Metabolites are the end products of all cell function. Metabolomics is the study of these small molecules such as sugars, lipids, and amino acids using analytical techniques including liquid chromatography coupled to mass spectrometry (LC-MS) and nuclear magnetic resonance (NMR). These instruments can measure differences between disease states and drug treatments as well as other physiological effects on cells. Ideal sample preparation is highly precise, accurate, and efficient with minimal metabolite loss. However, there is currently no universal method for isolating and analyzing these metabolites efficiently. Additionally, there is a lack of effective normalization strategies in the field of metabolomics, making comparison of results from different cell types or treatments difficult. The traditional method for data normalization uses protein concentration; however, the buffers used to extract metabolites effectively are not compatible with proteins, leading to protein precipitation and erroneous protein concentration measurements. Although time-consuming, cellular imaging would provide the best results in normalizing for cell number. We conducted a study comparing alternative filtration methods during sample preparation. Two types of filtration plates were analyzed for a variety of compound classes, features, and levels of sensitivity. Data normalization was tested by linearly increasing cell number and comparing whether protein concentration, cell debris weight, or DNA concentration increased at a rate that was proportional to cell number. Sample preparation was found to be more efficient using protein precipitation plates compared to phospholipid removal plates during lysate filtration. After testing several techniques to normalize data by cell number, DNA extraction from the cell debris demonstrated the most time-efficient, consistent, and reproducible results.
Metabolomics is the study of small molecules, such as lipids, amino acids, and sugars present within the cell.1,2 This field gives insight into cellular phenotype and function.3 Untargeted metabolomics is the common method used to repeatedly and quickly identify as many small molecules as possible at a specific occurrence using high resolution/high mass accuracy mass spectrometry.1-4 As a response to variations in environment or drug treatment, changes in metabolites can give insight into what is happening within the cell. Detection of these changes can facilitate identification of possible cancer oncogenesis.1
Despite the potential of metabolomics to provide new insights into cancer, it suffers from a lack of robust methodology. For example, at the time of this writing, the largest number of metabolites reported for a single study is approximately 400, whereas the number of metabolites in the human metabolome is estimated to be on the order of 40,000.2 Since the amount and types of measured metabolites are largely influenced by the method chosen for sample preparation, one focus of this report is sample preparation-related factors that could improve both the number of measured metabolites and the accuracy or precision with which particular metabolites are measured.2
Ideal sample preparation involves a procedure that is reproducible, fast, simple, and free of metabolite loss or degradation during extraction.2 During collection, the noise-to-signal ratio is problematic, but there are several ways to decrease it. One method focuses on protein precipitation. In this technique, metabolites are extracted from cells using solvents that quench metabolism and are filtered to remove protein and cell debris.5 Using a different approach, Ferreiro-Vera et al. found that using solid-phase extraction (SPE) to remove phospholipids yielded three- to four-fold up-regulation of metabolite signals compared to liquid-liquid extraction (LLE).5 We chose to focus on SPE due to its significant increase in metabolite signals and its simpler workflow.
Another problem associated with metabolomic data is the absence of robust methods for data normalization. Because raw metabolite measurements exhibit variation attributable to a number of sources, identification of a normalizer capable of reducing or eliminating the effects of such variation is important. Typically, total protein content or cell numbers are used.6 However, measuring total protein is not straightforward, as the buffers needed to extract metabolites are not compatible with proteins; for example, methanol in the buffers causes protein precipitation. While cell imaging as a surrogate for quantifying cell number is a good option, it is time-consuming and therefore not preferred because of the time-sensitivity of several drug treatments.6 Thus, an alternative, faster method that employs the same sample for both metabolite extraction and data normalization is needed.
Mass spectrometry (MS) was preferred over NMR spectrometry because MS has greater sensitivity, selectivity, and range of identifiable metabolites.11 One weakness of MS relative to NMR, however, is that the sample cannot be recovered. MS allows the user to easily identify the metabolite by comparing the molecular weight of the molecule to the Tandem mass spectrometry (MS/MS) spectra and retention time of the sample.11
The overall objectives of our research were to improve metabolomic data quality by 1) identifying a robust surrogate for the cell number method to facilitate data normalization and 2) optimizing sample preparation methods. Two different sample filtration strategies were tested—one using a protein precipitation plate and a second using a new phospholipid-depleting filter plate. Additionally, several data normalization strategies were evaluated. Finally, we applied the resulting optimizations to investigate the mechanism of action of L-asparaginase through its metabolic response.
MATERIALS AND METHODS
OVCAR-8 and OVCAR-4 are ovarian cancer cell lines from the subpanel of the NCI-60, one of the most highly profiled sets of cell lines in the world. Unlimited, easily obtained, and manipulated, OVCAR-8 cells are sensitive to L-asparaginase, while OVCAR-4 cells are resistant.10 The U251 central nervous system line and MDA-MB-435 melanoma line were also used.9 All cells were routinely maintained in RPMI-1640 media containing 5% fetal bovine serum (FBS) and 1% (2 mM) L-glutamine. Cultures were grown in an atmosphere of 5% CO2 and 90% relative humidity at 37°C.
Samples were prepared through filtration of the supernatant, vacuum-assisted evaporation to dryness, and reconstitution of the dried sample. Cell samples in 10 cm dishes were placed on dry ice and washed twice with 4-5 mL of 60% CH3OH, 40% HPLC-grade H2O, and 0.85% (NH4)HCO3, then extracted using 500 µL of CH3OH:CHCl3:H2O (7:2:1) buffer.7,8 The cells were homogenized for 30s to lyse the cells. The cells were then centrifuged at 20,000 x g at 4°C for 2 min. The lysate was removed and inserted into the protein removal plate while the remaining debris in the cell was saved. In order to prepare for the DNA extraction, the cell debris is air dried because speed vacuuming makes it difficult for the pellet to fully solubilize.
A phospholipid removal plate was used because Ferreiro-Veraet al. suggested it as a better way to effectively remove noise- or suppression-imparting metabolites.5 We compared the protein removal plate to the phospholipid removal plate. Each well was filled with 500Lfresh CH3OH with 10 µM xylitol. 200 L of the lysate was then added to each well. After using the vacuum to filter the lysate through the plates, the samples were dried using a speed vacuum and later reconstituted and analyzed in a hybrid ion trap MS.
Samples stored at -80°C were reconstituted in 50 µL of mobile phase A (deionized H2O + 0.1% Formic Acid (FA)) immediately prior to analysis with LC-MS. The LC parameters were as follows: 5µL of metabolomic sample were injected onto a C18 column (1.8µm particle size, 2.1 x 50mm), and analytes were eluted using a 25 min linear gradient of 100% H2O + 0.1% FA to 100% ACN + 0.1% FA. Analytes were detected in positive ion MS mode, and data was analyzed using XCMS Online software (Scripps Research Institute, La Jolla, CA). Averaged metabolite intensities for identified compounds were then plotted against cell number.
As a surrogate protocol for counting cell number, total genomic DNA was extracted and DNA concentration was quantified using a DNA extraction kit from IBI Scientific, following the Red Blood Cell Preparation protocol. After the cell pellets were completely air dried at room temperature, the pellet was reconstituted in 150 L of lysis buffer. The pellets were then stored at room temperature in the buffer for at least 24 h to allow for maximum resolubilization. DNA preparation was performed by adding GB buffer, shaking vigorously, and incubating at 60 °C for at least 10 min to allow for complete solubilization of the pellet. Afterward, the solution was treated with 100% C2H5OH and transferred to a GD maxi column. For large cell numbers, the solution was split between two columns to avoid saturation of the column. The columns were centrifuged, washed twice with wash buffer, and eluted with 50 L of elution buffer.
A phospholipid removal plate was evaluated against a protein precipitation filter plate in order to determine which plate yielded the highest intensity metabolites as well as the greatest variety of compound classes. As shown in Figure 1, both filtration plates yielded similar metabolite intensities for most of the known metabolites. However, in the case of the polyaminesputrescine and spermidine (Figure 1C), the protein removal plate significantly outperformed the phospholipid removal plate. Only one metabolite, glucosamine, displayed greater signal intensity with the phospholipid removal plate (Figure 1B). This result suggests that the phospholipid removal plate may also be useful for future analysis of other carbohydrates.
Protein Concentration and Cell Debris Mass
To attempt to normalize cell number, we tried measuring protein concentration. As shown in Figure 2, extraction of proteins with lysis buffer (Figure 2A) worked well in comparison to extraction with CH3OH/acetonitrile (Figure 2B), which showed low protein concentration. These were compared using levels of vehicle and L-asparaginase. To obtain better results, we tried another method for cell normalization using cell debris mass. Figure 3 shows that, in systems with fewer cell numbers, cell debris mass showed a low increase in weight; however, in samples with higher cell counts, the mass increased greatly. Neither cell debris nor protein concentration produced a linear relationship with cell number when using conditions and buffers required for metabolomics (Figure 2B and 3).
DNA concentration was measured for four different cell lines to evaluate its association with cell number. First, cell dilutions ranging from 0.25 – 4 million cells/dish were seeded and processed to determine DNA concentration. Figure 4 illustrates a positive linear correlation between cell number and DNA concentration among all four-cell lines investigated. Cell number was then evaluated by DNA concentration throughout a 48-hour incubation period. DNA concentration was recorded at 0, 4, 8, 24, and 48 hours after incubation. Analysis of the OVCAR-8 and OVCAR-4 cell lines suggests that a greater number of cells yielded faster growth (Figure 5). However, because cell doubling was not seen within the first 24 hours, these data suggest a slight lag phase in growth after feeding. This observation suggests that proximity between cells affects their growth (i.e., some lines grow faster once confluence exceeds a threshold level).
To test this hypothesis, OVCAR-4 cell growth was monitored by cell count (Figure 6) and DNA concentration (Figure 5B). Our results were consistent with the published doubling time of 39 hours for this cell line. However, the results also indicated a lag phase in the doubling of DNA concentration (Figure 5B) and cell number (Figure 6), although the latter lag phase was subtle.11 To validate the existence of the growth lag observed in OVCAR-4 cells, we investigated cell growth by DNA concentration for two additional cell lines: U251 and MDA-MB-435. Because cells of these two lines tend to grow individually, we expected to observe no lag phase. However, as shown in Figure 5C and 5D, both lines exhibited a growth lag. This lag is smaller than that of the OVCAR-4 line but greater than the growth lag of the OVCAR-8 line, as shown by the differences in DNA concentration between the 0 and 8 hour time points.
Metabolomic data quality can be dramatically affected by sample preparation and data normalization methods amongst many other factors. Ferreiro-Vera et al. reported that removing phospholipids provided improved signal intensity for majority of retrieved metabolites. Based on these results, we hypothesized that phospholipid depletion would improve metabolite signal intensity across a diverse set of metabolites extracted from the ovarian cancer cell line OVCAR-8.2 However, we found that a protein removal plate yielded greater signal intensity than the phospholipid removal plate for most metabolites. Specifically, the polyamines were significantly depleted in the phospholipid removal plate. Nevertheless, the phospholipid removal plate yielded a notable amount of glucosamine, suggesting that this plate may be helpful for the examination of carbohydrate metabolites.
Initial attempts to use protein concentration and cell debris mass from metabolomics samples failed to provide a solution for normalization. We initially tried normalizing cell number using protein. However, it was an issue because we miss the soluble proteins by measuring the debris. Protein concentration works well with bicine-chaps buffer, which keeps the proteins in solution. However, the buffers required for metabolite extraction do not preserve the proteins. Also, the methanol used for extraction causes precipitation of the proteins, making the concentrations measured extremely low. Realizing that the cell pellet could be put to use, we attempted to normalize cell number by cell debris mass. This method proved to be inefficient because the analytical scale was not precise enough to accurately give the same weight in repeated measurements. In the end, DNA extraction from the cellular debris proved to be an extremely effective method of normalization. Using OVCAR-8 and OVCAR-4 cell lines, DNA concentration was shown to increase proportionally to cell number. For time course experiments, cell number can therefore be normalized using DNA extraction from the metabolomic cellular debris.
In conclusion, this study has shown that protein precipitation plates are superior to phospholipid removal plates for most metabolite classes. However, for sugars, it may be more advantageous to use phospholipid removal plates. Rather than manually counting cells through a microscope, DNA extraction can efficiently normalize cell number without affecting the time required for metabolomic analysis.
I would like to thank Dr. John Weinstein for allowing me to have this wonderful research experience, Dr. Phil Lorenzi for supporting me and providing feedback to help make me a better scientist, and Dr. Leslie Silva for her daily guidance, advice, and support through my whole experience at MD Anderson Cancer Center. This research was supported by the University of Texas, MD Anderson Cancer Center.