Identification of drug metabolites with infrared ion spectroscopy – application to midazolam in vitro metabolism

The identification of biotransformation products of drug compounds is a crucial step in drug development. Over the last decades, liquid chromatography-mass spectrometry (LC-MS) has become the method of choice for metabolite profiling because of its high sensitivity and selectivity. However, determining the full molecular structure of the detected metabolites, including the exact biotransformation site, remains challenging on the basis of MS alone. Here we explore infrared ion spectroscopy (IRIS) as a novel MS-based method for the elucidation of metabolic pathways in drug metabolism research. Using the drug midazolam as an example, we identify several biotransformation products directly from an in vitro drug incubation sample. We show that IR spectra of the aglycone MS/MS fragment ions of glucuronide metabolites establish a direct link between detected phase I and phase II metabolites. Moreover, using quantum-chemically computed IR spectra of candidate structures, we are able to assign the exact sites of biotransformation in absence of reference standards. Additionally, we demonstrate the utility of IRIS for structural elucidation by identifying several ring-opened midazolam derivatives formed in an acidic environment.


Introduction
Drug compounds undergo a range of biotransformations in the body before they are excreted.The nature and kinetics of each of the metabolic steps are important characteristics of a drug candidate, determining its clearance rates and potential toxicity.4][5][6][7] However, the structural elucidation of the MS-detected metabolites remains challenging, as there are usually multiple isomers to consider.The type of metabolic reaction is often straightforward to infer from the mass difference between the metabolite and the drug molecule; for instance, hydroxylation corresponds to +16 u.However, determining the exact transformation site in the drug molecule is more challenging, as regioisomers tend to have very similar MS/MS fragmentation patterns that are also difficult to predict without reference standards. 8- 10Final elucidation of metabolite structures often relies on time-consuming isolation of the metabolite from the sample matrix, followed by NMR analysis, and/or comparison to synthetic reference standards, which are usually not commercially available.
To accelerate the metabolite identification process, there is an ongoing search for fully MS-based techniques that provide structural information on MS-detected ions.Among these technologies, infrared ion spectroscopy (IRIS) [11][12][13][14][15] is a promising candidate, as it provides IR spectra for massisolated m/z features.Vibrational resonances are highly sensitive to details of the molecular structure, such that closely related isomeric molecular ions can often be differentiated on the basis of their IR spectra.Indeed, a range of studies have shown that IRIS can distinguish isomeric small molecules [16][17][18][19][20][21][22][23][24][25][26][27][28] , including several phase-I drug metabolites [29][30][31] .A significant advantage in comparison to techniques based on MS/MS fragmentation patterns is that gas-phase IR spectra can be reliably and efficiently predicted using quantum-chemical calculations based on density functional theory (DFT).Therefore, molecular structures can often be elucidated in a reference-free manner; for final confirmation or quantification purposes, only a single synthetic standard is then required. 18,32 vious work has demonstrated IRIS-based differentiation of ortho-, meta-and para-isomers that may result from the hydroxylation of drug compounds containing a phenyl moiety. 29,31 n one of these studies, metabolites were extracted and analyzed from an in vitro drug incubation sample using a combination of LC-MS and IRIS. 29However, a drug molecule often undergoes several phase I (oxidation, hydrolysis or reduction) and phase II (conjugation) biotransformation reactions that occur sequentially.Elucidating a metabolic pathway involves identification of each of the (intermediate) metabolites and their place in the metabolic pathway.To explore the potential of IRIS for the mapping of full metabolic pathways, we demonstrate here the identification of several metabolites resulting from the incubation of midazolam in human hepatocytes.4][35][36] Here, we detect several MDZ metabolites using LC-MS and elucidate their full molecular structures in absence of reference standards, using a combination of MS/MS fragmentation patterns and comparison of experimental IR spectra to DFT computed reference spectra.Glucuronated drug molecules readily expel the glucuronide-group upon collisional activation in an MS/MS experiment, challenging the identification of the site of glucuronidation.Here, we use this feature of glucuronides by recording IR spectra for both the glucuronated precursor ion and its glucuronide-loss MS/MS fragment (aglycone).This allows us to suggest a metabolic link between phase I and phase II metabolites and to narrow down the list of candidate structures for the glucuronide metabolite.Subsequently, the location of the glucuronide group can be determined from the precursor ion IR spectrum.MDZ is well-known to undergo a ring-opening in an acidic environment. 34,37 s a further demonstration of the potential of IRIS for the identification of drug derivatives, we identify several ring-opened MDZ derivatives.

Sample preparation.
MDZ was incubated at 10 µM for 60 and 120 min with primary human hepatocytes in suspension at 0.05% final DMSO and 0.60% final ACN concentration.Control samples were prepared by quenching the metabolism at 0 min and by performing an incubation at 0 µM followed by spiking after quenching.The final hepatocyte density was 10 6 cells/ml.Two sample preparation procedures were performed.The first approach involved quenching 40 µl incubation volume with two volumes of DMSO.The second involved quenching the same amount of incubation sample with two volumes of ACN/HCOOH 90:10 [v/v].In both cases the sample was centrifuged for 10 minutes at 4000 rpm.The supernatant was used for LC-MS and LC-IRIS analysis.

Liquid chromatography-mass spectrometry
LC-MS experiments were performed using a Waters Acquity iClass/Synapt G2S mass spectrometer in positive electrospray (ESI) ionization mode and with ion mobility spectrometry (IMS) enabled.The scan time was 0.1 s over an m/z-range of 50-1200.An MS E method in resolution mode was set up that uses a transfer collision energy ramp from 15 to 40 eV for the high energy MS E data.The following ESI settings were used: capillary voltage = 0.5 kV, sampling cone = 30 V, source offset = 80 V, source temperature = 130 °C and desolvation temperature = 550 °C.Source gas flows were: cone = 50 L/h, desolvation = 950 L/h and nebulizer = 6 bar.For the IMS, the velocity and height of the step wave were set at 300 m/s and 10 V, respectively.IMS wave velocity and height were set at 650 m/s and 40 V, respectively, and IMS gas flow was set at 90 ml/min.Separation was performed using a Waters Acquity BEH phenyl column (100 × 2.1 mm i.d., 1.7 μm particles, 130 Å pore size) and the mobile phase consisted of 10 mM ammonium acetate in H20/ACN 95:5 [v/v] (mobile phase A) and ACN/MeOH 80:20 [v/v] (mobile phase B).The column was held at 40 C° and a flow-rate of 0.4 ml/min and a sample injection volume of 2 µl were used.After an initial time of 0.75 minute at 95% A, a gradient of 6.25 min was run to 95% B, followed by a hold at 95% B for 2.0 min.The column was equilibrated for 1 min between injections.
Data processing was done with MassMetaSite 4 and data review and metabolite assignments were performed in ONIRO (both from Molecular Discovery Ltd.).The resulting metabolites were partially identified using MS E data that were partially cleaned from co-eluting background ions based on both retention time and drift time alignment 38 .

Liquid chromatography -infrared ion spectroscopy.
Separations for the ion spectroscopy experiments were performed using a Bruker Elute UHPLC system consisting of a binary pump, cooled autosampler and a column oven.MS detection was performed using a 3D ion trap mass spectrometer (Bruker, AmaZon Speed ETD).The column was a Waters XBridge BEH phenyl column (150 × 3.0 mm i.d., 2.5 μm particles, 130 Å pore size) and the mobile phase consisted of 10 mM ammonium acetate in H2O/ACN 95:5 [v/v] (mobile phase A) and ACN/MeOH 80:20 [v/v] (mobile phase B).The column was held at 40 C° and a flow-rate of 1 ml/min and sample injection volumes of 2 µl were used.After an initial time of 0.9 minute at 95% A, a gradient of 7.7 min was run to 95% B, followed by a hold at 95% B for 2.4 min.The column was equilibrated for 2 min between injections.
IRIS experiments were performed with a 3D ion trap mass spectrometer (Bruker, AmaZon Speed ETD) modified for spectroscopy experiments (details on the modifications and synchronization between MS and laser instrument are described elsewhere 39 ) and the FELIX infrared free electron laser (see Ref. 40 ).FELIX was set to produce radiation in the ~600-1900 cm -1 range in the form of ~10 µs macropulses at a 10 Hz repetition rate.The bandwidth was ~0.4% of the centre frequency and the (wavelength-dependent) macropulse energy was 20-150 mJ.Mass-isolated ions were irradiated with a single FELIX macropulse.Photon absorption by the trapped mass-selected ions occurs when the laser frequency is on resonance with one of the vibrational transitions of the ion.Multiple-photon absorption leads to an increase in internal energy and eventually dissociation, which is detected by recording the MS spectrum after irradiation.Thus, the fraction of dissociated ions is a proxy for IR absorption.Recording a series of photofragmentation mass spectra as a function of the laser frequency and plotting the fragmentation yield [IRIS intensity = ln(ΣI(precursor + fragment ions)/I(precursor ion))] produces an IR spectrum.Each IR point was recorded as an average of six MS spectra and the IRIS intensity was linearly corrected for frequency-dependent macropulse energy variations 41 .The IR frequencies were calibrated using a grating spectrometer.
Recording an ion IR spectrum involves stepping the laser frequency through the range of interest and takes longer than the width of an LC peak.Therefore, fractions of LC eluent containing ions of interest were collected using a two-position six-port switching valve and an 80 µl sample loop.Here, the LC eluent is diverted to waste by the switching valve until the peak of interest arrives at the end of the column.At that point, the valve is switched such that the peak of interest is stored in the sample loop.After the elution of the peak, the valve is switched back and the contents of the sample loop are infused into the ion trap instrument by a syringe pump at reduced flow rate (120-180 µL h - 1 ).In separate experiments, solutions of reference standards (~1 µM in MeOH/H2O 50:50 [v/v]) were directly infused by the syringe pump for IRIS analysis.

Quantum-chemical calculations
Computed vibrational spectra of candidate structures were obtained by performing quantumchemical calculations using the Gaussian 16 software package. 42Input structures for the computations were generated using an automated workflow for conformational searching that uses the SMILES structure format [43][44] of the neutral molecule as input.Details of this computational workflow are described elsewhere. 32In short, using the cheminformatics toolbox RDkit 45 , the workflow generates protonation isomers considering all nucleophilic sites of the neutral molecule and performs a conformation search for each isomer using a distance geometry algorithm.7][48][49][50] The minimized structures are clustered based on similarity, resulting in a maximum of 10 conformers.These are minimized at the PM6 level of theory and structures with a relative energy (with respect to the lowest-energy conformer) of <40 kJ/mol are further optimized at the B3LYP/6-31++G(d,p) level of theory.For structures with relative energies below 20 kJ/mol, a (likely) more accurate energy value was obtained from a single-point MP2/6-311++G(2d,2p) calculation using the B3LYP optimized geometry; these values were used in the final energy ranking.Harmonic vibrational frequencies were calculated at the B3LYP/6-31++G(d,p) level.Frequencies were scaled by 0.975 and broadened with a Gaussian line shape of 20 cm -1 full width at half maximum to facilitate easy comparison with experimental spectra.
Unless otherwise indicated, structures and IR spectra reported in the manuscript correspond to the lowest-energy conformer for each isomer/tautomer.The relative energies of different isomers is considered not to be of much relevance in the identification of metabolites, as the underlying metabolic reactions are enzymatic.Relative energies for isomeric candidate structures are therefore usually not reported.

Method validation using reference standards
Reference standards of 1'-hydroxy-midazolam (1'-OH-MDZ) and 4-hydroxy-midazolam (4-OH-MDZ) are commercially available.To explore the feasibility of identifying MDZ metabolites using infrared ion spectroscopy and quantum-chemical calculations, we recorded IR spectra of the protonated ions ([M+H] + , m/z 342) of the two references.Figure 1a compares their IR spectra, showing that they are similar, but also present clear and distinct differences.Note for instance the doublets around 1100 cm -1 and 1150 cm -1 in the 1'-OH-MDZ spectrum, the band around 650 cm -1 in the 4-OH-MDZ spectrum and the very weak band around 900 cm -1 in the 4-OH-MDZ spectrum, which is redshifted compared to the corresponding peak in the 1'-OH-MDZ spectrum.Of special interest is the weak band just above 1700 cm -1 in the 4-OH-MDZ spectrum, which is attributed to a C=O stretch vibration.The structure of 4-OH-MDZ (inlayed in Figure 1a) appears not to include a C=O moiety, so that we rationalize the observed C=O stretch by enol-to-amide tautomerization (see Figure 1a).Likely, a small fraction of the ion population is in this tautomeric form.An analogous tautomerization is not possible for 1'-OH-MDZ, making the C=O stretch band very diagnostic in the differentiation of the two isomers.
Figure 1b and 1c compare the computed IR spectra of protonated 1'-OH-MDZ and the two tautomers of protonated 4-OH-MDZ to the experimental IR spectra.Computed geometries are inlayed in each panel.It is seen that the 1'-OH-MDZ spectrum is reasonably well explained by the computational spectrum, except for the intense band calculated just below 1200 cm -1 , which is redshifted compared to the experiment.This band corresponds to an OH-bending vibration of the 1'-OH-group, which may be redshifted due to the hydrogen-bond with the protonated nitrogen (see optimized geometry in the figure) that is not well modelled within the harmonic approximation.Moreover, a few predicted low-frequency bands are absent in the experimental IR spectrum, probably because of their low intensity, not allowing the ion to absorb enough photons to reach the dissociation threshold.The IR spectrum of 4-OH-MDZ is well explained by the computed spectrum of the enol-imine tautomer except for the obvious absence of the carbonyl stretching vibration, which is very intense in the predicted spectrum for the amide-tautomer.This suggests that the majority of the ion population corresponds to the enol-imine tautomer, despite its substantially higher predicted energy relative to the amide-tautomer (+49.0 kJ/mol).Perhaps the hydroxyl group of the enol-imine structure imparts greater solution-phase stability and no tautomerization occurs upon ESI transfer to the gas phase.Previous studies have provided ample spectroscopic evidence for such scenarios. 51- 53Note that an alternative enol-imine tautomer, having an N=C(-OH) double bond, of 4-OH-MDZ can be imagined as well.This tautomer is 11.2 kJ/mol lower in energy than the enol-imine tautomer in Figure 1c, but its predicted IR spectrum does not match well with experiment (see Figure S1 in the  .SI) and it is therefore not likely to contribute to the ion population.Figure 1d compares the experimental IR spectrum of 4-OH-MDZ to a weighted average of the computed IR spectra of the two assigned tautomers employing a 98:2 enol-imine/amide ratio.Note that the doublet around 1150 cm -1 in the 1'-OH-MDZ spectrum versus the singlet in the 4-OH-MDZ spectrum, is well predicted by the computations.This difference is due to the C-OH stretch vibration of the hydroxyl group, which shifts on top of a C-H in-plane bending mode depending on its attachment to an aromatic ring versus a methylene group, making this band of diagnostic value.

Identification of the main metabolites of midazolam
MDZ was incubated with human hepatocytes for 120 min.Samples taken at t=0 min, t=60 min and t=120 min were analyzed using an untargeted LC-MS approach (see method section) to identify MDZ derivatives and Table 1 contains the resulting compound list.Extracted ion chromatograms of the detected m/z-values in the t=60 min sample can be found in the SI (Figure S2).Compound A (Table 1) is recognized as protonated MDZ and the intensity of this compound decreased during incubation, indicating that it is being metabolized.Compound B and C differ by 16 u from the parent drug (A), identifying them as hydroxylation products of MDZ.Compound D has an m/z-value that is 176 u higher than the parent drug, and is therefore likely the product of glucuronidation of MDZ.Compound E has a mass difference of 176 u relative to compounds B and C, and is therefore likely the product of sequential hydroxylation and glucuronidation.We selected the most abundant compounds (A, B, D and E; see Figure S2 in the SI) for further identification using MS/MS and IRIS.It is seen that compound A and B show a rich fragmentation consistent with their structural assignment as protonated MDZ and protonated hydroxy-MDZ.In addition, the MS/MS spectrum of compound B shows an intense waterloss fragment, suggesting that the hydroxyl-group is located on an aliphatic position in the precursor ion.However, no distinction could be made between the two aliphatic positions of MDZ solely based on this fragmentation spectrum.The MS/MS spectra of compounds D and E both show a characteristic glucuronide-loss fragment (-176 u), which confirms their assignment as glucuronides.However, none of the observed fragment ions retains the glucuronide-group, making it impossible to derive the position of the glucuronide-group on the basis of these spectra.
For further identification, we recorded IR spectra of the protonated ions of the compounds and compared those to theoretical IR spectra of candidate structures.Figure 3a shows the IR spectrum of compound A compared with the theoretical IR spectrum of protonated MDZ ([MDZ+H] + , m/z 326).Overall, the observed vibrational bands are well reproduced in the theoretical spectrum, both in terms of peak positions as well as relative intensities.Some of the very weak bands in the 850 -1050 cm -1 range in the computational spectrum appear to escape observation.Their low absorption cross sections do not allow the ion to absorb enough photons within the duration of the laser pulse to reach the dissociation threshold.Moreover, some deviations between observed and computed relative band intensities are noticed.A slight redshift of the C=N and C=C stretching vibrations near 1600 cm -1 is observed, probably due to the anharmonic shift deviating slightly from average.However, in general, computed and theoretical IR spectra show a convincing match, confirming the structural assignment of compound A as protonated MDZ.
Figures 3b/c compare the IR spectrum of compound D to theoretical spectra of two candidate structures, see Figure .The two imine N-atoms of MDZ were considered as potential sites of glucuronidation.The majority of the experimental spectrum is well reproduced by the predicted spectrum of the candidate in Figure 3b, labelled MDZ-N-glucuronide 1 (red trace).Again, the C=N and C=C stretching vibrations near 1600 cm -1 are slightly redshifted in the experiment and some intensity differences are observed.The shoulder of the carbonyl stretching vibration near 1800 cm -1 suggests the co-existence of multiple conformations in the ion population.Indeed, at 10 kJ/mol above the global minimum conformer, the computations predict a conformer with a different orientation of the carbonyl group, leading to a slightly different C=O stretch frequency (see Figure S3 in the SI).
The presence of this conformer can also explain the high intensity observed around 1050 cm -1 .The predicted spectrum of the second candidate structure (labelled MDZ-N-glucuronide 2, yellow trace), provides an overall poorer match with the experiment.A feature of particular interest is the computed band near 1550 cm -1 , which corresponds to the C=N stretching vibration involving the glucuronated N-atom and is diagnostic for glucuronidation at this position.Moreover, the absence of a significant band around 1300 cm -1 in the MDZ-N-glucuronide 2 spectrum, where the MDZ-N glucuronide-1 computation predicts delocalized C-H bending vibrations, and the vibrational bands in the 600-900 cm -1 region, corresponding to delocalized vibrations, are diagnostic.These bands are likely affected by the largely different three-dimensional conformation induced by the different position of the glucuronide group.Overall, based on the better spectral match, we assign the MDZ-N-glucuronide 1 structure to compound D. Note that MDZ contains a third N-atom where

Figure 2. MS/MS spectra of compound A (panel a, m/z 326), compound B (panel b, m/z 342), compound D (panel c, m/z 502) and compound E (panel d, m/z 518). Observed m/z-values, molecular formulas of selected neutral losses and (partial) molecular structures are shown in each panel.
glucuronidation can be considered to take place.However, this reaction site is less likely due to severe steric hindrance at this position.Preliminary calculations for this isomer indicate that it is not very stable (several of its conformers converged to the protonated MDZ-N-glucuronide 1 structure during the geometry optimization) and that its predicted IR spectrum deviates significantly from the experimental IR spectrum (see Figure S4 in the SI).
To identify compound B, we generated nine candidate structures resulting from MDZ hydroxylation using the in silico enzymatic reaction tool BioTransformer [54][55] .As discussed above, the strong H2Oloss observed in the MS/MS spectrum of compound B suggests that hydroxylation occurs on an aliphatic position, pointing in the direction of 1'-OH-MDZ and 4-OH-MDZ.Figure 3d/e compare the IR spectrum of compound B to theoretical spectra of 1'-OH-MDZ and 4-OH-MDZ (see also Figure 1).A comparison to the theoretical IR spectra of all other candidate structures (with the OH-group in an aromatic position) is shown in Figure S5 in the SI.It is seen that the computed spectrum of 1'-OH-MDZ reproduces the experimental spectrum of compound B well, except for the OH-bending vibration of the 1'-OH group, which is blue-shifted in the experimental spectrum (as was also discussed above for the 1'-OH-MDZ reference spectrum).Computed spectra for most other candidates lack one or more diagnostic features observed in the experimental spectrum.For instance, several computed spectra fail to reproduce the intense vibrational band observed around 1000 cm -1 or the doublet around 1100 cm -1 .The spectrum predicted for aromatic OH-MDZ 4 (see SI) contains most of the observed features and shows a relatively good match to the experimental spectrum.However, relative band intensities and the overall band pattern deviate more from the experimental

Figure 3. Comparison of the experimental IR spectra of compounds A, B and D with computed spectra for candidate structures: (a) compound A versus protonated MDZ, (b) compound D versus protonated MDZ-N-glucuronide 1, (c) compound D versus protonated MDZ-Nglucuronide 2, (d) compound B versus protonated 1'-OH-MDZ, (e) compound B versus a weighted average of the enol-imine and amide tautomer of 4-OH-MDZ. Input structures and the conformations resulting from the quantum-chemical calculations are inlayed in the
spectrum than the spectrum predicted for 1'-OH-MDZ.Therefore, we assign the structure of 1'-OH-MDZ to compound B. The IR spectrum of a reference standard of protonated 1'-OH-MDZ is available (see Figure 1b) and indeed provides a close match to the IR spectrum of compound B, confirming our structural assignment (see Figure S6 for a direct overlay of the two IRIS spectra).
Identifying the molecular structure of compound E (m/z 518), which is the product of both glucuronidation and hydroxylation of MDZ, requires determination of both the site of hydroxylation and glucuronidation.Many structural isomers can therefore be considered as candidate, but we can expedite identification by making use of the fact that glucuronides generally expel the glucuronide moiety upon collision induced dissociation (CID).The resulting aglycone fragment ion retains the OH-group, so that an IR spectrum of the aglycone can reveal the position of the hydroxyl moiety, narrowing down the list of candidate structures for the glucuronide.Figure 4a compares the IR spectrum of the glucuronide-loss CID fragment ion (m/z 342) of compound E with the IR spectrum of compound B (m/z 342) from Figure 3.The two spectra are near-identical.Small differences in band intensities can be attributed to day-to-day fluctuations in IR laser power.Compound B was identified above as 1'-OH-MDZ, so that we can assign this structure to the aglycone of compound E as well.
As candidate structures for compound E, we considered glucuronidation on the OH-group and on the two imine N-atoms of 1'-OH-MDZ, see Figures 4b-d, where the experimental IR spectrum is compared with the theoretical IR spectra of these candidate structures.Comparing the computational spectra in Figures 4c-d and in Figures 3b-c shows that the predicted IR spectra of the candidate structures having the same glucuronide-position, but differing in the presence or absence of the OH-group, are almost identical (see Figure S7 in the SI for direct overlays).This suggests that the OH-group does not significantly affect the preferred conformation of the compounds, as can also be inferred from the computed 3D-conformations in the Figures.Similar to the discussion of the

Figure 4. (a) Overlay of the experimental IR spectra of compound C and the m/z 342 fragment of compound D. (b-d) Comparison of the experimental IR spectrum of compound D to theoretical IR spectra of protonated 1-OH-O-glucuronide (panel b), 1-OH-MDZ-N-glucuronide 1 (panel c) and 1-OH-MDZ-N-glucuronide 2 (panel d). Input structures and the conformations resulting from the quantum-chemical calculations are inlayed in the figure.
structural assignment of compound D, we can exclude the presence of 1'-OH-MDZ-N-glucuronide 2 (yellow trace) based on the absence of an experimental band near 1550 cm -1 (corresponding to the C=N stretching vibration of the glucuronated N-atom), the absence of the band around 1300 cm -1 and the mismatch in the diagnostic 600-900 cm -1 region.The distinction between the 1'-OH-MDZ-O-glucuronide and the 1'-OH-MDZ-N-glucuronide 1 is more subtle.Compared with the computed spectrum for 1'-OH-MDZ-N-glucuronide 1 (Figure 4c), the experimental spectrum does not contain the bands predicted around 1200 cm -1 and the match in the 600-950 cm -1 range is poor.On the other hand, the predicted spectrum of 1'-OH-MDZ-Oglucuronide (Figure 4b) does not reproduce the relative intensities of the bands around 1100 and 1400 cm -1 .However, overlaying the IR spectra of compound D and compound E (see Figure S8) shows that they are quite different, whereas computed spectra for structures with the same glucuronidation site are very similar; this suggests that compounds E and D do not share the same glucuronidation site.Additionally, the band predicted near 1200 cm -1 for 1'-OH-MDZ-N-glucuronide 1 (Figure 4c) is analogous to the feature predicted for MDZ-N-glucuronide 1 (Figure 3b), which is clearly observed in the experimental IR spectrum of compound D; this suggests that the computations perform properly for this band.The intensity of the band computed at 1400 cm -1 for protonated 1'-OH-MDZ-Oglucuronide is clearly at odds with the experimental spectrum; it corresponds to the OH-bending vibration of the carboxylic acid-group, which has a trans-configuration in the lowest-energy conformer.This causes the OH-group to be part of a strong hydrogen-bonding network, probably leading to an overestimation of the intensity of this band in the computation.Indeed, the higherenergy (+20.8 kJ/mol) conformer with a cis-configuration carboxylic acid group is predicted to show the correct relative intensity around 1400 cm -1 (see Figure S9 in the SI).Moreover, the relative intensity around 1100 cm -1 is also better predicted for this conformer.Therefore, we suspect that the intensities of these bands are incorrectly predicted at the level of theory used and we assign the 1'-OH-MDZ-O-glucuronide isomer to compound E.

Identification of ring-opened midazolam structures
A well-known property of MDZ is that it can undergo a spontaneous ring-opening via hydrolysis in an acidic environment, which enhances its solubility. 34,37 t physiological pH, the lipid soluble ringclosed form is dominant, allowing penetration of the blood-brain barrier.A study involving MDZ metabolism in mice suggested that its metabolites can undergo a ring-opening as well, although the exact structure of the detected H2O-addition product was not confirmed. 35Here, we employ IRIS to structurally identify several ring-opened MDZ derivatives.To promote the formation of the ringopened structures, we repeated the incubation of MDZ, but quenched the metabolism using a solution containing 10% formic acid (see method section).This led to the detection of three extra features, listed in Table 2, in addition to the compounds in Table 1.Compounds F, G and H are 18 u higher in mass relative to compounds A, B/C and D, respectively.We can therefore hypothesize that they are the ring-opened forms of these compounds.To explore whether IRIS can help identify the full chemical structure of these compounds, we selected compound F and H for further identification using IRIS.S10 for a direct overlay), with additional intense bands around 950 cm -1 and just below 1700 cm -1 in the spectrum of compound F (and some intensity differences).The band around 1700 cm -1 indeed suggests a ring-opened MDZ structure as this band may correspond to a ketone stretching vibration (see molecular structure in Figure 5a).Indeed, theory predicts a carbonyl stretching band at this frequency for ring-opened MDZ.
Overall, the predicted IR spectrum matches convincingly with the experimental spectrum, so that we assign the structure of protonated ring-opened MDZ to compound F.
Compound H has a mass difference of 176 u (corresponding to glucuronidation) with compound F, and a mass difference of 18 u with compound D and is therefore likely the result from ring-opening of MDZ-N-glucuronide 1 (compound D).In that case, we may expect the aglycone of compound H produced in a CID MS/MS experiment to correspond to ring-opened MDZ. Figure 5b compares the IR spectrum of the glucuronide-loss CID fragment ion (m/z 344) of compound H to the IR spectrum of compound F (m/z 344), confirming that these spectra are indeed highly similar, if not identical.This is another example of the ability of IRIS to link phase I and phase II metabolites.In contrast, the IR spectrum of compound H itself (m/z 520) and the theoretical IR spectrum for the lowest-energy conformer of protonated ring-opened MDZ-N-glucuronide 1 (Figure 5c) show subtle deviations.The ring-opened MDZ-N-glucuronide 2 isomer (Figure 5d) can not be excluded based on

Figure 5 (a) Experimental IR spectrum of compound F in comparison to the theoretical IR spectrum of protonated open-chain MDZ. (b) Comparison of the experimental IR spectra of compound F and the m/z 344 fragment of compound H. (c) Comparison of the experimental IR spectrum of compound H to theoretical IR spectra of two different conformers of protonated open-chain MDZ-N-glucuronide 1. (d) Comparison of the experimental IR spectrum of compound H to the theoretical IR spectrum of protonated open-chain MDZ-N-glucuronide 2. Input structures and the conformations resulting from the quantum-chemical calculations are inlayed in the figure.
its computed IR spectrum, as it shows a relatively good match with all bands in the spectrum.However, its presence is not expected, as the ring-closed version of this isomer was not observed.As an alternative, the predicted spectrum of a higher-energy conformer (+9.3 kJ/mol) of protonated ring-opened MDZ-N-glucuronide 1 shows a relatively good match to the experimental IR spectrum.
The most diagnostic difference between the predicted IR spectra of the two conformers is the relative position of the carboxylic acid C=O stretching vibration (~1800 cm -1 ).The frequency shift is likely due to an n -π + interaction with the charged imidazole ring or an H-bonding interaction with the primary amine, which are not present in the lowest-energy conformer.These arguments suggest that compound H corresponds to ring-opened MDZ-N-glucuronide 1 but that we do not probe the predicted lowest-energy conformation.This also reveals limitations of reference-free IRIS identification: especially for high-MW compounds, spectral differences may become subtle and the number of low-energy conformers may become large.Nonetheless, we believe that even in such cases, an IRIS analysis can greatly restrict the number of isomers to be synthetically prepared as reference standards, thus still expediting the identification.

Further discussion and conclusions
We demonstrated the reference-free identification of the main metabolites of MDZ formed by incubation in human hepatocytes using a combination of LC-MS/MS, IRIS and quantum-chemical computations.The metabolic pathway elucidated in this study is shown in Figure 6.We found that MDZ metabolizes via hydroxylation to 1'-OH-MDZ, which is followed by glucuronidation on the newly introduced OH-group.Additionally, we observed direct glucuronidation of the drug resulting in an MDZ-N-glucuronide.This is in general agreement with the literature, which reports these compounds as the major metabolites as well 33 .Some studies report minor amounts of 4-OH-MDZ and 1,4-dihydroxy-midazolam 33,36 .Indeed, we observed minor amounts of a second OH-MDZ isomer (m/z 342, RT 4.04 min), which was however not selected for further characterisation because of its low abundance, but it likely corresponds to 4-OH-MDZ.A metabolite with the molecular mass of 1,4-dihydroxy-midazolam was not observed in this study.
The results show that isomers resulting from hydroxylation at different sites are effectively differentiated on the basis of IRIS, as was also shown in previous work 29,31 .Where previous studies focussed on identification using reference standards, we demonstrated here identification in a reference-free manner aided by quantum-chemical computations.Glucuronide metabolites have much higher molecular weights than hydroxylation metabolites, but nonetheless present diagnostic IR spectra with many sharp spectral bands.Nevertheless, due to the large molecular size, the distinction between different glucuronidation-sites on the basis of computed IR spectra is less clear than for hydroxylation isomers.However, in most cases our reference-free approach suggests one candidate that is most likely, which may guide the synthesis of a small number of reference compounds for final confirmation.
We also showed that IRIS on an MS n -enabled platform can aid in linking phase I with phase II metabolites, thus revealing the metabolic pathways.An ion trap mass spectrometer in combination with a tunable IR laser allows us to record an IR spectrum of the glucuronide aglycone generated via CID MS/MS.Matching with IR spectra of observed phase I metabolites establishes a direct link between the phase II and phase I metabolites.IR spectra give direct information on the functional groups of a metabolite as well.We demonstrate the added benefit of this in practice by identifying several metabolites resulting from MDZ hydrolysis in an acidic environment.Here, the IR band corresponding to an ketone induced by MDZ ring-opening directly pointed towards a likely candidate structure.
In conclusion, we showed that IRIS in combination with quantum-chemically computed IR spectra of candidate structures can aid in the structural elucidation of drug metabolites to support drug discovery and development.Comparing IR spectra of the drug itself and various metabolites gives direct information on the formation of new functional groups, whereas computed IR spectra aid in final identification of biotransformation sites.The work here was performed at an open-access free electron laser facility, but IRIS can also be implemented with table-top laser systems 56 for easier and efficient application in pharmaceutical laboratories, although currently still over a restricted IR spectral range.

Figure 1 .
Figure 1.(a) Experimental IR spectra of protonated 1'-OH-MDZ (top panel) and 4-OH-MDZ (bottom panel).(b) Comparison of the experimental and theoretical IR spectra of protonated 1'-OH-MDZ.(c) Comparison of the experimental IR spectrum of protonated 4-OH-MDZ to the theoretical IR spectra of the enol-imine and amide-tautomer of protonated 4-OH-MDZ.(d) Comparison of the experimental IR spectrum of protonated 4-OH-MDZ to a weighted average of the theoretical IR spectra of the enol-imine and amide-tautomer (98:2 enol-imine/amide ratio).Input structures and the conformations resulting from the quantum-chemical calculations are inlayed in the figure.

Figure
Figure2b-d shows the recorded MS/MS spectra for compound A, B, D and E. The recorded mass of the main ions and the main neutral losses are inlayed in the figure.It is seen that compound A and B show a rich fragmentation consistent with their structural assignment as protonated MDZ and protonated hydroxy-MDZ.In addition, the MS/MS spectrum of compound B shows an intense waterloss fragment, suggesting that the hydroxyl-group is located on an aliphatic position in the precursor ion.However, no distinction could be made between the two aliphatic positions of MDZ solely based on this fragmentation spectrum.The MS/MS spectra of compounds D and E both show a characteristic glucuronide-loss fragment (-176 u), which confirms their assignment as glucuronides.However, none of the observed fragment ions retains the glucuronide-group, making it impossible to derive the position of the glucuronide-group on the basis of these spectra.

Figure 6 .
Figure 6.Overview of the in vitro metabolism of MDZ determined in this study.