Footprinting

Footprinting identifies transcription factor (TF) binding sites by analyzing the pattern of DNA cleavage events. When a TF is bound to DNA, it sterically hinders the access of enzymes like DNase I or Tn5 transposase, creating a protected region flanked by accessible chromatin. This results in a characteristic signal dip or 'footprint' in the sequencing read density profile. The algorithm scans for these localized minima in the continuous accessibility signal, distinguishing true footprints from sequencing noise or nucleosome-depleted regions through statistical modeling of the expected cleavage pattern.

Two primary assays enable footprinting analysis, each with distinct characteristics:

• DNase-seq: Uses DNase I enzyme for digestion, providing nucleotide-resolution cleavage patterns. The enzyme's larger size (~37 kDa) creates wider footprints (8-30 bp), offering high specificity but requiring 10-50 million cells.
• ATAC-seq: Employs hyperactive Tn5 transposase (~53 kDa) that simultaneously fragments and tags open chromatin. Requires only 500-50,000 cells, making it suitable for rare cell populations. However, the transposase's sequence bias and larger footprint size can reduce resolution compared to DNase-seq.
• Both methods generate strand-specific cleavage patterns that reveal the precise orientation and boundaries of TF occupancy.

Several specialized algorithms perform footprint detection from sequencing data:

• Wellington: Uses a binomial test on strand-specific cleavage counts flanking candidate motifs, comparing observed cleavage patterns to expected background distributions.
• HINT (Hmm-based IdeNtification of Transcription factor footprints): Applies Hidden Markov Models to segment the genome into footprint, flanking accessible, and background states based on cleavage density.
• CENTIPEDE: Integrates sequence motif information with DNase-seq cleavage patterns using a hierarchical Bayesian model to estimate posterior probabilities of TF binding.
• Basset: A deep convolutional neural network that learns to predict accessibility profiles and can identify footprints by analyzing in silico mutagenesis effects on predicted cleavage patterns.

The depth of a footprint—the magnitude of the signal reduction relative to flanking regions—correlates with TF occupancy and residence time. Key quantitative metrics include:

• Footprint Occupancy Score (FOS): The ratio of cleavage events within the protected region versus the surrounding accessible chromatin.
• Protection Score: A normalized measure of how effectively the bound protein blocks enzyme access, often calculated as 1 - (cleavage within footprint / cleavage in flanking regions).
• Differential footprinting: Statistical frameworks, often based on negative binomial models, compare footprint depths between experimental conditions to identify condition-specific TF binding.
• These metrics enable quantitative comparisons of binding affinity across cell types, treatments, or genetic backgrounds without requiring antibody-based enrichment.

Footprinting alone identifies binding locations but does not reveal which TF is bound. Motif analysis resolves this by scanning footprinted regions for known TF binding motifs:

• De novo motif discovery: Algorithms like MEME or HOMER identify overrepresented sequence patterns within footprinted regions.
• Motif database scanning: Footprints are intersected with position weight matrices from databases such as JASPAR or HOCOMOCO to assign candidate TFs.
• Footprint shape analysis: The cleavage pattern within a footprint—including its width and asymmetry—provides additional clues about the molecular identity of the bound protein.
• Combined footprinting and motif analysis enables genome-wide TF occupancy maps without the need for TF-specific antibodies.

Footprinting enables high-resolution mapping of regulatory element usage:

• Variant interpretation: Footprinting can assess whether non-coding genetic variants disrupt TF binding by comparing footprint depth between reference and alternate alleles, providing functional evidence for disease-associated SNPs.
• Regulatory network reconstruction: Genome-wide footprint maps reveal the hierarchical organization of TFs, identifying pioneer factors that open chromatin and downstream effectors that bind newly accessible sites.
• Cell-type-specific regulation: Comparative footprinting across cell types identifies differentially occupied binding sites that drive tissue-specific gene expression programs.
• Pharmacological profiling: Footprinting can measure the genome-wide effects of drug treatments on TF occupancy, enabling assessment of therapeutic mechanisms and off-target effects.