L. Renee Ruhaak1; Sandra Taylor1; Cynthia Williams1; UyenThao Nguyen1; Lauren Dimapasoc1; Sureyya Ozcan1; Carol Stroble1, 2; Suzanne Miyamoto2; Kyoungmi Kim1; Gary Leiserowitz2; Carlito B. Lebrilla1
1University of California, Davis, Davis, CA; 2UC Davis Comprehensive Cancer Center, Sacramento, CA
NOVEL ASPECT: Altered serum glycan profiles were validated as discriminating OC cases from controls, showing its great potential as a diagnostic tool.
Protein glycosylation plays important roles in cancer; aberrant glycosylation has been observed with malignant transformation, and altered glycosylation profiles were observed in serum and plasma of cancer patients compared to healthy controls. The survival rates of ovarian cancer (OC) are lower than most other cancers that affect women, but if tests with better accuracy than the test for CA125 were available for early detection then more lives could be saved. While the literature reports altered serum glycosylation profiles with ovarian cancer, the predictive values of such candidate biomarkers have not been determined and results have not been validated in independent test sets. We now report the potential of serum glycomics analysis using nLC-PGC-chip-TOF-MS as a diagnostic test for ovarian cancer.
Pre-operative sera of OC cases and healthy controls consisting of 43 stage III-IV OC cases and 49 age-matched controls were used as a training set for biomarker detection. Independently a set of patient sera was collected from 52 stage I-II cases, 52 stage III-IV cases, 52 cases with low malignant potential and 52 age-matched controls as a test set for validation. A high-throughput 96-well based nLC-PGC-chip-TOF-MS strategy, which was previously validated for its stability and repeatability, was employed to evaluate serum N-glycan profiles in each of the samples. N-glycan peak integrals were used for biostatistical analysis to evaluate the OC differential potential for the glycan features in the training set. These results were then validated in the independent test set.
Within the training set, 330 glycan compositions were detected in at least one of the samples analyzed. Of the glycan compositions that were consistently detected in the samples, levels of 36 compositions were shown to alter significantly (either over- or under-expressed) with OC at a false discovery rate of <0.05. Among the most significant glycan compositions were Hex5HexNAc4Fuc1, Hex4HexNAc4, Hex5HexNAc5Fuc2NeuAc1, Hex5HexNAc3, Hex6HexNAc3 and Hex5HexNAc6Fuc1NeuAc2. The most informative glycan (Hex6HexNAc3) singly yielded an AUC value of 0.896 with 93% sensitivity and 75% specificity. Multiplex classifiers combining one or more glycans together were developed with the highest accuracy of 91.2% (sensitivity 86.0%, specificity 95.8%) when combining nine glycans.Using the test set, we were also able to observe 330 glycan compositions, and levels of 33 compositions were significantly altered with OC. Twenty glycan compositions were significantly altered with OC stage III-IV in the same direction in both the training and the test set. The glycan composition with the highest classification accuracy in the stage III-IV samples of the training set continued to perform well in the independent test set. When testing the multiplex classifiers developed in the training set in samples of the independent testing set, an accuracy of 78% was achieved. Independent of the training set, which only contained stage III-IV samples, good separation was also obtained between the healthy controls and the stage I-II OC cases, with a classification rate of 83%. These results indicate that glycan profiles are promising new tools which could lead to improvement in the detection of ovarian cancer.