Crystalline Sponges as a sensitive and fast method for metabolite identification: Application to gemfibrozil and its phase I and II metabolites

Understanding the metabolism of new drug candidates is important during drug discovery and development, as circulating metabolites may contribute to efficacy or cause safety issues. In early phase of drug discovery, human in vitro systems are used to investigate human relevant metabolism. While conventional techniques are limited in their ability to provide complete molecular structures of metabolites (liquid chromatography mass spectrometry (LC-MS)) or require larger amount of material not available from in vitro incubation (nuclear magnetic resonance (NMR)), we here report for the first time the use of the crystalline sponge method (CS-XRD) to identify phase I and phase II metabolites generated from in vitro liver microsomes or S9 fractions. Gemfibrozil was used as a test compound. Metabolites generated from incubation with microsomes or S9 fractions, were fractionated using online fraction collection. After chromatographic purification and fractionation of the generated metabolites, single crystal X-ray diffraction (SC-XRD) of crystalline sponges (CS) was used to identify the structure of gemfibrozil metabolites. This technique allowed for complete structure elucidation of 5'-CH 2 OH gemfibrozil (M1), 4'-OH gemfibrozil (M2), 5'-COOH gemfibrozil (M3) and the acyl glucuronide of gemfibrozil, 1-O- β -glucuronide (M4), the first acyl glucuronide available in the Cambridge Crystallographic Data Centre. Our study shows that when optimal soaking is possible, crystalline sponges technology, is a sensitive (ng amount) and fast (few days) method that can be applied early in drug discovery to identify the structure of pure metabolites from in vitro incubations. show MS to the of low like we that can used the of of their metabolites promising different of structure


Introduction
During the process of drug discovery and development, assessing the identity of drug metabolites is crucial to understand the contribution of metabolites to efficacy or safety (Nedderman, 2009;Zhang et al., 2009). Guidelines have been drafted by regulatory agencies (EMA, 2013;FDA, 2016) to help the pharmaceutical industry to apply the right strategy regarding metabolites in safety assessment. The key element in all these guidances is to protect patients and healthy human volunteers from potentially toxic metabolites with a focus on human metabolites that are either absent in preclinical species or circulate in these animals at much lower concentrations than in humans (Ma and Chowdhury, 2011). It is therefore of paramount importance during drug development to gain an understanding of metabolism of new chemical entities in human as early as possible. To achieve this, in vitro systems from human origin (e.g., microsomes or hepatocytes) are used to metabolize the new active pharmaceutical ingredients (API) in drug discovery research (Brandon et al., 2003;Fasinu et al., 2012). Samples generated from these systems are analyzed with LC-MS of different types (e.g., tandem mass spectrometry (MS 2 ) and tandem-time-of-flight) by drawing inferences from fragmentation patterns to assess metabolism (Prakash et al., 2007;Prasad et al., 2011).
Although MS techniques have reached extreme sensitivity, these techniques fail in many situations, e.g., to identify the complete structure of metabolites. This forces scientists to use NMR, a less sensitive technique, requiring larger amounts of material (usually µg to mg range) (Murai et al., 2004), which is therefore not always suitable for analyzing samples from in vitro origin.
Besides NMR and LC-MS, another technique would be single crystal X-ray diffraction (SC-XRD), a direct technique that provides structural information at the atomic level by measuring electron density as diffraction pattern of a single crystal (Massa, 2004). SC-XRD can even provide information on the absolute configuration of chiral centers but has the intrinsic limitation due to requirements of single crystals of suitable sizes and shape. The process of crystallization is a time-consuming procedure, that requires availability of sufficient amounts of materials and cannot be performed with amorphous solids, liquids or volatile analytes. One This article has not been copyedited and formatted. The final version may differ from this version. DMD Fast Forward. Published on May 20, 2020as DOI: 10.1124 at ASPET Journals on July 9, 2020 dmd.aspetjournals.org Downloaded from DMD # 91140 6 approach to overcome the limitation of crystallization is the application of microcrystal electron diffraction from powder (Jones et al., 2018;Kunde and Schmidt, 2019). This technique requires solid material at higher amount.
For structural elucidation of metabolites in solution, a method with the information density of X-ray crystallography and the sensitivity in the ng range is needed for human related metabolites at this early stage of drug development research. In 2013, Makoto Fujita published a new technique commonly known as "crystalline sponge method", which enables crystal structure determination without crystallization with only ng to few µg amount of analytes (Inokuma et al., 2013). In one specific case only 80 ng of material were sufficient to clearly elucidate the X-ray structure of the analyte and presumably even smaller amounts are possible for synchrotron X-ray diffraction instruments. The CS-XRD uses pre-existing crystals of porous metal coordination complexes [(ZnX 2)3·(tpt)2·x(solvent)]n (X = Cl; tpt=2,4,6-tris(4-pyridyl)-1,3,5triazine), which can absorb organic molecules in its pores and make them observable by conventional single crystal X-ray crystallography. Via diffusion ("soaking") the analyte ("guest") is absorbed into the pores of porous CS and regularly ordered by intermolecular, non-covalent interactions ("guest soaking") (Inokuma et al., 2013;Inokuma et al., 2016;Sakurai et al., 2017).
As a result, the repetitive positioning of the analytes in each pore of the framework serves for structural analysis via X-ray diffraction. For a successful CS-XRD experiment regularly order of the analyte is important. Functionalization of the API´s (e.g., hydroxylation, demethylation) increases the type of non-covalent interactions with the framework and therefore increases the chances of regular order. This even works for otherwise non-crystallizable compounds . The CS-XRD has been shown to be able to successfully elucidate the structure of API´s (Sakurai et al., 2017), products from biosynthesis catalyzed by cytochrome P450 enzymes (Morita et al., 2019), natural plant products (Wada et al., 2018) and metabolic products prepared by reductive dichlorination of an insecticide as well as keton reduction of a steroid hormone with baker´s yeast ), but has not been applied for drugrelated material generated by incubation of API´s with human or animal tissues before. This article has not been copyedited and formatted. The final version may differ from this version. DMD Fast Forward. Published on May 20, 2020as DOI: 10.1124 at ASPET Journals on July 9, 2020 dmd.aspetjournals.org In this study, we apply the crystalline sponge technology for complete structure identification of gemfibrozil phase I and phase II metabolites after incubation with liver microsomes or S9.
Rat systems were used, however, the formation of the same metabolites were confirmed in human liver S9. This article has not been copyedited and formatted. The final version may differ from this version.
To generate gemfibrozil glucuronide, human liver microsomes were treated with alamethicin (4 °C, 20 min, final concentration: 25 µg/mL), a channel forming peptide to increase the metabolite formation. The reaction was initiated by the addition of UDPGA (1.5 mM) as cofactor.
HPLC-MS. The supernatants were analyzed on an Acquity Arc HPLC system combined with an isocratic solvent manager, a single quadrupole MS and a fraction manager (Waters Corporation, Milford, USA). The LC system included a quaternary solvent manager, a sample manager and a 2998 photodiode array detector. The isocratic solvent manager was used with a 1:50 splitter. Samples were analyzed with electrospray ionization mass spectrometry in the negative ion mode. The following source parameters were used: probe temperature 600 °C; capillary voltage 0.8 V; cone voltage 3.0 V and sampling frequency 2 Hz. Ions were acquired in an MS acquisition range from m/z 100 to m/z 650 in continuum mode.
HPLC separations were achieved on two Chromolith Performance RP18e columns as first chromatographic system (100-4.6 mm; Merck KGaA, Darmstadt, Germany) and Purospher Star RP18e Hibar HR column as second system (100-2.1 mm, 2 µm; Merck KGaA, Darmstadt, Germany) at a column oven temperature of 25 °C. For the first purification elution was performed at a flow rate of 1.0 mL/min over a period of 30 minutes with a mixture of solvent A (water + 0.1% formic acid) and solvent B (acetonitrile + 0.1% formic acid). Compounds were eluted using the following conditions: mobile phase B increased from 0% to 25% over 7 min, followed by a 13 min linear gradient to 60% B and a linear gradient to 100% B in 5 min; returning to 0% B in 0.1 minutes and re-equilibration at 0% B for 4.9 min. The first 6 min of eluent were directed to waste to reduce contamination, as it contains mainly microsomes and buffers salts from incubation matrix. For the second purification elution was performed at a flow rate of 0.45 mL/min over a period of 15 minutes using the following conditions: mobile This article has not been copyedited and formatted. The final version may differ from this version. DMD Fast Forward. Published on May 20, 2020 as DOI: 10.1124/dmd.120.091140 at ASPET Journals on July 9, 2020 dmd.aspetjournals.org Downloaded from phase B increased from 0% to 80% over 10 min, followed by linear gradient to 100% B in 2 min; returning to 0% B in 0.1 minutes and re-equilibration at 0% B for 2.9 min. The eluate was subsequently split between the photodiode array detector and fraction manager (98%), and the mass spectrometer (2%). The MS flow was increased to 0.3 ml/min with a makeup solvent (90% water + 10% acetonitrile). The fractionation of the analytes was triggered by the mass values (m/z 265, m/z 279 and m/z 425) registered by the MS detector. Data acquisition and sample fractionation were performed using the software MassLynx 4.2 combined with FractionLynx.
Collected fractions were pooled, evaporated to dryness under nitrogen flow at 40°C and resolubilized in acetonitrile/water (40%/60%) for further purification. The fractions from the second purification step were evaporated to dryness for the soaking process.
UPLC-MS 2 . The supernatants were analyzed on an Acquity UPLC Class I system (Waters Corporation, Milford, USA) combined with a linear ion trap quadrupole (Qtrap 5500) mass spectrometer (AB Sciex LLC, Framingham, USA). Samples were analyzed with an electrospray ion source in the negative ion mode. Source parameters were source temperature 600°C and ion spray voltage -4.5 kV. UPLC separations were achieved on an Acquity UPLC BEH C18 column (1.7 µm; 2.1 x 50 mm; Waters Corporation, Milford, USA) at a column oven temperature of 40 °C. Elution was performed at a flow rate of 0.7 ml/min over a period of 5.2 minutes with a mixture of solvent A (water + 0.1% formic acid + 10 mM ammonium formate) and solvent B (acetonitrile). Compounds were eluted using the following conditions: mobile phase B increased from 30% to 50% between 0.1 min and 4.0 min, followed by a 0.6 min linear gradient to 100% B, returning to 30% B in 0.02 minutes and re-equilibration at 30% B for 0.58 min. After injection, the column effluent was directed to the waste for 0.2 min, to reduce contamination. The software Analyst 1.6.3 was used for data acquisition. This article has not been copyedited and formatted. The final version may differ from this version. DMD Fast Forward. Published on May 20, 2020 as DOI: 10.1124/dmd.120.091140 at ASPET Journals on July 9, 2020 dmd.aspetjournals.org
For guest soaking one of the prepared [(ZnCl2)3(tpt)2•x(cyclohexane)]n single crystals was used as crystalline sponge, and then transferred with 50 µL cyclohexane to the sample vial.
Reference material was prepared as 1 mg/mL solution in DME, 1 µL of the solution pipetted into the sample vial, the vial closed with a screw cap with septum seal (VWR, Darmstadt, Germany) and pierced with a syringe needle for slow evaporation of the solvent during guest inclusion. The samples were placed in an incubator at 50 °C for 1 day. Soaking conditions for gemfibrozil glucuronide were conducted similarly. The volume of cyclohexane was reduced to 20 µL and the incubator temperature was 25 °C. Incubation samples were prepared similarly to the reference material. Differences were that 40 µL cyclohexane was transferred together with the crystal into the glass vial, which contained the pooled metabolite material. 4 µL of DME was added subsequently. The guest soaking for gemfibrozil glucuronide was performed using 20 µL of cyclohexane at 25 °C for 1 day.
Single crystal X-ray diffraction measurements were conducted using a Rigaku Oxford Diffraction XtaLAB Synergy-R diffractometer using Cu-Kα X-ray radiation (λ = 1.54184 Å), equipped with a HyPix-6000HE/HyPix-Arc 150° Hybrid Photon Counting (HPC) detector (Rigaku, Tokyo, Japan) at a temperature of 100 K using a Cryostream 800 nitrogen stream (Oxford Cryostreams, UK). The software CrysAlisPro (Rigaku Oxford Diffraction, 2018) was used for calculation of measurement strategy and data reduction (data integration, empirical and numerical absorption corrections and scaling). Crystal structures were modeled using OLEX2 (Dolomanov et al., 2009), solved with SHELXT and refined using SHELXL (Sheldrick, 2015). First, the refinement of the framework was performed by assigning atoms, modeling of disorder, applying anisotropic refinement and adding of H atoms. The electron density of the guest molecules could safely be identified in the residual electron density after refinement of the framework. The amount of analyte in the sponge pores is modeled as occupancy of the This article has not been copyedited and formatted. The final version may differ from this version.  To obtain more information, gemfibrozil and its metabolites were analyzed with UPLC-MS 2 .

LC-MS analysis of gemfibrozil metabolites prepared by incubation with rat/human
Tuning of instrument parameters provided the optimal transition for 4'-OH gemfibrozil reference material and was also applied for detection of M1 gemfibrozil. Precursor ion and product ion were adjusted to measure M3 gemfibrozil. Collision induced dissociation of gemfibrozil and its glucuronide generated a main product ion of m/z 121.0 (2',5'-dimethylphenoxy moiety).  . These parameters need to be optimized for each analyte. Different combinations of these parameters should be tried during this soaking process. Therefore, first analysis was conducted with 1000 ng for each experiment of commercially available gemfibrozil, 4'-hydroxy gemfibrozil and gemfibrozil 1-Oβ-glucuronide reference material to find the optimal soaking conditions for our tool compounds. The optimized soaking conditions were then applied to its metabolites generated from incubation. The reference material of 4'-hydroxy gemfibrozil was initially used to examine if this technology can be used for polar hydroxy and glucuronide metabolites. This article has not been copyedited and formatted. The final version may differ from this version. DMD Fast Forward. Published on May 20, 2020as DOI: 10.1124 at ASPET Journals on July 9, 2020 dmd.aspetjournals.org Downloaded from DMD # 91140 15 Both, the parent API and the reference metabolites 4'-hydroxy gemfibrozil ( Figure 5B) and gemfibrozil 1-O-β-glucuronide could successfully be structurally elucidated by CS-XRD using an in-house XRD system. The crystal structure of gemfibrozil revealed two molecules with occupancies of 100% and the final R1 and Rint values were 11.25% and 1.81%, respectively.
The crystal structure of gemfibrozil 1-O-β-glucuronide revealed one molecule with occupancy of 100% and final R1 and Rint values of 6.99% and 1.87%, respectively. Due to disorder, conformation of the glucuronide, i.e. the oxane ring and its substituents cannot be determined.
The presence of aliphatic rings in unfavorable conformations is often observed in the CS-XRD, especially for cyclohexane molecules. Nevertheless, the position of glucuronidation at the carboxylic acid function of gemfibrozil can be confirmed.
The successful results demonstrate that the application of the CS method can be expanded to the compounds containing highly polar hydrophilic groups introduced through metabolism of API´s, e.g., using cytochrome P450 enzymes, despite the hydrophobic pore of the CS.
Next, the CS method was applied to metabolites obtained by incubation. Incubating gemfibrozil with rat liver microsomes for M1 and M2, rat liver S9 for M3 and human liver microsomes for M4 resulted in the formation of four metabolites in different quantities as described above. The reference material of 4'-OH gemfibrozil and gemfibrozil 1-O-β-glucuronide allowed the quantification of M2 and M4, whereas the amounts of M1 and M3 were estimated. For pooling of M1, M3 and M4 an incubation solution of 20 µM gemfibrozil was used and 200 µM gemfibrozil for M2. The direct use of the separated metabolites was not possible due to the sensitivity of the CS-XRD against impurities as already mentioned in previous publications Kai et al., 2018). The impurities resulting from the incubation matrix, that is necessary for preparation of metabolites by enzymatic reaction (e.g., salts, cofactor, microsomal stock solution), could not be separated using only one purification step. Therefore, the pooled samples were further purified using a second column. CS soaking experiments were conducted with final amounts in a range of 500 to 1000 ng per experiment.
The soaking experiments of the obtained metabolites M1-M4 were carried out under the same conditions as used for the reference material. The soaked CS were then measured via XRD This article has not been copyedited and formatted. The final version may differ from this version. DMD Fast Forward. Published on May 20, 2020 as DOI: 10.1124/dmd.120.091140 at ASPET Journals on July 9, 2020 dmd.aspetjournals.org Downloaded from and the crystallographic analysis clearly revealed the positions of hydroxylation and carboxylation. M1 exhibits relatively high electron density in close proximity to the 5'-methyl carbon atom forming a benzyl alcohol, which is further oxidized at the same position to form a benzoic acid (M3). M2 shows high electron density at position 4' of the 2',5'-dimethylphenoxy moiety. The electron density map of the benzene core from gemfibrozil and its metabolites M1-M3 are shown in Figure 5B. Two molecules of 5'-hydroxymethyl gemfibrozil and one cyclohexane were clearly assigned by its electron density in the asymmetric unit with occupancies of 100% ( Figure 5A). The modeled structure shows a R1 value of 7.58% and a value of 2.18% for Rint. One guest is stabilized by π-π stacking interactions between its aromatic ring and the tpt ligand of the framework [distance 3.63 Å, angle 5.83 °].
Crystallographic data of 4'-hydroxy gemfibrozil revealed one analyte molecule with occupancy of 44% (R1 7.56%, Rint 2.09%) and two cyclohexane molecules. M2 interacts with the framework by π-π stacking interactions between the aromatic ring and the tpt ligand [distance 3.54 Å, angle 7.90 °], as well as CH···Cl interactions [distance 2.50 Å] between the 5'methylgroup of the analyte and the ZnCl2 part of the framework ( Figure 6). The data of 5'carboxy gemfibrozil provided low electron density of the analyte but clearly confirmed the position of metabolism in comparison with the crystal structure obtained from 5'-hydroxymethyl gemfibrozil ( Figure 5B). The XRD data of M4 showed the formation of an acyl glucuronide by conjugation of the carboxylic acid moiety of gemfibrozil with the C1-hydroxy group of glucuronic acid. This confirmed the position of glucuronidation, but due to low electron density of the analyte, the reference material was used to illustrate the structure in this publication. M3 and M4 could be assigned with low electron density compared to the hydroxy metabolites because of the higher polar and nucleophilic groups. The refined structures of gemfibrozil and its generated metabolites are shown in Figure 3. Crystallographic data and ORTEP diagrams of the asymmetric unit of the framework and analyte are shown in the Supplemental Data. This article has not been copyedited and formatted. The final version may differ from this version. DMD Fast Forward. Published on May 20, 2020as DOI: 10.1124 at ASPET Journals on July 9, 2020 dmd.aspetjournals.org

Discussion
In summary, we were able to analyze the structure of three different phase I metabolites and one phase II acyl glucuronide of our tool compound gemfibrozil at atomic level with only ng amounts of substance using the crystalline sponge method. Data from MS analysis provided us information on the type of metabolism, but could not pinpoint the exact position of the hydroxylation, carboxylation or glucuronidation reaction. Only the crystallographic data showed that hydroxylation occurred at the 4'-C and 5'-CH3 positions of the aromatic ring, as well as further oxidation at position 5'-CH3 to form a carboxylic acid and direct glucuronidation at the carboxylic acid moiety of the alkyl chain. The Cambridge Crystallographic Data Centre did not contain crystallographic data of an acyl glucuronide before and could be extended through M4 gemfibrozil.
The need to identify the structure of metabolites meets the need of industry to assess the contribution of drug metabolites to efficacy, toxicity or drug-drug interactions. It also meets the requirements from regulatory agencies to ascertain that human metabolites are present in preclinical species used to assess drug safety. Especially acyl glucuronides play an important role in the assessment of drug-related risk factors, as they contain the risk to be potentially reactive metabolites. An intramolecular rearrangement process, so called "migration", leads to the formation of chemically reactive species which can covalently bind proteins (Zhong et al., 2015;Bailey and Dickinson, 2003).
The application of the crystalline sponge technology in structure elucidation is a convenient technology that can be applied in drug discovery to accelerate the metabolite identification processes and allow for using human-derived systems to identify human metabolites from in vitro incubation, years before clinical samples are available.
This technology is, however, in its infancy. Therefore optimization of soaking conditions (time, solvent and temperature) as well as assessing the purity required for compound is still needed.
The amount of analyte per experiment was used in excess, but could be reduced as in the original publication mentioned (Inokuma et al., 2013). It has also been described that not all compounds can be applied in this technology as the size of the molecules may prevent them This article has not been copyedited and formatted. The final version may differ from this version. DMD Fast Forward. Published on May 20, 2020 as DOI: 10.1124/dmd.120.091140 at ASPET Journals on July 9, 2020 dmd.aspetjournals.org Downloaded from from entering the pores and their polarity may prevent some molecules to dissolve in the hydrophobic solvents typically used during soaking. It was therefore unclear, if the technology would be applicable to drug metabolites resulting from phase I and especially phase II reactions. To answer this question, we used reference metabolites (M2 and M4 gemfibrozil) to confirm the ability of CS-XRD to assess the structure of phase I and phase II metabolites.
Beside these 2 metabolites, we have applied this new technology to identify the structure of the 2 other unknown metabolites (M1 and M3 gemfibrozil) confirming its application in drug development of new chemical entities where only the structure of the candidate molecule is known.
Our results show that the combination of CS-XRD with MS data offers a great opportunity for scientists in drug metabolism and pharmacokinetics to assess the structure of metabolites produced in low amount from in vitro studies. The technology is also able to provide information on stereochemistry offering a significant advantage over other techniques like MS or NMR. In conclusion, we show that CS can be used to assess the structure of traces of drug-like molecules as well as their metabolites showing promising application at different stages of drug development where complete structure elucidation is needed. Besides metabolites, this technology could also be applied to identify degradation products and impurities during drug product development.
This article has not been copyedited and formatted. The final version may differ from this version.    Centroids are shown as grey spheres in the center of the benzene ring. This article has not been copyedited and formatted. The final version may differ from this version. DMD Fast Forward. Published on May 20, 2020as DOI: 10.1124 at ASPET Journals on July 9, 2020