Resolving isobaric interferences in direct infusion tandem mass spectrometry

Abstract

Rationale

The co‐fragmentation of precursors in direct infusion (DI) tandem high‐resolution mass spectrometry (HRMS) can complicate the fragment spectra and consequently lead to false hits during compound identification.

Methods

The method herein described, termed IQAROS (incremental quadrupole acquisition to resolve overlapping spectra), modulates the intensities of precursors and fragments by stepwise movement of the quadrupole isolation window over the mass‐to‐charge (m/z) range of the precursors. The modulated signals are then deconvoluted by a linear regression model to reconstruct the fragment spectra with less interference. The hardware to demonstrate the use of IQAROS was an orbitrap with electrospray ionization (ESI) or secondary electrospray ionization (SESI), although the method can also be applied to other ionization techniques or mass analyzers.

Results

Assessing the performance of IQAROS with isobaric standards revealed that the reconstructed fragment spectra match with spectra acquired from the pure standards and that more compounds were correctly identified compared with the classical approach with the quadrupole centered at the m/z value of the precursor of interest. Moreover, the strength of IQAROS is exemplified by the identification of two isobaric biomarkers directly from a breath sample with SESI‐HRMS.

Conclusions

With IQAROS, cleaner fragment spectra of co‐fragmenting isobars during DI‐HRMS analysis can be obtained. IQAROS can easily be set up by the standard graphical user interface of the instrument. Therefore, it facilitates the characterization of features of interest in samples analyzed by DI‐HRMS, for example, in high‐throughput or real‐time metabolomics.

1. INTRODUCTION

Given its speed and sensitivity, direct infusion (DI) mass spectrometry (MS) is the ideal technique for high‐throughput analysis ¹ and for real‐time on‐line monitoring. ² For example, DI‐MS‐based metabolomics ³ or on‐line MS‐based breath analysis ⁴ are applications where speed and high time resolution are crucial. The excellent separation of ions according to their mass‐to‐charge ratio (m/z) with high‐resolution MS (HRMS) has proven to compensate at least partially for the missing chromatographic separation. ⁵ , ⁶ In a typical DI‐HRMS metabolomics workflow, samples are analyzed on an MS¹ level, followed by statistical analysis which yields MS¹ signals of interest, which in turn have to be annotated to chemical structures in the next step. ⁴ For sufficient confidence, this identification step requires characterization by tandem mass spectrometry (MS²). ⁷

While molecules with differences of a few millidaltons can be distinguished on the MS¹ level with HRMS, their MS² characterization is more challenging because the quadrupole (Q) isolation window for precursor selection typically has a width of a few hundred millidaltons to a few daltons. Thus, isobaric precursors are often co‐fragmented and interfere with each another's MS² spectra. In the literature, the resulting complicated spectrum is often referred to as a chimeric MS² spectrum or chimera. ⁸ If a feature of interest makes up ≤50% of the total MS¹ intensity in the isolation window, a rule of thumb says that spectral matching is unrealiable. ⁹ Thus, for DI‐MS metabolomics, where hundreds of features are detected in the low‐mass region, ¹⁰ , ¹¹ spectral matching is often problematic.

This problem is well known in liquid chromatography/MS (LC/MS)‐based proteomics or metabolomics, where complex samples often lead to chimeras – even with the chromatographic separation and both for data‐independent acquisition (DIA) as well as for precursor‐dependent acquisition. Fortunately, several strategies to tackle this problem were developed: (1) re‐analyzing the sample under different LC conditions; (2) using the retention time correlation of precursors and fragments to deconvolute and reconstruct MS² spectra; ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ (3) deconvolute a chimera directly into a linear combination of library MS² spectra; ¹⁸ , ¹⁹ , ²⁰ or (4) use of in‐silico fragmentation, ²¹ , ²² , ²³ , ²⁴ , ²⁵ which ultimately disentangles chimeras by focusing on fragmentations with reasonable mass losses.

For DI‐MS, options (1) and (2) are not feasible because they rely on chromatography. Option (3) could be applied to DI‐MS, but it assumes that the interfering isobars are reported in a spectral library. However, this might often not be the case for adducts and in‐source fragments ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ or for ionization techniques other than ESI. Option (4) can have similar limitations, but is still very helpful because the focus on reasonable mass losses is independent of the ionization technique. Nonetheless, it might be that a fragment is consistent with two isobars simultaneously from a pure mass loss perspective, e.g., the benzyl fragment C₆H₅ ⁺ could arise from either one (or both) of two co‐fragmented aromatic precursors. Only a few studies have explicitly addressed the problem for DI‐MS, e.g., it is reported that ion mobility ³¹ or “collisional purification” ³² , ³³ on MS³‐ or pseudo‐MS³‐capable instruments can resolve chimeras.

Here, we report an alternative method dubbed IQAROS (incremental quadrupole acquisition to resolve overlapping spectra) to resolve co‐fragmented precursors in DI‐HRMS. It relies on small, millidalton differences between the accurate masses of the precursors. The method is based upon the fact that the width of the Q isolation window is relatively broad, but the window's center can be set quite precisely. Thus, by moving the Q over the precursor range in a stepwise manner, the precursor's transmission through the Q is regulated and the intensities of the precursors and their associated fragments are therefore modulated (Figures 1a–1c). Through visual interpretation of the modulation behavior or by mathematical deconvolution, the fragments can then be assigned to the correct precursor(s).

FIGURE 1.

Overview of IQAROS: a, Two co‐isolated, co‐fragmented and co‐detected isobars result in a chimeric MS² spectrum. b, The basic principle of IQAROS. The isolation window is stepwise moved over the m/z range of the two isobaric precursors. On the narrowest isolation width setting, the isobars cannot be isolated individually, but they can be modulated distinctively. c, Moving the isolation window through the precursor region modulates the intensities of the two isobaric precursors. Their scan s dependent intensities are denoted $P_{1,s}$ and $P_{2,s}$. Likewise, the fragment intensities are also modulated. d, The observed fragment intensity $F_{i,s}$ of fragment i in scan s can be expressed by a linear combination of the precursor intensities $P_{j,s}$ in the same scan s and a contribution coefficient ${\beta }_{i,j}$. The estimates $\widehat{\mathbf{\beta }}$ are obtained with two methods here: a non‐negative multi‐linear regression (method 1) or a non‐negative simple‐linear regression (method 2). The green rectangle highlights the difference between the two methods [Color figure can be viewed at wileyonlinelibrary.com]

In related work, it was previously recognized that the problems associated with chimeras can be addressed by Q modulation. A binary alternation of the Q position was implemented in a DIA LC/MS² method ³⁴ and in a precursor‐dependent LC/MS² method. ¹⁷ Similar to IQAROS, a Q movement with a small step size was reported for two DIA LC/MS² methods termed scanning SWATH ³⁵ and SONAR. ³⁶ Notably, scanning SWATH, SONAR and IQAROS have in common that they resolve chimeras by Q modulation but they also have important differences. Scanning SWATH and SONAR are DIA methods for LC/MS, where co‐eluting precursors are modulated by moving a broad Q window ³⁵ , ³⁶ (m/z 10–24) over the entire mass range to fragment all eluting compounds. In contrast, IQAROS is a method for DI‐MS, which modulates with the narrowest possible Q width (here, m/z 0.4) only a targeted precursor of interest and interfering isobars, e.g., a MS¹ feature which was statistically significant in a metabolomic study ⁴ together with neighboring signals. Also notable is the DI‐HRMS study of Wang et al, ³⁷ who visually interpreted how the signal intensities of two interfering precursors in their petroleum sample changed upon positioning the Q at two to three different locations. From this, they concluded which fragment belongs to which precursor. To the best of our knowledge, modulation with the Q isolation window's center has never been performed systematically in DI‐MS. In this study, multiple precursors are modulated over numerous small Q steps followed by a mathematical deconvolution. We first assess the performance of IQAROS by analyzing mixtures of isobaric standards with ESI and then we apply IQAROS to breath analysis with secondary electrospray ionization (SESI), which is a DI metabolomic method where a key bottleneck is biomarker identification from the complex samples. ⁴

2. METHODS

2.1. Chemicals

For preparation of the ESI and SESI buffer solution, water (H₂O, Optima, Fisher Chemical, LC/MS grade), methanol (MeOH, Optima, Fisher Chemical, LC/MS grade) and formic acid (FA, Merck, for analysis, purity 98–100%) were used. As model compounds, six readily available isobars, which are separable in MS¹ but co‐fragment in MS², were selected and are shown in Figure 2. Namely, benzothiazole (1, TCI, purity >96.0%), pyridine‐2,6‐dicarbaldehyde (2, Fluorochem, purity ≥98%), 3H‐pyrrolo[2,3‐d]pyrimidin‐4(7H)‐one (3, Fluorochem, purity ≥95%), adenine (4, Sigma‐Aldrich, purity ≥99%), acetanilide (5, Sigma‐Aldrich, purity ≥99.5%) and N,N‐dimethylbenzylamine (6, Sigma‐Aldrich, purity ≥99%) were used. For an exemplary application, azelaic acid (7, Sigma‐Aldrich, purity ≥98%) and 10‐hydroxydecanoic acid (8, Apollo Scientific, purity ≥95%), which were previously identified in a breath metabolomic study, ³⁸ were used for control measurements. Commercial calibration solutions (Pierce ESI Positive Ion Calibration Solution and Pierce ESI Negative Ion Calibration Solution, Thermo Scientific) were used to mass calibrate the instrument with the ESI source.

FIGURE 2.

Compounds used in this study. Their labels, structures, molecular formulae and exact monoisotopic masses for a given ionization mode are listed [Color figure can be viewed at wileyonlinelibrary.com]

2.2. Sample preparation

The SESI spray solution consisted of H₂O + 0.1% FA. For ESI, a 50:50 (v/v) MeOH/H₂O + 0.1% FA solution was prepared as a blank and to dissolve the standards. A 3 μM solution was prepared for standards 1 and 3 to 6 and a 6 μM solution for standards 2, 7 and 8 to analyze them individually. Accounting for the experimentally determined sensitivities of the standards 1 to 6, mixtures with equal signal intensities were prepared: a two isobars mixture called mixture II consisting of 9.82 μM 1 and 0.35 μM 6; a three isobars mixture called mixture III consisting of 9.82 μM 1, 0.77 μM 5 and 0.35 μM 6; and a six isobars mixture called mixture VI consisting of 9.82 μM 1, 14.80 μM 2, 2.51 μM 3, 0.97 μM 4, 0.77 μM 5 and 0.35 μM 6.

2.3. Mass spectrometry

The mass spectrometer used in this study was an orbitrap Q‐Exactive Plus (Thermo Scientific) operated with the manufacturer's standard control software (ExactiveTune, version 2.9, Thermo Scientific) and Xcalibur (version 4.1.31.9, Thermo Scientific). Mass calibration was performed according to the instrument manual and was always more recent than 7 days according to specifications.

MS¹ spectra were acquired with the following settings: narrowest scan range with a m/z 0.4 isolation width around the target center, profile mode, 5e6 automatic gain control (AGC) target, 500 ms maximum injection time (IT). Where not otherwise specified, the lowest resolution setting at 17,500 was selected.

Direct MS² spectra were acquired with the following settings: narrowest isolation width m/z 0.4 centered around the monoisotopic exact mass of the investigated compounds, profile mode, 3e6 AGC target, 500 ms maximum IT, 10, 35 and 50 eV stepped collision energies (CE) and 17,500 resolution.

Similar settings were used for IQAROS, but instead of centering around the monoisotopic exact mass, incremental target masses were manually entered via the graphical user interface of Xcalibur (Figure S1, supporting information). The list started −0.6 Da and ended +0.6 Da of the nominal mass of interest and contained entries in steps of 0.02 Da. For example, to apply IQAROS to an ion at m/z 136.11207, the list was constructed as follows: 135.40, 135.42, 135.44, …, 136.56, 136.58, 136.60 and contained a total of 61 entries. Moreover, a 500 ms maximum IT was used. The AGC target was set to 3e6 except for mixture VI, where it was set to 5e5 due to some intensity regulation artefact at AGC target 3e6. For the lower abundant SESI biomarkers, a repetition of every step (122 entries instead of 61) was used.

ESI measurements were performed with a standard source for this instrument (HESI‐II probe, Thermo Scientific). The ESI settings were: (+)‐ or (−)‐polarity, 4 kV spray voltage, 320°C ion transfer capillary temperature, 12 psi sheath gas, 0 aux gas, 0 sweep gas and 50 S‐lens RF level. Breath analysis with SESI was performed with a commercial SESI source (Super SESI, Fossiliontech) equipped with a 20 μm i.d. capillary (for Super SESI, Fossiliontech) and a flow sensor (Exhalion, Fossiliontech). Similar to prior studies with the same setup, ¹⁰ , ¹¹ a subject exhaled through a spirometry filter (MicroGard IIB, Vyaire Medical) with 8 L min⁻¹ of which 0.3 L min⁻¹ was directed through a transfer line at 130°C into an ionization chamber held at 90°C. For SESI, the analyzer's settings were: (−)‐polarity, 3.5 kV spray voltage, 250°C ion transfer capillary temperature and 60 S‐lens RF level.

2.4. Data analysis

MS¹.raw files were converted into.mzXML files with MSConvert (version 3, ProteoWizard ³⁹ ), processed and plotted in Matlab (R2018a, Mathworks). In contrast, MS² .raw files were converted into.mgf file format with MSConvert and processed with Matlab.

Direct ESI‐MS² data for standards and the blank were averaged over all scans for each selected precursor and peak‐picked with the Matlab function mspeaks. The MS² spectra of the standards were then blank subtracted. The resulting averaged and blank‐subtracted MS² spectra were written into a new.mgf file for every precursor and used to compare the results against the IQAROS output. The direct SESI‐MS² was averaged during exhalation and baseline‐subtracted. Then, the same processing as for the ESI standards was applied.

An overview of the IQAROS code is shown in Figure S2 (supporting information). Moreover, the Matlab code (provided as part of the archived data under DOI 10.3929/ethz‐b‐000520528) is fully commented; thus, only a brief description is given here. In a first step, the code determines which peaks are located within the m/z range where Q modulation took place. These peaks are considered precursors. Peaks with a lower m/z than the precursors are considered fragments. For all fragments and for all intense precursors (>5% of the maximum precursor intensity), the extracted ion currents (XICs) are calculated. For p scans and n fragments, the XICs are written into a n × p matrix herein called F. Likewise, the XICs for m precursors are written into a m × p matrix called P.

Similar to the work of Nikolskiy et al, ¹⁷ the underlying model is that the intensity $F_{i,s}$ of the fragment i in scan s can be expressed as a linear combination of all m precursor intensities $P_{j,s}$ in the same scan s multiplied with a contribution coefficient ${\beta }_{i,j},$ i.e., $F_{i,s}={\sum }_{j=1}^m{\beta }_{i,j}·P_{j,s}$. As an example, Figures 1c and 1d depict n = 3 fragments, m = 2 precursors and p = 3 scans. Instead of considering only one scan s and one fragment i, this can be rewritten for all p scans and n fragments: $\mathbf{F}=\mathbf{\beta }·\mathbf{P}$ with the n × p matrix F containing the fragment intensities for n fragments and p scans, the m × p matrix P containing the precursors intensities for m precursors and p scans, and the n × m matrix $\mathbf{\beta }$ describing the contribution to fragment i from every precursor j. The column j of matrix $\mathbf{\beta }$ represents the MS² spectrum for precursor j. For example, ${\beta }_{i,j}=0$ means that fragment i is not part of the MS² spectrum of precursor j. Likewise, ${\beta }_{i,j}=1$ or ${\beta }_{i,j}=2$ signifies that fragment i appears in the MS² spectrum of precursor j with the same or double the intensity of precursor j, respectively.

To obtain the estimates ${\widehat{\beta }}_{i,j}\mathit{,}$ two methods were used, as depicted in Figure 1d: method 1 is identical to that of Nikolskiy et al,¹⁷ i.e., ${\widehat{\beta }}_{i,j}$ are estimated by running a non‐negative multiple linear regression with the Matlab function lsqnonneg for every fragment i. Performing n times a non‐negative m‐multiple linear regression yields matrix $\widehat{\mathbf{\beta }}$. For method 2, it is assumed that every fragment is the product of only one precursor. The aim is to find the single precursor j which alone best describes the observation of fragment i. Thus, for every ith fragment m, non‐negative simple linear regressions are run with every jth precursor XIC as independent variable yielding in m ${\widehat{\beta }}_{i,j}$ per fragment i. For every jth regression, the coefficient of determination $R_{i,j}^2$ is calculated. Only the regression coefficient ${\widehat{\beta }}_{i,j}$ with the highest $R_{i,j}^2$ is written into matrix $\widehat{\mathbf{\beta }}$ for fragment i, all other values of ${\widehat{\beta }}_{i,j}$ are set to zero. Thus, every row i of $\widehat{\mathbf{\beta }}$ contains only one element >0.

From the $\widehat{\mathbf{\beta }}$ matrix, the mass spectra of the precursors are reconstructed, i.e. for a column j in $\widehat{\mathbf{\beta }}$ the entry ${\widehat{\beta }}_{i,j}$ represents the intensity of fragment i when precursor j has a normalized intensity of 1. Finally, the reconstructed mass spectra are saved as.mgf files.

2.5. Spectral matching

The deconvoluted spectra were compared with the blank‐subtracted direct MS² spectra. For this purpose, peaks with higher m/z values than the precursor were omitted and the intensities were filled into two incremental vectors ${\overrightarrow{I}}_1$ and ${\overrightarrow{I}}_2$ with 0.005 m/z increments. The MS² match score was then calculated as the squared cosine of the angle $\theta$ between the two vectors ⁴⁰ , ⁴¹ :

$$\text{score}={\cos }^2\left(\theta \right)={\left(\frac{{\overrightarrow{I}}_1·{\overrightarrow{I}}_2}{‖{\overrightarrow{I}}_1‖‖{\overrightarrow{I}}_2‖}\right)}^2.$$

To further assess the performance of IQAROS, the direct and the deconvoluted MS² spectra were processed with SIRIUS ²⁴ , ²⁵ (version 4.8.2). The database search was performed in all available databases for [M + H]⁺ or [M – H]⁻ species, respectively. Otherwise, orbitrap default settings were used except for the MS² mass tolerance (MS2 MassDev) which was set to 10 ppm and proposed structures with electron sextets were ignored.

3. RESULTS AND DISCUSSION

3.1. Deconvolution performance with isobaric standards

The problem of chimeric MS² spectra is shown in Figure 3a. The exemplary isobars 1 and 6 in mixture II are perfectly separable on the MS¹ level. When performing classical direct MS², i.e. centering the quadrupole at the precursor of interest, here m/z 136.0216 for 1 and m/z 136.1121 for 6, problems with chimeras arise. Since the two ions only differ by m/z 0.09, they cannot be isolated separately by the Q – even at the narrowest isolation width of m/z 0.4. In fact, when targeting 1 or 6 with direct MS², the two spectra are contaminated by the other isobar. Particularly, the spectrum of 1 is severely contaminated by the tropylium ion [6 – C₂H₇N + H]⁺ originating from 6 (Figure S4a, supporting information).

FIGURE 3.

Overview of IQAROS applied to mixture II containing 1 and 6. a, The MS¹ signal of the isobar mixture (the m/z 0.4 wide Q isolation window as well as the m/z 0.06 Q step size accuracy are also depicted). b, The normalized MS² XICs of [1 + H]⁺ and [6 + H]⁺ as well as their major fragments [1 – CHN + H]⁺ and [6 – C₂H₇N + H]⁺ during one modulation cycle with the Q isolation window. The XICs over all modulation cycles are shown in Figure S3a (supporting information). c, d, Plots of the normalized fragment intensities against the normalized intensities of the precursor [1 + H]⁺ and [6 + H]⁺, respectively. The three‐dimensional representation of the same plot can be found in Figure S3b (supporting information) [Color figure can be viewed at wileyonlinelibrary.com]

While m/z 0.4 is the narrowest Q isolation width, the Q isolation center can be entered in the instrument control software with a theoretical accuracy of m/z 0.00001. However, we found on our instrument that the Q hardware reacts to input by the software with an accuracy of approximately m/z 0.06, although this does not limit the resolving power of IQAROS, as will be discussed below. Thus, we set up IQAROS with a software‐defined arbitrary increment of m/z 0.02 (Figure S1, supporting information). Important is that the software‐defined increment is smaller than the hardware increment; here 0.02 ≤0.06, but different values are expected on other instruments. A larger step size within the hardware limit could make IQAROS even faster, but this was not optimized here. Therefore, the actual modulation corresponds to the one depicted in Figure 1b, i.e., the software moves the Q window after every scan by m/z 0.02 (Figure S1, supporting information) but the hardware reacts only after every third scan by moving m/z 0.06 ahead. In principle, an alternation between MS¹ and MS² is also possible, which would result in a method as depicted in Figure S5a (supporting information) and would be better to detect low‐abundance interfering isobars in the MS¹ scan. However, due to an instrument‐specific technical restriction, which does not allow the orbitrap scan range and the Q isolation range for MS¹ to be set independently from each another (details in Figure S5b and S5c, supporting information), the modulation here was only performed on the MS² level, as shown in Figure 1b.

Figure 3b shows the XICs of [1 + H]⁺ and [6 + H]⁺ in mixture II as well as their major fragments [1 – CHN + H]⁺ and [6 – C₂H₇N + H]⁺ over one modulation cycle. The total modulation involved six of these cycles and required approximately 360 scans or 3 min in total (Figure S3a, supporting information). The precursor‐fragment correlation can be depicted by plotting the normalized intensities of the fragments against the normalized intensities of the precursors as shown in Figures 3c and 3d or in three dimensions in Figure S3b (supporting information). [1 – CHN + H]⁺ correlates well with [1 + H]⁺ and also with [6 + H]⁺, but with significantly higher residuals for the latter. Thus, a relationship of [1 – CHN + H]⁺ with 1 but not with 6 can be deduced. Similar observations can be made with mixture III with 1, 5 and 6 (Figure S6, supporting information) and mixture VI with 1, 2, 3, 4, 5 and 6 (Figure S7, supporting information). Remarkably, even if isobars fall into the same m/z 0.06 Q step size, they and their fragments can be distinguished because their intensities are modulated distinctively, e.g., in scan 473 the normalized intensity ratios of [3 + H]⁺, [4 + H]⁺ and [5 + H]⁺ are 3%:10%:20% (Figure S7c, supporting information), although they are only spaced by m/z 0.025. Consequently, IQAROS has the potential to deconvolute isobars even if they are spaced closer to each other than the accuracy with which the Q's center of mass can be set.

To extract this relationship quantitatively, the two deconvolution methods were applied: method 1 establishes a fragment‐precursor relationship by describing every fragment intensity as a positive linear combination of precursor intensities. In contrast, method 2 assigns every fragment to that single precursor which best explains the intensity of the fragment. Mathematically, method 1 is based on a non‐negative multi‐linear regression. The reconstructed spectra of mixtures II, III and VI are compared with the pure standards in Figure S4, S8 and S9 (supporting information), respectively. For benchmarking with the standard method, the pure standards are also compared with the direct MS² measurements of the isobaric mixtures. The direct and the IQAROS spectra are compared with the pure standards by calculating the MS² match score, which is plotted in Figure 4. Except for compound 5 in mixture VI, all constellations have a higher MS² match score with IQAROS with deconvolution method 1. Indeed, the mean MS² match score for all eleven constellations for direct MS² is 0.42 and for IQAROS MS² 0.87. Interestingly, the score for 5 in mixture III is better than in mixture VI. In fact, with mixture VI, the deconvolution algorithm incorrectly assigns the fragment [5 – C₂H₂O + H]⁺ to 3, 4 and 6 instead of 5. Hence, a non‐negative linear combination of the three precursors seems to describe the XIC [5 – C₂H₂O + H]⁺ better than 5 alone. This finding can be interpreted as a problem with multicollinearity, i.e., the independent variables themselves are correlated with each another. This is the case for IQAROS where moving Q's isolation center into the isobars region increases the intensity of all precursors together. Multicollinearity can be quantified with a mathematical figure of merit called the variance inflation factor, 0 ≤ VIF < ∞, where a value of VIF >10 is considered problematic in regards of multicollinearity. ⁴² For mixture II, the two VIFs are around 10⁻⁵. For mixture III, the VIFs of 1 and 6 are at 7 and 11 and for 5 already at 120 indicating problematic multicollinearity. For mixture VI, the VIFs lie between 16 and 1060. Consequently, the poor result for 5 in mixture VI can be attributed to issues with multicollinearity of the method and the deconvolution model.

FIGURE 4.

MS² match scores of the direct and IQAROS deconvoluted spectra for mixtures II, III and VI. a, b, and c, MS¹ spectra for mixtures II, III and VI, respectively. d, MS² match scores between MS² spectra of the pure standards with the direct and the IQAROS spectra of the isobaric mixtures. For IQAROS, results for deconvolution methods 1 and 2 are shown [Color figure can be viewed at wileyonlinelibrary.com]

Multicollinearity can be tackled by alternative regression methods like ridge or lasso (least absolute shrinkage and selection operator). ⁴³ , ⁴⁴ However, these methods cannot be run on the standard numerical computing software used here with the non‐negativity restriction. Yet, determining $\widehat{\mathbf{\beta }}$ in this study is restricted to non‐negativity because it is assumed that a precursor never leads to a “negative intensity” fragment. This assumption is important, because it already eliminates some mathematical linear combinations which would otherwise be troublesome when dealing with the multicollinear data. Thus, an alternative to ridge or lasso regressions when dealing with multicollinearity is variable elimination in the model. ⁴³ This is implemented in method 2, where every fragment is assigned to the single precursor which best explains the XIC of the given fragment. Mathematically, this is done by picking the non‐negative simple linear regression model with the best fit, i.e., with the highest coefficient of determination. Method 2 will not suffer from multicollinearity as method 1. However, as a disadvantage, method 2 will not correctly describe when two precursors lead to the same fragment. Instead, the fragment will be assigned to only one precursor as, for example, depicted for fragment F₃ in Figures 1c and 1d. The IQAROS spectra deconvoluted by method 2 are shown for the three isobaric mixtures in Figures S10, S11 and S12 (supporting information) and the MS² match scores comparison in Figure 4. With method 2, IQAROS yields an average MS² match score of 0.99 and therefore the multicollinearity and the problem with compound 5 in mixture VI is eliminated.

3.2. Spectral search with isobaric standards

In the previous section, it was demonstrated that IQAROS leads to fragment spectra comparable with spectra of the pure standards. Here, the discussion focuses on how helpful this is for compound identification. For this purpose, the previously discussed spectra were processed with the spectral interpretation software SIRIUS. ²⁴ , ²⁵ The most challenging sample, i.e., mixture VI, is addressed.

The first column in Table 1 shows the structures of the six isobaric standards. When the pure standards are analyzed by direct MS², blank‐subtracted and processed with SIRIUS, the structures shown in the second column in Table 1 are obtained. This represents the best possible outcome because the inputs are the pure standards. In fact, SIRIUS correctly assigns 6 of 6 molecular formulae and 4 of 6 structures. The two incorrectly assigned structures are closely related isomers of the true structures. In contrast, when mixture VI, analyzed with direct MS², is processed, only 4 of 6 molecular formulae and 2 of 6 structures are correct. The wrong assignment of the molecular formulae arises from a combination of two adverse effects in high‐resolution tandem MS: (1) partial ion coalescence ⁴⁵ , ⁴⁶ , ⁴⁷ leads to significant shifts in accurate mass of the precursor (Figure S13, supporting information) and (2) the many fragments from other precursors possibly coincided with a fragment which would be consistent with the erroneous precursor mass. Here, IQAROS can help by eliminating fragments which coincidently fit a reasonable mass difference but actually stem from an interfering precursor. Indeed, 6 of 6 molecular formulae are assigned correctly with method 1 IQAROS. Moreover, 3 of 6 structures are correct – comparable with the 4 of 6 when the pure standards are measured. Only 5 is not reasonably assigned. Presumably, the aforementioned elimination of [5 – C₂H₂O + H]⁺ from the fragment list of 5 due to multicollinearity renders the structural assignment difficult because [5 – C₂H₂O + H]⁺ would indicate a N‐monosubstituted phenyl ring (N analog of the tropylium ion).

TABLE 1.

Compound identification of the standards using SIRIUS. ²⁴ , ²⁵ The first column shows the structures of the six isobaric standards. The second column shows the most likely structures according to the output of SIRIUS when the pure blank‐subtracted standards are processed. If the true structure is not the top hit, the position of the correct hit is listed instead. The third column shows the SIRIUS output when processing the direct MS² targeted mixture VI. The fourth and fifth columns show the SIRIUS output of mixture VI when analyzed by IQAROS with method 1 and method 2, respectively

If IQAROS with method 2 is applied, the true structure of 5 is at the 3rd hit position. As a disadvantage, method 2 eliminates too many features for 6. While it correctly assigns the main fragment [6 – C₂H₇N + H]⁺ to 6, it omits the minor fragment [6 – C₄H₉N + H]⁺, which corresponds to [C₅H₄ + H]⁺ and is found for many aromatic compounds. Only with one fragment was SIRIUS not able to run a search and thus left 6 unassigned. In summary, IQAROS with method 2 can avoid errors from multicollinearity but meanwhile might be too conservative in fragment elimination.

In conclusion, both IQAROS methods 1 and 2 lead to better formula assignments when analyzing isobaric mixtures compared with the classical direct MS² analysis of the mixed isobars. For structure assignment of the mixed isobars, IQAROS performs in most cases better or comparable to direct MS² analysis. However, the structure assignment does not meet the quality of the pure standards analyzed individually by direct MS².

3.3. Example: Identification of isobars in breath

An exemplary application will now be discussed where IQAROS turns out to be very useful. Secondary electrospray ionization (SESI)‐MS is an enormously sensitive and rapid technique for real‐time DI analysis of volatile metabolites and therefore is typically applied for breath analysis. ⁴ Depending on the acquisition technique and instrument, a few hundred to a few thousand features are found in the mass range from m/z 50 to 500 during exhalation. ¹⁰ , ¹¹ Taking into account that there are also some background peaks, it becomes apparent that many features must have neighboring isobars and thus often account for ≤50% of the isolated total intensity, which makes MS² spectral matching ambiguous. ⁹ Hence, reliable biomarker identification is usually done by collecting breath condensate followed by LC/ESI‐MS. ³⁸ However, this method is time‐consuming and can lead to contamination/analyte loss during sample preparation. As a consequence, it is highly desirable to obtain higher quality MS² spectra of biomarkers without the need for condensation and LC separation.

As an example, isobaric signals around m/z 187 acquired by SESI‐MS during the exhalation of a volunteer are shown in Figure 5a. In a previous study by Gaugg et al, ³⁸ two of the isobars were identified by breath condensation and LC/MS to correspond to fatty acids 7 and 8. Both in the study of Gaugg et al ³⁸ as well as here (Figures 5b and 5c in blue), a direct MS² experiment acquired over five exhalations leads to non‐interpretable fragment spectra due to the co‐fragmented isobars. In fact, if the direct MS² spectra are processed by SIRIUS, incorrect structural assignments are obtained (Figure 5f, second column). Conversely, with IQAROS over five exhalations and the more stringent method 2, the spectra (Figures 5d and 5e in blue) fit well with the ESI‐MS² spectra obtained from the pure standards (Figures 5d and 5e) in red). If the IQAROS spectra are processed with SIRIUS, they yield the same structures as the ones reported by Gaugg et al. ³⁸

FIGURE 5.

Example of two isobaric biomarkers in breath analyzed by SESI‐MS and identified with IQAROS: a, SESI‐MS¹ range during exhalation around m/z 187 where the previously identified biomarkers 7 and 8 lay. ³⁸ b, c, Direct SESI‐MS² spectra of 7 and 8 in blue. For comparison, the ESI‐MS² of the pure standards is shown in red. d, e, SESI spectra obtained with IQAROS method 2 for 7 and 8 in blue and the same ESI‐MS² spectra as before in red. F, In the first column, the previously identified biomarkers ³⁸ are shown, in the second column, the SIRIUS output from the direct MS² spectra b) and c) is shown, and, in the third column, the SIRIUS output from IQAROS from d) and e) is depicted [Color figure can be viewed at wileyonlinelibrary.com]

In conclusion, direct SESI‐MS² can lead to incorrect assignments due to co‐fragmented isobars. In contrast, with IQAROS biomarkers in SESI‐MS can be accurately identified without the need to collect breath condensate followed by LC/MS. Comparable results are obtained within minutes instead of several hours of sample preparation and analysis.

4. CONCLUSIONS

IQAROS (incremental quadrupole acquisition to resolve overlapping spectra) is an approach to disentangle chimeric MS² spectra that arise when a precursor of interest is co‐fragmented with neighboring isobars in DI‐HRMS. The method modulates precursor and fragment intensities by moving the isolation quadrupole's center over the precursor range in a stepwise fashion, followed by a mathematical deconvolution to reconstruct the MS² spectra of the individual precursors. The method can easily be implemented by simply creating a list with incremental precursor isolation center masses using the graphical user interface of the instrument.

For the analysis of a mixture containing up to six isobaric standards confined within a range of m/z 0.09, IQAROS delivered higher quality MS² data compared with a direct MS² approach. The match between the deconvoluted spectra and the spectra of pure standards was significantly higher and compound search with the spectra interpretation software SIRIUS led to a better hit rate. Depending on the mathematical model, problems might arise from multicollinearity (method 1 – multi linear regression) or from the incapability of the deconvolution algorithm to assign a fragment to more than one precursor (method 2 – simple linear regression). As an exemplary application, IQAROS was able to directly identify two biomarkers from on‐line breath measurements yielding the same structures as in a previous study, ³⁸ but without the need for time‐consuming breath condensate collection and LC/MS analysis. In general, IQAROS might be useful for all sorts of ambient or desorption ionization methods like DART, DBDI, DESI, EESI, DAPCI, and many more. ⁴⁸ For these techniques, hyphenation with chromatography to resolve interferences is often difficult or impossible. In contrast, IQAROS allows to mitigate the problem of co‐fragmentation without the need for chromatography or changes in the ionization process.

In this study, an isolation quadrupole in tandem with an orbitrap was used together with ESI‐based ionization sources. The deconvolution was performed by a non‐negative multi‐ or simple‐linear regression. It is important to note that IQAROS can be used with other setups as well: (1) it can be used for variable fragmentation techniques or ionization methods; (2) it can also be employed for other high‐resolution MS instruments such as time‐of‐flight or Fourier transform ion cyclotron resonance analyzers, and (3) other mathematical methods such as ridge, lasso or many more could possibly be used for deconvolution.

Limitations arise from multicollinearity when many precursors are deconvoluted. This problem can possibly be tackled with more advanced statistical methods. Moreover, the method needs sufficiently intense precursors, i.e., if a precursor is modulated to 10% of its maximum intensity, a fragment must still be abundant enough to be detected. An additional limitation, unique to the instrument used in this study, is that at 10 V collision energy (CE) the precursors must still be detectable. However, this criterion is repealed if modulation can be performed while alternating between MS¹ and MS².

PEER REVIEW

The peer review history for this article is available at https://publons.com/publon/10.1002/rcm.9266.

Supporting information

Figure S1. Screenshot of the method setup for IQAROS with the graphical user interface Xcalibur (version 4.1.31.9, Thermo Scientific). A normal targeted MS² method is set up, but instead of entering target precursors, an incremental list of m/z values is defined. For example, to apply IQAROS to an ion at m/z 136.11207, the list was constructed as follows: 135.40, 135.42, 135.44, …, 136.56, 136.58, 136.60 and contained a total of 61 entries. The method here is setup for 60 min and will just cycle through the 61 MS² acquisitions. Once sufficient modulations were performed (typically after 3 min), the acquisition was manually stopped by the operator. On the right side of the screenshot, other method parameters can be defined. The settings there can be adjusted to the operator's needs. Please note, that the 10 eV CE is important on the instrument used in this study (Figure S5). The above screenshot is unique to the instrument and software used in this study. However, a similar setup for IQAROS should be feasible on other tandem HRMS instruments.

Figure S2. Overview of the IQAROS Matlab code: main.m is the main Matlab file which calls other functions/files. The arrows represent input and output from these auxiliary functions/files. First, the user must define a deconvolution model i.e. either a non‐negative multiple linear regression or a non‐negative simple linear regression. Moreover, the quadrupole isolation width and peak‐picking parameters for the Matlab function mspeak must be defined by the user. The code then reads the.mgf filenames defines in the file “input_files.txt”, where the user before must have entered the.mgf file names to be processed. The.mgf files are read by “f_mgf_reader.m” and averaged over all scans by “f_spec_all_averager.m”. This averaged spectrum is then used to peak‐pick with mspeaks. The peaklist is then divided into a fragment list (peaks with m/z lower than the range where Q modulation was performed) and a precursor list (peaks with m/z in the range where Q modulation was performed). In the precursor list, only peaks with >5% intensity of the maximum precursor peak are kept for further processing. For reference, the untreated MS² spectra are plotted as .pdf with “f_print_overview.m” and saved as.mgf with “f_mgf_writer_untreated.m”. Then, XICs are calculated for all fragments from the fragment list and for all precursors from the precursor list. To these XICs, the user‐defined regression model is applied i.e. every i of the total n fragment XICs is described by the m precursor XICs multiplied with m regression coefficients β _i,j . From these regression coefficients β _i,j , the MS² spectrum for precursor j is reconstructed with “f_spectrum_reconstruction.m”. The reconstructed MS² are plotted with “f_print_spectrum_pdf.m”. For an additional overview, the precursors and fragment XICs are plotted with “f_print_XIC_pdf.m”, where related fragment/precursor pairs are plotted in red. Finally, the reconstructed MS² are written as.mgf files with “f_mgf_writer.m”.

Figure S3. Additional figures for the modulation applied to mixture II in Figure 3 in the main text. a) shows the normalized MS² XICs of [1 + H]⁺ and [6 + H]⁺ as well as their major fragments [1–CHN + H]⁺ and [6–C₂H₇N + H]⁺ for all six modulation cycles requiring approximately 3 min. A zoom into the last modulation cycle is shown in Figure 3b). b) shows the normalized fragment intensities of [1–CHN + H]⁺ and [6–C₂H₇N + H]⁺ on the z axis plotted against the normalized intensities of precursors of [1 + H]⁺ and [6 + H]⁺ on the x and y axis, respectively. The xz and yz projections of this 3 dimensional plot are shown in Figure 3c) and 3d).

Figure S4. Comparison of the direct MS² spectra of the pure standards 1 and 6 with the direct MS² and the IQAROS spectra deconvoluted with method 1 from mixture II containing 1 and 6. a) shows the direct MS² spectrum of pure 1 in red and the direct MS² spectrum of mixture II in blue. b) shows the direct MS² spectrum of pure 1 in red and the IQAROS deconvoluted spectrum according to method 1 for compound 1 of mixture II in blue. c) shows the direct MS² spectrum of pure 6 in red and the direct MS² spectrum of mixture II in blue. d) shows the direct MS² spectrum of pure 6 in red and the IQAROS deconvoluted spectrum according to method 1 for compound 6 of mixture II in blue.

Figure S5. Figures showing why an alternation between MS¹ and MS², similar to Nikolskiy et al's study¹ with a variation between 0 and 20 V CE, is not possible with the instrument used in this study. a) schematically depicts such an alternation. First, the narrowest possible range centered around a first m/z value is isolated by the Q and analyzed on MS¹ level by the orbitrap. Second, a MS² scan with the same center and width is performed followed by a third MS¹ isolation with identical center and width. Only then, the center is shifted by the selected step size and the cycle repeats. Importantly, the MS¹ orbitrap scan range and the isolation Q window are linked with each another on the instrument used in this study. For example, to isolate the range m/z 176.78–177.18 with the Q, the user must set this range for the orbitrap as well. Unfortunately, peaks outside the orbitrap MS¹ scan range are not returned to the user as will be discussed below. This makes sense for normal operation because no ions outside the quadrupole isolation window are expected to be detected on the orbitrap and thus no information outside the isolation window is stored to save computer memory. However, IQAROS is based on the detection of signals just around the edges of the isolation window and thus this information removal becomes problematic. b) shows a breath metabolic sample analyzed with SESI‐MS scanned on the MS¹ level. Three consecutive scan ranges with the narrowest m/z 0.4 width centered at m/z 176.94, 176.96 and 176.98 are shown. The right limit of the isolation window is shown as a dotted line. In the last scan plotted in green and centered at m/z 176.98, the peak at m/z 177.16 is fully detected. Parts of the same peak are also visible in the second scan plotted in turquoise and centered at m/z 176.96. However, the peak is abruptly cut and the read‐out intensities in the.raw file or in the instrument's software equal 0. The intensities for the same peak equal 0 for the first scan in blue centered at m/z 176.96. The phenomenon, that the peak outside of the set MS¹ is either completely 0 or abruptly cut, leads to the conclusion that the instrument's software does not return these data points to the operator. Instead, only data points within the selected scan range are returned. Because IQAROS requires the detection of ions outside the user‐defined isolation window limits, this phenomenon makes the MS¹ and MS² alternation impossible. c) In contrast, when the same mass range is analyzed with the same three consecutive scan ranges on the MS² level, the peak at m/z 177.16 is consistently detected. In principle, a MS¹ could be mimicked by an MS² scan with a 0 CE. However, the lowest possible CE allowed by the control software is 10 eV. As a compromise, the MS²‐only acquisition as depicted in Figure 1b) was implemented with 10, 35 and 50 eV stepped collision energies. A disadvantage of this compromise is that some precursor peaks aren't sufficiently detected anymore but still contribute to the modulated fragments, for example the peaks at m/z 177.04 or m/z 177.06 in figure b) and c). Please note that this limitation is instrument‐specific. It might be that on another instrument, a method with MS¹ and MS² alternation as shown in a) could be implemented.

Figure S6. Overview of IQAROS applied to mixture III containing 1, 5 and 6. a) shows the MS¹ signal of the isobar mixture. b) shows the normalized MS² XIC of [1 + H]⁺, [5 + H]⁺ and [6 + H]⁺ as well as their major fragments [1–CHN + H]⁺, [5–C₂H₂O + H]⁺ and [6–C₂H₇N + H]⁺ during the modulation with the Q isolation window. c) shows a zoom of the last modulation cycle. d) shows the normalized fragment intensities plotted against the normalized intensity of [1 + H]⁺. e) shows the normalized fragment intensities plotted against the normalized intensity of [5 + H]⁺. f) shows the normalized fragment intensities plotted against the normalized intensity of [6 + H]⁺. The fragment's intensities [1–CHN + H]⁺, [5–C₂H₂O + H]⁺ and [6–C₂H₇N + H]⁺ tend to align best with their corresponding precursor while having a significant deviation for the other two precursors.

Figure S7. Overview of IQAROS applied to a mixture VI containing 1, 2, 3, 4, 5 and 6. a) shows the MS¹ signal of the isobar mixture. b) shows the normalized MS² XIC of [1 + H]⁺, [2 + H]⁺, [3 + H]⁺, [4 + H]⁺, [5 + H]⁺ and [6 + H]⁺ as well as their major fragments [1–CHN + H]⁺, [2–C + H]⁺, [3–H₂O + H]⁺, [4–H₃N + H]⁺, [5–C₂H₂O + H]⁺ and [6–C₂H₇N + H]⁺ during the modulation with the Q isolation window. c) shows a zoom of the last modulation cycle. In scan 472–474, it can be seen that precursors and fragments spaced closer than the m/z 0.06 step size can be distinguished. [6 + H]⁺ and [6–C₂H₇N + H]⁺ are roughly at 45% of normalized intensities, [5 + H]⁺ and [5–C₂H₂O + H]⁺ at 20%, [4 + H]⁺ and [4–H₃N + H]⁺ at 10% and [3 + H]⁺ and [3–H₂O + H]⁺ at 3%. Schematically, this is explained in the inserted figure. Since the transmission efficiency of the quadrupole is not perfectly rectangular, signals are modulated distinctively depending on where exactly they lay in the transmission profile. d) shows the normalized fragment intensities plotted against the normalized intensity of [1 + H]⁺. e) shows the normalized fragment intensities plotted against the normalized intensity of [2 + H]⁺. f) shows the normalized fragment intensities plotted against the normalized intensity of [3 + H]⁺. g) shows the normalized fragment intensities plotted against the normalized intensity of [4 + H]⁺. h) shows the normalized fragment intensities plotted against the normalized intensity of [5 + H]⁺. i) shows the normalized fragment intensities plotted against the normalized intensity of [6 + H]⁺. The fragment's intensities [1–CHN + H]⁺, [2–C + H]⁺, [3–H₂O + H]⁺, [4–H₃N + H]⁺, [5–C₂H₂O + H]⁺ and [6–C₂H₇N + H]⁺ often align best with their corresponding precursor while having a significant deviation for the other five precursors.

Figure S8. Comparison of the direct MS² spectra of the pure standards 1, 5 and 6 with the direct and the IQAROS spectra deconvoluted with method 1 from mixture III containing 1, 5 and 6. a) shows the direct MS² spectrum of pure 1 in red and the direct MS² spectrum of mixture III in blue. b) shows the direct MS² spectrum of pure 1 in red and the IQAROS deconvoluted spectrum according to method 1 for compound 1 of mixture III in blue. c) shows the direct MS² spectrum of pure 5 in red and the direct MS² spectrum of the mixture III in blue. d) shows the direct MS² spectrum of pure 5 in red and the IQAROS deconvoluted spectrum according to method 1 for compound 5 of mixture III in blue. e) shows the direct MS² spectrum of pure 6 in red and the direct MS² spectrum of mixture III in blue. f) shows the direct MS² spectrum of pure 6 in red and the IQAROS deconvoluted spectrum according to method 1 for compound 6 of mixture III in blue.

Figure S9. Comparison of the direct MS² spectra of the pure standards 1, 2, 3, 4, 5 and 6 with the direct and the IQAROS spectra deconvoluted with method 1 for mixture VI containing 1, 2, 3, 4, 5 and 6. a) shows the direct MS² spectrum of pure 1 in red and the direct MS² spectrum of mixture VI in blue. b) shows the direct MS² spectrum of pure 1 in red and the IQAROS deconvoluted spectrum according to method 1 for compound 1 of mixture VI in blue. c) shows the direct MS² spectrum of pure 2 in red and the direct MS² spectrum of mixture VI in blue. d) shows the direct MS² spectrum of pure 2 in red and the IQAROS deconvoluted spectrum according to method 1 for compound 2 of mixture VI in blue. e) shows the direct MS² spectrum of pure 3 in red and the direct MS² spectrum of mixture VI in blue. f) shows the direct MS² spectrum of pure 3 in red and the IQAROS deconvoluted spectrum according to method 1 for compound 3 of mixture VI in blue. g) shows the direct MS² spectrum of pure 4 in red and the direct MS² spectrum of mixture VI in blue. h) shows the direct MS² spectrum of pure 4 in red and the IQAROS deconvoluted spectrum according to method 1 for compound 4 of the mixture VI in blue. i) shows the direct MS² spectrum of pure 5 in red and the direct MS² spectrum of mixture VI in blue. j) shows the direct MS² spectrum of pure 5 in red and the IQAROS deconvoluted spectrum according to method 1 for compound 5 of mixture VI in blue. k) shows the direct MS² spectrum of pure 6 in red and the direct MS² spectrum of mixture VI in blue. l) shows the direct MS² spectrum of pure 6 in red and the IQAROS deconvoluted spectrum according to method 1 for compound 6 of mixture VI in blue.

Figure S10. Comparison of the direct MS² spectra of the pure standards 1 and 6 with the direct and the IQAROS spectra deconvoluted with method 2 from mixture II containing 1 and 6. a) shows the direct MS² spectrum of pure 1 in red and the direct MS² spectrum of mixture II in blue. b) shows the direct MS² spectrum of pure 1 in red and the IQAROS deconvoluted spectrum according to method 2 for compound 1 of mixture II in blue. c) shows the direct MS² spectrum of pure 6 in red and the direct MS² spectrum of mixture II in blue. d) shows the direct MS² spectrum of pure 6 in red and the IQAROS deconvoluted spectrum according to method 2 for compound 6 of mixture II in blue.

Figure S11. Comparison of the direct MS² spectra of the pure standards 1, 5 and 6 with the direct and the IQAROS spectra deconvoluted with method 2 from mixture III containing 1, 5 and 6. a) shows the direct MS² spectrum of pure 1 in red and the direct MS² spectrum of mixture III in blue. b) shows the direct MS² spectrum of pure 1 in red and the IQAROS deconvoluted spectrum according to method 2 for compound 1 of mixture III in blue. c) shows the direct MS² spectrum of pure 5 in red and the direct MS² spectrum of the mixture III in blue. d) shows the direct MS² spectrum of pure 5 in red and the IQAROS deconvoluted spectrum according to method 2 for compound 5 of mixture III in blue. e) shows the direct MS² spectrum of pure 6 in red and the direct MS² spectrum of mixture III in blue. f) shows the direct MS² spectrum of pure 6 in red and the IQAROS deconvoluted spectrum according to method 2 for compound 6 of mixture III in blue.

Figure S12. Comparison of the direct MS² spectra of the pure standards 1, 2, 3, 4, 5 and 6 with the direct and the IQAROS spectra deconvoluted with method 2 from mixture VI containing 1, 2, 3, 4, 5 and 6. a) shows the direct MS² spectrum of pure 1 in red and the direct MS² spectrum of mixture VI in blue. b) shows the direct MS² spectrum of pure 1 in red and the IQAROS deconvoluted spectrum according to method 2 for compound 1 of mixture VI in blue. c) shows the direct MS² spectrum of pure 2 in red and the direct MS² spectrum of mixture VI in blue. d) shows the direct MS² spectrum of pure 2 in red and the IQAROS deconvoluted spectrum according to method 2 for compound 2 of mixture VI in blue. e) shows the direct MS² spectrum of pure 3 in red and the direct MS² spectrum of mixture VI in blue. f) shows the direct MS² spectrum of pure 3 in red and the IQAROS deconvoluted spectrum according to method 2 for compound 3 of mixture VI in blue. g) shows the direct MS² spectrum of pure 4 in red and the direct MS² spectrum of mixture VI in blue. h) shows the direct MS² spectrum of pure 4 in red and the IQAROS deconvoluted spectrum according to method 2 for compound 4 of mixture VI in blue. i) shows the direct MS² spectrum of pure 5 in red and the direct MS² spectrum of mixture VI in blue. j) shows the direct MS² spectrum of pure 5 in red and the IQAROS deconvoluted spectrum according to method 2 for compound 5 of mixture VI in blue. k) shows the direct MS² spectrum of pure 6 in red and the direct MS² spectrum mixture VI in blue. l) shows the direct MS² spectrum of pure 6 in red and the IQAROS deconvoluted spectrum according to method 2 for compound 6 of mixture VI in blue.

Figure S13. Partial ion coalescence observed for the six isobaric precursors. a) shows an overlay of multiple measurements. In grey, the six individual measurements of the pure compounds are overlaid. At the bottom, the compounds exact mass is depicted as a tick. The deviation from the exact mass to the accurate mass (measured peak center) is listed as grey number. In green, the mixture II with isobars 1 and 6 is plotted. The mass deviation now is tilted towards their common center. The effect becomes more pronounced for mixture III. 1 and 6 are tilted towards the common center around 5. For six isobars, the effect exhibits the same pattern i.e. peaks with lower m/z than the common center are shifted to higher m/z and vice versa. This effect, known as partial ion coalescence, is known to distort peaks with orbitrap², Fourier transform ion cyclotron resonance³ and time‐of‐flight⁴ mass analyzers. While the shown spectra are measured on a MS¹ level, the same behavior is observable for MS². Consequently, ion coalescence can be problematic for the identification of isobars because the measured precursor mass deviates significantly from the exact mass of the true molecular formulae. For a complex mixture of isobars, it might coincidently happen, that one of the fragments form another precursor is consistent with the shifted m/z value of the precursor. This is presumably what happened in Table 1 for compounds 4 and 5, which experience significant m/z shifts due to partial ion coalescence. b) shows the same at high resolution. Consequently, the problem of erroneous mass accuracy can not be circumvented by a higher resolving power.

Kaeslin J, Zenobi R. Resolving isobaric interferences in direct infusion tandem mass spectrometry. Rapid Commun Mass Spectrom. 2022;36(9):e9266. doi: 10.1002/rcm.9266

DATA AVAILABILITY STATEMENT

The original data used in this publication are made available in a curated data archive at ETH Zürich (https://www.research-collection.ethz.ch) under the DOI https://doi.org/10.3929/ethz-b-000520528. There, the fully commented IQAROS deconvolution Matlab code is provided. Additionally, the modulation experiments of mixtures II and III are provided as exemplary.mgf files such that the user can familiarize with the processing.