Anyone using ChIP-exo?

[jwfoley]

From the data analysis side, ChIP-exo (PMID: 22153082) looks much better than standard ChIP-seq - not because of the fine resolution (who really cares?), but just because of the much improved signal-to-noise. However, the only papers I've seen with it are by the same authors, and there are very few threads here about it, and I've heard rumors that it's not really working for anyone but the people who invented it.

Any anecdotes? Has anyone gotten it to work with the published protocol (PMID: 23026909) or some variant?

[FiReaNG3L]

It's working for me - it's not an easy technique as it is labor intensive and requires a lot of care, but at least for our application the results are well worth it. I'm using a slightly modified version adapted for Illumina.

[jwfoley]

Thanks. Is the Illumina version basically the same thing but with the universal Illumina adapter? Which steps are more difficult than standard ChIP-seq?

[FiReaNG3L]

My version is a custom one I implemented before their Methods paper got published - I never tested their solution.

It's not more difficult, it's just more labor intensive. It's basically a Chip + a library prep on the same day. If you follow their protocol to the letter, its 30 washes of 5 min, plus the enzymatic steps. The proper analysis also requires more work if you want to use the extra information the technique gives you.

Depending on your final goal, it could not be worth the extra work / costs or (as in my case) it could be essential. From your first post I would say if you don't really care about the fine resolution, you're better off with Chip-seq.

[cptzantar]

I've been trying to get chip-exo to work for about 2 months with little success. Would you be willing to share your protocol for chip-exo to see how it differs from what I am using?

Thanks.

[james hadfield]

We've one group making use of ChIP-exo the peaks look beautiful, are base-pair accurate for binding analysis and probably mean we only need 25% of the read depth. However the protocol is definitley not standardised, green-fingers seem to help and until it is really robust we'll keep running normal ChIP-seq for most work.

[Mammooth]

Hi!

I'm also working with the published protocol, for about... err.. half a year or so. I can't say with much success. I already changed a lot of things to make it more sound, but it seems there is still much work ahead.
Would any of you be so kind to provide a working protocol?
Of course any little help are welcome.

[jwfoley]

To those who report the protocol's failing: how? At what step do you get unexpected results?

[james hadfield]

Forgot to direct you to this: Development of an Illumina-based ChIP-exonuclease method provides insight into FoxA1-DNA binding properties.

http://genomebiology.com/2013/14/12/R147/abstract

[albireo]

I have been analysing illumina-based human Chip-exo data for a while - in my experience, the results are very TF-dependant.

At best, results (in terms of peak resolution and SN ratio) are almost within quoted specs.

At worst, results are comparable to ChIP-seq, with the added complication that a proper analysis will be lengthier (most off-the-shelf ChIP-seq analysis techniques will not work properly with ChIP-exo).

[jwfoley]

Thanks, albireo. Just to be clear, are you saying that in the worse cases, you have a ChIP-seq-like SN ratio, i.e. there is high background throughout the genome?

FYI I tested UniPeak (doi:10.1186/1471-2164-14-720) on the ChIP-exo data from the original paper and it seems to perform nicely with one or two non-default settings. Let me know if you'd like more information.

[albireo]

Thanks, albireo. Just to be clear, are you saying that in the worse cases, you have a ChIP-seq-like SN ratio, i.e. there is high background throughout the genome?

Hi, yes that is what I meant. I have performed a number of FRiP analyses (similarly to what done in the 2012 ENCODE Genome Research ChIP-seq methods paper) with very disappointing results in one case (~4-5% FRiP). This data is decent in other respects (high correlation of replicates, consensus motif present in 70% of the peaks).

FYI I tested UniPeak (doi:10.1186/1471-2164-14-720) on the ChIP-exo data from the original paper and it seems to perform nicely with one or two non-default settings. Let me know if you'd like more information.

I have not tested this specific peak caller, but have some experience with other methods, including Genetrack (the original peak caller used in the Pugh publication), Macs 2.0, Apex, Peakzilla and GEM, with wildly varying results. I should also add that my data is Illumina-based and was processed directly by Peconic.

[jwfoley]

I have performed a number of FRiP analyses (similarly to what done in the 2012 ENCODE Genome Research ChIP-seq methods paper) with very disappointing results in one case (~4-5% FRiP).

Interesting. The way ChIP-exo digests unprotected DNA, this sounds like it would come from a nonspecific antibody. My favorite negative control for ChIP-seq is a knockdown/knockout of the antibody target, though that's often biologically infeasible. I don't suppose you have that for your experiment?

[albireo]

The antibody should be sound.

We've been wondering if it is more a case of the exonuclease failing to stop at the crosslinking site, while stopping at various points before. This could be due to misfolding and be stricly dependant on the TF being analysed.

This exonuclease-stop hypothesis seems to be supported by the high variability in peak width I'm observing (which I verified via cross-correlation approaches to look at the start-stop distance - these fail to show anything apart from the phantom peak, suggesting there is no consensus start-stop distance). The fact that our TF binds in several configurations (monomeric/dimeric) does not help.

[TonyBrooks]

Forgot to direct you to this: Development of an Illumina-based ChIP-exonuclease method provides insight into FoxA1-DNA binding properties.

http://genomebiology.com/2013/14/12/R147/abstract

Thanks for that James. We're about to look into ChIP-Exo and designed our own primers, but we may just use these instead. Am I right in thinking that there is a T missing from their P7-exo adapters though? They are doing blunt-end ligation, but the Illumina R1 and R2 sequencing primers end in T due to the dA-tail ligation generally employed.