Jeffrey R. Enders

Lurasidone is an atypical antipsychotic that was approved by the FDA in 2010 to treat bipolar depression and schizophrenia. Like other antipsychotics, adherence to lurasidone is critical for successful disease treatment. Thus, therapeutic drug monitoring (blood testing) is often employed by clinicians to monitor adherence. Urine drug testing, with its advantages over blood testing, is another method used to confirm medication adherence. However, analytes used in blood testing are often very different than those used for testing in urine, where nonactive metabolites are often most prevalent. Choosing metabolites in urine that are relatively prevalent affords optimal method sensitivity, and thus improved testing results for adherence. To ensure optimal lurasidone adherence testing, an untargeted high-resolution mass spectrometry method was employed, using known positive human urine samples to identify the lurasidone metabolites and their relative abundance in urine. This testing identified a different primary urine metabolite from what has been reported in blood. The higher prevalence of this metabolite will improve lurasidone urine adherence monitoring.

	Medication monitoring has become increasingly important for successful treatment of patients with mental health diseases because adherence to treatment is generally poor, especially in the schizophrenic population (1–8). Urine has become an alternative to blood or plasma medication monitoring due to its noninvasive nature and ease of collection. Whereas blood or plasma drug testing usually involves the identification and quantitation of the parent compound or active metabolites, or both, the success of urine drug testing (UDT) is largely dependent on analysis of any metabolites of the parent compound. Although the parent compounds may be present in urine, often they are at very low concentrations relative to metabolites, and therefore do not provide the sensitivity required for medication monitoring. Urine metabolites are often predicted from identification in blood, plasma, or specific testing methods, such as gas chromatography-mass spectrometry (GC-MS), extractions, radioactivity, and using in vitro or animal samples. However, it has been shown that these methods are not always successful in identifying the most abundant urinary metabolite (9–12). Without suitable metabolites to test, a negative UDT result could prompt a clinician to alter treatment for a patient when treatment need not be altered. Therefore, a generic, untargeted approach is useful for the successful identification of urinary metabolites suitable for highly sensitive medication monitoring. Liquid chromatography–high resolution mass spectrometry (LC–HRMS) provides a sensitive and nonspecific detection method for setting up such an experiment.

	Lurasidone (Latuda) is an atypical antipsychotic that was approved for the treatment of acute symptoms of schizophrenia (13,14) and bipolar depression (15,16) in 2010 and 2013, respectively. It is commercially available as 20 mg, 40 mg, 60 mg, 80 mg, and 120 mg tablets, and is typically prescribed or administered at 40 or 80 mg per day. It is absorbed after oral administration with a bioavailability of 9–19%. Dosing is designed to be with food, which can increase the bioavailability by 100%. The mean elimination half-life is 18 h. Steady state serum concentrations for lurasidone are typically achieved after seven days of dosing (17–20). Lurasidone is metabolized in the liver primarily by CYP3A4. Metabolism includes oxidative N-dealkylation, hydroxylation of the norborane ring, S-oxidation, and reductive cleavage of the isothiazole ring, followed by S-methylation. Nearly two dozen metabolites of lurasidone have been previously identified, and only ~9% of the dose is excreted in urine (17–20). Typically, adherence to lurasidone therapy is monitored by evaluating levels of lurasidone and M11/ID-20219 (one of its metabolites) that were each predicted to be present in urine at approximately 12 and 24%, respectively. The structures for lurasidone and many of the confirmed metabolite structures can be seen in Table I.

	Previously, we reported the identification of novel metabolites for monitoring aripiprazole, brexpiprazole, haloperidol, and quetiapine in urine that were not originally predicted (9–12). Because there are some similarities of these antipsychotics to lurasidone, we decided to determine if the urinary lurasidone compound(s) predicted from plasma studies were indeed the most abundant prior to development of a confirmation method. This work reports the identification of lurasidone and prevalent lurasidone metabolites in urine using LC–HRMS from patients prescribed lurasidone. Additionally, confirmation of the most prominent metabolites was tested in a validated, targeted, quantitative liquid chromatography–tandem mass spectrometry method (LC–MS/MS), which are at odds with current reports of urine metabolites (17–20).

 were purchased from Cerilliant (Round Rock, Texas). Hydroxylurasidone was a custom synthesis product purchased from 

C Molecular (Greensboro, North Carolina). All solvents, including methanol (optima grade), formic acid (88%), acetonitrile (optima grade), ammonium acetate (optima grade), and isopropanol (optima grade), were purchased from VWR (Radnor, Pennsylvania, USA). Drug-free human urine was acquired from UTAK Laboratories (Valencia, California). Standards for S-methyl lurasidone and S-methyl hydroxylurasidone were not commercially available, and synthesis requests were unsuccessful.

	Identification of lurasidone metabolites using LC–HRMS was completed on 13 authentic urine samples from patients who were prescribed the medication. After metabolite identification was complete, an LC–MS/MS confirmation was validated. An additional 56 patients were prescribed lurasidone at different doses, with specimens collected over three separate days for each patient used to confirm the accuracy of the method. These samples were provided voluntarily, and anonymously, to assist with the development of a lurasidone confirmation method. No identifying or demographic information was collected on these volunteers, other than the prescribed lurasidone dose. There was an alphanumeric code from the clinic that was provided to track the patients who provided samples over the course of the three separate days. None of the results were shared with the clinician to assist with treatment. Ameritox is accredited by the College of American Pathologists (CAP) and abides by CAP, Clinical Laboratory Improvement Amendments (CLIA), and Health Insurance Portability and Accountability Act (HIPAA) requirements. Due to the secondary analysis nature of this work and the absence of clinical conclusions, neither the United States Food and Drug Administration (FDA) nor other clinical trial review or approval was obtained by Ameritox. Writing this manuscript did not involve human subjects, as defined by the U.S. Code of Federal Regulations (45 

46.102); thus, an Institutional Review Board (IRB) approval of these specific research activities was unnecessary.

	Thirteen patient urine specimens (100 µL) were diluted 5X with 400 µL of a reference standard, (0.25 µg/mL of hydrocodone-d

 was used as an internal reference standard for all LC–HRMS injections, to guarantee successful injection of the sample, and provide a retention time marker. Prepared samples were injected (5 µL) and separated on a Phenomenex Kinetex Phenyl-Hexyl, 2.1 x 50-mm, 2.6-µm column (Torrance, California) at 50 °C, and analyzed on an Agilent 6530 Q-TOF (quadrupole time-of-flight mass spectrometer) with an Agilent 1290 LC system (Santa Clara, California). The LC–QTOF method conditions are detailed in a previous publication (12). A lurasidone control in drug-free urine (75 ng/mL) was run, along with the patient samples, to assist in positive identification of the parent compound, if present. No other standards were available or purchased to assist in identification, until a confirmation method was developed. Each sample was injected and analyzed twice.

	The MS-only data were processed using Agilent Mass Hunter Qualitative Analysis and PCDL (Personal Compound Database and Library) manager software. A database of lurasidone and 11 of its possible metabolites' chemical formulas (Table I) was compiled, and used to search against the samples. The software matched compounds based on retention time (if available), mass (±20 parts per million or ppm), the isotopic distribution pattern, and the isotopic spacing theoretically derived from the chemical formula. To be identified as positive and a potential lurasidone metabolite, a compound had to have consistent retention times across multiple patient samples when a known retention time was lacking; otherwise, the retention times had to be within ±0.05 minutes of a control. The mass accuracy had to be within ±20 ppm; and the composite score of the mass accuracy and isotopic features had to be ≥70 (out of a possible 100). Compounds that had the highest area counts were also ranked and noted as the most abundant. To assist with differentiation of structural isomers, such as M8/M9 and M10, fragmentation spectra were obtained and reviewed to identify which isomer was present at an identified metabolite peak, as needed.

	Hydroxylurasidone was received as a neat solid that was dissolved into methanol at a concentration of 1 mg/mL, and lurasidone was received as a 100 µg/mL methanolic standard. Hydroxylurasidone and lurasidone were combined and diluted into a methanolic stock that was then further diluted into normal, drug-free human urine, to reach the appropriate calibrator (5, 25, 100, 500, and 1000 ng/mL) and quality control levels (75 ng/mL). Lurasidone-d8, 1 mg/mL methanolic stock, was diluted to 900 ng/mL in 0.1% formic acid in water solution. A 100 µL aliquot of the sample (patient sample, calibrator, or quality control stock) and 400 µL of lurasidone-d8 internal standard in 0.1% formic acid were added to a vial. Vials were then capped and vortexed for 10 s prior to injection of 5 µL.

	Samples were analyzed by LC–MS/MS on a Waters Acquity UPLC Xevo TQ-MS system (Waters Corporation, Milford, Massachusetts), a Waters Acquity UPLC CSH Phenyl-Hexyl 2.1 x 50-mm, 1.7-µm UPLC column. The LC method and MS conditions can be found in Strickland and associates (12). Analyte transitions are listed in Table II. The acquisition method was run in dynamic multiple reaction monitoring (MRM) mode, in order to maximize the number of points across the various analyte peaks. The validation of this method followed CAP and CLIA guidelines (21–25), and an internal SOP (standard operating procedure) that has been described in detail elsewhere (26). It should be noted that, due to the lack of standards for S-methyl lurasidone and S-methyl hydroxylurasidone, they were unable to be validated, and estimates of their concentration were made by comparing the quantiative peak area ratio to the lurasidone calibration curve. These compounds were included as a proof of concept, to show their estimated prevalence and relative importance for lurasidone compliance in UDT for when standards might be available. Also, due to the lack of standards, the transition parameters were estimated from hydroxylurasidone and are not optimized.

	The metabolite identification from the 13 patients analyzed by LC-QTOF can be seen in Table III. Compounds identified with the highest confidence (>90%) are highlighted in green, while less confident (>70% but <90%) compounds are highlighted in yellow. The three most abundant compounds for each specimen and replicate are noted with a star-asterisk in the respective square. For unidentified or not confidently identified compounds (<70%) for a given specimen replicate, the field is blank. It is clear that, although lurasidone was identified in almost all of the samples, it was not consistently among the most abundantly identified compounds. Metabolite M11, the predicted major metabolite, was rarely confidently detected in these samples. Instead of lurasidone and M11, metabolites M21 (S-methyl lurasidone), M22 (S-methyl hydroxylurasidone), and isomer M8/M9 (hydroxylurasidone), or isomer M10 (lurasidone sulfoxide), were frequently detected. To determine whether hydroxylurasidone or the lurasidone sulfoxide (isomers) was present, the collected fragmentation data from the QTOF were analyzed, and are shown in Figure 1. The identification of a peak at 

182 (red circle in Figure 1) confirmed the isomer as hydroxylurasidone by indicating a fragmentation of the hydroxylated norborane ring. If the identity was the lurasidone sulfoxide, expected fragmentation peaks of 

 152 or 237 from the oxidized sulfur atom on the isothiazole ring structure would be present. Additionally, the unhydroxylated norborane ring would have an expected 

 of 166. The absence of those expected peaks (

 152, 237, and 166) in the spectra confirms the identity of the metabolite as hydroxylurasidone, and a custom synthesis of the molecule was requested to validate a confirmation method. S-methyl lurasidone and S-methyl hydroxylurasidone were also requested as custom synthesis products, but attempts to synthesize for the method were unsuccessful.

	Upon receiving the hydroxylurasidone standard, an LC–MS/MS method was developed and validated. The results of validation of lurasidone and hydroxylurasidone are shown in Table IV. Although S-methyl lurasidone and S-methyl hydroxylurasidone were included in the method, without standards, validation was unable to be completed, and is the reason for their exclusion from Table IV. To ensure the ability to successfully detect and quantify lurasidone and hydroxylurasidone in patient specimens, samples from 56 additional patients (from three separate collection days) were provided for testing with the validated method. The results of these patient analyses are summarized in Table V, and separated by the prescribed dose. It is clear that testing for hydroxylurasidone helps with positive confirmation of taking lurasidone medication. It also appears, from the estimated concentrations of the S-methyl lurasidone and S-methyl hydroxylurasidone, that confirmation would be easier with these metabolites, because they are more abundant than both lurasidone and hydroxylurasidone. However, the lack of standards for these compounds makes it impossible to currently validate a method for reporting UDT results for these compounds.


	Figure 1: Fragmentation data from sample 8 Identified M8-M9-M10 Peak.

	The advantages of HRMS analysis have been reviewed in the literature, including the extreme selectivity of such methods (10–12,27–30). Using this method, authentic urine samples of human subjects who were known to be taking chronic doses of lurasidone were tested for the presence of lurasidone and 11 possible metabolites. Due to the high mass resolving power and low mass error on the QTOF, compounds that have similar mass to charge ratios, but different chemical formulas, were differentiated with the searching algorithm (hydroxylurasidone and S-methyl lurasidone). Also, by eliminating extraction preparation methods, compound loss was mitigated. Using liquid as opposed to gas chromatography also helped ensure that compounds with low volatility can still be accurately analyzed. Surprisingly, neither of the predicted major urinary metabolites M5 nor M11 were found to be consistently excreted through human urine in large detectable amounts. This result could be due to the fact that M5 and M11 both have a carboxylic acid moiety that would be a possible glucuronidation target. This was considered, and the glucuronidated versions were searched for in patient samples and poorly identified. While this may indicate that the glucuronide metabolites of M5 and M11 are not present in urine in any appreciable amount, it could also be due to the poor ionizability of glucuronidated compounds. To confirm, an additional hydrolysis study could be completed to see if there is an increase in the prevalence of M5 and M11 in patients after hydrolysis. However, with the significant presence of hydroxylurasidone (M8/M9), S-methyl lurasidone (M21), and S-methyl hydroxylurasidone (M22), it seemed unnecessary to pursue hydrolysis as a means for analyzing for lurasidone compliance.

	The presence of these metabolites in urine was not well predicted from results in blood. Hydroxylurasidone was estimated to have a prevalence of 2.8% as the M8 isomer, and 0.4% as the M9 isomer; S-methyl lurasidone and S-methyl hydroxylurasidone were not predicted at any measurable amount (17–20). It does appear that the estimate of lurasidone at 12% might be reasonable, as all but one of the 13 patients in metabolite discovery had detectable amounts of the parent compound (17–20).

	To better understand the relative amounts of each metabolite, lurasidone, hydroxylurasidone, S-methyl lurasidone, and S-methyl hydroxylurasidone present in urine, the results from the 56 patients used during method validation were analyzed. These results in Table V show that hydroxylurasidone, S-methyl lurasidone, and S-methyl hydroxylurasidone are present at approximately 2x, 7x, and 5x times, respectively, relative to lurasidone. All of the compounds show a general increase in concentration and percent positivity rate with increasing doses. It is clear that S-methyl lurasidone and S-methyl hydroxylurasidone provide slightly better positivity correlations at lower doses, but hydroxylurasidone does appear to provide enough benefit to help compensate for lower prevalence of lurasidone. Therefore, with the lack of available standards for S-methyl lurasidone and S-methyl hydroxylurasidone, hydroxylurasidone was validated to assist in UDT for lurasidone compliance.

	We successfully identified prevalent lurasidone metabolites in urine. Based on those identifications, we successfully validated a method for the purpose of UDT monitoring of lurasidone. The hydroxylurasidone metabolite provides benefit for lurasidone UDT monitoring by being more prevalent in the urine than lurasidone by ~2x, and providing more consistent positivity correlation at lower lurasidone doses. Although other metabolites are present in the urine in large concentrations, standards for those compounds are not available at this time. However, the proof-of-concept work with S-methyl lurasidone and S-methyl hydroxylurasidone shows that they are ~5x and ~7x greater in abundance than lurasidone, respectively, and would provide even better positivity correlation at low doses.

	(1) M. Ko, and T. Smith, Urine Drug Monitoring in Patients on Prescribed Antipsychotic Medications. 29th Annual US Psychiatric and Mental Health Congress Oct 21-24 (

http://www.ingenuityhealth.com/wpcontent/uploads/2016/11/Mental_Health_Populations_Study_US_Psych_2016_Poster.pdf

	(2) A.E. Cooper, P. Hanrahan, and D.J. Luchins, 

	(3) C.R. Dolder, J.P. Lacro, L.B. Dunn, and D.V. Jeste, 

	(4) S. Offord, J. Lin, D. Mirski, and B. Wong, I 

	(5) D.I. Velligan, F. Lam, L. Ereshefsky, and A.L. Miller, 

	(6) D.I. Velligan, Y-W. F. Lam, D.C. Glahn, J.A. Barrett, N.J. Maples, L. Ereshefsky, and A.L. Miller., 

	(8) R.A. Millet, P. Woster, M. Ko, M. DeGeorge, and T. Smith, Adherence to Treatment with Antipsychotic Medications Among Patients with Schizophrenia, Major Depressive Disorder, or Biopolar Disorder. Poster Presentation. (US Psychiatric and Mental Heal Congress (USPMHC) San Diego, CA,2015).

	(9) J. McEvoy, R.A. Millet, K. Dretchen, A.A. Morris, M.J. Corwin, and P. Buckley, 

	(10) J.R. Enders, S.G. Reddy, E.C. Strickland, and G.L. McIntire, 

	(11) O.T. Cummings, E.C. Strickland, J.R. Enders, and G.L. McIntire, 

	(12) E.C. Strickland, O.T. Cummings, A,A. Morris, A. Clinkscales, and G.L. McIntire, 

	(14) S. Caccia, L. Pasina, and A. Nobili, 

	(15) R. Franklin, S. Zorowitz, A.K. Corse, A.S. Widge, and T. Deckersbach, 

	(16) M. Ostacher, D. Ng-Mak, P. Patel, D. Ntais, M. Schlueter, and A. Loebel, 

The World Journal of Biological Psychiatry

http://doi.org/10.1080/15622975.2017.1285050

	(17) Product Information: LATUDA oral tablets, lurasidone hydrochloride oral tablets. Sunovion Pharmaceuticals Inc. Marlborough, MA (2018).

Disposition of Toxic Drugs and Chemicals in Man 

(Biomedical Publications, Seal Beach, CA, 11th Ed., 2010), pp 1232–1233.

	(19) Sunovion Pharmaceuticals Inc. Center for Drug Evaluation and Research Approval Package (Lurasidone). 

https://www.accessdata.fda.gov/drugsatfda_docs/nda/2013/200603Orig1s010.pdf

	(20) T. Ishibashi, T. Horisawa, K. Tokuda, T. Ishiyama, M. Ogasa, R. Tagashira, K. Matsumoto, H. Nishikawa, Y. Ueda, S. Toma, H. Oki, N. Tanno, I. Saji, A. Ito, Y. Ohno, and M. Nakamura, 

	(21) Association of Public Health Laboratories. CLIA-Compliant Analytical Method Validation Plan and Template for LRN-C Laboratories. December 2013.

	(22) F.T. Peters, O.H. Drummer, and F. Musshoff, 

 (AACC Press, Washington, DC, 2nd edition 2003) pp 114–115.

	(24) U.S. Department of Health and Human Services, Food and Drug Administration. Guidance for Industry-Bioanalytical Method Validation (2001).

	(25) National Laboratory Certification Program (NLCP). Manual for Urine Laboratories. October 2010.

	(27) J.M. Colby, K.L. Thoren, and K.L. Lynch, 

	(28) E. Partridge, S. Trobbiani, P. Stockham, T. Scott, and C.A. Kostakis, 

	(29) J.M. Colby,K.L. Thoren, and K.L. Lynch, 

	(30) S.K. Manier, A. Keller, J. Schaper, and M.R. Meyer, 

 is with Ameritox, LLC, in Greensboro, North Carolina. 

 is with the Molecular Education, Technology and Research Innovation Center of North Carolina State University, in Raleigh, North Carolina. 

 is with Premier Biotech, in Minneapolis, Minnestota. Direct correspondence to: 

Articles

For lurasidone treatment adherence testing, an untargeted high-resolution mass spectrometry method was employed, using known positive human urine samples to identify the lurasidone metabolites and their relative abundance in urine.

Determination of the Relative Prevalence of Lurasidone Metabolites in Urine Using Untargeted HRMS

Systems-wide measurements in life sciences research have considerable promise in areas ranging from personalized medicine to understanding dominant or prevailing biological processes. However, formidable challenges remain in the comprehensive analyses of complex living systems, such as the necessity for detection limits that encompass wide concentration dynamic ranges while maintaining molecular specificity, sensitivity, and throughput. Recent work has demonstrated the ability to address several of these challenges with multidimensional separations using gas-phase ion mobility–mass spectrometry (IM-MS). IM-MS provides the capability to analyze a wide breadth of biomolecular classes on a single instrument simultaneously by separating the molecules according to their apparent surface area (equating roughly to their size) as well as their mass-to-charge ratio. The addition of a size separation parameter distributes the molecular classes into different regions of conformation space based upon differences in the prevailing intermolecular folding forces. This distribution negates the predominance of endogenous and exogenous matrix interferences experienced with traditional MS and provides a rapid (millisecond) means for separating biomolecules similar to liquid chromatography (LC)–MS approaches. Because separations are performed following ionization, additional dimensions of molecular information can be obtained through combining LC–IM-MS and gas chromatography (GC)–IM-MS. In this report, IM-MS is outlined as a separations method, several examples of the utility of IM-MS for complex biological measurements are illustrated, and the implications of this approach for systems biology research are discussed.

Systems biology seeks to describe the function of a biological system using a holistic, multiscale approach. This approach encompasses the analyses of molecular classes such as the genome, transcriptome, proteome, and metabolome (among others) of a biological system. Most modern approaches choose to either rapidly sample a single aspect (for example, the electrochemical detection of glucose and the use of green fluorescent protein as a means of monitoring translation) or infrequently sample a more comprehensive set of data. Intermittent sampling results in a fractional snapshot of the system, upon which biological inferences must be placed. When considering the turnover rate of certain enzymatically catalyzed reactions (for example, millimolar per second rates for ATP, ADP, and cytosolic glucose) or the rate of ribosomal translation (that is, six to nine amino acids per second for a eukaryotic cell), for example, single minute sampling resolution is considered coarse (1–3). The rapid acquisition rate of time-of-flight mass spectrometry (TOF-MS) measurements allows for sampling at a timescale that is relevant for a greater number of biological processes. However, MS alone lacks the necessary peak capacity for systems biology analyses, even when considering high-resolution methods that can provide a resolving power greater than ~50,000 (4,5). This limited peak capacity may be alleviated through the coupling of mass spectrometry to one of several additional separations methods, notably ion mobility (IM), which is a post-ionization separation that resolves ions based on their relative size-to-charge ratio and occurs on the order of milliseconds (6,7). Nevertheless, the ability of MS and IM-MS to perform proteomic and metabolomic analyses is unquestionable (8,9). The sensitivity, dynamic range, and ability to obtain accurate mass measurements, among other figures of merit, are the stronger suits of IM-MS. 

In this article we describe the utility of IM-MS as a platform for systems biology research, the data that are acquired from these analyses, how it is interpreted, the strengths and limitations of the technology, and the outlook for future possibilities. 

Measuring the gas-phase electrophoretic mobility of ions, although popularized by the advances of electrospray ionization (ESI) and matrix-assisted laser desorption–ionization (MALDI) in the 1980s (10–12), really came to the forefront of separation science in the 1970s (13–17). There are currently several different forms of IM; however, all operate under similar guiding principles. Foremost, a mobility or separation cell is pressurized with a neutral background gas. The gas-phase ions from the sample are introduced into the cell and traverse the cell under the influence of a weak electric field at a rate inversely proportional to the number of collisions they experience with the background gas. Thus, smaller ions (for example, small-molecule metabolites) will traverse the cell with fewer collisions than large ions (for example, proteins). To a first approximation, these collisions are assumed to be completely elastic and the conditions of the cell (that is, electric field, temperature, and gas number density of the ion cloud) are tuned such that the collisions are of low enough energy to avoid inducing any structural or chemical changes. Thus, the elution order of ions is based upon their gas-phase packing density (that is, the physical density to which the individual molecule is folded and packed) in the gas-phase. 

The simplicity of the design of IM makes the technology versatile and robust enough to be found in the commercial and military sectors, in which it is used for detection of explosive compounds. In the research laboratory, the millisecond timescale of the separations facilitates the multiplexed combination of ion mobility with higher repetition-rate sampling techniques such as MS. The addition of IM to MS can be likened to the coupling of liquid chromatography (LC) or gas chromatography (GC) to MS, where pre-ionization separations (GC and LC), which inherently have a longer separations timescale (minutes to hours), interface with the higher sampling rate MS (for example, microseconds for time-of-flight MS), allowing many spectra to be taken for a given LC or GC separation. The resulting data comprise an LC trace with an associated MS spectrum acquired at subsecond intervals, which is typically displayed in two dimensions (Figure 1a). The hydropathic separation of LC is highly orthogonal to the 

 separation of MS and yields a spectrum where the molecular signals occupy a large region of separations space. IM-MS is a similar marriage of two temporally disparate separations, with IM dispersing ions based upon their relative gas-phase size and charge state and occurring on the order of milliseconds, allowing for hundreds of MS spectra to be taken for a given IM separation. These data produce two-dimensional spectra, as shown in Figure 1b, with the mass of the ion (

) displayed on the abscissa and drift time (the time it takes the ions to traverse the ion mobility drift cell in milliseconds) on the ordinate. The drift time of an ion, dictated by its gas-phase packing density, is not highly orthogonal to the mass differentiation of MS. This results in more concentrated occupation of separations space. 

 Figure 1: A comparison of the orthogonality of (a) LCâMS and (b) IM-MS.			

Advantages of the IM dimension include the ability to reduce chemical noise (by partitioning inherent chemical noise to other areas of two-dimensional conformation space) and to perform separations on the order of milliseconds (providing no significant reduction in acquisition time). The millisecond time-scale of these separations also allows for the addition of an LC- or GC-based separation before the mobility separation (yielding a three-dimensional data set). IM is also capable of separating analytes based upon their charge state and providing insight into prevailing intermolecular forces by effectively measuring gas-phase packing density, the implications of which will be discussed below. 

When compared to LC–MS data, two-dimensional IM-MS data exhibit several prevalent dissimilarities. Chief among these is the high degree of correlation seen in the two-dimensional IM-MS data (Figure 1). This correlation is inversely related to the orthogonality of the two measurements (18). The fact that the mass of an ion is intrinsically related to its size precludes ion mobility from being highly orthogonal to MS. Therefore, where LC and GC sample the chemical properties of polarity and volatility, which are often orthogonal to mass-to-charge ratio, ion mobility correlates to ion gas-phase packing density, which is related to mass. This correlation serves as both a challenge and a unique advantage. The challenge lies in the fact that at low masses (<500 Da) most of the data occupy a narrow region of two-dimensional space. However, for biological samples at larger masses, the data separate into distinct lines of correlation according to molecular class that are the result of differences in intermolecular folding forces that affect gas-phase packing density (19–21). This packing density separation becomes especially advantageous when considering complex biomolecular samples in which the prevailing intermolecular forces present in biomolecules lead to the partitioning of lipids, peptides, carbohydrates, and oligonucleotides into separate regions of conformational space (Figure 2). Lipids are the least dense, followed by peptides and then carbohydrates, with nucleotides occupying the highest degree of gas-phase packing density. Historically, these have been termed trendlines or mass-mobility correlations and have been used for the rapid classification of unknown biomolecules.

Figure 2: (a) Correlation based on gas-phase packing density and mass gives rise to correlation lines dictated by the particular chemical class (singly charged species in this example). Theoretical representations of correlation lines are displayed. (b) Data illustrating these correlation lines. (Figure 2a is adapted with kind permission from Springer Science+Business Media: L.S. Fenn and J.A. McLean, Anal. Bioanal. Chem. 391, 906 [2008], Fig. 2(a). Figure 2b is adapted with kind permission from Springer Science+Business Media: L.S. Fenn, M. Kliman, A. Mahsutt, S.R. Zhao, and J.A. McLean, Anal. Bioanal. Chem. 394, 235 [2009], Fig 1(a).)			

Ion mobility, being that it is a post-ionization separation technique, is easily integrated with traditional pre-ionization separations such as LC or GC. This added data dimensionality has several key benefits. For example, Clemmer and colleagues (22) have recently described LC–ESI-IM-MS studies of the human plasma proteome whereby three-dimensional separations provided an enhanced dynamic range in concentration that was estimated to be 10

 . Data from LC-IM-MS multiplexed arrangements illustrate the data density of such experiments from complex biological samples. An example spectrum of plasma from whole rat blood is shown in Figure 3 as a three-dimensional scatterplot. Each ellipsoid in the spectrum represents a uniquely identified peak with the relative shapes of the ellipsoids in each dimension being governed by the associated resolution of each measurement axis. The relative size of each ellipsoid represents the relative intensity of the ion signal. Note that there are over 100,000 uniquely defined features contained within this dataset, compared with approximately 10,000 when performing IM-MS in the absence of LC. This represents only ~10

  of the possible molecules that could be uniquely specified in this scatterplot — another testimony of the peak capacity afforded by the three-dimensional separation.

Figure 3: Three-dimensional scatter plot of an LCâESI-IM-MS spectrum displaying individual features of plasma from whole rat blood.			

Additionally, mobility separations can be used to aid in the interpretation of fragmentation spectra. If ion activation, by means of collision induced dissociation (CID), is performed following mobility separation, product ions that result from fragmentation can be correlated to precursor ions based upon their elution time from the IM dimension (Figure 4). This differs from in-source fragmentation, which is not mobility aligned, because this fragmentation occurs before the mobility cell. In contemporary LC–MS approaches, ions of a particular 

 are selected and subsequently activated for fragmentation. In untargeted approaches, such as Waters's MS

 selection; rather, each LC increment is subjected to activation of all masses. If the LC separation produces coeluted peaks, fragment ion spectra can be difficult to assign because they contain fragment ions arising from multiple precursor species. However, with the added IM dimension these fragmentation spectra are correlated to the precursor based on mobility, and thus enable assignment of the fragment ions to respective precursors. 

Figure 4: Ion mobility mass spectra of the small carbohydrate, lacto-N-fucopentaose 1, illustrating two IM-MS modes. (a) Spectra from fragmentation before IM-MS analysis resulting in dispersion of fragment ion signal across both dimensions of data analysis. (b) Spectra resulting from two stages of fragmentation, one before the IM analysis and one immediately following the IM analysis. In this latter case, fragment ion signal is dispersed across the mobility dimension and secondary fragment ion signals align in the mobility dimension, resulting in simultaneous fragmentation of all ions resulting from the first stage of activation.			

Because of the large amounts of data generated in an IM-MS analysis, translating the data to meaningful information can be challenging. Data acquired from IM-MS separations used in conjunction with any pre-ionization separation must first be discretized to finite retention time, drift time, and mass-to-charge values. Data must be aligned in all dimensions for cross-sample comparison. Open-source and commercial software is available that addresses several of these issues (23–25). The use of internal calibrants and spectral feature alignment algorithms compensate for minor separation incongruencies. Additionally, isotopic envelopes must be collapsed into monoisotopic peaks and intensities normalized, commonly using total ion count. This ensures relevant and relatively quantitative sample-to-sample comparability. To date, software that defines peaks using mobility in conjunction with 

 and retention time is not widely available, as the processing and alignment of raw data in three dimensions are computationally expensive. However, developments in this area are forthcoming. The availability of open-source or commercially available software for peak identification will be necessary to fully utilize the potential of three or higher dimensions of separation. 

Presently, these data are often discriminated using multivariate statistical analyses such as principal component analysis (PCA) and partial least squares-discriminate analysis (PLS-DA). In essence, these methods seek to distill data to two or three principal components that best describe the variability of the data. It is important that this approach allows the analytes that contribute most significantly to differences in sample sets to be identified from tens of thousands of molecular signals — the PCA components each represent a particular spectrum of analytes that is representative of a specific difference between the two samples. In this manner, needles may be drawn from the haystacks of data present in very complex biological systems. This method has been employed extensively in metabolomic analyses and is considered a reliable method for untargeted analyses, allowing for the prioritization of peaks of interest for database searching using publicly available databases (26–28).  

The advancement of systems biology research lies in the integration of multi-omic measurement strategies (for example, genomics, proteomics, lipidomics, glycomics, and metabolomics) with high temporal bandwidth; in other words, the ability to analyze many aspects of a biological system and track the products of the various biomolecular machinery with up to the second (or greater) time resolution. This ensures a more complete representation of the biological system and is essential to assembling a systems-level understanding. A single instrument platform, such as ESI-IM-MS or other multidimensional analyses, represents the best case scenario for temporal resolution because of the limited amount of sample manipulation that is required, thus preserving the biology in an unperturbed state. This approach also facilitates the ability to analyze samples with a multi-omic approach.  

When analyzing a complex system it is common to collect static samples or fractions to assess biomolecular changes. This offline method of sample collection becomes increasingly daunting as greater temporal resolution is desired, prompting excessive amounts of time to collect, process, collate, and then analyze the samples. The development of an online analysis platform could circumvent this issue by negating the need to handle the samples at all. In fact, when working with a continuous measurement system such as ESI-IM-MS, a continuously perfused sample is ideal.  

Aside from the vast analytical complexity of sample material and timing, there is still one major issue: How does one collect frequent multisystem samples without perturbing or destroying the biological system being studied? One solution is to run numerous duplicate systems in parallel, sacrificing one population for every sample collected; this assumes that all the populations are exact biological replicates of one another, which is highly unlikely. Another attractive option, which has recently gained momentum, is the use of the exometabolome to infer the phenotypic state (29,30). The exometabolome, or metabolic footprint, comprises any and all material that is secreted from the cell. This means the exometabolome has the potential to include proteins, lipids, glycans, metabolites, and in some bacterial and viral cases, nucleic acid material.  

Work is currently underway to integrate poly(dimethylsiloxane) (PDMS) cell-trapping microfluidic devices (31–33) online with ESI-IM-MS to sample the exometabolome of cellular populations in real time (Figure 5) (34). In this method, media is typically perfused using syringe pumps. Cell trapping microfluidic devices are fabricated from PDMS. These devices (not drawn to scale in Figure 5) are roughly 2 cm in total length and are composed of 7000 cell traps that measure 18 µm wide and 18 µm deep. The interior space has a typical height of approximately 20 µm. Following perfusion of cell culture media across the cells, the sample effluent is desalted and subsequently analyzed by IM-MS. This method could serve as an option for the integrated and comprehensive online systems biology measurement of biological systems, thus overcoming the difficulties related to daunting sample collection and system preservation. One of the key remaining challenges facing this endeavor is the susceptibility of electrospray ionization to the presence of salt in cell culture media. It has long been known that nonvolatile salts, even at moderately low levels, can cause substantial signal suppression, and at the levels usually found in cellular media it is difficult to observe any analyte signal against the salt cluster-filled background noise (35–37). Online sample processing to remove these salts can take on various forms; however, online solid-phase extraction (34) and microdialysis-based (38–40) methods seem to hold the most promise. Additionally, other online sample processing steps can be added, such as online protein digestion (41–43) or filtration, to give emphasis to the desired biomolecular class. The challenge is to optimize the desalting process without compromising the temporal resolution of dynamic biological processes that can be provided by both the microfluidic devices and IM-MS analysis.

Figure 5: Schematic diagram of a system used for the online ESI-IM-MS analyses of the cellular exometabolome. 			

The future of systems biology research will lie not in the division of divergent and highly specified analyses of individual components of biological systems, but rather in the holistic and convergent simultaneous measurement of all parts of a biological system. It is the optimization of combining these aspects into a single experimental platform, while also overcoming the inherent difficulties, that will result in the greatest insight into these systems. An integrated IM-MS platform coupled to real-time experiments on small populations of live cells with the possibility of feedback (44,45) will likely be useful for the present and future direction of systems biology–level experiments. 

Financial support for this work was provided by the NIH (1RO1GM092218-01 and RC2DA028981), the U.S. Defense Threat Reduction Agency (HDTRA-09-1-0013), the Vanderbilt University College of Arts and Science, the Vanderbilt Institute of Chemical Biology (Pilot project grant), and the Vanderbilt Institute for Integrative Biosystems Research and Education. We thank Allison Price for her editorial assistance. 

 (1) M. Rizzi, M. Baltes, U. Theobald, and M. Reuss, 

 (6) S.J. Valentine, M. Kulchania, C.A.S. Barnes, and D.E. Clemmer, 

 (7) B.T. Ruotolo, K.J. Gillig, E. Stone, and D.H. Russell, 

 (9) D. Katja, A.A. Pavel, and D. Bruce, 

(14) F.W. Karasek,H.H. Hill, and S.H. Kim, 

(15) F.W. Karasek, H.H. Hill, S. Kim, and S. Rokushika, 

 (Springer-Verlag, New York, 1984), p. 274. 

Ion mobility–mass spectrometry (IM-MS) is outlined as a separations method, several examples of the utility of IM-MS for complex biological measurements are illustrated, and the implications of this approach for systems biology research are discussed.

Jeffrey R. Enders

Articles

Determination of the Relative Prevalence of Lurasidone Metabolites in Urine Using Untargeted HRMS

Advanced Structural Mass Spectrometry for Systems Biology: Pulling the Needles from Haystacks