L-tyrosine-bound ThiH structure reveals C–C bond break differences within radical SAM aromatic amino acid lyases

2-iminoacetate synthase ThiH is a radical S-adenosyl-L-methionine (SAM) L-tyrosine lyase and catalyzes the L-tyrosine Cα–Cβ bond break to produce dehydroglycine and p-cresol while the radical SAM L-tryptophan lyase NosL cleaves the L-tryptophan Cα–C bond to produce 3-methylindole-2-carboxylic acid. It has been difficult to understand the features that condition one C–C bond break over the other one because the two enzymes display significant primary structure similarities and presumably similar substrate-binding modes. Here, we report the crystal structure of L-tyrosine bound ThiH from Thermosinus carboxydivorans revealing an unusual protonation state of L-tyrosine upon binding. Structural comparison of ThiH with NosL and computational studies of the respective reactions they catalyze show that substrate activation is eased by tunneling effect and that subtle structural changes between the two enzymes affect, in particular, the hydrogen-atom abstraction by the 5´-deoxyadenosyl radical species, driving the difference in reaction specificity.

Substrate modeling: The L-tryptophan amino group was modeled rather than an ammonium group due to its interaction with R323 and the carboxylate group was kept deprotonated as previously determined 5 .
Reactant scan: The substrate amino group, N(H1)(H2), orientation was investigated starting from the models we obtained for ThiH and NosL ( Supplementary Fig. 13) and by scanning the Cβ-Cα-N-H1 dihedral angle ( Supplementary Fig. 14). The QM part was reduced to 5´-dA•, L-tyrosine (or L-tryptophan), E183 (or E204) and R300 (or R323). Single-energy points were obtained with the M06-2X density functional 6 and the cc-pVTZ(-f) basis set. We also used the latter functional and basis set for all singleenergy point calculations below.
Hydrogen abstraction by 5´-dA•: To investigate the hydrogen atom transfer from the substrate (L-tyrosine or L-tryptophan) amino group to 5´-dA• ( Fig. 1 and Supplementary   Fig. 1), we i) scanned the hydrogen distance from the amino group nitrogen atom to the 5´-dA• C5´ atom and ii) searched for the transition state that was optimized.
Vibrational frequency calculations were performed to check that all minima (R and P) had positive frequencies and that the transition states had only one imaginary frequency. We further checked it was the relevant transition state for the reaction of interest by using the intrinsic reaction coordinate method (IRC). Starting points were the geometries of the 1 st point on the scan for ThiH (R in red in Supplementary Fig.   14c) and of the 5 th (productive) and 1 st (non-productive) points for NosL (R and R´, respectively, in green in Supplementary Fig. 14c). Reactants (RIRC) and products (PIRC) geometries issued from the IRC calculations were further optimized leading to Ropt for ThiH and Ropt and R´opt for NosL. For ThiH, RIRC and PIRC (before optimization) were used for tunneling evaluation (see section below). As expected, Ropt (ThiH) ≈ R (within the calculation precision) and Ropt (NosL) ≈ R (NosL) and the corresponding products had the correct geometries. Single-energy points of the results are presented in Fig. 4.
In R´opt (NosL) the L-tryptophan Cβ-Cα-N-H1 dihedral angle is at 23.8° on the flat potential region observed in the corresponding reactant scan ( Supplementary Fig.   14b,c). The scan exhibits a small barrier going from the 1 st to the 5 th point. We searched for the transition state between the 1 st and the 5 th point and found a transition state at 4.6 kcal.mol -1 from the 1 st point and at 0.6 kcal.mol -1 from the 5 th point, explaining why going down from TS´ we did not reach the global minimum. We have already observed this haziness around the first steps of the NosL reaction 5 . Indeed, we previously investigated NosL Cα-C bond break starting from a model of [L-Trp-NH•] already assuming the hydrogen abstraction by 5´-dA•. We started from the same X-ray model as the present study. At the time we had suggested a thermally-activated rotation of the amino-bearing arm in order to weaken the Cα-C bond and then described the reaction of the Cα-C bond break as well as the recombination of the resulting •COOto the substrate ring. In the present work, we have looked in further details at the very first step of the reaction and we observe alternatively that the reactant global minimum is not the one allowing the enzymatic reaction to proceed. Thus, the previous aminobearing arm rotation we had proposed now appears unnecessary, being here translated into the fact that the energy surfaces around the minima are flat in the case of NosL. This feature is actually most probably the one that differentiates NosL from ThiH.
Note: throughout the text, the dihedral called Φ refers to Cβ-Cα-N-H1 for the reactants [L-Tyr-NH2] and [L-Trp-NH2] models and Φ refers to Cβ-Cα-N-H for the transition states and the products (where H is the remaining hydrogen atom after hydrogen abstraction from -NH2 by 5´-dA•). Also the H that is transferred will be called Ht.  5 and main text) and (2) dehydroglycine is tightly held by two residues (arginine 300 and glutamate 158), we tried to find conformations leaving more space for p-cresyl• to separate from dehydroglycine. Thus, the frames we chose correspond to the largest distances between the substrate phenyl ring and tyrosine residues 76 and 181 (see Fig. 6b and main text). The minimal active site model consisted of L-Tyr, 5´-dA, Y76,   P78, Y80, L126, L128, E158, Y181, E183, S298, T299, R300, S318, S321, T323, F337, I339 and the two water molecules interacting with the carboxylate group of L-Tyr  Table 2) and that the associated transition states' frequencies were also high, we decided to quantify the impact of hydrogen tunneling on the energetic gain it provides for the hydrogen atom abstraction by 5´-dA• at the amino-nitrogen position of L-tyrosine or L-tryptophan, for ThiH and NosL, respectively.
There are quite a few different non-variational "rigid barrier" descriptions of the tunnel effect all based on the transition state [11][12][13][14][15][16][17] . Among them, Eckart 16 proposed an asymmetric potential barrier, later further explicated by others 17,18 , which takes into account the energy difference between reactants and products. This model has been successfully used to estimate tunneling correction 19,20 even though it tends to overestimate the values of the tunneling factors below 300 K 21 . This class of models rely on the use of a well-defined -hence "rigid" or "non-variational" -transition state.
Moreover, it is assumed that, in the neighborhood of this transition state, the motion along the direction of the one-dimensional reaction path can be separated from all other motions of the interacting species, thus neglecting contribution from the heavyatom environment. As a consequence, a potential barrier can be defined along this path from reactants to products via a single hydrogen transfer reaction coordinate which is orthogonal to all other modes of motions of the interacting species. A more complete picture of hydrogen tunneling would therefore include contributions of motions from the heavy-atom environment and a number of quantum-mechanical theories have been proposed, calculating Arrhenius curves from first principles, including tunneling. These theories (Small-Curvature Tunneling, Instanton, etc…) start with an ab initio calculation of the reaction surface (energies, gradients and hessians) before using either quantum or statistical rate theories in order to calculate appropriate rate constants 15,[22][23][24][25][26][27][28][29][30][31] .
Such involved calculations are however beyond the scope of our work. In effect, radical SAM enzyme structures all exhibit a tight proximity of 5´-dA (a good mimic of the highly reactive 5´-dA• radical) and the targeted substrate hydrogen to be abstracted 32 . This, along with the fact that even at high temperature MD simulations of L-tyrosine bound ThiH do not reveal drastic conformational changes around 5´-dA, justifies our choice of an analytic estimation of hydrogen tunneling effect using the Eckart approach (see below). Moreover, we will show that we already achieve, by Hydrogen tunneling is expected to lower the barrier from reactants R = amino group of substrate and 5´-dA• (NH2 + •C5´H2) to products P (•NH + C5´H3) via the transition state TS (HN…Ht…C5´H, where Ht is the tunneling hydrogen).
We use the following potential barrier function 16,17 : The

18.
We can now compute tunnel corrections for NosL and ThiH. From reference 17, the discriminating parameter for the Eckart model is: from which the following quantities are derived: (eq. S11) Finally, the reaction rate coefficient k(T) is given by: In turn, the correcting factor Q can be translated into an equivalent lowering of the barrier (from eq. S12): where R is the molar gas constant (1.9872 10 -3 kcal.K -1 .mol -1 ) and k is in s -1 .
Numerically:  Streptomyces actuosus, respectively) but also represented as a function of temperature in Supplementary Fig. 19 to take into account the optimal bacterial growth.
To check the above procedure, more specifically the internal consistency between the two energetic Eckart parameters A, B on the one hand and the internal length scale l (computed from the frequency ν: Supplementary to compute IRC x coordinates (cf. Eq. S1) centered on TS (set at x(IRC) = 0.0 Å). From R to TS, we selected p(rel.) distances, and from TS to P, we selected q(rel.) distances ('rel.' refers to a value calculated relative to that of TS). Table 5 allow us to report the IRC points on the ThiH Eckart potential curve (cf. Supplementary Fig. 18-20). It can be seen from Supplementary Fig.   20 that IRC points are close enough to the Eckart potential curve derived from QM/MM data and that IRC internal coordinates x(IRC) as defined in Supplementary Table 5 Fig. 4 and Supplementary Fig. 16). These values are those injected in the Eckart model and used in Supplementary Fig.  18, 20-21 to model the barrier potentials of TcThiH, SaNosL productive and nonproductive reactants (R), transition states (TS) and products (P). Energy values in black and parentheses are the values relative to the corresponding reactant global minima of ThiH and NosL ( Supplementary Fig. 14c). We report (in red) values of the frequency ν (cm -1 ) computed for each TS. We also report structural parameters involved in hydrogen tunneling evaluation, namely N-C5´, NHt ('t' stands for transferred) and HtC5´ distances, as well as ϴ (N-Ht-C5´) and Φ (Cβ-Cα-N-H (H1 for Rs')). Angles and dihedral angles are given in degrees. Supplementary * Values corresponding to the productive reactant (see main text and Fig. 4b). **k(T) values, calculated at the optimal bacterial growth temperature, can be compared to those typically observed for first order enzymatic systems 40 . Table 5 | N-Ht, Ht-C5´ and N-C5´ distances as well as N-Ht-C5´ angles (cf. Θ Scheme below) measured for ThiH for (QM/MM) IRC states: RIRC, TS and PIRC states as well as for two intermediate IRC states intR and intP. These five points follow the IRC path from TS toward both R on the one side and P on the other side. Are also computed p values relative to that of TS (p(rel.)), q values relative to that of TS (q(rel.)) and derived x(IRC) values selected from p(rel.) and q(rel.