3.1.4 Reproducibility of metabolite quantification
To evaluate the reproducibility of the methods detected, we used average relative standard deviation (RSD) as an indicator to evaluate the reproducibility of the extraction solvent. The RSD results of 43 metabolites obtained using 12 combinations of quenching and extraction methods were shown in Table 2 . The introduction of13C internal standards allowed us to quantitatively evaluate extraction methods with regard to biological reproducibility. In 2D Hela cells, the difference between the average RSD values of different methods was not significant, mostly between 20%-30%. S-A method had the lowest average RSD (18.9%) while the N2-M/C method had the largest average RSD (39.9%). The average RSD of the N2-A method which behaved excellent regarding metabolite coverage and abundance was 27.8%. The RSD values of 11 methods (except N2-M/C) were below 30%, indicating good reproducibility independent of quenching and extraction procedures. In addition, in all methods (except N2-M/C), the percentage of individual metabolites with RSD <30% was higher than 58% (Table 3 ). Guenther et al. found that in HCT116 cells, the percentages of metabolites with RSD < 30% cultured in 10% and 20% fetal calf serum (FCS) conditions using different quenching solvents, extraction ratios were ranging from 0% to 79% which had a large error [20]. In our study, the percentage of individual metabolites with RSD <30% for the optimal N2-A method could reach 61.0%, and the largest percentage of individual metabolites with RSD <30% of all methods was 69.9%, which proved that our methods were reproducible for most metabolites and N2-A method was relatively stable for the extraction of metabolites detected. By analyzing the RSDs of 12 methods, we found that the sample preparation methods using liquid N2 as the quenching agent had higher RSDs than the other two quenching agents, and the proportion of metabolites with RSDs less than 30% was also lower. We speculated that this might be caused by the difficulty in accurately determining the added volume and the volatilization time of liquid nitrogen, which made it impossible to ensure complete consistency of the quenching time (the contact time between samples and quenching agents) between parallel samples. However, for the metabolites with RSDs>50%, there were only three metabolites in N2-A method, which was the fewest among all 12 methods.