Towards Computing A Genetic Risk of Obesity Score

Genetic Makeup Can Exacerbate Effects of Fried Food Diet, published today, summarizes the findings in Fried food consumption, genetic risk, and body mass index: gene-diet interaction analysis in three US cohort studies, an Open Access article in the British Medical Journal by Qi et al, a group of Harvard professors. In this article, the authors reused a genetic risk of obesity score that they had previously published in Sugar-Sweetened Beverages and Genetic Risk of Obesity in the New England Journal of Medicine. Unfortunately, the latter article is not Open Access. So, I'll describe the Qi et al Genetic Risk of Obesity score (which I'll call QiGRO for short) in this post.

The information needed to compute QiGRO is presented in Table 1 of the Supplementary Appendix to Sugar-Sweetened Beverages and Genetic Risk of Obesity. This lists 32 SNPs (single-nucleotide polymorphisms) contributing to risk of increased BMI (body mass index), along with a β-coefficient describing the relative effect size for each. Here are the relevant columns from the table:

SNP	Nearest Gene	Chromosome	Effect	Other	Beta
rs543874	SEC16B	1	G	A	0.22
rs1514175	TNNI3K	1	A	G	0.07
rs1555543	PTBP2	1	C	A	0.06
rs2815752	NEGR1	1	A	G	0.13
rs2890652	LRP1B	2	C	T	0.09
rs887912	FANCL	2	T	C	0.1
rs713586	RBJ	2	C	T	0.14
rs2867125	TMEM18	2	C	T	0.31
rs13078807	CADM2	3	G	A	0.1
rs9816226	ETV5	3	T	A	0.14
rs13107325	SLC39A8	4	T	C	0.19
rs10938397	GNPDA2	4	G	A	0.18
rs4836133	ZNF608	5	A	C	0.07
rs2112347	FLJ35779	5	T	G	0.1
rs987237	TFAP2B	6	G	A	0.13
rs206936	NUDT3	6	G	A	0.06
rs10968576	LRRN6C	9	G	A	0.11
rs3817334	MTCH2	11	T	C	0.06
rs4929949	RPL27A	11	C	T	0.06
rs10767664	BDNF	11	A	T	0.19
rs7138803	FAIM2	12	A	G	0.12
rs4771122	MTIF3	13	G	A	0.09
rs11847697	PRKD1	14	T	C	0.17
rs10150332	NRXN3	14	C	T	0.13
rs2241423	MAP2K5	15	G	A	0.13
rs7359397	SH2B1	16	T	C	0.15
rs1558902	FTO	16	A	T	0.39
rs12444979	GPRC5B	16	C	T	0.17
rs571312	MC4R	18	A	C	0.23
rs29941	KCTD15	19	G	A	0.06
rs3810291	TMEM160	19	A	G	0.09
rs2287019	QPCTL	19	C	T	0.15

If you subscribe to 23andMe, you can get some idea about your QiGRO score. Not all of the above SNPs are included in the 23andMe data at this time, so you can't compute it exactly. You can compute the contribution to your QiGRO score from the SNPs that 23andMe does include. We'll call this the modified QiGRO score for short. Here's how to compute it.

Sign in to the service, click on the downward arrow next to your name in the upper right, and click "Browse Raw Data". This brings you to a page where you can "Jump to a gene" "or a SNP".

Copy the first SNP from the table above, rs543874, into the "or a SNP" box (the righthand one) on the "Browse Raw Data" page and click Go. 23andMe will list a result with your genotype at this SNP: it might be AA, AG, or GG. Since the above table lists G as the Effect allele, count the number of Gs in your genotype. If your genotype is AA, AG, or GG, you will get 0, 1, or 2, respectively. Multiply the number you get by the Beta coefficient, 0.22, from the first row of the above table. So if your genotype is AA, AG, or GG, you will get 0.0, 0.22, or 0.44, respectively. Start a running tally for your score using this number. Also start another running tally of the Beta coefficients, 0.22. Keep track of the number of SNPs you've checked. This is now 1.

In case you're curious, here's what this means. You got two copies of this locus (a site on a chromosome), one from your mother and one from your father. The above table says that there are two possible variants, or alleles, at this locus: A (the DNA nucleotide alanine) or G (the DNA nucleotide guanine). Here G is the effect allele, meaning that each G increases the risk of a higher BMI. The Beta coefficient describes how much it increases the risk.

Continuing, copy the second SNP from the table above, rs1514175, into the "or a SNP" box and click Go. In this case again your genotype at this SNP might be AA, AG, or GG. However, this time A is the Effect allele, so count the number of As in your genotype. If your genotype is AA, AG, or GG, you will get 2, 1, or 0, respectively. Multiply this by the Beta coefficient, 0.07, from the second row of the above table. So if your genotype is AA, AG, or GG, you will get 0.14, 0.07, or 0.0, respectively. Add this to your running tally for your score. For instance, if your genotype at rs543874 was AG, and your genotype at rs1514175 was AA, your running tally for your score would be 0.22 + 0.14 = 0.36. Also add the Beta coefficient, 0.07 to your running tally of the Beta coefficients, yielding 0.22 + 0.07 = 0.39. The number of SNPs you've checked is now 2.

Copy the third SNP from the table above, rs1555543, into the "or a SNP" box and click Go. Currently, 23andMe tells me: “No SNPs matching 'rs1555543' found in the data from your chip.” I deliberately did not include this information in the table above since 23andMe updates their chips and reruns samples from time to time, so by the time you read this post the data may be present. Try it and see. If there is no data, skip adding anything either to the running tally for your score or the running tally using the Beta coefficients.

Copy the fourth SNP from the table above, rs2815752, into the "or a SNP" box and click Go. In this case your genotype at this SNP might be AA, AG, or GG. Here A is the Effect allele, so count the number of A's in your genotype. If your genotype is AA, AG, or GG, you will get 2, 1, or 0, respectively. Multiply this by the Beta coefficient, 0.13, from the fourth row of the above table. So if your genotype is AA, AG, or GG, you will get 0.26, 0.13, or 0.0, respectively. Add this to your running tally for your score. For instance, if your running tally for your score was 0.36 as in the example above, and your genotype at rs2815752 is AA, your running tally for your score will now be 0.36 + 0.26 = 0.62. Add the Beta coefficient 0.13 to your running tally of the Beta coefficients, yielding 0.39 + 0.13 = 0.52. The number of SNPs you've checked is now 3.

Currently, rs2890652, rs887912, and rs713586 are not included in the 23andMe data. Just to be completely clear, I'll write out one more step.

Copy the eighth SNP from the table above, rs2867125, into the "or a SNP" box and click Go. In this case your genotype at this SNP might be CC, CT, or TT. Here C is the Effect allele, so count the number of C's in your genotype. If your genotype is CC, CT, or TT, you will get 2, 1, or 0, respectively. Multiply this by the Beta coefficient, 0.31, from the eighth row of the above table. So if your genotype is CC, CT, or TT, you will get 0.62, 0.31, or 0.0, respectively. Add this to your running tally for your score. For instance, if your running tally for your score was 0.62 as in the example above, and your genotype at rs2867125 is CT, your running tally for your score will now be 0.62 + 0.31 = 0.93. Add the Beta coefficient 0.31 to your running tally of the Beta coefficients, yielding 0.52 + 0.31 = 0.83. The number of SNPs you've checked is now 4.

Continue in this way until you've checked all 32 SNPs in the table. Currently there's no data for rs9816226, rs4836133, rs2112347, rs3817334, rs4929949, rs10767664, rs4771122, rs11847697, rs10150332, or rs7359397. So the number of SNPs you've checked would end up at 18, and your running tally of the Beta coefficients would end up at 2.84, assuming the same set of SNPs are included. Your running tally for your score will be some number between 0 and twice the sum of the Beta coefficients, i.e., 5.68 currently.

To get your modified QiGRO score, take your running tally for your score, divide it by your running tally of the Beta coefficients, and multiply it by the number of SNPs you've checked. You will get a number between 0 and twice the number of SNPs you've checked, i.e., between 0 and 36 currently.

Your actual QiGRO score includes the contributions of the other (currently 14) SNPs you weren't able to check, so it lies between your modified QiGRO score and your modified QiGRO score plus twice the number of SNPs you weren't able to check, i.e., currently your modified QiGRO score plus 28. Unfortunately this range is too large to provide useful information. According to Figure S1 in the Supplementary Appendix to Sugar-Sweetened Beverages and Genetic Risk of Obesity, almost the entire cohorts studied fall between a QiGRO score of 19 and a QiGRO score of 39, a range of only 20.  In the absence of any other information, you might look at the middle of the range, i.e., your modified QiGRO score plus 14. To get a better estimate we would need to know the distribution of these SNPs in the population.

In Fried food consumption, genetic risk, and body mass index: gene-diet interaction analysis in three US cohort studies, all study participants had a QiGRO score between 13 and 43. Qi et al divided their study participants into thirds by increments of 10 risk alleles, i.e., the lower third with scores between 13 and 23, the middle third with scores between 24 and 33, and the upper third with scores between 34 and 43. (They did not publish the exact cutoffs in their paper, so the intermediate boundaries may be off by one.) Qi et al found in these two studies that fried foods and sugar sweetened beverages increase BMI more among those with a higher QiGRO score. So, if it appears that your QiGRO score is on the high side, it may be more beneficial for you to cut back on sugar sweetened beverages and fried foods.

You might also note your genotype at rs1558902 in particular (which is currently included in the 23andMe data). Qi et al found that the association between BMI and this variant alone was significant, and strengthened as more fried foods were consumed.

Patrimony and the Evolution of Risk-taking

This paper by Michael D. Stern provides an interesting insight into the evolutionary dynamics of risk-taking: it is evolutionarily favored when and only when parental resources can be passed to offspring. The paper deserves to be widely read and followed up by testing its robustness to changing assumptions, probing its applicability in biological and social contexts, and teasing out its implications for the generation of diversity by evolutionary processes. One question that immediately springs to mind is how this fits in with Peter Richerson's and Robert Boyd's theory of cultural evolution—see Not By Genes Alone: How Culture Transformed Human Evolution (affiliate link). Another is to what extent this relates to Dean K. Simonton's historically-based viewpoint laid out in Origins of Genius: Darwinian Perspectives on Creativity (ditto). A third is the interplay between this dynamic and “relaxed selection.” I personally would be most interested in seeing to how this dynamic extends to cooperating groups, whether of kin or of nonrelated individuals—see Genetic and Cultural Evolution of Cooperation (ditto).

Patrimony and the Evolution of Risk-Taking

Abstract: The propensity to make risky choices has a genetic component, and recent studies have identified several specific genes that contribute to this trait. Since risk-taking often appears irrational or maladaptive, the question arises how (or if) natural selection favors risk-taking. Here we show, using a stochastic simulation of selection between two hypothetical species, “R” (risk-seeking) and “A” (risk-averse) that, when expected reproductive fitness of the individual is unaffected by the making of the risky choice (winnings balanced by losses) natural selection (taken to the point of extinction) favors the risk-averse species. However, the situation is entirely reversed if offspring are permitted to inherit a small fraction of the parent's increased or decreased fitness acquired through risk-taking. This seemingly Lamarckian form of inheritance actually corresponds to the human situation when property or culture are transmitted in families. In the presence of this “cultural inheritance”, the long-shot risk-taking species was overwhelmingly favored, even when 90% of individuals were rendered sterile by a losing choice. Given this strong effect in a minimal model, it is important to consider the co-evolution of genes and culture when interpreting the genetics of risk-taking. This conclusion applies, in principle, to any species where parental resources can directly affect the fecundity of offspring. It might also be relevant to the effects of epigenetic inheritance, if the epigenetic state of zygotes can be affected by parental experiences.

Protein Space Still Ripe For Exploration

Proteins are Nature's Robots: they carry out most of the activities of life, as well as constituting some of its major structural building blocks. Many of them fold into structurally stable conformations, which can then form crystals. Crystallographers “solve” the 3-dimensional structure of a protein by computing the coordinates of each of its constituent atoms from the X-ray diffraction pattern of these crystals. Thousands of such structures have been deposited in the Protein Data Bank (PDB). These beautiful macromolecules constitute The Machinery Of Life; author and illustrator David S. Goodsell highlights some of the fascinating stories of how they work in his Molecule of the Month column.

Clearly, only a finite number of proteins exist among all the living organisms on the Earth today. How far have we gotten in exploring the space of proteins? From a historical point of view, we can judge the extent to which a culture had explored the surface of the Earth over time by looking at the maps drawn by people of that culture. Had they encountered all the continents? How fine is the level of detail for each continent?

Current maps of protein space can be found at SCOP: Structural Classification of Proteins and CATH: Protein Structure Classification Database. For simplicity I'll discuss SCOP, although CATH is similar. From the Introduction to SCOP, “the different major levels in the hierarchy are: Family: Clear evolutionary relationship; Superfamily: Probable common evolutionary origin; and Fold: Major structural similarity.” Let's look at the growth of each of these categories in SCOP over time:

In each case I've set the height of the Y-axis at three times the value in July 2001 when SCOP version 1.55 was released. The rate of finding new folds (the “continents” of protein space) is neither taking off (as we might expect if there had been the same kinds of fundamental advances in protein structure determination as in sequencing) nor levelling off (as we might expect if we had explored most of the existing space). The conclusion is clear: there's plenty of unknown territory left for the pioneers of tomorrow.