Hacker News new | comments | ask | show | jobs | submit login
Potential DNA damage from CRISPR has been ‘seriously underestimated’ study finds (statnews.com)
288 points by valeg 7 months ago | hide | past | web | favorite | 49 comments



Note that this study covers 'on-target' effects.

As often covered in the news, CRISPR-mediated gene editing can be used to edit a DNA sequence to another, desired sequence. In these applications, it's very important that no other mutations besides the desired mutation are introduced. Such additional mutations would be 'off-target' mutations.

In a second kind of experimental approach, researchers can use CRISPR technology to make arbitrary mutations at a desired site in the genome. This approach cares less about the resulting damage; the goal is simply to destroy the existing gene.

The referenced article[1] seems to characterize the types of mutations that arise when people use CRISPR to ablate a gene (i.e., the second approach). It would be interesting to see if the range of DNA damage is restricted when CRISPR is instead used to change the sequence of a targeted DNA segment into another, desired sequence.

[1] https://www.nature.com/articles/Nbt.4192


The first thing to realize is that the crispr/cas9 strategy only produces damaged DNA (double stranded breaks, DSBs). It is not supposed to repair the DNA or do anything else.

You have five main responses to this:

1) Necrosis/Lysis (cell dies due to malfunction)

2) Apotosis (cell kills itself)

3) Senescence (cell doesn't die but stops dividing)

4) Non-homologous end joining (NHEJ, the free ends of the damaged DNA are just pasted back together by the cells repair machinery)

5) Homology-directed repair (HDR, other DNA around that has a similar sequence to the free ends gets used by the cells repair machinery to paste something in between; eg you have two of each chromosome so use the corresponding sequence on the other one to fill in the gap)

The second thing to realize is that cells containing the targeted DNA sequence will be more likely to have damaged DNA than those without. So once a cell is "mutated" (either because it just existed beforehand, or because it survived the crispr/cas9 treatment), it will have a survival advantage.

The third thing to realize is that often these cells they use can divide multiple times per day, I've seen as high as 12 times a day for activated T-cells. That means in two days a single mutant cell could possibly produce 2^24 = 16,777,216 offspring cells containing the mutation.

Anyway, I think the efficiency of this procedure may have been far overstated. There may be only a very few cells in each experiment that actually get edited, but a few days later when they check, those (already mutated) cells dominate the population due to their resistance to the crispr/cas9 treatment.

Eg, look at extended data fig 1h from this recent paper: https://www.nature.com/articles/s41586-018-0326-5/figures/5

The more cells that die, the higher percent of "edited" cells later.


> 5) Homology-directed repair (HDR, other DNA around that has a similar sequence to the free ends gets used to paste something in between; eg you have two of each chromosome so use the corresponding sequence on the other one to fill in the gap)

I wanted to elide the details, but this is what I was referring to when I discussed CRISPR applications that try to change a nucleotide sequence.

This of course leaves discussions of base-editing aside[1]

1. http://www.sciencemag.org/news/2017/10/novel-crispr-derived-...


Yes, you mentioned the two possibilities that everyone likes to focus on, but not the others which (I suspect) are much more common.

EDIT:

I really mean that I think in some of these experiments, where they only try to cause NHEJ, they just already had mutants in the population and selected for them. Its possible not a single cell was successfully edited but you still have so many cells with indels at the targeted sequence afterward.


Indel mutation rates aren't high enough, and the number of cells screened for mutations is too low for this explanation to suffice.


Do you have a paper in mind? From what I've seen 0.01% - 10% of controls have indels at any given site (obviously this depends on treatment, cell type, etc). You could call that noise of course, at the very least they can't detect rates below those using current methods.

I'll go find some refs on it if you ask.


You must be misunderstanding what you've read. There's no way cells survive with indels at 10% of their base positions, that's not possible.


Thats not what I said. I said some significant percent of cells (I remember as high as 10% from one paper, but usually its closer to 0.1%...) seem to have indels at any given locus. Either that or the methods used to measure this are too noisy to detect mutants that exist at those rates.


How is the per-base indel rate not 10% if 10% of cells carry an indel at any given locus?


  Reference: 
  GAATTC

  Variants:
  Cell 1: TAATTC
  Cell 2: GTATTC
  Cell 3: GATTTC
  Cell 4: GAAGTC
  Cell 5: GAATGC
  Cell 6: GAATTT
  Cell 7: GAATTC
  Cell 8: GAATTC
7/8 cells contain the reference nucleotide at any given site. Ie, 12.5% of cells contain an indel wherever you look.

For six cells, 1/6 nucleotides differ from the reference sequence (16.7% per base indel rate). For two cells, there is no difference (0 % per base indel rate).

Even though 12.5% of cells contain an indel at any given site, no single cell contains 12.5% indels. There is no reason the two values should be the same.


You clearly have no idea what you are talking about.

1) There are no indels in your example, only nucleotide substitutions.

2) There are six substitutions in 48 bases, for a per-base substitution rate of 12.5%. The fact that none of the individual cells contains 12.5% substitutions is totally irrelevant.


>"There are no indels in your example, only nucleotide substitutions."

It works perfectly fine as an example. A single nucleotide change could be due to an indel:

"Indel is a molecular biology term for an insertion or deletion of bases in the genome of an organism. It is classified among small genetic variations, measuring from 1 to 10 000 base pairs in length,[1][2][3][4][5][6][7]" https://en.wikipedia.org/wiki/Indel

>"There are six substitutions in 48 bases, for a per-base substitution rate of 12.5%. The fact that none of the individual cells contains 12.5% substitutions is totally irrelevant."

In the experiments they are counting how many cells contain a mutant (eg via GFP expression or not). Thus the number you want is percent of cells that contain a mutation at a given site.

You are the one calculating some other number... Remember what we are discussing? Its percent of cells with a given mutation and mutations per cell, the numbers I calculated:

>"There's no way cells survive with indels at 10% of their base positions, that's not possible."

>"You clearly have no idea what you are talking about."

There seems to be some heavy dunning-kruger going on with your post.


I asserted that saying that 10% of cells have an indel at any given site is equivalent to saying the per-base indel mutation rate is 10%.

You posted an example purporting to disprove this, but instead, illustrated that you don't know what an indel is, and that the two statements are indeed equivalent -- at any given site, 12.5% of cells are mutant, and that the per-base mutation rate is 12.5%.

The only one with Dunning-Kruger is you.


Here is what I am responding to (also keep in mind I was using indel to mean "indels and substitutions", ie whatever may be counted as an error during NHEJ):

> "There's no way cells survive with indels at 10% of their base positions, that's not possible."

> "How is the per-base indel rate not 10% if 10% of cells carry an indel at any given locus?"

I figured you were referring to per base within each cell. Your overall "per base" rate is getting the average (across cells) number of mutated sites. The survival of a cell is determined by its own genome, not the average of the population its been grouped into.

However, it is the case that if x% of cells have a mutation at any given site then there must be at least some cells in this group that contain x% or greater mutated sites. Is that what you meant? Because that's not a bad point. I don't know where that 10% value came from and have no attachment to it, just assume that one is measurement noise for now, but even cells with 0.1% of sites mutated seems like a lot.

There doesn't seem to be many whole genome single cell sequencing results available yet, which is what I think is needed here. Here is one that reports ~ 15k SNVs, 50-100 "micro-CNVs" (ie 10-100 kb copy number variants, fig S4), and 290 indels:

>"We call an SNV if there is a called NR allele and: (a) the total read depth in the bulk sample is above 15; (b) no bulk read has the NR allele; (c) if two SNVs are within 100bp from each other, both are discarded. With this procedure, we called 15,940 SNVs on the autosome of sample BJ1...We called 294 putative INDELs from the BJ1-BJ2 pair, the negative control" https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5538131/

So taking the smallest and biggest numbers of possible here gives:

  100%*(15e3 + 50*10e3   + 290*1)/3e9    = 0.02% of bases affected
  100%*(15e3 + 100*100e3 + 290*10e3)/3e9 = 0.4%  of bases affected
This seems to ignore any very small or large mutations that may affect a given site, and not all the mutations will simulate the effect of NHEJ... but this makes me think it isnt impossible for 0.001-1% to happen, especially in vitro.


typo in the last post: >"This seems to ignore any very small or large mutations"

I wrote this before I added the "smallest" calculation. So only the very large structural changes are ignored (> 100 kb).


Ok, I guess no one calls it an indel if theres no net change to the length. It doesnt really matter to the example if its an indel or substitution.


Actually just look at the controls in this paper:

  Fig 1b: 0.31%
  Fig 3b: 1.12%
  Fig 4b: 0.18%


Which paper? OPs, or the one about T-cells that you posted?



Those are not direct measures of indel rates, those are measures of PigA and CD9 protein levels above the threshold set by the investigators. The main text makes it clear that those percentages should be regarded as background noise, and they adjust for it when comparing CD9 and PigA targeting.


They can't detect the presence of mutants if hey occur at frequencies at or below the background rates. I mentioned this multiple times. Its simply that no one knows if they are present or not. Earlier:

>"You could call that noise of course, at the very least they can't detect rates below those using current methods."

>"Either that or the methods used to measure this are too noisy to detect mutants that exist at those rates."


Extended data figure 1h is testing the optimum pulse settings for the electroporation machine. There is a well-known trade-off between efficiency and viability (too small a pulse and no DNA is transferred into cells, too strong a pulse and the cells are destroyed). The efficiency increase at lower viability cannot be attributed to pre-existing mutants, because they're assaying for the presence of a GFP transgene inserted by homology-directed repair, not indels in general.


>"The efficiency increase at lower viability cannot be attributed to pre-existing mutants, because they're assaying for the presence of a GFP transgene inserted by homology-directed repair, not indels in general."

Yes, for the HDR case it is usually extremely unlikely (barring contamination) that the mutant already existed. If you reread my other post I specifically refer to the NHEJ case when discussing pre-existing mutants.

However, a single T-cell could (T-cells can divide up to 12 times a day[1]) expand to as many as 2^48 = 2.81475e+14 daughter cells in the 4 days of that study (see panel A of the same figure). They started with only 1e6 cells in the culture. Therefore these results could be explained by editing a single cell that then proliferates to make up the 30% of the population (300k cells) they see.

[1] http://journals.plos.org/plosone/article?id=10.1371/journal....


Try reading the methods section of your T-cell paper. After electroporation, they cultured at constant cell density, which only required adding media every 2-3 days.

So, on average, only one cell division happened before measurement, unless you want to claim GFP stimulates T-cells to hyperproliferate. My intuition is that cells with the transgene will actually be slower than their competitors, since they have to waste energy to make GFP.

The PLoS One paper you've linked is about in-vivo hyperproliferation of T-cells in response to an ongoing infection, not about in-vitro primary culture in minimal media.


>"So, on average, only one cell division happened before measurement, unless you want to claim GFP stimulates T-cells to hyperproliferate. My intuition is that cells with the transgene will actually be slower than their competitors, since they have to waste energy to make GFP."

No, the claim is that double stranded breaks lead to cell death and senescence amongst the non-mutated cells thus selecting for the cells that lack the target sequence (because they were just "edited" or otherwise). We don't know how many cells survived the treatment, or how fast the different populations were dividing, or the rate at which these cells were dying. From extended fig 1a just see that 2 days after treatment there were ~500k total cells from an original ~1 million and ~100k of those expressing GFP. Then four days after treatment there were ~1 million total cells with ~ 300k of them expressing GFP.

Are T-cells capable of dividing quick enough so a single cell can generate 100k in 2 days and 300k in 4 days? The literature says yes. Maybe there was an initial thousand edited cells that divided on average ~3 times each day for two days and then once the last two days as the surviving non-GFP+ cells recovered. Is that what happened? I don't know. They didnt report the details of what happened immediately after the treatment.

>"The PLoS One paper you've linked is about in-vivo hyperproliferation of T-cells in response to an ongoing infection, not about in-vitro primary culture in minimal media."

Yes, it only demonstrates how fast they could be dividing, setting an upper bound. We don't know under the conditions in this paper.


>Academic scientists were less dismissive of the new study, in Nature Biotechnology. One leading CRISPR developer called it “well-done and credible,” “a cautionary note to the [genome-editing] community,” and consistent with other research showing that the DNA cuts that CRISPR makes, called double-stranded breaks, “can induce the types of genomic DNA rearrangements and deletions they report.” He asked not to be identified so as not to jeopardize business relationships with genome-editing companies.

no. no. no. no. that is not good for scientific integrity


Here's another good writeup (imho better than OP's) on this study with some potential ways forward:

https://theconversation.com/crispr-cas9-gene-editing-scissor...

It's from Gaetan Burgio, who runs his own research group using CRISPR


CRISPR is a revolutionary research tool that enables researchers to efficiently study the effect of changing genes, commonly by "knocking them out". CRISPR basically cuts a specific piece of DNA from the genome, so you can study how a cell / animal / disease model is different with and without the gene -- testing causality of gene and phenotype

when CRISPR knocks out a gene, the cell tries to repair the DNA using its natural dna repair mechanisms. this can lead to unexpected insertions or deletions of DNA. researchers are widely aware of this risk and have studied it

however, this paper exposes shortcomings of prior assessments of this risk. specifically, prior studies thought that the unexpected insertions or deltions of DNA were usually small, less than 20 nucleotides. there were observations of larger "indels" but these were thought to be rare. there are a few other shortcomings of the research into "crispr induced lesions", the article lists them in detail

the scientists suspect that prior research has significantly underestimated the level of genomic alteration induced by crispr, and that some of this alteration could cause disease like cancer

the authors first tried to delete a gene with crispr by targeting the protein coding part of the gene, and also non protein coding parts. the crispr targeting the protein coding part of the gene showed significnat knockdown of the gene (over 97%). however, so did the crispr targeting the non protein coding part of the gene (up to 20% of cells did not express the gene). some crispr constructs targeted sites that were over 2,000 base pairs from the nearest protein coding site, and 5-7% of those cells did not express the gene. they determined that loss of the gene was due to loss of the protein coding region, not just regulatory elements in the non protein coding part. they did similar experiments in other cell lines and with other genes

so this suggests that researchers may have wildly underestimated the amount of genetic chaos induced by crispr. it is more possible than we thought that people receiving crispr therapies may also have pathogenic crispr induced dna damage

the paper references six studies, all in China, currently using crispr to treat humans. 3 of the studies use crispr to "knock out" PD1, a molecule that blocks the immune system from attacking cancer, 2 use CRISPR to make cells express CD19 to attack cancerous blood cells, and one study tests an HIV therapy. anti PD1 antibodies and anti CD19 cell therapies using other gene editing techniques are approved and highly effective. anti pd1 antibodies are expected to generate over $20B in sales in the next few years. so these treatments have strong scientific basis supporting their potential effectiveness, but we may have drasticlly underestimated the risks


> currently using crispr to treat humans

Correct me if I'm wrong, but my impression was that crispr is used for treatment only in cases where there is nothing else left and the alternative is sure death


CRISPR is not used for any treatments yet because we dont know if any crispr based treatments 1) will work in humans or 2) are safe. there are no FDA approved crispr based treatments. there is virtually no human data on whether crispr treatments are safe in humans.

in cases where a patient has exhausted all treatment options, they can choose to enroll in a clinical study of an unapproved medicine. there are generally dozens if not hundreds of clinical studies for a given cancer type. if a patient is eligible for trials, these days they have plenty of options (however many pts are not eligible for studies). crispr based treatments would be one of many options

the 5 cancer studies of crispr treatments cited in the paper are all chinese studies, and all of them are basically crispr versions of therapies that work. for example, three of them use crispr to remove a gene that makes a protein called pd1. there are approved drugs that treat cancer by blocking PD1, and the safety of these is much better understood than crispr therapies.

for the patients in the crispr studies in china, there are drugs that essentially target the same disease related protein that are known to be fairly safe. so the patients could get a drug that probably works roughly as well as the crispr treatments but with less unknowns about safety


Well, there's that crazy guy who's trying to make it available to the public for at home experimentation. He already tested it on himself, trying to get himself bigger muscles as I recall. One of the comments I saw from a microbiologist included the words "the most cancery cancer that ever cancered".


It's also likely that treatments where you can actually screen for mutations extensively will be more successful. The issue here is you get a lot of mutations that aren't desirable, if you can grow lots of different cultures and ID the ones that are OK, maybe that can work.


This isn't exactly surprising, in that CRISPR is a bacterial anti-bacteriophage defense system.


Yes, it it seems the bacteria jam it anywhere in their DNA (perhaps in many places at the same time) in the faint hope it will work to jam the machinery of the virus/phage it can work on in their successors. It looks as if a degree of specificity needs to engineered in to increase it's precision, like zinc fingers or something comparable.


No treatment can ever be proven to be safe. It can only be shown that you have yet to observe harm.


Even something such as ginseng that has a 5000+ year history of use with no real known side effects?


I suppose some people would argue that 'no known side effects' extends to 'no known effects' as well in this case. Scientific evidence that it does anything at all in humans appears to be non-existent to very weak at best.


Well, you can believe what you may, of course. But at least no one is dying from it...you can't say that about very many modern medicines, even Tylenol kills a lot of people.

Probably half of the world's peoples have had good effects from ginseng, whether that is "placebo" (as scientist and western doctors might say...) or based on chemical compounds that haven't been extracted yet (they are still extracting them from another plant medicine in use for over 5000 years called Cannabis...scientists have no idea about most of the 500 or so in that plant either), or based on some kind of spiritual element as is said about Ayahuasca and Peyote is really all besides the point. Ginseng has cured many people and kept many more from ever getting ill in the first place.

Explain it however you want.


I have family who work in TCM, and I think their guidance is pretty decent: if it's serious go see a doctor, get medicine, get better. If it's not serious (you're okay with living with it), go with the herbs, because the downsides are much smaller (and the effect sizes too), but that's okay because it's not a serious disease.

But I must stress this: if you have cancer, don't eat herbs and pray, go to a damn doctor.


Asbestos was in use for 3900 years (roughly :o) ) before it was noticed that it can be dangerous (we now treat it as utterly deadly in the UK at least, I wonder if that is proportionate though - I am not slightly qualified to say!)


good point. and people used to paint their faces with lead, use leeches as medicine, and so on and on. can't argue with you there! weren't asbestos used in early cigarette filters, even? i think they were.

look: i'm not suggesting to throw the baby out with the bathwater here! might as well use the tools we have the best we can. but why not look to nature and various medicinal traditions too?


as far as I know only in one brand of cigarette. it was a particularly hazardous form of as estos. ironic since it was marketed as being safest healthiest cigarette.


Leeches are used in medicine today, particularly after limb transplant microsurgery.


Another interpretation could be that the study found just how much our DNA deteriorates over time when replicated..


IMO it will be hundreds or thousands of years (if ever) until humans will fully understand the intricacies of DNA...the complexities are just enormous.


We went from not knowing how hereditary information is passed on to actually manipulating it in humans in the span of a few decades, don't be so pessimistic !


I ran into Liz Parrish at a longevity convention last year and she looked pretty damned good for being 2 years into CRISPRing herself. I cannot begin to fathom how wildly disruptive widespread modification of somatic cells would be to medicine and the drug industry. They'll just have to do it in China first I guess.


You should probably read this thread: https://news.ycombinator.com/item?id=11560943


Even if she lived to 150, a skeptic would just say it's a fluke and we have to wait 110 years for a large cohort double blind placebo controlled multi-center study of 40 year olds to finish at $800,000 a procedure before we can say that the intervention worked. It wouldn't get past the IRB either.

Longevity science, even if it works, will be impossible to prove to skeptics for purely practical reasons.




Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: