Spectrum of Apolipoprotein AI and Apolipoprotein AII Proteoforms and Their Associations With Indices of Cardiometabolic Health: The CARDIA Study.

Background ApoAI (apolipoproteins AI) and apoAII (apolipoprotein AII) are structural and functional proteins of high-density lipoproteins (HDL) which undergo post-translational modifications at specific residues, creating distinct proteoforms. While specific post-translational modifications have been reported to alter apolipoprotein function, the full spectrum of apoAI and apoAII proteoforms and their associations with cardiometabolic phenotype remains unknown. Herein, we comprehensively characterize apoAI and apoAII proteoforms detectable in serum and their post-translational modifications and quantify their associations with cardiometabolic health indices. Methods and Results Using top-down proteomics (mass-spectrometric analysis of intact proteins), we analyzed paired serum samples from 150 CARDIA (Coronary Artery Risk Development in Young Adults) study participants from year 20 and 25 exams. Measuring 15 apoAI and 9 apoAII proteoforms, 6 of which carried novel post-translational modifications, we quantified associations between percent proteoform abundance and key cardiometabolic indices. Canonical (unmodified) apoAI had inverse associations with HDL cholesterol and HDL-cholesterol efflux, and positive associations with obesity indices (body mass index, waist circumference), and triglycerides, whereas glycated apoAI showed positive associations with serum glucose and diabetes mellitus. Fatty-acid‒modified ApoAI proteoforms had positive associations with HDL cholesterol and efflux, and inverse associations with obesity indices and triglycerides. Truncated and dimerized proteoforms of apoAII were associated with HDL cholesterol (positively) and obesity indices (inversely). Several proteoforms had no significant associations with phenotype. Conclusions Associations between apoAI and AII and cardiometabolic indices are proteoform-specific. These results provide "proof-of-concept" that precise chemical characterization of human apolipoproteins will yield improved insights into the complex pathways through which proteins signify and mediate health and disease.

apolipoprotein AI

apolipoprotein AII

apolipoprotein B

Coronary Artery Risk Development in Young Adults study

proteoforms characterized by acylation at lysine residue 88 (K88) of ApoAI

liquid chromatography/mass spectrometry

post‐translational modification

Clinical Perspective

What Is New?

A proteoform is the precise molecular form of a protein, including allelic and splice variants, coding polymorphisms, and the nature, location(s) of any post‐translational modifications.

In this study, apolipoprotein AI (apoAI) and apolipoprotein AII (apoAII) present in human serum had 15 and 9 different proteoforms, respectively.

The magnitude and direction of association of proteoform abundance to cardiometabolic characteristics of 150 individuals varied substantially across proteoforms and between proteoforms and total protein abundance.

What Are the Clinical Implications?

Proteoform composition may vary significantly across biological states despite a smaller or undetectable difference in total protein concentration.

Thus, proteoform‐level characterization and quantification of apolipoproteins may enhance clinical and biological inference obtained from the measurement of total protein concentration alone.

ApoA1 (apolipoprotein AI) is the 11th most abundant protein in human serum. ApoAI serves as a major structural scaffold and binding ligand for high‐density lipoprotein (HDL) particles. , Thus, apoAI has a central role in lipid metabolism, mediating cholesterol transport and inflammatory, immunologic, and vasodilatory pathways. Like HDL‐cholesterol concentration (HDL‐C), apoAI has strong associations with markers of cardiometabolic health and risk of coronary heart disease (CHD). Although it is unclear whether the relationship between apoAI and CHD risk is causal, , strong associations between apoAI concentration and metabolic characteristics that are causally related to CHD risk, such as apoB (apolipoprotein B) and insulin resistance, support the hypothesis that ApoAI may be a mechanistically important mediator of cardiometabolic health. ,

The precise molecular form of a protein arising from a gene, including allelic and splice variants, coding polymorphisms, and the nature and location(s) of any post‐translational modifications (PTMs), is called a proteoform (Figure 1A). Previous reports have described a limited number of apolipoprotein PTMs, and several of those modifications are associated with differences in HDL functional properties. , , , , However, these reports are sparse, and have traditionally relied on protein digestion and subsequent PTM detection on small peptides (ie, bottom‐up proteomics), which forbids characterization of the precise proteoforms present in a sample. Consequently, data on apolipoproteins proteoforms and on their associations with differences in phenotypes are limited

Figure 1

Apolipoprotein proteoform analysis.

A, The proteoform profile of a gene product. The translation of different backbones from a single gene, be it because of allelic variation (shown here with apoAI [apolipoprotein AI] ) or splicing variation, combined with post‐translational modifications to these backbones contribute to the creation of a panel of different chemical species, all products of the same gene. These species—and not necessarily the unmodified coding product of the gene—are the actual molecules circulating and carrying out biological function in an organism. Each of these chemical variants of a gene product is called a proteoform. B, Study design. To search for associations between proteoform profiles and cardiometabolic phenotype, we used serum samples from 150 previously phenotyped CARDIA participants. Two samples were analyzed for each participant, corresponding to years 20 and 25 of the CARDIA study. Targeted top‐down mass spectrometry was used to discover, chemically characterize and quantify proteoforms of apoAI and apoAII. The percent contribution of each proteoform to the total amount of each apolipoprotein (ie, the quantitative proteoform profile) was compared with indices of cardiometabolic health for each individual studied. ApoAI indicates apolipoprotein AI; apoAII, apolipoprotein AII; apoB, apolipoprotein B; BMI indicates body mass index; CARDIA, Coronary Artery Risk Development in Young Adults; HDL‐C, high‐density lipoprotein cholesterol; and LDL‐C, low‐density lipoprotein cholesterol; PFR1‒3, proteoform 1‒3.

Recently, we have reported on a top‐down proteomic methodology to quantify whole proteoforms of apoAI in human serum, designed to capture a more comprehensive picture of apolipoprotein molecular variation. From a pilot group of 8 samples obtained from CHAS (Chicago Healthy Aging Study), we observed that proteoforms characterized by acylation (covalent fatty‐acid addition) at lysine residue 88 (K88) of ApoAI (K88acylApoAI) were positively associated with HDL efflux capacity. An association between an apoAI proteoform and higher efflux capacity had not been described previously, as other reports of apoAI proteoforms and/or PTMs showed associations with lower HDL efflux. In the CHAS study, however, we were unable to assess the associations between ApoAI proteoforms and other cardiometabolic characteristics because of the small sample size.

Thus, the objectives of the current study were to: characterize and quantify apoA‐I proteoforms in serum samples obtained from 150 well characterized adults from the Coronary Artery Risk Development in Young Adults (CARDIA) cohort collected at 2 different exam cycles 5 years apart; and to quantify the associations between apoAI proteoforms and cardiometabolic health indices including HDL efflux capacity (Figure 1B). We hypothesized that specific apoAI proteoforms would exhibit differences in association with cardiometabolic characteristics, and, specifically, that K88acylApoAI proteoforms would be positively associated with HDL‐C and HDL efflux and inversely associated with markers of poor cardiometabolic health.

Moreover, we set out to characterize the specificity of PTMs and overall proteoform chemistry and variation among apolipoproteins. We thus applied a similar proteoform characterization approach to apoAII (apolipoprotein A‐II), the second most abundant HDL‐associated protein. While some reports suggest apoAII is a mediator of HDL function , and that modifications of apoAII are associated differences in metabolism, no comprehensive characterization of apoAII proteoforms and their associations with cardiometabolic characteristics had yet been performed. Therefore, the present study also reports on associations between CARDIA characteristics and proteoforms of apoAII.

Methods

Study Cohort

Samples were obtained from the CARDIA study. Details of the CARDIA cohort and methods for risk factor measurement have been described elsewhere. Briefly, CARDIA is a community‐based cohort, designed to study the development of cardiovascular disease risk factors and their clinical sequelae in Black and White young adults in the United States. Between 1985 and 1986, 5115 participants between ages 18 to 30 years, 51.5% Black, 54.5% women were enrolled. At baseline and at each subsequent examination, participants underwent extensive in‐person measurement of CHD risk factors. Participants returned for examination at years 2, 5, 7, 10, 15, 20, 25, and 30, with 70% of the surviving cohort examined at year 30. In addition, health status and hospitalizations for cardiovascular endpoints are collected annually outside of clinic visits. Risk factors were measured, and samples were stored for all follow‐up examinations. At year 20, participants underwent coronary artery calcium (CAC) assessment. A total of 3547 participants returned for examination at year 20 and 3499 participants returned for examination at year 25. To date, the cohort has been followed through year 34.

Details of phenotypic characterization methodology, including CAC, HDL efflux, demographic and other serologic assessments are outlined in Data S1.

Sample Selection

CARDIA participants were eligible if they presented to the years 20 and 25 examinations, had serum samples available from both examinations, and had traditional risk factor measurement data, as well as CAC at the year 20 examination. Our sampling approach (described in detail below) was designed to: (1) allow for detection of the maximal number of apoAI proteoforms, (2) allow for analysis of apoAI proteoform intensity across a range of HDL‐C and efflux values, and (3) assess associations between proteoforms and subclinical atherosclerosis. We chose a sample size of 150 participants at years 20 and 25 (total number of samples: 300). Since this was discovery‐mode analysis, we did not perform a power calculation.

Since throughput is limited using current top‐down proteomics technology, we over‐sampled outliers to maximize the probability that we would detect low frequency proteoforms and detect associations with phenotype. We chose a 2×3 sampling approach based on prevalent CAC at year 20 and HDL‐C values at year 20 as outlined in Table 1. The cohort was stratified into each of the 6 categories, then we selected a random sample of 25 participants, who had sample available at years 20 and 25, from each CAC/HDL‐C subgroup.

Table 1

The 2×3 Sampling Used to Select Participant Samples for Proteoform Analysis

	CAC=0	CAC >0
HDL‐C >60 mg/dL	25	25
HDL‐C 40–60 mg/dL	25	25
HDL‐C <40 mg/dL	25	25

CAC indicates coronary artery calcium; and HDL‐C, high‐density lipoprotein cholesterol.

This study was approved by Northwestern Medicine's Institutional Review Board.

Incubation of ApoAI With Palmitic Acid

For detection of a potential non‐enzymatic reaction between apoAI and free fatty acids, in‐house‐purified non‐acylated apoAI was incubated with palmitic acid (Sigma‐Aldrich, Saint Louis, MO) at serum‐like conditions: (ApoAI)=1.5 g/L, (palmitate)=1.6 mmol/L, pH 7.4, 37°C. A 100 mmol/L ammonium bicarbonate buffer was used, for MS compatibility. Aliquots were collected at times: 2, 5, 24, and 48 hours, until apoAI oxidation/degradation prevented the potential detection of the acylated proteoform.

Proteoform Characterization and Quantification

Proteoforms were analyzed by liquid chromatography/mass spectrometry (LC‐MS) of participant apoB‐depleted serum samples. All peaks of similar charge, mass and retention time as apoAI or ‐aII were characterized by a classic top‐down proteomic approach. Serum samples of all individuals were run in randomized order in 6 LC‐MS analysis blocks (2 collection years of 25 individuals per block). Proteoform intensities were normalized by total ion current across runs and standardized across blocks. Standards of known apoAI concentration were analyzed along with samples in each block. Individuals' average percent abundance of each proteoform was compared with their characteristics in both CARDIA years 20 and 25. Associations between proteoforms and characteristics were assessed by either linear regression (for continuous characteristics) or t tests (for binary data). Then, either Pearson r and beta coefficients or fold differences were calculated, along with a p‐score. A Benjamini‐Hochberg multiple‐test correction was done, and significance was asserted at a 5% false discovery rate (2‐tailed type 1 error rate). Detailed descriptions of each of these analyses are present in the supplemental material.

Results

Participant Characteristics

Characteristics of participants grouped by HDL‐C levels are presented in Table 2. By sampling design, there were 50 in each HDL‐C group, half of whom had CAC >0 AU. The mean HDL‐C was 79, 49, and 34 mg/dL in the high, medium, and low HDL‐C groups with CAC; and 73, 50, and 35 mg/dL in the same groups without CAC, respectively. When compared with the middle and low HDL‐C groups, individuals in both high HDL‐C groups were more likely to be women, and individuals with high HDL‐C and CAC >0 were more likely to smoke. The high HDL‐C group had lower BMI, waist circumference, LDL‐C, triglycerides, and glucose levels. Individuals in higher HDL‐C groups had higher HDL efflux capacity. There was no difference in ABCA‐1‐dependent efflux across groups.

Table 2

Characteristics of the CARDIA Participant Sample at the Year 20 Exam Stratified by HDL‐C/CAC Groups

Group	CAC >0, HDL‐C >60	CAC >0, 40 ≤ HDL‐C ≤60	CAC >0, HDL‐C <40	CAC=0, HDL‐C >60	CAC=0, 40 ≤ HDL‐C ≤60	CAC=0, HDL‐C <40
n	25	25	25	25	25	25
Women, %	72	36	12	80	52	24
Age, y	47 (3)	46 (4)	47 (3)	47 (3)	44 (4)	44 (4)
Education, y	15 (3)	15 (3)	16 (2)	15 (2)	14 (3)	15 (2)
BMI, kg/m2	29 (6)	31 (7)	33 (5)	27 (5)	29 (8)	31 (4)
Height, cm	171 (11)	171 (9)	178 (9)	168 (9)	171 (10)	173 (10)
Weight, lbs	184 (40)	201 (43)	228 (38)	168 (34)	185 (42)	205 (33)
Waist circumference, cm	89 (13)	97 (13)	108 (13)	85 (12)	91 (14)	99 (11)
Physical activity	261 (198)	331 (253)	337 (222)	337 (273)	365 (311)	390 (273)
Alcohol, mL/dL	17 (18)	12 (12)	6 (14)	16 (22)	22 (68)	6 (15)
Current smoking, %	32	16	8	16	12	24
SBP, mm Hg	120 (16)	119 (17)	117 (12)	108 (11)	117 (11)	115 (12)
DBP, mm Hg	74 (11)	73 (13)	73 (8)	68 (8)	71 (8)	71 (9)
Blood pressure‐lowering medication, %	20	24	36	24	24	16
Total cholesterol, mg/dL	197 (41)	191 (37)	178 (36)	194 (29)	186 (54)	181 (32)
HDL‐C, mg/dL	79 (23)	49 (6)	34 (4)	73 (10)	50 (5)	35 (3)
LDL‐C, mg/dL	102 (34)	119 (37)	107 (31)	103 (30)	115 (50)	110 (29)
Triglycerides, mg/dL	82 (33)	114 (48)	194 (142)	92 (43)	107 (51)	177 (60)
Cholesterol‐lowering medication, %	16	28	36	0	12	8
Diabetes mellitus medication, %	4	4	16	4	12	12
Fasting glucose, mg/dL	96 (32)	99 (20)	120 (54)	92 (11)	97 (14)	102 (14)
Diabetes mellitus, %	4	4	20	4	4	8
Total CAC, AU	116 (241)	70 (107)	156 (330)	0 (0)	0 (0)	0 (0)
HDL efflux	1.3 (0.2)	1.1 (0.2)	1 (0.2)	1.2 (0.2)	1.1 (0.2)	1.1 (0.2)
ABCA1‐DEP. efflux	0.4 (0.2)	0.4 (0.2)	0.4 (0.2)	0.4 (0.2)	0.4 (0.2)	0.4 (0.2)
CRP	1.1 (1.7)	3.4 (4.8)	2.0 (2.3)	2.2 (2.3)	1.9 (2.6)	2.8 (3.1

Numbers represent either percent, for binary characteristics, or mean (SD), for continuous phenotype. ABCA1‐DEP. efflux indicates ABCA1‐dependent high‐density lipoprotein cholesterol efflux; BP, blood pressure; CAC, coronary artery calcium; CARDIA, Coronary Artery Risk Development in Young Adults; CRP, C‐reactive protein; DBP, diastolic blood pressure; LDL‐C, low‐density lipoprotein cholesterol; HDL, high‐density lipoprotein; HDL‐C, high‐density lipoprotein cholesterol; Rx, “under medication for”; and SBP, systolic blood pressure.

Proteoform Spectrum of Serum ApoAI

We set out to quantitate the full spectrum of proteoforms of apoAI detectable by top‐down LC‐MS in participant sera. Figure S1 shows raw mass spectrometric data for species that were identified as apoA‐I. LC‐MS data revealed a total of 15 apoAI species of distinguishable mass which were commonly detected across the individuals analyzed. Supporting fragmentation data can be found on the MassIVE database for this project, along with confidence metrics for proteoform identification (massive.ucsd.edu, dataset identifier: MSV000085676). Figure 2 outlines these species and depicts their differing chemical characteristics and abundance ranges in serum.

Figure 2

The proteoforms of apoAI [apolipoprotein AI].

Top: The 15 species targeted for characterization and quantification. Three protein backbones were observed, characterized here by their first and last amino‐acid residues in the canonical apoAI sequence (eg, R(‐6)‐Q243). Each single‐proteoform species was given a unique proteoform identifier. Upon characterization, several species were found to be mixtures of different proteoforms, either of the same or very similar mass. These included di‐oxidations; 16‐, 18‐, and 20‐carbon acylations on a canonical backbone and on a truncated backbone. Each of these similar‐mass sets was called a “proteoform group” and quantified as a single species. Proteoform abundance ranges were calculated based on total proteoform intensity divided by the summed intensity of all proteoforms of apoAI. Bottom: allelic backbones. An F71Y substituted backbone was observed, also presenting a set of 15 proteoforms and groups, created by the same modifications as the wildtype. The abundance range refers to the ratio of the summed intensity of proteoforms containing the allelic backbone to the sum of all ApoAI proteoform intensities in the heterozygotic individual observed. ApoAI indicates apolipoprotein AI; and PTM, post‐translational modifications.

Among common proteoforms, 3 apoAI backbone sequences were observed with different lengths: 1 containing the 6‐residue propeptide of apoAI (“ProApoA‐I”); the “canonical” backbone (residues D1‐Q243); and a “truncated” backbone, missing the C‐terminal Q243 residue. Furthermore, glycation of residue K133 by a hexose was observed, along with 3 levels of methionine sulphoxidation (1, 2 and 3 oxidized methionine residues), all on the canonical apoAI backbone. Lastly, 7 of the 15 apoAI species observed were covalent additions of fatty acids (acylations) to the lysine 88 residue (K88acylApoAI) happening both on the canonical and the truncated backbone sequences. Consistently, these species eluted from reversed‐phase liquid chromatography at a later retention time than the canonical proteoform because of their greater hydrophobicity (Figure S1). Further, while mass shifts of those proteoforms were consistent with modifications by fatty acids of 16, 18, 20, and 22 carbons, the combination of different unsaturated chains was not directly inferable from intact mass data, because of m/z overlap of apoAI ions differing by 2 hydrogens (1 unsaturation), as depicted in Figure S2. These species were therefore treated as “proteoform groups” based on the number of carbons of the fatty‐acid modification for purposes of proteoform quantification (vide infra).

In‐depth characterization of these acylated forms by ion fragmentation during tandem mass spectrometry allowed for further insight on the nature of apoAI acylation (Figure S2). Notably, fragmentation patterns were consistent with acylations at K88 by 16:0, 18:0, 18:1, 18:2, 18:3, 20:4, 20:5, and 22:6 fatty acids. Of these, only palmitoylation (16:0) of apoAI had been described before our group's targeted top‐down analyses of AapoAI. , Interestingly, these chain lengths and unsaturation states are consistent with the common types of fatty acids found in HDL particles. Furthermore, the relative MS intensity of each acylated proteoform group (16C, 18C, 20C, and 22C) is also in close agreement with the relative quantity of each fatty acid in HDL, as shown in A of Figure S2.

To inquire on the mechanism of acylation in vivo, we incubated canonical apoAI with palmitic acid at average serum concentrations and serum pH (7.4) for 48 hours (Figure S3). Notably, while signal of the canonical proteoform remained similar to the one observed in serum samples, where palmitoylated proteoforms could be detected, no signal for these forms was observed above 0.1% relative abundance (the detection limit for this study). Therefore, for these incubation conditions, apoAI acylation did not happen spontaneously at detectable levels.

Characterization of Allelic Variants of ApoAI

In the 2 serum samples of 1 of the individuals studied (in years 20 and 25 of the CARDIA study), an allelic variant of apoAI was observed (Figure S4). For this participant, 2 apoAI isoforms of roughly equal abundance (one of which was roughly 16 Da higher in mass) were observed and characterized by fragmentation. Intact mass and fragment mapping patterns were consistent with a F71Y polymorphism, indicating the individual was likely a heterozygote for APOAI. Interestingly, all the combinations of PTMs that were observed on the wildtype isoform of apoAI were also observed for this isoform.

Truncation and Dimerization of ApoAII

We also characterized the spectrum of apoAII chemical variation in the serum samples studied. Figure S5 shows raw LC‐MS data for apoAII. Nine proteoforms of apoAII could be observed and characterized, as outlined in Figure 3. The most abundant conformation of ApoAII observed was dimeric, with 2 chains linked by a disulfide bridge at C6. The 6 dimeric proteoforms observed were results of different levels of C‐terminal truncation of the 2 chains, as characterized by fragmentation and consistent with previous observations of apoAII dimers. Moreover, we observed 3 monomeric proteoforms of apoAII, also differing by their degree of backbone truncation. Notably, all monomeric forms were modified by cysteinylation at C6. On average, around 7% of the backbones of apoAII observed were cysteinylated, and thus monomeric.

Figure 3

Proteoforms of ApoAII [apolipoprotein II].

Top: Three monomers of apoAII. All monomeric forms observed were cystenylated (modified by disulfide bridge with a free cysteine) at C6. As depicted on the left, 3 backbones were observed, either intact, doubly, or singly truncated. These are characterized here by their first and last amino‐acid residues in the canonical ApoAII sequence (eg, Q1‐A75). Monomers of apoAII were each given a unique proteoform identifier. Bottom: 6 dimers of ApoAII. Dimers were 2‐way combinations of the backbones observed in the monomers, linked by disulfide bridge at C6. Proteoforms are named based on the last residues of their backbone and either cystenylation for monomers, or the last residues of the second backbone, for dimers (eg, AT/ATQ). Proteoform abundance ranges were calculated based on total proteoform intensity divided by the summed intensity of all proteoforms of ApoAII. PTM indicates post‐translational modifications.

Apolipoprotein Proteoform Profile of 150 Individuals

We analyzed the proteoform profile of 150 individuals, sampled at years 20 and 25 of the CARDIA study. Figure S6 depicts parameters of quality assurance for this proteoform quantification. Notably, total MS intensity was strongly correlated (R 2=0.97) with concentration of apoAI standards analyzed during LC‐MS quantification blocks. Moreover, the total MS intensity of apoAI was significantly correlated to participant HDL‐C values, at R=0.62, a coefficient consistent with previous results, acquired with different apoAI quantification tools. ,

Variation in the percent abundance of apoAI and ‐AII proteoforms across the 150 individuals is displayed in Figure 4. Canonical was the most abundant apoAI proteoform observed, while truncation and different levels of oxidation were the most common modified forms within individuals (Figure 4A). Furthermore, although acylated proteoforms were among the least abundant, ranging from 0.5% to 2% of the total MS intensity observed, this proteoform family was the one that accounted for most of the relative abundance variation across individuals (Figure S7). In contrast to apoAI, apoAII proteoform profiles varied more widely across individuals (Figure 4B). Notably, while dimeric proteoforms were consistently more abundant than monomeric ApoAII, no single proteoform was the most abundant overall.

Figure 4

Proteoform profile of 150 individuals.

Colors depict the average contribution of each proteoform to the total intensity of ApoAI (apolipoprotein AI) in each individual, as rank‐ordered by their high‐density lipoprotein cholesterol. Values are shown for CARDIA year 20 samples only. A, ApoAI proteoforms. “Acyl.” indicates acylations. B, ApoAII proteoforms. ApoAI indicates apolipoprotein AI; ApoAII, apolipoprotein II; and HDL‐C, high‐density lipoprotein cholesterol.

Covariance analysis of quantitative proteoform data (Figure S8) allowed for the characterization of proteoform “families,” which vary similarly across individuals. An unbiased factor analysis highlighted 3 strongly correlated clusters of proteoforms in ApoAI and also 3 in ApoAII. The 3 correlation factors corresponding to these groups explained, in both cases, >80% of proteoform variation in the individuals analyzed. Interestingly, the proteoforms pertaining to each covarying group contained similar types of PTMs, namely: acylations, oxidations, and truncation (for ApoAI), and single truncation, double truncation, and dimerization (for ApoAII). This suggests that proteoform covariation in these systems is mostly attributable to a small number of common underlying mechanisms involving certain PTMs. Moreover, truncated forms of ApoAI and ApoAII also significantly covaried, further suggesting that some PTM mechanisms might be shared across proteins.

Cross‐Sectional Associations of Proteoforms to Cardiometabolic Phenotype

We compared proteoform levels to individual cardiometabolic characteristics. In order to make this analysis independent of the association of total protein concentration to phenotype, we used the percent contribution of each proteoform to the total MS intensity of the apolipoprotein (Figure S8). Figure 5 shows heatmaps of the correlation coefficients relative to the association of each proteoform percent abundance to each continuous phenotype.

Figure 5

Correlation coefficient and significance of association of proteoforms percent abundance to continuous phenotype.

A, ApoAI proteoform associations. B, ApoAII proteoform associations. Left: CARDIA year 20. Right: CARDIA year 25. Abundances of each proteoform were compared with continuous phenotype in the 150 individuals studied. Clustering of both proteoforms and characteristics was unbiased. A Pearson r was generated for each correlation observed as well as a correlation P value. Colors show the strength and sign of each association. Statistical significance, symbolized by an asterisk, was asserted at 5% false discovery rate. ABCA1‐Dep. efflux indicates ABCA1‐dependent high‐density lipoprotein cholesterol efflux; Acyl, acylations; BMI, body mass index; CAC, coronary artery calcium; CRP, C‐reactive protein; DBP, diastolic blood pressure; HDL‐C, high‐density lipoprotein cholesterol; IMT, intima‐media thickness; LDL‐C, low‐density lipoprotein cholesterol; Phys Act, physical activity; and SBP, systolic blood pressure.

Notably, ApoAI proteoform heatmaps showed similar associations within covarying proteoform families and distinct associations between families. Moreover, these associations were similar between the 2 years studied. For instance, in both years 20 and 25 of CARDIA, hierarchical analysis clustered K88acylApoAI proteoforms together, and these forms showed the highest positive associations with HDL‐C and HDL efflux. These acylated forms also showed negative associations with BMI and waist circumference. Furthermore, a similar pattern, albeit with overall weaker associations and effect size, was observed with the non‐acylated truncated proteoform, which clustered with other truncated forms. These proteoforms also were negatively associated with common carotid thickness in both years. Conversely, the canonical proteoform showed an overall opposite association pattern, with significant negative association to HDL‐C and HDL efflux and positive associations to waist circumference, weight, and common carotid thickness. Mono‐oxidized ApoAI, as well as the dioxidized form in year 25, showed a significant positive association with HDL‐C, and, for year 20, mono‐oxidation showed a positive association with C‐reactive protein. Finally, as expected, we also observed a consistent and significant positive association between glycated ApoAI and serum glucose.

ApoAII proteoform heatmaps also showed distinct associations across the different proteoforms. In both years 20 and 25 of CARDIA, the dimer of singly truncated chains (AT/AT) of ApoAII had the strongest positive association to HDL‐C and HDL efflux. Concurrently, the fully non‐truncated monomer (ATQ/Cys) and dimer (ATQ/ATQ) were clustered together, showing significant positive associations with markers of obesity (BMI, weight or waist circumference), the dimer also showed a negative association with HDL‐C and HDL efflux. Finally, while most other forms showed no significant associations with phenotype, the overall clustering pattern suggests that proteoform‐to‐phenotype association is dependent both on number of truncations (singly truncated proteoforms were more strongly associated with higher HDL‐C and lower obesity indices) and the number of chains present in the molecule (only dimers were associated with higher HDL‐C and lower obesity indices).

For binary characteristics, a separate analysis was made (Figure S9). Notably, only one association was significant and consistent between both years of study: a higher level of glycated ApoAI was associated with diabetes mellitus.

Select proteoform‐to‐characteristic association data (R 2, beta coefficients and confidence parameters), including motif‐based aggregates of proteoform intensity, are shown in Table S1. Most noteworthy, canonical abundance variation explained (based on R 2 values) 20% and 10% of the variation in HDL‐C and HDL efflux in this dataset, while aggregate acylations explained 31% and 12%, respectively. However, associations with cardiometabolic indices varied in directionality: one positive standard deviation in canonical abundance was associated with −9.1 mg/dL of HDL‐C and +3.9 cm of waist circumference, while beta coefficients were +11.4 mg/dL and −5.9 cm, respectively for the same characteristics and K88acylApoAI.

Longitudinal Analysis of Change in Proteoform Abundance to Change in Phenotype

We analyzed intra‐individual changes in proteoform profile between years 20 and 25 of the CARDIA study and compared them to change in participant characteristics over time. Figure 6 shows a heatmap of the correlation coefficients observed in this analysis. Notably, several associations observed in the cross‐sectional analysis were also observed longitudinally. For instance, changes in truncated proteoforms were positively associated with changes in HDL‐C, while change in canonical abundance was inversely associated with change in HDL‐C. Change in acylated forms were negatively associated with changes in BMI and waist circumference, and glycation was positively associated with changes in serum glucose. No significant associations were observed between change in ApoAII proteoform abundance overtime and change in cardiometabolic characteristics.

Figure 6

Correlation coefficient and significance of association of intra‐individual changes in ApoAI proteoform profile to changes in continuous phenotype.

Changes in abundances of each proteoform were compared with changes in continuous phenotype within each of the 150 individuals studied. Clustering of both proteoforms and characteristics was unbiased. A Pearson r was generated for each correlation observed as well as a correlation P value. Colors show the strength and sign of each association. Statistical significance, symbolized by an asterisk, was asserted at 5% false discovery rate. ABCA1‐Dep. efflux indicates ABCA1‐dependent high‐density lipoprotein cholesterol efflux; Acyl, acylations; BMI, body mass index; CAC, coronary artery calcium; CRP, C‐reactive protein; DBP, diastolic blood pressure; HDL‐C, high‐density lipoprotein cholesterol; IMT, intima‐media thickness; LDL‐C, low‐density lipoprotein cholesterol; Phys Act, physical activity; and SBP, systolic blood pressure.

Discussion

Summary of Results

In this analysis, we identified and quantified 15 distinct ApoAI proteoforms in 150 CARDIA participants for both CARDIA exam years 20 and 25, representing the largest cohort for proteoform measurement to date. Interestingly, top‐down proteomics was able to identify gene‐level variation as well. We observed, in one participant, a well‐described and prevalent allelic variant of ApoAI (F71Y), which also contained 15 distinct proteoforms, suggesting that there are no significant differences in ApoAI PTM physiology for this allele.

Moreover, associations with cardiometabolic traits appear to cluster ApoAI proteoforms in 4 motifs: canonical plus 3 proteoform families (truncations, oxidations, and K88acylApoAI). We observed motif‐specific associations between proteoform percent abundance and participant characteristics. For instance, percent canonical and K88acylApoAI proteoforms had significant associations with HDL‐C, HDL efflux and markers of obesity, but the direction of association between these 2 proteoform motifs and indices of cardiometabolic health were opposite from each other (canonical was inversely associated with HDL‐C and positively associated with waist circumference; vice‐versa for K88acylApoAI).

Similarly, we observed motifs of proteoform covariation and associations with cardiometabolic phenotype in the 9 proteoforms of ApoAII. Importantly, while truncations of ApoAI and single truncations of ApoAII (both characterized by the loss of the C‐terminal glutamine residue) covaried across individuals, common side‐chain PTMs of ApoAI, such as acylations, had no analogous product in ApoAII. Furthermore, similar to ApoAI, the patterns of association of ApoAII proteoforms to cardiometabolic phenotype were largely proteoform‐specific.

The proteoform‐specific associations between ApoAI and AII and cardiometabolic health indices suggest that many proteoforms may be the cause or consequence of distinct (and possibly antagonistic) biological pathways involved in cardiometabolic health. The nature of this association to phenotype likely varies by proteoform. Figure 7A shows examples of hypothetical pathways through which metabolic, genetic and other biological factors may mediate the proteoform profile of ApoAI and AII. We posit that because specific proteoforms are the consequence of distinct biochemical pathways, which may be mediators or markers of phenotype, to the extent true, proteoform variation is more directly related to the biology underlying differences in phenotypic states than the observed variation in the aggregated concentration of proteins (Figure 7D through 7G).

Figure 7

The differential association of proteoforms to phenotype.

A, Proteoform characterization integrates the effects of genetic variation and elements of the metabolome and possibly the exposome on the proteome. For example, ApoAI (apolipoprotein AI) proteoforms are the result of allelic variations of the APOAI gene, which is likely modified by enzymes coded for by other gene loci. B, Further, glycation likely occurs because of a well‐described non‐enzymatic Schiff‐base reaction between serum glucose and proteins. C, Similarly, we hypothesize that acylation occurs on the high‐density lipoptotein particle, suggesting that K88Acyl AI may be a marker of increased metabolic activity of ApoAI in high‐density lipoptotein, which may explain the inverse associations between canonical ApoAI, acylated A‐I, and markers of cardiometabolic health. D, Proteoform composition may vary significantly across biological states despite a smaller, or undetectable difference in total protein. Thus, measurement of total protein concentrations, for instance using standard ELISA assays, may fail to detect significant differences in proteoform abundance, which could modify the associations detected (as we demonstrate with ApoAI) and give insight into the biology that mediates phenotype. E through G, Examples of proteoform‐specific associations with phenotype observed in this study. Crosses in (E) indicate participants with diabetes mellitus. For these panels absolute proteoform intensity was used, to compare with total apolipoprotein intensity. ApoAI indicates apolipoprotein AI; ApoAII, apolipoprotein II; HDL, high‐density lipoprotein; and HDL‐C, high‐density lipoprotein cholesterol.

Glycated ApoAI and Serum Glucose

Some proteoforms may result from enzymatic and non‐enzymatic interactions with metabolites and thus they may serve as indices of metabolic dysregulation, which may not be detectable by measurement of total protein concentration. For example, analogous to hemoglobin A1c, higher levels of glycated ApoAI are most likely the consequence of a well‐described Schiff base reaction between serum glucose and serum proteins. Thus, similar to the hemoglobin A1c assay, glycated ApoAI appears to be a candidate marker of the aggregate exposure to serum glucose (Figure 7B). Interestingly the association between higher blood glucose levels and higher glycated ApoAI appears to be driven by participants with diabetes mellitus, suggesting, as for hemoglobin A1c, that glycation of ApoAI occurs in the setting of prolonged exposure to elevated blood glucose levels. It follows then that the total intensity of glycated ApoAI is more strongly associated with serum glucose and diabetic status than total ApoAI (Figure 7E).

Acylation of ApoAI at Lysine 88 and HDL‐C/HDL Efflux

We suspect acylation at K88 is an enzymatically mediated process because incubation of canonical ApoAI with palmitic acid did not spontaneously produce acylated proteoforms. Furthermore, acylated proteoforms of ApoAII were not observed, which argues that acylation is not a random occurrence on abundant HDL‐associated proteins.

We also hypothesize that the abundance of K88acylApoAI may be a marker of the overall metabolic activity of ApoAI on HDL. ApoAI is known to exist in unbound and HDL‐bound forms. , , Interestingly, the fatty acids that are present on the acylated forms of ApoAI are the most prevalent fatty acids on HDL particles, implying that the fatty acids added to ApoAI are derived from fatty acids imbedded in HDL particles (Figure 7C). We hypothesize that ApoAI acylation may be a consequence of promiscuous activity of HDL‐maturation enzymes that use fatty acids as substrates (such as LCAT or PLTP, involved in cholesterol esterification and phospholipid formation, respectively). Thus, individuals with a larger percentage of K88acylApoAI may not only have more HDL‐active ApoAI, but also better HDL maturation (higher lecithin‐cholesterol acyltransferase or phospholipid transfer protein activity) and consequently more reverse cholesterol transport and TG exchange with apoB particles, and potentially lower risk for developing advanced or unstable atherosclerosis. Conversely, a higher percentage of canonical ApoAI may represent a larger relative amount of free, non‐metabolically active ApoAI, which may explain its inverse association with HDL‐C.

Dimerization, Truncation of ApoAII and HDL‐C

ApoAII is the second most prevalent protein on HDL particles. It is known to exist in monomeric, dimeric, singly‐ and doubly‐truncated forms as we report in this study. , , The role of ApoAII in HDL structure and function is less well defined than ApoAI. ApoAII is thought to serve as a competitive antagonist to ApoAI and thus modulate the functions of enzymes like lecithin‐cholesterol acyltransferase and hepatic lipase that use ApoAI as a cofactor. , Further, limited data suggest that some allelic variants of ApoAII are associated with visceral adiposity and CHD. This study is the first that provides a comprehensive assessment of ApoAII proteoform relative distributions and their associations with cardiometabolic phenotypes in humans. Given the prevalence of ApoAII in HDL particles, its previously described role in HDL structure and function, and the associations we report in this paper, it is possible that truncation of the terminal glutamine and dimerization of ApoAII may be involved with the activation of ApoAII's role in lipid metabolism. Moreover, cysteinylation of monomeric chains might regulate ApoAII dimerization and thus negatively affect its HDL‐associated function. Therefore PTM‐regulated differences in activity might underpin the stark difference in association to cardiometabolic indices observed between intensity of specific proteoforms and total ApoAII intensity (Figure 7G).

Implications for Lipid Biology

Our results add to a rapidly growing understanding of the HDL‐associated proteome. Over 120 proteins are associated with HDL particles and the proteome varies significantly by HDL particle composition, size, number, and function. , , The contribution of PTMs to the phenotypic and functional diversity of HDL particles in serum has not been well described and proteoform‐level analyses can help fill this knowledge gap.

We believe these data have broader implications for both proteomics technology and lipid biology. For the latter, it is noteworthy that other apolipoproteins undergo post‐translational modification, and these modifications likely have functional significance. , , , The findings we report in this analysis suggest that identification of apolipoprotein proteoforms may be a very useful way to gain novel insight into the complex biology of lipoprotein metabolism and its associations with states of health and disease. Furthermore, describing the differences in proteoform profiles across different tissues and disease states will provide important insights into the pathobiology of health and disease. Although we report ApoAI proteoforms in the serum compartment, there have been reports of a W72‐oxidized proteoform of ApoAI in human atheroma, which was not identified in this sample of human serum, suggesting that proteoform composition and the relative abundance of their PTMs may vary substantially by tissue compartment.

Strengths and Limitations

Using a methodology that we had previously employed in a small pilot sample from the Chicago Healthy Aging Study study, we identified similar ApoAI proteoform motifs, characterized new ApoAII motifs, and we greatly expanded up on our understanding of their associations with cardiometabolic phenotypes. CARDIA allows for linkage of proteoform data to high‐quality assessment of asymptomatic free‐living individuals. Thus, we are able to examine the spectrum of proteoform diversity in the context of extensive participant phenotyping.

However, this analysis should be interpreted in the context of its limitations, as well. First, the dynamic range of current top‐down proteomic technology does not allow for the identification of ApoAI proteoforms present in <0.1% relative abundance. Second, top‐down proteomics requires label‐free quantification, which provides relative quantification of proteoforms. However, MS intensity was highly correlated with absolute ApoAI concentration standards and total ApoAI MS intensity correlated with HDL‐C at correlation coefficients previously reported in the literature using standard, high‐fidelity assays. This suggests that the rank‐order and scale of ApoAI proteoform quantification by top‐down mass spectrometry is accurate. Third, the effect sizes reported are modest, but given the random error intrinsic to proteoform quantification the true biological associations are likely much stronger than what we report. Fourth, some of the proteoforms could be byproducts of sample preparation or electrospray ionization. In fact, we suspect this is the case of at least some of the oxidized proteoforms that we identified. However, the significant differences in the abundance of oxidized forms between individuals despite the same preparation technique and randomized injection, and a significant intra‐individual correlation of oxidation abundance between the 2 years analyzed suggest that a biological pathway mediates these differences, not electrospray ionization or sample preparation.

Conclusions

We report the correlation of 21 total proteoforms of ApoAI and ApoAII across 150 people and show their quantitative analysis down to <1% relative abundance. It is noteworthy that, while each proteoform is an independent mass spectrometric measurement, covariance of proteoforms was consistent within types of PTMs (motifs), the associations between these proteoforms and indices of cardiometabolic health were specific to and consistent within proteoform motifs and these associations portrayed externally consistent metabolic profiles. The most interesting example were acylations of ApoAI, which strongly covaried and were significantly and positively associated with indices of cardiometabolic health, such as a low waist circumference and high HDL‐C, whereas the canonical form of ApoAI had opposite directions of association with the same cardiometabolic characteristics. Moreover, the discordance in associations between total protein concentration and specific proteoform intensities with cardiometabolic characteristic(s) highlights the importance of being precise about the biochemistry of proteins for biomarker research. Our data provide the largest‐scale identification and associations between apolipoprotein proteoforms and human phenotypes. As such, the work serves as an example of using proteoform‐resolved measurements to improve efficiency in gaining insight into the complex pathways through which proteins mediate health and disease.

Sources of Funding

Work performed for this study was funded by the National Institutes of Health, under grants K23 HL133601‐03 (Wilkins), the American Heart Association, under grant SDG 27250022 (Wilkins), and the National Institute of General Medical Sciences, under grant P41 GM108569 (Kelleher). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The CARDIA study is conducted and supported by the National Heart, Lung, and Blood Institute in collaboration with the University of Alabama at Birmingham (HHSN268201800005I & HHSN268201800007I), Northwestern University (HHSN268201800003I), University of Minnesota (HHSN268201800006I), and Kaiser Foundation Research Institute (HHSN268201800004I). This article has been reviewed by CARDIA for scientific content.

Disclosures

None.

Data S1

Table S1

Figures S1–S9

References 12, 37, 38, 39, 40, 41, 42

Click here for additional data file.