NMR Spectroscopy in Complex Mixture Analysis and Structure Elucidation of Natural Products: Rethinking the Need for Separations

Abstract

Qualitative and quantitative analysis of complex mixtures and structure elucidation is generally impeded by the intrinsic complexity of the NMR spectra and the extensive signal overlap. The conventional approach to characterizing individual metabolites from complex crude extracts of natural products relies on multistep separation workflows employing diverse liquid chromatographic approaches and/or hyphenated techniques, which combine online integration of NMR with separation methods and other forms of spectroscopy. In recent decades, considerable efforts have been devoted to NMR applications in crude extracts without previous separation and isolation of the individual analytes. We present herein a critical overview of several NMR applications using chemical shift ranges of common organic functional groups, which can provide significant resolution advantages under specific experimental conditions. Particular emphasis is placed on: (i) characteristic chemical shift regions of strongly deshielded phenol OH groups, aldehyde CHO groups, hydroperoxide C-O-O-H groups and olefinic protons in conjugated double bonds; (ii) the advantages of using ¹³C chemical shift ranges through 2D ¹H-¹³C HSQC and HMBC experiments of strongly deshielded phenol OH groups, aldehyde CHO groups, hydroperoxide groups, conjugated double bonds, and deshielded aliphatic CH groups; (iii) selective 1D NMR-spin chromatography techniques (1D TOCSY, 1D NOE); (iv) multiple suppression of strong resonances for minor analyte identification and (v) band-selective excitation techniques for minor analyte identification and quantification. The complementary contributions of statistical heterospectroscopy and computational chemical shift prediction are also considered, together with a brief assessment of the NMR experimental parameters and performance characteristics.

1. Introduction

Nuclear magnetic resonance (NMR) is a principal spectroscopic technique for structural elucidation, stereoisomerism and conformational analysis of natural products [1,2,3,4,5,6,7]. Qualitative and quantitative analysis of complex mixtures, however, is generally hindered by the complexity of the NMR spectra and the strong signal overlap. The conventional approach to characterizing individual metabolites from complex crude extracts of natural products is through multistep separation procedures using a variety of liquid chromatographic approaches [8] (Figure 1). The isolated molecules are then identified with the combined use of tandem mass spectrometry (MS/MS) fragmentation and NMR experiments. This separation process involves laborious isolation work, which can take weeks or months and requires a high level of analytical skills and chemical expertise. Hyphenated techniques, which combine online integration of NMR with separation methods and other forms of spectroscopy, have also been extensively utilized for the analysis of multicomponent samples. Several variations have been utilized, including high-performance liquid chromatography NMR (HPLC-NMR), liquid chromatography-solid phase extraction NMR (LC-SPE-NMR), and LC-NMR-MS techniques [9,10,11,12] that involve expensive instrumentation and require a high level of instrumental expertise. Recent technological improvements in sensitivity using cryogenically cooled probes enabling detection at concentrations in the low μM range [13] and hyperpolarization or dynamic-nuclear polarization techniques [14], improvements in resolution and data processing have significantly expanded their application in natural product research and have enabled simultaneous separation, identification, and quantification of complex analytes and have advanced the discovery of novel analytes.

Figure 1. Schematic presentation of the various chromatographic, spectroscopic and hyphenated techniques that have been widely used in natural product research [8].

Metabolomics approaches have also been employed for a broad screening of the chemical diversity in complex extracts using high-resolution LC-MS/MS and/or 2D NMR approaches. To facilitate data processing, analysis and interpretation, a great variety of computational metabolomics tools have been developed, which can significantly accelerate and improve the structural elucidation of natural products in complex matrices [15]. Although the number of reference MS and NMR spectra has significantly increased in the last few decades, the matching rates are relatively low for certain analytes [16]. This can be attributed to the fact that several metabolites can share the same or similar biochemical and biosynthetic pathways, which are likely to result in similar spectral signatures. Thus, several substructure-based MS [17,18,19] and NMR [20,21,22,23,24,25] metabolomics workflows have been developed. Although computational metabolomic approaches can provide large-scale metabolomics analysis, further developments are necessary, especially in the field of NMR, due to the strong signal overlap of similar substructure motifs.

In recent decades, considerable efforts have been devoted to NMR applications in crude extracts without previous separation and isolation of the individual analytes using NMR-based diffusion experiments [26], multiple-quantum (MQ) NMR [27,28], homonuclear and heteronuclear NMR broadband decoupling techniques [29] and ultrafast 2D NMR [30]. We present herein a critical overview of several NMR applications using chemical shift ranges of characteristic organic functional groups, which can provide significant resolution advantages under specific experimental conditions. Particular emphasis is placed on: (i) characteristic chemical shift regions of strongly deshielded phenol OH groups, aldehyde CHO groups, hydroperoxide groups, ¹H chemical shifts in conjugated double bonds, (ii) 2D ¹H-¹³C heteronuclear single quantum coherence (HSQC) and heteronuclear multiple bond correlation (HMBC) experiments of strongly deshielded phenol OH groups, aldehyde CHO groups, hydroperoxide groups, conjugated double bonds, and deshielded aliphatic CH groups, (iii) selective 1D NMR spin-chromatography techniques which are based on extensive J coupling systems (1D total correlation spectroscopy (TOCSY)) or through space intramolecular proximity of protons (1D nuclear Overhauser effect (NOE)), (iv) multiple suppression of strong resonances for minor analyte observation and (v) band selective excitation techniques for minor analyte observation and quantification with improved sensitivity. The advantages of the complementary roles of statistical heterospectroscopy (STOCSY) and computational chemical shift prediction, together with key NMR experimental parameters and performance characteristics, are also briefly evaluated.

2. Characteristic Chemical Shift Regions

2.1. Strongly Deshielded Phenol OH Groups

Phenol type-OH groups are among the most important and frequently encountered functional groups in natural products [31,32,33]. The ¹H NMRs of the OH groups, however, usually appear at room temperature as very broad signals due to exchange with protons in protic solvents or residual traces of H₂O in aprotic solvents. Sharp phenol OH resonances are of vital importance for the unequivocal assignments of the ¹H NMR signals using ¹H-¹³C HMBC experiments [34,35,36,37,38,39]. Figure 2A shows the ¹H NMR spectrum, in DMSO-d₆, of caffeic acid which is a potent antioxidant found in a wide range of plants. Two extremely broad resonances of the C(4)-OH (Δν_1/2 ~ 203 Hz) and C(3)-OH (Δν_1/2 ~ 164 Hz) were observed. The addition of a minor amount of picric acid (Figure 2B) resulted in a significant reduction in the linewidths and, thus, excellent resolution of the C(4)-OH (Δν_1/2 ~ 1.2 Hz) and C(3)-OH (Δν_1/2 ~ 1.0 Hz) resonances. Further increase in the amount of picric acid resulted in a significant increase in the line widths (Figure 2C); the amount of added picric acid, therefore, should be carefully optimized. Due to the significant safety concerns associated with picric acid [38,40], several alternative acids have been investigated. The use of 2,2,2-trifluoroacetic acid (TFA) results in very effective sharpening of ¹H NMR signals of labile phenol and alcohol OH groups in natural products, which, in conjunction with the large ¹H NMR chemical shifts of up to 19 ppm and ¹H-¹³C HMBC experiments, provides an effective technique for the chemical analysis and direct monitoring of dynamic changes in metabolites without isolation steps in complex extracts of natural products. The use of TFA has several distinct advantages: it is a common reagent in organic synthesis, it shows excellent solubility in a variety of organic solvents, and the COOH group is undetectable in most cases due to the low concentration of TFA.

Figure 2. 500 MHz ¹H NMR spectra of the –OH region of caffeic acid (C = 5.7 × 10⁻³ M, T = 292 K, number of scans = 32), in (A) DMSO-d₆; upper trace: simulated line-shapes with two Lorentzian peaks for C-4 OH and C-3 OH resonances; (B,C) with [picric acid]/[caffeic acid] molar ratios of 12 × 10⁻³ and 219 × 10⁻³, respectively [35].

Solvent effects on phenolic OH groups have been extensively investigated under optimized resolution conditions, often with the addition of picric acid or TFA [37,41,42,43]. As shown in Table 1, the chemical shifts in the OH groups of phenol (1) and 4 methylcatechol (2) in various solvents are strongly dependent on the hydrogen bonding ability of the medium. For example, Δδ(δ_DMSO-d6 − δ_CDCl3) values of approximately 4.71 ppm and 3.69 ppm were observed for phenol (1) and 4 methylcatechol (2), respectively. The smaller shift difference in compound (2) has been attributed to a flip-flop intramolecular hydrogen bonding interaction between the C(1)–OH and C(2)–OH groups in chloroform solution [42]. The solvent-induced shift in the C(4′)–OH group of the natural product genkwanin (3) closely resembles that of phenol (1). In contrast, the C(5)–OH group of genkwanin exhibits an almost negligible chemical shift variation (<0.08 ppm) across different solvents. This behavior is consistent with the presence of a strong intramolecular hydrogen bond between the C(5)–OH and the adjacent C(4)=O group (Table 1), which effectively shields the hydroxyl proton from solvent interactions. Consequently, solvent independent chemical shifts can serve as diagnostic indicators of intramolecular hydrogen bonding in phenolic OH groups. The use of DMSO-d₆ in mixture analysis offers a distinct advantage: due to its strong solvation ability, the chemical shifts in solvent-exposed phenolic OH groups appear well outside the congested aromatic region, thereby improving spectral clarity and facilitating structural assignments [37,41,42,43,44,45,46,47,48].

Table 1. Chemical shifts in OH groups of phenol (1), 4-methylcylate (2) and genkwanin (3) in various solvents ^a at 298 K.

Compound	δDMSO-d6	δAcetone-d6	Δδ(δDMSO-d6 − δAcetone-d6)	δCD3CN	Δδ(δDMSO-d6 − δCD3CN)	δCDCl3	Δδ (δDMSO-d6 − δCDCl3)
(1)  C(1) –OH	9.36	8.29	1.07	6.90	2.46	4.65	4.71
(2)  C(1) –OH	8.57	7.58	0.99	6.42	2.15	4.88	3.69
C(2) –OH	8.71	7.65	1.06	6.52	2.19	5.02	3.69
(3)  C(5) –OH	12.95	13.03	−0.08	12.95	0.00	12.93	0.02
C(4′) –OH	10.56	9.26	1.30	7.70	2.86	5.36	5.20

^a Concentration C = 5 mM for (1) and (2); for genkwanin (3) C<< 5 mM due to the low solubility.

Figure 3 shows the ¹H NMR spectrum of a carob leaf acetone extract in DMSO-d₆ before and after the addition of a minor amount of TFA (mass ratio of CF₃COOH/plant extract ~ 2.2 × 10⁻²) which allowed the classification of various classes of phenol OH groups with the use of ¹H-¹³C HMBC experiments (see Section 3 and Section 7 for NMR experimental details) (Figure 4). The C(5)OH resonances of flavonoids (11.8 to 14.0 ppm) are significantly more deshielded than the other phenol-OH groups (Figure 3). The strong deshielding was attributed to the formation of a six-member intramolecular hydrogen bond with short O(5)H···O=C(4) distances [33,34,35,36,37,38,41]. This was verified with extensive DFT calculations of ¹H NMR chemical shifts which showed a strong dependence on hydrogen bond distances [39,42,43,44,45,46,47,48]. The C(5) –OH resonances are more deshielded in the presence of a C(2)=C(3) double bond due to extensive conjugation of this bond with the O(5)H···O=C(4) carbonyl bond (Figure 4B). Thus, in eriodictyol and naringenin, the C(5) –OH resonances are at 12.17 ppm and 12.18 ppm, respectively, while those of luteolin and apigenin, with a C(2)=C(3) double bond at 13.00 ppm and 12.99 ppm, respectively. The C(5) –OH resonances are more deshielded in the flavones (apigenin, 13.00 ppm) compared to those in flavonols (quercetin, 12.52 ppm) due to the presence and, thus, hydrogen bond cooperativity of the C(3) –OH group. The great sensitivity of the C(5) –OH resonances to long-range substitution is demonstrated in the case of eriodictyol and naringenin, which are clearly distinguished with a chemical shift difference of 0.01 ppm, although they differ in the substitution of the C(3′)–OH group, i.e., nine bonds apart. Several other regularities have been discussed in the literature [33,35,37,39]. Thus, the C(7) –OH resonances appear in a distinct and rather narrow region of 10.8–11.2 ppm, which greatly facilitates the assignment through ¹H-¹³C HMBC experiments (see below).

Figure 3. (a) ¹H NMR spectrum (500 MHz) of the phenol–OH region of 20.14 mg carob leaf acetone extract in 500 μL DMSO-d₆. (b) After titration with a TFA solution in DMSO-d₆. T = 295 K, number of scans 64, acquisition time 1.9 s and a relaxation delay 4.0 s [41].

Figure 4. Chemical shift ranges of the phenol –OH groups in DMSO-d₆ on the basis of the structure of the molecule and the nature of the substituent in the rings. (A): 15–20 ppm, (B): 10–15 ppm and (C) 8–10 ppm. The classification of the various phenol –OH groups was based on ¹H-¹³C HMBC experiments and/or sequential spiking with model compounds [33,37,38] (see Section 3 and Section 9 for experimental details).

Natural phenathroperylene quinones are a class of polycyclic aromatic compounds primarily found in fungi and plant species. The most prominent compounds are hypericin and pseudohypericin which exhibit resonances of the peri hydroxyl groups O(1)H, O(6)H, O(8)H and O(13)H in a very characteristic spectral region of 14 to 15 ppm due to the formation of strong hydrogen bonds with the C(7)=O and C(14)=O carbonyl bonds [49,50]. This allowed the facile identification and quantification of hypericin and pseudohypericin in H. perforatum extracts and commercial dietary supplements without prior HPLC separation.

¹H NMR of the strongly deshielded phenol-OH groups has also been utilized for the in-tube NMR monitoring of the enzymatic biotransformations of natural products [51]. A significant regioselective acylation of quercetin (ring B, Figure 5) and narigenin was found [52] using the Candida antarctica lipase B (CALB), although it has been reported in the literature that CALB has no detectable activity in flavonoid aglycones [53]. The identification and quantification of the reaction products, without any prior fractionation or isolation, was based on the C(5)OH resonances and the significant increase in resolution of the OH resonances upon the addition of picric acid (Δν_1/2 ~ 1.5 Hz, Figure 6) which resulted in the quantification of 3′-O-acetylquercetin(Q1), 10%, and of 4′-O-acetylquercetin(Q2), 35%, relative to unreacted quercetin (Q), 55%. Further applications included the use of the immobilized CALB enzyme in the bottom of the NMR tube but outside the active volume of the NMR coil [54,55]. The method has the advantage that the immobilized enzyme can be recycled for subsequent reactions and can easily be complemented with various NMR methods of ligand-observed biomolecular interactions [51].

Figure 5. Acylation of quercetin with the use of CALB (Novozyme 435) resulting in two products, Q1 (3′-O-acetyl quercetin) and Q2 (4′-O-acetyl quercetin). The OH(5) participates in the formation of an intramolecular hydrogen bond that is denoted with the red color [52].

Figure 6. 500 MHz ¹H NMR spectrum of the OH and aromatic regions of the enzymatically obtained monoacylated derivatives of quercetin in DMSO-d₆. (A,B) before and (C,D) after the addition of picric acid, respectively. In (C) the resonances of the OH(5) protons for unreacted quercetin, 3′-O-acetyl quercetin (Q1) and 4′-O-acetyl quercetin (Q2) are indicated with arrows [52].

2.2. Aldehyde CHO Groups

Aldehyde protons are strongly deshielded due to the presence of the carbonyl group and appear typically between 9.0 and 10.5 ppm with a characteristic J splitting pattern due to α-hydrogens. The CHO region has been extensively utilized to investigate, using various ¹H NMR techniques, lipid oxidation products of edible oils which were subjected to a variety of cooking processes and storage conditions [56,57,58,59,60]. Several alkenals, alkadienals, n-alkanals and related compounds were identified and quantified using the characteristic aldehyde proton chemical shifts and splitting pattern [59] (Table 2). When using spectrometers of 500 MHz or above, the improved spectral resolution can distinguish between lower and higher homologues of n-alkanal lipid oxidation products [61].

Table 2. ¹H NMR chemical shifts and spin multiplicities of aldehyde (CHO) oxidation products of edible lipids.

Chemical Shift (ppm)	Multiplicity	Functional Group
9.49	d	(E)-2-alkenals
9.52	d	(E,E)-2,4-alkadienals
9.55	d	4,5-epoxy-2-alkenals
9.57	d	4-hydroxy-(E)-2-alkenals
9.58	d	4-hydroperoxy-(E)-2-alkenals
9.60	d	(Z,E)-2,4-alkadienals
9.75	t	n-alkanals
9.78	t	4-oxo-alkanals
9.79	t	low molecular weight n-alkanals

d, doublet; t, triplet.

In recent years, secoiridoid compounds have received much attention because they have been associated with nutrition and health properties attributed to extra-virgin olive oil (EVOO) [62,63,64,65,66,67,68,69,70]. ¹H NMR was used as a simple and reliable method for monitoring, in the 9.1–9.8 ppm region, oleocanthal, oleacein, 3,4-DHPEA-EA (oleuropein aglycon) and p-HPEA-EA (ligstroside aglycon). For the NMR sample preparation, 5.0 g of olive oil was mixed with cyclohexane (25 mL) and acetonitrile (20 mL); after centrifugation, the acetonitrile phase was evaporated and the residue was dissolved in CDCl₃ (0.75 mL) (Figure 7). Oleocanthal + oleacein level and oleocanthal/oleacein ratio were suggested as a new index for the characterization of extra virgin oil [62]. Direct in situ ¹H NMR was used for the investigation of dynamic changes in composition, due to temperature and solvent effects, of secoiridoid derivatives of oleuropein 6-O-β-D-glucopyranoside (1) in complex olive leaf aqueous extracts (Figure 8) without prior isolation of the individual components. The equilibrium between the aldehyde hydrate and the dialdehyde forms of oleuropein aglycon in D₂O and the equilibrium between the two diastereomeric hemiacetal forms and the dialdehyde form in CD₃OD (Scheme 1) were investigated in detail [71].

Figure 7. The unique ¹H NMR (600 MHz, CDCl₃) profile in the 9.1–9.8 ppm region of two virgin olive oil varieties: (A) Koroneiki variety (with typical oleocanthal/oleacein ratio); (B) Koroneiki variety (with unusual oleocanthal/oleacein ratio). The signals for oleocanthal, oleacin, 3,4-DHPEA-EA (oleuropein aglycon) and p-HPEA-EA (ligstroside aglycon), are noted as 1, 2, 3 and 4, respectively. IS, internal standard. Number of scans 50, acquisition time 1.7 s, relaxation delay 1 s. Reprinted with permission from [62]. Copyright 2020, The American Chemical Society, Washington, DC, USA.

Figure 8. Aldehyde C–H regions of the ¹H NMR spectra (500 MHz) of 20 mg of an olive leaf aqueous extract in 0.6 mL (A) DMSO-d₆, (B) CD₃CN, and (C) CD₃OD; the upper traces represent the same spectra of the aldehyde H-5 and H-6 resonances of the compounds 2 an 3 of Scheme 1 after the application of resolution enhancement (T = 298 K, number of scans = 512, experimental time = 1 h) [71].

Scheme 1. Chemical structures of oleuropein 6-O-β-D-glucopyranoside (1) and its various aldehyde forms 2–4 [71].

The aldehyde ¹H NMR region was also used for the in situ monitoring of the reaction of proto catechuate 4,5-dioxygenase, from Pseudarthrobacter phenanthrenivorans sph3, with protocatechuic acid (PCA) and gallate [72]. The reaction product 4-carboxy-2-hydroxymuconate-6-semialdehyde (CHMS), Figure 9, exhibits characteristic resonances due to the keno-enol tautomerization process, which is slow on the NMR time scale. The two tautomeric forms were assigned with a combination of 2D ¹H-¹H COSY and TOCSY and 2D ¹H-¹³C HSQC/HMBC experiments.

Figure 9. Biotransformation of PCA (1 mM) by PcaA (8 μM), inside the 5 mm NMR tube. (a) ¹H NMR spectra 2 h (blue spectrum, (i)) and 24 h (red spectrum, (ii)) after the addition of PCA to protocatechuate 4,5 dioxygenase. ¹H NMRs of the substrate are shown in the blue spectrum. The strong imidazole resonances in the buffer are denoted with an asterisk (*). (b) Expanded region (7.8–9.4 ppm) of the ¹H NMR spectrum of Figure 9a(i) [72].

2.3. Chemical Shift Ranges of Hydroperoxides (R–O–O–H)

Hydroperoxides [73] are of great importance in several natural products, such as polyunsaturated fatty acids, steroids and terpenes, and their decomposition is related to several inflammatory diseases [74,75,76,77]. Hydroperoxides are significantly less acidic than phenol OH groups; therefore, using dry CDCl₃ and low solute concentrations to eliminate proton transfer, sharp ¹H NMRs can be obtained. For example, methyl linolenate, which was allowed to oxidize in atmospheric oxygen for 48 h at 40 °C, [78], resulted in a significant number of sharp ¹H NMRs (∆ν_1/2 ≤ 2.0 Hz) of the hydroperoxide protons in the deshielded region of 7.7 to 9.6 ppm (Figure 10). This allowed the identification and quantification of several hydroperoxide derivatives through 1D TOCSY and ¹H-¹³C HMBC experiments and DFT calculations, despite the limited literature data for comparison. The strongly deshielded resonances at ~9.5 ppm were attributed to the formation of an intramolecular hydrogen bond between the hydroperoxide proton and the O15 oxygen of the five-member endo–peroxide ring of the major 15,16-threo derivative, which is stronger than in the 15,16-erythro endo hydroperoxide molecule.

Figure 10. (a) ¹H NMR spectrum (800 MHz) of 20 mg methyl linolenate in CDCl₃ subjected to heating at 40 °C for 48 h; (b) the region of the C-O-O-H resonances of the oxidation products. Number of scans 64, acquisition time 1.02 s, experimental time 6.5 min, relaxation delay 5 s, T = 298 K [78].

Skiera et al. [79] utilized a similar methodology to that in phenolic compounds [80] in order to investigate the effect of various solvents (benzene-d₆, CDCl₃, acetone-d₆ and DMSO-d₆) and their binary mixtures, on the chemical shifts and linewidths of the hydroperoxide protons. CDCl₃/DMSO-d₆ (5:1 v:v) is an optimum binary mixture, which results in a significant deshielding in the range of 10.1–11.00 ppm and narrowing of the OOH signals. Comparison of the ¹H NMR method with the classical peroxide value (PV) analytical method was performed using an oxidized methyl linolenate standard and a triacylglyceride with a saturated fatty acid in the middle chain. It was concluded that although both methods have similar analytical performance characteristics, significant deviations were observed for olive oils, nut oils, black seed oils and pumpkin seed oils. The discrepancies between the ¹H NMR and PV methods were evaluated [79].

2.4. Olefinic Protons in Conjugated Double Bonds

Conjugated double bonds exhibit highly diagnostic resonances due to the olefinic protons in the region of 6.85–5.50 ppm. These resonances due to delocalization of the conjugated double bonds are deshielded with respect to the protons of isolated double bonds. Some regulations of the chemical shifts have been suggested: the “inner” olefinic protons are deshielded relative to the “outer” protons; in mixed cis/trans isomers, the cis bonds are deshielded relative to those of the trans bonds. DFT calculations of 18:3 conjugated linoleic acids and triene-containing model pheromone compounds showed that the strong deshielding of the closely spaced “inside” olefinic protons, which appear close to chemical shifts in aromatic systems, is primarily due to the effect of through-space C-H···H-C van der Waals steric interactions [81,82]. Four minor geometric (9-cis, 11-trans) 18:2, (9-trans, 11-cis) 18:2, (9-cis, 11-cis) 18:2, (9-trans, 11-trans) 18:2 conjugated linoleic acid (CLA) isomers were identified and quantified in the lipid fractions of lyophilized milk samples [83]. Resolution and sensitivity barriers have been successfully resolved with selective suppression of the major lipid resonances and 2D ¹H-¹³C HSQC experiments (see below).

Monounsaturated and polyunsaturated lipids are particularly susceptible to lipid peroxidation, which is a free radical process. NMR spectroscopy has been extensively utilized in the mixture analysis of oxidation products of food lipids as a function of composition and thermo-oxidation conditions without prior fractionation or isolation [56,57,58,59,60,61,84,85,86,87,88,89]. Table 3 presents chemical shifts and J couplings of the conjugated double bonds associated with hydroperoxides (OOH) and hydroxides (OH) [84].

Table 3. ¹H NMR chemical shifts and spin multiplicities of primary oxidation products of edible lipids.

Signal	Chemical Shift (ppm)	Multiplicity	Functional Group
–CH=CH–CH=CH–	6.58	dddd	(Z,E)-conjugated double
	6.00	ddtd	bonds of
	5.56	ddm	hydroperoxides (OOH)
	5.51	dtm
–CH=CH–CH=CH–	6.45	ddd	(Z,E)-conjugated double
	5.94	dd	bonds of
	5.64	dd	Hydroxides (OH)
	5.40	ddt
–CH=CH–CH=CH–	6.27	ddm	(E,E)-conjugated double
	6.06	ddtd	bonds of
	5.76	dtm	hydroperoxides (OOH)
	5.47	ddm
–CHOOH–CH = CH–	5.72	m	Single double bonds of
			hydroperoxides (OOH)
–OOH	8.3 to 8.9	—	Hydroperoxides

d, doublet; t, triplet; m, multiplet.

3. 2D ¹H-¹³C HSQC and HMBC Experiments

For the structure determination of natural products in crude extracts, the dispersion of ¹³C chemical shifts in the indirectly detected dimension of 2D hetero-correlated NMR experiments can significantly enhance the resolving power of NMR experiments, although at the expense of an increase in experimental time.

3.1. The Case of Strongly Deshielded Phenol OH Groups

2D ¹H-¹³C HMBC NMR experiments of sharp phenol OH resonances (Δν_1/2 < 2.5 Hz), optimized for ⁿJ(¹H, ¹³C) values of 10 to 2.5 Hz, result in a significant number of long-range cross-peaks (n = 2–6). This provides an excellent fingerprinting for the quantification of several classes of flavonoids (Figure 11) [37]. In addition to the great sensitivity of the OH resonances to hydrogen bonding and electronic phenomena, the ¹³C cross-peaks can be extremely informative for resonance assignment. For example, the ⁴J(¹³C, ¹H) cross-peak of the C(5) –OH to C(4)=O is highly diagnostic due to the sensitivity of δ(¹³C) of the carbonyl group to conjugation, substituent effects and hydrogen bond interactions [90,91]. In taxifolin, for example, δ(¹³C) = 198.62 ppm, while in narigenin δ(¹³C) = 182.62 ppm due to the conjugation of the carbonyl group with the C(2)=C(3) double bond. It was concluded that the combination of the δ(¹H) of C(5) –OH and δ(¹³C) = 198.62 of C(4)=O can serve as a molecular sensor in mixtures of closely related compounds [34,35,36,37].

Figure 11. The 500 MHz 2D ¹H-¹³C HMBC NMR spectrum of the C-5–OH resonances of an ethyl acetate oregano extract (30 mg) in 0.6 mL DMSO-d₆ (T = 290 K, number of scans = 88, experimental time = 11.5 h). Sorbifolin (1), ladanein (2), 2-(3,4-dimethoxyphenyl)-5,6 dihydroxy-7-methoxy-4H-chromen-4-one (3), thymusin (4), aromadendrin (5), taxifolin (6), naringenin (7), eriodictyol (8), luteolin (9). Optimization of the HMBC experiment for ⁿJ(¹H, ¹³C) values of 2.5 and 10 Hz (blue and red colors, respectively) [37].

Since the strongly deshielded OH region of 15–8 ppm is much less crowded than the aromatic region, the 2D ¹H-¹³C HMBC assignments can be verified with spiking of model compounds. Figure 12 shows the C(5) –OH (11.90–12.20 ppm) and C(7) –OH (10.80–10.95 ppm) spectral region after sequential spiking with naringenin, taxifolin and eriodictyol, which resulted in the unequivocal assignment and quantification [37]. Scheme 2 shows an experimental protocol for sequential proton and carbon resonance assignments.

Figure 12. The C-5 OH (A,a) and C-7 OH (B,a) spectral regions of the 500 MHz 1D ¹H-NMR spectra of 40 mg of a Greek oregano acetone extract in 0.6 mL DMSO-d₆ (T = 292 K, number of scans = 32, experimental time = 4 min). Sequential spiking with: (b) 0.17 μmol eriodictyol (8), (c) 0.20 μmol taxifolin (6), and (d) 0.36 μmol naringenin (7). The resonances of the spiking model compounds are denoted with asterisks (*) [37].

Scheme 2. Experimental protocol for sequential proton and carbon resonance assignments.

Yu et al. [92] performed a systematic classification of flavonoid metabolomes via data mining and quantification of ¹H-¹³C HMBC cross-peaks of the strongly deshielded C(5)-OH signals. Different types of flavonoids/chalcones in licorice were identified and expressed as barcodes, δ(¹H): 11.8–14.2 ppm and δ(¹³C): 80–180 ppm (Figure 13 and Figure 14). Notably, the number of NMR-detected metabolites of crude and enriched samples was greater than in the preparative HPLC method. Metabolomic analysis of three medicinal licorice species derived from the Glycyrrhiza uralensis (GU), G. glabra (GG) and G. inflaia (GI) species resulted in the identification of three major structural types and eight subtypes of biomedical interest.

Figure 13. Partial structures of different types of flavonoids/chalcones in licorice as well as their HMBC cross-peaks expressed as δ_H, δ_C in ppm and as barcodes (δ_H 11.8–14.2 ppm, δ_C 80–180 ppm). Reprinted with permission from [92]. Copyright 2020, The American Chemical Society, Washington, DC, USA.

Figure 14. The 11.8–14.2 ppm HMBC correlations of 41 licorice flavonoids/chalcones from Glycyrrhiza uralensis (GU), G. glabra (GG), and G. inflaia (GI). The literature (A) vs. experimental data (B) from the three licorice fractions GU-MF, GG-MF, and GI-MF (MF: methanol fractions). Colors identify the three main flavonoid/chalcone types, shapes, and subtypes. Reprinted with permission from [92]. Copyright 2020, The American Chemical Society, Washington, DC, USA.

3.2. The Hydroperoxide C-O-O-H Region

The application of the ¹H-¹³C HMBC experiment would be particularly valuable in mixture analysis of hydroperoxides, as the ¹³C chemical shifts of the hydroperoxide carbons fall within a highly characteristic region (82–88 ppm) [75,77,78]. However, optimizing the ¹H-¹³C HMBC experiment for these functional groups is very demanding since the ³J(¹³C-O-O-H) couplings are typically well below 4 Hz due to the presence of two electronegative oxygen atoms and the unfavorable torsion angles imposed by their lone electron pairs. Thus, very sharp resonances, Δν_1/2 < 1.5 Hz, and solute concentrations well above the low mM range are required for successful application. Nevertheless, the quantification of several hydroperoxides and diastereomeric endo-hydroperoxides was reported for complex mixtures of oxidation products of methyl linolenate with the combined use of ¹H-¹³C HMBC, 1D TOCSY and DFT calculations [77,78].

3.3. The Adlehyde CHO Region

Aldehyde carbons exhibit highly diagnostic resonances in the region of 190–210 ppm, which is distinctly separated from acids [93] and esters, 170–188 ppm. This region of chemical shifts, in conjunction with the strongly deshielded aldehyde protons, can provide an excellent means to investigate complex crude extracts of natural products. Figure 15 shows the 2D ¹H-¹³C HMBC spectrum of an olive leaf aqueous extract in CD₃CN. The dialdehyde form (2) of 6-O-β-D-glucopyranoside (1) (Scheme 1) shows connectivities of the 2:H-5 (δ = 9.51 ppm) to C-5 (δ = 202.9 ppm), C-4 (δ = 46.7 ppm) and C-3 (δ = 27.9 ppm) [71]. The 2:H-6 (δ = 9.20 ppm) shows connectivities to C-6 (δ = 196.6 ppm), C-8 (δ = 155.4 ppm), C-7 (δ = 144.0 ppm) and C-3 (δ = 27.9 ppm). The common connectivites of 2:H-5 and 2:H-6 with C-3 provide unequivocal assignment that the two carbonyls belong to the same molecular species. Such connectivity is also crucial for the identification of minor species (Figure 16).

Figure 15. The aldehyde region of the 2D ¹H-¹³C HMBC NMR spectrum (500 MHz) of an olive leaf aqueous extract (22.5 mg) in 0.6-mL CD₃CN (T = 292 K, number of scans = 144, experimental time = 18.5 h, ⁿJ values were optimized for 8 Hz) [71].

Figure 16. Selected region of the 2D ¹H-¹³C HMBC NMR spectrum (500 MHz) of an olive leaf aqueous extract (31.7 mg) in 0.6 mL CD₃OD (T = 320 K, number of scans = 128, experimental time = 14.5 h, ⁿJ values were optimized for 8 Hz) [71].

3.4. Conjugated Double Bonds

Olefinic carbons in conjugated systems typically appear in the region of 112 to 145 ppm, which is not distinctly different from that of the carbons of the isolated double bonds. This phenomenon has been investigated via DFT calculations [81]. Nevertheless, the clear deshielding of the olefinic protons in conjugated double bonds in combination with the large ¹³C chemical shifts dispersion can provide an efficient method in complex mixture analysis. The increased resolution with reduced ¹³C spectral width range from 112 to 142 ppm allowed the quantification of four minor (9-cis, 11-trans) 18:2, (9-trans, 11-cis) 18:2, (9-cis, 11-cis) 18:2 and (9-trans, 11-trans) 18:2 conjugated linoleic acid (CLA) geometric isomers in lipid fractions of lyophilized milk samples [83]. Figure 17 shows a selected region of the ¹H-¹³C HSQC spectrum of a lyophilized milk sample. In addition to the major resonance at 6.27 ppm with a cross-peak at 125.28 ppm due to (9-cis, 11-trans) 18:2 isomer, a minor component was observed at 6.29 ppm with a cross-peak at 125.85 ppm due to the (9-trans, 11-cis) 18:2 isomer [83]. The method has significant advantages for the analysis of CLA isomers relative to the laborious separation analytical methods [94,95].

Figure 17. Selected regions of ¹H-¹³C HSQC spectra (500 MHz) of the lipid fraction of a lyophilized milk sample. ¹³C shielding spectral range from 0 to 160 ppm and reduced ¹³C spectral range from 112 to 142 ppm for (A) and (B), respectively. 298 K, 40 repetitions of 256 increments, total experimental time 4 h 55 min and 5 h 49 min for (A) and (B), respectively [83].

3.5. Deshielded Aliphatic CH Groups

Deshielded aliphatic CH groups, due to electronically withdrawing substituents, can provide a method for the analysis of crude plant extracts without isolation of the individual components. Artemisia annua extracts are typical examples since the C(5) –H proton of artemisinin and deoxyartemisinin, due to the presence of two oxygen atoms (Scheme 3), appear in the less crowded region of 5.5–6.5 ppm and show highly diagnostic cross-peaks in the ¹H-¹³C HSQC and HMBC spectra (Figure 18) [96]. The C(5) –H signal of artemisinin at 5.83 ppm shows HMBC cross-peaks to carbons C(8) (23.6 ppm), C(11) (32.7 ppm), C(7) (44.9 ppm), C(6) (79.5 ppm), C(4) (105.1 ppm) and C(12) (172.2 ppm) and, thus, provides a fingerprinting of the complete carbon skeleton. It may be noted that the olefinic C(5) –H of dihydroartemisinic acid and C(13) –H of artemisinic acid and arteannuin B appear in the same less crowded region and, thus, are of importance in structure elucidation (Scheme 3, Figure 18).

Scheme 3. Chemical structures of compounds identified in Artemisia annua extract. The diagnostically important ¹H-¹³C HSQC and ¹H-¹³C HMBC correlations are denoted with red and blue colors, respectively [96].

Figure 18. Selected regions of the 2D ¹H–¹³C HMBC NMR spectrum (800 MHz) (a–c) of an Artemisia annua extract (20 mg) in 0.5 mL of chloroform -d (T = 298 K, number of scans = 64, and experimental time = 24 h and 25 min). The notations have as follows: Ch., Chrysosplenetin; Ch.-D, Chrysosplenol-D; 5-Hyd.-3,6,7-trimeth., 5-Hydroxy-3,6,7-trimethoxy-2-(4′-methoxyphenyl)-4H-chromen-4-one [96]. The asterisk denotes the doublet of the H13 of arteannuin B.

4. Selective 1D Spin Chromatography NMR Techniques

Selective one-dimensional (1D) NMR experiments were shown to be an excellent method in mixture analysis without previous separation and isolation of the individual analytes. The methods are based either on ¹H-¹H connectivities via J couplings which belong to a single chemical species (1D-TOCSY), or through-space (≤5 Å) intramolecular ¹H-¹H NOE effect.

4.1. 1D TOCSY

The frequency-selective 1D total correlation spectroscopy (1D TOCSY) can provide an important tool which allows the elucidation of an extensive spin-coupling network even in overlapped regions and in the presence of major components with 10²–10³ stronger NMR signal intensities. The method has been applied in the metabolomic analysis of amino acids [97], mango juice [98], lipid fraction of milk and dairy products [99,100] and determination of glutathione in white wine [101].

Figure 19 shows a selected region of the ¹H NMR spectrum of a solution of Artemisia annua extract. Selective 1D-TOCSY of the doublet at δ = 6.13 ppm (J = 3.3 Hz) of the olefinic C(13)–Ha resonance of the minor analyte arteannuin B (Scheme 3), shows very characteristic connectivities with C(13)–Hb (δ = 5.42 ppm), C(7)–H (δ = 2.74 ppm) and C(5)–H (δ = 2.72 ppm) although the above resonances are strongly overlapped or hidden under the resonances of the major analytes [96]. Similar 1D-TOCSY experiments were also performed with the aromatic resonances of flavonoids: C(6′)–H of chrysophenol D (δ = 7.57 ppm), C(6′)–H of chrysosplenetin D (δ = 7.64 ppm) and C(6′)–H of casticin (δ = 7.71 ppm) (Scheme 3, Figure 20). The quantitative NMR and HPLC results are summarized in Table 4 which showed an excellent linear correlation with R² = 0.998 and slope = 1.101 [96].

Figure 19. (a) Selected region of the ¹H NMR spectrum (800 MHz) of an Artemisia annua extract (20 mg) in 0.5 mL chloroform-d (T = 298 K, number of scans = 8, and experimental time = 1 min). (b) A 1D TOCSY spectrum of the solution (a) with mixing time 300 ms (number of scans = 256, acquisition time = 1.02 s, relaxation delay = 5 s, and experimental time ~17 min). The resonance at δ = 6.15 ppm that was excited by the use of a frequency-selective pulse is denoted with the arrow [96].

Figure 20. (a) Selected region of the ¹H NMR spectrum (800-MHz) of the solution of Figure 19. 1D TOCSY spectra with mixing time = 300 ms (T = 298 K, number of scans = 32 for (b,c), experimental time ~2 min, and number of scans = 64 for (d) and experimental time ~17 min). The resonances that were excited by the use of a frequency-selective pulse at 7.57 ppm (b), 7.66 ppm (c), and 7.71 ppm (d) are denoted with the arrows [96].

Table 4. Concentrations (±standard deviation) of compounds quantified in Artemisia annua extract (milligrams per gram dry extract) with the use of high-performance liquid chromatography (HPLC) and NMR [96].

Compounds	HPLC (mg/g Dry Extract)	NMR (mg/g Dry Extract)
Artemisinin	180.32 ± 2.17 a	180.75 ± 3.61
Dihydroartemisinic acid	39.36 ± 4.32 a	32.54 ± 1.16
Arteannuin B	b	3.14 ± 0.35
Artemisinic acid	b	14.78 ± 0.37
Camphor	b	16.17 ± 0.74
2,4-Dihydroxy-6-methoxy acetophenone	b	0.42 ± 0.05
Deoxyartemisinin	b	16.15 ± 0.61
Chrysosplenetin	20.02 ± 0.46 c	17.29 ± 0.15
Chrysosplenol-D	12.79 ± 0.53	11.65 ± 0.89
5-Hydroxy-3,6,7-trimethoxy-2-(4′-methoxyphenyl)-4H-chromen-4-one	3.73 ± 0.61	3.51 ± 0.22
Artemetin	2.12 ± 0.72	d

^a Quantitation of artemisinin and dihydroartemisinic acid was performed using liquid chromatography with tandem mass spectrometry. ^b Arteannuin B, artemisinic acid, deoxyartemisinin, camphor, and 2,4-dihydroxy-6-methoxy acetophenone were quantified only by ¹H NMR because standard compounds were not available. ^c Chrysosplenetin was coeluted with casticin. ^d Artemetin could not be quantified due to a strong NMR signal overlapping.

Selective 1D-TOCSY has also been used for the rapid identification and quantification of the minor species conjugated linoleic acids (CLAs), caproleic acid (CA) and 1,2-diglyceride (1,2-DAG) in the lipid fraction of milk and dairy products, even in the presence of major lipids with 3 × 10³ stronger NMR signal intensities (Figure 21) [99]. The same methodology was used to compare the metabolite profile of the lipid fraction of conventional and organic bovine milk [100]. Increased % contents of (9-cis, 11-trans) 18:2 CLA, linoleic acid and α-linolenic acid and decreased content of caproleic acid were observed in organic milk compared to conventional and, thus, of importance in NMR-based milk metabolomics and foodomics [102,103,104,105].

Figure 21. (a) ¹H NMR spectrum (500 MHz) of the lipid fraction of a lyophilized cheese sample in CDCl₃. The sn1, sn2, and sn3 indicate the stereospecific numbering of esterified glycerol. The inset shows a 512× magnification of the spectrum to display resonances from the 18:2 CLA and other minor species. (b–d) 1D TOCSY spectra of panel (a) with mixing time τ_m = 400 ms. The resonances that were excited by the use of a selective pulse are denoted with asterisks. In all cases, T = 298 K; number of scans = 256; acquisition time = 4.3 s; relaxation delay = 5 s; and total experiment time ~ 25 min [99].

The chemical shift selective 1D TOCSY [106] is an alternative and more effective technique, compared to the frequency-selective 1D TOCSY experiment, in investigating complex NMR spectra with strongly overlapped signals. The method has been applied in the analysis of raw honey samples dissolved in D₂O. Up to 22 oligosaccharides were identified and quantified in honey from several different botanical origins [107].

A statistical total correlation spectroscopy (STOCSY) was developed as a means of extracting 1D and 2D spectral reconstructions of biomarker molecules from patent and latent signals in 1D NMR spectra [108]. The STOCSY generates a pseudo-spectrum that reflects the correlation of the intensity variability in the ¹H NMR signals which belong to the same analyte. Even when the covariance matrix is resolution-limited, which is common in natural products datasets, STOCSY sufficiently extracts robust co-variation patterns that reflect shared spin-systems, structural motifs and biosynthetic origin. STOCSY, by selecting a driver peak, forces the algorithm to compute correlations relative to a single anchor, thus reducing the dimensionality problem. If a molecular species varies in concentration across a number of samples, its peaks co-vary, thus producing a molecular signature. This analysis allows the deconvolution and extraction of the spectroscopic characteristics of a given analyte in a complex extract.

Recently, a sensitive gradient-enhanced multiplet-selected targeted observation NMR experiment (GEMSTONE) has been applied in a number of 1D selective experiments in molecules with complex NMR spectra [109,110]. The GEMSTONE excitation was also employed with homonuclear decoupling [29], which shows all signals within an extensive J coupling system as singlets. The method greatly reduces spectral complexity and uncertainty [111] and was applied in several artificial mixtures, including glucocorticoids, which are used for the treatment of asthma. The concentrations used in the range of 100 mM limit the application of the method only to major analytes in crude natural product extracts.

4.2. 1D NOE

The 1D transient nuclear Overhauser effect (NOE) is an effective and time-efficient technique to investigate special intramolecular proximity (≤5 Å) between protons in a molecule [112]. It has been extensively used in structural analysis of bioorganic molecules; however, applications in mixture analysis of natural products are relatively few since it requires well-separated signals for successful selective excitation and, thus, reliable NOE connectivities. Figure 22 shows the ¹H-¹³C HMBC spectrum of a carob leaf acetone extract in DMSO-d₆ after titration with TFA. The C(5)–OH proton shows six cross-peaks to C(5), C(6), C(7), C(9), C(10) and C(4) (179.61 ppm) of ring C [41]. Application of the 2D ¹H-¹³C HSQC-TOCSY experiment resulted in the assignment of the rhamnopyranoside moiety based on the characteristic connectivities between the C(6″) methyl carbon and all protons of the α-rhamnopyranoside ring (Figure 23). Unfortunately, ¹H-¹³C HMBC correlation was not observed between C(1″)–H and C(3), C(4) and C(2) carbons. This problem was resolved with the use of selective 1D NOE excitation of the H1″ (which is well separated from the other signals), thus, resulting in NOE connectivities of all protons of the rhamnopyranoside moiety, the C(2′)–H and C(6′)–H and the C(5)–OH protons with very good sensitivity within 30 min experimental time (Figure 24). It was suggested that 1D transient NOE experiments could be of general interest in investigating the site and type of glycosylation in flavonoids [41].

Figure 22. Selected region of a ¹H-¹³C HMBC spectrum (800 MHz) of a carob leaf acetone extract (20.14 mg) in 500 μL DMSO-d₆. The connectivities of OH(5) and OH(7) with characteristic carbon atoms of myricetin-3-O-α-L-rhamnopyranoside are denoted with the double arrows. T = 298 K, 56 scans with 1024 increments, and total experimental time 14 h and 30 min [41].

Figure 23. Selected region of ¹H-¹³C HSQC-TOCSY spectrum (800 MHz) of the same solution as in Figure 22. Connectivities between the methyl carbon and all protons of the α-rhamnopyranoside ring are shown. T = 298 K, 64 scans with 1024 increments, mixing time = 100 ms, and the total experimental time 1 day and 7 h [41].

Figure 24. The 1D selective NOE NMR spectrum (800 MHz) of the same solution as in Figure 23. The irradiated H″ of myricetin-3-O-α-L-rhamnopyranoside is denoted with thunder, and the protons showing NOE connectivities are circled. T = 298 K, 256 scans, mixing time = 600 ms, and the total experimental time 30 min [41].

5. Multiple Suppression of Strong Resonances for Minor Analyte Observation in 1D and 2D NMR

The detection of minor analytes in the presence of intense resonances from major components remains a significant analytical challenge, primarily due to the dynamic range limitations of the analog-to-digital converter. When dominant signals occupy most of the available dynamic range, the effective sensitivity toward low-abundance species is markedly reduced. A widely adopted strategy to address this issue is the selective suppression of strong resonances. This approach has been successfully applied to the direct analysis of minor constituents in extra virgin olive oil without prior sample modification [113,114], in fish oils and their fractions [115], in the lipid fraction of dairy products [83], and in studies of selected phenolic compounds in extra virgin olive oil [116]. As illustrated in Figure 25, multi-suppression of the intense oleic acid resonances (0.80–5.50 ppm) leads to a substantial enhancement in sensitivity, enabling the quantification of several phenolic compounds—particularly secoiridoids—within the 9.2–11.8 ppm region. This methodology has facilitated the investigation of oleocanthal, oleacein, and total phenolic content across a range of olive oil samples from Epirus (Figure 26).

Figure 25. (A) Comparison of standard excitation (SE) and multiple spectral excitation (MSE) spectra of selected Epirus olive oil; (B) Aldehydic region of the SE and MSE spectra, with assignment of several phenolic compounds [116].

Figure 26. (A) The analyzed Epirus olive oil samples (N = 178). The sum of oleocanthal and oleacein distributions. The red line represents the mean, and the brown line the median. (B) The total phenolic distributions. The red line represents the mean and the brown line the median [116].

Multi-solvent suppression has also been extended to 2D NMR experiments. Mayer et al. [117] employed an interleaved ultrafast COSY sequence incorporating multi-solvent suppression for wine analysis. By suppressing the ethanol triplet and quartet along with the residual water signal, the method enabled the detection of 11 additional metabolites compared to conventional water-suppression schemes, within an experimental time of 23 min. However, the sensitivity of ultrafast COSY is typically 5–10 times lower than that of 1D experiments, making it less suitable for samples containing very low analyte concentrations [117].

6. Band-Selective Excitation Techniques for Minor Analyte Observation

Band-selective excitation of weak analytes in the presence of intense matrix signals represents a more general and efficient alternative to multiple-suppression schemes [118,119]. Band-selective excitation techniques have the advantage that they can be applied in a frequency range of arbitrary width, which allows an increase in the receiver gain and improved quantification of weak analytes. This approach enabled a significant reduction in acquisition time for a honey sample (0.2 g of honey in 0.6 mL DMSO-d₆): while a standard ¹H NMR spectrum was recorded in 7 h 22 min, a band-selective excitation experiment targeting the 7.0–11.0 ppm spectral region yielded significantly enhanced signal intensities in 45 min of experimental time, allowing excellent identification of 5-hydroxy methyl-furfural, furfural and o-p-methoxybenzaldehyde [119].

Band-selective NMR has also been successfully applied to suppress unwanted signals (≥98%) in complex matrices such as orange juice and rice root extract [120]. The resulting increase in the receiver gain and, thus, in sensitivity with the use of band-selective 1D-TOCSY and 1D-NOE enabled unambiguous assignments of various hydroperoxide resonances in binary CDCl₃/DMSO-d₆ solvents for oleic acid, linoleic acid, α-linolenic acid, as well as mildly oxidized mayonnaise samples [121]. Selective 1D NOE facilitated magnetization transfer from the strongly deshielded hydroperoxide proton to the adjacent allylic CH group (Figure 27), while subsequent 1D TOCSY of the identified allylic proton propagated magnetization through J coupling to the olefinic protons, enabling signal assignment of various hydroperoxides.

Figure 27. Signal assignment of hydroperoxides with the combined use of 1D NOE () and 1D TOCSY () [77].

Band-selective ¹H-¹³C HSQC and HMBC NMR experiments offer substantial improvements in digital resolution along the ¹³C dimension by reducing the heteronuclear spectral width without reducing spectral folding [122]. For example, Dais et al. [123] employed band-selective ¹H-¹³C HMBC experiments of marine dietary supplements with a selective ¹³C dimension of 2 ppm to unequivocally determine the positional distribution of ω-3, ω-6 and ω-9 fatty acids of the glycerol moiety. Wang et al. [124] utilized band-selective HSQC to identify chlorine-substituted ¹³C natural products. Band-selective ¹H-¹³C HSQC NMR was applied for species discrimination and identification of adulteration of Panax Linn extracts [125]. Selective excitation of anomeric ¹³C signals of sugars, which are typically in the region of 95–110 ppm, combined with non-uniform sampling (NUS), yielded excellent improvement in resolution (<0.5 ppm) and significant reduction in the experimental time (9 min 43 s) (Figure 28). Twelve well-separated resonances were identified for Rb1, Rc, Rd, Re, Rf and Rg1 ginsenosides from Panax Linn extracts (Figure 29, Table 5).

Figure 28. Comparison of HSQC and band-selective (bs) HSQC NMR (500 MHz, pyridine-d₅) spectra, all recorded with number of scans = 8. (A) HSQC of the Panax ginseng (PG) extract. TD = 128, acquisition time = 37 min 27 s. (B) HSQC of the PG extract. TD = 512, acquisition time = 2 h 46 min. (C) Bs-HSQC of the PG extract. TD = 128, acquisition time = 37 min 7 s. (D) Bs-HSQC with 25% non-uniform sampling (NUS) of the PG extract. (E) Bs-HSQC with 25% NUS of the Panax notoginseng (PN) extract. (F) Bs-HSQC with 25% NUS of the Panax quinquefoliam (PQ) extract, acquisition time = 9 min 43 s. The numbers 1–14 in the spectra (D–F) correspond to those in Table 5 [125].

Figure 29. Main ginsenosides in Panax Linn extracts. The attributed anomeric carbons are highlighted in red [125].

Table 5. The resonance assignment of each cross-peak of the main ginsenosides in Panax Linn extracts, in band-selective HSQC with 25% non-uniform sampling (NUS) (500 MHz, pyridine-d₅) [125].

No.	δ(13 C), ppm	δ(1H), ppm	Assignment
1	101.75	6.50	Re-1‴
2	103.55	5.93	Rf-1″
3	104.55	5.78	R1-1‴
4	109.94	5.66	Rc-1″
5	105.79	5.37	Rc-1⁗ + Rb1-1⁗
6	101.64	5.25	Re-1″
7	97.99	5.14	1H/13C-1′
8	105.16	5.09	Rb1-1″
9	105.72	5.01	Rg1-1″
10	104.52	4.99	Unknown1
11	106.78	4.93	Unknown2
12	103.33	4.91	R1-1″
13	103.66	4.91	Rf-1′
14	104.89	4.90	Rd-1″ + Rc-1‴ + Rb1-1‴

7. Complementary Use of NMR and MS Spectroscopy-Statistical Heterospectroscopy

Statistical heterospectroscopy (SHY) has emerged as a powerful approach for integrating high-resolution NMR and UPLC MS datasets acquired from the same samples [126,127]. By correlating NMR resonances with MS m/z features, SHY effectively disperses NMR signals into a second dimension defined by mass spectrometric information, thereby enhancing the structural interpretability of complex mixtures. This bioinformatics-driven methodology has proven particularly valuable in metabolomics. The SHY framework has been extended beyond NMR and LC MS combinations to include, for example, NMR–HPTLC analytical workflows, where multiple samples are analyzed by both techniques [128]. Through heterocovariance analysis [129] and statistical correlation spectroscopy, variations in ¹H NMR signals across neighboring chromatographic fractions have been correlated with their corresponding bioactivities [130,131,132,133,134]. This integrative strategy shows considerable promise for the rapid identification of bioactive constituents in complex natural matrices, potentially reducing the need for laborious and time-consuming isolation procedures.

8. Computational Chemical Shift Prediction in Mixture Analysis

Machine learning (ML) and, more recently, artificial intelligence (AI) have introduced robust and increasingly accurate frameworks for predicting NMR chemical shifts [135,136,137]. Among the most widely used open-source software is NMRShiftDB2 [138], which provides an extensive collection of 1D and 2D NMR data for more than 10,000 molecules, including over 44,000 ¹³C spectra [139]. The performance of predictive models is typically evaluated using quantitative metrics such as the mean absolute error (MAE), root-mean-square error (RMSE), and the coefficient of determination (R²) between experimental and predicted chemical shifts [137].

Despite these advances, compound identification in natural products and metabolomics remains highly challenging due to severe spectral overlap, the intrinsic complexity of crude extracts, the limited availability of high-quality experimental data, and the diversity of solvent systems used for NMR acquisition. For example, the Natural Products Magnetic Resonance Database (NP-MRD) currently contains NMR data for fewer than 7% of known natural products [140]. Recent developments in graph neural networks (GNNs) have significantly improved the prediction of ¹H and ¹³C chemical shifts for small organic molecules, particularly when solvent effects—critical for accurate prediction—are explicitly incorporated [141]. The latest NP-MRD module, PROSPRE, provides predicted ¹H NMR chemical shifts for approximately 280,000 molecules, representing a major step forward for the dereplication and identification of unknown natural products [142].

A further important development is the emergence of hybrid quantum-mechanical (QM) and machine-learning (ML) approaches, which enhance structural prediction with DFT-derived datasets. Guan et al. [143] introduced a GNN-based architecture trained on DFT-computed ¹H and ¹³C chemical shifts, while Novitskiy and Kutateladze [144] demonstrated an ML-assisted workflow for revising misassigned natural-product structures—particularly alkaloids—using experimental NMR data in conjunction with DFT-calculated chemical shifts and J-coupling constants. Bhadauria et al. [145] further integrated NMRShiftDB with DFT cross-validation, reporting sub-ppm accuracy for ¹³C chemical-shift prediction.

Nevertheless, even with these methodological advances and the emergence of increasingly automated structure-elucidation pipelines—including DFT-based workflows [146,147]—significant challenges remain. Many natural products and metabolites share closely related biosynthetic origins, nearly identical substructural motifs, and highly uneven concentration profiles, complicating unambiguous identification. In this context, systematic metabolome-level classification strategies based on characteristic chemical shift ranges can offer valuable complementary resolution. For example, data-mining approaches combined with quantification of ¹H–¹³C HMBC cross-peaks associated with strongly deshielded C(5)–OH signals in the relatively uncrowded 11–14 ppm region have demonstrated notable advantages for distinguishing flavonoid subclasses [92].

9. NMR Experimental Parameters and Performance Characteristics

9.1. ¹H NMR Experiments

For sufficiently well-resolved and sharp phenolic OH and hydroperoxide C–O–O–H resonances (Δν_1/2 ≤ 2), the quantification limit can reach the low-μM range when using a ≥500 MHz spectrometer equipped with a cryoprobe and a total experimental time of approximately 15 min. The limit of detection (LOD) may be as low as 1 μM, provided that an adequate number of scans is accumulated, corresponding to a total acquisition time of roughly 1 h. The method exhibits excellent linearity over the 0.1–40 mM concentration range (R² = 0.9996). The coefficient of variation (CV, %RSD), calculated from three replicate samples of identical origin, can be <1.8%, demonstrating very good repeatability.

9.2. ¹H-¹³C HSQC and HMBC Experiments

Two-dimensional ¹H-¹³C HSQC experiments are widely used in mixture analysis because they are straightforward to implement, offer substantial resolution enhancement in the ¹³C dimension, and provide sufficient sensitivity even in the low-mM range. However, minor components are often underestimated when strong resonances dominate the spectrum. Quantification is feasible when reference standards or calibration curves based on known concentrations are employed [83]. A recent study provides a comparative evaluation of different HSQC-based approaches designed to improve quantitative accuracy [148]. The technique shows very good linearity in the 3–15 mM concentration range (R² = 0.993). The limit of detection is typically 2–4 mM on a 500 MHz instrument equipped with a cryoprobe. This detection limit is expected to improve further with the use of band-selective HSQC combined with non-uniform sampling (NUS).

The ¹H-¹³C HMBC experiment and its variants constitute a powerful set of tools for mixture analysis, as they can be implemented routinely and enable the simultaneous identification of diverse classes of natural products, including flavonoids, aldehydes, phenolics, and compounds bearing conjugated double bonds. Optimization of the experiment typically involves two stages: one targeting ^2,3J(¹³C, ¹H) couplings of 8–10 Hz, and a second focusing on long-range ⁿJ(¹³C, ¹H) couplings (<4 Hz for n = 4–6). The latter is particularly important for establishing extended connectivities in highly substituted aromatic systems such as flavonoids. Although HMBC is exceptionally informative for structural elucidation, it does not inherently support quantitative analysis, as cross-peak intensities are not proportional to analyte concentration. Quantification can nevertheless be achieved indirectly by exploiting well-resolved and diagnostically useful ¹H NMR resonances—such as the strongly deshielded phenolic OH signals—once the corresponding structural assignments have been established through HMBC.

9.3. 1D Spin-Chromatography Techniques

In complex mixtures, 1D TOCSY experiments are optimized so that magnetization transfer occurs exclusively within J-coupled spin networks, thereby avoiding Hartmann–Hahn transfer between unrelated networks. Shaped pulses of 20–80 ms followed by MLEV-17 mixing sequences—with alternating phases to suppress off-resonance effects [97]—or DIPSI-2 sequences with continuous phase modulation [149] are effective in many cases [99,100,101]. Typical TOCSY mixing times for small molecules are up to 100 ms, although values as high as 300 ms have been used for systems with extensive J-coupling, such as lipids [100].

In transient NOE, a shaped pulse of ~50 ms combined with gradients to remove transverse magnetization and phase alternation to suppress zero-quantum artifacts has proven effective [41]. Consecutive mixing times of 100–600 ms are typically employed to ensure linearity and maintain direct cross-relaxation before spin diffusion becomes dominant.

9.4. Multiple Suppression Techniques

As demonstrated in Section 5, this approach has been successfully applied to the direct analysis of minor constituents—e.g., in extra-virgin olive oil—without prior sample modification [113,114]. However, selective presaturation of abundant analytes induces spin diffusion and NOE transfer, leading to non-quantitative intensities for minor analytes that are spatially close to major components. This issue is particularly pronounced in crude extracts containing analytes with diverse molecular weights, tumbling times, and extensive intermolecular contacts. In such cases, the use of 1D gradient-enhanced phase-sensitive NOESY (1D NOESY-GPPS) is recommended [113,114,116,150].

9.5. Band-Selective Excitation Techniques

Band-selective (narrow-band) excitation is among the most effective strategies for quantifying minor metabolites in the presence of overwhelmingly strong signals in crude extracts (see Section 5). Such pulses must ensure a uniform flip-angle excitation profile and consistent phase response within the selected bandwidth, while maintaining near-zero excitation outside the band. This is achieved using shaped pulses such as REBURP, EBURP, or adiabatic pulses, combined with double pulsed-field-gradient spin echoes (DPFGSE) to dephase residual transverse magnetization and phase-cycled coherence selection to eliminate undesired pathways [119,120].

10. Conclusions and Prospects for Future Research: “To Separate or Not to Separate?”

From the results detailed in the preceding sections, it is evident that there are several organic functional groups with ¹H and ¹³C chemical shift ranges which can provide significant resolution advantages under specific NMR experimental conditions. Chromatographic separations are, therefore, unnecessary in several cases of complex mixture analysis of natural products, as summarized below:

(i) Very sharp phenol-OH groups with Δν_1/2 ≤ 2.5 Hz can be obtained after careful optimization of experimental conditions that influence ¹H exchange rates, e.g., the addition of TFA, use of DMSO-d₆ as solvent and temperatures < 303 K [33,34,35,36,37,38,39,40,41]. This allows the large ¹H NMR chemical shift range of 8–19 ppm to be exploited in combination with 2D ¹H-¹³C HMBC, which provides an excellent means of investigating crude matrices of polyphenol natural products through a systematic classification of various classes of flavonoids via data mining and quantification of HMBC cross-peaks [37,92].

(ii) Aldehyde protons and carbons exhibit highly diagnostic resonances typically in the region of 9.0–10.5 ppm and 190–210 ppm, respectively. This, in combination with ¹H-¹³C HSQC and HMBC experiments, provides effective techniques for the investigation of lipid-oxidation products of oils which are subjected to a variety of cooking processes and storage conditions [56,57,58,59,60,61], and for the investigation of important secoiridoid compounds as a new index for characterization of extra virgin olive oils [62,63,64,65,66,67,68,69,70].

(iii) The extensive chemical shift range of hydroperoxide protons (7.0–11.8 ppm, depending on the solvent hydrogen-bonding ability) [77,78,79,121] in combination with selective 1D NOE, 1D TOCSY, and 2D ¹H-¹³C HSQC, could be of great importance in investigating edible oils and more generally food samples susceptible to oxidation.

(iv) Conjugated double bonds exhibit highly diagnostic resonances in the region of 6.85–5.50 ppm, which in combination with selective suppression of major lipid resonances and band-selective ¹H-¹³C HSQC experiments could be of importance in investigating conjugated lipids in dairy products [83] and oxidation products of polyunsaturated lipids [56,57,58,59,60,61,77,84,85].

(v) Deshielded aliphatic CH protons, due to electron-withdrawing substituents, and deshielded olefinic protons may appear in less crowded spectral regions and, thus, in combination with ¹H-¹³C HSQC/HMBC and 1D TOCSY experiments, can provide a valuable tool in mixture analysis, as in the case of Artemisia annua extracts [96].

(vi) Chemical shift-selective 1D TOCSY experiments can provide an important tool for elucidating an extensive spin-coupling network even in the presence of major analytes with significantly stronger NMR resonances. The technique has found numerous applications in metabolomics analysis of amino acids [97], juices [98], wine [101], milk and dairy products [99], raw honey samples [117] and crude extracts of natural products [96].

(vii) Band-selective excitation of minor analytes with significantly enhanced sensitivity is the most promising technique since it can be applied in any frequency-selective region and, thus, it can effectively suppress strong resonances from major analytes. The use of band-selective ¹H-¹³C HSQC/HMBC experiments [125] can provide significantly improved digital resolution in the ¹³C dimension and, thus, resolving power in mixture analysis.

The aforementioned NMR techniques and applications coupled with advances in machine learning [3,135,136,137,138,139,140,141,142,143,144,145] and computational metabolomic approaches, notably that of heterospectroscopy [126,127], advances in DFT calculations of atomic-level structures of natural products [7,46,145,146,147,148,149,150,151], and the accurate quantification of metabolites in complex mixtures, which can easily be accomplished and requires only a single well-resolved signal [1,152,153], will open new avenues of application in mixture analysis. The answer to the critical dilemma “to separate or not to separate?”, therefore, is “not to separate” in numerous investigations of natural product mixture analysis.