The authors significantly improved the description of the LC-MS measurements and the corresponding discussion. However, there remain some major weaknesses and wrong conclusions. To some degree, these are based on poor method development for the analysis approach (and maybe inexperience) with the Q Exactive mass spectrometer.
a) The authors did a good job in comparing the results from HILIC and RPLC, however, I was wondering why only the peak areas of 2-MT-OS were correlated. What about the 2-MG-OS?
b) I disagree with the conclusion that “there is no evidence of ion source induced artifacts” (L527). As the authors acknowledge, there is still some co-elution of sulfate and nitrate ions with detected OS species. Therefore, there is still a chance of adduct formation between inorganic ions and organic compounds. The “test for artefact formation” conducted by the authors is not convincing, as it is just proving that sulfate and 2-methylglyceric acid / 2-methyltetrol do not produce artifacts. This cannot be generalized for all organic compounds! (By the way, I totally agree with the authors’ conclusion that all previous direct infusion studies suffer from in-source adduct formation. Therefore, I’m actually very happy to see that people are now more and more using chromatographic separation before ionization. Nonetheless, if the separation is not done appropriately, we will not get away from ion source artifacts.)
c) The selected settings of the Q Exactive mass spectrometer are really far from appropriate for a quantitative analysis approach. With the settings applied here (i.e., full scan at R=70,000 combined with ddMS2 Top 10), the authors obtain ~1 full scan spectrum per second (i.e, full scan + 10 x ddMS2 = 256 ms + 10 x 64 ms = 896 ms). As a rule of thumb, for accurate quantification, about 12 data points are needed to get a good peak quality in the extracted ion chromatograms. Therefore, the chromatographic peak width needed here would have been roughly 11–12 seconds, which is a quite broad peak in typical UHPLC applications and only reached with quite concentrated samples. In particular, since all the OS peaks elute quite early, I would be surprised to see that the authors obtained more than ~5–6 data points for each of the peaks here. In principle, it is possible to increase data quality by measuring replicates. However, the authors decided to measure each sample only once. Therefore, I think the concentrations reported here are connected to very large uncertainties (which could have been avoided easily).
d) Regarding the uncertainties mentioned above, it is still unclear to me how the authors estimate an uncertainty of 60% for the measured OS concentrations.
Minor / Technical comments:
1) There are several mentions of “corplot” and “PollutionRose plot” in the manuscript, which might be disturbing to readers less familiar with R. I would recommend replacing these R-specific names with more general terms (e.g., corplot = correlation plot).
2) Figure 5: Some of the label names capitalized and some not.
3) L901: Ultra-performance liquid chromatography (UPLC) is actually specific for components from Waters Corporation. The more general term is ultra-high performance liquid chromatography (UHPLC).
4) Figure S3: I suppose the x-axis label should read 2-MG-OS.