Pushing the analytical limits: new insights into complex mixtures using mass spectra segments of constant ultrahigh resolving power

A new strategy has been developed for characterization of the most challenging complex mixtures to date, using a combination of custom-designed experiments and a new data pre-processing algorithm.


Sample details
Elemental analysis of the distillable fraction, non-distillable fraction, and the vacuum residue (D-F, ND-F, and VR, respectively) can be found in Table S1. According to the API gravity (American Petroleum Institute), the fractions D-F and ND-F can be classified as heavy and extra-heavy, respectively.

Distillable fraction.
A sub-fraction of a South American vacuum residue was obtained by supercritical fluid extraction (SFEF). The supercritical solvent was n-hexane and the conditions were a temperature of 265 °C and pressure of 4.5 MPa. The SFEF process has been described elsewhere. 1 As seen in Fig. S1, 90% of the constituents of these fractions have boiling points lower than 720 °C atmospheric equivalent temperature (AET).
Truly non-distillable fraction. A short path distillation unit operating at 687 °C AET was used to obtain a truly non-distillable fraction. According to the true boiling point curve (Fig.   S1), only 18% of its constituents evaporated below 720 °C AET. Thus, 82% of the constituents S-3 had boiling points with an unknown final boiling point; the heaviest components may not be vaporizable. 2 Considering the extraordinarily high complexity of this sample, the sample was fractionated by solubility in n-heptane (maltenes), which corresponded to 89% of the sample (51% resins, 33% aromatics, and 1% saturates). The maltene fraction was removed from the non-distillable sample by Soxhlet extraction using n-heptane for a period of 12 h.

Table S1
Elemental analysis of the South American vacuum residue, distillable fraction, and nondistillable fraction. Density was measured by means of a digital density meter (ASTM D4052), 3 SARA composition analysis by ASTM D2007, 4 where N and 8 9 are experimental parameters: the data set size and the sampling frequency, respectively. In accordance with the Nyquist theory, the sampling frequency must be twice the frequency of the lowest m/z being detected; hence, it determined by the low m/z cut-off values of the mass spectra. 9 Under ideal experimental conditions (e.g. no collisions with residual gas, ideal electric and magnetic fields, and no space-charge effects), FT-ICR mass spectrometers of lower magnetic field can reach the same resolving power as higher field instrumentation if the duration of the time domain signal is sufficiently high to compensate for the lower magnetic field (see Equation (1) and Table S2). For instance, a resolution of approximately one million (FWHM) can be obtained at m/z 850 using a 7 T FT-ICR with an acquisition time of 10.388 s at 4M, while the same resolving power could be obtained using a 12 T instrument if the transient lasts 6.060 s. It can be challenging, however, to produce longer transients, particularly due to spacecharge effects, and so higher field instrumentation offers an advantage. According to Equations (1) and (2), the resolving power can be increased by increasing the low m/z cut-off for detection. For instance, as can be seen in Fig  The low m/z cut-off and sampling frequency per segment of the stitched spectra are reported in Table S3 for both the D-F and ND-F samples.

Table S3
Low m/z cut-off, frequency, and acquisition time (1 23# ) per window of the data for the ND-F and D-F samples Segment center

ND-F (8 M data sets) D-F (4 M data sets)
# Low m/z cut-off

Rhapso
A flowchart of the steps required to stitch the data using the in-house software "Rhapso" can be seen in Fig. 1 in the main text. An example of the result of step 1 is illustrated in Fig. S4, with a before and after trimming of the segment between m/z 557-580. After further steps, Fig.   S5 shows the intensity correction for the segment m/z 595-616; mass spectra resulting from traditional broadband acquisition, stitching without intensity correction, and the OCULAR method could then be compared.

Distillable sample: D-F
As a result of the ultrahigh resolution of the stitched spectra and the improved dynamic range, the number of peaks assigned were increased by three-fold compared with the broadband absorption mode mass spectrum (4.6-fold compared with the broadband magnitude mode mass spectrum, see Fig. S6). The assigned peaks were plotted as contributions by heteroatomic class, CcHhNnOoSs, (Fig. S7). Compound classes with the label "[H]" denote protonated species, which those classes without the label were observed as radical ions. According to this Figure, approximately twice as many radical ions were observed in the stitched mass spectrum and approximately twice as many compound classes became accessible e.g. OS2, O3, N2O, N2, N4OV (see Fig. S7, among others). It is worth noting that the relative contributions of radical ions versus protonated species are influenced by experimental parameters, such as sample S-13 infusion rate. 10 The total number of peaks assigned were significantly increased for each class, although the relative abundance (in percentage) of the classes remained similar.

S-14
As can be seen in Fig. S8  The mean resolving power per narrow window is almost constant (average of 1,792,000 FWHM) over the full mass range (see Fig. S10). As shown in Fig. S11 and Table S5, about the half of the assigned peaks correspond to the isotopologues that were resolved and assigned with low ppm error (<0.5 ppm) in the mass spectrum, even at higher masses (Fig. S12). Thus, the isotopic contributions from 13 C4, 34 S2, and 13 C 34 S assist in the correct assignment of molecular formulae. S-17 Fig. S10 Comparison of the average resolving power as a function of m/z for the D-F sample using: a broadband mass spectrum using magnitude mode, a broadband mass spectrum using absorption mode ("phased"), and a corrected stitched data set using absorption mode (the OCULAR method), using the mean resolution of each segment.
Fig. S11 Mass distribution of a homologous series (17 DBE) for the N4OV class found in the OCULAR data for the stitched spectrum of the D-F sample, with a relative abundance of only 0.01% (see Figure S8). The confidence in the assignment is based on the low mass errors (horizontal bars in red) and the isotopic distribution.
S-18 Table S5 Isotopic number of peaks detected in broadband and OCULAR method for the D-F sample (not detected denoted by "-").  To illustrate that ultrahigh resolution is also necessary to improve mass accuracy, a comparison of a zoom-in of the mass distribution is shown in Fig. S13. At m/z 637, a resolution of 519,000 FWHM is not enough to baseline resolve and assign molecular compositions to peaks with a mass difference of 1.10 mDa, corresponding to a difference of 12 C4 vs 13 CH3S. At higher masses, such as m/z 793, peaks with this same mass difference are no longer resolved and coalescence of the peaks is observed. In this particular case, both assignments are possible within the tolerable mass error. Only after attaining the ultrahigh resolution achieved in the OCULAR data could it be demonstrated that both assignments were valid.

S-26
Table S6 Approximate number of peaks detected with smallest mass separations (below 3.4 mDa and rounded to two decimal places) specified below for the ND-F sample; the differences therefore count each peak once, instead of allowing a peak to have multiple mass separations. The asterisk denotes mass differences reported in previous works. 13