ChIP-seq in cancer cell lines

[captainentropy]

Hi, we're doing ChIP-seq in several cancer cell lines. Of course, common features of cancers can be translocations and polyploidy. We know that one cell line we use, K562, has I think 68 chromosomes and there are numerous translocations.

Discussing some recent ChIP-seq data a colleague in the lab was asking me about the reference genome and whether it factored in the translocations. I said "of course not" but it brought up a point that I (embarassingly) hadn't thought of and I'm wondering what the conventional wisdom is regarding ChIP-seq in cancer cells/cell lines?

The way I see it is the reference genome will have all the sequences but not necessarily in the same place for the cancer line in question. If we find some major peaks upstream of some gene in our ChIP, we don't know a priori whether the downstream (or upstream) sequences are present in our cell line as they are in the reference genome - they could be very far away.

For that matter if the sequences are on a rearrangement boundary any reads that span it will not map because they are composed of sequences from different regions. In fact, in this one ChIP, there were an unusually large number of reads that didn't map. Blast searches of many of these unmapped reads resultsed in multiple partial matches in the genome. This is of course what we see when doing deep-sequencing of 3C libraries which by design are chimeric sequences that are generated.

It seems to me that in this case it is imperitive to generate a reference genome for that cell line, otherwise the peak coordinates might be dramatically off. What's the accepted minimum coverage for sequencing a genome?

In the case of K562 the translocations haven't been mapped with high enough resolution to be confident in many regions. Of course peaks located in a region with no known rearrangements are probably fine.

Any thoughts on this?

[krobison]

30X is a general guide for coverage of a human genome. Higher is probably better in clinical samples, given heterogeneity. But, since for your purpose you aren't so worried about calling every last SNP but getting a general reference, perhaps you could go lower.

You could generate a much lower coverage mate pair library to identify the major rearrangements -- that would require much lower coverage & since you have the cell line generating the large sample requirements for this would not be difficult. You'd need to decide what size mate pair library to make -- i.e. are you worried about rearrangements on the scale of 5Kb or 20Kb?

Have you tried doing de novo assembly of the reads that didn't map? if they really were piling onto something, you might get some small contigs whose existence you could test by PCR.

If you take the two RNA-Seq papers on K562, do the fusion transcripts found help you interpret your data? Do you see lots of your fragmentary matches in the regions which would be implied by these fusions?

[captainentropy]

You're right, I'm not so worried about the SNPs. Because of cost/resources I would like to go as little coverage as possible. But basically one GAIIx flow cell (2x 100bp reads) would be ~18X coverage, yeah?

We haven't tried assembling anything yet. This is the first set of data we got back. The read numbers aren't good enough to do much in the way of assembly.

We'll have to be serious about sequencing the whole genome before we proceed with ChIP-seq. That's a chunk of change so we will proceed with caution.