Résumé
The use of sniffer dogs has expanded rapidly over the past two decades, particularly in conservation biology and medical diagnosis. Plant health surveillance systems could also benefit from this low-tech, real-time and mobile detection method that can be deployed over vast areas and diverse environments. While dogs have proven capable of detecting phytopathogenic bacteria and fungi, their ability to identify plant viruses remains undocumented. We developed a rigorous training and evaluation protocol to assess the performance of two dogs in detecting peach plants infected by the plum pox virus (PPV; Potyvirus plumpoxi), responsible for a serious viral disease of Prunus trees, currently monitored in France through symptom-based inspections. The dogs were first trained to discriminate symptomatic PPV-infected peach plants from certified virus-free plants and from plants infected by PDV, PNRSV and ACLSV. Using preliminary data collected during training, we used simulations to identify an experimental design allowing good statistical power (precision and accuracy) for performance estimates (detection sensitivity and specificity) while ensuring experimental feasibility. The dogs were tested in 20 runs of 18 plants each, where the number of PPV-infected plants was randomly assigned. A double-blind protocol was implemented to prevent behavioral bias. One dog achieved 85% sensitivity and 99% specificity. In a second phase, we assessed their ability to detect volatile organic compounds (VOCs) captured over 3–4 days from the same plants at full vegetation and bud break. For both dogs, detection performance using VOC-capture devices was excellent for plants in vegetation (86% sensitivity and 92% specificity) but was very poor for plants at bud break. The ability of dogs to detect weak VOC signals (winter dormancy, latent infections) needs further evaluation.