Quantifying the Molar Percentages of Cholesterol in Supported Lipid

Dec 3, 2012 - Quantifying the Molar Percentages of Cholesterol in Supported Lipid Membranes by Time-of-Flight Secondary Ion Mass Spectrometry and ...
0 downloads 0 Views 928KB Size
Download PDF
Article pubs.acs.org/ac

Quantifying the Molar Percentages of Cholesterol in Supported Lipid Membranes by Time-of-Flight Secondary Ion Mass Spectrometry and Multivariate Analysis Robert L. Wilson† and Mary L. Kraft*,†,‡ †

Department of Chemistry, and ‡Department of Chemical and Biomolecular Engineering, University of Illinois at Urbana−Champaign, Urbana, Illinois 61801, United States S Supporting Information *

ABSTRACT: The uneven cholesterol distribution among organelles and within the plasma membrane is postulated to be critical for proper cellular function. To study how interactions between cholesterol and specific lipid species contribute to the uneven cholesterol distribution between and within cellular membranes, model lipid membranes are frequently employed. Although the cholesterol distributions within membranes can be directly imaged without labels by using time-of-flight secondary ion mass spectrometry (TOF-SIMS), quantifying the cholesterol abundance at specific membrane locations in a label-free manner remains a challenge. Here, partial least-squares regression (PLSR) of TOF-SIMS data is used to quantitatively measure the local molar percentage (mol %) of cholesterol within supported lipid membranes. With the use of TOF-SIMS data from lipid membranes of known composition, a PLSR model was constructed that correlated the spectral variation to the mol % cholesterol in the membrane. The PLSR model was then used to measure the mol % cholesterol in test membranes and to measure cholesterol exchange between vesicles and supported lipid membranes. The accuracy of these measurements was assessed by comparison to the mol % cholesterol measured with conventional assays. By using this TOFSIMS/PLSR approach to quantify the mol % cholesterol with location specificity, a better understanding of how the regional lipid composition influences cholesterol abundance and exchange in membranes may be obtained.

C

(TOF-SIMS) is one of the few techniques that permit directly imaging the cholesterol distribution in membranes without the use of labels.21−25 By using TOF-SIMS to map the intensities of the cholesterol-specific fragment ions within phase-separated supported membranes, the preferential associations of cholesterol with sphingomyelin and dipalmitoylphosphatidylethanolamine have been imaged.21−24 The ability to quantify the cholesterol concentration at specific locations in model membranes would facilitate identifying the factors that control cholesterol distribution among and within biological membranes. Presently, cholesterol transfer can be quantified with the methods mentioned above (i.e., QCM or radiometric methods),11−14 whereas cholesterol content within different lipid phases can be measured with nuclear magnetic resonance (NMR).26,27 However, a chemically specific and quantitative imaging method is required to permit correlating membrane composition with morphology or location. Though TOF-SIMS can image unlabeled cholesterol with location specificity,21−24 the secondary ion yields are influenced by the local chemical environment, so secondary ion intensities are not directly proportional to the parent molecule’s concentration.21,22,28 Various SIMS techniques have been combined with selective isotope labeling to permit measuring

holesterol is a cell membrane component whose abundance influences a diverse assortment of fundamental cellular processes, including endocytosis, intracellular vesicle transport, and cell adhesion.1−3 Cholesterol is unevenly distributed between the membranes of cellular organelles, transitioning from low abundance at the beginning of the secretory pathway to high abundance at the end.4 In addition, differential affinities between cholesterol and various lipid species are postulated to drive the formation of functionally distinct plasma membrane domains that are enriched with cholesterol and select lipid species.5 Because changes in not only the total intracellular cholesterol level but also the cholesterol distribution among organelles and within individual membranes have been linked to disease, the mechanisms that control cholesterol distribution in biological membranes are the subject of much research.6−10 A variety of techniques have been employed to investigate the exchange and distribution of cholesterol between and within model membranes. Cholesterol transfer between membranes is usually assessed by using quartz crystal microbalance (QCM) or with radiometric techniques.11−14 The distribution of cholesterol within a membrane is often inferred from the membrane’s local biophysical properties.15−19 Though fluorescent cholesterol analogues can be directly visualized in membranes, the distribution of the fluorescent analogue between lipid phases may differ from that of unlabeled cholesterol.20 Time-of-flight secondary ion mass spectrometry © 2012 American Chemical Society

Received: July 3, 2012 Accepted: December 3, 2012 Published: December 3, 2012 91

dx.doi.org/10.1021/ac301856z | Anal. Chem. 2013, 85, 91−97

Analytical Chemistry

Article

solution used for bilayer formation. A mosaic of fluorescence microscopy images (Leica DM6000 B, Q-Imaging EXi Blue fluorescence microscope) was created of each flash-frozen and freeze-dried membrane to facilitate analyzing specific sample locations with TOF-SIMS.15 Preparation of Supported Lipid Membrane with a Composition Gradient. A supported membrane with a spatial variation in composition was made from two different vesicle solutions (99/0/1 and 82/17/1 mol %/mol %/mol % DLPC/ cholesterol/NBD-PC) following the previously reported procedure.29 The chrome-patterned oxidized silicon substrate was incubated on top of the two droplets (each 2 μL) of vesicle solutions for 30 min at room temperature (RT), rinsed in fresh water to remove the excess vesicles, flash-frozen, and freezedried. Exchange Membrane Sample Preparation. Exchange membrane samples with two different compositions were prepared. Samples with composition 1 were cholesterol-free supported lipid membranes that were exposed to vesicles containing 17 ± 1 mol % cholesterol. Samples with composition 2 were 19 ± 1 mol % cholesterol supported lipid membranes that were exposed to cholesterol-free vesicles. To create samples with compositions 1 and 2, chromepatterned oxidized silicon substrates in a Petri dish were covered with 2.4 mL of water, and then 2.4 mL of a 0.5 mg/mL vesicle solution composed of 99/0/1 or 80/19/1 mol %/mol %/mol % DLPC/cholesterol/NBD-PC, respectively, was added to the dish. After 30 min at RT, 2.4 mL of a 0.5 mg/mL vesicle solution composed of 82/17/1 or 99/0/1 mol %/mol %/mol % DLPC/cholesterol/NBD-PC was added to the dish that contained samples with compositions 1 and 2, respectively. Each dish was incubated for another 30 min at RT, and then the supported membranes were rinsed, flash-frozen, and freezedried. TOF-SIMS. TOF-SIMS was performed on a PHI Trift-III TOF-SIMS (Physical Electronics Incorporated, Chanhassen, MN) using a 22 kV 197Au+ primary ion beam. TOF-SIMS data were acquired in unbunched mode from 50 μm by 50 μm or 40 μm by 40 μm membrane regions that exhibited uniform fluorescence intensities. Several different locations were analyzed on each homogeneous lipid membrane sample. Positive ion spectra were obtained with a mass range of 0− 800 amu and a mass resolution of M/ΔM 1433 at the 184 m/z peak. An ion dose of 3 × 1013 ions/cm2 was used for all analyses. Note that TOF-SIMS analysis is usually performed using an ion dose below 1013 ions/cm2 in order to acquire molecular ions and avoid molecular fragmentation. However, our previous studies demonstrate that, using our instrumentation, ion doses within the static limit do not yield sufficient ion signals to permit detecting compositional changes within model membranes with multivariate techniques.38 Instead, an ion dose of approximately 3 × 1013 ions/cm2, which exceeds the static limit, yields the higher counts of fragment ions that are beneficial for extracting compositional information from model lipid membranes using multivariate analysis.38 Data Analysis. Unit mass binned spectra were extracted from sample locations that corresponded to the membrane, and not the chrome grid, using Wincadence software (Physical Electronics, Incorporated, Chanhassen, MN). Mass peaks outside of the m/z 50−300 range were removed from the spectra. The intensities of the peaks above 300 m/z, which include the DLPC and cholesterol molecular ions, were too low to be reproducibly detected with our instrumentation, and

the abundances of specific membrane components,15,29−31 but the need for isotope labels complicates sample preparation. A label-free TOF-SIMS approach has been used to quantify changes in relative cholesterol abundance, but the cholesterol concentration was not determined.32 Previous reports demonstrate that multivariate analysis enables extracting quantitative compositional information from TOF-SIMS data.33−36 For example, the compositions of unlabeled polymers and protein films have been quantified by constructing multivariate models of TOF-SIMS data from samples of known composition.33−35 Although multivariate analysis of TOF-SIMS data has been used to discriminate and image unlabeled lipids according to combinations of the low-mass fragment ions that are common to multiple lipid species,37−40 multivariate analysis has not been employed to quantify membrane composition. In this report, we quantified the molar percentage of cholesterol (mol % cholesterol) within supported lipid membranes by applying a multivariate analysis technique, partial least-squares regression (PLSR), to TOF-SIMS data. PLSR is a multivariate statistical analysis technique that relates the measured variables, which are the mass peaks that were collected with TOF-SIMS, to the unmeasured property of interest, namely, the mol % cholesterol in each membrane sample.41 A PLSR model was constructed from the spectra of model membranes that varied in mol % cholesterol (calibration samples) and used to measure the mol % cholesterol in supported lipid membranes in a test set. The accuracy of the PLSR measurements was assessed by comparison to the mol % cholesterol that was independently measured with enzymatic assays. We show that, by applying the TOF-SIMS/PLSR model to a single spectrum acquired at a small region in a supported lipid membrane, the mol % cholesterol could be measured with an accuracy of a few percent. We also used this TOF-SIMS/ PLSR approach to quantify the mol % cholesterol that was transferred between vesicles and discrete regions within supported lipid membranes.

■

MATERIALS AND METHODS Preparation of Homogeneous Supported Lipid Membranes. 1,2-Dilauroyl-sn-glycero-3-phosphocholine (DLPC), cholesterol, 1-palmitoyl-2-[12-[(7-nitro-2-1,3-benzoxadiazol-4yl)amino]lauroyl]-sn-glycero-3-phosphocholine (NBD-PC), and 1,2-stearoyl-sn-glycero-3-phosphocholine (DSPC) were purchased from Avanti Polar Lipids (Alabaster, AL) and used without further purification. Millipore water (18 MΩ) was used for all experiments. Oxidized silicon substrates that were patterned with a chrome grid were fabricated as previously reported.30 The chrome grid bars were 10 μm wide, and the area of the silicon substrate that was bound by the chrome grid was 90 μm by 90 μm. Lipid membranes consisting of various mol % of DLPC (79−99 mol %) and cholesterol (0−20 mol %) plus 1 mol % NBD-PC for visualization with fluorescence microscopy were formed on chrome-patterned oxidized silicon substrates, flash-frozen, and freeze-dried following previously reported methods.15,38 The mol % cholesterol in each cholesterol-containing vesicle solution used to form a supported lipid bilayer was measured with the Amplex Red cholesterol assay kit and Amplex Red phospholiase D assay kit (Invitrogen) as previously described.15 Eight replicate measurements were made on each vesicle solution, and the mean (μ) ± one standard deviation (σ) are reported for each sample. The mol % cholesterol in each homogeneous supported lipid membrane was assumed to be the same as that in the vesicle 92

dx.doi.org/10.1021/ac301856z | Anal. Chem. 2013, 85, 91−97

Analytical Chemistry

Article

peaks below m/z 50 are not chemically specific.42 Peaks from the chrome grid (m/z 52) and poly(dimethylsiloxane) (PDMS) contamination (m/z 73, 147, 207, 221, and 281)43 were also removed from the spectra. Note that the PDMS contamination and a major cholesterol fragment at m/z 14725 could not be resolved in these experiments; we experimentally found that better accuracy was obtained when peak m/z 147 was excluded. PLSR was performed on the data using the PLS Toolbox (v6.7.1, Eigenvector Research, Manson, WA) for MATLAB (7.14.0.739, R2012a, MathWorks Inc., Matick, MA). Detailed mathematical descriptions of PLSR can be found in previous publications.41,44 Each peak was normalized to the total intensity of the remaining peaks in the spectrum, and the spectra were mean-centered. The spectra that were used for model construction (calibration spectra) were loaded in the xblock, and the mol % cholesterol that was measured with the enzymatic assay in the calibration samples was loaded into the y-block. PLSR was used to calculate the latent variables (LVs), which are the linear combinations of mass peaks that are most relevant for predicting the mol % cholesterol in the membrane samples.41 The minimum number of LVs required to capture at least 95% of the variance in the mol % cholesterol were used to construct the PLSR model. Capturing this high percentage of variance in the mol % cholesterol helps to ensure that the model is able to accurately predict the relative cholesterol concentration in the membrane samples. For preliminary model refinement, the Y-student residual value and leverage of each sample were used to identify outlier samples that would be detrimental to the model. The Y-student residual value is the number standard deviations of error in the sample’s predicted mol % cholesterol compared to the mol % cholesterol measured with the enzymatic assay. Leverage is the amount to which a sample influences the PLSR model. Samples with Ystudent residual values outside of ±1.2 or leverage values >0.1 were removed from the preliminary model. When the majority of the spectra acquired from a single sample had either high leverage or Y-student residuals, all of the spectra acquired from the sample were omitted from further analysis. Cross validation was performed using a random subset with 10 splits and five iterations. The PLSR model was applied to the spectra of test lipid membranes that were not used for model construction to measure their mol % cholesterol. Variable importance in projection (VIP) plots were generated to visualize the VIP scores for each mass peak, which are an estimate of the importance of each mass peak in predicting the mol % cholesterol in the sample. VIP scores greater than one are important for the prediction.45,46

predict the mol % cholesterol in the membrane samples. The PLSR model also captured 99% of the mass spectral variance (50%, 36%, 7%, 4%, and 2% of the spectral variance was captured by LV 1, 2, 3, 4, and 5, respectively). This indicates that the vast majority of the TOF-SIMS data was useful for predicting the mol % cholesterol in the membranes. Though most of the spectral variance was captured with the first three LVs, the 6% of the spectral variance that was captured by LV 4 and 5 was useful for modeling 18% of the variance in the mol % cholesterol, and the inclusion of these two LVs improved model accuracy. Additional LVs did not significantly increase the variance captured by the model or its accuracy, so they were not included in the final model. The PLSR model was validated by using it to measure the mol % cholesterol in the calibration samples and four homogeneous supported DLPC membrane samples with known relative cholesterol concentrations that were not used for model construction (test samples). All of these samples had been prepared and analyzed with TOF-SIMS over period of approximately 12 months. The mean (μ) mol % cholesterol and the standard deviation (σ) of the individual TOF-SIMS/ PLSR measurements made on each homogeneous lipid membrane in the calibration and test sets are listed in Table 1. The relative cholesterol concentrations measured for the Table 1. Relative Cholesterol Concentrations in Calibration and Test Supported Lipid Membranes, Where an Enzymatic Assay Was Used To Measure the Mol % Cholesterol in the Membranes That Contained Cholesterola mol % cholesterol measured by assay (μ ± σ) 0 0 4 10 14 17 19 20

± ± ± ± ± ±

2 1 1 1 1 2

0 4±1 13 ± 1 19 ± 1 a

■

RESULTS Measuring Mol % Cholesterol in Homogeneous Supported Lipid Membranes by PLSR of TOF-SIMS Data. A PLSR model was constructed of numerous spectra from seven homogeneous supported DLPC membranes with relative cholesterol concentrations that ranged between 0 and 20 mol % (calibration samples). This range of mol % cholesterol was selected because high-quality membranes did not form when higher cholesterol levels were employed.15 The final PLSR model consisted of five LVs that captured 98% of the variance in the mol % cholesterol (36%, 19%, 25%, 8%, and 10% of the variance in the mol % cholesterol was captured by LV 1, 2, 3, 4, and 5, respectively). This high level of variance in the relative cholesterol concentration that was captured by the model helped to ensure that the model would accurately

mol % cholesterol measured by TOF-SIMS/PLSR (μ ± σ) Calibration Samples 1± 0± 5± 8± 14 ± 17 ± 19 ± 20 ± Test Samples 2± 4± 15 ± 21 ±

regions analyzed

1 1 1 1 1 1 1 1

13 14 12 8 14 15 12 16

1 1 1 1

11 14 17 12

See the Materials and Methods section.

calibration samples were nearly identical to the mol % cholesterol that were measured with enzymatic assays. The relative cholesterol concentrations (2 ± 1, 4 ± 1, 15 ± 1, and 21 ± 1 mol %) measured in the four test DLPC membranes by TOF-SIMS/PLSR were also in good agreement with the mol % cholesterol measured with the enzymatic assay (0, 4 ± 1, 13 ± 1, and 19 ± 1 mol %, respectively). This demonstrates that the cholesterol concentration-specific spectral variation did not change due to instrumental drift over the year in which the TOF-SIMS data were acquired, enabling accurate measurement of mol % cholesterol. To acquire location-specific compositional data, the PLSR model must be applied to a single spectrum acquired at the region of interest. Therefore, we also investigated the precision and accuracy of the PLSR measurements made with each 93

dx.doi.org/10.1021/ac301856z | Anal. Chem. 2013, 85, 91−97

Analytical Chemistry

Article

PLSR model to a single spectrum will enable quantifying relative cholesterol abundance at locations of interest in a membrane. The importance of each peak toward identifying the mol % cholesterol is shown in the variable importance of projection (VIP) scores plot (Figure 1, parts B and C). Peaks related to cholesterol (m/z 105, 107),25,47 DLPC (m/z 58, 59, 86, and 184),37 and NBD-PC (m/z 125) have VIP scores greater than unity (Figure 1, parts B and C), and thus, their normalized intensities vary according to the mol % cholesterol in the membrane. The high importance of the hydrocarbon peaks that are common to both lipids and cholesterol (m/z 55, 57, 91, and 95)25,47 suggests that the relative intensities of these peaks vary between the cholesterol and lipid spectra. Because the PLSR model related differences in relative intensities of the lipid- and cholesterol-related fragment ions to the mol % cholesterol in the membrane, accurate measurements were obtained despite our use of unit mass binned spectra with a limited mass range. However, we expect that this accuracy could be further improved by acquiring spectra with a TOF-SIMS instrument that reproducibly detects the more component-specific highmass (m/z > 300) secondary ions with high mass resolving power. Exploitation of the relative intensities of the lipid and cholesterol fragment ions to identify mol % cholesterol also had disadvantages; the lipid-related peaks were specific to DLPC, so the application of this model was limited to DLPC membranes. This PLSR model did not accurately predict the mol % cholesterol in membranes composed of DSPC, which has longer fatty acid tails than DLPC, likely because DLPC produces a higher ratio of headgroup to tail group fragment ions than DSPC.38 Consequently, a new PLSR model would need to be constructed to measure the mol % cholesterol in membranes that contain different lipid species. Prediction of the Mol % Cholesterol in a Supported Lipid Membrane with a Compositional Gradient. As a proof of concept, the TOF-SIMS/PLSR model was used to quantify the mol % cholesterol at regions of interest within a supported lipid membrane that exhibited a lateral gradient in composition. This compositional gradient membrane was created on an oxidized silicon substrate that was patterned with 10 μm wide chrome grid bars that partitioned the substrate into 90 μm by 90 μm patches of oxidized silicon. This patterned substrate was exposed to two adjacent droplets (2 μL each) of 99/0/1 and 82/17/1 mol %/mol %/mol % DLPC/ cholesterol/NBD-PC vesicle solutions such that the two droplets came into contact and mixed as the vesicles within them fused to the substrate (Figure 2A). Because vesicle fusion is usually faster than droplet mixing,29 the composition of the resulting membrane would vary from 0 to 17 mol % cholesterol from one side of the substrate to the other if cholesterol transfer between the vesicles and/or the supported membrane was negligible. However, cholesterol spontaneously transfers from vesicles with high cholesterol/lipid ratios to model and cellular membranes with lower ratios,11,13,14,48−50 so the supported membrane would likely have a smaller range of relative cholesterol concentrations. Note that lipid components within a supported lipid membranes can diffuse laterally within the corralled region but cannot cross the chrome grid or spontaneously transfer between membranes under these conditions.14,29 As a result, the membrane composition in each corralled region was homogeneously mixed, but the composition within each corral varied across the substrate.

individual spectrum. The mol % cholesterol measured by TOFSIMS/PLSR of the individual spectra acquired from the calibration and test lipid membranes are shown in Figure 1A.

Figure 1. (A) Identification of mol % cholesterol in calibration and test supported DLPC membranes by PLSR of TOF-SIMS data. The spectra from seven DLPC membranes of known mol % cholesterol (calibration samples) were used to construct the PLSR model. The PLSR model was then used to reassess the mol % cholesterol in the calibration samples and to measure the mol % cholesterol in DLPC membranes of known composition that were not used for model construction (test samples). (A) Graph shows the mol % cholesterol determined by PLSR at each location that a spectrum was acquired on the calibration (cal) and test membranes. (B) VIP scores plot shows which peaks have VIP scores >1 (red dashed line) and, thus, are important for determining the mol % cholesterol in the membrane. (C) The y-axis of the VIP plot shown in panel B was rescaled to facilitate identifying the peaks with VIP scores >1 (red dashed line).

The range of the individual measurements that were made on the same membrane was almost always less than 4 mol % cholesterol. Moreover, the individual PLSR/TOF-SIMS measurements were usually within 3 mol % of the cholesterol concentrations measured with enzymatic assays (Supporting Information Table S1). This ability to measure the mol % cholesterol with good accuracy and precision by applying the 94

dx.doi.org/10.1021/ac301856z | Anal. Chem. 2013, 85, 91−97

Analytical Chemistry

Article

concentration, which would dissipate a cholesterol concentration gradient.11,13,14,48−50 Therefore, the measurement of 2 mol % more cholesterol in the membrane at the right side of the substrate than in the vesicles with the highest cholesterol content must be caused by measurement error, and not spontaneous cholesterol transfer. However, the higher relative cholesterol concentration (6 ± 0 mol %) measured in the membrane at the left side of the substrate (regions 1−8) than in the vesicles deposited at that location (0 mol %) suggests that cholesterol had transferred into these cholesterol-deficient supported membrane regions. Next, we verified that the minimum mol % cholesterol measured in the gradient membrane was higher than that of the vesicles deposited at this region on the substrate (0 mol % cholesterol) due to cholesterol transfer, and not significant errors in the TOF-SIMS/PLSR measurements. Cholesterol-free supported lipid membranes were incubated for 30 min in the presence of 17 ± 1 mol % cholesterol vesicles. TOF-SIMS/ PLSR measured averages of 3 ± 1 to 5 ± 1 mol % cholesterol (Figure 3 and Table 2) in these exchange membranes, which

Figure 2. (A) Preparation of cholesterol gradient sample. A droplet each of 99/0/1 (red) and 82/17/1 (white) mol %/mol %/mol % DLPC/cholesterol/NBD-PC vesicle solution were placed on a glass coverslip approximately 2 mm apart, and an oxidized silicon substrate with a chrome grid was placed face down over the droplets for 30 min. (B) Mosaic of fluorescent microscopy images of the resulting freezedried supported lipid membrane. The fluorescent squares are the membrane patches (90 μm by 90 μm) that are bordered by the 10 μm wide chrome grid bars, which appear dark. Regions where the membrane was disrupted during freeze-drying are intensely fluorescent (i.e., bottom left). The periodic variations in fluorescence intensity are caused by uneven illumination within the individual images that make up the mosaic. Numbers indicate the locations where spectra were acquired with TOF-SIMS. On the basis of the positions of the scratches on the substrate (thick dark lines), the mol % cholesterol should increase from left to right. (C) Mol % cholesterol measured by TOF-SIMS/PLSR at each numbered region shown in panel B. TOFSIMS/PLSR measured (μ ± σ) 6 ± 0, 5 ± 1, 9 ± 1, 19 ± 2, and 19 ± 2 mol % cholesterol at regions 1−8, 9−16, 17−21, 22−24, and 25−30, respectively.

Figure 3. Graph of the mol % cholesterol measured by PLSR of individual spectra from supported lipid membranes consisting of 0 or 19 ± 1 mol % cholesterol that were incubated in the presence of vesicles that contained 17 ± 1 or 0 mol % cholesterol, respectively.

Table 2. Relative Cholesterol Concentrations Measured by TOF-SIMS/PLSR in Supported Lipid Membranes Consisting of 0 or 19 ± 1 Mol % Cholesterol That Were Incubated for 30 Min in the Presence of Vesicles Containing 17 ± 1 or 0 Mol % Cholesterol, Respectivelya mol % cholesterol measured with enzymatic assays in vesicles used to form supported membranes (μ ± σ)

The fluorescence microscopy images of the freeze-dried supported lipid membrane show the locations where spectra were collected (Figure 2B), where the mol % cholesterol in the corralled membrane patches was expected to increase from the left side of the substrate to the right. Consistent with this expectation, the maximum mol % cholesterol (μ ± σ) was measured by TOF-SIMS/PLSR at the lower right side of the substrate (19 ± 2 mol % for both regions 25−30 and 22−24), and the relative cholesterol concentration decreased as the analysis location moved up (9 ± 1 mol % at regions 17−21) and to the left (5 ± 1 mol % at regions 9−16 and 6 ± 0 mol % at regions 1−8) on the substrate (Figure 2C). The mol % cholesterol measured at the right and left sides of the substrate (19 ± 2 and 6 ± 0 mol %, respectively) was higher than those in the vesicles deposited at these locations (17 ± 1 and 0 mol %, respectively). These discrepancies in the relative cholesterol concentration must be due to either cholesterol transfer between membranes or to measurement error, as the transfer of phospholipids between membranes is insignificant under these conditions.14 Cholesterol spontaneously transfers from membranes with high concentration to those with low

0/17 ± 1 0/17 ± 1 0/17 ± 1 19 ± 1/0 19 ± 1/0

mol % cholesterol measured by TOFSIMS/PLSR (μ ± σ)

regions analyzed

± ± ± ± ±

10 16 17 12 11

3 5 3 15 17

0 1 1 1 1

a

The mol % cholesterol in the cholesterol-containing vesicles that were used to form the supported membrane was measured with the assay described in the Materials and Methods section.

are higher than that measured by TOF-SIMS/PLSR in cholesterol-free test membranes that were not exposed to cholesterol-containing vesicles (2 ± 1 mol % cholesterol, Table 1). This confirms that cholesterol transferred into the exchange membrane, though slightly (1−3 mol %) less than that observed in the gradient membrane. This small disparity is likely due to differences in the conditions used to prepare the gradient and exchange membranes (i.e., incubation media volumes, time between supported membrane formation and exposure to donor vesicles, ratio of cholesterol-free membrane 95

dx.doi.org/10.1021/ac301856z | Anal. Chem. 2013, 85, 91−97

Analytical Chemistry

Article

surface area in the vesicles and supported membranes, and total vesicle and cholesterol concentrations in incubation media). We also evaluated the validity of the maximum mol % cholesterol measured in the gradient membrane by assessing the relative cholesterol concentrations in 19 ± 1 mol % cholesterol supported membranes that were exposed to cholesterol-free vesicle solutions for 30 min. As listed in Table 2, 17 ± 1 and 15 ± 1 mol % cholesterol were measured in these membranes with TOF-SIMS/PLSR. Thus, slightly less cholesterol transfer occurred when the donor was a supported lipid membrane than when the donor was a vesicle, which is consistent with previous reports that the rate of cholesterol transfer decreases with decreasing curvature of the donor membrane.11,13,14,16,50 These results also indicate that cholesterol efflux from the exchange membrane was slightly higher than that from the gradient membrane. Again, this slight discrepancy was likely caused by differences in the volume and composition of the incubation medium that the exchange and gradient membranes were exposed to. Overall, we conclude that the mol % cholesterol measured in the compositional gradient supported lipid membrane by TOF-SIMS/PLSR were within a few percent of that measured with the enzymatic assay.

created by incubating cells in the presence of drugs that acutely lower (i.e., methyl-β-cyclodextrin) or increase (i.e., cholesterolmethyl-β-cyclodextrin complexes) cholesterol abundance in the plasma membrane.53 Then the average relative cholesterol concentration in the plasma membrane of the cell population could be measured with the enzymatic assays employed herein or with more sophisticated approaches to selectively measure the plasma membrane composition of cell populations.54 This would permit the construction of a PLSR model that correlated the spectral features in the TOF-SIMS data acquired from individual cells from each cell population to the relative cholesterol concentration measured in the plasma membrane. Such a PLSR model could then be applied to quantify the relative cholesterol concentration in the plasma membranes of mammalian cells from the same lineage. The accuracy of this approach, however, would likely be compromised by the cellto-cell variation in the relative cholesterol concentration within the cell population. Nonetheless, we expect that future efforts that build on the work reported herein may ultimately permit quantifying the relative cholesterol composition in membranes as complex as those found in real cells.

CONCLUSION We have used calibration spectra from membranes of known composition to develop a PLSR model that successfully captured the spectral variance that was correlated to mol % cholesterol, but not the spectral variation between membranes with the same mol % cholesterol. The resulting PLSR model accurately measured the mol % cholesterol in DLPC membranes that were prepared and analyzed over the course of a year, demonstrating the robustness of this approach. The greatest accuracy in the mol % cholesterol measured by TOFSIMS/PLSR was obtained when the PLSR model was applied to multiple spectra acquired from the membrane of interest. However, application of the PLSR model to a single spectrum yielded a relative cholesterol concentration that was typically within 3 mol % of the cholesterol composition measured with an enzymatic assay, which enables accurately measuring the mol % cholesterol at a specific membrane location. Further improvements in the accuracy of the measurements can likely be obtained through the use of the polyatomic primary ion sources that are known to enhance the yields of molecular ions51,52 that are specific to individual membrane components. Using this TOF-SIMS/PLSR approach, we found that cholesterol was more readily transferred from donor vesicles to cholesterol-deficient supported lipid membranes than from donor supported lipid membranes to cholesterol-deficient vesicles. While the model used in this work was specific for DLPC/ cholesterol membranes, analogous PLSR models can be constructed to permit quantifying the mol % cholesterol in membranes composed of other lipid species (i.e., sphingomyelin). The reported multivariate approach can be used to quantify cholesterol transfer between vesicles and discrete regions on supported lipid membranes or to correlate the mol % cholesterol with the morphology of distinct domains in phase-separated supported lipid membranes. Because this PLSR approach only requires the availability of calibration samples of known relative cholesterol concentration, this approach might enable quantifying the mol % cholesterol in the membranes of individual mammalian cells. For example, populations of cells that systematically differ in cholesterol concentration may be

* Supporting Information

■

■

ASSOCIATED CONTENT

S

Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org.

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS Portions of this work were carried out in the Frederick Seitz Materials Research Laboratory Central Facilities, University of Illinois, which is partially supported by the U.S. Department of Energy (DOE) under Grants DEFG02-07ER46453 and DEFG02-07ER46471. M.L.K. holds a Career Award at the Scientific Interface from the Burroughs Wellcome Fund. The authors thank Christopher Anderton (NIST) for assistance with supported lipid bilayer formation and William P. Hanafin (UIUC) for assistance with the enzymatic assays.

■

REFERENCES

(1) Takahashi, M.; Murate, M.; Fukuda, M.; Sato, S. B.; Ohta, A.; Kobayashi, T. Mol. Biol. Cell 2007, 18, 2667−2677. (2) Revertere, M.; Rentero, C.; de Muga, S. V.; Alvarez-Guaita, A.; Mulay, V.; Cairns, R.; Wood, P.; Monastyrskaya, K.; Pol, A.; Tebar, F.; Blasi, J.; Grewal, T.; Enrich, C. Mol. Biol. Cell 2011, 22, 4108−4123. (3) Norman, L.; Oetama, R.; Dembo, M.; Byfield, F.; Hammer, D.; Levitan, I.; Aranda-Espinoza, H. Cell. Mol. Bioeng. 2010, 3, 151−162. (4) Maxfield, F. R.; Menon, A. K. Curr. Opin. Cell Biol. 2006, 18, 379−385. (5) Lingwood, D.; Simons, K. Science 2010, 327, 46−50. (6) Lev, S. Nat. Rev. Mol. Cell Biol. 2010, 11, 739−750. (7) Liu, J.-P.; Tang, Y.; Zhou, S.; Toh, B. H.; McLean, C.; Li, H. Mol. Cell. Neurosci. 2010, 43, 33−42. (8) Wang, X.; Hinson, E. R.; Cresswell, P. Cell Host Microbe 2007, 2, 96−105. (9) Bosch, M.; Mari, M.; Gross, S. P.; Fernandez-Checa, J. C.; Pol, A. Traffic 2011, 12, 1483−1489. (10) Lange, Y.; Steck, T. L. Prog. Lipid Res. 2008, 47, 319−332.

96

dx.doi.org/10.1021/ac301856z | Anal. Chem. 2013, 85, 91−97

Analytical Chemistry

Article

(11) McLean, L. R.; Phillips, M. C. Biochim. Biophys. Acta 1984, 776, 21−26. (12) Wikstrom, A.; Svedhem, S.; Sivignon, M.; Kasemo, B. J. Phys. Chem. B 2008, 112, 14069−14074. (13) Thomas, P. D.; Poznansky, M. J. Biochem. J. 1988, 254, 155− 160. (14) McLean, L. R.; Phillips, M. C. Biochemistry 1981, 20, 2893− 2900. (15) Anderton, C. R.; Weber, P. K.; Lou, K.; Hutcheon, I. D.; Kraft, M. L. Biochim. Biophys. Acta 2011, 1808, 307−315. (16) Goksu, E. I.; Longo, M. L. Langmuir 2010, 26, 8614−8624. (17) Connell, S.; Smith, D. A. Mol. Membr. Biol. 2006, 23, 17−28. (18) Loura, L. M. S.; Prieto, M. In Methods in Membrane Lipids; Dopico, A. M., Ed.; Humana Press Inc.: Totowa, NJ, 2007; pp 489− 501. (19) Veatch, S. L.; Keller, S. L. Biochim. Biophys. Acta 2005, 1746, 172−185. (20) Shaw, J. E.; Epand, R. F.; Epand, R. M.; Li, Z.; Bittman, R.; Yip, C. M. Biophys. J. 2006, 90, 2170−2178. (21) McQuaw, C. M.; Sostarecz, A. G.; Zheng, L.; Ewing, A. G.; Winograd, N. Langmuir 2005, 21, 807−813. (22) McQuaw, C. M.; Zheng, L.; Ewing, A. G.; Winograd, N. Langmuir 2007, 23, 5645−5650. (23) Zheng, L.; McQuaw, C. M.; Ewing, A. G.; Winograd, N. J. Am. Chem. Soc. 2007, 129, 15730−15731. (24) Baker, M. J.; Zheng, L.; Winograd, N.; Lockyer, N. P.; Vickerman, J. C. Langmuir 2008, 24, 11803−11810. (25) Piehowski, P. D.; Carado, A. J.; Kurczy, M. E.; Ostrowski, S. G.; Heien, M. L.; Winograd, N.; Ewing, A. G. Anal. Chem. 2008, 80, 8662−8667. (26) Veatch, S. L.; Polozov, I. V.; Gawrisch, K.; Keller, S. L. Biophys. J. 2004, 86, 2910−2922. (27) Lindblom, G.; Oradd, G. J. Dispersion Sci. Technol. 2007, 28, 55−61. (28) Williams, P. J. Biol. 2006, 5, 18. (29) Kraft, M. L.; Fishel, S. F.; Galli Marxer, C.; Weber, P. K.; Hutcheon, I. D.; Boxer, S. G. Appl. Surf. Sci. 2006, 252, 6950−6956. (30) Kraft, M. L.; Weber, P. K.; Longo, M. L.; Hutcheon, I. D.; Boxer, S. G. Science 2006, 313, 1948−1951. (31) Lanekoff, I.; Sjovall, P.; Ewing, A. G. Anal. Chem. 2011, 83, 5337−5343. (32) Ostrowski, S. G.; Kurczy, M. E.; Roddy, T. P.; Winograd, N.; Ewing, A. G. Anal. Chem. 2007, 79, 3554−3560. (33) Wagner, M. S.; Shen, M.; Horbett, T. A.; Castner, D. G. J. Biomed. Mater. Res. 2003, 64A, 1−11. (34) Wagner, M. S.; Shen, M.; Horbett, T. A.; Castner, D. G. Appl. Surf. Sci. 2003, 203−204, 704−709. (35) Vanden Eynde, X.; Bertrand, P. Appl. Surf. Sci. 1999, 141, 1−20. (36) Vanden Eynde, X.; Bertrand, P. Surf. Interface Anal. 1997, 25, 878−888. (37) Anderton, C. R.; Vaezian, B.; Lou, K.; Frisz, J. F.; Kraft, M. L. Surf. Interface Anal. 2012, 44, 322−333. (38) Vaezian, B.; Anderton, C. R.; Kraft, M. L. Anal. Chem. 2010, 82, 10006−10014. (39) Biesinger, M. C.; Paepegaey, P.-Y.; McIntyre, N. S.; Harbottle, R. R.; Petersen, N. O. Anal. Chem. 2002, 74, 5711−5716. (40) Biesinger, M. C.; Miller, D. J.; Harbottle, R. R.; Possmayer, F.; McIntyre, N. S.; Petersen, N. O. Appl. Surf. Sci. 2006, 252, 6957−6965. (41) Wold, S.; Sjöström, M.; Eriksson, L. Chemom. Intell. Lab. Syst. 2001, 58, 109−130. (42) Kulp, K. S.; Berman, E. S. F.; Kinze, M. G.; Shattuck, D. L.; Nelson, E. J.; Wu, L.; Montgomery, J. L.; Felton, J. S.; Wu, K. J. Anal. Chem. 2006, 78, 3651−3658. (43) Oran, U.; Unveren, E.; Wirth, T.; Unger, W. E. S. Appl. Surf. Sci. 2004, 227, 318−324. (44) Geladi, P.; Kowalski, B. R. Anal. Chim. Acta 1986, 185, 1−17. (45) Chong, I.-G.; Jun, C.-H. Chemom. Intell. Lab. Syst. 2005, 78, 103−112.

(46) Musumarra, G.; Barresi, V.; Condorelli, D. F.; Fortuna, C. G.; Scirè, S. J. Chemom. 2004, 18, 125−132. (47) Christie, W. W. In The AOCS Lipid Library [Online]; American Oil Chemists’ Society, 2012. http://lipidlibrary.aocs.org/ms/ms21/ index.htm. Accessed June 20, 2012. (48) Backerz, J. M.; Dawidowicz, E. A. Biochemistry 1981, 20, 3805− 3810. (49) Thomas, P. D.; Poznansky, M. J. Biochem. J. 1988, 251, 55−61. (50) Fugler, L.; Clejan, S.; Bittman, R. J. Biol. Chem. 1985, 250, 4098−4102. (51) Ostrowski, S. G.; Szakal, C.; Kozole, J.; Roddy, T. P.; Xu, J.; Ewing, A. G.; Winograd, N. Anal. Chem. 2005, 77, 6190−6196. (52) Touboul, D.; Kollmer, F.; Niehuis, E.; Brunelle, A.; Laprévote, O. J. Am. Soc. Mass Spectrom. 2005, 16, 1608−1618. (53) Zidovetzki, R.; Levitan, I. Biochim. Biophys. Acta 2007, 1768, 1311−1324. (54) Bezrukov, L.; Blank, P. S.; Polozov, I. V.; Zimmerberg, J. Anal. Biochem. 2009, 394, 171−176.

97

dx.doi.org/10.1021/ac301856z | Anal. Chem. 2013, 85, 91−97