Sub-nanosecond tryptophan radical deprotonation mediated by a protein-bound water cluster in class II DNA photolyases

Light activation of class II DNA photolyases is enhanced by a unique cluster of protein-bound water molecules.


Introduction
The exposure of DNA to UV light causes serious damage of the genetic code and eventually results in fatal mutations. The most prominent forms of UV-induced lesions are cyclobutane pyrimidine dimers (CPDs) and the pyrimidine(6-4)pyrimidone photoproducts ((6-4)PPs). Several mechanisms evolved to restore the integrity of DNA, like nucleotide excision repair, base excision repair and photorepair. 1,2 The latter is catalysed by photolyases, a class of avoenzymes, which belong to the photolyase/cryptochrome-superfamily (PCSf). 3 Photolyases are substrate-specic and they are divided into  and CPD photolyases, which are further subdivided into classes I-III. Compared to other members of the PCSf, the class II CPD photolyases, which occur in plants, animals and many microbial organisms, are highly divergent in terms of their sequences, especially in functionally important parts of the catalytic C-terminal domain. [4][5][6] Cryptochrome (Cry) blue light receptors of plants and animals evolved from class I CPD and (6-4) photolyases, respectively, but they have mostly lost the ability to repair DNA. They are involved in multiple light-regulation processes 3 and possibly also in the light-dependent 'magnetic compass' of migrating birds and certain other animals. 7,8 For DNA repair, all photolyases depend on a fully reduced FAD cofactor, FADH À . Upon photoexcitation, FADH À transfers an electron to the damaged DNA and thereby catalyzes the repair of the lesion. Isolated photolyases oen contain a semireduced (FADHc) or even a fully oxidized avin (FAD ox ). 4,[9][10][11] Photocatalytically active FADH À is generated from oxidized avin species by a second light-induced reaction, usually referred to as photoactivation. In the case of initially oxidized FAD ox , the FADc À resulting from the primary photoinduced electron transfer (ET) becomes protonated (on a timescale of a few hundred milliseconds to a few seconds) to form FADHc. 12,13 Further reduction of FADHc to FADH À can be achieved by absorption of another photon, inducing transfer of a second electron to the avin cofactor.
The mechanism of photoactivation has been studied in detail in vitro for a class I CPD photolyase (from E. coli) and for a  photolyase (from X. laevis). [12][13][14][15][16][17][18] Upon photoexcitation, FAD ox or FADHc abstract an electron from the rst member of a chain of three 14 (or four 12 ) tryptophan residues to form a primary FADc À TrpHc + radical pair (in 0.5 to 0.8 ps) 17,18 or an FADH À TrpHc + pair (in $30 ps), 14 respectively. The electron hole then migrates along the Trp chain from the avin-nearest Trp towards the most distant one, stabilizing the pair thermodynamically and by its spatial separation within less than 100 picoseconds. [16][17][18] Further stabilization of the pair is achieved by deprotonation of the exposed terminal TrpHc + cation radical by the solvent, which typically occurs within a few hundreds of nanoseconds. 12,14,19 Finally, the resulting Trpc radical is scavenged by an extrinsic reducing agent. Similar light-induced electron and proton transfer reactions were observed in Crys. 20 Despite the low sequence identity with other photolyases and cryptochromes (<16%), 21 class II CPD photolyases such as the one studied here -MmCPDII from the archaeon Methanosarcina mazeishare the same overall structural fold, with an N-terminal domain comprising a Rossman-like fold and an a-helical C-terminus which harbors the catalytically active FAD cofactor. 4 Interestingly, class II CPD photolyases differ from other branches of the PCSf by the localization of the tryptophan cascade and by the presence of auxiliary tyrosine residues ( Fig. 1). A previous mutational study of this ET pathway revealed that the Trp triad can also be functional as a dyad including only the rst two FAD-proximal tryptophans. 4 In this study, we investigated the photoreduction of FAD ox to FADHc via FADc À in MmCPDII photolyase using transient absorption spectroscopy. We identied a cluster of water molecules involved in an unprecedentedly fast deprotonation of the distal tryptophan, Trp 388 . Additionally, we demonstrated that a conserved tyrosine, Tyr 345 , also participates in electron transfer to photoexcited FAD. This work represents the rst endeavor to thoroughly characterize the initial photoactivation step in a class II CPD photolyase.

Experimental
Multiple sequence alignment-based analysis of class II CPD photolyases 451 non-redundant class II sequences with a pairwise sequence identity of less than 90% were extracted from a previous sequence-similarity network analysis of the photolyasecryptochrome superfamily (combined PFAM protein families PF00875 and PF04244) 22 using the Cytoscape suite. 23 The multiple sequence alignment was done using Clustal omega. 24 WebLogo 25 was used for the visualization of the degree of position-specic conservation.

Protein preparation
Cloning of MmCPDII mutants. The generation of the E387Q mutant was based on pET-28a-MmCPDII and done according to the phusion protocol (NEB) using the 5 0 -phosphorylated primers listed below. The preparation of the W388F and Y345F mutants was described earlier. 4 E387Q primer Sense: 5 0 -Pho-CAG TGG AGC GAA TCT CCC GAA AAA-3 0 Reverse: 5 0 -Pho-CAG AAT TTT TTT TGC CCA GTA CAT GCG-3 0 Overexpression and purication of MmCPDII and mutants. Overexpression of MmCPDII and mutants was done as described previously using E. coli BL21-Gold(DE3) cells (Stratagene). 4 The cultivation was done in terric broth medium for 24 hours at 25 C (20 C for W388F). The proteins were puried using a NiNTA column (MACHEREY-NAGEL) with 50 mM NaH 2 PO 4 , 300 mM NaCl, pH 8.0 and SEC column with Superdex 200 material (GE-Healthcare) with 10 mM Tris-HCl of pH 8.0 and 100 mM NaCl.

Experimental conditions
Unless otherwise stated, the solutions of wild-type (WT) and all mutant MmCPDII photolyases studied herein contained 10 mM Tris-HCl buffer at pH 8.0 (measured at room temperature), 100 mM NaCl and 10% (v/v) glycerol. For the experiment in D 2 O, 10 mM Tris-HCl buffer (at pH 8.0) with 100 mM NaCl was lyophilised to dry powder and the sublimated H 2 O was replaced by D 2 O, adding up to the original volume of the H 2 O buffer; note that both H 2 O and D 2 O samples in this experiment (Fig. 5) were hence glycerol-free. In the experiments where cysteine was added as an external reducing agent, cysteine was rst dissolved in a more concentrated Tris-Cl buffer with NaCl. Aer titration by NaOH back to pH 8.0 and the addition of glycerol and water to the desired nal volume, a stock of 500 mM cysteine solution was nally obtained in the standard 10 mM Tris-Cl buffer with 100 mM NaCl and 10% (v/v) glycerol. The addition of cysteine to the photolyase sample has hence diluted the protein but the concentrations of other components (buffer, salt and glycerol) and the pH were kept constant. Cysteine was chosen rather than the more commonly used reductant dithiothreitol because the latter was shown to be less efficient in a similar system and to cause protein precipitation at high concentrations. 26 All samples were air-saturated and kept at 7 C during the measurements or on ice in between. Before each experiment, they were rid of free FAD and other low-molecular-weight impurities by ltration over size-exclusion columns (Micro Bio-Spin, Bio-Gel P-6). The UV/vis spectrum was checked before and aer each measurement to ensure that the sample was in a good shape, i.e., not aggregated and FAD was not released, but protein-bound and fully oxidized (>95% FAD ox ). UV/vis spectra were recorded on a Uvikon XS spectrometer (Secomam).

Transient absorption spectroscopy
Transient absorption kinetics were measured on three different setups described in detail in ref. 12, 13, 19 and 27. In experiments on ps/ns timescales, the photolyase samples were excited at 355 nm by a Nd:YAG laser (Continuum Leopard SS-10, pulse duration of 100 ps, repetition rate 1 or 2 Hz, and an energy in the order of 5 mJ per cm 2 ).
In all other experiments, the samples were excited at 470 nm by laser ashes of 5 ns duration and an energy #10 mJ cm À2 , delivered by a Nd:YAG pumped optical parametric oscillator (OPO; Brilliant B/Rainbow, Quantel, France).
Indicative values of excitation energies were obtained by measuring the laser pulse energy behind a cell lled with H 2 O using an energy meter (Gentec QE25SP-H-MB-D0).
For kinetic measurements on the ps/ns and ns/ms timescales (with 2 GHz and 100 MHz bandwidth limits, respectively), the monitoring light was provided by continuous-wave lasers as listed in ref. 13. 2 Â 2 Â 10 mm cells were used (excitation pulses entered the sample through the 2 Â 10 mm window; monitoring light through the 2 Â 2 mm window). The monitoring light beams were attenuated by neutral density lters and mechanically chopped to produce a rectangular light pulse of 140 ms duration and energy in the order of 1 mJ at the entrance of the cell, thus avoiding signicant actinic effects. This pulse was synchronized with the excitation laser ash (see ref. 27 for more details).
For experiments on millisecond to second timescales, the monitoring light was provided by a tungsten halogen lamp. For selection of a specic wavelength, an interference lter of 5 to 10 nm spectral bandwidth was inserted between the lamp and the sample. A similar lter was placed in front of the detector to block scattered light from the excitation ash and uorescence. The bandwidth of the amplier was limited to 30 or 100 kHz.

Signal analysis
Transient absorption signals were tted either individually or globally (with shared time constants) using the unweighted Levenberg-Marquardt least-squares minimization algorithm and monoexponential (Fig. 3, 4, 5, 7 (inset), 9, 10, 11 and 13) or biexponential ( Fig. 6) decay functions (ExpDec1 and ExpDec2, respectively) in Origin 8.6. A non-zero offset was allowed in the cases where there was a stable residual absorption at the end of the kinetics (due to formation of longer-lived species).
Note that the used value of 3 457 (FADc À ) was obtained from the published spectrum 31 of FADc À in an insect cryptochrome. The spectrum of FADc À in MmCPDII is not available but it is likely not exactly identical to that in the insect cryptochrome, which could have a certain impact on the accuracy of the calculated quantum yields.

Results
The primary goal of this study was to characterize the rst photoactivation step, i.e., the photoreduction of FAD ox to FADc À /FADHc, in the wild-type (WT) class II CPD photolyase (MmCPDII) and to nd out how its class-specic tryptophan triad, Trp 381 -Trp 360 -Trp 388 , participates in this process. Our transient absorption spectroscopic data have conrmed the functionality of the tryptophan cascade but they have brought about an additional issue: it turned out that upon electron transfer (ET) to the excited FAD ox , one of the oxidized tryptophans (TrpHc + ) in the triad undergoes an unprecedentedly fast deprotonation, that is by three orders of magnitude faster than the terminal tryptophans in other PCSf proteins (typically a few hundreds of nanoseconds). 12,14,19 This unusually fast rate indicated that the proton is probably transferred to a nearby, structurally-dened acceptor, rather than to disordered bulk solvent.
In order to identify the Trp undergoing the fast deprotonation and the proton acceptor, we have rst examined the W388F mutant protein, in which the third (terminal) tryptophan, Trp 388 , was replaced by a redox-inactive phenylalanine, which cannot participate in photoinduced electron transfer to the avin cofactor. Experiments on this mutant protein unequivocally pointed to the terminal tryptophan Trp 388 being the fast proton donor and we have hence searched for the likely proton acceptors in its vicinity.
For a better understanding of the following transient absorption data, absorption spectra of the species that could contribute to changes in transient absorption (FAD ox , FADc À , FADHc, TrpHc + , Trpc and Tyrc) are shown in Fig. 2.

Wild-type MmCPDII
Isolated MmCPDII photolyase (WT, as well as all mutants studied here) contains a fully oxidized FAD cofactor (FAD ox ), which has two pronounced absorption maxima in the near-UV/ vis region (see Fig. 2): a double maximum centred around 370 nm and a triple band centred around 445 nm. FAD ox in MmCPDII absorbs up to 500 nm and can hence, in principle, be excited anywhere below this wavelength. For our initial experiments, we have chosen an excitation wavelength of 470 nm (provided by a Nd:Yag-pumped optical parametric oscillator, OPO; pulse duration of $5 ns) to avoid interference with our monitoring light sources and to minimize artifacts due to hydrated electrons that are formed upon excitation in the UV.{ Based on our previous experience with other members of the PCSf, we have rst looked at timescales of a few ns to tens of ms, where we anticipated to observe formation of an FADc À TrpHc + pair and deprotonation of TrpHc + to Trpc. Transient absorption changes at representative wavelengths are shown in Fig. 3a, all signals are shown in Fig. S1. † All traces exhibit a step-like increase or decrease, the amplitudes of which (extrapolated to t ¼ 0) are represented by black squares in Fig. 3b. If the anticipated FADc À TrpHc + pair were formed, one would expect an absorbance increase below 415 nm and a bleaching between 415 and 485 nm (due to the transformation of FAD ox into FADc À ) Fig. 2 UV/vis spectra of species susceptible to contribute to transient absorption changes following the photoexcitation of FAD ox in MmCPDII. The FAD ox spectrum was measured in MmCPDII and scaled to 3 (at l max ) ¼ 11 300 M À1 cm À1 . 44 The FADHc spectrum was constructed as described previously 19 using the MmCPDII FAD ox spectrum and that of a mixture of FAD ox and FADHc in the same sample (obtained by partial photoreduction). The FADc À spectrum (from an insect cryptochrome) and spectra of Trp and Tyr radicals are taken from the literature. 31,45,46  combined with an absorption increase between 450 and 650 nm (due to the formation of TrpHc + )see red dashed line in Fig. 3b. Surprisingly, the signals at 562 and 594 nm (close to the absorption maxima of TrpHc + ) and at longer wavelengths exhibited only very small amplitudes compared to the bleaching around 450 nm, excluding the presence of signicant amounts of TrpHc + beyond the given instrument response time ($5 ns). On the other hand, the amplitudes of the absorption changes could be reasonably described by a difference spectrum reecting the formation of FADc À and a deprotonated tryptophan radical Trpc (see Fig. 3b). Assuming the signals are due to 100% FADc À Trpc pair, the quantum yield of FAD ox photoreduction in the rst nanoseconds reaches $55% (see the Experimental section for quantum yield determination).
In order to see if we can detect the FADc À TrpHc + pair (precursor of the FADc À Trpc pair), we have switched to another experimental setup enabling us to access faster timescales. 27,28 To avoid time limitations due to the 5 ns long excitation pulse duration provided by the OPO, we have used a frequency-tripled Nd:YAG laser at 355 nm with 100 ps pulses. The response time of this setup ($200 ps) has allowed us to observe signals ( Fig. 4a and S2 †), the initial amplitudes of which were indeed compatible with the presence of an FADc À TrpHc + radical pair (Fig. 4b).
The initial amplitudes partially decayed with a time constant of $350 ps to yield remaining amplitudes with a difference spectrum consistent with an FADc À Trpc radical pair (Fig. 4b). We hence attribute the 350 ps phase to deprotonation of TrpHc + to Trpc ($85%) in competition with FADc À TrpHc + recombination ($15%; the corresponding partial recovery of FAD ox is well visible, e.g., at 457 nm). Consistent with our attribution to a deprotonation reaction, the kinetics of this phase slows down from $350 ps in an H 2 O buffer to $800 ps in a D 2 O buffer (Fig. 5), corresponding to a kinetic isotope effect (KIE ¼ k H /k D ) of 2.3.
Within the 80 ms experimental time window shown in Fig. 3a, one can observe the beginning of a virtually uniform decay of an FADc À Trpc radical pair. Using a third setup adapted for monitoring of processes on milliseconds to seconds timescales, we have obtained signals containing the complete decay kinetics (Fig. 6a). It turned out that a biexponential decay function had to be used in order to obtain a good t, which was an indication of recombination of two distinct pairs of radicals. Global t of all signals yielded amplitudes for the two processes: those attributed to the faster process (with a time constant of 225 ms and amounting to $70% of the total signal amplitude at Fig. 4 (a) Flash-induced absorption changes on a ps/ns timescale for 64 mM WT MmCPDII at five selected wavelengths (see Fig. S2 † for all measured wavelengths). (b) Initial (extrapolated to t ¼ 0) and final amplitudes for all measured signals compared to difference spectra for the formation of FADc À + TrpHc + and FADc À + Trpc, respectively, and containing small amounts (7 and 4%, respectively) of hydrated electrons e À aq (see note { for more details). The sample was excited at 355 nm by a 100 ps pulse of E $ 5 mJ per cm 2 . The traces are averages of 16 to 64 signals recorded with a repetition rate of 1 Hz. 450 nm) were indeed consistent with the recombination of FADc À Trpc pairs but those attributed to the slower process (with a time constant of 1.1 ms and corresponding to the remaining $30% of DA 450 nm ) exhibited a near-zero absorption change at 540 nm, which was incompatible with a tryptophan radical being the recombination partner of FADc À . When looking at the structure of MmCPDII, 4 one can notice a tyrosine residue (Tyr 345 ) in the vicinity of the terminal tryptophan of the ET chain (3.8Å edge-to-edge distance from Trp 388 ; Fig. 1). An involvement of this tyrosine residue in ET to the excited FAD ox was anticipated, since the mutation of Tyr 345 to a red-ox inactive phenylalanine was previously shown to slow down the rate of FAD ox photoreduction in a steady-state experiment (compared to the WT protein). 4 Indeed, the 1.1 ms phase nicely ts the difference spectrum for disappearance of an FADc À Tyrc pair (Fig. 6b).
Finally, in the absence of external reducing agents, all transient species are completely lost due to recombination within less than 5 milliseconds in the WT MmCPDII (Fig. 6a) and the initial state of the protein with fully oxidized FAD is restored.
In order for the photoactivation reaction to be efficient, the FADc À anion radical has to be stabilized by scavenging of its recombination partner (be it Trpc or Tyrc) by extrinsic reductants. By adding sufficient amounts of cysteine to reduce transiently formed Trpc and/or Tyrc radicals, 26 we could compete against the FADc À Trpc/Tyrc recombination ( Fig. 7; the acceleration of the decay of DA 540nm reects the reduction of Trpc by cysteine; the subsequent rise at 0.3 M cysteine is attributed to protonation of FADc À , see below) and obtain an isolated metastable FADc À radical, which got further stabilized by protonation to yield a neutral FADHc radical (see inset of Fig. 7; formation of FADHc from FADc À is accompanied by a pronounced absorption increase at 610 nm and decrease at 380 nm, see spectra in Fig. 2). Under our experimental conditions (0.01 M Tris-HCl buffer at pH 8.0, 0.1 M NaCl, 10% (v/v) glycerol, 7 C), this protonation occurred with a time constant of $630 ms. However, a closer look at the 540 nm signal with the highest cysteine concentration (0.3 M cysteine; Fig. 7) suggests that, in a small fraction of proteins, FADc À can be protonated at a much faster rate, which is reected by the  growth phase (with s $ 2.2 ms) following the initial decay of the signal. Judging from the initial amplitude of the signal at 610 nm in the inset of Fig. 7, the fast protonation seems to be possible in $15% of proteins (note that only FADHc should absorb at 610 nm, contributions from FADc À or remaining Trpc/Tyrc radicals are expected to be zero at this wavelengthsee Fig. 2).

W388F mutant lacking the 3 rd tryptophan of the triad
In order to nd out which of the three tryptophan residues undergoes the unusually fast deprotonation in the WT MmCPDII ( Fig. 4 and 5), we decided to examine the behaviour of its W388F mutant, in which the terminal Trp of the triad was replaced by a non-reducing phenylalanine. Our structural data for this and the Y345F mutant (PDB codes 5O86, 5O8D) 32 show that there are no compensatory structural changes of residues lining the ET pathway, which could complicate the following analysis and interpretation. The transient absorption signals obtained for the W388F mutant on the ns/ms timescale upon excitation by 5 ns pulses at 470 nm (Fig. 8a) exhibit steep decays in the rst nanoseconds, followed by plateaus, the spectral footprint of which corresponds neither to an FADc À TrpHc + , nor to an FADc À Trpc pair, but is compatible with an FADc À Tyrc pair (Fig. 8b), formed with a quantum yield of mere $4% (see the Experimental section for quantum yield determination). The nearest tyrosine to the 2 nd Trp of the triad (Trp 360 ), which could serve as electron donor to Trp 360 Hc + , is Tyr 380 (4.6Å edge-toedge distance; Fig. 1). Tyr 380 is situated in the vicinity (4.0Å) of yet another tyrosine Tyr 445 , which is more exposed to solvent and could hence play the role of the terminal electron donor to FAD in the W388F mutant. Alternatively, Trp 360 Hc + could also be reduced by Tyr 345 (7.1Å edge-to-edge distance), which is involved in ET in the WT protein.
Analogously to the situation in the WT protein, we had to use a different experimental setup to resolve the fast process preceding the formation of the FADc À Tyrc pair. On the ps/ns timescale, we were able to resolve signals, which were decaying nearly completely with a time constant of $1.2 ns (Fig. 9a). The initial amplitudes spectrally t an FADc À TrpHc + pair (Fig. 9b). We tentatively assign the 1.2 ns decay in the W388F mutant protein to charge recombination in the pair FADc À Trp 360 Hc + . This recombination presumably competes with a much slower (in the order of 10 to 20 ns) side ET from Tyr 380 (or Tyr 445 ) to Trp 360 Hc + , yielding an FADc À Tyrc pair with a quantum yield of $4%.
In the absence of extrinsic reducing agents, the FADc À Tyrc pairs in W388F recombine with a time constant of $2.7 ms (Fig. 10), which is $2.5Â slower than the FADc À Tyrc recombination in the WT (the observed Tyrc radical is hence most probably not Tyr 345 c in the W388F mutant MmCPDII). A closer look at Fig. 10b (and also at Fig. 3 and 8) reveals that the amplitudes of signals above 550 nm are not zero, contrary to expectation for an FADc À Tyrc pair according to the reference spectra shown in Fig. 2. The absorption changes on these timescales can no longer be attributed to hydrated electrons because e À aq are much shorter-lived. Albeit small, the signals seem to be real and they decay with the same kinetics as the signals at other wavelengths. We hence tentatively attribute the absorption in the red to a particular shape of the FADc À spectrum in MmCPDII. Note that FADc À spectra with weak absorption in the red have been reported before for some other proteins of the PCSf. 33,34 Like the Trpc and Tyrc radicals in the WT, the Tyrc radical in the W388F mutant could be scavenged in the presence of cysteine and the resulting metastable FADc À was further stabilized by protonation to form FADHc (see Fig. S3 †) at the same rate as in WT MmCPDII under the same conditions (i.e., in $630 ms).

E387Q mutant: searching for an intra-protein proton acceptor
According to structural analysis of the environment of the distal tryptophan Trp 388 in MmCPDII (PDB entry 2XRZ), a straight- forward candidate for the intramolecular acceptor of the N1 proton from Trp 388 Hc + is deprotonatedk Glu 387 (see Fig. 12): its carboxylic group is as close as 3.6Å from the N1 atom of Trp 388 . We therefore prepared a mutant, in which the glutamate was replaced by glutamine, which could not serve as a proton acceptor. Fig. 11 compares signals recorded for the WT protein and the E387Q mutant at 562 nm, which is a suitable wavelength for direct monitoring of the deprotonation of TrpHc + to Trpc (and/or its reduction by a tyrosine). The gure clearly shows that the kinetics of TrpHc + disappearance was almost as fast as in the WT ($500 ps according to a monoexponential t in E387Q vs. $350 ps in WT; a biexponential t for E387Q yielded $400 ps and $3 ns at an amplitude ratio of 10 : 1 but the rootmean-square deviation was only marginally better, see Fig. S6 †), indicating that Glu 387 is not the primary proton acceptor. As outlined in the Discussion section, we identied a proteinbound water cluster ideally positioned to serve as the primary proton acceptor.
Y345F mutant: identication of the tyrosine involved in ET As mentioned above, there were indications that the tyrosine residue Tyr 345 also participates in electron transfer to the  photoexcited avin in MmCPDII. We have hence investigated the involvement of this residue in ET by direct comparison of time-resolved spectroscopic signals of the WT and the Y345F mutant MmCPDII.
Signals in Fig. 13 show a decay of ash-induced absorption changes in the Y345F mutant on the millisecond timescale. Unlike in the WT protein, the signals decay monoexponentially and the slower (1.1 ms) component present in WT (attributed to the FADc À Tyrc pair and representing approx. 30% of the radical pairs present at the beginning of the ms time scale) is obviously missing. The decay rate of the remaining faster phase (s $250 ms) is very close to the value obtained from the t for the faster process in the WT protein. We conclude that indeed a fraction of tyrosine Tyr 345 ($30%) gets oxidized during photoactivation of WT MmCPDII and forms the longest-lived (1.1 ms) radical pair with FADc À .
With respect to the kinetics and pathway of formation of the tyrosyl radical Tyr 345 c, we considered two (mutually nonexclusive) possibilities: (i) fast ET from Tyr 345 to Trp 388 Hc + (in competition with deprotonation of Trp 388 Hc + in 350 ps) and (ii) slow reduction by Tyr 345 of the deprotonated Trp 388 c radical (by proton-coupled ET in competition with recombination of the FADc À Trp 388 c pair in 225 ms). The similarity of the initial signal amplitude ratios at different wavelengths in the Y345F mutant and the WT protein on the ms time scale (compare Fig. 6a and 13) seems to contradict fast formation of a substantial amount of tyrosyl radical in WT. Nevertheless, we have also compared the signals at 408 nm (wavelength of Tyrc absorption maximum) obtained using our fastest experimental setup (Fig. S5 †). If the $30% of Tyrc radical were formed directly from TrpHc + (in competition with its deprotonation and/or its recombination with FADc À ), the amplitudes of the $350 ps decay in the 408 nm signal would have to be clearly different in the WT and in the mutant protein: we would expect that the decay due to $15% recombination would be visibly (almost fully) compensated by a growth due to the formation of the Tyrc radical in the WT protein. In the Y345F mutant, on the other hand, no Tyrc radical can be formed, and the phase should hence reect pure and uncompensated FADc À TrpHc + recombination (absorption changes due to TrpHc + deprotonation are negligible at 408 nm; see Fig. 2). Given that there is no signicant difference in the ps/ns signals for the two proteins, we can conclude that most of the Tyrc radicals are formed lateron the ms timescale in competition with the recombination of FADc À with neutral Trpc.

Discussion
Based on all the data obtained for WT and mutant proteins (in the absence of extrinsic reductants), we have constructed a reaction model (Scheme 1) of the primary processes following FAD ox photoexcitation in WT MmCPDII.
We observed formation of an FADc À TrpHc + pair within our experimental time resolution of 200 ps, followed by deprotonation of TrpHc + in $350 ps, concomitant with $15% charge recombination in the FADc À TrpHc + pair (Fig. 4). Assuming direct competition between deprotonation and charge recombination, the intrinsic time constants would be $0.4 ns and $2 ns, respectively. We assign the deprotonation to Trp 388 , the third (most distant to FAD) member of the triad Trp 381 -Trp 360 -Trp 388 (Fig. 1), because the deprotonation was lost (and replaced by charge recombination in $1.2 ns) in the W388F mutant (Fig. 9).
In analogy to the extensively studied class I CPD photolyase from E. coli (EcCPDI), we suppose that ET from Trp 388 to *FAD ox occurs by ultrafast hopping along the Trp triad also in MmCPDII. Losses may occur due to competition of forward ET with charge recombination in the radical pairs. We have obtained a quantum yield of $55% for the formation of the FADc À Trp 388 c pair (see Results and Experimental section). Taking into account charge recombination in competition with deprotonation of Trp 388 Hc + (see above), the quantum yield of formation of the FADc À Trp 388 Hc + pair should be $65%, i.e., the same as for the terminal FADc À TrpHc + pair in EcCPDI, 13 but substantially higher than in a (6-4) photolyase ($30%) 12 or in a plant cryptochrome ($20%). 19 The particular Trp triad in MmCPDII is hence as efficient in electron transfer as the best known "standard" Trp triad. For the time constants of the hopping steps along the Trp triad, our data can only set an upper limit of $100 ps for each of these steps. Given that the edge-to-edge distance between the avin and the proximal Trp in MmCPDII (4.8Å) is larger than in EcCPDI (3.7Å), the rst electron transfer step (from Trp 388 H to the excited avin) is likely to be slower than the 0.8 ps in EcCPDI. 17 The real kinetics of the electron hopping steps in MmCPDII remain to be resolved by ultrafast methods.
Our data ( Fig. 13 and S5 †) imply that the tyrosine residue Tyr 345 (situated 3.8Å from the third tryptophan of the cascade, Trp 388 ) acts as a fourth auxiliary member of the electrontransferring chain, though only about 30% of the Trp 388 Hc + /Trp 388 c radicals seem to eventually get reduced by Tyr 345 under our experimental conditions. Because of spectral congestion, we were not able to monitor the kinetics of tyrosine oxidation directly, but from the lack of a signicant effect of the Y345F mutation on the ps/ns signals at 408 nm (Fig. S5 †) we could conclude that most of Tyr 345 c was formed on a slower time scale, i.e., aer the deprotonation of Trp 388 Hc + . Direct oxidation of Tyr 345 by Trp 388 c seems highly unlikely because it would imply a hydrogen atom transfer over a distance of more than 6Å. We rather suggest that Tyr 345 is mostly oxidized by Trp 388 Hc + that is present in a very small amount in thermal protonation equilibrium with Trp 388 c. Charge recombination in the FADc À Trpc pair may also proceed via thermal reprotonation of the tryptophanyl radical. The observed biphasic recombination kinetics in WT MmCPDII (tted time constants of 225 ms ($70%) and 1.1 ms ($30%); see the Results section and Fig. 6a) can then be described in the framework of Scheme 1 assuming effective time constants of $300 ms and $1 ms for the charge recombination of FADc À Trp 388 c and the competing oxidation of Tyr 345 by Trp 388 c, respectively, followed by a 1.1 ms recombination of the FADc À Tyr 345 c pair (see ESI and Scheme S1 † for more details).
The ms/ms charge recombination reactions can be blocked (and FADc À hence stabilized) by extrinsic electron donors (which are abundant in living cells) that scavenge the Trpc/Tyrc radicals. In this situation, the isolated FADc À in MmCPDII becomes protonated in $630 ms to form a metastable FADHc (Fig. 7). This rate of protonation is slower than the reported $200 ms in the Xenopus laevis (6-4) photolyase 12 under similar conditions (0.05 M Tris-HCl buffer at pH 8.3, 0.05 M NaCl, 5% (v/v) glycerol, 10 C) but faster than the $4 seconds observed in the Escherichia coli class I CPD photolyase 13 (0.02 M phosphate buffer at pH 7.5, 0.2 M NaCl, 20% (v/v) glycerol, 7 C), in spite of a higher pH (8.0) used in the present experiment. The conversion of FADHc to the redox state active in DNA repair, i.e., the fully reduced FADH À , can follow directly aer the absorption of another photon. However, in the presence of oxygen (and in the dark), both FADHc and FADH À in isolated MmCPDII spontaneously revert to FAD ox (within a few minutes in an air-saturated solution).
It is a matter of controversy 35,36 whether photoactivation of DNA photolyases through their respective Trp chains is a vital process in vivo or whether the FAD cofactor is naturally and always fully reduced in the living cell, but given that at least two such ET chains have evolved independently in the photolyasecryptochrome superfamily and survived billions of years of evolution, we dare speculate that the efficient photoactivation of photolyases through the Trp (or Trp/Tyr) chains could be important, especially under intense solar irradiation (which causes simultaneously DNA damage and oxidative stress that could potentially deactivate DNA repair by photolyases by oxidation of their FADH À ). In any case, the electron transfer cascade of three tryptophans and one tyrosine (Trp 381 -Trp 360 -Trp 388 -Tyr 345 in MmCPDII) is almost strictly conserved within all class II photolyases (Fig. S7 †)

Previous observations in the context of our new data
MmCPDII mutant proteins, in which one of the tryptophans of the triad was replaced by non-reducing phenylalanines, were shown to exhibit slower rates of in vitro FAD ox photoreduction under steady-state irradiation in the presence of 25 mM dithiothreitol (DTT) as an extrinsic reducing agent. 4 Mutation of the rst Trp of the triad (W381F) had the strongest inhibitive impact and essentially blocked the photoreduction of FAD ox . Mutation of the last Trp of the triad (W388F) had the smallest impact, slowing down the FAD ox photoreduction rate by a factor of $2 (with respect to the WT protein under the same conditions). 4 This could seem to be in disaccord with the quantum yield of the FADc À Tyrc radical pair estimated here (4%), which is substantially lower than the 55% of the FADc À Trpc pair in the WT, but one has to bear in mind that: (a) the FADc À Tyrc pair in W388F is longer lived (2.7 ms) than the corresponding FADc À Trpc and FADc À Tyrc pairs in the WT (0.3 and 1.1 ms, respectively), which compensates for the lower yield by giving the extrinsic reducing agents more time to scavenge the Tyrc radical and stabilize FADc À in the W388F mutant, (b) the accessibility of the different radicals to the extrinsic reductants and thereby also the efficiency of the productive encounter of the redox partners is likely to be different, and (c) the systems in the steady-state experiment 4 were in dynamic equilibria, as the FADHc formed by illumination was continuously reoxidized back to FAD ox by molecular oxygen 38 present in the air-saturated samples. In any case, given the fast forward ET from Trp 388 H to Trp 360 Hc + in WT MmCPDII (<200 ps), the alternative electron transfer pathway involving Tyr 380 (oxidized by Trp 360 Hc + with an intrinsic time constant of 10-20 ns; see results on W388F), is essentially kinetically switched off and thereby unlikely to play any relevant role in the wild-type protein. By contrast and in spite of the fact, that it reduces only $30% of the Trp 388 c radicals, the tyrosine Tyr 345 could noticeably increase the yield of metastable FADHc, because the lifetime of FADc À Tyr 345 c (1.1 ms) is almost 5Â longer than that of the FADc À Trp 388 c radical pair (225 ms), which gives more time for exogenous compounds to reduce the recombination partner of FADc À .
A network of water molecules mediates fast deprotonation of Trp 388 Hc + With 0.35 ns, deprotonation of the cation radical of the distal member of the Trp triad (Trp 388 Hc + ) turns out to be three orders of magnitude faster than the corresponding reaction in other studied members of the PCSf: 200 ns in a plant cryptochrome, 19 300 ns in EcCPDI, 14 and 2.5 ms in the X. laevis (6-4) photolyase. 12 The latter rates are comparable to the deprotonation rates in aqueous bulk solution of the transiently formed free TrpHc + radical either alone ($700 ns) 39 or as part of synthetic ruthenium complexes (130-400 ns). 40,41 Accordingly, the very fast deprotonation of the Trp 388 Hc + radical in MmCPDII implies the existence of a structurally dened proton acceptor.
The nearest and most plausible protein-derived candidate for the proton acceptor is Glu 387 , situated 3.6Å from the deprotonating N1 atom of Trp 388 . Mutation of Glu 387 had only a minor effect on the deprotonation kinetics, excluding it as the direct proton acceptor. A closer look at the crystal structure shows that N1 of the Trp 388 indole ring forms an H-bond to the water molecule 247 (d N1-O ¼ 3.4Å). This water is ideally