High-dose atorvastatin causes a rapid sustained increase in human serum PCSK9 and disrupts its correlation with LDL cholesterol.

Proprotein convertase subtilisin kexin type 9 (PCSK9) is a key regulator of serum LDL-cholesterol (LDL-C) levels. PCSK9 is secreted by the liver into the plasma and binds the hepatic LDL receptor (LDLR), causing its subsequent degradation. We first demonstrated that a moderate dose of atorvastatin (40 mg) increases PCSK9 serum levels, suggesting why increasing statin doses may have diminished efficacy with regard to further LDL-C lowering. Since that initial observation, at least two other groups have reported statin-induced PCSK9 increases. To date, no analysis of the effect of high-dose atorvastatin (80 mg) on PCSK9 over time has been conducted. Therefore, we studied the time course of atorvastatin (80 mg) in human subjects. We measured PCSK9 and lipid levels during a 2-week lead-in baseline period and every 4 weeks thereafter for 16 weeks. We observed that atorvastatin (80 mg) caused a rapid 47% increase in serum PCSK9 at 4 weeks that was sustained throughout 16 weeks of dosing. Importantly, while PCSK9 levels were highly correlated with total cholesterol (TC), LDL-C, and triglyceride (TG) levels at baseline, atorvastatin (80 mg) completely abolished all of these correlations. Together, these results further suggest an explanation for why increasing doses of statins fail to achieve proportional LDL-C lowering.

relevant interaction with atorvastatin. A total of 84 eligible subjects went through a run in control period of two weeks where no treatment was administered, followed by initiation of atorvastatin 80 mg taken once daily. Subjects were seen every 4 weeks for a total of 16 weeks on high-dose atorvastatin. Subjects were assessed for compliance and adverse effects and had research labs drawn for PCSK9 evaluation at each visit. Routine lipid levels (including total cholesterol, LDL-C, HDL-C, and TG) were assessed following a minimal 8-12 h fast at baseline and after 8 and 16 weeks on atorvastatin. PCSK9 levels were assessed at 2 weeks prior to dosing, 0-week, 4-week, 8-week, 12-week, and 16-week visits. The two PCSK9 results for each subject at the minus 2-week visit and 0-week visit (both prior to the start of atorvastatin treatment) were averaged to determine a baseline PCSK9 level. All visits were conducted at the University of Florida General Clinical Research Center (GCRC). All subjects provided written, informed consent, and the study was approved by both the University of Florida's Institutional Review Board and the GCRC Scientifi c Advisory Committee (clinicaltrials.gov identifi er NCT00361283).

PCSK9 ELISA
PCSK9 levels in the serum samples were measured using our recently described PCSK9 dual monoclonal antibody sandwich ELISA ( 37,38,42 ) with minor modifi cations, including the use of a non-His-tagged recombinant PCSK9 standard. The exact epitopes recognized by the antibodies used in the ELISA are not known at this time. Human PCSK9 used as a standard in the ELISA was cloned from a human liver cDNA library, with a resulting construct used to generate an HEK293 stable cell line overexpressing PCSK9. The cDNA sequence used did not code for a His-tag. Cells were grown in serum-free media, and the secreted PCSK9 protein was purifi ed using an ion-exchange column followed by size-exclusion chromatography. Identity of the protein was confi rmed by N-terminal sequencing, and purity was judged to be greater than 95% based on SDS-PAGE followed by Coomassie blue staining. ELISA wells were coated overnight with anti-PCSK9 monoclonal antibody at a concentration of 5 g/ml. The following day, wells were aspirated, washed three times with assay buffer (50 mM HEPES, pH 7.40, 150 mM NaCl, 1% Triton X-100, 5 mM EDTA, 5 mM EGTA), and blocked for 1 h with TBScasein blocking buffer (Pierce). Next, 100 l of non-His-tagged recombinant PCSK9 standards (varying concentrations of recombinant protein in assay buffer) were added to the wells as a standard curve. Afterward, serum samples were diluted 1:20 in assay buffer, added to their respective wells, and the ELISA plate was incubated for 2 h at room temperature. Following aspiration, wells were washed three times with assay buffer, and 100 l of a 1:1000 dilution of conjugate antibody (HRP-labeled anti-PCSK9 monoclonal antibody, 1 mg/ml) were added to the wells for a 1 h incubation at room temperature. Following aspiration, wells were washed three times with TBST. After the last aspiration of TBST, 100 l of TMB development substrate (Pierce) were added to the wells and allowed to incubate for 30 min at room temperature. The reaction was stopped with an equal volume of 2N phosphoric acid, and plates were read at 450 nm. SigmaPlot version 8.0 was used for fi tting of the calibration curves. Serum samples were shipped on dry ice and stored at Ϫ 70°C prior to analysis. Reproducibility of the ELISA on frozen serum samples was tested by looking at the effect of up to four freeze-thaw cycles on samples, with at least 90% recovery observed for all samples after four freeze-thaw cycles.

Immunoprecipitation and Western blotting of PCSK9
Analysis of PCSK9 levels in serum samples by immunoprecipitation and Western blotting was performed as previously described ( 37,38,42 ) with minor modifi cations. For each immunoprecipi-subject, a 49-year-old male, heterozygous for two monoallelic dominant negative PCSK9 mutations, has also been described and had an LDL-C of 16 mg/dl ( 33 ).
Statins, which are the class of drugs most widely prescribed to lower LDL-C levels, have been shown to increase the activity/nuclear translocation of sterol regulatory element-binding protein-2 (SREBP-2), a transcription factor that activates both the LDLR and PCSK9 genes (34)(35)(36). Statins were originally reported to increase PCSK9 mRNA expression (34)(35)(36), and our group hypothesized that statin treatment in humans should increase circulating PCSK9 protein levels. Utilizing a novel PCSK9 sandwich ELISA ( 37 ), we were the fi rst to demonstrate that atorvastatin (40 mg), the most widely prescribed statin, significantly increased PCSK9 serum levels after three months of treatment ( 38 ). Since that time, at least two other laboratories have also reported that patients on statins have increased levels of circulating PCSK9 ( 7,39 ).
These observations of statin-induced PCSK9 increases in patients offer insight into the mechanism for the nonlinear statin dose-response relationship. These observations, however, are limited in that the detailed time course of statin-induced PCSK9 increases has not been described. It is not known, for instance, if PCSK9 levels plateau during statin therapy or continue to increase over time. It is possible that PCSK9 levels might increase initially and then decrease as a new steady-state level of hepatic PCSK9 secretion is reached. It is also not understood with certainty how statin treatment affects the correlation of PCSK9 levels with LDL-C levels over time or if PCSK9 levels predict treatment response to statins. The answers to these questions are not currently known because previous studies have been limited by their observational nature or sample size. In addition, previous studies have not assessed the effects of changes in PCSK9 over time or their potential role in determining statin response. Nor have the highest doses of statins been tested. As a result, in the current study, we determined 1 ) the detailed time course of the effect of high-dose atorvastatin on PCSK9 levels; 2 ) the possibility that baseline PCSK9 levels may predict the magnitude of the atorvastatin-induced LDL-C response; 3 ) whether atorvastatin-induced changes in PCSK9 levels correlated with atorvastatin-induced decreases in LDL-C; and 4 ) the effect of high-dose atorvastatin on the relationship between PCSK9 levels and LDL-C and other serum lipids over time.

Clinical study protocol
Our clinical study design has been reported previously ( 40,41 ). Briefl y, subjects had to be at least 18 years of age and not require treatment for dyslipidemia based on ATP III guidelines. Subjects were excluded if they had a history of coronary disease or risk equivalents, were pregnant, or had liver transaminases greater than two times the upper limit of normal. In addition, subjects could not have a history of cholesterol medication use or concurrent medication or complementary medicine use that could alter cholesterol or infl ammation or that posed a clinically shows, with regard to TG levels, the baseline TG level was 96 ± 6 mg/dl, and atorvastatin treatment resulted in signifi cant decrease at 8 and 16 weeks (72 ± 4 mg/dl and 70 ± 5 mg/dl respectively, both P < 0.01 versus baseline). Fig. 2A demonstrates that 80 mg/day atorvastatin treatment caused a rapid and sustained increase in circulating PCSK9 levels. At baseline, PCSK9 levels were 97 ± 4 ng/ml. After only 4-weeks of atorvastatin treatment, PCSK9 levels had increased 47% to 143 ± 5 ng/ml ( P < 0.01 versus baseline). This signifi cant increase in PCSK9 levels was maintained at the 8-week, 12-week, and 16-week time points with PCSK9 levels of 140 ± 6 ng/ml, 143 ± 7 ng/ml, and 142 ± 6 ng/ml, respectively (all P < 0.01 versus baseline).
In light of a recent report of a furin breakdown product of PCSK9 protein present in plasma migrating as a band below the PCSK9 band ( 39 ), we further investigated the atorvastatin-induced increase in PCSK9. To do this, we performed immunoprecipitation and Western blotting analyses of some representative samples included in Fig.  2A, which still had adequate volume remaining for immunoprecipitation (at least 100 l). Results of these experiments are shown in Fig. 2B , which demonstrates that atorvastatin treatment resulted in increases in the intact PCSK9 protein band that comigrated with the recombinant PCSK9 protein standard as well as the PCSK9 propeptide band.
We next investigated the effect of atorvastatin treatment on the correlation of PCSK9 levels with serum lipid levels. Fig. 3A shows the correlation between PCSK9 and TC levels at baseline and after 16 weeks of treatment with 80 mg/ day atorvastatin. At baseline, PCSK9 levels were correlated with TC ( r = 0.48, P < 0.01); however, 16 weeks of atorvastatin treatment completely abolished the correlation between PCSK9 and TC ( r = 0.05, P = NS). Similar results were observed with regard to the effect of atorvastatin treatment on the correlation of PCSK9 with LDL-C levels ( Fig. 3B ) and TG ( Fig. 3C ) levels. At baseline, PCSK9 levels were correlated with LDL-C ( r = 0.38, P < 0.01). After 16 weeks of atorvastatin treatment, however, the correlation between PCSK9 and LDL-C levels was completely disrupted ( r = -0.01, P = NS). Likewise at baseline, PCSK9 levels were directly correlated with TG ( r = 0.27, P = 0.02), but 16 weeks of atorvastatin treatment abolished this direct correlation ( r = Ϫ 0.15, P = NS). With regard to HDL-C ( Fig.  3D ), there was no correlation of PCSK9 to HDL-C levels either at baseline ( r = 0.21, P = NS) or 16 weeks ( r = 0.19, P = NS).
In light of these results, we next examined the correlation of atorvastatin-induced changes in PCSK9 levels with changes in LDL-C levels to determine if the largest increases in PCSK9 levels predicted the largest decreases in LDL-C levels. Fig. 4A shows the correlation of percent changes in PCSK9 levels (from baseline to endpoint) to percent changes in LDL-C levels (from baseline to endpoint). Interestingly, there was a trend toward an inverse correlation, although this trend did not achieve statistical signifi cance ( r = Ϫ 0.21, P = 0.06). Next we compared the baseline PCSK9 level to the absolute change in LDL-C observed (from baseline to endpoint) to determine if base-tation, 100 l of serum were added to 900 l of immunoprecipitation buffer (50 mmol/l HEPES, pH 7.40, 150 mmol/l NaCl, 10 ml/l Triton X-100, 5 mmol/l EDTA, 5 mmol/l EGTA). Next, PCSK9 was immunoprecipitated overnight with 1 g of anti-PCSK9 monoclonal antibody covalently coupled to trisacryl beads (Pierce). Afterward, beads were washed twice with immunoprecipitation buffer, and 40 l of 2× sample buffer (100 mmol/l Tris, pH 6.80, 40 g/l SDS, 200 ml/l glycerol, 20 mg/l bromophenol blue, 15 g/l dithiothreitol) were added to each tube. Samples were vortexed, boiled for 5 min, and stored at Ϫ 20°C prior to electrophoretic analysis. Western blotting was performed using SDS-polyacrylamide gels, with colored molecular weight markers (Invitrogen) run on each gel. Proteins were separated for 1.5 h at 175 V at room temperature and transferred to ECL nitrocellulose paper (Amersham) for 2 h (100 V, 4°C). Nitrocellulose blots were blocked for 1 h at room temperature in TBS-casein blocking buffer (Pierce) containing 1 ml Tween 20/L. After blocking, blots were probed with sheep polyclonal anti-human PCSK9 antibody in blocking buffer for 1 h at room temperature. Blots were washed three times (10 min each) with TBST (10 mmol/l Tris pH 7.40, 150 mmol/l NaCl, with 1 ml Tween 20/L). After washing, blots were probed with an HRP-labeled anti-sheep IgG for 1 h at room temperature. Following a fi nal three washes with TBST, blots were developed with ECL reagent (Amersham), air-dried, and exposed to Bio-Max X-ray fi lm (Kodak).  Fig. 1A demonstrates the effect of atorvastatin, 80 mg per day for 16 weeks, on serum TC levels. At baseline, the TC level was 180 ± 4 mg/dl. Following 8 weeks of atorvastatin treatment, there was a signifi cant decrease in TC compared with baseline (117 ± 3 mg/dl, P < 0.01 versus baseline). This decrease in TC was sustained at the 16week time point (118 ± 3 mg/dl, P < 0.01 versus baseline). As Fig. 1B demonstrates, atorvastatin also had a similar effect on LDL-C levels. At baseline, the LDL-C level was 101 ± 4 mg/dl. Following 8 weeks of atorvastatin treatment, there was a signifi cant decrease in LDL-C compared with baseline (45 ± 2 mg/dl, P < 0.01 versus baseline), and this decrease in LDL-C was sustained at the 16-week time point (45 ± 2 mg/dl, P < 0.01 versus baseline). In contrast, atorvastatin had no effect on HDL-C levels with baseline, 8-week, and 16-week HDL-C being 60 ± 2 mg/dl, 58 ± 2 mg/dl, and 58 ± 2 mg/dl, respectively ( Fig. 1C ). As Fig. 1D achieve statistical signifi cance. Finally, we compared atorvastatin-induced changes in PCSK9 levels to the fi nal LDL-C achieved after 16 weeks of treatment with atorvastatin. In this case, there was a signifi cant negative correlation line PCSK9 levels might predict who would respond most robustly to atorvastatin treatment. Again, as Fig. 4B demonstrates, there was a trend toward an inverse correlation ( r = Ϫ 0.20, P = 0.08), although this trend also did not  increased serum PCSK9 levels after 12 weeks. In the same study, baseline PCSK9 and LDL-C concentration were signifi cantly correlated, and the correlation was disrupted by atorvastatin treatment. Following this report, Mayne et al. reported that administration of atorvastatin signifi cantly increased PCSK9 levels while lowering LDL-C levels ( 43 ). Cariou et al. showed that, in patients with diabetes, statin treatment increased PCSK9 levels by 32% ( 44 ). In the same study, it was also shown that the correlation between PCSK9 levels and LDL-C levels was lost following statin treatment as statins increased PCSK9 levels and decreased LDL-C levels ( 44 ). Dubuc et al. ( 39 ) demonstrated that patients treated with statins had a 45% increase in circulating PCSK9 levels and that patients treated with a statinezetimibe combination had a 77% increase in PCSK9. Lakoski et al. ( 7 ) demonstrated that PCSK9 levels correlated directly with LDL-C levels in a large, ethnically diverse population and that statin treatment was associated with a signifi cant increase in circulating PCSK9 levels in both men and women.

RESULTS
As a result of these fi ndings, it was not unexpected that a high dose of atorvastatin (80 mg) would cause the baseline correlation of PCSK9 to LDL-C to be lost in our current study, although it was important to confi rm this effect. The positive correlation between PCSK9 and LDL-C at ( r = Ϫ 0.25, P = 0.03), which indicated that greater increases in PCSK9 levels tended to be associated with lower endpoint LDL-C levels, although the correlation itself was still relatively modest and of marginal statistical signifi cance.

DISCUSSION
Our results demonstrate that high-dose atorvastatin treatment (80 mg per day) causes a rapid and sustained increase in circulating PCSK9 protein levels. After only 4 weeks of treatment, PCSK9 levels increased 47% over baseline levels, and this increase was sustained at 8-week, 12week, and 16-week time points. Similar to what we and others have previously reported ( 7,29,37,38 ), baseline PCSK9 levels were highly correlated with TC and LDL-C levels. We further demonstrated a correlation between baseline PCSK9 and TG levels. After 16 weeks of atorvastatin treatment, however, these correlations were abolished, indicating that treatment with atorvastatin had completely disrupted the relationship between PCSK9 and these lipid parameters.
Several researchers have described statin-induced increases in PCSK9 levels in humans. In a previous smaller study ( 38 ), we observed that a lower dose of atorvastatin tein. Presumably, with each increased dose of a statin, the discrepancy in production rates of VLDL and PCSK9 would be further increased, and this makes it diffi cult to elucidate the signifi cance of the loss of the direct correlation between PCSK9 and TC, LDL-C, and TG that we observed.
One of the questions we wanted to address was whether baseline levels of PCSK9 might predict the magnitude of atorvastatin-induced decreases in LDL-C. We considered the possibility that subjects with the highest baseline PCSK9 levels might have the largest LDL-C responses to atorvastatin because serum PCSK9 levels were directly correlated with LDL-C, and atorvastatin acts to decrease LDL-C through increasing LDLR protein levels, in spite of increasing PCSK9 levels. Consistent with this hypothesis, there was a modest relationship between baseline PCSK9 levels and changes in LDL-C, with relatively higher baseline PCSK9 levels tending to be associated with numerically greater decreases in LDL-C. This correlation, however, did not achieve statistical signifi cance.
Another question that we wanted to address was whether atorvastatin-induced changes in PCSK9 levels would correlate with the magnitude of atorvastatin-induced LDL-C decreases. We expected that subjects who had the smallest atorvastatin-induced increases in serum PCSK9 levels might also have the most signifi cant atorvastatin-induced LDL-C lowering. This turned out, however, not to be the case. Rather, subjects that had the greatest increases in atorvastatin-induced PCSK9 levels also tended to have the largest atorvastatin-induced decreases in serum LDL-C, although this trend, similar to the previous one, did not reach statistical signifi cance. Interestingly, atorvastatininduced increases in PCSK9 were negatively correlated with endpoint LDL-C levels. This correlation was still relatively modest and just managed to achieve statistical signifi cance. In the case of all three of the above correlations, the fact that the trends either did not achieve statistical signifi cance or just achieved statistical signifi cance makes it diffi cult to draw defi nitive conclusions about the relationship between atorvastatin-induced changes in PCSK9 levels and LDL-C levels.
The loss in correlation may be in part due to the fact that statins increase the activity/nuclear translocation of sterol regulatory element-binding protein-2 (SREBP-2), a transcription factor that activates both the LDLR and PCSK9 genes (34)(35)(36). With PCSK9 and the LDLR thus both being increased by statin treatment, the amount of PCSK9 measured in the serum may not be totally refl ective of statin-induced increases in hepatic PCSK9 secretion because the increased amount of circulating PCSK9 also has more hepatic LDLR to bind to, which in turn removes it from the circulation. This aspect of the relationship between serum PCSK9 levels and hepatic LDLR levels makes it possible that statin-induced increases in serum PCSK9 levels may not refl ect the full effect to which statins modulate hepatic synthesis and secretion of PCSK9 protein.
Together, our data suggest that one possible explanation for why increasing doses of statins fail to achieve proportional LDL-C lowering may be due to statin-induced baseline exists under conditions where VLDL production is not inhibited by a statin. With each increasing dosage of a statin, a new steady state is reached in which hepatic LDLR expression is increased. Increased LDLR, in turn, can account for increased clearance of both LDL-C and PCSK9. It should be remembered, however, that while VLDL production in liver is decreased by the statin, production of PCSK9 is increased. Because of this, the statininduced loss of correlation between PCSK9 and LDL-C levels could occur even if PCSK9 did not affect LDLR pro- rapid and sustained increases in PCSK9 protein levels. For some time, it has been known that statins follow a rule of 6% in that whatever LDL-C reduction is achieved at a starting dose of a given statin is only improved upon an additional approximate 6% with each doubling of the dose. This important observation of non-dose-dependent response for commonly prescribed statins has led to much speculation about why statins should affect LDL-C levels in such a manner. Our current data, together with data that other laboratories have generated, suggest that statininduced increases in PCSK9 protein levels may account for the less than expected, incremental LDL-C level lowering from increasing doses of statins. This hypothesis, however, is unproven. To test this idea, future studies should prospectively test the effect of escalating doses of a statin on PCSK9 or test whether adding a pharmacological PCSK9 inhibitor to statin therapy results in additive or synergistic LDL-C lowering compared with statin monotherapy.
The authors thank Jayne Talbot for her support.