ChIP-seq Library Prep/Multiplexing?

[jazz710]

Hello everyone,

This is my very first post to the site, please be gentle! I have read a lot of good ChIP-seq papers, and I have scoured this board as hard as I could, but I think my questions may just be too specific to build an answer piecemeal. I really don't mean to impose on anyone's time, but I don't have anyone in my department I can really turn to for help on this...

My Question:
Where does the histone variant H2A.Z bind in the genome of my study organism and does H2A.Z binding change between treatments/developmental stages?

My System:
One vertebrate species (Genome size ~3.0Gb), two experimental treatments, 5 timepoints per treatment. (10 total samples) These are embryos so Timepoint 1 samples are significantly smaller than Timepoint 5.

My Samples:
For each 'sample' (10 as defined above) I have 8-10 embryos to pool [ie. Sample 1: Treatment 1, Timepoint 1 (8-10 embryos); Sample 2: Treatment 1, Timepoint 2 (8-10 embryos)...Sample 10: Treatment 2, Timepoint 5 (8-10 embryos)]

This is my first ChIP-seq experiment. I am in a lab that hasn't done ChIP before, let alone ChIP-seq. My experience so far with qChIP has been positive. I'm confident that my antibody is working, and that there's nothing too critically wrong there. But who knows...

I do ChIP and extract chromatin out of each embryo individually so that my qChIP data point can be an individual. I use the Input DNA control method, where I will take pre-IP DNA and post-IP DNA, and run them both in the qPCR for each gene to calculate Percent Input. For qChIP, I'm happy with that.

My questions come when trying to think about how to set this up for ChIP-seq. I just can't seem to crystallize it in my head.

I would like my ChIP-seq data to tell me a few major things:

1) Where does H2A.Z bind in my study organism?
2) Does H2A.Z binding at the same timepoint change between treatments?
3) Does H2A.Z binding change in the same treatment between developmental periods?
4) Does H2A.Z binding change differently between developmental periods depending on treatment? (Interaction term in ANOVA lingo)

Here are my concerns/questions. This is a lot, I know. I hate to sound needy or like I can't think for myself, but I'm just looking for advice from people that have done something similar:

1) My facility wants 30ul of 1ng/ul DNA (they'll settle for 10ng).
2) To start, there is much less IP product AND Input DNA in the Timepoint 1 samples than in the Timepoint 5 sample (Timepoint 5 > Timepoint 1)
3) I want to quantitatively compare H2A.Z occupancy at the same timepoint across treatments, and in the same treatment across timepoints.
4) If my 10 different ChIP/Input DNA preps start with different amounts of DNA, but then I dilute/concentrate them all to the same 1ng/ul, will that kill my quantitative analysis?
5) Input samples shouldn't change over timepoints (chromatin is chromatin), right? Can I pool my Input DNA together for my different timepoints/treatments (non-mutagenic) and just sequence 1 Input sample to get a baseline?
6) Is the quantification achieved by loading equal concentrations of each sample into each lane, and then looking at differences in read depth relative to the input? (ie. Each library was prepared with equal amounts of DNA, and determining differential binding relative to another sample is done by comparing fold change relative to Input?)

I'm also trying to figure out how many samples I can index and run in one lane. We can regularly generate ~70M reads/lane, and I think that 20M reads is suitable for proper coverage. My thought is then 3 samples/lane x 3, then 1 sample and the input in a 4th lane.

With money, I don't even want to think about biological replicates right now

Thank you so much for your time, and I hope to learn a lot from this discussion.

Bob

[mudshark]

some immediate thoughts:

a) try to keep the experiment as simple as possible. why 5 timepoints? (1) costs (2) statistical analysis incorporating all the data is fairly complicated. why not do a pioneer experiment an select two interesting time points first?

b) do NOT pool the inputs across time points as the input reads vary depending on the activity state of the chromatin. i.e. if expression changes your input reads change as well. you will need the corresponding inputs to normalize the IP properly. (Teytelman L, O ̈ zaydın B, Zill O, Lefranc ̧ois P, Snyder M, et al. (2009) Impact of Chromatin Structures on DNA Processing for Genomic Analyses. PLoS ONE 4(8): e6700. doi:10.1371/journal.pone.0006700)

c) if performing treatments you HAVE to replicate to get to a robust conclusion: better reduce experiment to 2 time points and replicate twice.

d) different amounts of DNA is not ideal. but if you cannot improve this and just qualitatively compare (peak, no peak) you still can derive useful info. for a true quantitative comparison, these differences might get problematic. my opinion.

[jazz710]

Thank you

Thank you so much for the quick reply!

a) try to keep the experiment as simple as possible. why 5 timepoints? (1) costs (2) statistical analysis incorporating all the data is fairly complicated. why not do a pioneer experiment an select two interesting time points first?

This is a good suggestion, and I think that for my first run, this would be ideal

b) do NOT pool the inputs across time points as the input reads vary depending on the activity state of the chromatin. i.e. if expression changes your input reads change as well. you will need the corresponding inputs to normalize the IP properly. (Teytelman L, O ̈ zaydın B, Zill O, Lefranc ̧ois P, Snyder M, et al. (2009) Impact of Chromatin Structures on DNA Processing for Genomic Analyses. PLoS ONE 4(8): e6700. doi:10.1371/journal.pone.0006700)

This citation will be very helpful, and it is not one that I had. Adding an input per StagexTreatment will add to my costs but it's not worth anything if I do it wrong.

c) if performing treatments you HAVE to replicate to get to a robust conclusion: better reduce experiment to 2 time points and replicate twice.

So here we're talking about biological replicates right? ie) if I have 8-10 total embryos per StagexTreatment, I should split them into sets of 4 or 5, then pool each set of 4, and do the ChIP-seq twice with the different sets?

If I pool samples together, I should pool their respective inputs correct (within a StagexTreatment)? So let's say Embryos 1-4 are in biological replicate 1 and Embryos 5-8 are Biological Replicate 2. I do the IP and get the Input for Embryos 1-4 (and 5-8). Then I pool the IPs, and pool the Inputs WITHIN each replicate (not across) and sequence each replicate as its own library?

This would mean that for 2 StagexTreatment pairs I have to sequence 16 libraries?! Do I need biological replicates for my Input too?

Stage1 - Treatment1 (2 IP, 2 Input)
Stage1 -Treatment2 (2 IP, 2 Input)
Stage5 - Treatment1 (2 IP, 2 Input)
Stage5 - Treatment2 (2 IP, 2 Input)

d) different amounts of DNA is not ideal. but if you cannot improve this and just qualitatively compare (peak, no peak) you still can derive useful info. for a true quantitative comparison, these differences might get problematic. my opinion.

This was my gut feeling, whether or not I wanted it to be true.

[mudshark]

So here we're talking about biological replicates right? ie) if I have 8-10 total embryos per StagexTreatment, I should split them into sets of 4 or 5, then pool each set of 4, and do the ChIP-seq twice with the different sets?

you should try to fully exploit the power of biological replication by full independence of experiments. do the replicates on different days with different batches of embryos and - ideally - different stock of treatment chemicals (in case it is a chemical.. )

If I pool samples together, I should pool their respective inputs correct (within a StagexTreatment)? So let's say Embryos 1-4 are in biological replicate 1 and Embryos 5-8 are Biological Replicate 2. I do the IP and get the Input for Embryos 1-4 (and 5-8). Then I pool the IPs, and pool the Inputs WITHIN each replicate (not across) and sequence each replicate as its own library?

This would mean that for 2 StagexTreatment pairs I have to sequence 16 libraries?! Do I need biological replicates for my Input too?

Stage1 - Treatment1 (2 IP, 2 Input)
Stage1 -Treatment2 (2 IP, 2 Input)
Stage5 - Treatment1 (2 IP, 2 Input)
Stage5 - Treatment2 (2 IP, 2 Input)

yes, given the fluctuations of input reads upon changes in gene expression you have to sequence the inputs corresponding to IP replicates, too. That is a lot of money, but I don't see any other way to come to reproducible and biologically valid results.

let me add one remark: if you KNOW that your input reads are not affected by treatment you might consider using a pooled treatment/no-treatment input to reduce costs. but how to be sure?

[jazz710]

Thank you again

you should try to fully exploit the power of biological replication by full independence of experiments. do the replicates on different days with different batches of embryos and - ideally - different stock of treatment chemicals (in case it is a chemical.. )

yes, given the fluctuations of input reads upon changes in gene expression you have to sequence the inputs corresponding to IP replicates, too. That is a lot of money, but I don't see any other way to come to reproducible and biologically valid results.

let me add one remark: if you KNOW that your input reads are not affected by treatment you might consider using a pooled treatment/no-treatment input to reduce costs. but how to be sure?

I don't know anything [that's why I'm doing the test ] so your advice will go heeded. It turns out that I was underestimating our read capabilities by a lot, so even though I still need to make lots of libraries, I can use less lanes.

I really appreciate your help and know that your responses gave me relief in a stressful time.

Bob