Chemical shift tensors as probes of chalcogen bonds: solid-state NMR study of telluradiazole-XCN⁻ (X = O, S, Se) salt cocrystals

The synthesis of 3,4-dicyano-1,2,5-telluradiazole was performed using a procedure reported by Semenov et al. (2012) with some modifications (see Supplementary Material). The preparation of the telluradiazole-XCN⁻ (X = O, S, Se) salt cocrystals similarly followed procedures adapted from the literature (Fig. 1; see Supplementary Material) (Semenov et al. 2018). TeCl₄, cyclic polyether 18-crown-6, KSCN, and KSeCN were obtained from Sigma–Aldrich. Diaminomaleonitrile and KOCN were obtained from Acros Organics. Isotopically enriched KOC¹⁵N (95%) was obtained from CortecNet. Solvents were dried with molecular sieves. Synthesis was performed using common Schlenk and vacuum-line techniques under argon. For work up, a rotary evaporator was employed to distill reaction solvents under reduced pressure. Further synthetic details are provided in the Supplementary Material.

To ensure that the compounds were successfully prepared as described above, and that their morphologies matched those reported previously in the literature, single-crystal structures were determined by using a Bruker AXS Kappa Apex diffractometer equipped with MoKα (λ = 0.7103 Å) radiation with an APEX II CCD detector (University of Ottawa, Canada). Crystals were mounted onto a glass fibre and cooled to 200 K (±2 K) before data collection using a liquid nitrogen cryogen system. Raw data collection and processing were performed with the Bruker APEX III software package. Crystal structures were solved using WinGX and Olex2 software packages. Packing diagrams were generated using Mercury 4.1.0 and POV-Ray, and the former was used to measure bond angles and bond distances.

Powder X-ray diffractograms were acquired using a Rigaku Ultima IV powder diffractometer (University of Ottawa, Canada) with a copper source and one diffracted beam monochromator. Relevant parameters include a 2θ scan angle of 5 to 50°, a scanning speed of 1°/min, a division slit of 1/3°, a scattering slit of 1/3°, a division height of 10 mm, and a rectangular slit of 0.6 mm. Experimental diffractograms were compared to simulated patterns generated from single-crystal structure crystallographic information files (CIFs) using Mercury software (CCDC 2021).

¹²⁵Te and ¹⁵N SSNMR spectra were acquired using a 400 MHz Bruker Avance III NMR spectrometer operating at 9.4 T with a 4 mm Bruker HXY probe (University of Ottawa, Canada). Samples were finely ground using a mortar and pestle, and packed into 4 mm o.d. zirconia rotors. Spectra were acquired using high-power decoupling (62.5 kHz) with magic angle spinning or cross-polarization MAS. ¹²⁵Te CSs were measured with respect to solid telluric acid, Te(OH)₆ (δ_iso = 685.5 and 692.2 ppm relative to Me₂Te at 0 ppm) (Collins et al. 1987). ¹⁵N CSs were referenced to isotopically enriched ¹⁵NH₄Cl (δ_iso = 39.3 ppm relative to NH₃(l) at 0 ppm) (Bertani et al. 2014). Data for each sample were collected at different MAS rates ranging from 1.500 to 11.000 kHz to determine the isotropic peak(s) and to measure CS anisotropy via a Herzfeld–Berger analysis (Herzfeld and Berger 1980) using the HBA software package (Eichele 1995). Typical recycle delays for the ¹²⁵Te direct-observe experiments were 60 s. Further experimental details are provided in the Supplementary Material.

DFT calculations of magnetic shielding tensors were carried out using two methods: a cluster-model approach within the Amsterdam Density Functional (ADF) software and an approach using periodic boundary conditions as implemented in the CASTEP software package. Both sets of calculations used the published single-crystal structures of compounds 1, 1a, 1b, and 1c (Semenov et al. 2018) as described below.

Cluster models of compounds 1, 1a, 1b, and 1c were built for DFT calculations carried out using ADF software, part of the Amsterdam Modeling Suite (2019.305) (te Velde et al. 2001; Baerends et al. 2019). In each case, a model was generated to ensure the coordination environment of the tellurium atom of interest was properly described (e.g., including close contacts) (see Supplementary Material). No geometry optimization was subsequently performed. The zeroth-order regular approximation (ZORA) method was used to include (i) scalar relativistic terms and, in a second set of calculations and (ii) scalar and spin–orbit (so) relativistic terms in the magnetic shielding tensor calculations (Schreckenbach and Ziegler 1995, 1996, 1997; Wolff and Ziegler 1998; Wolff et al. 1999; Autschbach and Zurek 2003; Autschbach 2013). The PBE0 functional (Adamo and Barone 1999) was used with the TZ2P basis set.

Experimentally established CIFs were used as input for CASTEP (version 20.11; Clark et al. 2005) calculations of nuclear magnetic shielding tensors using the gauge-including projector-augmented wave (GIPAW) DFT method (Pickard and Mauri 2001; Profeta et al. 2003; Yates et al. 2007). CCDC codes are as follows (Semenov 2018): BIHQOB (1a; ccdc 1837178), BIHQUH (1b; ccdc 1837179), BIHRAO (1c; ccdc 1837180). Calculations on KOCN were performed using a CIF based on the published single-crystal X-ray structure (Hendricks and Pauling 1925); however, this structure features a 50:50 (presumed) static disorder of the oxygen and nitrogen atoms. To model this disorder, the structure was manually modified in Materials Studio 5.0 to generate a supercell structure with P1 symmetry (lattice parameters a = b = 6.070 Å, c = 7.030 Å) wherein half of the disordered sites were randomly assigned as oxygen atoms and half as nitrogen atoms while ensuring each of the resultant ions were cyanates. Materials Studio 5.0 was used to create CASTEP .param and .cell files for all structures. The PBE (Perdew et al. 1996) and RPBE (Hammer et al. 1999) functionals were employed in separate sets of calculations using on-the-fly pseudopotentials. Cut-off energies were typically incremented to several hundred eV to ensure convergence of the computed magnetic shielding tensors (vide infra). Both the default Koelling–Harmon (Koelling and Harmon 1977) and ZORA-scalar relativistic treatments were used, in separate sets of calculations. The k-points used to sample the Brillouin zone were determined using the “fine” automated setting of Materials Studio. Convergence behaviour with respect to k-point setting is shown in the Supplementary Material.

Finally, two additional series of calculations were performed on small model systems using Gaussian 09 software (B3LYP/DGDZVP) and ADF software (ZORA-so, SAOP/QZ4P). The models were constructed from the available CIFs and specific distances and angles were varied to assess the ¹²⁵Te and ¹⁵N magnetic shielding tensor response.

EFGShield version 4.7 (Adiga et al. 2007) and MagresView version 1.6.2 (Sturniolo et al. 2016) were used to parse and visualize tensor output from the various sets of DFT calculations.

Magnetic shielding tensors were converted to CS tensors according to δ = (σ_ref-σ)/(1-σ_ref). The experimental absolute ¹²⁵Te shielding scale was used where σ_ref(Me₂Te) = 4333 ppm (Jameson and Jameson 1987). The experimental absolute ¹⁵N shielding scale was used where σ_ref(NH₃(l)) = 244.6 ppm (Jameson et al. 1981).

Compound 1, as well as cocrystals 1a, 1b, and 1c, were prepared according to adapted literature procedures. Single crystals of 1a, 1b, and 1c were grown and their crystal structures verified by single-crystal X-ray diffraction. In all three cases, the same structures were obtained as those reported in the literature. The morphology and phase purity of powders of these cocrystals were similarly assessed via powder X-ray diffraction (Fig. 2 and Supplementary Material). These findings allow for a confident interpretation of solid-state NMR data in the context of the diffraction-based structures (vide infra).

With the goal of investigating the impact on the cyanate nitrogen CS tensor of chalcogen bond formation to tellurium in cocrystal 1a, a second sample was prepared using isotopically ¹⁵N-enriched KOCN (95%). This isotopically enriched sample presented a powder X-ray diffractogram identical to that for the natural abundance sample and identical to that simulated based on the single crystal X-ray structure (Fig. 2). These data show that the crystal structure of both the natural abundance sample and the isotopically enriched sample prepared herein are consistent with that reported via single-crystal X-ray diffraction.

The ¹²⁵Te magic-angle spinning NMR spectra of solid powdered 1a, 1b, and 1c are presented in Figs. 3, 4, and 5. Spectra were acquired with high-power proton decoupling. Cross-polarization from protons did not yield good-quality NMR spectra, likely owing to the fact that the tellurium atoms in each sample are relatively distant from the only hydrogens in the system. Furthermore, these hydrogens are found on the 18-crown-6 rings, which are known to be mobile (Buchanan et al. 1987; Ratcliffe et al. 1992), thereby further weakening dipolar interactions between ¹H and ¹²⁵Te necessary for cross-polarization. The ¹²⁵Te MAS NMR spectra were acquired in a moderate magnetic field of 9.4 T, which was found to offer a good balance between sensitivity and spectral broadening due to CS anisotropy. Spectra were generally acquired with at least two different MAS rates (9.000 and 11.000 kHz) to distinguish the isotropic peaks from the spinning sidebands. These sideband manifolds were fit using the Herzfeld–Berger method (Herzfeld and Berger 1980) to determine the three principal components of the CS tensors (δ₁₁ ≥ δ₂₂ ≥ δ₃₃) as well as the span (Ω = δ₁₁-δ₃₃) and skew (κ = 3(δ₂₂-δ_iso)/Ω) of these tensors. The data are presented in Table 2, along with previously reported data for pure ChB donor 1.

Pure 1 crystallizes in the P2₁2₁2₁ space group and features a single crystallographically distinct tellurium atom. Each tellurium atom has two σ-holes available for chalcogen bonding with two additional molecules of 3,4-dicyano-1,2,5-telluradiazole, with electron donation coming from the ring nitrogens. The two resulting Te^…N chalcogen bonds have distances of 2.659 and 2.767 Å, and N-Te^…N angles of 152.1° and 149.1°, respectively. These interactions result in planar sheets of molecules within the crystal. The strength of these chalcogen bonds may be gauged via the reduced distance parameter (Table 1), R_ChB = d_ChB/Σ_vdW (where d_ChB is the distance between Te and N, and Σ_vdW is the sum of the van der Waals radii of Te and N). The R_ChB values for 1 are 0.74 and 0.77. The ¹²⁵Te CS tensor data for 1 (Table 2) have been reported previously; briefly, the isotropic CS is 2332.8 ppm and the span is 1268 ppm (Kumar et al. 2020b). The observation of one CS is consistent with the presence of a single crystallographically distinct tellurium atom in the unit cell of 1.

Salt cocrystal 1a crystallizes in the P2₁/c space group and features two crystallographically distinct tellurium atoms, each of which are engaged in chalcogen bonding via a single σ-hole to the nitrogen atom of the cyanate ion. These nitrogen atoms are also each in proximity to a potassium cation, each of which are complexed by an 18-crown-6 molecule. The two tellurium sites, and the geometries of the two Te^…N chalcogen bonds, are very similar. For example, the two Te^…N distances are 2.546 and 2.533 Å and the two N-Te^…N angles are 166.2 and 163.8°. The ¹²⁵Te MAS NMR experiments show inconclusive evidence for these two similar sites, with a possibly unresolved doublet with peak maxima at approximately 2315.7 and 2311.1 ppm (Fig. 3). The spinning sideband pattern was analyzed to obtain CS tensor parameters using one isotropic CS at 2313.1 ppm (Fig. 3). The resulting span of the tensor is 740 ppm, a decrease of 528 ppm relative to pure 1.

Salt cocrystal 1b crystallizes in the P1 space group and has one crystallographically distinct tellurium atom which engages in chalcogen bonding with the sulfur atom of the thiocyanate anion at a distance of 3.038 Å. The N-Te^…S chalcogen bond angle is 166.8°. ¹²⁵Te MAS NMR shows a characteristic spinning sideband manifold about the single isotropic CS, consistent with the X-ray crystal structure (Fig. 4). The isotropic CS is 2389.2 ppm and the span of the tensor is 1168 ppm, much larger than that for 1a, but still smaller than for the pure chalcogen bond donor 1.

Finally, salt cocrystal 1c crystallizes in the P1 space group and has one crystallographically distinct tellurium atom which engages in chalcogen bonding with the selenium atom of the selenocyanate anion at a distance of 3.126 Å. The N-Te^…Se chalcogen bond angle is 167.2°. Again, the ¹²⁵Te MAS NMR spectrum is consistent with the X-ray diffraction structure, showing one isotropic CS at 2376.3 ppm (Fig. 5). The span of the CS tensor, 1136 ppm, is again smaller than that seen for the pure donor 1.

The principal components of the ¹²⁵Te CS tensors measured presently, as well as the isotropic CSs and CS tensor spans, are consistent with data acquired for other similar tellurium-based systems featuring chalcogen bonds (Kumar et al. 2020b; Nag et al. 2022). When forming cocrystals of 3,4-dicyano-1,2,5-telluradiazole, it is a general conclusion that the isotropic ¹²⁵Te CS can either increase or decrease modestly, and that the span of the tellurium CS tensor almost always decreases, often by hundreds of ppm.

The ¹⁵N MAS NMR spectra of isotopically ¹⁵N-enriched KOC¹⁵N and ¹⁵N-1a are shown in Fig. 6. The X-ray diffraction structure of KOCN includes oxygen and nitrogen occupying the same sites with a 50:50 occupancy, thus suggesting the possibility of static or dynamic disorder in the structure. This compound packs in the I4/mcm space group and the nitrogen/oxygen sites have local C_2v symmetry. This symmetry means that the ¹⁵N CS tensor may have up to three unique principal components. Fitting of the ¹⁵N MAS NMR spinning sideband manifold results in a CS tensor with axial symmetry (δ_iso = 90.7 ppm and Ω = 231 ppm). Although the fitting process can be subject to errors (Hodgkinson and Emsley 1997), a perfectly axially symmetric tensor would be consistent with the two components which lie perpendicular to the cyanate axis having very similar magnitudes. DFT calculations of the nitrogen shielding tensor (vide infra) confirm that there is no dynamic disorder of the cyanate ions at room temperature, as the magnitude of Ω(¹⁵N) is not averaged compared to the values obtained by DFT at 0 K.

The ¹⁵N MAS NMR spectrum of ¹⁵N-1a presents two distinct CSs and spinning sideband patterns, consistent with the presence of two crystallographically distinct cyanate ions in the X-ray diffraction structure of this salt cocrystal (Fig. 6). The isotropic CSs are 79.5 and 77.8 ppm, and the spans of these tensors are 222 and 191 ppm, respectively. It is seen that both the isotropic shifts and the spans are reduced compared to those in pure KOCN. This is mainly attributed to changes in δ₃₃, the principal component which lies along the cyanate ion's axis. Its value increases from -64 ppm in KOCN to -55 and -38 ppm in each of the two sites in 1a. Furthermore, the axial symmetry of the ¹⁵N CS tensor in KOCN is lost upon formation of a chalcogen bond to tellurium in 1a. This can be quantified by the decrease in the skew of the tensor from 1.00 to 0.63, or alternatively by the decrease in the value of the intermediate component δ₂₂ from 168 ppm to 126 ppm (site 1) or 118 ppm (site 2). This loss of axial symmetry is also indicative of a decrease in symmetry at nitrogen in the cyanate ion.

Density functional theory calculations of the tellurium magnetic shielding tensors in 1, 1a, 1b, and 1c were carried out using a variety of approaches. The results of these calculations are summarized in Table 3. Four sets of data are presented. The first two, carried out using a ZORA relativistic treatment as implemented in ADF, use a cluster model, the PBE0 functional, and the TZ2P basis set. In one set of calculations, only a scalar relativistic treatment is used, while in the second one a spin–orbit relativistic treatment is used. The other two sets of calculations, carried out using the GIPAW approach as implemented in CASTEP, used the RPBE functional. In one set of calculations, the default Koelling–Harmon scalar relativistic method is used, while in the second one the scalar relativistic ZORA method is employed. These various sets of calculations help to determine the relative importance of including the crystal lattice in the model used (as is done using periodic boundary conditions in the GIPAW approach) compared to the model used to treat relativistic effects. A cursory inspection of the data in Table 3 suggests that the ZORA-spin orbit method, using a cluster model, provides results in best agreement with the experimental data.

Selected data from Table 3 are plotted in Figs. 7 and 8. In Fig. 7, the experimental ¹²⁵Te CS tensor spans (Ω), in navy blue, are compared to the DFT-computed values. Comparing the spans is one way to assess the computed values, as it removes the impact of imperfections in the computed isotropic magnetic shielding constant. The experimental values of the spans range from a low of 740 ppm for 1a to a high of 1268 ppm for pure chalcogen bond donor 1. The ZORA-spin orbit calculations (ADF) provide the best agreement with these experimental data; in the case of molecule 2 of salt cocrystal 1a, the computed value of 730.4 ppm is less than 2% below the experimental value. The computed value for 1b is in worst agreement with experiment, where an overestimation of 24% is seen (1453 ppm vs 1168 ppm). The ZORA-scalar calculations (ADF) provide the next-best agreement with experiment, but the spans are consistently underestimated by approximately 10 to 45%. GIPAW calculations using the Koelling–Harman scalar relativistic treatment consistently underestimate the experimental values by up to nearly 50%. Finally, the GIPAW calculations using the ZORA scalar relativistic treatment fare worst in comparison to experiment with most values underestimated by approximately 50%.

A further assessment of the DFT-computed tellurium magnetic shielding tensors is presented in Fig. 8, where the individual principal components are plotted against the experimental CS tensor values for the four computational methods used. Again, the ZORA-spin orbit approach provides the best agreement; the slope of the line of best fit (-1.1304) is the closest to unity among the four methods, the Pearson correlation coefficient of 0.9531 is the highest among the for methods, and the intercept of 3777.6 ppm is the closest to the experimental absolute shielding constant for dimethyltelluride, 4333 ppm. The remaining methods do not fare as well, with slopes as low as -0.4983 for the ZORA-scalar (CASTEP) calculations and R² values as low as 0.7768 for the Koelling–Harmon (CASTEP) calculations. The computed intercept for the latter set of calculations is also very low, 1869.7 ppm.

From all of these calculations, it may be concluded that the ZORA-spin orbit approach, using cluster models, provides the best results for the systems studied herein. The substantial role of spin–orbit relativistic effects seems to be the most important consideration. Upward spin–orbit shifts of several hundred ppm in each the principal components of the magnetic shielding tensor are seen when comparing the ZORA-scalar results to the ZORA-spin orbit results. These spin–orbit effects are much more important in the systems studied here than any potential effects on the magnetic shielding tensors from the surrounding molecules in the crystal lattice. This finding is consistent with previous literature reports on unrelated tellurium-containing systems (Alkan and Dybowski 2018).

DFT-computed ¹⁵N magnetic shielding tensors for KOCN and for the salt cocrystals 1a, 1b, and 1c are shown in Table 4. The same four methods used above for ¹²⁵Te magnetic shielding tensors were also used to calculate the ¹⁵N magnetic shielding tensors, except in the case of KOCN where building a cluster model was not realistic. The nuclear magnetic shielding values have also been converted to CSs using the experimental absolute shielding scale for nitrogen. As nitrogen is a lighter element, the impact of including spin–orbit relativistic effects is inconsequential for the systems studied here (i.e., changes in the components are less than 0.5 ppm). The computed ¹⁵N isotropic CS for KOCN, 80.7 ppm (GIPAW, RPBE, 500 eV, Koelling–Harmon) is 10 ppm lower than the experimental value of 90.7 ppm. The span of the ¹⁵N CS tensor for KOCN is calculated at the same level of theory to be 221.4 ppm, again about 10 ppm lower than the experimental value of 231(1) ppm. These discrepancies arise entirely due to underestimations of the δ₁₁ and δ₂₂ principal components, i.e., those which lie perpendicular to the cyanate principal axis. The δ₃₃ component is calculated more easily as, in the limit of an isolated linear system, the paramagnetic contribution to the shielding constant is zero. It is seen that the experimental and computed values of δ₃₃(¹⁵N) in KOCN are in agreement (-64(1) ppm and -63.9 ppm, respectively). The DFT calculations of ¹⁵N CS tensors for both crystallographic sites in salt cocrystal 1a, which features a chalcogen bond between the cyanate nitrogen atom and the tellurium atom, do not generally reproduce the experimentally observed decrease in span. When the same computational method is used for KOCN and for 1a (GIPAW, RPBE, 500 eV, Koelling–Harmon), an increase in the span from 221.4 ppm to 248.2 ppm is seen for site 1, and a small increase from 221.4 ppm to 222.6 ppm is seen for site 2. This is dominated by an increase in δ₁₁, which is not seen experimentally.

In addition to the impact of chalcogen bonding on the magnitudes of the tellurium and nitrogen magnetic shielding tensors, the impact on the orientations of these tensors in the molecular frame were considered. This absolute orientational information is not available from NMR studies of powdered samples; rather, approaches such as single-crystal NMR must be used. Recent experimental single-crystal NMR work from our group demonstrated how changes in CS tensor orientations could be correlated with halogen bond geometry in a series of phosphine oxide-iodoperfluorobenzene cocrystals (Xu et al. 2019). Presently, the results of DFT computations of the tellurium and nitrogen tensor orientations are assessed. As shown in Fig. 9, in both pure chalcogen bond donor 1, and in each of the cocrystals 1a, 1b, and 1c, the largest (and pseudounique) component of the ¹²⁵Te magnetic shielding tensor lies approximately perpendicular to the telluradiazole plane. In pure 1, the structure of which features self-complementary Te^…N chalcogen bonds, the direction of the least shielded σ₁₁ component bisects the N-Te-N angle of the telluradiazole ring. The orientation of σ₁₁ is modified away from this bisector as a result of chalcogen bonds to N, S, or Se. This change in orientation is also clearly reflective of the change in local symmetry about the telluradiazole ring. Conversely, the change in the orientation of the nitrogen CS tensor in 1a when compared to pure KOCN is negligible (Fig. 10); a deviation of the σ₃₃ component of only 3° from the cyanate axis is noted in 1a.

Finally, a simple model of 1 and the NCO⁻ anion was constructed and the Te^…N distance systematically modified to assess its impact on the ¹²⁵Te and ¹⁵N magnetic shielding tensors of the chalcogen bond donor and acceptor, respectively. B3LYP/DGDZVP and ZORA-so (SAOP/QZ4P) calculations were both performed. The general trend noted for ¹²⁵Te (Fig. 11) is that the span of the tellurium shielding tensor decreases as the chalcogen bond shortens and strengthens. This finding is consistent with the experimental data and with the other sets of calculations discussed above. In the case of ¹⁵N in the cyanate anion, an increase in the span is instead noted (Fig. 12) as the chalcogen bond shortens. This is generally consistent with the DFT results in Table 4, but in contrast with the experimental observation that the span of the ¹⁵N CS tensor decreases when engaged in a chalcogen bond in 1a. At the shortest relevant distances shown in Fig. 12, however, the span of the nitrogen magnetic shielding tensors becomes almost invariant to further reduction in the Te^…N distance, suggesting that in this regime other structural factors may contribute to the observed experimental decrease in span.