2.2.1. Tubocurarine (**4**)

The toxic alkaloid ("arrow poison") tubocurarine (**4**) was chosen as a model compound with only two stereogenic centers (C-1 and C-24) being nine bonds apart from each other (either counting the orange or the blue pathway in the macrocycle; see Figure 7). Additionally, along either way only three

out of the eight atoms in between have a proton attached, and thus **4** represents a prototype example where the relative configuration of the two remote stereogenic centers is expected to be indefinable on the basis of NOE data alone. The question is now: can RDCs contribute significantly to the assignment of the relative configuration of tubocurarine (**4**)?

**Figure 7.** Structure of tubocurarine (**4**). The two stereogenic centers (C-1 and C-24) are represented as orange circles. The nine bonds between the two centers (either way) are printed as bold orange or blue lines.

For the configurational assignment of tubocurarine (**4**), a total of 17 NOE-derived interproton distances and 16 <sup>1</sup>*D*CH RDCs were used for up to three independent alignment media, respectively, the RDC data for **4** was taken from Ref. [22] (see also Supplementary Tables S4 and S5).

Results for 1000 structures of tubocurarine (**4**) are shown in Figure 8a ("best 500") as a graphical representation of the total error for each structure, ordered according to ascending total errors. Following the methodology outlined in the previous chapter for **1** to **3**, one stereogenic center of **4** was set as reference and fixed by a single rDG chiral volume restraint (C-1), and therefore, only one center needed to be assigned in the calculations. Using NOE data exclusively, the first wrong structure is No. #80 (black curve/circle in Figure 8a). The energy difference between the two structures of opposite configuration at C-24 is extremely low (Δ*Etotal* = 0.04, see black symbols in Figure 8a). Accordingly, the total number of structures for tubocurarine (**4**) generated by rDG is almost equally distributed between both possibilities (the correct and the wrong configuration), and therefore a differentiation of the two alternative relative configurations of **4** by the NOE data set used here is impossible, as long as no further assumptions are made or additional experimental data is included.

The results can be significantly improved by adding RDC data to the restraints. A single alignment medium RDC data set with 16 individual <sup>1</sup>*D*CH RDCs added to the restraints of the rDG/DDD simulation leads to a clearly recognizable step in pseudo energy separating the first occurrence of a structure with wrong configuration (#334) from the energy minimum family of structures displaying the correct configuration of **4** (blue line and symbols in Figure 8a). This already pronounced diastereomeric differentiability is improved considerably when adding a second (Figure 8a, green line) or even third (Figure 8a, dark red line) RDC data set. Though these data sets require NMR measurements under different alignment conditions (alignment media) and are associated with quite some experimental effort, the resultant additional structural restraints add valuable information to the discrimination of diastereomers. With an increasing number *M* of alignment data sets used, the rDG/DDD simulations show a significantly increasing step in pseudo energy (*M* = 1: Δ*E* = 0.95, *M* = 2: Δ*E* = 1.73, and *M* = 3: Δ*E* = 11.33, cf. Figure 8a) between both alternate configurational assignments of **4**, and the total number of correctly identified structures increases consistently. The second and third AM RDC data sets remove the last remaining doubts on the configuration of **4** that might prevail after single-AM analysis. The predictive power of AM data sets cannot be estimated in advance of a measurement, but needs to be evaluated thoroughly after the NMR data has been acquired. For the experimentalist, this is of high significance, as adding further data and re-running a rDG/DDD simulation is very straight-forward–it simply requires adding a new RDC table in an additional input file–and within a couple of minutes a clear-cut answer on the decidability of a given structural problem is provided by the DG method presented here.

**Figure 8.** (**a**) Plot of the total "*pseudo energy*" of ranked rDG structures of tubocurarine (**4**), showing the first 500 out of 1000 structures generated (*K*NOE = 10.0 Å−2, *K*RDC = 1.5/*M* Hz−2) using only NOE restraints (black symbols) and an increasing number *M* of additional RDC data sets (blue, green, and dark red with *M* = 1–3 alignment media). The dashed lines and Δ*E* values on the right indicate the energy levels of the first wrong configuration identified (*M* = 1: #334, *M* = 2: #342, *M* = 3: #429), respectively, and thus the increasing differentiability of the correct configuration when using an increasing number of RDC restraints. The inset plot shows the corresponding data obtained using only RDCs (*M* = 1–3 data sets) without any NOE restraints. (**b**) Superposition of the first 428 DG structures (top plot) of correct configuration (*M* = 3), and backbone representation of the best-fit geometry (lowest pseudo energy, bottom plot, hydrogen atoms not connected to stereogenic centers have been omitted for clarity) of tubocurarine (**4**).

The main plot of Figure 8a shows the combined usage of NOE and RDC data, whereas the inset graph reveals, that the discriminative power for alternate configurations based on RDCs alone is smaller than the combined usage of NOE and RDC data. Though the level of differentiation still increases with the number of alignment media data sets applied, the energy is smaller and less significant (*M* = 1: Δ*E* = 0.41, *M* = 2: Δ*E* = 0.41, and *M* = 3: Δ*E* = 2.86, cf. Figure 8a, inset plot).
