Abstract
Introduction:
To further evaluate the utility of surface-enhanced laser desorption and ionization-time of flight mass spectroscopy (SELDI-TOF MS) proteomics in the diagnosis, prognosis, monitoring response to therapy, and follow up of patient with myeloma, we examined the ability of protein chips with 4 different surface chemistries to detect biomarkers of lytic bone disease.
Method:
Sera from 47 untreated myeloma patients were collected and stored at −80°C for future analysis. 22 serum samples were from patients with 1–26 lytic bone lesions as identified on X-ray skeletal surveys and 25 serum samples were from patients without lytic lesions. The sera were analyzed by SELDI-TOF MS using Ciphergen’s ProteinChip Biology System II (PBS II), to identify protein patterns associated with lytic bone disease. Each sample was applied in 2 replicates to randomly assigned spots on protein chips with different surface chemistries: immobilized metal affinity capture (IMAC30) activated with Cu++, weak cathion exchange (CM10), reverse phase (H50), and strong anaion exchange (Q10), all under conditions of low stringency, using a Biomek2000 robot. The mass spectra of proteins, generated using an average of 66 laser shots, were calibrated using peptide standards and normalized to total ion current using CiphergenExpress 3.0 software. We randomly selected 80% of patients for developing classification models that separate patients with lytic bone lesions from those without lesions, using a stepwise logistic regression method; 76 calibrated and normalized spectra from randomly selected patients, 18 with lytic bone lesions and 20 without lytic lesions were used for model development. The other 18 spectra from 4 patients with lytic lesions and 5 without lesions were used as a test dataset. The same datasets were used for all 4 chip types. Protein peaks that were significantly different between the two groups were used for modeling.
Results:
All 4 chip types yielded models with fit accuracies of 71%–92%. The spectra produced by each chip were different, reflecting the differences in surface chemistries, and each model used different protein peaks from each of the chip types; the IMAC30-based model used 11 protein peaks ranging in size from 2903 to 8226 kilodaltons (KDa); H50 model used 7 peaks, 2802–18837 KDa in size; CM10 model used 5 peaks, 2805–23307 KDa; and Q10 used 3 peaks, 6975–37210 KDa. Prediction accuracy was 89% for CM10, 78% for IMAC30, 76% for Q10 and only 33% for H50. CM10 and IMAC30 based models correctly assigned all patients with no lytic bone lesions. Wrongly assigned patients were different for the two chip types.
Conclusions:
Since each model wrongly assigned different patients and used different protein peaks, it is likely that a model that combines information from several surface types (consensus approach) will prove to provide the most accurate predictions, as needed for making clinical decision.
Author notes
Corresponding author
This feature is available to Subscribers Only
Sign In or Create an Account Close Modal