To read The Straight Dope on Cholesterol: 10 Things You Need to Know – Part 1 click here.
To start at the beginning of Peter’s 10-part series click here.
Because there is no experiment or clinical trial we can carry out to unambiguously resolve the exact role of inflammation in this process (and, by extension, the role of LDL particles), we rely on so-called “natural experiments.”
Such “natural experiments,” which help elucidate this role, are those of people with genetic alterations leading to elevated or reduced LDL-P. Let’s consider an example of each:
So why does having an LDL-P of 2,000 nmol/L (95th percentile) increase the risk of atherosclerosis relative to, say, 1,000 nmol/L (20th percentile)? In the end, it’s a probabilistic game. The more particles – NOT cholesterol molecules within the particles – you have, the more likely the chance a LDL-P is going to ding an endothelial cell, squeeze into the sub-endothelial space, and begin the process of atherosclerosis.
There are several examples of long-term studies examining the predictive power of LDL-C versus LDL-P with respect to cardiovascular disease. This study followed a Framingham offspring cohort of about 2,500 patients over a median time period of almost 15 years in each of the four possible groups (i.e., high-high, high-low, low-high, and low-low) and tracked event-free survival. In this analysis the cut-off points for LDL-P and LDL-C were the median population values of 1,414 nmol/L and131 mg/dL, respectively. So, “high” implies above these values; “low” implies below these values. Kaplan-Meier survival curves are displayed over a 16 year period – the steeper the slope of the line the worse the outcome (survival).
The same patterns observed in the other studies are observed here:
Interestingly the persons with the worst survival had low (below median) LDL-C but high LDL-P. The patients most likely to have high LDL-P with unremarkable or low LDL-C are those with either small LDL particles, or TG-rich/cholesterol poor LDL particles, or both (e.g., insulin resistant patients, metabolic syndrome patients, T2DM patients). This explains why small LDL particles, while no more atherogenic on a per particle basis than large particles (see point #9), are often a marker for something sinister.
Concordance is a statistical term that refers to variables that predict the same thing. Conversely, discordance refers to variables that do not predict the same thing. When LDL-C (what most doctors measure in your blood test) and LDL-P (what most doctors do not measure in your blood test) predict the same risk, they are said to be concordant. When they do not, they are said to be discordant. In the latter case, LDL-P is the one to follow.
To illustrate the prevalence of discordance between LDL-C and LDL-P, consider the figure below.
This figure shows data from patients with LDL-C between 100 and 118 mg/dL (i.e., second quartile of risk: 25th to 50th percentile) without metabolic syndrome (top) and with metabolic syndrome (bottom). In the patients without metabolic syndrome, LDL-C under-predicts cardiac risk 22% of the time. However, when you look at the patients with metabolic syndrome, you can see that 63% of the time their risk of cardiac disease is under-predicted.
The data above were collected from nearly 2,000 patients with diabetes who presented with “perfect” standard cholesterol numbers: LDL-C < 70 mg/dL; HDL-C > 40 mg/dL; TG <150 mg/dL. However, only in 22% of cases were their LDL-P concordant with LDL-C. That is, in only 22% of cases did these patients have an LDL-P level below 700 nmol/L.
Remember, LDL-C < 70 mg/dL is considered VERY low risk – the 5th percentile. Yet, by LDL-P, the real marker of risk, 35% of these patients had more than 1,000 nmol/L and 7% were high risk. When you do this analysis with the same group of patients stratified by less stringent LDL-C criteria (e.g., <100 mg/dL), the number of patients in the high risk group is even higher.
As a general rule, the more “metabolically deranged” an individual is, the greater this discordance between LDL-C and LDL-P, as shown in the figure below. The axis of this figure is adjusted so that the red bars and the blue bars should be of the same height when LDL-C and LDL-P are concordant.
Particle size is something many folks fixate on, and there is no doubt that smaller LDL particles are associated with greater risk. But, on a particle-by-particle basis are they, in fact, more atherogenic? Let’s find out.
This figure (one of the most famous in this debate) is from the Quebec Cardiovascular Study, published in 1997, in Circulation. You can find this study here.
This is kind of a complex graph if you’re not used to looking at these. It shows relative risk – but in 2 dimensions. It’s looking at the role of LDL size and apoB (a proxy for LDL-P). What seems clear is that in patients with low LDL-P (i.e., apoB < 120 mg/dl), size does not matter. The relative risk is 1.0 in both cases, regardless of peak LDL size. However, in patients with lots of LDL particles (i.e., apoB > 120 mg/dl), smaller peak LDL size seems to carry a much greater risk – 6.2X.
If you just looked at this figure, you might end up drawing the conclusion that both size and number are independently predictive of risk. Not an illogical conclusion…
What is not often mentioned, however, is what is in the text of the article:
“Among lipid, lipoprotein,and apolipoprotein variables, apo B [LDL-P] came out as the best and only significant predictor of ischemic heart disease (IHD) risk in multivariate stepwiselogistic analyses (P=.002).”
“LDL-PPD [peak LDL particle diameter] – as a continuous variable did not contribute to the risk of IHD after the contribution of apo B levels to IHD risk had been considered.”
What’s a continuous variable? Something like height or weight, where the possible values are infinite between a range. Contrast this with discrete variables like “tall” or “short,” where there are only two categories. For example, if I define “tall” as greater than 6 feet, the entire population of the world could be placed in two buckets: Those who are “short” (i.e., less than 6 feet tall) and those who are “tall” (i.e., those who are 6 feet tall and taller). This figure shows LDL size like it’s a discrete variable – “large” or “small” – but obviously it is not. It’s continuous, meaning it can take on any value, not just “large” or “small.” When this same analysis is done using LDL size as the continuous variable it is, the influence of size goes away, and only apoB (i.e., LDL-P) matters.
This effect has been observed subsequently, including the famous Multi-Ethnic Study of Atherosclerosis (MESA) trial, which you can read here. The MESA trial looked at the association between LDL-P, LDL-C, LDL size, IMT (intima-media thickness – the best non-invasive marker we have for atherosclerosis), and many other parameters in about 5,500 men and women over a several year period.
This study used the same sort of statistical analysis as the study above to parse out the real role of LDL-P versus particle size, as summarized in the table above.
This table shows us that when LDL-P is NOT taken into account (i.e., “unadjusted” analysis), an increase of one standard deviation in particle size is associated with 20.9 microns of LESS atherosclerosis, what one might expect if one believes particle size matters. Bigger particles…less atherosclerosis.
However, and this is the important part, when the authors adjusted for the number of LDL particles (in yellow), the same phenomenon was not observed. Now an increase in LDL particle size by 1 standard deviation was associated with an ADDITIONAL 14.5 microns of atherosclerosis, albeit of barely any significance (p=0.05).
Let me repeat this point: Once you account for LDL-P, the relationship of atherosclerosis to particle size is abolished (and may even trend towards moving in the “wrong” direction – i.e., bigger particles…more atherosclerosis).
LDL particles traffic not only cholesterol ester but also triglycerides. Each and every LDL particle has a variable number of cholesterol molecules which, because of constant particle remodeling, is constantly changing. In other words, of the several quadrillion LDL particles floating in your plasma, no two are carrying the exact same number of cholesterol molecules. It takes many more cholesterol-depleted LDL particles than cholesterol-rich LDL particles to traffic a given cholesterol mass (i.e., number of cholesterol molecules) per volume of plasma (i.e., per dL). Core cholesterol mass is related to both LDL particle size (the volume of a sphere is a third power of the radius – it can take 40-70% more small particles than large LDL particles to traffic a given cholesterol mass) and the number of TG molecules per LDL particle.
TG molecules are larger than cholesterol ester molecules, so as the number of TG molecules per particle increases, the number of cholesterol molecules will be less – in a very non-linear manner. Regardless of size, it takes many more TG-rich LDL particles (which are necessarily cholesterol-depleted) to traffic a given cholesterol mass than TG-poor LDL particles. The persons with the highest LDL particles typically (though not always) have small LDL particles that are TG-rich. These are incredibly cholesterol-depleted LDL particles.
Let’s start with what we know, then fill in the connections, with the goal of creating an eating strategy for those most interested in delaying the onset of cardiovascular disease.
There are several short-term studies that have carefully examined the impact of sugar, specifically, on cardiovascular risk markers. Let’s examine one of them closely. In 2011 Peter Havel and colleagues published a study titled Consumption of fructose and HFCS increases postprandial triglycerides, LDL-C, and apoB in young men and women. If you don’t have access to this journal, you can read the study here in pre-publication form. This was a randomized trial with 3 parallel arms (no cross-over). The 3 groups consumed an isocaloric diet (to individual baseline characteristics) consisting of 55% carbohydrate, 15% protein, and 30% fat. The difference between the 3 groups was in the form of their carbohydrates.
Group 1: received 25% of their total energy in the form of glucose
Group 2: received 25% of their total energy in the form of fructose
Group 3: received 25% of their total energy in the form of high fructose corn syrup (55% fructose, 45% glucose)
The intervention was relatively short, consisting of both an inpatient and outpatient period, and is described in the methodology section.
Keep in mind, 25% of total energy in the form of sugar is not as extreme as you might think. For a person consuming 2,400 kcal/day this amounts to about 120 pounds/year of sugar, which is slightly below the average annual sugar consumption in the United States. In that sense, the subjects in Group 3 can be viewed as the “control” for the U.S. population, and Group 1 can be viewed as an intervention group for what happens when you do nothing more in your diet than remove sugar, (which was the first dietary intervention I made in 2009).
Despite the short duration of this study and the relatively small number of subjects (16 per group), the differences brought on by the interventions were significant. The figure below shows the changes in serum triglycerides via 3 different ways of measuring them. Figure A shows the difference in 24-hour total levels (i.e., the area under the curve for serial measurements – hey, there’s our integral function again!). Figure B shows late evening (post-prandial) differences. Figure C shows the overall change in fasting triglyceride level from baseline (where sugar intake was limited for 2 weeks and carbohydrate consumption consisted only of complex carbohydrates).
The differences were striking. The group that had all fructose and HFCS removed from their diet, despite still ingesting 55% of their total intake in the form of non-sugar carbohydrates, experienced a decline in total TG (Figure A, which represents the daily integral of plasma TG levels, or AUC). However, that same group experienced the greatest increase in fasting TG levels (Figure C). Post-prandial TG levels were elevated in all groups, but significantly higher in the fructose and HFCS groups (Figure B). The question this begs, of course, is which of these measurements is most predictive of risk?
Historically, fasting levels of TG are used as the basis of risk profiling (Figure C), and according to this metric glucose consumption appears even worse than fructose or HFCS. However, recent evidence suggests that post-prandial levels of TG (Figure B) are a more accurate way to assess atherosclerotic risk, as seen here, here, and here.
The figure below summarizes the differences in LDL-C, non-HDL-C, apoB, and apoB/apoA-I (remember, apoB and apoA-I are good surrogates for LDL-P and HDL-P, respectively).
Again, the results were unmistakable with respect to the impact of fructose and HFCS on lipoproteins, and by extension the relative lack of harm brought on by glucose in isolation. [Removal of glucose and fructose/HFCS would have been a very interesting control group.]
In just a short period of time these dietary interventions had a profound impact on the markers of cardiovascular risk.
So, there you have it – everything you need to know about cholesterol in the length of time it takes to go to the grocery store, buy some high-cholesterol bacon and eggs, cook it, eat it, and savor it. Now it’s your turn to be the judge. Is the conventional wisdom about cholesterol being bad really true?