Should the amount of DNA input used for ChIP-seq library preparation be matched between the control and experimental groups?

Question

[Ian]

As a beginner in ChIP-seq experiments, I hope you understand that the following questions might be somewhat basic.

I am planning to perform ChIP-seq or MeDIP-seq analysis to investigate changes in global histone/DNA modifications in the experimental group compared to the control group. For example, if a decrease at the global level is observed in the experimental group, enriching the targets for analysis through immunoprecipitation might result in a relatively lower final DNA yield compared to the control group.

In such cases, I am conflicted about whether I should match the amount of DNA used to create the NGS library between the control and experimental groups. Conceptually, I am struggling with the idea that artificially matching the amount of DNA of the experimental group, which has a lower yield due to immunoprecipitation, to the control group might eliminate the differences that I need to observe between the control and experimental groups.

If my thinking is incorrect, I would like to understand why. Alternatively, if my thinking is correct, I would like to know how to proportionally set the amount of DNA used to create the library between the control and experimental groups.

Answer 1

[ATpoint]

You absolutely should not derive anything from the amount of ChIPed DNA. The yield can (in my experience) be largely different even on a replicate of the same condition. ChIP is extremely noisy.

Also, ChIP-seq and sequencing in general is relative, not absolute so it does not really suite for the quantification of absolute changes. If you just want to know whether the cell has generally more of a certain protein then maybe a simple Western will do. ChIP-seq is good for determining protein binding location and relative differences, e.g. differential peaks, but not general occupancy.

[Ian]

Thanks, ATpoint.

I realized that the focus is not on comparing absolute quantities. I now understand what kind of results we should expect from ChIP-seq. Thank you very much for taking the time to respond.

Would it be okay if I asked you the same question that I posted as a response to rfran010? I am very curious about your intellectual insights as well, and I would greatly appreciate your answer.

Answer 2

[rfran010]

You will not be able to detect the global change you describe because the libraries will need to be normalized for sequencing depth. As ATpoint points out, ChIP is also noisy, the background signal may change rep to rep and not be indicative of changes in target protein abundance.

So, for example you may pull down half as much DNA in your treatment sample. This may be due to lower target abundance, but could also be due to chromatin shearing differences or efficiency of any intermediate step (IP, reverse x-link, DNA purification...). There are protocols that suggest exogenous spike-ins can help normalize for global changes, but these come with issues of their own that need to be considered in context of your specific experimental design.

Furthermore, even if half as much DNA is indicative of lower target abundance and you use this as input and prep libraries and sequence them while keeping this relative input amount, there's still variation introduced with the number of sequencing reads produced. First, if you sequence half as much DNA you might expect half as many reads, then you map these and present them, showing on average, half the signal across the genome. I don't think any reviewers will accept this as valid. Furthermore, the read yield will also likely vary due to various factors, including clustering efficiency, pipetting error, etc. So usually you will need to normalize to the total number of reads, removing the whole point of starting with half as much DNA in the first place.

Additionally, during the library prep, if input amounts are vastly different that can introduce noticeable biases in the libraries, especially during the PCR step.

So I think if you have the ability, you should equalize your input DNA amounts, then hopefully differences in library composition are indicative of the IP'd DNA and not of your library prep method.

I hope that's relatively clear and helpful.

[Ian]

Thank you very much for providing such a detailed and kind response. It seems that it will be very helpful for planning future experiments.

Before I fully embrace the idea you mentioned, I have one last question.

For instance, let’s say I have confirmed global H3K27me3 decreases in treatment sample via Western blot, and I’m looking forward to conduct sequencing to identify the specific genomic regions with the indicated changes. I understood that constructing library by equalizing the IP-DNA input between the treatment and control groups as you suggested, we can still expect the differences in target abundance from the sequencing results.

However, I am imagining a situation where the background signal or unaffected target loci might increase in the treatment samples. In this case, would there be any analytical side effects where these regions are mistakenly considered as increased regions?

If this is correct, is there a well-known method or program to handle this in computational analysis phase?