To dissect DNA SSB recognition by PARP-1 at the single-molecule level, we used a single stranded DNA molecule designed to form a dumbbell like structure, in which an SSB is created between two hairpins (in the following, the stem carrying the free 5' terminus is called the 5' stem and that carrying the free 3' terminus is called the 3' stem) (Fig. 1b, Methods). The positions of two fluorophores were optimized on either side of the nick, ATTO 550 on the 3' stem and Alexa647 on the 5' stem, respectively, to sensitively monitor DNA conformations by measuring the smFRET efficiency. Using time-resolved fluorescence spectroscopy (Methods), smFRET efficiencies were determined from free DNA as well as from the DNA in presence of PARP-1 (Fig. 1c). Due to the design of the DNA ligand, smFRET data can be used to assess the kinking angle between the two DNA stems (θ in Fig. 1d).
Comparison of the smFRET data obtained from the DNA containing the SSB to that of a ds-DNA molecule obtained by ligation with T4 DNA ligase yielded virtually identical smFRET efficiency histograms (Fig. 1c), with peak FRET efficiencies of E=0.50±0.03 and of E=0.48±0.03, respectively (Supplementary Fig. S2a,b and Supplementary Table S1). Apparently, under the buffer conditions used here, the nicked DNA adopts a linear conformation in solution which is stabilized by stacking interactions, similarly to that of normal B-DNA.
PARP-1 binding leads to a pronounced kink in the DNA ligand containing a SSB
Binding of PARP-1 introduces a pronounced kink in the DNA and therefore leads to a shift of the observed smFRET efficiency histogram to higher values (high FRET peak at EHF=0.87, Fig. 1c, Supplementary Fig. S2e). Moreover, the observed histogram is asymmetric with a pronounced shoulder at lower FRET efficiencies, indicating conformational heterogeneity of the DNA-protein complex. Interestingly, when using a fragment of PARP-1 containing only its first two zinc finger domains (residues 1-214, F1F2), we observe a smFRET histogram practically indistinguishable from that in presence of the full length protein (Fig. 2, green EHF=0.86 see also Supplementary Fig. S2d, E and Supp. Table 1). Thus, binding of the first two zinc fingers is sufficient for forming the fully kinked DNA structure.
In contrast, when using a shorter fragment of PARP-1 containing only zinc finger domain 2 (residues 103-214, F2) the recorded smFRET data yielded a FRET efficiency histogram shifted towards higher efficiencies as compared to DNA alone, but the efficiencies are not as high as those observed in the presence of F1F2 (Fig. 2). The histogram for the F2 complex appears comparatively broad and again not entirely symmetric. This could be explained either by a remaining fraction of unbound DNA, by structural heterogeneity of the bound DNA complex, or by conformational changes on a time-scale faster than the experimental observation time (about 1ms). Taken together, these results show that F2 alone is able to induce a kink in the DNA, but kinking is not as pronounced as that caused by binding of F1F2.
While smFRET data provide a very sensitive tool for investigating structural changes, changes in smFRET efficiencies can also be caused by changes in the photo-physics of the dye molecules. We therefore performed control experiments with a slightly modified DNA construct as well as a different donor dye molecule. These experiments yielded closely comparable results (Methods, Supplementary Fig. S3), confirming our interpretation that the observed changes in smFRET efficiency histograms are caused by structural re-arrangement.
Quantification of kink angles
Next, we wanted to quantify the degree of kinking of the DNA bound to different PARP-1 fragments, namely F2 and F1F2. While smFRET data contains distance information, such an interpretation is not straightforward, since the 1D distance information about the inter-dye distance cannot be directly converted into a 3D model structure. However, if a model structure is available for a particular conformation, computational approaches can be used to compute the expected smFRET efficiency (Muschielok et al. 2008; Kalinin et al. 2012; Beckers et al. 2015; Peulen, Opanasyuk, and Seidel 2017; Reinartz et al. 2018; Nagy, Eilert, and Michaelis 2018; Eilert et al. 2018). This computed efficiency can then be compared to the experimentally measured FRET efficiency in order to establish whether or not the structure is in agreement with the experimental data, taking into account error estimates for both the smFRET efficiency measurement and the smFRET efficiency computation. Here, we integrated such a “backwards” approach with structural ensemble calculations in order to identify, at a single-molecule level, structural heterogeneity DNA conformations which agree with our experimental smFRET data and to analyze the distribution of kink angles in these ensembles in order to determine the extent of DNA kinking in each complex. The workflow is illustrated in Fig. 3a and explained in the following.
Simulations can predict the measured FRET efficiency for a given model structure
In order to simulate the FRET efficiency of a given structure, the first step is to calculate the accessible volume (AV) of both dyes, given the attachment points of the dyes and the geometry of the flexible linkers that attach them to the DNA (Methods) (Muschielok et al. 2008). Examples of the resulting AVs are shown for the structure of straight DNA (Fig. 3b), and represent the space that the dye can access. Once the AVs are calculated, we use Markov Chain Monte Carlo simulations together with Bayesian Parameter Estimation (Eilert et al. 2018) to determine the expected FRET efficiency for a given structure (Methods).
We first tested this approach using a model structure of the DNA in a linear, i.e. B-form, DNA conformation (Fig. 3b). The simulation yields a predicted FRET efficiency for this structure of Esim=0.51 ± 0.04 which agrees well with the measured value of Eexp=0.50 ± 0.03. The given error for the simulated FRET efficiency arises predominantly from the uncertainty in the measured isotropic Förster radius of 3% (Riso = 70 Å ± 2 Å, Methods), which translates to an uncertainty of ΔE=0.04 for the simulated FRET efficiency.
The next step towards determining the kinking angle of the nicked DNA molecule was to generate a large ensemble of sterically possible DNA conformations by keeping both stems in fixed conformations while varying their relative orientation (with the respect to the flexible region at the nick, Fig. 3c) in any way that did not lead to steric clashes between the stems (Methods). This leads to a very heterogeneous distribution of sterically allowed DNA conformations (Fig. 3d). For each of these conformation we then simulated the expected FRET efficiency. The resulting simulated FRET efficiencies show a broad distribution covering the range from Esim=0.42 to Esim=1 (Fig. 3E). Fig. 3f illustrates these very DNA structures in a spherical coordinate system where the angle between the two stems (3' stem represented by the black line and 5' stem represented by the grey dot) was computed for every structure (Methods).
Next, in order to compare the generated structural ensembles to the recorded smFRET, we selected those structures from the ensemble which agreed with the respective measured FRET efficiencies. To this end, for each structure we tested whether the simulated FRET efficiency was within ΔE=0.05 of the measured smFRET efficiency, where the uncertainty was obtained using error propagation, considering an error contribution of 0.03 from the experimental FRET efficiency and 0.04 from the simulated FRET efficiency (Methods). From the ensemble of 1000 structures, 87 structures were in agreement with the smFRET data for DNA alone (i.e. Esim=0.50±0.05, blue structures in Fig. 4a), 154 with the data for DNA+F2 (Esim=0.69±0.05, red structures in Fig. 4a) and 199 structures with the data for DNA+F1F2 (Esim=0.86±0.05, green structures in Fig. 4a). The respective ensembles of selected structures in Fig. 4a show that the three different FRET efficiency ranges lead to distinct distributions of kink angles θ (Fig.4b and c). For a better overview of the selected structures we chose a representation which displays the respective structures with reduced complexity, highlighting only the relative directions of the two DNA stems (Methods, Figure 4b). The side view shows that the allowed kinking angle regimes are well separated for the three ensembles (upper plot in Fig. 4c). From the view along the 3' axis (lower plot in Fig. 4b), it is apparent that the selected structures lie on circles around the 3' axis, indicating that our approach gives no information on the angle ϕ through which the 5' stem is rotated. This is not surprising, as the structure selection is based on a single distance between the two dyes. From the histograms we determined the mean kinking angles of <θ>=176°±24° for DNA only, <θ>=130°±20° for DNA+F2 and <θ>=90°±14° for DNA+F1F2 (Methods).
This kinking angle determination relies on the structural ensemble of the nicked DNA molecules. However, as a control we also computed an alternative structural ensemble by modeling F2 bound to the 3' stem (Methods) and again simulated the expected FRET efficiency for each structure in the ensemble (Supplementary Fig. S4). Again we compared the simulated FRET efficiencies to the experimental data given the uncertainty of ΔE=0.05 to define structural sub-ensembles for the complex in presence of F2 or F1F2 which are in agreement with the experimental data (Supplementary Fig. S4). From the respective structural ensembles, we again computed the mean kinking angles <θ>=127°±19° for DNA+F2 and <θ>=87°±15° for DNA+F1F2, which is in excellent agreement with the analysis based on the structural ensemble of the DNA only.
Dynamic analysis reveals fast dynamics for nick DNA in presence of PARP-1
As described above, the observed smFRET histograms for DNA in presence of F2, F1F2 and full length PARP-1 showed shoulders at lower FRET efficiencies (Fig. 2). A possible reason for this could be averaging over different structural states that interconvert on a time scale on the order of 1 ms. We therefore used several analytical approaches to test the data for evidence of dynamics in these structures (Supplementary methods). First, we investigated the dependence of smFRET efficiency on burstwise donor lifetimes in presence of the acceptor (Supplementary Fig. S5a). When comparing these lifetimes with the ratiometrically determined FRET efficiencies, one can analyze in two-dimensional histograms whether the observed histograms fall on the so called static FRET line or whether deviations can be seen that are caused by dynamic interconversion of linear and kinked conformations (Methods, Kalinin et al. 2010) . While for the free DNA the observed 2D histogram falls on the static FRET line (Supplementary Fig. S5a), there are clear deviations observed for the DNA in presence of F2, F1F2 or PARP-1, indicating the presence of a dynamic interconversion. Similarly, burst variance analysis (BVA, Torella et al. 2011) also showed indications of dynamic interconversion for DNA+F2, DNA+F1F2 and DNA+PARP-1, but not for the DNA in absence of proteins (Supplementary Fig. S5b). More direct information about the respective dynamics with time scales between 0.2 and 5 ms comes from time-window analysis, where histograms of FRET events were computed using different time-binnings (Methods, Supplementary Fig. S5c). Again, for the case of DNA alone, no dynamics were observed. Interestingly, for the DNA in presence of protein only very small changes were observed, indicating that dynamics occur on an even faster time-scale.
Besides actual dynamics, dye-photophysics could also influence the observed smFRET efficiency histogram. However, by using a Pulsed Interleaved Excitation scheme (Müller et al. 2005) we were able to investigate these effects directly. To this end, we performed FRET-FCS experiments for the single molecule burst data sub-ensemble consisting of the acceptor only species (Supplementary Fig. S6, Supplementary methods). This analysis revealed that the acceptor, Alexa647, attached to the nick DNA in the absence or in the presence of either F2 or F1F2 has a short relaxation time of 6-7 µs (Supplementary Fig. S6b, ). This is in good agreement with previous reports (Vandenberk et al. 2018; Baibakov and Wenger 2018), and is presumably caused by cis-trans photo-isomerization, characteristic of cyanine dyes (Widengren, Mets, and Rigler 1995; Widengren and Schwille 2000). For the full length PARP-1, the acceptor relaxation time is slightly longer (~10 µs) in comparison to results for the zinc finger fragments (Supplementary Fig. S6b, ), which is most probably due to direct interactions with the additional domains in a so called PIFE effect (Stennett et al. 2015).
To provide more evidence concerning the nature of the species involved in the dynamic interconversion between the linear and kinked DNA conformations, we employed FRET-FCS experiments (Schwille, Meyer-Almes, and Rigler 1997; Torres and Levitus 2007 Supplementary methods). These show that the translational diffusion times (Supplementary Fig. S7) in our confocal microscope for the low FRET population of DNA with PARP-1 (1990 µs) and for that of DNA with PARP-1 bound to niraparib (1976 µs), are clearly distinct from that of DNA alone (1303 µs). These diffusion times provide strong support for our interpretation that the protein remains bound to the DNA in these low-population states where the DNA is linear.
Next, we performed a quantitative analysis of the observed fluorescence dynamics caused by structural changes using filtered FCS (fFCS, Methods, Felekyan et al. 2012; Barth et al. 2018). From the observed smFRET efficiency histograms of both DNA+F2 and DNA+F1F2, we selected smFRET efficiency regions towards the edge of the histogram for interconversion analysis (Fig.5, blue and violet areas in the FRET efficiency distributions). By globally fitting the fFCS auto- and cross-correlation functions, the relaxation time for interconversion between the two respective species can be determined (Methods, Fig.5 and Supp. Table S2).
PARP-1 inhibitors can influence the dynamics of PARP-1
Next, we investigated whether the observed distribution of binding states can be modulated with PARP-1 inhibitors (PARPi). Some PARPi were recently shown to drive the PARP-1 allostery and activate or inhibit its function (Zandarashvili et al. 2020; Ogden et al. 2021). Here, we characterize effects caused by EB-47 and niraparib, which have been interpreted as “pro-retention” and “pro-release” inhibitors, respectively (Zandarashvili et al. 2020; Slade and Eustermann 2020). We observe that EB-47 traps the PARP-1-DNA complex in the kinked state, as the ELF (low FRET effciency) population decreases substantially in comparison to the situation without inhibitor (Fig. 6a, magenta). Interestingly, niraparib has a different effect on the observed smFRET histogram, since in this case the ELF population is increased rather than decreased (Fig. 6b). Note, a slight shift of the position of the peaks in the smFRET histograms in the presence of niraparib relative to the other observed smFRET distributions is not due to differences in the conformation, but instead is caused by the high fluorescence anisotropy (Supplementary Fig. S8).