sequencing exons in chip-seq

[mattanswers]

We have been having a really difficult time with Chip-Seq experiments in which we get 70% or more of our reads located in exons.

Also, our 'control' sample of DNA isolated without antibody (which should be all of the DNA) looks very similar (in terms of aligned reads) to our sample in which only DNA attached to transcription factor pulled down with Ab should be present. That is, control and treatment look very much alike when viewing aligned reads.

However, the above is not what we get when we do PCR on the samples before they are sequenced. In this case, when doing PCR around a gene expected to have TF binding we get the expected results of DNA in specific 5' regions and not in exons when using Ab, and DNA in both exons and 5' regions when no Ab.

Has anyone had a similar problem or have any insights ? Has anyone sequenced exons when doing Chip-Seq ?

[Chipper]

Sounds like there is a sample mixup or contamination from exon-capture experiments. ChIPs will only give a few nanograms of DNA so I would ask the sequence provider if they have run any exon-capture recently... You could also try PCR verification on the amplified libraries. Is there any sign of enricment around your known targets if you disregard exons?

[mattanswers]

Thanks for your reply Chipper.

We have seen this now in two different experiments. In the first experiment the Ab was against a protein involved in a complex that binds to DNA, but the particular protein itself does not bind DNA.
So, we tried a second experiment with Ab against a transcription factor and at the same time we used a different sequencing facility. The same thing happened in that control and sample look very similar in terms of viewing the aligned reads, and reads were mostly in exons. It is very striking in that the reads will not be in the introns. However, there is a greater difference between the two experiments than between the control and sample of each experiment. Typically we are getting 20 million reads per sample lane.

In the second experiment, reads were very abundant in the three exons of a target gene. They were also in the 5' region of the gene as well. However, the pattern of + strand reads and - strand reads in the 5' region from the sample 'suggests' possible binding sites. The control also has reads in the 5' region, but the +/- pattern of reads is not so much the staggard pattern you would expect for binding sites. It seems like the data is there, but that there is noise or background getting in the way.

By way of comparison, in the first experiment, there were only a few reads in the exons of this gene that was a target gene in the 2nd exp, and no reads in the 5' region.

[Chipper]

Interresting, perhaps the complex is involved in splicing? Is it a normal ChIP protocol with sonication? 70% exoninc reads sounds very high though since most ChIP-seq reads even for highly enriched TFs are background.

[mattanswers]

Thanks again for your reply,

But this has happened for two different proteins; one is in a complex and does not bind DNA directly, but the other protein does bind DNA directly.

Our protocol is fairly normal. We do not use a sucrose gradient, but a sucrose 'cushion' and spin for a pellet underneath the sucrose.

We seem to get a lot of reads, about 20 million, compared to some public data I have downloaded from GEO in which they where doing Chip-Seq for a different TF but in a very similar system. They were getting 2-10 million reads per sample.

[simonvh]

20 millions reads is a normal number for a lane on a current Genome Analyzer.

Are you sure your samples are not contaminated? Do the genes look familiar to you (ie. any specific constructs floating around in your lab)? If yes, then be very suspicious. We used to have some problems like that before switching to a more controlled environment for sample prep (room, filter tips, etc.). Although 70% is quite a lot...

[adamdeluca]

What reference are you using for genomic alignment? If you were aligning to an exon junction reference or something similar, that could really throw off the numbers.

[mattanswers]

I use the TAIR9 Arabidopsis genome in the original alignment.
TAIR also makes files of just exons, 5' regions, 3' regions, etc., so I use BEDTools along with the exons file and my sequences file to find the overlap. I also know what percentage of my sequences match to 500bp upstream 5' regions as well as other areas of the chromosome, but those numbers are in the lab.

[Chema76]

Hi,

In our experience, when reads for a ChIP-seq experiment map to exons is a clear signal that the ChIP didn,t work. In these cases, you are sequecing open chromatin. Genes being transcribed are in open chromatin

Jose Muino

[simonvh]

Is it really, exclusively exons only, or is there signal in the complete gene body including introns? How could exon-specific signals be explained by open chromatin? You would then sequence the introns too, right?

[mattanswers]

The reads are more in the exons than introns. I will try to post a picture.

[mattanswers]

example

This is representative of the whole genome.