Tris, ammonium hydroxide, sodium chloride, sodium bicarbonate, HPLC grade water and acetonitrile with 0.1% formic acid were purchased from Fisher Scientific (Fair Lawn, NJ, U.S.A.). Zwittergent 3–12 was purchased from EMD Millipore (Billerica, MA, U.S.A.). Calcium chloride and [Glu1]-Fibrinopeptide were purchased from Sigma-Aldrich (St. Louis, MO, U.S.A.). The IMER column, digestion and wash buffers were obtained from Perfinity Biosciences (West Lafayette, IN, U.S.A.). Protein standard materials, apoA-II, apoC-I, apoC-II and apoC-III were obtained from Academy Bio-Medical (Houston, Texas, U.S.A.). ApoA-IV and apoE were obtained from Novoprotein Scientific (Summit, NJ, U.S.A.). All protein standards were ≥98% pure, based on SDS-PAGE performed by the manufacturer, and were used without further purification. The concentrations of the protein stock solutions prepared from these solid materials were confirmed by amino acid analysis performed by Midwest Bio-Tech. (Fishers, IN, U.S.A.).

Five units of frozen human serum or EDTA plasma from deidentified individuals, used for preparation of calibrators and quality controls (QC), were purchased from Interstate Blood Bank (Memphis, TN, U.S.A.). Frozen reference sera used for evaluating method bias, SP1–01 (apoA-1) and SP3–08 (apoB-100), were provided by the Lipid Standardization Program at the Centers for Disease Control and Prevention (Atlanta, GA). Frozen human serum, used for cross-method comparison, was collected from 25 deidentified individuals at Solomon Park Laboratories (Kirkland, WA, USA). All specimens were viral tested. Collection and storage procedures met guidelines described in CLSI document C37-A. All samples were stored frozen at −70 °C and thawed in refrigerator overnight before analysis.

Labeled synthetic peptides (Table 1) were purchased from Midwest Bio-Tech. (Fishers, IN, U.S.A.); ¹³C and ¹⁵N isotopic enrichment was stated as ≥99%. Chemical purity for all peptides was confirmed in-house by LC-UV. Due to the multitude of analytes and associated cost, labeling occurred predominantly at internal residues, instead of the N-terminal Lys or Arg as is convention. However, only transitions containing the mass shifted residues were selected as quantification and/or qualifier ions. The individual IS peptide stock solutions were prepared with 0.1% formic acid to a nominal concentration of (50 pmol/μL), distributed into 400 μL aliquots and stored at −70 °C. A frozen aliquot of each IS peptide stock solution was sent to Midwest Biotech for amino acid analysis. Pooled IS solutions were prepared using 0.1% formic acid with 1 pmol/μL [Glu1]-Fibrinopeptide and contained between 0.1 and 2 pmol/μL of each labeled peptide to approximately match the native cleavage product peak intensities of 1:100 diluted sera.

Serum was chosen to prepare both as the calibrator pool and as QC to match the native matrix of the 25 unknown serum samples to be analyzed. The calibrator pool was prepared from an equal mix of sera collections from four individuals. This pool was vacuum filtered and distributed into 100, 75, 50, 20, and 10 μL aliquots into 2 mL storage vials, using a positive displacement repeater pipette and weighing the mass of each aliquot on an analytical balance for enhanced accuracy and precision. All weighed aliquots were stored at −70 °C until use. A weekly calibration series was prepared by addition of 1500 μL buffer (10 mM NaHCO₃, 150 mM NaCl, pH 7.4) to the stored aliquots giving 16, 21, 31, 61, and 151 fold dilutions. The latter four of these were further diluted 10 fold to obtain 210, 310, 610 and 1510 fold dilutions, giving a complete 9 point calibrator series. After dilution, the calibrator series was stored at 4⁰C and discarded after 5 days. Sera collections from three individuals (without pooling) were distributed into 1-mL aliquots and stored at −70 °C until use as quality controls (QC1, QC2, and QC3).

Quality controls and unknown serum samples were diluted 100:1 with buffer containing 10 mM NaHCO₃, 150 mM NaCl at pH 7.4. A 100 μL aliquot from each diluted sample was transferred to a 96-well plate. To each well, 50 μL of digest buffer containing 0.45% Zwittergent 3–12 was added. Samples were mixed on a shaker plate at 500 rpm for 5 min, then placed directly into the autosampler at 8 °C for online IMER-LC-MS/MS analysis.

The integrated digestion platform (Perfinity Biosciences) included an autosampler with a 500 μL loop, one quaternary pump, two isocratic pumps, two column oven compartments and four 2-position 10-port valves. The quaternary pump was used for the delivery of the IMER digestion buffer and column regeneration buffer. The two other LC pumps delivered 0.1% FA in water (eluent A) and 0.1% FA in acetonitrile (eluent B) for binary gradient elution.

The sample injection sequence started by drawing 5 μL of IS peptide solution from a reagent vial, followed by drawing 50 μL diluted serum from a 96-well plate. Assuming that a typical human serum sample contains 60–80 g/L of protein, with injection of 50 uL of 1500x–16x diluted calibrator serum pool, the total protein amount injected was between 2 and 250 μg. The IMER (Perfinity trypsin column 2.1 mm × 33 mm) was operated at 50 °C with a 25 μL/min flow rate, giving an injection-plug breakthrough time of 3–4 min (including pre- and post-column tubing). An additional 4 min was allotted to purge the IMER column and connection lines. The unreacted proteins, native cleavage products and internal standards were carried together directly to the trapping column (Halo^® C18 4.6 mm × 5 mm, 2.7 μm guard column). After the 8 min digestion period, the trapping column was idled between closed lines while the IMER was regenerated. During the consecutive sample run, native cleavage products and IS peptides from the previous run were eluted from the trapping column to the analytical column (Halo^® C18 core shell HPLC column, 2.1 mm × 100 mm, 2.7 μm), while the second trapping column was used for trapping the newly injected sample. The analytical separation was performed at 50 °C with a flow rate of 350 μL/min over a stepwise gradient ranging from 3% to 95% acetonitrile containing 0.1% formic acid (Figs. S2 and S3).

The mass spectrometer was a 6500 QTRAP^® (Sciex, Foster City, CA, U.S.A.) operated with a heated electrospray ionization probe in positive ion mode using the following settings: ion spray voltage 5500 V, ion source heater temperature 450 °C, source gas 50 psi, curtain gas 35 psi. The native and the isotopically labeled IS peptide chromatograms were acquired by multiple reaction monitoring (MRM) with unit mass resolution, in scheduled 60 s acquisition windows with a 0.65 s target scan time. Full MRM mass transition tables are summarized in Table S1.

For apoA-I and apoB-100 the serum samples were also analyzed by an off-line tryptic digestion coupled LC-MS/MS method, validated in our laboratory (Supplementary information), using native and isotopically labeled peptide calibrators. The concentrations of the peptide stock solutions used for the off-line digestion method calibration were determined by amino acid analysis in our laboratory [28]. For total cholesterol and total triglyceride measurements, we used a UPLC-MS/MS method developed in our laboratory (Supplementary information).

Molecular ion masses, product ion masses, and fragmentation potentials were generated using Skyline software version 3.1.0.7382. The raw mass spectrometry data was processed with SCIEX MultiQuant software version 3.0.2. For each peptide precursor ion, two product ions were collected; the product ion with the better signal-to-noise and reproducibility was selected for quantification, while the other was used to calculate product ion ratio for quality control (see Table S1 in Supplementary information). From native peptide vs. corresponding isotope labeled IS analog peak areas, response ratios were calculated. The linear regression calibration curves were calculated with 1/x weighting. In case of multiple target peptides per protein, the reported protein concentrations were calculated as the mean of the concentrations determined based on the individual peptide quantitation ions (Table 1). To note, we routinely use the average of multiple peptides to calculate the reported protein concentration in our laboratory. Once good agreement between two (or more) peptides has been demonstrated, calculation of the peptide-average by protein reduces the possible variation in individual peptide cleavage rates due to differences in protein tertiary structure from sample to sample. Furthermore, by monitoring the ratio of the individual peptide based concentrations helps to identify possible sample specific interferences. A more detailed description of the data processing work flow is presented in Supplementary information.

Because of the heterogeneous nature of the enzyme reaction in the IMER, the amount of peptide collected on the trapping column, as measured by the detected LC-MS/MS signal intensity, is limited by the protein diffusion across the stagnant liquid layer around the porous particles. The driving force of this diffusion, and the digestion rate, is the protein concentration difference on the two sides of the stagnant liquid layer. Because of the high excess of immobilized trypsin/surface area on the porous particles, the protein concentration at the particle surface is always lower than in the bulk mobile phase-side of the stagnant liquid layer. Therefore, the cleavage rate is limited by the concentration of the protein in the injection “plug”, which is determined by the concentration of the protein in the injection solution.

The process of peptide screening is important because the on-line digestion is always incomplete, but it needs to be fast to maximize method sensitivity. The rapid cleavage of the target peptides needs to be verified and needs to have a structurally sensible rationalization. Assuming that inside the pores the individual target peptides are cleaved with their unique and independent reaction rates, the pattern of cleavage efficacies along the protein sequence is a fingerprint of the protein structure and its accessibility to trypsin cleavage.

We performed a series of experiments digesting either purified proteins or diluted serum, where we monitored all possible trypsin cleavage products. For each apolipoprotein, an MRM acquisition method was generated by in silico digestion using Skyline software (https://skyline.gs.washington.edu). Dilution series of purified proteins or pooled serum were prepared in a 10–1000 nmol/L concentration range. For each selected transition, LC-MS/MS peak area count vs. protein concentration curves were calculated. Peptides with the highest regression slopes, lowest intercept/slope ratios and R² values ≥ 0.95 were chosen for quantification. The relative slopes for apoA-I, A-II, A-IV, C-I, C-II, C-III and E are represented by order of peptide sequence position in Fig. 2, where the marker sizes are proportional to the regression slope, relative to the highest value for each protein. In Fig. 3, apoB-100 peptides are shown in a similar format while grouped by structural domains. We observed that 8–12 residue peptides had significantly steeper regression slopes if they contained amino acid side chains with loop propensity (P or G). Consequently, most of the peptides that passed the screening process were from solvent exposed loop regions connecting relatively mobile amphipathic helices, as predicted by structural studies [7–10,29–32].

For apoA-II, C-I, C-II and C-III, because of their bihelical structure and single prominent loop region [9], only 3 or 4 peptides were detected. The larger proteins, digested in their purified forms, apoA-I, A-IV and E showed numerous peptides to choose from. However, when their lipid bound forms were digested from serum, relatively few peptides passed the desired concentration regression criteria. On published apoA-I structures, our selected peptides are located on helix 6 (THLAPYSDER-185), on helix 8 (ATEHLSTSEK-220) and on helix 9 (AKPALEDR-231). These regions have reduced lipid binding propensity on both discoidal and spherical HDL, consistent with more accessibility for proteolytic cleavage by trypsin [33]. The apoA-IV peptides that passed our screening criteria (LEPYADQLR-135 and LTPYADEFK-201) are located at two mobile loop regions that are involved in the monomer-dimer dynamics of apoA-IV [34]. Of the selected apoE peptides, SELEEQLTPAEETR-94 is located next to the LDL receptor binding N-terminal side of apoE, while LQAEAFAR-269 is located close to the C-terminal, consistent with higher surface exposure for trypsin cleavage [30,35].

The current consensus structure of LDL binding apoB-100 contains five domains: βα₁, β₁, α₂, β₂, and α₃ [36]. The βα₁ domain is the least attached to the lipid particle surface. We selected two peptides for quantification from this domain (TGISPLALK-220 and AAIQAR-526). A unique location on the apoB-100 structure is the beginning of the α₃ domain at residue 4050, the following 480 residues turn back toward the β₂ domain to form a functionally critical H-bond between residues 3500 and 4369. The matrix exposed sharp elbow of this turn contains ATGVLYDYNK-4077, which was selected as a third peptide for quantification by us, as by others [20].

The final list of peptides selected for quantitative analysis is summarized in Table 1. Because of the short digestion times, the probability for secondary reactions after trypsin cleavage was minimal. Even peptides with tryptophan (W) residues, known for their susceptibility for oxidation still gave acceptable method validation results (apoC-I EWFSETFQK, apoC-II ESLSSYWESAK and apoC-III GWVTDGFSSK).

Finding more than one suitable quantitative peptide for apoA-II was the most challenging. Only one peptide (EQLTPLIK) proved suitable for quantification. Two other peptides were detected in sequential proximity at 54–77: SPELQAEAK, and SYFEK. However, the product ions for both peptides showed relatively low signal-to noise, while SPELQAEAK exhibited poor native/labeled area ratio reproducibility.

With the original commercial set-up of one trapping column, each sample run required 25 min. During the first half of the injection run, the digestion and trypsin column wash period, the LC-MS/MS system was idle, and during the second half of the run, the LC-MS/MS acquisition the trypsin column was idle. The addition of an extra ten-port valve with 2 trapping columns allowed simultaneous digestion and LC-MS/MS analysis. Each consecutive sample was digested and trapped on one trapping column while the previously digested and trapped sample was eluted from the other trapping column and analyzed by LC-MS/MS. Therefore, except for the first and last sample of the batch, the LC-MS/MS sample throughput is only 12.5 min per sample, thus allowing for the same throughput as an off-line in-solution digestion LC-MS/MS run. We provide the 4 ten-port valve diagram in Supporting information (Fig. S1).

Because protein concentration range, injection volume, digestion buffer flow rate, detergent type, and detergent concentration are strongly interdependent method variables, the optimization was performed by application of a design of experiment (DoE) approach using JMP (SAS) software functions. With calculation of multivariate response surface models for each peptide, the optimization of the IMER conditions was accomplished with the following considerations. Of the detergents tested, Zwittergent 3–12 accelerated the digestion rate the most. The optimal Zwittergent concentration range was found at 0.15–0.2%, near the critical micelle concentration of Zwittergent. The optimal range of flow rate was 25–30 μL/min; minimizing the negative effect on LC-MS/MS signal intensity because of decreased reaction time, but maximizing the efficacy of product removal from the IMER. The optimal range of injection volume was 25–50 μL, by maximizing LC-MS/MS signal intensity at low protein concentration while avoiding the saturation of the IMER at high protein concentration. With these optimal conditions the carryover from the highest calibrator into a blank sample was <0.5% (Fig. S6).

The apolipoprotein concentrations in the calibrator serum pool were determined in triplicate analysis by either protein standard addition or external calibration methodologies. For apoA-II, A-IV, C-I, C-II, C-III, and E, standard addition methodology was used with commercially available purified proteins as secondary reference standards (Fig. S5). The protein reference standard solutions were value assigned by amino acid analysis performed at Midwest Bio-Tech., using NIST (U. S. National Institute of Standards and Technology) certified amino acids as primary reference materials. To examine matrix effects the standard addition experiment was performed with a serum pool, plasma pools and buffer matrix (Fig. S5). The relative differences between the response ratio-concentration regression slopes for the plasma vs. serum pool were <7%, and for plasma or serum pool vs. buffer were <8%. An intriguing exception was apoA-II, where a significant matrix effect was observed both for plasma pool vs. buffer (29%) and for serum pool vs. buffer (22%). We hypothesize that this matrix effect was due to unknown protease activity that seems to be apoA-II sequence specific. This hypothesis was supported by the negative curvature for the apoA-II calibration curves, especially at the 2 highest calibration points, i.e. prepared with the highest calibrator-serum-pool:buffer ratio (Fig. S7). We also observed that the analogous stable isotope labeled IS peptide for ApoA-II significantly degraded within 2–4 h when it was spiked into <50× fold diluted serum. For this reason, the IS mix was not spiked into the samples in the 96-well plates; instead the autosampler was programmed to draw the IS mix into the sampling loop from a separate vial immediately prior to the draw of the sample from the 96-well plate (Fig. 1).

For apoA-I and apoB-100, being non-exchangeable apolipoproteins with significantly different conformations in free and lipid bound forms, the ApoA-I and ApoB-100 content of the working calibrator serum pool was determined using the dilution series of a value assigned serum pool as a tertiary reference standard. This pool was value assigned by off-line tryptic digestion coupled LC-MS/MS in 24 independent runs, using synthetic peptides as secondary reference standards (Supplementary information). The peptide standards were value assigned by amino acid analysis performed in our laboratory, using NIST certified amino acids as primary reference materials.

A new dilution series of the value assigned calibrator serum pool was prepared on 5 consecutive days, and aliquoted in 96-well plates for analysis. At each dilution level, two injections were made from the same well. At the lower limit of quantification (LLOQ), variations in the measured response ratios were due to low signal-to-noise for the low abundance proteins in the serum calibrator. At the upper limit of quantification (ULOQ), digestion efficacy of the high concentration proteins was limited by the amount of trypsin in the IMER, leading to a nonlinear response. Representative calibration curves for each quantifier ion transition are provided in the supplementary (Fig. S7). The LLOQ and ULOQ for each peptide was determined either by applying ±20% and ±15% tolerance limit on accuracy, respectively, otherwise based on the lowest and highest calibrator concentrations (Table 2). For apoA-II, the two highest calibration points were systematically excluded because of the substantial negative deviation from the linear regression slope. Nonetheless, the R² correlation was at least 0.99 for all proteins, except for apoA-II (R² = 0.98).

Serum from three individuals were distributed into small aliquots and were stored frozen to be used as quality control materials (QC1, QC2 and QC3). Data from quintuplicate sample preparation on 5 days was used to calculate intra-assay and inter-assay variations. Protein concentrations were calculated as the average of measurements based on multiple target peptides (Table 3). The daily intra-assay variations were 1.8–4.4% for all proteins (N = 5), while the overall inter-assay variations were 2.5–8% (N = 25).

Thirty samples of human sera (25 individual clinical samples and 5 pools) were analyzed for the full panel of apolipoproteins. Deming regression of calculated concentrations based on apoA-I and apoB-100 peptides are presented in Fig. 4. In spite of their very different positions on the apoB-100 sequence, AAIQALR and TGISPLALIK within the βα₁ domain correlated well with ATGVLYDYVNK on the α₃ domain (slopes of 0.934 and 1.10 and R² ≥ 0.980), as well as with each other (slope 0.969 and R² 0.982). ApoA-I peptides gave excellent agreement (slopes of 0.984 and 1.024 and R² ≥ 0.984) while regression plots for apoA-IV, apoC-II and apoC-III showed <10% difference between peptides (Fig. 5). Some exceptions include four samples in which ApoC-I peptides disagreed >10%, indicating sample specific matrix interferences or unknown biological variations. Bland-Altman analysis plots are presented in Supplementary information (Fig. S8).

25 patient samples with a wide range of lipid profiles were analyzed by both on-line and off-line digestion methods in order to assess the agreement between the two methods in the concentration range of unknown samples. The IMER-LC-MS/MS method showed an average of −4.2% difference for apoA-I and +9.9% difference for apoB-100 relative to the off-line LC-MS/MS method (Fig. 6).

We acquired World Health Organization (WHO) harmonization standards of apoA-I (SP1–01) and apoB-100 (SP3–08) [14]. The quintuplicate analysis using the IMER-LC-MS/MS method showed 1.6% negative bias (2.6%CV, MW: 28 kDa) relative to the assigned value for apoA-I, and 13.1% negative bias (2.2%CV, MW: 515 kDa) for the apoB-100 harmonization standard. The off-line digestion coupled LC-MS/MS method (traceable to NIST certified amino acid calibrators) also had similar bias of 1% for apoA-I and 12% for apoB-100 relative to the WHO harmonization standards. Therefore, the observed bias originates from the independent calibration of our mass spectrometry based methods vs. harmonized immunoassays.

For the other apolipoproteins, harmonization standards were not available. Measurements in the 25 unknown samples were instead compared with the range of published values by various other authors and methods (Fig. 7) [18,20,37–40]. For most analytes, the values correlated well with previously reported literature data, while mean measurements were 50% lower for apoA-IV and apoC-III. The upper range of the ApoA-IV and ApoC-III values measured in the 25 samples were in the range of literature data. We observed that these relatively high ApoA-IV and ApoC-III levels were measured in hyperlipidemic samples (total cholesterol > 200 mg/dL and total triglyceride > 150 mg/dL).

Total cholesterol and triglyceride measurements were also acquired for the above mentioned 25 patient samples, which provided an opportunity to model total serum lipid levels as a function of apolipoprotein levels using multivariate least squares regression analysis (Fig. 8). The actual vs. predicted correlation for total cholesterol had R² = 0.89 (p-value < 0.001) and for total triglycerides had R² = 0.84 (p-value < 0.001). The multivariate regression analysis allows the assessment of the relative strength of correlations between lipid levels vs. individual apolipoprotein levels while correcting for other apolipoprotein levels in serum. Total cholesterol levels showed the most statistically significant correlation with apoB-100 levels (p-value < 0.001), while total triglyceride levels with apoC-III levels (p-value = 0.006). The strong correlation of apoC-III with total triglyceride levels is consistent with its known capability to inhibit both lipoprotein lipase activity and lipoprotein uptake by receptor sites [5,41], extending the circulation time of triglyceride rich lipid particles.