Data Processing of Product Ion Spectra: Redundancy of Product Ion Spectra of Small Molecules in Data-Dependent Acquisition Dataset

Abstract

Non-targeted metabolome analysis studies comprehensively acquire product ion spectra from the observed ions by the data-dependent acquisition (DDA) mode of tandem mass spectrometry (MS). A DDA dataset redundantly contains closely similar product ion spectra of metabolites commonly existing among the biological samples analyzed in a metabolome study. Moreover, a single DDA data file often includes two or more closely similar raw spectra obtained from an identical precursor ion. The redundancy of product ion spectra has been used to generate an averaged product ion spectrum from a set of similar product ion spectra recorded in a DDA dataset. The spectral averaging improved the accuracy of m/z values and signal-to-noise levels of product ion spectra. However, the origins of redundancy, variations among datasets, and these effects on the spectral averaging procedure needed to be better characterized. This study investigated the nature of the redundancy by comparing the averaging results of eight DDA datasets of non-targeted metabolomics studies. The comparison revealed a significant variation in redundancy among datasets. The DDA datasets obtained by the quadrupole (Q)-Orbitrap-MS datasets had more significant intrafile redundancy than that of the Q-time-of-flight-MS. For evaluating the similarity score between two product ion spectra, the optimal threshold level of the cosine-product method was approximately 0.8–0.9. Moreover, contamination of biological samples such as plasticizers was another origin of spectral redundancy. The results will be the basis for further development of methods for processing of product ion spectra data.

INTRODUCTION

High-resolution mass spectrometry (MS) has been used to obtain an accurate mass-to-charge ratio (m/z) value of observed ions in non-targeted metabolome analysis studies.^1,2) Moreover, product ion spectra have been comprehensively acquired from the observed ions by the data-dependent acquisition (DDA) or the information-dependent acquisition (IDA) modes of the quadrupole-time-of-flight (Q-TOF)-MS and Q-Kingdon trap or Orbitrap-MS.^3–8) In the DDA mode, a full-scan mass spectrum was obtained without fragmentation (MS1 scan). The product ion spectra were automatically obtained from a set of the most abundant ions as the precursor ions (MS2 scan).⁹⁾

The accurate m/z values and product ion spectra data provide a basis for a structural elucidation of metabolites.¹⁰⁾ However, errors in m/z values in product ion spectra inevitably restrict the narrowing of candidate molecular formulas. Moreover, detailed assignment of product ions was also confused from the presence of noise peaks in product ion spectra.^11–13) A previous study reported that the accuracy of m/z values and signal-to-noise levels of product ion spectra could be improved by generating an averaged product ion spectrum from a set of similar product ion spectra recorded in a DDA dataset.¹⁴⁾ Furthermore, it was demonstrated that the quality improvement contributed to reducing the number of candidate molecular formulas and elucidating a more detailed metabolite structure.¹⁴⁾

The averaged product ion spectra were generated using the redundancy of product ion spectra data in a DDA dataset. In this report, “a raw spectrum” and “an averaged spectrum” indicate a product ion spectrum recorded in a DDA dataset and an averaged product ion spectrum generated from a set of similar raw spectra. A single DDA data file often contains two or more closely similar raw spectra obtained from an identical precursor ion, depending on the settings of the DDA mode, such as the number of maximum candidate ions, the level of intensity threshold exceed, the time period to exclude former candidate ions, and the dwell time for one MS2 scan. In this study, the redundancy of raw spectra in one DDA data file is called the intrafile redundancy. Furthermore, typical metabolomics studies acquire a set of DDA data files from a series of biological samples with similar compositions of metabolites.^15–20) For example, a non-targeted metabolomics dataset consists of 248 DDA data files from different New York firefighters’ serum samples.²¹⁾ Thus, the dataset contains the interfile redundancy of product ion spectra of metabolites commonly existing among the biological samples.

In the averaging method tested in the previous study, two raw spectra data are considered “similar” when three indexes are within threshold levels.¹⁴⁾ The indexes are the retention time gap (ΔRT), the m/z gap of precursor ions (Δm/z), and the cosine product score of two product ion spectra (S). The previous study arbitrarily employed ΔRT_thres ≤1.0 min, Δm/z_thres ≤0.01, and S_thres ≥0.9 as the threshold levels for averaging raw spectra in a Q-TOF-MS dataset acquired in a yeast lipidomics study.^14,22) A similarity network was constructed after the determination of similarities among all raw spectra in a DDA dataset. Clusters or complete graphs of similar raw spectra were extracted from the similarity network. Averaged spectra were created by integrating all raw spectra in one cluster.¹⁴⁾ Here, the symbol n indicates the number of raw spectra used to produce one averaged spectrum.

As mentioned above, the averaging method depends on the redundancy in a DDA dataset. However, origins, degrees of redundancy, and variations among datasets were not well characterized. Moreover, an optimal level of the most critical threshold, S_thres, remains unclear. The aims of this study are an investigation of the nature of the redundancy in a DDA dataset and a determination of an optimal S_thres level. For this purpose, spectral averaging was performed for eight DDA datasets of non-targeted metabolomics studies. The analysis revealed that the nature of redundancy partly depended on the type of mass analyzers and that an optimal level of S_thres was 0.8–0.9. It was also demonstrated that the DDA datasets contain many similar raw spectra obtained from broadly eluting compounds. One of the compounds was deduced to be a plasticizer, di(2-ethylhexyl)phthalate (DEHP), indicating that contamination of biological samples was another origin of spectral redundancy.

EXPERIMENTAL PROCEDURES

Datasets

Table 1 shows eight datasets (Set 1–Set 8) used in this study.^21–25) The seven datasets (Set 1–Set 7) were downloaded from the Metabolomics Workbench²⁶⁾ and MetaboLights repositories.^27–29) The datasets were selected from search results using “DDA” and “IDA” as query keywords because they contain more than 30 DDA data files. For the Set 1 dataset, 46 data files repeatedly obtained from the CHEAR Pool (CHEAR Reference Plasma) and Study Pool (the pool of studied samples) were used out of the 248 data files because of limited computing power for the clustering.²¹⁾ The Set 8 dataset is a yeast lipidomics dataset obtained in our previous study, which will be open to the public elsewhere.²²⁾

Table 1. Datasets used in this study.

	Set 1	Set 2	Set 3	Set 4	Set 5	Set 6	Set 7	Set 8
Repository	Metabolomics Workbench	Metabolomics Workbench	Metabolomics Workbench	Metabolomics Workbench	Metabolomics Workbench	MetaboLights	MetaboLights	nd
ID	ST001171	ST002338	ST002338	ST001264	ST001264	MTBLS417	MTBLS417	nd
Title	Metabolomics of World Trade Center Exposed New York City Firefighters21)	Interplay between cruciferous vegetables and the gut microbiome: a multi-omic approach23)	Interplay between cruciferous vegetables and the gut microbiome: a multi-omic approach23)	Antibiotics and dietary minerals lipidomics24)	Antibiotics and dietary minerals lipidomics24)	Customized consensus spectral library building for untargeted quantitative metabolomics analysis with data independent acquisition MS and MetaboDIA workflow25)	Customized consensus spectral library building for untargeted quantitative metabolomics analysis with data independent acquisition MS and MetaboDIA workflow25)	Lipidome variation of industrial Saccharomyces cerevisiae strains analyzed by LC-Q-TOF/MS-based untargeted lipidomics22)
Author	Anna Nolan, Laboratory at NYU/Bellevue	Bouranis John, Oregon State University	Bouranis John, Oregon State University	Jacob Folz, University of California, Davis	Jacob Folz, University of California, Davis	Zhou Lei, Hyungwon Choi, National University of Singapore	Zhou Lei, Hyungwon Choi, National University of Singapore	Shuka Komori, Osaka University
Sample	Human, serums	Human, cultured cells	Human, cultured cells	Rattus norvegicus, plasma	Rattus norvegicus, plasma	Human, serums	Human, serums	Budding yeast cells, aerobic culture
Mass spectrometer	Q Exactive HF-X Orbitrap, Thermo Scientific	TripleTOF 5600, AB Sciex	TripleTOF 5600, AB Sciex	Q Exactive HF hybrid Orbitrap, Thermo Scientific	Q Exactive HF hybrid Orbitrap, Thermo Scientific	TripleTOF 5600+, AB Sciex	TripleTOF 5600+, AB Sciex	LCMS-9030, Shimadzu
HPLC	Reverse phase ODS	Reverse phase ODS	Reverse phase ODS	Reverse phase ODS	Reverse phase ODS	Reverse phase ODS	Reverse phase ODS	Reverse phase ODS
Ionization	ESI	ESI	ESI	ESI	ESI	ESI	ESI	ESI
Polarity	Positive	Positive	Negative	Negative	Positive	Positive	Negative	Positive
Fragmentation	CID	CID	CID	CID	CID	CID	CID	CID
Number of data files	248 (46)*	40	40	31	31	60	60	45

*Whereas 248 data files were available in the dataset, 46 data files of the CP and SP series were used in this study.

CP, CHEAR Pool; MS, mass spectrometry; nd, no data; NYU, New York University; Q-TOF, quadrupole-time-of-flight; SP, Study Pool.

Construction of averaged product ion spectra

The spectra averaging was performed using the in-house Python3 scripts developed in our previous study with modification.¹⁴⁾ All data files were converted into Hive format and loaded using the HiveSDK version 0.12 beta available from Reifycs (https://ja.reifycs.com/products/hive/). The averaging tasks were executed for each unit mass (m/z_unit) from 100 to 1,000 as follows:

1. 1. To extract precursor ions in one unit mass, m/z_unit, product ion spectra derived from precursor ions within the range from (m/z_unit−0.4)×1.0007 to (m/z_unit+0.6)×1.0007 were collected from all DDA data files in a dataset. This formula is arbitrary based on empirical rules.
2. 2. A similarity graph was created among the collected data based on the similarity of two product ion spectra. A pair of two product ion spectra data was considered to be similar when the retention time gap (ΔRT) and the m/z gap of precursor ions (Δm/z) were less than the threshold levels, ΔRT_thres and Δm/z_thres, respectively, and when the cosine product score of two product ion spectra (S) was greater than S_thres. The level of ΔRT_thres was arbitrarily set at 1.5 min. Levels of Δm/z_thres were set at 10 and 50 ppm for the Q-Orbitrap and Q-TOF-MS datasets, respectively. The Δm/z_thres levels were deduced from a variance between measured and theoretical m/z values of precursor ions. The cosine (dot) product is a method to evaluate the similarity between two mass spectra, whose score ranges from 0 (no similarity) to 1.0 (identical).³⁰⁾ The S_thres level ranging from 0.6 to 0.95 was used in this study.
3. 3. Clusters, including more than 10 product ion spectra, were extracted by finding complete graphs (cliques) in the created similarity graph using the NetworkX package (https://networkx.org/)³¹⁾ and an in-house script.
4. 4. All product ions were collected from all product ion spectra in a cluster to generate an average mass spectrum. The total number of product ion spectra in this cluster was denoted as n.
5. 5. A similarity graph of all product ions was created by considering that two product ions with Δm/z <20 and 100 ppm were similar for the Q-Orbitrap and Q-TOF-MS datasets.
6. 6. If the total number of product ions (k) in a clique was k >0.7×n, this clique was employed to create an averaged mass spectrum.
7. 7. The median m/z and relative intensity values of the product ions were used for the averaged mass spectra.
8. 8. The m/z value of the precursor ion of the averaged mass spectrum was determined as the median m/z value of precursor ions of all raw spectra.

RESULTS

Spectra averaging using public DDA datasets

This study used seven DDA datasets downloaded from the metabolomics data repositories (Table 1).^27–29) In addition, a yeast lipidomics dataset obtained in our previous study was also employed.²²⁾ The averaging tasks were performed for all datasets using the thresholds described in the Experimental procedure. The graphical summary of the averaging workflow is available in my previous report.¹⁴⁾ The threshold levels at S_thres=0.8 and ΔRT_thres=1.5 were commonly used in this section. Moreover, the minimal number of n was set at 10 because the previous study showed a good balance between the number of raw spectra and the degree of quality improvement.¹⁴⁾

For example, the Set 1 dataset consisted of 46 data files acquired from human serum samples by the Q-Orbitrap-MS (Table 1), whose total number of raw spectra was 606,149 (Table 2). Among them, 562,440 raw spectra were utilized to generate 8,104 averaged spectra for the spectral averaging (Table 2). Here, Total, Utilized, and Averaged represent the total number of raw spectra in a dataset, the number of raw spectra utilized to generate averaged spectra, and the number of averaged spectra, respectively. Moreover, to evaluate a degree of redundancy, the utilization ratio (UR), the mean number of raw spectra used to generate one averaged spectrum (n_mean), and the summary ratio (SR) were defined as follows:

  

$$UR=Utilized/Total$$

  

$$n_{\text{mean}}=Utilized/Averaged$$

  

$$SR=Averaged/Total$$

Table 2. Results of the spectral averaging at S_thres=0.8.

	Set 1	Set 2	Set 3	Set 4	Set 5	Set 6	Set 7	Set 8
Type	Q-Orbitrap	Q-TOF	Q-TOF	Q-Orbitrap	Q-Orbitrap	Q-TOF	Q-TOF	Q-TOF
Number of data files	46	40	40	31	31	60	60	45
The number of raw spectra in dataset, Total	606,149	213,879	188,954	212,316	215,967	548,647	512,375	504,072
The number of raw spectra utilized for averaging, Utilized*	562,440	102,414	71,642	178,747	211,528	119,680	125,827	121,834
The number of averaged spectra, Average*	8,104	2,869	2,158	683	831	3,808	3,857	2,400
Used ratio (UR)*	0.928	0.479	0.379	0.842	0.979	0.218	0.246	0.242
Mean number of raw spectra used to generate one averaged spectra (nmean)*	69.4	35.7	33.2	261.7	254.5	31.4	32.6	50.8
Summary ratio (SR)*	0.0134	0.0134	0.0114	0.0032	0.0038	0.0069	0.0075	0.0048
Mean number of ion signals in one averaged spectrum*	27.1	41.4	31.4	5.4	19.8	12.53	7.3	14.7
Ratio of signals observed more than 90% of raw spectra*	0.73	0.68	0.66	0.85	0.84	0.62	0.65	0.72
Summary ratio (SR) when ΔRTthres=∞**	0.0116	0.0127	0.0105	0.0022	0.0030	0.0060	0.0065	0.0046

*Data processing was performed using ΔRT_thres=1.5.

**Data processing was performed using ΔRT_thres=∞.

Q-TOF, quadrupole-time-of-flight.

For the case of Set 1, the numbers of Total, Utilized, and Averaged were 606,148, 562,440, and 8,104, from which the UR, n_mean, and SR levels were determined to be 0.928, 69.4, and 0.0134, respectively (Table 2). These results indicated that more than 92% of raw spectra were employed to generate averaged spectra and that 69.4 raw spectra were integrated into one averaged spectrum on average. Due to the redundancy, the number of Averaged was only 1.34% of the Total. The results indicated that interfile and intrafile redundancy contributed to the overall redundancy because the n_mean level, 69.4, was approximately 1.5 times larger than the number of data files, 46 (Table 2).

Large variation in spectra redundancy among DDA datasets

The results shown in Table 2 revealed a significant variation in spectra redundancy among DDA datasets. The UR, n_mean, and SR level ranges were 0.218–0.979, 31.4–261.7, and 0.0022–0.0127, respectively. For example, the UR and n_mean levels of Set 6 acquired by the Q-TOF-MS were the smallest among the eight datasets. The UR level at 0.218 means that more than 78% of raw spectra were irreproducible and did not contribute to generating averaged spectra. The n_mean level was 31.4, about half of the file number of the dataset, 60.

On the other hand, the largest level of UR was observed for Set 5 acquired by the Q-Orbitrap-MS. The UR level was 0.979, indicating that only 2.1% of raw spectra were useless. The n_mean value was 261.7, more than eight times larger than the file numbers of the datasets (Table 2). It indicates that similar raw spectra are acquired multiple times from one precursor ion during one analysis. Due to the large intrafile redundancy, the levels of Averaged and SR of Set 5 were 831 and 0.0038, respectively, which were the second smallest among the eight datasets (Table 2).

The comparative analysis also highlighted that there were different trends between the datasets acquired by the Q-TOF-MS and Q-Orbitrap-MS. For instance, the UR levels of the dataset acquired by the Q-TOF-MS were less than 0.479, while those of the Q-Orbitrap-MS were above 0.842 (Table 2). Moreover, the n_mean levels of the Q-TOF-MS datasets were smaller than those of the Q-Orbitrap-MS datasets (Table 2). These trends indicated that the Q-Orbitrap-MS datasets tended to have larger intrafile redundancy and smaller irreproducible raw spectra ratio than the Q-TOF-MS datasets.

The averaging tasks were performed iteratively for eight datasets using different S_thres levels, including 0.6, 0.7, 0.8, 0.9, and 0.95 (Fig. 1A–1C). For the case of Set 1 acquired by the Q-Orbitrap-MS, the SR levels increased in proportion to S_thres. The SR level at S_thres=0.95 was 0.0152, 22% larger than that of S_thres=0.6 (Fig. 1A). A similar trend was also seen in other Q-Orbitrap-MS datasets, including Set 4 and Set 5. On the other hand, the SR levels decreased as S_thres increased in the datasets acquired by the Q-TOF-MS (Fig. 1A). The UR and n_mean levels tended to decline in all datasets (Fig. 1B and 1C).

Fig. 1. Effect of threshold level of spectra similarity score (S_thres) on the results of spectral averaging. The spectral averaging tasks were performed for eight DDA datasets (Set 1–Set 8) at S_thres=0.6, 0.7, 0.8, 0.9, and 0.95. Here, Total, Utilized, and Averaged represent the number of raw spectra in a dataset, contributed raw spectra for the generation of averaged spectra, and averaged spectra, respectively. (A) Effect of S_thres on the summary ratio (SR=Averaged/Total). (B) Effect of S_thres on the utilization ratio (UR=Utilized/Total). (C) Effect of S_thres on the mean raw spectra used for one averaged spectrum (n_mean=Utilized/Averaged). (D) Effect of S_thres on the similarity network. Each node (dot) indicates raw product ion spectra. Each edge (line) indicates a similarity between two product ion spectra above the threshold level (S_thres). Two possible cases are shown as Case 1 and Case 2. The Averaged and Utilized values are also shown in the figure. DDA, data-dependent acquisition.

Figure 1D shows two possible relationships between the S_thres levels and the clustering results. One dot (node) indicates one raw spectrum. Each line (edge) represents a pair of similar raw spectra. One cluster is formed for the Case 1 when S_thres is 0.6. The cluster split into two when S_thres increased to 0.95 because there are six pairs of similar raw spectra with S levels more than 0.95. The number of Averaged elevates from 1 to 2, whereas the number of Utilized is constant at 6. This means that the levels of SR and UR are elevated and kept constant, respectively. On the other hand, the cluster disappears in Case 2 when S_thres increases to 0.95 (Fig. 1D). This is because the S levels of all pairs are less than 0.95. The numbers of Averaged and Utilized decreased to zero, resulting a decline in SR and UR levels.

The relationships suggest that the decrease in the SR, UR, and n_mean levels of the Q-TOF-MS datasets could be explained by the Case 2 relationship (Fig. 1A–1C). The Case 2 relationship also contributed to decline the UR and n_mean levels in the Q-Orbitrap-MS datasets (Fig. 1B–1C). Additionally, the cluster splitting by Case 1 explained the increase in the SR levels in the Q-Orbitrap-MS datasets (Fig. 1A).

Distinct nature of Q-TOF-MS and Q-Orbitrap-MS datasets

Figure 1D shows that the Case 1 relationship requires many pairs of raw spectra whose spectral similarity scores (S) are above 0.95. On the other hand, the Case 2 relationship occurs when all S are below 0.95. It means the Q-Orbitrap-MS datasets have some characteristics to elevate the S level. For this purpose, the natures of the eight DDA datasets were further compared.

First, the mean numbers of product ion signals in one averaged spectrum were investigated for the averaged spectra using S_thres=0.8. The cosine product method tends to provide larger scores when spectra contain only one product ion signal. The result showed that values of the mean numbers of product ion signals in one averaged spectrum ranged from 5.4 in Set 4 to 41.4 in Set 2 (Table 2). However, there was no relationship between the type of mass analyzers and these values (Table 2).

Next, the frequency of missing values in raw spectra was compared. Missing values are often observed in the product ion spectra.³²⁾ Thus, the averaging method used in this study employed product ion signals commonly observed in more than 70% of raw spectra to generate an averaged spectrum (See the Experimental section for a more detailed procedure.). Here, the frequency of missing values was investigated using a stricter condition. The ratio of product ion signals commonly observed in more than 90% of the raw spectra was determined to be 0.73 for the Set 1 dataset (Table 2). The ratio ranged from 0.73 to 0.85 for the Q-Orbitrap-MS datasets, while 0.62–0.72 for the Q-TOF-MS datasets. The results implied that the Q-Orbitrap-MS data tended to have a lower frequency of missing values in raw spectra than the data acquired by the Q-TOF-MS. The different frequency of missing values in raw spectra data could be one reason for the distinct nature between the Q-Orbitrap-MS and Q-TOF-MS datasets.

Effect of no threshold on retention time

During the ionization process, structurally similar compounds often provide an identical ion by the in-source fragmentation. For example, the proton adduct ion of kaempferol (a flavonoid) was commonly generated from kaempferol-3-O-rhamnoside and kaempferol-3-O-glucoside, which eluted at different retention times.³³⁾ Moreover, broad chromatogram peaks were often observed for some compounds. Here, a degree of redundancy caused by these compounds was investigated. The spectral averaging was performed at S_thres=0.8 without using the threshold of a retention time gap (or ΔRT_thres was set at ∞). For the case of Set 1, the SR level was determined to be 0.0116, which was 87% of the result at ΔRT_thres=1.5 min (Table 2). These results suggest that approximately 13% of averaged spectra generated using ΔRT_thres=1.5 min were derived from the in-source fragmentation or the broad peak metabolites. A similar reduction was commonly observed for other datasets (Table 2).

A compound responsible for the redundancy was investigated using structural elucidation of averaged spectra with large n. Among the averaged spectra of Set 8 generated at ΔRT_thres=∞, the largest n of the averaged spectra was 2,515. Unfortunately, no chemical structure was elucidated for the averaged spectra (precursor ion: m/z 279.1563. Product ion spectrum: m/z 98.9749: relative intensity (rel int) 999, m/z 149.0230: rel int 151). The number of raw spectra used for the second largest averaged spectrum was 2,482 (Fig. 2A). The extracted ion chromatogram of the precursor ion showed that this compound was eluted as a very broad peak in the DDA data file obtained from the wild-type Saccharomyces cerevisiae strains (BY4749, Fig. 2B).²²⁾ The structure of this compound was deduced to be DEHP from the molecular formula estimation using the m/z value of precursor ion as well as the MassBank search using the product ion spectrum (Fig. 2C).³⁴⁾ The compound seemed a ubiquitous contaminant in samples and solvents since DEHP is the most common phthalate in plasticizers. DEHP is also a target compound of human exposome studies.³⁵⁾ The averaged spectra of DEHP were also found from other datasets obtained at the positive ion mode, including Set 1, Set 2, and Set 6 (Figs. 2D–2F). Moreover, Set 8 also included an averaged spectrum deduced to be di-n-butyl phthalate (data not shown). These results indicated that plasticizers and other contaminant chemicals could be a source of redundancy in DDA datasets, although sources of the contamination remain unclear.

Fig. 2. Structural elucidation of most redundantly observed product ion spectra in the DDA datasets obtained at the positive ion mode. (A) Averaged product ion spectrum of second most redundant product ion spectra in Set 8. (B) Extracted ion chromatogram of the precursor ion (m/z 391.284) in the DDA data file obtained from the wild-type Saccharomyces cerevisiae strains (BY4749). (C) Product ion spectrum of DEHP (MassBank record ID: MSBNK-Antwerp_Univ-AN112902). (D–F) Closely similar averaged spectra generated from the Set 1 (C), Set 2 (E), and Set 6 (F) datasets. DDA, data-dependent acquisition; DEHP, di(2-ethylhexyl)phthalate; XIC, extracted-ion chromatogram.

DISCUSSION

This study investigated the degrees and origins of redundancy of product ion spectra in DDA data files. The DDA data obtained by Q-TOF-MS and Q-Orbitrap-MS were used because they were available from the metabolome data repository. The nature of DDA data obtained by other types of high-resolution MS was not investigated in this study. The results revealed a significant variation in intrafile redundancy among datasets (Table 2). It also revealed that the DDA datasets obtained by the Q-Orbitrap-MS datasets had larger intrafile redundancy than that of the Q-Orbitrap-MS (Table 2). One reason for this variation was the settings for the DDA mode. It seems that the DDA mode was operated under conditions allowing a certain degree of intrafile redundancy for the Q-Orbitrap-MS datasets, including Set 1, Set 4, and Set 5 (Table 2). It could be achieved by using a shorter time period to exclude former candidate ions for the settings of the DDA mode.

Moreover, this study showed that the frequency of missing values in a product ion spectrum of the Q-Orbitrap-MS datasets was lower than that of the Q-TOF-MS datasets (Table 2) in addition to other known distinct natures.^36,37) Due to the high intrafile redundancy and reproducibility of the Q-Orbitrap-MS datasets, most of the acquired data could be used for averaging. This explains the higher UR and n_mean levels of the Q-Orbitrap-MS datasets (Table 2). The larger intrafile redundancy is advantageous for spectral averaging since a larger n_mean level improves data accuracy through averaging. On the other hand, a too short time period to exclude former candidate ions would cause too much intrafile redundancy, which reduced the acquisition coverage, such as observed for Set 4 and Set 5 (Table 2).

For the case of the Q-TOF-MS datasets, higher frequency of missing values was observed. It suggested that the dwell time for MS2 scan might be too short to obtain reproducible data. Shorter dwell time was employed to increase the number of maximum candidate ions. However, the low UR levels of the Q-TOF-MS datasets (0.218–0.479) indicated that more than 50% of spectra were not reproducible and did not contribute intra- and interfile redundancy (Table 2). The result suggested that frequency of missing values could be reduced by using longer dwell time for MS2 scan.

The analysis of the DDA datasets also showed that the in-source fragmentation of similar compounds and a broad elution of an identical compound were other sources of redundancy. For the case of Set 1, approximately 13% of averaged spectra generated using ΔRT_thres=1.5 min were derived from the redundancy (Table 2). The database search deduced that one of the broad eluted compounds was a plasticizer, DEHP. Although the source of the DEHP remains unclear, the identification of contaminants in biological samples is important for quality improvement of DDA datasets to remove the unnecessary redundancy.

A suitable level of S_thres was estimated from the results of this study. The larger S_thres level is preferable for generating an averaged spectrum from a set of closely similar spectra. However, it is also unfavorable because the mean number of raw spectra for averaging n_mean is reduced. In the case of the Q-Orbitrap-MS data, a suitable S_thres level is between 0.8 and 0.9 due to a large drop in n_mean level when S_thres >0.9. Similarly, the S_thres level at 0.8 was optimal for the Q-TOF-MS data since the n_mean, SR, and UR levels decreased at S_thres >0.8.

These results indicated large variations in the origins and degrees of redundancy among datasets. It also demonstrated that an optimal level of the most critical threshold, S_thres, was approximately 0.8–0.9. Based on these findings, it is expected that the quality of product ion spectra data could be improved by the DDA data acquisition using an optimal setting and by the averaging task using appropriate thresholds.

ACKNOWLEDGEMENTS

We want to thank Profs. Eiichiro Fukusaki and Nobuyuki Okahashi at Osaka University; and Junko Iida, Jun Watanabe, and Atsuhiko Tohyama from the Shimadzu Corporation for their helpful comments and support. This work was supported by JST-NBDC Life Science Database Integration Project, grant number JPMJND2305.

Notes

Mass Spectrom (Tokyo) 2023; 12(1): A0138

REFERENCES

• 1)  T. Züllig, H. C. Köfeler. High resolution mass spectrometry in lipidomics. Mass Spectrom. Rev. 40: 162–176, 2021.
• 2)  B. Rochat. Quantitative/qualitative analysis using LC-HRMS: The fundamental step forward for clinical laboratories and clinical practice. Bioanalysis 4: 1709–1711, 2012.
• 3)  J. P. Koelmel, N. M. Kroeger, E. L. Gill, C. Z. Ulmer, J. A. Bowden, R. E. Patterson, R. A. Yost, T. J. Garrett. Expanding lipidome coverage using LC-MS/MS data-dependent acquisition with automated exclusion list generation. J. Am. Soc. Mass Spectrom. 28: 908–917, 2017.
• 4)  P. D. Hutchins, J. D. Russell, J. J. Coon. Accelerating lipidomic method development through in silico simulation. Anal. Chem. 91: 9698–9706, 2019.
• 5)  H. Lu, H. Chen, X. Tang, Q. Yang, H. Zhang, Y. Q. Chen, W. Chen. Ultra performance liquid chromatography–Q Exactive Orbitrap/mass spectrometry-based lipidomics reveals the influence of nitrogen sources on lipid biosynthesis of Mortierella alpina. J. Agric. Food Chem. 67: 10984–10993, 2019.
• 6)  H. Schoeny, E. Rampler, Y. El Abiead, F. Hildebrand, O. Zach, G. Hermann, G. Koellensperger. A combined flow injection/reversed-phase chromatography–high-resolution mass spectrometry workflow for accurate absolute lipid quantification with 13C internal standards. Analyst (Lond.) 146: 2591–2599, 2021.
• 7)  N. Danne-Rasche, S. Rubenzucker, R. Ahrends. Uncovering the complexity of the yeast lipidome by means of nLC/NSI-MS/MS. Anal. Chim. Acta 1140: 199–209, 2020.
• 8)  D. Schwudke, J. Oegema, L. Burton, E. Entchev, J. T. Hannich, C. S. Ejsing, T. Kurzchalia, A. Shevchenko. Lipid profiling by multiple precursor and neutral loss scanning driven by the data-dependent acquisition. Anal. Chem. 78: 585–595, 2006.
• 9)  W. M. Niessen. State-of-the-art in liquid chromatography–mass spectrometry. J. Chromatogr. A 856: 179–197, 1999.
• 10)  H. Tsugawa, T. Cajka, T. Kind, Y. Ma, B. Higgins, K. Ikeda, M. Kanazawa, J. VanderGheynst, O. Fiehn, M. Arita. MS-DIAL: Data-independent MS/MS deconvolution for comprehensive metabolome analysis. Nat. Methods 12: 523–526, 2015.
• 11)  T. Kind, O. Fiehn. Seven Golden Rules for heuristic filtering of molecular formulas obtained by accurate mass spectrometry. BMC Bioinformatics 8: 105, 2007.
• 12)  F. Matsuda, Y. Shinbo, A. Oikawa, M. Y. Hirai, O. Fiehn, S. Kanaya, K. Saito. Assessment of metabolome annotation quality: A method for evaluating the false discovery rate of elemental composition searches. PLoS One 4: e7490, 2009.
• 13)  F. Matsuda. Rethinking mass spectrometry-based small molecule identification strategies in metabolomics. Mass Spectrom. (Tokyo) 3: S0038, 2014.
• 14)  F. Matsuda, S. Komori, Y. Yamada, D. Hara, N. Okahashi. Data processing of product ion spectra: Quality improvement by averaging multiple similar spectra of small molecules. Mass Spectrom. (Tokyo) 11: A0106, 2022.
• 15)  V. Garikapati, C. Colasante, E. Baumgart-Vogt, B. Spengler. Sequential lipidomic, metabolomic, and proteomic analyses of serum, liver, and heart tissue specimens from peroxisomal biogenesis factor 11α knockout mice. Anal. Bioanal. Chem. 414: 2235–2250, 2022.
• 16)  L. Tao, J. Zhou, C. Yuan, L. Zhang, D. Li, D. Si, D. Xiu, L. Zhong. Metabolomics identifies serum and exosomes metabolite markers of pancreatic cancer. Metabolomics 15: 86, 2019.
• 17)  S. Yasuda, N. Okahashi, H. Tsugawa, Y. Ogata, K. Ikeda, W. Suda, H. Arai, M. Hattori, M. Arita. Elucidation of gut microbiota-associated lipids using LC-MS/MS and 16S rRNA sequence analyses. iScience 23: 101841, 2020.
• 18)  C. M. Henderson, M. Lozada-Contreras, V. Jiranek, M. L. Longo, D. E. Block. Ethanol production and maximum cell growth are highly correlated with membrane lipid composition during fermentation as determined by lipidomic analysis of 22 Saccharomyces cerevisiae strains. Appl. Environ. Microbiol. 79: 91–104, 2013.
• 19)  C. S. Ejsing, J. L. Sampaio, V. Surendranath, E. Duchoslav, K. Ekroos, R. W. Klemm, K. Simons, A. Shevchenko. Global analysis of the yeast lipidome by quantitative shotgun mass spectrometry. Proc. Natl. Acad. Sci. U.S.A. 106: 2136–2141, 2009.
• 20)  K. Tarasov, A. Stefanko, A. Casanovas, M. A. Surma, Z. Berzina, H. K. Hannibal-Bach, K. Ekroos, C. S. Ejsing. High-content screening of yeast mutant libraries by shotgun lipidomics. Mol. Biosyst. 10: 1364–1376, 2014.
• 21)  G. Crowley, J. Kim, S. Kwon, R. Lam, D. J. Prezant, M. Liu, A. Nolan. PEDF, a pleiotropic WTC-LI biomarker: Machine learning biomarker identification and validation. PLOS Comput. Biol. 17: e1009144, 2021.
• 22)  S. Komori, N. Okahashi, J. Iida, F. Matsuda. Lipidome variation of industrial Saccharomyces cerevisiae strains analyzed by LC-QTOF/MS-based untargeted lipidomics. J. Biosci. Bioeng. 135: 102–108, 2023.
• 23)  J. A. Bouranis, L. M. Beaver, D. Jiang, J. Choi, C. P. Wong, E. W. Davis, D. E. Williams, T. J. Sharpton, J. F. Stevens, E. Ho. Interplay between cruciferous vegetables and the gut microbiome: A multi-omic approach. Nutrients 15: 42, 2022.
• 24)  J. Folz, Y. T. Oh, I. Blaženović, J. Richey, O. Fiehn, J. H. Youn. Interaction of gut microbiota and high-sodium, low-potassium diet in altering plasma triglyceride profiles revealed by lipidomics analysis. Mol. Nutr. Food Res. 63: 1900752, 2019.
• 25)  G. Chen, S. Walmsley, G. C. M. Cheung, L. Chen, C. Y. Cheng, R. W. Beuerman, T. Y. Wong, L. Zhou, H. Choi. Customized consensus spectral library building for untargeted quantitative metabolomics analysis with data independent acquisition mass spectrometry and MetaboDIA workflow. Anal. Chem. 89: 4897–4906, 2017.
• 26)  M. Sud, E. Fahy, D. Cotter, K. Azam, I. Vadivelu, C. Burant, A. Edison, O. Fiehn, R. Higashi, K. S. Nair, S. Sumner, S. Subramaniam. Metabolomics Workbench: An international repository for metabolomics data and metadata, metabolite standards, protocols, tutorials and training, and analysis tools. Nucleic Acids Res. 44(D1): D463–D470, 2016.
• 27)  K. Haug, K. Cochrane, V. C. Nainala, M. Williams, J. Chang, K. V. Jayaseelan, C. O’Donovan. MetaboLights: A resource evolving in response to the needs of its scientific community. Nucleic Acids Res. 48(D1): D440–D444, 2020.
• 28)  N. S. Kale, K. Haug, P. Conesa, K. Jayseelan, P. Moreno, P. Rocca-Serra, V. C. Nainala, R. A. Spicer, M. Williams, X. Li, R. M. Salek, J. L. Griffin, C. Steinbeck. MetaboLights: An open-access database repository for metabolomics data. Curr. Protoc. Bioinformatics 53: 14.13.1–14.13.18, 2016.
• 29)  C. Steinbeck, P. Conesa, K. Haug, T. Mahendraker, M. Williams, E. Maguire, P. Rocca-Serra, S. A. Sansone, R. M. Salek, J. L. Griffin. MetaboLights: Towards a new COSMOS of metabolomics data management. Metabolomics 8: 757–760, 2012.
• 30)  S. E. Stein, D. R. Scott. Optimization and testing of mass-spectral library search algorithms for compound identification. J. Am. Soc. Mass Spectrom. 5: 859–866, 1994.
• 31)  A. A. Hagberg, D. A. Schult, P. J. Swart. Exploring network structure, dynamics, and function using NetworkX. in Proceedings of the 7th Python in Science Conference (SciPy2008) (Ed: G. Varoquaux, T. Vaught, J. Millman), Pasadena, CA, USA, 2008. pp. 11–15.
• 32)  F. Hamood, F. P. Bayer, M. Wilhelm, B. Kuster, M. The. SIMSI-Transfer: Software-assisted reduction of missing values in phosphoproteomic and proteomic isobaric labeling data using tandem mass spectrum clustering. Mol. Cell. Proteomics 21: 100238, 2022.
• 33)  F. Matsuda, R. Nakabayashi, Y. Sawada, M. Suzuki, M. Y. Hirai, S. Kanaya, K. Saito. Mass spectra-based framework for automated structural elucidation of metabolome data to explore phytochemical diversity. Front. Plant Sci. 2: 40, 2011.
• 34)  H. Horai, M. Arita, S. Kanaya, Y. Nihei, T. Ikeda, K. Suwa, Y. Ojima, K. Tanaka, S. Tanaka, K. Aoshima, Y. Oda, Y. Kakazu, M. Kusano, T. Tohge, F. Matsuda, Y. Sawada, M. Y. Hirai, H. Nakanishi, K. Ikeda, N. Akimoto, T. Maoka, H. Takahashi, T. Ara, N. Sakurai, H. Suzuki, D. Shibata, S. Neumann, T. Iida, K. Tanaka, K. Funatsu, F. Matsuura, T. Soga, R. Taguchi, K. Saito, T. Nishioka. MassBank: A public repository for sharing mass spectral data for life sciences. J. Mass Spectrom. 45: 703–714, 2010.
• 35)  D. K. Barupal, O. Fiehn. Generating the blood exposome database using a comprehensive text mining and database fusion approach. Environ. Health Perspect. 127: 097008, 2019.
• 36)  D. Szabó, G. Schlosser, K. Vékey, L. Drahos, Á. Révész. Collision energies on QTof and Orbitrap instruments: How to make proteomics measurements comparable? J. Mass Spectrom. 56: e4693, 2021.
• 37)  C. F. Wang, L. Li. Instrument-type effects on chemical isotope labeling LC-MS metabolome analysis: Quadrupole time-of-flight MS vs. Orbitrap MS. Anal. Chim. Acta 1226: 340255, 2022.