All over the world, the successive and frequently larger waves of coronavirus disease 2019 (COVID-19) have made it difficult to obtain adequate supplies of the testing kits and reagents required for the gold standard of infection detection: reverse transcriptase polymerase chain reaction (RT PCR). The time required to obtain a readout of the results is also long, creating transmission hazards for those with asymptomatic infection. Meanwhile, large-scale exposure to the virus occurs, and the pandemic gallops out of control.
A new study in PLOS ONE discusses the novel use of dogs to sniff out the virus in patient samples as a potential additional diagnostic route to ease the situation.
The virus responsible for this pandemic, namely, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has a mortality rate of 0.5-1%, by current estimates. The preponderance of asymptomatic cases and the long incubation from exposure to symptom onset has led to the rapid and extensive spread of the virus.
The current study explores an alternative testing method that is cost-effective, rapid, highly sensitive and specific. The principle of the test is the use of the volatile organic compound signature (VOC) of the virus.
Similar approaches have been used already to diagnose or monitor other diseases. These have made use of chemical sensors and biological sensors, in the form of gas chromatography-mass spectrometry, and canine detection.
Training dogs to pick up VOC profiles in SARS-CoV-2
Dogs can pick up specific VOCs produced by a particular disease condition, from a range of biological fluids, including urine and saliva. Such a profile is known to be present with SARS-CoV-2 infection, indicating that tracheal, salivary, and sweat samples can be used to train dogs to recognize it.
However, earlier studies used the same samples presented several times to demonstrate the ability of dogs to distinguish SARS-CoV-2-positive samples from negative samples.
The need to keep animals and laboratory staff safe during testing has prompted the use of urine samples, which have lower viral loads compared to saliva, while also being easily collectible.
What were the results?
The current study shows that dogs trained to distinguish samples from infected individuals, after detergent inactivation, can be trained to pick up heat-inactivated samples from COVID-19 patients, while ignoring samples from negative controls in the training phase.
During the training period, the dogs picked up 70% of positive samples while showing specificity of 99%. The overall accuracy was 94%.
When testing began, using novel heat-inactivated samples, the overall sensitivity was 68% (true positive results to the number of actual positives in the tested group). The specificity remained high at 99%.
The dogs were then exposed to virus-negative saliva samples along with the positive urine samples. This was followed by testing to see whether their training on urine samples helped them identify positive saliva samples.
This showed that all dogs also recognized the virus in the saliva from the same patients whose positive urine samples had been used in training.
Suboptimal testing results on novel samples
Finally, it is necessary that trained dogs should be able to distinguish the presence of SARS-CoV-2 in novel urine samples while ignoring negative samples. None of the samples in this round had been presented earlier to the dogs, either individually or as part of a mixed sample.
Here, the accuracy fell to 11%, as only a few dogs recognized positive samples.
False ‘false alert’
It is noteworthy that the majority of dogs repeatedly checked at two negative control samples which were used for training. These resulted in recognition as positive, whether used alone or in a mixture.
While identified as negative based on the results of an RT PCR at the time of entry, they came from a recently recovered individual who had been positive during the infection and from a person who had symptoms resembling COVID-19 but tested negative.
Obviously, these samples confused the training. However, the fact that most dogs showed recognition indicates that some degree of generalization of recognition had occurred.
Conclusions and future directions
The training utilized in this study did not result in documented generalization of a SARS-CoV-2 positive odor profile, despite dogs showing impressive discrimination between positive and negative samples.”
Earlier studies showed that dogs could keenly discriminate the VOC profile in positive SARS-CoV-2 samples from negative controls, but these were basically like revision tests, using the same samples over and over again. This could mean that the dogs were honing their ability to distinguish negative and positive samples in the training set, rather than learning the VOC of SARS-CoV-2 in urine and saliva.
The current findings show that dogs trained on a small number of samples then move to a more generalized mix of samples but are unable to generalize their discrimination when faced with completely novel samples.
Further studies will show if the dogs would have performed better if they had been exposed to fewer identical samples in the training period and if they had been exposed to novel samples in a blinded study before being trained.
The need to use carefully confirmed negative control samples is obvious from the confusion caused by samples that were repeatedly recognized as positive during the initial part of the training period before they were removed. The false-negative rate is significant with RT PCR, making this a difficult task.
Moreover, there is no current technology to understand how the sample from an infected or recently recovered individual ‘smells’ to a dog or how long this VOC profile lingers.
The number of samples and their presentations likely need to be refined to allow dogs not just to distinguish negative from positive controls, but also to generalize to novel samples.
Dogs are already being used to sniff out infected individuals in real-time. To help them generalize their scent discrimination, a larger number of positive, negative and novel samples will probably be crucial. The use of biological, chemical and electronic detectors during such training will help to identify the specific odor profile that is aimed at for recognition.