How do site-specific DNA-binding proteins find their targets?

Abstract

Essentially all the biological functions of DNA depend on site-specific DNA-binding proteins finding their targets, and therefore 'searching' through megabases of non-target DNA. In this article, we review current understanding of how this sequence searching is done. We review how simple diffusion through solution may be unable to account for the rapid rates of association observed in experiments on some model systems, primarily the Lac repressor. We then present a simplified version of the 'facilitated diffusion' model of Berg, Winter and von Hippel, showing how non-specific DNA-protein interactions may account for accelerated targeting, by permitting the protein to sample many binding sites per DNA encounter. We discuss the 1-dimensional 'sliding' motion of protein along non-specific DNA, often proposed to be the mechanism of this multiple site sampling, and we discuss the role of short-range diffusive 'hopping' motions. We then derive the optimal range of sliding for a few physical situations, including simple models of chromosomes in vivo, showing that a sliding range of approximately 100 bp before dissociation optimizes targeting in vivo. Going beyond first-order binding kinetics, we discuss how processivity, the interaction of a protein with two or more targets on the same DNA, can reveal the extent of sliding and we review recent experiments studying processivity using the restriction enzyme EcoRV. Finally, we discuss how single molecule techniques might be used to study the dynamics of DNA site-specific targeting of proteins.

Figure 1

Schemes for target site location. Three commonly discussed microscopic pathways for transferring a protein from one site to another along a long DNA molecule are ‘sliding’, ‘hopping’ and ‘intersegmental transfer’. (Top) A protein might ‘slide’ along the double helix, transferring from one base pair position to the adjacent one without dissociating from the DNA. Many repeated sliding events result in 1-D diffusion of the protein along the DNA contour. (Center) If dissociation occurs, the protein might re-encounter the same DNA, but at a new contour position: we term such an event a ‘hop’. (Bottom) On scales beyond the persistence length of the DNA double helix, 150 bp (50 nm), the DNA can run into itself as a result of its random thermally excited bending. Such encounters permit the protein to move from one DNA site to another via an intermediate in which the protein is bound transiently to both sites, a process called ‘intersegmental transfer’.

Figure 2

The trajectory of a diffusing protein is a ‘random walk’. A 3-D random walk of 10⁶ steps, each of length 1, is shown here as a projection on a 2-D plane. The end-points of the walk are at (0,0) and at (–300,–300). The overall size of the random walk is roughly the square root of the number of steps, or ∼1000 steps, as expected from the mean square law R² = Dt. Although this law holds quantitatively only in statistical terms when applied to many random walks, the overall size of one random walk is usually reasonably well estimated by the mean square law. It is important to note that the random walk has many voids; it does not completely explore the 3-D region it extends through.

Figure 3

Probability of finding a target of size a by 3-D diffusion. The diffusing protein, at an initial distance r from the target, will either collide with the target or diffuse off into bulk solution, never finding this target. During its random walk within a distance r of the target, the protein visits a fraction a/r of the ‘voxels’ of size a (see text). The probability of encountering the target is thus a/r.

Figure 4

The sliding length. A protein binds non-specifically to the DNA double helix (left) and then undergoes sliding steps randomly to the left and to the right, exploring the DNA contour through 1-D diffusion. Eventually, a dissociation event occurs; the characteristic distance explored between association and dissociation events is the sliding length. Due to the random nature of 1-D diffusion, the same DNA sites will be sampled repeatedly.

Figure 5

The targeting radius. A protein starting close to its target has a high probability of binding to the specific site; a protein starting far from its target has a low probability of binding specifically. The targeting radius ξ is the starting distance from which the probability of specific binding is 0.5. For a protein starting a distance ξ away, the two outcomes of specific binding (solid line pathway) and escape (dashed line pathway) occur with equal probability.

Figure 6

Association rate for the hopping + sliding model of Berg, Blomberg, Winter and von Hippel, normalized to the diffusion limit (k/Da), as a function of sliding length in units of the specific target size (l_sl/a). The case where sliding and 3-D diffusion occur at equal rates (D₁/D = 1) is presented. Two specific target concentration cases are shown: (upper line) low target concentration (a²Lc = 10^–6); (lower line) high target concentration (a²Lc = 10^–2). The total association rate shows a maximum at a certain optimal sliding length. For the optimal sliding length, 3-D motion keeps the protein from spending too long ‘oversampling’ any particular region of the DNA contour by 1-D diffusion. The optimal sliding length becomes shorter for higher target concentrations.

Figure 7

Plasmids and catenanes. The plasmid (on the left) contains a single recognition site (indicated with a hatchmark) within a specified segment of the circular DNA (indicated in blue). The remainder of the DNA (in red) contains only non-specific sequences. A protein (green sphere) bound to the non-specific (red) DNA can reach the recognition site by 1-D diffusion around the circle or by 3-D transfer (not shown). The plasmid is converted by a recombinase into a catenane (on the right) with two interlinked circles of DNA: a small circle (in blue) that carries the recognition sequence; a large circle (in red) with only non-specific sequences. A protein bound to the non-specific DNA in the large (red) circle cannot reach the recognition site in the small circle by 1-D diffusion but must instead dissociate from the large ring before re-associating with the small ring: since both rings are covalently continuous, the dissociation cannot occur from an end of a chain.