Dynamics of the HD regulatory subdomain of PARP-1; substrate access and allostery in PARP activation and inhibition

Abstract PARP-1 is a key early responder to DNA damage in eukaryotic cells. An allosteric mechanism links initial sensing of DNA single-strand breaks by PARP-1’s F1 and F2 domains via a process of further domain assembly to activation of the catalytic domain (CAT); synthesis and attachment of poly(ADP-ribose) (PAR) chains to protein sidechains then signals for assembly of DNA repair components. A key component in transmission of the allosteric signal is the HD subdomain of CAT, which alone bridges between the assembled DNA-binding domains and the active site in the ART subdomain of CAT. Here we present a study of isolated CAT domain from human PARP-1, using NMR-based dynamics experiments to analyse WT apo-protein as well as a set of inhibitor complexes (with veliparib, olaparib, talazoparib and EB-47) and point mutants (L713F, L765A and L765F), together with new crystal structures of the free CAT domain and inhibitor complexes. Variations in both dynamics and structures amongst these species point to a model for full-length PARP-1 activation where first DNA binding and then substrate interaction successively destabilise the folded structure of the HD subdomain to the point where its steric blockade of the active site is released and PAR synthesis can proceed.


i) Theory:
The fastest that an NH group can exchange with solvent corresponds to the "random coil" rate, which is the rate for an NH that experiences no protection through intramolecular hydrogen bonding; this is characterised by the rate constant krc (see point v) below). Participation in intramolecular hydrogen bonds within a folded structure retards this rate to an extent that depends on the strength of the hydrogen bond. This behaviour is usually represented using the simplified reaction scheme:  (1): Often this equation appears without the term kop in the denominator, e.g. in (2), corresponding to the assumption that all hydrogen bonds are strong, *+ ≫ ,-; however, here it is of interest to plot the behaviour for both strong and weak hydrogen bonds so the kop term on the lower line is retained. In the figure, those parts of the curves for which kcl > kop (implying that the closed form containing the hydrogen bond predominates) are shown using thick solid lines, while those parts for which kcl < kop (implying that the open form where there is no hydrogen bond predominates) are shown using thin dotted lines.
Because solvent exchange is a chemical reaction, it shows strong dependence on solution conditions, particularly pH and temperature, and in addition is sensitive to the chemical nature of nearby sidechains. For an amide group in a random coil conformation, i.e. one that is "unprotected" as it participates only in transient hydrogen bonds with solvent, the random coil rate solvent exchange constant krc can be roughly predicted from model compound studies (3); for a protein at pH 7.0 and 25 °C, krc would be expected to be very roughly 10 s -1 , with variations of up to about a factor of ten in either direction depending on sequence context. This is the value that was used in the simulations shown in the figure; for more details see point vi) below.

ii) EX2 and EX1 limits:
Two limiting cases are usually considered, which differ according to whether it is the first step (structural opening) or second step (exchange) that is rate-limiting; the first limit is referred to as EX1 and the second as EX2 (4). By far the commonest situation is for exchange to be in or near the EX2 limit, where structural fluctuations are much faster than intrinsic exchange, *+ + ,-≫ .* , so that Eq. 1 becomes: In effect, this implies that an equilibrium is rapidly established between the closed and open states, and the observed exchange rate then reflects both the fractional abundance of the opened state ( ,-) and the rate of exchange events it undergoes once formed. For hydrogen bonds strong enough that *+ ≫ ,-, the EX2 condition is often written as where Kop is the equilibrium constant for the pre-equilibrium between open and closed states.
For extremely slow rates, exchange can reach the EX1 limit, *+ + ,-≪ .* , and then Eq. 1 becomes simply: On the figure, the EX1 limit manifests as the coalescence into a single line of all of the curves for which kcl < 1 s -1 , in the region marked with a dashed red box. The EX2 limit is less obviously visualisable on the figure, but corresponds to those curves for which kcl > 100 s -1 (i.e. the bulk of the remainder of the plot), where increasing kcl leads to a proportionate movement of the kex vs kop line.
It may be noted that, formally, Eq. 1 is an approximation, since between the EX1 and EX2 limits (here corresponding approximately to the range 1 s -1 < kcl < 100 s -1 ) exchange follows a bi-exponential rate that cannot be represented analytically using a single rate constant (5); however, this complication is often ignored.

iii) Fastest events:
The NH protons that undergo the fastest solvent exchange are those situated in particularly mobile loops, where they participate only very weakly or not at all in intra-molecular hydrogen bonding; they are largely or completely unprotected from solvent exchange and consequently exchange at or near the random-coil rate. Under the conditions of the present study (25 °C and pH7), krc is roughly 10 s -1 (see vi below). However, the rate of underlying conformational motions for such protons is very much faster. For such motions to be detected in 15 N relaxation experiments, they must be faster than overall molecular tumbling, which for PARP-1 CAT domain at 25 °C in the present study has a correlation time of * ≈ 25 ns (see main text).
In other words, in such cases there are at least seven orders of magnitude difference between the rates of NH proton solvent exchange and the underlying rates of conformational exchange events involving those protons. This corresponds roughly to the part of the plot highlighted in red labelled "fastest". In this regime it would be expected that the conformational motions of individual peptide groups are largely un-coordinated with those of their neighbours, and any transient making or breaking of intramolecular hydrogen bonds involving these peptides would probably be largely or completely non-cooperative in nature.

iv) Slowest events:
In contrast, NH protons that exchange with solvent the most slowly can be at the EX1 limit, where the rate of solvent NH proton exchange is equal to the rate of the underlying conformational processes responsible for transiently breaking hydrogen bonds. Such cases could lie within the part of the plot highlighted in blue labelled "very slow" (though still slower cases of exchange corresponding to still stronger hydrogen bonds, and which would lie outside the limits of this plot, are also entirely possible). Hydrogen bonds involved in these processes are very strong indeed, and the processes that break them are likely to be highly co-operative in nature, involving the disruption of large sections of the structure (2,6,7); in the limit, for NH protons involved in the very strongest hydrogen bonds in the most stable of protein structures, solvent exchange can require global unfolding of the entire structure (8). Cases of EX1 exchange are relatively rare, although working at high pH makes EX1 more likely as it increases greatly the random coil exchange rate (2).

v) Between the extremes:
The great majority of hydrogen-bonded NH protons characterised in exchange studies fall between the limiting behaviours just described, and are mostly within the EX2 regime. Generalisations are difficult, but it seems reasonable to expect that the trend from fastest to slowest solvent exchange would correspond very approximately to a trend from least to most co-operativity in the conformational processes underlying exchange; for the great majority of hydrogen bonds, the conformational fluctuations required for transient opening will involve some other portions of nearby structure that are interconnected in the hydrogen bonding network, and the stronger the hydrogen bond, the larger the "structural reach" of such co-operative events is likely to have to be (8). Implications for the relationship between solvent exchange rates and rates of the underlying conformational fluctuations causing exchange are also hard to predict, but one may expect that the slower the exchange, the smaller is likely to be the extent by which conformational fluctuation rates are faster than solvent exchange. Even so, the EX2 regime covers a very broad range of NH solvent exchange rates, and in the majority of cases it is likely that the underlying conformational fluctuations that drive exchange are significantly faster than solvent exchange itself, potentially by orders of magnitude.
Without measured data concerning kop and kcl, quantitative statements are impossible.
However, the purpose of the analysis presented here is to extend the caution given elsewhere in the literature, e.g. in (2,7), against assuming that measured solvent exchange rates are similar to the rates of underlying conformational transitions responsible for that exchange; this is highly unlikely to be the case except for the very slowest exchanging NH protons that are at or near the EX1 limit.

vi) Random coil exchange rate at pH7 and 25 °C:
The prediction for krc used in the simulations was obtained from the data and analysis in Bai et al. (3) as follows: For a random coil amide group, the solvent NH exchange rate constant uncorrected for neighbouring sidechain effects is given by where kA is the rate constant for acid-catalysed exchange, kB is that for base catalysed exchange and kW is that for water catalysed exchange. At 20 °C, the temperature for which Bai et al. give tabulated data, the value of KD (the self-ionisation constant for D2O) is 15.049 (9). At pH 7, catalysis by acid and water can be neglected, so after conversion to log units Eq. Corrections for local sequence (see Bai et al. Table 2) can cause a maximal negative variation in kB of -0.97 log units (for the NH of Ile in the sequence Ile-Pro) and maximal positive variation in kB of +1.45 log units (for the NH of Cys in the sequence Cys-His + ). These corrections would correspond to a range in the rates between 0.91 s -1 and 241 s -1 . The average over all sidechain corrections (ignoring COOH forms, as these would be absent at pH 7) would give -0.02 to +0.14 log units, which corresponds to a range in the rates between 8.16 s -1 and 11.8 s -1 .
Using the Arrhenius equation to correct these values from a temperature of 20 °C to 25 °C gives: .* ( ) = .* (293) ( Thus, the random coil exchange rate constant at 25°C, uncorrected for sequence variations, is approximately 13.9 s -1 . This value was rounded to 10 s -1 for use in the simulations. Supplementary Figure S20. Schematic [ 15 N, 1 H] correlation plots showing chemical shift differences for the amide group 1 H and 15 N signals of residues in helix F for the comparisons indicated on the individual panels; in each case the WT free protein signal is indicated in orange and the mutant or inhibitor complex signal in light blue. The corresponding schematic for helix F of WT protein alone is also shown. showing that no large-scale perturbation of secondary structure results from binding of these inhibitors or from these mutations. Apparent scaling differences amongst the spectra of the mutants probably reflect errors in estimating concentrations between these different protein samples, whereas the curves for inhibitor complexes all rely on a single protein concentration estimate. All spectra were recorded at 20°C.  660  680  700  720  740  760  780  800  820  840  860  880  900  920  940  960  980 (Å)   660  680  700  720  740  760  780  800  820  840  860  880  900  920  940  960  980 αH   660   680  700  720  740  760  780  800  820  840  860  880  900  920  940  960  980 Figure S25. B factors for each chain in each of the structures of free PARP-1 CAT domain (7AAA, panels A and B) and its complexes with veliparib (7AAC, panels C and, and 2RD6, panel E), olaparib (7AAD, panels F and G), talazoparib (4UND, panels H and I, and 4PJT, panels J -M), and EB-47 (7AAB, panels N and O, and 6BVQ, panels P -S), as well as for chains C and F of the complex of PARP-1 F1, F3 and WGR-CAT with a DNA duplex (4DQY, panels T and U). Note the different vertical scales: panels A-O run to 150 Å 2 , panels P-S run to 200 Å 2 , and panels T and U run to 300 Å 2 . The atoms used for superposition were as follows: HD = 666-721, 730-743, 750-779 (N,C a ,C'); ART = 790-936, 939-1009 (N,C a ,C'); Set 1 atoms are used to calculate transformed co-ordinates resulting from fitting; Set 2 atoms are used to calculate an rms difference without further changing the co-ordinates.