2
$\begingroup$

I have ChIP-seq for H3K79me2 and H3K36me3 and RNA-seq data for treated and untreated samples. Those two histones mark active genes. Lets say, hypothetically, a peak caller finds differential sites at gene A for both of those histone modifications. However, when I run RNA-seq analysis tools (like edgeR or DESeq2) this gene is not marked as differentially expressed (well it has a FDR value of >0.05).

There might be several technical reasons for RNA-seq analysis methods not to find this gene as differentially expressed because

  1. it is really not differentially expressed
  2. or it is not differentially expressed enough for the tools to detect it

However, I am more interested in the biological aspect. If two histones mark gene A as active in treated samples, one would expect that it would be expressed in RNA-seq. What mechanism could lead to the result that however the genes is marked as active by histones, it will not be differentially expressed in RNA-seq?

$\endgroup$
  • 1
    $\begingroup$ Your statement that those two post-translational modifications of histones mark active genes is not supported by your experimental evidence (in every case). Perhaps a more accurate statement would be something like "often mark active genes" or "are typically associated with active genes?" $\endgroup$ – mdperry Jan 19 '16 at 3:20
3
$\begingroup$

It is difficult to draw any conclusion without further experimentation. There may be many other factors that prevent the expression of the gene including factors like post-transcriptional regulators. Some histone modifications like the ones you are mentioning are also a bit dicey and there can be bivalent modifications too. However, if there is a strong reason to speculate that these marks are associated with repression then you can try these:

  1. You can verify how many genes are there whose expression is not in accordance with the histone marks. If this for a small number of genes then you can investigate them separately. Use a statistical test to verify the significance of your correlation between histone mark and expression. If the discordance is widespread, then I would suggest you to repeat the RNAseq (easier and cheaper than repeating ChIP-seq). BTW, how were the read qualities? Depending on that you can decide which experiment needs repetition. Repetition will help you to know if the observation is indeed correct (a biological replicate); possibly you may end also up concluding that these marks are actually dicey.

  2. If a few genes show discordance, then just do Real-Time PCR for these genes. RT-PCR is more sensitive and another technique of quantification is anyways good. Use a few concordant genes also, in the experiment.

$\endgroup$
  • $\begingroup$ Which statistical test are you talking about? I have about 40 genes that are not supported by RNAseq, also I have genes that are supported by RNAseq but not by those histones. Sp, I have king of venn diagram of number of genes RNAseq vs active histone modification. $\endgroup$ – Alina Jan 18 '16 at 18:50
  • $\begingroup$ @Tonja 40 genes discordant: how many agree with the RNAseq? I am not sure about which statistical test to use here. Basically you have to see what is the probability of RNAseq not agreeing with Histone modification. $\endgroup$ – WYSIWYG Jan 19 '16 at 5:01
1
$\begingroup$

ChIP-seq isn't perfect. Even between technical replicates, you get a fair amount of variation, especially for broad marks like those you're using. It's rather uncommon to see people use H3K79me2 and H3K36me3 to determine if a gene's expressed or not. Using H3K4me3 and H3K27ac or H3ac is a more common method of marking promoters of transcribed genes.

40 genes discordant wouldn't worry me much unless you really have low reads for either set of sequencing experiments. @WYSIWYG's suggestion to use RT-PCR to verify your RNA-seq data for a few genes is a good one.

$\endgroup$
  • $\begingroup$ I checked the promoter histones as well and I have also some genes that are not supported by RNA-seq. I know I could validate the results but that is encode data. So, I would like just to know the theoretical reason for that. $\endgroup$ – Alina Jan 19 '16 at 15:57
  • $\begingroup$ What is encode data? The RNA-seq? $\endgroup$ – Jared Andrews Jan 19 '16 at 16:30
  • $\begingroup$ @JaredAndrews I think Tonja is referring to the TFBS ChIP-seqs.. $\endgroup$ – WYSIWYG Jan 19 '16 at 16:59
  • 1
    $\begingroup$ @WYSIWYG If that's the case, I'd suggest looking at different marks, as they certainly have them if they also have H3K79me2 and H3K36me3. The epigenome is complex, flexible, and variant. As such, you should look at the pieces as a whole whenever possible. $\endgroup$ – Jared Andrews Jan 19 '16 at 18:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.