The principle of ChIP-chip is simple. The first step is to cross-link the protein-DNA complex. This is done using a fixating agent, such as formaldehyde. The cross-linking can later be reversed with heat. Cross-linking kills the cell, giving a snapshot of the bound TF at a given time. The cell is then lysed, the DNA sheared by sonication and the chromatin[2] (TF-DNA complexes) is pulled down using an antibody (i.e. immunoprecipitated). If an antibody for the TF is available, then it is used; otherwise, the TF is tagged with an epitope targeted by commercially available antibodies (the latter option is cheaper, but runs the risk of altering the TF's functionality). Cross-linking is then reversed to free the bound DNA, which is then amplified, labeled with a fluorophore and dumped onto a DNA-array. The scanned array reveals the genomic regions bound by the TF. The resolution is around ~500 bp as a result of the sonication step.

DNA-arrays (or DNA-chips or microarrays) are flat slabs of glass, silicon or plastic onto which thousands of multiple short single-stranded (ss) DNA sequences (corresponding to small regions of a genome) have been attached. After performing a mRNA extraction in induced and non-induced cells, the mRNA is again reverse transcribed, but here the reaction is tweaked, so that the emerging cDNA contains nucleotides marked with different fluorophores for controls and experiment. Targets will hybridize by base-pairing with those probes that resemble them the most. The array can then be stimulated by a laser and scanned for fluorescence at two different wavelengths (control and induced). The ratio or log-ratio between the two fluorescence intensities corresponds to the induction level.

Electro-mobility shift-assays (or gel retardation assays) are a standard way of assessing TF-binding. A fragment of DNA of interest is amplified and labeled with a fluorophore. The fragment is left to incubate in a solution containing abundant TF and non-specific DNA (e.g. randomly cleaved DNA from salmon sperm, of all things) and then a gel is run with the incubated sample and a control (sample that has not been in contact with the TF). If the TF has bound the sample, the complex will migrate more slowly than unbound DNA through the gel, and this retarded band can be used as evidence of binding. The unspecific DNA ensures that the binding is specific to the fragment of interest and that any non-specific DNA-binding proteins left-over in the TF purification will bind there, instead of on the fragment of interest. EMSAs are typically carried out in a bunch of fragments, shown as multiple double (control+experiment) lanes in a wide picture. Certain additional controls are run in at least one of the fragments to ascertain specificity. In the most basic of these, specific competitor (the fragment of interest or a known positive control, unlabelled) is added to the reaction. This should sequester the TF and hence make the retardation band disappear, proving that the binding is indeed specific

In motif discovery, we are given a set of sequences that we suspect harbor binding sites for a given transcription factor. A typical scenario is data coming from expression experiments, in which we wish to analyze the promoter region of a bunch of genes that are up- or down-regulated under some condition. The goal of motif discovery is to detect the transcription factor binding motif (i.e. the sequence “pattern” bound by the TF), by assuming that it will be overrepresented in our sample of sequences. There are different strategies to accomplish this, but the standard approach uses expectation maximization (EM) and in particular Gibbs sampling or greedy search. Popular algorithms for motif discovery are MEME, Gibbs Motif Sampler or CONSENSUS. More recently, motif discovery algorithms that make use of phylogenetic foot-printing (the idea that TF-binding site will be conserved in the promoter sequences for the same gene in different species) have become available. These are not usually applied to complement experimental work, but can be used to provide a starting point for it. Popular algorithms include FootPrinter and PhyloGibbs.

Once the binding motif for a TF is known, this motif (which essentially defines a pattern) can be used to scan sequences in order to search for putative TF-binding site. This is useful, for instance, when trying to identify TF-binding site in ChIP-chip data. Searching for TF-binding site can be done in numerous ways. The most basic method is consensus search, in sequences are scored according to how many mismatches they have with the consensus sequence for the motif. A more elaborate way of searching involves using regular expressions, which allow to search for more loosely defined motifs [e.g. C(C/G)AT]. Common algorithms for this type of search include Pattern Locator and the DNA Pattern Find method of the SMS2 suite, but also some word processors. Finally, the mainstream way of conducting TF-binding site search is through the use of position-specific scoring matrices, which basically count the occurrences of each base at each position of the motif and use the inferred frequencies to score candidate sites. Algorithms in this last category include TFSEARCH, FITOM, CONSITE, TESS and MatInspector.