Friday, January 06, 2017

Genetic variation in the human population

With a current population size of over 7 billion, the human population should contain a huge amount of genetic variation. Most of it resides in junk DNA so it's of little consequence. We would like to know more about the amount of variation in functional regions of the genome because it tells us something about population genetics and evolutionary theory.

A recent paper in Nature (Aug. 2016) looked at a large dataset of 60,706 individuals. They sequenced the protein-coding regions of all these people to see what kind of variation existed (Lek et al., 2016) (ExAC). The group included representatives from all parts of the world although it was heavily weighted toward Europeans. The authors used a procedure called "principal component analysis" (PCA) to cluster the individuals according to their genetic characteristics. The analysis led to the typical clustering by "population clusters." (That term is used to avoid the words "race" and/or "subspecies.")


It's difficult to figure out which sequences were analyzed other than the fact they were exons. They say that all exons were sequenced but they don't say much about how many and the total sequence length. As near as I can figure out, they sequenced about 60 Mb (60 million base pairs) of exon sequences. This is way more than the total amount of coding sequence in the human genome (~26 Mb), That's because they sequenced 50 bp on each side of an exon for an additional 100 bp for every exon.

The authors identified 7,404,909 high-quality variants in the population of 60,706 individuals. Of these, the majority (>95%) were single nucleotide polymorphisms (SNPs). The rest were insertions or deletions. More than half of all the variants (54%) were singletons—mutations present in a single individual out of more than 60 thousand. The vast majority (99%) of all variants were present at a frequency of less than 1%. (There would be more than 400 million variants in the entire genome.)

Within the coding regions, there were 1.4 million synonymous variants—mutations that change the codon without changing the encoded amino acid. I think most of the other coding region mutations were missense mutations but it's difficult to extract this information from the paper. I suspect that a significant proportion of the 7 million variants were in the noncoding regions at each end of the exons. (There is also a significant amount of noncoding DNA within the first and last exons in a gene.)

It looks like there are about 10,000 genes that have premature stop codons or frameshift mutations that should prevent synthesis of a functional protein. These are loss-of-function mutations. They are not lethal because only one of the two copies is mutated. The data can be used to calculate how many genes cannot tolerate a loss-of-function mutation because losing one copy would be lethal. These are haploinsufficient genes and there are about 3,000 of them in the human genome.

By comparing the variants to presumed disease-causing mutations in medical databases, the authors calculated that the average person in their database is heterozygous for 54 such variants. Even though the individuals don't suffer from the genetic disease, this is too high a genetic load according to most population genetics. Many of the presumed disease-causing alleles are present at frequencies greater than 1% suggesting they may have been mischaracterized as disease-causing mutations.1

Several of the variants have been reclassified as benign based on the result of this study. The authors point out that filtering based on allele frequency is important—high frequency alleles are probably not disease-causing in most instances. They also point out that even many low frequency alleles previously suspected of causing genetic disease have probably been mischaracterized.2 The average individual in their group contains about 1 dominant disease-causing mutation and that's probably too high, suggesting that some of these mutations are probably benign.

There are 179,774 different variants classified as protein-truncating variants (PTV). These are mostly due to conversion of a normal codon to a stop codon. They are assumed to be loss-of-function mutations as described above. Many of them are present at significant frequencies in the population. The average person has 85 heterologous and 35 homologous PTVs. This suggests there are quite a few genes that can be inactivated (both copies) without significant effect. One of these would be the gene causing O-type blood [Is the high frequency of blood type O in native Americans due to random genetic drift?].

The highlight of the paper in both the abstract and the discussion is the fact there are 3,230 genes that are highly loss-of-function intolerant. Knocking out even a single copy is probably lethal. Most (72%) are not associated with any known genetic disease. This should not be a surprise. These genes are likely housekeeping genes such as those that encode RNA polymerase subunits or important enzymes. Any disruption of function will be lethal and they never appear as disease-causing. Most genetic diseases occur in genes that are not essential for cell survival. The most important genes are rarely associated with genetic diseases; a counter-intuitive fact that should be more widely disseminated.

The final comment in the paper concerns the total amount of variation in the entire human population. The authors point out that nearly every possible mutation should be present in a population of 7 billion individuals. The only exceptions should be dominant lethal mutations. Even with a sample size of 60,000 individuals they are only able to detect a fraction of the total variants present in humans.


1. Yes, the authors and I know about balancing selection and sickle-cell disease.

2. This is one of the problems with genetic testing.

Lek, M., Karczewski, K., Minikel, E., Samocha, K., Banks, E., Fennell, T., O'Donnell-Luria, A., Ware, J., Hill, A., Cummings, B. et al. (2016) Analysis of protein-coding genetic variation in 60,706 humans. Nature 536:285-291. [doi: 10.1038/nature19057]

16 comments:

  1. Larry writes: "The authors point out that nearly every possible mutation should be present in a population of 7 billion individuals. The only exceptions should be dominant lethal mutations"

    In fact, all nonlethal mutations should be present in dozens of independent copies. This is directly related to what I was telling Txpiper the other day, when he assumed that "standing genetic variation" was some kind of hypothetical that scientists just assumed without doing experiments.

    Let's get a ball park figure. For all nonlethal mutations, in the human population of 7 billion, about how many mutants should there be? Nonlethal SNPs only.

    The authors identified 7,404,909 high-quality variants in the population of 60,706 individuals. Of these, the majority (>95%) were single nucleotide polymorphisms (SNPs). The rest were insertions or deletions. More than half of all the variants (54%) were singletons—mutations present in a single individual out of more than 60 thousand.

    7,404,909 * 0.95 * 0.54 = 3.8 million singleton SNPs in the database of 60K individuals.

    Here I'm excluding SNPs that are not singletons because they would be duplicates. Thus, this is an underestimate of standing genetic variation.

    3.8 million singleton SNPs / 60,000 individuals = 63 singleton SNPs per individual

    For the whole Earth:

    63 singleton SNPs per individual * 7 billion individuals = 443 billion singleton SNPs in all human genomes

    This is approximate because there would be some duplicates.

    How many per base pairs in a human genome? 3.2 billion base pairs.

    443 billion singleton SNPs in all genomes / 3.2 billion base pairs per genome = 138 singleton SNPs per base pair per genome.

    But if there's (let's say) there's a C, it can mutate to T, G, or A. Assume equal propensity for each mutation = 33% probability for each.

    138 singleton SNPs per base pair per genome / 3 possible mutations per SNP =

    46 independent copies for every possible nonlethal SNP mutation per base pair in the genome.

    This is a very rough ballpark figure and an underestimate of standing genetic variation because I ignored SNPs that occurred in two or more individuals (non-singletons) in the database of 60,000. The actual figure could easily be almost twice as high: many dozens.

    Furthermore, we should expect much more genetic variation in non-coding DNA (98% of the genome) because 92% of non-coding DNA is known to NOT be conserved and is presumably non-functional.

    So, for every possible non-lethal mutation at every base pair in the whole human genome, 46 independent copies existing somewhere on Earth today is a significant underestimate.

    So "standing genetic variation" is a big deal and known to be so from experiments.

    ReplyDelete
    Replies
    1. The definition I've seen for standing genetic variation is "the presence of more than one allele at a locus in a population", not just SNP's.

      "So "standing genetic variation" is a big deal and known to be so from experiments."

      It will be as big a deal as flatulence when it has a Wikipedia entry. I've been discussing evolution with people and reading papers online for 15 years and had never heard of it till I bumped into the 2013 phys.org article I found when I was arguing about cave species in another forum. Perhaps it was called something else.

      Delete
    2. So you never knew that standing genetic variation existed, txpiper? Is everyone you know an identical clone of each other or something?

      Delete
    3. The concept of Standing Genetic Variation is apparently also referred to as "Cryptic Genetic Variation" or "Evolutionary Capacitance". The interest in 'stored' variation as a mechanism that could facilitate rapid adaptation seems to be relatively recent.

      Delete
    4. No, it's not recent. I kept asking you to name experiment(s) done in the 1940's by scientists who *assumed* that bacteria developed antibiotic resistance by some mechanism "frontloaded" into the bacteria that turned on when the antibiotic was added. Basically they assumed the current creationist idea of "esoteric information" or "frontloading." However, when they did the experiment, they found that the bacteria had lots of genetic variation before the antibiotic was added, and the antibiotic resistance started from mutations that existed *before* the antibiotic was added, and was then amplified by Natural Selection *after* the antibiotic was added. Standing genetic variation was not assumed, it was the conclusion from experiments.

      Txpiper never answered what those 1940's experiments were, so I'll name two of them: Luria and Dulbruck, aka the variance test, and replication plating. The simple English Wikipedia has a good article on L&B.

      Delete
    5. “Max Delbrück and Salvador Luria showed that in bacteria, DNA mutations happen randomly. This means they happen at any time, rather than being a response to selection.”

      So mutations don’t occur in bacteria because of selection pressure. Hurrah. Thanks for clearing that up. Lots of people seem to struggle with that salient point.

      The interest in 'stored' variation as a mechanism that could facilitate rapid adaptation seems to be relatively recent.

      Delete
    6. Another very rough estimate: the human population one generation ago was 5e9. Mutation rate is roughly 3e-8 per site per generation. That makes 150 new mutations per site, or 50 copies of each SNP arising in just the last generation. I think about 10% of all humans who ever lived are alive today, but older, deleterious-but-non-lethal mutation are less likely to have survived. Anyway, I reckon a lot more than 50 independent copies of each SNP.

      Delete
    7. The interest in 'stored' variation as a mechanism that could facilitate rapid adaptation seems to be relatively recent.

      By "recent", are you referring to those 1940's papers Diogenes has cited?

      Delete
    8. txpiper,

      "So mutations don’t occur in bacteria because of selection pressure. Hurrah. Thanks for clearing that up. Lots of people seem to struggle with that salient point."

      I hope you're not struggling any more, since I've known this since I was a little kid, quite a long time ago. Variation accumulates because mutations just happen.

      "The interest in 'stored' variation as a mechanism that could facilitate rapid adaptation seems to be relatively recent."

      They're not "stored" variation, it's unavoidable variation. Mutations cannot be stopped from happening. Variation cannot but accumulate. Though Darwin didn't know about mutations, his presentation of natural selection started by explaining that there can be lots of variation within populations. That from such variation selection could happen. This makes variation a very old concept in evolutionary biology.

      Delete
    9. Tx, you neglected to add these sentences, after your quote:

      "So, Darwin's theory of natural selection acting on random mutations applies to bacteria as well as to more complex organisms.

      Delbrück and Luria won the 1969 Nobel Prize in Physiology or Medicine partly for this work."

      Which makes the entry like this:
      "The Luria–Delbrück experiment, 1943, also called the 'Fluctuation Test', asks the question: are mutations independent of natural selection? Or are they directed by the selection?

      Max Delbrück and Salvador Luria showed that in bacteria, DNA mutations happen randomly. This means they happen at any time, rather than being a response to selection."

      Can you point me to the part where they say IDdidit?

      Delete
    10. Txpiper: "The interest in 'stored' variation as a mechanism that could facilitate rapid adaptation seems to be relatively recent."

      Exactly what I refuted. As Ed pointed out, L&D got the Nobel Prize in 1969 for this work they did in the 1940's.

      70 years ago is not "recent."

      It's just creationists that have become aware of it recently.

      Delete
    11. Another old-timey experiment on genetic variation existing BEFORE the environment changes (e.g. before antibiotic is added): Replica plating.

      The relevant experiment is by Lederberg.

      Lederberg, J and Lederberg, EM (1952) Replica plating and indirect selection of bacterial mutants. J Bacteriol. 63: 399–406.

      Delete
    12. Graham Jones writes: the human population one generation ago was 5e9. Mutation rate is roughly 3e-8 per site per generation. That makes 150 new mutations per site, or 50 copies of each SNP arising in just the last generation. I think about 10% of all humans who ever lived are alive today, but older, deleterious-but-non-lethal mutation are less likely to have survived. Anyway, I reckon a lot more than 50 independent copies of each SNP.

      That's another way to compute it. You're right that that's an underestimate because we inherit variation from our parents.

      Delete
    13. "The interest in 'stored' variation as a mechanism that could facilitate rapid adaptation seems to be relatively recent."

      Yeah if 1859 is "relatively recent". The whole fucking concept of natural selection is built on the premise (and observable fact) that in all hitherto natural populations of organisms, there is at any one time a diversity of variation in traits.

      Delete
  2. Thank you, Larry!
    A lot of interesting and useful information here! But when you say that there are about 3,000 haploinsufficient genes in humans, and later that they found 3,230 genes that are highly loss-of-function intolerant, is that the same thing? You say that the authors found in average 54 heterozygous disease-causing mutations in each person, and that this is too high a genetic load according to most population genetics. But what is the typical population genetics number?

    ReplyDelete
  3. Most genetic diseases occur in genes that are not essential for cell survival. The most important genes are rarely associated with genetic diseases; a counter-intuitive fact that should be more widely disseminated.

    This is another example of the phenomenon underlying the famous story of armor on planes in WW2. Armor keeps planes from being shot down, but makes them heavier, less maneuverable, and they use more fuel. So armor should only be put where it is needed. Planes were returning with bullet holes in the fuselage so the military wanted to add extra armor to the fuselage. Abraham Wald of the Statistical Research Group recommended the opposite. Put armor where you never see bullet holes, since planes with bullets in those areas never return. Those are the most essential parts of the plane. Similarly, the most essential genes don't exhibit mutations.
    https://medium.com/@penguinpress/an-excerpt-from-how-not-to-be-wrong-by-jordan-ellenberg-664e708cfc3d#.1uyw74qwg

    ReplyDelete