Reduce background signal in ChIPseq - (Jun/22/2011 )

Hi,

I just received results from my first transcription factor ChIPseq experiment. Although I get some nice peaks in the candidate regions that I used for optimization by ChIP-qPCR, I am worried about a lot of input-like background signal that I am seeing. Basically, almost 75-80% of the 10bp windows in a wiggle-file representing my data will have at least 1 read.
I certainly admit that our experiment may be a little difficult to do, as we do ChIP on a tissue, where only a fraction of cells express the TF of interest. However, we had a good idea about candidate target genes and I was able to consistently show good enrichment (15-100fold) for 7 of those candidates, while control regions were negative. Background in those experiments was comparable to previous ChIP experiments using other tissues in less limiting circumstances.
I can't think of much more what I could do to reduce background in my ChIP-assay: I make sure my sonication is good and does not give much signal above 600bp (typically less than 15%). I use siliconized Eppendorf tubes for the entire procedure to avoid DNA to stick to the tube walls. I looked at different amounts of chromatin and antibody per IP and stuck with the relation that gave the best enrichments. I optimized the salt concentration in my IP buffer to the highest concentration that would still give good enrichment. I know that the antibody is good - in fact it has already worked well for ChIPchip and many other applications. I use Dynabeads that have been blocked in BSA and tRNA over night. I do a multitude of washes (16 in total), including several high salt washes with both NaCl and LiCl. I do ProK and RNAse digests after elution and prep DNA by phenol/chloroform.
For the ChIPseq experiment, I pooled DNA from 12 individual IPs. I had only a nanodrop available to estimate DNA concentration after IP, which indicated I had ~800ng IP'ed DNA with a good 260/280 (1.8). I don't thinks this figure was correct and a tech form the core facility that processed the library for us confirmed that nanodrop would regularly give false high readouts for ChIPped DNA. qPCR showed about 100fold enrichment for a good candidate region. I could not confirm this on the library though, as it was processed in the core using a SPRIworks system.

Any suggestions how to proceed? Repeat ChIP and run DNA on a Bioanalyzer to confirm amount? Increase stringency of IP buffer? Add even more washes? Re-optimize amount of chromatin per IP?
I would appreciate any suggestions

Thanks

Makabe

Well first of all, it is quite difficult to understand from your post exactly what the problem is. The "input like background" is not necessarily a problem and actually that is the whole point of the input control. However, the critical point is that you have enrichment over this background in your TF sample.

It is probably best to work backwards to see what is going on, so lets deal with the following questions first:

1) How many reads did you generate for the input and TF ChIP-Seq (and what length)?
2) What platform did you use for the sequencing?
3) How many of your reads mapped to the reference genome, for the input and TF?
4) How many ChIP peaks did you detect in the TF sample?
5) How was your library generated? Did you match the amount of DNA going in to both the input and TF library prep?
6) On this point, there is no way you had 800 ng IP DNA......NO WAY!
7) You say you had good enrichment in your ChIP, but compared to what? You don't mention an IgG (or similar) control experiment.
8) You say you used negative control regions in your ChIP, and compared them to positive control loci. Is this correct?
9) What was the AVERAGE size of your IP DNA and what was the size of the library after size selection? You can run into major problems here if there is a mismatch between the two. For example, if your IP'ed DNA is 400 bp on average and you cut out 250 bp during size selection, you lose a lot of your enrichment. You NEED a bioanalyzer here; find one, borrow one, steal one, whatever. ChIP-Seq library prep without one is like working blind and that is not good.

My guess at this point would be that your library prep is the issue (it usually is), but we definitely need more info.