Planarizing cytosine: The S₁ state structure, vibrations, and nonradiative dynamics of jet-cooled 5,6-trimethylenecytosine

A number of femtosecond (fs) pump-probe studies of cytosine (Cyt), 1-methylcytosine (1-MCyt), and 5-fluorocytosine (5-FCyt) in the gas-phase have yielded ultrashort excited-state S₁ lifetimes that range from $τ=0.2$ to 3.2 ps.^1–5 However, more recent measurements of supersonically jet-cooled amino-keto Cyt, 5-methylcytosine (5-MCyt), and 5-FCyt based on Lorentzian lifetime broadening measurements at their respective $S0→S1$ electronic origins yielded lower limits on the S₁ lifetimes of 45 ps for Cyt and 75 ps for 5-FCyt.^6–8 More recent direct lifetime measurements of Cyt, 1-MCyt, and 1-ethylcytosine (1-ECyt) using the picosecond (ps) excitation/ionization delay technique revealed even longer values between $τ=450–900$ ps, which decrease rapidly with increasing vibrational excess energy (E_exc).⁹ 

Computational studies of the excited-state dynamics and nonradiative decay of amino-keto Cyt^10–20 have located three different conical intersections (CIs) between the ground and lowest excited ¹$ππ*$ states. The lowest of these, which dominates the $S1⇝S0$ nonradiative decay, is called (Eth)_X since the intersection structure is similar to the CI structure of ethylene. This CI is characterized by a puckering of the C⁶ atom and a twist around the C⁵–C⁶ bond, with a H–C⁵–C⁶–H torsional angle of $∼$120°.^{10–13,15,16,19,21,22} The next higher CI involves an N³ out-of-plane bending and a large out-of-plane amino deformation and is called (OP)_X.^10,16,19,22 The third CI, called $(nO,π*)X$⁠, has a semi-planar structure with sp³ hybridization of the C⁶ atom, shortening of the C²–N³ bond, and stretching of the C²-O bond relative to the ground state minimum.^{10–16,19,21,22}

Zgierski et al. have shown that covalently clamping the C⁵–C⁶ bond of Cyt with a trimethylene bridge in 5,6-trimethylenecytosine (TMCyt) increases the S₁ state fluorescence lifetime and quantum yield in room-temperature aqueous solution by $∼1000$ times relative to Cyt,^23–25 to $τ=1.2$ ns and $Φfl∼10$%.²⁶ Their configuration interaction singles (CIS) and second-order approximate coupled cluster (CC2) excited-state calculations predicted that this clamping shifts the (Eth)_X conical intersection of cytosine to $∼1500$ cm⁻¹ above the S₁ minimum, making this CI energetically less accessible.²⁶ The trimethylene bridge in TMCyt hardly affects the $π$-electron framework of Cyt, so the $S0→S1$ absorption band shifts from $λmax=267$ nm for Cyt to 280 nm in TMCyt.²⁶ In the clamped cytosine derivative pyrrolocytosine (PC), the C⁴-amino group (see Fig. 1) and the C⁵ atom are covalently connected, resulting in a pyrrole ring fused to the Cyt chromophore.²⁷ This extension of the $π$-electron framework significantly shifts the $S0→S1$ excitation maximum to $λmax=330–345$ nm, or about 70–80 nm to the red, compared to Cyt.^28,29 For PC, Thompson and co-workers have measured a lifetime of $τ=2.9$ ns and a quantum yield of $Φfl∼0.038$ in pH 7 phosphate buffer.²⁹ 

Intrigued by these observations, we have measured and analyzed the $S0→S1$ vibronic spectrum of supersonic jet-cooled TMCyt using two-color resonant two-photon ionization (2C-R2PI), UV/UV holeburning, and depletion spectroscopies. We also measured the S₁ state lifetime and triplet-state formation kinetics as a function of E_exc, using the nanosecond excitation/ionization delay technique, and report S₁ state nonradiative rate constants for internal conversion and intersystem crossing. In addition to the amino-keto tautomer 1 of TMCyt, we have also observed an intense R2PI spectrum that we assign to the $S0→S1$ transitions of the hydroxy-enol tautomers 2a/2b, see Fig. 1 for the tautomer structures. The measurements are accompanied by calculations of the lowest excited singlet (¹$ππ*$⁠) and triplet (³$ππ*$⁠) states of TMCyt using time-dependent (TD) density functional theory (DFT) (TD-CAMB3LYP), spin-component scaled CC2 (SCS-CC2), complete-active-state self-consistent field (CASSCF), and multi-state second-order perturbation-theory (MS-CASPT2) methods.

TMCyt was synthesized in three steps from adiponitrile according to Ref. 30 and characterized by ¹H-NMR spectroscopy, ¹³C-NMR spectroscopy, and electrospray ionization (ESI) mass spectrometry as described in the supplementary material. The experimental setup for two-color resonant two-photon ionization (2C-R2PI) measurements has been previously described.^31,32 Neon carrier gas at $∼$1.8 bar backing pressure was passed through a 20 Hz pulsed nozzle (0.4 mm diameter) heated to 230 °C that contains the TMCyt sample. 2C-R2PI spectra were measured by crossing the skimmed supersonic jet with the unfocused UV excitation and ionization laser beams in the source of a linear time-of-flight mass spectrometer (TOF-MS). Excitation was performed with 0.4–0.6 mJ UV pulses from a frequency-doubled Radiant Dyes NarrowScan D-R dye laser with a bandwidth of 0.05 cm⁻¹ (1500 MHz) or with an Ekspla NT342B optical parametric oscillator (UV-OPO) with a bandwidth of 4–6 cm⁻¹. The frequency scale was calibrated by measuring the fundamental frequency with the WS6 wavemeter.

Ionization light of $∼100$ $μ$J/pulse at 225 or 245 nm was produced by an Ekspla NT342B (UV-OPO). The measurements were typically done with 10–30 mV ion signal at the most intense vibronic band. A second frequency-doubled dye laser (⁠$∼400$ $μ$J/pulse) was used as the UV-depletion laser for UV/UV holeburning and depletion experiments. Excitation and ionization were performed as for the R2PI experiments. The depletion laser was fired 300 ns before the excitation/ionization laser pulses.

A uniform theoretical treatment of the ground- and excited-state potential energy surfaces of TMCyt is difficult, and we have combined several methods following a similar approach to our recent work on 1-MCyt.³³ The electronic ground state of all 14 tautomers and rotamers of TMCyt was first optimized using density functional theory (B3LYP) with the TZVP basis set. The ground state structures of the six most stable tautomers are shown in Fig. 1; these were re-optimized at the correlated level, using the second-order Møller-Plesset (MP2) method in the resolution-of-identity (RI) approximation, the SCS-MP2 method, and the CC2 method in the RI approximation, using the aug-cc-pVTZ basis set.

The adiabatic and vertical transition energies were calculated at the SCS-CC2 level of theory with the aug-cc-pVDZ basis set. Normal-mode calculations were performed for all geometry-optimized structures to assure that they correspond to true potential energy surface minima. The transition energies were also calculated at the MS-CASPT2 level of theory. For the $ππ*$ and ¹$nOπ*$ states, we used TD-CAM-B3LYP/6-311G** optimized geometries, whose MS-CASPT2 energy is lower than that of their CASSCF and SCS-CC2 analogues. For the optimization of (OP)_Min, we used the CASSCF(12,12)/6-311G** geometry because the other methods failed to converge to a minimum for that state.

For the calculation of the decay paths and conical intersections, we follow the MS-CASPT2//CASSCF approach, where the paths and structures are optimized at the CASSCF level of theory and the energy profiles along the paths are recalculated at the MS-CASPT2 level to include dynamic correlation energy (see our previous study of 1-MCyt³³). For the CASSCF and MS-CASPT2 calculations, we used, respectively, the 6-311G** basis and the ANO-L basis set contracted to 4s3p2d for C, N, and O and 3s2p1d for H. For the CASPT2 calculations, we used an imaginary level shift³⁴ of 0.1 a.u. and the standard ionization potential–electron affinity correction³⁵ of 0.25 a.u.

To calculate the reaction path to (OP)_X, we optimized the transition structure (TS) on S₁ and obtained the path by combining the intrinsic reaction coordinate³⁶ and initial relaxation direction³⁷ techniques. The calculated barrier includes the vibrational zero-point energy (ZPE) correction, which amounts to −475 cm⁻¹, based on CASSCF frequencies at (¹$ππ*$⁠)_Min (with 3N − 6 vibrational modes) and at the corresponding TS (including 3N − 7 modes). The CI was optimized using the recently developed double Newton-Raphson algorithm.³⁸ The active space of the CASSCF and MS-CASPT2 calculations was specifically tailored for each path; for details, see the supplementary material. We use (10,10) and (12,12) active spaces for the ethylene- and OP-type paths, respectively. With this approach, the MS-CASPT2 S₁/S₀ energy gaps at the CI structures were 1973 and 2265 cm⁻¹ (0.24 and 0.28 eV) at (Eth)_X and (OP)_X, respectively. The path to (Eth)_X, which has a sloped topology and does not involve a TS, was approximated with a linear interpolation in internal coordinates.

Density functional theory and CC2 calculations were performed using Turbomole 6.4.^39,40 The CASSCF optimizations were performed with a modified version of Gaussian09⁴¹ and the MS-CASPT2 calculations with Molcas 7.8.^42,43 Vibronic band simulations were done with the PGOPHER program.⁴⁴ As inputs, we used the SCS-CC2 calculated S₀ ground and S₁ excited state geometries and the corresponding normal-mode l matrices, employing conformer 1a. Additional diagonal anharmonicity constants⁴⁴ were included for some modes. The vibronic band intensities are based on full multidimensional Franck-Condon factors, including both mode displacements and mixing between modes (Dushinsky effect).⁴⁴ The vibronic simulations for conformer 1b are very similar to those for 1a.

Figure 1 shows the six most stable calculated tautomers and rotamers of TMCyt, and Table I summarizes their relative energies calculated at different levels of theory. All the correlated wave function methods predict the trans-amino-enol 2b tautomer to be the most stable one, with the corresponding cis-rotamer 2a $∼0.6$ kcal/mol higher. The amino-keto N1H tautomer 1 that is experimentally investigated below exists in conformer 1a, where the amino group and trimethylene ring are out-of-plane in the same direction, denoted Up-up (or Down-down), where the first (capitalized) orientation refers to the NH₂ group. In conformer 1b, the NH₂ group and trimethylene ring are arranged in opposite directions (Up-down or Down-up). The 1a and 1b forms are close in energy with 1b calculated to lie 3–4 cm⁻¹ above 1a. In the gas phase, both 1a and 1b are less stable than amino-enol conformers by 1.08 kcal/mol (CC2) or 1.54 kcal/mol (SCS-MP2). The B3LYP density functional method predicts the amino-keto N1H tautomer to be the most stable tautomer; however, it is known that this method predicts the order of the cytosine tautomers incorrectly.^6,45

The other TMCyt tautomers 2b, 2a, and 4 also exist as pairs of conformers analogous to 1a/1b, but only one form was calculated since the energy difference is expected to be very small. All the imino-enol forms lie $>13$ kcal/mol above the most stable tautomer 2b at the B3LYP/TZVP level; hence, we do not consider them any further.

Table II summarizes the calculated adiabatic and vertical transition energies of the 1a, 1b, 2a, 2b, and 3a conformers predicted by the SCS-CC2 method, together with the MS-CASPT2 transitions for 1a. Both methods are in good agreement, which validates our computational approach. They predict that the S₁ state minimum has ¹$ππ*$ character. The calculated adiabatic transition of 1a is 31 534 cm⁻¹ (SCS-CC2) and 31 831 cm⁻¹ (MS-CASPT2), in excellent agreement with the experimental value of 31 510 cm⁻¹. The vertical excitation energy of the ¹$ππ*$ state with the SCS-CC2 and MS-CASPT2 methods is 36 610 and 36 500 cm⁻¹, respectively. There are two low-lying ¹$nπ*$ states which arise from excitations out of the oxygen and nitrogen lone pairs, whose vertical excitation energies are 7000-10 000 cm⁻¹ higher. The structure of the ¹$nOπ*$ state was optimized at the TD-CAMB3LYP level. It is a minimum on S₂ with adiabatic energy of 37 597 cm⁻¹ (MS-CASPT2 single point). The structure optimization of the ¹$nOπ*$ state did not converge with SCS-CC2 because it reached a region of S₂/S₁ degeneracy, which is consistent with the small S₂/S₁ energy gap found at the ¹$nOπ*$ minimum at the MS-CASPT2 level. Optimization of the ¹$nNπ*$ state at the CASSCF level leads to (OP)_Min, with an adiabatic energy of 33 017 cm⁻¹. The electronic configuration at this structure is analogous to that described in our previous work on 1-MCyt.³³ 

The adiabatic transition energy of conformer 1a is calculated to lie slightly above that of 1b, differing by 51 cm⁻¹ at the SCS-CC2 level. With this method, the $S0→S1$ transitions of the major tautomers 2b and 2a are calculated to be ¹$ππ*$ and to lie at $∼35000$ and $∼34500$ cm⁻¹, respectively, or about 3500 cm⁻¹ further to blue than the transitions of the 1a/1b conformers. The lowest-energy electronic transition of the imino-keto tautomers 3a and 3b is predicted at 41 400 cm⁻¹ and 40 970 cm⁻¹, respectively. This is above the experimental spectral range covered in this work. On the other hand, the lowest ¹$ππ*$ transition of the 4 (N3H) tautomer is predicted to lie very close to that of the 1 (N1H) tautomer. However, tautomer 4 is calculated to be 5.5–5.8 kcal/mol less stable than tautomer 1; hence, we do not expect this tautomer to be observable in the supersonic jet.

In the SCS-CC2 S₀ optimized structure of 1a, the pyrimidinone framework is C_s symmetric, and the amino group and the trimethylene ring are bent slightly out of the ring plane. In the ¹$ππ*$ excited state, the SCS-CC2 and TD-CAMB3LYP methods predict (i) a stronger pyramidalization of the amino group, (ii) an in-plane deformation of the pyrimidinone framework, and (iii) an out-of-plane bend at the C⁶ atom (see Fig. 1 for the atom numbering). Figure 2 shows the SCS-CC2/aug-cc-pVDZ calculated geometries and geometry changes of TMCyt for both amino-keto N1H conformers. The TD-CAM-B3LYP optimized structure has similar out-of-plane deformations, see Fig. S1 in the supplementary material. This is in line with previous results for 1-MCyt,³³ for which both methods predict a substantial deplanarization at the ¹$ππ*$ state minimum.

As shown in Table I, three correlated quantum-chemical methods (MP2, CC2, and SCS-MP2) predict the energy difference between conformers 1a and 1b to be very small (1–10 cm⁻¹). At the typical T_vib = 5–7 K in our supersonic jet expansions, the relative population of 1a and 1b should thus be within a factor of 2–3. Given the computed adiabatic transition frequencies of 31 534 cm⁻¹ for 1a and 31 483 cm⁻¹ for 1b (see Table II), we should observe two spectra that are mutually shifted by about 50 cm⁻¹. However, the R2PI and UV/UV holeburning spectra discussed below show only a single ground-state species. The reason for this is the large-amplitude amino-inversion of TMCyt, which interconverts the conformers 1a and 1b.

In the S₀ state, the SCS-CC2/aug-cc-pVDZ calculated barrier height between 1a and 1b is $∼30$ cm⁻¹. We calculated the one-dimensional (1D) inversion potential at the same level by incrementing the H–N–C⁴–N³ and H–N–C⁴–C⁵ angles $𝜃inv$ from $𝜃inv=0°$ by $5°$ in the positive and negative directions, relaxing all other internal coordinates at every point. The resulting 1D potential is shown in Fig. 3. The inversion eigenfunctions in this potential were calculated by numerically solving the 1D vibrational Schrödinger equation. The reduced mass $μred,𝜃$ was determined by calculating the SCS-CC2/aug-cc-pVDZ amino-inversion imaginary normal-mode frequency at $𝜃inv=0°$⁠; the central five points of the inversion potential were taken to represent the harmonic potential at this angle, and $μred,𝜃$ for the 1D calculation was fixed such that the calculated normal-mode and 1D frequencies in this harmonic potential were the same. Figure 3 shows that the lowest-energy $vinv$ = 0 level lies $∼130$ cm⁻¹ above the barrier. Its wave function is delocalized over both the 1a and 1b geometries with its maximum near planarity (⁠$𝜃inv=0°$⁠). The fact that the vibrational ground state of TMCyt is quasiplanar (delocalized over both 1a and 1b) explains why the UV/UV holeburning spectra, discussed in Sec. III B, reflect the presence of a single ground-state species only. The second amino-inversion level $vinv$ = 1 lies 380 cm⁻¹ higher. It will be collisionally cooled out in the supersonic expansion and will not be considered further.

In the S₀ state, the planar (C_s symmetric) structure of TMCyt is an index-2 stationary point. Normal-mode analysis at this point yields imaginary frequencies for both the NH₂ inversion and trimethylene-ring out-of-plane vibrations. The S₀-state barrier to planarity is 307 cm⁻¹ at the SCS-CC2 level. In the ¹$ππ*$ excited state, the barrier to planarity is much higher, 1297 cm⁻¹. Four imaginary frequencies are obtained at the C_s stationary point.

Figure 4(a) shows a two-color R2PI overview spectrum of jet-cooled TMCyt in the 31 300–38 500 cm⁻¹ range, recorded with the UV-OPO. The $000$ transition of the amino-keto 1a/1b tautomer is observed at 31 510 cm⁻¹. At the $∼5$ cm⁻¹ resolution of the UV-OPO, vibronic bands are resolved up to +2100 cm⁻¹ above the $000$ band. The R2PI signal extends up 4000 cm⁻¹ above, but due to the high density of vibronic excitations and the moderate $∼5$ cm⁻¹ resolution, the band structure is washed out in this range. In Fig. 4(b), we show a PGOPHER vibronic band simulation based on SCS-CC2 structures and normal modes of TMCyt. They are seen to be in qualitative agreement with the R2PI spectrum and will be discussed in more detail in Sec. III F.

Starting at 35 930 cm⁻¹ or 4420 cm⁻¹ above the amino-keto $000$ band in Fig. 4(a), further intense narrow-band absorption features are observed. Based on the SCS-CC2 calculations of Sec. III A 2, we assign these to the amino-enol tautomers 2b predicted at 34 508 cm⁻¹ and 2a predicted at 35 008 cm⁻¹, see also Table II.

Figure 5(a) shows a higher-resolution 2C-R2PI spectrum of the amino-keto tautomers over the lowest 1500 cm⁻¹ range (31 400–32 900 cm⁻¹) using a narrow-band frequency-doubled dye laser for excitation. Because of its 0.05 cm⁻¹ bandwidth, the vibronic bands are much better resolved in this spectrum. Detailed vibronic assignments are given in Sec. III C. A high-resolution UV/UV holeburning spectrum is shown in Fig. 5(b) and was recorded with the burn laser at the intense band at $000+59$ cm⁻¹, marked with an asterisk in Fig. 5(a). It reproduces the 2C-R2PI spectrum in Fig. 5(a) in great detail. From this we conclude that all the observed vibronic bands originate from the ground-state level that gives rise to the transition at $000+59$ cm⁻¹. Figure 5(c) shows the corresponding UV/UV depletion spectrum in which the holeburning laser is scanned with the detection laser fixed at the intense $000+59$ cm⁻¹ band. The UV/UV depletion spectrum also reproduces the R2PI spectrum, although the signal/noise ratio is lower than that in the UV holeburning spectrum. At 900 cm⁻¹ above the electronic origin, the widths of the vibronic bands begin to increase, which indicates the onset of rapid non-radiative processes, see Sec. III F. Although no further bands can be observed in the depletion spectrum above +1000 cm⁻¹, the signal remains slightly below the baseline, indicating a constant depletion of the ion signal.

We first attempted to assign the vibronic bands in the R2PI spectrum of TMCyt in Fig. 5(a) based on the SCS-CC2, CC2, and TD-B3LYP harmonic frequencies of the ¹$ππ*$ state given in Table III. The lowest-frequency in-plane vibration is predicted to be $ν1′=254$ cm⁻¹ (SCS-CC2), $ν1′=248$ cm⁻¹ (CC2), or $ν1′=262$ cm⁻¹ (TD-B3LYP); hence, the vibronic bands below $∼250$ cm⁻¹ must arise from out-of-plane vibrations. Experimentally, the two lowest-frequency bands at 38 cm⁻¹ and 59 cm⁻¹ cannot belong to the same progression, so we assign these as fundamentals of the lowest-frequency out-of-plane vibrations $ν1′$ and $ν2′$ (that is, as $101$ and $201$⁠). Table III shows that the lowest two frequencies calculated with the SCS-CC2, CC2, and TD-B3LYP methods are two to three times larger. Previous experience with SCS-CC2, CC2, and TD-B3LYP excited-state calculations of cytosine derivatives and pyrimidinones has shown that while the in-plane S₁ state vibrational frequencies are well reproduced, the calculated out-of-plane vibrational frequencies are often 2-3 times higher than that observed experimentally.^6–8,31,33

For the PGOPHER⁴⁴ vibronic band simulations (see Sec. II B), we therefore decreased the out-of-plane frequencies to the experimental values. Figures 6(a)–6(c) show the simulated vibronic bands in red for the 0–420, 420–870, and 870–1320 cm⁻¹ sections of the spectrum and compare these to the high-resolution 2C-R2PI spectrum in black. We first fitted the S₁ state in-plane vibrational frequencies. The SCS-CC2 calculation predicts the lowest four in-plane fundamentals $ν6′$⁠, $ν11′$⁠, $ν12′$⁠, and $ν15′$ at 254, 461, 472, and 618 cm⁻¹, respectively, see Table III. These normal-mode eigenvectors are shown in Fig. 7. The $ν11′$ and $ν12′$ normal-modes correspond to the $ν6a′$ and $ν6b′$ in-plane vibrations that are characteristic of the $S0→S1$ spectra of benzene and its derivatives. We therefore assigned the bands at 449 cm⁻¹ and 471 cm⁻¹ as $1201$ and $1101$⁠, respectively, see Fig. 6(b); the order of these two vibrations was interchanged to obtain a better fit with the experimental R2PI spectrum. The $601$ transition was fitted to the band at 257 cm⁻¹; its intensity is rather small and it does not contribute further to the spectrum. The band at 615 cm⁻¹ was assigned to the $1501$ fundamental. The band at 944 cm⁻¹ is assigned as the in-plane fundamental $ν23′$⁠, as the overtone $1102$ had no intensity, see Fig. 6(c).

We then fitted the out-of-plane vibrations, see Fig. 6(a). The weak band at 38 cm⁻¹ is assigned as the $ν1′$ fundamental. Since the $ν2′$ and $ν4′$ vibrations involve structural changes of the trimethylene ring, see Table III, and hardly appear in the simulation, the intense 59 cm⁻¹ band is assigned as the $301$ “butterfly” vibrational fundamental. The $302$ overtone was fitted to the band at 126 cm⁻¹. The fundamentals of $ν5′$ and $ν7′$ were fitted to the bands at 93 cm⁻¹ and 221 cm⁻¹, respectively. The out-of-plane normal-mode eigenvectors $ν1′$⁠, $ν3′$⁠, $ν5′$⁠, and $ν7′$ are shown in Fig. 8. Note that the SCS-CC2, CC2, and TD-B3LYP harmonic frequencies in Table III differ from the fitted frequencies (Table IV) by a factor of 2–3, indicating that the S₁ state potential-energy surface is much flatter and more anharmonic along these coordinates than that predicted by the excited-state calculations.