Negative control sequencing

[captainentropy]

What are people in the field doing with respect to sequencing DNA captured in negative controls (IgG, preimmune serum, beads only, etc.)?

I hear very little about doing this. A labmate says she has talked with some labs doing major ChIP-seq projects and they always sequence the negative control.

Honestly, I don't always get measurable DNA from the neg control. One ChIP recently I recovered 480ng with the antibody and 7ng with the IgG+beads control and another was 580ng and 0ng. So, in these two cases it wouldn't even be possible. But many times I do pull down enough to make a library even though the enrichment of IP over negative is quite good.

Any thoughts/comments?

[mudshark]

given the fact that i once calculated that about 2% of my reads in my IP library actually were derived of regions bound by my target I am really surprised that you end up with so much less in an IgG or bead-only control.

anyway 480 ng with the antibody is hell of a lot, what is your target?

[captainentropy]

mudshark, could you tell me how you determined that only 2% of your reads were bound by your target? Assuming you had negligible DNA bound could it be a really bad antibody then?

As for my stuff keep in mind the amount of DNA pulled down is a function of antigen abudnance, amount of chromatin in the IP, and antibody quality. In my case I wasn't using much chromatin, 10 ug, I believe, but the protein is of very high abundance in the cells we use. Based on our calculation from past (non-ChIP-seq) experiments it occupies about 2% of the genome. My results were much higher, true, but perhaps the previous estimates, which were made with BACs, was too low. I also use an uncommon method of chromatin preparation. After lysing my crosslinked cells I spin the lysate through a cushion of 8M urea at 45-55K RPM for ~7-12 hours. The advantage of this is that the pellet that emerges at the bottom of the tube is 100% crosslinked material. There is no RNA, no free DNA, and no unbound proteins. No unbound DNA means less background. No unbound proteins means higher true signal. The antibody is pretty fabulous too though. Very clean. We have a couple of custom antibodies that aren't as strong as this commercial one. So all of that results in a lot of ChIPed DNA.

A nearby fly lab routinely recovers microgram quantities of ChIPed DNA. But they start with milligrams of chromatin and are ChIPing histones and histone PTMs.

[mudshark]

hi captainentropy

we usually chip from 50-60µg chromatin and get around 5-10 ng DNA in IP. i would not say that the antibody is particularly inefficient, it is comparable to many others. in comparison to anti histone ABs the yield is of course very low.

about the 2% i mentioned. i basically know all my binding sites because i have a cool prior knowledge system and anyway mapped the stuff previously by ChIPchip.

.. quite interesting your chromatin prep protocol!

[supertjejen]

hi captainentropy!

I'm having the same problem, I'm about to sequence and I have 50 ng in my IP and 4 in my negative control. for another antibody i have 230 ng in the IP and 1 ng in the negative control. I'm thinking about using the input as negative control instead apparently some do that. See "ChIP–seq: advantages and challenges of a maturing technology" Peter J. Park. But I'm knew to sequencing I'd appreciate feedback, have anyone any ideas?

[mudshark]

very good point. we only sequence the input as a reference. of course that does not control for any bias in unspecific bead-binding etc. an IgG control has its own bias anyway.

in ChIPchip the standard is to use the input as a reference. why should this be a bad choice for ChIPSeq?

[captainentropy]

supertjejen, since most people will use ChIP conditions that result in no background (chromatin/DNA sticking to the beads or Fc region of the antibody), one can't very well sequence that background right? Presumably then all the ChIP-seq signal is due to DNA recovered from the immunoprecipitation. However, it's now clear that a lot of those peaks are the same as what is seen from sequencing a control sample that was not immunoprecipitated (input). So be careful not to use the terms "negative control" and "input" interchangeably as they are two different things. The sequencing of the input is a good idea in order to control for the patterns that arise due to the uneven distribution of crosslinks in the nucleus due to regions of heterochromatin (more proteinaceous and thus more crosslinked) and euchromatin (less proteinaceous and thus fewer crosslinks). The regions with fewer crosslinks will sonicate/digest and decrosslink easier. The stuff that is harder to sonicate will remain behind in the insoluble pellet (after spinning the sonicated samples at high speed) thus reducing any potential signal from those regions and enriching any potential signal in the regions that had less crosslinking. Of course there is additional bias in the size selection step. When you select a narrow range of shorter fragments (~200-250) those are the fragments that sonicated very well which probably came from the chromatin with fewer proteins and crosslinking. The larger fragments are likely from the chromatin that was more resistant to sonication (heterochromatin likely). A little bias here, a little bias there, now the true signal might be a little skewed depending on where your protein of interest resides. Maybe. That's my thinking on it.

[NGene]

The sequencing of the input is a good idea in order to control for the patterns that arise due to the uneven distribution of crosslinks in the nucleus due to regions of heterochromatin (more proteinaceous and thus more crosslinked) and euchromatin (less proteinaceous and thus fewer crosslinks). The regions with fewer crosslinks will sonicate/digest and decrosslink easier. The stuff that is harder to sonicate will remain behind in the insoluble pellet (after spinning the sonicated samples at high speed) thus reducing any potential signal from those regions and enriching any potential signal in the regions that had less crosslinking. Of course there is additional bias in the size selection step. When you select a narrow range of shorter fragments (~200-250) those are the fragments that sonicated very well which probably came from the chromatin with fewer proteins and crosslinking. The larger fragments are likely from the chromatin that was more resistant to sonication (heterochromatin likely). A little bias here, a little bias there, now the true signal might be a little skewed depending on where your protein of interest resides. Maybe. That's my thinking on it.

I am facing this issue too. I guess I am enriching parts of the genome which are less crosslinked, losing likely the heterochromatin (which is what I am interested in...).
I believe that my statement is true as I clearly see differences in reads abundance also for H3 and H2A histones (no modifications, just nude histone) IPs I used to normalize.

Do you have any suggestion to overcome this issue?

Or even better, would one be able to use such disadvantage to his own advantage?
In other words, could I assume that wherever I see a clear "valley" in H3/H2A-IP reads, I am "seeing heterochromatin" and thus I could analyze these areas differently (e.g. different threshold for my sample data sets) from the ones that are more enriched?

[mudshark]

Do you have any suggestion to overcome this issue?

Or even better, would one be able to use such disadvantage to his own advantage?
In other words, could I assume that wherever I see a clear "valley" in H3/H2A-IP reads, I am "seeing heterochromatin" and thus I could analyze these areas differently (e.g. different threshold for my sample data sets) from the ones that are more enriched?

You can overcome this problem by normalizing to the released chromatin (input or mock ip).

as regards your valleys in H3/H2A: watch out, as there are regions in chromatin that are rather depleted of histones, nucleosome free/depleted regions, often to be found in promoters, preferentially active ones.

the differential release of chromatin has actually be turned into an assay:

http://www.ncbi.nlm.nih.gov/pubmed/19303047

[Chipper]

Here is another creative way of publishing non-random signals in the negative control...

http://www.ncbi.nlm.nih.gov/pubmed?term=sono-seq

[NGene]

You can overcome this problem by normalizing to the released chromatin (input or mock ip).

as regards your valleys in H3/H2A: watch out, as there are regions in chromatin that are rather depleted of histones, nucleosome free/depleted regions, often to be found in promoters, preferentially active ones.

the differential release of chromatin has actually be turned into an assay:
http://www.ncbi.nlm.nih.gov/pubmed/19303047

I am aware of regions that are depleted of histone/nucleosomes (especially active promoters, as you say). The point is that I am talking about very big domains (more than 5 Mbp) where the abundance is often 3-4 lower than a mean value I calculated.

Although I expected some "valleys", I did not expect them to be so large.

Thanks a lot for the reply, by the way.

PS: the two publications are brilliant!

[frozenlyse]

I am aware of regions that are depleted of histone/nucleosomes (especially active promoters, as you say). The point is that I am talking about very big domains (more than 5 Mbp) where the abundance is often 3-4 lower than a mean value I calculated.

Although I expected some "valleys", I did not expect them to be so large.

Thanks a lot for the reply, by the way.

PS: the two publications are brilliant!

I'm not sure what system you're using, but could you be seeing large genomic deletions? We see can see it contribute to both IP and Input signals in enrichment experiments eg http://www.ncbi.nlm.nih.gov/pubmed/21045081

[NGene]

I'm not sure what system you're using, but could you be seeing large genomic deletions? We see can see it contribute to both IP and Input signals in enrichment experiments eg http://www.ncbi.nlm.nih.gov/pubmed/21045081

I would exclude large deletions since the genome of the cell line I am using is sequenced and annotated. Besides, I assume I would get no reads if a deletion occurs (in a homogeneous population); I get lower number of reads, instead.

Anyways, I think that captainentropy pointed it out: I must be selecting (or at least, giving preference) to certain areas during the library prep.

I will keep investigating on this. Thanks for the replies!

[ETHANol]

I also use an uncommon method of chromatin preparation. After lysing my crosslinked cells I spin the lysate through a cushion of 8M urea at 45-55K RPM for ~7-12 hours. The advantage of this is that the pellet that emerges at the bottom of the tube is 100% crosslinked material.

I'd be really interested in knowing the details of your chromatin prep protocol.

As far as the issue of uneven chromatin fragmentation, my chromatin preps come out more when I fragment with micrococcal nuclease and the fragmentation seems to be more efficient, i.e. there is essentially no chromatin above 500 bp and it's simple to digest it all down to the mononucleosome size. There is evidence that it is easier on some epitopes as well. I can share what I do but I think it's too long to post so send me a PM with your e-mail if you like.

[NGene]

I would be interested in knowing captainentropy prep protocol too.

Concerning your protocol, ETHANol, I would be glad to read it too. I am sending a pm with my e-mail, if you do not mind.