Systematic Internal Standard Selection for Capillary Liquid Chromatography–Mass Spectrometry Time Normalization to Facilitate Serum Proteomics

Abstract

Because blood interacts with almost all tissues of the body, it is likely that changes in the overall health of an organism will be reflected in the quantities of specific serum peptides and proteins, making them biomarkers. Due to the complexity of serum, pre-analytical sample simplification and separation are needed prior to mass spectrometric analysis. Use of a reverse-phase capillary column coupled to a mass spectrometer allows for separation and analysis of serum as part of efforts to discover biomarkers. Even after sample simplification by organic solvent precipitation, data files for a single sample typically exceed one gigabyte, making it difficult to analyze complete serum mass spectrometry profiles with currently available software. However, with adequate safeguards, it appears possible to consider portions of mass spectra to find differences in peak intensities between clinical comparison groups visually. To facilitate this, the elution profile was divided into 2-min intervals in which mass spectrometry data were averaged. This required that molecular species had defined reproducible elution times. Given liquid chromatography coupled to mass spectrometry variation, misalignment of elution times of individual peaks occurred often. Hence, internal time controls were identified within each window and used for elution time normalization. This significantly reduced variability in data. This approach allowed for peak alignment across samples, improving biomarker discovery.

INTRODUCTION

Medical diagnosis has benefited from the measurement of serum molecules that are present in lesser or greater abundance in individuals having a specific disease. It is anticipated that additional biomarkers might provide new diagnostic capabilities, provide more accurate diagnosis allowing for the subclassification of disease or staging of disease progression, or even provide risk assessment as part of medical evaluation. If one could achieve a broad survey of biomolecules altered by a given disease, it may be possible to define those molecules that actually participate in the disease process and are consequently potential drug targets. Finding such markers then is of significant medical interest. Mass spectrometry (MS) has been gaining popularity as a means of searching for novel biomarkers in readily available bodily fluids such as serum, plasma,^1–3 urine,⁴ ocular fluids,^5,6 and nipple aspirate fluid.⁷ Capillary liquid chromatography coupled to mass spectrometry (cLC-MS) has been favored over other techniques in the search for new biomarkers in our laboratory due to its ability to provide a more comprehensive assessment of the peptides and small proteins present in complex samples than is possible with most other MS methods. This very ability to study hundreds to thousands of molecular ions in a single sample also generates large amounts of data, and so poses a major challenge to the use of such methods in finding new biomarkers due to increased complexity of data analysis. For example, a single cLC-MS run of a single serum sample produces an MS spectral data file that typically reaches one gigabyte or greater in size with 4000–5000 observable peaks. As one considers comparisons of clinical groups, each represented by dozens of specimens, the problem is compounded. One major challenge is having a straightforward way of addressing day-to-day or even run-to-run chromatographic variability so as to then allow meaningful comparisons of the spectra of samples run at different times.

Serum is an ideal medium in which to search for new biomarkers because it interfaces with the vast majority of cells in the body and so likely contains the imprint of diseased cells in the form of cell specific proteins and peptides, alterations in small signaling molecules, increased protein fragments, etc.⁸ Also, because it is routinely collected, it is considered to be a minimally invasive specimen and one that could be obtained repeatedly as part of prospective studies. Several groups have studied serum and plasma with the goal of proteomic biomarker discovery for the diagnosis of various cancers and other diseases. A recent review provides a good overview of putative biomarkers for human cancer discovered by serum proteome analysis.⁹ Serum, though easily acquired and well suited for biomarker discovery, presents challenges. For example, serum contains tens of thousands of different polypeptides ranging in size from a few to several hundred amino acids and spanning nine orders of magnitude in concentration.^10,11 The 22 most abundant proteins in serum account for 99% of the total protein in serum.²

Due to the extreme complexity of the protein complement found in serum, made even more complicated by variable post-translational modifications, including protein glycosylations, a separation step is required. Capillary column chromatography allows for the direct interfacing of columns with the mass spectrometer while providing a robust fractionation step. However, whole serum can easily foul capillary columns. Hence, removal of abundant, but typically uninformative, proteins is desirable. Additionally, MS instrumentation typically has a detection range of only two orders of magnitude, so the observation of low-abundance species is difficult without removal of the highly abundant serum proteins.¹² For all of these reasons, pre-analytical sample simplification of serum is needed prior to MS analysis.

A common method of simplifying a serum sample involves precipitating the large, high abundance proteins out of solution using an organic solvent. In addition to removing large proteins, this treatment has the added benefit of denaturing proteins that act as carrier molecules, liberating smaller bound factors. Because small proteins and peptides will be filtered from blood by the kidney,^10–14 small molecules that the body deems important enough to keep around are bound to larger proteins to prevent their loss, giving them a longer half-life within the bloodstream. Consequently, denaturing serum substantially increases the number of small proteins and peptides available for MS assay. However, many published methods of serum simplification remove these larger molecules without taking steps to ensure that any smaller proteins bound to them are released. It is in this low-molecular-weight low-abundance fraction of the human serum proteome that many biomarkers are likely to be found.^2,13

Even with over 99% of total protein removed from the serum samples via organic solvent precipitation, there remain in solution thousands of molecular species. We have developed a cLC-MS proteomics approach to search for potential biomarkers in large and complex data sets produced from analysis of human serum samples. Protein removal is followed by a reverse-phase capillary column step to help to separate the sample over time, reduce signal-suppression and allow us to obtain a more complete overall survey of the low-molecular-weight proteome of the patient that has been our focus.

When such a sample is analyzed using cLC-MS, a large and very complex three-dimensional data set is produced. Currently available software has been very good at three-dimensional comparisons of simple data sets, but has proven inadequate when working with the large, complex data sets produced using our methods. Lack of current software capable of analyzing these data sets has led to searching for peptides of interest within our data manually.

The complex nature and sensitivity of the LC instrumentation employed, as well as slight variations in the hand-packed columns used, cause the elution times of peptides and proteins to vary from day-to-day and run-to-run. Although the start time of the overall elution profile may vary from sample to sample, the elution order and the relative elution times of molecular species within a single sample run stay generally constant. Because most chromatographic peaks require 1–2 min to elute completely and in an effort to reduce the size of data sets and keep peak number manageable, we have defined 2-min elution windows throughout the chromatogram to allow comparison across several specimens in a sample set or between sample sets. Within each window we have located a central peak, typically a peptide, found in all specimens that could be used as an internal reference allowing for time alignment. Once aligned, the spectra can be overlaid and searched for quantitative differences between comparison groups. Any peaks demonstrating significant quantitative differences can then be further studied as possible biomarkers.

MATERIALS AND METHODS

All reagents were purchased from Sigma (St. Louis, MO) and used without further purification.

Sample Collection

Blood was obtained by antecubital venipuncture at the Brigham Young University Student Health Center from healthy volunteers. Blood was allowed to clot for 30 min at room temperature, and after centrifugation at 3500 rpm for 15 min, serum was collected, aliquoted, and frozen and maintained at −80°C until further processing.

Acetonitrile Precipitation

This method has been described in detail in a previous paper.¹⁵ Briefly, two volumes of high-performance liquid chromatography (HPLC)-grade acetonitrile (400 μL) were added to 200 μL serum, vortexed vigorously for 5 s, and allowed to stand at room temperature for 30 min. Precipitated proteins were removed by centrifugation. An aliquot of supernatant (550 μL) was then combined with 300 μL HPLC-grade water, mixed, and the mixture lyophilized to ~200 μL in a vacuum centrifuge (CentriVap Concentrator, Labconco, Kansas City, MO) to eliminate acetonitrile. Supernatant protein concentration was determined using a BioRad microtiter-plate protein assay (BioRad, Hercules, CA) according to manufacturer’s instructions. An aliquot containing 4 μg protein was transferred to a new micro-centrifuge tube and lyophilized to a volume < 20 μL. The sample was brought to 20 μL with HPLC water to which was added 20 L 88% formic acid and the sample was mixed vigorously.

cLC-MS analysis

Capillary liquid chromatography to fractionate or separate peptides and proteins was performed using a 15 cm ×250 μm i.d. capillary column, packed in-house using POROS R1 reversed-phase material (Applied Biosystems, Framingham, MA). Elution was accomplished employing a 2.2%/min gradient from 0% organic to an organic concentration of 60% acetonitrile containing 0.1% formic acid, followed by a 3.5%/min gradient up to a concentration of 95% organic phase. Chromatography used an LC Packings Ultimate Capillary HPLC pump system with a FamOS autosampler (Dionex, Sunnyvale, CA) controlled by the mass spectrometer software (Analyst, Applied Bio-systems). The cLC was coupled directly to the mass spectrometer and effluent was sprayed directly into a QSTAR Pulsar I quadrupole orthogonal time-of-flight mass spectrometer (Applied Biosystems). Data were collected for m/z 500–2500 over the entire chromatogram (55 min total including void volume, elution, and re-equilibration). Data collection, processing, and preliminary formatting were accomplished using the Analyst QS software package with BioAnalyst add-ons (Applied Biosystems).

cLC-MS/MS Analysis

A specimen containing 0.5 μg of total protein was loaded onto the column and an MS run was performed to determine the exact elution time of the peptide of interest. An MS/MS run was performed on the same amount of protein with collection of Multi-Channel Acquisition fragmentation data at a fixed collision energy of 30 for the 2-min time span in which the peptide of interest eluted. The fragmentation spectra were sent to MASCOT for potential identification. The first peptide used for elution time normalization had a monoisotopic mass of 1464.6 and was successfully identified as fibrinopeptide A.

Selection of Reference Peaks for Chromatographic Normalization

Using 2D and 3D visualizations of complete serum runs, multiple peaks that eluted at ~2-min intervals were selected as possible reference peaks. The extract ion chromatogram (XIC) function was used to check the chromatographic elution profile of a small m/z range that included the m/z of the possible reference peaks. To be further considered as possible reference peaks, chromatographic elution profiles of the selected peaks had to be relatively narrow (< 2 min), well shaped (close to Gaussian in shape), and having elution profiles that were well resolved and distinguishable from those of other molecular species of similar m/z. Those peaks that best met these criteria were further investigated in serum runs from various individuals prepared and run on different days. Ten peaks that eluted at ~2-min intervals and were ubiquitously present in all serum samples were selected as markers.

Normalization of Elution Times

Differences in elution time were corrected by aligning the central reference peaks. In practice this was accomplished by generating an averaged mass spectrum of the interval represented by exactly 1.00 min before and after a given reference peak, irrespective of its actual chromatographic elution time. Comparable mass spectra were generated for each specimen in the comparison groups for each of the 10 internal time controls. To evaluate the utility of the approach two peaks (one running before and one running after the reference peak for first window) were monitored in several samples of a single pool of normal serum. There were several aliquots of this pool assayed on the same day (n = 10 on two separate occasions) and several aliquots (n = 17) of the pool assayed over several days. Initially, the actual elution time of each test peak was obtained from the XIC plot of each of 37 runs, 20 run as part of two same day experiments and 17 on other days. As a quantitative indication of the effectiveness of time normalization, we looked at variability around two means. The first mean was the mean elution time of each of these two peaks as established for the whole set, as well as for same day and day-to-day runs. Then the mean elution time variance was calculated and this was used to find the absolute time difference between that mean and an individual peak elution time for each analysis. These differences between actual value and mean value were considered the measure of variability without alignment and were compared with normalized data. To determine variability of the time normalized data first the absolute time difference between the internal control and each of the two test peaks for each sample was calculated. This was termed the offset time. These offset times in turn were averaged and the absolute difference between a given offset time and the mean offset time was calculated. The variability in this difference was then considered to be the variability after time normalization and was compared to the unnormalized data statistically.

Statistical Analysis

Numerical data are presented as the mean ± SEM. Comparisons of variability were made by compiling the individual elution times for test peaks, determining the mean and finding the absolute difference between an individual elution time and the mean. The variability in the difference between the mean elution time and the actual individual elution times was then compared to normalized data. The two sets of time variability were compared by Student’s t-test. A p-value < 0.05 was considered significant.

Comparison of Two Groups of Human Sera

As proof of principle, the time alignment approach was applied to eight sera from normal male and eight normal female serum specimens.

RESULTS

In an effort to correct for differences in day-to-day and even run-to-run chromatographic variation between samples an approach was developed using internal controls. To do this, 10 molecular ion peaks were selected that were present in all human serum samples and that eluted at approximate 2-min intervals throughout the elution period (Figure 1). These were used for time normalization of serum samples run by cLC-MS. It was unimportant as to whether the peaks selected were in any particular charge state, whether they represented the monoisotopic peak within a charge envelope, or were even peptidic in nature (Table 1). Provided that these internal control species had a recognizable m/z value, eluted at a point of interest, were consistently found in all specimens and were chromatographically well behaved, their structure or chemical identities were not necessary for them to be useful. In reality, the identities of these control species are unlikely to be interesting in the context of biomarker discovery, given their abundant and uniform nature.

FIGURE 1.

Selection of internal elution time alignment standards. Labeled peaks represent the extract ion chromatogram (XIC) elution profiles of the 10 molecular species chosen as time alignment standards. TOF MS, time-of-flight mass spectrometry.

TABLE 1.

Properties of the 10 Selected Reference Peaks

Marker #	Reference Peak m/z	Charge State	Monoisotopic	Relative Elution Time (min)	XIC Window
1	733.3	+2	Yes	0.0	733–734
2	721.3	+2	No	2.2	721–722
3	1006.0	+2	Yes	3.7	1006–1007
4	1013.5	+5	Yes	6.3	1013–1014
5	547.3	+1	Yes	8.7	547–548
6	547.3	+1	No	10.9	547–548
7	1047.7	+1	Yes	12.8	1047–1048
8	637.3	+1	Yes	17.3	637–638
9	781.5	+1	No	19.8	781–782
10	1620.2	+1	Yes	22.1	1620–1621

The m/z value of the selected ten reference peaks is shown along with their charge state, position in the isotope envelope, average elution time (n = 160) relative to that of the first standard, and the m/z inclusion values used for each XIC window.

Nevertheless, once the 10 alignment controls were chosen as markers, attempts were made to sequence those that were peptides using MS/MS. These efforts provided partial but insufficient amino acid sequence data to allow for identification of all but one of the standards due to their large size (m/z 1013.5), the species being nonpeptidic in nature (m/z 547.3), fragmentation being poor (m/z 1006.0), fragmentation producing only very small fragments (m/z 721.3), or for other reasons. To date only the first standard (m/z 733.3 where z = 2) has been successfully identified. The peptide was fibrinopeptide A (Figure 2).

FIGURE 2.

The first serum standard was identified as fibrinopeptide A. Shown are the cLC-MS/MS fragmentation spectra of DSGEGDFLAEGGGVR found in gi|229185 and the resultant identification made by submitting the data to the MASCOT search engine. Dotted green lines connect m/z labels with corresponding peaks.

Time windows (2.0 min) centered on each of the XIC elution times of the 10 standards allowed for time normalization. This in turn allowed for the overlay of up to 16 different spectra averaged over that 2-min window. As a demonstration of the capability of this approach, several separate runs of a single human serum (to limit biologic variability) were carried out. Two peaks in the first time window, having m/z ratios of 675.4 and 1050.5, respectively, were selected to establish the effects of time alignment. The actual elution time for each of these two species was determined for 37 separate samplings. Then the relevant windows were aligned and the time differences between runs were calculated with respect to the reference peak (m/z 733.4). This brought about a marked and significant reduction in the variability of the time measurement. This is summarized in Table 2. Specimens run on the same column on the same day were more consistent in terms of their actual elution times than those run on separate days. Nevertheless, the internal time controls allowed for a significant reduction in variation in every case (Table 2).

TABLE 2.

Comparison of the Variability in Raw Elution Times Without Time Alignment with the Variability in the Offset Times After Time Alignment for Two Representative Peaks

Data Set	No Normalization	Time Normalized
Same Day Set 1 (n = 10)	Δ From Mean + SD	Δ From Mean + SD	P Value
Peak 1	0.090 ± 0.048 min	0.010 ± 0.014 min	0.00029
Peak 2	0.060 ± 0.040 min	0.010 ± 0.012 min	0.0016
Same Day Set 2 (n = 10)	Δ From Mean + SD	Δ From Mean + SD	P Value
Peak 1	0.110 ± 0.092 min	0.020 ± 0.017 min	0.0070
Peak 2	0.080 ± 0.065 min	0.010 ± 0.013 min	0.0044
Day-to-Day Set 1 (n = 17)	ΔFrom Mean + SD	ΔFrom Mean + SD	P Value
Peak 1	1.06 ± 0.88 min	0.15 ± 0.075 min	0.000175
Peak 2	0.90 ± 0.84 min	0.23 ± 0.152 min	0.003
All Determinations (n = 37)	ΔFrom Mean + SD	ΔFrom Mean + SD	P Value
Peak 1	0.78 ± 0.65 min	0.09 ± 0.07 min	0.002
Peak 2	0.80 ± 0.57 min	0.18 ± 0.11 min	0.003

Peak 1 = m/z 675.4, peak 2 = m/z 1050.5. Both peaks were found in the first elution window which also contained the central time alignment peak (m/z 733.4).

As a proof of concept, comparison of two sets of human sera, one set from males and one set from females was made after time normalization of one window. An initial review found reasonably well aligned peaks with consistent abundances for most peaks but with several peaks demonstrating obvious quantitative differences. A representative peak quantitatively different between males and females is shown (p < 0.0001, see Figure 3) and was located in elution window 7.

FIGURE 3.

Serum spectra overlay after time alignment for biomarker discovery. The extract ion chromatogram (XIC) elution time for the 7th elution standard was recorded for each of 8 male (blue) and 8 female (red) sample runs. A 2-min window (window 7) was centered on this time and the 16 spectra were overlaid. The resultant spectral compilation was then searched for biomarker candidates. Several were found, but only one (m/z 581.3) is shown. Notice that in this region the intensity of the blue peaks was typically somewhat greater than the intensity of the red peaks, but this pattern suddenly changed for the peak envelope (m/z 581.3 ¹²C monoisotopic peak) in the middle where female sera consistently contained more of this species than males (p < 0.0001). Such differences would indicate a candidate contained biomarker. TOF MS, time-of-flight mass spectrometry.

DISCUSSION

Ten internal serum peptides were selected as standards for cLC-MS time normalization. Many more peaks met the criteria needed to be considered as internal standards. Those chosen were selected in part for their location within the cLC chromatogram and their convenience in use. It was unimportant as to whether the peaks selected were in any particular charge state, whether they represented the monoisotopic peak within a charge envelope, or were even peptidic in nature.

Of those finally selected, most were monoisotopic but a few were not. If a monoisotopic molecular species did not conform to the selection criteria, a species within the ¹³C isotope envelope that did conform to the selection criteria was chosen instead. Use of these internal peaks as controls produced consistent elution time alignment across all sample runs and this in turn allowed for visual inspection of spectral overlays in the search for potential biomarkers. It is worth noting that a highly specific and useful proteomic control does not require knowledge of its actual composition provided it has a consistent and identifiable mass, is sufficiently abundant and resolved from other species. Attempts at identification of the ten time standards found that some of the current markers are not peptides but were still useful in this approach. Likewise, a highly specific proteomics pattern can be useful in the prediction or diagnosis of a disease in the absence of the complete knowledge of the chemical composition of the individual constituents in that pattern, although such information may prove important in the understanding of the disease process.

Using this approach, we demonstrated that normalizing elution times in the straight forward manner described here markedly decreased timing misalignment for any specific peak across multiple analyses of the same specimen. This was evident as a marked decrease in variability of elution time relative to the standard versus the actual cLC elution times. The impact of this is potentially large. For intra-assay variability, the average deviation from the mean was ~0.1 min without normalization with some specimens differing by as much as 0.3 min above or below the mean. Considering 2 SD, this creates a range of values that varies somewhat over 0.3 min. Even so, this is reduced to a mean of ~0.01–0.02 min and a range that spans 0.04 min after normalization. The interassay or day-to-day variability is even more impressive. In this case the mean elution time variability was ~1 min with a range (± 2 SD) of ~3.5 min. After time alignment the mean time variability fell to ~0.2 min with a range of ~0.4 min. Given that our elution windows are 2.0 min, it is entirely possible that a given peak might be inadvertently omitted if time normalization were not employed.

The significance of these results is primarily in the context of biomarker discovery. With advances in mass spectrometric technology has come the possibility of assessment of biological molecules, even in the serum. Likewise, with more innovative proteomic approaches there have been and will continue to be dramatic increases in both the size and complexity of the data sets accessed by the instrumentation. The need for additional biomarkers for medical diagnosis and clinical risk assessment of many diseases is widely acknowledged. Analysis of these data is becoming increasingly difficult and advances in software capable of analyzing these large and complex data sets has lagged behind advances in instrumentation. cLC-MS produces such massive data sets that most current software cannot manage such file size and it is necessary to parse the elution interval into discrete reproducible windows. While this reduces the file size and makes more manageable the absolute number of peaks to be considered, it creates the possibility that peaks of interest actually overlap two windows or partially fall outside the relevant time window. If time is misaligned, then more or less of a given molecular species will be included in the analyzed window potentially giving rise to differences in peak heights and areas due to location in the window, which is related to the uncertainties in elution time rather than to clinically related differences. We provided one such demonstration with the analysis and comparison of several sera from males with several sera from females. When spectra from two clinical groups are overlaid, our experience has clearly shown that the eye is very good at locating peaks that are quantitatively different (Figure 3). The approach assured that the time window considered was consistent across all subjects and increased confidence that the quantitative differences observed were due to biological differences and were not due to (or lost because of) elution time variance. Our data would suggest that the use of this time alignment approach involving endogenous internal controls dramatically reduces these time difference problems and allows for successful identification of peaks that differ between clinical groups. While this approach is somewhat laborious and time intensive, use of this method can and has led to the discovery of several statistically significant novel clinical biomarkers that are under active investigation.