In Table S1 (see supplementary material) were presents the results to conformational analysis carried out to the O, S and N structures. The dihedral angle analyzed was C1′-C7′-C8′-X (X = O, S or N) for the three molecular structures. These results indicate that the conformers to each case have similar energies, thus we have selected two conformational structures to each compound (O, N and S) to start DFT calculations.

The data referring to the binding angles, binding distances and dihedral angles calculated using the B3LYP/6-31G(d,p) and B3LYP/6-31+G(d,p) methods (Table S2) after geometry optimization were analyzed to find the most probable heteroatom. As shown in Table S2, the structures calculated using B3LYP/6-31G(d,p) and B3LYP/6-31+G(d,p) showed very close bond lengths and bond angles for both conformers of structures O, S and N, however there are differences on the structural parameters between the three possibilities that are relevant especially near the O, S or N heteroatoms, as expected. The dihedral angle C1′-C7′-C8′-X obtained for the three structures possibilities (O, S and N structures) showed a difference of about 110° when calculated using B3LYP/6-31G(d,p) and B3LYP/6-31+G(d,p). While the calculated dihedral angle O7′-C7′-C8′-X varied significantly with the geometry change, the calculated dihedral angles C8′-X-C9-O9 e X-C9-C8-C7 were quite close. The dihedral angles C9-C8-C7-C1, C2-C3-O3-CMe and O4-C4-C3-O3 showed that the conformers are more sensitive to the geometry changes than to the applied DFT methods. So both methods can be used to describe the geometry of these molecules.

The data for the FA, FADs and TMS (internal standard) (shielding constants of 32.1843 to ¹H and 186.3296 to ¹³C) were calculated in gas phase at the B3PW91/DGDZVP and B3LYP/6-31+G(d,p) levels. The experimental and theoretical chemical shift for the ¹³C and ¹H NMR data (chemical shifts) of the six structural possibilities of the FADs are showed in Tables 1 and 2 respectively, as well as the residue (RS) in ppm for each of the carbon and hydrogen atoms present in the structures. Tables 1 and 2 show the proximity existing between the values calculated by the DFT methodologies and those obtained experimentally for the O, S and N structures (Figures 1 and S1), which confirms the effectiveness of the computational model utilized for analyzing the possible structures. Consequently, the residual values in the structural region near the possible heteroatom provide an indication of which heteroatom may be the ligand of the studied FAD. This study showed that the proposed method can be used to identify unknown derivatives by the comparison between experimental and theoretical spectra. An example was carried out for the FA, confirming the effectiveness of computational methods (see Tables 3 and 4).

The structures with the nitrogen atom shows lower residue values for δ(¹³C) than those of the O and S structures. Residue values (RS) of 7.2; 1.9 ppm (B3PW91/DGDZVP) and 5.2; 7.2 ppm [B3LYP/6-31+G(d,p)], respectively were obtained for positions 8′ and 9, while for the same positions, the O and S structures showed RS = 26.9; 0.3 ppm (B3PW91/DGDZVP) and RS = 26.5; 2.3 ppm [B3LYP/6-31+G(d,p)] for carbon 8′, respectively, and for position 9 values of 4.7; 30.6 ppm (B3PW91/DGDZVP) and 5.9; 23.0 [B3LYP/6-31+G(d,p)], were found respectively. To explain the high residual values at position 8′ of the O structure (26.9 and 26.5 ppm) and at position 9 for the S structure (30.6 and 23.0 ppm) two electronic effects can be used. The first is the electronegativity difference between the three suggested heteroatoms: the oxygen is the most electronegative atom and deshields the carbon atom at position 8′, consequently, it will display the largest δ. The second is the electronic effect of resonance between the pair of free electrons of the suggested heteroatom with the carbonyl group in position 9. In the case of S structure, the atom is too large and the orbital overlap necessary for the resonance effect to take place, it is more difficult, leading to an electron shielding effect on the carbon atom of position 9, reveled in the largest δ (Figure 2). As expected, hydrogens of position 8′ of O structure are more deshielded by the heteroatom than those of N and S structures.

The indication of the N structure has also been suggested by analysis of the linear fit parameters presented in Table 4. The correlation coefficients (R²) acquired for δ(¹³C) were 0.99 [B3PW91/DGDZVP and B3LYP/6-31+G(d,p)] and 0.94 (B3PW91/DGDZVP) for δ(¹H) for the N structure and these values are higher than those found for the other structures. The MAE (mean absolute error) for the three structures was significantly corrected by the method employed. The general correlation of the data of Table 4 is satisfactory, given that all the signals coming from the different functional groups are divided into their own distinct regions. Thus, the general reliability of those calculations is confirmed

The linear setting performed with the experimental chemical shift data and calculated using the B3PW91/DGDZVP and B3LYP/6-31+G(d,p) methods for the three structural possibilities may be observed in Figures 3 and 4, respectively, where the linear correlation graphs for NMR data for ¹³C (a) and ¹H (b) are presented. Among the NMR data for ¹³C and ¹H of the three cases, the N structure is the best linearly adjusted one, including in the region where the electronic cloud is most dense. Statistical analysis shows that all models of ¹H and ¹³C presented good linear regression (90% < R² < 99%) (Table 4). However, the best models for the cross-evaluation procedure were in calculating the N structure for ¹³C NMR in B3LYP/6-31+G (d,p) (F = 1269.99; s = 3.4479; s_PRESS = 1.01 and Q² = 98.3%), and ¹³C NMR in B3PW91/DGDZVP (F = 1675.82; s = 3.2676; s_PRESS = 0.89 and Q² = 98.86%).

Frequency calculations of the normal vibration modes of the proposed structures were performed with the B3LYP/6-31G(d,p) and B3LYP/6-31+G(d,p) methods and the values obtained for the main absorption bands are showed in Table 5. A large absorption band was detected in 3370 cm⁻¹ characterizing the presence of hydrogen bonds (OH and NH groups) in the experimental spectrum. The value obtained utilizing theoretical calculations for NH were 3612 cm⁻¹ [B3LYP/6-31G(d,p)] and 3825 cm⁻¹ [B3LYP/6-31+G(d,p)]; the vibration of OH groups for O, S and N structures were observed in 3782–3823 cm⁻¹ [B3LYP/6-31G(d,p)] and 3828–3807 cm⁻¹ [B3LYP/6-31+G(d,p)], 3760–3822 cm⁻¹ [B3LYP/6-31G(d,p)] and 3828–3633 cm⁻¹ [ B3LYP/6-31+G(d,p)] and 3763–3822 cm⁻¹ [B3LYP/6-31G(d,p)] and 3824–3771 cm⁻¹ [ B3LYP/6-31+G(d,p)], respectively. The carbonyl stretching band was 1680 cm⁻¹ and the values obtained theoretically were, respectively, 1794 cm⁻¹ [B3LYP/6-31G(d,p)] and 1768 cm⁻¹ [B3LYP/6-31+G(d,p)] (O structure), 1754 cm⁻¹ [B3LYP/6-31G(d,p)] and 1701 cm⁻¹ [B3LYP/6-31+G(d,p)] (S structure) and 1758 cm⁻¹ [B3LYP/6-31G(d,p)] and 1727 cm⁻¹ [B3LYP/6-31+G(d,p)] (N structure). The asymmetric deformation of CH₃ groups, was 1510 cm⁻¹, and the values obtained by the computational method were 1504 cm⁻¹ [B3LYP/6-31G(d,p)] and 1509 cm⁻¹ [B3LYP/6-31+G(d,p)], 1502 cm⁻¹ [B3LYP/6-31G(d,p)] and 1509 cm⁻¹ [B3LYP/6-31+G(d,p)] and 1501 cm⁻¹ [B3LYP/6-31G(d,p)] and 1510 cm⁻¹ [B3LYP/6-31+G(d,p)] for the O, S and N structures, respectively. The data acquired with the two methodologies for the possible O, S and N structures, as expected, are close, because they are in the same range of absorption in the infrared experimental spectrum.

The [α]_D²⁵ = −16° obtained experimentally indicate that the enantiomer isolated has absolute configuration S in the position 7′. Similar result was obtained by Dellagreca et al. [18].

Stems of C. pullei (1.00 kg) were collected in the municipality of Peixe-boi (PA, Brazil) and identified by Ricardo Secco, a botanist at the Museu Paraense Emílio Goeldi (Belém-PA, Brazil). The stems was air dried, ground and extracted by percolation with hexane (7 days) and methanol (14 days), with filtrations every 3 or 4 days. The solutions were concentrated under vacuum in a rotary evaporator, resulting in hexane extract (0.65 g) and methanolic extract (80.00 g). Part of the methanolic extract (40.00 g) was submitted to partition with dichloromethane, ethyl acetate and n-butanol. The resulting solutions were concentrated in a rotary evaporator. The dichloromethane phase (4.80 g) was fractionated by column chromatography (CC) in silica, using mixtures of hexane, ethyl acetate and methanol in gradients of increasing polarities as eluents. The column fraction eluted with the mixture of hexane-AcOEt 60% was submitted to column chromatography procedure on Sephadex LH-20 using methanol as eluent, leading to isolation of 35 mg of the FAD. [α]_D²⁵ = −16° (c 0.01, CH₃OH), IR spectra were recorded in KBr in the spectrometer Nicolet IS10 FT-IR of Thermo Scientific) and ¹H and ¹³C NMR data were obtained at 300 and 75.4 MHz, respectively, in CDCl₃ using the solvent peak as the internal standard.

The O, N and S structures were drawn using HyperChem Release 7.5 software [23] and submitted to an initial optimization at PM3 [24]. In addition, the conformational analysis was carried out to confirm the minimum energy structure to the three possible structures (O, N and S), by carrying out a series of partial optimizations constraining the concerned dihedral angle step by step within the appropriate range, with a step size of 10°, these calculations were carried out using the HF/STO-3G basis set, the dihedral angle analyzed was C1′-C7′-C8′-X (X = O, S or N) for the three molecular structures. Previous studies about the importance of conformational analysis involving NMR calculations has been published [25,26] for a large number of natural products. The molecular structures were optimized with the Gaussian^® 03W [27] program, using the hybrid functional B3LYP together with the 6-31G(d,p) and 6-31G+(d,p) basis set. Vibrational analysis was performed using the procedure contained in the Gaussian^® 03W package [27] with the DFT method using the B3LYP/6-31G(d,p) and B3LYP/6-31+G(d,p) levels, in the gas phase. This ensured that each gradient optimization located indeed a true minimum energy structure (no imaginary frequencies). The normal vibration modes were visualized using the Hyperchem 7.5 program [23]. Data for NMR (¹³C and ¹H chemical shifts) were calculated using the DFT/B3PW91/DGDZVP and B3LYP/6-31+G(d,p) methods in vacuum. Recently, we have successfully used DFT/B3PW91/DGDZVP methodology to study ¹H and ¹³C NMR spectra of cordatin [28] and 8-epicordatin [29]. Two conformers each structure (O, N and S) were submitted to DFT calculations by different methodologies (see Table S3); we chose this methods because the energies of the conformers are similar. The Spartan ′08 program [30] was utilized to calculate electrostatic potential surfaces of the O, S and N structures utilizing the DFT/B3LYP/6-31G(d,p) method.

MINITAB14 [31] software was employed for statistical analysis of NMR linear regression data. The correlation coefficients (R²), the Fischer values (F) and the standard deviation (s) were the statistical parameters chosen for this analysis. For each one of the conformer of the molecules studied (O, N and S structures) parameters are presented for linear adjustment a and b: δ_calc = a + bδ_exp, mean absolute error: MAE = ∑|δ_calc – δ_exp|/n and corrected mean absolute error: CMAE = ∑| δ_corr – δ_exp|/n [32]. These parameters, calculated for experimental and theoretical data in these structures, allow the study of chemical displacements (ppm), as well as of residues: RS = |δ_exp – δ_calc|of the hydrogen and carbon atoms and the influence of the heteroligands involved in the region next to positions 8 and 9′.

The equations obtained were tested for their predictive power using a cross-validation procedure, which is a practical and reliable method for testing significance. This approach, known as “leave on out”, consists in developing a number of models with one sample omitted at the time. After these models are obtained, the omitted data are predicted and the differences between the real and predictor values are calculated. The sum of the squares of this difference is computed, and finally, the performance of the model (its predictive capacity) can be given by PRESS (predicted sum of squares) and by s_PRESS (standard deviation of the cross-validation) [33]:

where y is the experimental value, ŷ is the predictor value, n is the number of samples used for the construction model and k is the number of NMR parameters.

The predictive capacity [33] for the model was also quantified in terms of Q², which is defined as: