This is not the part 2 continuation that I had originally planned. As I continued to pursue my original line of thought, it became apparent that the results of this study were problematic as I attempted to fit them into the overall picture of ApoE research in Turkey. This is important, because as I have since come to understand, the whole matrix of disease, epigenetics and genetic predisposition, is nuanced and complicated. A single SNP in context A can have an entirely different effect when in context B.
What was the problem?
The results of the study would seem to suggest that the APO ε2 allele that is expressed worldwide at a 7% incidence is present in healthy individuals in Western Anatolia, Turkey at 17%. OK, that’s strange. But, additionally, in the PCOS group, ε2 was found at the more usual level (4%), while the presence of the most common ApoE genetic variant, ε3, was found to be detrimental. How can this be? What would be the evolutionary pressure that would favor a disease situation? Is it as Finch suggests that higher ApoE4 favored higher fertility at the expense of longevity and in this case, by means of protection from PCOS? This is indeed an extraordinary claim, and, as we know, extraordinary claims require extraordinary evidence. We need to use common sense, look at the journal reputation and look at other studies.
Although the study fails the common sense test, the journal, JARG, is among the top 50 in developmental biology. So, in checking further, I looked at other genetic studies from Turkey and came up with the following list (click on the table to view it separately in order to see the actual numbers):
The study in question is positioned at the top, pubmed ID 2649336. Notice, (columns AD-AF, highlighted in red), they are reporting a 17% incidence for Apo ε2 and a 75% incidence for ε3 in the healthy control population. (BTW, I marked the other anomalous study entries also in red.) In light of the allele distributions reported in the other studies, how likely is it that these numbers are truly representative of the local population?
The study was conducted in Izmir in the Aegean region on the Western coastline of Turkey:
Istanbul is about 200 mi away. A study there (line 9, PMC22277050) reports ε2, ε3 incidence in healthy controls at the more expected 5% and 88%, respectively, whereas another, PMC19368142 on line 13, has these numbers at 9% and 80%. For Ankara, about 250 miles away, one study, PMC21500980 (line 22), reports 3% and 92% for ε2, ε3, while another, PMC2935837 (line 16), says the control population there is 10% and 52% for the ε2, ε3 alleles. Meanwhile, the Turkish Adult Risk Factor Study (n=1774) collected data from all 7 regions and found the ε2, ε3 allele representation to be 7% and 87%.
These are huge differences, especially in Ankara. In these small Turkish studies, how are they collecting their control samples? If the control sample can vary by that much, how do they ensure that the control sample is applicable to the study question and hasn’t introduced ascertainment bias?
Another problem that bothered me was that if we were trying to show an evolutionary pressure raising ApoE4 for fertility protection at the expense of reduced ε3 and reduced longevity, then why are these studies (highlighted in red in column P under “Case as %”) showing higher ε3 associated with disease?
The clincher was the Hardy-Weinberg Equilibrium calculation (see columns at the right). This calculation is used to check for ascertainment bias. It looks at the internal consistency between the various genotypes. The original study paper failed to report the HW calculation, so I did it myself. For the study in question, the HW calculated probability estimates about a 92% chance that the sampling was in error.
To try to answer some of my questions, I drafted a very polite inquiry as to how the study control sample had been acquired and what had been done to protect against ascertainment bias. I then had a friend with some academic credentials send it off to the study author. Over a year has passed; we have not received a reply.
And the conclusion?
The Turkish population is rapidly growing and is ancestrally very diverse. The bottom line is that with a 92% likelihood that there was sampling error, I would hesitate to draw conclusions from this study. I could have just deleted my previous, part 1 post, but decided to leave it in with this as a follow up to remind both myself and others to be careful about attributing too much to a single study, especially when the sample size is small and the numbers just don’t make sense.