Evaluation of a viable-cell detection assay of Xanthomonas fragariae with latent class analysis

Authors: Turechek, William; WANG, H - Clemson University

Submitted to: PhytoFrontiers
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/26/2022
Publication Date: 3/20/2023
Citation: Turechek, W., Wang, H. 2023. Evaluation of a viable-cell detection assay of Xanthomonas fragariae with latent class analysis. PhytoFrontiers. 3:214-224. https://doi.org/10.1094/PHYTOFR-05-22-0052-FI.
DOI: https://doi.org/10.1094/PHYTOFR-05-22-0052-FI

Interpretive Summary: The development and evaluation of diagnostic tests is a principal area of research in plant pathology. In this paper, a latent class analysis (LCA) was conducted to evaluate the performance of a new, viable-cell detection assay (propidium monoazide (PMA)-qPCR) and compared to a simple plant bioassay and a standard Taqman qPCR assay for detection of Xanthomonas fragariae, the causal agent of angular leaf spot of strawberry. Results from the LCA and ROC curve analysis showed that the PMA-qPCR outperformed the qPCR for detection of viable cells, but that the LCA provided more robust estimates of the statistics used to evaluate test performance. Viability testing is extremely useful in certification and disease management applications, and with the information provided on test performance generated by LCA, the test can now be put to practical use to design sampling strategies to account for the errors associated with diagnostic testing.

Technical Abstract: The development and evaluation of diagnostic tests is a principal area of research in plant pathology. In this paper, a latent class analysis (LCA) was conducted to evaluate the performance of a new, viable-cell detection assay (propidium monoazide (PMA)-qPCR) and compared to a simple plant bioassay and a standard Taqman qPCR assay for detection of Xanthomonas fragariae, the causal agent of angular leaf spot of strawberry. Because LCA requires binary test results (+/-), a receiver operating characteristic (ROC) curve analysis was first conducted to identify optimal cutpoints to facilitate dichotomization of the test results for both the PMA-qPCR (Ct = 31) and the qPCR (Ct = 21.5) assays, as well to provide a secondary analysis to compare LCA results. In addition to the LCA with optimal cutpoints, a second LCA was conducted using Ct = 35 for the qPCR assay, which is considered a standard threshold for detection in many qPCR applications. Further, diagnostic test performances were evaluated with respect to the categorical or grouping variables ‘material’ (eleven materials were evaluated), storage ‘temperature’ (room temperature and -4 °C), and ‘days after inoculation’ (ten time points were evaluated). Results from the ROC curve analysis showed that the sensitivity and specificity of the PMA-qPCR assay were 0.80 and 0.79 at Ct = 31 and 0.72 and 0.79 for the qPCR at Ct = 21.5, compared to estimated sensitivities and specificities of 0.92 and 0.91 for the PMA-qPCR and 0.83 and 0.89 for the qPCR in an LCA at the same thresholds. In both analyses, the PMA-qPCR outperformed the qPCR for detection of viable cells, but estimated sensitivities and specificities were approximately 15% higher in the LCA. The sensitivity and specificity of the qPCR at Ct = 35 were 1.0 and 0.01 in the ROC curve analysis compared to sensitivities and specificities of 1.0 and 0.02 for the qPCR and 0.80 and 0.99 for the PMA-qPCR in the LCA (the sensitivity and specificity of the PMA-qPCR in the ROC curve analysis was the same). These values were nearly identical with the exception of the estimated specificity of the PMA-qPCR being 25% higher in the LCA. There were clear differences in the bioassay, PMA-qPCR and qPCR tests with respect to the grouping variable. However, with 220 combinations of material × storage temperature × storage times and because many combinations generated similar values of sensitivity and specificity, a simple approach to utilizing this information would be to rely on the average values generated with the model with no grouping variables. Viability testing is extremely useful in certification and disease management applications, and with the information on test performance generated by LCA, the test can now be put to practical use to design sampling strategies to account for the errors in testing.