TRIGGER WARNING: Long and detailed discussion of a genetics paper.
There are now several examples of modifications of an individual’s appearance and behavior by the environment, and of those modifications affecting the individual’s genes, usually by attaching methyl bases to specific nucleotides in the DNA sequence. This is a form of environmentally induced epigenetic modification. Usually, though, modification of DNA bases can also be coded by other parts of the DNA: bits of genes that “say” to the organism: “put a methyl base in position X of gene Y.” Most of the epigenetic changes we know of, and every example of such changes that are involved in adaptations, are caused not by the environment but by instructions from other genes. Maternal vs. paternal DNAs, for instance, are epigenetically and differentially modified by other genes, and fight it out in the fetus, since fathers have different reproductive interests from mothers.
So while environmental epigenetic modification of genes is known to exist, and even to be passed on for one or two more generations, this is not a common phenomenon, and is not known to be the basis of any adaptations that have evolved in organisms. It couldn’t be, in fact, since genetic changes involved in evolution must be passed on permanently. Environmentally-induced DNA changes, since they are “reset” and disappear within one to three generations, cannot in principle be responsible for adaptive evolution. Further, mapping of the genes causing adaptations within and among species show, in fact, that they are invariably caused by the substitution of DNA bases themselves, not epigenetic modifications of genes by the environment.
Despite the impermanence of environmentally-induced epigenetic changes, then, and the lack of evidence that they are involved in any adaptive evolution, people still keep banging the epigenetics drum. For epigenetics is a form of “Lamarckian inheritance”, and if the environment could truly cause the DNA to change in adaptive ways, and then be permanently inherited, that would be a non-neo-Darwinian form of evolution: it would, depending on its frequency, be a huge change in the way we think evolution works. (Darwin, in fact, suggested such a form of Lamarckian inheritance in The Origin, but experiments showed that he was wrong.)
Now, however, a new paper in Biological Psychiatry by Rachel Yehuda et al. (reference below; free download) appears to show not only epigenetic inheritance in humans caused by environmental influences, but that the environmental influence was the Holocaust, which induced trauma. The trauma, the researchers claim, methylated a particular DNA base in a gene related to stress response, FK506-binding-protein-5, or FKBP5. Compared to controls (Europeans who weren’t considered “survivors”), those adults with Holocaust exposure had more methylation at one DNA base in the gene—but not at two other bases examined. Further, the difference between controls and “experimentals” persisted in the next generation: the offspring of survivors had significantly lower methylation at the same site than did control offspring (the offspring of the controls who lived during the war but didn’t experience the Holocaust).
The authors are pretty careful in their statements, but do say this in the abstract:
This is the first demonstration of transmission of pre-conception parental trauma to child associated with epigenetic changes in both generations, providing a potential insight into how severe psychological trauma can have intergenerational effects.
Others have not been so careful, particularly science journalists, who either don’t read the paper or lack the expertise to evaluate it. Check out, for instance, this Guardian piece about the Yehuda et al. paper. Its author, Helen Thompson, seems completely unaware of the many problems with the study, and presents no caveats. It’s an example of bad science reporting. There’s another uncritical piece at Scientific American.
I won’t go into detail about Yehuda et al.’s methods and results, but will just give the major results and some of the problems that Matthew and I found with them.
Yehuda et al.’s paper stems from an earlier paper by Torsten Klengel et al. (reference and link below) showing that, among a sample of African-Americans, individuals classified as traumatized showed a decrease in methylation at several sites in the FKBP5 gene, a gene related to stress response. This demethylation was associated with an increase in the transcription (conversion into messenger RNA and then protein) of the gene, which itself was associated with changes in the stress-response system, presumingly leading to the increase of reported psychiatric disorders in the traumatized children.
Yehuda et al. looked at three regions of FKBP5, all in intron 7, associated with demethylation in the earlier study. What they found is shown below (“Holocaust-Affected Individuals” in red, “Controls” in white; parents on left [A]; offspring on right [B]). As you see by the asterisks, which mark statistical significance, one site (Bin 3, site 6) showed a significantly higher degree of methylation in the Holocaust parents compared to controls, while that same site in the offspring showed a lower degree of methylation in Holocaust offspring than in controls. There were no differences in the other two sites tested (one involves two nucleotides, the other three; see below).
This, then, shows the proposed epigenetic inheritance due to trauma. But there are formidable problems with and caveats about the data. Here are a few.
- Low sample size. In fact, extremely low sample size: 32 Holocaust survivors and 22 of their adult offspring, and only 8 parents and 9 offspring in the controls. While the results for the one site are significant, they are barely significant, with p values between 0.03 and 0.046 (0.05 is the cutoff). I realize that getting these individuals is difficult (most Holocaust survivors are now dead), but this bears repeating with a larger sample, or with (as the authors suggest) other forms of trauma.
- The sample size is inconsistent throughout the paper. As Mattthew wrote me: “Further confusion is caused if you check out the sample sizes – in Fig 4 n pairs = 10 (controls) and 23 (Holocaust), but in Figs 2A/B control n = 8 (parents) and 9 (offspring), while there were 29 Holocaust parents and 22 offspring. How can they get 23 Holocaust pairs if they only have 22 offspring?”
- The “Holocaust” experiencers didn’t necessarily include individuals who were traumatized. The paper notes the criteria: “Holocaust survivors were defined as being interned in a Nazi concentration camp, having witnessed or experienced torture, or having to flee or hide during WWII. Demographically comparable controls were living outside of Europe during WWII.” How many of these individuals were those that had to flee or hide versus those who were actually interned in camps vs those who witnessed or underwent torture? The data aren’t given. One might think, for instance, that those who fled Europe would be less traumatized than those interned in concentration camps.
- Most important from my viewpoint, the degree of methylation, compared to controls, is in the opposite direction between Holocaust parents and offspring. That is, an increased (and presumably environmentally induced) degree of methylation in parents is inherited as reduced methylation in offspring. This is not the way epigenetic inheritance is supposed to work: the parents’ genetic changes are supposed to be passed on, unaltered, to the children. If the environment causes a permanent epigenetic alteration of a gene (especially in the theories of neo-Lamarckian inheritance), then any adaptation must rest on the same alteration being handed down to offspring. The authors realize this problem, and try to get around it by proposing that the “hypermethylation” in parents causes lower levels of glucocorticoid hormones (a stress-related hormone) in the blood during pregnancy, and that causes “hypomethylation” in the DNA of the offspring. They call this “intergenerational biological accommodationism,” a fancy term for “unexpected reverse effects.” It’s not clear how this would be adaptive. (Since for all Holocaust parents include at least one mother, at least this hypothesis is plausible in principle.)
- The data are lumped in a weird way. The significant difference was seen at one nucleotide only (graph above), while the nonsignificant differences were seen in groups of 2-3 nucleotides lumped together. Why did they do this? As Matthew wrote me:
The combining is in the Bins, which are the bars on both figures. Bin 1 shows the % methylation for 2 CG sites. Bin 2 shows the % methylation for 3 CG sites. Bin 3 shows the % methylation for only *one* CG site. Bin 3 is also the only bin that shows a significant difference between controls and holocaust survivors/offspring.
The null hypothesis would be that this is just random noise – one way of testing this would be to see if similar single-site effects are seen for sites 1, 2, 3, 4 and 5 *separately* rather than pooled as they are on Figures 2A and 2B. Given that only site 6 has been linked with the effects of stress 6, they need it to be the only one that shows an effect. If the other single sites showed significant differences that would undermine their hypothesis.
With such small sample sizes, and barely significant effects of small amplitude, you’d want to see those single-site tests to be confident that the effect was limited to site 6.
- The authors don’t know the mechanism of epigenetic transmission. The classical route is that the DNA in germ cells (eggs and sperm) is altered by the environment, and then that alteration is passed directly on to offspring. This can’t be the case here because of the reverse changes between parent and offspring. Another way, which the authors suggest, is that the environmental modifications of DNA cause changes in the mother’s biochemistry or physiology in a way that affects the fetus’s DNA. While this is still a form of inheritance, it’s not one that’s especially reliable given the vagaries of physiology. Direct transmission via germ cells is more reliable, and would make epigenetic modification a more plausible way of causing permanent changes in the DNA. That doesn’t appear to be the case in this study.
- There appears to be a bit of inconsistency even in this negative correlation between degree of methylation between parents and offspring for Bin 6, site 3. The negative correlation and Figure above show a negative relationship: compared to controls, Holocaust parents are hypermethylated, while their offspring are hypomethylated. Yet the figure below shows a positive correlation, at least among Holocaust pairs, for methylation at the very same site. Maybe I’m misunderstanding something, but this is deeply confusing!
- I’m not convinced that the authors have completely ruled out environmental effects in the upbringing of the children, effects that themselves could have promoted hypomethylation of the genes. Although the authors eliminated the possibility of PTSD or trauma in offspring by giving them psychological tests, and by showing that the scores did not explain the correlation between hypermethylated parental genes and hypomethylated offspring ones, there could have been more subtle influences that weren’t tested for. This is still a form of “inheritance”, but it’s cultural inheritance rather than direct genetic inheritance. It resembles the “biochemical environment” hypothesis used by the authors to explain the negative correlation for methylation; but perhaps there’s a “cultural environment” effect as well. To rule this out, we’d have to remove children from their biological parents at birth and raise them in either a different “trauma” environment, or randomize them among environments. This of course is unethical and impossible in humans, though we can do it in lab animals like fruit flies and mice. Without such experiments, it’s hard to rule out subtle and even undetectable cultural influences that would affect how we judge this form of “inheritance.”
- Finally, there are no tests of whether the differences in methylation, be they hyper- or hypo-, persisted beyond the single parent-child generation. To be important in explaining evolutionary adaptation, such changes must last generation after generation after generation, world without end. This is not known for this study, nor has it been seen in any known case of Lamarckian “epigenetic inheritance.”
My conclusion is that while this paper is interesting and provocative, it suffers from formidable problems that call its conclusions into question. It needs to be repeated, preferably using other traumas (whose survivors are more numerous) and certainly using larger samples. The paper is intriguing, but certainly doesn’t mandate that we see this as a true case of epigenetically induced inheritance due to trauma, much less as a revision of how we think about evolution. (Note: the authors are not arguing that this kind of inheritance plays a role in evolution; that has been done by others.)
[UPDATE: There is another critique of the Yehuda et al paper, from a rather different point of view, but with similar conclusions, by John Greally, which you can read here – MC]
_______
Yehuda, R. et al. 2015. Holocaust exposure induced intergenerational effects on FKBP5 methylation. Biological Psychiatry, in press, http://dx.doi.org/10.1016/j.biopsych.2015.08.005
Klengel, T. et al. 2013. Allele-specific FKBP5 DNA demethylation mediates gene-childhood trauma interactions. Nature Neurosci. 2013 Jan;16(1):33-41. doi: 10.1038/nn.3275. Epub 2012 Dec 2.
First thing that jumped out at me was…how do they know that they haven’t discovered something peculiar to the European environment? Maybe even something as banal as the local allergens?
…and that, of course, is assuming that they actually picked up on anything at all, which seems quite questionable given the small sample size and effect.
b&
Yep. It is very curious to me that a 0.05 cutoff was used, as if this was some kind of psychiatric / sociology study and not a report of harder science. It reeks of a work that merely contributes to inappropriate publication bias.
Based on this synopsis at least, the paper fairly screams p-value fishing.
Obligatory XKCD: https://xkcd.com/882/
b&
That jumped right out at me too. They’re looking at multiple comparisons and seems might have been more wary of the “look elsewhere effect”. I would think at least a Bonferroni correction would’ve been appropriate, though I’m no statistician and I will probably not go read the paper.
Personally, I love the advice given to my close associates from a stats professor who we would occasionally run things by. “If it ain’t in the percentages, it ain’t there”.
Yeah just look at figure 2 from a distance. It’s pretty clear that there isn’t a whole lot to get excited about even if the sample size were enormous and something really is going on.
Yep. Geniuses think alike. I was looking at this yesterday, pulling back & using a straight edge, moving it around – just getting a feel for the proportions, and ultimately went… “meh”.
I worry a little about the reliability of Epigenisis as a theory. In gross terms Muslims and Jews should not have prepuces at birth! But, I can see a possibility that both high and low nutrition would have verifiable effects on a population over generations. Acquisition of ‘passed on’ genes from pathogens and parasites would also modify the human genome, as, no doubt,interaction with the human genome affects the development of pathogens and parasites. Now, about Toxicardia and cats…..
But don’t you see, Yahweh/Allah intervenes at conception and reverses the effect, to ensure that the boys still have foreskins, so that they can be cut off to fulfil his law. doG is great!
Toxocaria
Very interesting, and I agree with the critique that you provide. The paper should at least provoke further study with animal models, which is a positive thing.
I would leave animals out of it. There’s no shortage of people who went through all kinds of traumatic events (rapes, kidnappings, abuse by parents, human trafficking, wars, genocides, etc) in their lives and could be enlisted to participate in such studies. There’s no need to needlessly inflict additional trauma on animals.
But as Jerry points out, you can’t reliably obtain this information so only the fruit flies can help us.
Again, thanks for the clear explanation of the wonkery.
I cannot access the journal from home so I don’t know the answer, but how did they define their controls? Proper controls would be age-matched (people who lived in the same area at the same time but who were not subject to the horrors of the Holocaust) and their descendants.
I do not have access to the journal from home but I have a question about the controls. How did they define them?
Proper controls would be age matched (people who lived at the same time and place but who did not suffer the horros of the holocaust) and their descendants.
They’re sort of age-matched; there was no difference in age or other demographic criteria between controls and experimentals, though note that the sample of control adults was VERY small.
So they didn’t check if the effect was simply due to malnutrition?
I can’t seem to get a full text of the article, so maybe I’m wrong, but it seems to be a bit of an indirect approach. These days they could (try) to find a couple where one parent was a Holocaust survivor and the other was not. They could then directly follow the transmission of the methylated and un-methylated sites they are interested in through (hopefully more than a few) children and ideally even grandchildren. I know the survivors are becoming rarer and rarer but, following the transmission directly like this even in one family would be pretty convincing I think. A control family from the same area as the test family should be relatively easy to find.
Interesting though!
There seems to be a need for scientific papers to be summed up in the abstract so that journalists stop mis-interpreting.
Would sitting down to write an abstract cooperatively with a journalist help?
The abstract needs to be written for colleagues, and no,you don’t want a journalist in the room! But papers of potential interest to the lay public almost always come with a press release. Generally the latter is issued by the university or lab’s public relations officer, who is typically a trained technical writer or journalist. Usually the investigator will be able to preview and vet the press release. The problem is that this officer has every incentive to oversell the paper’s findings, and so, often, do the journalists who pick up the story.
… and the investigators.
Thanks for this.
Seconded
Thanks for sorting through this, and for collecting Matthew’s comments. The “inversion” of methylation in the offspring seems like a deal-breaker to me.
All very interesting, but if I was intending to study this phenomenon, I would look for easier subjects. There must be tons of war veterans with children from Vietnam to Pakistan. Granted not as many women as men, but why work with such a diminishing sample of people?
Also, if this phenomenon is real, it would be far easier to clearly demonstrate it using non-human subjects. So, once it is clearly understood in drosophila, or mice, it might be then worthwhile to check out humans.
Not to mention all the children of war rape, from Bosnia not that long ago to Iraq today….
b&
Yes, the last seventy years has not been lacking in comparable traumas.
Or Africa. Almost anywhere in Africa.
Cambodia; Indonesia in the sixties; the Cultural Revolution…
Given the age of Holocaust survivors, they should have been able to follow them for at least one, and probably two more generations. That they didn’t says something to me.
In Figure 4, I think you are indeed missing something. 🙂 You wrote:
>the figure below shows a positive
>correlation, at least among Holocaust
>pairs, for methylation at the very
>same site.
The text makes clear that the regression line in the figure, r=.441, is in fact for *all* the points in the chart (red and blue), not just the Holocaust pairs. The authors then add to this, saying:
>This association was primarily driven
>by the Holocaust-exposed families (r=.569,
>n=23, p= .005 for Holocaust-exposed,
>compared to r=.370, n=10, ns for controls).
But in so doing, they seem to be making the common error of comparing the significance of two results, not the results themselves. If you compare r=.569/n=23 and r=.370/n=10 with Fisher’s r-to-z transformation, you get z=0.56, which is hopelessly non-significant. In other words, there was no difference between the contribution of the survivors and controls to this regression line.
(This still doesn’t help with the problem that the regression line seems to show the opposite of the inverse relationship that was reported, of course; it just shows how bad the statistical analyses are here.)
Numbers guy says: it is the fact that the slope of the line is less that 1 that indicates F1 tends to be lower than F0. The overall upward trend is expected, that is that more methylated parents (F0 on X axis) will produce more methylated offspring (F1 on Y axis). If it were even, the slope would be 1, and a 60% parent would produce a 60% offspring, etc. If you go through the red points one a time, most of them are a parent which produced offspring with lower methylation.
Having said that, the overall effect seems to be a small +5% bump in F0, and a small -5% drop in F1. Pretty small potatoes for methylation in the 50% range.
You may be interested in why EpgntxEinstein, the blog of the Center for Epigenomics at the Albert Einstein College of Medicine judges the paper as the Over-interpreted epigenetics study of the week.
Was just going to post this. It’s pretty telling that at folks who are experts at epigenetics don’t seem to like this paper either.
Trigger warning appreciated since, as a layman who occasionally wanders into the morass of an academic paper, I no doubt suffer from acute PTSD.
Two things occurred to me as I read this. First, was the control group also from the same generation, i.e., the Second World War? I think you would have to be very careful to have a control group in Europe that didn’t have some sort of trauma from the period. Second, as Robert Seidel points out above, have the researches eliminated other environmental influences, especially malnutrition (which could also be post-war), that might cloud the sample, and, perhaps, also affect the control group? Aside from the sample size, I would think you could also test this against other victims in Russia and China (and other places), where there was extensive non-criminal detention under deplorable conditions.
Thanks very much for parsing the paper. Thanks, too, for bringing in Matthew’s expertise. I agree with Mark Sturtevant’s suggestion. This should have been reported as an interesting observation with much work left to be done. Further work using animals should be done after researchers develop a much more robust method to pair-match populations and isolate the variable.
But regarding epigenetics in general – isn’t it possible for even temporary changes in haplotype to result in a change to a population’s allele frequency for non-epigenetically induced genes?
Seems the epigenetic changes would have to affect reproductive fitness pretty dramatically . I’ve heard of such in plants but not animals. But I think it usually takes very strong, persistent selective pressure on a significant part of a population to change allele frequencies in a permanent way, much stronger than those usually attributed to epigenetics.
A fair point. I suspect that the math would prove you correct.
I don’t think there is an inconsistency between Figure 2 and Figure 5, if I understand you correctly. 🙂
For example, if F1 = 0.3 * F0 + random error. There is a positive correlation between F1 and F0. However, the mean of F1 is probably smaller than the one of F0 due to the multiplication of 0.3.
This was my thought as well. Are there no Holocaust survivors with adopted children? Or orphaned children of Holocaust survivors adopted by other families? Those would have been useful control groups.
(As an aside, I’m not completely clear on what environmentally induced methylation actually means. Is the claim that methyl groups attach to DNA spontaneously at different rates under different environmental conditions? Or is the attachment always mediated by an enzyme, but the expression of that enzyme varies with environmental factors? If the latter, then the distinction between genetically coded and environmentally induced methylation seems rather fuzzy.)
I wonder about that too, but with a problem of getting enough Holocaust survivors still alive as subjects, the number with adopted children will be lower still (and age of adoption might make such results fuzzy, too).
But you don’t actually need the parents to still be alive in order to assess methylation frequencies in their children.
But the experiment was with pairs of Holocaust survivors and their offspring, and the control of pairs of people who were alive at the time but unaffected by the Holocaust, and their offspring. So I think you do need a parent and child combination to test each time.
Thanks for the post, Professor Coyne. Because of your attention to the subject, I’m starting to think I ‘get’ epigenetic and the case against it!
Nice summary. Many thanks, PCC!
Reblogged this on Nina's Soap Bubble Box.
Meh. Small heterogeneous samples, convoluted analysis, convoluted explanation. Some of the local trainees had already asked me about this and I had skimmed it and triaged it to the ‘might be interesting, but probably not’ bin. Prof CC’s analysis (from someone more genetically astute than I) is quite helpful. I’d be interested to know if this can be replicated in, say, survivors of combat trauma. Until then, I remain poised over the fence.
I was hoping you would write about this study – I’ve seen it shared a few times already on Fb. I have lots of statistical thoughts, but it is really nice to read about the biological side of things here.
From a purely statistical point of view, the study gives no evidence at all of any claimed effect. The study is *extremely* underpowered, as Jerry noted in his first bullet point – *way* too small a sample size. What this translates too analytically is seriously inflated effect estimates that have significant possibilities of pointing in the wrong direction (the so-called “Type M” and “Type S” errors – see Gelman: http://goo.gl/8gJlOe). I would say it is far more likely than not that no effect exists between the case-control subjects because of this, a claim that is substantiated by the fact that the two proposed “effects” point in opposite directions.
To explain a bit further, the effects that the authors choose to focus on are the percentage methylation figures at site 6 in the parent and the offspring sample. I can’t access the article from my apartment, but just eyeballing their Figure 2 seems to indicate an effect size of about 8% between case and control groups at site 6 for the parent subset. Let’s focus just on the parent subset, since the same analysis goes through essentially verbatim for the offspring subset.
Again, I can’t currently access the article, but using the p-values Jerry reports above, we know that the standard error of the relevant estimate is between about 3.69 and 4.01 (corresponding to p-values of 0.03 and 0.046 respectively). Let’s be generous and use the lower standard error: 3.69. If the true effect between case and control parents at site 6 was only 1%, then we would *expect* to obtain an estimated effect size of about 8% with a sample this small; i.e., we would expect our estimate to exaggerate the true effect by a factor of about 8, and this is exactly what we see in the study. Furthermore, there is a 22% chance that this estimate is in the *wrong* direction; i.e. is positive when the true effect is negative.
Moreover, if the true effect between case and control parents at site 6 was only 0.1%, we would expect the estimated effect size to exaggerate the true effect by a factor of about 87. So again their estimated effect of 8% is exactly what we would expect in this case. Here too there would be a 46% chance that the estimate was in the wrong direction. The switched sign of the effect between parents and offspring further fits nicely into this picture. Even neglecting the other points that Jerry already made about the study, this analysis alone shows that the results of the paper are perfectly consistent with a zero true effect.
This type of design analysis (proposed by Gelman, et al.) should be standard practice, in my opinion, especially when it comes to evaluating small sample studies. While it’s certainly interesting from other points of view, statistically speaking, the study in question gives no evidence of a real effect of interest.
Also,
“The negative correlation and Figure above show a negative relationship: compared to controls, Holocaust parents are hypermethylated, while their offspring are hypomethylated. Yet the figure below shows a positive correlation, at least among Holocaust pairs, for methylation at the very same site. Maybe I’m misunderstanding something, but this is deeply confusing!”
This is a version of Simpson’s Paradox: https://goo.gl/nbk8QM. When we view the data grouped as 29 case parents vs. 22 case offspring and examine the trend, we see a negative relationship among the case subjects; i.e. offspring methylation is lower than parent methylation. But when we view the data grouped individually as 1 case parent vs. 1 case offspring 23 times (ignore the inconsistency in the sample size), we reduce that previously observed negative relationship to a single point: a red square in Figure 4. This negative relationship between (parent,offspring) pairs is now reflected in the fact that most of the red squares fall below the diagonal (parent > offspring). When we view the aggregate trend of these 23 pairings in total though, we see the positive relationship that reflects the fact that a parent with higher percentage methylation is likely to have offspring with higher percentage methylation than a parent with a lower percentage.
Oy just getting off work, and it looks like the majority of the week will be this way, but am bookmarking this to read over the weekend! Keep the science posts coming, please 🙂 Thanks for writing about this article – was hoping you would!
An Israeli friend remarked on the generation of children of Holocaust survivors who were given a special name in Israel (which I can’t remember) because of the psychological and social difficulties they often had – difficulties that derived, my friend said, from being brought up by people who were suffering from deep traumas. But, according to her, subsequent generations of Israeli children haven’t had the same sort of difficulties, which would suggest that there is no genetic reason for those difficulties.
I’ve just updated the post with a link to this critique, which has a different approach but comes to similar conclusions: http://epgntxeinstein.tumblr.com/post/127416455028/over-interpreted-epigenetics-study-of-the-week – MC
Fascinating to read even if I don’t understand the details. It seems to strongly confirm the skepticism seen on this forum.
I’m really grateful you wrote this post Prof. Coyne. I’d seen the original piece in the news and it certainly had me scratching my head.
Can the changes actually be environmental via the experiences being taught through conscious means like talking but also subconscious through the behaviour the children learn from without questioning. The Mark of Cain as it has been described.