that metabonomics bears significant potential for cancer type-specific diagnosis. Our model can potentially be applied for women at high risk for EOC, and our study promises to contribute to developing a screening protocol for the general population.

Figure 1

Representative 1D ¹H CPMG (top) and NOESY (bottom) spectra recorded for a serum specimen obtained from a patient diseased with early stage EOC. ¹H resonance assignments, for metabolites (see also http://www.hmdb.ca) for which significantly lower or higher concentrations were observed in this pilot study when comparing the spectra from early stage EOC and healthy control specimens are indicated in red and blue, respectively. Lower concentrations are observed, for alanine (p-value = 3.48×10⁻¹⁸), the choline moiety of phospholipids (4.44×10⁻²²), creatine/creatinine (<2.0 ×10⁻⁹), ‘LDL1’ representing CH₃(CH₂)n of lipid mainly in LDL (1.13×10⁻²⁰), CH₂CH₂CH₂CO of lipid mainly in VLDL (5.37×10⁻⁴), =CHCH₂CH₂ of unsaturated lipid (2.09×10⁻⁴), valine (6.64×10⁻⁹), ‘VLDL1’ representing CH₃CH₂CH₂C= of lipid mainly in VLDL (8.71×10⁻⁶). Higher concentrations are observed for acetoacetate (1.16×10⁻⁹), acetone (1.69×10⁻⁵), and β-hydroxybutyrate (1.07×10⁻⁸). Metabolites for which marginal concentration differences were detected are indicated in gray. Future studies have to reveal if those are statistically significant (see ‘Conclusions’).

Figure 2

(A) Score plot of first and second principal components obtained from Training Set. Using the same components, the data from (B) Test Set, and (C) Validation Set are displayed to show separation of early stage EOC patients and healthy women (see text). EOC patients (red) and healthy controls (black) are also separated in the third and fourth components (not shown). The predictive model was built with logistic regression from Training Set and accuracy was assessed with the Test and Validation Set (see Supporting Information).

Figure 3

Probability of early stage Epithelial Ovarian Cancer (p-EOC) calculated for each spectrum in (A) Training, (B) Test, and (C) Validation Set. The predictive model was constructed by logistic regression using four Principal Components of the joint CPMG and NOESY bin arrays based on the Training Set (Figure 2A). For all three sets, early stage EOC patients (red) exhibit significantly higher p-EOC values than their healthy controls (black) (see Supporting Information).

Figure 4

Receiver Operator Characteristic (ROC) Curves for the three logistic regression models built with CPMG bin arrays (‘CPMG’ model), NOESY bin arrays (‘NOESY’ model), and concatenated CPMG and NOESY bin arrays (‘joint’) as obtained for the Validation Set. Area under the ROC Curve (AUC) values measure the quality of predictive models. Those values (Table S1) are similar for the three predictive models, with the joint model being slightly superior for both the Test Set and Validation Set.