In total, 165 compounds were incorporated into our assessment, and their CCS values (TWCCSN2) were determined following LC-IMS-MS analysis (SI, Table S5). The entire data set, covering an overall mass (m/z) and TWCCSN2 range of m/z 146–862 and 120–280 Å2 respectively, was utilized to compare mean measured (TWCCSN2, meas) with predicted CCS (TWCCSN2, pred) values (see “Correlation between TWCCSN2, pred and mean TWCCSN2, meas values”). Moreover, a reduced data set (n = 139) was further used to verify if the previously observed repetitive shifts in relative TWCCSN2, meas between parent drugs (or precursors) and their metabolites were biotransformation-specific. The selection and inclusion of those compounds for such assessment was conducted based on the possibility of creating compound pairs (n = 86; SI, Table S6) covering three common types of phase I biotransformation [hydroxylation (n = 19), N-oxidation (n = 9), demethylation (n = 13)] and five common phase II biotransformations [O-glucuronidation (n = 14), N-glucuronidation (n = 9), sulfation (n = 10), glutathione conjugation (n = 7), and acetylation (n = 5)]. Although most investigated compounds represented already marketed drugs and their corresponding metabolites, additional internal compounds were included for our assessment to extend the number of compound pairs for certain types of biotransformation. The only prerequisite for internal compound inclusion was to have synthesized reference material with a nuclear magnetic resonance-confirmed structure.

Reproducibility of TWCCSN2, meas values

Similar to other published data [22, 29,30,31], the generated analyte-specific TWCCSN2, meas values were highly reproducible over the entire period of our assessment (8 months) which was not only demonstrated with the reference mix used for system suitability testing (SI, Table S4 and Fig. S1) but also with the entire dataset (SI, Table S5). For the latter, the SD between individually determined TWCCSN2, meas values (up to n = 3) was maximum 1.8 Å2 resulting in a precision ≤ 1.0%.

Correlation between TWCCSN2, pred and mean TWCCSN2, meas values

After achieving excellent reproducibility in TWCCSN2, meas values for our entire data set (n = 165), a better understanding of its accuracy was required. For this, mean TWCCSN2, meas values were further compared with their TWCCSN2, pred values (Waters CCSonDemand) which overall resulted in a good correlation (R2 = 0.9594) as depicted in Fig. 1a. The obtained linear regression with a slope of 0.9036 (95% confidence intervals ranging from 0.8748 to 0.9323) and an intercept at + 18.14 (95% confidence intervals ranging from 12.49 to 23.78) was also close to its optimum (dashed black line). On the individual level, the obtained bias between TWCCSN2, pred and mean TWCCSN2, meas ranged from − 6.5 to + 11.9% (SI, Table S5) while the mean absolute error over the entire data set was 5.1%. The bias was within ± 10% (red area/dashed lines in Fig. 1) for 164 of the investigated compounds (99.4%). By further narrowing the acceptance criterion, most of the investigated compounds (87.3%, n = 144) met the ± 5% threshold (orange area/dotted lines in Fig. 1) while only 64.8% (n = 107) met the most stringent acceptance criterion of ± 3% (green area/solid lines in Fig. 1). It is worth mentioning that additional efforts were conducted to elucidate the origin of obtained bias in our presented data set. Neither the molecular weight (data not shown) nor an increase in TWCCSN2 could be identified as a root cause. The latter demonstrated an equal distribution in negative (n = 93) and positive bias (n = 72) over the investigated TWCCSN2 range (Fig. 1b). Hence, additional investigations would be necessary to clarify why certain TWCCSN2 values could not be predicted as accurately as other ones. Besides CCSonDemand, other computational tools or algorithms, previously reporting excellent correlations between CCSpred and CCSmeas values [32, 33], could further be tested with our publicly shared TWCCSN2, meas values (SI, Table S5 or the separately provided.csv file in the online resources). Such assessment, however, was out of scope for this study. Overall, a high confidence was associated with the generated data set which was used for our assessment of biotransformation-specific relative shifts in TWCCSN2, meas as discussed next.

Fig. 1

a Correlation between predicted (TWCCSN2, pred) and mean measured CCS values (TWCCSN2, meas) for 165 investigated compounds. b Bias distribution across investigated TWCCSN2 range. The colored areas or lines represent the deviation of ± 10% (red/dashed), ± 5% (orange/dotted), and ± 3% (green/solid) from the optimum (dashed black line)

Biotransformation-specific relative shifts in mean TWCCSN2, meas values

An almost linear relationship between shifts in mean TWCCSN2, meas values relative to the increase or decrease in mass was obtained for each type of investigated biotransformation (Fig. 2a). For instance, the elimination of a methyl group from the parent drug (− 14 Da) decreased the mean TWCCSN2, meas relative to the parent drug or precursor by - 6.5 ± 2.1 Å2 on average (Table 1). On the other hand, the mean TWCCSN2, meas relative to the parent drug or precursor increased by + 13.5 ± 1.9 Å2 on average following acetylation (+ 42 Da) while sulfation (+ 80 Da) increased the mean TWCCSN2, meas in comparison to the parent drug or precursor even further (+ 17.9 ± 4.4 Å2). Oxygenation as phase I biotransformation, including hydroxylation (+ 3.8 ± 1.4 Å2) and N-oxidation (+ 3.4 ± 3.3 Å2), also perfectly fitted into the almost linear relationship (Fig. 2a) which agreed with other published data [23]. However, both phase I biotransformations could not be differentiated from each other purely based on IMS-MS data since similar relative mean shifts in TWCCSN2, meas were observed (Fig. 2b). Fortunately, other experiments such as hydrogen–deuterium exchange [34, 35] or selective reduction of N-oxides to amines with titanium (III) chloride [36] enable the discrimination between both types of biotransformation. Secondary (n = 3), tertiary (n = 4), and quaternary N-glucuronides (n = 2) included in our assessment exhibited on average a slightly higher relative mean shift in TWCCSN2, meas (+ 41.7 ± 7.5 Å2) compared to the investigated O-glucuronides (+ 38.1 ± 8.9 Å2) with almost identical SDs (Table 1 and Fig. 2c). On some occasions, a much more discriminative relative shift in TWCCSN2, meas values was obtained when the parent drug was either O- or N-glucuronidated (see “Potential to elucidate structural relationships between metabolites”) while the tendency of N-glucuronides exhibiting higher relative shifts in TWCCSN2, meas values compared to O-glucuronides remained constant.

Fig. 2

a Correlation between mean measured mass changes and relative mean TWCCSN2, meas shifts. Zoom into b hydroxylation and N-oxidation as well as c O- and N-glucuronidation. d Example for TWCCSN2 calculation (grey italics values) based on the obtained polynomial regression from a in comparison to TWCCSN2, meas values (black values)

Table 1 Obtained mean m/z and relative mean TWCCSN2 shifts based on measured or predicted CCS values

It was further observed that the variation in relative mean shifts in TWCCSN2, meas values was significantly increased for phase II compared to phase I biotransformations which was somehow expected: simple modifications following phase I reactions typically introduce relatively small changes in the molecular structure of the parent drug resulting in fairly consistent shifts in relative TWCCSN2, meas values (except for major dealkylations within the molecule). On the other hand, conjugation of a large and rather flexible residue such as glutathione (tripeptide, + 305 Da) alters the molecular structure significantly which resulted in a mean relative TWCCSN2, meas increase by + 49.2 ± 13.2 Å2 on average (Table 1). Two main factors were identified as a potential origin of the increased variation in relative mean TWCCSN2, meas shifts following phase II biotransformation: (i) bulky entities could exist in different gas phase conformations and (ii) the site of conjugation can further influence changes in relative mean TWCCSN2, meas values. For instance, glucuronic acid conjugated to a flexible aliphatic side chain could have a significantly different TWCCSN2, meas compared to its isobaric version where conjugation occurs on a rather rigid aromatic ring system (see example in “Potential to elucidate structural relationships between metabolites”).

Nonetheless, almost non-overlapping bands in relative mean TWCCSN2, meas shifts were obtained for each type of investigated biotransformation (Fig. 2a and Table 1). The plateau towards the upper end of the correlation between the mean measured mass and relative mean TWCCSN2, meas shift was most likely caused by the underestimation of the TWCCSN2, meas values for the glutathione conjugates which rather tend to be multiply charged instead of singly charged. It is noteworthy mentioning here that IMS and the CCS value of a molecule strongly depend on its charge state [29, 37]. For our work, however, the TWCCSN2, meas of a singly charged ion from a metabolite had to be compared with the one of its parent drug (see also “Limitations”). The obtained relative mean TWCCSN2 shifts based on TWCCSN2, meas values further agreed with the theoretical relative shifts when taking TWCCSN2, pred values into consideration (Table 1 and SI, Fig. S2).

Based on the obtained polynomial regression (y =  − 0.0005x2 + 0.3202x − 1.4506) between the mean measured mass changes and relative mean shifts in TWCCSN2, meas (Fig. 2a), TWCCSN2 values of metabolites could readily be calculated which is illustrated for an internal compound in Fig. 2d: the TWCCSN2 of the parent drug (INE963) following hydroxylation was calculated 207.2 Å2 (TWCCSN2, calc) which agreed with the actual TWCCSN2, meas for metabolite M13 (207.7 Å2). Even after subsequent N-acetylation and demethylation of M13, resulting in metabolite M8, the TWCCSN2, calc (215.4 Å2) was in excellent agreement with the TWCCSN2, meas (216.0 Å2). Hence, the discovered biotransformation-specific relative shifts in mean TWCCSN2, meas values (and the combination thereof) can also be considered for metabolite assignment/confirmation (rather than their absolute TWCCSN2, meas values) complementing the conventional approach where changes in m/z values are compared. Only for glucuronidation or glutathione conjugation, special care must be taken exhibiting increased variabilities and slightly overlapping bands in relative mean TWCCSN2, meas shifts (Fig. 2a and Table 1).

Similar biotransformation-specific shifts in relative TWCCSN2, meas obtained on fragment ion level

Identical biotransformation-specific shifts in relative TWCCSN2, meas were also obtained on fragment ion level as illustrated for hydroxylation, demethylation, and N-acetylation in Fig. 3. Phase II biotransformations such as sulfation and glucuronidation were not considered for this assessment since the major fragment ion from such metabolites is often only the parent drug or corresponding precursor following the characteristic neutral loss of the sulfate (80 Da, SO3) or the glycone (176 Da, C6H8O6), respectively [38,39,40,41]. One prerequisite for the assessment of biotransformation-specific shifts in relative TWCCSN2, meas on fragment ion level was the presence of common fragment ions between the metabolite and parent drug or precursor (with and without the biotransformation). The relative shift in TWCCSN2, meas for mephenytoin after hydroxylation, leading to 4-hydroxy mephenytoin, was + 4.1 Å2 on precursor ion level (Fig. 3a). An almost identical relative shift (+ 3.4 Å2) was obtained after comparing the TWCCSN2, meas values of the major fragment ion for mephenytoin at m/z 134 and 4-hydroxy mephenytoin at m/z 150 (Fig. 3b). For demethylation, the relative shifts in TWCCSN2, meas between diazepam and nordazepam on precursor and fragment ion level were again in the similar range: − 4.6 Å2 on precursor ion level (Fig. 3c) and − 3.8 Å2 on fragment ion level (Fig. 3d) considering the fragment ions at m/z 222 and m/z 208 for diazepam and nordazepam, respectively. The best agreement in biotransformation-specific relative shifts in TWCCSN2, meas determined on precursor (+ 8.1 Å2, Fig. 3e) and fragment ion level (+ 8.0 Å2, Fig. 3f) was obtained with an internal compound (NVS1) and its N-acetylated metabolite M7. The combination of fragment ion TWCCSN2, meas values, allowing the discrimination of ortho-, para-, or meta-conjugated metabolites due to minor changes in TWCCSN2, meas values [16] and the potential to utilize relative shifts in TWCCSN2,meas values for metabolite mapping purposes (see “Potential to elucidate structural relationships between metabolites”), would represent a powerful analytical tool for future metabolism studies to elucidate structural relationships between positional isomers and their subsequently conjugated metabolites, e.g., glucuronides.

Fig. 3

Biotransformation-specific shifts in relative TWCCSN2, meas values obtained either on precursor ion (a, c, e) or fragment ion level (b, d, f) exemplified for hydroxylation, demethylation, and N-acetylation (proposals for charge localization and fragment ion structures are only tentative)

Potential to elucidate structural relationships between metabolites

One major benefit to considering relative shifts in TWCCSN2, meas values for biotransformation studies is the potential to elucidate structural relationships between metabolites in order to design metabolic pathways. During in vivo metabolism assessment of an internal compound (NVS1), several hydroxylated and glucuronidated metabolites were detected in pooled plasma samples (AUC0-24 h) following oral (rat) and intravenous administration (dog), respectively. In both studies, conventional MS fragmentation did not assist in structure elucidation and made the localization of the biotransformation impossible since similar product ion spectra were obtained (data not shown). In contrast, relative shifts in TWCCSN2, meas values provided hints about their structural relationships (Fig. 4). The direct O-glucuronide (M2) was conjugated to a rather flexible side chain whereas the N-glucuronide (M4) was conjugated to a rigid aromatic ring system. Consequently, both direct glucuronides exhibited different TWCCSN2, meas values (234.3 Å2 for M2 and 254.1 Å2 for M4). By knowing that hydroxylation of NVS1, leading to metabolite M1, increased the TWCCSN2, meas by + 6.2 Å2 relative to the parent drug, structural relationships of the detected hydroxylated and glucuronidated metabolites (M9, M10 and M11) relative to both direct glucuronides could readily be elucidated: the differences in TWCCSN2, meas of M10 (+ 25.2 Å2) and M11 (+ 25.4 Å2) were too high to originate from M2 (O-glucuronide). On the other hand, when the TWCCSN2, meas of M10 (259.5 Å2) and M11 (259.7 Å2) were compared to the N-glucuronide (M4), then the relative shifts in TWCCSN2, meas values matched with the expected one for NVS1 hydroxylation (+ 5.4 Å2 for M10 and + 5.6 Å2 for M11). Consequently, M10 and M11 were most likely structurally related to M4. On the other hand, M9 was most likely structurally related to M2.

Fig. 4

Potential of relative shifts in TWCCSN2, meas values to elucidate structural relationships between metabolites as illustrated with an internal compound (NVS1) and its in vivo obtained hydroxylated and glucuronidated metabolites. Please note that the displayed mapping is based on relative changes in TWCCSN2, meas and not on enzymatic formation (hydroxylation with subsequent glucuronidation)

Limitations

Despite the above-highlighted benefits to consider relative shifts in TWCCSN2, meas values for biotransformation assignment/confirmation and metabolite mapping purposes, several limitations are also associated with the presented methodology: (i) biotransformation-specific relative shifts in TWCCSN2, meas values could only be obtained when TWCCSN2, meas values of the same ion species were compared with each other. It is not recommended to compare the TWCCSN2, meas of a protonated parent drug with one of its deprotonated metabolites. Moreover, TWCCSN2, meas comparison of a protonated parent drug with the sodium or potassium adduct of its metabolite will neither be successful. (ii) So far only singly charged ions were considered for our assessment. Nevertheless, this already allowed the comparison of a broad range of parent drugs/precursors with their corresponding metabolites since low molecular-weighted drugs with a mass of up to 500–600 Da tend to be mainly singly charged. For future activities, the proposed concept of biotransformation-specific relative shifts in TWCCSN2, meas values may also need to be verified for multiple charged ions. (iii) The current data set is still rather limited and needs to be further extended, e.g., by either adding more compound pairs to already investigated types of biotransformation or by incorporating additional types of biotransformation such as dealkylation reaction causing a significantly greater loss in mass.