Genetic and secondary causes of severe HDL deficiency and cardiovascular disease1

We assessed secondary and genetic causes of severe HDL deficiency in 258,252 subjects, of whom 370 men (0.33%) and 144 women (0.099%) had HDL cholesterol levels <20 mg/dl. We excluded 206 subjects (40.1%) with significant elevations of triglycerides, C-reactive protein, glycosylated hemoglobin, myeloperoxidase, or liver enzymes and men receiving testosterone. We sequenced 23 lipid-related genes in 201 (65.3%) of 308 eligible subjects. Mutations (23 novel) and selected variants were found at the following gene loci: 1) ABCA1 (26.9%): 2 homozygotes, 7 compound or double heterozygotes, 30 heterozygotes, and 2 homozygotes and 13 heterozygotes with variants rs9282541/p.R230C or rs111292742/c.-279C>G; 2) LCAT (12.4%): 1 homozygote, 3 compound heterozygotes, 13 heterozygotes, and 8 heterozygotes with variant rs4986970/p.S232T; 3) APOA1 (5.0%): 1 homozygote and 9 heterozygotes; and 4) LPL (4.5%): 1 heterozygote and 8 heterozygotes with variant rs268/p.N318S. In addition, 4.5% had other mutations, and 46.8% had no mutations. Atherosclerotic cardiovascular disease (ASCVD) prevalence rates in the ABCA1, LCAT, APOA1, LPL, and mutation-negative groups were 37.0%, 4.0%, 40.0%, 11.1%, and 6.4%, respectively. Severe HDL deficiency is uncommon, with 40.1% having secondary causes and 48.8% of the subjects sequenced having ABCA1, LCAT, APOA1, or LPL mutations or variants, with the highest ASCVD prevalence rates being observed in the ABCA1 and APOA1 groups.

Although many studies have examined potential genetic causes of low HDL-C, there is no prior study to our knowledge that has carefully examined both secondary and genetic causes of severe HDL-C deficiency in a population of more than 250,000 subjects and related such findings to ASCVD (3). Our purpose in this investigation was to assess the prevalence of severe HDL deficiency in a large reference laboratory population and to determine potential underlying secondary and genetic causes by direct DNA sequencing of genes that are germane for lipoprotein metabolism, including APOA1, ABCA1, LCAT, and LPL, as well as to ascertain the prevalence of ASCVD in the various genetic HDL deficiency states.
Decreased levels of serum HDL-C (<40 mg/dl in men and <50 mg/dl in women) have been found to be a very significant independent risk factor for ASCVD and in this regard were superior to serum apoA-I levels in the Framingham Offspring Study (4,5). An even better marker of ASCVD risk may be decreased HDL function as assessed by cellular cholesterol efflux (6,7). Pre-1 HDL, the smallest HDL subparticle separated by two-dimensional gel electrophoresis, serves as the acceptor of cellular free cholesterol and phospholipids via the ATP binding cassette transporter 1 (ABCA1) and, in the process, is converted to larger  HDL particles, while 1 HDL serves as both a donor and acceptor of cholesterol in its interaction with liver cells via scavenger receptor B1 (8)(9)(10). In patients with low HDL and premature ASCVD, compared with control subjects, there are significantly lower apoA-I levels in 1 HDL and higher apoA-I levels in pre-1 HDL than in other HDL particles (11)(12)(13)(14). Hypertriglyceridemia is also associated with increased levels of pre-1 HDL (15). Increasing apoA-I levels in 1 HDL by means of treatment with the simvastatin/ niacin combination has been associated with decreased coronary atherosclerosis (16).
In prior studies of families with premature ASCVD, we reported that 19% had lipoprotein (a) excess, 15% had familial dyslipidemia (high TGs and low HDL-C), 14% had familial combined hyperlipidemia (high LDL-C, high TGs, and usually low HDL-C), 4% had isolated low HDL (hypoalphalipoproteinemia), and 1% had familial hypercholesterolemia, according to the 10th and 90th percentile values from the Framingham Offspring Study (17,18). Our data indicate that low HDL-C in families with premature ASCVD is frequently associated with either elevated TGs or elevations of both TGs and LDL-C. In the general population, we have also documented that low HDL-C levels are significantly associated with hypertriglyceridemia, obesity, diabetes, men, sedentary lifestyle, and cigarette smoking (19). Patients with markedly low plasma concentrations of HDL-C (<20 mg/dl) in the absence of marked hypertriglyceridemia, marked inflammation, diabetes, liver disease, or the use of anabolic steroids are therefore uncommon (3) and are most likely to have homozygous, compound heterozygous, or heterozygous defects involving the APOA1, ABCA1, or LCAT genes (1).
The overall data presented in this investigation are consistent with the concept that approximately 40% of patients with severe HDL deficiency have secondary causes, including severe hypertriglyceridemia, increased inflammation, and uncontrolled diabetes. In the absence of such secondary causes, about half of patients with markedly low HDL-C have mutations or variants at the ABCA1, LCAT, or APOA1 gene loci as determined by DNA sequencing. Moreover, compared with nonaffected control subjects, patients with ABCA1 or APOA1 mutations or variants have an increased prevalence of ASCVD. In contrast to the conclusions of some investigators, the data indicate that genetic causes of severe HDL deficiency can, in turn, cause premature ASCVD (1)(2)(3).

METHODS
The subject population consisted of 112,776 men and 145,476 women referred to our reference laboratory over a 3-year period (2014-2016) that had the following parameters assessed in serum or plasma after an overnight fast using methods as previously described (20): total cholesterol, TGs, direct LDL-C, small dense LDL-C (sdLDL-C), HDL-C, apoA-I, apoB, HDL particle analysis, glucose, insulin, adiponectin, high-sensitivity C-reactive protein (hsCRP), fibrinogen, myeloperoxidase (MPO), and liver transaminases. We utilized results from only the first laboratory sample if a subject was sampled more than once. The median age of this population, the gender percentage, and the fasting lipid concentrations were quite similar to values previously obtained from subjects participating in cycle 6 of the Framingham Offspring Study (11). However, because our subjects were not randomly selected but were rather selected by their healthcare providers to have their serum and plasma biochemistries characterized by a reference laboratory, they may not be representative of middle-aged and elderly subjects in the general US population.
The lipid assays were standardized by the Lipid Standardization Program of the Centers for Disease Control (Atlanta, GA), and the apolipoprotein assays were standardized by the Northwest Lipid Research Clinics Program of the University of Washington (Seattle, WA). All assays except the HDL particle analyses and the MPO assays were performed using automated enzymatic analyses on Roche COBAS automated analyzers with reagents obtained from Roche Diagnostics (Indianapolis, IN). The within- and between-run coefficients of variation for these assays were <4%. The MPO measurements were performed using a chemiluminescence assay obtained from Siemens (Malvern, PA) on a Siemens Dimension 200 EXL platform, with within- and between-run coefficients of variation of <5%. The apoA-I content of HDL particles was analyzed by two-dimensional gel electrophoresis followed by immunoblotting as previously described, with within- and between-run coefficients of variation of <10.0% for all particles except for pre-1 HDL, which was <15.0% (11).
In this population, 0.33% of men (n = 370) and 0.099% of women (n = 144) had fasting HDL-C concentrations <20 mg/dl. From this population, we excluded all subjects with hsCRP concentrations >10.0 mg/l, fasting TG concentrations >600 mg/dl, glycosylated hemoglobin (HbA1c) levels >8.0%, and liver transaminase levels >120 U/l, as well as men receiving testosterone replacement. These cutoffs for TGs, MPO, and liver enzymes were chosen because they represent values greater than the 99th percentile of the values in our population, while the remaining cutoffs were chosen because they represent significant inflammation, uncontrolled diabetes mellitus, and/or hormonal effects that are known to lower HDL-C levels (21).
Under a protocol approved by the Schulman Investigational Review Board (Cincinnati, OH), a total of 201 of 308 (65.3%) eligible subjects were studied, provided that their healthcare providers consented to an anonymized genetic analysis at no cost to either the patient or the healthcare provider. The remaining 107 subjects were not tested because 1) we never received a sample for DNA analysis (the main reason), 2) the healthcare provider did not give consent because he/she was no longer following the patient or did not want to participate in the study, or 3) the healthcare provider did not respond to our letter or phone calls. Data on these 107 subjects are provided in the Results section.
Clinical information from the study subjects was obtained from the healthcare provider by telephone interview using a questionnaire and was placed in an anonymized database. The questionnaire asked about the subject's history of angina pectoris, myocardial infarction, coronary artery bypass grafting or angioplasty, stroke, transient ischemic attacks, carotid or peripheral artery disease, and/or surgery for these latter conditions. If any of these responses was positive, the subject was deemed to have ASCVD. The questionnaire also asked about the subject's history of any form of kidney disease, decreased kidney function, proteinuria, neuropathy or neurological disease, seizures, hepatosplenomegaly, and/or corneal opacification or visual impairment. DNA was isolated from whole blood of the 201 eligible subjects and analyzed by next-generation sequencing at 23 gene loci (APOA1, ABCA1, LCAT, LPL, LIPC, LDLR, APOB, PCSK9, LDL-RAP1, STAP1, ABCG5, ABCG8, APOE, LIPA, GPIHBP1, APOA5, APOC2, APOC3, CETP, SCARB1, CYP27A1, DHCR24, and LIPG) on an Illumina MiSeqDx platform using 2 × 150 paired-end reads as previously described (22). With the exception of CYP27A1, these genes were selected by Johansen et al. (22) because of their known role in causing monogenic lipid disorders. We added the gene CYP27A1 because of its role in causing cerebrotendinous xanthomatosis. For each gene, sequencing was performed on all known exons plus an additional 25 base pair padding at each intron-exon border. The median read depth was 517×, with 100% sensitivity and specificity for SNPs and 100% sensitivity and 91% specificity for insertions/deletions. FASTQ files were processed using a custom workflow in CLC Biomedical Genomics Workbench version 3.2 (Qiagen, Valencia, CA). All mapping, variant calling and primary annotation, and quality control were performed using validated protocols. Variant call format files containing all identified variants in targeted regions were annotated and further analyzed using Ingenuity Variant Analysis (Qiagen) with a customized filtering cascade designed to identify potentially disease-causing variants. The analysis incorporated the classification guidelines described by the American College of Medical Genetics and Genomics (23). In addition, copy number variations and large structural variants were assessed using built-in tools from Biomedical Genomics Workbench (Qiagen) and VarSeq copy number variant analysis (Golden Helix, Bozeman, MT). Bioinformatic analysis of the sequencing results was confirmed by Dron et al. (24) as previously described. Our genetic analysis focused on three gene loci directly involved in HDL particle metabolism (APOA1, ABCA1, and LCAT). We also selected four common SNPs (ABCA1 rs111292742, ABCA1 rs9282541, LCAT rs4986970, and LPL rs268) known to be associated with HDL-C and TG plasma concentrations (discussed below) and analyzed their prevalence and minor allele frequency in our study population of 201 subjects versus the reported global population (25). Results of the analysis were given to the healthcare provider if requested.
Further analysis was required for variants that were novel or of uncertain significance to ascertain the potential pathogenicity. Prediction scores used for this assessment included 1) Genomic Evolutionary Rate Profiling scores to evaluate conservation; 2) DANN pathogenicity scores, which use computational deeplearning methodologies to classify variants; and 3) dbNSFP, which aggregates several informatics prediction algorithms and uses a proportion of these algorithms to indicate the number of scores predicting the mutation to be deleterious versus the number predicting low functional impact. These scores were assessed using Varsome (Saphetor, Lausanne, Switzerland) and other computational and literature-based tools in Ingenuity Variant Analysis.

Prevalence of secondary causes of severe HDL deficiency
As shown in Table 3, among men with severe HDL deficiency, 19.1% had fasting TG values >600 mg/dl versus 0.1% of normal controls; 18.9% had hsCRP values >10.0 mg/l versus 3.7%; 14.2% had HbA1c values >8.0% versus 3.1%; 4.1% had MPO values >1,000 pmol/l versus 0.6%; and 3.3% had elevated liver transaminase levels >120 U/l versus 0.5% (all P < 0.001). These cutoffs for TGs, MPO, and liver enzymes were chosen because they represent values greater than the 99th percentile of the values in our population, while the remaining cutoffs were chosen because they represent significant inflammation and/or uncontrolled diabetes mellitus. Each of these parameters is known to be associated with HDL deficiency (3). Men with low HDL-C (20-39 mg/dl) generally had prevalence rates for these abnormalities that were intermediate between men with severe HDL deficiency and normal controls. All men with these abnormalities were excluded from the genetic studies, as were 38 men (10.2%) with an HDL-C value <20 mg/dl who were receiving testosterone replacement, which is known to lower HDL-C levels significantly (21). Table 3 also shows that among women with severe HDL deficiency, 24.5% had hsCRP values >10.0 mg/l versus 5.4% of controls, 21.9% had fasting TG values >600 mg/dl versus 0.0%, 11.2% had HbA1c values >8.0% versus 1.6%, 5.1% had MPO values >1,000 pmol/l versus 0.7%, and 1.6% had elevated liver transaminase levels >120 U/l versus 0.3% (all P < 0.001). Women with low HDL-C (20-49 mg/ dl) generally had prevalence rates for these abnormalities that were intermediate between women with severe HDL deficiency and those with normal HDL-C. All women with these abnormalities were excluded from the genetic studies. In both men and women the major reasons for exclusion were because of severe hypertriglyceridemia, markedly increased inflammation (markedly elevated hsCRP and/or MPO levels), and/or uncontrolled diabetes mellitus.

Genetic causes of severe HDL deficiency
Of a total of 308 eligible subjects, there were 107 (72% men) for whom no DNA was available. Their mean age was 55 years, and their mean lipid and lipoprotein concentrations were as follows: non-HDL-C, 154 mg/dl; direct LDL-C, 107 mg/dl; TGs, 198 mg/dl; HDL-C, 18 mg/dl; and apoA-I, 87 mg/dl, with an HDL-C/apoA-I ratio of 0.207. Of these subjects, 5 (4.7%; all men) were reported by their healthcare provider to have established ASCVD, with a mean age b HDL-C/apoA-I ratio medians in men for the normal HDL-C, low HDL-C, and severely deficient HDL-C groups are 0.341, 0.288, and 0.214, respectively. of onset of 61 years. There were 201 subjects (65.3%) with HDL-C levels <20 mg/dl and without known secondary causes of low HDL-C who underwent DNA analysis. Of these subjects, 172 (85.6%) were Caucasian, 16 (8.0%) were Hispanic, 10 (5.0%) were African American, and 3 (1.5%) were Asian. Of 94 of the 201 subjects sequenced (46.8%; 72 men and 22 women), no ABCA1, LCAT, or APOA1 mutations, selected variants (ABCA1 rs111292742, ABCA1 rs9282541, LCAT rs4986970, LPL rs268), or other mutations were found (mutation-negative group). The HDL-C/apoA-I ratio medians in women for the normal HDL-C, low HDL-C, and severely deficient HDL-C groups are 0.374, 0.310, and 0.197, respectively.
Data are expressed as the number (%) of subjects in the HDL-C group. ALT, alanine aminotransferase; AST, aspartate aminotransferase; NC, not calculated. a data on these subjects were similar to the data on the subjects for whom no DNA was available for sequencing. The mean age of the mutation-negative group was 53 years; the mean lipid and lipoprotein concentrations were as follows: non-HDL-C, 142 mg/dl; direct LDL-C, 99 mg/dl; TGs, 237 mg/dl; HDL-C, 16 mg/dl; and apoA-I, 81 mg/dl, with an HDL-C/apoA-I ratio of 0.198. Of these 94 subjects, 6 (6.4%; 4 men and 2 women) were reported to have established ASCVD by their healthcare provider, with a mean age of disease onset of 64 years.

ABCA1 mutations and variants
Of the 201 subjects sequenced, 54 (26.9%; 38 men and 16 women) were noted to have ABCA1 mutations (Tables 4 and 5) or SNPs ( Table 6); 18 of the mutations were determined to be novel. Among the 39 subjects (19.4%) having ABCA1 mutations, 2 men, subjects ABCA1-01 and ABCA1-02, were homozygous for the novel mutations ABCA1 p.G851R and p.P1412R, respectively. Their plasma concentrations of HDL-C, apoA-I, nonHDL-C, LDL-C, and TGs were 5 and 6 mg/dl, 10 and 24 mg/dl, 35 and 91 mg/dl, 5 and 77 mg/dl, and 312 and 67 mg/dl, respectively. ABCA1-01 was also heterozygous for LCAT rs4986970 (p.S232T) (see below) and at the age of 56 years had evidence of splenomegaly, anemia, and peripheral neuropathy (numbness and tingling of his right arm) but no evidence of ASCVD. At the age of 47 years, he had undergone coronary angiography at the National Institutes of Health that showed no evidence of coronary atherosclerosis. It should be noted that his direct LDL-C value was very low (5 mg/dl). HDL particle analysis indicated that almost all of his apoA-I was in pre-1 HDL. ABCA1-02 was recently reported, after we supplied the healthcare provider with the genetic information; unfortunately, in the published report the mutation was mislabeled as ABCA1 p.P1312R (26). At the age of 38 years, this patient was asymptomatic with no evidence of ASCVD and had a marked decrease in apoA-I in all HDL particles, especially in very large 1 HDL, which was approximately 2% of normal.
As shown in Table 4, there were 5 ABCA1 compound heterozygotes (2 men, 3 women) with a mean age of 46 years and mean lipid and lipoprotein concentrations as follows: HDL-C, 9 mg/dl; apoA-I, 30 mg/dl; non-HDL-C, 118 mg/ dl; LDL-C, 94 mg/dl; and TG, 180 mg/dl. Three of the mutations found in these subjects were novel. Subjects ABCA1-03 and ABCA1-06 had established ASCVD (with the age of disease onset of 46 and 58 years, respectively), with non-HDL-C concentrations >80 mg/dl. When analyzed together, the 7 subjects (4 men, 3 women; mean age: 46 years) who were either homozygous or compound heterozygous for mutations in ABCA1 (Table 4) had mean plasma lipid values as follows: HDL-C, 8 mg/dl; apoA-I, 26 mg/dl; nonHDL-C, 102 mg/dl; direct LDL-C, 79 mg/dl; and TG, 182 mg/dl, with an HDL-C/apoA-I ratio of 0.37. Of these subjects, 29% had evidence of ASCVD.
Subjects ABCA1-08 and ABCA1-09 were double heterozygous, having one mutation in ABCA1 and the other in LCAT and APOA1, respectively (Table 4). They were men, aged 35 and 38 years, with no evidence of ASCVD and with the following respective values: HDL-C, 16 and 17 mg/dl; apoA-I, 79 and 95 mg/dl; non-HDL-C, 164 and 116 mg/dl; direct LDL-C, 136 and 75 mg/dl; and TG, 111 and 260 mg/dl. Their HDL-C/ apoA-I ratios were 0.20 and 0.18, respectively. All 9 ABCA1 homozygotes, compound heterozygotes, and double heterozygotes had a marked deficiency of apoA-I in very large 1 and large 2 HDL (<10% of normal; data not shown). Table 5 shows that there were 20 men and 10 women heterozygous for ABCA1 (includes 13 novel mutations) with a mean age of 53 years and mean values as follows: HDL-C, 15 mg/dl; apoA-I, 68 mg/dl; non-HDL-C, 143 mg/dl; Novel mutation (n = 5). direct LDL-C, 107 mg/dl; and TG, 227 mg/dl, with an HDL-C/apoA-I ratio of 0.22. Among these 30 subjects, 15 (50.0%) had evidence of ASCVD (11 men, 4 women; mean age of disease onset: 58 years); none had evidence of neuropathy, splenomegaly, or anemia. ABCA1-20 and ABCA1-25 were also found to be heterozygous for LCAT rs4986970 (see below), while ABCA1-12 and ABCA1-29 were also heterozygous for LPL rs268 (see below). Data regarding the study subjects having only ABCA1 variant rs9282541 (p.R230C) and/or rs111292742 (c.-279C>G) and no ABCA1, APOA1, or LCAT mutations are provided in Table 6. An 80-year-old Hispanic man (SNP-01) and a 34-year-old unrelated Hispanic woman (SNP-02) were found to be homozygous for the ABCA1 rs9282541 variant, which has been reported in the Mexican population to be associated with low HDL-C levels (27). Their respective values were as follows: HDL-C, 7 and 19 mg/dl; apoA-I, 44 and 75 mg/dl; non-HDL-C, 108 and 120 mg/dl; direct LDL-C, 58 and 91 mg/dl; and TG, 155 and 204 mg/dl. SNP-01 also had evidence of ASCVD (postmyocardial infarction at the age of 71 years). SNP-03, SNP-04, and SNP-05 were heterozygous for the ABCA1 rs9282541 variant. None of these subjects had ASCVD. Their mean values were as follows: HDL-C, 15 mg/dl; apoA-I, 66 mg/dl; non-HDL-C, 149 mg/dl; direct LDL-C, 105 mg/dl; and TG, 244 mg/dl. This allele was found in 1.74% of our subjects versus 1.10% in the global population (not significant).
The ABCA1 rs111292742 (ABCA1 c.-279C>G) variant was found in 10 subjects, one of whom (SNP-31) was also heterozygous for LPL rs268. ABCA1 rs111292742 has previously been associated with low HDL-C and hypertension in a Japanese population, while LPL rs268 has previously been reported to be associated with hypertriglyceridemia and low HDL-C levels (28,29). These 10 subjects (all men; mean age: 51 years) had the following mean values: HDL-C, 16 mg/dl; apoA-I, 86 mg/dl; non-HDL-C, 143 mg/dl; direct LDL-C, 92 mg/dl; and TG, 318 mg/dl, with an HDL-C/apoA-I ratio of 0.19. The ABCA1 rs111292742 allele was found in 3.48% of our subjects versus 2.46% in the global population (not significant). Evidence of ASCVD was noted in 3 of the 15 subjects with ABCA1 variants (SNP-01, SNP-06, and SNP-10). Overall, 20 of the 54 subjects (37.0%) with ABCA1 mutations or variants had evidence of ASCVD (mean age of disease onset: 59 years), significantly higher than the prevalence (6.4%) observed in our mutationnegative group (P < 0.05). Three of these subjects had evidence of neuropathy: ABCA1-01 with right-arm neuropathy, and ABCA1 SNP-01 and SNP-10 both with numbness and tingling in their lower extremities. SNP-01 and SNP-10 also had diabetes mellitus, which is well-known to cause neuropathy.

LCAT mutations and variants
LCAT mutations or selected variants were found in 25 of the subjects (12.4%); 4 of the mutations were novel ( Table 7). Subject LCAT-01, a 52-year-old man, was found to have a homozygous LCAT c.950T>G (p.M317R) mutation with the following values: HDL-C, 7 mg/dl; apoA-I, 44 mg/dl; non-HDL-C 80 mg/dl; direct LDL-C, <5 mg/dl; and TGs, 145 mg/dl, with an HDL-C/apoA-I ratio of 0.16. The direct LDL-C value of <5 mg/dl was below the detectable limits of the direct LDL-C assay, presumably due to the lack of a cholesteryl ester core in the LDL particles. As reported previously, this subject had marked corneal opacification and renal insufficiency, subsequently requiring hemodialysis (30). He was the first lecithin-cholesterol acyltransferase (LCAT)-deficient patient to receive LCAT enzyme replacement therapy, which transiently stabilized his renal disease over approximately a 4-month period, until such therapy was no longer available (31). HDL particle analysis indicated that his apoA-I was mainly in very small pre-1 and small 4 HDL, with irregularly sized and poorly lipidated large HDL particles (30). After enzyme replacement, his apoA-I-containing particle distribution was transiently normalized (31). He had no evidence of ASCVD.
Three subjects were found to be LCAT compound heterozygotes. LCAT-02 and LCAT-03, with heterozygous LCAT c.110C>T, p.T37M and LCAT c.101C>T, p.P34L mutations, respectively, were siblings and have been previously reported (32). They were aged 23 and 26 years with the following respective values: HDL-C, 6 and 3 mg/dl; apoA-I, 36 and 14 mg/dl; non-HDL-C, 190 and 214 mg/dl; direct LDL-C, 157 and 156 mg/dl; and TGs, 158 and 290 mg/dl, with HDL-C/apoA-I ratios of 0.17 and 0.21. They presented with peripheral corneal opacification, normal kidney function, and no evidence of ASCVD (32). However, because of their elevated LDL-C and a strong family history of ASCVD, they were placed on statin therapy. LCAT-02 had an excellent response to the therapy; her LDL-C concentration decreased from 157 to 62 mg/dl (32). The diagnosis of  the siblings was consistent with fish-eye disease (FED) (32). LCAT-04, the third compound heterozygote (LCAT c.491G>A, p.R164H and LCAT c.382G>A, p.G128S), was a 69-year-old man with the following values: HDL-C, 13 mg/dl; apoA-I, 79 mg/dl; non-HDL-C, 96 mg/dl; direct LDL-C, 54 mg/dl; and TGs, 216 mg/dl, with a HDL-C/apoA-I ratio of 0.16. He had no evidence of ASCVD or kidney disease but did have mild corneal opacification. As also shown in Table 7, heterozygous LCAT mutations were noted in 13 subjects (9 men and 4 women; mean age: 49 years), with the following mean values: HDL-C, 17 mg/dl; apoA-I, 85 mg/dl; non-HDL-C, 150 mg/dl; direct LDL-C, 113 mg/dl; and TGs, 221 mg/dl, with an HDL-C/apoA-I ratio of 0.20. One of these subjects (LCAT-08) was also homozygous for ABCA1 rs111292742, while another (LCAT-15) was heterozygous for a previously reported LPL mutation (29). None of these subjects had ASCVD or kidney failure, and none was reported to have corneal opacification. Eight additional subjects (4 men, 4 women; mean age: 48 years) were found to be heterozygotes for the LCAT rs4986970 (p.S232T) variant (Table 6). Their mean values were as follows: HDL-C, 12 mg/dl; apoA-I, 61 mg/dl; non-HDL-C, 162 mg/dl; direct LDL-C, 115 mg/dl; and TGs, 241 mg/dl. None of these 8 subjects was noted to have kidney disease, and only subject SNP-22, aged 76 years, was noted to have ASCVD (age of onset of 75 years with transient ischemic attacks). The LCAT rs4986970 variant has previously been associated with low HDL-C values in an Iranian population (33). We noted that the minor allele frequency for this variant in our severely HDL-deficient study population was 2.74% compared with a reported population frequency of 1.76% (not significant), but we had no data regarding Iranian descent.

APOA1 mutations and variants
Mutations in the APOA1 gene were found in 10 subjects (6 men, 4 women; 5.0% of the total study population) ( Table 8). APOA1-01, a 30-year-old man, with an HDL-C of 2 mg/dl and an apoA-I of 15 mg/dl, was found to be homozygous for the novel APOA1 c.409G>T, p.E137* mutation and the ABCA1 rs111292742 SNP. He had a truncated form of apoA-I and was asymptomatic, with no evidence of ASCVD. The remaining 9 subjects (mean age: 59 years) had heterozygous APOA1 mutations and the following mean values: HDL-C, 18 mg/dl; apoA-I, 96 mg/dl; non-HDL-C, 127 mg/dl; direct LDL-C, 90 mg/dl; and TGs, 260 mg/dl, with an HDL-C/apoA-I ratio of 0.20. Subjects APOA1-02, APOA1-03, and APOA-04 were from a kindred with the APOA1 c.718C>T (p.Q240*) mutation, which also results in a truncated form of apoA-I. They all had ASCVD, as previously reported (34). APOA1-05, a 67-yearold woman who was heterozygous for APOA1 c.517C>A (p.R149S), had no evidence of ASCVD, but she did have a positive family history. In addition to the HDL deficiency, her plasma had a decreased ability to esterify cholesterol. She and her kindred have been previously reported (35). An APOA1 in-frame deletion c.391_393delAAG (p.K131del) was found in four subjects, one of whom (APOA1-07), a man aged 71 years, had ASCVD. Subject APOA1-10 with the APOA1 c.178T>G, p.S60A mutation had no evidence of ASCVD. Therefore, in our study population of 201 subjects with severe HDL deficiency, we found 10 subjects with Novel mutation (n = 1) .
APOA1 mutations that were mainly either truncations or in-frame deletions. Of these 10 subjects, 4 (40%) were noted to have ASCVD (3 men, 1 women; mean age of disease onset: 62 years). Table 6 indicates that 8 subjects with no other mutations (7 men, 1 woman; 4.0%) had the LPL rs268 (p.N318S) variant, one of whom (SNP-23) was homozygous. These subjects had a mean age of 52 years and the following mean values: HDL-C, 16 mg/dl; apoA-I, 85 mg/dl; non-HDL-C, 155 mg/dl; direct LDL-C, 107 mg/dl; and TGs, 299 mg/dl. SNP-25 was a 78-year-old woman who developed ASCVD at the age of 71 years (coronary angioplasty and stent placement). SNP-31, who was heterozygous for LPL rs268 and ABCA1 rs111292742 (as noted above), was a 63-year-old man with no evidence of ASCVD. He had the following values: HDL, 6 mg/dl; apoA-I, 35 mg/dl; non-HDL-C, 199 mg/dl; direct LDL-C, 116 mg/dl; and TGs, 407 mg/dl, with an HDL-C/apoA-I ratio of 0.16. The LPL rs268 variant has previously been reported to be associated with hypertriglyceridemia and low HDL-C levels (29). In total, 11 subjects, including the 3 subjects who also had a mutation or another variant (SNP-31, ABCA1-12, and ABCA1-29), carried the LPL rs268 variant. The frequency of this allele was significantly more prevalent in our study population (2.99%) than in the reported global population (1.29%; P < 0.01).

Other lipid-related mutations
As shown in Table 9, 10 subjects with severe HDL deficiency had mutations in genes related to monogenic lipid disorders characterized by elevated LDL-C and/or TG concentrations and/or by sitosterolemia. OTHER-01, found to be homozygous for an ABCG8 mutation, had markedly elevated -sitosterol levels, consistent with sitosterolemia (36). Two subjects (OTHER-02 and OTHER-03) were compound heterozygous for LDLR mutations and APOB mutations, and five subjects were heterozygous for LDLR or APOB mutations. Some of these subjects had LDL-C values >190 mg/dl, and those that did not were aggressively being treated with lipid-lowering agents (combination statin/ ezetimibe). OTHER-05, aged 65 years, had established ASCVD (age of onset: 60 years) and was being aggressively managed with lipid-lowering therapy. OTHER-09, a 62-year-old man who was heterozygous for an LIPC mutation resulting in an amino acid substitution in hepatic lipase (p.T405M), had a TG concentration of 483 mg/dl. He had evidence of ASCVD, with angioplasty and stent placement at the age of 61 years. OTHER-10, a 46-year-old man who was heterozygous for the LPL p.T45N mutation, had moderate TG elevation (284 mg/dl) and no evidence of ASCVD.

ASCVD prevalence rates
The ASCVD prevalence in the 107 subjects (72% men) without secondary causes and for whom no DNA was available was 4.7%, with a mean age of onset of 61 years. The ASCVD prevalence in the 94 subjects (73% men) without secondary causes who did have DNA sequencing and in whom no mutations were found was 6.4%, with a mean age of onset of 64 years. Therefore, in these 201 pooled subjects with severe HDL deficiency with no secondary causes (72% men), the prevalence of ASCVD was 5.5%, with a mean age of onset of 62 years. In the 25 subjects with severe HDL deficiency with LCAT mutations or variants, only 1 subject, a 76-year woman, developed transient ischemic attacks (age of onset: 75 years). In this subject group the ASCVD prevalence was 4.0%. In contrast, in the 54 subjects with severe HDL deficiency and ABCA1 mutations or variants, the ASCVD prevalence was 37.0%, with a mean age of disease onset of 59 years (significantly higher than in the mutation-negative group; P < 0.05). Similarly, in the subjects with severe HDL deficiency and APOA1 mutations or variants, the ASCVD prevalence was 40%, with a mean age of disease onset of 62 years. Pooling the ABCA1 and APOA1 subjects (n = 64), the ASCVD prevalence was 37.5%, with a mean age of disease onset of 60 years. The ASCVD prevalence in these 64 subjects was 7-fold greater than in the 201 subjects with severe HDL deficiency who were not found to have any mutations or were never sequenced (5.5%; P < 0.01).

DISCUSSION
The first form of severe HDL deficiency to be described was Tangier disease by Fredrickson in 1964 (37). Homozygotes with the disease, named after the Chesapeake Bay island where the first two cases originated, were noted to have diffuse cholesteryl ester accumulation in macrophages in their tonsils, liver, spleen, and other tissues, as well as rapid catabolism of their HDL proteins, apoA-I, and apoA-II (38)(39)(40). Homozygotes were found to have HDL-C levels <10 mg/dl, moderate hypertriglyceridemia, decreased LDL-C, occasional peripheral neuropathy, mild corneal opacification as seen on slit-lamp examination, and an increased risk of premature ASCVD (41)(42)(43)(44). The defect in these patients was found to be an inability to efflux cholesterol onto HDL particles from their cells due to defects in the ABCA1 gene and its gene product ABCA1 (45-52). On HDL particle analysis, homozygotes were found to have apoA-I mainly in pre-1 HDL (53). Heterozygotes were found to have a 50% reduction in cellular cholesterol efflux, very low levels of very large 1 and large 2 HDL particles, and approximately 50% of normal HDL-C levels (53,54).
Population studies indicated that three common ABCA1 variants (p.G596A, p.A2589G, and p.G3456C) were more common in men with low HDL-C (<40 mg/dl) and ASCVD than in control subjects (55). Cohen et al. (56) sequenced the APOA1, ABCA1, and LCAT genes in 284 subjects with HDL-C concentrations below the 5th percentile (<30 mg/dl) and in 236 subjects with HDL-C levels above the 95th percentile (>75 mg/dl). They noted that nonsynonymous mutations or variants were significantly more common in subjects with low HDL-C than in those with high HDL-C (12% vs. 2%). They found that 9.9% of the subjects with low HDL-C had ABCA1 mutations, 2.2% had LCAT mutations, and 0.2% had APOA1 mutations. Therefore, mutations in ABCA1 were more common in patients with low HDL-C than mutations in LCAT or APOA1, in agreement with our findings; however, no clinical or other laboratory information was provided in their study (56).
Fricke-Schmidt et al. (57,58) did ABCA1 genotyping in 9,259 Danish subjects followed for 25 years and noted that 5 of 6 SNPs (ABCA1 p.V771M, p.V825I, p.I883M, p.E1172D, and p.R1587K) predicted increased risk of ASCVD, with ABCA1 p.V771M, p.I883M, and p.E1172D being the most important risk predictors. They compared ASCVD risk in heterozygotes versus noncarriers for four ABCA1 variants (p.P1065S, p.G1216V, p.N1800H, and p.R2144X) and noted that these variants reduced HDL-C levels by 17 mg/dl, associated with an adjusted hazard ratio for ASCVD of 1.70 (P < 0.001) (57,58). Akao et al. (59) assessed the ABCA1 p.R219K variant in 5,414 subjects in a statin trial for the elderly and noted an increased ASCVD event rate for those with the variant.
Abdel-Razek et al. (60) assessed potential differences in clinical and subclinical atherosclerosis (history of ASCVD or positive cardiac calcium score) and plasma cholesterol efflux capacity in 72 patients (mean age: 54 years) with HDL-C concentrations below the 10th percentile based on whether they had ABCA1 or APOA1 mutations. Disease-causing or likely pathogenic ABCA1 mutations were found in 13 patients, and likely pathogenic APOA1 mutations were found in 3 patients (n = 16; 22%), with 83% of these patients having evidence of atherosclerosis compared with 38.6% in patients with low HDL-C without such mutations or variants (P < 0.01). Patients with mutations also had lower cellular cholesterol efflux capacity (60). Dron et al. (61) recently reported 4 patients, aged 34-59 years, with low HDL-C associated with largescale deletions within the ABCA1 gene. Three of these patients had some evidence of ASCVD. The above data, like ours, indicate an increased risk of ASCVD in patients with low HDL-C and ABCA1 mutations. The next form of severe HDL deficiency, known as familial LCAT deficiency (FLD), was first described in 1967 by Norum and Gjone (62). These patients presented with marked corneal opacification, hyperlipidemia, anemia, proteinuria, and marked HDL deficiency (62). In contrast to normal subjects, almost all plasma cholesterol in these patients was unesterified due to a deficiency in LCAT activity. FLD patients were subsequently noted to have homozygous or compound heterozygous mutations in the LCAT gene and to develop kidney failure in the fourth or fifth decade of life (1,62). We have documented that FLD homozygotes generally have apoA-I in plasma present only in very small pre-1 and small 4 discoidal HDL particles (63). When the LCAT-deficient homozygote in the present study was treated with LCAT enzyme replacement therapy, there was transient appearance of -HDL particles in the normal HDL-size range and stabilization of the patient's compromised kidney function, as reported previously (29). However, the patient ended up on dialysis when enzyme replacement was no longer available. In our view, the treatment of choice in homozygous FLD patients will be enzyme replacement or gene therapy.
A variant of FLD known as FED was first described by Carlson and Philipson in 1979 in a Norwegian man and his three daughters, all of whom had marked corneal opacification (64). They had normal concentrations of total cholesterol but elevated concentrations of TG, VLDL cholesterol, and LDL-C and marked HDL deficiency (1,64). The disease was characterized further as being associated with ASCVD in later life, visual impairment, and dense corneal opacification. The two compound heterozygotes we report here as part of this study have previously been reported (32). In our view, the treatment of choice in FED patients is to optimize their LDL-C levels with statin therapy to prevent ASCVD.
Haase et al. (65) examined the effects of low HDL-C on ASCVD causality in 10,281 Danish subjects based on the presence of an LCAT variant associated with low HDL-C (LCAT rs4986970, p.S208T; seen in 4% of their population). They found that low HDL-C was robustly associated with increased ASCVD risk, but genetically decreased HDL-C due to impaired LCAT activity was not (65). Similar observations were reported in a recent head-to-head comparison of FLD- and FED-associated LCAT mutations: FLD mutations were associated with decreased atherosclerosis, while FED mutations were associated with increased atherosclerosis (66). The difference appears to relate to the capacity of LCAT to modulate the cholesteryl ester content of apoB-containing lipoproteins. In FLD this capacity is lost. Overall, the data indicate that FLD causes kidney failure, while FED causes ASCVD (3,66). In our study we did not see an increased ASCVD prevalence in subjects with severe HDL deficiency associated with LCAT mutations.
A third form of severe HDL deficiency, familial apoA-I deficiency, was first described by Schaefer et al. (67,68) in 1982 in a woman with severe HDL deficiency, undetectable plasma apoA-I, and death at the age of 42 years due to severe premature ASCVD. In 1983, Karathanasis, Zannis, and Breslow (69) became the first investigators to clone the APO1 gene. This kindred was subsequently noted by Ordovas et al. (70) to be affected with a large deletion of the entire APOA1/C3/A4 gene complex, with heterozygotes having about 50% of normal HDL-C, apoA-I, apoC-III, and apoA-IV levels in plasma. The disorder is now known as familial apoAI/CIII/AIV deficiency. In addition, in 1982 Norum et al. (71) described two sisters with severe HDL deficiency, undetectable plasma apoA-I and apoC-III levels, and premature ASCVD at ages 28 and 29 years. They were subsequently shown by Karathanasis,Ferris,and Haddad (72) to have a DNA rearrangement affecting the adjacent APOA1 and APOC3 genes. This disorder is now known as familial apoAI/CIII deficiency (69). Both forms of severe HDL deficiency are characterized by low TG and normal LDL-C concentrations.
Familial apoA-I deficiency (undetectable plasma apoA-I) with premature ASCVD due to an APOA1 codon 84 nonsense mutation was first described by Matsunaga et al. (73) in 1991. Other investigators have subsequently described patients with premature ASCVD that had this disorder caused by a homozygous codon 2 nonsense mutation (74,75). Haase et al. (76) sequenced the APOA1 gene in 10,330 subjects and found that 0.27% of 10,330 subjects were heterozygous for APOA1 nonsynonymous variants associated with reductions in apoA-I and/or HDL-C levels and a hazard ratio of 1.72 (P < 0.01) for myocardial infarction, largely driven by the APOA1 p.A164S variant. The data presented herein are in agreement with prior studies indicating that mutations in APOA1 are less common than those in ABCA1 and LCAT in patients with HDL deficiency and are associated with premature ASCVD.
Genetic variations at the LPL gene locus are often associated with both hypertriglyceridemia and low HDL-C levels. The frequencies of three LPL polymorphisms (p.D9N, p.N291S, and p.S447X) were noted to be increased in men with low HDL-C levels (<40 mg/dl) and ASCVD versus control subjects (77), with the LPL N9 and S291 alleles, which raise TG levels, being more frequent, and the S477X allele, which lowers TG levels, being less frequent in the low HDL-C subjects than in control subjects (65). In the present study we observed that the LPL p.N318S allele occurred in 3.0% of our severely HDL-deficient population compared with a reported population frequency of 1.3%. We have carried out extensive genotyping (>1,000 SNPs) in 699 men with low HDL-C (<40 mg/dl) and ASCVD and 705 age-matched ASCVD-free men (78). After adjusting for multiple testing within each gene, SNPs at gene loci significantly associated with ASCVD and low HDL-C were 1) LIPC (hepatic lipase) rs4775065 (P < 0.0001 vs. control subjects), 2) CETP rs5882 (P = 0.0002), 3) ABCA1 rs2249891 (P = 0.0126), and 4) CUBN rs7893395 (P = 0.0246) (78). Dron et al. (24) examined the effects of polygenic determinants on extremes of HDL-C in three population cohorts (Western University Lipid Genetics Clinic, Montreal Heart Institute Biobank, and the University of Pennsylvania) that included a total of 686 subjects with low HDL-C levels (30.9 mg/dl in men and 38.7 mg/dl in women) and 1,165 subjects with high HDL-C (54.1 mg/dl in men and 69.6 mg/dl in women). The prevalence of CVD by history was highest in the Montreal Heart Institute Biobank cohort (ranging from 31.3% to 67.1%) and lowest in the University of Pennsylvania group (ranging from 4.1% to 12.9%) but was always highest in the low HDL groups (24). They reported that 18.7% of the low HDL group had rare heterozygous large-effect variants, mainly in the ABCA1, LCAT, and APOA1 genes, with an additional 12.8% having an extreme polygenic trait score (24). The nine SNPs used in their polygenic trait score in order of effect size on HDL-C were CETP rs3764261, LIPC rs1532085, ABCA1 rs1883025, SCARB1 rs838880, GALNT2 rs48469914, MVK rs7134594, CMIP rs2925979, ZNF648 rs1689800, and ANGPTL4 rs7255436, with an r 2 value of 0.8996 (P < 0.0001) (24).
The data presented herein, as well as prior data based on DNA sequencing, would indicate that HDL deficiency due to mutations or variants at the ABCA1 or APOA1 gene locus are associated with premature ASCVD . There may be, however, significant heterogeneity with regard to ASCVD risk in such patients depending on their LDL-C levels. In patients with Tangier disease and hypersplenism, there may be markedly increased catabolism of LDL apoB with very low LDL-C levels, which may protect such patients from ASCVD, in contrast to Tangier patients with normal LDL-C (3). There may also be heterogeneity in patients with HDL deficiency due to APOA1 variants, with variants associated with decreased LCAT activation, such as ApoA-I Milano , having no increased ASCVD risk, while other variants that do not affect LCAT activation having increased ASCVD risk (3).
As indicated earlier, we consider the HDL Mendelian randomization studies by Voight et al.  (1,2). Both studies did see significant relationships between the intronic variants APOA1 rs6589566 and APOA1 rs10790162 and ASCVD risk (1, 2); however, the APOA1 variants were excluded from their analysis because of effects on other lipoproteins. Voight et al. (1) also relied almost entirely on the LIPC variant rs61755018 (p.N396S), which had no significant relationship with ASCVD. In our view, relying on intronic SNPs to assess the genetic contributions of a given gene to disease causation is flawed. Instead, investigators are relying increasingly on high-throughput next-generation DNA sequencing, which allows for more precise assessments of genetic variation at any given gene locus with disease, including ASCVD.
It should be stated that our study has its limitations in that the number of subjects that were sequenced was relatively small, and the reference laboratory population may not be representative of the general population. In our view, the measurement of HDL-C and apoA-I levels is helpful in identifying HDL-deficient patients, and the quantification of the distribution of apoA-I in HDL subpopulations by two-dimensional gel electrophoresis and immunoblotting analysis is helpful in making the diagnosis (3). Nevertheless, the definitive diagnosis of the various forms of severe HDL deficiency must rely on DNA sequencing of the ABCA1, LCAT, and APOA1 genes (3). Another limitation of our study is the lack of data concerning the known polygenic causes of HDL deficiency, as documented recently by Dron et al (24).
The data presented herein are consistent with the following concepts with regard to severe HDL deficiency (HDL-C <20 mg/dl): this condition is 1) uncommon in the general population and is significantly more frequent in men than in women; 2) frequently associated with severe hypertriglyceridemia, increased inflammation, diabetes, and in men with testosterone replacement; 3) most commonly caused by mutations in ABCA1, which can increase ASCVD risk depending on the underlying LDL-C levels; 4) may be caused by mutations in LCAT that can lead to renal failure; and 5) least commonly caused by mutations in APOA1 that can lead to premature ASCVD.