Mediator-lipidomics: databases and search algorithms for PUFA-derived mediators

Lipid mediators (LMs) derived from PUFAs play important roles in health and disease. Databases and search algorithms are crucial, but currently unavailable, for accurate and prompt analysis of LMs via liquid chromatography-ultra-violet-tandem mass spectrometry (LC-UV-MS/MS). A novel algorithm and databases, cognoscitive-contrast-angle algorithm and databases (COCAD), were developed for the identiﬁcation of LMs based on the integration of standard MS/MS spectra with chromatograms and UV spectra. Segment naming and empirical fragmentation rules were introduced to determine MS/MS ion identities, along with ion intensities used by COCAD in matching the unknown to those of authentic standards. The structures of potential LMs without synthetic and/or authentic products as standards were identiﬁed by developing theoretical databases and algorithms based on virtual LC-UV-MS/MS spectra and chromatograms. The performance of these databases and algorithms was tested by identifying LMs in murine tissues. These results indicate that COCAD has many advantages for proﬁling and identiﬁcation of LMs compared with the conventional dot-product algorithm. —Lu, Y., S. Hong, E. Tjonahen, and C. N. Serhan. Mediator-lipidomics: databases and search algorithms for PUFA-derived mediators. J. Lipid Res. 2005. 46: 790–802.

. LC-UV-MS/MS can also provide more spectral information and criteria for those compounds with specific UV chromophores and assist in the elucidation of both known and novel LM structures. In this context, many LMs derived from PUFAs possess conjugated double bond systems that are critical components for their bioactions; each gives a characteristic UV spectrum (2)(3)(4)(7)(8)(9). Unesterified PUFAs frequently coexist in samples, and these compounds contain 1,4cis -pentadiene subunits that do not give a specific UV chromophore signature. For example, leukotrienes contain a conjugated triene chromophore and lipoxins possess a conjugated tetraene chromophore.
Mediator-lipidomic databases containing MS/MS as well as UV spectra and chromatographic profiles with appropriate search algorithms are imperative for accurate and timely analysis of LMs (12). Mediator-lipidomic analysis is summarized in Scheme 1 .
Although GC-MS can be used in many cases for structure identification, the required high column temperatures limit its use because many of the PUFA-derived LMs are thermolabile (7). Search algorithms for GC-MS electron impact ionization spectra have been well studied (13)(14)(15), as have those for both GC-MS chromatograms and mass spectra (16). The contrast-angle (or dot-product, ϭ cosine of the contrast-angle) algorithm is widely used. The contrast-angle is the angle between two spectra represented as vectors composed of ordered peak intensities (13). Because databases and search algorithms for LMs on comprehensive LC-UV-MS/MS are not yet available, we initially used MassFrontier™ (ThermoFinnigan) GC-MS mass spectral commercial software to construct a LM database and search algorithm. The search algorithm for Mass-Frontier is dot-product and was developed by Stein and Scott (13,14).
The dot-product algorithm in MassFrontier uses the intensities of all peaks in MS/MS as a vector for computation and is not concerned with the identities of ions, whether they are molecular ions, ions derived from molecular ions, or ions from interfering substances in the samples (13)(14)(15)(16). Using the ions generated from interfering background can decrease the percentage correct for best match of the search. Even the ions generated from unknown LMs contribute differently to the identification. For LMs, ions generated by cleavage of the carbon-carbon bond inside the carbon chain have greater diagnostic yield for determining the structures than the ions formed by the loss of water or CO 2 , which are common features of most PUFA-derived LMs (3,4,7,8,11).
The relationships between ESI-MS/MS spectra and LM structures have been studied extensively (2)(3)(4)(7)(8)(9). The identification and elucidation of the unknown structures of novel LMs by LC-UV-MS/MS are based on the relationships between structural features (such as functional groups and double bonds) and characteristics of the spectra and chromatograms (MS/MS ions, UV spectra, and LC retention times). On C18 reversed-phase LC, chromatographic retention times (CRTs) generally increase in the following order: trihydroxy LMs, dihydroxy LMs, monohydroxy LMs, and then the nonoxidized PUFA precursors (such as arachidonic acid) are eluted, depending on the mobile phase used. For positional isomers of hydroxy-containing LMs, the CRTs decrease as the hydroxy groups locate closer to the methyl group at the -end of the long-chain PUFA. For instance, the CRT of 5 S ,14 R ,15 S -trihydroxy-6 E ,8 Z ,10 E ,12 Eeicosatetraenoic acid (LXB 4 ) is usually shorter than that of 5 S ,6 R ,15 S -trihydroxy-7 E ,9 E ,11 Z ,13 E -trans -11-cis -eicosatetraenoic acid (LXA 4 ). The CRT of native LXA 4 (containing a 15 S -hydroxy at carbon 15) is shorter than that of the aspirin-triggered 15-epi-LXA 4 but greater than that of 11trans -LXA 4 , which is the natural all-trans -containing isomer of LXA 4 (17).
The band multiplicity and maximum absorbance wavelength ( max ) of UV spectra are additional class-or seriesspecific signatures of LMs, such as the presence of an asymmetric singlet band with max ‫ف‬ 235 nm for the conjugated diene present within the lipoxygenase pathways. These include monohydroxy-eicosatetraenoic acids (mono-HETEs); a triplet with max ‫ف‬ 270 nm for the leukotrienes, such as the conjugated triene within LTB 4 (5 S ,12 R -dihydroxy-6 Z ,8 E ,10 E ,14 Z -eicosatetraenoic acid); a triplet with max ‫ف‬ 300 nm for conjugated tetraene, as present within LXA 4 and LXB 4 ; and, for example, an asymmetric singlet with max ‫ف‬ 242 nm from the two conjugated dienes interrupted by a methylene group, as present in the double lipoxygenation product 5 S ,15 S -di-HETE (17). For compounds without a class and/or series specific chromophore or with max in vacuum UV range (actually undetectable on our LC-UV-MS/ MS instrument) such as 1,4cis -pentadiene-containing PUFAs as well as some prostaglandins, namely PGE 2 (11 ␣ ,15 Sdihydroxy-9-oxo-prosta-5 Z ,13 E -dien-1-oic acid) and PGF 2 ␣ (9 ␣ ,11 ␣ ,15 S -trihydroxy-prosta-5 Z ,13 E -dien-1-oic acid), which do not possess conjugated double bond systems (17,18), the UV chromophore component is not used for identification. When the conjugated tetraene of LXA 4 isomers is in the all-trans geometry, the max is shifted to 302 nm instead of 300 nm (17). The relationship between MS/MS spectra and LM structures is of interest and presented vide infra.
With current chemical analytical technologies, most LMs are identified manually by comparing the spectra and chromatographic behaviors acquired from sample tissues with those of authentic standards of known LMs; when au-thentic standards are not available, as in the case of novel LMs and their metabolites, basic chemical structures can be obtained on the basis of the relationship between structures and features of their spectra and chromatographic behaviors compared with those of synthetic and biogenic products prepared to assist in the assignment. We routinely identify LMs by comparing the unknown spectra (MS/MS, GC-MS, and UV spectra) and CRTs with those of authentic and synthetic standards if available or with a theoretical database that consists of virtual UV and MS/MS spectra and CRTs for discovering potentially novel LMs (3,4,12) if standards are not available. We initially developed a theoretical database and algorithm according to the relationships between LM structures and their spectral and chromatographic characteristics (12). The proposed structures of novel potential LMs in the theoretical databases were based on PUFA precursors and established biosynthetic pathways.
In this report, we focus on constructing mediator-lipidomic databases and search algorithms useful for the identification of LM structures using LC-UV-MS/MS with the following objectives: 1 ) to assemble a database using currently available mass spectral software; 2 ) to construct a cognoscitive-contrast-angle algorithm and databases (COCAD) to improve the identification of LMs using MS/MS ion identities that currently cannot be determined with available software; and 3 ) to develop a theoretical database and algorithm for assessing potentially novel and/or unknown structures of LMs and their metabolites in biologic matrices. It is quite meaningful to develop mediator-lipidomic databases and algorithms using ion trap mass spectrometers that are relatively cheaper and popular. Moreover, the fragmentation rules and patterns for CID spectra from triple-quadruple mass spectrometers, another popular MS instrument, are similar to those we encounter using the ion trap (7,19,20). The databases and algorithms described in this report should be a useful initiation that can be extended to other types of MS instruments.
We used a LUNA C18-2 (Phenomenex, Torrance, CA) column packed with 5 m spherical C-18 particles. The dimension was 100 mm (or 150 mm) long and 2 mm inner diameter. The column was kept in a column heater (30 Њ C). The LC system used a P-4000 quaternary LC pump (ThermoFinnigan). When the column was 100 mm long, the mobile phase (methanol-water-acetic acid, 65:35:0.01) flowed at 0.2 ml/min in C channel from 0 to 8 min, linearly ramped to methanol in D channel from 8.01 to 30 min, eluted as methanol for 5 min, and finally eluted as C again. When the column was 150 mm, the mobile phase (methanol-wateracetic acid, 65:35:0.01) flow rate was 0.2 ml/min through the C channel from 0 to 15 min, linearly ramped to methanol in D channel from 15.01 to 48 min, eluted as methanol for 8 min, and return eluted as C again.
UV spectra were acquired using a tandem photodiode-array UV detector (UV-6000; ThermoFinnigan) that scanned from 200 to 360 nm. Samples were injected manually with a 7125E injector equipped with a 100 l loop (Rheodyne, Rohnert Park, CA). The injection volume was 20 l, and the amount of each LM and related compounds in each set of analyses ranged from 0.5 to 100.0 ng.
The ion source conditions for the LCQ were set as follows: electrospray voltage, 4.3 kV; heating capillary, 230 Њ C and Ϫ 39 V; tube lens offset, 60 V; sheath N 2 gas, 80 units (near 1.2 l/min); auxiliary N 2 gas, 3 units (0.045 l/min). The ion trap analyzer typically scanned from m/z 200 to 800 in MS mode to determine the m/z of ESI-ion source-generated ions. The parameters for MS/MS were as follows: damping and collision gas, He (at ‫ف‬ 0.1 Pa); isolation width, m/z 1.5; collision time, 30 ms; normalized collision energy on base of the maximum available tickling voltage (5 V), ‫ف‬ 38% for prostaglandins and dihydroxy-and trihydroxy-containing acyclic eicosanoids, 42% for monohydroxy-containing eicosanoids, and 50% for PUFAs. The settings for data-dependent scan MS/MS were as follows: isolation width, m/z 1.5; normalized collision energy, 40%; minimum signal required, 100,000; minimum MS n signal required, 10,000; repeat count, 3; repeat duration, 0.5 min; exclusion duration, 1.5 min. Automated gain control was set at ‫ف‬ 2 ϫ 10 7 for MS and MS/MS full scan.
Deuterium-labeled PGE 2 ( d 4 -PGE 2 ), d 4 -LTB 4 , and d (n) -PUFA were routinely used in our laboratory as the internal standard to calculate recoveries during sample preparation and quantification. Deuterium-labeled eicosanoids were also routinely used to calibrate CRTs and account for potential daily variations in relative chromatographic behavior. CRTs for LMs were routinely within Ϯ 0.6 min.

Generation of eicosanoids by polymorphonuclear leukocytes and stimulated spleen
Results from LC-UV-MS/MS for eicosanoids acquired from biological samples were used to demonstrate the performance of the databases and algorithms constructed. One sample set was obtained by incubations of rabbit polymorphonuclear leukocytes ( ‫ف‬ 50 ϫ 10 6 cells per incubation) isolated from rabbit whole blood and stimulated with ionophore A23187 (15 M) and arachidonic acid (10 M, 20 min, 37 Њ C) as described (19). Samples were also analyzed from murine spleens (kindly provided by J. Aliberti and coworkers) prepared as described (20). The spleens were collected from mice injected with homogenate of Toxoplasma gondii (RH strain). Incubation of samples was stopped with two volumes of methanol, and samples were maintained at Ϫ 20 Њ C before isolation and further analyses. All samples were extracted using a solid-phase extraction technique (21). In short, samples rapidly preequilibrated at pH 3.5 were loaded onto solid-phase extraction cartridges (100 mg of C18 adsorbent per cartridge; Varian, Palo Alto, CA) and eluted with 8 ml of methyl formate (EM Science, Gibbstown, NJ). Samples were taken to dryness using a flow of nitrogen gas. Each sample was suspended in 20 l of mobile phase C and injected for LC-UV-MS/MS analysis.

Software
For LMs and related compounds for which synthetic and authentic standards were available, the databases were constructed with MassFrontier (version 2.0; HighChem, Ltd., San Jose, CA) (22), a commercial software designed for use with positive-ion MS spectra generated by electron impact ionization. We implemented cognoscitive-contrast-angle and theoretical algorithms and developed databases using Microsoft Visual Studio 6.0 for by guest, on October 5, 2019 www.jlr.org Downloaded from mass spectra obtained in the negative-ion mode. All of the spectra generated were processed with Microsoft Excel 2000.
UV patterns are simply presented in the database as max and divided into six clusters: ‫103ف‬ nm, ‫872ف‬ nm, ‫072ف‬ nm, ‫242ف‬ nm, ‫532ف‬ nm, and vacuum UV. In contrast, each MS/MS spectrum has dozens of ions and therefore is a much larger data set than a UV max . If the UV max is matched first, then only the MS/MS subdatabase having LM standards with this UV max , which is a fraction of the whole MS/MS database, needs to be searched. This type of narrowed search is faster than that through the entire MS/MS database. This will become more significant when more and more standard MS/MS data are entered into the database. We tested the route with MS/MS spectra before UV; the search results are the same as having MS/MS after UV data. If the UV pattern is unclear, we can just search the MS/MS and CRT to avoid the assignment errors.

Databases constructed with commercial mass spectral software containing LC-UV-MS/MS chromatograms and spectra measured from standard LMs
A mediator-lipidomic database using LC-UV-MS/MS spectra and chromatograms acquired from authentic LMs was constructed with GC-MS spectral software MassFrontier. The UV max of standard LMs were written into the subdatabase names and the CRTs were written into the LM names, so that MassFrontier could handle the acquired UV spectral results and CRTs for the identification of the unknown LMs according to the logic diagram in Scheme 2.
The demonstration of the efficacy of this database was assessed with the identification of eicosanoids biosynthesized by rabbit leukocytes using LC-UV-MS/MS (Fig. 1). For peak I in the inset of Fig. 1A, a subdatabase "mTOz 351Vacuum UV" was selected with the molecular ion of interest and matched UV max . The search showed that the MS/MS spectrum of peak I matched best with PGE 2 , with the highest matching score of 907 (Fig. 1B). Furthermore, the CRT of peak I and standard PGE 2 matched. Therefore, peak I was assigned as PGE 2 , consistent with the manual identification and assessment of MS and UV spectral and LC chromatographic features of PGE 2 .

COCAD: using MS/MS ion identities and intensities as well as UV spectra and CRTs for LM identification
Identification of mass spectral ions generated from LMs. The system with COCAD that we developed can be used to elucidate the fragmentation of LMs in mass spectrometry and to match unknown MS/MS spectra with those of synthetic and/or authentic standards. In the process of matching, the intensity of each peak is treated differently based on the ion identity. MS/MS ions are clustered into three types: "peripheral-cut" ions, formed by neutral loss of water, CO 2 , amino acid, or amines derived from functional groups linking to the LM carbon chain as hydroxy, hydro- peroxy, carbonyl, epoxy, carboxy, amino acid group, or amino group; "chain-cut" ions, formed by the cleavage of a carbon-carbon bond along the LM carbon chain; and "chain-plus-peripheral-cut" ions, formed by a combination of chain cut and peripheral cut. Molecular ions formed during ESI can easily be converted to peripheralcut ions in the MS/MS process. Similarly, chain-cut ions can also be readily converted to chain-plus-peripheral-cut ions (Scheme 3).
Typical chain-cut ions for LMs in MS/MS are formed by ␣-cleavage of the carbon-carbon bonds connecting to the carbon with a functional group directly attached (2-4, 7, 8). LMs readily undergo ␣-cleavage (7). We proposed the nomenclatures illustrated in the LXA 4 structure presented in Scheme 3 to systematically name the segments formed via chain cut and chain plus peripheral cut without concern for hydrogen shifts occurring during mass spectrometric analysis of PUFA-derived products. Each of these LMs derived from PUFA has a carboxyl terminus and a methyl terminus. An Arabic number was used to designate the position of the functional group on the LM carbon chain where the cleavage occurs. The uppercase letter right after the number indicates the side of the functional group on which the cleavage occurs: C is for cleavage on the carboxyl side of the functional group, and M is for cleavage on the methyl side of the functional group. Each cleavage can directly generate two segments. The lowercase letter that follows indicates the side of the cleavage on which the segment forms: c is for a segment formed on the carboxyl side of the cleavage, and m is for a segment formed on the methyl side of the cleavage. Segments formed through ␤-cleavage or ␥-cleavage toward the functional group are named by adding ␤ or ␥ between the uppercase and lowercase letters. Because three hydroxy groups of LXA 4 are located at C 5 , C 6 , and C 15 of the 20 carbon linear chain, ␣-cleavages can occur on the bonds at C 4 -C 5 , C 5 -C 6 , C 6 -C 7 , C 14 -C 15 , and C 15 -C 16 to generate 10 chain-cut segments (Scheme 3). All of the possible chain-cut, peripheral-cut, and chain-plus-peripheral-cut segments for LXA 4 are listed in the table in Scheme 3.
A MS/MS ion detected from LM samples in the negative-ion mode generally is formed from a specific segment with the addition or subtraction of hydrogen(s) caused by hydrogen shift during the cleavage. The charge (z) of the LM negative ion is usually equal to 1; therefore, the massto-charge ratio (m/z) of a LM ion is usually equal to its mass (m). Previous reports (7,8,11)  Modification of MS/MS ion intensities according to their identities. In nature, isotope 13 C constitutes 1.1% of elemental carbon. If the intensity of an ion with mass M (ion charge z is 1) containing only carbons, hydrogens, and oxygen(s) is 100, the contribution of 13 C to the intensity of the ion with mass (M ϩ 1) (the ion with mass 1 unit higher) is statistically equal to 100 ϫ 1.1% ϫ the number of carbons in ion M, which is listed in Table 2 is much less than 100 ϫ 1.1% ϫ carbon number. The 13 C contribution of ion M to the intensities of ions (M ϩ 1) and (M ϩ 2) was subtracted in the computation of the matching score for the COCAD and theoretical algorithms. If ion M could not be identified, the carbon number in the ion (M ϩ 1) or (M ϩ 2) is used because M, (M ϩ 1), and (M ϩ 2) generally have the same carbon number.
Chain-cut ions are most informative and can be diagnostic for determining LM structures such as the position of functional groups and double bonds. Generally, there are more chain-plus-peripheral-cut ions than chain-cut ions, because the former are derived from the latter by one or several neutral losses. It is likely that some chainplus-peripheral-cut ions have the same m/z even though they originate by different mechanisms and represent different structural features. Therefore, chain-cut ions are more specific than chain-plus-peripheral-cut ions for defining the LM structure. Peripheral-cut ions in MS/MS spectra are similar among LM isomers; therefore, they were not specific enough for differentiation of individual LM isomers.
According to the fragmentation rules, the nth MS/MS peak can be identified as one or several chain-cut (C) ions, peripheral-cut (P) ions, and/or chain-plus-peripheral-cut (CP) ions. The weighted intensity ( y I n ) of each identified ion is as follows: where y is the MS/MS ion type identified as C, P, or CP; I n Ј is the relative intensity of the nth peak in the MS/MS spectrum; is the number of chain-cut ions identified for the nth MS/ MS peak; is the number of chain-plus-peripheral-cut ions identified for the nth MS/MS peak; is the number of peripheral-cut ions identified for the nth MS/MS peak; and y W is the weight measuring the importance of the identified ion to determine the LM structure. It is 10 as C W and 1 as CP W or P W. The fingerprint features of chain-cut ions are used to define LM structure by multiplying their intensities by 10, which was determined to be the best among the values (2, 10, 20, and 100) tested. Weighted MS/MS ion intensities are used for COCAD and the theoretical system described below. represents the contribution of peripheral-cut ions to I n Ј ( ϭ 3 for peripheral-cut ions formed via the loss of one CO 2 from each molecular ion, ϭ 10 for peripheral-cut ions formed via the loss of one H 2 O from each molecular ion, and ϭ 1 for other peripheral-cut ions formed via multiple loss of CO 2 and/or H 2 O from each molecular ion). The assignment of values is arbitrary and based on the observation of relative intensities of peripheral-cut ions in MS/MS spectra of LMs. COCAD contrast angle. COCAD used a contrast-angle algorithm to match an MS/MS spectrum between sample and standards. The contrast angle is calculated as follows: U v is equal to C v , CP v , or P v for unknown spectrum to be identified; S v is equal to C v , CP v , or P v for standard spectrum.

(Eq. 6)
where v is the vth virtual ion and V is the total number for one type of virtual ion formed via chain cut, chain plus peripheral cut, or peripheral cut for a specific LM.
is equal to 1 if the nth MS/MS peak can be identified as the vth virtual ion formed via chain cut and is equal to 0 if not; or has a similar meaning but for ions formed via chain plus peripheral cut or peripheral cut; N is the total number of peaks in the MS/MS spectrum. D C is the dot-product between the virtual vectors of U (unknown sample) and S (standard) formed via chain cut; D CP and D P are the dotproducts for chain-plus-peripheral-cut and peripheral-cut ions, respectively. D C , D CP , or D P in equation 5 represents the similarity of ions formed via chain cut, chain plus peripheral cut, or peripheral cut between an unknown spectrum and a standard spectrum. None of them is greater than 1. The vth virtual ion is not used for the calculation of the corresponding D C , D CP, or D P if either U v or S v is 0. If every C v , CP v , or P v within the vectors is 0, then D C , D CP, or D P , respectively, is assigned the value 0.
The COCAD contrast angle in equation 6 represents how well the spectrum of the sample matches the standard: if it is 0Њ, the two spectra match exactly; if it is 90Њ, the two spectra do not match at all. The smaller the contrast angle between 0Њ and 90Њ, the better the match (13,24). The value is integrated and normalized from dotproducts D C , D CP , and D P (equation 6). The numeric coef- by guest, on October 5, 2019 www.jlr.org Downloaded from ficient 10 in equation 6 was found to be the best value (2,20, and 100 were also tested) that emphasizes the fingerprinting feature of chain-cut ions, because chain-cut ions are more important for determining the LM structure than are other types of ions. To normalize [(10 ϫ D C ϩ D CP ϩ D P ) Ϭ (11 ϩ CP )] in equation 6 to be no more than 1, 11 was used in the denominator of equation 6, and CP is equal to 1 if at least one MS/MS ion is identified as a chain-plus-peripheral-cut virtual ion or is equal to 0 if no such ion is identified. No chain-plus-peripheral-cut ion is identified in a few LM standard spectra. Therefore, CP is introduced in equation 6 to normalize the COCAD contrast angle to 0 when matching these types of spectra against themselves. Unidentified ions were excluded for matching in equations 2 to 6.
Integrating MS/MS spectra with UV spectra and chromatograms in COCAD to identify LMs. COCAD databases incorporate the LC-UV-MS/MS chromatograms and spectra measured from standard LMs and the segments formed by cleavage illustrated in Scheme 3. The features of UV spectra and chromatograms are represented with the UV max and CRTs, respectively. The search of COCAD databases was conducted stepwise, as depicted in the logical diagram shown in Scheme 2, from UV max , to MS/MS spectra, and then to CRTs. Using COCAD, we identified LXA 4 in murine spleen stimulated during exposure to T. gondii ( Fig.  2A, B). The search was narrowed to the subdatabase with UV max 301 nm ( Fig. 2A, middle inset) and molecular ion m/z 351, which were detected for the unknown peak II in the chromatogram (m/z 251 of MS/MS 351) for the stimulated spleen sample. The CRT (5.2 min) of peak II was then matched against the candidates with the highest MS/MS match scores ( Fig. 2A, left inset). For comparison, searching was also conducted with the system developed with MassFrontier (Fig. 2C). Both demonstrated that LXA 4 was the best match for the unknown peak II shown in Fig. 2A. However, COCAD and MassFrontier could not be used alone to identify the structures of unknowns without LC-UV-MS/MS chromatograms and spectra obtained for authentic and/or synthetic standards.

Theoretical database and search algorithm for identification of potentially novel LMs
Theoretical databases consist of the segments (Scheme 3), the UV max , and CRTs predicted for potentially novel LMs. Searching against a theoretical database is also performed stepwise as described in Scheme 2, from UV max , to MS/MS spectra, and then to CRTs. Equation 7 is the matching score for an unknown MS/ MS spectrum compared with the virtual spectrum based on the segments and the fragmentation rules. is the total number of chain-cut ions via ␣-cleavage formed from the f th functional group and matched to the nth MS/MS peak. F is the total number of functional groups in one LM. f is counted from the carboxyl terminus of LM.
summates the weighted intensities of MS/MS peaks identified as chain-cut ions formed from the f th functional group via ␣-cleavage. The smallest MS/MS ion detectable in ion trap tandem mass spectrometry is generally ‫ف‬m/z 95 for LMs with a molecular ion of ‫004ف‬ Da (25). To compensate for the bias caused by the inability to detect an MS/MS ion of m/z Ͻ 95, factors ( f T C Ϭ f A C ) 0.5 , ( f T CP Ϭ f A CP ) 0.5 , and (T P Ϭ A P ) 0.5 are used in equation 7. f T C (or f T CP ) is the total number of chain-cut (or chain-plus-peripheral-cut) ions formed from the f th functional group via ␣-cleavage. T p is the total number of peripheral-cut ions formed from one LM. f A C ( f A CP or A P ) is the fraction of f T C ( f T CP or T P ) representing the ions within the MS/MS Matching score is used in equation 7 for normalization to eliminate the impact on the matching scores of the total peak intensities in MS/MS spectra. The unknown MS/MS spectrum was assigned to the LM and/or potentially novel LM that has the highest matching score in addition to having the predicted UV spectral and chromatographic features (e.g., max and CRTs). This theoretical system was used successfully to identify 15-HETE in murine spleen (Fig. 3). Peak III displayed at CRT of 20.4 min on the chromatogram (at m/z 219 of MS/MS 319; left inset) and had a UV max of 235 nm (Fig. 3A, middle inset). Therefore, the search on the theoretical system was narrowed to the subdatabase with molecular ion m/z 319, UV max 235 nm, and CRT 21 min. In this case example, 15-HETE gave the highest matching score among all compounds in the subdatabase (Fig. 3B). The MS/MS peaks identified were annotated with the ion interpretation that also shows a fragmentation mechanism (Fig. 3A). Segments of 15-HETE that matched the MS/MS peaks according to the fragmentation rules are italicized (Fig. 3B). For comparison, MassFrontier was also used to identify peak III, which also identified it as 15-HETE (Fig. 3C).

Evaluation of LC-UV-MS/MS mediator-lipidomic databases and algorithms
We used LC-UV-MS/MS chromatograms and spectra acquired from murine tissue and organ extracts spiked with 34 LMs and related compounds for the evaluation. For every known compound added (i.e., spiked) in all samples, the search of its LC-UV-MS/MS data through a database gives a "hit list" showing what compound matches best. The best match has the highest matching score using the MassFrontier or theoretical database and the smallest contrast angle using COCAD. Hence, the best match is either correct or incorrect. The mean and SEM of percentage correct for best match indicated the performance of each database and algorithm (Table 1). Generally, COCAD offered the greatest percentage correct (Table 1), which was reasonable because COCAD takes into account both MS/ MS ion intensities and identities measured from authentic eicosanoid standards and unknowns. The percentage correct from the theoretical system is comparable with that from the MassFrontier system (Table 1).
Only the identified ions were used for the COCAD and theoretical databases. The test results in Table 1 demonstrate that it is not necessary to use all of the MS/MS ions to identify a LM or other compounds. Moreover, in some cases it is not obvious and cost-efficient to identify all MS/ MS ions. Rigorous identification can be done using isotope labeling and derivatization, which may not be feasible or realistic for routine analyses of the biologically relevant samples. We can identify a LM with precision using one or several chain-cut ions per functional group along with peripheral-cut ions and chain-plus-peripheral-cut ions according to the MS/MS fragmentation rules. For exam-  n ϭ 212). Therefore, the threshold score for a 95% confidence interval of matching score for correct best match is 2.02 Ϫ 1.96 ϫ 0.71 ϭ 0.62. The databases and search algorithms were developed on the basis of LC-UV-MS/MS data of LMs. The ion intensity patterns of MS/MS spectra generated from an ESI-triple quadrupole mass spectrometer are quite similar to those from an ESI-ion trap mass spectrometer, because the collision energy for both types of instruments is in the low-energy region [a few to 100 eV (laboratory kinetic energy of ions)] (7,19,20,27). Therefore, the constants and algorithms described in this report may fit the CID spectra from triple quadrupole MS/MS without much modification. For high-collision-energy (10 2 to ‫01ف‬ 3 eV) CID spectra generated via sector or spectrum time-of-flight/time-of-flight analyzer, the relative intensity patterns are quite different compared with the low-energy spectra, although many ions occurred for both energy situations (7,8,19,20,27); for example, the peripheral-cut ions are less abundant than the chain-cut ions. For ion trap and triple quadrupole MS/ MS, peripheral-cut ions are more abundant than chaincut ions. Our constants and algorithms give chain-cut ions more weight than peripheral-cut ions because chain-cut ions are more important to define LM structures. Therefore, they may still fit high-collision-energy CID spectra. Nevertheless, the constants and algorithms should be thoroughly tested and modified accordingly to fit other instruments.
The CRTs used in this set of experiments were obtained using specified chromatographic conditions (for a column of 100 mm length and some for 150 mm length; see Systems and Methodology for details) and because several fundamental issues were our initial focus. Hence, the databases and algorithms were programmed so that the new LC-UV-MS/MS data, including other chromatographic conditions, can be easily entered and used in the databases.

CONCLUSIONS
Databases and search algorithms are imperative to identify LMs in a large number of complex biological samples by LC-UV-MS/MS. Integrating MS/MS spectra with UV spectra and chromatograms for LM identification as demonstrated in the present studies significantly improves the percentage identified correctly. For the three systems of databases and algorithms we constructed, only MS/MS spectra with UV spectra matched to the unknown were used for further searching. Next, the CRTs were matched for the compounds with a good match on both UV and MS/MS spectra. MS/MS ions were identified by the empirical fragmentation rules. The ion intensities were modified according to ion identities to calculate COCAD contrast angles and the theoretical matching score for database searching. The best performance was obtained with COCAD, which counts peak intensities differently based on ion identities during direct comparisons. Also, the theoretical systems presented here can be used to identify novel structures of potentially new LMs without using standard LMs, with percentage correctly identified comparable with that of the system developed with MassFrontier when tested with LM chemical standards. Thus, the present results indicate that both the COCAD and theoretical systems can be used to provide identification and interpretations of negative-ion MS/MS fragmentation generated during profiling of LMs from complex biologic matrices.
The three mediator-lipidomic systems described in this report were constructed with results from LC-UV-MS/MS www.jlr.org Downloaded from obtained using specific LC conditions and parameters. Although these low-collision-energy CID spectra were quite similar to those acquired from triple quadruple mass spectrometers (7,19,20), the lipidomic systems need to be thoroughly tested and modified wherever necessary to be useful with results from triple quadruple mass spectrometers, not to mention those from intermediate-and highcollision-energy mass spectrometers. Theoretical databases and algorithms can be powerful when used appropriately to reduce significantly the tedious work that experienced biochemical researchers have to do when identifying the structures of novel LMs. However, they should consider other information or constraints, such as biosynthetic and metabolic pathways, stereochemistry, and bioactions, to ensure that the results are correct. Moreover, new LC results can be entered into the systems to extend the mediator-lipidomic systems for use with other LC conditions. Copies of the program and libraries can be made available to investigators on request.
Many thanks to Mary Halm Small for expert editorial advice and assistance in manuscript preparation. The authors thank Katherine Gotlinger for laboratory assistance and Dr. Nan Chang for critiques. The authors also thank Dr. Lucila Ohno-Machado of BWH Bioinformatics and Decision Systems Group for helpful discussions. This work was supported in part by