Data were presented as means ± standard deviation. One-way ANOVA followed by either Tukey’s or Dunnett’s multiple comparisons test was performed to identify significant variations at 95% confidence interval, using GraphPad Prism 7.0 (GraphPad Software Inc., USA).

To statistically investigate the EV phenotypic changes in response to treatment, SERS intensities at peaks of 1075, 1375, 1335, and 1000 cm−1—representing the expression of each biomarker—were used as LDA input variables. LDA then generated discriminant functions that consisted of different linear combinations of input variables. The resulting discriminant functions were uncorrelated with each other, and each function maximized the difference between groups on that function. The first two discriminant functions that explained most input variables were selected for EV phenotypic clustering. Discriminant scores generated from these two discriminant functions were plotted to describe the differences between each data point. LDA was performed with SPSS 19.0 software package (SPSS Inc., USA).