Multidimensional profiling of plasma lipoproteins by size exclusion chromatography followed by reverse-phase protein arrays.

The composition of lipoproteins and the association of proteins with various particles are of much interest in the context of cardiovascular disease. Here, we describe a technique for the multidimensional analysis of lipoproteins and their associated apolipoproteins. Plasma is separated by size exclusion chromatography (SEC), and fractions are analyzed by reverse-phase arrays. SEC fractions are spotted on nitrocellulose slides and incubated with different antibodies against individual apolipoproteins or antibodies against various apolipoproteins. In this way, tens of analytes can be measured simultaneously in 100 μl of plasma from a single SEC separation. This methodology is particularly suited to simultaneous analysis of multiple proteins that may change their distribution to lipoproteins or alter their conformation, depending on factors that influence circulating lipoprotein size or composition. We observed changes in the distribution of exchangeable apolipoproteins following addition of recombinant apolipoproteins or interaction with exogenous compounds. While the cholesteryl ester transfer protein (CETP)-dependent formation of pre-β-HDL was inhibited by the CETP inhibitors torcetrapib and anacetrapib, it was not reduced by the CETP modulator dalcetrapib. This finding was elucidated using this technique.

( 2-6 ) components. Apolipoproteins are of particular importance, as demonstrated for apoB, the major apolipoprotein of LDL ( 7 ). The proteins in HDL constitute more than half of its mass and include those with well-characterized functions (apoA-I, apoA-II, apoE, apoC), enzymes (LCAT, paraoxonase 1), and also numerous but less-abundant apolipoproteins with poorly characterized functions ( 8,9 ). In addition, most are considered exchangeable apolipoproteins that can distribute to lipoproteins of different densities and might have different functions or might indicate different health status. For example, apoE is mainly associated with large HDL particles in the general population, but is also associated with VLDL and LDL particles in hyperlipidemia and dyslipidemia ( 10 ).
Other examples whereby protein function may be infl uenced by its association with a particular lipoprotein include phospholipid transfer protein (PLTP), which has been reported to be active when associated with HDL but inactive when associated with LDL or VLDL ( 11 ). In contrast, lipoprotein-associated phospholipase A 2 (Lp-PLA 2 ) is active when associated with LDL but less active in HDL ( 12 ) and considered a biomarker for cardiovascular disease risk assessment. The implications of these differences on a benefi cial versus detrimental role of these proteins are poorly understood, partly due to lack of suitable techniques with which to study their associations in large sample sets.
Because the highly complex protein composition of plasma poses a challenge to analysis ( 13 ), separation into Abstract The composition of lipoproteins and the association of proteins with various particles are of much interest in the context of cardiovascular disease. Here, we describe a technique for the multidimensional analysis of lipoproteins and their associated apolipoproteins. Plasma is separated by size exclusion chromatography (SEC), and fractions are analyzed by reverse-phase arrays. SEC fractions are spotted on nitrocellulose slides and incubated with different antibodies against individual apolipoproteins or antibodies against various apolipoproteins. In this way, tens of analytes can be measured simultaneously in 100 l of plasma from a single SEC separation. This methodology is particularly suited to simultaneous analysis of multiple proteins that may change their distribution to lipoproteins or alter their conformation, depending on factors that infl uence circulating lipoprotein size or composition. We observed changes in the distribution of exchangeable apolipoproteins following addition of recombinant apolipoproteins or interaction with exogenous compounds. While the cholesteryl ester transfer protein (CETP)-dependent formation of pre-␤ -HDL was inhibited by the CETP inhibitors torcetrapib and anacetrapib, it was not reduced by the CETP modulator dalcetrapib. This fi nding was elucidated using this technique. Supplementary key words HDL proteomics • protein-protein interaction • protein-lipid interaction • fast protein liquid chromatography • size exclusion chromatography • reverse-phase arrays Lipoprotein particles are complex structures and the subject of several studies in which different functions have been attributed to both lipid ( 1 ) and associated protein fl uorescence assay employing cholesterol esterase, cholesterol oxidase, and peroxidase, together with the fl uorogenic substrate p -hydroxyphenylacetic acid.

Preparation of reverse-phase arrays
Fractions from SEC separation were thawed, and 50 µl of each fraction was transferred to a 384-well plate, the "spotting plate" (Eppendorf low-protein bind 200 µl) using an automated pipetting robot (Sephyr and Twister, Caliper Life Sciences; Hopkinton, MA). Fractions were spotted in 0.4 nl aliquots onto nitrocellulose membrane-coated microscope slides (Whatman/ GE Healthcare FastSlides or Schott Nexterion white nitrocellulose slides) in a predefi ned pattern using a contact-free inkjet dispenser with a piezo-actuated nozzle (Nanoplotter 2.1E, GeSiM, Grosserkmannsdorf, Germany). Four concentrations of each fraction were generated by applying up to four drops on one location, referred to as a "spot." Arrays held a maximum of 4,608 spots, equivalent to 48 fractions from 24 samples spotted at four concentrations. Up to 60 replicate slide copies were produced from each fraction set. Each slide copy was incubated with a different primary antibody raised against a particular protein and a corresponding fl uorescently labeled secondary antibody. To reduce the infl uence of slide variation, all fractions from one individual (i.e., at different time points or doses) were combined on one slide. Data relating to each slide and associated fraction set were identifi ed using a barcode system and collated.

Analysis of reverse-phase arrays
Reagents. Sterile fi ltered assay buffer [0.1% PBS, Tween 20, 5.0% BSA, 0.02% sodium azide dissolved in deionized water (Millipore MilliQ-Advantage)] was used for blocking, dilution of antibodies, and washing. ApoA-I was detected using an in-house mouse MAb directly conjugated to Alexa647. All other antibodies and detection reagents were from commercial sources and were used as recommended by the supplier. Reagent suppliers, dilution factors, fi nal antibody concentrations, and incubation times of primary and secondary antibodies can be found in Table 1 and Table 2 .
Procedure: incubation with antibodies. Slides were blocked in assay buffer for 5 min, rinsed for 10 s with deionized water, and blow-dried with a nitrogen jet. Slides were stored dry until use. Each slide was transferred to a polypropylene incubation tube (Glass Hybridization Chamber, Clontech; Mountain View, CA) and fi lled with 1.8 ml of primary antibody solution appropriately diluted in assay buffer. After incubation at room temperature for 2 h for MAbs and 16 h for polyclonal antibodies (PAbs), slides were washed with assay buffer and incubated with an Alexa647conjugated secondary antibody (except for apoA-I, where an Alexa647-conjugated primary antibody was used), or Alexa647conjugated streptavidin for 12 min. Slides were washed with assay buffer, then for 10 s with deionized water, and again blow-dried with nitrogen. Alternatively, the slides were automatically processed using the DiscoveryXT system (Ventana Medical Systems; Tucson, AZ), employing the same conditions.
Fluorescence detection and data analysis. Slides were scanned in a GenePix 4200AL fl uorescence microarray scanner (Molecular Devices; Sunnyvale, CA) with appropriate fi lter settings for Alexa647 dye. Each scan yielded a TIF image, with each spot representing lipoprotein subpopulations may initially reduce this complexity. Notably, the separation of plasma lipoprotein particles by ultracentrifugation or isotachophoresis may change lipoprotein composition by detaching loosely bound proteins ( 5,14 ). Gel electrophoresis, and 2D gels in particular, has a nonparalleled separation power, but gel preparation ( 15,16 ), protein extraction, and characterization are laborious. Protein association is well preserved by using size exclusion chromatography (SEC), a method that also allows reasonably high throughput and is applicable to large clinical studies. The similarity of separation by SEC to ultracentrifugation has been shown, and rigorous assessments of SEC have been made ( 17,18 ). Reverse-phase arrays (RPAs) are ideally suited for rapid analysis of tens of analytes from small volumes of fractionated plasma. Analyte-containing samples are immobilized on a solid support and assessed using appropriate antibodies, i.e., in contrast to forward-phase arrays where the antibody is immobilized. Use of RPAs is well established for characterization of proteins in cell and tissue lysates ( 19 ).

Blood donors and SEC
Human plasma samples were provided by healthy volunteers from a clinical study conducted in accordance with the Declaration of Helsinki and approved by independent ethics committees. All participants provided written informed consent. A general schematic of the technique described is provided in Fig. 1 . In brief, EDTA plasma samples were centrifuged for 4 min at 1,200 rpm (113 g ) and 100 µl of supernatant diluted with 100 µl of SEC solution (0.15 M NaCl, 1 mM EDTA, pH 8.0) prior to being loaded onto the separation system. SEC was carried out using an ÄKTApurifi er 10 (GE Healthcare; Uppsala, Sweden) consisting of a compact separation unit, equipped with an autosampler A900, a fraction collector Frac-950, and a Superose 6 column (10/300GL, 10 × 300-310 mm, 24 ml). The system was controlled by a UNICORN control system version 4.10 (GE Healthcare). The column was equilibrated with 25-30 ml SEC buffer (fl ow rate 500 µl/min) prior to loading 200 µl of prediluted sample and separation at 10°C. Eluting fractions, usually 32 or 48 of 500 l were collected in 96 deep-well plates (Eppendorf Protein Low-Bind; Hamburg, Germany). After fractionation, samples were frozen at Ϫ 20°C before further processing. Selected fractions obtained by Superose-6 SEC were further investigated by onedimensional agarose gel electrophoresis (Paragon LIPOgel kit, Beckman Coulter, Inc.). Proteins were transferred from wet gels to nitrocellulose membranes by capillary force for 2 h at room temperature under slight pressure and assayed for apoA-I using an anti-human monoclonal antibody (MAb) clone 3F10 (Intracel, #RP-062) conjugated to HRP using EZ-Link™ Plus Activated (Pierce, #31487) according to the manufacturer's instructions.
Total cholesterol concentration was determined in each fraction as previously described ( 20 ) using a microenzymatic  one concentration of each fraction. Spot intensity was analyzed by GenePix Pro, v 6.1. To determine protein concentration within individual fractions, the intensities of the four spots of different concentrations were linearly fi tted. The square root of the sum of squared residuals was taken as a quality parameter (quasi-SD) of linearity and scatter of the intensities of the four concentrations. The center of the linear fi t was taken as the relative fl uorescence intensity (RFI), which represented the signal intensity of each fraction.  (3 µM), and torcetrapib or anacetrapib (1 µM). All compounds studied were synthesized using standard procedures. rhCETP was expressed by a cell line described by Weinberg et al. ( 21 ), kindly provided by Professor Alan Tall (Columbia University) and purifi ed by hydrophobic interaction chromatography and SEC [modifi ed from Ohnishi, Yokoyama, and Yamamoto (22)].

ELISA for the detection of CETP
NUNC maxisorp plates were coated with anti-CETP capture antibody clone 6/2 (5 g/ml) in a total volume of 100 l/well using washing buffer (PBS, 0.05% Tween-20) overnight at 4°C in a humid box. After washing with washing buffer, wells were blocked with blocking buffer [assay buffer (50 mM Tris, 140 mM NaCl, 5 mM EDTA, 0.05% NONidet P40, 0.25% gelatin, all adjusted to pH 7.4) plus 1% BSA] for 1 h at 37°C. Fifty microliters of each fraction was applied to each well and incubated for 1 h at 37°C. Wells were rinsed and incubated with 50 l HRP-conjugated antibody clone 6/17 (0.5 g/ml) for 1 h at room temperature. After thorough rinsing of wells, the plates were incubated for 5 min at room temperature with 100 l/well of substrate mix containing 2 M citric acid (pH 3.95), 0.95 mM tetramethylbenzi-dine, and 0.006% H 2 O 2 . The reaction was stopped by addition of 100 l 0.5 M H 2 SO 4 /well. Optical density was determined at a wavelength of 450 nm.

Distribution of proteins is consistent with previous fi ndings
The spotting process, using SEC-derived fractions, allowed several dozen identical slides to be produced automatically. This approach to the analysis of spotted samples with antibodies is similar to a dot blot. However, the RPA approach is dramatically miniaturized. Using the classic macroscopic dot blot technique, typically the entire fraction of 500 µl is used to prepare only one blot; here, less than 20 µl of each fraction is used while producing up to 60 identical "blot copies" in one spotting run. Each copy can then be incubated with a different antibody ( Fig. 1 ). The turnover of the technology is fast and requires little "hands-on" time, inasmuch as many steps are automated. Typical turnover for a set of 8-12 samples includes fractionation (Day 1), spotting (Day 2), antibody incubation (Day 3), and readout (Day 4). Machine usage could be scheduled to provide a maximum throughput of 40 plasma samples per week; this produced up to 1,800 protein chromatograms from 45 different assays in one particular study (data not shown).
With this technology, as expected, the major apoA-I peak coincided with the HDL-cholesterol (HDL-C) peak in samples from untreated healthy volunteers, with no detection of apoA-I in other particle populations. The apoB peak coincided with the cholesterol peaks representing VLDL and LDL, also as expected ( 23 ). ApoE was detected in particles larger than those comprising the major HDL-C peak but smaller than particles comprising the major LDL-cholesterol (LDL-C) peak ( Fig. 1B-G ), consistent with previous fi ndings ( 24 ). This illustrated and a much better signal-to-noise ratio than planar-modifi ed surfaces (epoxy-or dodecyl-functionalized, data not shown).
Assay sensitivity SEC separation was highly reproducible. In absolute terms, as assessed by absorption at 280 nm, retention time measures showed a coeffi cient of variation (CV) between 0.2% and 0.3%. Expressed in relative terms of the fraction volume, this was between 3% for early fractions and 12% for later-eluting fractions. Peak amplitude had a CV between 2.3% and 9.5%.The overall variation of the RPA technique, as assessed by incubation of 10 replicate slides, had median CVs of 5.4-10.5%, depending on the antibody used. Lower limits of detection (LLODs) and assay linearity were determined for those analytes in which purifi ed proteins were available by spotting dilution series of the respective proteins. LLODs were 310 ng/ml for apoA-I and 6 ng/ml for CETP, levels signifi cantly below published concentrations in full plasma.

Assessment of antibody specifi city
For each analyte, we tested several different commercially available or in-house-produced antibodies on sets of SEC fractions representing different conditions. Each antibody was tested at different concentrations and with different incubation times. The position of analyte peaks from the resulting profi les were compared with published reports of associations of analyte to lipoprotein particle subfractions (e.g., 25 ). While the signal generally increases with rising antibody concentration, nonspecifi c binding also increases. Nonspecifi c binding was assessed from signals in fractions containing only running buffer and from fractions in which no substantial occurrence of an antigen was expected, i.e., apoA-I in LDL or apoB in HDL. Antibody concentrations were optimized to yield maximum signals in fractions in which an antigen was expected and minimal nonspecifi c binding. To further assess nonspecifi c binding to other proteins the ability of this technology to identify and describe lipoprotein particles by means other than their cholesterol content alone.
The active form of PLTP, detected with a PAb, was associated with HDL, whereas the inactive form, detected in addition to the active form by an MAb, was also associated with VLDL and LDL, as per previous observations ( 11 ). The technology is therefore also capable of distinguishing different forms or conformations of a given protein.

Sensitivity and surface saturation effect
Inherent to separation by SEC, analytes will, in general, be more diluted in the fractions than in total plasma, which could compromise sensitivity toward low-abundant analytes. To overcome this, during spotting, we deposited multiple drops of each fraction onto nitrocellulose membrane-coated slides, thus increasing the concentration of analytes per unit area. However, full saturation of the surface area available had to be avoided to maintain measurements within the linear range. The deposition of 1, 2, 3, and 4 drops of each fraction, each at a given spot, facilitated the concentration of low-level analytes and the assessment of linearity during the analysis. The linearity was inspected by plots of RFI for the four concentrations ( Fig. 2A ). If saturation was approached, the four concentrations would display high RFI values, but a slope approaching zero. Hence, mean RFI versus the slope of four concentrations of each fraction were routinely plotted as a means of quickly controlling linear surface coverage in all experiments ( Fig. 2B ). In this way, the entire process was optimized to completely avoid surface saturation; in particular, the amount of plasma loaded onto the column (ideal: 200 l, 1:1-diluted EDTA-plasma). Further dilution of fractions prior to spotting was explored also, but undiluted fractions proved more effective. We also tested different microarray surface coatings; slides coated with a nitrocellulose membrane yielded a dynamic range of more than three orders of magnitude in protein concentration The incubation of human plasma in vitro with rhCETP and detection with different anti-CETP antibodies by RPA revealed that both endogeneous CETP and added rhCETP eluted at long retention times. Under basal conditions, some CETP eluted with other lipoproteins, but also at longer retention times, i.e., under conditions in which the salt eluted from the column ( Fig. 3A , fractions 36-40). Added rhCETP increased only the peak in relatively late fractions, also found during purifi cation of rhCETP; the protein from these late fractions remained functional. Unexpectedly long retention times for CETP have been reported previously, albeit referred to as lipid transfer protein at that time ( 22 ).
In the presence of the CETP inhibitor torcetrapib, CETP was not detected in late fractions, but coeluted with LDL and HDL, indicating a stronger association to lipoproteins when inhibitor is present, a fi nding confi rmed by ELISA employing different antibodies for the capture and detection of CETP. The redistribution of CETP upon inhibition has been described previously ( 26 ) providing further validation of the technology.

CETP-induced formation of pre-␤ -HDL particles
Most striking was the observed formation of an additional apoA-I peak with increasing concentrations of rhCETP, which appeared in fraction 29, i.e., following the principal HDL-C peak (fractions 23 to 27). All of the fractions shown in Fig. 4 were also analyzed for other apolipoproteins. Of those lipoproteins detected in fraction 29, and specifi c binding to the analyte, we prepared slides with concentration series of purifi ed plasma proteins and purifi ed apolipoproteins and incubated all used antibodies to those slides. Particular fi ndings, such as lipoprotein redistribution or change in conformation of some apolipoproteins, were typically confi rmed by alternative methods, e.g., Western blot or sandwich immunoassay of selected fractions. In most cases, the alternative techniques utilized antibodies different from those used for RPA, thus avoiding potential bias introduced by using the same antibody.
In addition to thorough evaluation of primary antibodies, it was crucial to assess the cross-reactivity of detection reagents. Streptavidin binding to fractions was negligible, and biotinylated primary antibodies were therefore preferred. Inasmuch as such conjugates are not always available, and biotinylation can change the specifi city of primary antibodies, we carefully evaluated cross-reactivity of secondary antibodies to the fractions. Optimal results were obtained with commercially available secondary antibodies, which were cross-absorbed against IgGs from several different species (so-called highly cross-absorbed).

Association of CETP with lipoproteins is increased by a CETP inhibitor
The proper activity and function of rhCETP was confi rmed by a cholesterol transfer assay and decrease of HDL-C with increase of LDL-C in cholesterol profi les (data not shown). presence of 3 µM of the CETP modulator dalcetrapib, CETP-dependent formation of pre-␤ particles was conserved ( Fig. 4B, E ).

DISCUSSION
SEC followed by RPA (SEC/RPA) allows the detection of many lipoproteins in many fractionated samples. With only a low initial sample volume required (100 µl plasma), data on multiple lipoprotein parameters may be generated from a single sample set. This technology is capable of detecting differences in lipoprotein composition or changes only apoA-I levels were changed by rhCETP ( Fig. 5 ). These small particles, which contained almost exclusively apoA-I, were identifi ed using agarose gel electrophoresis followed by Western blotting ( 27 ) as pre-␤ -HDL particles, a nascent form of HDL with a low cholesterol content and involved in reverse cholesterol transport (RCT). The activity of CETP toward increasing apoA-I in pre-␤ -HDL particles has been previously described ( 28 ). When 1 µM of either CETP inhibitor torcetrapib or anacetrapib was coincubated with plasma and increasing concentrations of rhCETP, CETP-induced formation of pre-␤ particles was completely abolished ( Fig. 4C, D ). In contrast, in the detect low-abundant proteins. Other techniques use pooled fractions, which further limits resolution of subparticles and does not seem to detect all proteins, e.g., CETP ( 25 ).
SEC/RPA allowed us to distinguish between CETP in a lipoprotein-bound state and an unbound state with singlefraction resolution, the latter eluting at unusually long retention times, i.e., after all soluble plasma proteins, and as previously reported ( 22,26 ). However, we suspected that this specifi c behavior of CETP in SEC may have been overlooked in some studies, particularly if fractionation was stopped prematurely before the elution of unbound CETP molecules. In addition, SEC/RPA revealed that CETP not only transfers cholesteryl esters from HDL to LDL but also remodels ␣ -HDL and generates pre-␤ -HDL. Signifi cantly, pre-␤ -HDL particles are the primary acceptor for excess cholesterol following effl ux from macrophages via the ABCA1 transporter, the fi rst step in the RCT pathway, and are believed to play an important role in the reduction of atherosclerotic plaque burden. These two different functions of CETP have been found by SEC/RPA to be differentially impacted by three compounds that decrease CETP activity to different extents: torcetrapib (IC 50 0.012 µM), anacetrapib (IC 50 0.016 µM), and dalcetrapib (IC 50 0.425 µM). Whereas torcetrapib and anacetrapib inhibit both functions of CETP, dalcetrapib modulates CETP activity; i.e., it neither inhibits the transfer function of CETP within HDL particles nor affects the remodelling of ␣ -HDL into pre-␤ -HDL ( 27 ).
The technique may also be used to screen for new analytical antibodies, possibly recognizing certain epitopes or selectivity for particular protein associations. Screening antibodies against a large fraction set, with conditions carefully selected to represent true positive, true negative, and intermediate conditions, improves the likelihood of fi nding selective antibodies.
SEC/RPA technology may help to develop understanding of proteins whose activity status depends on their associated lipoproteins, as demonstrated in this study for PLTP. Other proteins, such as apoF, a putative CETP modulator, apoM, or Lp-PLA 2 , are thought to have different functionalities, depending on their lipoprotein associations. It will be interesting to study the different conformations and their differential association with lipoproteins by SEC/RPA. Another application may include the screening of plasma from patient populations to identify reliable biomarkers indicative of different patient conditions, differing susceptibility to treatment, or for the identifi cation of new risk factors.
In conclusion, this technology adds to existing knowledge ( 30 ) in further showing how multidimensional testing can be used to illustrate the breadth of biological diversity in lipids and lipoproteins while keeping the number of individual sample handling steps low.
The authors thank Walter Schilling for the initiative and indispensible support in the early phase of this project. Christof Fattinger, Michael Hennig, and Markus Böhringer are acknowl edged for their support during all phases of this project.
in the association of proteins with particular lipoprotein particles; this in turn facilitates the study of lipoprotein remodeling in different physiological or pathological conditions, or due to treatment.
The separation of plasma to reduce its complexity is widely used. Fractionation of plasma by SEC offers several advantages over ultracentrifugation or electrophoretic separation, in that it is much faster, provides highly reproducible retention times, maintains the composition of lipoprotein particles, and requires only a small volume of plasma. In addition, full automation with standard equipment is well-established ( 1 ). Although we used 100 µl plasma, separation of mouse plasma from volumes as low as 10 µl has been reported ( 29 ). Fraction sizes from such a separation would still be suffi cient for RPA, thus offering a further advantage where sample volume is limited.
Further optimization of the SEC/RPA technique for the analysis of apolipoproteins is desirable, particularly to overcome limitations in the separation characteristics of the employed Superose 6 column. One major drawback is that soluble plasma proteins coelute mostly with HDL, such that distinction between a lipoprotein-associated protein and a soluble protein is not possible. Additionally, different species of HDL are not well separated from each other. This could be overcome by using different column materials or different separation methods. We are currently evaluating silica-based separation materials (e.g., TSKgel columns, Tosoh Bioscience; Stuttgart, Germany) and separation by asymmetric-fl ow fi eld-fl ow-fractionation (AF4). The advantage of AF4 is a fully tuneable separation characteristic based simply on employed fl ow rate. Also, the interactions between proteins and column material, which result in a separation not purely determined by hydrodynamic radius, are of lesser concern with AF4. If other separation characteristics were desired, for example to improve the separation of different VLDL and LDL subparticles, another column material or other AF4 settings could be employed to provide fractions available for analysis by RPA.
Particular strengths of the antibody-based detection by RPA are the high throughput and potential sensitivity to