Molecular Models of Fis Binding to DNA

In order to understand the sequence logo for Fis binding sites [5], we constructed three dimensional models of Fis-DNA binding. The models of Fis interacting with DNA were built using Insight II software from Biosym Technologies, Inc., on an IRIS computer (Silicon Graphics, Inc.), and displayed with RasMol 2.5, available at
http://molbiol.soton.ac.uk/rasmol.html
or
ftp://ftp.dcs.ed.ac.uk/pub/rasmol/ .
The Fis protein coordinates are those of the Protein Data Bank
( http://www.rcsb.org/pdb/ ) entry 1fia .

In Fis-DNA models built by earlier workers either B-form DNA with no particular base sequence was smoothly bent and placed next to the X-ray crystal model of Fis [25,26], or the single distal hin Fis binding site (V01370, 180, 8.9 bits) was used and kinks added to the DNA at positions based on the DNase I hypersensitive pattern [27]. Since the sequence logo gives a detailed model for the sequence conservation at Fis sites, we used this information to construct three models:

• a single Fis homodimer binding to the consensus sequence as determined by the logo (Fig. 3A).

• two Fis homodimers binding to two Fis sites spaced 11 base pairs apart on DNA (Fig. 3B).

• two Fis homodimers binding to two Fis sites spaced 7 base pairs apart on DNA (Fig. 3C).

We used several criteria for making the first model. As noted by previous workers, Arg85 and Lys91 in the D helix are likely to be involved in DNA binding since mutations at these amino acids interfere with DNA binding [25,26]. The logo shows highly conserved G-C base pairs at -7 and +7in major groove positions, as is common when contacts are made by proteins with helix-turn-helix motifs [27,28]. In addition, methylation of these bases blocks Fis from binding [29] and Fis protects these bases against methylation [30]. However, as mentioned in previous studies, with B-form DNA the G's at $\gamma = 16$would be further apart than the distance between the two DNA-contacting D helices of Fis. Fis can deflect DNA from 40 degrees to 90 degrees [2,25,27,31], so we can attain a shorter distance between the two DNA contact points at $\gamma = 16$either by `bending' (a gradual curvature of the helix axis over a sequence of more than two bases) or by `kinking' (a sharp change in the DNA strand's axial direction between two adjacent bases) at one or more locations between the two contacts [32].

If specific contacts are not formed, a smoothly bent DNA might be expected to produce smooth sequence conservation over the binding region, but the logo shows strong variable conservation of base sequences at positions -5 through +5, so it is likely that the deflections are not smooth. Furthermore, both CA and TG, which are conserved at positions -4 to -3 and +3 to +4[26] (i.e. positions $$H_g = e(G) -\sum_{b=A}^{T} p(b) \log_2 p(b)$$, have been shown to display unusually large roll (rotation about the long axis between base pairs [33]. The CRP-DNA complex displays a 90 degrees bend that results from CA and TG kinking [34,35]. (See [28] for the CRP sequence logo.)

Additionally, the other two pyrimidine-purine (Py-Pu) base sequences conserved in Fis sites, TA [36] and CG [37,38,39,40] show this high-roll characteristic. These four kinking dinucleotides appear frequently at $$H_g = e(G) -\sum_{b=A}^{T} p(b) \log_2 p(b)$$ and $$H_{after} = \sum_{l=1}^L \biggl( e(n(l)) -\sum_{b=A}^{T} f(b,l) \log_2 f(b,l) \biggr),$$. They account for the base frequencies in the logo and they identify kink locations which could result in the observed DNA deflection [41]. Examination of the individual Fis binding sites confirms this observation. With the exception of seven Fis sites in our set: fis X62399 292, nrd K02672 3266, aldB L40742 153, hin V01370 68, cin X01828 289, gin M10193 389, and hns X07688 655, sites which do not have a Py-Pu step at $$H_g = e(G) -\sum_{b=A}^{T} p(b) \log_2 p(b)$$do have one displaced just one position left or right [34]. In addition, these Py-Pu kinks form during energetic minimization of a Fis-DNA complex [42]. We therefore constructed our model using two 42 degree kinks at $$H_g = e(G) -\sum_{b=A}^{T} p(b) \log_2 p(b)$$. More kinking would prevent Fis from fitting into the two successive major grooves. Two 50 degree kinks were also added at positions $$H_{after} = \sum_{l=1}^L \biggl( e(n(l)) -\sum_{b=A}^{T} f(b,l) \log_2 f(b,l) \biggr),$$because this was the largest observed kink found for CAP protein binding [35].

The midsection of the logo is AT rich and this should create a higher twist, leading to a compression in the minor groove [32,43,26]. We incorporated this compression into our model by making the helical repeat through the central A-T tract (bases -2 to +2) 10 base pairs per turn and setting the remainder of the model to 10.6 base pairs per turn for B-form DNA [44,45,46].

We tried to maximize the overall DNA bend angle of Fis bound to DNA and found that the best we could do, while avoiding stereochemical collisions, was 60 degrees. This model of a single Fis homodimer contacting DNA (Fig. 3A) has two kinks and one compression that allow the four nitrogens of Arg85 to be within contact distance (<3.5 angstroms) of either the O6 or N7 acceptor of each G at $\gamma = 16$and Lys91 to be within 3.5 angstroms of the phosphate backbone at $\sim$with no major structural conflicts.

To investigate the consequences of two Fis molecules binding to nearby sites, we constructed 3 dimensional models with Fis sites separated by 11 or 7 base pairs. For the two overlapping Fis models, we extended the general scheme of modeling a Fis homodimer to the synthetic DNA sequences used in our gel shift experiment.

For the overlap 11 model (Fig. 3B), different kinks from the two adjacent sites would be in the same place. Because 50 degree kinks would prevent Fis from fitting into either site, we made both 42 degrees. We incorporated 50 degree kinks on the outside two positions, as in the single site model. We also increased the twist to -36 degrees at all A-T pairs within the central region [36].

For the overlap 7 model (Fig. 3C), the two outside TG and CA ends were kinked 50 degrees to conform to the single Fis model. However, two pairs which would normally be kinked were not Py-Pu pairs in this sequence. We excluded kinking of the left site at $$H_{after} = \sum_{l=1}^L \biggl( e(n(l)) -\sum_{b=A}^{T} f(b,l) \log_2 f(b,l) \biggr),$$ due to a Py-Py pair (CT) and we excluded kinking of the right site at -7.5 due to a Pu-Pu pair (AG). Therefore we only added 42 degree kinks at three positions: at two Py-Pu pairs (left -3.5 CA and right +3.5 TG) which correspond to the single Fis model, and at one in the center (TA). As in the overlap 11 model, twists of -36 degrees were made throughout any A-T tract to be consistent with the single homodimer model.

We found that two Fis proteins separated by 11 base pairs would strongly interpenetrate. On the other hand, two Fis proteins separated by 7 base pairs have a minimal van der Waals force conflict between the two central D helices, but this might be accommodated for by flexibility of the DNA-protein complex. Given that the D helices cannot fit directly into B-form DNA, there is some uncertainty as to how Fis binds DNA. We thought that 11-base separated Fis molecules (Fig. 3B) would compete for binding but that a 7-base separation (Fig. 3C) might allow simultaneous binding.