Simca p version 15

Quantitative variables were presented as means; median with standard deviations and group differences tested by Student’s t-test. Spearman correlation coefficients were used to assess co-variations between C. matruchotii and bacteria in saliva or tooth biofilms. These analyses were performed using SPSS version 26 (IBM Corporation, Armonk, NY, USA). The tests were two-tailed and p-values < 0.05 considered significant. Taxa associated with high levels of C. matruchotii were identified in multivariate partial least squares (PLS) regression. The results are presented in a PLS loading plot with influential variables (i.e., a variable importance in projection (VIP) score >2.0). The PLS models evaluated the abundance of C. matruchotii in saliva and tooth biofilm as dependent variables (y-variable) against an independent x-block composed of all other identified bacterial species (x-variables, n = 311). All variables utilized for PLS regression analysis were auto-scaled and logarithmically transformed as needed to improve normality. SIMCA P+ version 15.0 (Umetrics AB) was used for these analyses.

The GC-MS data were converted to a netCDF format file and processed using MetAlign software for peak detection and alignment [59 (link)]. MetAlign parameters were set according to the AIoutput scaling requirements: a peak slope factor of 2, peak threshold of 10, average peak width at half height of 25, and peak threshold factor of 4. These settings corresponded to a retention time of 3–26 min and mass range of 85–500 m/z. The result of the data (CSV) was imported into AIoutput software for peak prediction and identification [60 (link)]. After feature intensities were normalized relative to the intensity of the internal standard (retention time 11.205 min, m/z 147), multivariate statistical analyses were performed. To visualize the variance of metabolites, PCA and PLS-DA of GC-MS data were performed using SIMCA-P, version 15.0 (Umetrics, Umea, Sweden). For model validation, a 200-fold cross validation was performed. Metabolites with a VIP score greater than 1.0 and p-value < 0.05 using the Student’s t-test were considered to have discriminatory power to distinguish between the two groups. Multiple testing was corrected, using the positive FDR (type 1 error) by computing the q-values after the t-test. Metabolites were identified by comparing their mass spectra with the AIoutput software, NIST library, and the human metabolome database (HMDB).

All the experiments were independently performed in, at least, triplicate under the same environmental conditions. Data is expressed as mean ± SE of three independent replicates. Significant differences between groups were determined by ANOVA, followed by two-tailed multiple t-tests with Bonferroni correction, performed with XL-STAT 2019 biostatistics software (Addinsoft). All results were considered significant at p < 0.05, represented by different letters. Principal component analysis (PCA) was performed using SIMCA P+ version 15.0 (Umetrics AB, Umeå, Sweden). Variables were mean-centered and unit variance-scaled prior to PCA. Hierarchical clustering analysis and Pearson correlation coefficient analysis were obtained with PAST 3.0, with significant thresholds at p < 0.05, p < 0.01 and p < 0.001 represented by *, ** and ***, respectively.