Evaluating sampling uncertainty in the quantitative 1H nuclear magnetic resonance analysis of lignin

Pawade, S. S., Toom, L., Herodes, K., and Leito, I. (2025). "Evaluating sampling uncertainty in the quantitative 1H nuclear magnetic resonance analysis of lignin," BioResources 20(1), 2234–2242.

Abstract

In recent years, lignin analysis utilizing quantitative nuclear magnetic resonance (qNMR) has attracted considerable interest and has been the subject of numerous studies. However, evaluating the measurement uncertainty of qNMR results of lignin remains a challenge. Specifically, uncertainty originating from lignin sampling or subsampling has been overlooked in a large majority of articles. Although lignin is a reasonably homogeneous substance, it is nevertheless a solid, and individual samples collected from the same bulk may have somewhat different compositions depending on mixing and the amount of sample taken. The objective of this study was to evaluate the influence of sampling uncertainty on qNMR analysis of lignin-based analysis as a case study, with an exclusive focus on the relative quantification method. The results from this study demonstrate that sample-to-sample variations can contribute to approximately half of the variability in actual qNMR measurements. The relative standard deviation (RSD) of sample-to-sample variability was 2.4%. In contrast, the other sources of variability related to qNMR, including measurement, baseline irregularities, and partial peak overlap, caused an RSD of 4.4%. The total variability RSD was 5.0%. In this article, two calculation approaches were presented for evaluating the uncertainty due to sampling from replicate measurement data of different samples, which may be helpful for practitioners in the field.

Download PDF

Full Article

Evaluating Sampling Uncertainty in the Quantitative ¹H Nuclear Magnetic Resonance Analysis of Lignin

Shrikant Shivaji Pawade ,* Lauri Toom , Koit Herodes , and Ivo Leito

DOI: 10.15376/biores.20.1.2234-2242

Keywords: Lignin; NMR; Quantitative NMR; Sampling; Uncertainty

Contact information: University of Tartu, Institute of Chemistry, Ravila 14a, 50411 Tartu, Estonia;

* Corresponding author: shrikant.shivaji.pawade@ut.ee

Graphical Abstract

INTRODUCTION

Lignin is the second-most abundant type of biopolymer on Earth, accounting for 30% of organic carbon (Boerjan et al. 2003; Lu et al. 2017; Pawade et al. 2023). Lignin is obtained in large amounts (approximately 60 to 70 million tonnes annually) as a by-product of the paper and pulp industry. As such, it is one of the most important renewable organic feedstocks. However, lignin is currently significantly underutilized compared with cellulose (An et al. 2015). Lignin is one of the few renewable aromatic biopolymers and stands out for its aromatic nature among other biopolymers (Wang et al. 2021). Approximately 98% of lignin is currently used as a source of energy and heat (Constant et al. 2016).

Based on the plant species, tissue type, and specific cell wall layer, lignin is composed of aromatic units with different structures and ratios of aromatic units (Happs et al. 2021). In general, lignin consists primarily of p-hydroxyphenyl (H), guaiacyl (G), and syringyl (S) units. The S, G, and H units interlink with each other via different linkers/bonds: β-O-4 (β-ether), α-O-4, β-β (resinol), and β-5 (Balakshin et al. 2011). β‑O‑4, β‑β, β‑5, PhOMe, PhOH, aliphatic methoxy and aliphatic OH, aromatic H, and aldehyde substructures and linkages are the prevalent structural fragments that are typically quantitatively analyzed (Balakshin et al. 2011; Shimizu et al. 2017). Analytical techniques can determine the types of monolignols and their contents, the distribution of inter-unit linkages, and the functional groups that make up the chemical structure of lignin. For advanced applications of lignin, regardless of the depolymerization approach, fundamental knowledge of its structural features and physicochemical properties is essential (Nayak et al. 2020).

Quantitative nuclear magnetic resonance (qNMR) spectrometry seems to be the most widespread and versatile technique for the quantitative analysis of lignin (Capanema et al. 2004, 2005). It enables determining numerous structural characteristics of lignin by using ¹H (Faix et al. 1994), ¹³C (Balakshin and Capanema 2015a,b; Balakshin et al. 2016), or ³¹P (Gracia-Vitoria et al. 2022) (after derivatization) as the nuclei, as well as different two-dimensional techniques (Zhang and Gellerstedt 2007; Constant et al. 2016; Amiri et al. 2019). The authors recently developed an interest in determining the possible accuracy attainable through the qNMR analysis of lignin, and the authors conducted a literature survey (Pawade et al. 2023). The literature analysis revealed several uncertainty sources inherent in the measurement, for example, the repeatability of spectra (especially if the signal-to-noise ratio is low), the accuracy of peak integration (especially in the case of baseline irregularities), an overlap of signals of interest with other signals (Balakshin and Capanema 2015), and different NMR-specific uncertainty sources, such as deviations in the coupling constant, resonance offset effects, and effects of ¹H T₁ relaxation (Zhang and Gellerstedt 2007; Amiri et al. 2019).

Interestingly, however, very little attention has been devoted to the possible uncertainty arising from the sampling or subsampling of the lignin. In addition, the sampling protocol used has not been reported in most publications. At the same time, it is widely recognized that sampling (or subsampling) is among the most crucial uncertainty sources in most chemical analyses (Ramsey et al. 2019; Medeiros et al. 2022; Sano and Lourenço 2023).

Although lignin can be reasonably homogeneous, it is still a natural solid material, and such materials are always, at least to some extent, inhomogeneous. This can be caused, e.g., by raw material variability (powder composition can change during the fill run), segregation (smaller particles tend to move to the bottom, and larger particles tend to stay on top), or absorption of moisture (topmost layer will absorb more moisture than the middle or bottom layers). Thus, subsamples taken from the same bulk of lignin can have somewhat different compositions and, therefore, give different NMR spectra.

Uncertainty from sampling was mentioned only in a couple of reports (Froass et al. 1998; Balakshin and Capanema 2015), and only in one of them was some quantitative data presented (Froass et al. 1998). This information enabled the authors to approximate the subsampling and sample preparation variability, which ranged from 5% to 8% relative standard deviation (RSD). At the same time, in most reports, even the most in-depth ones, sampling/subsampling is not mentioned. As a result, in many cases, it is not known whether replicate measurements were performed using the same or different samples/subsamples. In such cases, when an RSD estimate between replicate measurements is provided, its meaning remains obscure.

Considering this, the authors decided to investigate the uncertainty due to the inhomogeneity of subsampling in the analysis of commercial lignin and how it compares with the variability of NMR spectrometric analysis itself.

EXPERIMENTAL

Reagents and Samples

Softwood kraft lignin (alkali, low sulfate content) from pine wood was procured from Sigma-Aldrich USA, and DMSO-d₆ (99.8% with 0.03% tetramethylsilane (TMS)) was acquired from Deutero GmbH Kastellaun, Germany.

Sampling and Sample Preparation

The sampling protocol was as follows: Lignin samples were collected from various parts of a lignin container containing 1.0 kg of lignin. Altogether, 15 samples were obtained from the bulk: 5 from the top layer, 5 from the middle layer, and 5 from the bottom layer. Every sample was analyzed in quadruplicate from the same solution over a time span of up to a little more than a week. To prepare the samples for ¹H nuclear magnetic resonance (NMR) analysis, each sample was weighed to 5.0 mg of lignin in an NMR tube and dissolved into 0.6 g DMSO-d₆. A small sample size and low concentration of the resulting solution were used to ensure the complete dissolution of lignin. A larger sample size of a larger lignin sample was not used for three reasons: (1) complete dissolution – very important to guarantee good-quality spectra – would have required large amounts of the expensive deuterated solvent; (2) in many investigations, in contrast to this work, only small samples may be available; and (3) in this work the uncertainty of weighing does not influence the results, as all the quantification is done using ratios of peak areas. The NMR tube was firmly closed and held in an ultrasonic bath for 15 min to completely dissolve the lignin residue. All samples were prepared using the same method within 2 to 3 days. The ¹H NMR spectra of each sample were measured four times, and replicate analyses were performed in random order within a two-week period.

NMR Spectrometry

Proton nuclear magnetic resonance (¹H NMR) spectrometry was used in this work. All spectra were acquired with a Bruker Avance-III 700 MHz NMR spectrometer from Switzerland with a 5-mm BBO (broadband observe) probe. The sample temperature was maintained at 25 °C for all measurements. The ¹H NMR spectra were acquired using 81920 data points, 30° pulse, recycle time of 5.91 s (acquisition time 2.9 s, relaxation delay 3.0 s) (Zhang and Gellerstedt 2007), and 2048 scans preceded by 4 steady-state scans.

For reference, the authors measured T₁ values and obtained 1.0 to 2.0 s, depending on the specific structural fragment. Thus, it is reasonable to argue that the chosen NMR acquisition parameter (mainly the D₁ value) may have been insufficient to obtain full relaxation of signals between scans, and therefore, signal areas can be biased. However, absolute quantification was not the focus of this study. All results and conclusions are based on the between-sample variability of the ratios of integrals. This information can still be obtained because the signal inaccuracies either cancel out or remain the same for replicate measurements.

All acquired ¹H NMR spectra were subjected to the same data treatment using TopSpin software (Bruker TopSpin 3.2). The spectra were zero-filled to 256,000 points, line broadening (LB) of 0.1 Hz was used, Fourier transformation, manual phase correction, and baseline correction was automatically done using the spline baseline correction method (the high importance of accurate baseline correction has been stressed in literature) (Balakshin and Capanema 2015a,b). The spline baseline correction was based on a predefined set of data points, which were considered part of the baseline. The authors chose the same baseline points (31 points, which were saved in a baslpnts file that was used for spline baseline correction on every spectrum) on every spectrum to define the baseline. The TopSpin software then was used to fit the regions between these points and subtract from the measured spectrum (the command for this procedure in the TopSpin software is sab).

All the ¹H NMR spectra were calibrated using the signal corresponding to TMS at δ = 0.0 ppm. Then, all signal regions of the spectra were integrated, excluding the TMS and DMSO-d₅ signals. A certified NMR reference standard by Bruker was routinely used to calibrate the 90° ¹H pulse, and the Prosol table (where the pulse lengths were stored) was updated regularly.

All quantitative analyses were performed using a reference peak; that is, all the peak intensity ratios in all regions, both aromatic and aliphatic, were calculated relative to the integral of this peak. The reference peak should be strong and well-separated. Two such peaks were identified: one in the region of 8.40 to 8.65 ppm, belonging to a formate salt in the lignin product, and another one in the region of 1.56 to 1.60 ppm, belonging to aliphatic fragments. Data analysis was performed separately with these two reference peaks. The signals or signal ranges that were used for integration were the aldehyde peak (9.18 to 9.30 ppm), aromatic region (5.80 to 8.00 ppm), peak related to O-CH/O-CH₂ (4.13 to 5.10 ppm), methoxy and hydroxy peaks, and residual water peak from DMSO solvent (2.80 to 4.10 ppm). The signals at 0.60 to 1.50 ppm were due to aliphatic protons that were not oxygenated (An et al. 2015). From all samples, 11 different signal intensity ratios, listed in Table 1, were quantified (both peaks that were used as reference peaks were quantified only when the other peak was the reference). The peaks and ranges, as well as their identifiers, are shown in Fig. 1.

Fig. 1. (A) Representative example of an acquired ¹H NMR spectrum of kraft lignin sample in DMSO‑d₆, (B) Aliphatic region from 0.70 to 2.30 ppm; See the Supporting Information for more spectra

To reveal the effect of sample preparation on the variability of integrals, as opposed to NMR measurement and integration, such signals and ranges were selected for further analysis that were sufficiently (but not overly) intense and where satisfactory baseline correction was possible. For this reason, the region corresponding to methoxy and hydroxy (2.80 to 4.10 ppm) groups were excluded (very high intensity of the signal and residual water involved, making it difficult to correct the baseline) as well as the peak in the region of 9.18 to 9.30 ppm due to its very low intensity.

Data Analysis

The data obtained were checked for outliers using the Dixon Q test. The Dixon Q test was carried out at 95% confidence level (without p-value adjustment), separately for every type of integral for each type of sample (top, middle, bottom), pooling together the data of the 5 replicates. This resulted in 30 Dixon tests, each with 20 datapoints. To check the normality of the data, the data of all integrals were normalized, and the resulting normalized datasets were pooled into a 600 data point set. The normality of the distribution was evaluated by visual comparison with the cumulative normal distribution curve, as well as the linearized normality plot (see the Excel file in the Supporting Information).

Data analysis was performed with the aim of dissecting the overall variability of the relative peak intensities into two components: Variability from measurement and variability from sampling. Data analysis was performed separately for each of the 10 intensity ratios. Two approaches were used. One was the analysis of variance (ANOVA) approach, described in detail in Van Der Veen and Pauwels (2000).

The other approach (termed as “RSD approach”) looks at the overall variability of the results as relative standard deviation (RSD_Total), which can be regarded as composed of two components: Variability caused by sampling (RSD_Sampling) and variability caused by measurement (RSD_Meas):

(1)

Sampling variability cannot be directly evaluated because any evaluation of variability always involves measurements. Therefore, an indirect approach was used, as follows: The variability found as standard deviation over the results obtained with all samples taken from the bulk material (RSD_Total) includes variabilities originating from sampling as well as variabilities from measurement. Variability found from replicate measurements with the same sample (RSD_Meas) includes the variability of measurement only. This can be expressed by the following Eq. 2:

(2)

For every individual signal intensity ratio, the RSD_Total was determined as the RSD of the individual intensity ratios of all replicates of all 15 samples. The RSD_Meas for every individual signal ratio was obtained as pooled RSD (for explanations, see section 6 of the course presented in Leito, Helm, and Jalukse (2015) of the 4 replicate measurement results of the same signal intensity ratio from the 15 samples. The RSD_Sampling for each individual signal ratio was obtained from Eq. 2.

To obtain an estimate of the average RSD across the signal ratios corresponding to the different peaks, the individual RSD values were pooled by calculating the root mean square (RMS) of the RSD values (“Pooled RSD values” in Table 1).

RESULTS AND DISCUSSION

The outlier check revealed one outlier data point out of a total of 600. The authors did not find the reason for the deviation and accordingly did not consider it justified to eliminate it because (1) the outlier tests were carried out at a 95% confidence level, which in the case of 30 tests means that there is a probability of 1 – 0.95³⁰ = 0.79 that there is at least one outlier and (2) given that the overall amount of data is large and leaving the data point out would not change anything in the conclusions of this work. The overall cumulative distribution of the data was indistinguishable from the normal distribution (see Supporting Information).

As expected, the two data analysis approaches yielded almost identical results. There was a somewhat more difference between the results obtained with the different reference peaks. The RMS averages of all four sets of results (two data-analysis approaches and two reference peaks) are presented in Table 1. A detailed calculation file containing the calculations, all individual results obtained with both approaches, and both reference peaks is provided in the Supporting Information. The ANOVA approach may be considered fundamentally more rigorous. At the same time, the RSD approach gives a number of RSD values as interim results, which are more easily interpretable than the interim quantities of ANOVA and may be usable by themselves, for example, for tracking shortcomings in experiments.

Table 1. Overall Relative Standard Deviations of Signal Ratios (RSD_Total), RSD due to measurement (RSD_Meas), and RSD due to sampling (RSD_Sampling)

Data presented as RMS averages of 2 data analysis approaches and 2 different reference peaks. See the Supplementary Information using the weblink provided in the Appendix for the complete calculation and individual results.

As shown in Table 1, the RSD_Total, RSD_meas, and RSD_sampling values varied from 3.3% to 7.4%, 2.7% to 6.4%, and 1.6% to 3.9%, respectively. In most cases, RSD_Meas was larger than RSD_Sampling. There did not seem to be any clear pattern in these variabilities, nor are there any significant outliers. Therefore, it is reasonable to assume that the observed differences were caused by statistical fluctuations.

The pooled RSD values demonstrate that under the experimental conditions used, the RSD values of NMR measurement (together with peak integration) and sampling to the overall RSD of the measurement results differ by approximately two times. This means that uncertainty due to sampling is by no means a negligible source of uncertainty in this type of analysis and, in contrast to general practice (Pawade et al. 2023), should always be considered.

One of the challenges in the measurement was the poor separation of some signals. In addition, baseline correction represents an important difficulty in data processing if quantitative results are desired. In the replicate measurements of lignin, there were minor differences in the peak shapes and slight changes in the chemical shifts. However, by using a larger number of scans, it was possible to address such issues. This technique allows the calculation of the sampling uncertainty of various types of lignin.

While studies of effects of NMR-related parameters on the accuracy of quantitative analysis of lignin are numerous, this is essentially the first study to examine uncertainty/variability due to sampling, and only one material type was examined. In future studies, it might be interesting to investigate the variability due to the sampling of lignins obtained using different technologies and possibly other natural materials. Additionally, the lignin used in this study was relatively homogeneous, being a commercial product in the form of a fine powder. In contrast, samples obtained from pulp mills or industrial sources are likely to exhibit higher inhomogeneity and thus higher sampling uncertainty compared to the results obtained in this study.

CONCLUSIONS

The results demonstrated that under the experimental conditions, the uncertainty due to the sample-to-sample variability was approximately half of the uncertainty accounting for the variability in qNMR measurement and integration. Thus, sampling/subsampling is, by all means, an important uncertainty source.
Two additional aspects are worth mentioning. The sample size was at the low end of what is typically used for such an analysis. Thus, by using larger samples, it may be possible to observe smaller sample-to-sample variability. However, this experimental setup and the results presented illustrate a “good case,” as the sample was a commercial product from an established industrial process, which can be assumed to be reasonably well mixed. For this reason, the sample-to-sample variability found in this study may not necessarily be applicable in more exploratory situations, such as the analysis of less homogeneous crude products or products from experimental processes.
The presented calculation file in Supporting Information can serve as a template for practitioners interested in evaluating the uncertainty due to sampling from replicate measurement data of different samples.

ACKNOWLEDGMENTS

This research was funded by the Estonian Ministry of Education and Research (TK210) and the Estonian Research Council grant PRG690. The experiments were performed using the instrumentation at the Estonian Centre of Analytical Chemistry (TT4, www.akki.ee). This study is part of the Ph.D. thesis of Shrikant Shivaji Pawade.

REFERENCES CITED

An, Y.-X., Li, N., Wu, H., Lou, W.-Y., and Zong, M.-H. (2015). “Changes in the structure and the thermal properties of kraft lignin during its dissolution in cholinium ionic liquids,” ACS Sustain. Chem. Eng. 3(11), 2951-2958. DOI: 10.1021/acssuschemeng.5b00915

Balakshin, M., Capanema, E., Gracz, H., Chang, H., and Jameel, H. (2011). “Quantification of lignin–carbohydrate linkages with high-resolution NMR spectroscopy,” Planta 233(6), 1097-1110. DOI: 10.1007/s00425-011-1359-2

Balakshin, M., and Capanema, E. (2015a). “On the quantification of lignin hydroxyl groups with ³¹P and ¹³C NMR spectroscopy,” J. Wood Chem. Technol. 35(3), 220-237. DOI: 10.1080/02773813.2014.928328

Balakshin, M., and Capanema, E. A. (2015b). “Comprehensive structural analysis of biorefinery lignins with a quantitative ¹³C NMR approach,” RSC Adv. 5(106), 87187-87199. DOI: 10.1039/C5RA16649G

Balakshin, M., Capanema, E. A., Santos, R. B., Chang, H., and Jameel, H. (2016). “Structural analysis of hardwood native lignins by quantitative ¹³C NMR spectroscopy,” Holzforschung 70(2), 95-108. DOI: 10.1515/hf-2014-0328

Boerjan, W., Ralph, J., and Baucher, M. (2003). “Lignin biosynthesis,” Annu. Rev. Plant Biol. 54(1), 519-546. DOI: 10.1146/annurev.arplant.54.031902.134938

Capanema, E. A., Balakshin, M. Y., and Kadla, J. F. (2004). “A comprehensive approach for quantitative lignin characterization by NMR spectroscopy,” J. Agric. Food Chem. 52(7), 1850-1860. DOI: 10.1021/jf035282b

Capanema, E. A., Balakshin, M. Y., and Kadla, J. F. (2005). “Quantitative characterization of a hardwood milled wood lignin by nuclear magnetic resonance spectroscopy,” J. Agric. Food Chem. 53(25), 9639-9649. DOI: 10.1021/jf0515330

Constant, S., Wienk, H. L. J., Frissen, A. E., Peinder, P. de, Boelens, R., van Es, D. S., Grisel, R. J. H., Weckhuysen, B. M., Huijgen, W. J. J., Gosselink, R. J. A., et al. (2016). “New insights into the structure and composition of technical lignins: A comparative characterisation study,” Green Chem. 18(9), 2651-2665. DOI: 10.1039/C5GC03043A

Faix, O., Argyropoulos, D. S., Robert, D., and Neirinck, V. (1994). “Determination of hydroxyl groups in lignins evaluation of ¹H-, ¹³C-, ³¹P-NMR, FTIR and wet chemical methods,” Holzforschung 48(5), 387-394. DOI: 10.1515/hfsg.1994.48.5.387

Froass, P. M., Ragauskas, A. J., and Jiang, J. (1998). “Nuclear magnetic resonance studies. 4. Analysis of residual lignin after kraft pulping,” Ind. Eng. Chem. Res. 37(8), 3388-3394. DOI: 10.1021/ie970812c

Gracia-Vitoria, J., Rubens, M., Feghali, E., Adriaensens, P., Vanbroekhoven, K., and Vendamme, R. (2022). “Low-field benchtop versus high-field NMR for routine 31P analysis of lignin, a comparative study,” Ind. Crop. Prod. 176, article ID 114405. DOI: 10.1016/j.indcrop.2021.114405

Happs, R. M., Addison, B., Doeppke, C., Donohoe, B. S., Davis, M. F., and Harman-Ware, A. E. (2021). “Comparison of methodologies used to determine aromatic lignin unit ratios in lignocellulosic biomass,” Biotechnol. Biofuels 14(1), article 58. DOI: 10.1186/s13068-021-01897-y

Leito, I., Helm, I., and Jalukse, L. (2015). “Using MOOCs for teaching analytical chemistry: Experience at University of Tartu,” Anal. Bioanal. Chem. 407(5), 1277-1281. DOI: 10.1007/s00216-014-8399-y

Lu, Y., Lu, Y.-C., Hu, H.-Q., Xie, F.-J., Wei, X.-Y., and Fan, X. (2017). “Structural characterization of lignin and its degradation products with spectroscopic methods,” J. Spectrosc. (2017) 1-15, article ID 8951658. DOI: 10.1155/2017/8951658

Nayak, K. K., Parkhey, P., and Sahu, R. (2020). “Analysis of lignin using qualitative and quantitative methods,” in: Lignin, Springer Series on Polymer and Composite Materials, S. Sharma and A. Kumar (eds.), Springer International Publishing, Cham, Switzerland, pp. 115-138. DOI: 10.1007/978-3-030-40663-9_4

Pawade, S. S., Toom, L., Herodes, K., and Leito, I. (2023). “Accuracy of quantitative NMR analysis: A case study of lignin,” J. Chem. Metrol. DOI: 10.25135/jcm.88.2304.2753

Shimizu, S., Akiyama, T., Yokoyama, T., and Matsumoto, Y. (2017). “Chemical factors underlying the more rapid β-O-4 bond cleavage of syringyl than guaiacyl lignin under alkaline delignification conditions,” J. Wood Chem. Technol. 37(6), 451-466. DOI: 10.1080/02773813.2017.1340957

Talebi Amiri, M., Bertella, S., Questell-Santiago, Y. M., and Luterbacher, J. S. (2019). “Establishing lignin structure-upgradeability relationships using quantitative ¹H-¹³C heteronuclear single quantum coherence nuclear magnetic resonance (HSQC-NMR) spectroscopy,” Chem. Sci. 10(35), 8135-8142. DOI: 10.1039/C9SC02088H

Van Der Veen, A. M. H., and Pauwels, J. (2000). “Uncertainty calculations in the certification of reference materials. 1. Principles of analysis of variance,” Accredit. Qual. Assur. 5(12), 464-469. DOI: 10.1007/s007690000237

Wang, S., Zhang, K., Li, H., Xiao, L.-P., and Song, G. (2021). “Selective hydrogenolysis of catechyl lignin into propenylcatechol over an atomically dispersed ruthenium catalyst,” Nat. Commun. 12(1), article 416. DOI: 10.1038/s41467-020-20684-1

Zhang, L., and Gellerstedt, G. (2007). “Quantitative 2D HSQC NMR determination of polymer structures by selecting suitable internal standard references,” Magn. Reson. Chem. 45(1), 37-45. DOI: 10.1002/mrc.1914

Article submitted: November 14, 2024; Peer review completed: December 28, 2024; Revisions accepted: January 17, 2025; Published: January 27, 2025.

DOI: 10.15376/biores.20.1.2234-2242

APPENDIX

Additional spectra, as well as the file containing all the integral values and all calculations (in the MS Excel format), are available as Supporting Information at https://doi.org/10.6084/m9.figshare.25516477