Methodological considerations in the development of HPLC-MS methods for the analysis of rodent plasma for metabonomic studies

Abstract

A study of the factors involved in obtaining valid global metabolite profiles from the HPLC-MS of rat or mouse plasma for the purposes of metabonomic analysis has been undertaken. Plasma proteins were precipitated with three volumes of either methanol or acetonitrile. Chromatographic separations were performed on a C18-bonded stationary phase using 3.5 and 5 μm particles packed into 2.1 and 4.6 mm i.d. formats, respectively, and on a C8 phase using 3.5 μm particles and a 2.1 mm i.d. column. Three reversed-phase gradient solvent systems, based on acidified water–acetonitrile, acidified water–methanol and acidified water–methanol–acetonitrile mixtures, were investigated. The column eluent was analysed with both positive and negative electrospray ionisation using a quadrupole-linear ion trap mass spectrometer. These studies revealed that while accurate classification of sample type can be made, there are a number of methodological problems associated with the analysis of plasma with respect to factors such as repeatability and column longevity. In particular, special care has to be taken to ensure that the analytical system is properly “conditioned” by the repeated injection of matrix samples. The use of biological quality control (QC) samples provided an important means of monitoring method performance. Finally, the source of the plasma (Zucker wild-type or (fa/fa) rat or mouse tumour model) also appeared to have an effect on the repeatability of the methodology.

---

Introduction

Metabolic profiling, as performed for the purposes of metabolomic¹ or metabonomic^2,3 studies, is heavily dependant upon the delivery of reliable analytical data. Indeed method “validation” is becoming an increasingly important topic for this area (inasmuch as it is possible to validate methods where many of the analytes are unknown at the start of the study and remain unidentified at the end of it). For techniques such as NMR spectroscopy, there is evidence that it is possible to perform the analysis in such a way as to be confident of acquiring reproducible data.^4–6LC-MS-based methods are, however, rather more susceptible to problems due, for example, to inter- and intra-run changes in instrument sensitivity and mass accuracy as well as chromatographic properties (retention, peak shape, selectivity etc.). We have recently described the results of studies examining analytical variation for the HPLC- and ultra performance (UP) LC-MS analysis of human and animal urine,^7–9 and proposed simple procedures for the quality control (QC) of such analyses involving the use of a standard urine (in this case made by making a pooled sample from the samples under investigation). One finding from these studies was that, in order to obtain repeatable data, a number of injections of the QCs were mandatory to condition the system in order to stabilise retention times and signal intensities.^8,9 As well as reliable methods for urine, it is clearly also necessary to have suitable profiling methods for analysing blood-derived samples such as plasma and serum. Plasma represents an important biofluid for metabonomic analysis but, unlike urine, which can be analysed with minimal sample preparation, requires sample preparation prior to injection to remove proteins, which would otherwise lead to the rapid destruction of the column. In addition, the presence of both polar analytes such as amino acids, glucoseetc., and non-polar lipids means that this biofluid represents a more formidable challenge than urine, which is composed mostly of polar and ionic molecules. A number of recent studies have described methods designed to produce global metabolite profiles from either serum or plasma, using either GC^10,11 or LC-MS (HPLC and UPLC),^12–17 or indeed GC and LC technologies in combination (e.g., ref. 16 and 18). In addition we have recently described methods for the analysis of human plasma samples using UPLC¹⁹ following either solvent precipitation or solid phase extraction. In that study, we again found that system conditioning provided an important foundation for successful method development. Similar conclusions were obtained for method development on human serum undertaken as part of the HUSERMET project.¹⁷ Here, we describe studies aimed at developing reliable and repeatable HPLC-MS-based methods for the analysis of plasma from rodents (rat and mouse) using both normal animals and disease models (dyslipidaemic and tumour-bearing).

Experimental

Solvents and reagents

All solvents used for LC-MS analysis (LC-MS Chromasolv grade) were purchased from Sigma-Aldrich (Dorset, UK). Formic acid, of analytical grade, was also purchased from Sigma-Aldrich. Water (18.2 Ω) was obtained from a Purelab Ultra system from Elga (Bucks, UK). The standards used for preparing test mixtures, including acetanilide, acetophenone, valerophenone, caffeine, glycoholic acid, pimelic acid, hydroxyhippuric acid, and adipic acid, were purchased from Sigma-Aldrich (Gillingham, Dorset, U.K). The Waters® test mixture containing theophylline, caffeine, hippuric acid and nortriptyline was also used.

Samples

All animal experiments were conducted in full accordance with the United Kingdom Home Office Animal (Scientific Procedures) Act 1986. Plasma samples, prepared using lithium heparin as the anticoagulant, were obtained from either 20 week old control male Zucker (fa/fa) obese and wild-type (−/−) rats (ten of each), or from control and human tumour xenograft-bearing Swiss nude mice (nu/nu genotype; AstraZeneca) (85 samples in total, of which 18 were control and the remainder derived from a number of different tumour cell lines ). The plasma was prepared by centrifugation at 3000 g for 5 min at 4 °C. The samples were stored for up to 2 months at −20 °C until analysed.

Sample analysis

Before analysis, the 20 rat or 85 mouse plasma samples forming the test sets were randomized. In the case of the Zucker-derived samples, for each analytical run the samples were analysed in triplicate (i.e., an aliquot of each sample was injected in three batches comprising all of the samples) in three separate batches, one batch after the other (each batch randomized differently), in order to provide an analytical run containing 60 samples (not including QCs or standards). When only the lean subset was analysed, each of the ten individual samples was analysed six times (i.e., an aliquot of each sample was injected six times), in a single analytical run, in six separate batches (each batch subject to a different randomization), thus also resulting in 60 samples in the data set (not including QCs or standards). The mouse-derived samples were analysed only once each, in a single analytical run of 85 samples (not including QCs or standards). In addition to the samples themselves, the conditioning, QC and test mixture samples were included for each set as described below.

Routine sample preparation. Aliquots (50 μl) of the rodent plasma samples were prepared by protein precipitation with 150 μl of cold methanol (both acetonitrile and methanol were evaluated but we, like others^10,11 found that the latter gave superior results). Precipitated proteins were removed by centrifugation (Centrifuge 5417, Eppendorf) at 17 900 g for 10 min. Subsequently, 120 μl of the supernatant was diluted with 60 μl of water and vortexed.

A biological quality control (QC) sample^7–9 was prepared at the same time by mixing equal volumes from each of the test samples (20 μl) and treating this pooled sample as described above. For analysis, 10–15 of these biological QC samples were run prior to commencing the main analytical run, as described below, to ensure that the system had stabilized, and one QC sample was run every ten samples thereafter.

Two different standard mixture solutions were used to monitor the quality and stability of the separation. The mixture used for positive electrospray ionisation (ESI) consisted of acetanilide, acetophenone, valerophenone and caffeine, all at concentrations of 5 μg ml⁻¹, or the Waters test mix, composed of theophilline, caffeine, hippuric acid and nortriptyline, at concentrations of 60, 60, 60 and 22.5 μg ml⁻¹, respectively. A mixture of glycoholic acid, pimelic acid, hydroxyhippuric acid and adipic acid, at concentrations of 20.48, 24.44, 21.74, 20.30 and 20.60 μg ml⁻¹, respectively, was used in negative ESI. These test mixtures were run at the beginning (i.e. after the “conditioning injections), middle and end of the analytical run.

HPLC-MS conditions

For HPLC-MS, 10 μl of each sample was analysed by HPLC using a Perkin-Elmer series 200 high pressure LC micro solvent delivery system (Perkin-Elmer Life Sciences, Cambridge, UK) and a CTC PAL autosampler (CTC Analytical, Switzerland). The autosampler was maintained at 4 °C. Separations were performed on either a Symmetry C18 3.5 μm (2.1 × 150 mm) column (Waters Ltd, Elstree, UK), a Symmetry C18 5 μm (4.6 × 50 mm) column (Waters) or a C8 3 μm (2 × 50 mm) column (Phenomenex, Macclesfield, U.K.). A fresh Waters Sentry™ 2.1 × 10 mm guard column was used for each analytical run (Waters Ltd, Elstree, UK). All of the columns used in this study were maintained at 40 ± 0.2 °C during the analysis by using a heater controller to ensure temperature stability (Jones Chromatography, Hengoed, U.K.).

For method development purposes, two reversed-phase gradient systems, based on binary solvent systems composed of either acetonitrile or methanol (solvent B) versuswater (solvent A), were evaluated. All eluents were acidified with 0.1% formic acid v/v. For these solvent systems, separations were performed using gradient elution with the initial conditions set at 90% A : 10% B for the period 0–0.5 min post injection, followed by a series of linear gradients to 50% B at 3 min, 80% B at 11 min and finally 100% B at 12 min. The solvent composition was then held at 100% B for 2 min, after which the column was returned to the starting conditions and held for 2 min prior to the next injection, making a total cycle time of 16 min/sample. A flow rate of 400 μl min⁻¹ was used for the 2.1 mm i.d. columns and 1000 μl min⁻¹ for the 4.6 mm i.d. column

Prior to analysis of the samples, it has been our standard practice to run a number of the QC samples for system “conditioning”. As the study progressed it became clear that much more system conditioning was needed for plasma compared to urine,^8,9 and so a shorter gradient program was used to achieve this without wasting too much analysis time. The final methodology adopted here for the conditioning phase of the analysis was 15 injections of 20 μl of the QC sample, with chromatography as follows. The starting conditions were 90% A and 10% B (methanol) changing in a linear gradient to 100% B over 3 min, which was held at this composition for 2.5 min before returning to the starting condition, which was also held for 2.5 min before the next conditioning QC injection. This gave an overall cycle time of 8 min per conditioning sample. This was then followed by sample analysis using the conventional gradient system described above using methanol as the organic modifier.

Mass spectrometry

All of the MS data were acquired on a 4000 QTRAP hybrid triple quadrupole linear ion trap mass spectrometer, (Applied Biosystems/MDS Analytical Technologies, Warrington, U.K.). The TurboIonSpray inlet was operated at 350 °C in both positive and negative ESI modes in separate experiments. Enhanced MS scan data were acquired using Q3 as a linear ion trap operating in the dynamic fill time mode over a range of 100–900 m/z, with a scan rate of 1000 amu s⁻¹ and a step size of 0.08 amu. The TurboIonSpray voltage was set at ±4500 V, curtain gas at 20 psi, auxiliary gases at 40 psi and declustering potential at 40 V.

Data analysis

MarkerView software version 1.2.0.1 was used to process the raw spectrometric data acquired by Analyst 1.4.2 software (Applied Biosystems/MDS Analytical Technologies). The parameters for peak finding, alignment and filtering were as follows: noise threshold set at 1e5, minimum spectral peak width at 0.25 amu, minimum RT peak width at three scans, maximum RT peak width at 100 scans, retention time tolerance of 0.3–0.8 min depending on the run, mass tolerance at 0.25 amu and maximum number of peaks at 3500. For pattern recognition, Principal Component Analysis and a range of statistical tools were used including Principal Component Variable Grouping (PCVG)²⁰ in order to facilitate the discovery of relationships between peaks. Using PCVG peaks that are correlated with one another can be manually or automatically assigned to groups and the resulting groups can be manually inspected from within MarkerView.

Results

Initial method development

Column conditioning and washing. As indicated in the introduction, our previous experience with the analysis of urine samples, by either HPLC or UPLC, showed that conditioning of the LC-MS system with injections of the biofluid under study was necessary to establish stable analytical conditions.^7–9 Preliminary investigations were undertaken to determine whether this was also a requirement for system stability in the case of plasma. Initially this was investigated on Zucker rat-derived samples (normal (−/−) and (fa/fa) obese) analysed on a 3.5 μm C18 phase using a 2.1 mm i.d. × 150 mm long column. The Zucker sample set was chosen for these preliminary studies for the pragmatic reason that the sample volumes that can be obtained from rats are greater than those available from mice, and therefore allow multiple analysis of the sort required for method development. The development of the analytical LC-MS conditions was based on our previous findings on the HPLC-MS of human urine⁸ and used an acidified water–acetonitrile gradient for chromatography. As a result of these experiments it became clear that a minimum of at least eight injections were required before the system achieved anything like adequate system conditioning. This was in some contrast to our findings for urine where stability was usually achieved within five injections.^7–9 We therefore proceeded to further evaluate the methodology using a minimum number of ten injections to obtain some measure of system stability. A similar requirement, for a minimum of ten serum sample-extracts, has also recently been shown by the HUSERMET consortium for the analysis of human-derived serum samples.¹⁷ During the studies described here, we sometimes observed that the chromatography deteriorated during the run, with complete column failure after a few hundred injections (as evidenced by high column back pressures and poor peak shapes that could not be restored even with extensive column washing). This was again in contrast to our previous experiences with urine^8,9 where many more of injections were possible with no reduction in column performance. The deterioration in the column with injections of plasma extracts was presumed to be due to contamination with strongly retained, non-polar, sample constituents. In an attempt to counter this problem, more extensive column washing was performed at the end of the gradient, with the time taken for elution with 100% acetonitrile increased from 2 to 5 min and the precolumn changed for each run. This increase in the duration of the washing cycle at the end of each analytical run did result in a qualitative improvement, to the extent that this methodology gave a more reliable separation of the strains in PCA (PC1 vs. PC2, data not shown), and at the same time also in longer column lifetime (in excess of 1000 injections). Whilst this resulted in a separation of the samples derived from (fa/fa) and normal (−/−) animals using PCA, there was, however, generally poor grouping of the QC samples (which showed clear evidence of run order effects). In addition, the technical replicates, when all 60 samples in the three batches were considered as a single block, showed poor repeatability of the method compared to our previous experiences with urine samples. The use of either positive or negative ESI for detection gave similar results (here and in subsequent method development), and for the sake of brevity, only positive ESI will be considered further.

Evaluation of different analytical columns and solvent systems

A conclusion from these initial evaluations was that, although a degree of system conditioning had taken place as a result of the injection of ten QC samples prior to the commencement of the analysis, the system was still not completely equilibrated. Indeed, it seems likely that the system was still undergoing modification during the analytical runs (based on a general increase in the retention times of a number of peaks throughout the run). On the assumption that excessive retention of lipids was responsible for this poor repeatability, we briefly examined the use of an alternative, less retentive, C8-bonded packing material. However, the C8 phase coupled with an acetonitrile-based solvent gradient did not provide any obvious improvement and was not pursued further, which contrasts with the work of Bruce et al.¹⁴ who concluded that for UPLC-MS of human plasma, the C8 phase was to be preferred. We also briefly investigated the use of an alternative column geometry in the form of a 4.6 mm i.d. column packed with the same 3.5 μm C18-bonded stationary phase, reasoning that the significantly higher ratio of stationary phase to sample might modify the effects of column contamination to the benefit of the analysis. However, the results obtained still showed major time-dependant changes in system properties during the run and certainly did not offer any improvement over the 2.1 mm i.d. column format with respect to repeatability (data not shown). Overall, the conclusion from this initial evaluation was that the acetonitrile–water gradient on the C18 column, whilst perhaps suitable for short analytical runs of a few tens of samples, was not sufficiently robust for longer analyses.

Methanol–water gradient elution

Examination of alternative solvent gradients, based on either water, methanol or water–methanol : acetonitrile (2 : 1 v/v) using the same experimental conditions led us to the general conclusion that water–methanol systems appeared to give the best results and further studies to optimise the system were therefore undertaken using this type of gradient. Again, the initial evaluation of the methanol–water system employed ten conditioning injections of the QC sample for system equilibration. Although the methanol–water gradient system did appear qualitatively better than the acetonitrile-based systems, providing a good separation in the case of (−/−) and (fa/fa) rats in the individual batches, examination of all the data as a single block by PCA still showed a clear dependence of the results on the run order. The poor repeatability of the technical replicates was again consistent with insufficient column conditioning.

Rapid gradients and large volume injections for system conditioning

Analysis of the results emerging from this series of experiments on Zucker rat plasma, and parallel studies using UPLC,¹⁹ led to the general conclusion above that the initial ten QC injections provided only a limited measure of system conditioning and probably needed to be increased in number. However, these conditioning injections have the disadvantage that they contribute nothing to the subsequent data analysis, whilst consuming a significant amount of instrument time. To make the conditioning step more efficient and reduce the time taken, the effect of injecting either 10 × 20 μl or 15 × 20 μl of the plasma extract, rather than 10 × 10 μl, was investigated. Also, to improve throughput, the time taken to run the conditioning samples was reduced by shortening the gradient to 3 min from 11 min (see experimental section and also ref. 19). Examination of the resulting data from the Test Mixture injections showed that both retention times and responses were generally stable, and further examination of the QC data also showed that greater retention time stability was achieved using this approach than that obtained previously. The best results were obtained when 15 × 20 μl injections of the conditioning QC samples were employed, with both retention time stability and repeatability of the peak area response improved by this measure. (In the context of this work, the term repeatability is used as defined in ref. 21, as follows: “repeatability expresses the precision under the same operating conditions over a short time interval. Repeatability is also termed intra-assay precision. Within day or within-run precision are also often used to describe repeatability.”) However, despite this exhaustive system conditioning, some dependence on run order was still observed for the first few samples (data not shown).

Lean Zucker rat plasma analysis

The results described above, where repeatability for the plasma samples proved difficult to obtain, were somewhat different from our experiences with the equivalent UPLC method for human plasma¹⁹ (and that of others for human serum¹⁷). However, one difference between these rat-derived samples and the human plasma analysed was the very high lipid content of the plasma samples obtained from the (fa/fa) strain. Given the obvious potential for lipid contamination of the stationary phase to adversely affect the results, we therefore investigated the possibility that the sample type had an effect on the repeatability of the system. Thus, the analysis was rerun, but this time using only samples from the phenotypically normal Zucker (−/−) wild-type rats, with QC samples also made up only from samples obtained from these animals. For this experiment, 15 conditioning injections of 20 μl were performed, followed by the Test Mixture, and then three 10 μl injections of the QCs (only the last of these, “QC18”, was used in the subsequent data analysis). Thereafter one QC injection was used per ten sample injections, with the Test Mixture run in the middle and at the end of the analysis. The analysis of the data from the Test Mixtures provided highly repeatable data with coefficients of variation (CVs) for retention times less than 0.4% and peak area variability ranging from 2.7–7.0%. Similarly, following the conditioning step, excellent retention time stability was also seen for the ions detected in the QC samples over the whole run.

The results obtained on PCA for this analysis of samples from lean animals only were much more satisfactory than those obtained when samples from (fa/fa) animals were also present, with reasonable clustering of technical replicates (though some dependence on run order was still detectable), as shown in Fig. 1A. Run order effects are seen in PC2, but when higher PCs are examined (e.g. PC3 vs. PC4, Fig. 1B), then grouping by animal was obtained, clearly illustrating the potential for the metabolic phenotyping of even very closely related and inbred animals.

Fig. 1 (A) PCA (PC1 vs. PC2) on the data acquired when only the normal Zucker (−/−) wild-type rat samples were analyzed in a single run with six replicates per sample (as six blocks of ten samples/block). The dependence on run order is shown in PC2. Each animal has been allocated a different colour to show the variability in the samples as well as the QCs. (B) PCA (PC3 vs. PC4) on the data acquired for the normal Zucker (−/−) wild-type rat samples were analyzed in six replicates. Clear grouping of the technical replicates is observed. As in part A, each animal is represented by a different colour to show the excellent clustering when the run order effects dominating PC1 and 2 are removed.

As we have discussed elsewhere,^8,9 acceptable criteria for repeatability in metabolic profiling experiments are still being defined; however, for bioanalytical methods used to monitor drugs, the FDA recommends that a CV of 15% of the nominal value is considered acceptable (although at concentrations near to the method’s limit of quantification (LOQ), CVs of 20% can be used).^22,23 For biomarkers , which probably better reflect the type of analysis contemplated in these untargeted metabolic profiling studies, the current FDA guidance allows up to 30% of total error for targeted LC-MS analysis.²³ Examination of the peak area data obtained here for the ions present in the QC samples showed good repeatability by these criteria. Thus, when all ions detected by the processing software for the QCs (n = 1421) were examined, it was found that ca. 80% of the ions gave CV values of less than 15%, 88% of the ions less than 20% and 97% of the ions less than 30%, as illustrated in Table 1 (the full set of QC data is provided for all 1421 ions in Table S1 of the ESI† ). For the QCs to be accepted as a reliable means for assessing repeatability, it is clearly critical that they must accurately reflect the behaviour of the samples themselves. When the repeatability of the six technical replicates of the plasma samples obtained for each of the individual rats were examined using the same process, an average of ca. 80% of the ions gave CV values less than 15%, 89% less than 20%, and 96% less than 30% (Table 1, samples L1–L10). This behaviour is essentially the same as that of the QCs, validating the use of the latter for performance monitoring.

Table 1 Data from the technical replicates from the lean (−/−) Zucker rat (animals L1 to L10) dataset and the corresponding QC data showing the number of ions that achieved 15%, 20% and 30% CV

Sample	L1	L2	L3	L4	L5	L6	L7	L8	L9	L10	Average of test samples (L1–L10)	QC
Number of ions	1083	1172	1175	1146	1213	701	1205	1197	1195	1202		1112
% of total ions with CV <15%	76.2	82.5	82.7	80.6	85.4	49.3	84.8	84.2	84.1	85.9	79.6	78.25
Number of ions	1224	1307	1290	1285	1315	895	1316	1328	1308	1314		1250
% of total ions with CV <20%	86.1	92.0	90.8	90.4	92.5	63.0	92.6	93.5	92.0	92.5	88.5	87.97
Number of ions	1350	1400	1381	1384	1395	1171	1386	1394	1398	1393		1360
% of total ions with CV <30%	95.0	98.5	97.2	97.4	98.2	82.4	97.5	98.1	98.4	98.0	96.1	95.71

---

As we have noted previously for urine samples, there was a clear relationship between signal intensity and repeatability. Thus, the most intense peaks, whilst in the minority, were also the most repeatable (see Table 2 for a summary of the repeatability of ion signal intensity). In addition to this, the repeatability was also examined with regards to the mass of the detected ions. This is shown in Table 3, where it can be seen that ions of higher mass (700–900 amu) show a significant reduction in repeatability.

Table 2 Data for ions from the three sample sets examined here showing the effects of ion intensity on repeatability (<15, <20 and <30% CV), together with the numbers of ions at each intensity range

Ion intensity range	Lean Zucker rat data set				Lean and (fa/fa) Zucker rat data set				Tumour-bearing mice data set
	% of total ions	CV <15%	CV <20%	CV <30%	% of total ions	CV <15%	CV <20%	CV <30%	% of total ions	CV <15%	CV <20%	CV <30%
e5	0.00				0.22	0.00	0.00	0.00	0.00
e6	5.07	29.17	51.39	86.11	3.18	6.98	13.95	39.53	15.07	18.92	54.05	79.73
e7	41.83	72.90	87.88	96.80	33.60	55.95	67.18	79.52	69.86	53.06	70.26	82.22
e8	50.99	86.74	91.30	95.72	58.92	71.73	78.39	87.06	15.07	67.57	74.32	85.14
e9	2.11	100.00	100.00	100.00	4.07	90.91	90.91	96.36	0.00

---

Table 3 Data for ions from the three sample sets examined, showing the relation of the mass of the ion detected with the repeatability of its signal (<15, <20 and <30% CV), in the mass spectrometer, together with the numbers of ions (as % total) at each mass range

Mass range (amu)	Lean Zucker rat data set				Lean and (fa/fa) Zucker rat data set				Tumour-bearing mice data set
	% of total ions	CV <15%	CV <20%	CV <30%	% of total ions	CV <15%	CV <20%	CV <30%	% of total ions	CV <15%	CV <20%	CV <30%
100–300	22.40	80.19	90.88	98.74	20.50	72.20	83.39	91.34	39.30	65.60	84.38	94.27
300–500	35.70	80.08	88.17	95.07	33.10	66.07	73.44	85.49	41.50	47.29	63.05	78.33
500–700	24.60	85.96	93.70	99.14	26.70	77.07	81.49	91.71	14.30	24.29	45.71	67.14
700–900	17.10	62.40	77.27	90.08	19.30	39.46	49.81	59.77	4.70	30.43	60.87	73.91

---

Signal repeatability was also assessed using the visualization tools of the MarkerView software to study spectrum plots and extracted ion chromatograms (XICs). In Fig. 2A a spectrum plot of masses found in the QC samples over the range 802–840 amu for the time period 9.2–11.0 min is shown. Poor stability was observed for several of these ions, in agreement with the findings described above (Table 3). In contrast, excellent stability was seen for ions eluting in the same time window but in the mass range 508–550 amu (Fig. 2B) and similar behaviour was found for the majority of the detected ions. Another perspective is provided in Fig. 2C, which illustrates the XIC of an ion of m/z 258.32 eluting at 7.5 min, which showed low variation in signal intensity, shown by its CV of 4.1% (see ESI, Table S1† ). This plot shows good signal repeatability in addition to satisfactory peak alignment, as would be expected from the low CV achieved by this ion.

Fig. 2 Characteristic XIC and spectrum plots used for the examination of ion stability. In dark blue is the first injection (QC18) and in cyan is the last injection (QC24). (A) Spectrum plot for the analysis of the QCs in the mass range 802–840 amu and the time period 9.2–11 min. Poor stability was observed for most of the ions shown, with a significant loss of signal observed along the run. (B) Spectrum plot for the analysis of the QCs in the mass range 508–550 amu and the time period 9.2–11 min. Excellent stability was generally observed for these ions with the exception of the ions at m/z 518.4 and 542.4, which showed some variability. (C) A plot of XIC of the ions in the range 257.8–259.0 amu for the time period 6.5–8.5 min showing stable intensity and good peak alignment along the run.

All of the ions that fell outside the repeatability criterion of CV >30% (ca. 4% of the total) in the QCs were then manually examined to understand the reasons for their failure to achieve the required level of repeatability. For most of these ions, the reason was that they showed a much higher intensity in QC18 (the first analytical QC injection of the data set following conditioning), but thereafter, were generally stable in subsequent QCs. Another type of signal variability that was observed was related to the run-order of the samples with some ions showing a trend of decreasing signal intensity, whilst others exhibited an increase. There were also a small number of ions that showed an irregular pattern of fluctuations in intensity along the run. Irrespective of the reasons for these various types of behaviour, it is clear that this subset of poorly repeatable ions would be poor candidates for consideration as potential biomarkers (examples of these ions are given together with the loadings plot in the ESI, Fig. S1† ).

Lean and (fa/fa) Zucker rat plasma analysis

When the same analysis was performed on a mixed set of lean and (fa/fa) Zucker rat-derived plasma samples (ten of each, in three batches), the equivalent data showed some differences to that obtained when the normal (−/−) Zucker animals were analysed on their own. Thus, the run order effect seen for the QCs and technical replicates was more pronounced, and even after ten injections of 20 μl of the QC sample, the system did not seem to have attained full equilibration (in a subsequent analysis, with the same samples, 15 conditioning injections of 20 μl were performed with similar results). However, class separation was seen using PCA when the individual batches of samples were analysed by PCA and, when all of the data was analysed as a single block, separation was also seen between the two classes in PC1 vs. PC2, as shown in Fig. 3A.

Fig. 3 (A) PCA (PC1 vs. PC2) on the data acquired when the normal and the (fa/fa) Zucker rat samples were analyzed in triplicate. A run order effect is evident (see the QC sample trajectory and also the numbering of some of the samples themselves representing the order of injection). Nevertheless, a class separation of the normal Zucker (−/−) wild-type rats (blue dots) from the (fa/fa) strain (green dots) can be seen. (B) PCA (PC1 vs. PC2) after removal of variables contributing to time trend (variables were found by applying PCVG). Better QC clustering and separation of the strains is observed and the run order effect is not as evident.

However, the repeatability of the method when used on these “mixed” samples was slightly worse than when the samples from the normal (−/−) strain were analysed alone, as described above. Thus, of the 1351 ions detected in the QC samples, 878 (65.0%), 985 (72.9%) and 1124 (83.2%) gave CVs within 15, 20 and 30%, respectively, as shown in Table 4 (the full QC data set is given in the ESI, Table S2† ). Similar results were obtained when the technical replicates (n = 3) of each sample were examined according to these criteria (Table 4). These values are somewhat lower than those for the lean animals on their own (e.g., at the 30% CV level ca. 96% of the ions for the normal animals passed, compared to only 83% for the mixed normal and (fa/fa) run). As is clear from Table 4, the level of repeatability in the mixed run never attained that of the phenotypically normal animals when analysed on their own. As we have noted for previous analyses of biological fluids,^8,9,19 the most intense ions generally exhibited the best repeatability. Thus, the small group of intense ions (>e9) show CVs of less than 15% (see Table 2) for both lean only and mixed lean, and (fa/fa) rat sample sets. We also examined ion intensity repeatability in connection with ion mass (Table 3); again, a noteworthy reduction in repeatability is seen for higher masses (700–900 amu). Interestingly these irreproducible ions of high mass (700–900 amu) show high variation in intensity with a significant portion of them (close to 40%) not fulfilling the 30% criterion. Once again the behaviour of the technical replicates and the QCs was similar, providing confidence that the QCs did indeed adequately reflect the analytical repeatability of the method. Overall, these results support the notion that the problem of repeatability was, at least in part, a result of the high concentrations of lipids present in the Zucker (fa/fa) obese rat samples.

Table 4 Data for ions from the three technical replicates of each Zucker rat plasma sample showing the repeatability at the CV level <5%, together with the number of ions detected. L = Zucker lean and F = Zucker (fa/fa) obese animals, together with the number and percentage of ions in the QC samples showing repeatability at the 15%, 20% and 30% CV levels

Sample	F1	F10	F2	F3	F4	F5	F6	F7	F8	F9	Average F
Number of ions	627	1003	798	901	1046	967	703	971	1170	945
% of total ions with CV <15%	46.4	74.2	59.1	66.7	77.4	71.6	52.1	71.9	86.6	69.9	67.6

---

Sample	L1	L10	L2	L3	L4	L5	L6	L7	L8	L9	Average L
Number of ions	891	1016	754	779	1105	946	606	1125	723	1002
% of total ions with CV <15%	65.9	75.2	55.8	57.7	81.8	70.0	44.9	83.3	53.5	74.2	66.2

---

QC
Number of ions	878		Number of ions	985		Number of ions	1124
% of ions with CV <15%	65.0		% of ions with CV <20%	72.9		% of ions with CV <30%	83.2

---

To examine in more detail the source of the run order effects seen in these data, the dataset was examined using a t-test between the first ten injections and the last ten injections of the run. In this way, ions that differed to a great extent (in their intensity) between the first and the last ten injections were identified, and their contribution and trend plots along the run were examined. In addition, manual inspection of the ions exhibiting high variation (CV >30%) in ion intensity was conducted. These studies showed that the majority of ions showing time-related trends exhibited a loss of signal intensity. For example in Fig. 4, the ion at m/z 433.1 eluting at 0.7 min shows a continuous decline in its signal (see specifically the red dots representing the QC injections). Such a trend might perhaps be attributed to ion source contamination and loss of ionisation efficiency along the run. However, a second ion at m/z 883.8 and an 11.9 min retention time (also shown in Fig. 4) showed a large variation between the QC injections in a much less predictable fashion. A small number of ions exhibited this trend. Finally, a number of ions exhibited an increase in intensity along the run. Fig. 5 shows the behaviour of the ion at m/z 505.3, eluting at 8.6 min, plotted by run order within the individual groups (normal strain (−/−), (fa/fa) obese and QCs) of samples (X axis). An increase in the intensity during the run is evident in all three sample groups.

Fig. 4 Profile plots of characteristic ions showing a time of analysis trend. Ions (mass/rt pair) 433.1/0.7 (blue line) and 883.88/11.9 (pink line) are plotted, sorted in the X axis by the run order. The lean group (−/−) is indicated by L (blue dot) and the (fa/fa)-derived rat samples are denoted by F (green dot). The red dots represent the QC samples.

Fig. 5 Profile plots of a characteristic ion showing a time of analysis trend. Ion 505.3/8.6 plotted in X axis by run order within the sample group. As for Fig. 4, the lean (−/−) animals are indicated by L (blue dot) and the (fa/fa) animals by F (green dot) with red dots representing the QC samples.

The PCVG utility was also applied in order to reveal such underlying trends.²⁴ This utility groups variables together according to their trends.²⁰ Three groups of variables that showed a significant time related trend were identified and the contribution to the profile plot for each variable was manually checked. Variables showing a significant time-related trend (either an increase or decrease in the signal) were excluded from the data set and PCA repeated. This resulted in tighter clustering of the QCs (see Fig. 3B) and an enhanced separation of the two strains. However, such “thinning” of the data needs to be undertaken with some care to ensure that important biological information is not lost, and clearly it is important to indicate where and when such procedures have been performed when reporting the data.

Mouse plasma analysis

As well as rat plasma, we also examined the use of the developed LC-MS method for samples obtained from both control and tumour-bearing mice of different tumour types, including colorectal, skin and non-small cell lung cancer-derived tumours (to be reported elsewhere). For these samples, the injection of 10 × 20 μl of the conditioning samples (with chromatography using the fast gradient conditions) resulted in very stable responses and retention times, together with acceptable peak shapes, for the compounds in the test mixture. The CVs for retention time variation were less than 0.3% for all test components and CVs for peak areas ranged from 2.5–8.8%. When the QC injections were examined for repeatability, it was found that ca. 50% of the total ions gave CVs of less than 15%, 68% less than 20%, and 82% lower than 30% (the data for the measured ions in the QC samples are given in the ESI, Table S3† ).

PCA of these mouse plasma samples showed a clear clustering of the bulk of the QCs, but with QCs 18, 19 and 20 lying together as a distinct “satellite” group a little out of the main group (encompassing injections 95 (QC18), 107 (QC19) to 115 (QC20), respectively (see Fig. 6A). Re-examination of the QC data concentrating on the differences between QCs 11–17 and 18–20 showed that, while retention times were not different, some (but not all) of the ions, particularly those at higher mass (see Table 3, and for more detail Table S3 of the ESI† ) showed a marked change in response, generally a decrease in intensity, suggesting that perhaps source contamination in the MS was the reason for these data being less reliable towards the end of the run. Of course alternative explanations for these changes in signal intensity in addition to instrument contamination are also possible, including analyte instability. Irrespective of the reason for these changes, it was found that they contributed to the run-order effect evident in PC2 and the effect was suppressed when the scores plot was projected as PC1 vs. PC3 (see Fig. 6B). The loadings plots and the raw LC-MS data were examined in detail in MarkerView; the PCVG utility was again applied to assist in revealing variables (ions) that exhibited a clear run-order trend. Representative ions are shown in Fig. S2 in the ESI.† Based on the findings of the variable grouping utility, a group of 75 variables showing such behaviour was selected and excluded from subsequent PCA. The resulting PCA scores plot shows a much tighter cluster of QC injections, with PC1 and PC2 explaining more than 60% of the variation (see Fig. 7).

Fig. 6 PCA scores plot of the mouse plasma analysis. (A) PC1 vs. PC2 scores plot showing a run effect in PC2, and (B) PC1 vs. PC3. Different colour codings represent different types of xenograft.

Fig. 7 PCA (PC1 vs. PC3) of the mouse plasma sample analysis after the removal of variables that contributed to the run-order effect. Different colour codings represent different types of cancer.

Discussion

The development of methods for use in the analysis of rodent plasma in metabonomic/metabolomic studies by HPLC-MS has proved to be more complex than for biofluids such as urine. It seems quite clear that, as found for urine, the build up of compounds on the column requires the use of a number injections of matrix in order to condition the system and stabilise retention times.^7–9 However, unlike urine, where relatively few injections of matrix were required to achieve this system stability, in the case of plasma, much more “conditioning” was required to equilibrate the system for rodent plasma. We have noted similar requirements for the UPLC-MS analysis of human plasma,¹⁹ and this also seems to be true of the UPLC-MS of human serum.¹⁷ The poor repeatability seen with the acetonitrile–water-based gradient seemed, at least in part, to be associated with changes in the chromatographic properties of the column, which was consistent with a build-up of contaminants that were not being removed even with extended column washing. This is perhaps not unexpected if the retention mechanism for these contaminants was based on hydrogen bonding, as acetonitrile is not particularly effective at disrupting this type of interaction. Methanol, which is better at breaking hydrogen bonding interactions than acetonitrile, was therefore used as the organic modifier. Suitable conditioning of the system was finally obtained here by an increased number of larger injections of the plasma extract. The use of larger injections of matrix, combined with rapid gradients is useful for increasing the efficiency of this analytical “deadtime”.

As well as the need for adequate pre-equilibration of the system, disease status may also have an effect on performance, as shown by the case of rat-derived samples where a method suitable for normal animals gave poorer results when used to analyse plasma from Zucker (fa/fa) animals containing much higher concentrations of lipids. There also seems, with the current methodology, to be a practical limit on the number of plasma samples that can be included in a single analytical run before performance deteriorates, as assessed by the use of the biological QC samples (see also ref. 17). This seems to be the result of changes in signal intensity rather than changes in retention time and, as shown by the mouse tumour plasma samples, analytical runs of less than 100 samples would seem to be indicated. The use here of biological QC samples provided a simple means of monitoring method performance, and appears to provide a more sensitive indicator of performance than simple test mixtures in revealing changes in response/retention time.

The results presented here describe only a portion of the effort directed towards obtaining a robust and repeatable HPLC method for plasma and it seems clear from this work that run order effects will be a continuing issue for those working with this matrix. If it is accepted that metabolomic/metabonomic analysis of plasma/serum with the current technology will encounter run-order effects, then the problem is probably best addressed with medium length runs, close monitoring of the analytical system, and thorough maintenance of the columns, LC systems and mass spectrometers, together with tight monitoring of the quality of the data (see also the conclusions of the HUSERMET study for serum¹⁷). PCVG has proven to be a useful tool for detecting these effects and the variables involved.

Conclusions

Plasma represents a greater analytical challenge for global metabolite profiling than samples such as urine, even following the removal of the bulk of the proteins using an efficient solventprecipitation procedure. However, with extensive system equilibration using “conditioning” injections of the matrix, a degree of repeatability can be obtained. However, such methods do not yet appear to be robust for use over long periods, or for samples with e.g., high lipid content such as those obtained from Zucker (fa/fa) rats. This implies that methods should not be developed on normal “control” samples but on those that mimic the actual samples to be analysed. Although statistical tools are now available that can identify trends in large data sets, and may assist in removing some time-related variables, the high quality of the raw data is of the utmost importance. Hence monitoring of the analytical system and ensuring that it operates as close as possible to the ideal conditions remains very important for the generation of useful and trustworthy global metaboliteMS profiles. In this respect, the use of a biological QC, the analytical behaviour of which closely models that of the sample set under investigation, appears to provide a pragmatic means of assessing repeatability.

References

1. J. K. Nicholson and J. C. Lindon, Nature, 2008, 455, 1054–1056.
2. J. K. Nicholson, J. C. Lindon and E. Holmes, Xenobiotica, 1999, 29, 1181–1189.
3. J. K. Nicholson, J. Connelly, J. C. Lindon and E. Holmes, Nat. Rev. Drug Discovery, 2002, 1, 153–161.
4. E. J. Saude and B. D. Sykes, Metabolomics, 2007, 3, 19–26.
5. A. D. Maher, S. F. M. Zirah, E. Holmes and J. K. Nicholson, Anal. Chem., 2007, 79, 5204–5211.
6. M. Lauridsen, S. H. Hansen, J. W. Jaroszewski and C. Cornett, Anal. Chem., 2007, 79, 1181–1186.
7. T. Sangster, H. Major, R. Plumb, A. J. Wilson and I. D. Wilson, Analyst, 2006, 131, 1075–1078.
8. H. G. Gika, G. A. Theodoridis, J. E. Wingate and I. D. Wilson, J. Proteome Res., 2007, 6, 3291–3303.
9. H. G. Gika, E. Macpherson, G. Theodoridis and I. D. Wilson, J. Chromatogr., B: Anal. Technol. Biomed. Life Sci., 2008, 871, 299–305.
10. J. A, J. Trygg, J. Gullberg, A. I. Johansson, P. Jonsson, H. Antti, S. L. Marklund and T. Moritz, Anal. Chem., 2005, 77, 8086–8094.
11. P. Begley, S. Francis-McIntyre, W. B. Dunn, D. I. Broadhurst, A. Halsall, A. Tseng, J. Knowles, R. Goodacre and D. B. Kell, Anal. Chem., 2009, 81, 7038–7046.
12. M. S. Sabatine, E. Liu, D. A. Morrow, E. Heller, R. McCarroll, Wiegand, G. F. Berriz, F. P. Roth and R. E. Gerszten, Circulation, 2005, 112, 3868–3875.
13. E. J. Want, G. O’Maille, C. A. Smith, T. R. Brabdon, W. Uritboonthai, C. Qui, S. A. Trauger and G. Siuzdak, Anal. Chem., 2006, 78, 743–752.
14. S. J. Bruce, P. Jonsson, C. Clorarec, J. Trygg, S. L. Marklund and T. Moritz, Anal. Biochem., 2008, 372, 237–249.
15. W. B. Dunn, D. Broadhurst, M. Brown, P. N. Baker, C. W. G. Redman, L. C. Kenny and D. B. Kell, J. Chromatogr., B: Anal. Technol. Biomed. Life Sci., 2008, 871, 288–298.
16. R. E. Williams, E. M. Lenz, A. J. Wilson, J. Granger, I. D. Wilson, H. Major, C. Stumpf and R. Plumb, Mol. BioSyst., 2006, 2, 174–183.
17. E. Zelena, W. Dunn, D. Broadhurst, S. Francis-McIntyre, K. Carroll, P. Begley, S. O’Hagan, J. D. Knowles, A. Halsall, HUSERMET Consortium, I. D. Wilson and D. Kell, Anal. Chem., 2009, 81, 1357–1364.
18. P. Gao, C. Lu, F. Zhang, P. Sang, D. Yang, X. Li, H. Kong, P. Yin, T. Tian, A. Lu and G. Xu, Analyst, 2008, 133, 1214–1220.
19. F. Michopoulos, L. Lai, H. Gika, G. Theodoridis and I. Wilson, J. Proteome Res., 2009, 8, 2114–2121.
20. G. Ivosev, L. Burton and R. Bonner, Anal. Chem., 2008, 80, 4933–4944.
21. Applications of LC-MS in Toxicology, ed. Aldo Polettini, Pharmaceutical Press, London, UK, 2006, p. 80.
22. FDA Guidance for Industry, Bioanalytical method Validation, Food and Drug Administration, Centre for Drug Evaluation and Research (CDER), May 2001. A guidance.
23. C. T. Viswanathan, S. Bansal, B. Booth, A. J. DeStefano, M. J. Rose J. Sailstad, V. P. Shah, J. P. Skelly, P. G. Swann and R. Weiner, Am. Assoc. Pharm. Sci., 2007, 9, E30–E42.
24. L. Burton, G. Ivosev, S. Tate, G. Impey, J. Wingate and R. Bonner, J. Chromatogr., B: Anal. Technol. Biomed. Life Sci., 2008, 871, 227–235.

---

Footnote

† Electronic supplementary information (ESI) available: Additional figures and tables. See DOI: 10.1039/b910482h

---