Enantiomeric characterization and structure elucidation of LH601A using vibrational circular dichroism spectroscopy

Abstract

LH601A is a novel non-reactive chiral molecule inhibiting Keap1-Nrf2 protein-protein interaction. The absolute configuration (AC) was independently determined in this study using vibrational circular dichroism (VCD) spectroscopy. Because of band overlapping and broadening in the IR spectrum, a direct VCD spectrum comparison method is devised without the conventional IR band alignment. Being an unbiased AC inquiry, all possible chiralities are evaluated based on the statistical analysis of VCD similarity, S_v. The AC of three-center stereoisomer LH601A is unambiguously assigned to (S,R,S). A comparative study was also carried out to investigate the structural and energy differences of calculated conformers using the polarized continuum model of dimethyl sulfoxide.

Keap1-Nrf2 protein-protein interaction is considered a critical point of a signaling pathway that can be targeted for intervention in cancer and inflammation.^1,2 A novel non-reactive small molecule, LH601A (ML334), was reported as a first-in-class direct inhibitor of the Keap1-Nrf2 protein-protein interaction.³ As a stereoisomer with three chiral centers, LH601A was assigned to an absolute configuration (AC) of (S,R,S) based on the x-ray structures of an enantiomeric pair and a stereospecific synthesis. However, lacking direct evidence leaves some doubts about the AC of LH601A, which led us to the vibrational circular dichroism (VCD) spectroscopy.^4–10

Although VCD has been widely used for AC determination of various chiral molecules,^11–13 it remains challenging for the application of flexible stereoisomers with molecular weight of 300 or above, such as LH601A. On the experiment side, the band broadening and low signal-to-noise ratio were frequently observed due to a greater number of vibrational modes in the limited VCD sensitive region (1100–1500 cm⁻¹) and multiple populated conformers. As a result, fewer well-defined bands can be identified and used to compare with those from a calculated spectrum. On the model side, the number of spectrum calculations is increased by the number of flexible torsions and the power of number of chiral centers. Even though the amount of calculations can be handled by today’s computer resources, the subsequent AC assignment becomes a daunting manual task that requires comparing visually subdued IR and VCD spectra with tens to hundreds of calculated ones.

Fortunately, a few automated VCD spectrum-comparing techniques^14,15 have emerged to ease the task. In SimIR/VCD approach, the similarity indexes, S_I and S_V, are computed for each modeled spectrum, and used as the basis for assigning AC and estimating error. It has been demonstrated that the method is effective for flexible enantiomers and stereoisomers and was implemented as a routine procedure in chiral analysis^16,17. One study¹⁸ also shows that the result using S_V directly without frequency shifting is well correlated with that using both S_I and S_V. While it has not been applied to an unknown stereoisomer, the direct S_V (SimVCD) method appears suitable for cases where observed IR bands are too weak and too broadened to be considered as the reference for frequency alignment. One such example is the IR 1400-1100 cm⁻¹ region of LH601A.

Here, we report the first SimVCD approach for the AC determination of LH601A. Being unbiased, the study assumed no information about the chirality from x-ray and synthesis, thereby investigating all eight AC possibilities. The likelihood of AC was first identified from the chirality-grouped S_V statistics. The conformation of leading S_V conformers was then sampled to generate a set of best fit spectra. Finally, the conformation averaged spectra were compared to the experimental spectra and the assignment of AC was given based on the pre-established S_V criteria. A comparative study was also carried out to investigate the structural and energy differences of calculated conformers using the polarized continuum model (PCM)¹⁹ of dimethyl sulfoxide.

2. Experiments

LH601A, prepared as described in the publication.³ was dissolved in DMSO-d6 (4.6mg/0.15mL) and placed in a 100 μm path length cell with BaF2 windows. IR and VCD spectra were recorded on a ChiralIR2XTM VCD spectrometer (BioTools, Inc.) equipped with dual PEM accessory, with 4 cm-1 resolution, 10-hour collection, and instrument optimized at 1400 cm-1. The IR spectrum of DMSO-d6 was also measured using the same cell. The baseline of the IR spectra were corrected by subtracting the IR of the solvent from those of the samples.

The conformations of four stereoisomers (RRR, RRS, SRS, SRR) associated with ML334 were generated and energy minimized using Knime v2.12.1 RDKit^20,21 with MMFF94 force field²². Those of the other four (SSS, SSR, RSR, RSS) were implicitly included because of the symmetry nature of enantiomeric structures and VCD spectra. An energy cutoff of 20 kcal/mol and a RMSD of 2 Å were used to generate starting conformers. For leading conformers (|Sv| > 0.1), smaller RMSD down to 0.5 Å in conformational search as well as manual conformation adjustments were used to cover all nearby important conformations. About 150 conformers have been subjected to density functional theory (DFT) calculations using Gaussian 09²³ with 6–31G(d) basis set and PBEPBE functional. For implicit solvent modeling, the PCM model of DMSO²⁴ in Gaussian 09 was used. The DFT calculations were carried out on XSEDE resources.²⁵ A Lorentzian half-width of 12 cm⁻¹ was used to generate the calculated spectra.

The spectrum comparison SimVCD was implemented in KNIME v2.12.1 similar to the SimIR/VCD protocol¹⁵ without the option of frequency shifting. The two similarities IR and VCD spectra are calculated as between calculated (c) and observed (o) IR and VCD spectra, respectively, are defined as:

$$S_I=\frac{Ico}{Icc+Ioo-Ico}$$ (1)

$$S_V=\frac{Ico}{Icc+Ioo-\left|Ico\right|}$$ (2)

where I_ij are the self and overlap integrals of spectra i and j, which can be observed (o) or calculated (c):

$$I_{ij}=\int Fi\left(v\right)Fj\left(v\right)dv$$ (3)

The sign of S_V indicates whether the model and the sample have the same AC (+) or they are an enantiomeric pair (−). For comparison and plotting, the strengths of the measured and calculated spectra were scaled to 0–1 for IR and −1–1 for VCD. While the automated frequency shifting¹⁵ was not used in this study due to the weak IR bands in VCD sensitive region, a manual shifting of −23 cm⁻¹ for carbonyl stretch bands (1525-1550 cm⁻¹) was applied to the spectra calculated in gas phase for better visual and S_v comparisons.

The spectrum averaging was performed by either energy-weighted (Boltzmann) or S_V-weighted¹⁶ methods. The later was done by adding an increasing fraction (starting at 0.1) of a next similar spectrum to the most similar (with highest S_V) spectrum until the S_V of the mixed spectrum reaches the maximum. This maximized S_V spectrum is then treated as the most similar one, and the process is repeated for the rest of similar spectra. The final fractions of each spectrum were normalized as the populations of corresponding conformers.

The decision for spectrum match and AC assignment is based on the statistical test theory and the statistical values of S_V. We conceptually divide all modeled conformers into two groups: spectrum-fit conformer (SFC) and spectrum-unfit conformer (SUC). We postulate that the S_V of SFC S_V differs significantly from the mean S_V of SUC. The number of SUC is usually large enough for estimating the mean, <S_V>₀, and the standard deviation (SD) of SUC spectra. As derived previously, a modeled conformer is SFC (the spectrum matches) if

1. |S_V - <S_V>₀| > 2SD ;

   and

2. |<S_V>| > |S_V|.

The first criterion says the S_V of a matched spectrum (of SFC) is significantly different (>2SD) from the mean S_V of all unmatched spectrum. The second one says the S_V of a conformationally-averaged spectrum is greater than any individual spectrum. In practices, inclusion of all conformers in SUC statistics may increase the SD a little, which makes the first criteria more stringent. Of cause, visual examination of matched spectra and conformers is the final approval.

3. Results and discussions

The experimentally observed IR and VCD spectra of LH601A are shown in Figure 1. The three VCD bands (a,b,c) in 1640-1780 cm⁻¹ are associated with the four carbonyl stretches in the IR spectrum. The next two VCD bands (d, e) in 1390-1440 cm⁻¹ are associated from the vibrational modes involving the two nitrogens. The rest of VCD bands (f-o) in 1100-1390 cm⁻¹ correspond to a flat IR spectrum. Without distinctive IR bands as references, the one-to-one VCD band links between experiment and calculation can hardly be established, which is a prerequisite in conventional VCD AC assignment. The situation led us to use the S_V directly for VCD spectrum comparison without IR band alignment.

Figure 1.

Experimental IR (bottom) and VCD (top) spectra measured in DMSO-d6. The alphabetic letters denote the corresponding bands in VCD and IR.

To narrow down the possibilities of AC and to reduce unnecessary DFT calculations, 20 low energy conformers from each stereoisomer were first screened for the likelihood AC matching. The approximate normal densities of S_V for grouped and all stereoisomers are shown in Figure 2. The S_V of all (dot curve) has a mean of −0.02 and a SD of 0.13, which is close to the S_V statistics (<S_V>₀~ 0, SD ~ 0.1) of SUC revealed in other studies.^15,18 Comparing the stereospecific S_V with the overall S_V, we found that SRS has the largest mean shifts of 0.14, implying that the SRS group likely contains SFC.

Figure 2.

The S_v distribution curves for each (n = 20) and all stereoisomers (n = 80) using normal distribution approximation.

Indeed, further conformational search in the neighborhood of leading SRS conformers yielded several high S_V conformers. Their spectra, structures and calculated properties are shown in Figure 3, Figure 4, and Table 1, respectively. Among the spectra, SRS11b resembles the experimental spectrum most with the highest S_V of 0.47. It not only matches bands in the carbonyl stretch region of LH601A, but also has a good overlap for the f – n bands.

Figure 3.

Calculated VCD and IR spectra of leading SRS conformers and the similarity-averaged VCD and IR spectra (AV/S) overlaid with the experimental ones (red). A part of calculated spectra (1525-1550 cm⁻¹) is removed and the frequency values greater than 1550 is shifted −23 to show the overlap of carbonyl stretch bands. The AV/S VCD spectrum has a S_V of 0.50 and populations as SRS11b (59%), SRS11c (6%), SRS9d (23%) and SRS9c (12%).

Figure 4.

Stick model of SRS conformer structures. Carbon, oxygen, nitrogen and hydrogen atoms are colored in grey, red, blue and white respectively. The heavy atoms of x-ray structure (xLH601A) are colored in green. The green dash line of SRS16b0 represents an intramolecular H-bond.

Table 1.

S_v, energy and geometry of SRS conformers calculated without solvent

			Torsions (degree) a
Conformer	Sv b	E (kcal/mol) c	1	2	3	4	5
SRS11b	0.47	0.271	−50.0	147.3	58.4	65.6	−18.0
SRS11c	0.31	1.027	−48.6	152.5	60.5	65.0	151.7
SRS9b	0.30	1.237	54.8	156.1	56.1	62.9	51.5
SRS9c	0.30	2.995	53.0	158.6	60.5	62.0	−145
SRS16b0	0.07	0.000	64.7	147.7	48.5	−13.2	137.6

^a.The torsion IDs are defined in the 2D structure of H601A.

^b.Calculated with a manual shifting of −23 cm⁻¹ for carbonyl stretch bands (1525-1550 cm-1).

^c.Relative to the DFT energy of SRS16b0.

In contrast, the spectrum of SRS16b0, the lowest energy conformer, does not fit the observed one, either visually or by its low S_V (0.07). The corresponding conformation in Figure 4 shows the formation of an intramolecular H-bond. This is probably unrealistic due to the geometry optimization in gas phase. In fact, this conformation is not stable in a PCM as we will show later. Using the S_V-weighted averaging, we obtained the average spectrum (AV/S of Figure 3) with a S_V of 0.50 and populations as SRS11b (59%), SRS11c (6%), SRS9d (23%) and SRS9c (12%).

To avoid erroneous AC assignment, we examined the other possibilities. It was noticed that the S_V density of RRS has the most negative tail in Figure 2, indicating that some RRS conformers have S_V significantly different from −0.02, the mean of SUC. There are two spectra from RRS group with the most negative S_V, −0.27 and −0.29, which are in the board line of criterion 1 (|S_V –[−0.02]| > 0.26). The corresponding enantiomeric spectra, SSR15d and SSR15e are shown in Figure 5 with the experimental spectrum and the best matched SRS11b spectrum. Although the SSR spectra appear similar to LH601A, a close look at the VCD sensitive region in Figure 6 shows that SRS11b is a better match in many details, which agrees with its greater value of S_V. Further discounting of AC (S,S,R) comes from the second S_V criterion for spectrum matches. While the S_V of averaged SRS spectrum is greater than any individual SRS spectrum, we did not find any SSR spectra averaging can improve its S_V beyond 0.35 (SSR15e). Together, AC (S,S,R) has been ruled out. The other ACs are ruled out by the lower S_V of SRR and RRR spectra, see supplements. Therefore, the AC of LH601A is unambiguously determined to be (S,R,S).

Figure 5.

Calculated VCD spectra of RRS enantiomer (black) overlaid with those of experiment (red) and SRS11b. A part of calculated spectra (1525-1550 cm⁻¹) is removed and the frequency values greater than 1550 are shifted −23 to show the overlap of carbonyl stretch bands. The original RRS spectra were flipped upside-down to represent as SSR spectra. The S_V of SSR15c, SSR15d and SSR15e are 0.17, 0.30 and 0.35 respectively. SSR15c is the most stable conformer in the SSR group.

Figure 6.

Calculated VCD spectra of SSR15e (black) and SRS11b (Blue) overlaid with that of experiment (red) in the frequency region of 1100-1500 cm⁻¹ .

The four spectrum-fit structures elucidated from VCD are very similar; all have their torsion 3 around 60 degree to minimize the van der Walls clash between the two fused rings. The main difference between SRS11b/c and SRS9b/c is the equatorial versus axial carboxylate. A small difference between SRS11b and SRS11c, or between SRS9b and SRS9c, is the torsion, or protonation site, of the carboxylate. Here we see that S_V is sensitive to the proton position. We also noted that SRS11c closely resembles the XDR structure.

To explore the performance of implicit solvent model on LH601A, we carried out comparative study on several conformers using PCM model of DMSO. As table 2 indicated, SRS11b is now the lowest energy conformer, in agreement with its S_V ranking. The conformation of SRS11b remains similar to that from the gas phase calculation. In comparison, SRS16b0 has the highest energy among the conformers. In fact, the gas phase conformation is not stable in PCM as the two conformations differ 54 degrees in torsion 5. The energetic agreement of PCM model can also be seen in the spectrum alignment with the experimental observations. Taking SRS11b spectra for example, the PCM spectrum has better band alignment with the experimental one in the lower frequency region, e.g. b band in Figure 5, 6. In high frequency region, the gas phase spectrum seems slightly better as the matching of i band.

Table 2.

S_v, energy and geometry of SRS conformers calculated using PCM model of DMSO

			Torsions (degree)a
Conformer	Sv	E (kcal/mol)b	1	2	3	4	5
SRS11b	0.35	0.000	−49.1	154.8	56.7	65.4	−20.0
SRS11c	0.15	0.012	−48.2	157.4	56.2	65.0	147.3
SRS9b	0.17	0.058	54.7	157.1	55.1	63.2	50.0
SRS9c	0.17	1.232	52.6	156.6	55.6	62.8	−139
SRS16b0	0.15	6.064	53.3	93.1	49.0	−3.6	83.5

^a.The torsion IDs are defined in the 2D structure of H601A.

^b.Relative to the DFT energy of SRS11b.

As for the predicted relative strength, the PCM spectrum appears overestimated for the region of 1300-1450 as shown in Figure 6, including the negatively enhanced d band. This phenomenon can also be seen across other spectra in Figure 5. That is probably why the individual and overall similarities did not improve in PCM model over the gas phase despite the better energetic agreement. Another reason could be the imperfection of PBEPBE functional used with PCM. The selection of PBEPBE was based on the balance of accuracy and speed based on the gas phase calculation for neat (1S)-(−)-α-pinene.¹⁵ The best combination of basis set, functional and PCM model for calculating VCD spectrum remains to be seen. Other studies^26,27 have shown that explicit solvent models with or without PCM are superior to PCM alone for particular molecules. In addition to the accuracy of a solvent model, cost and time are also considerations in the real world. For a routine AC determination, it is wise to try some simple calculations first. This work has shown that the gas phase or PCM calculations with a direct S_V analysis are sufficient for LH601A measured in DMSO.

4. Conclusion

The AC of LH601A has been determined independently to be (S,R,S) using VCD, which validated the assignment from the expectation of the asymmetric synthesis and the x-ray structure of the enantiomeric pair. The direct S_V analysis has been demonstrated to be capable of predicting complex AC without the alignment of IR spectrum, which reduces the amount of model calculations and suitable for AC determination of flexible multi-chiral-center stereoisomers. The calculation using PCM model of DMSO yields a better energetic agreement with observations than the gas phase model.

Figure 7.

Calculated DMSO VCD spectra of leading SRS conformers and the Boltzmann-averaging VCD and IR spectra (AV/E) overlaid with the experimental ones (red). The AV/E VCD spectrum has populations as SRS11b (33%), SRS11c (32%), SRS9d (30%) and SRS9c (5%).

Figure 8.

Calculated gas phase (SRS11b/G, blue) and DMSO (SRS11b/D, green) VCD spectra overlaid with that of the LH601A (red).