Prediction of ecological niches
For bat SCoVrCs, the ecological niche was inferred using GPS data collected for viruses published during the last two decades. The list of the 19 available geographic records is provided in online supplementary Table S2. For bat SCoV2rCs, the ecological niche was initially predicted using the four geographic localities where viruses were previously detected1,5–7: two in Yunnan, one in northern Cambodia, and one in eastern Thailand (data set A). However, the use of only four records is questionable since Van Proosdij et al.22 have estimated that a minimum of 13 records is required to develop accurate distribution models for widespread taxa. For that reason, we used a genetic approach to increase the number of geographic records. Since the detection of identical CO1 sequences in different bat populations is indicative of recent dispersal events of females, we also selected the 17 geographic records where bats showed the same CO1 haplotypes than virus-positive bats (data set B: 21 points; see online supplementary Table S1).
For each of the three data sets (bat SCoVrCs; data sets A and B for bat SCoV2rCs), the 19 bioclimatic variables available in the WorldClim database31 were studied for an area corresponding to the minimum and maximum latitudes and longitudes of the selected points (19 points for bat SCoVrCs; 4 and 21 points, respectively for the SCoV2rCs data sets A and B) and the caret R package32 was used to determine the least correlated variables (|r|< 0.7)33. For bat SCoVrCs, the following five predictor bioclimatic variables were retained: Bio3 (isothermality), Bio4 (temperature seasonality), Bio5 (maximum temperature of the warmest month), Bio15 (precipitation seasonality), and Bio18 (precipitation of the warmest quarter). For data set A, the following seven predictor bioclimatic variables were retained: Bio3, Bio7 (temperature annual range), Bio10 (mean temperature of the warmest quarter), Bio13 (precipitation of the wettest month), Bio14 (precipitation of driest month), Bio15, and Bio18. For data set B, the following seven predictor bioclimatic variables were selected: Bio2 (mean diurnal range), Bio3, Bio7, Bio10, Bio13, Bio15, Bio17 (precipitation of the driest quarter), and Bio18. Ecological niche modelling was performed with the MaxEnt algorithm using ENMTools in R34. The MaxEnt approach was chosen for its ability to work with presence-only data sets and to produce results with a low sample size35. The area under the curve (AUC) of the receiver operating characteristic plot was used as a first measure of model accuracy, a value of 0.5 indicating model accuracy not better than random, and a value of 1 indicating perfect model fit36,37. To test for sampling bias, the distribution model using all selected localities was tested against a null model developed by 1000 times drawing an equal number of random points from the entire study area37. The position of the AUC value was tested against the 95% confidence interval (CI) of the 1000 AUC values of the null-models. If the AUC value is ≥ 95% CI null-model’s AUCs, the model is considered performing significantly better than a random model37.