Rapid Detection of Clostridium difficile Toxins in Stool by Raman Spectroscopy.

Area of Science:

• Biomedical Diagnostics
• Spectroscopy
• Infectious Diseases

Background:

• Clostridium difficile infections (CDI) are defined by diarrhea and a positive stool test or pseudomembranous colitis.
• Current diagnostic methods like nucleic acid amplification testing can detect toxigenic C. difficile but may lead to overdiagnosis due to asymptomatic colonization.
• Toxin enzyme immunoassays for detecting pathogenic TcdA and TcdB have low sensitivity and moderate specificity.

Purpose of the Study:

• To evaluate Raman spectroscopy (RS) as a novel, reagent-free method for sensitive detection of C. difficile toxins (TcdA and TcdB) in stool samples.
• To address the overdiagnosis of CDI by improving the accuracy of toxin detection.
• To compare the performance of RS with existing diagnostic methods.

Main Methods:

• Stool samples negative for CDI were spiked with known concentrations of TcdA and TcdB.
• Raman spectroscopy was performed on photo-bleached stool supernatant samples.
• Machine learning models (SVM, Random Forest, Gradient Boosting, PCA-LDA) were used to analyze spectral data, with datasets split for training and testing.

Main Results:

• Raman spectroscopy successfully distinguished spiked stool samples from unspiked samples across various toxin concentrations (0.1 pg/mL to 1 ng/mL).
• Machine learning models achieved accuracies ranging from 64% to 77%.
• The best performing models (Gradient Boosting Machine, PCA-LDA, SVM Linear Kernel) demonstrated sensitivities from 69% to 90% and specificities from 43% to 78%.

Conclusions:

• Raman spectroscopy can detect TcdA and TcdB in stool with moderate-to-high sensitivity and low-to-moderate specificity.
• RS shows potential for rapid, reagent-free detection of C. difficile toxins at clinically relevant levels.
• Further improvements in sensitivity and specificity are anticipated with deep learning methods, offering a solution to mitigate CDI overdiagnosis.