Using LC Retention Times in Organic Structure Determination: Drug Metabolite Identification.

BACKGROUND: There is a continued need for improvements in the efficiency of metabolite structure elucidation.
OBJECTIVE: We propose to take LC Retention Time (RT) into consideration during the process of structure determination.
METHODS: Herein, we develop a simple methodology that employs a Chromatographic Hydrophobicity Index (CHI) framework for standardizing LC conditions and introduce and utilize the concept of a predictable CHI change upon Phase 1 biotransformation (CHIbt). Through the analysis of literature examples, we offer a Quantitative Structure-Retention Relationship (QSRR) for several types of biotransformation (especially hydroxylation) using physicochemical properties (clogP, hydrogen bonding).
RESULTS: The CHI system for retention indexing is shown to be practical and simple to implement. A database of CHIbt values has been created from re-incubation of 3 compounds and from analysis of an additional 17 datasets from the literature. Application of this database is illustrated.
CONCLUSION: In our experience, this simple methodology allows complementing the discovery efforts that saves resources for in-depth characterization using NMR. Copyright© Bentham Science Publishers; For any queries, please email at epub@benthamscience.org.

Introduction

The lowering of intrinsic metabolic clearance is a key part of the optimization process that takes place during drug discovery. A low intrinsic metabolic clearance is one of the properties that could lead to a lower dose and an adequate circulating half-life that, in turn, allows for oral delivery of a drug. Metabolic stability optimization is conducted by in-depth analyses of the types of metabolic biotransformations a drug undergoes and the modification sites of the drug, investigations which are conducted primarily via liquid chromatography coupled to mass spectrometry (LC-MS) [1]. Recent advances of high-resolution MS allow for a very accurate method for the identification and characterization of highly complex organic mixtures. This methodology has been amenable to higher throughput applications for monitoring and quantifying a drug and its metabolites. Metabolites are detected and characterized based on their mass over charge ratio (m/z), and the interpretation of fragment ions are based on Collision-induced Dissociation (CID) spectra. This process assists in the determination of metabolite structures but often results in a number of possible positional isomers that have identical molecular ions and fragment ion spectra; this is true particularly for the addition of a single oxygen atom, the most frequently occurring biotransformation. Definitive proof of a structure often requires further characterization such as total synthesis and/or NMR. The latter requires metabolite generation followed by a laborious isolation. For marketed drugs, metabolites can be important for forensic or environmental reasons but are typically not available as reference standards. Here, we propose that standardized chromatographic behavior (i.e., LC Retention Time (RT)) reflects certain physicochemical properties of a metabolite, that if considered together with MS information, can allow for greater insight into a metabolite’s structure.

The prediction of RT from structure when authentic standards are unavailable is becoming more common in the fields of natural products, environmental contamination, metabolomics and forensics [2-7]. Firstly, new LC protocols have made RTs more reliable between runs, between labs and even with changing gradients and instruments [8, 9]. Annotating compound databases with RT has regained popularity, as 20 years ago this was commonplace before the advances in electrospray MS [10, 11]. Secondly, a maturing understanding of reverse phase LC retention mechanisms allows for reliable Quantitative Structure–Retention Relationship (QSRR) modeling between calculated molecular descriptors and experimental RTs of authentic compounds [7, 12]. The QSRR concept has been applied to drug metabolite identification, as is reported in the literature [13-17], and software packages include this type of modeling as a feature [12]. As an example, Herre and Pragst [15] proposed a biotransformation RT shift using data from C8 column chromatography to report 55 parameters for 29 biotransformations. Despite these advances, the QSRR concept has not really taken hold in the routine workflow of metabolite identification efforts in drug discovery.

One obstacle in RT prediction is that, unlike past experiences with gas chromatography [18], RTs for reverse phase LC tend to be less reproducible due to subtle effects of pH, underivatized silanols and mobile phase compositions; however, the robustness of RT assessment can be improved with RT indexing, for example using nitroalkanes [3], or amides [8] or drugs [11]. Valko’s Chromatographic Hydrophobicity Index (CHI) [19] is attractive and uses language familiar to drug metabolism scientists. The CHI roughly approximates the percent acetonitrile that is required in a gradient for a solute to be equally distributed between stationary and mobile phases. This system converts RT to a CHI value, by calibrating a given linear gradient method with a set of drug-like CHI standards. CHI values are calculated using Equation 1, where a and b are constants.

RT=a(CHI)+b (1)

As with other quantitative structure-property predictions, we assume functional group additivity and define a CHIbt as the change in chromatographic hydrophobicity due to a biotransformation.

CHIbt = CHImetabolite -CHIparent (2)

When conducting chromatographic separations of metabolites, bioanalytical chemists have largely settled on the use of reverse phase columns (mainly C18) with a combination of acetonitrile and water acidified with formic acid as the mobile phase. We, therefore, concentrated our efforts on the use of C18 columns in the determination of LC RT in this study. The theoretical basis of reverse phase LC retention has been recently reviewed [20]. Snyder and associates proposed a hydrophobic subtraction model for reverse phase LC selectivity [21]. Their theory relates the RTs of two compounds to their physical properties and to corresponding properties of the column, with hydrophobicity as the main retention mechanism in reverse phase, especially in C18 columns. Secondary factors are steric interactions (which prevent insertion of a solute into the hydrocarbon chains of the stationary phase), hydrogen bonding (between donor and acceptor in the solute as well as between donor and acceptor in the stationary phase) and ion exchange interactions between a charged solute and the column.

The model developed by Snyder et al. is used primarily for comparing LC column behaviors, and not solute behavior. In addition, their predictions of solute RT are more descriptive than quantitative. However, the selectivity parameters from different columns show the similarities shared by many of our favorite columns (Table ). As can be seen in Table , the values of hydrophobicity are one to two orders of magnitude larger than most of the other parameters, indicating the importance of the hydrophobic component in chromatographic separation. The ion exchange term is relevant mainly for cationic solutes (bases, quaternary amines) interacting with ionized silanols from the column. Working at an acidic pH has become the preferred LC-MS method (pH 2.8; 0.1% formic acid) because silanols remain neutralized, thus minimizing ion exchange effects. For most type B silica C18 columns, the ion exchange term is relatively small and unchanged by metabolism; however, this term must be considered when a drug’s basic site is affected by metabolism (N-oxidation, N-acetylation). The last column of Table shows the parameters for a type A silica column. Literature data from such a column may be very different in terms of both hydrogen bonding and ion exchange and will not be included in this analysis.

Experimental section

Measurement of CHIbt

Triclocarban, imipramine and atorvastatin were incubated with human liver microsomes using a standard protocol. The metabolites were identified by LC/MS/MS on a Thermo Fisher Orbitrap Velos and comparison of MS/MS spectra and RT to literature values. The Valko CHI standards were run with the same chromatography. All RT data was pasted in an Excel (2007) spreadsheet (Microsoft, WA; Supplemental Data, S-1) along with reference information, structures and details of each chromatographic method. Equation 1 was used with the literature CHI values for standards at pH 2.6 to calculate slope and intercept for conversion of each metabolite’s RT to a CHI value.

Theory for Production of a QSRR for Estimation of CHIbt from Literature RT Data

Equation 3 is used to relate CHI values to the octanol–water partition coefficient (logP) [22] for neutral molecules. Pharmaceutical research departments use this concept to estimate hydrophobicity in a high throughput laboratory using LC gradients [19, 22-25].

logP=0.047(CHIn)+0.36(HBC)–1.1 (3)

HBC is the hydrogen bond donor count, obtained by summing up all the NH and OH groups in a molecule and the superscript n shows that this only applies for neutrals. An increase in the hydrogen bond acidity of the solute decreases LC retention, whereas hydrogen bond acidity does not practically affect logP. This is because reversed phase C18 stationary phases have much less hydrogen bonding than do acetonitrile–water mobile phases, and octanol has almost the same level of hydrogen bonding as water [26]. Thus, a correction term for solute acidity is needed to get a valid relationship between an LC retention parameter and logP.

Equation 4 replaces logP with clogP (calculated logP) and solves for CHIn.

CHIn=21.3clogP–7.67HBC+23.4 (4)

Most drugs are charged, however, and we must, therefore, consider the effect of charge on hydrophobicity and on ion exchange properties. At pH 2.8 (the pH of 0.1% formic acid), most carboxylic acids are protonated hence not charged, and therefore neutral (especially at a high percent of acetonitrile [27]), but amines will be protonated, and therefore positively charged. Predicting the hydrophobicity of charged molecules is very complex. The RT of a charged molecule is sensitive to temperature, counterions, pH and mobile phase composition. To take into account the presence of charged molecules, we introduce a compound specific variable, Fp, to include those charged molecule stationary phase interactions that are not incorporated into HBC or clogP or are miscalculated by the simplified equation 4.

CHIp=CHInp+Fp (5)

Again we assume additivity and define an Fbt so that the

CHIm=CHInm+Fp+Fbt (6)

For the time being, we will not focus on Phase 2 conjugation reactions such as sulfonation, glucuronidation and glutathione adduction that introduce new charge states to a molecule. We will also ignore major dealkylations or amide cleavages, which dramatically alter a molecule’s weight and/or charge state. With these limitations, the parent molecule will often mimic the charge-related attributes of the metabolite, so that for some metabolites, the Fbt value of the metabolite will be zero. Substituting equation 5 back into equation 1, yields equation 7.

RTp=a(CHInp+Fp)+b (7)

If one or more metabolites,m1, have Fbt=0, then equation 8 will hold.

RTm1=a(CHInm1+Fp)+b (8)

We can take the fixed Fp out of the parenthesis and combine it with b to make a new constant b`=aFp+b. generating equation 9 for the parent, 10 for the well-behaved metabolite and 11 for the general case

RTp=a(CHInp)+b’ (9)

RTm1=a(CHInm1)+b’ (10)

RTm=a(CHInm+Fbt)+b’ (11)

Detailed Procedure for Applying the QSRR to Literature RT Data

For each of the 20 examples (the 3 compound datatsets from section 2.1 and 17 literature datasets; all 20 compounds shown in Table ), the structure was redrawn in MoKa (Molecular Discovery; version 2.6.4) [28], and clogP was calculated and added to the spreadsheet. HBC was determined by manual inspection as the total number of NH and OH bonds. CHIn was calculated using equation 4.

In most cases, the parent molecule and a metabolite with unchanged nitrogen atoms and no new major intramolecular hydrogen bonds provide at least two reference points to establish a linear correlation between RT and CHIn and generate a reference fitting line. The slope, a, and y-intercept of the linear correlation, b`, were then used to calculate the CHIn+Fbt for each molecule with equation 11. Cases in which no metabolite had Fbt=0 were not considered. Finally substituting in equations 5 and 6 generates equation 12. The unknown Fp values are eliminated giving equation 13 as the method we used to estimate CHIbt.

CHIbt=(CHInm+Fp+Fbt)-(CHInp +Fp) (12)

CHIbt=(CHInm+Fbt)-(CHInp) (13)

Many metabolism projects resulted in multiple CHIbt results for the same biotransformation. These were averaged in each spreadsheet.

Results and Discussion

A single dedicated laboratory can achieve highly reproducible LC RTs [14]; however, interlaboratory reproducibility of RT is not so simple [29], demanding scrupulous attention to mobile phase, temperature, LC pump dead volume, if not identical column manufacture. We chose to evaluate the CHI retention index system to produce reproducible CHIbt’s. We ran the 10 recommended standards and found that 3 coeluted (colchicine, 8-phenyltheophylline and acetophenone) at pH 2.8. We replaced the former two with benzamide [19, 22] to get better coverage. After conversion of RT to CHI for parent and identified metabolite the calculation of CHIbt is obvious. We chose 3 molecules (atorvastatin, imipramine and triclocarban) to incubate with microsomes in-house to generate metabolites and placed the results in the Supplemental Excel file sheets 2-4. Once enough molecules have been tested a table of CHIbt values such as Table can be used to estimate where in a chromatogram a predicted biotransformation product would elute. Note that of the three molecules we tested, triclocarban yields a metabolite with a CHI value of 150, well outside the linear portion of the CHI calibration curve. It might be valuable to extend the set of CHI reference standards to cover this range, but such an intensely hydrophobic drug metabolite is rare.

Prediction of Structure for Unknown Metabolites with Known CHIbt

Many labs will not have the CHI system or wish to evaluate literature RT data, so we developed a QSRR. To generate an initial table of data we chose an additional 17 molecules from the literature to test our RT prediction method. These literature reports utilized reverse phase LC with acetonitrile and acidic mobile phase and most undergo mono-oxidation biotransformations. The need for better methods of exactly specifying the site of mono-oxidation biotransformation has been pointed out in a recent review [30]. The complete results for all compounds are presented in the Supplemental Material, tabs 2-21, but the process is illustrated here with atorvastatin.

Atorvastatin

Fig. ( illustrates the structure determination process. It shows the extracted m/z 575.255 for the mono-hydroxylated metabolites plotted to a CHIbt axis. The colored bars take the average CHIbt for the important carbon mono-oxidations and set a bandwidth at one standard deviation. MS/MS had shown that the main peaks are results of hydroxylation of the acylaniline function. Comparison to the bars indicates that the early eluter must be the non-hydrogen bonded meta or para isomer while the late eluter must be the hydrogen bonded ortho isomer. In solution, atorvastatin is present in both free acid and lactone forms, so a total of six major peaks (supplemental file, Table ) are observed in the LC-MS chromatogram following human liver microsome incubation [31]. The parent lactone is readily identified. The identification of the two hydroxylated lactones is simple in that they elute in the same relative pattern as the unlactonized forms. The effects of dual biotransformations that occur on separate portions of a drug, for example, the lactone formation and aromatic hydroxylation of atorvastatin, are additive. Importantly, the difference in LC RT values allows for the assignment of regiochemistry as ortho or meta/para in this case.

Fig. (1)

The extracted m/z 575.255 for the mono-hydroxylated metabolites of atorvastatin plotted to a CHIbt axis. The colored bars take the average CHIbt for the important carbon mono-oxidations and set a bandwidth at 1 standard deviation. Comparison to the bars indicates that the early eluter must be the non-hydrogen bonded meta or para isomer while the late eluter must be the hydrogen bonded ortho isomer.

The application of the QSRR prediction is illustrated in the supplemental file, Table . The initial correlation between RT and CHIn is fairly poor (first graph, r2=0.49). Both the ortho hydroxylation and the lactonization effect intramolecular H-bonding, not reflected in clogP or the nominal HBC count. Excluding the ortho-hydroxy point and the 3 lactones allows a two point “corrected line” (middle graph). Applying this new slope and intercept to the RT values for the others allows for the calculation of corrected CHIn+Fbt (column g). CHIbt is then calculated by subtracting the column g value for the metabolite from that of the appropriate precursor.

When atorvastatin and metabolites were analyzed inhouse with the CHI standards the CHIbt results were quite comparable to the predicted results (ortho-hydroxylation, predicted -2.9, observed -3.6; para hydroxylation predicted -16.4 observed -20.6; lactonization, predicted 6.2, observed 7.7). Similarly, comparable results were observed for imipramine and triclocarban.

All the changes in CHI due to metabolism in the compounds used in this study are collected in Table of the supplemental file. Table summarizes the average CHIbt for each biotransformation along with the standard deviations and n values. Where we had measured CHIbt results and predicted results, Table shows only the measured results. The effects of different functional group biotransformations on CHI are discussed below.

Aliphatic C-oxidation

Aliphatic C-oxidation is typically distinguished from aromatic C-oxidation by the loss of water fragment ion in the CID spectrum. The CHIbt is typically more negative for aliphatic C-oxidation than for aromatic C-oxidation or N-oxidation. However, subtle differences in chemical structure cannot be easily predicted. A literature report [32] describes two side-chain hydroxylated derivatives of pioglitizone with similar CID spectra, and so synthesis and chemical derivatization were required to differentiate between the two isomers. If only one isomer was present, identifying it by RT prediction would be impossible since the prediction error would be greater than the difference in RT. In this case, clogP cannot distinguish the isomers which show a slightly lower hydrophobicity for the primary alcohol vs. the secondary. This difference in hydrophobicity is reflected in a lower RT for the primary alcohol. The example of pioglitazone (and the following example, glibenclimide) demonstrates a general rule: with two closely related hydroxylation isomers, the one with the new hydroxyl that is furthest from the unchanged polar part of the parent drug will elute the earliest. This observation is intuitive in that the new hydroxyl group would disrupt the insertion of the more hydrophobic part of the metabolite into the C18 chains of the stationary phase.

Glibenclimide contains a mono acylamino substituted cyclohexyl ring which was extensively oxidized. The 4-trans isomer showed the largest CHIbt while the values for the 4-cis,3-trans and 3-cis isomers were intermediate. The 2-trans isomer had the lowest CHIbt, likely due to increased intramolecular hydrogen bonding. These results are not reflected in clogP and HBC. Using a complex QSRR combining descriptive parameters and Canvas fingerprints Falchi et al. [14] could accurately predict the relative retention of 2-, 3- and 4-hydroxycyclohexylamines but could not distinguish 4-cis vs. 4-trans isomers because they only used 2D descriptors. Herre and Pragst [15] describe 4 isomers of hydroxytestosterone which are well separated on LC but have comparable predicted hydrophobicities. These stereochemical issues require modeling of minimized conformational structures to calculate their hydrophobicities more accurately. Such techniques are not typically available for high throughput applications by non-specialists.

Buspirone is metabolized to four mono-hydroxylated isomers. The N-oxidized metabolite has the longest RT, while the shortest RT is the result of a remote aliphatic oxidation. The two intermediate isomers are due to an oxidation of a pyrimidine and an aliphatic oxidation next to a carbonyl, which creates an intramolecular hydrogen bond.

Aromatic C-oxidation

Determining the regiochemistry of aromatic C-oxidation is not typically possible by MS/MS. Ion mobility MS has been used for distinguishing regioisomeric metabolites [33, 34], but it is not widely available. As demonstrated herein for atorvastatin and triclocarban, LC retention is a simple alternative that can often identify the aromatic substitution of a mono-hydroxylated metabolite. In particular, ortho-oxidation will yield a much longer RT than meta- or para-oxidations if hydrogen bonding is possible. Ortho-oxidation leading to intramolecular hydrogen bonding decreased hydrophobicity in many other substituted phenols (nitro, methoxy, catechol, acetophenone, benzamide) [35-38]. Relative retention of the isomers of acetaminophen [39] also follows the “furthest from the unchanged polar part of the parent drug rule” with the relative retention in the order parartho.

We report two examples of heteroaromatic oxidation, a benzimidazole and a pyrimidine. These often occur with aldehyde oxidase involvement in metabolic clearance. These cases are hard to predict and possibly not well modeled by clogP. But in general, this should decrease CHI less than phenyl oxidation. As with aliphatic oxidation, there appears to be a general correlation between CHI and the distance of the new hydroxyl group to the next most polar part of the molecule. Greater adoption of the simple rules that relate RT to metabolite structure might obviate the need for compound synthesis to prove the structures of all mono-hydroxylated aromatic metabolites.

Oxidation of Amines

Conversion of basic amines to hydroxylamines or N-oxides typically causes an increase in retention at pH 2.8 [40]. As illustrated with nine of our examples, this increased RT distinguishes them from most C-oxidations. The related aliphatic oxidation of the carbon next to a basic cyclic amine, creating an aminol, leads to a minor change in CHI.

Oxidation of Sulfur Compounds

Our table does not contain many sulfur oxidations but the CHIbt for conversion of a sulfide to a sulfoxide is the largest we have observed. Subsequent oxidation to the sulfone has a smaller positive CHI effect.

N,O-Demethylation

Demethylation is a very common biotransformation pathway. The removal of an N-methyl group from a basic amine causes little change to its RT. These metabolites should be much more polar than the parent according to clogP and HBC. Perhaps the loss of the methyl group increases access to charged silanols with a subsequent increase in ion exchange interaction. Loss of a methyl group from an amide (loperamide) or a methyl ether (4-methoxy-α-PVP, rivoglitazone, diltiazam) has the expected decrease in LC retention due to new polar functional interactions with the solvent. However, O-demethylation of cisapride creates a new intramolecular hydrogen bond and a smaller CHIbt.

Other Biotransformation Reactions

Other biotransformation reactions are not the focus of this paper- their structures will typically be obvious from the mass change or yield to MS/MS analysis. Biotransformation which involves loss of the positively charged amino group will be especially hard to model but will likely lead to an increase in retention time.

Discussion

Note that there is considerable variability in the CHIbt values for a particular biotransformation but some general conclusions can be made:

Aliphatic oxidation most remote from the other polar functionalities causes the greatest loss of CHI.

Aromatic oxidation is generally a smaller CHI effect than aliphatic.

Nitrogen atom oxidation typically increases CHI.

Biotransformation which introduces an intramolecular hydrogen bond results in a smaller change in CHI than that predicted by clogP.

The CHI effects of multiple biotransformations are additive.

The graphical abstract takes the average CHIbt for the important mono-oxidations and sets a bandwidth at one standard deviation. The extracted ion chromatogram for the mono-oxidation metabolites of imipramine is overlaid. There is 1 dominant and two minor metabolites. These have CHIbt values of 2.3,-14.1 and -18.9 (supplemental Excel file, worksheet 4). Inspection of the color coded CHIbt regions would indicate that the first is very likely N-oxidation while the latter two could be hydrogen bonded aliphatic or aromatic oxidation. These hypotheses can then be taken into the evaluation of the CID spectra [41], simplifying the interpretations.

The prediction of RT would benefit from further improvements to QSRR modeling [24]. The simple rules described here can supplement common metabolite identification methods to substantiate the results and minimize the need to prove a compound’s structure by scale-up incubation plus NMR analysis or chemical synthesis.

Conclusion

Tools for predicting the MS/MS [42] spectra and RT behavior of drug metabolites will allow for the creation of a database of possible structures. A similar development occurred in proteomics, where the process of identifying peptides by comparing theoretically generated predicted MS/MS spectra and RTs obviated the need for unavailable reference standards. The prediction of LC RT can greatly aid in structure determination in related fields such as untargeted metabolomics, natural products, forensics and environmental analysis [43]. The prediction of electrospray LC-MS/MS sensitivity [44] when reference standards are unavailable is daunting, but knowledge of the percent acetonitrile at elution (one interpretation of the CHI value) could assist such a calculation.

Recent developments with retention indexing schemes raise the hope of interlaboratory reproducibility of reverse phase retention behavior. True interlaboratory reproducible reverse phase behavior could supply compound identification criteria useful for cases in which species have identical mass spectral behavior. Much more effort is needed in defining the protocols for measuring reverse phase retention reproducibly and for predicting RT from a structure. One advantage we have in this work is that an additional standard, the parent drug, is present in all samples, thus giving more confidence in the reproducibility of the CHIbt values. We hope to extend this approach to other areas (such as identification of glucuronidation position) where mass spectrometry cannot always provide sufficient structure information. We have herein proposed a paradigm in which a metabolite’s mass is first defined via high-resolution MS, followed by a prediction of LC RTs for the possible structures formed to assist in the determination of the exact metabolite structure.

Table 1

Selectivity parameters used in the hydrophobic subtraction model for various C18 reverse phase HPLC columns [21, 45].

Column Type	Hydrophobicity	Steric Interaction	Column Acceptor Hydrogen Bonding	Column Donor Hydrogen Bonding	Ion Exchange (pH 2.8)
Zorbax Eclipse Plus C18	1.03	0.007	–0.072	–0.02	–0.004
Luna C18	1.018	–0.02	0.07	0.008	–0.36
Polaris C18A	0.929	0.007	–0.227	0.062	0.149
ACE C18	1	0.026	–0.095	–0.006	0.143
Hypersil Gold C18	0.881	–0.002	–0.017	0.036	0.162
Inertsil ODS-2	0.994	0.032	–0.045	–0.005	–0.116
Xterra MS C18	0.985	0.012	–0.141	–0.014	0.133
Zorbax C18 (type A silica)	1.089	0.055	0.474	0.060	1.489

Table 2

List of compounds studied. New data was generated for the first 3. Literature data was utilized for the latter 17.

Atorvastatin	Buspirone	Glibenclimide
Imipramine	ZLN005	Terfenadine
Triclocarban	Rivoglitazone	Cisapride
RO9237	Clemizole	Diltiazem
Loperamide	Axitinib	Thioridazine
Pioglitizone	Quetipine	Dextromethorphan
4-Methoxy-alpha PVP	Nefazodone

Table 3

Changes to CHI for the compounds used in the study following different biotransformations (n is the number of occurrences in the supplemental file).

Biotransformation	Average CHIbt	Standard Deviation	n
N-Oxidation tertiary amine	4	3	8
-secondary amine	32		1
Aromatic oxidation	-17	4	8
-with HB	4	10	2
-heteroaromatic	-7	1	2
Aliphatic oxidation	-30	6	6
-with HB	-15	5	5
N-demethylation- basic amine	-2	2	5
-amide	-15		1
Aromatic O-demethylation	-18	1	3
Aliphatic O-demethylation with HB	-8		1
Alcohol oxidation to aldehyde	12		1
Alcohol oxidation to ketone	6	5	2
Alcohol oxidation to acid	-1	6	4
Sulfide oxidation to sulfoxide	-36	25	3
Sulfoxide oxidation to sulfone	17	10	1
Acid dehydration to lactone	8		1
Primary amine oxidation to alcohol	53		1
Cyclic amine oxidation to iminium	9	23	2
Cyclic amine oxidation to lactam	29	31	2