Rapid single-tier serodiagnosis of Lyme disease

Abstract

Point-of-care serological and direct antigen testing offers actionable insights for diagnosing challenging illnesses, empowering distributed health systems. Here, we report a POC-compatible serologic test for Lyme disease (LD), leveraging synthetic peptides specific to LD antibodies and a paper-based platform for rapid, and cost-effective diagnosis. Antigenic epitopes conserved across Borrelia burgdorferi genospecies, targeted by IgG and IgM antibodies, are selected to develop a multiplexed panel for detection of LD antibodies from patient sera. Multiple peptide epitopes, when combined synergistically with a machine learning-based diagnostic model achieve high sensitivity without sacrificing specificity. Blinded validation with 15 LD-positive and 15 negative samples shows 95.5% sensitivity and 100% specificity. Blind testing with the CDC's LD repository samples confirms the test accuracy, matching lab-based two-tier results, correctly differentiating between LD and look-alike diseases. This LD diagnostic test could potentially replace the cumbersome two-tier testing, improving diagnosis and enabling earlier treatment while facilitating immune monitoring and surveillance.

Fig. 1. Overview of the paper-based multiplexed vertical flow assay (xVFA) and point-of-care diagnosis of Lyme disease (LD).

a Transmission of Borrelia burgdorferi through the bite of Ixodes ticks and the presentation of various antigens generating an immune response from the host. b Comparison of incidence of LD cases in the northeastern US from 2000 and 2019 indicating an increase in the incidence of cases due to the growing population of ticks. Worldwide incidence of LD in 2019. Legend indicates the number of cases per 100,000 people. c Centralized laboratory-based two-tier serology of LD uses relatively expensive instruments and trained personnel, resulting in high turnaround time and cost per test. d Point-of-care xVFA assay using low-cost paper layers and a smartphone reader that provides results for a multiplexed LD assay in <20 min. e The xVFA contains a selected peptide panel immobilized on a nitrocellulose membrane that reacts with IgM and IgG antibodies from LD patient serum. f Stability of modVlsE-FlaB peptide and VlsE recombinant protein indicating a loss in performance of a protein immobilized assay by more than 50% over a 90-day period. Data were presented as mean values ± SD, with standard deviation indicating three replicates (N = 3). g Combining IgM and IgG detection in a single xVFA assay enhances the sensitivity of an individual immunoreaction spot (N = 3). Data were presented as mean values ± SD. h Smartphone-based portable reader and automated image processing of the signals from the peptide panel before and after the assay, yielding normalized signal intensities. The individual peptide spots are analyzed using a multiplexed model to classify samples as either LD positive or negative. Panels a, c, d were created with BioRender.com released under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International license (https://creativecommons.org/licenses/by-nc-nd/4.0/deed.en).

Fig. 2. B. burgdorferi antigenic peptide library screening and selection of prevalent LD-specific epitopes.

a Overview of the screening of epitopes to select the most relevant peptide antigens that function in the paper-based xVFA platform. b Heatmap representing the normalized signal intensities of peptides screened against patient samples positive (+) and negative (−) for LD, utilized in the development of the paper-based multiantigen xVFA platform. c Variance in the normalized signal intensity for each antigen peptide, determined through xVFA screening of control patient sera. The variance is plotted in descending order of reactivity, illustrating the comparative activity of the peptides against both positive and negative patient serum samples, highlighting the most active antigens. d Heatmap representing the correlation of peptides against each other. Inset shows examples of highly correlated peptides (Rec164 and OspC1) and non-correlated peptides (modVlsE-FlaB and OppA4), which yield additional information for a diagnostic panel. Panel a was created with BioRender.com released under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International license (https://creativecommons.org/licenses/by-nc-nd/4.0/deed.en).

Fig. 3. Training of the multiantigen xVFA panel using early-stage LD patient samples obtained from the Lyme Disease Biobank (LDB).

a The multiantigen coated sensing membrane and map of each antigen location spotted in duplicates along with the positive and negative control reaction spots. b Heatmap displaying the normalized signal intensities calculated from two immunoreaction spots for each of the three xVFA replicate tests performed on the individual patient samples in the training subset (40 samples × 3 replicate, 120 xVFA tests). The color scale on the top represents the average %CV in measurement for all peptides per patient sample. c Receiver operator characteristics (ROC) resulting from the neural network-based multiplexed diagnostic model comparing the model’s performance when different numbers of peptides are used to train the network. The inset shows the confusion matrix for a three-peptide model. d Bar plot showing the area under the curve (AUC) for the different input features used to train the diagnostic model. A three-peptide model including single immunoreaction spots from modVlsE-FlaB, Var2FlaB, and OppA4 yields the highest AUC of 0.94. e The final prediction outcome from the neural network-based diagnostic model, indicated as values from 0 to 1, where each dot represents a test that was performed on the xVFA using a patient sample (40 patient samples tested in triplicate). The red dotted line at 0.5 indicates the threshold for positivity of the diagnostic model.

Fig. 4. Assessment of diverse samples from the CDC using the multiantigen xVFA and the optimized neural network-based diagnostic model.

a Prediction outcome from the neural network-based diagnostic model using the trained network indicated as values from 0 to 1, where each dot represents a test that was performed on the xVFA using patient samples that had different stages of LD, healthy control samples from regions where LD is endemic and non-endemic, cross-reactive samples from look-alike diseases such as fibromyalgia, rheumatoid arthritis, multiple sclerosis, mononucleosis, syphilis, and severe periodontitis. The red dotted line at 0.5 indicates the threshold for positivity of the diagnostic model. Three xVFA tests were run per sample. b Bar plot displaying the percentage concordance of the xVFA assay with the standard two-tier testing (STTT) and modified two-tier testing (MTTT) IgM diagnosis. c Bar plot comparing the accuracy of the single-tier xVFA assay with individual centralized lab-based STTT and MTTT reference tests, including the overall diagnostic outcomes using each algorithm.

Fig. 5. Validation of the single-tier xVFA and the optimized diagnostic model using samples from the Lyme Disease Biobank in cohort 3.

a Prediction outcome from the neural network-based diagnostic model using the trained network indicated as values from 0 to 1, where each dot represents a test that was performed using the xVFA (30 samples × 3 replicates, 90 xVFA tests). The dotted horizontal line represents the threshold for a positive diagnosis. The cohort consisted of 15 LD positive samples and 15 LD negative samples that were confirmed using standard two-tier testing for ground truth. b Confusion matrix summarizing the performance of the xVFA in terms of true positives, false negatives, false positives, and true negatives, with calculated sensitivity and specificity. The assay demonstrates high diagnostic precision, validating its effectiveness for Lyme disease detection.