Performance of revised STO(1M)-3G basis set for prediction of 5-fluorocytosine chemical shifts

Nuclear shieldings and chemical shifts of 5‐fluorocytosine (5FC) were predicted in the gas phase and DMSO solution modeled by polarizable continuum model using B3LYP density functional and revised STO(1M)‐3G basis set. For comparison, eight arbitrary selected basis sets including STO‐3G and medium‐size Pople‐type and larger dedicated Jensen‐type ones were applied. The former basis sets were significantly smaller, but the calculated structural parameters, harmonic vibrational frequencies, were very accurate and close to those obtained with larger, polarization‐consistent ones. The predicted 13C and 1H chemical shieldings of 5FC and cytosine, selected as parent molecule, were acceptable (root mean square for 13C chemical shifts in DMSO of about 5 ppm and less) though less accurate than those calculated with large basis sets, dedicated for prediction of nuclear magnetic resonance parameters.

Keywords: 5FC; 5‐fluorocytosine; GIAO NMR; modified basis sets; STO(1M)‐3G

Nuclear shieldings and chemical shifts of 5‐fluorocytosine were predicted in the gas phase and DMSO solution modeled by polarizable continuum model using B3LYP density functional and revised STO(1M)‐3G basis set. For comparison, eight arbitrary selected basis sets including medium‐size Pople‐type and larger dedicated Jensen‐type ones were applied.

GRAPH

Prediction of nuclear shieldings and chemical shifts has been routinely used to support experimental works.[[1]] It is well known that the predicted nuclear magnetic resonance (NMR) parameters are very sensitive to the level of theory and basis set flexibility and completeness.[[3], [5]] However, molecular modeling of large biomolecular structures and ordered carbon nanostructures has been a big challenge for the theory due to the system size leading to nonlinear increase of calculation time and lowering their predictive accuracy.[[7]] In 2012, Voronkov et al.[15] modified small STO‐3G basis sets for a very efficient ones with relatively small number of basis functions, allowing an efficient calculation of electronic structure and magnetic properties of the studied systems.

In our previous study, we demonstrated the excellent performance of "Leszczynski basis sets" in predicting single‐walled carbon nanotube and fullerene chemical shifts.[14] Their high accuracy and efficiency in predicting NMR parameters were also tested on several smaller molecules.[[1], [16]] Just recently, these basis sets were remodeled and improved by Kapusta and coworkers.[17] The currently available ones, called STO(1M)‐3G, have been improved and reported at the end of 2018.[17]

In our current study, we decided to verify the performance of recently redesigned STO(1M)‐3G basis sets for prediction of nuclear shieldings and chemical shifts of an important antifungal drug and anticancer prodrug—5‐fluorocytosine (5FC).[[18]]

5FC belongs to the group of the fluorinated pyrimidine analogues. It was used for the first time in 1968 in the treatment of systemic fungal infections.[[18]] Nowadays, it is widely used in combination with amphotericin B in the treatment of serious infections caused by Candida (septicemia and endocarditis) and Cryptococcus (meningitis and pulmonary infections).[24] The major mechanism of 5FC action is based on entering the fungal cell via cytosine permease where it is metabolized to 5‐fluorouracil, which is subsequently converted to 5‐fluorodeoxyuridinemonophosphate, which inhibits fungal DNA synthesis and 5‐fluorouridinetriphosphate, which disturbs the biosynthesis of fungal RNA.[25] Recently, 5FC is used also as a prodrug for delivery of 5‐fluorouracil to the tumor cells.[26]

Several heteroatoms, carbonyl and amide groups, as well as a double bond form the planar heterocyclic molecule of 5FC. Thus, the structure of 5FC[27] is quite challenging for theoretical prediction of accurate nuclear shieldings and NMR chemical shift (for atom numbering and detailed structure see Figure ). Thus, the justification for consideration of a single molecule for testing a new basis set is related to its peculiar structure, difficult to accurately predict using density functional theory (DFT) and well‐performing basis sets.

For efficient theoretical modeling, DFT calculations are often used due to their acceptable accuracy to speed ratio.[[28]] Among the very high number of available functionals, the B3LYP[[30]] hybrid functional is widely used for reliable prediction of structure, energy, and NMR parameters of small‐ and large‐size molecules. Because our experimental proton and carbon spectra of 5FC are measured in DMSO‐d6, we will model both gas‐phase NMR data and in DMSO solution, using a simple polarizable continuum model (PCM)[33] approach. In addition, there is a direct impact of geometry quality on the predicted NMR parameters. Thus, in the first step, we will check the performance of the used basis set for 5FC structure and harmonic frequencies prediction. The performance of STO(1M)‐3G basis sets will be critically compared with experiment and basis sets used for NMR calculations.

Routine room temperature 1H and 13C NMR spectra of a few mg 5FC (Aldrich) were measured in 0.6 ml of DMSO‐d6 using 400‐MHz Bruker Ultrashield spectrometer and dedicated acquisition and processing topspin software. Both 1D and 2D carbon‐proton heteronuclear multiple bond correlation spectra were collected using tetramethylsilane as reference.

Gaussian 2016[34] was used for all DFT calculations with B3LYP hybrid functional. The subsequent frequency calculations produced all real wave numbers. The vibration second‐order perturbation theory[[35]] method was used to predict anharmonic frequencies. An attempt to include a Grimme D3 dispersion correction[[37]] in a form of B3LYPD3 functional produced many unrealistic anharmonic frequencies (significantly higher than the corresponding harmonic modes). To get the most accurate NMR parameters, both structure optimization and gauge‐including atomic orbital[[3], [40]] (GIAO) calculations were calculated with the same basis set. The results of calculations with recently available STO(1M)‐3G basis set[17] were compared with those obtained using initial STO‐3G and medium‐size and large Pople‐type 6‐311++G** and 6‐311++G(3df,2pd) ones.[42] In the subsequent text, these basis sets will be called small and medium. For more accurate calculations, Jensen's polarization‐consistent basis sets aug‐pc‐2 was applied.[[43]] The most accurate calculations were conducted using dedicated Jensen's basis sets for nuclear shieldings and indirect spin–spin coupling constants: aug‐pcS‐2, aug‐pcS‐3, aug‐pcJ‐2, and aug‐pcJ‐3.[[43], [47]] These basis sets will be abbreviated as apcS2, apcS3, apcJ2, and apcJ3, respectively. Their accuracy in the prediction of nuclear shieldings and spin–spin couplings was verified in several reports.[[7], [47], [49]] The accuracy of carbon chemical shift prediction was expressed as deviation of theoretically predicted parameters from available experimental values in a form of root mean square (RMS) deviation. However, in case of proton chemical shifts, two bands (N1H and NH2) are involved in strong H‐bonding with DMSO solvent, and their positions are shifted by about 3–5 ppm. Becasue in our simplified PCM we cannot implicitly account for this effect, only deviations for nonexchangeable H6 signal, as measure of accuracy for proton chemical shift, will be presented. No correction for zero‐point vibrational energy[[50]] was applied. For brevity, the detailed results of calculations are gathered in the Supporting Information.

There is no available gas‐phase structure of 5FC. Therefore, the quality of our structure prediction will be compared with that of the crystal structure of anhydrous 5FC.[27] Obviously, the presence of hydrogen bonding and crystal packing forces will somehow distort its molecular structure, and we should not expect a very good agreement of our predicted structure with experimental data. In addition, the reported experimental positions of hydrogen atoms are about 0.2 Å shorter than those of typical C–H and N–H bonds. In Table  are gathered experimental bond lengths for 5FC and the corresponding deviations of B3LYP predicted parameters using STO‐3G, STO(1M)‐3G, and seven other basis sets. For brevity, all calculated bond lengths and dipole moments of free 5FC and in DMSO are included in Table S1. It is apparent from Table  that all basis sets are suitable for fairly good prediction of 5FC geometry with RMS of about 0.02 Å in the gas phase. In general, the improvement of basis set size and flexibility is accompanied by a small decrease of RMS (from about 0.022 to 0.020 Å in the gas phase). Interestingly, the modified Leszczynski's basis set,[17] in both the gas phase and DMSO, produces the best agreement with experiment (RMS of 0.016 and 0.009 Å). On the other hand, significantly larger RMS is produced by initial STO‐3G basis set (0.068 Å). It is obvious that C═O bond is elongated by about 0.03 Å in the crystal state due to the presence of strong H‐bonding, and our theoretical results for an isolated molecule of 5FC resemble gas‐phase data for compounds with a carbonyl group. In addition, the predicted N1C2 bond for a free molecule is overestimated by about 0.04 Å. This could be a result of H‐bonding of H‐N1 with the neighboring molecule leading to shortening of N1C2 bond in the crystal state. Obviously, our gas‐phase calculations cannot account for this. Thus, as expected, the geometry of 5FC is very sensitive to the surrounding, and in the presence of DMSO, introduced with a very simple PCM, it closely resembles the experimental X‐ray structure.[27] In fact, the RMS deviations of bond lengths between nonhydrogen atoms in DMSO are about two times smaller than those in the gas phase (about 0.01 vs. 0.02 Å). Thus, it is very pleasing to observe a very good performance of STO(1M)‐3G basis set in prediction of interatomic separations in 5FC, in both the gas phase and DMSO solution. The data in Table  also point out an additional and unexpected benefit from designing STO(1M)‐3G as relatively small and inexpensive basis sets for prediction of NMR shieldings and chemical shift. Thus, they perform very well in prediction of 5FC geometry. We could expect similar performance of this modified basis set for other nucleic bases and their modifications.

Deviations of B3LYP predicted bond lengths (in Å) from X‐ray values in 5FC using STO‐3G, STO(1M)‐3G, and seven other basis sets in the gas phase and DMSO solution

1 Note. The RMS data are also shown.

• 2 Abbreviations: 5FC, 5‐fluorocytosine; RMS, root mean square.
• 3 6‐311++G**.
• 4 6‐311++G(3df,2pd).

It is also evident from Table S1 that 5FC is highly polar in the gas phase and that its polarity is higher in DMSO.

There are 33 vibration modes in 5FC, and all calculated wave numbers are included in Table S2. Unfortunately, modes 1 and 30 were not observed experimentally.[52] In Table  are gathered deviations of B3LYP‐calculated harmonic vibrations, obtained with STO‐3G, STO(1M)‐3G basis set, and selected Pople and Jensen ones, from experimental data in the gas phase.[52] The RMS value obtained with initial STO‐3G basis set is very large (103 cm−1). As before for interatomic distances, the revised basis set STO(1M)‐3G performs similarly to other tested ones (RMS of about 70 cm−1). Harmonic frequencies above 1,400 cm−1 are significantly overestimated (by more than 100 cm−1). Interestingly, many anharmonic frequencies predicted with vibration second‐order perturbation theory method produce significantly higher anharmonic frequencies than did the corresponding harmonic ones (see Table S2). The anharmonic frequencies were only calculated with a few smaller basis sets due to very long computational time and inferior quality of results. Thus, for both STO(1M)‐3G and 6‐311++G** basis sets, the RMS values are about twice the values for the corresponding harmonic ones (69 vs. 120 and 73 vs. 144 cm−1). As for bond lengths, the STO(1M)‐3G basis set seems to perform fairly well in comparison with significantly larger Pople‐ and Jensen‐type ones.

Deviations of B3LYP predicted harmonic vibrations (in cm −1) in 5FC from experimental gas‐phase values using STO‐3G, STO(1M)‐3G, and seven other basis sets

• 5 Note. The RMS data of harmonic and selected anharmonic vibrations are also shown.
• 6 Abbreviations: 5FC, 5‐fluorocytosine; RMS, root mean square.
• 7 6‐311++G**.
• 8 6‐311++G(3df,2pd).

In Table  are shown B3LYP‐calculated nuclear shieldings of 5FC in vacuum and DMSO. A comparison of these results, obtained with STO‐3G, STO(1M)‐3G, and selected Pople and Jensen basis sets, indicates that nuclear shieldings are very sensitive to the quality of basis set. Thus, depending on the nuclei, in some cases, the absolute values differ by up to 10–30 ppm. Obviously, the smallest basis set used (STO‐3G) produces both carbon and proton nuclear shieldings deviating markedly from the more complete basis sets. For brevity, in Figures S1 and S2 are shown 13C and 1H nuclear shieldings calculated with the selected basis sets versus the experimental chemical shifts. It is encouraging to see a significantly better performance of modified STO(1M)‐3G basis set in predicting carbon shieldings. On the other hand, it is more difficult to judge the proton results due to H‐bonding shifting experimental signal positions. From these results, one could not recommend STO‐3G basis set for GIAO NMR prediction of absolute carbon shieldings, and the modified version, STO(1M)‐3G, performs significantly better.

B3LYP‐calculated nuclear shieldings of 5FC in the gas phase and DMSO using STO‐3G, STO(1M)‐3G, and selected Pople and Jensen basis sets

• 9 Abbreviation: 5FC, 5‐fluorocytosine.
• 10 6‐311++G**.
• 11 6‐311++G(3df,2pd).

The 5FC proton and carbon chemical shifts in Table  were obtained from data included in Table  using benzene as a reference molecule (in case of molecules with double bonds and/or aromatic character, benzene works better as reference molecule for C‐13 and H‐1 chemical shifts than does tetramethylsilane). For completeness, in Table S3 are listed benzene shieldings, obtained with the same combinations of basis sets as for 5FC. Interestingly, the most accurate carbon shielding in benzene (about 58 ppm) is obtained with STO(1M)‐3G basis set and is close to complete basis set result[7] obtained from very sophisticated Coupled Clusters with single and double excitations with perturbational treated triples (CCSD(T))[53] calculations. This result is also very close to an experimental value of 57.105 ± 0.009 ppm.[54] The best used basis sets, apcS3 and apcJ3, produce about 18 ppm smaller shieldings. This is a known deficiency of DFT, leading to about 20 ppm lower shieldings due to overestimation of the paramagnetic component. However, due to systematic error cancellation, the chemical shifts for the same nuclei, calculated with selected basis sets, are fairly similar and less scattered (up to 15 ppm; see Table ). Moreover, the carbon shift RMS values for free 5FC are acceptable and are in a range from 2.58 to 3.63 for aug‐pcS‐2 and 6‐311++G** basis sets. The corresponding carbon RMS value, calculated with the revised STO(1M)‐3G basis set, is nearly two times larger (5.49 ppm). This value is about 2.5% of typical 13C chemical shifts (most spectra are in a range of 200 ppm). Therefore, the 13C chemical shifts for 5FC in the gas phase, obtained with modified STO(1M)‐3G basis set, though acceptable, are less accurate than those obtained with traditional basis sets.

B3LYP‐calculated 1 H and 13 C chemical shifts (in ppm) of 5FC in the gas phase using STO‐3G, STO(1M)‐3G, and selected Pople and Jensen basis sets and benzene as reference

• 12 Note. The carbon RMS, as well as, the absolute deviations from experiment for nonexchangeable H6 proton, is also shown.
• 13 Abbreviations: 5FC, 5‐fluorocytosine; RMS, root mean square.
• 14 In DMSO, this work.
• 15 6‐311++G**.
• 16 6‐311++G(3df,2pd).

The performance of different basis sets in prediction of 5FC carbon atoms chemical shifts is illustrated in Figure . It is apparent from Figure a that chemical shifts of carbons C2, C4, and C6 are underestimated from −10 to −5 ppm for all studied basis sets. The worst deviation is observed for carbon C4 in case of STO(1M)‐3G basis set (about −10 ppm). A significant improvement in prediction of carbon chemical shifts in 5FC is observed as a result of DMSO solvent inclusion (Figure b). Interestingly, in the presence of DMSO, all chemical shifts are predicted with significantly higher accuracy (deviations of about −2 to 2 ppm) for all basis sets. Only for C4 chemical shift calculated with STO(1M)‐3G basis set is the improvement less significant (deviation of −10 lowers to −9 ppm).

It is also very encouraging to observe smaller 13C RMS values for all basis sets in case of both geometry optimization and GIAO NMR calculations in DMSO. Therefore, the best results are obtained by using both optimization of 5FC geometry and subsequent GIAO calculations of NMR shieldings in DMSO. It is worth reminding that the initial STO‐3Gmag basis set[15] performed very well in predicting geometry and carbon chemical shifts in single‐walled carbon nanotubes and fullerenes.

Interestingly, the deviation of H6 chemical shift in the gas phase is smaller (−0.69 ppm) than that obtained with significantly larger and better basis sets (about −0.9 ppm). The corresponding deviations in DMSO are even better (−0.5 and −0.7 ppm).

Recently, we reported on theoretical NMR parameters for cytosine,[55] that is, the parent molecule of 5FC. Using STO(1M)‐3G and 6‐311++G** basis sets, we compare in Table  the 13C RMS values and deviations of H6 and H5 chemical shifts for free cytosine and 5FC molecules and in DMSO solution. It is apparent from Table  that carbon and proton chemical shifts of cytosine and 5FC are more accurate in DMSO than in the gas phase. In the gas phase, both STO(1M)‐3G and 6‐311++G** basis sets perform better in predicting carbon and proton data of 5FC versus cytosine. The 13C data calculated with 6‐311++G** basis set in DMSO are significantly better than those obtained with STO(1M)‐3G (RMS of about 1.8 vs. 5 ppm). Deviations of H5 proton chemical shifts of free cytosine are very large (up to about 30% of typical proton chemical shift range) and somehow smaller in DMSO. Interestingly, the deviation of H6 chemical shift in 5FC is significantly smaller (−0.5 to −1.0 ppm).

RMS values of B3LYP‐calculated cytosine and 5FC 13 C NMR chemical shifts, in both the gas phase and DMSO, using STO(1M)‐3G and 6‐311++G** basis sets

• 17 Note. The absolute deviations from experiment in case of nonexchangeable H5 and H6 protons are also shown.
• 18 Abbreviations: 5FC, 5‐fluorocytosine; NMR, nuclear magnetic resonance; RMS, root mean square. The experomental values for 5FC and cytosine are from refs.[[55]]

Finally, in Table  are shown numbers of basis functions for the basis sets involved in the current study. Size of calculations and number of basis sets have a direct impact on the calculation time. It is clear that the STO(1M)‐3G basis set is constructed with very small number of functions (169) used for calculations on 5FC in comparison with the remaining ones, containing from 226 to 1,270 basis functions. On the other hand, the modified STO(1M)‐3G basis works about two times longer than does 6‐311++G**. Besides, the central processing unit (cpu) time for the latter and a very small original STO‐3G basis set (49 basis functions) are similar (16 and 17 min). Thus, the reason for this is probably related to the different number of primitive Gaussians and Cartesian basis functions (see Table ).

Number of basis functions used for 5FC optimization and gauge‐including atomic orbital NMR calculations in the gas phase and cpu time (in days, hours, minutes)

• 19 Abbreviations: 5FC, 5‐fluorocytosine; NMR, nuclear magnetic resonance.
• 20 6‐311++G**.
• 21 6‐311++G(3df,2pd).

We used a newly redesigned STO(1M)‐3G basis set for B3LYP geometry optimization, harmonic frequency calculations, and prediction of GIAO NMR parameters of free 5FC and in the presence of DMSO solvent, modeled using the PCM approach. For a direct comparison, we performed additional calculations using significantly larger 6‐311++G**, 6‐311++G(3df,2pd), aug‐pc‐2, aug‐pcS‐2, aug‐pcS‐3, aug‐pcJ‐2, and aug‐pcJ‐3 basis sets. The original STO‐3G basis set performs significantly worse in predicting 5FC geometry, harmonic frequencies, absolute nuclear shieldings, and chemical shifts. Additionally, we checked the performance of STO(1M)‐3G and 6‐311++G** basis sets for free cytosine and in DMSO. The STO(1M)‐3G basis set, designed for prediction of NMR parameters, performed unexpectedly well in prediction of 5FC geometry and harmonic frequencies. The predicted B3LYP/STO(1M)‐3G carbon chemical shifts in the studied heteroatom molecules, though acceptable (RMS of about 5 ppm), were less accurate than those obtained with larger Pople‐type and dedicated Jensen‐type basis sets. The newly tested STO(1M)‐3G basis set is significantly better than the original STO‐3G for prediction of NMR parameters of 5FC. However, the traditional Pople basis set 6‐311++G** seems to be faster and more efficient for calculating NMR shieldings.

T. K. and M. A. B. were partly supported by the Faculty of Chemistry, University of Opole. We gratefully acknowledge the Wroclaw Centre for Networking and Supercomputing for letting us use their software and hardware. We are also grateful to Sergiy Okovytyy for supplying STO(1M)‐3G basis set.

GRAPH: Table S1. Comparison of B3LYP predicted bonds (in Å) and dipole moment (in D) in free 5FC and in DMSO using STO‐3G, STO(1M)‐3G, and seven other basis sets with X‐ray values 1.Table S2. B3LYP predicted harmonic and anharmonic vibrations (in cm−1) in free 5FC vs. experiment2Table S3. B3LYP calculated nuclear shieldings of benzene in the gas phase and DMSO using STO‐3G, STO(1M)‐3G and selected Pople and Jensen basis sets.Figure S1. Correlation between C‐13 nuclear shieldings of 5FC in (top) vacuum and (bottom) in DMSOFigure S2. Correlation between H‐1 nuclear shieldings of 5FC in (top) vacuum and (bottom) in DMSO