Love your factory analogy! I have been pondering the penetrance puzzle for years, and the Apple factory seems to capture many of its features and subtleties.
An additional wrinkle to perhaps consider is genetic variation. To extend your analogy, say Apple were to dip into its bags of cash and build multiple additional factories. Just as no two humans are identical and might differ at thousands or even millions of genetic loci, any pair of factories might, for example, have slightly (or markedly) different levels of security (i.e. immune variation), say, or might be built in cities/countries with significant differences in drinking/drug use, work habits, education level, etc.
Whatever the source(s) of variation, the point is that it's hard to imagine that any two of these factories would be precisely the same. And so, even if these factories used the same centralized operations manuals and produced iPhones to the same specs, inevitably there would be differences in how they looked and perhaps even operated internally.
Now, going back to your huntingtin analogy, say Apple hired only alcoholic solar panel technicians in two factories, one in China and the other in India or Mexico City. Given the size and complexity of the networks in these factories, it's similarly hard to imagine that they would respond identically to the same insult.
Great post! I have been trying to understand this recently. I was recently at a talk where someone was predicting whether mutations/variants within a particular gene will be benign or harmful, using something like a Bayesian method. What they predict is actually a probability, because the genotype-phenotype relationship is probabilistic - it will have an effect in some fraction of the population that carry it. It's like predicting what fraction of flips from a weighted coin will come up heads. You end up with a distribution of probabilities. Another way of saying this is that, rather than saying "there's a 75% chance it is harmful based on what we know from similar variants", you have a distribution of probabilities that it will be harmful. ie, a distribution of what fraction of people who have the variant will exhibit the harmful phenotype.
So, the outcome of the predictive model is a statement (with uncertainty!) like "we predict that 75% +/- 20% of people with the gene will exhibit a harmful phenotype".
Sure. Let's compare it with two cells that would normally interface, like a nerve cell and a muscle cell.
This is pretty simple signaling: the nerve cell releases a chemical, ACh, the muscle cells pick it up and react. This requires specialized equipment at the end of the nerve cell, between the nerve and the muscle, and at the edges of the muscles. But these are just terminals. The inner workings of the nerve and the muscles don't have to worry about this.
Immune cells also have terminals where they interact with somatic cells. However, they can't be satisfied with just interacting with somatic cells through terminals, because they need to know everything that's going on within the somatic cell. Likewise, somatic cells can't afford to ignore or mess up their signaling to the immune cells, because they can easily get killed.
So, somatic cells are burdened with a bunch of extra equipment to communicate their state to the immune cells, along with the terminals. This equipment includes:
1. Cytokines, chemicals which get released into extracellular space as a general distress signal
2. DAMPS, a class of chemicals which are SOS signals
3. Surface proteins, which communicate a wide range of signals, including just general presentation of what's going on inside the cell
4. Exosomes and microvesicles, packages that get shipped to immune cells for more information on what's going on inside the cell
All of this signaling is really complex. When it goes haywire, the immune cells can and do cause major damage, so there are additional layers on top of the first layer of immune signaling to avoid costly mistakes. If there were no immune cells, somatic cells would never need to communicate their internal state to any cells, and could just connect with all cells through terminals. But that sort of simplicity isn't really an option for a human.
The topic is interesting but the conclusions far too hyperbolic,
transcription factors are highly complex but iirc they mostly regulate the degree of protein expression, not its final chemical composition?
As for post translational transformations, there are many but their role can often be neglected.
One also has to distinguish between nonsense and misense mutations.
While nonsense mutations can besides hypoexpression, create both functional and dysfunctional proteins, iirc misense always create the same broken protein?
e.g. in muscular dystrophies
the varieties in prognosis can be explained by varieties in the level of protein expression and in other factors, but empirically is it unreasonnable to assume a misense always lead to the same protein?
btw the role of partial genomic imprinting is understudied
It's going to depend on the type of recessive lethality (the gene, the affected developmental pathway, etc) that we're talking about.
I highlighted the general idea of constraint at the top of my comment to illustrate why at some basic level Mendelianism is a reasonable model for survival-essential phenotypes (of which there are many otherwise what's the point of sex) without genetic heterogeneity (as understood to be downstream of the normal function of the protein encoded by the relevant gene).
This conflicts with the principle of constraint. There are simply cases where two true null alleles will be guaranteed to prevent life (recessive lethality). There are other variant-phenotype relationships that are virtually without uncertainty too. The Mendelian model is obviously good enough that we use it in medical genetics to make diagnoses and guide treatment.
The concepts of penetrance, expressivity, epistasis are simply layers on top of the basic Mendelian framework. The Mendelian framework still informs many assumptions in quantitative and population genetics. The idea of linkage disequilibrium is a modification of Mendel's law of independent assortment. Many of the assumptions made in the oft more respected quant/popgen field are more fuzzier than Mendel's model.
It simply depends on the specific gene and developmental pathway we're talking about. For example, Notch in fruit flies is well described. Humans have several Notch paralogs likely in part to avoid this vulnerability.
This is why I highlighted the concept of constraint in my initial comment. There are survival-essential phenotypes (certain molecular activities that must occur for an organism to survive) that are entirely dependent on the function of one protein encoded at a single locus. This is part of why sex exists of course. For each of these cases, Mendelianism is quite plainly the most appropriate model.
I found this a really interesting supporting argument for genes not being destiny.
Love your factory analogy! I have been pondering the penetrance puzzle for years, and the Apple factory seems to capture many of its features and subtleties.
An additional wrinkle to perhaps consider is genetic variation. To extend your analogy, say Apple were to dip into its bags of cash and build multiple additional factories. Just as no two humans are identical and might differ at thousands or even millions of genetic loci, any pair of factories might, for example, have slightly (or markedly) different levels of security (i.e. immune variation), say, or might be built in cities/countries with significant differences in drinking/drug use, work habits, education level, etc.
Whatever the source(s) of variation, the point is that it's hard to imagine that any two of these factories would be precisely the same. And so, even if these factories used the same centralized operations manuals and produced iPhones to the same specs, inevitably there would be differences in how they looked and perhaps even operated internally.
Now, going back to your huntingtin analogy, say Apple hired only alcoholic solar panel technicians in two factories, one in China and the other in India or Mexico City. Given the size and complexity of the networks in these factories, it's similarly hard to imagine that they would respond identically to the same insult.
Great post! I have been trying to understand this recently. I was recently at a talk where someone was predicting whether mutations/variants within a particular gene will be benign or harmful, using something like a Bayesian method. What they predict is actually a probability, because the genotype-phenotype relationship is probabilistic - it will have an effect in some fraction of the population that carry it. It's like predicting what fraction of flips from a weighted coin will come up heads. You end up with a distribution of probabilities. Another way of saying this is that, rather than saying "there's a 75% chance it is harmful based on what we know from similar variants", you have a distribution of probabilities that it will be harmful. ie, a distribution of what fraction of people who have the variant will exhibit the harmful phenotype.
So, the outcome of the predictive model is a statement (with uncertainty!) like "we predict that 75% +/- 20% of people with the gene will exhibit a harmful phenotype".
Can you say more about why insistence on legibility is the biggest source of added complexity?
Sure. Let's compare it with two cells that would normally interface, like a nerve cell and a muscle cell.
This is pretty simple signaling: the nerve cell releases a chemical, ACh, the muscle cells pick it up and react. This requires specialized equipment at the end of the nerve cell, between the nerve and the muscle, and at the edges of the muscles. But these are just terminals. The inner workings of the nerve and the muscles don't have to worry about this.
Immune cells also have terminals where they interact with somatic cells. However, they can't be satisfied with just interacting with somatic cells through terminals, because they need to know everything that's going on within the somatic cell. Likewise, somatic cells can't afford to ignore or mess up their signaling to the immune cells, because they can easily get killed.
So, somatic cells are burdened with a bunch of extra equipment to communicate their state to the immune cells, along with the terminals. This equipment includes:
1. Cytokines, chemicals which get released into extracellular space as a general distress signal
2. DAMPS, a class of chemicals which are SOS signals
3. Surface proteins, which communicate a wide range of signals, including just general presentation of what's going on inside the cell
4. Exosomes and microvesicles, packages that get shipped to immune cells for more information on what's going on inside the cell
All of this signaling is really complex. When it goes haywire, the immune cells can and do cause major damage, so there are additional layers on top of the first layer of immune signaling to avoid costly mistakes. If there were no immune cells, somatic cells would never need to communicate their internal state to any cells, and could just connect with all cells through terminals. But that sort of simplicity isn't really an option for a human.
The topic is interesting but the conclusions far too hyperbolic,
transcription factors are highly complex but iirc they mostly regulate the degree of protein expression, not its final chemical composition?
As for post translational transformations, there are many but their role can often be neglected.
One also has to distinguish between nonsense and misense mutations.
While nonsense mutations can besides hypoexpression, create both functional and dysfunctional proteins, iirc misense always create the same broken protein?
e.g. in muscular dystrophies
the varieties in prognosis can be explained by varieties in the level of protein expression and in other factors, but empirically is it unreasonnable to assume a misense always lead to the same protein?
btw the role of partial genomic imprinting is understudied
https://en.wikipedia.org/wiki/Genomic_imprinting
It's going to depend on the type of recessive lethality (the gene, the affected developmental pathway, etc) that we're talking about.
I highlighted the general idea of constraint at the top of my comment to illustrate why at some basic level Mendelianism is a reasonable model for survival-essential phenotypes (of which there are many otherwise what's the point of sex) without genetic heterogeneity (as understood to be downstream of the normal function of the protein encoded by the relevant gene).
This conflicts with the principle of constraint. There are simply cases where two true null alleles will be guaranteed to prevent life (recessive lethality). There are other variant-phenotype relationships that are virtually without uncertainty too. The Mendelian model is obviously good enough that we use it in medical genetics to make diagnoses and guide treatment.
The concepts of penetrance, expressivity, epistasis are simply layers on top of the basic Mendelian framework. The Mendelian framework still informs many assumptions in quantitative and population genetics. The idea of linkage disequilibrium is a modification of Mendel's law of independent assortment. Many of the assumptions made in the oft more respected quant/popgen field are more fuzzier than Mendel's model.
Would you call death a phenotype?
It's a supra-phenotype. The lethality occurs because of the impact on a specific developmental process or for some other reasons.
So, there's still no guarantee on the exact, non-supra phenotype that the alleles would have? There's only a guarantee of lethality?
It simply depends on the specific gene and developmental pathway we're talking about. For example, Notch in fruit flies is well described. Humans have several Notch paralogs likely in part to avoid this vulnerability.
This is why I highlighted the concept of constraint in my initial comment. There are survival-essential phenotypes (certain molecular activities that must occur for an organism to survive) that are entirely dependent on the function of one protein encoded at a single locus. This is part of why sex exists of course. For each of these cases, Mendelianism is quite plainly the most appropriate model.