Tiling microarrays represent an extension of expression arrays. They are designed to interrogate both coding and non-coding sequence across an entire genome. Their design is straight forward, simply place a probe every Xbp across each block of non-repetitive sequence on each chromosome. Tiling arrays are being used for a variety of purposes including the mapping of binding sites for DNA binding proteins (transcription factors, polymerases, modified histones, etc.), mapping of DNA modifications (methylation), mapping of coding and non-coding RNA transcripts (transcriptomics), and identifying sequence gain and loss (aCGH). Although static maps have their uses, what is turning out to be more useful is identifying regions that change under different experimental conditions.
At present, our core uses oligonucleotide based tiling arrays purchased from three companies: Affymetrix, Agilent, and NimbleGen. Each has it's advantages and disadvantages.
Currently we support analysis using the T2 package.
- I think it important to have three replicas for every experiment. These don't necessarily have to be biological replicas, e.g. three different isolates of a fibroblast cell line processed in parallel. Technical replicas in most cases are preferred, e.g. pool multiple plates of fibroblasts and split it after fixing/lysing/preparing chromatin and process in parallel. Technical replicas are extremely informative. They provide a measure of the sample prep/ microarray processing consistency without any confounding biological variability.
- If you choose to only go with two replicas be prepared for lots of qPCR validation. The symmetric null method of confidence estimation works reasonably well for non complex organisms but rarely for human or mouse. Three IPs and three input replicas are the minimum needed for a random label permutation confidence estimation.
- It is rather critical to perform all of these preps using the same reagents, same equipment, at the same time, etc. Be prepared to throw out all of the data generated during the pilot/ optimization phase of the experiment and repeat it along side the other samples.
- For two color arrays, there is no need to have a matching input control on every chip. You can treat the different colors independently and think of them as a way of getting two readouts from the same chip. For example, if I performed a ChIp-ChiP experiment with 4 antibodies on the same fibroblast cells I would use 9 chips. For each antibody, I would prepare three technical replicas, and label them with Cy3 or Cy5 and place them on different chips. I would do likewise with one input sample split to three technical replicas. For the last sample, I would perform mock IPs, three technical replicas, using IgG or a non specific antibody, e.g. anti-GST or anti-FLAG. See below.
- For your experiments involving an IP, consider performing a mock IP using IgG or an antibody that doesn't bind anything within your sample, e.g. anti-GST or anti-FLAG. It is best to use a type matched antibody. TiMAT2 is set up to generate empirical FDRs based on the mock IP and is a great way to estimate confidence in your real IP data without having to perform exhaustive qPCR. One consideration though, it must be performed in parallel with the real IPs. Subtle differences in washing and amplification really effect the output.
- Don't over wash your beads during the IP. Use the minimum number of washes (3x 5min?) that give good fold enrichment for known targets by qPCR. Over washing can lead to a PCR bottle neck effect where a small number of regions get overly amplified and will look exactly like real regions on the microarray even though nothing specific was pulled down by the IP.
- Commercial chips are very consistent. Technical repeats (using the same hybridization solution on different chips) are unnecessary and can emphasize false positives in certain testing situations.
- You might want to prep twice the amount of chromatin you will need to be sure to not run out if and when parts of the experiment need to be repeated.
- Save enough pooled pre amplification real IP and input material for ~50 qPCR rxns each.