Development of an automatic linear calibration method for high-resolution single-particle mass spectrometry: improved chemical species identification for atmospheric aerosols

The mass resolution of laser desorption ionization (LDI) single-particle aerosol mass spectrometry (SPAMS) is usually low (∼ 500), which has been greatly improved by the recent development of the delayed ion extraction technique. However, due to large fluctuations among LDI processes during each laser shot, accurate calibration of the mass-tocharge ratio for high-resolution SPAMS (HR-SPAMS) spectra is challenging. Here we developed an automatic linear calibration method to improve the accuracy of mass-tocharge (m/z) measurement for single atmospheric aerosol particles. Laboratory-generated sea spray aerosol and atmospheric ambient aerosol were tested. After the calibration, the fluctuation ranges of the reference ions’ (e.g., Pb and SO+4 ) m/z reaches ±0.018 for sea spray aerosol and ±0.024 for ambient aerosol in average mass spectra. With such m/z accuracy, the HR-SPAMS spectra of sea spray aerosol can easily identify elemental compositions of organic peaks, such as Cx , CxHy and CxHyOz. While the chemical compositions of ambient aerosols are more complicated, CxHy , CxHyOz and CNO peaks can also be identified based on their accurate mass. With the improved resolution, the time series of peaks with small m/z differences can be separated and measured. In addition, it is also found that applying high-resolution data with enhanced mass calibration can significantly affect particle classification (identification) using the ART-2a algorithm, which classify particles based on similarities among singleparticle mass spectra.


HR-SPAMS was made and we proposed the principle of this calibration method can be applied into some similar instruments, such as single particle mode Aerosol Mass Spectrometer (AMS)."
Line 273-280: "More importantly, 139[C2H3O5S -] (the theoretical m/z value : -138.97) can be clearly distinguished from other possible assignments, such as 139 [C11H7 -] with the m/z value  with the m/z value  with theoretical m/z value of -152.986 can be distinguished from other possible assignments, such as 153 [C12H9 -] with m/z value  with m/z value of -152.883. These two important organic ion peaks have been suggested to be the characteristic ion peaks for the organosulfates in secondary organic aerosols (Surratt et al., 2010;Surratt et al., 2007)." Line 319-347 : "The ART-2a classification of the HR-SPAMS results (Fig.S6) show that the signal at 23 [Na + ] + ], which are known to be harmful to human health (Das et al., 2018;Peng et al., 2020). Furthermore, particles of Type 2HR containing abundant secondary inorganic components like ,  and , which originated from the aerosol aging processes (Dall'Osto and Harrison, 2012;Ma et al., 2016). In contrast, these two first particle types were lumped together into Type 1LR in the LR-SPAMS classification results (Fig.S5). Due to the merge of these two particle types, [Pb+]   , suggesting that these three types were from biomass burning or residential cooking burning. [K + ] is also another feature of this type particle emission (Bi et al., 2011;Hudson et al., 2004). There were obvious relative ion intensity differences at 26 [CN -], 42 ] and 62[NO3 -] among these types, which implied that these three particle types might be from different burning sources or experienced different levels of aging (Luo et al., 2020). While these three particle types were lumped together as Type 2LR. This critical information which could be potentially used to distinguish particle sources and aging processes was lost. Additionally, Type 7HR can be assigned as ECOC type, based on its strong signals at [Cx + ], [CxHy + ] and [CxHyOz + ]. Particles of this type may come from the primary emission sources, and the emitted black carbon particles would also form this type particles after absorbing some low volatile organic compounds in the atmosphere (Sodeman et al., 2005). 97  can be observed to have a weaker signal than 62[NO3 -] and 46 , which implied that the secondary reaction of SO2 as the precursor of HSO4was not significant in the particle surface for ECOC type particles in this study (SULLIVAN and PRATHER, 2007). In contrast, the classification results of the LR-SPAMS were not so clear and generated less particle types. Given HR-SPAMS spectra have much more detailed chemical information about particles, we would propose that the ART-2a classification of HR-SPAMS might be more accurate." ambient aerosols and 8 a.u in the sea spray aerosols. And a particle was discarded from the spectrum database if it did not have enough reference ions ( the minimum number of reference ions was set to be 5) in either positive or negative mass spectrum. And now we have made a clear reasoning in the section 3.2 and also stated in the conclusion. Changes in manuscript: Line 210-213: "And 4,624 sea spray particles and 1,409 ambient particles were successfully calibrated. As some fraction of particles had been filtered because their mass spectrum did not have 5 or more reference peaks to conduct the calibrations. To mitigate this problem, we proposed some adjustments in the next section." Line 370-373: "There is a deficiency of this HR-SPAMS calibration method, which has been showed that some fraction of particles cannot be calibrated due to the presence of weak signals of the marker ions. It can be mitigated by applying some additional marker ions." 4. What is the matrix size produced by the HR-SPAMS? Is there a limit for the ART-2a to classify the matrix of particle mass spectra? Such information should be included. As there is 25,000 bins in either positive or negative single mass spectrum and we handled about 1,400 particles in this study. The matrix size is about 25,000×2×1400=7e7. There was no absolute limit for ART-2a as long as the database was not extremely large, like the experiment data lasting for several months. And we have made the following additional statement to include this necessary information in our revised manuscript. Changes in manuscript: Line 305-306 : "The previous ambient aerosol SPAMS dataset (1,400 particles) was used for the matrix size of the ART-2a is around 7×10^7."

Section 4.4: What kind of new information is provided when new matrix is included in the classification? I think it would be interesting if there is new information after the classification of newly calibrated mass spectra.
It is the same comment as the referee's major comments. We agree with the referee's comment that the discussions of our new classification results are not quite sufficient in our original manuscripts. And now we have made a more comprehensive discussions, as we responses to the referee's major comments above. Conclusion: It would be better to include some atmospheric implications for the identification of additional peaks, in particular, organic peaks. Currently, the authors showed that more particle types can be obtained, but it might not be meaningful enough for the scientific community.
We agree with the referee's comments and here is the revision: Changes in manuscript: Line 358-370: "With this method, HR-SPAMS can also determine the time series of organic and inorganic peaks, whose m/z are very close to each other (e.g. 41K + (Manuel et al., 2006;Wenzel et al., 2003)." Line 79: "A higher mass resolution version of the SPAMS with better m/z accuracy is needed." Line 103-104: "In addition, the impact of using high resolution SPAMS data on particle classification by ART-2a algorithm was assessed." on manuscript "Development of an Automatic Linear Calibration Method for High Resolution Single Particle Mass Spectrometry: Improved Chemical Species Identification for Atmospheric Aerosols" We thank Referee #2 for the comments and suggestions. We have addressed every comment and made significant changes to improve the paper. Again, the referee's comments are greatly appreciated.

Referee Comments in black bold.
Authors' Response in blue. Changes in manuscript in Red italic.

Major comments
This study reports the development of an automatic linear calibration method for analyzing mass spectral data acquired with single particle mass spectrometers with mass resolution of ∼ 2000. The paper also shows the successful application of this method to analyzing lab generated sea spray particles and some ambient aerosols. This work is important given the broad application of single particle mass spectrometry in atmospheric studies and aerosol research and improvement of the chemical resolution of this technique is important. The scope of the work fits well within AMT and the manuscript is generally well written. I recommend acceptance for publication after following comments are addressed.

Response:
We are grateful for the comments given by the referee and hope our new automatic linear calibration method can be applied into more versions of aerosol mass spectrometry for atmospheric studies. Also we have addressed all the comments in the following paragraphs.

Specific comments： 1. Line 119,change to "dried by"
We have made correction to that. Changes in manuscript: Line 135-136: "The ambient particles were dried by a diffusional dryer before being sampled by the HR-SPAMS." 2. Fig. S2, the caption for this figure needs to be rewritten to better present the information content.
We have rewritten the caption of the Fig.S2 to better present the information content. Changes in supplementary material: " Fig.S2 partial enlarged detail in the sing particle mass spectra for the

The description on
Step 3 given in the paragraph on pages 6 and 7 is a bit hard to follow. How exactly is the calibration conducted? Are the measured m/z bins determined from the "traditional method" mentioned in Step 0? What exactly is the "traditional method" involved? How many bins are selected for each m/z?
(1)The detailed statement of the Step 3 is that: First, we will get the measured m/z bins determined from the "traditional method", just as described in the Step 0. And these bins were not sufficiently accurate. Then, we picked up some reference ions to make a linear regression between the two set of variables (measured vs. theoretic reference ion m/z bin values). The obtained calibration parameters (a slope and an intersect) from linear regression were used to calibrate every bin value for this mass spectrum. However, the m/z bin values are fixed numbers (they are not continuous). So we had to assign the calibrated m/z value to its closest m/z bin value. Finally, the correct aerosol mass spectra can be acquired. We had made some revisions to make it easier to read in the Step3; (2)The measured m/z bin values were determined from the "traditional method" mentioned in the Step 0; (3)The "traditional method" was more like a coarsely-calibrated method, which usually selected a few particles with distinct ion patterns. Fig3 and Table S1&S2 had reported there were around five of larger bin numbers offset for the ionized species. What's more important is that every particle mass spectra were different from each other and needed its own calibration parameters. So the "traditional method" is not enough for the SPAMS data processing.
(4)One specific bin was selected for each m/z as shown in the  [197][198][199]: "Then we used these parameters to make the calibration for every bin value in this mass spectra. Finally, the m/z of the whole spectrum had been corrected."

Give units for "measurement m/z" and "theoretical m/z" on the axis labels in all the Figures presented in this paper.
The unit is Dalton. We have updated all m/z axis labels in in the revised paper. Figure 3, the symbols are hard to differentiate, consider to revise. The spectra a and b look identical, are they really represent sea spray aerosol and ambient aerosol respectively?

6.
Thanks for pointing this out. It was a mistake and has been corrected now. We have also separated the figure into 2 panels to make them easier to differentiate. Changes in manuscript: Line 548-553: "a. b.

Figure 3. Probability distributions of the marker peak locations before and after Automatic
Linear Calibration (AL-Cal) for (a) sea spray aerosol and (b) ambient aerosol" 7. Figure 4, what is "error limits"? How was is calculated?
The "error limit" is the delta bin number which is concluded from the Table S1&S2 and Fig.3, representing the accepted error range (around 3 bin numbers) after the calibration for a m/z bin value. Changes in manuscript: Line 252-254: " Figure 4 reports the average positive and negative mass spectra for the laboratory generated sea spray aerosols and the error limits mean the concluded accepted error range." 7. Line 250, change to "Ca2+" Actually the ion fragment generated from the SPAMS can only carry one charge, so the Ca + is reasonable.

Line 266, define "LR-SPAMS"?
We have give a definition for that. Changes in manuscript: Line 295-297: "In contrast, it is impossible for a low resolution (LR)-SPAMS to provide such detailed time variation measurement of these peaks." June 3 rd , 2020 are more complicated, CxHy, CxHyOz and CNO peaks can also be identified based on their accurate mass. With the improved resolution, the time series of peaks with small m/z 48 differences can be separated and measured. In addition, it is also found that applying high 49 resolution data with enhanced mass calibration can significantly affect particle 50 classification (identification) using the ART-2a algorism, which classify particles based on 51 similarities among single particle mass spectra.

Introduction
Atmospheric aerosols can significantly impact radiative forcing, cloud formation and 57 human health (Ackerman et al., 2004;Zhang and Kin-Fai, 2012). They originate from 58 various sources and undergo many atmospheric aging processes, resulting in an extremely 59 complicated mixture of particles with a large range of sizes and chemical compositions.

60
This mixture is usually referred as "mixing state". Measurement of aerosol mixing state  in the range of ±10 ns and the ion start position could be varied in the range of ±150μm.

88
As a result, substantial peak jittering is observed when switching between mass spectra of 89 each individual particle. This peak jittering leads to a fact that isotopic pattern identification In this study, we report a calibration method for single particle high resolution mass spectra 99 data. Based on the assumption that the sea spray aerosol has relatively simple chemical 100 composition while the ambient aerosol has more complex chemical composition, the 101 performance of the calibration method had been evaluated in detail for these two aerosol 102 systems with different complexity. In addition, the impact of using high resolution SPAMS

139
To improve the accuracy of m/z for HR-SPAMS spectra, an automatic linear calibration 140 method has been developed. Noticeably, due to the technical limitation of data acquisition, 141 the whole HR-SPAMS spectrum is not continuous but divided by a large number of m/z 142 bins, which are described in Fig.S2(a partial enlarged detail in the single particle mass 143 spectra) and can be viewed as the probability density histogram of the m/z. Here we denote 144 "m/z bin value" as the median m/z value of each bin.

146
The linear calibration method is described as the following steps: Step 0: The SPAMS data was coarsely-calibrated by the traditional method, which usually 149 selected a few particles with distinct ion patterns, i.e. the molecular composition of some 150 distinct peaks in the mass spectra can be easily identified. Then, the time of flight of these 151 peaks and the true m/z of the corresponding ions were used to calculate a set of calibration 152 parameters for positive and negative spectra. The parameters were finally applied to the 153 whole mass spectra dataset, and coarsely-calibrated was completed.

155
Step 1: a pool of ion peaks in the single particle mass spectra were selected as the potential 156 m/z calibration reference ions. The selection criteria are (1)  Step 2: a set of reference ions was chosen from the potential reference ion pool for each 185 spectrum. The selection was based on the absolute ion intensity of the reference ions in this 186 spectrum. They must be greater than a threshold, e.g. we set 15 a.u. for ambient aerosol 187 and 8 a.u. for sea spray aerosol, respectively. A particle was discarded from the spectra 188 database if it did not have enough reference ions (the minimum number of reference ions 189 was set to be 5) in either positive or negative mass spectrum.

191
Step 3: the reference ions were used to calibrate m/z for mass spectra of each particle. As

208
In this study, a total of 5,130 sea spray aerosol particles and 5,007 ambient aerosol particles were analyzed. And 4,624 sea spray particles and 1,409 ambient particles were successfully 210 calibrated. As some fraction of particles had been filtered because their mass spectrum did 211 not have 5 or more reference peaks to conduct the calibrations. To mitigate this problem, 212 we proposed some adjustments in the next section. Figure 2 shows that the calibration 213 curves for a random selected sea spray aerosol particle and ambient aerosol particle. The

Automatic linear calibration method with a larger reference ion pool 224
It is important to note that a large number of ambient particles were filtered because their 225 spectra did not have 5 or more reference peaks to conduct calibrations. We needed to identify them using additional information other than their integer m/z values.

232
The specific reference ions was determined by their coarsely-calibrated m/z.  Figure 4 reports the average positive and negative mass spectra for the 252 laboratory generated sea spray aerosols and the error limits mean the concluded accepted 253 error range. Similar to the low-resolution sea spray aerosol mass spectra, they contain 254 major peaks of Na + , Mg + , K + , Na2Cl + , CN -, Cl -, CNO -, NaCl -, NaCl2and MgCl3 -, as well 255 as many smaller peaks, such as Ca + , SiO2 -/SiO3 -, and KCl2 -. With the improved m/z 256 measurement, many peaks, which cannot be determined by integer resolution mass spectra, 257 now can be clearly identified (Table 2). For example, the ion with m/z at 27.0267 is C2H3 +

258
rather than Al + . The ion with m/z at 76.9336 is CaCl + rather than C6H5 + . And some sulfur 259 containing organic ions, such as CS + , can be determined. Surprisingly, we can identify the 260 presence of HCO2and CaCO3 -, demonstrating that carbon hydrates are contained in sea 261 spray aerosols. many organic ions, such as Cx, CxHy, and CxHyOz can be directly identified (Table 3). Also,

269
we can separate the organic and inorganic species more directly with the high mass   Figure 6 shows that HR-SPAMS was able to separately 294 measure the time series of these two peaks with a small m/z difference. In contrast, it is 295 impossible for a low resolution (LR)-SPAMS to provide such detailed time variation 296 measurement of these peaks. The ART-2a classification of the HR-SPAMS results (Fig.S6)