Omnigenics

May 10, 2020 7:23 AM

The more closely geneticists look at complex traits and diseases, the harder it gets to find active genes that don’t influence them.

After the paper proposing omnigenics (An Expanded View of Complex Traits: From Polygenic to Omnigenic) was published

The reaction was swift. “It caused a lot of discussion,” said Barbara Franke, a geneticist at Radboud University in the Netherlands who studies attention deficit hyperactivity disorder (ADHD). “Everywhere you went the omnigenic paper would be discussed.”

Lectures by Jonathan Pritchard explaining the idea:
The Omnigenic Architecture of Human Complex Traits
Pritchard: Omnigenic Architecture of Human Complex Traits

Psychiatrists discuss the implications for the understanding and treatment of mental illness:
Journal of Psychiatry and Brain Science Virtual Special Issue
Life after GWAS – where to next, for psychiatric genetics?

Ed Yong sums it up over at the Atlantic: What If (Almost) Every Gene Affects (Almost) Everything?

More specifically, it means that all the genes that are switched on in a particular type of cell—say, a neuron or a heart muscle cell—are probably involved in almost every complex trait that involves those cells. So, for example, nearly every gene that’s switched on in neurons would play some role in defining a person’s intelligence, or risk of dementia, or propensity to learn. Some of these roles may be starring parts. Others might be mere cameos. But few genes would be left out of the production altogether.

This might explain why the search for genetic variants behind complex traits has been so arduous. For example, a giant study called… er… GIANT looked at the genomes of 250,000 people and identified 700 variants that affect our height. As predicted, each has a tiny effect, raising a person’s stature by just a millimeter. And collectively, they explain just 16 percent of the variation in heights that you see in people of European ancestry. That’s not very much, especially when scientists estimate that some 80 percent of all human height variation can be explained by genetic factors. Where’s that missing fraction?

Pritchard’s team re-analyzed the GIANT data and calculated that there are probably more than 100,000 variants that affect our height, and most of these shift it by just a seventh of a millimeter. They’re so minuscule in their effects that it’s hard to tell them apart from statistical noise, which is why geneticists typically ignore them. And yet, Pritchard’s team noted that many of these weak signals cropped up consistently across different studies, which suggests that they are real results. And since these variants are spread evenly across the entire genome, they implicate a “substantial fraction of all genes,” Pritchard says.

The team found more evidence for their omnigenic model by analyzing other large genetic studies of rheumatoid arthritis, schizophrenia, and Crohn’s disease.

posted by clawsoon (29 comments total)

A question for those who know more about this: Polygenic and omnigenic conclusions are based on data from large-scale SNP data. How likely are those conclusions to change if we get better at large-scale screening of other types of genetic variation, copy number variants and whatnot?
posted by clawsoon at 7:42 AM on May 10, 2020

This is fascinating. Ever since learning years ago that height had been found linked to multiple genes, not a single one--among other factors such as timing of childhood illnesses--I've suspected it would play out more like this.

I'm just a lay-man with a fairly decent grasp of biology, but there's definitely a bit of satisfaction in seeing my suspicions borne out. The model I've had in my head all these years wasn't that it was the genes--or rather not just the genes, but more that they produce a standing interference or wave pattern, which makes it a whole lot more complicated to figure out. Like tossing pebbles in a pond, turning one off or on would change the entire pattern of ripples. Mapping that would be mind-boggling.

Considering the relative paucity of genes in the human genome compared to many other species on Earth--not to mention the vast number conserved between wildly disparate species--it's clear that given our complexity, something interesting is going on.
posted by los pantalones del muerte at 7:56 AM on May 10, 2020 [5 favorites]

My recollection of reading this paper, as a quantitative geneticist, was, "Oh, cool, someone who hasn't read Fischer, Wright, Haldane, or Falconer has reinvented the infinitesimal alleles model and epistasis." On further reflection that wasn't entirely fair, but I can't say that this work has really had any notable impact in my field. I work in livestock, though, not humans, and maybe it's different over there. I can't find it right now, but I do recollect that some reanalysis of the GIANT data (possibly by Wray and Visscher) found a much higher heritability with some improvements in the model. Similar traits in cattle have heritabilities greater than 50%.

It's definitely proven to be much harder than anticipated to identify true causal effects and estimate the magnitude of those effects in most populations, but particularly in humans, where there's so much variation within and between populations. However, SNP data have been extremely useful for making predictions of individual genetic merit because it doesn't matter if we know gene functions to make good predictions.

We've done some work on adding CNV to our prediction models and it helps a little. The challenge with CNV is that calling them from SNP data doesn't work real well (there are lots of false positives and negatives), you need sequence data for that.

What the model is really saying, as best I can tell, is that there's epistasis everywhere -- probably true -- and that adequately addressing it in our statistical models is going to be very difficult. Also true because the number of interactions scales so quickly.
posted by wintermind at 8:05 AM on May 10, 2020 [8 favorites]

BTW Ken Burns's The Gene: An Intimate History (4 hours, based on the book of the same name by Siddhartha Mukherjee) was released last month and might still be available to anyone with PBS on cable TV...
posted by XMLicious at 8:16 AM on May 10, 2020 [3 favorites]

One of the reasons eugenics is bullshit is that genes don't work that way. Genetics is astoundingly complex and the interactions of individual genes is so mind-bogglingly complicated and so profoundly poorly-understood that even experts have what amounts to a cursory knowledge on the scale of the actual domain.
posted by Pope Guilty at 8:22 AM on May 10, 2020 [3 favorites]

I think the reason that eugenics is BS is that the "traits" eugenicists want to select for are overwhelmingly controlled by environment, not genetics (e.g., level of education attained or economic "success"). We are very, very good at making long-term genetic changes in plant and animal species because our selection programs are based on actual biological characteristics of individuals (e.g., yield of an ear of corn or number of eggs laid).
posted by wintermind at 8:30 AM on May 10, 2020 [9 favorites]

for whatever reason, I've always found it somehow reassuring when science discovers that the universe is even more imponderably vast than we previously realized (though I suppose now we should be saying multiverse).

What If (Almost) Every Gene Affects (Almost) Everything?

This sort of gives me the same feeling. As above so below and all that. With the added bonus that yes, it puts the lie to f***ing eugenics.
posted by philip-random at 8:33 AM on May 10, 2020

At last! A Magic Bullet for magic bullets.
posted by Conrad-Casserole at 8:56 AM on May 10, 2020

I think the reason that eugenics is BS is that the "traits" eugenicists want to select for are overwhelmingly controlled by environment, not genetics (e.g., level of education attained or economic "success"). We are very, very good at making long-term genetic changes in plant and animal species because our selection programs are based on actual biological characteristics of individuals (e.g., yield of an ear of corn or number of eggs laid).

Selective breeding and eugenics are not the same thing at all; eugenics is based on the idea that certain genes are better on a flat, objective level while selective breeding is based on the idea that certain traits are more desirable for particular purposes and breeding organisms which share those traits can produce offspring which are more likely to possess those traits. Selective breeding can produce offspring with a greater propensity for certain traits, but it's a clumsy process that a) doesn't always work, because genetics is, again, vastly more complex than people realize and b) is, due to that complexity, extremely prone to producing propensities for traits that are not desired by the breeders, like the pulmonary hell that virtually all pug dogs live in.

Part of the point of papers like this is that those aren't the same idea and we don't understand the relationship with anything like the knowledge or certainty we think we do.
posted by Pope Guilty at 8:56 AM on May 10, 2020 [3 favorites]

Thank you for explaining genetics to a geneticist.
posted by wintermind at 9:46 AM on May 10, 2020 [9 favorites]

I'm not the one conflating selective breeding and eugenics.
posted by Pope Guilty at 10:11 AM on May 10, 2020 [1 favorite]

I'm pretty sure Francis Galton was a eugeneticist and I'm pretty sure he advocated selective breeding, so I'm not sure about the distinction?
posted by dmh at 10:31 AM on May 10, 2020 [1 favorite]

What the model is really saying, as best I can tell, is that there's epistasis everywhere -- probably true -- and that adequately addressing it in our statistical models is going to be very difficult.

I actually don’t think that’s what it’s saying. I think the omnigenic model is still mainly focusing on narrow-sense heritability resulting from linear combinations. Pervasive epistasis is a different model that’s been proposed to explain “missing” heritability. (In the model organism work I’m thinking of, where you can actually get a large enough population size to explain most of the variance in a complex trait, I’m pretty sure that epistasis was sometimes but not typically found to be a major contributor. Sorry, on my phone so can’t link.)

Also one major difference with livestock is that I would expect selective breeding to mean that way more of the variance in directly selected traits would be explained by a relatively small number of mutations, compared to “natural” populations.
posted by en forme de poire at 10:59 AM on May 10, 2020 [1 favorite]

We expect nature to look like our idea of good design. In software or engineering, good design separates each piece so it does one or two specific things, or it has a really general purpose. This makes it easier to improve the design or repair it when there is a problem.

Evolution does the opposite where for efficiency each piece is used over, and in combination for many purposes. There is no need in evolution for someone to understand how it works. In software this would be called "spaghetti code", an unmaintainable mess.
posted by bhnyc at 11:02 AM on May 10, 2020 [6 favorites]

This is a surprise? Did the medical community ever think to ask a fucking computer programmer?
posted by metametamind at 11:53 AM on May 10, 2020

The difference between "eugenics" and "selective breeding" is essentially the context in which you're doing it: eugenics is performed on humans in order to select for some kind of vague... betterness... of person, and selective breeding is performed on non-human organisms usually for considerably more specific goals. The techniques are similar (killing or sterilizing individuals with traits you don't want to reproduce, for example), and eugenicists are usually pretty shitty at genetics qua genetics as well as defining the traits they want, but the general idea really is the same. The horror of eugenics derives from the effects that eugenics has on people; the technology is neutral, but can be used to evil ends.

A question for those who know more about this: Polygenic and omnigenic conclusions are based on data from large-scale SNP data. How likely are those conclusions to change if we get better at large-scale screening of other types of genetic variation, copy number variants and whatnot?

Extremely unlikely--the notion behind the SNPs isn't necessarily that the copy number variants that turn up on a large-scale GWAS analysis are causing the phenotype of interest, but that the copy number variants are probably closely linked to whatever genetic variation is having a particularly strong effect on the genotypic variation. Once you figure out the copy number variant you're interested in, then you engage in mapping near it to figure out exactly what it's signifying.
posted by sciatrix at 12:21 PM on May 10, 2020 [4 favorites]

metametamind: This is a surprise? Did the medical community ever think to ask a fucking computer programmer?

That would make more sense if intelligent design were true. It isn't, though, and we have nothing precisely analogous to a billion years of random mutations resulting in working systems, so we have to ask life itself to find out how it works.

This isn't a completely obvious result. A few years ago, they found that most variation in dog breeds is explained by only a handful of genes. Why shouldn't humans be the same? Are humans the same? You can't find the answer with assumptions or analogies or insights from other fields. You can't even answer it with data from other species.

Evolution has a way of exploring all the options, and multiple options can be successful. Traits can spread through a species on the basis of a handful of mutations, and they can spread through a species on the basis of hundreds of thousands of mutations. Both are possible; we can't answer the question of which possibility applies to humans until we actually gather the data.
posted by clawsoon at 12:23 PM on May 10, 2020 [1 favorite]

I work in livestock

Inside of a cow, it's too dark to read scientific papers.
posted by Greg_Ace at 12:25 PM on May 10, 2020 [16 favorites]

sciatrix: the technology is neutral

That might depend on how you feel about monoculture crops and breeding animals to serve our ends. :-)

(My own moral position on this is ambiguous, but I do know that the strong version of the moral position exists.)
posted by clawsoon at 12:29 PM on May 10, 2020 [2 favorites]

We expect nature to look like our idea of good design.

Here's the thing, and I keep coming back to this both personally and professionally:

On a genetic scale, what is good, anyway? We talk a lot about genetic quality and high-quality individuals in my subfield, and we define these in certain specific ways: which individuals can resist a certain parasite best, or can forage most efficiently, or is the densest, or or or or...

And the thing I always wind up coming back to is that it's often not possible to disentangle these things from stochastic variation in environment--or, well, luck of the draw. The selective microclimate changes all the time, and individuals do the best they can to cope with it, but the "best" genotype is a perpetually moving target. So we consider, well, perhaps individuals hook up the most desirable traits to condition dependence, so that the "highest-quality" animals are simply those who are best matched to the microclimate of the moment. But that is inextricable, in many species, from who was best matched to the environment ten years ago, or who had a mother who was particularly able to invest at the time, and so forth. "Best" is a moving target.

What does "high quality genetics" even mean? With respect to selective breeding, we can define the term narrowly: we want, say, the cow who produces the most milk possible in a modern high-output dairy setting. But if your context for livestock management changes--if you need, for example, a dairy cow that can resist biting flies, or one that isn't vulnerable to milk fever around calving--those cattle aren't necessarily the best options. We are very good at optimizing animal and plant lines to extremes that serve our agriculture, which is fine so long as we can sustain those production levels--but you change the circumstances just a smidgen, and suddenly our fine "good design" looks like a child's first failure.
posted by sciatrix at 12:35 PM on May 10, 2020 [6 favorites]

(Yeah, I see you, but... I still think the technology is neutral, and to advocate otherwise, I can only imagine a hardline animal rights activist who is so concerned about monoculture that they're also advocating that we start swapping out our maize for teosinte. You can have a lot of nuanced positions about selective breeding and genetic diversity and so forth, but if theoretical-you are going to argue seriously that selective breeding qua selective breeding is inherently no-exceptions an evil idea, I have a lot of questions about your worldview.)
posted by sciatrix at 12:39 PM on May 10, 2020 [2 favorites]

I was visiting a college that had an intensive agriculture program with my daughter and we took a walk past some of the animal-raising areas. The piglets were so musclebound that walking was laborious and the breeder pigs literally could not walk. I don’t know if this is typical of modern hog farming but it’s hard to view a technology that breeds such freaks as being entirely neutral.
posted by sjswitzer at 12:49 PM on May 10, 2020 [1 favorite]

sciatrix: the "best" genotype is a perpetually moving target.

I like the idea that it's not an evolutionary landscape, it's an evolutionary seascape.
posted by clawsoon at 1:02 PM on May 10, 2020 [1 favorite]

The surprising thing about the omnigenic model isn't that lots and lots of indirectly-related genes can be relevant to a particular phenotype, or that molecular biology does not look like intelligent design. In my experience, it's been the computer programmers and engineers who have most needed to be disabused of the idea that biology is orderly or modular, or that we understand enough about how gene networks operate to make analogies to even the most chaotic forms of software. These things were known to "the ancients"; Fisher was mentioned upthread, and the people who first studied inheritance knew about things like pleiotropy before we even really knew that genes existed.

The most surprising thing about this model, at least as I understand it, is the extent to which the "peripheral" genes -- i.e., the ones with tiny effect sizes and only indirect effects on the trait of interest -- explain vastly more of the variance in complex traits than the "core" genes, the ones with relatively large effect sizes that seem to act directly in determining that trait. That's what's so counterintuitive and still controversial about this hypothesis: the idea that all these indirect garbage-bin effects, which individually are so tiny you need millions of people to even measure them with any degree of accuracy, collectively do a much better job of predicting how tall you'll be than the sum total of genes that are actually directly involved in processes like bone growth or hormone production.
posted by en forme de poire at 4:40 PM on May 10, 2020 [4 favorites]

it’s hard to view a technology that breeds such freaks as being entirely neutral.

I have yet to see any technology whatsoever that cannot be used unethically by those who lack ethics or don't care to apply them.
posted by wierdo at 5:11 PM on May 10, 2020 [5 favorites]

To go "Yes, and!" at en forme de poire, since we're not all population geneticists in the room--for the curious, the thing with Fisher that is being referred to is Fisher's geometric model (or as I always think of it, Fisher's mutational target, which is way less easy to Google). This model is a mathematical prediction based on Fisher's work that considers effect size of mutations which turn out to be beneficial, which roughly finds that beneficial mutations are much more likely to be small.

If a trait exists with some fitness maximum, and organisms exist at some distance from that optimal measure of the trait, a mutation may move either towards that fitness optimum (in which case it might be beneficial), or away from it (at which point it is definitely maladaptive). However, for mutations of very large effect sizes, you might easily overshoot the desired optimum trait by some distance. If you only overshoot it by a little bit, that is still beneficial--but if you overshoot it by a lot, your mutation winds up being maladaptive even if it is moving the phenotype closer to the true optimum. So Fisher predicts that mutations which are beneficial and therefore selected on are most likely to be small in many successive mutations rather than being mutations of large effect sizes. Amplify that out into many many many traits all acting at once, and you wind up with the situation described in this article.

Now, R. A. Fisher worked this out in 1930, so this idea isn't necessarily new from a population genetics perspective. This omnigenetics paper advances an idea that builds on Fisher's work (among others), which is that these mutations of small effect are so pervasive that variation in every gene which is currently active at the time of trait development creates some small effect on the function of the whole, which add up to larger variations in the aggregate.

Which.

Yeah, that tracks.
posted by sciatrix at 8:58 PM on May 10, 2020 [3 favorites]

it’s hard to view a technology that breeds such freaks as being entirely neutral.

So this is what I mean about the technology being neutral: you can breed those hogs which are hyperspecialized for very specific things, but you can also selectively breed for a balance of many traits at once that confers something in between. Consider, for example, the Alaskan Husky, which is a dog selectively bred for (roughly) the ability to run quickly over long distances. They are endurance running freaks of nature compared to dogs which have not been bred for a century as racers. But there is nothing truly unethical about being an Alaskan Husky--the desired outcome does not cause pain or difficulty for the animals as a consequence of their unusual ability to run.

On the other hand, we can have a discussion about other purely recreational effects of selective breeding. There are fervent arguments about whether it is ethical to breed for, for example, bracycephalic dogs like Pugs or Bulldogs whose cranial anatomy predictably restricts the dogs' ability to breathe, especially during sleep. Pugs are familiar and beloved and do not seem strange to us today, but the attributes for which they have been selected interfere with the dogs' ability to perform species-typical behaviors.

From the perspective of a stalk of maize, corn is every bit as wild and bizarre as a Pug might be to someone who has only seen a wolf. The ancestral plant from which corn is grown is called teosinte, and you can barely see the little ears at the tip of the stalks. The ears we eat have been wildly exaggerated to yield more food that can be harvested and consumed more quickly than the original naturally selected teosinte. Compare any crop to its wild cousins and you will see similar enormous changes in the wild type structure; in fact, something like eight of the cruciferous vegetables we commonly eat come from the same species of Brassica, each line selected to over-emphasize some aspect of the plant's development.

The pigs you saw may have been double-muscled pigs, like Belgian Blue cattle. I would personally categorize double muscling as case-by-case and ask about animal welfare for animals living with the condition. There certainly is argument about the effects of these kinds of double-muscling alleles, but that's an argument about the ends to which we turn artificial selection. It's not an argument about whether artificial selection is ever acceptable, it's an argument about to what extent artificial selection is acceptable and what the limits of that selection should be. That's what I mean by saying "the technology is neutral."
posted by sciatrix at 8:58 PM on May 10, 2020 [6 favorites]

Thanks sciatrix! Just wanted to link the paper I was thinking of above, that used large sample sizes in a yeast cross to look at the sources of “missing heritability”. The contributions from epistasis (essentially, non-additive interactions between alleles) that they found were a bit larger than I remember, but still were typically small relative to the plain vanilla additive effects. (Interestingly, almost none of the epistasis they did find appeared to be explained by considering pairwise interactions, which suggests to me that it would be really difficult to detect and explain a significant part of those non-additive effects.) Basically, this suggests the main problem may not be gene interactions.

Of course this doesn’t consider gene-environment interactions, which are probably relevant in e.g. human psychiatric phenotypes like schizophrenia (empirically, we know there are both genetic risk factors like having a close relative with schizophrenia, as well as environmental risk factors like urban vs. rural environments and adverse childhood experiences).
posted by en forme de poire at 11:17 AM on May 11, 2020 [2 favorites]

en forme de poire: Of course this doesn’t consider gene-environment interactions, which are probably relevant in e.g. human psychiatric phenotypes like schizophrenia (empirically, we know there are both genetic risk factors like having a close relative with schizophrenia, as well as environmental risk factors like urban vs. rural environments and adverse childhood experiences).

Not to mention the stochastic nature of development.
posted by clawsoon at 4:25 PM on May 11, 2020 [2 favorites]