Distinguishing patients with active tuberculosis (TB) from those with latent TB is an
important clinical problem. The SELDI-TOF MS (Surface Enhanced Laser
Desorption Ionisation – Time of Flight Mass Spectrometry) platform allows for high
throughput detection of multiple proteins in biological fluids. Proteomic patterns
reflecting host-pathogen interaction can be used as a tool to aid our understanding of
the Natural History of Tuberculosis.
Methods: Plasma samples were collected prospectively in a shanty town in Lima,
Peru. Latent and active TB status was defined using the Tuberculin Skin Test (TST),
Quantiferon (QFN) assay and TB culture. Crude plasma and fractionated plasma
samples were analysed on weak cationic CM10 chip surfaces using a Biomek 3000
Laboratory Automation Workstation. Spectra were generated using a ProteinChip
System 4000 Mass spectrometer. Data was analysed using a Support Vector Machine.
Results:
Samples were collected from 154 patients with active TB, 112 patients with
respiratory symptoms suggestive of TB and 151 healthy controls. Multiple peaks
differed significantly between active TB patients and unhealthy controls. Trained
optimal classifiers discriminate between:
i) active TB and unhealthy controls with 84% accuracy (87% sensitivity, 79%
specificity) in crude plasma and up to 89% accuracy (90% sensitivity, 88%
specificity) in fractionated plasma
ii) active TB and latent TB with 89% accuracy (90% sensitivity, 89% specificity)
iii) latent TB and no TB in healthy controls with 77% accuracy (67% sensitivity, 84%
specificity).
Conclusions:
SELDI-TOF MS proteomic profiles in combination with trained optimal classifiers
accurately discriminate active TB from other respiratory disorders. The classifier for
latent TB was not as accurate, but active TB could be discriminated from latent TB.