Application of phase correction to improve the characterization of photooxidation products of lignin using 7 Tesla Fourier-transform ion cyclotron resonance mass spectrometry

Methanol, acetonitrile, ammonium hydroxide, and alkali lignin powder were purchased from Sigma-Aldrich (Steinheim, Germany). The lignin powder was dissolved in a methanol–acetonitrile–ammonium hydroxide mixture (50:50:1 v/v/v) prior to analysis. Organic-free water was generated using a Millipore (Bedford, Massachusetts, USA) Direct-Q8 purification system.

Photooxidation experiments were performed using a custom-built oxidation chamber (see Fig. S2) consisting of a light source array of 8 white light-emitting diodes (LED; λ = 400–800 nm; λ maximum at 620 nm; luminous intensity 7.8 cd at 65 mA; luminous flux 23.3 lm at 65 mA), 10 blue LED (λ = 410–540 nm; λ maximum at 465 nm; luminous intensity 2.5 cd at 20 mA), and 5 UV LED (λ = 355–420 nm; λ maximum at 385 nm; radiant flux 350 mW at 500 mA), which simulated the intensity and spectral profile of the solar radiation on the earth’s surface (more technical details of the implemented LEDs are given by Nichia Corp. (2017)). Two lignin samples were used in the photo-degradation experiment: a control sample, and a solar-irradiated sample that was kept in the oxidation chamber for 14 d. For the control, the lignin powder was placed in a sealed conical flask that was wrapped in aluminum foil to prevent exposure to light.

Samples were analyzed using electrospray ionization in negative ion mode. Mass spectra were recorded using a 7 T FTICR-MS instrument (Bruker, Bremen, Germany), equipped with an Infinity Cell (Caravatti and Allemann 1991). For each mass spectrum, 100 individual transients (duration of 1.89 s) were collected and co-added to enhance S/N (Marshall and Verdun 1989).

Broadband phase correction of the mass spectra was performed using the Bruker FTMS processing software version 2.1.0 (a screenshot of the instrument settings is shown in Fig. S3). Both magnitude mode and absorption mode spectra were analyzed using a combination of DataAnalysis 4.2 (Bruker) and Composer 1.5.0 (Sierra Analytics, Modesto, California, USA). The full-scan mass spectra were internally calibrated throughout the m/z range from 121 to 927 using a series of homologous lignin compounds consisting of well-characterized building blocks. Elemental formulae were only assigned to the peaks inside the calibrated m/z range, with the following constraints for the lignin compounds: composition was restricted to C, H, and O; H/C ratios from 0.6 to 2, O/C ratios from 0 to 0.9, and double bond equivalents (DBE) from 4 to 30; the acceptable mass error was set to ±1 ppm for singly-charged ions.

When a Fourier transformation is performed for a finite transient recorded by FTICR-MS, it will generate “tails” for the peaks, the extents of which vary depending on the damping mode of the signal. These tails can interfere with adjacent low-intensity peaks (Marshall et al. 1979). For this reason, the time-domain transient is often multiplied by a window function prior to Fourier transformation to smooth peak shapes and to minimize the sideband intensities (Marshall and Verdun 1989). This procedure is called “apodization”. An optimal window function removes the sidebands at a cost of resolving power and S/N (Qi et al. 2013). However, the success of apodization depends on the transient acquisition time and its damping constant; detailed investigations have shown that its effect varies significantly from sample to sample and from spectrum to spectrum (Kilgour and Van Orden 2015). In principle, it has been shown that a full-window apodization is optimal for the magnitude mode (Lee and Comisarow 1987), whereas a half-window is suitable for the absorption mode (Lee and Comisarow 1989). In general, this practice is widely accepted to optimize the different mass spectra obtained from FTICR-MS (Beu et al. 2004; Xian et al. 2012), in particular, in petroleomics applications (Xian et al. 2010; Cho et al. 2014). In this study, lignin degradation provided very complex mixtures consisting of tens of thousands of species with variable relative abundances (Reale et al. 2004; Owen et al. 2012), similar to petroleum compounds. Hence, the magnitude and absorption mode spectra acquired were apodized based on the above rule to achieve optimum performance.

Direct analysis of complex mixtures—without chromatographic pre-separation—is commonly performed with FTICR or orbitrap mass spectrometers, as these instruments provide much higher resolving powers than instruments such as quadrupole-time-of-flight-MS. In crude oil analyses, the petroleum compounds appear regularly in the mass spectrum according to the heteroatom, hydrogen deficiency, and DBE (Rodgers and McKenna 2011). For lignin, in contrast, no universal methodology has yet been proposed for classification; however, several strategies have been reported to identify the characteristic lignin components. For example, the classic Kendrick mass defect (KMD) plot was modified to quickly characterize the different linkages in the lignin oligomers (Qi et al. 2016b). These two-dimensional (2D) matrix plots provided systematic lineups of the different lignin linkages using structure-specific KMD bases, and higher lignin oligomer structures originating from the same linkages were immediately visualized in these matrix plots.

In the present work, broadband mass spectra of lignin were acquired using a 7 T FTICR-MS instrument. Mass spectra were then plotted for both magnitude and absorption modes. The original FTICR-MS signal is a composite signal of all sinusoidal waves of the frequencies of different ions. The sinusoidal waves are Fourier transformed to extract the frequency information from the recorded signal, followed by calibration to obtain the m/z domain. Unfortunately, Fourier transformation produces a complex output expressed in polar terms—magnitude and phase (Comisarow and Marshall 1974; Marshall and Roe 1978). Therefore, the projection of the frequency spectrum ranges from positive to negative as a result of the phase. To overcome this problem, the vector sum is calculated to yield a phase-independent magnitude-mode spectrum, which is displayed by most commercial and custom-made FTICR instruments (Marshall and Verdun 1989). However, the absorption-mode spectrum can be obtained if the phase shift correction is possible. The absorption-mode display offers superior mass resolving power (up to a factor of 2), mass accuracy, and sensitivity over the conventional magnitude mode (Qi et al. 2012a; Kilgour et al. 2013).

In the magnitude mode display, the average mass resolution (full-width at half maximum; FWHM) of the peak at m/z 400 was about 285 000 (Fig. 1), whereas the absorption mode spectrum demonstrated a resolution of ∼485 000, together with improved S/N of the same order. For low magnetic field FTICR-MS, the resolving power is often not sufficient to resolve the critical mass differences of 3.4 mDa (SH₄ vs. C₃) and 4.5 mDa (CH vs. ¹³C). When m/z exceeds 500, this can lead to incorrect assignments of elemental composition to mass spectral peaks and discontinuous DBE vs. carbon number plots (Cho et al. 2014). Fortunately, after phase correction, the absorption mode spectra provided enhanced spectral resolution to at least partially address this problem, as outlined in the following demonstration.

Lignins are cross-linked molecules containing various characteristic chemical linkages, and consequently, lignin molecules with unique functional groups appear at predictable regular positions relative to each other in the mass spectra (Qi et al. 2016b). In our experiments here, we used 2D KMD plots (Kendrick 1963) to recognize series of lignin compounds that contain a dibenzyl ether (C₇H₇O) linkage and methoxylation (OCH₃) throughout the entire mass spectrum (Fig. S4); m/z windows of ∼10 mDa were expanded for the chosen compounds to compare the performance of the two mass spectral modes. As shown in Fig. 2, at low m/z, the elemental formula of compound 1 was calculated to be C₁₆H₁₂O₈, with a small adjacent isobaric signal (compound 2, C₁₃H₁₆O₈S) at 3.4 mDa distance (SH₄ vs. C₃). Because the m/z values of these two compounds were relatively low, the two peaks were successfully resolved in the magnitude mode spectrum. In the corresponding absorption mode display, the S/N of compound 2 increased sharply, and this improvement reduced the mass error of the assignment from 0.78 to −0.13 ppm (Fig. 2), even though the additional resolving power did not provide any new information in this particular example. In general, the benefits of the absorption mode display are less pronounced for compounds of low m/z. At higher m/z, however, due to the decrease of mass resolving power in FTICR and the large increase of possible elemental compositions, the improvements from phase correction become significant. This is evident in our lignin analyses, where a series of dibenzyl ether-linked polymers with increasing m/z values were identified in the mass spectra, starting from the isobaric species C₁₆H₁₂O₈ and C₁₃H₁₆O₈S (C₃ vs. SH₄). As illustrated in Fig. 2, the resolution of the two isobaric peaks C₃ and SH₄ decreased with increasing m/z. At m/z 679, the two close mass doublets (C₃₈H₃₂O₁₂ and C₃₅H₃₆O₁₂S) were only partially resolved in the magnitude mode display, whereas in absorption mode, they were almost baseline resolved, accompanied by a significant improvement of mass accuracy. This was generally observed in our lignin experiments; the phase correction method often resolved very close peaks that remained unresolved in the conventional magnitude mode mass spectra, along with greatly increased mass accuracy and a resulting lower number of possible elemental composition assignments in the lignin mass spectra obtained from low magnetic field FTICR-MS. To further demonstrate the magnitude/absorption mode differences, DBE vs. carbon number plots of the O₇ lignin compounds class (C _x H _y O₇) were plotted from both magnitude and absorption mode mass spectra (Fig. 3). It was immediately obvious that the dots drawn from the magnitude mode data exhibited gaps in the plot, and the chemical composition was not distributed continuously, especially for species with higher carbon numbers, because the mass resolving power of the 7 T FTICR-MS was not sufficient to resolve the peaks at higher m/z. Very likely, some of the neighbouring peaks (e.g., SH₄ vs. C₃, Fig. 2) merged; thus, only one of the compounds could be identified in the magnitude mode spectrum. The dots generated from the absorption mode display, in contrast, exhibited a more continuous pattern (Fig. 3, bottom).

In addition to the improved spectral quality, it was also interesting to discover that the low m/z region (<200) of the absorption mode spectrum consisted of a variety of signals with negative intensities (Fig. 1, inset). A closer look at the spectrum revealed that these signals were also present in the magnitude mode spectra as regular mass spectral peaks. In fact, the negative signals were due to spectral artefacts from harmonic signals, electronic noise, or radio-frequency interference, which are common in some mass spectrometers (e.g., FTICR, orbitrap, and ion trap) (Mathur and O’Connor 2009; Qi et al. 2013). Unfortunately, these artefacts are unavoidable mass spectral features; they contain no chemical information but can seriously complicate data interpretation. For very complex mass spectra, these artefacts can overlap with real peaks, cause wiggles in the baseline and deteriorate the S/N of the spectra, and thus, have the potential to give false positive assignments. Normally, identifying such artefacts requires a comparison and correlation of the peaks in question to higher m/z primary signal peaks (Qi and O’Connor 2014); however, such a procedure would be time consuming and challenging for the complex spectra of lignin. As an example, in Fig. 1 the third harmonic region is expanded in the lignin mass spectrum. In the conventional magnitude mode the harmonics appear as typical peaks, whereas in absorption mode all artefacts could be directly recognized because artificial signals cannot be corrected properly. This ease of identifying artefacts is an important additional advantage of absorption mode spectra, in addition to the advantages described above. Thus, all lignin characterization experiments described in the following section were performed in absorption mode.

The photooxidation study performed here consisted of simulated solar irradiation of the lignin sample and comparison of the measured product distribution using absorption mode FTICR to an identical control sample that was not exposed to the light source. The simulated solar irradiation treatment was terminated after 14 d, and no visible physical differences were observed between the two sample sets. First, a van Krevelen diagram (van Krevelen 1950) was plotted to visualize the transformation of lignin compounds on a broader scale (Fig. 4). It is immediately obvious that the range of O/C ratios for the lignin compounds extends much farther in the solar-irradiated sample, as illustrated by the product distribution shifts on the x-axis (O/C ratio). This readily demonstrates the formation of new oxygenated species during simulated solar irradiation. In contrast, the y-axis exhibited similar H/C ratio ranges for the two samples. As the H/C ratio directly corresponds to the number of DBE of the lignin compounds, the unchanged H/C ratio suggests that the degree of unsaturation and the core structure of lignin compounds were likely not influenced by the solar light. The newly formed compounds were mostly located in two areas of the van Krevelen diagram: (1) O/C > 0.7, indicating addition of oxygen; and (2) O/C < 0.2 and H/C < 1, corresponding to aromatic compounds at low m/z resulting from photo-induced degradation reactions. A similar behavior was previously reported for lignin and crude oil compounds under UV irradiation (Islam et al. 2013; Qi et al. 2016a) and was explained by UV-triggered radical ion fragmentation of the benzene compounds. It was interesting to observe that the simulated solar irradiation produced far fewer fragment compounds as compared with the UV degradation experiment (Fig. S5). This was probably because the simulated solar light used here exhibited much longer wavelengths compared with the UV light of the previous study and, hence, lacked energy to excite the ions for fragmentation. In addition, some peaks detected in the low m/z region of the mass spectrum were artificial signals generated by electronic noise or harmonic peaks, which caused false positive assignments in the magnitude mode mass spectrum previously applied. Here, these noise artefacts were directly eliminated from the absorption mode display.

In addition to the van Krevelen diagrams, bar charts were used to visualize the different oxygen compositions of the lignin samples (Fig. 5). The oxygen number distribution shows that the abundance of compounds with two to five oxygen atoms decreased sharply and shifted to classes with higher O content (O_6–10). That is, the compounds with low oxygen content were either degraded or oxidized during solar irradiation. Furthermore, compounds with more than 10 oxygen atoms in their structures appeared to be less reactive under photooxidation conditions, as the change of abundance was not significant. This phenomenon was probably caused by the photo-stability of the conjugated π-systems in lignin. Compounds with a low oxygen number are likely lignin monomers with a simple aromatic ring structure and methoxy groups; hence, they have a strong tendency to undergo rapid excited electronic state quenching by photooxidation, which results in either compound fragmentation or oxidation (Ashfold et al. 2010). In contrast, lignin compounds with a higher oxygen number are obviously polymers, likely consisting of several adjoined aromatic rings. Such conjugated π-systems can readily redistribute vibrational energy over the entire system and stabilize their core structures.

The result of simulated solar irradiation can also be observed on a micro-scale, from individual peaks of expanded segments of the mass spectra. Figure 6 shows a ∼0.25 m/z segment of the two samples, allowing 11 and 19 peaks to be assigned with unique elemental formulae in the two spectra. As lignin is composed of only three elements(C, H, and O), only six species in the control and nine species in the oxidized sample were confirmed to be lignin compounds. The other peaks were not related to lignin as they contained nitrogen or sulphur. This finding agrees with our previous oxidation experiment using UV light (Fig. S6). Several other abundant low peaks were detected in comparison with the previous UV-oxidation experiments, which was due to the enhanced capabilities of the absorption mode mass spectral analysis used in the present study, offering greatly improved S/N and mass resolution over the previous study. Nevertheless, the new lignin compounds generated by the simulated solar irradiation experiment were consistent with the compound distribution from UV experiments, indicating very similar oxidation pathways. In the solar-irradiated FTICR spectra, the overall peak abundance shifted to species with lower mass defects, due to the negative mass defects of oxygen, except for three new lignin compounds with much higher O content that were observed here.